Machine learning is becoming essential in Earth System modelling – from European Centre for Medium-Range Weather Forecasts – ECMWF’s AIFS forecast model to the design of foundation models, as pursued in our projects on weather and climate (e.g. WeatherGenerator and RAINA). It opens up new ways to handle complexity, improve efficiency, and extend predictive skill.
CESOC has launched a new Machine Learning Working Group to jointly address these challenges. At the kick-off, researchers from University of Cologne, Forschungszentrum Jülich, Deutscher Wetterdienst, ECMWF and others came together in a collaborative, forward-looking atmosphere.
We were inspired by two keynote speakers: Nikki Vercauteren on “Unresolved Variability of Turbulent Fluxes in Atmospheric Models: How Can Machine Learning Help Derive Physically Consistent Emulators,” and Wolf Ketter on “Strategic Economic Management for Developing EV Charging Hubs: Asset Planning and Investment Decision-Making.”
We had a lively discussion on the potential and limitations of ML in physics-based applications and on what it takes to make data ML-ready. Many overlapping interests and opportunities for collaboration became clear – across topics like physical consistency, data quality, and access to HPC resources such as the Jülich Supercomputing Centre (JSC) JUPITER.
We’re looking forward to exploring ideas together – taking on today’s and tomorrow’s challenges in Earth system science.
—
If you would like to register for membership and for more information, please visit the page: https://cesoc.net/become-member/ or contact us info@cesoc.net.