Introduction

Since the late 1960s when the first long cores were recovered from Greenland and Antarctica, ice cores have provided unique details about the nature of the Earth’s climate system with emphasis on glacial–interglacial cycles. Recently, greater attention has been focused on reconstructing high-resolution (optimally annual) proxy histories for the Holocene, with emphasis on the last few thousand years. This attention results, in part, from the need to assess the well-documented 20th-century warming of the Earth’s surface temperatures from a longer-term perspective (Reference Mann, Bradley and HughesMann and others, 1999; Reference Levitus, Antonov, Boyer and StephensLevitus and others, 2000; Reference HansenHansen and others, 2001; Reference Jones and MobergJones and Moberg, 2003; Reference Moberg, Sonechkin, Holmgren, Datsenko and KarlénMoberg and others, 2005). Restricted to very cold and/or high locations, ice-core-derived proxy histories provide a partial picture of past climatic and environmental conditions. These frozen archives are powerful contributors to multi-proxy reconstructions, providing multiple lines of evidence (e.g. changes in atmospheric dustiness and chemistry, in mass accumulation and in temperature) from remote locations where other proxy data are unavailable.

Over the last few decades, the Ice Core Paleoclimate Research Group at The Ohio State University (OSU) has collected a suite of tropical and subtropical cores from the South American Andes and the Tibetan Plateau, as well as polar cores from Antarctica, Greenland and Franz Josef Land (FJL), Russian Arctic. The analyses of these ice cores have emphasized the highest time resolution possible. More recently, a number of new annually resolved ice-core records from Greenland have been added to the OSU collection. The cores were acquired between 1995 and 1999 as part of the NASA–NSF (US National Science Foundation) PARCA (Program for Arctic Regional Climate Assessment) project (Reference Thomas and InvestigatorsThomas and others, 2001). These cores are particularly valuable as they come from sites that are widely distributed along the western side of Greenland. Previously, most available high-resolution Greenland ice-core histories that extend back more than a century were drilled in central and north-central Greenland or in southern Greenland near the Dye 2 and Dye 3 radar stations (now abandoned).

Ice-Core-Derived δ¹⁸O Records

Figure 1 presents the decadally averaged δ¹⁸O records organized geographically (north to south), along with regional maps showing their locations. The records from Tibet, the Andes and Antarctica have been published separately, but not collectively with the newer δ¹⁸O histories from Greenland and FJL. This synthesis is limited to the last millennium, as many of the records are relatively short, extending back just a few centuries to one half-millennium. Six of the δ¹⁸O records extend to earlier millennia, four into the Last Glacial Stage. The longer records include those from the Dunde and Guliya ice cores from the Tibetan Plateau (Reference ThompsonThompson and others, 1989, Reference Thompson1997, respectively), the Quelccaya and Huascarán cores from Peru (Reference Thompson, Mosley-Thompson, Dansgaard and GrootesThompson and others, 1986, Reference Thompson1995, respectively), the Sajama ice cap in Bolivia (Reference ThompsonThompson and others, 1998) and the Plateau Remote site from East Antarctica (Reference Mosley-Thompson, Jones, Bradley and JouzelMosley-Thompson, 1996). Table 1 provides basic information about each core. Details of the dating procedures are in the references.

Fig. 1. Decadally averaged δ¹⁸O histories are shown from north to south (top to bottom). Horizontal bars in each record show the average δ¹⁸O values for the pre- and post-1900 time intervals. The 1000 year Northern Hemisphere reconstruction is from Reference Mann, Bradley and HughesMann and others (1999) and updated by Reference Mann and JonesMann and Jones (2003). Superimposed is the observed near-surface temperature record (Reference Jones and MobergJones and Moberg, 2003). Maps show locations for the cores.

Table 1. Information for all ice cores discussed in the text Timescale under revision.

