Upload
khangminh22
View
0
Download
0
Embed Size (px)
Citation preview
EDGAR A. WORK, JR.Environmental Research Institute of Michigan
Ann Arbor, MI 48107DAVID S. GILMER
U. S. Fish and Wildlife ServiceJamestown, ND 58401
Utilization of Satellite Data forInventorying Prairie Pondsand Lakes
LANDSAT-1 data were used to discriminate ponds and lakes forwaterfowl management.
INTRODUCTION
T HE PRIMARY BREEDING AREAS of NorthAmerican waterfowl are the Dakotas,
the southern portions of the prairie provinces, north-western Canada, and parts ofAlaska. These areas are annually the sites ofsystematic surveys conducted by the U.S.
lished sampling transects. Numbers and distributions of natural ponds and lakes as tabulated during these flights are an importantcriteria used in assessing annual waterfowlproduction. In order to explore procedureswhich could enhance FWS capabilities formonitoring annual and seasonal changes in
ABSTRACT: By using data acquired by LANDSAT-1 (formerly ERTS1), studies were conducted in extracting information necessary forformulating management decisions relating to migratory waterfowl.Management decisions are based in part on an assessment ofhabitatcharacteristics, specifically numbers, distribution, and quality ofponds and lakes in the prime breeding range. This paper reports on astudy concerned with mapping open surface water features in theglaciated prairies. Emphasis was placed on the recognition of thesefeatures based upon water's uniquely low radiance in a single nearinfrared waveband. The results of this recognition were thematicmaps and statistics relating to open surface water. In a related effort, the added information content of multiple spectral wavebandswas used for discriminating surface water at a level of detail finerthan the virtual resolution of the data. The basic theory of thistechnique and some preliminary results are described.
Fish and Wildlife Service (FWS) in cooperation with the Canadian Wildlife Service andvarious states and provinces. The surveysserve to aid management decisions relatingto annual hunting regulations and researchneeds by providing an estimate of population size and annual reproduction. Thesesurveys rely chiefly upon observations madefrom low-flying light aircraft traveling estab-
water conditions, an evaluation ofLANDSAT-l (formerly ERTS-l) sensors wasconducted (Work 1974; Work et al., in press).Objectives of that investigation were to mapand tabulate statistics on surface water conditions and to determine changes in wetnessbetween spring and summer during a twoyear period in a glaciated prairie region located in east-central North Dakota. This
PHOTOGRAMMETRIC ENGINEERING AND REMOTE SENSING,
Vol. 42, No.5, May 1976, pp 685-694.685
686 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976
paper reports on the techniques used to detect and characterize open surface water features.
PRELIMINARY CONSIDERATIONS
Level thresholding of a radiation signal in,a near-infrared waveband is a reliable andsimple technique for delineating surface water. This technique is effective because atnear-infrared wavelengths the apparentradiance of water is usually uniform andlower than for other terrain objects. Thus,using an appropriate near-infraredwaveband, water may be delineated by accepting scene points with low radiance values (classified as water) while rejecting allvalues above a certain threshold (non-water).In order to appreciate the use of and limitations to this technique a brief discussion follows.
The apparent radiance of a body of wateris the result of (1) reflections at the air-waterinterface, (2) reflections from particulatematter suspended in the water volume, and(3) reflections from the bottom. Because thefield-of-view of the LANDSAT scanner islimited to near vertical observations (nominally to within 5.78 degrees of satellite nadir)and because water surfaces reflect specularly, radiation reflected by water to the scanner must emanate from a sky position nearthe zenith. Given the northerly latitudeswhich characterize the glaciated prairies, theLANDSAT scanner does not view waterreflected direct solar radiation (i.e., theground specular point is at a considerabledistance outside the field-of-view of thescanner).* This leaves only that fraction of
• Strong (1973), through the combined use ofimagery collected over ocean surfaces fromLANDSAT-l and NOAA-2 satellites, found thatwhen the sun elevation exceeds 55 degrees, theLANDSAT imagery was subjected to considerablecontamination by sunlight even though the specular point was nearly 550 km from nadir. Strumpfand Strong (1974) termed this condition "diffuseglitter." The condition is due to the sea state or theslope of wind driven surface waves. In NorthDakota the maximum solar elevation angle duringthe time of the LANDSAT-l overflight (approximately 1015 local sun time) was 59.5 degreesalong the southern boundary of the state at thesummer solstice. This would indicate that waterbodies at these latitudes were capable of producing diffuse glitter for a limited period of time centered on 22 June. However, surface conditions oneven the largest water bodies found in the NorthDakota prairies were considerably smoother thanthose found in the oceans. No evidence of diffuseglitter was observed during the course of thisstudy.
