Post on 20-Dec-2021
transcript
c. J. TUCKER*NASA/Goddard Space Flight Center
Greenbelt, MD 20771E. L. MAXWELL
Colorado State UniversityFort Collins, CO 80523
Sensor Design forMonitoring Vegetation CanopiesOptimization of sensor wavelengths and bandwidths has beeninvestigated based upon the analysis of in situ spectralreflectance data collected from experimental plots of bluegrama grass.
INTRODUCTION
DESCRIPTION OF RESEARCH UNDERTAKEN
T HE RESEARCH was undertaken to evaluatevarious wavelengths and bandwidths,
corresponding to simulated sensors, by in-
study because they are morphologically oneof the least complex vegetation types. Thisreduced the variability inherent in the studyand allowed for a more meaningful interpretation of the results.
Two data sets, one from June and one fi'om
ABSTRACT: Optimization of sensor wavelengths and bandwidths hasbeen investigated based upon the analysis of in situ spectral reflectance data collected from experimental plots of blue grama grass.Sensor characteristics have been simulated by integration of spectral data over the region from 0.350 to 1.000 J1.m. Subsequently, theintegrated reflectance values were regressed against the canopy orplot variables (total wet biomass, total dry biomass, leaf water content, dry green biomass, dry brown biomass, and total chlorophyllcontent) to determine the relative significance between integratedreflectance and the canopy variables for the various wavelengthsand bandwidths simulated. Three spectral regions of strong statistical significance (0.35-0.50, 0.63-0.69, and 0.74-1.00 J1.m) were identified and found to be persistent both early and late in the growingseason. In addition to quantifying the significance ofvarious sensorwavelengths and bandWidths, the additive effects of adjacent spectral regions were also quantified. The results of this work enablequantitative judgments to be made regarding existing and hypothetical sensors and their effectiveness in monitoring green functioningvegetation.
tegrating narrow bandwidth (0.005 J1.m)spectral reflectance curves of blue gramagrass (Bouteloua gracilis (H.B.K.) Lag.) plots.Grass canopies were selected for detailed
* National Academy of Sciences-NationalAcademy of Engineering Research Fellow.
September, were chosen for analysis (TableI). The June data represented a less advanced phenological state when most of thestanding vegetation was alive or green,whereas the September data represented amore advanced phenological state of approximately equal amounts of dry standing live
1399PHOTOGRAMMETRIC ENGINEERING AND REMOTE SENSING,
Vol. 42, No. 11, November 1976, pp. 1399-1410.
1400 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976
and standing dead vegetation. The June dataprovided a much greater range of plot variables (total wet biomass, total dry biomass,leaf water content, dry green biomass, drybrown biomass, and total chlorophyll content) which will enable the results of thisanalysis to be extended to other higherbiomass ecosystems.
The data were obtained at the IBP Grassland Biome Pawnee Site on native shortgrassprairie at the Pawnee National Grassland,about 35 miles northeast of Fort Collins,Colorado. Field measurements were madein the Ecosystem Stress Area (ESA) on control, irrigated, and/or nitrogen fertilizedplots.
Prairie vegetation is dominated by variousspecies of grasses. One species, blue grama,comprises about 75 per cent of the dryweight of the gramineous vegetation at thePawnee Site (Uresk, 1971). For this reason,plots of blue grama grass were selected forexperimentation purposes.
In situ measurements of spectral reflectance were obtained with the field spectrometer laboratory designed and constructed for the IBP Grassland Biome Program to test the feasibility of spectro-opticallymeasuring the aboveground plant biomassand plant cover (Miller et al., 1976).
Simulation of sensors of differentwavelengths and bandwidths was accomplished by integration of the narrowbandwidth spectral reflectance data. Subsequent regression analysis determined thedegree of statistical significance, in coefficient of determination (r2 values) terms, between the integrated reflectances and theplant canopy variable in question. The results of the regression analysis identified regions of high sensitivity and provided aquantitative basis for sensor design. The results clearly show that vegetational remotesensing can be improved in the future assuming, as we do, that grass canopy resultsare applicable to other vegetation types.
It should be noted that this research addresses only the target spectral factors affecting sensor selection for the 0.35 - 1.00 /Lmregion. The trade-offs between bandwidth,spatial resolution, signal/noise ratio, and atmospheric transmission were not consideredas they apply to ground, aircraft, and satelliteremote sensing of vegetation.
CANOPY PHYSIOLOGY AND REMOTE SENSING
The physiological factors affecting or actually determining reflectance from leaveshave been the subject of recent research.Knipling (1970) and Woolley (1971) are ex-
cellent references concerning leaf morphology and physiology and the resulting spectral reflectance, transmission, and absorption.Hoffer and Johannsen (1969) have addressedthe question of spectral signature analysisfor vegetation monitoring. Gausman et al.(1973) have investigated biological criteriafor sensor selection and found that the spectral wavelength intervals centered around0.680, 0.850, 1.650, and 2.200 /Lm provideoptimum discrimination of vegetation.
Colwell (1974) investigated grass canopybidirectional spectral reflectance usingcomputer modeling which was qualitativelyvalidated by field experimental measurements. Interested readers are directed to thiswork for an excellent treatment of grasscanopy bidirectional reflectance where different soil backgrounds, leaf area indices,percent vegetation covers, amounts of liveand dead vegetation, solar zenith angles,look angles, and azimuth angles were considered for the wavelengths of 0.55, 0.65, and0.75 /Lm (Colwell 1974).
The majority of basic research to date concerning plant physiology and resulting reflectance has been done with leaf spectrameasured in the laboratory or with aircraft orspacecraft multispectral imagery ofheterogenous vegetation scenes. Little insitu canopy spectral data have been collected, although the US/IBP GrasslandBiome has collected approximately twothousand spectral curves of prairie scenes.
