sG I S&

REMOTE SENSING NEWSFIJI USER GROUP, Number 15 (9602), August 1996

Welcome to thesecond news-letter for thisyear and againthe net or the

web or the superhighway is animportant issue. An �on ramp� tothis information highway is com-ing to or has arrived at a serviceprovider near you. It�s here to staybut it�s a moving target and infor-mation providers, informationpublishers and information con-sumers must get aboard as soonas possible and tool up to get thebenefits of information exchange.

SOPAC in Suva has built thefirst intranet service in Fiji and thecontent is growing rapidly withother sites connecting everyweek and an international gate-way is imminent. New services(content) were started in mid Au-gust which allows users to viewavailable satellite images of theregion as well as access a metaGIS database. This is describedin �WWW and GIS - a casestudy� with details on how youcan access the growing numberof sites and content. All this forthe cost of a local phone call.Don�t forget that as advised in thelast newsletter this current newsletter will also be published in PDFformat and can be viewed on the SOPAC intranet using your favour-ite web browser with Adobe Acrobat.

Prakash Narayan, from Fiji, who is currently undertaking remotesensing research at Universitaire Weiss, Strasbourg, France hasprovided an update to the list of web sites of interest to the GIS andRemote Sensing community. In addition, he has covered some ba-sics of classification as well as another image processing tutorial.He will be keeping us posted with future updates.

Management Services Division, Forestry, has contributed sev-eral articles ranging from image analysis through database devel-opment to map production. But whatever the field, there is a com-mon theme which is the building of a forest inventory containingaccurate information which can be compared over time to assistplanners, policy makers and decision makers. Correct decisions willresult in preservation of the natural resources which will lead in turnto maintaining biodiversity, erosion control, improved water quality,disaster mitigation and preservation of the coastal zone.

The three monthly Fiji GIS/RS User group meetings of May,June and July are reported. The June and July meetings were bothmilestones with the farewell of Michel Larue (SOPAC) and WolfForstreuter (MSD-Forestry). They are both founding members andactive participants of the user group and it was an appropriate timeto review why the group had been formed, what the goals of thegroup were and whether those goals were being met. The answersrespectively were: the sharing of information and technology andthe development of standards; the ability for relevant departmentsand organisations to implement and develop GIS/RS capacity inhouse while having access to technical support and data; and fi-nally Yes! the goals were being met. It should again be noted that aGIS/RS user group was formed in Kiribati and modelled on the sameguidelines as this group. The Kiribati group also meets on a monthlybasis.

There is also much more and again we welcome contributionsfrom national, regional and international users. Yes, you can reachus by Internet - see back page for the addresses!c

NEWSMAY MEETING

Fiji Land Information Support Cen-tre, 2:30 pm Tuesday 14 May 1996.

UPDATES AND NEWS

Chang Surveyors

• Fiji Sugar Corporation is still un-decided on implementing the GISproject.

Fiji Land Information SupportCentre (FLIS)

• Three consultants from New Zea-land evaluating the progress left lastweek.

• The census boundary capturingfor the Bureau of Statistics (1996boundary) is continuing.

• A hazard mapping pilot project ison the way, two coloured maps havebeen produced already and handedto parliament.

• Netware 4.0 training courses washeld by Computec, 20 staff membersattended

• SOPAC conducted MapInfo train-ing for FLIS.

• Coloured info material is availableabout FLIS for public awareness.

Fiji Posts and TelecommunicationsLtd (FPTL)

• The tender documents for the sup-ply and commissioning of a GIS sys-tem have been provided to 21 inter-ested parties.

Lukemine Enterprises

• USA must be testing the position-ing the GPS satellites as last weekbetween 12:00 and 12:50 local timethe PDOP was absolutely out ofrange. Later everything was runningfine.

Mineral Resources Department(MRD)

• 18 people attended a data man-agement training course.

• Tests being conducted withSOPAC to establish a Web Server.

Management Services Division,Forestry Department (MSD)

• The Forest Type / Forest FunctionMaps arrived, 100 sheets per map,coloured off set printing. Thesemaps cover also Vanua Levu in thenew topographic map sheet parti-tion.

• Digital data showing the road net-work and river system (includingthe smallest creeks) arrived on CD-ROM and will be available for everyuser.

• All plantation areas are mappedat 1:10,000 scale except two planta-tions in Viti Levu where a coordi-nate shift has to be carried out,which is still not clarified yet.

• MSD is working on the update ofthe Forest Type / Forest Functionmaps, these maps will be plottedagain for all sheets where a notice-able change in plantation area oc-curred.

• MSD is able to plot maps withdetailed road network and river sys-tem

• E-mail transfer of GPS rover datafrom Vanua Levu to MSD-forestryvia SOPAC is working operational

• Osea Tuinivanua is co-investiga-tor of the MOMS program. He man-aged that data of Fiji will be re-corded in June and August this year.MOMS is a German sensor based inthe Russian permanent space stationproviding 4.5 m data in panchro-matic mode and 14 m data in

2n2

multispectral mode.

• Josua Wakolo is attending a 10 weeks trainingcourse in Japan on remote sensing. Japan is pro-viding this course to assist the actual users ofspace borne data.

• MSD is asking about status of data sharing

Native Lands Trust Board (NLTB)

• the data translation is in progress, NLTB is nowworking testing MapInfo.

Public Works Department (PWD)

• PWD is investigating the ownership of sew-age pump stations. The stations have been GPSsurveyed with the help of Forestry and others.PWD is now investigating the accuracy of handheld GPS rovers.

• one staff member has been sent to a MapInfocourse to learn accurately digitising.

South Pacific Applied Geoscience Commission(SOPAC)

• Steve Solomon from Canadian Geological Sur-vey who replaced Don Forbes at SOPAC was in-troduced and announced that the Suva vulner-ability study has progressed to the point wherethe data collection is nearly complete on thecoastal zone along the Suva waterfront. Datacollection involved surveying elevations andmapping coastal types and attributes. Most ofthe data has been input into MapInfo GIS. Theanalysis phase will start in the next few weeks.

• SOPAC completed the a one week MapInfotraining workshop for 20 participants from thegovernment sectors in Fiji where FLIS was theadministrator. A further training workshop willbe conducted for 15-20 participants in Tarawa,Kiribati, in June. Again, a training CD will beoptionally provided to each government depart-ment in addition to course notes provided to allparticipants. The CD will contain relevantdatasets from Kiribati provided by both SOPACand Kiribati government as well as course ma-terial which will allow the departments to con-duct their own in-house follow-up training. Thiswill mean that participants will have relevantmaterials sufficient to become trainers withintheir own departments.

• Work on the Environmental Information Sys-tem at DOE is progressing on schedule and thenetwork is operational.

University of the South Pacific (USP)

• Next semester an advanced GIS course will beoffered.

• Progress is being made towards the GIS Cer-tificate Course which should be available nextyear.

World Wildlife Foundation (WWF)

• No developments or planned implementationof GIS.

PRESENTATIONS

SmallWorld - Steven Robertson (SmallWorldsoftware)

SmallWorld is a UK based company whichstarted in Cambridge in 1988 following an in-depth market survey. There are now 325 instal-lations word wide with some 3,000 licenses. Since1994 the company has an office in New Zealand.The headquarters for the South Pacific IslandCountries is in Melbourne.

The software is primarily targeted at utilities(water, electrical, gas, telecomms) due to its net-working capabilities and asset management fea-tures. The software is running under WindowsNT, before it was written for UNIX. The packagecosts about $AUS 10,000. The program has a in-built vector to raster conversion and vice versa.

New Technologies for Teaching GIS - FranckMartin, SOPAC

The use of screen capture software which allowsvoice overlay was used for the recent MapInfocourse conducted by SOPAC and administeredby FLIS. This software enabled a series of tutori-als to be recorded to disc files for subsequentplayback. The files were then transferred to a CDtogether with baseline data, SOPAC KnowledgeBase and MapInfo’s own tutorial. The Knowl-edge Base is an Access database of questions andanswers from the MapInfo-I news list and is nowin excess of 3 MB in size. Solutions can besearched by subject. There was interest by sev-

eral organisations who requested copies of theCD.

FLIS Network Upgrade & Tests - MarkWilliams, FLIS

The problems associated with network growthwere detailed and the need for network manag-ers to have access to appropriate measuringequipment to ensure that the heath of their net-works is maintained. Problems associated withtotal cables lengths being in excess of the recom-mended 185 metre maximum were identifies withtwo areas at 212 metres. The solution was to breakthese segments with a router or bridge.

FLIS is migrating from 3.11 to 4.1 and an up-grade strategy is in place.

Digital River System and Road Network of1:50,000 Maps of Fiji new Map Series availableat MSD, Forestry Department - Wolf Forstreuter

Wolf explained the digital infrastructure dataavailable at MSD-Forestry. The Lands Depart-ment provided two sets of transparencies show-ing: a) the creeks and rivers, b) the road network,villages, village names, road names and rivernames. These transparencies have been scannedin Germany employing a high resolution scan-ner. For Vanua Levu the transparencies were onlyavailable in the old topographic map series cov-erage. These transparencies have been cut and

WWW & GIS– a case study: SOPAC Services –

n by Franck Martin, SOPAC, E-mail: franck@sopac.org.fj

Every day we read about the benefits of Internet and the World Wide Web(WWW). In our GIS community, someone may wonder how it could improve theirwork, and most important what is available now and at what cost.

SOPAC is deeply involved in the promotion and development capabilities in thegeosciences in the Pacific Island Countries and we use GIS and remote sensingas well as new communication technologies to provide information about the Pa-cific. Part of these new technologies is the installation of a WWW server at SOPACheadquarters in Suva. This WWW server was designed to provide information firstto SOPAC staff, secondly to other geologists and scientists, thirdly to the people ofthe Pacific, and finally to the public.

First: information for SOPAC staffInformation at SOPAC is in different formats. There are reports, maps and

databases. Databases are developed under MS-Access. For one who knows MS-Access or other database software, it is cumbersome to load a 5Mb software ap-plication plus a 5Mb database in your computer to just require a simple phonenumber. Fortunately the WWW is here to help us. Microsoft Internet InformationServer (IIS) has a built-in interface to request data via Open Database Connectiv-

n3

reassembled to the new sheet coverage beforescanning. Now all infrastructure data is avail-able in digital form related to the new map sheetcoverage. Wolf explained that it is difficult to usethe high resolution data set because of the filesize which creates difficulties for the hardwareinstalled at MSD-Forestry or Lands Department.However, MSD provided a way to resemble thefiles to a resolution easy to handle and accurateenough to be used for map printing. he also dem-onstrated maps plotted at MSD-Forestry with theinfrastructure. The Lands Department will usethe infrastructure data sets.

FLIS-GIS Upgrade New Features - AnselmHaanen, FLIS

The new Intergraph MGE system for CCMS wasdetailed and the provision of increased user ac-cess to the topographic database outlined.

Investigations underway with IntergraphVistaMap as a front end to Intergraph MGE.

The improved set of tools with the newIntergraph MGE was detailed as well as an en-hanced MGE to MapInfo translator.

The usual problems of dependant upgradeswas outlined and in the case of FLIS there havebeen the anticipated teething problems with mi-grating the GIS, backend database (Oracle 6 to 7)and network server (Netware 3.11 to 4.1) at the

same time. However, the systems are now fullyoperational.

