Global surface temperatures over the past two millenniaMichael E. MannDepartment of Environmental Sciences, University of Virginia, Charlottesville, Virginia, USAPhilip D. JonesClimatic Research Unit, University of East Anglia, Norwich, UKReceived 24 May 2003; revised 17 June 2003; accepted 1 July 2003; published 14 August 2003.[1] We present reconstructions of Northern and SouthernHemisphere mean surface temperature over the past twomillennia based on high-resolution ‘proxy’ temperaturedata which retain millennial-scale variability. Thesereconstructions indicate that late 20th century warmth isunprecedented for at least roughly the past two millenniafor the Northern Hemisphere. Conclusions for theSouthern Hemisphere and global mean temperature arelimited by the sparseness of available proxy data in theSouthern Hemisphere at present.INDEX TERMS: 1620Global Change: Climate dynamics (3309); 3309 Meteorologyand Atmospheric Dynamics: Climatology (1620); 3344Meteorology and Atmospheric Dynamics: Paleoclimatology;4221 Oceanography: General: Dendrochronology; 4215Oceanography: General: Climate and interannual variability(3309). Citation: Mann, M. E., and P. D. Jones, Globalsurface temperatures over the past two millennia, Geophys.Res. Lett., 30(15), 1820, doi:10.1029/2003GL017814, 2003.1. Introduction[2] It is important to place modern climate warming in alonger-term context. A number of previous studies have thusfocused on climate change of the past few centuries tomillennium, based both on modeling experiments employingestimated climate forcings [e.g., Crowley, 2000; Gerber etal., 2003] and empirical reconstructions based on climateproxy data. The requirements for the latter such reconstruc-tions are (a) high (annual to decadal) resolution and exactdating, so that proxy records may be calibrated againstinstrumental data, (b) reliability of the low-frequency infor-mation so that millennial-scale variability can be faithfullyreconstructed, and (c) adequate spatial distribution of data sothat the regionally-variable nature of past climate variability[e.g., Bradley and Jones, 1993] is appropriately taken intoaccount in reconstructing a large-scale (e.g., hemispheric)mean. It is also important to keep in mind the seasonality ofproxy indicators in forming a climate reconstruction [seeJones et al., 1998; 2001; Mann, 2002; Mann et al., 2003].[3] Several recent studies emphasize spatial reconstruc-tion of climate fields, typically based on multivariate climatefield reconstruction (CFR) techniques [e.g., Mann et al.,1998; 1999; Luterbacher et al., 2002; Evans et al., 2002].Such spatial reconstructions of past surface temperature canbe averaged to yield hemispheric mean temperature est imates[Mann et al., 1998; Briffa et al., 2001]. Other hemisphericmean temperature reconstructions have employed a moreelementary approach, in which various temperature proxyrecords are composited (perhaps after applying some weight-ing factor) and then simply scaled against the avail ableoverlapping instrumental record to yield a hemisphericreconstruction [Bradley and Jones, 1993; Overpeck et al.,1997; Jones et al., 1998; Crowley and Lowery, 2000]. Thesimilarity of these latter reconstructions to those determinedbased on more elaborate approaches [e.g., Folland et al.,2001; Mann, 2002] suggests that the more elementaryapproach can yield a reliable reconstruction. Previous workhas emphasized the Northern Hemisphere, and the past 1000years for which adequate proxy data have been available forhemispheric mean temperature reconstructions. Preliminaryreconstructions of Southern Hemisphere temperatures in pastcenturies have also been attempted [Jones et al., 1998; 2001;Mann et al., 2000].[4] During the past few years, a number of additionallong, high-resolution (annual or decadal) temperature proxyreconstructions (or extensions of previous, shorter records)with reliable millennial-scale variability have been pro-duced. These new records allow an extension of proxy-basedhemispheric mean temperature reconstructions for bothhemispheres. Such extensions are the focus of this study.2. Data[5] For the Northern Hemisphere (NH), we make use oftemperature reconstructions from 8 distinct regions (basedon 23 individual proxy records). Each employs eitherindicators (e.g., sediments and ice cores) with no knownlimitations in resolving millennial-scale variability, or tree-ring records in which various ‘conservative’ standardizationmethods have been used, which, though not all identical,share the common aim of seeking to preserve millennial-scale temperature variability. Since some of t he proxyindicators may contain non-temperature influences, the var-iable reliability of the temperature signal in different proxyestimates is dealt with in our methodology (section 3). Theavailable dataset, though modest in size, samples a range ofseasons, tropical through polar latitudes, both major con-tinents, and marine as well as terrestrial environments in theNorthern Hemisphere (Figure 1). The data set includes:[6] (1) multiproxy composite reconstruction of annualmean temperatures over eastern Asia back to 1 AD based ona composite of 11 historical, lake sediment, tree-ring, andice core proxy records from China, Japan, the TibetanPlateau, and Taiwan [Yang et al., 2002].[7] (2) conservatively standardized tree-ring temperaturereconstruction from Mongolia (annual mean temperaturesback to AD 264 [ D’Arrigo et al., 2001]).GEOPHYSICAL RESEARCH LETTERS, VOL. 30, NO. 15, 1820, doi:10.1029/2003GL017814, 2003Copyright 2003 by the American Geophysical Union.0094-8276/03/2003GL017814$05.00CLM 5--1[8] (3) northern Eurasia tree-ring temperature reconstruc-tion (warm-season temperature back to AD [Briffa andOsborn, 1999]; the record is a composite of 3 ring-widthchronologies processed using Regional Curve Standardiza-tion (RCS) [see Briffa et al., 2001, and references therein]spanning the Eurasian treeline, and includes the twolongest series used by Esper et al., 2002 in their recon-struction of ex tratropical Northern Hemisphere summertemperature).[9] (4) western North American tree-ring temperaturereconstruction, (warm season temperature [Mann et al.,1999]; we employ an extension of the first principalcomponent of the western North American tree-ring databased on 6 ultra-long lived, temperature-sensitive WesternNorth American tree ring records available back to AD200—the resulting series is virtually indistinguishable fromthe corresponding