The results ofTucker (1975) and Tucker etal. (1975) report on two of the few groundbased studies of in situ measured canopy reflectance for which detailed information isavailable concerning the biophysical composition of the plant canopy.
Statistical analyses have identified five regions of the spectrum between 0.350 and1.000 /Lm where different biological and/orphysical processes result in different relationships between canopy reflectance andthe amount of vegetation present (Figure 1).The five spectral regions described byTucker (1975) where different relationshipsexist include:
(1) The 0.350 to 0.500 /Lm region(ultraviolet-blue), characterized by strongabsorption by the carotenoids andchlorophylls and strong relations betweenspectral reflectance and the plot variables.
(2) The 0.500 to 0.600 /Lm region (green),characterized by a reduced level of pigmentabsorption and hence higher reflectance (the"green" color of healthy vegetation). Thisspectral region exhibits a weaker relationship between spectral reflectance and the
SENSOR DESIGN FOR MONITORING VEGETATION CANOPIES 1401
:/-':.~"n/.-.""':" '. /'----------,,--- ",
'"-fotol ••ttll_.
~ CII'OfOphW Il4( L.ol_l..
.60 .m <>U ~ ~ 90 tOO
*AV[L[IilIGTH l .. ml
e = Napier's number (i.e., -2.71)f31)' = regression derived estimate
of f31 at wavelength AXi = canopy variable i (see Table
1)
or
where
f30)' = regression derived estimateof f30 at wavelength A
The area of reduced absorptance in thegreen region of the spectrum exhibits somedegree of absorptance, albeit to a lesser extent than adjacent regions. The relationshipbetween reflectance and the plot variablescan be approximated in this spectral regionby either Equation 1 or Equation 2 depending upon the range of plot variable Xi(Tucker, 1975).
The transition from strong pigment absorption in the far red to strong reflection atthe lower end of the photographic infraredresults in weaker relations between spectralreflectance and the plot variables over thisspectral region. Note from Figure 1 that thecoefficient of determination approaches zeroat the midpoint of the interval (0.72 JLm).This effect could be caused by very similarsoil and vegetation reflectance characteristics in this interval, i.e., the vegetation couldhave the same per unit reflectance contribution as the soil background. Thus, therewould be little effective discrimination offunctioning vegetation. Or because the previously mentioned change from strong absorption to enhanced reflection occurs over asmall wavelength interval (-0.040 JLm), thespectrometer may introduce a measurementerror into this region because of slightwavelength calibration errors. A combination of the two hypotheses is the most logicalproposition (Tucker, 1975).
Strong relationships were found betweenspectral reflectance and the plot variables inthe photographic infrared region. The function which best approximated the relationship between spectral reflectance and theplot variables in this region was of twoforms, depending upon the range of the plotvariables in question:
Fit. 1. Spectral coefficient of determination curves for total wet biomass, totalchlorophyll, and leaf water content. otethe similarity among these three curvesresulting from the regressions betweenthe respective plot variables and spectralreflectance for 35 plots sampled in June1972.
plot variables because of the reduced levelof pigment absorption.
(3) The 0.600 to 0.700 JLm region (orangered), characterized by strong chlorophyll absorption and a strong relationship betweenspectral reflectance and the plot variables.
(4) The 0.700 to 0.740 JLm region, a transition region between strong chlorophyll absorption in the far red and enhanced reflectance in the photographic infrared whichexhibits weak relations between spectral reflectance and the plot variables.
(5) The 0.740 to 1.000 JLm photographicinfrared region, exhibiting high or enhancedreflectance and a strong relationship between spectral reflectance and the plot variables. The term "enhanced reflectance" isused to denote the higher levels of photographic infrared reflectance by healthy vegetation, which result from internal leaf scattering mechanisms in the absence of absorption. The canopy geometry also contributesto the enhancement of photographic infraredreflectance by inter-leaf scattering.
Thus the different relationships exist between spectral reflectance and the sampledplot or canopy variables depending upon thedominant physiological and physical processes in the various spectral regions. Regionsof strong absorption can be represented byfunctions of the form
(1)where
RFL). = canopy reflectance at wavelength A
A). = regression derived coefficient at wavelength A *
RFL). = f3o). + f31)' Xi (3)
* Standard regression notation is used afterDraper & Smith (1966).
and
1402 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976
where
s~ = asymptotic reflectance atwavelength A
Asymptotic spectral reflectance is reachedby adding leaf layers or biomass until a stable (i.e., unchanging) reflectance is obtained(see also Gausman et aI., 1976; Tucker,1976).
METHODS AND ANALYSIS
Several thousand curves of grassland vegetation have been collected in the fieldusing the field spectrometer laboratory. Asubset of this data base was selected for thisexperiment. It consisted of the spectralradiances and reflectances of circular 1/4 m2
plots of blue grama measured in an irrigatedarea. Thirty-five plots were measured inJune of 1972 and 40 in September of 1971.
Immediately after the reflectance measurements were completed, the plot wasclipped of all standing vegetation. An aliquotwas extracted for chlorophyll analysis andimmediately qUick-frozen on dry ice (Horwitz 1970). the clipped vegetation was putinto a plastic bag, sealed, and placed in anicebox. When the clipped vegetation fi'omfour or five plots accumulated, it was transported to the Pawnee Site's laboratory building and stored in a refrigerator.
