Cover Image Credits

Cover Photo: Reconstructed mean annual precipitation (MAP) from Late Triassic paleosols at Petrified Forest National Park. Three proxies are shown, based on the chemical index of alteration (CIA-K; Sheldon and others, 2002, Journal of Geology), Paleosol-Paleoclimate Model version 1.0 (PPM_1.0, Stinchcomb and others, 2016, American Journal of Science), and the new random forest model for mean annual precipitation version 1.0 (RF-MAP_1.0). The RF-MAP_1.0 results suggest that semi-arid phases began far earlier than previous reconstructions indicated, and that the middle Norian climate shift was marked by the introduction of semi-arid to arid pulses rather than a secular drying trend. White line shows the MAP reconstruction of Nordt and others (2015, GSA Bulletin).

Paper summary: In this issue, Lukens and others (2019, p. 819–845) apply machine-learning techniques to develop two new proxies for rainfall based on paleosol major and minor elemental geochemistry. These random forest models for mean annual precipitation (RF-MAP versions 1.0 and 2.0) were built using data science best practices, are widely applicable, and reduce prediction error compared to other recent efforts. Lukens and others (2019) argue for a transition to a more robust modeling framework on diverse soil data sets for future paleosol proxy development.