WV-Net: SAR Ocean Foundation Model

A self-supervised learning approach to train a SAR open ocean foundation model on 12+ million images from the ESA Sentinel-1 satellite. WV-Net enables retrospective analysis of 9 years of global-coverage data and identifies extremely rare patterns with above 90% mean average precision.

The model improves classification and regression scores by up to 40% across downstream problems in climatology, atmospheric sciences, ocean sciences, and vessel monitoring.

Published in AI for the Earth Systems (AMS, 2025).