Moon: IBM and NASA launch AI to map ice and craters

The model was trained on a dataset comprising over 30 spatially aligned layers of information from nine instruments aboard four missions. This dataset includes data from the Lunar Reconnaissance Orbiter and NASA’s GRAIL mission, as well as complementary observations from the Japanese SELENE/Kaguya mission.

IBM and NASA state that the AI can be adapted for various tasks, including detecting areas likely to contain ice, mapping craters, and studying lunar volcanic formations. These operations are crucial for site selection, terrain risk assessment, and preparing for a sustainable human presence.

In benchmark tests, the model outperformed some widely used methods for identifying geographical features on the lunar surface by up to 23%. For areas with high ice potential, researchers report an error reduction of up to 22% compared to the SwinV2-B model. At a resolution of approximately 100 meters, crater detection improved by nearly 19% with half the amount of training data.

The project addresses a longstanding challenge in lunar research: missions have accumulated vast volumes of images, maps, and measurements from different sensors, often at incompatible resolutions. The new model seeks to unify this information into a common representation to more quickly highlight relationships that are difficult to detect with specialized tools.

An open AI for lunar exploration

The NASA-IBM Lunar Foundation Model joins the Prithvi family, developed by IBM and NASA around open models applied to Earth observation, weather, and heliophysics. IBM notes that the new lunar dataset associated with the model is also open, allowing scientific teams to adapt it for other uses.

This announcement comes as NASA prepares for the next phase of the Artemis program. The agency aims for its first new crewed mission to the lunar surface with Artemis IV in 2028, followed by Artemis V and the gradual development of a sustainable presence around the lunar South Pole.

Lunar ice is one of the primary objectives of this exploration. In regions that are permanently shadowed, it could provide water and oxygen for crews and, ultimately, contribute to the production of propellants for more distant missions. The model and its technical report have been publicly available on Hugging Face since September 10, 2026.