Laboratory technicians then began processing the clipped vegetation. The first determination made was the total wet biomassweight measurement. After this measurement was completed, the clipped vegetationwas transferred to a paper bag and force-airdried at 50°C for 48 hours. Upon completionof the drying cycle, the total dry biomassweight measurement was made. The leafwater content was calculated as simply thedifference between the total wet biomassand total dry biomass. The leaf water contentrepresents the water present in the leaf andstem material.
The total dry biomass was then separatedmechanically with manual finishing intogreen and brown fractions, and weighed(Van Wyk 1972). The chlorophyll contentwas determined for the representative 5 galiquot after Horwitz (1970). This was thenmultiplied by the total wet biomass to yieldthe chlorophyll concentration in mg/m2 units.
The total wet biomass, total dry biomass,leaf water content, dry green biomass, anddry brown biomass were all expressed ing/m2 units (Table 1). Per unit area measurements such as the various biomass determinations, leaf water content, and total
chlorophyll content will be used in describing the research results. Projected areas ofthese canopy variables will not be used although simple relationships exist betweenbiomass and leaf area indices.
INTEGRATION OF NARROW BANDWIDTH
SPECTRAL CURVES
The narrow bandwidth reflectance data(0.005 f.Lm bandpass) were numerically integrated to approximate a variety ofbandwidths. Reflectance values were usedsuch that the results are implicit propertiesof vegetation.
The integration procedure was carried outin the following fashion:
RFLA = (It RFL(I)2 IlA) Y2 (5)
where
RFLA = reflectance in band from 1..(1)to ,\(n)
RFU})2 = square of reflectance at ,\(1)Ill.. = 0.005 f.Lm
This method of integration was used toapproximate the reflectance integral fromthe narrow bandwidth (0.005 f.Lm) in situspectral measurements.
Reflectance is a measure of the relative intensity with which electromagnetic waveswill be reflected from a surface. The squareof reflectance, therefore, is a measure of therelative amount of energy reflected from thesurface. Since the sum of narrow bandwidthreflected energies must equal the energywhich would be measured in a widebandwidth system of equal total bandwidth,Equation 5 should provide a valid estimateof reflectance for the wide bandwidth system. This procedure is consistent with thewell-known relationships for blackbodyradiation and noise intensity, both of whichshow a linear variation of power or energywith bandwidth. This procedure does assume a constant surface impedance and noncoherent energy across the bands being integrated. The surface impedance probably isnot constant, particularly for bandwidthswhich span from strong absorbingwavelengths to highly reflecting wavelengths, but for a first approximationthe results should be reasonably valid. Theassumption of noncoherence is certainlyvalid.
The results obtained with Equation 5 areconsistent when the integrated reflectancesare regressed against the plot variables andcompared to the results of the narrow
SENSOR DESIG FOR MONITORING VEGETATIO CANOPIES 1403
TABLE 1. STATISTICAL SUMMARY OF THE BIOPHYSICAL CHARACTERISTICS OF THE SAMPLE PLOTS.A STATISTICAL DESCRIPTION OF THE VEGETATIVE CANOPY CHARACTERISTICS FOR
(A) THE THIRTy-FIVE 114 M2 SAMPLE PLOTS OF BLUE GRAMA SAMPLED IN JUNE 1972 AND(B) THE FORTY 114 M2 SAMPLE PLOTS OF BLUE GRAMA SAMPLED IN SEPTEMBER 1971.
Standard Coefficient Standard errorSample Range Mean deviation of variation of the mean
A. June 1972Wet total biomass 52.00-1230.40 339.52 316.94 93.35 50.11
(glm2)Dry total biomass 13.04- 528.84 134.07 130.25 97.15 20.59
(glm2)Dry green biomass 12.48- 343.36 105.11 93.46 88.93 14.78
(glm2)Dry brown biomass 00.16- 185.48 28.96 40.23 138.91 6.36
(glm2)Leaf water 38.12- 701.56 205.46 187.83 91.42 29.70
(glm2)Chlorophyll 62.27-2108.06 414.41 515.56 124.41 81.52
(mglm2)
B. September 1971Wet total biomass 70.83- 491.22 261.31 134.40 51.44 21.25
(glm2)Dry total biomass 41.50- 337.84 168.55 90.81 53.88 14.36
(glm2)Dry green biomass 17.12- 185.04 89.38 50.15 56.11 7.93
(glm2)Dry brown biomass 20.40- 186.42 82.41 48.54 58.90 7.68
(glm2)Leaf water 28.03- 190.80 92.75 50.93 54.91 8.05
(glm2)Chlorophyll 53.02- 778.97 319.58 238.73 74.70 37.75
(mglm2)
bandwidth analyses. The integrated regressed values closely approximate the narrowbandwidth results.
The integrations were carried out in twomodes:
(1) Initially the spectral region between0.350 to 0.800 JLm was divided up into 15, 9,5, and finally 3 equal bandwidths. Thespectral reflectances corresponding to eachspectral curve were integrated according toEquation 5. Results were punched ontocards along with the six plot variables forsubsequent regression screening. Regression screening denotes the use of an algorithm to screen various univarite combinations of variables (Frayer et al., 1971).
(2) A particular bandwidth was positionedat a wavelength of interest; and (a) whileholding the lower limit ofthe band constant,the upper limit was reduced by 0.010 JLmsteps. Each diminished bandwidth was integrated in the same manner. (b) Then theupper limit of the band was held constantand the lower limit was increased at 0.010JLm steps until the bandwidth was equal to orgreater than 0.010 JLm. Each interval was in-
tegrated and evaluated in an identical manner to (1) above.
REGRESSION SCREENING
Regression screening was used to evaluatethe relationship between the various integrated reflectances and the plot variables interms of coefficients of determination (r2
values). This was advantageous because thecomputer program used (FSCREEN) calculates the various r2 values and then outputsan ordered list of the r2's (Frayer et aI., 1971).Because of the number of bandwidth intervals simulated, the FSCREEN approach allowed for a substantial savings in computertime.
