Dhruv Balwada (Lamont-Doherty Earth Observatory, Columbia University,, United States); Tatsu Monkman (Courant Institute of Mathematical Sciences, New York University, United States); Shafer Smith (Courant Institute of Mathematical Sciences, New York University, United States)

Estimating mesoscale and submesoscale driven eddy heat transport in global ocean circulation remains challenging due to the lack of fine-resolution and co-located velocity and tracer observations. Different satellites provide broad spatial coverage for resolving eddy dynamics across different observable variables, but inferring true surface currents and the corresponding tracer transport based on the satellite data is nontrivial.

To address this, we propose a method utilizing diffusion models and other approaches to downscale and transform sea surface height (SSH) and tracer data into higher-resolution surface velocity and tracer flux estimates. As a proof of concept, we are currently testing our approach in idealized quasi-geostrophic models, and comparing the machine learning (ML) approaches with traditional data driven optimal interpolation and stochastic filtering strategies. We hope to develop a hierarchy of approaches that would be capable of providing us refined tracer transport estimates using SWOT and other eddy resolving satellites.