Remote Sensing

Environment

w .elsevierncornlocate/sna

Optical-Biophysical Relationships of VegetationSpectra without Background Contamination

Xiang Gao,' Aifredo R. Huete,' Wenge Nif and Tomoaki Miura'

For a better evaluation of the accuracy of VIs in estimat-ing biophysical parameters, a "true" VI value attributedonly to the vegetation signal andfree of any contaminationis needed. In this article, pure vegetation spectra wereextractedfrom a set of open and closed canopies by unmix-ing the green vegetation signalfrom the background compo-nent. Canopy model-simulation and reflectances derivedfrom graph-based linear extrapolation were used to unmixand derive a "true" vegetation signal, equivalent to a perfectabsorber (free boundary) canopy background reflectancecondition. Optical-biophysical relationships were then de-rived for a variety of canopy structures with differencesinfoliage clumping, horizontal heterogeneity, and leaftype.A 3-dimensional canopy radiative transfer model and ahybrid geometric optical-radiative transfer model (GORT)were used to simulate the directional-hemispherical reflec-tances from agricultural, grassland, andforested canopies(cereal and broadleaf crop, grass, needleleafi and broadleafforest). The relationships of the extracted red and near-infrared reflectances and derived vegetation indices (VIs)to various biophysical parameters (leaf area index,fractionof absorbed photosynthetically active radiation, and per-cent ground cover) were examinedfor the pure vegetationspectra. The results showed normalized difference vegeta-tion index (NDVI) relationships with biophysical parame-ters to become more asymptotic over the pure vegetationcondition. The extraction of pure vegetation signals hadlittle effect on the soil-adjusted vegetation index (SAVI),which had values equivalent to those obtained with thepresence of a background signal. NDVI values were fairlyuniform across the different canopy types, whereas the SAVI

' Department of Soil, Water and Environmental Science, Universityof Arizona, Tucson, AZ

l Department of Geography, University of Maryland, CollegePark, MD

Address correspondence to X. Gao, Dept. of Soil, Water, and Envi-ronmental Science, Univ of Arizona, Tucson, AZ 85721. E-mail: xgao@ag.arizona.edu

Received 16 December 1999; revised 12 May 2000.

REMOTE SENS. ENVIRON. 74:609 620 (2000)(©Elsevier Science Inc., 2000655 Avenue of the Americas, New York, NY 10010

values had pronounced differences among canopy types,particularly between the broadleaf and cereal/needleleafstructural types. These results were useful not only in select-ing suitable vegetation indices to characterize specific canopybiophysical parameters, but also in understanding a "true"VI behavior, free of background noise. ©2000 ElsevierScience Inc.

INTRODUCTION

Remote sensing plays an important role in the study of thebiosphere through its ability to make repeatable measure-ments of vegetation characteristics at global scales. Datafrom different wavebands (often visible and near-infraredwavelengths) have been combined to produce spectral veg-etation indices (VIs), which are sensitive measures of bothspatial and temporal variations in vegetation photosyntheticactivity and canopy structural variations. VIs have also beenshown to be well correlated with vegetation parameterssuch as leaf area index (LAI), biomass, canopy cover, andthe fraction of absorbed photosynthetically active radiation(fAPAR) (Tucker, 1979; Asrar et al., 1989; Sellers, 1985).

fAPAR is an important variable in studies of the energybudget and hydrology of the vegetated land surface, whileLAI is closely related to a variety of canopy processes,such as interception, evapotranspiration, photosynthesis,respiration, and leaf litterfall (Sellers et al., 1992a, b; Run-ning and Coughlan, 1988; Potter et al., 1993).

Generally, vegetation indices approach a saturationlevel asymptotically for a certain range of LAI (Sellers,1985) and respond linearly tofAPAR. However, a biophysi-cal explanation of the relationship between these indicesand observable vegetation phenomena is still subject tomuch discussion. Many studies have concluded that VI toLAI/fAPAR relationships are canopy structure and landcover dependent, varying with changes in leaf angle distri-bution, vegetation clumping, row orientation, spacing, andoptical properties of canopy components (leaf, stem, etc.)

0034-4257/00/$ see front matterPI' S0034-4257(00)00150-4

ELSEVIER

610 Gao et al.

(Asrar et al., 1992; Baret and Guyot, 1991; Choudhury,1987; Coward and Huemmrich, 1992; Roujean and Breon,1995). Different canopy types exhibit drastic variations incanopy structures and reflectance properties, which canproduce different VI values while having identical LAI orfAPAR values. If satellite data are to be used as a measuringtool to determine LAI/fAPAR over large areas where thereare differences in canopy characteristics, then an under-standing of these relationships specific to a given type ofcanopy must be developed.

In addition, solar zenith angle, sensor view angle, at-mospheric conditions, and background influences from soiland litter alter remotely sensed spectral signatures and thederived vegetation indices significantly (Baret et al., 1991;Huete, 1987; Deering et al., 1992; Deering et al., 1994).As noted in numerous studies, darker soil substrates resultin much higher vegetation index values for a given amountof vegetation when the ratio vegetation index (PNTR/Pred)or the normalized difference vegetation index (NDVI,(PNIR P,,d)/(PNIR+PYed)) were used as vegetation measures,while opposite soil brightness influences occur with theperpendicular vegetation index (PVI) (Huete et al., 1985;Elvidge et al., 1985; Roberts et al., 1990). Atmosphericturbidity generally inhibits reliable measures of vegetationand sometimes renders atmosphere-induced variations oncanopy spectra to exceed those due to vegetation develop-ment. These effects make the accurate and quantitativetranslation of VIs more difficult and complicated.

Myneni et al. (1995) reported that there are more than12 vegetation indices in the optical region and that theyhave been correlated with vegetation amount,fAPAR, un-stressed vegetation conductance, and photosynthetic ca-pacity. The choice and suitability of a VI is generally deter-mined by its sensitivity to the characteristics of interest,and/or its sensitivity to disturbing factors. Many effortshave been made to optimize vegetation indices and renderthem insensitive to variations in sun-surface-sensor geom-etries, atmosphere, calibration, and canopy background.Solar zenith, sensor view angle, and atmospheric influencesare increasingly being handled with improvements in atmo-spheric correction algorithms and bidirectional reflectancedistribution function (BRDF) models. Global, NOAA Ad-vanced Very High Resolution Radiometer (AVHRR)-NDVI processing includes correction for molecular scatter-ing and ozone absorption (Los et al., 1994). In the caseof the Moderate Resolution Imaging Spectroradiometer(MODIS), BRDF and atmospheric corrections are appliedprior to VI computation. Canopy background "brightness"effects on VIs, on the other hand, are not easily correctedand must be handled within the VI equation itself. The soiladjusted vegetation index (SAVI) (Huete, 1988) attemptedthis correction, with a soil adjustment factor L, to theNDVI equation to account for first-order soil-vegetationoptical interactions and differential red and NIR extinctionthrough the canopy [Eq. (1)J:

SAVI (1+L)-(pNIR pred)/(pNIR+p,,d+L), L 0.5. (1)

The enhanced vegetation index (EVI) is one of the pro-posed MODIS VI products and includes a soil adjustmentfactor as well as atmosphere resistance term, using the blueband (Liu and Huete, 1995; Huete et al., 1997) [Eq. (2)J:

EVI 2 5-(PNIR pYed)'(L+PNIR+CIpRed C2 'pbl.,) (2)

The atmosphere resistance concept was developed byKaufman and Tanr6 (1992). The current coefficients sug-gested for MODIS are L-1, C1 -6, and C2 7.5.

