Blab AI
NASA and IBM Release Open-Source AI Model to Map Lunar Ice, Craters and Volcanic Features
On September 10 2026, NASA and IBM announced the launch of the NASA‑IBM Lunar Foundation Model, a free‑to‑use artificial‑intelligence system that turns more than a decade’s worth of lunar observations into detailed, actionable maps.
The new foundation model was trained on over 30 distinct data layers collected by nine instruments aboard four NASA missions, including the long‑running Lunar Reconnaissance Orbiter (LRO). By fusing these layers—ranging from high‑resolution imagery to topographic, thermal and spectroscopic measurements—the AI can spot subtle features that single‑instrument analyses often miss.
According to the developers, the system boosts feature‑detection accuracy by up to 23 % compared with conventional methods. In benchmark tests, the model correctly identified key lunar features—such as crater rims, volcanic vents and, most importantly, potential ice deposits in permanently shadowed craters—more often than the algorithms that have been used for years. The increased precision is expected to help scientists locate safe landing sites and assess resource availability for future missions.
"Collecting data is only part of the job," said Kevin Murphy, NASA’s chief science data officer. "The foundation model is intended to make that data easier for researchers to explore and use. By integrating observations from multiple instruments, the AI can generate comprehensive maps without requiring scientists to process each dataset separately."
The model’s focus on permanently shadowed craters near the lunar poles aligns directly with NASA’s Artemis program. Water ice trapped in these cold‑trap regions could supply drinking water, oxygen and hydrogen for rocket fuel. Artemis IV, scheduled for early 2028, will land astronauts in the lunar south‑pole region, and the AI’s ice‑mapping capability could help pinpoint the most promising locations for future habitats.
Juan Bernabe‑Moreno of IBM Research added, "Uncovering the mysteries of the Moon requires an ability to learn from an extraordinary volume of scientific data." The open‑source release means that researchers worldwide can download the model, adapt it to their own data, and contribute improvements.
The LRO, launched in 2009, has delivered high‑resolution imagery and topographic data that cover 98 % of the lunar surface. The NASA‑IBM model builds on that legacy by combining LRO data with observations from earlier missions such as the 1960s Lunar Orbiter program and newer instruments. The fusion of datasets allows the AI to detect subtle features that may be invisible in single‑instrument analyses.
The foundation model is among the first publicly available AI systems specifically built for planetary science. Its release follows a growing trend of large‑scale machine‑learning models being applied to scientific data, but the lunar focus sets it apart from models developed for Earth observation or medical imaging.
Because the model is open source, it invites collaboration across the scientific community. Researchers can submit new data layers, refine the training set, and share results on platforms such as GitHub. This collaborative approach aligns with NASA’s broader open‑data strategy, which has made thousands of planetary datasets freely available through the NASA Space Science Data Coordinated Archive.
While the AI model does not yet replace the need for ground‑truth validation, it provides a powerful tool for prioritizing areas for future robotic or crewed missions. By highlighting potential ice deposits and mapping crater distributions, the system can help mission planners assess landing risks and resource prospects.
The release comes at a time when the Artemis program is preparing for its first crewed lunar landing. NASA’s schedule calls for Artemis IV in early 2028, followed by annual landings thereafter. Accurate lunar maps will be essential for the safe navigation of landers and for the long‑term establishment of lunar infrastructure.
In summary, the NASA‑IBM Lunar Foundation Model represents a significant step forward in lunar data analysis. By combining multi‑instrument observations with AI, the model improves feature detection, supports resource mapping and provides an open platform for the global lunar science community. The next phase will involve integrating the model into mission‑planning workflows and expanding its training data as new lunar missions return additional observations.
The new foundation model was trained on over 30 distinct data layers collected by nine instruments aboard four NASA missions, including the long‑running Lunar Reconnaissance Orbiter (LRO). By fusing these layers—ranging from high‑resolution imagery to topographic, thermal and spectroscopic measurements—the AI can spot subtle features that single‑instrument analyses often miss.
According to the developers, the system boosts feature‑detection accuracy by up to 23 % compared with conventional methods. In benchmark tests, the model correctly identified key lunar features—such as crater rims, volcanic vents and, most importantly, potential ice deposits in permanently shadowed craters—more often than the algorithms that have been used for years. The increased precision is expected to help scientists locate safe landing sites and assess resource availability for future missions.
"Collecting data is only part of the job," said Kevin Murphy, NASA’s chief science data officer. "The foundation model is intended to make that data easier for researchers to explore and use. By integrating observations from multiple instruments, the AI can generate comprehensive maps without requiring scientists to process each dataset separately."
The model’s focus on permanently shadowed craters near the lunar poles aligns directly with NASA’s Artemis program. Water ice trapped in these cold‑trap regions could supply drinking water, oxygen and hydrogen for rocket fuel. Artemis IV, scheduled for early 2028, will land astronauts in the lunar south‑pole region, and the AI’s ice‑mapping capability could help pinpoint the most promising locations for future habitats.
Juan Bernabe‑Moreno of IBM Research added, "Uncovering the mysteries of the Moon requires an ability to learn from an extraordinary volume of scientific data." The open‑source release means that researchers worldwide can download the model, adapt it to their own data, and contribute improvements.
The LRO, launched in 2009, has delivered high‑resolution imagery and topographic data that cover 98 % of the lunar surface. The NASA‑IBM model builds on that legacy by combining LRO data with observations from earlier missions such as the 1960s Lunar Orbiter program and newer instruments. The fusion of datasets allows the AI to detect subtle features that may be invisible in single‑instrument analyses.
The foundation model is among the first publicly available AI systems specifically built for planetary science. Its release follows a growing trend of large‑scale machine‑learning models being applied to scientific data, but the lunar focus sets it apart from models developed for Earth observation or medical imaging.
Because the model is open source, it invites collaboration across the scientific community. Researchers can submit new data layers, refine the training set, and share results on platforms such as GitHub. This collaborative approach aligns with NASA’s broader open‑data strategy, which has made thousands of planetary datasets freely available through the NASA Space Science Data Coordinated Archive.
While the AI model does not yet replace the need for ground‑truth validation, it provides a powerful tool for prioritizing areas for future robotic or crewed missions. By highlighting potential ice deposits and mapping crater distributions, the system can help mission planners assess landing risks and resource prospects.
The release comes at a time when the Artemis program is preparing for its first crewed lunar landing. NASA’s schedule calls for Artemis IV in early 2028, followed by annual landings thereafter. Accurate lunar maps will be essential for the safe navigation of landers and for the long‑term establishment of lunar infrastructure.
In summary, the NASA‑IBM Lunar Foundation Model represents a significant step forward in lunar data analysis. By combining multi‑instrument observations with AI, the model improves feature detection, supports resource mapping and provides an open platform for the global lunar science community. The next phase will involve integrating the model into mission‑planning workflows and expanding its training data as new lunar missions return additional observations.
