NASA and IBM Release an AI Model Built to Study the Moon
foundation model
An AI that learns broad patterns first and can later be adapted for different tasks.
LRO
NASA’s spacecraft that has photographed the Moon for many years.
crater
A hole made when an object strikes the Moon.
What happened
NASA, the U.S. space agency, and IBM Research, IBM’s research group, released the NASA-IBM Lunar Foundation Model on September 10, 2026. It is one of the first open-source AI models made specifically for lunar science. It learned mainly from NASA’s Lunar Reconnaissance Orbiter, or LRO. Researchers can use it to map craters, find unusual young volcanic features, and estimate where polar ice may stay stable. The model is public on Hugging Face. Its code is available on GitHub. The team also released machine-learning datasets and benchmark collections.
Why this data matters
LRO has studied the Moon for 17 years. Its images cover almost the whole lunar surface in high detail. The model trained on about two million image tiles. More than one million were high-resolution camera images at one-meter resolution. Nearly 964,000 were multispectral images at 100-meter resolution. The training also used data from NASA’s GRAIL and Lunar Prospector missions. It included JAXA’s SELENE mission.
What a foundation model does
A foundation model does not learn only one task. It first learns broad patterns from a large dataset. Scientists can then fine-tune it for a new task with a smaller set of labeled examples. This can save researchers from building a new model from scratch each time.
Why the model matters
Craters are marks left by impacts. Counting and measuring them helps scientists estimate the age of lunar surfaces. It also helps reconstruct parts of solar system history. Faster mapping could leave researchers more time to interpret the results.
The Moon’s poles contain permanently shadowed regions. They are cold enough to preserve ice for extremely long periods. Mapping that ice could reveal more about lunar history. It could also identify possible resources for future exploration. The model can also help locate irregular mare patches. These young-looking volcanic features may challenge older ideas about how quickly the Moon cooled.
What has been shown
NASA says the model matched or exceeded several strong baseline models across the evaluated tasks. It performed comparably for crater mapping and volcanic-feature segmentation. It showed a clear advantage when estimating polar ice stability.
One test examined a new crater made when a SpaceX rocket body hit the Moon. The post-impact image was kept out of pretraining. Researchers later fine-tuned the model to detect the change. It identified existing craters and highlighted the new one. This suggests the model can adapt to unfamiliar surface changes. However, different lighting can make small craters harder to see.
What remains uncertain
These results come from NASA’s reported evaluations. They do not prove that the AI can confirm new discoveries by itself. Scientists still need to check its predictions against images and other observations. It is also unclear how consistent the model will be across every region, lighting condition, and research task.
What to watch next
The open release lets researchers worldwide reproduce, compare, and improve the work. Future studies will show whether it changes lunar maps, ice estimates, or our understanding of the Moon’s geological past. The important shift is practical: more scientists can begin with a shared lunar model instead of starting from zero.
A New AI Helps Scientists Read the Moon
📰 Full story: NASA and IBM Release an AI Model Built to Study the Moon
NASA and IBM made a public AI model for studying lunar pictures.
foundation model
An AI that learns general patterns before doing specific jobs.
LRO
NASA’s spacecraft that takes detailed pictures of the Moon.
crater
A hole made when a space rock hits the Moon.
💡 The gist
- It studies many pictures of the Moon.
- It can help find craters, volcano features, and possible ice.
- Scientists around the world can test and improve it.
NASA, the U.S. space agency, and IBM Research, IBM’s research team, released the model. Its name is the NASA-IBM Lunar Foundation Model. It learned mainly from NASA’s Lunar Reconnaissance Orbiter, or LRO. The spacecraft has photographed almost the whole Moon in detail.
The model studied about two million image tiles. More than one million showed the surface at one-meter detail. Nearly 964,000 showed different kinds of light. This training helps the AI recognize shapes on the Moon.
A foundation model learns broad patterns first. Scientists then adjust it for smaller tasks with labeled examples. This is faster than building a separate AI for every question. It does not mean the AI understands the Moon like a person.
A crater is a hole made when a space rock hits the Moon. Counting craters helps scientists estimate the age of lunar ground. It also helps them study the history of the solar system. The model can help make those maps faster.
It can also help locate unusual young volcanic patches. These features may reveal how the Moon cooled. The model can estimate where ice may stay near the lunar poles. Some polar places are always dark and very cold. Ice might remain there for a very long time. Finding it could teach us about the Moon’s past. It could also help plan future exploration.
NASA says the model performed as well as, or better than, several strong comparison models. Its clearest advantage appeared when estimating stable polar ice. It also helped detect a new crater after a SpaceX rocket body hit the Moon. The model had not trained on the later image.
The result is promising, but it is not perfect. Different sunlight can hide small craters. Scientists must check the AI’s answers. The model’s code and data are public. Researchers can now compare results and improve the tool together.
A Picture Helper for the Moon
📰 Full story: NASA and IBM Release an AI Model Built to Study the Moon
NASA and IBM made a computer helper that studies Moon pictures.
AI
A computer helper that finds patterns in pictures.
LRO
A spacecraft that takes pictures of the Moon.
crater
A hole made when a rock hits the Moon.
NASA is the United States space agency. IBM Research is IBM’s computer research team. Together, they made an AI for the Moon.
AI means a computer helper that finds patterns. This AI learned from many Moon pictures. NASA’s LRO spacecraft took many of them.
The AI can look for holes on the Moon. These holes are called craters. Rocks make craters when they hit the Moon.
It can also look for old volcano shapes. It can point to dark places where ice might stay. The AI is like a picture helper. It helps people search more quickly.
But it does not know everything. Sunlight can hide small holes. People still check the AI’s answers.
NASA and IBM shared the AI’s code and learning pictures. Other scientists can try it too. They may make better Moon maps later.