There have been extensive studies made on the sensi-tivities of VI-LAI/fAPAR relationships to internal and ex-ternal factors with radiative transfer models for both homo-geneous and discontinuous canopies (Baret and Guyot,1991; Coward and Huemmrich, 1992; Huemmrich andCoward, 1997; Asrar et al., 1992; B6gu6, 1993; Myneniand Williams, 1994). In those studies, a reference or stan-dard case was defined in terms of parameter values consid-ered typical from a remote sensing point of view. A sensitiv-ity analysis was performed by changing or perturbing thebase case parameter values one at a time and evaluatingthe influence of each disturbing factor. Some studies uti-lized the mean VI value as a "true" VI from which noiseanalyses were conducted to quantitatively characterize theinfluences of individual and combined external factors(Huete and Liu, 1994; Baret and Guyot, 1991).

Simulation studies with canopy reflectance modelshave provided sufficient grounds for relating fAPAR toNDVI by a simple linear model since the relationships areindependent of pixel heterogeneity and variations in leaforientation and optical properties (Myneni et al., 1994).However, such relationships remain very sensitive to thebrightness of the background materials. Huemmrich et al.(1997) stated that the effects of background reflectancewere dramatic with a range of possiblefAPAR values fora given NDVI that could vary as much as 50%, if thebackground reflectance was not known. Simulation studiesin forest canopies have also shown major problems in usingVIs for overstory canopy characterization, result from diffi-culties in discriminating the signal contribution of theoverstory from that of the background (including un-derstory vegetation) (Spanner et al., 1990). Many authorshave reported a lack of sensitivity in NDVI to overstoryLAI in open canopy conifer forests with dense understories(Spanner et al., 1990; Chen and Cihlar, 1996).

No matter how robust a VI algorithm is, influencesfrom external and nonvegetated factors cannot be com-pletely removed, resulting in some variabilities of the VI-LAI/fAPAR relationships. For validation purposes and er-ror/uncertainty analyses, a "true" VI value attributed onlyto the vegetation signal and free of any contamination isneeded. In order to assess the accuracy in a VI product, thetrue VI value must be established for any given vegetationcondition or amount. In this study we investigate a totalcorrection for background by removing the canopy back-

ground signal, equivalent to a perfect absorber canopybackground condition. We utilize data sets without an at-mosphere component, enabling an analysis of the opticalcharacteristics and VI behavior of the remaining "vegeta-tion" signal only as a function of various biophysical param-eters (leaf area index, fraction of absorbed photosyntheti-cally active radiation, and percent ground cover), withoutatmosphere and background contamination.

ANALYSIS APPROACH

A 3-dimensional canopy radiative transfer model and a hy-brid geometric optical radiative transfer model (GORT)were employed to simulate the directional-hemisphericalreflectances from a set of open and closed canopies withdifferences in canopy amount and structure (foliage clump-ing, horizontal heterogeneity, leaf type, etc.) (Myneni,1991; Myneni et al., 1991; Li et al., 1995; Ni et al., 1997;Ni and Woodcock, 2000). The four canopy types simulatedand analyzed include: homogeneous cereal crop/grass anddiscontinuous broadleaf crop under various percent coverand clump leaf area index (CLAI) conditions; needleleafforest and broadleaf forest under various crown cover andfoliage area volume densities (FAVD).

Model SimulationA 3-dimensional radiative transfer model, which takes intoaccount leaf clumping, hot spot, mutual shadowing, andlateral and vertical heterogeneity over different structuralland cover types (Myneni, 1991; Myneni et al., 1991; My-neni et al., 1997), was employed in this study to simulateradiation scattering and absorption over cereal crop/grassand broadleaf crops. The radiative transfer equation isparametrized by the leaf area density function, the proba-bility density of leaf normal orientation, and the leaf scatter-ing phase function and is numerically solved by discreteordinates method-by discretizing the angular variable intoa finite number of directions and introducing finite differ-ence schemes for the spatial derivatives (Myneni, 1991;Myneni et al., 1991). The model has been validated throughcomparisons with field measurements of soybean andmaize reflectance for trends and accuracy (Shultis andMyneni, 1988) and with AVHRR data over the First Inter-national Field Experiment (FIFE) sites in a grassland prai-rie (Privette, 1994). This model is currently being usedextensively by the MODIS and Multi-angle Imaging Spec-troRadiometer (MISR) Terra instrument science teams.

A hybrid geometric optical radiative transfer model(GORT), which incorporates multiple scales of clumping(shoots into branches, branches into whorls, whorls intocrowns, and crowns into stands which comprise the land-scape) in discontinuous canopies, was used to model theradiation regime of needleleaf and broadleaf forests. Themain difference in the simulation of these two forests is

Optical-Biophysical Relationships of Vegetation Spectra 611

the effect of the horizontal whorl, which allows more lightto pass through the canopy, present only in the needleleafforest canopies (Ni et al., 1997). In the GORT model, thediscontinuous canopy layer is modeled as an assemblageof randomly distributed tree crowns of ellipsoidal shapewith specified horizontal and vertical crown radii and arange of crown center heights. Within each single crown,the foliage and branches are characterized by the foliagearea volume density (FAVD). The effect of the heteroge-neous canopy structure due to all levels of clumping onthe radiation regime is characterized by modeling two typesof gap probabilities (within-crown and between-crowngaps) as a function of crown size, density, height, andfoliage area volume density (Li et al., 1995; Ni et al., 1997).The surface hemispherical reflectances are calculated bydividing all of the radiation scattered out of the canopylayer (upwelling scattered radiation at the top of the canopylayer) by incident solar radiation (Ni and Woodcock, 2000).This hybrid model was validated using vertical PAR trans-mission measurements over the old jack pine and old blackspruce stands in BOREAS (Ni et al., 1997). Recently itwas also validated using daily surface hemispherical reflec-tances collected during the winter and summer of 1995over a sparse old jack pine and a dense old black spruceforest in the northern area of BOREAS with good accuracy(Schewchuk, 1997; Ni and Woodcock, 2000).

Our analyses were performed by varying one variableof interest (background brightness, clumping leaf area in-dex/FAVD, ground cover/tree density) and setting the re-maining variables at nominal values which are believed torepresent an average state of each canopy type. The maincharacteristics and differences of the two agricultural cano-pies are shown in Table 1 (Myneni et al., 1997).

Input SpecificationsThe primary input variables and values specified in thetwo models used to simulate PAR absorption and canopyreflectances in blue, green, red, and NIR bands are listedin Table 2.

The tree geometry parameters (hl, h2, r, and b) col-lected in the old black spruce stands in the southern areasof BOREAS (called SOBS site) were used for GORT modelinput (Chen, 1996; Ni et al., 1997). The values of FAVDand tree density were set arbitrarily to study the effects ofcrown thickness and crown cover in both needleleaf andbroadleaf forests. According to Ni and Woodcock (2000),usually hl and h2 are not sensitive parameters and thecrown shape parameter bir can be assumed constant forany species of certain land cover type. The most sensitiveparameters with certain land cover type are the crownradius, tree density, and foliage areavolume density. Crownradius and tree density combine to yield the single mostimportant variable-crown cover. Branches are assumedto be perfect absorbers with zero transmittance and re-flectance.

