Copyright @ Stephen A. Drury 2001

The right of Stephen A. Drury to be identified as author of this work has been asserted byhim in accordance with the Copyright, Designs and Patents Act 1988.

All rights reserved. No part of this publication may be reproduced or transmitted in anyform or by any means, electronic or mechanical, including photocopy, recording or anyinformation storage and retrieval system, without permission in writing from thepublisher or under licence from the Copyright Licensing Agency Limited, of 90 TottenhamCourt Road, London WIT 4LP.

Any person who commits any unauthorised act in relation to this publication may be liableto criminal prosecution and civil claims for damages.

First published in 1987 by Allen and UnwinSecond edition published in 1993 by Chapman and HallThird edition published in 2001:

Published in the UK by:Nelson Thomes LtdDelta Place27 Bath RoadCHEL TENHAMGL53 7THUnited Kingdom

Published in the USA and Canada by:Blackwell Science Inc.Commerce Place350 Main StreetMalden, MA 01248.,5018USA

01 02 03 04 05 / 10 9 8 7 6 5 4 3 2 1

A catalogue record for this book is available from the British Library

ISBN 0 7487 6499 2

Library of Congress Cataloging-in-publication Data

Drury, S. A. (Stephen A.), 1946-Image Interpretation in Geology (Steve Drury .-3rd ed.

p. CIn.Includes bibliographical references and index.ISBN 0 632 05408 51. Geology-Remote Sensing. I. Title

QE33.2.R4D78 2001550' .22'2-dc21

For further information on Nelson Thomes visit our website:www.nelsonthomes.comFor further infonrlation On Blackwell Science Inc. visit our website:www.blackwell-science.com

The inclusion of TNTlite in the accompanying CD-ROM to Image Interpretation in Geology(Third edition) does not imply any endorsement of the soffwarebr either the publishers orthe author.

Set by Graphicraft Limited, Hong Kong

Printed and bound in Q1ina by L.Rex

Contents

Preface, viiAcknowledgements, x

Electromagnetic Radiation and Materials, 11.1 The nature of electromagnetic radiation, 1

1.2 The generation of electromagnetic radiation, 21.3 Matter and electromagnetic radiation, 4Associated

resources on the CD-ROM, 14Further reading, 14

5 Digital Image Processing, 1225.1 The image histogram, 1235.2 Contrast stretching, 1255.3 Spatial-frequency filtering, 1335.4 Data reduction, 1385.5 Band ratioing, 1435.6 Pattemrecognition,145Associated resources on the CD-ROM, 157Further reading, 157

6Human Vision, 162.1 The eye and visual cortex, 16 .

2.2 Spatial resolving power, 192.3 Seeing brightness, 242.4 Producing, seeing and representing colour, 262.5 Perception of depth, 292.6 Dangerous illusions, 30Further reading, 31

Thermal Images, 1606.1 What a thermal image shows, 160

6.2 Qualitative interpretation of thermalimages, 1646.3 Semiquantitative analysis, 1696.4 Multispectral thermal data, 172Associated resources on the CD-ROM, 174Further readinj?;, 174

3 7 Radar Remote Sensing, 1767.1 Interactions between radar and stJrfacematerials, 1767.2 Interpretation of radar images/ISO7.3 Radar interferometry ,203Associated resources on the CD-ROM, 204Further reading, 204

8

How Data Are Collected, 323.1 Photography, 32

3.2 Vidicon cameras, 373.3 Line-scanning systems, 383.4 Pushbroom systems, 423.5 Microwave imaging, 433.6 Imaging spectrometers, 493.7 Gamma-ray spectrometers, 493.8 A short history of remote sensing, 503.9, Airborne data, 513.10 Basic characteristics of orbiting satellites, 533.11 Data from staffed spacecraft, 553.12 Data from unstaffed spacecraft, 573.13 Future prospects, 65Associated resources on the CD-ROM, 6/Further reading, 67

Non-Image Data and GeographicalInformation Systems, 2068.1

Forms of non-image data, 2078.2 Non-image data in raster format, 2088.3.

Data analysis in geographical informationsystems, 2218.4 Concludi~g note, 224Associated resources on the CD-ROM, 225Further reading, 225

49 Geological Applications of Image Data, 227

9.1 Geomorphology,2289.2 Geological mapping, 2319.3 Exploration, 2339.4 Engineering applications, 2419.5 Geochemical hazards, 243Further reading, 244

Photogeology, 684.1 Destructional landforms, 69

4.2 The recognition of rock types, 764.3 Stratigraphic relationships, 904.4 Structural relationships, 954.5 Superficial deposits and constructionallandforms, 107Associated resources on the CD-ROM, 120Further reading, 120

vi Contents

Appendix A Stereometry, 249Further reading, 252

Installing resources,8260Working with the IIG resources, 262Getting help, 262

Appendix D Sources of Image Data, 264Image atlases, 264Websites,264

Appendix B Image Correction, 253B.l Geometric rectification, 253B.2 Replacing dropped lines, 255B.3 De-striping,255BA Removal of random noise, 257Associated res<>\lrces on the CD-ROM, 257Furtherreading,258 .

Glossary, 266Index, 279

Appendix C The CD-ROM Resources, 259Contents of the CD-ROM, 259

Colour plates fall between pp. 148 and 149.

~~~

.

form of anaglyphs that can be viewed:using the viewing spectacles packed with the CD-ROM.This supplements the need to use a lens stereoscope toobtain full benefit from the text figures. Finally there is acollection of various types of image that contain import-ant geological features.

Because there may be readers who wish to study thebook simply as a textbook, and some who have no easyaccess to a computer, the text does hot depend on the CD.At the end of each chapter is a brief guide to the relevantCD-ROM resources. Appendix C gives instructions forinstalling TNTlite, and also appears as Resources.rtf onthe CD-ROM.

Remote sensing roughly means extending humansensory perception to distances greater than we canachieve unaided and to information that is far beyondour physiological capabilities. Vision is far and away ourmost powerful and flexible sense, and so the strict focusis on capturing information about the Earth's propertiesin the nearly continuous, two-dimensional fashion thatis characteristic of images. This is possible only for thoseproperties that control how Earth materials interactwith electromagnetic radiation-not only visible light,but a spectrum that extends from gamma- to microwaveradiation. Other attributes relc1ting to natural variationsin density, magnetic and electrical properties are detect-able, but only in a discontinuous fashion-from pointto point or along surveyed lines. The same holds forvariations in chemistry, topographic elevation and thegeometric structure of rocks, both at and beneath thesurface, plus a host of other kinds of geological informa-tion. Although some of these attributes can be measuredfrom a distance, the immediate results are not images.They are not excluded from the book, however, becausethere are means of recasting numbers distributed irregu-larly in two cartographic dimensions into the form ofimages. Visual perception is unsurpassed in extractingnuances from any kind of image, whatever its meansof derivation. So, there is an overlap between remotesensing and more familiar means of gathering geosci-entific information. Part of it is through images, and partthrough data analysis itself. One of the most importantnew tools is using computers to find patterns and cor-relations among far more layers of information than thehuman intellect can grasp. We deal as a matter of routinewith spatial and to a lesser extent time dimensions, but ageological problem often involves tens of different dimen-sions. The vast bulk of the information is distributed interms of geographical co-ordinates-it can be registeredto maps. An extension from remote sensing is a sort of

The first two editions of Image Interpretation in Geology wona wide readership among undergraduate and profes-sional geologists, since 1987, but advances in the techno-logy of remote sensing and its application demand someupdating. In addition, there are always better ways ofexpressing concepts, so I have revised the style in mostof the chapters. I have replaced several of the images withmore instructive ones, and the further reading extends to1999. Most important, I have added a CD-ROM that I hopewill supplement and extend the usefulness of the book.In its new form, Image Interpretation in Geology transcendsthe original textbook to become potentially a cQmplete,distance-learning course on remote sensing, image pro-cessing and digital mapping for geologists.

Thanks to MicroImages of Lincoln, Nebraska, theCD includes the students' version of their professionalmapping and image processing system, TNTmips. Thispackage, called TNTlite, is a unique example of highlysophisticated freeware, in that it is equally as func-tional as the professional version, limited only in termsof the maximum usable image size and with exportfrom TNTmips format to other formats disabled. TheCD includes both Windows and MacOS versions, sothat almost everyone with access to a modem desktopor portable computer will be able to learn essentialskills to a high level of sophistication. The packagecontains the full and comprehensive TNTmips refer-ence manual and a series of Adobe Acrobat format,Getting Started tutorials, which cover all aspects ofthe system's functionality. Moreover, I have added 11exercises that focus on all the geological applicationsof remote sensing covered by the text, which are basedon a variety of image and nonimage data, mainly cover-ing an area of stunning geology in the semiarid terrainof north-east Africa. They begin at an elementary leveland progressively guide the user through the meanswhereby TNTlite manages, enhances and analysesdigital image data that bear on geological problems.As well as the image data that the exercises use, manyof the other digital images on the CD form a resourcethat instructors can use to create their own exer-cises and assignments, together with advice on howto import their own data in a TNTlite-compatible form.My choice of TNTlite as a teaching tool was solelybecause of MicroImages' unique policy of freely dis-tributing fully functional software, and does not consti-tute an endorsement-readers must make their own

judgement.The CD-ROM includes the full set of stereoscopic pairs

of aerial photographs that appear in Chapter 4, in the

vii

viii Preface

within the military intelligence community. Some of thefirst civilian systems that emerged with declassificationwere orientated to a ,broad and coarse view of meteoro-logy and the oceans. The most revolutionizing aimed atvisible and near-visible infrared parts of the spectrum,where vegetation is readily distinguished, and used aresolution less than 100 m to image small areas of theEarth'~ surface. The Landsat series, initiated in 1972,aimed at providing strategic economic information onthe global cereal crop. This never fully materialized,but as well as many vegetation-orientated users, geo-logists seized on the new, wide and detailed view oflandscape. The potential for complete, near-global coverin a standard form revitalized the ambition to extendgeological mapping at scales better than 1 : 250 000 from25% of the continents to the remainder, by using Landsatimages as supersynoptic aerial photographs-they aremore than just that.

Because the platforms that carried the sensors wereunmanned and in quite high orbits, they had to transmitimages in electronic form. Rather than using ordinarytelevision methods, the decision was taken at an earlystage to code and transmit the data digitally. In Chapter5 you will grasp the enormous opportunities for 'tuning'digital images to human vision &nd extracting hiddeninformation from them. It focuses on the spectral pro-perties of materials in that part of the solar spectrum thatthe Earth's surface reflects. Appendix B complementsthat by a brief account of digital means of removing theblemishes and distortions that can affect many kinds ofelectronically captured data. '

Conceptually more difficult to interpret are images thatdepend on the emission of thermal energy by the surface,and the strange interactions of artificial microwavesused in radar remote sensing. These form the topics ofChapters 6 and 7. Perhaps more so than systems oper-ating at the familiar end of the spectrum, thermal andradar studies are attuned to geological phenomena, butthere are relatively few geoscientists who are well versedin their intricacies, partly because most images fromthem derive from experimental airborne systems. Thefirst decade of the 21st century promises to open newvistas with the deployment of such systems on satellites,and an early grasp of the possibilities will allow readersto take advantage of novel opportunities.

Geoscientists collect all manner of information relatingto natural force fields, environmental geochemistry andexploration records, as well as lithological and structuraldata. A gravity determination, an analysis of soil, a bore-hole log or even a measurement of dip and strike atan outcrop are costly. In past decades, such data weregathered for a single purpose, perhaps an explorationprogramme, then set aside. Simple graphical expressionof these variables rarely exploited their full content, andoften treated them in isolation. Transforming spatially

~ultidimensional aid to geological skills; a geographicalinformation system (GIS).

Chapter 1 sets out to explain in relatively simple termshow matter and electromagnetic radiation interact. It isthese interactions that laid the foundation for designingremote-sensing instruments, planning their use and inter-preting the images that they generate. Although com-puters play an ,ever increasing role in geoscience, they willnever replace the geologist. At the end of the day it is theeye and a mental kit-bag of skills, experience and intu-ition that geneI:ate ideas from perception of the real world.Chapter 2 makes some important points about how ourvisual system functions, and its power and limitations,both from a physiological standpoint and some of itsmore psychological aspects. This might seem strangein a geological text, but it is the key to matching theinformatio~cqntained within images to the make-up ofthe interpreter. The necessary background is completedin Chapter 3, which describes many of the instrumentsinvolved in generating images.Jhis chapter also reviewsthe most. important sources of images. The Internet isnow the single most important means of discoveringwhat information is available, and Appendix C listssufficient, reliable URLs for beginning a search-many ofthe sites include Links to others in the same field, andof course the possible sources evolve continually.

Although remote sensing spans much of the electro-magnetic spectrum, before 1970 the only public sourceof images was from aerial photography. The geologicaluse of aerial photographs since the 1930s has left a price-less heritage of interpretative methods, on which remotesensing builds. Chapter 4 is the core of the book and con-centrates on photogeological interpretation. Most of theimages there are panchromatic photographs, often com-bined in stereopairs that can be viewed using a standardlens stereoscope (the CD contains all of these as anaglyphs,which you can view using the spectacles packed with it).Some are images derived from digital remote sensingsystems, but are used for their content of informationon geologically relevant landforms rather than theirspectral features. Part of photogeology traditionally hasbeen orientated to the extraction of geometric informa-tion and transfer of interpretations to maps; in other wordsto photogrammetry. Space prevented a proper treatmentof this subject, other than aspects of photogrammetrythat need to be carried out in the laboratory or the field,which are covered in Appendix A. There has been littleadvance in photogrammetry for geologists since the 1960s,and I have no hesitation in referring readers to older andmore comprehensive texts on the subject, rather than topresume to improve on our forebears.

Remote sensing dramatically entered the geologicalscene with the launch of sensors aboard continuouslyorbiting, unmanned satellites in the 1970s. The techno-logy had been available for 10 or more years, but locked

data into image form, applying digitaldraws us to

of between 1 anding on the level of detail.at the reconnaissance levelfrom U5$0.7-5 per squareefficiencies between 50 andper day I depending on whether

imageryimages. Although I must emphasizeing is no substitute for hitting rocks

spectral attributes of surface materialsfield -

fusion through GIS methods brings everylever to bear on resolving problems and grasping novel

opportunities.

Steve DruryCumbria, England

variate analysis techniques opens new vistas. Not onlycan the data be revitalized in later programmes, but GIStechniques permit a kind of information fusion that drawstogether many lines of evidence in solving problems orgenerating new ideas. Full and often unsuspected valuecan be squeezed from the records. Moreover, the fusionalsb helps bring the combined experience and differentskills of several people to bear on problemsi not any easytask otherwise, as many a manager will verify. This newand growing approach to information forms the topicof Chapter 8.

No course on geological remote sensing would becomplete without an explicit guide to its practicalapplications. Ideally, the best option would be to useactual case histories, but the problem is to precis thebest while extracting useful generalizations within setpage limits. In Chapter 9, I have chosen to let the expertsspeak fully for themselves by referring to important casestudies, and concentrate on some general principles ofstrategy and tactics. This covers several topics: landformstudies, geological mapping, hydrocarbon, metal andwater exploration, engineering applications and studiesof natural hazards. They are set in the framework of con-ventional field surveys, but show how remote sensingcan provide vital clues at critical stages.

Remote sensing aided by digital image processingis aesthetically pleasing, because all of us like colourfulpictures. If we did not then it would be a pretty sterile

NOTE

Some technical terms and concepts appear in the

Glossary.

and commented from the software designers' standpointon first drafts of the Exercises. Margaret was the first totest them as a beginner in remote sensing, and pointedout shortcomings of the early drafts. Most of the exercisesuse Landsat TM data of the newly independent stateof Eritrea, where I have been working for 15 years on avariety of practical and research topics. Some of thosedata were purchased by Jim Howard (formerly of theTechnical Department, Oxfam) and Martine Bilanou(forinerly Eritrean field director of the Eritrean Inter-Agency Consortium). The stream-sediment geochemicaldata of part of Scotland, used in Exercise II, are fromthe G-BASE archive of the British Geological Survey,courtesy of its Director and thanks to help from JanePlant and Phil Green of BGS.

Thanks are due to Ian Francis, Sarah Shannon andDave Frost at Blackwell Science for their work in produc-ing this edition of the book, and to Harry Langford, BillHouston and Philip Aslett for their careful checking ofthe text at various stages in the process.

I am grateful to the following individuals andorganizations who contributed or gave permission forthe reproduction of figures:

Figure 1.5(a) is reproduced from S. Valley, Handbookof Geophysics and Space Environments, 1965, McGraw-HillBook Company; John Wiley & Sons Inc. (1.6, 1.8, 1.12,1.13, 2.ge, 3.10, 3.14b and c, 3.22b, 3.25, 5.32, 6.1, 8.4);Gordon and Breach Science Publishers (1.7); EconomicGeology Publishing Co., University of Texas 0.9, 1.10,1.16, 1.17, Plate 3.2); G. Bell and Sons Ltd. (2.3, 2.15);Fig. 2.9(d) reproduced with permission from Photo-grammetric Engineering and Remote Sensing, 49, 3, Daily,p.352 @1983, by the American Society for Photogram-metry and Sensing; Aeroffims Ltd (3.3,3.12, Plate 3.1);W.H. Freeman & Co. @1978 (3.7); J.P. Ford, Jet Pro-pulsion Laboratory, California Institute of Technology(JPL), Pasadena (3.20, 3.27, 4.16a, 7.19, 7.20); H.C.MacDonald, the University of Arkansas (3.22); StuartMarsh, Sun Oil, Houston (Plate 3.2); Ronald Blom, JPL,Pasadena (7.18, 7.26, Plates 7.3 & 7.4); NOAA NESDIS,Washington DC (3.29); European Space Agency (3.34a);National Space Development Agency of Japan (3.34b);National Space Science Data Center/World DataCenter A for Rockets and Satellites, NASA (7.4-7.17a);Fig. 4.1 reproduced from V.C. Miller, Photogeology,1961,McGraw-Hill Book Company; US Department of Agri-culture, Agricultural Stabilization and ConservationService (4.2, 4.38,4.44a, 4.46,4.47b, 4.50); Figs 4.5,4.16(c),4.43, 4.48, 4.51 and 4.52 reproduced with permissionof the Department of Energy, Mines and Resources,

Like many geologists, I stumbled into remote sensing outof necessity, in my case by a need in 1977 for an idea ofthe re~onal structure of the Precambrian craton of SouthIndia. I had no access to aerial photographs or large-scale maps of the region. Without Landsat-2 an ambi-tious research programme on crustal evolution wouldhave foundered. At about that time Dave Rothery andMike Baker began postgraduate research at the OpenUniversity, which involved using Landsat data in theOman and the Andes, but had to use image-processingfacilities behind the chain-link fence of a secure militaryestablishment, The late Ian: Gass, then Chairman of EarthSciences at. the Open University, saw our plight and,typically, more long-term opportunities. His fight forand winning of funds for image-processing facilities laidthe groundwork from' which this book emerged. Ideasabout how to use digital image processing do not fallfrom trees. The'spin' of Image Interpretation in Geologyemerged from shared mistakes, exchanges of ideasand differences of opinion, first with Dave and Mike andlater with Gavin Hunt, Barros Silva, Beto de Souza andMargaret Andrews, all research students at the OpenUniversity from the 1970s to now. We have been lucky,sharing excellent facilities and funds to get us into thefield and to purchase data. Just as important to me havebeen opportunities to teach geologists the 'black arts' ina number of less well-favoured countries-Syria, Egypt,Arabia, India and Eritrea-which brought home theimportance of clarity and demystifying jargon. Thatexperience depended on being invited by the Food andAgriculture Organization of the UN, The US GeologicalSurvey Mission in Jiddah, Oxfam (UK) and the EritreanInter-Agency Consortium to run courses in geologicalremote sensing, and on the trainees, who hopefully didlearn something practical.

A book on image interpretation depends as much onits pictures as on its text. A fair proportion stem frommy work, but the majority, including some of the mostpleasing colour images I have ever seen, were providedfrom their collections by many colleagues around theworld. In this edition, many of the new images owe alot to resourceful use of image processing by Beto andMargaret. Staff in the Open University photographic andprint workshops printed many of the halftone images,and Andrew Lloyd and John Taylor of Earth Sciencesdrew several of the line figures. The major addition tothe book is the set of exercises on the CD-ROM thatuse TNTlite. Terry Peterson and Randy Smith of Micro-Images Inc (the producers of TNT software) donated aprofessional version of TNTmips, helped create the CD

x

~

Canada Copyright Her Majesty the Queen in Rightof Canada; Clyde Surveys Ltd. (4.6a, 4.11b, 4.12b,4.31a, 4.45a); National Cartographic Information Center(NCIC), US Geological Survey (4.6b, 4.7b, 4.11a, 4.26,4.35c, 4.38a); US Geological Survey (USGS) (4.7a, 4.8,

4.10a, 4.15a, 4.18, 4.26, 4.27, 4.35c, 4.38a, 4.45b, 4.47,4.49);Fig. 4.10(b) reproduced by permission of the Directorof Planning, Department of Planning, West YorkshireMetropolitan County Council; Figs 4.12 and 4.24(a)reproduced by courtesy of Amoco Inc.; Federal Ministryof Works, Nigeria (4.14b); Department of NationalDevelopment and Energy, Australia (4.14c); I.G. Gass,Open University (4.15); University of Illinois AerialPhotography and Remote Sensing Committee (4.20);Michael Smallwood, Optronix Inc. (Plate 4.2); DaveRothery, Open University (Plate 4.3, 4.41b, 5.33);P.W. Francis, Open University (4.31b); National Oceanicand Atmospheric AdIninistration, Washington (4.31c);Institut Geographique National, Paris (4.35); Figs 4.36,4.38(b), 4.40, 4.41 and 4.42 reproduced by permissionof the Direcci6n General de Fabricaciones Militares ofInstituto Geografico Militar, Chile; Pat Chavez, USGS(5.18, 5.20); Gary Raines, USGS (Plate 5.7); Environ-mental Research Institute of Michigan, Anne Arbor

~

Kahle, JPL,Haydn (Plate 6.2) Fig. 6.15Rowan, Geology, 8, 1980 FigsThe Geological Society of America;Pasadena (Plates 6.5 & 9.6); InteraCalgary (7.5, 7.23); Earl Hajic,Santa Barbara, and Goodyear(7.22); Gerald Schaber, USGS, ArizonaFarr, JPL, Pasadena (7.25); AmericanPetroleum Geologists (Plate 7.3); Earth

Remote-Sensing Laboratory, WashingtonSt Louis (Plate 8.2b); Charles Trautwein,Center (8.16, 8.17); Stanley Aronoff (8.2,

.-

of Oceanographic SciencesrUK (9.1); }PL,9.3, --

Sensing (Plates 9.3 & 9.4); Director of thelogical Survey (Plates 8.3 & 9.6, geochemical data inTNllite Exercise 11); Eosat Inc. (Plate 9.8); Trish Wright,Marconi Research Centre (Plate 9.9a); Falk Amelung,Stanford University (Plate. 9..9b); Didier Massbnnet,CNES, Toulouse (Plate 9.9c); Adobe Systems Inc. (Plate2.1).

Electromagnetic Radiation andMaterials

I

I

I

]

electricfield~ r.'

I ,./

distance ,magnetic ( ~ ~

field '. '\ jJJ

The way in which electromagnetic radiation (radiationfor short) is generated, propagated and modified is thekey to designing remote-sensing systems, their deploy-ment and understanding the images that they produce.The physical inseparability of radiation and other formsof energy from matter means that when matter andradiation interact, both are inevitably modified in someway. To account for these transformations fully involvescomplex mathematics with which most readers will notbe familiar. For this reason, Chapter 1 takes a simplifiedlook at the phenomena involved. For those readers withthe background and inclination, the Further ReacJjng forChapter 1 provides deeper theoretical insights.

1.1 The nature of electromagneticradiation

You do not need to know a great deal about the physicsof generation and propagation of radiation to begininterpreting remotely sensed data. Light and all formsof radiation behave both as waves and as particles. Theyhave the form of linked electric and magnetic forcefields, transmitted in packets (quanta or photons), which

~ '" ,have zero mass at rest. A particle of matter, such as anelectron, displays wavebehavi~ur under certain condi-tions. Pure energy and pure matter are totally abstractconcepts. Mass and energy are inseparable and are relatedthrough Einstein's famous relationship:

E=mc2

in nuclear fission and fusion, and by the transforma-tion of energy into subatomic particies in high-energyparticle accelerators.

More familiar waves, such as sound, ripples on waterand those seen moving along a shaken rope, are propag-ated by the motion of particles of matter. The motion maybe back and forth along the travel direction (longitudinalsound waves), in a circulating manner (orbital motion iI;1water waves) or perpendicular to the travel direction(transverse waves in a shaken rope). Those associatedwith radiation are transverse waves, and involve vibra-tions at right angles to their direction of travel. However,radiation can traverThrough a vacuum, and althoughit can affect particles in a physical medium by changingtheir electronic, vibr,a~nal and rotational properties,it must pro~gate itself in the absence of matter. Eachquanhlm has associa1ed electric and magnetic fieldsthat oscillate as sine waves at right angles to each ot~erand

to the direction of propagation (Fig. 1.1). The dis-tance between wave crests is the wavelength (?;.), andthe

number of vibrations passing a point in 1 s is thefrequency

(v). Wavelength is expressed in metres withthe prefixes pico (pm = 10-12 m), nano (nm =1Q-9 m), micro

(~m = 10-6 m) and the rest are familiar. Frequency isexpressed

as herz (Hz or per s~cond) with the prefixesmega (MHz = 106 Hz) and kilo (kHz = 103 Hz). Knowingone

enables the other to be calculated from the equation:A.V=

c (1.1)

The electric and magnetic vibrations associated with aquantum can be in any orientation at right angles to thedirection of propagation. However, if the fields for allquanta

are lined up in one direction by some means, theradiation becomes plane-polarized-a familiar concept

where E is energy, m is mass and c is the velocity ofradiation in a vacuum. This fundamental relationshiphas been demonstrated in practice by the net loss of mass

~

~/f1-

~~~1

sPeed oflight

Fig. 1.1 Electromagnetic radiation(EMR) comprises waves in electricaland magnetic fields. These fields areat right angles to each other and tothe direction of propagation of thewaves. The waves represent regularfluctuations in the fields and aredescribed by sine functions. Thedistance occupied by a completecycle from peak to peak is thewavelength (A), and the number ofcycles passing a fixed point in asecond is the frequency of theradiation (v). As EMR has a constantvelocity in a vacuum (c) it is easy tovisualize the relationship between A,v and c shown in Equation 0.1).

--

2 Chapter.. 1

gives the radiant flux per unit[.olid angle, which is calledradiance. The units of radiance are W m-2 sr-I (watts persquare metre per steradian, where steradian is the unitof solid angle).

Sometimes it is useful to consider the quantity ofradi-ation measured at a specific wavelength only. For examplethe spectral radiant flux is the power received or radiatedby a body per unit area per unit wavelength. For con-venience, this is often quoted in W m-2 ~-I. Similarlyspectral radiance is measured in W m-2 srI ~-I.

Although it is more correct to quote the amount ofradiation coming from a surface as radiance or spectralradiance, when writing or speaking informally or non-numerically the more familiar term brightness is oftenused instead. This term can refer either to the amount ofradiation coming from the surface or to the appearanceof a region within an image. For example, if a particulararea in an image is said to be brighter-than another area,it is cl~ar what we n)ean, even though we may not be ableto quantify the difference in terms of radiance units.

The generation of radiation is essentially a simple pro-cess. It is produced Whenever the s~e or direction ofan eleciPc or magnetic field fluctuates with time. Radiowaves can be produced by the flow of rapidly alternat-ing currents in a conducting body or antenna. Thealternation is, in effect, the repeated accele~ation anddeceleration of electrons. At the shortest wavelengths,gamma-rays result from disruption of atomic nuclei dur-ing nuclear fission or fusion reactions. X-rays, ultravioletradiation and visible radiation are.generated by electronsjumping from one stable orbiting shell around an atomto another. When an electron moves from a higher orbitto a lower one, the energy which it loses is converted intoa photon of a specific wavelength. Infrared and microwaveradiation is produced by thermally induced vibration androtation of molecules. Microwaves are also generated byfluctuations in electric and magnetic fields.

Wavelengths of electromagnetic radiation span manyorders of magnitude, from shorter than 1~~m~~~~8,-a,ill!!!~::ray~to longer than 100~!.~rveryl~4iQwayes. Convenience demands a divisionof this vast range into several arbitrary regions, eachwith its own name. Figure 1.2 shows this spectrum ofwavelengths and conventional names (with further sub-division of the near-infrared shown later in Fig. 1.5(b).

In nature all processes that generate radiation arerelated in some way to the temperature of the bodyemitting it. All matter in the Universe, even that in thenear-perfect vaCUum between galaxies, is above absolutezero (-273.15°C) and emits some form of radiation. Justhow much is emitted and the range of its wavelen~hsis a complex function both of temperature and thenature of the body itself. Matter capable of absorbingall electtomagnetic energy that it receives and emittingradiation perfectly according to its temperature is known

for any geologist who has used a polarizing microscopefor petrography.

The frequency, or wavelength of radiation is a func-tion of the energy of the quanta. Max Planck formulatedthis relationship (Planck's Law) as:

E = vh = ch/L (1.2)

where h is Planck's constant (6.62 x 10-34 J s). Equation(1.2) means that the shorter the wavelength or the higherthe frequency of radiation, the greater the energy of eachquantum. As the most important radiation-matter inter-actions take place at the quantum level, Planck's Law isimportant in understanding them (Section 1.2).

The final basic property of radiation is its int~ty,equivalent to the brightness of visible light. This may beregarded as either the nu~er of quanta or the ampli-tudes of the electric and magnetic fields. The more q~antaat a particular wavelength, the greater the energy thatis transI;nitted. The, energy of a single long-wavelengthquantum is less than tn.a"t of 9fie at short wavelength.Conseq~nt1y; more long-wavelength quanta must fall ona detector to produce a measurable response comparedwith the number of shorter wavelength quanta that pro-duce the same response. In general, therefore, systemsaimed at long-wavelength radiation need to collect radi-ation either from a larger target area or over a longertime than those used for shorter wavelengths. This hasimportant consequences for the resolution of remote-sensing systems and their ability to discriminate realobjects from syst~matic noise. In reality, things are a gooddeal more complicated than this, because instruments usedifferent kinds of detectors at different wavelengths.

""" 1.2 The generation of electromagneticradiation

As electromagnetic radiation crops up frequently inremote sensing literature,' this is a convenient point atwhich to introduce some of the terminology that isused for radiation, Electromagnetic radiation is a formof energy, and the amount of radiation per unit time(power) is measureg in units of JS-1 (joules per second),or W (watts), The power incident on or emanating froma body is known as the radiant flux, but it is usuallymore useful to express it as the power per unit area-theradiant flux density (W m-2), The radiant flux densityfalling on a sul:face is known as the irradi~nce, whereasthat leaving a surface is called the emittance (sometimescalled the exitance),

Limitations to the size of measuring devices mean thatwe can rarely measure directly all the radiation leavinga surfa~e, Instead, what is measured is the amount ofradiation intercepted by a small detector, which collectsradiation travelling through a given solid angle, This

Electromagnetic Radiation and Materials 3

rain and fogattenuation..scattering--

different parts of the spectrum (see Fig.1.Sh). Those portionscovered by this book are highlighted, together With theprocesses relevant to geological remote sensing. The na'rrowvisible band is useful as a reference. ,The wavelength andfrequency scales are logarithmic,

Fig. 1.2 This surtunary of that part of the electromagnetic(EM) spectrum which is detected routinely by scientists showsthe relationship between wavelength and frequency, thephenomena that are involved in generation and interaction ofele<;tromagnetic radiation (EMR), and the nomenclature for

asa blackbody. The total energy emitted by a blackbody-its emittance (H) in W m-2-isproportional to the fourthpower of its absolute temperature (T in Kelvin or K).Thisis the Stefan~Boltzmann Law:

H = aT4 (13)

visible energy

/

..Sun'ssurface.-,; temperature

~

IE~

~~QJuCta

+"'+"'

°EQJ+"'Cta

::c~"(ij

tQJa.

V\

~

incandescent lamptemperature

Earth's surfacetemperature

where ais the Stefan-Boltzmann constant (5,7 x 10-8 Wm-2 K-4).

At any particular temperature, a blackbody emits radi-ation with a range of wavelengths. However, its absolutetemperature determines which wavelength transmitsthe maximum amount of energy. This dominant wave-length ().,m in p,m) is given by Wien's Displacement Law:

).,m = 2898/T (1.4)

SoJ as temperature incr{!ases, total energy emitted risesvery rapidly and the wavelength carrying most energybecomes shorter. The shape of the curve relating emit-tance to wavelength is important (Fig. 1.3), and stemsfrom both the Stefan-Boltzmann and Wien's laws. Forany tem~~~e!s aminimumwav:el~~<:ii-ation,. a nearby wavelength ofmaxim~~~ti~I;;~_~ndalon~tlr~~JQ!!~~~llit1hs. Th~s a bla(:kbod!afOOOOK-,-the Sun's surfacetemperature-does not emItradiation with wavelengths shorter than 0.1 p.m, has anenergy peak at 0.5 p,m (in the part of the spectrum that isvisible to us as green), butemits all wavelengths beyondthat up to about 100 p.m. The total energy emitted isgiven by the areas beneath the curves in Fig. 1.3.

1 I0.1 0.2 0.5 1 2 6 10 20 50 100

Wavelength (\Jm) '""?'-

Fig. 1.3 This family of curves on logarithmic axes expresseshow energy emitted by a square metre of a blackbody atdifferent temperatures varies with wavelength, and howthe wavelength of maximum emittance and the range ofwavelengths emitted change with absolute temperature.The area und~r each curve represents the total energy emittedat each temperature. Both the Stefan-Boltzmann a~d Wienlaws control the shapes.

4 Chapter 1

No natural objeet is a perfect blackbody. In the case ofthe Sun-the source of most radiation exploited in remotesensing-many processes other than simple heating areinvolved. Consequently, the solar irradiation curve(see Fig. 1.5) is a little different from the ideal case. Aswell as radiation in the 0.1-100 ~m range, the Sun emitsgamma-rays resulting from thermonuclear processes andlong-wavelength radiation resulting from fluctuationsin its predigious magnetic and electric fields.

Remote sensing is concerned with two categories ofradiation from the Earth's surface-that which falls on itand is absorbed or reflected, and that which is emittedby the surface itself. Reflected radiation derives in the--main from the Sun, and systems detecting it are termedp.assive because no artificially induced energy is required.An active system involves artificial 'illumination', as inflash ph«;>tography. In rem«;>te sensing the mo~t widelyused active systems utilize radar (r~dio ~etection andranging) transmissions and detection of the radarenergy reflected back to a se11$or by the Earth's surface.Experi~ents have demonstrated thCit other active sys-tems utilizing artificial radiatipn, usually in the form ofultraviolet lasers,ca~ produce data for.1imited ranges ofapplications.

Because ~e Earth's ambient temperature is about300 K, Wien'sDisplacement Law implies that it has amaximum emittance at 9.7 ~, in the mid-infrared (MIR)range.. The energy involved in producing ,this emittedradiation derives from three sources: the flow of radio-genic heat from the Earth's interior, the heating of thesurface by solar radiation, and human activities. Long-wavelength infrared is not the only radiation emiliedby, the Earth. All rocks, and the materials derived fromthem, contain variable proportions of the unstable iso-topes 4OK, 232Th, 2350 and 2380, which emit gamma-rayswhen they decay. These too can be detected remotely,and add to the range of true remote sensing techniques.

1.3 Matter and electromagnetic radiation

The ~ey to understanding how remotely sensed datahelps us recognize different materials at .the Earth's sur-face lies in the way radiation interacts with matter. Todiscuss this exhaustively means examining the inter-actions at the molecular or atomic level, which involvesErwin Schrodinger's wave mechanics (covered by someof the Further Reading). Only a very simplified accountis given here.

For a single chemical element there are several pos-sible states in which it can exist, E:'ach of which has acharacteristic energy level. Such states involve the typesof bonds (covalent or ionic) and the coordination stateof atoms within molecules, the energy level of an atom'soutermost orbital electrons and much more besides.

The range of states and tqe associated energy levels areunique for every element and compound. An atom ormolecule may pass through a transition from one state toanother if it is excited by just the right frequency of radi-ation. A simple example is fluorescence when radiationof one frequency is absorbed to cause a transition, andreversal of the transition emits lower frequency radi-ation. There are three basic types of transition.-.electronic,vibrational and rotational.

Electronic transitions involve shifts of electrons inthe outermost orbitals in an atom that impart its ele-ment's valency and many of its chemical properties. Suchtransitions are the reverse of one means of generatingradiation. A photon of a specific wavelength inducesan outer electron to move from an orbital at the lowestenergy level permitted by wave mecharilcs (its groundstate) to one with a higher energy level (its excitedstate), thereby absorbing the photon. The wavelengthsassociated with electronic transitions are determinedby the principal quantu~ numbe~angular momentumand spin associated with electron orbitals within a par"-ticular element. Elec~nic tran§!tions occur in solids,liquids,and gases, but are particularly important for ele-ments such as iron and chromium, which have severalpossible valence states, and various p!;!§itions and co-ordinations in naturally occurring molecules. These dif-ferences account for subtl~. changes in the wavelengthsof electronic transitions depending on the element's host.As electronic transitions require high excitation energies,they are most common at the short wavelengths of ultra-violet and visible light.

Vibrational transitions result in changes in the relat-ive disposition of the component atoms of molecules.The most easy to visualize are distortions of bonds,either stretching or bending from one equilibrium stateto another. Analogous to sound, as well as a funda-mental wavelength or 'note' associated with a vibrationaltransition, there are mathematically related harmonicsor overtones at other wavelengths. Like electronic trans-itions, those associated with vibrations of mol~ar bondscharacterize solids, liquids and gases. They require lowerenergies than electronic transitions, however, and sooccur with longer wavelength radiation in the infrared

region.Transitions also may occur in the rotational proper-

ties of molecules, but they are restricted to gases. Theyrel"te to changes in the moment of inertia of the rotat-ing molecules of gas. Rotational transitions are of mostimportance, together with vibrational transitions, in theinterac~on between radiation and the atmospheric gasesthrough which the Earth's surface must be viewed by allremote-sensing systems (Section 1.3.1).

The energy detected by remote-sensing systems overthe spectrum of radiation is therefore a function of howenergy is partitioned between its source and the materials

E lectroma gnetic

(1.6)

(P..), absorptance (a..) iind transmittance ('t..),iveiy, so that:

(Ex(EJ..+(EA/EJ.. + {Ei/EJx = 1

i.e.

()

P" + a" + 'f" = 1 (1.7)

The vast bulk of geological materials are opaque andtransmittance is zero, so that Equation (1.7) reduces to:

p,,+a,,=l (i.~)which means in effect that reflectance and absorptanceare interchangeable (generally the adjective spectral isomitted from both terms), but reflectance spectra arenearly always used. The ratio of the total radiant fluxreflected by a surface to the total radiant flux incidenton it (in both cases spanning a range of wavelengths)is known as the albedo of the surface, Although it isnot exactly the same, we .perceive albedo as the overallvisible brightness of a reflective object.

The value of reflectance for a surface specifies theproportion of the incident energy that is reflected ata specified wavelength, but not the direction in whichthe reflected energy travels. This depends on whetherthe surface produces specular reflection like a mirror; ordiffuse reflection like a sheet of matt paper. In specularreflection, all the reflected energy is directed away atan angle equal and opposite to the angle of incidence. .

In diffuse reflection, the Teflected energy is directednearly equally in all directions, irrespective of the angleoj incidence (Fig. 1.4). A perfect diffuse reflector is oftencalled a Lambertian reflector. Many surfaces combine~pecular and Lambertian properties, in that they reflectsome energy in all directions, but reflect a larger propor-tion in the specular direction (Fig. l.4c).

A surface behaves as a specular reflector if it is smooth,and as a diffuse reflector ifit is rough. Smoothness androughness depend on the wavelength of radiation. Gen-erally, a surface behaves as rough if its textur~ is on ascale comparable wjth,or greater than that of the wave-length of the radia~n, and in a smooth fashion if itstexture is on a finer scale t~a~thewavelength. Most sur-faces, such as rock, soil ol'grass, are diffuse reflectors in

with which it interacts on its way to the detector. Theenergy of any particulaf wavelength of radiation maybe transmitted through the material, absorbed within it,reflected by its surface, scattered by its constituent par-ticles, or reradiated at another wavelength after absorp-tion. In nature, all possibilities combine together in onedegree or another. .

Three kinds of spectra can be measured for any material-absorption (and its inv~rse, transmission), reflectionand emis&ion spectra. An absorption/tr,ansmissionspectrum is produced when the material is interposedbetween sou~e and sensor, A reflection spectrum ismeasurec;i when both source and sensor are on thesame side of the material. For an emission spectrumthe material itself is the source. In each set-up a prismor diffraction grating spreads the composite radiationout into its component wavelengths, when intensitiesat discrete wavelen.;;ths, measured by a variety of detec-tors, may be related to specific emission or absorptionprocesses. This is the technique used by astronomers todetect and measure element abundances in stars fromatomic absorption bands in stellar spectra. The remotesensor, however, is more concerned with continuousspectra, which show the variation in energy/intensityover a range of wavelengths. Such spectra are more orless smooth curves in which peaks and troughs indicatemaxima and minima around wavelengths that corres-pond to some characteristic transition. Many macroscopicand microscopic factors conspire together to determinethe width, strength and abruptness of these features,some of which are discussed later in this section and inother chapters.

The principle of conservation of energy determinesthat for any radiation-matter interaction, the inci~entradiant flux at one wavelength (EJA is distributed betweenreflection (ER}A' absorption (E A}A and transmission (ET}Aby the material involved

(EJA = (ER}A + (EA}A + (ET}A (1.5)J

Dividing Equation (1.5) throughout by (EJA producesan expression allowing the spectral properties of thematerial to be defined in terms of the ratios (ER/Ei}A(E A/ EJA and (ET/ EI)').f which are the spectral reflectance

\ \'\ \.\

Fig. 1.4 Schematic diagramsshowing (a) specular, (b) diffuseor Lambertian and (c) reflectioninvolving both specular and diffusecomponents. (a) (b) (c)

6 Chapter 1

the visible spectrum-they look equally bright in what-ever direction they are viewed, even though tiny partsof a surface (individual mineral crystals, for example)behave specularly.

(IJm) Wavelength (cm)

Fig. 1.5 Various gases in the atmosphere absorb solar energyin different wavebands by vibrational and rotationaltransitions: As a result the solar irradiation curves measuredin outer space-upper curve in (a)-and at the surface-lowercurve in (a)-are very different. The energy available forinteractions with matter at the surface is divided into discreteatmospheric windows separat~d by bands dominated byatmospheric absorption (grey). In (b) the main atmosphericwindows throughout the whole of the useful part of theelectromagnetic (EM) spectrum are shown on a logarithmicscale, in terms of the percentage transmitted through theatmosphere. These two graphs, together with the spectralproperties of natural materials, form the basis for designingremote-sensing systems.

matter in the~tmosphere. The type of scattering changeswith t;he size of the particles responsible. Where radi-ation interacts with particles smaller1han its wavelength,such as molecules of oxygen and nitrogen, the degree ofscattering is in.versely proportional to the fourth powerof the wavelength. This is known as Rayleigh scatteringafter its discoverer, Lord Rayleigh. The relationship means

1.3.1

The effect of the atmosphereRemote

sensing of bodies such as the Jovian moon 10 orthe planet Mars is a geologist's delight. Both have very thin

almost transparent atmospheres, except when volcaniceruptions in the case of 10 or sand storms on Mars takeplace. Virtually the whole of the radiation spectrum isavailable for surveillance, given a suitable range of sensors.For the Earth, however, all radiation must pass througha dense atmosphere. Before reception by a satellite-mounted sensor, solar radiation must pass down throughthe atmosphere and tbenback again to the ~ensor. Fora sensor measuring radiation emitted by the Earth thepath is single, but there is some effect, nevertheless.

As well as nitrogen and oxygen, the atmosphere con-tains significant amounts of water vapour, ozone (°3)'carbon dioxide and traces of other gases. All interactwith radiation by vibrational and rotational transitions,the net effect of which is absorption of energy in specificwavebands (Fig. 1.5a), The absorption of short-wave solarratliation is one process that heats the at~osphere. Thegrowth in industrially emitted CO2 in the atmosphere isthe source of-the so-called 'greenhouse effect', which issomewhat different. Carbon dioxide's principal effect inthis regard is at the longer infrared wavelengths domin-ated by the Earth's thermally emitted radiation (Fig. 1.3).Outgoing thermal radiation is absorbed by CO2 andstored temp~rarily before ultimate re-emission to space.This process is dominated by much more abundant watervapour, but CO2 is currently increasing so giving rise tofears about global warming. Methane and ozone have asimilar effect. This 'lag' in heat loss results in the atmo-sphere warming above the temperature it would attainwithout absorbers of thermal radiation.

At short wavelengths atmospheric absorption bandsare narrow, but increase in width in the infrared andmicrowave regions. Figure 1.5(b) shows that about 5Q%of the radiation spectrum is unus~le for remote sensingof the Earth's surface, simply because none of the cor-responding energy can penetrate the atmosphere. In thecase of emitted gamma-rays, only by flying close to theground ~an some energy be detected. It is possible torecord absorbed wavelengths relating to gases, but thisis for atmospheric studies that are beyond the scopeof this book.

Another irritation to the remote sensor is blue sky,strange as it might seem. Obvious when we look upwardon a clear day, exactly the same 'tinting' is presentlooking downwards. It is caused by one of a number ofphenomena resulting from the scattering of radiation by

The other constraint-on system design, and the mostimportant factor in remote-sensing strategy, is the inter-action between radiation and those solids and liquidsthat constitute the Earth's surface. There are only threeimportant components: water, vegetation and the min-erals making up rock and soil. For the geologist, theinteractions between radiation and rocks and soils arecritical. However, because they can contain water or havevegetation growing upon them, these materials mustbe considered too. The next three sections introducesome of these surface interactions, and they are expandedupon in later chapters.

that the effects of Rayleigh scattering increase dramatic-ally at short wavelengthSr--hence blue sky and distant bluemountains. The effect on a view from above the Earth'ssurface is to swamp the real reflected blue and ultra-violet radiation with a high scattered component, andto reduce the contrast.

Where atmospheric particles are similar in size to thewavelength of incoming radiation, as is the case for thegiant molecules of water and for dust, Mie scatteringresults. This affects wavelengths longer than that of bluelight, and is a problem in clear but humid or dusty atmo-spheric conditions. Red sunsets are attributed to the Miescattering effect of very fine dust blown up from desertsor microscopic ash particles and acidic water dropl.etsinjected into the atmosphere by volcanic eruptions.Aerosol droplets in cloud or fog, which are much largerthan most radiation wavelengths of interesL in remotesensing, scatter all wavelengths in the visible and infraredspectra. Aerosols are impenetrable, except by radiationbeyond 100 ~m wavelength-microwave and radar. Evenat such long wavelengths, heavy rain or snowfall cancause nonselective scattering of this kind; and as a resultcan be detected and even measured.

On a clear night stars appear to shift and wink, as dodistant objects on a hot day. These distortions are pro-duced by temperature variations in the air, which producefluctuations in its refractive index, and a range of opticalanomalies. These same effects are also present whenthe Earth is viewed from above. Atmospheric shimmerforms an important constraint on just how small an objectremote sensing can detect, irrespective of the theoreticalresolving power of each system.

All this implies that remotely sensed ima'ges of theEarth are unavoidably degraded in various ways by theatmosphere. Selective atmospheric absorption also meansthat only some wavebands are available for surveil-lance (Fig. 1.5b). Those that pass relatively undiminishedthrough the air present atmospheric windows, anddetermine the framework wherein different methods ofremote sensing can be devised. This is an appropriatepoint at which to introduce a broad division of thespectrum according to the source of the measured radi-

." ation in remote sensing. Figure 1.5(a) shows that radiantenergy from the Sun falls to very low values beyondabout 2.5 Iln\. Of that in the 0.4-2.5 ~m region a goodproportion is reflected from the surface, depending onthe material, so allowing remote sensing of the propertiesof the reflected radiation. This is the reflected region. Thetwo windows between 3 and 5 ~m and between 8 and 14~m are dominated by radiant energy that is emitted bysurfaces heated by the Sun. This is the emitted region.The transparent region beyond 1 mm is the microwaveregion. Quite different technologies are needed to acquireremotely sensed data in these three regions, and they areoutlined in Chapter 3.

1.3.2 Interaction of electromagnetic radiation withrocks and minerals

In this section only the most general features are covered,interactions specific to particular rock and soil types beingdealt with in later chapters. Here the effects of commonrock- and soil-forming minerals have primary importance.They are considered for three major ranges of radiationwavelength, from 0.4 to 2.5 ~ (viSible VNIR and SWIR),8-14 ~ (emitted or thermal MIR) and 1 mm to 30 cm(microwaves). Radar interactions in the microwaveregion are discussed separately in Chapter 7, becausethey are fundamentally different from those of shorterwavelengths. The most important processes involvedare electronic and vibrational transitions (Section 1.3);only gases exhibit rotational transitions.

Rocks are assemblages of minerals, and so theirspectra are mixtures of those for each of their constitu-ents proportional to their abundance. Minerals in turncomprise various assemblages of different elements,held together as molecules by different kinds of bonds.Electronic transitions within atoms themselves requiremore energy than vibrational transitions within mole-cules, so the former characterize the short wavelength,ultraviolet to visible range, whereas the latter dominatethe longer wavelength SWIR. There is, however, someoverlap between the ranges of these two fundamental

processes.The most common ingredients of rocks and the

minerals that make them up are o~gen, si!i~on andaluminium, together with varying proportions of iron,magneSium, calcium, sodium and potassium, and smalleramounts of the other elements. Oxygen, silicon andaluminium atoms have electron orbitals in which energylevels are such that transitions between them have littleor no effect on the visible to near infrared range. Thespectra of minerals are dominated by the effects of lesscommon elements and the molecular structures in whichthey are bonded. The characteristic energy levels ofisolated elements are changed when they are combinedin minerals because of the valence states of their ions,the type of bonding and their relationships to other ions

8 Chapter 1

i~QIut:ratQI

'+=QIa:

i~

~QJuc:

tQJ

'+=QJ~

0.40.50.6 0.7 0.80.91.0 1.1 1.2 1.3

Wavelength (IJm)

Fig. 1.7 Iron oxides and hydroxides display absorptionfeatures resulting from Fe-O charge transfer and crystal-fieldeffects in their reflectance spectra (T -shaped symbols).Substitution of iron in clay minerals superimposes similarfeatures on clay spectra. The spectra are offset for clarity.

O.b " 1.0 1.5 l.O 2.5

Wavelength (~m)

Fig. 1.6 The number and position of features resulting fromelectronic tr~itions in iron min~r!lls (T -shaped symbols)depend 01) theco-ordination of FeZ" ions in the molecularstructures of the minerals concerned. The spectra are offsetvertically for clarity. In this figure and in Figs 1.7-1.10 and1.15-1.17 the vertical bands indicate the widths of spectralbands sensed by the Landsat Thematic Mapper (Chapter 3),the most widely used orbiting remote-sensing system.

spectra interfere. More important, they are rarely freshbut covered with thin veneers of weathering products.As visible and NIR radiation interacts only with the outerfew micrometers of the surface, spectra of fresh mineralsrarely affect the reflected radiation used in much remote

sensing.Another type of electronic transition results from the

presence in metal ions of electrons that have sufficientenergy that they are not strongly attached to any particu-lar ion and may transfer from one ion to another. This isthe property that imparts high electrical conductivity tometals. In a mineral a similar transition, called a charge-transfer, can occ1,lr. It too is induced by energy in nar-row wavebands of radiation, giving rise to absorptionfeatures. The most common charge-transfer is involvedin the migration of electrons from iron to oxygen, andresults in a broad absorption band at wavelengths shorterthan about 0.55~. It is common to all iron-bearing Y'

.-minerals and is responsible for a steep decline in reflect- ~ance towards the blue end of the spectrum. The mostnoticeable effect is with iron oxides and hydroxides(Fig. 1.7), and is the reason why these minerals and the~,rocks containing them are coloured yellow, orange, red :,.and brown. Such mineral~ form the main colourants inweathered rocks. They too display crystal-field absorp-tions, the most prominent being around 0.&--{).9 ~m. Asshown in Fig. 1.7, the location and shape of these bandsvary subtly from mineral to mineral and, as discussedin Chapter 5, form a means for discriminating betweenthese important minerals.

(their coordination). Because they can exist as ions withseveral different valel1ces, the transition metals iron,copper, nickel, chromium, cobalt, manganese, vanadium,titanium and scandium exhibit a great range of possibil-ities. As iron is by far the most abundant of these metals,its effects are the most common and the most obvious.

Figure 1.6 shows reflectance spectra of several iron-bearing minerals that display features resulting fromelectronic transitions in ferrous (Fe2+) ions. The differ-ent wavelengths of the features relate to the symmetry,degree of lattice distortion and coordination of Fe2+ inthe different minerals. The features are troughs, indicat-ing that absorption of energy takes place over the bandof wavelengths involved, in order to shift outer electronsto excited orbitals. All these features are the products oftransitions in discrete ions and their breadths result fromcrystal-field effects. Although there seems to be a widerange of possibilities for distinguishing the iron mineralsin Fig. 1.6, in reality they are only useful in a laboratorysetting using fresh minerals. In the fiel~ minerals areassembled in different proportions as rocks, so that their

0.6 1.0 1.5 2.0 2.5

Wavelength (~m)

Fig. 1.8 Minerals containing chemically bound watt;$ haveparticularly distinctive absorption features close to thetheoretical overtone wavelengths for the H-O-H bond-stretching transitions (T -shaped symbols). None of the featuresfor sulphates, such as gypsum, relate to SOi-ions. The spectraare offset for clarity.

1.0 1.5 2.0 2.5

Wavelength (IJm)

Fig. 1.9 The bending of AI-GH and Mg-GH bonds in clayminerals and micas produce disti!l$!iveabso~tion fe~tures intheir reflectanc~ spectra (T -shape,d symbols). Tog~her withother features of the spectra, they form a potentially powerfulmeans of discriminating these minerals,which are importantproducts of hydrothermal and sedimentary processes. Thespectra are offset for clarity .

~

rise to a number of absorption features in the SWIR ofwhich that around 2.3 ~m is most prominent (Fig. 1.10).That at 2.55 ~m lies outside the atmospheric window.

Although the reflected part of the spectrum has limitedpotential for rock discrimination, sufficient diagnostic

.features are present that great effort has been put intodevising methods of capturing data with sufficientspectral resolution to separate the different features. Thebroad wavebands from the Landsat Thematic Mapper(Chapter 3), highlighted in Figs 1.6-1.10, are capableof detecting but not discriminating between hydroxy-lated and carbonate minerals. The effectiveness of moresophisticated systems forms an important topic in laterchapters.

In that part of the spectrum where the Earth's emittedradiation reaches a peak, an atmospheric window between8 and 14 ~m allows radiation to be sensed remotely. Anideal blackbody, from which the total energy emittedand its distribution between different wavelengths

.are governed by the Stefan-Boltzmann and Wien laws(Section 1.2), has a distinctly shaped emission spectrum

In the SWIR part of the spectrum the most importantvibrational transitions in minerals are those associatedwith the presence of OH- ions or water molecules boundin the structure or that are present in fluid inclusions.The water molecule has three fundamental vibrationtransitions, owing to stretching of the H~H bond at 3.11and 2.90 ~m and bending at 6.08 ~m. As a result of over-tones and their combination, these produce absorptionfeatures at 1.9, 1.4, 1.14 and 0.94 ~m, which are diagnosticof the presence of molecular water in minerals (Fig. 1.8).These features, however, are completely swamped bythe nearly identical effects of water vapour in the atmo-sphere, and are only useful in a laboratory setting.

Many silicates and alteration minerals contain hydroxyl(OH-) ions in their molecular structure, for which thereis only one fundamental ~ond-stretching transition ofthe O-H bond, at 2.7 ~m. This may form overtones incombination with other transitions, the most importantof which are bond-bending transitions, involving distor-tion of the metal-hydroxyl bonds Mg-OH and AI-OH toproduce absorption features near 2.~ and 2.2 ~m, respect-ively. Such features are prominent in aluminous micas Iand clay minerals (Fig. 1.9) and dominate signatures of.hydroxylated minerals that contain magnesium, suchas talc, chlorites, serpentinesand magnesium-rich clays(saponites). Provided that the absorption features canbe res~lved, these spectral characteristics form a power-ful tool in discriminating between chemically differentrock types.

Similar vibrational transitions and overtones charac-terize carbonate minerals. They derive from stretchingand bending of the C -0 bond in the CO~- ion. They give

10 Chapter 1

quartz

i~QIut:ruuQI'i=QI~

three-

orthoclase'-.. '"" I

::,~""'~-ralbite

AI-O-Si

1.0 1.5 2.0 2.5

Wavelength (IJm)

Fig.l.lO Vibrational transitions related to C-o bottds produceabsorption features in the reflectance spectra of carbonates(T -shaped symbols). The most distinctive is that near 2.35 ~m,which potentially can distinguish between carbonates andclays. The spectra are offset for clarity.

rc:0

'ij;VI

"EVIc:ro.=

labradorite'",,---:-- ..

muscovitesheet

chain

.:::-:---~--rolivine isolatedtetrahedra

6 7 8 9 10 12 15 20 3040

Wavelength (IJm) Structure

Fig. 1.12 Because of differences in the structure of silicates,the positions of both the Si-O bond-stretching trough and its'shoulder' at shorter wavelengths (arrows) occur at slightlydifferent positions in the mid-infrared spectra of differentsilicate minerals. The spectra result from experiments usingtransmitted energy, but would appear very similar foremission.. The spectra are offset for clarity.

0 2 4 6 8 10 12 14 16

Wavelength (~m)

Fig. 1.11 Jhe emission spectrum for quartz, at 600 K, deviatesfrom that for a true blackbody becausce of a strong featureinduced by Si-Q bond stretching: Quartz therefore is as~lective radiator and not a greybody.

(Fig. .1.3) but no spectral features. A measure of a naturalmaterial's deviation from the ideal is its emissivity (eA)-the ratio between its radiant emittance for a particu-lar wavelen ~~) ata given temperature and that of ablackbo .A greybody~~~ emissiVity less

an 1.0 for all wavelengths; Most natural materials,however, have emissivities that vary with wavelength.They are selective radiators because vibrational tI:ans-itions of bonds in their molecular structure !!!!1?e.delemission at characteristic wavele_n.gih§,(QU-artZ-'is a good-6a:ii1pre:as1FIg~T.l1shows.

It radiates asa nearly idealblackbody up to 6 ~m, but deviates from the ideal atlonger wavelengths.

Because good emitters are also good absorbers of radi-ation, their emissivities are equal to their absorptivities.The latter is difficult to measure and this relationship,known as Kirchh..Qff's Law after its initiator, can be trans-formed using Equation (1.8) to:~

e" = I-p" (1.9)

e trough in the emittance curve for quartz between

and 9 ~m is the result of Si-O bond-stretching vibra-tions. This and related spectral structures are best shownby transmission spectra, and they occur in both silicates{Fig. 1.12) and nonsilicates(Fig.1.13). This is because mostminerals have chemical bonds with energies of vibrationthat fall in the thermal infrared region, and emission ishindered at the energy which coincides with each bond

"'~::J...raQjc.EQ)

...

...c:ra:cra

IX:

apatite

t:0

";;jVI

°eVIt:

IV

~ hematite

kaolinite

24.00 04.00 08.00 12.00 16.00 20.00midnight noon

Time (h)

Fig. 1.14 Materials with high thermal inertia, such as metals,show little range in diurnal temperature because they heat andcool slowly. Those with low thermal inertia, such as soils, areprone to rapid heating and cooling, and so they reach highdaytime and low nil!ht-time temDeratures.

24.00midnight

c ., I I '" I I

5 .': 6 7 8 9 10 12 15 20 3040

Wavelength (IJm)

Fig. 1.13 Spectra of nonsilicates in the mid-or thermal infraredpart of the spectrum show completely different patterns ofabsorption features from silicates. They are also very differentfrom one another, suggesting great potential for lithologicaldiscrimination by remote sensing of emitted thermal radiation.The spectra are offset for clarity.

;':' ~~~~~:'i-.;A"'H d'~T.r'~..C"~/(: {)r'

surface varies, heating to a maximum at the hottest partof the day and cooling by radiation to a minimum justbefore dawn (Fig. 1.14). The extremes and rates of thisvariation depend on a material's absorptance, trans-mittance and thermal capacity. These variables can beexpressed empirically by thermal inertia; a measure ofthe time-dependent response of the material to temper-ature changes (Chapter 6). A rock with a high thermalinertia heats up and cools stowly, so showing a low rangeof diurnal temperatures. Those with low thermal inertia~splay lar e fl ctuati()RS-during-th~c;y_cle. ~

At 3 K the Earth's surface emits radiation-.itw~-lengths in the microwave region as well as infrared, albeitat intensities that decrease to very low levels as wavelengthincreases. The atmosphere absorbs most of the energyin the thermal range from 14 ~xIl10 1 mm. In the ~cro-wave region, however, to which the atmosphere is trans-parent, intensities are still high enough to be measuredby passive remote-sensing systems. A signal detectedabove the surface will include components emitted bythe surface, emitted by the atmosphere and transmittedfrom below the surface. This last component is possiblebecause rocks and,soilshave much larger transmittancesat these wavelengths than they have in the visible andinfrared 'parts of the spectrum. Thus passive microwaveremote sensing is capable of providing informationabout buried materials as well as surface materials. Itspotential use in geology is expanded in Chapter 7.

Rock spectra are composites of those of their con-stituent minerals. Depending on the structure and com-position of these minerals, they may be detected if theyare sufficiently abundant and their spectral features arestrong enough. Discussion of rock spectra, their inter-pretation and uses is continued in later chapters.

vibration. The most important feature in the silicate fam-

ily of spectra (Fig. 1.12) is that the minimum of the main

absorption trough shifts according to the type of silicate

structure involved. So too does the peak at the short wave-

length edge of the absorption trough. A partial explana-

tion for this is that in silicates with different structures

the SiO4 tetrahedra share oxygens in different ways. The

advantage for the geologist is that the progressive shift

of the short wavelength peak and the main trough towards

longer wavelengths corresponds to a transition from

felsic to increasingly mafic minerals (Chapter 6).

In this region of the spectrum, various vibrational

transitions in nonsilicates produce spectral features that

are different from those of silicates (Fig. 1.1.3). The most

important are those associated with carbonates and iron

oxides, which are so distinct that even small amounts of

these nonsilicates in dominantly silicate rocks drastically

alter their spectra. Indeed, limestones and ironstones

should be easily distinguishable from each other, as well

as from silicate ro~~l.!!!tbiS1?~!tQithe-spectmm /:~~~~~gy emitted by rocks in the thermal

fufrared region is related to their temperature. By assum-

ing the Stefan-Boltzmann Law it is possible to calculate

the radiant temperature of a surface from its total radi-

ant emittance (Chapter 6). The temperature of a rock

is contributed to by two sources of energy: the Earth's

ternal heat flow and solar energy absorbed during

d time. During a 24-h cycle the temperature of the

A/~A~~ );1){

12 Chapter 1

~Q)uc:IU+"'UQ)-.:Q)~

1.3.3 Interaction of electromagnetic radiation withvegetation

Depending on the climate and whether soils are deriveddirectly from underlying bedrock or have been trans-ported, vegetation may show variations that relate togeology. Plants use solar energy to convert water andcarbon dioxide to carbohydrate and oxygen through theprocess of photosynthesis. HoW they do this has a stronginfluence on their interaction with radiation. Being livingorganisms, their metabolism is strongly depend~nt onwater-based vascular systems and cell structures. Theabundance of water in their structure therefore controlsthese interactions too.

The catalyst for photosynthesis is the pigment chloro-phyll, a complex protein containing iron. Chlorophyllabsorbs solar radiati9n to :boost the energy le¥els of elec-trons and so drive the proton pumping across plant-cellmembranes that is the basis for their metabolism. This is~chieved by absorption bands near 0.45 and 0.68 J.lm-inthe blue and red parts of the visible spectrum (Fig. 1.15).That is why most healthy leaves appear green. In addi-tion to its absorption features, chlorophyll can be madeto emit light, or fluoresce, in two narrow bands near0.69 and 0.74 J.lm if illuminated with a strong beam oflig4t, such as a laser. This is the basis of special laserremote se~~g techniques used to assess the chlorophyllcontent of leaves or plankton Chlorophyll, however, isunstable above about 70°C. To protect it from thermalbreakdown plants have evolved means of balancingenergy. This is achieved by strong reflection of nearinfrared radiation (Fig. 1.15), partly by shiny coatingsto leav~s but mainly by the internal cells. The structureof plant cells is such that up to 50% of incident, nearinfrared radiation-is reflected internally to re-emerge. The

U.4 U.:> U.b U./ U.8 0.9

Wavelength (IJm)

Fig.L16 .AJI the different attri!?utes of chlorophy~ contentJeafshape, area and number, together with overall plant structurecontribute to the spectral reflectance properties of a plantspecies. Whereas all four plants shown have rather similarproperties in the visible spectrum, they are clearlydistinguished by their near-WJ;ared reflectance.

remainder is transmitted directly through the leaves.Water in the cells does absorb some energy at its char-acteristic overtone~ around 1.4 and 1.9 ~m (Fig. 1.15),the absorbtance depending on the proportion of cellwater. Beyond about 2.0 ~m leaves absorb near infraredradiation.

The different cell structures, proportions of chlorophylland other pigments, water content and surface morpho-logy of different plants have a marked effect on theirspectral properties in the visible to near infrared (VNIR)spectrum. The spectral reflectance of vegetation increasesvery steeply as wavelength increases beyond about 0.7 ~and 0.75 ~m. This sharp change in spectral reflectance issometimes known as the red edge~Figure 1.16 illustratesthis effect for vegetation of qifferent types. It is clear thatnot only does the height of the VNIR plateau depend onthe species of plant responsible, but the exact positionof the red edge may vary according to plant type aswell These two factor~ can also fluctuate within an indi-vidual species if the plants are under stress, as a resultof deprivation of water or nutrients, or of poisoning byan excess <;>f a toxic trace ele~ent such as chromium."Moreover, pl~nts are assemblages of leaves, spaces, twigsandsom~times br~nches, with different leaf shapes and

~QIuCtotQI

'i=QI~

0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4

Wavelength (IJm)

Fig. 1.15 The typical spectral reflectance curve of a leaf showsthe strong effect of absorption by chlorophyll in the visible partof the spectrum, the efficiency of reflection of near-infraredwavelengths by its cells and the distinctive absorption featuresof water contained in its structure. Shown for comparison arespectra typical of soil and clear standing water.

Electromagnetic Radiation and Materials 13

water supply to toots- and light (which controls theopening and closing of pores). The use of emitted thermalinfrared radiation therefore can provide clues to many ofthese processes.

3

,

~OJuc:."1::OJ-.:OJ~

.~

~

2

0.4 0.8 1.0 1.2

Wavelength (~m)Fig. 1.17 Spectra 1-5 show the progressive stages of colourchange in beech leaves preceding autumn leaffall, ~om darkgreen (1) throtigh light yellow-green (2), orange-red (3), brown(4) and- dead, dry leaves (5). "

0.6

sizes and so on. All the individual interactions in suchcompound structure can interfere and further broadenthe range of responses. This facilitates discriminationbetween species, and between healthy and stunted mem-bers of the same species.

As plants have life cycles of various durations theirspectral properties are not fixed. As a deciduous leafmatures before falling, its chlorophyll content decaysaway thereby removing the strong red absorption(Fig. 1.17). Consequently it changes colour from greenthrough yellow to red. As cells shrink and dry theybecome less effig~nt at refle£ting near infrared radiation.As leaves fall, progressively less of the plant interceptssolar radiation and the reflectance becomes dominatedby the soil and leaf litter beneath it. When new leavesappear, near infrared reflectance is well developed,coupled with high yellow reflectance. As chlorophyllcontent increases the blue and red ~bsorption bandsdevelop, until visible reflectance reaches a minimumat the height of the growing season. Coniferous treesdo not shed their leaves during one season, aIl4_~lwaysdisplay high chlorophyll abundances. As t~eir leayes aresmall, however, less solar radiation is intercepted andso reflectance of the whole plant is lower compared withdeciduous trees in full leaf.

The response of vegetation in the thermal infraredregion is complex. Most of the energy absorbed at shorterwavelengths is re-emitted to maintain the energy balance.The radiant temperature of a plant can be as much as10-1Soc above air temperature at night, and up to 5°Cbelow during the day; Plants also control their temper-ature by transpiration-"the exhalation of moisfure frompores on their leaves and thereby the loss of latent heat inwater vapour. Many factors playa role in determiningthe rate of transpiration: acttial temperature; humidity,

1.3.4 Interaction of electromagnetic radiationwith water

Bodies of water have a rather different response toradiation than that of water bound up in the moleculesof minerals. They do not exhibit the discrete vibrationaltransition bands so characteristic of molecular water.Instead, the spectral response curves show broad features(Fig. 1.15). In the visible range, the interactions dependon a variety of factors. Just considering reflectanceproperties, the amount of visible light reflected from awater surface depends on the illumination angle and thepresence and nature of waves; smooth water can show'sun glint' whereas rippled water does so less often.In general, less than 5% of incident visible radiation isreflected by water.

Water has a high transmittance for all visible wave-lengths, but it increases as wavelength decreases. Asa result, only blue light:penetrateswater beyond a cer-tain depth, longer wavelengths having been absorbedat shallower levels. In clear water therefore, it is possibleto estimate the depth from the intensity of visible radi-ation reflected from the bottom, particularly that of bluelight (Plate 1.1). For depths greater than about 40 in,however, all visible radiation is absorbed and water

,bodies appear dark.

There is also a certain amount of scattering of lightWithin water, which is responsible for the blue colour ofclear water even when it is too deep to see to the bottom..There are two factors contributing to this effect: Rayleighscattering ensures that short wavelengths are scatteredmore than longer ones, and the decrease in transmittanceWith increasing wavelength-ensures that scattered bluelight is unlikely to be absorbed before it escapes at thesurface.

Suspended sediment, plankton and natural dyes,such as tannin from bogs, all increase the reflectance ofvisible light from water. It is thus possible to estimate theamount of suspended material in water from remotelysensed da~.

In the near infrared region water acts almost like a per-fect blackbody, and absorbs virtually all incid!?nt energy.It is the only natural material With this property i andso water bodies can be distinguished easily from othersurface features in this part of the radiation spectrum,even if they are shallow or contain much suspendedmaterial. Being a good approximation of a blackbody,water is a nearly perfect emittet: of infrared radiation,as well as being a good absorber (Fig. 1..18). This meansthat measurements of emitted infrared radiation in the

14 Chapter 1

blackbody

/- Associated resources on the CD-ROMQjuc:IV~'E:Qj+"'c:IV::c~"iij

~Co

VI

'\water

You can access the resources by opening the HTML fileMenu.htm in the Imint folder on the CD using your Webbrowser. More detailed information about installing theresources is in Appendix C.

Exercise 2 uses one of the TNTlite processes thatallow you to browse a library of spectral curves for awide variety of minerals.

~..,.,;I, ,", II0 .2" 4 6 8 10 12 14'.16

Wavelength (IJm)

Fig. I.is Experimental measurements of the spectral radiantemission of water reveal that it is a near-perfect blackbody.This ought to mean that measurements of energy emitted inthe mid-infrared region by water should give a true surfacetemperature. In practice, however, there are complicationsowing to the cl.1illing or warming effects of air on the surfacefilm, so that only approximate, relative temperaturemeasurements are possible.

Further reading

The lists of recommended reading are not exhaustive,but guide readers towards the most useful and up to dateliterature relevant to each chapter. Each paper or bookcontains its own list of references, through which a morein-depth literature survey can be attempted. Exampleshave been taken from as many different parts of the worldand as diverse a range of geological environments aspossible. The bias is towards publications that highlightthe role of remote sensing in geology. Other applicationareas, particularly those c<;>ncemed with vegetation andgeomorphology, can often reveal useful new appr<;>achesand unsuspected links. Remote sensing is not just a toolthat happens to be useful to geologists, but increasinglyplays a coordinating role in multidisciplinary studiesthat unify all aspects of the Earth's natural systems.

General reading

These references comprise the main sources of informa-tion on remote sensing in general and a number of geo-logically orientated texts.

8-14 ~m region can be used to calculate the surfacetemperatures of water bodies very accurately.

A combination of the properties of water in bulkand those of molecular water controls the interactionof radiant energy with water held in the pore spaces ofsoils and rocks. Pore water increases the absorptance ofrock and soil, and so decreases their reflectance. Moistrocks and soils therefore appear darker than when theyare dry: In the infrared region, wet soil or rock displaysin a muted fashion the reflectance troughs produced byvibrational transitions that are so distinct for mineralscontaining molecular water and for vegetation (Figs 1.8and 1.15). In the thermal infrared region, the responseof wet soil and rock is complicated by various environ-mental factors, such as wind cooling and humidity, inan analogous fashion to the effects of transpiration in

plants.Fresh snow is one of the most reflective natural

surfaces at visible wavelengths, but in the near infraredregion its reflectance decreases with 1he presence ofsome of the broad H2O vibrational features. As showages, it recrystallizes, forming larger crystals, and theeffect of this is to reduce the reflectance, especially inthe infrared region. Ice has a similar spectral response,except that its visible reflectance is below 70%, but if itcontains impurities (as in many glaciers) visible reflect-ance can be below 20%. Changes such as these in thespectral response of snow and ice as they recrystallize withage and incorporate impurities can be readily observedby remote sensing, and are useful in hydrological studiesas well as in glaciology. In particular, very old, clean iceis generally blue in colour.

Beaumont, E.A. & Foster, N.H. (eds) (1992) Remote Sensing.Treatise of Petroleum Geology Reprint Series No. 19, Amer-ican Association of Petroleum Geologists, Tulsa, Oklahoma.

Colwell, R.N. (ed.) {1960) Manual of Photographic Interpreta-tion. American Society of Photogrammetry. Falls Church,

Virginia.Colwell, R.N. (ed~) (1983) Manual of Remote Sensing (2 Volumes),

2nd edn. American Society of Photogrammetry. Falls Church,

Virginia.Curran, P.J. (1985) Principles of Remote Sensing. Longman,

London.Drury, S.A. (1990) A Guide to Remote Sensing. Oxford University

Press, Oxford.Drury, S.A. & Rothery, D.A. (1989) Remote Sensing (Distance-

learning course jointly produced by the Dutch and BritishOpen Universities. British Open University code PS670).Open universiteit, Heerlen, The Netherlands.

Goetz, A.F.H., Rock, B.N. & Rowan, L.C. (1983) Remote sensingfor exploration: an overview. Economic Geology 78, 573-590.

Goetz, A.F.H. & Rowan, L.C. (1981) Geological remote sensing.Science 211, 781-791.

Electromagnetic Radiation and Materials 15

.Further Reading for Chapter 1

These references are the most important for expandingknowledge about the physics of remote sensing, in par-ticular the spectral characteristics of Earth materials.

Gupta, R.P. (1991) Remote Sensing Geology. Springer-Verlag,Berlin.

Henderson,F.B. & Rock, BN. (eds) (1983) Frontiers for GeologicalRemote Sensing from Space. Report Fourth Geosat Work-shop, American Society of Photogrammetry. Falls Church,Virginia.

Hord, R.M. (1982) Remote Sensing-Methods and Applications.Wiley, New York.

Lillesand, T.M. & Kiefer, R.W. (1987) Remote Sensing and ImageInterpretation, 2nd edn. Wiley, New York.

Reeves, R. (ed.) (1975) Manual of Remote Sensing (2 Volumes),1 st edn. American Society of Photogrammetry. Falls Church,Virginia.

Sabins, F.F. (1987) Remote Sensing: Principles and Interpretation,2nd edn. Freeman, San Francisco.

Short, N.M. (1982) The Landsat Tutorial Workbook. NASA Refer-ence Publication 1078. NASA, Washington DC.

SiegaL B.S. & Gillespie, A.R. (eds) (1980) Remote Sensing inGeology. Wiley, New York. "-

Smith, J.T. & Anson, A. (eds) (1968) Manual of Colour AerialPhotography. American Society of Photogrammetry. FallsChurch, Virginia.

Swain, P.H. & Davis, S.M. (1978) Remote Sensing: the Quantitat-ive Approach. McGraw-Hill, New York.

The most fruitful volumes in which to seek literatureon every conceivable application of remote sensing ingeology are the Proceedings of the Thematic Conferences onRemote Sensing for Exploration Geology sponsored by theEnvironmental Research Institute of Michigan (ERIM).There have been 14 of these conferences up to 2000, andtheir proceedings contain hundreds of excellent, thoughunrefereed, papers, which are too numerous to list here.The ERIM also sponsors the annual Conferences on RemoteSensing of the Environment, which cover all aspects ofremote sensing, including sections on geological applica-tions. Other conference proceedings should be soughtby using remote sensing as a keyword in bibliographysearches. The main journals for remote sensing are TheInternational Journal of Remote Sensing, PhotogrammetricEngineering and Remote Sensing, Remote Sensing of Environ-ment and IEEE Transactions on Geoscience and RemoteSensing. Many geological journals contain papers on theresults of remote sensing studies. A convenient means ofseeking more recent papers on remote sensing is to lookfor references to older publications using the ScienceCitation Index.

Bartholomew, M.J., Kahle, A.B. & Hoover, G. 0989) Infraredspectroscopy (2.3-20 IJ.m) for the geological interpretation ofremotely sensed multispectral thermal infrared data. Inter-national Journal of Remote Sensing 10, 529-544.

Hunt, G.R. 0977) Spectral signatures of particulate minerals inthe visible and near-infrared. Geophysics 42, 501-513.

Hunt, G.R. 0979) Near-infrared 0.3-2.4IJ.m) spectra of alterationminerals: potential for use in remote sensing. Geophysics 44,1974-1986.

Hunt, G.R. 0980) Electromagnetic radiation: the communica-tion link in remote sensing. In: Remote Sensing in Geology (edsB.S. Siegal & A.R. Gillespie), pp.91-115. Wiley, New York.

Hunt, G.R. & Salisbury, J.W. 0970) Visible and near-infraredspectra of minerals and rocks: I silicate minerals. ModernGeology 1, 283-300.

Hunt, G.R. & Salisbury, J.W. 0970) Visible and near-infraredspectra of minerals and rocks: III oxides and hydroxides.Modern Geology 2,195-205.

Hunt, G.R. & Salisbury, J.W. 0971) Visible and near-infraredspectra of minerals and rocks: II c;arbonates. Modern Geology2, 23-30. , ,

Lyon, R.J.P. 0965) Analysis of rocks by spectral infrared emis-sion 08-25 microns). Economic Geology 60, 715-736.

Lyon, R.J.P. 0972) Infrared spectral emittance in geologicalmapping: airborne spectrometry data from Pisgah crater, CA.Science 17S, 985-988.

Milton, N.M. 0983) Use of reflectance spectra of native plantspecies for interpreting multispectral scanner data iI:t the EastTintic Mountains, Utah. Economic Geology 78, 761-769.

Milton, N.M., Ager, C.M., Collins, W. & Chang, S.H. 0989)Arsenic- and selenium-induced changes in spectral reflect-ance and morphology of soybean plants. Remote Sensing ofEnvironment 30, 263-269.

Smith, J.A. 0983) Matter-energy interaction in the opticalregion. In: Manual of Remote Sensing (ed. R.N. Colwell), 2ndedn, pp. 61-113. American Society of Photogrammetry. FallsChurch, Virginia.

Suits, G.H. 0983) The nature of electromagnetic radiation. In:Manual of Remote Sensing (ed. R.N. Colwell), 2nd edn pp. 37-60.American Society of Photogrammetry. Falls Church, Virginia.

Vincent, R.K., Rowan, L.C., Gillespie, R.E. & Knapp, C. 0975)Thermal infrared spectra and chemical analysis of 26 igneousrock samples. Remote Sensing of Environment 4, 199-209.

Human Vision

eye is physiologically capable of superb perception, ithas remarkably little innate ability. From the immedi-ate post-natal state of being able to recognize just crudeface-shaped patterns, the diversity of human vision, itscoordination with other functions and the intellectualcapacity to analyse and describe the results are learntentirely through experience (Fig. 2.1). This means thatthoroughly unfamiliar ways of displaying something,like the view vertically downwards from an aircraft, ora representation of the way thermal infrared is emittedby an object, can only be interpreted through tuition,practise and the experience of success. It also means thattonal, textural and shape-related attributes of images,which in the normal course of life would be disregarded,eventually can be extracted efficiently from images, givenmotivation. In this chapter some of the functions of thehuman visual system are described in the context ofremote sensing.

2.1 The eye and visual cortex

In the field or in the laboratory, geologists rarely usesenses other than their sight. Even the surface texture ofa rock or mineral is better defined by looking rather thanby feeling. With the exception of the distinction betweensilt and clay by the palatability test (if you nibble a pieceof rock with grains too fine to see, the silt is crunchy,whereas clay has the 'mouth-feel' of chocolate!), pre-liminary classification of rocks and minerals is based ontheir appearance, either to the naked eye or with the aidof a petrological micro~ope. To delve much deeper needspurely artificial means-at the simplest by tests of hard-ness and at the extreme by using a mass spectrometer.Because of this, geologists have had to devise their ownhighly specialized jargon to organize the visible attri-but~s of .r°c~. Given an understanding of how naturalmaterials and radiation interact, it is therefore no greathardship for geologists to use their well-practised visualcap~bilities in interpreting images. Things are that mucheasier and more efficient, however, if images are pro-duced in a form that is attuned to the characteristics ofhuman vision.

Of all the higher functions of animals, sight is theone with the longest evolutionary history. Although thisis not the place to venture deeply into the processes ofnatural '$election that have been involved in the develop-ment of , vision, the survival advantages of being able

to detect and estimate motion, distance and size, and torecognize shapes, patterns and colours are fairly obvious.Eyes have evolved independently in many animal groups,such as the compound eyes of insects and the eerie sim-ilarity of those of squids-humble molluscs-to our own.A curious thing about human vision is that although the

There are few fundamental differences between theoptics of an eye and those of a camera (Section 3.1). Bothuse a lens to form an inverted image on a light-sensitivesurface. In the eye this surface-the retina-is a nearlyspherical curved surface instead of a plane. The lens in aneye serves merely to adjust focus for objects at differentdistances and the main refraction is at the cornea (Fig. 2.2).

>-~::c

t1j

iO:J

b.QIu...QI

Q.

0 10 20 30 40 50 60 70

Age (years)Fig. 2.1 This schematic graph shows the relationship betweenperceptual ability and age. Most rapid development happensbetween birth and the age of eight. The ability of-adults can bechanged significantly by training and new experience.

16

Human Vision 17

The diameter of the retina is about the same as the24 mm depth of a 35-mm film, so the eye can be com-pared meaningfully with a 35-mm camera. The lens ina camera that produces an undistorted view has a focallength (f) of 50 mm. That of the eye is about I? mm,producing a field of view of 1800 compared with one of500 for the camera. The curvature of the retina on whichthe image is focused compensates for the fish-eye dis-tortion inherent in such Wide-angle lenses. The range ofdistances in focus using a lens-the depth of field-':variesinversely with the square of focal length, so the humaneye has about nine times the depth of field of a normalcamera at the same relative aperture. Rather than the eyeelongating and contracting to focus objects at differentdistances, as happens with a camera, muscles attachedto the lens change its thickness and therefore the overallfocal length of the eye.

The amount of light gathered by the eye is controlledby much the same mechanism as in a camera. The pupildiameter is changed by muscles in the iris over the range2-8 mm. These correspond to camera a,pertures (f-stops)of fl8to f12, respectively; Compare this with the filA to fl32of a sophisticated camera. Pupil diameter also affects depthof field in much the same way as a camera's aperture.It helps overcome optical aberrations, and so improvesimage definition. Not only the intensity of light governspupil size, however, but interest and mood too. Theoptimum pupil size for image interpretation is about4 mm or f14. There is a great deal more fun in image inter-pretation than you might think, because this optimumcan be achieved readily by occasional visual and, muchbetter, physical contact with an attractive partner.

The retina is the eye's image-recording device, ana-logous to film and to the charge-coupled device (CCD)arrays in digital cameras. The grains of silver salts orphotosensitive dyes in a film emulsion, or individualCCDs are equivalent to some 130 million light-sensitivereceptors in the retina (Fig. 2.3). They are not immedi-ately at the surface of the retina, To reach them light hasto pass through several layers. The first is formed ofoptic nerve fibres, which gather to a bundle at the opticnerve (Fig. 2.2). The second is a network of blood vesselsand neural cells, which perform some information pro-cessing. The receptors themselves are not distributedrandomly, nor are they of one single type.

All receptors function through bleaching of a pigmentwhen light is intercepted. The pi~ent, rhodopsin, hasan extremely ancient pedigre~, as it is found in purplebacteria that inhabit-highly saline lakes and evolved morethan 3 billion years ago. The bleaching of rhodopsinreduces the receptor cell's permeability to sodium ions.In the absence otlight sodium ions enter the cell to main-tain a current, and light decreases the nerve si~al. Thisis the opposite of what we experien<;e, and again seemsto reflect the antiquity of vision; for a simple animal,

darkening, as by the shadow of a potential predator, isa more relevant stimulus thiln lightening. In. terms ofnerve signals, animal vision is more like a negative thana photographic print. ..

There are two kinds of receptor. Those equivalent topanchromatic film are known as rods becaus~ of theirshape. They cover the whole retina, except for the pointdirectly on the visual axis, known as the fovea.. Rods aremost densely packed at about 200 away from the fovea,and gradually fall off in number towards the edge of theretina. Cones contain three different kinds of pigmentsensitive to red, green and blue light: They number onlyabout 7 million, but are concentrated at and immedi-atelyaround the fovea (Fig. 2.4), although there are soin~even in the outer area of the retina. At the point wherethe optic nerve leaves the eye there are neither rodsnor cones (Fig. 2.4). This is known as the blind spot. Itis rarely noticeable because the brain compensates forthe discontinuity.

Rods cannot distinguish colours, and only respond t9lightness. The peak of their sensitivity occurs in the bluepart of the spectrum (Fig. 2.5), which happens to cpr-respond with the wavelength of maximum atmosphericscattering. Rods therefore are sensitive to the compon-ent of surface illumination derived from the sky itself.They are more efficient than cones under conditionsof low illumination, when their wavelength sensitivity

18 Chapter 2

fovea blind spot

160000 r I'" \

\1 \

\\IIII

"I

I \J \

120000

7'E5~ 80000+"'CoQ)uQ)~

40000

rods .-

II

...

\ rods\

"\

\ ,.I

...I

cones

I" II III III II, II, III II, IIJ II

IIIIIIIIII

r.- ~)ii~!, , ,\J I I' cones "',:1 I:

060 40 20 0 20 40 60 80

Visual angle (degrees)

Fig. 2.4 A plot of the density of receptors in the retina showstheir uneven distribution, from a p1aximum around the foveato a mipimum at the periphery of vision.

comes into its own-at d~sk and at night illumination isdominated by sky light. This scotopic or night vision onlygives images of brightness, which contain no colour. Blueobjects are seen more clearly than red. Definition is alsosacrificed in the same way as it is in a fast film. Visionunder brighter conditions is described as photopic, andinvolves both rods and cones. The peak of sensitivity ofcones and rods acting together in photopic vision is inthe green part of the spectrum and corresponds to themaximum energy region in the solar spectrum (Fig. 1.5).The sensitivity curves for the three main types of coneare shown in Fig. 2.5. Red and green are distinguished bycones only around the fovea, together with blue. Towardsthe periphery of vision only blue and yellow contributeto colour vision. By having a friend bring a colouredobject through your periphery of vision to directly in frontof you, while you stare fixedly ahead, you will find thatyou can see the object in your periphery but only per-ceive its colour when it is in front of you.

Motion or changes in lightness detected by rods on theperiphery of vision signal the viewer to direct the fovealpart of the retina towards the source of the change. Itthen can be examined in colour, with better resolution andwith an element of information processing. The neurones,or nerve cells in the retina can transmit impulses directlyfrom the receptors to the visual cortex of the brain, andare in fact an integral part of the brain. Information pro-cessing begins on the retina itself. Neurones can inhibitor enhance the signals and also ignore them. In Fig. 2.3one kind of neurone (H) links many receptors, thus result-ing in high sensitivity but low discrimination in thespatial domain or acuity. The same receptors are alsoconnected through more specific neurones (B) for highacuity. Which of the alternative circuits are used by wayof a switching mechanism depends on the stimulus itselfand return signals from the visual cortex. Some of theconnecting neurones respond only to specific shapes,orientations or directions of movement, based on assem-blies of receptors forming a receptive field. When theyare triggered they cause a specific cortical cell to fire.

The links between receptors, retinal neurones and cor-tical neural cells are grouped in circuits as cell assemblies.Every new visual experience excites a common core ofneurones and in a sense imprints them with the experi-ence. Exposure to a similar experience more easily re-excites the grouping and adds new neurones to broadenthe experience. These imprinted neurones are some-times unconnected with the visual cortex. For instance,exposure to a waterfall in real life imprints an assemblyin the cortex based on neurones associated with visual,auditory and even tactile stimuli. Looking at an image ofa waterfall later, not only reactivates the visual neuronesbut can bring back the experience of the sounds and feelof falling water. There is nothing mystical in the feelinginduced by a photograph or painting which so closely

(b)blue green red

1

c: 0.80

.~C-

o 0.6III.ara

~ 0.4.~ra

""Qi~ 0.2

0.4 0.5 0.6 0.7

Wavelength (IJm)Fig. 2.5 (a) The overall sensitivity of the eye peaks atdifferent wavelengths for scotopic and photopic vision.(b) The spectral absorption curves for the three differentkinds of cone, normalized for comparison, show that twopeak at blue and green but that normally associated Withred light peaks at orange.

Human Vision

particle-wave duality of raqiation, Heisenberg's Principleof Uncertainty and several other fundamental relation-ships. The human visual system produces an imperfectrepresentation of the real world for a whole variety ofreasons. Foremost amongst these is the way an image ofreality is subdivided into small elements of the naturalcontinuum of spatial and tonal attributes. As the retina isa mosaic of rods and cones, a continuous image fallingon to it is recorded in the form of this mosaic. It, is dis-sected into discrete packages. The unconscious scanningaction of the eye and the aggregating function of thevisual cortex smoothes out the discontinuities, so thatthe impression is of a smooth whole.

All artificial images are dissected too. They may bemade up from brush strokes ina painting, from randomgrains of metallic silver in a photograph or the dots inthe line raster of a television picture. Examining the half-tone and colour images in this' book reveals them to bemade up of assemblages of ink dots and spaces on thepaper. As will be discussed in later chapters, dissectionin a regular format is essential for the transmission andcomputer processing of many kinds of remotely sensedimage. Image dissection by the visual system has animportant bearing on what the human visual system canand cannot see.

Interpreting the spatial attributes of an image relieson a very complex series of interactions. They involveshapes, patterns, assemblages of tones and colours asdifferent. textures, the scale of the image and its context.This function is the most ~ortant in image inter-pretation. In many cases, however,' ~n image containsfar more spatial information than the eye really needs inorder to make a decision. A newspaper photograph of apop star or a widely reviled politician contains between250 000 and 1 million dots, which are sufficient not onlyfor instantaneous recognition but even to make the sub-ject reasonably attractive. On the other hand, a skilledcartoonist or a mi~ic can reproduce instant recognitionby a few sketched lines or suitably exaggerated grimaces.The newspaper image contains much redundant spatialinformation for the purpose of recognition. However, forthe purpose of definite identification-the politician mayeasily have been substituted by a dummy-the imagemay not contain enough information. The question,'Is this the real Adolf Hitler or a rubber doll?', can beresolved only by more information about small physicalcharacteristics, such as the presence of a mole or the pre-cise geometry of a moustache;

For the sake of economy or convenience many..

images have reduced spatial redundancy, and attenuatedtonal or colour redundancy too (Section 2.3). For someinterpretative purposes this makes little difference tothe observer. Figure 2.6 shows an image taken froman aircraft that has been dissected to give differe~tdegrees of spatial redundancy. Some parts of the scene

2.2 Spatial resolving power

The natural world is a continuum of matter in motionand cannot be broken into i~s constituent parts in isola-tion, except by abstractthought. This is expressed by the

20 Chapter2

(a) (b) (c) ,.

(d) (e) (f)

Fig. 2.6 This series of images shows the appearance of the same data after dissection to component picture elements, which increasein size on the ground from 10 m (a) through 20 (b), 30 (c), 40 (d) and 60 (e) to 80 m (f).

are recognizable on all images, others appear only in theleast dissected image. Most observers will notice a jumpin quality about halfway in the sequence, above whichthere is no clear improvement, and below which thereis little apparent degradation. Viewing the images fromincreasing distances seems to cause the position of the

transition to change, until all the images appear much thesame. On the other hand, increasing the magnificationof a dissected image eventually results in the compon-ent parts interfering with the impression of the whole(Fig. 2.7). Below this magnification the elements arementally fused into a coherent whole.

Human Vision 21

and yellows than for red~ and blues. This results fromthe increase in chromatic aberration in the eye at the twoextremes of the visible spectrum.

Contrast is defined as the ratio of the differencebetween the brightness of an object and that of itssurroundings to the sum of the two brightnesses, and isexpressed as a percentage. The higher the contrast ofan object, the more easily it is perceived. Moreover,there is a lower limit of contrast that the eye can dealwith, around 0.5%. The effect of contrast is illustrated byFig. 2.8(a), which illustrates the response of the eye toa point source of energy, such as a star. Although thereal boundary between the point source and its back-ground is sharp, the eye's response gradually changeswith distance away from the point source,. giving a bell-shaped curve termed the point-spread function (PSF).The steeper the gradients involved in this curve, themore likely it is that the point will be perceived. Thehigher the contrast, the steeper are these gradients and

(a)

t pointsource'"

'"

QIc:+"'.c01

":.Q-a~"a;u...QI~

(b)

v.v.OJC....r.C1.~.c"COJ>

"0;v

QjDo

Distance

Fig. 2.8 The change in response of an imaging systemacross a point source of energy produces a variation of imagebrightness with distance away from the point (a). This isknown as the point-spread function (PSF). The more rapid thechange away from the point to the level of the background;the better the resolutio~ of the system. The rate of change withdistance also depends on the ratio between the energy emittedby the point and by its surroundings. This ratio is the contrast.Low contrast and small brightness gradients (dashed lines)result in poorer resolution than high contrast with highgradients of brightness (solid lines). The two point sourcesin (b) can be resolved when contrast is high, but are barelydistinguishable with low contrast.

The chain of processes involved in visual interpretationof the spatial features of an image begins with detection.This is the near-threshold perception that something ofinterest exists in the image. Being able to define the posi.tion of this 'thing' is localization. Recognition that it is aparticular category of 'thing' , an automobile for example,and identification of its unique attributes-a 1969 VWBeetle with rusted fenders, driven by a friend-dependson observing increasingly intricate features of the object.This may not be possible if the scale of view is too smallfor the eye to resolve the component parts of the objecton the retinal image. At too large a scale the imagestructure may dominate the view and destroy the visualform of the object. A view of only a few parts of the objectin an unfamiliar context may also hinder recognitionand identification.

The elements of image interpretation depen~ on theacuity of the visual system, mental fusion, the degreeof dissection of the image, its scale, and on a combina-tion of all these factors, which is expressed as resolution.Another, less well-defined factor is the context of objectswithin the image.

The acuity of human vision is a measure of ability todistinguish spatial detail. It depends partly on factorsinherent in the visual system and partly on purelyexternal factors.1 Results from experiments depend to some extent onwhat the subject is called upon to do. Acuity is differentfor detecting the presence of a test object compared withdiscerning whether there is one object or two. Likewise,different results occur for localizing the position of theobject, and for recognizing what it is.2 The polarity of the target also has an effect. Brightobjects on a dark background are more easily detectedthan dark on light because scattering in the fluids withinthe eye broadens a bright stimulus.3 Contrast is important, a black object on a white back-ground is more easily detected than dark grey on lightgrey.4 If the pupil is dilated beyond the optimum of 4 mm,spherical and chromatic aberration in the eye degradesthe retinal image. Constriction of the pupil results indegradation by diffraction.S The eye takes time to adjust to changed illumina-tion conditions, so there is a time-dependent effect onacuity.6 The illumination or overall brightness of an imagehas an effect depending on whether photopic vision ispossible or not. Retinal rods have poor acuity comparedwith cones, and below a certain brightness they take overand scotopic vision is used. Very bright images degradethe acuity of cones as a result of overwhelming the neuralchemistry involved in rhodopsin.7 Under normal illumination conditions the colours inan image help determine acuity. It is higher for greens

22 Chapter 2

actual frequencies depending on the grain size of the filmor paper. The MTF of the eye is different for coloured(chromatic) compared with black and white (achromatic)images (Fig. 2.9d). For frequencies lower than 2 cyclesper degree at the retina (spacings in the image greaterthan about 0.2 mm) the chromatic MTF shows that theeye discriminates colour changes in space very well.This colour acuity drops rapidly for more closely spacedobjects. The achromatic MTF, however, shows a clearmaximum of fidelity at around 7 cycles per degree(spacings in the image between 0.03 and 0.06 mm). Ablack and white image is obviously better for analysingspatial detail. At higher spatial frequencies the modula-tion falls because of the physical size and distributionof receptors, and as a result of the blurring effect ofunconscious eye movements. The fall at lower frequenciesis explained shortly. In a sense, therefore, wheJ:l lookingat black and white scenes, the eye performs as a spatialfrequency filter (Chapter 5) and enhances frequenciesaround 7 cycles per degree at the expense of others. Itsorts the trees from the wood, but is unconcerned withthe leaves.

In reality, matters are a good deal more complicated andinteresting than this. Assemblies of receptors that func-tion to pick up boundaries make comparisons in spacebetween the amounts of light in neighbouring parts ofthe image. The receptive fields are roughly circular, andtherefore operate irrespective of boundary orientation.A central subregion of each responds in the oppositeway to an outer 'ring' in the field, thereby enhancing thestimulus, As you will see in Chapter 5, this is how certaintypes of image-processing filters work with digital images;they were devised before their homologues in the eyewere known. Receptive fields vary in size, so that theeye itself performs image enhancement over a range ofscales. This goes some way to explaining the differencebetween the achromatic and chromatic MTFs, rods beingmore closely spaced than cones and therefore able toform smaller cell assemblies and receptive fields.

Any image or natural scene can be regarded as, andreconstructed from a 'spectrum' of sine waves with dif-ferent directions, wavelengths and amplitudes. Such areconstruction is known as a Fourier synthesis, and thedisassembly of an image into a family of sine waves is aFourier analysis (Chapter 5). The spots, lines and barsthat form the targets for tests of acuity, and form imagestoo, are themselves made up ofsme waves. Small dotsand closely spaced lines are dominated by high ampli-tudes at high frequency. This is why they are difficult todiscern. As size increases, high frequency componentsqecome lower in amplitude and those in the frequencyrange to which the eye is attuned increase. Perceptionand distinCtion become easier. Beyond this, however,very low frequency components increase in amplitude;effectively degrading perception and distinction.

the scale at which the image is presented is a major fac-tor in interpretation. The retinal frequency (f cycles perdegree) of features on an image is a function of imagescale (1:5) and viewing distance (X in metres). The fol-lowing equation relates these to spatial wavelength onthe surface (A. in metres):

A. = 5Xtan (1If) (2.1)

The relationship in Equation (2.1), together with the effectof varying contrast, is of crucial importance in designingmethods of image interpretation of different terrains.Even the best available image Will have little effect on thelimitations posed by the human MTF if it is presentedat an inappropriate scale. To improve significantly thedetail in an interpretation requires a dramatic increasein the scale. To examine very large features means anequally large decrease in scale. For example, a blackand white remotely sensed image at a scale of 1 : 500 000being viewed from a distance of 1 m, gives most visualimpact to features with dimensions of a1?out 1.25 kmon the ground, i.e. to those appearing on the retina at7 cycles per degree. Those with dimensions less than0.5 km and more than 3.5 km will excite less than halfthe response in the humanvisual system (Fig. 2.9d),For these purely physiological reasons, therefore, imageinterpretation is best achieved by combining imagesobtained from different altitudes, in which the degreeof dissection need be sufficient only to avoid the obviousappearance of segmentation at the scale used (Figs 2.6and 2.7). Using a single image at different scales andviewing diStances also helps extract more of the spatialinformation.

In practice, resolution combines the acuity of the eyewith the resolving power of the imaging system, themeans of reproduction, the scale of the image and theconditions under which it is viewed. It is defined asthe ability to discriminate individuals in a set of smallsimilar items. It is not really measurable in practice, beinga threshold betWeen can and cannot. Discrimination ofobjects is easier using a black and white image than witha colour lInage at the same scale.

There are some other interesting and useful featuresof the eye's performance in the spatial domain. First,lines can be separated more easily than points with thesame width and spacing. This is because receptors arelinked in arrays aimed at line detection. Second, it iseasier to detect offsets in lines than to separate linesWith a spacing equal to the offset. The absolute limits are2 s of arc subtended at the eye for offsets and about a

24 Chapter 2

minute of arc for lines. The area on the retina with thegreatest power for spatial resolution is near the fovea,where receptors are most densely packed. Optimuminterpretation is, as a result, achieved by flicking directattention from one part of an image to another. Inother words! a good interpreter peers at images for longperiods, often many times; so they are best pinned up onthe office wall!

2.3 Seeing brightness

The eye has a remarkable capacity for detecting light.In a darkened room a flash of light corresponding to10 quanta absorbed by an area of retina containingabout 500 rods can just be detected. This suggests thata single quantum has a realistic probability of exciting asingle rod. A rod by itself, however, cannot produce asensible signal to .the brain, but needs the simultaneousresponse of about 10 other rods. Curiously, this phenom-enal detection ability by no means enables the visualsystem to distinguish brightness variations in a con-tinuous, quantum by quantum fashion.

Experiments on human light detection are based<;>n the noticeability to an observer of changes in theluminance of a lamp or in those of sheets of exposedphotographic paper with different grey tones. Theterm luminance is used to distinguish the objectivemeasurement of the amount of light involved fromthe perceived brightness of a lamp and the lightnessof a reflecting surface. One set of experiments showsthat the difference-detection capacity of the eye is non-linear (Fig. 2.10), The perceived brightness or lightnessincreases rapidly at low luminance and more slowly athigherlu~ances.

Another set of experiments focus on the eye's con-trast-detection powers. Contrast can have positive ornegative values: dark object and light surround, and viceversa. ,.For objects and surroundings with a large visual

angle-the ratio between the dimensions of an object andits distance from the eye, expressed in radians-and highluminance, the minimum detectable contrast is 0.5%.This suggests that about 200 steps of contrast can be dis-criminated in a grey scale made up of sharply boundedluminances. This is very interesting. Although it is aphenomenal detector of light, the eye is not a good dis-criminator of perceived brightness or lightness. In fact,as the visual angles of objects decrease to those mostcommonly found in images, the range of discerniblecontrast differences decreases to about 20-30 steps.

Returning for the moment to objects with large visualangles, it is interesting to note that contrast detectionfalls off towards lower luminances of object and sur-rounding. As a result the lightness perceived as themid-tone of a sequence from black to high luminance ina grey scale is darker than the real mid.,tone (Fig. 2.11).The perception of a grey tone also depends on whetherthe contrast is positive or negative. A grey patch ona black background appears lighter than an identicalpatch on a white background (Fig. 2.12a). The eye is thuscapable of limited contrast enhancement. The reverseeffect is produced when fine elements of black or whiteare introduced into a grey object (Fig. 2.12b), The poorperformance of the eye with contrast differences isimproved to some extent by a peculiar illusion associ-ated with tonal boundaries. On Fig. 2.11 narrow bandsare seen immediately adjacent to the tone boundaries-light bands in the paler blocks and dark bands in thedarker. These Mach bands disappear on covering theboundary. Their effect is to enhance and sharpen tonalboundaries. When the boundary is covered it is diffi-cult to distinguish the two adjacent grey tones. TheMach-band illusion stems from the image processingperformed by assemblies of receptors, and the same thinghappens at tonal boundaries in images when they arefiltered by algorithms that are homologous to those inthe visual system (Chapter 5).

Black and white photographs or analogue televi-sion pictures can contain hundreds or thousands of dis-crete grey tones. The vast majority are redundant as faras an observer is concerned. If they are progressivelyreduced, little difference is noticed until the numberfalls below about 16 (Fig. 2.13). The obvious implica-tion is that the tonal range of a black and white image-can be severely truncated, without loss of interpret-able information, This is in fact inevitable when high-quality photographs are screened before printing, andwhen- limits have to be imposed on the tonal rangeof data gathered by digital remote-sensing systems (orin digital television transmission) for the sake of con-venience and computing economy. The eye's limitationsin the domain of grey tones, however, is amply com-pensated for by its extraordinary efficiency in the colourdomain.

100

dark surround

light surround

V\V\

Q)

C+"'.c"~.c

Q)>".-ro""Qi~

i 0Luminance relative to white

Fig. 2.10 The relationship betw~en tneperceived brightness ofa lamp and its luminance showsthatth~ difference-detectingability of the~ye is nonlinear.

80

60

40

20

Human Vision

~

perceptualmid-tone

~

"The observed n'iid-tone between black and\white on a slightly to each side of a boundary between grey tones

(bottom). These Mach bands disappear when the boundary iscovered and it is barely possible to separate the two adjacentgrey tones. The steps in reflectance are each 5%.

~

(a) (b)

,(a).~

~impleillustration

appears darker against the whitebackground. The effect is morepronounced if the vertical divide ismasked with a pencil. (b) A uniformgrey background appears darkerwhen partly filled with black linesand lighter when white lines arepresent.

K

Fig. 2.13 In this series of photographs the same image is shown with 128, 64, 32, 16, 8 and 4 grey levels (from top left to bottom right)to show the futility of trying to perceive more than 20-30 grey tones distributed as small picture elements.

50%reflectance

26 Chapter 2

2.4 Producing, seeing and representingcolour

Additive primary cQlours are used in colour televi-sion screens coated with spots of red-, green- and blue-glowing phosphors. The subtractive primaries are usedin colouring paints, in lithographic printing and in re-versal colour ffim. The important point about primarycolours though, is not their usefulness nor the physicsinvolved, for the same effects are produced by broadspectral bands in ffiters as well as by truly monochro-matic light. Colour for us is a product of cone receptorsin the retina and analysing their combined response inthe brain. Young did not know this, but he did realizethat there must be a connection between his discoveryand the way the eye works. He postulated a tristimuluscolour theory, relating the three additive primaries tothree different kinds of optical 'nerves', or trichromaticvision. Incidentally, some birds have quadrichromaticvision, using different coloured oil droplets as filters intheir cones, rather than having three different varietiesof rhodopsin in cones. Our colour vision is in the mannerof the goldfish, whereas other animals have no colourvision (achromatic) o.vision based on two colours only(dichromats).

That cones have three ranges of sensitivity, peaking inred, green and blue (Fig. 2.14) was not demonstrated until1964, but this does not explain all perceptual oddities infull. When green is perceived through one eye at the sametime as red is seen by the other, the sensation is yellow.This cannot be due to the cones alone. For this reason,and several others, refinement of Young's hypothesisled to the assumption in the 19th century of three pairsof processes involving white-black, blue-yellow andred-green, in the 'opponents' colour theory. This toohas been confirmed by experiments, which suggest that

Although visual perception is limited in practice toabout 20-30 grey tones, the eye can easily distinguisha huge range of different colours. The limit is thoughtto be about 7 million. Colour dominates human vision.For this reason there is a long history of research intohow- colour is produced and perceived, enormous diffi-culty in truly cataloguing colours, and bitter controversyabout the contribution of colour to aesthetics. In com-mon practice, it is the source of great annoyance in tryingto touch up chipped paint on automobiles-it truly is alaw of nature that no autoshop carries matching paintfor our cars.

Thomas Young (1773-1829)-famed fQJ Young'sModulus and the wave theory of light-discovered thatby projecting light through red, green and blue filters toproduce overlapping circles several remarkable effectsare produced. First, where all three filters overlap-where the eff~s of all three add together-white light isproduced. As the spectrum of white light produced by aprism or in a rainbow contains-a continuum of colours,this was unexpected. Second, where two colours over-lap, other colours are produced. For a red and bluemixture the resulting reddish-blue or magenta is pre-dictable. So too is the production of cyan by an equalmixture of green and blue. The result of adding red andgreen is yellow, which has no red or green attributesat all. This is completely unexpected. By varying therelative luminosities of red, green and blue in the over-lapping zone, Young was able to produce all the otherspectral colours and many others too, such as brownand pastel shades. These phenomena are illustrated inPlate 2.1. Another of Young's observations was that red,green or blue cannot be produced by mixtures of theother two.. Those three ~olours therefore must be so dis-tributed in tne visible spectrum as to ~onstitute separatevisual entities. Young termed them additive primarycolours, because their effects result from addition. Other~Qmbinations of spectral cololJrsalso act additively, butonly red, green and blue can produce the full range ofcolour sensations.

A complementary effect can be produced by subtract-ing red, green and blue components fro~ white light,using filters. Subtraction of red produces cyan, whiteminus green gives magenta, and the result of subtract-ingblue produces yellow. All three filters together trans-mit no light. Magenta plus yellow filters produce red,blue results from equally combined cyan and magenta,whereas yellow plus cyan transmits only green. Mixedcombinations of all three subtractive primary colourswith different luminances result in the whole gamut ofcolour sensations.

0.4 0.5 0.6 0.7

Wavelength (IJm)Fig. 2.14 The three types of cone show different spectralsensitivity curves. Although the overall sensitivity forphotopic vision peaks in the green, blue cones are mostsensitive. This compensates for the smaller number of blue-sensitive cones in the retina.

Human Vision 27

l take the form of one

two sets relatingJo colour

and deriving yellow information

.The present consensus is

the tristimulus and opponents processes are

Independent research on the transmission of

I that the most conveni-

--' 1 red-green

blue-yellow signals. These are converted in the

streams emitted by guns focused on the three sets ofdifferent phosphors on the screen.

As if this was not sufficiently complex, 'ithe visualresponse to colour has many other peculiarities, some ofwhich are significant to the interpretation of colour images.Some are physiological, others are entirely psychological.

Figure 2.12 showed that perception of brightness isaltered by the contrast between an object and its sur-roundings. Very similar effects occur when identicalcolour patches are placed on light or dark backgrounds,or on differently coloured backgrounds. Not only are theapparent brightnesses different, the perceived coloursare different too. This effect is physiological. A psycho-logical effect also governs how colours are remembered.In general, light colours are remembered as lighter anddark colours as darker. Impure colours, such as brown,reddish-yellow and bluish-green are remembered asbeing more strongly tinted by the dominant primary, andare interpreted as more red, yellow Or green, respect-ively. Memories of colour images are more vivid andsimpler than reality. This a source of great irritation amongremote sensors; the enhanced colour images that theymake by various image processing techniques (Chapter 5)never again look as 'good' as when they first appearedon the computer monitor.

Learning by experience and with guidance providesan association of colours with familiar objects, any per-ceived deviations from this being disturbing. Pale blueskin, purple bananas and green meat are not attractive.The influence of learned associations is so strong thatwith lists of names of colours printed in the 'wrong'colours it is almost impossible to identify the coloursused without some confusion. Remote sensing covers awide spectral range in which the reflectance or emittancecharacteristics of familiar materials deviate from thosein the visible range (Section 1.3). Images produced byassi~ng data from nonvisiblebands to red, green andblue give a totally 'false' impression. The commonestexample is where near-infrared data are assigned to red,red data to green and green to blue in a monitor display.The appearance of vegetation in various shades of red is

aesthetically disturbing. From an objective standpoint itis much more useful than a pleasing natural colour image,however, because of the great variation of infrared reflect-ance among plants (Section 1.3.3). If the information con-tained in electromagnetic (EM) radiation outside thevisible range is to be exploited, this visual dissonancemust be overcome by practice.. So, you may ask, why notdisplayth~ infrared data as green, so that vegetation looksmore 'natural'? There is a good reason for not doing so;we are able to perceive a greater range of red shadesthan green, and blue is the least-well perceived of all.Long experience shows that the most visuallystimulat-ing colour images use the most information-rich data tocontrol red, and the least to control blue.

Psychophysiology may be the main control over theinterpretation of and information extraction from colourimages, but it is conditioned by an individual's sense ofaesthetics. As a general rule, the most pleasing image isoften the most suitable to work with. Human aestheticsis exceedingly curious. Our home planet Earth is regardedas a beautiful and stimulating environment, for the mostpart at least. The human visual system'evolved in thisenvironment. Considered objectively though, it is not par-ticularly colourful. A vast proportion comprises shadesof blues, greens, greys and browns. The full capacityof colour vision is reserved for only a tiny volume offlowers, fruits and animals in the natural world. Althoughwe do not have vividly coloured genitalia or plumage,we have developed aesthetics out of natural experienceand exercise colour vision much further through art,clothing, furnishings and cosmetics. A few guidelines canbe abstracted from this controversial area.1 Many people label some colours as 'warm' and othersas 'cool', associated with fire or sunlight, and foliageand water, respectively. There is a general preferenceior'warm' colours.2 Given a choice of six spectral colours, experimentsshow that there is a clear order of preference from mostliked to least liked of blue, red, green, violet, orange andyellow.3 For any given colour there seemS to be a pr~ference fora lightness that corresponds with the relative lightnessof the colour in the spectrum-light yellow is preferredto dark yellow, but dark reds and blues seem morepleasing than light shades.4 How colours are arranged relative to one another inthe spectrum appears to govern perceptual harmony. Ifthey are arranged in a circle, adjacent colours, such asyellow and green; or violet and red, harmonize well. Sotoo do diametrically opposite colours, such as yellow andblue. The greatest clash is between combinations of redand yellow-green, for instance, which are about 900 apart.5 Different colours harmonize best when they are equ-ally light and dark, or have the same saturation with aspectral colour.

28 Chapter 2

It cannot be overemphasized that how pleasing a colourimage is varies widely from person to person. How-ever, these general rules of thumb, governed by personalfoibles, can be used to arrive at the most stimulating andmost easily interpreted images of remotely sensed data.Although a certain amount of 'matching' to personalityand mood can be achieved during photographic process-ing, the ultimate control depends on image processingcomputers (Chapter 5).

The enormous range of humanly sensible colours posesan immediate problem of nomenclature, measurementand cataloguing. Means of satisfying these requirementsare necessary for many practical reasons as well as forteaching and research. The major step in systematizingcolour was made by Albert Munsell in 1905, when hetried to produce a colour atlas. He based his approach(the Munsell system) on subjective comparison, but itcontains the elements of all subsequent colour-coding

systems.Any coloured surface appears to have an attribute

related to one of the spectral colours or to nonspectralpurple. Munsell designated this the hue of the surface.His system has five principal hues-blue, green, yellow,red and purple or B, G, Y, Rand P-and five intermedi-ate hues-BG, GY, YR, RP and PH. These 10 hues arearranged in a colour circle so that the apparent differ-ence in hue between adjacent surfaces are constant. The10 segments are divided into 10 subdivisions, to give auseful range of 100 hues. A hue coded as lOGY is fivesteps from green-yellow (5GY) towards pure green (5G)(Fig. 2.15).

Any coloured surface has a luminosity equivalentto a shade of grey between black and white. This is itsMunsell value, and ranges from 1 (black) to 10 (white).Colours also appear to be a mixture of a grey and a purehue. A hue which contains no grey component. is said

white

110 fva'ue s~:~e

15P, ,1///, IU"/~//5A!

106Hue sy",

106G;:::

,5'f'B

~10RP

~

to be saturated, and it possesses a high chroma. Onewith a high grey component has a low chroma. Thechroma scale in the Munsell system ranges from 0 forachromatic greys to 16 or 17 for the highest conceivablesaturation. As Munsell value decreases, so the numberof discernible chromas decreases. The Munsell system istherefore three dimensional, as shown in Fig. 2.15, butoccupies an irregular solid. This is because it is basedon human perception of colour, which is not regular.However imperfect it might be it is still the only inter-nationally accepted standard for visual comparison. Itis used by soil scientists, textile and paint manufacturers,and many other groups.

In 1931 the Commission Internationale de l'Eclairage(CIE) produced the nearest approach yet to a fully object-ive means of describing and measuring colour. TheCIE colour co-ordinate system is based on a tristimulusset of co-ordinates, where the three primary colours areimaginary, yet produce a colour space which containsall perceptible colours. The CIE primaries stem from thereal primaries red, green and blue. In the same way asall spectral colours can be made from different propor-tions of monochromatic red, green and blue light, it ispossible to express the CIE tristimulus values involvedin producing any monochromatic colour by three inter-fering curves shown in Fig. 2.16(a). For light with equalluminance at any visible wavelength the contributingtristimulus values can be read from the graph. Theseare divided by their sum to give three chromaticityco-ordinates, x, y and z. These co-ordinates sum to 1.0,so any colour can be specified by just two co-ordinatesand a value for the luminance. On a plot of x against y allperceivable colours fall within the curved line shownon Fig. 2.16(b). The numbered section of the line refersto saturated monochromatic light from violet (0.38 ~m)to red (0.65 ~m). The straight base of the enclosed area

represents nonspectral purples.There are two difficulties with the CIE chromaticity

diagram. First, equally perceived differences in colourare not represented linearly. Second, it represents ideal-ized conditions. How the chromaticity co-ordinates ofsamples are measured to classify materials in this systemare beyond the scope of the book.

Various attempts have been made to simplify theMunsell and CIE systems, as aids to expressing remotelysensed data in the most visually stimulating way pos-sible, with the least expenditure of computing time. Onemethod that simplifies the Munsell system to threemeasures of colour-hue, saturation and intensity-hasreceived particular attention. These approximate Munsell'shue, chroma and value in a conical colour space derivedmathematically from red-green-blue colour space. Thistransformation allows interesting means of enhancingcolour and combining unrelated types of data; It is dis-cussed further in Chapter 5.

:L]

)R -I

ma scale

black

Fig. 2.15 A schematic representation of the axes used byMunsell in devising his Book of C%ur shows the divisionsthat he used for hue, chroma and value. The value axis isa grey scale.

(b)

QJ:J""jij>\II:J""5E~O':+"'

""jij

tQJa.

III

yyx

0.4 0.5 0.6 0.7

Wavelength (~m)Fig. 2.16 (a) These curves represent the variation withwavelength of tristimulus values for the imaginary primariesused in constructing the Commission lntemationale de

x

l'Eclairage (CIE) chromaticity diagram. (b) The dEchromaticity diagram is based on plots of the tristimulusvalues x and y in (a) for any given luminosity.

If a pencil is held at about 30 cm away, in front of amore distant object, alternate winking of each eye givesthe familiar impression that the pencil moves relativeto the distant object. Holding the pencil at arm's lengthreduces the apparent displacement. Alternately focusingon the pencil and the distant object with both eyes givesthe stereoptic impression that the two objects are indeedseparate distances away. With one of the objects ster-eoptically fused, the points of focus of each object aredisplaced on the retina. This is why we see two images ofthe distant object when we focus the pencil at the fovea.Figure 2.17 shows that the image of the more distantobject falls further from the fovea in the right eye thanit does in the left eye. To focus the distant object at thefovea requires a change in the eyes' angle of convergence.The difference between the two angles of convergenceis the relative parallax of the two objects. It is this dif-ference in parallax that is the most important stimulusfor depth perception in the nearby environment. As inthe case of estimation of distance, it can function onlyfor angles greater than 1 minute of arc. This binocularstereopsis is the basis for three-dimensional viewing ofpairs of remotely sensed images in which large relativeparallaxes are produced by the method of capturingimages. This is explained in Appendix A, and used extens-ively in the book, Chapter 4 especially, using differentviewing devices.

2.5 Perception of depth

After colour vision, the most important aspect of thehuman visual system is based on the fact that it is bin-ocular. A visible object subtends a perceptible angle atthe retina of one eye which depends on the physicalsize of the object and its distance. The remembered sizeof a familiar object allows the brain unconsciously toestimate roughly how far away it is. For normally sightedpeople, an image focused on the retinas of both eyesenables them to perceive a more exact impression of theposition of the object in three dimensions. The funda-mental reason for this is the 6-7 cm horizontal separationof the eyes, known as the eye base.

Rays extending from a point on an object to theeyes are separated by an angle which decreases as thedistance of the object increases (Fig. 2.17). The angle inradians is approximated by the ratio between the eye-base and the distance to the object. That angle is theobject's absolute parallax at which the eyes must be con-verged to focus the object at or near the fovea. The visualcortex unconsciously fuses the two images, the outcomebeing a stereomodel of the location of the object in threedimensions. Because the retina can resolve only angularseparations greater than about 1 minute of arc, there is alimit to stereoptic vision.

30 Chapter 2

b"

L-~

~ ~ ; ~

C/~~

Fig. 2.17 Points on two objects atdifferent distances have differentabsolute parallaxes (aJ and az)measured in radians as a/E. Therelative parallax is the differencebetween the two angles.

£,E2 '.-.'-.

The limited angular resolving power of the eye meansthat we cannot see stereoptically beyond about 400 m.We learn to estimate distance beyond that from variousfeatures, such as the expected sizes of familiar objects,perspective effects and the decrease in luminosity causedby increasing absorption and scattering in the atmo-sphere. The absence of this last factor is responsiblefor the 'unreal' appearance of photographs taken on theMoon, where there is no decrease in luminosity of objectswith distance. As most remotely sensed images representa view from vertically above the Earth's surface, manyinherent means of depth perception are of little use. Onethat is important is the interpretation of shadows-ormisinterpretation of naturally dark surfaces-accordingto psychological rules that have evolved in the humanvisual system.

Shadows result when illumination is from the side ofan object. Much of the texture in a view of the Earth'ssurface results from shadows cast by hills into valleys.Shadows give a very real impression of depth, but unlessthe direction of illumination is known the impressionis ambiguous. Figure 2.18 contains perspective cluessuggesting an oblique view of a landscape with circu-lar features. Most observers will interpret the featuresas craters in the surface. Inverting the figure gives theopposite impression. Staring at the figure rotated through900 for a while enables both alternatives to seem pos-sible, although for most of the time no depth is seen.Two interesting observations can be made. In the lastexperiment the perspective effect, the shading and thestrange orientation of the 'horizon' cause visual dis-sonance. The first two experiments, however, suggestthat there is a subconscious assumption that illumina-tion is from the top of the figure. Strangely, an even moreconvincing illusion of depth results from oblique shading.Cartographers have long exploited this by shading con-toured topographic maps as if they were illuminatedfrom the top left, or north-west on a Mercator projection.Remotely sensed images usually show no perspective,

Fig. 2.18 Viewed this way up the image is clearly that of acrater with a central peak. Viewed from the top of the page thimage should 'flip' from reality to inverted relief.

and shadows allow a moderately good impression oftopographic relief. Many of the examples in later chapterscan show both 'positive' and 'negative' topographies,however, depending on how they are viewed. In imagesfrom satellites, such as Landsat or SPOT (Chapter 3),the time of capture is mid-morning, with illuminationfrom the south-east in the Northern Hemisphere. Thisinvariably gives the illusion of inverted relief in correctlyorientated images.

