¿eh depends on 
icients inside a 
ited and, in this 
listribution was 
\rea Index. 
cted, Prod , and 
iplex algorithm 
i parameters p, 
hemispherical 
n coefficient, C 
m if a spectral 
he main feature 
¡rent sites, 
2 
XS3 
2 
1.45 
0 
3.52 
7 
0.63 
8 
0.80 
9 
1.17 
2 
1.00 
7 
1.75 
12 XS3 
0.96 
0.77 
0.99 
0.99 
0.97 
0.98 
0.94 
f the leaves as 
re these optical 
albedo and the 
Another model, 
tentative of the 
dve very similar 
an be observed 
Figure 3.a Relative r_m.s.e as a function of wavelength 
over cotton for various models. - Hapke — Verstraete 
Deering .. Rahman o Roujean x Shibayama + SAIL 
Wavelength 
Figure 3.c Same as figure 3.a except for the correlation Figure 3.d Same as figure 3.c except over bare soil 
coefficient 
over pecan. The SAIL model also seems to give good agreement with these two models. At last, Deering model gives 
most of the time very different results. It has to be underlined at this point that this last model was not developed for 
inversion purposes. 
It can also be underlined that leaf reflectances look more reasconable than the transmittances, even if both values appear 
to be a bit high. In any case, these values have to be compared with ground measured values, which are unprocessed at 
this time. 
The SAIL model also involves explicitly the LAI. But the retrieval of this parameter, through the described method, 
appeared to be very erratic. Actually, the retrieved LAI varies widely with the considered spectral band. The retrieval of 
this specific parameter will be the subject of an upcoming study. 
5. Consequences for use of satellite data 
5.1 Importance of bidirectional effects 
From what has been said, we can underline the influence of bidirectional effects for use of remote sensing measurements 
and especially satellite data. Data acquired along the principal plane are clearly the most influenced by directional 
effects, and even if we limit the view zenith angle to a range of -30° to 30° (corresponding to SPOT possible geometric 
configurations), we can observe variations from -20% to 70% over the different types of canopy. For AVHHR 
configurations, unavalaible in our data sets, the only way to estimate the amplitude of the effects is to simulate 
bidirectional reflectance, with the parameters retrieved from the fits, in a range of -65° to 65°. The variations computed 
this way fall in a range of -50% to 150%. 
5.2 Limited samp ling 
Since most of the operating satellites can only gather measurements in a very limited range of geometric configurations, 
h appears essential to test the possibility of confidently fi tting a model with this kind of limited sampling. The partial test