IPCC Fourth Assessment Report: Climate Change 2007
Climate Change 2007: Working Group I: The Physical Science Basis

6.6 The Last 2,000 Years

6.6.1 Northern Hemisphere Temperature Variability What Do Reconstructions Based on Palaeoclimatic Proxies Show?

Figure 6.10 shows the various instrumental and proxy climate evidence of the variations in average large-scale surface temperatures over the last 1.3 kyr. Figure 6.10a shows two instrumental compilations representing the mean annual surface temperature of the NH since 1850, one based on land data only, and one using land and surface ocean data combined (see Chapter 3). The uncertainties associated with one of these series are also shown (30-year smoothed combined land and marine). These arise primarily from the incomplete spatial coverage of instrumentation through time (Jones et al., 1997) and, whereas these uncertainties are larger in the 19th compared to the 20th century, the prominence of the recent warming, especially in the last two to three decades of the record, is clearly apparent in this 150-year context. The land-only record shows similar variability, although the rate of warming is greater than in the combined record after about 1980. The land-only series can be extended back beyond the 19th century, and is shown plotted from 1781 onwards. The early section is based on a much sparser network of available station data, with at least 23 European stations, but only one North American station, spanning the first two decades, and the first Asian station beginning only in the 1820s. Four European records (Central England, De Bilt, Berlin and Uppsala) provide an even longer, though regionally restricted, indication of the context for the warming observed in the last approximately 20 to 30 years, which is even greater in this area than is observed over the NH land as a whole.

Figure 6.10

Figure 6.10. Records of NH temperature variation during the last 1.3 kyr. (a) Annual mean instrumental temperature records, identified in Table 6.1. (b) Reconstructions using multiple climate proxy records, identified in Table 6.1, including three records (JBB..1998, MBH..1999 and BOS..2001) shown in the TAR, and the HadCRUT2v instrumental temperature record in black. (c) Overlap of the published multi-decadal time scale uncertainty ranges of all temperature reconstructions identified in Table 6.1 (except for RMO..2005 and PS2004), with temperatures within ±1 standard error (SE) of a reconstruction ‘scoring’ 10%, and regions within the 5 to 95% range ‘scoring’ 5% (the maximum 100% is obtained only for temperatures that fall within ±1 SE of all 10 reconstructions). The HadCRUT2v instrumental temperature record is shown in black. All series have been smoothed with a Gaussian-weighted filter to remove fluctuations on time scales less than 30 years; smoothed values are obtained up to both ends of each record by extending the records with the mean of the adjacent existing values. All temperatures represent anomalies (°C) from the 1961 to 1990 mean.

The instrumental temperature data that exist before 1850, although increasingly biased towards Europe in earlier periods, show that the warming observed after 1980 is unprecedented compared to the levels measured in the previous 280 years, even allowing for the greater variance expected in an average of so few early data compared to the much greater number in the 20th century. Recent analyses of instrumental, documentary and proxy climate records, focussing on European temperatures, have also pointed to the unprecedented warmth of the 20th century and shown that the extreme summer of 2003 was very likely warmer than any that has occurred in at least 500 years (Luterbacher et al., 2004; Guiot et al., 2005; see Box 3.6).

If the behaviour of recent temperature change is to be understood, and the mechanisms and causes correctly attributed, parallel efforts are needed to reconstruct the longer and more widespread pre-instrumental history of climate variability, as well as the detailed changes in various factors that might influence climate (Bradley et al., 2003b; Jones and Mann, 2004).

The TAR discussed various attempts to use proxy data to reconstruct changes in the average temperature of the NH for the period after AD 1000, but focused on three reconstructions (included in Figure 6.10), all with yearly resolution. The first (Mann et al., 1999) represents mean annual temperatures, and is based on a range of proxy types, including data extracted from tree rings, ice cores and documentary sources; this reconstruction also incorporates a number of instrumental (temperature and precipitation) records from the 18th century onwards. For 900 years, this series exhibits multi-decadal fluctuations with amplitudes up to 0.3°C superimposed on a negative trend of 0.15°C, followed by an abrupt warming (~0.4°C) matching that observed in the instrumental data during the first half of the 20th century. Of the other two reconstructions, one (Jones et al., 1998) was based on a much smaller number of proxies, whereas the other (Briffa et al., 2001) was based solely on tree ring density series from an expansive area of the extratropics, but reached back only to AD 1400. These two reconstructions emphasise warm season rather than annual temperatures, with a geographical focus on extratropical land areas. They indicate a greater range of variability on centennial time scales prior to the 20th century, and also suggest slightly cooler conditions during the 17th century than those portrayed in the Mann et al. (1998, 1999) series.

