NASA and IBM Research have jointly unveiled an open-source artificial intelligence model designed to automate the geological mapping of the Moon’s surface, identifying volatile water ice deposits, impact craters, and historical volcanic formations from satellite imagery.
The model was developed by training deep neural networks on decades of lunar exploration data, including high-resolution digital elevation models, multispectral observations, and radar imagery captured by instruments aboard NASA’s Lunar Reconnaissance Orbiter (LRO).
Geological analysis of the lunar surface has historically required extensive manual annotation by planetary scientists, a process that limits the volume of data that can be systematically surveyed. The new foundation model automates the segmentation and classification of surface features, enabling automated detection of geological structures across vast lunar coordinate areas. The architecture is derived from the Prithvi geospatial foundation model family jointly developed by NASA and IBM for Earth observation.
NASA and IBM have released the model code, training architecture, and pre-trained weights as open-source software, making it accessible on public developer platforms for academic institutions, planetary science teams, and commercial space enterprises. By providing open access, the partnering organizations aim to accelerate collaborative research into the Moon's stratigraphic history and volatile distribution.
Planetary scientists noted that foundation models trained on remote sensing data offer significant advantages over traditional rule-based image processing algorithms by generalizing effectively across varied illumination angles and solar elevation conditions. The software pipeline allows mission operators to input raw orbit data and generate standardized digital elevation overlays, helping autonomous lander guidance systems identify landing hazard zones during approach phases.
The mapping of permanently shadowed craters and subsurface volatile deposits carries immediate operational significance for NASA’s Artemis program. Identifying viable water ice concentrations in the lunar south polar region is considered essential for future in-situ resource utilization (ISRU), where extracted ice could be converted into drinking water, oxygen, and liquid hydrogen-liquid oxygen rocket propellants for deep space exploration missions.