IBM and NASA Launch Open-Source AI Model to Map the Moon and Support Future Lunar Exploration
IBM and NASA have released a new open-source artificial intelligence model designed specifically for lunar science, giving researchers a new way to analyze decades of observations of the Moon. Called the NASA-IBM Lunar Foundation Model, the system is designed to help scientists map craters, study volcanic features and identify areas where water ice could potentially exist.
The model was officially released on September 10, 2026, and is publicly available for researchers and developers. NASA said the model is among the first open-source foundation models created specifically for scientific exploration of the Moon.
The development comes as NASA and other space agencies prepare for a new era of lunar exploration, with future missions increasingly focused not only on visiting the Moon but also on understanding its surface, resources and suitability for sustained human activity.
What Is the NASA-IBM Lunar Foundation Model?
The NASA-IBM Lunar Foundation Model is an AI system trained on large quantities of lunar imagery and geophysical information collected by multiple space missions.
Instead of creating a separate machine-learning system for every scientific task, foundation models are trained to develop a broad understanding of their subject. Researchers can then adapt the model for specific applications using comparatively small amounts of labeled data.
NASA said the model was trained primarily using observations from the Lunar Reconnaissance Orbiter (LRO), along with data from missions including NASA’s GRAIL and Lunar Prospector and Japan’s SELENE/Kaguya spacecraft.
The training dataset contains roughly 2 million image tiles, including more than 1 million high-resolution camera images and nearly 964,000 multispectral images.
AI Can Help Map Moon Craters
One of the major applications of the new model is crater detection and mapping.
The Moon’s surface contains an enormous number of impact craters created over billions of years. Studying their size, distribution and geological characteristics can help scientists determine the relative ages of different lunar regions and understand the history of the early solar system.
Manually analyzing these features across the Moon is extremely time-consuming.
The NASA-IBM model can help researchers identify and classify craters much more efficiently. NASA said the model can be adapted to map craters at different scales, allowing scientists to process large quantities of lunar imagery and focus more of their time on interpreting the results.
Crater mapping could also have practical applications for future missions. Detailed knowledge of terrain can help researchers evaluate potential landing locations and identify hazards such as steep slopes and boulders.
Searching for Ice on the Moon
Perhaps the most important application of the IBM NASA Lunar AI Model is its ability to help researchers investigate potential water-ice deposits.
Scientists believe permanently shadowed regions near the Moon’s poles can preserve ice because some of these locations receive little or no direct sunlight and remain extremely cold.
Finding and mapping lunar ice is important because water could become a valuable resource for future human missions.
Water could potentially provide drinking supplies, while its components — hydrogen and oxygen — could also be used for life-support systems and rocket propellant.
The NASA-IBM model combines different types of lunar observations to estimate areas where ice could be stable. NASA reported that the model demonstrated a clear advantage in this task, while IBM said its testing showed up to a 22% reduction in error compared with a widely used SwinV2-B model for identifying areas with high potential for lunar ice.
Studying the Moon’s Volcanic History
The AI model can also help scientists investigate the Moon’s volcanic past.
Although the Moon is not generally considered to be volcanically active today, its surface preserves evidence of extensive volcanic activity from its geological history.
Researchers are particularly interested in structures known as Irregular Mare Patches, unusual volcanic features that may provide clues about how the Moon cooled and evolved.
The NASA-IBM model can help identify these structures across large areas of the lunar surface. IBM reported that the model captured the extent of these volcanic features more effectively than the baseline system used in its comparison while requiring less fine-tuning.
Mapping these features could help scientists refine their understanding of the Moon’s thermal and geological evolution.
Why Open-Source AI Matters for Lunar Science
A major part of the announcement is that the model is being released openly.
NASA said the model is publicly available through Hugging Face, while its codebase and related resources are also available for researchers to experiment with. NASA and IBM have released machine-learning-ready datasets and benchmark collections alongside the model.
This could allow scientists around the world to work with the same foundation model rather than developing independent systems from scratch.
Open access could also make it easier for researchers to fine-tune the model for new lunar science applications that NASA and IBM have not yet prioritized.
