Diagnosing and assessing uncertainties of terrestrial ecosystem models in a multimodel ensemble experiment: 2. Carbon balance

摘要

This paper examines carbon stocks and their relative balance in terrestrial ecosystems simulated by Biome-BGC, LPJ, and CASA in an ensemble model experiment conducted using the Terrestrial Observation and Prediction System. We developed the Hierarchical Framework for Diagnosing Ecosystem Models to separate the simulated biogeochemistry into a cascade of functional tiers and examine their characteristics sequentially. The analyses indicate that the simulated biomass is usually two to three times higher in Biome-BGC than LPJ or CASA. Such a discrepancy is mainly induced by differences in model parameters and algorithms that regulate the rates of biomass turnover. The mean residence time of biomass in Biome-BGC is estimated to be 40-80 years in temperate/moist climate regions, while it mostly varies between 5 and 30 years in CASA and LPJ. A large range of values is also found in the simulated soil carbon. The mean residence time of soil carbon in Biome-BGC and LPJ is similar to 200 years in cold regions, which decreases rapidly with increases of temperature at a rate of similar to 10 yr degrees C-1. Because long-term soil carbon pool is not simulated in CASA, its corresponding mean residence time is only about 10-20 years and less sensitive to temperature. Another key factor that influences the carbon balance of the simulated ecosystem is disturbance caused by wildfire, for which the algorithms vary among the models. Because fire emissions are balanced by net ecosystem production (NEP) at steady states, magnitudes, and spatial patterns of NEP vary significantly as well. Slight carbon imbalance may be left by the spin-up algorithm of the models, which adds uncertainty to the estimated carbon sources or sinks. Although these results are only drawn on the tested model versions, the developed methodology has potential for other model exercises.