Integration of principal component analysis and supervised classification for mineral alteration mapping using satellite imagery, Kafta Humera, Ethiopia

Abstract

Remote sensing is an important technique for lithological mapping and hydrothermal alteration detection; yet, achieving high classification accuracy remains difficult in complex geological environments. This study addresses the difficulty of delineating lithological units and alteration zones in the Kafta Humera district of Ethiopia, a region within the East African Rift characterized by intricate volcanic and sedimentary formations. The primary objective was to evaluate the integration of Landsat-8 OLI and ASTER multispectral datasets for mapping basaltic rocks, Quaternary deposits, and mineralized zones while comparing the performance of supervised classification algorithms. The methodology involved atmospheric correction (FLAASH), Optimum Index Factor (OIF) analysis, and Selective Principal Component Analysis (SPCA) to reduce data redundancy and enhance spectral signatures. To classify these features, the study compared Maximum Likelihood Classification (MLC) and Support Vector Machine (SVM) methods. Results indicated that MLC outperformed SVM, achieving an overall accuracy of 80% compared to 73.5%. The MLC-derived maps accurately identified significant rock units—including granite, sandstone, and basalt—alongside diagnostic clay, iron oxide, and ferrous oxide minerals. ASTER’s shortwave infrared (SWIR) bands were particularly instrumental in distinguishing argillic, phyllic, and propylitic alteration halos. Furthermore, lineament analysis revealed critical structural conduits for hydrothermal fluids. In conclusion, the integration of PCA and MLC with multi-sensor datasets provides a robust, cost-effective framework for mineral exploration in rift-influenced terrains. To further enhance classification precision in geologically "noisy" environments, the study recommends the future adoption of hyperspectral imagery and advanced machine learning architectures.