The ‘hockey stick’ reconstruction of Mann et al. (1999) has been the subject of several critical studies. Soon and Baliunas (2003) challenged the conclusion that the 20th century was the warmest at a hemispheric average scale. They surveyed regionally diverse proxy climate data, noting evidence for relatively warm (or cold), or alternatively dry (or wet) conditions occurring at any time within pre-defined periods assumed to bracket the so-called ‘Medieval Warm Period’ (and ‘Little Ice Age’). Their qualitative approach precluded any quantitative summary of the evidence at precise times, limiting the value of their review as a basis for comparison of the relative magnitude of mean hemispheric 20th-century warmth (Mann and Jones, 2003; Osborn and Briffa, 2006). Box 6.4 provides more information on the ‘Medieval Warm Period’.

Box 6.4: Hemispheric Temperatures in the ‘Medieval Warm Period’

At least as early as the beginning of the 20th century, different authors were already examining the evidence for climate changes during the last two millennia, particularly in relation to North America, Scandinavia and Eastern Europe (Brooks, 1922). With regard to Iceland and Greenland, Pettersson (1914) cited evidence for considerable areas of Iceland being cultivated in the 10th century. At the same time, Norse settlers colonised areas of Greenland, while a general absence of sea ice allowed regular voyages at latitudes far to the north of what was possible in the colder 14th century. Brooks (1922) described how, after some amelioration in the 15th and 16th centuries, conditions worsened considerably in the 17th century; in Iceland, previously cultivated land was covered by ice. Hence, at least for the area of the northern North Atlantic, a picture was already emerging of generally warmer conditions around the centuries leading up to the end of the first millennium, but framed largely by comparison with strong evidence of much cooler conditions in later centuries, particularly the 17th century.

Lamb (1965) seems to have been the first to coin the phrase ‘Medieval Warm Epoch’ or ‘Little Optimum’ to describe the totality of multiple strands of evidence principally drawn from western Europe, for a period of widespread and generally warmer temperatures which he put at between AD 1000 and 1200 (Lamb, 1982). It is important to note that Lamb also considered the warmest conditions to have occurred at different times in different areas: between 950 and 1200 in European Russia and Greenland, but somewhat later, between 1150 and 1300 (though with notable warmth also in the later 900s) in most of Europe (Lamb, 1977).

Much of the evidence used by Lamb was drawn from a very diverse mixture of sources such as historical information, evidence of treeline and vegetation changes, or records of the cultivation of cereals and vines. He also drew inferences from very preliminary analyses of some Greenland ice core data and European tree ring records. Much of this evidence was difficult to interpret in terms of accurate quantitative temperature influences. Much was not precisely dated, representing physical or biological systems that involve complex lags between forcing and response, as is the case for vegetation and glacier changes. Lamb’s analyses also predate any formal statistical calibration of much of the evidence he considered. He concluded that ‘High Medieval’ temperatures were probably 1.0°C to 2.0°C above early 20th-century levels at various European locations (Lamb, 1977; Bradley et al., 2003a).

A later study, based on examination of more quantitative evidence, in which efforts were made to control for accurate dating and specific temperature response, concluded that it was not possible to say anything other than ‘… in some areas of the Globe, for some part of the year, relatively warm conditions may have prevailed’ (Hughes and Diaz, 1994).

