Crucial Geochemical Signal Identification for Cu-Fertile Magmas in Paleo-Tethyan Arc Based on Machine Learning.

Whole-rock geochemical signals of porphyry intrusions can provide clues for key metallogenetic processes and the potential for porphyry Cu deposits (PCDs). However, it cannot be guaranteed that these signals are the best choices, because they are proposed more based on the individual experience of researchers or some local and limited research data. In this study, it was found that the fertility potential for contemporaneous porphyry intrusions from different locations (fertility field) could be identified according to the composition statistics of the porphyry intrusions in the southern Sanjiang Tethyan Orogen. Therefore, a crucial signal-ranking Monte Carlo-based framework was implemented using the feature importance acquired from a random forest machine learning method. The first six crucial variables among 31 oxides and elements for fertile magma identification were then grouped into three sets using principal component analysis and hierarchical cluster analysis. Several optimal two-dimensional discrimination plots were built based on the k-nearest neighbor classification method according to the three sets of oxides and elements. The results indicate that the fertile magmas have signatures of high Sr, Ba, and P₂O₅ and low heavy rare earth elements (REEs) and SiO₂. These signatures indicate that hydrous, oxidized, and volatile-rich Cu-fertile magmas originated from a thick arc crustal environment and underwent amphibole and garnet fractionation in the deep crust and the subsequent addition of mafic melt in shallow chambers. The application of these discrimination plots in the Paleo-Tethyan shows that the Zhongdian Triassic–Jurassic PCDs that developed in the sub-Paleo-Tethyan subduction zones are special cases. [ABSTRACT FROM AUTHOR]