Is a bipolar seesaw consistent with observed Antarctic climate variability and trends?


  • David P. Schneider,

    1. Climate and Global Dynamics Division, Earth System Laboratory, National Center for Atmospheric Research, Boulder, Colorado, USA
    2. Department of Atmospheric and Oceanic Sciences and Cooperative Institute for Research in Environmental Sciences, University of Colorado at Boulder, Boulder, Colorado, USA
    Search for more papers by this author
  • David C. Noone

    1. Department of Atmospheric and Oceanic Sciences and Cooperative Institute for Research in Environmental Sciences, University of Colorado at Boulder, Boulder, Colorado, USA
    Search for more papers by this author


[1] A bipolar seesaw of Arctic and Antarctic temperature anomalies has been reported to be evident in instrumental data on decadal timescales during the last century. This finding hinges upon a global temperature data set that for the area poleward of ∼60°S is derived from only one sub-Antarctic station prior to the mid-1940s, and does not include a substantial number of Antarctic stations until the late 1950s. The timeseries of the single-station record for the early period spliced to the data based on broader coverage for the latter period is an artificial estimate of the Antarctic climate trend and its variability. We estimate the real variability using the original timeseries from the sub-Antarctic station, a reconstruction of the Southern Annular Mode index, and an ice-core based reconstruction of Antarctic temperature. None of these Antarctic timeseries are significantly correlated with Arctic or North Atlantic climate records, nor with the index of the Atlantic Multidecadal Oscillation, which was proposed as the driving mechanism of the seesaw. Instead, each of these records is consistently correlated with tropical Pacific sea surface temperatures. However, neither the seesaw nor the tropics alone can fully capture the complexity of Antarctic climate variability and climate change.

1. Introduction

[2] In the context of a warming global climate, whether warming in Antarctica during recent decades has been suppressed by internal variability, perhaps originating from the tropics [e.g., Bertler et al., 2004] or from the North Atlantic [Chylek et al., 2010], is an important question with implications for predicting the rate of 21st Century Antarctic warming and its potential impacts, including sea level rise and changing surface albedo. Thus it is important that we identify a recent error in the application of a global, gridded temperature data set to the question of a bipolar seesaw of Arctic and Antarctic temperature anomalies [Chylek et al., 2010].

[3] Several studies have suggested that significant warming has occurred over Antarctica as a whole since the late 1950s at a rate of about 0.1°C/decade [e.g., Jacka et al., 2004; Steig et al., 2009; Schneider et al., 2012a]. Since 1979, Antarctic temperature trends are essentially zero, though there are notable regional exceptions, including strong warming in the Peninsula and West Antarctica [Schneider et al., 2012a]. The mixed signals from Antarctica stand in sharp contrast to the rapid warming of nearly the entire Arctic in recent decades [e.g., Turner and Overland, 2009]. Could a bipolar seesaw explain this contrast? If not, what other mechanisms of decadal scale variability could amplify or suppress regional Antarctic warming?

[4] Here we re-evaluate the seesaw hypothesis using both the data set employed byChylek et al. [2010] and alternative data. We discuss the evidence for other origins of decadal scale variability in Antarctica, namely the tropics. Although we will focus on patterns of climate variability, it is important to recognize that both geographic and external factors have been proposed as moderating influences on Antarctic climate change [e.g., Turner and Overland, 2009]. The vast Southern Ocean absorbs and sequesters heat and may be limiting warming of the overlying atmosphere. Additionally, stratospheric ozone depletion has been most pronounced in the Antarctic and has led to a strengthening of the polar vortex, limiting meridional atmospheric heat transport to the Antarctic. However, ozone depletion's main effects on the surface climate occur in austral summer [e.g., Thompson et al., 2011] and certainly do not explain all aspects of Antarctic temperature trends [e.g., Schneider et al., 2012a]. Thus, other explanations for Antarctic climate change require consideration.

