• Open Access

Air temperature variations on the Atlantic-Arctic boundary since 1802



[1] A two-hundred year instrumental record of annual surface air temperature in the Atlantic-Arctic boundary region was reconstructed from four station-based composite time series. The new time series adds seventy-six years to the previously available record. Credibility is supported by ice core records, other temperature proxies, and historical evidence. This record provides new perspective on past climate fluctuations in a region where pivotal climate system processes occur and where unexplained low-frequency variations were observed during the 20th century. It is well correlated with sea-surface temperature anomalies, both regionally and in the vicinity of the North Atlantic western boundary current and southern recirculation gyre. The extended instrumental record reveals an irregular pattern of decadal-scale temperature fluctuations over the past two centuries. The early 20th century warming event is by far the most striking historical example.

1. Introduction

[2] Surface air temperature (SAT) records for the Atlantic-Arctic boundary region reflect a curious low-frequency pattern of climate variation in the 20th century. This pattern is defined by two distinct periods of warming with an interval of somewhat cooler temperatures between. The first SAT increase, associated with the early twentieth-century warming (ETCW) event, began about 1920 and persisted until mid-century. Positive SAT anomalies were largest in the Atlantic-Arctic region during this period. The second is the present and more widespread warming mainly attributed to anthropogenic forcing. Both warming periods are linked to considerable social and environmental impacts.

[3] An explanation for this apparent low-frequency variability in the regional SAT record has been elusive. The problem is amplified because the cause of the ETCW, a major feature in the record, is unclear [Brönnimann, 2009]. Several recent studies suggest that the ETCW could be largely a product of intrinsic variability within the climate system [e.g., Bengtsson et al., 2004; Wang et al., 2009]. The cooler interval from ∼1950 may be related to higher emissions of anthropogenic aerosols [Shindell and Faluvegi, 2009]. Others propose that the ETCW is part of an underlying multidecadal pattern connected to the meridional overturning circulation (MOC) of the North Atlantic [Chylek et al., 2009; Polyakov et al., 2009]. Resolution of the question is confounded by a lack of long high-quality temperature records.

[4] Here we present an extended instrumental SAT record (designated TNA) for the Atlantic-Arctic boundary region for the period 1802 – 2009 reconstructed from a sparse network of instrumental observations (Figure 1). This record adds seventy-six years to the previously available instrumental SAT record. TNA is verified by comparison with regional ice core records and other local temperature proxies. It is well correlated with sea-surface temperature (SST) anomalies, both regionally and in the vicinity of the mid-latitude western boundary current (WBC) in the North Atlantic, and with variations in sea ice extent. SAT fluctuations in the record are also consistent with historical accounts.

Figure 1.

Extended annual mean SAT record for the Atlantic-Arctic boundary region (TNA). 95% confidence limits are shown. Decadal-scale variations are emphasized with a 2-way Butterworth low-pass filter constructed to remove frequencies higher than 0.1 cycles year−1 (bold black line). The early 20th century warming (ETCW) episode is marked. Regions represented by station-based composite SAT records used in TNA are indicated in the map.

2. Data and Methods

[5] TNA is a weighted annual average of four composite time series distributed across the Atlantic-Arctic boundary region, as shown in the map in Figure 1. Records come from recently published [Klingbjer and Moberg, 2003; Vinther et al., 2006] and historical sources [Wahlén, 1886]. Weight coefficients were derived from a multiple linear regression onto a modern land-based SAT reference (CRUTEM3v [Brohan et al., 2006]) for the same region (north of 60° N – 60° W to 45° E). 1950–1979 was selected for the calibration period to avoid gaps in Russian data in the 1980s. The resulting formula is:

equation image

where T1 = S.W. Greenland, T2 = Iceland, T3 = Tornedalen (Sweden), and T4 = Arkhangel'sk (Russia). Gaps in T1 result in 8 missing annual values between 1803 and 1828. In the early part of the record (1802–1851) there are 19 gaps caused by missing monthly observations (predominantly May–September). These gaps were filled using a linear model derived from complete data, which for consistency was applied to the entire record. Correlation between estimated and observed T1 is 0.98 (correlations reported throughout refer to annual data unless otherwise noted). There are ten years with missing values in either T3 or T4. These are 1807–1812, 1815, 2003, 2004 and 2009. These gaps are filled by using only one of the time series in the computation after adjusting coefficients. Additional data from Nuuk and Aasiaat (Greenland) and Haparanda (Sweden) are used to bring TNA up to date (2005–2009). See auxiliary material for details.

[6] Performance of TNA was evaluated by comparison with annual land-based SAT from CRUTEM3v for the overlapping period (1873–2009). Correlation between the two data sets was 0.96 for the calibration period (1950–1979). Split calibration and verification correlations (1950–1964 and 1965–1979) are also high (r ≥ 0.95). The minimum 30-year correlation was 0.88 (1925–1954), likely due to disruption of the larger network during World War II. Correlation over the 67 years prior to 1940 is 0.97. Additional discussion and analysis of data quality is provided in the auxiliary material. TNA and unpublished subsidiary data are supplied in Tables S2S4.

