Anticipating risks is one of three priorities expressed in the RPO (Référentiel Pluriannuel d'Objectifs) TR (Thème de Recherche) Arceau 2014-2018. The thesis proposed aims at improving an anticipatory approach to flood in non-gauged environment, applied to the whole French territory (AIGA method implanted in the service "Sudden floods" of Schapi: Vigicrue Flash - early 2017) .
This approach, based on a distributed hydrological modeling in cells of 1 km², using information from weather radar, has room for improvement on two particular points that we want to address in this thesis:
- Relevance in "downscaling" or "space weathering", that is to say when the method is applied on non-gauged sub-basins. This requires a timing and regionalization of parameters that must be adapted.
- The ability to take into account the real-time modeling errors when making predictions: for a forecast in ungauged site, this means assimilating the information available on neighboring sites.
These two points ultimately correspond to the same data assimilation problems with a model, for calibration or recalibration of parameters. These points have been addressed so far by more or less conventional approaches in hydrology. The new skills in 4D-Var variational data assimilation carried by Irstea lead us to want to apply these methods in our approaches. The objective of this thesis is to implement efficient data assimilation methods for the treatment approaches large (distributed model of hydrological forecasting across the French territory).