PEER.REVIEWED ANTIC IE

Detection of Holocene Lakes in the SaharaUsing Satellite Remote Sensing

E.F. Lambin, J.A. Walkey, and N. Petit-Maire

AbstractDuring the Holocene warm optimum, the Saharo-Sahelianlimit was located at 22" to 23" N Latitude, sahelian biotopesand surface water being recorded all across the cunentsoutJrern Soharo. The extension of lakes during the lastwarm Holocene optimum is related to the range and fre-quency of monsoonal rains and/or of Atlantic cyclones. Ourobjective is to produce a record of Holocene surface watetextension in an area presently hyperarid. This study demon-strates the feasibility of detecting the localization of pdeo-lakes by analyzing the spectral information of rcmotelysensed data. The spectral signature of evaporitic deposits re-lated to the drying up of paleolakes is separable in the redand near-infrared wavelengths from the spectral signature ofthe surrounding land-cover classes. The most appropriatespatial rcsolution - jn the range of resoiution cell sizes ofCurrent sensors - to discriminate the residual recotds of pa-leolakes is B0 metres.

lntrcductionGeological and biological evidence in the Saharan basins,east to west, demonstrates that past cold (glacial) global epi-sodes and past warm (interglacial) global episodes have re-sulted in, respectively, extensions and regressions of thedesert belt. The Saharo-sahelian limit can be defined asroughly corresponding to the 100-mm isohyet. During theLas{ Glacial Maximum, at 20,000 B.P., this limit was locatedaround 13'to 14" N latitude (Talbot, tg8+). During the Holo-cene warm optimum, this l imit was located at 22'to 23'NIatitude (Petit-Maire, 1991). Sahelian biotopes and surfacewater (lakes or swamps) were recorded at those latitudesthroughout the large basins from the Atlantic to the Red Sea(Petit-Maire and Kropelin, 19S1). To obtain a synoptic repre-sentation of surface water extension in the present Saharanarea, there is a need to connect spatially all the local obser-vations. Satellite remote sensing is the most appropriate toolto provide such a coarse scale representation because geolog-ical surface features show distinctive spectral patterns on re-motely sensed data (Petit-Maire and Page, 1992). In thispreliminary study, we test the feasibility of detecting Holo-iene paleolakes on the basis of the spectral information of

E.F. Lambin and J.A. Walkey are with the Department of Ge-ography and Centre for Remote Sensing, Boston University,675 Commonwealth Avenue, Boston, MA 0221'5.

N. Petit-Maire is with the Laboratoire de G6ologie du Quater-naire, C.N.R.S. Luminy, Case 907, 13288 Marseil le Cedex 9,France.

E.F. Lambin is presently with Monitoring Tropical Vegeta-tion (IATD) |oint Research Center, I-2L020 Ispra (Varese), It-aly.

PE&RS

remotely sensed data. If this feasibility is demonstrated, wecan deflne the spatial resolution of remotely sensed data thatis the most appropriate to detect paleolakes. We shall also at-tempt, if posiible, to characterize by remote sensing theirsurface and contours.

Radar imaging from space has been used to detect sand-buried channeli of ancient river and stream courses in desertenvironments (e.g., McCauley et al., 1sB2), but the detectionof paleolakes doel not necessarily require the ground-pene-traiing capability of radar systems. Passive satellite remotesensiig techniques have never been applied before to themappiig of paleolakes, but a few authors have shown thep.utiica-bititv of fine resolution satellite data in detection ofialine soil (Sineh et al., 1.983; Singh and Dwivedi, tg8g;Manchanda, 19-g+; Sommerfeldt ef 01., 1985). Some of theminerals formed at the surface of agricultural fields by salini-zation and waterlogging are the same as found in the resid-ual deposits of paleolakes. Gore and Bhagwat (1991) haveshown that higlily saline soils correspond on Landsat MSSdata to the brightest white patches. These authors could ex-tract different ilasses of saline areas using a Soil BrightnessIndex. In the work of Stoner and Baumgardner {1980), soilswith gypsic mineralogy, a very common feature for residualpaleolile deposits, were also found to have the highest spec-tral reflectanies on average, for all observed wavelengths,among over 240 United States and tropical soil series.

