Synergistic algorithm for estimating vegetation canopy...

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 103, NO. D24, PAGES 32,257-32,276, JANUARY 27, 1998

32257

Synergistic algorithm for estimating vegetation canopy leafarea index and fraction of absorbed photosynthetically activeradiation from MODIS and MISR data

Y. Knyazikhin,1 J.V. Martonchik,2 R.B. Myneni,1 D.J. Diner,2 and S. W. Running3

Abstract. A synergistic algorithm for producing global leaf area index and fraction ofabsorbed photosynthetically active radiation fields from canopy reflectance data measured byMODIS (moderate resolution imaging spectroradiometer) and MISR (multiangle imagingspectroradiometer) instruments aboard the EOS-AM 1 platform is described here. Theproposed algorithm is based on a three-dimensional formulation of the radiative transferprocess in vegetation canopies. It allows the use of information provided by MODIS (singleangle and up to 7 shortwave spectral bands) and MISR (nine angles and four shortwavespectral bands) instruments within one algorithm. By accounting features specific to theproblem of radiative transfer in plant canopies, powerful techniques developed in reactortheory and atmospheric physics are adapted to split a complicated three-dimensional radiativetransfer problem into two independent, simpler subproblems, the solutions of which arestored in the form of a look-up table. The theoretical background required for the design ofthe synergistic algorithm is discussed.

1. Introduction

Large-scale ecosystem modeling is used to simulate arange of ecological responses to changes in climate andchemical composition of the atmosphere, including changes inthe distribution of terrestrial plant communities across theglobe in response to climate changes. Leaf area index (LAI) isa state parameter in all models describing the exchange offluxes of energy, mass (e.g., water and CO2), and momentumbetween the surface and the planetary boundary layer.Analyses of global carbon budget indicate a large terrestrialmiddle- to high-latitude sink, without which the accumulationof carbon in the atmosphere would be higher than the presentrate. The problem of accurately evaluating the exchange ofcarbon between the atmosphere and the terrestrial vegetationtherefore requires special attention. In this context the fractionof photosynthetically active radiation (FPAR) absorbed byglobal vegetation is a key state variable in most ecosystemproductivity models and in global models of climate,hydrology, biogeochemestry, and ecology [Sellers et al.,1997]. Therefore these variables that describe vegetationcanopy structure and its energy absorption capacity arerequired by many of the EOS Interdisciplinary Projects[Myneni et al., 1997a]. In order to quantitatively andaccurately model global dynamics of these processes,differentiate short-term from long-term trends, as well as todistinguish regional from global phenomena, these two

parameters must be collected often for a long period of timeand should represent every region of the Earth’s lands.Satellite remote sensing serves as the most effective means forcollecting global data on a regularly basis. The launch ofEOS-AM 1 with MODIS (moderate resolution imagingspectroradiometer) and MISR (multiangle imagingspectroradiometer) instruments onboard begins a new era inremote sensing the Earth system. In contrast to previoussingle-angle and single-channel instruments, MODIS andMISR together allow for rich spectral and angular sampling ofthe radiation field reflected by vegetation canopies. This setsnew demands on the retrieval techniques for geophysicalparameters in order to take full advantages of theseinstruments. Our objective is to derive a synergistic algorithmfor the extraction of LAI and FPAR from MODIS- and MISR-measured canopy reflectance data, with the flexibility to usethe same algorithm in MODIS-only and MISR-only as well.Although a prototyping of the algorithm with data was also afocus of our activity, these results are not discussed in thisarticle. Plate 1 demonstrates an example of the prototype ofthe MODIS LAI/FPAR data product.

Solar radiation scattered from a vegetation canopy andmeasured by satellite sensors results from interaction ofphotons traversing through the foliage medium, bounded atthe bottom by a radiatively participating surface. Therefore toestimate the canopy radiation regime, three important featuresmust be carefully formulated. They are (1) the architecture ofindividual plant and the entire canopy; (2) optical propertiesof vegetation elements (leaves, stems) and soil; the formerdepends on physiological conditions (water status, pigmentconcentration); and (3) atmospheric conditions whichdetermine the incident radiation field. Photon transport theoryaims at deriving the solar radiation regime, both within thevegetation canopy and the radiant exitance, using the above

___________1Department of Geography, Boston University, Massachusetts.2Jet Propulsion Laboratory, California Institute of Technology.3The School of Forestry, University of Montana, Missoula.

Copyright 1998 by the American Geophysical Union

Paper number 98JD02462.0148-0227/98/98JD-02462$09.00

KNYAZIKHIN ET AL.: SYNERGISTIC MODIS-MISR LAI&FPAR ALGORITHM

32258

a)

b)

Plate 1. (a) Global LAI and (b) FPAR in September-October 1997 derived from SeaWiFs (sea-viewing widefield-of-view sensor) data. This data set includes daily atmosphere-corrected surface reflectances at eightshortwave spectral bands. Surface reflectances at red (670 nm) and near-infrared (865 nm) at 8 km resolutionwere used. The algorithm was applied to daily surface reflectance data for all days from September 18 toOctober 12, 1997. For each pixel, LAI and FPAR values corresponding to the maximum NDVI during thisperiod are shown in these pannels. The look-up table for biome 1 (grasses and cereal crops, Table 1) was usedto produce global LAI and FPAR for all biome types.


32259

mentioned attributes as input data. This theory underliesnumerous canopy radiation models (see, for example, reviewsby Myneni et al. [1989] and Ross et al. [1992]). Usuallyretrieval techniques rely on a model, which providerelationships between measured data and biophysicalparameters. It allows for the design of fast retrievalalgorithms. However, such algorithms can retrieve only thosevariables that are explicitly represented in the canopyradiation models. They exclude the use of a rather wide familyof three-dimensional models in which desired variables maynot be in the model parameter list directly [Ross and Marshak,1984; Myneni, 1991; Borel et al., 1991; Kimes, 1991;Knyazikhin et al., 1996]. They are also based on someassumptions which may not be fulfilled. For example,numerous canopy radiation models presuppose that thecanopy angular reflectance measurements can be performedabout the plane of the solar vertical which providesinformation on the hot spot effect [Kuusk, 1985; Simmer andGerstl, 1985; Marshak, 1989; Verstraete et al., 1990; Myneniet al., 1991]. This suggestion may be appropriate formultiangle instruments such as MISR or POLDER(Polarization and Directionality of the Earth’s Reflectance)[Deschamps et al., 1994]. For the single-angle andmultichannel MODIS instrument, this suggestion is notfulfilled. There is yet another problem encountered when oneincorporates a particular model in the inverse mode. A ratherwide family of canopy radiation models designed to accountfor the hot spot effect conflict with the law of energyconservation (Appendix); that is, they are not “physicallybased” models.

In designing the synergistic algorithm, we cast aside theidea of trying to relate a retrieval technique with a particularcanopy radiation model. Our approach incorporates thefollowing tenets: (1) a retrieval algorithm can use any field-tested canopy radiation model; that is, the retrieval algorithmis model independent; (2) the more measured information isavailable and the more accurate this information is, the morereliable and accurate the algorithm output would be, i.e., con-vergence of the algorithm; (3) the algorithm must be as simpleas the one linked to a particular canopy radiation model; (4)spectral and angular information are synergistically used inthe extraction of LAI and FPAR. Because three-dimensionalmodels include all diversity of one- and two-dimensionalmodels as special cases, property (1) of the algorithm can beachieved, if one formulates the inverse problem for three-di-mensional vegetation canopies: given mean spectral, and inthe case of MISR data, angular signatures of canopy-leavingradiance averaged over the three-dimensional canopyradiation field, find LAI and FPAR. It is clear that the giveninformation is not enough to solve the inverse problem. Forexample, the three-dimensional canopy structure can varyconsiderably with LAI essentially unchanged. Therefore oneneeds to limit the range of variation of the variables deter-mining the three-dimensional radiative regime in plantcanopies. It can be achieved by using a vegetation coverclassification parameterized in terms of variables used by

photon transport theory [Myneni et al., 1997]. It distinguishessix biome types, each representing a pattern of the architectureof an individual tree (leaf normal orientation, stem-trunk-branch area fractions, leaf and crown size) and the entirecanopy (trunk distribution, topography), as well as patterns ofspectral reflectance and transmittance of vegetation elements.The soil and/or understory type are also characteristics of thebiome, which can vary continuously within given biome-de-pendent ranges. The distribution of leaves is described by theleaf area density distribution function which also depends onsome continuous parameters. A detailed description of biometypes is presented in section 2.

The canopy structure is the most important variabledetermining the three-dimensional radiation field invegetation canopies. Therefore section 3 starts with a precisemathematical definition of this variable and how variouscanopy radiation models treat this variable. This allows us tospecify some common properties of the present canopyradiation models. The basic physical principle underlying theproposed LAI/FPAR retrieval algorithm is the law of energyconservation. However, a rather wide family of canopyradiation models (described in the Appendix) conflict withthis law. Therefore the three-dimensional transport equationwhich includes a nonphysical internal source is taken as thestarting point for the derivation of the algorithm. In section 4,a technique developed in atmospheric optics is utilized toparameterize the radiative field in terms of reflectanceproperties of the canopy and ground, as well as to split theradiative transfer problem into two independent subproblems,each of which is expressed in terms of three basic componentsof the energy conservation law: canopy transmittance,reflectance, and absorptance. These components are elementsof the look-up table (LUT), and the algorithm interacts onlywith the elements of the LUT. This provides the requiredindependence of the retrieval algorithm to a particular canopyradiation model. The next important step in achievingproperty (3) is to specify the dependence of canopytransmittance, reflectance, and absorptance on wavelength. Itis precisely derived in section 5; this dependence is describedby a simple function which depends on the unique positiveeigenvalue of the transport equation. The eigenvalue relatesoptical properties of individual leaves to canopy structure.This result not only allows a significant reduction in the sizeof the LUT but also relates canopy spectral reflectance withspectral properties of individual leaves, which is a ratherstable characteristic of green leaves.

In spite of the essential reduction of possible canopyrepresentatives by introducing a vegetation coverclassification, the inverse problem still allows for multiplesolutions. A technique allowing the reduction of nonphysicalsolutions is described in section 6. A definition of the LUT isgiven in this section as well. A method to estimate the mostprobable LAI and FPAR, accounting for specific features ofthe MODIS and MISR instruments, and providingconvergence of the algorithm is discussed in sections 7 and 8.The maximum positive eigenvalue and the unique positive


32260

eigenvector corresponding to this eigenvalue, detailed insection 5, express the law of energy conservation in a compactform. The results of this section allow us to relate theNormalized Difference Vegetation Index (NDVI) to thisfundamental physical principle. Relationships between FPARand NDVI are also used in our algorithm as a backup to theLUT approach, and so we discuss these in section 9.

2. Canopy Structural Types of GlobalVegetation

Solar radiation scattered from a vegetation canopy andmeasured by satellite sensors results from interaction ofphotons traversing through the foliage medium, bounded atthe bottom by a radiatively participating surface. Therefore toestimate the canopy radiation regime, three important featuresmust be carefully formulated [Ross, 1981]. They are (1) thearchitecture of individual plants or trees and the entirecanopy; (2) optical properties of vegetation elements (leaves,stems) and ground; the former depends on physiologicalconditions (water status, pigment concentration); and (3)atmospheric conditions which determine the incident radiationfield. Photon transport theory aims at deriving the solarradiation regime, both within the vegetation canopy andradiant exitance, using the above mentioned attributes as inputdata. This underlies a land cover classification [Myneni et al.,1997] which is compatible with the basic physical principle oftransport theory, the law of energy conservation. Global landcovers can be classified into six types (biomes), depending ontheir canopy structure (Table 1). The structural attributes ofthese land covers can be parameterized in terms of variablesthat transport theory admits as follows.

