SWOT-hydro2-learning: Learning regionalization of hydrological-hydraulic models and discharge laws over river networks with SWOT and multi-source data assimilation

Principle Investigator(s): Pierre-André Garambois (INRAE, Aix-Marseille University, RECOVER), Jérôme Monnier (INSA & Institut de Mathématiques de Toulouse (IMT))

Co-Investigator(s): Sylvain Biancamaria, Robin Bouclier, Stéphane Calmant, Kévin Larnier, Thomas Ledauphin, Jérôme Maxant, Benjamin Renard, Olivier Roustant, Hélène Roux, Adrien Paris, Hervé Yésou

Collaborator(s): Cedric David, Renato Frasson, Mike Durand, Harry Lee, Daniel Medeiros-Moreira, Rodrigo Paiva, Fabrice Pappa

---

Using new differentiable hydrology-hydraulic coupled models and innovative combinations of data assimilation and machine learning techniques, we will develop disruptive approaches for hydrological-hydraulic modeling and discharge inversion algorithms that optimally exploit the unprecedented SWOT hydraulic visibility along with complementary multi-source data (altimetry, imagery, in situ data, etc), from the river section to the river network scale. Model Parameters Learning and regionalization capabilities for ungauged rivers, as well as Uncertainty Quantification (UQ), will be investigated on contrasted basins demonstrators at multiple scales and resolutions, while aiming at worldwide and operational applicability of the proposed methods. Our interdisciplinary and highly complementary team includes experts in remote sensing, hydrology, hydraulics, applied mathematics, Data Assimilation (DA) and numerical modeling who already collaborate closely and involved in SWOT Science Team and mission preparation for more than 10 years. The SWOT-Hydro2-Learning project is completely aligned with current and planned activities of the SWOT ST and related working groups with active contribution/lead of SWOT community publications. The project will be based on and contribute to the open source numerical models and hybrid data assimilation-learning algorithms we develop for research and operational applications (French flood forecasting platform, CNES operational hydrology). Expected project benefits, in addition to high quality datasets, computational codes and basin-scale demonstrators potentially transferable to operational, will pertain to methodological advances in forward-inverse and cartographic hydrological-hydraulic modeling with built-in learning capabilities from SWOT and multi-source data. This project studies the internationally-recognized paradigm shift in integrated basin modeling and hydrological learning. We propose here high-level research breakthroughs based on the development of hybrid approaches for learning hydrological and hydraulic models with multi-source data assimilation. Such technologies will enable unprecedented use of Earth observation to inform hydrology (first principles yet to be discovered).