Training
Machine Learning for Earth Systems Modelling – three online courses
Understanding and modelling our planet is rapidly evolving as machine learning becomes increasingly central to Earth system sciences. Within the framework of the Destination Earth (DestinE) initiative of the European Commission (DG CNECT), three free online courses on Machine Learning (ML) for Earth systems modelling are offered throughout 2026. This series of three short online courses introduce participants to the fast-moving world of ML in Earth system modelling, from foundational concepts to advanced prediction and hands-on applications.
The courses form a progressive learning pathway. Course 1 provides an accessible introduction and broader context, while Courses 2 and 3 are aimed at a more technical audience. Course 2 focuses on the architectures, data and prediction workflows behind modern AI weather prediction systems, while Course 3 moves on to advanced applications and future directions across weather, climate and Earth system science.
Whether you are a researcher, practitioner, or technical specialist from a meteorological service, climate centre, academic institute, or industry, this training series will help you build the skills and confidence to:
- understand the fundamentals of machine learning in Earth system science,
- apply ML concepts within the DestinE ecosystem, and
- engage with next-generation artificial intelligence (AI) modelling tools.
All three courses are taught by experts from ECMWF and institutions across Europe and beyond, with expertise in AI and machine learning in weather prediction and climate science. The course series is developed by ECMWF in collaboration with the Karlsruhe Institute of Technology (KIT) and Wageningen University and Research.
Course 1 – Foundations and New Frontiers and Course 2 – Architectures, Data, and Prediction are available in self-paced format.
Registration is now open for Course 3 – Applications and Future Directions, starting on 5 October 2026.
What makes this training unique?
What sets this training series apart is that participants move from understanding the foundations of ML in Earth system science to working with modern AI weather prediction systems in practice and applying ML to real Earth system challenges.
A strong hands-on component connects the science directly to practice. In Course 2, participants work with code, data and models through Jupyter notebooks. This includes applied examples using ECMWF’s AI Forecasting System (AIFS) and Anemoi, the open-source machine learning framework co-developed by ECMWF and several National Meteorological Services, its Member Stat and Co-operating States. Participants learn how to prepare data, define and train an ensemble graph-transformer model, and generate a weather prediction.
Course 3 takes the next step: applying and critically assessing ML approaches for challenges such as extreme weather, downscaling, coupled Earth systems and data assimilation, while exploring emerging directions including explainability and trust, foundation models, hybrid modelling, long-range prediction and direct observation prediction.
How can AI models generate ensemble predictions?
| In this short video, Peter Düben, Head of the Earth System Modelling Section at ECMWF, explains how AI models can be used to generate ensemble prediction – one of the core ideas that Course 2 builds towards. |
Course information
.
.
Explore core ML concepts, how data-driven methods complement physics-based modelling, the role of ML within DestinE, and key questions around responsible and explainable AI. Designed for a broad audience and as a conceptual foundation for the more technical Courses 2 and 3. Learning activities include short videos, interactive exercises, quizzes and discussions.
For more details and registration, go to Course 1.
.
.
.
.
Explore how modern AI-based weather prediction systems are designed, trained and evaluated, from neural architectures and data preparation to deterministic and probabilistic prediction, model evaluation and benchmarking. Through Jupyter notebooks, learners work directly with code, data and practical prediction workflows, including examples using AIFS and Anemoi. Learning activities also include short videos, quizzes, discussions and expert-led sessions. The course builds on the foundations of Course 1 and prepares learners for the advanced applications explored in Course 3.
For more details and registration, go to Course 2.
–
.
Explore advanced applications of machine learning in weather, climate and Earth system science, including extreme events, downscaling, coupled Earth systems, data assimilation and long-range prediction. The course also examines explainability and trust, foundation and hybrid models, data-driven scientific discovery and direct observation prediction.
Learning activities combine expert lectures and talks with Jupyter notebooks using real-world workflows, readings, quizzes and discussions. By the end of the course, you will be better equipped to implement ML workflows for real Earth system applications and critically assess the strengths and limitations of emerging approaches.
For more details and registration, go to Course 3.
.
