27 
Valensole 
250 300 
NDVI (—) and 
Inserted in the 
he local rainfall 
om Leroy and 
reases and the 
his effect may 
site in Figure 
;es shadowing 
etween 1 and 3 
te near infrared, 
tions, as can be 
andard deviation 
Roujean, 1994). 
lefined with the 
Dtted in Figure 5 
d with the kO in 
he time series of 
aspect and often 
the Beauce site, 
y spring and late 
3VI through the 
sting to note that 
ed NDVI of the 
irve of observed 
ation activity is 
supported by the rainfall data from a local meteorological station (Figure 5, Valensole site), which show 
evidence of strong rainfall events around day 250 after two months of drought. 
Site 
Model 
Ref. 
5 visible 
5 near infrared 
arid Crau 
Roujean et al. (1992) 
in 
0.013 
0.017 
humid Crau 
" 
0.013 
0.013 
Valensole plateau 
" 
0.014 
0.018 
Beauce plain 
" 
0.018 
0.027 
Champagne 
" 
0.023 
0.032 
Remoulins 
" 
•• 
0.013 
0.021 
Landes 
'• 
" 
0.017 
0.015 
Niger 
Ross (1981) 
(2) 
0.025 
0.023 
« 
Verstraete et al. (1990) 
0.025 
0.027 
» 
Shibayama and Wiegand (1985) 
0.024 
0.024 
- 
Deering et al. (1990) 
0.024 
0.024 
» 
Roujean et al. (1992) 
•• 
0.025 
0.025 
- 
Rahman et al. (1993) 
0.025 
0.022 
» 
Hapke (1981) 
" 
0.026 
0.023 
Kansas prairie 
Ba et al. (1994) 
(3) 
0.007 
0.014 
Table 1 : Value of residue 5 given by Eq.(3), for different sites, in the visible and near infrared. Ref (1): Leroy and 
Roujean (1994); Ref. (2); Cabot and Dedieu (1994); Ref. (3): Ba et al. (1994). 
3 3 Extracting surface albedo 
Surface albedo extraction needs an estimation of bidirectional properties of the surface. This can be achieved 
with satellite measurements for which the directional configuration is sufficiently variable, combined with 
surface bidirectional reflectance models as described in § 3.1. In that case, the corrected reflectance R^^. 
given by Eq. (4) is simply the surface direct spectral albedo. Cabot and Dedieu (1994) have used this method 
with a set of daily AVHRR/NOAA data acquired over a test site located in Niger, north of Niamey (15.02°N, 
3.66°E), from May to October 1991. Atmospheric corrections are performed using the SMAC method 
(Rahman and Dedieu, 1994) based on the 5S radiative transfer code (Tanrd et al., 1990). Atmospheric water 
vapor content is derived from quasi-daily radiosoundings performed by the Meteorological Service of Niger at 
Niamey airport No in situ measurement is available for the optical depth of the aerosols, which are assumed 
of the continental type with an optical depth of 0.2 at 550 nm. Mean monthly ozone content is taken from a 
climatology. Since the site is a semi-arid site which shows very small temporal evolution, the period of 
composition is chosen to be the whole study period (6 months). A number of different analytical reflectance 
models of various philosophies, empirical (Shibayama and Wiegand, 1985), semi-empirical (Roujean et al., 
1992; Rahman et al., 1993), geometrical (Deering et al., 1990), and radiative transfer based (Hapke, 1981; 
Ross, 1981; Verstraete et al., 1990) have been tested against the satellite data. 
The level of fluctuations 5 remaining after modelling (Eq. (3)) is shown in Table 1, for the various 
models under test. Table 1 shows that 8 is of the order of 0.025 for the 2 spectral bands, practically 
irrespectively of the chosen model. This value, in the visible, is somewhat larger than the values of 8 obtained 
in test sites in a temperate region (see Table 1), but is of the same order when expressed in relative value, the 
visible reflectance of semi-arid sites being rather large. Figure 6 shows that the time profile of modelled 
reflectances (using here the model of Shibayama and Wiegand, 1985) resemble that of the observed 
reflectances. 
An attractive reason for using bidirectional reflectance models in the compositing process is that they 
can supplement the limited angular sampling provided by a sensor such as AVHRR, not only for the 
reconstruction of radiative quantities such as albedo, but also for the prediction of the reflectances that an other 
sensor would see on the same target. Thus Cabot and Dedieu (1994) have derived the surface bidirectional 
reflectance in the shortwave METEOSAT band (0.35-1.1 p) from a linear combination of bidirectional 
reflectances obtained from AVHRR visible and near infrared bands, and have simulated the TOA reflectance 
as could be seen by the shortwave band of METEOSAT. The comparison between predicted TOA reflectances 
and reflectances actually measured by METEOSAT gives very satisfactory results for every tested model on 
the Niger semi-arid site, with rms errors of reflectance of about 0.017.