Using Remote Sensing to Detect andMonitor Land-Cover and Land-Use Ghange

Kass Green, Dick Kempka, and Lisa Lackey

AbstractOur ability to change ourselves and our surroundings in-crcases with each technological advance. Changes today aremore ertensive and occur more rapidly than ever before. Thesignificance of these changes incrcases as the world's popu-lation grows, the available land base declines, and the resi-liency of our environment becomes increasingly taxed. As aresult, many organizations need to monitor change in landcover and land use. While ertensive rcsearch has been com-pleted on various change detection methods, Iittle work hasbeen done to implement the technologies in a commercialenvironment. The goal of Pacific Meridian Resources' NASAE a rth Ob s e rv atio n s C o m me rc i al Ap p I i c ati o n s P ro gr a m(EICAP) project is to take existing geographic informationsystem and satellite image processing technologies and de-velop production-ortented methods and sewices fo.r ossess-ing and monitoing land-use and land-cover change. Thegoal of the project is not to develop new technologies, butrather to tailor existing technologies for commercializationand introduction into the marketplace. This paper reviewsthe toc,tt project to date by presenting the findings andmethods of severcl completed pilot projects.

IntroductionAs technology advances, so does our ability to change oursurroundings, Changes made on the surface of the Earth to-day are more extensive and occur more rapidly than ever be-fore. The ramifications of these changes have become moresignificant as the world's population grows, the availableland base declines, and the resiliency of our environment be-comes increasingly taxed. Planners and resource managersneed a reliable mechanism to assess these consequences bydetecting, monitoring, and analyzing land-use changesquickly and efficiently. The demand for an efficient land-cover detection system currently exists among a variety of Io-cal, state, federal, and private organizations. While extensiveresearch has been completed on the effectiveness of variouschange detection technologies, little work has been done toimplement the technologies in a production environment.

To meet this demand, Pacific Meridian Resources' EarthObservations Commercial Applications Program (EocAP)project with the National Aeronautics and Space Administra-tion (NASA) takes existing geographic information system(cn) and image processing technologies and develops pro-

K. Green is with Pacific Meridian Resources, 5915 HollisStreet, Building B, Emeryville, CA 94608.

D. Kempka is with Pacific Meridian Resources, 9823 OldWinery Place, Suite 16, Sacramento, CA 95827,

L. Lackey is with Pacific Meridian Resources, 5200 SWMacAdam, Suite 570, Portland, OR 97201.

PE&RS

duction-oriented methods and services for assessing andmonitoring land-use and land-cover change. The primarythrust of this project, the Land Use and Cover Change Analy-sis System (LUCcAs), is the development of methods to assessland-cover change over time. Land-cover changes will be re-lated to land-use change which in turn will be associatedwith demographic and economic change. The goal of theproject is not to develop new technologies, but rather to tai-Ior existing technologies for commercialization and introduc-tion into the market place. The result will be a set of servicesand products which wili bring existing technologies from theresearch sector into the commercial and public sectors.

BackgroundTechnology DevelopmentThe overall goals in change detection and monitoring are (1)to compare spatial representations of two points in time bycontrolling all variance caused by differences in variables notof interest (i.e,, variation in orbital and platform altitudes)and (z) to measure change caused by differences in the varia-bles of interest. Currently, change detection, monitoring, andupdating rely primarily upon two types of techniques: map-to-map comparisons and image-to-image comparisons.

Map-to-map comparisons rely on the identification ofdifferences between two maps of different dates. This type ofcomparison requires that images (either digital or analog)from two different dates be classified by interpreters, withthe maps then compared to one another. In the past, thecomparison was performed on a light table and was the onlytype-of spatially referenced change detection possible. Withthe introduction of cIs, map-to-map comparisons can now beperformed digitally, either by the area in each polygon typeor on a pixel-by-pixel basis.

The usefulness of map-to-map comparisons is con-strained by the assumptions and techniques used to producemaps of the same area at different times. Maps of the samearels on different dates will vary for three reasons: (1) differ-ent classification systems, (2) different mapping techniques,and (3) actual differences in land cover. Thus, change detec-tion using maps or cls coverage alone is risky becausechanges Eetween maps can be caused by factors other thandifferences in land cover or land use. A more powerfulmethod of change detection incorporates the analyses of re-motely sensed imagery and/or field examinations into the GIS(Verbyla, 1990).

