Periodic Reporting for period 1 - SEED-FD (Strenghtening Extreme Events Detection for Flood and Drought)
Reporting period: 2024-01-01 to 2025-06-30
Climate change is causing significant impacts, including floods and droughts, which necessitate early warning systems to save lives. However, one third of the global population in least developed countries lacks coverage from these systems.
The EU-funded SEED-FD project aims to address this gap by improving flood and drought detection and predictions globally, particularly in lower and middle-income countries.
Using advanced science, Earth Observation (EO), and non-EO technologies, the project aims to enhance the accuracy and reliability of hydrological simulations and extreme events forecasts.
SEED-FD will collaborate with the Copernicus Emergency Management Service (CEMS) to provide skilful forecasts, thereby improving all aspects of the CEMS Hydrological Forecasting Modelling Chain (HFMC). By integrating state-of-the-art science with new data, the project seeks to transform new observational information into high-quality hydro meteorological extreme event forecast products.
SEED-FD will invest in better representing hydrological processes and parameterization techniques of the CEMS core hydrological engine (LISFLOOD) and combine the model enhancements with innovative techniques to integrate EO and non-EO data with the near real-time hydrological processing chain for reducing hydrological forecasting errors, hence breaking the current limitations of hydrological simulation accuracy where there are no or few in situ data available and make skillful forecasts available anywhere in the world. The enhancements will also drastically improve the monitoring capacity of hydrological status and extreme events worldwide. Using Artificial Intelligence-based event-detection algorithms, SEED-FD will create a new global flash flood forecast product for better anticipation of high-impact events and will create global drought forecast indicators for anticipating food, water, and energy shortages. The evolutions developed in SEED-FD will be domain-agnostic and could be applied to both European or Global domains. However, the benefits are expected to be largest in the global south for lower and middle-income countries, typically the most impacted by extreme hydrological events but also where the current knowledge gap in hydrological simulation and forecasting is highest.
The EU-funded SEED-FD project aims to address this gap by improving flood and drought detection and predictions globally, particularly in lower and middle-income countries.
Using advanced science, Earth Observation (EO), and non-EO technologies, the project aims to enhance the accuracy and reliability of hydrological simulations and extreme events forecasts.
SEED-FD will collaborate with the Copernicus Emergency Management Service (CEMS) to provide skilful forecasts, thereby improving all aspects of the CEMS Hydrological Forecasting Modelling Chain (HFMC). By integrating state-of-the-art science with new data, the project seeks to transform new observational information into high-quality hydro meteorological extreme event forecast products.
SEED-FD will invest in better representing hydrological processes and parameterization techniques of the CEMS core hydrological engine (LISFLOOD) and combine the model enhancements with innovative techniques to integrate EO and non-EO data with the near real-time hydrological processing chain for reducing hydrological forecasting errors, hence breaking the current limitations of hydrological simulation accuracy where there are no or few in situ data available and make skillful forecasts available anywhere in the world. The enhancements will also drastically improve the monitoring capacity of hydrological status and extreme events worldwide. Using Artificial Intelligence-based event-detection algorithms, SEED-FD will create a new global flash flood forecast product for better anticipation of high-impact events and will create global drought forecast indicators for anticipating food, water, and energy shortages. The evolutions developed in SEED-FD will be domain-agnostic and could be applied to both European or Global domains. However, the benefits are expected to be largest in the global south for lower and middle-income countries, typically the most impacted by extreme hydrological events but also where the current knowledge gap in hydrological simulation and forecasting is highest.
In WP1 we collected satellite and in-situ observations that will be used during the project to improve flood and drought forecasting in the study areas. Specifically, the integrated satellite and reanalysis precipitation product for the Danube basin for the period 2007-2022 was produced and distributed to the partners. Similarly, satellite-derived river discharge proxies for 14 virtual stations in the Danube, Niger and Parana river basins were distributed. Satellite soil moisture products available under Copernicus Land Monitoring Service were proposed for their use in the project. The VorteX-io water level data from the micro-stations installed in Europe were made available to the project and the installation of the new stations is underway.
