Fig. 3. Global and basin-averaged sampling error compared with reconstructed temperature change.

(A to G) The sampling errors for different truth fields subsampled by historical observation locations are shown as green dots, with the green solid line and the error bar for the mean and ±2σ, respectively. Blue stars denote the 16 different truth fields. The gray line is the reconstructed monthly temperature anomaly time series from 1960 to 2015 with ±2σ interval in gray shading, and the dark line is the time series after a 7-year low-pass filter (see Materials and Methods). The orange curve is the NCEI result, along with a 1-SE bar (dashed orange). (A) Global, (B) tropical/subtropical Pacific, (C) North Pacific, (D) Indian, (E) tropical/subtropical Atlantic, (F) North Atlantic, and (G) southern oceans. (H and I) S/N ratio of the temporal variability of our reconstruction on (H) decadal scales (solid lines) and (I) interannual scales (triangle lines) compared to the sampling error.