2.6 Dangerous illusions

As well as the pseudostereoscopic tricks that shadowedimages can play on the observer, certain aspects of our

image interpretation. There are hundreds

which are discussed in text books on humanMost are based on fairly simple, sharply outlined

.-, -,

References to the psychophysiology of vision, so import-ant in presenting data as images for visual interpreta-tion, are replete within the fields of psychology. Manyare difficult to understand or contain only small sectionsof direct relevance here. The following references areuseful guides.ism. For the world of the familiar this is fine. However,

the eye can 'invent' lines. It can also link separate pointsin smooth curves, and is prone to sensing the unfamiliarhidden in a familiar background-as children, we haveall stared at complicated wallpaper and seen dreadfulbeasts lurking therein. Similar phantasms lie in wait forthe unwary image interpreter.

Remotely sensed images are, by and large, less fam-iliar than images of our immediate surroundings. Theycontain information relating to the unknown, or at leastthe partly known. The positions from which they areacquired are totally outside our normal experience, andmany relate to invisible parts of the spectrum. All con-tain a huge variety of detail, with both organized andrandom elements, and therein lies the main source ofillusions. Our visual system can help us to abstract mean-ingful spatial information, but it also can overestimatethe importance of those kinds of features to which it isattuned-lines in particular. Moreover, there are suspi-cions that any interpreter perceives regular geometricpatterns in objectively homogeneous images. There hasbeen little research in this area. Some studies seem toindicate that the same image excites different responsesin the spatial domain for different interpreters. Realfeatures may be there, but some of those identified bythe different subjects are probably spurious. Althoughremote sensing with image interpretation is probably

Buchanan, M.D. (1979) Effective utilization of colour in multi-dimensional data presentation. Proceedings of the Society ofPhoto-optical Instrumentation Engineers 199, 9-19.

Burns, K.L. & Brown, G.H. (1978) The human perception ofgeological lineaments and other discrete features in remotesensing imagery: signal strength, noise levels and quality.Remote Sensing of Environment 7, 163-167..

Burns, K.L., Sheperd, J. & Berman, M. (1976) Reproducibility ofgeological lineaments and other discrete features interpretedfrom imagery: measurement by a coefficie~t of association.Remote Sensing of Environment 5, 267-301.

Canas, A.A.D. & Barnett, M.E. (1985) The generation andinterpretation of false-colour principal component images.International Journal of Remote Sensing 6, 867-881.

Cornsweet, T.N. (1970) Visual Perception. Academic Press, NewYork.

Halliday, T. (ed.) (1995) The Senses and Communication. Book 3 inBiology, Brain and Behaviour (SD206). The Open University,Milton Keynes.

Padgham, C.A. & Saunders, J.E. (1975) The Perception of Lightand Colour. G. Bell and Sons, London.

Siegal, B.S. (1977) Significance of operator variation and the angleof illumination in lineament analysis on synoptic images.Modern Geology 6, 75-85.

Stroebel, L., Todd, H. & Zilla, R. (1980) Visual Concepts forPhotographers. Focal Press, London.

Wise, D.U. (1982) Linesmanship and the practise of linear goo-art.Geological Society of America Bulletin 93, 886--888.

How Data Are Collectedu

Up to the mid-1970s, geologists used remotely sensedimages acquired by photographic cameras mounted onaircraft. Electronic data-gathering methods developedrapidly for two main reasons. First, the increasedemphasis on planetary monitoring by unstaffed satel-lites and probes meant that it was impossible to retrievefilm, except by costly return and re-entry manoeuvres.This , demanded systems that would transmit data usingmicrowave signals. Second, ventures into that part of theelectromagnetic spectrum where film will not respondrequired the development of new kinds of instrumentsbased on electronic imaging. Some of them, such asvideo cameras, collect and transmit images that cor-respond to a photographic 'snap shot'. They consist ofrepresentations of the intensity variations in a scene inthe form of modulated signals similar to those used fortransmitting and recording television broadcasts. Thesignals corresponding to radiance are continuous andproduce pictures that are described as analogue imagesin much the same way as are photographs. They are,however, by no means perfect representations of thecontinuum of reality on the ground, one reason beingthat they are dissected in a similar fashion to images col-lected by the human eye. In photographs the dissectionis based on the grain of film emulsion, and is not regularin detail. In electronic devices the dissection is regularby virtue of how the means of detecting radiation arestructured.

Electronic systems gather data as a sequence of smallparts of a scene. The data have a structure similar to thatof a tiled floor or a regular mosaic, which is reconstructedto form a picture. lr\dividual 'tiles' are known as pictureelements, or pixels. Electronic, analogue devices-TV andvideo cameras before the digital broadcasting revolution-produce tiled images that are reproduced on a videomonitor by the sweeping or 'refreshing' of the screen byelectron beams. The refreshed display of the tiled matrixof pixels is termed a raster. Precise analogue measure-ments, such as 1.763 mV, are cumbersome to transmit,however, and the information that the signals carry iseasily degraded by interference or by copying at relays.Rescaling and transmitting these numbers expressed bybinary arithmetic-digital numbers (DN)-largely removesthese problems. In binary a string of zeroes and ones,or 'offs' and 'ons! can express any number, so that thesignal becomes as simple as possible. This means that,provided the signal is detectable, the binary digits areunmistakable. There is no degradation and the data canbe copied or relayed infinitely. The range used mostcommonly has eight binary bits (a binary bit is a single

'on' or 'off'-l or O-in a binary number) or one byte, rep-resenting a range of intensities by digital numbers from 0(00000000 in binary) to 255 (01111111). These are 8-bitintegers, but any number, floating-point or complex, canbe expressed in binary form too as 16-, 32-, 64- or 128-bit'words'. Binary data, of course, are those that digitalcomputers were designed to manipulate, and digitalimage processing (Chapter 5) increases analytical and

interpretative power enormously.Recent developments do not imply that aerial or satel-

lite photography is redundant. There are vast archives ofsystematically gathered photographs, and they will con-tinue to grow as long as photography maintains advant-ages of cost, flexibility and resolution over digital imagecapture. At the time of writing photography is capable offiner spatial detail over larger scenes than commerciallyavailable digital systems. The latter need to be flown atlower elevations than conventional cameras to matchtheir resolution, with a reduction of area of coverage.Photographic film also can be scanned digitally to renderit useful in image processing of various kinds. Perhapssooner Tather than later, photography will be replacedcompletely by digital images as technology develops.One of the pressures for that is the question of stor-age. Photographs are bulky and ultimately deteriorate,whereas digital images can be kept indefinitely on op-tical storage media that progressively take up less andless space-hundreds of frames on a single CD-ROM.The other reason for a long life for the photographic,analogue image is that approaches for the interpretationof aerial photography forms the historic basis for geolo-gical analysis of most other forms of imagery.. Chapter 4focuses on photogeology as an introduction to the inter-pretation of other kinds of image.

In this chapter, Sections 3.1-3.9 give outlines of theprinciples and techniques for each major method of datagathering used in remote sensing. Section 3.10 givesinformation about the different platforms from whichimages have been acquired and the characteristics of theimages themselves.

3.1 Photography

The earliest means of remote sensing, photography, isalso the most enduring and still plays an important rolein geological applications. Of all methods of gatheringdata from a distance it is by far the most simple. All thatis required is a light-proof box, a lens and shutter system,and film coated with an emulsion sensitive to radiation.

32

~o14 ;

image

lens

diaphragm (diameter d)

shutter (speed t)

Fig. 3.1 Refraction by the glass in a lens enables rays from anobject to be focused on film through a much larger aperturethan would be possible with a primitive pin-hole camera. Moreenergy is available to sensitize the film and so expo~ure timecan be much shorter. Moreover, the aperture and exposuretime can be varied using an iris diaphragm and a timedshutter. These further refine the flexibility of a lens camera,as does the provision of films with different speeds orsensitivies to light.

In a normal camer~, objects at different distances 0must be focused on the film by changing the distancefrom lens to ffim i, governed by the relationship:

1 I focal length = 110+ Iii (3.2)

In remote sensing, 0 is of course such a large distance thatfocusing is unnecessary, the ffim being located at thefocal plane of a fixed lens system.

In all remote-sensing systems, two attributes are para-mount: the minimum energy and the smallest object thatcan be detected on an image. The two are interrelated.The minimum separation between two objects at whichthey can be distinguished-usually after enlargement-isthe spatial resolution. It is often described by the largestnumber of lines per millimetre that can be seen in theimage, which can be related to dimensions on the groundusing the optical geometry of the system. The fasterthe film, the coarser the resolution, and commonly usedfilm ranges from ISO 8 to 640, with resolutions from 250to 25 lines mm-1, which correspond to ground resolu-tions of 6-60 cm at an image scale of 1 : 15 000. In practicespatial resolution is rated using test charts with high-contrast dark and light lines at different spacings. Abetter measure is the system's response to a point sourceof energy, such as a star. This will be blurred and spreadout to a greater or lesser extent, forming a pattern knownas the point-spread function (Sectio~ 2.2, Fig. 2.8).

In photographs limits on resolution are posed by thequality and size of the lens,And are affected by defectssuch as spherical and chromatic aberration. Both affectthe sharpness of focusing. Chromatic aberration resultsfrom the variation of refractive index and therefore focallength of lenses according to the wavelength of radia-tion. It produces different focusing of the componentradiation on the film, so giving blurring. The ffim itselfis coated with grains of photosensitive chemicals in anemulsion, the size of the grains varying with the speedof the ffim. A fast ffim has coarser grains and lowerresolving power than fine-grained, slow film, but is ableto record lower energies at a particular shutter speed andf-stop. Because fast ffim is more sensitive, there must be atrade-off between resolving power and detecting power.

Black and white film depends on the photosensitivityof silver salts. When a photon strikes a grain in the ffiman electron is freed from the silver salt molecule. Itmigrates until it lodges at a defect. in the lattice, whereit converts a silver ion into a metallic silver atom. Thistriggers a chain reaction converting all silver ions inthe grain into metal. The larger the grain, the greater thechance of an encounter with a photon; that is why fastfilms have coarser grain than slow films. Developingdissolves away the unconverted silver salt leaving anegative image of black silver grains, the density ofwhich represents the energy distribution falling on thefilm. Printing is the reverse of this process. ..

In a camera, radiation reflected from an object isrefracted by the lens to form an inverted image onthe film (Fig. 3.1). The most important attributes of thelens are its focal length and its diameter or aperture (forspecial applications, such as the use of stereoscopic aerialphotographs in extracting elevation data-Chapter 8-the precise geometry of the lens, or the camera model, isneeded). The aperture is changed using a diaphragm,and so controls the radiation-gathering capacity of thelens. How much radiation falls on the film is also gov-erned by the speed of the shutter-the length of time thefilm is exposed. All cameras have shutter speeds in aconventional sequence of fractions of a second, decreas-ing by factors of two-1/2, 1/4, 1/8. ..1/250, 11500,1/1000. The energy E falling on the film during exposureis given by:

E =sd2tl4f2 (3.1)

where s is the radiant energy of the scene {W rom-1 S-1),d the aperture (rom), t the exposure time (s) and f thefocal length of the lens (rom). Only t and d may be variedon the camera. The aperture is expressed conventionallyas the ratio between focal length and aperture diameter,known as f-stop. It is arranged in a sequence of fixed set-tings that complement the shutter speed, so that a rangeof combinations of speed and aperture giving identicalexposure of the film to radiant energy can be chosen.As the energy is inversely proportional to the square ofthe f-stop, to achieve this a halving of exposure timemust be matched by an increase in f-stop of :-J2 or 1.4times. F-stops are then in the order ..J2, 2, 2..J2,22, or f/1.4,f/2, f/2.8, f/4, etc., familiar to all photographers.

34 Chapter 3

fig. 3.2 There is considerable overlapbetween the sensitivities of blackand white panchromatic films andthose sensitive to reflected infrared.To remove the swamping effect ofscattering in the blue part of thespectrum, panchromatic film used inremote sensing generally is exposedthrough a yellow, or 'minus-blue'filter. This only transmits wavelengthsto the right of its transmittance curve.To remove the effects of reflectedvisible radiation, infrared film isexposed through a dark red filter.

There are tWo main types of black and white filmused ~n remote sensing, those sensitive to radiation withwavelengths up to 0.7~, covering the whole visiblepa~ of the spectrum-panc):1romatic-and those with asensitivity that extends to about 0.9 ~m, including partof the near-infrared region (Fig. 3.2). Infrared-sensitivefilm was devised originally as a means of distinguish-ing between vegetation, which strongly reflects VNIRradiation (Section 1.3.3), and green camouflage nettingor paint, which does not.

As explained in Section 1.3.1, the blue end of the spec-trum is dominated by atmospheric scattering and haze.To improve the contrast of panchromatic photographs theyare normally exposed through a yellow, blue-absorbingfilter, and are known as minus-blue photographs. Sim-ilarly, to reduce the effects of visible light, infrared film isusually exposed through a dark-red filter (Fig. 3.2), whichabsorbs blue and green light. Examples of both types ofimage are shown in Fig. 3.3.

Whereas the human eye and colour television operate

(a) (b)Fig. 3.3 This pair of 1 : 40 000 minus-blue panchromatic (a)and infrared (b) aerial photographs are of the same area on thecoast of West Africa. They were taken under heavy cloud coverand are dimly lit as a result. The panChromatic image showsmost of the details of the township, roads and the coastal sandspit, but contains very little information relating to vegetation.The infrared image has higher contrast, and is ~n excellentmeans of discriminating vegetation, although the culturalfeatures are poorly displayed. The silt-laden waters of the

lagoon show in black, because water strongly absorbs. Thismakes the water courses more clearly visible on the infraredimage. Vegetation isa strong reflector of near-infraredradiation, and most of the variation in the fields is the result ofcrops being at different stages of growth. Individual trees,especially the mangroves near the lagoon, show up as brightspeckles: The panchromatic image in vegetated areas showsbare soil as light tones and crops and trees as a monotonousdark grey. Copyright Aerofilms Ltd.

How Data Are Collected 35

with additive primary colours-red, green and blue-colour film uses subtractive colour, for which the prim-aries are magenta, yellow and cyan (Section 2.4). Eachresults from subtracting one of the additive primariesfrom white light. A yellow filter subtracts blue, magentasubtracts green and cyan subtracts red. In colour positivefilm there are three layers of emulsion that are sensitiveto blue, green and red light (Fig. 3.4a). When developedthese emulsions retain yellow ,magenta and cyan dyes,respectively, in inverse proportion to the amounts of blue,green and red light to which they have been exposed.White light passing through the developed illm hasits primary additive colours illtered by the dye layersto produce a positive colour image. As an example,developing a film that has been exposed to blue lightbleaches the yellow layer completely but does not affectthe magenta and cyan layers. These layers illte,~ out greenand red from white light transmitted through the film,leaving only the blue component.

Colour infrared film aJso includes yellow, magenta andcyan dye layers, but each has a different sensitivity com-pared with normal colour film (Fig. 3.4b). In this case allthe layers are sensitive to blue light, as well as to green,red and near-infrared light, respectively. Removing theswamping effect of blue means placing a blue-absorbing,yellow filter in front of the lens.

The appearance of natural colour photographs needsno amplification, except to mention that their fidelity mayvary with different proprietary materials and develop-ing processes. Colour-infrared photographs, however,express infrared, red and green radiation as red, greenand blue, respectively, and so have an unusual appear-ance (Plate 3.1). They are examples of what are termedfalse-colour images. In their case, vegetation which reflectsVNIR strongly and absorbs a high proportion of visiblelight appears as various shades of red and magenta. Redmaterials, such as haematite-stained sand, reflect infraredstrongly too. The combination of weak blue, intense

(a)

Fig. 3.4 The ranges of sensitivity ofthe yellow, magenta and cyan dyelayers in natural colour (a) andcolour-infrared (b) positive filmsare different, as are the ranges oftransmi~~ce of the filters usedwith them.

36 Chapter 3

Fig. 3.5 Vertical aerial photographs of urban areas dominatedby high-rise buildings show clearly the effects of radial reliefdisplacement, and of the increase in scale with increasedelevation of the surface.

displacement of the top increases radially outwards fromthe principal point. The wider the lens' field of view andthe lower the camera height, the more pronounced theeffect becomes. From the geQmetry of Fig. 3.(:i(a) and theprinciple of similar triangles you can see that the heightof a building h relates to its distance from the principalpoint r, the radial displacement. of its top relative to itsbase d and the flying height Hby:

h=Hd/r (3.3)

green and intense red in the false colour image impartsa yellow or orange tint to such surfaces (Section 2.4).Since water absorbs infrared totally, damp soil frequentlyappears in blue or cyan hues (Plate 3.1).

The interpretation of panchromatic, infrared, naturalcolour and false colour photographs forms an import-ant topic in later chapters. However, they are limited tothat part of the spectrum with wavelengths shorter than0.9 ~m. Investigation of longer wavelengths requireselectronic image gathering devices. Before discussingthem, some of the geometrical attributes of photographsneed to be introduced.

Photographs of the Earth's surface may be taken fromany angle, but the most useful are those taken lookingvertically downwards to give a map-like view. Obliquephotographs are sometimes taken, and they have twoadvantages: they show perspective, which ~llows thebrain to estimate distance and scale; and they displayoutcrops of low-angled planar structures that are diffi-cult to pick out in vertical images because they followirregular topography closely. The rapid changes in scalein oblique photographs, however, makes them unsuit-able for transferring information to maps.

Vertical photographs normally are taken one frameat a time by a mapping or metric camera. They are builtto high precision with very high quality lenses, andare desi~ed to take photographs from which accuratemeasurements of the surface can be taken. As they areused most widely to produce systematic, overlappingcoverage of the surface, metric cameras expose film ata rate which is matched to that of their platform, eitheran aircraft or a satellite. To avoid motion blur, at eachexposure the film is drawn across the focal plane ata rate which compensates for the ground speed of theplatform. The optical system incorporates fiducial marksat the centre points of each side of the image, as a frameof reference for measurements: Images of an altimeter, aspirit level; a clock and a counter on the photograph'smargins help check on the flying height and attitude,time and position in sequence.

A vertical photograph gives a view of the Earth thatwe rarely experience, so at first sight it has an unusualappearance..Over a familiar subject, such as the centre ofa city (Fig. 3.5), one cause of this unfamiliarity is evidentat once. Tall buildings appear to lean outwards from thecentre or principal point of the photograph. Thi$radial~relief displacement is easily explained by the opticalgeometry involved (Fig. 3.6). Tracing rays from the topsand bases of tall buildings through the lens to the focalplane shows that the angle subtended by the buildingat the focal plane changes from zero when the buildingis at the principal point to a maximum at the edge of theimage. Away from the principal point the top projectsfurther from it than the base. As the lens is circular thesame effect occurs everywhere, and the amount of relative

So, it is possible to make accurate measurements ofelevation from vertical photographs, provided that h/His significantly greater than zero. That holds for mostelevations of any importance, provided that the flyingheight is not of the order of tens or hundreds of kilo-metres, when the displacement is very small. If the needis to measure cha~ges of elevation of the ground of theorderofa metre or less, as in some engineering applica-tions, then low-level flights are necessary.

Another effect of the variations in height of surfacefeatures that can be seen on Figs 3.5 and 3.6 is that theycause variations in scale on the image. The separationof the tops of buildings A and B on Fig. 3.6 is noticeablygreater than that of their bases, whereas both are equidist-ant. Other distortions result from different attitudes ofthe camera platform (Fig. 3.7) that cause slightly obliquephotographs, the scale of which changes across theimage. Scale also may change from one photograph toanother in a sequence because of variations in flying

How Data Are Collected 37

~fli~tion flight direction

~

:::~-~~~

--~ /-'.I

"

~

(a) (b)

~:~j:::::~ ~:~~=::;;;:fD(b) ~I

~c ,.--"

~

prin~ipal ~ Bpoint "'"

~

.(r

~A(c)

Fig. 3.7 Apart from the distortions in scale owing to varyingelevation of the surface, the main distortions in aerialphotographs are a result of tilting of the aircraft (a and b)so that the view is slightly oblique. A perspective effectof changing scale is introduced. Lonvection cells in theatmosphere may cause the aircraft to change acltitudeso that adjacent images have different scales (c).

Fig. 3.6 Tracing the ray paths from the tops and bases ofequally tall buildings through the lens of a camera to the film(a) shows how radial relief displacement is produced in avertical photograph. The further the building is from the pointdirectly below the lens-the nadir, which is imaged as theprincipal point of the photograph-the larger the angle itsubtends at the film. On the film (b) the top of a building isimaged further from the principal point than its base, andboth base and top are joined by a radial line to the principalpoint. The building therefore appears to lean away fromthe centre of the image (Fig. 3.5). Moreover, the scale at thelevel of the bases of the buildings is smaller than the levelat their tops.

on at least two separate photographs. This impartsparallax to objects with different heights. Adjacentoverlapping photographs represent the views by leftand right eyes, but with the eye separation increased tohundreds of metres or even several kilometres. Viewingoverlapping photographs with appropriate binocularlens or mirror systems simulates stereoscopic vision withan exaggerated three-dimensional view of the ground.Such stereopairs form the basis for both manual anddigital elevation measurements (Appendix A).

3.2 Vidicon cameras

Television originally used a camera known asa vidicon.Like a photographic camera, a vidicon gathers radiationthrough a lens system and passes it through variousfilters to focus on a flat target. Instead of a photosensitiveemulsion, the coating on a vidicon target is a transparentphotoconductive material The electrical conductivity ofthe target increases with the intensity of the illuminatingradiation. Scanning the target with an electron beam as

height. Distortions of these kinds are unavoidable andcan be corrected only by computer (Appendix B).

Another defect of photographs that is systematic-ally related to their geometry is a radial decrease inoverall brightness called vignetting. This is a result ofthe radially increasing angle between camera lens andsurface, and a corresponding decrease in the apparentaperture or light-gathering capacity of the lens. Filtersin which density increases towards their centre can com-pensate this.

Taking vertical photographs usually involves sequencesof frames along parallel flight lines. The interval betweenexposures is arranged to allow for sufficient overlapbetween images so that every point on the ground appears

38 Chapter 3

a series of lines builds up the picture. This maps thevariation in conductivity, and so the illumination, as ananalogue electrical signal. The time involved is aroundthat involved in photography, so the signal is effectivelya sn~pshot. Using the signal to modulate the intensityof an electron beam sweeping over a phosphorescentscreen reconstructs the image.

Vidicons produce only one signal to be rendered asgrey tones. A colour system needs three vidicons withdifferent filters. Separate electron beams, each focusedon phosphor grains giving red; green or blue response,make up the final image on a colour video monitor.like film emulsions, the materials used in vidicons aresensitive only to visible and near-infrared radiation.Thi~, together with the physical limitations on resolutionimparted by the size of the vidicon target and by theaccuracy of the scanning electron beam, restricts the use-fulness of vidicon cameras. To a large extent vidicons havebeen superseded in video cameras by charge-coupleddevices (Section 3.4) mounted in arrays, but the principleapptied to the early Landsat series (Section 3.12.2) andinterplanetary missions.

strips of the terrain to.be monitored. A mirror directsradiation on to the detector and a motor rotates themirror back and forth through a small angle in arepeated cycle. If the sweep is perpendicular to the plat-form's motion it gathers data from immediately belowand to either side of the flight path. This builds up animage from straight swaths or lines orientated across theflight line by matching the frequency of sweeps to theground speed of the platform.

Direction of radiation by the mirror on to banks ofdetectors occurs after splitting it into a series of wave-bands using diffraction gratings. This means that thedetectors sense all the chosen wavebands simultaneously(Fig. 3.8). Consequently, the response of each det~ctorcorresponds to exactly the same portion of terrain as allthe others, at precisely the same instant. This ensuresthat the data for each waveband are in exact register.To improve resolution one sweep of the mirror directsradiation to several detectors for each waveband. Theswath of firiite width on the ground comprises severalstrings of signals, each of which represents a smaller linewithin it.

Lin~scannex:s escape the spectral limitations (between0.4 and 1.1 ~m) of photographic and vidicon methods.Radiation-sensitive detectors now span the spectrum fromultraviolet through to the region where energy emittedby the Earth reaches a peak (Fig. 3.9), although no singledetector covers .the whole range. Those detectors forradiation beyond 1.1 ~m give measurable responses onlywhen cooled, either by liquid nitrogen or by exposure tothe very low temperatures of outer space.

3.3 Line-scanning systems

A.similar effect to the raster from a vidicon comes fromusing a single photosensitive detector cell to scan radia-tion emanau~g from a series of strips of the Earth'ssurface. One means of doing this involves a line-scanner.The forward motiQn of the platform allows successive

dispersingelement

rotating scanmirror

.I //-

entranceslit

~~

/ /

individualdetectors

Fig. 3.8 Radiation gathered by a scanmirror in a line-scanner is focused byan optical system through a dispersionsystem, either a prism or a diffractiongrating, on to banks of detectors.There is at least one detector for eachspectral band being sensed, and inmost cases there are several so thatone sweep of the mirror across thesurface produces a number of linesof data for each band.

~

How Data Are Collected 39

siliconlead

sulphide indiumsulphide mercury

cadmiumtelluride

15.01.1 .2,5 7.0

Wavelength (IJm)

Several photoelectronic detectors cover the whole

At

is:

(3.4)

histhe

energy. This generally a~pears as speckles on an imageof an otherwise uniform surface. In practice the per-formanceof a detector is assessed by its signal-to-noiseratio (SIN). Noise is generally constant and so there is amihimum signal, below-which distinction is not possible.The signal-to-noise ratio at a particular wavelength (S/M..is affected by many variables:1 quality of the detector at that wavelength (Q.t>;2 IFay (1);3 width of the waveband being detected (AA);"4 the spectral radiant flux of the surface (E.t>;5 time taken to sweep the portion of surface correspond-ing to a pixel (t).The variables express (SIN).. empirically by;

(SIN);"=Q;"xI2xAAxE;,,xt (3.5)

This empirical relationship underlies .the design of line-scanners.

Several options help improve, i.e. increase the signal-to-noise ratio, withouthaving'to develop a better qualitydetector. These are: increasing the IFay (coarsening thespatial resolution), widening the waveband, or increas-ing the time over which energy is collected for each pixel(slowing the scan speed of the mirror, or the velocity ofthe platform). Narrowing the waveband 'to detect smallspectral features is possible only by increasing the IFay,or by increasing the time taken to sweep each pixel. Thisis important because many of the spectral features ofinterest shown in Chapter 1 are extremely narrow, and asmall AA is needed 'to resolve them.

Clearly there are trade-offs in trying to 'improvespatial and spectral resolution while keeping an accept-able SIN. The most important limiting factor is the timedwelt on each pixel, because this involves the velocityof the platform and the need to cove~ all parts of the sur-face. Spacecraft have a fixed orbital velocity, dependingon altitude, and there is a lower limit for aircraft too-their stalling speed. Moreover, the performance of thedetectors is not constant with wavelength. The spatialand spectral resolutions, and quality of data from opera-tional line-scanners have to be optimized.

There are two methods of recording line-scan data, inanalogue or digital form, either on magnetic tape aboardthe platform or after telemetry down to a ground receiv-ing station. In the analogue case the signal modulatesa light source that moves in lines across a photographicffim, so building up an analogue image in the form ofa strip. More usually the recording is in digital format.The first step involves dividing the continuous signalfrom the detector into discrete blocks, either mecha,nicallyusing a rotating shutter blade, or electronically usinga timer. In either case the block of signals correspondto finite lengths of recording time and thus to distancesalong the scanne.d swath. The signals in each block areaveraged, rescaled and converted" to a digital number

area, the time taken to focus energy fromit on the detectors and the size of the scanning mirrordetermine the total energy that is available for detectionat all wavelengths. The system's ability to discriminatedifferent energy levels is controlled by the sensitivity ofthe detectors. Sensitivity forms the main practical limiton resolution. Whereas it may be technically feasible tohave a very small IFOV, the detectors would show onlybroad variations in radiance from the surface. The imagewould be of poor quality. The broader the range ofwavelengths directed on to a detector the more energy isavailable for it to detect. So for a small IFOV, the broaderthe waveband, the better is the discrimination of differentsignals from the surface. With a large IFOV, a narrowerwaveband can be monitored and still discriminate dif-ferent si~als from the ground.

Any detector experiences random fluctuations causedby electronic or structural defects in its desi~ or powersupply. To a greater or lesser degree all produce aresponse to this noise in the absence of any incident

40 Chapter 3

(b)

(PN), which is recorded. Ea<;:h ON represents the radia~tenergy in a particular waveband associated with a pixelon the ground. An image recorded in digital formattherefore comprises a geometrically precise array of pixelsin lines that is identic~l in structure for each waveband,so that they can be registered exactly. This array, or rasterlends itself to both computer manipulation and to video

display.The size of each pixel is a function of IFaV and flying

height. Although th~ resolution of the system is definedas the finest spacing of objects on the ground that canbe discriminated in an image, it is the pixel size thatis quoted most commonly as 'resolution'. Whether anobject is detectable or not depends on its size, its con-trast with its surroundings and its shape and orientationwith respect to the raster pattern. Figure 3.10 shows thatpixels often overlap several objects or classes,of gro~ndC9ver. The response for each of these mixed pixels hasa contribution from each enclosed object or part of anobject, depending on their size and contrast relative

.~~

+[1--1 I I 1 1 1

1 II

altitude variation

(c) (d)

pitch variationDA B c

~

yaw variation

~

:ta~~

"\~ "\

1 (e) (f)

r \.-:_--~J3 I."f1 .

1L- .roll variation

2

(g) (h)

3

platform velocity ~

Fig. 3.11 Because the distance on the ground covered bythe same angle subtended at the scanning mirror increasesaway from the platform (a), all line-scan images decreasein scale away from the ground track of the platform (b).The scalealongrhe ground track does not change, sostraight ~ear features running diagonally across thesurface are distorted in a characteristic sinusoidal fashionon the image (b). Tall objects appear to lean away from theground track and the shapes of objects become distorted.As well as this effect, distortions resulting from instabilityof the platform result in the rectangular format of an image(dashed outlines) representing more or less irregular ordisplaced areas of the surface (c tog). In the case of satellites,the Earth rotates beneath the platform as it passes over.As a result the area imaged as a series of lines scanneddUririg a period of time represents a parallelogram on thesurface (h):

Fig. 3.10 A series of 30 m pixels superimposed on the ~oundsurface shows that most pixels include several objects andsurface categories. Only pixels AI, BI and D2 are 'pure'.The rest are mixed pixels. If the farm had a sufficiently highcontrast with grass its effects on pixel B2 could be highenough that the pixel had a different jJrightness from thosesurrounding it. The farm might be detectable, but it could notbe recognized as a farm. Similarly the road could affect pixelsA3, B3,C2, C3and D3,andprobably others beyond the map.Because road-containing pixels would be linked,.however,the crude shape and orientation of the road could be seen.The track may be too narrow to have a noticeable effect.Pure luck has resulted in the boundary between the ploughedfield and grass falling along a boundary separating four pixels.On the image it would be clear and represented accurately.All the other boundaries fall within pixels. Although theywould contribute to the brightness of these pixels, on theimage the bound!lries would become steps in the brightnessbetween pixels dominated by large classes. Their position andorientation would become 'blurred' by the rectang\llar raster.

Earth rot

7

ation

How Data Are Collected 41

lines not parallel to scan tines or thegroundtrack take ona curved shape as a result (Fig. 3.11). As the separationbetween sensor and ground increases along the scanline, so the resolution becomes coarser towards theimage edge. The higher the platform and the smaller thesweep angle, the smaller these distortions become, andwith some satellite systems they are barely noticeable.

Distortions resulting from platform motion are pre-sent in all line-scan images to a greater degree thanin near-instantaneous photographs, because they arebuilt up over a relatively long period of time. Platformaltitude and flight direction may vary. There can also beroll, pitch, yaw and heave. The Earth rotates beneathsatellites while the image is being scanned, so that a rect-angular image is really a parallelogram on the ground(Fig. 3.11h). Analogue images (Fig. 3.12) can rarely becorrected for such distortions. Digital images, however,permit computerized registration to maps (Appendix B).

to their surroundings. An object smaller than a pixeltherefore may be detectable if it has a high contrast. Thisdoes not mean that it can be recognized, however, forrecognition depends on shape, and shape is masked bythat of the pixel itself.

Like vertical photographs, line-scan images suffer froma variety of distortions, resulting from both the opticalsystem and the motion of the platform. Whereas opticaldistortions on photographs increase radially outwardsfrom the principal point, those on a line-scan imageincrease outwards from the ground track along each line.This is because the mirror changes the viewing angleof the ground continuously during each sweep. Bothscale and geometry change away from the groundtrack,so that tall objects lean in parallel away from the mid-lineof the image, and scale decreases in the same fashion.Because of varying lateral scale, the apparenldirectionof ground features changes across the image. Straight

Fig. 3.12 This line-scan image of thermally emitted infraredwavelengths over a power station in Nottinghamshire, UK,shows distortions of the type featured in Fig. 3.11(b). They are

particularly noticeable for the road and the array of coolingtowers. The light specks at lower left are cattle.'CopyrightAerofilms Ltd.

42 Chapter 3

The final point about defects in line-scan images stemsfrom their structure. Any fluctuation in the responseof an individual detector or miscalibration betWeen de-te<;:tors in an array for one waveband results in stripingparallel to the scan lines. Sometimes, a detector may failtemporarily, or its response may go out of phase with theothers in an array. These failures produce lines with nodata, and pixels displaced relative to those either side,respectively. Random electrical discharges in the systemmay also introduce spurious numbers into an image,which show as bright pixels that relate to nothing on theground. Appendix B covers various defects and theirremoval in more detail.

3.4

Pushbroom systems

Line-scanners contain mechanical components thatsometimes fail and have 'a poor signal-to-noise ratio.Most important, they cannot have both high spatial andhigh spectral resolutions,' because they dwell on eachpixel during a sweep of the mirror for a very short time.Tiny «10 mm across) radiation-sensitive cells, calledcharge-coupled devices (CCDs) help solve the last prob-lem; Charge'-coupled devices use the same materials asthose shown in Fig. 3.9. B~cause1hey respond to wave-lengths up 10 about 2.4 ~m, CCDs can monitor the fullvisible and near:.infrared (NIR) range in which manyspectral features of interest occur. A CCD becomes chargedin proportion to the incident radiation, and dischargesvery quickly to produce either an analogue or digital

signal. In effect, a COD is a light-sensitive capacitor.A rectangular array of CCDs forms the light-gatheringdevice in both video and digital cameras. Arranged inthousands along a linear array on to which an opticalsystem directs radiation from the ground, no movingparts are required to build up an image from a movingplatform. The array simply and directly measures radia-tion from pixels along a ground swath swept by it. Sucha device is known as a pushbroom system-it works inan analogous fashion to the pushing of a wide broomacross the floor. Pushbrooms build up images of success-ive lines perpendicular to the flight path, each columnof pixels corresponding to the response of each CCD inthe array (Fig. 3.13a). In addition to their advantagesof simplicity together with direct measurement fromgeometrically exact pixels, pushbroom systems are reli-able and require little power. Th~ SPOT pointable push-broom (Section 3.12.3) can image areas to either side ofits ground track by using a variably orientated mirror(Fig. 3.13b). The spatial resolution of a pushbroom sys-tem is a function of the siz~ of each CCD, the optics andthe flying height. Its spectral resolution, like that of aline-scanning system, depends on the time during whicheach CCD gathers radiant energy.

A satellite carries two imaging systems, one a line-scanner, the other a pushbroom. Each produces imageswith pixels that are 30 m square arranged in lines3000 pixels long. One sweep of the line-scanner's mirrortakes a time t, so the time dwelt on each pixel is t/3000.The pushbroom captures "all 3000 pixels in the sameline at once, in time t. So a pushbroom's CCDs gather

(a) (b)strip-selection mirror CCD detector arrays

array ofdetectors

6000

scanning bysatellite motion

/"

/'om60km(6000 x 10 m)

Fig. 3.13 A pushbroom system produces an image by usingthe motion of the platforin to sweep an array of charge-

coupled devices (CCDs) across the surface (a). Radiation isdirected on to the array by a mirror (b), the angle of which cart

oblique vertical obliqueviewing viewing viewing

be changed to allow imaging ahead of the platform or to oneside. This enables stereoscopic images to be acquired and,in the case of sideways direction allows the same area to beimaged on several separate occasions without direct overpass.

",-How Data Are Collected 43

radiation for 3000 times longer than do the detectors in aline-scanner. This gives three advantages for pushbroomsystems. First, their signal-to~noise ratio is,much better,and so they can monitor lower radiation levels. Second,they can detect smaller pixels, by increasing the numberof CCDs in the pushbroom and by reducing the time ofdata collection for each line, thereby making it narroweron the ground. Thirdly, it becomes possible to r~duce theterm ~)., in Equation (3.5). In other words, the spectralresolution can be improved to match the narrownessof important spectral features. The design of imagingspectrometers (Section 3.6) exploits this last option.

Table 3.1 CQding Qf divi~ons in the microwave region usedby radar.,systems

Band code Wavelength (~m)

KaK

~XCSLP

0.8-1.11.1-1.71.7-2.42.4-3.83.8-7.57.5-1515-3030-100

3.5 Microwave imaging

The longest wavelengths used in remote ~ensing aremore than a thousand times longer (1-25 cm)than thoseaccessible to line-scan systems « 15~m) This is themicrowave region of the radiation spectrum. All mater-ials above absolute zero emit microwaves, but for theambient temperature of the Earth's surface (300 K) theenergy curve (Fig. 1.3) shows that energies in this regionare low. Gases and clouds hardly attenuate microwaves,however, and they can pass through several metres ofdry solids. Detection of natural microwaves from theEarth, like most other techniques, is a passive method.Microwaves also can be generated electronically usingklystron tubes, when coherent radiation of a singlewavelength is produced. Detection and analysis of thisartificial radiation when scattered back to a detectorby the surface is therefore an active method, originallytermed radio detection and ranging, or radar.

Passive microwave remote sensing collects data usingan antenna, similar to a radio-telescope. Most surveysproduce profiles of microwave emittance along lines,r~ther than images. Images, in the familiar form of line-scans, become possible if the antenna is swept from sideto side beneath the platform. These scanned imagessuffer from the same distortions as those of line-scansystems operating in the visible and near-infrared regions,but their spatial resolution is very much coarser, andeach wavelength requires a separate antenna.

The IFOV of a passive microwave system is expressedas a solid angle (8), and is given by:

8= 1.2A./D (3.6)

where A. is wavelength and D is the diameter of theantenna. For microwave wavelengths and fine resolu-tion D is too large to be portable, and except for verylow flying heights the spatial resolution is coarser thanwith other imaging systems. For this reason, there havebeen few experiments in geological applications thatusepassive microwave technology. Its main use has beendirected at climatology and oceanography.

During World War II, when radar was being developed,security meant giving letter codes to dt££erent microwavewavebands (Table 3.1). Those used most colpmonly areKa' X, C and L bands.

The operation of active microwave or radar imagingsystems differs fundamen~lly from that of all other sys-tems. Microwave radi~tion 'illuminates' the surface witha downwards look direction that is to the side of the plat-form. It is sideways-looking radar. Producing imagesdoes not involve using an ante~a to: sweep a groundare~; that would limit the ~ize of the antenna and thusthe resolution. Instead, a fixed antenna detects returningradiation from a long, narrow strip perpendicular to theflight direction of the platform (Fig. 3.14a).

To avoid interference between the et:nitted radia-tion and that returning fr6m the surfac~, illuminationinvolves pulses. Timing of the pulses is crucial, so thatall radiation returning from the surface arrives beforeanother transmitted pulse. The timing depends on themaximum distance to the side, or the range, from whichreturns are expected. The frequency of the pulses isinversely proportional to the range, because the furtherthe range, the longer returns take to reach the antenna.Because radar travels at the speed of light, the timing isin microseconds.

The response of the antenna to energy returning fromthe surface from each pulse comprises both a measureof the returning energy and the time elapsed since thepulse. This time (t) is a measure of -distance to the surfacefrom which energy was returned. The distance is twicethat from the platform to the object (the slant range). Asradar waves travel at the speed of light (c), the slantrange is ct/2. As Fig. 3.14(a) shows, the distance on theground or ground range (r) is slant range multiplied bythe cosine of the depression angle (IX) for the radar beamat that point:

r = (ct cosa)/2 (3.7)

The depression angle decreases as range increases.A line on a radar image is ther~for~ a plot of ~n~rgy

,;:::ith time, r~sulting from th~ outward propag.ation of a

44 Chapter 3

(a)

Fig. 3.14 (a) The geometry involvedin sideways-looking radar governsmany of the intrinsic properties ofradar images. The depression angle(a) is the angle between horizontaland a radar ray path. The look angle,sometimes quoted instead, is simplythe angle between vertical and a raypath (90°-a). The incidence angle isthe angle betweeQan incident radarray and a line at right angles to thesurface. For a horizontal surface it isthe same as the look angle, but varieswith the slope of the surface. Thebeam width (angle /3) determineshow the illumination of the surfacespreads out from near to far range.The slant range is the direct distancefrom the antenna to object, and isrelated to the true or ground range bythe clepression angle (Equation 3.7).(b) The solid curves on this cartoonare spherical radar wavefrontsemitted by the antenna, and theirnumbers indicate the time since theywere emitted. The dashed curves areradar wavefronts back-scatteredfrom the house and tree. They havethe same number conventionindicating time. The wavefrontfrom the house which has justbeen received at the antenna hastravelled for 17 time units, thatfrom the tree for 13. The graph(c) shows how radar energyreturned from the house andtree is recorded at the antenna.

(b)

(c)

high energyoutput pulse

return from house

!"

!UCCI~

return from tree

t., ~ )---!" '! -I r I J I T =i=r==, I '-;-1'" II '"'-

0 2 4 68 10 12 14 16 18

Time (number of pulses)

i~ the range direction. To distinguish two objects onthe ground, the returned signals from each object for thewhole pulse must reach the antenna at different times.Any overlap between the two sets of signals means that'the objects merge into one on the image. As Fig. 3.15shows, the minimum separation between the objects,along the slant range must be half the pulse length, orc'r/2. This limiting distance, the slant-range resolution,is independe~t of distance from the platform. The cor-responding distaI:tce on the ground, the ground-range

Isingle... pulse and jts return from objects on the ground(Fig. 3.14b). With the next pulse the platform has movedforwards, to acquire another line of data. At its mostsimple the image is built up by modulating the intensityof a single line cathode-ray tube with the energy-timesignal. Exposing film to this at a rate proportionat to theplatform speed builds up a strip-like image.

Resolution on a radar image depend~ on two para-meters:. pulse length and beam width. The length of~ach pulse Jtime 'f/ distance C'f) controls the resolution

How Data Are Collected 45

(a)antenna

'y-~

~'--\~

FB Fc

,

'\A

'-' R;~

1/l!c"""!",, 'i ,..",t,;:I'i Fif L/2 "",,/

"'"

Fig. 3.15 In (a), the surface isilluminated by pulses of radar fromthe antenna (grey) of a fixed durationr. The pulse length L is equal to cr.The fronts of back-scattered returnsof a pulse from objects at A andB;are at FA and FB' and the rears atRA and RB. The signals that thesereturned pulses generate at theantenna (b) overlap so that they arejust indistinguishable, and form asingle broad peak. Pulses returnedfrom objects with a closer spacingthan A and B would overlap evenmore. At spacings greater than AB,however, return signals are separate(c) and the two objects, such as Aand C, can be distinguished. Thelimiting slant range correspondingto the separation between A and Bis half the pulse length L. This isthe slant-range resolution. In theground range this resolutionbecomes L/2 cosa.

~

(b)

~

-=--~,=-c:~-~-~"~

FA Fa RA Hs

resolution (R.)' however, varies with the depression angle(a), so that:

Rr = (C'f cos a) /2 (3.8)

and it increases away from the platform. However,it is theoretically independent of the altitude of theplatform.

The width of the radar beam is expressed by an angle(/3) (Fig. 3.14a), and is directly proportional to the radar

.wavelength ().,) and inversely proportional to the length

.~f the antenna (L). The width of the line on the groundilluminated by each pulse depends on the distancetromthe platform. As distance increases, so the beam fans out.The resolution parallel to the ground track of the plat-form, the azimuth resolution (Ra) is the width illumin-ated by each pulse. Azimuth resolution deteriorates indirect proportion to the ground range (r)

Ra"= rf3= ()./L)(cTcosa)/2 (3.9)-Each line on an image therefore represents data thatbecome increasingly degraded away from the platform.The azimuth resolution also deteriorates as the altitudeof the platform becomes greater. So a satellite radarsystem based on sensors developed for aircraft wouldhave equally good range resolution, provided the powerwas high enough, but hopeless azimuth resolution.There are, however, ingenious ways of avoiding thislimitation. ..

The most obvious way to decrease the beam width isby extending the antenna. From an altitude of 100 kinan azimuth resolution of 100 m requires a beam widthof 1 ~rad. For 25 cm wavelength rad;ar, substituting inEquation (3.9), shows that an antenna 250 m long willbe required. Clearly, there are physical limits using this

46 Chapter 3

achieved with the same antenna used in a brute-force system.The further down r~nge an object is, the more times it canback-scatter successive radar pulses as the platform flies by.Thus the discQrnination of ahead, in-line and behind positionsrelative to the moving platform is independent of range.Azimuth resolution theoretically is constant in SAR images.In practice it is degraded with increasing range because theradar energy becomes weaker with distance as a result ofbeam spread. Eventually, energy back-scattered from thefurthest range becomes too weak to be detected.

Fig. 3.16 In a synthetic aperture radar system interferenceoccurs between a reference signal and the increased 9rdecreased frequencies of radar waves returning fromilluminated areas ahead of and behind the platform. Wavesreturning from a narrow strip at right angles to the flightpath do not suffer a Doppler shift, and no interference occurs.This allows returns at any instant from the in-line strip to bediscriminated from those from the rest of the illuminatedsurface dUring signal processing. This makes possible a muchfuter resolution in the azimuth direction than could be

solution. Radars for whichbeamwidthis controlled bythe length of their antennae are known as real-aperturesystems. they have simple designs and recording sys-tems, as outlined earlier, but they are useful only whenmounted in aircraft. Real-aperture radar is now obsolete.More complex systems use the coherent nature ofartificial microwave radiation to simulate an extremelylong antenna. They are called s~thetic-aperture radar(SA:R) systems, and are the only useful means for high-altitude and satellite-radar remote sensing.

-Within the ground area illuminated by the pulsedradar beam (Fig. 3.16), parts of the surface are ahead ofthe platform, parts behind, and only a narrow strip, justas wide as the antenna itself, is in line at any instant. Thereturns from the areas ahead and behind are subject toa Doppler shift, just as sound would be. Imagine a trainwith its horn sounding. Sound echoed from ahead wouldhave a higher pitch to an observer on the train than wouldthat echoed from behind. Only echoes from the sidewould sound the same as the horn. The Doppler effecthas altered the wavelength of the echoed sound. Exactlythe same thing happens to radiation, including radar,although the shift is very small. Because radar is coher-ent thes~ small differences can be detected electronically.

A SAR system records the time, the energy and thefrequency of radar returns from the surface, with dif-ferent frequencies signalling the position of the objectresponsible for a frequency shift relative to the motionof the platform. The frequency is measured by allowingthe returns to interfere with a reference signal with a

frequency that is the same as that of the radar pulse itself.Those frequencies affected by Doppler shift produce inter-ference patterns. A constructive interference produces ahigh signal, a destructive interference gives a low signal.These can be recorded opticany as bright and dark spotsarranged in lines parallel to flight direction, or digitallyfor computer processing. Each object on the ground,defined by the time of the response for successive radarpulses, is represented by several spots, from when it wasahead, in line and behind the moving platform. Owingto the beam spread, the further an object is from theplatform the more times the electronics 'look at' it. Thattransforms the moving real antenna into a much longer,synthetic antenna, thereby improving the azimuth resolu-tion. Because the spread of the beam illuminates far-rangeobjects for longer than those at near range, azimuthresolution is independent of range. That electronic trickmakes SAR imaging possible from satellites, subject onlyto the constraints of weight, power requirements andthe rate of data transmission. The last is enormous bycomparison with data from optical systems, because ofthe great complexity of 'raw' SAR data.

Recording SAR signals on film produces a one-dimensional interference, in the form of a radar holo-gram (Fig. 3.17a). By itself it is meaningless, but whenilluminated by a laser it is optically correlated to producea radar image (Fig. 3.17b). Digital SAR signals can beunpacked by very powerful software to produce a digitalimage in raster format, which suffers none of the degrada-tion inherent in optical correlation.

How Data Are Collected 47

(b)

Fig. 3.17 The raw data from a synthetic-aperture radar systemis expressed on film asa hologram (a), which cannot beinterpreted until it has been optically correlated.Th~patternson the film represent interference between the Doppler-shiftedfrequencies of back-scattered radar waves and a reference

frequency generated by the system. Projection of a laserthrough a SAR hologram using a complex lens systemoptically correlates the interference patterns Qn the hologramto produce an image of the surface (b). Courtesy of EricKasischke, Environmental Research Institute of Michigan.

in slant-range images. A hyperbolic transformation ofthe data down range converts them to a nearly uniformscale, which equates that of a flat surface, th~reby pro-ducing a ground-range image.

Surfaces with high relief produce a strange effect. Thetop of a mountain may be nearer to the platform thanits b~se. Because radar produces time images, the topappears closer to the platfQrm than the base (Fig. 3.19).The mountain appears to 'lean' tow~rds ,the platform,giving a phenomenon termed layover. Many radar imagesof mountainous terrain show arrays of layovers (Fig. 3.20),which resemble the flatirons nQrmally associated withdipping strata (Chapter 4). They have no relationship todip and are purely artificial. The layover effect dependspartly on the relief, but mainly on the depression angle.For high depression angles-when the surf~ce is morenearly below the platform-the effect is worst. With ashallow depression angle the effect is noticeable only forthe strongest relief. The effect at its most bizarre can dis-playa mountain overlaying a river or glacier (Fig. 3.22b).

Radar transmission can be in different modes ofpolarization, with vibrations restricted to a vertical orhorizontal plane, as can receiving the returns. Such con-figurations enable the rotations induced by reflectionsfrom different types of surface materials to be analysed.This and other surface-dependent features of radar imagesform the central part of Chapter 7. We turn now to theappearance of radar images.

Because radar looks sideways, and as radar imagesrepresent energy and time, it produces unusual geo-metric features as well as those resulting from platforminstability and Earth rotation. Times recorded in a radarimage represent direct or slant distances of objects fromthe platform, and each surface object subtends a differ-ent angle to the platform. As a result, uncorrected radarimages vary in scale along range. They are compressedat near range and only approach true scale at far range,the opposite of line-scan images. This results from thedifferent incidence angles of radar wavefronts at dif-ferent ranges (Fig. 3.18) and is an inherent distortion

Fig. 3.18 In a slant-range image thescale down-range increases becausethe depression angles of sphericalwavefronts which meet the surfacedecrease away from the antenna. Theapparent distances between objectsand between parts of the same objectare measured in time. The time takento illuminate A is shorter than for Band C. Likewise it takes longer forawave to move from B to C than fromA to B. The equal distances on theground increase away from theplatform on the slant-range image.

T~flying \height \

H -..j , appearance..jC' I- on slant-

/ ~ ~ange radarImage

I,,/

:!

48 Chapter 3

E~&~SRt-

\ "SRfS R b """

\

SRt< SRf< SRb

'"

(b)

Fig. 3.19 Because the whole of its slope that faces the platformis at a steeper angle than a radar wavefront, the top of themountain is illuminated by the wavefront before its flankand base. This is translated into slant ranges-S~, S~ andSRb -which produce an image \Vhere the top is nearer to theplatform than the base or flank. The mountain appears on theimage to lean towards the platform.

Fig. 3.21 Where the surface slope is at a lower angle thanradar wavefronts (a) the slant ranges of top, flank and baseof a mountain are less than their true ranges on the ground.The result is that the slope facing the platform is compressedor foreshortened. Sometimes the surface has approximatelythe same angle as a wavefront (b). In this case the slant rangesfor all parts of the surface are the same and the whole slopeis compressed to a single line.

a single point (Fig. 3.21b). Examples of both phenomenacan be seen in Figs 3.20 and 3.34(a).

The most striking feature on radar images of variablerelief is their sidelit character (Figs 3.20 and 3.34a). Slopesfacing the platform are most strongly illuminated andreflect most of the energy back to the antenna, so show-ing up as bright. Slopes facing away receive least energyand appear darker. Where such slopes are steeper thanthe radar ray paths, then they are in radar shadow,receive no energy and appear totally black (Fig. 3.22).The farther the range, the lower the depression angleand the lower the relief needs to be to produce a radarshadow.

Where depression angle varies greatly from near tofar range, as in all images acquired from aircraft, lay-over, foreshortening and shadowing vary markedlyin the range direction (Figs 3.20 and 3.34a). This canbe so grotesque in areas of only moderate relief thatsuch images are of little use for detailed interpretation.One escape from this problem is to mosaic strips withthe same depression angle from overlapping swaths. Abetter solution is to make images from such an altitudethat only a small range of depression angles covers avery broad swath on the ground. The best images arethose from orbit. Despite their irritations, these peculi-arities inherent in radar image have one very usefuladvantage. Differential relief displacements caused bydifferent depression angles and look directions causeimage parallax. That, as in overlapping aerial photo-graphs,. enables stereoscopic viewing (Chapter 7 andAppendix A).

Fig. 3.20 Layover dominates both the ne~r- and far-rangeparts of this radar image ofa mountainous area inJapan. Itta:kes the for:m of arrays of triangular fea~res that look likeflatirons, not to be confused with real 'flatirons' which are socharacteristic of eroded dipping strata on aerial photographsand other images in the visible and infrared. Note how someof the mountains appear to 'lie over'drainage courses.CourtesyNASDA, Japan.

Related to layover is another effect typical of radarimages: foreshortening. This occurs where the surfaceslope is less steep than that of the radar wavefront. Thepulse reaches the base before the top, so that the slopingsurface appears shorter on the image than on the ground(Fig. 3.21a). ~ere the surface is parallel to the wave-front all points on it are exactly the same slant distancefrom the platform, so that';ll fall together on the image as

How Data Are Collected 49

=--~~varying depression angles

~ radarwavefront

for top andbase

"\" "

ground-rangeimage T B BT B T

layover of foreslope point image foreshortening of foreslopemedium return from of foreslope. shadow on backslopebackslope weak return

from backslope

slope away from the platform receive less energy and soalways appear darker on a radar image. If their slope is greaterthan the depression angle they receive no energy and areiI1total shadow. As depression angle decreases towards the farrange, this part of an image is more likely tocontaiJ:l shadowsthan the near range. Figure 3.21. shows examples of layoverand foreshortening from near torar range.

Fig. 3.22 This diagram summarizes the peculiarities of radarimage geometry by showing the effects on mountains with thesame shape at different ranges. Layover dominates slopesfacing the platform at near range, and foreshortening fallsfrom a maximum at medium depression angles to lesssignificance in the far range. Wherever they are, slopes steeperthan wavefronts always suffer layover, whereas shallowerslopes are always foreshortened to some extent. Surfaces that

The most revolutionary outcome ofCCD arrays is thedevelopment of imaging spectrometers. Instead of pro-ducing a few broad spectral bands for each pixel in animage, they capture hundreds of very narrow bands inimage form. One method uses a long linear array of CCDs,very similar to that in the pushbroom device describedin Section 3..4, but having CCDs sensitive to differentparts of the spectrum arranged along it. A mirror focusesradiant energy through a diffraction grating, so thatthe resulting spectrum spreads along the CCD array.This combination of line-scanning with a long CCDarray is a whiskbroom imaging spectrometer. By com-bining several linear CCD arrays, each tuned to adjacentnarrow spectral bands, in an area array n bands long bym elements wide, an even simpler pushbroom imagingspectrometer is possible. This operates in exactly thesame general fashion as the system shown in Fig. 3.13.

The many spectral channels, combined with millionsof pixels in images from imaging spectrometers, poseproblems of data storage and means of analysing thedata. Some of these techniques are covered in Chapter 5.

3.6 Imaging spectrometers

The spectral curves shown in Chapter 1 are products ofspectroradiometers. They direct radiant energy from asurface through a diffraction grating to spread the radia-tion into its constituent spectrum. Detectors appropri-ate for different wavelength ranges scan the spectrum,recording radiances for very narrow wavebands, so con-structing accurate spectral curves. Although useful inestablishing the spectral characteristics of materials, suchinstruments are not appropriate for mappipg purposes.Although it is possible to carry spectroradiometers onremote-sensing platforms, the various factors discussedin Section 3.3 around Equation (3.5) limited them untilrecently to only a few narrow wavebands in producingprofiles along flight lines. Carefully chosen bands couldallow important information on the distribution of dif-ferent minerals to be produced, but such systems wereunwieldy and limited in the data that they produced.

The development of CCDs sensitive across the wholereflected region of the spectrum, and the ability to mini-aturize them in large arrays, allows much more flexiblesystems. Exposing such an array, with different CCDelements 'tuned' through appropriate optics to adjac-ent narrow wavebands, produces detailed spectra forsmall areas along a profile. Plate 3.2 shows an examplethat has been enhanced digitally in order to empha~izespectral features.

3.7 Gamma-ray spectrometers

The naturally occurring unstable isotopes 4OK, 23%, 235Uand 23BU all emit gamma-rays when they decay. Eachisotope (more precisely, various daughter isotopes intheir decay schemes) emits a range with discrete energy

50 Chapter 3

cover absorb the radiation, thereby attenuating it. Forterrestrial uses only aerial surveys, flown at ground-hugging altitudes, are capable of producing useful data.For airless bodies, such as the Moon, gamma-ray surveysfrom orbit have allowed the extrapolation of a tiny num-ber of geological observations by the Apollo astronautsto much of its surface.

3.8 A short history of remote sensing

Remote sensing depends on the ability to get suffici-ently far from the Earth's surface to obtain a worthwhileview and the invention of means of capturing an image.Although the first staffed balloon flight was in 1783,daguerreotype photography appeared only in 1839. It wasnot until 20 years later still that these two achievementswere combined, when Gaspard Tournachon ascendedin a balloon and photographed a village near Paris.

Military applications have always dominated thedevelopment of remote sensing since then. During theCivil War (1861-65) in the USA, the Union forces photo-graphed Confederate d~fensive positions from captiveballoons, and using remotely operated cameras flownon kites. Attempts were even made in Germany to usedelayed-action miniature cameras carried by pigeons.During World War I (1914-18), aerial photographic recon-naissance using powered aircraft systematized aerialphotography. That enabled its first significant uses incartography, forestry and ~ology during the 1920s and1930s. The first major geological applications were for oilexploration in what was then Persia, by the Anglo-PersianOil Company. Aerial photo-reconnaissance dominatedplanning and execution of many operations duringWorld War II. However, three new developments wereachieved under the spur of military necessity. Radarwas developed, together with VNIR-sensitive film forcamouflage detection and trials were made of thermal-infrared sensing devices.

During the 1950s, colour infrared photography founduses in vegetation studies, side-looking airborne radarwas refined and synthetic-aperture radar became pos-sible with the development of a successful optical pro-cessor. In the same decade photographic technologyand interpretation skills continued to develop, with theclandestine flights of US Air Force spy planes such asthe supersonic U2 over Soviet and other territory.

Since the early 1960s there have been large numbers ofmilitary surveillance satellites in orbit. Many of th~ earlierones recorded images on film., which was recovered onre-entry of the platform, ejected in canisters to be scoopedup by recovery aircraft or located electronically on theground. Increasingly, however, intelligence gathering hasmoved towards electronic means using line-scanners andnow pushbrooms, depending on ever increasing spectral~

levels that a gamma-ray spectrometer can detect andmeasure. Such instruments depend on the emissionof bursts of light by certain crystals, such as sodiumiodide doped with thallium, when they absorb gamma-rays. The intensity of the light is proportional to theenergy and inversely proportional to the wavelengthof the gamma-ray photons, whereas the frequency oflight pulses is a measure of the amount of radioactivity.By filtering the discrete energy levels and measuringpulse frequency for narrow energy bands it is possible toestimate the proportion of each isotope in the surfacewhich emits the gamma-rays.

In practice a gamma-ray spectrometer comprises avery large crystal surrounded by photomultiplier tubesand appropriate shields, which amplify the bursts oflight and block out cosmic rays, respectively (Fig. 3.23).Measurements are made with respect to another crystalwhich receives continuous gamma-rays from a standardradioactive so~rce.

When first devised, this technique was capable ofaccurate analysis only under laboratory conditions, orfor very crude analogue measurements from the air.Improvements in electronics, photomultipliers and thesize of suitable crystals now mean that quantitative air-borne data are possible. Systematic overflights in a gridpattern, together with reformatting and interpolationof data to a raster produces images that represent theapparent uranium, thorium and potassium contents ofsurfaceJJ;laterials (Chapter 8).

The main hindrances to gamma-ray imaging resultfrom the way in which the atmosphere and vegetation

How Data Are Collected 51

again at the demand of .the military. Since then theiracquisition has become a matter of routine for a widevariety of agencies, followingth~ transf~r of technologyfrom military to civilian applications.

Aerial photographic surveys are commissioned byboth national and local government agencies and by con-sulting and exploration companies. In the first case theyare acquired primarily for monitoring changes in landuse and quality, and for updating topographic maps. Inmany cases the photographic campaigns are co-ordinatedwith conventional surveying, when requirements fordifferent scales sponsor data acquisition fro~ a varietyof altitudes. Consequently, for any area of interest to thegeologist, there may be photographs with different fieldsof view, scales and resolutions for a number of differentyears and seasons.

The basic form of an aerial survey consists df anum-ber of approximately parallel flight lines, along whichexposures are timed to give at least 60% overlap betweensuccessive photographs, and' the spacing of which issuch that adjacent lilies of images 'have s~fficient sid.e-lap to ensure complete coverage. (Fig. 3.24). .More oftenthan not, an aircraft cannot maintain a perfect course,because of crosswinds and turbulence. If no correction

of the solid-state detectors. These devicesmilitary secrets, but as pressures grew for

-the technologies wereprogressively declassified in the USA. In retrospect, itseems clear that that secrecy lay not in the technologicalprinciples involved, but the limits to which they couldbe pushed and the precise areas on which attentionwas directed. Various lines of indirect evidence suggestthat satellite imagery with resolution better than 1 m hasbeen available for two decades. The present estimateof 10-15 cm visible-VNIR resolution from the latestUS 'Keyhole' satellite series (KH-12) may indeed havebeen superseded. Ultrafine resolution radar and thermalimagery is also indicated by indirect evidence. Probableglobal coverage and certainly global potential is possessedby USA and the former Soviet intelligence communities,but declassification for civilian applications is likely tobe slow. The reason for this is likely to be a n{ixture ofdiplomatic nicety and the advantage of potential oppon-ents not knowing precisely what can be hidden and what<;an be seen. In 1996 the US Government authorizedphased public release of satellite photography from theCorona series of intelligence satellites through the USGeological Survey, but commercial systems now offerbetter resolution (around 1 m).

In the next three, sections remotely sensed data thatcan be acquired openly, albeit at some cost, and whichare of use in geology are described.

3.9 Airborne data

Remote sensing from aircraft has several advantagesover orbital remote sensing, but there are also disadvant-ages. It is easier to change sensors, to suit a particularstudy, and to record at any time of day. The flying heightcan be varied, from less than 100 m above ground to30 km to give flexibility of scale, and to reduce variousatmospheric effects. Major disadvantages are that anaircraft is subject to various perturbations which cancause distortions in the i~ages, and that it is sometimesnot possible for the aircraft to fly in sensitive areas.

3.9.1

Aerial photographs

Aerial photographs are the most easily available of allremotely sensed data, and have been acquired at variousscales for much of the land area of the Earth. Their firstwidespread use was in World War I as a source of militaryintelligence. Soon afterwards oil companies began to usethem in geological mapping. The basic theory and practiceof making cartographic measurements from stereoscopicpairs of overlapping vertical aerial photographs weredevelop~d in the interwar years. Major advances in filmand camera technology took place during World War II,

52 Chapter 3

is made for crosswinds the photograph edges remainparallel to the intended flight line, but they drift awayfrom that line (Fig. 3.24). Correction normally meansturning the aircraft's nose into the wind, which causes itto crab, and if the camera is not reoriented the photo-graph edges cease to parallel the flight line, althoughtheir principal points will lie along it (Fig. 3.24). Usuallythese deviations are not so severe as to affect seriouslythe usefulness of the photography. Turbulence doescause problems, however. The aircraft may rise and fall,changing the scale of each image. It may pitch or roll,imparting a tilt to each image which produces unrealslopes in the stereomodel (Fig. 3.7). Finally, the groundspeed may change, thereby altering the overlap betweenadjacent images. If the overlap is less than 50% there willbe strips across the images for which stereoviewing isimpossible. If it is greater than 70% the stereomodel willshow insufficient relief for proper interpretation.

experience in how to \lse and interpret similar datafrom future satellite systems. Table 3.2 shows some ofthe specifications for two such airborne systems operat-ing in the reflected and emitted regions, each of whichhas produced large volumes of data both for publicuse and confidential exploration. The most interestingof these is the Geoscan, operated by Carr-Boyd Pty ofAustralia, which incorporates spectral coverage fromboth the reflected and emitted regions. The wavebandsin the reflected region are sufficiently narrow and posi-tioned to enable discrimination of specific mineralogicalfeatures in spectra. The bands in the emitted regionare, likewise, targeted on molecular features of emissionspectra. It combines fine spectral resolution within astrategically placed but limited number of wavebandsthat keeps data volume within manageable limits with-out sacrificing the system's potential for detecting anddiscriminating different surface materials.

More innovative airborne systems include theNASA imaging spectrometers. The first of these (1983)was the Airborne Imaging Spectrometer (AIS), whichused a pushbroom, area-array detector (Section 3.6).This was superseded by the Airborne Visible/InfraredImaging Spectrometer (A VIRIS) based on the whisk-broom principle. Together with a number of similar,

3.9.2

Other airborne dataAirborne

scanners and pushbroom systems are flowncommercially by a number of private companies andby national military and civil organizations. Some ofthem have been used in campaigns to give scientists

Table

3.2 Spectral range of three COl1"irnercial airborne multispectral imaging systems

Daedalus

AADS 1268[FOV 2..5 or 1..25 mrad

Geoscan AMSS Mk IIIFav 2.1 mrad

nMSlFav (2.5 mrad)

Band

numberCentral wave-length (~m)

Bandwidth(Jlm)

Band

numberCentral wave-length (1JJn)

Bandwidth(~m)

Band

numberCentral wave-length (~)

Bandwidtl(~)

0.4350.4850.5600.6100.6600.7200.8300.9801.6502.150

10.500

1234567891011

0.0300.0700.0800.0200.0600.0600.1400.1400.2000.2704.000

123456789101112

13

0.5220.5830.6460.6930.7170.7400.8300.8730.9150.9552.0442.0882.1362.1762.2202.2642.3082.3528.6409.1709.700

10.22010.75011.280

0.0420.0670.0710.0240.0240.0230.0220.0220.0210.0200.0440.0440.0440.0440.0440.0440.0440.0440.5300.5300.5300.5300.5300.530

123456

8.48.89.29.8

10.711.7

0.40.40.40.81.01.0

15161718192021222324

How Data Are Collected 53

Specifications of two imaging spectrometers satelite

A VIRIS HIRIS

Aircraft2.56140.40-2.45 ~m22410

Satellite8.8 (30 mpixels)800

OAO-2.451iffi19210:.-20

Spectral rangeNumber of channelsBandwidth (nm)

but less-widely deployed imaging spectrometers, thesedevices are precursors to the High-Resolution Imag-:ing Spectrometer (HIRIS) intended to fly in orbit in theearly 2000s (Section 3.13) Details of the specificationsof A VIRIS and HIRIS are given in Table 3.3.

Airborne imaging radar is now used routirtely by oiland mineral exploration companies oyer terrains thatare often cloud covered. Country-wide SAR surveys bygovernments are becoming widespread, the most not-able being the Brazilian RADAM project in the 1970s andthe ongoing US Geological Survey programme for thewhole USA. Extensive multipolarized and multispectralradar surveys have been flown by NASA, principally inNorth America, as forerunners of Earth orbital systemsand interplanetary probes. Examples of airborne SARimages are given in Chapter 7.

Fig. 3.25 The orbital inclination of a satellite is given by anangle (J between the plane of the Qrbit and that of ~heEquator.If the satellite passes directly over the poles then the inclinationis 90°. Conventionally, for orbits that take the satellite roundthe globe in the same direction as the Earth's rotation, (Jislessthan 90°; if it travels in the opposite direction then (J is grelitel:than 90°.

3.10 Basic characteristics of orbitingsatellites

The orbit of a satellite is elliptical in shape, but remotesensing satellites are usually put in orbits that are veryclose approximations to a circle. The laws of gravity dic-tate that a satellite in a higher orbit travels more slowlythan one in a lower orbit, and because higher orbits arelonger than lower ones this means that a high orbitingsatellite takes a good deal longer to circle the Earth than alow orbiting satellite.

For a satellite in a circular orbit of radius Ro from thecentre of the Earth its velocity v is given by:

v = (GM/Ro)1/2 (3.10)

where G is the universal gravitational constant(6.67 x 10-11 Nm2 kg-2) and M is the mass of the Earth(5.98 x 1024 kg).

The time taken to complete an orbit, the orbital period,P is given by:

P = 2nRo/v (3.11)

The radius of the Earth (Re) is 6371 km, and so the height(h) of the orbit about the Earth is:

h = Ro -Re (3.12)

Below 180 km the Earth's atmosphere is too dense forsatellites to orbit without uncontrollable frictional heat-ing. Above 180 km there is still a small atnlo.sphericdragon a satellite, causing its orbit to gradually spiral down-ward until eventually it reaches thicker atmosphere andburns up. Satellites are launched into such low orbits,but they do not last long. Above a few hundred kilo-metres there is so little atmospheric drag that a satellitewill remain in high orbit indefinitely.

The plane of an orbit must always p~ss through theEarth's centre but can be in any orientation. If its orbit isinclined at more than 450 to the equatorial plane then asatellite is in polar orbit. One that is less steeply inclinedis called an equatorial orbit (Fig. 3.2.5). Various forcesother than atmospheric drag can perturb an orbit, suchas the gravitational attraction of the Sun and Moon. Jetsof gas maintain the orbit and also control the spin of thesatellite.

An orbiting satellite normally changes its position inthe sky continuously. However, if a satellite has an orbitalperiod exactly the same as that in which the Earth rotatesand orbits in the same direction as the Earth rotates, it isin a geosynchronous orbit. To do this its orbital altitudemust be 35 786 km. If a satellite is in a circular geosyn-chronous orbit with zero inclination (above the Equator)then it appears to remain stationary over the same pointon the ground, and is in a geostationary orbit. Geosta-tionary orbits are useful for communications and meteoro-logical satellites, which need to be sited sufficiently far

54 Chapter.

resolution systems with ~wath widths of 100-200 kIn cancompletely image the globe within about 20 days, whereaslow-resolution systems, with swath widths of over1000 kIn, can image the globe in the course of a single day.

One disadvantage of Sun-synchronous orbits is that itis not always possible to transmit an image to the groundas soon as it is gathered, in real time. To send a signaldirectly to the ground means that the satellite has to beabove the horizon as seen from a ground receiving sta-tion. At typical orbit heights, the satellite has to be withinabout 3000 kIn of the station. If there is no ground receiv-ing station in line of sight from the satellite then theimage is either transmitted to the ground via a relaysatellite in a higher, geosynchronous orbit or is recordedon-board the satellite to be transmitted to a ground sta-tion when it comes within range. Tape recorders usepower and make the satellite heavier to launch, so not aUremote-sensing satellites carry them.

In the case of radar imagery a Sun-synchronous orbitis unnecessary, because illumil}ation is active, However,only polar orbits allow full global.coverage.

Spacecraft that carry personnel are much heavier thanunstaffed ones, because they must include life-supportsystems. This means that it is difficult to lift them intohigh orbits and expensive to put them in polar orbit.Equally important, it is difficult to recover personnelsafely. Usually their orbital inclination is about the sameas the latitude of the launch site. Most staffed missionsare of short duration, which makes them unsuitable formonitoring at a fixed solar time and incapable of acquir-ing comprehensive cover even within the limits of theirorbital inclination. Staffed missions, such as the SpaceShuttle, tend to be used as test-beds for experimentalsystems. For example, spaceborne imaging radar wasadvanced considerably by Shuttle experiments.

from the globe that. they can monitor almost a wholehemisphere on a nearly continuous basis. The altitudeof geostationary satellites is so great that the groundreso:lution is low because the smallest IFOV of sensors inthe reflected region is about 20 ~rad, which at an altitudeof 35 786 kIn gives about 720 m on the ground.

It is useful to obtain repeated images at the same timeof day, to ensure similar illumination conditions. The timecan be selected to avoid early morning mists or afternooncloud in the tropics, to take advantage of the diurnalheating and cooling cycle in thermal infrared studies, orto ensure a low Sun angle that highlights topographicfea~res. To do this a satellite must pass over all points atthe same local solar time-it must be Sun-synchronous.A satellite placed in a steep polar orbit, the motion ofwhich carries it westwards over the ground at a ratecomparable with the Earth's rotation, achieves this con-dition. Most Sun-synchronous polar orbiters pass fromnorth to south over the sunlit hemisphere and return fromsouth to north over the night-time hemisphere. Becausethe Earth rotates beneath the orbital plane, successiveorbits cover different ground tracks. As the orbital periodis much less than 1 day, images of several ground trackscan be acquired within 24 h. Most Sun-synchronousremote-sensing satellites are placed in orbits with alti-tudes between 600 and 1000 kIn. At 800 kIn, a sensor withan instantaneous field of view of 20 ~rad would have awound resolution cell of 16 m, compared with the 720 machieved from geosynchronous orbit.

The, type'." of ground coverage offered by a S1Jn-synchrooous satellite 1sshown in Fig. 3.26. Unless theswath is ,very wide, there is a gap between the swathsimaged in successive orbits. Part of the design oforbitingsystems is to calculate an orbital geometry that ensuresthat the gaps are filled as efficiently as possible. High-

Fig. 3.26 The groundtrack of a Sun-synchronous satellite (in this casethe first of the Landsat series) for asingle day. In fact the local solartime varies from 1030 hours at 600Nto 0830 hours at 60°5, a~d a trulySun-synchronous orbit is impossible.

systems. Th~ second flig~ in.c. .-Data from staffed spacecraft

satellites have provided platforms for a wideof remote-sensing instruments since the early

of space exploration. They have not, ~s yet, pro-.--

limited coverage. In many cases the instru-

Multispectral Infrared Radiometer (SMIRR), was a non-imaging spectroradiometer that measured reflectancein 10 wavebands over the range 0.5-2.4 Jlm, Includingthree with a width of 0.04 Jlm in the vicinity of theAI-OH feature near 2.2 Jlm (Fig. 1'.9). Its objective wasto test the theoretical possibilities for rock and mineraldiscrimination from orbit (Section1'.3.2).

The second system Was the first Shuttle ImagingRadar (SIR-A) experiment. This deployed a 23-cm (L-band) SAR system similar to that on the unstaffedSeasat of 1978 (Section 3.12.1). Unlike Seasat, with itsmain target being the ocean surface, SIR-A was designedprimarily for land applications. To reduce layoverproblems (Section 3.5) in areas of rugged topography,without producing too much shadowing, SIR-A useda depression angle fixed at 43° to cover a ground swath50km wide. Ground resolution was 40 m in bothazimuth and range directions. Continuous holographicfilms representing 8 h of data were' correlated optic-ally after the mission to radar images on positive film(Fig. 3.27, see Chapter 7). This record amounts to about10 million km2 of land and ocean between 41°N and 35°S(Appendix D).

the data. Except for datacollected by Soviet cosmonauts, which are now becom-ing available, all stem from USA programmes operatedby the National Aeronautics and Space Administration(NASA). Archives for many remotely sensed data fromstaffed NASA missions are held by the US GeologicalSurvey's EROS Data Center and at NASA Goddard SpaceFlight Center, whc>se World-wide Web pages appear inAppendix b, along with those of other agencies.

The Mercury and Gemini programmes providedover 1000 normal colour and a few false-colour infraredoblique and near-vertical photographs, taken by hand-held 70 mm cameras. Many of them were directed atgeologically interesting areas. Both Gemini and Earth-orbiting Apollo missions were limited to orbits between35°N and 35°S. As well as 70 mm colour cameras, Apollo 9(1969) also carried the first orbital multispectral experi-ment. This consisted of an array of four hand-triggered70 mm cameras mounted in the port of the commandmodule. Three of the cameras exposed black and whiteffim to give images of the green (0.47-0.62 ~m), red(0.59-0.72 ~m) and very-near-infrared (0.72-0.90~)regions, and the fourth used infrared colour ffim. Groundresolution proved to be about 100 m. This ex~rimentwas directed only at test sites in the southern USA andnorthern Mexico, and was co-ordinated with aircraftremote sensing flown at the same time as the mission.The experiment was designed as a test of the conceptfor an unstaffed multispectral scanner, which became theEarth Resources Technology Satellite (July 1972), sub-sequently renamed Landsat-1 (Section 3.12.2). The successof Landsat-1 prompted NASA to extend its remote-sensing ambitions by a series of experiments aboardSkylab (May 1973 to February 1974), a 100 tonne orbitalworkshop based on the third stage of a Saturn~ V launchvehicle, which orbited between 500N and 500s.

N

~ I 10km Iillumination \Fig. 3.27 This SIR- A L-band radar image of the AppalachianPlateau and the adjacent fold belt in Kentucky and Virginiashows the advantages over rugged terrain of a low depressionangle. There is little distortion and many of the majorgeological features are displayed well. The same area1sshownas a Seas'"\t SAR image in Fig. 3.33. Courtesy of J .P, Ford, JetPropulsion Laboratory, Pasadena.

3.11.1

Space Shuttle data

NASA's staffed spaceflight programme is centred on thereusable Space Shuttle, the payload capacity and flexibil-ityof which is enormous compared with earlier launch

56 Chapter3

a pushbroom system with a ground IFOV of 20 m andtwo bands (0.58-0.63 and 0.83~0.98 flm). The later MOMS(1996 carried by a Russian staffed satellite) used threepushbroom systems, two pointed fore and aft to givepanchromatic stereoscopic capability, and one aimedvertically downwards to gather data with 25 m resolu-tion in four wavebands between 0.45 and 1.05flm.Neither MOMS provided images of more than a few ofthe planned scenes.

As well as the SII{-B system, the October 1984 Shuttlemission carried the Large Format Camera (LFC), whichproduced panchromatic natural colour and false-colourinfrared images. The effective resolution of LFC panchro-matic photographs is between 8 and 15 m, dependingon the altitude of the Shuttle orbits. Each frame coversareas of 250 x 500 to 350 x 700 km, in an archive of2160 photographs, many of geologically interesting areasin arid and largely unknown terrains. Various overlapsbetween frames give all LFC images stereoscopic poten-tial. This instrument was never flown again, and NASAwas reported to have sold the archive of negatives toa commercial distributor for a peppercorn sum. Figure3.28 is an example of an LFC photograph for comparisonwith Landsat images of the same area (Plate 3.3). SeveralLFC images are used in Chapter 4.

In October 1984 the Shuttle carried a system similar toSIR-A, called simply SIR-B, but with a variable depres-sion from 75 to 30° and hnproved resolution. This resultedin optimum depression angles for different kinds ofapplication, and several images with varying parallaxthat permit pseudostereoscopic viewing. The pointableantenl1a meant that test sites could be viewed from avariety of look directiQns. Data in both optical hologramform and as digital records cover swaths between 57°Nand 57°S. Images were expected to be of even better qual-ity than those from SIR-A. Unfortunately, technical prob-lems during the mission thwarted many of the plannedacquisitions, and only a few images are of useable qual-ity .The 1994 SIR-C/XSAR experiment, conducted jointlyby NASA and the Germafispace agency, involved the useof multispectral radar (L, C and X bands) with variouspolarizations (Chapter 7), and a polar orbit.

In early 1984, a Shuttle mission carried two Euro-pean imaging systems that are useful for geologicalapplications. One was a Metric Camera, primarily torcartographic applications, with a ground resolution of30 m: and frame size of 23 cm by 23 cm, which exposedfalse-colour infrared and p~nchromatic film. The otherwas the first test of the German Modular Optical-Electr<?nic Multispectral Scanner (MOMS) which used

Fig. 3.28 Large Format Cameraimage of semiarid mountains inEritrea, north-eastern Africa. Theeffective resolution is about 10 m.See Plate 3.3 also.

How Data Are Collected 57

Remote-sensing experiments scheduled for testsaboard the Shuttle until the Challenger disaster includedthe Thermal Infrared Multispectral Scanner (TIMS),collecting emitted radiation in six bands between 8and 12 ~m (Table 3.2). Results from the aircraft-mountedversion of this are discussed in Chapter 6. Anotherimportant but unflown candidate was an imaging spec-

--

the astronauts took manyphotographs from the ports of the Shuttles with a varietyof cameras. These are easily obtained from NASA'sGoddard Space Flight Center (Appendix D).

3.12 Data from unstaffed spacecraft

There are three advantages of unstaffed orbital platforms.(i'

1 By not requiring complex and heavy life-support sys-tems, they are cheap to build and launch, and need notbe recovered.2 They can be placed in Sun-synchronous polar orbitsthat guarantee the possibility of almost complete globalcoverage (Section 3.10).3 Subject to the deterioration of equipment and exhaus-tion of propellants required for orbit adjustments, theycan operate for very long periods gathering repetitivedata. Humans can endure the conditions of orbital travelfor only a matter of months.

3.12.1 Meteorological and oceanographic satellites

The most enduring, successful and consistent produc-tion of re~otely sensed images has been by unstaffedmeteorological satellites. To describe and illustrate themall and the variety of devices that they have carried isbeyond the scope of this book. Primarily, they monitorglobal climatic features, with sensors tuned to cloudcover and patterns, atmospheric water content, pre-cipitation and both land- and sea-surface temperatures.Some also carry instruments aimed at oceanographicfeatures, such as the abundance of phytoplankton andparticulate matter in sea water. Their role is monitoringthe entire planet on a day to day basis. Consequentlytheir design and their orbits ensure coverage of vast areasat coarse resolution, between 800 m and 25 kill. For manygeological applications the features of interest are toosmall to be resolved. Their supersynoptic view, however,does serve to gain a gross overview of regional geology.In the same way that high-resolution satellite images helpco-ordinate the information drawn from aerial photo-graphs and field work, meteorological images can helpdraVf together the findings from a study of better resolu-tion satellite images.

The first image-gathering unstaffed satellite wasthe US Television and Infrared Observation Satellite

(TIROS-l), launched op 1 April 1960. It carried twominiature television cameras on an 800-kIn orbitbetween about SooN and 500S, with a period of about100 minutes. This began an ongoing series of low-altitude environmental satellites launched from theUSA. For administrative reasons a deep moat of con-fusion surrounds their naming, partly because beforelaunch they had one name and then were rechristenedin orbit. Their administering body also changed name.For simplicity they are known as the TIROS/ESSA/NOAA series, which were placed in near-polar orbitsto gather data for most of the globe. They were firstadministered by the US Environmental Science Service(ESSA), which became the National Oceanic and Atmo-spheric Administration (NOAA).

The first geostationary image-gathering satellite, withan orbit at about 41 000 kIn, was NASA's ApplicationsTechnology Satellite (ATS-l), launched in December1966. It provided a view of almost an entire hemi-sphere from a position above the Equator, and was suc-ceeded by a series of similar platforms: ,Soviet GOMS,the Japanese Himawari (over the western Pacific), theEuropean Space Agency Meteosat (over the easternAtlantic), and the Indian INSAT (over the Indian Ocean)series. Together these comprise the GOES family, admin-istered by the World Meteorological Organization inGeneva. The bulk of instruments aboard these plat-forms produce a range of data on a nearly real-timebasis, but much is of too coarse a resolution to be usefulgeologically. The panchromatic visible sensors aboardthe geostationary platforms have a resolution of 900 m,however, which does enable some gross geologicaldetail to be seen.

The most geologically useful instrument aboard thecurrent TIROS/ESSA/NOAA series is the AdvancedVery High Resolution Radiometer (A VHRR). This is aline-scanning system, with a mirror that sweeps through56° either side of the nadir of the ground track. The sys-tem's IFOV together with the orbital altitude (833 kIn)leads to a ground resolution of about 1.1 km at nadir,degrading outwards because of the large scanningangle. Each scan records data for 2048 pixels, which,together with the off-nadir distortion, means that imagescover a swath width greater than 2500 kIn. All theNOAA series satellites use a near-polar, Sun-synchronousorbit; and cross the same ground area repeatedly at thesame local time. Two satellites are always functioning,one passing over at 0730 hours and the other at 1430hours, on descending orbits (southwards). They crossthe night-time side of the Earth 12 h later on ascend-ing orbits. The orbital period of 102 minutes results in14.1 orbits per day, ensuring complete global coverageevery 12 h.

The A VHRR instrument has five channels sampl-ing radiation from the visible red to emitted infrared

58 Chapter 3

Table 3.4 Spectral bands cov!"red by five geologically ~seful orbiting systems

Landsat

AVHRR

SPOT JERS-l

ASTER

BandBand I{ange (11m) Band Band Range (I.lm) Band Range (11m)Range(!!ffi) Range(~)

MSS 1 (4)MSS~,<5)MSS 3(6)MSS 4 (7)MSS (8)TMITM,2TM3TM4TM5TM61M7

c 0:5-0.6

0.6-{).70.7-0..8

i0.8-1.110.4-12.6OA5-0;~20.52-0.600.63-0.690.76-0.901.55-1.7510A-12.52.08-2.35

12345

0.58-10.73-3.55-10.5-11.5'-

XSIXS2XS3P

0.5-0.590.61-0.680.79-0.890.51-0.73

0.52-0.600.63-0.690.76-0.860.76-0.861.60-,,1.712.01-2.122.13-2.252.27-2.40

0.52~0.60 -}0.63-0.69 -~I

0.76-0.86 -~}1.600-1.700'"2.145-2.1852.185-2.2251.235-2.2852.295-2.3652.360-2.430..8.125-8.47~8.475-8.8258.925-9.27510.25-10.9510.95-11.656

123(+:4567891011121~14

1234 (stereo)5(;78

Nimbus-7 was the last of the series, launched in Octo-ber 1978 into a 955-km Sun-synchronous orbit with aninclination of 99.30 and a repeat cycle of 6 days, and wasretired in September 1984. Nimbus-7 carried a varietyof radiometers ranging from the ultraviolet to micro-waves, including the Coastal Zone Color Scanner (CZCS).The CZCS covered a swath width of 1566 kIn, with apixel-size of 825 m at nadir. There were six spectralbands, four of which were. narrow channels (0.02 J.lm)in the visible region for mapping algal chlorophyll andother suspended material in the oceans, a VNIR channelfor mapping surface vegetation and a thermal infraredchannel aimed at ocean temperature variations. The CZCSdata had great potential for regional geological over-view, but have rarely been used because of the patchynature of its cover. In 1997 NASA replaced the CZCSwith a similar, but finer resolution system known as theSea Wide-field Sensor (Sea-WIFS) (Appendix D).

Seasat was launched into a 790-km, 1080 inclinationpolar orbit in June 1978. It carried a variety of radiome-ters and a radar altimeter (Chapter 8), but more signific-antly it had the first SAR system to be used in orbit.This was an L-band (23.5 cm) system with a depressionangle of 700 giving a swath width of 100 km and a resolu-tion of about 25m. A steep depression angle results intopographic layover (Section 3.5) and other distortionsin rugged terrains (Fig. 3.30), but is especially suitablefor imaging the ocean surface to assess its roughness.In fact, as its name suggests, Seasat was designed foroceanographic applications, but the radar also proveduseful over land. ,The area of SAR coverage for Seasatwas limi~ed to North America and the northern Atlantic,region by the avililable ground stations. Unfortunatelya pow~r failur,e terminated Seasat operations after only

region (Table 3.4). The detectors output intensity levelsin digital form in the range 0-1023 (10-bit precision).Transmission to the ground takes two forms. Data arebroadcast continuously at full resolution, when any suit-ably equipped ground-receiving station can record anddisplay it-High Resolution Picture Transmission (HRPT).Five on-board recorders can each store 10 minutes of fullresolu~on data-Local Area Coverage (L;\C)-and a fullorbit (1)0 mjnutes) of data, resampled to 4 km resolution-Glo~l Area Coverage (GAC). These stored data aretransmitted to NOAA ground-receiving stations in theUSA, when the satellites are in line of sight with thereceiving antennae. Figure 3.29 is an example of a visible-red LAC image from the NOAA-7 A VHRR. One advant-age of the daily capture of images from the A VHRR isthat eventually a cloud free image or one with minimalcloud cover will be acquired. It therefore becomes pos-sible to mosaic seyeral such images to build up super-synoptic images. This has been done for the whole world,and Plate 3.4 shows a simulated natural-colour infraredimage of western Europe and North Africa.

Although suiteq. to investigations of very largegeological features, A VHRRima~ryis currently under-us~d by geologists. Estimation of rainfall from clouddensity and temperature, and monitoring of snow masses,on a daily basis, provide input to models of stream run-off used by hydrologists. The spectral range means thatthe Stefan-Boltzmann relationship permits estimationof temperatures from those normally affecting land andsea surfaces, to those associated with active lava flows.Pote~tially, twice-daily A VHRR data could become thebasis for volcano monitoring (Chapter 9).

Between 1964 and 1978 various sensors were tried outon satellites of the Nimbus series, operated by NA$A.

0.681.103.9311.512.5

;tereo)

How Data Are Collected 59

I

10km I

N

illumination \Fig. 3.30 This extract from a Seasat L-band radar image is ofthe same area in north-east USA as Fig. 3.27. The radarlookdirection is at 450 to that for the SIR-A image in Fig. 3.27. This,combined with the steeper depr~ssion.anglefor the Seasat :SAR compared with that for SIR-A, makes the image virtually.unintelligible. The look-direction is parallel to the mainmountain ridges and valleys in the Appalachian foldbelt andso they are suppressed.' The rugged topography hasresulteqin the image being plagued by for~horlening and layovereffects, which further disrupt its structure. However, in theAppalachian Plateau (northern part of the image) there areadv~ntages. The dominant southwards drainqge is enhanced,allowing discordant linear features, probably major faults,to be picked out.

3.12.2 The Landsat series

Experiments from staffed satellites and the success ofthe semipermanent meteorological satellite systemsencouraged NASA to initiate a series of high-resolutionorbital systems aimed at the biological and physicalresources of the Earth. The first Earth Resources Tech-nology Satellite (ER~I) was launched by NASA on23 July 1972, using a Nimbus platform (Fig. 3.31a). Itwas subsequently renamed Landsat-I, and eventuallythe administration of the Landsat series passed firstto NOAA and then to a commercial company EOSA T(now Earth Imaging) (Appendix D). The 918 kIn Landsatorbits passed within 90 of the poles, and circled theEarth once every 103 mmutes. Descending orbits crossedthe Equator at an angle of 90 at around 0930 hours localtime. So, the Landsat series Was Sun-synchronous, togive illumination of the surface by the mid-morning Sun.This avoids early morning cloud at mid-latitudes, but

Fig. 3.29 The gross geomorphology of an area of about4 million km2, covering parts of Mghanistan, PaJdstan, part ofthe former USSR and Tibet is shown on this Chaimell (red)image from the A VHRR abOard NOAA-7. The High Himalayaand Tibetan Plateau in the east are largely snow-or cloud-covered. The Himalayan foothills are densely forested andappear dark. The fertile p.ains of the I~dus and its tributariesare dark, in contrast to the lightcoloured desert areas. Thegeological advantages of a supers~optic view are amplyillustrated by the interconnected complexities of the Sulaimanand Salt Ranges of north-west Pakistan. They are trUncated inthe far west by a huge strike-slip fault, along which the IndianSubcontinent has slipped during its collision with Asia.

105 days. of operation. Nevertheless it provided sufficientevide~ce'of the advantages of all-weather, day or nightorbital SAR imaging to increase the impefus for moregeologically orientated systems, such as the SIR series.

60 Chapter 3

(a)

Fig. 3.31 The first three Landsats occupied spare Nimbus platforms (a), whereas that for Landsat-4 and -5 isa modified TIROSplatform (b).

;/~;;:~:~ay M~1 Orb?t N, djY M+ 1

/0

't i'\:"