Photogrammetric Engineering & Remote Sensing,Vol . 60, No. 3, March 1994, pp. 331-337.

oogg - r 1 12 I I 4/6003-3 3 1 $ 03. 0 0/oO1994 American Society for Photogrammetry

and Remote Sensine

The introduction of digital imagery, including airbornescanners, digital orthophotography, and satellite imagery, hasexpanded the techniques available for change detection (Jen-sen, 1982; Martin, 1986; Jampoler and Haack, 1990). LandsatThematic Mapper (ru) and other remote sensing data havebeen used to detect and quantify forest decline associatedwith air pollution (Vogelmann and Rock, 19BB) and insectdamage (Leckie ef a1., 19BB). Digital ru and multispectralscanner data have been used for urban-suburban change de-tection (Jensen and Toll, 1982; Adeniyi, 1985), and spor datahave been used to monitor land conversion at the urban-ruralinterface (Martin and Howarth, 1989; Ehlers ef o1., 1990J.Visually enhanced images produced from merged sPoT pan-chromatic and multispectral image data have been found tobe very useful for visual land-use interpretation and urbangrowth delineation (Carper ef o1., 1990; Welch and Ehlers,1987). Landsat data have been used to differentiate croptypes and monitor crops over time (Wagner, 1992; Oden-weller and Johnson, 1984), and SPoT data have been used todifferentiate between various crop types manually (DeGloria,19B5). Image differencing (Jensen, 1985; Hame, 1.986; Yokotaand Matsumoto, 19BB) and principai components analysisand/or tasseled cap transformation (Richards, 1984; Fung andLedrew, 1987; Byrne ef o1., 1.990; Toll et a1., 1980; Kauth andThomas, L976; Crist and Cicone, 1984) are the two methodsof image-to-image comparison used most often.

Market DevelopmentDespite the success of these techniques, they have not beenwidely implemented for the following reasons: (1) most stud-ies completed to date have been designed to assess the effec-tiveness of particular technology rather than trying to adapttechnology to solve existing problems (Martin and Howarth,19BO), and (2) most studies have been single-medium ap-proaches with limited practical applications. As a result, theabsolute and comparative utility of existing techniques is notwidely understood among potential users in the commercialor public agency sectors.

In addition, change in land cover does not always implya change in use. For example, timber harvesting may effectdrastic changes in cover without a shift in land management.The implications of changes in land use and land cover are afunction of the economic, social, polit ical, and ecologicalcontext in which they occur. Planners and policy makers re-quire methods that (1) accurately relate changes in landcover to changes in land use and (2J facilitate assessment ofthe impacts of these changes within appropriate analysisframeworks.

A set of operating procedures that allorvs potential usersto choose and adapt existing technologies to their individualneeds is still lacking from previous investigations. Becausemost studies are designed to assess the capabil it ies of imageprocessing, most focus on the use of imagery alone. Geo-graphic information systems provide opportunities for creat-ing more dynamic and meaningful analysis opportunities forpotential users by integrating multiple layers of spatial infor-mation. To assess the impacts of different tlpes of changes inIand cover and use, an analysis framework needs to be de-vised that incorporates relevant social, economic, political,and ecological factors. The results of change detection analy-sis using satellite imagery may be interesting and revealing,but when placed in context with other spatial layers, such astopography, political boundaries, property ownership, orother data meaningful to users, this information gains power-ful analysis capabilities.

332

A system is needed that integrates the capabilities of im-age processing and cts with understanding of the economic,political, and biological issues contributing to a particularIand use problem at hand. To serve a variety of potentialusers, this system should be capable of rapidly identifyingand implementing methods and tools most suited for a par-ticular situation in terms of scale, budget, information needs,and frequency of repetition.

MethodsMulti-temporal analysis of satellite imagery is effective forchange detection only because there is high correlation be-tween spectral variation in the imagery and land-coverchange. However, like map-to-map comparisons, image-to-image comparisons also require that variation caused by fac-tors other than land cover be understood and, where possi-ble, controlled.

The type of change under investigation - data quality,availability, and scale, and the suitability of the algorithm forthe change detection required - all influence the choice of anoptimal technique to be used in analysis (Hodgson et a/.,19BB; Jensen, 1981). The LUCCAS project relies on pilot inves-tigations to test which methods appear to be appropriate(cost effective) to yield optimal change results for the givenecological situation and economic and polit ical conditions.

The initial plan for the development of tuccas was acomparative analysis of methods and data sources to be per-formed on several types of change in a case study area in Or-egon. However, since the award of the EOCAP contract,several of Pacific Meridian's clients have approached thecompany with change detection and monitoring problemsthroughout the West. Because the project is part of a com-mercialization program, the comparative analysis of one casestudy area was abandoned to respond to market needs.