The models given in Equations 1 and 2were transformed from their original nonlinear form into linear models for regressionscreening purposes. In each case, the integrated reflectances were regressed against thevarious plot variables.
To facilitate comparisons between theJune 1972 and the September 1971 statistics,the plot variables of total wet biomass, leafwater, and total chlorophyll content were
1404 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976
used. Of these sampled variables, leaf waterwas used principally because it is indicativeof the amount of the alive, green, and photosynthetically active biomass present (Tuckeret al., 1975).
EXPERIMENTAL RESULTS
EQUAL BANDWIDTH APPROACH
Coefficients of determination for the integrals for five equal bandwidth intervalswithin the 0.350 and 0.800 J.Lm region werefound to closely resemble the continuousnarrow band r2 plots (Figure 2). The samesimilarity was found when the region wasdivided into 15, 9, and 3 equal bandwidths(Tucker, 1975).
Figure 2 demonstrates the statistical similarity between the narrow bandwidth (0.005J.Lm) regression results and the much widerbandwidth results. In fact, it is obvious thatintegration of the coefficient of determination curves would have produced almostidentical results. The only exception to thisis seen for the fifth band (0.71 to 0.80 J.Lm) onFigure 2 where the area for that band islarger than the area under the curve. Thisimplies that the random error or noise in the0.71 to 0.80 J.Lm band did not appreciably de-
1.00
.80z0
.80~zi:it;;0..0
...::"~0
"
WAVELENGTH (I'm)
FIG. 2. Comparison between the continuous coefficient of determination curve(0.005 /-Lm bandwidth) and the coefficientsof determination resulting from dividingthe 0.350 to 0.800 /-Lm region into five equalbandwidth intervals for the September1971 data. The continuous curve resultedfrom the series of regressions betweenspectral reflectance and total wet biomass atninety-one 0.005 /-Lm intervals. The histograms resulted from the five regressions between integrated reflectance and total wetbiomass. ote how the five equalbandwidth intervals closely approximatethe area under the continuous coefficient ofdetermination curve.
grade the strong infrared relationship between spectral reflectance and total wetbiomass. When the interval was made wider(0.65 to 0.80 J.Lm) and began to include moreof the strong chlorophyll absorption region,as in the case of the three band simulation, amarked reduction in significance occurred.
DECREASI G BANDWIDTH APPROACH
The following bands were selected forevaluation by this method:
(1) 0.37 to 0.55 J.Lm (September dataonly),(2) 0.50 to 0.68 J.Lm (June and Septemberdata), and(3) 0.60 to 0.78 J.Lm (June and Septemberdata).
The spectral estimation of total wetbiomass was investigated in the 0.37 to 0.55J.Lm band. The area of greatest significancewas in the ultraviolet region as noted inTable 2. The strong ultraviolet-blue sensitivity which existed between the various plotvariables and reflectance in this region hasnot previously been reported. The possibility that this region shows a strong sensitivityas the growing season wanes is of particularinterest for the estimation of grass canopybiomass.
The 0.50 to 0.68 J.Lm band includes theorange-red chlorophyll absorption region. Ofparticular interest here were the differencesbetween the June and September data sets.Although the total bandwidth of 0.50 to 0.68J.Lm was the most significant for both datasets (Table 3), the June r 2 value of 0.88 wasfar greater than the September r 2 value of0.56. The lowest June r 2 value, 0.65, washigher than the highest September value.
The spectral interval of 0.60 to 0.78 /-Lm
was considered in tripartite fashion:
(I) 0.60 to 0.72 J.Lm: absorption +noise.
(II) 0.72 to 0.78 J.Lm: noise + enhancedreflectance.
(III) 0.60 to 0.78 J.Lm: absorption +noise + enhanced reflectance.
This was done to evaluate the effect(s) ofpigment absorption, noise, and enhancedspectral reflectance and the combinations ofthese three effects taken together or two at atime.
One would expect a slight correlation degrading of situation (I) and (II) of the abovedue to the noise in the 0.70 to 0.74 J.Lm regionand a serious correlation degrading of situa-
SENSOR DESIG FOR MONITORI G VEGETATIO CA OPIES 1405
TABLE 2. INTEGRATED SIMULATION RESULTS FOR THE 0.37 TO 0.55 !Lm INTERVAL REGRESSED AGAINSTTOTAL WET BIOMASS. NOTE THE GREATER SIGNIFICANCE IN THE ULTRAVIOLET REGION.
(A) 0.37 TO 0.55 !Lm; (B) 0.37 TO 0.55 fLm. (EVERY OTHER LINE DELETED FOR BREVITY).
Rank
(A)13579
111315
Orderedr2's
y = aeb:r
0.690.680.660.650.650.640.610.55
Bandwidth(fLm)
0.370-0.4200.370-0.4300.370-0.4700.370-0.4600.370-0.4500.370-0.5000.370-0.5200.370-0.540
Rank
(B)13579
111315
Orderedr2's
y = aebr
0.520.510.480.460.440.410.350.27
Bandwidth(fLm)
0.370-0.5500.390-0.5500.410-0.5500.430-0.5500.450-0.5500.470-0.5500.490-0.5500.510-0.550
TABLE 3. ORDERED COEFFICIENTS OF DETERMINATION VALUES RESULTING FROM THE REGRESSIONSBETWEEN INTEGRATED REFLECTANCE INTERVALS AND LEAF WATER CONTENT FOR THE 0.50 TO 0.68 fLm
REGION. (A) REPRESENTS THE RESULTS FROM THE THIRTy-FIVE PLOTS SAMPLED IN JUNE 1972 AND(B) REPRESENTS THE FORTY PLOTS SAMPLED IN SEPTEMBER 1971. NOTE THE SIMILAR ORDERINGS
BETWEEN THE COEFFICIENTS OF DETERMINATION AND ASSOCIATED BANDWIDTHS FOR THESE Two SAMPLEPERIODS. (EVERY OTHER LINE DELETED FOR BREVITY).