(

-'I

~~~//'1

, -orbit N, day Morbit N+1, day M ~ orbit N, day M+18

Fig. 3.32 The diagram shows successive Landsat orbits ontwo days. On day M, orbit N + 1 is shifted 2100 km west oforbit N at this latitude. Orbit N on day M + 1 is only 120 kmwest of the same orbit of the previous day, so the imagestaken on both days overlap. The higher the latitude thegreater the overlap between adjacent scenes. The sequencerepeated itself every 1~ days for Landsat-I, -2 and -3. Thisrevisit frequency was reduced to 16 days for Landsat-4 and -5:Very.mt1ch the same principle applies to all S11i\-synchronous.polar'-Orbiting systems. In the case of the NOAA A VHRR,the revisit frequency is every 12 h to give day and nightcoverage.

also highlights topography because of the low Sun angle.Successive orbits were 2760 km west from their prede-cessor (Fig. 3.32), so Landsat-l crossed the same point onthe ground every 18 days. Every part of the globe, exceptfor the near-polar regions, which are never crossed, cantherefore be imaged up to 20 times per year. Unlikethe A VHRR, Landsat imaging systems cover a narrow(185 km) swath, so there are gaps in the ground cover-age in successive orbits (Fig. 3.32). Over an 18-day period(251

orbits), however, Landsat-l to ~3 gave a complete setof ground tracks, spaced at 159 km apart at the Equatorand giving a 14% sideways overlap, after which time thecycle repeated. The sideways overlap increased towardsthe poles, and reached 50% at 540 latitude, so that everypoint at this latitude was covered twice as frequentlyas most points on the Equator. The ground track of aLandsat.l to -3 satellite over a single day is illustrated inFig. 3.26, and the relationship between adjacent orbitson successive days is shown in Fig. 3.32. This repeatedpattern of overpasses, common to all Sun-synchronoussatellites, is the basis for the Landsat world,.wide refer-ence system for individual scenes, the principle of whichis followed by other operational systems.

In 1982 NOAA adopted a new orbital plan forLandsat-4 and its successors. This involved anew,heavier platform (Fig. 3.31b) and a lower orbit (705 km),resulting in a shorter orbital period (99 minutes), ashorter repeat cycle of 16 days and a greater number oforbits to give global coverage. This was to give greaterorbital stability and to help improve the resolution ofthe on-board sensors.

How Data Are Collected 61

and -2 were equipped with two imaging

Vidicon

system, the other a four-channel Multispectral

(MSS), both aimed at visible and VNIR bands

features related to chlorophyll (Table 3.4).

original Landsat concept was orientated towards

of vegetation rather than geology. The one

orientated topographicThere can be little doubt that Landsat was

to maintain a permanent high-precision

customer for wheat, the former Soviet Union.

The RBV s aboard Landsat-l and ~2 provided little data.they left a legacy in the naming ?f the fouron Landsat-l to -3, which were designated

4, 5, 6 and 7 because they followed the RBVThe Landsat-l to -3 MSS system had

using an oscillating mirror3.8). This collected radiation for banks of six sensors

of the four wavebands, which were directeda dispersing element to the detectors by optic

-.

of the system, which was 109 jlrad, resulting inground resolution of 79 x 79 m. The sampling rate,

, was such that each pixel overlapped those.5 m, resulting in an actual pixel size

79 m depth and 56 m width. With the lower orbitand -5 the MSS scan angle was increased

swath width at 185 kIn, and the opticsadjusted to preserve the approximately 80 m

The spectral bands were identical,although renumbered as bands 1-4 as mentioned above,so that old and new MSS data are very closely compar-able to maintain data continuity.

The analogue data from each MSS detector areelectronically chopped and digitized 10 a 0~63 range ofDN (6-bit precision). On Landsat-l10 -3 the data wereeither relayed directly to a ground-receiving stationor recorded until the platform was above a ground-receiving station. On Landsat-4 and -5 there are no taperecorders and data are telemetered to the groundeither directly or via a series of tracking and data relaysatellites (TORS). The data are rescaled on the ground

bands 4, 5 and 6 (now 1, 2 and 3~ and 0":63for band 7 (now 4), and output either to film using adigital film writer, or to digital storage. In both cases thedata use the range 0-255 (8-bit precision), but of courseonly 128 different real levels are possible in the data.The detectors are calibrated by lamps built into the sys-tem in such a way that the darkest expected surface-deep clear water or black rock-and the lightest-cloudor snow-both give responses within the range. Because

cloud or snow is so mpch brighter in the four bandsthan any surface vegetation, soil or rock, this means thatcloud-free images generally have their data compressedinto a relatively small range of low digital numbers andrequire contrast enhancement (Chapter 5).

The Landsat-1 to -3 MSS collected data for 3240 pixels56 m wide during every sweep of the mirror, each 474 mdeep-79 m for each of the six detectors. The groundswath imaged on each overpass therefore was about185 km wide. The continuous stream of data was sam-pled about every 25 s, giving a total of 2340 lines, so thateach sample represents an area 185 x 185 km on theground. This is termed a scene. Each scene acquired byLandsat does not represent a square area on the ground,but a parallelogram (Fig. 3.11h), because of the effect ofthe Earth rotating in the time taken to gather the data.Neither are the scenes orientated precisely relative to truenorth. This is because the orbits are .inclined to latitude;in fact they are slightly curved (Fig. 3.26). Apart fromdata from Landsat-1 and -2, scenes are corrected forEarth rotation effects, although the user must registetthem to local map co-ordinates (Appendix B). An extractfrom a Landsat MSS image is shown in Plate 3.3(a).

The other addition to Landsat-3 was a line-scansystem for emitted infrared. This thermal data had aresolution of 120 m, with .a similar structure to theMSS data. However, the thermal channel failed shortlyafter launch.

The gradual deterioration in the performance of allsystems aboard Landsat-3 k>rced an early launch of itssuccessor, Landsat-4 in September 1982. As well asa platform design based on that of the TIROS series(Fig. 3.31b) and the replacement of onboard recordingcapability with direct telemetry of data by multipurposedata-relay satellites, Landsat-4 carried an entirely newinstrument-the Thematic Mapper (TM)~as well as aslightly modified MSS. The TM is a line-scan sys~em,but incorporates several innovations compared withMSS. Instead of using fibre optics to carry radiation tothe detectors, they are placed directly at the focal planesof the optical system. One of these focal planes carriesfour banks of 16 detectors for four wavebands in thevisible to VNIR range. Another is refrigerated by aradiative cooler, and carries two banks of 16 detectorsfor two wavebands in the SWIR region, and one baQkof four detectors for emitted infrared. Table 3.4 showsthe specifications for TM wavebands, where there isan apparent anomaly in the numbering of the bands-band 7 is in the SWIR, and band 6 contains thermaldata. This arose from the late addition of the SWIRband at the request of a vociferous geological .lobby.The arrangement allows the mirror to scan a groundstrip with approxi.mately the same width as that scannedby the MSS, but radiation is ~ollected by 16 instead ofsix detectors for the visible and near-infrared channels:

62 Chapter 3

The SPOT system maJ'ked two significant advances inremote sensing; it was the first unstaffed satellite to usea pushbroom system, and was the first remote sensingsatellite to offer stereoscopic cover. By pointing bothHRV sensors downwards it is possible to image a strip117 km wide, with a 3-km overlap between instruments(Fig. 3.33b). Alternatively/by means of steerable mirrorsthat direct radiant energy on to the CCD arrays andpoint up to 270 either side of vertical, one or both HRVinstruments can image a swath centred up to 475 kmon either side of the ground track. This ability meansthat a stereoscopic pair of images can be acquired byimagi~g the same area from different directions, duringdifferent orbits (Fig. 3.33c). A stereoscopic SPOT imageis used to illustrate part of Section 4.4.1. The same areacan also be revisited up to 11 times during the 26-dayperiod, depending on the latitude (Fig. 3.33d). As wellas providing routine high-resolution images, SPOT isalso capable of responding to short-lived events, suchas floods and volcanic eruptions.

The SPOT platforms carry a tape recorder and trans-mit images recorded from around the world, as well asimages of Europe in real time, to ground stations atKiruna (Sweden) and Toulouse (France). In addition,many of the Landsat ground stations are also licensedto record and distribute SPOT data. The next two inthe series, SPOT -2 and -3 were the same as the first twosystems. SPOT -4 carries an additional spectral band at1.58-1.75 Ilm and also a 1-4 km pixel vegetation-sensingsystem, using five spectral bands (blue, green, red, VNIRand SWIR), with a swath 2200 km wide (extending 500either side of nadir). SPOT -5 carries improved sensorscapable of 10 and 5 m resolution for mUltispectral andpanchromatic modes, respectively.

The ground IFOV in these channels is therefore 30 m,an improvement of about 2.6 times compared with thatof the MSS. Resolution for the thermal channel is 120 m.To match the IFOV to the platform's orbital velocity,the mirror gathers data in both forward and backwardsweeps. This means that the actual scan lines are notexactly parallel on the ground because of orbital motion.Correction for this distortion result~ in an actual pixelsize of 28.5 m.

Thematic Mapper scenes are ~lmost the same size asthose from the accompanying MSS, but because of theimproved resolution and larger number of channels theycomprise 8.5 times as many digital numbers. Landsat-4ceased to operate early in 1983 because of problems withpower supply and serviceability of its relay system.Fortunately it was possible.to replace it with the alreadyprepared, but slightly modified Landsat-5 system, inMarch 1984. Plate 3.4 compares a Landsat TM imagewith one from the MSS system. Landsat-5 continues tofunction, with Landsat-4 as a back-up, but both arelong past their planned lifetime. Their replacement,Landsat-6, ignominiously fell into the sea after lift off.After much wrangling, Landsat-7 with an additional15 m panchromatic band, was launched successfully in1999. At the time of writing NASA was about to launchan experimental, low-cost platform to test componentsfor fufure members of the Landsat series.

Because of the overlap between the image swaths foradjacent Landsat paths, which increases with increasinglatitude, it is possible to view parts of the images stere-optically (Appendix A). A system designed to producestereoscopic pairs of images by pointing the detectingsystem off-nadir, thereby introducing parallax differ-ences between images, is the French Systeme Probatoirede l'Observation de la Terre (SPOT).

3.12.3 SPOT data

The French Centre National d'Etudes Spatiales (CNES)launched SPOT -1 into an 830-km Sun-synchronous polarorbit (98.70 inclination) in February 1986. The SPOT sys-tem is operated by SPOT-Image. The platform and itssuccessor SPOT -2 cross a point on the Equator on descend-ing orbits at 1030 hours solar time every 26 days. TheSPOT system uses two pushbroom devices (Section 3.4)(Haute Resolution Visible-HRV) comprising arraysof 6000 CCDs (Figs 3.13 and 3.33a). These may operatein either a panchromatic mode with a ground IFOV of10 m, or multispectrally to gather green, red and VNIRreflectance (Table 3.4) with an IFOV of 20 m. The imagedswath, when directly below the platform, is 60 km wide.In the panchromatic mode scenes comprise an array6000 pixels square, and in .the multispectral mode eachscene is 3000 pixels square.

3.12.4 }ERS-1 data

In February 1992 the Japanese National Space Develop-ment Agency (NASDA) launched what was then the mosttechnically sophisticated, multipurpose remote sensingsystem in the civilian domain. It was called simply theJapanese Earth Resources Satellite-1 (JERS-1), and car-ried two imaging systems, an Optical Sensor (OPS) anda synthetic aperture radar system (SAR). The JERS-1orbited at an altitude of 568 km with a 97.70 inclination,which allows a revisit period of 44 days.

The OPS covered the reflected region of the spectrum,being intended for detailed discrimination of vegetation,soils and rocks. It used two radiometers, one for thevisible and near-infrared regions, the other for the SWIR.Both pointed vertically downwards, but that carryingvisible and near-infrared sensors could also point 15.30ahead of the nadir to obtain stereoscopic images. Theseven spectral bands capturing data from the nadir

How Data Are Collected 63

HRV Instruments

(c)day of overpass

D+~O- D

..::,~~~~S ~I

D-5/-.

'~

(d)

~

/(swath, observed

~

v

Fig. 3.33 The SPOT platform carries two HRV sensors (a) thatimage two 6O-km swaths directly below the satellite with 3kmoverlap (b). Pointable mirro~s in the HRVs allow viewing atvarious angles to the side tq allow revisit~ at frequencies up toevery 5 days (c) or stereoscopic imaging (d).

(Table 3.4) each had a dedicated 4096 element CCD arraycapable of me~surements with 6-bit (64 DN) precision.This configuration, together with an IFOV of 32.2 ~rad,produced images 75 kIn across with a pixel size of18.3 x 24.2 m. The forward-looking radiometer used onlya single VNIR band to correspond with one from thenadir-pointing mode for stereoscopic imagery~

The three bands in the 2;0-2.4 ~m region were aimedspecifically at Al-OH, Mg-OH and C-O bohd transitions(Section 1.3.2} and .should have enabled more confid-ent separation of clays, hydroxylated ferromagnesianminerals and carbonates than had been possible before.Technical considerations limited the dynamic range ofthe data to 64 ON compared with the 256.oN of Landsat

64 Chapter 3

TM data, which trimmed this optimistic view some-what. Sadly, technical problems with the SWIR sensorsintroduced seriously disturbing noise, which requiredtime-consuming cosmetic processing before JERS-l datawere useable.

The JERS-"l SAR also had potential for geologists. Its23 cm L-band system had a depression angle of 550 aimedat suppressing the layover and shadowing in high-reliefareas that plague SAR images with 1arger and smaller

depression angles, respectively (Fig. 3.22).

3.12.5 Other systems of interest

The Heat Capacity Mapping Mission (HCMM) experi-ment launched by NASA in April 1978 had a near-polarorbit at 620 km (97.60 inclination). It ceased operationin September 1980. The HCMM orbits were Sun-synchronous with times of equatorial passage at 1400and 0300 hours local time, chosen to match those of theaverage maximum and minimuhl daily soil temperat-ures. Potentially, it could estimate variations in sur-face thermal. inertia, when the day and night-time datawere registered (Section 1.3.2). It carried two sensors,one covering visible and near-infrared (0.5-1.1 J.Lm)regions,. the other the emitted infrared (10.5-12.5 J.Lm)region. The ground resolution of the thermal scannerwas about 600 m, reduced to 480 m after onboard pro-cessing. The imaged ground swath was about 716 kmwide. Because day and night orbits were inclined toone another, each scene of registered thermal data waslozenge-shaped. Without an onboard recording system,data ar~ available only for areas within range of suit-ably equipped ground-receiving stations. The potentialfor extracting important information relating to rocktypes and to soil moisture is sufficiently great, however,that HCMM results are discussed at some length inChapter 6.

The first Indian Remote Sensing SatelJite (IRS-l A)was launched into a 904-km Sun-synchronous polarorbit in March 1988 by a Russian rocket. Its orbitalrepeat was 22 days, with an equatorial overpass time of1000 hours. The satellite carried three pushbroom LinearImaging Self Scan~ng (LISS) cameras, each recordingthe same four visible and near-infrared spectral bands(0.4-0.5,0.5-0.6,0.6-0.7 and 0.7-0.9 J.Lm). LISS-l coversa 148-km swath with a ground IFOV of 73 m. LISS-2A and LISS-2B are identical and image adjacent swathson either side of the ground track at 36.5 m pixel size.Most data are transmitted to the Indian ground stationin Hyderabad (limiting the coverage to India and itsvicinity), but there are plans to receive some data at otherreceiving stations. The latest, IRSI-C, also carries a 5-mresolution panchromatic, pushbroom sensor, providinguseful, medium-cost image data for several geologicalapplications.

How Data Are Collected 65

Illumination direction Illumination direction

I::0

':e~:c+"or.,21;0::QI~

"'Q;+"10

VI

50 km

Fig. 3.34 ERS-l (a) and JERS-l (b) gAR images of part of Japan,showing the volcano Mount Fuji. The fact that Mount Fuji is anearly perfect cone with a circular summit crater serves todemonstrate the inappropriate depression angle of ERS~ gAR

by its apparently lying on its side. Many other ruggedtopographic features are also completely distorted by extremelayover. The JERS-l image preserves the shape of the volcano,but still contains layover.

3.13 Future prospects

For the period up to 2005 there are plans for enoughcivilian remote-sensing satellites to fill a separate chapter.Plans, however, only materialize given money andsuccessful launches. Owing to technical difficulties, evenfunded ventures do not leave the ground on time. Thefirst of the Earth Observation System platforms (below),which is to carry the ASTER instrument, eventually

reached orbit after an IS-month delay (December 1999).Some do not work properly/leading to near-suicidaldisappointment among scheduled investigators, whomay have planned for years. Limiting the scope of thissection is consequently not so difficult. I discuss onlythose future systems that have a geological impact, andwhich seem likely to fly.

The early years of the 21st century seem likely not tosee great innovation in land-orientated remote sensing,except perhaps for the wider availability of hyperspectral

66 Chapter 3

data. The Lewis satellite, which was to carry a 384-bandimaging spectrometer, failed to achieve orbit in 1998.Perhaps that is just as well. The vast bulk of geologistsare not 'up to. speed' with Landsat, let alone withhyperspectral capabilities of imaging spectrometers. Thesame cannot be said for atmospheric and oceanographicstudies, which are farmore advanced. The growing prob-lems of global warming, pollution and environmentalchange have forced a change of focus from simplemeteorological data to the application of arcane physicallaws in the quest for means of understanding what theEarth is currently going through. For most users of land-orientated remotely sensed data, the most pressing needis for assured continuity of tried and tested systems.The Landsat series is guaranteed up to the launch of theseventh platforn1 in the series, with a few modifications,and SPOT and IRS are similarly ensured.

The 1990s witnessed great activity, with new ven-tures into multifrequency ,multipolarization and select-able depression angles for SAR imagery, imagingspectrometry and multispectral thermal infrared sys-tems, all tested using airborne platforms. Apart fromthe possible development of active systems aimed atexciting mineral fluorescence and improvements ingaq1ffia-ray spectrometry, there seems little likelihoodof entirely new developments in the foreseeable future.Instead, much aspiring research depends on the carryingof experimental systems into orbit for further testingbefore truly operational satellites are launched, Theseambitions centre on the joint USA/European/JapaneseEarth Observing System (EOS) project, which plansup to three unstaffed polar-orbiting platforms startingin 1999.

The bulk of plans for the EOS are not high-resolutionland-orientated systems, but three are of immediate inter-est to geologists, although not all are equally assured.The first EOS platform (Eas AM-I, renamed Terra aftera public competition in the USA for an imaginativename) successfully entered orbit on 18 December 1999,where it carries a development from the principle of theGeoscan and JERS-l systems; that of using a few narrowwavebands strategically placed around mineralogic-ally useful spectral features. The Advanced SpaceborneThermal Emission and Reflection Radiometer (ASTER)incorporates three spectrometers; one for the visibleand VNIR with three bands (15 m resolution), one for theSWIR region with six bands (30 m resolution), and oneaimed at mineralogical features in the thermally emittedpart of the spectrum with five bands (90 m resolution).The ASTER also uses a forward-pointing VNIR sensor toprovide stereoscopic imagery. Details of the band posi-tions and widths are in Table 3.4. At the time of writingASTER had produced no scientific data, but was func-tioning properly. The goal of the mission is to generatecloud-free cover for the entire continental surface during

its 6-year planned lifetime; it could supplant Landsat for

many geological applications.The A VIRIS imaging spectrometer has had such success

that a similar instrument-the High Resolution ImagingSpectrometer (HIRIS)-seems an ideal candidate for EOS,particularly with the unseemly demise of Lewis. Technicalconstraints in planning, mainly to do with weight andpower requirements, however, have resulted in its defer-ment to later EOS platforms, which are not yet assured.The HIRIS will incorporate 192 bands (Table 3.3), suf-ficient for adequate discrimination of many importantspectral features. The images will be 800 pixels wideat a resolution of 30 m, giving a swath 24 km wide.Similar technology is incorporated in the ModerateResolution Imaging Spectrometer (MODIS) but this hasa 500-m ground IFOV that makes it of marginal interestto geologists. The MODIS is aboard EOS AM-I (Terra).In April 2000, NASA launched a low-cost satellite (EO-I)to test follow-on instruments for the Landsat series, butit also carries a narrow-swathe (7.5 km) hyperspectralsystem called Hyperion.

One of the experimental victims of the aftermath of theChallenger disaster was lengthy delay in the testing of afollow-on to SIR-B, the SIR-C/XSAR system with 3.0,5.3and 24 cm wavelengths, which could operate in four dif-ferent polarization modes and with variable depressionangles. Experiments with multispectral, multipolariza-tion SAR systems, such as SIR-C/XSAR, have had mixedsuccess in geological applications. Such a system wasproposed initially for EOS; but is in limbo at the timeof writing. There are considerable problems with SARsystems in or~it, not the least of which are weight, highpower consumption and high demands on data trans-mission. NASA launched a Space Shuttle Mission dedic-ated to measuring topographic elevation over most ofthe land surface using radar interferometry (Chapter 7)(Shuttle Radar Topography Mission-SRTM) in February2000. As discussed in Chapter 8!digitalelevation models(OEMs) are sources of an immense amount of informa-tion relating to terrain. That from the SRTM will have aspatial resolution of about 30 m, with an elevation pre-cision of 10 m, and promises to add a new direction inmany geological studies.

Aerial photographs continue to be the workhorses forgeological applications, because of their high resolution,as you will see in Chapter 4. However, costs are risingand storage of film has always posed problems. Similarresolutions are now achieved by entirely digital cameras,and the data can be stored compactly on CD-ROMs aswell as being designed for computer analysis. The sametechnology potentially has a large market for illustrat-ing news items, and several commercial outfits plan tolaunch 'mediasats' with resolutions of the order of 1 m.Most, such as Earthwatch, Eyeglass and Space Imaging,will capture stereoscopic images to serve increasingly

How Data Are Collected 67

sophisticated cartographic and planning requirements.Clearly they will find a ready use among geologists-ifthey are cost-effective--particularly for areas where it isdifficult to arrange aerial surveys.

Geologists seem to have rosy prospects in remotesensing for the next decade. This period is likely to beone of consolidation rather than innovation, giving themajority of geologists the time to get to grips with whathas been happening over the last three decades in geo-logical remote sensing research, to apply the new data toexciting new geological problems instead of repeatedlyporing over tiny test areas, and to catch up with theircolleagues in other fields. The first step along this roadis to review the basic methods of image interpretationusing simple photographic data from both aircraft andsatellites in Chapter 4.

Associated resources on the CD-ROM

You can access the resources by opening the HTML fileMenu.htm in the Imint folder on the CD using your Webbrowser. More detailed information about installing theresources is in Appendix C.

On the Menu page, select the link Miscellaneous:Image data, which will take you to a page that givesexamples of many kinds of image.

Further reading

Elachi, C. (ed.) (1983) Micro~ve and infrared satellite remotesensors. In: Manual of Remote Sensing (ed R.N. Colwell), 2ndedn, pp.571-650. American Society of Photogrammetry, FallsChurch, Virginia.

Freden, S.C. & Gordon, F. (1983) Landsat satellites. In: Manualof Remote Sensing (ed. R.N. Colwell), 2nd edn, pp. 517-570.American Society of Photogrammetry, Falls Church, Virginia.

Goetz, A.F.H. & Herring, M. (1989) The high resolution imag-ing spectrometer (HIRIS) for Eas. IEEE Transactions onGeoscience Remote Sensing GE-27, 136-144.

Goetz, A.F.H., Vane, G., Solomon, J.E. & Rock, B.N. (1985)Imaging spectrometry for earth remote sensing. Science 228,1147-1153.

Hiller, K. (1984) MOMS-O1 Experimental Missions on SpqceShuttle Flights STS-7 June '83, STS-11 (41:-B) February '84. DataCatalogue. DFVLR, Overpfaffenhofen.

Jensen, H., Graham, L.C., Porcello, L.J. & Leith, E.N. (1977)Side-looking airborne radar. Scientific America 237,84-95.

Kahle, A.B., Shumate:, M.S. & Nash, D.B. (1984) Active airborneinfrared laser system for identification of surface rock andminerals. Geophysical Research Letters 11, 1149-1152.

Lowe, D. (1980) Acquisition of remotely sensed data. In:Remote Sensing in Geology (eds B.S. Siegal & A.R. Gillespie),pp. 48-90. Wiley, New York.

Lowman, P.D. (1980) The evolution of geological space photo-graphy. In: Remote Sensing in Geology (eds B.S. Siegal & A.R.Gillespie), pp. 91-115. Wiley, New York.

Moore, R.K., ed. (1983) Imaging radar systems. In: Manualof Remote Sensing (ed. R.N. Colwell), 2nd edn, pp. 429-474.American Society of Photogrammetry, Falls Church, Virginia.

NASA (1976) Landsat Data Users' Handbook. GSFC Document76S05-4258, NASA Goddard Space Flight Center.

Norwood, V.T. & Lansing, J.C. (1983) Electro-optical imagingsensors. In: Manual of Remote Sensing (ed. R.N. Colwell), 2ndedn, pp. 335-367. American Society of Photogrammetry,Falls Church, Virginia.

Slater, P.N., ed. (1983) Photographic systems for remote sens-ing. In: Manual of Remote Sensing (ed. R.N. Colwell), 2nd edn,pp. 231-291. American Society of Photogrammetryi FallsChurch, Virginia.

Vane, G. & Goetz, A.F.H. (1988) Terrestrial imaging spectro-scopy. Remote Sensing of Environment 24, 1-29.

Vane, G. & Goetz, A.F.H. (1993) Terrestrial imaging spec-trometry: current status, future trends. Remote Sensing ofEnvironment 44,117-126.

Vane, G., Green, R.O.; Chrien,T.G., Enmark, H. T., Hansen, E.G.& Porter, W.M. (1993) The Airborne Visible/Infrared Imag-ing Spectrometer (A VIRIS). Remote Sensing of Environment 44,127-143.

Allison, L.J. & Schnapf, A. (eds) (1983) Meteorological satellites.In: Manual of Remote Sensing (ed. R.N. Colwell), 2nd edn,pp. 651-679. American Society of Photogrammetry, Falls

Church, Virginia.Chavez, P .5. & Bowell, J .A. (1988) Comparison of the spectral

information content of Landsat Thematic Mapper and SPOTfor three different sites in the Phoenix, Arizona region. Photo-grammetric Engineering and Remote Sensing 54, 1699-1708.

Chiu, H. Y. & Collins, W. (1978) A spectroradiometer for air-borne remote sensing. Photogrammetric. Engineering and Remote

Sensing 44,507-517.CNES (1982) SPOT Satellite-Based Remote Sensing System. CNES,

Toulouse.Doyle, F.J. (1985) The Large Format Camera on Shuttle mission

41-G. Photogrammetric Engineering and Remote Sensing 51, 200.

Photographic tone refers to the colour or relativebrightness of parts of the surface making up a scene. In ablack and white photograph it is expressed as differentshades or levels of grey. Without tonal variations theshapes, patterns, textures and context of objects couldnot be seen. As it is related to the reflectance propertiesof surface materials, tone depends on the part of thespectrum covered by the imaging system and the com-position of the surface itself. Tone also may be affectedby processing and printing, and by the illumination con-ditions too. For example, the same surface will exhibitdifferent tones on slopes facing the Sun, those facingaway and those in shadow. As a result the absolute toneis of less use in interpretation than the relative tonal dif-ferences between different objects. For instance, a standof deciduous trees in full leaf will have a lighter tonethan conifers, irrespective of lighting conditions or themethod of printing. Except under unusual circumstancesin some arid or semiarid climates, rocks are generallymasked by soil and vegetation. Even when cropping outat the surface they are frequently veneered with lichens,mosses or inorganic patinas. As a result rocks only veryrarely show their true colour or tone. None the less tonaldifferences do play an important role in lithological dis-crimination on photographs (Section 4.2).

Texture is a combination of the magnitude and fre-quency of tonal change on an image. It is produced bythe aggregate effect of all the many small features thatmake up a particular area of surface. The texture of aforest may be made up of the effect of branches, indi-vidual trees and stands of individual species. The scaleand resolution of the image, however, will determinewhich of these dominates texture. On a Landsat MSSimage with 80 m resolution only stands of trees willaffect texture, but on a low-altitude aerial photograph allthe attributes of the forest except the leaves will con-tribute. Likewise, the texture associated with a particularrock type will change with scale and resolution.

Patterns on an image result from the spatial arrange-ment of the different tones and textures that make it up.They may be arrangements of vegetation, topographicfeatures, drainage channels or different types of rock,which can be traced discontinuously over a significantarea. The patterns used most commonly are those relat-ing to drainage networks (Section 4.1.1)/ which are often

Photographic images are the cheapest and most readilyavailable remote-sensing data for geological interpreta-tion. They may be acquired directly by cameras mountedon aircraft or satellites, or produced from digital imagedata. Images in this form, either black and white orcolour, are available for all parts of the EMR spectrumaccessible to remote-sensing systems. For simplicity,however, this chapter deals only with those displayinginformation from the visible and near-infrared regions.Analysis of longer wavelength data and those in digitalform is covered in Chapters 5, 6 and 7. The simple photo-geological principles outlined in this chapter 'form thebasis for interpretation of any images, whatever their

origin.The majority of images used here are vertical black

and white photographs from aircraft, together with somesingle-band photographic hard copy from the LandsatMSS and RBV systems. For many users, colour photo-graphs are a luxury, but they do have certain advantagesstemming from the human eye's better chromatic dis-crimination. A few natural and false-colour photographsand some discussion of their information content areincorporated in the chapter. Colour imagery is discussedat greater length in Chapter 5, however, because theresults of computer processing of digital data are nearlyalways displayed in colour. Single photographs canbe interpreted in two dimensions, indeed for Landsatimages this is the only option, except for small stripsof overlap between adjacent paths. However, stereopticviewing of overlapping photographs adds the extradimension of relief, which usually is exaggerated. Aswell as showing topography, stereoptic viewing per-mits the measurement. of differences in altitude andthe calCulation of slope angles. These factors are of greatimportance both in highlighting geological featuresand enabling field sites and traverses to be locatedaccurately. Where possible, therefore, the images in thischapter are in the form of stereopairs, set up for viewingwith a pocket Jens stereoscope. Appendix A containssome discussion of stereometry.

Interpretation of photographs for any purpose relieson several basic characteristics of surface. These are tone,texture, pattern, shape, context and scale. They are allmore or less qualitative attributes, and their use is verymuch a matter of experience and personal bias.~

Many of the figures in Chapter 4 are stereopairs of aerial photographs set up for viewing with a standard lens stereoscope. If you donot have stereoscope, or find difficulty In achieving stereoptic 'fusion', you will find all the stereopairs as anaglyphs in the ImageInterpretation in Geology directory on the CD, within a separate folder. Use the anaglyph viewer packed with the CD to view them inany graphics package, such as Adobe Photoshop or MS Photo Editor.

68

Photogeology 69

to underlying rock type or geological structure.

Precisely how particular tones, textures, patterns and..

..Their context is important. A V-shapedin the Sahara Desert is unlikely to have been

by glacial action, and most probably resultswater erosion during sudden floods. Similarly,-feature in an area of metamorphic rocks is

.By the same token, the scale of

a karstic feature, but may well be a product of,- --,- -

The criteria outlined in the following sections dependvarious combinations of these basic characteristics

., -J

together with topographic relief expressed in" viewing, where possible. Obviously, tone,

partly

may change depending on the time of day andIn many cases there is an optimum

a best season when geologically relatedVery subtle slope changes

to geology in a flat terrain may be highlightedlow Sun angles or by remnants of snow or frost

--'-- -depressions.. The same viewing conditions

rugged terrain may degrade the interpretability ofmuch shadow or snow relics, so

Asrocks vary in their porosity, and because

that were taken after rains rather

by the influence of dense crop canopies. These-,. --

are made. The transp<»"t and evenmal deposition oferosional debris has a contrary effect. It masks feamresrelating to the hard rocks buried by it. These two funda-mental aspects of the Earth's surface processes producetwo basic types of landform. Destructionallandforms laybare to some degree the strucmre and composition of con-solidated and lithified rock, and indicate the dominantenvironmental influences over the development of land- -scape. Constructional landforms result from a superficialveneer of relatively recent rocks and sediments. Theirsurface attributes also give clues to environmental pro-cesses as well as to the namre of the veneers themselves.

In this section the emphasis is on destructionalland-forms, as they lead into the criteria for 'hard rock' inter-pretation covered in Sections 4.2-4.4. Constructionallandforms and their associated superficial deposits areconsidered in Section 4.5.

Different rock types respond to weathering anderosion in different ways. This depends on many factors.The most important are the composition and degreeof lithification of the rocks themselves, together with thetype of weathering and erosion. These last two factorsare mainly determined by climate, and sometimes byaltimde. A region in the tropics may have a hot, humidclimate near sea level, but mountainous areas experienceconditions somewhat similar to those of temperatelatimdes. Likewise, mountains with a moderate rainfallmay flank an arid or semiarid desert.

An important criterion in geological interpretation OfJimages is the degree to which different rock types resisterosion. A sandstone strongly cemented with an inertcompound such as silica will be resistant under virtuallyany conditions. A rock containing reactive minerals suchas feldspars or carbonates, however, will soon succumbto the agents of chemical weathering that predominatein hot humid climates. The same rock may respond inthe opposite fashion in an arid or cold climate wherechemical weathering is less significant.. A rock's resist-ance to erosion is also relative. A basaltic dyke intrud-ing soft mudstones may give rise to a prominent ridge,but where it cuts gneisses it may have a relatively lowresistance and occupy a narrow trough.

There are four important types of erosion that canattack the products of weathering-those associated withflowing water, wind, wave action and moving ice. Marineand wind erosion can produce distinct destructionallandforms, but they are of minor importance in imageinterpretation. Marine erosion is restricted, for obviousreasons, and most geological investigations are con-ducted well away from the sea. Aeolian erosion has arelatively minor effect at the scale normally used inremote sensing. The major sculpting agents are flowingwater and ice. Water dominates, even in the most aridof modern environments, either because the main land-forms were produced during more humid episodes or

parts of the world dry season images are best

Destructionallandforms

rock-sculpting activities of the various agents of

It is

70 Chapter 4

water continues to flow at the surface after intermittentbut spectacular rainstorms.

a less obvious relationihip to structure. By no meansall streams conform to this simple classification. Mostbear no obvious relationship to geology or former landsurfaces. They are referred to as insequent streams.

The evolution of a landscape can be complex, particu-larly during protracted or episodic uplift and downwarderosion. A stream system may begin by developing on ageologically simple surface of uniformly dipping strata,producing consequent, subsequent and other streams.This simple sequence may be merely a veneer unconform-ably above more complex older rocks. The older rockseventually may be crossed by drainage that bears littleresemblance to their structure, but is inherited from anearlier, higher level system. This results in a superim-posed drainage system. Relics of it may persist even afterprolonged erosion establishes the control of the deepstructure, leaving many insequent streams. Figure 4.1shows the main classes of stream in a diagrammatic way.Figure 4.2 is a stereopair of aerial photographs illustrat-ing real examples and their use in interpretation.

In areas where tectonic activity resumes after theestablishment of a drainage system, the old streamsmay continue in their original courses across the axes ofuplift or folding to produce antecedent drainage. Thiscan happen if the uplift is quite slow or if the streamsinvolved are sufficiently powerful to erode through therising crust fast enough. The best known examples ofantecedent drainage are those of the Colorado Rivercutting through the Kaibab Uplift to form the GrandCanyon (Plate 4.1), and the Indus and Brahmaputrawhich flow southwards through the Himalaya, despitethe phenomenally rapid rates of uplift in this tectonic-ally active area. Active tectonic uplift in many cases is toorapid for antecedent drainage to develop, and streamsmay be reversed 01: ponded to form lakes.

Any cycle of stream erosion passes through variousstages, which relate to how close the drainage systemhas adjusted to its erosional base-level. The famousgeomorphologist W.M. Davis compared these stages tothose in a human life, and his general terminology is stillin common use today. A cycle begins when the base-level falls, either through uplift, tilting, fall in sea level,down-cutting of a major river, increase in stream flowor various permutations. Such a change is referred to asrejuvenation~ In the initial stages the pr~vious drainagesystem remains or a new one develops in depressionsin a I;lewly emerged land surface. The youthful stage ischaracterized by active erosion and the formation ofdestructive landforms. Streams flow in V-shaped valleys,and they almost completely occupy the valley floors.Maturity begins as the stream ~ystem begins to developwide valleys as a result of the destruction of ridgesand the beginning of deposition. Mature str~ams havereached a le~thwise profile in equiUbrium with .thebase-,level. Old age is characterized by the dominance of

4.1.1 Landforms and dr~inage

Except in strongly glaciated areas, drainage patterns con-form to some degree to the regional slope of the terrainand to its underlying geological structure. Because ofthis it is often possible, even where outcrop is poor, torecognize the gross structural elements. The first stepin using drainage, usuaUy the most obvious elementof a landscape, is to seek genetic connections betweenstream patterns and the disposition of rocks. This involvesassessment of the geomorphological history of the areaand the stage of d~velopment of landscape dissection. Itis not possible tQ catal9~e all possible cQJnbinations ofrock type, structure and stream development, or to out-line all the many methods of describing drainage. Thissection uses a simple classification based on the directionof stream courses relativ~ to geological controls.

A channel that has developed on some initial slope,and which continues to follow its original directionis said to be consequent. The original slope may haveresulted from regional tilting of an earlier peneplaneds~face, or may result 4irectly from the direction .of dipof~strangraphicunit. The point to remember is that theterm relates only to the topographic, and not any struc-turalaspect of a consequent stream's origin. Streams thatclearly flow down the dip of a particular stratum or litho-logical boundary are termed resequent streams. They have..developed

ona surface already stripped by earlier, con-sequentstre~s and are structurally controlled.

Streams propagate themselves and develop tributariesby head ward erosion. They tend to follow the lines ofleast resistance, On a surface underlain by dipping stratathat resist weathering and erosion differently, somewill follow lines determined by the least resistant layers.These are subsequent streams. By the same token, sub-sequent streams- can follow and etch out the lines offractured rockd~veloped along faults. Consequently thestrike of. strata and fault lines, so important in unravel-li~ geological S;trlicture, are most likely to be reflectedin subsequent stream courses.

Continued downward and headward erosion even-tually begins to etch out the basic geological fabric ofan area, in the form of cuestas, valleys and other topo-graphic features. Slopes are produced on the valleysides of original consequent and subsequent streams inconcert with the underlying geological div~rsity. Theymay be in several directions. Streams that follow themare obsequent. In areas of dipping strata, slopes developin subsequent stream valleys, which are in the oppositedirection to the local dip. Small obsequentstreams fre-quently develop on suchsl9pes. Others may flow downthe sides of major consequent orresequent valleys, with

~

(a)'--~~

L.i.:_~..:. .r<;

7,~

s 0

(b)

These

two diagrammatic sketches of how underlying., ,-

streams, respectively. That marked '?' is possibly a consequentstream that developed upon a dipping planar surface, whichhas been incised into the underlying tilted strata during uplift.Reproduced with the permission of McGraw-Hill Books Inc.

Many of the dry stream channels in this semiarid of Agriculture, Agricultural Stabilization and ConservationService; CSL-Clyde Surveys Limited, UK; DNDE-Department of National Development and Energy, Australia;EMRC -Department of Energy ,Mines and Resources, Canada;IGN-Institute Geographic NatiQnal, Paris; NCIC-NationalCartographi~ Information Centre, US Geological Survey;USGS-US Geological Survey. Where possible;photographreference numbers are shown.

: 40 000. Source: ASCS UPRM-13 20 + 21.

72 Chapter 4

wide flat flood plains with meanders and redistributionof superficial sediments. The final result is a peneplaindevoid of control by underlying geological structures..The mature and old-aged stages produce constructionallandforms (Section 4.5).

How quickly an erosional cycle passes through thesestages depends on a number of factors. The closer astream is to the base-'level, the faster it reaches equilib-rium. Mor~ remote areas pass from the initial to youthfulstages at rates dependant on headward erosion in themain streams. Easily eroded rock types will achievematurity more quickly than resistant ones. In an area ofvaried lithologies youthful features may be mixed withthose of maturity. The rate of erosion, determined byboth volume and rate of flow, and by the erosive contentof solids in streams, plays its part too. Fall in base-level israrely instantaneous, and the rate of change also playsa role in the evolution of 'landscape. Episodic changesare common, and their interruption of the progresstowards equilibrium canbe marked by several levels in adrainage system, all characterized by youthful, matureand aged features. In such cases, the boundaries betweenlevels can be marked by sudden changes in the longit-udinal profiles of streams.

Adding to the complexity of landforms dominatedby flowing water is the opposite of rejuvenation, whenth~ base-level rises. This hastens 1he maturation of adrainage system: and the onset of depositional features.Such aggradational episodes may indeed be interspersedwith those of -rejuvenation. Further consideration ofthese complex geomorphological processes is beyondthe scope of this book, but are covered in many moderntexts on physical geology.

The shapes of valleys in cross-section; which areoften apparent in stereoptic views of photographs, canbe important-criteria in identifying underlying litho-logies. Similarly, the drainage density can play animportant discriminatory role (Section 4.2). However,both are strongly affected by climate. Surprisingly, drain-age density in an arid area on any particular rock typeis usually greater than one developed under humid con-ditions. Less obvious is the contrast in valley profilesbetween humid and arid terrains. Humid conditionsencourage the rapid formation of thick soil profiles.Lubricated by water, these tend to creep slowly down-slope, thereby limiting slope angles and helping smooth"the overall terrain. Under arid conditions neither soilnor lubrication is present to any marked degree. Con-sequently much steeper slopes to valley sides can besustained, even in rocks that are not strongly resistantto erosion. Moreover, most streams in deserts are inter-mittent, and when they do flow it is often during storms.Under these conditions downward erosion is rapidbut short-lived. It produces slot-like gorges along mainchannels, and, because torrential rainfall contributes

mainly to surface run-o~, innumerable rills and gullieson the slopes. The soil and vegetation mantle in humidterrains slows surface run-off and encourages infiltration.Gullying is therefore more subdued, if present at all.Figures 4.2, 4.6, 4.7 and 4.8 show the contrasts betweentypical humid and arid terrains for these two importantcharacteristics.

As implied earlier in this section, the patterns de-veloped in a drainage system as it evolves containimportant clues to geological structure. They can alsomake some contribution to distinguishing some rocktypes. These aspects are considered in Sections 4.2 and4.4. Here the main pattern types are described andillustrated with some typical examples.

Drainage patterns of all types may be found on avariety of scales, from those defined by major river sys-tems to those associated with rills and gullies. They helpdefine the textural features of a scene, and dependingon the spacing of the individual streams may be definedas coarse-, fine- or intermediate-textured patterns. Thestreams that define the patterns can be of any genetictype: consequent, insequent, subsequent and so forth.

Areas drained primarily by insequent streams show alack of any preferential channel direction. The streamsflow in irregular, branching courses with many randombends. Tributaries enter larger streams 1;1sually at an acuteangle to the general direction of regional flow, itselfdetermined by regional slope. The pattern produced islike the complex branching of a tree, and because of thisit is termed dendritic (Fig. 4.$a). Dendritic drainage pat-terns are characteristic of terrains showing lithological,structural and topographic homogeneity.

In contrast, drainage may be dominated by a numberof parallel or subparallel streams, indicating a unidirec-tional flow. For obvious reasons this is termed a paralleldrainage pattern (Fig. 4.3b). Regular deviations ofcourses from this result in fan patterns, which reach theirultimate in a completely radial pattern, where streamsall flow away from a single point (Fig. 4.3c). The oppositecase, where streams flow radially towards a commonpoint (Fig. 4.3d) is called centripetal drainage, Suchsimple patterns all derive from topographic featuresand cannot be assumed to reflect any particular structureor lithology, although this is often revealed by closer

inspection.Other categories of drainage pattern can be regarded

as embellishments upon the last four. They include sub-sequent streams that have been controlled by lithologicaland structural weaknesses, and so give more geologicalinformation. Patterns dominated by many straight streamswith a variety of trajectories may define rectangular orangular patterns, depending on their angular relation-ships (Fig. 4.3e). A particular variant of these two types,where short tributaries connect to long straight majorstreams produces a pattern reminiscent of trained vines

Photogeology

73

(b)"" \ ~~~

~

i~I~

\/{'~

/ I

(d)

\~

-~

~

specific elements may be picked out by the interpreter.Typically, an overall pattern or series of patterns may con~tain streams that do not cQnform to the overall. picture.They are anomalous, and attract attention simply becausethey are different. Such drainage anomalies ill an areaof clearly defined geology may highlight geolQgical sub-tleties, which otherwise would go unnoticed ina p~relyfield study. On the other hand, in areas where geologyis obscured because dissection is not well advanced or athick blanket of superficial sediment masks the under-lying structure, anomalies can give clues leading togeological revelations.

Whatever patterns are displayed on a p4otograph, it isalways important to bear in mind that they reflect thefull range of factors behind the development of land-scape. Climatic, topographic, lithological, and stru~turalcontrols all combine together with the stage reached inthe evolution of the drainage system. In poorly-exposedareas a crude idea of the structure can be appraisedbest simply by tracing the drainage patterns revealedby a photograph on to a transparent overlay. Sooner orlater, however, interpretations have to be confinned ande~tended by field studies.

//~~

~

K"

/

(hi

~'~--v

I;;

/"\- r

Patterns

formed by drainage networks can take on-,

4.1.2

Glacially eroded landformsMoving

masses of ice are the most P9werfullandscape-sculpting agents imaginable. Although velocities ofice

flow are very slow, even in the most active valleyglaciers, the enormous masses involved ensure that their

energy is enormous. During the Pleistocene glaciationsit is estimated that some of the continental ice sheetsexceeded

3 kIn in thickness. The process of glacial erosionbegins with the plucking of debris from the substrate,

often weakened by freezing and thawing in advanceof the ice front. The incorporation of this debris in iceprovides

the ice base with a renewable supply of abras-ive material, which acts in the manner of a giant rasp.No rock- type can resist for long the protracted passageof

ice above it. The destructive landforms that resultdepend for their shapes and patterns on the type of

glaciation involved. There are two main types-,-,valleyand continental glaciation.

Valley glaciation begins with the accumulation of per-manent snow masses in depressions on mountain slopes.There

it undergoes a transformation to ice through a seriesof stages. Once a certain mass of ice has accumulated itis able to flow downslope under its own gravitational

potential energy. The climatic conditions suitable forice accumulation ensure that exposed rock on peaksundergoes

powerful physical weathering, dislodgingangular fragments and thereby 'sharpening' the peaks.These

fragments falling on tp the ice, together withthose plucked from the base of the ice, contribute to theabrasive

capacity of the glacier. At its source the glacier

4.3f) referred to as a trellis pattern. Subsequenton a radial pattern result in a con-

series of valleys defining an annular pattern(Fig. 4.3g).

In reality, the several basic pattern types are rarelydeveloped to perfection. Several generally combinetogether to give topologically complex systems, in which

74 Chapter 4

Fig; 4.4 The Highlands of Scotland contain many examples of the products of upland glacial erosion. This stereopair shows typicalcirques, aretes and hanging valleys, as well as the rugged bare rock surfaces resulting from subglacial plucking. Scale 1 : 25 000.Source: author.

carves a bowl-like depressioR or cirque by the localmovement of ice through a vertical arc (Fig. 4.4). Onceit has spilled over the lip of this bowl, the ice moveSdownslope, usually along a pre-existing valley.

Valleys are eventually remoulded..to a V-shaped cross-section and straightened by the removal of interlockingspurs. Apart from their typical cross-section, glaciatedvalleys can be recognized easily by the abundance oftruncated spurs.. The other effect of glaciation is deepen-ing of the earlier valleys, the amount depending onthe energy of the glacier responsible. The pre-existingvalley patterns determine toa marked degree the courseof later glacial flow paths, which comprise major valleyswith tributaries. The energy difference$ involved in theglacier system result in the courses of less powerfulglaciers being less deeply eroded than those of majorice streams. This produces hanging valleys at. the con-fluences (Fig. 4.4). Because valley glaciation is gener-ally superimposed upon pre-existing stream and river~ourses, except in the Arctic and Antarctica, the remain-ing patterns conform to some extent to those of theirprecursors. The difference is that the minor valleys arewiped out and those remaining are grossly straightel;led.The net result is to preserve some of the indicators ofgross geological structure, but to erase those indicatingthe details.

Examination of large-scare photographs of glaciallyeroded uplands enables some of the smaller featurestypical of ice erosion to be recognized. Among theseare roches moutonees and major ice striations (Figs 4.4and 4.5). The typical shape of roches moutonees, withsmooth, rounded surfaces facing towards the former iceflow and steep, glacially plucked faces pointing downflow, enables estimates of ice flow directions to be made.Striations give similar information, but are ambiguousat these scales. In upland areas, ice-flow directions aregenerally obvious from the overall topography and therelative orientation of glaciated valleys, and these minorfeatures are of more importance in glacial analysis of flatareas of continental glacial erosion.

During the Pleistocene huge ice sheets, originatingfrom snow accumulation at high latitudes and to a lesserextent from ice-capped mountain ranges, spread south-wards over large tracts of northern Eurasia and NorthAmerica. They produced two main types of terrain,those dominated by glacial erosion and those where theproducts of erosion were deposited. The former are nowthe enormous rocky wastes of the Canadian, Baltic andSiberian Shields, from which most of the previous coverof Phanerozoic sediments was stripped. This revealed theunderlying Precambrian basement. The areas dominatedby glacial deposition occur further to the south where

~~spite!he ~ard~ess of .the ~et~morphosed darker metavolcanic rocks, with the granite having resistedglaciation better. The intricate jointing in the granite, itsintrusive contact and basic dykes cuttIng it have beenselectively exploited by glacial action resulting in linear,vegetation-filled depressions and narrow lakes. Scale1 : 50 000. Source: EMRC A13781 52 + 53. Copyright HM theQueen in Right of Canada (195~.

The forward move- transient but massive load of the former ice sheets wassufficient to depress the crust hundreds of metres belowits original level. Slow post-glacial isostatic rebound,at different rates and to different extents depending onthe amount of glacial depression; has imparted regionalslope to the shields. These ensure that the drainage doeshave an overall orientation, albeit tortuous.

loads of debris. The typical features of theseareas are considered in Section 4.5. Here

comments about the erosional features of large,

Unlike upland glaciers, for which erosional features

sheets have a uniform and dramatic effect on the

the geologist. The all-pervading erosionby the debris-loaded ice base emphasizes every con-ceivable heterogeneity in the rocks over which it passes.The softer rocks and zones of structural weakness areetched out; sometimes completely, leaving the moreresistant rocks as positive features. Pre-existing drainagepatterns are obliterated, so that when the ice retreatsnew drainage is controlled by the intricacies of the newlyscoured surface (Fig. 4.5). These patterns can be long-lived, simply because the grinding flat of the land sur-face reduces stream energy toa minimum. However, the

4.1.3 Other destructionallandforms

Marine erosion is capable of producing landforms that tosome degree are controlled by the underlying geology.They are restricted, however, to a narrow coastal swath,and in general are obvious only on large-scale photo-graphs. In somecaSe8, a rise in sea level and flooding ofa landscape formed by other processes can reveal grossgeological elements from the patterns of the islands leftabove sea level. Where geological interpretation of coastalareas is demanded, the reader is referred to existing textson coastal geomorphology.

Erosion by wind in deserts is indeed an importantprocess, but its products are generally Visible only in thefield. The reason for its relatively minor importance to theimage interpreter is that wind energies are low compared

76 Chapter 4

with those of flowing water. Because of this only a smallload of abrasive particles can be carried. Wind erosionis rarely powerful enough not at a sufficient rate to over-whelm the dominant effects of stream erosion in deserts,even though streams flow only intermittently.

4.2 The recognition of rock types

The next. set of steps. depends on the systematicdescription of the probable areas of bedrock. If possible,the terrain should be examined closely for any obviousboundaries between different types of surface. After apreliminary division of contrasted surface types, eachcan be described using the following key headings.1 The topography should be described, in particularnoting the relative resistance to erosion of the unit andany anomalous features.2 What is the drainage pattern and what is the spacingbetween individual channels?3 If possible the cross-sectional shape of the smallestdrainage features-'gullies-should be described.4 The vegetation cover and land use directly related tothe unit must be noted.5 Finally, if bare soil or rock outcrop is visible its photo-graphic tone or colour compared with other units, andany texturaHeaturesshould be described.

4.2.1 Sedimentary rocks

Sedimentary rocks are usually stratified owing to varia-tions in the conditions of their deposition. As a resultthey typically display a banded appearance on photo-graphs, although the width of bands depends on theirattitude as well as on the thickness of individual strata.As volcanic lava flows and pyroclastic rocks, some intrus-ive igneous rocks and most metamorphic rocks alsoshow layering, there is considerable scope for confusion.Avoiding this relies on the geological context of the lay-ered rocks being examined.1n most cases this is obviousfrom previous knowledge of the study area. In relat-ively unknown areas the context must be assessed firstby looking for signs of volcanoes, cross-cutting contactsand highly complex structures. In their absence, layeredrocks can be assumed to be sedimentary until field studyproves the contrary.

The topographic expression of sediments depends ona variety of factors.- Principally it is determined by theattitude of bedding, ranging from horizontal to vertical.This can result in a series of landforms from whichestimates of dip and strike can be made, but this isconsidered in detail in Section 4.3. Topography is alsogoverned by resistance to erosion, which is owed to avariety of factors in its own right.

Sediments can exhibit various stages of compaction andcementation. For example, a compact but uncementedsandstone is less resistant to some types of erosionthan one that is completely cemented. The porosityand permeability playa role too. A permeable rock willallow water to seep into it thereby lessening the effectof erosion by surface run-off compared with that on animpermeable rock. The main weak zones in all sedimentsare bedding surfaces. Their spacing makes an importantcontribution to l'elative resistance. So a thinly bedded

In order to recognize or distinguish between differentlithologies on a photograph the geologist needs twoprerequisites-a sufficient correlation between geology andlandform and a combination of experience, patience, per-ception and ingenuity. The attributes of photographic tone,texture, patterns, context and scale need to be weldedtogether with a knowledge of how different rock typesrespond in different climates to weathering and erosion.

In this Section fhebasic division is into sedimentaryrocks, extrusive and intrusive igneous rocks and meta-morphic rocks. To some extent the treatment pre-emptslater Sections by introducing some of the stratigraphicand structural criteria required in rock discrimination.Most of the distinguishing criteria, particularly thoserelating to textures and patterns, are equally useful inthe interpretation of images of other parts of the EMRspectrum covered in later chapters. As many rock typeshave different appearances under different climaticconditions, wherever possible they are illustrated withseveral images; It would be useful to give examples ofthe kinds of appearance most likely to be encounteredby the interpreter-unspectacular and poorly expressedon the image. However, the world is a diverse place, thisis an introduction and space is limited. The illustrationshere are the best that could be found.

There is no rigid routine for the recognition and dis-crimination of lithologies in. photogeological interpreta-tion. However, a systematic approach of one kind oranother is essential. It must be backed up by note takingand annotation of the images, so that a permanent recordis kept and experience is built up in a way that can bereferred to at any future time. As a guide to this approachthe following general scheme is recommended.1

The climatic environment has to be assessed-humidor arid, boreal, temperate or tropical.2 Is the vegetation, if any/natural or agricultural?3 The erosional environment must be assessed in termsof its energy/its stage of development and the relat-ive contributions of stream, glacier and wind erosionestimated.4 The area should be divided into parts that are domin-ated by superficial deposits and those probably havingbedrock at or close to the surface.S Areas of outcrop or near-surface bedrock should beexamined for signs of lithological banding or its appar-ent absence, and distinguished accordingly.

Photogeology

77

of the same composition exploitation of differenres in resistance can then leaveresistors as ridges of various kinds separated by parallelsubsequent valleys in the less resistant units. The netresult is a gross texture of more or less parallel elementsof the landscape (Fig. 4.2), characterized by some form oftrellis drainage pattern. The degree of parallelism becomesmore

nearly perfect the steeper the dip of the beds. Inmany humid terrains the ridges are wooded, whereasvalleys

are farmed.

than the coarser ones. This is because the

--the hydraulic action of moving fluids, their

.".

rocks. Rocks such as sandstone and silt-composed dominantly of quartz, which is both

and chemically inert, resist both mechanical and

rainwater containing

morphological features of sediments, their

on exactly the same fundamental factors as,Vegetation cover and outcrop tone or colour

-

need be considered in photogeology: coarserocks such as conglomerates and sandstones,-,' , , mudstones,and

and chemical sediments including lime-

nearly always composed..' ,The differences into

erosion impart a distinctive grain of.--

textures

depends

Coarse ~lastic Sediments

Sandstones and conglomerates are composed mainly ofquartz grains, with varying proportions of other resist-ant minerals such as feldspars and rock fragments. Theirdegree and nature of cementation varies considerably,and so does their resistance to erosion. When interbeddedwith equally cemented finer clastic rocks, however, theygenerally form positive features, often forming cliffs andsteep escarpments. Cements based on iron compoundsor silica are particularly strong, but carbonate cementsare prone to attack by acids in rainwater.

In arid climates, quartz grains dislodged by weather-ing are easily removed by wind. Sandstones and con~glomerates therefore tend to display bare surfaces underthese conditions (Fig. 4.7). Except in tropical rain forest~where laterites may develop, co,?-rse clastic rocks inhumid climates rarely develop de~p, fertile soils. Theirpermeability makes them too well-drained to supportextensive grass cover. As a result vegetation is relativelysparse and often dominated by trees with deep rootsystems (Fig. 4.7).

As many sandstones are very permeable, rairifall infiltr-ates them, thereby reducing surface run-off. This helpsensure {hat they have relatively wide.,spaced drainagetextures (Fig. 4.7b).Gullies ~re relatively rare and smallstreams tend to develop V -shaped cross-sections owingto the well-supported nature of sandstone's granularstructure. Being unable to deform except by brittle dis-location, one of the most distinctive features of coarseclastic sediments is their abundance of joints, often inseveral tectonically related directions. The zones of weak-ness associated with jointing exert a marked controlon the finer structure of drainage on sandstones, whichmay take on a prominent rectangular or angular pattern(Fig. 4.7a). Oiffsand cuestas in sandstone are frequentlyjoint-controlled.

The bareness of sandstone outcrops in arid areas,combined with their well-drained nature, usually per-mits them to show their true photographic tone or colour;unaffected by near-surface pore water.. Silica- andcarbonate-cemented sandstones are usually pale, but thosecontaining iron compounds can be any shade of yellow,orange, brown or red. In panchromatic black and whitephotographs red sandstones appear dark. The frequently

by distinctly stepped topography, espe-the resistors are more than 5 m thick. Each

controlled by a resistant unit and separated bysteep slopes underlain by soft rocks. In arid clim-resistors frequently form plateaux incised by major

flowing in steep-sided canyons with steppedwalls. Headward erosion of tributaries can isolatebuttes and mesas from the main plateau (Fig. 4.6a). Soilformation and its movement downslope under humidconditions mutes such topographic expression (Fig. 4.6b).The lack of original slopes in horizontally bedded terraintends to encourage the development of dendritic drain-age patterns. This, together with the tendency for majorunits to crop out parallel to contours results in complex'patterns and textures (Fig. 4.6). Such complexities areoften paralleled by vegetation communities, which maydiffer from one lithology to another.

Tilted sediments that have been incised by streamerosion provide the conditions whereby consequent, sub-sequent and other stream classes can develop. The full

:hapter

4

..-~.6

Both stereopairs show the most typical feature)ography developed upon horizontally layeredi-outcrops

that closely parallel topographic contours.d terrains (a) the intermittent violent erosion develops,-sided

gullies and valleys. Variable resistance to erosion,s case by sandstones and siltstones, leads to suddenges in slope angle and step-like features on the valley.One noteworthy feature is the restriction of bushes

ly one level in the succession. This probably is adary between water-carrying sandstone and underlyingrmeable

siltstone. In a more humid climate such as)f eastern Kansas (b) the topography is more muted.pectacular 'candy-stripe' appearance results fromjedded

white limestones and grey shales oflian age. Scales: (a) 1 : 20000 and (b) 1 : 20 000.ces: (a) CSL Series B 3 + 4 (b) NCIC Kans.2 A + B.

Ie forest cover of sandstones in humid climates also5 them a dark colour on panchromatic photographs.

lently show a strong fissility parallel to bedding,:h is accentuated by the reorientation of flaky clay~rals

during lithification. It is this fissile nature whichlcterizes shale, a term that is commonly applied to

fissile mudstones and siltstones. Fine-grained clasticnents may comprise a variety of grain sizes, some-

s in graded units, and show bedding on a scale fromto 1 m. However, except in detailed stratigraphicthey are usually lumped together as homogeneous; up to hundreds of metres in thickness.

Clastic

Sedimentslstones

and siltstones contain a high proportion of clay~rals together with varying amounts of quartz. They

esentdepositional environments with low energies.any cases they contain fine laminations between 1 mma

few centimetres thick. Because of this such rocks

Photogeology 79

main clue to the presenc~ of sandstone is the wide spacing ofdrainage and the roundness of topography resulting from itsoften weak cement and susceptibility to c;:hernicalweathering.Compared with fine clastic sediments it is resistant to physicalweathering, and is ~ot prone to solution like carbonates.Scales: (a) 1 : 20 000 (b) 1 : 33000. Sources: (a),USGS GS-WI-12146 + 147 (b) ASCSAPL-4 V 98 + 99.The

Cements can be of various compositions, but evenwhen totally uncemented, rocks with a high clay contenthave good cohesion owing to the electrostatic charges onthe clay mineral particles. The poor permeability of fineclastic rocks, however, means that in an unconsolidatedsequence of fine and coarse sediments rainfall infiltra-tion is at a minimum on the fine rocks and surface run-off has more chance of eroding them than the coarsermaterials. Irrespective of degree of cem.entation andclimate, therefore, fine clastic rocks are more prone to

erosion than coarse clastic rocks. They tend to be poorresistors and form negative topographic features..

As well as promoting relatively rapid erosion, high run-off results in the development of closely spaced drainagetextures. Clay-rich rocks are more able to deform plast-ically than sandstones and as a result tend not to showprominent joint systems, although they may have finelyspaced joints as a result of shrinkage. This encouragesthe development of randomly orientated streams in hori-zontal strata and an overall dendritic drainage pattern.

:hapter

4

:\

in dipping sequences, fine clastic units often display!ndritic element within an overall trellis pattern of

;equent and subsequent streams. The low resistanceosion, particularly when cement has been dissolved orkened, together with the continual exposure to sur-run-off

means that gullies in fine clastics soon develop.:\ arid climate gullies form during short-lived heavy

fall and aggregate to form a minutely dissected terrain,~lly walls retaining steep angles. The continuous

off on fine clastic;rocks in humid climates encour~the degradation of gully sides, the formation of less~ly spaced drainage and a rounded topography.rlY rock with a high clay content is likely to contain aigher complement of plant nutrients than those com-~d mainly of silica or carbonates. These are releasedIse

by plants during weathering and soil formation., soils are not only nutrient rich, but are able to retain\utrientsand

high moisture contents because of the

--

1.8 Gently dipping mudstones and siltstones typicallylop dendritic and closely spaced drainage patternsLIse of their poor internal drainage. In arid areas, such as(a), individual gullies have a sharp V-shaped cross-In despite the low resistance to erosion of fine clasticnents.

Paradoxically, drainage on shales in humidnes (b) is often more widely spaced than under dryitions.

As shown clearly by the stereopair of part of

inia,

the smallest channels have vanished as a result ofase of lateral erosion during continuous stream flow.'{ profiles are smooth and shallow. The high contentltrients

in shales together with their poor internallage helps ensure that they have a luxuriant natural

t cover~ Scales (a) 1 : 27000 and (b) 1 : 43000. Sources:SGSG5-WI-36 68 + 69 and (b) USGSG5-AZ-2119 + 120.

Photogeology 81

vegetation and are often intenselyIn arid climates the converse applies. Unlike

r -

scale of metres. Iron, the main colouring agentcan be chemically bound in the clay minerals

The low depositional energies of

This often results in a high proportion

a complex of iron oxides and hydroxides,, -"- --, .-which can impart strong

red coloration to outcrops.---

Provided

that resolution

.Chemically Precipitated Sediments

There are two main categories of chemical sediments,those precipitated by evaporation of restricted bodiesof water-highly soluble halite (NaCl) and other salts-and those precipitated mainly as the hard parts oforganisms,-calcite and dolomite, the carbonates of cal~cium and calcium with magnesium-which form variouslimestones.

Being highly soluble in water, evaporites rarely cropout in humid terrains. However, their presence in asequence may be detected by the haphazard presenceof more or less circular depressions where collapse intosolution cavities has occurred. Where the evaporites atein the form of salt diapirs, such collapse structures maybe arranged in distinctive patterns related to the shape ofthese gravity-driven structures. Near-surface salt mayaccumulate in soils as a result of capillary rise of salinesolutions. As few plants are sal.t-tolerant this may giverise to barren patches in otherwise fertile ground.

Evaporites are crystalline solids, and being devoidof pores do not allow infiltration of rainwater. In aridclimates, where they are attacked by only infrequentrain storms, salt bodies may crop ouL They generallyhave distinctive textures composed of finely spacedgullies with sharp promes (Fig. 4.9). Patches of residual

fine clastic rocks rarely crop.-" --

Figure 4.8 shows the typical appearance of horizontally.-.

--

Fig. 4.9 Anhydrite-rich evaporites in this area of Somaliadisplay many of the peculiar landforms that make them soeasy to recognize in semiarid terrains. The dip slope in theoverlying limestones shows an excellent example of changing

outcrop shapes controlled by varying valley-bottom slopesuperimposed on uniformly dipping strata. Scale: unknown.Source: unknown.

82 Chapter 4

(a)

(b)

Fig. 4.10 Prolonged chemical weathering of carbonates inhumid climates leaves a thin residual veneer of clay-rich soil,which is often quite fertile, as in Kentucky (a~.However, theextremely good internal drainage may preclude all but thelargesfsurface streams. The often low-lying surface isgenerally pock-marked by many depressions over sink-holes.In recently glaciated areas, such as northern England (b),soil development has not been possible over limestones

despite very humid conditions. Limestones have resistedglaciation better than associated sandstones and shales, toform prominent crags. Drainage is almost absent a~d the barerock surface is filled with innumerable solution-widenedjoints. Vegetation occupies only those surfaces veneeredwith glacial till. Scales (a) 1 : 25 000 and (b) 1 : 10000.Sources: (a) USGS GS-XS-6 38 + 39 and (b) MetropolitanCounty of West Yorkshire 42-68, 176 +177.

Photogeology

83

evaporitestend to resist erosion quite well com-

with the fine clastic rocks with which they are.-.'. They generally stand out. as posit-

features. When in the form of diapirs,nature of the intrusions often controls a

radial drainage pattern, sometimes with an..." .J, -'-

lesser extent dolomite, is slightlyin rainwater with dissolved carbon dioxide-

In humid climates, therefore, topographicfrom solution dominate areas under-

by carbonates. As carbonates generally are very.cemented, shrink in the course of lithification

most of which form by the introductioninto the carbonate structure during or

of a limestone, are yet more stronglybecause the calcite to dolomite transformation

..by a volume decrease. Solution is

conditions allows the d~velopment of extensive steep-sidedvalley systems in a glacially dominated topographyof rugged carbonate scars (Fig. 4.10b). Improvementof the climate 'unplugs' the subsurface drainage and theperiglacial drainage is left as dry valleys.

Carbonates may vary in purity, but generally containvery little clastic material. As a result they are cappedby very thin soils in humid climates, allowing theirinternal jointed structure to show up as distinctive pat-terns (Fig. 4.10b). The exception is in areas of very pro-tracted and intense chemical weathering, as in a tropicalrain forest, where residual material accumulates. As themain clastic components are clay minerals, the result-ing soils are poorly drained and dense vegetationmay accumulate on a terrain almost completely lackingin large streams. In many cases the soil is washed intosinkholes, where particularly luxuriant vegetation maythrive.

In arid and semiarid terrains, carbonates become lesseasy to recognize on photographs. They resist erosion,have thin soil cover, are well-drained and display pro-minent joint systems. It is easy to confuse them withcoarse clastic sediments. On large-scale photQgraphswith good resolution they can be discriminated by theirlack of cross-bedding, their often thickly bedded natureand the peculiarities of their drainage. In most cases,however, they can be distinguished from coarse clasticrocks only by their photographic tone or colour. Manycarbonates are very pure, and being composed of whiteminerals generally appear light on photographs. Thereare anomalies though. Some limestones have a high con-tent of hydrocarbons and their dark coloration may Deconfused with that of basic igneous rocks. Such organic-rich limestones accumulated under reducing conditionsand sometimes contain iron sulphide grains. Weatheringand formation of limonite in these cases can impartstrong reddish and brown stains to outcrops. Whateverthe initial difficulties in arid terrains, the building up ofgeneral signatures for each of the dominant rock typeswill eventually produce sufficient differences for thecarbonate component, if ~ny, to be outlined.

begins, joints are accentuated thereby increas-.-

..infiltration of rainfall into carbonates is

else the distinguishing feature of carbon-humid climates is the comparative absence ofdrainage. In the case of more easily dissolved

this feature is accompanied by abundantdepressions where rainfall most easily

c 4.10a). These sinkholes maymonumental proportions where underground

The solubility of limestone in areas having UITdergonehumid climates for long periods results in their

low-lying aspect. Dolomites under the same.--

4.2.2 Igneous rocks i

Igneous rocks take on a variety of guises because theyhave a wide range of compositions and grain sizes andmay be in various forms as extrusive or intrusive masses.Much effort in photogeology is assigned to establishingthe relationships of intrusive bodies to country rocks,which is dealt with in Section 4.3. The most distinctivefeatures of extrusive igneous rocks are those associatedwith their accumulation at the surface as constructivelandforms (Section 4.5). The emphasis here is on dis-tinguishing igneous rocks from sediments and fromone another after they have been eroded.

compact nature of carbonates, together withheadward erosion and the lack of surface drain-

age, leaves carbonates as long-lived uplands often with

steep-sided major valleys.Carbonates strongly resist glacial erosion. This com-

bined with the closing of subterranean fissures bypermafrost during subsequent periglacial conditionsimparts a peculiar aspect to carbonate terrains in borealareas. The preclusion of infiltration during periglacial

84 Chapter 4

silicate minerals, they vsually are resistant to erosionand form positive topographic features. Unless highlyjointed, they are impermeable compared with sandstonesand carbonates and so develop closely spaced drain-age textures. Highly jointed lava flows have excellentinternal drainage and may have few surface streams.Joint systems are produced by shrinkage during cool-ing and tend to be closely spaced with a polygonalpattern, forming columns perpendicular to the flowbases. At the scales normally seen in photographs thesepatterns are not visible. This structural and composi-tional homogeneity ensures that dendritic patterns ofminor drainage channels occur on their eroded surfaces,in contrast to the angular patterns of jointed resistantsediments.

The principal division of lavas in the field is basedon colour, itself determined by the varying proportionsof coloured ferromagnesian minerals in the rock. Inthe fresh state basic lavas are dark, intermediate com-positions show as intermediate tones and acid lavas arealways light toned. Except under conditions of extremechemical weathering, the constituent silicates of lavasdo not break down in situ to release the strong colour-ing agents of iron oxides and hydroxides. In arid andsemiarid climates lavas therefore take on monotonousshades of grey. There is one important exception to thisgeneralization. Some lavas may be altered soon aftereruption by percolating hydrothermal fluids. Dependingon original composition, the products may be stronglycoloured by secondary minerals, including iron oxidesand hydroxides. Such altered igneous rocks are of con-siderable importance in economic geology, because thehydrothermal fluids may have redistributed valuablemetals to form ore deposits in association with thealtered zones.

Under humid conditions the response of lavas tochemical weathering depends on the proportions ofunstable minerals such as ferromagnesian minerals andfeldspars relative to stable quartz. Acid lavas are clearlymore resistant than basic, and the thickest residualsoils form on basic lavas. The minerals quartz andfeldspar, which dominate acid and intermediate com-positions, contain relatively little in the way of plantnutrient, except in the case of potassium-rich alkalifeldspar. Such lavas, except for alkaline varieties, formpoor soils and are characterized by thin vegetation.Exposed soils are rich in quartz and clay minerals formedby feldspar breakdown and are pale coloured. Highlyalkaline lavas, because of their abundance of potassiumand sometimes phosphorus, often support anomalouslyluxuriant vegetation. More basic lavas break down togood iron-rich soils and support dense vegetation andintensive agriculture. Where soils are exposed they takeon a distinctive brown or red hue reflecting high ironcontents.

Extrusive Igneous Rocks

The greatest problem in identifying the products ofvolcanic activity is that they are frequently interbeddedwith sediments as parallel-sided layers. The problemsare removed if they are still associated with -relics ofvolcanic landforms, such as dissected volcanic cones.Then -the lavas and associated pyroclastic rocks formroughly circular features with radial and annular drain-age patterns, and the only problem lies in discriminatingbetween lavas and pyroclastic deposits.

The classic cone-shaped volcanic edifice charact~rizesthe viscous and explosive:eruption of intermediate to acidmagmas. With the exception of some areas of within-platealkaline activity, magmatism of this kind is restrictedto long, narrow zones near to destructive plate margins.As well as short-travelled viscous lavas, island-arc andAndean-type volcanism produces abundant pyroclasticmaterials of several kinds. The coarsest agglomeratestravel only very short distances from vents. On photo-graphs they show little difference from coarse clasticsediments, but may be recognized by their restrictedoutcrops and rapid changes in thickness. Finer grainedtuffs, irrespective of their composition and grain size,travel long distances tose;ttle as layered beds of fairlyuniform thickness, either subaerially or in bodies ofstanding water. Because of this they are almost imposs-ible to distinguish from clastic sediments. Their grainsize and degree- of later cementation determines theirresponse to erosion, on which they can be subdividedin exactly the same ways as clastic sediments. Identifica-tion as products of volcanism relies almost exclusivelyon context-whether or not they are associated with anydistinctiye volcanic feature.

The exception concerns those pyroclastic rocks thatformed from incandescent clouds of volcanic debriswith sufficient heat to maintain the particles in a nearmolten state after they came to rest. In such ignimbritesthe particles are annealed together to form a glassy rock,which may contain pumice fragments and abundantpores formed by residual gases. They are indeed toughro<;ks and forIn positive topographic features. Theycan be of any thickness up to several hundred metres.Their porosity ensures that they are well-drained withwidely spaced channels. The process of their formation-almost instantaneous deposition of thoroughly mixeddebris-ensures that they are homogeneous and lackinternal bedding. However, the resolution of photographsdetermines whether this feature can be discerned. In adissected terrain ignimbrites therefore can be confusedwith sandstones and limestones.

Lavas, unless they are undissected and show dis-tinctiye surface features (Section 4.5), are difficult todistinguish from sediments with which they may beinterbedded. Being crystalline rocks composed of hard

leir colour and resistance to chemical

iistance they can flow under gravity.mediate lavas are very viscous, flow

..-as a result display rapid

viscous lavas are those of basaltic com-

~ -

rols over magmatism. They can flow for...". --.'c

-.re dissection. Once erosion is under-

---r common name plateau basalts. The

' -~- 's often allows soilsnents

to develop between flows. The

~ver. The pervasiveing in flood basalts makes them par-

--\ cliffs owing to the vertical

.j"Ularntrast to the blocky aspect of

clastic

rocks are of little use

-NS some of theIres of dissected lava flows in arid and

ographic

tone or colour of intrusive rockslle same way as that of lavas, and so too, support of soils and vegetation. The main

lres of intrusions derive from their.internal structures and resulting photo-

-s that they intrude. These relationships are.letail in Section 4.3. Here it is sufficient

post-date those surrounding them to

laced.

They may be narrow/long strips resultingI the planation of vertical parallel-sided sheets or?s.

Diapiric plutons take on a variety of forms, rang-from elliptical bodies to irregular masses withfing stocks, bosses and apophyses, depending on, three-dimensional shape and the depth of ero-

Various kinds of ring intrusions take on annularrops, as may laccoliths and lopoliths. There arelerous

other minor variants. The most difficult totify as intrusions are sills, because their emplace-t parallel to sedimentary bedding allows them to beused

with lava flows or even resistant sediments.ltever their form, to describe them fully from photo-,hs

the boundaries of intrusions with their coun-'ocks must be continuously traceable. More oftennot this is not possible. Estimates of outlines from

ographs have to backed up by detailed mappingle ground, including measuremenm of the attitudes)undaries.gures

4.5 and 4.12 show a variety of the moremon intrusive forms.;

intrusions are emplaced at depth within thet/ where temperatures are elevated by the intrusions\selves and by the geothermal gradient, they cool~ slowly than lavas. This enables them to develop.er crystalline textures than lavas. Cooling also pro-!s

joints through shrinkage. The slower the coolingnore widely spaced. are the joints. The orientation

\e resulting joint systems is often complex, but isly controlled by the outer surfaces of the intrusion

llgh which heat flows because of temperature gradi-Dykes frequently illustrate this last factor very weIl,ng regular joints perpendicular to their margins.patterns therefore give valuable information on

lepth of emplacement, the spet!d of cooling and the'e of the intrusion.~ain

size, as well as composition of intrusive rocks isnportant factor in controlling how they respond to:hering.

Different silicates expand and contract toTent degrees during heating and cooling. This is an>rtant process during physical weathering. Similarly /~ change volume as they are transformed to other!rals during chemical weathering, thereby assistingag of the rock. Both processes increase in effective-as the grain size increases. So a fine grained intrusionbe more resistant to erosion than its coarse-grainedIterpart.Ie

great variety of compositions, forms and graini of intrusions precludes a comprehensive treatment, As an illustration of most of the important pointsxtensive description of grarlitic rocks, among the: common intrusions, is given here.~arlitic

rocks contain high pToportions of the resist-nineral quartz. They are also homogeneous, owinge

mixing processes involved in the convective uprise

le

shape of intrusions when they outcrop at therh they were

86 Chapter 4

(a)

(b)obvious primary features of the flows. The centralparl of thestereomodel shows subdued flow ridges in the pale grey areaand possible flow fronts in the dark, jungle areas. Abundantdrainage channels together with their association with clearvolcanic features suggests that the deeply dissected parts ofthe area are underlain by lavas too. Scales: (a) 1 : 55 000and (b) 1: 40 000. Sources: (a) NCICAriz.12 B + C and(b) cSt Series C 45 + 46.

Fig. 4.11 The high resistance of dissected lavas in arid terrains,such as that of Arizona (a), results in plateaux with flat topsand steep bounding cliffs. Isolated outliers often assume theform of spectacUlar mesas and buttes. Also shown are light-colour~d buttes developed on diatremes where the disruptionof strata can be seen. In humid tropical areas, such as theCameroons (b), the high nutrient <:ontem of weathered lavasresults in luxuriant i«;"get~tion, which masks all but the most

of granitic magmas. As a result granites tend to resisterosion strongly. There is a notable exception... In aridclimates, where weathering is dominated by physicalprocesses, such as heating and cooling, the large differ-ences in the coefficients of thermal expansion of quartzand feldspar rapidly cause crumbling of granite outcrops.Under these conditions granites may be expressed bydepressions whereas related intrusions of different com-position still form positive features (Fig. 4.13).

The homogeneity of granites tends to exert a strongcontrol over drainage networks, which are domin-antly dendritic. However, having cooled slowly atdepth granitic intrusions have frequently been fissuredby large joints which may show as controls of minordrainage. As both form light-coloured outcrops andare strongly jointed, granites and sandstones can beconfused. The main distinguishing features are theirregular nature of granitic joint patterns compared with

spheroidal weathering of basic rocks in humid climates androlling ofqoulders derived in this way often broadens andmas;ks the outcrops of dykes. In this example from Zimbabwethe dyke is easily recognized where it cuts light Precambriangneisses, but is only discernible frOI1;' the dark metavolcanicu:nit when viewed stereoptically. Scales: (a) 1 : 20 000 and(b) 1 : 40 000. Sources: (a) courtesy of Amoco Inc. and(b) CSL Series C 39 + 40.

88 Chapter 4

~

generally show up (Fig. 4.14c)..Basic and ultrabasic intrusions exhibit many of the

characteristics of granites on photographs, particularlyin temperate, humid climates; However, bare outcropsare, of course, much darker.. Another distinguishingfeature of basic intrusions and ophiolitic masses is theircommon banding (Plate 5.9), which results from thegreater tendency of more fluid basaltic magmas toundergo efficient fractional crystallization. The jointingin such rocks is generally more closely spaced than ingranites, and in arid climates this is expressed by ex~tremely jagged topography (Fig.. 4.15). All the mineralsin basic and ultrabasic rocks are prone to decompositionunder warm, humid conditions. This produces deep soilswith luxuriant vegetation, and a tendency to more sub-dued topography than that underlain by granites.

rectan~lar systems in sandstone, and the irre~larelevations of granite surfaces compared with those ofsandstones, which are controlled by bedding planes(Figs 4.7 and 4.14a).

The crystalline nature of granite, together with itstendency to form poorly bound soils and its hardnessmeans that rainfall runs off efficiently, stripping thesurface without forming ~llies. Soils may develop injoint-controlled depressions, so that in temperate humidclimates bare surfaces criss-crossed .with dark lines ofthin vegetation are common. The ov~rall effect is a ruggedtopography of rounded rocky knolls with widely spaceddendritic drainage (Figs 4.5 and 4.14b).

Granites are capable of storing elastic strains producedin them during cooling. They are orily released whenerosion brings the rock close to the surface, when theyare expressed by joint systems that closely parallel pre-existing topography. Further erosion exploits these weak-nesses by a process known as onion-skin weathering orexfoliation. In its most extreme form, under humid trop-ical conditions, this results in the formation of immensedomes known as inselbergs, typified by the Sugar Lqafof Rio de Janeiro in Brazil. Inselberg topography ischaracterized by arcuate drainage pattemsbetween indi-vidual domes. Even in areas affected by intense chemicalweathering and the development of substantial soil~ andvegetation upon granites, their distinctive landforms

4.2.3 Metamorphic rocks

Metamorphism is by definition the growth of newminerals in what were previously sediments and igne-ous rocks under the influence of elevated temperat-ures and pressures. Metamorphic rocks therefore maybe of virtually any composition. The varied responsesto erosion of sediments and igneous rocks by virtueof their composition and grain size, however, is largelymasked by the growth of new interlocking minerals

Photogeology

89

(c)

The

stereopair in (a) is from an arid part of Wyoming, rocks eroded under humid conditions in Nigeria andnorthern Australia, respectively. Image (b) shows a typicalinselberg surrounded by a flat pediment of debris derivedfrom the granite. Scales (a) 1 : 37000, (b) 1 : 30 000 and(c) 1 : 25 000. Sources: (a) USGS GS-CMA-32 88 + 89,(b) Federal Ministry of Works, Nigeria 37-668 + 69and (c) DNDE CAB-151-25125 + 5126.

The

,rugged, bare surface is reminiscent

Drainage

is controlled by the..-

metamorphism. Except for moderately soluble

-rocks are impermeable. They do

contain well-developed joint systems into which

-can infiltrate, having cooled slowly during

from the deep crust over millions of years. As

drainage is both finely spaced and dendritic

they are vegetated, very few distinctive characters showup compared with sedimentary rocks (Fig. 4.16a). In areasof

bare rock, in deserts or glaciated terrains, however,they have very distinctive attributes. Most metamorphicrocks contain compositional banding on a variety ofscales, representing original layering of sedimentaryor i~eous origin. Alternatively it may have resultedfrom the redistribution of the rocks' constituent elementsduring metamorphism, by the processes of pressuresolution during deformation or partial melting at deep

Being composed of strong interlocking crystals, meta---T --rocks are tough and so form positive topo-

features. In humid temperate climates, where

90 Chapter 4

Fig. 4.15 The mountains of north-east Oman are dominatedby exotic masses of oceanic lithosphere-mainly gabbros andperidotites-thrust over sediments of the continental margin.In this stereopair the Oinan ophiolite shows as a dark jagged

area with a complex and fine-textured drainage system. Itsboundary with the block-strewn outcrop of white carbonatesis a folded thrust. Scale: 1 : 40 000. Source: I.G. Gass, OpenUniversity.

Under the intense directed stresses involved in thefolding of metamorphic rocks newly forming platey androd-like minerals tend to orientate themselves, and smallmasses of rock slide by one another by shear deforma-tion. Both processes result in cleavage, usually parallel tothe axial surfaces of the folds. Cleavage forms on scalesfrom millimetres to metres, and only shows clearly asstrong parallel linear textures on photographs with thehighest resolution. It frequently pervades large areas ofmetamorphic terrain, however, and may control minortopographic features and small-scale drainage, therebymasking the compositional banding and folds. This isparticularly common in glaciated terrains, where the mostsubtle weaknesses are etched out by the scouring actionof ice. Because glaciation produces finely spaced groovesit is easy to confuse these with cleavage (Fig. 4.5).

crustal levels. This can impart striking striped texturesto bare metamorphic rock surfaces. The influence of theintense deformation which almost inevitably accom-panies all but contact metamorphism modifies thesepatterns into even more distinctiv~ shapes. Repeatedfolding and shearing results in complex structures,which may be etched out by weathering or glacial action

(Fig.4.16b,c).It is only under conditions of strong chemical-weather-

ing that the compositional variations in metamorphicrocks manifest themselves in different resistances toerosion. As in the case of igneous rocks, the response iscontrolled mainly by the relative proportions of quartzand other stable silicates, compared with the amountsof unstable, usually ferromagnesian silicates. Siliceousmetamorphic rocks are virtually inert and form strongpositive features. In a sequence of metasediments theseare not necessarily metamorphosed sandstones, but canequally be former siltstones or mudstones in which theoriginal clay minerals have become inert aluminosilic-ates. Layers with a high iron and magnesium contenttend to weather easily. These contrasting responses tochemical weathering are important, because they are theonly means whereby the distinctive structural featuresof metamorphic rocks show up in densely vegetated rainforest (Fig. 4.16b).

4.3 Stratigraphic relationships

The keys to successful interpretation of geological historyfrom image interpretation are no more than a few rulesused in stratigraphy and the interpretation of geologicalmaps. As in rock identification, observations should bemarked on images in some way and details of the criteriaused noted down.

The dissected metamorphic rocks of the Cumberland Several of the resistant ridges define complex patterns offolding that typify metamorphic rocks. This diagnostic featureis far more obvious on aerial photographs, particularly ofglaciated metamorphic terrains, such as the Canadian Shield(c). The stereopair shows an excellent example of foldinterference patterns. Scales: (a) 1 : 500000, (b) 1 : 250 000 and(c) 1 : 50000. SoUrces: (a) courtesy of J.P. Ford, Jet PropulsionLaboratory, (b) author and (c) EMRCA1378119 + 20,copyright HM the Queen in Right of Canada (1953).

Dip and strike

as bedding, joints, faults and intrusive contacts,be measured with fair accuracy using a compassclinometer. On photographs these measurements

surfaces having some detectable topo-

resistance to erosion may result in the formation ofcuestas of one kind or another (Section 4.1.1). They pro-vide the best opportunities for estimates of dip.

Along a cuesta maintained by a resistant layer theslop~ that has the same direction as the dip is usually aconsequent or resequent slope (Section 4.1.1). One with adirection that is opposed to dip is an obsequent slope.Where dip is less than about 300 the obsequent slope

92 Chapter 4

hori~ontal

5-\Q

:J~5"OJ.-+

o':J

~0--5"'"0-

"tJID

is usually steeper thanothe resequent slope (Fig. 4.17).This can give a good indication of the direction of dip..However, there are exceptions.. Asymmetric cuestas-candevelop along faults (Section 4.4.1), but this is usuallyobvious. If the layers immediately above the resistantunit have been stripped from the resequentslope, thenthe resulting dip slope can provide a very good estimateof the average dip of the resistant layer. Making a dipestimate from photographs depends on being able toview the slope in a stereomodel.

As dip angle increases there is less chance thathigher layers in a sequence are completely stripped fromresequent slopes. As a result the cuesta becomes moresymmetrical (Fig. 4.17). With steep dips the accumula-tion of talus on the obsequent slope and the increasedopportunity for higher layers to slide under gravity downthe resequent slope means that the asymmetry of a cuestacan be the reverse of that associated with shallow dips.Although angles of dip can be measured quite accuratelyfrom stereomodels (Appendix A), it is a tedious process.

very steep dip

dip -scarpslope -slope

Fig. 4.17 (left) The relationship between the surface form ojcuestas and the dip of underlying layered sequences.

Fig. 4.18 (below) This stereopair of an area in Wyoming showshow the outcrop patterns of dipping strata are controlled byboth angle of dip and topography. The structure displayedisa moI1ocline with vertical strata to the west andshallow northward dips in the east. Scale: 1 : 32 000.Source: USGS GS- VHI-2 54 + 55..

(a) Upstream dfpto

to the apparent dips observed. Figure 4.18from cuestas and dip slopes.

(b) Steep downstream dipof dip are not possible.upon sequences of dipping

taking on shapes familiar to anyone who hasa geological map. In valleys the boundaries

more arcuate, with

(c) Downstream dip less than stream gradientUsing the law of Vs is essential in rapidlythe variation in direction of dip over large

the only means of detecting them and dis-antiforms and synforms. Figure 4.18 givesof this useful technique. Occasionally it is

steep, or in areas of subdued relief,do not result in V-shaped outcrops. Traces of

and their boundaries provide only an indica-This is not reliable in the case of shallow

in subdued terrain, because the slightest slopea marked effect on the trend of outcrops. Where

. Fig. 4.19 These block diagrams show how the interplaybetween topographic slope and the dip of strata controlthe shapes of outcrops in valleys and on spurs. Generally,V -shaped outcrops in valleys point in the direction of dip,but if topographic slope is locally greater than a dip inthe same direction, the relationship is reversed (Fig. 4.9).The apical angles of the Vs become more obtuse as the dipbecomes steeper, until no topographic displacement ofoutcrop occurs with vertical layers.

valleys, strike determination is about the, Provided

are small compared with the

are fairly accurate. This generally is the case-

they intrude. Cuestas are rarely involved. However, thecontact may display V s in valleys and reveal its directionof dip, and perhaps clues to the dip angle. As plutonicigneous contacts are often highly irregular, informationof this kind may be misleading. Provided it is not vertical,the hade of faults may be revealed by V -shapes in valleys.

method of assessing the direction of dip,

absent, is by examining drainage

As Figs 4.1 and 4.17 indicate, where dips are

than about 450 consequent and resequent streams---

are present, mapping them out can reveal a

Dip and strike estimations are not restricted to bound-aries in layered rocks. Faults, the margins of igneousintrusions and unconformities are all surfaces for whichgeometry is important to geologists. Unconformitiesfrequently form cuestas and their dips can be estimatedeasily. Igneous intrusions, such as batholiths, generallyform uniform positive features relative to the rocks that

4.3.2 Superposition'

If the dip direction of any geologically important surfaceis definable on a photograph then it is a simple matterto work out the vertical sequence of rocks cropping outon either side of it. In a layered s~uence the rock crop-ping out in the dip direction is above the surface, thatcropping out in the opposite direction is below. In

94 Chapter4

wall. In terrains.dominiited by thrust tectonics, such asthe Rockies of western North America and the EuropeanCaledonides! and by nappes as in the Alps, recognitionof thrust surfaces and their dip directions is the key toestablishing a vertical sequence of tectonic units (seeSection 4.4)..

4.3.3 Unconformities

Unconformities are among the most important strati-graphic features because they represent periods of timeseparating episodes of deformation, uplift and erosionfrom the re-establishment of deposition. They allow thegeological history of an area to be subdivided into con-veni~nt episodes. By definition an unconformity separatesolder, more complex sequences from stratigraphicallyyounger rocks with simpler structure that lie abovethe unconformity. This usually is revealed by the over-stepping of the unconformity surface from one level inthe underlying sequence to others; Overstep shows onphotographs as an irregular and discontinuous line,marking the unconformity, which truncates traces ofbedding and other structures that lie topographicallybelow it (Fig. 4.20).

most cases the rock above the surface is younger andthe rock below is older. By using all the boundariesbetween layers that are expressed on a photograph it istherefore possible to build up a more or less completestratigraphic column of the most obvious rock units. Thereis one obvious assumption, that there has been no majorrecumbent folding to invert the stratigraphy. Assessingthis possibility is based on observations of structuralcomplexity and on context. Field observations, whichare vital to fill in the detailed stratigraphy, will in mostcases show whether tectonic inversion is present fromway-up indicators. Figures 4.2, 4.9, 4.16(a) and 4.18 givesome examples of simple stratigraphy which can beestablished once dip direction is known.

Identification of the relative age of important or pro-minent units in a layered sequence is not only usefulin establishing a rough stratigraphic framewprk, it alsois a useful aid to recognizing the presence of folds andtheir sense of closure. The relative stratigraphic posi-tion of units juxtaposed across faults allows the senseof movement to be established, and in some cases theamount of throw too.

For faults with hades less than 900 exactly the sameprinciple is used to discover the hanging wall and foot-

the two clearly truncates the ridges, and the horizontal unitlies unconformably upon the steeply dipping strata. The widespacing of drainage in the younger unit suggests that it is amassive; coarse clastic rock. The older unit comprises shalesand limestones. Scale: 1 : 22000. Courtesy of the Committeeon Aerial Photographs, University of illinois.

Fig. 4.20 On this stereopair of aerial photographs of an area inOklahoma can be seen two distinct units. One is characterizedby a dendritic drainage pattern and is sparsely vegetated. .

Faint traces of bedding parallel to contours suggests that it ishorizontal. The other appears banded; with several almoststraight, wooded ridges, which appear to be controlled bysteep dips to the north.north-west. The boundary between

Photogeology

95

involved, the regional smpeof the outcrop of anuncon-formity can be used to estimate quite subtle featuresrelating to original basin extent, and the effects of regionalwarping and faulting younger than the unconformity(Fig. 4.21),

4.3.4 Cross-cutting relationships

As mentioned in Section 4.3.3, igneous rocks generallyare the only units in an area that can provide reliableradiometric ages for calibration of the stratigraphy, Intru-sions give the minimum age of the rocks that they cutand a maximum age for those which sit unconformablyupon them. Similarly, they allow the time relationshipsof structural features that they transect, or which disturbthem, to be assessed. Sedimentary sequences are built upover millions of years, with perhaps very limited evid-ence of age such as fossils or radiometrically dated units.They provide only crude age liinits for tectonic activity.Intrusions however, are emplaced over geologically shortperiods of time and provide a more sensitive means ofworking out tectonic evolution.

Apart from sills and other intrusions with marginsconcordant to older layering, all intrusions have marginsthat cut across older structures, as shown in Figs 4.5, 4.13and Plate 4.2. In the case of massive intrusions it is rare tofind evidence of Ihei. later deformation. Narrow dykesare particularly sensitive to distortion, however, andprovide near-ideal markers for structural investigations.

A mosaic of several of the Landsat MSS band 5 scenes

1 : 2.5 million. Source: author.

Unconformities are usually near-planar boundaries,-the presence of erosion surfaces that have

Where erosion is interrupteddenudation by sudden increase in

local sediment load, a rugged landscape may be.-.

4.4 Structural relationships

Folds, faults and thrusts, and their relationship to.features such as unconformities and intrusions are used

to work out a qualitative tectonic history for an area.They also 'aid quantitative estimates to be made of thedisplacements involved and the disposition of principalstre~ses. Using structures as indicators of the mechanicsof deformation is largely beyond, th,e scope bf this book.Only a few important comments are made, full detailsbeing available in standard textboo~ on structural geo-logy. In this section the emphasis is on dete<;:ting anddescribing major structures..

examination of a stereo model will allow such

Although unconformities permit the ages and intens-.' -folding and faulting to be related to the

rocks, their most revealing relationships are

igneous intrusions. As igneous rocks can be dated

means, their age relative to the uncon-

formity~ as w~ll a.s ~othe rocks and structures divided by

Unconformities can be tilted and folded, in which casetheir attitude can be estimated using the same methods asthose applying to dipping1ayers (Section 4.3:1). Becausethey are major time boundaries separating rocks ofcontrasted str;ucture, unconformities can often be tracedover huge areas. Satellite photographs with their super-synoptic cover are then very useful. Because elevationdifferences are effectively 'ironed ouf at the small scales

4.4.1

Faults, lineaments and arcuate features

Faults are surfaces across which blocks of crust havemoved relative to one another. Identifying a feature ona photograph as a fault therefore means establishing thatdisplacement has indeed taken placE! across i,t. Exceptfor thrusts, most faults dip steeply and crop out as moreor less linear .features that are relatively unaffected bytopography. There are, however, many ways in whichobvious linear features may be formed. They may be

96 Chapter 4

.Fig. 4.22 These block diagrams showthe relationships between the threemain classes of fault and the requiredarrangements of the maximum andminimum principal stress axes.Theoretically, for each arrangementof principal stresses, faults shoulddevelop with two orientations. Theacute angle between them is bisectedby the maximum principal. stressdirection, and the two sets form aconjugate pair.

(a) Normal (b) Reverse and thrust (c) Strike-slip

Faults manifest the brittle response of the crust tostress. They are surfaces along which the crust has failedand which allow the stresses involved to be dissipated bycrustal extension or shortening. The attitude adopted byfaults is governed approximately by simple mechanicallaws relating the orientation of the maximum, minimumand intermediate stress axes acting in the crust, thestrength of rocks and the resulting surfaces of potentialfailure. Figure 4.22 illustrates these principles and showsthe main classes of faults that develop. At the instant offailure one of two possible fault surfaces begins to breakfirst and strain is concentrated and propagated across it.Thus, although theoretically faults should develop equ-ally, parallel to two conjugate directions, in most casesone direction dominates. The conjugate set of faults tothis is either suppressed or plays a relatively minor rolein crustal deformation. "

Heterogeneities in the crust and in the stresses involvedduring a period of crustaideformation mean that all thetypes of fault shown in Fig. 4.22 may occur in an area. Ifthey do, then estimates of their attitudes and the direc-tions of movement along them allow the overall tectonicsof the area, and their relationship to major crustal forcesto be assessed (Fig. 4.23). An important parameter toassess from such composite fault systems is the overall

faults, joints, dykes, steep to vertical strata and culturalfeatures, such as roads and boundaries between areas ofdifferent agricultural use. Certain defects in some imagesproduce artificial lines too, known as artefacts,(Chapter 5and Appendix B).. It also may be possible to link variouselements of a scene by lines, such as isolated boundariesbetween surfaces of different tone, meanders in a num-ber of streams, notches in ridges, truncated spurs andfeatures relating to other geological structures. These maybe purely fortuitous connections,but could have somegenetic relationship to subtle or deeply buried faults.

With the increasing availability of small-scale, super-synoptic satellite images of the Earth there has been atendency among some geologists to cover them with real,inferred and entirely imaginary lines. In some cases theseextend across entire continents. Repeated analysis by thesame person or by" different interpreters often resultsin different arrangements of many of the lines, The nearobsessive nature of this simplistic approach has pro-voked a measure of scepticism and cynicism towards thewhole of remote sensing by many scientists. There canbe no doubt that linear features of geological importancedo occur,and that they are visible only on small-scalephotographs because they form broad, subtle and verylong connections. However, all potential lines need tobe viewed with caution and subdivided into categorieswith different degrees of confidence (Section 2.6).

A note on nomenclature is necessary here, to avoidadding to the already burgeoning confusion surround-ing lines on images.'" Linear is an adjective describingthe line-like character of an object or array of objects. It is'often misused asa noun designating clear, short lines onan image, in contrast to a lineamentwllich1s correctlyapplied to long, often subtle linear arrangements of vari-ous topographic, tonal, geological and even'geophysicaland geochemical features. Linear feature is used hereto describe relatively short (less than 5% of the extent ofthe image being used) lines and" regular arrangementsof surface 'phenomena. It is used together with the termlineament in an entirely nongenetic sense. When eithercan be recognized or confidently inferred as a geologicalentity, then it is given its correct, geological name.

~"---Y/

Fig. 4.23 Because the Earth's crust is constrained in twodimensions, one major episode of deformation, such as thatcharacterizing the TibetanPI~teau today, can produce all threeclasses of faults. This cartoon shows how they might formrelative to one another.

Photogeology

97

movements, the relative displacements--, --.

The orientation and sense of fault movement

turn gives a context within which other, more

majority of faults that are easily visible on~1 hades, as explained later.

Asa result their surface exp~~ssion maywith the depth of erosion. No fault continues

.," _1.' ---'- It may die out, the strain1 It may split or develop

result from the reorientation of principal stress,' "T_' large

, '"- have a similar but less widely distributed

and develop minor splays of their own. The result" "

photointerpretation of faults is often more reli-their detection in the field. Because they are

features, from which

to faults are difficult tothe geologist is close to them. The synoptic

photographs enables widely separated pieces--to be linked as sharp and semicontinuous

,- --'-, The vertical exaggeration

floor (Fig. 4.24c). The ~of may be metamorphosed base-ment and the floor composed of sediments with simplestructure. Distinctive lithologies in the roof rocks, suchas red beds, may also indicate an inverted sequence andthe suspicion of thrusting. Iff the main, however, thrustsare notoriously elusive in photogeology; because theyhave low angles of dip their outcrop often parallels thatof the layers that they have displaced. Often they canbe inferred only from the context of other observations.One of these is the identification of local truncation ofcompositional layering in the hanging wall or footwall(Fig. 4.25). These are known as strike cut-offs, and aremore easily recognized on images covering large areasthan on the ground, because the angle between thrustand local strike is often very small.

The steeper the angle of a fault the less it conformsto topography. Steep faults are conspicuous on photo-graphs because of their near independence of valleys,ridges and uplands. The steeper they are the more theyresemble regular lines. Because of intense fracturing alongfaults they often form local weaknesses, become erodedout and form linear depressions (Figs 4.24, 4.30 andPlate 4.3). They are often followed by streams. The almostirresistible tendency of streams to meander means thatany straight segments immediately raise the suspicion offault control. There are relatively rare cases of faults thathave been cemented by percolating fluids or followed byigneous dykes, in which case anomalous straight ridgescan form.

The high permeability offault zones means that theycan be saturated by water. They provide easy accessfor root systems, and when forming depressions pro-tect plants from wind and desiccation. Lines of vegeta-tion encouraged by these conditions are also potentialindicators of faults, but joints provide exactly thesame conditions and other evidence is n~ded for con-firmation. Where faults have thrown permeable rocksagainst impermeable, spring lines may form. In thecase of limestone thrown against impermeable rocks,local drainage may disappear into lines of sinkholesmarking the fault. Where a fault juxtaposes rocks ofgrossly different resistance to erosion, the fault can beexpressed as a line of cliffs, waterfalls or abrupt changesin slope.

Whatever linear features can be discerned on a photo-graph, they can be identified as faults only if other geo-logically significant features in the scene are displaced ortruncated. Even truncation is a dubious criterion, becausegeological structures can terminate abruptly for a hostof reasons. Normal and reverse faults displace the crustvertically. In terrains dominated by horizontal or gentlydipping strata the displacement shows up only by changesin elevation of easily recognized beds or boundariesacross the fault. Where dips are greater than about 5°,vertical displacements also produce apparent lateral

most subdued topographic features and reveals clearlyany vertical displacements associated with faulting.Despite the scepticism about 'linear geoart' , image inter-pretation is the most powerful means of detecting crustalfractures and analysing their tectonic and economic

significance.As with all geological boundaries, the dip on fault

surfaces governs how they are affected by landforms.The lower the angle the more a surface tends to followtopographic contours and adopt complex outcrop pat-terns. Thrusts are controlled strongly by topography,and their expression is often identical to that of beddingsurfaces. On occasion the floor of a thrust, which maybe coated by tough mylonite, stands out as a prominentdip slope (Fig. 4.24c). Sometimes the tones, patternsand textures of rockS forming the roof of a thrust clearlyindicate that they are older than the rockS forming the

98 Chapter 4

(a)

(b)

Fig. 4.24 (a) The sharp boundary separating dark from lightterrains is a very large normal fault in the Sultanate of Oman.The light unit-consists of Tertiary limestones, which weredeposited upon the dark, late Cretaceous ophiolite complex.The ophiolite here is composed of layered gabbros in whichcompositional banding is clearly visible. It has been thrustover the pale-grey Precambrian granites seen at top right.

The thrust itself has been diSplaced by a smail strike-slip faulton the right. (b) The sharp linear feature running diagonallyacross this aerial photograph of part of the Canadian Shield is astrike-slip fault. The diSplacement across it can be judged fromthe relative diSposi~onof the vertical contact between granite(light grey) and metavolcanic rocks (darker grey) marked bythe dashed line. It is about 5 km.

Photogeology 99

(c) This stereopair shows a spectacular thrust faultin the Caledonides of north-west Scotland, which droveArchaean gneisses over Cambro-Ordovician sediments.The thrust plane is marked by the pronounced bench onthe peninsula, which is coated with mylonite. The typical

knobbly topography of the heavily glaciated gneisses is quiteobvious in the upper part of the pair. Scales: (a) 1 : 20000,(b) 1 : 50 000 and (c) 1 : 20000. Sources: (a) courtesy of Amocomc., (b) EMRC A13621177, copyright HM the Queen in Rightof Canada, and (c) author.

displacements in outcrops. Features marking resistantunits are truncated by the fault and reappear across itat another position, depending on the direction of dipand the sense of movement of the fault. Knowing thedirections of dip of the unit and of the fault surface--perhaps from V -shaped outcrops-allows the sens~ offaulting to be worked out. Strike-slip faults produce truelateral displacements. However, the resulting truncationand displacement of dipping resistant units can haveexactly the same appearance as those associated withnormal and reverse faults. The only reliable criterion isthe displacement of vertical surfaces, such as other faults,vertical strata, igneous dykes and the axial surfaces ofupright folds.

As well as revealing the present tectonics of anarea, active faults present major environmental hazards.Locating them and identifying their movement sensesis doubly important to the geologist. World:.wide moni-toringof seismic events has provided such a wealth ofdata on the location of earthquake epicentres that all

Fig. 4.25 (left) On this Landsat TM band 5 image of an area inEritrea the presence of a fault almost parallel to compositionalbanding, probably a thrust, is indicated by a marked linearfeature to either side of which the strike of smaller linearfeatures corresponding to layering is truncated. These areramps in the hanging wall and footwall of the thrust.Scale 1: 100000. Source: author.

100 Chapter4

displacements along an active fault. The body of water at thecentre is a typical sag pond caused by fault-induced tilting.Scale: 1 : 28 000. Source: NCIC Utah 2 B + C.

Fig.. 4.26 Running across the centre of this stereopair of anarea in Utah are several connected scarps. They are developedwithin Pleistocene moraines, and resulted from normal

The most obvious sign of a fault's activity is its dis-turbance of cultural features, such as roads and straightfield boundaries. Equally as clear are displacements ofRecent superficial deposits, often with the presence ofclear breaks and low bluffs (Fig. 4.26). Small-scale tiltingassociated with fault movements produces depressions,often filled with water to form sag ponds (Fig. 4.26).Where streams cross active strike-slip faults they areperiodically displaced. This produces sudden bends intheir courses along the fault line. Such offset streamsare often present in aR drainage crossing the fault, anddefine both the line and the sense of most recent move-ment (Fig. 4.27). Physical disruption by a major earth-quake of topographic features formed during periodsof quiescence, results in the formation of truncated orfaceted spurs (Fig. 4.27). Such displaced ridges canbe transported so far along a strike-slip fault that theyblock the upper ends of valleys, the formative streamsof which have been shifted elsewhere. This producesshutter ridges and headless valleys (Fig. 4.27).

The supersynoptic view provided by satellite imagesis ideal for re-evaluating fault patterns and other reflec-tions of brittle tectonics over large areas. Small, region-ally insignificant features are suppressed and very large,subtle and previously unsuspected ones may show up.The smaller the scale the more subtle, more completelyconnected and larger are the features that can be discerned.

areas of major seismic activity are now known. Seekingactive faults in them therefore is aided by known con-text. The world-wide increase in potentially hazardousand seismically sensitive nuclear installations, however,means that-even the smallest earthquakes pose a seriousthreat.. Currently many areas hitherto considered stableate being re-evaluated for seismic risk. This concentrateson delineating old fault systems that may be reactivatedbYJ;ninor tectonic processes, such as isostatic readjust-ment after glacial loading..

Large acqve faults are recognized in the same waysas old ones etched out by erosion. Their activity hasextended for thousands, if not millions of years, so thaterosion has had time to pick them out. In many casesthey are in areas of considerable relief resulting fromthe rapid uplift and erosion that accompanies intensetectonic activity. They are easily seen in such terrains.An active tectonic zone may also contain faults thathave ceased to move. So it is important to distinguishthose that pose a: threat from those which are probablysafe. An active fault may remain dormant for tens andhundreds of years between earthquakes, with stressesslowly being built up before explosive release. Theremay well be no recprds of seismic events related to them.Fortunately, faults that have been active in the last fewthousand years still preserve unique features which canbe seen on photographs.

ridge and faceted spurs. The sense of active movement appearsto be sinistral strike-slip. Scale: 1 : 12 000. Source: USGS GS-GF.1 39 + 40 (colour original).

the interconnectedness of tectonic featuresBearing in mind the cautionary comments

--

very quickly and related to notions of the.., The best results are from

magmas. Both may encourage mineraliza-c_- --.,' --,-- "-

Fig. 4.28 This NOAA-6 A VHRR band 2 image of WestAsia shows the relationships between the tectonicallylinked Oman ophiolite (SW), the SSE-NNW Lut Block-amicrocontinent that indented Iran-the E-WfoldedMakran Range and the N-S folded Kirthar and SuleimanRanges, which are bounded to the west by a large sinistralstrike~slip system. All formed at roughly the same timein the late Cretaceous to early Eocene when the Tethysocean closed and Indiaimpacted Asia. Scale: 1 : 15 millien.Source: author.

concentrated on the mapping of linearand ~eaments, and their relation to known

of deep-seated faults. Further incentive stemmed from thesynoptic analysis of regional patterns of mineralizationthat began to reveal linear arrangements of active minesand minera,l prospects. One suggestion was that they

Chapter 4

~~1.29

This sketch map of Plate 4.3 distinguishes betweenare thought to be traces of compositional layering andr faults. The N-S faults bound a major pull-apart basin,lble cored by Pleistocene volcanics. The NE-SW and"I-ESE faults are conjugate strike-slip faults, the dominant~ing

the former. This pattern has been interpreted as:ting northward crustal shortening and E-W extension.

:rust in fact deforms as a framework of lozenge-shaped;sbounded by strike-Slipfaults'and which contain the~ns.

Scale: 1: 1.8 million.

,.30 In Eritrea the western lowlands contain large areas of.ltially fertile, clay-rich soils yet they record much lowerill than the nearby upland areas to the east. Runofffromplands

is heavily charged with sediment and dams to~st it for irrigation would soon become silted. Equally,hemeral streams are so powerful during storms thatwould

destroy channelled irrigation schemes. On this~ Format Camera image several marked linear features:ting

large crustal fractures trend from the high rainfall\ds to plains with clay-rich soils but low rainfall.eivably, they could be transferring large volumes ofquality ground water to the area of agricultural potential.~ could be tapped quite easily. Scale: 1 : 700 000.:e:

author.

tverdeepcrustalweaknesses,

Another was that theyrded the passage of the crust over mantle hot-spots,>retically,

both mechanism~ should be reflected it}Ie tectonic and magmatic features visible at the sur-

given a wide enough perspective,l1other regular anq sharp type of feature that haslcted

co~si.derable attention from explorationists isircle, The eye is very good at seeing circles. Indeedrather too good, and can pick out circles in nearlylom patterns. Nevertheless, there arf!O many nearlyliar and elliptical features produced by crustal pro-~s, and some of them are indeed related genetic-to

mineralization and hydrocarbon accumulations.most obvious are salt domes produced by the,iric

uprise of low-density halite and the productionl1nular fault systems and overlying circular domes4.31~). They are well-known traps for oil at}d gas.ulargranite

intrusions form in a directly analogous, and when they are peralkaline, have associated tin,liu~

and rare-earth mineralization (Fig. 4.13). Anortant proportion of di~onds are found in kimber-pipes, the products of cryptoexplosiop structuresciated

with ultramafic magmas rising from the deep

tIe.

Such pipes too have a nearly perfect circular,-section.. However, kimberlite is very soft and pipe

tation often shatters the country rock so that circularlTes are suppressed. Within large igneous plutons,iorite

to granodiorite composition, the patterns ofrective flow of magma and hydrothermal fluids can

it in circular cross-sections of flow cells. These may~ to mineralization of the porphyry type. Another

of real circular feature relates to the erosion ofcal volcanoes and their frequently associated annular:s

produced in calderas. Many porphyry-type oresound at different levels in such.structures, and may(posed by erosion (Fig.4.31b).

view of the Moon reveals the most dramatic of cir-r featureS--,-the cr~ters associated with the impactrge meteorites. There is no way in which the Earthd

have escaped the bombardment responsible for the(-marked lunar surf~ce. However, the vast majority

(d)

The circular feature at the centre of (d), a Landsat MSS band5 image, is located in the granulite-facies gneisses of theAnaimalai Hills in South India. Although it is boundedby a clearly defined boundary, on the ground there is noevidence for what may have produced the structure.Scales: (a) 1 : 90 000, (b) 1 : 1 million, (c) 1 : 14 million and(d) 1 : 1 million. Sources: (a) CSL SeriesB 11, (b) courtesyofP.W. Francis Open University, (c) National Oceani<: andAtmospheric Administration, Washington, and (d) author.

largest to have had a diameter of 6000 kIn-have beenobliterated. Such impact structures do exist where littlehas happened to the crust for a billion :years or more.They are among the easiest features to spot from orbit

104 Chapter4

(Fig. 4.31c). On an: entirely different plane are circularfeatures that are defined by mineral occurrences on acontinent-wide scale, and which show up, if at all, onlyon satellite images encompassing millions of squarekilometi'es. Several authors have interpreted such fea-tures as relics of the early bombardment of the Earth,and continually seek others as a regional guide to newmineral districts. This is an entirely illusory quest, becausethe thermal effect of even the largest impact would lastonly about 50 million years. Moreover, the mineraliza-ti<;>n associated with such features affects rocks with ageranges that span up to 2.5 billion years. Whatever thesegiant features are-and if they are not imaginary-theyprobably reflect long-lived anomalies in mantle heatproduction, and their episodic effect on crustal uplift andthermal evolution. Finally, there are circular featuresthat undoubtedly exist, but to which no mecltanism canbe assigned (Fig. 4.31d).

4.4.2 Shear zones

Near to the surface the crust deforms by brittle processesand by buckling. With increasing depth the influenceof higher temperatures and hot fluids decr~ases theviscosity of rocks, so that they are more prone to deformplastically. Instead of crustal displacement being takenup across sharp fault surfaces, the shear strain is dis-tributed through a large volume of rock, but the neteffect is exactly the same. Such deep-seated, ductile faultmovements produce shear zones, the forms of which arevery different from those of faults. Like faults, shear zonesresult iI1 thrust, normal and strike-slip displacements ofthe crust. Their typical form is illustrated in Fig. 4.32.

Deformation .in a shear zone involves changes inangular relationships, so that linear features outside the

shear system are &:een ~ swing into it. This can be seenin Fig,,' 4:32 to be accompanied by a narrowing in thespacing between thelineai features as the shear zone isentered. The reason .for this is simply the tectonic envir-onment in which the shearing takes place. The crustis being compressed in the direction of the maximumprincipal stress and extended at right angles to it (seeFig. 4.22). Some of this crustal deformation is taken up bythe displacements along the shear zones, but becauserocks are responding plastically, local compressive andextensional strains develop along the shear zone itself.The rock is flattened in the plane of the shearing.

The more strain is taken up by a shear zone, the fur-ther displaced are once adjacent points either side ofit, and the wider the zone of flattened and stretchedrocks becomes. .If marker units are present then theirdisplacement is immediately apparent. In most casesthis is not possible, but theoretical methods allow theshear strain to be estimated from the width of the beltO.f flattening. Even if this is not measurable, the senseof strike-slip movement is quite obvious from the wayin which linear features are rotated into the shear zone(Figs 4.32 and 4.33).

As well as deforming linear features, shear zones alsorotate and flatten folds, often into unique and bizarreforms (Fig. 4.33b). Any relatively narrow belt of finelyspaced linear features with flattened folds separatingterrains with simpler structure can be suspected of beinga shear zone with a strike-slip component. Thrust-likeshear zones are not so obviQus, but can be detected fromtheir association with complex fold forms (Section 4.4.3).

shearzone

//

I'

4.4.3 Folds

Folding of the crust presents the most stimulatingand challenging images to the geologist. As well as pro-viding an element of variety for the interpreter, andanother means of evaluating strain and stress patterns,folds are of great economic importance. Anticlinal struc-tures form traps for oil and gas, whereas synclines oftenhost artesian water supplies. Locating and analysingfolds is an important element in predicting subsurfacegeology, and in reconstructing the geological historyof an area.

Folds may be classified using a great range of geo-metric properties, for which there are excellent guidesin standard texts on structural geology. In photointer-pretation most of these criteria are not visible, so herea gros$ly simplified system is used. It is based on threecriteria-the interlimb angle of folds (Fig. 4.34a), theplunge of fold axes (Fig. 4.34b) and the dip of fold axialsurfaces (Fig. 4.34c). As the way-up of folded layers isnot discernible from photographs, an upward-closingfold is properly described as an antiform, the oppositebeing a synform.

~/ !

!

Fig. 4.32 This sketch shows the main elements of a shear zone.Layers are displaced by shearing in the same way as in a fault.However, instead 'of a sharp break, the displacement i!!accomplished birotation, "extension and flattening of thelayers, so that the shear zone itself comprises highly deformed,banded and often lineated rocks.

South India. In the Red Sea Hills of Sudan (b) an indistinct,NW-SE shear belt has sinistrally displaced an earlier WSW-ENE zone of shearing by about 30 kIn. Scale: 1 : 1 million.Source: author.

!\/1\\

300-100

tight

(a) Interlimb angle::::::::::::===~:~:::

..,-::~~;:::~~~~~,1800-1200

gentle1200-700

open

(b) Dip of axial surface1

/~~~~~\upright

\

/~:~:~;~"asymmetrical

-~

recumbent

~

overturned

(c) Plunge of hinge-- ~c~~-~.", ""

""q

Fig. 4.34 The nomenclature of foldsis based partly on three mainfeatures-the angle between thelimbs of a fold, the plunge of its axisand the dip of its axial surface.Rarely more than these features canbe seen on remotely sensed images.

A\

~

sideways-closinghorizontal plunging

axes in such cases depends upon systematic observationof dip direction, using valley Vs (Fig. 4,35a,b). Like faults,however, folds do not propagate to infinity. At somepoint they die out, the strain being taken up by another

Where a fold has a shallow plunge, closures are notcommon. The limbs crop out as dipping series of layers,taking forms appropriate to the angle of dip and the topo-graphy (Section 4.3.1). Detecting folding and locating fold

Q100-00

isoclinal70°-30"

close

Chapter 4

It"1l'

-L35

Folds are the most spectacular features on remotely~d images, particularly when they are well exposed.opair (a) gives an example of inverted topography inria, where an asymmetrical, plunging synform folding~zoic

sediments controls high ground. The nature.of

lId could be worked out from the varying directionsngles of valley Vs by using one image. This would beble

on st~reopair (b) too, which shows two plunging,metrical antiforms separated by a synform in Morocco.

Even the most regular folds have a plunge on theirsomew.hereiand this may show as a closure of out-s

of some distinctive unit. The more steeply thataxes plunge, the more obvious and more common:losures

are (Fig, 4.35a,b). The forms taken by foldsphotograph are strongly controlled by the asym-

y of the dips on the limbs. The limbs of an upright

appear almost like mirror images of each other4.35b). In an extremely asymmetric fold the steeperis less controlled by topography than the shallowand adopts a more nearly linear form (Fig. 4.18).~turned

folds present problems. The tighter and the~ nearly recumbent they are, the more complex arelutcrops

of their limbs (Fig. 4.35c). Where stereoptic

Ster~pai:

~cl contains several overturned folds in the affected by thrusts, which disrupt some of the folds, and a fewnormal faults. Scales: (a and b) 1 : 40 000 and (c) 1 : 63 360.Sources: (a) IGN ALG-1952-'231-233123 + 124, (b) IGNMaroc-009

424 + 425 and (c) NCIC Tex.1 A +B.

is

possible, the correct description of folds iseasy. For satellite images the process can be

problems, the most common of which isSynforms (Plate 4.4).

evidence of valley Vs is

4.5

Superficial deposits andconstmctionallandforms

More

even than destructional landforms, those res¥lt-ing from the constructive action of volcal)ism an~ fromsediment transport and deposition are enormously yaried.So

many facto~s are involved that the number of per-mutations and combinations is near infinite. Space does

not permit a comprehe~sivetreatment;merely an accountof some common and easily recognized features. Mosttexts

on volcanic, desert, glacial fluviatile and coastalgeomorphology provide sufficient diversity of descrip-tions

and photography for this brief account to be easilysupplemented.

the most weird patterns that can be seenimages are those produced by the

folding of. layered rocks. When a fold withis refolded by another with a different

a three-dimensional interference structure isThe distinctive patterns are produced whe~

are planed by erosion. They can adopt--but three types form theind

most commonly seen end-members. Where., --.by another set with a differ-

trend, domes and basins result (Fig. 4;16c).folds refolded by upright folds produce

.,

4.5.1

Volcanic landformsThe

popular concept of an active volcanic landscape isthat it is typified by symmetrical volcanic cones, eachpossessing a summit crater. This is only sometimes thecase

for areas dominated by viscous magmas, of inter-mediate to acid composition. These magmas cannot

flow far from the vent from which they issue, and beingunable to release their gas content easily, are prone to~xplosive

activity. The result is that lavas"and ejecta from3. vent build up close to it as a layered conical structure

The effect of strike-slip shearpre-existing folds is often to transform them

..(Fig. 4.33). The presence of

108 Chapter4

terrped a stratovolcano, which may be as much as sev-eral thousand metres high. The sides of a stratovolcanotypicaUy slope at angl~saround 30°. The vent itself iskept relatively clear 6f debris by repeated explosions, so

forming a crater. The strengthening of the volcanic edificeby lava flows ensures that the crater walls are steep andcliff-like. Figure 4.36 shows an ~xample of the more typ-ical,complex stratovolcano.

Fig. 4.37 Cinder cones, such as this typical example fromArizona,are closer to the popular .concept of a volcano thanmost stratovolcanoes. The cone is symmetrical, despite theapparent distortion as a result of radial-reliefdisplacement.The funnel shape of the crater, which results from the

instability of unconsolidated ash, is clear. In this case, explosiveactivity succeeded earlier eruption of moderately fluid lava,the typical morphology of which is obviously overstepped bythe cone. Scale: 1 : 20000. Source: ASCS BT 1952 + 1953.

Photogeology 109

Basaltic magmas a~ much more fluid than those ofintermediate to acid composition. They tend to releasetheir gas content with a minimum of explosive activity,and travel long distances down shallow slopes. Theyrarely build steep cones, although there may be a minorbasaltic component in stratovolcanoes. Two major kindsof basaltic landform are produced. The magmas mayissue from fissures induced by crustal extension. Suchfissure eruptions (Fig. 4.38a) typify basaltic volcanism atconstructive plate margins. Where basaltic magmatismis controlled centrally, as in a mid-plate setting, lavasflow in all directions to build up vast domes with very

are characterized by having a smooth,

kmin diameter.

~

Fig. 4.38 On the flanks of Mauna Loa shield volcano onHawaii, eruptions of fluid, gas-charged basaltic lavasemanated from long fissures (a). Small spatter cones arealigned along some of the fissures, as are cinder cones. Theblack dots on a pale background are collapse pits above a lavatube. Silicic lavas tend to be far more viscous than those of

basaltic composition. Stereopair(b) shows an excellentexample from the Andes of Chile.. It has not flowed far fromthe dissected stratovolcano that was its source. Its distinctlobate form shows ridges perpendicular to flow direction.Scales: (a) 1 : 41 000 and (b) 1 : 60 000. Sources: (a) NCICHawaii 7 and (b) Instituto Geografico Militar, Chile.

110 Chapter4

Fig. 4.39 Lava flows of different ages are distinguished on thisLandsat MSS band 7 image of part of Idaho by their differentreflectances. The youngest flows are black and successively olderflows are increasingly reflective as a, result of being maskedby blown sediment and colonization by vegetation. The flowsemerged from a NW-SE fissure, which can be seen faintlynear the centre of the largest area of lavas. Scale: 1 : 1 million.Source: courtesy of Dave Rothery, Open University.

