Parker Robinson Liautaud , Peter Huybers

A number of groups attempted to predict atmospheric CO2 concentrations between 420 to 800 ka prior to publication of the Dome C ice-core record by the European Project for Ice Coring in Antarctica, EPICA. The predictions that fared best assumed that the relationships between CO2 and proxies of air temperature remained consistent over the past 800 ky [7]. Here we extend predictions of atmospheric CO2 concentrations over the last 2 Ma under a similar assumption of consistent physical relationships between CO2 and climate over time and test this assumption against existing observations. Our principal approach is to use a recently-developed Bayesian paleoclimate model to infer CO2 values conditional on past sea level. An ensemble of seven different CO2 histories are inferred from an equal number of sea-level reconstructions. Five of the ensemble members give a consensus prediction that CO2 in the early Pleistocene, 2-0.8 Ma, averaged 241 ppm (238 ppm - 245 ppm 95% c.i.) with 95% of CO2 values within 206 ppm and 275 ppm. Uncertainty estimates account for contributions from orbital forcing, the ice-albedo feedback, age uncertainties, and other factors. The other two ensemble members indicate 20-50 meter higher sea level during the early Pleistocene and imply much higher CO2 levels. Our consensus prediction aligns well with a compilation of previously published δ11B-based CO2 reconstructions that, after calibration to late-Pleistocene ice-core CO2 values, average 237 ppm (95% of CO2 values within 195 ppm to 273 ppm). Furthermore, 94% of consensus CO2 predictions fall within the range indicated by 60 early-Pleistocene CO2 measurements from air trapped in discontinuous ice segments from the Allan Hills in East Antarctica. Our consensus prediction can be definitively tested by obtaining continuous ice-core atmospheric CO2 records that extend into the early Pleistocene.