DISCUSSION

• Kevin McConnell (Lukemine Enterprises) ad-vised on a project which had checked the loca-tion of 200 pumping stations and how whilemost were very close to FLIS data, some wereseveral tens of metres out. Field work is neededto check these fixed installations and a systemfor revising the digital data will be implemented.

• USP requested feedback on what was the mostpopular GIS/Desktop Mapping application inuse in Fiji and was advised by participants thatthe common standard was MapInfo. SOPACstated that they had been quoted USD 43,000 forArcInfo for NT which did not favour any pro-motion for this product.

JUNE MEETINGManagement Services Division, Forestry Depart-ment, 2:00 pm Tuesday 4 June 1996

UPDATES AND NEWS

Osea Tuinivanua highlighted during his wel-come that the European Union is the largest do-nor to forestry. Besides the plantation mappingproject which is ending now, the EU finances theForestry Management School which will be es-tablished in Colo-i-Suva.

ity (ODBC) to any kind of database and to display it in Hypertext Mark-up Lan-guage (HTML). HTML is the format read by WWW browsers such as Netscape orInternet Explorer. These browsers are simple and fast as they are just �viewers� ofdata. Another advantage is their availability on a wide range of platforms such asPC, Macintosh, Unix�. Your database developments suddenly reach a wider au-dience. For instance, SOPAC has a contacts database of people known by SOPACwho receive information from us, such as member country representatives, do-nors, scientists. To obtain the phone number or e-mail address of one of thesepersons takes several minutes with MS-Access but using Netscape, the wholeoperation is completed in less than 1 minute.

Second: Scientific information for the scientific communityOur geologists and coastal scientists are producing more than 40 technical

reports every year, highlighting the geology and survey work in the South Pacific.These reports are freely available in SOPAC but you need to visit our premises tosee it or request for a hard copy. We are currently converting these reports toAdobe Portable Document Format (PDF). This format saves the document com-

Figure 1. SOPAC spatial dataset index.

4n4

Chang Surveyors

Their project was presented and the need for aGIS explained. However, there are still some fi-nancial problems to purchase a system.

Fiji Land Information Support Centre (FLIS)

• Three consultants from DOSLI visited FLISfor three weeks for consolidation andstabilisation of operational systems.

• CCMS continues on extracting metric cadas-tral sheets and should be completed soon.

• topographical data base has been establishedwith three staff working on the verification andcleaning of digital topo data sent from NZ.

• Two NLC files on DXF format had been suc-cessfully translated by FLIS to DGN format

• the hazard/earthquake pilot project is cur-rently in progress.

• cadastral data transfer from FLIS to NLTB forALTA review exercise is completed

Lukemine Enterprises

new GIS and GPS equipment will arrive at hiscompany by end of June. This new generation of

GIS will provide sub centimetre accuracy..

Mineral Resources Department (MRD)

• MRD has now a GTM package for mining con-trol.

• the Internet connection to SOPAC is fully in-stalled.

Management Services Division, Forestry De-partment (MSD)

• The EU funded plantation is completed. 71maps are plotted at 1:10,000 scale. These mapswere also imported to the Forest Cover / ForestFunction maps to update the plantation area.Plantation image maps were produced at1:50,000 scale.

• the transformation problem for two plantationsin Viti Levu is solved by employing GCPs col-lected by GPS survey and ARC-INFO transfor-mation modules

• Another SPOT scene is ordered which coversthe north of Viti Levu

• Digital data capture by stereo-mapping is nowconcentrating on logging areas after completingplantation areas

• a consulting team from NZ is looking for thepotential to privatise Fiji’s hardwood plantations

South Pacific Applied Geoscience Commission(SOPAC)

• explained problems converting data from UTMto FMG. There will be a note in the next GIS &Remote Sensing newsletter.

• Franck Martin will be conducting a MapInfoworkshop in Tarawa, Kiribati during the weekcommencing 17 June. The audience will be stafffrom government departments. A GIS & RemoteSensing User group will be formed and regularmonthly meetings recommended. The aims ofthe group and the structure of meetings will bemodelled on the Fiji group. A report on the out-come will be presented at the next meeting.

• The Suva Vulnerability Study is continuing anda report will be presented at the next meeting.

PRESENTATIONS

Overview on the EU Funded Forest MappingProject, Wolf Forstreuter, MSD-Forestry

The purpose of the plantation mapping projectwas not only to carry out a the mapping at1:10,000 scale, but to enable MSD to do the map-ping and to integrate the mapping in the forestmonitoring system. Wolf explained that this isnecessary because a change of plantation areainduces a change of natural forest cover. He alsodemonstrated the limits of GPS survey in forestenvironment. GPS was supposed to be used formapping all plantation boundaries, however, itwas partly replaced by stereo-mapping due tolow accessibility of positioning satellites inforested valleys. Wolf also highlighted the needfor satellite data for monitoring the developmentof the plantations. A detailed article will be pub-lished in the next GIS and Remote Sensing News-letter.

Plantation Image Maps, Asesela Wata, MSD-Forestry

Asesela Wata explained the way to produce sat-ellite images showing the forest canopy and theplantation boundaries at a scale of 1:50,000. Sub-sets covering the plantation area have to be cutout of georeferenced satellite images. Knowingthe location and age of plantation sub-areas(compartments and stands) it is possible to carry

pressed with all its settings allowing anyone to view and print the document with-out altering its presentation. At the same time we have converted GIS cataloguesfrom different organisations as well as the regular Fiji GIS and Remote SensingNewsletters. Each document takes about 10 minutes to download on a 14.400kbps line.

It is also interesting to browse through the data sets available and retrieveinformation such as: data type, custodian, accuracy, availability, updates, loca-tion� This is done through a link to the Spatial DataSet Index database developedby SOPAC for the Department of Environment (see Figure 1, previous page).

SOPAC has purchased some SPOT images. Someone may wish to have aquick look at these images before planning any work. This is already available andyou can request this information by pointing to a location on a map. This creates asmall picture of the SPOT image giving enough detail for you to consider the im-age worthwhile (see Figure 2, facing page).

Third: PEACESAT to provide information to the PACIFICPEACESAT is a network of Satellite Stations implemented in many Pacific Is-

lands Countries. In addition to voice, these stations allow computers to exchangedata. Therefore with a simple configuration, anyone using PEACESAT can accessthe SOPAC PEACESAT terminal and from there browse our WWW server.

Last: Information FOR the publicIf you own a computer with a modem, you can freely connect to our WWW

server by dialling 381023 and providing the username meredith with the pass-word 123xyz. Your computer must have TCP/IP PPP software with PAP authenti-cation. For example, this is built-in to Windows 95 and Windows NT. You set yourgateway as 202.52.1.155 and the DNS as 202.62.0.129, and you can connect tothe URL: http://www.sopac.org.fj/. Please contact us if you need more information.The only cost is a normal phone call to this number (381023). Of course you arelimited to browsing the local resources but it will take you time to visit everything

and the number of WWW servers con-nected to our Intranet hub is growing.

And much more...If you want to be kept updated about

the GIS/RS community of the SouthPacific, you can subscribe to the mail-ing list: GIS-PACnet@sopac.org.fj bysending a mail containing your e-mailaddress to GIS-PACnet@sopac.org.fj.c

n5

out an object specific contrast enhancement. Thenthe plantation boundaries are overlaid onto theimage. Asesela demonstrated that it is possibleto show with such an product sub-areas of a plan-tation in which:

• Mahogany dominates the canopy,

• the natural forest dominates the canopy,

• the natural forest cover is still present, but dy-ing due to poisoning.

Further investigations will enable MSD to sepa-rate different plantation species. Asesela Wataexplained that plantation image maps cannot re-place a 1:10,000 mapping, but it is a managementtool for monitoring forest change and develop-ment. A more detailed article will be presentedin the next GIS and Remote Sensing Newsletter.

Results of Forest Stratification Employing Dig-ital Satellite Data, Osea Tuinivanua, MSD-For-estry

Stratification of Fiji’s natural forests into forestspecies classes is necessary for better forest man-agement. The availability of digital satellite datahas cost-effectively enhanced the approach. Byapplying Landsat TM data, using appropriateimage processing software (ERDAS) with localknowledge and in local situation preliminary re-sults have been obtained. Using the Navua RiverCatchment (103,000) as a test site, approximatelyfive tree species association have been classified.Field validation is continuing to authenticatethese findings. This is quite a new developmentin the country as Fiji is solely relying on aerialphotographs in the past. Digital satellite data willbe readily available in the future and will becomecheaper too.

Map Production with ARC-INFO, VelemaniLatiLevu, MSD-Forestry

ARC-INFO software was already available atMSD, when the plantation mapping projectstarted. It provides raster vector conversion andvice versa and can be easily linked to a dBASEdata bank. The spatial plantation data is storedin ARC-INFO. Velemani explained the differentdata input which was done by stereo-mapping,digitising and GPS survey. The link to the rela-tional plantation data bank updates the actualstocked area. The connected plotter allowed theproduction of more than 70 maps on transpar-encies from dyeline copies can be produced tolow cost. An article will be included in the nextGIS and Remote Sensing Newsletter.

Forestry Application of MapInfo, OwenSpringford, Consultant, MSD-Forestry

Owen Springford is involved in the evaluationof Fiji’s hardwood plantations. He found the spa-tial data resource needed at MSD-Forestry.MapInfo which is installed on his laptop com-puter has not the technical potential of ARC-INFO, but he was able to utilise all spatial datawithin a short time and without any transfor-mation software. He also was able to take hislaptop containing the spatial and non-spatialdata about the plantations into the field for veri-fication.

Purpose of Forest Monitoring, History + Dis-cussion, Wolf Forstreuter, MSD-Forestry

Wolf gave a review of the GIS and Remote Sens-ing User’s group history. This meeting startedin Forestry about five years ago. With the estab-

lishment of FLIS the meeting grew and becamevery regular. During the years the contentchanged. In the beginning the information aboutupdates dominated the meeting later presenta-tions about special application or new technol-ogy. The meeting became an important informa-tion platform and Fiji is one of the few countrieswhere people agreed on a common data format(DXF) and this is because staff on technical levelhave the mandate to meet regularly. For the fu-ture, SOPAC will become an important role be-cause a regional organisation with the mandatefor GIS and Remote Sensing can provide techni-cal expertise required by national organisations.Wolf’s job was also to introduce satellite RemoteSensing to Forestry. This specially required byForestry because Forestry Departments in tropi-cal countries have to monitor the change of for-est cover and digital satellite images are the ap-propriate information. In the field of RemoteSensing the meeting still has to forward someideas. Data sharing between Government organi-sation must be addressed to the satellite sellingcompanies. The idea of a transportable groundreceiving station must be discussed on decisionmaker level and then between the South PacificIsland Countries.

DISCUSSION

The question was raised why compartmentboundaries within a plantation do not follow themataqali boundaries, because this would avoidmuch problems during the harvesting of the tim-ber later. At present the silvicultural managementunits are not linked to the ownership bounda-ries which will create discussion on a later stage.

MSD answered that it is already a technicalproblem to establish the plantation boundariesalong the mataqali boundaries. To mark mataqali

Figure 2. SOPAC SPOT Image search toolkit.

6n6

boundaries within a plantation cannot be man-date of a forestry division. MSD is waiting forthe digital boundaries provided by NLTB viaFLIS.

The question was raised which type of poi-son is used for chemical thinning.