In medieval times, as now, climate was unlikely to have changed in the same direction, or by the same magnitude, everywhere (Box 6.4, Figure 1). At some times, some regions may have experienced even warmer conditions than those that prevailed throughout the 20th century (e.g., see Bradley et al., 2003a). Regionally restricted evidence by itself, especially when the dating is imprecise, is of little practical relevance to the question of whether climate in medieval times was globally as warm or warmer than today. Local climate variations can be dominated by internal climate variability, often the result of the redistribution of heat by regional climate processes. Only very large-scale climate averages can be expected to reflect global forcings over recent millennia (Mann and Jones, 2003; Goosse et al., 2005a). To define medieval warmth in a way that has more relevance for exploring the magnitude and causes of recent large-scale warming, widespread and continuous palaeoclimatic evidence must be assimilated in a homogeneous way and scaled against recent measured temperatures to allow a meaningful quantitative comparison against 20th-century warmth (Figure 6.10).

A number of studies that have attempted to produce very large spatial-scale reconstructions have come to the same conclusion: that medieval warmth was heterogeneous in terms of its precise timing and regional expression (Crowley and Lowery, 2000; Folland et al., 2001; Esper et al., 2002; Bradley et al., 2003a; Jones and Mann, 2004; D’Arrigo et al., 2006).

Box 6.4 Figure 1

Box 6.4, Figure 1. The heterogeneous nature of climate during the ‘Medieval Warm Period’ is illustrated by the wide spread of values exhibited by the individual records that have been used to reconstruct NH mean temperature. These consist of individual, or small regional averages of, proxy records collated from those used by Mann and Jones (2003), Esper et al. (2002) and Luckman and Wilson (2005), but exclude shorter series or those with no evidence of sensitivity to local temperature. These records have not been calibrated here, but each has been smoothed with a 20-year filter and scaled to have zero mean and unit standard deviation over the period 1001 to 1980.

The uncertainty associated with present palaeoclimate estimates of NH mean temperatures is significant, especially for the period prior to 1600 when data are scarce (Mann et al., 1999; Briffa and Osborn, 2002; Cook et al., 2004a). However, Figure 6.10 shows that the warmest period prior to the 20th century very likely occurred between 950 and 1100, but temperatures were probably between 0.1°C and 0.2°C below the 1961 to 1990 mean and significantly below the level shown by instrumental data after 1980.

In order to reduce the uncertainty, further work is necessary to update existing records, many of which were assembled up to 20 years ago, and to produce many more, especially early, palaeoclimate series with much wider geographic coverage. There are far from sufficient data to make any meaningful estimates of global medieval warmth (Figure 6.11). There are very few long records with high temporal resolution data from the oceans, the tropics or the SH.

The evidence currently available indicates that NH mean temperatures during medieval times (950–1100) were indeed warm in a 2-kyr context and even warmer in relation to the less sparse but still limited evidence of widespread average cool conditions in the 17th century (Osborn and Briffa, 2006). However, the evidence is not sufficient to support a conclusion that hemispheric mean temperatures were as warm, or the extent of warm regions as expansive, as those in the 20th century as a whole, during any period in medieval times (Jones et al., 2001; Bradley et al., 2003a,b; Osborn and Briffa, 2006).

McIntyre and McKitrick (2003) reported that they were unable to replicate the results of Mann et al. (1998). Wahl and Ammann (2007) showed that this was a consequence of differences in the way McIntyre and McKitrick (2003) had implemented the method of Mann et al. (1998) and that the original reconstruction could be closely duplicated using the original proxy data. McIntyre and McKitrick (2005a,b) raised further concerns about the details of the Mann et al. (1998) method, principally relating to the independent verification of the reconstruction against 19th-century instrumental temperature data and to the extraction of the dominant modes of variability present in a network of western North American tree ring chronologies, using Principal Components Analysis. The latter may have some theoretical foundation, but Wahl and Amman (2006) also show that the impact on the amplitude of the final reconstruction is very small (~0.05°C; for further discussion of these issues see also Huybers, 2005; McIntyre and McKitrick, 2005c,d; von Storch and Zorita, 2005).

Table 6.1. Records of Northern Hemisphere temperature shown in Figure 6.10.