The Holocene Lakes in the Taoudenni Basin (Nofthem Mali)In the hyperarid central Sahara, a string of Holocene paleo--lakes is iocated along a 125-km long depression, between EIGuettara and Taoudenni. A detailed description of these isgiven by Fabre and Petit-Maire (1988) and Petit-Maire (1991).A bri"f

-r.r*-ary of their main characteristics will be given

here.Between 9000 and 4000 years B'P. + 500 years, perma-

nent lakes existed in this area. They were fed by runoff andkarstic emergences foom a limestone Carboniferous plateau tothe north, into a flat red clay country. The hydrological opti-mum was between 8500 and 6700 year B.P. when rainfallseasonality strongly decreased, indicating both monsoonaland Atlaniic preclpitation patterns. Between 4400 and 3500year B.P., all lakes dried up,some leavinS^saline evaporites-(Agorgott), others leaving carbonates, still forming visible pil-lari and yardangs (Haijad), or travertines with gypsum insmall depressions (Telig) (Figure 1). The upper layer of tt 'edepositsbf the Agorgott brackish lake consists of pale red

Photogrammetric Engineering & Remote Sensing,Vol . 61, No. 6, fune 1995' pp. 73L-737.

oogg-1112 I 95/6 106-73 1 $3. 00/0O 1995 American Society for Photogrammetry

and Remote Sensing

PEER.REVIEWED ARI ICTE

sebkha sandy clays with halite brought to the surface by cap-il lary action.

Detection of Paleolakes on SPOT Data

DataWe have processed and interpreted one seor image in themultispectral mode coverins the Taoudenni area. In thatmode,ihere are three specti l bands (green, red, and near-in-fraredl with a nominal soatial resolut ion of 20 metres. Thesedata were acquired on 6 May 1986 in clear-sky condit ionsand include the three oaleolakes described in the last section(Figure 2). The image iras been geometrically corrected andregistered to the 1:200,000-scale topographic map of Taou-denni in the UTM projection. No atmospheric correction hasbeen performed. A contamination of the satel l i te image by at-mospheric aerosols would only be a problem for this study i fthe optical thickness of the atmosphere had varied throughthe scene. As reference data, we have used the photogeologicinterpretation of the area fFabre, 1983); field data from Fabreand Peti t-Maire (rgee), Fabre (rgsr), and Oxnevad (19s1);and a prel iminary study of SPOT images (Page ef o,1., rgsr).

The digital image processing has been performed with thesof tware PCI EASI/PACE 5.0.

Spectral Separability MeasureA procedure commonly used in remote sensing is to deter-mine, before classifying the data, the mathematical separabil-ity of classes. We are particularly interested in knowing howseparable the spectral signature of paleolakes is from thespectral signature of the surrounding land-cover classes, de-fined through field surveys (Fabre, 1991). The spectral re-sponse of salts and carbonates has been compared to theresponse of four other classes: "Red Country-1" (sandstonecovered by sand), "Red Country-2" (exposed sandstone), car-bonates and Hammadian marls, and Hammadian l imestone.To perform the interclass separability analysis, the Jeffries-Matusita (Jtvt) distance (also called the Bhattacharrya dis-tancel has been calculated between the evaporites and theother classes. This distance is a measure ofthe average dis-tance between two class density functions which have a nor-mal distribution. It has been dernonstrated that it performsconsiclerablv better as a separabilitv measure for multivariatenormal speitral class models than hivergence and is equiva-

il

Figure 1. Field map of the area berween Taoudenni and El Guettara, with paleolakes location (from Fabreand Petit-Maire, 1988).

732 PE&RS

PEER.REVIEWED ARTICTE

Figure 2. lmage subset of the SpoT xs Band 3, with the paleolake locations. Paleolacustrine evaporitic de-posits appear as bright white patches. A : Agorgott; T : Telig; H : Haijad. Dikes can be seen as l inearfeatures. @ SPOT image Copyright 1990 CNES.

lent to transformed divergence (Swain et ol., L97L), By con-trast to the Mahalanobis distance, the IM distance has asaturating behavior with increasing class separation, as doesthe probabil i ty of correct classif icat ion. A fM distance of 2.0between two spectral classes would imply classif icat ion ofpixel data into those classes (assuming they were the onlytwo) with 100 percent accuracy, i f probabil i ty distr ibutionclass models are employed and i f the density functions arenormal (Richards, 19BG).

