Changing climates and disturbance regimes of the 21stCentury will push tropical ecosystems into novel states that have no-analog in the recent historical record. In addition, inaccurate simulations of tropical forests contributes to multiple biases in Earth System Models (ESMs) and results in disagreement among models regarding whether tropical forests will be future carbon sources or sinks. The goal of this study was to compare ecosystem demography model predictions to field measurements across gradients in tropical carbon cycling and community composition in Amazonia, with some datasets spanning up to 24 years of measurements. Models include a dynamic, individual-based demographic gap model, ZELIG-TROP, and the Ecosystem Demography v.2 (ED2) and CLM-ED dynamic vegetation models. Field measurements were used to generate both model parameterization and benchmarking datasets for the Amazon Basin as a whole, and the northwest Amazon (NWA) and the Central Amazon (CA) separately.
Results/Conclusions:
When comparing models of different scales, CLM-ED over-estimated biomass in the >100 cm size class by 90+ Mg ha-1, thus largely overestimating aboveground biomass, and over predicts growth and mortality rates compared to field data. ED2 underestimated the aboveground biomass stock (231 vs. 312 Mg ha-1) for a CA forest, and similar to CLM-ED exhibits fast growth and mortality rates leading to an over prediction of the Central Amazon carbon sink. Model validations of ZELIG-TROP confirmed that the NWA and CA simulations successfully reproduced observed values of net primary productivity, biomass, and leaf area index, and mortality rates, but predicted slightly lower growth rates. Modeling the existing tree mortality gradients across Amazonia is a complex task, yet essential to reliable prediction of carbon storage in a warmer climate. Using ZELIG-TROP, we evaluated the response of Amazonian forests to elevated mortality rates. The simulated NWA with doubled mortality rates (from 2% to 4%) was found to have a significant decrease in biomass (29.6%) and a slight decrease in NPP compared to a control simulation. However, there was a non-significant shift in community composition in the NWA forests (Wilcoxon rank sum, Z=0.95, p=0.34). When mortality was doubled in the CA significant changes in basal area and community composition were observed (Z=2.28, p=0.02) but this shift did not generate a community composition representative of the observed NWA. Our modeling results suggest that species composition in CA is more sensitive to a doubling of mortality rates than in NWA leading to a larger decrease in biomass in CA (41.9%).