Hélène Roux

Hydrological models are widely used for large-scale flood modeling but typically rely on simplified routing schemes and a grid-based discretization that poorly capture river flow dynamics and geometry. Conversely, hydraulic models enable more accurate representation of flow dynamics through physically based routing schemes and vector-based geometry, but their application to large domains is constrained by significant parametrization challenges. Bridging these limitations requires integrated hydrological-hydraulic (H&H) modeling approaches capable of reconciling the spatial scale and structural mismatch between the two models while ensuring seamless coupling and computational efficiency for large-scale applications. We present an integrated raster-vector H&H modeling framework that leverages sub-grid representation of both river networks and drainage areas derived from high-resolution topography. The framework is implemented within the open-source smash modeling platform and integrates an end-to-end workflow from DEM preprocessing to internally coupled H&H simulations. The framework was evaluated over the Garonne river basin (France, 50 100 km²) through one-directional coupling of a grid-based conceptual hydrological model with a vector-based 1D hydrodynamic model solving the zero-convective inertia approximation of the shallow water equations. Geometric preprocessing analysis performed across spatial resolutions ranging from 0.5 km to 10 km demonstrates that sub-grid information enable to maintain high spatial accuracy of DEM-derived sub-grid networks across scales in relation to the mapped hydrography, and reduces significantly catchment area errors compared to the default grid-based area delineation method. H&H simulations conducted at 1 km, 5 km, and 10 km without recalibration show robust preservation of flow timing across scales and demonstrate more stable streamflow bias across resolutions when using sub-grid drainage areas, while grid-based areas exhibit scale-dependent volume biases reflecting drainage area misrepresentation. The proposed H&H framework demonstrates scalable and efficient simulations at large domains within a unified modeling environment, offering promising perspectives for assimilation of multi-source water surface observations to infer key model parameters, addressing critical parametrization challenges in data-sparse regions.

Deep-learning models have demonstrated strong performances in reproducing stream temperature dynamics, which is promising for the reconstruction of missing stream temperature records at ungauged locations. However, model accuracy over the range of high, summer stream temperature values has been usually overlooked, raising the question of the suitability of using deep-learning methods during this crucial season. In this study, we investigated strategies to improve the performances of a stream-temperature model based on LSTM (Long Short-Term Memory) cells over the extreme, highest 10 % observed values at 21 stations located in the Garonne river catchment. We quantified the gain in model performance thanks to regional multi-catchment training with static attributes, exploiting hydrologically relevant variables, and further penalizing the errors at extreme temperature values using custom loss functions. Our key results are: (1) Regional multi-catchment training is the best strategy to improve the performances of LSTM models not only over the extreme, top 10 % values but also over the whole range of observations. (2) The gain in performances was mainly brought by the use of static, catchment and reach attributes. (3) Customizing the loss function to emphasize the model errors on extreme temperature values did not lead to significant gains in test performances. This study further confirms the suitability of regionally trained LSTM models that exploit static attributes for the reproduction of extreme stream temperature values, offering significant advantages for water management at data-sparse regions.

Hydrological models are widely used for large-scale flood modeling but typically rely on simplified routing schemes and a grid-based discretization that poorly capture river flow dynamics and geometry. Conversely, hydraulic models enable more accurate representation of flow dynamics through physically based routing schemes and vector-based geometry, but their application to large domains is constrained by significant parametrization challenges. Bridging these limitations requires integrated hydrological-hydraulic (H&H) modeling approaches capable of reconciling the spatial scale and structural mismatch between the two models while ensuring seamless coupling and computational efficiency for large-scale applications. We present an integrated raster-vector H&H modeling framework that leverages sub-grid representation of both river networks and drainage areas derived from high-resolution topography. The framework is implemented within the open-source smash modeling platform and integrates an end-to-end workflow from DEM preprocessing to internally coupled H&H simulations. The framework was evaluated over the Garonne river basin (France, 50 100 km²) through one-directional coupling of a grid-based conceptual hydrological model with a vector-based 1D hydrodynamic model solving the zero-convective inertia approximation of the shallow water equations. Geometric preprocessing analysis performed across spatial resolutions ranging from 0.5 km to 10 km demonstrates that sub-grid information enable to maintain high spatial accuracy of DEM-derived sub-grid networks across scales in relation to the mapped hydrography, and reduces significantly catchment area errors compared to the default grid-based area delineation method. H&H simulations conducted at 1 km, 5 km, and 10 km without recalibration show robust preservation of flow timing across scales and demonstrate more stable streamflow bias across resolutions when using sub-grid drainage areas, while grid-based areas exhibit scale-dependent volume biases reflecting drainage area misrepresentation. The proposed H&H framework demonstrates scalable and efficient simulations at large domains within a unified modeling environment, offering promising perspectives for assimilation of multi-source water surface observations to infer key model parameters, addressing critical parametrization challenges in data-sparse regions.

