×

You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Title
Mesures physiques et signatures en télédétection

1086
2 - MODEL DESCRIPTION
The linkage of a vegetation model and global remote sensing data requires that the model is able to predict
radiative properties of the surface at the same time and space scales as available satellite measurements.
Evolution of the canopy structure with time has to be explicitly simulated, especially when seasonal changes
occur. These changes, which are driven by both environmental and internal factors, will be referred as
phenological events (for instance, leaf shooting, senescence, abscission, dormancy). In what follows, we
present the general frame of the generic vegetation model that we developed, and provide a more extensive
description and parameter set in Kergoat et al. 1994. The system is described by a soil water bucket model and
4 carbon compartments which average the leaves, stems, roots and storage over a given area. This choice of
compartments comes from differences in radiometric response and in carbon kinetics. The fluxes between
these reservoirs depend on daily climate input (daily minimum and maximum temperature, precipitation,
irradiance, net radiation, vapor pressure deficit) and parameters associated with phenological stage and
vegetation type (e.g. annual/perennial evergreen/deciduous...). Carbon and water fluxes are coupled through
the stomatal resistance. The fluxes sub-models are derived from Running and Coughlan (1988), and Bonan
(1991) formalism. Nutrient cycle has not been considered yet. For simulation over a 10 years period, nutrient
limitation may be implicitly prescribed in the parameter set. As far as carbon is concerned, the model consists
in a set of differential equations:
~=ax(t).[f(G,t)+h(t)]-g(G,t)
' dMi r , (1)
— = cu(t).[f(G,t) + h(t)]~
where G is the leaves carbon content
Mj the carbon content of stems, roots and storage (i = 2 to 4)
f(G) is the photosynthesis rate
h(t) is the rate of carbon translocation from storage
g(Mj) is the sum of carbon losses due to autotrophic respiration and mortality.
ai(t) is the allocation of carbon to Mj
Our assumption is that ai(t) functions are driven by the phenology and link the carbon fluxes and the canopy
development. These equations are solved with a one-day time step. The main outputs are the
atmosphere/vegetation CO2 fluxes, litterfall, and carbon compartments. A complete description or validation
of the model is beyond the scope of this paper, but we assume, for this study, that the system dynamics is
reasonably well simulated. Once the relevant processes have been identified and modelled, the model requires
parameter sets which, ideally, are measured and accurate enough to allow satisfying simulation of vegetation
functioning. This is a critical issue, especially when spatialization is considered. For instance, Bonan (1993)
pointed out the importance of vegetation type and LAI in estimating boreal forest carbon fluxes. At the global
scale, modelling autotrophic respiration, carbon partitioning and phenology for example are still problematic,
for process models are sensitive to these parameterizations. The methodology that we propose in this paper is
to use satellite observations to correct the model course and hence to be more confident in C02 fluxes
estimations.
3 - ASSIMILATION: A SYNTHETIC EXPERIMENT
3.1. Method
In this paper, we propose to use comparison between this prognostic model outputs and radiometric
measurements to improve the model prediction. This methodology, hereafter referred as control or re
calibration, is related to variational assimilation developed in atmospheric research and oceanography (Le
Dimet and Talagrand 1986). The purpose is to find the model trajectory which leads to the best agreement to a
set of observations distributed in time. If X represents the model variables, the model can be written as
F(X,u)=0, where u, the control variables, are inputs leading to a unique model evolution (e.g. initial conditions
or parameter values). The discrepancy between the model and the data can be measured by