NASA and IBM Launch AI Model to Analyze Lunar Data
NASA and IBM Research have unveiled an open-source artificial intelligence model designed to analyze nearly two decades of lunar data. Announced on September 10, 2026, the NASA-IBM Lunar Foundation Model is intended to support future crewed missions to the Moon by providing new insights from extensive datasets.
The model was trained using approximately 2 million image tiles, including over 1 million high-resolution camera images and about 964,000 multispectral images. These datasets were sourced from NASA's GRAIL and Lunar Prospector missions, as well as JAXA's SELENE/Kaguya mission.
Performance evaluations indicate significant improvements over existing models. The new model reduces root mean square error by up to 22% when identifying regions with high potential for lunar ice, compared to the SwinV2-B model. It also shows a 3% improvement in capturing volcanic features and nearly 19% better crater detection at a 100-meter resolution.
The AI model integrates multi-modal, multi-resolution lunar observations into a unified framework, enhancing the accessibility and actionability of lunar data for the scientific community. This initiative aligns with NASA's strategy to prepare for sustained human presence on the Moon.
The model and its dataset are publicly available on platforms like Hugging Face, with source code accessible on GitHub, allowing for reproducible research and further development.