low slope angle~typically 5-10°. These shield volcanoescan be hundreds of kilometres across, as in the case ofthose in Hawaii.

Basaltic volcanism sometimes produces fire fountainsas lava is forced violently to the surface by the pressureof rapidly rising magma. The lava forms droplets andglobules, which are still partly molten when they land.This ensures that they stick together to form spattercones (Fig. 4.38a). These tend to be associated with fissureeruptions, but may also form on a minor scale when lavaerupts from the flanks of shield volcanoes or squirt_sthrough the crust of a major flow as a result of thepressure head involved. Spatter cones have very steepsides owing to rapid solidification. They are usuallyless than 20m high.

Truly explosive activity with cinder and ash produc-tion is associated with basaltic activity only wiwre watercan enter the magma chamber. Basaltic cinder and ashcones are found in ~ssociation with coastal volcanoes,and where volcanQes form beneath ice sheets, as inIceland. In the last case truly bizarre geomorphologicalevents, called jOkulhlaups may occur,. which involveglacial collapse, catastrophic floods and mudflows aswell as the formation of volcanic edifices.

Undissected lava flows have very distinctive appear-~ncesresultingfrom the process of flow itself. The mostooviQuS is the formation of lobate margins, where thec~ol end of a flow is highly viscous and spreads slowlyinto topographic depressions. The solid crust bucklesand breakS, into blocks, which can produce levee-likeridges at...t~esiqes and ends of flows (Fig. 4.37). In vis-Cous lavasftheformation of ridges perpendicular to flowdirecti~n Eixtends right across flows (Fig. 4.38b). Such

c

lavas flow only down steep slopes, and usually extendonly short distances. Basaltic flows travel much further,and tend to develop flow lines parallel to flow direc-tion (Fig. 4.38a). Sometimes late-stage lava movementis beneath an already solidified crust. The resulting lavatubes can collapse to form distinctive sinuous depres-sions on the flow surfa~e(Fig. 4.38a).

In areas of repeated lavaerupttons distinguishingbetween flows of different ages is sometimes a relativelysimple matter. The stratigraphic rUles of-superpositionand overlap generally canbe applied: Moreover, the oldera flow the greater the tendency for its surface roughnessto be degraded by soil formation, and for it to supportvegetation (Fig. 4.39).

The issue of fine-grained pyroclastic material fromvolcanoes prone to explosive activity produces a blan-keting effect over older landforms. The resulting tuffssmooth out the topography (Fig. 4.40) and kill the localvegetation. They are, however, prone to rapid erosion,and are soon dissected. The most spectacular resultsof explosive activity relate to the flow of high-density,incandescent mixtures of ash and gases. These nuees

ardentes are emitted from volcanoes with such viscousand gas-charged magmas that they cannot easily beextruded. Gas pressure builds up in the magma chamberuntil the entire contents are belched out in a single event.The dense mixture of hotash and gas has a low viscosity,so that its immense gravitational energy causes the mix-ture to travel at velocities in excess of 100 km per hour.It therefore can flow over and around large objects inits path to swamp vast areas and will eventually solidifyas an ignimbrite (Section 4.2.2). The high energy of nueesardentes results in deep fluting of the ignimbrite sur-face, parallel to the direction of flow (Fig. 4.40). Recentignimbrites are thus easy to identify, even from satellitealtitudes.

Stratovolcanoes, be~ng formed by explosive mag-matism, are prone to catastrophe. The expulsion of thecontents of a magma chamber during a nuee ardenteevent leaves a cavity at depth, into which the super-structure may collapse. This is accomplished by annularsystems of faults defining a caldera (Fig. 4.41a). Theresulting circular depression is occasionally filled with

.

'1\

the southern flank of the volcano. The complex tonal andtextural patterns in the debris can be used to identify theoriginal location of the material in the volcanic superstructureand

to reconstruct the mechanics of the avalanche. Scales:(a).1 : 60000 and (b) 1 : 400 000. Sources: (a) InstitutoGeografico Militar, Argentina and (b) courtesy of DaveRothery, Qpen University.

112 Chapter 4

--...,,'

Fig. 4.42 Alluvial fans typically develop at the exits ofintermittent streams draining arid mountainous regions.In this part of the Andes of western Argentina, steeplydippingPaIaeozoic strata have received a thin veneer ofvolcanic de,bris from a massive ignimbrite event resulting

-

in a peculiar smoothing of the older topography. Erosion ofthe veneer has not long been underway, and is marked bylocalized, complex systems of sharp rills. Scale: 1 : 60000.Source:lnstituto Geografico Militar, Argentina.

ash and debris, which has many of the morphologicalfeatures of a lava flow. Befug soft, however, they are sooneroded away.

4.5.2 Fluviatile and lacustrine landforms

Deposition by flowing water to form constructionallandforms results when flow velocity falls below thatneeded to retain sediment in suspension or to move italong the stream bed by bouncing or rolling.

High-energy streams issuing from uplands carryabundant coarse particles. Once the slope changes as thestreams emerge from constricted valleys, this coarse loadis quickly dumped to form low cones with their apicesat the mouths of the mountain valleys. The streams areable to flow in any radial direction on these cones, and sosplit into numerous divergent channels. This gives riseto alluvial fans, which typify the flanks of uplands insemiarid terrains (Fig. 4.42). Stream flow is sporadic andreaches a maximum during storms or rapid snowmeltat high altitude. The constricted flow spreads at thevalley mouth to form sheet floods which deposit andredistribute debris uniformly across the fans. The highlyporous nature of the fans ensures that a high proportionof flow infiltrates to produce local groundwater sup-plies. Only where vegetation is unable to bind the debrisand stabilize stream courses do alluvial fans in humidterrains develop on steep slopes as screes.

a lake, CQntinued magmatism may exploit the caldera toform minor cones along the annular faults, and centraldomes of resurgent, viscous magma (Fig. 4.40). Insteadof resulting in nuee ard~nte activity, build up of pres-sure withina volcano may cause the flanks to bulgealarmingly. If the bulge develops unstable slopes, thevolcano flank may collapse to form a massive debrisflow (Fig.. 4.41b). The release 0.£ pressure on the flank canthen allow the accumulated gas-rich magma to explodesideways, breaching the entire superstructure. This wasthe process associated with the May 1980 eruption ofMount St Helens in north-western USA.

Many stratovolcanoes reach high elevations, and evenin equatorial latitudes may be capped with snow and ice.This is a hazardous association. A combination of ice-bound ash and heat associated with episodes of activitycan produce sudden melting and the formation of mudavalanches. These can assume nearly the same monu-mental proportions as ignimbrite and volcanic collapseevents. The mud flows in much the same fashion as lava,.usually faster and further. The energy of these volcanicmudflows is high enough to sweep all before them,leaving sinuous scars through vegetation, and blockinglocal drainage when they come to rest. Their process ofsolidification involves a fall i~ the velocity of ~rbulentflow, below which the mud ceases to behave as a liquidand assumes solid proportions. It sets, and slowly thewater content seeps away to leave a resistant mass of

Photogeology

113

reaches

of a mature or old-aged stream.When the stream leaves its course

e, "-,,

depositiononth£ insid~ of bends (Fig. 4.43a). Sometimesmeanders curve back upon themselves to such a degreethat the narrow ground separating parts of a bend is cutthrough. The bend is cut off to form a stagnant oxbowlake (Fig. 4.43a). Features such as these typify floodplaiI'lswhere the alluvium is sufficiently well-bound byvegetation or because of its fineness to support discretestream courses. In semiarid terrains or in flat valleysfilled with poorly bound alluvium, stream courses areephemeral. Flow may change from one course to anotherrepeatedly, thereby forming a: braided stream system(Fig.4.43b).

Streams

flowing across

reworks the alluvium deposited there to...

towards their outside banks, as wellThis produces a system of low, arcuate

known as meander scrolls, which represent

Fig. 4.43 Continuous flow of streams over nearly featurelessplains of unconsolidated sediments results in migratingmeanders. In the example shown from the Canadian Arctic (a)traces of former meanders are preserved as meander scrollson the insides of bends, aI\d abandoned channels in the form

of oxbow lakes. Braided sh-eam systems form on similarsubstrates when flow is intermittent. As in (b), they are bestdevewped in semiarid terrains. Scales: (a) 1 :70 000 and

(b) Unknown. Sources: (II) EMRC A13139 82 + 83, copyrightHM the Queen in Right of Canada (1951),and (b) unknown.

114 Chapter 4

(a)

Fig. 4.44 Deltas formed by shifting distributary channels takeon the classic arcuate form, typified by that shown in (a). Itresults 'from the entry of a large stream into a lake in New York.Where sediment is deposited in standing water by anumberof fixed distributaries, most material is deposited near the

channels and builds up deltas of the birdsfoot kind (b).This system is part of the Mississippi delta in Louisiana.Scales: (a) 1 : 20000 and (b) 1 : 60000. Sources: (a) ASCSARU-IEE 74+ 75 and (b) NCICLa.8.

4.5.3 Aeolian landforms

The action of wind is not restricted to deserts, but canmould constructional landforms wherever sedimentsof a size small enough to be transported in air are notbound together. They may contribute to both coastal andglacial landscapes. The physics of wind and its meansof transporting particles are complex and sensitive tomany subtle factors. The dominant process is the effici-ent sorting of grain sizes into the bouncing and rollingload and that carried in suspension. For a particular windspeed the ranges of grain sizes present in both loadsis very narrow. Aeolian deposits are therefore very wellsorted. The rolling and bouncing load moves much moreslowly than suspended material, and so the two sizeranges are deposited in completely different areas..

Dunes result from the deposition of rolling and bounc-ing particles. Their form is determined by the amountof sand available in an area, the amount of vegetationcover and by the constancy of the wind. The classicmigrating, horn-shaped dune or barchan (Fig. 4.45b)forms where sand supply is low, little vegetation is avail-able to halt the movement of sand and wind directionis roughly constant. The horns point downwind. The leeslope, where deposition takes place, is steeper than thewindward slope up which grains are blown. Barchansmigrate slowly down wind. Where vegetation is able

,Changes in erosional base-levelina flood plain res1!ltin the stream system cutting into earlier alluvial deposits.Thi~ produces terraces of flat ground above the activeflood plain, on which old fluviatile features may be pre-served, anq into which minor drainage may have beenincised. It is often possible to recognize several terracelevels. Those With coincident height~ on either side ofthe flood plain are the res~lt of re~venation. Those thathave different heights either side are the products ofslow downward erosion and migration back and forthof the stream system.

Where streams flow into bodies of standing water,either lakes or the sea, the flow is halted and all sedimentis deposited. The result is a delta. If the delta developsshifting distributary channels, similar to those in an allu-vial fan, then it adopts an arcuate shape as a low-angledcone (Fig. 4.44a). Where persistent channels are main-tained deposition is from these and the delta propagatesas a branching birdfoot' delta (Fig. 4.44b).

Deposition in lakes, other than that associated withdeltas, is in the form of horizontal layers or in beaches.Drainage of a lake results in almost perfectly flat topo-graphy, interrupted only by the relics of former islands.Old lake beds can be distinguished from abandonedflood plains by the absence of meander scrolls and oxbowlakes, and the presence of curvilinear features markingthe sites of old beaches (see Section4.5A).

Photogeology

115

Fig.

4.45 .An abundant supply of sand, dry conditions andstrong prevailing winds favour the linear or sell dunes shown

from an area in the Middle East by (a). The major dunes areparallel to the ptevailingwind and on their crests have smallmigratory dunes, the horns of which point downwind.

Where

sand supply is restricted, as on the Kaibab Plateauof Arizona (b), parabolic, transverse and barchan dunesdominate.

Scales: (a) 1 : 20000 and (b) 1 : 40 000.Sources: (a) CSt Series B 9 + 10 and (b) NCIC Ariz.7 B + C.

to bind sand as it accumulates, dunes with the sameoverall shape as barchaJ:ls develop, but instead of thehorns pointing downwind they are fixed by vegetationand point windwards.

An abundance of sand and a lack of vegetationencour-age the formation of large dunes with crests that line upperpendicular to the prevailing wind. These transversedunes are typical of the world's largest sandy deserts.Where wind velocities are high, however, sand becomesstabilized in longitudinal or seif dunes running parallelto the wind (Fig. 4.45a). These can extend for hundredsof kilometres, and are so large and stable that they can be

preserved after climate has become more humid, whenthe dunes become vegetated.

.In deserts subject to seasonal fluctuations in winddirection, the dunes take on shapes conforming to theinterfering effects of several directions of sand transport.They may have a random pattern, but sometimes developdistinctive star shapes.

The material carried in suspension by wind travelslong distances until it reaches climatic regions with lowwind speeds. These usually are associated with the highpressu~ systems over continental interiors. Settling ofwind borne dust produces a peculiar sediment known as

116 Chapter4

Washington. The peculiar texture in the area stripped of loess,known as scabland, is a product of scouring and giant rippleformation by powerful stream action in glacial-lake overflowchannels. Scale: 1 : 20 000. Source: ASCS AAP-1G 123 + 124.

Fig. 4.46 The very finest material stripped from dry, baresediments Py the wind is earried long distances ill suspensionto be deposited as homogeneous mantles of loess. Streamerosion of loess produces a characteristic pinnate, or featnereddrairiage pattern well displayed by this stereopair of an area of

loess. It is exceptionally well sorted and homogeneous,and frequently lacks bedding. One consequence of thishomogen-eity and 'the process of deposition is that loesshas a tendency to break along irregular vertical surfaces.This imparts a distinctive form of drainage pattern toloes~-mantled areas in North America and Eurasia. Theyare dominated by dendritic networks of vertical-sided,U-~h~ped valleys and gullies. Each stream network looksremarkably like a feather, so they are described aspinnate (Fig.. 4.46). Most of the largest areas of loess stemfrom ablation of fine dust from glacial deposits duringthe Pleistocene, when wind systems controlled by highpressure over the ice masses blew from west to east inthe Northern Hemisphere.

4.5.4 Glacialland£orms

Constructional landforms resulting from glacial periodspresent the greatest variety of all. Not only ice and itsload of gouged~up sediment were involved, but alsothe meltwater and wind, which reworked the sediment,and the freezing temperatures, which imposed variousstructures upon and within it. It is impossible to give acomprehensive account of all the possibilities here, onlythose distinctive enough to identify terrains dominatedby glacial and related deposits. A thorough analysis ofphotographs of such areas requires reference to special-ized texts.

The material deposited solely by melting ice duringits retreat is given the catch-all nameilll, which covers amultitude of sins. In an area-of upland glaciation, till takesdistinctive forms related to various kinds of moraine.Ice sheets in lowland areas deposit till in two distinctforms. At the ice front where supply is balanced bymelting and ablation, debris is supplied continually. It issimply dumped and accumulates in a zone of exceed-ingly dirty ice. This forms an end'mor'1ine of one ofseveral varieties. They may mark the ultimate extent ofthe ice sheet, or temporary halts during its retreat. Thoseassociated with continental glaciation may extend asmore or less continuous and sinuous lines for thousandsof kilometres. They are characterized by a narrow belt ofhummocky topography in an otherwise almost feature-less plain. As till has a high content of fines, morainesare poorly drained and often have small lakes preservedin irregular depressions (Figs 4.47a and 4.48). Theseresult either from the irregular nature of the moraine orfrom the melting of buried ice masses and consequentcollapse. Collapse structures of this kind are known askettle holes.

Till deposited at .the base of a moving glacier hasoften been planed almost flat. Apart from sporadic kettleholes it presents a depressingly boring prospect. Itsuniformity, impermeability and low relief ensures poordrainage and the establishment of complex dendriticand deranged patterns. Subtle variations in relief and

planed to an almost flat surface. Drainage takes no particulardirection and rainfall eventually seeps into the till. The area ofbasal till in Indiana, shown in (b), is characterized by a mottledappearance owing to variations in soil moisture controlled bythe irregular but subdued topography. Scales: (a) 1 : 60000and (b) 1 : 20000. Sources: (a) NCIC N.Dak.2 B + C and(b) ASCS BWI-3BB 95 + 96.

permeability help produce a mottled texture of vegeta-tion, particularly in spring (Fig. 4.47b). The grindingmonotony of basal till is relieved by features producedby the once overriding ice mass. The most impressive ofthese are rounded, oval-shaped knolls known as drum-lins (Fig. 4.48). The typical 'basket of eggs' topography ofdrumlin country results from irregularities in ice flow,brought on perhaps by obstructions beneath or by con-fluence of several sheets. The long axes of drumlins areambiguous indicators of ice-flow direction.

Melting of the Pleistocene continental glaciers pro-vided huge volumes of water. Where this meltwaterflowed within the ice, meandering tunnels formed. Inthem the glacial debris was sorted and transported.Some of it was deposited in the base of the tunnels, toresult in sinuous ridges of well-dra,ined sand and gravel,termed eskers, when the ice melted (Fig. 4.49). Some-what similar tracts of well-drained sandy terrain formedfrom deposits in ice-margin streams, where the glacierabutted areas of high relief. These well drain~d kame

118 Chapter 4

Fig. 4.48 The masking effe<;t of regular fields in this stereopair of part of New York completely obscures the topography until thescene is viewed stereoptically. Then a series of drumlins becomes quite obvious, which has the typical 'basket-of-eggs' landscape.Scale:l : 60 000. Source: NCr<;:N.Y.3 A+B.

Fig. 4.49 The dominant material beneath this area of northernCanada is till deposited in an area of 'dead' ice, with its typicalknobartd kettle topography. Running across this ill-drainedlandscape is a sharp ridge of light-coloured, well'-drained

sand, lacking vegetation. It is an esker deposited in a subglacialmelt stream. Scale: 1 : 54 000. Source: EMRC Al4887102 + 103,copyright ~the Queen in Right of Canada (1955).

with sediment. The ice sheet, often up to a kilo.metrein thickness, formed prominent relief. The high ~nergystreams flowing from it slowed rapidly and depositedtheir sedimeJlt load to form outwash deposits. They are

terraces have a contrasted vegetation to the impervioustills that they cross.

In front of the ice ~argin water flowing from the topof the ice or from tunnels within it was heavily charged

Jhotogeology

119

)road

belts of sandy sedi~nt that become progressivelymer away from the ice front. In detail they are fans which

:oalesce, and on which braided streams flowed. This)asic structure is still easily recognized in many casesFig. 4.50). Where outwash engulfed patches of ice left)ehind

by glacial retreat, kettle holes formed, so that a)itted surface is sometimes seen. As they are so porous,

)utwash deposits are well drained and now have widely;paced surface drainage.

Pre-glacial drainage may have been dammed tempor-lrily by ice so that large glacial lakes formed. The deposits:esulting

from such lakes are identical. in most respects:0 those from any other drained lake. They are even more:tat and tedious than ground moraine. Drained glacialakes may even preserve relics of deltas from the melt-.\'ater streams that fed them. The sudden removal of thece dam during glacial retreat released huge floods, which:arved out deep valleys. Once the lake had drained, thesevalleys

became occupied by streams too small to have)een capable of forming them. The valleys have a typicalJ-shaped

cross-section, revealing that they were in facthe channels of these short-lived torrents. They are some-imes characterized by huge potholes and giant ripples)roducing scabland (Fig. 4.46). Episodic drainage of glacialakes resulted in successive shore lines being preserved1S

more or less parallel beach ridges (Fig. 4.51).

pits and bars during a perioo when the glacially depressed:rust was slowly rebounding isostatically. Scale: 1 : 36 000.iource:

EMRC Al6347 29 + 30, copyright HM the Queen intight of Canada (1958).

120 Chapter4

duces all the stereopa~ in this chapter as anaglyphs.There are additional anaglyphs of stereoscopic aerialphotography in the link Additional linages: Anaglyphs,where you can practice the skills that you have masteredin Chapter 4.

Further reading

These references comprise a useful compendium ofexamples of geomorphological, lithological and struc-tural interpretation from aerial photographs and satel-lite images in the reflected re&on. Many of the selectedreferences for Chapter 9 are also useful in geologicalinterpretation.

Fig. 4.52 Where unconsolidated sediments in northernCanada and other boreal areas are permanently frozen atdepth, the seasonal freezing and thawing of the upper soillayer produces regular patterns. This is accomplished by thevolume changes involved in ice formation and melting, whichslowly circulates and sorts grains of different grain sizes.Another contributory factor is the formation of ice lenses anddoming of the surface, followed by melting and collapse.Scale: 1 : 36 000. Source: EMRC A16347 30, copyright HMthe Queen in Rigl"\t of Canada (1951).

The unifying feature of all glaciated terrains is that theywere subject to intense cold for long periods. In commonwith modem boreal areas they were prone to permafrost.Among the many effects of permanently frozen ground,an easily recognized one is the formation of various kindsof patterned ground. The sediments most conducive toits formation are silts and clays, especially those of glacialla~e beds and alluvium (Fig. 4.52). In now temperatelatitudes, the patterns show up most clearly either inspring when vegetation is beginning to grow, or duringperiods of drought. The former cracks, now filled withporous debris, are relatively dry so that crops growingover them become stunted or discoloured.

Associated resources on the CD-ROM

Abrams, M.J., Rothery, D,A. & Pontual, A. (1988) Mapping inthe Oman ophiolite using enhanced Landsat Thematic Mapper

images. Tectonophysics 151,387-401.AI Khatieb, S.O. & Norman, J.W. (1982) A possibly extensive

crustal failure system of economic interest. Journal of PetroleumGeology 4,319-327.

Allum, J.A.E. (1966) Photogeology and Regional Mapping. Per-gamon, London.

Allum, J .A.E. (1980) Photogeology-c-early days. Geoscience Canada7,155-158.

Avery, T .E. & Berlin, G.L. (1985) .Interpretation of Aerial Photo-graphs, 4th edn. Burgess, Minneapolis, Minnesota.

Babcock, E.A. (1971) Detection of active faulting using obliqueinfrared aerial photography in the Imperial Valley, California.Geological Society of America Bulletin 82,3189-3196.

Bailey, G.B. & Anderson,P.D..£1982) Applications of Landsatimagery to problems of petroleum exploration in QaidamBasin, China. Bulletin, American Association of Petroleum Geo-logists 66, 1348-1354.

Baker,M.C.W. (1981) The nature and distribution of UpperCenozoic ignimbrite centres in the Central Andes. Journal ofVolcanology and Geothermal Research 11, 293-315.

Bodechtel, M., Kley, M. & Munzer, U. (1985) Tectonic analysisof typical .fold structures in the Zagros mountains, Iran, bythe application of quantitative photogrammetric methodson Metric Camera data. Proceedings, DFVLR-ESA Workshop,Oberpfaffenhofen, pp. 193-197. ESA SP-209, European SpaceAgency, Paris.

Crosta, A.P. & Moore, J .McM. (1989) Geological mapping usingLandsat Thematic Mapper imagery in Almeria Province,south-east Spain. International Journal of Remote Sensing 10,505-514.

Dikkers, A.J. (1977) Sketch of a possible lineament pattern inNorth-west Europe. Geologie en Mijnbouw 56, 275-285.

Drury, S.A. (1982) A regional tectonic study of the ArchaeanChitradurga greenstone belt, Karnataka, based on Landsatinterpretation. Journal of the Geological Society of India 24,167-184.

Drury, S.A. (1983) A Proterozoic intracratonic basin, thermalevolution and dyke swarms in South India. Journal of theGeological Society of India 25, 437-444.

Drury, S.A. (1990) SPOT image data as an aid to structuralmapping in the southern Aravalli Hills of Rajasthan, India.Geological Magazine 127, 195-207.

You can access the resources by opening the HTML fileMenu.htm in the Imint folder on the CD using your Webbrowser. More detailed information about installing theresources is in Appendix C.

On the Menu page, select the link Anaglyphs (Chapter 4stereopairs), which will take you to a page that repro-

ery, D.A. & Drury, S.A.(1984) The])etan Plateau. Tectonics 3, 19-26.

13-1489.In, L.C. & Wetlaufer, P.H. (1981) Relation between regionaleament systems and structural zones in Nevada. Bulletin,1erican

Association of Petroleum Geologists 65, 1414-1432.

t, N.M. & Blair, R.W., Jr (1986) Geomorphology from Space.\SA SP-486, US Government Printing Office, Washington,

.1, J. T. & Anson, A. (eds) (1968) Manual of Color Aerial Photo-rphy. American Society of Photogrammetry Falls Church,rginia.

n, M., Arvidson, R.E., Sturchio, N.C. & Guinness, E.A. (1987)hologic mapping in arid regions with Landsat ThematicIpper data: Meatiq dome, Egypt. Geological Society of America!lletin

99, 748-762.tn, M., Arvidson, R.E., Duncan, I.J., Stem, R.J. & EItliouby, B. (1988) Extension of the Najd Shear System fromudi

Arabia to the central Eastern Desert of Egypt basedintegrated field and Landsat observations. Tectonics 7,91-1306.

In, M., Becker, R., Arvidson, R.E., Shore, P., Stem, R.J.,Alfy, Z. & Guinness, E.A. (1992) Nature of the Red Sea: antroversy revisited. Geology 20, 593-596.ey, I.J. (1988) The reconstruction of palaeodrainage andgional geological structures in Australia's Canning andfficer Basins using NOAA-A VHRR satellite imagery. Earth'ience

Reviews 25,409-425.katakrishnan, R. & Dotiwalla, F.E: (1987) The Cuddapahlient: a tectonic model for the Cuddapah Basin, India,lsed on Landsat image in~rpretation. Tectonophysics 136,,7-253.Bandat, H.F. (1962) Aerogeology. Gulf Publishing Company,ouston.

dai, T. (1983) Lop Nur (China) studied from Landsat andR-A imagery. ITC Journal 4, 253-257.litz, E.R. (1932) Drainage patterns and their significance.urnal

of Geology 40, 498-521.

Increasingly, remotely sensed images are in digital form(Chapter 3) as two-dimensional arrays made up of pixels.Each pixel is assigned a digital humber (ON) that rep-resents the energy or radiance of the waveband beingmonitored. Digital image handling converts these arraysof ON into signals that modulate the beam of electronsinvolved in producing a video display. The beam sweepsa phosphorescent screen repeatedly, so refreshing thedisplay continuously. Images generated by this meansare called rasters, so such data are commonly said to bein raster format. Desk-top scanners reproduce photo-graphs in digital form, including the additive red, greenand blue components of colour photographs. Chapter 8explains how software convert data that do not relate toradiation, such as topographic elevation, gravitationaland magnetic potential, and geochemical data that arer~corded at isolated points or semicontinuously alongsurvey lines, into raster format for display and variouskinds of image processing.

The ON generally are coded in the 8-bit (1 byte) binaryrange corresponding to 0-255, where ~ach step cor-responds'to a range in the original real data. Such byte-format data are ideal for computer manipulation usingimage-processing software, although modem computersare designed to handle 64- or 128-bit floating pointnumbers, and do so with sufficient efficiency for an 8-bitformat no longer to be necessary. The degradation isfor economy of storage and transmission, particularly onorbiting platforms where data rates, power and storagemedia are limited. It may seem that the degradationfrom real information presents a1imit to image analysis.So it does, especially for conversion of radiance data tomeasures of reflectance. Moreover, the devices are tunedfor all likely radiances upwelling from the surface, fromthe most reflective to the most absorbent. That gener-ally means that data for anyone scene are compressedto only a fraction of the available 0-255 range, ratherthan being optim(il for the radiant properties of eachscene. Where quantitative information is essential, as insensors aimed at atmospheric composition or tempera-tures, such as those on the NOAA A VHRR, analogue todigital conversion involves a greater range of ON. Forqualitative interpretation and many semiquantitativeapplications, however, the data preserve sufficient pre-cision for them to be useful in practice. A colour imagecomprising RGB channels that span a real range of 6 or7 bits (64 or 128 ON) will show up to 221 or 2 millioncolours. That is close to the limit of human colour dis-crimination, and sufficient for various kinds of automatedclassification.

The design of digital imaging devices ensures thatdata from different parts of the spectrum are in per-fect spatial register with one another. Thus any pixel ina scene can be thought of as a stack of DN, each layercorresponding to one waveband. Data from differentdevices or processes can be registered to one another,particularly to a common cartographic base using geo-metric correction and rectification software (Appendix B),thereby increasing the depth of the 'stack'. This registra-tion has two important advantages. First, a colour videomonitor displays three 'layers', or mathematical derivat-ives from them, together as additive (RGB) colour images.This is possible because video pictures are themselvesmade up from exactly positioned pixels of red, greenand blue phosphors on the screen. The phosphors glowin proportion to the electron flux from red, green andblue colour guns, as the raster repeatedly sweeps fromside to side and top to bottom to make up the picture.The DN in the image data modulate each electron fluxon a pixel-by-pixel basis. The exact registration of pixelsalso enables the DN for different data 'layers' to be com-pared and combined in many different ways withoutdisrupting the basic structure of the image.

The objective of this chapter is twofold: to introducethe methods of digital image processing, and to showhow they can be applied to geological remote sensing.Some image-processing tasks aim at removing distor-tions and blemishes that result from imperfect meansof gathering data. They are introduced in Appendix B.Those covered here relate to enhancing the quality ofimages for improved visual interpretation, to extract-ing information that cannot be perceived and displayingit as an image, and to measuring and defining degreesof statistical similarity within a scene. This last topicinvolves using the computer as a trained but unbiasedmeans of recognizing spatial and spectral patterns, orclasses, within the image data.

As far as possible, the treatment is descriptive ratherthan mathematical. There are many different image-processing software packages, and several approachesto the tasks that they perform. The account therefore isgeneral and does not venture deeply into computer hard-ware and software architectures. Any geologist using animage processing package will have to become familiarwith its individual foibles, its internal algorithms andthe way in which it interacts with the user. This chapterprovides a basis for understanding how image pro-cessing works, a guide to selecting appropriate meansof processing images for various purposes, "and someexamples of their application.

Digital Image Processing 123

(a)

5.1 The image histogram

ON

(b)

~~

Cj'c:Q)::JC"Q)...

'+-Q)>

"+".!J.::JE::JU

DN

Fig. 5.1 The probability density function (PDF) of animagecan be expressed by two fo~s of histogram. In (a) the numbeor frequency of pixels assigned to each digital number (DN)are plotted as bars against DN. Each division on the DN axisis te~ed a bin and is the width of one of the bars. In (b) thevertical axis is the cumulative frequency of pixels haVingDN less than a particular value. For example, the cu,mulativefrequency for DN = 50 is fso + f49 + f48 + ...+ fo- so that thecumulative frequency for the maximum DNis 100%. Thesame distribution of DN appears as peaks and sometimestroughs in (a) and as changes of gradient in a roughlyS-shaped curve in (b).

An image of a single band of digital remote-sensing datais a representation of how the radiant energy reflectedor emitted by the surface is distributed in two spatialdimensions. The energy is expressed as ON, which ina display are represented by a variation in brightnessshowing as different grey levels. Since the eye is capableof distinguishing only about 30 grey levels in a blackand white image (Section 2.3), a display of up to 256 greylevels appears to be continuous. In fact it is made up ofdiscrete steps. An image also can be expressed statist-ically as the probability of finding ON of a given valuewithin it. This measure is properly termed the prob-ability density function (POF). The POF is representedmost conveniently by a histogram offue numbel;, of pixels,which, regardless of spatial position within an image,have a particular ON (Fig. 5.1).

The histogram of ON frequencies is probably thesingle most useful measure in digital image processing.Its shape indicates the contrast and homogeneity of thescene. For instance, a scene of a homogeneous surfacewith a low contrast will produce a histogram with asingle, sharp peak. A broad single peak suggests homo-geneity but a wide range of contrast. Images that containseveral distinct types of surface cover may show multi-ple peaks. If each has a significantly different averagebrightness there will be a clear separation between thesepeaks. As their average brightness becomes more alike,so the peaks will begin to merge. Other shape attributesof the histogram give a kind of statistical shorthand foran image. The presence of 'tails' and the degree of asym-metry of the peak both indicate important structuralfeatures in an image that will rarely be obvious fromthe picture itself. Figure 5.2 shows examples of imagesand their associated histograms.

The POF also may be represented as a cumulativehistogram, with the cumulative frequency of pixels witha particular ON plotted against ON. For each ON thecumulative frequency is that of the ON plus the frequen-cies of all ON less than it. So the cumulative frequency ata ON of 128 is f128 + f127 +.£126 +. ...+ fo' In this case thesteepness of the curve is a measure of the contrast, andheterogeneities show up as kinks in the curve (Fig. 5.1);

It is of course possible to produce histograms fordifferent parts of an image, so that the data structure invarious types of terrain can be compared. These, togetherwith the histogram of all the data, form the basic require-ments for contrast stretching (Section 5.2), because itis effectively the histogram that is manipulated in thisoperation. A crude idea of the contribution of known typesof surface to the histogram can be obtained simply byidentifying known pixels and dumping their ON in dif-ferent bands to a line printer or the computer console.

Multispectral data can be expressed in multi-dimensional histograms, the simplest of which is thetwo-dimensional histogram (Fig. 5.3). These are simplybivariate plots of DN for one band against those foranother. They display visually the degree of correlationbetween the two bands and identify instances wherea surface has distinctly different kinds of respon~e tothe two bands. The density of plots of individual pixelsis a measure of where peaks in the one-dimensionalhistograms coincide. Highly correlated data, or classesof surface materials that have similar appearances onimages of the two bands plot near straight lines on thetwo-dimensional histogram. Poorly correlated data showas shapeless clouds of points, as in Fig. 5.3, in which theoverall poor correlation stems from the scene being a

lL4 Lhapter 5

(a)

~~ 5c:QI:JcoQI~

lJ ;,v IV\) I;)U LUU L:>UDN

(b}

15[

~ 10~c:Q):JC"Q)...u.

u 50 lUU 150 200 250DN

Fig. 5.2 (above) Diffetentkinds of terrain sometimesproduce contrasted histograms. Both images are ofLandsat MSS band 7. In (a) the bright areas of saltflats and dark areas of lavas in the Andes are morecommon than surfaces with intermediate digjtalnumbers (DN). The histogram has two peaks and issaid to be bimodal. Image (b), also from the Andes, hasa simple probability density function (PDF) whichgjves a nearly symmetrical, unimodal histogram, eventhough areas with distinctly differetlt DN are visible.

Fig. 5.3 (left) This bivariate plot of digital numbers(DN) in Landsat TMband 4 against those in band 3 isfor an area comprising a mixture of open water, baresoil and rock, and variable vegetatiol} cover. Becausechlorophyll absorbs red (band 3) energy and plant cellsreflect infrared (band 4) very strongly, vegetation-dominated pixels have higher DN in band 4 than theydo in band 3. Vegetated pixels form a cluster whichdeviates from the roughly linear distribution of pixelscontain~g rock and soil.

Digital Image Processing 125

-vegetated surfaces, bare rock andThe rock and soil show high correlation,

-to produce enhanced colour images.

manipulated

as the basis for

screen of a digital image processor, or thein a digital film writer, expresses ON inrange

of intensities from 0 (black~ to 255.., The simplest func-

using a look-up table (L{!JT). A LUT converts an input(ON(i,j)J to an output (ON(i,;)o), where (i,j) expressesthe position of a pixel in the image. For a linear stretchthe function is the basic equation for a straight linerelating ON(i';)i and ON(i,j)o:

ON(i,j)i = m(ON(i,j)o) + c,

where, m is the gradient 'and c the intercept on theON(i,j)o axis. The equatio~ can be expressed as a graphof input against output (Fig. 5.4a). Because input and out-put ON have a limited range (0-255) the computer doesnot have to spend time performing the same calculationon each of the millions of pixels in an image. It simplyreplaces the input value with the corresponding outputvalue from the LUT.

A simple linear stretch of the data in Fig. 5.4(a) pro-duces an image that at least shows some of the surfacedetails (Fig. 5.4c). It still appears to have a rather poorcontrast, however-it appears hazy. The reason for thisis twofold. First, each ON in the original data has a ,con-tribution from atmospheric scattering (Section 1.3.1),the amount decreasing as wavelength increases. Part ofthis is the scattering of upwelling radiance, and part theway scattered solar radiation ('sky light') illuminates thesurface. The second is that the maximum ON is actuallyfor small clouds in the image that are much brighter thanthe brightest part of the surface. The effect of the atmo-sphere is seen on the histogram of raw data (Fig. 5.4a),where there are no pixels with a ON of O. On an airlessplanet total shadows would be completely black, soone way of correcting for the scattered component isto identify shadowed pixels, find their ON and set allpixels with this and lower values to O. The chosen ONis known as the cut-off. This is termed a dark-pixelcorrection. Care must be taken not to choose dark pixelsin water, for in the visible region their ON may be theresult of suspended sediment, or a shallow bottom, notatmospheric effects. Using their value for cut-off wouldresult in blacking out areas of dark but unshadowedsurface. Another method of atmospheric correction isto plot ON for a long-wavelength band-usually in theinfrared, where scattering is minimal-against thosefor a band in the visible region, where scattering hasa greatf!r effect. A line of best fit drawn through the dis-tribution will intercept,the short wavelength axis at aON approximating the scattered component (Fig. 5.5).This value is then used as the cut-off. A third methodis by inspection of the histograms for all the bandsand estimating the shift attributable to scattering ineach. Whichever method is chosen, some care is needed.Valuable information caR be present in shadowed areasbecause they are lit by the scattering effect of the sky.Correction will lose this information. Loss of informa-tion must be traded-off against the improvement inimage contrast resulting from atmospheric correction.

-.to any of these 256 intensity levels. It can

range of ON to another. This is the

-..transformation may be required, but the most

, the way in which digital image

objective of all land-orientated remote-sensing

In

other words

be

expressed as differences in ON. The

As

a result the histograms for--

--5.4a). This results in loss of data,

--for unwanted bright materials, such asThe human eye is capable only of dis-

30 grey levels in the black to whiteonly if they are adjacent and sharply

5.4b), even though it might contain 100 differentr .-means expanding

ilistogram to the full range available in

The simplest means of improving the contrast in

-the o-2,?5 range. The minimum DN is setto 0, the maximum to 255. Each histogram bin relating toeach DN originally present is moved to a new position,and the new bins are eq~ally spaced (Fig. 5.4a). This isa linear stretch. The computer accomplishes this by

126 Chapter 5

(a)

0 50 150 200 255255

200Fig. 5.4 (a) Diagram showing thehistogram for raw Landsat TM band3 data (left) for a mixed terrain onthe Sudan-Eritrea border, whichare compressed towards low digitalnumbers (DN). The simplest contraststretch uses a linear look-up table inwhich DN of zero remain zero andthe highest in the data (180) are set to255, to produce an output histogramshown at the top. Display of the rawdata produces a dark, low-contrastimage (b) in which few details canbe seen. Using the linear contraststretch shown in (a) introducesdiscernible contrast in the resultingimage (c). This image isapproximately 30 x 25 km.

saturation ON

150linear stretch

look-up table (LUT)

100

50

cut-off DN

0

(b) (c)

sharpens the high spatial-frequency features in the image(Section 5.3). Enhanced images of single bands or false-colour images comprising three contrast-stretched bandscan-then be interpreted geologically with a fair measureof success. They can be improved further using the inter-active possibilities of the computer to suit the aestheticpreferences of the interpreter. More contrast can beintroduced by increasing the value of the cut-off ON anddecreasing the value of ON above which all values are

Figure 5.6 shows the improvement in image qualityresulting from atmospheric correction of Fig. 5.4(c). As asmall cloud is present in the image, and has a DN of 180(Fig. 5.4a), the contrast has been further increased bysetting the maximum DN on soil or rock to 255; therebysaturating all pixels with higher DN.

A linear contrast stretch with atmospheric correction isin most cases sufficient to produce an image of high qual-ity/ because the removal. of haze and increased contrast

0 50 100 150 200 255

band 7

~

Fig. 5.5 A bivariate plot of Landsat TM band 5 againstband 7 (a) has a line of best fit (the major axis of theelliptical plot), which extrapolates to the origin (0, 0).This shows that atmospheric scattering is minimal inboth short-wavelength infrared (SWIR) bands. A plotof band 5 against band 2 (b), which is badly affectedby atmospheric scattering, gives an estimate of thescattering component in band 2 from the intercept ofthe distribution withthe band 2 axis.

0 ~U 100 15u LOO 255

band 2

set to 255 or maximum. Increasing the cut-off will darkena larger percentage of a scene. Decreasing the ON formaximum brightness increases the image area showingas light. However, some information is lost in both cases.

Occasionally a scene has areas dominated by both lightand dark surfaces, as in the example used here, both ofwhich contain geologically interesting information. A

simple linear contrast stretch will not perform an enhance-ment ideally suited to both. Contrast in the central partof the range will be enhanced at the expense of the lowand high ends of the range. Again the imperfections ofhuman vision will not allow a complete interpretation.There are many ways of overcoming these limitationsby more complex methods of stretching. The simplest is

128 Chapter 5

histograms for the contr~sted areas of geological interestand to devise a piece-Wise stretch. In this operation thegeologically unimportant parts of the overall histogramare compressed, and the remaining space is occupiedby linear stretches of the light and dark ranges selected.Sensible contrast is thus introduced into both light anddark areas. The LUT for such a manipulation and theresulting image is shown in Fig. 5.8.

More sophisticated stretches involve creating LUTsusing nonlinear transformations. To avoid the inherentloss of contrast in the tails of the original histogram inlinear stretches, the raw data can be forced to resemblea normal distribution. This obeys Gaussian laws of stat-istical distribution, produces a bell-shaped histogramwith its peak at a DN of 127 and results in a Gaussian ornormalized stretch. The manipulation of bins in the rawhistogr~ to produce an equal population.' density ofpixels along the ON axis results in the spacing of binsbecoming inver~ely proportional to the number of pixelsthat theycontam;this is an equalization stretch (Fig. 5.9).This is achieved by forcing the cumulative histogram(Section 5.1) to become a straight line, and because of thisthe operation is sometimes known as a ramp stretch.Ramp stretches result in the greatest contrast stretchbeing applied to the middle, most populated range ofDN, with compression in the less densely populated lowand high DN tails. Sometimes it may be necessary tostretch the dark part of a~ image as much as possiblewhile still retaining some contrast in the light range.This is possible using a .logarithmic stretch (Fig. 5.9). The

Fig. 5.6 This image is of the same data as Fig. 5.4(b),.but theyhave been corrected for atmospheric scattering and for smallbright clouds (DN = 180 on Fig. 5.4a). All the data that relate tothe surface have been stretched linearly to occupy the whole0-255 range.

to select cut-off and maximum for linear stretching ofthe dark and light parts of the scene, to produce twoseparate images. Figure 5.7 illustrates this. The imagein which dark areas are stretched (Fig. 5.7a) has allthe light areas washed out. The converse holds for theenhancement of the light areas (Fig. 5.7b). This is a time-consuming and inconvenient approach in all but themost extreme cases. A compromise is to investigate the

(a) (b)Fig. 5.7; In (a) the low digital numbers (DN) of Fig. 5.6 havebeen stretched linearly to occupy the full 0-255 range. Thedark terrain now shows more detail, btit the light areas are

washed out. Image (bjhas had all the low DN set to 0, sothat maximum contrast in the light alluvial flats allowstheir variation to be seen.

~WT~~~C'0 ~ \.J~~~ ~~ L-UC '0 QAA"\.

~-"

Digital Image Processing 129

(a)

0 50 100 150

Input DN

200 250

(b)

5.2.1 Choosing bands for colour images

The Landsat MSS and SPOT systems record data fromonly three regions in the visible and VNIR parts of thespectrum. Effectively, only one kind of false-colour com-posite image can be generated, where VNIR controls red,visible red controls green and green controls blue, togive the unfamiliar rendition of vegetation in reds andapproximately natural colours for other kinds of surface.Contrast stretching can improve the range of coloursin such an image, but there are limits imposed byt_hestrong correlation between the MSS and SPOT bands(Section 5.2.2). Newer systems, such as the LandsatThematic Mapper, JER5:-1, various airborne multispec-tral scanners and imaging spectrometers provide morebands in the visible, SWIR and thermal infrared regions,centred on particular spectral features associated withdifferent kinds of surface material. The data in theseregions are less correlated and the number of possiblefalse-colour images is increased. ,An example of justhow useful this flexibility can be is shown in Plate 5.1,comparing a natural colour Landsat TM image with astandard false-colour image and one where red, greenand blue (RGB) are controlled by SWIR, visible red andblue wavebands. However, the more bands that areavailable the greater is the choice of possible three-bandcombinations, including the order of their renditionin RGB. Considering the last point involves attentionto the proclivities of human vision, aesthetics in par-ticular. Quite simply there is usually a single renditionof three particular bands that is more visually stimul-ating (and therefore more readily interpreted) than theother five possible RGB combinations, even though theinformation content in all six RGB renditions is iden-tical. Plate 5.1(c,d) illustrate this point. Although thereasons for this are not fully understood, a simple ruleof thumb is to render the most informative band fora particular purpose in red, the next in green and theleast informative in blue.

Choosing three bands for RGB display, when animportant objective is to use as few images as possible ininterpretation, is an increasingly difficult task for datawith large numbers of bands. For the six reflected bandsin a Landsat TM image there are 20 combinations, foreach of which there are six RGB orders, compared withonly six possibilities for three-band SPOT data. Clearlythe potential for confusion with 120 possible colourimages, let alone a variety of enhancements for each,is enormf;)us. The numbers involved in imaging spec-trome~ data are greater than there are protons in theuniverse! Stati~tical analysis of the correlation betweenthe bands can help in selecting two or three combina-tions for a particular objective. Inspection of Fig. 5.10,showing all six Landsat TM reflected bands, reveals thatall are very similar, owing to the high correlation-in the

Fig. 5.8 The look-up table (LUT) (a) for a piece-wise stretch ofFig. 5.4(b) shows two different linear stretches, for the lowrange and the high range of geologically important digitalnumbers (ON). The resulting image (b) has improved contrastin both dark and light terrains. In a clearly bimodal image,such as Fig. 5.2(a), the best strategy would be to stretch thepeak of low ON, compress the trough of intermediate ON andstretch the peak of high ON. In the LUT shown in (a) thiswould introduce a low-angle segment between the twostretched components.

converse is achieved by an exponential stretch (Fig.5.9c). Virtually any mathematical transformation can bedevised for contrast stretching, but in most cases theresults are not a sufficient improvement over the simpleroptions to warrant the time involved.

Plate 5.1 gives examples of false-colour images inwhich the component bands have been interactivelystretched to produce the optimum contrast and colourbalance for the geological information of interest.

130 Chapter 5

(b) (c)(a)

output histOgra~IIIIIIIIIIIIIIII.llrlltlm IIIIIIIIIIIII

Output

0 50 100 150 200 255

~11111!1111111111111111~'~~'~~~ .~i.:tOgram

Output

0 50 100 150 200 255

255 ;

ou~ut histogram

Output

.0 .50 100 150 200 255

255 j~E'"c,0~

:c...OJC-

.!:

E'"c-o~:i:..."C-

.!:

200

...150~CoC

-100

+-'OJC-o!:

exponential stretchlook-up table (LUn

logarithmic stretchlook-up table (LUT)

50

0

Fig. 5.9 The effects of (a) an equipopulation or ramp stretch, (b) a logarithmic stretch and (c) an exponential stretch on Fig. 5.4(b) areshown as the look-up table (LUT), input and output histograms (top), and resulting image. "

ones are: Fe~ oxides and hydroxides, which govern'redness' and absorption around 0.9 ~m; overall reflect-ivity or albedo of common rocks, which peaks around1.6 ~m; minerals that contain OH-, which have absorp-tions around 2.1-2.2 ~m, and carbonates with absorp-tions around 2.35-2.4 ~m (Section 1.3.2). The reasonthat Plate 5.1(c) (TM bands 5, 3 and 1 as RGB) is farmore useful geologically than Plate 5.1(a) or (b) is thatit expresses albedo and redness well, and the overallreflectivity and iron content of the local rocks are morevariable than any other mineralogical characteristics.The narrower the spectral bands, the more strategicallyplaced they are relative to spectral features of minerals,and the more features that are covered, the greater thepossibilities for lithological discrimination.

In the thermal infrared many more minerals, includ-ing the bulk of rock-forming minerals as opposed tominor minerals and surface coatings, influence the detailsof emission spectra (Figs 1.12 and 1.13). Consequently,given multispectrcl:l thermal data such as TIMS or thoselikely to become available from ASTER, the direct map-ping of true petrographic categories becomes a distinctpossibility (Chapter 6).

reflected part of the spectrum because of the dominanceof albedo. They do contain differences, but they are quitesubtle. Computer analysis of the correlation can identifythreefold groups of bands in order of their nonredun-dant information content.

However attractive, such an automatic proceduremakes no distinction between the reasons for theorder, which in the vast majority of cases arises fromthe spectrally distinct nature of vegetation (Fig. 5.3),even when it forms only a very sparse cover to an area.Although RGB combinations that emphasize vegeta-tion differences can be useful in geological mapping,a better choice is to select bands on the basis of whetheror not they contain spectral features resulting fromthe mineral content in soil and rock. Compared withthe effects of vegetation, such features will have a verymuted influence, but if the minerals involved are presenttheir influences willbe.there in the data, unless vegeta-tion cover is total. Section 5.5 discusses inspection ofmineral spectra in designing uses of ratio images, but afew guidelines are appropriate here.

In the reflected region only a few groups of mineralshave marked effects on band response, The dominant

(b)

(e) (f)

Images (a) to (f) are of the six reflected Landsat TM bands (1, 2, 3, 4, 5 and 7, respectively) for the area at the Sudan-Eritrea."..- 5.4.

both emitted and reflected data, high correla-

First, it means that there is a great dealdata, that is, each band contains more

the same information apart from subtleties,for general purposes anyone band could be sub-

they are, the more the redundancy.that is a problem for computer storage as well

source of confusion. A seven-band Landsat TMoccupies 250 megabytes of magneti,c storage.

.~ ,

Clearly some means ofreduction are needed (Section 5.4) to extract the

The full range of 2563 colours possible in

5.2.2 Contrast stretching in different colour spaces

Figure 5.11(a-<:) demonstrates schematically the way inwhich high correlation between two bands is retained dur-ing linear contrast stretching, and Fig. 5.11 (d) expressesthe limit of such stretching in RGB space. Using anyof the methods of contrast enhancement discussed sofar would produce the same result. The stretched datafall within an elongated ellipsoid, thereby producingonly a very limited range of the colours that are possible.The full range .would occupy the entir~ volume of thecolour cube in Fig. 5.11(d). To fill that range wouldinvolve drastically reducing the degree of correlationin the data, which is the object of decorrelation stretch-ing. This is discussed in Section 5.4, because it relies onstatisti<;al manipulation of data. The methods coveredhere depend on changing the co-ordinate system usedto represent colour.

The major axis of the ellipsoid in Fig. 5.11 (d) definesthose points where the red, green and qlue componentsare equal, thereby pr~ucing shades of grey. This is thegrey or achromatic axis inRGB space, and the rest of the

the correlation mu;st be sought

132 Chapter 5

(a) (b) (c)(d)band 2

255

255band 1

255band 1

r

,

Fig. 5.11 Bivariate plots of two bands, showing (a) the rawdata, (b) the data after linearly stretching band 1 only and(c) the data after stretching both bands. The high correlationbetween the two bands is retained. (d) Perspective view of

RGB coloqr space, showmgthe volume occupied by ONfrom three correlated bands of reflected data in the form ofan ellipsoid with its principal axis along a line where red,green and blue ON are equal.

Fig. 5.12 Intensity, saturation and hue (HIS) defined in RGBspace. See Plate 5.2.

ellipsoid represents pastel shades and the colour limitsof the associated image. Only points closer to the RGBaxes represent vivid colours, and high correlation mean~that there can be none, except as a result of signal noise.A position on the achromatic axis, or a projection on toit from a coloured pixel is a measure of the intensity(Section 2.4), or the brightness of the colour. The actualcolour can be expressed by two other measures. The closerit is to the achromatic axis the more pastel the colour, thefurther away the more vivid or pure is the colour. Thismeasure is the saturation, which is measured at rightangles to the axis. The third measure of colour in relationto the achromatic axis is the hue. This varies around acircle, also at. a right angle to the axis. Hue is essentially ameasure of the dominant colour, whether it is a pastelshade or more pure. Figure 5.12 shows the relationshipbetween the descriptions of colour in the RGB andhue-intensity-saturation (HIS) co"ordinate systems. TheHIS system is in the form ofa symmetrical cone with itsapex at the black point of RGB co-ordinates. Plate 5.2 isa section at right angles to the achromatic axis, showingthe relationship between hue and saturation in the formof a colour wheel.

Intensity is a number from 0 at the apex of the HIScone to an arbitrary 255 at the limit of RGB space.Saturation is an angle between 0° and half the apicalangle of a cone that extends to the limits of RGBspace.Hue is also an angle between 0° and 360° around thecolour wheel. Red is chosen as the hue correspondingto 0°, green is at 120°,. blue at 240° and the circle iscompleted by 360° at red again. To enable HIS space tobe exploited in image processing, the RGB co-ordinatesof each pixel are converted to HIS measures by a mathe-matical transformation. Rather than leaving saturationand hue in the form of angles they are expressed inthe 0-255 range, when their ranges, and that. of intensity,can be modified independently by simple stretching.Stretching intensity is akin to using the same contrast

stretch on the red, green and blue bands in RGB space.Because stretching hue leads to a complete modifica-tion of colour, it is normally left unchanged. Stretchingsaturation is the most important manipulation, for this isthe same as inflating the RGB ellipsoid of highly corre-lated data to fill a greater proportion of available colourspace. After such manipulation the modified intensity,saturation and hue measures are converted back again toRGB space by the inverse of the mathematical transform,so that the enhanced image can be displayed on a videomonitor. However, most image processing software nowincorporates options for direct display of HIS ffies as

Digital Image Processing 133

intensity and saturation only, onThe result is a far more vivid image, but

results stem from transforming RGB

takes less time. There

other manipulations on intensity/such as

one as intensity another as hue. Some--

filtering

of spatial-frequency--

scene brightness variations are expresseddistributed as rectangular pixels. The variation

--

Distance

Fig. 5.13 Variations in digital number (DN) along a line on animage (a) can be expressed as a range of s~e waves (b) withdifferent amplitudes and frequencies. Their cumulative effectis the same as the real spatial frequency distribution of theimage.

image. The fluctuations on such a graph

and amplitudes. When superimposed

give the same effect

5.13). In an image, sine waves in two

express the variation in DN rather than in.~

with different amounts ofthat result from topographic effects,

those between brightly lit and shadowedc. c ..

to another. In the most extremeis an abrupt change from an area of

--

On a graph of ON with distance this is repres-a steep or even vertical gradient. Boundaries

kind are known as edges. They occupy only aand are thus high-frequency features. They

than they are long, and can define lines---

There is of course a complete range from high-frequencyfeatures through medium to low frequency. The scaleand gradient associated with changes in DN within animage are closely interwoven, and both help in recogniz-ing different types of boundary. For geological purposesthe most important boundaries are often the edges. Theymay be shadow effects that give information on topo-graphic features, some of which represent rock types withdifferent resistances W erosion, or tectonic weaknessessuch as faults and 'joints. Some edges are boundariesbetween rocks with different reflective properties, soilsderived from different rock types, or vegetation bound-aries that have an underlying geological control. Edgesmay occur in isolation, as in the case of faults ~r otherboundaries separating large masses of uniform butdifferent rock types. They also may be closely spaced, .,especially when they represent compositional bandikgor joints.

Whereas edges may represent small-scale geologicalfeatures, medium- and low-frequency spatial featuresoften sh<?w the gross geological features of an area.;..-Folds may repeat stratigraphic sequences on the scaleof kilometres. They define medium-frequency spatialfeatures. Phenomena such as batholiths, major fault-blocks, sedimentary basins and orogenic belts controllow'-frequency features with dimensions in tens or hundredsof kiIometres. The lower the frequency of a feature on

between fields with different crops, roads-and between shadows

expected where natural vegetation changes with altitude,have low gradients, occupy large spa~s and thereforeare low-frequency features. Any linear attributes theymay have are vague and difficult to define.

134 Chapter 5

domain, or in the spatial domain of the image itself bya process known as convolution. Frequency-domainffitering is more powerful, but is also the more expensiveof computer time, involves highly complex mathematicsand the result of a transform is not easily visualized interms of the image itself. Most image processing systemsroutinely use convolution ffiters, with the option offrequency-domain filtering for special purposes.

The easiest way to understand convolution filtering isto follow the actual process that takes place in the com-puter itself. Convolution is accomplished by what isknown as a box filter. An image can be represented bythe expression or function P(i,j), the original DN at everyco-ordinate in the raster, where i and j are the line andpixel (row and column) co-ordinates. A box filter is amatrix of dimension 2M + 1 rows by 2N + 1 columns, sothe number of rows and columns is odd and one cellalways lies at the exact centre of the matrix. Each cell inthe box ffiter has a weighting C, and the whole matrixcan be expressed by the functio; C(k,l), with k and I beingthe cell co-ordinates within the box. An output imageO(i,j) is computed by convolving the image f(i,j) withthe matrix C(k,i). This operation is mathematicallyexpressed by:

M N

O(i,j) = 1 1 P(i + k,j + l)C(k + M + 1,1 + N + 1)k=-M I=-N '\ -

an image the more subtle it tends to be, and {he moredifficult it is to perceive, depending on the associatedcontrast (Section 2.2).

The spatial frequency of any feature Is determinedby the scale of the image and its resolution or pixel size.With decreasing scale the higher the frequency of aparticular kind of feature will tend to become. As resolu-

~ tion coarsens, the highest frequency features disappearaltogether because they are smaller than an individual,.pixel. What would be undetectable subtle variation on alarge-scale image often become clearly distinguished at,small scale with coarse resolution. They are transformedin some cases from low-frequency features to edges. Thehuman visual syste~ responds to varying spatial fre-quencies in images in a complex way, depending on thescale and viewing distance of an image and on the eye'sMTF, which is different for black and white 'Vld colourimages (Section 2.2, Equation 2.1).

The process involved in improving the appearanceand the interpretability of the spatial distribution ofdata in an image is spatial-frequency filtering. It consistsof selectively enhancing the high-, medium- and low-frequency variations of ON in an image. It is not alwaysnecessary. Simply by manipulatirig the contrast of animage the properties of the human MTF (Section 2.2)can be exploited. Reducing the contrast leaves visibleonly the low- and medium-frequency features, whereasincreasing it accentuates the sharp changes in contrastassociated with edges. There is a limit to which either ofthese simple contrast transformations can be taken, how-ever, when the visual quality of the image breaks downand interpretability begins to degrade. A more powerfultechnique is to use various mathematical transforms toextract or accentuate selectively the high-, medium- andlow-frequency variations.

A mathematical technique for separating an imageinto its various spatial-frequency components is Fourieranalysis, the details of which are beyond this book.The basic principle is illustrated by Fig. 5.13 for asingle dimension. The Fourier transform of an image,the result of such a separation, expresses the spatialattributes of an image in terms of their frequencies,amplitudes and their orientation. It is a transform thatenables certain groups of frequenci.(ls and directions tobe emphasized or suppressed by algorithms knownas filters. Those that emphasize high frequencies andsuppress low frequencies are high-pass filters. Sim-ilarly there are medium- and IQ'w-pass filters. Moreover,selected ranges of spatial freq~ncies can be removed orretained in the~esulting image, using !:>~n~t~ and~and-p~s~ filters. The process is analogous to the elec-

, tronic filtering in amplifiers to reduce hiss and rumble,enhance the bass or treble and so on in a sound record-ing. Filtering can be implemented through the Fouriertransform, when it is said to operate in the frequency

i.e., O(i,j) = the sum froIrt k = -M to M of the sum from1 = -N to N of P(I + k,j + 1) multiplied by C(k + M + 1,1 + N+1). ; '\

In familiar terms this means that the Jatrix C is over-lain on the image with its central cell (M + 1,N + 1) ontop of a pixel (i,j), and the other cells lying on top of theimmediately surrounding pixels. The ON for each pixeloverlain by the convolution matrix is multiplied by thecorresponding weighting factor and the products aresummed. It is this sum of the local area operations that isused to compute the value to be used in place of the ONof the pixel beneath the centre of the convolution matrix.An output image is produced by the convolution ma~systematically being moved over and transforming theON of every pixel in the image.

The simplest convolution is that used to generate alow-pass filter. This functions by replacing the ON of each..-pixel with the average ON of the pixels surrounding it.Figure 5.14(a) shows a convolution matrix that could beused. In this case it is a 3 x 3 matrix, each cell has an iden-tical positive weighting and the sum of the weights addsto 1.D.This ensures that the overall statistics of the imageare not changed by the convolution. Figure 5.14(b-d)shows how a portion of an image is converted to a low-pass version. The dimensions of the convolution filter--sometimes called its kernel or window size-determinethe extent to which the image is smoothed. Put another

;>

,..

Digital Image Processing 135

26127132120 121 1191241

23127119125122118125

27 261

23 27or

19

~22117119125127130131

20

121 125123121 1191171

(a) (b)

26/25123122122

25 I 26 I 25 I 24 I 26 I[ 24+26+22 J=~X +28+30+27 =27

+28+32+29 2~..t-271 ~71271261

23 1 26128 1291271

,22 I 25 I 27 I 27 I 25 I

(c) (d)The

convolution matrix for a low-pass filter (a)~ ~ .,

way, the larger the filter the lower the frequency ofspatial variations tn ON that are expressed in the result-ing image. Figure 5.15 shows the results of two low-passconvolutiol) filters with different dimensic;>ns comparedwith the original image.

The simplest means of high-pass filtering is accom-plished by subtracting a low-pass filtered image fromthe original image. This shows all the deviations in ONfrom the local mean. It can be achieved directly by con-vol4tion. A simple high~pass convolution filter producingdifferences from the local mean is given in Fig. 5.16(a). Inthis case the sum of the weightings is zero. The deviations ~from the local mean may be either positive or negative.Moreover they most often have smaller values thanthe original ON. 50 that the high-pass filtered image canbe displayed, its ON must be rescaled to occupy the -0-255 range. As with low-pass filters, the size of a high-pass convolution matrix has an effect on the resultingimage. In this case, the smaller the matri~ the more finelyspaced are the differences tha! showup, and vice versa.An example 9f a high-pass filtered image of this kindis shown in Fig. 5.17(a). One of its functions is a ,meansof edge detection, Many other types of convolution filtercan be devised for edge detection, and some of them areshown in Fig. 5.16. They may be used incases where thesimple mean difference matrix does not give satisfactoryresults.

A ~ser generally i}opes to perform all interpretationson a single image, using both tone and colour variatiol)stogether with textural feature~Jepresented by edges.The image of detected edges in Fig.5.17(a) is noticeablylacking in contrast. Producing an imlige with both tonliland enhanced spatial information means adding theresults of edge detection to the original image. 5~ch edgeenhancement also can be achieved by a singleconvolu-tion with a suitably designed filter. Figure 5.16 shows

--This calculation is shown in (c).

.,All convolution works in this way and the results depend

.-, , .,. -, -_.

~

(a) (b)

Fig. 5.15 Image (a) is a linearly stretched image of Landsat TMband 5 data for the Sudan-Eritrea area used in Fig. 5.4 andPlate 5.1. Images resulting from 3 x 3 and 13 x 13.low-pass

(c)

convolution are shown in (b) and (c). As the dimensionsof a low-pass filter increase the smoothing effect becomesmore

obvious.

~136 Chapter 5

19

Ix2

1-x2

(a) (b) (c)

1

912"

lx2

(d) (e) (f)Fig. 5.16 Convolution matrices for edge detection or high-pass--filtering can be achiev~..!>y subtracting the results of a low-pass convolutiop from the original image (a), or by devisingmatrices based on more complex algorithms (b and c). Edge-enhancement matrices result from adding the edge detector tothe original image (d, e and f). "'",

longer its computation. takes. Figure 5.18 is an exampleof the use of a large filter to detect regional fault patterns.

As edges and other spatial boundaries in an imageare expressed by changes in ON over a small dis-tance, they can be detected in images of the gradientof ON. Gradient is defined as the magnitude of thefirst derivative and this measure also can be calculatedby convolution techniques. Changes in gradient arerelated to edges too, but in a more complex way. They~re displayed by the rate of change of the gradient,or the second derivative. Derivative techniques usuallyproduce images of remotely sensed data that bear littleresemblance to the landscape in a scene, and are diffi-cult to use as a result. Such images of remotely senseddata are not illustrated here. In geophysics, however, itcan be very useful to render aeromagnetic data as images(Chapter 8) and then convert these into images of thevariation in magnetic gradient, or gradient of the gradient(the second derivative) over an area.

So far, all the convolution matrices used have involvedsymmetrical distributions of weightings. Sometimes asym-metrical weightings are required for specific applications.The most important of these are for enhancing spatialfeatures arranged in different directions. Figure 5.19shows two asymmetrical filter designs that produceapproximately the value of the first derivative in twodifferent directions. Using 3 x 3 matrices it is possible toachieve directional filtering to the eight principal pointsof the compass. Figure 5.20 shows an example of direc-tional filtering to enhance edges running north to southacross an image. In fact edges between 22.50 either side

three edge-enhancement matrices corresponding to thethree edge-detection matrices. In each case the weightingof the central cell of the matrix is increased by 1, therebyadding the ON of each pixel automatically to the convolu-tion. The sum of the weights in each edge-enhancementmatrix is thus 1. Figure 5.17(b) is an edge-enhancedversion of Fig. 5.15(a) produced by this method.

A high-pass spatial filter normally enhances featuresthat are less than half the dimensions of the convolutionmatrix used. In the case of geological features of regionaldimensions, such as major faults and intrusive contacts,very large matrices must be used. The larger a filter, the

(a) (b)

Fig. 5.17 Image (a) is the result of a simple edge-deiectionconvolution (Fig. 5.16a) for the data displayed in Fig. 5.15(a). This wasadded to the original image to produce the edge-enhanced version shown in (b).

(b)The contrast-stretched image of Landsat MSS band 7

.,- .,discernible on an image of m!dium-scale edges detected by a31 x 31 mean-difference matrix (b). Courtesy of Pat Chavez,US Geological Survey. '

(a) (b)5.19 Directional filters based on square convolution

-.northward direction, matrix (b) in a south-westward direction.

Fig. 5.20 The image of Jordan shown in Fig. 5.18 (a) has beensubjected to an eastward directjonal first derivative filter. Theresulting image has been smoothed by a 3 x 3 low-pass filter andshows many faults more clearly than Fig. 5.18 (6). However, thisenhancement of linear features is at the expense of tonal variationsin the original image, which reJate to different kinds of surfacematerial. Courtesy of Pat Chavez, us Geological Survey.

of1h~ chosen direction are enhanced preferentially.Figure 5.20 also demonstrates the suppression of broadtonal variations and the rendering of edges as if theywere shaded topowaphic features. In this case this helpsovercome the directional bias in satellite images as aresult of illumination by the Sun. The pseudoshadow-ing effect of directional filters is extremely useful inenhancement of nonimage data which lack the clues todepth that actual shadows give (Chapter 8).

A danger in convolution filtering is that the filter-weighting designs are approximations only to those thatwould be feasible in frequency-domain filtering. More-over, the ffiters are restricted to rectangular matrices.

138 Chapter 5"

Fig. 5.21 Directional filtering by a one-dimensional mean-difference matrix-l pixel wide and 27 deep in this case-doeshighlight N-S edges in this image (see Fig. 5.17b for comparisOn).However, it has produced regular-artefacts in two directions,roughly north-north-eastand north-north-west, which couldeasily be misidentified as geologically significant features.

These deviatio:ns frOI\1 the idea~ result in spurious featuresor artefacts appearing in some filtered images. Figure 5.17contains examples of a common type of artefact, knownas ringing. In the raw image (Fig. 5.15a) a dark line cross-ing pale groun9 at the bottom le!t is an igneous dyke. OnFig. 5.17(a) this is represented by a dark line with brightfringes on either side, ~s are other linear features. This isringing, which in some cases can help detect small edges,but can introduce imaginary features to the interpreter.An even more disturbing type of artefact results from thegeometry of the convolution: matrix, which can intro-d~ce totally new geometric features (Fig. 5.21).

Whereas using square filters avoids directional arte-fa<;"ts, riI}ging is inherent in those square filters in whichijll the cells have weights, the larger the filter the moregross the rin~g,so that in some cases data are repeatedas 'ghosts'. For small filters ringing can help accentuatesmall edges, but if the topography is dissected by manysmall features, ringing merely disrupts the patterns. Usingdesigns with symmetrical crosses defined by zeroes andfinite weights, as in Fig. 5.16(b), helps suppress ringingwhile enhancing edges.

curves of natural mat~ials are relatively smooth in theOA-2.5I.lm range. In most cases a high reflectance inone waveband is matched by similarly high reflectancesin the others; which is due to the result of the overallalbedo of the surface. In such cases the data are said to behighly correlated. High correlation indicates that there isa high degree of redundancy ~mong the data-for manypurposes the information in one band is very much thesame as that in the others. This is illustrated graphic-ally in Fig. 5.10, which shows all six reflected bands ofLandsat TM data for a scene largely bare of vegetation.A very high proportion of inter band variability in sceneswith abundant vegetation is the result of the unique andvery broad spectral features associated with living plantcells (Section 1.3.3). High correlation between the threebands used to produce a false-colour composite resultsin little improvement over a single-band image. Thereis little colour variation. Even. where bands have beenselected to highlight the effects of known spectral features,redundancy can still be a major problem to the interpreterowing to the lack of colour contrast and the exploitationof only a limited region in RGB space. Very similar strongcorrelation between different wavebands can charac-terize any part of the EM spectrum where the bands areclose to one another. In the thermal infrared region thecorrelation is even stronger than in the reflected regionowing to the close approximation of many natural sur-faces to blackbodies. To some extent redundancy in colourimages can be overcome by manipulations such as the HIStransform (Section 5.2.2). 'There are, however, methodsthat use powerful statistical analysis to re-express theinformation contained in multispectral data accordingto its variability, so achieving a full description of theinformation in less variables. This is data reduction!Inherent in these methods is the possibility for inflatingthe data to fill more fully the multidimensional spacethat they occupy, thereby achieving decorrelation. Thissection explores some of the possibilities.

If ON from one band are plotted against those fromanother band near to it in the EM spectrum the major-ity of points lie on, or near to a diagonal line passingthrough the origin of the graph (Fig. 5.11a). The closer tothe diagonal the data are the greater their correlation.Poorly correlated data plot away from the diagonal.The result is an elliptical. distribution. Extending this tothree bands often reveals the data to be mainly containedwithin a cigar-shaped ellipsoid (Fig. 5.11b).!

A means of improving the spread of data is to redistri-bute them about another set of axes in multidimensionalspace, which maximizes the separation of differences inthe data (in this case the transform is in Cartesian space,rather than the spherical system of co-ordinates used inthe HIS transform). The procedure used most commonlyis principal component (PC) analysis, also known as the

arhunen-Loeve transform after its originators. Means

5.4 Data reduction

Apart from the effects of spectral features resulting fromvarious transitions (Section 1.3), the spectral reflectance

Digital Image Processing 139

(a) I (b) I (c)best-fitline = PC1 ~I

~ I ~ea~ zc

~I ""~Oft~

9 ()

u~.z;

.,'" PC2(0,0)~ mean band Xr I~

Band X ON Band X DN -ve 0 +ve

PC2

This axis becomes the first principal component on to whichnew values of the data, co!)1bining contributions from bothbands, are projected. The second axis expresses the variancethat cannot be expresSed by the first principal component.When these residual data are projected on to this axis theycomprise digital numbers (DN) for the second principalcomponent.

Fig. 5.22 Bivariate plots of data from any two bands in thevisible and near-infrared regions generallypr<;>duce anelongate ellipse of points (a) because of strong correlation.Principal component analysis begins by sh.ifting the originof the plot to the point defining the means (mx,.my) of the twosets of data (b). The axes are then rotated through an angle Osothat one is aligned with the maximum variance in the data (c)..and variances of data from two bands show graphicallyas scatter in a bivariate plot (Fig. 5.22a). The varianceof a single band expresses the spread of values aboutits mean. A measure of the joint variation of two vari-ables is their covariance. When covariance is positive thedata are positively correlated, as indicated by the crudeslope towards the origin of the cloud in Fig. 5.22(a),and when it is negative an inverse relationship holds.When the covariance is zero, the two channels of dataare completely independent of each other; a bivariateplot would show a formless cloud. For multispectraldata (topologically speaking multidimensional) the fullstatistical relationship between all the axes is expressedas a covariance matrix. A matrix of correlation coeffici-ents expresses the degree of data interdependence inanother way.

The first step in PC transformation shifts the origin ofthe n-dimensional distribution to that point which rep-resents the means of all the bands. This sets the meansto zero-the procedure is easier to visualize in the two-dimensional case (Fig. 5.22b). This step allows rotationof the axes, until one coincides with the line along whichall the data show the greatest spread, along the prin-cipal axis of the ellipse of data (Fig. 5.22c). This new,rotated axis is the first PC. Values associated with PClare projections of the 'raw' data points on to it at right

angles. An axis perpendicular tolhis now defines the lineexpressing all the remaining variation in two-dimensions;again by orthogonal projection to the axis from the datapoints. This is the second PC. In space with more than twodimensions this operation continues to define orthogonalaxes that progressively consume all the variation thatdoes not project on to lower,.order PCs It results in asmany PCs as there were original data axes. Th~ mathe-matics to achieve this transformation uses an expressionof the intervariability of the constituent 'raw' data bands,either the covariance or the correlation matrix. The resultsfrom the two alternative '<!riving' statistics are slightlydifferent, but the reasons need not be explained here(for simplicity the covariance matrix is used). In short,PC analysis is a little like finding the axes of a deformedpebble and measuring them for strain analysis, one bigdifference being that it deals with hyperspace.

Projection of the original n bands of data on to nnew PCs uses eigenvectors to produce linear, additivecombinations of the original data for each PC. Eacheigenvector is derived from the 'driving' statistics, andis a weighting or loading factor for the additive contribu-tion of each band to a PC that can be positive or negative,depending on the particular geometry of the data dis-tribution. Table 5.1 includes a matrix of eigenvectors usedto project ON from each band in the Landsat TM scene in

Table 5.1 Eigenvectors for eachprincipal component of the LandsatTM data shown in Fig. 5.23 indicatethe relative loadings of the differentbands in each component. Theeigenvalues indicate the relativeproportion of the overall scenevariance encompassed by eachcomponent

Eigenvectors

Eigenvalues Band 1 Band 2 Band3 Band 4 Band 5 Band 7

PClPC2PC3PC4PC5PC6

2242284

181161

0.27

0.58

-0.48

-0.44

0.31-0.27

00

-0

0-0

0

0.370.400.11,0.37

-0.670.32

0.310.270.67Q230.570.00

0.68-0.48

0.18

-0.50

-0.16

0.00

0.42-0.32

-0.52

0.600.290.00

.21

.32.11.01.14.91

140 Chapter 5

with variations in solar jllumination that result fromtopography. In the case of multispectral thermal data, thefirst PC is dominated by surface temperature variations.Atmospheric effects reside in the first componentbeca1,lSethey are always highly correlated. The higher ordercomponents express deviations of various kinds fromthe average in PC1, and contain information relating tospectral variations in the scene, which stem from differ-ent types of surface. Because noise in the data generallyis the least correlated between bands, it ends up in thehigher order components. Figure 5.23 shows the six PC~derived from the data shown in Fig. 5.10. The decreasein variance and increase in noisiness as the order of thecomponents increases is obvious. But what do these newimages show, after such a complicated process?

Making sense of PC images involves examining theeigenvectors, or r~lative loadings of each input band toeach component. As Table 5.1 shows, the loadings are allpositive and high for the first component, indicating thatit expresses the average of all the bands quite closely.

Fig. 5.10 on to each Pc. The valu~ assigned to a pixel's P<;:is simply the addition of the products of multiplyingthe DN for each band by its. associated eigenvector (thecolumns in Table 5.1). As well as tral)sposing the axesthrough the n bands of data, PC analysis transforms thecovariance matrix too, so that the covariances betweenthe resulting PCs are set to zero. In other words thePCs are decorrelated. The variation within the 'raw' datais recast as variances on the PCs, which are known aseigenvalues. The highest eigenvalue is associated withthe first PC, and progressively lower eigenvalues witheach higher order PC. Figure 5.22(c) sbow that the valuesassigned to a PC can be positive or negative, because ofthe initial repositioning of the origin in the distribution.For them to be ~isplayed as an image means expressingthe range of values in the familiar 0-255 range using alook-~p tableQt some kind. ..

The first PC of highly correlated data is generally aweighted average of all the data. In the case of reflecteddata, this approximates an image of albedo combined

(a) (b) (c)

(d) (e)

Fig; 5.23 Images (a) to (f) represent data in the six principalcomponents that are possible for Landsat TM reflected dataThey cover the same area as Fig. 5.1 O.The high quality ofthe image of the first principal component results from theexpUlsion of noise to higher order components, an,d from

(f)

combination of highly correlated contributions from allsix TM bands. The bulk of noise resides in the sixth PC (f).The other PCs are additions of each band weighted by therespective eigenvectors as explained in the text. A false-colourimage derived from these datajs shown in Plate 5.4.

Digital Image Processing 141

of the eigenvector matr~, for no two scenes are alikeand the covariance matrices for each produce uniquesets of components and associated eigenvectors. So,images of different terrains and images of the same sceneacquired on different dates will not show similar geolog-ical materials in the same way. Rather than attemptingto identify the reasons for the appearance of specificdivisions of the surface that are not seen in conventional" images, it is better to use colour PC images to map those

divisions and the structures that they reveal arbitrarily,and then seek reasons for their separation in the field.There are other strategies though.

I@°irected

principal component analysis (PCA)

Figure 5.3 shows how different materials express them-selves in bivariate piots, in this case for a plot of VNIRagainst red radiance for an en~e Landsat MSS scene. Thescatter is dominated by an~'e showing high positive.correlation that correspondsto'bare soil and rock. Watershows itS characteristic low VNIR and significantly higherred radiances and forms a separate cluster towards thebottom. Vegetation wit~ strong VNIR reflectance andred absorption forms the negatively correlated 'lump' onthe main cloud. Similarly, ~o_n- and clay-rich surfaceswould form identifiable clusters on plots, say, Landsat,TM bands 3 and 1, and 7 and 5, respectively. In thecase of the VNIR-red distributiontn Fig. 5.3, the first PCaxis clearly would not follow the axis of the main highlycorrelated ellipse, and its position in n-dimensions alsowould be significantly shifted in areas with a high pro-portion of vegetation. The outcome of this is that in suchscenes PCA becomes dominated by the vegetation effect,and increasingly suppresses the separation of geologic-ally interesting surfaces, depending on the proportion ofvegetation cover.

One means of using PCA to accentuate geologicallycontrolled spectral variation is deliberately to omit denselyv.egetated parts of a scene from the statistics that 'drive'the process. An arbitrary 'rul1" is applied, as metaphor-ically in ecclesiastic law! So this approach is known ascanonical analysis. In practice the operator outlines alarge area with the lowest vegetation ~ver, generallyusing a VNIR, red and green rendition in an RGB image.This training set then becomes the source for thecovariance matrix, and the PCA begins with the PCI axisforced to follow the principal axis of the soil and~ockpixels. Two things should happen: variance as a resultof vegetation appears strongly and almost uniquely inone PC-often the second; variance as a result of rockand soil spectra becomes more accentuated in the otherhigh-order PCs, and related more easily to spectral prop-erties of important minerals. Theoretically, this shouldproduce more useful results than plain, vanilla PCA.In practice it suffers from one main drawback; the

However, for the other components there is a mixture ofpositive and negative, and low and high eigenvectors.Understanding these numeric clues in the eigenvectormatrix means returning to specific features in the spectra(Chapter 1). Considering vegetation first, this has a uni-quely high reflectance in Landsat TM band 4 and a lowone in band 3, relative to soils and rocks (Fig. 1.15). Acomponent that has a high eigenvector for band 4 and ahigh value with the opposite sign for band 3 (e.g. PC5 inTable 5.1) is clearly expressing variations in vegetationdensity. Figure 5.23(e) shows dark features along drain-ages that relate to vegetation, which are where expectedfrom its red appearance on the 432 image in Plate 5.1 (b),but which occupy a larger area than suspected simplyfrom a conventional false-colour image. (Note: whethermaterials associated spectrally with this or that PC showprominently as dark or light relative to their s~rrourid-ings is impossible to predict.) Clay minerals' high reflec-tance in band 5 and strong absorption in band 7 (Fig. 1.9)mean similar high loadings in a co~onent that containsinformation about clay variability (e.g. PC4 in Table 5.1).Figure 5.23(d) shows dark areas, but to know if theprediction is verified would mean field checks in thiscase (in fact most of them are clay-rich alluvium). Muchthe same argument for the spectral charffcters of ferriciron minerals in the visible and VNlR (Fig. 1.7) canhelp identify a component- that expresses iron variabil-ity. Unfortunately thingS" are rarely that simple, andthe eigenvectors generally are a guide only to under-standing PC images. There is, however, a ruse to forcePC analysis to enhance specific materials selectively(Section5.4.1).

As the effect of PC analysis is to compress the varianceof the original data into the lower order components, it isa means of data reduction. This is rarely useful for datawith few, broad bands, such as Landsat TM. Importantinformation for small, spectrally unusual parts of a scenemay lurk in the higher order components from such data,so it is important to check all components carefully. A PCapproach is much more useful as a means of data reduc-tion in analysing the huge volumes of hyperspectral datafrom imaging spectrometers (Section 5.6.2). With sucha wealth of data there are many possible combinations toexamine, and confusion is inevitable using simple band

techniques.Principal components may be displayed in any com-

bination as red, green and blue to produce a false colourimage. Instead of the frequently limited range of coloursresulting from normal band composites, a vivid arrayis common (Plate 5.4). This effect of the decorrelationprocess can be extremely useful in separating subtlydifferent categories of surface. However, the coloursbear little relationship to the true composition and EMproperties of the surface. To explain them would meanconsiderably more than grappling with the intricacies

..

Chapter 5

~gative) loadings fo.bands 1 and 3 in the eigenvectorix gives a measure of 'redness' that is related toized

iron minerals. Similarly, an OH/C-O PC will~ high loadings for bands 5 and 7 in the appropri-igenvector matrix. Forming a composite channel, byng

the two PCs together, gives a third channel to use:\ RGB image. Such images have limited ~ buthis

simplification they clearly outline areas rich in., OH/C-O or both mineral types. The last is of con-rable

importance, for many zones of hydrothermal'ation show both in abundance, and often haveciated

metal mineralization.>tentially, as explained earlier in this section, this~ of directed PCA should permit the discrimination~veral more mineral groups. The grouping of TMis

2, 3, 4 and 5 should distinguish haematite fromhite, because of the significantly higher reflectance Ile

latter in the green regio~ Using TM bands 1, 3, 5 17 should give some information about the Mg-OH I !C-Obond distortions as well as that due to AI-OH.~ful selection of high-order PCs from the two groupsl reference to the respective eigenvector matrices;lisplay as RGB images, and some knowledge of theogy covered by a scene, potentially results in highlyured

images and discrimination of subtle features.rith the failure of the Lewis platform with its hyper-:tral sensor, and the indecision regarding the deploy-It of HIRIS in the EOS series, Landsat TM and ASTERl will remain the main source of geologically orient-I image data for some time. In the last case, directed

~ that extends into the multispectral thermal bandsr well present a quantum leap in routine geologicalge

interpretation. However, hyperspectral data willritably colour the future of geological remote sensing:\

orbit, and Section 5.6.2 focuses on that, including'e on directed PCA.

~ Decorrelation stretching

re 5.24 shows in two-band form the principle of.rrelation or D-stretching. Figure 5.24a shows twoly correlated bands relative to original band axes andew PC axes. Figure 5.11 and Section 5.2.2 explainthere are limits to the exploitation of RGB colour

:e by simple contrast stretching. The rotation of axeslIved in PC analysis enables the cloud of data to be:ched in two directions at right angles, instead of just

g the major axis of the original elliptical distribu-The space defined by the new axes can be filled moreiently (Fig. 5.24b,c). Simply by rotating the axes backleir original position in band space (Fig. 5.24d) results[\

extreme, and otherwise impossible, 'inflation' ofdata. Moreover, this procedure retains the relative

tion of the pixels in the original distribution. Thisns, in the three-dimensional case used in RGB images,

ults

can never be repeated for other scenes, even those cluired on the same date-the adjacent rows along a rtform'

s pat~. In addition, it becomes tempting to refin~ carea for the statistics, or to try other ar~as where, 1

'haps, more is known of the field geology. Time soon aises by, and with it an increasing and a confusing mass a

different PCs results. Where geological knowledge ilacking, canonical analysis can be hit or miss, but 1:~s prove useful where some areas are known to show i~resting

mineral associations, as in the case of hydro- srmal alteration. Using them to define a training Se.1, ,mId

result in other, unknown areas with similar prop- ,les to show up in the resulting PCs. The other waydirecting

PCA is to focus on general knowledge of t~ctral properties, without making arbitrary decisions ()ut whether this or that part of a scene is most rep- 1entative

of geology. ..~~xamining the spectra in Chapter 1 shows that in some (

~ feature characterizes a particular material. Vegetation ,I case in point, and can be defined from the relative,liance

in VNIR and red bands. A ratio between the two (Iuite sufficient to highlight vegetation (Section 5.5). ,

e same could be said for ferric oxides and hydroxides, jthey are uniquely red-tinged and ought to show in !

iti9 image involving red and blue radiance. There are (ferences between the oxides and hydroxides, how-

~r, the latter being orange-brown and the former more ![dish. Besides that, both have absorption features in ]!

VNIR. If data for these regions are available, as they (! in the cases of Landsat TM and ASTER, using all ,

them would be much more effective, particularly if ]ta more suitable for other materials were excluded. ]lch the same applies to minerals containing AI-Of! j~-OH

or C-O bonds that produce absorptions in the j-2.4~ range (Figs 1.9 and 1.10). Using Landsat TM jIld 7 and 5 data alone would lump the~allin a single ]egory because the breadth of band 7 covers all the,tures.

However, clays (Al:-OH) also show a higher I1 than blue radiance (th~y are 'cream' coloured). .rbonates

are spectrally flat in the visible region (theyilly are 'white'), whereas Mg-OH minerals, such as jlorite

are often greenish. So data from both the visible ]d SWIR regions has the potential for at least somebtle

discrimination, when broad-band Landsat TM is~only sensor available.The

general strategy adopted by most researchers~eFurther Reading), who have ventured into this par-ular 'haunted wing' of PCA, is based on acceptingit the only clear discrimination possible with Landsat,f data is between ferric oxides/hydroxides and OH-dC-a-bearing minerals. They group the available TMnds

into .1, 3, 4 and 5 (for ferric minerals), and I, 4,5.d 7 (for OH- and C-O-bearing minerals); Principal

mponent analysis, of these groups produces four PCs "'th associated eigenvectors. That PC with high (positive

Digital Image Processing 143

(a) (b) transform is one method. Because the highest PC rarelycontains any meaningful information, simply omitting itfrom the inverse transform can achieve a dramatic re-duction in the speckle and striping that result from sen-sor malfunctions, with little loss of spectral information.

5.5 Band ratioing

Band X PC2

(c) (d)

./

~l~I

//.J "1 r::; jrl'ii

<I: , ,-J-}J'"

if'Band X

Fig. 5.24 (a) is a bivariate plot of two bands with principalcomponent (PC) axes superimposed. In (b) the first PC hasbeen stretched after rotation of the axes to PC space, the secondPC having been stretched in (c). This shows how it is possibleto produce a decorrelation of the data in PC space. (d) showsthe rotation of the decorrelated data back to original bandspace, the original correlation having been drastically reduced,although each pixel is still in its original position relative to allthe others. Compare with Fig. 5.11.

Just as the data in an image can be displayed as grey-toneimages of single bands, or as three-band colour images,so can various arithmetic combinations of bands. Themost useful qf these is the ratio of one band to another.It is prepared simply by dividing the DN of each pixelin one band by that for the corresponding pixel inanother band. Theoretically this should produce a rangeof new values for the pixel from zero to infinity/becausesome DN Will have zero values. Because of the restrictedrange in each band of data and the strong correlationbetween bands, in practice ratios seldom fall outside therange 0.0-4.0. Displaying an image of the ratio requiresrescaling these floating-point values to the 0-255 rangeof integers.

Random noise in different bands generally is uncor-related, so ratios for pixels with spurious DN in eitherthe numerator or denominator band produce extremevalues. This enhances noise and most ratio images sufferfrom strong speckling and striping. They can be 'cleaneduP/ using the digital processes described in Appendix B.The contribution from atmospheric effects (Section 1.3.1)varies With wavelength, which means that ratioing alsotends to accentuate its degrading effects. So/ before ratiosare calculated the bands involved should be correctedfor atmospheric scattering (Section 5.2).

Individual bands of remotely sensed images showthe effect of varying illumination caused by topography.The Sun illuminates a flat surface homogeneously. Wherethere is any appreciable relief, the lighting of slopes thatface the Sun is stronger than for horizontal surfaces, andslopes facing away from the Sun receive less radiation.As a result, a surface With uniform reflectance propertieswill show varying DN across a scene. In a scene Withmany different surface types this can lead to addedconfusion in both detailed visual interpretation and incomputer-assisted classification (Section 5.6.1) that usebands alone.

Theoretically any surface should receive the same pro-portions of energy in all wavebands irrespective of itsorientation to the Sun. It also should reflect energy inthe proportions controlled by its spectral reflectance pro-perties. Therefore the ratio between two bands for pixelscontaining the same kind of surface should be the sametirrespective of the slope direction. For this to be true thesurface must reflect radiation equally in all directions.Under natural conditions this behaviour is uncommon

that a colour image can be stretched by this means togive nearly all conceivable colours, without altering thebasic hues of each typ~ of surface in the scene. This isachieved by stretching the PCs to the full 0-255 range andthen reversing the transform. Complex as it may seem, inpractice this is merely adding together the products ofmultiplying all the stretched PCs by their associatedeigenvectors-the rows in Table 5.1-for each band toproduce the new D-stretched bands. As an example, band1 is reconstructed by adding the products of each PC andits associated eigenvector in the top row. Plate 5.5 showsa D-stretched version of Plate 5.1 (c) for comparison withPlate 5.3. The effect is quite similar. Some of the uniquepossibilities presented by this technique, however, havebeen used to sharpen the image and to reduce the noisethat is inevitable in extreme stretches.

As in the manipulation of intensity in the HISstretch, using edge enhancement sharpens the first PC(Section 5.3) and careful contrast stretching removesall atmospheric haze. Moreover, as noise increasinglydominates the higher order PCs, especially the last one(Fig. 5.23£), noise reduction is possible. Using medianfilters (Appendix B) on noisy PCs before the inverse

144 Chapter 5

(a) (b)

Fig. 5.25 An image of enhanced Landsat MSS band 7 data foran area in South India (a) shows patches of strong reUectancethat could be either bare sOil or healthy vegetation. Displayingan image of the band-7 : band-S ratio (b) exploits the spectralproperties of vegetation to highlight vegetated areas and to

give some indication of the proportions of plant cover to baresoil. Pixels devoid of vegetation ave the lowest value of theratio and appear dark, whereas the brightest pixels have densegreen vegetation cover.

,

accentuates features in the spectral signature curve of aparticular surface material. If the bands used cover peaks,absorption troughs or changes in slope of the spectralcurve, then, combined in pairs as ratios they express wellaspects of a material's spectral signature. The simplestexample relates to vegetation. The characteristic featuresof plant spectra are the chlorophyll absorption featurein red and the strong reflection of infrared by living cellsin the 0.7-1.2 ~m VNIR range (Section 1.3.3). In imagesof red and VNIR radiance, pixels that contain a high pro-portion of living plants show as relatively dark and light,respectively. Inastandard false colour composite imagethey will stand out from bare soil or rock as strongly red-coloured areas. Within these fairly obvious areas may beall sorts of variations. The proportion of a single speciesrelative to unvegetated soil may change from place toplace. Soil moisture or other environmental variationsmay affect the health and spectral properties of the spe-cies. The type of vegetation may vary, and there may beall manner of combinations of these important factors.Images expressing only data from various wavebandsoften conceal these differences, simply because the humaneye has a limited range of perception. However, surfacevariations do change the striIcture of the spectral reflect-ance curve. In particular, the depth of the chlorophyllabsorption feature, the peak height of VNIR reflectanceand the gradient of the curve between red and infraredWill change. QUite small changes in these parameters havea disproportionately large effect on the ratio of VNIR tored radiance (Fig. 5.25). The changes become easier toevaluate on the ratio image. Vegetation studies use ratioimages of different kinds to estimate the proportion ofgreen leaf cover and to help in discriminating betweenand classifying different plant species.

and reflection reaches a maximum in a direction con-trolled by the structure of the surface and the angle ofillumination. Another disruptive factor is the effect ofillumination by diffuse radiation from the atmosphere;so-called 'skylight'. Radiation from this source, whichis responsible for the illumination in deep shadow, hasa variable contribution to the whole Scene. Its spectralrange is different from direct solar radiation, becauseit stems from wavelength-dependent scattering. Con-sequently, it changes the ratios on differently orientatedslopes in a subtle fashion. Nevertheless, a ratio image doesreduce the effects of slope and shadows to a markeddegree (Plate 5.6a), and atmospheric correction beforeratio generation reduces the effect considerably. Verymuch the same argt!ments hold for surfaces that vary intheir albedo, but have similar spectral properties. Theywill appear very differently in images that use bandsalone, but are expressed in the same way in a ratio image(Fig. 5.25).

Irritatingly, in these advantages lie serious disadvant-ages of ratioing. In its 'ironfug-out' of topographic effectsand the suppression of differences in albedo, ratioinghides important information. However, image process-ing is one endeavour in which it is possible to get a freelunch, or maybe even two. Using the HIS transform(Section 5.2.2) to convert the colour in a three-ratio imageinto saturation and hue, the 'lack' of spatially importantinformation and the wealth 'of inevitable noise ends upin the intensity component. One ruse to restore spatiallyimportant information replaces the intensity from the HIStransformation of three ratios with a suitably stretchedand edge-enhanced band. Plate 5.6b shows an exampleof this technique.

The most important property of a ratio image is that it

tal

Image Processing 145

rt communiti~s ar~ sometimes excellent indicators;eology in humid terrains. In some cases, specific;s

provide favourabie chemical conditions for growthestricted plant communities, such as calcium- and;nesium-loving

species, and even species that willr grow over rocks rich in transition metals. Sometimes,ubstrate

threatens the health of plants. This may beto the result of anomalously high concentrations ofc elements, as might. be the casein the vicinity of cer-mineral deposits. More commonly, however, such.sed

vegetation is merely a response to either water-;ing or a shortage of water, in which case it generallyseas<:>nal phenomenon. Different plants have dif- .t

nt tolerances to environmental stress, so anotherogically related feature of vegetation is the inter-ation of species that are favoured or stressed by parti-r geological conditions. Geobotanical indicators can!xamined

most successfully using ratio techniques~doQcareful reference to field reflectance spectra.gure

5.27 shows how the reflectance spectrum ofant changes with the moisture content. of its leaves.progressive deepening of the absorption features atlt 1.45 and 1.9 11m as leaf moisture increases is veryr.

Although atmospheric effects mask this portion of;pectrum, the related changes in the regions ~roundLm

relative to the shorter infrared plateau, and in theLm region, show that ratios can be devised to high-t variable moisture stress. With Landsat TM datai/5

and 5/7 ratios should increase as l~af-moistureent increases. Geochemical stress, either through'tages

of essential nutrients or the presence of toxicpounds, sometimes results in the near absence of!tation over the offending rocks and soils (Plate 5.7).e

common is the stunting and discoloration of plantsving on them. Disruption of the pigmentation in

,ts produces shifts in the position of the chlorophyllIrption features associated with the red and blues of the spectrum. It also may affect the shape of the

trum near to these features. These signs of geo<;:hem-,tress are often most discernible at particular stages

lant growth, commonly just before the onset of leaf,scence (Fig. 1.17). At present the wavebands most

ily available from orbital systems are too broad tooit these changes. However, the shifts and distor-; are definable by ratios between very narrow b,ands

rperspectral data.

1

Ratios and mineral mappingI.

crude sense, ratios between qands describe the:tral'colour' of an object, although colour perceivedllumans

corresponds only to the visible range. Anlortant geological example does relate to percept-colour. Character.istically, the Fe-() charge-transferLSition (Section 1.3.2) is a broad absorption feature~avelengths

less than 0.55 11m. It is responsible forstrong red coloration of rocks rich in iron oxideshydroxides. Sometimes this coloration is maskednixing

of iron minerals with large amounts of othererals, which reflect strongly at all wavelengths, suchuartz.

The albedo of such a mixture will be so highit appears white in a natural- or false-colour image.

'lever, the ratio of red t<;> blue reflectance will enhancesmall contribution of iron minerals, giving pixelsron-bearing

rocks a higher value than those com-~d of pure quartz. In Landsat TM data this measure:edness'

corresponds to the ratio between bands 31. Moreover, the position of the Fe-() feature is

~rent for hematite (FezO3) and goethite (Fe(OH)3)'latite being more absorbent in green than is goethite,

ch results in the former being cherry red and the!r an orange brown. Potentially the red/green ratiodiscriminate between the two minerals; goethite ismost common weathering product of pyrite (FeSz)meral deposits. Iron also produces an absorptioni between 0.85 and 0.92 11m, owing to an electronicsition (Section 1.3.2). This feature falls within the;e

of Landsat TM band 4, whereas a reflectance highill minerals is found in band 5. The ratio of band 5and

4 therefore shows higher values for oxidized-rich rocks than for other types. The AI-OH and-OH rotational transitions associated with clays and~r hydroxylated minerals result in absorptions withinisat TM band 7. Dividing this band by band 5 resultsay-rich rocks showing up as dark areas.sing

Landsat TM data these four spectral features1 the practical limit to surface discrimination using

) techniques, even though there are 30 possible ratiobinations of the six reflected bands. Moreover, the~ band widths mean that only gross differences areIrable.

Many other features can be investigated usingower bands within the same reflected range, as from'-I,

ASTER and hyperspectral devices (Section 5.6.2).ltroduce the potential of the technique, Fig. 5.26 shows'atios

of Landsat TM 5/4,3/1 and 7/5, together with,gle band to show spatial features of the area of study.

~ 5.6 shows them in this order as an RGB image.Pattern recognition

~arch in military and various scientific fields sinceIt 1970 has developed several mathematical meanslutomatically extracting information £rom digital~s.

The techniques fall under the general heading~tificial intelligence. They were devised originally to

:

Ratios and geobotanyts

depend on the underlying soils and rocks for, supply of water and nutrients. In their natural state

(a)

(c) (d)Fig. 5.26 Images (a) to (c) show the TM ratios of bands 5/4,311 and 715 compared with a band 3 image (d) for the area on theSudan-Eritrea border used in Fig. 5.4 and Plate 5.1. See also Plate 5.6.

Fig. 5.27 (left) The leaves of Indian com have reflectancespectra that show considerable changes with increasingmoisture content. Similar curves have been obtained fromvarious plant species.

Digital Image Processing 147

geologists to 'train' comp1Jters in making some geologicaldecisions in a less biased way..

5.6.1 Multispectral classification

Automatically discriminating surface materials on digitalimages using computer software, or classification, isbased on statistically separable differences betweentheir radiances in the available barid§ (Fig. 5.28). Suchdifferences stem from the materials' spectral properties(Fig. 5.29), which hyperspectral images record almostfaithfully, and which are the most reliable means of quan-titative discrimination (Section 5.6.2). Hyperspectialdata are, however, a far cry from the most cheaply andeasily available multispectral images that use broad bands.In them the subtle features of a material's spectrum aredegraded to average ON for widely spaced portionsof the full spectral range, even though the wavebandsaim at highlighting the main spectral features of specificmaterials.

Inspection of Figs 5.28 and 5.29 suggests that acomputer could perhaps achieve a good discriminationbetween some of the different surface types using datafrom one band alone. For instance, water has very lowON in band 10 compared with vegetation and all rocks.Although there are also differences among the rocktypes, and between them and vegetation in all thebands, there is no guarantee that there is not consider-able overlap. The bars and spectral curves relate only tospecific localities within a scene. Over the whole scenethe reflectance properties of different surface categoriesvary within limits determined by natural variations in thematerials themselves and mixtures between materials.Figure 5.30 shows that the overall histogram for anyoneband is built up from individual ones that represent dif-ferent materials. Many of these 'sub histograms' overlap.

aid in the recognition of specific objects, such' as tanksor en~eering components, to speed up intelligencegathering and develop robotic vision. Such pattern re-cognition is focused on three main aspects. First is thedetection of different kinds of surface materials, such ascamouflage or colour-coded objects, which relies on thetonal or spectral characteristics of images. Most cheaplyavailable digital data have limited numbers of spectralbands, which generally are so broad that they span severalspectral features.. Nonetheless, different surface materialsdo show sufficient variability in their broad-band reflect-ance or emittance' for them to be distinguished visually onsuitably enhanced and transformed images, as describedin Sections 5.2, 5;4 and 5.5. The differences can be sys-tematized and exploited digitally in semiquantitativeclassification. If hyperspectral data are available, thecoverage of visible to SWIR wavelengths by ~any nar-row bands produces near facsimiles of spectra for eachpixel in the scene. This opens the possibility of spectralpattern recognition and mapping variations in surfacecover quantitatively (Section 5.6.2). Of equal importanceis the discrimination of distinctive shapes, such as war-ships or tanks, from natural spatial patterns. Third isthe recognition of changes in an area with time, as in thecase of the progressive build-up of mijitary forces or theredeployment of missile silos. Although most of the soft-ware involved stemmed from the military, medical andengineering communities, much of it is very useful inenvironmental science applications.

As geological features are so variable and because theyare often masked by soils and vegetation, visual inter-pretation, even by an experienced person, is often ambi-guous. This is made even more of a problem by the sheernumber of theories that a geologist can use in assessingthe context of features in an image. Artificial intelligenceand pattern recognition open up the possibility for

(a) (b) (c) (e)

z0

Fig. 5.28 Spectral reflectancecurves can be represented crudelyby bar charts corresponding todigital numbers (DN) in therestricted number of wavebandscovered by most imaging systems.In this case (a) represents water,(b) vegetation, (c) hydrothermallyaltered granite, (d) shale and (e)basalt. The waveband numbersshown are those in a multispectralline-scanner.

I I3 6101213

water

OL36101213

alteredgranite.

3 6101213

shale

3 6101213

basalt

148 Chapter 5

scanner.-~~-

--

.'

~Q)uCco

tQ)

'+=Q)~

-_/,-

1

---; , /,?::=_:

~;;/"

3 ~.4-

/ ~\

.~-=::-:=:-

"5

\ --/- --.9

"'8'. 7

0.4 1.2 1.6 2.0 2.4

Wavelength (IJm)Fig. 5.29 Superimposed on the reflectance spectra of a number of surface materials from Utah,. USA are the wavebands usedin Fig. 5.28. From Podwysocki et at. (1983).

0.8

0 50 100 150DN

Fig. 5.30 The histogram of one band for a whole scene concealswithin itsmaller histograms for particular classes of surface. Inthis case (a) to (d) are hypothetical class~s.

which almost inevitably is weathered to some extent. Rock-forming minerals are replaced by their altered products,at least to some degree. Imagine a granite formed fromfeldspar, quartz and muscovite. Its weathered surfacewill be quartz with unweathered feldspar and mica, plusvarying proportions of clays formed by feldspar weath-ering, let alone the minute but highly coloured productsfrom conversion of iron compounds in the rock. Rocksrich in ferromagnesian amphiboles, pyroxenes and olivineswill have surfaces full of clays, serpentines, chlorites andoxidized iron minerals. Although rocks are divisible intothe familiar classes using a hand lens or microscope, evenif they are weathered a little, petrography is impossiblefrom more than a few metres away!

The second difficulty is this. Except in the most aridterrains with little superficial cover, rock outcrops arereally quite rare, though often very prominent. They aremantled to various degrees by soil, transported sedimentsand vegetation (Fig. 5.31). Although vegetation and soilsmay be related to rock type, even uniquely in some cases,the geologist requires an outline of where different rocksare present beneath the veneer-that is what a geological.map sets out to show. Even in areas of complete exposure,classification faces problems. In deserts, thin surfaceveneers of manganese and iron oxides known as desertvarnish coat rock surfaces, which, irrespective of theircomposition, appear dark in visible and near-infraredimages. A sI:inilar problem i~ encountered with highlyreflective encrustations of salts form~d by evaporation.In some areas of diverse vegetation and soil cover the onlypossibility in classification is to select subclasses withineach lithological category, such as bare rock, derived soiland vegetated surfaces with different degrees of plant

Some are narrow and others are broad. Each may bedescribed by the mean DN for the category that it rep-resents and the variance, which expresses the spreadof DN within a category, and is considered shortly.The differences can be enhanced using ratio techniques(Section 5.5), w~ere the high VNIR reflectance relativeto that in red for vegetation distinguishes it uniquelyfrom other natUral materials. Things are never as simpleas they might seem, however. Pixels themselves form thelimit to discrimination, and frequently are so large thatthey represent mixtUres of exposed materials in smallpatches that comprise pixels on the ground (Fig. 3.10).Even with near-continuous spectra, a surface class islikely to be 'impure'. That is bad enough for categories ofvegetation, but worse for geological materials because oftwo conspiring factors. The first of these is that remotelysensed information, from whatever part of the electro-magnetic spectrum, emanates from the surface only,

channels

~

--2---,..

-6

Plate 1.1 The clear water of the Red Sea off Mersa Fatma inEritrea allows green light to penetrate to depths cj 20-30 m. Inimage (a), a Landsat TM false-colour image, the blue colours ofthe nearshore waters are to the result of reflection from coralreefs and sand in shallow water. Image (b) shows colour bands

assigned to narrow ranges of blue reflectance to produce arepresentation of the variation in bathymetry. Grey-tones areabove sea level, red is the shallowest water and the 'rainbow'progression towards dark blue then magenta is from shallowto deep water.

Plate 2.1 The huge range of colourshas been produced by combiningthe three additive primary coloursin the proportions shown as red,green and blue. Courtesy of AdobeSystems Inc.

[facing INIge 148]

(a) (b)

Plate 3J In this pair of vertical aerial photographs of a ruralscene in Devon, UK (a) is natural 'colour and (b) false-colourinfrared, Although the natural colour image is more pleasantto look at, because it is dominated by wavelengths near thepeak of human visual sensitivity, it contains less informationthan the false-colour infrared image, This is particularlystriking in the wooded area, where not only are the

conifers sharply distinguished from the sparse, bright redbroad-leaved species in the dark swath of trees, but manystands of different broad-leaved species, different stagesof growth and different canopy types can be separated too.On the natural colour image these features are by no meansobvious. Copyright Aerofilms Ltd.

1.93

2.01

2.10

2.18

2.27

2.35

E.3-.c...CIc:QI

""Qj>~5

2.44

350 300 250 200 150 100 50

Flightline pixel number

Plate 3.2 This figure consists of many spectra from an airborne along the flight lme of peaks and troughs in the surfacespectroradiometry campaign. The spectra are stacked for reflectance spectrum. These in turn may indicate precise,lyindividual ground resolution cells along a flight line from the abundance of minerals that have distinctive absorptionright to left. The vertical axis spans wavelengths from 1.93 to bands. The magenta patches between 2.1 and 2.35 ~ suggest2.44 ~m in 20 nm bands. For each narrow spectral band the abundant clay minerals (Fig. 1.9) at the surface fromspectrum for a single ground resolution cell is colour coded samples 70-80, 120-210 and 300-330 along the flight path.from red (high value) through a rainbow sequence to violet Courtesy of Stuart Marsh, Sun Oil, Houston.(low value). The colour patterns indicate the diStribution

~

i 4.3 This enhanced false-colour composite of Landsatdata shows an area in the central part of the Tibetan~au.

Although it is of marginal use in this form foriminating lithologies, the many linear features that it

esses are the key to understanding the region's activenics. Several of them clearly diplace Recent superficial

Isits, as at centre right and centre left. Scale 1 : 1 million.ce: courtesy of Dave Rothery, Open University.

~ 4.1 The Colorado River of Arizona is a classic example of:edent drainage. In this false-colour Landsat TM linage atIe

of about 1 : 3 million it has carved the spectacular Grand,on through the rising Colorado Plateau because its rate ofnward

erosion exceeded the rate of crustal uplift.

!

4.2 The curved dark bands on this enhanced Landsatfalse-colour composite of part of Andhra Pradesh inh India are outcrops of sediments that filled the huge

iapah Basin between 1.7and 1.3 Ga ago. The yellow areae south-west comprises 2.6 Gil Archaean granitic gneisses.~s cutting the Archaean show up very clearly as -

secting

dark lines, but stop at the boundary with the lateImbrian rocks. This boundary is clearly an unconformity.nt dates on the dykes show that they are about the samelS the oldest of the basinal sediments. Scale: 1 : 1 million.ce:

author, written to film by Michael Smallwood,onix Inc.

Plate 4.4 The western part of the Qaidam Basin in QinhaiProvince, China, shown on this LandsatMSS false-colourcomposite, is floored by folded lacustrine and othe. nonmarinesediments of Tertiary to Quaternary age. They occupy themainly grey texturally flat area. The basin is bounded to theNW by the sinistral Altun strike-slip fault, which forms aprominent linear valley. The rugged area in the north is anuplifted massif of older, Mesozoic rocks. Scale: 1 : 1 million.Source: author.

(a) (b)

I (d)

assignment of bands to colour components is shown by (d),which has bands 3/ 1 and 5 in RGB. Exactly the sameinformation as in (c) is shown, but the choice of coloursprevents the eye from appreciating the full range of variability.The geology is of a major boundary between a high-grademetamorphic terrain with granite intrusions and basalticdykes to the left, and a very complex, younger terrain ofperidotites, metabasalts and metasediments to the right.The boundary is marked by deep alluvium along a majorriver cl;)ntrolled by a huge shear rone. Each image is30 km across.

(c)

Plate 5.1 These linearly stretched false-colour Landsat TMimages of part of the Sudan-Eritrea border are of the areashown in Figs 5.4-5.9. They reveal more details than theband 3 images in both light and dark terrains, but each bandcombination contains different information. The naturalcolour image (bands 3, 2 and 1 as RGB) (a) is least informative.The standard false-colour composite (bands 4, 3 and 2 as RGB)(b) shows vegetation in shades of red, but little more than (a).Combining bands 5, 3 and 1 in RGB (c) reveals better detailof the distribution of different rock types, while suppressingvegetation as dark tones. The importance of a correct

Plate 4.5 Superimposition of intense shear strains and uprightfolds on an earlier set of recumbent folds has produced severalfold interference structures in a Precambrian metamorphiccomplex in north-east Sudan. They show up well on thisenhanced Landsat TM false-colour composite because thedifferent lithologies are strongly contrasted in colour. Theimage is 30 kIn across. Source: author.

re.d

Plate 5.3 The Landsat TM band 5, 3 and 1 data in Fig. 5.10(c)have been transformed to intensity-hue-saturation (IHS)co-ordinates, with the intensity and saturation files stretchedand the hue unchanged. The results have been transformedback to RGB space to produce a vivid range of colours.Comparison of the two images reveals that the basic colourdifferences in the image have not been changed, so that colourcan still be related easily to spectral properties.

cyan,..Plate 5.2 Cross-section through the conical repreSentation of

intensity-hue-saturation (IHS) colour space, or colour wheel,showing how pure spectral hues on the perimeter gradethrough pastel shades to grey at the axis of the cone. In thiscase red is 0 at the top.

Plate 5.4 This image combines the first-, second- and third-order principal components from the area of Plate 5.1 andFig. 5.10 in red, green and blue. The result is a vivid colourrepresentation of surface variations in which much geologicaldetail is visible, particularly different kinds of rock. Thecolours are not easily related to the rocks' spectral properties,however, and the compression of most of the variance intothe first principal component (PC) means that the other PCsare noisy and noise interferes with interpretation in thefalse-colour image.

Plate 5.5 Decorrelation- or D-stretched image of Landsat TMbands S, 3 and 1 in RGB for the same area as PlateS.l (c).The first principal component (PC) was stretched and edge-enhanced, and higher order PCs filtered to remove noisebefore rotation back to original band space. It shows littledifference compared with Plate S.3, but the D-stretch techniqueoffers more °PPQrtunities for cosmetic improvement than theintensity-hue-saturation (IHS) transform.

(b)

Plate 5.6 (a) False-colour ratio image of the area of Fig. 5.26,using TM 5/4 as red, 3/1 as green and 5/7 as blue. Althoughproducing an excellent discrimination of rock types, the imagelacks some topographic detail and is noisy. By using theintensity-hue-saturation (IHS) transform and substituting a

contrast stretched TM band for the ratio intensity (b) bothdefects can be avoided without changing the colour renditionof different rocks, which is dependant on spectral featuresrelated to iron minerals, albedo and hydroxyl-bearingminerals. Both should be compared with Plates 5.3 and 5.4.

Plate 5.7 Vegetation in the humid climate of northernCalifornia is luxuriant over most rock types. However,ultramafic rocks produce soils that inhibit growth. On thiscolour composite ,of L~ndsat MSS band ratios-4/5, 4/6 and6/7 as red, green and blue-the bluish areas have sparsevegetation compared with the red areas. They c~rrelate verywell with known outcrops of ophiolites-black outlines.Courtesy of Gary Raines, US Geological Survey.

Plate 5.8 This image is a density-sliced version of the ratio ofLandsat TM bands 5/7 (see Fig. 5.26). This ratio should becorrelated with the proportion of hydroxyl-bearing minerals insurface rocks and soils. The colours are assigned to ranges ofratios, the 'cool' colours representing low hydroxyl-mineralcontents, the 'warm' colours higher contents.

"'...'"f;:;,.'"',

(a)

assigned to hartzburgite, layered gabbro and sheetedbasaltic dykes, respectively, which comprise the ophiolite,the brown indicates zones of pervasive serpentinizationin the hartzburgite, and the pink, yellow and orange relateto volcanic rocks and limestones, which are overlaintectonically by the ophiolite. Courtesy of Dave Rothery,Open University, UK.

""~..~' r

""'I'd. ,.t:t'v:,i ~ fr

Plate 5.9 The Cretaceous ophiolite complex of the Sultanateof Oman is difficult to interpret from a false-colour LandsatMSS image (a). Most of the ophiolite is very dark. However,careful manipulation of the data and use of superv'fsedmaximum-likelihood classification techniques allows a crudediscrimination between the different major lithologies to bemade. In (b) the colours light blue, green and dark blue are

(a) (b)

wet season (b) shows a spectacular improvement. This is theresult of st,rong geological control over vegetation andvariations in soil moisture. Important features are theBushveldt layered basic-ultrabasic intrusion in the north-east,the circular Pilansberg intrusion at top centre and previouslyunsuspected stratigraphic variations in the PrecambrianTransvaal d,olomites (lower halO. Courtesy of ERIM, USA.

Plate 5.10 Southern Africa has a strongly seasonal climate,which has a marked effect on the appearance of remotelysensed images. Image (a) is an enhanced false-coIourcomposite of Landsat MSS data for an area in the Transvaal,South Africa during the dry season. Although some geologicalfeatures show up, they are muted by the effect of deadvegetation and dry soil. The same area imaged just after the

Plat~ 6.1 This colour image ofHCMM data is of the same area asthat shown in Fig. 6.11. The red, greenand blue components are controlledby negative images of Day-IR,Night-IR and Day-VIS, so a warmarea during the day imparts a darkred to the image, a cold area at nightcontributes bright green, a surfacewith high albedo is representedby a dark blue addition, and so on.The resulting colours combineinformation on albedo and thermalinertia in a striking way. As well asa much better discriminationbetween many different major rockunits compared with Fig. 6.11(see Units 15, 2, 3, TK, KIn) there areseveral important points to note.Sandstone Unit 10 shows up as blue.It has low albedo,and is warm duringthe day and cold at night, implyinga low thermal inertia. This is rela:tedto its red colour and high silt content.Other sandstones, such as Units 12,13 and 14, appear dark orange (highalbedo, low daytime temperatures,warm at night, and hence highthermal inertia). They are pale,coarse-grained, massive sandstones.Courtesy of Anne B. Kahle, JetPropulsion Laboratory, Pasadena.

Plate 6.2 This image results fromdisplaying Landsat MSS band 7 data(representing albedo) with HCMMDay-IR and Night-IR as intensity,saturation and hue, respectively. Thearea is of the High Atlas Mountainsin Morocco, which are formed bya fold belt of Permo-Triassicsandstones and Jurassic toCretaceous carbonates resting on aPrecambrian and Palaeozoic igneousand metamorphic complex. Thevalleys are partially filled withTertiary to Recent sands andalluvium The various colours can beinterpreted in terms of thermalproperties by reference to Table 6.2.Courtesy of Rupert Haydn,University of Munich.

34°45' Plate 6.3 Combining data fromvisible, near- and thermal-infrared.parts of the spectrum throughprincipal component analysis andRGB display results in spectacularcolour images. In this case, .

components loaded with visible andnear-infrared data control red andgreen, whereas blue represents acomponent containing most of theday and night thermal information.It covers the same area as Figs 6.7 and6.12, Pisgah Crater and associatedlava flows in the Mojave Desert ofsouth-eastern California, USA. It isimpossible to relate image colours toa multitude of contributing surfaceproperties, and the image must beinterpreted empirically. The centraldark-red and orange units are aa andpahoehoe flows from Pisgah Crater,which is brighter orange-it is a cindercone. The yellow and magenta area atbottom centre represent a playa lake.Different igneous units are clearlydiscriminated in the lower left andlower right quadrants, and manysubtle differences in the alluviumat the top left also show up well(see the map in Fig. 6.12). Courtesyof Anne B. Kahle, Jet PropulsionLaboratory, Pasadena.

(a) (b)

in a roughly circular zone just left of the centre, which bothimages highlight in mainly red colours. In the ratio imageclay-rich parts of the altered zone have a high 1.6/2.2 ratio andappear bright red. Those with a high iron content and a high1.6/0.48 ratio appear green. Silicified rocks have lost their claycontent and are dark red, brown and bluish. Yellow patchesreflect zones rich in both iron and clay minerals. The TIMSimage expresses these variations rather differently, and thecolours are affected most by silica content-either quartz oropal. Silicified rocks are bright orange, and clay-rich opalizedrocks are magenta. Both types of image complement eachother in the search for this type of mineral deposit. Courtesyof Anne B. Kahle,jet Propulsion Laboratory, Pasadena.

Plate 6.4 Image (a) is a false-colour image of TIMS bands 5,3 and 1, displayed as red, green and blue after contraststretchjng of principal components and rerotation into theoriginal data space (Section 5.4). Image (b) is a colour ratiocomposite of 1.6-2.2 ~m (red), 1.6-0.48 ~m (green) and0.6-1.1 ~ (blue) reflectances from iln airborne multispectralscanner. Both are of the Cuprite mining district in Nevada,USA. The TIMS image is much better than the ratio image indistinguishing tuffs, basalts, carbonates and siltstones, plusexamples of outcrops enriched in clay minerals. The ratioimage is better for separating alluvial deposits. The areacontairis rocks that have suffered two different kinds of "alteration during hydrothermal activity and mineralization,

L J.aIe 0.:7 J.IUS lIM~ lInage was produced in the same way ashat in Plate 6.4. It is of part of the Mesozoic Wind Riverledimentary basin in central Wyoming, USA. Because of theirrod-infrared

spectral features (Fig. 6.15), silica-rich sedimentsLfe portrayed in red to red-orange, and carbonates in greenmd

blue-green. Compared with the Thematic Mapper image,f the same area (Plate 7.1), many significant variations in

tratigraphy, which are invisible in the reflected part of the

,pectrum, ate quite evident in this image. The blue-green:arbonates contain sporadic patches and discontinuous layers

Jf red sandstones. The brightly coloured, dominantlyuenaceous rocks on the left owe their variation in colour

pinks, greens and yellows) to variable proportions of:arbonate and silica in their cement. Courtesy of Harold Lang,

et Propulsion Laboratory, Pasadena.

~J ..' ) INr

'Iumination

direction

ti'

---

~athering in the rock unit there. The other, more subtle, huespinks, creams, browns, greens and blues relate more closelyvariations in surface roughness, Compared with the TMlage, the multipolarized radar image does not show such,arly defined geological.boundaries and lithological;tinctions.

However, it does show additional boundaries andits that are not seen on any other kind of image, particularlythe north-west. An advantage of the control possible with

tar is the suppression of shadows resulting fromImination parallel to the main ridges in the area. Bothages

clearly complement each other and Plate 6.;;. Courtesyfom Farr, Jet Propulsion Laboratory, Pasadena.

~ Km --QI ,ate 7.1 The images here cover a larger area of the Windver Basin than do those in Fig. 7.24. Image (a) is a false-Jour composite of Landsat TM bands 4, 3 and 2 as red, greenld blue. The same primary colours in (b) are controlled by thei, HH and W polarized gAR data shown in Fig. 7.24. Thelours in (b) therefore show the effects of surface morphologyl back-scatter and of matter-radar interactions on theIlarity of radar pulses. Those in (a) represent electronic and,rational transitions. Of particular interest are !;he red-hued~as in the south-east of (b), where strongly depolarizedrums

were produced. The lack of vegetation revealed by (a)this area means that subsurface volume scattering may be;ponsible, possibly as a result of significantly deeper

Plate 7.2 SIR-C/XSAR radar image of Erta A'le volcano inEthiopia (see Fig. 7.25) using L-band as red, C-band as greenand X-band as blue. The coloration helps distinguish'iavaflows of different ages and with different surface textures. Thebluish flows are 'rough' only in X-band and may be pahoehoeflows, perhaps of obsidian, which is common on Erta A'le.Brownish flows are relatively smooth in all bands, and this,together with their being crossed by the bluish af\d pinkishflows, may indicate that they are old flows with surfaces thathave been degraded by weathering and erosion. The white

areas face towards the radar illumination and are highlightedrather than being especially rough. Lakes are picked out asdark features. Much of the alluvial surround to the volcano isdark, suggesting that it has a smooth surface of silts and clays.The green areas at the top are of lush vegetation alongdrainages and spring lines at the base of the volcano. X-bandradar penetrates the vegetation completely, whereas to L-bandit is 'smooth', and only C-band wavelengths are back-scatteredby leaves and twigs.

Plate 7.3 Recasting a frequency-filtered digital Seasat image inthe intensity-hue-saturation (IHS) colour system restores thefull information content of the image, but adds moreeasily interpreted colours relating to varying roughness anddielectric constant. In this case the Wamsutter Arch ofWyoming, USA, is more clearly defined by rock-relatedcolour bands than in the original grey-tone image (Fig. 7.26a).Courtesy of Ronald 810m and Michael Daily, Jet PropulsionLaboratory, Pasadena.

Plate 7.4 The San Raphael Swell of Utah, USA, has been thetarget to test the geological usefulness of many remote-sensingsystems (Plate 6.3 and Fig 6.12). In this image Se~at radar datahave been combined with Landsat MSS data. The first, secondand third principal components of Landsat MSS data havebeen modulated by the Seasat SAR data and displayed as red,green and blue. The extent to which the radar data haveimproved the image can be judged by comparing the fourcorners of the image, which Seasat did not cover, with thecentral parts. There are three distinct advantages with thismethod. The better resolution of the Seasat data resolutionsharpens the expression of the Landsat data, topographicfeatures are enhanced by radar, and variations in back-scatterresulting from surface properties control the brightness of thecolours stemming from the decorrelated MSS data. A muchmore detailed geological interpretation is possible than wouldbe the case with any single imaging system. Courtesy ofRonald Blom and Bill Stromberg of the Jet PropulsionLaboratory, Pasadena. I//umlnatlo/"!

Plate 8.1 Version of Fig. 8.11(b) using look-up tables for red,green and blue to assign the 'rainbow' range from blue (low)to red (high) to the digital numbers (DN) of total magneticfield anomaly.

(a) (b)Plate 8.2 (a) 'illuminated' aeromagnetic data from Fig. 8.13combined with a coloured rendition. (b) Gravity anomaly datafor North America colour coded from magenta through blue,green and yellow to red and 'illuminated' from the north-west.The large negative anomaly over the western USA is

associated with active crustal thinning in the Basin and RangeProvince. Much detail of deep crustal structure is revealed forthe central plains, which are veneered by almost featurelessMesozoic sediments. Courtesy of the Earth and PlanetaryRemote-Sensing Laboratory, Washington University, St Louis.

(a) (b)through the superficial sediments as a result of active upliftfpllowing deglaciation, and sample bedrock directly. The redareas associate with granitic intrusions (red on a), being rich inRb and low in both Sr and Cr, as expected from the generalgeochemistry of granites. Similarly, the blue areas-richin Cr and poor in Sr and Rb-match the distribution ofbasic-ultrabasic intrusions (dark green on a). Greenareas-high Sr, low Rband Cr-occur in the dominantlymetasedimentary areas (browns and pale green on a),although some of the high-Ca granitic masses (pink on a)are Sr-rich. The diffuse reddish area extending to top centrecoincides with older migmatitic gneisses (blue on a).The coincidence of the geochemistry with known geologyis sufficiently good that anomalies that do not relate to thegeological map are likely to represent important unknownfeatures, hidden by superficial sediments. Courtesy of theDirector of the British Geological Survey.

Plate 8.3 The geology of the north-east of Scotland (a) isamong the most complex in Britain, comprising ametamorphic complex of Neoproterozoic sediments that havebeen repeatedly deformed by nappes and later regional folds.This complex also contains intrusions of granitic andbasic-ultrabasic composition. The geological map is the resultof over i\ century of field mapping in an area where exposure isat best 10-15%, and in some places is virtually nonexistentexcept along the coast. Moreover, it has been subject torepeated glaciation and most of the superficial sediments areexotic, albeit only a few metres thick. Image (b) is a renditionin RGB order of rubidium, strontium and chromium datainterpolated from stream-sediment analyses. Sampling wasin first-order streams, with catchments that are of the'order ofI km2 or less (Fig. 8.7). Despite the exotic nature of most soils,the geochemical image bears a close resemblance to themapped geology. This is because the streams have incised

