PermaBN: A Bayesian Network framework to help predict permafrost thaw in the Arctic.

Rising global temperatures are a threat to the current state of the Arctic. In particular, permafrost degradation has been impacting the terrestrial cryosphere in many ways, including effects on carbon cycling and the global climate, regional hydrological connectivity and ecosystem dynamics, as well as human health and infrastructure. However, the ability to simulate permafrost dynamics under future climate projections is limited, and model outputs are often associated with large uncertainties. A model structured on a Bayesian Network is presented to address existing limitations in the representation of physically complex processes and the limited availability of observational data. A strength of Bayesian methods over more traditional modeling methods is the ability to integrate various types of evidence (i.e., observations, model outputs, expert assessments) into a single model by mapping the evidence into probability distributions. Here, we outline PermaBN, a new modeling framework, to simulate permafrost thaw in the continuous permafrost region of the Arctic. Pre-validation and expert assessment validation results show that the model produces estimations of permafrost thaw depth that are consistent with current research, i.e., thaw depth increases during the snow-free season under initial conditions favoring warming temperatures, lowered soil moisture conditions, and low active layer ice content. Using a case study from northwestern Canada to evaluate PermaBN, we show that model performance is enhanced when certainty about the system components increases for known scenarios described by observations directly integrated into the model; in this case, insulation properties from vegetation were integrated to the model. Overall, PermaBN could provide informative predictions about permafrost dynamics without high computational cost and with the ability to integrate multiple types of evidence that traditional physics-based models sometimes do not account for, allowing PermaBN to be applied to carbon modeling studies, infrastructure hazard assessments, and policy decisions aimed at mitigation of, and adaptation to, permafrost degradation. • Bayesian networks can integrate various types of evidence into a single model. • Model performance is enhanced when certainty about the system components increases. • Model estimates of permafrost thaw depth consistent with current research. [ABSTRACT FROM AUTHOR]