Fig. 2: Climate anomalies in the reconstructions and model simulations.

Comparison of reconstructed and simulated anomalies is shown for (a) annual water balance and (b) July mean temperature (Tjul) for the Midwest, Great Lakes and Northeast regions. Reconstructions from fossil pollen sequences are expressed as the mean of all reconstructions interpolated at 50-year time step (dotted line) and a five-point running mean (solid line). The uncertainty bands for the reconstructions represent 95% errors of the ensemble mean calculated using 1000 bootstrap samples of all datasets. Three transient climate simulations are shown, including TraCE-21ka (past 12 ka), MPI-ESM (past 8 ka), and EC-Earth (past 8 ka), with dotted lines indicating the means of annual values calculated for 50-year bins and the solid lines the five-point averages of the bin means. The 95% error bands for the transient simulations were calculated as ± 2σ of the values in each 50-year bin divided by the square root of the sample size (50). Squares indicate anomalies from CESM1 equilibrium simulations for 12, 11, and 9 ka. The anomalies are expressed relative to the preindustrial period (0.25–0.75 ka) for the reconstructions and the transient simulations, and relative to a preindustrial control run (Supplementary Table 2) for CESM1. The SiZer maps (lower panels) show the significant features of the BRT-based reconstructions when smoothed at a range of bandwidths, with the bandwidth used at each point on the vertical axis indicated by the horizontal distance between the white lines. For each point in time and each bandwidth (h), red indicates a significant rising trend, blue a significant falling trend, purple a lack of a significant trend, and gray a lack of sufficient data for meaningful inference. For reconstructions from the individual fossil data sites, see Supplementary Figs. 1–10.