shown in blue, red and orange. Courtesy of the GeologicalSurvey of New South Wales. Comparing the radiometricimage with the map reveals some interesting features. Sharpboundaries on the image at top left and bottom right relate tomajorlaults. The red to pink body on the image towards topleft links to the redgranite{)n the map, but extends over alarger area, perhaps indicating associated permeation of themetapelites by V-rich pegmatites. Some of the amphibolitesshow as dark streaks. The blue granite appears to be equallyrich in all three radioactive elements, whereas the glacigenicsediments have low radioactivity. The bulk of the metapelitesappear to be Th-rich. Possibly more detail is revealed by theimage than by the map. Courtesy of CSIRO, Mineral PhysicsDivision, Australia.

Plate 8.4 (a) Colour rendition of gamma-ray intensities foruranium, thorium and potassium as red, green and bluecomponents for an area in Australia. Reds, greens and bluesshow areas that have high values for U, Th and K, respectively,and low values for the other two elements. Equally high valuesfor all three elements are shown by grey tones and pastelshades. Dark areas have low radioactivity. Courtesy iSf BMRand CSIRO, Canberra. (b) Geological map of thecorresponding area. Black lines indicate important faults andfold axes. Yellow areas are Quaternary unconsolidatedsediments and all other shades relate to various Precambrianrocks. The brownish units at top right are glacigenic sedimentsresting unconformably upon high-grade metamorphic rocks,the bulk of which are aluminous metapelites (pale purple),together with green amphibolites and various granitic rocks,

Plate 8.5 Image combining digital elevation data withBouguer gravity anomaly data for the same area as Fig. 8.15,using the DEM illuminated from the north-west as intensity,a constant saturation and gravity as hue in an intensity-hue-saturation (IHS) transform. The colours range fromblue

for negative anomalies through a rainbow spectrum tomagenta for the highest gravity anomaly. The green to yellowtransition, being the most easily perceived spectral boundary,is keyed to the zero Bouguer anomaly.

Plate 8,6 This image includes the area in Plate 8.4 and usesthe intensity-hue-saturation (IHS) tr-ansform, withaeromagnetic data as intensity, a constant saturation andhue derived from U, Th and K data in Plate 8.4 (a). Althoughthe faults are well displayed in both magnetic and radiometricdata, the unconformity beneath glacigenic sediments has noeffect on the magnetic data, however, it is prominent in theradiometric data.

Plate 9.1 Enhanced Landsat TM image (bands 7,4 Wd

2 as RGB) of an unmapped area in Eritrea showing clearlydifferentiated lithologies. The yellow and brown unitsare different marbles, red-magenta corresponds withmetamorphosed basalts, and blue-grey units are chloriteand graphite schists. Bare and vegetated alluvium showsdistinctly in the major drainage channels. The sharp contactsbetween different lithologies, and abundant examples offine compositional banding help delineate major structuresin the area. In the north-west quadrant, fine compositionalbanding (red, p,aleand bluish) has been shift;ed by strike-slip

L__1_~ 11"T "'£_..1L- 'r\._£~..I'~A_~_I,~I:~

within a tectonic unit that is bounded to the south-east bya sharp feature that cuts across the strike of the displacedbanding to its north-west. This feature is a major shear zone.The yellow marbles that dominate the northern part of thearea define several large folds. To the south-east of the marblezone are two thrust-bounded tectonic units, the thrustsbeing revealed by strike cut-offs. The lower of these unitscontains brown marbles and pale overlying rocks, whereasthe upper unit is dominated by red metabasalts and blueschists. The complex is part of an accreted assembly ofNeoproterozoic terranes that formed in a series of island

C:_-- ,,1\ " ,,~ L-

Plate 9.2 Perspective view combining enhanced Landsatbands 7, 4 and 2 as RGB with a digital elevation model (DEMderived from stereoscopic aerial photography. This shows adissected unconformity in north-east Africa that separatesTertiary continental flood basalts (dark grey with vegetation-~--~_\ £_~- ,",--,1.. "',--,-- D_~___1..~'~~ _h'~~""';_O~'~

with a strike that trends from lower left to upper right in themultihued terrain. The Tertiary cover forms a plateau at uppe:left, and a relict ridge at right. The horizon is about 15 km longExaggeration of the vertical scale gives a false impression ofthe slopes in the topography. The range of elevation in the are,.r~ 1nnn L- "'Ann_~

Image

of part of Nova

Plate 9.4 By combining SAR data asintensity and airborne gamma-rayspectrometer data (K, Th, U as RGB)as hue, an intensity-hue-saturation(IHS) transform reveals considerablestructural and lithological variationfor this area in Nova Scotia.Comparison with Plate 9.3 showsthat the granites mapped in the fieldas similar bodies have considerabledifferences in U, Th and K content.

Plate 9.5 This false-colour ratio image of a geologicallyunknown area in Eritrea uses the ratios 5/4, 3/1 and 7/5 inRGB order to exploit the spectral features of the ferric iron-and hydroxyl-bearing minerals, which are often abundant inhydrothermally altered rocks, to assign a yellow colour toareas bearing them. The yellow areas on the left proved tocoincide with hydrothermal alteration of metavolcanic rocksin which scheelite (CaWO4) was found in quartz veins.The larger yellow zone running north-south is an area ofgossanous material overlying a metamorphosed sulphide-richshale, thereby showing that remote sensing has to be usedwith field verification. Size: 30 by 30 km.

(a) (b)

(c) (d)

I (f)

cover in green. (e) Landsat TM band 5 and tin values combinedthrough the intensity-hue--saturation (IHS) transform, withmagnetic field intensity shown in contours. (f) Directionallyfiltered Landsat.TM band 5 image (to enhance fractures,the most important of which are in white) combined withBouguer gravity anomaly data through the IHS transform.Superimposed in red and green are areas with significanttin and zinc-bismuth anomalies, respectively. The last twoare spatially associated with fractures at the margin of thenegative gravity anomaly above a buried granite. Courtesyof the Director of the British Geological Survey.

(e)Plate 9.6 Data and results from an image-orientatedgeographical information system (GIS) exercise in targetingtin mineralization in the English Lake District. (a) Geologicalmap: within the Borrowdale Volcanics greens are basaltsand yellows rhyolites. (b) Colour sliced map of Bouguergravity anomaly, ranging from magenta for lowest negativeanomalies through blue to red at highest positive anomalies.Superimposed are contours for magnetic field intensity.(c) Tin, copper and zinc values sliced at 50-75-95-99percentile levels (Fig. 8.8b) and displayed in RGB order..'(d) Landsat TM 742 image showing the dominant vegetation

Plate 9.7 Stand of prematurelysenescent maple trees (orange)related to seepage of natural gas.Courtesy of Harold Lang, JetPropulsion Laboratory, Pasaden.

Plate 9.8 Landsat TM 742 imageof part of the East African Rift inKenya, with vegetation related tosprings near faults showing greenin an otherwise arid terrain. Widtl140 km. Courtesy of Eosat.

(a)(b)

(c) (d)Plate 9.9 Examples of interferometric radar images that showchanged ground elevations. (a) Interferogram of part of theSelby coalfield in northern England, produced from ERS SARdata separated by 35 days. The blue to red interference fringeseach signify 28 mm of settlement, the total in this area beingabout 90 mm in the period covered. The inset is for a differentperiod, and shows details of both subsidence and miningactivities. The ornamented rectangles show the extents ofsubsurface workings in the coalfield. Courtesy of Trish Wright,Marconi Research Centre: @Marconi Electronic Systems.(b) Interferogram of the Las Vegas area, Nevada, producedfrom ERS SAR data from April 1992 and December 1997. Theinterference fringes (see bottom left) each span 100 mm ofsettlement, giving a total over 5.8 years of 20 cm. In this case,

the settlement is due to abstraction of groundwater frombeneath the city. Courtesy of Falk Amelung, StanfordUniversity. (c) Interferogram superimposed over a perspectiveview of Mount Etna, in which several 28 mm fringes indicatethe changes in surface elevation resulting from magmamovement. Courtesy of Didier Massonnet, CNES, Toulouse.(d) 28 mm colour fringes related to ground subsidence causedby the 1993 Eureka Valley, California, earthquake (magnitude6.1). The elongate basin subsided by around 10 cm during andafter the earthquake. The dark areas are to the product ofincoherent returns in the radar data, whereas the colouredareas beyond the seismic basin reflect shifts of a fewmillimetres, possibly as a result of soil swelling after rains.Courtesy of Jet Propulsion Laboratory, Pasadena.

:ig. 5.31 Diagrammatic representation of the variq,ps possibleayers that obscure or cryptify true geological relationships. To:his

may be added the limited number of mineral types that~asily acquired remotely sensed data can distinguish. Even if:ompletely bare of cover, rocks can rarely be classified on;tandard

petrographic grounds by remote sensing alone.

Figures 5.28 and 5.29 ihow that some surface categories:lisplay contrasted reflectance at different wavelengths.Using two bands of data, selected because of such con-:rasts, can enhance differences between categories andmprove their separability. This is shown diagrammat-cally in Fig. 5.32(a). To train a computer to make these;eparations means giving it a set of rules on which to baseyes' or 'no' decisions. The simplest means of doing thisis

to divide the bivariate distribution into rectangularboxes (Fig. 5.32b). The boundaries of the boxes representthe

ranges of ON for the two bands within small, knownIreas of the interesting surface categories. Such areasIre

selected during field work and are termed trainingIreas, because they are used to train the computer. The:omputer compares the ON of unknown pixels With:he boxes. If they fall within one they are assigned to:he relevant class. If they fall in none then they remainunclassified, and possibly indicate a need to refine themeans

of classifying or increase the number of classes. Thesame principle can be applied to any number of dimen-

sions, and usually is. This is known as parallelepiped:lassification. An important limitation is that natural cat-~gories

generally plot as ellipsoids in N-dimensional space,md parallelepipeds are only a crude representation ofthis. A refinement is to represent the volume occupied:>y the data from training areas by an assembly of smallerparallelepipeds

(Fig. 5.32c).Providing the computer with a more sophisticated

set of rules for decisIons of the 'yes, if type means usingthe statistics of data from training areas. The simplest ofthese

classification routines is to identify the mean ONfor each training-area category in each band being used.The distance from each of these means in N-dimensionalspace of an unknown pixel can then be cal'culated. Thepixel is assigned to the class with the closest mean.Figure

5.32(d) shows this minimum-distance-to-meansclassification for the two-dimensional case. A refinementof this is to bring the variance of ON within training areasinto

play. Assuming that the ON within a training areaform normal distributions for each band, the histograms:an be considered to be beU-shaped. Depending on thevariance (a measure of the width of the distribution) thefurther a ON is from the mean the less the probabilitythat it represents the category in question. In the two-dimensional case (Fig. 5.32e) ON from a training areadefine

an elliptical cluster. Each cluster can be contouredto show how the probability decreases away from the

mean point. Figure 5.32(f) expresses this more graphic-ally as a 'topography' of probability. The computer canthen assess the plot of ON from an unknown pixel inthis probabilistic context. It calculates the likelihood ofthe pixel belonging to each of the predefined classes andassigns it to that class where the likelihood is maximized.This

method is maximum-likelihood classification. Thethree multispectral classification methods shown in

:over. More often than not so many classes would be.equired that the results are so confusing as not to'epay the effort expended. The only option is to visuallyinterpret enhanced images of different kinds on the:raditional basis of colour, texture, pattern and geolo-~cal

context. In the majority of applications this is theiominant approach, aided only where appropriate byudicious use of classification. Despite this pessimisticview, it is possible to detect and outline spectrally unusual~ocks by automated means-many are interesting because:heir oddness stems from them being eco~omicallynteresting concentrations. For that reason, briefly intro-iucing

some classification techniques is appropriate forthe image-orientated geologist.

Inspection of the histograms for different surface:ategories allows different ranges of DN in the overallrustogram to be assigned to those categories or groups of:ategories that can be separated by this means. Assigning

1 colour to each of these ranges highlights each pixelwith a DN that falls within the range in the appropriate:olour. This simplification of the data, in effect a crudeclassification, is called density slicing (Plate 5.8). WhereJnly

a few, distinct surface types make up an area, thiscan be a rapid and accurate means of mapping their.patial distribution. A more frequent use is to simplify a.cene

to assess gross variations, or to express continuousvariation in a single type of surface. One example of thelast application is density slicing of the VNIR/red ratioto express variations in density of vegetation cover.I\nother is slicing the DN range of blue or green reflectedradiance

from the bed of a body of clear water to expressdepth variations (Section 1.3.4, Plate 1.1).

150 Chapter!

(a (b

z()).'1I:II

CI

c).."0,"

<t5'o°cf>o corn

AAAA A A fore

CCCCC hay

san!

+. + + ..rh+ + u. udl

XxXll wate':)("---I X~ XIX XI X XI X )I X~L_.;x!.

xx xxXX

x xx '

x;x

-cr- - -,r:.~r;- -+-.:j::- T I.~~1 + + + + + I~-l1+ + + + I+ + + .+J r Q1 + 1 r\O

1++ +++1 1 O~OfI + +1 ~ 1 I + + 1 --+ i - 1+++ +Ir --1 I IL :f l I ~I

-I 0-1 0

: ° °cP

~~---~P- '6-h1"-

I ~4~ I1 4 ~44 I~44 ~ I~~~~ I1t.~~16 J

Band X Dr Band X DI

«(r-' T ' r.~...+ + 11.-'4: + ++ + + ++: :~-I. + + Iif" + ++ -II

~:+ + +++: io II ++ + +I ~I I ~r+ + + +! JL I= J !~-C c

r-;tI CrPr--: ~-_JIrPP

P-~~6~1 6 66~bI~66 61

~664-.J~_6- ~

z0>-"0CIV

01

~--:l~ x j.V XXI

14:ll( .j:1.)( )

~~Band X Dr Band X 01

(e (1

~

sand -

rcorlI

Ar-ha'~

c)-"t:£:1\

IX

urbalr- fore!

~

equiprobabilit;contours ~

~

---:;.;;;:..~~~--=;:-;

~.'""""~~~~

~>::~;';~~~~(v~"

~~

~-::

.,(.:~~ 0'

V>3(\UBand X Dr

water ~ ---,

represents the ~iIPum-distance-tO-means method.The contours in (e) express the probability that any poinbelongs to a particular class, which is the basis of themaximum-likelihood method of classification. Theprobabilities are illustrated more graphically as athree-dimensional topology in (f).

Fig. 5.32 Different me.thods of spectralcl~ssification can berepresented diagrammatically by reference to bivariate plotsof two of the barlds used. In (a) several clusters in the datacorrespond to areas with different kinds of surfaCe. The simpllparallelepiped classification is shown in (b), that based onmore precise boundaries for each class in (c), and (d)

+ + 0+ + ++++ cola

+ + + + + ~o+ + cj5)- cB

++ + ++ + q.,~~

Dr-£!:0 ~

$U~0 ~~0 ~~~~~~

~ ~~ ~~~~~ ~L,t.~~~~~~~

Digital Image Processing 15

Fig. 5.32 form the basis for more sophisticated method~that are beyond the scope of this book; but the associatedfurther reading gives more information.

Before making a maximum-likelihood classification,an interpreter may be able to assess roughly how mucl1of a scene is likely to be occupied by each surface cat-egory. This may.result from field observations or froma rapid inspection of an image. This crude estimate canthen be used to weight the probabilities involved in th~classification. For example, beach sand will be limitedto narrow strips around bodies of water. The probabil-ity of any randomly selected pixel being beach sand i~therefore low, and its calculated probability in that con-text is given a low weighting. Conversely, vegetationin a humid climate is very common, so the probabilityof any pixel being classified as such is given a higl1weighting. Another anticipated factor is tha,! not aUreal surface categories are represented in the trainingareas. The classification can be weighted accordingly sothat the computer does not try to force 100% of the sceneinto classes. Both these refinements help to ensure thatthe classification is as close to reality as permitted by thequality of the data and the precision of the classifica-tion method.

All the above classification methods rely on theoperator making decisions about the areas on the groundsthat are most representative of the surface categories 01interest. This approach is termed supervised classifica-tion. It is also possible to allow the computer to examinethe data in each band and sort out particular correla-tions amongst them. This in effect means dividing theN-dimensional histogram into arbitrary segments purelyon the basis of heterogeneities in the distribution 01DN. The operator can control the sharpness and numberof the divisions, but basically leaves the computer to itsown devices in assigning pixels to different classes.This is unsupervised classification, and expresses differ-ences purely on the basis of spectral properties, i.e. it isempirical. It is useful in totally unknown areas, when itprovides the operator with a means of identifying spec-trally different kinds of terrain. This can then orientatethe future field work, or be used as an adjunct to moreconventional visual interpretation of the image.

Usually, the more independent variables, or dimen-sions, used in a classification the more easily spectr-ally distinct classes can be separated. However, in the0.4-2.5 J.Lm range there is a high degree of correlationbetween bands. This is quite clear from the spectralreflectance curves in Fig. 5.29, particularly those for rocks.Differences are expressed in only two or three bands-in the visible range expressing colour, that around 0.8-0.9 J.Lm with an absorption feature related to oxidiZediron minerals and that.around 2.2 J.Lm for the OH- vibra-tional transition. These bands are the main means ofdiscriminating iron-rich and clay-rich rocks from others.

By themselves these bailds would perform a roug!and ~eady classificatiqn. It might seem that using all th(bands available from a rem@te-sensing system woulcsharpen the discrimination even further. However, th(effect of strong interband correlation is to dilute the con.tribution of the most discriminating data. What happeruin practise in this part of the EM spectrum is that a~the number of bands used for classification increase~beyond a certain limit the accuracy of the results actuall)declines. A strategy n~eds to be devised to use only thos(bands that do the job well. This approach also speeds uFthe operation of the softWare involved.

An irritating problem with classification based orspectral data alone is th~ variation in ON resulting frOIrvarying illumination. In terrain of high relief or wher(Sun elevation is low, the presence of many shadows dis.rupts the process. Moreover, the spectra of rocks contaironly subtle features, which can be hidden in the naturavariations. Variation in illumination is reduced by ratio.ing. Given carefully chosen bands, ratios can enha:cflc(these subtleties (Section 5:5). Another advantage of ratictechniques is that they combin~ data from several band~in a smaller number of paramet~rs, thereby improvin~classification. A disadvantage is that they have a 10"signal-to-noise ratio, which tends to counteract some ojtheir advantages, with noise pixels being rnisclassified.

Figure 5.32(e) shows that four of the six categoiie~overlap completely in the bands defining the horizo~taand vertical axes, even though the overall separatiorlooks good. Unfortunately computers do not have eye~and work systematically with Cartesian co-ordinates irthis case. A classification based 011 Fig, 5..32(e) would notbe very good. If, however, the data could somehow b(plotted on other axes, so that differences are maximized,that would improve the classification based on tWo band~alone. A more powerful approiich is to perform a PCanalysis before classification. This not only derives newaxes, but also expands the data to fill N-dimensional spac~as fully as possible. More important still, it combin~together the hlghlycorrelated parts of the N-dimensionaJdistribution along the different axes (Section 5.4). Thisreduces automatically the number of axes containirigsignificant data. Each axis has a contribution from allihebands and the new data sets are uncorrelated. Furtherimprovements deliberately attempt to arrange.the'newaxes to maximize separation betWeen the fields of knownor suspected classes using directed PCA (Section 5.4.1),as in Fig. 5.33.

Classification for geological applications has tWopractical approaches. One is to map surface classes thatcorrespond to several different kinds of underlyingrock. This can be useful in producing a reconnaissancelithofacies map of a whole area, to be refined by fieldche~king. In otherw<?rds, the common rocks are sought.Plate 5,9 gives an example of. this approach. Another

152 ChapterS

zQ>--0t:"'

aJ

Band X DN

Fig. 5.33 Lying within the highly correlated ellipse of data onthis bivariate plot of digital numbers (DN) for two bands arethree classes of surface t~t need to be separated. It is clear thatclassification based on the two axes shown would produceconfusion between the classes. However, by basing: principalcomponent (PC) analysis on the means and covariances of datafrom within training areas of these classes produces newcanonical axes, which maximize the differences between them.

strategy is classification to pick out those small, usuaijyimperceptible areas on an image that contain rare butdistinctive rock types, This is of great importance incommercial mineral exploration. First, ore deposits areunusual rarities by definition. Second, to find them bypainstaking field survey is slow and expensive, espe-cially in poorly known areas. If classification of remotelysensed data, even if it is quite imprecise, can narrow thearea of exploration, ~o much the better. Field geologists,geochemists and geophysicists are then able to con-centrate their talents much more cost-effectively.

Whether or {lot a particular patch of a surface categorycan be classified accurately depends on its size relativeto the spatial resolution of the imaging system. It alsodepends on its contra~t to the surrounding surface. Inmany cases an individual pixel's ON is contributed toby several surface components with sizes that are lessthan that of the pixel itself (Fig. 3.10). The ON of sucha mixed pixel is an average of those of each of its com-ponents weighted by their relative size and their con-trast. Strongly contrasted components may have suchan effect on the pixel that their contributions to its ONare disproportionately large compared with their actualsize. Under certain conditions very small isolated areastherefore can be detected and classified.

the scene corresponds tQ spectra of radiances for eachpixel that makes it up. Consequently, these spectra canbe compared with laboratory or field spectra of puresurface materials in ways that make possible identifica-tion of the mineralogical composition of natural surfaces.Before you become over-excited at the seeming pos-sibility of geological mapping without the rigours offield work, remember that reflectance spectra addressonly the top few micro metres of the surface. That is, thespectra are dominated largely by minerals that formwhen rock-forming minerals are weathered, and they arenot reflections of the petrographic and textural criteriathat geologists use for rock classification. Moreover,as Fig. 5.31 shows, over most parts of the Earth rocksare obscured by soils (often transported) and varioustypes and densities of vegetation. As if that were notdepressing enough, there is a further complication; rocks,even when in a pristine state, are usually made up ofmixtures of several prominent minerals. Nevertheless,hyperspectral data are so rich in information that theycan more than while away those rainy Friday afternoonsin the office!

A typical set of hyperspectral data involves 224 bands(A VIRIS), which means factorial !224 (224 x 223 x 222x ...) possible RGB false colour images. If I am notmistaken, that is somewhat in excess of the numberof hydrogen atoms in the known universe. Clearly,exploring the possibilities is not to be undertaken lightly;indeed for most combinations it is fruitless becauseof the high correlation between nearby narrow bands.The key to such an approach, which is useful becauseit maintains the data in visually appreciated form, isto remember that the sequence of bands builds-up foreach pixel a reflectance spectrum that is almost as pre-cise as one collected by a laboratory spectroradiometer.Knowledge of the wavelengths associated with variousmineralogical transitions-absorption features, such asthose associated with Fe3+, AI-OH and Mg-OH, andC-O-allows bands that occur in and away from com-mon features to be displayed as RGB combinations. Theywill show areas of surface that contain an abundanceof the chosen mineral in vivid colours, whereas areascontaining little will be dull greys. For an unknownar:ea, that would be a good opening strategy. Some hyper-spectral software packages show transects across a sceneas stacked, one-dimensional RGB combinations of spec-trally associated bands in increasingly long-wavelengthregions of the data, allowing bands to be skipped to avoidthe problem of redundancy. Where there are areas alongthe transect that contain high abundances of mineralswith prominent features, the RGB 'slices' in differentparts of the spectrum show as bright colours. Variousmeans of enhancing contrast (Section 5.2) allow mutedc910urs to be made more vivid and interpretable in thesegenerally strongly correlated bands. Spatial filtering to

5.6.2 Spectral mapping

By collecting image data in many narrow, adjoiningwavebanq,s, hyperspectral imaging devices opened upa new dimension in remote sensing. As well as vastlyincreasing ti)e number of possibly discriminatory com-binations of bands, as RGB images, ratios and so on,

Digital Image Processing 153

classification according t~ variable shading of the surface.Other empirical methods use, in one way or another, thecombined effect of hyperspectral bands to define spectrafor each pixel in the scene.

Earlier in Chapter 5 you became familiar with bivari-ate plots of DN from broad-band data, where eachpixel is expressed in two-dimensional space. These oftenshow broad distinctions between different materials, asin Fig. 5.32. Hyperspectral data occupy so many bandsthat bivariate plots cannot even come close to expressingthe full variability in a scene. For a computer, multidi-mensional or hyperspace is trivial, despite our difficultyin grasping it. The band values in a spectrum for a pixelplot as a point in hyperspace. Other pixels with similarspectral properties cluster in the same region of thisn-dimensional space, whereas those with dissimilarspectra plot in distinctly different regions. The outcomeis a series of 'clouds' in hyperspace. Two-dimensional'slices' through the n-dimensional distribution wouldappear similar to the bivariate plots in Fig. 5.32, butclearly do not show all the 'clouds' that exist-moreof this later. In this form, it might seem possible arbitrar-ily to define spectral' clouds' or classes using any ofthe methods expressed by Fig. 5.32. However, this isinefficient. A faster method uses a process analogousto human brain functions-:a neural-network technique(beyond the scope of this book). It starts by randomlyselecting a number of positions in hyperspace, say 256,that arbitrarily represent the central co-ordinates ofspectral classes. The position"of each is joined by a line orvector through hyperspace to the origin of the distribu-tion (0,0,0, ...0,0). How similar each pixel's spectrumis to these classes is measured by the n-'dimensionalangle between its vectot and that of the class centre.The first pixel assessed in this way will be closest to oneof the arbitrary class centres, the position of which isadjusted to improve the fit. Progressively all pixels areconsidered, and the class centres evolve to positionswith vectors that best fit all the spectra withimhe scene.Such an unsupervised, neural~network classifier is welldescribed as a self-organizing mapper.

The same principle of exploring hyperspace lies behindmethods of supervised classification, i.e. when there issome knowledge about the types of surface in an area.One begins with a training area that a known, miner-alogically simple surface dominates, but includes othertypes of surface. The pixel spectra for the training areaplotted in hyperspace will contain a cluster correspond-ing to the known material. By exploring plots of two-dimensional slices through the distribution, generally bya random rotation of the axes, that cluster can be isolatedand defined in hyperspace. Its centre forms the basis forlater classification. If field spectra are available, they toocan be used to create n-dimensional centres. Using bothmethods of centre defutition for classification depends

sharpen boundaries is less useful, as the narrow band-widths mean that signal-to-noise ratios (Sections 3.3, 3A)are considerably lower than in broader band data-edgeenhancement makes the noise more obtrusive, whereasnoise-removal filters may remove spectral features ofsmall areas with unusual mineralogy .

The large number of bands, together with the highredundancy among most of them, suggests that datareduction by principal component analysis is an excel-lent strategy, both to save time and to highlight inter-esting areas and materials. As you saw in Table 5.1, thekey to understanding the results of PC analysis is theeigenvector matrix. That shows the loading of differ~ntbands in each PC. In the case of bands that encompassdifferent materials' absorption features, PCs that arestrongly loaded by them express most clearly the dis-tribution of those materials. For hyperspectra~ data, theeigenvector matrix is very large-about four pages of thisbook-and therefore difficult and confusing to use. Analternative strategy is to express the eigenvectors for eachband as a curve for each PC. As the bands ar~ arranged inincreasing wavelength, this rendition can be comparedwith reference spectra of different materials. In this waythe operator can select the PCs most affected by dis-tinctive features in them. Similarly, the eigenvalues, orvariances for the PCs can be plotted graphically. Yet it isstill tedious to plough through as many as 224 plots-for high-order PCs may contain information on rare andspectrally odd materials as well as noise. The directedPC analysis approach helps overcome that, by perform-ing the transform for mineralogically selected subsetsof the data (Section SA. 1). The division of Landsat TMbands, and those of other broad-band sensors into'Fe3+, and 'OH' regions (visible and VNIR, a\ld SWIR,respectively) for directed pC analysis expands with hyper-spectral data to groups of bands that encompass, Fe3+,AI-OH, Mg-OH and C-O regions, and in a few casesspecific minerals, such as alunite.

Using ratios between bands that express specific absorp-tion features (Section 5.5) is also a possible strategy withhyperspectral data, being possible for several specificmineral groups. As well as reducing redundancy andhelping focus on specific minerals for visual image inter.:pretation, PC analysis and ratioing of hyperspectral dataalso assist in their digital classification, by reducing thenumber of dimensions involved.

As field identification of many of the minerals thatshow prominent spectral features is difficult, particu-larly when they are mixed, supervised classification ofhyperspectral data is difficult without field or referencespectra. In their absence, there are several variants ofempirical,. or unsupervised classification. Data reductionby PC analysis or ratioing aims at the techniques forunsupervised classification outlined in Section 5.6.1,with ratios having the advantage of reducing spurious

154 Chapter 5

ProVided field spectra are obtained underillumination ..

-preferably

suffer from none of these effects, being calibrated

on how the distribution of spectra thr()ughout the imageare treated. One method focuses on calculating the anglebetween spectral vectors and those of the class centres.Low values indicate a close similarity of pixel spectrato the centres, and high values represent pixels that areunlikely to be of the same kind. Another assumes that allpixels contain varying proportions of the material thatdefines a centre, and therefore their spectra are mixturesof its spectrum and many unknown spectra. This methodthen calculates the proportion of the spectrum of eachpixel that could be produced by incorporating that ofthe centre, a perfect match giving a result of 1.0 or 100%,and a complete absence of signs of the centre spectrumyielding O. A third approach calculates correlation coeffici-ents betWeen each image spectrum and that of the centre,1.0 showing a good match and 0 no match at all. Thesemethods are not 'yes/no' classifiers, but proQuce con-tinuous tone images that express probabilities. In thatform their use is for assessing mineralogical variationin an area, butby using threshold values, distinct classesbecome possible. Variants combine results from severaldifferent centres as RGB images.

Field knowledge, either as spectra or training areasfor surfaces dominated by specific minerals, can be usediri. another way. For example, let us assume that train-ing areas or field spectra define all the most importantmaterials (minerals and vegetation) that contribute tothe surfaces in a scene. By assuming that all pixels con-tain spectral contributions from all the materials andtheir spectra result from linear mixing of these end-members, the linear uturiixing method produces imagescorresponding to the calculated contributions of all theend-'members. The sum of the fractional contributionsin the perfect case will be 1.0 or 100% for all pixels.The critical stage 'in linear unmixing is ensuring that allpossible end-members contribute to the calculation; adifficult undertaking even in well-known areas.

ln a perfect world, geology would provide us with atleast some outcrops containing absolutely pure mineralsin' which we are interested-that never happens! In alaboratory setting, however; such pure end-memberscan provide definitive spectra. Using laboratory spectrais the key to high-confidence spectral mapping, but hyper-spectral images gathered from air- or spacecraft sufferfrom all manner of effects that are beyond our control.They measure radiance-the spectral r~diant energy,originating in solar radiation, that is reflected from theEarth's surface. The spectral properties of the surfacecontribute to that, but so do other things. Solar energyvaries in intensity with wavelength, partly because ofthe Stefan-Boltzmann and Wien laws (Section 1.2),partly from atmospheric scattering and absorption bygases in the air (Section 1;3,1), and also from the angle atwhich solar radiation strikes the surface (varying withsun angle and with topography)...

first to the second. This can be done bymodels of atmospheric transmittance, but most con-veniently by mimicking the way a laboratory spectrora-diometer is used. Areas of quartz-rich sand or salt,and particularly concrete, have flat reflectance spectra(uniformly reflective at all wavelengths) and are highlyreflective. If they are also topographically flat (as theyoften are), then an image spectrum of them will be analmost pure signature of atmospheric effects at the timethe image was acquired, in the same manner as that ofa laboratory spectroradiometer's reference target. Flat-field calibration divides each image spectrum by thatof such a spectrally and topographically flat area, to pro-duce a close approximation to a reflectance spectrum. Ofcourse, in many parts of the world bare, flat expanses ofany such spectrally monotonous materials are rare. Anapproximate calibration to comparability with laboratoryreflectance spectra depends on estimating the atmo-spheric effect from all pixels in the scene. This involvescalc?lating an average of all the image spectra, which,if the scene is sufficiently diverse, cancels out most ofthe features resulting from 'Specific materials. Variousadjustments prior to averaging, such as correcting foratmospheric scattering and instrumental gain by sub-tracting the average spectrum for the darkest pixels inthe scene from all others, and adjusting all scene spectraso that the area beneath them is the same-effectivelynormalizing the brightness in every band-renders theaverage spectrum as a close approximation to that of a flatfield. Each normalized spectrum divided by the averagebecomes an approximate reflectance spectrum.

Calibrated image spectra, being comparable withreflectance spectra of pure minerals obtained underlaboratory conditions, potentially can be analysed quan-titatively in terms of the mineralogical composition ofthe surface. Methods based on single class centres, definedby laboratory spectra, and on linear unmixing providepowerful tools for mapping. Many investigations whenhyperspectral data become more widely and cheaplyavailable will undoubtedly focus on exploration, that ison finding small occurrences of minerals that are goodindicators of mineralization; such as various Fe3+ -andAI-OH-bearing products of hydrothermal alteration.Using laboratory spectra of anticipated indicator min-erals, together with spectra ofunmineralized, commonsurfaces, linear unmixing helps target areas on whichground exploration can focus.

Frequently, areas occur on various images that arequite clearly highly anomalous, yet their composition isunknown. With calibrated hyperspectral data, a spectrumfrom such an area can be matched against all referencespectra. Its degree of fit with standard minerals allowsthe composition of the anomalous surface to be assessed.There is a means of picking out such anomalies, however,without seeking them in the vast number of possibleimages that hyperspectral data present.

The exploration of image spectra in hyperspace byreference to arbitrary vectors though it forms the basisfor another approach. Any pixel in a scene comprisesvarious proportions of materials, either patches of purematerials, as analogized by the mixed pixels in Fig. 3.10,or intimate mixtures of say mineral grains or individualplants of different species. The spectrum of a pixel is theadditive effect of those of all the materials that it contains,according to their proportions at the surface. Knowingthe materials and their proportions would allow usto reconstruct the pixel's spectrum. Pixels composed ofa single material have spectra that, by definition, aredifferent from those of several mixed together-they arepure end-members, analogous to orthoclase, albite andanorthite in the feldspar solid solution series. So, in ahyperspace plot of image spectra, the spectra of pureend-members would be extreme values. The chances offinding pixels consisting of a pure materials decreasesas pixel size increases, but if it was possible to find thosethat are extreme relative to the much more abundantmixed pixels, that would give a good idea of the range ofmaterials that contribute to the scene as a whole. Onceidentified, their spectra can be matched to those of refer-ence materials (see below). This gives a guide to analysisby the various techniques that I have outlined.

Estimating pixel purity relies on a computer assessingthe position in hyperspace of each pixel spectrum rel-ative to random vectors from the origin. Each analysisrecords the extreme pixels for a vector. As the processiterates through many such random vectors, it counts thenumber of times each pixel is an extreme. The numberassigned eventually is an index of pixel purity, thehigher it is the more likely a pixel is to consist of a pureend-member. Using colour density slicing (Section 5.6)helps highlight the most extreme pixels, and theirspectra, matched against reference materials, suggestthe regional end-members.

frequencies dominate, textural analysis is second~ryin importance to evaluation of tone. As introduced inChapter 2, human vision is very perceptive in recogniz-ing textural features. However, it is very difficult for aninterpreter to make accurate decisions on the boundariesbetween areas of subtly different texture-different inter-preters would place the boundaries in different places.The use of computer-assisted measures of image textures,although very difficult and time-consuming, may addto the power of image classification. One method is tocalculate the variance of ON within predetermined squareboxes across an image. Higl1 variance suggests fine,sharp textures, whereas low variance indicates coarse,smooth textures. Another approach is to extract differentspatial frequencies from an image using a variety offilters. Each expresses different kinds of texture, and thedifferent images can form the basis for a semiquantitat-ive classification of texture (Chapter 7).

One approach is to use statistical measures of thespatial frequency of variations in ON. To do this meansthat limits have to be set on gradients of ON and thespacing between changes in gradient must be representedin some way. This approach is likely to become increas-ingly useful in geological applications. Another line ofattack is to use mathematical expressions of the spatialstructure of the data. Basically, a sort of mathematicalgrammar is devised to describe each spatial pattern inthe image data in terms of its parts. Each subpattern isbroken down to its components, the smallest distinguish-able category of which comprises primitives. Primitivesmay be the result of some kind of statistical approach totextural features. They are signified to the computer by amathematical 'sentence', and each class of patterns isdescribed by a set of such sentences. So far the use of pat-tern recognition of this kind has focused on simple orpredictable patterns such as print characters in documentimage processing, chromosomes and military objects.Simple pattern recognition has been applied to geometricfeatures in geology, such as the detection of circularand linear features. Although these are of considerableimportance, the fact is that such features are found equ-ally as easily by visual inspection (Chapter 2), and theautomat~d methods often find spurious features.

Where a precise classification scheme exists forextremely complex patterns and textures, as in the caseof fingerprints, considerable advances have been made.Th~ problem with geological patterns is that, as wellas being complex in their own right, they are also partof a complex whole. They have to be detected as wellas being described and classified. Fingerprints are notdetected by computers, they are sought, found andphotographed by detectives. Elegant though automatedspatial pattern recognition may be, it remains to be seenif it can be develgped for new geological applications.More important is the question of whether it will repay

5.6.3 Spatial pattern recognition

The surface texture of different rock types, as expressedby drainage networks, colour banding and mottling forinstance, is an important parameter in photointerpreta-tion (Chapter 4). It represents the spatial distributionof tone variations. It predominates where tone, or ON,in an image varies with a high frequency. Where low

156 Chapter 5

the effort in comparison with the biased, but. extremelyeffecti~e analysis by trained geological interpreters. Onegreat difference between the military pattern analystand the geologist is ~hat the former generally is not per-mitted to visit the site of interest!

The spatial variation of DN in an image can be usedin another way that might pay dividends for geolo~calinterpretation. That is by extending the methods of spec-tral pattern recognition using the concept of context.A particular pixel on a surface may be classified betterif there is some knowledge of the classification of itssurroundings. Contextual c.1asstfication at its simp.1estconsists of using the classification of other neighbour-ing pixels to weight the classification algorithm. Thecomputer performs a check of results within a specifiedradius in order to achieve this. Conceivably it couldbe impr.oved to take into account the context of otherassociated ciasses.. Thus a pixel, which could be eithergranite or sandstone on the basis of its spectral properties,would be more likely to be sandstone if it were near topixels identified unambiguously as limestone.

Any classification seeks to locate the boundariesbetween surface areas of different types and the extent ofareas of each class as accurately as possible. A means ofdoing this is image segmentation. It consists of creatingareas of several pixels that show spectral similarities.In a simple example a box of predetermined size passesover the image and tests the pixels falling within it forsimilarity according to the analyst's instructions. It is atechnique that finds a ready use in analysis of imagingspectrometer data, such as those from A VIRIS, wherethe spectra of individual pixels are compared with spec-tral libraries of diagnostic minerals, such as differenthydroxyl-bearing minerals (Plate 3.2). Either the spectraare matched directly or a kind of mathematical short-hand for the spectral shapes is devised to speed up thecomparisons. If they fall within predetermined boundsof similarity they are grouped together as members of aclass. The effect is to cause areas with similarity to spec-tral categories- to grow to their predetermined spectrallimits. By combining this approach with a means ofedge detection (Section 5.3) the growth of areas of eachclass can be stopped at what are hoped to be naturalboundaries in the scene. This approach is most success-ful in images where the surface categories are distinctand spectrally homogeneous, as in agricultural land.As emphasized in Section 5.6.1 geolo~cal classes areoften spectrally heterogeneous.

5.6.4 Change detection

The monitoring of changes at the Earth's surface by theuse of images recorded on several dates-multi-temporalanalysis.,-has been important for studies of vegetation,cultural and military features throughout the hif,tory of

remote sensing. With photl'graphic data it consists simplyof visual comparison. The advent of regular repetitivedigital imaging, first from meteorological satellites thenthe early Landsats, meant that more automated methodsbecame possible. Most applications have been for ana-lysis of the evolution of weather systems, crop rotationand diseases, changes in land-use patterns and manyother topics for which these satellites were designed.One of the two main methods of analysis is based ondetection of spectral changes in fixed surface materials.One example exploits the seasonal changes in vegeta-tion canopy, which can discriminate between differentspecies if the correct times for comparison are chosen.Another is detection of spatial change at the surface, suchas the replacement of agricultural land by buildings, orthe development of military installations. At its simplestthe comparison is achieved by flicking from an imagetaken at one time to that taken at another.

Early in the Landsat programme it became clear thatthere was considerable potential for exploiting imagestaken at different times of year in geological investiga-tions. As seasonal climates become more contrastedso lithologically controlled variations in soil moistureand vegetation become more variable during the year.Increasing soil moisture in bare areas results in lowerreflectance in all bands, whereas increasing vegetationcover increases infrared reflectance and diminishes thatin red and blue bands. The differences are most clearbetween dry- and wet-season images (Plate 5.10). Formost purposes geologists e"Ploit these differences inselecting an optimum time of year when geologicalinformation content is highest. In some climates the dryseason is best because wet-season images are maskedby vegetation, often with agricultural patterns. Othersshow geology better just after the wet season when soiland rock moisture varies most, according to porosity,and natural vegetation is strongly controlled by it.

Highlighting differences by using data from two ormore seasons can be a very powerful technique. As wellas expanding the variety of geologically related informa-tion it can show up geobotanical anomalies related todifferent kinds of geochemical stress and selective growthof plants. As the spectral effects may be at a maximumfor only short periods, dates of imagery need to bechosen carefully. Sometimes particularly severe weatherin one year is the only reason why such phenomena aredetectable, and comparison is made between severalyears as well as several seasons.

A special case of multi-temporal analysis of spectralchange is the use of diurnal variations in surface tem-perature to estimate the thermal inertia of the surface.This is considered separately in Chapter 6.

Short-lived geological and hydrological events canbe detected, monitored and assessed by multi.,temporalanalysis. Among these spatial changes are volcanism,

Digital Image Processing 157

slumping, floods, snow accumulation, glacial-.

, The data used are of the 'before and after'With Landsat data the minimum frequency of

-, ---,- ,

the SPOT system increases this frequency toonce every 4 days. For really vast events and those ofshort duration the TIROS-N /NOAA satellites producecoarse resolution data every 6 h.

The simplest means of automated change detectioninvolves the subtraction of data for one day from thatfor another. This can present problems, however. Unlessthe two images have nearly the same structure-similarvegetation and atmospheric conditions-the new datasets can be overwhelmed by nongeological differences.It is best applied to data from the same season fo~ differ-ent years. Another method depends on changes beingpoorly correlated with the rest of the data. They com-prise anomalies on plots of DN from the same bands fordifferent dates. Principal component analysis (Section 5.4)of selected bands for two or more dates casts the highlycorrelated data from the separate images into the firstand second components. The major differences resultingpurely from gross seasonal changes in vegetation appearin the second and third components. The unusual, small-scale differences are generally found in higher ordercomponents. Although the potential for becoming lostin a labyrinth of decorrelated data sets is high, carefulselection of components and their display can highlightsmall but important changes very proficiently.

Associated resources on the CD-ROM

Since Landsat-1 made multispectral, digital imageryavailable publicly and cheaply in 1972, the potential foradding value to 'raw' remotely sensed data was enorm-ous. There has been an explosion of software aimedat image enhancement and analyst.; over the last threedecades. Some packages have disappeared as com-petition sharpened, whereas others have emerged asnew operating systems made complex tasks more easilygrasped by inexpert users of computers. Competition,and the appearance of new kinds of data, such as digitalSAR, hyperspectral imagery and OEMs, has meant thatthe functionality of these packages has evolved rapidly.At the time of writing, this growth seems to have reacheda plateau. There has never been a more propitious timefor entering the image processing and analysis fray,as the speed, precision and cost of computers and usefulperipherals dwarf those when it all began to take off inthe 1970s. Equally, the explosion in data sources, bothcommercial and freely available in the public domain,means that geologists risk being marginalized if theyremain unskilled in this area.

The leading software vendors maintain lively Websites, where you can browse their products. The marketleaders are:

ERDAS-imagine and related software

http://www.erdas.comENVIhttp://www.rsinc.comERMapper ..http://www.ermapper.comPCIhttp://www.pci.comImage Analysthttp://www-nihon.intergraph.com/photogrammetry/rs/ianaly.aspTNTmipshttp://www.microimages.com

An excellent review of these packages is in Limp,F. W. 1999. Image processing software: system selectiondepends on user needs. Ceo World May, 36-46. Twomonthly magazines provide breaking news of softwareand associated hardware developments-EOM (EarthObservation Magazine http://www.eomonline.com) andMapping Awareness (http://www.geoplace.com).

Further reading

The image-processing techniques that I have outlined inChapter 5 are covered by a series of practical exercises,which use a professional image-processing and digital-mapping package (TNTlite) on the CD-ROM. You canaccess the text ffies for these by opening the HTML ffieMenu.htm in the folder Imint on the CD using yourWeb browser. More detailed information about instal-ling TNTlite is in Appendix C. As well as the data files(Imint\Exercises\Data) that the exercises use, you canimport other data into TNTlite, provided that they arewithin the size limits that the vendors have imposedfor commercial reasons. This means that designing newexercises is possible, and you can use TNTlite for smallresearch projects.

My choice of TNTlite as a 'hands-on' resource is becauseMicroImages lnc (its designer and vendor) is the onlysoftware company that offers a size-limited, but fullyfunctional version of its product as freeware for unlimitedredistribution. Other than that, neither I nor BlackwellScience imply any endorsement of Microimages products.

Section 5.6.2 Spetral mapping is condensed from TNTlite'Getting Started' booklets Introduction to HyperspectralImaging, and Analysing Hyperspectral Images.

The following references form a framework for ex-tending your understanding of digital image processing

techniques.

158 Chapter 5

Anuta, P .E. (1977) Computer-assisted analysis techniques forremote sensing data interpretation. Geophysks 42,468-481.

Berhe, S.M. & Rothery, D.A. (1986) Interactive proce~sing ofsatellite images for structural and lithological mapping innorth-east Africa. Geological Magazine 123, 393-403.

Blodget, H.W. & Brown, G.F.(1982) Geological mapping by useof computet-enhanced imagery in western Saudi Arabia. U.S.Geological Survey Professional Paper 1153.

Buchanan, M.D. &Pendgrass,. R. (1980) Digital image process-ing: can intensity, hue and saturation replace red, green andblue? Electro-optical Systems Design 12, 29-36.

Canas, A.A.D. & Barnett, M.E. (1985) The generation andinterpretation of false-colour principal component images.International Journal of Remote Sensing 6, 867-881.

Castleman, K.R. (1977) Digital ImRge Processing. Prentice Hall,Englewood Cliffs, New Jersey.

Chavez, P.S. (1988) An improved dark object subtraction tech-nique for atmospheric scattering correction of multispectraldata. Remote Sensing of Environment 24, 459-479.

Chavez, P .S. (1989) Radiometric calibration of Landsat ThematicMapper multispectral images. Photogrammetric EngineeringRemote Sensing 55, 1285-1294.

Chavez, P.S. & Bauer, B. (1982) An automatic optimumkernel-size selection technique for edge enhancement. RemoteSensing of Environment 12, 23-38.

Chavez, P.S., Berlin, G.L. & Sowers, L.B. (1982) Statisticalmethod for selecting Landsat MSS ratios. Journal of AppliedPhotographic Engineering 8, 23-30.

Condit, C.D. & Chavez, P .S. (1979) Basic concepts of computer-ized digital image processing for geologists. U.S. GeologicalSurvey Bulletin 1462.

Crippen, R.E. (1988) The dangers of underestimating theimportance of data adjustment in band ratioing. InternationalJournal of Remote Sensing 9,767-776.

Davis, P.A. & Berlin, G.L. (1989) Rock discrimination in thecomplex geological environment of Jabal Salma, Saudi Arabia,using Landsat Thematic Mapper data. PhotogrammetricEngineering and Remote Sensing 55,1147-1160.

De Souza Filho, C.R., Denniss, A.M., Drury, S.A., Rothery, D.A.& Carlton, R. W. (1996) Restoration of noise-corrupted opticalFuyo-1 (JERS-1) data using frequency domain techniques.Photogrammetric Engineering and Remote Sensing 62, 1037-1047.

Drury, S.A. (1985) Applications of digital image enhancementin regional tectonic mapping of South India. Proceedings of the18th International Symposium on Remote Sensing of Environ-ment, Paris 1895-1904.

Drury, S.A. (1986) Remote sensing of geological structurein temperate agricultural terrains. Geological Magazine 123,113-121.

Drury, S.A. & Hunt, G.A. (1988) Remote sensing of lateritizedArchaean greenstone terrain: Marshall Pool area, north-eastern Yilgarn Block, Western Australia. PhotogrammetricEngineering and Remote Sensing 54, 1717-1725.

Drury, S.A. & Hunt, G.A. (1989) Geological uses of remotelysensed reflected and emitted data of lateritized Archaeanterrain in Western Australia. International Journal of RemoteSensing 10, 475-497.

Dykstra, J.D. & Birnie, R.W. (1979) Reconnaissance geologicalmapping in Chagai Hills, Baluchistan, Pakistan, by computerprocessing of Landsat data. Bulletin, American Association ofPetroleum Geologists 63, 1490-1503.

Evans, D. (1988) Multise~or classification of sedimentaryrocks. Remote Sensing of Environm~t 25, 129-144.

Evans, D. & Lang, H. (1985) Techniques for multisensor imageanalysis. Proceedings of the 18th International Symposium onRemote Sensing of Environment, Paris.

Fabbri, A.G. (1984). Image Processing of Geological Data. VanNostrand Reinholt, New York.

Gillespie, A.R. (1980) Digital techniques of image enhancement..In Remote Sensing in Geology (edsB.S. Siegal & A.R. Gillespie),

pp. 139-226. Wiley, New York.Gillespie, A.R., Kahle, A.B. & Walker,R.E. (1986) Colour enhance-

ment of highly correlated images: I~ecorrelation and HSIcontrast stretches. Remote Sensing of Environment 20, 209-235.

Gillespie, A.R., Kahle, A.B. & Walker, R.E. (1986) Colourenhancement of highly correlated images: lI-channel ratiosand chromaticity transformation techniques. Remote Sensingof Environment 22, 343-365.

Gong, P. (1996) Integrated analysis of spatial data from multi-ple sources: using evidential reasoning and artificial neural-network techniques for geological mapping. PhotogrammetricEngineering and Remote Sensing 62, 512-523.

,Gonzalez, R.C. & Wintz, P. (1977) Digital Image Processing.Addison-Wesley, Reading, Massachusetts. -

Grasso, D.N. (1993) Application of the IHS colour transforma-tion for 1 : 24 OOO-scale geological mapping: a low-cost SPOTalternative. Photogrammetric Engineering and Remote Sensing59,73-80.

Hardcastle, K.C. (1995) Photolineament factor: a new computer-aided method for remotely sensing the degree to whichbedrock is fractured. Photogrammetric Engineering and RemoteSensing 61, 739-747.

Hord, R.M. (1982) Digital Image Processing of Remotely SensedData. Academic Press, New York.

Hunt, G.A. (1988) Improvements in the forward and inverseprincipal component transformations for geological mappingin a semiarid terrain. Proceedings, IGARSS Symposium 1988,Remote Sensing: Moving towards the 21st Century, pp. 1061-1062. Special Publication 284, European Space Agency, Paris.

Jensen, J.R. (1986) Introductory Digital Image Processing. PrenticeHall, Englewood Cliffs, New Jersey.

Knepper, D.H. Jr & Raines, G.L. (1985) Determining stretchparameters for lithologic discrimination on Landsat MSS bandratio images. Photogrammetric Engineering and Remote Sensing51,63-70.

Lang, H.R., Adams, S.L., Conel, J.E., McGuffie, B.A., Paylor,E.D. & Walker, R.E. (1987) Multispectral remote sensing asa stratigraphic and structural tool, Wind River Basin andBig Horn River Basin areas, Wyoming. Bulletin, AmericanAssociation of Petroleum Geologists 71, 389-402.

Mah, A., Taylor, G.R., Lennox, P. & Balla, L. (1995) Lineamentanalysis of Landsat Thematic Mapper images, NorthernTerritory, Australia. Photogrammetric Engineering and RemoteSensing 61, 761-773.

Mather, P.M. (1987) Computer Processing of Remotely SensedImages: an Introduction. Wiley, New York.

Mather, P.M., Tso, B. & Koch, M. (1998) An evaluation ofLandsat TM spectral data and SAR-derived textural informa-tion for lithological discrimination in the Red Sea Hills, Sudan.International Journal of Remote Sensing 19, 587-604.

Moore, G.K. & Waltz, F.A. (1983) Objective procedures forlineament enhancement and extraction. PhotogrammetricEngineering and Remote Sensing 49, 641-647.

Digit~ Thermal Images 161

Podwysocki, M.H., Segal, D.B. & Abrams, M.J. (1983) Use ofmultispectral scanner images for assessment of hydro-thermal alteration in the Marysvale, Utah, mining area.Economic Geology 78, 675-687.

Pratt, W.K. (1978) Digital Image Processing. Wiley, NewY ork.Renz, Andrew, N. (ed.) (1983) Remote Sensing for the Earth

Sciences. Manual of Remote Sensing, 3rd edn, Vol. 3. Wiley,New York.

Rothery, D.A. (1984) Reflectances of ophiolite rocks in the LandsatMSS bands: relevance to lithological mapping by remote sens-ing. Journal of the Geological Society of London 141, 933-939.

Rothery, D.A. (1985) Interactive processing of satellite imagesfor geological interpretation. Geological Magazine 122, 57-63.

Rothery, D.A. & Francis, P.W. (1985) A remote sensing study ofa sector collapse volcano. Proceedings of the 18th InternationalSymposium on Remote Sensing of Environment, Paris.

Rowan, L.C. & Bowers, T.L. (1995) Analysis of linear featuresmapped in Landsat Thematic Mapper and side-looking air-borne radar images of the Reno 10 x 20 Quadrangle, Nevadaand California: implications for mineral res~urce studies.Photogrammetric Engineering and Remote Sensing 61, 749-773.

Rowan, L.C., Goetz, A.F.H. & Abott, E. (1987) Analysis ofShuttle multispectral infrared radiometer measurement ofthe Western Saudi Arabian shield. Geophysics 52, 907-923.

Rubin, T.D. (1993) Spectralmappingwithima~tance is difficultPhotogrammetric Engineering and Remote Sensl~ Kirchhoff's

Ruiz-Arrnenta, J.R. & Prol-Ledesma, R.M. (1998) T~ing Equa-enhancing the spectral response of hydrothermal,"minerals in Thematic Mapper images of Central 1\,.,International Journal of Remote Sensing 19, 1981-2000. -(~.9)

Schowengerdt, R.A. (1983) Techniques for Image Processing a~Classification in Remote Sensing. Academic Press,N ew York.

Siegal, B.S. & Abrams, M.J. (1976) Geological mapping usingLandsat data. Photogrammetric Engineering and Remote Sensing42,325-337. .

Siegrist, A.W, (1980) Optimum spectral bands for rock dis-crimination. Photogrammetric Engineering and Remote Sensing46,1207-1215.

Siizen, M.L. & Toprak, V. (1998) Filtering of satellite images ingeological lineament analyses: an application to a fault zonein Central Turkey. International Journal of Remote Sensing 19,1101-1114.

Vane, G., Duval, J.E. & Wellman, J.B. (1993) Imaging spectros-copy of the Earth and other Solar System bodies. In: RemoteGeochemical Analysis: Elemental and Mineralogic Composition(eds C.M. Pieters & P.A.J. Englert), pp. 121-143. CambridgeUniversity Press, Cambridge.

Digital

Image Processing 159

Podwysocki, M.H., Segal, D.B. & Abrams, M.J. (1983) Use ofmultispectral scanner images for assessment of hydro-thermal alteration in the Marysvale, Utah, mining area.Economic Geology 78, 675-687.

Pratt, W.K. (1978) Digital Image Processing .Wiley, NewY ork.Renz, Andrew, N. (ed.) (1983) Remote Sensing for the Earth

Sciences. Manual of Remote Sensing, 3rd edn, Vol. 3. Wiley,New York.Rothery,D.A.

(1984) Reflectances of ophiolite rocks in the LandsatMSS bands: relevance to lithological mapping by remote sens-ing. Journal of the Geological Society of London 141, 933-939.

Rothery, D.A. (1985) Interactive processing of satellite imagesfor geological interpretation. Geological Magazine 122, 57-63.

Rothery, D.A. & Francis, P. W. (1985) A remote sensing study ofa sector collapse volcano. Proceedings of the 18th InternationalSymposium on Remote Sensing of Environment, Paris.

Rowan, L.C. & Bowers, T.L. (1995) Analysis of linear featuresmapped in Landsat Thematic Mapper and side-looking air-borne radar images of the Reno 10 x 20 Quadran~, Nevadaand California: implications for mineral resource studies.Photogrammetric Engineering and Remote Sensing 61, 749-773.

Rowan, L.C., Goetz, A.F.H. & Abott, E. (1987) Analysis ofShuttle multispectral infrared radiometer measurement ofthe Western Saudi Arabian shield. Geophysics 52, 907-923.

Rubin, T.D. (1993) Spectral mapping with imaging spectrometers.Photogratrimetric Engineering and Remote Sensing 59,215-220.

Ruiz-Armenta, J.R. & Prol-Ledesma, R.M. (1998) Techniques forenhancing the spectral response of hydrothermal alterationminerals in Thematic Mapper images of Central Mexico.International Journal of Remote Sensing 19,1981-2000.

Schowengerdt, R.A. (1983) Techniques for Image Processing andClassification in Remote Sensing. Academic Press, New York.

Siegal, B.S. & Abrams, M.J. (1976) Geological mapping usingLandsat data. Photogrammetric Engineering and Remote Sensing42,325-337.

Siegrist, A.W. (1980) Optimum spectral bands for rock dis-crimination. Photogrammetric Engineering and Remote Sensing46,1207-1215.

Suzen, M.L. & Toprak, V. (1998) Filtering of satellite images ingeological lineament analyses: an application to a fault zonein Central Turkey. International Journal of Remote Sensing 19,1101-1114.

Vane, G., Duval, J.E. & Wellman, J.B. (1993) Imaging spectros-copy of the Earth and other Solar System bodies. In: RemoteGeochemical Analysis: Elemental and Mineralogic Composition(eds C.M. Pieters & P.A.J. Englert), pp. 121-143. CambridgeUniversity Press, Cambridge.

6reflected region of the EM spectrum. As a result thermalimages are often poorly correlated with those from thevisible and near-infrared range. They provide an extradimension for the discrimination of different rocks, soilsand vegetation. They can be used in three main ways forinvestigating geological features. First, they may beinterpreted in just the same way as photographs, givenan understanding of the thermal properties of the sur-face. Second, images of the same area taken at differenttimes of day give semiquantitative information abouthow different materials respond to heating and cooling.This is the basis of estimating variations in thermalinertia (Section 1.3.2). The third application dependson quantum mechanics and laboratory investigations,which show that troughs in the emission spectra ofminerals, related to Si-O bond stFetching and otherphenomena, occur at different wavelengths depend-ing on chemical composition and molecular structure(Section 1.3.2). It relies, therefore, on gathering thermaldata in narrow wavebands to highlight these distinctivefeatures.

6.1 What a thermal image shows

The variations in tone or DN in a thermal image are mea-sures of the surface's radiant emission (Section 1.3.2). Inthe 8-14 Jlm waveband natural materials are efficientabsorbers of solar radiation and very little is reflected(Fig. 6.1). If the surface was a blackbody the image toneor DN would be an accurate measure of the true surfacetemperature. Radiant emission is controlled by a pro-perty of matter known as emissivity (Section 1.3.2). Ablackbody has an emissivity of 1 whereas all naturalmaterials have values less than 1. Most natural materialsare selective emitters rather than true greybodies andtheir emissivity varies with wavelength (Section 1.3.2).The temperature derived from thermal infrared meas-urements by applying the Stefan-Boltzmann Law-the apparent blackbody temperature of the surface-istherefore always less than the true surface temperature.

Materials with high emissivity absorb and emitlarger proportions of incident energy and heat energytransmitted into them than those with low emissivity.Whether the temperature of a surface increases ordecreases depends on increase or decrease in the heatcontent of the surface material. The law of energy con-servation determines the heat balance at the surface.Heat flows to and from the surface by many differentmeans. Geothermal heat flows from below by conduction

Beyond a wavelength of about 4 J.Lm, energy from theEarth's surface is almost exclusively the result of radiantemission from natural materials. Stefan-Boltzmann'sLaw relates the total energy emission to the fourth powerof a material's absolute temperature. Wien's Displace-ment Law describes the spread of emission among arange of wavelengths, rising to a maximum determinedby absolute temperature (Section 1.2, Fig. 1.3). Both lawsrelate to the ideal behaviour of blackbodies-perfectabsorbers and emitters of radiant energy. Natural mater-ials are not blackbodies, and they only approximate theserelationships. The Earth's average surface temperatureof 300 K means that its energy emission peaks at 9.7 J.Lm.Gases in the atmosphere absorb energy in distinct wave-bands, so that only a small proportion of the emittedspectrum is transmitted. Incidentally, this lies behindthe so-called 'greenhouse' effect of gases such as water,carbon dioxide and methane, that slow down heat lossby radiation and so allow the mean surface tempera-ture to be 33 K above that likely if the planet had noatmosphere. The main atmospheric window for thermalremote sensing is between 8 and 14 J.Lm (Fig. 1.5).

Thermal infrared imaging uses devices similar tothose suitable for the reflected region (Sections 3.3 and3.4), based on cooled solid-state detectors (Fig. 3.9).Generally the detectors give a lower signal-to-noise ratiothan those used for shorter wavelengths, resulting incoarser spatial and spectral resolution. The earliest ther-mal images were of the analogue type, signals from aline scanner modulating the brightness along a line on acathode-ray screen to build up an image from successivelines on film moving across the screen. Digital thermalimages from modern systems can be enhanced and ana-lysed by the image-processing methods described inChapter 5.

One advantage of thermal infrared images is that theEarth's surface is always far above absolute zero, andemits thermal radiation equally well at night as it doesduring the day. The possibility of night-time surveil-lance drove the invention of thermal imaging devices bymilitary scientists. Fortuitously this adds to the possibil-itiesfor geological interpretation. The reason for this isthat different materials lose heat at different rates duringthe night, and show up in quite different ways comparedwith their daytime appearance. The only limitations arethe degree of cloud cover, which thermal radiation c~n~not penetrate, and the atmospheric window of reducedattenuation.

The thermal properties of surface materials a're verydifferent from those that control energy transfer in the

160

lected

"'-"',/ . d'/ emltte .

II

II

II

and the two are interchanieable.to measure and 1Law after its initiator, can be transformedtion (1.8) to:

eA=l-PA (l.9)

This theoretical relationship is amply illustrated by amotorist wearing shorts sitting on a black seat on a hotday. A white seat is distinctly more comfortable. Theblack surface has low reflectance and Ngh absorptance-it heats up very efficiently-and vice versa. Most solarenergy that penetrates the atmosphere is carried byvisible and near-infrared radiation (Fig. I.Sa). Heating ofthe surface is therefore the result mainly of absorption ofthese wavelengths and their transformation to vibrationof the material's component molecules. Solar radiationin the mid-infrared region is completely absorbed byany surface, but contributes relatively little to heating. Amajor factor in the heat balance of a surface therefore isits ability to reflect radiation over the whole visible andnear-infrared range. This is the albedo of the surface,which controls its overall reflectivity.

To appreciate the surface heat balance and how thesurface temperature of an area changes with tim~ meansconsidering the heat adde~ to and removed from thesurface. As geothermal heat is usually constant and lowcompared with the heating effect of solar radiation, it canbe disregarded outside volcanic areas. The amount ofsolar radiation at all wavelengths that is available to heatthe surface depends on: ,-1 the elevation of the Sun above the horizon-a functionof latitude, time pf year and time of day;2 atmospheric conditions, such as cloudiness, humidityand air densii;y-a function of weather and altitude; .3 the' topography in relation to solar azimuth andelevation-slopes facing the Sun intercept more radi-ation than those facing away. If slopes are steep enoughor the Sun is low enough there may be shadows thatreceive only radiation from the sky.

Three main processes contr01 removal of heat from theEarth's surface. .1 Radiation of thermal infrared.2 Heat transfer by convection-this depends on the move~ment of heated air, either by convection celis or wind.Cool air is heated by a hot surface, but equally a coldsQrface may receive heat from warm air. The amount ofheat transferred by convective means depends on tem-perature gradients in the atmosphere, wind speed andturbulence" the area of surface exposed-,-this depends onthe surface roughness-and on the density of the air andhence altitude.3 Transfer of latent heat. Evaporation of water from thesurface involves adding the latent heat of vaporizationfrom the surface to ~he water to form water vapour.This depends on the same variables as convective heat

12 I refl~ 1° 1 '

~ 8QI 6 1u f

.~ 4 1"U .~ I

L I

'.

"'.I

'

---I.-/: .--, , , , , ., , , , , ,

0 4 8 12 16 20 24

Wavelength (IJm)

Fig. 6.1 The radiance of electromagnetic (EM) energyreflected by the Earth's surface falls rapidly towards zerobeyond

about 2.5 .urn. Absorption is virtually complete.Energy received from the Earth at wavelengths longerthan this is almost entirely that emitted because of thetemperature

of the surface.

through rock and soil from the Earth's interior. Exceptin volcanic areas, this component is more or less con-stant over very large areas and far below the additionof solar energy. Solar radiation transmitted through theatmosphere is absorbed at all wavelengths, including aproportion in the visible and near-infrared region. Down-ward radiation contains a component of solar energyoriginally absorbed by the atmosphere at wavelengthsbetween the atmospheric windows, but re-emitted atlonger wavelengths. Energy is lost from the surfaceby an even more complex set of mechanisms. Mostimportant is that radiating away from the surface, butsignificant losses involve the heating of air or evapora-tion of water involving latent heat at the ground surfaceand its movement by wind.

Radiative heat transfer dominates the surface heatbalance. The principle of conservation of energy deter-mines that for any radiation-matter interaction, the incid-ent radiant flux at one wavelength (EJA is distributedbetween reflection (ER)A' absorption (E A)A and transmis-sion (ET)A by the material involved:

(EJA = (ER)A + (EA)A+ (ET)A (L5)

Dividing Equation 0.5) throughout by (EJA producesan expression allowing the spectral properties of thematerial to be d~fined in terms of the ratios (ER/EJA(EA/EJA and (ET/EJA' which are the spectral reflectance(PA), absorptance (aA) and transmittance ("A)' respect-ively, so.that:

PA+(XA+"A=l {1.7)

The .vast bulk of geological materials are opaque andtransmittance is zero, so that Equation 0.7) reduces to:

PA+aA=l 0.8)

Because good emitters are also good absorbers of radi-ation their emittances are equal to their absorptances,

~q","('Ii,"

i~q",, °0 ~/:," /r- "8~"

""

162 Chapter 6

The rate of change iQ temperature in a material iscontrolled by its thermal diffusivity (IC). This combinesthermal capacity, thermal conductivity and density inthe relationship:

IC= K/cp (6.1)

Thermal diffusivity is the most convenient measure of amaterial's ability to transfer heat from the surface to theinterior during heating and vice versa.

Thermal inertia (P) is a measure of the resistance of amaterial to a change in its temperature in response to achange in the temperature of its surroundings, given by:

P = VPC1( (6.2)

As the Earth's ambient temperature varies fairly regu-larly throughout the 24 h.diurnal cycle, thermal inertia isthe most appropriate measure of the combined thermalproperties of. surface materials. It is also relatively simpleto estimate from digital thermal images captured at dif-ferent times of day (Section 6.3.1). Values for thermalinertia together with other parameters measured forvarious surface materials in the laboratory are given inTable 6.1.

How the layer of soil and rock immediately beneaththe surface heats and cools is a complex interplay betweenall these parameters. Heat flows from hotter to coolerregions in this layer at a rate deterni.ined by thermal con-ductivity. Daytime heating of the surface causes heat tobe conducted downwards. Once surface cooling beginsin the afternoon the heat stored at depth flows back tothe surface. During the diurnal cycle the temperatureproffie in the top layer therefore changes. The shape of

transfer, together with the moisture content of surfacematerials and the atmospheric humidity. It is com-plicated by the transpiration from any vegetation cover.The reverse is dew formation when latent heat is addedto the surface as water condenses on it.

~learly, interpretation of thermal images is a weightymatter. As well as the addition and loss of heat at the sur-face, the heat balance depends on the movement of heatjust below the surface, how that changes the temperatureof rocks and soils, and the time factor involved. In addi-tion to the albedo and emissivity of the surface, there areother factors at work.

The ability of a volume of material to store heat isexpressed by its thermal capacity (c). This is the amountof heat required to raise the temperature of a unit vol-ume of material by 1°C. Thermal capacity is the productof specific heat and density (p) and is e~pressed in heatunits per unit volume per degree Celsius. Specific heatvaries very little for rocks and soils, so density differ-ences have most effect on their ability to store heat. Thehigher the density the more heat can be absorbed by thesurface with only a 1 °Cris~ in temperature. Low-densitymaterials undergo a greater change in temperature asheat is added or lost.

An expression of the ability of a material to pass heatby conduction is its thermal conductivity (K), defined asthe amount of heat passing through unit area for a unitdistance in unit time for a temperature difference of 1°C.Rocks and soils are poor conductors of heat, but varyover an order of magnitude. As water and air have evenlower values of K, their presence in porous r04s extendsthe I,"ange further.

Table 6.1 Thermal properties ofrocks and water measuredexperimentally at 293 K.

Thermal Thermalcapacity i C diffusivity, x

Density", Okg-lK-l) (m2s-1)(kg m-3) (x 10-2) (x 10-6)

Thermalinertia, P

(J m-2 S-I/2 K"

(x 10-3)

Thermalconductivity, K(J m-ls-l K-l)Material

320030002800260010001800240027002700280023002500260025002100180017001000

4.62.5-2.1

3.10.31.22.65.D.2.32.1i.82.05.05.02.5,Q.6

1.30~5

.~

8.47.18.46.76.78.49.67.18.87.17.17.17.58.08.4

10.014.742.3

1.71.20.91.60.40.81.32.61.01.10.81.12.61.31.40.30.50.1

3.52.32.22.20.4.J.32,63.12.32.11.41.93.12,32.11.01.81:5

Peridotite"

GabbroBasalt'GranitePumice ~dry)IgnimbriteSerpentiniteQuart;ziteMarbleSlateShaleLimestoneDolomiteSandstoneGravelSandy soil .Clay soilWater

Thermal Images 1'63

80

(a)

60

40

20

'0.01

dawnsunset

30 L

Fig. 6.2 The results of modelling the thermal behaviour of athin layer at the Earth's surface show that temperature varieswith depth, The profiles of temperature against depth changeaccording to the time of day down to a depth of about 30 CIll.The greatest variation ocCurs in the toplO cIll or so. Bxaminingthe curves in a time sequence shows that a heating 'wave'rises through the ground from dawn to noon. Cooling thenprogresses downwards, becoming ~lower during the night.The curves were modelled for relatively low thermal inertiaand density, so that they represent extreme rates of heatingand cooling. After Kahle (1980).

~ (b)

40

0.120

0

12.00 18.00 24.00 06.00 ..~- .-.

Time (hrs)

Fig. 6.3 The diurnal variations in temperature of surfacesunderlain by ma~e~als Qf differe~~thermal iI1ertia (a) showdifferent rates of heating and cooling. As a result, they crossone another about2h after dawn and 1 h before sunset.Temperature-time curves for similar materials With differentalbedos (b) are approximately parallel. Those With low albedo(dark surfaces) consistently have higher surface temperaturesthan those with higher albedos (light surfaces). Both sets ofcurves result from mathematical models rather than actualmeasurements. After Watson (1975)..

lL.UO

the profile and how it changes with time are governed bythe surface heating and the various thermal propertiesof the materials involved.

Figure 6.2 shows a time sequence of temperatureprofiles modelled for varying surface heat input anda typical set of thermal parameters for a uniform,near-surface layer. The profiles show how temperaturechanges down the layer as a heating wave from dawnto the hottest part of the day, and as a wave of coolingas solar heating wanes after noon. Below about. 30 cm,temperature stays the same throughout the daily cycl~.For a given material, this depth varies over the courseof a year because of seasonal variations in solar eleva-tion and weather conditions. The shapes of such profilesvary with the values of the different thermal parametersof the near-surface materials. An increase in conduct-ivity allows the heating wave to penetrate further, sodistributing the solar energy through a greater volume.As a result, surface temperature does not rise so highor so quickly and changes more slowly with depth.An increase in thermal capacity-controlled mainly bydensity-means that more heat is required to raise thetemperature of the topmost part of the layer. Less is leftto penetrate to depth. Temperature at the surface doesnot rise so high or so quickly, but changes more rapidlywith depth. Both these parameters are combined as aproduct in thermal inertia (Equation 6.2), and so conspireto magnify these two effects as thermal inertia increases.One result of this modelling is that although the shapesof profiles change for materials with different conductivit-ies and thermal capacities, if they have the same thermal

inertia their surface temperatures at any time are thesame. Because of this, thermal inertia can be estimatedfrom digital images with a fair degree of confidence.

How surface temperatures vary through the diurnalheating and cooling cycle (Section 1.3.2, Fig. 1.14) dependson the thermal properties of the materials beneath.Figure 6.3 shows diurnal variations in surface t~mper-ature calculated for different values of two importantthermal..parameters-:thermal inertia and albedo-withother parameters held constant. The family of curvesfor different thermal inertias show clear thermal cross-overs. The reason for them is that a material with a lowresistance to change in temperature-low thermal inertia-heats up quickly to a high temperature during the dayand cools in a similar fashion. Conversely, ri1aterialswith high thermal inertia heat and cool more slowlyto less extreme temperatures. Metals and any materials

164 Chapter6

12.00 18.00 24.00 06.00 12.00

Time (hrs)

Fig. 6.4 The shapes of diurnal temperature curves of differentnatural materials depend largely on their thermal inertias, towhich moisture content makes an important contribution.The actual temperatures that they achieve depend mainlyon albedo and, toa lesser extent, on geothermal heat flow.The curves are merely sketches of relative behaviour.

containing water (damp soil and vegetation) experiencethe least diurnal change in surface temperature, becausethey have very high thermal inertias. Although the curvesfor different albedos have different positions on the graph,they are essentially the same shape. Curves derived forall other thermal parameters are similar to those foralbedo. This confirms the importance of thermal inertiain mid-infrared remote sensing. Measuring the differ.,ence between day- and night-time surface temperaturesgives a clue to thermal inertia. A large difference impliesa low thermal inertia and a small variation suggests theconverse. For materials with the same thermal inertiabut very different values of all other thermal parametersthis difference will not vary much.

Figure 6.4 is a sketch of diurnal temperature curvesfor a number of surface materials.. It allows us to pre-dict relative grey levels relating to these materials onthermal infrared images taken at different times of day(Section 6.2). It is also useful in selecting the times whengreatest contrast exists between different materials.

6.2 Qualitative interpretation ofthermal images

This section discusses day and night-time images ofenergy emitted in the 8-14 ~m waveband, in eitheranalogue or digital format, first as grey-tone images, andthen combined with other data as false colour images.

texture, pattern, shape. and context. Unlike images ofshorter wavelength radiation, the tone of a thermal im-age expresses surface temperature and not reflectance.Cooler areas have dark tones and warmer areas appearlight. The greatest and most common problem in inter-preting images of reflected radiation occurs in areas ofuniform albedo. They are typical of terrains with a thinskin of soil, vegetation or desert varnish. In such cases,thermal images are useful, because they should revealdifferences in surface temperature related to variousthermal properties to depths of 30 cm or so.

During daylight hours, as well as the differential heat-ing effect of direct sunlight on materials with differentthermal properties, images are affected by topography.Areas in shadow are cool and appear dark in much thesame way as they do in more conventional images. Sunlitslopes are heated differentially according to their orienta-tion. South-facing slopes are heated most strongly in theNorthern Hemisphere. Exactly the same enhancementof topographic features normal to the Sun's azimuth,and suppression of parallel features occur as they do inimages of shorter wavelengths. Daytime images expresstopography and geological structure very well, but indoing so obscure variations resulting from differentthermal properties of the surface.

After sunset the effects of differential solar heatingremain for a time. Loss of heat by radiation is controlledby both thermal inertia and surface temperature. Formaterials with the same thermal inertia, the hottestsurfaces cool most quickly",and topographic effects aregradually removed. As Figs 6.3 and 6;4 show, surfacetemperatures at night become stable, particularly in thefew hours before dawn. During the day temperaturefluctuates rapidly as the Sun rises to its zenith and thenbegins to set. Because topographic effects are suppressedand temperatures are stable, night-time images arebetter for lithological interpretation. However, bothtypes complement each other, when available. Figure 6.5shows the different attributes of day and night-timeimages of the same area. As the properties that controlthe temperature of the surface are complex and unfamil-iar, interpretation is best achieved with as much groundcontrol as possible. The identity of some features, suchas water bodies, can be suggested by their shape andimage texture.

Planning the time of day to acquire images with thegreatest information content relies very much on thenature of the problem. Figures 6.3 and 6.4 reveal thatthere are two brief periods during the day, 2-3 h beforesunset and 2-3 h after dawn, when no matter what thevariations in thermal inertia of materials surface tem-perature should be quite uniform, according to theory.These times of thermal cross-over are unsuitable formany applications. The same figures show that max-imum temperature contrasts prevail during the hottest

6.2.1 Grey-tone thermal infrared images

Single-channel thermal infrared datil produce grey-toneim~ges- Like those -of visible and near-infrared wave-bands, the main elements of interpretation rely on tone,

Thermal Images 165

Fig. 6.5 The top image shows the emitted infrared energy at1100 hours from an area in the Front Range ofColorpdo, USA.The effect of solar shadowing is clear and highlights the localtopography. The lower image is of the same area, but wasacquired at 0400 hours, when surface temperatures are verystable. There is virtually no topographic effect and the imageshows clear distinction between warm (bright) areas of dryoutcrop and soil, and darker areas of cool damp soil. Thebright patches on both images are small ponds. North is tothe left in both cases and the scale is about 1 : 100000.Courtesy of Ken Watson, US Geological Survey, Denver.

part of the day and before dawn. These times are thosenormally chosen for thermal imaging.

Dry rocks and soils have relatively low thermal inertias,whereas vegetation, standing water and wet soils havehigh values (Fig. 6.4). Merely to discriminate betweenthese types of surface cover should be possible usingday or night. images. The effects of topography duringdaytime help to delineate geological structures, but.hinder lithological discrimination. Although the contrastin surface temperatures over rocks of different thermalinertias is lower during the night, the suppression oftopographic effects means that night images are best fordiscrimination.

Variations in density affect the thermal inertia ofdry rocks strongly (Table 6.1). A dense crystalline rock,such as peridotite, has a high thermal inertia. Its surfacetemperature remains fairly stable compared with lessdense rocks, such as granites. The porosity of sedimentsand pyroclastic volcanic rocks mainly determines densityand thermal inertia. A highly porous rock, such as pumice,displays rapid diurnal variations in temperature becauseof its low thermal inertia. Less extreme differences helpdistinguish between, for instance, sandstones with vari-ous degrees of cementation. The mineralogical com-position of rocks also contributes to thermal inertiabecause of differences in thermal conductivity. The car-bonates calcite and dolomite provide an interes.ting case.Dolomite is both denser and more conductive than

calcite, the combined fifect being to give dolomites athermal inertia roughly twice as h;gh as those in lime-stones. Although the two carbonate rock types appearidentical on shorter wavelength images, they are con-trasted on thermal images (Fig. 6.6). As quartz has a veryhigh conductivity compared with clay minerals thereshould be a good contrast between otherwise similarmudstones and siltstones.

These observations suggest that thermal images aidlithological discrimination where dry rocks and soilsare near to the surface. Thermal inertia, however, doesnot entirely control the actual surface temperatures andimage tones of different rocks. The other importantfactor is the interplay between albedo, emissivity andabsorptivity. If two rocks have identical thermal inertia,but one has a higher albedo, areas underlain by thatrock will show the lower temperatures and darker tonesat any time of day. As rocks with other combinationsof thermal inertia and albedo may well achieve sin'lilartemperatures, there is ample opportunity for confusionin only one image. The rocks in Fig. 6.6 all have similaralbedo. Figure 6.7 shows examples of day and nightthermal images from a dry terrain, where albedo variesmarkedly.

In humid terrains complications relating to watermake their presence felt. The diurnal temperature curvefor standing water on Fig. 6.4 does not have much to dowith the thermal inertia of water, which in thelaboratoryis of the same order as those of rocks. The apparentlyhigh value in nature is caus~d by the fact that water is aliquid. It convects when temperature varies with depth.During the daytime the surface water in a lake achievesthe highest temperature. This cools at night to be replacedby warmer water rising from below, thus stabilizingthe surface temperature to some extent. This effect ismagnified because water has a very high thermal capa-city and therefore can store heat very efficiently. The 11,eteffect, as shown by Fig. 6.4 and illustrated by Fig. 6.6(stream at top right), is that relative to dry rock and soilwater appears cool (dark) during the day and warm(light) at night.

Vegetation introduces other complexities and anom-alies. Because of its high water content green deciduousvegetation stores heat efficiently and appears warm onday images. During the night, water vapour is transpiredfrom the leaves thereby shedding latent heat. This givesa relatively cool signature (Fig. 6.6). 0verall the diurnaltemperature pattern is similar to that of water (Fig. 6.4).Curiously, the composite effect of the needles on conifer-ous trees is to simulate a blackbody. This results in suchvegetation having a brighter night-time signature thanthat during the day. Dead vegetation, such as grassesor agricultural stubble, has a low water content and ahigh content of air, both mstems and trapped betweenthe foliage. It forms such a good insulating layer that

166 Chapter6

(a)

Fig. 6.6 The aerial photograph (a) of an area in Oklahoma,.USA, shows mapped boundaries between limestone (L),dolomite (D) and granite (G). Apart from variations in texture,there is little to distinguish between the three rock tYpes, andt~ dplomite appears identical to the granite. A thermal imageacquired just before dawn (b) reveals contrasts between thethree rock types with similar albedo, which are the result

of different thermal inertias. The high thermal inertia of thedolomite ensures that it remains warm during the night,whereas the limestone and granite, with lower thermalinertia, lose more energy and appear darker. Irregular darkpatches in the dolomite represent wet soils in dry valleys.Courtesy of Ken Watson, US Geological Survey, Denver.

low resolution are also available from a variety of satel-lites. These include images from the Landsat TM with aresolution of 120 m. They are acquired at 0930 hourslocal time, which is close to the cross-over point onFig. 6.3(a). As surface temperature shows the least con-trast at that time and topographic effects are markedbecause of low Sun elevation, these images are of littleuse for geological interpretation-they appear like im-ages of shaded relief. Although night-time TM thermalimages can be acquired, this requires special requests,and in any case the overpass time is not predawn, butat around 2130-2230 hours, Wh,en differential heatingeffects have not had time to subside. Because land-surface and cloud temperature data are useful to mete-orology, and global estimates of sea temperature areuseful to fisheries, all meteorological satellites carry athermal imaging system. That with the best resolution(1.1 km) is the AVHRR aboard the TIROS-N/NOAAseries (Section 3.12.1). Since 1978 this has produced con-tinuous day and night coverage of the whole globe at

the ground retains heat and has a brighter night-timesignature than bare soil.

Moisture contained in soils at or near th~ surface has aprofound effect on the thermal behaviour of rocks. Thishas nothing to do with its intrinsic thermal propertiesbut results from the cooling effect of evaporation. Themore porous a rock is the greater the effect. Areas ofdamp terrain appear coolon images taken at any time(Fig. 6.6). The amount of cooling increases with air tem-perature and decreases as humidity rises. The greatestcontrast between damp and dry soil therefore is duringthe daytime, particularly when the weather is hot anddry. Another enhancing factor is the speed of the wind,as this 'blow-dries' the surface, as well as increasing theconvective loss of heat. FiguTes 6.8 and 6.9 illustratesome of the typical features o[thermal images in humidterrains.

The thermal images shown up to this point atefrom airborne instrumeJ:\ts-that cover narrow swaths ofground at a large scale. Small-scale thermal images with

Thermal Images 167

4 km resolution and selective imaging at full resolution.The system has a ground swath about 2500 km wide andhas great but underused potential for regional studies ofgeology and hydrology. An example of NOAA thermalimagery is shown in Fig. 6.10. One of the two constantlyoperating NOAA platforms has overpasses at 1430 and0230 hours, which are better suited to geological inter-pretation of the thermal bands.

The Heat Capacity Mapping Mission (HCMM) (Sec-tion 3.12.5) was designed specifically for measurementof thermal inertia and thermal discrimination of surfacematerials. Although the amount of data and their cover-age was limited by termination of the mission in 1980after 2 years and data were received by only a few groundstations, HCMM data form a valuable resource forregional geological applications. The resolution is 600 mfor swath.widths of 716 km. Data were acquired duringthe day in a broad band between 0.55 and 1.1 ~m toestimate surface albedo (Day-VIS), and in a 10.5-12.5 ~mthermal band both at night (Night-IR) and in daytime(Day-IR). Because of the timing of day and night orbits,and variation in cloud cover, only a limited number ofareas have both kinds of thermal image close enough intime .to be compared meaningfully. Figure 6.11 showsexamples of these. Although they reveal features of greatgeological interest, they are by no means as useful ascombinations of the data with those from other systems,such as false colour composites (Section 6.2.2) andimages of thermal inertia derived from the HCMM data(Section 6.3).

Fig. 6.7 (right) The area of Recent volcanic activity and intensedesert erosion at Pisgah in the Mojave Desert, California, USA,is a classic test site for remote-sensing techniques. The digitalimage of reflected visible radiation in (a)-equivalent toalbedo-shows very little information in the nearly blackareas of lava flows, or in the very bright areas of superficialsediments and the evaporites in the dry lake at the bottomcentre of the image. Images (b) and (c) are, respectively, of day-and night-emitted infrared. Comparison between the threeimages enables qualitative assessment of the effects of albedoon energy absorption, and of thermal inertia variations. Forexample, the high-albedo lake deposits are cooler during theday than the less reflective alluvial deposits surrounding them,as expected. The black lava flows have not achieved muchhigher daytime temperatures than the surrounding reflectivesediments"however, which is contrary to first expectations.On the night image the lavas are much warmer than thesediments. This suggests that the lavas heat up and cool downmore slowly than the sediments, and therefore have a higherthermal inertia. Some details of the lavas appear in the thermalimages, but a fuller analysis is reserved for the images in Fig.6.12 and Plate 6.3. The thermal images also reveal a clear linearfeature running from top to bottom, just to the left of centre,.which is not visible on the albedo image. This is probably anactive fault that cuts both lavas and sediments. Courtesy ofAnne B. Kahle, Jet Propulsion Laboratory, Pasadena.

168 Chapter 6

Fig. 6.8 The top of this pair of thermal images of the FrontRange in Colorado, USA, was acquired at 1100 houR;,the bottom just before dawn, and they show much the samecontrasts as noted for Fig. 6.5. On both, but particularly on theday ~age,the complex folds at the mountain-plain boundaryare clear, as are many linear features (probably faults) in theRockies proper. The night image shows more clearly thedifferent sedimentary uI:lits along the mountain front, anddistin~ishes between them and the dominantly graniticrocks of the Rockies in this area. The particularly bright foldedlayers at the front are quartzites, which have a very highthermal inertia. Courtesy of Ken Watson, US GeologicalSurvey, Denver.

6.2.2 Using thermal dAta in false colour images

Despite the complexities of interpreting thermal infrareddata in the form of grey-tone images, various methodshave been devised to display them in colour. The attrac-tion of doing this stems from ,the contrasted appearanceof day and night images and the lack of correlationbetween thermal data and those from other parts of thespectrum. In theory, poorly correlated data displayedtogether as components of a colour image should pro-duce a wide range of hu~s.

There are various options for colour display. Thesimplest is to use suitably contrast-stretched raw data tocontrol red, green and blue components of an image.Despite the poor resolution the result in Plate 6.1 isamuch stronger discrimination among the rock typespresent than in Fig. 6.11.

A means of adding fine topographic detail to low-resolution HCMM data is to combine them with betterresolution data from another system. Merely to use, say,80 m resolution Landsat MSS data to control one of thethree primary colours adds little. The coarser HCMMdata dominates the other two colour components, stillresulting in an image with poor spatial resolution. Toavoid this requires that the brightness or intensity of theimage is controlled by the data of better spatial quality,while the colour incorporates thermal information. Plate6.2 is an image using ISH colour space (Section 5.2.2). It

(a)

Fig. 6.9 This pair of daytime digital images contrastsreflected visible radiation (a) with emitted infrared (b )from

.'. ,

an area of complex Precambrian and Lower Palaeozoicrocks in northern Scotland, which are largely blanketedby vegetation, boulder clay and peat. The albedo imageshows very little detail; In contrast, the thermal image is

aff~cted by subtle differences in vegetation, soil moisturecontent and thickness of the thin superficial veneer.This reveals several interesting structural features thatwould otherwise go unnoticed. Source: Open University,courtesy of NERC, Swindon.

Thermal Images 169

(b)

represents an exotic block of old continental crust that wasimpacted into this geologically young terrain during theclosure of Tethys. The younger, folded sediments of theMakran Range, running east-west in the south, and the lessdeformed sediments flanking the Lut Block have relativelyhigh surface temperatures. Considering the time of day, theyounger rocks possibly have higher thermal inertia than thecrystalline rocks by the Lut Block, although there will be someconfusion as a result of the time of the image being close to thatof cross-over on Fig. 6.3(a).

Fig. 6.10 The TIROS-NOAA series of meteorological satellitesprovide coarse-resolution image data in the visible ari'd near-infrared regions, and in the thermally ernittedpart of thespectrum, using the on-board A VHRR system. Image (a)shows A VHRR red reflectance (band 1) over part of south-eastern Iran, taken approximately 1 h after dawn. This isequivalent to an albedo image, and shows considerablevariation over this tectonically complex area. The image ofemitted radiation in the thermal band 4 (b) is much simpler.The most obvious feature is the zone of relatively cool rocksextending roughly north-south along the Dasht-e-Lut. This

night as thermal inertia, together with data from otherspectral regions to improve the separability and there-fore classification of surface materials.

combines Landsat MSS band 7 data with HCMM Day-IRand Night-IR.

A powerful means of colour display involving thermaldata combines them with many sets of data from otherparts of the spectrum. A convenient means of doing thisis by principal component analysis (Section 5.4). In adisplay, those components loaded strongly by thermaldata are combined with components with a strong load-ing for other data sets. Plate 6.3 is an image where redand green are controlled by PCs loaded with visibleand near-infrared information, whereas blue is drivenby a PC containing most of the thermal information.Although vivid in colour and clearly discriminating dif-ferent kinds of surface, interpreting such images in termsof the spectral properties of different types of surface isdifficult.

6.3 Semiquantitative analysis

6.3.1 Thermal inertia estimates

Only sophisticated laboratory experiments can obtainaccurate measurements of thermal inertia. Estimates usingimages derive from combining information on albedoand the temperature change of the surface during thediurnal heating cycle. Because it is impossible to measurethe true surface temperature precisely using thermal data,the calculations use values of thermal inertia derivedfrom the emitted energy, assuming that the surface actsas a blackbody. The mathematical relationship for this isfairly simple, and the conversion from ON to equivalentblackbody temperature is done by look-up tables for theinstrument used. The average ON in the visible andnear-infrared regions reflected from the surface duringdaytime provide albedo. -

Although expensive, it is quite easy to acquire thethree data sets needed for thermal inertia estimationsusing airborne line-scan systems. Finding suitable datafrom a satellite is by no means so simple. The only orbitalsystem designed for this purpose was the HCMM. Its600 m resolution data for albedo and day and nighttemperature are available only for areas covered by a

Extracting information relating to surface thermal prop-erties from images relies on their being in digital format.Semiquantitative analysis has two main objectives. Firstly,it exploits the diurnal changes in surface temperatureat surfaces with different thermal properties to expressthem as thermal inertia. Secondly, it uses thermalinformation, either of one time or combined for day and

170 Chapter 6

(a) (b)

Fig. 6.11 Both images are of digitally processed HCMM datafor the San Rafael Swell in Utah, USA, which is an erodedasymmetrical anticline. It exposes Permian sediments in itscore, which are succeeded outwards by Triassic, Jurassicand Cretaceous sediments, mainly sandstones, shales andcarbonates~ Image (a) shows Day-IR data and (b) Night-IRdata. Below eacn image is an interpretation based on thethermal data and the known stratigraphy. Units 17 and 16comprise Permo-Triassic limestones and red shales. Unit 15is a Triassic sandstone.. Units 14, 13 and 12 are all pale, massive

sandstone of Triassic to Jurassic age. Units 11 and 10 areJurassicshaley limestone and fine red sandstone. Units 5, 3and 2 are Cretaceous shales, and unconsolidated Tertiarystrata are represented by TF and TK. On each image all arecrud~ly distinguishable. An important feature is that Units 12,13 and 14 show cool day and warm night signatur~s, comparedwith the similar looking Unit 1O, which is warm during the dayand cool at night. Units 12, 13 and 14 have high thermal inertia,whereas Unit 10 has a low thermal inertia. Courtesy of Anne B.Kahle, Jet Propulsion Laboratory, Pasadena.

Thermal Images 171

Like conventional day and night thermal images, fromwhich they are derived, A TI images contain anomal-ous features. As well a~ being affected by topography,they contain features that result from variations in soilmoisture content, varying wind speed, au temperature,humidity and various other factors. As it is not possibleto correct for all these factors, interpretation must bebacked up by surface observations on the day of the

image acquisition.

6.3.2 Thermal data in classification

Images of thermally emitted infrared radiation and theirderivatives express variations in density, porosity, mois-ture content and thermal inertia of surface materials.These have little if any effect on images of reflectedradiation. Apart from the effects of narrow absorptionbands, visible to near-infrared images are dominatedby variations in albedo. Thermal infrared images con-tain an albedo effect, so that bright areas on an image ofreflected energy are relatively dark (cool) on a thermalimage. There is a negative correlation between the twotypes of image. The albedo 'effect is removed by thecalculations involved in producing an AT! image andso they are more or less uncorrelated with images in thevisible to near-infrar~d spectrum. Clearly, this improvesthe potential for discrimination between different typesof surface.

Images of AT! allow discrimination in areas of uni-form albedo, where yisible and near-infrared imagesare least effective (Fig. 6.13). Examining the range ofthermal inertias for some common rocks (Table 6.1and Fig. 6.14) shows several interesting features thathelp orientate the use of thern;lal data in classification.Igneous rocks of widely different composition are notwell separated by thermal data, with the exception ofultrabasic rocks. They are best discriminated by reflectedradiation. Sedimentary rocks on the other hand showan excellent range of values, reflecting large variationsin their porosity, density and mineralogy. Of particularimportance is the strong contrast between limesto~eand dolomiter which are virtually indistinguishableby any other remote-sensing technique. Thermal dataare also useful in distinguishing between crystalline,partly cemented and uncemented sediments, becauseof variations in porosity and density. Quartzite, sand-stone and sand are c~mposed mainly of quartz andtherefore difficult to distinguish on images of short-wavelength radiation. They show strong contrast onthermal inertia images. Even where the data availabledo not permit calculation of thermal inertia values, aswith Landsat TM band 7, a single thermal dimensioncan be used as a powerful discriminant in classifica-tion, simply because of low correlation with otherdimensions.

few ground-receiving stations. Global data are availableat 1.1-4 km resolution from A VHRR. Although Landsatgathers thermal data, this is donerbutiriely only duringdaytime overpasses. By special request it is possible toturn on the sensors at night. However, the unfortunatetimings of the Sun~synchronous Landsat orbits (0930and 2130 hours) coincide with the times of least contrastin surface temperatures (Fig. 6.3). ,

The first step in calculating the pixel-by::pixel vari-ation of thermal inertia over an area is to register imagesof albedo, day and night thermal data. Accurate registra-tion of the images from tWo times is difficult, particu-larly with airborne data that ~clude relief displacement.Satellite data for day and night during the same 24-hcycle are captured from ascending and descending orbits,so that only lozenge-shaped areas where both groundtracks intersect are available for study. In tbe case of theshort-lived HCMM experiment, the scope for study ofthermal inertia is limited. The continuous, world';widerecording of thermal data from the NOAA A VHRR sys-tem offers much greater potential, although it has beenlittle used.

The simplest measure of the rate of heat loss from thesurface by radiation is the difference between day andnight temperature (AT). This does not give an accurateestimate of thermal inertia because of the albedo effect.A surface with a high albedo heats up relatively slowly,has less heat to lose at night and so has a small valueof AT. Low albedo surfaces soak up more heat, achievehigher day temperatures and lose heat very rapidly atnight to give a high AT. A more meaningful estimate ofapparent thermal inertia (All) combines albedo (a), ATand a correction factor for varying solar energy input tothe surface (C), as follows:

ATI = C(1 -a) IAT (6.3)

The correction factor C depends on season, latitude,altitude and the angle and direction of slope. For mostpurposes corrections are made only for season andlatitude. To remove topographic effects requires a digitalimage of elevation and extensive computations. In mostcases the results show no striking improvement onuncorrected All. Values for a and AT are derived fromthe average visible to near-infrared reflectance, and fromday and night temperature-estimates based on thermalimage data. For HCMM these are Day-VIS, and Day-IRand Night-IR, respectively. The relevant AVHRR dataare daytime Channels 1 and 2 for albedo, and Channels 4and 5 for thermal data. The calculation is performed foreach pixel in the registered scene of day and night data.The A TI values fpr each pixel are rescaled to the conven-tional 0-255 range, when ATI itself can be displayedas an image. Figure 6.12 compares images of AT and A TIderived from the aircraft data for the area shown inFig. 6.7.

172 Chapter 6

(a)

I~

~'\.,.

4

Fig. 6.12 Image (a) is of temperature difference between dayand night derived from aircraft data of the same volcanicarea in the Mojave Desert of California, USA, as Fig 6.7 andPlate 6.3. Image (b) is of apparent thermal inertia. The map (c)$hc;>ws rough outlines of the major outcropping rock units..Unit 1 is alluvium, 2 represents !j;aline flats on a playa lake,3 is basalt (3a) with an aa surface), 4 indicates rhyoliteintrusives, 5 is granite and 8 is andesite. The image helps todiscriminate igneous rocks with high thermal inertia fromalluvium with low, and identifies a division between aa andpahoehoe surfaces within the central Pisgah flow. However,it makes no clear distinction between igneous rocks suchas 3, 4 and 8 with very different compositions. Courtesy ofAnne B. Kahle, Jet Propulsion Laboratory, Pasadena.

3

4 '~I 't,

~ 3\\8~0 5km(c)

gabbro ~6.4 Multispectral thermal data

U black shale

d turbidites rt--- black chertbasalt ~ ~ Within the ~ange 8-14 ~m the emission spectra of silicate

minerals contain a prominent, broad absorption troughand associated features caused by Si-Q bond stretching.The position of this trough and the shape of its shouldersare controlled by the co-ordin~tion of silicon with oxygen.As t~e proportion of silicon in a silicate increases, theabsorption feature shifts to shorter. wav~lengths. Becausesubdivision of silicates is a~cording to Si-O co-ordination,tl:lis shift is potentially a means of discriminating betweenthem using remote sensing (Section 1.3.2, Fig. 1.12).Similar features characterize carbonates (C-O bondstretching); iron minerals (Fe-O), clay minerals (Si-O-Si,Si-O, AI-OH) and various other groups of minerals

1000 2000 3000 4000

Thermal .inertia (Jm-:2 (_'/2 K-1)

Fig. 6.13 Landsat MSS data fail to discriminate between rocktypes With similar, low albedos. In thermal imaging, constantalbedo is a distinct advantage, as variations in surfacetemperature then relate ohly to differences in density, thermalcapacity and conductivity. These differences are bestexpressed by thermal inertias, shown in these histograms,where the spread among dark rocks is very Wide. Data fromWatson (1978).

Thermal

Images 173

Igneous

~granitesquartzite~

felsic

gabbroCb

A.

maficmonzonite

oasal

latiteSedimentary

shales /

olivine

basalt

limestones1__5

san~~~~/:~I

~I

dol O~~~:::~~~ ~

q U~~~,,-";:-~

montmorillonite

1000 2000 3000

Thermal inertia (J m-2 c-'i2 K-')

Fig. 6.14 A comparison b~tween histograms showing therange of thermal inertia in igneous and sedimentary rocksshows that thermal imagery is more helpful in mappingsedimentary basins than areas of crystalline rock.Metamorphic rocks show much the same overlaps as igneousrocks. Data from Hummer-Miller and Watson (1977).

4000

limestone +clay and quartz-.

limestone

(Section 1.3.2, Fig. 1.13). The wealth of different, com-positionally controlled features in the thermal infraredpart of the spectrum suggests a bright future for multi-spectral analysis within the 8-14 ~band. It is the regionmost likely to provide direct discrimination between rocktypes, because the most prominent features relate directlyto rock-forming minerals rather than minor componentssuch as lini.onite and clay minerals. Figure 6.15 showsspectra for a Wide range of igneous and sedimentary rocks.

Despite the great attraction of multispectral thermaldata, the difficulties in separating narrow wavebands inthe emitted Tegion has enabled only two experimentalimaging devices to be produced to date-the TIMS andthe Geoscan systems. Their spectral -ranges are shown inTable 3.2. Both have been operated only from aircraft,although the ASTER instrument incorporates fivenarrow bands (Table 3.4) in the mid-infrared. ASTER isaboard the Terra satellite launched in December 1999.

dolomite

6 7 8 9 10 12 16 20 30

Wavelength (IJm)Fig. 6.15 These thermal-infrared transmission spectra ofigneous and sedimentary rocks depend on the componentminerals in each rock (Figs 1.12 and 1.13): The vertical barsindicate the wavebands detected by the experimental ThermalInfrared Multispectral Scanner (TIMS). Thespectra are stackedto show relative shapes. From Kahle and Rowan (1980).

kaolinite

174 Chapter 6

major roles for this appr~ach. The first relates to igneousrocks, which of course contain a blend of various silicateminerals. Figure 6.16 shows the progressive shift of theSi-O bond-stretching featUre in rock spectra towardslonger wavelengths as bulk silica content decreases.Plate 6.4 shows how powerful this line of attack could be.If ultrabasic rocks were present in the area even betterresults would have been possible. The other approachconcentrates upon the differences in quartz, clay mineraland carbonate spectra and their effects on the signaturesof sedimentary rocks. Plate 6.5 gives an example of sedi-mentary discrimination using this method.

Associated resources on the CD-ROM

You can access the resources by opening the HTML fileMenu.htm in the Imint folder on the CD using yourW ebbrowser. More detailed information about installing theresources is in Appendix C.

On the Menu page, select the link Additional Images:Image Data, where you will find examples of imagesbased on thermally emitted data.

Further reading

8 9 10 11 1213

Wavelength (IJm)Fig. 6.16 Falling SiO2 content of igneous rocks (andmetamorphic rocks of the same range of compositions) resultsin a progressive shift of the Si-o bond-stretching absorptionfeature to longer wavelengths in thermal emission spectra.The positions of the TIMS channels (vertical bars) suggest thatmultispectral thermal studies could become a powerful toolin geological mapping of crystalline terrains. The spectra arestacked to show their shapes more clearly. Data from Vickersand Lyon (1967).

In theory, choosing an appropriate combination ofbands based on laboratory spectra and suitable ratiosshould highlight the effects of different minerals. Inpractice, the radiance emitted by the surface is domin-ated by temperature and hence topography, albedo andthermal inertia. There is a very high degree of correlationbetween thermal bands. Colour composites of raw bandsor ratios, even with contrast stretching, are very bland.The best approach is using the decorrelation stretchdiscussed in Section 5.4. This fully exploits colour spaceand produces images in which colours can be related tolaboratory spectra and interpreted in terms of differentrock types.

Experimental work on a number of test sites hasshown up several interesting results, a few of which arehighlighted by Plates 6.4 and 6.5. There seem to be two

Abrams, M.J., Kahle, A.B., Palluconi, F.D. & Schieldge, J.P.(1984) Geological mapping using thermal images. RemoteSensing of Environment 16, 13-.'3.

Byrne, G.F. & Davis, J.R. (1980) Thermal inertia, thermal admitt-ance and the effect of layers. Remote Sensing of Environment 9,295-300.

Eberhardt, J.E., Green, A.A., Haub, J.G., Lyon, R.J.P. & Prior,A.W. (1987) Mid-infrared remote sensing systems and theirapplication to lithological mapping. IEEE Transactions,Geoscience and Remote Sensing GE-25, 230-236.

Hummer-Miller, S. & Watson, K. (1977) Evaluation of algo-rithms for geological thermal inertia mapping. Proceedingsof the 11th International Symposium on Remote Sensing of Envir-onment, pp. 1147-1160.

Kahle, A.B. (1980) Surface thermal properties. In: Remote Sens-ing in Geology (eds B.S. Siegal & A.R. Gillespie), pp. 257-273.Wiley, New York.

Kahle, A.B. & Abbot, E., eds (1986) The TIMS Data UsersWorkshop. Technical Publications, Jet Propulsion Laboratory,Pasadena, CA.

Kahle, A.B. & Goetz, A.F.H. (1983) Mineralogic informationfrom a new airborne thermal infrared multispectral scanner.Science 222, 24-27.

Kahle, A.B. & Rowan, L.C. (1980) Evaluation of multispectralmiddle infrared aircraft images for lithologic mapping in theEast Tintic Mountains, Utah. Geology 8, 234-239.

Kahle, A.B., Abrams, M.J., Alley, R.E. & LeVine, C.J. (1981) Geo-logical Application of Thermal Inertia Imaging using HCMMData. Publication 81-55, Jet Propulsion Laboratory, Pasadena,CA.

Kahle, A.B., Palluconi, F.D., LeVine, C.J., Abrams, M.J., Nash,D.B., Nash, R.E. & Scheildge, J.P. (1983) Evaluation of Thermal

~

Data for Geological Applications. Publication 83-56, Jet Pro-

pulsion Laboratory, Pasadena, CA.Nash, D.B. (1985) Detection of bedrock topography beneath

a thin cover of alluvium using thermal remote sensing. Photo-

grammetric Engineering and Remote Sensing 51, 77-88.Ninomiya, Y., Matsunaga, T., Yamaguchi, Y., Ogawa, K.,

Rokugawa, S., Uchida, K., Muraoka, H. & Kaku, M. (1997) Acomparison of thermal infrared emissivity spectra measuresin situ, in the laboratory, and derived from thermal infrared

multispectral scanner (TlMS) data in Cuprite, Nevada, USA.International Journal of Remote Sensing 18, 1571-1582.

Offield, T. W. (1975) Thermal-infrared images as a basis for struc-ture mapping, Front Range and adjacent plains in Colorado.

Geological Society of America Bulletin 86,495-502.Oppenheimer, C. & Francis, P.W. (1997) Remote sensing of

heat, lava and fumarole emissions from Erta' Ale volcano,Ethiopia. International Journal of Remote Sensing 18, 1661-1692.

Pohn, H.A., Offield, T.W. & Watson, K. (1974) Thermal inertiamapping from satellites-discrimination of geological",unitsin Oman. U.S. Geologica} Survey Journal of Research 2, 147-158.

Pratt, D.A. & Ellyett, C.D. (1979) The thermal inertia approachto mapping of soil moisture and geology. Remote Sensing ofEnvironment 8, 151-168.

~owan, L.C.,Offield, T.W., Watson,K., Cannon, P.J. & Watson,R.D. (1970) Thermal infrared investigations, Arbuckle

Mountains, Oklahoma. Geological Society of America Bulleting.. ']~An ,,~,,~

[hermal Images 175

Sabins!F.F. (1969) Thermal infrared imagery and its applica-tion to structural mapping in Southern California. GeologicalSociety of America Bulletin 80! 397-404.

Sabins! F.F. (1980) Interpretation of thermal infrared images. InRemote Sensing in Geology (eds B.S. Siegal & A.R. Gillespie),pp. 275-295 Wiley, New York.

Short, N.M. & Stuart, L.M. (1982) The Heat Capacity MappingMission (HCMM) anthology. NASA Special Publication 465.Washington DC.

Vane, G. (ed.) (1987) Proceedings of the 3rd Airborne ImagingSpectrometer Data Analysis Workshop. Publication 87-30, JetPropulsion Laboratory, Pasadena, CA.

Vickers, R.S. & Lyon, R.J.P. (1967) Infrared sensing fromspacecraft-a geological interpretation. Proceedings, Thermo-physics Special Conference, American Institute of Aeronauticsand Astronautics, Paper 67-284.

Watson, K. (1975) Geological applications of thermal infraredimages. IEEE Transactions, Geoscience and Remote Sensing GE-63,128-137.

Watson, K. (1978) Thermal phenomena and energy exchangein the environment. In: Mathematical and Physical Principles ofRemote Sensing, pp.109-174. CNES, Toulouse.

Wolfe, E.W. (1971) Thermal IR for geology. PhotogrammetricEngineering and Remote Sensing 37,43-52.

Wood, J.A., Lasserre, M. & Fedosejevs, G. (1990) Analysis ofmid-infrared spectral characteristics of rock outcrops and anevaluationnfthpJ(""hlo ~~~I' "-". ..

">2

subsurface materials;3 the wavelength, polarization and depressionthe radar, which are controlled variables;4 the electrical properties of the surface-the complexdielectric constant of surface materials.In decreasing order of importance, all help determinethe proportion of incident microwave energy that thesurface scatters back directly to the antenna aboard anaircraft or orbiting platform. This governs the tone of theradar image. The brighter the tone the larger the back-scattered energy. As with images acquired from any partof the spectrum, the fundamental basis of interpretationlies in the spatial distribution of tones to give familiartextures, shapes and patterns. The tone at any pointgives information relevant to the actual nature of thesurface itself. This section considers the effects on back- .scatter of variations in surface electrical properties androughness and how these effects change with wave-length, polarization and depression angle. Going furthermeans introducing a couple of important definitions.

A measure of the intensity of back-scattered energy -from a point target is its radar cross-section. This is the.-area of a hypothetical surface that scatters radar energy-equally in all directions crnd which would return thesame energy to the antenna as the real point target. Ameasure of the energy back-scattered from a target witha large area, such as a field, is its radar scatteringcoefficient. This is the average radar cross-section perunit area. It is a dimensionless quantity, and as it variesover several orders of magnitude we express it as 10 timesits logarithm, in decibels (dB). The radar scatteringcoefficient is a fundamental measure of the radar pro-perties of a surface, and governs the tone of the surfaceon a radar image.

The microwave region of the EM spectrum presents twoopportunities for gathering remotely sensed data. First,like radiation in the 8-14 ~m range, the Earth's surfaceemits microwaves as a result of its te:mperature, accord-ing to the Stefan-Boltzmann relationship and Wien'sLaw (Section 1.2). Second, microwave radiation can begenerated artificially as coherent waves. Not only doesthe atmosphere fail to attenuate microwaves to any greatextent beyond about 0.3 cm, but electronic systems candetect very low energies. The first opportunity relatesto passive microwave remote sensing, the second makespossible radar imaging, the only operatWnal activeremote-sensing system. Radar plays the dominant rolein this chapter because passive microwave methods,although theoretically promising, have failed to produceuseful results for geologists..

Radar imaging possesses many advantages relatingto the control that is possible with an active system. Theprinciples were covered in detail in Section 3.5. It too has-a day or night, all-weather capability and can be directedat the surface or at the atmosphere. It depends on thescattering of energy by materials that lie in the path ofan artificial microwave beam, known as back-scatter,and reception of the proportion of energy that returnsback to the antenna. Radar. waves are generated incbherent form, that is as radiation with a single wave-length where all waves are in phase. Coherent radarmakes possible the use of Doppler shift in generatingsynthetic-aperture radar (SAR) images, where resolu-tion is fine and independent of range (Section 3.5). Anymicrowave wavelength can be generated with anypolarization. The angle of illumination of the surface canbe varied by adjusting the antenna. Simply changingthe flight line of the platform changes the directionof radar propagation. These controls help optimizeillumination conditions for many kinds of application.Flexibility and complete control is essential because theinteractions between radar waves and the surface areboth complex and very different from those for shorter

wavelengths.

7.1 Interactions between radar andsurface materials

What happens to the electromagnetic energy in a radarpulse when it meets the surface depends on four mainfactors:1 the attitude of the surface-dealt with at length inSection 3.5;

7.1.1 Complex dielectric constant

A material's dielectric constant controls the proportionof radar energy reflected by a material and that penetrat-ing into it. Materials with a high dielectric constant, suchas metals and water, are excellent reflectors and absorbvery little energy. The lower the dielectric constant themore energy is absor~ed and can re-emerge, giving thepotentialfor penetration beneath the surface. The diele£;tric constant is not a simple function of wavelength, butvaries relatively little for most rocks and soils over therange of commonly used wavelengths.

Dry soils and rocks have dielectric constants in therange 3-8. These are low ~nough to permit penetration

176 '- '-;,...

Radar Remote Sensing 177

L x Ka

+"'t:co1;;t:0u

.~

tQJ

""Qj

(5

Because the leaves oi growing plants have a highmoisture content, vegetation has a higher dielectric con-stant than dry natural materials. Plants therefore areexcellent radar reflectors. As a result only a proportion ofradar energy is able to penetrate to the surface beneath,depending on the nature and density of the vegetationcanopy. Figure 7.1 suggests that longer wavelength radaris more capable of penetrating vegetation to give informa-tion about the surface than that of short wavelength.

Except for its effect on penetration, dielectric constantis a minor factor in controlling the tone and textureof radar images. They are dominated primarily by slope.effects

(Section 3.5) and by surface roughness.

10' 109 10'0 10'1 1012[300] [30] [3] [0.3] [0.03]

Frequency (Hz)[Wavelength (cm)f

Fig. 7.1 As the wavelength of microwave radiation increases,so too does the effective dielectric constant of water. At thewavelengths of Ka- and X-band radar the dielectric constantis between 30 and 45, whereas for L-band radar it is at itsmaximum of about 80. Rocks and soils rarely conta~ morethan about 15% water, so the effect of moisture on theirdielectric properties is considerably reduced comparedwith that suggested by the graph. It might appear from thegraph that, because the dielectric constant of water decreasesat shorter wavelengths, Ka-band radar is most likely topenetrate beneath the surface. However, the physics of thephenomenon show that roughly an equal number of radarwaves penetrate material with the same moisture content,irrespective of wavelength. So, contrary to expectations,it is L-band radar that is most effective at revealingsubsurface features. Experiments indicate that unlessthe moisture content is less than 1 %, penetration isnegligible.

of a significant proportion of incident radar energy intodry materials to depths up to 6 musing L-band or 25 mradar. Penetration is only possible, however, where thetopographic surface is radar-smooth. Subsurface returnsemerge only when the subsurface has a radar-roughcomponent (Section 7.1.2). As water has a maximum

",. dielectric constant of 80, the greatest cause of variationin this parameter in rocks and soils is their moisturecontent. As moisture content increases the dielectricconstant increases in a roughly linear fashion. However,the dielectric constant of water increases as wavelengthincreases (Fig. 7.1), and this has an effect on the radar-related properties of damp soil too. As even a saturatedsoil contains relatively little water the effect is not marked.

Penetration of radar into soil is expressed in terms ofthe number of wavelengths. The complex equations gov-erning this behaviour suggest that for constant dielectricproperties the same number of waves penetrate, irre-spective of wavelength. Thus the longer the wavelengththe greater the penetration. Recent studies indicate thatmoisture content greater than 1 % effectively rules outany penetration, so this phenomenon can be exploitedonly in hyperarid terrains of terrestrial deserts and onplanets such as Venus and Mars.

7.1.2 Roughness

A perfectly smooth surface of a material with a highdielectric constant acts as a mirror to radar, as it wouldto all forms of radiation. Being directed to the side ofthe platform, radar pulses meet a horizontal surface at anacute angle and are reflected away from the antennaat the same angle, without being scattered. This specu-lar reflection results in a totally black signature for asmooth surface (Fig. 7.2a). Where a smooth surface isat right angles to the radar beam, the reflection is dir-ectly towards the antenna, giving an intensely brightresponse. Sand dunes and ocean waves I;;ommonly showthis effect. Smooth reflectors with facets arranged atright angles result in multiple reflections between thefacets. This is the principle behind the corner reflectorsplaced at the mastheads of small boats, so that they aredetectable by navigational radar; Regardless of the angleat which a radar wave enters the cavity of a cornerreflector, reflections between the facets ensure that theenergy is returned directly to the antenna (Fig. 7.2b).Corner reflectors occur naturally at the intersectionsbetween. bedding and rectangular jointing in rocks suchas sandstones, limestones and lavas. They give rise tobright speckles, or blooms, within the area of more uni-form tone that characterizes such rocks.

A Tough surface is made up of countless irregularities,some of which mimic corner reflectors, whereas othersgive rise to more complex interactions. The net effect isto scatter radar energy diffusely in all directions. Somereturns to the antenna as a measurable signal (Fig. 7.2c).Roughness, for the purpose of interaction with EM radi-ation is a relative term depending on wavelength andangle of incidence: Two relationships (Equations 7.1 and7.2) allow roughness to be quantified. In them the heightof irregularities-the roughness (h)-is determined bythe wavelength )" and the angle of incidence e. A surfacethat appears smooth to radiation satisfies the Rayleighcriterion:

h < ),,/25 sine. (7.1)

178 Chapter 7

(b)

Fig. 7.2 How radar is back-scattereddepends on the incidence angleand on surface roughness. Fourpossibilities are shown: (a) for aperfectly smooth horizontal surface;(b) for a corner reflector; (c) for arough surface: and (d) for a ~moothnatural surface.

+10 rTable 7.1 Surface roughness (in cm) relative to radarwavelengths, dedved from the Rayleigh cdteriono"

smooth

m~+-'t:OJ

.0

~OJ0u

C1t:

W

i::to

U'"

0

", moderately rougt',,\

'. \"..\

"""\ '. -.\ \\,-'

-10

'. '. '.""" rough

""""

""

:u"CIQIX: -20

""

" -"""

...

~

..

--0 20 40 60

Angle of incidence (degrees)

Fig. 7.3 The proportion of emitted energy back-scatteredfrom a surface to the antenna-the radar scatteringcoefficient-depends on both the surface roughness andthe angle of incidence. Both the depression angle andthe surface slope control the incidence angle.

One that is rough satisfies the criterion:

H> A/4.4 sin£} (7.2)

The behaviour of radar waves at natural surfaces iscomplex. Rough surfaces scatter energy diffusely in alldirections, whereas surfaces of intermediate roughnesscombine specular and scattered components (Fig. 7.2d).The tone of a horizontal natural surface on a radar imagetherefore is a combined result of its roughness and toa much lesser extent of the dielectric constant of thematerials from which it is formed.

Quite clearly the Rayleigh criterion implies that radarwavelength helps determine what is rough or smooth.Table 7.1 shows limiting values of the mean height ofsurface irregularities associated with different categoriesof roughness for three radar wavelengths in common use.They were derived using the Rayleigh criterion. Roughnessalso depends on the angle of incidence of radar waves atthe surface. For radar images gathered from aircraft thisis a major problem. Because the depression angle changesmarkedly from near to far range (Section 3.5), ensuringthat the antenna can record from a wide swath of surface,means that the incidence angle changes too. Theapparenlroughness of a surface of uniform roughness thereforechanges across the swath. As only a narrow range ojdepression angles can encompass a very broad swath fromorbital altitudes, this effect on image tone is not apparenlon Seasat and SIR images (Section 3.12.1). Depressiorangle can be varied systematically on orbital SAR systems

It is possible theoretically to exploit the variation ofapparent roughness with incidence angle to gain moredetailed information about the nature of the surface, orto select a depression angle that gives the greatest dis-crimination between surfaces of different roughness.

Figure 7.3 shows how the radar scattering coefficientdepends on surface roughness and incidence angle. Fo!low incidence angles smoo'th surfaces reflect a largeproportion of energy directly back to the antenna. Afthe incidence angle increases, more and more energy i!reflected away from the antenna, until at about 400 al

A

BcD

E F

1

2

3

4

5

10 km-became hyperarid. Several nea;by archaeological sites confirmthat the area was humid enough 6000 years ago to supportagriculture. One theory holds that the patterns result fromradar penetration and back-scatter from subsurface variationsin grain size. An alternative is that the radar is picking upsubtle variations in surface roughness related to the oldstream courses. This image from the Shuttle Imaging Radarprogramme, together with most of those illustrating thischapter, was produced by the Jet Prop~lsion Laboratory,Pasadena.

energy is reflected away. Smooth horizontal surfacestherefore show as bright on images taken with a steepdepression angle and dark on those taken looking fur-ther to the side of the platform. As rough surfaces scatterenergy diffusely in all directions, the proportion return-ing to the antenna is virtually independent of incidenceangle. Their brightness on an image usually does notchange much with the angle at which radar waves arereceived at the surface. They appear dark relative tosmooth and intermediate surfaces when incidence angleis steep-depression angle shallow-but form bright fe~-tures at all incidence angles less than about 20°. Figure7.3 implies that images taken with a judicious selectionof depression angles can give a powerful discriminationof variations in surface roughness of horizontal surface.This is very useful where the tonal variation of a single

image is controlled by an unknown variation of bothroughness and dielectric constant.

As well as scattered energy from the ground surface orthe top of a vegetation canopy, the radar returns fromthe Earth's surface contain components from within thesoil and the vegetation canopy itself. These comprise theresults of complex volume scattering, mainly multiplereflections from electrical heterogeneities within thesoil and among plant structures such as branches. Soillayers sometimes reflect and scatter some of the energypenetrating the soil surface, as can the buried soil-rockinterface. In many cases it is impossible to separateall the different components and all contribute to theoverall roughness of the surface but, in relatively simpleterrains, it is possible to identify effects that result fromsubsurface features (Fig. 7.4).

illumination \

Fig. 7.4 This SIR-C image of part of the Sahara Desert showsthe ability of radar to penetrate dry sediments. Landsat imagesshow that the area is a nearly featureless sand and gravel plain.The image shows patterns which suggest a drainage network.Because tributaries of the main channel 'V' towards the south,water that produced the network was flowing southwards.The Nile system, some 250 km to the east of this area, flows tothe north, as do most of the widely spaced active wadis in thisarea. The drainage pattern is thought to date from a Holocenewet period and to have been buried beneath sand as climate

180 Chapter7

viewing (Section 7.2.4). Multipolarized (Section 7.2.5)and multifrequency radar images (Section 7.2.6) provideadditional information about the COmp9$ition and tex-ture of surface materials. Radar image data areincreas-ingly available in digital form. They contain quantitativeinformation and can be analysed by image processingtechniques to extract concealed informa~on (Section 7.2.7)or combined iorcomparative purposes with other kindsof data (Section 7.2.8).

7.1.3 Polarization

Like visible light, radar transmission and reception canbe in different modes of polarization. Polarized radia-tion has its electrical component of wave vibration in asingle plane perpendicular to the direction of propaga-tion. For radar transmission and reception this usually isarranged to be either horizontal (H) or vertical (V). Themost common combination is to transmit horizontallypolarized radar and receive the horizontally polarizedcomponent of back-scattered radiation. This is assignedthe coding HH, and gives like-polarized images. Hori-zontal transmission and vertical re<;;eption (HV) producesa cross-polarized image. There are lour possibilities-HH,VV, HV and VH. The most usual strategy is-to transmitho~ontally and receive both horizontal and verticalcomponents-HH and HV.

Interaction between the surface and radar waves usu-ally leaves the sense of polarization unchanged. However;it can cause depolarization or rotation of the plane ofpolarization. So the mode of polarization can influ-ence how objects appear on images. A proportion of thedepolarized component of back-scatter and that rotatedthrough 90° is detected in a cross-polarized image. Howthese effects are produced is not perfectly understood,but a widely held view is that some of these phenomenaresult from multiple reflections at or beneath the surface.Multiple reflections are at a minimum with bare rock orsoil with a high dielectric constant. Where veg~tation isthick and varied in structure there is a greater chance ofmultiple reflections between leaves, twigs and branches,and depolarized returns are common from vegetatedsurfaces. It is possible that subsurface back-scatter bysoil particles or rock heterogeneities may give rise tosimilar effects.

7.2 Interpretation of radar images

Processes that produce the tonal and textural features onradar images are varied and complex. A general accountof how the full r~nge of rock types and geological struc-tures respond to radar illumination requires a book in itsown right. A mere impression of the possibilities is givenin this section. The strategy adopted is first to applysome of the photogeological principles introduced inChapter 4 to a series of radar images showing importantgeological phenomena (Section 7.2.1). Section 7.2.2 thendiscusses the advantages for structural geology of imagestaken with different radar look directions. Images takenwith different depression angles reveal more informa-tion about surface roughness (Section 7.2.3). Pairs ofradar images with different look directions or differ-ent depression angles introduce parallax, and this, asin overlapping aerial photographs, allows stereoscopic

7.2.1 Geological features on radar images

A radar image combines the attributes of tone, texture,pattern, shape, context and scale in much the same wayas grey-tone images of shorter wavelength radiation.They differ in the way these interpretable attributes areproduced, but the same photogeological principlesapply (Chapter 4).

The most striking feature of a radar image is how itaccentuates surface topography. To a large extent theamount of energy back-scattered to the antenna is gov-erned by the attitude of slopes. Those perpendicularto .the direction of wave propagation return nearly allincident radiation and appear bright. Those slopingaway at a steeper angle than the radar are in shadowand show as totally black. Other slope angles producereturns in which intensity combines influences of slopetogether with those stemming from the properties ofsurface materials. Slope attitude produces layover andforeshortening (Section 3.5) ~o that topography is some-times displayed in an unusual form. Despite this, radar'soverall effect is to sharpen topography to a greaterdegree than seen in other kinds of image. In this respect,radar images are analogous to edge-enhanced, low Sun-angle images. Strong shadowing helps the eye appreci-ate topography because of the pseudostereoscopic effect(Section 2.5).

Topographic emphasis is most important for inter-pretation of geolo~cal structures underlying majordestructionallandforms. Their control is often the atti-tude of compositional layering and faults, folding, andthe boundaries of igneous intrusions. Figures 7.5 and 7.6show examples of this enhancement of layering andfolds. Spectacular examples of topographic enhance-ment relate to faulting (Figs 7.7 and 7.8). Recognitionof igneous intrusions often relies on tonal and texturaldifferences as well as on the expression of cross-cuttingcontacts. Figure 7.9 gives some particularly good ex-amples of both plutonic intrusions and dykes.

Another special property of radar aids enhancementof structures. Features with rectangular boundaries arenatural corner reflectors (Section 7.1.2). All the radarenergy that illuminates the angles returns to the antenna.Even where such a corner reflector is much smallerthan a resolution cell, its back-scatter can saturate the

A B c E FD

I 10km .'

