IceNav: Risk aware In-Ice navigation using machine learning emulation

Navigation through sea-ice entails considerable risk, with periods when the ship is trapped in ice leading to large fuel expenditure and possible damage to the vessel. British Antarctic Survey, Turing Institute and Warwick University are undertaking a collaborative research project to develop risk aware tools for in-ice navigation in highly dynamic sea-ice conditions. This project aligns with the BAS 2040 carbon reduction targets to meet its net-zero goals.

Sea-ice is a highly dynamic phenomenon with linear openings in the sea-ice, termed leads, allowing quick navigation but with the risk of closure over very short time-scales. Leads open up in the sea-ice depending on wind and other conditions, and allow the opportunity for reduced fuel navigational routes. Quantifying the risk of following these routes requires a detailed understanding of how the ship operates in icetaking into account all the possible predictive sea-ice conditions and their probabilities. 

Current numerical models of route construction could be used but would have to be re-run for all possible environments taking into account the variation in the sea-ice conditions and returning a probabilistic estimate of the optimal low risk routes that could be taken. This approach is computationally expensive and is not-scalable for decision support during underway operations. Recent advances in machine learning techniques allow the emulation of environmental models, generalisable for a range of parameterised conditions. These tools are currently leveraged in weather forecasting  for a speedup of ~50,000x ( FourCastNet cite) with minimal loss in clarity of the forecasts. 

In this project  we are using these cutting-edge techniques to emulate the route construction for in-ice navigation for a range of sea-ice conditions, allowing the recovery of probabilistic risk distribution of different navigational routes through the predicted environment.