GraphCast AI Weather Model Adapted to Mars in New Study

A recent study published on arXiv (2608.05054) explores how Earth-based weather foundation models can be adapted for use in planetary atmospheres, specifically by modifying GraphCast, a graph neural network designed for weather forecasting, for Mars. Although GraphCast excels in terrestrial predictions, its effectiveness in non-Earth settings had not been previously assessed. Utilizing the Mars Climate Database (MCD), which offers global atmospheric data across vertical altitudes akin to Earth’s pressure levels, the researchers analyzed both zero-shot and fine-tuned GraphCast forecasts of Martian temperature and wind. The initial zero-shot predictions were surprisingly accurate but struggled with diurnal variability and quickly reverted to climatological averages. To improve this, the team fine-tuned GraphCast with MCD data and solar radiation inputs while keeping humidity constant. This adjustment allowed for quicker adaptation to Martian thermal changes, indicating that leveraging Earth weather models could enhance planetary atmospheric studies. The findings underscore the promise of AI weather models for other planets, potentially advancing our understanding of Martian climate and aiding future explorations. The study was noted as a cross-type submission on arXiv, suggesting it may have been presented at a conference or journal.