IBM and NASA have released an open-source AI model that transforms decades of lunar observations into insights, helping scientists identify features and accelerate exploration.
TURNING LUNAR DATA INTO DISCOVERYIBM and NASA have announced the open-source release of the NASA-IBM Lunar Foundation Model, one of the first publicly available foundation models designed for scientific exploration of the Moon.
MAPPING THE MOON IN GREATER DETAILPotential lunar ice deposits are one area where the model could support future exploration.
The presence of lunar ice could provide water and oxygen resources for a future Moon base, as well as rocket fuel for missions to Mars.
The NASA-IBM Lunar Foundation Model joins IBM and NASA’s Prithvi family of open foundation models, which spans geospatial, weather, heliophysics, and now lunar science.
IBM and NASA have released an open-source AI model that transforms decades of lunar observations into insights, helping scientists identify features and accelerate exploration.
TURNING LUNAR DATA INTO DISCOVERY
IBM and NASA have announced the open-source release of the NASA-IBM Lunar Foundation Model, one of the first publicly available foundation models designed for scientific exploration of the Moon.
Trained on an extensive lunar observation dataset curated by IBM and NASA researchers, the model enables scientists to work across decades of complex, multi-instrument data, helping uncover relationships and patterns that may otherwise remain difficult to identify.
The Moon has been continuously observed for decades, with sensors and instruments generating petabytes of data on its surface and subsurface. Yet analysing this information has traditionally required scientists to sift through maps and images manually or rely on lower-resolution, task-specific machine learning models.
The NASA-IBM model is designed to address these limitations by bringing multimodal and multi-resolution lunar observations into a common framework. In testing, it exceeded widely used approaches by up to 23 percent when identifying key geographic features, including potential ice deposits, craters, and volcanic formations.
MAPPING THE MOON IN GREATER DETAIL
Potential lunar ice deposits are one area where the model could support future exploration. Permanently shadowed regions are among the Moon’s most challenging environments to observe, but may contain ice beneath the surface. The presence of lunar ice could provide water and oxygen resources for a future Moon base, as well as rocket fuel for missions to Mars.
According to a NASA-IBM authored technical paper, the model reduced error in identifying areas with high potential for lunar ice by up to 22 percent compared with the SwinV2-B ImageNet model.
The technology can also help researchers investigate the Moon’s volcanic history. By identifying Irregular Mare Patches, scientists can better understand lunar volcanic activity and thermal evolution, while also gaining information relevant to future surface operations. The model captured the extent of these volcanic features 3 percent better than SwinV2-B when working with imperfect labels.
Crater detection represents another important application. Lunar craters can reveal information about the age and geology of different terrains, as well as the chemical composition of the early lunar interior. More detailed crater mapping can also support the selection of safe landing sites and help identify hazards and locations for future infrastructure.
At around 100-metre context-scale resolution, the model outperformed SwinV2-B by nearly 19 percent while using only half the training data.
AN OPEN FOUNDATION FOR EXPLORATION
Alongside the model, IBM and NASA scientists have created an open-source lunar dataset that brings more than 30 spatially aligned layers from nine instruments across four missions into a machine learning-ready framework.
The dataset combines tens of thousands of images and maps from NASA’s Lunar Reconnaissance Orbiter and GRAIL missions, alongside complementary lunar observations from the Japanese Aerospace Exploration Agency’s SELENE/Kaguya mission.
For NASA, the opportunity extends beyond simply collecting scientific data. Kevin Murphy, Chief Science Data Officer and Acting Chief Data and AI Officer at NASA Headquarters, highlights the potential for AI to make NASA’s vast scientific record easier for researchers to explore and use.
IBM’s Juan Bernabe-Moreno, Director of IBM Research Europe, UK and Ireland, similarly emphasises the model’s ability to connect observations across instruments and provide an open platform for the wider research community.
The NASA-IBM Lunar Foundation Model joins IBM and NASA’s Prithvi family of open foundation models, which spans geospatial, weather, heliophysics, and now lunar science. By providing a shared foundation that researchers can adapt to different scientific questions, the collaboration aims to accelerate discovery and support the next phase of lunar exploration.
