Ecosystem model optimization using in situ flux observations: benefit of Monte Carlo versus variational schemes and analyses of the year-to-year model performances

摘要

Terrestrial ecosystem models can provide major insights into the responses ofEarth's ecosystems to environmental changes and rising levels of atmosphericCO₂. To achieve this goal, biosphere models need mechanistic formulationsof the processes that drive the ecosystem functioning from diurnal to decadaltimescales. However, the subsequent complexity of model equations isassociated with unknown or poorly calibrated parameters that limit theaccuracy of long-term simulations of carbon or water fluxes and theirinterannual variations. In this study, we develop a data assimilationframework to constrain the parameters of a mechanistic land surface model(ORCHIDEE) with eddy-covariance observations of CO₂ and latent heat fluxesmade during the years 2001–2004 at the temperate beech forest site of Hesse,in eastern France.As a first technical issue, we show that for a complex process-based modelsuch as ORCHIDEE with many (28) parameters to be retrieved, a Monte Carloapproach (genetic algorithm, GA) provides more reliable optimal parametervalues than a gradient-based minimization algorithm (variational scheme). TheGA allows the global minimum to be found more efficiently, whilst thevariational scheme often provides values relative to local minima.The ORCHIDEE model is then optimized for each year, and forthe whole 2001–2004 period. We first find that a reduced (<10) set ofparameters can be tightly constrained by the eddy-covariance observations,with a typical error reduction of 90%. We then show that includingcontrasted weather regimes (dry in 2003 and wet in 2002) is necessary tooptimize a few specific parameters (like the temperature dependence of thephotosynthetic activity).Furthermore, we find that parameters inverted from 4 years of fluxmeasurements are successful at enhancing the model fit to the data on severaltimescales (from monthly to interannual), resulting in a typical modelingefficiency of 92% over the 2001–2004 period (Nash–Sutcliffecoefficient). This suggests that ORCHIDEE is able robustly to predict, afteroptimization, the fluxes of CO₂ and the latent heat of a specifictemperate beech forest (Hesse site). Finally, it is shown that using only1 year of data does not produce robust enough optimized parameter sets inorder to simulate properly the year-to-year flux variability. This emphasizesthe need to assimilate data over several years, including contrasted weatherregimes, to improve the simulated flux interannual variability.