Pilot project partners include the U.S. Department of Ag-riculture Forest Service, Washington State Department ofNatural Resources, Ducks Unlimited, CREST, the NationalOceanic and Atmospheric Administration, Potlatch Corpora-tion, and the Portland Metropolitan District. Each of theclients has become a corporate partner in the EOCAP projectby contributing funds, imagery, and/or personnel time.

Projects initiated with these partners have led to the fol-Iowing characterization of land-cbver and land-use changeanalysis techniques/products :

o Techniques that make the detection of change possible;o Techniques for the measurement of extent and magnitude of

cnange;o Techniques for updating an existing map or GIS layer to in-

corporate change; ando Techniques for estimating the impacts of change on environ-

mental, economic, and political conditions.

Pilot projects have been used to evaluate techniques, de-velop markets, and produce sample products. Most of the in-itial pilot projects concentrated on the development ofmethods for detection and measurement of change in forestIand vegetation, including

o Developing image maps for on-the-ground planning andchange detection,

o Assessing land-use and land-cover change in the PortlandMetropolitan Area,

e Analysis of pest damage in the forests of eastern Oregon,o Updatin8 managed stand layer,o Monitoring harvest levels for cumulative effects prioritiza-

tion.

PE&RS

o Assessing coastal wetlands change in the Iower ColumbiaRiver and Willapa Bay, and

. Monitoring changes in wetland habitat availability and its re-lationship to carrying capacity of waterfowl populations.

The following section summarizes the methods and results ofthe Portland metropolitan area proiect.

Podland Metropolitan DistrictAs populations have grown, the pattern of land cover anduse in and around the Portland metropolitan area haschanged. The pattern shifts compromise the continuity ofnatural areas, resulting in increasingly limited options for fu-ture urban expansion. Pacific Meridian worked with thePortland Metropolitan Service District to develop change de-tection methods so that the District can deveiop a long-termprogram for utilizing remotely sensed data in change detec-tion.

Plate 1. Multitemporal TM Landsat imagery.

The primary interest in the change detection study wasto identify urban natural areas that had sustained significantIoss of vegetation, primarily through timber harvesting anddevelopment-related activities. Significant vegetation iosswas determined to be a change in vegetative canopy of atleast 30 percent. A secondary interest was identification ofareas of vegetation gain. The classification system consists ofthree categories:

. vesetation Gain: ::Jifii:,$greater

increase in

. Vegetation Loss: ai...rrrt or greater decrease invegetative cover.

involves subtraction of one image date from another (jensen,1986). For this project, a comparison was made of two sepa-rate difference flles and implementation of simple GISmodels. One difference file was developed from subtractionof Band 7 for each date. A second file was created from sub-traction of a vegetation index for each date composed of aratio of Band g1o Band 4. The results were compared andcombined to take advantage of the individual strengths ofeach technique. Analysis was aided through incorporation ofan existing digital natural area coverage-and multi-date1:24,000-siale cn aerial photography. This integrated ap-proach capitalized on the strengths of individual techniques.^

Othei ancillary data used in the analysis included an ex-isting 1989 digital ARCAI.TFO coverage of-Urban Natural Areasand i:24,000-Jcale cIR aerial photography acquired in June1989 and June 1991. Processing and analysis were completedusing both Et{DAs and ARC^NFo software. Change -ana.l}siswas iccomplished in several steps integrating multiple imageprocessing and cIS analysis techniques.

Band 7 was chosen as the primary analysis band becausethis portion of the spectrum (middle infrared ) is less sensi-tive io atmospheric variation. A single band, or differencefile, was created through subtraction of the two Band 7s fromeach other. Prior to subtraction, the two Band 7s were evalu-ated for minor differences in reflectance unrelated to changesin cover type. Unchanged areas of very low reflectance (e.9.,water bodies) and very high reflectance (e.g', barren andpaved areas) were compared in Band 7 of both dates. Theisoa image'had an aveiage of z brightnes-q values Sreaterthan the ioot i-ug" in both the areas of low reflectance andhieh reflectance. Vilues for the 19BB image were thereforeshlfted downward by 2 prior to subtraction. A constant of100 was added to the difference file values to compensate forsoftware limitations prohibiting negative values. The singleband difference file then was created as

o No Change:Less than 30 percent increase invegetative cover and less than 30percent decrease in vegetative cover,