diffuse skylight which emanates from anear-zenith sky location to impinge upon thewater surface and thence to be reflected tothe scanner. In relative magnitude, however,diffuse skylight is much weaker than directsolar radiation, especially in the nearinfrared and under clear sky conditionswhen optimal satellite observations are possible. McDowell (1974) illustrated the magnitude and spectral differences between diffuse skylight and direct solar radiation (Figure 1). In addition, the reflected skylightcomponent is further diminished becausewater surfaces are a uniformly weak reflectorof radiation which impinges at any but veryoblique angles (Figure 2).
In considering reflections emanating fromparticulates within the water volume andfrom the bottom surface, water's absorptivitymust be considered. In the near-infrared,that fraction of radiation which penetratesthe air-water interface is largely absorbed,
. the extent of absorption being dependentupon the wavelength and the length of thewater path. This situation is shown quantitatively in Figure 3, which illustrates thespectral transmission of pure water for a variety of path lengths. Consequently, a sensorviewing a water body in a near-infrared bandreceives little or no radiation that may havebeen reflected by the bottom or volume suspended particulates.
The relative utility of different nearinfrared bands for detecting water shouldnot be predicated purely on the basis oflonger wavelengths. For example, the use ofa waveband in the 2.0- to 2.6-/-l.m atmospheric window is not optimal because of thedecreasing amount of solar radiation at thesewavelengths. It must be remembered thatmost terrestrial objects are relatively strongdiffuse reflectors and that with adequatesolar illumination such targets will contrastsharply with darker surface water features.An ideal waveband for delineating surfacewater is the 1.5- to 1.8-/-l.m spectral interval(Work and Thomson 1974; Work 1974). Sufficient solar illumination is available in thiswaveband to illuminate background objects.By using this waveband, it is also possible inmany instances to distinguish water which isoccluded by aquatic plants. Thesecapabilities are of particular importance forlower-altitude aircraft data in which case it isboth possible and desirable to identify smallponds, sheet water, and shoreline patterns oflarger ponds and lakes. AnalyzingLANDSAT-l scanner data, however, wefound band-7 (0.8 to 1.1 /-l.m) to be satisfactory except for an occasional omission error
SATELLITE DATA FOR INVENTORYING PONDS & LAKES 687
0.4 0.5
WAVELENGTH ("tm)
1.0 1.1
FIG. 1. Scene irradiance components for a clear day. (AfterMcDowell, 1974.) Spectral irradiance levels attributable toskylight alone may be significantly different and variable withwavelength due to variations in atmospheric haze. The conditions shown were recorded on an exceptionally clear day inNew Mexico on 30 October 1970 at a sun elevation angle of 43degrees.
in the case of shallow ponds carrying highsediment loads.