The heterogeneity of the plant canopy can be described bythe three-dimensional leaf area distribution function uL. Itsvalues at spatial points depend on trunk distribution,topography, stem-trunk-branch area fraction, foliagedispersion, leaf and crown size, and leaf clumping [Myneniand Asrar, 1991; Oker-Blom et al., 1991]. The three-dimensional distribution of leaves determines various modelsto account for shadowing effects [Kuusk, 1985; Li and

Strahler, 1985; Verstraete et al., 1990].The leaf area index LAI is defined as

,)(1

LAI LSS∫⋅

=V

drruYX

(1)

where V is the domain in which a plant canopy is located; XS,YS are horizontal dimensions of V. If the vegetation canopyconsists of Nc individual trees, LAI can be expressed as

,LAI)(1

LAICC

1

L

1∑∫∑

==

⋅==N

k

kk

V

N

k kk pdrru

Sp

k

where Sk is the foliage envelope projection (e.g., crown) of thekth plant or tree onto the ground; pk=Sk/(XS⋅YS) and LAIk is theleaf area index of an individual plant or tree. Thus LAI is

LAI = g⋅LAI 0 ,

where ∑=

=C

1

N

k

kpg is the ground cover and

∑=

⋅=C

1

0 LAI1

LAIN

k

kkpg

is the mean LAI of a single plant or tree. The spatialdistribution of plants or trees in the stand is a characteristic ofthe biome type and is assumed known. For each biome type,the leaf area density distribution function is parameterized interms of ground cover and mean leaf area index of anindividual plant or tree, each varying within given biomespecific intervals [gmin, gmax] and [Lmin, Lmax], respectively.Thus the vegetation canopy is represented as a domain Vconsisting of identical plants or trees in order to numericallyevaluate the transport equation.

To parameterize the contribution of the surface underneaththe canopy (soil and/or understory) to the canopy radiationregime, an effective ground reflectance is introduced, namely,

Table 1. Canopy Structural Attributes of Global Land Covers From the Viewpoint of Radiative Transfer ModelingGrasses and

Cereal Crops Shrubs Broadleaf Crops Savannas Broadleaf Forests Needle Forests

Horizontal heterogeneity no yes variable yes yes yes

Ground cover 100% 20-60% 10-100% 20-40% > 70% > 70%

Vertical heterogeneity

(leaf optics and LAD) no no no yes yes yes

Stems/trunks no no green stems yes yes yes

Understory no no no grasses yes yes

Foliage dispersion minimalclumping

random regular minimalclumping

clumped severeclumping

Crown shadowing no not mutual no no yes mutual yes mutual

Brightness of canopy

ground medium bright dark medium dark dark


32261

.),()(

),(),(

)(

2

b

2

b,b

2eff,

∫∫∫

−

Ω′Ω′′Ω′

Ω′ΩΩ′′ΩΩ′

= +−

πλ

πλλ

π

µπ

µµ

λρdrLq

ddrLR

q (2)

Here Lλ is radiance at a point rb of the canopy bottom; Rb,λ isthe bidirectional reflectance factor of the canopy bottom. Thefunction q is a wavelength-independent configurable functionused to better account for specific features of various biomes,and it satisfies the following condition:

.1)(

2∫

−

=Ω′Ω′π

dq (3)

Note that the effective ground reflectance depends on theradiation regime in the vegetation canopy. It follows from thedefinition that the variation of ρq,eff satisfies the followinginequality:

)4(;)(

),(

max

),()(

),(

min

2

,b

2

beff,2

,b

2

Ω′

ΩΩΩ′

≤

≤Ω′

ΩΩΩ′

∫

∫

+−∈Ω′

+−∈Ω′

q

dR

rq

dR

q

π

µ

λρπ

µ

πλ

π

πλ

π

that is, the range of variations depends on the integratedbidirectional factor of the ground surface only. Thebidirectional reflectance factor of the ground surface Rb,λ andthe effective ground reflectance are assumed to be

horizontally homogeneous; that is, they do not depend on thespatial point rb. The pattern of the effective groundreflectances (ρ1, ρ2, …, ρ11), ρi=ρq,eff(λi), at the MODIS andMISR spectral bands (Table 2), is taken as a parametercharacterizing hemispherically integrated reflectance of thecanopy ground (soil and/or understory) and can varycontinuously within the interval defined by equation (4). Thelower and upper bounds of equation (4) depend on biometype. The set of various patterns of effective groundreflectances is a static table of the algorithm, i.e., element ofthe look-up table. The present version of the look-up tablecontains 25 patterns of effective ground reflectances evaluatedfrom the soil reflectance model of Jacquemoud et al. [1992],using model inputs presented by Baret et al. [1993]. Figure 1demonstrates spectral ground reflectances ρq,eff for biome 1

Table 2. MODIS and MISR Spectral Bands

Bands

Center of Spectral

Band, nm Instrument

1 648 MODIS2 858 MODIS

3 470 MODIS

4 555 MODIS

5 1240 MODIS

6 1640 MODIS

7 2130 MODIS

1 446 MISR

2 558 MISR

3 672 MISR

4 866 MISR

Figure 1. Spectral effective ground reflectance for 25 different soils. It includes three soil types described asmixtures of clay, sand, and peat. Each soil type is characterized by three moisture levels (wet, median, dry)and from two to three soil roughnesses (rough, median, smooth, or rough and smooth). These effectiveground reflectances were evaluated from the soil reflectance model of Jacquemoud et al. [1992] using modelinputs presented by Baret et al. [1993].


32262

(grasses and cereal crops).To account for the anisotropy of the ground surface, an

effective ground anisotropy Sq is used,

,),()(

),(),(

)(

1),(

2

b

2

b,b

,b

∫∫

−

−

Ω′Ω′′Ω′

Ω′Ω′′ΩΩ′

=Ω

πλ

πλλ

µπ

µ

λρ drLq

drLR

rSeffq

q (5)

,0n, bbb <•Ω∈ Vr δ

where nb is the outward normal at point rb. The effectiveground anisotropy Sq depends on the canopy structure as wellas the incoming radiation field. We note the followingproperty:

,1),(

2∫

+

=ΩΩπ

µ drS bq

that is, the integral depends neither on spatial nor on spectralvariables. For each biome type, the effective groundanisotropy is assumed wavelength independent. The six covertypes presented in Table 1 can now be expressed in terms ofthe above introduced variables.

2.1. Biome 1, Grasses and Cereal Crops

Canopies exhibit vertical and lateral homogeneity,vegetation ground cover of about 1.0 (gmin=gmax=1), plantheight generally about a meter or less, erect leaf inclination,no woody material, minimal leaf clumping, and soils ofintermediate brightness. The one-dimensional radiativetransfer model is invoked in this situation. Leaf clumping isimplemented by modifying the projection areas with aclumping factor generally less than 1. The soil reflection isassumed Lambertian; that is, Rb,λ=Rlam,λ. We also set q=1. Theeffective soil reflection and anisotropy then have thesimplified form

ρq,eff(λ)=Rlam,λ , Sq(rb,Ω)=1/π . (6)

2.2. Biome 2, Shrubs

Canopies exhibit lateral heterogeneity, low (gmin=0.2) tointermediate (gmin=0.6) vegetation ground cover, small leaves,woody material, and bright backgrounds. The full three-dimensional (3-D) model is invoked. Hot spot, i.e., enhancedbrightness about the retrosolar direction due to absence ofshadows [Privette et al., 1994], is modeled by shadows caston the ground (no mutual shadowing because ground cover islow). This land cover is typical of semiarid regions withextreme hot (brush) or cold (tundra/taiga) temperature regimesand poor soils. For this biome we represent the bidirectionalsoil reflectance factor Rb,λ as

,),()(),( 0,2,1,b ΩΩ⋅Ω′=ΩΩ′ λλλ RRR (7)

where Ω0 is the direction of the direct solar radiance. We set

.)()( *,1,1 λλ ρΩ′=Ω′ Rq (8)

The effective soil reflection and soil anisotropy then have theform

)()( 0*,2

*,1eff, Ω= λλ ρρλρq , ,

)(

),()(

0*,2

0,2

ΩΩΩ

=Ωλ

λ

ρπR

Sq (9)

where

,)(1

2

,1*,1 ∫

−

Ω′′Ω′=π

λλ µπ

ρ dR

.),(1

)(

2

0,20*,2 ∫

+

ΩΩΩ=Ωπ

λλ µπ

ρ dR

The functions q and Sq are assumed wavelength independentand serve as parameter of this biome. This biome ischaracterized by intermediate vegetation ground cover. Theuse of the above model for the bidirectional soil reflectancefactor means that only the incoming direct beam of solarradiation which reaches the soil can influence the anisotropyof the radiation field in the plant canopy.

2.3. Biome 3, Broadleaf Crops

Canopies exhibit lateral heterogeneity, large variations invegetation ground cover from crop planting to maturity(gmin=0.1, gmax=1.0), regular leaf spatial dispersion,photosynthetically active, i.e., green, stems, and dark soilbackgrounds. The regular dispersion of leaves (i.e., thepositive binomial model) leads to a clumping factor that isgenerally greater than unity. The green stems are modeled aserect reflecting protrusions with zero transmittance. The three-dimensional radiative transfer model is invoked in thissituation. The soil reflection is assumed Lambertian, i.e.,Rb,λ=Rlam,λ. The function q=1. The effective soil reflection andanisotropy are expressed by equation (6).

2.4. Biome 4, Savanna

Canopies with two distinct vertical layers, understory ofgrass, low ground cover of overstory trees (gmin=0.2,gmax=0.4), canopy optics, and structure are therefore verticallyheterogeneous. The full 3-D method is required. The interac-tion coefficients have a strong vertical dependency. Savannasin the tropical and subtropical regions are characterized asmixtures of warm grasses and broadleaf trees. In the coolerregimes of the higher latitudes, they are described as mixturesof cool grass and needle trees. The effective soil reflectionand soil anisotropy then are simulated by equation (9).

2.5. Biome 5: Broadleaf Forests

Vertical and lateral heterogeneity, high ground cover(gmin=0.8, gmax=1.0), green understory, mutual shadowing ofcrowns, foliage clumping, trunks, and branches are included,so the canopy structure and optical properties differ spatially.Mutual shadowing of crowns is handled by modifying the hot


32263

spot formulation. Therefore stand density and crown sizedefine this gap parameter. The branches are randomlyoriented, but tree trunks are modeled as erect structures. Bothtrunk and branch reflectance are specified frommeasurements. For this biome the three-dimensional transportequation is utilized to evaluate the effective soil reflection andanisotropy as a function of LAI and Sun position. These areintermediate calculations and are used to precomputeparameters stored in the LUT.

2.6. Biome 6: Needle Forests

These are canopies with needles, needle clumping onshoots, severe shoot clumping in whorls, dark vertical trunks,sparse green understory, and crown mutual shadowing. This isthe most complex case, invoking the full 3-D method with allits options. A typical shoot is modeled to handle needleclumping on the shoots. The shoots are then assumed to beclumped in the crown space. Mutual shadowing by crowns ishandled by modifying the hot spot formulation. The branchesare randomly oriented but the dark tree trunks are modeled aserect structures. Both trunk and branch reflectance arespecified from measurements. The effective soil reflection andanisotropy are evaluated the same way as for biome 5.

