The Machine Learning for Earth System Modelling series has concluded its second course, following a hands-on programme exploring the methods and workflows behind AI weather prediction systems. The course series is developed within the European Commission’s Destination Earth (DestinE) initiative and supports researchers, practitioners, and technical professionals in understanding and applying machine learning (ML) in weather, climate and Earth system modelling. Registration for the third and final course is now open, which will focus on applications and emerging directions in ML for Earth system science.

The moderated live run of Course 2 in the Destination Earth (DestinE) Machine Learning for Earth Systems Modelling series has concluded. Course 2 – Architectures, Data, and Prediction – attracted 2,390 enrolled participants, and so far 305 participants have obtained a certificate of completion.

Course 2 built on the conceptual foundations introduced in Course 1 and focused on the theory and methods behind modern AI weather prediction systems. Participants progressed from deep-learning principles and neural architectures to data preparation, deterministic and probabilistic prediction, and model evaluation and benchmarking.

From theory to working with AI weather models

Jupyter notebooks were central to the course’s hands-on approach. Throughout the modules, participants could work directly with code, data, and models, applying the concepts introduced in the lectures to practical ML workflows. The notebooks covered topics ranging from the construction of simple machine learning algorithms and the data handling pipeline, to applied examples showing how to run ECMWF’s AI Forecasting System (AIFS).

The final notebook brought the main elements of the course together in a complete workflow using Anemoi, the open-source machine learning framework co-developed by ECMWF and several National Meteorological Services in its Member and Co-operating States. Participants were guided through preparing a dataset, defining and training an ensemble graph-transformer model of the ensemble-version of AIFS, and running the model to generate a weather prediction.

This allowed learners to explore some of the key developments that AI models are able to offer. In particular, the course demonstrated why single-model AI forecasts generate overly diffuse weather systems, and how the introduction of ensemble models in the past year has helped both sharpen and calibrate forecasts.

Watch: In this short video, Peter Düben explains how AI models can be used to generate ensemble predictions, one of the core ideas that Course 2 builds towards.

The course also contained a three-part Discover Anemoi webinar series. Experts from ECMWF introduced the open-source Anemoi framework from end to end, covering datasets, graphs and models, training and inference. Anemoi also underpins both ECMWF’s operational AIFS and the ML Earth system components for waves, sea-ice, ocean, land and hydrology built by ECMWF and its partners in DestinE. Together, the webinars and notebooks showed how the individual components connect within a framework designed to support both research and operational AI weather prediction and Earth system modelling.

Finally, the live programme featured a panel discussion, titled Towards Regional High-Resolution Weather Forecasting with Machine Learning. Experts from European meteorological services, ECMWF and academia discussed the advantages and drawbacks of the many emerging approaches to regional ML weather modelling which are being used across Europe including in DestinE: Limited-area models, stretched-grid models and downscaling. They also discussed challenges related to training data, probabilistic prediction, verification and operational trust. The discussion highlighted that the most suitable approach depends on the available data, computing resources, operational setting and intended application.

While the moderated live run has concluded, Course 2 remains available in self-paced format through the ECMWF learning platform.

Applications and future directions in Course 3

Registration is now open for Course 3: Machine Learning for Earth Systems Modelling – Applications and Future Directions, starting on 5 October 2026.

After having explored how modern ML prediction systems are designed, trained, run and evaluated, in the third and final course in the series participants will focus on how these methods can be applied to extreme weather prediction, Earth system and climate modelling, and data assimilation. The course will also examine how to interpret the models, how they fare in sub-seasonal forecasting, and new, emerging approaches.

Topics will include:

• ML applications for extreme events and downscaling
• explainability, trust and model failure
• foundation models and hybrid modelling
• sub-seasonal and long-range prediction
• coupled Earth systems and ML-based data assimilation
• end-to-end modelling, Forecast-in-a-Box and data-driven scientific discovery

Course 3 continues the applied approach of the series through expert lectures, Jupyter notebooks, readings, quizzes and community discussions. Learners will explore how ML workflows can be implemented for real Earth system applications and how the strengths and limitations of new approaches can be assessed critically.

The course is aimed primarily at:

• researchers in weather, climate and Earth system science, including advanced PhD students and postdoctoral researchers
• operational numerical weather prediction and climate-model developers and practitioners
• ML researchers working with environmental or geophysical data

Completion of Courses 1 and 2 is recommended for participants who do not already have equivalent technical knowledge and experience.

The free online course starts on 5 October 2026 and has an estimated study load of approximately 16 hours.

Register now for Course 3.

Find more information on the ML training page.

The course series is developed by ECMWF under the Destination Earth initiative in collaboration with the Karlsruhe Institute of Technology, Wageningen University and Wageningen Research, with contributions from experts across the weather, climate and Earth system science community.