The spectral signatures of the different classes have beendefined by selecting training areas from the ref'erence data.Each class was represented by a sample of prototype pixels

TneLE 1. JEFFRTES-MArusrr^ ,j[L_t;lli::s BETwEEN Evnpontres AND rHE

which have, on the photogeologic map, the attributes of thatclass. Means, standard deviat ions, and covariance matricesfor the three sPoT bands were extracted for each class, andpairwise JIra distances were calculated between evaporitesind al l the other classes, f i rst, taken individual ly, and thenmerged in coarser categories (Table 1).

These results demonstrate that the evaporite and carbon-ate spectra are highly separable from the spectra of the otherclasses. I t can also be seen that, for the maximum-l ikel ihoodclassif icat ion, the potential for discriminating the evaporitesfrom the rest is larger if the other classes are treated sepa-rately rather than clustered together. This is explained by anincrease in covariance when dissimilar classes are clustered.

ClassificationThe scatterplot of the red and near-infrared bands has beenproduced (Figure 3). As can be expected (Crist and Cicone,isa+), most of the pixel data fal l along the so-cal led "soi l

l ine" (the diagonal of the scatterplot). Al l the evaporite pix-els are located at the extreme end of the soi l l ine, having thebrightest spectral response both in the red and near-infraredwavelengths. This observation suggests that an automatic de-tect ion of this class should be possible using the spectral in-formation in these two bands only. A classification of theevaporites spectral class has been attempted using a simple

CarbonatesSand and

covered Exposed Hammadiansandstonesandstone mar ls

Hamnradian Avr: rage lMl imestono distance

Evapor i tes 2.OO 2.OOEvapor i tes 2.OOEvapor i tes 1.951Evaporites

1 . 9 5 1 1 . 9 8 11 . 9 3 8

1 . 9 8 11 .904

1 . 9 8 31 . 9 6 91 .9661 .904

PE&RS

PEER.REVIEWED ARTICTE

o7

iat

ooo

v

b.AHA

Band 3

Bard 2T9?

b t i g h t n a ! ! v a l u a

Figure 3. Scatterplot of the red and near-infrared bands.

thresholding of the red and near-infrared bands. This isequivalent to a two-band parallelpiped classification. Thethird sPoT band was not used in the classif icat ion because.for this image, it is highly redundant with the red band (cor-relat ion coeff icient of 0.96). The photogeologic interpretat ionof the area has been used to select training areas over severalknown evaporite deposits. The threshold to extract paleolakepixels has been defined as the value two standard deviationsbelow the mean of the training data for that class, in the twobands. This interval is displayed on Figure 3 and includesabout 95 percent of the values of the class. The brightnessvalues ofthe threshold are 140 in the red band and 159 inthe infrared band. All pixels with a spectral response abovethat threshold correspond to evaporit ic pixels. Al l pixelswith a spectral response below that threshold correspond toother classes. This simple approach al lowed us to map theextension of the evaporit ic deposits.

Evaluation of the MethodGiven the great difficulty in performing field work in thispart of the Sahara, field data have not been systematicallymeasured at the pixel level, according to a pre-defined sam-pling scheme and with a location accuracy satisfactoryenough to control the satel l i te-derived map at the pixel scale.However, in addit ion to the photogeologic interpretat ion ofthe area, very good reference data from past expeditions inthe Taoudenni basin are available (Figure 1). Therefore, ourproduct was evaluated at the scale of the paleolakes.

Concerning the location of the paleolakes, the classif ica-tion results were in very good agreement with the referencedata. For the estimation of the areal extent of the oaleolakes.the evaluation is more comolex because the lakes-extensionhas varied throughout the Fiolocene. However, the areal ex-tent of the apparent evaporit ic deposits of the paleolakes, asestimated by remote sensing, can be compared to the totalareal extent of the topographical depression, as estimated in

734

the field. These estimates are based on an interDretation ofthe topography and geology of the landscape (Fabre, 1991),and correspond to the maximum limit of the pofenfial lakearea.

