Evaluation of the Mg-in-magnetite geothermometer and development of new magnetite geothermometers using machine learning

Abstract Determining the formation temperatures of rocks and minerals is essential in understanding geological processes. Recently developed Mg-in-magnetite geothermometer has been used widely to estimate the temperature of magmatic and hydrothermal systems accurately. However, the applicability of this approach to natural samples has yet to be evaluated fully. We collected 4831 LA–ICP–MS magnetite analyses from various ore deposits, including Magmatic, Fe oxide–apatite (IOA), Fe oxide–Cu–Au (IOCG), porphyry, and skarn deposits. Our results suggest that the Mg geothermometer may not be applicable to magnetite in some of these systems. The geothermometer is typically effective in determining mineralization temperatures in magmatic Fe–Ti–V deposits and in high-temperature hydrothermal IOA, IOCG, and porphyry deposits; however, there are discrepancies when they are applied to skarn deposits, where temperatures obtained using the Mg-in-magnetite geothermometer (800–1000 °C) are much higher than corresponding fluid-inclusion homogenization temperatures (200–500 °C). To address this problem, we used a machine learning Random Forest regression model to estimate temperature based on various elements, including Ti, V, Mg, Mn, Co, Si, Al, and Zn contents. We introduce an empirical geothermometer based on the Ti and V contents of magnetite. The temperature estimated by this empirical geothermometer is in good agreement with the homogenization temperatures of fluid inclusions in hydrothermal magnetite, as well as with the temperature of magmatic magnetite derived from other independent geothermometers. The result demonstrates the robustness and applicability of the new geothermometer in various geological environments.