Prediction of navigation risks in the Arctic Northeast Passage based on Bayesian and neural networks

This study proposes a hybrid method for predicting and analysing navigation risks in the Arctic Northeast Passage, where the accelerated melting of sea ice due to global warming has led to increased maritime traffic. Ensuring the safe navigation of ships in ice-covered waters has become a critical issue. A Bayesian network model is developed to quantify the risks associated with ship besetting in ice and ship-ice collisions. The model comprehensively incorporates environmental, ship-related, human, and organisational factors. Expert knowledge and satellite-based environmental monitoring data are combined to parameterise the Bayesian network, while probability distributions and Monte Carlo simulation are employed to address the uncertainty arising from experts’ subjective judgements. The probabilistic risk estimates generated by the Bayesian network are subsequently used to train a neural network, enabling the model to capture complex non-linear relationships and enable rapid prediction of navigation risks. This integration leverages the interpretability of probabilistic reasoning and the adaptability of machine learning. The proposed method exhibits strong predictive performance and produces spatial visualisations of navigation risk under varying conditions, providing robust decision-support for route planning and enhancing navigational safety in Arctic waters.