Abstract's details
Estimation of Discharge and River Channel Parameters through Hydrology–Hydraulic (H&H) Coupling with SWOT Altimetry: Toward Learnable Parameterizations
Event: 2025 SWOT Science Team Meeting
Session: Hydrology: Global Hydrology Modeling Working Group
Presentation type: Oral
The unprecedented hydraulic visibility of river surface deformation provided by the SWOT satellite offers a wealth of information for enhancing hydrological-hydraulic (H&H) models and improving discharge estimations worldwide. However, estimating uncertain or unknown parameters in hydraulic models—such as inflow discharges, riverbed bathymetry, and friction parameters embedded in the momentum equation and subject to equifinality—constitutes a high-dimensional inverse problem that remains ill-posed when based solely on altimetric observations. To address this challenge, we couple a river network hydraulic model with regional hydrological modeling, leveraging physically consistent inflow estimates at upstream boundaries and lateral contributions to constrain the solution space and improve parameter identifiability across the river network (after Pujol et al. 2020). A robust variational data assimilation (VDA) of water surface elevation (WSE) observations into the 1D Saint-Venant river network model DassFlow 1D enables the joint inference of inflow hydrographs, effective bathymetry, and spatially distributed friction at the network scale starting from prior parameters infered with HiVDI (Larnier et al. 2020, Larnier and Monnier 2023).
This approach is demonstrated on the large, complex, and poorly gauged Maroni basin in French Guiana (Larnier et al., 2025). The pre-processing chain includes: (i) constructing effective hydraulic geometry using drifting ICESat altimetry and Sentinel-1-derived river widths; and (ii) filtering noisy SWOT Level 2 WSE data prior to assimilation. Results show systematic
improvements in both the fit to assimilated WSE data—achieving an 85% cost reduction—and discharge validation at five gauges across the network. When assimilating SWOT data alone, 70% of WSE residuals fall within the [−0.25; 0.25 m] range, and normalized root-mean-square error (NRMSE) of discharge estimates ranges from 0.05 to 0.18, representing an improvement of 18% to 71% over prior estimates. The dense spatial coverage of SWOT WSE observations enables the inference of detailed spatial variability in riverbed elevation, friction coefficients, and inflow time series.
Next, the hybrid physics–AI hydrological model SMASH (Huynh et al. 2024, 2025) is coupled with DassFlow 1D (Brisset et al. 2018). This numerical integration of two differentiable models enables VDA to provide information feedback from hydraulic observables to the hydrological model (after Pujol et al., 2022). This feedback capability is demonstrated on the Maroni River by calibrating the hydrological model directly from SWOT-derived WSE observations. The approach is currently being adapted to the Garonne basin, where the methodology of Pujol et al. (2022)—embedding a hydrological model within a seamless 2D–1D-like hydraulic framework—is also under investigation.
This work is part of the SWOT-Hydro2Learning TOSCA project, which develops disruptive H&H modeling and discharge inversion methods by coupling differentiable models with data assimilation and machine learning, leveraging SWOT, multi-source Earth observation, and in situ data.
References
• Pujol et (in prep) Enhancing River Discharge Estimation from SWOT through Basin-Scale Hydrological-Hydraulic Variational Data Assimilation.
• Pujol, L. et al. (2020) Estimation of Multiple Inflows and Effective Channel by Assimilation of Multi-satellite Hydraulic Signatures: Case of the Ungauged Anabranching Negro River, JoH, https://doi.org/10.1016/j.jhydrol. 2020.125331.
• Larnier, K. et al. (2020) River discharge and bathymetry estimation from SWOT altimetry measurements, IPSE, https://doi.org/10.1080/17415977.2020.1803858.
• Larnier K., Monnier J. (2023) "Hybrid neural network - variational data assimilation to infer river discharges from SWOT-like data". CG. https://link.springer.com/article/10.1007/s10596-023-10225-2
• Brisset, P. et al. (2018) On the Assimilation of Altimetric Data in 1D Saint-Venant River Flow Models, AWR, https://doi.org/ 10.1016/j.advwatres.2018.06.004.