The δ¹⁸O histories from two different drilling projects at the Summit Site in central Greenland are included for comparison with the newer Greenland records (Fig. 1). The Greenland Ice Sheet Project 2 (GISP2) record (Reference Grootes, Stuiver, White, Johnsen and JouzelGrootes and others, 1993) is shown along with a composite from two cores, called Site T cores, drilled 4 km apart at the Summit in 1989 by OSU (Reference Mosley-Thompson, Thompson, Dai, Davis and LinMosley-Thompson and others, 1993). The δ¹⁸O history from the Windy Dome Ice Field on Graham Bell Island, FJL, (Reference HendersonHenderson, 2002) is included along with the five new Greenland δ¹⁸O histories: GITS, NASA-U, D2, D3, Raven (Fig. 1), collected by the PARCA initiative. With the exception of the GISP2 core, all records discussed here were analyzed and dated by the authors using consistent procedures. Moreover, unlike many of the earlier Greenland records, the timescales for the PARCA cores were constructed using three seasonally varying parameters: the insoluble dust concentrations and oxygen isotopic ratios (δ¹⁸O), both measured at OSU, and hydrogen peroxide (H₂O₂) measured at the University of Arizona (Reference McConnellMcConnell and others, 2001). The timescales were further confirmed with beta radioactivity horizons from thermonuclear bomb tests and by identification of known volcanic eruptions (Reference Mosley-ThompsonMosley-Thompson and others, 2001, Reference Thompson, Mosley-Thompson, Davis, Lin, Henderson and Mashiotta2003).

To facilitate their comparison, each record has been normalized with respect to its entire length (for cores shorter than one millennium) or with respect to the last millennium for the longer cores. The annual Z score is the standardized deviation from its respective mean (annual Z score = (annual value-record mean)/record standard deviation). The annual Z scores are shown (Fig. 1) as unweighted 10 year (decadal) averages plotted at the midpoint of the decade. The goal here is to examine a global (pole-to-pole) array of ice-core-derived (δ¹⁸O) proxy temperature histories to assess 20th-century climate changes within a longer-term perspective. Thus, the averages for the pre-1901 period (1900 to the oldest year in the record or AD 1000) are compared with those from the post-1900 part of the record (1901 to the most recent year in the record). It would be ideal if all the records were of equivalent temporal length; however, the time, expense and logistics required to collect ice cores dictates the depth to which cores are recovered at a given site.

Before comparing the δ¹⁸O records and drawing general conclusions, the advantages and limitations of using δ¹⁸O as a proxy for regional air temperature are reviewed briefly. With respect to ice cores, the δ¹⁸O–air-temperature (T ^a) relationship has been most extensively investigated in Greenland and Antarctica. In Antarctica, a strong linear relationship between δ¹⁸O and T_a has been demonstrated (Reference Aldaz and DeutschAldaz and Deutsch, 1967; Reference Mosley-Thompson, Bradley and JonesMosley-Thompson, 1992; Reference Peel, Bradley and JonesPeel, 1992; Reference JouzelJouzel, 1999), with slope (a) values ranging from 0.76 to 0.92%˚C⁻¹. In Greenland the slope of the linear relationship ranges from 0.62 to 0.67%˚C⁻¹ (Reference Dansgaard, Johnsen, Clausen and GundestrupDansgaard and others 1973; Reference Johnsen, Dansgaard and WhiteJohnsen and others, 1989). More recently, using central Greenland borehole temperatures, Reference Cuffey, Clow, Alley, Stuiver, Waddington and SaltusCuffey and others (1995) calibrated the paleothermometer, δ¹⁸O = αT_a + β. They concluded that δ¹⁸O provides a ‘faithful’ proxy for long-term average temperature at that site (GISP2); however, they caution that it is inappropriate to use a single set of constants (α and β) to infer past climate changes, as they depend upon factors that change with time. Thus, in keeping with the general practice in ice-core paleothermometry (Reference JouzelJouzel and others, 1997; Reference JouzelJouzel, 1999), we adopt the convention for the polar cores such that more negative δ¹⁸O values reflect cooler air temperatures and less negative δ¹⁸O values reflect warmer air temperatures at the time of condensation.

