Researchers led by UC San Diego's Scripps Institution of Oceanography have developed GOFLOW, a deep learning technique that converts thermal images from geostationary weather satellites into high-resolution maps of ocean surface currents. It reveals fast-changing, sub-10-kilometer features vital for climate, heat/carbon uptake, and marine ecosystems, with results published in Nature Geoscience (DOI: 10.1038/s41561-026-01943-0).
Luc Lenain, an oceanographer at Scripps, and Kaushik Srinivasan (now at UCLA) led the development of GOFLOW—Geostationary Ocean Flow—after spotting dynamic temperature patterns in North Atlantic Gulf Stream satellite data from 2023. The neural network, trained on simulated ocean currents, analyzes real sequences of thermal images from satellites like GOES-East (captured every five minutes), tracking how patterns bend, stretch, and shift to infer current speeds and directions. Co-authors include Roy Barkan (Tel Aviv University) and Nick Pizzo (University of Rhode Island), with funding from the Office of Naval Research, NASA, and the European Research Council.
Unlike traditional polar-orbiting satellites (revisit every 10 days), ships, or radars (limited coverage), GOFLOW generates hourly maps of small-scale features like eddies that drive vertical mixing—previously observable only in models. Validation against 2023 Gulf Stream ship data showed strong matches, outperforming topography-based methods in resolution.
'Weather satellites have been observing the ocean surface for years,' Lenain said. 'The breakthrough was learning how to turn that time-lapse into hourly maps of currents by tracking how temperature patterns bend, stretch and move from one hour to the next.' He noted it enables real observations of intense currents key to heat and carbon uptake studies.
No new satellites are required, though cloud cover remains a challenge. The team plans global expansion and has publicly released data and code.