Mohamed Saadi (Institut de Mécanique des Fluides de Toulouse (IMFT), Université de Toulouse, CNRS, Toulouse, France); Pierre-André Garambois (INRAE, Aix-Marseille Université, RECOVER, Aix-en-Provence, France); François Colleoni (INRAE, Aix-Marseille Université, RECOVER, Aix-en-Provence, France); Truyen Huynh (INRAE, Aix-Marseille Université, RECOVER, Aix-en-Provence, France); Kevin Larnier (Hydro Matters, Toulouse, France); Ludovic Cassan (CERFACS, Toulouse, France); Hélène Roux (Institut de Mécanique des Fluides de Toulouse (IMFT), Université de Toulouse, CNRS, Toulouse, France)

Accurate simulation of flood propagation and submersion at regional to continental scales requires proper representation of river channel and floodplain geometry, which significantly influences fluvial hydrodynamics. However, detailed bathymetric data (e.g., channel storage, thalweg elevation, bank slope) are often unavailable across large domains, leading to a simplified representation of river bathymetry in hydrological-hydrodynamic modelling applications. This poor representation of river bathymetry introduces uncertainties in model outputs (e.g., peak time and magnitude, depth, inundation extent), particularly during extreme flooding events. The increasing availability of multi-scale remote sensing altimetry data (e.g., SWOT, LIDAR) offers a valuable opportunity to fill this gap in data-sparse regions. By assimilating these remote sensing observations (e.g., water surface elevation) into integrated hydrological-hydrodynamic modeling frameworks, hydraulic model parameters such as bathymetry and roughness can be refined through inverse modeling enabling more accurate flood prediction in data-sparse regions. To address these challenges, this work introduces a hybrid raster-vector hydrologic-hydrodynamic modeling framework combining grid-based hydrological modeling with vector-based river routing. This approach provides the spatial structure needed for effective assimilation of remote sensing observation data. We developed a fully automated preprocessing workflow that requires minimal inputs: a fine-resolution digital elevation model (DEM) and a reference vector hydrography. The preprocessing workflow consists of four sequential steps: (1) DEM conditioning through stream burning and depression filling; (2) flow direction computation at the DEM's native resolution followed by upscaling to target hydrological model resolution; (3) coarse-scale hydrography delineation by tracing reference vector river pathways along upscaled flow directions rather than using the conventional support area threshold approach; and (4) coupling the hydrological grid to the DEM-derived hydrography through direct cell-to-cell index matching. This workflow preserves the fine-scale river features through a subgrid-scale representation of the river network at the native DEM resolution, thereby overcoming dependency on the grid resolution of the hydrological model. The preprocessing workflow was implemented within SMASH (Spatially distributed Modelling and ASsimilation for Hydrology), a computational software framework for hydrological modeling. We evaluated the workflow on the Garonne basin in France using MERIT DEM and the SWORD river network across three spatial resolutions (approximately 250m, 500m, and 1 km). The quality of the DEM-derived subgrid network was assessed through two metrics: mean separation distance (MD) and length ratio (LR). Results showed excellent performance across all resolutions, with the subgrid network maintaining close spatial alignment to the SWORD reference network (MD of 16-17 m) while successfully preserving fine-scale features (LR consistently at 1). A modeling numerical experiment was conducted within SMASH on the Garonne basin at 1 km resolution coupling the grid-based conceptual GR hydrological model to a vector-based routing model solving a simplification of the 1D shallow water equation without convective acceleration terms (sufficient approximation for low Froude regimes in SWOT context). For this experiment, river channels were represented using a simple rectangular geometry with width and depth parameters estimated from drainage area-based empirical relationships. The results demonstrated excellent mass balance conservation between the hydrological grid and the vector network, with very low errors. This high precision confirms the effectiveness of the cell-to-cell index matching approach for spatial coupling between raster and vector domains. This hybrid raster-vector modeling framework provides a flexible and scalable foundation for large-scale hydrological-hydrodynamic applications, enabling effective assimilation of satellite river observations to enhance modeling accuracy in data-sparse regions. Ongoing work aims to extend the framework for inferring effective bathymetry-friction (Larnier et al 2025) using parameterized shapes and geomorphological constraints, within a learnable regionalization framework (Huynh et al. 2025) enabling optimal integration of SWOT and LiDAR elevation data, along with in situ measurements when available.
References:
Larnier K., et al. (2025) Estimating channel parameters and discharge at river network scale using hydrological-hydraulic models, SWOT and multi-satellite data. https://hal. inrae.fr/hal-04681079
Huynh, T., et al. (2025). A Distributed Hybrid Physics-AI Framework for Learning Corrections of Internal Hydrological Fluxes and Enhancing High-Resolution Regionalized Flood Modeling. (2025) HESS. https://doi.org/10.5194/egusphere-2024-3665