MSD answered that Forestry was using ar-senic poison and not the type employed in Viet-nam by USA.

The subject of standards for exchange of datawas raised and it was agreed that new stand-ards are available in Australia which will betterserve the needs of users and overcome the re-strictions of DXF. The chairman suggested thatthis become a subject or theme for the next meet-ing.

The discussion of having a spatial data cata-logue for every organisation ended with follow-ing statements:

• An Intranet/Internet Web site is the most ef-fective.

• The catalogues should updated as normal textdocuments without standard format for the timebeing.

• The catalogues should be updated as soon aspossible and stored at both FLIS and SOPAC.

• SOPAC will set up a home page on Intranet/Internet for remote access.

JULY MEETINGSouth Pacific Applied Geoscience Commission,2:30 pm Tuesday 9 July 1996

UPDATES AND NEWS

Fiji Land Information Support Centre (FLIS)

Stage 1 Programme

• Extraction of all metric cadastral sheets com-pleted.

• 1:1000,1:5000,1:10000 scales

• Data verification and cleansing continue onseveral existing operational systems like VKBand Titles Journal.

• Tests were conducted on cables/file servers in

relation to access response time and connectiv-ity. As a result Cables and antenna on all FLISsystems have been repaired or changed due tocorrosion.

• Policy on Data Sharing between Governmentagencies had been endorsed by FLIC and a semi-nar is to be conducted on this policy before im-plementation.

Stage 2 Programme

• Cleaning/verification of digital topo data isprogress. Digital contour data is now available.

Projects

• Hazard/Earthquake Pilot continues

• NLC Pilot Project is in progress

• Draft-Scooping document for Natural ResourcePilot Project completed.

• Extraction of Bureau of Statistics maps willcommence tomorrow (10/7/96) and the finalbatch of the 1996 boundaries have been receivedand digitising in progress.

Training

• Four officers from FLIC member agencies havebeen nominated to attend the three months GIStraining in Australia towards the end of this year,1996. (PWD, NLTB, DTCP, Environment)

• 28 officers from FLIC member agencies sat theirpapers at the USP during the month of June andare still awaiting results.

n by Prakash Narayan, Cité Universitaire WEISS C120, 7 Quai du Bruckhof,67089 Strasbourg, France

Histogram, Transfer function, Palette, Composite colour, Image raster-vector vec-tor-raster, Superpostion, Geometric rectification, Digitial Elevation Model(DEM).

Histogram

The histogram is a very important tool for image processing applications. It is agraphical representation of the distribution of shades of colour values used or inanother words a visual representation showing the number of pixels belonging toeach of the colour values in an image. With the help of the histogram we are ableto see which intervals of colour values appear most in an image. The interest tosee this interval is to determine whether this interval is a narrow or a wide one. Ifthe interval is a narrow one then it means only a very limited numbers of colourvalues are used from among all those colour values available.

The humain eye can not differenicate the changing of very close colours thereforemaximum information from an image is not revealed if the interval is a narrow one.This narrow interval of colour values are streched to include values of all the colourvalues available. This transformation enables the humain eye to differenciate bet-ter the changing of colours which leads to a better visual representation of theimage.

A Practical Approach to ImageProcessing in Remote Sensing

Transfer function

After having visualized the histogramof an image, it is then to determine ifthe interval of color values used is anarrow or a wide one. If the interval is anarrow one, it is concluded that there isa poor contrast in the image. The con-trast is the variation of the shades ofcolours in relation to one another. Thelower and upper limits of this interval aredetermined from the histogram. A trans-fer function is then applied within thisinterval. Several types of transfer func-tions are available in an image process-ing software such as linear,equipopulation, logarithms, constantand point transfer functions. The newlook up table is produced starting fromthe lower limit to the upper limit of thechosen interval of the histogram. Thecolour values outside this interval areattributed the value zero or any otherdesired value.

n7

Publicity/Awareness Programme

• FLIS news Edition 14 has been completed andnow ready for distribution.

• FLIS newly designed brochure are currentlybeing distributed.

• The Permanent Secretaries meeting was hostedby FLIS. FLIS Support Centre made a presenta-tion on the FLIS Programme and also organisedsite visits around existing operational projectsafter the meeting.

Housing Authority

• Although the first time at these meetings thisorganisation was very interested in developmentin GIS and in particular standards which havebeen developed and adopted in Fiji. Future par-ticipation is expected and it was noted that therehas been involvement with PWD, FEA, Telecom.

Management Services Division, Forestry De-partment (MSD)

Completion of Wolf Forstreuter’s EU funded projecton all hardwood plantation

The following targets have been met:

Data capture - based map 1:1000 and schemeplan

• Digital re-mapping projected to FMG. Estab-lishment of spatial data base link to:

• dBASE inventory plantation data bank

• Erdas raster GIS (1:50000) scale)

Production of Maps

Updates of Arc/Info System - Correction ofBaravi plantation:

Field visits using GPS - for locations of GCPs

GPS data - Arc/info for co-ordinate transforma-tion

Update existing BRV plantation map

Update of Forest Cover forest function - 1:50000

GIS Remote Sensing

Update forest cover forest function map - 028Lands Department Map.

Information overlay district boundary with for-est cover forest function.

Area statistics calculation.

Update of natural forest data bank.

Process digital river and road network

An example of vector based information con-verted to raster TIFF format, initially extractedfrom the original transparencies from Lands De-partment and scanned for production of digitalreproduction of transparencies then convertedand stored in compressed file format for spaceefficiency

Map Distribution - Forest cover forest functionmap - distributed to some organisations, includ-ing NLTB, Director of Lands, Koronovia LandSection, DFO N, S, W

Another SPOT scene, covering North of Viti Levuhas just arrived.

South Pacific Applied Geoscience Commission(SOPAC)

• Intranet web services established and testsclearly demonstrate that this media is the mosteffective method for distribution and publishingof GIS information for corporate users, nationaland international users.

• The Kiribati GIS & Remote Sensing User Groupinitiated following the MapInfo workshop con-ducted by SOPAC in Tarawa during June 1996.

• The Suva Vulnerability Study is nearing com-pletion and brief report to be presented at thismeeting.

• Michel Larue’s final meeting.

Palette

The palette defines the number of shades of colour values available to displayan image file. For 8 bit visualization 256 shades of colours are possible (2^8 =256),for 16 bit visualization 65536 shades of colours are possible (2^16= 65536). Themore shades of colours available the better it is to differenicate between shades ofcolours. To have a visual representation of colours, a combination of Red, Blueand Green (RBG) colour components are attributed to each colour values.

Composite colour

The notion of the three fondamental colours Red, Blue and Green is utilised.These three basic colours are affected to one of the image bands. It is evident thatwhen using this technique only three bands at a time can be used to display animage. For those images having more than three bands, a transformation is per-formed to produce only three new bands using information from all of the bands ofthe image.

An example using a SPOT image : band XS1 is affected to colour Blue, bandXS2 is affected to colour Green and band XS3 is affected to colour Red.

Image raster-vector, vector-raster

An image is a matrix of pixel numerical values. Each matrix coordinates has anumerical value attributed to it. This is a representation of the whole image and isrefered to as a raster image file.

When only a limited amount of information is extracted from an image, it be-comes evident that fewer stocking space is required. Thus if a matrix representa-tion is used, there will be some coordinates that will not be used, a waste of stock-ing space.

So only for the coordinates for which the pixels are used, the pixel�s numericalvalues are stocked in a file which is referred to as a vector image file. It is possibleto superpose a vector file onto a raster file or vice versa using the appropriate fileconversion transformations.

Superpostion

This is the superpostion of two im-age files that contain two different typesof information for the same image file.

An application example : Superpose acomposite colour image file with animage file that contains only vegetationinformation of that image file.

To obtain the vegetation information, theNDVI is calculated using the band datavalues. A filtre is then passed to smoothout the contrast (ex. average filter). Aspecfic filter is then passed to find thecontours that are supposed to containonly vegetation information. Using com-mands appropriate to the software beenused the composite colour image is su-perposed onto the image containing onlythe vegetation information. We will havea composite colour image with the con-

8n8

tours of the vegetation boundaries on topof this composite colour image. It is thecomposite colour image that is modified.

Geometric rectification :

A reference image is used to carryout a geometric rectification of an im-age. A reference image is an image thatis supposed to be «correct», a map foran example.

Ground control points are chosenfrom the image to be rectificied with theircorresponding points in the referenceimage. An image is rectified by using oneof the rectifying methods proposed bythe image processing software beenused.

An example using the triangulationmethod: The control points are graphi-cally chosen using the zooming facilitiesfor finer point locations. The trianglepoints chosen are visualized to see thatthe triangle do not cross each other.More ground control points are addedor removed as the need maybe. Apolynom degree is chosen (degree 1,2or 3 ) for which the residue is examined.

Ground control points with a largeresidue are also removed. Triangles areconstructed before the actual rectifica-tion process is carried out by using theappropriate software modules.

Digitial Elevation Model (DEM)

DEM values are obtained from twoimages of the same scene taken fromtwo different angles using stereoscopic.The DEM is used to represent an im-age in a 3 dimension representation.The DEM value file is used in conjunc-tion with an image file to produce the 3dimensional representation.c

University of the South Pacific (USP)

As announced in past meetings, USP will offeran advanced GIS course next semester, calledGE305: Spatial Topics in Geography, AdvancedGIS. Tentatively, it will be taught by Mele Rakaiand/or Bruce Davis. The venue will be SSED,rooms S009 and S008. The times will be Monday4:00 - 6:00 pm lecture. Tuesday 5:00 - 7:00 pmlab.

The times are to support working profession-als as well as students. Alternate times may beavailable. It will be a hands-on course, with afull week to complete assignments usually.

Please note that it should not be as high-pres-sure as GE204.

Topics include: Data quality and standards,integration of remote sensing data into GIS, useof GPS, advanced raster and advanced vectoranalysis, GIS database design and development,project design and development, and others. Itis meant to be a follow-on to the introductoryGE204 course.

Prerequisites: GE204 (Introductory GIS) orequivalent. Some experience with GIS softwarestrongly recommended. Second year studentssuccessfully completing GE204 are eligible.

For further information, please contact BruceDavis (davis_b@usp.ac.fj) or George Saemane(saemane_g@usp.ac.fj). For registration information,please contact the Registrar’s Office at USP.

Keresi Fonmanu (Student Melbourne Univer-sity)

Keresi described her project on NLC boundariesand advised that she has been using IDRISI, Lati-tude and Sage to analyse the data. She stronglyrecommended that students who are consider-ing undertaking post graduate studies in Aus-tralia should undertake the full range of avail-able GIS courses at USP before studying over-seas as the standard is extremely high.

PRESENTATIONS

New tools for Metadata catalogues. AnslemHaanen, FLIS

The importance of metadata was stressed whichcan make datasets more accessible, avoid dupli-cation, increase efficiency. Metadata is a databaseof data and an essential tool for an organisationhandling multiple GIS data sets. World-widestandards are being developed and a presenta-tion was made of the ANZLIC system whichaddresses privacy where this has become anemerging concern. ANZLIC has proposed aminimum number of core elements via a nationalstrategy which trades off the minimum numberof elements and the needs of the users.

It was noted that Microsoft Access has beenselected by ANZLIC as the database of choicefor developing and collecting the Australian andNew Zealand data for subsequent publishing ofthe national or regional metadata catalogue.

