Trace element variations of pyrite in orogenic gold deposits: Constraints from big data and machine learning

Abstract

Orogenic gold deposits represent key global gold resources, yet its internal differences are poorly constrained. Pyrite is ubiquitous in orogenic gold ores, and its trace element geochemistry has significant implications for the ore-forming processes. However, the diffusion regularity and controlling factors of pyrite trace element geochemistry of orogenic gold deposits lack effective constraints. To address these issues, pyrite trace element geochemical data (4092 sets of data) from 67 orogenic gold deposits were compiled for big- data and machine learning analyses. The principal component analysis (PCA) divides the pyrite data into four groups according to trace element correlations: I) Au-As; II) Sb-Cu; III) Ag-Pb-Bi; and IV) Co-Ni-Te-Se-Mo. This division is likely ascribed to the element occurrence in pyrite. Partial least squares-discriminant analysis (PLS-DA) and Random Forests classification were used to identify pyrite trace elemental variations among orogenic deposits of different mineralization ages, wallrocks, and tectono-metallogenic settings. The results show that the influence on pyrite geochemistry from temporal variation is not obvious, and only Triassic and Jurassic orogenic-type auriferous pyrite shows enrichments of Cu, Pb, and Sb, which may reflect the multiple mineralization events in these deposits as Pb commonly occurs as cation in the fluids, while Sb usually occurs in the melt. The spatial difference is more distinct with less overlapping of confidence circles in PLS-DA and higher accuracy in Random Forests, which may have been associated with fluid source variation influenced by the geodynamic setting (mantle model, supposed by distinctive geodynamic variation from metallogenic belts in China) and source strata (metamorphic model, supposed by Archean deposits mostly hosted in green-schist phase besides China). Pyrite in orogenic gold ores hosted in different wallrocks can be differentiated by the different As, Pb, and Cu contents, as the contents of these elements vary widely in the wallrocks and participate in pyrite lattice through fluid-rock reactions. Random Forests regression reveals that pyrite As-Sb contents in pyrite show distinct negative correlations with ore-forming temperature, but Se content shows a non-linear influence. This suggests that the pyrite As-Sb-Se contents in orogenic gold ores can be used as a geothermometer. Furthermore, we suggest that the pyrite Co-Ni contents are not good indicators for the ore-forming temperature, as the low salinity fluid in orogenic gold systems makes Co-Ni difficult to form chloride complexes which is sensitive to temperature variation.