METHODS
Our delineation of surface water was accomplished by using a high-speed digitalcomputer which examined each scene pixel(picture element) and classified the pixel aseither water or non-water. In training thecomputer for this automated classification,
100,
tr.l
80
~«E-o() 60~..:lr..[;l
~ 40r.l
liir.lPo 20
o.f:=:+==+==+==+==+-+---+-t-----Io W ~ W ~ ~ ~ ro ~ 00
ANGLE OF INCIDENCE MEASUREDFROM NORMAL
FIG. 2. Variation of reflecting power of theair-water interface as a function of the angleof incidence (Natural or unpolarized light ofwavelength 0.589 JLm).
data samples were examined in order to determine typical water radiance levels andthe radiance levels of other scene materialsalso known to have low radiance characteristics. Experience has shown that variousscene objects exhibit radiances that may approach the low radiance of water dependingupon the specific near-infrared wavebandunder consideration, the geographic locale,and the phenologic state of scene objects. Ineastern North Dakota, dark prairie soils,
100,_~_=-.....,,:::--__
90
~ 80
'"~ 70
~~ 60..~... 50o'"fi 40
0:!:::>; 30
fl!~ 20
10
0,4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4
WAVELENGTH (jJm)
FIG. 3. Spectral transmittance of pure waterfor different path lengths. (Plotted after datafrom Sverdrup et 01., 1942.)
688 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976
50r----------------,
FIG. 4. Histograms used to locate a decision or threshold boundary for separatingwater from non-water terrain features .
fewer potholes or basin features and, because of its low relief, has been subjected tonumerous wetland drainage projects. Thisdifference in wetland occurrence warranteda stratification of the numerical results basedon these physiographic variations.
A computer-generated thematic map identifying surface water in a portion of the studyarea located almost wholly within the coteauphysiographic region was produced (Figure5). Although a digital map of the type illustrated graphically portrays water conditions,it does not represent an efficient form of datamanagement. Thematic maps as such, and
25
Soil
o
r'"i ~water
40 i,: i:, DecisionBoundary
~ ! i I;; 30 i! I~ i i~ ,.-~: I50 20: i!Xl i -'1 I~ ! I
10 : i II iI L, I
oil r-\..r-
Mollisols (formely referred to as Chernozems), have consistently approached thelow radiance values of water in the 0.8- to1.1-J.tm waveband. In such a situation, sampled water and soil radiance values wereexamined in order to discern relative differences. Typical histograms or frequency distributions are shown in Figure 4. In thesedata, both water and soil had distributionswhich were displaced from each other although slight overlap occurred in the tail regions. We have found that shallow water features are often represented within the leading tail of such a water distribution; as a result, the threshold or decision boundary waslocated nearer to the soil's density peak inorder to group the shallow water situationswith other surface water categories.
RESULTS
The intent of the recognition process wasto monitor changes in surface water conditions for a seasonal (May-to-July) and an annual period (July-to-July). Data gathered byLANDSAT-Ion 31 July 1972, 14 May 1973,and 7 July 1973 were analyzed. The samearea comprising 3311 km2 (1278 mi2) was observed on each of these dates. The study areaincluded portions of two different physiographic regions: a coteau or moraine featurecreated by stagnation ice, and a drift plain orlow relief feature of numerous groundmoraines. The drift plain inherently has
, .L'
.:.~ .,... "~.'
'. '. .~ / 1 i"
.. '
...''''-./
;.~ :' ...~, ..~....., ....
..\t ;' ...,., -/
l .,-
.f.1 .•.' ..'"'.",
.~.
. ~-
••
•.~
~ .--. .
,. '".' ."
. ..
FIG. 5. Digital water recognition map from ERTS observation 1008-16594 of31 July 1972. The map, obtained by thresholding channel MSS-7, depicts anarea northwest of Jamestown, North Dakota.
SATELLITE DATA FOR INVENTORYING PONDS & LAKES 689
the required manual interpretation, are notfeasible for wide area surveys on a routinebasis. For purposes of analysis, the data generally has greater value if it is presented interms of statistical tables. Figure 6 illustratesa computer-generated statistical tabulationof ponds and lakes observed on 7 July 1973in that portion of the study area lying in thecoteau physiographic region. The upper listing itemizes all ponds recognized in thescene and provides the location of each pondbased on a coordinate system derived from
the sensor's scanning geometry and also on ageographic coordinate system. A size frequency distribution of ponds in the entirescene is presented in the lower tabulation.
