Anemoi: open-source winds of change in European weather forecasting

By: Prof Hannah Cloke, Department of Meteorology, University of Reading and ECMWF Fellow

Earlier this month I opened my inbox to the welcome news that the European Meteorological Society has chosen the Anemoi framework for its 2025 Technology Achievement Award. In the EMS announcement Anemoi is praised as “an excellent example of a European collaborative effort, offering enhanced forecast accuracy through advanced machine-learning methodologies. The flexible open-source approach enables many stakeholders to integrate AI in operational forecasting.” (ECMWF, European Meteorological Society)  

What is Anemoi? 

Named after the Greek gods of the winds, Anemoi is a freely available set of Python tools that turn raw weather data into machine-learning forecasts. Developed by ECMWF together with more than a dozen national weather services, the framework lets users mix-and-match modules for data preparation, model training and real-time inference—rather like building with Lego bricks. Because the code is Apache-licensed, everyone from large national centres to master’s students can stand on the same technological foundations. (ECMWF Events (Indico)) 

Real-world impact you can already feel 

The headline success story so far is AIFS, ECMWF’s AI-based global weather forecast. Since February it has been running four times a day alongside the traditional physics-based forecasting system (IFS), matching—or even beating—it for several near-surface measures while finishing a ten-day forecast in just minutes. (ECMWF) 

Crucially, the trained AIFS weights are openly shared, so other services have been able to fine-tune regional versions—Norway’s Bris in the Nordics and Germany’s AICON over central Europe, to name two—within weeks rather than years. 

Learning together: “Discover Anemoi” 

If you’d like to peek under the bonnet yourself, ECMWF has just published a six-part “Discover Anemoi” webinar series. Each recorded session—now freely available—walks through a different stage of the workflow, from creating datasets to running a model in real time. Slides, notebooks and videos are all online, making it easy for students and practitioners alike to follow along at their own pace. (ECMWF Events (Indico)) 

Why this matters 

Back in my 2020 post “House-building ban on floodplains isn’t enough,” I argued that clever science only helps if it reaches the people who need it. (University of Reading Blogs) Anemoi helps in two ways: 

• Speed – forecasts that once tied up a supercomputer overnight can now be produced on a single high-end graphics card. 

• Skills – our MSc students can spin up toy Anemoi experiments during term-time and feed the rainfall output straight into flood-risk models, gaining hands-on AI experience that was unimaginable a few years ago. 

What’s next? 

The Anemoi team are already testing AIFS-CRPS, a probabilistic upgrade that generates whole ensembles of forecasts instead of a single best guess. ECMWF plans to bring this version into operations later this year, giving decision-makers a clearer picture of forecast uncertainty. (arXiv) 

The EMS award is a well-earned pat on the back for colleagues across Europe, but it also reminds us of something more fundamental: open collaboration accelerates progress. By sharing code, data and ideas, we shorten the journey from scientific insight to real-world benefit. Long may these favourable winds keep blowing.