COMPTON J. TUCKER*LEE D. MILLER
NASA/Goddard Space Flight CenterGreenbelt, MD 20771
Soil Spectra Contributions toGrass Canopy SpectralReflectanceThe contribution of a soil-litter background to the compositegrass canopy spectra can, in some grassland cases, beextracted and quantified.
INTRODUCTION
T HE CONTRIBUTION of the underlying soilspectra to the composite canopy spectral
reflectance of vegetated surfaces has recently been addressed (Colwell, 1974; Driscoll and Spencer, 1972; Richardson et al.,1975; Wiegand et al., 1974). Remotelysensed data of vegetated surfaces could beanalyzed more accurately if the contributionof the underlying soil spectra were knownand thus possibly could be removed from thecanopy spectral radiance to yield more information about the vegetation. It would
spectra-optical techniques for measuringaboveground standing crop biomass of grasslands. Early in the study it became apparentthat the soil or background spectra dominated low biomass grass canopy spectralradiance or reflectance. It was thereforenecessary to examine the contribution of thesoil spectra to the composite grass canopyspectral reflectance as a function of biomassand wavelength.
BACKGROUND
Let us assume that the soil background(realizing that the "soil" background is actu-
ABSTRACT: The soil or background spectra contribution to grasscanopy spectral reflectance for the 0.35 to 0.80 fJ-m region was investigated using in situ collected spectral reflectance data. Regressionanalysis was used to estimate accurately the unexposed soil spectralreflectance and to quantify maxima and minima for soil-green vegetation reflection contrasts.
generally be impractical to take detailed soilspectra measurements of the area in question because the vegetation canopy obscuresthe soil surface and the time involved forthese in situ measurements is usually prohibitive.
The work reported in this article was partof a larger effort sponsored by the u.S. International Biological Program's GrasslandBiome. The data were collected with the objective of evaluating non-destructive
"National Academy of Sciences-NationalAcademy of Engineering Research Fellow.
ally a soil-litter background) has a characteristic spectra for a given particular area(Figure 1). The soil spectra for the fieldstudy site was a monotonically increasingfunction with wavelength over the spectralrange of 0.35 to 1.00 fJ-m. The dry soil surfacewas more highly reflective than the wet soilsurface. Dry refers only to the uppermostlayer as determined by visual inspection.
The plant canopy on the grassland soil surface will be viewed as some statistical ensemble of foliage elements superimposedover the soil-litter background. The densityof the ensemble offoliage elements will be a
PHOTOGRAMMETRIC ENGINEERING AND REMOTE SENSING,
Vol. 43, No.6, June 1977, pp. 721-726.721
722 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1977
function of biomass. The incoming spectralirradiance will interact with the grass canopyand, depending upon the vegetational density or biomass, can also interact with thesoil background. The interaction(s) with thesoil background become less and less as thevegetational density or biomass increasesuntil the asymptotic spectral radiance or reflectance is reached (Tucker, 1977). Increases in the vegetational density orbiomass effect no change in the canopyspectra when the asymptotic spectralradiance or reflectance has been reached.This can be explained because the canopy isof sufficient density and thickness to preventthe penetration of the incident spectral irradiance to lower biomass levels of thecanopy. Hence, the incident spectral irradiance does not interact with additional(and lower level) biomass. As the vegetational density increases to the point wherethe spectral reflectance begins to asymptoteat a given wavelength, the soil spectra contribution to the canopy spectra is minimal atthat wavelength. When the canopy is of sufficient density or biomass to result in theasymptotic spectral reflectance, there is nosoil spectral reflectance contribution to thecomposite canopy spectral reflectance. Thusthe relative contribution of the soil spectra tothe composite canopy spectra is inversely related to the biomass or vegetation density.
The asymptotic spectra for green grasscanopies were quite different from those forthe soil surface at the study site (Figure 1).As plant growth and development result inincreasing amounts of green plant materialabove the dry soil surface, the canopyspectra changes. In regions of the spectrumwhere absorption occurs, the compositecanopy spectra decreases and approachesthe asymptotic green reflectance spectra. In
OO.r-----------------,...