As shown in Table 2, the remote sensing analysis detectsthe surface features associated with paleolakes but only clas-sif ies a small portion of the topographical depressions whichare at the origin of these paleolakes. The bright signatureswhich are extracted by thresholding the red and infraredbands correspond to the surface of exposed salt, carbonate,and gypsum deposits, e.g., the residual records ofthe pastlake extension. This approach gives only a mlnimum figureof the past lake surface, and therefore leads to omission er-rors. The surface of the topographical depression sets the up-per limit of the lake extension, but this does not mean thatthe lake necessarily filled the whole depression at one periodof the Holocene. This raises the question of what are the realIimits of the paleolakes, knowing that there have been signif-icant variations of the lake level throughout the Holocene(Fabre, 19S1). If the threshold to discriminate the brightevaporite deposits is lowered, a commission error is pro-duced; pixels corresponding to the Hammadian carbonatesand limestone at the northeast of the lake boundaries are in-cluded in the class at the same time as pixels within the lakedepression.

DiscussionIn summary, the location of residual evaporitic paleolake de-posits by remote sensing is feasible, using a simple thres-holding of the red and near-infrared bands. Thisclassification technique can be easily automated to process alarge amount of satellite data in order to detect paleolakesover a much wider area of the Sahara. Concerning our sec-ond objective, the characterization by remote sensing of thesurface and contours of paleolakes, the results are disap-pointing. While it is not possible, with the field data cur-rently available, to quantify the magnitude of our omissionerror, it is likely to be large when considering the maximumactual lake extension, which is unknown but assumed to beapproximate to the area of the topographic depression. Thisunderestimation is explained by the fact that evaporites donot cover the entire extent of the original lake surface due tothe original physical process of evaporite formation and tothe subsequent aeolian deflation of evaporite deposits.

0ptimal Spatial ResolutionIn order to estimate the cost, in terms of data acquisition andprocessing, of mapping evaporite deposits throughout the Sa-hara, it is important to determine what is the optimal resolu-tion-cell size to discriminate the residual records ofpaleolakes. The selection of an optimal spatial resolution and

Tngre 2. CoMpARrsoN or tne EsrrvnteD RESTDUAL EvlpoRrrrc AREA wtrH THETornr SuRrncr oF THE DEpRESStorus nr TElrc. Heuno. nlto Aconcorr

Satel l i teestimation of

evaporitesArea of thedepression

Proportion ofthe

depressionarea

Tel igHaifadAgorgott'

7.3 km'?55 .0 km '52 .3 km '

0.2 km'6.0 km'?2.91 km'?

2.7 4o /o

1 .D9o/o

5 .CO" /o

1 Estimations only for the portion of the lake which is on the SpoTimage (44% of the total area of the lake).

PEER.REVIEWED ARI ICTE

spatial scale depends on the spatial structure of the land-scape and the type of information that is to be extracted(Woodcock and Strahler, 1,987).

MethodThe sPor data have been analyzed at their original spatialresolution [20 metres in the multispectral mode) and havethen been degraded to successively coarser resolutions, ap-

PE&RS

proximating the response of other satellite sensor systems: B0metres (Landsat Multispectral Scanner) and 1.1 kilometres(nvunn LAC/HRPT data). Landsat Thematic Mapper data havenot been simulated because their spatial resolution (30 me-tres) is only slightly coarser than t[e resolution of spot data.The simple aggregation and averaging of sror pixels do notproduce a perfect simulation of the actual Landsat andAVHRR data, because these sensors are characterized by dif-ferent point spread functions, modulation transfer functions,and spectral resolutions. However, this study only exploresranges of resolutions and there is no need to mimic exactlythe response of particular sensors. At every resolution andfor every paleolake, the data have been reclassified using thesame threshold values, and the new maps have been com-pared to the map produced at the 20-metre resolution.

