Assessing the ability of three land ecosystem models to simulate gross carbon uptake of forests from boreal to Mediterranean climate in Europe
Abstract. Three terrestrial biosphere models (LPJ, Orchidee, Biome-BGC) were evaluated with respect to their ability to simulate large-scale climate related trends in gross primary production (GPP) across European forests. Simulated GPP and leaf area index (LAI) were compared with GPP estimates based on flux separated eddy covariance measurements of net ecosystem exchange and LAI measurements along a temperature gradient ranging from the boreal to the Mediterranean region. The three models capture qualitatively the pattern suggested by the site data: an increase in GPP from boreal to temperate and a subsequent decline from temperate to Mediterranean climates. The models consistently predict higher GPP for boreal and lower GPP for Mediterranean forests. Based on a decomposition of GPP into absorbed photosynthetic active radiation (APAR) and radiation use efficiency (RUE), the overestimation of GPP for the boreal coniferous forests appears to be primarily related to too high simulated LAI - and thus light absorption (APAR) – rather than too high radiation use efficiency. We cannot attribute the tendency of the models to underestimate GPP in the water limited region to model structural deficiencies with confidence. A likely dry bias of the input meteorological data in southern Europe may create this pattern.
On average, the models compare similarly well to the site GPP data (RMSE of ~30% or 420 gC/m2/yr) but differences are apparent for different ecosystem types. In terms of absolute values, we find the agreement between site based GPP estimates and simulations acceptable when we consider uncertainties about the accuracy in model drivers, a potential representation bias of the eddy covariance sites, and uncertainties related to the method of deriving GPP from eddy covariance measurements data. Continental to global data-model comparison studies should be fostered in the future since they are necessary to identify consistent model bias along environmental gradients.