PH 
the retrieved values. This process should be enhanced for low to medium leaf area indices, which is what we 
actually observed (Figure 4). Chlorophyll concentration is less accurately estimated as compared to the 
inversion process performed on the 6 input variables. Except for the white backgrounds, there is a general 
overestimation of chlorophyll concentrations, that could compensate the general underestimation of leaf area 
index. The same general pattern is observed for the leaf equivalent water thickness. The lesser retrieval 
capabilities observed here as compared when estimating the 6 biophysical parameters is to be related to the 
poorer reflectance spectra reconstruction capabilities in the red edge and the water absorption domains. 
The compensation observed between leaf area index and chlorophyll concentration or leaf 
equivalent water thickess can be explained as follows for natural and bright soils. In the visible and middle 
infrared spectral regions characterized by pigments or water absorption, both an increase in leaf chlorophyll or 
water contents, and in leaf area index induces a decrease in canopy reflectance. However, this explanation did 
not hold for black soils for which an incease in leaf area index would lead to an increase in canopy reflectance. 
However, these compensation features lead us to investigate the performances of a "synthetic" canopy variable 
such as the canopy content in chlorophyll or water, which is the product between leaf area index and leaf 
chlorophyll concentration or leaf equivalent water thickness. Canopy chlorophyll or water contents are now 
estimated with a better accuracy as compared to leaf chlorophyll or water contents or even leaf area index as 
presented in figure 5 and 6 . However, the estimates of these canopy variables over the white backgrounds are 
Pop. Size 
LAI 
c ah 
C, 
LAI.C nh 
LAI.C,,, 
85 (°) 
2.29 
10.9 
0.013 
50.1 
0.033 
188 Bands 
78 i 1 ) 
2.39 
10.6 
0.011 
52.3 
0.034 
82 i 2 ) 
2.33 
11.1 
0.013 
50.9 
0.033 
6 Variables to be retrieved: 
75 ( 3 ) 
2.43 
10.8 
0.011 
53.2 
0.035 
[LAI, 6„ s, C nh) C^, N] 
88 (°) 
2.74 
10.1 
0.014 
46.1 
0.039 
6 Bands 
77 i 1 ) 
2.92 
10.4 
0.012 
49.3 
0.042 
84 ( 2 ) 
2.75 
11.0 
0.015 
45.9 
0.038 
73(3) 
2.94 
10.6 
0.012 
49.2 
0.041 
96 (°) 
2.61 
19.7 
0.025 
38.3 
0.051 
188 Bands 
81 0) 
2.64 
16.2 
0.020 
41.1 
0.055 
3 Variables to be retrieved 
82 ( 2 ) 
0.68 
16.0 
0.023 
23.4 
0.022 
[LAI, C nh , CJ 
70 (3) 
0.72 
12.2 
0.018 
25.0 
0.024 
with 
96 (°) 
2.57 
16.3 
0.018 
33.1 
0.042 
[6j= 28.6°, 5=0.33, V=1.23] 
6 Bands 
78(1) 
2.63 
13.5 
0.014 
35.8 
0.046 
82 ( 2 ) 
0.69 
13.6 
0.017 
21.0 
0.019 
70(3) 
0.74 
11.0 
0.012 
22.5 
0.021 
Table 1. Root mean square error values ( rmse ) observed on 3 canopy' biophysical variables: LAI, 
C ah (pg.cnr 2 ), C w (cm). The canopy contents of chlorophyll ( LAI.C ab in pg.cnr 2 ), and water 
(LAI.C W in cm) were also investigated. The population size is presented for a selection of cases: (°) All plots 
with successful inversion. ( : ) All plots with successful inversion and LAIX).5. f 2 ) All plots with successful 
inversion but without a white background. ( 3 ) All plots with successful inversion, with LAIX).5 but without a 
white background. The inversion was applied with retrieval of 6 or 3 variables, and using either the 188 
AVIRIS like bands, or the 6 TM like bands, 
3.2. Using few broad bands. 
We first tried to invert the PROSPECT+SAIL model by retrieving the 6 biophysical parameters using the 6 
Thematic Mapper broad bands. From a mathematical point of view, this inversion process was not obvious 
because the number of unknowns was the same as the number of reflectance data. However, results show that 
the inversion converged regularly for 88 plots over 96. The spectra simulated from the PROSPECT+SAIL 
model using the retrieved values of the 6 biophysical variables show good reconstruction peformances with a 
rmse value of 0.0212. The same features of biases and rmse as for the inversion using high spectral resolution 
information are observed (figure 2a), but in an enhanced manner (figure 2c). However, the 4 biophysical 
canopy variables are poorly estimated (table 1). This is not the case surprisingly for the 2 biochemical 
composition variables, C ab and, C w that have rmse values very close to the one observed from the inversion 
using high spectral resolution information. This result is again repeated when inverting to retrieve only 3 [C ab , 
C ^ LAI] of the 6 biophysical variables, the 3 others being assigned the same values as previously [6,=28.6°, 
s=0.33, A/= 1.23] White soils lead also to poor estimates of the canopy variables for the same reasons as 
188