Towards Reliable Sea Ice Drift Estimation in the Arctic: Deep Learning Optical Flow on RADARSAT-2

Accurate estimation of sea ice drift is critical for Arctic navigation, climate research, and operational forecasting. While optical flow has advanced rapidly in computer vision, its applicability to satellite SAR imagery remains underexplored. We present the first large-scale benchmark of 48 deep learning optical flow models on RADARSAT-2 ScanSAR data, evaluated with endpoint error (EPE) and Fl-all metrics against GNSS-tracked buoys. Several models achieve sub-kilometer accuracy (EPE 6-8 pixels, \~300-400 m), capturing coherent regional drift patterns and demonstrating that deep learning computer vision models can be effectively transferred to polar remote sensing. Optical flow provides full spatial drift fields at resolutions beyond sparse buoy networks, offering new opportunities for navigation and climate modeling.