Unsupervised Machine Learning for Lithological Mapping Using Geochemical Data in Covered Areas of Jining, China

Abstract

Application of (supervised and unsupervised) machine learning algorithms to big geoscience data can facilitate intelligent lithological mapping and interpretation in a data-driven fashion. Among them, unsupervised machine learning could play more important role in lithological classification because lithological labels are often missed in covered places. In this study, principal component analysis (PCA) and self-organizing maps (SOM) were applied to delineate major lithologies in the Jining region, Inner Mongolia, China, by integrating ten major-element stream sediment geochemical (K2O, Na2O, CaO, MgO, Al2O3, ΣFeO, MnO2, SiO2, P2O5 and TiO2) datasets. The results suggest that unsupervised machine learning using major-element stream sediment geochemical data is effective in capturing main lithological variations in a covered area. In detail, the PCA method identified only three interpretable components among nine principal components, whereas the SOM clustering recognized four categories including Archean TTG (tonalite–trondhjemite–granodiorite) gneisses besides felsic series, Cenozoic clastics and Cenozoic basalts. Apart from coinciding well with the pre-existing geological map in outcropping areas, the mapping results added new details regarding spatial distributions of felsic and basaltic rocks, which were previously recognized as Cenozoic clastics. This study highlights the ability of unsupervised learning for intelligent lithological mapping in covered areas using major-element stream sediment geochemical data.