NASA Releases Open-Source AI Model Trained on Lunar Data

NASA has released an open-source artificial intelligence model designed specifically for lunar science, trained on nearly two decades of Moon observation data. The NASA-IBM Lunar Foundation Model, developed through a collaboration between NASA, IBM Research, and several universities, is now publicly available on Hugging Face with its complete codebase on GitHub.

The model was trained primarily on data from NASA's Lunar Reconnaissance Orbiter (LRO), which has accumulated more data over 17 years than all other NASA planetary missions combined. Training data included roughly 2 million image tiles: more than 1 million high-resolution camera images at 1-meter resolution and nearly 964,000 multispectral images at 100-meter resolution. The model also incorporated imagery from NASA's GRAIL mission, Lunar Prospector, and JAXA's Selenological and Engineering Explorer.

Why it matters

Foundation models represent a shift in how scientific AI tools are built. Rather than training specialized algorithms from scratch for each research question, scientists can now adapt a single pre-trained model to multiple lunar research tasks using only small amounts of labeled data. This approach dramatically reduces the computational resources and time required to analyze planetary datasets, potentially accelerating discoveries across NASA's archive of petabytes of scientific observations.

Practical applications for lunar research

The model supports several specific research tasks. For scientists studying lunar ice, it estimates where ice patches are likely stable in permanently shadowed regions near the Moon's poles—areas that remain cold enough to preserve ice for billions of years. This capability has direct implications for mapping potential resources for future exploration missions.

Researchers investigating the Moon's volcanic history can use the model to identify irregular mare patches, unusual volcanic features that appear relatively young and challenge established timelines for when the Moon cooled. The model also automates crater detection and measurement, a labor-intensive process essential for dating the lunar surface and reconstructing solar system impact history.

In testing, the model matched or exceeded the performance of other baseline models across evaluated tasks. It achieved comparable results on crater mapping and irregular mare patch segmentation while demonstrating clear advantages in estimating polar ice stability.

Part of a broader AI strategy

The lunar model joins other AI tools developed through the NASA-IBM partnership, including the Prithvi family of models trained on Earth observation data for disaster monitoring and crop prediction, and the Surya model for predicting solar flares and space weather events.

"NASA has spent decades building an extraordinary scientific record of the Moon, but collecting data is only part of the job," said Kevin Murphy, NASA's chief science data officer and acting chief data and AI officer. "We also have to make data easier for scientists to explore and use."

The development team included experts from NASA's Marshall Space Flight Center, Goddard Space Flight Center, and Ames Research Center, along with researchers from the Universities Space Research Association, SETI Institute, University of Maryland Baltimore County, and Howard University. The team released machine learning-ready datasets and benchmark collections alongside the model, integrated into the open-source TerraTorch toolkit.

These details were first reported by NASA Science.