NASA Launches Open-Source Lunar AI Model Trained on 2 Million Moon Image Tiles
Updated
Updated · Science@NASA · Sep 10
NASA Launches Open-Source Lunar AI Model Trained on 2 Million Moon Image Tiles
3 articles · Updated · Science@NASA · Sep 10
Summary
NASA released the NASA-IBM Lunar Foundation Model on Hugging Face and GitHub, giving researchers an open-source tool built specifically to analyze the Moon’s surface.
Roughly 2 million image tiles trained the model, drawing mainly on 17 years of Lunar Reconnaissance Orbiter data, including more than 1 million 1-meter-resolution images and about 964,000 multispectral images.
The pre-trained system can be fine-tuned with limited labeled data for crater mapping, irregular mare patch detection and polar ice stability estimates, where NASA said it matched or beat several baseline models.
Permanently shadowed polar regions and relatively young volcanic features are key targets because they could reveal usable ice resources, refine lunar geologic timelines and support future exploration planning.
The release extends NASA and IBM’s broader AI-for-science partnership, adding a lunar model to earlier Earth-observation and heliophysics systems while publishing datasets and benchmarks for reproducible research.