ProceduresChange detection was accomplished by comparing two Land-sat TM scenes of 7 July 1991 and 30 August 1988. The 7 July1991 image date was selected because it was cloud free andclose to the 31 July date ofavailable color infrared (cIR) pho-tography used as ancillary data. Although the 30 August19BB scene was acquired nearly two months later in the sea-son than the 1991 image, the 19BB scene was also cloud freeand could be used at no additional cost to the project. Plate1 displays a combined image of Band 7 from each of theLandiat rM dates. Band z in the 1991 image is displayed inthe blue and red color planes. Band 7 of the 19BB image isdisplayed in the green color plane. In general, areas with lessvegetation in 1991 than in 19BB appear in magenta, Areaswith an increase in vegetation over the same time period ap-pear in green,

Both images were geo-coded, terrain corrected, and re-sampled to 25-metre pixels. To verify co-registration accu-rac1,, bands of each image date were displayed together indifferent image planes. Stable linear features, such as roads,were inspected throughout the area to assess variation in reg-istration between the two images. Observed variation rangedfrom 0 to a maximum of 7.7 pixels in localized areas. Be-cause resampling would result in a certain amount of datadegradation, this minimal level of misregistration was deter-mined to be acceptable.

Several different types of algorithms are available for de-tecting change using multi-date imagery. Image differencing

PE&RS

L , , : W n r 1 1 - I W u ' t i - 2 1 + 1 0 0

A,, : change Pixel value,

where

V7 stii : reflectance value of rl'r Band 7 in 1991,

V7 eeii : reflectance value of TM Band 7 in 1988,

j : row number. and

I : Iine number.

Values of ro0 in the resulting difference file indicate areaswith no change in reflectance value between 19BB and 199t,Areas of increased reflectance are indicated by values greaterthan 100, and areas of decreased reflectance are indicated byvalues less than 100.

The difference file was visuallv analvzed to determinerelationships between difference ,rilues ind changes in vege-tative cover between dates. Each value was highlighted onthe imagery and compared to the 1989 and 1991 cIR photog-raphy. In this manner, the reflectance values were dividedinto five ranges representing relative Ioss and gain of vegeta-tive cover on the ground (see Table r).

Visual evaluation of the Band 7 difference file identifiedthe relative strength of relationships between the value dif-ferences and actual changes in vegetative cover. The differ-ence file alone appeared to be an exceptional indicator ofclearcuts and other areas of significant vegetative loss. Thestrength of the Band 7 difference file as an indicator of vege-tative change weakened, however, under four types of condi-tions: shadow differences, herbaceous and grassy areas,riparian areas, and water level differences.

The August 19BB image was nearly two months furtherfrom the summer solstice than the July 1991 image. Shadowscast across northwest slopes, therefore, appear longer in the19BB image than in the 1991 image. The increased brightnessof areas on the 1991 image that Iie in shadow on the 1988image provided a false indication of vegetation loss in theBand 7 difference file.

Variation in water level between the two image datesalso resulted in considerable confusion. Water gauging sta-tions on the Columbia, Clackamas, and Tualatin Rivers re-corded a difference of 40 Dercent in water levels between the19BB and 1991 image dates. The infrared portion of the spec-trum sensed by Band 7 is absorbed by water, thereby yield-ing very low (near zero) reflectance values, Therefore, Iowelevation areas that were exposed in 1988 but submerged inlggL erroneously appear in areas of vegetation gain in thedifference file,

Because the 1991 image was acquired in both a morevigorous part of the season and a wetter year than the 1988image, herbaceous and grassy meadows tended to appeardarker in the difference file, providing a false indication ofvegetation gain, Moist vegetation absorbs more infrared lightthan dry vegetation. Herbaceous and grassy vegetation, there-fore, appeared brighter in the dryer period of late August of19BB than in the wetter period of early July 1991. Similarly,deciduous trees and other vegetation in riparian areas re-

TneLE 1. CrnrucE CereeoRrEs BAsED oru Beno 7 DTFFERENcES VALUES

Band zDifference Value Change Category

= >- 30Yo vegetation gain= < 30% vegetation gain_ - ^ ^ L ^ - ^ ^- uu uudu5g: < SoYo vegetation loss= > 30Yo vegetation loss

flected less light in the more vigorous, wetter environment ofthe 1991 image. The errors resulting under these four condi-tions were conected through subsequent differencing, model-ing, and editing steps.

