NASA and IBM Built an AI to Find Ice on the Moon

We have decades of data about the Moon. For years, satellites and lunar missions have beamed back high-resolution images, topographical maps, and thermal readings. But sifting through all of that information to find exactly what you need is incredibly difficult.

To solve this, IBM and NASA just released the NASA-IBM Lunar Foundation Model. It is a massive, open-source artificial intelligence model designed to process all of that complex lunar data and help scientists prepare for a long-term human presence on the Moon.

Organizing Decades Of Lunar Data

Before this AI model, researchers had to manually combine data from different instruments. A map of the Moon’s gravity might not easily align with one of its surface temperature.

To fix this, IBM and NASA researchers built a unified dataset. They took more than 30 layers of spatial data from nine different instruments across four separate lunar missions and aligned them perfectly. Then, they trained the AI model on this unified dataset. Now, instead of looking at observations in isolation, scientists can use the AI to spot hidden patterns across all of those different measurements at once.

Hunting For Ice And Safe Landing Zones

One of the most important jobs for this new AI is finding resources. The model is specifically trained to identify potential ice deposits hidden in the permanently shadowed regions of the Moon. Water is heavy and expensive to launch from Earth, so finding lunar ice is critical for future astronauts who will need it for drinking water and rocket fuel. According to IBM, the new model reduces the error rate in predicting high-potential ice areas by up to 23% compared to general-purpose vision models.

The AI is pretty impressive when it comes to finding craters. It can figure out exactly where the craters and steep slopes are, which is a huge help for NASA as they look for the safest places to land the next Artemis missions. Planners get a detailed, almost street-level look at the Moon’s surface hazards, thanks to the AI’s sharp mapping. It even spots some of the Moon’s most unusual volcanic features—Irregular Mare Patches—that tell scientists more about how the Moon has changed over time.

An Open Source Tool For Everyone

Perhaps the biggest news is that this technology is not locked behind closed doors. IBM and NASA released both the AI model and the massive unified dataset to the public on the open-source platform Hugging Face.

Because the model is highly efficient, researchers do not need a massive computing budget to use it. This means universities, startups, and international space agencies can all access a cutting-edge tool to accelerate their own lunar research. By providing a shared foundation, IBM and NASA hope to accelerate discovery across the entire space industry.