A new generative AI model has filled in the gaps in satellite ocean data, revealing that narrow submesoscale fronts do a disproportionate share of the work moving energy through the ocean's currents.
Researchers trained a generative deep learning framework on high-resolution ocean simulations, then paired it with multi-source satellite observations to reconstruct gap-free surface currents at kilometer scale, finer than existing instruments can measure directly. They applied the method to the Agulhas Current, one of the ocean's most eddy-dense systems, off the southern tip of Africa. The reconstruction let them trace kinetic energy across scales: above 10 kilometers, submesoscale motion feeds energy upward into mesoscale eddies and shapes their seasonal cycles, while below 10 kilometers, convergence at sharp fronts drains energy downward toward dissipation. Both transfers concentrate heavily within the fronts themselves, where energy moves up to ten times more efficiently than in surrounding water.
That matters because the ocean models used for weather and climate forecasting simplify eddy behavior with parameterizations that have never had direct evidence for what's happening at these small scales. This study gives modelers a specific, data-backed target: fronts cover a small fraction of the ocean's surface but account for a large share of its energy cascade. Get that wrong, and forecasts of eddy behavior, and the heat and carbon they move, inherit the error.
It's the same trick behind recent AI weather models: learn physics from simulations, then use it to see what instruments can't measure directly. Here, that means watching an energy pathway nobody could previously observe at all.