Instrumental temperatures  
Series  Period  Description  Reference  
HadCRUT2va  1856–2005  Land and marine temperatures for the NH  Jones and Moberg, 2003; errors from Jones et al., 1997 
CRUTEM2vb  1781–2004  Land-only temperatures for the NH  Jones and Moberg, 2003; extended using data from Jones et al., 2003  
4 European Stations  1721–2004  Average of central England, De Bilt, Berlin and Uppsala  Jones et al., 2003  
Proxy-based reconstructions of temperature  
Series   Period  Reconstructed Season   Region  Location Of Proxiesc  Reference  
JBB..1998  1000–1991  Summer  Land, 20°N–90°N  ◢ ◢  □ □ Jones et al., 1998; calibrated by Joneset al., 2001  
MBH1999  1000–1980  Annual  Land + marine, 0–90°N  ■  ■  ◢  ◢  Mann et al., 1999  
BOS..2001  1402–1960  Summer  Land, 20°N–90°N  ■  ◢  □  □  Briffa et al., 2001  
ECS2002   831–1992  Annual  Land, 20°N–90°N  ◢  ◢ □ □ Esper et al., 2002; recalibrated by Cooket al., 2004a  
B2000   1–1993  Summer  Land, 20°N–90°N  ◢  □ □ □ Briffa, 2000; calibrated by Briffa et al., 2004  
MJ2003   200–1980  Annual  Land + marine, 0–90°N  ◢  ◢  □  □  Mann and Jones, 2003  
RMO..2005  1400–1960  Annual  Land + marine, 0–90°N  ■  ■  ◢  ◢  Rutherford et al., 2005  
MSH..2005   1–1979  Annual  Land + marine, 0–90°N  ◢  ◢  ◢  ◢  Moberg et al., 2005  
DWJ2006   713–1995  Annual  Land, 20°N–90°N  ■  ◢ □  □  D’Arrigo et al., 2006  
HCA..2006   558–1960  Annual  Land, 20°N–90°N  ◢ ◢ □  □  Hegerl et al., 2006  
PS2004  1500–2000  Annual  Land, 0–90°N  ◢ ■  □  □  Pollack and Smerdon, 2004; reference level adjusted following Moberg et al., 2005  
O2005  1600–1990  Summer  Global land  ◢  ■  □  □  Oerlemans, 2005  


a Hadley Centre/Climatic Research Unit gridded surface temperature data set, version 2 variance adjusted.

b Climatic Research Unit gridded land surface air temperature, version 2 variance corrected.

c Location of proxies from H = high-latitude land, M = mid-latitude land, L = low-latitude land, O = oceans is indicated by □ (none or very few), ◢ (limited coverage) or ■ (moderate or good coverage).

Since the TAR, a number of additional proxy data syntheses based on annually or near-annually resolved data, variously representing mean NH temperature changes over the last 1 or 2 kyr, have been published (Esper et al., 2002; Crowley et al., 2003; Mann and Jones, 2003; Cook et al., 2004a; Moberg et al., 2005; Rutherford et al., 2005; D’Arrigo et al., 2006). These are shown, plotted from AD 700 in Figure 6.10b, along with the three series from the TAR. As with the original TAR series, these new records are not entirely independent reconstructions inasmuch as there are some predictors (most often tree ring data and particularly in the early centuries) that are common between them, but in general, they represent some expansion in the length and geographical coverage of the previously available data (Figures 6.10 and 6.11).

Figure 6.11

Figure 6.11. Locations of proxy records with data back to AD 1000, 1500 and 1750 (instrumental: red thermometers; tree ring: brown triangles; borehole: black circles; ice core/ice borehole: blue stars; other including low-resolution records: purple squares) that have been used to reconstruct NH or SH temperatures by studies shown in Figure 6.10 (see Table 6.1, excluding O2005) or used to indicate SH regional temperatures (Figure 6.12).

