Central peaks of impact craters as probes of lunar crustal composition: Results from laboratory and remote spectral data analysis

The central peaks of 109 lunar impact craters are examined in Clementine UVVIS camera multispectral data. The peaks (from 40-180 km diameter craters) may have excavated material from 5-20+ km depth. The abundance and distribution of various peak lithologies are examined in light of crustal evolution models. The central peak compositions indicate a crust that is consistent with predictions based on global differentiation of a magma ocean, with a gradual trend towards more mafic compositions with depth, and intrusion of mafic units. These intrusions occur at all levels, but their compositions appear to be linked to depth within the crust. The number of distinct lithologies within each crater's peaks have also been determined. The lunar crust is locally diverse, with half the peaks containing multiple lithologies. The survival of distinct lithologies at many central peaks suggests that, as at terrestrial craters, pre-impact stratigraphy is preserved during peak formation.; Interpreting the central peak compositions depends upon distinguishing between rocks uplifted from depth, and those created during the impact process, such as impact melts. To better characterize the spectral properties of impact melts in remote data, two suites of lunar samples were measured in NASA's Reflectance Experiment Laboratory (RELAB) at Brown University. The results suggest that abundant quenched glass is a key component of impact melts found at lunar craters. More crystalline melts have unusual spectral features that may be related to cooling rate and rock texture.; Finally, future analyses of lunar multispectral images would benefit from more quantitative analytical approaches. Spectral mixture analysis can be used to model the spectral variability in multi- or hyper-spectral images and to relate the results to the physical abundance of surface constituents represented by spectral end-members. A new approach to end-member selection is presented in which end-members are derived mathematically from the image data subject to user-defined constraints. Detailed analysis of several examples reveal that this approach provides physically realistic end-members that may represent purer components than could be found in any image pixel.