MACHINE LEARNING ANALYSIS OF TRACE ELEMENT DATA DISTINGUISHES SEDIMENTARY AND HYDROTHERMAL PYRITE

Sedimentary pyrite has long been used as an indicator of marine environmental conditions in Earth history. To capture reliable paleoenvironmental signals, however, we need to first evaluate chemical signals in pyrite as it can be altered and masked by later diagenetic, hydrothermal, and/or metamorphic processes. In this study, we trained two supervised machine learning algorithms on a large LA-ICP-MS pyrite trace element database to distinguish between pyrite of different origins. The analysis validated 12 trace elements (Co, Ni, Cu, Zn, As, Mo, Ag, Sb, Te, Au, Tl, and Pb) as excellent predictors of pyrite origins. Further statistical analysis suggests four trace element clusters behaving differently among sedimentary (syngenetic and early diagenetic), synsedimentary hydrothermal (syngenetic hydrothermal), and post-sedimentary hydrothermal (epigenetic hydrothermal) pyrite, which is probably driven by chemical and physical properties of source fluids, interactions between elements, competition among coprecipitating minerals, and pyrite growth rate. Armed with this initial success, we then demonstrated the efficacy of this approach in identifying the origins of isotopically superheavy pyrite in the Cryogenian Tiesi’ao and Datangpo formations and pyritic rims associated with fossiliferous chert nodules in the Ediacaran Doushantuo Formation of South China. For the superheavy pyrite, the models consistently show high confidence levels in identifying its genesis type, and most samples were given sedimentary origins. The pyritic nodule rims are suggested by the models to be hydrothermally altered early diagenetic pyrite, i.e., with mixed signals. The study highlights the importance of pyrite trace elements in deciphering marine paleoenvironments and the origin of pyrite in sedimentary strata.