2. Antarctic Data

[5] The Chylek et al. [2010] results rely on the annual mean Antarctic surface air temperature (SAT) timeseries from the Goddard Institute for Space Studies temperature analysis (GISTEMP) [Hansen et al., 2010]. We use a 2° × 2° degree version of GISTEMP, constructed from 1200 km smoothing of land air temperature and ocean sea surface temperature (SST) anomalies. We apply cosine of latitude weighting to compute timeseries representing the area-average of the domains 64°N–90°N (Arctic) and 64°S–90°S (Antarctic). GISTEMP incorporates Antarctic observations from the Reference Antarctic Data for Environmental Research (READER) archive [Turner et al., 2004]. A key record in READER is from Orcadas station (60.7°S, 44.7°W) for which the years 1904–2008 are complete except for 2002 and 2003, which we fill in with climatology (anomalies of zero with respect to 1961–1990).

[6] In addition to Orcadas, another relevant long instrumental timeseries is the reconstructed annual mean Southern Annular Mode (SAM) index [Visbeck, 2009]. These timeseries both have the drawback that they do not include true Antarctic data (south of ∼65°S). Key proxy data come from ice cores; we use a reconstruction of Antarctic temperatures [Schneider et al., 2006], hereafter called ICECORE, which is representative only of the main part of the Antarctic continent, excluding the Peninsula and Orcadas region.

[7] To assess the representativeness of the GISTEMP data, we compare it with ICECORE and with three spatially complete Antarctic temperature reconstructions for the late 1950s to late 2000s period [Monaghan et al., 2008; Steig et al., 2009; O'Donnell et al., 2011].

3. Data Characterizing North Atlantic and Tropical Variability

[8] To characterize variability associated with the North Atlantic, we use the Atlantic Multidecadal Oscillation (AMO) index of Trenberth and Shea [2006]. The AMO is commonly defined as the average of detrended SST anomalies in the North Atlantic basin and a quasi-periodicity of roughly 50–70 years [Delworth and Mann, 2000; Trenberth and Shea, 2006; Ting et al., 2009]. The distinction of the Trenberth and Shea [2006] index is that the global mean SST anomaly timeseries was linearly subtracted from the SST data before computing the average over the North Atlantic. We also use a timeseries of North Atlantic regional SAT from land stations [Wood et al., 2010].

[9] To characterize tropical variability, we use an “optimal tropical index” (OTI) [Deser et al., 2004], a multivariate composite of SST anomalies in the Indian and southeastern tropical Pacific oceans, rainfall anomalies in the South Pacific Convergence Zone, cloudiness in the central equatorial Pacific, and the difference in sea level pressure (SLP) anomalies between the southeastern tropical Pacific and south Indian Oceans. The OTI is relevant to atmospheric teleconnections in the extratropics associated with low-frequency ‘ENSO-like’ variability [Garreaud and Battisti, 1999] and prominent regime shifts of Pacific climate in the 20th Century [Deser et al., 2004]. The index of the Pacific Decadal Oscillation (PDO) [Mantua et al., 1997] is also used.

4. Global Gridded Data and Methods

[10] The global gridded SST and SLP data sets that we use included the NOAA Extended Reconstruction (ERSSTv3b) [Smith et al., 2008] and the Hadley Centre SLP version 2 [Allan and Ansell, 2006]. All of our data sources and methods used are described in the auxiliary material.

5. Results

[11] The timeseries of annual average Antarctic temperature anomalies from GISTEMP (Figure 1a) exhibits a significant long-term upward trend and more variance before ∼1960 than after. A first indication of a problem, the linear trend of the Antarctic timeseries is greater than that of the Arctic timeseries for the same dataset [see alsoChylek et al., 2010, Figure 1]. From 1904 to 1943, the Antarctic timeseries is positively correlated at r > 0.99 (p < 0.05) with the single sub-Antarctic station record from Orcadas and from 1958–2008 it is negatively correlated with Orcadas (r = −0.36 detrended, p < 0.05). Over the latter period, it is significantly positively correlated (p < 0.05) with Antarctic-average temperature anomalies fromMonaghan et al. [2008], Steig et al. [2009], and O'Donnell et al. [2011], at r = 0.92, r = 0.78 and r = 0.93, respectively. The GISTEMP timeseries is also significantly correlated (r = 0.58, p < 0.05) with the ICECORE series for 1958–1999. Thus the analysis captures variability near Orcadas only for 1904–1943 and provides a reasonable estimate of Antarctic-wide temperatures for 1958 to present. During the intervening period 1944–1957, a number of additional stations began reporting, but a substantial number did not exist until the 1957–1958 International Geophysical Year.

