Temperature Reconstructions

In this section, the reconstructions of MBH98, the associated uncertainties, and raw data used in calibration and verification are available in a variety of formats.

A. Large-Scale Trends   B. Spatial Patterns   C. Influence of Climate Forcings


B. Spatial Patterns

Yearly global temperature maps for annual-mean, Boreal cold-season and warm-season are available below for the reconstructed temperature fields (1730-1980), the raw temperature data (1902-1993) used for calibration, and the sparse raw "verification" temperature data (1854-1901) used for cross-validation. Also available are the "EOF-filtered" instrumental data from 1902-1993. In the latter case, only that data variance during the calibration period described by the actual eigenvectors used to calibrate the multiproxy network (see "Data and Methods" section) are retained. These filtered versions of the raw data are thus in some sense a more appropriate standard for comparison to the multiproxy-reconstructed patterns than the raw data itself.

Raw data and reconstructed patterns can be compared side by side where available. These maps are "clickable" so that time series for particular regions (with uncertainties, in the case of reconstructions) can be obtained. An ANIMATION of the temperature fields is also available.

To investigate the spatial patterns and time histories of the global temperature reconstructions on your own, you can begin by clicking here or on the annual-mean anomaly map below for the year 1730.




In part due to the especially strong El Nino of 1997/1998, there has been renewed interest in past variations in the El Nino phenomenon, and the context in which they place prominent recent (1997/1998 and 1982/1983) events. Two excellent examples of past very strong El Ninos in the reconstructions are those evident in the temperature patterns for 1791 and 1878 (Figure 13). Independent corroboration of these events is provided by the historical chronology of Quinn and Neal, 1992 [see MBH98]. The reader may note that an alternative historical El Nino chronology has recently been produced by Ortlieb (2000), with conclusions that sometimes differ from those of Quinn and Neal. We note however that, to the extent that the Quinn chronology used here is imperfect, it will provide a very conservative corroboration of our own chronology, as mismatch may be due to uncertainties in the chronology as well as in our reconstruction]. Both events shown exhibit the classic eastern tropical Pacific warming and horseshoe pattern of warming and cooling in the North Pacific. The details of ENSO-related patterns of variation in the annual-mean temperature reconstructions is discussed by Mann et al (2000).


El Nino events during 1791.



Figure 13:
Global temperature pattern reconstructions for two historically documented very strong El Nino events during 1791 (top) and 1878 (bottom).

El Nino events during 1878.

The reconstructed annual mean NINO3 index provides an estimate of El Nino-related temperature variability in our reconstructions. Based on this index, the 1997-1998 (and 1982-1983) events appeared (see the discussion in Mann et al, 2000) to be among the strongest events back to at least AD 1650. It could not however, be concluded with much certainty at that time that that they are stronger than any other events during that period, owing to the appreciable uncertainties in the reconstructions for the tropical Pacific region and the sub-optimal calendar-mean basis for that reconstruction.

However, as evident in the statistics, the cold-season NINO3 index calibrates/cross-validates a considerably larger share of the instrumental data variance than the annual-mean series (about 50% in calibration and verification back to 1780, and about 40% back to 1650). This is not surprising, as a boreal cold-season window is a more appropriate basis for defining the ENSO phenomenon than a calendar mean. Our winter NINO3 reconstruction exhibits a highly significant correlation with the largely independent reconstruction of the winter (DJF) Southern Oscillation Index(SOI) of Stahle et al (1998). The two reconstructions are correlated at r=0.63 over the full period of overlap (1705-1976) and r=0.60 during the pre-calibration interval (1705-1901). This is nearly as high as the observed correlation(r=0.7) between the instrumental SOI and NINO3 series during the 20th century. The similarity of these two reconstructions, as well as the significant correspondance with the historically-based El Nino chronology of Quinn and Neal (1992) discussed earlier (see cold- season calibration/verification statistics) suggests considerable reliability in the ENSO-related features of our surface temperature reconstructions. Using this improved, seasonal reconstruction of NINO3 (Figure 14) we find added evidence that the two recent events 1982/1983 and 1997/1998 stand out as somewhat anomalous in the long term record. There is evidence that certain events (such as the 1877/1878 El Nino) may be more underestimated in their amplitude in our reconstruction than would be expected from random calibration uncertainties. This is difficult to determine, as the instrumental surface temperature record is quite sparse during that period of time. Moreover (see Mann et al, 2000) the winter SOI, used as a substitute for the NINO3 index, shows quite similar behavior to our reconstruction at that time. Only further work with both the instrumental record and proxy climate records in ENSO-sensitive regions will further elucidate this issue.

Larger Image

Figure 14:
Reconstructed boreal cold-season NINO3 index back to 1650. Shown for comparison is a partially independent (dendroclimatic rather than multiproxy) winter (DFJ) reconstruction of the Southern Oscillation Index (SOI-Stahle et al, 1998). Yellow shaded region indicates the 95% confidence bounds for the NINO3 reconstruction.


As commented upon earlier, there are important distinctions between regional and hemispheric trends, with regional trends exhibiting considerably greater variability and heterogeneity. Cold and warm periods in different parts of the globe, for example, are not in general synchronous. Even during the "Little Ice Age" (Bradley and Jones, 1992) not all areas were uniformly cold; geographical and temporal variations were apparent, as highlighted by an examination of the reconstructions presented here. The "Medieval Warm Period" or "Medieval Optimum" (Hughes and Diaz, 1994) is even more enigmatic.

