River bathymetry and discharge estimation using video surface velocity data assimilation in the 2D differentiable numerical shallow water model dassflow

This contribution presents a Variational Data Assimilation approach for the integration of water surface velocity fields into a 2D shallow water hydraulic model and the inference of distributed bathymetry and inflows. It describes the addition of flow velocity observations into an existing assimilation framework. A regularizing effect is introduced in the form of a simple bathymetry model. Synthetic experiments are designed to study the identifiability of the sought parameters from velocity observations and show that the assimilation method can leverage the informational content of synthetic observations in a twin experiment setup. The method notably allows good inferences of distributed upstream inflows from observations of their mixed velocity signatures at a downstream confluence. The method is then applied to a real river reach where fair inferences of the local cross-section are obtained. This combination of the ANDROMEDE and DassFlow platforms could enable fast, automated, physically-based estimations of distributed bathymetry and inflows from unintrusive measurements.