Integral IntegralOrdered bandwidth Ordered bandwidth
Rank r 2 's (!Lm) Rank r 2 's (!Lm)
(A) y = a + b/x (B) y = aebr
1 0.88 0.500-0.680 1 0.56 0.500-0.6803 0.86 0.500-0.660 3 0.51 0.500-0.6605 0.85 0.500-0.650 5 0.46 0.500-0.6407 0.84 0.500-0.620 7 0.43 0.500-0.6309 0.81 0.500-0.600 9 0.39 0.500-0.550
11 0.76 0.500-0.580 11 0.37 0.500-0.62013 0.70 0.500-0.560 13 0.35 0.500-0.59015 0.65 0.500-0.540 15 0.33 0.500-0.570
ro ~ 00 00 100 lro I~ 100 IOO~
LEAF WATER (g/m2)
ro ~ 00 00 100 lro I~ 100 IOO~
LEAF WATER (g/m2)
3.0
2.8
"' 2.6uz
2.4>'!u"' 2.2...Ju.
"' 2.0'""' 1.8>ii 1.6...J
"'a:14
1.2
1.00
oo
oo 0
0-
•o 0
oo 0
00. . ..
,'= 0.68
o0,
50
4.6
"' 4.2uz
38>'!u"' 34...J.."' 30a:
"' 26>;::<t 22...J
"'a: 1.8
14
100
•oo.- ..\~
o\ ...
o00
o 0
FIG. 3. Relative integrated reflectance plotted against leaf water content for the spectralintervals of (a) 0.37 to 0.50 fLm and (b) 0.60 to 0.70 !Lm (ultraviolet-blue and orange-redabsorption regions respectively). These data from the September sampling period indicatethe inverse relationship between integrated reflectance and the leaf water content for theseregions of strong pigment absorption. Relative reflectance is used because of the unitchanges which resulted from integration by Equation 5.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976
PRIMARY EFFECTS
Three primary effects were evident between 0.350 and 1.000 /Lm and were persistent for both data sets:
(1) Absorption by plant pigments whichoccurs in the 0.350 to 0.500 /Lm and 0.600 to
LEAF WATER (g/m2)
FIG. 4. Relative integrated reflectanceplotted against leaf water content for the~ectral interval of 0.75 to 0.78 /-Lm (photographic infrared enhanced reflectance).These data from the September samplingperiod show the direct relationship between integrated reflectance and the leafwater content.
tion (III) resulting from the confounding interaction of pigment absorptance, noise, andenhanced spectral reflectance.
As the upper limit of o. 78 /Lm remains constant and the lower limit is increased from0.60 /Lm toward 0.75 /Lm in 0.01 /Lm increments, a marked improvement in r 2 valuesoccurred for both the June and Septemberdata sets (Table 4).
0.700 JLm regions (Figure 3). The orange-redand ultraviolet-blue regions of the spectrumshowed similar relationships between integrated ref1ectance and the plot variables.These results identified the greatest sensitivity to be in the 0.37 to 0.46 and 0.67 to0.69 /Lm wavelengths, respectively .
There was a decrease in r 2 values in theorange-red region as the upper limit of integration decreased from 0.68 to 0.54 /Lm (Table3). This is undoubtedly due to weaker absorption by the chlorophylls as the upperlimit of integration moves away from 0.68/Lm. This also indicates that the -0.67 to 0.69/Lm region is where maximum chlorophyllabsorption occurs in the orange-red.
(2) The photographic infrared region (0.74to 1.00 JLm) showed the enhancement ofspectral reflectance by the functioning vegetation present and the strong sensitivitywhich existed between the functioningbiomass and the resulting spectral ref1ectance (Figure 4). Note the decrease in r2 values as the lower limit of integration is decreased (Table 4). This is undoubtedly dueto the inclusion of random effects (noise) inthe 0.70 to 0.74 /Lm region and the absorption in the 0.60 to 0.68 /Lm region.
(3) The 0.70 to 0.74 /Lm portion of thespectrum demonstrated the random effectswhich occur in this transition region (Figure5). The regression results from this spectralregion indicated nonsignificance when integrated reflectances were regressed againstthe plot variables.
In addition to the 0.70 to 0.74 /Lm region ofnonsignificance, the green region (0.50 to0.60 /Lm) was found to not be significant forthe September data set (Figure 5). As the
....... , .... ..-....
r2~ 0.73
..,
20 40 60 80 100 120 140 160 leo 200
1406
5.0
4.7
w 4.4UZ
4.1<1...Uw 3.8--'"-w 3.'a:
w 3.2?...<1 2.'--'wa: 2.6
23
2'°0
TABLE 4. ORDERED COEFFICIENTS OF DETERMINATION VALUES RESULTING FROM THE REGRESSIONS
BETWEEN INTEGRATED REFLECTANCE INTERVALS AND LEAF WATER CONTENT FOR THE 0.60 TO 0.78 /-LmREGION. (A) REPRESENTS THE RESULTS FROM THE THIRTy-FIVE PLOTS SAMPLED IN JUNE 1972 AND
(B) REPRESENTS THE FORTY PLOTS SAMPLED IN SEPTEMBER 1971. NOTE THE SIMILAR ORDERINGS
BETWEEN THE COEFFICIENTS OF DETERMINATION AND ASSOCIATED BANDWIDTHS FOR THESE Two SAMPLE
PERIODS. (EVERY OTHER LINE DELETED FOR BREVITY).
