Chaopeng Shen (Pennsylvania State University, US); Colin Gleason (University of Massachusetts, Amherst, US); Theodore Langhorst (University of Massachusetts, Amherst, US); Kostas Andreadis (University of Massachusetts, Amherst, US); Casey Brown (University of Massachusetts, Amherst, US)

The Surface Water and Ocean Topography (SWOT) satellite, launched in December 2022, has introduced transformative capabilities for monitoring surface water dynamics. We present an early investigation of the information value of using SWOT-derived lake and river data across the conterminous United States (CONUS) for streamflow simulation. Two distinct approaches — a data-driven model and a physically-based model — were first employed to “reconstruct” sparse SWOT observations (2023-2025) into continuous daily data from 2017 to 2024. In other words, we simulated a time series in response to meteorological forcings in a highly consistent way to SWOT-like data. These reconstructions were then integrated into a data-driven modeling framework to simulate final gauge-observed discharge. Our results demonstrate that incorporating SWOT-related lake and river data can improve streamflow long-term simulation by better informing either the physical parameters or data-driven networks. The improvement is most noteworthy for median flows, presumably due to observational frequency. Notably, absolute values of the SWOT lake water surface elevation data without de-meaning provided more substantial improvements, and applying a roughness penalty during the reconstruction of SWOT lake signals also proved beneficial. Incorporating hydrological prior knowledge was beneficial to effectively extracting and utilizing information from SWOT observations. Expanded temporal coverage of SWOT data expected to enhance our simulations' precision. By demonstrating the readiness of differentiable modeling for global hydrological applications, this study also lays the foundation to assimilate SWOT data for streamflow prediction.