Benchmarking land surface models in the Arctic-Boreal region

Over the past 20-30 years, the Arctic-Boreal region warmed at twice the global rate, leading to widespread ecosystem impacts, including the thawing of permafrost soils. As climate continues to warm, permafrost degradation is projected to increase, exposing a significant quantity of previously frozen carbon to microbial degradation. Understanding when and how much carbon will be released is crucial for determining whether the Arctic will act as a net carbon source or sink in the future. However, models still have large uncertainties when simulating carbon cycle dynamics under changing climate conditions, particularly in permafrost ecosystems. This study uses the International Land Model Benchmarking (ILAMB) software to benchmark TRENDY models against various sets of global and regional datasets to evaluate current model performance both globally and in the Arctic-Boreal region. We found that inferred model skill is somewhat dependent on the regional specificity of the reference datasets used in model benchmarking. Inferred model skill was higher when using global reference datasets of component fluxes (overall Gross Primary Production (GPP) and Ecosystem Respiration (ER) scores of 0.63 and 0.60, respectively, out of a possible score of 1.0) compared to ABoVE datasets specific to the northern high latitudes (overall GPP and ER scores of 0.50 and 0.55, respectively). In addition, inferred model performance was more variable when using ABoVE datasets as benchmarks. Assuming that regionally specific datasets more accurately represent carbon cycle dynamics in high-latitude systems (compared to global datasets), the observation that inferred model skill degrades when using regionally specific datasets provides a false sense of model skill when only using global products. Findings from this study suggest the importance of regional benchmark datasets, particularly in high-latitude ecosystems.