Reconstructing ocean mixed layer variability from SWOT using machine learning

The recently launched Surface Water and Ocean Topography (SWOT) satellite is capable of measuring sea surface height anomaly (SSHA) at fine spatial scales (1–10km). We use machine learning to identify a relationship between these fine-scale SSHA measurements and upper ocean density profiles, particularly in the mixed layer, which mediates critical air-sea interactions impacting climate and weather patterns. We train models to learn vertical density profiles as a function of SSHA, and optionally Sea Surface Temperature (SST), representing density profiles in a low-dimensional space using principal component analysis. Our results indicate that using high-resolution SWOT patches improves prediction compared to single-point inputs. Trained on data from April to July 2023, the model can accurately predict density from October 2023 to October 2024, capturing seasonal variations in mixed layer density even without explicit seasonal indicators such as SST.