1115 
R - f vD R d + ivO-^G + o ' fk-D ' *vG)-^S 
where fy is the cover fraction and subscripts D, G and S stand for dry, green and soil respectively. 
The areal extent of bare soil varies throughout the growing season, and even at the peak of biomass present, the 
percentage of soil visible may be more than 50%. Therefore, a soil reflectance model has been judged necessary 
to account for the effect of the soil on the scene reflectance. The model used here, is one derived from Hapke’s 
model (Hapke, 1981; Pinty et al., 1989) and validated on different soils (Jacquemoud et al., 1992). The model 
used for the vegetation components is one of the most widely used canopy reflectance model, the SAIL model 
(Verhoef, 1984). The only parameters used to describe the vegetation are the LAI, the LAD (Leaf Angle 
Distribution) and the optical properties of the leaves. The input needs of the canopy reflectance model are 
therefore compatible with the structural vegetation variables simulated by the ecosytem model. 
The model is run separately for the green and dry components. The LAI used in the model to calculate the 
reflectances of each component is given by LAI/f v , which represents the LAI simulated by the ecosystem model 
confined in a fraction f v of a total unit surface. The leaf optical properties used as input parameters have been 
estimated with the PROSPECT model (Jacquemoud and Baret, 1990) which computes leaf optical properties 
spectra from the chlorophyll concentration and a structure parameter of the leaves. As parameter values for 
sahelian grasses are unavailable, those for crops have been used instead. In the absence of more detailed 
information, a spherical distribution is used to characterize the leaves inclinations inside the homogeneous 
canopy. The soil underlying the vegetation is considered a Lambertian surface and its hemispherical reflectance 
has been estimated using the soil reflectance model mentioned above. A more complete description of the 
modelling can be found in Lo Seen et al (1993). 
Modelling atmospheric effects 
Atmospheric processes such as absorption by gases and scattering by molecules and aerosols, are responsible 
for the contamination of satellite data acquired within the solar spectrum. A direct consequence of this is the 
deterioration of the relationships established between vegetation indices and vegetation biophysical 
characteristics. The effects of atmospheric constituents on the specific case of NOAA AVHRR visible and near 
IR data have been investigated and presented in Tanrd et al. (1992), where the magnitude of each decoupled 
effect on the reflectances in both channel and on the NDVI are discussed. Particularly, a number of studies have 
been carried out on the variability of transmission properties of the atmosphere in the Sahel in relation with the 
same NOAA AVHRR data (Holben et al., 1991; Justice et al., 1991; Soufflet et al., 1991). These studies have 
stressed on the importance of water vapor absorption in the near IR band, recommending that correction for 
this effect be made a priority especially for sparsely vegetated areas. The correction for aerosol effects, however, 
is quite difficult to perform as they are highly variable in space and time, and necessitates measurements of 
aerosol optical thicknesses which are often unavailable. In the present study the atmospheric effects on the 
simulated reflectances are accounted for using the SMAC method (Rahman and Dedieu, 1993) which is a 
computational simplification of the 5S radiative transfer model (Tanrt et al., 1990). Gas absorption (HjO, 0 3 , 
0 2 and C0 2 ) is computed as for a US62 standard atmosphere (McClatchey et al., 1971) profile of pressure, 
temperature and gas concentrations. Here, the vertically integrated content of 0 3 is assumed constant and equal 
to 0.25 cm atm. The water vapor data used are the 0.5°x0.5° gridded outputs of the operational General 
Circulation Model (GCM) of the European Center for Meteorological and Weather Forecast (ECMWF, 
Reading, UK) and furnished by METEO-FRANCE (Toulouse). In the absence of data on the aerosol loading of 
the atmosphere, a continental aerosol model with an optical thickness = 0.2 (corresponding to a relatively 
clear atmosphere) are considered for all simulations. 
Comparison with satellite data 
The satellite data set used as the test case has been extracted from the AVHRR Global Area Coverage (GAC) 
data set of the African continent archived at the Institute for Remote Sensing Applications, CEC Joint Research 
Center (Ispra, Italy). The daily images retained are the TOA NDVI and reflectance images derived from 
radiances measured in the channels 1 (risible 0.58-0.68 pm) and 2 (near infrared 0.72-0.98 pm), together with 
a cloud probability image, during the period June to December 1987. The images are furnished in a Mercator 
projection with a pixel size of 5x5 km in the Sahel. More information about the archive can be found in 
Belward et al. (1992). At the sites of the Ferlo where the ecosystem model simulations had been validated, 
temporal profiles of the NDVI and reflectances are extracted with the condition that a value averaged over a