A graphical summary of pond frequencyfor two dates illustrates the pond size distribution for the coteau portion of the studyarea (Figure 7). The two curves show that adecrease in pond numbers occurred in the1973 May-to-july interval. Information onchanges in pond numbers over this seasonalperiod is an important data input to models
01JUlB C N47-221"098-22 N /lll101-21,w098-01 IH9-1tlS4)- SUH ElSl AlUI 192-486&-1\1 0-1-0-1 H EA.I" CUTS DATU
lA8UlA110H Of Kf(UGNIlED lilAlER BODIES
AREA AREA
LA' LONG SCAN LINE POINT !ACRES} IHECTARESIitT. HID 99."194 1439 '" 20.867 8.""'S/01.3909 99.5~H 101040 " 1.098 .44"'o7.37Jl 9<;.44£3 1'042 ". 1.098 .4....41.3t:H 99.4051 144b lib. 1.4118 3.11141.Jbb8 ..... ·,j99 1449 110 1.09' .44.... 1.HZO 9').Z':lt>O 1458 ." 1.098 ......."T.llse ;O;.21'n 111058 "0 1.098 .4"' .... 1. 3181 'lI9.2"'03 IllbO ... 4.393 1.778lt7.3':'89 99.)953 t4t>4 ". 2.1'll> .889.. 1. ~282 99.259'1 1460" ... 2.1'i6 .889'o7.319\) i9.21"9 l'tbl ." 3.295 1.333'01.3170 99."C'j3 14U ". 2.196 .889'01.3208 99.2b99 14b8 .., 4.39) 1.118,,1.3532 (,lq.I,~tll, H71 ". 1.C.q8 .1,1,1,
1,1. )l91, QQ.Z50q 11,11, M, U. 2.196 .889
f...I,1.llb" 9Q.2I,58 1411 .u 1.098 ... 4 .... 1.);:10 99.)19) 11,18 ,.. 3.195 1.U)
\\ II,b."~.·Z Qy.t7ub ZI!J1o 'ij ." 7b.lf11 31.112 'ij"b.!ll~ ) 9,.2i91 21'11 '0. 2.IV6 .iIUl9106o.!Jl6o) "'i ... ',IO 2158 ." 21."~b 11.111..b.8H9 99."::ill 2159 ••0 I.V98 .,.1,,,
I,b.H: .. 9·1.)t.')5 Zhtl ." 83.lob6o 33.1791,60.8122 f,JLZBe 21600 'is I.(,I'U .410".. 6.cl2Jd 99.JI61 11U '0' 460.120 Ilt.Ob7"b,~"18 99."8102 ZIT2 '" 2.196 .889"6.80'51 ~9.Z""b 2112 '0' 19.708 8.0004tl.80,H 9'Ll1011 2175 710 10.393 1.178
StENE N.lI. COTUU 1I NES 13ee. THRU 1428LINES 1429 THRU 1452LINES 1"53 THRU 1'099LINES lSOO THRU 1571LINeS 1512 THRU 2115
SCENE AREA- 922 SQ. M'.2389 SQ. K".
