1105 
This formula was constructed with another constraint: namely to minimize the sensitivity of the index to soil 
brightness changes in order to reduce the dependency of the index value on processes not related to the dynamic 
of vegetation. Kaufman and Tanrd (1992) have also suggested a new vegetation index for use with future space 
instruments such as MODIS. Their Atmospherically Resistant Vegetation Index (ARVI) takes advantage of an 
additional channel in the blue region to improve the surface rcpresentativity of the index based on remote 
measurements. The various vegetation indices described above have been used for a great number of 
applications. One property of vegetation canopies is of particular interest: the fractional vegetation cover. 
2- 3 Vegetati on indices and canopy properties 
Two properties of vegetation canopies are of particular interest: the fractional vegetation cover and the 
leaf area index (LAI). The latter is a well defined concept used in agrometeorology and agronomy, it is a 
parameter of great import since it directly controls the amount of photosynthetically active radiation (PAR) 
absorbed by the canopy (e.g. Tucker and Sellers, 1986) and therefore the energy, water and carbon fluxes 
through living plants. 
The concept of fractional vegetation cover is not as well defined, however, because it is dependent on 
the scale of observation. But it is widely used, especially in General Circulation Model (GCMs), to describe the 
proportion of vegetation vs. bare soil present in the environment. In the rest of this paper, we focus on 
fractional vegetation cover, because it is us uall y considered with LAI as essential ingredients to model 
biosphere-atmosphere interactions (e.g. Avissar and Verstraete 1990) and because it has been shown that of all 
the vegetation properties of interest, vegetation indices may be most sensitive to this fractional vegetation cover 
(e.g. Verstraete and Pinty, 1991). 
3- EVALUATION OF INDICES USING SIMULATED SURFACE SPECTRAL DATA 
Five representative vegetation indices have been selected : NDVI because of its wide use, WDVI and 
SAV1 because they are designed specifically to be less sensitive to soil effects, MSAVI because of its 
improvement with regard to the SAVI and GEMI because it was constructed to be less sensitive to soil and to 
atmospheric effects. 
3-1 Overview of the model 
We have chosen to use a simple model to represent heterogeneous surfaces as a straightforward linear 
combination of bare soil and full vegetation canopy. We have then generated data sets for which the relation 
between vegetation indices and the fractional plant cover or leaf area index can be controlled. To this end, we 
have built on the model of Sellers (1985, 1987), which uses a two-stream approximation to represent the 
radiation fluxes in the plant-soil system. The model computes the spectral albedos of a coupled heterogeneous 
soil-vegetation system in the two spectral bands AVHRR by linearly combining the contributions of both bare and 
vegetated areas in each spectral channel (e.g. Hanan et til., 1991; Verstraete and Pinty, 1991). 
3-2-Selectinn of surface conditions 
The optical properties of the leaves in the first two AVHRR channels can be estimated as a function of 
the foliar pigment concentration and the leaf internal structure (Pinty et al., 1993). This approach is based on the 
PROSPECT model which computes these properties on the basis of the chlorophyll and water contant of the 
leaves, as well as an internal structural parameter (Jacquemoud and Baret, 1991). For this study , we have 
chosen a set of parameters typical of a green representative green leaf (See Table 1) and a uniform distribution of 
leaf angle was assumed. Together, the leaf model and the canopy model permit the simulation of the effects of 
changes in plant and canopy properties (in particular the fractional vegetation cover and the LAI) on the 
reflectances (albedos), and hence on the computed vegetation indices. Soil properties were derived from the data 
set published by Bowker et al. (1985). A subset of 10 soils types has been extracted to represent a wide range of 
conditions in three classes (dark, medium and bright). (See Figure 1). Figure 2 exhibits the total surface spectral 
reflectance produced by the joint leaf-canopy model for 11 vegetation fractional cover (a) over 10 soil types; 
isolines link points of equal a, assuming a LAI of 5. Figure 3 shows the value of the five vegetation indices over 
each of the selected soils, ranked by increasing brightness. Compared to NDVL SAVI and MSAVI display a 
reduced variability to dark soils, in accord with its design objectives. Figure 4 shows the overall sensibility of the 
indices to variation in soil type and o by displaying the maximum and the minimum value of each index over the 
set of all 10 soils and for each value of a. To be a useful predictor of a, an index must exhibit appreciable 
sensitivity to that parameter, but also provide little or no sensitivity to soil type. It can be seen that NDVI is