SOPAC Intranet/Internet Web Server as an ac-cess tool for Metadata catalogues. Franck Mar-tin, SOPAC

The dramatic increase in acceptance of Web serv-ers as a method of distributing and publishinginformation world-wide has led many compa-nies to adopt Web servers and browsers as ameans of accessing data at the corporate, nationaland international level. The areas of GIS andRemote Sensing have been traditional users ofthe Internet for publishing their data. Two ex-amples are the Canadian Geological Survey andthe satellite image provider, SPOT. SOPAC hasdeveloped a Web Server for use within the or-ganisation as well as other users in Fiji via anIntranet. MapInfo is shortly releasing an upgradeto MapInfo Pro (4.1) as well as MapInfo Pro

Server which will allow clients running popularBrowsers such as Netscape and Internet Explorerto query Web servers with GIS functionally pro-vided by MapInfo Pro Server which passes CGIor API requests to a MapInfo session which ac-cesses databases. The promise of this new tech-nology was demonstrated via the Intranet.

Kiribati GIS & Remote Sensing User Group.Franck Martin, SOPAC

Franck Martin reported on MapInfo workshopin Tarawa, Kiribati during the week commenc-ing 17 June conducted by SOPAC. The audiencewas staff from government departments. A GIS& Remote Sensing User group was formed andthe first of regular monthly meetings held. Theaims of the group and the structure of meetingswere modelled on the Fiji group. A newsletterwas produced which should serve as a record ofthe meetings as well as publicising the activitiesof the group within Kiribati. In addition, train-ing was provided on newsletter layout and de-sign using Microsoft Publisher.

A Coastal GIS for Assessing Vulnerability toRising Sea Level. Jens Kruger, SOPAC

This project is nearing completion with all sur-vey data collected and stored in a GIS. Complexcalculations have been applied to determine pos-sibility of damage through sea encroachment andtsunamis using past data for tide levels, windvelocity and directions as well as the topogra-phy of the reef. The outputs of the calculationscan be overlaid on coastal areas with high vul-nerability and varying height above sea level todetermine a vulnerability index for appropriateareas. The final results will be presented to De-partment of Environment to enable them to as-sist in the development of the Suva peninsularand enable steps to be taken to mitigate disas-ters.

Forest cover change in Western Samoa usingMapInfo. Leslie Allinson, SOPAC

In 1989 a GIS of Western Samoa was created inArc/Info under an ADB funded land use projectfor the government departments responsible forland survey, agriculture and forestry. The datain the GIS was nearly lost due to hard drive fail-ure as well as a tropical cyclone which ruinedthe computer system. However, the data wasrecovered and taken to the Australian NationalUniversity for the background in a land use pub-lication and later the data returned to Samoa.However, the usefulness of the data was severelylimited due to it being in Arc/Info format andstaff at the Department of Lands and Survey ar-ranged for both spatial and attribute data to betranslated into MapInfo at another Australianuniversity. The data set was then converted tothe Western Samoa Integrated Grid datum andthe Department began adding new data whichincludes land parcels (Apia only at present), me-teorological stations and rainfall data. This valu-able data set also contains historical land coverfrom aerial surveys for 1954 and 1989 whichshows a 19% loss (1,348 to 1,090 sq km) of forestcover in this period in a Savaii with a 32% loss(739 to 503 sq km) in Upolu where these islandshave land areas of 1,695 and 1,130 sq km respec-tively. The purpose of the presentation was todemonstrate the ease with which data could be

n9

analysed and displayed in MapInfo.

Overview of GIS and Remote Sensing Programat SOPAC. 1992-1996. Michel Larue, SOPAC

An overview of the GIS and Remote Sensing Pro-gram at SOPAC showed that this relatively newtechnology has now evolved into a mature prod-uct and is widely accepted as a tool for a host ofapplications, in particular land management andland cover monitoring. The GIS components ofdata entry, processing, storage and retrieval wereoutlined and examples given using bathymetrydatasets. The applications required for theseprocesses range from high end specialist UNIXtools through Arc/Info, ERDAS, ER Mapper andmid range entries such as AutoCAD,Microstation and MapInfo. The bundle selectedby SOPAC as appropriate for the geology sectorin the Pacific Island Countries is MapInfo, Surferand IDRISI.

The entry stage has been typified by ana-logue to digital using scanners, raster to vectorconverters; use of digitising A0 and A3 tableswith AutoCAD and MapInfo while the process-ing stage uses Surfer, Quicksurf (AutoCAD add-in) and Vertical Mapper (MapInfo add-in) forgridding and contouring, IDRISI, AdobePhotoshop and Corel Photopaint for imageanalysis. Specialist Visual Basic tools have beenrequired for much of the offshore survey data.The storage phase uses networked file serverswith high capacity hard drives, DAT backup aswell a Exabyte tapes. CD-ROMs are used for gen-eral storage and distribution of the data whilemetadata catalogues. The holdings now include2,400 MapInfo layers

The final stage is retrieval and all roads leadto the Intranet/Internet. It was noted that prob-lems encountered included quality control, da-tabase design, GPS uncertainty and sustainabilityof data holdings.

The future lies in a regional organisation be-ing responsible for co-ordinating the data acqui-sition, data holdings and method of publishingto reduce costs and increase efficiencies.

DISCUSSION

• Following proposal by FLIS, the group thankedMichel Larue for his GIS support during the pastfour years and it was noted that Michel was anactive participant in these monthly meetings.

• General discussion on the future of Intranetservices in Fiji.

• There was general consensus that feedbackfrom FLIC would be invaluable for this meetingto determine whether the cooperation engen-dered by these meetings was providing positiveresults as perceived by policy makers.c

S A T E L L I T E

N E W SThe satellite news will concentrate

on three main topics, the EarthWatchProgramme, the Indian Satellite and theMOMS programme. There is no newsabout SPOT, JERS-1 or Landsat data,the systems work normal. For Fiji,Landsat scenes are only available ashistorical images. The latest SPOTscene of Fiji arrived at MSD in July thisyear; the scene is nearly cloud free andcovering the Monasavu area.

1. EarthWatch ProgrammeThe EarthWatch Programme will set

a new dimension in image data applica-tion in terms of availability, coverage anddelivery. As reported in earlier newslet-ters, the company EarthWatch Incorpo-rated was formed in January 1995 bymerging two companies Ball Aerospace& Technologies Corp. And WorldViewImaging Corporation. Further, the Com-pany has developed a range of world-wide strategic partnerships includingHitachi (Japan), Nuova Telespazio(Italy), CTA (USA), Datron Systems(USA), MacDonald Dettwiler (Canada).This group of companies cover satellitetechnology, downlink and data distribu-tion experience as well as RemoteSensing and GIS knowledge.EarthWatch announced that they willbuild and launch their satellites at one-tenth the cost of competing satelliteprojects. This is likely to lead to pricereductions for data.

The programme will deal two typesof satellites the Early Bird and the QuickBird. The Early Bird will be launchedDecember 1996 and Quick Bird in late1997. Both satellites have onboard stor-age capacity and will download the im-age data in either Alaska or Coloradowhere the company has build groundantennas. Early Bird is able to store 500scenes per orbit, which is an improve-ment compared to existing satellites.Both satellites have each two sensorson board The satellite and sensor char-acteristics:

Early Birdsatellite characteristics:

An Update of GIS &Remote Sensing Internet Sites

Remote Sensing/Geography Information Systems (RS/GIS) is a developing discipline therefore ac-cess to new information is very essential to keep abreast with the changes. It can be Software devel-opment, Modelling techniques, Imaging material and techniques, applications, new products or justsimply curiosity. Large amount of these information is available on the Netscape or mosaic throughInternet. Knowing where and what information is available really helps to reduce utilisation time.

1) http://www.asprs.org/asprs2) http://www.geomatics.com/rti3) http://www.acsm-hqtrs.org/acsm4) http://www.ersi.com5) http://www.gisworld.com6) http://www.bentley.com/7) http://www.bluemarblegeo.com8) http://www.thinkspace.com9) http://www.tydac.com10) http://www.alexandria.sdc.ucsb.edu/11) http://ewse.pci.on.ca12) http://www.nbs.gov/nbii/

13) http://www.gcmd.gsfc.nasa.gov14) http://www.northwoodgeo.com15) http://www.eosat.com16) http://www.gislab.teale.ca.gov17) http://www.urisa.org18) http://world.std.com/~able19) http://ogis.org20) http://www.lib.unb.ca/GGE/21) http://www.ccrs.emr.ca22) http://www.stakart.no/isotc2111/It is personally not possible to go through all of

these RS/GIS sites available on the Internet because it will take a lot of time. Therefore comments aremost welcomed to find the better sites and in the process new sites can be added to update the listregularly .c

n Prakash Narayan, Cité Universitaire WEISS C120, 7 Quai du Bruckhof, 67089 Strasbourg, France

10n10

orbit: sun-synchronousaccessible swath: + 280 kmheight: 470 km

panchromatic sensor:scene coverage: 3 x 3 kmspatial resolution: 3 mspectral coverage: 450 - 800 nm (panchromatic)

multispectral sensor:scene coverage: 15 x 15 kmspatial resolution: 15 mspectral bands: 500 - 590 nm (green)

610 - 680 nm (red)790 - 890 nm (infrared)

Quick Bird:satellite characteristics:orbit: medium inclinationaccessible swath: + 360 kmheight: 600 km

panchromatic sensor:scene coverage: 27 x 27 kmspatial resolution: 1 mspectral coverage: 450 - 900 nm (panchromatic)

multispectral sensor:scene coverage: 27 x 27 kmspatial resolution: 4 mspectral bands: 450 - 520 nm (blue)

520 - 600 nm (green)610 - 680 nm (red)790 - 900 nm (infrared)

The company will develop a systemto send the data by Internet to the user.The user does not become the ownerof the data, but leases it for the purposeof a project.

2. The Indian Satellite IRS-1CThe Indian Satellite IRS-1C is the

only operational digital sensor providinghigh resolution data (5 m panchromaticdata, 25 m multispectral). The data cannow be received by the following groundreceiving stations:

� Norman, Oklahoma (USA)

� Shadnagar (India)

� Neustrelitz (Germany)

� Johannesburg (South Africa)

� Alice Springs (Australia)

� Haloyama (Japan)

� Cotopazi (Ecuador)

� Bangkok (Thailand)

For areas not covered by theseground receiving stations like South

Pacific, the satellite has a onboard tapeto store images until passing a groundreceiving station. IRC-1C has the samedata format as Landsat TM.

Questions and data purchase can beaddressed to:

EOSAT4300 Forbes Boulevard,Lanham, MD 20706Fax: +1-301-552 3762

NRSADept of Space, Govt. of IndiaBalanagar, Hyderabad 500 037Andra Pradesh, IndiaFax: +91-40-278 664, +91-40-278 648

3. The MOMS ProgrammeThe MOMS-2P sensor is onboard

the permanent Russian space stationMIR. On 7 May 1996, the space shuttlePROGRESS carryiny MOMS sensorwas connected to MIR. Data recordingis postponed until October 1996. MOMSwill record multispectral and panchro-matic data from Fiji, because OseaTuinivanua1 is a co-investigator in theMOMS programme.c

____________1For further information contact: Osea Tuinivanua,Fax: +49-761-203 3701. He is head of MSD-For-estry

Digital River System and RoadNetwork available at MSD

n by Wolf Forstreuter, MSD-Forestry

IntroductionOne target of Fiji’s Natural Forest InventoryProject was the production of Forest Cover /Forest Function maps. During the planning proc-ess it was assumed that the infrastructure wasavailable in digital form in Fiji, which was notthe case. The first maps provided were not ac-ceptable. After the visit of a GOPA representa-tive to Fiji it was decided to present the revisedmaps in the new Lands Department maps sheetcoverage. The only acceptable way was to scanthe infrastructure from available informationand combine this data digitally with the forestinformation. This was an additional and neces-sary step so that Fiji’s requirements could be met.Fiji’s Forestry Department wanted to benefitother Departments and asked for the digital datasent with the physical maps to Fiji. Both arrived,during the last weeks.