,OINTSPOINTSPOUHSPOiNTSPOINTS
] THRU '''0] THRU 314) THRU 491] THRU 622] THFtU 74T
FrIEL LENGTH _ 79.00 "ETERS PIXEl "10TH. 560.26 METERS
"Ouf. • NORMALPOI"fTS COUNTED If VULTAGE IS .GE. 0.0 ANa .LE. 1.0 WOLTS
DISTRI8UTlON OF LAKES IN THE SCENE
BY AREA
AREA IACRESI.25 TO .50.50 TO 1.00
1.00 TO 2.002.00 TO 3.003.00 Tu "0.(,0"0.0(, TO 60.006.00 TO 8.008.00 TO IO.CO
10.00 To 15.CO15.00 TO 20.00",v.oo TO 25.0025.Ju TO 30.003\).001 to 1,0.004~.00 10 !JU.OO~O.OO T(J 1S.0a7S.CO 10 100.00
11I0.00 TO lSG.OOIS?·)\) TO 200.00OVER 200.00
UtA IHECTARESI.10 TO .20.20 TO .40.40 TO .81.81 TO 1.21
1.21 TO 1.621.602 TU 2.432.4) TO ).2"3.24 fa ".OS4.05 TO tI.OTb.Ol TO 8.091.09 fO 10.12
10.12 TO 12. H12.14 fa 16.191t1.19 10 10.BIO.B TO 30.3530.35 JU 40.41100.'" f" 60.1060.10 fO 80.9'"OVER 80.9"
FREIllUEHCYoo
1.9.0.,••,.21,.281018..11181.13,13
FIG. 6. Example of computer printout of pond and lake statistics for an area withinthe Coteau du Missouri physiographic region of North Dakota. Data collected byERTS-l at 1654 GMT on 7 July 1973.
690 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976
currently used for predicting waterfowl production (Geis et al., 1969).
In examining these data it should be recognized that the pond sizes listed by thecomputer should in practice be termed "apparent size." This is because each pixel ofdata was examined and determined to beeither totally water or not water. Many pixelslying on the perimeters of ponds and lakesundoubtedly contained some unrecognizedand untabulated water. This caused the surface areas of virtually all water features to beunderestimated. In terms of percentage, theerror were greater for the smaller ponds andfor those of irregular shape (Le., those havinga high ratio of perimeter length to area). Thevery small ponds, of course, would not berecognized. Generally a pond must havebeen at least 0.4 hectare (1.0 acre) in size tobe recognized. Recognition of a 0.4 hectarepond was dependent upon whether the pondwas wholly included in one digital sample(i.e., within a pixel) or fractionally distributed over several pixels. In general, it isproblematic whether ponds in the 0.4- to1.6-hectare (1.0- to 4-acre) size class wererecognized. Above 1.6 hectares, ponds werenearly always recognized but not necessarilyrecognized at their full areal extent.
IMPROVEMENTS IN SPATIAL RESOLUTION
Because prairie ponds are frequentlysmaller than 0.4 hectare (1.0 acre), it was apparent that many water bodies were not
-,
L
1L~
~--l
'-'1L_-
1Ln--d--l
tabulated. We , therefore, undertook a radically different recognition technique whichoffered promise for improving the data's resolution limit. The remainder of this paperdescribes that approach and discusses somepreliminary results.
When the instantaneous-field-of-view(IFOY) of a scanner is large with respect tothe scene objects being scanned, a single resolution cell may contain a mixture of materials. In attempting to improve upon the data's resolution capabilities, we have utilizedthe added information content of multiplespectral wavebands to estimate the proportions or fractions of materials which fill theIFOY. The technique termed "proportionestimation" or "mixtures extimation" wasfirst outlined by Horwitz et ai. (1971) andfurther described by Nalepka et ai. (1972).Before this study, the application of thistechnique was largely developmental in nature, and its use in this and anotherLANDSAT-l study (Malila and Nalepka,1973 and 1974) must be considered to beamong the first attempts to test its applicability in an operational context.. A discussion of proportion estimation
theory is beyond the scope of this paper,however the essence of the technique can bedescribed in geometric terms. Assume that adata set comprised of two spectral channels,~I and ~2' contains three pure and unique
materials - A, B, and C. This situation is depicted in Figure 8 where the signature
Legend14 May 1973
-- 7 July 19'73
---------------,
12346810o.oo14+-t-t--l--+--+----t---t-1_--+I_---III-- ---:I=--=--=n=-=n=n=n~n~
15 ~ ~ ~ 40 50APPARENT POND SIZE (ACRES)
FIG. 7. Summary ofsize distribution ofponds in the coteau stratumfor a seasonal period. Where the pond size increments are greaterthan one-acre, the data have been averaged over the increment.
