From 23 to 27 February 2026, the European Centre for Medium-Range Weather Forecasts (ECMWF) hosted a five-day training course in Bonn dedicated to machine learning (ML) in the framework of the Destination Earth (DestinE) initiative of the European Commission. The course brought together scientists and software engineers from meteorology, climate science, Earth system modelling and ML from across Europe and beyond, to exoplore how Artificial Intelligence (AI) is transforming numerical weather prediction and Earth system science.

Within DestinE, artificial intelligence and machine learning are playing an increasingly central role. These techniques support innovative modelling approaches that complement the physics-based digital twins and contribute towards the development of an AI Earth system model.  

The five-day training course introduced experts in meteorology and climate science to the concepts and methods of machine learning, drawing on Earth system modelling examples from DestinE as well as on ECMWF’s Artificial Intelligence integrated Forecasting System (AIFS) and the Anemoi software framework, developed in close partnership with national meteorological services. 

A course tailored to Earth system sciences and machine learning  

ECMWF scientists guided participants through the foundations of machine learning before moving to the deep learning approaches that underpin cutting edge ML modelling, including graph neural networks, convolutional neural networks and transformers.

The programme also addressed deep generative models for ensemble forecasting, along with practical aspects such as model evaluation, training datasets and preparing models for operational use. Throughout the week, lectures and discussions were complemented by hands-on coding sessions using the European Weather Cloud. 

From research to operations: data-driven forecasting in practice 

A central focus of the training course was how machine learning methods are being increasingly used both for research and for operational forecasting. In a session on data-driven weather prediction, ECMWF Scientist Mariana Clare introduced the scientific foundations of AIFS and explained how it complements ECMWF’s physics-based forecasting portfolio.

Participants then explored these concepts in practice, running components of AIFS during hands-on sessions supported by ECMWF Scientists Jakob Schlör, Mariana Clare and Mario Santa Cruz Lopez. This provided first-hand insight into how data-driven models are trained, evaluated and prepared for operational use. 

Participants collaborating during hands-on machine learning coding sessions.

The course also featured dedicated sessions and practical exercises on Anemoi, a collaborative open-source software framework developed by ECMWF together with several national meteorological services across Europe. 

Through hands-on work with datasets, model configurations and training strategies, participants learned how Anemoi enables organisations to develop and run ML forecasting systems using their own data while benefiting from shared tools and community expertise. 

Another highlight of the course was the Forecast-in-a-Box concept, introduced by Harrison Cook, Research Software Engineer at ECMWF.

In his presentation, he illustrated how data driven weather models built with, Anemoi and related tools can be combined into modular, ready-to-use forecasting workflows. By packaging open-source software and AI models from ECMWF and Member States into reproducible pipelines covering data input, simulation, post-processing and visualisation, Forecast-in-a-Box enables users to run and adapt AI simulations even with limited computational resources.  

Extending machine learning across the Earth system 

Participants also learned about the development of AI-based models of Earth system components within DestinE, including land, hydrology, ocean and sea ice.

ECMWF Scientis Maria Luisa Taccari presented ECMWF’s global deep learning model “AI for Flood Forecasting” (AIFL) model as an example of how deep learning can extend predictive capabilities in hydrology.

AIFL addresses flooding hazards through a global deep learning approach. The model is fine-tuned with forecasts from ECMWF’s Integrated Forecasting System, IFS, and delivers fast, scalable predictions with lead times of up to ten days, supporting flood preparedness and early warning.

Strengthening Europe’s expertise in machine learning for weather and climate prediction  

By combining theoretical foundations, practical coding sessions and operational examples, the course demonstrated how machine learning is revolutionising weather and climate predictions. It also showed how data-driven methods are emerging as powerful complementary tools within the Destination Earth ecosystem. 

Through training initiatives such as this, ECMWF and Destination Earth continue to strengthen expertise across Europe and support the development of the next generation of climate and weather prediction capabilities. 

ECMWF is launching a series of ML online training courses. Find more information here.