MANY MINDEDLunar Resource Mapping Model Shows Promise for Future Analysis
NASA and IBM have released an open-source lunar foundation model trained on 17 years of lunar orbiter data. The model processes nearly 2 million tile bundles across 11 modalities, integrating observations from nine instruments across four missions including NASA’s Lunar Reconnaissance Orbiter, GRAIL, Lunar Prospector, and JAXA’s Kaguya/SELENE. It uses about 1 million high-resolution images (roughly 1 meter per pixel) from the Lunar Reconnaissance Orbiter’s Narrow Angle Camera and 964,000 multispectral images (100 meters per pixel) from its Wide Angle Camera.
The model reduces prediction error for polar ice deposits by up to 22% compared to SwinV2-B and achieves 19% better performance on coarse-scale crater detection (100-meter scale). It processes lighting geometry explicitly rather than inferring it from raw pixels and uses FlexiViT to adapt to different image patch sizes without retraining. The model is publicly available on Hugging Face and GitHub as part of NASA-IBM’s 'AI for Science' collaboration.
This advancement supports future lunar resource analysis but does not create immediate abundance. The model cannot provide absolute geodetic positioning—latitude/longitude errors can reach dozens of degrees—and requires physical measurements for precise location. Performance metrics are based on limited test datasets, and the 22% ice prediction improvement specifically applies to polar ice deposit tasks. While the model shows potential for better resource mapping, it does not replace physical instruments or deliver immediate access to lunar resources.
What to watch: Controlled experiments isolating individual innovations and the model’s need for physical measurements to achieve accurate positioning. The source notes this is a foundation for future analysis, not a current abundance solution.
Source: The Decoder