En savoir plus

The hydrological community has been paying particular attention to the advantages of using machine-learned hydrologic models or model components in the last years. Their aims span over a variety of open opportunities for research and operations in the water sector, such as:

• to unlock data sets and data services from earth observations initiatives that are underexploited by public and private stakeholders in water resources planning;
• to improve the accuracy and spatial resolution of hydrological models and enhance the evaluation of impacts on water resources availability for ecosystems, agriculture and industries;
• to bring together observed and modelled data available from different hydrological components and for different water management objectives within a river basin, such as water quality management, reservoir operations, surface and groundwater supply systems;
• to enhance our ability to describe the feedbacks between climate change and people, drought and flood risks, and to quantify the human influence on droughts and floods.

From input data processing to model output prediction, novel techniques have been explored to support a large number of hydrological applications:

• forecasting of extreme events (floods and droughts),
• predicting water use patterns and projecting future water demand,
• detecting patterns in reservoir management and predicting releases,
• regionalizing hydrological model’s parameters over large river basins,
• regionalizing groundwater information at the catchment scale.

This workshop brought together hydrologists and data-driven experts to explore pathways to bring new data sets and data-driven modelling approaches from hydrologic research to the more practical and operational planning level in river basins.

The workshop is a joint-initiative of the Horizon Europe STARS4Water Project and the French national ANR project DRHYM

Chairs:

• Maria-Helena Ramos (INRAE) and Nathan Rickards (UKCEH): presentations
• Daniel Klotz (JKU) & Helen Baron (UKCEH): Hands-on session

Speakers:

• Martin Gauch (Google Research, CH): Deep Learning for large-scale hydrologic modeling
• Peter Nelemans (Deltares, NL): Towards a fully distributed hydrological deep learning model with Graph Neural Networks
• Leandro Avila (FZ Jülich, DE): Downscaling GRACE total water storage based on LSTM models and TSMP (Terrestrial Systems Modelling Platform) simulations
• Jing Deng (Deltares, NL): Operational low-flow forecasting using LSTMs
• Victor Gomez-Escalonilla and Pedro Martinez-Santos (UCM, SP): Predictive mapping by means of ML algorithms
• Basil Kraft (ETH Zurich, CH): Combining machine learning and process knowledge for global hydrological modeling
• Antoine Degenne (INRAE, FR): Hydrological modeling using hybrid process-based and machine learning modeling
• Annine Kenne (JKU, AT): Progress in reservoir modeling with ML
• Helen Baron (UKCEH, UK): Forecasting reservoir storage with the Random Forest

Organizing committee:

• Maria-Helena Ramos (INRAE UR HYCAR, France) (Chair)
• Helen Baron (UK Centre for Ecology & Hydrology, United Kingdom)
• Antoine Degenne (INRAE UR HYCAR, France)
• Harm Duel (Deltares, The Netherlands)
• Virginie Keller (UK Centre for Ecology & Hydrology, United Kingdom)
• Daniel Klotz (Johannes Kepler University Linz, Austria)
• Stefan Kollet (Forschungszentrum Jülich, Germany)