https://solarsystem.video/videos/watch/df4f10e9-57ca-49e5-a41f-79813632960a Presented by Saurabh Shubham (University of Maryland, College Park) ML4PSP Seminar September 2026 Understanding the compositional diversity of the lunar surface is essential for reconstructing its geological history and guiding future exploration. Traditional methods for lunar geological mapping rely on the mapper’s interpretation of spectral and morphological data. However, such an approach is subjective and relies on human judgment, which can introduce unintended bias. A data-driven spectral mapping approach can produce maps that are independent of the mapper’s goals. In this research, the spectral variability on the Moon is mapped using unsupervised clustering analysis. Here, an agglomerative "Ward" clustering approach on JAXA’s Kaguya Multiband Imager global data reveals seven spectrally consistent units. These units reflect differences in surface exposure age and global variations between mafic and feldspathic materials. Some units are associated with freshly excavated, plagioclase-rich materials. Other units show evidence of mixing between the bright, plagioclase rich surface with darker, more mafic materials that may have originated from deeper crustal layers, nearby volcanic terrains, or impact-driven transport. Our results show that machine learning can provide an independent way to map lunar surface diversity and identify regions where the surface preserves relatively fresh feldspathic crust or has been mixed with more mafic material. This approach helps reveal how the Moon’s crust has been modified by excavation, impact mixing, and space weathering through time. Tue, 20 Oct 2026 02:27:41 GMT https://validator.w3.org/feed/docs/rss2.html PeerTube - https://solarsystem.video https://solarsystem.video/client/assets/images/icons/icon-1500x1500.png https://solarsystem.video/videos/watch/df4f10e9-57ca-49e5-a41f-79813632960a All rights reserved, unless otherwise specified in the terms specified at https://solarsystem.video/about and potential licenses granted by each content's rightholder.