GREEN VEGETATION PLOT 20074(TOTAL DRY BIOMASS =530 g/m2)
\-----_...--------_......--_.--_._-----
~.36 OAO 0.45 0.50 0.56 O.eo 0.86 0.70 0.75 0.80 0.85 0.90 0.95 1.00
WAVELENGTH hIm)
FIG. 1. Spectral reflectances for dry soil, wet soil,and the asymptotic green reflectance. The dry soiland wet soil are for five bare soil plots measuredwhen dry and wet, respectively. The asymptoticgreen reflectance curve is from a plot of bluegrama grass having a total dry biomass of530 g/m2 .
the near infrared region of -0.71 to 0.74 J.Lm,the composite canopy spectra does notchange appreciably. In spectral regionswhere minimal or no absorption occurs, suchas the -0.74 to 1.20 J.Lm region, the composite canopy spectra increases and approaches the asymptotic green reflectancespectra. Discrimination of vegetationbiomass, for example, is strongly dependentupon the soil surface-vegetation spectral reflectance or radiance contrast. For thisreason, some wavelengths are far superior toothers for discrimination of green vegetationbiomass (Tucker and Maxwell, 1976).
The effectiveness of some wavelengthsdecreases while that of others increaseswhen the soil surface is wet (Figure 1). Thesoil-vegetation reflectance contrast decreases in the red and blue regions while itincreases in the photographic infrared region. This has also been reported by Colwell(1974).
Theoretical considerations indicate thatthe soil spectra can be extracted by regressing canopy spectral reflectance againstsome measured biophysical characteristic ofthe canopy such as total biomass, greenbiomass, brown biomass, chlorophyll, andleaf water concentration. The simple modelused for this extraction was a general linearregression model of the form:
Reflectance). = !3o). + !31)' (plot characteristic)+ Error). (1)
Note that the Reflectance)., !3o)., !31)., and Error). are all functions of wavelength.
METHODS AND ANALYSIS
STUDY LOCATION
The experimental results reported hereinwere obtained on native shortgrass prairie atthe IBP Grassland Biome Pawnee Site, thefield research facility of the Natural Resource Ecology Laboratory, Colorado StateUniversity, located on the USDA Agricultural Research Service Central Plains Experimental Range about 35 miles northeastof Fort Collins, Colorado. Field measurements were made in the Ecosystem StressArea (ESA) on control, irrigated, and/or nitrogen fertilized plots.
Prairie vegetation is dominated by variousspecies of grasses. One species, blue grama(Bouteloua gracilis (H.B.K.) Lag.), comprises about 75 percent of the dry weight ofthe gramineous vegetation at the PawneeSite (Uresk, 1971). For this reason, plots ofblue grama grass were selected for experimentation purposes.
SOIL SPECTRA CONTRIBUTIONS 723
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 above ground plant biomassand plant cover (Miller et aI., 1976).
DATA USED
Forty III m2 plots of blue grama grass weresampled in situ by spectroradiometric measurement over the 0.350 to 0.800 Mm regionat every 0.005 Mm interval with the mobilefield spectrometer laboratory. All measurements were made normal to the ground surface and were sampled in early September,1971.
Immediately after the reflectance measurements were completed, each plot wasclipped of all standing vegetation. The clipped vegetation was placed in a plastic bag,qUick-frozen on dry ice, and subsequentlytaken into the laboratory for weighing,chlorophyll extractions, and drying.Laboratory determinations were made fortotal wet biomass, total dry biomass, leafwater content, dry green biomass, drybrown biomass, and total chlorophyll content(Table 1).
REGRESSION ANALYSIS
A regression approach was undertaken toapproximate the relationship(s) existing between the six sampled canopy variables andthe spectral reflectance at each 0.005 Mm interval. Standard regression notation afterDraper and Smith (1966) will be used anddenoted as a function of wavelength by thesubscript A.
Statistically, both variables were sampled
with error. However, the reflectance variance for all 40 replications was approximately the same. The reflectance measurements were considered as 91 independentregressions over the 0.350 to 0.800 Mm spectral region because the reflectance was measured at a fixed wavelength. Because of thestatistical model used, reflectance could betreated as the dependent variable eventhough it was sampled with error for a fixedwavelength. The biomass and spectral reflectance measurements were considered asa bivariate normal distribution of the form
and, for fixed X = x,
whereYx = measured spectral reflectance;X =measured biomass, chlorophyll, leaf
water, etc.; and(T~, = measurement error associated with
spectral reflectance.Although the regression model used was a
simple linear regression model, the interaction of solar irradiance with a plant canopy isvery definitely nonlinear. However, thefunctional relationship between spectral reflectance and the biophysical plot variablessampled for the experimental plots in question could be accurately approximated by alinear model. If the range of total wetbiomass values for the experimental plotswould have exceeded 500 gim 2 , then a nonlinear model would have been necessary.
