Evaluation of simulated soil carbon dynamics in Arctic-Boreal ecosystems

Given the magnitude of soil carbon stocks in northern ecosystems, and the vulnerability of these stocks to climate warming, land surface models must accurately represent soil carbon dynamics in these regions. We evaluate soil carbon stocks and turnover rates, and the relationship between soil carbon loss with soil temperature and moisture, from an ensemble of eleven global land surface models. We focus on the region of NASA's Arctic-Boreal vulnerability experiment (ABoVE) in North America to inform data collection and model development efforts. Models exhibit an order of magnitude difference in estimates of current total soil carbon stocks, generally under-or overestimating the size of current soil carbon stocks by greater than 50 PgC. We find that a model's soil carbon stock at steady-state in 1901 is the prime driver of its soil carbon stock a hundred years later-overwhelming the effect of environmental forcing factors like climate. The greatest divergence between modeled and observed soil carbon stocks is in regions dominated by peat and permafrost soils, suggesting that models are failing to capture the frozen soil carbon dynamics of permafrost regions. Using a set of functional benchmarks to test the simulated relationship of soil respiration to both soil temperature and moisture, we find that although models capture the observed shape of the soil moisture response of respiration, almost half of the models examined show temperature sensitivities, or Q10 values, that are half of observed. Significantly, models that perform better against observational constraints of respiration or carbon stock size do not necessarily perform well in terms of their functional response to key climatic factors like changing temperature. This suggests that models may be arriving at the right result, but for the wrong reason. The results of this work can help to bridge the gap between data and models by both pointing to the need to constrain initial carbon pool sizes, as well as highlighting the importance of incorporating functional benchmarks into ongoing, mechanistic modeling activities such as those included in ABoVE.