Data-driven hybrid hydrological models: revisiting the role of river basins as information providers for a better assessment of risks and water resources – DRHYM

Hydrological models which represent the catchment behaviour in transforming climate inputs into streamflow and other hydrological variables are still of limited efficiency, in spite of long terms efforts to improve their skills. As an example, the development of GR hydrological models started in the 1980s in France, and these models are now used in many countries for various applications. A key feature of their empirical development is the use of large set of catchments. Today the blooming development of artificial intelligence (AI) techniques and the increasing available of data sets are the global scale offer an opportunity to revisit the development of these hydrological and improve their predictive capacity. The main paradigm of DRHYM will be to consider the catchment as a source of information to identify improved hybrid model structures with wider applicability in time and space. This will be done by putting the search for improved model structures in hydrology in the framework of AI techniques and large worldwide data sets. Three main challenges will be addressed in the project, namely (1) setting up an innovative methodological framework to develop improved hybrid hydrological models, (2) designing GR models with improved skills for a wide spectrum of time and space scales, and (3) address the limits of machine-learning-based and GR-based approaches in a context of limited data. The main deliverables of the project will a methodological framework for hybrid-model development in hydrology, a new generation of GR models and open-source codes to share with the end-users community.