978 
control the timing of these events and the value of the allocation coefficient of assimilates to leaves. In a first step, 
the modeling has focused on the simulation of a temperate deciduous forest vegetation type. With the description 
chosen for the various processes relevant for the NFP estimation, three pertinent parameters are shown to be 
important for the time course of the LAI: cumulative temperatures required for the budburst, for the senescence, 
and allocation coefficient to leaves. 
Kergoat and Dedieu (1994) describe an example of an assimilation test with simulated data. Simulated 
reflectance data are computed in the view conditions of NOAA/AVHRR measurements, and some noise is added 
to take into account residual perturbations remaining after atmospheric correction and filtering process. The 
assimilation experiments generally lead to the very good retrieval of these parameters values. Consequently, the 
estimated NPP is greatly improved The choice of the minimization algorithm used requires a specific attention 
The relations which describe the biological processes have great non-linearities, and same af them are simple 
“threshold functions”. Consequently, the cost function exhibits discontinuities in response to the variations of 
parameters. The dow nhill simplex method generally gives satisfactory results, even if the procedure cannot 
always avoid to stop in local minima depending on the first guess. 
This result shows that the rules and parameterizatians which are defined for some ecological processes in 
particular places can be generalized in space, and in time by the use of long-time series observations. If the model 
is able to reproduce the interannual variations observed which result from variability or from changes in the 
environmental conditions, it will be a useful contribution to the general modeling of the vegetation dynamics. An 
example of a rather complete model published these last years to answer the question of ecosystems dynamics 
and functionning is HYBRID (Friend et al. 1993). It takes into account the dynamic aspects of forests stands 
(succession, mortality...) on the long term and the daily functioning of the canopy cover by the association of a 
Gap-model and a biogeochemical exchanges model. It does not describe the seasonal course of LAI because it 
has been first established for coniferous forests, but the modeling of the phenolocical processes would be an 
improvement which is certainly required for using such a model with confidence over non-evergreen canopies. 
The design of HYBRID was possible for specific ecosystems on particular sites where the laws used to describe 
the various processes (photosynthesis, respiration, évapotranspiration, growth and senescence, mortality, 
dispersion, recruitment, species competition...) were established, either in a mechanistic way, or just empirically, 
from many ground studies. Remote sensing is a unique tool to extent such a model to new geographic areas by 
the determination of the spatial variability of parameters, and perhaps by modification in the description of sane 
processes. 
5 - DISCUSSION 
Global and continuous coverage of Earth by satellite observations are the unique tool which allows the inversion 
of variables used as inputs into diagnostic models fa the estimation of the Net Primary Productivity at the global 
scale for example. The analysis of the spatial features of such outputs and their interannual variability is then a 
very exciting subject (Maisongrande et al. 1994), which could not be done without the use of satellite data. 
Beside this “forcing strategy”, the purpose of this paper was to demonstrate that satellite data have, in some way, 
the potential to be used as “control”. They can validate ecological models, by the means of the comparison of the 
spatial and temporal variations of radiometric values over terrestrial ecosystems which are simulated and 
observed. They can help to determine which factors are the most sensitive fa the different processes, and have a 
strong impact on the satellite signal. Consequently, they can help in the definition of the functions which are the 
most relevant fa the description of such processes. 
Assimilation techniques provide the calibration of the values of some poaly known parameters, 
as well as some initial conditions. Some of the available minimization algorithms are usable to control biological 
models. The problem of local minima will always exist, and a careful protocol designed with different sets of 
“first guess” must lead to experimental configurations where we are sure to find the absolute minimum However, 
the absolute minimum can be obtained with a specific combination of parameters far from their realistic range of 
possible values, and then providing a totally incoherent description of the vegetation behavia. Generally, the 
functional models built on serious (which does not mean complex) biological and ecophysiological rules, contain 
by themself enough coherency between the various described mechanisms to not diverge in drastic ways. 
Nevertheless, the range of possible values fa each parameter, as well as the range of their possible associations 
would help to constrain the minimization procedure by taking into account these information as well as the 
information cm reflectances.