Potential Above-Ground Biomass data assimilation to constrain slow processes in ORCHIDEE (v4.3) land surface model.

Above Ground Biomass (AGB) and its temporal and spatial dynamics are key to monitor carbon budgets of land forest ecosystems, especially under climate change, but they are currently poorly represented in land surface models (LSMs). However recent advances in LSMs allow to have more explicit representations of stand dynamics, although key parameters associated to C allocation, mortality and recruitment processes are largely uncertain. Data assimilation methods can help to better parameterize these processes, but few studies focused on the use of AGB, compared to fast-varying fluxes like Gross Primary Productivity (GPP). This study aims to bridge this gap in the last version of the ORCHIDEE land surface model, focusing on the African tropical forest which was much less studied than other biomes. We investigate the optimal strategy to assimilate AGB products from remote sensing observations in combination with other classical C flux products in order to improve ORCHIDEE’s representations of C fluxes and stocks of African ecosystems. We assimilate the ESA-CCI AGB product along with the FLUXCOM GPP data to optimize key model parameters for two Plant Functional Types in Africa linked to photosynthesis, C allocation and mortality, using either a Genetic Algorithm or a variational approach. The fast processes are first constrained with GPP (FLUXCOM data) while the slow processes are optimized with AGB (ESA-CCI data). We select potential maximum AGB for each model pixel (~50km), using the upper quartile of the high-resolution data (~30m), which represents the likely AGB of an undisturbed ecosystem. This choice reflects the fact that the current ORCHIDEE version is more suitable to represent ecosystem response to climate drivers rather than to disturbances. The final objective will be to use raw AGB to define an additional regional or pixel-based disturbance layer to ORCHIDEE. Key parameters involved either in fast (GPP) or slow (AGB) processes are selected by sensibility analysis. This two-steps assimilation allows us to significantly reduce the RMSD against the observations, for both GPP and AGB. This study highlights the potential of remote sensing AGB to constrain slow processes of LSM to better capture the dynamic of AGB in African tropical forests. While requiring a specific methodology, the assimilation of AGB induces significant changes in the C allocation, mortality and regrowth simulation by the ORCHIDEE model, thus impacting the carbon budgets of African tropical forests as well as increasing the overall confidence in future projections.