Fusion of Arctic sea ice drift results from multi-frequency and multi-polarization AMSR2 data using a random forest approach

Accurate monitoring of sea ice drift has become increasingly crucial for climate research and safe navigation in polar regions. Existing sea ice drift extraction methods based on passive microwave data utilize single-frequency data, which have limitations in both accuracy and spatial coverage. In this study, we propose a random forest (RF)-based fusion method for extracting sea ice drift by integrating multi-frequency sea ice drift results obtained from Advanced Microwave Scanning Radiometer 2 (AMSR2) data. Validation against buoy measurements demonstrates that our fused results consistently outperform both single-frequency and single-polarization, averaging fusion, and linear fusion results throughout the year. During the freezing season, the RF fusion results yield [Formula: see text] improvements of up to 0.07 and reduce the root mean square error (RMSE) by up to 0.42 km/d. Even greater enhancements are observed for the melting season, with [Formula: see text] increases of up to 0.13 and RMSE reductions of up to 0.83 km/d. Additionally, the method improves the spatial coverage of valid sea ice drift extractions by at least 6% each month, peaking at 14.5% in June. The RF fusion sea ice drift results offer valuable data support for global climate change research and ensure safe navigation in Arctic regions.