A model intercomparison project to study the role of plant functional diversity in the response of tropical forests to drought

Uncertainty in how the land carbon (C) sink will change over time contributes to uncertainty in Earth system model (ESM) projections of climate change. Much of the land sink is thought to reside in old-growth tropical forests, but recent analyses suggest a diminishing C sink in these forests due to rising temperatures and drought. Thus, there is an urgent need to better understand tropical forest responses to drought and to incorporate this understanding into ESMs. Previous work with vegetation demographic models (VDMs) – which represent the dynamics of individuals or cohorts, along with hydrology and biogeochemistry − suggest that functional diversity can enhance tropical forest resilience to climate change. However, there is little understanding of how different approaches to representing trait diversity and demography affect model outcomes. To explore the potential for trait diversity to moderate tropical forest responses to drought, we explored the behavior of nine VDMs, ranging from models with detailed site-level parameterizations to more generalized land models designed as ESM components. The behavior of each model was studied using soil and meteorological data collected at each of two tropical forest sites: Paracou Research Station, French Guiana, and Tapajos National Forest, Brazil. Low and high trait-diversity scenarios were simulated for each model using historical meteorology, as well as reduced rainfall scenarios. Few models showed strong effects of trait diversity on drought resistance (short-term response of forest biomass to rainfall reduction), but most models showed positive effects of diversity on resilience (long-term recovery of forest biomass following the initial biomass loss due to rainfall reduction). Long-term recovery was always associated with shifts in community composition towards greater drought-tolerance. However, there were large differences among models in the degree and time-scale of recovery. These differences were unrelated to th