Hemispherical reflectance and transmittance values

612 Gao et al.

Table 1. Structural Attributes of Two Agricultural Canopies (from Myneni et al., 1997)

Grass/Cereal Crop Broadleaf Crop

Horizontal heterogeneity (ground cover) No (100%) Variable (10 100%)Stems Green stems (10%')Leaf size (m) 0.05 0.10Leaf normal orientation Erectophile SphericalFoliage dispersion Minimal clumping (<1.0) Regular (clumping factor >1.0)Radiative transfer model 1-D 3-DPlant/clump LAI 0-7 0-7Vertical heterogeneity (leaf optics and LAD) No NoUnderstory No NoCrown shadowing No No

Stem fraction refers to the fraction of canopy LAI; green stems are modeled as erect reflecting protrusions with zero transmittance.

(400-2450 nm) were obtained for the leaves of cereal crop/grass, needleleaf forest, and broadleaf forest using an ASDspectroradiometer (Analytical Spectral Devices, Inc., Boul-der, Colorado), a BaSO4 integrating sphere (LI-1800, LicorInc., Lincoln, Nebraska), and a light source modified forfull-range spectral measurements (Asner, 1998; Asner etal., 1998). Each reflectance and transmittance sample wasa mean of 200 individual full-range spectral measurementsand then convolved to Channels 1-4 of Landsat TM (Table3). The hemispherical reflectance and transmittance valuesof cotton leaves, representing a broadleaf crop, were mea-sured with a PS2 (Personal Spectrometer II) spectroradi-ometer, a Licor integrating sphere, and a light source. Eachreflectance and transmittance sample was a mean of 10individual spectral measurements and also convolved toChannels 1-4 of Landsat TM (Table 3).

Four soils in dry condition were used in the simula-tions: i) bright, yellowish-brown Superstition sand (sandy,mixed, hyperthermic Calciorthid), ii) a high-iron, redWhitehouse sandy clay loam (fine, mixed, thermic UstollicHaplargid), iii) a brown Avondale loam (fine-loamy, mixed,hyperthermic Typic Torrifluvent), and iv) dark, organic-rich Cloverpsring Loam (fine-silty, mixed Cumulic Cryo-boroll). Their optical properties (Table 3) were obtainedusing a Barnes Modular Multispectral Radiometer (MMR)(Barnes Engineering Co., Stanford, Connecticut), whichmeasured radiant flux simultaneously in seven spectralbands (0.45-0.52 pm, 0.52-0.60 pm, 0.63-0.69 pm,0.76-0.90 pm,1.15-1.30pm,1.55-1.75pm, and2.08 2.30pm) with 15° field of view (Huete, 1987).

Extraction of Pure Vegetation SpectraTwo approaches were utilized to extract the pure vegeta-tion signal: 1) graph-based linear extrapolation, and 2)model simulation.

Theoretical Basis for Graph-Based Linear ExtrapolationHuete (1987) used a simple, first order interaction modelto decompose measured spectra over incomplete plant can-opies into a soil-dependent component and a pure vegeta-tion component free of soil influences. The model can beexpressed with Eqs. (3) and (4):

di(Ql) -Eo(2)rn()t2(2) +Eo(2)r6QL),

E,(i) Idr(2 r=rm(2l)r6 (2l)±rsQ)tcQ%),

(3)

(4)

where d, is the spectra of the soil-canopy mixture; E0rst2is the soil-dependent component, which is the product ofglobal irradiance at the top of the canopy Eo, soil reflectancer1, and the downward and upward global transmittancethrough the canopy t2; E0 r, is the vegetation componentand rJ(A) equals canopy reflectance.

The above equation is based on two assumptions: 1)second-order, soil-plant interactions are negligible; 2)downward and upward canopy transmittances are equal.When soil background behaves as a perfect absorber(r, 0), then dr, Eor, Utilizing a simple analytical canopyRT model, Yoshioka et al. (1999) also presented a similarequation to describe the reflectance of the canopy-soilsystem by assuming a single canopy layer and a soil layerunderneath the canopy. They, however, used the averagecanopy transmittance defined as the logarithmic averageof the downward and upward transmittance of the canopylayer for t, in (4). They also included the higher-orderinteraction terms but found them to account for only mi-nor differences.

Thus, in a plot of canopy reflectance against back-ground reflectance for an identical canopy cover but withdifferent backgrounds, the pure vegetation signal r, can beextrapolated from the intercept (canopy reflectance withbackground of zero reflectance). The slope of such a line isthe two-way global transmittance. Band-by-band derivationyields pure vegetation reflectance spectra free of back-ground influences. In this study, modeled canopy reflec-tance with four soils of varying brightness as backgroundwere utilized in deriving the linear regression and extrapo-lating to the zero soil case.

Model SimulationModel simulation was accomplished by inserting "zero"for background reflectance in each band and canopy typesimulation run. This also provided a zero-casefAPAR value(fraction of absorbed radiation free of background in-fluence).

Optical-Biophysical Relationships of Vegetation Spectra 613

A Lambertian soil case was assumed in all simulations.fAPAR was approximated by the total absorbed radiationfraction within the red wavelengths. As noted by Myneniand Williams (1994), fAPAR can be estimated to 95% ofits true value by the waveband 0.589-0.685 pm. In the 3-D radiative transfer model,fAPAR is calculated as the sumof the following four parts: absorbed direct solar radiation,absorbed diffuse sky radiation, absorbed soil-reflected inci-dent radiation, and absorbed scattered radiation. In theGORT model, fAPAR is calculated based on the verticaldistribution of solar radiation absorption density. In addi-tion, fAPAR used in this study represents instantaneousvalues at the same solar zenith angle as that of the reflec-tance calculations.

The pixel-based canopy leaf area index (canopy LAI)and effective leaf area index (effective LAI, Le) were usedin agriculture/grass and forest canopies, respectively, foranalyses of their relationships with VIs. Canopy LAI in-cludes the contributions from stem and is calculated as[ground cover-(/clump LAI+stem silhouette area index)].In the case of cereal crop/grass, stem silhouette area indexis 0.0, and canopy LAI is equivalent to /clump LAI. Weassumed that a clump LAI (CLAI) of 0.01 is small enoughto be considered as zero vegetation density. Effective LAI(Le), which takes into account the clumping effect of nee-dle into shoots, is related to foliage area volume densityFAVD, crown volume Vcrown, horizontal crown radius r,vertical crown radius b, and tree density A by Eq. (5) (Chenet al., 1991; Ni et al., 1997):

Le - FAVDPVc_-WP2 FAVD-4/37TAb2O. (5)

Combining with equation crown cover(%)= 1 e`2(Table 2), we obtain Eq. (6):

Le4FAVD-4/3b-ln( c c%)1 I crown cover%

(6)

So, the "FAVD-b-4/3" is similar to the function of CLAIin agriculture/grass. Both ground cover and crown coverrefer to the percent of pixel covered by the green vegetationin this study.

We limit our results to the NDVI and SAVI. The EVIwas also analyzed but not presented due to its similarity toSAVI under the "no atmosphere" conditions of this study.

