Advancing lithological mapping using artificial intelligence and remote sensing data: A case study of the El hammam district (Moroccan Central Massif)

Lithologic mapping has been considered an accurate means of mineral exploration. However, traditional methods using fieldwork can be limited by accessibility and logistical issues. The current research combines multispectral remote sensing datasets with Artificial Intelligence algorithms. The region of El Hammam, which is bounded by major Hercynian faults, is characterized in terms of facies by the presence of quartzites, schists, and Paleozoic carbonates covered by Namurian volcanic rocks.The datasets of ASTER and Landsat-9 sensors were classified by using the Support Vector Machine (SVM), Random Forest (RF), and Artificial Neural Network (ANN). The accuracy of ASTER-ANN was the highest (F1 measure of 94.61%), followed by RF (F1 of 91.28%), then SVM (F1 of 86.53%). For Landsat-9 data, the highest accuracy was again for ANNs, with Kappa of 0.7364 and accuracy of 84.41%. The best differentiation was seen for dolerite and dolomitic limestone with ANNs and RFs, and less so for schist and flysch. Our results show that multispectral datasets, combined with artificial intelligence techniques, can ease the lithological characterization and can provide a useful guide for mineral exploration in highly deformed geological domains such as El Hammam.