ABSTRACT
The spread of C4 grasses in the late Neogene is one of the most important ecological transitions of the Cenozoic, but the primary driver of this global expansion is widely debated. We use the stable carbon isotopic composition (δ(13)C) of bison and mammoth tissues as a proxy for the relative abundance of C3 and C4 vegetation in their grazing habitat to determine climatic and atmospheric CO2 controls on C4 grass distributions from the Last Glacial Maximum (LGM) to the present. We predict the spatial variability of grass δ(13)C in North America using a mean of three different methods of classification and regression tree (CART) machine learning techniques and nine climatic variables. We show that growing season precipitation and temperature are the strongest predictors of all single climate variables. We apply this CART analysis to high-resolution gridded climate data and Coupled Model Intercomparison Project (CMIP5) mean paleoclimate model outputs to produce predictive isotope landscape models ("isoscapes") for the current, mid-Holocene, and LGM average δ(13)C of grass-dominated areas across North America. From the LGM to the present, C4 grass abundances substantially increased in the Great Plains despite concurrent increases in atmospheric CO2. These results suggest that changes in growing season precipitation rather than atmospheric CO2 were critically important in the Neogene expansion of C4 grasses.