fault trending south-west-north-east. Displacement of ridgessuggests that it has a downthrow to the south-east. The ridgesare densely wooded, hence their relatively bright signatures. Apower line cut through the forest between A6 and B5 shows uIclearly. Back-scatter is dominated by the vegetation canopy,and does not relate to variations in surface roughness.Nevertheless, the wealth of textural information is useful inidentifying different rock types. The ridges are well-drainedsandstones, and the valleys are floored mainly by shales andcarbonates, which are less resistant to weathering and erosion.The impermeable shale units from A3 to E6 and B1 to H3 / 4 arEcharacterized by a much finer drainage network than the morepermeable strata. Courtesy of Intera Technologies Ltd, Calgary

at D5 Vs to the south-east, suggesting that the band of ridgesfrom A3 / 4 to E6 are strata dipping steeply to the south-east.The eastern limb of the syncline at F4 has been displaced by c

illumination

Fig. 7.5 This 1 : 35 000 image of an area near Pittsburgh,Pennsylvania, USA, is from an airborne X-band SAR systemlooking from west to east. It is 50 kIn wide, has a resolution cellof 6 x 12 m (azimuth x range) and is a digitally processed

.'---"~--'"I> T."1 1'~-_'.I_-

182 Chapter 7

10 km

AS contains a layered sequence"of ironstones, shales andcarbonates. The radar-smooth Unit in the synform consistsmainly of shales veneered by alluvium. From DI to GI /2 is arugged area underlain by a massive ironstone, weathering ofwhich has produced the huge iron ore deposits of MountTurner (Dl) and Mount Tom Price, which is just north of HI.As well as defining the major fold structures and lithologicaldivisions magnificently, SIR-A also shows up many hithertounmapped geological features. The most obvious form a seriesof wide bright linear features in the area from CI to B3. Theseare thick quartz veins, many of which follow minor faults. Theimage fails to detect pisolitic Tertiary limonite and hematitedeposits that form the main iron ores, except near MountTurner (Dl), where they show as a bright patch.

illuminationFig. 7.6 This SIR-A image shows part of the HammersleyRange in Western Australia, at a scale of 1 : 500 000. The area isunderlain by Lower Proterozoic volcanics, which areinterbedded with some of the world's1argest iron ore-deposits, in the form of banded ironstone formations. Thelargest structure is the Rocklea Dome (C2 to E2 and C3 to E3)with a core of Archaean granite. A similar basement-coreddome occurs at H3 / 4. The bright linear feature from D3 to C2is a dyke cutting the granite, but which is unconformablyover~in by the Proterozoic volcanics. Other fine lines in thegranite are also dykes. The dark unit with bright ridgessurrounding the dome is an easily weathered sandstone.Succeeding it to the west and between D2, ES and F4 arepillowed metabasalts, the rough surface of which results in abright signature. The synform with an axial trace from D4 to

A B c D E F G

I 10 km I

in A/B / C4 are northward and are to the south inA/B / CS,indicating an east-west striking antiform. The east-weststructure from G2 to H2 is a synform. Thedifficult)r inassessing the older structures IS ma~ified by the effectsof later faulting. rhelinear feature from C1 to CS clearlydisplaces structures on either side in a)eft-lateral sense andis a major strike-slip fault. Anottler importantf,ault trendsfrom F4 to G2. In both cases the faulting also affects drainageand alluvium, and obviQusly has been active recently. Anearthquake of ma~itude 6.8 onthe Richter scale occurredin the area in 1961. In some places (F/G4) huge rock slides!possibly induced by seismicity, have obliterated the layering inthe mountains. In the alluvial deposits tone decreases towardsthe valley axes as granularity decreases, and the distriqutarygullies are well defined.

illumination I

Fig. 7.7 One of the most striking SIR-A images is of the KalpinTagh mountains in north-western Xinjiang, China. The areacomprises Lower Palaeozoic carbonates and clastic sedimentsaccreted to Asia by plate tectonic events in the Triassic Period.Their structures were reactivated and disrupted during theCretaceous to Eocene collision of India with Asia, whichformed the Himalayas far to the south. The rugged terrain hasresulted in some layover.. The best displayed of the olderstructures is a thrust from C2 to G2, which is expressed as abright line representing a slope angled steeply to the radarbeam. It transgresses the layering to the south, ~specially nearthe southward lobe atG2, and dips to the north. Determiningthe dip in the layered strata is di;fficult because of theprominence of layover flatirons (B/C5, F /G4). Clear Vs inescarpments at E4 and E/F2 show that the two blocks dipnorth and may be separated by another east-west thrust. Dips

184 Chapter 7

A B c D E F G H

1

2

4

5

N

~ I 10km I

valleys, deflected drainage and shutter ridges, all reminiscentof an active fault. Its sense of movement is left-lateral strike-slip. Another prominent, possibly active fault trends from H2to E/F2, and many others can be picked out, especially if theimage is viewed obliquely from several directions. Obliqueviewing also helps to distinguish different surface texturesrelated to the bedrock. Three units can be distinguished easilyThe dissected terrain (H5) is underlain by a Lower Palaeozoicmetamorphic complex. The flat surface at E/F4/5 is almostdevoid of drainage and probably is underlain by limestone.Along the major fault is a rugged bright area of thrust blocks,but to the north and south are dark, smoother terrains thathave developed on sandstones.

.illumination

Fig:. 7.8 In densely forested tropical terrains, such as this partof the Vogelkop Region in western Irian Taya, field mapping ishindered by the terrain. Aerial photography and other meansof gathenng data in the visible and infrared spectra arethwarted by near-continuous cloud cover. Although radarprovides useful images, like this one from SIR-A, they are notnearly so sharp as those over arid terrains, because thevegetation canopy intercepts a high proportion of the energyand only a muted representation of the ground surfaceemerges. However, given experience it is possible to recognizemany geological features. The most prominent here is a linearfeature trending roughly east-west from Hl to B5. Carefulexamination of the feature around D /E4 reveals truncated

Radar Remote Sensing 185

nearly parallel to the look direction, and are at least partlyburied by sand. The fact that they show so clearly suggests thatthey consist of aligned, joint-bounded blocks buried by sand,each of which acts as a corner reflector. Careful mappingshows a complex pattern with a variety of orientations,including an arcuate set at F2. The granites do not seem to havebeen cut by the dykes and may be later intrusions. The dykesare useful in identifying less obvious granite bodies. AtA/B4/5 a swarm of about 10 east-west dykes are not seen inan area bounded by a vaguely circular set of features. Thiscould be another granite. The dominance of dykes in the imageobscures most details of the older metamorphic complex thatthey cut, probably because the older rocks have radar-smqothsurfaces. Some curvilinear features around B / C3/4 maydefine folded banding in the basement.

illuminationFig. 7.9 Like Fig. 7.4, this SIR-A image of part of eastern Malimay have penetrated a thin veneer of sand, the high reflectivityof which swamps Landsat MSS images with large areas ofbright tones. The dark area in the top left corner is a sand-filled, possibly fault-controlled valley, which is so smooth thatall energy is reflected away .The large circular feature at CSis a diapiric granite penetrating late Proterozoic metamorphicrocks. Its bright rim and dark interior suggest that its outerzones are more resistant than the now sand-covered core.Other granites occur at DIES, G/HS and E1. The last oneconsists mainly of inselbergs set in a sand and gravelpediment, but the roughly circular shape can be picked out. Atits south-western boundary it is truncated by a straight linearfeature-probably a fault. The scene is dominated by hundredsof dykes, which show as bright lines. Most are orientated

186 Chapter 7

I 10km I

illumination

Fig. 7.10 This SIR- A image begins just to the south-east ofFig.'7.8. The top left quadrant is dominated by rugged terraindeveloped on metamorphic rocks. They are overlain by a smalltongue of clasti~ rockSuending north-west-south-east fromA4. The clastic rocks are less rugged, and they have a coarserdrainage texture. They ar~ succeeded to the south-east, beyonda line roughly £Tom B/C5 to HI, by rocks on which there isvirtually no drainage, and"a surface that is pitted with largedeptessions~ They are particularly noticeable in the areaG"/Hl/2Sutfaceslike these are typical of tropical karstdeveloped on carbonates. Only three major streams drain thearea in deep gorges, and most rainfall, of which Irian laya hasno shortage, disappears into sink-holes. A low NNW-SSE

scarp between D4 and E4/5 possibly defines a major faultthrough the carbonates. Indeed, a whole series of mutedlinear features are visible in this monotonous terrain.A roughly east-west line from D /E5 to H2/3 separates thecarbonates from a subtly different kind of landscape to thesouth. Along the boundary can be seen several small scarpswith clear V-shapes (G/H3/4). These are clastic sediments,which dip to the south. The line of V s is, in fact, a zone oflocally steepened dip (a monocline) flanking a majorsedimentary basin to the south. Within the basin thedrainage becomes clear once more because the rocks thatilll it are impermeable.

tropical rain forest of various rock-related textures thatradar has enhanced.

A further result of the highlighting effect of radar is tomake shapes that characterize erosional or constructiveprocesses more distinctive. This is particularly usefulin glacial terrains, where strong reflection from snowand ice saturates it);lages of shorter wavelengths. Radarimages show variations in surface roughness and there-fore ranges of grey over snow and ice. The topographyshows quite clearly (Fig. 7.1.1), Figure 7.12 picks outvolcanic cones and the effects of dissection quite well.Much smaller and more intricate drainage channels

response from th~ cell that contains it and thereby showup on an image. Sometimes the back-scatter swampsseveral cells to produce a star-like bloom. Alignedcomers resulting from joint blocks along a small faultor dyke help highlight a linear feature that wouldotherwise be invisible on an image of reflected shortwavelength radiation (Fig. 7.9). .

Topographic sharpening also magnifies the contrastassociated with smaller scale features, such as drainagepatterns and other textural attributes of the surface. Theybecom~ easier to distinguish and to interpret in litho-logical terms. Figures 7.8 and 7.10 show examples in

Radar Remote Sensing 187

A B c D E F G

~

N

illumination \Fig. 7.11 The knot of mountains at the juncture of the Pamir,Salt and Karakoram Ranges formed during the indentation ofTethyan marine sediments, island arcs and microcontinentswhen India finally collided with Asia some 90-70 Ma.Together they form one of the most rugged terrains on Earth,with the most active erosion known. This SIR-A image of partof the Karakoram on the India-Pakistan border is, as a result,dominated by layover and shadow effects. Very few, if any,details relating to geological structure can be seen becauseof these distortions. However, it contains probably themost spectacular features of mountain glaciation to be seenapywhere. Snow and ice caps at B2/3 and E/FJ/3show asrelatively homogeneous grey areas through which the majorpeaks show as bright, foreshortened slopes facing the antennaand dark shadows on the opposite side. The relatively brightreturns from the ice fields indicate moderately rough surfaces.Each ice field forms the source for major valley glaciers severalkilometres across. The glacier in the centre is about 50 km long.Dark and light stripes in it correspond tp ice and rpugher

10 km

medial moraines. Each of the1noraines extends to the entrypoints of tributary glaciers. Bright strea~ running across thecentral glacier are probably ice falls riven with crevasses, thosenormal to the look direction being preferentially highlighted.The snout of the main glacier, vaguely defined as a down-valley lobe, is at E/F5. Below it the valley has a dark signature,suggesting that it is flat-bottomed. Towards the snqut, theglacier becomes brighter and more uniform in tone as meltingand ablation reveal more debris and disrupt the pattern ofmedial moraines. Two tributary glaciers at C3 and 03 appearanomalous, with a uniform black signature. This probablyresults from their sloping steeply away from the antennaand receiving very little incident radar energy. Altho~hdistorted, the radar image shows far more topographicdetail than images of other parts of the spectrum do. Theywould be overwhelmed by the uniform reflectivity and lowtemperatures ofsRQw and ice, which formed a near totalcover in the area at the time of the SI!{-A overpass<Novemb!!r 1981).

wavelengths. The worst are those with IQw relief andmono~onous reflectivity, and where natural patterns aremasked by regular agricultural fields. Both are commonin several parts of the world. Included among them areareas with thin highly reflective sand cover, dark desertvarnish, burn scars and thin vegetation. The partialdependence of radar back-scatter on surface roughness

showup on radar images than on those from other sys-tems (fig. 7.13). Fi~re 7.14 shows how radar images ofsandy desert landforms are often disappointing becauseof the general smoothness of sand and its specularreflection of radar energy .

Some terrains present problems for geological iJlter-pretation of imi;lges that use reflected and thermal

188 Chapter 7

10 kmillumination

Fig. 7.12 The volcanic edifice of Kuhha-ye Sahand, south ofTabriz in north-~estern Iran, clearly reveals its structure onthis SIR~A image. The m,arkedradial drainage ts a commonfeature of dissected andesitic stratovolcanoes. The valleysradiate outwards from what was a central caldera complex atD /E3, but which is now barely discernible because of deeperosion. Each major valley has an intricate dendritic systemof tributaries, in which no annular element can be seen.This suggests a relatively homogeneous structure, perhapsdominated by pyroclastic rocks. Thick lava units separated

by more easily eroded tuffs coutd tend to impart a concentricelement to the drainage. Several yeunger, parasitic volcanoescan be seen on the flanks of the major edifice from E2 to H3,and around the central zone. Each is clearly a cone, and one haspreserved a summit crater (H2/3). On the flanks of these smallvolcanoes dissection has also b!'!gun to impart a radial drainagepattern. In several cases the development of parasitic conesappears to have deflected drainage on the flanks of the mainvolcano.

allows nonnally invisible differences in microrelief andvegetation cover to show through such optical masks.In many cases, patterns of great geological significanceare revealed as the following examples show.

Basaltic lavas are dark in the visible and near-infrared"-

regions. Radar potentially is capable of discriminatingrough aa flows from smooth pahoehoe flows, andbetween flows of different ages for which roughness isvariably modified by vegetation, soil cover and degrada-tion: (Fig: 7.15). The response to weathering of the rocksfrom which it fonned often controls soil granularity.Likewise, different grain sizes of alluvium and othersuperficial deposits relate directly to varying energiesof transport and deposition. Chemical sediments, suchas salt and gypsum in enclosed evaporating ba~ins canshow variations in roughness owing ,to shrinkage and

localized solution. Such differences show up as tonalpatterns on radar images (see Fig. 7.20). They are addedto by the changes in dielectric constant induced bydiff~rent moisture cont~nts. In flat-lying areas, there-fore, the tonal variations on radar images can be useddirectly for geological mapping (Fig. 7.16). However, inmore rugged terrains the same effects, though present,are modified by the influence of slope. The radar sign-ature of a uniform surface type changes with the angleand direction of slope and therefore incidence angle(Fig. 7.3). If slopes are chosen carefully, it is possible tos~ through this and map lithologies with a fair measureof "accuracy .

U;nder conditions of extreme dryness, a proportionof the incoming radar energy penetrates soils. Drynessalso permits its return to the surface after subsurface

Radar Remote Sensing 189

10 kmillumination

Fig. 7.13 The Huang He (Yellow River) in Shanxi Province,China (AI to AS) carries a high load of suspended sediment,owing to its catchment comprising large areas mantled bywind blown dust, or loess. This SIR-A image illustrates thecomplex dendritic patterns of the innumerable tributaries ofthe Huang He flowing westwards from the mountaills in thecentre of the image. The changes in structure of drainage fromthe Hei-ch' a Shan and Luliang Shan mountains to the lowerterrains on either side mark major lithological boundaries. Themountains are complex Palaeozoic basement unconformablyoverlain by Triassic sediments in the lowlands to the east andfaulted against the Triassic rocks to the west. The brighterreturns from the mountains are to the result of volumescattering from ~he dense forest canopy above 1500 m. The

lowlands have sparse vegetationexcept in the valleys, aridthey show typical badland topography. Many of the smallertributaries to the west of the mountains display a pinnatedrainage pattern. This is typical of areas ma'ntled by thick,.homogeneous loess deposits. In som~areas this mantleremains, but much of the drainage is probably inherited froma now-eroded loess cover. Although little geological structureis immediately apparent in the lowlan!;is, there is a hintof a roughly north-south grain immediately west of themountains. Several major valleys have nearly straightcourses, which may reflect faults cutting the Triassicsediments (A2, A4 to D4). The bright areas from A4 to D4and round H4 are major towns in this otherwise sparselypopulated region.

scattering. In hyperarid deserts this means that radarimages sometimes give direct information about buriedfeatures. Radar is the only remote-sensing methodwhere this is possible, but only field studies can confirmthat penetration has indeed taken place. A possibleexample was shown in Fig. 7 A, but in that case thesignificance of the patterns related only to buried rivercourses. They probably stemmed from variations insubsurface grain size of the sand. Of more importanceto the geologist is energy scattered back to the surfacefrom the interface between sand and bedrock. Thisallows the topography of the buried surface to be ana-lysed. Erosional patterns pre-dating the inundation by

sand can reveal the sam!,! structural information as barerock surfaces. The only difference is that energy sc~t-tered beneath the surface is attenuated, and the patternsappear relativelyC~ark on radar images (Fig. 7.17). Asradar refraction at the surface reduces the incidenceangle and wavelength, however/the back-scatter some-tim:escanbe increased as more of the subsurface satisfiesthe roughness criterion (Equation 7.2).

7.2.2 Varying look direction

The position of the Sun restricts the illumiriation direc-tions that' other remote-sensing' systems can exploit.

190 Chapter 7.illumination

A B c D E F G

1

2

3

1

2

N

solar illumination tFig. 7.14 The top of this figure is part of a SIR-A image of theBadain J aran desert ~ Inner Mongolia, and tJte bottom shows aLandsat MSS band 5 image of the same area. The comparison isqliite instru<;tive. The Badain Jaran contains the world's largestsand dunes, up t0300m high. On the Landsat im~ge theysbow clearly, together with s~aller dunes superimposed onthem. In the west they are huge composite dunes with acrescentic or barchan-like component. Towards the mountainsthey grade into sharp-crested ridges with a north-westerlytrend. Some of the interdune depressions contain spring-fedlakes. On the SIR-A image very little information on dunestructure is visible. This is because sand is smooth to L-bandradar, and only dune slopes facing south return energy to theantenna. The main trend of dunes is parallel to the lookdirection, so the whole area appears mainly dark. The bright

10 km

patches are favourably orientated dune faces and thickets ofreeds surrounding the interdune lakes. The low informationcontent of radar images of sand-cover~d areas is in sharpcontrast with the response from areas of outcrop. There, theLandsat image is 'washed out' by highly reflective thin sandcover, whereas the radar image is sharp and informative. Thisis a result of low returns from deep sand-filling depressions,and high returns from both bare rocks and surfaces with onlythin, dry sand cover, which th~ radar has penetrated. Muchmore structural information is present in the radar image.Rock surfaces in deserts are generally rough because of winderosion. At E/F3 on the radar image can be seen evidence forthis in the form of narrow ribs of rock trending north-east forabout 10 km. These are yardangs, or aerodynamic ridgessculpted by sand-laden prevailing winds.

Radar Remote Sensing 191

GA B c D E F H

2

5

N

illuminationFig. 7.15 The Jebel al Dmz in western Syria is a largeQuaternary volcanic field dominated by basaltic lavas, thenorth-western part of which is shown on this SIR-A image.Much of the surface is extremely rough, so the lava field showsup as medium to light grey-tones. On Landsat MSS images it isvery dark. Here the boundary between lavas and the smooth,dark, older surface on to which they flowed is sharply defined.Its lobate form from E1 to H3 is typical of the flow fronts offluid lavas. Much of the lava surface is speckled with blackpatches. These are pits and depressions in the flows, nowpartly filled with wind-blown sand. As a general rule, lavasurfaces become smoother with age as surface processesdegrade their original morphology, so the youngest lavasshould be the brightest on radar images. At D jE5, E2 and C1are particularly bright surfaces. In each case a small cone isassociated with these rough surfaces, and at 1::::1 the area ismade up of manytongue-shaped lobes, confirming the

\ I 10 km I

presence of young flows. In the area from A2/3 to C2/3 are anumber of small, young cones with vents. Those at A2/3 aresurrounded by a dark patch, suggesting that they producedpyroclastic deposits which form a locally smooth surface.Streaks running to the north-east from this area show how thefiner ashes have been transported by the prevailing south-westerly wind. To the south-east of this cluster of cones is anorth-west-south-east line, which proves to be an array ofsmall cones on closer inspection. These are probably controlledby a fault in the basement. A similar feature extends betweenAl /2 and C3. These north-west-south-east arrays of minorvents are common throughout the Jebel al Druz, and they arefound as far south as southemJordan. They seem to beconnected to faults which splay out from the major left-lateralstrike-slip system along the Jordan valley and the Gulf of

Aqaba.

192 Chapter 7

A B c D E F G H

1

2

3

4

5

N

illumination IFig. 7.16 The Devonian sediments of Western Sahara, whichare shown on this SIR-A image, are mainly siliciclastic, rangingfrom conglomerates to siltstones. They are approximatelythe same in composition; and present similar signatures onimages in the visible and near-iIifrared regions. Here they aremagnificently divided into easily mapped units on the basisof their radar back-scattering properties. The desert surfaceis nearly flat, and the tonal banding represents variations inoutcrop roughness. Because individual beds can be mapped,small-scale stnicturesshow up well, such as the folds from D4to E5 and small faults from F2 to F4 and G2 to G5. The presenceof V-shaped outcrops controlled by small valleys shows thatdip is to the north and the oldest rocks are at C/D5. The bright,irregular boundary from Cl to C3 is an escarpment, facets ofwhich act as comer reflectors to give specular returns. The

I 10 km I

escarpment truncates the Devonian bedding and representsa nearly horizontal unconformity, above which are muchyounger sediments. Their radar signature is monotonousbecause they are undissected. Lines running south-west-north-east on the surface may be sand streakS: parallel to theprevailing wind. Another possibility is that they are relatedto joints. This would explain the uniformly bright signature ofthe escarpment, irrespective of its orientation. Joints with thisdirection would produce many facets on the escarpment thatare perpendicular to the radar look direction, and wouldensure specular returns. Another interestingieature is that thevery few drainage channels in the area have bright signatures.This suggests that they are filled with pebbles andboulderymaterial. The supply of fine-grained material in such a rockydesert is probably low.

enlargement of Seas at radar data for part of the semiaridMojave Desert in California, USA. Image (c) is an aerialphotograph of the left-central part of the area covered by theSeasat image. The radar image shows severalNNW-SSE-trending bright lines, whereas the photograph shows no signof these features, which are in a monotonous area of alluvium.Detailed field mapping, and geophysical surveys across theradar features showed them to be buried quartz-monzonitedykes. That they show up on the Seasat image is the result offour factors: the dryness of the sand cover at the time the imagewas acquired, the fortuitous orientation of the dykes across theradar look direction, marked subsurface relief on the dykeswhich caused them to act as elongated corner reflectors andthe rubbly nature of the products of dyke weatheringcompared with those derived from country rock. Courtesyof Ron Blom, Jet Propulsion Laboratory, Pasadena.

Fig. 7.17 (opposite) (a) The Red Sea Hills, a narrow swathof which is shown by this SIR-A image running south-west-north-east near the Sudan-Egypt border, are composedof a complex assemblage of Precambrian and LowerPalaeozoic metamorphic and i~eous rocks.. The bright areasare free of sand cover;and they are very rough. They arealmost black in the field, owing to desert varnish. The darkerareas are mantled by high-albedo sand. As the sand is verydry, radar has penetrated it to show features at the sand-rockinterface, but has been attenuated. Large numbers of linearfeatures show up, some of which control subsurface extensionsof the wadis that cut the exposed rock surface. As well asaiding geological mapping of the areas, the penetratingcapacity of radar provides a unique means of predicting wheregroundwater may have accumulated in the subsurfacechannels. Image (b) was produced after digital processing and

Radar Remote Sensing 193

(a)

(c)(b)

~illumination

..illumination

I 1 km I I 1 km I

194 Chapter 7

Radar Remote Sensing 1.95

N110 km

illumination IFig. 7.19 The circular patches on this SIR-A image of northernUbya are fields irrigated by centre-pivot sprinkler systems,supplied by deep boreholes. Only furrow slopes normal to theradar illumination are able to reflect energy back.to the sensor.Closer examination shows that .the circles have structureshaped like bow ties because the curvature of the furrowsbecomes less extreme towards the perimeters of the fields.This effect is the result of corrugations{arsmaller than theimage resolution and may alsoocc4f on ribbed rock outcrops.

7.2.3 Varying depression angle

Whereas any radar image can give information on sur-face roughness, radar science presents two possIbilitiesfor finer division of this attribute. Section 7.2.5 examinesthe potential of different radar frequencies. This sectionconcentrates on what may be possible with different radardepression angles. Figure 7.3 implies that steep anglesgive a maximum response from surfaces that are smooth

Radar is subject to no such hindrance because it ispointable, although orbital parameters restrict this fromsatellites. Topographic enhancement by low SUn angle islimited to early morning and late afternoon. Satellite sys-tems, such as Landsat, gather data at fixed times of day.This suppresses topographic features roughly orientatedalong the solar illumination direction. Although thedegree of suppression of features parallel to illumina-tion is greater in radar images than with other methods,only two or three carefully planned sorties reveal every-thing that is detectable. Orbital systems permit fourlook directions-to either side from both ascending anddescending orbits. Selective pointing of the antenna byelectronic means can increase the range of directionseven further.

There are two main advantages of multiple look direc-tions. First is the selective enhancement o{ differentlyorientated topographic features. Careful planning allowsthe analysis of structural features with all trends andgeometries. Moreover, features hidden in shadow fromone look direction are revealed from another direction.Figure 7.18 shows examples of SIR-A and Seasat imageryof the same area taken from orbits inclined at about 450to each other.

Radar's other advantage relates to directional featuresat the surface that are too small to be resolved, such asparalle1 corrugations in ploughed fields and on weath-ered outcrops of finely banded or foliated rocks. Radarlooking along the corrugations responds as though thesurface is smooth-it appears qark. A look direction acrossthem gives the effect of roughness-the surface appearsbright-provided the amplitude and wavelength of thecorrugations satisfy the criterion in Equation 7.2. Forsand ripples these conditions are not met-they are notsharp enough. Although some research has been directedat detecting hitherto undiscovered potato crops, thepotential for geologicalanalysis so far remains untapped.Figure 7.19 shows this effect in a rather unusual agricul-turaJ context.

There is one major disadvantage of using differentlook directions. The appearance of rugged topography isvery different on radar images looking in two separatedirections (Section 3.5; Fig. 7.18) thereby disrupting

interpretation.

Fig. 7.18 (opposite) The top image is from SIR-A and the lower;from Seasat. The image swaths were gathered from orbits 'inclined at about 450 to each other. The area is the Santa YnezMountains of coastal California, USA, centred on Santa Barbaraat E/F4. There are obvious differences between the two images,owing to the different depression angles of the two systems. TheSeasat image has stronger layover, and it shows subtle featuresat the sea surface. However, many of the differences stem fromthe contrasted look directions. The strike of layered rocks alongthe coast is shown well by SIR-A, because it is favourably

orientated towards the look direction, This attribute is disruptedon the Seasat image by strong returns from north-south valleysides. Running from VI to B/CS is the Santa Ynez Fault, whichis clear on the SIR-A image, together with indications of thediSplacement of bedding' It is nearly undetectable on the lowerimage. Seasat, however, highlights a major fault fromBI to A/B4that is not visible on the top image because of its unfavourableorientation. The smaller depression angle of SIR-A results in amuch clearer expression of variations in surface roughness.Courtesy of J.P. Ford, Jet Propulsion Laboratory, Pasadena.

196 Chapter 7

a moderately rough surface; however, the Seasat imagecontains more information. In contrast, the eastern chottappears monotonously dark apart from a swirling bright bandon the Seasat image, whereas the SIR-A image shows strongerand more vaIjable ~eturns. The differences could result fromsubtle variations in the roughness of the salt deposits that floorthe chotts, and which are variably enhanced by the differentdepression angles, Courtesy of J. P. Ford, Jet PropulsionLaboratory, Pasadena.

Fig. 7.20 The top image is fromSeasat and the lower from SIR-A L-band radar. Although the illumination directions ate atright angles, the area has too little relief for this to have anysignificant effect. The obvious differences between the twoimages derive mainly from the different depression angles(70° and 43°, respectively). The area is the Sahara Desert ofnorth-eastern Algeria, dominated by evaporite basins, knownlocally as chotts, which are below sea level (B2 to C2 throughB5 to C5, and D2 and G2 through F4 to K4). The western chotthas roughly the same brightness on both images, suggesting

Radar Remote Sensing 197

(a)

~

~

parallaxdifference

(b) ~L--1 ~ r:::J:

for a particular wavelength, according to the Rayleighcriterion. Shallow angles eliminate back-scatter from allbut the roughest surfaces. Angles between 30 and 40°give different responses from all sorts of surface.

Variations in depression angle also accentuate surfacetopography in different ways because of the differentdegrees of layover, foreshortening and shadowing(Section 3.5). In particular steep angles P1aximiZe thediscrimination of subtle changes in areas of low re,lief.This has to be traded against the distortiqn of high-relieffeatures, which are better expressed by shallowdepres-sion angles. In turn, low-angle illumination may misssubdued features in the landscape.

Using different depression angles is then a means ofincreasing information potentially related to both lithol-ogy and structure. Although most research has beenconducted with aircraft data, the most widely flvailablecomparisons come from the Seasat and SIR-A orbitalmissions-both deploying L-band radar with depressionangles of 70° and 43°, respectively. Figure 7.20 givessome idea of tbe potential of systems, such as Radarsat,that permit multiple depression angles.

~

~

parallaxdifference

(c)

::I:::,,=!=~ =---1 1 =!= &: ~:

"""~ (. ""--<::

""1

parallaxdifference

Fig. 7.21 Radar parallax suitable for stereoscopic viewingresults from three main configurations: (a) illumination fromopposite sides; (b) illumination.from the same side but fromdifferent altitudes; and \c) illumination from the same side,at the same height but with different depressionangles.Configurations (a) and (c) are possible from orbit. Then:-ethod that produces best stereoptic models is (c).

7.2.4 Stereoscopic radar images

An inherent characteristic of side-looking radar is itsinevitable introduction of relief displacement on images(Section 3.5). As it measures the down-range distance of anobject in terms of fune taken for energy to reach and returnfrom the object, the displacement is towards the nadir ofthe system. In images produced by optical systems, reliefdisplacement is away from the nadir. Either kind of reliefdisplacement produces parallax between images takenfrom different positions, and therefore the possibility ofstereoptic viewing (Section 2.5, Appendix A). How it isproduced makes little difference to the observer.

There are several radar configurations that producestereoscopic potential (Fig. 7.21). Images with the samedepression angle from each side of an area, or those fromthe same side but with different depression produce

stereoscopic convergence.Depression angle and its geometric effects vary con-

siderably across airborne radar images from near to farrange. On a pair of images the conditions for stereopticfusion and for vertical exaggeration change from placeto place. The observer may need to be an ocular athlete.This variation is almost absent from orbital images, sothey are potentially much easier to use.. The main prob-lems are those of steep slopes and rugged relief, andtheir influence on layover, foreshortening and shadoweffects. Provided such effects all have the same sense,as they do on pairs of images acquired from the sameside of an area, information is sensible and can producephotogrammetric quality stereopairs. Images taken withopposed look directions are all but useless in areas of high

relief. For tms reason, the majority of stereoscopic radarmissions adopt the same-side, same-height configuration(Fig. 7.21c). Tms is possible from a fixed orbit, providedthe antenna can be pointed. Figures 7.22 and 7.23 giveexamples of stereoscopic radar images from aircraft.

7.2.5 Multipolarization radar

A variety of interactions between radar, the surface coverand subsurface materials can alter the polarization of

198 Chapter

Fig. 7.22 In southern V e~ezuela, near to the Guyana border,the spectacular plateaux formed by fault-bounded blocks ofthe Cambrian Roraima Formation were the inspiration forConan Doyle's 'Lost World'. Although not infested withdinosaurs, they are genera~y shrouded in cloud andcomprehensive field work is not possible. This L-bandstereopair~ produced from an aircraft by the same-side,

same-height method, reveals many details of the localstructure, including evidence for an unconformity wherethe Roraima Formation oversteps a dyke in the crystallinebasement in the left-hand part of the pair. Courtesy of EarlHajic, University of California, Santa Barbara. Imagesproduced .by Goodyear Aeros~ace Corporation.

radar waves (Section 7.1.3). By examining different com-binations of transmitted and received polarizations it there-fore should be possible to add yet more flesh to geologicalinterpretation of radar images. In a way analogous tomultifrequency radar it helps discern subtle differencesin surfac~ ro~ghness, vegetation cover and subsurfacegranularity, which are not visible on single radar imagesor on those from other parts of the spectrum. Informationof this kind is us~ful in an empirical sense for rock dis-crimiIlation and highlighting anomalies. It is of more valueto t~e botanist. or agricultural specialist in quantitativemeasures of the ki!1ds and densities of vegetation cover.

Different send-receive combinations can be comparedas separate images, but as always, their combination asfalse-colour images makes far better use of the eye's acuityin comparative studies. Figure 7.24 and Plate 7.1 showexamples of the potential of multipolarization radar imagesapplied to the study of varied sequenc;:es of bedrock.

the wavelength dependency of the Rayleigh criterion(Section 7.1.2, Table 7.1). This requires illumination of anarea by radar using several frequencies. As well as aid-ing ~he mapping of different lithologies and superficialdeposits in terrain bare of vegetation, images of back-scattered energy for a range of frequencies can provideinteresting information from densely vegetated areassuch as jungles.

At relatively short wavelengths (Ka- and X-bands) thereturns are mainly from any plant canopy that is present,rather than from the surface. The high dielectric constantof vegetation, owing to its water content, makes all partsof plants highly reflective to these wavelengths. Radartnerefore is scattered primarily by leaves, and to a lesserextent by twigs and branches. At longer wayelengths(L-band) the dielectric constant is smaller and greaterpenetration is possible. This means that the back-scatteris influenced by the ground surface beneath grasses. Inforests, even L-band radar fails to penetrate the canopy,and back-scatter is mainly from twigs and branches.Combining different wavelengths therefore should giveinformation relating to three attributes of the surface-itstopography ,its roughness and the density of vegetation

7.2.6 Multifrequency radar

The most powerful means of discriminating betweensurfaces with different degrees of roughness is to exploit

Radar Remote Sensing 199

Fig. 7.23 This same-side, same-height stereopair of X-bandairborne radar images is of part of the same area shown in Fig.7.5.

It shows considerable detail of the folding and faulting inthe Carboniferous sediments of the Appalachians near

Pittsburgh, Pennsylvania, USA, and its quality is good enoughfor accurate photogz:ammetry. The im~ges were produced bythe

STAR digitalSAR system, anq are provided by courtesy ofIntera Technologies Ltd, Calgary.

such information is in providing an additional dimensionfor geological interpretations of an area.

cover. This potentially is of interest in the search forgeobotanical features related to geochemical anomalies.It is certainly a means of looking at the different vegeta-tion covers that sometimes distinguish one lithologyfrom another.

Radar images of unvegetated areas remove the com-plicating effects of vegetation, so that varying responsesto different radar frequencies are more easily inter-preted geologically. Figure 7.25 shows three images thatuse different radar frequencies over a bare volcanicterrain.

Despite the restriction of its deployment to a limitednumber of orbits, the SIR-C/XSAR experiment carriedby the Shuttle in 1994 produced the largest availablearchive of multifrequency radar images. Three wave-lengths (L, 23 cm; C, 6 cm; X,3 cm), together with vari-ous polarization modes provide a variety of RGB imagepossibilities. Because the Rayleigh criteria for each band(Equations 7.1 and 7.2) give> 7.5, 1.9 and 0.97 cm for'rough', and < 1.3,0.34 and 0.17 cmfor 'smooth', respect-ively, for L-, C- and X-band wavelengths. The colour insuch images is a sensitive guide to the nature of rock andsoil surfaces. Plate 7.2 is a good examPle of how useful

7.2.7 Digital processing of r~dar images

Processing of digital radar -data permits geometriccorrection and removal of noise (speckle) inherent tothe way in which they are captured. Multifrequencyandmultipolarization data, combined as RGB images, lendthemselves to various kinds of enhancement that helpsmake the images more easily interpreted visually. Itmight also seem that variation in DN, resulting partlyfrom surface roughness, can contribute to classificationwhen combined with other data. However; the domin-ant control over radar tone at all scales is topography.This has a much greater contribution than in images 6freflected and emitted radiation. Consequently it tendsseriously to disrupt the accuracy of classification in allexcept the most subdued terrains.

The main opportunity for automated informationextraction from radar data lies in texture analysis andenhancement. There is far more textural informationin a radar image than the eye can handle and sort

200 Chapter

(a)

N

\N\~(azimuth)

.-illumination

~(azimuth)

.illumination

(b)

Fig. 7.24 These images are from an airbomeL-band radarstudy of variations in back-scattered energy using variousmodes of polarization: (a) HH, (b) VV and (c) VH. The targetarea is in the Wind River Basil1 of central Wyoming, USA, andcomprises a uniformly dipping sequence of Triassic to lateCretaceous limestones, siltstones, shales and sandstones. Theoldest rocks outcrop in the west of the image, and the dip isroughly eastwards. The illumination direction was chosen tominimize the effects of topography, mainly north-southescarpments. At first sight there appears to be little differencebetween the images of different polarization arrangements.Indeed, in this form the data are highly correlated so that anyone image is as useful as the others. However, combining thedata in a false-colour image shows the subtle effects ofdifferent surfaces on polarization (Plate 7.1).

N

\~(azimuth)..

illuminatioh

into categories. So, processing radar image textures bycomputer aims to simplify and categorize them. Themost obvious tr~atment is to use edge enhancementand directional filtering (Secnon 5.3) to extract the max-imum amount of information about curvilinear featuresthat are present in an image. This, however, ignoresa lot of the variation ~ the image ,that lies betweenlinear features and is distributed in more or less homo-geneous patches, and serves to decrease the alreadypoor signal-to-noise ratio. To extract information relat-ing to this aspect of the data mea~ a more quantitativeapproach.

There are two main ways of quantifying spatial informa-tion, of which texture forms a major part. The simplest isspatial frequency ffitering (Section 5.3). Tonal variationswith low spatial frequencies are most usually influencedby the overall changes in roughness in an area. Thisis fundamental lithological and vegetation information.The higher frequency variations stem from all the differ-ent topographicfeatures-,-scarps, ridges, valleys, gullies,pits and hummocks. They contain much of the structuraland textural information. Separating the two gross com-ponentsby low- and high-pass ffitering (Fig. 7.26) enablesa colour rendition to be extracted from a monochrome

Radar Remote Sensing 201

Fig. 7.25 The area shown is the hugeErta' Ale volcano astride an activemantle plume in the DanakilDepression of Ethiopia. The imagesare from the 1994 SIR-C/XSARmission of the Space Shuttle: (a) L-band (23 crn), (b) C-band (6 cm) and(c) X-band (3 crn), with flight andillumination direction, and northarrows at lower right of each. Themarginal ticks have a 10-kIn spacing.The roughly elliptical volcanooccupies the central part of all threeimages. It is surrounded by areas ofalluvial sediments, and the blackareas at the left and right of all threeimages are lakes. To some extent thebrightness of surfaces on the volcanodepends on the age of individuallava flows, as lava surfaces are oftenblocky and angular immediatelyafter eruption, but degrade throughweathering and erosion to smoothersurfaces. The brightest surfacesare probably the youngest flows.Brightness of the volcano surfacebecome greater and more uniformfrom L- to X-band, as a result ofthe decreasing dimension of theRayleigh roughness criterion atshorter wavelengths. L-band ismore effective at discriminatingindividual flows and their relativeages. Alluvial areas are generallysmoother than bare rock, but theincrease in overall brightness of thesurroundings of the volcano fromL- to X-band images shows howsuccessively shorter wavelengthsrespond more to subtle variationsin surface roughness. This is mostnoticeable at top right. The lack ofany returns from the lakes indicatesthat the water surface was barelyrippled at the time of the overpass.More information becomes apparentafter combining the multifrequencydata in a false-colourimage (Plate 7.2).

(a)

(b)

(c)

widely and rapidlyDN vary, and the finer and more con-trasted the texture. Relatively smooth textures shouldbe expressed by low variance. The procedure involvesmeasuring the variance of ON in the surroundings ofeach pixel, and assigning that to the pixel as a measure ofvariance. The art lies in choosing boxes of the right sizeto express different textures properly. With the radardata in this form it is possible to display an image ofvariance. It is important to note that variance, like bandratios of shorter wavelength data, is little affected byvariations in the angle of illumination.

image, often with spectacular results. The filtered dataare expressed in the HSI system (Section 5.2.2), wheredifferent hues are interpreted more easily in terms oftheir control by terrain features (Plate 7.3). Separationinto high-, medium- and low-frequency components isalso possible, when the image can be reconstructed incolour through red, green and blue filters.

A more complex option is to express texture statistic-ally. As texture amounts to the spatial variation of tonein an image, an appropriate measure is the varianceof ON within an area. The higher the variance the more

202 Chapter 7

(b)

'Surlace' component

Fig. 7.26 Digital separation of different spatial frequencycomponents from a Seasat image of theW amsutter Arch inWyoming, USA {a) results in the production of (b) an imageof low-ft~quency variations related to gross variations inroughness and possibly rock type, and (c) an image of high-frequency variations containing topographic and textural

'Topograhy' componentinfonnation. The high-pass image can be used to controlintensity and the low-pass image to express hue in colourHSI image (Plate 7.3) where saturation is assigned a constantvalue to give the best colour rendition. Cqurtesy of RonaldBlom and Michael Daily, Jet Propulsion Laboratory.Pasadena.

The final step expresses differences in texture by severalimages that have been spatially filtered to enhance arange of spatial frequencies. Each is then recast as avariance image in its own right. The images are thentreated as independent spectral bands, but as part of aspatial spectrum r'lcther than an EM spectrum. These canbe displayed together as colour images or used as ameans of texture classification by multivariate methods(Section 5.6.3). Elegant as this approach might seem,little success has yet been achieved.

7.2.8 Radar combined with other data

Radar images are very different from those of otherspectral regions, so it might seem appropriate .to usethem in.the raw form together with other data sets as anuncorrelated dimension in classification. This is not afruitful option. The great sensitivity to topography wouldmean that landforms played a greater role in classifica-tion than differences in surface properties. There wouldbe gross misclassification. One option IS to use measures

Radar Remote Sensing 203

(a)of topographic texture taken from radar images, such asthe variance data described in Section 7.2.7. However;uses of this kind discard most of the visual informationcontent inherent in radar images. An important use ofradar images therefore is as a means of introducing orenhancing topography and texture into colourful, but lesssharp combinations of other wavebands. The best way ofdoing this is through the HIS transform (Section 5.2.2),where radar data substitutes for the intensity in an RGBimage of other data that contributes hue and saturation.The radar contributes a mass of structural information,the other data provide distinctions between differentrocks (Plate 7.4). Chapter 9 shows other possibilitiesthat introduce topographic information from radar intoimages of geophysical data.

baseline

7.3 Radar interferometry

(b)

baseline

You may be familiar with interferometry from opticalphysics. A puddle of water with a film of oil on it pro-duces patterns of colour known as Newton's fringes onthe surface. These bands result from light rays reflectedby the smooth surfaces of the oil and the water beneathit. Tiny variations in thickness of the oil film causeinterference between the phases of the two sets ofreflected light, to give different colours. The patternsrelate to variations in thickness of the film. Radar inter-ferometry (sometimes given the acronym InSAR) issimilar and generates interference patterns between twosets of radar signals scattered from the Earth's surface.The difference is that interference between the twosignals results from the time lag between one signal andthe other, both of which emanate from the same objecton the ground. This results from the differences in lengthof the two radar paths. These lengths are measurablein terms of the radar wavelength. If they are exactly thesame then the two waves will be in phase. If there is adifference, then the signals will be out of phase to somedegreerand will interfere in the same manner as do lightwaves in Newton's fringes.

There are two common configurations for radar inter-ferometry. Two antennas separated by an accuratelyknown distance form the set-up for the fixed baselinemethod (Fig. 7.27a). An alternative is using back-scattered signals from overpasses on different dates.

Combining two interferometric radar signals producesan interferogram or fringe map, which looks similar tothose seen in a film of oil. In the case of a fixed-baselineusing two antennas .the fringes represent differencesin elevation of the surface,and can be recast as a digitalelevation model (Chapter 8). Interferometric fringesproduced by registering radar signals from two over-passes of a single receiving antenna, as on ERS-l or -2,

Fig. 7.27 (a) Antennas for fixed baseline radar interferometry(b) Radar wave paths to fixed-baseline antennas for points atdifferent surface elevations.

-represent any changes in elevation at different pointson the scene. Such changes might occur during groundsubsidence that results from landslips, mining or move-ment on a fault line, or because of inflation of a volcanoas magma is emplaced prior to an eruption.

In order to see how fixed-baseline radar interferome-try is sensitive to the height of the target, Fig. 7.27(b)

204 Chaptl

Further readingshows two different points with differ~ntelevations, inrelation to radar waves. You can see that the differentdistances of each of these targets from the two antennasthat constitute the baseline depends on their elevation.For the higher target (target 2), the difference is greaterthan for the lower one (target 1). Consequently, theinterferometric phase shift between the radar waves fortarget 2 is therefore larger than it is for target 1. As differ-ential distance gets larger so too does the incidenceangle at the target (so, 91 < 92), The interferometric phaserelates to the incidence angle by:

interferometric phase =B sin«(J)/wavelength (7.3)

where B is the baseline length. Incidence angle (9) can becalculated fr:om the measured interferometric phase andknown radar wavelength. The topographic elevation thenstems from geometry. Accuracies of the orde£.of a fewmetres are possible from orbit.

The calculation of changes in elevation from dual-dateradar interferometry is more complicated. As the radarequivalent of Newton's fringes are in steps of one wave-length of phase shift, the changes can be assessed inmillimetres-each fringe in a 10-cm radar system isdivisible into many lesser steps.

The European Space Agency's ERS-1 is an operationalsource of dual- or multidate interferometry. NASAlaunched a Shuttle mission (the Shuttle Radar TopographyMission-SRTM) in February 2000, which carried a fixed-baseline InSAR system aimed at providing a digital ele-vation model for 80% of the Earth's land surface between600N and 56°S. The SRTM reused C- and X-band SARinstruments from the 1994 SIR-C/XSAR mission, withthe addition of second antennas on a mast extendingfrom the Space Shuttle's payload bay. The resultingDEMs will have a horizontal resolution of 30 m and oneof up to 4 m for elevation. This comprehensive mappingto very high geodetic standards took a mission last-ing only three weeks. By November 2000, NASA JetPropulsion Laboratory and the German Space AgencyDLR had released a few preliminary SRTM data, globalcoverage being scheduled for 2002.

Associated resources on the CD-ROM

You can access the resources by opening the HTML fileMenu.htm in the Imint folder on the CD using your Webbrowser. More detailed information about installing theresources is in Appendix C.

On the Menu page, select the link Miscellaneous:Image Data, where you will find examples of imagesbased on a variety of SAR techniques. Of great interestare OEMs made using InSAR, which you can find in thelink Miscellaneous: Digital elevation.

Amelung, F., Galloway, D.L., Bell, J.W., Zebker, H.A. &Laczniak, R.J. (1999) Sensing the ups a!1d downs of LasVegas: ISAR reveals structural control of land subsidenceand aquifer-system deformation. Geology 27, 483-486.

Berlin, G.L., Schaber, G.G. & Horstman! K.C. (1980) Possiblefault detection in Cottonball Basin, California: and applica-tion of radar remote sensing. Remote Se:l1sing of Environment10,33-42.

Blom, R.G. & Daily, M. (1982) Radar image processing for rock-type discrimination. IEEE Transactions, Geoscience and RemoteSensing GE-20, 343-351.

Blom, R.G., Crippen, R.E.& Elachi, C. (1984) Detection of sub-surface features in Seasat radar images of Means V alley,Mojave Desert, California. Geology 12, 346-349.

Blom, R.G., Schenk, L.R. & Alley, R.E. (1987) What are the bestradar wavelength, incidence angles and polarization fordiscrimination among lava flows and sedimentary rocks?A statistical approach. IEEE Transactions, Geoscience Remote

Sensing GE-25,208-212.Curlis,J.D., Frost,V.S. & Dellwig, L.F. (1986) Geological mapping

potential of computer-enhanced images from the ShuttleImaging Radar: Lisbon Valley anticline, Utah. PhotogrammetricEngineering and Remote Sensing 52, 525-532.

Daily, M. (1983) Hue--saturation-intensity split-spectrum pro-cessing of Seasat radar imagery. Photogrammetric Engineeringand Remote Sensing 49, 349-355.

Daily, M., Elachi, C., Farr, T. & Schaber, G. (1978) Discrimina-tion of geological units inpeath Valley using dual frequencyand polarization imaging radar data. Geophysical ResearchLetters 5,889-892. -

Daily, M., Farr, T., Elachi, C. & Schaber, G. (1979) Geologicalinterpretation from composited radar and Landsat imagery.Photogrammetric Engineering and Remote Sensing 45,1109-1116.

Dellwig, L.F. (1969) An evaluation of multifrequency radarimagery of the Pisgah crater area, California. Modern Geology

1,65-73.Dellwig, L.F. & Moore, R.K. (1966) The geological value of

simultaneously produced like- and cross-polarized radarimagery. Journal of Geophysics Research 71, 3597-3601.

Elachi, C. (1983) Microwave and infrared satellite remote sen-sors. In: Manual of Remote Sensing 2nd edn (ed. R.N. Colwell),pp. 571-650. American Society of Photogramrnetry, FallsChurch, Virginia.

Elachi, C., Brown, W.E., Cimino, J .B., Dixon, T., Evans, D.L., Ford,J.P., Saunders, R.S., Breed, C., Masurski, H., Schaber, G., Dellwig,L., England, A., MacDonald, P., Martin-Kaye, P. & Sabins, F.(1982) Shuttle imaging radar experiment. Science 218, 996-1003.

Elachi, C., Roth, L.E. & Schaber, G.G. (1984) Spaceborne radarsubsurface imaging in hyperarid regions. IEEE Transactions,Geoscience Remote Sensing GE-22, 382-387.

Evans, D.L., Farr, T.G., Ford, J.P., Thompson, T.W. & Werner,C.L. (1985) Multipolarization radar images for geologicalmapping and vegetation discrimination. IEEE Transactions,Geoscience Remote Sensing 23, 275-281.

JPL (1980) Radar Geology, an Assessment. Publication 80-61, JetPropulsion Laboratory, Pasadena, CA.

JPL (1982) The SIR-B Science Plan. Publication 82-78, JetPropulsion Laboratory, Pasadena, CA.

Radar Remote Sensing 205

Kasischke, E.S., Schuchman, A.R., Lysenga, R.D. & Meadows,A.G. (1983) Detection of bottom features on Seasat s~h~ticaperture radar imagery. Photogrammetric Engineering andRemote Sensing 49, 1341-1~3.

Lowman, P.D., Harris, J., Masuoka, P.M., Singhroy, V.H. &Slaney, V.R. (1987) Shuttle iinaging radar (SIR-B) investiga-tions 0.£ the Canadian Shield: initial report. IEEE Transactions,Geoscience Remote Sensing GE-25, 55-66.

MacKay, M.E. & Mouginis-Mark, J.P. (1997) The effects of vary-ing acquisition parameters on the interpretation of SIR-Cradar data: the Virunga volcanic chain. Remote Sensing ofEnvironment 59, 321-336.

Martin-Kaye, P.H.A. & Lawrence, G.M. (1983) The applicationof satellite imaging radars over land to the assessment, map-ping and monitoring of resources. Philosophical Transactionsof the Royal Society of London Serif;sA 309, 295-314.

M~auley, J.F., Schaber, G.G., Breed, C.S., Grolier, M.J.,Haynes, C.V., Issawi, B., Elachi, C, & Blom, R.G. 1982. Sub-surface valleys and geoarcheology of the Fjstem Sahararevealed by Shuttle radar. Science 218, 1004-1020.

Pravdo, S.H., Huneycutt, B., Holt, B.M. & Held, D.N. (1982)Seasat Synthetic-aperture Radar Data User's Manual. Publica-tion 82-90, Jet Propulsion Laboratory, Pasadena, CA.

Rudant, J.P., Derom, J.-P. & Polidori, L. (1994) Multi-resolutionanalysis of radar iinages and its application to lithological

and structural mapp~: Lafzac{southern France). InternationalJournal of Remote Sensing 15,2451-2468.

Sabins, F.F. (1983) Geological interpretation of Space Shuttleradar images of Indonesia. Bulletin, American Association ofPetroleum Geologists 67,2076-2099.

Sabins, F.F., Blom, R. & Elachi, C. (1980) Seasat radar image ofSan Andreas Fault, California. Bulletin, American Associationof Petroleum Geologists 64, 612-628.

Schaber, G.G., Berlin, G.L. & Brown, W.E., Jr (1976) Variationsin surface roughness within Death Valley ,California: geolog-icalevaluation of 25-cm wavelength radar images. GeologicalSociety of America Bulletin 87, 29-41.

Schaber, G.G., Elachi, C. & Farr, T.G. (1980) Remote sensing ofSP Mountain and SP Lava Flow in North-Central Arizona.Remote Sensing of Environment 9, 149-170.

Schaber, G.G., McCauley, J.F. & Breed, C.S. (1997) The useof multifrequency and polarimetric SIR-C/XSAR data ingeological studies of, Bir Safsaf, Egypt. Remote Sensing ofEnvironment 59, 337.,-363.

Wadge, G. & Dixon, T.H. (1984) A geological interpretationof Seasat-SAR imagery of Jamaica. Journal of Geology 92,561-581.

Wright, P. &. Stow, R. (1999) Detecting mining subsidence fromspace. International Journal of Remote Sensing20,1183-1188.

The symbolism used in maps varies (Section 8.1), andis sometimes out of tune with human visual perception.Many maps show variations in a third dimension ascontour lines. Human vision treats lines and tonal varia-tions in different ways (Chapter 2). Our perception of acontour map depends partly on the tonal variation pro-duced by the spacing of contours-the closer the lines arethe darker the appearance. The spacing of contour linesrelates to gradients in the third dimension, however, sowe see a contour map as an inverse representation ofgradient. So even if contours are carefully labelled thereis conflict between what is learned about contour mapsand their interpretation and innate human perception.Contour maps are notoriously difficult to relate to thereal world. Section 8.2 centres on converting informationthat is generally rendered as contour maps into imageform. Images are more readily understood and amenableto enhancement using di~tal image processing.

The variety of data that a geologist may need to as-semble, brood over and interpret can be represented as aseries of layers (Fig. 8.1). Each layer contains informationpertinent to a specific element of the task. To assess wateJresources for domestic and agricultural uses a hydrogeo-lo~st might need layers for geology, soil type, rainfall.land use, hydrology, data from wells, topography,land ownership, transportation routes and settlementsFrom these basic layers it would be necessary to extrac1information on the distribution of aquifers, well yieldsstream gauging data, faults, topographic slope, the fac.ing direction of slopes (slope aspect), and so on. This is c

soi /-}

Earlier chapters have dealt almost exclusively with datathat in one way or another were gathered in the form ofimages. These contain a wealth of information fromwhich geological inferences can be made and synthe-sized in the form of maps. They are restricted to the useof EM radiation from the Earth's surface, however, andthe spectral response of rocks poses fairly tight limits ongeological interpretation. Except where structural fea-tures can be extrapolated in three dimensions,. it also lim-its interpretation to, at most, a few kilometres of thecrust, where uplift and erosion have produced extremerelief. Geologists areconcemed with ever deeper levels,partly because the search for resources has,to extendits scope as near-surface deposits become depleted, andpartly because other kinds of data enable deep levels tobe investigated. Indeed, the one presupposes the oilier.

Data that give clues to subsurface geology includewell logs, seismic reflection sections, potential-field datafrom gravity afld magnetic surveys, and results fromvarious kinds of geophysical survey orientated at theelectrical properties of rocks, soils and minerals. Thereare many other kinds of data with an indirect bearingon surface geology, such as measurements of emittedgamma-radiation, and geochemical surveys of rock andsoil samples, vegetation, stream sediments and water.To these 'raw' data can be added information carriedon existing maps. As well as defining lithological andstratigraphic units, and geological structures, maps caninclude sites of resource extraction, finds of economicallyinteresting mineral associations (prospects or showings)and the range of information on surface elevation,hydrology, communications, habitations, land use andownership generally associated with conventional topo-graphic maps.

For decades geologists have used many of theseother data sources in particular tasks, such as genera]mapping, geomorphological. studies, exploration foJhydrocarbons, metals and water, site investigations fOJengineering projects, and assessing environmental con-straints on various operations, as well as in fundamentaJresearch (Chapter 9). The unifying factor among all thesEtypes of information is that they are geographical illnature. They relate to points, lines or areas on the Earth'ssurface expressed in various cartographic co~ordinatEsystems. As paper maps such data present a numbeIof problems. They may be at a variety of scales and illdifferent projections. Even as transparent overlays, thEspace required and the geologist's finite powers of men.tally retaining information pose limits to integrating thEinformation properly.

/7/ land use

2

/

/

';./

/

Fig.8.1 Layered structure of primary geological and relatecdata.

206

Non-Image Data and Geographical Information 207

daunting body of conventionally separate information.An agriculturist working on the same programme wouldneed some of this information too, plus more detail on,say, soil texture, soil depth, drainage properties, contentof nutrients, pH, etc. Map information in paper formclearly poses problems for both, indeed, some informa-tion may well not appear on maps. Inevitably the ana-lysis is slow and largely dependent on intuition andarbitrary selection of information.

An organized catalogue of geographical informationas a series of exactly registered layers in digital form isthe basis for management and interrogation of map datain a geographical information system (GIS). A GIS notonly rationalizes data, but greatly speeds access to them.They become amenable to a variety of objective and effici-ent means of analysis, which encompass all pertinentinformation and thereby broaden the scope for inter-pretation, planning and execution. Section 8.3 introducessome of the basic concepts and techniques in GIS.

8.1 Forms of non-image data

The raw data available for use in geological and manyother kinds of investigation can take a variety of forms:1 values for some variable at a point, for instance eleva-tion, geochemical and gravity data, borehole records,dip angle and direction;2 continuous records of a variable along a line, as fromseismic profiles, airborne and ground geophysical surveys;3 areas designated as homogeneous categories, such aslithostratigraphic, soil and vegetation maps, and resultsof classification;4 lines expressing relationships or linear categories-faults, intrusive contacts, roads and property boundaries.5 continuous or semicontinuous grey-tone or colourimages of remotely sensed data.Types I, 3 and 4 often involve information that is notpart of some continuous variation, but expresses definedcategories or attributes. Obvious examples are the metal-bearing minerals found at a mining prospect, names ofrock types, different soil classes and various senses of faultmovement. Moreover, a named category often has otherassociated attributes. A stream will have a particular orderin relation to its connections with other larger and smallerstreams-the higher the order the greater the number oftributaries-and will have a unique catchment size. Itmay be named in its own right or as a tributary of a largersystem. A particular soil type may have variable colour,texture and drainage properties, and also can be categor-ized in terms of how prone to erosion it is. Roads havedifferent numbers of lanes and various paving surfaces, aswell as a designated code. Rocks have many attributes, afew being mineralogy, chemical composition, radiometricage, porosity and even associated plant community.

There are two basiL-models that encode geographicalinformation, known as raster and vector models. Araster model represents the real world by dividing two-dimensional space into cells with definite shapes. Cellsmay be rectangular, triangular, hexagonal or any othershape that allows a systematic tiling or mosaic to coverthe surface. Most rasters use square cells arranged in agrid of rows and columns, in the manner of a remotelysensed image. Their precision is limited by the sizechosen for the cells. Objects and properties referring tothem are represented by attributes assigned to the cells.A cell's attribute applies to every location within it,and the cell cannot be divided. The clearest examplesof raster-format data are provided by digital images,including paper maps captured by various scanningdevices. Maps are a great deal simpler than images,and comprise a limited number of divisions, comparedwith the 224 divisions possible with a colour image of theEarth's surface. Raster models of maps might representthe boundaries between divisions accurately, but that isat the expense of redundancy for the areas that they sep-arate; they are expensive in terms of file size. Other typesof raster model are renditions of numerical nonimagedata recast in the form of a grid of values (Section 8.2.1),which are more image-like than maps.

In the vector model, points and lines define theboundaries of objects or conditions, much as if they weredrawn on a map. A small number of elements make up avector model-points, lines and closed loops (polygons)made of straight-line segments. Lines are defined bytheir end-points or nodes, and by the vertices joiningstraight segments between the nodes, so that complexcurves comprise many segments and vertices. Polygonsare lines closed by a single node. The accuracy of vector-format data is limited only by the precision ofsurveying,the encoding length of computer words (now commonlyup to 64 bits) and the number of points used to representcurved lines and polygons in the real world. Primarily,vector format represents spatial entities, but many attri-butes can be recorded for, and assigned to, each point,line or polygon within a single me. Data input to vectorformat can take many different forms, by keying in loca-tion and attributes at a terminal from field records,through data logging and positional devices such as asatellite positioning system used in the field, by manu-

ally digitizing existing maps, using complex patternrecognition to extract vectors from raster-format images,by line-following scanners, and by input of previouslycompiled vector files.

Figure 8.2 shows the essential differences betweenraster and vector formats. Raster data files, because theycover the whole of an area, each contain millions or tensof millions of cells. In this regard vector-format filesare much more compact, comprising up to tens of thou-sands of elements. Each model has its advantages and

208 Chapter 8

Fig. 8.2 Differences between raster-and vector-format data. A simple sceneof reality (a) is represented as a rasterby (b), where each element of the sceneis made up of square pixels, and invector format (c) by points, lines andpolygons labelled with an attribute.

100 200 300 400 500

easting

600

disadvantages. The raster model has a far simpler datastructure than the vector model. It allows much easiercombinations of different data than the vector model,represents spatial variability much more efficiently,and permits a great variety of digital manipulationsand enhancements. Vector-format data, as well as beingmore compact, encode topological relationships whereasrasters do not. Being very precise representations ofdefinite spatial features, vector files are better suitedto graphical representation than rasters, which presentboundaries with a: blocky structure as a result of theirbeing represented by arrays of cells. Both raster and vectordata are used in geographical information systems(Section 8.3), but raster-forinat renditions of nonimagedata have many uses in purely visual interpretation andcombination with the ir1formation in remotely sensed

images.

raster maps lies in their retention of the cartographer'sart and the ability to combine them with other kinds ofraster data in composite images (Section 8.2.6).

All other data involving quantifiable, natural attributes,which vary continuously over an area, initially requireconversion from records at points or along survey linesto the regular grid that is the hallmark of the rastermodel. This involves interpolation from the values atsurveyed positions to areas for which there are no data.

8.2 Non-image data in raster format

8.2.1 Interpolation and grid ding

Producing a raster rendition of nonimage data gatheredat more or less irregularly spaced positions involvesmathematically predicting from nearby measured values,and the overall variation, values that might be foundwithin regularly spaced cells covering the whole areaoccupied by the data. This is interpolation. Predictingvalues that lie outside the surveyed area is extrapolation,which is rarely used, except in predicting values that areat the boundaries of the area of cover.

Interpolation relies on expressing the known datain the form of a three-dimensional mathematical model.The simplest approach is to assume that the value at anypoint is the same as that at the nearest surveyed station.This divides the area into polygons containing surveylocations, the sides of which are equidistant betweenadjoining positions. The model consists of steps, andunless the original data were gathered in a rectangulargrid, the polygons are more akin to a vector representa-tion than a raster. This method is rarely useful in geo-logical applications, and the favoured models embodypredictions of gradual change. .

The simplest raster~formatdata are facsimiles of the red,green and blue components of a colour map, producedas raster files by a scanning device. Although in theorythree-colour maps can be classified into categoriescomprising lithological units; boundaries and structuralfeatures, in practice the cartographic use of hachuredtones, similar colours for different features and the pres-ence of symbols, such as dip arrows, tend toihwart suchefforts. A better representation of "the inforrnationcon-tent in the map is achieved by using the vector model,but that entails manually digitizing boundaries if themap is available only as a paper print. The appeal of

Non-Image Data and Geographical Information 209

z

(b) ./"

./

~ Fig. 8.4 Deviation of a best-fit line from the actual data points.

./

V

~~

The simplest of these predictive models is that invqlvedin regression analysis, which fits surfaces to a data dis-tribution to minimize the sums of squares of the distancesbetween real points and the surface. Figure 8.3(a) showsa best-fit line that conta~s the variation of a property zwith distance x. It is a cross-section of a simple curvedsurface in three dimensions. The distance in question isthat between the points and the line in the z direction-there

is no need to illterpolate the distance, which im-plicitly is correct. The method uses polynomial surfacesof any order (Fig. 8.3b). This method, also known astrend-surface

analysis, predicts the long-range variationof the third, nonspatial dimension and effectively smoothslocal

variations, rarely passing exactly through the origjnaldata points (Fig. 8.4). Effectively, trend-surface analysis

treats the 'raw' data as though they contain errors;Modem geophysical and geochemical techniques areremarkably precise, and the variability in data gener-ally is real. So, a raster produced in this way degradesthe information content. However, the deviations, orresiduals, from trend surface models of different orderscontain anomalous deviations from the general patternsrepresented by the surface itself (Section 8.2.5). A si;mi1,ar

surface-fitting interpolatiol1 involves Fourier analys~(Section 5.3), but this too omits deyiations from sl}looth

surfac,es.Before digital fitting of curves to data, fitting was

achieved using flexible rulers, technically known :assplines. The important feature of a spline is that it can fitcurves with a form that varies from place to place. Trendsurfaces and Fourier series accommodate the whole inone complex function, and the points usually d~viatefrom the computed prediction. A spline fit can be per-fect, and the joins between one part of a curve and otherdifferent parts are continuous. This does not mean that aspline predicts exact values of the third dimension atunknown points, but adding extra points merely allowsthe spline to become a more precise reflection of reality.However, a spline makes no allowance for errors amongthe data. A line drawn using a spline is approximatelymade up of segments defined by different cubic polyno-mials, and has continuous gradients and gradients ofgradient (first and second derivatives). Consequently,splines lend thez:nselves to mathematical expressions,and can be applied to two- and three-dimensional datasets using a computer. For lines and surfaces these areknown as cubic-spline and bicubic-spline functions,respectively. Mathematical splines can fit data exactly,or, if errors are suspected, can incorporate a degree ofsmoothing that allows for imprecision. Figure 8.5 s~owsa spline fit to ..several points, and the modification thatm~st occur if one point is revised subsequently, empha-s~ing thelil}lit.ations of this approach.

One of the problems with using bicubic splines in grid-ding, ~~ implied by Fig. 8.5, is that they can introduce

210 Chapter 8

\,

x

Fig. 8.5 Cubic-spline function fitted exactly to data points, thethinner line showing how the fit changes with the modificationof one point (c to c').

artificial anomalies because of the way that they work.One means of avoiding this is to assign to an unknowncell the average value from actual points in its localneighbourhood. This is a moving-average approach, inwhich the size of the neighbourhood or 'window' can bevaried. Another adjustment incorporates weighting thecontribution of each point according to its distance fromthe unknown cell. However, this leaves several dif-ficulties: What is the optimum size for the 'window'?What shape and orientation should it have? Is distancethe best control over weighting? What errors are associ-atedwith the interpolated values? A mining engineer,D.G. Krige, addressed these problems in attempting to-optimize

predictions of spatial factors in mine planning.He recognized that many properties vary too irregularlyto be modelled by smooth functions based solely onthree-dimensional geometry. They require knowledge ofsome of the statistics involved. The mathematics involvedis complex, but kriging, as this approach is known,allows great control on the precision and accuracy ofgridding, as is needed in mapping variations of metalgrade in ore bodies and in the production of digitalelevation data (Section 8.2.4). However, methods basedon splines and weighted moving averages tend to bethe most frequently used means of producing rasters.

with a fairly even spaciag, problems are miDimal withgravity data and similar information measured at dis-crete points. Many surveys are conducted along surveylines, however, either continuously, as with aeromagneticsurveys, or at regular along-line sampling intervals, as inairborne gamma-ray spectrometry and electrical ground

surveys.The main problem with line surveys is that the sam-

pling interval along lines generally is spaced far moreclosely than the separation between lines-there is adistinct linear bias to the data. Moreover, in terms ofthe variability encountered, some segments ~ontain fargreater deviations from regional trends than others.Using all the data encounters two problems. First, thequality of the interpolation decreases away from thelines, and that imparts artificial, geometric featuresparallel to the lines. Second, the same weight is given to'smooth' areas as to complex areas, thereby involvingmuch redundant information and tending to extendsmoothness into areas that should be complex. Somemeans of reducing redundancy and avoiding bias hasto be sought.

There are two strategies to follow. The line spacinglimits the degree of spatial accuracy in areas between thelines, and so the operator selects an appropriate grid-cellsize. In general, a size between a quarter and an eighthof the line spacing avoids artefacts parallel to the lines.Another option is to adopt a sampling strategy for theraw data along the lines themselves. If the spacing ismuch smaller than the eventual cell dimensions thendirectional bias remains. If sampling is infrequent, thensome of the information is lost. The usual solution isa compromise. Wide sampling in smooth areas loseslittle information, whereas sampling in complex areasconcentrates on local minima and maxima, and, moreimportant, on defining the shape of the variation wherethere are changes in gradient. Doing this automaticallyinvolves incorporating measures of local complexity,such as variance and the second derivative, plus ident-ification of maxima and minIma and an upper limitimposed on sampling density. Inevitably, some informa-tion is always lost, to be replaced by interpolated values,but this is offset by the advantages of rendering the datain raster form.

Much geophysical information dates from beforethe 'digital revolution'. Raw data as analogue profilesalong survey lines needs conversion to digital form, eithermanually or using line-following software. Contourmaps may w~ll have been produced by manual inter-polation. As well as the visual disadvantages of contourmaps discussed earlier, their very nature in expressingcontinuous variations in discrete steps degrades theoriginal data. In such cases the only option is to extractinformation by digitizing contoured data. Figure 8.6shows three methods that differ in sampling strategy.

8.2.2 Geophysical data

Today, virtually every geophysical survey records datain digital form, with. extremely precise geographicalpositioning by satellite navigation systems. The raw datatherefore are available immediate:ly-given payment ofthe appropriate fee-for interpolation using the methodsoutlined in the last section. Gridding may already havebeen done and the results stored, because this is essentialin the computer production of conventional contour mapsfrom survey data. There are, however, a few remainingproblems. As point measurements generally are produced

Non-Image Data and Geographical Information 211

contours are position~ accurately, and samples alongthem (Fig. 8.6b). This maintains, and even extends thedegradation implicit in contouring. However, the lastmethod can be partly automated by applying line-following software to scanned maps, and the variationsin contour spacing assists the gridding software in fittinga better surface to the data than is possible using the othermethods. None have particular advantages, but becauseproducing raster maps generally involves input frommany contour maps, the same method is essential so thatthe end product has a consistent mathematical basis.

(a) .~ ~' I."Ii " ,I

,r; I

'1

;~Ie;

'---8

,.

~,'

~

""~

~'.,

.! I ~(c)

8.2.3 Geochemical data

Measurements of the concentration of an element ina sample of rock, soil or vegetation relate only to the geo-graphical location of the sampling. Gridding treats themin exactly the same way as geophysical data from pointsurveys. However, such surveys are slow and expensiveto conduct and rarely figure in regional evaluation.A much quicker, cheaper and convenient approachexploits the natural sampling of soils and rocks by ero-sion, transport and deposition by flowing water. Most ofthe continental surface has drainage channels that areeasy to locate on maps or aerial photographs, and easy tosample. Results of drainage geochemical surveys, usinganalyses of either stream sediments or stream water,are by far the most widespread kind of geochemicalexploration. However, they present a peculiar problemin conversion to raster format.

The concentration of a chemical element in streamsediment or water does not represent the composition ofthe soil or rock in the immediate vicinity of the samplepoint. Instead, it represents the entire area supplyingwater and sediment moving past the sample point.That is, it represents the catchment area upstream ofthe sample point. Figure 8.7 illustrates the difficulty. Thearea represented by each sample increases the furtherdown the drainage the sample is collected. This relatesto the order of the stream-the more tributaries enter astream, the higher is its order and the larger its catch-ment area. Consequently, in sampling all streams, irre-spective of stream order, each sample not only has itsown unique concentration but a unique significance interms of the area that it represents. It is impossible tointerpolate such results to a regular raster that has anyuseful meaning. In fact it would be grossly misleading todo this. The only valid representation would be assign-ing the measured values to the associated catchments.If there are several sample points along a drainagethis clearly leads to great confusion. Furthermore, as theorder of a stream increases, so the chances of anomalousvalues decrease owing to the diluting effect of water and~diment being contributed from a wider area. Muchmetal exploration involves drainage geochemical surveys

./~

.,...

~/,.

.,

:;

.~:~

,~

~.e'/'.

,.w'

'",-;;//.~

//:/.---,~'--

8,

~,.-~c

r,,,8

Fig. 8.6 Three methods of digitizing data from contour maps.

The first systematically estimates a value for the variableat points in a regular grid (Fig. 8.6a). Using the reasoningthat more points are needed to define areas of increas-ing complexity means sampling with variable spacing(Fig. 8.6c). Both methods 'second guess' the methodo-logical approach used in compiling the original map,inevitably assuming linear variation between contoursin manual digitizing. The third method assumes that the

212 Chapter 8

(a)

v.QJv.

.?:-cocco

"0...QJ.aE:JZ

(b) 100

t99%

~

:~~ "',j, 97%,j, 95%

90%85%

90

?c

~Q,

.8-~CW:J0-W

.:::w>

~10"5E:JU

80

70

60

50

40

3

20 -

10 I~

v1.0 1.5 2.0 2.5 3.0

10 ppm 100 ppm 1000 ppm

log [CuIFig. 8.8 (a) Histogram of copper conc~ntrations in ageochemical survey with a smooth curve fitted to thevariability. (b) Cumulative frequency distribution for thecopper data ill (a) showing different percentile levels.

infrequent occurrences may be significant anomalies.Background values generally comprise a more or lessdistinct, broad peak in the histogram with a clear max-imum and minimum. Anomalies lie outside this range,and exploration geochemists usually consider onlyabnormally high values, for obvious reasons (Fig. 8.8a).The maximum expected value within the backgroundrange is sometimes termed the threshold, and it is pos-sible to separate anomalies from the normal by settingall cells with values below the threshold to an arbitrarylevel and using contrast enhancement to highlightanomalies. More sophisticated statistics help refine thisscreening. On~ is to use the cumulative histogram anddensity slice at a variety of percentile levels (Fig. 8.8b).As many elements show a log-normal distribution ofvalue~many occurrences over a restricted range (suchas 10-50 parts per million) and a few occurrences over avery wide range (such as 100-1000 parts per million)-the histogram is often used with a logarithmic axis.

of this style, seeking anomalies low in a drainage andthen tracing them up tributaries to their source. Thatlargely rules out the data being useful in raster form.However, there is a strategy to overcome the problem.

The lower a stream's order, the smaller is its catchmenta:nd the mQre likely are the concentrations of chemicalelements in stream samples to represent nearby rocksand soils: Provided sampling involves only streams ofthe same order, preferably the first. or second order, andinterpolation uses grid cells that have a similar size tothat of the associated catchments, the resulting raster isrealistic. Because of the wealth of possibilities for displayand enhancement of geochemical data, and their combi-natiqn with other kinds of data, either in raster images(see Plate 8.3) or GIS, this form of sampling is becomingthe norm for drainage g~ochemical surveys.

Multi-element geochemistry is a powerful meansof discriminating rock types and gaining insights intopetrogenetic relationships, because of the different.behavi-ours of various suites of elements during chemicalfractionation by different geological processes. So, oneuse of raster-format geochemical data is in basic geolog-ical mapping, when the full range of variability is used.Geochemical exploration, primarily for metals but some-times for hydrocarbons, relies on locating areas whereconcentrations of some elements rise well above thegeneral levels or backgrounds associated with commonrock types to form anomalies. A problem lies in decidingwhich values constitute anomalies within the full rangeof variation. A variety of solutions centre on statisticalanalysis of the data. The most frequently occurringranges of values constitute the background, whereas

Non-Image Data and Geographical Information 213

Another approach is to use some arbitrary rule that, forexample, expresses cells with values below the threshold(t), between t and t plus the average value (a), t + a tot + 2a, t + 2a to t + 3a, and greater than t + 3a with differ-ent intensities or colours, each constituting anomalouscells of increasing potential significance. Because of thewide variation in chemical composition of common rocks,particularly for some trace elements, a concentration thatis anomalous for one rock type may be within normalbounds for another. So, a further complicating factor inusing geochemical data is screening them according tothe known distribution of different lithological units.

spective views of unknown terrain, introduction of reliefdisplacement into remotely sensed images to give stere-optic viewing potential and the correction of remotelysensed data for variable solar illumination.

Digital elevation models (DEM) take a variety offorms. The simplest are raster arrays of elevation values,but these use the same density of cells irrespective of thecomplexity of the topographic surface, and so occupylarge storage volumes. A number of methods use vari-able densities of data points to reduce storage space,but all must be gridded if the data are needed in rasterformat. The most commonly used of these efficient stor-age formats is the triangular irregular network (TIN). ATIN comprises a number of triangular facets (technic-ally known as Delaunay triangles) linking points withknown elevation, the greater the relief, the larger thenumber of points and facets. Usually, producing a TINinvolves converting a DEM to the low density format(see Fig. 8.9). Capturing elev~tion data from existingcontour maps and interpolating them can us~ any of themethods described in Section 8.2.1 and 8.2.2. For som~areas DEMs in various forms are available from statemapping agencies. The US Department of Defense'sNational Intelligence Mapping Agency (NIMA) main-tains a global archive of DEMs derived from surveillancesatellites and aircraft for most of the planet at very highresolution. This is known as Digital Terrain ElevationData (DTED) Level-lC. At the time of writing DTEDLevel-lC for areas outside the USA was accessible onlyto US Government Agencies"or those of allied powers,but this may change if pressure from potential users isapplied strenuously and continuously. A Shuttle Missionin February 2000 deployed an interferometric radar systemto measure topographic elevatibn with a geographical

8.2.4 Elevation data

The Earth's surface is the most familiar example ofcontinuous variation in three spatial dimensions, andis almost universally represented by contoured maps oftopographic elevation. As well as containing accuraterepresentations of topography, in which there may beclues to geological features that exert a control over land-forms, topographic data have immense potential for theextraction of other information with a variety of uses.Topography controls the rates and directions in whichsurface water flows, and it involves drainage networksthat can be established without hydrological survey.Where there are known geological boundaries, whichperhaps appear on remotely sensed images, their trendsin relation to topography define their dip and strike, andeven structure contours from the simple three-pointsolution. Topographic slope estimates give measuresof the likelihood of soil erosion or risk from landslip.Incorporating topographic data with other kinds opensup new ways of visualizing relationships, such as per-

Fig. 8.9 (a) Digital elevation modelshown as a grey-tone image.(b) Triangu)arirregu)ar network(TIN) representing the same surface.Note how the density and size oftriangles in the nN reflect varyingdegrees of complexity in the digitalelevation model (DEM). (a) (b)

214 Chapter 8

resolution of 25 ro, and a vertical r~solution of 4m. Earlyresults from the Shuttle Radar Topography Mission(SRTM) show that the quality of the OEMs that it willproduce surpass those previously restricted by NlMA..

Most topographic maps use elevation data derivedfrom stereoscopic measurements of aerial photographs,using a variety of manual and semiautomatic photo-grammetric instruments, backed up by precise geodeticsurveying. The methods rely on measurement of the par-allax differences inherent in the relief displacement instereoscopic image pairs (Appendix A). Satel1i~e systems,such as SPOT and JERS-1, produce digital stereopairsand it is possible to digitize photographic images accu-rately. This makes it possible to derive OEMs automatic-ally, through the use of complex software. The parallaxestimation that is essential in such automated cartographyrelies on the computer being able to recognize the samefeatures on each image of the stereopair. An operatoridentifies a number of points on the stereopair that act as'seeds' for pattern recognition, which builds up manymore such ties points automatically. The parallax dif-ferences associated with these points form the basis forcalculating elevation relative to accurately surveyedground-control points. As elevation precision dependson that associated with parallax, the spatial resolution ofthe imagery is crucial, because parallax can be measuredonly in multiples of pixels. For1D--m resolution SPOTimages, elevation precision is around 20-30 m-too poorfor all but. qualitatiye use of the resulting OEMs. Aerialphotographs have resolutions in the 0.1-1 m range, andso provide OEMs that are suited to detailed cartography.

~n the absence of readily available, precise OEMs, thebest means of producing them for detailed, quantitativeapplications is by painstaking digitization of contourson large-scale (>: :25000) maps and then interpolation(Section 8.2.1). This should ensure vertical precision ofthe order of a few metres.

Because the travel time of radar waves scattered fromthe surface can be measured very precisely, it mightseem reasonable to expect that radar could be used formeasurements of surface elevation to within a few cen-timetres. Seasat and ERS-'1 deployec,isuch radar altime-ters, but results are available only for the ocean surface.The problem over land lies in both the difficulty infocusing the radar beam and in the response fromrugged surfaces. From orbital altitudes the radar beamspreads to several kilometres, thereby limiting the resolu-tion in the horizontal plane. Within this 'footprint' thehighest features reflect the radar wave first and the low-est give a' later return of the radar pulse. Consequently,all that can be measured in the vertical dimension is theaverage elevation and the range of elevations. For such awide 'footprint' that information is of little use over allbut the most gentle terrain. Radar altimetry thereforeis of little use over land, waves at the ocean surface are

r;irely higher than 1~20 m, however, and the averageocean':surface elevation can be measured to within a fewcentimetres within the wide radar 'footprint'. Waveheight obviously is an interesting measurement foroceanographers and meteorologists, but the usefulnessof ocean-surface elevation to geologists is not immedi-atelyobvious.

Sea level is closely controlled by gravitational factors,hence the diurnal and monthly patterns of tides. Butthe Earth's gravitational field also p.lays a role, as well asthose of the Moon and Sun. Averaged over several monthsthe ocean-surface elevation mainly relates to variationsin terrestrial gravity. Over the oceans the gravitationalfield, and therefore the surface elevation is dominatedby variations in ocean depth. Because water is about aone-third as dense as the rocks of the oceanic litho-sphere, the shallower the water the higher the gravita-tional potential and vice versa. This means that waterflows to shallow areas to form a bulge, and converselythe water surface is depressed over deep ocean basins. Afeature on the ocean floor with a relief of 1 kIn results in a2-m disturbance of the surface. With a vertical precisionof 5 cm, radar altimetry therefore can represent ocean-floor features as small as 25 m, provided that they extendover several kilometres. So, asa result of this convolutedrei(cltionship, images of radar altimetry data over theoceans form bathymetric maps of such precision thatsubtle features of the ocean floor are readily apparent(Fig. 8.10). Because conventional bathymetric maps arecompiled from soundings",taken by ships, their qualityvaries according to the density of marine traffic. Over mostof the oceans bathymetry derived from radar altimetry isfar more informative. It is possible to use conventionalbathmetry to remove the effect of varying water d~pth toanalyse changes in gravitational potential resulting fromvariations in the density of the underlying crust andmantle. This can be used to map features possibly relatedto thermal convection in the mantle, and to locate deepsedimentary basins in areas of continental shelf.

Although radar altimetry appears unlikely to generateuseful DEMs over land, because of beam-focusing prob-lems, laser altimeters potentially can produce very highresolution in all three dimensions over land. Indeed,laser methods are used from orbit for precise site posi-tioning in studies of plate dynamics. There is one majordifficulty of a nonscientific nature associated with veryprecise DEMs. They are essential for the guidance sys-tems of cruise missiles, as well as other military applica-tions, and therefore are shrouded in secrecy. As cruisemissiles have been around for well over a decade, so toohave DEMs with the requisite precision.

With SRTM interferometric radar data (Section 7.3), allthis set to change rapidly, provided the results are notplaced under a security blanket when their quality andinformation content become clear,

Non-Image Data and Geographical Information 215

Fig. 8.1.0 Image of sea-surfaceelevation derived from Seasat radaraltimetry data, showing details ofocean-floor topography as a result ofthe effect of gravitational potentialon the sea surface.

(a)

Fig. 8.11 (a) Contour map of magnetic total field anomaly over an area in northern England. (b) Image of the data shown in (a).The area is 100 kIn square.

Although a great improvement over contoured data,the image in Fig. 8.11(b) suffers from two main prob-lems. Human vision distin~ishes between only 16 and32 grey levels, so the 256 levels are not distin~hable.The image also lacks clues to 'depth' and therefore appearsbland. These factors mask any high spatial-frequencyfeatures. Contrast stretching (Section 5.2) gives someimprovement, but the full range of values for the mag-netic anomaly only emerge graphically if the data aredisplayed in colour. This is most conveniently achievedby assigning the ON to a colour palette to produce a'rainbow: range of colours, where low values are in bluesand greens, intermediate values appear as greens andyellows, and high values show as orange to red and

8.2.5 Raster processing methods

Raster data of the kind discussed so far in this sectioncontain a wealth of informatioh from which geologistscan make interpretations and combine them with otherkinds of data for analysis in a GIS {Section 8.3), AlthoughGIS methods provide awesome analytical power, formost applications visual interpretations are extremelyeffective. Figure 8.11 shows the improvement. in visualquality that stems from converting contoured data to animage. In this case the data are aerial measUrements ofanomalies in the total magnetic field potential, convertedfor display from negative and positive values of nano-tesla to an 8-bit (0-255) range of po~itive integers.

216 Chapter8

magenta. The resulting pseudocolour image of the datain Fig. 8.1 is shown in Plate 8.1. Most magnetic andgravity data show deviations from the average potentialfield, and the anomalies may be positive or negative.Careful colour assignment can ensure that the zeroanomaly coincides with the green to yellow transition inthe rainbow spectrum, perceptually the most distinctboundary to us, so that 'cold' colours represent negativeanomalies and 'warm' colours associate with positive.

Although pseudocolour renditions express more fullythe data's information content, colour masks high-frequency spatial features in the data (Section 2.2). A lackof clues to depth also hides spatial detail. The mostimportant clues to depth in our visualizing topographyare shadows, as seen in all aerial and satellite imagery.Simulating solar shading of nonimage raster data usesa variety of algorithms, the simplest of which are thedirectional filters discussed in Section 5.3, which mimictl:le effect of a constant Sun elevation from the eightprincipal points of the compass. More sophisticatedalgorithms allow various illumination angles and typesof lighting to be selected. Figure 8.12 shows the dramaticrealism that results from such a transform of a OEM.You should see valleys and ridges realistically, but turnthe page through 180 degrees and the illusion 'flips' to aninverse of topography (Section 2.5).

Raster renditions of ~y spatial variable are in effectmaps of the 'topography' of the variable, and can beenhanced just as easily by simulated shading to tune animage to human vision. Figure 8.13 shows shaded imagesof magnetic 'relief' incorporating illumination from twodirections at right angles. They show how high-frequencyspatial features with different orientations can be high-lighted separately. By combining oblique 'illumination'with pseudocolour, using multiplicative or ISH methods(Section 5.2.2), the advantages of both result in graphicimages, ideally suited to interpretation (Plate 8.2).

(b)Fig. 8.12 Digital.elevation data (GTPO30) of East Africafiltered to simula~e the shadowing by solar illumination fromthe north-west.

Fig. 8.13 Aeromagnetic data shown in Fig. 8.11(b)'illuminated'from the north (a) and west (b).

Non-Image Data and GeagTClphical Infornration 217

(a)

.(b)

Fig. 8.14 (a) Image of gravitational potential in an area ofAlaska, compared with (b) an image of the first derivative ofthe gravitational potential. Courtesy of Charles Trautwein, USGeological Survey, EROS Data Center.

As well as enhancement of gridded nonimage data, avariety of mathematical functions can extract informa-tion from them. These methods include differentiation toproduce various derivatives and trend-surface analysis(Section 8.2.1). The magnitude of the first derivative of aspatial variable expresses the gradient of the topologicalsurface. It is particularly useful in analysing OEMs bygiving regional estimates of topographic slope. This canbe useful in assessing slope stability and rates of runoff,as well as in geomorphological analysis. Some geolog-ical applications benefit from derivatives of geophysicaldata. Figure 8.14 shows an image of gravity data togetherwith one expressing the slope of the gravitational field.The latter highlights sudden changes in gravitationalpotential, which may correspond with subsurface dens-ity boundaries.

The magnitude of the second derivative expresses therate of change of gradient of a topological surface, andis thereby a measure of the variability or 'roughness' ofa surface. Inflections in a surface by definition have asecond derivative of zero, and so are defined uniquely.In force-field data the main element in contributing tohigh-frequency variations is the crystalline basement,in which magnetic susceptibility and density vary con-siderably in contrast to their relative uniformity in theoverlying veneer of sediments. The more deeply buriedbasement features are the smoother their expressionin gravity and magnetic data. In magnetic data, therelationship between anomaly patterns and the bodiesresponsible for them is often complex owing to bipolar-ity. Many magnetic anomalies resulting from a singlebody have both negative and positive features, and thesource for the anomaly frequently resides beneath theline of its inflection. Consequently, second-derivativeimages of gravity and magnetic data express depth tobasement qualitatively, and locate the sources of magneticanomalies. Quantitative use of force-field data involvescomplex three-dimensional modelling, in which geolog-ically reasonable bodies of different density and magneticsusceptibility are manipulated to simulate the observedgravitational or magnetic field anomaly patterns. Theyare beyond the scope of this book.

Another approach to analysing deeply buried features,especially from gravity data, is trend surface analysis(Section 8.2.1), which fits polynomial surfaces to the ori-ginal point data. A first-order surface expresses the grosstrends as a dipping plane, whereas higher-order surfacesassume more complex shapes to express regional patterns.The higher the order of trend-surface, the more closely itmatches the actual variation in the data. However, evenfor 10th-order surfaces, the patterns are grossly smoothedcompared with surfaces interpolated by spline or krig-ing functions, and represent the low-frequency spatialvariations contained within the data. Low-frequencyvariations in gravity data stem from broad variations in

the deep crust and upper mantle and from the relativelyuniform but variably thick veneer of surface sediment'-ary cover. The higher frequencies stem from variationswithin near-surface basement, or perturbations in thedepth to the cover-basement interface, often related todeep faulting. Subtracting the trend surface or regionalfield from the interpolated field topology results in aresidual field, in which may reside important and previ-ously indiscernible information about cover-basementrelationships. Figure 8.15 shows the results of differentorders of trend-surface and the associated residual fields

218 Chapter 8

(a) (b) (c)

Fig. 8.15 Images of Bouguer gravityanomaly data (a) and the 4th (b) and10th order (c) trend surfaces fittedto an area in northern England.Subtracting these regional trends fromthe 'raw' gridded data results in imagesof the residual fields for the 4th and 10thorder surfaces (d and e), in which can beseen hidden features that relate to deep-crustal structure, after 'illumination'from the west.(d) (e)

from Bouguer gravity anomaly data for the area shownin Figs 8.11 and 8.13, and Plates 8.1and 8.2(a).

8.2.6 Multivariate raster images

Enhanced images of single rasters are well suited tointerpretation, using all the skills and intuition embod-ied in a trained geologist. Using them singly, however,when there are many different kinds of data, produces awelter of interpretations that need to be co-ordinated insome way. One approach is to transfer them to a GIS,where a variety of computer applications can assist inthis task (Section 8.3), but this relies on mathematicalmeans of seeking relationships. Human vision is by far themost flexible means of assembling connections betweendisparate sets of information, and it makes sense toexhaust the full range of possibilities presented byimages before resorting to computer analysis.

Several methods combine multiple data sets as com-posite images, most relying on techniques covered inChapter 5. The simplest of these uses three sets, i.e. thered,. green and blue components of colour images.Seeking relationships between the sets relies on Young'stheory of additive colour (Section 2.4). Where all three

variables are strongly correlated, the DN used in colourdisplay are all similar, so that the result is a shade ofgrey, ranging from white to dark grey depending onwhether all the values are high or low. If two of thevariables are well correlated and the third is unrelatedor anticorrelated, this is signified by the area being ren-dered in shades of yellow, magenta or cyan, dependingon the colour assignment of the variables and their DN.Areas showing the three additive primary colours sug-gest that there is no significant correlation between anyof the variables, and one is high whereas the other twohave low values. Pastel and nonspectral colours resultfrom partial correlation among the variables.

As RGB images assign the same weight to all threevariables displayed, low information content in onevariable tends to swamp that in the others, particularly ifthere are gross disparities in the spatial resolution of thedata sets. Therefore, a simple rule is to combine data setswith approximately the same spatial resolution. Nor is itwise to combine data that have fundamentally differentrelationships to geological features. Consider an imagemade up from a Landsat band; gravity and magnetic data.By no stretch of the imagination can any meaningfulcorrelation in an RGB image be expected, because the

Non-Image Data and Geographical Information 219

properties being displayed relate to spectral reflectance,density and magnetic susceptibility i which have no rela-tionship. Although all three may contain vital informa-tion on the distribution, say, of major faults, the resultingimage inevitably will look like a pizza! A second ruletherefore is to combine data in RGB form where thereis some expectation of real correlation between thevariables. For most geological applications, the unifyingfactor is likely to be different relationships to rock type.

A good example of variables to expressiI} RGB orderfor lithological discrimination, if they are available, wouldbe nickel, chromium and gravity anomaly data. Asultramafic rocks are rich in Ni and Cr and have high den-sity, then areas underlain by peridotites would show aslight grey tones on the image. Basalts generally havelower Ni/Cr ratios than ultramafic rocks and slightlylower densities. They might be revealed by greenish or

,,'.cyan areas. Granites and sediments havelow values forall three variables, and consequently would show asdark patches. Ironstones, with high density and low Crand Nii would be distinguished by blue signatures..Occurrences of nickel sulphide ores, being depleted inchromium but associated with dense ultramafic rocks,should be revealed as magenta or red areas.

Because much is now known about the affinities of dif-ferent chemical elements for different lithologies, andtheir behaviour relative to one another in geological frac-tionation processes, geochemical data are ideally suitedfor simple RGB combination as a route to lithologicaldiscrimination. In particular the common association ofrestricted ranges of elements during different kinds ofmineralization episode makes this approach extremelypowerful in metal exploration. Some granites, with whicha variety of pegmatite, skarn and hydrothermal depositsare frequently associated, consistently carry tin-tungsten-lithium-fluorine. In exploration for gold mineralizationthe 'pathfinder' elements arsenic-bismuth-antimony aresought routinely in drainage and soil surveys. However,high-resolution data from geochemical surveys are notoften available outside of mining companies. Britain isparticularly well endowed with public-domain data ofthis kind, and Plate 8.3 shows-some highly informativeresults from them.

More widely available, particularly in arid and semiaridterrains, are the results of airborne gamma-radiationspectrometry (Section 3.7), which are often flown togetherwith magnetic surveys. Gamma-radiation data relateto the abundances of radioactive isotopes of uranium,thorium and potassium in soils and rock. Fortuitol,lsly,these three elements behave very differently in a greatvariety of geological processes, are found in a large rangeof concentrations and assume quite different relativeabundances in a wide variety of common rock types.Although often noisy because of the blocking of radia-tion by vegetation and air, RGB images comprising U,Th

and K estimates from.spectrometry are very colourful,the colours relati~g well to the surface distribution ofdifferent rocks and soils, as Plate 8.4 shows. Interpretingsuch images relies on both an understanding of additivecolour theory and knowledge of the common relativeabundances of the three elements in rocks. Granites usu-ally contain all three elements in abundance, whereasbasic and ultrabasic igneous rocks are strongly depleted.Sandstones are often depleted in both U and K, beingdominated by quartz grains, but can have high Th con-tents locked in stable heavy minerals such as zircon andmonazite. Clay-rich rocks generally contain high K con-tents bonded withfuclay minerals, and if deposited underreducing conditions generally have elevated U contentsowing to precipitation and adsorbtion of the insoluble U6i-ion. They contain little Th owing to its retention in inertheavy minerals that rarely reach the low-energy environ-ment associated with the deposition of clays. Limestonesusually are depleted in all three, but if rich in organicremains appreciableU may have been precipitated fromsea water or hydrothermal solution& because of thereducing conditions that accompany hydrocarbons. '

Provided that the result~ can be understood easily, it ispossible to combine information from upto six variablesin an RGB, image ~ using'rqtios. This is most suited togeochemical data,. where ratioing is commonly used inexpressing 'petrogenetic differences:

Despite the earlier warning about unwise combina-tion of data in RGB form, meaningful fusionbetweenvariables with fundamen~lly unrelated structures butimportant geological content is feasible. The appropriatemethod is the intensity-saturation-hue aSH) transform(Section 5.2.2). Although this was devised for the enhance-ment of RGB images by manipulation of the intensityand saturation derived from them, it is possible to sub-stitute data of different kinds for the intensity and hue inthe inverse aSH to RGB) transform. The simplest appli-cation is using a single data set, such as aeromagneticraster data. The raw data are used as the hue and a syntheticsolar-shaded version incorporating high-frequency spa-tial features is used as the intensity, with saturation set toa constant (in the range 64-255) to ensure vivid coloursin the result. This is another means of producing imagessimilar to Plate 8.2.

Combining two data sets using the inverse ISH .trans-form employs the strategy of using the data with thegreatest spatial information content, often some kind ofremotely sensed image, as the intensity input. This ensuresthat the spatial information is best rendered for visual-ization as brightness variations in the 01,ltput colourimage. Hue is controlled by the data with the greatestcontribution to understanding lithologica~ variation, suchas gravity or geochemical data, so that the resultingcolours are meaningful. Saturation is again set to a con-stant. Plate 8.5 gives an example of this approach.

220 Chapter 8

Fig.. 8..16 The parallax in thisstereopair of Landsat MSS imagesof Alaska was introduced using adigital elevation model. Courtesy ofCharles Trautwein, US GeologicalSurvey, EROS Data Center.

N

t10 km

Fig. 8.17 Th~ parallax in thisstereopair of Landsat MSS imagesof Alaska was introduced using thegravity data shown in Fig. 8.14(a),and although the stereomodellooksodd, it helps resolve evidence formboth the surface and subsurface.Courtesy of Charles Trautwein, USGeological Survey, EROS DataCenter.

N

t10 km

Because an RGB image comprising three variablescan be converted to intensity, saturation and hue valuesusing the ISH transform, the colour information, expressedby the saturation and PUE:, can be combined with anotherdata set. that is substituted for the intensity. Again, thebest strategy is using data with potential for lithologicaldiscrimination as a source of colour and allowing inten-sity to be controlled by a variable containing high-frequency spatial information. Plate 8.6 incorporatesside-illuminated aeromagnetic data as intensity and U,Th and K estimates from airborne radiometric dataas saturation and hue for the same area as Plate 8.4. Bycareful use of ratios,..this method allows the combinationof up to seven variables, but at the expense of increasingthe difficulty of interpretation. Usually such extrememethods use variables that are selected on the basis ofa well understood model relating to the task in hand.For instance, epithermal gold mineralization is oftenassociated with faults, generally results in clay-mineralalteration of host rocks, has elevated concentrations of

As-Bi-Sb and the hydrothermal fluids often deplete thehosts in a number of alkali metals. From that model asophisticated use of the ISH method could be based onhigh-resolution imagery for intensity and a mixture ofgeochemical data and spectral ratios as the source of hueand saturation controls, but it may be less confusing touse GIS methods.

As parallax derived from stereopairs of images is aroute to modelling variations in topographic elevation,so the inverse is possible. In the absence of stereoscopicimages, which greatly improve the detection of geo-logical structures and lithological boundaries, syntheticstereopairs can be generated from satellite images byadding parallax derived from a OEM. Figure 8.16 usesLandsat data and a OEM of part of Alaska. This approachis not limited to the use of elevation data, because thesame procedure can convert any three-dimensionaltopology into parallax shifts in synthetic stereopairs.Figure 8.17 shows an example using gravity anomalydata to introduce relief into the Landsat image used in

Non-Image Data and Geographical Information 221

Fig. 8.16. Neither is this method limited to a pair of datasets, but synthetic stereopairs can be produced for colourimages that already may incorporate information fromup to seven variables, as discussed above. Bizarre andpotentially confusing as such a monumental fusion ofdata might seem, given careful blending of data to a par-ticular geological model can produce images that fullyexploit the geologist's powers of observation and mentalsynthesis (Chapter 9).

8.3 Data analysis in geographicalinformation systems

nitude more storage apd computing capacity is requiredthan with a vector-based system, but the downward spiralof cost per unit of storage, processor speed and number-crunching power largely obviates these problems.

Much of the output from a GIS is in the form ofmaps that result from data fusion and analysis, and it isoften necessary to display or print these to acceptablecartographic standards. For this, vector format has thedistinct edge of sharp, accurate boundaries and ease ofgraphical modification. So, the most appropriate GIS forgeology incorporate means of conversion from raster tovector format, at least for the simple boundaries and thecategories within them that result from analysis.

The functions availablE! in a GIS are numerous andoften complex. Many were devised with the planneror manager in mind, not the geologist. The followingsections summarize some of the most relevant functions,and more detailed accounts can be found in a number ofspecialist te~ts. Functions relevant to geological applica-tions in a GIS break down to: retrieval and measure-ment (Section 8.3.1); overlay operations (Section 8.3.2);neighbourhood operations (Section 8.3.3); operationsinvolving DEMs (Section 8.3.4); operations involvingfaults and fractures (Section 8.3.5).

Like different wavebands in multispectral remote-sensingdata, the data layers residing in a GIS are registered, butwith the additional advantage that the registration baseis a geodetic co-ordinate system of some kind. A GISis geocoded. Both raster and vector data, together withlayers of attributes for each vector category, are likely tobe present and used in a geologically orientated GIS.Unlike remotely sensed data, the contents of each filemay not be related directly to the others. As will beshown, the purpose of a GIS is to derive new, topic- ortask-orientated attributes and layers from interrogationof the original data. Consequently, large numbers of newlyderived ffies are likely to be generated, many of whichmay be intermediate steps to some ultimate goal.

The most important general cautionary note is thatusing a GIS demands a far more ordered approach thandoes digital image processing. Its analytical potentialshould not simply be explored in the quest for visuallystimulating images. It requires a carefully thought-outstrategy that exploits information in the context ofwell-defined goals, and is constrained by the analyticalfunctions that reside in the system itself. Without thisdiscipline, the user undoubtedly will experience acuteconfusion and obtain results that have no significant

meaning.Geographic information systems use several different

architectures. These involve two main types; those hav-ing a dominantly vector-based approach and those thatare orientated to raster format. The first type mainly sup-ports the handling of existing map data in digital form,and is more or less a kind of digital mapping system ofinterest to planners and cartographers. Raster-handling~apabilities are essential when remotely sensed and othersemicontinuous, spatial variables form an importantproportion of the source data, as is undoubtedly the casein geological applications. Fortunately, there are nowmany GIS that combine both raster and vectorcapabil-ities. The link between the two kinds of function is theease of conversion of vector files to raster format withvarious cell sizes, so that vector data can be treated asrasters. Obviously this means that several orders of mag-

8.3.1 Retrieval and measurement functions

Retrieval operations are the most simple functions ina GIS. They involve the search for specific attributes,values or ranges of values residing within the dataavailable, and the transfer of the relevant points, lirles,polygons or raster cells to a new me. Examples would beextracting all areas where granite outcrops, points wheregold has been found, fold axes, areas of positive Bouguergravity anomalies or where the abundance of a chemicalelement exceeds the threshold plus three times theaverage concentration (Section 8.2.3). More complexcategories, such as all granites with high zirconium instream sediments and low topographic relief also can beretrieved by specifying the data layers within whichthe relevant information lies. It is also possible to useretrieval operations to combine several classes as ameans of simplifying information. An example would becombining all granites, adamellites, granodiorites, tonalitesand monzonites of any age in an area as a general classof granitoids. The purpose of retrieval functions is toisolate specific features of interest, usually so that theymay be combined with other information at a later stage.One possibility is setting all cells or polygons found to avalue of 1.0 and other areas to zero. Multiplying otherdata sets by the retrieved file would lead to actual valuesonly remaining for the retrieved class. This is a form of

masking.Measurement of the distance between points, the

lengths of lines, perimeters and areas of polygons, the

222 Clmpter 8

areas occupied by raster cells conforming to some limits,and volumes of specified segments of a OEM are func-tions common to any GIS.. Potential usesin geology maybe as a first step in analysis of faults and fractures, assess-ing the relative importance of different rock types thatmay contribute toxic elements to drainage in a catch-ment area, measuring the areas and volumes of potentialsources of aggregates, or even the road travel distancesfrom one important locality to the next.

A NOT B

AORB A X OR B

Logical operators can be applied to more than twoattributes or conditions, but the operators must be insome form of priority. Figure 8.18 shows the differentresults of A and B OR C, the primary operator beingparenthesized in the two cases. Consider the exampleof a search for mineralization in limestones. As wellas fluorite, baryte (BaSO4) is a common hydrothermalmineral in such terrains, and areas of limestone witheither F or Ba anomalies might be prospective. The state-ment (limestone AND (F> 100 ppm OR Ba > 200 ppm»would narrow the search, b~t a statement with the ANDoperator prioritized would also include areas with highF and Ba that are not associated with limestones, butperhaps with barren shales.

Results of logical overlay operations can be the endproduct of a search or the basis for masking, or they maybe a step in a more complex sequence involving otherfunctions.

Arithmetic overlay operations are appropriate whenconsidering data that combine both attributes andvariables, generally assembled to exploit some concep-tual model.. The general idea is to use the data and themodel to rate or score areas or cells according to howwell they satisfy the model, A good example is providedby targeting possible sites of porphyry-style coppermineralization. These always occur in association withintermediate to acid granitoid intrusions, 50 information

8.3.2 Overlay operations

Overlay operations can involve both arithmetic andlogical approaches, and seek to combine informationfrom two or more layers of data to derive another.Combinations involving arithmetic overlays can becomeextremely complex, so logical overlays are consideredfirst. Both require considerable forethought and plan-ning in the context of some model.

Logical operations involve the rules of simple Booleanalgebra, which use the operators AND, OR, XOR andNOT to see whether a particular condition is true or false.These logical relationships are understood most easilyusing sets and Venn diagrams. Figure 8.18 shows a num-ber of sets of attributes as Venn diagrams for a variety oflogical statements~ the shaded portions indicating theconditions where the statements are true. As an exampleof a simple but realistic modeli consider a search for areasof potential hydrothermal mineralization in an area con-taining limestone. Limestones are prone to various typesof hydrothermal mineralization, as they are attendedby reducing conditions and are subject to easy solu-tion, by hydrothermal fluids. A common hydrothermalmineral in limestone country is fluorite (CaF2). Given ageological map and stream-sediment fluorine deter-minations, an indicator of possible areas of mineraliza-tioRwould be derived by using the statement (limestoneANDF >100 ppm).

To illustrate the OR operator, consider a model for the10catioR of useful wearing-course- aggregate. In an areadominated by weak shales and sandstones there aresmall outcrops of basalts and slightly metamorphosedgreywackes that would be ideal. The statement (basaltOR greywacke) would find all mapping units that areeither basalt or greywacke, or both together.

For the NOT operator an illustration would becondi-tionsfor constructing a road. Areas of low slope wouldbe ideal, but those occupied by peat deposits w9uldbe potentially disastrous. The statement (slope < 20 NOTpeat) would show all gentle slopes except those occu-pied by peat.

The XOR operator is a little more complicated, andis equivalent to 'either A or B, but not both together'.This operator is rarely used in geological applications,but is useful in selecting areas for specific land-uses.

Non-Image Data and Geographical Information 223

A second functiol\ that might help in this sitingexercise would be measuring the distance of every cellfrom established transport routes. In the porphyry-copper example, this would be the means of establishingproximity to granite outcrops.

Many geological applications of GIS use proximityanalysis in relation to various types of boundary. Forexample, many mineral deposits produced by move-ment of solutions through .ock, such as hydrothermal,contact metasomatic and pegmatite types, are spatiallyrelated to faults, unconformities and contacts of igneousintrusions. There is a higher probability that they lie closerto such features than in homogeneous rock. Corridors orbuffer zones of a specified width astride boundaries ofthese kinds open up more complicated, but appropriateoverlay operations.

Many other neighbourhood operations rely on movingsearch areas that produce statistical summaries, includ-ing averages, totals, maximum and minimum values,and measures of diversity, for the immediate vicinity ofpoints or cells in vector or raster data. Specialized textswill contain accounts of other neighbourhood opera-tions that are more appropriate for land-use and other

nongeological applications.

about their distribution is needed, perhaps combiningareas where they crop out with measures of proximity tothem (Section 8.3.3). As porphyry deposits occur nearthe top of such intrusions, seeking buried intrusionscould prove important. Their low density and low mag-netic susceptibility relative to country rocks producesnegative anomalies on both gravity and magnetic maps.Copper concentration in drainage or soil geochemistrydata is an obvious part of the model, but high copper canbe associated with unmineralized basaltic rocks, so thereneeds to be some means of screening such false signals.Basalts contain far higher chromium concentrations thangranitoids, so high Cu/Cr ratios should characterizemineralized granitoids. Porphyry mineralization is formedby the transport of metals in hot, watery fluids derived ata late-stage in the fractional crystallization of granitoidmagma. Typically they result in alteration °t feldspars toclay minerals, which have distinctive spectral signatureson remotely sensed images covering the SWIR. Por-phyry deposits also contain large volumes of dissemin-ated iron and copper sulphides, which are broken downto brightly coloured ferric hydroxides during weather-ing to give gossans. The model therefore should includedata from remote sensing that highlight such alteration.Suitable data sets would be Landsat TM 7/5 and 3/1band ratios.

Each criterion in the model must be assessed in termsof how the values of the variable involved can be dividedinto ranges that carry a score in favour of the presenceof porphyry-copper mineralization. It is useful to setthresholds for some critical factors beyond which valuesdefinitely militate against a satisfactory conclusion, inorder to exclude the relevant cells from the model.In developing the model, some criteria will be moreencouraging than others, so the scoring often involvesdifferently weighting the data sets used. The complexoverlay operation then sums the weighted scores foreach cell, resulting in a map that prioritizes the regionfor more detailed exploration and expenditure of funds.Virtually every kind of mineral deposit, hydrocarbonplay or groundwater source could be expressed in such aform, so that the search is narrowed and rationalized.

8.3.3 Neighbourhood operations

Neighbourhood operations evaluate the characteristicsof the area surrounding a specified location. One func-tion is to count the number of specified features within aparticular distance from a point, line or boundary. Anexample would be to count the number of mines withinan economic trucking distance of every cell in an area asa means of deciding on possible sites for ore-processingplants that will serve several mines. That example wouldalso use other considerations to highlight the mostfavourable sites.

8.3.4 Operations involving OEMs

As well as estimations of topographic slope and simulationof solar shading using first derivative filters (Section 8.2.5),OEMs form the basis for extracting specific geomorpho-logical features, such as lines of drainage, stream order,ridges and catchment boundaries. These operationsinvolve a search outwards from a starting location usinga specific decision rule. In seeking the path of water flowover a OEM the rule used might be to start at the highestelevation in the area and then move to the cell among theeight surrounding cells that has the lowest elevation.The operation would then be repeated until either it metthe edge of the area, or a cell in a depression with no out-let. This would trace one drainage path. Starting againat the next highest elevation would trace another, andso on, until all. topographically defined drainages hadbeen defined. The connectivity between tributaries helpsdefine the number of tributaries, and thus stream order,for all the drainages. Catchments above specified pointscan be defined by spreading only to adjacent cells Withthe same or higher elevation, until there are no adjacenthigher cells, thereby defining watersheds. Such opera-tions have great potential in hydrological and hydrogeo-logical applications, where the land unit of interest isthe catchment basin. Figure 8.19 shows watersheds anddrainages paths that such analysis has extracted from aOEM of part of north-east Africa.

Although mainly of interest to planners, intervisibilityfunctions define 'the areas in direct line of sight from

224 Chapter 8

(a) (b)Fig. 8.19 Digital elevation model (GTOPO30) of central part of Fig. 8.12 (a), and (b) vectors for watersheds and drainage pathsderived from it.

specified locations. This uses the DEM to find featuresthat screen straight-line connections between the targetlocation ilnq parts of the surrounding landscape. Thismight be of use in satisfying the ever increasing environ-mental and amenity constraints placed on developmentof extractive industry and route orientation, with whichgeologists might easily become involved.

economic connotation of fracture-controlled mineraldeposits, hydrocarbon migration paths and ground-water resources, as well as leakage potential fromfractured waste-disposal sites and seismic risk in heavilyfaulted terrain. Hence the near obsession of someremote sensing geologists with fracture analysis, andthe growth of a not inconsiderable cottage industry.Geological remote sensors are sometimes regarded witha jaundiced eye in consequence of a welter of learnedpapers in the 1970s and 1980s, which rested on masses oftedious, web-like maps and largely meaningless rosediagrams indicating the length, density and azimuth oflines on images.

Given that it is possible to weed the illusory from thereal, the digitization of mapped and interpreted faultsand fractures in vector format now allows useful statist-ical information to be extracted. Software is availableto define fracture spacing and density, the density offracture intersections, spatial trends and a variety ofmore sophisticated measures. These derivatives can thenbe combined with other data layers using operationscovered in Sections 8.3.1-8.3.3.

8.3.5 Operations involving faults and fractures

As well as the quest for hitherto undiscovered potatofields, lost cities and hidden oceans, it is tempting for thegeologist to relax by drawing dense masses of straightlines on remotely sensed images. This often is supportedby the misguided belief that all apparent linear con-nections between elements of terrain represent faultsor fractures that surface mapping has missed. In truth,this once popular pastime owes much to imaginationand the line-seeking predilection of human cortical cellassemblies (Section 2.1), for individuals rarely agreeon all linear features observed on images. Nonetheless,suitably enhanced images do reveal many known andunsuspected real faults and fractures, the trick being todevise means of ascertaining their reality .

N ow, the presence of fractures in the crust, particularlythose involving dilatation, has considerable significancefor the channelling of various kinds of fluids. Exactlywhat the fluid dynamics are in fractured rocks asopposed to those in porous and permeabie media ISnot well known. What is abundantly clear is the strong

8.4 Concluding note

The use of non-image data in raster form and applicationof GIS methods to multivariate information is at the highend of image inte!pretation in terms of costs in timeand money. Set against this are a number of mitigating

Non-Image Data and Geographical Info.

factors. Much non-image data initially is generated forlimited uses, and if they fail to deliver the 'goods' theyare often discarded, despite the frequently high costsof acquisition. For many of them, ingenuity can easilydevise other applications. Indeed, hidden in their originalform is important information with unexpected connota-tions. Recasting them in digital form, either as rasters orin vector format, helps reveal these surprises, particularlyby using routine enhancement techniques. Multiple usesdivide survey and data-processing costs among severalprojects. The basic element in GIS methods is integrationof data, and there is no reason why this should notlead to integration of seemingly unrelated projects. Asan example, it requires no stretch of the imaginationto see that drainage geochemistry can serve mineralexploration, monitoring risks to humans, livestock andenvironments, and in some cases may help explain fea-tures of natural vegetation. All three have other commonlinks, such as soil type.

By establishing an ethos of using spatial data basesand GIS methods for all aspects of the environment in itsbroadest sense, public administrations from the local,through national to regional and global levels can fostercloser links between all the disciplines involved. Each hascontributions to make to all the others. Social, economicand environmental problems are not unrelated in real-ity. Treating different problems separately often leads tounsuspected difficulties in another field. So such a fusionof effort as well as data can chart a course that takes as itsstarting point the interconnectedness of all things.

Associated resources on the CD-ROM

You can access the resources by opening the HTML meMenu.htm in the Imint folder on the CD using your Webbrowser. More detailed information about installing theresources is in Appendix C.

On the Menu page, select the links Additional Images:Geophysical, Geochemical Data or Digital Elevation,where you will find examples of images based on avariety of nonimage data.

Further reading

Anfiloff, V. & Luyendyk, A. (1986) Production of pixel mapsof airborne magnetic data for Australia, with examples forthe Roper River 1 : 1 000 000 sheet. Exploration Geophysics 17,113-117.

Aronoff, S. (1989) Geographic Information Systems: a ManagementPerspective. WDL Publications, Ottawa.

Arvidson, R.E., Guinness, E.A., Strebeck, J.W., Davies, G.F. &Schultz, K.J. (1982) Image processing applied to gravity andtopographic data covering the continental United States.EOS (Transactions of the American Geophysical Union) 63,261-265.

Batson, R.M., Edwards, K. & Eliason, E.M. (1976) ,stereo and Landsat pictures. Photogrammetric Enginee,Remote Sensing 42,1279-1284.

Bolivar, S.L., Balog, S.H. & Weaver, T.A. (1982) ResO\characterization for uranium mineralization in the MontrLlox 20 Quadrangle, Colorado. Proceedings of the 5th Annul.Uranium Seminar, New York, pp. 37-48. American Institute ofMining Metallurgy and Petroleum Engineering, New York.

Bonham-Carter, G.F., Agterberg, F.P. & Wright, D.F. (1988)Integration of geological data sets for gold exploration inNova Scotia. Photogrammetric Engineering and Remote Sensing54,1585-1592.

Burrough, P.A. (1986) Principles of Geographical InformationSystems for Land Resources Assessment. Oxford UniversityPress,Oxford.

Campbell, A.N., Hollister, V.F., Dutta, R.V. & Hart, P.E. (1982)Recognition of a hidden mineral deposit by an artificial intel-ligence program. Science 217,927-928.

Drury,S.A. & Walker,A.S.D. (1987) Display and enhancement ofgridded aeromagnetic data of the Solway Basin. InternationalJournal of Remote Sensing 8, 1433-1444.

Drury, S.A. & Walker, A.S.D. (1991) The use of geophysicalimages in subsurface investigations of the Solway Basin,England. Surveys in Geophysics 12, 565-581.

Edwards, K. & Davis, P.A. (1994) The use of intensity-hue-saturation transformation for producing colour-shaded reliefmaps. Photogrammetric Engineering and Remote Sensing 60,1369-1374.

Gibson, P. (1992) Evaluation of digitally processed geophysicaldata sets for the analysis of geological features in NorthernIreland. International Journal of Remote Sensing 14,161-170.

Graham, D.F. & Bonham-Carter, G.F. (1993) Airborne radio-metric data: a tool for reconnaissance geological mappingusing a GIS. Photogrammetric Engineering and Remote Sensing59,1243-1249.

Guinness, E.A.R.E., Arvidson, C.E., Leff, M.H., Edwards, K. &Bindschadler, D.L. (1983) Digital image processing appliedto geophysical and geochemical data for Southern Missouri.Economic Geology 78, 654-663.

Harrington,H.J., Simpson, C.J. & Moore, R.F. (1982) Analysis ofcontinental structures using a digital terrain model (DTM) ofAustralia. BMR Journal of Australian Geology and Geophysics 7,68-72.

Hastings, D.A. (1983) Synthesis of geophysical data with space-acquired imagery: a review. Advances in Space Research 3,157-168.

Hoffman, P.F. (1987) Continental transform tectonics: GreatSlave Lake shear zone (-1.9 Ga), north-west Canada. Geology

15,785-788.Jenson, S.K. & Domingue, J.O. (1988) Extracting topographic

structure from digital elevation data forgeographicalinfor-mation system analysis. Photogrammetric Engineering andRemote Sensing 54, 1593-1600.

Junkin, B.G. (1982) Development of three-dimensional spatialdisplays using a geographically based information system.Photogrammetric Engineering and Remote Sensing 48, 577-586.

Karpuz, M.R., Roberts, D., Olesen, 0., Gabrielsen, R.H. &Herrevold, T. (1993) Applications of multiple data sets tostructural studies on Varanger Peninsula, Northern Norway.International Journal of Remote Sensing 14, 979-1003.

Kowalik, W.S. & Glenn, W.E. (1987) Image processing of aero-magnetic data and integration with Landsat images forimproved structural interpretation. Geophysics 52,. 875-884.

Non-Image Data and Geographical Infonnation 225

factors. Much non-image data initially is generated forlimited uses, and if they fail to deliver the 'goods' theyare often discarded, despite the frequently high costsof acquisition. For many of them, ingenuity can easilydevise other applications. Indeed, hidden in their originalform is important information with unexpected connota-tions. Recasting them in digital form, either as rasters orin vector format, helps reveal these surprises, particularlyby using routine enhancement techniques. Multiple usesdivide survey and data-processing costs among severalprojects. The basic element in GIS methods is integrationof data, and there is no reason why this should notlead to integration of seemingly unrelated projects. Asan example, it requires no stretch of the imaginationto see that drainage geochemistry can serve mineralexploration, monitoring risks to humans, livestock andenvironments, and in some cases may help e~plain fea-tures of natural vegetation. All three have other commonlinks, such as soil type.

By establishing an ethos of using spatial data basesand GIS methods for all aspects of the environment in itsbroadest sense, public administrations from the local,through national to regional and global levels can fostercloser links between all the disciplines involved. Each hascontributions to make to all the others. Social, economicand environmental problems are not unrelated in real-ity. Treating different problems separately often leads tounsuspected difficulties in another field. So such a fusionof effort as well as data can chart a course that takes as itsstarting point the interconnectedness of all things.

Associated resources on the CD-ROM

You can access the resources by opening the HTML fileMenu.htm in the Imint folder on the CD using your Webbrowser. More detailed information about installing theresources is in Appendix C.

On the Menu page, select the links Additional Images:Geophysical, Geochemical Data or Digital Elevation,where you will find examples of images based on avariety of nonimage data.

Further reading

Anfiloff, V. & Luyendyk, A. (1986) Production of pixel mapsof airborne magnetic data for Australia, with examples forthe Roper River 1 : 1 000 000 sheet. Exploration Geophysics 17,113-117.

Aronoff, S. (1989) Geographic Infonnation Systems: a ManagementPerspective. WDL Publications, Ottawa.

Arvidson, R.E., Guinness, E.A., Strebeck, J.W., Davies, G.F. &Schultz, K.J. (1982) Image processing applied to gravity andtopographic data covering the continental United States.EOS (Transactions of the American Geophysical Union) 63,261-265.

Batson, R.M., Edwards, K-. & Eliason, E.M. (1976) Syntheticstereo and Landsat pictures. Photogrammetric Engineering andRemote Sensing 42,1279-1284.

Bolivar, S.L., Balog, S.H. & Weaver, T.A. (1982) Resourcecharacterization for uranium mineralization in the Montroselox 20 Quadrangle, Colorado. Proceedings of the 5th AnnualUranium Seminar, New York, pp. 37-48. American Institute ofMining Metallurgy and Petroleum Engineering, New York.

