MLESM 2026: Advancing Machine Learning for Earth System Science
From 24 to 26 August 2026, researchers from the Earth system science and machine learning (ML) communities got together in Bonn and online for the fourth Workshop on Machine Learning for Earth System Modelling (MLESM). The workshop gathered more than 100 participants to talk about recent developments in machine learning for weather, climate, and Earth system applications.
Following the MLESM 2026 hackathon held the week before, the three-day event gave scientists a platform to present their latest work, exchange ideas, and discuss the challenges and opportunities of ML for Earth system modelling. Topics ranged from machine learning-based numerical weather prediction and high-resolution forecasting to climate models, Earth system components, observations, foundation models, and evaluating AI-based models.
The workshop was organized through a collaboration between researchers from Forschungszentrum Jülich, the Transdisciplinary Research Area (TRA) Modelling group from the University of Bonn, the University of Cologne, and the European Centre for Medium-Range Weather Forecasts (ECMWF), with support from the Center for Earth System Observation and Computational analysis (CESOC) and the HClimRep project. Martin Schultz, Head of the Earth System Data Exploration (ESDE) group at Jülich Supercomputing Centre (JSC), was part of the local organising committee.
Connecting research through weather and climate
Across the three-day workshop, the programme covered a broad series of developments in the rapidly evolving field of ML for Earth system science. Presentations addressed topics such as ML models, climate modelling, ocean and sea-ice applications, Earth observations and the development of foundation models.
Alongside the talks, poster sessions provided the opportunity for participants to discuss ongoing research in more detail and exchange ideas with colleagues in a more relaxed setting.
ESDE contributions to MLESM 2026
Researchers from ESDE also participated and contributed to the scientific programme. Savvas Melidonis presented a poster entitled HClimRep, a foundation model for capturing atmosphere, ocean and sea-ice interactions. His poster highlighted the ClimateAI group’s work on developing the first foundation model for climate science that simultaneously integrates atmospheric, oceanic, and sea-ice data and observations.
As in previous years, the MLESM workshop featured some of the most fascinating work from top researchers. What was different this year was the more honest, courageous, and active interaction among scientists, who were not afraid to formalize and diagnose what is often called the “black box” of AI. It was encouraging to see people extending the success of AI from weather to AI for climate, one of the most challenging topics of our time.
For some of our members, MLESM 2026 was also a chance to attend the workshop for the first time. We asked them what they found most interesting about the event, what surprised them the most, and what they enjoyed the most.
MLSEM provided a great platform to learn about the next generation of Earth system modelling. A mix of academics, industry researchers and entrepreneurs from across the world attended, bringing interesting perspectives into conversations. I particularly enjoyed the discussions on evaluation metrics beyond RMSE.
As a first time attending MLESM, I was glad to be in the same room with the top researchers of the field and getting to know the different school methods towards our research problems. What struck me the most is how open the researchers are to talk about failures as much as talking about success, which took the whole discussion into a new level that benefited everyone!
Each year MLESM workshop reflects the increasingly growing interest in ML as a tool for weather and climate research, while also highlighting the importance of interdisciplinary research, bringing together experts from the Earth system science, mathematics, and computational science communities. For ESDE, this year’s workshop provided an opportunity to once again contribute to the discussion and exchange of perspectives on the development of AI-based approaches for weather and climate modelling.
The next MLESM workshop is already scheduled for 30 August-1 September 2027. Mark your calendars!
Read more about the MLESM workshop on the official CESOC website.