Climate models as a test bed for climate reconstruction methods: pseudoproxy experiments

(a) Comparison of two pseudoproxy networks that approximate the most populated nests in the multiproxy networks of Mann et al.72 (MBH98, red dots) and Mann et al.73 (gray squares). Bottom plots show the total and categorized proxy abundances in the global multiproxy network of Mann et al.73 for: (b) the full dataset; and (c) the culled dataset as screened and used by Mann et al.73 The plotted abundances are as published in the Supporting Information of Mann et al.73

A ‘signal’ time series sampled from the annual surface temperature field of the ECHO‐G ERIK2 millennial simulation35 after interpolating to an even 5° latitude–longitude grid (the geographic center of the sampled grid is 57.5°N, 2.5°E). The time series has been normalized to have a standard deviation of one. Shown on the left below the sampled temperature signal is a single white noise time series that has been scaled to have variances of 1, 4, and 16. The sum of the pure‐noise and temperature time series is shown in the lower right panels and represents the pseudoproxy time series used in pseudoproxy experiments (PPEs). Signal‐to‐noise ratios (SNRs) and correlations between the original signal and the signal‐plus‐noise series are also shown in the lower right‐hand plots.

Index and climate field reconstruction (CFR; bolded) reconstructions of NH temperature variations during the Common Era (CE) using multiple climate‐proxy networks and methods; the HadCRUTv2 instrumental temperature record82 is shown in black. All series have been smoothed using a Gaussian‐weighted filter to remove fluctuations on time scales less than 30 years. All temperatures represent anomalies from their 1961 to 1990 mean83–95. Reconstructions derived from CFRs have been bolded by the author. (Reprinted with permission from Figure 6.10(b) in Ref 21. Copyright 2007 Cambridge University Press).

Pseudoproxy results derived from an experiment using the NCAR CCSM1.4 millennial simulation, a pseudoproxy sampling scheme approximating the most populated nest in the multiproxy network of Mann et al.72 (MBH98), an signal‐to‐noise ratio (SNR) of 0.5, and a representative instrumental temperature mask.59 Both pseudoproxies and the instrumental temperatures were sampled from the annual surface temperature field of the CCSM1.4 simulation, after properly interpolating to a 5° latitude–longitude grid.57 Results are derived from the hybrid RegEM truncated total least squares method using a calibration interval from 1856 to 1980 Common Era (CE).59 Panel (a) compares the low‐pass filtered NH mean temperature anomalies (relative to the calibration interval) derived from the applied climate field reconstruction (CFR) method and the known model mean. Panel (b) plots the local correlation coefficients computed between the reconstruction and known model field during the reconstruction interval (850–1855 CE). Panel (c) is the same as in (b), but for the difference between the reconstructed and known model means, i.e., the reconstruction bias.