3. Radiative Transfer Problem for VegetationMedia

The domain V in which a vegetation canopy is located, is aparallelepiped of horizontal dimensions XS, YS, and biome-dependent height ZS. The top δVt, bottom δVb, and lateral δVl

surfaces of the parallelepiped form the canopy boundaryδV=δVt+δVb+δVl. The structure of the vegetation canopy isdefined by an indicator function χ(r) whose value is 1, if thereis a phytoelement at the spatial point r, and zero otherwise.Here the position vector r denotes the Cartesian triplet (x,y,z)with (0<x<XS), (0<y<YS), and (0<z<ZS), with its originO=(0,0,0) at the top of the canopy. The indicator function istreated as a random variable. Its distribution function, in thegeneral case, depends on both macroscale (e.g., randomdimension of the trees and their spatial distribution) andmicroscale (e.g., structural organization of an individual tree)properties of the vegetation canopy and includes all three ofits components, absolutely continuous, discrete, and singular[Knyazikhin et al., 1998]. In order to approximate thisfunction, a fine spatial mesh is introduced by dividing thedomain V into Nε nonoverlapping fine cells, ei, i= 1,2, … , Nε,of size ∆x=∆y=∆z. Each realization χ(r) of the canopystructure is replaced by its mean over the fine cell ei as

.,)()()(

1)(L i

ei

erdrmrem

ru

i

∈= ∫ χ (10)

Here m is a measure suitable to perform the integration ofequation (10). The function uL is the leaf area densitydistribution function. In the general case, (10) is the Lebesgueintegral and it may not coincide with an integral in the “true

sense.” This integration technique provides the convergenceprocess uL→χ/m(V) when ε→0 [Knyazikhin et al., 1998], andso equation (10) can be taken as an approximation of thestructure of the vegetation canopy. The accuracy of thisapproximation depends on size ε of the fine cell ei. To ourknowledge, all existing canopy radiation models are based onthe approximation of (10) by a piece-wise continuousfunction, e.g., describing both the spatial distribution ofvarious geometrical objects like cones, ellipsoids, etc., and thevariation of leaf area within a geometrical figure [Ross andNilson, 1968; Nilson, 1977; Ross 1981; Norman and Wells,1983; Li et al., 1995]. Therefore we proceed with thesuggestion that uL is the random value whose distributionfunction is described by a piece-wise continuous function. Foreach realization, the radiation field in such a medium can beexpressed as

)11(.),(),()(

),()(),(),(

4

L

L

Ω′Ω′Ω→Ω′Γ=

ΩΩ+Ω∇•Ω

∫ drLrru

rLrurGrL

πλλ

λλ

π

Here Ω•∇ is the derivative at r along the direction Ω; Lλ is themonochromatic radiance at point r and in the direction Ω,

,),(2

1),( L

2

LLL ΩΩ•ΩΩ=Ω ∫+

drgrG

ππ

is the mean projection of leaf normals at r onto a planeperpendicular to the direction Ω; gL is the probability densityof leaf normal distribution over the upper hemisphere 2π+;

,),,(),(2

1

),(1

2

LL,LLLL∫+

ΩΩ→Ω′ΩΩ•Ω′Ω=

Ω→Ω′Γ

π

λ

λ

γπ

π

drrg

r

is the area-scattering phase function [Ross, 1981], and γL,λ isthe leaf-scattering phase function. Unit vectors are expressedin spherical coordinates with respect to (−Z) axis. It followsfrom the above definitions that the solution of the transportequation is also a random variable. For each biome type, theangular distribution of radiance leaving the top surface of thevegetation canopy is defined to be the mean value, <Lλ>bio, ofLλ over different realizations of the given biome type. Thefollowing definitions of biome-specific reflectances are usedin this paper.

The hemispherical-directional reflectance factor (HDRF)for nonisotropic incident radiation is the ratio of the meanradiance leaving the top of the plant canopy, <Lλ(r t,Ω)>bio,Ω•nt>0, to radiance reflected from an ideal Lambertian targetinto the same beam geometry and illuminated under identicalatmospheric conditions [Diner et al., 1998a]; that is,


32264

0n,n),(

1),(

),( t

2

tt

biot0 >•Ω

Ω′•Ω′Ω′

>Ω<=ΩΩ

∫−π

λ

λλ

πdrL

rLr .

Here nt is the outward normal at points r t∈δVt; <⋅>bio denotesthe averaging over the ensemble of biome realizations; and Ω0

is the direction of the monodirectional solar radiation incidenton the top of the canopy boundary.

The bihemispherical reflectance (BHR) for nonisotropicincident radiation is the ratio of the mean radiant exitance tothe incident radiant [Diner et al., 1998a], i.e.,

.n),(

n),(

)(

2

tt

2

tbiot

0hem

∫∫

−

+

Ω′•Ω′Ω′

Ω•Ω>Ω<

=Ω

πλ

πλ

λdrL

drL

A

In order to quantify a proportion between direct and diffusecomponent of incoming radiation, the ratio fdir(Ω0) of directradiant incident on the top of the plant canopy to the totalincident irradiance is used. If fdir=1, HDRF and BHR becomethe bidirectional reflectance factor (BRF), and the directionalhemispherical reflectance (DHR). Here rλ(Ω,Ω0) and

)( 0hem ΩλA denote, depending on the situation (fdir=1 or fdir≠1),

HDRF and BHR or BRF and DHR.In spite of the diversity of canopy reflectance models, they

can be classified with respect to how the averaging over theensemble of canopy realizations is performed. In terms ofequation (11), this is equivalent to how the averaging ofuL(r)Lλ(r,Ω) is performed. In the turbid medium models, thevegetation canopy is treated as a gas with nondimensionalplanar scattering centers [Ross, 1981]. Such modelspresuppose that

biobioLbioL ),()(),()( Ω=Ω rLrurLru λλ . (12)

As a result, equation (10) is reduced to the classical transportequation [Ross, 1981] whose solution is the mean radiance<Lλ(r,Ω)>bio. This technique allows the design of conservativeradiation transfer models, i.e., models in which the law of en-ergy conservation holds true for any elementary volume. Suchan approach cannot account for the hot spot phenomena be-cause it ignores shadowing effects. This motivated the devel-opment of a family of radiative transfer models based on thefollowing fact: the two events that a point inside a leaf canopycan be viewed from two points r1 and r2 are not independent[Kuusk, 1985]. The mean of uL(r)Lλ(r,Ω) is presented as

,),()(),,(),()(bioLbioL Ω⋅Ω′Ω=Ω rLrurprLru

bio λλ

where p is the bidirectional gap probability [Kuusk, 1985; Liand Strahler, 1985; Verstraete et al., 1990; Oker-Blom et al.,1991]. Such models account accurately for once scatteredradiance, taking Gp<uL> as the extinction coefficient. Forevaluation of the multiply scattered radiance, assumption (12)is usually used [Marshak, 1989; Myneni et al., 1995b]. Thesetypes of canopy-radiation models can well simulate BRFs.

However, they are not conservative (Appendix 1). Theproblem of obtaining a correct closed equation for the meanmonochromatic radiance was formulated and solved byVainikko [1973], where the equations for the mean radiancewere derived through spatial averaging of the stochastictransport equation (11) in a model of broken clouds. Thisapproach was studied in detail by Titov [1990]. Anisimov andMenzulin [1981] utilized similar ideas to describe theradiation regime in plant canopies. The stochastic modelsincorporate the best features of the above mentionedapproaches. The aim of this paper is to derive some generalproperties of radiation transfer which do not depend on aparticular model and which can be taken as the basis of ourLAI/FPAR retrieval algorithm. Equation (11) express the lawof energy conservation in the most general form. Thereforeour aim can be achieved, if this equation is taken as a startingpoint for deriving the desired properties. In order to includecanopy reflectance models with hot spot effect intoconsideration, a transport equation of the form

)13(),(),(),()(

),()(),(),(

4

L

L

Ω+Ω′Ω′Ω→Ω′Γ=

ΩΩ+Ω∇•Ω

∫ rFdrLrru

rLrurGrL

λπ

λλ

λλ

π

will also be considered in this paper. Here Fλ is a functionwhich accounts for the hot spot effect (Appendix).

Equation (13) alone does not provide a full description ofrandom realizations of the radiative field. It is necessary tospecify the incident radiance at the canopy boundary δV i.e.,specification of the boundary conditions. Because the canopyis adjacent to the atmosphere, and neighboring canopies, andthe soil or understory, all which have different reflectionproperties, the following boundary conditions will be used todescribe the incoming radiation [Ross et al., 1992]:

,)()(),,(),( 0ttop

,m0ttop

,dt Ω−Ω+ΩΩ=Ω δλλλ rLrLrL (14)

,0n, ttt <•Ω∈ Vr δ

)15(,)()(),,(

n),(),(1

),(

0llat

,m0llat

,d

0n

ll,ll

l

Ω−Ω+ΩΩ+

Ω′•Ω′Ω′ΩΩ′=Ω ∫>•Ω′

δ

π

λλ

λλλ

rLrL

drLRrL

,0n, lll <•Ω∈ Vr δ

)16(,n),(),(1

),(

0n

bb,b

b

b

∫>•Ω′

Ω′•Ω′Ω′ΩΩ′=

Ω

drLR

rL

λλ

λ

π

,0n, bbb <•Ω∈ Vr δ

where Ltopd,λ and Ltop

m,λ are the diffuse and monodirectionalcomponents of solar radiation incident on the top surface ofthe canopy boundary δVt; Ω0∼(µ0,φ0) is the direction of themonodirectional solar component; δ is the Dirac delta func-


32265

tion; Llatm,λ is the intensity of the monodirectional solar radia-

tion arriving at a point r l∈δVl along Ω0 without experiencingan interaction with the neighboring canopies; Llat

d,λ is the dif-fuse radiation penetrating through the lateral surface δVl; Rl,λ

and Rb,λ (in sr-1) are the bidirectional reflectance factors of thelateral and the bottom surfaces, respectively; and nt, nl, and nbare the outward normals at points r t∈δVt, r l∈δVl and rb∈δVb,respectively. A solution of the boundary value problem, ex-pressed by equations (13)-(16), describes a random realizationof the radiation field in a vegetation canopy.

4. Mathematical Basis of the Algorithm

The aim of this section is to parameterize the contributionof soil/understory reflectances to the exitant radiation field.We closely follow ideas used in atmospheric physics[Kondratyev, 1969; Liou, 1980]. It follows from the linearityof equation (13) that its solution can be represented as thesum

Lλ(r,Ω) = Lbs,λ(r,Ω) + Lrest,λ(r,Ω) . (17)

Here Lbs,λ is the solution of the “black-soil problem” whichsatisfies equation (13) with boundary conditions expressed byequations (14), (15), and

Lbs,λ(rb,Ω) = 0, rb∈δVb, Ω•nb < 0 .

