Improved Detection of Lunar Water Ice Using Supervised Machine Learning Approach

Mini-SAR data for characterization of lunar craters with water ice has been done by the use of enhanced radar Circular Polarization Ratio (CPR) as an indicator of water ice. In this study, we have examined the ability of supervised Machine Learning (ML) technique to classify craters having anomalous high CPR in the cold traps of water ice in polar region. Since elevated CPR values alone, can be an ambiguous signature, caused by wavelength scale corner reflectors and presence of low volatiles such as water ice, study attempts to recognize dominance of anomalous class inside craters rim. In addition to CPR - a key indicator of frozen volatiles, considering backscattering coefficient, surface roughness and surface temperature as input parameters to support vector machine algorithm. The results obtained from supervised ML classification has enabled detection of additional 14 anomalous craters including Cabeus A, having favorable factors of surface temperature less than 120K, low surface roughness and low backscattering coefficient (SI) ~ −21.1 dB, Thereby enhancing detection of craters with water ice.