The model has been integrated into the open-source TerraTorch toolkit and joins IBM and NASA’s broader family of scientific foundation models.
Turning Decades of Lunar Data Into New Discoveries
NASA has collected enormous amounts of information about the Moon over decades.
The Lunar Reconnaissance Orbiter alone has been observing the Moon for approximately 17 years, producing a huge archive of high-resolution imagery and other measurements.
The challenge is no longer simply collecting information. Researchers also need tools capable of processing and connecting information from different instruments and missions.
The NASA-IBM model is designed to address that challenge by bringing different types of observations into a unified AI framework.
IBM and NASA researchers also created an open-source lunar dataset containing more than 30 spatially aligned data layers from nine instruments across four missions.
This allows AI systems to analyze multiple characteristics of the lunar environment rather than relying on a single type of image.
Supporting Future Crewed Missions
The timing of the model’s release is significant as NASA continues developing its long-term plans for returning humans to the Moon.
Future lunar exploration will require detailed knowledge of terrain, resources and environmental conditions.
AI could help researchers process this information before spacecraft and astronauts reach particular locations.
For example, automated crater mapping could assist with terrain analysis, while ice-prospecting tools could help identify promising regions near the lunar poles. Geological mapping could also provide scientists with a better understanding of areas that may be important for future surface operations.
The model is therefore not simply an AI experiment. It represents part of a broader effort to turn existing scientific data into practical tools for future exploration.
A New Role for AI in Space Exploration
The IBM-NASA collaboration reflects a broader shift in how artificial intelligence is being used in space science.
Instead of AI being limited to individual applications, researchers are increasingly developing foundation models that can be adapted to multiple scientific problems.
NASA and IBM have already worked together on other foundation models, including Prithvi, which focuses on Earth observation, and Surya, which uses solar observations for heliophysics research.
The lunar model extends that approach beyond Earth and into planetary science.
As lunar datasets continue to grow, AI could become an increasingly important tool for discovering patterns that would be difficult or impractical for researchers to identify manually.
What Comes Next for the Lunar AI Model?
The initial focus is on three major areas: crater mapping, volcanic-feature detection and lunar ice prospecting.
However, the open-source nature of the project means researchers can potentially adapt the model to other problems.
Future applications could include detecting changes between different lunar observations, improving geological maps, identifying previously overlooked surface features and supporting the selection of potential exploration sites.
NASA has already demonstrated that the model can be fine-tuned to detect newly formed impact features in lunar imagery, suggesting that it could eventually help researchers monitor changes across the Moon over time.
The broader goal is to make decades of lunar observations easier to use and transform them into new scientific discoveries.
Frequently Asked Questions
What is the IBM NASA Lunar AI Model?
The IBM NASA Lunar AI Model, officially called the NASA-IBM Lunar Foundation Model, is an open-source AI foundation model designed to analyze lunar imagery and geophysical data for scientific research and exploration.
What can the NASA-IBM Lunar Foundation Model do?
The model can be adapted for tasks including mapping lunar craters, identifying volcanic features and estimating areas where ice could potentially exist, particularly near the Moon’s poles.
Is the NASA-IBM Lunar Foundation Model open source?
Yes. NASA and IBM have released the model openly, along with related datasets and resources for researchers. The model is publicly available through Hugging Face, and its codebase is available through GitHub.
Why is lunar ice important?
Lunar ice could provide an important resource for future human missions. Water could potentially be used for drinking and life support, while hydrogen and oxygen derived from water could potentially support fuel production.
How does AI help lunar exploration?
AI can process enormous amounts of lunar imagery and scientific data much faster than manual analysis. It can help identify patterns, map geological features and highlight areas that researchers should investigate more closely.
What missions provided data for the model?
The model was trained using data from missions including NASA’s Lunar Reconnaissance Orbiter, GRAIL and Lunar Prospector, as well as Japan’s SELENE/Kaguya mission.
Could the model help future astronauts?
Potentially. By improving maps of craters, volcanic structures and possible ice deposits, the model could provide information useful for scientific planning, resource assessment and future lunar surface operations.