Data SourceThe Lands Department provided copies of theoriginal transparencies of the infrastructure1

without additional costs besides material andovertime. These transparencies were scanned bya company in Bremen/Germany. In order to scanthe road network and rivers for Vanua Levu thetransparencies of the old map sheet series werecut and assembled to fit into the new map sheetseries coverage.

The digital data of the scanned transparen-cies is available on CD-ROM at MSD. For everymap sheet the CD-ROM contains a dedicated di-rectory. Thirty-eight map sheets or directories areaccessible (see Table 1).

For map sheets covering Viti Levu the usersfind in every directory a file containing the

_________________

1this was possible in Fiji, it might be difficult in other countries!

L-26 L-27 L-28 L-29

M-26 M-27 M-28 M-29

N-26 N-27 N-28 N-29 N-32

O-23 O-24 O-26 O-27 O-28 O-29 O-32

P-22 P-23 P-24 P-27

Q-22 Q-23 Q-24 Q-25-26 Q-28 Q-29

R-22 R-23 R-24

S-21 S-22 S-23 S-24 S-25

Table 1. Directories available on CD-ROM containing digital map information.

n11

scanned road network indicated by “_S” and the river system indicated by “_B”and a “FCIF_” filecontaining the Forest Cover and Interpreted Forest Function map2. For most map sheets of VanuaLevu the road network and river system is combined in one file.

Directory M_28

FCIF_M28 TIF 1,922,156 08-02-96 8:05p

M28_B TIF 3,864,132 06-10-95 10:14p

M28_S TIF 4,420,762 02-02-96 10:04p

Figure 1. Listing of one directory in this case for map sheet M28.

From Compressed *.TIF to ERDAS *.GIS FilesMSD-Forestry is equipped with software such as ERDAS, ARC-INFO, MicroStation, ImageAlchemythat is able to read TIF format. However, there are different TIF formats, which was unknown toMSD. Further, the data format description was not provided. The format was analysed and theconversion was carried out at SOPAC by using ENDOBE PhotoShop software.

Then MSD-Forestry was able to convert normal TIF files to ERDAS GIS files. The file size createdthe next problem. The corresponding ERDAS GIS file for a 40 MB TIF file has about 350 MB! Thesolution was to work on a Magneto Optical Disk (MOD) which has 1.2 GB or 600 MB on each side3 .ImageAlchemy and ARJ software were loaded on the MOD. Then a MS-DOS batch file was createdcontaining for every compressed TIF file the following steps:

• decompression with ARJ software

• conversion from uncompressed TIF to ERDAS GIS by ImageAlchemy

• compression of ERDAS GIS by ARJ software

• deleting the TIF and the GIS files

These steps took about three to four hours for every file!

Compressing the Resolution and Geometric CorrectionThe created GIS files have 15,700 lines and 22,600 columns. The smallest linear elements which arethe grid lines have a width of five pixels. In order to create manageable files a cluster of four by fourpixels had to be converted to one pixel.

2 The digital FCIF maps have been created at MSD as part of the forest inventory project.3 MSD-Forestry is using SONY MOD

Figure 2. Aggregation of information, a matrix of 4 x 4 pixels were combined to one pixel by ERDAS moduleAGGIE. This decreases the information content, see Figure 7.

It was noted that the scanned transparencies have a slight distortion that prevents them from beingcopied directly to the Forest Cover / Forest Function information layer. More importantly theselayers have to have the same pixel range as the forest cover / forest function layer. This has beenarchived by a geometric correction program forcing the four corners of every layer to fit.

To find the four corners of a map layer an interactive process is required where the operatoridentifies the X- and the Y-value by features visible in the form of roads and rivers in comparison

with the physical maps. For each individual mapsheet the four corners of the actual map contenthave to be transferred to the corner pixel of amatrix containing 4800 columns and 3600 lines(see Figure 3).

new file old file

co-ordinates co-ordinates

1) 1 1 871 129

2) 4800 1 5604 101

3) 4800 3600 5627 3645

4) 1 3600 895 3673

Figure 3. GCP file for geometric correction of thealready resembled data set. The geometric correctionensures that the infrastructure fits exactly over theforest type information, however, performed like thisit creates a further reduction of resolution.

Decreasing the Pixel Size ofthe Forest Cover InformationAfter the described geometric correction of theinfrastructure files, it is necessary to adjust thepixel size of the forest cover information to bepresented in the map. One pixel has to be con-verted into a matrix of 3 x 3 pixel, so that theforest information of one map sheet covered by1600 columns and 1200 lines is transformed to afile of 4800 columns and 3600 lines. To carry outthis transformation a geometric correction isemployed again, but this time the process in-creases the file size for the forest information.

new file old fileco-ordinates co-ordinates

1) 1 1 1 12) 4800 1 1600 13) 4800 3600 1600 12004) 1 3600 1 1200

Figure 4. GCP file for file transformation, the processdecreases the pixel size and increases the file size ofthe forest cover information.

After the process the files containing the infra-structure and the file containing the forest coverinformation have the same orientation and pixel

Figure 5. The file transformation process creates outof one pixel nine pixels storing the same thematicinformation.

12n12

size. For the first maps the pixel size was onlyreduced by factor four (one pixel was convertedto four pixel). This required a resembling of theinfrastructure producing numbers and lettersdifficult to read (see Figure 5)

Comparison of Resolution

Figure 6. Upper left corner of map sheet O28 in theoriginal data file of the scanned transparency.

Figure 7. Upper left corner of map sheet O28 in theresembled data set, 4 x 4 pixels are condensed to onepixel.

Figure 8. Upper left corner of map sheet O28 aftergeometric correction employing a further reductionof resolution by factor 3. This is no longer carriedout, the pixel size of the forest cover information isdecreased by factor three instead.

ConclusionThe time requirements are not listed in the arti-cle. For one map sheet showing the forest infor-mation combined with the river system and roadnetwork 22 hours computer time are necessary.Mainly the steps of aggregation and geometriccorrection are very time consuming. However,it is possible to carry out such a job in Fiji. Fur-ther, once the infrastructure files are converted,aggregated and geometrically corrected it is easyto overlay them on any existing map informa-tion. MSD-Forestry provides an excellent tool formap production. The infrastructure layer readyfor overlay are available for seven map sheetsand further processing is under way.c

Figure 9. Subset of a digital Forest Type Forest Function map showing Colo-i-Suva Plantation after re-mapping.The infrastructure and numbers are clearly visible although the resolution is reduced by factor 5.

IntroductionUnder EU funding all hardwood plantations in Fiji had to be re-mapped in order

to link them to the forest monitoring system. The target was to establish a spatialdata base which can be linked to ERDAS raster GIS (1:50,000 scale), to link it tothe dBASE plantation data bank storing timber-related information and to plot trans-parencies to produce dyeline copies. For this activity MSD employed ARC-INFObecause:

� the scale of 1:10,000 required a vector data based spatial data base

� ARC-INFO provides an ERDAS in-built vector to raster conversion

� ARC-INFO provides real raster to vector conversion

� ARC-INFO provides the link to dBASE which was difficult with MicroStation, avector data based software also available at MSD

n Velemani Latulevu, MSD-Forestry

Production of PlantationMaps with ARC-INFO

n13

Knowing about the high time require-ments of ARC-INFO because of its userunfriendliness all jobs are reduced to theminimum amount which has to be han-dled by ARC-INFO.

Results showed procedures used byMSD to carry out the mapping wasstandard operation for such a mappingpurpose.

Converting Existing MapsThe Lands Department provided

plots on transparency for plantations al-ready mapped by them. However, thesemaps have no link to the forest monitor-ing system and no link to the plantationdata bank either. The spatial data cap-ture was based on conventional stereo-mapping which does not create a digitaldata base. These maps have been digi-tised with ARC-INFO1 and are availablein a digital data set. The conversion fromCassini to Fiji Map Grid (FMG) was car-ried out with Mike Poidevin�s program,by using the surrounding map edges asGround Control Points (GCPs). The timeconsuming part of the map productionis the fine editing, such as adding com-partment and stand number, species,road names etc. for all plantations wherea mapping of unstocked areas was car-ried out this information was stored in adifferent layer and combined with theplantation boundaries by the ARC-INFOmodule UNION.

Converting Existing Maps ofUnknown Projection

Two maps which were digitised weredelivered without any coordinates. It wasimpossible to define the coordinate sys-tem and projection used for these maps.To transform these two maps to FMG,GCPs were collected and distributedproportionally over the plantation. TheseGCPs were then visited in the field withthe GPS (Global Positioning System)and the satellite delivered position re-corded. Employing these coordinates adata bank was established keeping forX, Y of the digitising table also the FMGof these GCPs. Then the ARC-INFOmodule TRANSFORM transferred thedigitised spatial data into a coveragerelated to FMG.

Producing New MapsNo maps were available for planta-

tions on Vanua Levu. At the beginningof the project it was planned to carry outa complete GPS survey recording the

boundaries, main creeks and mainroads. This was impossible due to timeconstrains and poor satellite receptionin the valleys. Stereo-mapping of1:50,000 aerial photographs was ap-plied. The stereo-mapping was carriedout in MSD using a ZEISS Visopret. TheFMG position of all creeks, rivers, roadsand of all ridges was recorded. The GPSsurvey later was only2 used to identifythe position of features not visible in thephotographs. The data set captured bystereo-mapping and the data recordedby GPS were then pre-processed inMicroStation before converting to DXFand shifting into ARC-INFO.

Non-stocked areas were mappedwith a Baush&LompZoomtransferscope from 1:25,000 col-oured photographs3 , by superimposingthem on to the newly plotted 1:10,000maps and drawing their boundaries intothe maps for digitising.

Results and FutureInvestigations

72 maps were plotted on transpar-ency paper. Of every map, three dyelinecopies were produced and handed tothe field staff, to help with managing theplantation. It will help to avoid furtherover-planting4 and will provide exactarea figures as payment base for allsilvicultural activities.

Any additional change of area cor-rections can now be considered in a fastand cost effictive way. It is important tonote, that every further change will au-tomatically change the area estimationat the 1:50,000 level, which means thatevery future conversion of natural for-est to plantation will also update the database of natural forest.

The data transfer from the raster databased GIS to ARC-INFO storing thespatial plantation data base was notperformed, because in was not neces-sary at this stage. However, it was car-ried out on a trial basis in order to showthe operationality. If there are no aerialphotographs satellite images could betaken as a provisional data source. ARC-INFO allows the data link.

