EarthGenerator

A Foundation Model for Earth System Modelling

EarthGenerator will deliver a foundation model of the Earth system. It extends WeatherGenerator by integrating atmosphere, ocean, and land. The model is physically consistent and based on generative artificial intelligence (AI). It provides a general‑purpose capability for multiple downstream tasks. These include seasonal forecasting and multi‑annual climate projections with minimal training.

The project is organised into four themes. First, data curation produces AI‑ready datasets. Second, model development advances architectures with physical constraints and enriched latent spaces. It also enables scalable high‑performance computing (HPC) training. Third, use cases address climate extremes, land carbon cycles, food security, and climate‑induced migration. Fourth, engagement and dissemination support open science and community adoption. The infrastructure is FAIR (Findable, Accessible, Interoperable, Reusable) by design and interoperable with European services.

EarthGenerator accelerates research in climate science and AI for scientific use. It improves prediction accuracy and supports decision‑making on global challenges. The project strengthens Europe’s sovereignty in weather and climate modelling. It supports open science and bridges knowledge gaps across domains.

CSC – IT Center for Science (CSC) contributes HPC expertise and data infrastructure. In Work Package 1 (WP1) and Work Package 6 (WP6), CSC supports data distribution and unified access application programming interfaces (API). In Work Package 2 (WP2), CSC contributes to scalable training and optimisation on EuroHPC systems. CSC co‑leads Work Package 7 (WP7) tasks on model efficiency, scaling, and distillation. CSC ensures efficient use of LUMI and future systems, and supports large‑scale model training and deployment.