Comments on “Testing the Fidelity of Methods Used in Proxy-Based Reconstructions of Past Climate”: The Role of the Standardization Interval*

Abstract

Mann et al. (2005, hereafter M05) conclude that they “find no evidence for the suggestion that real-world proxy-based temperature reconstructions are likely to suffer from any systematic underestimate of lowfrequency variability.” This conclusion is based on multiple pseudoproxy experiments using the National Center for Atmospheric Research (NCAR) Climate System Model (CSM) millennial integration and the climate field reconstruction (CFR) method known as regularized expectation maximization (RegEM; Schneider 2001). RegEM was used by Rutherford et al. (2005, hereafter R05) to reconstruct historical Northern Hemisphere climate from the Mann et al. (1998) proxy network, which prompted the follow-up study by M05 to test, in part, the veracity of the R05 millennial climate reconstruction. We have used the publicly available codes published by R05 and M05 to perform a new suite of pseudoproxy reconstructions with the CSM data. Our findings contradict the M05 conclusion and highlight an important methodological choice that was different from R05, not reported by M05, and has significant impacts on the derived reconstructions. Testing climate reconstruction methods with simulated climates relies on proper application of real-world constraints. For instance, it is important to perturb pseudoproxy networks with realistic noise models such that the noise is representative of actual proxy records. A variety of colored noise models have been adopted (Mann and Rutherford 2002; von Storch et al. 2004, 2006; M05), but these may not fully mimic the nonlinear, multivariate, nonstationary characteristics of noise in many proxy series (e.g., Jacoby and D’Arrigo 1995; Briffa et al. 1998; Esper et al. 2005; Evans et al. 2002; Anchukaitis et al. 2006). Therefore, improving the representation of noise in pseudoproxy networks is an ongoing and important area of research. What is more obvious, however, is that the methodological constraints of real-world climate reconstructions must be preserved in pseudoproxy tests if they are to have any direct applicability to actual reconstructions of historical climate. Using information or techniques that would never be possible in real-world settings sheds little light on climate reconstruction methods. The principal motivation of this comment is to note that M05 used information prior to the period of widespread observational evidence, thereby significantly affecting the outcome of their reconstructions. RegEM requires an input data matrix that is a composite of both instrumental and proxy data. A time-byspace matrix for the instrumental data is first formed in which rows correspond to years in the calibration and reconstruction periods, and columns correspond to grid cells in the instrumental field. For instance, a reconstruction for the Equator–70°N region of the NH on a 5° 5° latitude–longitude grid and spanning A.D. 850– 1980 would fill a matrix of 1131 rows by 1008 columns. This matrix of course would be initially empty, except for the instrumental data in the calibration period (rows 1007–1131 for an 1856–1980 calibration interval). The second part of the composite matrix is formed from the proxy data, composing a matrix of 1131 rows and n columns, where n is the number of proxies (104 in the case of M05). Thus, the instrumental and proxy matrices are concatenated by column and compose the input matrix for the RegEM algorithm (Fig. 1). As is standard with most reconstruction procedures, * Lamont-Doherty Earth Observatory Contribution Number 7077.