There is large uncertainty about whether the Amazon will be a carbon sink or source over the next century under a changing climate and rising atmospheric CO2 levels. To investigate this uncertainty, we simulated the ecosystem dynamics of a lowland moist tropical forest using three models: the Ecosystem Demography model 2 (ED2) (structured vegetation), the Community Land Model 4.5 (CLM4.5-BGC) (unstructured vegetation), and the Functionally-Assembled Terrestrial Ecosystem Simulator (FATES) (structured vegetation; which combines ED and CLM), driven with local climate and CO2forcing from the preindustrial period to 2100. Tree inventory data from a site north of Manaus, Brazil, with repeated demographic measurements from 1996-2011 were compared against simulations from the same time period.
Compared to field observations, ED2 and FATES showed good agreement with observed biomass and forest size structure, but predicted higher growth and mortality fluxes than observed. These biases led to high, continual growth in all size classes and functional groups, whereas the field data indicate that a quarter of canopy trees showed no detectable growth, and the site had neutral biomass accumulation over the 15-year period. CLM4.5-BGC showed better agreement with measured growth and mortality fluxes. With a doubling of CO2 by 2100, all models predicted an appreciable forest sink (1.04 and 0.52 Mg ha-1 yr-1 in ED2 and CLM4.5-BGC, respectively), but due to contrasting process representations, different trends in biomass accumulation and opposing responses of vegetation turnover rate emerged. The differences were attributed to phenology response, nutrient constraints and down-regulation of photosynthesis (native to CLM4.5-BGC only), inability to capture accurate density dependent processes (ED2 only), and large woody net primary productivity responses in ED2. While ED2 and FATES predicted a forest sink similar to field-based inventory studies averaged across the entire Amazon basin, the magnitude of the site-specific sink was large. Determining appropriate process-level model benchmarks for constraints on carbon accumulation rates with rising CO2 is an important focus for future research.