344 
3.3. Pigment analysis 
The pigment concentration was immediately measured after completion of reflectance measurement. Chlorophyll 
a, chlorophyll b, and total carotenoids were extracted in acetone 80%. The concentration was determined by 
optical densitometry according to the methodology defined by Lichtenthaler (1987). We observed very strong 
correlation between chlorophyll a and chlorophyll b concentrations (Fig. la). For this reason, we decided to 
consider only explicitly chlorophyll a concentration. A weaker correlation exists between chlorophyll a and 
carotenoids (Fig. lb). 
Figure 1. Relationship observed between chlorophyll a (C a ) concentration and chlorophyll b (C b ) (a) or 
carotenoid (C c ) (b) concentrations. Dots correspond to the data points. The solid lines correspond to the linear 
regression, (a): C b = 0.29 C a +0.72; RMSE= 1.3; R 2 =0.91; n=35; (b): C c =0.23 C a +1.32; RMSE=l.5\ R 2 =QM, 
«=35. 
3.4. Parameter fitting 
The following power and exponential functions were used to describe the relationship between C c /C a and the 
three pigment indices (PI) shown earlier: 
C c /C a = a+b.PIS (5) 
C c /C a - a+b.exp(g.PI) (6) 
We estimated the parameters a, b, and g of functions (5) and (6) using classical non linear fitting techniques with 
the least square as the criterion to minimize, die minimum simplex being the minimization algorithm (Nelder 
and Mead, 1952). For each of the 6 models (3 PI times 2 functions), we looked systematically for the best 
couple of wavebands [Xj, X,] to be used in the 400 nm-800 nm spectral domain. We investigated only cases 
where Xj<Xj because of the obvious symmetry observed between the role of R { and R 2 in the definition of Pis : 
SRPI(R kl JR k2 ) = SRPI(R k2 R xl y l \ NDPI(R U R X2 ) =-NDPI(R k2 R kl )\ SIPI(R kl R k2 ) = SIPI(R k2 P kl y l . 
4. RESULTS AND DISCUSSION 
To select the best set of wavebands [X ; , X 2 ], we plotted the R 2 values corresponding to each set of wavebands 
[Xj, X 2 ] used to compute the PI s. Figure 2 a, b and c presents the results as gray level maps for the three 
pigment indices SRPI , NDPI, and SIPI. For each plot, the upper part corresponds to the power model (5) and the 
lower part to die exponential model (6). The two functionally linked indices (SRPI and NDPI) behave very 
similarly. On the average, the power model performs better (Table 1).