Explainable machine learning reveals apatite fertility and porphyry copper mineralization processes in the syn- and post-subduction settings

Abstract

Apatite, as a common accessory mineral in igneous rocks, is important in evaluating the mineralization potential of porphyry Cu deposits (PCDs). Although apatite has been widely studied, its potential for recording differences in the ore-forming processes of PCDs between syn- and post-subduction settings has not been fully explored. In this study, we collected global chemical compositions of apatite from major orogenic belts and used a Random Forest (RF) model to accurately classify apatite from ore-barren and ore-bearing igneous rocks. The model exhibited exceptional classification performance, with all class accuracies exceeding 90% on the test set. To further analyze the model’s performance, the information gain and interpretability properties of Shapley Additive Explanations (SHAP) were used to assess the importance of various geochemical features. This study underscores the critical roles of Cl and Sr/Y in discriminating porphyry Cu-related apatite in subduction settings from barren ones, La/Yb and heavy rare earth elements (HREE) in differentiating unfertile from fertile apatite in post-subduction PCDs, and the significance of HREEs, F, and Eu in identifying subduction-related fertile apatite from its post-subduction counterparts. These results may reflect the garnet fractional crystallization and the fluid exsolution process that predate apatite crystallization under post-subduction conditions. These insights not only enhance our understanding of the model’s reliance on specific elemental features to discriminate between different types of apatite, but also reveal differences in mineralization mechanisms between syn- and post-subduction PCDs. Therefore, apatite in general shows significant potential for applications in mineral exploration and tectonic setting discrimination.