Spectral Parameter-Based Prediction of Lunar FeO Content Using Random Forest Regression

The distribution of iron oxide (FeO) across the lunar surface is a key parameter for reconstructing the Moon’s geological evolution and evaluating its in situ resource potential for future exploration. This study applies a spectral-based approach to estimate FeO concentrations using remote sensing reflectance data combined with a Random Forest (RF) regression model. The model was trained on a dataset comprising 89 lunar samples from the Reflectance Experiment Laboratory (RELAB) database, supplemented with compositional data from Apollo samples available via the Lunar Sample Compendium and reflectance spectra from the Clementine mission. Spectral data spanning the visible to shortwave infrared range (415–2780 nm) were analysed, with diagnostic absorption features centred around 950 nm, typically associated with Fe2+. Model validation was conducted against FeO estimates from independent nearside locations not included in the training set, as reported by an external remote sensing study. The trained model was also applied to produce a new global FeO abundance map, demonstrating strong spatial consistency with recent high-resolution reference datasets. These results confirm the model’s predictive accuracy and support the use of legacy multispectral data for large-scale lunar geochemical mapping. This work highlights the potential of combining machine learning techniques, such as Random Forest, with remote sensing data to enhance lunar surface composition analysis, supporting the planning of future exploration and resource utilisation missions.