Improving paleoclimate predictions from paleosol geochemistry

Paleosol climofunction-based methods for paleoclimate reconstruction are an important component of deep-time archives, but current proxies are subject to major gaps in terms of modern dataset quality and the suite of predicted climate parameters. Here we compile an expanded modern soil dataset and add soils from unique climate zones, pair these with updated high-resolution climate datasets, and apply novel partial-least squares and principle components methods to develop an improved predictive soil geochemistry paleoclimate model. This model provides more accurate prediction of common proxy response variables (temperature, precipitation) with lower root mean square predictive error than previous principle components-based methods, and introduces new proxy response variables associated with seasonality (range of temperature, growing season precipitation). We apply the model to key Cretaceous sites from the Western Interior Basin (Two Medicine and Kaiparowits formations) and show that the model predicts comparable mean annual temperature (~13℃) and precipitation (~600 mmyr-1) to other proxies, while also uniquely predicting mean annual range of temperature (~26℃) and growing season precipitation (~450 mmyr-1) that match regional climate model projections. Our model is also broadly available for new paleoclimate studies using open-source coding.