MAGLEARN: A Global Calibration Framework for Mg/Ca Paleothermometry of Planktic Foraminifera Based on Machine Learning

Abstract Mg/Ca paleothermometry of planktic foraminifera is one of the most important tools in quantitative paleoclimatology. Mg/Ca paleothermometry relies on the robustness of the calibration that maps shell Mg/Ca values to calcification temperatures. While the existing calibration schemes offer reasonable accuracy toward different paleoenvironmental applications, their generalization to a wide environmental range and geographically distinct regions could still be improved. This can be achieved by adequately representing the complexity of multivariate factors that govern shell Mg/Ca and by employing independent validation (out‐of‐sample testing). Here we present a unified framework—MAGLEARN for Mg/Ca paleothermometry of commonly used planktic foraminifera species ( Globigerinoides ruber , Trilobatus sacculifer , Neogloboquadrina pachyderma , Neogloboquadrina incompta , and Globigerina bulloides ). The framework is rigorously validated on independent data and incorporates secondary non‐thermal influences, including seawater carbonate chemistry, salinity, post‐depositional alteration, and shell cleaning methodology. Both forward and inverse versions have been developed that show improvements over the conventional forms. MAGLEARN adds to the diversity of existing calibration frameworks that are essential for quantitative reconstructions of paleoenvironmental conditions.