Inverting a new hydrogen map on the moon derived from Chang'E-2 gamma-ray spectrum data

This study addresses the challenge of extracting subtle ²H characteristic γ ray (²H@2.223 MeV) count from the Chang'e-2 Gamma-Ray Spectrometer (CE2-GRS) data for inversing lunar water. Traditional methods interfered with by radioactive nuclides, such as ²¹⁴Bi, ²⁷Al, ³⁰Si, ⁴⁹Ca, etc., cannot retrieve the ²H@2.223 MeV count. The machine learning algorithms, which can efficiently uncover potential correlations between hydrogen and lunar geological features, have provided new insights for addressing inverse problems. Via integrating the Levenberg-Marquardt- Back Propagation (LM-BP) algorithm, three BP neural network models are trained to produce a new lunar global map of ²H@2.223 MeV count. The South Pole-Aitken Basin (SPA) basin on the lunar farside, the Mare Imbrium, and the Oceanus Procellarum on the lunar nearside exhibit pronounced high count(above 12 per 3 s). In contrast, the highland regions exhibit relatively lower counts (below 3 per 3 s). The high abundance of hydrogen in certain lunar regions is the cumulative result of various geological activities. In permanently shadowed regions (PSRs), the presence of dry ice (CO₂), which absorbs more thermal neutrons, may lead to differences in hydrogen concentration derived from neutron and spectral data. Furthermore, it is proposed that there is a higher probability of water in the De Gerlache crater.