ECMWF is taking a prominent role in the Horizon Europe research project SEED-FD to improve the global prediction of extreme floods and droughts.
SEED-FD (Strengthening Extreme Events Detection for Floods and Droughts) is intended to develop prototypes that could be used to update the Global Flood Awareness System (GloFAS). This system provides global flood forecasts for the EU-funded Copernicus Emergency Management Service (CEMS), led by the Joint Research Centre (JRC).
ECMWF is the computational centre for the hydrological forecasting activities of CEMS, and it is the science leader of SEED-FD.
The three-year project, which started in February 2024, is also intended to develop prototypes for the Global Drought Observatory (GDO) of CEMS.
Aims of the project
SEED-FD aims to widen the use of GloFAS. “At the moment, GloFAS only forecasts floods in relatively large rivers, it frequently misses localised events associated with intense precipitation. The project aims to equip us to provide forecasts for such events,” says Christel Prudhomme, who leads ECMWF’s Hydrological Monitoring and Forecasting team and the CEMS hydrological forecast computational centre.
“In addition, currently the GDO doesn’t provide any drought forecasts, and the project aims to research the best methods to fill this gap.”
SEED-FD will also try to improve GloFAS flood forecasts more generally. Errors in GloFAS forecasts are, for example, to be reduced through machine learning and satellite information. “Currently we don’t use satellite information at all,” Christel says.
Satellite information from EUMETSAT satellites and the European Space Agency’s Copernicus Sentinel constellation is to be used to support flood forecasting, for example from the Sentinel-2 mission shown here. (Credit: ESA/ATG medialab)
ECMWF’s role
Part of ECMWF’s contribution is to develop methods to detect and forecast the risks of localised flooding arising from intense precipitation. “At the moment, GloFAS is designed for relatively large rivers, and quite often floods in smaller rivers and urban landcovers associated with intense precipitation will not be detected,” says ECMWF scientist Calum Baugh.
The intention is to use products for extreme rainfall, such as ECMWF’s Extreme Forecast Index (EFI) and a new product developed by ECMWF colleagues in the Horizon-Europe-funded CENTAUR project.
These will be enhanced by including information about the vulnerability of different rivers, based on a range of different, mostly external sources. “The idea is to refine our forecasts with dynamic information about the rivers’ susceptibility to localised flooding, for example the degree of urbanisation, river response time to rainfall and whether soil moisture is already saturated,” Calum says.
There will also be work on hydrological forecasts through post-processing to improve the detection of floods.
Post-processing can take the form of error correction or uncertainty quantification. “Essentially, it is trying to make a hydrological forecast more like the real world,” says ECMWF scientist Gwyneth Matthews.
Post-processing methods are already used in the European Flood Awareness System (EFAS), whose modelling chain is run by ECMWF for CEMS, but GloFAS is currently missing this component. “In SEED-FD, we're developing statistical methods that can be applied at a global scale with a focus on capturing high flows and floods correctly,” Gwyneth says.
To achieve this, a particular machine learning method that keeps the memory from past time steps is used.
One aspect will be the increased use of observations. Currently, GloFAS uses observations only to calibrate its hydrological model. “We’ll be using observations of river flow, precipitation and water level,” says Gwyneth.
SEED-FD will look into using a hindcast–forecast long short-term memory (LSTM) error model. This is capable of combining observations from multiple sources, such as satellites (‘EO Data’) and micro-sensors on bridges, with model output from ECMWF’s Integrated Forecasting System (IFS) and GloFAS to estimate the forecast error.
River discharge observations are, however, not universally shared across the world. “We are looking at satellite products to see what information they can provide, and we hope that they can improve the system,” Christel explains.
Pilot sites
The new methods will be tested on pilot river basins. Two of these will be used in the development phase of the project.
The Danube basin in Europe and the Bhima River basin in India will be considered in the development phase of SEED-FD.
This will be followed by the validation phase, in which the improvements will be applied in three basins with different hydrological conditions.
In the validation phase, improvements will be applied in the Paraná basin in South America and the Niger and Juba-Shebelle basins in Africa.
The goal is to see which of the proposed changes have the best potential for operational implementation.
From prototypes to implementation
The initial work will be to develop prototypes of the various systems. These will then be handed over for technical implementation to run the new code on the pilot sites.
This research-to-operations aspect will be handled by ECMWF scientific software engineer Nicola Martin.
“I work with all partners in the project to help them to use ECMWF systems and then to integrate the prototypes into a new test system which mimics that used for GloFAS,” Nicola says. “There’s a data hub on ECMWF’s high-performance computing facility, and all partners are aware of it.”
Nine partners
The project comprises nine partners. In addition to ECMWF, they are:
The French technology company Magellium, which manages and coordinates the project; the Italian Research Institute for Geo-Hydrological Protection (CNR-IRPI); the French company vorteX-io; the International Institute for Applied Systems Analysis (IIASA) based in Austria; Italy’s Politecnico di Milano; Design & Data based in Germany; the IGAD Climate Prediction and Applications Centre (ICPAC) based in Kenya; and the European Commission’s Joint Research Centre (JRC), the entrusted entity for CEMS.
The partners work on a range of issues, such as collecting new data; using the data to improve hydrological simulations and developing drought forecast algorithms; improving model structure and processes; testing the proposed improvements; and the benefits for users.
As an outcome of the project, the JRC will decide which aspects should be implemented by CEMS-Flood.
Further information
Further information about SEED-FD is provided in four videos on the project.