TABLE 1. STATISTICAL SUMMARY OF THE BIOPHYSICAL CHARACTERISTICS OF THE SAMPLE PLOTS.
A STATISTICAL DESCRIPTION OF THE VEGETATIVE CANOPY CHARACTERISTICS OF THE 40 1/4 m 2 SAMPLE
PLOTS OF BLUE GRAMA SAMPLED IN EARLY SEPTEMBER, 1971.
Coef. ofSample Range Mean SD variation SE
Wet total 70.83- 261.31 134.40 51.44 21.25biomass (g/m2 ) 491.22Dry total 41.50- 168.55 90.81 53.88 14.36biomass (g/m2 ) 337.84Dry green 17.12- 89.38 50.15 56.11 7.93biomass (g/m2 ) 185.04Dry brown 20.40- 82.41 48.54 58.90 7.68biomass (g/m2 ) 186.42Leaf water 28.03- 92.75 50.93 54.91 8.05(g/m2 ) 190.80Chlorophyll 53.02- 319.58 238.73 74.70 37.75(mg/m2 ) 778.97
724 PHOTOGRAMMETRIC ENGINEERING & REMOTE SE SING, 1977
WAVELENGTH h.m)
FIG. 2. F value curve resulting from the simplelinear regression model analysis of variance forthe regression between reflectance and dry greenbiomass for each of the 91 wavelength intervalsbetween 0.35 and 0.80 p.m. The horizontal (-5-)line represents the 0.5 percent level of significance for 1 and 38 df.
The regression model in Equation 1 wasestimated at the 91 0.005 /Lm intervals between 0.350 to 0.800 /Lm using the 40 replicated measurements of spectral reflectanceand the canopy variables. The simple correlation coefficient (r), the coefficients of determination (r2 ), the resulting regressionmodel analysis of variance F values (F) (Figure 2), and the final regression equations ateach of the 91 0.005 /Lm wavelengths werecalculated (Table 2). These statistics, expressed as functions of wavelength, definedthe relative sensitivity on a spectral basis between the various biophysical variables ofthe sample plot and the plot's spectral reflectance.
The regression equation intercept (/3ox in
Equation 1) was considered as the contribution of the soil surface to the compositecanopy spectral reflectance. Restating Equation 1 we have
CANOPY RFLx = SOIL RFLx + /31X(BIOMASS (g/m2)) (4)
The terms "CANOPY RFL" and "SOILRFL" were percentages, the biomass (in thisexample) was in glm2, and /3/A has units ofpercent/(glm2) or the weighting coefficient ofthe respective biophysical plot variable.
The regression model simplifies when thebiophysical plot variable was zero to the expression:
CANOPY REFLECTANCEx= SOIL REFLECTANCEx (5)
The relative contribution of the underlyingand somewhat obscured soil surface was related to the composite canopy spectral reflectance and plot variable by the generallinear regression model with a minimum ofassumptions.
These assumptions included the following:
(1) The biomass range of the sampledgramineous canopies varied from 0.0 glm2
to 500 g/m2 (Table 2),(2) A linear relationship existed between
canopy spectral reflectance and the respective plot variable, and
(3) The canopy spectral reflectance was measured in situ.
RESULTS
The estimated regression equation intercepts plotted as a function of wavelengthclosely resemble a soil spectra curve (Figure3). This curve was considered as the result-
TABLE 2. TABULAR RESULTS FOR THE SIMPLE LINEAR REGRESSION BETWEEN REFLECTANCE AT0.775 p.m AND THE DRY GREEN BIOMASS CLIPPED FROM 40 1/4 m2 IN SITU SAMPLE PLOTS OF
BLUE GRAMA.