ResultsThe results demonstrate that, among the spatial resolutionsof current satellite sensors, the B0-metre range (Landsat MSS)will lead to the highest classification accuracy. To illustratethis point, brightness values at three resolutions have beenplotted along an east-west transect which crosses the topo-graphical depression and the evaporitic deposits of the Hai-iad lake, a medium size paleolake. The different transects areshown on Figure 4 for the red band, and Figure 5 for the in-frared band. At the sPor resolution (Figures 4a and 5a), thebrightness values along the transect exhibit many local varia-tions, especially through the evaporite deposits (brighter pix-els). This is consistent with the findings of Woodcock andStrahler (1987) who have shown that, when the resolution-cell size is much smaller than the obiect to be discriminated,the local variations of brightness values corresponds tochanges in scene composition at a scale finer than the scaleof the obiect. This within-class variance decreases the spec-tral separability of the evaporite class and results in a lowerclassification accuracy. At the BO-metre resolution (Figures4b and 5b) some of the within-lake variation has beensmoothed by the averaging effect related to the larger resolu-tion-cell size. The spectral separabil ity of the paleolake wil lthus be higher at this resolution. At the 1.1-kilometre resolu-tion (Figures 4c and 5c), only one pixel has a value larger -but only slightly - than the threshold in the red band and,in the near-infrared band, the paleolake is barely detectable.This resolution is therefore too coarse compared to the sizeof the lake: most of the 1.L-km pixels fall across evaporit icdeposi ts and surrounding matei ia ls . and average thei r br ight-nesses values. The oresence of these mixed pixels removesthe possibility of loiating evaporite deposits reliably. Anysmaller or narrower paleolake would stay undetected. As aconclusion, the most appropriate spatial resolution in therange of resolution cell sizes of current sensors is 80 metres.

ConclusionThe digital analysis of remotely sensed data could detect andlocate the evaporitic deposits related to the drying up of pa-leolakes through an empirical analysis of radiances in thered and infrared wavelengths. Even though we demonstratedthat the areal extent of paleolakes cannot be estimated, theirsimple detection and location remains a useful appl icat ion,especially if applied at the scale of tropical deserts. It will al-Iow an inventory of previously unknown paleolakes to bemade. From that, it will be possible to infer changes in at-mospheric paleocirculat ion based on the presence of surfacewater and modifications of the Precipitation/Evaporation ra-t io. This wil l contr ibute to the improvement and test of cl i-

o:6

dootr

EED

@

200

180

160

't40

120

1 0 0

200

180

1 6 0

140

't20

100

200

180

160

140

120

100

threshold

2 0 0 0 4 0 0 0Dls tance (m)

(a)

6 0 0 0

o:6

6ooc

E

-9@

ot6

o6occIo

2 0 0 0 4 0 0 0 6 0 0 0

Dls tance (m)(b)

threshold

o 2000 4000 6000 8000Dls tance (m)

{ c )Figure 4. Transect of brightness values across Haijad atthree resolut ions: red band. (a) sPor resolut ion (20 m);(b) 80 metres resolut ion; (c) 1.1 ki lometres resolut ion.

PEER REVIEWED ARI ICTE

o:6

t,tooE

E

I@

200

180

160

140

120

1 0 0

threshold

2000 4000Dletance

( a )

6 0 0 0 8 0 0 0( m )

o:6

odoc

E

Io

1 8 0

1 6 0

140

't20

1 0 02000 4000 6000

Dlstance (m)( b )

og6

oooc

EE t

@

200

'| 80

1 6 0

1 4 0

120

1 0 0

threshold

0 2000 4000 6000Dlstance (m)

8 0 0 0

t c )Figure 5. Transect of brightness values across Haijad atthree resolutions: near-infrared band. (a) SpoT resolution(20 m); (b) 80 metres resolution; (c) 1.1 kilometres reso-lut ion.

matic models. In order to map and quanti fy the exfension ofsurface water during the humid optimum, remote sensingtechniques wil l have to be combined with f ield surveys oraerial photointerpretation for a finer estimation of the actualpast surface of individual lakes. The project of detecting andlocating most Holocene paleolakes in the now arid part ofnorthern Africa would involve the processing of a very large

736

amount of data. In order to cover the region lying between30'and 17" N, from the Atlantic shore of Mauritania and Oc-cidental Sahara to the Red Sea shore of Sudan, the handlingof 300 Landsat MSS scenes would be reouired.

In addit ion to i ts contr ibution to paleocl imatology, themethod presented here has potential applications in archae-ology. The presence of surface waters in the present Sahara,as in other dry areas, attracted animal and human popula-tions. Many Holocene lacustrine shores are rich in archaeo-logical sites. The detection of paleolakes can therefore beused to locate privi leged sites for f ield investigations.

