Animation Caption: Sea Surface Height Anomaly (SSHA) from -30 cm (blue) to 30 cm (red) over the global ocean, arrows on top represent the ocean currents. The circular shapes of those anomalies are mesoscale eddies. The panels present a side-by-side comparison between the NeurOST method and the conventional method widely used. First, the SSHA, then the derived ocean current speed and finally the vorticity are shown to illustrate NeurOST’s capability to retrieve fine details of the ocean eddies.

Mesoscale eddies, circular currents of water with diameters spanning from 50 to 300 kilometers, transport freshwater, heat, carbon, nutrients, etc. around the world and within the ocean, profoundly impacting marine ecosystems and the Earth's climate. They also account for over 80% of the ocean’s kinetic energy, the energy created by the movement of ocean water. Mesoscale eddies are therefore critical components of the global ocean circulation and climate, and need to be better characterized and understood.

Satellite altimetry measuring Sea Surface Height (SSH), from which ocean currents can be deduced, has been a valuable tool in tracking these ocean mesoscale eddies over the past three decades¹. However, tracking eddies from SSH observations poses significant challenges due to the limitations of conventional nadir altimeters, such as Sentinel-6 Michael Freilich. These instruments measure sea level along the nadir track, which leaves large gaps between tracks. Therefore, it is necessary to interpolate these observations using an a-priori estimation of the relationship between different measurements in space and in time. This results in global grids of SSH with spatial and temporal resolutions of ~100km and 10 days, respectively, which limits the two-dimensional eddy reconstruction and tracking. Researchers are starting to leverage the recent advancement of machine learning to overcome those limitations. The recently published SSH gridded product, NeurOST, on PO.DAAC is a notable example.

The novel method combines Sea Surface Temperature anomaly (SSTA) and along-track nadir altimeter SSH anomaly (SSHA) data to create a new daily higher resolution SSHA and ocean current dataset². This artificial intelligence (AI) method called deep learning teaches computers to recognize patterns and connections in space and time in a combination of data, here SSHA and SSTA. The resulting animation above, spanning from January 2018 to December 2023, displays NeurOST SSHA estimates with ocean currents on top as arrows. The height of the sea surface has highs (red) and lows (blue) indicating ocean eddies and currents. The animation then zooms in a very energetic area in the western Pacific Ocean and shows some comparisons between this new artificial intelligence method and the observations from conventional gridded altimetry products (Data Unification and Altimeter Combination System; DUACS). First, the SSHA are displayed, then the ocean current speed computed from both products and finally, the vorticity, that represents the rotation of the ocean water mass. We can see the improved ability of this new AI product to capture features such as ocean eddies. This suggests that AI methods such as deep learning can be powerful tools to study our oceans.

The machine learning methodology and data product were developed by an Ocean Surface Topography Science Team at the University of Washington and sponsored by NASA Physical Oceanography program (PO). The method was published in two publications^2,3 and welcomes further investigations. The project is emerging as an essential component of the newly established NASA Ocean AI Working Group, an important step forward by NASA PO teams to advance modern satellite data synthesis. By leveraging cutting-edge machine learning techniques, researchers can explore how to effectively process and analyze vast amounts of satellite data to enhance understanding of the Earth's oceans and ultimately improve climate predictions.