IBM AND NASA RELEASE OPEN-SOURCE AI MODEL FOR LUNAR DATA ANALYSIS

IBM and NASA have released the Lunar Foundation Model, an open-source artificial intelligence system designed to process decades of lunar observation data and map the moon's surface more accurately. The model is available to download from Hugging Face, an open-source AI repository, and will allow researchers to analyse petabytes of data gathered by robotic lunar missions over the past five decades. NASA and IBM also released a multimodal dataset alongside the model to enable wider access to lunar research. The partnership aims to support NASA's preparation for future Artemis missions and the establishment of sustained human presence on the moon.

The Lunar Foundation Model processes data from multiple sources including high-resolution optical cameras, laser altimeters, radar reflectance tools and spectrometers that measure elemental density—datasets that were previously difficult to unify into a cohesive picture of the lunar surface. Researchers compared the new model's performance against SwinV2-B, a Microsoft-trained vision system commonly used as a baseline for image analysis tasks. The NASA-IBM model reduced errors by 23 per cent when identifying areas where subsurface ice might be located and outperformed SwinV2-B by 19 per cent in identifying and classifying craters whilst using half the training data. The model recently demonstrated its capability when it correctly identified a new crater created by a SpaceX Falcon 9 rocket impact on 5 August, despite significant overlap with an existing crater.

Scientists can use the foundation model to address specific research questions without building a new AI system for each task. Potential applications include generating reliable crater maps to plot safe landing zones, analysing crater composition for clues about the moon's chemical makeup and history, and locating heavily shadowed sites that may conceal subsurface ice critical for long-term lunar bases. Dr. Juan Bernabé-Moreno, director of IBM Research Europe, UK and Ireland, stated the model's reusable architecture addresses a significant computational bottleneck in lunar data analysis.