NASA and IBM Research launched an open-source artificial intelligence model on September 10, 2026, designed to extract insights from nearly two decades of lunar data and prepare for upcoming crewed missions to the Moon. The system automates geological mapping, crater identification, and ice deposit analysis across petabytes of planetary archives.
Operational Takeaways for the Lunar Foundation Model
- Infrastructure Efficiency: Built on 2 million image tiles from NASA’s Lunar Reconnaissance Orbiter, the open-source model eliminates the need for researchers to construct custom machine-learning systems from scratch.
- Exploration Targets: The AI identifies potential water ice stability zones near the lunar poles and tracks irregular mare patches to secure resources for future crewed bases.
Deployment Ecosystem: Hosted publicly on Hugging Face with source code published on GitHub, the project integrates directly into the TerraTorch geospatial toolkit for global research standardization.
Seventeen Years of Lunar Data Trained on Open-Source AI Architecture
The NASA-IBM Lunar Foundation Model relies primarily on observations collected by NASA’s Lunar Reconnaissance Orbiter, which has mapped the lunar surface in detail since 2009. Researchers also incorporated terrain and gravity signals from NASA’s GRAIL and Lunar Prospector missions alongside data from Japan’s SELENE spacecraft to build a comprehensive compositional index.
“NASA has spent decades building an extraordinary scientific record of the Moon, but collecting data is only part of the job,” Murphy stated regarding the launch. By deploying the model inside the open-source TerraTorch toolkit, the agency aims to streamline access to petabytes of scientific data and accelerate discoveries ahead of surface returns.
Automating Crater Mapping and Locating Polar Ice Deposits
Beyond surface photography, the foundation model targets critical resources required for long-term lunar habitation. Identifying these stable zones provides prospective access points for drinking water, breathing oxygen, and rocket fuel production.
The system also automates the manual measurement of impact craters and rare volcanic formations known as irregular mare patches. By bypassing months of manual image labeling, the AI allows planetary scientists to focus directly on geological interpretation and mission site selection.
| Dataset Component | Volume / Resolution | Primary Mission Source |
|---|---|---|
| High-Resolution Captures | 1 million tiles (1-meter resolution) | Lunar Reconnaissance Orbiter (LRO) |
| Multispectral Frames | 964,000 tiles (100-meter resolution) | Lunar Reconnaissance Orbiter (LRO) |
| Gravity & Composition | Planetary terrain signals | GRAIL, Lunar Prospector, SELENE (Kaguya) |
Standardizing Geoscience Tools Across Global Research Teams
The release of the Lunar Foundation Model marks the latest collaboration between NASA and IBM, following earlier Earth-focused “Prithvi” models and the “Surya” space-weather system. Internal development brought together the Impact AI team at Marshall Space Flight Center alongside scientists from the Planetary Science Division, Goddard Space Flight Center, and Ames Research Center.
By publishing the complete source code on GitHub, NASA enables independent research institutions to test novel hypotheses against identical baseline data. This standardized approach replaces fragmented, one-off machine learning systems with reusable assets designed to scale across upcoming planetary disciplines.
Disclaimer: The information provided in this article is for educational and informational purposes only and does not constitute financial advice.