Integral IntegralOrdered bandwidth Ordered bandwidth
Rank r2's (/Lm) Rank r2's (/-Lm)
(A) y = a + bx (B) y = a + bx1 0.82 0.750-0.780 1 0.73 0.750-0.7803 0.78 0.730-0.780 3 0.66 0.730-0.7805 0.70 0.710-0.780 5 0.55 0.710-0.7807 0.58 0.690-0.780 7 0.39 0.690-0.7809 0.46 0.670-0.780 9 0.23 0.670-0.780
11 0.35 0.650-0.780 11 0.12 0.650-0.78013 0.26 0.630-0.780 13 0.05 0.630-0.78015 0.19 0.610-0.780 15 0.01 0.610-0.780
SENSOR DESIGN FOR MONITORING VEGETATION CANOPIES 1407
40
38
36 .. ......34 .32 ..30 •w
> 28;::
".
...J 26.
w0:
24
22
o
..
5.0 b
4.5r'oOOO
wuz" 4.0.... .u ..w
"' ..w .0: 3.5 '10. :., .w . .~
..... 3.0
,"...JW0:
25
200 20 40 60 80 100 120 140 IW 180 200
LEAF WATER (g/m')
2.00 20 40 60 80 100 120 140 160 100 200
LEAF WATER (g/m')
FIG. 5. Relative integrated reflectance plotted against leaf water content for the spectralintervals of (a) 0.50 to 0.60 J-Lm and (b) 0.70 to 0.74 J-Lm transition region. Note the lack ofanyapparent relationship between integrated reflectance and leaf water content for thiswavelength interval for the September sampling period.
growing season wanes, the significance between integrated reflectance in the 0.50 to0.60 ILm region decreases.
COMBINATION OF TWO PRIMARY EFFECTS
The inclusion of the 0.50 to 0.60 ILm and0.70 to 0.74ILm regions (green and transitionregions, respectively) of reduced significance with wavelengths showing strongpigment absorption (the 0.350 to 0.500 ILmand 0.600 to 0.700 ILm regions) demonstratedconclusively (Figure 6) the degrading effectof these regions. Furthermore, the integralsevaluated in Table 4 show the reduction in r
values which occur as proportionally more ofthe 0.70 to 0.74 ILm region is included withthe 0.74 to 0.78 ILm spectral interval of enhanced reflectance. This is also evidentwhen comparing Figures 4 and 7.
COMBINATION OF ALL THREE PRIMARY EFFECTS
Sensor bandwidths were also evaluatedwhere the integrated reflectances encompassed areas of strong pigment absorption,lessened spectral sensitivity, and enhancednear infrared reflectance (0.60 - 0.78ILm )(Table 4) (Figure 8). Results of these evaluations demonstrated conclusively that any
. . .
r Z =0.48
...............
:.:..
8.0
7.5
w 7.0uz 6.5"....uw 6.0...Ju..w 5.50:
w 50>~.. 4.5...Jw0: 4.0
3.5
3.00 ~ 40 60 00 100 I~ 140 160 100~
LEAF WATER (g/m')
FIG. 7. Relative integrated reflectanceplotted against leaf water content for thespectral interval of 0.70 to 0.78 J-Lm. Notethe degrading effect indicated by this plotwhen one compares this figure to Figure 4.The added variability is introduced by theinclusion of the 0.70 to 0.74 J-Lm region intothis interval. These data are from the September sampling period.
,
r'. 0.43
.....
.. :.
... .:.•
7.0
6.6
w 6.2uz" 5.8....uw 5.4...Ju..w 5.00:
w 4.6~...... 4.2...Jw0: 3.8
3.4
3.00 ~ ~ 60 00 100 I~ I~ 160 100~
LEAF WATER (g/m')
FIG. 6. Relative integrated reflectanceplotted against leaf water content for thespectral interval of 0.60 to 0.74 J-Lm. otethe degrading effect upon the relationshipbetween integrated reflectance and leafwater content when the 0.70 to 0.74 J-Lm region of reduced significance is included withthe 0.60 to 0.70 J-Lm region of strong absorption (see Figure 3). These data are from theSeptember sampling period.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976
INFLUENCE OF TIME OF SAMPLING DURING THE
GROWING SEASON
The phenological influence can be inferred from comparisons between early in the
sensor bandwidth where absorption and enhanced reflectance both occur was extremely insensitive for spectrally estimatinggrass canopy vegetational status. Sensorssuch as this should most certainly be avoided.
growing season data (June) and late in thegrowing season data (September). The Junedata showed a higher degree of significance,and sensor wavelength and bandwidthcharacteristics were not crucial. The September data, however, indicated that as thegrowing season progresses and the transferor conversion of live to dead standing vegetation occurs, a lower level of significanceexisted. Furthermore, sensor wavelength andbandwidth criteria were extremely important.
The September data represented a morecomplex interaction between incident spectral irradiance and the grass canopy becauseof the presence of more dead vegetation.The results of the more complex canopystate represented by the September datawere therefore the most useful to developoptimum sensor wavelength and bandwidthcriteria for remote sensing sensors.