AcknowledgmentsWe are most grateful to Farouk El-Baz and Jean Fabre fortheir comments on the manuscript. We also thank N. Pageand M. Bobo for their collaboration. The SPOT data were ac-quired thanks to financial support from CNES/INSU (ISIS Pro-gram) (N. Petit-Maire, I-cq Marseille) and nactc (J.Riser,Avignon University).

ReferencesCrist , E.P. , and R.C. Cir ;one, 1984. A phvsical lv based t ransformat ion

of Thematic Mapper data: the tasieierl-cap, IEEE Geoscience anrlRenote Sensing Transactions, GE 2213:256-263

Fabre,1. , 1983. Esquisse strat igraphiquc pr6l iminaire des d6pots la-cr.rstres quaternaires, Snhoro ou SahelT Qttaternaire rdcent dubossin du Taoudenni (N. Petit-Maire and /. Riser, editors), Edi-t ions du CNRS, Marsei l le , France. pp. 421.441. .

1991. Les lacs Hai iad et Agorgot t : cadre g6ologique et6volution, Pal1oenvironnenlents du Sahara. Lacs Holocdnes itTaoudenni lMol i l (N. Pet i t -Maire, edi tor) , Edi t ions du CNRS,Par is, France, pp. B1-S0 and pp. 1.1.7-1.3O.

Fabrc., /., and N. Petit-Maire, 1988. Holocene climatic evo.lution at22"-23'N f rom two paleolakes in the Taoudenni area (NorthernMali), Poleogeography, Paleoclintatology, Paleoecologv, 65:1311-"t48.

Gore, S.R., and K.A. Bhagwat, 1991. Sal ine degradat ion of Indian ag-rir;ultural lands: a case study in Khambhat Taluka, Gr-rjarat State(tndia), using satellite remote sensin g, Geocarto Internotional, 3..5 - 1 3 .

Manchanda, M.L. , 1984. LIse of remote sensing techniques in thestudy of d ist r ibut ion of sal t af fected soi l in N.W. India, /ournolIndian Societv Soil Science, 32:7O1-7 12.

McCatr ley, I .F. , G.G. Schaber, C.S. Breed, M.f . Grol ier , C.V. Haynes,B. Issawi, C. Elachi , and R. Blom, 1982. Subsurface val levs andgeoarcheology of the eastern Sahara revealed bv shuttle radar,Sr: ience, 21 8:1OO4-1 O2O.

Oxnevad, I . , 1s91. Le lac d 'Agorgot t : s6dimentologie du pal6olac, Pa-ldoenvironnenenh^ du Sahora. Lacs Holocdnes it Taoudenni[Mo1t j (N. Pet i t -Maire, edi tor) , Edi t ions du CNRS, Par is, France,p p . 1 3 1 - 1 4 0 .

Page, N. , B. .Sinron, and J. Fabre, 1991. Apport dcs images.SPOT, InPo]6oenvironnenel-tts du Sahara. Locs Holocnes Taoudenni(Mal i ) (N. Pet i t -Maire, edi tor) , Edi t ions du CNRS, Par is, Frant ;c,pp . 65 -21 .

Pet i t -Maire, N. , 1989. Interglacia l environments in present ly hyper-ar id Sahara: paleocl imat ic impl icat ions, Paleocl intato logv andPaleonteteorolo3;v: Modern and Past Pntterns oJ GlobalAtnospheric Transpot't (M. Leinen and M. Sarnthein, editors),Kluwer Acad. Publ . , Ber l in, pp. 632-661.

(editor), "1991. Pal1oenvironnentents du Sohara. Lacs Holoci-ne.s d Tooudenni (Mal i ) , Edi t ions du CNRS, Par is, 236 p.

Pet i t -Maire, N. , and S. Kr i ipel in, 1991. Les c l imats holocdnes du Sa-hara le long du Tropique du Cancer, Paleoenvironnentents rlLtSahara. Lacs Holocines d Taoudenni (Malil (N. Petit-Maire, edi-tor) , Edi t ions du CNRS, Par is, France, pp. ZOS-21O.

PE&RS

PEER.REVIEWED ARI ICTE

Petit-Maire, N. and N. Page, 1992. Remote sensing and past climaticchanges in tropical deserts: Examples of the Sahara, Episodes,1.5i113-11.7.

