Scale-dependent performance of earth system models in simulating terrestrial vegetation carbon
Model evaluation using observation data and model intercomparison is essential for better understanding their performances and potentially identifying sources of uncertainty. In this study, we investigated ten earth system models (ESMs) that were involved in Coupled Model Intercomparison Project Phase 5 (CMIP5) for their performances in simulating vegetation carbon of terrestrial ecosystems at four scales (i.e., grid, vegetation, biome, and globe). Our aim is to evaluate the ability of CMIP5 ESMs to reproduce vegetation carbon stocks in their historical simulations at different scales and explore the causes for the uncertainties.
The ESMs showed very scale-dependent performances and opposite conclusions could be drawn when evaluating at different scales. Grid cell goodness-of-fit of CMIP5 ESMs was poor but has been greatly improved at the vegetation type scale, the analogues of plant function types in models. Large variability existed in tropic and boreal regions where vegetation carbon stocks were very high. The simulated global vegetation carbon differed as much as 4-fold although six out of the ten ESMs could reproduce global vegetation carbon less than 20% variation of reference data, 556.6 Pg C. Vegetation dynamics and coupled nitrogen cycle might cause global vegetation carbon to decline over time from their initial states. The variation in global vegetation carbon simulated by ESMs partly resulted from their differences in the initial states. Carbon residence time (or plant longevity) could explain the majority of the variability of vegetation carbon across ESMs. To improve the behaviors of the ESMs, more benchmarks with uncertainty analysis of measurement errors are needed to constrain models for their historical simulations; including initial states and plant function types, especially longevity of biomass pools.