Continuous Water Surface Elevation Estimates Using Deep Learning with Legacy Altimetry and Surface Water and Ocean Topography Data

Chinmay Deval, Alqamah Sayeed, Ashutosh Limaye

We present the development of a high-temporal-resolution global dataset of daily river water surface elevation (WSE), spanning January 2008 through May 2025. By utilizing a deep learning framework to integrate legacy satellite altimetry and the Surface Water and Ocean Topography (SWOT) mission data, we produced a continuous record covering 9,184 river reaches, 5,926 rivers, and 1,342 basins. The densification process successfully reconciles the sparse temporal sampling of missions such as Jason and Sentinel with the spatial precision of the SWOT Ka-band interferometer. Evaluation metrics demonstrate robust performance, with a median correlation (R) of 0.46 against Hydroweb calibration data and a superior median R of 0.62 against independent SWOT observations. Spatial analysis confirms that the model maintains near zero global median bias relative to legacy data. This multidecadal resource provides a critical analytical foundation for global monitoring and the management of freshwater resources under varying climatic zones. Ultimately, this dataset bridges the strategic gap between sparse legacy altimetry observations and the high-resolution capabilities of the contemporary SWOT mission.