EarthGenerator a foundation model for Earth system modelling

EarthGenerator will deliver a foundation model of the Earth system, extending the WeatherGenerator model by integrating atmosphere, ocean, and land in a single, physically consistent model. By adopting a generative AI approach, EarthGenerator will provide a general-purpose capability adaptable to multiple downstream tasks from seasonal forecasting to multi-annual climate projections with minimal additional training. The project is organised into four themes: (1) data curation and optimisation, producing AI-ready datasets; (2) model development and training, advancing architectures with a module to enforce physical constraints and and enriched latent spaces, and scalable HPC training; (3) use-case applications, including improved prediction of climate extremes, quantification of the land carbon cycle, multi-scale early warning systems for food insecurity, and analysis of climate-induced migration and displacement; and (4) engagement and dissemination, ensuring open science and community adoption. By combining AI innovation with climate science and HPC expertise, EarthGenerator will accelerate research and development in climate change science, advance AI tailored for scientific needs, and strengthen Europe’s sovereignty in weather and climate modelling. Its results will directly contribute to the objectives of the Work Programme by advancing foundation models for science, bridging knowledge gaps across domains, supporting open science, and delivering societally relevant solutions to global challenges.