Briffa (2000) produced an extended history of interannual tree ring growth incorporating records from sites across northern Fennoscandia and northern Siberia, using a statistical technique to construct the tree ring chronologies that is capable of preserving multi-centennial time scale variability. Although ostensibly representative of northern Eurasian summer conditions, these data were later scaled using simple linear regression against a mean NH land series to provide estimates of summer temperature over the past 2 kyr (Briffa et al., 2004). Esper et al. (2002) took tree ring data from 14 sites in Eurasia and North America, and applied a variant of the same statistical technique designed to produce ring width chronologies in which evidence of long time scale climate forcing is better represented compared with earlier tree ring processing methods. The resulting series were averaged, smoothed and then scaled so that the multi-decadal variance matched that in the Mann et al. (1998) reconstruction over the period 1900 to 1977. This produced a reconstruction with markedly cooler temperatures during the 12th to the end of the 14th century than are apparent in any other series. The relative amplitude of this reconstruction is reduced somewhat when recalibrated directly against smoothed instrumental temperatures (Cook et al., 2004a) or by using annually resolved temperature data (Briffa and Osborn, 2002), but even then, this reconstruction remains at the coldest end of the range defined by all currently available reconstructions.

Mann and Jones (2003) selected only eight normalised series (all screened for temperature sensitivity) to represent annual mean NH temperature change over the last 1.8 kyr. Four of these eight represent integrations of multiple proxy site records or reconstructions, including some O isotope records from ice cores and documentary information as well tree ring records. A weighted average of these decadally smoothed series was scaled so that its mean and standard deviation matched those of the NH decadal mean land and marine record over the period 1856 to 1980. Moberg et al. (2005) used a mixture of tree ring and other proxy-based climate reconstructions to represent changes at short and longer time scales, respectively, across the NH. Seven tree ring series provided information on time scales shorter than 80 years, while 11 far less accurately dated records with lower resolution (including ice melt series, lake diatoms and pollen data, chemistry of marine shells and foraminifera, and one borehole temperature record from the Greenland Ice Sheet) were combined and scaled to match the mean and standard deviation of the instrumental record between 1856 and 1979. This reconstruction displays the warmest temperatures of any reconstruction during the 10th and early 11th centuries, although still below the level of warmth observed since 1980.

Many of the individual annually resolved proxy series used in the various reconstruction studies cited above have been combined in a new reconstruction (only back to AD 1400) based on a climate field reconstruction technique (Rutherford et al., 2005). This study also involved a methodological exploration of the sensitivity of the results to the precise specification of the predictor set, as well as the predictand target region and seasonal window. It concluded that the reconstructions were reasonably robust to differences in the choice of proxy data and statistical reconstruction technique.

D’Arrigo et al. (2006) used only tree ring data, but these include a substantial number not used in other reconstructions, particularly in northern North America. Their reconstruction, similar to that of Esper et al. (2002), displays a large amplitude of change during the past 1 kyr, associated with notably cool excursions during most of the 9th, 13th and 14th centuries, clearly below those of most other reconstructions. Hegerl et al. (2006) used a mixture of 14 regional series, of which only 3 were not made up from tree ring data (a Greenland ice O isotope record and two composite series, from China and Europe, including a mixture of instrumental, documentary and other data). Many of these are common to the earlier reconstructions. However, these series were combined and scaled using a regression approach (total least squares) intended to prevent the loss of low-frequency variance inherent in some other regression approaches. The reconstruction produced lies close to the centre of the range defined by the other reconstructions.

Various statistical methods are used to convert the various sets of original palaeoclimatic proxies into the different estimates of mean NH temperatures shown in Figure 6.10 (see discussions in Jones and Mann, 2004; Rutherford et al., 2005). These range from simple averaging of regional data and scaling of the resulting series so that its mean and standard deviation match those of the observed record over some period of overlap (Jones et al., 1998; Crowley and Lowery, 2000), to complex climate field reconstruction, where large-scale modes of spatial climate variability are linked to patterns of variability in the proxy network via a multivariate transfer function that explicitly provides estimates of the spatio-temporal changes in past temperatures, and from which large-scale average temperature changes are derived by averaging the climate estimates across the required region (Mann et al., 1998; Rutherford et al., 2003, 2005). Other reconstructions can be considered to represent what are essentially intermediate applications between these two approaches, in that they involve regionalisation of much of the data prior to the use of a statistical transfer function, and so involve fewer, but potentially more robust, regional predictors (Briffa et al., 2001; Mann and Jones, 2003; D’Arrigo et al., 2006). Some of these studies explicitly or implicitly reconstruct tropical temperatures based on data largely from the extratropics, and assume stability in the patterns of climate association between these regions. This assumption has been questioned on the basis of both observational and model-simulated data suggesting that tropical to extratropical climate variability can be decoupled (Rind et al., 2005), and also that extratropical teleconnections associated with ENSO may vary through time (see Section 6.5.6).

