Lithological Mapping using Artificial Intelligence and Remote Sensing data: A Case Study of Bab Boudir region, Morocco

Abstract

Lithological mapping is a crucial component of geological analysis, providing valuable insights into a region's mineralization potential and aiding mineral prospecting efforts. Manual execution of this task, especially in remote and resource-intensive areas, poses significant challenges. The integration of artificial intelligence (AI) techniques with remotely sensed data offers a swift, cost-effective, and precise approach to lithological mapping. In this study, machine learning algorithms (SVM, RF, and ANN) and deep learning techniques (CNN) were employed to map lithological units in an area, half of which lacked any published geological map. The study area is situated in the Bab Boudir rural municipality within the Taza province, geologically located in the Meso-Cenozoic cover of the Tazzeka inlier and characterized by moderate vegetation. Furthermore, the study evaluated the effectiveness of two types of remote sensing data: multispectral data from Sentinel-2 and hyperspectral data from Hyperion. The results revealed that the SVM and CNN methods achieved the highest overall accuracy and kappa coefficient, followed by the RF classifier, while the ANN approach yielded lower accuracies.