Tel Aviv University's GOFLOW uses AI to map fine ocean currents from satellite images

An international team led by Prof. Roy Barkan of Tel Aviv University's Department of Geophysics has developed GOFLOW, an artificial intelligence system that reconstructs detailed ocean current patterns directly from satellite pictures. The work was published in Nature Geoscience and publicised by the university in spring 2026. Co-authors came from the Scripps Institution of Oceanography, the University of California, Los Angeles and the University of Rhode Island.

Many of the ocean's most important movements happen at scales of a few tens of kilometres or less. These small fronts and eddies stir the water, carry heat and dissolved gases between the surface and the deep, and shape how the sea trades heat and carbon dioxide with the air. Standard satellite methods mostly miss them, because they rely on measurements of sea-surface height that are too coarse, or on simplified physics that does not hold at these scales.

GOFLOW takes a different approach. It was trained on high-resolution computer simulations of the ocean and then applied to sequences of infrared satellite images that record sea-surface temperature. From the way temperature patterns shift between images, it infers how the surface water is moving, without the restrictive assumptions of earlier techniques. Applied to the Gulf Stream, it revealed processes smaller than 30 kilometres linked to sharp temperature fronts and converging currents.

The team says GOFLOW also delivered the first direct satellite estimates of horizontal divergence, a measure of where water is rising or sinking. Until now that could only be estimated from models or scattered ship measurements. The researchers expect the tool to help scientists study storm strength, marine heatwaves, the spread of pollutants and the ocean's uptake of carbon dioxide, and to feed better data into climate models.