SATELLITE DATA FOR INVENTORYING PONDS & LAKES 691
B
A1--'
FIG. 9. Geometric interpretation of estimate for a special case. The unknown, Z,lying outside the signature simplex is amixture of materials A and C.
A1--'
FIG. 8. Geometric interpretation of meansof signature mixtures. In the case illustrated, the unknown, X, is a mixture of threepure materials - A, B, and C - which formthe vertices of the signature simplex.
verse ratio by which the simplex triangle'sleg A-C is divided by a line drawn from Zorthogonally to that leg. If the unknown isquite distant from the signature simplex (described in terms ofax2 distance), the algorithm designates the unknown as an alienobject or an object composed of none of thesimplex materials.
Although the above description has beenlimited to three pure and unique materials intwo-dimensional space, the concept is easilyexpanded to situations where many objectmaterials exist in a spectral hyperspace. Inapplying the algorithm, however, it is necessary that two operational constraints be observed: (1) at least n-l spectral channels ofinformation are required to satisfactorily estimate mixtures ofn-object materials, and (2)the signatures for the materials in a mixturemust not be similar and no one signaturemust come close to the weighted average ofthe other signatures. If either of these latterconditions occur, the signature simplex issaid to be degenerate, and the estimate ofproportions may be poor or invalid.
In our study, the intent was to delineatewater in a mixture of scene materials,thereby making it possible to improve thesize estimates of larger ponds and, more significantly, to recognize small ponds whichotherwise would not have been detected.Examining the data collected in the July1973 time frame, it appeared that only two ofthe four LANDSAT wavebands were uniquein representing the majority of scene materials present in eastern North Dakota. Bands5 (0.6 to 0.7 J.tm) and 7 (0.8 to 1.1 J.tm) appeared to be best for distinguishing mostscene materials while bands 4 (0.5 to 0.6 J.tm)and 6 (0.7 to 0.8 J.tm) were respectively redundant. This being the case, it appearedunlikely that the proportion estimation algorithm could function in using more thanthree signatures. As a result, we chose forprocessing signatures representing water,bare soils, and green vegetation. Using thesesignatures, the computational algorithm wasapplied to an area of 287 km2 (110 mi2). Theresultant computer output was a set of waterproportions for each scene pixel. A waterrecognition map generated from this outputis shown in Figure 10 and, for comparison, amap generated by using the singlewaveband thresholding algorithm is shownin Figure 11. In the proportion estimationmap, the symbol density is related to theproportion of water estimated for that pixel.In order for the map accurately to portray thescene, certain percentage or acceptance
c
c
z1I
A
B
means for the three materials are shown intwo-dimensional signal space. In the nondegenerate case, each pure signatureis a distinct vertex of the closed geometricfigure or signature triangle. If an unknownscene element (IFOV) consists of a mixtureof all three materials, the signature generated by this unknown element, X, lies withinthe simplex. An estimate of the proportion ofeach pure material constituting the unknownelement is obtained by drawing a line from avertex through the signal to be classified tothe opposite leg of the simplex. The fractionof the line between the crossover point andopposite leg defines the proportion of thecorresponding vertex material in the unknown. For the case illustrated in Figure 8,the unknown happens to lie at the centroidof the triangle, and its composition would be inthe ratio of 1/3, 1/3, and 1/3 of materials A, B,and C respectively. A case requiring specialgeometric interpretation is shown in Figure9. In this instance the unknown, Z, lies outside or on the edge of the signature simplex.The unknown is determined to be comprised of only materials A and C in the in-
692 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976
BarnesLake
, ..
""....... ,.","..~
'.,II" T .,'
.'
:
AlkaliLake
,J..... \
\FIG, 10, Water recognition obtained by use of the proportion estimation algorithm. The symbol density is related tothe proportion of water estimated for that pixel.
