adjust the value of a parameter at the regional scale. 
Figure 1: Coupling satellite data and production models. 
Stategies a, b, and c are described in text. 
3 2 Calibration of the production model by satellite data: assimilation principle 
Moulin et al. (1994) discuss the linkage of different reflectances models with the winter wheat model 
AFRCWHEAT2 (Porter et al., 1992), to simulate the visible and near infrared reflectances observed on SPOT 
data during the growing season over a major French agricultural region. The number of available observations is 
very limited in that study: one image in April (beginning of heading), and 3 images (very close in time) in May, 
at the maximum of winter cereals activity. Consequently, we choose to adjust only one parameter by 
minimiz ation of the variation between modeled and observed reflectances. The April image controls die timing of 
the growth, mainly determined by the sowing date, so it seems possible to adjust this parameter from the 
ass imil ation of the SPOT data. Results showed that the use of die near infrared reflectances observed on the 
wheat fields generally allow to retrieve a sowing date value close to that which is statistically the most 
represented over the studied area according to the technical agricultural services. Some attention must also be 
carried out on the choice of the radiative transfer model, and an the level of incertitude regarding die required 
parameters: optical properties of leaves, soil reflectances, mclinaison of leaves... The mare the biological model 
would be able to describe the behavior of the structure of the canopy (including information on the leaves color, 
like in AFRCWHEAT2), the mere we could be confident in the ‘model to satellite’ approach (Hollingsworth, 
1990). 
We can conclude that we are confident in the interest of die combined use of satellite data and 
functional models with this ‘assimilation’ stategy. It is interesting to notice that the use of images in April and 
May allows to retrieve an initial condition which takes place 6 months earlier (figure 2). just because we benefit 
from an a priori knowledge of the temporal behavior established on biological and physical bases. This is the 
same principle used by Meteorologists and Oceanographs in what is called “variational assimilation”. The 
consequence is the improvement of the description of the processes all along die growing period, at time when 
data are not available: during winter far example, and also at the end of spring, in a predictive mode. Another 
point to underline is the fact that the description of the processes at die period where data are available (May for 
example) is essentially controlled by the temporal dimension, and not by the simple inversion of a characteristic 
variable of the canopy using these May images as unique information. Consequently, the problems of accuracy of 
the quantitative value of the measurement at one date appear to be less important, as well as the problems of the 
insensitivity of reflectances to changes in LAI values higher than 3 or 4. 
Many points must be precised: what is the precision an the adjusted parameters? What is the