The function Lrest,λ also satisfies equation (13) with Fλ=0 andboundary conditions expressed as

Lrest,λ(r t,Ω) = 0 , r t∈δVt , Ω•nt < 0 ,

)18(,n),(),(1

),(

0n

llrest,,l

lrest,

l

∫>•Ω′


Ω

drLR

rL

λλ

λ

π


)19(,n),(),(1

),(

0n

bb,b

b,rest

b

∫>•Ω′


Ω

drLR

rL

λλ

λ

π

.0n, bbb <•Ω∈ Vr δ

Note that Lrest,λ depends on the solution of the “completetransport problem.” The boundary condition (19) can berewritten as

Lrest,λ(rb,Ω) = ρq,eff(λ)Sq(rb,Ω)Tq,λ., (20)

where ρq,eff, and Sq are defined by (2) and (5), respectively,and

.),()()(

2

bb, ∫−

Ω′′Ω′Ω′=π

λλ µ drLqrTq (21)

The function q is defined by (3). The coefficient ρq,eff is as-sumed to be independent of the point rb. It is taken as the pa-rameter describing the reflectance of the surface underneath

the canopy and can vary continuously within a biome-depend-ent interval (section 2). The biome-dependent function Sq isassumed to be wavelength independent and known (section2). We replace Tq,λ in (20) by its mean value over the groundsurface. This implies that the variable Tq,λ is independent onthe space point rb (this is automatically fulfilled if a one-dimensional radiative transfer model is used to evaluate theradiative field in plant canopies). Taking into accountequation (20), we then can rewrite the solution of thetransport problem, equation (17), as

Lλ(r,Ω) = Lbs,λ(r,Ω) + ρq,eff(λ)Tq,λLq,λ(r,Ω) , (22)

where Lq,λ(r,Ω) satisfies equation (13) with Fλ=0, boundarycondition expressed by equation (18), and

Lq,λ(r t,Ω) = 0, r t∈δVt, Ω•nt < 0 , (23)

Lq,λ(rb,Ω) = Sq(rb,Ω), rb∈δVb, Ω•nb < 0 . (24)

Thus Lq,λ(r,Ω) describes the radiation regime in a plantcanopy generated by anisotropic and heterogeneous sourcesS(rb,Ω) located at the canopy bottom. We term the problem offinding Lq,λ(r,Ω) an “S problem.” Substituting (22) in (21), weget

)()()()( b,,eff,,bsb, rTrTrT qqqbq

q λλλλ λρ r+= , (25)

where

,),()()(

2

b,bsb,bs ∫−


λλ µ drLqrT q

.),()()(

2

b,b, ∫−


λλ µ drLqr qqr

We then average equation (25) over the ground surface. Thisallows us to express Tq,λ via Tq

bs,λ, r q,λ, and ρq,eff. Substitutingthe averaged Tq,λ into equation (22), we get

)26(.),()(1

)(),(

),(

,,bs,eff,

eff,,bs Ω

−+Ω≈

Ω

rLTrL

rL

qq

qq

qλλ

λλ

λ

λρλρ

r

Here qT λ,bs and r q,λ are averages over the canopy bottom. Notethat we can replace the approximate equality in equation (26)by an exact equality if a one-dimensional canopy radiationmodel is used to evaluate the radiative regime. It follows fromequation (26) that the BHR, hem

λA , HDRF, rλ, and the fractionof radiation absorbed by the vegetation, hem

λa , at wavelength λcan be expressed as

)27(,)()(1

)()(

)(

0,hem

,bs,eff,

eff,,0

hem,bs

0hem

Ω−

+Ω≈

Ω

q

qq

qq

A

λλ

λλ

λ

λρλρ

tr

tr


32266

)28(,)()(1

)()(),(

),(

0,hem

,bs,eff,

eff,,0,bs

0

Ω−

Ω+ΩΩ≈

ΩΩ

q

qq

qqr

r

λλ

λλ

λ

λρλπρ

τ tr

)29(,)()(1

)()(

)(

0,hem

,bs,eff,

eff,,0

hem,bs

0hem

Ω−

+Ω≈

Ω

q

qq

qq λ

λλλ

λ

λρλρ

tr

aa

a

where r hembs,λ, a

hembs,λ, and rbs,λ are the BHR, HDRF, and the

fraction of radiation absorbed by the vegetation, respectively,when the canopy ground reflectance is zero. Here

∫−

Ω′Ω′′=Ω

πλ

λλ

µ2

t

,bs0

,hem,bs

),()(

drL

T qqt

is the weighted canopy transmittance,

∫+

Ω′Ω′′=π

λλ µ2

t,, ),( drLqqt

is the transmittance resulting from the anisotropic source Sq

located at the canopy bottom, and

),()( t,, Ω=Ω rLqq λλτ

is the radiance generated by Sq which leaves the top of theplant canopy, and aq,λ is the radiance generated by Sq andabsorbed by the vegetation. The radiation reflected,transmitted, and absorbed by the vegetation must be relatedvia the energy conservation law,

,1)( hem,bs

,hem,bs0,

hem,bs =+Ω+ λλλλ atr q

qk (30)

,)(

)()(

0,hem

,bs

01,hem

,bs0,

Ω

Ω=Ω

≡

q

q

qkλ

λλ

t

t

.1,,, =++ λλλ qqq atr (31)

Note that all the variables in equations (27) and (28) are meanvalues averaged over the top surface of the canopy.

It follows from equation (27) that

)32(.)()(1

)(

)()(

0,hem

,bs,ef,

eff,

0hem

,bs0hem

Ω−

≈

Ω−Ω

q

qfq

qq,

A

λλ

λ

λλ

λρλρ

tr

t

r

Thus the contribution of the ground to the canopy-leavingradiance is proportional to the square of canopy transmittanceand that the factor of proportionality depends on ρq,eff. If theright-hand side is sufficiently small, we can neglect thiscontribution by assigning a value of zero to the effective soilreflectance.

Thus we have parameterized the solution of the transportproblem in terms of ρq,eff and solutions of the “black-soilproblem” and “S problem.” The solution of the “black-soil

problem” depends on Sun-view geometry, canopy architec-ture, and spectral properties of the leaves. The "S problem"depends on spectral properties of the leaves and canopystructure only. At this stage, these properties allow a signifi-cant reduction in the size of the LUT because there is no needto store the dependence of the exiting radiation field onground reflection properties. Since the solution of the “black-soil problem” and “S problem” determine the size of the LUT,we focus on the solution of these problems, using equation(26) as the basis of the algorithm. The next step is to specifythe wavelength dependence of the basic algorithm equation.

5. Spectral Variation of Canopy Absorptance,Transmittance, and Reflectance forConservative Models

Let us consider equation (11) with boundary conditionsexpressed by equations (14)-(16). This boundary value prob-lem can be reduced to the solution of the “black-soil problem”and “S problem.” In the LAI/FPAR retrieval algorithm theboundary conditions (15) for the lateral surface of domain Vare replaced by vacuum condition, i.e., Lλ(r l,Ω)=0 if r l∈δVl

and Ω•nl<0 [Diner et al., 1998b; Knyazikhin et al., this issue].The boundary condition of the “S problem” expressed byequations (18), (23), and (24) are wavelength independent inthis case. The incoming radiation (14) can be parameterized interms of two scalar values: fdir,λ and total flux F0,λ of incomingradiation. It allows representing the “black-soil problem” as asum of two radiation fields. The first is generated by themonodirectional component of solar radiation incident on thetop surface of the canopy boundary and, the second, by thediffuse component. Dividing the transport equations andboundary conditions which define these problems by fdir,λF0,λ

and (1-fdir,λ)F0,λ, one can reduce them to transport problemswith wavelength-independent boundary conditions. Thus thespectral variation of the radiative field in vegetation canopiescan be described, when the spectral variation of the solutionof the transport equation with wavelength-independentboundary conditions is known. Therefore we consider thefollowing boundary value problem for the transport equation

)33(,),(),(

),(),(),(

4,s Ω′Ω′∫ Ω→Ω′=

ΩΩ+Ω∇•Ωdrr

rrr

πλλ

λλ

ϕσϕσϕ

.0n,,),(),( <Ω•∈Ω=Ω rVrrBr δϕλ (34)

Here B is a wavelength-independent function defined on thecanopy boundary δV, and nr is the outward normal at the pointr∈δV. Differentiating equations (33) and (34) with respect towavelength λ, we get

)35(,),(),(

),(),(),(

4

,s Ω′Ω′Ω→Ω′=

ΩΩ+Ω∇•Ω

∫ drrd

d

rurru

πλλ

λλ

ϕσλ

σ

,0n,,0),( <Ω•∈=Ω rVrru δλ (36)


32267

where

.),(

),(λ

ϕλ d

rdru

Ω=Ω

The following results from eigenvector theory are required toderive a relationship between spectral leaf albedo and canopyabsorptance, transmittance, and reflectance.

An eigenvalue of the transport equation is a number γsuch that there exists a function ϕ which satisfies

[ ])37(,),(),(

),(),(),(

4

,s∫ Ω′Ω′Ω→Ω′=

ΩΩ+Ω∇•Ω

πλ ϕσ

ϕσϕγ

drr

rrr

with boundary conditions

ϕ(r,Ω)=0 , r ∈ δV = δVt+δVb+δVl , nr•Ω < 0 .

The function ϕ(r,Ω) is termed an eigenvector correspondingto the given eigenvalue γ.

The set of eigenvalues γk, k=0,1,2, … and eigenvectorsϕk(r,Ω), k=0,1,2, … of the transport equation is a discrete set[Vladimirov, 1963]. The eigenvectors are mutuallyorthogonal; that is,

lklk

V

drdrrr ,

4

),(),(),( δϕϕσπ

=ΩΩΩΩ∫∫ (38)

where δk,l is the Kroneker symbol. The transport equation hasa unique positive eigenvalue which corresponds to a uniquepositive (normalized in the sense of equation (38))eigenvector [Germogenova, 1986]. This eigenvalue is greaterthan the absolute magnitudes of the remaining eigenvalues.This means that only one eigenvector, say ϕ0, takes onpositive values for any r∈V and Ω. This positive couplet ofeigenvector and eigenvalue plays an important role intransport theory, for instance, in neutron transport theory. Thispositive eigenvalue alone determines if the fissile assemblywill function as a reactor, or as an explosive, or will melt. Itsvalue successfully relates the reactor geometry to theabsorption capacity of the active zone. Because the reactor iscontrolled by changing the absorption capacity of the activezone (by inserting or removing absorbents), this value iscritical to its functioning. The similarity to the problem athand is that we need to relate canopy architecture (“similar” toreactor geometry) with leaf optical properties (“similar” to theabsorption capacity of the active zone). The expansion of thesolution of the transport equation in eigenvectors has mainly atheoretical value because the problem of finding these vectorsis much more complicated than finding the solution of thetransport equation. However, this approach can be useful ifwe want to estimate some integrals of the solution. Thereforewe apply this technique to derive a relationship betweenspectral leaf albedo and canopy absorptance, transmittance,and reflectance.

Equation (35) with boundary conditions (36) is a linearhomogeneous differential equation with respect to λ in a

functional space [Krein, 1972]. Its solution ϕ can be expandedin eigenvectors,

∑∞

=

Ω+Ω=Ω1

00 ),,()(),,()(),(k

kk rarar λϕλλϕλϕλ , (39)

where coefficients ak do not depend on spatial or angularvariables. Here we separate the positive eigenvector ϕ0 intothe first summand. As described above, only this summand,a0ϕ0, takes on positive values for any r∈V and Ω. Substituting(39) into equation (35), we get

[ ]

)40(,),,(),(

),,(),(),,(

4

,

0

0

∫∑

∑

Ω′Ω′Ω→Ω′=

ΩΩ+Ω∇•Ω

∞

=

∞

=

πλ λϕσ

λ

λσλ

drard

d

rurru

kks

k

k

kk

where uk=d(akϕk)/dλ. Substituting (37) into (40), furtherresults in

[ ]

[ ] .0)(

),,()(),,()(1

),(0

=

Ω−Ω−×

Ω+∇•Ω∑∞

=

λλγ

λϕλλλγ

σ

d

draru

r

kkkkk

k

Here γk(λ) is the eigenvalue corresponding to the eigenvectorϕk. It follows from this equation, as well as from theorthogonality of eigenvectors, that

[ ] [ ]),,()()(1

)(),,()(

Ω−

=Ω

rad

d

d

radkk

k

k

kk λϕλλγ

λλγ

λλϕλ

.