3. Results

[7] TNA is a credible instrumental record of annual SAT for the Atlantic-Arctic boundary region for the period 1802 – 2009. The record shows that decadal-scale SAT fluctuations have occurred in this region over the past two centuries at irregular intervals. These fluctuations are superimposed on a warming trend of 0.062°C decade−1. The major contributor to the trend is the SAT increase beginning in the early 20th century. This trend is consistent with other studies.

[8] It is often impossible to validate SAT reconstructions beyond the limited range of data typically available for calibration and verification. General reliability must be assumed. Here the credibility of the extended TNA record is independently supported by comparison with a regional mean stable isotope (δ18O) record computed using three annually resolved ice core records. These are from Greenland (GRIP/GISP2 stack) [White et al., 1997], Svalbard (Austfonna, 79° 50 N – 24° 01 E) [Isaksson et al., 2005] and Severnaya Zemlya (Akademii Nauk ice cap, 80° 31 N – 94° 49 E) [Fritzsche et al., 2005; Opel et al., 2009]. The averaging suppresses uncorrelated noise in the individual records [e.g., Vinther et al., 2008] and ensures consistency with the regional definition of TNA. The TNA time series and mean δ18O are well correlated (Figures 2a and 2b and Table 1). Maximum correlation occurs at zero lag and trend differences cannot be distinguished from the null hypothesis.

Figure 2.

Concurrent variations in multiple climate records. (a) TNA. (b) Mean δ18O. (c) SST0′. (d) Teigarhorn SST′. (e) Mean sea ice index (inverted). (f) Koch (Iceland) sea ice index [Wallevik and Sigurjónsson, 1998]. A low-pass filter was applied as in Figure 1 (bold black lines). Notable deflections are marked with dashed gray lines. The term ‘ice years’ refers to the period in the 1960s when sea ice conditions in Iceland were unusually severe by 20th century standards.

Table 1. Correlation Between TNA and Other Independently Collected Variables for the Mutually Overlapping Period 1886–1985
Correlationδ18OSST0Teigarhorn SST′Sea Ice

[9] The credibility and broader climate significance of the TNA record can also be inferred from its association with variations in SST anomaly and sea ice extent (Figures 2c2f and Table 1). More descriptive historical records also provide qualitative evidence that fluctuations in TNA are associated with expected environmental impacts.

[10] TNA is well correlated with annual SST anomaly, both in the mid-latitude western North Atlantic Ocean (SST0′, centered at 38° 30 N – 65° W) [Rayner et al., 2006], and regionally, as represented by an eastern Iceland SST time series (Teigarhorn, 64° 41 N – 14° 21 W).

[11] The TNA record also closely tracks variations in sea ice extent. Figure 2e shows the mean of four historical sea ice records for the Atlantic-Arctic region: Newfoundland ice extent [Hill, 2008], the Storis index [Rosing-Asvid, 2006], and two Nordic Seas indexes [Vinje, 2001]. Averaging is justified for the same reasons given above, and here also serves to dampen noise related to the wind-driven displacement of the marginal ice zone associated with the winter atmospheric circulation (i.e., the North Atlantic Oscillation) evident in winter-spring sea ice indexes (i.e., the Newfoundland and Nordic Seas indexes). Qualitative agreement between TNA and the Koch sea ice index is also noteworthy, especially the minima ∼1840–55 and during the ETCW, and the maximum in the 1960s (the “ice years” in Iceland).

[12] While difficult to assess quantitatively, descriptive accounts provide secondary evidence that TNA is a good reflection of regional climate history. Concurrent change in the abundance of North Atlantic fish stocks is an example. There were two periods in the early 19th century when cod was reported to be especially abundant in western Greenland: the 1820s and 1840s [Jensen, 1939]. A small commercial fishery emerged in the latter period [Buch et al., 1994]. The 1830s were colder and cod fishing was poor. A similar pattern has been described with respect to both the cod fishery of Iceland and annual temperature estimated from the prevalence of sea ice on the Icelandic coast [Jónsson, 1994]. The regional environmental impacts of the ETCW and the ‘ice years’ are well documented.

[13] Correlation of TNA with the historical North Atlantic SST anomaly field [Rayner et al., 2003] yields the spatial pattern shown in Figure 3a. Significant correlations occur near high surface heat flux regions associated with the WBC (Figure 3b) and extend into the area of the southern recirculation gyre around the Sargasso Sea. Decadal-scale fluctuations in TNA and SST0′ are nearly synchronous (Figure 3c). TNA shows a different pattern of climate variability than several other indexes of mean North Atlantic SST anomaly [e.g., Enfield et al., 2001; Gray et al., 2004] (Figure 3d).

Figure 3.

