AI and the World's Hard Problems · Earth & climate science · arXiv · 2026-10-19
Atmospheric Chemistry-Climate Modeling with an AI Foundation Model
Chemistry and aerosol processes are typically the most expensive part of atmospheric models and are often heavily simplified. The authors fine-tuned the Aurora Earth system foundation model on simulations from the chemistry-climate model UKCA, and the result reproduces the coupled system's daily fields, with 23 of 27 fields above R² 0.85. The model runs about 100 times faster than UKCA and stays stable over 50-year simulations, and adding MLS satellite ozone retrievals to training reduced zonal-mean ozone error by 23% between 250 and 20 hPa.