EVALUATION OF LA DSAT CHANNELS
The seven LANDSAT sensors wereevaluated by simulating the bandwidths ofthe various sensors. The utility of the sevenLANDSAT Return Beam Vidicon (RBV) andMulti Spectral Scanner (MSS) channelsranged from excellent to poor in terms of optimal channel bandwidths for monitoringgrass canopy vegetational status (Table 5).MSS bands 5 (0.60 to 0.70 JLm) and 7 (0.80 tol.l0JLm) and RBV band 2 (0.58 to 0.68 JLm)are well situated to sense blue grama canopyradiances, which in turn were highly relatedbiologically to the canopy vegetationalstatus.
RBV band 1 encompasses the 0.50 to 0.56
r2 • 0.01
.. ...... .... ..... .. . .. ...20 40 60 80 100 120 140 160 180 200
LEAF WATER (g/m2 )
FIG. 8. Relative integrated reflectanceplotted against leaf water content for thespectral interval of 0.60 to 0.78 /Lm. otethe confounding effect which results whena region of absorption, a region of reducedsignificance, and a region of enhanced reflectance are combined into the same spectral interval. Refer to figures 3, 4, and 5b forthe three primary effects included in thisfigure. These data are from the Septembersampling period.
1408
9.0
8.6
w 8.2uz
7.8<l>-Uw 7.4...Ju.w 7.0a:
w 6.6>;:::
6.2<l...Jw
5.8a:
5.4
5q)
TABLE 5. RBV AND MSS COEFFICIENTS OF DETERMINATION RESULTING BETWEEN THE REGRESSIONS OFINTEGRATED REFLECTANCE AND THREE PLOT VARIABLES FOR Two SAMPLING PERIODS. THE JUNE DATA
INCLUDED THIRTy-FIVE PLOTS AND WAS MEASURED FOR THE 0.50 TO 1.00 /Lm REGION; THE SEPTEMBERDATA INCLUDED FORTY PLOTS MEASURED FOR THE 0.35 TO 0.80 /Lm REGION.
Highest r 2 values
June (n = 35) September (n = 40)
Total wet Total wetBandwidth biomass Leaf water Chlorophyll biomass Leaf water Chlorophyll
Channel (/Lm) (g/m2) (g/m2 ) (mg/m2 ) (g/m2) (g/m2 ) (mg/m2)
RBV-l 0.475-0.575 0.72 0.76 0.77 0.32 0.42 0.25RBV-2 0.580-0.680 0.88 0.91 0.91 0.38 0.62 0.32RBV-3 0.690-0.800 0.68 0.68 0.68 0.49 0.44 0.41RBV-4 0.500-0.600 0.77 0.81 0.82 0.26 0.37 0.21MSS-5 0.600-0.700 0.88 0.91 0.91 0.39 0.65 0.33MSS-6 0.700-0.800 0.72 0.72 0.71 0.54 0.51 0.45MSS-7 0.800-1.100* 0.70 0.71 0.75
• The September data was not collected over this region. The June data was, but suffers from a progressively lower signal to noise ratiobeyond -0.90 /LlTI and thus is seriously degraded beyond -0.95/LlTI.
SENSOR DESIGN FOR MONITORING VEGETATION CANOPIES 1409
ILm region of lessened significance and doesnot include enough of the blue region to beeffective as the growing season wanes. RBVband 3 includes the 0.69 to 0.70 ILm region ofabsorption by chlorophyll, the 0.7C to 0.74ILm region of noise, and the 0.74 to 0.80 ILmregion of enhanced reflectance. Thus RBVband 3 is degraded by some pigment absorption at its lower wavelengths and by thenoise present in the 0.70 to 0.74 ILm region.
MSS band 6 is redundant to band 7 andincludes the noisy 0.70 to 0.74 ILm region.Bands 5 and 7 are well situated for vegetational remote sensing of the blue gramacanopy studied. Band 4, however, is placedover the 0.50 to 0.60 ILm region of lessenedstatistical significance and what physiological information is sensed by this band is better sensed by band 5.
It should be noted that MSS 6 has beenshown to be of greater utility than MSS 7 formonitoring rangelands by Rouse et al.(1974). This has been corroborated byJohnston (1976) for LANDSAT imagery ofthe Pawnee National Grassland, Colorado.In any event, MSS 6 does include the 0.70 to0.74 ILm transition region of noise and theinformation that is highly correlated to vegetational density or biomass for this sensor isreceived in the 0.74 to 0.80 ILm region.
We thus feel that MSS 6 and MSS 7 arehighly redundant in that the same vegetational information is equally as well sensedover the 0.74 - 1.30 ILm. However, sensordetector sensitivities and atmospheric effects make direct comparisons between MSS6 and MSS 7 difficult to draw.
SUMMARY A 0 CONCLUSIO s
(1) Simulation-integrations for varioussensor wavelength and bandwidths for twosample periods, one early and one late in thegrowing season, indicated that greater spectral sensitivity existed earlier in the growingseason between reflectance and the grasscanopy variables.
(2) The spectral regions of 0.37 - 0.50,0.63 - 0.69, and 0.75 - 0.80 ILm were foundto be statistically significant, in a regressionsense, both early and late in the growingseason.
(3) Inclusion of noisy bands (green andtransition region) degrade sensor results andshould be avoided to optimize vegetationalremote sensing.
(4) Inclusion of a bandwidth whereorange-red absorption and enhanced reflectance both occur result in a near total degrading of any spectral sensitivity and should beavoided.
(5) Sensor location and bandwidth criteriacan be accurately approximated by integration of spectral coefficient of determinationcurves.
(6) LA DSAT MSS bands 5 and 7 andRBV band 2 are well situated for biologicalremote sensing.
(7) MSS band 6 is redundant to band 7(biologically speaking) and is degraded bynoise from the 0.70 to 0.74ILm region. However, actual MSS 7 performance may be degraded by sensor characteristics and/or atmospheric effects.
