A DATA-DRIVEN EMPIRICAL MODEL THAT PREDICTS PALEOCLIMATE FOR A DIVERSE RANGE OF CONDITIONS FOR PALEOSOLS

Here we evaluate a data-driven empirical model that uses subsoil geochemistry from wide-ranging soil forming environments to predict Mean Annual Precipitation (MAP) and Mean Annual Temperature (MAT) as a joint response with few initial assumptions. The paleosol-paleoclimate model, PPM_1.0, was developed using a combined partial least squares regression (PLSR) and thin-plate spline (TPSPLINE) approach on 685 mineral soil B horizons currently forming under MAP ranging from 130-6900 mm and MAT ranging from 0-27 ºC. The PLSR results on 11 major and minor oxides (Al₂O₃, ZrO₂, TiO₂, Fe₂O₃, P₂O₅, MnO, CaO, MgO, Na₂O, K₂O, and SiO₂) show that four linear combinations of these oxides (Regressors 1-4) have potential for predicting climate. Regressors 1-4 are related to fundamental controls on weathering including, ionic potential and silicate and carbonate mineral weathering, which are driven by water flux and temperature-dependent dissolution. The TPSPLINE model fit on Regressors 1-4 results in a RMSE_MAP of 228 mm and RMSE_MAT of 2.46 ºC. These RMSE values are lower than some preexisting MAT models. The PPM_1.0 results show that subsoil weathering processes operating under a wide-range of soil forming factors possess climate prediction potential, which agrees with the state-factor model of soil formation. We argue that although the PPM_1.0 does not account for effects from diagenesis, it should always be used initially to predict MAT and even MAP in deep time where few paleosol property and assumptions can be made a priori. The PPM_1.0 approach may also yield less misapplication because it covers a broader range of MAP than previous geochemical-based models