There has been much less investigation of the δ¹⁸O–T _a relationship in non-polar ice cores, in part because records have been limited. Reference Rozanski, Araguás-Araguás and GonfiantiniRozanski and others (1992, Reference Rozanski, Araguás-Araguás, Gonfiantini, Swart, Lohmann, McKenzie and Savin1993) reviewed an extensive set of in situ data collected over three decades, mainly from mid-latitude, lower-elevation sites. They suggested that as a first approximation, temperature controls the isotopic composition of precipitation at high and mid-latitudes while the amount of precipitation (the amount effect) controls the isotopic composition in tropical regions (Reference Rozanski, Johnsen, Schotterer and ThompsonRozanski and others, 1997). They also concluded that the relationship between longer-term changes in δ¹⁸O and T _a for a given location is more appropriate for paleoclimate investigations (vs short-term applications).

Additional studies have addressed the roles of air temperature and the precipitation ‘amount effect’ in determining the δ¹⁸O values in precipitation (snowfall) from a regional perspective. Reference Yao, Thompson, Mosley-Thompson, Zhihong, Xingping and LinYao and others (1996) conducted the first study of the δ¹⁸O–T _a relationship in discrete precipitation events at three sites on the Tibetan Plateau. They reported that averaging δ¹⁸O and air temperatures for discrete precipitation events over longer periods (months to years) removed the synoptic effects and pointed to air temperature as the dominant control on longer-term δ¹⁸O variations at individual sites.

Reference Thompson, Mosley-Thompson and HendersonThompson and others (2000) address the conundrum with Andean precipitation whereby the seasonal δ¹⁸O–T _a relationship is opposite that in the polar regions. In the Andes, summer snow has more negative δ¹⁸O values than winter snow, and the seasonal δ¹⁸O range can be large, up to 20%, although the seasonal temperature range is small. Reference Grootes, Stuiver, Thompson and Mosley-ThompsonGrootes and others (1989) used a three-step model, employing Rayleigh fractionation in each step, to explain the δ¹⁸O values measured on the Quelccaya ice cap. They concluded that air circulation and air mass stability, rather than temperature, determine the seasonal δ¹⁸O cycle over Quelccaya. They did not address the longer-term relationship between δ¹⁸O and T_a in the tropics that appears consistent with that in the higher latitudes (more negative δ¹⁸O implies colder T _a at condensation). Reference Thompson, Mosley-Thompson and HendersonThompson and others (2000) suggest that seasonal changes in the cycle of deep tropical convection modulate the height of the mean condensation level (hence temperature of condensation). Deep summer convection produces precipitation that is more isotopically depleted than winter precipitation that condenses at lower levels in the atmosphere. Reference Henderson, Thompson and LinHenderson and others (1999) and Reference VuilleVuille and others (2003) have investigated the controls on recent snowfall over Sajama, Quelccaya and Huascaran. Reference Henderson, Thompson and LinHenderson and others (1999) note that the spatial distribution of temperature anomalies in the western tropical Atlantic influences atmospheric circulation at 500 hPa and thereby isotopic fractionation over the Amazon Basin, the primary source of precipitation for Huascaran and Quelccaya. Reference VuilleVuille and others (2003) note that their modeled δ¹⁸O values depend strongly on precipitation amount, but conclude that the δ¹⁸O signal in precipitation is influenced by a combination of mechanisms (precipitation amount, temperature, variability of the moisture source and changes in atmospheric circulation). Reference Bradley, Vuille, Hardy and ThompsonBradley and others (2003) report strong linkages between sea surface temperatures across the equatorial Pacific Ocean and δ¹⁸O in ice cores from the tropical Andes as well as the Himalaya (Dasuopu glacier). They, like Reference Henderson, Thompson and LinHenderson and others (1999), report a strong link between δ¹⁸O and El Nino-Southern Oscillation (ENSO) variability, which is not surprising as ENSO integrates all the mechanisms mentioned above. The δ¹⁸O– T _a relationship on longer timescales (decades to centuries) is of greater importance here. Reference Thompson, Mosley-Thompson and HendersonThompson and others (2000, Reference Thompson, Mosley-Thompson, Davis, Lin, Henderson and Mashiotta2003) present evidence that on longer timescales, δ¹⁸O variations in Andean precipitation more strongly reflect variations in air temperature than precipitation. Nevertheless, this issue is ripe for further investigation, as the dearth of records and their short duration present challenges to fully addressing the controls on the δ¹⁸O– T _a relationship at high-elevation sites in the low latitudes. Here the trends in δ¹⁸O, when averaged over decadal and longer timescales, are assumed to reflect primarily trends in air temperature rather than in precipitation.