Hydrological models are widely used for large-scale flood modeling but typically rely on simplified routing schemes and a grid-based discretization that poorly capture river flow dynamics and geometry. Conversely, hydraulic models enable more accurate representation of flow dynamics through physically based routing schemes and vector-based geometry, but their application to large domains is constrained by significant parametrization challenges. Bridging these limitations requires integrated hydrological-hydraulic (H&H) modeling approaches capable of reconciling the spatial scale and structural mismatch between the two models while ensuring seamless coupling and computational efficiency for large-scale applications. We present an integrated raster-vector H&H modeling framework that leverages sub-grid representation of both river networks and drainage areas derived from high-resolution topography. The framework is implemented within the open-source smash modeling platform and integrates an end-to-end workflow from DEM preprocessing to internally coupled H&H simulations. The framework was evaluated over the Garonne river basin (France, 50 100 km²) through one-directional coupling of a grid-based conceptual hydrological model with a vector-based 1D hydrodynamic model solving the zero-convective inertia approximation of the shallow water equations. Geometric preprocessing analysis performed across spatial resolutions ranging from 0.5 km to 10 km demonstrates that sub-grid information enable to maintain high spatial accuracy of DEM-derived sub-grid networks across scales in relation to the mapped hydrography, and reduces significantly catchment area errors compared to the default grid-based area delineation method. H&H simulations conducted at 1 km, 5 km, and 10 km without recalibration show robust preservation of flow timing across scales and demonstrate more stable streamflow bias across resolutions when using sub-grid drainage areas, while grid-based areas exhibit scale-dependent volume biases reflecting drainage area misrepresentation. The proposed H&H framework demonstrates scalable and efficient simulations at large domains within a unified modeling environment, offering promising perspectives for assimilation of multi-source water surface observations to infer key model parameters, addressing critical parametrization challenges in data-sparse regions.

Topographic data are becoming increasingly accurate and available thanks to advances in remote sensing, including LIDAR and satellite imagery; yet, significant uncertainty remains in the bathymetric data. These data are highly valuable for constructing hydraulic models and characterizing continental hydrosystems. To transform raw data into a suitable form we developed a QGIS plugin called CEPHEE, which enables a wide range of users to obtain a hydraulic overview of a catchment. From high resolution digital elevation model, the plugin automatically provides a description of the hydrological network and river. Bathymetry can be corrected to produce an equivalent geometry consistent with prescribed flow rates. It identifies bank positions and subsequent river widths for various flow rates, integrates a steady-state calculation, a steady backwater curve calculation, and provides a rating curve for each cross-section. The chained procedure has been validated against empirical hydromorphological laws as well as full 2D shallow water models simulations.

Hydrological models are widely used for large-scale flood modeling but typically rely on simplified routing schemes and a grid-based discretization that poorly capture river flow dynamics and geometry. Conversely, hydraulic models enable more accurate representation of flow dynamics through physically based routing schemes and vector-based geometry, but their application to large domains is constrained by significant parametrization challenges. Bridging these limitations requires integrated hydrological-hydraulic (H&H) modeling approaches capable of reconciling the spatial scale and structural mismatch between the two models while ensuring seamless coupling and computational efficiency for large-scale applications. We present an integrated raster-vector H&H modeling framework that leverages sub-grid representation of both river networks and drainage areas derived from high-resolution topography. The framework is implemented within the open-source smash modeling platform and integrates an end-to-end workflow from DEM preprocessing to internally coupled H&H simulations. The framework was evaluated over the Garonne river basin (France, 50 100 km²) through one-directional coupling of a grid-based conceptual hydrological model with a vector-based 1D hydrodynamic model solving the zero-convective inertia approximation of the shallow water equations. Geometric preprocessing analysis performed across spatial resolutions ranging from 0.5 km to 10 km demonstrates that sub-grid information enable to maintain high spatial accuracy of DEM-derived sub-grid networks across scales in relation to the mapped hydrography, and reduces significantly catchment area errors compared to the default grid-based area delineation method. H&H simulations conducted at 1 km, 5 km, and 10 km without recalibration show robust preservation of flow timing across scales and demonstrate more stable streamflow bias across resolutions when using sub-grid drainage areas, while grid-based areas exhibit scale-dependent volume biases reflecting drainage area misrepresentation. The proposed H&H framework demonstrates scalable and efficient simulations at large domains within a unified modeling environment, offering promising perspectives for assimilation of multi-source water surface observations to infer key model parameters, addressing critical parametrization challenges in data-sparse regions.