• Pujol, et al. (2022) Multi-dimensional hydrological-hydraulic model with variational data assimilation for river networks and floodplains, Geoscientific Model Develop- ment, https://doi.org/10.5194/egusphere-2022-
• 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. (2024) Learning Regionalization within a Differentiable High-Resolution Hydrological Model using Accurate Spatial Cost Gradients. WRR. https://doi.org/10.1029/2024WR037544
• 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. https://doi.org/10.5194/egusphere-2024-3665
Back to the list of abstractThis approach is demonstrated on the large, complex, and poorly gauged Maroni basin in French Guiana (Larnier et al., 2025). The pre-processing chain includes: (i) constructing effective hydraulic geometry using drifting ICESat altimetry and Sentinel-1-derived river widths; and (ii) filtering noisy SWOT Level 2 WSE data prior to assimilation. Results show systematic
improvements in both the fit to assimilated WSE data—achieving an 85% cost reduction—and discharge validation at five gauges across the network. When assimilating SWOT data alone, 70% of WSE residuals fall within the [−0.25; 0.25 m] range, and normalized root-mean-square error (NRMSE) of discharge estimates ranges from 0.05 to 0.18, representing an improvement of 18% to 71% over prior estimates. The dense spatial coverage of SWOT WSE observations enables the inference of detailed spatial variability in riverbed elevation, friction coefficients, and inflow time series.
Next, the hybrid physics–AI hydrological model SMASH (Huynh et al. 2024, 2025) is coupled with DassFlow 1D (Brisset et al. 2018). This numerical integration of two differentiable models enables VDA to provide information feedback from hydraulic observables to the hydrological model (after Pujol et al., 2022). This feedback capability is demonstrated on the Maroni River by calibrating the hydrological model directly from SWOT-derived WSE observations. The approach is currently being adapted to the Garonne basin, where the methodology of Pujol et al. (2022)—embedding a hydrological model within a seamless 2D–1D-like hydraulic framework—is also under investigation.
This work is part of the SWOT-Hydro2Learning TOSCA project, which develops disruptive H&H modeling and discharge inversion methods by coupling differentiable models with data assimilation and machine learning, leveraging SWOT, multi-source Earth observation, and in situ data.
References
• Pujol et (in prep) Enhancing River Discharge Estimation from SWOT through Basin-Scale Hydrological-Hydraulic Variational Data Assimilation.
• Pujol, L. et al. (2020) Estimation of Multiple Inflows and Effective Channel by Assimilation of Multi-satellite Hydraulic Signatures: Case of the Ungauged Anabranching Negro River, JoH, https://doi.org/10.1016/j.jhydrol. 2020.125331.
• Larnier, K. et al. (2020) River discharge and bathymetry estimation from SWOT altimetry measurements, IPSE, https://doi.org/10.1080/17415977.2020.1803858.
• Larnier K., Monnier J. (2023) "Hybrid neural network - variational data assimilation to infer river discharges from SWOT-like data". CG. https://link.springer.com/article/10.1007/s10596-023-10225-2
• Brisset, P. et al. (2018) On the Assimilation of Altimetric Data in 1D Saint-Venant River Flow Models, AWR, https://doi.org/ 10.1016/j.advwatres.2018.06.004.
• Pujol, et al. (2022) Multi-dimensional hydrological-hydraulic model with variational data assimilation for river networks and floodplains, Geoscientific Model Develop- ment, https://doi.org/10.5194/egusphere-2022-
• 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. (2024) Learning Regionalization within a Differentiable High-Resolution Hydrological Model using Accurate Spatial Cost Gradients. WRR. https://doi.org/10.1029/2024WR037544
• 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. https://doi.org/10.5194/egusphere-2024-3665