Solving this ordinary differential equation results in

[ ].),,()()(1

)(1),,()( 00

0 Ω−−

=Ω rara kkk

kkk λϕλ

λγλγ

λϕλ (41)

Thus if we know the kth summand of the expansion inequation (39) at a wavelength λ0, we can easily find thissummand for any other wavelength.

We introduce e, the monochromatic radiation atwavelength λ intercepted by the vegetation canopy,

,),(),()(

4∫∫ ΩΩΩ=π

λϕσλ rrddr

V

e (42)

and e0 as

.),,(),(),()(

4

00 ∫∫ ΩΩ⋅ΩΩ=π

λ λϕϕσλ drdrrr

V

e (43)

Given e, we can evaluate the fraction a of radiationabsorbed by the vegetation at the wavelength λ as

a(λ) = [1-ω(λ)]e(λ) , (44)

where


32268

0 .1

0 .2

0 .3

0 .4

0 .5

0 .6

0 .7

0 .8

0 .9

1

0 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6 0 .7 0 .8

Frac

tion o

f ene

rgy a

bsor

bed

L ea f a lbe do

U n ifo rm le a ve s

L AI=1.1L AI=5.1L AI=9.1

0 .1

0 .2

0 .3

0 .4

0 .5

0 .6

0 .7

0 .8

0 .9

1

0 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6 0 .7 0 .8

Frac

tion o

f ene

rgy ab

sorbe

d

L ea f a lb ed o

P la n op hile lea ves

L A I= 1 .1L A I= 5 .1L A I= 9 .1

Figure 2. Spectral variation of fraction of absorbed radiation by vegetation for uniform (left) and planophile(right) leaves evaluated with canopy radiation model (points) and from equation (46).

Figure 3. Spectral variation canopy transmittance for uniform leaves evaluated with canopy radiation model(points) and from equation (47) for LAI=1.1 (left) and 4.1 (right).

0 .0 4

0 .0 6

0 .0 8

0 .1

0 .1 2

0 .1 4

0 .1 6

0 .1 8

0 .2

0 .2 2

0 .2 4

0 .1 0 .2 0 .3 0 .4 0 .5 0 .6 0 .7 0 .8 0 .9

Cano

py tr

ansm

ittanc

e

L ea f a lb ed o

U n ifo rm lea ves

L A I= 4 .1

0 .3 8

0 .4

0 .4 2

0 .4 4

0 .4 6

0 .4 8

0 .5

0 .5 2

0 .5 4

0 .5 6

0 .5 8

0 .6

0 .1 0 .2 0 .3 0 .4 0 .5 0 .6 0 .7 0 .8 0 .9

Cano

py tr

ansm

ittanc

e

L ea f a lb ed o

U n ifo rm lea ves

L A I= 1 .1


32269

),(

),(1

)( 4

Ω′

ΩΩ→Ω′Γ

=∫

rG

dr

πλ

πλω (45)

is the leaf albedo. Below an estimation of e0 will beperformed. This value is close to e. We skip a precisemathematical proof of this fact here. An intuitive explanationis as follows: Putting (39) in (42) and integrating the seriesresults in only the positive term containing a0ϕ0. As a result,e(λ)/e(λ0)≈e0(λ)/e0(λ0). Let us derive the dependence of e onwavelength. Substituting equation (39) into equation (43) andtaking into account equation (41) as well as the orthogonalityof eigenvectors, equation (38), we obtain

,)()(1

)(1)( 00

0

000 λ

λγλγ

λ ee−−

=

where γ0 is the positive eigenvalue corresponding to thepositive eigenvector ϕ0. Taking into account equation (44), wecan also derive the following estimation for a:

.)()(1

)(1

)(1

)(1)( 0

00

00 λλωλω

λγλγ

λ aa−−⋅

−−

= (46)

Thus given canopy absorptance at wavelength λ0, we canevaluate this variable at any other wavelength. Figure 2 showsspectral variation of the fraction of energy absorbed by thevegetation canopy a for uniform and planophile leaves. Equa-tion (46) can also be used to specify the accuracy of a canopyradiation model to simulate the radiative field in the canopy.On can see (Figure 2, right) that our radiation model is errone-ous in the case of planophile leaves when LAI>5 and the leafalbedo ω>0.5. At a given wavelength, a is a function of can-opy structure and Sun position in the case of “black-soil prob-lem,” and a function of canopy structure only in the case ofthe “S problem.” We store a at a fixed wavelength λ0 in theLUT.

A somewhat more complicated technique is realized toderive an approximation for canopy transmittance,

,)(

,)(1

)(1

)(, ,

00

00,

−−

=

λω

λλγλγ

λωλ λλ DD rr

tt (47)

where rD,λ is the spectral reflectance of the leaf element. Theratio rD,λ/ω(λ) is assumed to be constant with respect towavelength for each biome. Thus given the canopytransmittance at wavelength λ0, we can evaluate this variablefor wavelength λ. Figure 3 shows spectral variation of canopytransmittance for uniform leaves evaluated with our canopyradiation model and with equation (47). At a fixedwavelength, t is a function of canopy structure and Sunposition in the case of the “black-soil problem,” and afunction of canopy structure in the case of the “S problem.”We store t at a fixed wavelength λ0 in the LUT.

The canopy reflectance r is related to the absorptance andtransmittance via the energy conservation principle

r (λ) = 1 - t(λ) - a(λ) . (48)

Thus given canopy transmittance and absorptance at a fixedwavelength, we can obtain the canopy reflectance for anywavelength. Figure 4 demonstrates an example of equation(48).

The unique positive eigenvalue γ0, corresponding to theunique positive eigenvector, can be estimated as [Knyazikhinand Marshak, 1991]

0

0 .0 5

0 .1

0 .1 5

0 .2

0 .2 5

0 .3

0 .3 5

0 .4

0 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6 0 .7 0 .8

Direc

tiona

l hem

isphe

rical

reflec

tance

L ea f a lb ed o

U n ifo rm lea ves

L A I= 1 .1L A I= 4 .1

Figure 4. Spectral variation of the DHR for uniform leavesevaluated with canopy radiation model (points) and fromequation (48) for LAI=1.1 and 4.1.

0

0 .2

0 .4

0 .6

0 .8

1

1 .2

1 .4

1 .6

1 .8

2

0 1 2 3 4 5 6 7 8 9 1 0

Coeff

icien

t K

L ea f a re a in de x

U n ifo rm lea ves

A b s o rbt ion

Figure 5. Coefficient K as a function of LAI for canopyabsorptance.


32270

γ0(λ) = ω(λ)[1 - exp(-K)] , (49)

where K is a coefficient which may depend on canopystructure (i.e., biome type, LAI, ground cover, etc.) and Sunposition but not on wavelength or soil type. Its specificationdepends on the parameter (absorptance or transmittance) andtype of transport problem (“black-soil problem” or “Sproblem”). The coefficient K, however, does not depend onthe transport problem and sun position, when it refers tocanopy absorptance. Figure 5 shows the coefficient K for the“S problem” and canopy absorptance as a function of LAI.This coefficient is an element of the LUT. Note that theeigenvalue γ0 depends on values of spectral leaf albedo (45)which, in turn, depends on wavelength. It allows us toparameterize canopy absorptance, transmittance, andreflectance in terms of canopy structure, Sun position and leafalbedo.

6. Constraints on Look-Up Table Entries

In spite of the diversity of canopy reflectance models,their direct use in an inversion algorithm is ineffective. In thecase of forests, for example, the interaction of photons withthe rough and rather thin surface of tree crowns and with theground in between the crowns are the most important factorscausing the observed variation in the directional reflectancedistribution. These phenomena are rarely captured by manycanopy reflectance models. As a result, these models are onlyslightly sensitive to the within-canopy radiation regime. Thisassertion is based on the fact that a rather wide family ofcanopy radiation models are solutions to (13), includingmodels with a nonphysical internal source Fλ (Appendix).Within such a model the sum of radiation absorbed,transmitted, and reflected by the canopy are not equal to theradiation incident on the canopy. The function Fλ is chosensuch that the model simulates the reflected radiation fieldcorrectly, i.e., these models account for photon interactionswithin a rather small domain of the vegetation canopy. On theother hand, it is the within-canopy radiation regime that isvery sensitive to the canopy structure and therefore to LAI.The within-canopy radiation regime also determines theamount of solar energy absorbed by the vegetation. Ignoringthis phenomenology in canopy radiation models leads to alarge number of nonphysical solutions when one inverts acanopy reflectance model. Therefore (27) and (28) must betransformed before they can be used in a retrieval algorithm.

Let us introduce the required weights

,1),(,)(

),(),(

2

0,bs0

hem,bs

0,bs1

0,bs =ΩΩΩΩ

ΩΩ=ΩΩ ∫

+

−

πλ

λ

λλ µ

πdw

rw

r

(50)

,1)(,)(

)(

2,

, =ΩΩΩ

=Ω ∫+π

λλ

λλ µ

τdww q

q

qq

t (51)

With this notation, (28) can be rewritten as

)()(1

)()(

),(

0,hem

,bs,eff,

eff,,0

hem,bs,bs

0

Ω−

+Ω≈

ΩΩ

q

qq

qq

qww

r

λλ

λλλλ

λ

λρλρ

ππ tr

tr

(52)

and from (30) and (31), the canopy reflectances hem,bs λr and r q,λ

can be written as

hem,bs

1,hem,bs

hem,bs 1 λλλ atr −−= ≡q , (53)

λλλ ,,, 1 qqq atr −−= . (54)

Thus (52) is sensitive to both factors determining thedirectional reflectance distribution of plant canopies (theweight wbs,λ) and to the within-canopy radiation regime[ 1,hem

,bs

≡q

λt , hem,bs λa , tq,λ, aq,λ]. Equations (52)-(54) also allow the

formulation of a test for the “eligibility” of a canopy radiationmodel to generate the LUT. First the weights wbs,λ areevaluated as a function of Sun-view geometry, wavelength,and LAI by using a field-tested canopy reflectance model.Then with the same model, r hem

bs,λ and λ,qr are evaluatedfrom (53) and (54), and inserted into (52). A canopy radiationmodel is “eligible” to generate the LUT file if (50) and (51)are satisfied to within a given accuracy for any Sun-viewcombination, wavelength, and LAI. We do not know of acanopy reflectance model which can pass the above test. Thatis because there is no published model thus far which satisfiesthe energy conservation law. Although a conservativetransport equation for a vegetation canopy has not yet beenformulated, solutions of this equation satisfy propertiesderived in a previous section. These properties can be used tocorrect existing canopy radiation models for the “eligibility”to generate the LUT. An algorithm to correct a canopyradiation model is presented by Knyazikhin et al. [this issue]which was used to generate the LUT for the MISR LAI/FPARretrieval algorithm.

