Reducing uncertainty of Arctic ecosystem models through identification of key parameters

Changes in vegetation and permafrost dynamics are altering carbon budgets and might push the Arctic from a carbon sink to a source. Yet, projections from terrestrial ecosystem models remain highly uncertain, limiting their relevance for climate change risk management. The significant mismatch between data availability and data needs for modeling leaves many parameters poorly constrained, and this parameter uncertainty then propagates through the models, making up a significant proportion of the outputs uncertainty. Additional field samples of those parameters can thus be an efficient way to refine Arctic ecosystem models. For this study, we compared three models, DVM-DOS-TEM, ED2, and SiPNET, to evaluate how model structure can impact parameter uncertainty. To identify which parameters should be prioritized in field efforts, we identify the mismatch between data needs and data availability in various databases, including the TRY Plant Trait Database, the Biofuel Ecophysiological Traits and Yield database (BETY), the Arctic Long Term Ecological Research database (LTER), the Next-Generation Ecosystem Experiments Arctic data catalogue (NGEE Arctic), and the Fine-Root Ecology Database (FRED). To further refine which parameters to target, we compare the models to a causal loop diagram of the Arctic ecosystem, which includes unquantified, and thus unmodeled, processes. We map model parameters to processes in the causal loop diagram and identify important feedbacks via the internal network structure. One important network substructure, the so-called feed forward loops, quantify processes that are linked through both a direct and an indirect process, via one intermediary variable. These indirect processes might be implicitly included in the model if parameter values are accurately depicting field conditions. We find that the parameters related to minimum and optimum temperature for photosynthesis are associated with a particularly high number of feed forward loops, but are not constrained well by existing data. Through both methods combined, i.e. the identification and use of parameters from the databases and analysis in the causal loop diagram, we identify which model parameters play a key role in model precision and should be prioritized for further sampling in the field to reduce model uncertainty.