This five-day course focuses on machine learning (ML) for numerical weather prediction (NWP) and earth system modelling, in the context of the Destination Earth programme.
Destination Earth (DestinE) is an EU-funded initiative to develop digital twins of the Earth to better model and understand climate change and extreme weather. It is jointly implemented by ECMWF, EUMETSAT and the European Space Agency.
AI and ML are increasingly playing a central role in DestinE, and are being used to build ML-based earth system components such as ocean, sea ice and waves to complement the physics-based models underpinning DestinE's digital twins, supporting uncertainty quantification and enhanced interactivity.
This course will introduce those working in meteorology and climate science to the concepts and methods of modern machine learning, drawing on DestinE examples and applications and ECMWF's data-driven AIFS weather models. As there are many general courses on machine learning available – including free online courses – this course will have a particular focus on the use of machine learning in the domain of Earth System Sciences (ESS).
Main topics
The course will cover the following themes:
- An overview on the use of machine learning in ESS
- Introduction into the most important machine learning methods that are relevant for ESS, including deep learning approaches.
- Examples for the use of specific machine learning tools across the weather and climate prediction workflow and how they can be prepared for use in operational predictions.
- Data-driven forecasting, based around AIFS and the Anemoi framework
- An overview of the DestinE programme and examples of how machine learning is being applied within DestinE.
The course will consist of lectures, discussions, and hands on sessions with code examples.
Requirements
Participants should have a good meteorological or climate-science background, and a good familiarity with statistics. Participants should also have some limited experience with Python code and Jupyter notebooks. Basic experience with machine learning applications in Earth system sciences and the handling of Earth system data would be advantageous. Some practical experience in numerical weather prediction is an advantage.
All lectures are in English, and the training will take place at the ECMWF office in Bonn, Germany. As an EU-funded course, priority will be given to EU member states as well as ECMWF member states.