From Sea Surface Height to Velocity - A Machine Learning Approach

From Sea Surface Height to Velocity - A Machine Learning Approach

Yaswanth Reddy Yaradoddi, Jinbo Wang and Elyes Serghine report that a deep-learning model can infer upper-ocean velocity from sea surface height (SSH) and environmental data, including at scales below 50 km. Such estimates are challenging because conventional methods rely on geostrophic or cyclogeostrophic balance. The authors trained a U-Net encoder--decoder on hourly fields from a pre-SWOT Cape Basin extraction of the LLC4320 ocean simulation, using SSH, sea surface temperature (SST) and wind stress as inputs. At 50 m, they report a speed RMSE of 0.049 m s−1, 29% lower than the geostrophic approximation, and relative-vorticity correlation of 0.93, versus 0.75 for that approximation. They also report divergence correlation of 0.66, while geostrophic velocity is nondivergent by definition. Their training tests found that SSH gradients outperformed raw SSH, wind stress was the most useful covariate for total velocity prediction, SST benefits varied seasonally, and including vorticity and divergence in the loss improved accuracy. The authors present these results as a path toward SWOT applications, where kilometer-scale SSH snapshots contain balanced and unbalanced signals. The reported evaluation uses simulated fields, so how well the approach transfers to SWOT observations remains an open question.

Publication: 8deacb30-284c-4e6d-843c-2b6cd72f1dce

attribution — pass: The reading correctly attributes the work and its reported results to Yaswanth Reddy Yaradoddi, Jinbo Wang and Elyes Serghine. It presents the findings as their reports rather than as established fact.

Authors:Yaswanth Reddy Yaradoddi, Jinbo Wang, Elyes Serghine

summary — uncertain: The figures are faithful, but central terms are not explained for a general reader. SWOT is never expanded or described as a satellite mission. Geostrophic balance, U-Net, RMSE, relative vorticity, divergence, and balanced versus unbalanced signals are also unexplained, so the significance stays opaque. Smallest repair: expand SWOT as the Surface Water and Ocean Topography mission, add plain-language glosses or term tooltips for geostrophic approximation, RMSE and vorticity, and note that the velocity targets are at the surface and at 50 m. The title and question are fine.

scales now resolved by the Surface Water and Ocean Topography (SWOT) mission

date — pass: The displayed original publication date, 2026-11-03, matches the v1 submission date shown on the arXiv page. It is not the later feed date of 2026-11-04.

[v1] Tue, 3 Nov 2026 03:58:05 UTC (2,164 KB)

support — uncertain: The summary's opening says the model can infer velocity including at scales below 50 km. The abstract describes sub-50-km scales as the challenge and reports results on a 1.5-km grid. It does not state a scale-specific evaluation showing success below 50 km. The draft also says estimates are challenging because conventional methods rely on balance assumptions; the source states these points side by side, not as a cause. Smallest repair: drop 'including at scales below 50 km' or attribute it as the motivating problem, and replace 'because' with neutral juxtaposition. The simulation-only interpretation is properly labeled.

Inferring upper-ocean velocity from altimetric sea surface height (SSH) remains challenging below 50~km

media — pass: The selected media is the verified direct link to the arXiv abstract page for this paper. It is the most relevant way for readers to reach the original source.