IBM and NASA Unveil Open-Source Lunar Foundation Model

IBM and NASA have collaborated to release an open-source lunar foundation model, which is now available on Hugging Face. The model has been trained on decades of observations of the Moon and combines data from nine instruments across four missions, including NASA's Lunar Reconnaissance Orbiter and GRAIL missions.

The dataset itself could be more important than the model, as it provides a unified, machine-learning-ready dataset of the Moon that was previously lacking. Scientists often work through maps and images manually or build separate models for individual tasks, but neither approach works well at scale.

One clear application of the model is finding ice on the Moon's surface. Permanently shadowed craters are promising locations for subsurface ice, which could provide water, oxygen, and raw material for rocket fuel. The model estimates where ice could be found by combining observations captured at different resolutions, reducing the error in identifying areas with high ice potential by 23% compared to SwinV2-B.

The model is part of IBM's Prithvi family, which includes models for geospatial data, weather, heliophysics, and the Moon. The same partnership previously produced a model with ESA designed to give an intuitive understanding of Earth.