Machine learning-based garnet thermometer program

Garnet, found in nearly all lithologies of the lithospheric mantle >~60 km depth, can be used to constrain the petrological and thermal structure of the deep lithosphere and its diamond potential, because its composition is sensitive to the temperature of last equilibration. Traditional mineral-based thermometers involving garnet either require the compositions of coexisting minerals or costly trace element analysis for Ni. Here, we developed a machine learning (ML)-based garnet thermometer that utilises the major and minor element compositions of garnets alone, including SiO₂, TiO₂, Al₂O₃, Cr₂O₃, FeO, MgO, MnO, CaO, and Na₂O. These elements can be accurately measured by electron probe microanalysis (EPMA). To construct a predictive model for garnet temperatures, we applied XGBoost (Extreme Gradient Boosting), a widely used, powerful, and scalable ML algorithm. This program provides a valuable tool for researchers studying mantle thermal structures and diamond exploration.