Multi-source in situ and satellite variational data assimilation into a fully distributed hydrological model for floods and droughts modeling over poorly gauged and ungauged areas

Estimating water flows and stocks in surface hydrology is crucial for addressing important socio-economic issues, such as managing water resources and predicting extreme events like floods and droughts. These challenges become more significant with the ongoing global climate change, which may intensify the hydrological cycle. Advanced modeling tools are necessary for making precise and reliable local forecasts. However, hydrological models, regardless of their complexity and status, encounter difficulties in accurately and reliably predicting quantities of interest such as river flows or soil moisture states, and in accounting for meteorological-climatic effects on hydrology. Given the complexity of the physical processes involved and their heterogeneous and limited observability, hydrological modeling is a challenging task, and internal flows often have significant uncertainties. These uncertainties could be reduced by integrating new observations from remote sensing applied to continental surfaces, which is rapidly evolving. A variety of satellites and sensors now allow the observation of watershed surface characteristics and hydrological responses with increasing spatial-temporal resolutions. In particular, products of soil moisture, evaporation, and land use are now available at relatively high spatial-temporal resolution. This work focuses on improving the integration of satellite and in-situ land surface data into spatially distributed hydrological models. The Hybrid Data Assimilation and Parameter Regionalization (HDA-PR) approach incorporating learnable regionalization mappings, based on neural networks into the differentiable hydrological model SMASH, is modified to account for satellite moisture maps in addition to discharge at gauging stations and basins physical descriptors maps. Regional optimizations are performed on flash-flood-prone areas located in the South of France and their accuracy and robustness is evaluated in terms of simulated discharge and moisture against observations.