Temperature Variability on a continental scale over the Past 2 Millennia

Australia: The Land Where Time Began

Temperature Variability on a continental scale over the Past 2 Millennia

Changes of climate from the past had strong regional expression. The PAGES 2k Consortium reconstructed the past temperatures of 7 continental-scale regions for the past 1-2 millennia, finding that in almost all the regions they had reconstructed temperature for, the most common feature they showed was a long-term cooling trend that ended in the late 19th century. Temperature variability patterns are distinctly different from regional patterns at multi-decadal to centennial scales, with more similarity occurring in each hemisphere than between the hemispheres. The reconstructions showed no multi-decadal warm periods, such as the Medieval Warm Period, or cold intervals, such as the Little Ice Age, that were globally synchronous, though between AD 1580-1880 all reconstructions indicate conditions were generally cold, in some regions being punctuated by warm decades in the 18th century. The region where the transition to these cold conditions was first experienced was the Arctic, Europe and Asia, later affecting regions in North America and the Southern Hemisphere. Over the period ad 1971-2000 the long-term cooling trend was reversed by recent warming with the area-weighted average reconstructed temperature being higher than at any other time in almost 1,400 years.

The climate of the Earth has undergone significant climate variations that have not yet been quantified at the continental scale during the current interglacial period, where it is arguable that climate variability is more relevant to ecosystems and societies than conditions that have been globally averaged. In order to distinguish anthropogenic impacts from the background range of natural variability, determination of these ranges is necessary (Rockstrӧm et al., 2009). Reconstructing spatiotemporal patterns of climate variability of the past aids in gaining an understanding and quantifying the influence of externally forced and intrinsic dynamics of the global climate system (Snyder, 2010), and to understand natural climate variability, which needs to be considered in scenarios of future climate (Braconnot et al., 2012; Deser, Phillips, Bourdette, & Teng, 2012). In this paper the PAGES 2k Consortium presents a global data set of proxy records as well as temperature reconstructions for 7 regions on a continental scale. They describe the most prominent features of continental-scale temperature changes at multidecadal to millennial timescales. Contrasting to other recent reconstructions on a global scale (Mann et al., 2008; Mann et al., 2009), an intercontinental perspective is provided by this study of the evolution of temperature over the past 1-2 thousand years.

The aim of the ‘2k’ Network of the IGBP Past Global Changes (PAGES) project is to produce a global array of regional climate reconstructions for the past 2,000 years (www.pages-igbp.org/workinggroups/2k-network). There are 9 PAGES 2k working groups representing 8 continental scale regions and the oceans. Critical expert knowledge of proxy data sets is brought by regional representation, which is essential for improving reconstructions of palaeoclimates (Frank et al., 2010). The PAGES 2k Network is coordinated with the National Oceanic and Atmospheric Administration (NOAA) World Data Center for Climatology to establish a benchmark database of proxy climate records for the last 2 millennia (Supplementary Note A).

Reconstruction domains for the PAGES 2k regions that are reported here encompass 36 % of the surface of the earth. The average mean temperature averaged over these regions explains 90 % of the global mean annual temperature variability in the instrumental record (Supplementary Fig. S1), though the regions largely coincide with the continents and not the climatological criteria. At the present suitable proxy records from Africa are too sparse for a reliable temperature synthesis (Nicholson et al., 2013), and analysis of Palaeoceanographic data by the 2k Ocean2k group, that has been formed recently, is in progress (PAGES/Ocean2k Working Group). The proxy climate records that were found to be best suited for reconstructing annual or warm-season variability of temperature within each region that was found by groups in those regions, using criteria that are a priori established. There are 511 time series of tree rings, pollen, corrals, sediments of lake and marine environments, glacier ice, speleothems and historical documents that record changes in biological or physical processes that have been observed that can be used to reconstruct temperature variations.

The reconstructions by PAGES 2k have annual resolution, with the exception of North America. There are 3 of the temperature reconstructions that span approximately 2,000 years (Arctic, Europe and Antarctica), and 3 that cover 1,000-1,200 years (Asia, South America and Australasia). Included in the North American region is a shorter decadally resolved reconstruction based on tree rings, that goes back to AD 1200, and a longer reconstruction that is based on pollen that goes back to AD 360.