RESULTS

Agriculture

Uniform Case (Cereal Crop/Grass and Broadleaf Cropwith 100% Cover)Spectral Signatures. In Figure 1, simulated canopy reflec-tances of cereal crop/grass in the red and NIR bands areplotted against bare soil background reflectance for variouslevels of clump LAI (CLAI). Each line represents a con-stant vegetation amount with different soil backgrounds.The intercepts of such a plot represent modeled canopyreflectances with zero background, while the slopes repre-sent the two-way global canopy transmittances [see Eq.(4)]. As can be seen, modeled canopy reflectances behavelinearly with soil background reflectance with regressionanalyses indicating RI values larger than or close to 0.99.In the visible bands (only the red band is shown), modeledcanopy reflectances generally decrease with increases invegetation amount. In the NIR band, the change in mod-eled canopy reflectance with vegetation amount is depen-dent on the underlying soil brightness with a "critical" soilreflectance value of -0.30. Brighter soils (>0.30) result

Table 2. Parameter Specifications in Model Simulations

Direct composition of incident radiation (0.8)Solar zenith angle (400)

Illumination and viewing Sensor view zenith and azimuth (hemispherical)conditions Agriculture Forest

Canopy and landscape conditions * Plant/clump leaf area index, CLAI * Lower bound of crown center height, hi (3.0 m)(m /m') (0.01, 0.25, 0.50,0.75, 1.00, * Upper bound of crown center height, h2 (8.5 m)2.00, 3.00, 4.00, 5.00, 7.00) * Horizontal crown radius, r (0.76 m)

* Ground cover (10%, 25%, 50%, 75%, * Vertical crown radius, b (2.7 m)and 100% for broadleaf crop only) * Leaf orientation factor, G (1.5)

* Leaf angle distribution (Table 1) * Foliage area volume density, FAVD (m/m')(0.0028, 0.0694, 0.1389, 0.2083, 0.2778, 0.5556,0.8333)

* Tree density, 2l (no. of trees/m') (0.1585,0.1966,0.2815, 0.3820, 0.505,0.6635, 0.764, 1.0455,1.2689, 1.6509)

* Leaf angle distribution (spherical)

Optical properties of components in Leaf: reflectance and transmittancethe canopy system (Table 3) Background (soil): reflectance

CLAIL one-sided total surface area of leaves per unit ground area.b 2: it is related to the crown cover by the equation {crown cover(%) 1 e Ir) (Ni et al., 1997). Thus, the listed tree densities correspond to 25%,

30%, 40%, 50%, 60%, 70%, 75%, 85%, 90%, and 95% crown cover, respectively.

614 Gao et al.

Table 3. Optical Properties of Soils and Green Leaves of Four Canopy Types Used in the Model Simulations

Reflectance'

Blue Green Red NIB Transmittancea(0.45-0.52) (0.52-0.60) (0.63-0.69) (0.76-0.90) Blue Green Red NIB

Grass/cereal crops 0.07 0.10 0.08 0.40 0.02 0.10 0.04 0.38Broadleaf crops 0.10 0.15 0.11 0.52 0.01 0.05 0.01 0.39Needleleaf forest 0.08 0.15 0.08 0.45 0.05 0.08 0.03 0.30Broadleaf forest (EG)b 0.06 0.11 0.05 0.46 0.01 0.08 0.03 0.41Superstition sand 0.18 0.256 0.337 0.378Avondale loam 0.087 0.128 0.188 0.232Whitehouse sandy clay loam 0.041 0.078 0.158 0.206Cloverspring loam 0.033 0.044 0.062 0.093

'The wavelength range of each wavelength region (blue, green, red, and NIR) for transmittance is same as that for reflectance.bEGCevergreen species in Brazil.

in decreasing modeled canopy reflectance with increasingamounts of vegetation, while darker soils (<0.30) resultin increasing modeled canopy reflectance with vegetationamount. Soil brightness always has a positive effect onmodeled canopy reflectances in all wavebands and CLAIlevels. Note that there is also a critical soil reflectance valuein the red ( 0.025) below which canopy spectra wouldincrease with vegetation amount.

The model-simulated and graph-derived, "zero" back-ground canopy reflectances were compared and showedlittle difference in magnitude with maximum RMS errorsin the red below 0.0002 and in the NIR below 0.003. Thisindicates that a first-order, interaction model can ade-quately describe the manner in which soil and vegetationspectra mix to produce a composite canopy response. Thetwo-way global transmittance spectra (slopes in Fig. 1) andextracted pure canopy reflectance spectra with "zero" soilbackground (intercepts in Fig. 1) are shown in Figure 2over four bands and various levels of CLAI. The resultingpure canopy reflectance spectra are free of backgroundcontamination and start at 0.0 for zero vegetation amount(CLAI 0.01) and increases with CLAI at all wavelengths,including the visible. Sensitivity to vegetation amount isgreatest in the NIR and very low in the visible. The two-way global transmittance curves, which are independent

Figure 1. Relationship between simulated canopy reflectanceand underlying bare soil reflectance in red and NIR bandsfor various levels of clump LAI (.01, .25, .50, .75, 1.0, 2.0,3.0, 4.0, 5.0, and 7.0) for cereal crop and grass.

O4L Avorde0 Superst t1

Vh tehouse0°3 - l0 CoversPig

02 -

04

° 001 Wu

' 03

> W

m 020050 a0

0 1

' r 01I, - C

700D0 '

00 0.1 02 03 04

So I red reflectance

of soil background, ranged from 1.0 at zero vegetationamount in all wavebands to 0.0 at dense vegetation amounts(CLAI 7.00). The magnitude of transmittance in theNIR band is higher than those of the visible bands, espe-cially at intermediate vegetation densities. The visiblebands have similar magnitudes of transmittance with thegreen band slightly higher. These results agree well withknown wavelength-dependent flux extinction through veg-etated canopies.

Reflectance!Vts-canopy LAI relationships. The purecanopy reflectance in the red and NIR bands, as well asthe mixed spectra of vegetation with dark (Cloverspring)and bright (Superstition) soil backgrounds for cereal crop/grass and broadleaf crop are compared in Figure 3. In thered band, the two canopies present a similar trend in thatreflectances over both dark and bright soils decrease withcanopy LAI, yet increase with canopy LAI over the "zero"soil. In contrast, NIR reflectances behave quite differentlyfor the two canopies. The NIR reflectance of all canopiesincreases with canopy LAI for all backgrounds except inthe case of the cereal crop with bright soil. We also notethat the NIR band shows the best discrimination of canopytypes with canopy reflectance levels of the broadleaf cropmuch higher than those encountered in the cereal cropcase. Thus, from a pure vegetation signal perspective (zero-soil), we can conclude that it is the NIR that shows the best

Figure 2. Derived canopy reflectance spectra for "zero" soilbackground and transmittance spectra for various levels ofclump LAI (.01, .25, .50, .75, 10, 2.0, 3.0, 4.0, 5.0, and 7.0)for cereal crop/grass.