Figure 1.

(a) Annual mean temperature anomalies for 1904–2008 for the GISTEMP Antarctic record (blue line), the observed Orcadas station record (black line) and the Monaghan et al. [2008]Antarctic land temperatures (red line). (b) Timeseries of the GISTEMP Antarctic and Arctic records, the Orcadas temperature record, and ICECORE. The correlations in the legend indicate the correlations of the timeseries with the Arctic and Antarctic GISTEMP records, respectively. (c) Low-pass filtered (cutoff of ∼10 years) and linearly detrended timeseries of the AMO index, the GISTEMP Arctic record, and the Arctic Atlantic sector record described in the text. (d) As in Figure 1c, but for timeseries of the OTI, PDO index, SAM index, and the ICECORE, Arctic Pacific sector, and Orcadas records.

[12] In considering the appropriateness of the GISTEMP Antarctic record for evaluating the bipolar seesaw, it is important to recognize that strong dipoles of temperature anomalies exist within Antarctica, arising from SAM and ENSO-related variability [e.g.,Kwok and Comiso, 2002; Bromwich et al., 2004; Marshall, 2007; Yuan and Li, 2008]. It has long been argued that the record from Orcadas is not representative of the Antarctic as a whole [Raper et al., 1984] and shown that anomalies at Orcadas tend to be anti-correlated with anomalies on the continent [e.g.,Schneider et al., 2006]. In the case of GISTEMP, the correlations above indicate that the Antarctic timeseries calculated by Chylek et al. [2010]and reproduced by us essentially splices Orcadas temperature anomalies for the first part of the 20th Century with broader-scale, Antarctic-wide temperature anomalies for the late 1950s to present. This timeseries is therefore dominated by the changing Antarctic data coverage and is not physically based. The changing availability of Antarctic observations is acknowledged by the developers of the GISTEMP analysis [Hansen et al., 2006, 2010].

[13] We can essentially reproduce the reported bipolar seesaw (Figure 1b), however it is largely an artifact in the context of the GISTEMP analysis. Considering other, continuous Antarctic records, the Orcadas record has a weaker anti-correlation with the Arctic timeseries than the century-length GISTEMP record and a strong anti-correlation (r = −0.79) with the Antarctic ICECORE record, while the ICECORE record is actually positively correlated with the Arctic instrumental record (Figure 1b). Although the Arctic timeseries could also have sampling issues, the GISTEMP Arctic timeseries is well correlated with other published Arctic temperature timeseries [e.g., Johannessen et al., 2004]. Similar results are obtained using the consistently sampled North Atlantic region SAT record of Wood et al. [2010], confirming that any deficiencies in the GISTEMP Arctic data have little impacts on the findings here.

[14] While the Arctic temperature record has a high correlation with the AMO index, the Antarctic ICECORE and Orcadas records have very small correlations (Table 1). Additionally, linearly detrended annual mean values of the Orcadas SAT record and North Atlantic regional SAT from land stations [Wood et al., 2010] are not significantly correlated over 1904–2008 (r = −0.19). Thus, evidence for the seesaw is considerably weakened when alternative Antarctic and/or North Atlantic data are used.

Table 1. Correlations of the AMO Index and OTI With Selected (Smoothed and Detrended) Timeseries, as Well as SST and SLP Regression Pattern Correlations (Area-Weighted) for the Latitudes 60°N–60°S, for 1904 to ∼2008 (End Date Varies With The Indices)a
 AMO Temporal 10-Year FilteredAMO Spatial SST, SLPOTI Temporal 10-Year FilteredOTI Spatial SST, SLP
  • a

    The Atlantic sector of the Arctic is the area averaged land SAT anomalies for 64°N–90°N, 90°W–35°E from GISTEMP. The Pacific sector of the Arctic is the area averaged land SAT anomalies over 64°N–90°N, 150°E–90°W. Italicized text and bolded text indicate the spatial pattern correlations that are considered significant at the 95% and 90% levels, respectively.