As a next step the plantation databank stored in dBASE IV could be seton top of the spatial data base stored inARC-INFO. Therefore ARC-View can beemployed.c

____________

1 this would have been also possible withMicroStation, but the system was occupied dur-ing this time and ARC-INFO was available.

2 which was still sufficient to keep one surveycrew busy!

3 it was technically impossible to employ theseaerial photographs for the stereo mapping.

4 planting areas outside the plantation due to in-accurate maps

Improving the Understanding of WaterCatchment Forest Cover with

Thematic Mapper Data

n Osea Tuinivanua, MSD-Forestry

AbstractThe exercise is a continuation of the application of TM data through additional image enhancementof spectral properties to exhibit the potential differentiation within the Navua water catchment for-est cover. However the complexity and variability in the forest species composition remains. Sec-ondly the ever present problems of cloud and haze. Spectral information with original raw TMbands was increased by deriving new bands. Additional bands from principal component and tasseledcap transformations improved the understanding of forest cover within the catchment area.

IntroductionThe classification of actual forest types is essential for effective forest management planning. Classi-fications of forest types based on water catchments have been recommended (Wakolo, 1995., MSD;1995). An important and cost effective means of carrying this out is through satellite imagery as

14n14

being performed in other countries. Although,we are restricted with quantitative and qualita-tive data, we are optimistic for a positive changein the future. However the spectral differencesbeen observed can be used as basis for furtherforest classification.

Test AreaThe test area includes the same Navua Catch-ment. The water catchment is a mosaiking ofsouth-east quarter scene TM 075/072 and north-west qaurter scene TM074/073. Date of data ac-quisition ranges from 25/12/91 for 075/072 and6/3/92 for 074/073. Other features describingthe characteristics of the catchment area are con-tained in previous issues of this newsletter, seelist of references at the end of this article.

Training AreasThe selection of training areas were predeter-mined using a GIS layer of forest types deter-mined under Howard and Berry (1973). The areaselection considered the representation of vari-able forest types to be investigated and allocatedon the Fiji Map Coordinates. The map coordi-nates were then recorded into the roving GPS forthe field identification of plots. Once these plotshave been located in the field, another GPS read-ing must be recorded from each plot preferablythe plot centre for the GPS Differential Correc-tion. The GPS Differential Correction is now pos-sible and operational with an accuracy that isacceptable for forest applications in Fiji (Poidevin,1994, Poidevin, 1995).

Each training area comprises of 4 blocks (25m * 25 m) with north - south orientation to re-semble pixel layout. The only quantitative meas-urement taken was that of tree dbh over 10 cm.The remaining descriptive assessment includetree species, canopy layer, canopy structure andleaf structure. However, other important featureslike slope, aspect, extent of sunlight penetrationto ground level and so on were noted.

TM Imagery dataThe Landsat 4 Thematic Mapper data were re-corded around December 25th, 1991 and March6th, 1992 for scenes 075/072 and 074/073 respec-tively. The satellite imagery was made availableto the National Invenatory Project from 1991 -1994 and eventually used for the exercise.

Data AnalysisData analysis for the exercise was performedusing ERDAS version 8.2 under UNIXworksattion while other original processing wasperformed on PC based ERDAS version 7.5.

The original raw Thematic Mapper bands 3to 5 and 7 were retained. Other derivative bandsinclude the principle component transformation(PC1) which accounts for the best visible vari-ability and three Tasseled cap transformationbands explaining the brightness (TC1), greeness(TC2) and wettness (TC3). The principal compo-nent transformation was performed without TM

band 1 and band 6. For this report the final image consists of 9 bands. At this stage no ratio bandsand other transformation bands will be included.

The differentially corrected GPS readings (FMG coordinates) of the training areas were used tolocate the map coordinates with the AOI Tool in the image. These were then expanded to the desiredconstraining factors under the SEED module and recorded in the signature file if required for furtherprocessing. The statistics of the selected training area spectral patterns can then be calculated anddemonstrated, for example, in separability via ellipse plots or contingency matrix and divergence.

The traditional supervised technique was used in the classification. The classification algorithmuses a maximum likelihood classifier that considered the statistical spectral variability of trainingareas and similarly, using the covariance matrices of the training areas for the differentiation. Al-though, maximum likelihood classifier has inherent weaknesses, its has greater advantages as wellwhich should be exploited in the exercise.

ResultsThe spectral pattern differentiation by individual selected areas can be graphically demonstrated viasignature variability (Figure 1).

The results maintain that TM4 and TM5 (more significantly than TM3 and TM7) continue to providevaluable forest cover differentiaion for the training areas in the test site. However, the derived bands

%DQG��

'

1

�

9

D

O

X

H

�

��

��

��

��

���

���

70� 70� 70� 70� 3&� 7&� 7&� 7&�

ZS�IW� ZS�IW� ZS��IW� ZS��IW� ZS��IW�

ZS��IW� ZS��IW� ZS��IW� ZS��IW� ZS��IW�

ZS��IW� EDPEX PDKRJDQ\

Figure 1. Signature Mean Plot of the Training Areas.

principle component band (PC1) and the Tasseled Cap (TC1, TC2 and TC3) showed great potentialsin differentiating these training areas or supposedly forest typing. The supervised classification re-sult was very encouraging. Certain misclassifications were detected and could be resolved withfurther processing.

DiscussionThe intention of the exercise was to continue the classification of forest types based on catchmentareas. Secondly, using these various techniques and methodologies to establish more friendly prac-tical link to operational level. It is extremely impossible to locate pure stands of native forests inlogged areas and presumably to a large extent in our unlogged forest as well.

It can be derived from this preliminary result that observed spectral differentiations among thisoriginally designated one forest type (ft1) has great potential for more classes. Bamboo growths andthe old pure mahogany plantations can be clearly distinguished from the others. The persistentseparation of unlogged forest proved very interesting. It can be interpreted from the signature graphthat the behaviour of most native forest cover was distinct and consistent throughout. However,water moisture dropped significantly for mahogany, yet its wettness proved highest among all oth-ers. A valid explanation could be the resulting impact of its annual refoliation. The bamboo growths

n15

showed the least wettness and this was not surprising due to its leave characteristics.

The driving force here is to derive forest types based on its detectable behaviour from satelliteimagery. This is only the beginning as more possibilities yet to be tested. Individual characteristics offorest types are becoming obvious and should continue to be investigated and validated with ad-vancing techniques and methodologies.

AcknowledgementsThanks to the MSD staff members for the additional GPS readings and support.

ReferencesBerry, M.J., Howard, W.J., 1973. Fiji Forest Inventory - The Environment and Forest Types. Foreign

and Commonwealth Office ODA, Land Resource Study No. 12. Land Resources Division, Surrey,Eng. 1973.

Poidevin, M., 1995, A Basic Understanding of Global Positioning System (GPS). Fiji User Group. GISand Remote Sensing News No. 12. 1995 pp 11-12.

Poidevin, M., 1995, GPS Differential Corrections Now Available in Fiji. Fiji User Group. GIS andRemote Sensing News No. 9. 1994 pp 2.

MSD, 1995. GIS Documentation Data at Scale 1:50,000 Management Services Division (MSD), FijiForestry Department ver. 30. August, 1995. Internal Report 53-55.

Wakolo, J. 1995. Water Catchment as a Management Unit ? . Fiji User Group, GIS and Remote SensingNews No. 12, November, 1995. pp 7-8.c

n by Wolf Forstreuter, MSD-Forestry

Mapping and Monitoring ofFiji's Hardwood Plantations

IntroductionThe European Union financed and

assisted a hardwood plantation mappingproject in Fiji which is now completed.The purpose was the mapping at1:10,000 scale and the integration of themapping into the country wide forestmonitoring system based at the Man-agement Services Division (MSD) ofFiji�s Forestry Department.

Before the project started only sketchmaps were available for most plantationareas and these were of unknown pro-jection with inaccurate boundaries. Inorder to produce accurate maps withinthe given time frame, the project usedmodern equipment and techniques suchas Global Positioning System (GPS),stereo mapping, digital satellite imagesand Geographical Information System(GIS). The experience showed that thecombination of these techniques is theright choice for forest mapping. Further-more, the equipment and technique pro-vided the basis for a continuous updateof the spatial plantation information,

which can be handled in the country. Itis no longer necessary to order this serv-ice from overseas companies.

The GPS ApplicationAt the planning stage of the project it

was believed that GPS can be used asthe only information source for accurateplantation mapping. The experienceshowed, that this technique has its limi-tations and has to be supplemented byother ways of mapping.

The field teams followed tracks,roads, creeks and boundaries with thehand held receiver (rover) and recordedthe positioning data received from thesatellites. Opening the sunroof of the carit was possible to stand upright with theGPS rover and survey forest roadswithin a short time. After completing thesurvey of one or two features the roverdata was uploaded to a laptop. At theend of a day (in remote areas after twoor three days) the data files were trans-ferred via modem and telephone line toMSD.

MSD has its own base station fas-tened on the roof of the building. Thearriving positioning data gathered in thefield can be differentially corrected us-ing the data files recorded by the basestation at the same time. This correc-tion increases the accuracy to betterthan 5 metres and this was confirmedon several examples. The disadvantageof the GPS survey was the poor recep-tion of positioning data in valleys underforest cover.

Stereo Mapping ApplicationThrough a former AusAID project,

MSD is equipped with an analyticalstereo mapping instrument and recentblack and white aerial photographs cov-ering the plantation areas were availableas well. Due to time constrains and thedescribed difficulties of the GPS survey,this survey was partly replaced by stereomapping. The aerial photographs havea scale of 1:50,000. In combination withthe instrument (ZEISS Visopret) and thenewly determined Ground ControlPoints (GCPs)1 these photographs pro-vided the necessary accuracy (betterthan 5 metre). Within a short time allcreeks, tracks roads and ridges werestored in files as digital positioning data.

These files was then shifted to theGIS. These features often making theboundaries of sub-areas within the plan-tations such as compartments or stands.A forest operator deleted all features notmarking such sub-areas. Then he im-ported the GPS survey data to comple-ment boundaries and highlighted areasfor which additional information was re-quired. For these areas the GPS fieldteams visited the plantations again andcaptured additional data. Later the roadnetwork and river system was overlayedwith different label allowing a later printin different colour.

The black and white aerial photog-raphy at 1:50,000 scale did not allow theseparation between plantation and natu-ral forest cover. Only some mature plan-tation stands can be recognised by thetypical texture. This was insufficient forthe required monitoring.

Applying Satellite ImagesTo monitor plantation areas it is nec-

essary to provide information about:

� which areas are planted and whereplantation establishment took place out-side the leased area

� which areas are still stocked with un-

16n16

treated natural forest

� which areas are logged and preparedfor planting2

Coloured aerial photographs at1:25,000 scale or larger provide the bestinformation source. However, such in-formation is required regularly and aerialphotographs are very expensive to ac-quire. Digital satellite images3 proved tobe the best alternative. Three steps arenecessary to produce plantation imagemaps:

1. geometric correction, to fit the imageinto the required projection

2. object specific contrast enhance-ment to visualise areas dominatedby planted species from areas wherenatural forest dominates the canopyand to separate untouched naturalforest from logged and chemicallytreated parts

3. overlay of the plantation boundariesderived from 1:10,000 mapping

The produced plantation imagemaps at 1:50,000 scale4 showed areaswhich have been planted outside theplantation boundary and unplanted ar-eas within the plantation. However, thetime factor has to be considered. It isimpossible to identify these parts directlyafter planting, because they are plantedunder the umbrella of the natural forest.Before they can be identified on the sat-ellite image the planted trees must domi-nate the canopy. Nevertheless, the pro-duced image maps proved to be an op-erational tool for monitoring the planta-tion development.

