Mapping the spatial distributions of oxide abundances and Mg# on the lunar surface using multi-source data and a new ensemble learning algorithm

Abstract

The spatial distribution of oxide abundances and Mg# (Mg/(Mg + Fe)) on the lunar surface is of great significance for in-depth understanding the origin and evolution of the Moon. China's Chang’E−5 (CE-5) mission returned young lunar soils for the first time, providing a new ground truth for the inversion of oxide abundances. In this study, the relationship between multi-source remote sensing data (including Chang’E−1 Interference Imaging Spectrometer (CE-1 IIM) data and the new global Christiansen feature (CF) product, named IIM-CF data), and the abundances of six oxides (FeO, TiO2, MgO, SiO2, Al2O3 and CaO) measured at 40 lunar sampling sites including CE-5 were analyzed. The use of IIM-CF data as the input features of the selected inversion models for obtaining the abundances of oxides, and the oxide abundances measured at the 40 sampling sites were used as the ground truth. The models selected for this investigation contain three typical algorithms − random forest (RF), extreme gradient boosting (XGBoost) and partial least squares regression (PLSR), and a new method integrates RF, XGBoost and PLSR together named RXP was developed in this study. The modeling results of the abundances of the six oxides derived from the above four algorithms show that the RXP algorithm outperforms the other three algorithms. The distributions of the six oxides and Mg# on the lunar surface covering the range from 70° N to 70° S (70° N/S) with a resolution of about 200 m/pixel were generated using the proposed RXP algorithm. Our results indicate that, compared with the result of a single data source, the use of IIM-CF data improved the accuracy of the modeling of oxide abundances and Mg#. It is expected that the CE-5 samples can bring additional references to the studies of the inversion for the oxides of the lunar surface and deepen our understanding over this issue.