OSISAF-to-MASAM2: Deep Learning for Sea Ice Concentration Super-Resolution Product

Abstract

This letter presents a deep learning approach for Arctic sea ice concentration (SIC) downscaling based on multisensor fusion applied during the model training stage. While conventional downscaling methods primarily enhance spatial resolution, the proposed approach also improves physical consistency by fusing the widely used Ocean and Sea Ice Satellite Application Facility (OSISAF) SIC dataset with the high-resolution multisensor-corrected MASAM2 product. This synthesis not only increases spatial resolution from 25 to 4 km, but also yields more realistic ice patterns, as validated by a high structural similarity metric and a mean absolute error (MAE) of 0.02. A comparison of U-Net-based and attention U-Net architectures with classical interpolation methods, a conditional diffusion baseline, and a manual hybrid approach demonstrates the advantages of the proposed solution. The projection change between the two products is embedded in the network as a fixed resampling layer, removing the external remapping stage from the pipeline. The presented model has strong potential to improve sea ice forecasting quality and generate high-resolution historical datasets. Code and weights are provided on https://github.com/ITMO-NSS-team/GRSL_masam2_downscaling