🌊 An AI maps the Gulf Stream and other currents like never before

Weather satellites are not only used to forecast the weather: they also provide valuable information about ocean currents.

By observing their thermal images, scientists have identified the patterns of the Gulf Stream and other major circulations. From this information, GOFLOW was born, an artificial intelligence system that, every hour, converts these snapshots into maps of water movements without needing any new satellite.


Indeed, ocean currents play a central role in regulating climate, transporting heat and carbon, and distributing nutrients. Yet, measuring them on a large scale remains a challenge. Altimetry satellites only overfly a given area every ten days, too infrequently to capture rapid changes. Ships and coastal radars offer better temporal resolution, but only over limited areas.

Thus, GOFLOW bypasses these limitations by using geostationary satellites such as GOES-East, which take images every five minutes. The team led by Luc Lenain trained a neural network on high-resolution ocean simulations. The model learns to relate deformations in temperature patterns to current velocities. By analyzing sequences of images, it can infer the currents responsible for the observed changes.

Tests compared GOFLOW's results to direct measurements taken by ships in the Gulf Stream in 2023. The agreement was excellent, and the new method revealed much finer details than traditional techniques. In particular, it detected eddies and boundary layers that previously disappeared into averages. These structures are important for understanding vertical exchanges between the surface and the depths. These results were published in the journal Nature Geoscience.

This breakthrough opens promising avenues for climatology. The method could be integrated into climate models and weather forecasts, thereby improving our understanding of air-sea interactions and the transport of marine debris.

However, cloud cover remains an obstacle, as clouds block thermal images. The team plans to combine other types of satellite data to achieve continuous coverage. Work is already underway to extend the method to a global scale.

The codes and data are publicly available, allowing other researchers to develop new applications.