Discussion

Placing the well-documented 20th-century warming of the Earth’s globally averaged temperature within a temporal perspective longer than existing meteorological (Reference Jones and MobergJones and Moberg, 2003) and oceanographic (Reference Levitus, Antonov, Boyer and StephensLevitus and others, 2000) observations (∼1860 to the present) requires proxy records such as ice-core-derived δ¹⁸O histories. Figure 1 reveals that the averaged δ¹⁸O for post-1900 precipitation is enriched (to differing degrees) relative to that for pre-1901 precipitation in all cores except those from Siple Station, West Antarctica, and Plateau Remote, East Antarctica. The observed regional trends in temperatures and the associated ice-core-derived δ¹⁸O records are discussed below. If the 20th-century δ¹⁸O trends among the ice cores are broadly consistent with the contemporaneous temperature trends for their respective regions, then the earlier portions (pre-20th-century) of the δ¹⁸O records should also reflect regional temperature trends and provide a viable longer-term perspective for the recent warming.

Large regional differences exist among the Greenland δ¹⁸O histories, even between the GISP2 and Summit 1989 records that are ∼ 4 km apart. The 20th-century enrichment appears larger in Summit 1989 than in GISP2, but this does not result from the different record lengths (∼500 vs ∼1000years, respectively). To facilitate comparison, the average δ¹⁸O value was calculated for the GISP2 segment that overlaps the Summit 1989 record, and, as Figure 1 reveals, the 500 year δ¹⁸O average is only slightly less than that for the 1000 year segment. Thus, the difference cannot be resolved with data in hand, and likely reflects some combination of differences in dating approaches, sampling schema and real spatial differences among the cores.

The 5 year unweighted running means of the annual δ¹⁸O values (Fig. 2) highlight the regional variability that exists over Greenland on annual to decadal timescales. This variability arises from (1) differences in regional climate forcing, (2) more localized conditions (e.g. an isotopic record is only produced when snow is falling), and (3) post-depositional modification by deflation, drifting and redeposition of surface snow. Despite this regional variability, the coherency of the Greenland δ¹⁸O records (Fig. 2) provides strong evidence that δ¹⁸O of the precipitation (snow) does reflect large-scale, multi-year variations in air temperature. The vertical bars in Figure 2 highlight three periods of marked isotopic depletion (cooling) that are widely recorded along the west side of the ice sheet from the northwest (GITS) to the southwest (Raven). The two closely spaced shaded bars highlight two decades of well-documented volcanically induced cooling: (1) 1810−20 with the eruptions of Unknown in 1809 and Tambora, Indonesia, in 1815 (Reference Dai, Mosley-Thompson and ThompsonDai and others, 1991; Reference Thompson, Mosley-Thompson, Davis, Lin, Henderson and MashiottaMosley-Thompson and others, 2003) and (2) 1830−40 with the eruptions of an unidentified volcano in ∼1832 and Coseguina, Nicaragua, in 1835 (Reference Cole-Dai, Mosley-Thompson and ThompsonCole-Dai and others, 1997). The third bar highlights a brief, but large and spatially coherent, cooling from 1917 to 1919 that is documented in Arctic temperature composites (see e.g. Reference Jones and MobergJones and Moberg, 2003, fig. 2).

Fig. 2. The 5 year running mean of the annual δ¹⁸O histories from five new Greenland ice cores are shown along with the 1989 Site T record collected at the Summit (GISP2) site. Vertical shaded bars highlight specific time intervals discussed in the text. Horizontal bars in each record show the average δ¹⁸O value before and after the large cooling in 1917–19 that was followed by a rapid, widespread warming over much of the Arctic.