As an aid in solving the shadow and water problems inthe Band 7 difference file. a second difference file was devel-oped through subtraction of vegetation indexes. A vegetationindex for each date was created using a ratio of TM Band 3over TM Band 4. This ratio is the inverse of a typical vegeta-tion index, which is usually expressed as Band 4 over Band3. The inverted ratio was used so that the range of numberswas equivalent to that in Band 7 (i.e., higher values indicateless vegetation). Each index was scaled with a factor of 198.A vegetation index difference file was created by subtractingthe 19BB index from the 1991 index. The mathematicalexpression for the vegetation index difference file is

Aa: fV3rri l + V4"r11* 1981 - fVSuu,t+ V4 asit* 19Bl + 100

where

Al : change Pixel value,

V3u : .uglu"tance value of rL,l Band 3,

V4, : ,"11""tance value of ru Band 4,

BB : 1988,

91 : 1991 ,

j = row number, and

j : Iine number,

As with the Band 7 difference file, the index difference filewas visually analyzed to determine breaks indicating gxeaterthan 30 percent vegetation change. Each value was high-lighted on the imagery and compared to the 1989 and 1991cn photography. The reflectance values were divided intothree ranges representing relative loss and gain of vegetativecover on the ground (see Table 2),

Visual assessment of the index difference file indicatedthat it did not identify vegetation Ioss as consistently as theBand 7 difference file. In addition. the seasonal effect ingrassy and herbaceous areas appeared even more pronouncedin the vegetation index difference file. The index file didovercome the oroblems associated with shadow and waterIevel variationi.

A simple comparison model was constructed to identifyareas of conflict between the Band 7 and index differencefiles. Specifically, pixels with conflicting values were identi-fied as follows:

Index difference value < 75 and Band 7 difference> 106; orIndex difference value > 95 and Band 7 difference< 95.

Examination of the output file indicated that these areasof conflict consisted of either areas subject to excess shadowon the 198B image or areas affected by differences in waterlevel between the two image dates. These areas were subse-quently re-coded as "no change,"

A series of majority raster scans were conducted in sn-DAS to create homogeneous pixel areas of land-cover change,firm up boundaries between classes, and eliminate outlyingpixels from the classification. The output file was then visu-

1 - 9 49 5 - 9 9

100101 - 106707 - 255

PE&RS

77771 1 1 87773

TraLe 2. Curuce Cnreconres Beseo on VEcgrelor.r Inoex DrrreREr.rcE VALUEs

Vegetation IndexDifference Value Change Category

1 - 74 : > 30olo vegetation gain75 - 95 : no significant change

96 - 255 : > 30o/o vegetation loss

Tlau 3. Exluplr or Aner CxerucE Reponr

Urban Natural AreaPolygon-ID#

Acres by Change Class

Vegetation Gain Vegetation Loss No Change

blowdown typical of the edge effect of clearcuts on neighbor-ing forest stands.

Vegetation gain was due primarily to increased crownclosure in lower woody vegetation. The highest degree ofvegetation gain occurred in clearcuts where shrubs and smalltrees rapidly occupy cleared sites. Vegetation gain in forestand shrub areas typically occurred as ingrowth within smallcanopy openings. Areas that showed vegetation gain withinclosed forest polygons of the 1989 natural area coverage weresmall openings where shrubby and herbaceous plants couldbe seen by the satellite. These small areas are generalized bythe photo interpreter in Metro's 1989 coverage as part of alarger stand.

The change summary indicates some vegetation gain inrocky areas, Upon examination of the 1989 photography,these areas appear to be predominately cleared areas thatwere mislabeled as rock in the photo-interpreted 1989 origi-nal coverage.

ConclusionsLike the Portland Metropolitan Area project, most of the ini-tial projects were successful at developing image processingtechniques that identify change, measure the extent ofchange, and update a map of cls data layer to incorporate thechange. However, knowledge that a change has occurred isrelatively uninformative unless the change can be Iinked toan impact on resources or on benefits and costs of those re-sources. Therefore, an assessment of change needs to includean analysis of impact of the change. The second-year investi-gations focus on linking change in Iand cover and land useto environmental and economic impacts. Second-year plansalso include (1.) finalization of change detection productsthrough the completion of techniques and a survey of themarket, and (2) increased marketing activities to boost theexposure and acceptance of change detection products.

The pilot projects to date have brought focus to the proj-ect and meaning to the comparative analyses, The followingpreliminary technical conclusions have been developed:

r Assessing land-cover change in forested areas is fairlystraightforward and easily implementable with image subtrac-tion. The immediate adoption of these techniques by severalPacific Meridian clients supports its commercial value.