Representing hydrological and hydraulic processes consistently across spatial scales remains a major challenge for large-scale flood modelling. Besides using simplified routing schemes that struggle to accurately represent in-channel river flow dynamics, most large-scale hydrological models adopt a Cartesian-grid discretization scheme due to their compatibility with widely available gridded datasets. However, this grid-based structure poorly captures river geometry and oversimplifies natural drainage boundaries, leading to scale-dependent biases in runoff production and streamflow simulations, commonly referred to as the “catchment size problem”. In contrast, hydrodynamic models implement more physically-based routing schemes with fine-scale geometric representation of river channels, but face important parametrization challenges at larger scales. To address these challenges, we introduce an integrated hydrological–hydrodynamic (H&H) modelling framework that enables a seamless coupling between coarse-resolution gridded hydrological modelling and fine-scale vector-based river routing, leveraging a sub-grid representation of the river network derived from high-resolution topography. Notably, sub-grid information is propagated into the hydrological model by replacing regular grid-cell areas with realistic drainage areas derived from sub-grid topography, thereby addressing the aforementioned “catchment size problem”. The integrated H&H framework is implemented within the SMASH modelling platform (Spatially distributed Modelling and ASsimilation for Hydrology, https://smash.recover.inrae.fr/). For this application, we coupled the grid-based conceptual hydrological model GR4 to a vector-based hydrodynamic model solving a simplification of the 1D shallow water equations without convective acceleration terms. We evaluated this framework over the Garonne River catchment (France, ~50,000 km²) using the MERIT digital elevation model (resampled at 100 m) and the reference national river network BD TOPAGE®. We conducted H&H simulations across three spatial resolutions: 1 km, 5 km, and 10 km, where we considered the 1 km configuration as a baseline and kept the same hydrological and hydrodynamic parametrization across resolutions (no recalibration): semi-distributed hydrological parameters, uniform channel friction, and simplified rectangular channel geometry where widths and depths are estimated from geomorphological relationships, and bathymetry is subsequently derived from geomorphological depths and elevation. Results show that the sub-grid river representation maintains a consistently high spatial accuracy across spatial scales, with mean separation distance from the reference hydrography of around 25 m and minimal omission of the mapped network (

Deep-learning models have demonstrated strong performances in reproducing stream temperature dynamics, which is promising for the reconstruction of missing stream temperature records at ungauged locations. However, model accuracy for the range of high, summer stream temperature has been usually overlooked, raising the question of the suitability of using deep-learning methods during this crucial season. In this study, we investigated strategies to improve the performances of a stream-temperature model based on LSTM (Long Short-Term Memory) cells over the highest 10 % observed values at 21 stations located in the Garonne river catchment. We quantified the gain in model performance thanks to regional multi-catchment training with static attributes, exploiting hydrologically relevant variables, and further penalizing the errors at extreme temperature values using custom loss functions. Our key results are: (1) Regional multi-catchment training is the best strategy to improve the performances of LSTM models not only over the top 10 % values but also over the whole range of observations. (2) The gain in performances was mainly brought by the use of static, catchment and reach attributes. (3) Customizing the loss function to emphasize the model errors on extreme temperature values did not lead to significant gains in test performances. This study further confirms the suitability of well-trained LSTM models for extreme stream temperature values, offering significant advantages for water management at data-sparse regions during summer periods.

Deep-learning models have demonstrated strong performances in reproducing stream temperature dynamics, which is promising for the reconstruction of missing stream temperature records at ungauged locations. However, model accuracy for the range of high, summer stream temperature has been usually overlooked, raising the question of the suitability of using deep-learning methods during this crucial season. In this study, we investigated strategies to improve the performances of a stream-temperature model based on LSTM (Long Short-Term Memory) cells over the highest 10 % observed values at 21 stations located in the Garonne river catchment. We quantified the gain in model performance thanks to regional multi-catchment training with static attributes, exploiting hydrologically relevant variables, and further penalizing the errors at extreme temperature values using custom loss functions. Our key results are: (1) Regional multi-catchment training is the best strategy to improve the performances of LSTM models not only over the top 10 % values but also over the whole range of observations. (2) The gain in performances was mainly brought by the use of static, catchment and reach attributes. (3) Customizing the loss function to emphasize the model errors on extreme temperature values did not lead to significant gains in test performances. This study further confirms the suitability of well-trained LSTM models for extreme stream temperature values, offering significant advantages for water management at data-sparse regions during summer periods.

Deep-learning models have demonstrated strong performances in reproducing stream temperature dynamics, which is promising for the reconstruction of missing stream temperature records at ungauged locations. However, model accuracy for the range of high, summer stream temperature has been usually overlooked, raising the question of the suitability of using deep-learning methods during this crucial season. In this study, we investigated strategies to improve the performances of a stream-temperature model based on LSTM (Long Short-Term Memory) cells over the highest 10 % observed values at 21 stations located in the Garonne river catchment. We quantified the gain in model performance thanks to regional multi-catchment training with static attributes, exploiting hydrologically relevant variables, and further penalizing the errors at extreme temperature values using custom loss functions. Our key results are: (1) Regional multi-catchment training is the best strategy to improve the performances of LSTM models not only over the top 10 % values but also over the whole range of observations. (2) The gain in performances was mainly brought by the use of static, catchment and reach attributes. (3) Customizing the loss function to emphasize the model errors on extreme temperature values did not lead to significant gains in test performances. This study further confirms the suitability of well-trained LSTM models for extreme stream temperature values, offering significant advantages for water management at data-sparse regions during summer periods.