Multivariate Analysis of Chemcam First Calibration Samples

Introduction: ChemCam is an active remote sensing instrument suite being built for MSL [1,2]. It uses laser pulses to remove dust and to profile through weathering coatings of rocks up to 9 m away. Laser-induced breakdown spectroscopy (LIBS) obtains emission spectra of materials ablated from the samples in electronically excited states over an analysis spot < 1 mm in diameter at the focal point of the incident laser beam. The flight model (FM) of the instrument composed of the body unit including the spectrometers and their electronics, and of the mast unit including the laser and the microcamera (RMI) and their electronics, went through intensive testing and calibration at CESR and LANL before delivery to JPL. The samples and calibration procedures are described in more detail in another abstract [3]. LIBS, with its narrow emission peaks and large number of channels, lends itself very well to multivariate analysis (MVA). Here we describe investigations of several MVA techniques aimed at rock classification using the ChemCam calibration data. The elemental analysis is not treated here per se. Samples and Calibration Details : This abstract reports on a subset of the 65 standards used for Chem-Cam calibration. The standards include, in pressed powder form, basalts, andesites, gypsums, olivines, limestones, dolomites, a graphite (for carbon), sediments of marine, river, and stream origin, and glass and ceramic standards to simulate Mars basalts and sedi-mentary rocks, respectively. Most of the details of the calibration are reported in another abstract at this meeting [3]. Briefly, the samples were analysed by ChemCam LIBS in 7 Torr of CO 2 using 50 laser pulses per analysis spot and 5 spots per standard at various distances. The data used in this study were taken during thermal-Mars-atmosphere testing up to 7 meters distance, a relatively long reach for ChemCam. Methodology: In spectroscopy, it is considered that the observed spectra are linear combinations of pure sources or poles that we want to estimate. We used two different multivariate analysis methods to achieve this goal with the LIBS spectra: the well-known Principal Component Analysis (PCA) and the Independent Component Analysis (ICA). PCA calculates principal components (PCs) which are linear combinations of the variables that convey the major part of the information contained in the global dataset. Spectra are decomposed in linear combinations of the