Application of Machine Learning Techniques to Distinguish between Mare, Cryptomare, and Light Plains in Central Lunar South Pole−Aitken Basin

Abstract

We apply machine learning techniques to identify and map resurfacing units in the central South Pole−Aitken (SPA) basin using three Lunar Reconnaissance Orbiter (LRO) mission data sets: 321/415 nm and 566/689 nm band reflectance ratios from Hapke photometrically standardized albedo maps and a Terrain Ruggedness Index map using the Wilson et al. method. Other data were considered, but albedo and topography data were key in distinguishing between maria, cryptomaria, and light plains. A two-step image classification approach was applied to the data sets, an unsupervised K-Means algorithm followed by a supervised Maximum Likelihood Classification (MLC) algorithm. K-Means identified four units, one associated with dark smooth maria, two not associated with any particular features, and a fourth associated with edge effects. To further discriminate between the two nonassociated units, the K-Means unit map and an LRO morphologic basemap were used to select multiple training areas for three defined units in the MLC algorithm: mare, cryptomare, and cryptomare/light plains. From the training area values, the MLC unit map showed a distinction between the two prior indistinguishable K-Means units. Our results show (1) that the cryptomare from the MLC algorithm is in good agreement with cryptomaria mapped by J. L. Whitten & J. W. Head, (2) that the presence of scattered maria within large patches of cryptomaria indicates possible incomplete and/or uneven ejecta deposits or sheet flows covering cryptomare surfaces, and (3) a 79% increase in the total extent of cryptomaria compared to that by J. L. Whitten & J. W. Head for the same given study area in central SPA.

Keywords

• Lunar maria
• Lunar features
• Lunar science
• The Moon