A Landmark Week of Machine Learning for Earth System Modelling (MLESM) in Bonn
From 25 to 29 August, both a workshop and a hackathon in Bonn were dedicated entirely to the topic of Machine Learning for Earth System Modelling (MLESM) – for the JSC's Earth System Data Exploration (ESDE) research group, it was a particularly productive and rewarding event.
Vibrant Workshop: Thematic Sessions, Poster Presentations, and Networking
From 25 to 27 August, the third MLESM workshop took place at the Universitätsclub Bonn, bringing together around 140 participants on-site and another 100 online. The workshop created a vibrant space for exchange between the Machine Learning and Earth System Science communities. It was organised by the Center for Earth System Observation and Computational Analysis (CESOC) and the European Centre for Medium-Range Weather Forecasts (ECMWF) and was supported by the Transdisciplinary Research Area (TRA) Modelling at the University of Bonn, the University of Cologne and Forschungszentrum Jülich.
The programme featured thematic sessions, poster presentations, and networking opportunities, covering topics such as:
- Machine learning-based numerical weather prediction
- Climate emulators and data-driven climate models
- Earth system component models (ocean, land, atmosphere)
- Integration of Earth system observations
- Evaluation and explainability of machine learning models
- Datasets for training, benchmarking, and validation
The ESDE research group was involved in the organisation of the workshop, and several members gave talks on ongoing projects, including RAINA, the WeatherGenerator and HClimRep. These sessions sparked lively discussions and highlighted the role of machine learning in advancing forecasting and climate modelling.
High-Impact Hackathon: Coding Challenges and Expert Lectures
Right after the workshop from 27 to 29 August, around 25 early career researchers joined a three-day hackathon, supported by tutors and lecturers from ECMWF, NVIDIA, Forschungszentrum Jülich, and Helmholtz-Zentrum Hereon. The focus shifted to hands-on collaboration: coding, testing, and experimenting with new methods.
Participants worked on the following challenges:
- Model evaluation and post-processing for extreme events
- AI-based weather forecasting workflows and Anemoi
- Satellite data integration
- Model stability and long-term climate scales
- Diffusion Transformers for Earth system applications
In addition to coding, participants attended expert lectures on topics ranging from graph neural networks and foundation models to evaluating weather and climate models. The ESDE group took an active part in the hackathon, contributing both to the coding challenges and the exchange of ideas.
“My group worked on implementing a generative forecasting system based around diffusion models, like those used in image generation. Through the close supervision of the tutors on this project I learned a lot about the theory behind the models, beyond implementing them in code. It was a useful experience to learn how to prioritise the delivery of a finished product in the given time frame.” (Moritz Hauschulz, University of Oxford and intern at JSC)
A Week that Builds Community
Taken together, the workshop and hackathon showcased the breadth and depth of MLESM research: from theoretical discussions and project presentations to coding sprints and problem-solving in teams. For the ESDE research group, it was a week of sharing their own work, connecting with colleagues from across the globe, and contributing directly to the growth of a community at the intersection of machine learning, weather, and climate science.
It was a full week of MLESM – and one that will continue to shape how the ESDE group will work, collaborate, and innovate in the years to come.
More insights into the workshop and hackathon:
https://cesoc.net/key-insights-from-mlesm25/
https://cesoc.net/mlesm2025-hackathon/