Bonham-Carter, G.F., Agterberg, F.P. & Wright, D.F. (1988)Integration of geological data sets for gold exploration inNova Scotia. Photogrammetric Engineering and Remote Sensing54,1585-1592.

Burrough, P.A. (1986) Principles of Geographical InformationSystems for Land Resources Assessment. Oxford UniversityPress, Oxford.

Campbell, A.N., Hollister, V.F., Dutta, R.V. & Hart, P.E. (1982)Recognition of a hidden mineral deposit by an artificial intel-ligence program. Science 217,927-928.

Drury,S.A. & Walker,A.S.D. (1987) Display and enhancement ofgridded aeromagnetic data of the Solway Basin. InternationalJournal of Remote Sensing 8, 1433-1444.

Drury, S.A. & Walker, A.S.D. (1991) The use of geophysicalimages in subsurface investigations of the Solway Basin,England. Surveys in Geophysics 12,565-581.

Edwards, K. & Davis, P.A. (1994) The use of intensity-hue-saturation transformation for producing colour-shaded reliefmaps. Photogrammetric Engineering and Remote Sensing 60,1369-1374.

Gibson, P. (1992) Evaluation of digitally processed geophysicaldata sets for the analysis of geological features in NorthernIreland. International Journal of Remote Sensing 14, 161-170.

Graham, D.F. & Bonham-Carter, G.F. (1993) Airborne radio-metric data: a tool for reconnaissance geological mappingusing a GIS. Photogrammetric Engineering and Remote Sensing

59,1243-1249.Guinness, E.A.R.E., Arvidson, C.E., Leff, M.H., Edwards, K. &

Bindschadler, D.L. (1983) Digital image processing appliedto geophysical and geochemical data for Southern Missouri.Economic Geology 78, 654-663.

Harrington, H.J., Simpson, C.J. & Moore, R.F. (1982) Analysis ofcontinental structures using a digital terrain model (DTM) ofAustralia. BMR Journal of Australian Geology and Geophysics 7,68-72.

Hastings, D.A. (1983) Synthesis of geophysical data with space-acquired imagery: a review. Advances in Space Research 3,157-168.

Hoffman, P.F. (1987) Continental transform tectonics: GreatSlave Lake shear zone (-1.9 Ga), north-west Canada. Geology15,785-788.

Jenson, S.K. & Domingue, J.O. (1988) Extracting topographicstructure from digital elevation data for geographical infor-mation system analysis. Photogrammetric Engineering andRemote Sensing 54, 1593-1600.

Junkin, B.G. (1982) Development of three-dimensional spatialdisplays using a geographically based information system.Photogrammetric Engineering and Remote Sensing 48, 577-586.

Karpuz, M.R., Roberts, D., Olesen, 0., Gabrielsen, R.H. &Herrevold, T. (1993) Applications of multiple data sets tostructural studies on Varanger Peninsula, Northern Norway.International Journal of Remote Sensing 14, 979-1003.

Kowalik, W.S. & Glenn, W.E. (1987) Image processing of aero-magnetic data and integration with Landsat im~ges forimproved structural interpretation. Geophysics 52, 875-884.

226 Chapter 8

O'Sullivan, K.N. (1986) C<'rnputer enhancement of Landsat.Magnetic and other regional data publications. Volume 2,Geology and Exploration, 13th CMMI Congress, Singapore,pp.1-10.

Paterson, N .R. & Reeves, C. V .(1985) Application of gravity andmagnetic surveys: the state of the art in 1985. Geophysics 50,2558-2594.

Reeves, C.R. (1985) Airborne geophysics for geological map-ping and regional exploration.lTC Journal 3, 147-161.

Salvi, S. (1995) Analysis and interpretation of Landsat syntheticstereopairs for the detection of active fault zones in the Abruzziregion (central Italy). Remote Sensing of Environment 53,153-163.

Seemuller, W.W. (1989) The extraction of ordered vectordrainage networkS from elevation data. Computer Vision,Graphics Image Processing 47, 45-58.

Wadge, G., Young, P.A.V. & Mason, D.C. (1992) Simulation ofgeological processes using an expert system. Journal of theGeological Society of London 149,455-463.

Lee, M.K., Pharaoh, T.C. & Soper, N.J. (1990) Structural trendsin central Britain from images of gravity and aeromagneticfields. Journal of the Geological Society of London 147,241-258.

Majumdar, T.J., Mohanty, K.K. & Srivastava, A.K. (1998) On theutilization of ERS-1 altimeter data for offshore oil exploration.International Journal of Remote Sensing 19, 1953-1968.

Marble, D.F. & Peuquet, D.J. (1983) Geographic informationsystems and remote sensing. In: Manual of Remote Sensing(ed. R.N. Colwell), 2nd edn, pp. 923-958. American Societyof Photogrammetry, Falls Church, Virginia.

McGuffie, B.A., Johnson, L.F., Alley, R,E. & Lang, H.R. (1989)IGIS: computer-aided photogeologic mapping with image-processing, graphics and CAD/CAM capabilities. GeobyteOctober, 8-14.

McMahon, M.J. & North, C.P: (1993) Three-dimensional inte-gration of remotely sensed geological data: a methodologyfor petroletml exploration. Photogrammetric Engineering andRemote Sensing 59,1251-1.256.

O'SuVivan,K.N. (1983) The role of image processinsin mineral,exploration: why stop at Landsat? Advances in Space Research

3, 16~-17l.

Most applications of remote sensing in geology involvethe delineation of structures and the discrimination ofdifferent rock and soil types, often in spite of variationsin natural vegetation cover but sometimes exploitingthem. This chapter examines the contexts and potentialroles for remote sensing in the day-to-day activities ofprofessional geologists. Much can be learned from iitdi-vidual case studies, but such is the range of possibilitiesthat to illustrate them well would require a book on eachapplication area in its own right. Any project has its owninternal logic, geared to the nature of the terrain, theavailability of image data, costs and efficieI}.cy, and thespecific objectives of the investigation. Many case studiesskate over the generalities or omit them for the sake ofbrevity, assuming that readers have some knowledgeof geological image interpretation, and even for reasonsof commercial confidentiality .

Selecting a remote sensing strategy involves consider-ing how geological models comprise a number of generalcomponents, and seeking means whereby informationfrom images or other graphic data can assist in assem-bling these components. Chapter 9 covers several topicsin which geologists playa crucial role, examines the kindsof information that they need to contribute in helping toachieve the objectives of a particular programme, andsuggests the kinds of data and the way that they can behandled appropriately to provide this information. It isnot a 'cookbook' but a guide to orientating the methodsof remote sensing to the observational and analytical skillsthat have long been the geologisf s stock in trade.

Can remote sensing supplant field observations? Theanswer is a resounding 'No!', for a number of reasons.First, many of the features that are essential in buildingup a geological picture, such as petrographic texture,minor tectonic and sedimentary structures, and detailedstratigraphic and petrogenetic relationships are far toosmall to be resolved by any.conceivable remote sens-ing instrument. Second, truly fresh rock is rarely, if everexposed, being covered by weathering patinas, soil andvegetation of different types. The hammer, hand lens andultimately the microscope and geochemical analyticaltechniques ferret out the mineralogical and textural fea-tures that geologists use in classifying rocks and under-standing their genesis. Third, even if fresh rocks wereexposed completely, detailed knowledge of the spectraof rock-forming and accessory minerals reveals that fewcan be identified and separated uniquely, nor can theirproportions be assessed precisely because of variationsin the strength of their associated spectral features inpure form. In the reflected region detectable minerals

are restricted mainly to those carrying ferric, hydroxyland carbonate components, together with differences inoverall reflectivity that stem from a multitude of factors.The thermally emitted region is more promising becauseof the shifts in spectral features associated with variationsin silicon and aluminium co-ordination with oxygenand other molecular features, which give informationon rock-forming minerals. Although giving some guideto rock composition, overlaps and different intensitiesof spectral features still mean that multispectral thermaldata are unlikely ever to provide a.comprehensive map-ping capability. Radar data provide information on theweathering and erosionaltextures of rocks, which helpdetermine variations in surface roughness, and thepresence of comer reflectors smaller than the method'sresolution dimensions. Although helpful in some casesby providing another dimension to image interpretation,neither is radar an all-purpose geological mapper in itsown right.

So information provided by remote sensing has tobe verified and amplified by field work, in much thesame way as the subsurface information provided bygeophysical surveys needs to be interpreted in the lightof surface observations, and ultimately by the drill.However, remote sensing views the surface in a detailedmap-like form, providing a synopsis of areas that arefar larger than can be encompassed by a ground viewand much faster than field mapping. It reveals featuresthat are difficult to map at the surface and allows fieldobservations 10 be extrapolated into unvisited areas. Inthis regard remote sensing reduces the amount of fieldwork needed to achieve project objectives to, and alsodirects those investigations to areas containing the mostcritical evidence. All this helps increase a geologist's effici-ency. By venturing into nonvisible parts of the spectrumand through digital image processing, remote sensingincreases the chances of making discoveries that wouldbe overlooked in conventional field work. It amplifiesthe geologist's powers of observation and detection.

Remote sensing provides a kit of tools that a geologistcan deploy as and where necessary. Making a selectionfrom this toolbox involves several considerations:1 Application definition. What geological features andprocesses relate to the application? Is it possible to extractinformation about them from image data?2 Resolution and scale. What is the minimum spatialresolution needed to detect the required geologicalinformation? Over what range of scales do the geolo-gical features and processes manifest themselves? Ingeneral, the finer the resolution the larger is the scale of

227

228 Chapter 9

and Plate 5.10. This iflvolves the presence of livingvegetation (sometimes this is a distinct aid, sometimes ahindrance, depending on the local climatic conditions),varying soil-moisture content and the Sun angle atthe time of overpass. At moderate to high latitudes thewinter Sun is lower in the sky than in the summer, sothat the shadows of muted topographic features are moreapparent. In a few applications that involve short-livedphenomena, such as studies of volcanoes, changing rivercourses and environmental changes associated withresource-based industry, repetitive cover is essential. Formost geological applications, however, a single imagewith favourable atmospheric and vegetation conditionsis often all that is needed. Moreover, the date of theimage is frequently immaterial (this can be financiallyimportant, as data from early years are often cheaperthan new data). There are, of course, areas that are soplagued by perpetual cloud cover that reflected andthermal images rarely reveal the surface. Whatever theapplication in such an area, only radar remote sensingprovides any assistance to the geologist.

9.1 Geomorphology

The study of landforms and landscape involves de-scription, classification and analysis of the evolutionof the planet's surface. Although sometimes seen asa marginalized topic within geology, geomorphologyis of fundamental importance because it is one of thedriving forces of biological evolution, and controlshabitability. In landscape studies the element of detail isessential, for landforms are enormously complex on allscales. The information base has to match reality, andimages at different scales are the only realistic base forfull interpretation.

Remotely sensed images lend themselves ideally togeomorphological studies at any required scale and areaof coverage. As well as specific topographic and hydro-logical feature:), image:) can provide detail:) on vegetationtype and cover, on which landscape evolution dependstogether with climatic variations, underlying geologyand time. The images in Chapter 4 provide a rich com-pendium of geomorphological examples, as well as illus-trating most aspects of geology.

Until the widespread accessibility of satellite imagesthe science had drawn in on itself, after the megascopicideas of Davis fell into disfavour. It focused on the smallscale, the short-term and the study of process, becomingalmost divorced from geology. This trend coincided withthe rise of ideas on systems analysis and the applicationof statistical methods, to which detailed information onsmall areas lent itself. The landscape became idealizedas elements linked by flows of mass and energy. Suchan approach found itself incapable of extrapolation to

presentation. The maximum scale needed is often aguide to the minimum resolution size. If the requirementinvolves a maximum scale of 1 : 1 million then there islittle point in using data with a resolution of 10 ni orless. As Equation (2.1) reveals, what the eye can see in animage depends on the scale and viewing distance, andthere are definite limits. A project involving products at1 : 10000 scale needs a resolution less than 5 m.3 Spectral coverage. Is multispectral, single band orpanchromatic imagery required, and what regions of thespectrum are appropriate for the features being sought?The answers depend on whether the main emphasis islithological discrimination, mapping geological struc-tures or evaluating terrain. The first hinges on the degreeto which rocks and the soils derived from them are freeof vegetation cover, and needs selection of wavebandsthat mineralogically determined spectral featUres affect.Structural mapping and terrain analysis depend to alarge extent on the interpretation of high-frequency topo-graphic features. So, our better perception of fine detailin grey-tone images (Section 2.2) than in colour imagesoften means that single-band or panchromatic imagesare more appropriate than false-colour composites, par-ticularly if they can be viewed stereoptically. Structuralgeology and lithological mapping are complementary,however, and some tectonic features can be inferredonly by their displacement, or relationship to lithologicalboundaries, when colour images can be useful. Similarly,geomorphology involves considering the relationshipbetween physical processes of landscape developmentand the effects of vegetation cover, when false-colourimages are the best means of providing informationon plant distribution. For some applications nonimagedata are appropriate, such as topographic elevation,potential-field or geochemical data in a form suitable foruse in image processing or GIS operations.4 Data processing. Chapters 5 and 8 discussed a widerange of possible manipulations used in enhancing orextracting the information contained in images. BecausenIl involve "me nnd therefore CO5t, Qnd in 50me c;t:;e:;produce new images that require varying depths ofinsight into what they reveal, a choice of techniquesis important in planning a project. Some are aimed atimproving the visual qualities of images, such as contraststretching and edge enhancement, others at expressinghidden information and relationships between differ-ent data sets, such as ratioing and principal componentanalysis. A further suite extracts information from manyspatially related data sets, as in classification and GISoperations.

More mundane considerations are whether the datarequired are indeed available, how much they wouldcost, and if they have acceptable cloud cover and atmo-spheric haze. An important point concerns the time ofyear for data acquisition, as emphasized in Section 5.6.3

Geological Applications of Image Data 229

large features and time spans of geological magnitude.It was an exclusively 'bottom upwards'approach. Studiesof other planetary bodies, s\:tch as the Moon, Mars andsatellites of the Outer Planets based themselves entirelyon images of successively inct:easing scale, forcing a'top downwards' approach and much different methods,which largely relied on stratigraphic rules of observation.Elements concerned with planetary evolution beganto emerge. As well as discovering features of giganticdimensions, this approach needs considerable ingenuityto interp!et the processes involved. Planetology foundan echo in geomorphology, with the focus shifting onceagain to grand theories.

Climatic geomorphology builds on Davis' stageist theoryof evolution. It is based on the notion that today's relief-forming mechanisms vary according to zoned climaticconditions, and this relationship can be extrapolated to'"the past. Much of the terrestrial landscape is seen as beinginherited from processes during past climatic periods.As climate currently is in transition from a major warm-ing only 10 000 years ago this seems inescapable. Beforethe end of the last glacial period the world was a muchdifferent place from today for about 100 000 years, andimposed features that have not yet come to balance withmodem climatic conditions. Indeed, the last 2-4 millionyears have been a period of great climatic instability, andthe pace of succeeding glacial and interglacial periodshas a far higher frequency than climatic shifts in earlier,more 'normal' geological times. Elements of a steady-state, pre-Pliocene set of landscapes still play an under-lying role in setting today's geomorphological scene. Butclimate is not the only dynamic in the evolution of theEarth's surface.

Structural geomorphology focuses on the fundamentalcontrol of geology and tectonics over landscape, partlyby different resistances of rocks to erosion and partlybecause of active vertical movements Qf the crust throughdifferent mechanisms. Climate plays a role but its effectsare conditioned by rocks and their movements. A unityof the two strands of terrestrial planetology arose fromthe growing recognition that rapid erosion by glaciersand powerful rivers unload,s the lithosphere, therebyallowing uplift. Removing mass, mainly from valleys,unloads the lithosph~re beneath the remaining peaksand higher ground so that they continually rise, t~erebygenerating mountain ranges of Himalayan and Alpinemagnitude. This recognition stems from the discoverythat much of the rapid Himalayan uplift took place inthe Pliocene and Pleistocene, whereas the underlyingtectonic forces that generated thick, buoyant continentalcrust were initiated tens of million years befQre. The mostrapid uplift was delayed until rapid erosion becamepossible when climate changed so that glaciers and themeltwater derived from them could sculpt the landscapeat an accelerated rate.

Oearly, geomorphol"gy is scale-de~ndent and unifica-tion is only possible by studies at all scales from thatshowing minor gullies to ~ontinent-wide features. Froma lower limit of around 1 : 10000 supported by aerialphotography and the highest resolution satellite data,remotely sensed data presents a continuum of scalesdown to 1 : 10 million or less through platforms such asLandsat and SPOT, and the coarse-resolution meteoro-logical satellites. Even terrain features on the submetrescale, as expressed by surface roughness, are amenableto radar imaging.

Although most landscapes change at rates that aretoo slow to be monitored within the 60-y-ear historyof systematic remote sensing, som~ processes are quickenough for change detection to be possible. These in-clude aeolian, coastal, glacial, volcanic, lacustrine, mass-wasting and some fluvial processes. Judicious selectionof time series of an area, if available, can help monitor thedynamics involved in landscape cha~ge..

Because landscapes are three-dimensiona,l, stereo-scopy is of great importance in geomorphology. Whereorganized agriculture camouflages topographic fea-tures with patchworJ< patterns (Fig. 4.48) it is essential.Gradually th~ SPOT programme, and later stereoscopicsystems, are providing global stereoscopic coveragethat previously was possible only through large-scaleaerial photographs, which do not allow continuity ofthree-dimensional visuali.r;atlon. Although stereoscopicLarge Format Camera images covered several. millionsquare kilometres, sadly the deployment of this low-costinstrument on the Space Shuttle was limited to just onemission in 1984.

In rigorous statistical analyses of land£orm .st~reo-scopic images include extraneous clutter as a r~sQJt ofvariations in the reflective properties of the surface at awide range of spatial frequencies. Moreover, proceedingfrom a mental stereomodel to actual measurements ofrelief is a tedious process. Accurate expressions of thevariation in topographic elevation through digitaleleva-tion models form the base for quantitative landformanalysis (Section 8.2.4). Digital elevation models may bederived from topographic contour maps, by automatedparallax estimation from digital stereoscopic imageryand more recently by interferometric radar. Such OEMsenable the scale-dependence of terrain features to bemonitored, amounts and rates of mass transfer to beestimated and drainage channels and wiltersheds tobe extracted automatically. The entire continental surfaceis covered at about 1 km resolution by the GTOPO30OEM, to be supplanted at about 100 m by the SRTMinterferometric-radar system (Chapters 7 and 8) plannedfor completion by 2001-2 by JPL-NIMA.

Geomorphological mapping poses considerabl~ prob-lems .as a result of the many different approaches thatare in,yogue and ,the sheer complexity of landforms, All

230 Chapter 9

mapping involves simplifying the complex continuurnofreality through symbols, categorization and areal diVision-ca sort of data reduction in its own right. Three possiblegeneral approaches to 'division of landfprms are:pUTelyas functions of shape and pattern; according to inter-pretation of the. genetic mechanisms involved; and byrecognizing landforms of different ages using simplestratigraphic rules. It is difficult, if not impossible, to avoidwelding the three approaches together, because evidencefor all three manifests itself to the interpreter on images.Drainage systems provide ready examples of the confu-sion. These can be categorized purely on form (Fig. 4.3),but the forms always have some underlying controls, oftendue to geological factors, and may have been induced byan earlier fluvial episode subsequently incised into a newlyrising block of crust ~r as a result of land subsidence orrising sea-level. Changes in erosional baseJevel, bothfalling and rising, dominate those parts of the NorthernHemis;phere once covered by continental ice-cap$, andall continents have responded in terms of erosion anddeposition to the global sea-level rise that followed Ietreatof those ice-caps. Depending on tectonic activity, sucheustatic changes are accompanied by isostatic changesin base-level.. It is quite common in any climatic zone tofind meandering drainages that developed on an earlyplanation surface, which are now incised into rugged,rising mountains (Fig~ 4.15).

Only the most grossly generalized geomorphic cat-egories are widely agreed upon. In practice no two geo-morphological maps produced by different interpretersare comparable in detail, and few correspond except ina very general way. Producing maps is a graveyard formany ageomorphologist with a bureaucratic outlook!Essentially, terrain maps cater for specific needs, not forall. Some have attempted to map geomorphological units,which are genetically homogeneous landforms producedby specific constructional or destructional processes. Allland-forming processes overlap, however, and many land-forms owe their features to several processes of differentage. A more productive line of attack relies on a hierarchyof forms depending on scale. In such a multitiered approach,one tier of large-scale map units disappears or blurs asscale and perceptual detail decrease, to be replaced by asuccession of tiers at increasingly smaller scales. In thisway it is possible to make links between manifestationsof process and evolution from the scale of the smallestelement of interest to those that characterize entire con-tinents. This engenders a unification between understand-ing the detailed dynamics of currently active systems,through the interplay between fundamentally differ-ent types of erosional. and constructional processes inan evolutionary context, to the grandiose relationshipsbetween surface and internal processes that involve thelithosphere as a whole, areas the size of continents andtime spans of t~ns, hundreds and perhaps even thousands

of million years.. Trying,to achieve su<::t1a symbolic rela-tionship, rich in the dynamics" involved is the challengeposed to geomorphologists by planetologists workingon extraterrestrial bodies.

Even if this systematization were to be achieved, thereis st~l the likelihood that geomorphological maps willbe made by experts for experts. There is no problem orharm in the context of academic research, but it oftenseems that remotely sensed images themselves are themost useful geomorphological maps, because all theinformation is there for producing thematic maps ofany type for any purpose. However, practical decisionshave to be made and boundaries drawn. Consequentlygeomorphologists are often called on to make inter-pretations and maps to assist a sPt7cific social, environ-mental or economic programme. Co-ordinating imageinterpretation involves isolating the elements Qf the land-scape that are relevant to the programme's objectives.Geomorphology with an end other than human curiosityis often termed terrain analysis. It contributes to environ-mental and resource management, land-use and routeplanning, military strategy, arid engineering applications(Section 9.4), increasingly through GIS combined withother types of information, such as variations in vegeta-tion cover, soil type, geology and hydrology.

Chapters 4 and 7 contain many useful guidelines forgeomorphological image interpretation, as do some ofthe articles in the further reading for this chapter andmodem textbooks on geomorphology. Images from thereflected region are quite sufficient for most geomorpho-logical purposes, and it is rarely impossible to achieve sub-stantial results with single-band images alone. However,false colour images, if affordable, can add significantlyto interpretations. As the dominant features on imagesare geomorphological in origin, little enhancement isnecessary i except for cosmetic improvement by contraststretching and edge enhancement.

Thermal data and radar images increase the breadthof detectable features, with radar adding information onmicrorelief that forms a direct link to field observations.Where cloud cover dominates, as in equatorial regions,radar sometimes forms the only source of informationabout terrain, because of its all-weather capabilities.Attempts have been made in such obscured regions toproduce radar mosaics, such as Project RADAM in Brazil.Of the existing orbital SAR instruments, that carried byERS-l is inappropriate for geomorphology in all but areasof low relief because of its steep depression angle andthe dominance of images in rugged terrain by layover(Fig. 3.34). The ]ERS-l SAR has a less steep depressionangle and produces images as useful as those fromSIR-A, as can Radarsat. Radar has another importantattribute to offer geomorphologists, in .-the form ofshallow penetration of very dry sand and ice. Figure 7.4gives a dramatic example of the potentialfor discovering

Geological Applications of Image Data 231

flat-topped shield volcanoes at right and a conical volcanowith a funnel-shaped crater at left. Separating these featuresis a hummocky terrain of huge pillow-like flows alongthe axial rift. A fault scarp is prominent at left. The imageis 3 by 4 km. Courtesy of Bramley Murton, Institute ofOceanographic Sciences, UK, and Roger Searle, Universityof Durham, UK.

Fig. 9.1 Sidescan sonograph of part of the axial zone of theMid-Atlantic Ridge. Sonar images are similar to thoseproduced by radar, looking sideways from the platform trackand responding to timed acoustic returns. They are bothshadowed and vary in tone according to surface roughness.In this case, the dark upper strip is directly beneath theplatform-a submersible. Prominent on the image are

completely hidden geomorphological information usingSAR in hyperacid areas. The potential global coverageof JERS-1, ERS-1 and RADARSA T promises an upsurgein geomorphological discovery.

It is often forgotten that the 70% of Earth's surfacecovered by oceans has landforms too. Conventionalbathymetry is limited in its contribution to mapping themorphology of the ocean floor through its restriction toshipping lanes. The Seasat and ERS-1 radar altimetry data,through measurement of sea-surface elevation containinformation relevant to water depth through gravita-tional effects (Fig. 8.9). Side-scanning sonar reveals manyunsuspected features of the sea bed (Fig. 9.1) that isbeginning to revolutionize understanding.

9.2 Geological mapping

defining boundaries between rock units of different type,expressing their disposition relative to each other and totime, and delineating the tectonic structures about whichthey are disposed. Geological maps are two-dimensionalexpressions of four dimensions-three spatial dimensionsand time, and incorporate clues to Earth history and theinterplay between internal and surface processes.

Remote sensing cannot provide this information unlessit is blended with field work and laboratory investigations.As emphasized many times, image data do not directlyprovide conventional divisions between rock types, whichare and will depend on mineralogy, texture in handspecimen and thin section, and geochemistry. There aretwo main, general approaches to mapping from images;one based on image units (see Figs 6.11 and 6.12) the otheron lithofacies units.

Image units combine the textural and pattern attri-butes of the surface with the way it interacts with EMradiation in terms of albedo or spectral characteristicsexpressed on false-colour images. It is a more or lessobjective approach to mapping, and implies no- geneticconnotations. For extraterrestrial surfaces, with little ifany chance of ground verification, such an approach is in-evitable. The relative ages of the units derive from simplestratigraphic rules, such as discordant relationships. In

Although Image Interpretation in Geology is about ex-tracting geological information from images, the mainresponsibility of geologists is rationalizing this informa-tion in the form of graphic maps. How such maps aredesigned and produced is mainly outside the book'sscope, but is covered using practical examples in theCD-ROM exercises. Essentially, mapping consists of

232 Chapter 9

Occasionally there is sutficient detail to recognize generalshapes that are reminiscent of processes on the Earthproduced by the flow of fluids of different viscosity,including water, ice,l~vas and landslips (Fig. 9.3). Usingimage units has the appeal of being definable in termsof the available data,. but different units emerge fromusing different data (Fig. 6.11). More importantly, suchunits are of little use to other geologists, and the mainrequirement of a geological mapping programme is tolayout information that has a practical use, without theuser having to go through the rigmarole of understand-ing the particular means of subdivision.

In terrestrial mapping, there are few cases wherefield data or earlier geological maps are not available.Con~equently the:re are often opportunities to assignlithologies to recognizable image units, extrapolate fromknown to unvisited areas and to trace structural featuresover broad areas.. Even in cases of completely unknownterrain-increasingly rare-image interpretation can pro-ceed directly to a reconnaissance map of conventionaltype. Enpugh will be known from adjoining, mappedterrains to make an educated guess as to the likely rocksand structures there, backed up by spectral knowledgeand the appearance of different rocks in climaticallysimilar terrains. Geologists are unlikely to find manysurprises in terms of rock composition that require acomplete revision of terrestrial petrogenesis, such as vastflows of sulphur lava or methane ice that occur on otherplanetary bodies. Geologists make mistakes during imageinterpretation of new areas without field data, but wherethere is no information virtually any new observationis valuable, and new ideas help guide field work to testthem out. Armed with experience of the ways in which

Fig. 9.2 Voyager-2 image of Miranda, a moon of Uranus,showing a grooved terrain next to an older cratered terrainthat has b~n partially resurfaced to mute the craters.Younger impact craters clearly show as sharp circularfeatures. Courtesy of the Jet Propulsion Laboratory,Pasadena, USA.

studies of many planetary bodies variations in the dis-tribution of impact craters become useful (Fig. 9.2). This isbased OQ ~e observation, initially from the Moon, thatthe older the surface the greater the chance of cratershaving fofn)ed, the rate of impacting having decreasedwith ~e .t~oughout the Solar System. Other such signsof relatiyeage include the masking of features by trans-ported debris, such as ejecta from impacts or aeolian du~t.

(a)

Fig~ 9.3 (a) Magellan SAR image of the Alpha Regio region ofVenus showing volcanic domes, around 2S kilometres across,formed by extrusion of viscous magma. The curious fracturepatterns on the domes suggest extrusion after the formationofa solid skin. Courtesy of Jet Propulsion Laboratory,

Pasadena, USA. (b) Viking image of V allesMarineris on'Mars, in which the brancWng form of some of its sidevalleys ~uggest the former action of flowing water on thisnow dry planet. Courtesy of Jet Propulsion Laboratory!Pasadena, USA.

Geological Applications of Image Data 233

geological boundaries are modified by topographic reliefit should be possible to predict where hidden boundariestrend and reappear.

In general, the vast majority of geological boundariesreflect either roughly parallel compositional layering ordiscordances cutting across that layering, such as faults,igneous contacts or unconformities (Figs 4.20, 4.24a and7.10). Recognition of layer-parallel features on images,conforming to the relationship between dip and strikeof planar fabrics and relief is of primary importance, forthis enables folding and angular discordances to bediscovered, evaluated and mapped. Occasionally, layerswith distinctive spectral properties or texture (Plates 6,5and 7.2) serve as marker horizons that help in moreaccurate delineation of fold structures and relative dis-placements on faults {Plate 9.1). Although the amount ofinformation that permits such photogeologicil mappingvaries with vegetation and soil cover, this more subject-ive approach based on inferred genetic relationships leadsto results that have a predictive content of immediateuse to any geologist. It forms the bones on which moregeological flesh can be sculpted as investigations proceed.Achieving this often depends on experimentation withdigital image processing to reveal the best combinationof data, and access to a variety of renditions helps gleanall the clues available.

Of an possibilities, stereoscopic viewing is undoubtedlythe most important aid to grasping the relative dispositionof rock types and the dip of surlaces in relation to relief.It shows clearly the attitude of layers and their thickness,and using simple instruments they can be quantifiedaccurately. The law of V s (Section 4.3.1) on single imagesis a good guide to the direction of dip. Topographicelevation data used together with this help quantify thedip by using the simple three-point solution for eleva-tions on an assumed plane. Elevation values at manypoints on outcropping boundaries as they cross aDEMprovide structure contours, after appropriate surlacefitting. Alternatively, a DEM can introduce parallax intoa single image, so that when viewedstereoptically withthe original, three-dimensional relief is combined withthe image features (Fig. 8.16). Becafise of the way in whichlow-angle planar featUres follow topography closely,they are extremely difficult to detect and map on imagestakenirom directly above, eyen if they are stereoscopic.Powerful software allows the use of DEMs in creatingoblique views that incorporate perspective, from anyazimuth and altitude above the surlace. 'Draping' imageson such views simulates the 'Swiss hammer' approachto low-angle faults and unconformities (Plate 9.2).

Many existing geological maps were constructed fromfield work and aerial-photograph interpretation, oftenwith generalization about lithological variation, or to suita specific purpose that limited the number of divisions.Very often the 'nose-to-the-ground' approach prevented

the recognition of large or subtle features. Importantdivisions were frequently missed, simply because theywere invisible. Mapping huge tracts at small scales (1 :100000 and smaller) relies on traverses where the terrainpermits easy and rapid access, rather than by systematicvisits to every outcrop and the follo\".ring of boundaries.Inevitably, such reconnaissance maps omit much qetail,but a general framework does exist. Remotely senseddata helps revise these maps at larger scale. Digital carto-graphic methods open up new possibilities. Scanning anexisting colour geological map into red, green and blueseparates and then registering it to an image, allows ageologist to combine the two by usmg techniq~es suchas the HIS transform (Section 5.2.2) (Pla,te 9.3). Digitizingthe boundaries from existing maps, either automaticallyor manually, is an alternative. Merging the -boundaries\¥ith images provides guides to the distribution of litho-logies andstructuies.Either approach allows these Vectorsto be edited and new ones to be added as the first step inconversion from paper to digital-maps: Comprehensiveinterpretation of all conceivable and accessible geologicalattributes in an area is difficult to express graphically ina single paper map without distracting clutter and con-fusion. The high costs of mastering and printing litho-graphic paper maps generally mean that much of thisinformation, although available, is rarely accessible exceptby consulting archives. Storing all interpreted attributesas vector lines, polygons and points in a digital,.mappirlgsystem lays the basis for entirely new styles of map pro-duction. Trying to jam as much information as possibl~on to a conventionally published map results in com-promises, omissions and simplifications. Instead theobjective becomes to generate maps that extract detailpertinent to specific applications from the mapping database as 'one-offs'.Moreover, all information is availablefor further integrated analysis usingG1S appr9a~hes.

Geological mapping ~ims at understan<iing th~subsurface as well as surface geological features, so itbenefits from using geophysical data, such as those frommagnetic and gravity surveys (Section 8.1.2, Figs 8.1-:-8.15). Not all means of subdividing rocks are provid(CYdby field or image observations. Plate 8.3 demon$trat~sthat gamma-radiometric surVeys contain a wealth ofinformation relateq to lithological variation. Plate 9.4shows the potential of merging such data with morespatially 'clean' remotely sensed images.

9.3 Exploration

Although the guidelines for finding mineral deposits,hydrocarbon reservoirs or water are geological, explora-tion itself is fundamentally an economic activity. Noanomaly is sought merely out of curiosity. Findingabnormally high concentrations of metals, oil Of water

234 Chapter 9

beneath the surface is not the end of the story either, forthey do not necessarily constitute resources. To becomeeconomic entities, they must satisfy a wide variety ofcriteria. They must have a ready market, and be extract-able and refineable to a useful product at a total cost thatis less than their intrinsic value, thereby ensuring profit.As well as the multifarious technicalities associated withextracting the sought-for products from beneath theground and refining them, there are the purely logisticaspects of access and supply of services. l1rilike any othercommodities physical-resource production inevitablydepletes the assets of the operating organization; mostphysical resources are nonrenewable. Industries based onrenewable resources and manufacturing either increaseassets or at least maintain them ata stable level. IntrinSicvalues or market prices of physical resources fluctuatein a largely unpredictable fashion, part of whic;h is linkedto changes in the available supplies. Consequently, suc-cessful exploration can rebound on the market, as wasthe case ~th the discovery of large petroleum reservesoutside the ambit of the OPEC cartel, which since 1973had attempted to maintain oil prices at a high level.Together with other economic forces and measures toconserve energy in the 1980s, these discoveries led to adepression in the world price of oil that continues to thisday, and means that exploration is continually subject tostringent limits. In the early 1990s these limits were sorigid that major oil companies began to disband the largeteams of remote sensers that had been assembled in theirexploration departments since the mid-1970s. Much thesame trend affected virtually every company involvedwith metals and lower value resources, partly owing togluts on the market stemming from new discoveries andmore cost-effective extraction technologies, and partly toincreasingly unpredictable markets. The concept of whatconstitutes a resource or a reserve changes with time.Water is in a unique position, for it is a vital requirementfor survival and for the development of the largelyagricultural economies of the so-called 'Third World', aswell as for the smooth running of the social and industrialsystems of the developed countries,

More than any other industrial. activity, exploitationof and exploration for physical resources are bound upwith risks to invested capital. The attraction of capitaldepends on these higher risks being offset by theopportunity for a greater return on investment. No-onewill seek even gold if the required capital can be moreprofitably and safely invested elsewhere. Explorationis akin to research and development-it carries no guar-antee of profit in its own right, and certainly its success-ful outcome after evaluation is not immediate. Mostresource operations require a lead time of up to 15 yearsbefore offsetting the costs of exploration against in-come from production. However, the depletion of assetsdemands that exploration continues, except in periods

of fundamental econo~ic depression, such as that fol-lowing the Wall Street Crash of 1928.

This is not the place for a full account of the non-geological factors relating directly and indirectly toexploration. Suffice it to say that it centres on decisionsabout what to look for, where to concentrate the search,a minimum quantity of the resource in question for itto generate profit by exploitation, a minimum unit value,depending on the concentration of the commodity beingsought and its unit price, and a minimum rate of profit.This in a nutshell, is an exploration objective. Note that onlythe location of the search area involves geological con-siderations, and they are mixed up with other factors, suchas varying fiscal conditions, accessibility and communica-tions, and political stability .To a variable extentexplora-tion involves a gamble-at the outset there is only a set ofoptimistic notions, particularly when most of the depositsof the 'trip-over' variety have long been discovered. Thegreatest risk therefore is that of 'gambler's ruin', or theloss of capital through a string of unsuccessful ventures.Avoiding this disagreeable situation means estimatingthe probability of satisfying the exploration objective,and assigning funds in proportion to the risk of failurethroughout the whole operation. Exploration thereforeis conducted in stages, which gives every opportunity toproceed or stop according to the accumulation of informa-tion, not as a blind gambler would. Each stage involves areview of information gained and the degree to which therisk of not satisfying the exploration objective has changed.The expected value of the infurmation from the next stageis set against the cost of getting it. When the value exceedsthe cost exploration continues. When cost exceeds thevalue of expected information then it stops. This cumu-lative process amounts to setting lost opportunities againstthe chances of lost investment. Figure 9.4 shows the kindsof method available to metal-orientated exploration, andthe stages at which they are often deployed, in relationto growing costs. Table 9.1 shows some approximatefigures for costs and efficiencies of various explorationtechniques involved in metal exploration. Although out-dated in terms of the actual costs, it shows the relativeexpense and efficiency of each method.

The low cost and high efficiency of remote sensingmakes it a favoured method early in exploration. It bothreduces early costs and helps narrow the focus of laterstage$. This is important because narrowing the area ofsearch at the end of each exploration stage goes handin hand with reducing the degree of risk to investment.As well as helping geological reconnaissance, remote sens-ing is useful in targeting smaller areas for more detailedand costly follow-up. Moreover, its use does not endwith the preliminary stage. The interpretation can beupdated in the light of other data, and imagery forms anideal base upon which to co-ordinate this laterinforma-tion through a GIS approach.

Geological Applications of Image Data 235

detailed exploratTonreconnaissancepreliminarystudies

evaluation

field reconnaissance

x

x

remote sensing analysis

surface geological mapping

underground geologicaimapping (from drill core;

exploration geophysics

exploration geochemistry

drilling for geological data

ore body sampling I

x

I cumulativex

x

laboratory support

!geotechnical investigations

economic-technologicaiIstudies

x

x

x -.-/

time/yearsarea of coverage/km2

2 3 4

1000-300000 2-50 1-10

Fig. 9.4 Chart showing the general stages involved in a mineral exploration programme. The graph of costs is only an indication ofthe way cumulative costs mount. '

Table 9.1 Approximate costs andaverage efficiencies of someexploration methods (1988 data)

0.020.7105

>1.06104

50050

25 L<m-110i<m-2

'2,50 day-l

50025

160 krn-2

600 krn-2

15 krn-2

50 krn-2

750 krn-2

101

5025

2

160 km-l5000 km-l.

40 m-l5000 m-l

PreliminariesSatellite remote sensingInterpretation and mapAirborne remot~ sensingInterpretation and map

Airborne geophysicsMagnetic a~d electromagneticInterpretation ~nd mapLiterature search

Field studiesGeological reconnaissanceDetailed geological mapping,Geochemical surveying orientationDrainage surveySoil or biogeochemical survey

Geophysical surveyingResistivity and spontaneous potentialInduced potentialDiamond drill coresShaft sinking

10, 0.5

-'

-

Exploration is no longer a haphazard process basedon a prospector's intuition; individual experience andluck, although all three still playa role. As easily f0undresources become more scarce and economic viability of

production subject to increasing stringency, the searchrequires a more scientific basis. Targeting or play gener-ation express this ethos. Explorationists base a geologicalmo~el on the fund of knowledge ,about regional and local

km-2km-2

km~2

km-2

236 Chapter 9

structural controls on known mineralization, hydrocarbonor other resource accumulations. What is known fromresearch about the processes involved in their formationand the blend of geological features commonly associ-ated with such p~ocesses adds to the modeL Targetingalso helps define the physical and chemical peculiaritiesof different resources. These indicate the specifi{: kindsof information that are needed by the explorationistto narrow down the search and eventually pin-pointa deposit. This blend of critical information, known orto begathered,forms the basis for planning and inter-rogating a GIS.

9.3.1 Metal exploration

Deposits of metalliferous ores form by a host of pro-cesses within the igneous and sedimentary parts Qf the rockcycle. There; are few, if any, examples of ores originat-ing through exclusively metamorphic processes. A largeproportion of deposits show evidence which suggeststhat metals concentrate to ore grade through the agencyof watery fluids moving beneath the surface, as in peg-matite, porphyry, contact-metasomatic;, hydrothermal ands~ondarily enriched deposits. Such fluids commonlycarry hydrog~n ions, and have low pH. Hydrogen ionsare the main agent that breaks the covalent bonds inmany silicates and induces them to change to more stablecompounds. This induces the alteration in ordinary rocksthat is commonly associated with these deposit types.Moreover, the freedom for electrons to move in wateryfluids means that ~hey have a wide variety of redox poten-tials (Eh), which playa major role, together with pHvariatil;)ns, in oxidation~ndreduction,and the solutionand pr~cipitation of many compounds, including oreminerals. This broad group of mineralizing processesthen induces considerable changes in the common rockswith which the ores are sometimes associated.

Ore formation that largely excluded the influence ofsubsurface watery fluids, such as magmatic segregationand placer deposits, takes place during the formationof the igneous or sedimentary hosts to the ore. Con-sequently,. the primary process leaves products .that arepart and parcel of.. their host. This renders them diffi-cult, if not impossible to discern on remotely sensedimages, because they are orders of magnitude smallerthan comQlon features in the host. In their unalteredstate, such ores often blend imperceptibly into the back-ground. Occasionally, spectral features peculiar to theore minerals, as with those Qf chromite, make it possibletheoretically to hightight them using image processing.This, however, is feasible only if the ore bodies are com-parable in size with, or larger than, the spatial resolutionof the sensor. Most commercial chromiteores are tabularand rarely more thana few metres in thickness, and fewhave been discovered using remote sensing alone.

However, the geochemical affinities of metals com-monly sought in magmatic segregations, such as titanium,chromium, nickel and platinum-group metals,. meansthat it is easily possible to reduce the area of search tolikely host rocks. All four cases involve basic to ultra-basic igneous rocks..In lightly vegetated terrain their lowalbedos readily distinguish them on images from otherigneous lithologies, and from sedimentary and metamor-phic rocks. In humid areas rocks rich in ferromagnesianminerals produce fertile soils rich in montmorilloniteand illite clays that retain both water and nutrients, andcarry different plant associations and densities.

Placer deposits are even more intractable to remotesensing, for they are often buried beneath ordinary sedi-mentby virtue of their content of high-density minerals.However, experience suggests many likely sites for placeraccumulation in drainage, shoreline and aeolian environ-ments, Careful analysis of geomorphological features fromimagery thus helps in locating potential targets.

Remote sensing becomes a powerful exploration tool inits own right when the primary and secondary processesof mineralization result in the formation of spectralanomalies. This is especially so when the anomalies extendwell beyond the extent of the ore itself. Deposits formedor modified through the influence of migrating wateryfluids frequently exhibit such alteration haloes spreadthrough large volumes of rock because of pervasive fluidmigration. These quite literally can appear as targets,hopefully with a paying bull's eye at the centre.

The most common style of alteration is the break-down of feldspars and ferromagnesian minerals to avariety of clays and other hydroxyl-bearing minerals,where the SWIR range of their spectra exhibits absorp-tions with patterns that are unique to each species(Fig. 1.9). The general presence of hydroxylated mineralsin lightly vegetated terrain (but little potential for theiridentification) is provided by the contrast betweenLandsat TM bands 7 and 5, usually expressed by a ratioimage. The three SWIR bands of JERS-1 and ASTERprovide more opportunity for discrimination, but themost useful data are those from hyperspectral imagingradiometers, such as AVIRIS.

A considerable number of ore deposits contain abund-ant sulphide minerals, of which the most common ispyrite (FeS2)' if their deposition was associated with theavailability of sulphide ions and reducing conditions.Oxidizing conditions, associated with weathering andin migrating groundwater, render iron-rich sulphidesunstable. They break down to sulphuric acid and anumber of ferric hydroxides and complex sulphates,which are both strongly coloured and possess crystal-fieldabsorptions in the VNIR (Fig. 1.7). As well as gossansformed over secondarily enriched sulphide deposits,simple weathering imparts these spectral features to soils.Moreover, the ferric compounds are often in the form of

Geological Applications of Image Data 237

mon plants burgeon as a-result of extremely favourablelevels of nutrients. This particularly is the case with sedi~mentary phosphate deposits and carbonatites containingabundant apatite (Ca5(PO4)3).

Potentially useful as remote sensing is, there are fewore deposits that owe their discovery to the use of imagesalone. In fact much of the literature on remote sensingin metal exploration focuses on features associated withwell~known deposits. Exploration more generally usesspecific mineralization models, such as that outlinedfor porphyry-style deposits in Section 8.3.2. These fusemany kinds of data that have a bearing on the model,either in the form of complex image products or withina geographical information system. The products of aGIS are so numerous and often have little immediatemeaning outside the exploration team that they arenot illustrated here. Instead Plate 9.6 gives an exampleof graphic combination of various data sets for visual

interpretation.The area is part of the English Lake District, with

its physiography shown by Plate 9.6(d). It is clearlydominated by dense vegetation. The generalized geologyemerges from Plate 9.6(a), and the gravity and magneticanomaly patterns are shown in Plate 9.6(b). The explora~tion model is orientated to hydrothermal tin, zinc andcopper mineralization associated with the granite intru-sions Plate 9.6(c) is a colour image combining stream-sediment Sn, Cu and Zn analyses as RGB. Where allthree elements have high values is signified by whiteand cream colours-these are "areas of potential interest.Experience suggests, however, that significant mineral-ization of this type is closely associated with granitecontacts, particularly in the roof zone. In Plate 9.6(e-,f) thefocus shifts to the vicinity of the geochemical anomalynear thegramte at left of centre on the regional images,where there is also a prominent magnetic anomalyaIida negative gravity anomaly that extends beyond the out-crop of the granite.

Plate 9.6(e) is an RGB image using Landsat TM dataas intensity, a constant saturation and gridded Sn valuesas hue, with magnetic contours overlaid. The magnetichigh probably relates to magnetite formed during con'"tact metamorphism at the roof of the buried granite.The high Sn values are spatially related to two indis-tinct linear features~possibly fracture zones related todeep-seated faults-'trending WSW-ENE and NW-SE.Plate 9.6(f) supplements this by combining direction-ally filtered TM data as intensity, gravity anomaly as hueand significant Sn and Bi anomalies as red and greenoverlays. The purple area indicates where the granite islurking at shallow depths. The conclusion is that tinanomalies here are spatially related to the buried graIiiteand arranged along deep-seated fractures, and so con-stitute an ideal targeUor detailed follow-up in this poorlyexposed area.

colloids that move in suspension to affect broad areas.Such anomalies are characterized by higher red reflect-ance than in green or blue, and can be identified usingratios of Landsat TM bands 3, 2 and 1. The crystal-fieldabsorptions in the VNIR express themselves in ratiosof TM bands 4 : 3 and 5 : 4, although the positioning ofLandsat MSS bands 4 and 3 is more appropriate for thesefeatures. It is worth noting that the acid conditions whichsulphide weathering creates increases the abundanceof hydrogen ionsJso weathered sulphide ores acceleratethe breakdown of silicates to clays.

Occasionally the geochemistry of hydrothermal fluids,particularly those that are highly alkaline, induces thesolution and reprecipitation of silica around ore deposits.Although quartz has no unique spectral features inthe reflected region, cryptocrystalline or opaline silicacontains abundant water. As well as the hi~ albedoof siliceous surfaces, cherty or opaline alteration mayexhibit absorption in TM band 5 on the flank of the1.4 11m water feature (Fig.I.B).

This suite of alteration features has proved extremelyuseful in targeting a variety of hydrothermal deposits,notably supergene-enriched, porphyry and hot-springepithermal types. Plate 6.4 shows dramatically how thecombination of all three alteration types near a porphyrycopper deposit express themselves, after using carefullyselected combinations of both reflected and thermallyemitted data. Plate 9.5 uses three Landsat TM band ratiosselected to express ferric minerals, clays and albedo tohighlight possible zones of hydrothermal alteration in acompletely unknown area.

The effect of fluid migration on rocks is nor restrictedto alteration and oxidation. If conditions are reducing toslightly oxidizing and very acid then iron in its ferrousstate and a number of other transition metals are soluble.They can be leached from rocks through which thefluids pass producing bleached zones. This commonlyis associated with the formation of secondarily enrichedsedimentary uranium and vanadium deposits. Formerlyiron-rich sandstones, either grey because of finely dividedpyrite or red to orange from their contained ferric minerals,become white. The anomalous- colour contrast therebyallows efficient detection of areas affected by potentiallymineralizing fluids of this kind.

The presence of vegetation does not necessatily pre-clude a search for mineralization using images alone.The high concentrations of metals, the peculiar soilEh and pH associated with many ore deposits, and theleaching of nutrients as a result of migration of acidfluids from sulphide oxidation often suppresses the properdevelopment of common plants. They become stunted,discoloured or simply fail to grow, thereby forming bar-ren areas. Such geobotanicalanomalies form yet anotherfocus for image processing in target generation. Anothertype of anomaly encountered occasionally is where com-

238 Chapter 9

9.3.2 Hydrocarbon exploration

Oil and natural gas fields form much larger targets thanmetal ore deposits, but for obvious reasons they arealways buried and -therefore hidden. Locating them hasto be precise, and relies on detailed interpretation ofseismic reflection data and, at the end of the day, a com-mitment to the drill. The focus on a particular areaon which to commit costly seismic surveys and drillingrelies almost entirely on play generation. A hydrocarbonplay is a combination of favourable geological circum-stances for the likely presence of oil or gas, conditionedby the economics involved in developing any finds in aparticular region. The geological aspects of a play can besummarized as the coincidence of a number of favour-able factors: the presence of an organic-rich source rock;its burial to a depth where local geothermal conditionspermit the maturation of organic compounds to oil andgas; a path whereby migration from source to reservoircan occur; the presence of rocks with sufficient porosityand permeability to act as a reservoir; the presence ofa suitable trap with an impermeable cap of s(jme kindwhere the hydrocarbons can accumulate and be pre-served. Previous drilling records in partly developedoil fields, together with stratigraphic and structuralinterpretation of seismic sections and exposed geologypermit the assessment of many of these factors. The hugeamount and diversity of data in well-known hydro-carbon basins, including those from existing fields, arenow handled and analysed routinely in what amount toa three-dimensional GIS. Remote sensing methods some-times complement such analyses, but assume a moreimportant role in unexplored frontier areas. That is themain focus here.

All significant hydrocarbon fields are found in or nearsedimentary basins that subsided sufficiently to permitthe deposition of thick and varied sedimentary sequences.There are many different tectonic situations where thesebasins might have formed, and most large basins arealready known. Without expensive seismic surveys, how-ever, the details of their architecture are sketchy. Thickaccumulations of sediments with densities and magneticsusceptibilities lower than crystalline basement give riseto negative gravity anomalies and magnetically 'smooth'zones. So a first approximation of their innards arisesfrom examination of gravity and magnetic data. Usinggriddeddata and derivatives from them in image form,as in Figs 8.11-8.15, helps locate variations in basin depth.Knowledge of subsidence rates and geothermal conditionshelps assess zones within the oil and gas maturation'windows'. Sometimes, gravity and magnetic data helpvisualize deep fracture zones that may have focusedsedimentary facies variations and fluid migration. Theradar altimetry of the oceans discussed in Section 8.2.4(Fig. 8.10) contains information on regional variations in

gravitational potential. So actual gravity measurementsare not always essential. Many offshore basins show upon derivatives of radar altimetry.

Analysis of surface exposures gives clues to the inter-nal structure of sedimentary basins. Chapters 4, 6 and 7show many examples of the structural and lithologicalinformation that can be won directly from remotelysensed images. Although this is unlikely to locate oil fieldsdirectly, it provides a basis for basin analysis and under-standing of the tectonic controls over basin evolution.Plates 4.3, 4.4, 6.5 and 7.1, and Figs 7.5,7.8 and 7.10 areparticularly instructive in this respect.

Natural oil and gas seepages, sometimes flaring due toignition by lightning, were the primary indicators of anumber of oil fields, both onshore and at sea. Every oilexplorationist hopes to make such an 'easy' discovery, forlittle remains other than to investigate by drilling. Thereis no such thing as a completely 'tight' trap, particularlywith respect to light hydrocarbons and gas. Every hydro-carbon field undergoes some leakage, or has leaked inthe past. Whether or not this reaches the surface, andhas a sensible effect if it does, is a matter of chance.Remote sensing for seep detection is consequently a topicin which considerable resources have been invested. Itmust be emphasized that evidence for seepage is rarely asignpost to 'drill here'. Light petroleum and gas migratealong the lines of greatest permeability, either a suitablesedimentary stratum or a zones of fracturing. Sometimesthese might be directly above a trap, but equally theymay be orientated laterally: Also, the signs may indicatepast leakage, that has exhausted the source which maynow contain only highly viscous tars. A good example ofthe latter are tar sands, from which the light fraction hasescaped, and which require mining and heat treatmentto win useful products.

The most obvious seepages contain liquid oil and tarryresidues from evaporation, and everyone knows thattar is black. Good examples are the La Brea tar ponds ofthe Ventura basin in Los Angeles, and the major Iranianfields were found from flaring seeps known for centuriesby local people. Dark signatures on images from thereflected region in prospective basins are always worthnoting. However, most seeps are far more subtle. Therehave been reports of a sort of physical blurring of minorphysiographic features above several major, on-shoreoil fields in unvegetated terrains. This has not yetbeen explained, but may be due to periodic upwelling offormation water charged with gas to create quicksandconditions. Of more use are the potential effects of suchfluids on the chemical and physical properties of thematerials through which they have passed.

Accumulations of hydrocarbons require extremelyreducing conditions, and the formation waters associatedwith them have low Eh. The passage of reducing fluidsthrough rocks containing oxidized iron minerals, such as

Geological Applications of Image Data 239

on overlying plant life. They may become stunted or dis-coloured, wholesale die-off may occur, and the timingof seasonal colour change in the canopy (Section 1.3.3,Fig. 1.17) may change, commonly resulting in prematuresenescence (Plate 9.7). In vegetated areas remote sensingprovides a wide range of opportunities to locate suchseep.related geobotanical anomalies.

Hydrocarbon seepage is not restricted to land areas,but it would be easy to dismiss as a focus for offshoreexploration because the sea floor is not usually accessibleto remote sensing. But consider for a moment what gasseepage beneath water might do. It forms rising plumesof bubbles. Bubbles do not rise passively, but inducea slow upward component of movement in the water:This is the principle of the air-lift dredge used in marinearchaeological excavations. Such plumes can draw cold,nutrient-rich bottom water to the sea surface. Of course,other phenomena such as tidal and current activityresult in upwellings, but these change position with theseasons. A seep-related upwelling should remain in thesame place. Theoretically therefore images of sea-surfacetemperature and colour associated With phytoplanktonblooms, taken over a period, should reveal fixed coldspots and planktonic blooms that are possibly relatedto seepage. As offshore interesfis in large hydrocarbonfields with commensurately large seeps, such an indirectexploration method can be conducted on a daily basisusing low-cost meteorological data With coarse resolu-tion. A number of major oil companies have conductedresearch along such lines under conditions of 'greatsecrecy. Related to it is the deployment of low-level air-borne surveys of aerosols at -the sea surface that resultfrom the bursting of small bubbles. The hydrocarbonchemistry of the sucked-up samples is monitored usinggas chromatography. However, growing knowledge thatmany areas of the sea floor are underlain at shallow depthsby sediments saturated With methane hydrates, formedby reaction between decay products from plankton aridcold, high-pressure water, means that many submarineseepages are not related to commercial oilfields.

hematite, induces the reduction of insoluble ferric iron(Fe3+) to ferrous (Fe2+) ions, which are soluble. One pos~sible effect of escaping hydrocarbons and the movementof formation water is bleaching, particularly of red bedsin the sequence. Not only iron can be mobilized in thisway, but many other metals too. So, seepage zones can beassociated with anomalous soil geochemistry. Geochem~ical soil surveys, for hydrocarbons as well as metals, canfigure in hydrocarbon exploration, where the methodsdescribed in Section 8.2.3 are appropriate.

One of the most controversial methods in hydrocarbonexploration centres on another likely phenomenon inhighly reducing environments. Low Eh can partly reducethe common ferric oxide and hydroxide colourants inmany sediments to black magnetite (F~O 4)' little imagina-tion is required to guess that seepages associated withsuch diagenetic magnetite may induce higp-frequencymagnetic anomalies. The controversy lies in whether ithappens at all, and if it does whether the anomalies canbe screened from others with different sources. In activeexploration areas the ground will be cluttered with allsorts of ironmongery that would produce very similaranomalies. That it does occur is partly confirmed byincreased magnetic susceptibility in cores from product-ive wells to depths as great as 1 km compared with thatin dry holes. This has been backed up by the discovery oftiny magnetite spherules in the cores. Their form encour-ages the idea that as well as reducing conditions some kindof biological activity is involved, of which more shortly.Conceivably, the diagenetic magnetite phenomenon addsmore value to the use of high-resolution aeromagneticsurveys in oil exploration, in which sophisticated spatial-frequency filtering and image enhancement may isolateseepage-related anomalies. As unconsolidated sedimentson the sea floor have more potential for the passage ofseeping fluids, and offshore exploration is dominatedby high-cost seismic surveys, related methodsiincludingremote measurements of magnetic susceptibility of thesea bed, are under development.

Far less controversial are methods related to theeffects of noxious seeping fluids on terrestrial plant life.Plants obtain their nutrients and water through com-plex symbioses of bacteria and fungi at their root tips.If these are damaged then the plants die, fail to thriveor show unusual features. Enhanced hydrocarbon con-tent in groundwater encourages the growth of anaerobicbacteria that reduce sulphate (SOi-) to sulphide (52-) ions.This is the source of hydrogen sulphide or 'sour gas' inmany oil fields, and its presence encourages the precipita-tion of heavy-metal sulphides carried by groundwateror seeping formation water. Sour-gas generation is thereason for the association of geochemicaf anomalieswith some seeps, and is possibly involved in diageneticmagnetite formation. It and the 'dumping' of heavy metalswreaks havoc on root-tip symbioses, with a range of effects

9.3.3 Water exploration

In the midst of a world economic depression, many ofthe talents of remote-sensing geologists are being shedby oil and mining companies, as demand and price oftheir products stagnate or fall. At the same time, in bothdeveloped and underdeveloped regions, the -problemsof water shortage-and decline-in its quality grow to dis-astrous proportions. A sage once commented that thefirst provision in any civilized society, after a means ofcollecting taxes, is that of a safe and dependable watersupply, yet this baSic need is under considerable threat.To a large extent geologists can assemble the informa-tionneeded to alleviate, if not solve these problems.

240 Chapter 9

rocks.

These thresholds ere often floored by thin mantlesof permeable alluvium so providing firm footings for theconstruction of dams or barrages. Thresholds are easilylocated

on stereoscopic aerial photographs or satelliteimages (Fig. 4.13). Upstream erosion always poses the

problem of siltation of artificial reservoirs, the usefullifetime of which may be reduced to only a few years inareas of high soil erosion. Although this might seemto suggest that surface dams are best avoided in suchareas, there is an interesting means of turning siltationto advantage. Provided the sediment fill behind a damhas good porosity and permeability, once completelysilted it can still retain useful water supplies, accessedby infiltration galleries built deliberately to be buried bysedimentation. Such artificial groundwater reserves havethe advantage over open water reservoirs of being lessprone to loss by evaporation in hot areas. The mainreason why transported sediments are unsuitable forthis secondary storage potential is the filling of theirpore spaces by fine grains or clays. If the supply of sedi-ment is free of fines and stream velocities are energeticenough to move sand and gravel, dams designed to illland produce from their eventual sedimentary fill are apromising option. The best sediments are those com-posed mainly of quartz particles. Remote sensing helpsevaluate potentially useful catchments for this newstrategy in terms of the bedrock exposed to erosion. Areasof granite or older sandstones should source sedimentswith good hydraulic properties. Rocks that easily rot toclays, such as basic igneouS-rocks, mudstones and fine-grained metamorphic rocks produce sediments with nosecondary supply potential.

Hydrogeological assessment of groundwater re-sources involves primarily the mapping of lithologiesarid structures to help asSess the subsurface dispositionof sedimentary aquifers. This is akin to basin analysis forhydrocarbons, with the difference that groundwatermigrates laterally and downwards. The conditions forsubsurface water accumulation, involving various kindsof trap, are familiar to most geologists. Images pro'videabundant information that can be blended with observa-tions on runoff versus infiltration potential in rechargeareas. Here the intention is to concentrate on specificadvantages offered by remote sensing.

Among the many sites for groundwater accumulationare those provided by subsurface dams. These includefaults that juxtapose aqUifers with aquitards, and igneousdykes with a crystalline nature that blocks flow wherethe dyke trends across the direction of subsurface flow.Both faults and dykes exhibit discontinuous linearfeatures at the surface that often are more easily seenand extrapolated on images (Figs 4.12, 4.24 and 7.9, andPlates 4.2 and 4.3) than on the ground. Radar is particu-larly useful in locating small dykes because it highlightsthe aligned"cornerreflectors formed by jointing (Fig. 7.9),

Water supply is served by two main options, outsideof desalination and other technological fixes. Thesecentre on surface water and that stored in rocks andsoils as groundwater. The use of remote sensing in hydro-logy and hydrogeology relates mainly to suggestingfavourable areas for development in terms of the relat-ive amounts of water available and various factors thatmight condition development. Indirectly, it has somebearing on water quality. An example would be the iden-tification of zones of hydrothermal alteration and beds ofmarine shales, both of which can contribute potentiallytoxic compounds to waters floWing over or through them.Screening for biological pollution would in most cases beserved through a GIS by data layers on habitations andland use. A geologically or hydrologically favourablesite may not necessarily be useful. To evaluate it in eco-nomic and social terms requires information on the typeand location of usage, as well as on various engineeringconsiderations. Again, a multifaceted GIS is required, ofwhich geological information from remote sensing andother spatial data would form a part.

The surface water regime in an area can be expressed bythe relationship between the four factors of precipitation,surface flow, evapotranspiration and infiltration. Onlywhen channelled orin lakes does surface water become apotential resource. Aside from losses by evapotranspira-tion and infiltration, the flow through a channel or enter-ing a lake depends on the area of upstream catchmenttogether With precipitation. The suitability of the water,in terms of sediment load, depends mainly on the slopeangles in the catchment and the density of vegetationcover that binds soil.

.Precipitation information can be derived from theuse of monthly averages of cloud-top temperature onmeteorological satellite' (metsat) images as a gauge ofthe likelihood of rain falling from any clouds, and by theuse of multi-temporal images shoWing snow cover andsnow melt rates. Evapotranspiration rates can be estim-'ated from surface temperafure averages derived frommetsats, and by models based on vegetation densityderived from its unique contrast in VNIR and red reflect-ance. Inffifration is a thornier problem and is discussedlater. Catchment area and slope estimation, together Withanalysis of drainage networks, are derivatives from OEMs,as covered by Section 8.2.4. Combined With remotelysensed estimates of vegetation cover, this allows thequalitative assessment of each catchment for likely levelsof turbidity resulting from sOil erosion.

The main input by geologists to studies of surfacewater resources focuses on the location and evaluationof sites for capturing flow and harvesting it for drinkingwater, irrigation or hydropower generation. Whereaslong earth barrages or bunds pond wate.in broad valleyson gently undulating terrain, the best sites are wherevalleys narrow as they pass through more resistant

Geological Applications of Image Data 241

distribution of vegetation in an arid, faulted are~, andclearly indicates areas favourable for development ofsubsurface supplies. Phreatophytes in fact consume largevolumes of groundwater. One estimate is that one squarekilometre of dense desert scrub transpires sufficientwater vapour in a year to supply all the needs of a villageof 1000 inhabitants.

9.4 Engineering applications

Geologists have long made important contributions tothe siting of major constructions, such as roads, dams,power plants and waste disposal sites. Much of this hascentred on aspects of safety and stability in relation tovarious kinds of slope failure, seismicity and flooding, aswell as to the engineering properties of foundations andsources of potential construction materials.

Slope stability and risk from landslips are most appro-priate to routing of roads and railways. Although thereis a variety of contributory factors to instability, the com-mon factor to all such hazards is variation in topographicslope. Stereoscopic images help us visualize $lope, butthese generally incorporate a degree of vertical exagger-ation. It requires painstaking use of photogrammetricinstruments to produce accurate estimates of slopes andslope profiles. A OEM makes it a simple matter to plottopographic profiles, to extract slope values and to high-light those so steep as to pose a hazard of failure.

In humid tropical areas, where thick soil and regolithpotentially pose problems of mudflows and arcuate slips,extremely steep slopes are commonly stabilized by densevegetation and the associated root mat. Vegetation cover,or the lack of it, is an important parameter in slope-stability assessment in soil-covered areas. Together withgradient data, multispectral means of vegetation classi-fication from images clearly forms an important input toplanning. In areas largely free of soil, vegetation is likelyto give little assistance to the engineer. The rock typeand associated structures govern the likelihood of rockfalls and landslips. The most risky combination is that ofinterbedded permeable and impermeable strata whichdip parallel to the topographic slope. Both gravity andlubrication by water along bedding surfaces conspireto encourage detachment and mass wasting. Generally,satellite imagery has too coarse resolution to permit detec-tion of such features, alt~ough multispectral data mayassist in delimiting lithologies with contrasted engineer-ing properties. Stereoscopy using aerial photographs isa decided advantage in locating potential hazards and,equally important, evidence of previous slope failure.

Steep slopes are not the only areas posing potentialhazard to heavy construction. Soft sediments, such aspeats and clay-rich rocks, are easily compacted as wateris expelled during loading. If the compaction varies

even when the dykes are buried by veneers of alluviumin arid areas (Fig. 7.17). Because dykes are frequentlymore resistant than their hosts, when a landscape becomesmantled by superficial soft sediments older dykes mayform corrugations at the interface, which pond ground-water or slow its flow.

Unconsolidated sediments, especially piedmont fans,alluvium, aeolian sand and beach deposits (Figs 4.42,4.43,4.45 and 4.51), provide limited supplies of ground-water that are easily accessible. In semiarid climates thewater table in such sediments may be deeper than theextent of the potential host, so estimates of thickness areneeded. Apart from geophysical surveys at the surface,the only means of visualizing this parameter is the use ofradar images. Provided soil moisture is low, radar canpenetrate as deep as 4 or 5 m to give a signal from sub-surface features (Section 7.1.1). Figure 7.17;~a) shows adramatic example from the Red Sea Hills of Sudan, whereotherwise completely hidden intricacies of a drainagesystem buried by alluvium indicate the most likely sitesfor obtaining water from the main distributaries of the olddrainage. Conversely, in areas of variable soil moisture-a good guide to deeper groundwater-radar scatterometrygives a guide to moisture content.

Whereas faults in sedimentary sequences provideopportunities for subsurface dams, in areas domin-ated by impermeable crystalline rocks the shatteringassociated with faulting provides the only opportunityfor infiltration and eventual recovery of groundwater.Remote sensing is pre-eminent in delineating such zones.Of special interest is the extrapolation of large fracturesystems from elevated areas of high precipitation tolower, drier ground. The fractures potentially can trans-fer water laterally over tens to hundreds of kilometres.Because the secondary porosity involved in fracturingcan become sealed by fines, evaluating fracture zonesrequires consideration of the rock types involved andthe nature of weathering. In all climates micaceous rocksare easily sealed. In humid areas, or those once subjectto high chemical weathering, feldspars in granitic andgneissic rocks, and ferromagnesian minerall) in moremafic rocks break down to clays that clog the potentialpathways. In dry climates many granites break down bymechanical means to form gravely regoliths and fracturezones still retain their permeability. In all climates, inertquartzites and soluble carbonates are the ideal hosts forfracture-controlled ~quifers. As discussed in Section 4.2,most images carry clues to the general distribution oflithologies, and multispectral data can help further dis-tinguish rock types.

Because phreatophyte plants can put down roots todepths in excess of 100 m to reach the water table, thepresence of vegetation in arid areas is a sure guide tothe presence of water, although not necessarily to itsquality. Plate 9.8 shows an example of the discontinuous

242 Chapter 9

laterally, the differential subsidence can cause failure.Former lake beds, glacial tills and flood-plain alluviumare replete with such hazards,. yet they are often signifiedon images by the distinctive landforms associated withtheir formation (Figs 4.43, 4.47 and 4.50) that are noteasily detectable on the ground.

Section 4.4.1 illustrated several of the features associ-ated with active faults, most of which require the fineresolution of aerial surveys. Now it would be a pre~ dimgeologist who failed to recognize the hazards associatedwith a project entering seismically active terrain, becausevirtually all sizeable earthquakes above magnitude 4 arereadily and systematically located by a global networkof seismographs. The crucial issue is to relate themto fault zones, although destructive earthquakes alongsubduction zones have no surface manifestation, hav-ing deep epicentres A first call would be to delineate allpotential faults that might fail in a seismically risky area,using aerial photographs. Most damage in seismicallyactive areas, however, occurs not in high-relief areasof hard rocks, where faults are easily spotted, but inareas of unconsolidated sediment that behave plasticallyduring earthquakes, even when the seismic event is fardistant. The catastrophes of Mexico City (1988) and Turkey(1999) bear grim witness to this. Areas of silts and clays,often associated with former lake beds and estuarinedeposition, need to be delimited. This is no easy task, asall low-lying flat areas are suspect. Careful searches formorphological features, such as strand lines (Fig. 4.51),that indicate such sediments must be conducted.

Equally, it would be an engineer verging on the psy-chotic who wilfully built a bridge, a dam or a powerplant straddling a fault of any age, for the loading itselfcould induce movement. These criminal errors do occur,however. Old lines of weakness are often exploitedby modem earth movements, such as those related topost-glacial unloading. Displacements of centimetrescould cause catastrophic events in structures containingmoving parts with low tolerances, such as the controlrods ina nuclear plant. However, it is not always easy torecognize old fractures on the ground or even on images,because their active control over landforms is longextinct. In areas of low relief and" intensive agriculturethe patchwork of fields with different crops producesan efficient camouflage for muted topographic featuresthat might indicate faults (Fig. 4.49). Various image pro-cessing methods are available to reduce this culturalcamouflage and reveal previously unsuspected featuresof tectonics (Fig. 9.5). Radar interferometry helps expressthe kinds of small-scale ground movement associatedwith earthquakes (Plate 9.9d). Although this is of nouse in mitigating effects of any particular earthquake,by showing the breadth subject to ground movements,research using InSAR might serve to indicate corridorsof risk along known fault lines.

Fig. 9.5 Enhanced early spring LandsatTM band 5 image ofpart of an intensely farmed area in southern England. At thistime of year grasses and winter cereals have similar moisturecontents and the field patterns that generally obscuretopographic features (Fig. 4.48) are uniquely removed to revealmany major fault zones that are unreported by conventionalgeological mapping.

Whether they move or not, fractures form pathwaysfor the large-scale movement of fluids (Sections 9.3.2and 9.3.3). This is the last thing that is desired in a waste-disposal site, particularly one for long-lived nucleardebris. Besides using images to screen areas with faultsfrom consideration, many of the techniques deployedfor water exploration are useful in selecting waste-disposal sites. Apart from methane generation producedby fermentation of organic waste that vents upwards andlaterally, leakage follows exactly the same rules as themovement of groundwater, and hydrogeological inter-pretation of image data is essential in siting.

One hazard that engages the undivided attention ofinsurance actuaries in urban areas is that of ground sub-sidence, which can cause minor though expensive damageto buildings and services. It occurs most frequently intwo main ways, through subsurface excavations, suchas coal mining or solution mining of salt, or becauseof the volume reduction of aquifers when water extrac-tion exceeds rates of groundwater recharge. Movements

Geological Applications of Image Data 243

because they are minoi features, often smaller than asystem's resolution, and partly because their signaturesare easily overlooked. A large body of detailed evidencesuggests that the slow rise of magma before a volcanicevent induces slight changes in both the gravitationalfield around volcanoes and their topography, as wellas producing swarms of minor earth tremors. petectingthese and other signs by continual field studies is in itsinfancy and focused on famous volcanoes such as Etna.Interferometric radar provides a means of preciselydocumenting volcano shape changes (Plate 9.9c) thatpotentially might become a means of early warning. TheERS-l and -2 orbiting radar imaging systems are idealfor monitoring of that kind, but such is the complexityof InSAR processing that few institutions yet have thecapacity for keeping watch over one volcano, let alone allthose that pose real threats.

9.5 Geochemical hazards

are rarely more than a few centimetres, although theircumulative effects can build to metres. Consequently, theyare difficult to detect remotely, although the damageis all too easily witnessed. The ability of interferometricradar imaging to create both highly precise OEMs andimages of elevation change opens up an entirely newfield of subsidence detection and policing, lor the victimsare otherwise very rarely able to prove who, if anyone,is responsible. Plate 9.9(a,b) shows InSAR results forsubsidence related to both underground mining andground water abstraction.

Although not ~trictly speaking an issue for which thereare engineering remedies, volcanism and the associatedhazards of debris flows form an extreme risk in volcan-ically active regions. There is absolutely no problem inrecognizing volcanoes, but the most hazardous are thosethat remain dormant for long periods be~j;';en majoreruptions. Soils formed from lavas and ash are amongthe most fertile there are, and farmers are drawn to tillthem despite the risk, particularly-where good land is ata premium. Communities grow, and in the Andes someof the most hazardous areas on the planet support highpopulations on the lower slopes of volcanoes. Eruptionsof lavas pose little hazard to life,- simply because theyare quite obvious, and travel relatively slowly, althoughflows will destroy everything in their path. Having tem-peratures in excess of 1000°C, active flows and lava lakesemit radiation at wavelengths as short as the visible range.They closely approximate black bodies, and the radiancein several wavebands allows temperature estimationusing the Stefan-Boltzmann Law. Detecting eruptionsby remote sensing is easy, and although seemingly altru-istic is akin to telling the grandmothers on the slopes of avolcano how to suck eggs. It is an excuse for academicstudy of volcanism-perfectly justifiable but an excusenonetheless. Detecting signs that a volcano is about to'go off' is altogether more urgent, for two main reasons.First,. high-altitude dormant cones are often cappedby snow and ice, which if melted-even at their base-may generate avalanches and huge mudflows known aslahars. These travel at speeds far in excess of lavas andaccount for the majority of fatalities. The second are evenmore rapid incandescent debris flows, formed whenvolcanoes being inflated by viscous magma fail side-ways instead of vertically through a vent to produce anash plume. Figure 4.41 (b) shows features typical of sucha catastrophic event, as do the film sequences of the fateof Mount St Helens in May 1980. The ruins of Pompeiiand Herculaneum near Vesuvius be~r witness to theawful cons~quences for anyone unfortunat~ enough tolive close to volcanoes with such tendencies.

Signs of pending eruption, of whatever kind, aresubtle. There may be an incr~sein fumarole activity,perhaps heating at the site of a future eruption. Neitherof these is easily monitored by remote sensing,. partly

Statistical analysis of medical records, particularly thoseof morbidity and mortality (sickness and cause of death),are beginning to reveal disturbing evidence for healthblackspots that correlate with a variety of environmentalparameters. These are in addition to the known associa-tions of mental retardation with lead pollution, leukaemiaand other cancers with nuclear plants, a variety ofhorribleafflictions with mercury and cadmium pollution andbirth defects clustered arouftd sites for hazardous wastedisposal, and seem to stem from pur~ly natural environ-mental variations.

Some environmental hazards have been known for along time, such as the associations of goitre and fluorosiswith areas deficient in iodine and with abnormally highfluorine contents in soil, respectively. However, new rela-tionships are appearing regularly. Some cancers are morefrequent over certain uranium-rich granites and shales,and have been traced to radon-gas emissions from thesoil accumulating in houses. There is some evidence forincreased incidence of multiple sclerosis in fluorine-richareas. Men living in areas supplied with water fromlimestones are more at risk from heart disease than thosein soft-water areas. High contents of organic pigment fromp~at in water correlates with increased risk of stomachcancer. To these few examples of human risk from geo-chemical anomalies can be added incidence of ill thrift inlivestock from excess molybdenum, copper deficiencyand lead anomalies. Whereas the associations are known,currently the g~ochemical data to assess variable risk arelacking for most parts of the world.

The most useful information would be results fromgeochemical surveys of soil, as carried out routinelybut in restricted areas for metal exploration. 'This isextremely expensive (Table 9.1), and unlikely ever to be

244 Chapter 9

remain in the commerceial domain at a price beyond thepurses of most of those who would benefit. from them. Itremains

to be seen whether the proliferation of orbitingsystems will flood the market and force down prices, orif

a kind of cartel among the operating companies willmaintain them artificially.

Further reading

These

references are but a few. of the many relating toapplications of remote sensing and GIS in geology. The

Proceedings of the Thematic Conferences on Remote Sensingfor Exploration Geology, sponsored by the Environmental

Research ~stitut.e of Michigan (ERIM), include manypapers in t~s .very general area, Occasionally, trade

journals, such as the Mining Magazine and the Oil and GasJournal will contain review articles on this topic. Journals,

such as Economic Geology and the Bulletin of the AmericanAssociation of Petroleum Geology, often carry papers onremote

sensing in exploration, but so do most EarthScience journals. A good way to keep abreast of papers

on remote sensing in geology is using the search optionsin the Science Citation Index, or in the electronic journalslists

of various publishers, using appropriate searchterms and keywords.

conducted

globally, At much lower cost, and availablenow for many developed countries, are data from drain-ag~

surveys (Section 8.2.3), In image form (Plate 9.6) theyprovide an extremely efficient means of screening for

geochemical hazards, as they are for exploration, andcould be incorporated easily into health-orientated GISsto

be assessed with the many other variables. Risk fromnaturally high radioactivity is readily assessed by refer-ence

to images of gamma-ray emissions derived fromairborne radiometry (Plate 8.4).

In the absence of analytical data, the known con-centration or depletion of some of the geochemicalculprits

with specific rock types and with relativelyunusual processes, such as those involved in mineral-ization, means that geological maps are a rough guideto

potential problems. Maps derived from images are insome cases even more useful in this regard)han thosecompiled

by more conventional means. The discussionin Section 9.3.1 on the signatures of hydrothermallyaltered

areas and those where sttlphides have beenoxidized to products including sulphuric acid shows

that remote sensing has a unique contribution in thisrespect~ Hydrothermal and other types of mineraliza-

tion invplving pervasive fluids boost the levels of manyhazardous elements in rocks and their associated soils.The

natural release of acids takes some of these elementsinto solution, further dispersing them in the naturalenvironment.

There is one important point of an ethical nature,which is often overlooked in assessment of hazards ofwhatever kind. It may well be possible to establish a pre-viously unsuspected risk. The problem lies in informingthe

people exposed to it. If there is no remedy, other thanmigration at great cost and carrying perhaps yet otherrisks,

should they be told?A final point of ethical. content is perhaps obvious, but

rarely stated. As well as intelligence apalysts, industrialand academic geologists in the developed countries areprivy,

through remote sensing and digital image pro-cessing, to a great deal of strategic economic information

about less-fortunate areas of the world. Often they areable to learn more, about resources and terrain than

their counterparts in the so-called 'Third World' becauseof their technological advantage. In an ideal world, such

information shottld be transmitted on acompletelyequit-able basis, but frequently it is not. The simplest recourseis

the spread of the enabling technology of remote sens-ing, and costs of hardware and ~oftware to achieve whatis

outlined in this book have now been reduced to levelsthat are affordable by any government. The question ofaffordability of data, for a continuous supply is neededto put theory into practice, isa different matter. WhereasLandsat-7 is back in the domain of public access at lowcost, and data from the EOS programme will be free tobona:.fide researchers, most data, both existing and planned,

Abrams, M.J., Brpwn, D., Lepley, L. & Sadowski, R. (1983)Remote sensing for copper deposits in Southern Arizona.Economic Geology 78, 591-~.

Abrams, M.J., Conel, J.E. & Lang, H.R: (1985) The Joint NASAlGeosat Test Case Study, 2 Volumes. American Association ofPetroleum Geologists, Tulsa, Oklahoma.

Abrams, M., Bianchi, R. & Pieri, D. (1996) Revised mapping oflava flows on Mount Etna, Sicily. Photogrammetric Engineeringand Remote Sensing 62, 1353-1359.

Abrams, M.J., Ashley, R.P., Rowan,L.C., Goetz, A. & Kahle, A.B.(1977) Mapping of hydrothermal alteration in the Cupritemining district, Nevada, using aircraft scanner linages for thespectral region O.46~2.36 ~. Geology 5, 713-718.

Allum, J.A.E. (1981) Remote sensing in mineral exploration-case histories. Geoscience Canada 8, 87-92.

Alwash, M:A. & Zilger,J. (1994) Remote sensing-based geolog-ical mapping of the area west of AI Madinah, Saudi Arabia.International Journal of Remote Sensing 14, 163-172.

Bailey, G.B. & Anderson, P.D. (1982) Applications of Landsatlinagery to problems of petr<;>leum exploration in QaidamBasin, China. Bulletin, American Association of PetroleumGeologists 66, 1348-1354.

Baker, M.C.W. & Baldwin, J.A. (1981) Application of Landsatmultispectral classification for iocating gosans in North Chile.In: Remote Sensing in Geological and Terrain Studies (eds J.A.Allan"& M. Bradshaw), pp. 25-33. Remote Sensing Society,London. .

Berne, S.M. (1988) A regional tectonic study of NE and E Africaand its implication for mineral exploration: a synoptic viewfrom satellite i~agery. Proceedings, IGARSS '88 Symposium,Edinburgh, pp. 363-366. ESA SP-284, European SpaceAgency, Paris:

Geological Applications of Image Data 245

Hills of NE Africa. Asslfciation of Geoscientists In Development(AGID) News 67/68, 17-20.

Drury, S.A. & Berhe, S.M. (1993) Accretion tectonics in northernEritrea revealed by remotely sensed imagery. GeologicalMagazine 130, 177-190.

EI Rakaiby, M.L. (1995) The use of enhanced Landsat TMimage in the characterization of uraniferous granitic rocks inthe Central Eastern Desert of Egypt. International Journal ofRemote Sensing 16, 1063-1074.

Ellyett, C.D. & Pratt, D.A. (1975) A Review of the PotentialApplications of Remote Sensing Techniques to HydrogeologicalStudies in Australia. Technical Paper 13, Australian WaterResearch Council, Canberra.

Farrand, W.H. (1997) Identification and mapping of ferric oxideand oxyhydroxide minerals in imaging spectrometer dataof Summitville, Colorado, USA, and the surrounding SanJuan Mountains. International Journal of Remote Sensing 18,1543-1552.

Feder, A,F. (1986) Integrated interpretation for exploration.Oil and Gas Journal May, 180-184.

Feldman, S.C. & Taranik, J. V. (1988) Comparison of techniquesfor discriminating hydrothermal alteration minerals withairborne imaging spectrometer data. Remote Sensing of Environ-

ment 24, 67-83.Fookes, P.G., Sweeney, M., Manby, C.N.D. & Martin, R.P. (1985)

Geological and geotechnical engineering aspects of low-costroads in mountainous terrain. Engineering Geology 21, 1-152.

Francis, P. W. & De Silva, S.L. (1989) Application of the LandsatThematic Mapper to the identification of potentially activevolcanoes in the Central Andes. Remote Sensing of Environ-ment 28, 245-255.

Francis,P.W. & Rothery, D.A. (1987) Using the Landsat ThematicMapper to detect and monitor active volcanoes: an examplefrom Lascar volcano, northern Chile. Geology 15; 614-617.

Fraser, A., Huggins, P., Rees, J. & Cleverly, P. (1997) A satelliteremote sensing technique for geological structure horizonmapping. International Journal of Remote Sensing 18, 1607-1616.

Goosens, M.A. & Kroonenberg, S.B. (1994) Spectral discrimina-tion of contact metamorphic zones and its potential formineral exploration, Province of Salamance, Spain. RemoteSensing of Environment 49, 331-344.

Gupta, R.K. & Badarinath, K.V.S. (1993) Volcano monitoringusing remote sensing data. International Journal of Remote

Sensing 14, 2907-2918.Gupta, R.P. & Bodechtel, J. (1982) Geotechnical applications

of Landsat image analysis of Bhakra dam reservoir, India.Remote Sensing of Environment 12, 3-13.

Gupta, R.P. & Joshi, B.C. (1990) Landslide hazard zoning usingthe GIS approach-a case from the Ramganga catchment,.Himalayas. Engineering Geology 28, 119-131.

Halbouty, M. T. (1976) Application of Landsat imagery to petro-leum and mineral exploration. Bulletin, American Associationof Petroleum Geologists 60, 745-793.

Halbouty, M.T. (1980) Geological significance of Landsat datafor 15 giant oil and gas fields. Bulletin, American Association ojPetroleum Geologists 64, 8-36.

Harding, A.E. & Forrest, M.D. (1989) Analysis of multiplegeological data sets from the English Lake District. lEEETransactions, Geoscience Remote Sensing GE-27,732-739.

Harris, J.R., Renez, A.N., Ballantyne, B. & Sheridan, C. (1998)Mapping altered rocks using Landsat TM and lithogeochem-ical data: Sulphuret&-Bmcejack Lake district, British Columbia,

,

Billings, W.P. (1950) Vegetation and plant growth as affected bychemically altered rocks in the Western Great Basin. Ecology

30,62-74.Birnie, R.W. & Francica,J.R. (1981) Remote detection of geobotan-

ical anomalies related to porphyry copper mineralization.Economic Geology 76, 637-647.

Bolviken, B., Honey, F., Levine, S.R., Lyon, R.J.P. & Prelat, A.(1977) Detection of naturally heavy-metal-poisoned areas byLandsat-1 digital data. Journal of Geochemistry and Exploration

8,457-471.Bushara, M.N. (1996) Neogene geological structures using

enhanced Thematic Mapper over eastern Zagros Mountains.International Journal of Remote Sensing 17,303-313.

Chang, S.H. & Collins, W. (1983) Confirmation of the airbornebiogeophysical mineral exploration technique using laborat-ory methods. Economic Geology 78, 723-736.

Cole, M.M. (1982) Integrated use of remote sensing imagery andgeobotany in mineral exploration. Transactions of the GeologicalSociety of South Africa 85, 13-28.

Collins, W., Chang, S.H. & Kuo, J.T. (1981) Detection of HiddenMineral Deposits by Airborne Spectral Analysis of Forest Canopies.NASA Contract NSG-5222, Final Report.

Collins, W., Chang, S.H., Raines, G., Channey, F. & Ashley, R.(1983) Airborne biogeochemical mapping of hidden mineraldeposits. Economic Geology 78, 737-749.

Conradsen, K. & Harpoth, O. (1984) Use of Landsat multispec-tral scanner data for detection and reconnaissance mappingof iron oxide staining in mineral exploration, central EastGreenland. Economic Geology 79, 1229-1244.

Crosta, A.P., Sabine, C. & Taranik, J.V. (1998) Hydrothermalalteration mapping at Bodie, California, using A VIRIS hyper-spectral data. Remote Sensing of Environment 65, 309-319.

Darch,J.P. & Barber,J. (1983) Multi-temporal remote sensing ofa geobotanical anomaly. Economic Geology 78, 770-782.

Das, J.D., Saraf, A.K. & Jain, A.K. (1995) Fault tectonics of theShillong Plateau and adjoining regions, north-east India usingremote sensing data. International Journal of Remote Sensing

16,1633-1646.Dennis, A.M., Harris, A.J.L., Rothery, D.A., Francis, P.W. &

Carlton, R.W. (1998) Satellite observations of the April 1993eruption of Lascar volcano. International Journal of RemoteSensing 19, 301-321.

Deroin, J .-P., Motti, E. & Simonin, A. (1998) A comparison of thepotential for using optical and SAR data for geological mappingin an arid region: the Atar site, Western Sahara, Mauritania.International Journal of Remote Sensing 19, 1115-1132.

De Souza Filho, C.R. & Drury, S.A. (1998b) Evaluation of JERS-1(Fuyo-1) OPS and Landsat:I'M images for mapping of gneissicrocks in arid areas. International Journal of Remote Sensing 19,3569-3594.

Deutch, M. & Estes, J .E. (1980) Landsat detection of oil fromnatural seeps. Photogrammetric Engineering and Remote Sensing

46,1313-1322.Donovan, T.J., Forgey, R.L. & Roberts, A.A. (1979) Aeromagnetic

detection of diagenetic magnetite over oil fields. Bulletin,American Association of Petroleum Geologists 63, 245-248.

Drury, S.A. (1990) Satellite aid. New Scientist 126, SO-53.Drury, S.A. (1990) SPOT image data as an aid to structural

mapping in the southern Aravalli Hills of Rajasthan, India.

Geological Magazine 127,.195-207.Drury, S.A. & Berhe, S.M. (1991) Remote sensing in ground-

water exploration in arid regions: examples from the Red Sea

246 Chapter 9

Canada. Photogrammetric Engineering and Remote Sensing 64,309-322.

Horler, D.N.H., Barber, J. & Barringer, A.R. (1980) Effects ofheavy metals on the absorbance and reflectance spectra ofplants. International Journal of Remote Sensing 1, 121-136.

Huntington, J.F. & Green, A.A. (1988) Recent advances andpractical considerations in remote sensing applied to goldexploration. In: Bicentennial Gold 88, Melbourne.

Irvine, J.M., Evers, T .K., Smyre, J.L., Huff, D., King, A.L., Stahl, G.& Odenweller,J. (1997) The detection and mapping of buriedwaste. International Journal of Remote Sensing 18, 1583-1596.

Jansma, P.E. & Lang, H.R. (1996) Application of spectral strati-graphy to Upper Cretaceous and Tertiary rocks in southernMexico: Tertiary graben controls on volcanism. Photo-grammetric Engineering and Remote Sensing 62, 1371-1378.

Jutz, S.L. & Chorowicz, J. (1993) Geological mapping and detec-tion of oblique extensional structures in the Kenyan Rift Valleywith SPOT/Landsat TM data merge. International Journal ofRemote Sensing 14, 1677-1688.

Keighley, J.R., Lynn, W.W. & Nelson, K.R. (1980) Use ofLandsat images in tin exploratioR, Brazil. Proceedings of the14th Symposium on Remote Sensing of Environment, San Jose,

pp.341-343.Kenea, N.H. (1997) Improved geological mapping using

Landsat TM data, Southern Red Sea Hills, Sudan: PC andHIS decorrelation stretching. International Journal of RemoteSensing 18, 1233-1244.

Kezheng, B. (1980) The application of remote sensing tech-niques to prospecting for metal deposits. Proceedings of the14th Symposium on Remote Sensing of Environment, San Jose,

pp.135-'-.142.Kierein- Young, K.S. (1997) The integration of optical and radar

data to characterize mineralogy and morphology of surfacesin Death Valley, California, USA. International Journal ofRemote Sensing 18,1517-1542.

Koch, M. & Mather, P.M. (1997) Lineament mapping forgroundwater resource assessment: a comparison of digitalSynthetic Aperture Radar (SAR) imagery and stereoscopicLarge Format Camera (LFG photographs in the Red SeaHills, Sudan. International Journal of Remote Sensing 18,1465-1482.

Krishnamurthy,. J. (1998) The evaluation of digitally enhancedIndian Remote Sensing Satellite (IRS) data for lithological andstructural mapping. International Journal of Remote Sensing 18,3409-3437.

Kruse, F .A. (1988) Use of airborne imaging spectrometer data tomap minerals associated with hydrothermally altered rocksin Northern Grapevine Mountains, Nevada and California.Remote Sensing of Environment 24, 31-52.

Kruse, F.A., Kierein-Young, K.S. & Boardman, J.W. (1990)Mineral mapping at Cuprite, Nevada with a 63-channel imag-ing spectrometer. Photogrammetric Engineering and RemoteSensing 56, 83-92.

Labovitz, M.L., Masuoka, E.J., Bell, R., Siegrist, A. W. & Nelson,R.F; (1983) The application of remote sensing to geobotanicalexploration for metal sulphides-results from the 1980 fieldseason at Mineral, Virginia. Economic Geology 78, 750-760.

Labovitz, M.L., Masuoka, E.J., Bell, R., Nelson, R.F., Latsen, E.A.,Hooker, L.K. & Troensegaard, K.W. (1985) Experimentalevidence for spring and autumn windows for the detectionof geobotanical anomalies through the remote sensing ofoverlying vegetation. International Journal of Remote Sensing 6,195-216.

Lattman,L.H. & Parizek, R.a. (1964) Relationship between frac-ture traces and the occurrence of groundwater in carbonaterocks. Journal of Hydrology 2, 73-91.

Legg, C.A. (1990) Remote Sensing, an Operational Technology forthe Mining and Petroleum Industries. Institute of Mining andMetallurgy, London.

Liggett, M.A. & Childs, J.F. (1976) An application of satelliteimagery to mineral exploration. Proceedings of the 1st AnnualWilliam T. Pecora Memorial Symposium, Sioux Falls, SouthDakota, pp. 253-270.

Loughlin, W.P. (1990) Geological exploration in the westernUnited States by use of airborne scanner imagery. In: RemoteSensing, an Operational Technology for the Mining and PetroleumIndustries (ed. C.A. Legg), pp. 223-241. Institute of Miningand Metallurgy, London.

MacDonald, H.C., Clark, B.C.., Taylor, D.S., Rainwater, G.L.,Kaupp, V.H; & Waite, W.P. (1989) Assessment of Shuttle ImagingRadar and Landsat.Imagery for Ground-water Exploration in AridEnvironments. Miscellaneous Paper EL-79-6, U.S. Army Corpsof Engineers, Washington DC.

Marsh, S.E. & McKeon, J.B. (1983) Integrated analysis of high-resolution field and airborne spectroradiometer data foralteration mapping. Economic Geology 78, 618-632.

Massonnet, D., Briole, P. & Arnaud, A. (1995) Deflation ofMount Etna monitored by spaceborne radar interferometry.Nature 375, 567-570.

Memmi, J.M. & Pride, D.E. (1997) The application of DiamondExploration Geoscientific Information System (DEGIS) forintegrated diamond exploration in the north-central UnitedStates of America. International Journal of Remote Sensing 18,1439-1464.

Milton, N.M., Collins, W., Chang, S.-H. & Schmidt, R.G. (1983)Remote detection of metal anomalies on Pilot Mountain,Randolph County, North Carolina. Economic Geology 78,605-617.

Missalati, A.,Prelat, A.E. & Lyon,R.J.P. (1979) Simultaneous useof geological, geophysical and Landsat digital data in uraniumexploration. Remote Sensing of Environment 8, 189-210.

Murphy, R.J. (1995) Mapping of jasperoid in the CedarMountains, Utah, USA, using image spectrometer data.International Journal of Remote Sensing 16, 1021-1041.

Newton, A.R. & Boyle, T.P. (1993) Discriminating rock and sur-face types With multispectral satellite data in the Richtersveld,Nw cape Province, South Africa. International Journal of RemoteSensing 14, 943-959. .

Norman, J.W. (1980) Causes of some old crustal failure zonesinterpreted from Landsat images and their significance inregional mineral exploration. Transactions of the Institute Miningand Metallurgy (Section B: Applied Earth Science) 89, B63-B72.

Offield, T.W. (1976) Remote sensing in uranium exploration. In:Exploration of Uranium Ore Deposits, pp. 7317744. InternationalAtomic Energy Agency, Vienna.

Offield, T.W., Abbott, E.A., Gillespie, A.R. & Loguercio, S.O.(1977) Structural mapping on enhanced Landsat images ofsouthern Brazil: tectonic control of mineralization and specu-lations on metallogeny. Geophysics 42, 482-500.

Oppenheimer, C (1998) Volcanological applications of meteoro-logical satellites. International Journal of Remote Sensing 19,2829-2864.

Oppenheimer, C., Rothery, D.A., Pieri, D.C., Abrams, M.J.& Carrere, V. (1993) Analysis of Airborne Visible/InfraredImaging Spectrometer (A VIRIS) data of volcanic hot spots.International Journal of Remote Sensing 14, 2919-2934.

Geological Applications of Image Data 247

Rowan, L.C., Wetlaufer, ~H., Goetz, A.F.H., Billingsley,F.C. &Stewart, J.H. (1974) Discrimination of rock types and detec-tion of hydrothermally altered areas in South-central Nevadaby the use of computer-enhanced ERTSimages. U.S. GeologicalSurvey Professional Paper 883.

Sabins, F.F. (1999) Remote sensing for mineral exploration. OreGeology Reviews 14, 157-183.

Saraf, A.K., Cracknell, A.P. & MacManus, J. (1989) Geobotanicalapplications of airborne Thematic Mapper data in Sutherland,NW Scotland. International Journal of Remote Sensing 10, 545-555.

Schmidt, R.G. (1976) Exploration for porphyry copper depositsin Pakistan using digital processing of Landsat-1 data. U. S.Geological Survey Journal of Research 4, 27-34.

Schmugge, J.T. (1983) Remote sensing of soil; moisture: recentadvances. IEEE Transactions, Geoscience Remote Sensing GE-21, 336-344.

Segal, D.B. (1983) Use of Landsat multispectral scanner dataforthe definition of limonitic exposures in heavily vegetatedareas. Economic Geology 78, 711-722.

Short, N.M; & Blair, R.W., Jr (1986) Geomorphology from Space.NASA SP-486, U.S. Government Printing Office, Washington,DC.

Siegal, B.S. & Goetz, A.F.H. (1977) Effect of vegetation on rockand soil type discrimination. Photogrammetric Engineering andRemote Sensing 43, 191-196.

Simpson, C.J. (1978) Landsat: developing techniques and applica-tions in mineral and petroleum exploration. BMR Journal ofAustralian Geological Geophysics 3,181-191.

Spatz, D.M. (1997) Remote sensing characteristics of thesediment- and volcanic-hosted precious metal systems: imageselection for explorations and development. InternationalJournal of Remote Sensing 18, 1413-1438.

Vincent, R.K. (1977) Uranium exploration with computer pro-cessed Landsat data. Geophysics 42, 536-541.

Vincent, R.K. (1994) Remote sensing for solid waste landfillsand hazardous waste sites. Photogramlrretric Engineering andRemote Sensing 60, 979-982.

Vogel, A. (1985) The use of space technology in earthquakehazard assessment. Proceedings of the 18th International Sym-posium on Remote Sensing Environment, Paris.

Wang, Q.M., Nishidai, T. & Coward, M.P. (1992) The TarimBasin, NW China: formation and aspects of petroleum geology.Journal of Petroleum Geology 15, 5-34,

Whitney, G., Abrams, MJ. & Goetz, A.F.H. (1983) Mineral dis-crimination using a portable ratio-determining radiometer.Economic Geology 78, 688-698.

Williams, R.S. (ed.) (1983) Geological applications. In: Manual ofRemote Sensing (ed. R.N. Colwell), 2nd edn, pp. 1667-1953.American Society of Photogrammetry, Falls Church, Virginia.

Windeler, D.S. (1993) Garnet-pyroxene alteration mapping inthe Ludwig skarn(Y erington, Nevada) with Geoscan airbornemultispectral data. Photogrammetric Engineering and RemoteSensing 59,1277-1286.

Paradella, W.R., Bigneli, P.A., Veneziani, P., Pietsch, R.W. &Toutin, T. (1997) Airborne and spaceborne Synthetic ApertureRadar (SAR) integration with Landsat TM and gamma-rayspectrometry for geological mapping in a tropical rain forestenvironment, the Carajas Mineral Province, Brazil. Inter-national Journal of Remote Sensing 18, 1483-1502.

Peltzer, G. & Rosen, P. (1995) Surface displacement of the17 May 1993 Eureka Valley, California, earthquake observedby SAR interferometry. Science 268, 1333-1336.

Peters, D.C. (1983) Use of airborne multispectral scanner datato map alteration related to roll-front uranium migration.Economic Geology 78, 641-653.

Plant, J .A., Jones, D.G., Brown, G.C., Colman, T .B., Cornwell,J.D.,Smith, K., Walker, A.S.D. & Webb, P.C. (1988) Metallogenicmodels and exploration criteria for buried carbonate-hostedore deposits: results of a multidisciplinary study in easternEngland. In: Mineral Deposits within the European Community(eds J. Boissonnas & P.Omenetto), pp. 321-377. Springer-Verlag, Berlin. *'

Podwysocki, M.H., Segal, D.B. & Abrams, M.J. (1983) Useof multispectral scanner images for assessment of hydro-thermal alteration in the Marysvale, Utah, mining area.Economic Geology 78, 675-687.

Prost, G. (1980) Alteration mapping with airborne multispec-tral scanners. Economic Geology 75, 894-906.

Raines, G.L. & Wynne J.C. (1982) Mapping of ultramafic rocksin a heavily vegetated terrain using Landsat data. Economic

Geology77,1755-1769.Ramasamy, S.M. & Balaji, S. (1995) Remote sensing and

Pleistocene tectonics of Southern India penisula. InternationalJournal of Remote Sensing 16, 2375-2391.

Ramos, V.A. (1977) Basement tectonics from Landsat imageryin mining exploration; Geologie en Mijnbouw 56, 243-252.

Resrnini, R.G., Kappus, M.E., Aldrich, W.S., Harsanyi, J.C. &Anderson, M. (1997) Mineral mapping with HyperspectralDigital Imagery Collection Experiment (HYDICE) sensordata at Cuprite, Nevada, USA. International Journal of RemoteSensing 18, 1553-1570.

Richards, D;M., Jones, V.T., Matthews, M.D., Maciolek, 'I.,Pirkle, R.J. & Sidle,W.C. (1986) The 1983 Landsat soil-gasgeochemical survey of Patrick Draw area, Sweetwater County,Wyoming. Bulletin, American Association of Petroleum Geologists70,869-887.

Rivard, B. & Arvidson, R.E. (1992) Utility of imaging spectro-metry for lithologic mapping in Greenland. PhotogrammetricEngineering and Remote Sensing 58, 945-949.

Rothery, D.A., Francis, P.W. & Wood, C.A. (1988) Volcanomonitoring using short-wavelength infrared data fromsatellites. Journal of Geophysics Research 93, 7993-8008.

Rowan, L.C. & Bowers, T.L. (1996) Remote mineralogic andlithologic mapping of the Ice River alkaline complex, BritishColumbia, Canada, using A VIRIS data. PhotogrammetricEngineering and Remote Sensing 62, 1379-1385.

Viewing remotely sensed images in three dimensionsrelies on the stereoptic capabilities of the human visualsystem. As outlined in Section 2.4, for an object to be per-ceived in three-dimensional space the absolute parallaxof its images on the retinas of both eyes must be greaterthan the minimum resolution of the visual system. Portwo objects to be discriminated in three dimensionssimilarly means that their relative parallax or parallaxdifference at the eyes also must exceed this minimum.If the positions of an imaging platform at high altitudeor in orbit are spaced widely enough during the captureof two overlapping images, then the resulting abso-lute and relative parallaxes on the images ci}ll be largeenough for them to be resolved visually. This being thecase, a stereoptic model of the surface topography is seenif each image can be viewed separately by each eye. Theadjacent images are substituted for the natural viewsusing both eyes, and the effective eye base is increasedfrom 6 or 7 cm to as much as 100 km. Relief is perceivedbecause a continuum of parallax difference exists for allof the points that constitute the surface.

Parallax is normally achieved in images by differentpositions of the platform as in overlapping aerial pho-tographs taken from adjacent positions along the sameflight line, or from different flight lines with differentviewing geometries, as in the case of radar and line-scanor pushbroom systems (Chapters 3 and 7). Landsatimages from adjacent paths have between 15 and 80%overlap, from the Equator to the highest latitudescovered by the system. Stereoptic potential is availablefrom both the SPOT and MOMS systems (Section 3.12)by selective pointing off-nadir and a combination offorward- and nadir-looking devices, respectively. Italso can be introduced synthetically into a single image

~

/ photo ~

~

7

) pri pal

p t

I

Ll.

~~

-.,., J

j

249

250 Appendix A

I-.- eye base -;,ojl

liens stereoscope mirror

I \

photo 1photo 2

left photo right photo

Fig. A3 The lens system in a reflecting stereoscope opticallystraightens the line of sight, as in a lens stereoscope. Theprisms and mirrors enable the pair of images to be separatedsufficiently for the whole area of overlap to be viewedstereoptically.

~: convergence for );.~ comfortable vision

Fig. A2 This simple ray diagram shows how the lenses of astereoscope optically straighten the line of sight from each eyewhen the eyes are focused at and converge towards a point atthe normal distance for comfortable reading. The effect is toenable each eye to see a separate image and to create theillusion of viewing the stereoptically fused pair of imagesfrom a great distance.

method of achieving unaided stereopsis of an image pairis 'to stare'through' them, as in a daydream, and thento coax the eyes to regain focus without convergence.With luck, a blurred stereoptic model is achieved,which becomes sharper with perseverance. Of course,to accomplish this means that common parts of bothimages must have toughly the same separation as theeyes. A simple aid for the unsuccessful beginner is toseparate the two views physically with a card held atthe bridge of the nose. Once fusion and focus can beachieved, then the card can be dispensed with. As usefulas unaided stereoptic viewing of image pairs is, only aminority of people can manage to do it. The majorityrequire some kind of optical assistance-a stereoscope.

In a simple lens stereoscope (Fig. A2) the lenses refractrays from an .object portrayed on both images in such away that they correspond to the convergence normallyexperienced with close-range vision. The two imagesof the object are focused on the left and right retinaswith no strain, and the stereoptic model of relief quicklysprings into being, as can be tested with any of thestereopairs of images in Chapters 4, 7 and 8. The problemis that only a portion of the overlap between the twoimages as wide as the eye base can be viewed. To viewthe full area of overlap means using a system of lensestogether with prisms or mirrors, or both, in a reflectingstereoscope (Fig. A3). This combines the refraction ofray's to the normal convergence angle with an increasein the effective eye base, in much the same way' as

binoculars. All stereoscopes, including those incorporatedin sophisticated plotting and measuring instruments, arebased on these simple principles.

The stereoptic potential of image pairs with parallaxdifferences also can be exploited by other optical prin-ciples. One involves projecting both images on to ascreen, one through a filter that polarizes light in thehorizontal plane, the other through a vertically polariz-ing filter. The screen is viewed through spectacles withfilters having corresponding polarities. Each eye receiveslight from only one of the two projected images, andstereoptic fusion results. This is a handy method of pre-senting and teaching the principles of stereoptic photo-geological interpretation. Another way of ensuring thateach eye sees only one of a pair of images is to print orproject the images in separate primary colours. Viewingwith spectacles having filters with the same arrange-ment of primaries ensures that only light from theleft image is received by the left eye, and vice versa.Coloured light from the other image is absorbed by thefilter. This is the principle of the anaglyph, once popularin children's books and three-dimensional movies. Allthe stereoscopic figures in Chapter 4 are presented asanaglyphs on the CD, which has an accompanying pairof red-green spectacles.

Viewing most of the pairs of images in Chapter 4with a lens stereoscope reveals a marked degree ofvertical exaggeration of the topographies involved. Thisphenomenon is very useful for highlighting subtle topo-graphic features that otherwise would go unnoticed.However, it also causes all slopes to appear steeper thanthey really ate. This is particularly irritating when dipsof geologically important surfaces need to be estimated.In most cases vertical exaggeration means that estimatescan be made only in terms of horizontal, gentle, moder-ate or steep dip. The amount of vertical exaggeration

Stereometry 251

.

......Fig. A4 Viewing these converging rows of dots with a lensstereoscope produces a stereoptic illusion that they form asingle row which dips into the page. This is the principlebehind both the parallax wedge and the stereometer. In aparallax wedge, each pair of dots is accompanied by a scaledivision indicating the absolute parallax between the dots.In a stereometer this distanc~ IS measured using a micrometerto which the plates carrying the dots are attached.

h = ~p/(ta~92 -ta~91) (A2)

where 91 a~d 92 are the depressio~ angles from theantenna to objects on the two images. .

Because parallax differences in most images are small,they cannot be measured precisely using a ruler. It ismore convenient and useful to use the principles ofstereopsis during stereoptic viewing itself, The simplestmethod is to use a transparent plate engraved accuratelywith two converging rows.of dots, crosses or other marks,known as a parallax wedge (Fig. A4). When viewed witha lens stereoscope the two converging rows of marksappear as a line dipping through the stereoptic model,because of the regular variation in absolute parallaxbetween adjacent marks. By moving the wedge overthe stereopair at right angles to the flight line, one of thefused pairs of marks can be 'sat' on a particular topo-graphic feature. The corresponding absolute parallaxof the two marks is read from the scale. Another pair ofmarks is fused so the floating mark 'sits' on a feature ata different elevation, to give the absolute parallax at thatpoint. The difference between the two readings is thenused in Equation (AI) to give a difference in elevationbetween the two points. Clearly, a parallax wedge togetherwith a lens stereoscope is a useful tool in the field.

The parallax bar or stereometer uses the same prin-ciples as the parallax wedge. It consists of two smalltransparent plates, each engraved with a dot or a cross,which are linked by a micrometer screw so that th~irseparation can be measured precisely. With the baraligned parallel to the flight direction the separation ofthe spots is adjusted so that the fused spot 'sits' on .thesurface, first at one PQsition and then at anoth~r. Theparallax difference is read from the micrometer, and isconverted to a difference in elevation..

depends on the geometries of both the imaging andviewing systems.

To ensure complete stereoptic cover aerial photo-graphs -generally have a 60% overlap or more along aflight line. Two factors contribute to the overlap. First,the focal length of the lens governs the field of viewand, for a fixed frame size, the scale of the image. Theshorter the focal length, the wider the field of view andthe smaller the scale. The distance flown by the plat-form between successive exposures-the air base-is theobvious second factor. At a fixed altitude therefore 60%overlap is achieved by a larger air base for a lens withshort focal length than for one with a longer focal length.The two images in the first case contain a greater relativeparallax for two objects with different elevations. Thisproduces an enhanced three-dimensional eff~ct whenthe two images are viewed stereoptically. From these

'*simple relationships the degree of vertical exaggerationcan be deduced to be directly proportional to the ratiobetween the air base and the altitude.

Because of the advantages of vertical exaggerationfor interpretation, and because slope angles can bemeasured by other methods, aerial photographs usuallyhave an air base: altitude ratio of between 0.4 and 0.6.This gives exaggeration factors of between three andfour times normal. Beyond exaggeration factors of 4, theamount of relative parallax between objects with highand low elevation may be larger than the visual systemcan accommodate. To view the whole scene stereoptic-ally requires different image separations for the low andhigh parts. The field of view of orbiting systems, althoughlarge in absolute terms, is small relative to altitude toensure that distortions at the margin of images are keptto a minimum. As the equivalent of air base for over-lapping Landsat images decreases towards the poles froma maximum at the Equator, the vertical exaggerationchanges from about 1.2-0.2. A useful stereoptic model isonly possible for latitudes below about 50°. The stereo-scopic imaging arrangements for SPOT and MOMS aimat exaggeration factors of around two.

The parallax differences between overlapping imagesprovide the basis for quantit.ative measurements of therelative elevations of objects. Where b is the air baseof the images, H is the altitude of the platform and Apis the parallax difference between the images of twoobjects, their vertical separation h is given by:

H=HAp/b (AI)

provided measurements of Ap and b are in the sameunits. Converting the parallax differences between objectson pairs of overlapping radar images to relative eleva-tions involves different equations, which depend onthe particular configuration of the sensors. For the com-monly used same-side, same-h~ight arrangem~nt theelevation difference h is given by:

252 Appendix A

two images to place thfm both in the same frame ofreference. This involves rotation and aligning the imagesalong the flight path that connects the two principal points,as well as resampling them to the same pixel dimensions.This is much the same as the set-up for viewing withan optical stereoscope (Fig. Al). In a similar fashion tothe eyes' adaptation to parallax in creating stereopticfusion, the software finds the parallax in the along-flightdirection to calculate elevation differences for everycommon pixel. Calibrated to elevation values for controlpoints, this gives a DEM with absolute elevation abovethe sea-level datum. The difference lies in the softwareusing pattern-matching algorithms to 'recognize' commonpoints automatically. The efficiency with which it doesthis depends on the images' quality. This means thata high percentage of pixels are -not recognized andomitted from the calculations. Instead, those points forwhich parallax can be measured form the basis ofa TIN(Section 8.2.4), which the software interpolates to a con-tinuous topographic surface. Consequently, we can seedetails in stereopairs that do not appear in DEMs.

Further reading

More sophisticated instruments used in stereometryincorporate means of calculating elevation differences,together with mechanisms for removing tilt produced inthe stereopairs during flying and for transferring informa-tion from the images to a base map. A pair of markssimilar to those used in the parallax bar are on linked arms,so the fused mark can be moved along topographic orgeological features. Another linkage incorporating theappropriate scale conversion drives a pencil across a basemap to record the features. If the fused mark is main-tained apparently on the surfac~, then it is possible tomap topographic contours directly. Further refinementcombines the results from such plotters and measur-ing systems through microprocessors to produce digitalmaps. Further details of the various instruments used inphotogrammetry are beyond the scope of this book.

The angle between the horizontal and a line that joinstwo points can b~ defined by the ratio between theirdif-ference in elevation and their hQrizontalseparation. Thisratio is the ~ngent. The angle therefore can be estimatedby stereometric measurement of parallax difference forthe two points and their horizontal separation. If the twopoints. lie on the same dipping surface, such as a topo-graphic slope or a bedding plane, then this angle is thetrue dip of the surface. The advantages of this possibilityfor the geologist equipped with aerial photographs, astereoscope and a parallax wedge or bar are obvious.One in particular is that the minor variations in dip thatplague field measurements are removed to give a localaverage. .

Relative elevations of objects on images can be mea-sured without involving parallax, but with less precisionby using th~ lengths of $hadows that they cast. If theangle lJ at which the object is illuminated by the Sunor by radar is known, then the elevation of the top ofthe object above its base h is related to the length of its

sbadowlby:h = 1 tanlJ ; (A3)

Useful as this method may be for estimating topographicelevations of landforms, it is of little use in estimatingdips,. because the very presence of shadows obscuresgeological features...,Automated derivation of elevation data from digital

stereopairs of images uses the basic principles of stereo-metry,; One of the images is geocoded using the geo-graphical co-ordinates and elevations of ground controlpoints. The operator creates common tie points on the

Coupland, D.H., Vincent, R.K. & Pleitner, P.K. (1985) Examplesfrom anew digital elevation mapping algorithm for stereo-scopic image data. Proceedings of the 18th InternationalSymposium Remote Sensing of Environment, Paris.

Day, T. & Muller, I.-P. (1989) Digital elevation model produc-tion by stereomatching SPO'Pimage pairs: a comparison ofalgorithms. Image and Vision Computing 7,95-102.

Lattman, L.H. & Ray, R.G. (1960) Aerial Photographs in FieldGeology. Holt, Rinehart and Winston, New York.

Orti, F.(1981) Optimal distribution of control points to minimizeLandsat image registration errors. Photogrammetric EngineeringAnd Remote Sensing 47, 101-110.

Rochon, G., Haja, S.R. & Simard, R. (1985) Automatic digitalelevation model extraction from digital stereo images. Pro-ceedings of the 18th International Symposium Remote Sensing ofEnvironment, Paris.

Shlien, S. (1979) Geometric correction, registration and resam-pling of L~ndsat imagery. Canadian Journal of Remote Sensing5,74-89.

Slama, C.C. (ed.) (1981) Manual of Photogrammet!Y, 4th edn.American SoCiety ofPhotograrnmetry, Falls Church,Virginia.

Thompson,M.H. (ed.) (1966) Manual of Photogrammetry. 3rd edn.American Society of Photograrnmetry, Falls Church, Virginia.(2 Volumes.)

Wolf, P.R. (1974) Elements of Photogrammet!Y. McGraw-Hill,New York.

B.1 Geometric rectificationMost kinds of remote-sensing device produce unavoid-able geometric distortions in the images that they pro-duce. These departures from normal map projections ofthe Earth's surface include the warping effects of devia-tions in the flight direction and attitude of the platformsfor remote sensing (Chapter 3). As well as distortions,defects in the radiometric fidelity of the images are alsocommon. These range from mistakes in the exposure,processing and printing of photographic film to theresults of systematic foibles in digital imaging devices.

Section 5.2 covered poor contrast resulting from atmo-spheric scattering and compression in digital images,and how to compensate for them. Other common blem-ishes in digital images are caused by malfunctions inthe hardware, or are peculiarities of the method itself.The manufacture of line-scanning devices and, to a lesserextent, pushbroom systems fails to achieve a perfectmatch between the responses of individual detectors tothe same stimulus. Each detector may also drift or deteri-orate with time, especially if its temperature fluctuates.The net result is striping parallel to the scan lines on theimage. In the Landsat MSS and TM instruments detec-tors for visible and near-infrared radiation are mountedin banks of 6 and 16, respectively. The effect of mis-matches between the detectors is repeated patterns ofstriping, each stripe being 6 or 16 lines deep.

Brief failures in the circuits of a detector result inspurious signals during a scan cycle. This may producelines without data or, where the data have uniformly highvalues, which appear as black or white line drop-outs,respectively. Build-up of minute electrostatic charges inthe circuits of the system and their eventual dischargecan result in a spurious response for a single pixel.This noise effect generally is distributed as randomspeckles in an image. A very similar result stems fromthe complex returns of microwave signals, so that radarimages are systematically infested with speckle. In fact,the speckle does contain valid information relatingto attributes of the surface, but this is rarely of muchinterest to the geologist, and it disturbs the topographicpatterns of interest.

Not only do defects degrade the visual quality ofan image, but many of them have spatial frequenciesclose to where the achromatic MTF of the human visualsystem rises to a maximum (Section 2.2). We notice themmore readily than natural variations, and even a slightdefect disturbs interpretation.

All images contain geometric distortions of some kind.There are many contributory effects-different kindsof platform instability, optical and mechanical distor-tions in the radiation-gathering set-up, Earth rotationbeneath the platform, and both parallax and scale effectsresulting from topographic relief. Apart from the lasttwo, simple corrections of the geometry remove mostdistortions. Where precise information about platformattitude and track relative to the Earth's rotation isavailable, this can be built into automatic correction pro-grammes. In polar-orbiting satellite images the correc-tion for Earth rotation is simply a systematic line-by-linedisplacement of pixels, very much like giving a slant tothe edge of a pack of cards.

At large scales1he Earth's surface approximates a planeon which the topography is superimposed. Rectificationof an image essentially restores objects to their correctrelative disposition in two-dimensional space as if theywere all viewed from directly above. Small-scale imagesfrom metsats contain distortion effects resulting fromEarth's curvature. They are projections of a sphericalsurface into two dimensions. All maps have similardistortions. Different scares and different uses call fordifferent kinds of projection, however, such as stereo-graphic, conical, cylindrical and more complex rendi-tions. All aim to maintain a common scale over a wholemap, a minimum of distortion, and a uniform repres-entation of shapes and areas. As maps are the stock-in-trade for all users of remotely sensed images, geometricmanipulations of images attempt to register images tothe map projection required for a particular project.

Map projections involve two main features: a geomet-rical model that closely approximates the shape of theEarth, and a co-ordinate system for referring. to loca-tions. Bulging because of rotation means that the Earth isslightly flattened-its polar radius is slightly shorter thanthat to points on the Equator. The planet is close to beingan ideal ellipsoid. However, topography and variationsin density within the Earth result in local changes in itsgravitational field. For surveyors this is something of aproblem, because defining a horizontal surface dependson the angle of a plumb line! The imaginary surfaceat which gravity is the same as at mean sea-level-thegeoid-is not an ellipsoid, and can deviate from the idealby up to 100 m.

Co-ordinates that express map position always referto a geodetic datum on which the map is based. Such a

253

254 AppendixB

describe the deviation ill two dimensions of the imagefrom the map projection. The pair of polynomial trans-forms allow a computer to warp the image to registerwith the map, by treating the image mathematically likea rubber sheet. In the production of a new rectifiedarray of pixels for the image, it is possible to change thedimensions and shapes of pixels. This can involveshifting, rotating and distorting the original pixels.

Each new pixel in a rectified image must haveassigned to it a ON derived from those in the originalimage. This is resampling. As the new pixels are dis-placed relative to the raw array of ON, simply assigningthe original ON of a shifted pixel will produce errorsin the new image. The simplest means of allowing forthis effect is to assign the ON of the nearest raw pixel toa pixel in the new array. This is the nearest-neighbourmethod (Fig. Bla). Being a crude approximation it resultsin the resampled image appearing ~locky'. It retainsthe original statistics of the data, however, and thus iscommonly used when the resampled image is intendedfor multispectral classification. A more geometricallyprecise approach is to assign the distance-weightedmean ON of several raw pixels immediately surround-ing the position of a rectified pixel using convolution(Section 5.3). Best results involve the mean of the 16 sur-rounding raw pixels using a method known as cubicconvolution (Fig. Bl). This reduces theblockiness in therectified image, and keeps disturbance of the originalfrequency distribution of ON to a minimum.

The parallax and scale variations that cannot beremoved from an image by its digital registration to amap depend on both elevation differences and position

Fig.

Bl In the nearest-neighbour method of resampling thedigital number (DN) for an output pixel (a) is taken unchangedfrom the input pixel (shaded) lying nearest to the new one.Resampling by cubic convolution produces an output DN at b,which is the mean value of the DN from those 16 input pixelsgrouped around the position of the output pixel (shaded). Amore sophisticated version weights the contribution of the DNof each input pixel t6 the mean according to its distance fromthe

new pixel position.

datum uses an ellipsoidal surface, but no single one fitsevery part of the planet. So different areas need differentellipsoids, and a datum specifies both the ellipsoidappropriate for a region of Earth's surface and the originfor its co-ordinates.. In some co-ordinate systems/such asthe metric grid used in the Universal Transverse Mercatorsystem, different ranges of latitude and longitude con-stitute different zones. Finally there are projections. Theobject of a projection is to render the ellipsoidal Earth'ssurface as another surface, which can be flattened with-out tearing or distortion on toa flat piece of paper. Onlythree kinds of surface can do this: planes thatlie tangen-tial to the surface,. cones and cylinders. These form thebasis for the .many projections that cartographers use.Each involves distorting surface features in some way;in terms of scale, area,or shape. The first is impossible toachieve everywhere, and the second and third, althoughpossible, are mutually exclusive. Such is the life of thegeographer! However, maps are what we use to refer toland and its geology; they are the arbiters of accuracy.That is why transforming images to conform with mapsis called rectification! Things go awry when co-ordinatesrefer to inappropriate datums, and that can lead to errorsof up to ahnost a ki.lometre.

As most digital images consist of rectilinear arrays of DNfor columns and rows of pixels, rectification consists of atransformation from one set of Cartesian co-ordinates toanother. The three~di~ensional co-ordinates of points,expressed. as latitude and longitude, can be expressedby l'eference to a rectilinear array on maps. However,describing a position accurately in degrees, minutesand seconds of arc is difficult; Points in a metric grid ofeastings and northings are more easily assessed. Themost commonly us~d grid of this kind is the UniversalTransverse Mercator (UTM) projection found on mostmodem maps.

Image distortions; except those caused by parallaxdue to differences in surface elev~tion, can be removedby registration of a digital image to a map base. Theprocess begins by identifying topographic or culturalfeatures visible on both the image and the map. Themap co-ordinates are assigned to the pixel and lineco-ordinates of a number of these ground control pointson the image. A very similar procedure can be used toregister an image of one type to another. Dependingon the number of control points, a pair of polynomialequations, such as:

X\II=a + bX+ cY+dX2 +eXY+ fy2+ gX2y+ hxy2'+ iX3.y3+1 +...

Y m = Z + yX + xY + WX2 + vXY + Uy2 + tX2y + SXy2 +

rX~+qy3+:. .

where Xin and Y m are map co-ordinates and X andY areimage co-ordinates, can be generated. These equations

Image Correction 255

of features relative to the nadir of the imaging system.For most satellite images these do not produce distor-tions that are much greater than the imprecision inherentin surveying and printing small-scale maps. For instance,an elevation difference of 1 kIn at the edge of a Landsatimage gives a displacement equivalent to only 160 m.Parallax does introduce important distortions in aerialphotographs with scales greater than 1 : 100000.

To overcome parallax and scale distortions, inter-pretations from stereopairs of aerial photographs aretransferred to maps using various optical-mechanicalplotters. Where maps do not exist at a scale suitablefor the interpretation they must be constructed from thephotographs using triangulation methods that allowfor radial-relief displacements. The parallax involved inradial-relief displacement can be removed from aerialphotographs using an orthophotoscope. In.. this instru-ment the image is rephotographed through a narrowslit. The operator views the stereomodel and attemptsto keep an image of the slit at the ground surface (seeAppendix A) as the image is scanned in one direction asa series of narrow swaths. Through various mechanicallinkages this manipulation is translated into up-and-downmovements of the film on to which the image is projected.This controls the local scale of the new image on the film.The process can involve direct rephotography, or thethree-dimensional co-ordinates controlling the film andslit positions can be stored and replayed through theorthophotoscope when the task is complete. The finalproduct is an orthophotograph, which represents eachpoint as though it has been viewed from directly above,just as in a map. Where digital elevation data are avail-able complex computer programs can remove parallaxfrom digitized photographs, again resulting in an ortho-

