SAR-Based Sea Ice Drift and Lagrangian Tracking for Evaluating Sea Ice Drift Forecast Models

Kurzfassung

We use sea ice drift information obtained from Synthetic Aperture Radar (SAR) observations to assess the usability of sea ice drift forecast models for multi-day sea ice analysis. This work demonstrates the capabilities of a vector approach on a small case study of twenty analysed SAR-scene pairs in winter 2022/2023 in the Baffin Bay with 12.5km and 0.5km grid spacing. Finding optimal shipping routes through sea ice becomes increasingly important for navigation in polar regions. Sea ice drift forecasts such as TOPAZ4 and neXtSIM provide a prediction of the sea ice situation several days (up to 10) into the future. We apply an approach that combines methods from the interdisciplinary field of environmental physics, geoinformation technology and remote sensing. Sea ice drift vector fields are obtained from successive Sentinel-1 image pairs using the phase correlation technique applied in a hierarchical resolution pyramid. The derived drift is compared to historical multi-day sea ice drift forecast of TOPAZ4 and neXtSIM, provided in the EU CMEMS (Copernicus Marine Environment Monitoring Service) Artic analysis and forecast products PHYS_002_001_a and PHY_ICE_002_011. The forecast model trajectories and SAR-based measurements are both calculated starting from a regular grid. This allows working either in a raster or a vector model. We use the vector model for data processing to benefit from high flexibility and floating-point number coordinate resolution. An object-oriented-programming (OOP) approach, a topology, hashing and spatial indexing yield a computational performance that can compete with raster analysis. Forecast model trajectories are derived with Lagrangian Tracking. The deformation parameters divergence, vorticity and shear are calculated by applying Sobel kernels. With this developed workflow, sea ice motion and deformation predicted by TOPAZ4 and neXtSIM can be evaluated by SAR-based drift measurements. The present work is realized in the context of the FAST-CAST 2 project, funded under grant 19F2191A by the German Federal Ministry for Digital and Transport’s mFUND programme.