This brief cold event preceded an abrupt warming in the high Arctic (60−90˚ N) from ∼1920 to ∼1940 (Reference RogersRogers, 1985). The warming was strongest in the Kara and Barents Sea region and only slightly weaker in western Baffin Bay. It is hypothesized to have been linked to the retreat of sea ice in response to enhanced westerly flow (Reference Bengtsson, Semenov and JohannessenBengtsson and others, 2004). Over the ice sheet the post-1920 warming is recorded by slightly enriched δ¹⁸O values at all sites except NASA-U (Fig. 2). Note that the δ¹⁸O-inferred warming is most pronounced at the two southernmost sites (Raven and D3). The new Greenland δ¹⁸O records are relatively short and provide a limited temporal perspective, but when coupled with the longer records from Summit they suggest modest 20th-century warming over Greenland with alternating multi-decadal periods of warming and cooling, consistent with the limited available meteorological observations (for a summary see Reference BoxBox, 2002).

An ice-core record from the Windy Dome ice cap in FJL yields quite a different picture (Reference HendersonHenderson, 2002). The decadally averaged δ¹⁸O record (Fig. 1) suggests a large warming that began ∼1910 and was sustained until mid-century, consistent with observations in the Barents Sea region as discussed above. Longer ice cores from this region are needed to assess the 20th-century warming relative to that associated with the Medieval Warm Period (MWP).

The available high-resolution δ¹⁸O records from Northern Hemisphere high-latitude ice cores are generally consistent with 20th-century observations. Specifically, they record (1) the short-term cooling in response to explosive tropical volcanic eruptions that injected sulfate aerosols into the tropical stratosphere where they were globally distributed, (2) the brief large-scale cooling from 1917 to 1919, and (3) the subsequent rapid (few years), large-scale warming in the Arctic (Reference RogersRogers, 1985; Reference OverpeckOverpeck and others, 1997; Reference Jones and MobergJones and Moberg, 2003) that persisted for several decades. Thus it is realistic to assume that δ¹⁸O variations in the pre-20th-century part of the ice-core records also reflect regional temperature trends. The collection of annually dated δ¹⁸O records discussed here suggests that 20th-century warming has been spatially variable over Greenland but, on balance, temperatures have been modestly warmer than in previous centuries, particularly over the southern part of the ice sheet. With the exception of the two southern sites (D3 and Raven), the magnitude of the warming over the 20th century remains smaller than the decadal-scale variability. Eastward in the Russian Arctic, the FJL δ¹⁸O record suggests that 20th-century warming in that high-Arctic region is well outside the norm for the last half-millennium.

Ice cores from tropical and subtropical regions are limited, but new cores are being slowly added to the archive. Two low-latitude ice-core δ¹⁸O composite records have been previously published (Reference Thompson, Mosley-Thompson, Davis, Lin, Henderson and MashiottaThompson and others, 2003) and are included in Figure 1. The composite for the South American Andes is derived from the Quelccaya, Huascaran and Sajama cores, while the Tibetan Plateau composite is based on the Dunde, Guliya and Dasuopu cores. Note that these two composite δ¹⁸O histories reveal some notable regional differences over most of the last millennium. The Andean composite shows a clear Little Ice Age or recent neoglacial cool period (AD ∼1450−1880) as well as warmer conditions from AD∼1100 to 1350, correlative with the MWP. The Tibetan composite shows neither of these multi-century climate variations. The only consistent, century-scale feature between the two regional composites is the marked isotopic enrichment in 20th-century precipitation. Although high-resolution, well-dated tropical and subtropical ice-core records are scarce, they paint a consistent picture, specifically, that the 20th-century warming in the Andes and over the Tibetan Plateau is now outside the range of natural variability for the last millennium. The recent warming in lower-latitude, high-elevation regions is consistent with atmospheric temperatures (Reference Diaz, Eischeid, Duncan and BradleyDiaz and others, 2003; Reference Jones and MobergJones and Moberg, 2003), Intergovernmental Panel on Climate Change model predictions (Reference Cubasch and HoughtonCubasch and others, 2001) and the widespread recession of alpine glaciers (Reference Thompson, Mosley-Thompson, Davis, Lin, Henderson and MashiottaThompson and others, 2003; Reference OerlemansOerlemans, 2005).

