Feature-Relative Deployment Optimisation of Ice-Tethered Buoy Arrays in the Southern Ocean

James Hepworth , Edward Boje, Amit Kumar Mishra

The Southern Ocean marginal ice zone is critical for climate dynamics and sea-ice model validation, yet remains severely undersampled due to costly deployment and short buoy lifetimes. The sea-ice edge shifts by hundreds of kilometres between years, so fixed-coordinate optimisation fails when future ice conditions are unknown. This paper presents a robust multi-objective framework that encodes buoy positions relative to the dynamic ice-edge feature, so one genotype adapts to inter-annual variability. NSGA-II optimises spatial coverage and ship-operating cost. Training uses 39 years of Polar Pathfinder data (1979--2017). Blind validation covers seven held-out years (2018--2024), and cross-product transfer uses two independent drift products (DPGSIV AMSR2 and OSI-405-c). The feature-relative strategy retains 99.4\% coverage on unseen years and outperforms fixed-coordinate, grid, random and historical fan baselines at lower cost. It transfers to alternative products, giving a quantified decision-support tool for pre-cruise planning.