1132 
32. Observations assimilation in a crop growth model 
Since the growth model is able to simulate interesting variables regarding the productivity, we attempt to use 
satellite measurements to control the model : poorly known parameters are adjusted to obtain the best 
agreement between simulation and observation. Since only 4 observations are available, we focus our attention 
on one particular parameter : the sowing date. This input drives the phenology and thus, as said above, allows 
the estimation of carbon fluxes. 
The present study concerns the region of 40km by 40km. According to the land use classification, 
winter cereals represent more than 52% of the surface. So, we attempt to retrieve the most probable sowing 
date relative to this vegetation group over the region. Observations are relative to the test-sites (wheat 
'Soissons' sowed on 15 October), thus, as we know the real sowing date, we are able to conclude about the 
validity of the assimilation. We successively use 7 first guess dates (in a range of plus or minus one month 
around the 15 of October) as inputs of the growth model. Then for each of these dates, an optimization 
algorithm (downhill simplex method, Matlab, from The Math Works, Inc.) estimates the sowing date that 
minimizes the difference between simulation and observation. We apply this technique using successively the 
4 radiative transfer models. Although we observe a sensitivity to the initial sowing date (over the 2 months 
period), whatever the transfer model used, it converges towards 2 retrieved sowing dates. 
reference date retrieved dates 
Baret model: 288(-) 298 (..), 304 (--) 
SAIL model: 288(-) 270 (..), 275 (-) 
EXTRAD model: 288(-) 287 (..), 278 (--) 
Nilson-Kuusk model: 288(-) 299 (..), 277 (--) 
The retrieved values have been introduced into the growth model to compute n.i.r. reflectance time profile and 
Net Primary Production (NPP), (Fig. 3). 
Baret model 
SAIL model 
EXTRAD model 
Nilson-Kuusk model 
Figure 3 : Reflectance time profiles. Observations (*) with error bars which are 2 times standard deviation in 
height. Curves represent Nadir simulations using respectively the Baret, SAIL, EXTRAD and Nilson-Kuusk 
model. Reference profile : real sowing date (-), profiles obtained with the retrieved sowing days (..) & (-)■ 
NPP values corresponding to these simulations are in gClnFly.