Oerlemans (2005) constructed a temperature history for the globe based on 169 glacier length records. He used simplified glacier dynamics that incorporate specific response time and climate sensitivity estimates for each glacier. The reconstruction suggests that moderate global warming occurred after the middle of the 19th century, with about 0.6°C warming by the middle of the 20th century. Following a 25-year cooling, temperatures rose again after 1970, though much regional and high-frequency variability is superimposed on this overall interpretation. However, this approach does not allow for changing glacier sensitivity over time, which may limit the information before 1900. For example, analyses of glacier mass balances, volume changes and length variations along with temperature records in the western European Alps (Vincent et al., 2005) indicate that between 1760 and 1830, glacier advance was driven by precipitation that was 25% above the 20th century average, while there was little difference in average temperatures. Glacier retreat after 1830 was related to reduced winter precipitation and the influence of summer warming only became effective at the beginning of the 20th century. In southern Norway, early 18th-century glacier advances can be attributed to increased winter precipitation rather than cold temperatures (Nesje and Dahl, 2003).

Changes in proxy records, either physical (such as the isotopic composition of various elements in ice) or biological (such as the width of a tree ring or the chemical composition of a growth band in coral), do not respond precisely or solely to changes in any specific climate parameter (such as mean temperature or total rainfall), or to the changes in that parameter as measured over a specific ‘season’ (such as June to August or January to December). For this reason, the proxies must be ‘calibrated’ empirically, by comparing their measured variability over a number of years with available instrumental records to identify some optimal climate association, and to quantify the statistical uncertainty associated with scaling proxies to represent this specific climate parameter. All reconstructions, therefore, involve a degree of compromise with regard to the specific choice of ‘target’ or dependent variable. Differences between the temperature reconstructions shown in Figure 6.10b are to some extent related to this, as well as to the choice of different predictor series (including differences in the way these have been processed). The use of different statistical scaling approaches (including whether the data are smoothed prior to scaling, and differences in the period over which this scaling is carried out) also influences the apparent spread between the various reconstructions. Discussions of these issues can also be found in Harris and Chapman (2001), Beltrami (2002), Briffa and Osborn (2002), Esper et al. (2002), Trenberth and Otto-Bliesner (2003), Zorita et al. (2003), Jones and Mann (2004), Pollack and Smerdon (2004), Esper et al. (2005) and Rutherford et al. (2005).

The considerable uncertainty associated with individual reconstructions (2-standard-error range at the multi-decadal time scale is of the order of ±0.5°C) is shown in several publications, calculated on the basis of analyses of regression residuals (Mann et al., 1998; Briffa et al., 2001; Jones et al., 2001; Gerber et al., 2003; Mann and Jones, 2003; Rutherford et al., 2005; D’Arrigo et al., 2006). These are often calculated from the error apparent in the calibration of the proxies. Hence, they are likely to be minimum uncertainties, as they do not take into account other sources of error not apparent in the calibration period, such as any reduction in the statistical robustness of the proxy series in earlier times (Briffa and Osborn, 1999; Esper et al., 2002; Bradley et al., 2003b; Osborn and Briffa, 2006).