. Assessment of land-cover change in non-forested areas ismore difficult because of both the rapid occurrence of changeand the Iess-definable relationships between spectral changeand land-cover change.

o Using map subtraction to assess land-cover change is muchless desirable than image subtraction because apparent land-cover differences from map to map often result more from

0I U . J

20.0

ally edited. Editing focused primarily on errors in vegetationgain in riparian areas. The L:24,000-scale CIR 19Bg and 1991photography were visually compared to the change outputfile. Inconsistencies were manually corrected. The ERDAS ras-ter coverage was then converted to ARC/INFO polygons of veg-etation loss, vegetation gain, and no vegetation change,

Once the ARCANFO coverage was created, it was joinedwith the original natural area coverage and an attribute filewas created that indicated how much vegetation had beengained or lost from each polygon. An ASCII file was madefrom these data in the format shown in Table 3.

ResultsTable 4 summarizes the comparison of the vegetation changecoverage with the original 1989 natural area coverage. Sub-jective visual analysis of the photography indicated that Iossof woody vegetation was identified with a high degree ofprecision through the change detection process. The vast ma-jority of cover loss occurred within forested cover types,Mosi vegetation Ioss appeared to be due to timber liaivestingor clearing for development,

The change summary presented in Table 4 indicatessome minor vegetation loss in areas identified in 1989 asclearcuts. This anomaly may in part be the result of a combi-nation of vector-to-raster conversion error and the edge effectof clearcuts on adjacent forest stands. Conversion of the orig-inal vector coverage to 2s-metre cells resulted in minor in-clusions of forested areas in adjacent clearcut polygons.Vegetation loss identified in clearcuts may also be due to

03 9

000

Teele 4. CNeruee Suuunnv

neighbor-

Original Cover

TweBareClearcutClosed ForestOpen ForestScattered ForestMeadowRockClosed ShrubOpen ShrubScattered Shrub

Total

Total Unchanged

Acres702.00

7,375.7093,884.7010,525.904,610.40

10,488.009.30

2,758.401,983.102,609.00

128,S46.50

AcresZ O . J U

789 .517,245.98

648.78633 .41

27.674 .97

224 .98225 .76

L ,743 .47

4,964.4t

percent

5 7 . 3 91 . 3 36 .16

73.740.27

c J . q s

o . r o1 1 ? <

4 3 . 8 3

J . O J

Acres3.29

49.414,123.33

290.O7770.67

0.0047.O725.8787.26

4,838.02

percento.473 . 5 9{ . J Ut .1R

3.700.450.00

1 . 3 03.34

3 . 7 5

Acres682.00576.00

88,547.009,688.003,846.00

10,420.004.00

2,489.007,742.O01,401.00

119,397,00

percent97 .0042.0094.00s2.0083.0099.0045.0090.0088.0054.00

93.00

differences in methods used to make the maps rather thanfrom real changes in land cover.

o Relating iand-use change to land-cover change is also difficult because of varying correlations between spectral changeand land-use change. For example, parcels in the hardwoodrangelands of California are frequently subdivided into five-acre "ranchettes." Often, roads, fences, and homes are con-structed on the ranchettes. The land use has changod fromranching to second-home development. The overall vegeta-tion and spectral response of the site may change liltle, butthe impact on wildlife populations resulting from the fencesand the presence of humans can be major.

o For certain site-specific applications, even high spatial reso-lution imagery does not have enough detail for the produc-tion of image maps that wiii be used in on-the-groundresource management and inventory updating. However, re-cent software developments make it relatively easy to ortho-couect small-scale aerial photography (1:31,000) for theproduction of image maps.

Pacific Meridian's marketing of change detection productshas identified four issues in the commercialization of chaneedetection technologies:

o Product Awareness - While manv clients oerceive the needfor change detection, they lack

"rr"."tr"s of the potential for

the use of satellite imagery in change detection. Successfulmarketing requires both introducing the client to the productand creating a level of comfort with the new technology. Theproducts produced to date with the EOCAP contract have ex-tremely high value as concrete evidence to increase prospec-tive client comfort with the technology.

. Image Cost - The cost of imagery comprises up to 90 percentof a change detection project in sharp contrast with vegeta-tion{ype mapping where the cost of imagery totals approxi-mately 10 percent to 30 percent of the cost of a project. Highimage cost adversely impacts the growth of the change detec-tion market. Nevertheless, even with high data cost, imagesubtraction has been popular with clients.

o Ease of Technical Implementation of Change Detection - La-bor costs and expertise requirements for image subtraction inchange detection are relatively low. Often, clients are moreinterested in performing change detection in-house ratherthan contracting the work out. This suggests a focus concen-trated in training and image sales, as well as the need identified in image processing services.