Z0 r Oso (deied)

700

200 11

Crit al Soil0 50 Refe tace Va e'

o01

I

t 02o

0 o

L 00

I02002

Is

I050 XI

00 co3

00 I00 0O 1 0 2 0.3 0 4 0 0 6 97 08 O

Soil NIR reflectance Wavelength (um)

10 001

0850

06

04

02 2

0D0 7 00

05 03 07 08 09

Wavelength (urn)

Optical-Biophysical Relationships of Vegetation Spectra 615

04 -C ral Bradlea

0 v Zro0s o Dark O

Q °9ght ®

02 Q

t30 1t

0.00 2 4 6

Canopy LAJ

06O

2 04

02

000 2 4 6

Canopy LAI

05

z r->02

at 021<i

X 0048 0 2 4

Canopy LAI6

04

C 03-

02

00

00 02

Ceaf B-oadleaf05

Dak C 0 4 E

Brg 0

Ai 0 3

Z 02

01, .

04 06 08 1,0fAPAR

V /H

00 l .00 02 04 06 0

fAPAR

0806

> 04

02

008 0 2 4

Canopy LAI

0Z 0A

xi ?1

q/ 4'

0000 02 04 06 08

6 8fAPAR

> 04u9

02

10

Figure 3. Relationship between canopy LAI and canopy redand NIR reflectance and VIs for "zero," dark, and brightsoil backgrounds for cereal crop/grass and broadleaf crop(uniform case).

discrimination of vegetation structural variations (cereal vs.broadleaf). The visible bands provide very little discrimina-tion and most changes in the visible bands are associatedwith soil background differences instead of vegetation.

Although soil brightness always has a positive effecton canopy reflectance, vegetation indices give differentresponses to soil brightness (Fig. 3). Darker soils generallyresult in higher NDVI values for incomplete canopies.However, extrapolating to a "no soil" case, further raisesthe NDVI, particularly over low canopy LAI conditions,where the NDVI value varied by more than 0.6 units overthe three backgrounds (Fig. 3). The vegetation index valuefor zero soil determines the upper limit of the NDVI.Figure 3 shows that the NDVI is not only background-sensitive, but most of its dynamic range occurs only withthe presence of a soil background, the brighter the back-ground the greater its dynamic range. The NDVI exhibitedvery little sensitivity to vegetation for the zero soil case,approaching saturation throughout the entire range of can-opy LAI. The presence of a soil background restores anexponential dynamic range of NDVI, but in a mannerdependent on the background optical properties.

For both canopy types, the SAVI had "zero" soil behav-ior not too different from those of the bright and dark soilcases, all responding exponentially to canopy LAI. TheSAVI also had slight soil brightness problems, but oppositeto that of the NDVI, with brighter backgrounds producingslightly higher values. In addition, the NDVI-canopy LAIrelationship was more sensitive to soil background thancanopy type as NDVI values varied little between the twotypes (Fig. 3). This mimicked the lack of variations foundin the red band. The NDVI also saturated at lower canopy

Figure 4. Relationship between fAPAR and canopy red andNIR reflectance and VIs for "zero," dark, and bright soilbackgrounds for cereal crop/grass and broadleaf crop(uniform case).

LAI values (-2) than the other indices (canopy LAI -4).The SAVI was not so sensitive to soil background but verysensitive to canopy type. The asymptotic maximum valuesof the SAVI were 0.45 and 0.65, respectively, for cerealcrop/grass and broadleaf canopies. This agrees with thelarge separation in asymptotic NIR reflectances (0.3 and0.48) encountered between the two canopy types.

Reflectance!Vls-fAPAR relationship. The relation-ships betweenfAPAR and red/NIR canopy reflectance un-der "zero," dark, and bright backgrounds for the two canopytypes are shown in Figure 4. They are very similar to thoseof red/NIR canopy reflectance with canopy LAI exceptthat fAPAR is more linearly related to canopy reflectance.Note that red reflectances converge at maximum fAPARvalues while NIR reflectances separate with maximumfA-PAR values according to canopy type. Soil brightness hada positive effect on fAPAR magnitude with the brightestsoil resulting in the highestfAPAR for a constant canopyover both canopy types (Fig. 4). This was due to the highersoil-reflected PAR which could be absorbed by the canopy.Soil brightness also affected the relationship between vege-tation indices and fAPAR with SAVI responding nearlylinearly to JAPAR for zero, dark, and bright soil back-grounds of both canopy types. The relationship of NDVIwithfAPAR was mostly linear with a soil-dependent slopeor sensitivity which became very small for the zero soilcase. As in the relationships with canopy LAI, the NDVIdeveloped a nearly saturated relationship with fAPAR forthe "no soil" case and had a greater dynamic range onlyin the presence of brighter soil backgrounds. Canopy typeinfluences on VIs-fAPAR relationships were less significantthan the VI-canopy LAI relationships. The NDVI was

10

00 i00 02 04 06

fAPAR

10

v

616 Gao et al.

0 4 0 504 ; 25 % 50 %100 % 0

054o i iv Zero v a, ^o Dank z 04

o 020 03

00 00

0 2 4 6 0 2 4Canopy LAI Canopy LAI

1 0

080 0

06

>Z 0o4 1: l: I .

0.2 id ~:: :00

0 20 40 60 60 100 12C

Ground ove

10

008

069700

0 5

- 082

* 0 20 40 60 60 100 120G oLird cove

06

06

0 6

Z 04

02

00

06

<040A

002

000 2 4 6 8 0 2 4 6 a

Canopy LAI Canopy LAI

Figure 5. Relationship between canopy LAI and canopy redand NIR reflectance and VIs for "zero," dark, and brightsoil backgrounds for broadleaf crop with 25%, 50%, and 100%ground cover.

nearly insensitive to canopy type while the SAVI respondeddifferently to each type atfAPAR values greater than 0.4.

Effects of horizontal heterogeneity (Broadleaf Crop)Spectral signatures. One of the main differences in canopycharacteristics between broadleaf crop and cereal crop/grass is the horizontal heterogeneity. Two parameters areneeded to describe such a canopy: clump LAI and %ground cover. Under heterogeneous canopies, one mayhave significant soil background signals and high canopytransmittance values along with high clump LAI values.As the canopy develops from low to full cover, the heteroge-neity of the incomplete canopies decreases until a homoge-neous, full cover is achieved. Regression analyses showedthe same linear relationships between canopy response andunderlying soil reflectance for any combinations of clumpLAI and ground cover, enabling the derivation of 'pure'

Figure 6. Relationship between fAPAR and VIs for "zero,"dark, and bright soil backgrounds for broadleaf crop with25%, 50%, and 100% ground cover.

10 -

08

086

04

02

00

06

>

02-

Figure 7. Relationship between ground cover and VIs for"zero" (solid line) and bright soil (dotted line) backgroundsfor various levels of clump LAI (.01, .25, .50, .75, 1.0, 2.0,3.0, 4.0, 5.0, and 7.0) for broadleaf crop.

canopy spectral signatures (zero soil) and two-way globaltransmittance spectra through graph interpolation.

Reflectance/VIs-canopy LAI relationship. The rela-tionship between canopy LAI and canopy reflectance un-der "zero," dark, and bright soil backgrounds for broadleafcrop canopies of 25%, 50%, and 100% cover is shown inFigure 5. Canopy red reflectances under "zero" and darksoil backgrounds show near saturation to variations of bothcanopy LAI and %cover, but is very responsive over thebright soil background. In contrast, NIR canopy reflec-tances are more sensitive to variations of foliage clumpingand horizontal heterogeneity over "zero" and dark soils.Note that NIR canopy reflectances over the bright soilbackground show different variations with canopy LAI be-tween heterogeneous and uniform canopy cases (Fig. 3).Thus, the broadleaf crop of 100% cover was the only case

Figure 8. Derived canopy reflectance spectra for "zero" soilbackground for various levels of FAVD (0.003, 0.07, 0.14,0.21, 0.28, 0.56, and 0.83) for needleleaf and broadleaf forestwith 50% and 95% crown covers.

