The Copernicus Emergency Management Service (CEMS) has taken an important step forward in its operational flood forecasting capabilities. Since 10 September 2025, with the releases of European Flood Awareness System (EFAS) version 5.5 and Global Flood Awareness System (GloFAS) version 4.4, both flood forecasting systems now incorporate forecasts from ECMWF's Artificial Intelligence Forecasting System (AIFS) Single model.
The inclusion of AIFS in EFAS and GloFAS marks a milestone in the operational use of artificial intelligence (AI) for hydrometeorological applications. In EFAS, AIFS complements the existing meteorological forecast inputs (the IFS ENS and COSMO-LEPS ensembles and the ICON deterministic forecast) within the multi-model decision-making workflow. In GloFAS, it represents the first step toward a multi-model forecasting system, which is currently based solely on IFS ENS.
Why was AIFS introduced?
Multi-model advantages: The inclusion of AIFS Single in EFAS and GloFAS strengthens the multi-model approach of both flood forecasting systems. It increases the diversity of meteorological forecast inputs and opens the door for future AI model additions, such as the ensemble system of AIFS ENS, which became operational in July 2025.
Alignment with ECMWF developments: With the last two IFS cycles and IFS ENS upgrade to 9 km resolution, the high-resolution deterministic forecast (HRES), a key component of EFAS, has become redundant with the unperturbed member (Control, IFS CF) of the ensemble. In EFAS, AIFS Single has replaced HRES. For now, it inherits the HRES's forecast-skill-based weights in the flood warning workflow, with plans to recompute the weights of the different forecast inputs once a sufficiently long AIFS Single forecast period becomes available.
Competitive accuracy: Evaluation exercises show that AIFS often delivers higher forecast skill than the corresponding IFS CF across a range of variables and lead times. For precipitation, the main advantage of AIFS Single is in the medium range, while extreme events benefit less, mainly due to the currently lower resolution of AIFS Single (28 km vs 9 km). Further analysis will inevitably be necessary to fully understand the performance and added value of AI models in the multi-model context of the CEMS-Flood systems.
Operational innovation: By introducing AI into daily operations, EFAS and GloFAS combine scientific robustness with the speed and scalability of machine learning.
Future-readiness: The adoption of AIFS reflects a strategic shift toward hybrid forecasting systems that combine the strengths of physics-based and AI-based approaches.
What does this mean for EFAS and GloFAS users?
For users of EFAS and GloFAS, the integration of AIFS Single means flood forecasts will now deliver enhanced multi-model experience by including both physics-based and data-driven methods. This translates into richer and more complementary information for decision makers, humanitarian responders and water managers critical in situations where every possible new forecast scenario might make a difference.
The move also underscores a broader transformation underway within Copernicus and ECMWF: embracing artificial intelligence as a trusted component of operational Earth system forecasting. While traditional models will likely continue to provide the scientific backbone and necessary training datasets, AIFS and its integration into CEMS-Flood demonstrates the power of AI to strengthen and potentially revolutionise resilience and preparedness in the face of increasing flood risks worldwide.
With AIFS now fully operational in EFAS and GloFAS as of 10 September 2025, CEMS has entered a new era of innovation in flood forecasting.
Feedback from EFAS and GloFAS users about the new AI-enriched CEMS-Flood systems will be crucial and much welcome in the future for our quest to improve flood forecasting on the European and global scales.