Evaluating state-of-the-art process-based and data-driven models in simulating CO₂ fluxes and their relationship with climate in western European temperate forests

Abstract. This study evaluates two process-based (LPJ-GUESS and SMAP-L4C) and two data-driven (CarbonSpace and FLUXCOM) models to capture the temporal variability of CO₂ flux exchanges (GPP, RECO and NEE) of evergreen needleleaf and deciduous broadleaf forests (ENFs and DBFs) in temperate western Europe and its relationship with climate. Three sites from the FLUXNET network are considered together with two non-instrumented sites located in Burgundy (North-East France). The focus is put on the representation of the annual cycle, annual budget, interannual variability and “long-term” trend. The data-driven models are the best models for representing the mean annual cycle and mean annual budget in CO₂ fluxes despite magnitude uncertainties. In particular, the models accounting for plant functional types in their outputs tend to simulate more marked annual cycle and lower annual CO₂ sequestration for DBFs than ENFs in Burgundy. At the interannual timescale, the CO₂ flux – climate relationshipis stronger for GPP and RECO than NEE, with increased CO₂ fluxes when 2 m temperature, vapor pressure deficit and evapotranspiration increase and when precipitation and soil moisture decrease. The models forced by dynamic climate conditions clearly outperform those driven by static climate conditions. The “long-term” trend is not obvious for NEE neither in the observations nor in the simulations, partly because both GPP and RECO tend to increase in western Europe. Our results suggest that the spatial resolution of the climate drivers is likely very important for capturing spatial and temporal patterns in CO₂ exchanges and point towards the need to choose the appropriate model and spatial resolution according to the scientific question to deal with.