ORCADAS−0.10−0.30, 0.03−0.31−0.59, −0.72
ICECORE−0.11−0.19, −0.040.510.84, 0.84
SAM0.390.37, 0.290.640.78, 0.86
ARCTIC GISTEMP0.660.80, 0.730.540.47, 0.70
ARCTIC PACIFIC0.330.41, 0.450.720.80, 0.87
ARCTIC ATLANTIC0.840.92, 0.870.390.32, 0.55
AMO  0.04−0.02, 0.13

[15] In Figure 2we compare SST and SLP regression patterns using the AMO index and OTI, together with regressions using the ICECORE, Orcadas and SAM records for the Antarctic. For the Arctic, we use the 64°N–90°N GISTEMP timeseries, along with separate land SAT records for the Pacific and Atlantic sectors of the Arctic. Both the OTI and the AMO are associated with negative SST anomalies in the South Atlantic. In the North Atlantic, the AMO emphasizes positive SST anomalies immediately south of Greenland and the OTI is associated with positive but weaker anomalies further south. In the tropical Indo-Pacific, SSTs associated with the OTI are predominantly positive and SSTs associated with the AMO are predominantly neutral or negative. In the western North Pacific, the AMO is associated with positive anomalies and the OTI associated with negative anomalies. The ICECORE record and the Arctic-Pacific record both exhibit SST regression patterns consistent with the OTI. There are differences in locations of maximum regression coefficients, with the Arctic-Pacific SAT associated with stronger anomalies in the North Pacific and the North Atlantic than the ICECORE record. Consistent with the SSTs, the OTI, Arctic-Pacific SAT and ICECORE records all exhibit a pattern in SLP resembling the negative phase Southern Oscillation. Although the observational data constraining the sea level dataset are sparse in the SH high latitudes, the high pressure anomaly in the Amundsen-Bellingshausen Sea is consistent with the typical pattern associated with warm-phase ENSO events and low-frequency ‘ENSO-like’ variability [Garreaud and Battisti, 1999]. This circulation pattern implies colder than average conditions in the Antarctic Peninsula region. It is also consistent with warmer conditions in other parts of West Antarctica and generally with the strong decadal variability found in West Antarctic ice core records [Schneider and Steig, 2008]. Positive temperature anomalies in the Orcadas record are associated with warmer than normal conditions in the south Atlantic, and a global pattern in SST and SLP roughly opposite to the positive phase of the OTI. The AMO and Arctic-Atlantic SAT show strong North Atlantic signatures but little coherent patterns in tropical Indo-Pacific SSTs.

Figure 2.

Regressions of global fields of SST (colors) and 65°N–65°S SLP (contours) anomalies onto various indices for 1904–2008. Prior to regression, all indices and the data at each gridpoint were smoothed with a low-pass filter, preserving timescales >10 years and the linear trends were removed. Note that the Tropical Index ends in 1997 while the Antarctic icecore index ends in 1999, and that the filtering removes approximately 5 years from the start and end of each series. The SST units are in °C per standard deviation of the index. The SLP contour interval of the darker lines is 0.3 hPA per standard deviation of the index, with positive values as solid lines (the zero contour is the thicker solid line) and negative values dashed. Also shown in lighter black lines are SLP contours of −0.1 and 0.1 hPA to better illustrate the regression patterns in the tropics. The green dots in the ice core regression indicate the locations of the individual ice cores used for the ICECORE composite and the green dot in the Orcadas regression indicates the location of Orcadas station. (bottom row) SST and SLP anomalies associated with the juxtaposition of a cool tropical phase and a positive AMO phase (bottom left)—suggested by the trends in the indices since about 1980—compared with the pattern of trends (SST: °C/decade; SLP: hPA/decade, contoured as described above) for 1979–2008 (bottom right). Stippling indicates the trend is statistically significant at the 95% confidence level or above in the SST field. Green hatch lines indicate that the trend is significant in the SLP field.