Richards, J.A., 1S86. Remote Sensing Digital Image Analysis,Springer-Verlag, Berlin, 281 p.

Singh, A.N., and R.S. Dwivedi , 1989. Del ineat ion of sal t af fectedsoils through digital analysis of Landsat MSS data, International

lournal of Remote Sensing, 10:83-92.

Singh, A.N,, L. Venkatarataman, A.K. Sinha, K.R. Rao, and R.S.Dwivedi, 1983. Utilization of Landsat data for delineating, map^ping and managing of soil resources. The problem and prospectsunder Indian conditions, Proceedings of the 17th InternationalSymposium on Remote Sensing of Environmenf, Ann Arbor,Michigan, pp.23L-240.

Sommerfeldt , T.G., M.D. Thompson, and N.A. Pront , 1985. Del inea-tion and mapping of soil salinity in southern Alberta fromLandsat data, Conadian lournd of Remote Sensing, 10:104-118.

Stoner, E.R., and M.F. Baumgardner, 1980. Physicochemical, Siteand Bidirectional Reflectance Foctor Characteristics of Uni-

formly Moist Soils, LARS Technical Report 111679, Purdue Uni-versity, West Lafayette, Indiana.

Swain, P.H., T.V. Robertson, and A.G. Wacker, '1.97L. Comparison ofDivergence and B-Distance in Feature Selection, InformationNote 020871, Laboratory for Applications of Remote Sensing,Purdue University, West Lafayette, Indiana.

Talbot, M., 1984. Late Pleistocene rainfall and dune building in theSahel, Poleoecology of Africa, 16:203-274.

Woodcock, C.E., and A.H. Strahler, 1987. The factor of scale in re-mote sensing, Remote Sensing of Environment, 21.13"17-332.

(Received 14 December 1992; revised and accepted 1 June 1993; re-v ised 22 Iune 1993)

Eric F. LambinEric F. Lambin is Assistant Professor at the De-partment of Geography and Research Associatewith the Center for Remote Sensing at BostonUniversity. He received his PhD in Geographyfrom the Universit6 Catholique de Louvain, Lou-

vain-la-Neuve, Belgium, in 1988. His main research interestis the analysis of land-cover change processes in intertropicalregions using remote sensing techniques. He is also inter-ested in the application of remote sensing in human ecology.

f.A. WalkeYl.A. Walkey received his B.A. in Archaeologyand Anthropology from Boston University. He iscurrently pursuing research in the application ofremote sensing techniques in the field of archae-ology.N. Petit'Maire

I N. Petit-Maire, a Director of Research at thetr- ,;| Centre National de Recherche Scientifique,

"-"'f France. has been working in the world deserts,iq"' mainlv the Sahara. over the last 20 vears. spe-l;$.i mainl.y the Sahara, over the last 20 years, spe-' wlr cializine in field research of past environmental-.'Y"- cralrzrng rn tield researcn oI past envrronmental

change across the most desolate hyperarid areas. Since 1.987,she has led IGCP Project 252 (Deserts: Past, Present, Future)and was elected one of the Vice Presidents of IUGS duringthe last International Geological Congress in Washington,D.C.

1'4th. Biennial Workshop on

Color Aerial Photography andVideography for Resource Monitoring

The Fourteenth Biennial Workshop on Color Aerial Photography and Videography for Resource MonitoringWorkshop was developed to bring together those individuals who are involved in the growing number ofapplications being developed using aerial photography and videography. This proceedings contains 22 of thepapers that were presented at this meeting.

The workshop was sponsored by the American Society of Photogrammetry and Remote Sensing (ASPRS)and the Department of Biological and Irrigation Engineering and Conference and Institute Division at Utah StateUniversity. Held in Logan, Utah, May 1993. Topics include:

. Systems and Image Processing I o General Applications . Agricultural Applicationso Systems and Image Processing II o Forestry Applications o Delineation of Wetlands and. Delineation of Wetlands and . Fisheries Habitat Monitoring Riparian Vegetation II

Riparian Vegetation I

1994. 224 pp. 6 colorplates. $55 (soficover); ASPRS Members $35. Stock # 4721.

For detallc orr otderlngr cee the ASPRS ctone ln thle louraat.

PE&RS