From 25–27 August 2025, the third workshop on Machine Learning for Earth System Modelling (MLESM) took place at the Universitätsclub Bonn, Germany, under the umbrella of the Center for Earth System Observation and Computational Analysis (CESOC), the European Centre for Medium-Range Weather Forecasts (ECMWF), TRA (Transdisciplinary Research Areas, University of Bonn). The event gathered around 140 international on-site participants and 100 virtual attendees, who listened to 46 presentations on cutting-edge science, explored more than 30 posters, and engaged in lively discussions on the future of machine learning in weather and climate research. The workshop and the following hackathon were co-organised by Forschungszentrum Jülich (FZJ), ECMWF, and the University of Bonn, with support from RAINA (DLR, BMFTR).
The programme highlighted the rapid progress in applying machine learning to Earth system modelling. Sessions addressed topics such as machine learning for numerical weather prediction, climate emulators, foundation models for atmosphere and hydrology, and the integration of Earth observations into these approaches. Discussions also covered model verification, explainability, and the creation of high-quality datasets for training and evaluation.
For the RAINA project, which contributes to the development of a foundation model for extreme precipitation and wind, several presentations showcased ongoing work:
- Poster: Verification of AI models – Adaptation and extension of an operational verification framework as part of the RAINA project (Britta Seegebrecht, DWD)
- Talks: RAINA Nowcasting – High-resolution nowcasting of precipitation and wind extremes with a foundation model for the atmosphere (Michael Langguth, FZJ)
- RiverMamba: A State Space Model for Global River Discharge and Flood Forecasting (Mohamad Hakam Shams Eddin, University of Bonn)
- The WeatherGenerator: a foundation model for weather and climate (Christian Lessig, ECMWF)
The workshop offered the RAINA community an excellent opportunity to exchange ideas with leading experts, gain inspiration from related projects, and strengthen collaboration on advancing machine learning methods for extreme weather prediction.
Further insights can be found here (CESOC) and here (ECMWF), and here (CESOC) about the MLESM Hackathon.
