1133 
Net Primary Production estimation. One of the main interests of this adjustment technique is the estimation of 
the NPP using the retrieved sowing date as an input of the growth model. For each model, we note a variation 
of the seasonal NPP with the sowing date used (first guess compared with adjusted). But conclusions depend 
on the radiative transfer model considered. 
Sensitivity to the radiative transfer model Results presented on Figure 3 show a dispersion of the retrieved 
dates with the reflectance model used 
* According to Figure 2, for the model of Baret, in n.ix, the simulated values were often higher than 
observations. However, the 2 values of adjusted NPP are close to the reference. Actually, adjusted dates are 
later than the reference, and NPP does not vary anymore. So, in this case, NPP estimation seems correct even 
if reflectance modelling is not perfect. 
* Considering the SAIL model results, simulations are rather good in May (Fig. 2) but in April the value is too 
low. That is why, the adjustment supplies early sowing dates. In consequence, due to the April observation, 
NPP estimations are far from the reference. 
* The EXTRAD simulations fit well with observations (Fig. 2) for all dates. Thus the retrieved sowing dates 
are quite close to 288. But a little variation of sowing day around this period (beginning of October) leads to 
an important change of productivity. 
* For the Nilson-Kuusk model simulation (Fig. 2), the agreement is better in April than in May. Whatever the 
sowing date given by the minimization procedure, the simulated reflectances cannot be in agreement with the 
4 observations at the same time. The 2 retrieved sowing dates are earlier for the first, and later for the second 
than the first guess day. 
We are interested in the estimation of NPP by adjustment of the sowing date. But, as we have seen, 
the sensitivity of this variable to sowing date is more or less important if sowing occurs earlier or later. So, the 
shift between retrieved NPP and reference is not the same whether we obtain earlier or later sowing date than 
288. Despite the sensitivity of the results to the reflectance model used, these tests of satellite data assimilation 
into a vegetation model have shown that the simulation of reflectances is obtained with a sufficient accuracy to 
perform the linkage with remotely sensed measurements. 
4 - CONCLUSION AND PERSPECTIVES 
The first part of this article displays a first step towards the simulation of time reflectances profiles. The 
simulation has been improved with considering the impact of the cover fraction which is an important 
parameter. The study performed at the field scale is an essential step to validate the methodology of linkage 
between a vegetation and a radiative model. Indeed, many information concerning crops are available both 
regarding land use classification and model parameters (sowing date, variety). Although the number of SPOT 
pictures remains low, the confrontation of simulation with these observations allows us to have confidence in 
the simulation methodology used. However, many improvements have to be done concerning the modelling 
process. We have to consider the variations of the leaf inclination with the development of the crop. In fact, a 
great improvement would be obtained if the structure of the plant could be simulated in the growth model. To 
consider the cover fraction, we could generalize by the use of a profile fitting the LAI instead of local ground 
measurements. 
The second part of the text has shown that observations can help us to retrieve some pertinent 
parameters especially concerning the phenological plant development. Pertinent parameters are the ones that 
have a great influence both on description of interesting state variables behaviour (carbon fluxes ...), and on 
time profile of the radiometric signal. This information can then be introduced into growth model, allowing an 
improved description of all the biophysical mechanisms, and thus an estimation of crop production or carbon 
dioxide fluxes. The sowing date has been retrieved at the field scale by assimilation of SPOT reflectances in 
the growth model. However, we have to be careful concerning the adjustment of modelling with observations. 
The minimization procedure may not find the absolute minimum (but a local one) of the cost function that is 
why we have to test several 'first guess' dates to find the 'real' date. Here, the tests performed with several 
reflectance models allow us to check the method used. The assimilation of satellite observations is a means to 
access to a spatialization of the exchanges (matter, energy) versus time. Study of adjustment techniques of 
pertinent parameters of crop models from radiometric measurements will be the ultimate purpose. In 
particular, we will check if assimilation can be performed from regional satellite observations, over large 
agricultural countries.