Wavelength = 0.775 p.m:Correlation coefficient (r)Coefficient of determination (r2 )
Standard error of the estimate
Regression equation: Estimated reflectance (0/0) = 16.5695 + 0.0527* (Green Biomass-glm2 )
= 0.8493= 0.7213= 1.6657
SourceRegressionResidualTotal
df1
3839
Analysis of variance tableSum of Meansquares squares
272.8683 272.8683105.4277 2.7744378.2959
F ratio98.3517
P(F>comp F)0.0000
SOIL SPECTRA CONTRIBUTIONS 72550,.---------- --,
45
40
35
interesting to note that the minimum valuesfor /3110. lie in the 0.67 to 0.69 J.Lm region ofstrong in vivo chlorophyll absorption whilemaximum values for /3110. lie in the 0.75 to 0.80
30
2550,.--------------,
-O·~.3S':---,-OL40---.,.O....,..--,O..L.50,--O:-'.56=--.,-'O....,.--,JO...'---,-O.L70---.,.O":::.75----:-'O..
WAVELENGTH (pm)
-0.05
30
15
40
35
~
zw~ 25
20
-0.04
-0,03
...~ 0.01U~ o.oo,I---~-------__t--.,
8 -0.01
X -0.02
0.02
0."
0."
0.05
0.06,.-----------__.,,--
----. Plusond m,nul(±)on. sTonClOf'd a,,,lolion45 fur ~ '-Ofl ,~",o (urns
--s-- M.a" ...I1~I... (;U'...s
---1-- I""1'C1!P15 ploUe<! ..., WO....'-"9IFl
0.45 0.50 0.55 0.60 0.65 0:70 0.75 080
WAVELENGTH ("m)
FIG. 4. Comparison among the mean of six computed soil spectral reflectances and the mean ofsoil spectral reflectance measured in the samearea. -S- represents the mean spectral reflectanceof 25 soil curves. - represent plus and minus onestandard deviation of the mean. The + curve represents the mean of the six regression interceptcurves. Note the close similarity between the twomean curves from 0.350 JLm to 0.800 JLm. The meansoil curve was formed by averaging 25 soil spectralreflectance curves for five bare soil plots with fiverepetitive curves per plot. The intercept meancurve was formed from the individual interceptcurves (Figure 3) resulting from the simple linearregression between spectral reflectance and theplot variables for the 40 plots of blue grama grass.
FIG. 5. f3 .. values plotted as a function ofwavelength for the 910.005 JLm intervals between0.350 and 0.800 JLm. The f31A values were derivedfrom the series of regressions between spectral reflectance and the dry green biomass for the 40plots sampled. Maxima and minima values for f31Arepresent the greatest SOil-dry green biomass reflectance contrast(s).
040 Q45 050 055 060 0.65 0.70 075 0.80
WAVELENGTH (/1m)
ing series of soil spectral reflectances fromEquation 5.
The spectral intercepts (i.e., /30/0.) curvesutilizing each of the six different availablebiophysical measures of the vegetationcanopy variables were very similar, as in allcases the underlying soil surface was in factthe same and should have the same spectralreflectance (Figure 3).
Regardless of the biophysical measure ofthe canopy used, the regression interceptsclosely resembled those of the other five intercept curves. All six of these curves in turnclosely resembled the independently measured spectral reflectance of bare soil measured in the same area (Figure 4).
It was interesting to note that the soilspectra estimate was reasonably accurate inregions of the spectrum where there was alack of regression significance betweencanopy spectral reflectance and the respective plot variable(s). The coefficient of determination (r2 ) values and F ratios (Figure2) resulting from the regression betweencanopy spectral reflectance and dry greenbiomass indicated a lack of statistical significance in the 0.530 to 0.600 J.Lm and 0.700 to0.740 J.Lm regions of the spectrum. However,the estimate of the soil spectra in these sameregions was as accurate as in other areas ofthe spectral interval studies where strongsignificance existed. The accuracy of the soilspectra estimate was apparently not dependent upon regression significance.
The plot of /3, /0. as a function of wavelengthindicates the departure from the soil spectrafor green vegetation (Figure 5). Maxima andminima correspond to those wavelengthswith the greatest soil-green vegetation contrast for the soil present at the study site. It is
20
FIG. 3. Comparison between the spectral intercept curves for the six independently sampledcanopy variables. Note the very close similaritybetween the six curves.
15
726 PHOTOGRAMMETRIC ENGI EERING & REMOTE SENSING, 1977
j.l.m region of enhanced photographic infrared reflectance.