Map ProductionFor change detection of forest cover

and any area assessment, digital mapsare sufficient. For field work and for visu-alisation the status and any planning aforestry department must be able to pro-duce their own physical maps.

Maps at the scale of 1:10,000 wereproduced by a pen-plotter on transpar-ency allowing cost effective dyeline cop-ies.

At 1:50,000 scale coloured mapswere produced. An recent natural for-est inventory project produced a com-plete set of maps in Germany. Thesemaps show the forest cover, however,they reflect the situation before the plan-tation mapping was carried out. Fromthis map production received the infra-structure in the form of the road network

and river system as digital data set.These data was created by scanningcorresponding transparencies providedby the Fiji�s Lands Department. The for-est cover information was updated byimporting the newly mapped plantationareas into the digital forest cover infor-mation. Interpretation of satellite imageswas carried out for areas which werewrongly mapped as plantation before.Before overlaying the digital infrastruc-ture the pixel size of the forest cover in-formation was reduced to keep the finestructure of linear elements such ascreeks, labels etc. An A0 ink-jet plotterwas used for final map output.

Such a map does not reach the qual-ity of an offset printing product, however,it provides already an expectable stand-ard and shows the latest forest situa-tion.

ConclusionThe plantation mapping project pro-

vided more than accurate plantationmaps, it updated the forest monitoring

system to an extent that the ForestryDepartment is able to produce their ownmaps at 1:50,000 and 1:10,000 scale. Itis important to install mapping facilitiesin the country rather than ordering theproduct from overseas companies. Fur-ther, it was important to install a moni-toring and mapping system within theForestry Department, rather than em-ploying the facilities and knowledge ofother Departments. A good forest man-agement requires permanently updatedinformation about the forest cover. To-day, a forest map cannot be used formore than five years.

__________1 the GPS facilities were used to establish GCPsnecessary for stereo modelling

2 normally chemical thinning is carried out

3 the project used multispectral SPOT data

4 1:50,000 is the largest scale which can be pro-duced with satellite data available at MSD

n by Asesela Wata, MSD-Forestry

Plantation Image MapsIntroductionThe increasing area of Fiji’s hardwood planta-tions has to be managed. Remote sensing data isan excellent tool to monitor the situation in thefield. It is necessary to know the location oflogged, over-planted, not planted or vegetation-free areas. Coloured aerial photographs at1:25,000 scale are the right type of data. How-ever, it is difficult to get frequent coverage. Sat-ellite images cannot be employed to map at1:10,000 scale, but they provide an overview andthey are frequently available.

Producing Plantation ImageMapsBefore satellite data can be included into Fiji’smapping system a geometric correction has totransform this data to fit the projection and mapsheet coverage. Corrected data is available atMSD, and the geometric correction was ex-plained.

To avoid handling unnecessarily large filesout of image files covering one map sheet, sub-sets have to be separated out just covering the

area of the plantation. ERDAS software allowsthe recording of the geographic reference of thesubset. Every picture element has the Fiji MapGrid coordinates of its centre.

Digital images allow an object-specific con-trast enhancement. Parts of the image showinggrassland, agricultural area etc. are not of inter-est, but the difference between natural forest andparts of the canopy dominated by mahogany isimportant. Digital image processing can improvethe contrast between these different forest cano-pies while equalising the features outside forest.The reflection of areas with known forest canopyis analysed by the operator in relation to theirrange of reflection in each channel of the satel-lite image. The same is necessary for trainingareas of natural forest to identify the lowest andhighest pixel value in each channel. Then theimage is displayed setting all pixels lower thanthe lowest reflection of forest such as water bod-ies to zero and all parts of the image showingreflection higher than bright parts of the forestcanopy to 255. The range between 0 and 255 isthen available for stretching the contrast of for-est cover. Agricultural areas cannot be differen-tiated from cloud cover, and water is displayed

n17

unique in black, but the colour showing the for-est cover differs in a wide range allowing theidentification of mahogany, natural forest andother canopy.

The plantation outline as well as the com-partment and stand boundaries are available asARC-INFO coverage at MSD-Forestry. ERDASsoftware allows the import of polygons as arcsand converts the corresponding pixel value ofan existing image to a value related to the ARCshowing the boundary. The process has to becarried out separately for every band of the im-age. Therefore the image has to be separated intoits three bands and later the three bands have tobe joined again to one image file. This processcan be handled by a batch file.

CorelDraw software is used to superimposethe plantation name and the origin of the satel-lite data. Before doing this the image has to beconverted from ERDAS.LAN file to TIFF file for-mat, which is handled by ERDAS.

ImageAlchemy software is handling theprinting processes of image data. This softwareconverts the TIFF file format to an HP-RTL printfile. Because of the large space required this printfile is not stored on disk. The software allowssending it directly to the printer. Specifying thesize of the printed image enables the scale to bedetermined accurately

ResultsThe image maps allow a clear separation be-

tween:

• natural forest

• natural forest heavily logged or poisoned

• bare soil / grass

• mahogany dominating the canopy

Further it is possible to see:

• areas dominated by Nokonoko velau(Casaurina equisetifolia). This can be noted in com-partment 4 of Wainunu plantation which reflectsmore intensive red than the surrounding forestcanopy.

• bamboo if dominating the area after loggingthe natural forest. This can be noted in compart-ment 41 of Korotari plantation.

Summary and FutureActivitiesIt is impossible to separate areas planted withmahogany from un-planted areas, because theplanting is carried out under the remainingcanopy of the natural forest. However, satelliteimages can be used to identify compartmentswhere the planted trees dominate the canopywhich occurred after 12 to 14 years. At least af-ter this time over-planted areas are visible. It isalso possible to identify untouched remnantnatural forest if the surrounding area is affectedby chemical thinning. Satellite data proved to auseful tool to help in plantation monitoring.

Further investigations have to be carried outto determine after which time mahogany domi-nates the canopy, depending on silviculturaltreatment and soil. A more detailed investiga-tion is required for the separation of differentspecies such as Cordia subcordata or Anthocephaluschinensis.c

Figure 1. Display of a plantation image map. The original file is reduced and the coloured image is convertedto an one band (infrared band) black and white image. It is still possible to see the difference between mahoganyand natural forest cover.

n Prakash Narayan, Cité Universitaire WEISS C120, 7 Quai du Bruckhof, 67089 Strasbourg, France

Classification in a Simple FormAbstract

The purpose is to overview the generalities(a simplified version) of classification. Nonew ideas are introduced. No prerequiste is assumed but some prior Remote Sensing andImage Processing knowledge would be an advantage.

IntroductionSpecific examples of classification for remote sensing application include determining:

1) Tree-species composition in a forest2) hot spot of incipient forest fires3) natural vegetation cover-types4) crop types5) state of health or stressed vegetation6) percent of sedimentation in a river or lake7) percent of pollutant in a river or lake8) geological formation and rock types9) lineament patterns10) degree of mineralisation11) urban land-use patterns

To carry out a classification, we must first define the classes of entities of interest, thatis the kinds of objects between which we must discriminate. We must provide a methodol-

18n18

ogy permitting the recognition of an object in the class of objects of interest. Using thismethodology we also must construct a decision rule which will decide what kind of objecta particular object is, on the basis of the measurements made from the observed small-area ground patches.

Most classification of remotely sensed image data is done by processing each pixel�sinformation separately or independently. This means that a category assignment is madeto each pixel purely on the basis of its own information. When it is the decision rule whichmakes category assignment that uses the information from a pixel and some of its neigh-bouring pixels then the classification is called spatial classification or spatial spectral clas-sification.

The training data is used for each class to determine the class sample-mean. Themeasurement space is then partitioned so that each class is associated to the measure-ment class closest to its sample mean.

Feature selectionMultispectral (data obtained using several bands) remotely sensed imagery can pro-

duce a very high dimensional data vector for each pixel. The data has inherent redundan-cies and processing all of the data or storing all of them may not be cost effective. Featureselection procedures are used to select those dimensions most suitable for processing.

Some of the two kinds of feature selections depending on whether the classes andtheir statistic are known or not. If they are not known the best feature selection procedureis called principal components. If they are known the easiest-to-use feature selection isbased on Bhattacharyya distance.

Pricipal components is a standard statistical technique for selecting that subspace of agiven dimension in which the most data variance lies. If Xi ......Xn are the sample datavectors, µ the sample-mean vector and S the sample covariance-matrix, the best K di-mensions in which to project the data would be the K eigenvectors of S having the lowesteigenvalue.

Bhattachararyya distance is a measure of the separability between two classes. Fortwo Gaussian classes having means and covariance µ1, S1 and µ2, S2 respectively, theBhattacharayya distance is given by:

Classification processingRemote sensing classification is the process of sorting pixels into a fixed predeter-

mined number of classes, or categories of data based on their data file value. If a pixelsatisfies a certain set of criteria, it is assigned to the class that corresponds to that criteria.This process is also referred to as image segmentation.

Supervised classification: The indentification of pixels is done based on previous knowl-edge. Identify an area of the image that can be identified using ground truth information.This process is called taking a training sample. Statistical computation is carried out forthis area (mean, covariance, standard deviation). A set of training samples is called aspectral signature. A pixel by pixel classification is then perfomed on the multispectralimage data.

Unsupervised classification: The pixels that look alike or have the same characteristicsare grouped together and is given a label. The user then interprets the classes chosen bythe algorithm used. Requires very minimal initial input from the user, that is to identify thenumber of classes desired.

For instance to visualize the classification, each class is given a symbolic colour torepresent its characteristic.

SignaturesThe result of a training sample is a set of signatures that defines a training sample or

cluster. Each signature corresponds to a class, and is used with a decision rule to assignthe pixel in the image file. The two types of signatures are:

A parametric signature is based on sta-tistical parameters (eg. mean andcovariance matrix) of pixels that are in thetraining sample or cluster.

A non-parametric signature is not basedon statistics but on discrete objects (eg. poly-gons or rectangles) in a feature space im-age. These feature space objects are usedto define the boundaries for the class

Distance definitionA signature is a point in a spectral space

(up to n dimensions). To produce sub-spaceof this space, it should be possible to fix lim-its, which comes down to measuring dis-tance between signatures. There are sev-eral methods to measure the distance in agiven space, which are usually grouped inthree major categories:

1) circular distances (euclidean, chebychevetc )

2) elliptical distances (quadratic, by corre-lation etc )

3) conical distances (chi squared )

note: A distance between two points in the spec-tral plot bears no relationship to ground distancebut instead represents a spectral distance be-tween these two points.

Minimum distance : Is the spectral dis-tance between a given pixel and the meanof each signature. This pixel is assigned tothe class for which the mean is the closest.

The equation for Euclidean distance :-

n = number of bands (dimensions)i = a particular bandc = a particular classXxyi = data file value of pixel x , y in band iµci = mean of data file value in band i for the sample

for class cSDxyc = spectral distance from pixel x , y to the mean

of class c

Mahalanobis distance: Is similar tominimum distance, except that thecovariance matrix is used in the equation:-

D = Mahalanonbis distancec = a particular classX = the measurement vector of the

candidate pixelMc = the mean vector of the signatureCOVc = the covariance matrix of the pixels in

the signature of class c

COVc −� = inverse of COVc

T = transposition function

n19

The pixel is assigned to the class, c, for which D is the lowest. Mahalanobis distancetakes the variability of classes into account. However it tends to overclassify signatureswith relatively large values in the covariance matrix.

