973 
NOAA/AVHRR NDVI over 128 reserves in Africa to validate the spatial variations of LAI values simulated by 
an ecological concept of equilibrium between precipitation and canopy structure (Woodward and McKee, 1991). 
2.3 Modeling the functioning for each kind of ecosystem. 
The gener alization of FOREST-BGC to all kinds of biomes was made possible by the establishment of a specific 
par ame ter set for each functional type (BIOME-BGC, Running and Hunt, 1993). There is no phonological model 
in the present versimi, which means that, for non-evergreen canopies, the time profile of the LAI, as well as its 
value during the active period must be prescribed. The simulations made by that model have shown that the 
various outputs (fluxes, NPP...) were extremely sensitive to LAI, which is a major characteristic of the canopy. 
One solution could be to use temporal series of satellite data in die short wavelengths which appear to be an 
efficient tool to monitor vegetatimi phenology, as we are gcnng to discuss. 
231. Characterisation of vegetation cycle for African savannahs. GVI time series were used to map 
informatimi about the seasonal activity of savannahs, i.e. the phonological cycle characterized by length and 
be ginning of active period. At a given time, a Markovian model of radiometric behavior gives the probability for 
the canopy to be in the following states: dormancy, growth, and senescence depending cm the kind of ecosystem 
(Viovy and S aint, 1993) This results in an a priori knowledge of what the GVI temporal profile could be. in a 
stochastic sense, allowing uncertainties in the description of the vegetarian behavior. The temporal GVI profile 
can also carry some informatimi not directly related to the biological activity of the canopy: remaining noise, soil 
influence... The combination of the Markovian model with the satellite signal has provided better confidence in 
the phenological cycles obtained by the indicatimi of “what the most probable temporal profile” should be. In a 
second step, the same GVI time series over savannahs were coupled to the “qualitative” functional model of 
Fomès and Blasco (1988). This model simulates the temporal behavior of the canopy activity for several classes 
of sav annah as a function of the distributi cm of the precipitatimi during the previous monthes. This model was 
applied to the FAO vegetatimi map, and the comparison of the simulated profiles with the GVI data has allowed a 
control of the vegetation map, with correction of the classification in some places to improve the consistency 
between simulation and observations (Viovy and Saint, 1991). 
These results show that it is very interesting to use an a priori knowledge about the temporal 
radiometric response of the surface. After these qualitative descriptions, we want to discuss the modeling of the 
seasonal behavior in a quantitative way. Clearly, what we expect from a functional model, is: a) the modeling of 
the exchanges depending on the state of the canopy, b) the modeling of the changes in the state of the vegetation 
itself and their links with the previous exchanges and the internal development. Among the various processes 
which must be taken into account to correctly describe the functioning, the phenology and the allocatimi of 
assimilates to the different organs are poorly known mechanisms. Depending cm the ecosystems, snowmelt, 
temperature or soil water availability drives the phenological calendar, which controls in part the partitimi of 
assimilates between above-ground biomass, roots or storage pool. If we want a correct carbon budget, for 
example .at the adequate time step, an improvement of the modeling of these processes is requiered 
232 A regional Sahelian grassland model. Lo Seen et al. (1994) describe an example of a regional ecosystem 
process model able to simulate fluxes, growth and senescence at a daily time step far annual Sahelian grasslands. 
The ecological model has been validated with 15 years of biomass measurements in two regions and appears to 
reproduce the interannnal variations in the production as well as the spatial ones. Some specific daily outputs of 
this model are the LAI and the vegetatimi cover fraction. These two variables allow the coupling with the 
radiative transfer model SAIL (Verhoef, 1984) to simulate the temporal profile of the visible and near infrared 
reflectances (and then the NDVI) over the canopy. These simulated radiometric temporal profiles are in good 
accordance with vegetatimi indices derived from NOAA/AVHRR data after adequate calibration and atmospheric 
corrections. 
The ability of an ecological model, coupled to a radiative transfer model, to reproduce the 
radiometric observations, suggests that these observations could be used to control some pertinent parameters of 
the ecological model if die info rmati on required to drive the temporal feature of the canopy development are 
missing. Agronoms have already tested the use of temporal radiometric measurements over crop fields in 
complementarity with crop production models in order to improve the outputs of interest, especially the yield 
estimation. They are available for major crops, thanks to the great amount of knowledge for these species. We 
present same examples of the use of satellite data with crop models in the next section.