Recent advances in satellite altimetry, most notably the Surface Water and Ocean Topography (SWOT) mission and its Ka-band Radar Interferometer (KaRIn), enable sea-level measurements at unprecedented spatial resolution and with significantly improved spatial coverage. These developments provide new opportunities to investigate ocean dynamics at sub-mesoscale scales that have previously been difficult to observe.

This study explores how the use of a high-resolution geoid serving as a realistic vertical datum representing the Earth’s equipotential surface allows the derivation of dynamic topography (DT) from SWOT sea-level observations. The resulting DT reveals sub-mesoscale eddy structures that are not adequately resolved by conventional hydrodynamic models.

To further enhance spatial resolution, we investigate a reconstruction methodology based on a generative deep-learning framework, employing conditional diffusion models (CDM-SL). Coarse-resolution SWOT sea-level observations (approximately 2 km horizontal resolution) are used as inputs to reconstruct sea-level fields at a spatial resolution four times finer than the original data.

The combined use of high-resolution geoid information and deep-learning-based reconstruction is demonstrated in the Baltic Sea, focusing on the Baltic Proper, Bothnian Bay, and the Gulf of Finland. The results reveal the presence of persistent and semi-persistent eddies that influence both surface and subsurface dynamics, highlighting the potential of this approach to improve the observation and understanding of sub-mesoscale processes in semi-enclosed seas.