(a) Spatial correlation pattern in annual North Atlantic SST anomaly [Rayner et al., 2003] associated with TNA (1870–2004). Mean SST contours (16–22°C) are shown in gray. Location of SST0′ is indicated (box). (b) Mean annual net surface heat flux from the NCEP-NCAR reanalysis [after Rhines et al., 2008]. (c) TNA (black) and SST0′ (blue) detrended and filtered as in Figure 1. (d) TNA (black) and reconstructed mean North Atlantic SST anomaly (green) [Gray et al., 2004].

4. Discussion

[14] A face-value interpretation of the TNA record and its apparent association with multiple climate-relevant variables adds considerably to the stock of information on the climate history of the Atlantic-Arctic boundary region. It provides added perspective on pivotal atmosphere-ocean-cryosphere interactions that occur in the region and that have far-reaching influence on the climate system. Evidence of a strong teleconnection between TNA and SST0′ is relevant to the ongoing debate on the role of the Atlantic WBC and MOC [e.g., Rhines et al., 2008]. On a more basic level, the TNA record provides a longer and more nuanced historical context in which to place recent observations.

[15] Low-frequency climate variation was conspicuous during the 20th century, and the ETCW was a major component of this pattern. TNA shows weaker decadal-scale fluctuations occurred in irregular episodes during the 19th century; none were as distinctive as the ETCW in terms of persistent SAT change. Since lesser fluctuations are recorded in independently collected data sets it is improbable that an earlier event on the scale of the ETCW would escape notice, even given the imperfect methods and instrumentation of the early 19th century, nor is it likely that the small amplitude trend in the data related to increasing anthropogenic forcing masked any previous events of this nature. Thus, the low-frequency pattern of the 20th century does not appear to have a clear analog in the previous century. We see no obvious evidence of a regular climate oscillation as a leading mode of variability reflected in TNA.

[16] The teleconnection between TNA and SST0′ offers an opportunity to reframe the problem of low-frequency variability in the North Atlantic. We have shown that TNA and SST0′ are well correlated and closely related in time (Figure 3c and Table 1). Indexes of mean North Atlantic SST, on the other hand, lag the onset of the ETCW by up to six years and are out of phase with TNA during the 19th century. By constraining the problem to the potential mechanisms that could govern the low-frequency behavior of both TNA and SST0′ interpretation becomes more tractable to theory and empirical investigation [e.g., Kwon et al., 2010].

[17] An alternative explanation for the observed low-frequency pattern can be sketched as follows. TNA and SST0′ are both located in geographically limited regions of strong horizontal temperature gradient. Advection is an important process that regulates the transfer and storage of heat in both regions. This is because heat flux is a strong function of the air-sea temperature contrast [Kelly et al., 2010] produced by oceanic and atmospheric advection. Positive fluctuations in both TNA (and by extension other variables associated with it) and SST0′ might be initiated by persistent variations in the large-scale atmospheric circulation that promote the advection of warm maritime air into the Atlantic-Arctic region and simultaneously limit flux-induced cooling in the Nordic Seas and the WBC/recirculation gyre region. Unusual atmospheric circulation patterns certainly occurred during the ETCW and included a meridional component that enhanced southerly advection into the Arctic [Bjerknes, 1959; Petterssen, 1949]. The source of decadal-scale memory must be within the ocean, perhaps due to a ‘reddening’ of stochastic atmospheric variability [e.g., Frankignoul and Hasselmann, 1977] leading to enhanced heat storage in the Nordic Seas and in the southern recirculation gyre [Kelly et al., 2010]. Anomalous atmospheric circulation patterns may be reinforced as a consequence [e.g., Minobe et al., 2008].

[18] As for the future, with no other examples in the record quite like the ETCW we cannot easily suggest how often—much less when—such a comparably large regional climate fluctuation might be expected to appear. Extrapolating the 20th century pattern is unlikely to produce useful insight. Correlation between TNA and mean δ18O suggests more perspective may come with the prospective extension of the latter record by ∼500 years. How anthropogenic forcing will affect the historical pattern in the future is unknown, but if past is prologue it would be reasonable to expect substantial regional climate fluctuations of either sign to appear from time to time.

5. Summary

[19] TNA is a credible instrumental record of annual SAT in the Atlantic-Arctic boundary region for the period 1802–2009. An irregular pattern of decadal-scale SAT fluctuations occurred over the past two centuries. Near simultaneous low-frequency variations in both TNA and SST0′ may be initiated by fluctuations in the atmospheric general circulation coupled to regional ocean heat transport and storage processes. We see no obvious evidence of a regular climate oscillation. Singular episodes of regional climate fluctuation should be anticipated in the future.


[20] This project was supported by Arctic Research of the NOAA Climate Program Office, by JISAO under NOAA Cooperative Agreement NA17RJ1232, through the Arctic System Science Program (NSF grant 0531286), and in part by NSF grant ATM-0812802 (BVS). JISAO contribution 1784. PMEL contribution 3492.