In WP2 we corrected the station location for GloFAS calibration. We use an Intersection over Union ratio approach to selected station locations on a coarser grid-scale, reducing the errors in assigning stations to the correct upstream basin. The Glofas stations are already in a good shape as many stations are shifted manually to the right location. We improved the station location in the Danube basin for six stations. The tool is now evaluated at JRC to be used for the next calibration round. To improve performance for ungauged basins new calibrations and regionalization methods were tested. We tested Budyko and GRACE calibration. For data assimilation, it was necessary to wait for the prototypes of the HFMC running on the different basins to be generated.. We are now carrying out the preliminary tasks for integrating data assimilation into the system (daily storage of GloFAS simulations and ensemble simulations for the Ensemble Kalman filter). The integration of the data assimilation module will be done in the ecFlow workflow, allowing for seamless management and scheduling of tasks within the forecasting system.
The task in WP3 is to improve the hydrological model Lisflood. We work on a better groundwater representation in Lisflood due to integration of the groundwater model MODFLOW. This will be done by using the python interface flopy to connect Lisflood to Modflow. For a test region (Morava/Danube) necessary data and maps have been produced.
We work at a better representation of wetlands (e.g. Inner delta Niger, Sudd of Nile, Murray-Darling). In a first step we used the global dataset of Wetland Area and Dynamics for Methane Modeling (WAD2M) to estimate the area and the magnitude of the seasonal variability of wetlands (at the example of the Niger inner delta). The next step is to simulate the increase and decrease of the wetland with Lisflood.
In WP4 the input data required for the tasks on flash flood and drought development have been obtained, input data for the task on post-processing are ongoing. A benchmark post-processing method has been selected that in general improves the GloFAS forecasts by 10%. Developed post-processing methods will be compared against this benchmark. The methodology to create a flash flood product has been defined, work will start shortly to optimise the definition of flash flood warning areas. The drought indicators have been defined for each of the three time scales, the next steps are to begin integrating forecast data.
WP5 activities have primarily focused on technical preparation and support. Best practices for software development have been established, and a central Data Hub has been set up to support the project. Guidance on utilising ECMWF systems and software has been provided. Initial modifications to the Hydrological Forecast Modelling Chain have been implemented in preparation for prototype use, and both input and control datasets have been generated. Discussions with WP1-4 have been conducted to gain a deeper understanding of their technical requirements.
In WP2 we corrected the station location for GloFAS calibration. We use an Intersection over Union ratio approach to selected station locations on a coarser grid-scale, reducing the errors in assigning stations to the correct upstream basin. The Glofas stations are already in a good shape as many stations are shifted manually to the right location. We improved the station location in the Danube basin for six stations. The tool is now evaluated at JRC to be used for the next calibration round. To improve performance for ungauged basins new calibrations and regionalization methods were tested. We tested Budyko and GRACE calibration. For data assimilation, it was necessary to wait for the prototypes of the HFMC running on the different basins to be generated.. We are now carrying out the preliminary tasks for integrating data assimilation into the system (daily storage of GloFAS simulations and ensemble simulations for the Ensemble Kalman filter). The integration of the data assimilation module will be done in the ecFlow workflow, allowing for seamless management and scheduling of tasks within the forecasting system.
The task in WP3 is to improve the hydrological model Lisflood. We work on a better groundwater representation in Lisflood due to integration of the groundwater model MODFLOW. This will be done by using the python interface flopy to connect Lisflood to Modflow. For a test region (Morava/Danube) necessary data and maps have been produced.
We work at a better representation of wetlands (e.g. Inner delta Niger, Sudd of Nile, Murray-Darling). In a first step we used the global dataset of Wetland Area and Dynamics for Methane Modeling (WAD2M) to estimate the area and the magnitude of the seasonal variability of wetlands (at the example of the Niger inner delta). The next step is to simulate the increase and decrease of the wetland with Lisflood.
In WP4 the input data required for the tasks on flash flood and drought development have been obtained, input data for the task on post-processing are ongoing. A benchmark post-processing method has been selected that in general improves the GloFAS forecasts by 10%. Developed post-processing methods will be compared against this benchmark. The methodology to create a flash flood product has been defined, work will start shortly to optimise the definition of flash flood warning areas. The drought indicators have been defined for each of the three time scales, the next steps are to begin integrating forecast data.
WP5 activities have primarily focused on technical preparation and support. Best practices for software development have been established, and a central Data Hub has been set up to support the project. Guidance on utilising ECMWF systems and software has been provided. Initial modifications to the Hydrological Forecast Modelling Chain have been implemented in preparation for prototype use, and both input and control datasets have been generated. Discussions with WP1-4 have been conducted to gain a deeper understanding of their technical requirements.
It is too early at this stage to include results beyond the state of the art and their potential impacts.