High-resolution maps of six lunar oxide abundances derived from MI–CF multi-source data using IPSO-optimized ensemble learning

This dataset provides high-resolution global maps of six major lunar oxides (FeO, TiO₂, Al₂O₃, CaO, MgO, and SiO₂) within the latitude range of 65°N–65°S at a spatial resolution of 59 m/pixel. The oxide abundance maps were generated using a multi-source remote sensing dataset that integrates KAGUYA Multiband Imager (MI) reflectance data and Diviner Christiansen Feature (CF) products (hereafter referred to as MI–CF data). An ensemble learning framework was employed, in which multiple regression algorithms—including support vector machine regression, extreme gradient boosting, and linear regression—were combined using a stacking strategy (SXL). To further enhance inversion accuracy, the SXL ensemble was optimized using an improved particle swarm optimization (IPSO) algorithm with nonlinear inertia weights and learning factors. The inversion models were trained and validated using oxide abundance measurements from returned lunar samples collected during the Apollo, Luna, and Chang’E missions. Compared with previous lunar oxide inversion products based on single data sources or traditional regression approaches, the IPSO-SXL model significantly improves prediction accuracy and demonstrates the advantage of jointly using MI and CF data. This dataset is suitable for studies of lunar petrology, crustal composition, and the Moon’s geological and thermal evolution, as well as for comparative analyses with future lunar missions and in-situ measurements.