model is able to predict 
satellite measurements, 
when seasonal changes 
>rs, will be refereed as 
). In what follows, we 
ovide a more extensive 
water bucket model and 
ren area. This choice of 
cs. The fluxes between 
iperature, precipitation, 
phenological stage and 
tes are coupled through 
dan (1988), and Bonan 
0 years period, nutrient 
med, the model consists 
J(u) = JJC.X — Od|| dt where Od is the set of observations, belonging to [0 T], and C a mapping from the 
state space to the observation space. The problem is then to minimize J(u) under the constraint F(X,u)=0. In 
our case, C is a ’scene' model which computes reflectances from the vegetation state variables. For 
meteorological studies, the control is usually the initial state of some 3D fields, because this causes the main 
errors in the prediction, when compared, for instance, to physical parameterizations schemes. This is not the 
case for ecological studies, where general consensus on model laws usually does not exist This means that 
control 'variables' can be chosen among model parameters as well as among initial conditions. 
For this synthetic experiment 'observations' consist in reflectances computed from the outputs of a reference 
simulation of the model (labelled 'ref hereafter) using a reference parameter set The basic assumption is that 
the model is perfect We then suppose that some of these parameters are poorly known, and give them biased 
values. These a priori values are the 1st guess parameter set The discrepancy between model (with 1st guess 
parameters) and 'observation' is then minimized, and this leads to a retrieved parameter set (labelled 
'retrieval'). 
3.2. Reference simulation: a deciduous temperate forest 
With this model, we simulate a deciduous forest ecosystem, as an example of canopy which shows a 
pronounced seasonality. Daily meteorological inputs have been collected at Boigneville (latitude 48°, France) 
for year 1992. At this site, the vegetation growing season takes place from April to October, with winter 
dormancy and possible limited summer water stress. The purpose here is to have a reasonable simulation of 
natural vegetation, not to model a real forest site. 
l fluxes and the canopy 
nain outputs are the 
escription or validation 
he system dynamics is 
led, the model requires 
im ulation of vegetation 
instance, Bonan (1993) 
>n fluxes. At the global 
le are still problematic, 
propose in this paper is 
nfident in C02 fluxes 
tputs and radiometric 
red as control or ie- 
and oceanography (Le 
the best agreement to a 
•del can be written as 
; (e.g. initial conditions 
an be measured by 
3.2.1. Control parameters. Suitable control parameters must clearly be important drivers of the surface 
radiative properties, suffer from large uncertainties and, finally, condition the carbon fluxes simulation. 
Phenology modelling and allocation of carbon to the leaves both fit these criteria. 
-Phenology: For 'seasonal' ecosystems, canopy development is known to be a fundamental driver of gas 
exchange, when time scales of day to months are considered (KOrner 1993). For different vegetation types, 
internal control on development can be modelled as a response to environmental factors: heat sum, 
photoperiod and chilling requirement for example, for temperate ecosystems (e.g. Murray et al.1989, Nizinsky 
and Saugier 1988), also soil water availability for arid zones and savannahs. As our simulation aims to 
reproduce a deciduous temperate forest, the development has been parameterized as a function of cumulated 
daily (positive) mean temperature starting on the 1st of January. Two phenological events have been modelled: 
a) the start of the growing period: This occur as soon as a threshold Lj of cumulated temperatures is reached. 
Carbon is translocated from the storage compartment (until it is empty) and allows leaves growth and 
photosynthesis, b) the start of the senescence period: as soon as a cumulated temperature threshold £2 ^ 
reached, abscission of leaves occurs, with a constant rate. 
- Allocation of assimilates: eq (1) shows that the daily growth of the various compartments \depends on 
allocation coefficients applied on photosynthesis and carbon translocation. Carbon partitioning is one of the 
major difficulties in process oriented vegetation modelling. Feedback between fluxes and structure occurs in 
annual herbaceous vegetation and also in perennial ecosystems, particularly when behavior over several years 
is considered. Assessing these feedbacks is not an easy task, because growth can be carbon-limited but also be 
dependent on water and nutrient status, or result from 'internal' control characterizing adaptation to the local 
environment. For this forest simulation, we assume that a specific allocation pattern is associated with the two 
phenological events previously defined, and that the allocation coefficients are interpolated, between these 
events, as a linear function of cumulated temperature. We further assume that, at the beginning of the growing 
season, allocation to leave (respectively stem, roots, storage) is a (respectively (l-a)/2, (l-a)/2,0) and 0 
(respectively 0,0,1) at the beginning of the senescence period. 
These three control parameters (£^£2 ^ °0 l^d to a reasonable LAI time profile (figure 1). Quasi-linear 
increase at the beginning of the growing season is due to carbon translocation from storage, which stops when 
the storage compartment is empty, resulting in a sudden change in slope. Quasi-linear decrease corresponds to 
the senescence phase. The net CO2 flux (photosynthesis minus autotrophic respiration) is shown chi figure 2 a). 
Gross Primary Productivity (GPP) is 1011 g C m-2 year-1 and NPP 625 g C m-2 year-1, these values are 
within the range of published data for temperate deciduous forests (e.g. Lieth 1975) 
3.2.2. Reflectance model The foliar surface, predicted by the vegetation model, along with prescribed optical 
properties and real view and solar angles are used to compute, with the SAIL model (Verhoef 1984), red and 
1087