1986) is used to interpolate daily estimations of LAI, the shape of the temporal variations of LAI being fitted to 
the “observed” LAI. Then, the daily generated “observed LAT are read by AFRCWHEAT as input variables. 
The procedure, applied to wheat fields of various varieties and various known sowing dates, leads to important 
improvements in yield assesment. 
3.12 Calibration of the production model by LAI derived from satellite data. Mass (1988) performed LAI 
inversion from Landsat/TM data over sorghum fields in Texas. Again, the LAI values retrieved along the growing 
period are not very frequent They are not used to adjust, like in the previous case, one or two parameters of a 
statistical model for the LAI profile, but to adjust a pertinent parameter of the crop model GRAMI itself. The 
parameter adjusted is the initial value of the LAI at the emergence, which determines in a very important way the 
canopy behavior and the production. The procedure was applied to fields of known sowing dates, and proved to 
greatly improve the yield estimation. 
A general comment can be done about these two examples. The LAI was derived freon satellite data by the mean 
of classical relationships using vegetation indices, the NDVI in case a), and the PVI in case b). Even if satellite 
measurements can be corrected from atmospheric effects, even if the use of indices can lead to minimization of 
direcdonnal effects or soil influence, some perturbations still remain. Moreover, as soon as the LAI reaches 
values of 4, its inversion from the NDVI becomes very doubtful. The PVI/LAI relation used by Maas was locally 
calibrated. It is then difficult to apply that method in places where ground truth is not available and so it does not 
totally agree with the aims of satellite users who expect information about “non-surveyed” areas. In case a), the 
statistical model may easily be fitted to inaccurate LAI values, because it does not contain in itself the biological 
knowledge required for a realistic canopy description. Case b) avoids such a problem by the adjustment of the 
LAI profile simulated by the model itself. However the extrem simplicity of GRAMI does not always retain the 
model to diverge: the recalibration which leads to a simulated LAI profile coherent with the LAI derived from the 
observations can simulate, in the same time , a non realistic production. We also notice that, in a farcing strategy 
like in case a), the knowledge of the sowing date is required as an essential initial condition. If die timing of the 
be ginning is shifted in the model, we may be suspicious toward the capability of the temporal behavior simulated 
by the model to be coherent with the “observed LAT’. 
3.13. Calibration of the production model by ground radiometric measurements. The third example suggests a 
new strategy, where the remotely sensed observations are not used to derive a canopy variable, but are taken for 
what they really are: radiative measurements. Bouman (1991) adds to the SUCROS model a radiative transfer 
model to simulate the visible and near infrared reflectances at the canopy nadir, and Cloud equations to simulate 
the radar backscatiering. The model is then able to describe the temporal profile of radiative measurements. 
Ground radiometric observations over sugarbeet fields were used to adjust sensitive parameters. These 
parameters are the sowing date, which was not considered to be known in that case, and rather empirical species 
dependant parameters, for which the range of possible values is uncertain: the ‘relative growth rate’ which has an 
effect cm the phenological development, the ‘light use efficiency’, and the ‘maximum leaf area’. The adjustemenl 
of these key parameters allows the model to simulate a temporal description of the canopy behavior very dose to 
the observations. This result leads to important improvements in the yield estimation, because the internal 
coherency of the mechanisms described within the modd is kept Consequently, we may think that the 
description of all variables is better. 
This last approach consists in a complete modeling of the temporal radiometric signal aquired 
over a vegetative canopy. This needs the linkag e of a functional model and a radiative transfer model. To apply 
this strategy with satellite data, we must add an atmospheric correction model either in the direct mode, to 
simulate data at the top of the atmosphere, either in inverse mode, to correct the satellite data and then work with 
radiative measurements at the canopy level. The current available models for the radiative transfer through the 
atmosphere make these two strategies equivalent in quality and weakness. But die satellite acquistion angles must 
be taken into account when the radiative transfer model is used, if we want to reproduce the tune profile of 
radiometric measurements in the exact conditions of observation. Following, the assimilation technique leads to 
the adjustment of poorly known parameters, important for the description of the vegetation cover and for the 
satellite signal, by the retrieval of the best coherency between the model and the data. The parameters which are 
expected to be retrieved from assimilation of satellite data refer to the relevant questions for the production at the 
regional scale. At the field scale, where fanning practices are known, the use of frequent ground radiometric 
measurements allow to focuss at biophysical parameters, which is very interesting for the understanding of the 
biological processes where a high level of precision is required. At the regional scale, the main questions are 
rather die fa rming practices: sowing date, choice of the variety These factors mainly determine the annual 
production of the different species in a given region. We discuss in the next section the use of satellite date to