As the largest tropical forest in the world, Amazonia accounts for 15% of terrestrial productivity. Precipitation across the basin is predicted to decline as global climate change and regional land transformation continue through this century. This study evaluated predictions of site-level gross, net and component ecosystem C-fluxes by five land-surface models (ED2, IBIS, JULES, CLM3.5, SiB3) and one site-specific carbon and hydrodynamic model (SPA) for two Brazilian Amazon rainforests (Tapajos and Caxiuana National Forests) when precipitation was reduced by 0%, 30%, 50% or 80% for 7 years. All simulations followed a standardize protocol that included common site-level meteorological drivers, spin-up procedures, and edaphic properties. Model output was compared against reported C-fluxes from long-term, large-scale, in situ drought experiments conducted at the two rainforest sites.
Results/Conclusions
For both sites, the ensemble annual mean prediction of net ecosystem production of carbon shifted from a sink to a source when precipitation was reduced by 50%. However, considerable variation in model predictions of the component C-fluxes implicated contrasting mechanisms causing the shift in net ecosystem production. Also, no single model consistently outperformed any of the others across precipitation levels, sites or C-flux variables. Three key findings emerged from this study that explained much of the discrepancy in predicted C-fluxes among the models and between model predictions and observations. First, the soil water-stress function was too simplistic in the five regional models to realistically characterize the stomatal response to drying soil. This consequently led to unrealistic diurnal and/or longer-term patterns of gross primary production at both sites under drought conditions. Second, the large differences in the magnitude of the change of simulated heterotrophic respiration under drought were explained by relatively large differences both in the change of the soil C-pool and the sensitivity to soil moisture prescribed in each model. Third, in contrast to reported observations, most models incorrectly predicted a significant reduction in autotrophic respiration as drought conditions persisted. This study elucidates where model development efforts should be concentrated in order to increase our confidence in predicting the fate of a drier future Amazon forest.