It follows from (32) and (52) that the HDRF can berepresented as

.)]()([)(),( 0hem

,bs0hem

0hem

,bs,bs0 Ω−Ω+Ω≈ΩΩ λλλλλλ π rr Awwr q

(55)

For each pixel the MISR instrument provides the spectralBHR and DHR. Therefore this expression is used to evaluatethe HDRF and BRF in the case of MISR data, setting retrieved

hemλA in (55). Equation (28) is used to evaluate the BRF in the

case of MODIS data.Thus the BHR described by (27) and the HDRF described

by (55) can be expressed in terms of optical properties of aleaf and the energy conservation law, as well as in terms ofsolutions of the “black-soil problem” and “S problem” at areference leaf albedo value of ω(λ0). This facilitatescomparison of spectral values of the BHR or HDRF withspectral properties of individual leaves, which is a ratherstable characteristic of a green leaf. It also can be interpretedas “inclusion of additional information” into the algorithm,


32271

thus allowing a significant reduction in the number ofretrieved solutions. Canopy transmittances and absorptances,and coefficients p=1-exp(-K) where K is defined by (49) forthe “black-soil problem” and “S problem” at a reference leafalbedo value of ω* as well as the weights (50) and (51) areprecomputed and stored in the LUT. It allows the use of thesame LUT for MODIS and MISR instruments. A detaileddescription of such a LUT is presented by Diner et al.[1998b].

7. LAI Retrieval From MODIS and MISR Data

For each pixel the MODIS instrument can provide atmos-phere-corrected BRF in one view direction and at seven bandsin the solar spectrum every day [Vermote et al., 1995]. TheMISR instrument covers the whole globe within 8 days. Foreach pixel, it provides the HDRF, BHR, BRF, and DHR innine view directions and at four spectral bands of solar spec-trum [Diner et al., 1998a; Martonchik et al., 1998]. Thusevery 8 days, one has the set of pixel reflectances correspond-ing to 16 different Sun positions, 15 view angles, and at 11spectral bands. These canopy reflectances and Sun-view ge-ometry are input for the algorithm. Note that this is the maxi-mum amount of information which may be available. Inreality, however, it may be less, e.g., because of cloud coverand performance of preprocessing algorithms. Let r0,λ(Ω′,Ω′0)be the BRF retrieved from MODIS data and rλ(Ω,Ω0) and

)( 0hem ΩλA be the BRF and BHR retrieved from MISR data.

Here Ω′ and Ω are the view MODIS and MISR directions, Ω′0and Ω0 are the direction of direct solar radiation during timesof MODIS and MISR observations, and β, λ denote the centerof the MODIS and MISR spectral bands, respectively. Theseretrieved reflectances are the input for the algorithm which weexpress in the vector-matrix form as

,

),(),(),(

),(),(),(

),(),(),(

8,08,02,02,01,01,0

8,08,02,02,01,01,0

8,087,02,02,01,01,0

0

777

222

1`1

Ω′Ω′Ω′Ω′Ω′Ω′

Ω′Ω′Ω′Ω′ΩΩ′Ω′Ω′Ω′Ω′Ω′Ω′

=

βββ

βββ

βββ

rrr

rrr

rrr

r

,

),(),(),(),(

),(),(),(),(

),(),(),(),(

)(

094093092091

024023022021

014013012011

0

ΩΩΩΩΩΩΩΩ

ΩΩΩΩΩΩΩΩΩΩΩΩΩΩΩΩ

=

Ω

λλλλ

λλλλ

λλλλ

rrrr

rrrr

rrrr

r

[ ].)()()()(

)(

0hem40

hem30

hem20

hem1

0hem

ΩΩΩΩ=

Ω

λλλλ AAAA

A&

Here βk, k=1, 2, … ,7 and λm, m=1,2,3,4 are centers of theMODIS and MISR spectral bands listed in Table 2. We willuse r0,λ(Ω,Ω0), rλ(Ω,Ω0), )( 0

hem ΩλA , 0r , )( 0Ωr , andA&

hem(Ω0) to denote modeled canopy reflectances (i.e.,

evaluated from equation (52) for MODIS and equations (55)and (27) for MISR instruments) and ),(~ 00 ΩΩλr ,

),(~ 0ΩΩλr , A~ hem

λ(Ω), r~ 0, r~ (Ω0), and A&~ hem(Ω0) to denote

observations of these variables.To establish relationships between measured surface re-

flectances and canopy structure, we introduce the space ofcanopy realization P. This space is represented by canopystructural types of global vegetation (biome), each represent-ing patterns of the architecture of an individual tree and theentire canopy, and spectral leaf albedo (45) at MODIS andMISR bands. Each biome is characterized by ground cover g,mean LAI of an individual tree L, and pattern of effectiveground reflectances (ρ1, ρ2, …, ρ11) in the MODIS and MISRbands (section 2). The element p of this space is the vectorp=(bio, ω1, ω2, … , ω11, ρ1, ρ2, … , ρ11, L, g). Here bio cantake six values only; one pattern (ω1, ω2, … ,ω11) of the spec-tral leaf albedo per biome. Ground cover, the LAI ofindividual vegetation, and effective ground reflectance canvary within given biome-dependent ranges (section 2). Thusthe space of canopy realization is supposed to representpatterns of existing vegetation canopies. The set P is the sumof six biome-dependent subsets; that is,

6

1=

=bio

bioPP .

The element of Pbio is the vector (ρ1, ρ2,…, ρ11, L, g).For each biome type, the modeled reflectances 0r ,

)( 0Ωr , and A&

hem(Ω0) are functions of p. In order tocharacterize the closeness between modeled and retrievedreflectances, the following merit functions are introduced

[ ] ,

),(

),(

),(~),(),(

~,8

1

0

7

1

8

1

2

0

,0,00

7

1000

∑∑

∑∑

==

==

Ω′Ω′−Ω′Ω′

=∆

ji

j

jjjj

l

jl

jl

rrjl

rr

ll

ν

σν ββ

[ ]

,

),(

),(

),(~),(),(

)(~),(

9

1

4

1

9

1

200

4

1

00

∑∑

∑∑

==

==

ΩΩ−ΩΩ

=

ΩΩ∆

j

r

i

j r

jjr

l

r

jl

jl

rrjl

rr

ll

ν

σν λλ

)56(.

)(

)(

)(~

)()(

)(~

),(

4

1

4

1

20

hem0

hem

00

∑

∑

=

=

Ω−Ω

=

ΩΩ∆

l

A

l AA

hemhemA

l

l

AAl

AA

ll

ν

σν λλ

&&

The first and second functions characterize the closeness be-tween modeled BRFs and those obtained from MODIS and


32272

MISR data. The third function compares modeled and re-trieved BHRs. Here ν0(l,j) and νr(l,j) take on the value 1 if theBRF at wavelength βl and λl, in Sun-sensor directions(Ω′j,,Ω′0,j) and (Ωj,Ω0), exists, and zero otherwise; νA(l)=1 ifthe BHR at wavelength λl exists, and 0 otherwise; σ0, σr, andσA are uncertainties in the BRFs and BHR retrievals. Thus themerit functions are defined and normalized such that a modelwhich differs from the retrieved canopy reflectance values byan amount equivalent or less than the retrieval uncertainty willresult in values of ∆0, ∆r, and ∆A of the order of unity. In termsof these notations we formulate the inverse problem asfollows: given biome type, bio, and atmosphere correctedcanopy reflectances r~ 0, r~ (Ω0), and A

&~ hem(Ω0) find all p∈Pbio

for which ∆(p)≤h where h is a configurable threshold valueand

[ ] [ ] .)(~

),()(~),(~,)( 00hem

00000

ΩΩ∆+ΩΩ∆+∆=∆ AArrrrp Ar

&&

Any p∈Pbio for which ∆(p)≤h must be considered a candidatefor a true p. Let us introduce a set of candidates for thesolution as

.)( and:);( 0 hpLgLAIPpPLQ biobio ≤∆<⋅∈=

This set is subset of Pbio and contains such p from Pbio forwhich the leaf area index LAI=LAI0⋅g is less than a givenvalue L from the interval [Lmin⋅gmin, Lmax⋅gmax] and ∆(p)≤h. Theset Q(Lmax⋅gmax; Pbio) contains all p∈ Pbio for which a canopyradiation model generates output comparable with measureddata.

In order to quantify acceptable candidates for the solution,we introduce measures (distribution functions) defined on theset Pbio as follows [Knyazikhin et. al., this issue]. The subsetPbio is represented as a sum of nonintersected subsets

jkPPPP jbiokbio

N

k

kbiobio ≠∅===

,, ,,

1

, .

Let N(L;Pbio) be numbers of subsets Pbio,k containing at leastone element from the set Q(L;Pbio). As measures of Q(L;Pbio),we introduce biome specific function Fbio(L) as

.);(

);(lim)(

maxmax bio

bio

Nbio PgLN

PLNLF

⋅=

∞→ (57)

The subset Pbio,k specifies a set of canopy realizations whoserange of variation is “sufficiently small.” N(Lmax⋅gmax;Pbio) istotal number of solutions of ∆(p)≤h; N(L;Pbio) is the numberof these solutions when the leaf area index LAI0⋅g is less thena given value L in the interval [Lmin⋅gmin, Lmax⋅gmax]. Thefunction (57) is the LAI conditional distribution functionprovided p∈Pbio and ∆(p)≤h. Note that the function (57)depends on L, A

~ hemλ(Ω0), r~ 0, r~ (Ω0), and A

&

~ hem(Ω0). Thevalue

∫⋅

⋅

=maxmax

minmin

)(

gL

gL

biobio ldFlL

is taken as solutions of ∆(p)≤h and the value

( )∫⋅

⋅

−=maxmax

minmin

)(22

gL

gL

biobiobio ldFlLd (58)

is taken as the characteristic of the solution accuracy. Biometype bio is expected to be derived from the MODIS land coverproduct. Therefore the synergistic LAI/FPAR algorithm musthave interfaces with MODIS/MISR reflectances product andthe MODIS land cover product. If the inverse problem has nosolutions (i.e., Fbio=0), we assign a default value to (58) and abackup algorithm is triggered to estimate LAI usingvegetation indices [Myneni et al., 1997b]. Plate 1demonstrates an example of prototyping of the LAI/FPARalgorithm with atmospherically corrected SeaWiFS (sea-viewing wide field-of-view sensor) data. The functions νr andνA were set to zero.

Given r~ 0, r~ 0(Ω0), and A&

~ (Ω0), it may be the case thatLAI algorithm admits a number of solutions, covering a widerange of LAI values. When this happens, the retrieved reflec-tances are said to belong to the saturation domain [Knyazikhinet. al., this issue], being insensitive to the various parametervalues of Pbio. Under this condition, the function (57), whichdescribes the number of times a solution has a particular LAIvalue, will appear flat over the range of LAI, illustrating thatthe solutions all have equal probability of occurrence. Herewe skip a description of this situation and how this situationcan be quantified. For details of these results as well as aprecise mathematical investigation of this approach and somenumerical examples illustrating its various aspects, the readeris referred to [Knyazikhin et. al., this issue].