All of the large-scale temperature reconstructions discussed in this section, with the exception of the borehole and glacier interpretations, include tree ring data among their predictors so it is pertinent to note several issues associated with them. The construction of ring width and ring density chronologies involves statistical processing designed to remove non-climate trends that could obscure the evidence of climate that they contain. In certain situations, this process may restrict the extent to which a chronology portrays the evidence of long time scale changes in the underlying variability of climate that affected the growth of the trees; in effect providing a high-pass filtered version of past climate. However, this is generally not the case for chronologies used in the reconstructions illustrated in Figure 6.10. Virtually all of these used chronologies or tree ring climate reconstructions produced using methods that preserve multi-decadal and centennial time scale variability. As with all biological proxies, the calibration of tree ring records using linear regression against some specific climate variable represents a simplification of what is inevitably a more complex and possibly time-varying relationship between climate and tree growth. That this is a defensible simplification, however, is shown by the general strength of many such calibrated relationships, and their significant verification using independent instrumental data. There is always a possibility that non-climate factors, such as changing atmospheric CO2 or soil chemistry, might compromise the assumption of uniformity implicit in the interpretation of regression-based climate reconstructions, but there remains no evidence that this is true for any of the reconstructions referred to in this assessment. A group of high-elevation ring width chronologies from the western USA that show a marked growth increase during the last 100 years, attributed by LaMarche et al. (1984) to the fertilizing effect of increasing atmospheric CO2, were included among the proxy data used by Mann et al. (1998, 1999). However, their tree ring data from the western USA were adjusted specifically in an attempt to mitigate this effect. Several analyses of ring width and ring density chronologies, with otherwise well-established sensitivity to temperature, have shown that they do not emulate the general warming trend evident in instrumental temperature records over recent decades, although they do track the warming that occurred during the early part of the 20th century and they continue to maintain a good correlation with observed temperatures over the full instrumental period at the interannual time scale (Briffa et al., 2004; D’Arrigo, 2006). This ‘divergence’ is apparently restricted to some northern, high-latitude regions, but it is certainly not ubiquitous even there. In their large-scale reconstructions based on tree ring density data, Briffa et al. (2001) specifically excluded the post-1960 data in their calibration against instrumental records, to avoid biasing the estimation of the earlier reconstructions (hence they are not shown in Figure 6.10), implicitly assuming that the ‘divergence’ was a uniquely recent phenomenon, as has also been argued by Cook et al. (2004a). Others, however, argue for a breakdown in the assumed linear tree growth response to continued warming, invoking a possible threshold exceedance beyond which moisture stress now limits further growth (D’Arrigo et al., 2004). If true, this would imply a similar limit on the potential to reconstruct possible warm periods in earlier times at such sites. At this time there is no consensus on these issues (for further references see NRC, 2006) and the possibility of investigating them further is restricted by the lack of recent tree ring data at most of the sites from which tree ring data discussed in this chapter were acquired.

Figure 6.10b illustrates how, when viewed together, the currently available reconstructions indicate generally greater variability in centennial time scale trends over the last 1 kyr than was apparent in the TAR. It should be stressed that each of the reconstructions included in Figure 6.10b is shown scaled as it was originally published, despite the fact that some represent seasonal and others mean annual temperatures. Except for the borehole curve (Pollack and Smerdon, 2004) and the interpretation of glacier length changes (Oerlemans, 2005), they were originally also calibrated against different instrumental data, using a variety of statistical scaling approaches. For all these reasons, these reconstructions would be expected to show some variation in relative amplitude.

Figure 6.10c is a schematic representation of the most likely course of hemispheric mean temperature change during the last 1.3 kyr based on all of the reconstructions shown in Figure 6.10b, and taking into account their associated statistical uncertainty. The envelopes that enclose the two standard error confidence limits bracketing each reconstruction have been overlain (with greater emphasis placed on the area within the 1 standard error limits) to show where there is most agreement between the various reconstructions. The result is a picture of relatively cool conditions in the 17th and early 19th centuries and warmth in the 11th and early 15th centuries, but the warmest conditions are apparent in the 20th century. Given that the confidence levels surrounding all of the reconstructions are wide, virtually all reconstructions are effectively encompassed within the uncertainty previously indicated in the TAR. The major differences between the various proxy reconstructions relate to the magnitude of past cool excursions, principally during the 12th to 14th, 17th and 19th centuries. Several reconstructions exhibit a short-lived maximum just prior to AD 1000 but only one (Moberg et al., 2005) indicates persistent hemispheric-scale conditions (i.e., during AD 990 to 1050 and AD 1080 to 1120) that were as warm as those in the 1940s and 50s. However, the long time scale variability in this reconstruction is determined by low-resolution proxy records that cannot be rigorously calibrated against recent instrumental temperature data (Mann et al., 2005b). None of the reconstructions in Fig. 6.10 show pre-20th century temperatures reaching the levels seen in the instrumental temperature record for the last two decades of the 20th century.