03 03

Zerosol (ded)

N-edl -f (50%)

-0,2

o 01

0,0

04

t 0 3

> 0 2o01c 1

. 0,

Zco (dC-d)

Bradle /50%1

014

0 003

05 06 07 08 09

Wavelength (urn)

Z z i (derivd)

Ne- eeaf (95%)

00

00 02 04

fAP!

06 08 1 0 08 1 0fAPAR

05 06 07 08 09

Wave ergth (um)

083

014

01

00

05 06 07 08

Wavelength (um)

0.4 Ze s (dOd)

Ba-def (95%)

03

02

01

00

09

, 014

I003

05 06 07 08 09

Wavelength (urn)

>z

v

Optical-Biophysical Relationships of Vegetation Spectra 61 7

10

08

04

Figure 9. Relationship between effec-tive LAI and VIs for "zero," dark, andbright soil backgrounds for needleleafand broadleaf forest with 25%, 50%, and95% crown covers.

02

00

08

> 0.6

0

04

25% 50% 959,v ZroZ Darkh Bright

Effective LAI

8 9

02

00

07

06Broad oaf

05

0o425 04

25% 50% 05 0 2

Zeoo VDark O X

0 1 2 3 4 5 6 7 8 9

Effective LAI

01

00

25% 50%

, ... - I

Needlet- Broacdlea

1 Dark 0 0i o Brght e

0 1 2 3 4 5 6 7 L 9

Effective LAI

where NIR canopy reflectance increased with clump LAI(CLAI) over a bright soil.

Several features can be observed in the influence ofhorizontal heterogeneity on the relationship between eachVI and canopy LAI for the bright, dark, and "zero" soilbackgrounds (Fig. 5). The relationship between VI andcanopy LAI was not unique but dependent on the %coverof the canopy and "soil-noise" variability. Thus, a canopyLAI of 2 may result in NDVI values ranging from 0.35to 0.85, depending on the interplay of %cover and soilbrightness. Generally, canopies with sparse clumps andhigher ground cover (e.g., CLAI 2.0 and 50% groundcover) have higher VI values than those with dense clumpsand lower ground cover (CLAI 4.0 and 25% groundcover). This was true for all indices with one difference:When canopy LAI is less than 1.0, SAVI values are rela-tively insensitive to the various configurations of groundcover and CLAI. This implies that VIs are not indicativeof canopy LAI or absolute amount of leaf area, but respon-sive to the spatial distribution of such leaf area. Thus, theconcept of canopy LAI is more meaningful only in theuniform canopy case (complete ground cover). By contrast,the SAVI reduced soil variability very well for both incom-plete and uniform canopies (Fig. 5), allowing one to ob-serve the unique relationships between VI and canopy LAIfor various clumping configurations. As in the cereal crop/grass case, the NDVI showed little variation over the entirerange of canopy LAI and %cover conditions under "zero"soil background due to saturation effects.

VIs-fAPAR relationships. There is a fairly linear rela-tionship of both VIs with fAPAR (Fig. 6). This relationship,however, is strongly soil dependent in the NDVI. Soilinfluences were stronger in the NDVI-fAPAR relationship

than in NDVI-canopy LAI, because a brighter backgroundhas the effect of enhancingfAPAR (through backscatter)while depressing NDVI values. As with the cereal crop/grass canopies, the zero soil case yielded a smaller rangeof NDVI values (0.68-0.80) with varying %cover. TheSAVI-fAPAR relationship at different percent ground cov-ers were relatively insensitive to the underlying soil bright-ness and linear throughout the VI dynamic range. One canalso observe a nearly unique correspondence between bothVIs and fAPAR regardless of the spatial distribution ofleaf area in a pixel. This indicates that the relationship isindependent of pixel heterogeneity, as observed in previousstudies (Myneni et al., 1994). Nevertheless, at a given can-opy LAI, PAR absorption can differ depending upon howground cover and CLAI are distributed in different config-urations in a canopy. Canopies with greater ground coverand lower CLAI are more absorptive than canopies withlower ground cover and higher CLAI. Thus, the patternof leaf area distribution on the ground is a better determi-nation of the radiation regime than the absolute amountof such leaf area. In summary, the spatial distribution ofleaf area determines the magnitude of both canopy reflec-tance and PAR absorption. Although the relationship be-tweenfAPAR and NDVI is indeed insensitive to the spatialdistribution of leaf area, its sensitivity to the underlyingsoil brightness makes it difficult to extrapolatefAPAR fromNDVI or vice versa, not as in the strong, unique relation-ships between SAVI andfAPAR.

VIs-ground cover relationships. The relationships ofVIs with percent ground cover over "zero" and bright soilbackground are plotted in Figure 7 for the various levelsof clump LAI (CLAI). The relationships are quite differentfor NDVI under the two cases with NDVI relatively insen-

10

0.8

06

0

04

Figure 10. Relationship between fA-PAR and VIs for "zero," dark, and brightsoil backgrounds for needleleaf andbroadleaf forest with 25%, 50%, and 95%crown covers.

02

0c

10Needlelaf

08

06

04

, 25%~, 50% 95%

2 zeo v 0.2/ oDark 2

A Bright A * jO 8 I 00+

00 02 04 06 068 I 1 0 0

fAPAR

02 04

fAP!

/050% 95%

Zero vvDark ::Bright L

06 06 1 0

07N-dcoleaf Broadlif

0.6 v Zer ...

O Dark * 0

0,5 0 Brght 0

004

02

0 1

0 000 02 04 06 08 1

fAPAR

,,, T

i .4 H

AR

618 Gao et al.

10

08

06

04

-J _ NRDVJ = 025 + 1 III 'SAV

,r'=0574 (al cases, 2 - 0 837 (rea so I case)

0 0 0 02 4 06 08 1 0

SAVI

08

06

> 04

02

00

EV 3 -0 039 + 1 136 ' SAVIr 0993 (a aes)

Ceral Crop /GassBroadlef CropNeed e ef f orestBoadleaf crestRegssion iFO

00 02 04 06 08

SAVI

Figure 11. Crossplots of NDVI and EVI vs. SAVI for all casesof the four canopy types.

sitive to percent ground cover over the "zero" background.In contrast, the "zero" soil case had no major influence onthe SAVI-%cover relationship. For both VIs, the resultingrelationships were generally linear and strongly CLAI de-pendent. With an increase of CLAI, the sensitivity of VIto ground cover generally increased significantly, exceptin the case of NDVI over "zero" background. Thus, a VI-%cover relationship in a broadleaf crop or shrub canopywould require knowledge of the mean CLAI of the shrubplant, broadleaf plant, or row crop.

ForestSpectral signatures. The derived canopy reflectance spec-tra without soil contamination for needleleaf and broadleafforests of 50% and 95% crown cover are depicted in Figure8. The differences between needleleaf and broadleaf forestare very similar to those between cereal crop/grass andbroadleaf crops. NIR reflectances responded to bothFAVD, crown cover, and canopy type. With the samecrown cover and FAVD, the magnitude of "pure" canopyreflectance in the NIR band is higher in the broadleafforest. In contrast, there were very few differences in themagnitudes of "pure" canopy reflectances in the visiblebands under 50% and 95% crown covers, various levels ofFAVD, and two canopy types (needleleaf vs. broadleaf).