An interesting extrapolation of Figures 1and 5 may explain some of the findings ofRouse et al. (1974) who reported greatergrassland biomass mapping utility usingLandsat MSS6 (0.70 to 0.80 j.l.m) than MSS7(0.80 to 1.10 j.l.m). If we assume that the soilspectra data presented in Figure 1 applies tothe American Great Plains test areas used byRouse et al. (1974), it is apparent that greatersoil-green vegetation contrast occurs inMSS6 (0.70 to 0.80 j.l.m) than MSS7 (0.80 to1.10 j.l.m). This could explain the greater utility ofMSS6 vs. MSS7 for Rouse et al.'s (1974)study areas.
SUMMARY
(1) The contribution of a soil-litterbackground to the composite grasscanopy spectra can, in some grasslandcases, be extracted and quantified.
(2) Applied regression analysis was foundto accurately estimate the soil spectralreflectance and to quantify thewavelengths of maximum soil-greenvegetation reflectance contrast.
(3) The near infrared soil-green vegetation reflectance contrasts may explainwhy Landsat MSS6 has been found insome situations, superior to MSS7 forbiomass monitoring.
ACKNOWLEDGMENTS
I wish to thank Mike Garratt and Lee Miller for suggestions on the manuscript.
This paper repOlts on work supported inpart by National Science Foundation GrantsGB-7824, GB-31862X, GB-31862X2, GB41233X, and BMS73-02027 A02 to the Grassland Biome, U.S. International BiologicalProgram, for "Analysis of Structure, Function, and Utilization of Grassland Ecosystems." This is a drastically revised version ofa paper presented at the 3rd Annual Remote
Sensing of Earth Resources Conference(1974).
REFERENCES
Colwell, J. E. 1974. Vegetation canopy reflectance. Remote Sensing of Environment 3:175-183.
Draper, . R., and H. Smith. 1966. Applied regression analysis. John Wiley and Sons, NewYork, N.Y. 467 p.
Driscoll, R. S., and M. M. Spencer. 1972. Multispectral scanner imagery for plant canopycommunity classification. Proc. Eight Int.Symp. on Remote Sensing of Environ. 11:1259-1278.
Miller, L. D., R. L. Pearson, and C. J. Tucker.1976. Design of a mobile field spectrometerlabor:ltory. Photogramm. Eng. Remote Sensing 42(4): 569-572.
Richardson, A. J., C. L. Wiegand, H. W. Gausman,J. A. Cuellar, and A. H. Gerbermann. 1975.Plant, soil, and shadow reflectance components of row crops. Photo. Eng. and RemoteSensing 41(11): 1401-1407.
Rouse, J. W., R. H. Hass, J. A. Schell, D. W. Deering, and J. C. Harlan. 1974. Monitoring thevernal advancement and retrogradation(greenwave effect) of natuml vegetation.Type 111 Final Report, Goddard Space FlightCenter, Greenbelt, Md. 342 p.
Tucker, C. J. 1977. Asymptotic nature of grasscanopy spectral reflectance. Appl. Optics16(5): 1151-1157.
Tucker, C. J., and E. L. Maxwell. 1976. "Sensordesign for monitoring vegetation canopies".Photogramm. Eng. and Remote Sensing41(11): 1399-1410.
Uresk, D. M. 1971. Dynamics of blue gramawithin a shortgrass ecosystem. Ph.D. Diss.Colorado State Univ., Fort Collings. 52 p.
Wiegand, C. L., H. W. Gausman, J. A. Cuellar, A.H. Gerbermann, and A. J. Richardson. 1974.Vegetation density as deduced from ERTS-lMSS response. Third ERTS Symposium,NASA SP-351. Vol. I, Sect. A, U.S. Government Printing Office, Washington, D.C. 93116.
Workshop Proceedings AvailableAt no cost, a limited number of copies of the Proceedings for the U. S. Army Engineer
Topographic Laboratories/American Society of Photogrammetry Workshop for Environmental Applications of Multispectral Imagery, 11-13 November 1975, are available on request.Those desiring copies of these Proceedings should write to the Commander and Director,U. S. Army Engineer Topographic Laboratories, ATTN: ETL-LO, Fort Belvoir, VA 22060.