Recent advances in AI models trained on Earth’s weather data have opened up new ways to model planetary atmospheres. But it’s still unclear whether these models can work on other planets like Mars. In this study, we test whether a weather prediction model trained on Earth can be used to forecast Martian weather and climate.

Our early results show that the Earth-trained model doesn't fully generalize to Mars, but it can still capture important patterns, like the daily temperature cycle. To improve its accuracy, we try two new approaches: (1) fine-tuning the model using simulated Mars weather data, and (2) training the model with added knowledge about Mars, such as its thin CO₂-rich atmosphere and different sunlight patterns.

This study shows that AI models built for Earth may provide a useful starting point for forecasting weather on Mars and other planets where data is limited.