Deep gold prospectivity modeling in the Jiaojia gold belt, Jiaodong Peninsula, eastern China using machine learning of geometric and geodynamic variables

Abstract

Gold mineralization in the Jiaojia gold belt was formed in a structurally-dominant hydrothermal mineral system showing a close spatial association with the Jiaojia detachment fault. This study delves into the Jiaojia gold belt from the perspective of coupled spatial association and ore-forming processes by employing spatial analysis of three-dimensional (3D) models, 3D ore-forming numerical modeling, and 3D prospectivity modeling using machine learning techniques (random forest (RF) and multilayer perceptron (MLP)). The overarching goal is to gain insight into the structural-hydrothermal gold system and pinpoint potential areas of deep-seated gold deposits for future exploration endeavors. The spatial analysis of ore-controlling faults uncovers a close correlation between gold enrichment and specific fault geometrical attributes, including a dip angle ranging from 20° to 40°, minimal variations in dip angle (less than 5°), and convex topographical features. These attributes likely stem from the influence of fault morphology on the flow and pooling of fluids. In conjunction with this, 3D ore-forming numerical modeling of structural deformation and fluid flow reveals that gold mineralization is intertwined with moderate volumetric strain and shear strain of rock and fluid divergence. This interaction seems particularly pronounced in areas characterized by channel-like or gentle features. Consequently, it is plausible that gold distribution in the Jiaojia region is the outcome of a comprehensive coupling process involving strain localization, rock deformation, fluid flow, heat transfer and/or interaction. The deep gold prospectivity models of RF and MLP for the Jiaojia district jointly using the predictive variables of fault geometry features and ore-forming simulation data (volume strain, shear strain, temperature variation, and fluid flux) exhibit higher AUC (area under the curve) values compared to models employing individual predictor variable datasets. This improvement underscores their enhanced predictive capability. The prospectivity results thus were used for identifying gold potential within the Jiaojia region, where five promising gold targets at depth were ultimately determined.