Biao Lu (Technical University of Denmark, DTU Space, Denmark); Farshad Salajegheh (School of Engineering, University of Newcastle, Callaghan, NSW, Australia); Benjamin J. Phrampus (United States Naval Research Laboratory, Stennis Space Center, MS, USA); Jonathan Kirby (Technical University of Denmark, DTU Space, Denmark); Paul Elmore (Applied Physics Laboratory, Johns Hopkins University, Laurel, MD, USA); Xiaoli Deng (School of Engineering, University of Newcastle, Callaghan, NSW, Australia); Luis Altamirano (The University of Southern Mississippi, Division of Marine Science, MS, USA); Yao Yu (Scripps Institution of Oceanography, University of California San Diego, La Jolla, CA, USA); Walter H. F. Smith (8 Center for Satellite Applications and Research, National Oceanographic and Atmospheric Administration, College Park, MD, USA); Ole B. Andersen (Technical University of Denmark, DTU Space, Denmark); David Sandwell (Scripps Institution of Oceanography, University of California San Diego, La Jolla, CA, USA)

The introduction of high-resolution sea surface observations provided from the Surface Water and Ocean Topography (SWOT) satellite has enabled the generation of marine gravity fields at 8-km resolution with 1-2 mGal accuracy using just one year of data, representing a major improvement over 30 years of conventional nadir altimetry. The low noise in the short-wavelength observations from SWOT is critical for bathymetric inversion, where gravity precision strongly impacts seafloor predictions. To take full advantage of the possibilities provided with the new data from SWOT, an international workshop at the Technical University of Denmark (DTU) was held in November 2024 with the goal of utilizing the high-resolution gravity field from SWOT, together with Machine Learning models, to provide the best possible bathymetric maps from ocean gravity.

We present key results from this workshop, including:
1. Bathymetry predictions produced by five research groups using different ML architectures. All models demonstrated significant global improvements in accuracy (20–30%) compared to the previous bathymetry release (V32), with the largest enhancements over continental shelves and deep ocean trenches.
2. Cross-model comparisons that revealed how different ML design choices affect predictive performance and identified persistent challenges posed by the sparse distribution of ship-based depth soundings, which serve as training labels.
3. Comparison with conventional physics-based inversions, showing that machine learning methods resolve nonlinear components in the gravity-topography transfer function, which improve prediction over seamount summits.