(8) RBV band 3 includes the 0.69 to 0.70ILm region of absorption, the 0.70 to 0.74ILmtransition region of noise, and the 0.74 to0.80 ILm region of enhanced spectral reflectance and should be moved to a longerwavelength interval (say 0.75 to 0.85ILm) forbetter results.
(9) Remote sensing missions concernedwith vegetation monitoring could increasethe potential information content by including these criteria in system design.
ACK OWLEDGMENTS
We would like to thank Marvin Bauer,Harold Gausman, and Jerry Richardson fortheir assistance in reviewing and commenting upon the manuscript.
This paper reports on work supported inpart by National Science Foundation GrantsGB-31862X2, GB-41233X, and BMS7302027 A02 to the Grassland Biome, U. S. International Biological Program, for "Analysisof Structure, Function, and Utilization ofGrassland Ecosystems."
REFERENCES
Colwell, J. E. 1974. Grass canopy bidirectionalspectral reflectance. In Proceedings of the 9thInternational Symposium on Remote Sensingof Environment. ERIM, Univ. Michigan, AnnArbor, pp. 1061-1085.
Draper, N. R., and H. Smith. 1966. Applied Regression Analysis. John Wiley and Sons, NewYork. 417 pp.
Frayer, W. E., R. W. Wilson, and G. M. Furnival,1971, FSCREEN: A computer program forscreening all combinations of independentvariables in univariate multiple linea?' regressions, Dept. Forestry and Wood Science,Colorado State Univ., Fort Collins. 23 pp.
Gausman, H. W., W. A. Allen, R. Cardenas, andA. J. Richardson, 1973, "Reflectance discrimination of cotton and corn at four growthstages," Agron.]. 65:194-198.
Gausman, H. W., R. R. Rodriquez, and A. J.Richardson, 1976, "Infinite reflectance ofdead compared to live vegetation," Agron. ].(in press).
Hoffer, R. M., and C. J. Johannsen, 1969, "Ecolog-
1410 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976
ical potentials and spectral analysis," In P. L.Johnson (ed.) Remote sensing and ecology.Univ. Georgia Press, Athens. pp. 1-16.
Horwitz, W. (ed.), 1970, Official methods ofanalysis, 11th Ed. Association of AnalyticalChemists, Washington, D.C. pp. 53-55.
Johnston, G. R., 1976, Remote estimation of herbaceous biomass, M.S. thesis, Colorado StateUniv., Fort Collins. 120 pp.
Knipling, E. B., 1970, "Physical and physiologicalbasis for the reflectance of visible and nearinfrared radiation from vegetation," RemoteSensing Environment 1(3): 155-159.
Miller, L. D., R. L. Pearson, and C. J. Tucker,1976, "Design of a mobile field spectrometerlaboratory," Photogram. Eng. and RemoteSensing 42(4):569-572.
Rouse, J. W., R. H. Haas, J. A. Schell, D. W. Deering, and J. C. Harlan, 1974, "Monitoring thevernal advancement and retrogradation(green-wave effect) of natural vegetation,"NASA Type III Final Report, GSFC. 347 pp.
Tucker, C. J., 1975, Spectral estimation of grass
canopy vegetational status, Ph.D. dissertation, Colorado State Univ., Fort Collins. 106pp.
Tucker, C. J., L. D. Miller, and R. L. Pearson,1975, "Shortgrass prairie spectral measurements," Photogram. Eng. Remote Sensing41(9):1157-1162.
Tucker, C. J. 1976. Asymptotic nature of grasscanopy spectral reflectance. Applied Optics(in press).
Uresk, D. M., 1971, Dynamics of blue gramawithin a shortgrass ecosystem, Ph.D. dissertation, Colorado State Univ., Fort Collins. 52pp.
Van Wyk, J. J. P., 1972, "A preliminary report onnew separation techniques for live-deadaboveground grass herbage and roots from drysoil cores," USIIBP Grassland Biome Tech.Rep. No. 144, Colorado State Univ., Fort Collins. 16 pp.
Woolley, J. T., 1971, "Reflectance and transmittance of light by leaves," Plant Physiol.47:656-662.
New Sustaining Member
Col-East, Inc.
COL-EAST, INC., had its beginning in 1946and has literally grown up with the
aerial mapping and photogrammetric industry. It became involved with the preparationof topo mapping via photogrammetric meansin 1956 and so has completed twenty years ofthis particular type of work.
Col-East, Inc., has its main office at NorthAdams, Massachusetts with a small operating unit located in Norwood, Massachusetts.It is a wholly contained operation, operatingthree aircraft, utilizing a modern Zeiss camera, and a complete ground control crewusing EDM's and other precision equipment. Col-East does in-house semi-
analytical bridging and much of its compilation is turned out on a Kern PG-2/AT-12 plotter. The final drafting is usually presented tothe client on mylar with ink.
Col-East's clientele includes some of thelarger engineering firms on the East Coast,and by maintaining good quality it has obtained repeat contracts from these companies.
Col-East reportedly is in a position to provide assessors' plans, complete through allphases, or provide base map and photo packages for the use of others such as surveyors,engineers, or town engineering departmentsin constructing property maps.
ASP Needs Old MagazinesBecause of an unexpected demand for journals and student requests, the supply of some
back issues of PHOTOGRAMMETRIC ENGINEERING has been depleted. Consequently, until further notice, National Headquarters will pay to the Regions-or to individual members$1.00 for each usable copy of the following issues sent to Headquarters, 105 N. VirginiaAve., Falls Church, Va. 22046:
YearFebruary. 1975-Vol. XLIMarch 1975-Vol. XLI
NumbersNo.2No.3