.~ '.~~es;,'"
"''00.~..,' .
~
1
f·
"........
','~~~1
; ..~.~. .
9'
FIG. 11. Water recognition obtained by thresholdingchannel MSS-7. The decision criteria is such that eachpixel has been classified as either totally water or not water.
SATELLITE DATA FOR INVENTORYING PONDS & LAKES 693
limits were determined for the output of thealgorithm. For example, it seemed appropriate to count pixel values of 86 per cent andabove as totally water. This proceduretended to account for the likelihood that avalue close to the signature mean (Le., closein terms of its probability of being that material) may in fact have been a pure samplerelated to that mean. Similarly, pixels showing less than 30 per cent water were assumed to contain no water at all.
In general, a detailed comparison of theclassification maps and related imagery indicated that proportion estimation significantly improved pond shape definition. FishLake and several nearby lakes shown in Figures 10 and 11 illustrate this fact. AlkaliLake (Figure 10), although recognized, wasnot accurately delineated. As implied by itsname, the lake apparently was high in dissolved and precipitated solids and as suchwas a lacustrine feature for which the watersignature utilized was not representative. Aninspection of this same lake in aerial photography indicated a color hue similar to thehighly saline (c.f., alkaline) lakes found occasionally throughout this area of NorthDakota particularly during periods of lowwater. It should be noted that Alkali Lakewas not omitted from our enumeration, onlymis-recognized in terms of size. Our primeobjective in applying the proportion estimation algorithm was to recognize many ofthe smaller water features which otherwisewould not have been detected. From anexamination of the results, it appears that allponds larger than 0.5 hectare (1.3 acres)were consistently recognized and that manysmaller ponds to 0.13 hectare (0.33 acre) as aminimum were recorded. This was athreefold improvement relative to thesingle-waveband thresholding technique.For the 2B7 km2 area illustrated both in Figures10 and 11, an increase of 131 per cent in thenumber of water bodies tabulated was observed relative to the number tabulated byusing the thresholding algorithm.
SUMMARY AND CONCLUSIONS
The mapping of open water as an indicatorof waterfowl habitat quality has been carriedout by using two different recognitiontechniques, a single waveband thresholdingapproach and a multiple waveband approachtermed "proportion estimation." The singlewaveband technique has proven simple toimplement. Its computer algorithm wasrapid and accurately recognized prairie lakesand large ponds. The resultant products of
this processing technique were thematicmaps and statistical tabulations describingopen s~rface water conditons. The mapsserved to portray visually the location andfrequency of surface water bodies but usually necessitated additional interpretation.Statistical tabulations provided a useful collation and data summary.
The proportion estimation technique,which utilizes the increased informationcontent of multiple spectral bands, estimatesthe fraction of a resolution cell which may becomposed of open water. This technique allowed for the recognition of a greaternumber of small ponds not previously identified and improved the apparent spatial resolution of the data by a factor of three incomparison to the single wavebandthresholding technique.
Because of the effectiveness of the severalrecognition techniques described, development of an operational system using highaltitude sensors for synoptically characterizing wetland habitat conditions appears to bea realistic goal. In future operational systemswhich require the delineation of water features smaller than a resolution cell in size,we feel that a combination of the singlewaveband and the proportion estimationtechniques is likely to be employed for thesake of accuracy and efficiency. Such a dualprocessing approach would effectively usethe single waveband technique to identifythose resolution cells comprised totally ofwater. The proportion estimation algorithmwhich is slower running and consequentlymore costly would be used to delineate onlywater peripheral to already identified pondsand lakes and those water features which areless than a resolution cell in size.
ACKNOWLEDGMENTS
This work was supported in part by NASAContract S-70243AG-4 to the U. S. Fish andWildlife Service, Northern Prairie WildlifeResearch Center, Jamestown, North Dakotaand their subcontract, USDI 14-16-0008-75to the Environmental Research Institute ofMichigan, Ann Arbor. We are especiallygrateful to Harvey K. Nelson who was theprincipal investigator on this program and toA. T. Klett who ably assisted in much of theassociated field work.
