Remote sensing and machine learning for lithology and mineral detection in NW, Pakistan

Abstract Multispectral and hyperspectral remote sensing represent critical instrumental methodologies that significantly enhance geological mapping and facilitate the identification of mineral resources. Collisional remnants of Himalaya-Hindukush northern Pakistan offer a unique opportunity to explore the hydrothermal deposits hosting economic minerals, inaccessible due to rugged and steep topography. The primary applications of ASTER and ZY1-02D remote sensing data include the extraction of alteration zones to detect economically significant minerals and depicting lithological units. To assess potential mineralization zones in the rugged terrain of northern Pakistan, herein, we implement integrated remote sensing techniques like false color composite, decorrelation stretching, band ratio, machine-learning techniques (maximum likelihood classification, spectral angle mapper, neural net classification, and match filtering), and validation (field observations). The results revealed 11 lithological units and several mineralization zones with key indicator minerals. These zones contained clay minerals, i.e. kaolinite, muscovite/sericite, and propylitic minerals, including calcite and chlorite, indicating copper and iron mineralization. The detected minerals emphasize that hydrothermal alteration in the region, controlled by the hot fluids from the granitic plutons (Tarichmir, Garam Chashma, and Kafiristan) of northwest Pakistan. The acquired mineral signature from remote sensing data was validated with ground truth observations, spectrometry, and petrographic analysis. Furthermore, this study proved the effectiveness of implemented approaches in delineating mineralization zones and discriminating lithological units.