NASA and IBM Launch Open-Source AI Model to Accelerate Lunar Research
Space & Artificial Intelligence: NASA and IBM have released a new open-source artificial intelligence model designed to help scientists analyse the Moon’s surface and extract useful information from decades of lunar observations.
The NASA-IBM Lunar Foundation Model was launched on September 10, 2026. The model has been trained using a large collection of lunar imagery and geophysical information and is publicly available for researchers to experiment with and adapt to different scientific tasks.
AI Designed Specifically for Lunar Science
The new model is intended to help researchers process the enormous volume of information collected by lunar missions. NASA’s Lunar Reconnaissance Orbiter has spent years gathering detailed observations of the Moon, creating a vast archive that can be difficult to analyse manually.
The model was trained using approximately 2 million image tiles, including more than one million high-resolution camera images and nearly 964,000 multispectral images. Data from other lunar missions was also incorporated into the training process.
Searching for Ice Near the Lunar Poles
One of the model’s applications is identifying areas where water ice may be stable near the Moon’s poles.
Permanently shadowed regions can remain extremely cold because sunlight does not reach them directly. Scientists believe some of these locations may preserve ice for very long periods.
The AI model can help researchers analyse these difficult-to-study areas and generate maps indicating where ice may potentially be present or stable.
Faster Mapping of Lunar Craters
The system can also assist scientists in identifying and measuring impact craters.
Crater counts are important in planetary science because the number and characteristics of impacts can provide clues about the age of lunar surfaces and the history of the Solar System.
NASA and IBM say the model can process lunar surface information more efficiently than traditional approaches for several mapping tasks.
Studying Volcanic Features
Another potential application involves identifying unusual volcanic formations known as irregular mare patches.
These features are particularly interesting because some appear relatively young compared with traditional estimates of when major lunar volcanic activity ended. Mapping them could help scientists investigate the Moon’s geological and thermal history.
Multi-Mission Data Combined
The project goes beyond analysing images from a single spacecraft. NASA and IBM created a unified, machine-learning-ready dataset combining more than 30 spatially aligned data layers from nine instruments across four missions.
This gives researchers a broader view of the lunar environment by bringing different types of observations into a common framework.
Open Access for Researchers
The model and associated datasets have been released openly, allowing researchers around the world to examine the technology, adapt it to new scientific questions and develop additional applications.
The project is part of NASA and IBM’s broader effort to build AI foundation models for scientific research, following earlier work involving Earth observation and space-weather data.
The new lunar model could therefore become a useful research tool as scientists prepare for increasingly detailed investigations of the Moon. By combining large volumes of historical observations with AI-based analysis, researchers may be able to examine lunar features more rapidly and identify patterns that would be difficult to detect through conventional analysis alone.
