NASA and IBM Release Open-Source Lunar Foundation Model for Moon Science

NASA and IBM Research have unveiled an open-source lunar foundation model trained on 17 years of lunar observations. The model integrates data from multiple instruments and missions, creating a reusable resource for lunar science. Key applications include crater detection, ice deposit prediction, and segmentation of Irregular Mare Patches. The model excels in predicting potential ice deposits at the lunar poles, a critical resource for future missions.

The dataset, called SomBench, comprises nearly 2 million data bundles from 11 modalities and two spatial scales. It includes high-resolution images from the Narrow Angle Camera and multispectral images from the Wide Angle Camera. The data primarily comes from NASA’s Lunar Reconnaissance Orbiter, supplemented by missions like GRAIL and JAXA’s Kaguya/SELENE. The model accounts for lunar lighting conditions by incorporating imaging geometry as explicit context, helping distinguish illumination effects from actual surface properties.

The model outperformed common baselines, particularly in ice prediction, reducing error by up to 22%. It also showed strong performance in coarse-scale crater detection, improving by nearly 19% over the SwinV2-B baseline. The researchers emphasized that the model is a tool for analysis, not a substitute for physical measurements, as it cannot provide absolute geodetic positioning.

The NASA-IBM Lunar Foundation Model is publicly available, along with its code and datasets. It is part of the NASA-IBM AI for Science collaboration, which also developed the Prithvi model for Earth observation. The project contributes to the growing application of foundation models in scientific research, providing a common framework for analyzing lunar data.