Initial Results from SAR-Based Validation of Sea Ice Drift Forecast Models

TOPAZ4 and neXtSIM sea ice drift forecast model data, provided in the CMEMS artic analysis and forecast products PHYS_002_001_a and PHY_ICE_002_011, is assessed using Sentinel-1 Synthetic Aperture Radar (SAR) image pairs. This work helps to improve the high-resolution forecast Predictive Ice Image (PRIIMA) (https://business.esa.int/projects/priima) application which generates multi-day forecasted SAR scenes. Results of PRIIMA form the basis for the calculation of possible ship routes in ice-infested waters, which are then made accessible for navigators in near real-time.

The true sea ice drift is derived from two successive Sentinel-1 scenes using phase correlation applied in a hierarchical resolution pyramid. The measured ice displacement vectors are compared to predicted drift trajectories derived from multi-day TOPAZ4 and neXtSIM forecasts available at acquisition time of the first SAR scene. Differences in sea ice drift direction and magnitude between models and actual SAR measurements are then analysed visually and statistically. A feature-based approach of data handling enables the comparison of drift data stored in vector format and model data stored in raster format. The analysis benefits from the high coordinate precision of vector models and a fine granularity of a grid spacing down to 500 m.

This initial study focuses on time differences of 1 to 3 days in two regions of interest (RIOs). The first ROI is located offshore close to Cape Morris Jesup at the transition between Lincoln and Wandel Sea. The second ROI is placed at open sea above Alpha Ridge near Ellesmere Island. A small initial sample of seven image pairs is analysed to test the approach, from which some preliminary conclusions can already be drawn: The root-mean-square separation distance between the end points of the measured drift vectors and the derived model trajectories (RMS separation distance) is around 4 to 5 km per day for both TOPAZ and neXtSIM. The RMS separation distance shows a high variability across the investigated image pairs. In the first ROI, a high RMS separation distance in the neXtSIM model was found near Cape Morris Jesup in the case of coastward drift or drift around the cape.

In future works, the conducted analysis will be extended to a one-year analysis in selected regions arctic wide, taking into consideration the influences of sea ice deformation, atmosphere and ocean.