Implementation of Machine Learning Algorithms for Mapping the Geological Units in the Vicinity of Posidonius Crater and the Eastern Rim of Mare Serenitatis

Document Type

Article

Publication Date

7-11-2025

Department

Department of Geological and Mining Engineering and Sciences

Abstract

Remote Sensing facilitates the systematic mapping of lunar geological units by integrating spectral, morphological and topographic data. The Moon’s diverse lithology, impact structures and space weathering effects provide an ideal site for the automated geological classification. This study utilized the Moon Mineralogy Mapper (M3) and Lunar Orbital Laser Altimeter (LOLA) data to implement two machine learning algorithms (random forest (RF) and support vector machine (SVM)) for classifying geological units in the Posidonius crater and eastern rim of Mare Serenitatis. Published geological maps were used to define training regions with spectral profiles of geological units serving as key parameters, supplemented by LOLA-derived elevation parameters. Classification was tested using training datasets of 30%, 50% and 70% to evaluate the data density effects on model performance. The highest classification accuracy was achieved with 70% training data, yielding an overall accuracy (OA) of 90.09% for RF and 78.4% OA for SVM, demonstrating RF’s superior performance. To evaluate the significance of mineralogical versus multi-sensor datasets, an independent RF classification was conducted using only M3 spectral bands. The standalone spectral data yielded an OA of 86%, demonstrating that mineralogical information alone provides robust classification but benefits significantly from the integration of topographic parameters. Inclusion of LOLA-derived parameters improved lithological boundary delineated by enhancing geomorphic contrast, mechanical competence differentiation and space weathering effects. For comparative assessment, a spectral matching technique using the Spectral Information Divergence (SID) algorithm was applied to the M3 dataset. The SID-based classification produced significantly lower accuracy (OA = 27.56%), highlighting the limitations of traditional spectral similarity approaches in geological mapping compared to machine learning models. This study highlights the potential of integrating multi-sensor datasets with machine learning algorithms for high accuracy lunar geological mapping. The proposed methodology provides a scalable and reproducible framework for automated geological classification on the Moon and other planetary bodies.

Publication Title

Journal of the Indian Society of Remote Sensing

Share

COinS