VIs-biophysical relationships. The relationships be-tween VIs and biophysical parameters (effective LAI andJAPAR) of both needleleaf and broadleaf forests with vari-ous backgrounds and crown covers are depicted in Figures9 and 10, respectively. For the same VI and biophysicalparameter, their relationships are very similar, except thatthe broadleaf forest consistently resulted in higher VI val-ues. Figure 9 shows the NDVI-effective LAI relationshipsto be more sensitive to canopy background and relativelyinsensitive to crown cover and canopy structure type. Incontrast, the SAVI-effective LAI relationships were pre-dominantly canopy structure dependent (broadleaf vs. nee-dleleaf). There is a generally linear to slightly curvilinearrelationship between VIs andfAPAR with, once again, theNDVI sensitive to canopy background and insensitive tocanopy structure while the SAVI sensitive to canopy struc-ture and insensitive to background (Fig. 10).

C;2

In this study, all analyses were also performed for theenhanced vegetation index (EVI). We did not include anEVI analysis here since these results were similar to SAVIunder no atmosphere (r2 0.99 for the four canopy typeswith different foliage clumping/FAVD, horizontal hetero-geneity, and background brightness) (Fig. 11). Thus, all ofthe properties found in SAVI were also applicable to theEVI In comparison, the correlation between NDVI andSAVI for the four canopy types were weaker with r2 valuesof 0.84 over the dark to bright canopy backgrounds (Fig.11). This relationship was further degraded when the zerosoil case was included in the regression (r2 -0 57)

SUMMARY AND CONCLUSION

In this article, a 3-dimensional canopy radiative transfermodel and a hybrid geometric optical-radiative transfermodel (GORT) were employed to simulate the directional-hemispherical reflectances from agricultural, grassland,and forested canopies (cereal and broadleaf crop, grass,needleleaf and broadleaf forest). Pure vegetation spectra,free of background contamination and equivalent to a"zero" canopy background reflectance condition, were de-rived by both model simulation and graph-based linearextrapolations. These spectra and derived VIs were ana-lyzed to understand the "true" VI relationships to variousbiophysical parameters characteristic of four canopy typesexhibiting variations in canopy structure, foliage clumping,horizontal heterogeneity, and leaf type.

The results showed significantly greater saturationproblems in the relationships between NDVI and variousbiophysical parameters with removal of soil backgroundcontamination (zero soil). It was the presence of a soilbackground that restored the dynamic range of NDVI ina manner dependent on the background optical propertiesand enabled the use of NDVI to assess variations in vegeta-tion amount. In contrast, the extraction of pure vegetationspectra did not have a major influence on the SAVI valueor its relationship with biophysical parameters. NDVI-biophysical relationships were found to be more sensitiveto soil background than canopy structure type while SAVI-biophysical relationships were not sensitive to backgroundbut responded to canopy structure type and amount.

The pattern of leaf area distribution on the groundwas found to be more determinant of the canopy radiationregime than the absolute amount of such leaf area. Thus,canopy LAI /effective LAI was only meaningful in theuniform canopy cases, while ground cover and clump LAI/FAVD were more important in characterizing the radiationregime and canopy reflectances in open canopies. Thespatial heterogeneity of vegetation canopies did not signifi-cantly affect the relationship between VIs and JAPAR.However, the sensitivity of the NDVI-fAPAR relationshipto the underlying soil brightness made it difficult to extrap-olatefAPAR from NDVI or vice versa. This was in contrast

to the strong, unique relationship found between SAVIand fAPAR.

We also found NDVI to be advantageous in yieldingbiophysical relationships applicable across varying canopytypes (cereal crop vs. broadleaf crop, or needleleaf forestvs. broadleaf forest). However, NDVI sensitivity to soilbackground affected these relationships and requiredknowledge of soils for biophysical estimations. The SAVIprovided biophysical relationships free of soil influencessimilar to those in the real soil cases, but their relationshipswere sensitive to canopy structure, thus requiring knowl-edge of canopy type for biophysical assessments. In theMODIS-era, land cover maps, would be available to aidin the application of the SAVI and EVI to yield biophysicalvegetation information. In contrast, the NDVI would bene-fit from the availability of soil maps for improved biophysi-cal information extraction.

These studies have provided some insights into select-ing suitable VIs to derive specific canopy biophysical pa-rameters across different canopy types, and evaluating theaccuracy of derived biophysical canopy parameters withremote sensing measurements. An accuracy analysis nor-mally requires a standard or "true" value from which onecan measure error and uncertainty. An atmosphere-freeVI value determined from nadir view and reference sunangle still suffers from background contamination unlessa "reference" soil can be assigned. In reality the canopybackground is composed of a variety of materials, suchas litter, rock, and vegetation understory, whose opticalproperties are quite different from that of general soils.The SAVI appears to eliminate this dilemma. Nevertheless,it is of future interest to further test the zero background/"true" VI behavior over additional background materialsand to validate this in the field over different canopy types.

The authors wish to thank G. P. Asner, W. van Leeuwen, and J.Qi for providing the leaf optical property data of the relatedcanopy types. Also the author is grateful to R. B. Myneni forproviding the source code of 3-D radiative transfer model. Thisstudy was funded by NASA-MODIS Contract NA5-31364 (A.Huete).

REFERENCES

Asrar, G., Myneni, R. B., and Kanemasu, E. T. (1989), Estimationof plant-canopy attributes from spectral reflectance measure-ments. In Theory and Applications of Optical Remote Sensing(G. Asrar, Ed.), Wiley, New York, pp. 252-296.

Asrar, G., Myneni, R. B., and Choudhury, B. J. (1992), Spatialheterogeneity in vegetation canopies and remote sensing ofabsorbed photosynthetically active radiation: a modeling study.Remote Sens. Environ. 41:85-103.

Asner, G.P. (1998), Biophysical and biochemical sources of vari-ability in canopy reflectance, Remote Sens. Environ. 64:234-253.

Asner, G. P., Wessman, C. A., Schimel, D. S., and Archer, S(1998), Variability in leaf and litter optical properties: implica-

Optical-Biophysical Relationships of Vegetation Spectra 619

tions for BRDF model inversions using AVHRR, MODIS andMISR. Remote Sens. of Environ. 63:243-257.

Baret, F., and Guyot, G. (1991), Potential and limits of vegetationindices for LAI and APAR assessment. Remote Sens. Envi-ron. 35:161-173.

Begue, A. (1993), Leaf area index, intercepted photosyntheticallyactive radiation, and spectral vegetation indices: a sensitivityanalysis for regular-clumped canopies. Remote Sens. Envi-ron. 46:45-59.

Chen, J. M. (1996), Optically-based methods for measuring sea-sonable variation of leaf area index in boreal conifer stands.Agric. For. Meteorol. 80:135-163.

Chen, J. M., and Cihlar, J. (1996), Retrieving leaf area index ofboreal conifer forests using Landsat TM images. Remote Sens.Environ. 55:153-162.

Chen, J. M., Black, T. A., and Adams, R. S. (1991), Evaluationof hemispherical photography for determining plant area indexand geometry of a forest stand. Agric. For. Meteorol. 56:129-143.

