3x3 pixels window is retained in the profile if the corresponding pixel value in the cloud probability image is 
below a given threshold. Days of extreme illumination/viewing conditions are rejected. 
In order to illustrate how the modelling approach reproduces satellite informations, simulations of NDVI and 
reflectances profiles are compared to profiles extracted from the satellite data set mentioned above (figure 3). 
100mm 
0mm 
Figure 3. : Comparison of simulated and observed NDVI profiles, (simulated : solid line, observed : dashed 
line, dots correspond to days with observations). Bottom profiles are 9 days maximum value composites. 
The profiles of the upper graph contain daily values, whereas those of the lower graph are maximum value 
composites (MVC) over periods of 9 days. The hyetogram used by the ecosystem model for the simulation is 
also plotted. Rainfall amounts to an adequate total of 501 mm, well distributed over the rainy season. Some 
early rains in the month of June have triggered the emergence of vegetation one month before the start of the 
growing season. An early peak of biomass (~ 200 kgDM/ha) can be seen in the simulated profile, but not in the 
observed profile. The growth stages are fairly well tracked until ten days before peak biomass when a sharp 
drop, followed by a relative increase two weeks later, occurs in the observed profile. These anomalies in the 
observed NDVI profiles cannot be explained in terms of incidents during the growth stages, as the distribution 
of rain throughout the season was quite good. It is interesting to note that during the months of September and 
October, once the cloudy days and the days with unfavourable sun/view conditions are rejected, only 12 
observations are left. The probability of obtaining clear days (in terms of water vapour content and aerosol 
loading) are then very low. 
However, in spite of atmospheric perturbations which cannot be taken into account with the available 
meteorological informations, the observed NDVI profiles depict the essential features of the evolution of 
vegetation amount during a growing season. They provide a fairly good indication of the date when installa tion 
and active growth of the vegetation occur, while peak biomass can also be time-marked, but with at least 10*15 
days uncertainty. This type of information, which has not proven very useful up to now, can be particularly 
interesting when driving an ecosystem model with satellite data. Because, in addition to the quantitative 
estimate of vegetation amount usually sought, informations obtained about the phases timing may be used to 
bound the ecosystem model simulations, thus compensating a possible lade of precision in the model inputs (for 
example when only ten-days instead of daily rainfall is available).