Each temperature reconstruction on a continental scale was derived by the use of different statistical methods, and each 2k Network group tailored their procedures to the strengths of their regional proxy records and calibration targets. A scaling approach to adjust the mean and variance of a predictor composite to an instrumental target was used by most groups, or to extract a common signal from the predictors used a technique that was regression based, to extract a common signal from the predictors by the use of principal components or distance weighting. Among the regional temperature reconstructions some of the heterogeneities might result from differences in methods of reconstruction, which may underestimate the amplitude of low frequency variability differently (Christiansen, Schmith & Thejll, 2009).

Also, differences between regional reconstructions may reflect differences in the targets of their reconstruction (‘land only’ or ‘land and ocean’), proxy types and proxy dating uncertainties. There is also a difference in temperature variability between summer and annual reconstructions, though in meteorological records from the study regions the 2 are correlated (mean r-value = 0.73 ± 0.08). Based on their statistical similarity to an instrumental target, as was done for most regional reconstructions, can hamper interpretations involving the comparing the temperature variability during the 20^th century to the variability of temperature of earlier periods (Esper & Frank, 2009). In order to assess the extent to which the reconstruction method choice might influence their conclusions, they also applied procedures that were uniform to the same proxy data in order to generate 3 additional reconstructions from each proxy region; referring to these as ‘alternative reconstructions’.

The analysis by the PAGES 2k consortium of multi-decadal variability focuses on 30 year mean temperatures (3 calendar decades) and their standardised values. The inclusion of 30-year-resolved reconstructions, which are based on pollen from North America is enabled by this temporal resolution. Regional difference in the magnitude of the temperature variability, which depends of geographical factors and can be influenced by seasonal biases in some proxies, are circumvented by the use of standardised values.

Millennial cooling

All regions show a long-term cooling trend over their respective record lengths that were followed in the 20^th century, except Antarctica, by recent warming. The cooling trend is significant, P < 0.05, prior to AD 1900, in all except North America, where there is a weakly significant trend, P < 0.10, in the tree ring reconstruction, and in the pollen reconstruction, not significant. The regional rate of cooling is variable between 0.1^oC and o.3^oC per 1,000 years. Since AD 1000 - the interval represented in all regions - is the trends are also significant, with the exception of Europe (P = 0.13). In general the overall trends in the 30-year-averaged Pages 2k Network reconstructions agree with those in the alternative reconstructions. They are also consistent with the global sea surface cooling that occurred from AD 1 to 1800 which is exhibited in the synthesis of PAGES Ocean2k (PAGES/Ocean2k Working Group).

In order to determine the extent to which the long term cooling trend is a common feature the individual site-level proxies were also analysed, and this approach is independent of the procedures of reconstruction. In general, the longer the proxy record, the more likely it will exhibit a long-term cooling record that is significant. E.g. 20 of the 24 records from all regions that extend back to AD 1, and 1900, have negative slopes between years AD 1 and 1900, which is extremely unlikely to occur by chance, 1-tailed sign test, P < 0.001. ¾ of these have slopes that are statistically significant (P < 0.05). Of the 90 records extending back to AD 1000, 54 have negative slopes (P < 0.03), and more than ⅓ of these are statistically significant (P < 0.05).

Temperature differences at multi-continental scales support the regression analysis of the cooling trend. High temperatures during the early centuries of the first millennium are shown by the 3 longest reconstructions, the Arctic, Europe and Antarctica, compared with those of the late (pre-20^th) centuries of the 2^nd millennium. During the last millennium commensurate cooling occurred; e.g. that first 4 centuries (AD 1000-1800) in all regions, in spite of cooling that begins between AD 1200 and 1300 in the Arctic, Europe and Asia. Between the 4-century intervals averaged over all of the regions was about 0.1^oC, which is consistent with a cooling trend (about -0.2^oC per 1,000 years).

