Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

MarsCast: Transfer Learning of AI Weather Foundation Models to Planetary Atmospheres

Source

arXiv

arxiv.org/abs/2608.05054v1

Read the source

Source

Plain-Language Brief

Researchers adapted the GraphCast graph neural network weather forecasting model to Mars, achieving accurate predictions of current conditions with zero-shot training. Fine-tuning improved forecasts by capturing diurnal cycles, seasonal, and vertical temperature structures. This suggests that Earth-trained AI weather models can be used for Martian atmospheric simulations.

What the Document Says

  • The study investigates the transferability of Earth weather foundation models to Mars.
  • GraphCast graph neural network is adapted for Mars.
  • Zero-shot GraphCast forecasts provide accurate current conditions but fail to replicate diurnal variability and decay towards climatological mean states.
  • Fine-tuning GraphCast with MCD variables and top-of-atmosphere solar radiation improves predictions, capturing diurnal cycles and forecasting up to 10 days.
  • Improvement increases with training sample size and is sensitive to seasonal initialization.
  • The research shows that Earth-trained AI weather models can simulate Martian atmospheric dynamics.

Key Concepts Mentioned

  • GraphCast
  • Mars
  • weather forecasting
  • zero-shot learning
  • fine-tuning
  • Mars Climate Database
  • top-of-atmosphere solar radiation
  • diurnal cycle
  • seasonal temperature structure

Claims Made by the Document

  • GraphCast can be adapted for Mars.
  • Zero-shot GraphCast provides accurate current conditions but fails to replicate diurnal variability.
  • Fine-tuning GraphCast improves predictions by capturing diurnal cycles and seasonal temperature structures.
  • Improvement in forecasts increases with training sample size.

Evidence Used

  • The study states the adaptation of GraphCast for Mars.
  • The text mentions the results of zero-shot and fine-tuned GraphCast predictions.
  • The research provides data on improved forecasts after fine-tuning.

Methods / Approach

  • Adaptation of GraphCast for Mars, zero-shot learning, fine-tuning with MCD variables and top-of-atmosphere solar radiation

Limitations Stated by the Authors

  • The authors mention that the fine-tuned model still fails to fully replicate diurnal variability.

Limitations Noted by the Agent

Agent interpretation — not from the source.

  • Not stated in source: It is unknown if the study discusses the limitations of the used data or the generalizability of the results to all Martian regions.

Open Questions

  • What is the accuracy of the fine-tuned GraphCast model in predicting Martian wind fields?

Notes for Future Q&A

May include interpretation. Labeled where it does.

  • The research primarily focuses on temperature predictions, but it might be interesting to investigate the accuracy of wind field predictions.

Raw Abstract / Excerpt

We investigate the transferability of Earth weather foundation models to planetary atmospheres by adapting the GraphCast graph neural weather forecasting model to Mars. While GraphCast achieves state-of-the-art performance for terrestrial forecasting, its applicability to non-Earth environments remains unexplored. Using the Mars Climate Database (MCD), which provides global atmospheric fields across vertical altitude levels (similar to Earth pressure levels), we evaluate zero-shot and fine-tuned GraphCast predictions of Martian temperature and wind fields. Zero-shot forecasts produce a surprisingly accurate depiction of current conditions but fail to reproduce diurnal variability and rapidly decay toward climatological mean states. To address this limitation, we fine-tune GraphCast using MCD variables and top-of-atmosphere solar radiation forcing while holding humidity constant. Fine-tuning enables rapid learning of Martian thermal variability. Within as few as 10 training epochs, the model begins to capture the diurnal cycle and forecasts up to 10 days reproduce seasonal and vertical temperature structure. Prediction quality improves with training sample size and exhibits sensitivity to seasonal initialization. These results demonstrate that Earth-trained AI weather models can be adapted to simulate Martian atmospheric dynamics, providing a pathway toward rapid planetary weather prediction to support mission operations, dust storm risk mitigation, and future human exploration.