NASA and IBM Open-Source AI Model to Map the Moon

NASA and IBM released the open-source Lunar Foundation Model, an AI trained on 17 years of lunar data that outperforms existing methods by up to 23% at mapping ice, craters and volcanoes.

NASA and IBM have announced the open-source release of the NASA-IBM Lunar Foundation Model, described by both organizations as one of the first publicly available foundation models built for scientific exploration of the Moon. The model is hosted on Hugging Face with its codebase on GitHub, and it was trained primarily on data from NASA's Lunar Reconnaissance Orbiter (LRO). "NASA has spent decades building an extraordinary scientific record of the moon, but collecting data is only part of the job," said Kevin Murphy, NASA's chief science data officer and acting chief data and AI officer. The release includes machine-learning-ready datasets and integration with the open-source TerraTorch toolkit.

Unlike traditional approaches that require building specialized algorithms from scratch for each task, foundation models are pre-trained on vast unlabeled datasets and can then be fine-tuned quickly for many applications. According to phys.org, the model was trained on roughly 2 million image tiles from LRO's 17-year record — more than 1 million high-resolution camera images at 1-meter resolution and nearly 964,000 multispectral images at 100-meter resolution — plus terrain data from NASA's GRAIL and Lunar Prospector and JAXA's SELENE mission. phys.org notes LRO's data output exceeds that of all other NASA planetary missions combined, making it ideal training material.

In benchmark testing, NASA and IBM said the model identified key lunar surface features with up to 23% greater accuracy than widely used existing methods, a figure reported by The American Bazaar, India Today and AiThority. Its clearest advantage came in estimating the stability of polar ice deposits. According to AiThority, the model reduced error (RMSE) in identifying high-potential lunar ice areas by up to 22% compared with the SwinV2-B (ImageNet) model, and outperformed SwinV2-B by nearly 19% on crater mapping at context-scale resolution using just half the training data. On crater segmentation of irregular mare patches it delivered comparable results with greater efficiency.

The surprise is that AI is now being applied to a planetary body beyond Earth at this scale — and that it is fully open rather than locked behind a corporate or government gate. The model joins IBM and NASA's Prithvi family of open science models, which spans geospatial, weather and heliophysics applications such as the Surya solar-weather model. IBM's Juan Bernabe-Moreno said the model gives scientists "a foundation to explore the Moon at scale, connecting observations across instruments, revealing patterns that are difficult to see in isolation." The collaboration paired NASA's Impact AI team at Marshall Space Flight Center with Goddard, Ames and academic partners including USRA and the SETI Institute.

The applications map directly onto NASA's Artemis program, which The American Bazaar and India Today report aims to return astronauts to the Moon in 2028. Ice in permanently shadowed polar regions could yield water, oxygen and rocket fuel for missions to Mars, and better prospectivity maps could help planners choose where robotic and human missions should go. Crater maps also matter for safety: AiThority reports crater mapping helps NASA select safe landing sites, avoid steep slopes and boulders, and plan locations for long-term lunar infrastructure. Studying irregular mare patches, which appear surprisingly young, could also revise timelines for how quickly the Moon cooled.

Because the model, code, benchmark collections and datasets are all public, researchers worldwide can now fine-tune it for lunar questions NASA has not yet addressed. Alongside the model, AiThority reports IBM and NASA scientists built the first open-source lunar dataset of its kind — a unified, machine-learning-ready collection aggregating more than 30 spatially-aligned layers from nine instruments across four missions. India Today notes the open release lets the science community explore the lunar landscape and, in IBM's words, help "the next generation of astronauts find their way around." Expect rapid follow-on studies as lunar exploration enters an increasingly active phase.