The cause of cooling between the 10^th and 16^th centuries have been investigated by several studies, which suggest a potential role of solar irradiance, volcanic activity, changes of land cover and insolation that is driven by the orbit of the Earth (Mann et al., 2009; Jones & Mann, 2004; Crowley, 2000; Bauer, Claussen, Brovkin & Huenerbein, 2003; Hegerl et al., 2007). It was shown by an ensemble of simulations that were performed with a climate model that was of intermediate complexity that these 4 forcings are the dominant cause of the annual mean cooling trend simulated between AD 900 and AD 1850, with the relative contribution of each forcing varying among regions. It is believed the same forcings are likely to have played a similar role, though on a longer scale, that includes the entire first millennium. Some records of volcanic (Goosse et al., 2005) and solar (Steinhilber et al., 2012) activities, e.g., show significant negative forcing on the long term covering the last 2,000 years (Mann-Kendall trend test, P < 0.05). Also, in the mid to high latitudes of the Northern Hemisphere the millennial-scale cooling that occurred can be ascribed to a decrease in solar insolation that was orbitally driven, as has been suggested for the Arctic (Kaufman et al., 2009) and northern Scandinavia (Esper et al., 2012). A climate model of intermediate complexity simulates a multi-millennial cooling in summer in the Southern Hemisphere as a delayed response to the decrease in insolation in the local spring, modulated by the thermal inertia of the Southern Ocean (Renssen, et al.,. 205).

Variability – multi-decadal to centennial

Temperature was not uniform among all regions, which highlights the evolution of temperature at multidecadal to centennial time scales that was regionally specific. However, a sustained warm interval was generally encompassed in all 4 Northern Hemisphere regions, over the period from about AD 830 to 1100. A sustained warm period occurred Later in South America and Australasia, from around AD 1160 to 1370. Temperatures were generally high in the Arctic and Europe in the first centuries AD. Most other reconstructions are too short to infer temperatures prior to around AD 1000.

Between AD 1200 and 1500 the transition to colder regional climates is evident earlier in the Arctic, Europe and Asia than in North America or the Southern Hemisphere. Non-forced variations that involved the major modes of atmospheric variability (Fernández-Donado et al., 2013; Goosse et al., 2012) could be reflected in differences among regions. All regions with the exception of Antarctica entered a protracted cold period by around AD 1580. Cold conditions prevailed until late in the 19^th century, with the exception of intervals or regional warmth, especially during the 18^th century. Again, the alternative reconstructions show patterns similar to those of the PAGES 2k regional reconstructions.

Palaeoclimatic records that span the past millennium are often characterised as including some manifestation of a warm Medieval Warm Period (MWP) that was followed by a cool Little Ice Age (LIA) (Goosse et al., 2012). A tendency for temperature anomalies on a centennial scale have been shown by previous reviews of these intervals, though they have also emphasised their heterogeneity through space and time (Mann et al., 2009; Bradley, Hughes & Diaz, 2003; Matthews & Briffa, 2005; Ljungqvist et al., 2012; Diaz et al., 2011). The regional temperature reconstructions that have been presented in this paper also show little evidence of multidecadal shifts that are synchronised globally that would mark MWP and LIA intervals that are well defined. Instead, the specific timing of peak warm and cold intervals varies regionally, with the result that multidecadal variability in temperature departures from an underlying global cooling trend that are regionally specific.

The regional temperature reconstructions presented in this paper can be compared with solar irradiance and volcanic activity records to examine the extent to which changes in temperature coincide with these climate forcings. The time series of area-weighted, standardised 30-year-mean averaged temperatures across all regions displays broad similarities with climate forcings records, a feature that was also apparent in reconstructions of average temperature for the Northern Hemisphere that had been published previously. Between 1251 and 1820 9 30-year periods when volcanic-solar forcing was at its most negative in 5 distinct 30- to 90-year intervals, as was determined by using criteria that are specified in methods, generally correspond to a decrease in average temperature. The averaged temperature during these perturbed intervals was shown by Monte Carlo sampling of the detrended global time series to be significantly lower (P < 0.01) than the means of 9 30-year periods between AD 830 and AD 1910. These volcanic- and solar-perturbed periods includes the coolest 30-year period of each reconstruction, though in Australasia and North America at the regional scale the response appears to lag by 1 30-year period during the 19^th century, when there was a cluster of volcanic eruptions between AD 1823 and 1835 (Crowley, 2000), and Antarctica was at its coldest in the middle of the 20^th century. During each of the strong solar-volcanic downturns not all regions cooled. Tree rings in Australasia, North America and South America do not show multidecadal cooling during the earliest interval of strong negative forcing (AD 1251-1310).

