A methodology for quantitative analysis of hydrated minerals on Mars with large endmember library using CRISM near-infrared data

Abstract

Hydrated minerals are good indicators of aqueous environments on Mars. Accurate mineral species identification and quantitative estimation of their abundance is very important for understanding past and present geologic and climatic processes of Mars. We present a novel methodology to analyze hydrated minerals using the Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) data. The methodology is driven by a sparse unmixing model, which can select the optimal subset of signatures from a large endmember library to best model mixed spectra. Laboratory measured spectra of minerals and rocks, scene-derived spectra of common components such as dust and synthetic flat and pure slope spectra are used to populate an endmember library. CRISM and the endmember library reflectance spectra are converted to single-scattering albedo with the Hapke radiative transfer model. The single-scattering albedos of the CRISM and endmember library spectra are then L1 normalized to eliminate the effects of the different measurement conditions between laboratory and orbital data. The sparse unmixing algorithm is finally applied to CRISM normalized single-scattering albedo data using the spectral library. Mineral identification and quantitative analysis are accomplished at the same time. The methodology shows good performance in synthetic and laboratory mineral mixture suites. Examples of six CRISM targeted mode images in three locations (Nili Fossae, Northeast Syrtis Major and Kashira crater), which have been previously thoroughly studied, were analyzed in detail in this study. In comparison with previous studies, our methodology identifies the mineral deposits detected through more manually intensive methods by expert analysts while simultaneously determining the mineral abundances. Furthermore, this new method provides an objective pathway to the determination of mineral presence from a large library of potential options. As an example of unexpected results, we detected putative serpentine in the Nili Fossae images for the first time. Even though the typical diagnostic spectral features used to identify minerals directly from CRISM data are obscured due to the mixing among mineral phases, T-test analyses of serpentine abundance provide critical statistical proof of its presence. Overall, the methodology we present works well for hydrated mineral detection with CRISM data and could provide a rigorous, objective method to map mineralogy in planetary hyperspectral remotely sensed data.