Objectives
These three online courses introduce the principles, methods, and applications of machine learning in Earth System Science. Together, they provide a structured learning pathway from conceptual understanding to the technical foundations of AI-based prediction and advanced applications across weather, climate and Earth system science. The courses explore how machine learning complements physical modelling and how AI-based systems are developed, evaluated and applied in practice.
After completing all three online courses, you will be able to:
- Understand core ML concepts and their relevance to Earth observations and prediction
- Understand real-world AI use cases, including AIFS and other prediction models, and their role within DestinE
- Identify and describe the ethical, regulatory, and societal aspects of AI adoption, including data privacy, bias, fairness and explainability
- Develop practical understanding of different machine learning types, training processes, and validation techniques in forecasting contexts
- Explain neural architectures (CNNs, GNNs, Transformers) and their role in representing atmospheric dynamics
- Explore and evaluate data management, optimization strategies, and compute requirements for large-scale model training
- Apply machine learning workflows to real-world applications in prediction and Earth system modelling, including extreme events, downscaling, coupled Earth systems and long-range prediction
- Explain advanced topics, including foundational models, hybrid modelling strategies, direct observation prediction and ML-based data assimilation
- Evaluate the operational readiness and limitations of ML systems and critical assess emerging research directions
Target audience
The training series is designed for different levels of prior knowledge.
Course 1 – Foundations and New Frontiers targets a broad audience of high-level information users and policy/decision makers from both public and private sectors, academia, and industry, as well as professionals in weather, climate and Earth system science who want an introduction to machine learning.
Course 2 – Architectures, Data, and Prediction is aimed at a technical audience in meteorology, climate science, Earth system modelling or related fields. Participants should already have basic programming experience and introductory knowledge of statistics and machine learning. Course 1 is recommended as a conceptual introduction.
Course 3 – Applications and Future Directions is an advanced technical course for researchers, developers and advanced PhD students and postdoctoral researchers in weather, climate, Earth system science and ML, including those working in operational NWP and climate modelling. Familiarity with Python-based ML workflows and Earth system modelling is expected. Prior completion of Courses 1 and 2 is recommended for participants without equivalent knowledge and experience.
More to explore
ECMWF is offering a wide range of training in the field of ML in Weather and Climate:
- Massive Open Online Course (MOOC) in Machine Learning in Weather and Climate is a self-paced and fully-online course, guiding you from the basics of machine learning up to practical implementations of ML models in Python.
- Discover Anemoi is a six-part series of training webinars that uncovers the key components of the Anemoi framework. Watch the recordings of all webinars below and see the presentation slides to learn how Anemoi works and how to use it.
- Browse our training catalogue for all of our training resources on machine learning.
Find out more about AI in DestinE, how AI is being used in weather forecasting at ECMWF in the AIFS Blog, browse our machine learning Jupyter Notebooks on GitHub, and see all of our upcoming training courses.
FAQ
Is there a cost to attend the courses?
No, the courses are free of charge.
Is it mandatory to take all courses sequentially?
No. None of the courses require any of the other courses as prerequisites. You can take Course 2 without taking Course 1 or 3, for example. Each course is designed to stand on its own
Are the courses self-paced?
The courses are largely self-paced. Each of the three courses will initially be delivered as a “live run” on specified dates, which will include interactive elements such as webinars and participant discussions.
After the live run, all course content will remain available and can be accessed at any time, allowing participants to learn at their own pace.
Do I need prior knowledge of machine learning?
The level of prior knowledge required depends on the course:
- Course 1 is designed for a broad audience, including high-level information users, policy and decision makers from the public and private sectors, academia, and industry. It is also suitable for technical professionals, such as non-ML meteorologists, who would like an introduction to machine learning concepts.
- Course 2 and Course 3 are aimed at a technical audience, typically with a scientific background in Earth sciences or related fields. Participants are expected to have at least a basic familiarity with programming and a background in statistics. Some of the more advanced sections assume prior knowledge of machine learning.
Where can I register?
Course 1 – Foundations and New Frontiers and Course 2 – Architectures, Data, and Prediction are available in self-paced format.
Registration is now open for Course 3 – Applications and Future Directions, starting on 5 October 2026.
© Title image: best – stock.adobe.com