A novel satellite embedding-driven SWOT framework for river discharge estimation: a case study of the Yangtze river

Accurate river discharge estimation is essential for water cycle modeling, water hazards prevention, water ecosystem protection and water resource management. However, conventional ground-based gauging networks are often sparsely distributed and insufficient to capture the complex hydrodynamic characteristics within large and heterogeneous river basins. With the rapid advancement of remote sensing technologies, particularly the launch of the Surface Water and Ocean Topography (SWOT) mission, new opportunities have emerged for large-scale and continuous river discharge estimation. In this study, we develop and evaluate a SWOT-based river discharge estimation framework for the Yangtze River by integrating SWOT RiverSP hydraulic observations organized on the SWORD river network with Satellite Embedding V1 data in a random forest model to capture both hydrodynamic and geomorphological characteristics. The framework is evaluated using temporally separated training and testing observations from nine gauging stations along the Yangtze River mainstream, with station-specific nonlinear empirical models used for comparison. The main conclusions are as follows: 1) The random forest model substantially improves discharge estimation relative to the empirical formulation, with a maximum NSE gain of 0.4372 and an NRMSE reduction of up to 0.1175; 2) The model demonstrates reliable performance across the nine gauging stations, with all stations achieving NSE ≥ 0.72 and the overall performance reaching NSE = 0.8910 and NRMSE = 0.0664; 3) The spatially explicit discharge estimates show strong agreement with reach-scale results (correlation ≈ 0.9), indicating that water level exerts primary control on discharge, while channel width and geometry significantly modulate discharge responses in morphologically complex reaches. These results demonstrate the potential of combining SWOT hydraulics with satellite foundation embeddings to improve discharge estimation. Physics-embedding framework improves discharge estimation across nine gauging stations.Random forest improves discharge estimation accuracy (NSE ≥ 0.72).Gridded discharge reveals level dominance with width modulation. Physics-embedding framework improves discharge estimation across nine gauging stations. Random forest improves discharge estimation accuracy (NSE ≥ 0.72). Gridded discharge reveals level dominance with width modulation.