It is sometimes erroneously argued that the globe was as warm or even warmer than present during the early part of the millennium (e.g., AD 1000-1200) based on historical or anecdotal considerations (e.g., the early colonization of Greenland, unusually bountiful agricultural yields and wine harvests in Europe early in the millennium, etc). Mann et al (2000) use a careful statistical analysis to show that the sparse regional information available earlier than AD 1400, while allowing for verifiable hemispheric temperature reconstructions back to about AD 1000, are associated with self-consistent estimates of uncertainties that are greatly expanded beyond those during more recent centuries.

These limitations notwithstanding, the best evidence, based on the extension of hemispheric climate reconstructions back a full millennium is that late 20th century conditions are probably warmer than those which prevailed any time this millennium, though conditions during the 11th thru 14th centuries appear warmer than those which prevailed during the 15th through 19th centuries in general. This conclusion is supported by independent estimates based on composites of modest numbers of Northern Hemisphere proxy records (Jones et al, 1998; Crowley and Lowery, 2000). The 19th century was particularly cold for both Europe and North America (the reader is referred to the regionally averaged temperature series for North America and Europe here). This period comes closest to being a truly "global" cold period (see Mann et al, 1999) although, as noted earlier, even in this case the cooling is not nearly as evident in the Southern Hemisphere.

To illustrate some of the problems inherent in estimating hemispheric mean temperature from limited regional information, consider (Figure 15) the warmest (1834), second warmest (1822), and coldest (1838) years in Europe prior to the 20th century, based on the MBH98 temperature pattern reconstructions.


1834 - Warmest Europe

1822 - Second Warmest Europe

1838 - Coldest Europe

Figure 15:
Global annual-mean temperature pattern reconstructions for three years associated with unusually warm or cold anomalies in the European sector during 1834 (top left), 1822 (top right), and 1838 (left).


While 1834 was the warmest year in Europe, it was colder than typical conditions (by 20th century standards) over large parts of the Northern hemisphere. This is especially true for 1822, the 2nd warmest year in Europe, but a cold year over most of the Northern Hemisphere. In contrast, the coldest year in Europe (1838) was indeed one of the coldest over much of the Northern hemisphere (see discussion below), but in fact, temperatures were nonetheless warm, relative to typical 20th century conditions, over significant portions of Greenland and Alaska. The coldest years in Europe might, by analogy with this example, have been quite unusually mild in Greenland, and a favorable opportunity for its colonization. It becomes readily evident from such examples (let alone, more careful statistical diagnostics) that inferences into hemispheric or global-scale temperature variations based on limited regional (e.g., European) information is perilous. In fact, the considerable low-frequency "noise" in the Atlantic and neighboring regions, due in large part to modes of ocean circulation variability (see e.g., Delworth and Mann, 2000), and the substantial overprint of the North Atlantic Oscillation on climate variations in this region during past centuries (see e.g., Cullen et al, 2000, Mann, 2000) particularly obscures hemispheric trends in this region. There is modeling evidence, in fact, that suggest that Medieval warmth was restricted to regions influenced by the North Atlantic (Overpeck, 1998). Statistically speaking, estimates of European temperature variability provide a very poor indication of large-scale temperature trends in past centuries, and should be strictly avoided for hemispheric, let-alone global-scale climate inferences.

Note that 1816 (the so-called "year without a summer"--Figure 16) in addition to appearing to have indeed been an especially cold summer (see Briffa et al, 1998) and a cold year for the NH temperature as a whole (though not anomalous relative to other years during that very cold decade), was an anomalously cold year only in Europe and parts of North America. In fact, conditions in the Middle and Near east were warmer than normal by 20th century standards.


1816 -

Figure 16:
Annual-mean Global temperature pattern reconstructions for the so-called "Year without a summer" 1816.

1816 - Warm

1816 - Cold

A number of other years (1870, 1864, 1838, 1820, 1700, 1642, and many years during the 1450s and 1460s decades) appear to have been substantially colder than 1816 for the hemisphere as a whole in the temperature reconstructions presented here. Our notions of this year as a particularly cold one may thus arise in large part from the fact that the coldness was most pronounced in those regions--Europe and North America--which figure most prominently in the western anecdotal and historical framework. The regional overprints of warming (e.g., in the Middle East) and extreme cold (e.g., Europe) which are superimposed on generally cold hemispheric conditions, in regions neighboring the North Atlantic may be attributed to the North Atlantic Oscillation or "NAO" (see Luterbacher et al, 1999; Cullen et al, 2000, Mann, 2000 for a discussion of inferences into past NAO-related climate variability). We believe that both the cold hemispheric conditions, and a strong NAO-like atmospheric circulation anomaly, were due to the explosive Tambora eruption in Indonesia during spring of 1815 (see MBH98). Seasonally-specific reconstructions of 1816 for the cold (Oct-Mar) and warm-half (Apr-Sep) years (see Figure 16) indicate that this pattern is clearer and more dominant during the cold-season, wherein the quadrapole pattern of warm and cold anomalies in continental regions bordering the North Atlantic is quite distinct. This is as expected, since the NAO is primarily, though not exclusively, a cold-season mode of atmospheric circulation variability. There is some evidence of the persistence of this pattern, albeit more weakly, into the warm-season. In particular, distinct cooling in the eastern United States and over much of Europe is clearly expressed during the warm-season, consistent with the notion of 1816 having been a "year without a summer" in those regions. Some of the coldness of the early 19th century might also be due to weakened solar irradiance forcing at that time. The possible influences of external climate forcings on hemispheric temperatures are discussed below.

On to... Influence of Climate Forcings