The number of annually resolved and carefully dated δ¹⁸O records available from Antarctica is also limited. Three δ¹⁸O records, each representing a different part of the continent, are shown in Figure 1. The annually dated Dyer Plateau (DP) core (Reference ThompsonThompson and others, 1994) is located along the spine of the Antarctic Peninsula, a region where temperatures have increased strongly since the 1950s (Reference Marshall, Lagun and Lachlan-CopeMarshall and others, 2002; Reference King, Turner, Marshall, Connolley, Lachlan-Cope, Domack, Burnett, Leventer, Conley, Kirby and BindschadlerKing and others, 2003). Temperature records here are short, with observations extending back to the late 1940s. The decadally averaged 500 year δ¹⁸O history suggests cooler temperatures from AD∼1700 to 1950, followed by ¹⁸O enrichment that is contemporaneous with the observed warming in the region (Reference Mosley-Thompson, Thompson, Domack, Burnett and LeventerMosley-Thompson and Thompson, 2003). The 20th-century average δ¹⁸O value exceeds that for the previous 400 years, and the isotopically inferred warming of the last few decades lies well outside the range of natural variability for the last half-millennium.

On the high dry polar plateaus of East and West Antarctica, δ¹⁸O records are available from Plateau Remote and Siple Stations. At Siple Station (SS), the high annual accumulation (461 mmw.e.) results in an excellently preserved and easily interpretable annual record (Reference Mosley-Thompson, Bradley and JonesMosley-Thompson, 1992). At Plateau Remote (PR), the low accumulation (∼40mmw.e.) makes annual resolution difficult, but well-known time-stratigraphic volcanic horizons provide excellent time constraints for the last millennium (Reference Mosley-Thompson, Jones, Bradley and JouzelMosley-Thompson, 1996; Reference Cole-Dai, Mosley-Thompson, Wight and ThompsonCole-Dai and others, 2000). The δ¹⁸O history at PR shows a very modest ¹⁸O depletion (cooling) in 20th-century precipitation relative to that over the last millennium, while the ¹⁸O depletion is stronger at SS. Both suggest a recent cooling relative to the earlier part of the record. Antarctic meteorological records are short, few in number and biased toward coastal stations and stations in the Peninsula (Reference TurnerTurner and others, 2005). Two long records exist for the interior of the continent (Vostok and Amundsen-Scott South Pole Station) and both show long-term cooling trends. At South Pole (SP), average annual near-surface temperatures have declined slowly by ∼0.5˚C since record keeping began in 1958 although interannual variability is high (data provided to E.M.-T. by SP meteorological personnel in July 2005).

Thus, recent δ¹⁸O trends in Antarctic precipitation appear regionally consistent with the observed temperature trends. The recent ¹⁸O enrichment in the DP core from the spine of the Peninsula and the modest ¹⁸O depletion in the SS and PR cores on the high polar plateau are consistent with the warming and cooling trends observed in those regions, respectively (Reference Vaughan, Marshall, Connolley, King and MulvaneyVaughan and others, 2001 and references therein; Reference Thompson and SolomonThompson and Solomon, 2002; Reference TurnerTurner and others, 2005). Assuming that the earlier portions of these δ¹⁸O records also provide credible temperature proxies, the recent warming in the Peninsula region is unusual within the longer (500years) perspective available, and the slight cooling in East and West Antarctica lies within the range of past variability (1000years for PR and 500 years for SS).

Conclusions

High temporally resolved ice-core-derived δ¹⁸O records have been examined to assess whether they provide reliable proxies for examining regional 20th-century temperature trends within a longer-term perspective. The largest isotopically inferred warming is found in the Russian Arctic. Warmer 20th-century temperatures characterize the Tibetan Plateau and the South American Andes and extend into the Antarctic Peninsula. The 20th-century warming in these regions lies outside the range of natural variability as discerned from the earlier part of their respective δ¹⁸O records. Over Greenland a modest warming that is most pronounced over the southwestern part of the ice sheet has ensued since ∼1920. Modest cooling has dominated 20th-century temperatures over the high plateaus of East and West Antarctica. The unique climate histories preserved in the glaciers and ice caps from the Russian Arctic through the tropics to the Antarctic Peninsula may soon be degraded and/or lost if the current warming in those regions persists.