graphic image. Fig. B2 The Landsat-2 MSS image of an area in the EthiopianHighlands (a) is severely affected by lines of spuriously lowdata. In image (b) the spurious lines have been replaced by,data derived from those in the adjacent lines. At the scale thedefect is hardly noticeable.

B.2 Replacing dropped lines

Figure B2(a) illustrates graphically the disturbing effectsof a relatively few lines in images that contain spuriousdata. The problem can be redvced by replacing each linewith data derived from those adjacent to it Each pixel inthe dropped line is assigned the DN of the mean valueof the pixels immediately above and below it. Althoughthe new line is entirely artificial, it is so similar to it~neighbour that it becomes barely noticeable, except underhigh magnification. The degree of cosmetic improvementthat can be achieved is clear from Fig. B2(b).

B.3 De-striping

TM images, it appears as a regularly repeated defectthroughout an image. A simple and rapid means of cos-metic treatment of such simple striping begins by com-paring histograms for all of the detectors in a bank over arepresentative part of the scene. The histogram with thehighest mean ON and the greatest variance is selected asthe standard, and the others are then normalized to it.Individual ON for pixels contributing to each histogramare changed to fit the 'master' histogram using look-uptables (Chapter 5).

Histogram-matching methods remove stripes effect-ively from images of relatively homogeneous scenes.Where scenes contain large areas with very different over-all brightness, the correction may leave residual stripes

Where image striping stems from different responsesfrom each detector in a bank, as in Landsat MSS and

256 Appendix B

(a) (d)

(b) (c)Fig. B3 A variety of problems in theJER5-1 short-wavelengthinfrared (SWIR) sensors produced severe defects in image(a) of band'8 covering part of the Eritrean Highlands. (b) Themagnitude of the Fourier transform of image (a), in which thenoise shows as a variety of 'spikes' and lines, as explained in

(c). Using ruters specially designed for this unique set ofdefects almost completely removes the noise, as shown inthe restored image (d). Courtesy of Beto de Souza Filho,University of Campinas, Brazil.

in particularly dark and light are~s. More comprehensivede-s~ping methods involve more complex computations.One is to use convolution filtering based on matriceswith dimensions that match the number of sensors inthe imaging system. The filter effectively smooths outthe striping. Other methods rely on identifying the

frequency and amplitude of striping and other all-pervasive, systematic defects, which are superimposedon the real variatio~, by analysing a Fourier transformof the image. The periodic noise can then be identifiedas a signal and suppressed by filtering in the frequencydomain. The inverse transform to the spatial domain

Image Correction 257

then restores the cleaned-~pimage. Figure B3 shows thecomplex noise to which JERS-l SWIR images are prone.Without cosmetic treatment they are difficult.to inter-pret. In the frequency domain the noise is isolated as dis-tinct signals, and almost wholly removed by speciallydesigned filters that operate in the frequency domain.

B.4 Removal of random noise

Fig. B5 An image of the ratio between Lands~t MSS ban'qs 4and 5 for an area in .the Chilean Andes (a} is plagued by an'.increase in thesignal-to-noise ratio, which is meyitableirl ""

ratio images. By convolvmgthe raw image with a 3 x 3median filter the speckle and residual striping is virtuallyremoved (P).

Random noise, or speckle, degrades the visual impact ofthe raw images (see Fig. B5a), and therefore should beremoved by some means. Being randomly distributed,it is uncorrelated with the real data in a multispectralimage. In the decorrelation process involved in principalcomponent analysis such poorly correlated data ~ndup in the highest order components. Perfgrming PCA,discarding the highest PC, and then reconstituting thedata by the inverse transform is a logical, although by nomeans quick, solution to the visual disturbance intro-duced by noisy pixels.

For some noisy images, such as those produced byband ratioing techniques or by radar systems, principalcomponent analysis is not an appropriate means of noiseremoval. The interesting information is best retained inthe original data set. A hypothetical DN resulting fromnoise is represented in Fig. B4 by the number 255, andis surrounded by valid DN. Calculation of the mean ofthe nine DN to replace the DN of the central noisy pixelusing a convolution filter produces a value of 51, andalso blurs the image. The new pixel is still sufficientlydifferent from its neighbours to appear as a speck in theimage. However, the median DN of the sequence 24, 25,26,26,27,27,255 is 26. As this is more representative of thesurrounding DN, the resulting image is effectively clearedof noise, whereas real data remain almost unchanged.Such median filtering is possible using a 3 x 3 squareconvolution matrix in the spatial domain. Figure B5(b)shows the efficiency of median filtering.

Most image processing software packages includea range of filtering methods to remove image defects.Deciding on the most appropriate for a particular prob-lem is a matter of trial and error.

Associated resources on the CD-ROM

You can access the resources by opening the HTMLfileMenu.htm in the lmint folder on the CD using your Webbrowser. More detailed information about installing theresources is in Appendix C.

Exercise 6 includes TNTlite processes that are usefulin cosmetic treatment of image defects, and showssome of the artefacts that appear when using spatialfiltering. Exercise 7 covers georeferencing artd rectifica-tiort of images.

Fig. B4 This array of digital numbers (DN) fora 3 x 3 part ofan image consists of eight valid DN surrounding a spuriouslyhigh DN, which constitutes system noise. Replacing the valueof 255 with the median DN for all nine pixels removes thenoise effect..

258 Appendix B

Further reading

Bernstein, R. (eds) (1983) Image geometry and rectification.In: Manual of Remote Sensing, 2nd edn (ed. R.N. Colwell),pp. 873-922. American Society of Photogra~etry, FallsChurch, Virginia.

Billingsley, F.C. (1983) Data processin:g and reprocessing.In: Manual of Remote Sensing, 2nd edn (ed. R.N. Colwell),pp. 719-792. American Society of Photogrammetry, FallsChurch, Virginia.

Crippen, R.E. (1989) A simple spatial filtering routine for thecosmetic removal of scan-line noise from Landsat TM P-tapeimagery. Photogrammetric. Engineering and Remote Sensing.55,327'-331.

view them while reading Chapter 4 by using the red-blue spectacles packed with the CD, either in a graphicspackage, such as Adobe PhotoShop or MS Photo Editor,or in TNTlite after importing these JPEG files to Micro-Images format (see later). Other folders contain a col-lection of different kinds of images, which illustratevarious geological themes and types of imagery, andhave appended brief descriptions. You should viewthese using a Web browser because the images arelinked to text using HTML-i.e. the compressed imagesare in a separate sub-folder and therefore accessible bythemselves. Many of the images in the collection havesizes that are compatible with TNTlite, so they can beimported as a means of adding interactive, digital inter-pretation to Exercise 9. They are 24-bit RGB JPEGs andso are limited to three channels. That somewhat limitstheir use with the other exercises, but the JPEGs can beunpacked to three 8-bit components within TNnite, forexperiments with contrast stretching, some ratioing andfiltering.

Contents of the CD-ROM

The CD contains several folders or directories, most ofwhich relate to TNTlite.

TNnite resources

There are many folders associated with the TNTliteresources on the CD, some of which are copied auto-matically when you install TNTlite on your hard disk,and others that you have the option to install or usefrom the CD. The most relevant of those which you canchoose to use from the CD are:

Data contains data compatible with TNTlite (inData \ Litedata) that are formatted as MicroImagesproject files (*.rvc), each of which contain several dataobjects of various kinds. MicroImages' own tutorialexercises (documented in the Getting Started Acrobat-format *.pdf files) use them.

Getstart contains illustrated Adobe Acrobat docu-ments (*.pdf files) for each of the MicroImages GettingStarted tutorials. Folder Acroread contains the latestAdobe Acrobat Reader software that automatically runsfrom the CD when you open an Acrobat document fromGetstart by double-clicking on a *.pdf file. If you opted toinstall the Getting Started documents on your hard diskalong with TNTlite (see Installing TNTlite below) youcan access them from the Help option associated with

The CD-ROM at the rear of the book contains resourcesaimed at extending and applying what you should havegrasped from reading the text. I hope that it transformsImage Interpretation in Geology from a textbook support-ing courses in geological remote sensing to a course in itsown right, that will help professional geologists developnew practical skills through home study, and can sup-port college instructors in enriching existing courses orsetting up new ones for undergraduate geologists.

The central component uses a version of the profes-sional Map and Image Processing System (TNTmips)software developed and marketed by MicroImages Inc.of Lincoln, Nebraska, USA. MicroImages are the onlysoftware developers in this field to offer a freeware ver-sion (TNTlite), which is fully functional. TNTlite differsfrom TNTmips only insofar as it limits images to a maxi-mum size of 314368 cells (e.g. 512 614, 1024 307pixels), vector objects to 500 points, 1500 lines and 500polygons, CAD objects to 500 elements, TIN objects to1500 nodes and database objects to 1500 records pertable, and has disabled export capabilities. These limits,however, do not prevent results being printed, andTNTlite can be used for small research projects.

The inclusion of TNTlite in Image Interpretation inGeology does not imply any endorsement by eitherthe publishers or myself.

On the CD are 11 Exercises linked to Image Interpreta-tion in Geology, which are designed to allow users to mas-ter the skills associated with TNTlite processes. Theseskills are essential in all aspects of geological remotesensing, image interpretation and the production of dig-ital maps to a publishable, professional standard. Eachexercise uses data files specially prepared to be compati-ble with TNTlite. The CD also contains MicroImages'own tutorials and supporting data files, as a series of 54Getting Started booklets in Adobe Acrobat format (thelatest version of Acrobat Reader is on the CD too). Thoseresources cover aspects of remote-sensing and othergeospatial applications (land use, urban and environ-mental) that will be broadly useful to Earth scientists.The full TNTmips Reference Manual, which providesmore detailed guidance, is also on the CD-ROM.

Randy Smith, PhD of Micolmages Inc and MargaretAndrews of the Open University helped enormously ingetting the Exercises working, technically and as teach-ing material.

Two other kinds of resource reside in separate folderswithin the Image Interpretation in Geology directory(Imint) on the CD. One folder contains anaglyph ver-sions of all the stereopair figures in Chapter 4. You can

259

260 Appendix C

eachTNTlite process window as you are working, or atany time from the main TNTlite menu (Display/GettingStarted).

Refman contains the full, 2500 page TNTmips Refer-ence Manual as HTML ffies and linked GIF images (insubfolders Ci and Li). You can start this from Netscapeor Internet Explorer by using File/Open Page orFile/Open, respectively, selecting the CD (e.g. D:), thenD:/Refman/Html and double clicking on Toc.htm (tableof contents). If you opted to install the Reference Manualon your hard disk along with TNTlite (see InstallingTNTlite below) you can access relevant parts of the man-ual from the Help option associated with each TNTliteprocess window as you are working, or at any time fromthe main TNTlite menu (Display/Reference Manual).The manual is comprehensive, and you may find iteasier first to refer to the Getting Started document appro-priate to a process which you find difficult, and goingto the Reference Manual if you need details about how aprocess works.

If you have plenty of space on your hard disk, youwill firid it simplest to install all these folders, aswell as the TNTlite software, on your hard disk (seeInstalling TNTlite, below).

Mb if you wish to install all the Getting Started data sets,but you can access them from the CD if your hard-diskspace is limited.

CD-drive: Obviously!Processor: PCs-Pentium III, Pentium II, Pentium Pro,

Pentium or 486 computer (or compatible); MacIntosh-Power Mac.

Operating system: PCs-Windows 95/98/2000;Macintosh-MacOS 8.0 {version 8.6 or higher is recom-mended); LINUX.

Random access memory (RAM) At least 16Mb(32 Mb recommended, especially for MacIntosh). ForMacs, virtual memory must be set to at least twice that ofRAM.

Display: Minimum area of 640 x 480 pixels displaying256 colours (recommended minimum for Image Interpre-tation in Geology exercises is 1024 x 768 pixels and 16-bitcolour).

HTML browser: Netscape Communicator or Internet

Explorer.Adobe Acrobat Reader (the latest version is in

D:I Acroread/Win/Win95nt).

IIG

Resources

Image Interpretation in Geology resources

The folder Imint is where"you will find all the materialexplicitly connected with Image Interpretation in Geology.Within it are three folders:

Exercise This contains folders for the text and the datathat you will use in the TNTlite geological exercises.

Anaglyl This contains all the anaglyph versions ofthe stereopairs iri Chapter 4. You can easily access themfrom the CD using either TNnite or a graphics package,but copy the folder onto your hard disk if you haveplenty of space.

Anagly2 and Imagery. These contain selections of in-structive images from various sources and linked HTMLtext-more anaglyphs and different types of images,respectively. Again, you can opt either to access themfrom the CD or copy the folder onto your hard disk.

If you have plenty of space on your hard disk, youwill find it simplest to install all these folders on yourhard disk (see Installing Resources, below).

Sufficient hard-disk space for ti:le data files associatedwith the Image Interpretation in Geology Exercises. Theytake up about 140Mb (including about 100 Mb for thehyperspectral data) and text files for the Exercises (lessthan one megabyte). If you are short of space, then youcan access the folder (Imint/Exercise/Data/Cuprite) con-taiillng hyperspectral data from the CD-ROM. You mustin any event load all the other data associated with theExercises (Area1.rvc etc. in Imint/Exercise/Data) ontoyour hard disk, because you will be saving results intothese files. Images and HTML files that illustrate the textof Image Interpretation in Geology occupy about 30 Mh.

A graphics package, such as Adobe PhotoShop or MSPhoto Editor;

A Web browser, suchils Intem~t Explorer or NetscapeCommunicator (as high a version as you can get), as allthe Exercises are in HTML.1f you do not have a browser,or the HTML versions do not run properly, you will alsofind copies of the exercises in Rich Text Format (RTF,*.rtf) which should be readable in their origirtal formatusing any word processor less than 10 years old (NOTNotepad or other text-only packages).

Installing resources

Installation

PC users: autorun

Minimum requirements

TNTlite

Hard-disk space: At least 80 Mb for the TNTlite soft-ware. Getting Started and Reference Manual folders takeup 60 and 35Mb, respectively, plus an additional 150

If your system supports Autorun CDs, a few secondsafter loading the CD a "splash" window will appear onscreen. This has 6 buttons:

The CD-ROM Resources 261

Note: Each time lflat you need to use the "splash"buttons, you should eject the CD tray and then ~insertit.

PC users: manual

If for some reason Autorun does not display the"splash", you will need to perform the above manually,using Windows Explorer, as follows:

IIG Menu Display the contents of D:/Imint, anddouble click on Menu.htm.

Install IIG Resources Copy D:/Imint and all its con-tents to your hard drive C:.

Install TNTlite Double click on Setup.exe in D:, oruse Start/Run with the command line D: \setup.exe.

Note: Unlike many software packages, installingTNTlite does not create strangely named files andstealthily place them in some other hard-disk folder,where you can never find them. So there is no need tocarefully uninstall the software. If you want temporarilyto create space for something else, simply delete thewhole C:\TNT_WIN folder, and re-install TNTlitewhen you need to.

Run TNTlite Installing TNTlite will, automaticallyadd a line to your Start Up menu (Start/Programs) fromwhich you can run TNTlite 6.4. But see Tips below.

Browse You can browse the CD, as above, directlyfrom My Computer or Windows Explorer.

IIGMenu...InstalllIG Resources...Install TNTlite...RunTNTlite...Browse...Exit

IIG Menu Clicking on this button loads an HTMLmenu page into your default browser. This provideslinks to all the llG Resources on the CD (in folder Imint),and allows you to use the resources-including theExercises and data that relate to the TN11ite geospatialanalysis software-directly from the CD.

InstalllIG Resources Clicking on this button immedi-ately copies all the llG resources in the folder D:/Imintfrom the CD to your hard drive, using the DOS utilityxcopy. The path to the new- folder containing the IIGresources will be C:/Imint, i.e. it will be a tog-level folderon your hard disk.

Install TNTlite Clicking on this button starts theTNTlite installation procedure. You will first have toconfirm that you wish to Continue with installation(ignore the instruction to attach the software licencekey). There now follow some informative windows cov-ering system requirements that you have already read inthe previous section. The next important window dealswith Setup Options, of which you can select only oneat a time. Select Install/Setup the TNTlite version ofTNTmips and press Continue. Setup offers you theoption of taking the default destination for TN11ite-Path: C:\Tnt_win~r choosing another. Accept thedefault, when Setup installs the software and shows youprogress. When done, the installation procedure returnsto Setup Options window. Click in the window once,and you can decide whether or not to install the ReferenceManual, the Getting Started booklets and the Sample Dataassociated with them.

Run TNllite Clicking on this button launches theTNTlite software, provided that you have alreadyinstalled it in C:/Tnt_win. If you have not done this,nothing happens (TNTlite will not run from the CD). Thebutton is provided so that you can use resources anddata on the CD if you chose not to copy them to yourhard disk because of space problems.

Browse Clicking on this button opens a windowshowing the contents of the CD. By double clic~ng on aparticular folder, you will see a new window showingits contents and so on. You can open RTF, HTML andPDF files from here, by double-clicking on their icons,which will launch your default word processor, browseror Adobe Acrobat Reader. This is particularly usefulfor the TNTlite Getting Started booklets (in folderD:/Getstart), the TNTlite Reference Manual (in folderD:/Refman/html) or the IIG Resources (various foldersin D:/lmint), if you have decided not to copy them toyour hard disk.

Useful tips

Using the Resources can be made easier if you creqteShortcuts to several things.

TNTlite When you have coP1pleted the installationyou will see a small window (Free TNTlite products)with several icons. Left-hold-drag the icon labelledTNTmips 6.4 onto your desktop. Double-clicking on theshortcut will launch TNTlite,

If you have installed the Reference Manuill on yourhard disk, you can create a shortcut to the Table ofContents document on your desktop, so that you canstudy the manual without having to start TNTlite. Go toC:/Tnt_win/Refman/Html in Windows Explorer andright-click on Toc.htm. Select Create Shortcut from thepull-down menu, and then click-hold-drag the shortcutme onto your desktop from Windows Explorer. Double-clicking on the shortcut will open the Table of Contentsin your default browser, and you can select links tQ allthe features of TNTlite.

IIG Resources If you have copied all the Resourcesto your hard disk, go to C:/Imint and right click onMenu.htm. Select Create Shortcut from the pull-downmenu, and then click-hold-drag the shortcut file ontoyour desktop from Windows Explorer. Double-clickingon the shortcut will open the IIG menu in your default

262 Appendix C

browser, and you can select links to all the resourcesassociated with the book.

the resources, except fOl" TNTlite. You Reed to launchTNnite separately, either from Start/Programs or adesktop shortcut (PC), or using the recent applicationsand documents in the Apple menu (PowerMac).

TNTlite uses MI/X, MicroImages' X-server that oper-ates within Windows or MacOS (it is similar in somerespects to the UNIX windowing system). The X win-dow that opens, and within which allTNTlite windowsappear, when you launch TNnite, has one problem-you cannot easily resize it.. That does not mean thatTNTmips and other Windows or Mac software will notwork at the same time. The X Window merely hidesthem. That is no great problem, as both PCs andPowerMacs have means whereby you can bring otherwindows tQthe front. of the display-the Auto Hide andOn Top functions for the Windows and MS Office TaskBars, in the case of a PC. However, every time you clickinside the X Window the others hide again. You willneed to find ways to shift from one set of windows to theother. A crude option is to minimize the X Window-the'-' box at top right, but it is better to bring Windows orMacOS windows to the front when you need to, usingthe Task Bar in the first case and toggling the X-windowusing the Finder Icon on a Mac. So that you can viewboth the text for the Exercises and TNnite windows, thelarger the display area for which you have set yourmonitor, the more you will be able to see when readinginstructions. An area of 1024 x 768 pixels is OK, but1280 x 1024 is obviously better.

Once you begin the IIG Exercises, resize the windowthat you use to view the text to reveal as much as possi-ble of the TNnite windows. HTML files automaticallyfit the width of a browser window, so will pose noproblems. If you view the RTF format text files for theExercises in a word-processor window, you may findthat some of the text is hidden, depending on your soft-ware. Using the View/Online Layout option in recentversions of MS Word automatically wraps text to thewidth that you choose. If this is not possible, then youwill need to reset the line length for the text files to thewidth of your resized window.

PowerMac users

On a PowerMacAutorun will not function. Instead youwill see the contents of the CD in a window when youopen the CD icon from the desktop.

Install TNTlite Double click on the Installer icon. Thenext window allows you to select the License Options,so select Install TNTlite... and press OK. CheckTNTmips 6.4 for PowerMac, (if you wish to install theReference Manual check the Reference Manual HTMLoption) then press Install. TNTlite now loads into afolder on the hard drive automatically named TNTProducts 6.4.

Note: Unlike many software packages, jnstallingTNTlite does not create strangely named files andstealthily place them in some other hard-disk folder,where you can never find them. So there is no need tocarefully uninstall the software. If you want temporarilyto create space for something else, simply delete thewhole TNT Products 6.4 folder, and re-install TNTlitewhen you need to. Because files copied from the CD maybe locked, you will need to hold down the Alt key whenselecting the Empty Wastebasket option in order tocompletely delete them from the hard drive. This alsoapplies if you wish to empty the Imint folder from thewastebasket (see below).

Run TNTlite Open the TNT Products folder anddouble click on the TNTmips icon. TNT displays aMicroImages logo screen and then opens a full screenwindow named MicroImages X Window System(MI/X). All TNT processes take place within this uniqueX Server. Treat the Micro-Images X Window System as asimple background window. Any time during a TNTsession, you can use the normal Macintosh techniques(such as the Apple menu or the Application menu) tojump from MI/X back to the Mac desktop. TNTliteshould appear thereafter as a recently used applicationin the Apple pull-down.

InstalllIG Resources Copy the Imint folder to yourhard disk. For ease of access to the IIG Resources youwill need to launch Menu.htm from this folder. It willappear as a recently opened document in the Applepull-down menu thereafter.

Getting help

Like all image processing and desk-top mapping pack-ages, TNTlite is complex and completing the exercisesinvolves your mastering many different windows. It ispossible to get stuck or lost! The Getting Started bookletsare very useful in resolving common problems, as is theReference Manual. Two booklets entitled Getting Started:Displaying Geospatial Data and Getting Started: Navigatingare particularly useful in helping you learn to navigatearound in TNTlite and display your results. I have triedto follow MicroImages' style in the IIG exercises, but to

Working with the IIG resources

The whole package of resources that relate to ImageInterpretation in Geology is designed for access using abrowser (Netscape or Internet Explorer) from the HTMLfile Menu.htm in the Imint folder on your hard disk(or as a desktop shortcut on a PC), which links to all

The CD-ROM Resources 263

fessional version of TWmips. MicroImages software-support engineers give preference to those clients.However, as time allows, they will respond sympatheti-cally to queries from TNTlite users. Use support@micoimages.com for e-mail: and #1-402-477-9559 for faxenquiries (please do not phone) to contact MicroImagesSoftware Support.

MicroImages continually develops its products,generally releasing new versions on a quarterly basis.Anyone can obtain upgrades for TNTlite either bydownloading from the Web (http:/ / www .micro-images.com) or requesting a CD (MicroImages mayimpose a small charge to cover shipping the CD) from:MicroImages, Inc.11 th Floor-LBL Plaza South206 South 13th StreetLincoln,NE 68508,.-2010U&AFax:#1-402-477-c-9559 I

e-mail: info@microirnages.com

save space they do not repeat the basic TNTlite X-Windows operations. It is essential that you begin withExercise 1, which explains the most important, basicaspects of that style of operation. There are reminders inlater exercises of these basics, but not often, so as to buildup your skills in as short a space as possible. So, you canresolve any difficulties as you mount the learning 'ramp'if you return briefly to the earlier exercises to refresh

your memory.If you find any parts of the exercises difficult to follow,

even after looking for general guidance in the GettingStarted booklets, please contact me via Blackwell Science(but do persevere for longer than a few moments beforewriting). That will also help me to refine the instructionsfor a revised version of the CD. I will NOT answerquestions relating to operating systems or hardware,expecting you to be thoroughly familiar .,with yourown computer or having direct access to someonewho is.

MicroImages is a commercial company, with primaryduty to those clients who have purchased the pro-

Image atlases

These may be quite dated references, before the useof CD and Web image resources, but they constitute atimeless source of useful images, particularly suited forgeological interpretation. For individual use, it is pos-sible to use a desktop scanner to import your selection ofimages, making sure that the remove screens option isset because all use photolitho printing.

Image As: from the drop-down menu, and then assignyour own name and a suitable folder before saving. Inmost cases the files will be either JPEGs or GIFs, withthe appropriate extension. However, some sites thatallow you to browse global archives of data as reducedresolution images assign odd or nonexistent extensionsin the hope that no:'one will think of doing this. They areusually JPEGs, so assign the extension .jpg when saving.Also, remember all modern computers have a PrintScreen button that saves the screen to RAM. Try usingAlt+Print Screen, which copies the active window toRAM, and then pasting to a graphics package. You canhave lots of fun piecing together mosaics of such globalbrowse images, and can find educational or researchuses for them. Eritrea.bmp in the Images folder withinthe IIG directory is such an image.

Note: Some producers of remotely-sensed image datahave copyright in them, whereas others waive rights.You should check carefully on the websites for the copy-rights held by the data producers or suppliers. If youwish to disseminate downloaded data then you mustobtain permission from the source.

Ford, J.P., Blom, R.G., Bryan, M.G., Daily, M.l., !Axon, T.H.,Elachi,C. &Xenos,E.C. (1980) SeasatViews North America, theCaribbean, and Western Europe with Imaging Radar. Publication80-67, Jet Propulsion Laboratory, Pasadena, CA.

Ford, J.P., Cimino, J.B. & Elachi, C. (1983) Space Shuttle ColumbiaViews the World with Imaging Radar-the SIR-A Experiment.Publication 82-95, Jet Propulsion Laboratory, Pasadena, CA.

Hamblin, W.K. (1980) Atlas of Stereoscopic Aerial Photographsand Landsat Imagery of NQrth America. Tasa, Minneapolis,Minnesota.

lnstitut Geographique National (1984) l'Atlas Des Formes DuRelief. lnstitut Geographique National, Paris.

Mollard, J.D. (1974) Landforms and Surface Materials of Canada: aStereoscopic Airphoto Atlas and Glossary. J.D. Mollard, Regina,Saskatchewan.

Sheffield, C. (1981) Earthwatch: a Survey of the World from Space.Sidgwick and Jackson, London.

Sheffield,C. (1983) Man on Earth. Sidgwick and Jackson, London.Short, N.M. & Blair, R.W., Jr (1986) Geomorphology from Space.

NASA SP-486, US Government Printing Office, Washington,DC.

Short, N.M., Lowman, P.D., Freden, S.C. & Finch, W.A. (1976)Mission to Earth: Landsat Views the World. NASA SpecialPublication 360. NASA, Washington DC.

Short, N.M. & Stuart, L.M. (1982) The Heat Capacity MappingMission (HCMM) anthology. NASA Special Publication 465.NASA, Washington DC.

Web sites

These URLs lead you to pages either where digitalimages can be viewed and downloaded, or where youcan find useful items on remote sensing and related sub-jects. Many of the sites have lists of useful links, so thatyou can build up a comprehensive list of bookmarksyourself.

Tips: Most images displayed through an Internetbrowser will end up in your temporary cache, fromwhere you can copy them to permanent storage, but theywill often have incomprehensible names. A better optionis to right-click on the displayed image, select Save

ASTER HOME PAGE (JPL)

http://asterweb.jpl.nasa.goy/

ASTER site, Japan

http://astweb.ersdac.or.jp/ao/

British Geological Survey-Geoscience InformationResources

http://www.bgs.ac.uk/bgs/relat.htmi

Digital Chart of the World

http://www.maproom.psu.edu/dcw/

EarthWatch

http://www.digitaiglobe.com/

Environmental Research Institute of Michiganhttp://www.erim-int.com/

EOS Homepagehttp://eospso.gsfc.nasa.gov 1

Eurimage S.p.A.http://www.eurimage.com/

Goddard Space Flight Centerhttp://www-vOims.gsfc.nasa.gov IvOimslDA T APRODSERV IData_Prod- Types_Serv .html

Indian Remote Sensing Satelliteshttp:/ 1202.54.32~ 1641

264

Sources of Image Data 265

Landsat 7

http://landsat7.usgs.gov/

Marine Geology & Geophysics Images

http://www.ngdc.noaa.gov/mgg/image/images.html

Marine Terraserverhttp://www.ngdc.noaa.gov I mgg I image I images.html

MicroImages Atlaseshttp://microimages.com/ atlasserver

Microsoft Terraserverhttp://terraserver.microsoft.com/ default.asp

NASA Earth Science Enterprisehttp://www.earth.nasa.gov I

NASA Jet Propulsion Laboratoryhttp://www.jpl.nasa.gov/ .,

NOAA Home Page

http://www.noaa.gov/

RADARSAT

http://radarsat.space.gc.ca/

Remote Sensing Linkshttp://www.cent.org/ geo 121 focus5e.htm and

http://www.ersi.bc.ca/links.htmi

Sea WiFS Project-Homepagehttp:// seawifs.gsfc.nasa.gov ISEA WIFS.html

Shuttle Radar Topogrcrphy Mission (SRTM)

http://www-radar.jpl.nasa.gov/srtm/

SIR-C Precision Data

http://edcwww.cr.usgs.gov/landdaac/sir-c/precision/ precision.html

Space Imaging-IKONOShttp://www.spaceimagingeurope.com/

Spot Imagehttp://www.spotimage.fr/

TERRA (EOS AM -1 )http:// terra.nasa.gov /

TOPEX/Poseidonhttp://topex-http:/ /www.jpl.nasa.gov /science/science.html

USGS Global Land Information System (GLIS)http://edcwww.cr.usgs.gov/webglis/ '

USGS Spectral Laboratory

http://speclab.cr.usgs.gov/

Virtual Field Course

http://www.geog.le.ac.uk/vfc/index_2.html

WWW Virtual Library: Remote Sensing

http://www.vtt.fi/aut/rs/virtual/

X-SAR data at DLR, Germany

http://www.op. dLr.de/n~-hf/SRL.html

This Glossary defines many terms and concepts withwhich readers may not be familiar and which crop up inthe book and in other remote-sensing literature. It shouldbe used in conjunction with the Index, which showswhere the terms are defined more fully and in context.

absolute temperature Temperature measured on theKelvin scale, the base of which is absolute zero(~273°C). DoC is expressed as 273 K.

absorptance A measure of the ability of a materialto absorb electromagnetic (EM) energy at a specificwavelength. .

absorption band A range of wavelengths int~ electro-magnetic (EM) spectrum where a material absorbs EMenergy incident upon it, often resulting from variousenergy-matter transitions.

achromatic line A line in red, green, blue image displayspace,n:mnfug at 450 to the axes. A three-channel com-bination from most images tends to be correlated, andso the multispectral data will plot close to this line, andthe image will not be strongly coloured unless specialimage processing techniques (such as decorrelationstretching) are applied.

achromatic vision The perception by the human eye ofchanges in brightness, often used to describe the per-ceptionof monochrome or black and white scenes.

active remote sensing A system based on the illumina-tion of a scene by artificial radiation and the collec-

, tion of the reflected energy returned to the system.

Examples are radar artd systems using lasers.acuity A measure of human ability to perceive spatial

variations in a scene..Jt varies with the spatial frequency,shape and contrast of the variations, and depends onwhether the scene is coloured ormonochrome.

additive primary colours The spectral colours red,green and blue, which Thomas Young (1773-1829) dis-covered to be capable of reproducingaRother colourswhen mixed by projection through filters, and each ofwhich Cannot be produced by mixtures of the other two.

AIS Airborne Imaging Spectrometer.albedo The fraction of the total electromagnetic (EM)

energy incident on a material that is reflected in alldirections.

analogue image An image where the continuous varia-tion inihe property being sensed is represented by acontinuous variation in image tone. In a photographthis is achieved directly by the grains of photosensit-ive chemicals in the film, in an electronic scanrier, theresponse in say millivolts is transformed to a displayon a cathode-ray tUbe where it may be photographed.

artefact A feature on an image that is produced by theoptics of the system or by digital image processing,and sometimes masquerades as a real feature.

ASTER Advanced Spacebome Thermal Emission andReflection Radiometer to be carried by EOS-IAM(Terra), launched in December 1999. ASTER willacquire images in 14 spectral bands in the reflectedand thermally emitled regions, selected particularlyfor geological applications.

atmospheric shimmer An effect produced by the move-ment of masses of air with different refractive indices,which is seen most easily in the twinkling of stars.Shim1ner results in blurring on remotely sensed images,and is the ultimate control over the resolution of any

system.atmospheric window A range of electromagnetic (EM)

wavelengths where radiation can pass through theEarth's atmosphere with relatively litlle attenuation.

atlribute Property assigned to a line, polygon or pointin vector-format data. Several may be combined togive a description of a vector that is useful in a geo-graphical information system (GIS).

A VHRR Advanced Very High Resolution Radiometer,a multispectral imaging system carried by the TIROS-NOAA series of meteorological satellites.

A VIRIS Airborne Visible/l11frared Imaging Spectro-meter.

azimuth In general this is the compass bearing of aline given in degrees clockwise from north. In radarterminology it refers to the direction at right angles tothe radar propagation direction, which is parallel tothe ground track in a sideways-looking system.

band In remote sensing, a range of wavelengths fromwhich data are gathered by a recording device.

band-pass filter Spatial filter designed to accentuateonly features with frequencies between set limits.

band-stop filter Spatial filter designed to selectivelyremove features with frequencies between set limits.

bin One of a series of equal intervals in a range of data,most commonly employed to describe the divisions ina histogram.

binary A numerical system using the base 2. Ex~mplesare 0 = 0, 1 = 20 = 1, 10 = 21 = 2, 11 = 21 +20 = 3.

bit An abbrevi~tion of binary digit, which refers to aneXponent of 2. A bit is represented by 0 or 1 for 'on' or'off' in a digital computer.

blackbody A perfect radiator and absorber of electro-magnetic (EM) energy, where all incident energy isabsorbed and the energy radiated from the body ata particular temperature is ~t the maximum possible~

266

Glossary 267

produced by the .Commission Intemationale de

l'Eclairage.classification The process of assigning individual pixels

of a multispectral image to categories, generally basedon spectral characteristics of known parts of a scene.

Coastal Zone Colour Scanner (CZCS) A multispectralimaging system carried by the Nimbus series of meteo-

rologicalsatellites.coherent radiation Electromagnetic radiation who.se

waves are equal in length and are in phase, so thatwaves at different points in space act in unison, as in alaser and synthetic aperture radar.

cones Receptors in the retina that are sensitive to colour.There are cones sensitive to red, green and blue com-ponents of light.

context The known environment of a particular featureon an image.

contrast The ratio between the energy emitted or refl-ected by an object and its immediate surroundings.

contrast stretching Expanding a measured range ofdigital numbers (DN) in an image to a larger range, toimprove the contrast of the image and its component

parts.convolution Arithmetic combination between two

arrays of numbers, generally implemented by a smallbox filter whose weights multiply with DN in thearray overlain by it, the sum of the pixel-by-pixel oper-ations being substituted for the pixel at the centre ofthe area over, lain by the box. The convolved image isgenerated by systematically passing the bQx filter overevery pixel in the scene.

comer reflector A cavity formed by three planarreflective surfaces intersecting at right angles, whichreturns radar directly back to its source.

crystal-field effect Dependence of spectral features onco-ordination of ions in a mineral molecule. Most com-monly shown by minerals dominated by ferric iron (Fe3t-).

cut off The digital number in the histogram of a digitalimage that is set to zero during contrast stretching.Usually this is a value below which atmospheric scat-tering makes a major contribution.

dark-pixel correction Use of the darkest pixels in ascene, either total shadows or deep, clear water, to ap-proximate the contribution by atmospheric scattering.

decorrelation stretching A way of making a three bandmultispectral image more colour£ul by stretching thedata distribution along axes related to the naturalelongation of the data distribution (the principalcomponent directions). After stretching, the data aredisplayed on the original red, green and blue axes,thus retaining the original, straightforward colourrelationships. Sometimes abbreviated to D-stretch.

density slicing The process of converting the full rangeof data into a series of intervals or slices, each of whichexpresses a range in the data.

rate for each wavelength, as governed by the Stefan-Boltzmann Law. No natural material has these idealproperties, although water is a close approximation.

blind spot The point of entry of the optic nerve to theretina where no radiation is detected by the eye.

bond-stretching and bond-bending transitionsProduce spectral features due to distortion of bondsin molecules, particularly the metal-hydroxyl bondsAI-OH and Mg-OH.

box classification A classification technique in whichthe decision boundary around the training set is a rect-angular box. The term parallelepiped classification is

synonymous.box filter Rectangular array of numbers used in con-

volution with an image to implement some kind of

spatial-frequency filtering.brightness Often used loosely to refer ei!her to the

amount of radiation coming from a surface or to theappearance of a region within an image.

byte A group of 8 bits of digital data in binary form.A byte represents digital numbers up to 255, and isthe standard adopted by most digital remote-sensingsystems where the range of energies is coded from 0to 255.

canonical analysis A variant of principal componentanalysis that uses a selected range of the input vari-ables, usually relating to specific surface materials, togenerate the statistics used in the selection of principalcomponent axes.

cell assemblies The linked receptors, retinal neurons andneural cells in the visual cortex of the brain that enableinteraction between perception and past experience.

change detection A range of techniques to identifyimportant differences between scenes recorded ondifferent dates.

channel The range of wavelengths recorded by a singledetector to form an image. A multispectral image isrecorded in several channels simultaneously. Theterm also can be used to refer to a synthetic channel(i.e. not a simple spe~tral band as originally recorded),such as one created by ratioing or principal compon-ents transformation. See also band..

charge transfer A mechanism of spectral absorption ofEMR, when an electron is transferred across a chemi-cal bond.

charge-coupled device (CCD) A light-sensitive capa-citor whose charge is proportional to the intensityof illumination. They are able to be charged anddischarged very quickly, and are used in push-broom devices, spectroradiometers and modernvideo cameras.

chromatic vision The perception by the human eye ofchanges in hue.

CIE colour system Tristimulus colour theory basedon mathematically derived chromaticity co-ordinates,

268 Glossary

emittance A term for th~radiant flux of energy per unitarea emitted by a body. Now obsolete.

emitted region Part of the spectrum between 3 ~m and1 cm dominated by energy emitted by bodies aboveabsolute zero.

EOS Earth Observing System, a proposed multinationalseries of heavily instrumented remote sensing satel-lites to be deployed in the decade beginning in themid-1990s. It probably will include American (NASA),European (ESA) and Japanese polar orbiting platformsand a manned space station.

EOSA T Earth Observation Satellite Company, basedin Lanham, Maryland" USA, A private company con-tracted to the US Government since September 1985to market Landsat d'ata and develop replacements forthe Landsat system.

equalization stretch C::ontrast stretch that forces thecumulative histogram of DN to a straight line, therebyspacing bins according to their frequencies. Thisachieves an equal population density of pixels alongthe DN axis. Also known as equipopulation or rampstretch.

equatorial orbit An orbit of a satellite around the Earthin which1he orbital plane makes an angle of less than450 with1he Equator.

equipopulation stretch See equalization stretch.EROS Earth Resources Observation System, based at

the EROS Data Center, Sioux Falls, South Dakota,USA. Administered by the US Geological Survey,it forms an important source of image data fromLandsat-I, -2 and -3 and for Landsat-4 and -5 data upto September 1985, as well as airborne data for theUSA.

ERS-l Earth Remote Sensing Satellite-I, a European(ESA) satellite scheduled for launch in 1992, targetedprincipally on marine applications by the use of activeand passive microwave techniques.

ESA European Space Agency, based in Paris. A con-sortium between several European states for thedevelopment of space science, including the launchof remote.;sensing satellites.

exitance The radiant flux density (of electromagneticradiation) leaving a surface;

exponential stretch Contrast' stretch u~ing ac look-uptable that is an exponential curve. Accentuates con-trast in light regions~

false-colour image A colour image where parts ofthe non-visible electromagnetic (EM) spectrum areexpressed as one or more of the red, green and bluecomponents, so that the c010urs produced by theEarth's surface do not correspond to normal visualexperience. Also called a false-colour composite (FCC).The most commonly seen false-colour images displaythe very-near-inf:rared region as red, red as green andgreen as blue.

depression angle In radar usage this is the anglebetween the horizontal plane passing through -theantenna and the line connecting the antenna to the tar-get. It is easily confused with the look angle.

depth of field The range of distances in focus using a lens.diffuse reflection Reflection of light (or other electro-

magnetic radiation) from a surface approximatelyevenly in all directions, as opposed to specular(mirror-:-like) reflection. A diffuse reflector has a sur-face that is rough on the scale of the wavelength ofelectromagnetic radiation concerned.

digital image An image where the property beingmeasured has been converted from a continuousrange of analogue values to a range expressed by afinite number of integers, usually recorded as binarycodes from 0 to 255, or as one byte.

digital number (DN) The value ofa-variable/ecor.dedfor each pixel in an image as a binary integer, usu-ally in the range 0-255. An alternative term used insome texts is brightness value (BV). The plural of itsacronym is ON.

directional filter A spatial-frequency filter that en-hances features in an image in selected directions.

D LR German Space Agency .Doppler shift A change in the observed frequency of

electromagnetic (EM) or other waves caused by therelative motion between source and detector. Usedprincipally in the generation of synthetic-apertureradar images.

edge A boundary in an image between areas with dif-ferent tones.

edge enhancement. The process of increasing the con-trast between adjacent areas with different tones onan image.

eigenvalue Proportion of the variance in multispectraldata that is redistributed to ~ principal component.

eigenvector Crudely; the contribution of an input bandto a principal component.

electromagnetic radiation (EMR) Energy transported-by the propagation of disturbances in the electric and

magnetic fields. Detection andmeasurememofnaturalor artificial EMR is the basis for remote sensing.

electronic transition A mechanism of spectral absorptionof electromagnetic radiation (EMR), when an electronis moved from a lower orbit to a higher, as the result ofabsorption of a photon. Also a mechanism of genera-tion of EMR, when an electron goes from a higher orbitto a lower, and emits a photon in the process.

emissivity A measure of how well a surface emits, orradiates, electromagnetic radiation (EMR) thermally;defined as the ratio between the thermal exitancefrom the surface and the thermal exitance from a blackbody (perfect emitter) at the same temperature. Ab.l~ckbody therefore has an emissivity of 1 and naturalmaterials range from 0 to 1.

Glossary 269

geostationary orbit.An orbit at 41000 km inthedirec-tion of the Earth's rotation, which matches speed sothat a satellite remains over a fixed point on the Earth'ssurface.

greybody A material with constant emissivity less than1.0 for all wavelengths.

grid format The result of interpolation from values of avariable measured at irregularly distributed points, oralong survey lines, to values referring to square cellsin a rectangular array. It forms a step in the process ofcontouring data, but also can be used as the basis for araster format to be displayed and analysed digitallyafter the values have been rescaledio the 0-255 range.

ground-control point A point in two dimensions that iscommon to both an image-and a topographic map,and can be represented by (x,y) co-ordinates based onthe map's cartographic projection and grid system;Used in geometric correction of distorted images, andtheir registration to a convenient map projection. ,-

ground range Distance on the Earth's surface from theposition of an object to that directly. below a radarantenna.

ground truth Observations made on ihe ground -thatcan be used to verify the interpretation of an image.Ground truth includes field-based identification ofsurface types and iheir distribution (to,help performand verify classification). Comparable observations atsea (e.g. phytoplankton concentration, wave height,etc.) are usually known as sea truth.

harmonic Secondary spectral feature related to vibra-tional transitions at wavelengths related to that of thefundamental (analogous to those in sound). Alsoknown as an overtone.

heat capacity A measure of how much energy it takes towarm upa given mass of a material through 1°C.

Heat Capacity Mapping Mission (HCMM) NASAsatellite launched in 1978 to observe thermalprop~r-ties of rocks and soils. Remained in orbit foronly:a fewmonths.

high-pass filter A spatial filter that selectively. enhancescontrast variations with high spatial frequencies inan image. It improves the sharpness of images and is amethod of edge enhancement., -.

HIRIS High ResolutionlmagirigSpectrometer, possiblyto be carried by a future platform of the Earth

Observing System.histogram A means of expressing the frequency of

occurrence of values in a -data set within a series ofequal ranges or bins, the height of each bin repres-enting the frequency at which values in the data setfall within the chosen range. A cumulative histogramexpresses the frequency of all values falling within abin and lower in the range. A smooth curve derivedmathematically from a histogram is termed the prob-ability density function (PDF).

flat iron A term used to describe roughly triangularfeatures seen on images, derived from their similarityto the household appliance used for smoothing cloth.On aerial photographs they result from dipping rocklayers. On radar images they may be an artefact causedby high relief and steep depression angle.

flat-field calibration Use of spectrally featureless materi-als, such as halite, to judge atmospheric effects in hyper-spectral data. The observed spectrum of spectrally flatmaterial is used to divide those of other pixels, therebymaking them roughly equivalent to reflectance spectradetermined under laboratory conditions.

fluorescence A property of some materials where electro-magnetic (EM) energy of one wavelength is absorbedand then re-emittedat a longer wa¥elength.

foreshortening A distortion in radar images causing thelengths of slopes facing the antenna to c\ppear shorteron the image than on the ground. It is produced whenradar wavefronts are steeper than the topographicslope.

Fourier synthesis and analysis Construction and dis-assembly, respectively, of a complex waveform in termsof sine waves with different wavelengths and ampli-tudes. Used in frequency-domain filtering of images.

fovea The region around that point on the retina inter-sected by the eye's optic axis, where receptors are mostdensely packed. It is the most sensitive part of the retina.

frequency The number of waves that pass a referencepoint in unit time, usually 1 s.

f-stop Ratio between focal length and aperture diameterof a lens.

fundamental Refers to the main spectral feature pro-duced by a vibrational transition.

gamma-ray spectrometer Based on light emission by acrystal when exposed to gamma radiation.. By tuningthe system to the energy levels of gamma radiationemitted by 4OK, 23% and 238U, it discriminates betweenand measures the radiation emitted by potassium,thorium and uranium contained by surface materials.

Gaussian stretch See normalized stretch.geographical information system (GIS) A data handling

and analysis system based on sets of data distributedspatially in two dimensions. The data sets may be maporientated, when they comprise qualitative attributesof an area recorded as, lines, points and areas in vectorformat, or image orientated, when the data are quanti-tative attributes referring to cells in a rectangular gridin raster format.

georeferencing Correlation of positions on images' withgeographic co-ordinates, using ground-control points.Sometimes referred to as geocoding.

Geoscan Remote-sensing device operated by Carr-Boyd Pty of Australia, which incorporates narrow-band spectral coverage., from both the reflected andemitted regions.

270 Glossary

URV Haute Resolution Visible, the visible and very-near-infrared pushbroom imaging system carriedby SPOT; 20 m pixels in colour and .10 m pixels inpanchromatic mode.

hue One of the three characteristics that define the colourof a pixel in an image in terms of intensity, hue andsaturation. Hue is a measure of the relative amountsof red, green and blue contributing to the colour.

Hyperion Imaging spectrometer with 7.5 km swathwidth to be carried on long-delayed NASA satellite(EO-I); part of NASA's latest strategy of low-cost

imaging systems.hyperspectral data -Often used for multi-channel, nar-

row band data captured by imaging spectrometers.image dissection The breaking down of a continuous

scene into discrete spatial elements, either by thereceptors on the retina, or in the process of capturingthe image artificially.

image striping A defect produced in line-scannerand pushbroom imaging devices produced by non-uniform response of a single detector, or amongst abank of detectors. In a line-scan image the stripes areperpendicular to flight direction, but parallel to it in a

pushbroom image.imaging radar A radar that constructs an image of the

terrain or sea surface by complex processing of theechoes reflected back to the antenna.

imaging spectrometer A sophisticated remote sensinginstrument that records an image in a large number ofnarrow spectral channels.

incidence angle The angle between the surface and anincident ray of electromagnetic (EM) radiation-mostusually referring to radar.

infrared radiation Electromagnetic radiation (EMR) ofwavelength between 0.7 ~m and 100 ~m, comprisingreflected infrared 0.7-3.0 ~m and thermal infrared3.0-100~.

instantaneous field of view (IFOV) The solid anglethrough which a detector is sensitive to radiation.It varies with the intensity of the radiation, the timeover which radiation is gathered, and forms one limitto the resolution of an imaging system.

intensity A measure of the energy reflected or emittedby a surface. Specifically, the overall brightness of acolour in the IUS colour system.

intensity-hue-saturation (IUS) colour system Based ona cone whose axis has equal values for red, green andblue (the grey axis). Intensity of a colour is measuredby projection of its RGB co-ordinates on to the greyaxis. A colour's hue is the angular position (from0-360°) of its RGB co-ordinates around the circum-ference of a circular section through the cone, knownas a colour wheel. Saturation measures the distanceof a colour's'RGB co-ordinates radially from the greyaxis.

intensity-hue-saturatioa (IUS) processing After con-verting a colour image from Cartesian RGB colouraxes to the conical IUS colour system, its intensity,saturation or hue can be modified or substituted. Mostcommonly, this is used in enhancing the visibility ofcolours in an image by stretching the intensity andsaturation, without altering the hue. This maintainsthe original relationship between colour and spectral

properties.interferometric radar (InSAR) See radar interferometry.irradiance The radiant flux density falling on a surface.IRS-l Indian Remote Sensing satellite launched in 1988,

which carries two pushbroom scanners covering fourbands in the visible and VNIR, one with 73 m resolu-tion the other <:reating pixels half that size. Most dataare restricted to the Indian subcontinent.

JERS-l Japanese Earth Resources Satellite, launchedin 1992, when it was renamed 'Fuyo-1' (the acronymJERS-1 seems to have stuck outside Japan). JERS-l car-ries four sensors. Pushbroom sensors in the reflectedregion produce two bands in the visible range, two inthe VNIR, one being a forward-pointing band to pro-duce streoscopic images, and four bands in the short-wavelength infrared (SWIR). Each band has a pixel sizeof 18.3 by 24.2 m The fourth sensor is an L-band SARsystem with a ground resolution of 18 m. Data are avail-able globally through onboard recording facilities.

Kirchhoff's Law Simple relationship between reflect-ance and emissivity for good emitters of radiation,whose emissivities and absorptivities are equal.

Lambertian reflector A perfectly diffuse reflector, whichreflects incident electromagnetic radiation (EMR)equally in all directions.

Landsat A series of remote-sensing satellites in Sun-synchronous polar orbit that began in 1972. Initiallyadministered by NASA, then NOAA, and since 1985by EOSAT. Carried MSS, RBV and TM systems.

laser Light artificially stimulated electromagnetic radi-ation. A beam of coherent radiation with a single

wavelength.layover A distortion in radar images when the angle of

surface slope is greater than that of radar wavefronts.The base of a slope reflects radar after the top of theslope, and because radar images express the distanceto the side in terms of time, the top appears closer tothe platform than the base on an image, giving theimpression of an overhanging slope.

Lewis Satellite carrying a 384-band imaging spectro-meter, which sadly failed to achieve orbit in 1998.

line drop out The loss of data from a scan line caused bymalfunction of one of the detectors in a line-scanner.

line-scanner An imaging device that uses a mirror tosweep the ground surface normal to the flight pathof the platform. An image is built up as a strip com-prising lines of data.

Glossary 271

mixed pixel A pixel in which the digital number (ON)represents the average energy reflected or emittedby several types of surface present within the areathat it represents on the ground. Sometimes calleda mixel.

modulation transfer function (MTF) A measure of thesensitivity of an imaging system to spatial variationsin contrast.

MOMS Modular Optical-electronic Multispectral Scan-ner, a German imaging system carried experimentallyby Shuttle missions, and which is based on a push-broom system.

MSS Multispectral Scanner, a remote sensing instru-ment carried on all the members of the Landsat series,recording pixels about 80 m across.

multispectral scanner A line-scanner that simultaneouslyrecords image data from a scene in several differentwavebands. Most commonly applied to the four-channel system with 80 m resolution carried by theLandsat series of satellites.

multispectral space The multidimensional space (some-times known as feature space) represented mathe-matically by plotting the digital number (ON) in eachspectral band of an image on an orthogonal set of axes.If there are six spectral bands under considex:ation(for example when using the six reflected bands inLandsat TM data), then the multispectral space issix-dimensional.

Munsell colour system Arbitrary divisioR of coloursbased on hue, value and chroma.

nadir The point on the ground vertically beneath thecentre of a remote-sensing system.

NASA National Aeronautics and Space Administration,USA.

NASOA National Space Development Agency, Japan.near-infrared (NIR) The shorter wavelength range of the

infrared region of the electromagnetic (EM) spectrum,from 0.7 to 2.5 J.!m. It is often divided into the very-near-infrared (VNIR) covering the range accessibleto photographic emulsions (0.7-1.0 ~), and the short-wavelength infrared (SWIR) covering the remainderof the NIR atmospheric window from 1.0to 2.5 J.!m.

Nimbus-7 An experimental polar orbiting satellitelaunched in: 1978, notable for carrying the CoastalZone Colour Scanner (CZCS).

NOAA National Oceanic and Atmospheric Admin-istration, USA.

noise Random or regular artefacts in data, whichdegrade their information-bearing quality and are dueto defects in the recording device.

nonselective scattering The scattering of electromag-netic (EM) energy by particles in the atmosphere thatare much larger than the wavelengths of the energy,and which causes all wavelengths to be scatteredequally.

linear stretch Contrast stretch that redistributes mini-mum and maximum DN in raw data linearly between0 and 255 respectively, thereby stretching all inter-mediate DN over the full range of brightnesses.

logarithmic stretch Contrast stretch using a look-uptable that is a logarithmic curve. Accentuates contrastin dark regions.

look angle The angle between the vertical plane con-taining a radar antenna and the direction of radarpropagation. Complementary to the depression angle.

look direction The direction in which pulses of radarare transmitted.

look-up table (LUT) A mathematical formula used toconvert one distribution of data to another,most con-veniently remembered as a conversion graph.

luminance A measure of the luminous intensity of lightemitted by a source in a particular directi2n.

Mach band An optical illusion of dark and light fringeswithin adjacent areas of contrasted tone. It is a psycho-physiological phenomenon that aids human detectionof boundaries or edges.

mask A geographical area to be separated from otherareas for some purpose, usually to perform some image-processing or GIS operation without influencing areasoutside the mask. A Bimple example would be bodiesof water on which contrast enhancement is to be per-formed, either to render all in a single tone or to dis-play subtle differences, without affecting land areas.

maximum-likelihood classification A classification tech-nique in which unknown pixels are assigned to a classaccording to statistically defined probabilities.

median filter A spatial filter that substitutes the medianvalue of digital number (DN) from surrounding pixelsfor that recorded at an individual pixel. It is useful forremoving random noise.

Meteosat The European (ESA) member of the GOESgeostationary weather satellite series.

microwave region That part of the spectrum beyond1mm.

microwaves Electromagnetic radiation (EMR) of wave-length between 100~m and 1 m.

mid-infrared (MIR) The range of electromagnetic (EM)wavelengths from 8to 14 ~m dominated by emissionof thermally generated radiation from materials. Alsoknown as thermal infrared.

Mie scattering The scattering of electromagnetic (EM)energy by particles in the atmosphere with comparabledimensions to the wavelength involved.

minimum-distance-to-means classification A classifica-tion technique in which unknown pixels are assignedto whichever class whose mean they are closest to, in

multispectral space.minus-blue photograph A panchromatic black and white

photograph from which the blue part of the visiblerange has been removed using a yellow filter.

272 Glossary

piece-wise stretch ContMst stretch that assigns severalranges of ON in raw data to different output ranges,each being transformed linearly.

pixel A single sample of data in a digital image, havingboth a spatial attribute-its position in the image anda usually rectangular dimension on the ground-anda spectral attribute-the intensity of the response in aparticular waveband for that position (ON). A con-traction of picture element.

Planck's Law An expression for the variation of emitt-ance of a blackbody at a particular temperature as afunction of wavelength.

point spread function (PSF) The image of a point sourceof radiation, such as a star, collected by an imagingdevice. A measure of the spatial fidelity of the device.polar

orbit An orbit that passes close to the poles, therebyenabling a satellite to pass over most of the surface,except the immediate vicinity of the poles themselves.

polarized radiation Electromagnetic radiation in whichthe electrical field vector is contained in a single plane,instead of having random orientation relative to the pro-pagation vector. Most commonly refers to radar images.

principal component analysis The analysis of covariancein a multiple data set so that the data can be pro-jected as additive combinations on to new axes, whichexpress different kinds of correlation among the data.

principal components Directions, or axes, in multispec-tral space, defined by approximating the data from awhole image, or selected area in an image, to a multi-dimensional ellipsoid. The 'first principal componentdirection is parallel to the longest axis of this ellipsoid,and the other principal components are orthogonal toit. The first principal component of the image is thedata displayed relative to an axis parallel with the firstprincipal component direction.

principal point The centre of an aerial photograph.probability density function (PDF) A function indicat-

ing the relative frequency with which any measure-ment may be expected to occur. In remote sensing it isrepresented by the histogram of digital numbers (DN)in one band for a scene.

pushbroom system An imaging device consisting ofa fixed, linear array of many sensors, usually CCDs,which is swept across an area by the motion of theplatform, thereby building up an image. It relies onsensors whose response and reading is nearly instan-taneously, so that the image swath can be segmentedinto pixels representing small dimensions on theground.

quantum (pl. quanta) The elementary quantity of elec-tromagnetic (EM) energy that is transmitted by a par-ticular wavelength. According to the quantum theory,EM radiation is emitted, transmitted and absorbed asnumbers of quanta, the energy of each quantum beinga simple function of the frequency of the radiation.

nonspectral hue A hue which is not present in thespectrum of colours produced by the analysis of whitelight by a prism or diffraction grating. Examples arebrown, magenta and pastel shades.

normalized difference vegetation index (NOVI) Avegetation index that calculates how far a pixel plotsfrom the soil line, using the relationship: NOVI =(VNIR'- red) / (VNIR + red).

normalized stretch Contrast stretch in which a range ofraw ON are transformed by forcing their histogram tothe shape of a normal or Gaussian distribution thatspans the full 0-255 range.

opponents colour theory Representation of colourin terms of three components, white-black, blue-yellow and red-green.

orbital period The time taken by a satellite to completean orbit. .'

orthophotograph A vertical aerial photograph fromwhich the distortions due to varying elevation, tilt andsurface topography have been removed, so that it rep-resents every object as if viewed directly from above,as in a map.

orthophotoscope An optical-electronic device whichconverts a normal vertical aerial photograph to an

orthophotograph.overtone See harmonic.panchromatic 'Covering the range of wavelengths in

the visible (and sometimes the VNIR) part of thespectrum.

parallax The apparent change in position of an objectrelative to another when it is viewed from differentpositions. It forms the basis of stereopsis.

parallax difference The difference in the distances onoverlapping vertical photographs between two points,which represent two locations on the ground withdiff~rent elevations.

parallelepiped classification A classification techniquein which the decision boundary around the trainingset is a rectangular box. The term box classification issynonymous.

passive microwaves Radiation in the 1 mmto 1 mrangeemitted naturally by all materials above absolute zero.

passive remote sensing The capture of images repres-enting the reflection or emission of electromagnetic(EM) radiation that has a natural source.

pattern A regular assemblage of tone and texture onan image. Often refers to drainage systems.

photographic infrared That part of the near-infrared (orreflected infrared) spectrum to which photographicfilms respond, i.e. wavelengths between 0.7 11m andabout 1.0 11m. Virtually synonymous with very-near-infrared.

photon A quantum of electromagnetic (EM) energy.photopic vision Vision under conditions of bright illu-

mination, when both rods and cones are employed.

Glossary 273

quantum theory A theory about electromagnetic radi-ation (EMR) which dictates that there is a minimumamount of energy that can exist (or be emitted) atany particular wavelength. This m~nimum amount isa quantum, or photon.

radar The acronym for radio detection and ranging,which uses pulses of artificial electromagnetic (EM)radiation in the 1 mm to 1 m range to locate objectswhich reflect the radiation. The position of the objectis a function of the time that a pulse takef? to reach itand return to the antenna.

radar altimeter A non-imaging device that recordsthe time of radar returns from vertically beneath aplatform to estimate the distance to and hence theelevation of the surface. Carried by Seasat and theESA ERS-l platforms.

radar cross-section A ll)easure of the int~1Jsity of back-scattered radar energy from a point target. Expressedas the area of ahypoth~tical surface that scatters radarequally in all directions and which would return thesaI1)eenergy to the antenna.

radar interferometry Use of dual antennae to recordbackscattered radar energy, so that the antenna sep-aration is analogous to the distance between stereo-scopic aerial photographs.. Radiation travels differentdistances from each point on the ground to each~ntenna, so that when th~ signals are combined theyinterfere. Using the geometry of the radar set-up andthe known wavelength of the radiation the interfer-ence record is converted to topographic elevation ofeach point. This produces a digital elevation model ofthe overflown surface.

radar scattering coefficient A measure of the back-scattered radar energy from a target with a large area.Expressed as the averC1ge radar cross-s~ction per unitarea in decibels (dB). It is the fundamental measureof the radar properties of a surface.

radar scatterometer A non-imaging device that recordsradar energy back-scattered from terrain.-as a functionof depression angle.

radial relief displacement The tendenc;;y of verticalobjects to appear to lean radially away frOll) the centreof a vertical aerial photogr~ph. Caused by the conicalfield of view of the camera lens.

radiance The radiant flux density falling on a surfacemeasured per solid angle. A useful c<?ncept becalJ-Sewe rarely measure the radiant flux density leaving asurface in all directions; instead we collect it with adetector responding to electromagnetic radiation (EMR)arriving at the detector from a finite solid angle.

radiant flux The power incident on or leaving a body inthe form of electromagnetic radiation (EMR).

radiant flux density The power inl::ident on or leaving abody in the form of electromagnetic radiation (EMR)per unit area of surfal::e.

ramp stretch See eqqalization stretch.range In radar usage this is the distance in the direc-

tion of radar propagation, usually to the side of theplatform in an imaging radar system. The slant rangeis the direct distance from the antenna to the object,whereas the distance from the- ground track of theplatform to the object is termed the ground range.

range resolution In a radar image, the resolution in thedown-range direction, i.e. at right angles tp the flightline, as distinct from azimuth resolution. Rangeresolution depends on the pulse length and the inci-dence angle, and is not ma<;ie worse by flying at highaltitude.

raster The scanned and illuminated area of a videodisplay, produced by a modulated beamQf electronssweeping the phosphorescent screen line-by line fromtop to bottom at a regular rate of repetition.

r~ster format A means ;of representing spatial data inthe form of a grid of digital numbers (DN), each lineof which can be used to modulate the lines _of a videoraster.

ratio The digital number (DN) of one band of a multi-spectral image divided by the DN ~f another bandusually rescaled on to an 8-bit scale. This ~mphasizesthe relative differences between the two channels andignores brightness/intensity / albedo.

Rayleigh criterion A way of quantifying surf&c~ rough-ness with r~spect to wavelength of ~l~ctromagneticradiation (EMR), to determin~ wheth~r the surfacewill act as a specular (s~ooth) r~flector or as a diffuse(rough) reflector. A surface can be consid~redrough ifthe root mean square height of surface irreg:ulariti~s isgr~ater than on~ ~ighth of th~ wavel~ngth of EMRdivided by th~ cosine of the incidence angle;.

Rayleigh scattering Selective scattering of light in .theatmosphere by particles that are small (:omparedwiththe wavelength of light.

RBV The Return-Beam Vidicon system aboard Landsat-3, which produced panchromatic digital images witha resolution of 40 m.

real-aperture radar An imaging radar system wherethe azimuth resolution is determined by the physicallength of th~ antenna; tb~ wavelength and the range.Also known as brute-force radar.

receptor Light-sensitive cell embedded in the eye'sretina. See rods and cones.

red edge The sharp in~rease in spectral reflectanceof healthy leafy vegetation, which Q~curs with increas-ing wavelength between about 700nm and 750 nmwavelength, i.e. in thered/very-near-:infrared part ofthe spectrum.

redundancy Information in an image that is either notrequired fox: interpretation or cannot be seen. Redund-ancy may be spatial or spectral. Also refers to multi-spectral data where the degree of correlation between

274 Glossary

bands is so high that one band contains virtually thesame information as all the bands.

reflectance/reflectivity The ratio of the electromagnetic(EM) energy reflected by a surface to that which fallsupon it. It may be qualified as spectral reflectance. Thesuffix -ance implies a property of a specific surface.The suffix -ity implies a property for a given material.

reflected infrared That part of the infrared spectrumthat, in terrestrial remote sensing, is mostly made upof reflected solar electromagnetic radiation (EMR), asopposed to the thermal infrared part of the spectrum,which has longer wavelengths. The reflected infraredregion is defined as wavelen~ from 0.7 ~m to 3.0 /:lIn,and is synonymous with the near-infrared.

reflected region That part of the spectrum between0~4-2.5~m where most energy is reflected fromsurfaces.

region of interest (ROI) See mask.resampling The calrulation of new digital numbers

(DN) for pixels created during geometric correction ofa digital scene, based on the values in the local areaaround the uncorrected pixels.

resolution A poorly defined term relating to the fidelityof an image to the spatial attributes of a scene. Itinvolves the instantaneous field of view (IFOV) andmodulation transfer function (MTF) of the imagingsystem, and depends on the contrast within the image,as well as on other factors. It is usually expressed asline pairs per lnillimetre of the most closely spacedlines that can be distinguished, and therefore dependson human vision, scale and viewing distance.

RGB colour system Based on the additive primarycolours red, green and blue. This is the system used byall video monitors in which various data are assignedto red, green and blue colour guns.

ringing Fringe-like artefacts produced at edges bysome forms of spatial-frequency filtering.

rods The receptors in the retina that are sensitive tobrightness variations.

rotational transition A mechanism of spectral absorptionof electromagnetic radiation (EMR), when a photon ofthe correct wavelength is absorbed to cause a gaseousmolecule to undergo a change in its rotational energystate.

saturation In colour theory it means the degree of mix-ture between a pure hue and neutral grey. Sometimesused to refer to the maximum brightness that can beassigned to a pixel on a display device, and corre-sponds to a digital number (DN) of 255.

scale In cartography this refers to the degree of reduc-tionfrom reality that is represented on a map, usuallyexpressed as a ratio (e.g. 1 : 250 000), a representat-ive fraction (e.g. 1/250 000) or an equivalence (e.g.1- cm = 2.5 kin). A large-scale map represents grounddimensions by larger cartographic dimensions than

a smaller-scale map. lliis is the opposite of the com-mon usage wherein a large-scale feature has largerdimensions than a smaller-scale feature.

scattering An atmospheric effect where electromagnetic(EM) radiation, usually of short visible wavelength,is propagated in all directions by the effects of gasmolecules and aerosols. See Rayleigh, Mie and non-selective scattering.

scene The area on the ground recorded by a photographor other image, including the atmospheric effects onthe radiation as it passes from its source, to the groundand back to the sensor.

scotopic vision Vision under conditions of low illu-mination, when only the rods a:re sensitive to light..Visual acuity under these conditions is highest in theblue part of the spectrum.

Seas at Polar-orbiting satellite launched in 1978 byNASA to monitor the oceans, using imaging radar anda radar altimeter. It survived for only a few months.

Sea-WIFS Sea Wide-Field Sensor-replacement remotesensing system for the CZCS carried by a satellitelaunched in 1997.

selective radiator A material that has emissivities thatvary with wavelength.

short wavelength infrared (SWIR) That part of the nearor reflected infrared region with wavelengths between1.0 ~m and 3.0 ~m. These wavelengths are too long toaffect infrared photographic film, hence the distinc-tion from VNIR.

Shuttle Radar Topography Mission (SRTM)Interferometric radar mission to map surface eleva-tion of the continents between 600 Nand 580 S, with aresolution up to 25 m. OEMs derived from the missionwill be publicly available fromNASA/JPL and theGerman Space Agency (OLR) from mid-2000.

signal-to-noise ratio (SIN): The ratio of the level of thesignal carrying rea1.information to that carrying spuri-ousinformation as a result of defects in the system.

signature In remote sensing this refers to the spectralproperties of a material or homogeneous area, mostusually expressed as the range of digital number (ON)in a number of spectral bands.

SIR Shuttle Imaging Radar, synthetic-aperture radarexperiments carried aboard the NASA Space Shuttlein 1981, 1984 and 1994.

slant range Direct distance from a radar antenna to apoint at the Earth's surface. Because radar uses energytransmitted obliquely downwards, slant range isdifferent from actual distance on the ground, orground range, and varies through the scene in thelook direction.

SMIRR Shuttle Multispectral Infrared Radiometer, anon-imaging spectroradiometer carried by the NASASpa:ce Shuttle covering 10 narrow wavebands in the0.5-2.4 ~m range.

Glossary 275

soil line A diagonal line at 450 on a plot of red digitalnumbers (ON) against very near-infrared (VNIR) ON.Bare soil plots close to this line, but leafy vegetationplots away from this line, having much higher VNIRreflectance.

spatial-frequency filtering The analysis of the spatialvariations in digital numbers (ON) of an image andthe separation or suppression of selected frequency

ranges.specific heat The ratio of the heat capacity of unit mass

of a material to the heat capacity of unit mass of water.spectral hue A hue that is present in the spectral range of

white light analysedbya prism or diffraction grating..spectral quantities A measure of electromagnetic radi-

ation (EMR) per unit wavelength, as opposed tooverall wavelengths. Thus we have spectral radiance,which is radiance per unit wavelength, aqd so on.

spectral radiant flux The power received or radiated bya body per unit area per unit wavelength. FOr con-venience, this is often quoted in Wm-2 ~m-l (spectralradiance is the same measure in terms of solid anglesas Wm-2 sr-l ~m-l).

spectral response A general term referring to the amountof light reflected by a surface at different wavelengths,as seen in an image. The spectral response ofa surfaceis controlled principally by its spectral reflectance.

spectroradiometer A device that measures the energyreflected or radiated by materials in narrow electro-magnetic (EM) wavebands.

specular reflection Reflection of electromagnetic radi-ation .(EMR) from a smooth surface, which behaveslike a mirror. A specular reflector has a surface thatis smooth on the scale of the wavelength of EMRconcerned. If the surface is rough on the scale of thewavelength of the EMR, it will behave as a diffusereflector.

SPOT Satellite Probatoire pour l'Observationde la Terre,a French satellite carrying two pushbroom imagingsystems, one for three wavebands in the visible andvery-near-infrared (VNIR) regions with 20 m resolu-tion, the other producing panchromatic images with10 m resolution. Each system comprises two deviceswhich are pointable so that off-nadir images are pos-sible, thereby allowing stereoptic viewing. Launchedin February 1986 with several later versions.

SRTM Shuttle Radar Topography Mission. Carried outin 14 days in February 2000, this Shuttle mission wasdedicated to deployment of two interferometric radarsystems (X- and C,.band) aimed at mapping topo-graphic elevation with a 30 m resolution for all landareas between 600 north and south. Data should beavailable by 2002.

Stefan-Boltzmann Law A radiation law stating that theenergy radiated by a blackbody is proportional to thefourth power of its absolute temperature.

stereopsis The ability for objects to be perceived in threedimensions as a result of the parallax differences pro-duced by the eye base.

stereoscope A binocular optical instrument used toview two images with overlapping fields that con-tain parallax differences, as a means of stimulating, stereopsis.

subtractive primary colours The colours cyan, magentaand yellow, the subtraction of which from whitelight in different proportions allows all colours to becreated.

Sun angle When considering the ill~mination of asurface by electromagnetic radiation (EMR) from theSun, the angle between a normal to the surface andthe direction of the Sun.

Sun-synchronous orbit A polar orbit where the satellitealways crosses the Equator at the same local solartime.

supervised classification A classificatioR techniquewhereby the human operator identifies training areason the image that are intended to be representativeexamples of each surface type.

synthetic-aperture radar (SAR) A radar imaging sys-tem in which high resolution in the azimuth directionis achieved by using the Doppler, shift of back-scattered waves to identify waves from ahead of andbehind the platform, thereby simulating a very longantenna.

TDRS Telemetry and Data Relay Satellites, launched bythe US Department of Defense into geostationaryorbit, primarily for military communications but usedfor data transfer by Landsat.

Terra The imaginative name chosen through a competi-tion for the first of NASA's Earth Observing Systemplatforms, EOS-IAM, after its lohg-delayed launch inDecember 1999.

texture The frequency of change 'and arrangement oftones on an image, often used to describe the aggregateappearance of different parts of the surface; but some-times used for the spacing of drainage elements.

thematic map An image on which are overlaid theresults of classification, usually by showing each classin a distin~tive colour. The term classified image is

synonymous.Thematic Mapper (TM) An imaging device carried

by Landsat-4 to -7, which records scenes in sevenwavebands, six in {he visible and near'-infrared (NIR)regions with a resolution of 30 m, and one in the mid-infrared (MIR) region With a resolution of 120m. TheAdvanced Thematic Mapper on Landsat 7 includes a15 m resolution panchromatic band and improvedresolution for its thermal sensor.

thermal capacity The ability of a material to store heat.thermal conductivity A measure of the rate at which

heat passes through a material.

276 Glossary

tl'lermal emission ~mission of electromagnetic radiation(EMR) frQm a material owing to thermal vibration asa result of its temperature.

thermal inertia A measure of the response of a materialto changes in its temperature. The apparent thermalinertia is calculated from the diurnal change in emittedthermal energy by a material

thermal infrared (TIR) That part of the spectrum :withwavelengths between 3.0 ~m and 100 ~m. These arethe wavelengths at which thermal emission is greatestfor surfaces at normal environmental temperatures.However, hot areas (fires, parts of volcanoes and soon) emit thermally at wavelengths shorter than thethermal infrared.

thermal vibration Vibration of atoms or molecules asa result of their temperature. Because these atomsof molecules contain moving electric char,ges, thisresults in thermal emission of electromagnetic radi-ation (EMR).

tie point A point on the ground that is common to twoimages. Several are used in the co-registration ofimages.

TIMS Thermal Infrared Multispectral Scanner, an air-borne device used by NASA to measure and createimages of mid-infrared (MIR) energies emitted by thesurface in six wavebands.

TIROS-N The most widely used series of Sun-synchronous meteorological satellites, carrying theA VHRR visible to thermal infrared imaging system.

tone Each distinguishable shade from black to whiteon a monochrome image, sometimes called grey-tone.

training area A sample of the Earth's surface with knownproperties, the statistics of the imaged data withinwhich are used to determine decision boundaries inclassification.

transmittance/transmissivity The ratio of the electro-magnetic (EM) energy passing through a material tothat falling on its exposed surface.

transpiration The production and emission of watervapour and oxygen by plants.

Triangular Irregular Network (TIN) A set of irregu-larly spaced three-dimensional nodes that represent asurface as a set of adjacent, conterminous triangles orfacets.

tristimulus colour theory A theory of colour relating allhues to the combined effects of three additive primarycolours corresponding to the sensitivities of the threetypes of cone on the retina.

ultraviolet That part of the spectrum with wavelengthsbetween about 0.5 nm and 400 nm. Not useful formost aspects of terrestrial remote sensing, because ofscattering in the atmosphere and absorption by ozone.

unsupervised classification A classification techniquein which classes are identified by a computer-drivensearch for clusters in multispectral data space.

variance A measure of t~e dispersion of the actual val-ues of a variable about its mean. It is the mean of thesquares of all the deviations from the mean value of arange of data.

vector format The expression of points, lines and areason a map by digitized Cartesian co-ordinates, direc-tions andyalues.

vegetation index A technique, usually involving ratio-ing, whereby channels from a multispectral imageare combined to show up variations in the amountof vegetation. A simple example is to divide very-near-infrared digital numbers (ON) by red ON.

Venn diagram Schematic representation of logical rela-tionships between sets of attributes.

vertical exaggeration The extent to which the verticalscale exceeds the horizontal scale in stereoptic view-ing of two overlapping images with parallax differ-ences. It is directly proportional to the base heightratio.

very-near-infrared (VNIR) The shortest wavelengthpart of the near-infrared (reflected infrared), withwavelengths between 0.7 Jlm and 1.0 Jlm. Virtuallysynonymous with photographic infrared.

vibrational transitions A mechanism of spectralabsorption of electromagnetic radiation (EMR), whenthe vibration of a chemical bond in a molecule ofcrystalline lattice changes from one state to another.Vibrations may occur either as stretching or bendingof a bond.

vidicon An imaging device based on a sheet oftransparent material in which electrical conductivityincreases with the intensity of electromagnetic (EM)radiation falling on it. The variation in conductivityacross the plate is measured by a sweeping electronbeam and converted into a video signal. Now replacedby cameras employing arrays of charge-coupleddevices (CCOs).

vignetting A gradual change in overall tone of an imagefrom the centre outwards, caused by the imagingdevice gathering less radiation from the peripheryof its field of view than from the centre. Most usu-ally associated with the radially increasing anglebetween a lens and the Earth's surface, and the cor-responding decrease in the light-gathering capacityof the lens.

visible radiation Electromagnetic radiation (EMR) inthat part of the electromagnetic spectrum that humaneyes respond to. It lies between the ultraviolet andthe infrared regions, with wavelengths from 400 nmto 700 nm.

VISSR Visible Infrared Spin-Scan Radiometer carriedby the GOES satellites.

visual dissonance The disturbing effect of seeing afamiliar object in an unfamiliar setting or in an unex-pected colour.

Glossary 277

volume scattering Scattering of electromagnetic (EM)radiation, usually radar, in the interior of a material. Itmay apply to a vegetation canopy or to the subsurfaceof soil.

wavelength The mean distance between maxima orminima of a periodic pattern. In the case of electro-magnetic (EM) radiation, it is the reciprocal of thefrequency multiplied by the velocity of light.

whiskbroom Imagin& device that sweeps radiantenergy from lines on the ground across a CCD arraymade up of several hundred devices, each designed torecord radiance in a narrow spectral band. Used in

imaging spectrometers.Wien's DisplacementLaw A radiation law stating that

the peak of energy emitted by a material shifts to shorterwavelengths as absolute temperature increases.

Index

ATS-1 (Applications Technology Satellite),57

augite, 10Australia, 89,182, Plate 8.4, Plate 8.6A VHRR see Advanced Very High

Resolution Radiometer (A VHRR)A VIRtS (Airborne Visible/Infrared Imaging

Spectrometer), 52-3,57,66,152

Note: page numbers in italics refer to figures; those in bold refer to tables.

absolute temperature, 3 Alaska, 217,220absorptance, 5,161 albedo,S, 130, 144, 145, 162, Plate 6.1absorption spectra,S constant, 172absorptivity, 165 estimation, 167achromatic axis, 131-2 and metal exploration, 237,Plate5.6active remote sensing, 4 and surface temperature, 163-4acuity, 21 and thermal inertia, 165,166,167,169,

and colour vision, 21,23 171spatial, 23-4 albite, 10

additive primary colours, 26,34-5,219, Algeria, 106,196Plate 2.1 alluvial fans, 112,114

Advanced Spaceborne Thermal Emission alluvium, 113,120,241, Plates 6.2-3, Plate 9.1and Reflection Radiometer (ASTER), radar images, 19365,130,145 AI-OHi>onds, 9,142,152,172

spectral bands, 58 altimetersspectrometers, 66 laser, 214thermal images, 173 radar, 214,231,238

Advanced Very High Resolution aluminium, energy level, 7Radiometer (AVHRR), 57-8 AmericanCivilWar(1861-5),50

images, 58 amphibolites, Plate 8.4sensors, 122 anaglyphs, 250spectral bands, 58 Anaimalai Hills (India), 103thermal images, 166-7,169 analogue-digital conversion, 122thermal inertia estimation, 171 Andes, 124,243,257

aeOlian landforms, 114-16 volcanoes, 103,108,109,111,112aerial photographs, 51-2 Andhra Pradesh (India), Plate 4.2

distortions, 36-7 AND operator, 222elevation data, 214 annular drainage, 72, 73geological applications, 66 anorthite, 10image interpretation, 36,68-121 antecedent drainage, 70, Plate 4.1infrared, 34 antennaeminus-blue, 34 radar, 43,46oblique, 36 Space Shuttle, 204principal point, 36,37 anticlines, 104,170radial-relief displacement, 36,37 apatite, 11stereoscopic, 37,68-121,249,251 Apollo missions, 50,55vertical,36,37 AppalachianPlateau(US),55,59,181,vignetting, 37 199

aerial photography apparent thermal inertia (An)applications, 50,51 estimation, 171flight lines, 37, 51-2, 249 images, 171historical background, 50-1 Applications Technology Satellite (A 1'5-1),overlap, 51 57sidelap, 51 Arctic (Canada), 113stereoscopic, Plate 9.1 Argentina, 111,112

aerial surveys, commissioning, 51 Arizona (US), 70,86,108,115, Plate 4.1aeromagnetic data see magnetic data artefacts, 96, 138aesthetics, and colour, 27-8 artificial intelligence, and patternAfghanistan, 59 recognition, 145,147Africa, 224,Plate9.2 aretes, 74age, and perceptual ability, 16 ash cones, 110agriculture, monitoring, 61 ASTER see Advanced Spaceborne Thermalairborne data, 51-3 Emission and Reflection RadiometerAirborne Imaging Spectrometer (AlS), 52 (ASTER)airborne radar, applications, 53 An see apparent thermal inertia (An)Airborne Visible/Infrared Imaging atmospheric correction, 125-7

Spectrometer (A VIRlS), 52-3,57,66, atmospheric effects, 6-7,140,228152, 236 gas absorption, 6

aircraft haze, 126advantages, 51 and ratioing, 143-4cameras, 32 scattering, 6-7, 17, 127, 128disadvantages, 51 shimmer, 7tilting, 37 atmospheric windows, 7,9

AlS (Airborne Imaging Spectrometer), 52 atoms, coordination states, 4

backscatter (radar)blooming, 180-6coefficient, 176,178dielectric constant, 176-7mechanisms, 178and polarization, 180and roughness, 177-,,9,187,189surface materials, 176

bacteria, anaerobic, 239Badain Jaran (Mongolia), 190balloons, aerial photography, 50band-pass filters, 134band ratioing see ratioingband-stop filters, 134barchans, 114-15baryte, 222basalts, Plate 6.4

flood, 85, Plate 9.1outcrops, 222plateau, 85raster images, 219reflectan~, 147

basins, 107batholiths, 93bathymetry, 214,231, Plate 1.1bedding, 76-7beryl, 8best-fit lines, 209binary numbers

in data transmission, 32in image processing, 122

bismuth, anomalies, 237bivariate plots, 139,143,150,152,153blackbodies, 138,169

emittance, 3-4,10,160water as, 13-14

blind spot, 17blooms (radar), 180-6blue sky, 6-7bond bending, 9bonds, distortions, 4Boolean algebra, 222Borrowdale (UK), Plate 9.6Bouguer gravity anomalies, 218,221, Plate

8.5, Plate 9.6box filters, 134Brahmaputra, 70braided streams, 113, 119brightness, 2

and contrast, 25,27detection, 23,24

bronzite, 8buried landscapes, 95buttes, 77byte-format data, 122

279

280 Index

contour maps, 30,215digitI'zing, 211elevation data, 213-14sampling, 210-11visual perception, 206

contours, optical illusions, 30contrast, 21,24

and brightness, 25,27contrast stretching, 123, 125-33, 137

colour images, band choice, 129"'-31in colour spaces, 131-3decorrelation,131,138,141,142-3equalization, 128exponential, 129Gaussian, 128goals, 125linear, 125logarithmic, 128magnetic data, 215nonlinear, 128piecewise, 128ramp, 128in thermal images, 168

convolution, 134-5limitations, 137-"8matrices, 135,136,1.37

copperanomalies, Plate 9.6concentration, 212,223mineralization, 222-3,237

coral reefs, Plate 1.1corn, spectra, 146cornea, 16corner reflectors, 177,180-6,227Corona satellites, 51correlation coefficients, 139cortex (visual) see visual cortexcovariance, 139covariance matrices, 139cover-basen1ent relationships, 217craters, 102-4,232

optical illusions, 30cross-cutting relationships, 95cruise missiles, guidance systems, 214crustal deformation, 96-7,104-7crystal-field effects, 8cubic splines, 209,210CuddapahBasin(India),95,Plate4.2cuestas, 91-2,93Cumberland Plateau (US), 91cut-off (image correction), 125CZCS (Coastal Zone Color Scanner), 58

calcite, 10,11calderas, 102,110-12,188Caledonides (Scotland), 99California (US), 101, 194, Plate 5.7, Plate 9.9

MojaveDesert,167,172,193,Plate6.3camera lenses

aberration, 33fieldofview,36refraction, 33

camerason aircraft, 32aperture,17,33eyeanalogy,16-17film speed, 33focal length, 17,33fstop,17,33on manned spacecraft, 55metric, 36, 56video,32see also digital cameras; Vidicon cameras

Cameroons, 86camouflage, 242

detection, 147 ' ;reflectivity, 34

Canada, 118,119, 120, Plate 9.3Canadian Arctic, 113Canadian Shield, 75,91,98cancers, 243canonical analysis; 141,152capital investment, and ~xploration, 234carbonate minerals, 83, PlateS 6,4-5

radar images, 183spectra, 9,10,130,142thermal inertia, 165weathering, 82

carbon dioxide, 6, 160cartography

digital, 66-7,233see also maps

catchments, 211,212,223,240CCDs see charge-coupled devices (CCDs)CD-ROM resources, 259-63CD-ROMs, data storage, 66cements, 77,79Centre National d'Etudes Spatiales (CNES),

62 .centripetal drainage, 72, 73change detection, 156-7

automated, 157charge-coupled devices (CCDs), 38,42,62,

63incameras, 17in imaging spectrometers, 49

charge-transfer transitions, 8Chile, 108,109,111,257China, 183, 189, Plate 4.4chlorite, 9chlorophyll, 12-13,61,144,145chromaticity diagrams; 28,29chromium

anomalies, 219concentration, 223, Plate 8.3

CIE see Commission International del'Eclairage (CIE)

cinder cones, 108,109,110circular features, detection, 102-4cirques, 74classification

geological applications, 151-2methods, 149-51,153-4,156multispectral, 147-52problems, 151and radar data, 199-200

surface materials, 147-52,169thermal data in, 171

clastic sediments; 192coarse, 77-8erosion, 79fine, 77,78-81radar images, 183

clay minerals, 78-9,80-1,Plate6.4detection, Plate 3.2discrimination, 151isotopes, 219and ratio images, 145reflectance, 141spectra, 9, 236thermal inertia, 165

clayrocks see mudstonescleavage, metamorphic rocks, 90climate

and geomorphology, 229and water supplies, 240,

CNES (Centre National d'Etudes Spiitiales),62 c .

coalfiel.s, Plate9..9Coastal Zone Color Scanner (CZCS)j

58coasts, geological interpretation, 75C-Obonds, 9,10,142,152,172coherent radiation, 43Colorado Plateau (US), Plate 4.1Colorado River (US), 70,Plate4.1Colorado (US), 165,168colour atlas, 28colour cubes, 131,132colour images

band choice, 129-31colour range, 122contrast stretching, 131-3see also red-green-blue (RGB) images

colour perception, 145colour photography, 34-6,68colour representation, in infrared images,

27colour theory

Munsell's, 28tristimulus, 28Yourtg's, 26-7,218

colour transformations, 219-20colour vision, 17-18,26-8

and acuity, 21,23and aesthetics, 27-8early studies, 26imagememory,27physiological effects, 27psychological effects, 27quadrichromatic, 26trichromatic, 26

colour wheels, 27, 132, Plate 5.2ComffiiSsion International de 1 'Eclairage

(CIE)chromaticity diagram, 28,29colour co-ordinate system, 28, 133tristimulus values, 28

cones (vision), 17-18,21,26-7tonal sensitivity, 26types of, 18,26

coniferous trees, 13consequent streams, 70,93 .

conservation of energy, 5constructional landforms, 69,107-20context, 69contextual classification, 156continental glaciation; 74-5,116,117,

119

daguerreotype photography, 50dams, and water supplies, 240-1Danakil Depression (Ethiopia), 201dark-pixel corrections, 125Dasht-e-Lut (Iran), 169data collection, images, 32-67data reduction, 131,138-45data sources, 264-5data storage

CD-ROMs, 66elevation data, 213magnetic tape, 39

data transmissionanalogue, 32digital, 32

Davis,W.M., 70,228,229deciduous trees, 13decorrelation stretching, 131,138,141,

142-3Delaunaytriangles, 213

Index 281

DLR (German Space Agency), 204DNs see digital numbers (DNs)dolomite, 10,83

thermal inertia, 165,166,171domes, 107Doppler shift, 46, 176drainage patterns, 70-3,224, Plate 4.1

categorization, 230tributaries, 223see also dendritic drainage; trellis drainage

drainage surveys, 211-12, 244dropped lines, replacing, 2q5drumlins, 117,118dryness, and radar images, 188-9D-stretching see decorrelation stretchingDTED (Digital Terrain Elevationpata),

213ductile faults see shear zonesdunes, 114-16,190dyk~,~5/96;97 ,. 138

basaltic, Plate 5.1, Plate 5.9dejection, 87,240-1,Plate~,~(, ';.'jOInts, 85

deltas, 114,119DEMs see digital elevation models (DEMs)dendritic drainage, 72,73,86,88,91,94

development, 79~80density slicing, 149depression angles, 43,66

and radar images, 176,178,180,195-7depth perception, 29-30deserts, radar images, 187,190,192,193,196desert varnish, 148,187de-striping, 255-7destructionallandforms, 69-76detectors

area-array,52faults, 42photoelectronic, 39photosensitive, 38radiation-sensitive, 38sensitivity, 39 .'

signal-to-noise ratio, 39Devon (UK), Plate3.1dew, 162diapirs, 88dielectric constant, surface materials, 176~7,

Plate 7.3differentiation, in raster data processing,

217diffraction gratings, 38diffuse reflection, 5-6digital cameras

charge-coupled devices, 17data, 66-7

digital cartography f 66-7,233digital elevation models (DEMs), 66,204,

213-14,216,252engineering applications, 241forms, 213geomorphological applications, 223~4,

229hydrological applications, 223-4and parallax, 220,233and perspective views, Plate 9.2

digital image processing, 122-59goals, 122

digital imagesapplications, 66-7data reduction, 131directional bias, 137high-frequency featill;es, 133histograms, 123-5low-frequency features, 133redundancy, 131registration, 122

digital numbers (DNs), 32,39-40, 122in colour displays, 218. .in digital image processing, 123, 125-8,

129254

257

133-6

213

v~etation interactions, 12-13water interactions, 13-14

electronic transitions, 4elements

concentrations, 211-12energy levels, 7-8

elevation data, 213-14,233,252emission spectra, 5emissivity,10,160,165emittance, 2,3-4,161EMR see electromagnetic radiation (EMR)EM spectrum see electromagnetic (EM)

spectrumenergy, 2

conservation of, 5detection, 4-5" ;

engineering applications, 241-3England,82,215,218,Plate8.1;Plate8.5

Devon, Plate 3.1 .Lake District, 237, Plate .9.6Selby Coalfield, Plate 9.9

environmental hazardsgeochemical, 243-4geophysical, 241-3

Environmental Science Service (ESSA),satellites, 57

EOS see Earth Observation SysteIJ1 (EOS)EOSA T (Earth Observation Satellite

Co~ration), 59equalization stretch, 128.Eritrea, 56,99, Plate 3.3, Plate 9.1, Plate 9.,5

Mersa Fatma, Plate 1.1western lowlands, 102see also Sudan-Eritrea border

Eritrean Highlands, 256EROS Data Centre, 55erosion, 69-70

cycles, 70glacial,. 73-5, 83 ;headward, 70,72 '.marine, 69,75 .

rates, 72soils, 240stream, 70-2wind, 69,75~6

erosion resistance, 69,76-7,90ERS-1 (satellite), 64,230-1

images, 65radar altimeters, 214

Erta' Ale volcano (Ethiopia), 201, Plate 7.2ESA see European Space Agency,(ESA)escarpments, 91-2,93eskers, 117ESSA (Environmental Science Service),

satellites, 57ethics, and exploration, 244Ethiopia, 201,Plate 7.2Ethiopian Highlands, 255Eureka Valley (US), Plate 9.9European Space Agency (ESA), 57satellites, 64, 204 .-

evaporites, 81-3evapotranspiration rates, 240exfoliation, 88exitance, 2exploration, 233-41

and capital investment, 234economics, 234, 235ethical issues, 244goals, 234market-led, 234offshore, 239risks, 234

Earthellipsoid, 253geoid, 253intemalheatflow, 11microwave emissiQn, 176radiance, 161surface geometry, 253-4surfaceheatremoval,161-2surface temperature, 4, 11,43,160

profiles, 163-4Earth Observation Satellite Corporation

(EOSAT), 59Earth Observation System (EOS), 142

future prospects, 66launch, 65platforms, 66thermalimages,173

earthquakes, 99-100,242,Plate9.9Earth Resources Technology Satellite

(ERTS-l) see LandsatEarthwatch, 66East Africa, 216East African Rift (Kenya), Plate ~.8edges

detection, 135enhancement, 135-6representations, 133-4

Egypt, 193eigenvalues, 139-41,.153eigenvectors, 139-41,142,153Einstein's equation, 1electromagnetic (EM) spectrum, 2,3,6electromagnetic radiation (EMR), J:':'~5

absorption, 5atmospheric effects, 6-7characteristics, 1-2emitted region, 7fields, 1generation, 2-4intensity, 2andmatter,4-14microwave region, 7,11,43,176polarized, 1-2reflected region, 7reflection, 5reradiation, 5rock interactions, 7':'11"scattering, 5,6-7transmission, 5

282 Index

low-pass, 134, 135median, 143,257medium-pass, 134spatial, 134-8square, 138see also spatial-frequency filtering

fire fountains, 110first derivative filters, 136First World War, aerial photography, 50,51flat-field calibration, 154flatirons, 47flight lines, 37,51-2,249

spectroradiometry, Plate 3.2flood basalts, 85, Plate 9:1flood plains, 72,113-14fluorescence, 4; 12fluorite, 222fluorosis, 243fluviatile landforms, 112-14folding , .

metamorphic rocks, 90, Plate 4:5 .-.

rada~ima~!!~, 199unconformities, 95

folds, 104-7classification, 104dip, 105-6nomenclature, 105

foreshortening, 48,49Fourier analysis, 23,134,209Fourier synthesis, 23Fourier transforms, 134fovea, 17,18,24,29fracture analysis, 224fractures

and geographical information systems,224

and water supplies, 241frequency, 1frequency-domain filtering, 134Front Range (US), 165,168fundamental wavelength, 4Fuyo-1 see Japanese Earth Resources

Satellite (JERS-1)

exploration (cont.)stages, 234,235targeting, 234-6see also geochemical exploration;

hydrocarbon exploration; metalexploration; water exploration

exponential stretch, 1+9extrapolation, 208eye base, 29Eyeglass, 66eyes

compound, 16evolution, 16see also human eye

faceted spurs, 100,101 '

false-colour images, 35':'6,126-7" 228,230

band choice, 129-31, Plate 3.1thermal data in, 168-9

fan draiIlage; 72faults, 95-104, Plate 8.4, Plate 9.1active, 100 ..;', .c.,

angles, 97and construction projects, 242detection,97,98,240

issues, 224asenvironmentalhazards,99-100,242and geographical information systems,

224 i

mapping, 233normal, 96,97,98radar images, 199reverse, 96,97splay,97strike-slip, 96,97,98,99,101,102thrust, 96,97types of, 96

FCCs see false-colour imagesfeldspars, 84,86,736Fe-Obonds, 172ferric minerals see iron mineralsferromagnesian minerals, 84,236field studies

economics, 235and remote sensing, 227

filmblack and white, 33-4colour

infrared, 35natural, 35sensitivity, 35structure, 35

digital.scanning, 32exposure, 33holographic, 55infrared, 34 ,panchromatic, 34,56 .'

resolution, 33sensitivity, 34speed, 33storage, 32

problems, 66structure, 33

filtersband-pass, 134band-stop, 134box, 134coloured, 26, 34directional, 136-7, 138frequency-domain; 134high-pass, 134,,135,136image-processmg, 23,37

gabbros, 90, 98, Plate 5.9GAC (Global Area Coverage), 58gamma-rays, 4

detection, 6generation, 2

gamma-ray spectrometers, 49-50,219gases

radiation absorption, 6volcanic, 110

Gaussian stretch" 128Gemini programme, 55geobotany

anomalies, 156,237indicators, 145 .

andratioing, 145geochemical data, 211-13, Plate 8.3

anomalies, 212,233-4background threshold, 212drainage surveys, 211-12histograms, 212-13ratioing, 219risk assessment, 243-4sampling, 211-12

geochemical exploration, 21.1-12,220pathfinder elements, 219

geochemical hazards, 243'-4geochemical stress, 145geochemical surveys, soils, 243-4geodetic datum, 253-4 .

geograf>hical information systems (GIS),206-26

data analysis, 221-4and digital elevation model, 223-4in exploration, 234-6fault studies, 224furictions, 221 ..geological applications, 221-2goals, 221layered structures, 206measurement furictions, 221-2- .

and metal exploration, 237multivariate data, 206-7neighbourhood functions, 223organizational issues, 221overlay functions, 222-3raster-based, 207-8,221,225registration in, 221retrieval functions, 221-2vector-based, 207-8,221,225

geoid, 253geological applications, 50,51,56,58,227-47

aerial photographs, 66classification, 151-2data processing, 228digital images, 66-7geographical information systems, 221-2potential, 64remote sensing strategies, 227-8spatial pattern recognition, 155-6and spectral coverage, 228

geological mapping, 231-3,Plate8.3,Plate8.4, Plate 9.3, Plate 9.6

boundaries, 231,233image units, 231-2lithofacies units, 231,232-3subsurface, 206

geometric correction, 253-5geomorphological applications, 228-31

change detection; 157and digital elevation models, 223-4,229terrain analysis, 228,230

geomorphological images, interpretation,230

geomorphological" mapping, 229-31multitiered approach, 230

geomorphological theoriesbottom upwards approach, 228-9climatic, 229Davis's, 228,229megascopic, 228stageist, 229structural, 229top downwards approach, 229

geomorphological units, mapping, 230geomorphology ;228-31

coastal, 75geophysical data, 210-11

gridding, 210line surveys, 210sampling strategies, 210-11

Geoscan system, 52,173geostationary satellites, 53-4

resolution, 54geosynchronous orbits, 53-4geothermal heat, 160-1German Space Agency (DLR), 204GIS see geographical information systems

(GIS)glacial erosion, 73-5,83glaciallandforrns, 116-20

erosional, 73-5radarimages,187

Index 283

fi16lgnetic anomalies, 239offshore, 239radar altimetry, 238

hydrocarbon resourcesformation, 238occurrence, 238and pollution, 239seepages, 238-9, Plate 9.7

hydrogen ions, 236hydrogen sulphide, 239hydrological mapping, and digital elevation

models, 223-4hydrothermal fluids, 84, 101, 102, 237hydrothermal mineralization, 222hydroxides

and ratio images, .145spectra, 8,130,142

hydroxylates, spectra, 9hydroxyl-bearing minerals, spectra, 156,

236, Plate 5.6, Plate 5.8, Plate 9.5hydroxyl (OH-) ions, vibrational transitions,

9Hyperion system, 66hyperspectral data, 66,145,147,152-5

glaciationcontinental,74-5,116,117,-119valley, 73-4

Global Area Coverage (GAC), 58global warming, 6, 66gneisses, Plate 8.3Goddard Space Flight Center (US), 55,57GOES (satellite), 57goethite, 8,145goitre, 243gold mineralization, 219,.220GOMS (satellite), 57gossans, 236gradient change, 136GrandCanyon(US),70,}>late4.1granites, 85-8,89, Plate 8.4, Plate 9.2-3, Plate

9.4anomalies, 237 ,health risks, 243 ~intrusions, 87,102isotopes, 219radar images, 185 iraster images, 219reflectance, 147and sandstones comp~ed, 86-8thermal inertia, 165,166weathered, 148

gravitational potential, 217gravity data

anomalies, 218,219,221, Plate 8.2, Plate8.5, Plate 9.6

from ocean-surface elevation, 214pseudocolours, 216

greenhouse effect, 6,160greybodies, emittance, 10grey scales, 24,25, 28,}23

human discrimination, 125,215,228gridding, 208-10 ,

contour maps, 210-11geophysical data, 210kri~g, 210moving averages, 210regression analysis, 209splines, 209-10trend-surface analysis, 209

ground control points, 254ground stations, 62, 64groundwater

abstraction, Plate 9.9accumulation, 240-land subsidence, 242-,.3

gullies, 72,77, SO, 81, 116Guyana, 198gypsum, 9,..11

182

62,63

heat transfer, radiative, 161Hei-ch' a Shan (China), 189 .,' .hematite, 8,11,145 rir"jHerculaneum (Italy), 243High Atlas Mountains (Morocco), P/ate6.2Highlands (Scotland), 74high-pass filters, 134,135, 136High-Resolution Imaging Spectrometer

(HIRIS), 53,66,131,142High Resolution Picture Transmission

(HRPT), 58Himalayas, 59Himawari (satellite), 57HIRIS (High-Resolution Imaging

Spectrometer), 53,66,131,142HIS system see hue-intensity-saturation

(HIS) systemhistograms, 123-5

in classification, 147,148,149,151cumulative, 123

H-O-H bond~, 9 .hornblende, 10 ..; .HRPf(HighResolutionPictU~." "

Transmission), 58HRV (Haute Resolution Visible), 62,63Huang He (China), 189hue, 132hue-intensity-saturation (HIS) system, 28,

138,143,168, Plates 5.2-3, Plate 7.3transforms, 132-3,144,233, Plate 5.6,

Plates 8.5-6, Plates 9.3-4non-image data, 216,219-20

human eye, 16-19angular resolution, 30blindspot, 17camera analogy, 16-17chromatic aberration, 21cornea, 16cross-section, 16depth of field; 17grey-scale discrimination, 125,215,

228lens, 16

focal length, 17modulation transfer function, 22-3parallax,249 .

responses, 21-2sensitivity, 18spatial acuity, 23-4spherical aberration, 21 .see also pupil; retina

human vision, 16-31acuity, 21binocular, 29brightness detection, 23,74colour perception, 145contour map perceptiqns, 206depth perception, 29-30and image dissection, 19-21'and memory, 19night, 18photopic, 18,21scotopic, 21spatial frequency response, 134spatial resolving power, 19-24stereoptic, 29stereoscopic, 19,249see also colour vision; optical illusions

humidity, 162Hyderabad (India), 64hydrocarbon exploration, 234,238-9

geochemical data, 238-9geophysical data, 238

..,Iceradar images, 186,187spectra, 14striations, 74

ice darns, 119ice sheets, 73,74-5,118-19Idaho, 110IFOV see instantaneous field of view

(IFOV)igneous rocks

colour, 84extrusive, 84-5intrusive, 85-8radar images, 193recogrlition, 83-8 ,silicon dioxide content, 1740.thennal inertia,171,113thenn"l spectra" 173,174 .

ignimbrites, 84,110,111, 111image atlases, 264image correction, 253-8defects, 253 .

distortions, 253-5image data sources, 264-5image dissection .

and human vision, 19-21 :and perceptual quality, 20

image interpretationaerial photographs, 36,68-'21'geomorphological images, 230and optical illusions, 30-1,96precautions, 96psychophysiology, 27and spatial resolution, 19-24surface characteristics, 68

imagesdata collection, 32-67data sources, 264-5geological applications, 227-41geomorphological, 230rectification, 254redundancy in, 19-20resamplmg, 254seasonal, 69,156, P/ate5.10sonar, 231see a/so colour images; digital images;

false-colour images; fufrarerlimages;radar images; red-green-blue (RGB)

284 In4ex

images, 65sensors, 62-4spectral bands, 58stereoscopic images, 214

jarosite, 8Jebel al Dmz (Syria), 191JERS-1 see Japanese Earth Resources Satellite

(JERS-1)joints, 96

carbonate minerals, 83columnar, 85dykes, 85granites, 86intrusions, 85lavas, 84sandstones, 77, 79

Jordan, 137Jupiter, moons, 6jokulhlaups, 110

-"Kaibab Plateau (US) 70 115 ' '",. ,

, ,Kal~in Tag~i~,;>unt~(Cmha);:183'kaines,.W7-"1'S'" .Kansas (US), 78 .

kaolinite, 9,11, Plate 9.2Karakoram Range (Asia), 187Karhunen-Loeve transform see principal

component analysis (PCA)Kentucky(US),55,59,82,91Kenya, Plate 9.8kettle holes, 116,119Keyhole satellites, 51kimberlite pipes, 102Kirchhoff's law, 10,161Kirthar Range (Pakistan), 101Kiruna (Sweden), 62Krige, D. G., 210kriging, 217

in gridding, 210Kuhha-ye Sahand (Iran), 188

images (cont.)images; stereoscopic images; thermalimages

image segmentation, 156image striping, 255imaging spectrometers, 43,49,52-3

specifications, 53spectral range, 52

impact structures, 102-4India, 91,105,144,187,Pl4te4.2Indiana (US), 117Indian Remote Sensing Satellite (IRs-I),

64Indus, 70infiltration, 240infrared images, 160-74 .

colour representation, 27,50see also thermal images

infrared photography, 34colour, 50

infrared radiation, generation, 2InSAR (interferometric synthetic-aperture

radar) 203 204 242 243:' -'.', ". ( '.'INSAT (satellite), 57 .

inselbergs, 88insequent streams, 70instantaneous field of view (IFOV), 39,40,

62in geostationary satellites, 54in microwave imaging, 43

intensity, 2,132intensity-saturation-hpe (ISfIrsystem see

hue-intensity-saturation (HIS)system

interferometric synthetic-aperturetadar(InSAR),203,204,242,243

interferometry, Plate 9.9radar, 203~4, 213-14;242

Internet; data sources,. 264~5interpolation, .,208-10intrusions, 85::'8,93,95 " -;-~

granites, 87,102 '

10, remote sensing; 6 .'ionic bonding" 7-8Iran, 101,103,169,.188Irian Jaya, 184,.186iron minerals .

colloidal, 236-7detection, 236discrimination, 151and ratio images, 145reflectance, 141spectra, 8; Plate 5.6, Plate 9.5

iron oxides, 148spectra, S;'130, 142

iron sulphides, 81irradiance, Z "

IRS-l (Indian R~II1ote 5ensmg Satellite),64

ISH system see hue-intensity-saturation(HIS) system

isotopesabundance, 219gamma-rays, 49-50unstable, 49

Italy, 243,P/ate9,~

Labrador (Canada), 103labradorite, 10LaBrea(US),238LAC (Local Area Coverage), 58lacustrine landforms, 112-14lahars, 243Lake District (UK), 237, Plate 9.6lakes, Plate 7.2

drainage, 114glacial, 119oxbow, 113,114

Lambertian reflectors, 5landforms

aeolian, 114-16constructional, 69,107-20 :

destructional, 69-76division, 230.' ..

and drainage, 70-3fluviatile, 112-14geomorphological mapping, 230lacustrine, 112-14in oceans, 231see also glacia:llandforms

Landsat, 59-62,220-1,229applications, 61development, 55,59-60future modifications, 62future prospects, 66ground stations, 62Multispectral Scanner, 61,68,129,139,

141,168-9orbits, 60

Retu~-beam Vidicon, 61,68seasonal studies, 156stereoscopic potential, 62Thematic Mapper, 9,61-2,138,142,145,

153,166,237thermal data, 61

landscape change, 229landslips, risk assessment, 241Large-Format Camera (LFC), 56,229laser altimeters, 214lasers, in remote sensing, 12Las Vegas (US), Plate 9.9lava flows, 107-12lavas, 84-5,124,188,191, Plate 7.2

dissected,86environmental hazards, 243radar images, 188thermal inertia, 167viscosity, 85weathered 86

law of V's, 93,105,233layovers, 47,48leaves

dielectric constant, 177senescence, 145,Plate 9.7spectra, 12,13,146

Lewis satellite, 66,142LFC (Large-Format Camera), 56,

229Libya, 195light, detection, 24light scattering, in water, 13limestones, 81-3,84,98,Plate5.9

drainage patterns, 83and health risks, 243hydrothermal mineralization,

222isotopes, 219radar images, 200thermal inertia, 16!?, 166, 171

limonite, 8, 8J, 83lineaments, 95-104

use of term, 96linear, use of term, 9"6linear features, precautions, 96Linear Imaging Self SCanning (LISS)

cameras, 64linear stretch, 125linear unrnixing, 154line defects, drop outs, 253lines, optical illusions, 31line scanners; 38-42aerial, 52 .

analogue iiata, 39arrangement, 3q,detectors, 38,42

sensitivity, 39digital data, 39-41images

analogue, 41defects, 42digital, 41distortions, 41structure, 40-':1 ,-

limitations, 42resolution, 40-1

radiometric, 39spatial, 39,42spectral, 39,42

signal-to-noise ratio,. 39telemetry, 39in thermal inertia estimation, 169-71

line surveys, 210

Japan, 48Japanese Earth Resou.rces Satellite (JERS-1),145 .

applications, 230-1,236data, 62-4false-colour limitations, 129

Index 285

LISS cameras (Linear Imaging Self Scanningcameras), 64

listric surfaces, 97lithological discrimination see rock

discriminationLocal Area Coverage (LAC), 58loess, 116,189logarithmic stretch, 128logical operators, 222look-up tables (LUTs), 125,128,.129,130,

Plate 8.1LosAngeles(US),238Louisiana (US), 114low-pass filters, 134, 135Luliang Shan (China), 189luminance, 24 , ,Lut Block (Iran), 169LUTs (look-up tables), i25,.128,.129,130,

Plate 8.1

metal exploration, 236-7dra~age surveys, 211-12geobotanical methods, 237geochemical data, 236-7and geographical information systems,

237geophysical data, 236targeting, 237, Plate 9.6

metal-hydroxyl bonds, 9metals, dielectric constant, 176metamorphic rocks, 91, Plate 5.1; Plate6.2,

Plate 8.3, Plate 8.4cleavage, 90erosion resistance, 90folding, 90radar images, 186,193 ,.'recognition, 88-90

metamorphism, mechanisms, 88metapelites, Plate 8.4metasediments, Plate 9.2-3meteorite impacts;,. 103m$orologica1 satellites (me~(sYi" $7-:?,h.?,

resolution, 57Meteosat, 57methane, 6,160Metric Camera, 56Mexico City, 242Mg-OHbonds, 9,142,152microwave imaging, 43-8

resolution, 43see also passive microwave imaging; radar

images

Mach bands, 24,25 ' '-'magmas, 84,86,88,101,102 '.

basaltic, 109and volcanism, 107-12, Plate 9.9

magnesite, 10magnetic data

anomalies,215,217,Plates8J-2,Plate9.6in hydrocarbon exploration, 239

pseudocolours, 216magnetic tape, data storage, 39 ;Makran Range (Iran), 101,169Mali, 185manned spacecraft

cameras, 55data, 55-7disadvantages, 54missions, 55Space Shuttle, 55-7

mapsfrom images, 206-7geomorphological, 229-30hydrological, 223-4production standar~ 221projections, 253-4 ;reconnaissance, 233rectification, 254structural, 228surface materials, 15}-2 ,symbology, 208 .

terrestrial, 232see also contour mapsi geological

mapping; geomorphologicalmapping; mineral-mapping; spectralmapping,

Marathon Basin (US), 107'1. ,-marbles, Plate 9.1 -c-,t'; -,-

marine erosion, 69,75 .Mars, 177,232 '.

moaulation, 22modulation transfer function (MTF), 22-3,

134achromatic, 23,253chromatic, 23

moisture contentand radar images, 177,188and thermal images, 166

Mojave Desert (US), 167,172, 193, Plate 6.3MOMS (Modular Optical Electronic

Multispectral Scanner), 56,249Mongolia, 190monoclines, 92montmorillonite, 9Moon

craters, 102,-3,232, ..'.,gamma-ray surveys, 50 " ,geomorphology, 229 '

optical illusions, 30moraines, 116,117,119Morocco,106,Plqte6.2mountains, fonnation, 229Mount Etna (Iialy)~ -243, Plate 9,9MountFuji(Japan),65 .MountStHelens(US),111,11;2;243Mount Tom Price (Australia), 182Mount Turner (Australia), 182moving averages, in gridding, 210MS (multiple sclerosis), 243MSS see Multispectral Scanner ~MSS)MTF see modulation transf!!r fUnctio}:'l (MTF)mud avalanches, 112mudflows, 110,112;241,.243mudstones, 77,(8,80,90

thermal inertia, 165multiple sclerosis (MS), 243multispectral data, hi~ograms,: 123-5 '.Multispectral Scanner (M$S), 61,62,68,139,141 ..

false-colour limitations, 129~'Plate 3.3thermal images, 168-9 -

multispectral thermal data, 172~4multitemporal analysis, 156-7multivariate data, in geograp~al

information systems, 206-7multivariate raster images, 218-:21Munsell, Albert Henry (1858-:1918), coloiiratlas, 28 .,

muscovite, 9,10mylonite, 99

...NASA see Natiqnal Aeronautics and Space

Administration (NASA)NASDA (National Space Development

Agency(Japan)),62 ,National Aeronautics and Space

.A4ministrapo~ {NASA), 52,53,56,57,204

Landsat series, 59-62missions, 55,58; 66

National Intelligence Mapping Agency(NIMA) (US), 213,214,229

National Oceanic and AtmosphericAdministration (NOAA)

orbital plans, 60satellites, 57,58,157,166-7

National Space Development Agency(NASDA) (Japan), 62

natural gas see hydrocarbon resourcesneighbourhood functions, in geographical

information systems, 223nervecells,17,18-19neural networkS, 153

bands, 43detection, 43emittance, 43,176generation, 2,43

microwave systemsactive, 43instantaneous field of view, 43passive, 43see also radar

Mid-Atlantic Ridge, 231Middle East, 115mid-infrared (MlI(), 4, '7Mie scattering, 7mineralization, 219,220,222-3,244, Plate

9.6models, 237

mineral mappingand classification, 151-~and ratioing, 145spectral, 154subsurface, 206

mineralsband choice, 130detection, Plate 3.2, Plate 9,6ferromagnesian, 84, 236spectra, 7-11see also carbonate minerals; clay minerals;

hydroxyl-bearing minerals; iro~minerals

minimum-distance-to-means classification,149,150

minus-blue panchromatic images, 34Miranda(moon),232MlR (mid-infrared), 4,7missiles, guidance systems, 214Mississippi (US), 114Moderate Resolution Imaging Spectrometer

(MODIS), 66Modular Optical Electronic Multispectral

Scanner (MOMS), 56,249

286 Index

overstep, 94overtones, 4oxbow lakes, 113,114oxygen, 6

energy level, 7ozone, 6

Nevada (US), Plate 6.4, Plate 9.9Newton's fringes, 203, 204New York (US), 114,118nickel, anomalies, 219Nigeria, 89night vision, 18Nile, River, 179NIMA <National Intelligence Mapping

Agency (US», 213,214,229Nimbus series (satellites), 58,59,60NOAA see National Oceanic and

Atmospheric Administration(NOAA)

noise, random, 143,257non-image data, 206-26

forms of, 207-8hue-in tensi ty'-Sa tura tion transforms,

216,219-20raster format, 208-21instereomodels, 220-1vector format, 208

nonlinear stretch, 128 -:: 0',.nonselective scattering;'7 -,nonsilicates, spectra, 10,. 11North Africa, Plate 3.4North America, Plate8.2North Dakota (US), 117NOT operator, 222Nottinghamshire (UK), 41Nova Scotia (Canada), Plate 9.3nuees ardentes, 110-12

obsequent streams, 70,93oceanographic satellites, 57-9,66oceans

floor topography, 214, 215landforms in, 231surfaceelevatioI'l,. 214

offset streams, 100,.1.01o-H bonds, 9oil exploration see hydrocarbon explorationOklahoma (US), 94,166olivine, 8,10Oman, 87,90,98,101,Plate5.9onion-skin we'athering,. 88'..ophiolites, 90,101, Plate 5.7, Plate 5.9.OPS (Optical Sensor),62optical illusions, 19

in aerial photographs, 36-7contours, 206and image interpretation, 30-1,96Mach-band, 24

Optical Sensor (OPS), 62optic nerve, 17orbiting satellites

advantages, 51characteristics, 53..:4disadvantages, 51spectral bands, 58see also geostationary satellites; Sun-..synchronous satellites

pixels, 3i, 40-1, 61,122classification, 151distortions, 254histograms, 123purity, 155ratioing, 143spectra, 155

Planck's law, 2planetology, 229,232plankton blooms, 239plant cells, structure, 12plants

anomalies, 237constituents, 12-13dielectric constant, 177energy absorption, 12hydrocarbon contamination, 239nutrients, 80-1roots, 241spectra, 144see also trees; vegetation

plateau basalts, 85point-spread function (PSF), 21-2,33polarized radiation, 1-2,180pollution, 66polynomial transforms, 254Pompeii (Italy), 243pores, water-containing, 14porphyry, 237

mineralization, 222-3potassium isotopes, 219,220power, 2

exitance, 2irradiance, 2radiance, 2

spectral, 2radiant flux, 2

spectral, 2radiant flillS density, 2

emittance, 2precipitation, and water supplies,

240primary colours

additive, 26, 34-5, 219, Plate 2.1subtractive, 26, 35

primitives, 155principal component analysis (PCA),

138-43canonical, 141,152in change detection, 157and classification" 151covariance matrices, 139decorrelation, 138-9,140directed, 141-2eigenvalues, .1.39,-,41,153eigenvectorsJ)-39-;-41, 142, 153in image processing, Plates 5.4-5, Plates

6.3-4, Plate 7.4in spectral mapping, 153in thermal image processing; 169

principal point, 36,37probability density function (PDF), 123projections, 253-4proximity analysis, 223pseudocolours, in raster-format data,

215-16pseudoshadowing, 137, 216PSF (point-spread function), 21-2,33psychophysiology, and image

interpretation, 27

orbitsgeosynchronous, 53-4polar; 54

ores, detection, 152,236-7OR operator, 222orthoclase, 10orthophotographs, 255outcrops, 78,86; 148, 154, Plate.4,2

displacements, 99outwash deposits, 119overlay functions, in geographical

information systems, 222-3

Pakistan, 59,101, 18zPamir Range (Asia), 187panchromatic film, 34, 56parallax, 29,30,37

and digital elevation models, 220,233

distortions, 254, 255and elevation data, 214and radar images, 180,197and stereometry, 249-52

parallax bars, 251parallax wedges, 251 .parallelepiped classification, 149",.,,-,.

150 , ::'particle;'~'C\ly~d~ty, 19passive microwave imaging, 176

resolution, 43passive remote sensing, 4pathfinder elements, in geochemical

exploration, 219pattern, 68-9pattern recognition, 145-57

and artificial intelligence, 145,147spatial, 155-6see also classification

PCA see principal component analysis(PCA)

PDF (probability density function) seehistograms

peat, health risks, 243peneplains, 72Pennsylvania (US), 181perceptual abiJjty, and age, 16peridotites, 90, Plate 5,1

raster images, 219thermal inertia, 165

permafrost, 120perspective views, Plate 9,2petroleum exploration see hydrocarbon

explorationpH, and metal exploration, 236photoelectronic detectors, 39photogeology, 68-121,180photographic film see filmphotographs (aerial) see aerial photographsphotography, 32-7

colour, 34-6,68daguerreotype, 50 .:"infrared, 34, 50minus-blue, 34oblique, 36vertical, 36,37vignetting, 37see also aerial photography; colour

photographyphotomultipliers, 50photons, 1photosensitive detectors, 38photosynthesis, catalysis, ~2phreatophytes, 241physical resources, exploration, 234piecewise stretch, 128pigeonite, 8pinnate drainage, 116 ,Pisgah Crater (US), 167,172, Plate 6,3Pittsburgh (US), 181,199

pupildiameter, 17dilated, 21

Index 287

pushbroom systems, 42-3,56,249,253advantages, 43aerial, 52arrangement, 42resolution

spatial, 42spectral, 43

signal-to-noise ratio, 43pyrite, 145,236pyroclastic rocks, 84, 85

thermal inertia, 165

Qaidam Basin (China), Plate 4.4QinhaiProvince (China), Plate4.4quanta, 1quantum energy, 2quartz, 9,10,84,85-6 '.',' .

cementation, 77spectra, 10,237thermal inertia, 165,171

Quebec (Canada), 103 c,.

RADAM project (Brazil), 53,230radar, 43-8

airborne, 53altimetry, 214,231,238antennae, 43, 46beam width, 44-6blooms, 180-6coherence, 176corner reflectors, 177, 180~6,227depression angles, 43,47,48,176,178,

195-7development, 43,50elevation data, 214flatirons, 47ground range, 43incidence angles, 47letter codes, 43look direction, 189-95parallax, 197penetration, 176,179,189polarization, 47,180.pulse length, 44range, 43

real-~perture, ,464

44-54551

43,44,50; 231

177176-80

178,180,

reIflote sensing missionsfuture prospects, 65-7manned, 55-7tmmanned, 57-64 ,. ,

remote sensing systemsactive, 4constraints, 7energy detection, 4-5passive, 4

resampling, 254cubic convolution, 254nearest-neighbour, 254

resequent streams, 70,93resolution

film, 33radiometric, 39 -' ..spatial, 19-24,39,42,227-8 ..spectral, 39,42

Resurs (satellites), 64retina, 16,29

cross-section, 17fovea, 17,18,24,29nervecells,17,18-'"'19receptors, 17-18 .;

retinal frequency, 23Return-beam Vidicon (RBV), 61,68RGB images see red-green-blue (RGB)

imagesrhodopsin, 17rills, 72ringing, 138 'risk assessment

in exploration, 234geochemical data, 243~4.subsidence, 241-2,243volcanoes, 243

roches moutonees, 74rock discrimination, 227, Plate 9..1. , ,

colour rendition, 219generalapproach,76

RockleaDome(Austral:ia),182 ,rocks

colour, 77-8,81, 83, 84dielectric constant, 176-7hydrothermal alteration,. Plate 9.5sampling, 211thermal inertia, 165,171thermal properties, 162ultramafic; 219volcanic, Plate5.9wet, 14see also igneous rocks; metamorphic rocks;

pyroclastic rocks; sedimentary rocksrock spectra, reflection, 7-11rock types

and erosion, 69' .recognition, 76-90-and water exploration, 240-1

rods, 17-18,21,24 '

Roraima Formation (Venezuela), 198rotational transitions, 4roughness

and backscatter (radar), 177-9,187,189surface materials, 177-9,217, Plate 7,3

rubidium, concentration; Plate 8.3Russia, 59

sag ponds, 100Sahara Desert, 179,192,196salt diapirs, 102,103Salt Range (Asia), 187salts, 81salts flats, 124

side-lighting, 48stereoscopic, 48,180,197topographic accentuation, 180,197topographical effects, 180-9

radar interferometry, 203-4, 213-14, 242Radarsat, 64,197,230-1radar scattering coefficient, 176, 178radarscatterometry, 241radial drainage, 72, 73radial relief displacement, 36,37radiance, 2,154,161radiant flux density, 2radiant temperature, 11radiation-matter interactions, 161radioactivity, health risks, 244radon, health risks, 243ramp stretch, 128raster-format data, 122,207-8,225

non-image, 208-21processing, 215-18pseudocolours, 215-16si'e also multivaii~tefaster images .

rasters, 19,32,40,122raster-vector conversions, 221ratioing, 143-5

disadvantages, 144andgeobotany,145andgeochemicaldata,219and mineral mapping, 145

Rayleigh criterion, 177-8,197,198,199Rayleigh scattering, 6-7,13RBV (Return-beam Vidicon), 61,68receptors

cones,17-18,21,26-7density,. 18rods,17-18,21,24

rectangular drainage, 72, 73rectification, 254red edge, 12red-green-blue (RGB) images, 199,233,

237, Plates 5.3-5bandchoice,129,Plate5.1contrast stretching, 131~decorrelation stretching,. 142~multivariate rasters, 218-21

Red Sea, Plate 1.1Red Sea Hills (Sudan), 88,105; 193,.241redundancy

spatial, 19-20spectral, 131tonal, 19-20

reflectance, 5spectral,138,147,148,154,161reflection '.

diffuse, 5-6Lambertian, 5seismic, 238 ,;"specular, 5,177

reflection spectra, 5,7-13registration, 122, 233

in geographical information systems,221

regression analysis, in gridding, 209rejuvenation, 70,72remote sensing

atmospheric effects, 6-7criticisms, 96economics, 235and field observation, 227future prospects, 65-7historical background, 50-1spectral range, 27strategies, 227-8

288 Index

seismic reflection, and hydrocarbonexploration, 238

Selby Coalfield (UK), Plate 9.9selective radiators, 10self-organising mappers, 153shadows

optical illusions, 30radar images, 48, Plate 7.1

shales, 78,Plate9.5health risks, 243 ""'1radar images, 200 '"reflectance, 147

Shanxi Province (China), 189shapes, 69

optical illusions, 31shear zones, 104,105 "shield volcanoes, 109-10short-wavelength infrared (SWIR), 7,9,236shutter ridges, 100 .Shuttle Imaging Radar (SIR)' ,

SIR-A, 55,195,197 ,{"":"\~"'"SIR-B 56 66 i, ,

SIR-C/XsAR, 56,66,204Shuttle Multispectral Infrared Radiomet~r

(SMIRR), 55Shuttle Radar Topography Mission (SRTM),

66,204,214,229siderite, 10sideways-looking radar, 43,44,50,231sight see human visionsignal-to-noise (SIN) ratio, line sca~ners, 39silicates, 236

spectra, 10,11,172structure, 9thermal spectra, 174

silicon, energy level, 7silicon dioxide,in igneous rocks, 174sills, 85,95siltstones, 77,78,80,90,192, Plate 6.4

radar images, 200thermal inertia, 165

sinkholes, 83Si-Obonds,10,160,172SIR see Shuttle Imaging Radar (SIR)Skylab, 55skylight, 144slopes

stability assessment, 241topographic, 213types of, 91-2

SMIRR (Shuttle Multispectral InfraredRadiometer), 55

solar shaaowing, 164,165,216,223Somalia, 81sonar images, 231sour gas, formation, 239South Africa, Plate 5.10Space Imaging, 66Space Shuttle, 54

antennas, 204Challenger disaster, 57,66data, 55-7imaging systems, 56,229remote-sensing experiments, 57topographic surveys, 213-14

spatial acuity, 23-4spatial filters, 134-8spatial-frequency ffitering, 133-8,200-1

definition, 134scale issues, 134

spatter cones, 110 ...specific heat, 162 .

spectraabsorption, 5emission, 5reflection, 5,7-13transmission, 5

spectral coverage, ~d geologicalapplications, 228

spectral mapping, 52-5neural-network teChniques, 153self-organising, 153

spectral radiance, 2spectral radiant flux,. 2spectrometers, 66,141

gamma-ray, 49-50,219imaging, 49

spectroradiometers, 49,57,152flat-field calibratlori; 154output spectra, Plate 3.2

specular reflection, 5,177spessartine (garnet), 8:splines, 217

cubic, 209,210ingridding, 209-10

SPOT see Systeme Pr6batoire del'Observation de la Terre (SPOT)

sprinklers, radar images, 195spy planes, 50square filters, 138 .SRTM (Shuttle Radar Topography Mission),

66,204,214,229staffed spacecraft see manned spacecraftstaurolite, 8Stefan-Boltzmann constant, 3Stefan-Boltzmann Law, 3,9,11,58; 154, 160,

176,243stereometers, 251 ,stereometry, 24~52stereomodels, 29,220-1stereopsis, methods, 249-50stereoptic viewing, 29-30,249-50

anaglyphs, 250vertical exaggeration, 250-1

stereoscopes, 68,250,252 c,stereoscopic images

aerial photographs, 37,68-121and dip estimation, 93elevation data from, 214engineering applications, 241from non-image data, 220-1 .

in geomorphology, 229radar, 48,180,197

stereoscopic viewing, 68-121in geological mapping, 233

sand dunes, 114-16,190sandstoI)es, 77,79, 84,Plates6.1-2

and granites compared, 86-8isotopes, 219radar images, 200thermal inertia, 165,171

San Rafael Swell (US), 170, Plate 6.1, Plate 7.4Santa Barbara (US),.194Santa Ynez Mountains (US), 194saponite, 9SARsee synthetic-aperture radar (SAR)satellite images, resolution, 51satellites

geostationary, 53-4military, 50-1oceanographic,. 57-:9'(i.6, ,,' ,orbits, 53-4 '

tracking and data relay, 61see also meteqrological satellites (metsats);

orbiting Satellites;$un-synchronoussatellites '

saturation, 125,126,132Saturn-V laUI)ch vehicl~:15fj" '.~; 'i:JJ: ,~

scabland; 119 ,'"scale, 69,227-8'" "

scarps, 100scattering

Mie,7nonselective, 7Rayleigh, 6-7,13in water, 13see also backscatter{radar)

scheelite, Plate 9.5Scotland,74,99,168,Plate8.3sea level, factbrs affecting; 214Seasat, 55

radar altimeters, 214,231radar images, 195,197SAR system, 58-9

seasonal images, 69,156, Plate 5.10Sea Wide-field Sensor {Sea-WIFS), 58second derivative filters, 136Second World War, aerial photography, 50,

51sedimentary basins, structure, 238sedimentary rocks

erosion resistance, 76-7and igneouS rocks, 83-4permeability, 76-7porosity, 69, 76recognition, 76-83thermalinerna,171,173thermalspectra, 173,174vegetation cover, 77

sedimentschemically precipitated, 81-3groundwater, 241lacustrine, Plate4.4marine, 187morphology, 76-7outcrops, Plate4.2radar penetration, 179raster images, 219sampling, 211suspended,13thermal inertia, 165see also clastic sediments

seepages, hydrocarbon resources, 238-9,Plate 9.7 .

seifdunes, 115seismic events

monitoring, 99-100risk assessment, 242

snowradar images, 186jS7 !,

spectra, 14SIN ratio (signal-to-noise ratio); 39Socompa (Chile), 111 :

soilsanomalies, 14dielectric constant, 176-7.erosion, 240formation, 77,80-1geochemical surveys, 243-4roughness, 179sampling, 211 ,thermal inertia, 165

solar energy, 154absorption, 6

solar radiationabsorption, 161diurnal variations, 164and remote sensing, 4surface heating, 161

Index 289

stereoscopic vision, 19stomach cancer, 243strata, 76

horizontal, 77, 78tilted, 77

stratigraphic relationships, 90-5stratovolcanoes, 108,109,188

eruptionkinetics,110-12streams

braided, 113,119catchments, 211,212,223classification, 71consequent, 70,93drainage patterns, 70-3insequent, 70obsequent, 70,93offset, 100,101order, 211-12,223resequent, 70,93sampling from, 211-12 .,

subsequent, 70, 72types of, 70

strike, 91-3,213,233strike cut-offs, 97striping, 253

image, 255strontium, concentration, Plate 8.3structural mapping, 228structural relationships, 95-107 .

subsequent streams, 70,72subsidence

detection, 243mechanisms, 242-3risk assessment, 241-,-2

subsurface dams, 24Q-lsubsurface data, 206subtractive Prima ry colours, 26,35Sudan, Plate4.5

RedSeaHills,88,105,193,241Sudan-Eritrea border, :126, 131, 135, 136,

138,140,146, Plate 5.1, Plates 5.3-:6,Plate 5.8

Suleiman Range (Pakistan), )01sulphides, 81,236-7,7~ .Sun ~

illumination effects, 143-4,151;189-95radiation, 4surface temperature, 3

sunsets, and Mie scattering, 7Sun-synchronous satel!ites, 58,64, 171

advantages, 57-54

54

107-20-~- 70.

151,.153-4

71

thermal images, 160:;-75,230apalogue, 160day-time, 164,168,170,171,1.72, Plates

6.1-3detectors, 160digital, 160false-colour, 168-9grey-tone, 164-7interpretation, 160,162

qualitative, 164-9semiquantitative, 160,169-71,

night-time, 164,165,168,170,171,172,Plates 6.1-3

small-scale low resolution, 166-7and thermal cross-over, 164-5topographic effects, 1(i4,16f>see also infrared imageS "':'-

thermal inertia, 11,162,166,167,173, Plate6.1 .

estimation, 160.,163,167, 16~1surface materials, 165and surface temperature, 163-4see also apparent,thermal iner.tia (An)

Thermal Infrared Multispectral.Scanner(TIMS), 57,130,173

thorium isotopes, 219, 220thrust tectonics, 94Tibet, 59Tibetan Plateau, 96, 102, Plate~4.3tides, 214till,116,117-18tilting

faults, 100unconformities, 95

TIMS (Thermal Infrared MUlti,5pectralScanner), 57,130,173

tin, anomalies, 237, Plate 9.6 cTIN (triangular irregu\ar ri~tw.ork), 213,252nR05-1 (Television.and)nfrar~ .

Observatio;n Satel1jte-1), 57,60nRQS;;N (TeleVision and Infra,red

Observation Satelpte-N),)57, 166-7TM see Thematic Mapper '(TM).tone, 68topography

effects .on radar images, ,180-9 : .on thermal images, 164,16$ .

elevation data, 213-14visualization, 216

Toulouse (France), 62Tournachon, Gaspard-Felix (1820-1910), 50tracking and datarelaysatellit~ (TDRSs).,

61'training areas, 149,153-4training sets, 14)-2 "transition metals, spectr"".,2 ,~!"transitions '

bond-bending, 9bond-stretching, 9charge-transfer, 8electronic, 4,8rotational, 4vibrational, 4 .transmission spectra, 5 '

transmittance, 5,161water, 13

transpiration, 1.3Transvaal (South Africa), Plate 5.10transverse dunes, 115trees, Plate 9.7

false-colour images, Plftte 3.1spectra, 13

surfaces, reflectance, 143-4surface temperature, 161-2

diurnal variations, 163-4,169Earth,4,11,43,160

profiles, 163-4Sun,3

surface water, 240

surveysaerial, 51drainage, 211-12, 244geochemical, 243-4geophysical data, 210line, 210

SWIR (short-wavelength infrared), 7,9,236synclines, 104synthetic-aperture radar (SAR), 46

applications, 53,58-9,64;2~0-razimuth resolution, 46depression ~ngles, 66 .

Doppler shift, 46holograms, 46,47 ,,::r-:,images, 65,176 ..:" :"':'"SYria, 191 .'.'.. .,

Sytte~e P~batoire de l'Observation delaTerre(5POT),157

data, 62false-colour limitations, 129ground stations, 62sensors, 62-3spectral bands, 58stereoscopic images, 214,229stereoscopic potential, 62,249

Tabriz (Iran), 188tar see hydrocarbon resourcestargeting, 234-6,237,Plate9,6TDRSs (tracking and data relay satellites),

61tectonics, 95-107telemetry ; line scanners, 39television

additive primary colours, 26,34-5colour transmission, 27

Television and Infrared ObserVationSatellite (nR05-1), 57,60

Television and Infrared ObservationSatellite (nR05-N), 157,166-7 .

terraces,114terrain analysis, 228,230Terra (satellite), 66Texas (US), 107texture" 68'

image variance, 155radar data analysis, 199-200,201.-:2,203,

227Thematic Mapper (TM)

applications, 237 \,¥~:l. .characteristics, 61-2 ":'c'" .

false-colour images, Plate 3,3limitations, )29-30

image size, 131ratio images, 145reflected bands, 138thermal images, 166wavebands, 61,142,153

thermal capacity, 162,163thermal conductivity, 162thermal cross-over, and thermal images,

164-5thermal data

in classification, 171see also multispectral thermal data

thermal diffusivity, 162

290 Index

gases~ 110parasItic, 188radar images, 188,191,199,201risk assessment, 243shield, 109-10thermal images, 167see also lavas; stratovolcanoes

trellis drainage, 72-3,77,80trend-surface analysis

in gridding, 209regional fields, 217residual fields, 217-18

triangular irregular network (TIN), 213,252

tristimulus colour theory, 28truncated spurs, 100tuffs, 84,110, Plate 6.4Turkey, 242

UTM projection (Universal TransverseMercator projection), 254

valley glaciation, 73-4valley shape, 72

U-shaped,74,116,119V-shaped, 70,93

valley V's, 93,105,106,107variance, 139,155vector-format data, 207-8,221,225

non-image, 208vegetation

anomalies, 237, 239band choice, 130classification, 148-9,241dielectric constant, 177and fault detection, 97in flood plains, 113geological control, 77,88,133, Plate 5.10image processing, 124,141-2,228monitoring, 61multit~mporal analysis, 156pattems, 12t}inradarimages,184,198-9inratioimages,l44recognition, 76, 77red edge, 12reflectance, 147reflectivity, 34roughness, 179and slope stability assessment, 241spectra, 12-13,144and springs, Plate 9.8stressed, 145on thermal images, 165-6thermal inertia, 164,165in water exploration, 240,241see also plants

Venezuela, 198Venn diagrams, 222Ventura Basin (US), 238Venus, 177,232very-near infrared (VNIR), 7,12,34,236-7Vesuvius (Italy), 243vibrational transitions

bond-bending, 4bond-stretching, 4

video cameras, 32Vidicon cameras, 37-8

colour, 38see also Return-beam Vidicon (RBV)

vignetting, 37Virginia(US),55,59,80vision (human) see human visionvisual cortex, 16-19,29,31

cells, 224complex, 19hypercomplex, 19simple, 19

visual perception, tonallirnitations, 26VNIR (very-near infrared), 7,12,34,

236-7Vogelkop Region (Iran), 184,186volcanic rocks, Plate 5.9volcanoes, 107-12,231,232,P/ate 7.2, Plate

9.9cones, 84,102,103,107-9,186eruption detection, 243eruption kinetics, 110-12explosive activity, 110

WamsutterArch(US),202,Plate7.3Washington (US), 116waste disposal

environmental hazards, 243sites, 242

water, 160asblackbody,13-14colour, 13depth estimation, 13dielectric constant, 176,177economic importance, 234light penetration, 13light scattering, 13molecular, 14pore, 14reflectance, 147spectra, 13-14surface, 240thermal inertia, 164thermal properties, 162transmittance, 13see also groundwater

water exploration, 239-41and rock types, 240-1

watersheds, 223,224water supplies

pollution, 240strategies, 240

water vapour, 6wavelength, 1wave propagation, 1weathering, 69,85

carbonates, 82onion-skin, 88

websites, 264-5weighting, 151West Africa, 34Western Europe, Plate3.4whiskbroom system, 52Wien's Displacement Law, 3,4,9,15;.4,160,

176wind erosion, 69,75-6Wind River Basin (US), 200, Plate 6.5, Plate

7.1Wisconsin (US), 119World Meteorological Organization, 57World War I, aerial photography, SO, 51World War II, aerial photography, 50,51Wyoming (US), 89,92,202, Plate 7.3

Wind River Basin, 200, Plate 6.5, Plate 7.1

Xinjiang (China), 183XOR operator, 222x-rays, generation, 2

Yellow River (China), 189Young's modulus, 26Young, Thomas (1773-1829), colour

theories, 26~7,218

U2 (spy plane), 50ultramafic rocks, raster images, 219ultraviolet radiation, generation, 2uncertainty principle, Heisenberg's, 19unconformities, 93,94-5,233, Plate 4.2, Plate

9.2folding, 95tilting, 95

United Kingdom (UK) see England; ScotlandUnited States (US), 86,108,115

Appalachian Plateau. 5.>,59,181, 199,Plate 4.1

Arizona, 70,86,108,115, Plate 4.1California, 101, 194, 238, Plate 5.7, Plate 9.9

Mojave Desert, 167,172, 193,Plate 6.3Colorado, 165,168Colorado River, 70,Plate4.1Goddard Space Flight Center, 55,57Grand Canyon, 70, Plate 4.1Indiana, 117Kansas, 78Kentucky,55,59,82,91Louisiana, 114Mississippi, 114Mount St Helens, 111, 112, 243Nevada, Plate6.4,Plate9.9New York, 114,118North Dakota, 117Oklahoma, 94,166Pennsylvania, 181Pittsburgh, 181,199Texas, 107Utah,71,79,80,100

San Rafael Swell, 170, Plate 6.1, Plate 7.4Virginia,55,59,80Washington, 116Wisconsin, 119Wyoming, 89,92, 202, Plate 7.3

Wind River Basin, 200, Plate 6.5, Plate7.1

United States Geological Survey, 53missions, 55

Universal Transverse Mercator (UTM)projection, 254

unmanned spacecraftadvantages, 57data, 57-64historical background, 32

unstaffed spacecraft see unmannedspacecraft

unsupervised classification, 151uranium

health risks, 243isotopes, 219,220

Uranus, 232Utah (US), 71,79,80, lOO

San Rafael Swell, 170, Plate 6.1, Plate 7.4surface materials, 148

Zimbabwe, 87zinc, anomalies, Plate 9.6