Maximum likelihood / Bayesian: The maximum likelihood is based on the probabilitythat a given pixel belongs to a particular class. Assume that these probablities are equalfor all classes and that the input bands have normal distributions.

The equation for the maximum likelihood/Bayesian is:-

D = weighted distancec = a particular class cX = the measurement vector of the candidate pixelMc = the mean vector of the sample of class cAc = percent probability that any candidate pixel is a member of class cCOVc = the covariance matrix of the pixel in the samlple of class c|COVc| = determinant of COVc

COV −� = inverse of COVc

ln = natural logarithm functionT = transposition function

The pixel is assigned to the class, c, for which D is the lowest.

Types of decisionsParallelepiped: A signature is represented only by a minimum and a maximum data

value in each band of the image.

Minimum distance: The decision is based on the spectral distance between the meanpixel value and the pixel to be classed.

Thresholding: The user specifies a maximum allowable value for distance values, whichis the threshold value. If the distance of a pixel is greater than this threshold value, the pixelis assigned to a background or unclassified class.

ConclusionIn many Remote Sensing applications, the final stage of image processing is to per-

form a method of classification of some kind from the list of methods proposed on animage file. Classification enables to highlight some certain characteristics in an image andvaries according to ones needs. As an example a urban developper would highlight onlybuilding and road structures while a forestery person would highlight only vegetation re-gions etc etc. Non parameteric and Neuronal network methods are now being proposed inmany articles.c

n by Wolf Forstreuter, MSD-Forestry

E-mail for Forest MappingIntroductionThe Management Services Division (MSD) of theForestry Department has e-mail access viaSOPAC1 . At the same time problems exists re-garding transfer GPS survey data from VanuaLevu to Colo-i-Suva for further processing. Itseemed logical to investigate the transfer of datausing the e-mail access. Normally, GPS survey iscarried out for one or several weeks until the teamreturns together with the survey data on laptopcomputer disk and floppy disk backup. At MSD,a differential correction is performed and thenthe data is displayed for a check of completeness.If data is lost, inaccuracy is visible due to bad

satellite reception, the survey team plus car hasto go back by ship to Vanua Levu to repeat thework. This is time consuming and costly. TheEuropean Union saw the problem and providedas part of the funded inventory project a laptopcomputer with a modem for data transfer.

Sending Data from OtherIslandsIn the field, the GPS rover2 data is down loadedonce or several times a day. Each file containingsurvey data of one boundary, one road or one orseveral ground control points has a file nameindicating the rover unit and the month, day andhour of receiving. If there are more than one filerecorded per hour they can be distinguished bythe additional letter (see Figure 1 below).

For every file the forester writes an ASCIIfile with the same file name but an extension TXTcontaining the:

1. start of the rover file (year, month, day, hour)local time.

2. feature surveyed e.g. boundary betweencompartment 11 and 7.

3. noticed survey problems e.g. bad receivingat 11:35.

4. end of the rover file (year, month, day, hour).

At the end of a day the team leader condensesall files recorded during the day into one ARJ3

file. The filename indicates year, month, and day.This sequence allows the automatic order bycommon software.

Whenever the team visits a forest stationwith a functional telephone line the team leaderconnects the modem of the computer with theGPS mailbox at SOPAC. The ARJ file containingall survey data is attached to an e-mail he sendsto the mailbox.

Receiving Data at MSD andAdvising PerformanceWhenever MSD connects to its mailbox atSOPAC it will receive mail with attached surveydata files, because these are sent automaticallyfrom the GPS mailbox to the normal MSDmailbox. MSD normally connects three times aday to its mailbox at SOPAC.

If MSD receives a message with an attachedsurvey data file it can be differentially corrected4

within a short time, because the base station datais downloaded every day. The corrected data thencan be overlayed and displayed in the GIS that

Figure 1. Filename of a GPS rover file.

R041723C.SSF

rover unit "R"

April (04)

17th of this month

11 at night in Greenwich

the third file within this hour(1 = A, 2 = B, 3 = C)

20n20

Organisations within Fiji who receive and pro-vide contributions to this newsletter:

ORGANISATION/FAXA & P Development Consultants, 384766Australian Embassy, 300900Bureau of Statistics, 303656Delegation of the Commission of the Euro-

pean Community, 300370Department of Energy, 386615Department of Environment, 303515Department of Town and Country Planning,

303515Digital Equipment Corporation, 300889Embassy of the Republic of Marshall IslandsEnvironmental Consultings Fiji Ltd, 370012Fiji Forest Industries Ltd, 813088Fiji Pine, 661784Fiji Posts and Telecommunications, 313362Forestry Department, Management Services

Division, 320311Forum Secretariat, 305573French Embassy, 300937Federated Stated of Micronesia EmbassyHarrison and Grierson, 301986Japanese International Cooperation

Agency, 302452Lands Department, FLIS Support Centre,

305029Lukemine Enterprises (TRIMBLE), 370858MacPacific, 303681Marine Department, Fiji Hydrographic Serv-

ice, 303251Mineral Resources Department, 370039Ministry of Agriculture, Fisheries and Forestry,

Drainage and Irrigation, 305546Ministry of Foreign Affairs, Civil Aviation & Me-

teorologyMinistry of Primary Industries, Drainage and

Irrigation, 305546Ministry of Primary Industries and Coopera-

tives, Land Use Section, 400262Native Land Trust Board, 303164Papua New Guinea Embassy, 300178Public Works Department, Hydraulics Section,

303023Public Works Department, Water and Sewer-

age Section, 315244Queensland Insurance, 300285South Pacific Applied Geoscience Commis-

sion (SOPAC), 370040South Pacific Commission, 370021South Seas Computing, 370875Tuvalu Embassy, 301023UNDP, Department of Humanitarian Affairs,

304942UNDP, Regional Water and Sanitation Project,

302487UNDP/FAO, South Pacific Forestry Develop-

ment Programme, 305212University of the South Pacific, GIS Unit, 301487University of the South Pacific, School of Pure

and Applied Sciences, 302890Wood & Jepson Consultants, 303361

Organisations outside Fiji (sorted alphabeti-cally by country) who receive and providecontributions to this newsletter:

ORGANISATION/COUNTRYACE Technology Australia Pty Ltd, AustraliaBunnings Tree Farms Pty Ltd, AustraliaCSIRO, Division of Fisheries, AustraliaSPOT Imaging Services Pty Ltd, AustraliaInternational Development Planning & Man-

agement Co, CanadaMinistry of Agriculture, Cook IslandsMinistry of Foreign Affairs, Cook IslandsMinistry of Marine Resources, Cook IslandsDepartment of Resources & Development,

Federated States of MicronesiaDepartment of Conservation & Resources Sur-

veillanceDivision of Forestry, Federated States of Micro-

nesiaGOPA Consultants, GermanyBureau of Planning, GuamDepartment of Agriculture, GuamMinistry of Environment and Natural Resource

Development, KiribatiMinistry of Foreign Affairs & International

Trade, KiribatiMinistry of Home Affairs and Rural Develop-

ment, Lands and Survey Division, KiribatiMarshall Islands Marine Resources Associa-

tion, Marshall IslandsMinistry of Foreign Affairs, Marshall IslandsCentre ORSTOM de Noumea, New Caledo-

niaService des Mines et de l'energie, New Cal-

edoniaSouth Pacific Commission (SPC), New Cal-

edoniaCritchlow Associates, New ZealandDepartment of Survey & Land Information,

Aeronautical Charting, Wellington, NewZealand

Landcare Research, Lincoln, New ZealandMonitoring & Evaluation Research Associates,

New ZealandRNZAF Base Auckland, New ZealandUniversity of Otago, Spatial Information Re-

search Centre, New ZealandDepartment of Mining and Petroleum, Papua

New GuineaDepartment of Mining and Petroleum, Cor-

porate Services Division, Papua NewGuinea

Department of Mining and Petroleum, Min-erals Division, Papua New Guinea

University of Papua New Guinea, Departmentof Geography, Papua New Guinea

Forum Fisheries Agency (FFA), Solomon IslandsMinistry of Foreign Affairs & Trade Relations,

Solomon IslandsMinistry of Natural Resources, Forestry Division,

Solomon IslandsMinistry of Natural Resources, Geological Sur-

vey Division, Solomon IslandsNew Zealand High Commission, Honiara,

Solomon IslandsDepartment of Agriculture, TongaMinistry of Fisheries, TongaMinistry of Foreign Affairs, TongaMinistry of Lands, Survey and Natural Re-

sources, TongaDepartment of Foreign Affairs and Economic

Planning, TuvaluDepartment of Lands & Survey, TuvaluMeteorological Office, TuvaluMinistry of Natural Resources, TuvaluOffice of the Prime Minister, TuvaluPublic Works Department, TuvaluEOSAT, USAUnited States Forestry Service, USADepartment of Geology, Mines and Water

Resources, VanuatuDepartment of Forestry, VanuatuLands and Survey Department, VanuatuMinistry of Foreign Affairs, External Trade & Im-

migration, VanuatuApia Observatory, Western SamoaDepartment of Agriculture, Forests and Fish-

eries, Western SamoaDepartment of Lands, Surveys and Environ-

ment, Western SamoaMinistry of Foreign Affairs, Western SamoaSouth Pacific Regional Environment Pro-

gramme (SPREP), Western SamoaWestern Samoa Water Authority, Western

Samoa.c

Requests for inclusion in the mailing listfor this newsletter as well as the submis-sion of articles for publication should be sentto:

GIS AND REMOTE SENSING NEWSSOPAC, PRIVATE MAIL BAG, GPOSUVA, FIJI

Attention: Les AllinsonTel: 381377 Fax: 370040E-mail: les@sopac.org.fj

Dr Wolf ForstreuterGOPA-ConsultantsHindenburgring 1861348 Bad HomburgGermany

E-mail: gopacon@ibm.netFax: +49-6172-930 350Tel: +49-6172-930 398

It would be appreciated if contributionscould be sent on floppy disk in Word forWindows (preferred), Wordperfect for Win-dows or Wordperfect for DOS format.c

identifies any inaccurate survey caused by lowreceiving visibility.

MSD then sends a message to the GPSmailbox at SOPAC telling the field team if theyhave to repeat the survey, which normally is anexception. The field team receives this messagewhen they connect to the GPS mailbox again toupload the next survey data. If the team has torepeat a survey it is still in Vanua Levu and mostlikely still in the vicinity of the plantation site.

ConclusionThe modem provided by European Union savestime and effort. It should be investigated if ageneral data link via modem to the northern di-vision can cut down on communication prob-lems. The described example also shows that a(regional) - technical organisation such asSOPAC becomes an integral part of the processin providing access to new technical facilities,which are not always experienced in practice.c

________________1 SOPAC = South Pacific Applied Geoscience Com-mission2 a GPS rover is a hand held GPS receiver storing theposition data3 ARJ is the extension indicating a file created by ARJcompressing software4 GPS survey requires a differential correction to in-crease the accuracy. Therefore a base station is neces-sary which is based at MSD in Colo-i-Suva

CONTACTS