8. Description of Synergistic FPAR Retrieval

It follows from (29) and (32) that the fractional amount ofincident photosynthetically active radiation (PAR) absorbedby the vegetation canopy (FPAR) can be evaluated as

)59(,),,(),,(

)()(),(FPAR

00

700nm

nm400

0hem

Ω+Ω=

Ω= ∫pbioQLAIbioQ

depbio

qbs

λλλa

where

,)()(),,(

700nm

nm400

0hem

,0bs ∫ Ω=Ω λλλ deLAIbioQ bsa (60)

∫ −Ω=

Ω700nm

nm400

,hem,bs

,eff,

eff,0,

0

)61()()(1

)()(

),,(

λλλρλρ

λλ

λ de

pbioQ

q

qq

qq

q

tr

a


32273

[ ] .)()()(~

)(

)(700nm

nm400

0hem

,bs0hem

0,

0,∫ Ω−ΩΩΩ

= λλλλλ

λdeA

q

qr

t

a (62)

The Qbs term describes the absorption within the canopy for ablack-soil condition, and Qq term describes the additionalabsorption within the canopy due to the interaction betweenthe ground (soil and/or understory) and the canopy. Herep∈Pbio; e is the ratio of the monochromatic flux incident at thetop surface of the canopy boundary to the total downwardPAR flux which can be expressed as

,

)(

)()(

700nm

nm400

0hem

,0

0hem

,0

∫ Ω

Ω=

λ

λ

λλ

λλ

deE

eEe

where E0,λ is the solar irradiance spectrum that is known forall wavelengths; hem

λe is the normalized incident irradiancedefined as the ratio of the radiant incident on the surface toE0,λ [Diner et al., 1998a]. The mean over those p∈Pbio whichpassed the test ∆(p)≤h is taken as the estimate of FPAR, i.e.,

,),(FPAR1

FPAR1

∑=

=PN

kPbio pbio

N

where NP is the number of canopy realizations p∈Pbio passingthis test. When there is no solution (i.e., Fbio=0), the algorithmdefaults to a NDVI-FPAR regression analysis to obtain anestimate of FPAR [Myneni et al, 1997b].

The normalized incident irradiance and the BHR are pro-vided by the MISR instrument at three spectral bands withinthe PAR region. We assume a piece-wise linear variation inthese variables over regions [446, 558 nm], [558, 672 nm],and a constant over regions [400, 446 nm], [672, 700 nm].Substituting these piece-wise linear functions into (59) and(62), one can express FPAR as a function of hem

λe and A~ hem

λ

[Diner et al., 1998a]. Note that the dependence of FPAR onground reflection properties is included in A

~ hemλ which is

provided by the MISR instrument; that is, expression (59) is afunction of the biome type, Sun position, ground cover, meanleaf area index of an individual plant, and retrieved BHR.

If only MODIS observations are available for a givenpixel or the MODIS-only mode is executed, e(λ) isapproximated by

,

)5200(

)5200()(

nm700

nm400

0

0

∫=

KE

KEe

λ

λλ

where Eλ(T) is the Planck function [Kondratyev, 1969, p.230]. In this case, the Qq term is a function of the biome type,Sun position, ground cover, mean leaf area index of an indi-vidual plant, and pattern of the effective ground reflectance.Expression (61) is used to evaluate this term. The Qbs and Qq

terms are precomputed and stored in the look-up table.

9. Theoretical Basis of NDVI-FPAR Relations

The measured spectral reflectance data are usuallycompressed into vegetation indexes. More than a dozen suchindexes are reported in the literature and shown to correlatewell with vegetation amount [Tucker, 1979], the fraction ofabsorbed photosynthetically active radiation [Asrar et al.,1984], unstressed vegetation conductance and photosyntheticcapacity [Sellers et al., 1992], and seasonal atmosphericcarbon dioxide variations [Tucker et al., 1986]. There aresome theoretical investigations to explain these empiricalregularities [Vygodskaya and Gorshkova, 1987; Myneni et al.,1995a; Verstraete and Pinty, 1996]. Results from the previoussection allow us to relate the vegetation indexes to thefundamental physical principle, i.e., the law of energyconservation. Here we consider the normalized differencevegetation index (NDVI) whose use is included in theLAI/FPAR retrieval algorithm.

Let us consider NDVI defined as

,NDVIhemhem

hemhem

βα

βα

AA

AA

+

−= (63)

where hemλA is the BHR or DHR, and α and β are near-IR

and red spectral wavebands, respectively. These variables area function of Sun position Ω0, but this dependence has beensuppressed in the notation of this section. For the sake ofsimplicity, we consider the NDVI for the “black-soil” problemand “S problem.” It follows from equations (48), (47), and(46) that equation (63) can be rewritten as

,)(),()(),()(2

)(),()(),(NDVI

ββαββαβββαββα

tarta

mk

mk

−+−= (64)

where

,)(1

)(1

)(1

)(11),(

,0

,0

βωαω

αγβγ

βα−−⋅

−−

−=a

ak

.1)(1

)(1),(

,0

,0 −−−

=αγβγ

βαt

tm

Here γ0,a and γ0,t are defined by equation (49) with K=Ka (forcanopy absorptance) and K=Kt (for canopy transmittance),respectively. Here the ratio between the leaf spectralreflectance and the leaf albedo is assumed to be constant withrespect to wavelength, and so it is excluded from theargument list of t. After simple transformations, one obtains

( ),,)(NDVI ,, ββθβ rta ss⋅=

where the function θ has the following form

,),(),(2

),(),(),(

xmky

xmkyxs

⋅−+⋅−=βαβα

βαβα

,)(

)(, β

ββ a

tt =s .

)(

)(, β

ββ a

rr =s


32274

Thus NDVI is proportional to the canopy absorptance at thered band. It follows from Eqs. (46) and (64) that

.),()(1

)(1

)(1

)(1

)()(1

)(1

)(1

)(1)(

,,,0

,0

,0

,0

ββθβωλω

λγβγ

ββωλω

λγβγ

λ

rta

a

a

a aa

ss

NDVI

−−

−−

=

−−

−−

=

Let e(λ) be the ratio of monochromatic radiant energy incidenton the top surface of the canopy boundary to the total PARflux. Integrating e⋅a over the PAR region of solar spectrum,we get

FPAR=k⋅NDVI ,

where

( ) .)()(1

)(1

)(1

)(1

,

1700

400 ,0

,0

,,

−−

−−

= ∫ λλλγ

λωβωβγ

θ ββde

ssk

a

a

rt

Thus if the canopy ground is ideally black, FPAR is propor-tional to NDVI. The factor of proportionality k depends on theratios st,β and st,β, the coefficients Ka and Kt, and the leaf al-bedo at the red and near-IR spectral bands. A relationship be-tween NDVI and FPAR which accounts for the soil contribu-tion can be derived from equation (27) in a similar manner.Other types of vegetation indexes can be derived in an analo-gous way.

10. Concluding Remarks

This paper presents the theoretical basis of the algorithmdesigned for the retrieval of LAI and FPAR synergisticallyfrom MODIS and MISR data. A three-dimensional formula-tion of the radiative transfer process is used to derive simplebut correct relationships between spectral and angular biome-specific signatures of vegetation canopies and the structuraland optical characteristics of the vegetation canopies. How-ever, these relationships are not directly used to obtain thebest fit with measured spectral and angular canopyreflectances. Accounting for features specific to the problemof radiative transfer in plant canopies, we adopt powerfultechniques developed in nuclear reactor theory andatmospheric physics in the retrieval algorithm. This techniqueallows us to explicitly separate the contribution ofsoil/understory reflectance to the exitant radiation field, torelate hemispherically integrated reflectances to opticalproperties of phytoelements and to split the complicatedradiative transfer problem into several independent simplersubproblems, the solutions of which are precomputed andstored in a form of look-up table, and then used to retrieveLAI and FPAR. The solutions of the subproblems arecomponents of various forms of energy conservation principle(e.g., canopy transmittance and absorptance of a vegetationcanopy bounded by vacuum on all sides). They are determinedfrom general properties of radiative transfer and areindependent of the models used to generate the LUT. Thus we

express the angular and spectral signatures of vegetation cano-pies in terms of the energy conservation principle. It allowsthe design of an algorithm that returns values of LAI andFPAR which provide the best agreement not only to measureddata but which also conform to the energy conservation law.Since the algorithm interacts only with the elements of theLUT, its functioning does not depend on any particularcanopy radiation model. This flexible feature allows the use ofthe best canopy radiation models for the generation of theLUT.

Appendix

A rather wide family of canopy radiation models includethe following steps in their formulation:

1. The attenuation of direct and diffuse incident radiationLλ,0 is evaluated. It satisfies the equation

0),(),(),( 0,0, =ΩΩ+Ω∇•Ω rLrrL λλ σ (A1)

and boundary conditions (14)-(16). The solution of thisboundary value problem can be explicitly expressed in manypractical cases. Here σ is the total interaction cross sectiondefined as

.)(),(),(bio

rurGr LΩ=Ωσ (A2)

2. The upward once-scattered radiation Lλ,1 is evaluated. Itsatisfies the equation

(A3)),(),()(

),(),(),(

4

0,bioL

1,11,

∫ Ω′Ω′Ω→Ω′Γ=

ΩΩ+Ω∇•Ω

πλλ

λλ

π

σ

drLrru

rLrrL

and the vacuum boundary condition; that is,

,0n,,0),(1, <•Ω∈=Ω rVrrL δλ

where nr is the outward normal at point r∈δV. The totalinteraction cross-section σ1 is defined as

,)(),(),,(),(bioL01 rurGrpr ΩΩΩ=Ωσ (A4)

where p is the bidirectional gap probability (section 2). Thisboundary value problem allows for an explicit solution inmany practical situations.

3. The multiply scattered radiance is evaluated by solvingthe transport equation

[ ]∫ Ω′Ω′+Ω′Ω→Ω′Γ=

ΩΩ+Ω∇•Ω

πλλλ

λλ

π

σ

4

1,M,bioL

M,M,

),(),(),()(

),(),(),(

drLrLrru

rLrrL

(A5)with the boundary conditions expressed by

,0n,0),( tM, <•Ω=ΩtrLλ


32275

,n),(),(1

n),(),(1

),(

0n

ll1,,l

0n

llM,,llM,

∫

∫

>•Ω′

>•Ω′

Ω′•Ω′Ω′ΩΩ′+

Ω′•Ω′Ω′ΩΩ′=Ω

l

l

drLR

drLRrL

λλ

λλλ

π

π


,n),(),(1

n),(),(1

),(

0n

bb1,,b

0n

bbM,,bbM,

∫

∫

>•Ω′

>•Ω′

Ω′•Ω′Ω′ΩΩ′+

Ω′•Ω′Ω′ΩΩ′=Ω

b

b

drLR

drLRrL

λλ

λλλ

π

π

.0n, bbb <•Ω∈ Vr δ

The monochromatic radiance is given in such models as

.),(),(),(),( M,1,0, Ω+Ω+Ω=Ω rLrLrLrL λλλλ (A6)

There may be some differences in formulations of thesubproblems 1, 2, and 3. However, all such models have oneproperty in common: the original total interaction cross-section (A2) is replaced by another coefficient (A4) when oneevaluates the distribution of the single-scattered radiationfield. This trick allows the inclusion of the hot spot effect intocanopy radiation models.

Equation (A3) can be rewritten in an equivalent form as

[ ] A7)(,),(),(

),(),()(

),(),(),(

1,1

4

0,bioL

1,1,

λ

πλλ

λλ

σσ

π

σ

Lrr

drLrru

rLrrL

⋅Ω−Ω+

Ω′Ω′Ω→Ω′Γ=

ΩΩ+Ω∇•Ω

∫

It follows from summarizing Eqs. (A1), (A5), and (A7) thatthe radiance (A6) satisfies equation (13) with F defined as

[ ] ,),(),(),( 1,1 λλ σσ LrrrF Ω−Ω=Ω

and boundary conditions expressed by Eqs. (14)-(16). Thussuch models describe radiation regime in a vegetation canopygenerated by incoming radiation and an internal source Fλ.This source appears due to the changes in the extinctioncoefficient when one tries to account for the hot spot effect.

Acknowledgments. This research was carried out byDepartment of Geography, Boston University, under contract withthe National Aeronautics and Space Administration.

ReferencesAnisimov, O. A., and G. V. Menzulin, The problem of modeling the

radiative radiation regime in the plant cover, Sov. Meteorol. andHydrol., 10, 70-74, 1981.

