Automatic Detection of Water and Mafics in M Radiance Images

Introduction: We describe the detection of water (OH/H 2 O) and mafic mineralogy absorption features in Ryder crater from a fully automated search for spectral anomalies in the Moon Mineralogy Mapper (M 3) lunar catalog [1]. We employ superpixel endmember analysis [2] to detect spectral outliers. Our approach operates on M 3 radiance data without explicit thermal emissivity or illumination corrections. The resulting spectra are not formal endmembers for linear unmixing or abundance estimation. However they are qualitatively interpretable and suggest the range of spectral diversity in the scene. These discovered features include water-rich areas of Ryder crater, revealing these materials without prior direction from spectral libraries or analyst-selected band ratios. The demonstration independently supports the significance of the ~3µm water signal as an important component of the scene's spectral diversity and shows the utility of this automated method to recognize geologically meaningful materials in M 3 data. Background: The M 3 imager aboard Chandrad-rayaan-1 collected 85 channels over the 430-3000 nm range with resolution of ~140 m/pixel. These data have previously facilitated large-scale maps of mafic mineralogy and provided direct evidence of water on the lunar surface [1]. As with any large hyperspectral database, analysis is complicated by noise and the natural size of the images. It is possible or even likely that aspects of local or regional spectral diversity still lie undiscovered near the level of noise. Automated anomaly detection algorithms such as endmember detection can assist the discovery process by directing analysts' attention to key surface mineralogies. These fully unsupervised approaches can complement directed methods such as band ratios; they are sensitive to a wide range of unexpected features which provides additional confidence that discovered signals are a significant component of the scene's spectral diversity. This work considers an M 3 image of Ryder crater (M3G20090125T172601). The crater ejecta has been previously associated with the presence of water [2]. Specifically, the spectrum contains water absorption features near 3µm. These absorption features have been independently corroborated in Cassini VIMS images of the lunar surface [3]. We consider the ability to detect this feature an important prerequisite for an anomaly detection method. Methods: We preprocess each large M 3 radiance image by breaking it into subframes of 3000 or fewer scan lines. Average scene emissivity and illumination are found to vary somewhat smoothly as a function of latitude [1]. This subframing provides piecewise constant regions that mitigate this variation. …