Unmixing hypserspectral imaging datasets - a case study with martian meteorites

Abstract:

The scaling between the microscopic characteristics to the macroscopic spectral features links the in situ observations to orbital measurements of the planetary surfaces. Insights from spectral unmixing could help provide quantitative information to characterize the microscopic petrography of rocky exposures. Unlike in terrestrial applications, the origin and effect of endmember variability relevant to planetary surfaces is much less understood. In this study, we acquired hyperspectral images in visible and near infrared on nine Martian meteorites with various resolutions with the HySpex SWIR-384 imaging instrument. Based on the high-resolution images (24-50 microns/pixel) where the major mineral grains can be resolved, we test the linear spectral unmixing algorithms on the coarser-resolution images and the bulk meteorite spectra. From the analysis of the Martian shergottite NWA480, the bulk meteorite spectrum represents well the average composition of the major minerals of the meteorite. The linear spectral unmixing results with the Fully Constrained Least Square (FCLS) method for three images of different resolutions indicate consistent contributions of selected spectral endmembers. However, dark components, including maskelynite and Fe-pigeonite, are difficult to differentiate using spectral data alone. The trade-off between the abundances of dark components and the reflectance variability in the same mineral grain prevents an accurate representation of the surface with spectral unmixing. Algorithms considering the spectral variability of minerals, like those proposed for terrestrial canopy mapping, and supervised endmember selection would be crucial to improving the spectral unmixing results. The insights into the microscopic mineralogy and texture would help decipher future in situ measurements, such as those made by the SuperCam instrument onboard the Perseverance Rover.