Asrar, G., M. Fuchs, E.T. Kanemasu, and J.L.Hatfield, Estimatingabsorbed photosynthetic radiation and leaf area index fromspectral reflectance in wheat, Agron. J., 76, 300-306, 1984.

Baret, F., S. Jacquemoud, and J.F. Hanocq, The soil concept inremote sensing, Remote Sens. Rev., 7, 65-82, 1993.

Borel, C.C., S.A., Gerstl, and B.J. Powers, The radiosity method inoptical remote sensing of structured 3-D surfaces, Remote Sens.Environ., 36, 13-44, 1991.

Deschamps, P.Y., F.M. Bréon, M. Leroy, A. Podaire, A. Bricaud,J.C. Buriez, and G. Sèze, The POLDER mission: Instrumentcharacteristics and scientific objectives, IEEE Trans. Geosci.Remote Sens., GE-32, 598-615, 1994.

Diner, D.J., J.V.Martonchik, C. Borel, S.A.W. Gerstl, H.R. Gordon,Y. Knyazikhin, R. Myneni, B. Pinty, and M.M. Verstraete, MISR:Level 2 surface retrieval algorithm theoretical basis, JPL InternalDoc. D-11401, Rev. C, Calif. Inst. of Technol., Jet Propul. Lab.,Pasadena, 1998a.

Diner, D.J., W.A. Abdou, H.R. Gordon, R.A. Kahn, Y. Knyazikhin,J.V.Martonchik, S. McMuldroch, R.B. Myneni, and R.A. West,Level 2 ancillary products and data sets algorithm theoreticalbasis, JPL Internal Doc. D-13402, Rev. A, Calif. Inst. of Technol.,Jet Propul. Lab., Pasadena, 1998b.

Germogenova, T.A., The Local Properties of the Solution of theTransport Equation (in Russian), 272 pp., Nauka, Moscow, 1986.

Jacquemoud, S., F. Barret, and J.F. Hanocq, Modeling spectral andbidirectional soil reflectance, Remote Sens. Environ., 41, 123-132, 1992.

Kimes, D.S., Radiative transfer in homogeneous and heterogeneousvegetation canopies, in Photon-Vegetation Interactions:Applications in Plant Physiology and Optical Remote Sensing,pp. 339-388, edited by R. B. Myneni and J. Ross, Springer-Verlag, New York, 1991.

Knyazikhin, Y., and A. Marshak, Fundamental equations of radiativetransfer in leaf canopies and iterative methods for their solution, inPhoton-Vegetation Interactions: Applications in PlantPhysiology and Optical Remote Sensing, pp. 9-43, edited by R. B.Myneni and J. Ross, Springer-Verlag, New York, 1991.

Knyazikhin, Y., J. Kranigk, G. Miessen, O. Panfyorov, N.Vygodskaya, and G. Gravenhorst, Modelling three-dimensionaldistribution of photosynthetically active radiation in slopingconiferous stands, Biomass Bioenerg., 2/3, 189-200, 1996.

Knyazikhin, Y., J. Kranigk, R.B. Myneni, O.Panfyorov, andG.Gravenhorst, Influence of small-scale structure on radiativetransfer and photosynthesis in vegetation cover, J. Geophys. Res.,103, 6133-6144, 1998.

Knyazikhin, Y., J.V. Martonchik, D.J.Diner, R.B. Myneni, M.M.Verstraete, B.Pinty, and N. Gobron, Estimation of vegetationcanopy leaf area index and fraction of absorbed photosyntheticallyactive radiation from atmosphere-corrected MISR data, this issue.

Kondratyev, K. Y., Radiation in the Atmosphere, 912 pp., Academic,San Diego, Calif., 1969.

Krein, S.G (Ed.), Functional Analysis, 352 pp., Foreign Technol.Div., Wright-Patterson Air Force Base, Ohio, 1967.

Kuusk, A., The hot-spot effect of a uniform vegetative cover, Sov. J.Remote. Sens, 3, 645-658, 1985.

Li, X., and A. Strahler, Geometric-optical modeling of a coniferousforest canopy, IEEE Trans. Geosci. Remote Sens., GE-23, 207-221, 1985.

Li, X., A. Strahler, and C. E. Woodcock, A hybrid geometric optical-radiative transfer approach for modelling albedo and directionalreflectance of discontinuous canopies, IEEE Trans. Geosci.Remote Sens., 2, 466-480, 1995.

Liou, K.-N., An Introduction to Atmospheric Radiation, 392 pp.,Academic, San Diego, Calif., 1980.

Marshak, A., Effect of the hot spot on the transport equation in plantcanopies, J. Quant. Spectrosc. Radiat. Transfer, 42, 615-630,1989.

Martonchik, J.V., D.J.Diner, B.Pinty, M.M.Verstraete, R.B.Myneni,Y.Knyazikhin, and H.R.Gordon, Determination of land and oceanreflective, radiative and biophysical properties using multi-angleimaging, IEEE Trans. Geosci. Remote Sens., 36, 1266-1281,1998.

Myneni, R.B., Modeling radiative transfer and photosynthesis inthree-dimensional vegetation canopies, Agric. For. Meteorol., 55,323-344, 1991.

Myneni, R. B., and G. Asrar, Photon interaction cross sections foraggregations of finite dimensional leaves, Remote Sens. Environ.,37, 219-224, 1991.


32276

Myneni, R. B., and D. L. Williams, On the relationship betweenFAPAR and NDVI, Remote Sens. Environ., 49, 200-211, 1994.

Myneni, R.B., J. Ross, and G. Asrar, A review on the theory ofphoton transport in leaf canopies in slab geometry, Agric. For.Meteorol., 45, 1-165, 1989.

Myneni, R.B., A.L. Marshak, and Y. V. Knyazikhin, Transporttheory for a leaf canopy of finite-dimensional scattering centers,Quant. Spectrosc. Radiat. Transfer, 46, 259-280, 1991.

Myneni, R.B., F.G. Hall, P.J. Sellers, and A.L. Marshak, Theinterpretation of spectral vegetation indexes, IEEE Trans. Geosci.Remote Sens., 33, 481-486, 1995a.

Myneni,R.B., S. Maggion, J. Iaquinta, J.L. Privette, N. Gobron, B.Pinty, D.S. Kimes, M.M. Verstraete, and D.L. Williams, Opticalremote sensing of vegetation: Modeling, caveats and algorithm,Remote Sens. Environ, 51, 169-188, 1995b.

Myneni,R.B., C.D. Keeling, C.J. Tucker, G.Asrar, and R.R. Nemani,Increased plant growth in the northern high latitudes from 1981 to1991, Nature, 386, 698-702, 1997a.

Myneni, R.B., R.R. Nemani, and S.W. Running, Estimation of globalleaf area index and absorbed PAR using radiative transfer model,IEEE Trans. Geosci. Remote Sens., 35, 1380-1393, 1997b.

Nilson, T., A theory of radiation penetration into non-homogeneousplant canopies, in The Penetration of Solar Radiation into PlantCanopies (in Russian), pp. 5-70, Estonian Acad. of Sci., Tartu,1977.

Norman, J.M., and J. M. Welles, Radiative transfer in an array ofcanopies, Agron. J., 75, 481-488, 1983.

Oker-Blom, P., J. Lappi., and H. Smolander, Radiation regime andphotosynthesis of coniferous stands, in Photon-VegetationInteractions: Applications in Plant Physiology and OpticalRemote Sensing, pp. 469-499, edited by R. B. Myneni and J.Ross, Springer-Verlag, New York, 1991.

Privette, J. L., R. B. Myneni, W. L. Emery, and C. J. Tucker,Invertibility of a 1D discrete ordinates canopy reflectance model,Remote Sens. Environ., 48, 89-105, 1994.

Ross, J., The Radiation Regime and Architecture of Plant Stands,391 pp., Dr. W. Junk, Norwell, Mass., 1981.

Ross, J.K., and A. L. Marshak, Calculation of the solar radiationreflection from plant cover using the Monte-Carlo method, Sov. J.Remote. Sens., 5, 58-67, 1984.

Ross, J., and T. Nilson, The calculation of photosynthetically activeradiation in plant communities, in Solar Radiation Regime inPlant Stand (in Russian), pp. 5-54, Estonian Acad of Sci. Inst.Phys. Astron., Tartu, 1968.

Ross, J., Y. Knyazikhin, A. Kuusk, A. Marshak, and T. Nilson,Mathematical Modeling of the Solar Radiation Transfer in PlantCanopies (in Russian, with English abstract), 195 pp.,Gidrometeoizdat, St. Peterburg, Russia, 1992.

Sellers, P.J., J.A. Berry, G.J. Gollatz, C.B.Field, and F.G. Hall,Canopy reflectance, photosynthesis and transpiration, III, Areanalysis using improved leaf models and a new canopyintegration scheme, Remote Sens. Environ., 42, 187-216, 1992.

Sellers, P.J., et al., Modeling the exchange of energy, water, andcarbon between continents and atmosphere, Science, 275, 602-509, 1997.

Simmer, C., and S.A. Gerstl, Remote sensing of angularcharacteristics of canopy reflectances, IEEE Trans. Geosci.Remote Sens., GE-23, 648-658, 1985.

Titov, G.A., Statistical description of radiative transfer in clouds, J.Atmos. Sci.,47, 24-38, 1990.

Tucker, C. J., Red and photographic infrared linear combinationformonitoring vegetation, Remote Sens. Environ., 8, 127-150,1979.

Tucker, C.J., Y.Fung, C.D. Keeling, and R.H. Gammon, Relationshipbetween atmospheric CO2 variations and a satellite-derivedvegetation index, Nature, 319, 195-199, 1986.

Vainikko, G.M, Equation for mean intensity in broken clouds (InRussian), Meteorol. Issled., 21, 28-37, 1973.

Vermote, E., L.A. Remer, C.O. Justice, Y. J. Kaufman, and D. Tanré,Atmospheric correction algorithm: Spectral reflectances(MOD09), version 2.0, Algorithm technical backgrounddocument, NASA EOS-ID 2015 Doc., 42 pp., 1995.

Verstraete, M.M., and B. Pinty, Designing spectral indexes forremote sensing applications, IEEE Trans. Geosci. Remote Sens.,34, 1254-1265, 1996.

Verstraete M.M., B. Pinty, and R.E. Dickenson, A physical model ofthe bidirectional reflectance of vegetation canopies, J. Geophys.Res., 95, 11765-11775, 1990.

Vladimirov, V.S., Mathematical problems in the one-velocity theoryof particle transport, Tech. Rep. AECL-1661, At. Energy of Can.Ltd., Chalk River, Ontario, 1963.

Vygodskaya, N.N., and I.I. Gorshkova, Theory and Experiment inVegetation Remote Sensing (in Russian, with English abstract),248 pp., Gidrometeoizdat, St. Petesburg, Russia, 1987.

D.J. Diner, and J.V. Martonchick, Mail Stop 169-237, Jet

Propulsion Laboratory, California Institute of Technology, 4800 OakGrove Drive, Pasadena, CA-91109. (e-mail: [email protected];jvm@ jord.jpl.nasa.gov)

Y. Knyazikhin, and R. B. Myneni, Department of Geography,Boston University, 675 Commonwealth Avenue, Boston, MA022150. (e-mail: [email protected]; [email protected])

S. W. Running, The School of Forestry, University of Montana,Missoula, MT 59812. (e-mail: [email protected])

(Received January 8, 1998; revised May 1, 1998;accepted July 22, 1998)

Synergistic algorithm for estimating vegetation canopy...

Documents

Transcript of Synergistic algorithm for estimating vegetation canopy...