Identification of a deposit type in an early stage of exploration can save money and time. However, despite several studies, there is a lack of simple deposit classifiers based exclusively on the major and trace element chemistry of minerals, which is relatively easier to measure. Furthermore, the identification of fluid source is crucial for any ore-genetic study, as it exerts a first order control on physicochemical characteristics of fluid, which ultimately control the alteration mineralogy, metal solubility and mobilization, and metal association in a deposit. The stable isotope composition (e.g., δ34S, δ13C, δ18O, δ11B, etc.) of rocks and minerals is conventionally used to constrain the source of ore-forming fluids, however, the potential of mineral chemistry for constraining fluid source is not well-explored. In this study, we compile the major element composition (wt% of oxides) of a large number (n = 1932) of hydrothermal biotite from various ore settings and execute multivariate statistical analysis and machine learning-based predictive modeling to explore the potential of using biotite chemistry as an indicator of deposit type and fluid sources. The atoms per formula units (i.e., apfu values), and calculated parameters (e.g., Xphl, chlorine intercept value, log(fH2O/fHCl)fluid) is used to perform principle component analysis (PCA) and partial least square discriminant analysis (PLS-DA). The results show that among the different ore settings, iron-oxide copper‑gold, Archean gold, porphyry, and magmatic-hydrothermal rare earth metals deposits and among the fluid sources, (meta-)evaporitic, magmatic-hydrothermal, and metamorphic fluid sources can be reliably discriminated using biotite chemistry. We also show that skarn, hydrothermal platinum group elements, and carbonatite-related rare earth elements deposits may not be discriminated using biotite chemistry. Similarly, mafic rock-derived, carbonatite-derived, and meteoric fluid sources may not be discriminated by biotite chemistry. We propose multiple binary and ternary classifier diagrams, which can be used as discriminator of deposit types and fluid sources. Furthermore, we have developed four different PLS-DA predictive modeling, two for deposit types and two for fluid sources, which can predict deposit types and fluid sources with good accuracy. We provide a Windows© based program that can be used to discriminate deposit types using our PLS-DA predictive modeling. The binary and ternary classifiers and the Windows© programs are tested with recently published biotite data, and it has been shown that these classifiers and programs can efficiently separate deposit types and fluid sources. We also discuss about the limitations of the classifiers.
Read full abstract