It is important to recognise that in the NH as a whole there are few long and well-dated climate proxies, particularly for the period prior to the 17th century (Figure 6.11). Those that do exist are concentrated in extratropical, terrestrial locations, and many have greatest sensitivity to summer rather than winter (or annual) conditions. Changes in seasonality probably limit the conclusions that can be drawn regarding annual temperatures derived from predominantly summer-sensitive proxies (Jones et al., 2003). There are very few strongly temperature-sensitive proxies from tropical latitudes. Stable isotope data from high-elevation ice cores provide long records and have been interpreted in terms of past temperature variability (Thompson, 2000), but recent calibration and modelling studies in South America and southern Tibet (Hoffmann et al., 2003; Vuille and Werner, 2005; Vuille et al., 2005) indicate a dominant sensitivity to precipitation changes, at least on seasonal to decadal time scales, in these regions. Very rapid and apparently unprecedented melting of tropical ice caps has been observed in recent decades (Thompson et al., 2000; Thompson, 2001; see Box 6.3), likely associated with enhanced warming at high elevations (Gaffen et al., 2000; see Chapter 4). Coral O isotopes and Sr/Ca ratios reflect SSTs, although the former are also influenced by salinity changes associated with precipitation variability (Lough, 2004). Unfortunately, these records are invariably short, of the order of centuries at best, and can be associated with age uncertainties of 1 or 2%. Virtually all coral records currently available from the tropical Indo-Pacific indicate unusual warmth in the 20th century (Cole, 2003), and in the tropical Indian Ocean many isotope records show a trend towards warmer conditions (Charles et al., 1997; Kuhnert et al., 1999; Cole et al., 2000). In most multi-centennial length coral series, the late 20th century is warmer than any time in the last 100 to 300 years.

Using pseudo-proxy networks extracted from GCM simulations of global climate for the last millennium, von Storch et al. (2004) suggested that temperature reconstructions may not fully represent variance on long time scales. This would represent a bias, as distinct from the random error represented by published reconstruction uncertainty ranges. At present, the extent of any such biases in specific reconstructions and as indicated by pseudo-proxy studies is uncertain (being dependent on the choice of statistical regression model and climate model simulation used to provide the pseudo-proxies). It is very unlikely, however, that any bias would be as large as the factor of two suggested by von Storch et al. (2004) with regard to the reconstruction by Mann et al. (1998), as discussed by Burger and Cubash (2005) and Wahl et al. (2006). However, the bias will depend on the degree to which past climate departs from the range of temperatures encompassed within the calibration period data (Mann et al., 2005b; Osborn and Briffa, 2006) and on the proportions of temperature variability occurring on short and long time scales (Osborn and Briffa, 2004). In any case, this bias would act to damp the amplitude of reconstructed departures that are further from the calibration period mean, so that temperatures during cooler periods may have been colder than estimated by some reconstructions, while periods with comparable temperatures (e.g., possible portions of the period between AD 950 and 1150, Figure 6.10) would be largely unbiased. As only one reconstruction (Moberg et al., 2005) shows an early period that is noticeably warmer than the mean for the calibration period, the possibility of a bias does not affect the general conclusion about the relative warmth of the 20th century based on these data.

The weight of current multi-proxy evidence, therefore, suggests greater 20th-century warmth, in comparison with temperature levels of the previous 400 years, than was shown in the TAR. On the evidence of the previous and four new reconstructions that reach back more than 1 kyr, it is likely that the 20th century was the warmest in at least the past 1.3 kyr. Considering the recent instrumental and longer proxy evidence together, it is very likely that average NH temperatures during the second half of the 20th century were higher than for any other 50-year period in the last 500 years. Greater uncertainty associated with proxy-based temperature estimates for individual years means that it is more difficult to gauge the significance, or precedence, of the extreme warm years observed in the recent instrumental record, such as 1998 and 2005, in the context of the last millennium.