IBM and NASA Release Open-Source AI Model for Lunar Exploration

IBM and NASA have released an open-source AI foundation model designed to help scientists analyze the Moon using decades of observations from multiple missions. The NASA-IBM Lunar Foundation Model combines different types and resolutions of lunar data to help identify potential ice deposits, map craters and study volcanic features.

The model is trained on a unified dataset containing more than 30 spatially aligned data layers from nine instruments across four missions. The dataset brings together observations from NASA’s Lunar Reconnaissance Orbiter and GRAIL missions, along with complementary data from Japan’s SELENE/Kaguya mission.

AI Model Combines Multiple Lunar Data Sources

Studying the Moon has traditionally required researchers to work with observations collected by different instruments at different resolutions. IBM and NASA designed the new model to bring those datasets into a common representation, allowing researchers to use the same foundation model for several scientific tasks rather than developing a separate machine-learning system for each one.

The model can work with multimodal and multiresolution observations, including data describing the Moon’s topography, temperature and other surface characteristics. IBM and NASA have made both the model and the associated lunar dataset openly available so researchers can adapt them for additional applications.

Potential Ice Deposits and Craters

One of the model’s main applications is identifying areas that could contain lunar ice, particularly in permanently shadowed regions near the Moon’s poles. The system combines several measurements to estimate where conditions may be favourable for ice. Importantly, these are predictions of ice prospectivity, not direct confirmation that ice exists at a particular location.

The model also improves automated crater analysis. IBM and NASA reported that it matched the accuracy of specialised models when working at one-metre resolution and outperformed a SwinV2-B model by nearly 19% at a coarser 100-metre resolution while using half the training data. Crater maps can help researchers study lunar history and assist in evaluating potential landing and infrastructure sites.

Model Helps Study the Moon’s Volcanic History

Another application involves irregular mare patches, volcanic features that provide clues about the Moon’s geological and thermal history. IBM and NASA reported that the new model mapped these features 3% better than the comparison SwinV2-B model while achieving comparable accuracy with lower fine-tuning costs.

The results suggest that a single adaptable AI system can support several types of lunar analysis. Researchers can fine-tune the foundation model for individual tasks instead of starting from scratch each time a new scientific question arises.

Open Model Supports Future Lunar Research

The NASA-IBM Lunar Foundation Model is part of IBM’s wider Prithvi family of open foundation models for scientific applications. The release gives researchers access to the model and a machine-learning-ready lunar dataset rather than limiting the technology to the teams that developed it.

For NASA, the system could eventually support research tied to longer-term lunar exploration by making large volumes of existing observations easier to process. The model does not replace spacecraft or direct measurements, but it could help scientists narrow down areas that deserve closer investigation as lunar missions expand.