A data-driven Bayesian network methodology for predicting future incident risk in Arctic maritime-based cargo transit

This paper develops an Arctic incident risk evaluation and prediction tool for Arctic maritime transit of cargo along three key Arctic passageways. The tool computes incident probabilities by incident category (grounding, foundering, equipment failure and collision/allision), severity and Arctic subregion through a data-driven Bayesian network methodology. These probabilities, along with estimated consequences by incident type and severity level, are employed in the computation of historical Arctic subregion and route-based risks, where the route-based risks account for exposure. The structure and parameters of the proposed BN are learned from assessing causal relationships using historical incidents and their consequences, environmental conditions, and expert knowledge. The trained BN is applied on predicted navigation conditions with suitable fidelity to forecast future subregion and route-based incident probabilities and risks in future Arctic passage. The tool is applied on a climate scenario with changing environmental conditions given probabilistic predictions obtained from a state-of-the-art climate model over a 50-year time horizon. Arctic-route incident and risk predictions are provided through 2069. Additional models were employed to derive dynamically downscaled projection of sea ice and provide estimates of future Arctic wave climate, enabling high-resolution simulation of Arctic shipping conditions. Resulting predictions can support planning for future Arctic operations.