Estimation of elemental abundances of the lunar regolith using clementine UVVIS+NIR data

In this study we propose a regression model for the estimation of lunar elemental abundances from spectral features extracted from Clementine multispectral imagery in the visible and near-infrared domain. We extract a set of spectral features, including the continuum slope, the FWHM of the ferrous absorption trough near 1000 nm, and the wavelengths and relative depths of the absorption minima and inflection points present in the trough. As a “ground truth” for the elemental abundances we rely on the Lunar Prospector gamma ray spectrometer (LP GRS) data. With respect to the elemental abundances of the Apollo and Luna landing sites independently derived from returned samples, the best examined regression model is a second-order polynomial. The proposed regression-based approach allows an estimation of the elemental abundances of Ca, Al, Fe, Mg, and O at an accuracy of about 1 wt% and some tenths of a weight percent for Ti. We examine the influence of calibration of the Clementine UVVIS+NIR data and find that its effect on the results obtained with the regression approach is minor. Furthermore, we define a three-endmember model which allows the petrographic mapping of the lunar surface materials in terms of their Fe, Mg, and Al abundances. We examine the global distribution of Mg-rich rocks, the distribution of cryptomaria, and the occurrence of aluminous mare basalts in the Frigoris region. A possible regional compositional anomaly in northwestern Oceanus Procellarum is found, which corresponds to an extended area displaying spectral characteristics consistent with mare basalt containing significant amounts of olivine. On local scales, we examine in terms of our regression model the highland craters Proclus and Tycho, the compositionally anomalous central peaks of the craters Copernicus and Bullialdus, and the pyroclastic deposits on the floor of Alphonsus and on the northern rim of Petavius. As a general result, we show that the regression-based approach allows the detection of the main lunar terrain classes and rock types based on multispectral imagery in the visible and near-infrared domain.