REFERENCES
Geis, A. D., R. K. Martinson, and D. R. Anderson.1969. Establishing hunting regulations and allowable harvest of mallards in the UnitedStates.]' Wildl. Mgmt. 33(4): 848-859.
Horwitz, H. M., R. F. Nalepka, P. D. Hyde, and J.
694 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976
P. Morgenstern. 1971. Estimating the proportions of objects within a single resolutionelement of a multispectral scanner. Proc. of7th Internat'l Symp. on Remote Sensing ofEnvironment, Willow Run Labs., Inst. of Sci.and Techno!., The University of Michigan,Ann Arbor, pp. 1307-1930.
Malila, W. A., and R. F. Nalepka. 1973. Atmospheric effects in ERTS-l data and advancedinformation extraction techniques. Proc.Symposium on Significant Results from theEarth Resources Technology Satellite-I, Vo!.I, 5-9 Mar. 1973, New Carrollton, Md., sponsored by NASNGoddard Space Flight Center,Greenbelt, Md., pp. 1097-1104.
Malila, W. A., and R. F. Nalepka. 1974. Advancedprocessing and information extractiontechniques applied to ERTS-l MSS data.Proc. Third Earth Resource TechnologySatellite-I Symposium, Vol. 1. 10-14 Dec.1973, Washington, D. C. sponsored byNASNGoddard Space Flight Center, Greenbelt, Md., pp. 1743-1772.
McDowell, D. Q. 1974. Spectral distribution ofskylight energy for two haze conditions.Photogrammetric Engineering 40(5): 569-571.
Nalepka, R. F., H. M. Horwitz, and P. D. Hyde.1972. Estimating proportions of objects frommultispectral data. Tech. Rep. No. 3165073-T, Willow Run Labs., Inst. of Sci. and
Techno!., The University of Michigan, AnnArbor. 47 pp.
Strong, A. E. 1973. ERTS-l anomalous darkpatches. NASA ERSP Weekly Abstracts, NTIS(E73-10747), U. S. Dept. of Commerce,Washington, D. C., pp 119-120.
Stumpf, H. G. and A. E. Strong. 1974. ERTS-lviews an oil slick? Remote Sensing of Environment 3: 87-90.
Sverdrup, H. U., M. W. Johnson, and R. H. Fleming. 1942. The Oceans. Prentice-Hall, Inc.,Englewood Cliffs, N. J. 1087 pp.
Work, E. A., Jr. 1974. Application of the Earth Resources Technology Satellite for monitoringthe breeding habitat of migratory waterfowlin the glaCiated prairies. Master's Thesis. TheUniversity of Michigan. 107 pp. available fromUniversity Microfilms, Ann Arbor, Mich.(Thesis Abstr. M-6698.)
Work,E. A., Jr., D. S. Gilmer, A. T. Klett. (in press).Utilization of ERTS-I for appraising changesin continental migratory bird habitat. TypeIII Final Report prepared for NASA, GoddardSpace Flight Center, under contract.
Work, E. A., Jr. and F. J. Thomson. 1974. A studyof waterfowl habitat in North Dakota usingremote sensing techniques: phase II. Tech.Rep. No. 101000-12-T, Environmental Research Institute of Michigan, Ann Arbor. 96pp.
The American Society of Photogrammetry
publishes two Manuals which are pertinent to its discipline:
Manual of Photogrammetry (Third Edition)Price to Price to
Members Nonmembers
1220 pages in 2 volumes, 878 illustrations,80 authors. (Sold only in sets of 2 volumes)
Manual of Color Aerial Photography
550 pages, 50 full-color aerial photographs, 16 pagesof Munsell standard color chips, 40 authors
$19.00
$21.00
$22.50
Send orders, or requests for further infonnation toASP, 105 N. Virginia Ave., Falls Church, Va. 22046