Reconstructed temperature in the 20^th century

The 20^th century was the warmest or nearly the warmest century in all regions, with the exception of Antarctica, where warming may have been dampened by the large thermal inertia of the surrounding ocean (Stouffer, Manabe & Bryan, 1989). When Antarctica is excluded, the average temperature among the 6 regions was about 0.4^oC higher than the averaged temperatures of the preceding 500 years. When compared with the preceding 500 years, warming in the 20^th century for the 4 regions in the Northern Hemisphere was, on average, about twice that of the Australasia and South America, that are strongly ocean-dominated regions (about 0.5^oC compared to 0.2^oC), and the greatest differences are at northern high latitudes. In the Arctic 20^th century warming (0.9^oC) was about 3 times the average of the other 5 non-polar regions.

The best estimate obtained by this study of reconstructed temperature for AD 1971-2000 can be compared with all other consecutive 30-year periods within each regional reconstruction. Reconstructed temperature was higher during 1971-2000 in Asia and Australasia than other 30-year periods. Temperatures were warmest around the Arctic around AD 935, but the second warmest was the 20^th century with the 1941-1970 period being warmer than 1971-2000. In South America, the reconstructed temperature for AD 1971-2000 was similar to the record maximum in AD 1251-1280. The reconstructed temperature for the 1971-2000 interval in North America does include the warm decades since 1980, and therefore underestimates the actual temperature for that interval. Slightly higher reconstructed temperatures were registered in Europe in AD 741-770, and the interval from AD 21-80 was substantially warmer than 1971-2000. It is probable that Antarctica was warmer than 1971-2000 for a time period as recent as AD 1671-1700, and the entire period from 141-1250 was warmer than 1971-2000. The relative magnitude of recent warming in the alternative reconstructions generally supports these interpretations.

This study involved analysing each individual proxy record that contributed to the regional reconstructions in order to evaluate whether the values during 1971-2000 indicate temperatures were higher for that period than for any other 30-year period, independent of the procedures that were used to calibrate the temperature reconstructions. According to this analysis, more sites in the 323 individual proxy records that extend to AD 1500 seem warmest during the 1971-2000 period than any other 30-year period, both in terms of the total number of sites and their proportion in each region. Similarly, of the 52 records that extend to AD 500, more sites, as well as a higher proportion, seem warmest during the 20^th century than during any other century. The fraction of individual records indicating the highest temperatures during 1971-2000 decreases with decreasing length, which is consistent with the overall cooling trend over the past 2 millennia.

It is indicated by the area weighted average of the best estimate of past temperature derived from all 7 regions that 1971-2000 was warmer than any other time in nearly 1,400 years, though this analysis does not consider the uncertainty that is associated with the temperature measurements, and that the length of the reconstructions vary. Area-weighted averages of the 3 alternative reconstructions generally support this result. There are large uncertainties that remain, especially during the first millennium, when only some regions are represented. Regardless, a persistent long term global cooling trend has been reversed by the global warming that has occurred since the end of the 19^th century. Between the 19^th and 20^th centuries the increase in the average temperature exceeded the temperature differences between all other consecutive centuries in each region, with the exception of Antarctica and South America.

Clear regional expressions of variability of temperature at multidecadal to centennial scales are shown by PAGES 2k Network reconstructions, while a long term cooling trend prior to the 20^th century is evident globally. Temperature changes on a centennial scale in Australasia and South America generally follow those of the Northern Hemisphere; though in Antarctica temperature changes do not correlate with those of other regions. The long-term global trend prior to the 20^th century is evident in differences in average temperature between multicentennial periods, as well as in slopes of least-squares linear regressions for both the regional reconstructions and the individual site-level proxy records. In this study the compilation of the proxy data and reconstructions will be useful in studies in the future by serving as a benchmark for comparisons with simulations of climate models that are aimed at understanding the cause of global cooling, and the extent to which temperature fluctuations and trends at the continental scale can be explained by externally forced and unforced variability.

