Operational SAR-Based Sea ICE Concentration Products for Copernicus Marine Service

The Arctic is currently facing unprecedented changes due to anthropogenic warming, leading to an accelerated melting of sea ice and an increase in human activity in the region. The expanding maritime community accessing wider areas of the Arctic emphasizes the need for detailed and timely information on the state of Arctic sea ice for maritime safety and planning. Synthetic Aperture Radar (SAR) imagery have been playing a crucial role in providing year-round mapping of sea ice conditions due to its high spatial resolution (approx. 100m), independent from solar illumination, and the ability to penetrate cloud cover. National Ice Centers worldwide primarily rely on SAR imagery, where ice analysts manually interpret the data to produce sea ice charts for maritime users. However, the task becomes laborious and time-consuming with the growing availability of satellite imagery. To address this challenge, there is a move towards partial automation of the process, aiming to assist ice analysts in delivering high-resolution sea ice products in near-real-time. A fully automated sea ice mapping system is also envisioned to integrate high-resolution sea ice products into forecast models, potentially improving forecast quality. We compiled a vast matched dataset of Sentinel-1 HH/HV imagery and AMSR2 brightness temperatures to train a Convolutional Neural Network-based algorithm with regional ice charts as labels. The retrieval methodology, denoted ASIP, achieves an R-score of 95.5% against a held-out test dataset of regional ice concentration charts.