From Orbits to Insights: A Foundation Model for Lunar Science

Foundation Models (FMs) are large, general artificial intelligence (AI) models that can be applied to a range of AI and machine learning tasks. NASA’s Office of the Chief Science Data Officer plans to create an FM for each division in NASA’s Science Mission Directorate. Here, we report progress on the Planetary Science Division Lunar FM, an exciting and timely target for such an endeavor. Recent missions have gathered a large and diverse array of multi-modal datasets informing us about the Moon’s interior structure, surface geology and processes operating from the surface to the core. These datasets span a variety of modalities like high-resolution and multi-band imaging, spectroscopy, and synthetic aperture radar, covering a broad swath of the electromagnetic spectrum. Through these modalities, scientists can investigate a variety of planetary processes. This includes geologic mapping, surface process analysis, discovering evidence of water, hazard assessment for landing sites, and characterizing anomalously young volcanic features. This broad range of applications motivates an FM that can be applied to such diverse problems. This initial FM would be a basis for downstream machine learning models developed by the community that can power tools and approaches for advancing NASA’s long-term lunar exploration and discovery goals. In this presentation, we will highlight key successes and challenges encountered during the development of LunarFM, and outline how this effort scales toward a broader, large-scale approach to lunar science. Also, this first lunar FM may help inform the development of future FMs targeted to other solar system bodies.