Pacific Meridian considers the NASA EOCAP II LUCCASproject to be a vital component to the development of futureinnovative services for our clientele. It has afforded us theopportunity to stay on the cutting edge of change detectiontechniques by supplying Research and Development (R&D)funding. An R&D component is a very valuable commodityrarely available to young, rapidly growing companies.

AcknowledgmentsThe authors wish to thank both NASA and EOCAP corDoratepartners for their continuing support and enthusiasm for thisproiect.

ReferencesAdeniyi, P.O., 19S5. Digital Analysis of Multitemporal Landsat Data

for Land Use,lland Cover Classifications in a Semi-Arid Area ofNigeria, Photogrcmmetric Engineering & Remote Sensing,57(71):77 67-77 7 4,

Byrne, G.F., P.F. Crapper, and K. K. May, 1990. Monitoring LandCover Change by Principal Component Analysis of Multitem-

336

poral Landsat Data, fiemofe Sensing of Environment, 1O:775-184.

Carper, W. Ioseph, Thomas M. Lillesand, and Ralph W. Kiefer, 1990.The Use of Intensity-Hue-saturation Transformations for Merg-ing SPOT Panchromatic and Multispectral Image Data, Photo-grammetric En gine ering & Remote Sens ing, 56 (4) :4 5 9-467.

Crist, 8,P., and R.C. Cicone, 1984. Applications of the Tasseled CupConcept to Simulated Thematic Mapper Data, PhotogrammetilcEngineering & Remote Sensing, 50:343-352.

DeGloria, S.D., 1985. Evaluation of Simulated SPOT Imagery for theInterpretation of Agricultural Resources in California, P.holo-grammetric En gine ering & Remote Sensrng, 5 1 (8) :1 1 03-1 1 08.

Ehlers, Manfred, Mark A. fadkowski, R. Richard Howard, and DavidE. Brostuen, 1990. Application of SPOT Data for RegionalGrowth Analysis and Local Planning, Photogrammetric Engi-neering & Remote Sensrng, 56(2):175-180.

Fung, Tung, and Ellsworth LeDrew, 1987. Application of PrincipalComponents Analysis to Change Detection, Photogrammetric En-gi n eerin g & Rem ote Sensrng, 5 3 (1 2) : 1 649-1 65 8.

Hame, T.H., 1986. Satellite Image Aided Change Detection, RemoteSensing-Aided Forest Inventory, Research Notes 19, Departmentof Forest Mensuration and Management, University of Helsinki,Helsinki, Finland, pp. 47-60.

Hodgson, Michael E., fohn R. Jensen, Halkard E. Mackey, Jr., andMalcolm C. Coulter, 1988. Monitoring Wood Stork ForagingHabitat Using Remote Sensing and Geographic Information Sys-tems, Photogrommetdc Engineering & Remote Sensing,

fampoler, Susan M., and Barry N. Haack, 1990. The Use of SPOTData to Map Urbanization and Analyze Change: Kathmandu, Ne-pal, Technical Papers, ACSM-ASPRS Annual Convention, ImageProcessing/Remote Sensing, Vol. 4.

fensen, J.R., 1981. Urban Change Detection Mapping Using LandsatDigital Data, The American Geographe4 8(2):727-747.

1985. Urban Change Detection Mapping Using Landsat Digi-tal Data, The Suweillant Science: Remote Sensrns of the Envi-ronment, Second Edjtion (R.K. Holz, editor), Johi Wiley andSons Inc., New York, pp. 310-325.

1 986. Introduc tory D igital Image Pro c e ssing, Prentice-Hall,Englewood Cliffs, New Jersey, 379 p.

Jensen, ).R., and D.L. Toll, 1982. Detecting Residential Land Use De-velopment at the Urban Fringe, Pftotogrommetric Engineering &Remote Sensing, 48:629-643.

Kauth, R.f. , and G.S. Thomas, 1976. The Tasseled Cap - A GraphicDescription of the Spectral-Temporal Development of Agricul-tural Crops as Seen by Landsat, Proceedings, Symposium onMachine Processing of Remotely Sensed Data, Purdue Univer-sity, West Lafayette, Indiana, pp. 4B-41-48'51.

Leckie, D.G., P.M. Tei l let, D.P. Ostaff, and G. Fedosejevs, 1988. Sen-sor Band Selection for Detection of Current Defoliation Causedby the Spruce Budworm, Remote Sensing of Environment,Z O : J I - J U .