Choudhury, B. J. (1987), Relationships between vegetation indi-ces, radiation absorption, and net photosynthesis evaluated bya sensitivity analysis. Remote Sens. Environ. 22:209-233.

Deering, D. W., Eck, T. F., and Grier, T. (1992), Shinnery oak bi-directional reflectance properties and canopy model inversion.IEEE Trans. Geosci. Remote Sens. 30(2):329-348.

Deering, D. W., Middleton, E. M., and Eck, T. F. (1994), Reflec-tance anisotropy for a spruce-hemlock forest canopy. RemoteSens. Environ. 47:242-260.

Elvidge, C. D., and Lyon, R. J. P. (1985), Influence of rock-soilspectral variation on the assessment of green biomass. RemoteSens. Environ. 17:265-279.

Goward, S. N., and Huemmrich, K. F. (1992), Vegetation canopyPAR absorptance and the normalized difference vegetationindex: an assessment using the SAIL model. Remote Sens.Environ. 39:119-140.

Huemmrich, K. F., and Goward, S. N. (1997), Vegetation canopyPAR absorptance and NDVI: an assessment for ten tree spe-cies with the SAIL model. Remote Sens. Environ. 61:254-269.

Huete, A. R., Jackson, R. D., and Post, D. F., (1985), Spectralresponse of a plant canopy with different soil backgrounds.Remote Sens. Environ. 17:37-53.

Huete, A. R. (1987), Soil-dependent spectral response in a devel-oping plant canopy. Agron. J. 79:61-68.

Huete, A. R. (1988), A Soil-adjusted vegetation index (SAVI).Remote Sens. Environ. 25:295-309.

Huete, A. R., and Liu, H. Q. (1994), An error and sensitivityanalysis of the atmospheric and soil-correcting variants of theNDVI for the MODIS-LOS. IEEE Trans. Geosci. RemoteSens. 32:897-905.

Huete, A. R., Liu, H. Q., Batchily, K., and van Leeuwen, W.(1997), A comparison of vegetation indices over a global setof TM images for EOS-MODIS., Remote Sens. Environ.59:440-451.

Kaufman, Y., and Tanre, D. (1992), Atmospherically resistantvegetation index (ARVI) for EOS-MODIS, IEEE Trans.Geosci. Remote Sens. 30:261-270.

Li, X., Strahler, A. H., and Woodcock, C. E. (1995), A hybridgeometric optical-radiative transfer approach for modelingalbedo and directional reflectance of discontinuous canopies.IEEE Trans. Geosci. Remote Sens. 33:466-480.

Liu, H. Q., and Huete, A. R. (1995), A feedback based modifica-

620 Gao et al.

tion of the NDVI to minimize canopy background and atmo-spheric noise. IEEE Trans. Geosci. Remote Sens. 33:457-465.

Los, S. O., Justice, C. J., and Tucker, C. J. (1994), A global 10by 10 NDVI data set for climate studies derived from theGIMMS continental NDVI data. Int. J. Remote Sens. 15:3493-3518.

Myneni, R. B. (1991), Modeling radiative transfer and photosyn-thesis in three-dimensional vegetation canopies. Agric. For.Meteorol. 55:323-344.

Myneni, R. B., and Williams, D. L. (1994), On the relationshipbetween FAPAR and NDVJ. Remote Sens. Environ. 49:200-211.

Myneni, R. B., Marshak, A. L., Knyazikhin, Y., and Asrar, G.(1991), Discrete ordinates method for photon transport in leafcanopies. In Photon-Vegetation Interactions: Applications inOptical Remote Sensing and Plant Physiology (R. B. Myneniand J. Ross, Eds.), Springer-Verlag, Heidelberg, pp. 45-109.

Myneni, R. B., Hall, F. G., Sellers, P. J., and Marshak, A. L.(1995), The interpretation of spectral vegetation indices. IEEETrans. Geosci. Remote Sens. 33:481-486.

Myneni, R. B., Nemani, R. R., and Running, S. W. (1997), Estima-tion of global leaf area index and absorbed PAR using radiativetransfer models. IEEE Trans. Geosci. Remote Sens. 35:1380-1393.

Ni, W., Li, X., Woodcock, C. E., Roujean, J. L., and Davis, R.(1997), Transmission of solar radiation in boreal conifer for-ests: measurements and models. J. Geophys. Res. 102(D24):29555-29566.

Ni, W., and Woodcock, C. E. (2000), The effect of canopy struc-ture and the presence of snow on the albedo of conifer forests.J. Geophys. Res. 105(D9):11879-11888.

Potter, C. S., Randerson, J. T., Field, C. B., Matson, P. A.,Vitousek, P. M., Mooney, H. A., and Klooster, S.A. (1993),Terrestrial ecosystem production: a process model based onglobal satellite and surface data. Global Biogeochem. Cycles7(4):811-841.

Privette, J. L. (1994), An efficient strategy for the inversion ofbi directional reflectance models with satellite remote sensingdata, Ph.D. thesis, University of Colorado, Boulder.

Roberts, D. A., Adams, J. B., and Smith, M. 0. (1990), Transmis-

sion and scattering of light by leaves: the effect on spectralmixtures. In Proc. 10th Annual IGARSS, College Park, MD,20-24 May, pp. 1381-1384.

Roujean, J. L., and Breon, F. M. (1995), Estimating PAR absorbedby vegetation from bidirectional reflectance measurements.Remote Sens. Environ. 51:375-384.

Running, S. W., and Coughlan, J. C. (1988), A general model offorest ecosystem processes for regional applications. I. Hydro-logical balance, canopy gas exchange and primary productionprocesses. Ecol. Model 42:125-154.

Schewchuk, S. R. (1997), Surface mesonetfor Boreas.J. Geophys.Res. 102(D24):28,077-29,082.

Sellers, P. J. (1985), Canopy reflectance, photosynthesis and tran-spiration. Int. J. Remote Sens. 6(8):1335-1372.

Sellers, P. J., Berry, J. A., Collatz, G. J., Field, C. B., and Hall,

F. G. (1992a), Canopy reflectance, photosynthesis and transpi-ration. III. A reanalysis using improved leaf models and a newcanopy integration scheme. Remote Sens. Environ. 42:187-216.

Sellers, P. J., Heiser, M. D., and Hall, F. G. (1992b), Relations

between surface conductance and spectral vegetation indicesat intermediate (100 m2 to 15 km2 ) length scales. J. Geophys.

Res Atmos. 97(ND17):19,033-19,059.Shultis, J. K., and Myneni, R. B. (1988), Radiative transfer in

vegetation canopies with anisotropic scattering. J. Quant.Spectrosc. Radiat. Transf 39:115-129.

Spanner, M. A., Pierce, L. L., Peterson, D. L., and Running, S.

W. (1990), Remote sensing of temperate coniferous forestleaf area index: the influence of canopy closure, understoryvegetation and background reflectance. Int. J. Remote Sens.11:95-111.

Tucker, C. J. (1979), Red and photographic infrared linear combi-nations for monitoring vegetation. Remote Sens. Environ. 8:127-150.

Yoshioka, H., Huete, A. R., and Miura, T. (1999), Derivation of

vegetation isoline equation in red-NIR reflectance space.IEEE Trans. Geosci. Remote Sens. 38:838-848.