Relation between proxy records and temperature

The sign, either positive of negative, of the relation between each proxy record and temperature is listed in Supplementary Database S1. Different approaches for different regions were used to determine these relations. The relation in the Arctic was adopted from the original records that were published. The 11 inputs to the regional reconstruction in Europe include 10 tree ring composites and documentary evidence, that were all calibrated to temperature. Regional syntheses that were previously transformed into temperature were also relied upon for the North American pollen-based reconstruction. For Asia and Antarctica, only records that had demonstrated positive correlations with local or regional (respectively) instrumental data were included in the reconstructions. The Asia, North American tree ring, South American and Australasia reconstructions were based on methods of principal regression methods, where individual proxy records could contribute with either a positive or negative temperature relation with different principal components, or for ensemble members with different instrumental targets. The sign of the North American tree-ring reconstruction for each record could vary spatially as well. For these regional reconstructions that are listed in Database S1the sign of the relationship with temperature is based on the direction of the relation between the proxy records and the area-weighted average mean annual temperature from the domain areas in Fig. 1 using the HadCRUT4 (ref. 29) data series.

Significance of trends

The significance of the cooling trends on the long term in the reconstructions and in the individual site level proxy records was calculated at the P = 0.05 confidence level proxy records by using the 1-sided Student’s t-test, and accounting for the lag-1 autocorrelation in the calculation of standard error as well as indexing the t-values (Santer et al., 2000). In the individual proxy records (site level) trends were determined by first inverting those proxy records that had a negative correlation with temperature (as described above). In each record the pre-1900 trend was analysed for successively shorter segments of time, starting with the entire time series, then truncating each series by 30-year intervals, each segment ending in AD 1900. In each truncated series the trend was determined by the slope of its linear regression.

In order to determine the strongest agreement and magnitude of forcing 2 time series of solar forcing and 2 of volcanic forcing from ref. 31were analysed, the volcanic series from ref. 16, and the solar forcing from ref. 17. Each time series was binned into 30-year intervals that coincide with the temperature reconstructions (bin centres from AD 845-1985). The 8 bins, comprising 21 % of each record, with the most negative forcing was selected from each series; then the subset of bins that were overlapping among at least half (≥3) of 6 time series was identified. The 9 30-year bins that were defined in this way comprise 5 intervals that were termed ‘volcanic-solar downturns’: AD 1251-1310, 1431-1520, 1581-1610, 1641-1700 and 1791-1820.

Temperatures during volcanic-solar downturns

In order to evaluate whether temperatures were significantly lower during intervals of negative climate forcing, a Monte Carlo procedure was implemented to sample randomly the regional temperatures that were area-weighted between AD 831 and 1910, the warm 20^th century from AD 1911-2000, was excluded to make the test more conservative. The mean temperature of 9 30-year intervals was calculated for each of 10,000 iterations. The lower temperatures during these intervals may be in part due to long-term cooling as most of the intervals of negative forcing occur later in the record. The test was also conducted on the detrended area-weighted mean to assess this possibility. The mean of the 9 intervals of negative solar and volcanic forcing for both the original and detrended tests was lower than more than 99 % of the means of intervals that had been selected randomly, i.e., P < 0.01.

Warmest intervals with individual proxy records

Site-level warmest-interval analysis was carried out on 30-year intervals that spanned from AD 1491 to 2000 and for 100-year intervals between AD 501 and 2000. Proxy records that had negative correlation with temperature were inverted. Only records with at least 1 measurement in each of the 30-year or 100-year bins were used. The 3 tree-ring records from northern Fennoscandia, which had been used in both the Arctic and Europe reconstructions, were included only once in the site-level analysis. Ranking the values of 30- or 100-year bins for each record was used to identify the warmest intervals. The values were presented as the fraction of records in each region, instead of the number of records, in order to better represent regions with fewer records. The fractions were scaled in such a way that a value of 1 corresponds to all records in all regions.

Sources & Further reading