NASA and IBM have released an open-source artificial intelligence model called the NASA-IBM Lunar Foundation Model. It is designed to help scientists study the Moon and support future lunar missions. The system can analyse large volumes of lunar observation data. It can also identify features including potential ice deposits, craters and volcanic formations known as irregular mare patches.
The model was trained using data collected by multiple lunar missions. Its dataset includes more than 30 spatially aligned data layers from nine instruments across four missions.
Much of the data comes from NASA’s Lunar Reconnaissance Orbiter, which has mapped most of the Moon’s surface in high resolution. It also draws on data from NASA’s GRAIL and Lunar Prospector missions and Japan’s SELENE/Kaguya mission.
In benchmark testing against a SwinV2 transformer baseline, the model reduced errors in identifying high-potential ice areas by up to 22%. It also improved crater detection accuracy by nearly 19% at 100-metre resolution while using half the training data. NASA said the system can outperform widely used methods by up to 23% overall when identifying key lunar surface features.
The technology could help scientists study permanently shadowed polar regions that may contain ice beneath the surface. The resource could be important for future human missions and potential rocket fuel production.
The model’s weights are published on Hugging Face, while the full codebase is available on GitHub for researchers to test and adapt.
The release is the fourth in IBM and NASA’s Prithvi family of open foundation models. The family also covers Earth observation, weather and heliophysics.
