Work

Estimation of River Discharges from SWOT Satellite Measurements with Uncertainty Quantification

Year: 2023
Type: preprint
Abstract: Estimating river discharge for ungauged or poorly gauged basins is a critical challenge in hydrology. The SWOT satellite provides measurements of water surface elevation, width, and slope but does not measure discharge or bathymetry. This paper introduces a novel version of the HiVDI algorithm, named HiBaVDI (Hierarchical Bayesian Variational Discharge Inference), which combines Bayesian inference and Variational Data Assimilation (VDA) to estimate river discharge from SWOT measurements while quantifying uncertainty. A key point of the method consists in relaxing the strong dependencies between friction and discharge through a physics-informed change of variables, enabling probabilistic estimation of bathymetry, friction, and discharge. The Bayesian step provides first estimates with uncertainty quantification, while VDA subsequently calibrates hydrodynamic models. The paper also examines the ill-posedness of the inverse problem, demonstrating that the classical 1D flow model yields infinitely many solutions, while more complete shallow flow models would theoretically resolve this issue. Evaluated across four scenarios (depending on the available priors), HiBaVDI achieves high accuracy on the synthetic datasets Pepsi-1 and Pepsi-2 (covering 48 rivers worldwide), with median posterior errors lower than 15%.
Source: HAL
Language: English
Open Access status: green
OpenAlex ID: W7155584236
Cites:
Cited by:
Related to: 10