Martin, L.R.G., 1986. Change Detection in the Urban Fringe Employ-ing Landsat Satellite Imagery, PIan Can., 26(z):782-190.

Martin, Larry R.G., and Philip J. Howarth, 1989. Change-DetectionAccuracy Assessment Using SPOT Multispectral Imagery of theRural-Urban Fringe, Remote Sensing of Environment, 30:55-66.

Odenweller, fulie 8., and Karen I. Johnson, 1984. Crop IdentificationUsing Landsat Temporal-Spectral Profiles, Remote Sensing ofEnvironment, 14:3 9-54.

Richards, I.A., 1984. Thematic Mapping from Multitemporal ImageData Using the Principal Components Transformatiorr, nemoteSensing of Environment, 16:35-46.

Toll, D.L., J.A. Royal, and I.B. Davis, 1980. Urban Area Update Pro-cedures Using Landsat Data, SP Fall Technical Meeting, Niagara

Falls, American Society of Photogrammetry, Falls Church, Vir-ginia, pp. RS-1-E-1-RS-1-E-17.

Verbyla, D.L., 1990. Building Predictive Models from Geographic In-formation System, Proceedings of the Symposium on State-of-the-fut Methodologr of Forest Inventory, USDA Forest Service,Pacific Northwest Research Station, General Technical ReportPNW-GTR-z63, pp. 403-408.

Vogelmann, James E., and Barrett N. Rock, 1988. Assessing ForestDamage in High Elevation Coniferous Forests in Vermont andNew Hampshire Using Thematic Mapper Data, fie.rtrofe Sensrngof Environment, 24:tO9-727.

Wagner, David G., L992. Integating nemote Sensrng and EcologicalModeling for Assessmg Terrestrial Ecosystems, Dept. of ForestSciences, College of Natural Resources, Colorado State Univer-sity, 23 p.

Welsh, R., and M. Ehlers, 1987. Merging Multiresolution SPOT HRVand Landsat TM Data, Photogrammetric Engineering & RemoteSensing, 53 (3):301-303.

Yokota, T., and Y. Matsumoto, 1988. Detection of Seasonal andLong-Term Changes in Land Cover from Multitemporal LandsatMSS Data, Proceedings of the 9th IG, XSS Symposium, ESA SP-284, Edinburgh, Scotland, pp. 215-278.

1993 ACSM/ASPRSTECHNICAT PAPERS

"Looklng to the Future with an Eye on the Past"

Proceedings of the 1993 ACSM/ASPRS Annual Meetingheld in New Orleans, Louisiana, February 1993.

Even if you couldn't attend the 1993 ASPRS/ACSMAnnual Convention in New Odeans, this 3 volume setcan help you discover what's new in the industry.karn the latest research and theory in GIS, RemoteSensing, Photogrammetry, and Cartography.

VOL. T - AGSMStock # 4933-1. pp.422.

r The 1990 Census and GIS: A Case Study of Day-care Accessibility in East Baton Rouge Parish

r Real-Time Kinematic GPS for Surveying: Centi-meters in Seconds

r Examination of the Validity of State Plane Projec-tion Computations in PLSS Applications

r Surveying and Mapping &lucation: An Interdisci-plinary Approach

I Disputes Between Adjoining l-andowners

r Automated Classification of l-andsat Data Using anInduction-Based Model

r AVHRR for Forest Mapping: National Applica-tions and Global Implications

I Development of Modular Courses for RemoteSensing, Photogrammetry, and LIS/GIS Instruction

Forest and Wetland Monitoring Using an Integrat-ed cISChange Detection Using Multi-Temporal FeatureSpace With Digital TM Data

VOL.3 - ASPRS: GlS,Photograrnmetly. and Modellng

Stock # 4933-3. pp. 458.

Spatial Autocorrelation in Remotely-Sensed andGIS DataFeature Mapping in a Photogrammetric GISApplication of Neural Networks in Forest Eco-logical ClassificationPractical Guidelines for the use of GPS Photogram-metryAn Approach to Automated Terrain ClassificationFrom Digital Elevation Model DataIntegrating a Resource Assessment Model intoARC/Info GIS: A Spatial Decision Support Sys-tem Development

1993. Three volumes. 1,354 pp. $30 each(sofrcover); I.SPRS Members $20 each. Stock #4933-1, 4933-2, 4933-3.

To Order. See The ASPRS Store.

I

T

I

I

I

t

t

# 4933-2. pp.474.