Accurate mineral mapping in structurally complex terrains is essential for effective resource exploration. This study investigates the application of ASTER satellite imagery integrated with spectral processing techniques and supervised machine learning algorithms to identify key industrial minerals in the Semirom kaolin deposit, located in the Zagros fold-and-thrust belt, Iran. Methods applied include Band Ratio (BR), Relative Band Depth (RBD), spectral indices, Principal Component Analysis (PCA), Directed PCA (DPCA), Band Ratio Color Composite (BRCC), Random Forest (RF), and Support Vector Machine (SVM). Field surveys and X-ray diffraction (XRD) analyses were used for ground validation.
Kaolinite-rich horizons were reliably delineated using the Kaolinite Index (KLI), PCA, DPCA, and both machine learning classifiers, with SVM achieving an overall accuracy of 77.2% (Kappa = 0.67) and outperforming RF (69.1%, Kappa = 0.55). Calcite was best detected using DPCA applied to SWIR bands in combination with the Calcite Index (CLI), while hematite was only partially discriminated in mixed-pixel zones through BR and SVM classification. Major limitations included spectral overlaps between minerals, erosional cover, and mixed-pixel effects caused by heterogeneous surface conditions.
The results highlight the potential of integrating ASTER data with advanced image processing and machine learning as a cost-effective and reliable framework for preliminary exploration of clay-rich deposits. Furthermore, kaolinitic horizons at the Ilam–Sarvak boundary coincide with known bauxite occurrences,