NASA and IBM Release Open AI Model for Lunar Science

NASA and IBM released a publicly available artificial-intelligence model on September 10, 2026, designed to help researchers analyze decades of Moon observations across multiple instruments and image resolutions.

The NASA-IBM Lunar Foundation Model is available for research, with the model hosted on Hugging Face and its complete codebase available through GitHub. NASA and IBM also released machine-learning-ready datasets and benchmark materials. NASA said the model is integrated into the open-source TerraTorch toolkit, giving researchers a common starting point for testing and adapting the system.

What trained the model

The model was trained primarily on roughly 2 million image tiles from NASA’s Lunar Reconnaissance Orbiter. That includes more than 1 million high-resolution camera images at 1-meter resolution and nearly 964,000 multispectral images at 100-meter resolution.

The training data also included high-resolution Moon imagery and terrain information from NASA’s GRAIL and Lunar Prospector missions and Japan’s SELENE/Kaguya mission. Combining those sources gives researchers a way to work across different types and scales of lunar information rather than relying on a single camera or instrument.

NASA described the release as part of its open-science effort to make large scientific datasets easier to use. Instead of building a separate machine-learning system from scratch for every research question, scientists can start with a model that has already learned broad patterns from lunar data and fine-tune it for a specific task.

What researchers can use it for

The model can support crater mapping, identification of irregular mare patches and detection of surface changes between observations. Crater maps help scientists study the age and history of lunar terrain, while irregular mare patches are volcanic features that may provide clues about how the Moon cooled and evolved.

It can also estimate where ice may be stable on or below the surface near the lunar poles. Permanently shadowed regions can remain cold enough to preserve ice for long periods, making them important targets for lunar science and future exploration planning.

Those ice-related results are estimates of ice prospectivity, not confirmation of an accessible deposit. The release does not announce a usable water discovery or select a final landing site. Nor is the model intended to control spacecraft or make safety-critical mission decisions without human review.

How the testing performed

NASA said the model matched or exceeded several baseline systems across the evaluated tasks, with a particular advantage in estimating polar ice stability. IBM reported that the cited testing showed up to a 22% reduction in error for ice-prospectivity estimation, nearly 19% better crater performance at 100-meter context resolution while using half the training data, and a 3% improvement in mapping irregular mare patches compared with cited baseline methods.

Those percentages come from NASA and IBM’s technical testing and should not be read as universal guarantees for every lunar dataset. The comparisons were made against specified baseline models and tasks. NASA also noted that changing illumination and observation conditions can affect the visibility of smaller features when researchers compare images taken at different times.

Why the open release matters

Public access gives universities, independent researchers and international science teams a common starting point for studying NASA’s lunar records. It also makes it easier to reproduce the reported results, test the model on new data and adapt it to questions the original team did not address.

The near-term value is therefore practical rather than autonomous: faster mapping, better prioritization of promising areas and broader use of existing Moon data. The model may support future exploration analysis, but scientists will still need to verify its findings with additional observations and mission-specific engineering work.

Sources

Look for updates to this story

Discover more from Interactive News

Subscribe to get the latest posts sent to your email.