level of improvement in the interesting output variables, depending mi the number and the temporal distribution 
of the data used? However, this assimilation stategy is the only one usable in order to retrieve the probable value 
of important parameters over large areas where local information mi the initialization or the parameterization of 
relevant processes are missing. We discuss in the next section the potential of satellite data assimilation to precise 
the modeling of the behavior of natural vegetation. 
Initial condition 
t 
Data assimilation 
Figure 2: principle of the “variational assimilation”. 
Retrieval of the most probable trajectiry leads to the processes description at any time. 
4 - MODELING THE SEASONAL DYNAMICS OF NATURAL VEGETATION. - TEST OF 
SATELLITE DATA ASSIMILATION. 
As a consequence of what has been said before, from satellite observations in the short wavelengths, we can 
expect to retrieve the parameters which drive the seasonal profile of file canopy. In case of natural vegetation 
modeling, major unknown are phenology and allocation. The correct description of these processes is important 
fa NPP es timation , for water and C02 fluxes, and fa vegetation itself. To expect some help from satellite data, 
the model must be conceived with the specific objective of a possible control by satellite observations. It means 
that the model must describe file tempaal change in variables directly linked to radiative measurements. Fa 
example, the only model able to simula te the seasonal behavia of natural vegetation at the global scale, TEM 
(Terrestrial Ecosystem Model Raich et al. 1991) cannot be controlled by satellite in its current state. It simulates 
the seasonal behavia of fire cover by the description af 5 state variables (carbon and nitrogen contents in the 
vegetation and in the soil) at a monthly time-step. Exchanges are simulated as functions of climate, soil and 
vegetation type, and the hydric budget drives a kind of phonological submodel. But linkage with reflectances 
needs the knowledge of the LAI. which is not described by TEM: the carbon content of the various parts in the 
vegetation (leaves, stems, roots) are not distinguished. 
So, the use of satellite data to improve the modeling must be kept in mind when file model is 
built. Kergoat et al. (1993) propose a functional model with the following objectives: a) the description of the 
exchanges and the sensorial dynamics of natural vegetation at the global scale, b) the assimilation of satellite data. 
Visible and near infrared reflectances are. in the short term, the more immediate data available fa the whole 
globe every day from NOAA/AVHRR and are especially intersting in the case of vegetation with an annual 
cycle: grasslands, deciduous forests, toundra, crops. It is then important to give a correct description of the 
seasonal canopy development, by the means of variations of carbon content of green leaves, Le. the LAI profile. 
At the be ginning the growth of LAI is driven mainl y by the storage pool, then as soon as photosynthesis occurs, 
a part of the daily assimilates is allocated to leaves, until senescence. Phonological calendar and various stresses 
977