Washington, D.C. – September 10, 2026
NASA and IBM have announced the open-source release of the NASA-IBM Lunar Foundation Model, a unified artificial intelligence system trained on petabytes of lunar observation data to help scientists map the Moon's surface, detect subsurface ice, and plan future exploration missions.
The model brings together data from NASA's Lunar Reconnaissance Orbiter, GRAIL mission, and complementary observations from Japan's SELENE/Kaguya spacecraft into a single framework. Scientists can now adapt the system to detect craters, identify volcanic features, and estimate polar ice stability using only small amounts of labeled data.
Breaking Through Data Barriers
For decades, researchers have collected extensive lunar data, but without a unified, machine-learning-ready framework, scientists had to manually sift through maps and images or rely on task-specific models that could not easily adapt to new research questions.
"NASA has spent decades building an extraordinary scientific record of the Moon, but collecting data is only part of the job," said Kevin Murphy, chief science data officer and acting chief data and AI officer at NASA Headquarters in Washington. "We also have to make data easier for scientists to explore and use. The NASA-IBM Lunar Foundation Model shows what's possible when we bring AI to NASA's petabytes of scientific data."
How It Works
The foundation model combines multimodal and multi-resolution observations to make predictions about the lunar surface. Researchers using the system have already demonstrated that it can match or exceed the performance of several baseline models across tasks including crater mapping and identification of irregular mare patches.
For polar ice detection, the model reduces error in identifying areas with high potential for ice by up to 22 percent compared to traditional approaches. At scale resolution, it outperforms previous methods by nearly 19 percent while using half the training data.
"The NASA-IBM Lunar Foundation Model gives scientists a foundation to explore the Moon at scale, connecting observations across instruments, revealing patterns that are difficult to see in isolation, and providing an open platform the global research community can build on," said a co-author of the technical paper accompanying the release.
Beyond the Model: Open Dataset and Tools
Alongside the foundation model, IBM and NASA scientists released the first open-source lunar dataset of its kind. This unified, machine-learning-ready dataset aggregates over 30 spatially aligned layers from nine instruments across four missions, providing the lunar science community with a rich, multi-modal view of the Moon's surface and subsurface.
The dataset and model are hosted publicly on Hugging Face, with the complete codebase available on GitHub for testing and experimentation. The system is integrated into the open-source TerraTorch toolkit, making it accessible to researchers who want to build custom models for specific lunar science tasks.
Support for Artemis and Beyond
The timing of the release aligns with NASA's Artemis program, which aims to establish a sustained human presence on the Moon. Lunar ice is critical for this effort, as water ice can be processed into drinking water, oxygen, and rocket fuel for missions to Mars and beyond.
Scientists can already use the NASA-IBM model to investigate multiple lunar phenomena, including craters, Irregular Mare Patches that reveal the Moon's volcanic history, and potential ice deposits near the poles. The model can help identify landing sites that minimize radiation exposure, map terrain to guide rover routes, and detect surface changes over time.
Research Impact and Future Work
The technical paper accompanying the release, authored by researchers from NASA, IBM, and multiple academic institutions including the Universities Space Research Association, SETI Institute, University of Maryland Baltimore County, Howard University, NASA Ames, and NASA Goddard, will be published through peer-reviewed channels.
Support for this release came from the Office of the Chief Science Data Officer's strategy for AI for science, part of a broader ongoing collaboration between NASA and IBM to apply advanced AI to scientific discovery across the planet and solar system.
The open-source nature of the release ensures that scientists worldwide can contribute to improvements, validate findings, and build custom applications for their specific research needs. As lunar exploration accelerates toward the late 2020s, tools like this will help the global research community turn vast amounts of satellite data into actionable insights for both scientific discovery and mission planning.
Source: NASA Science, September 10, 2026. IBM Research, September 10, 2026.
FIRAT Editorial Board
Institutional Research Desk · Foresight Institute of Research and Translation
The collective editorial and research translation board of FIRAT, synthesising peer-reviewed evidence, policy briefs, and division milestones across our seven foundational research pillars.