[16] Pattern correlations (Table 1; see auxiliary material for methods) show that the three Antarctic indices—Orcadas, ICECORE, and SAM—all exhibit significant relationships with SST and SLP patterns associated with tropical interdecadal variability. In contrast, the Arctic Atlantic sector record is strongly associated with the AMO pattern. However, the Arctic Pacific sector record does not have significant correlations with the AMO pattern, but rather shows strong correlation with the OTI SST and SLP patterns.

6. Discussion

[17] Our results suggesting a direct linkage of the Antarctic with tropical climate are supported by a number of recent studies. We find limited linkages with the North Atlantic or Arctic. With respect to North Atlantic/Arctic linkages, the results of a number of relevant modeling experiments reinforce our observationally based findings. In interpreting model experiments, it is important to distinguish those that force the North Atlantic with large amounts of freshwater (so-called hosing experiments) from those that are unforced, representing the internal variability of the coupled ocean-atmosphere system. The unforced experiments are most relevant to assessing the seesaw variability described byChylek et al. [2010]. The main mechanism to evaluate is the AMO and its potential linkages with the variability of the Atlantic Meridional Overturning Circulation (AMOC) [Knight et al., 2005], and consequent impacts on the oceanic transport of heat between the Northern and Southern hemispheres. In this regard, it is notable that a robust AMO connection to SH climate has not been reported in observations or unforced model integrations [Knight et al., 2005; Ting et al., 2009], and it is not clear if the bipolar seesaw mechanism is relevant outside of the last glacial and deglacial periods [Broecker, 1998], when proxy data and model experiments suggest that the AMOC was vulnerable to slowing down after large pulses of freshwater entered the North Atlantic, strongly cooling the North and weakly warming the South [e.g., Stouffer et al., 2006].

[18] Tropical variability is evident in a number of Antarctic data sets and variables, as reviewed by Turner [2004] and Schneider et al. [2012b]. The main mechanism for tropical signals to reach the Antarctic is a Rossby wave-train driven by anomalous tropical deep convection [e.g.,Turner, 2004]. The anomalous wave activity can occur on both interannual ENSO timescales [e.g., Mo and Higgins, 1998; Mo, 2000] as well as decadal timescales associated with ‘ENSO-like’ low-frequency variability [e.g.,Karoly, 1989; Garreaud and Battisti, 1999]. Karoly et al. [1996] found the leading mode of decadal variability in SH atmospheric circulation to be related to ENSO and concentrated in the Pacific sector. Tropical variability can also impact the SAM [e.g., L'Heureux and Thompson, 2006; Fogt et al., 2011] due to anomalous heating of the tropical troposphere during ENSO and its impacts on the meridional temperature gradient [Seager et al., 2003]. Thus, the SAM is not independent of tropical variability and this is an explanation for the correlation between the SAM and OTI SST patterns reported in Table 1. Taken together, there is much stronger evidence for tropical modulation of Antarctic climate than there is for a bipolar seesaw or oscillation between the Arctic and Antarctic.

[19] With respect to recent Antarctic trends, there is a complex interplay among multiple mechanisms, both internal and external. A positive SAM trend has occurred largely since the 1970s and the mid- and high-latitude SH is one of the few regions of widespread statistically significant SLP trends since 1979 (Figure 2, bottom right). As in other studies [Turner and Overland, 2009], we infer that the circulation trend is largely responsible for the lack of significant Antarctic-averaged warming since ∼1979. A bipolar seesaw is not needed to explain the lack of warming in some regions and seasons in Antarctica [Marshall, 2007; Schneider et al., 2012a]. However, both the SAM and Antarctic temperature trends have a marked seasonality, and there has been significant regional warming in the western Antarctic Peninsula and West Antarctic Ice Sheet [Schneider et al., 2012a]. These seasonal and regional warming patterns are linked with atmospheric circulation changes in the South Pacific that are consistent with the observed pattern of tropical and extratropical SST trends [Schneider et al., 2012a; Ding et al., 2011]. The pattern of SST trends since 1979 (Figure 2, bottom right) in part reflects tropical variability and the transition from a warm to a cool tropical phase over recent decades that is evident in the OTI and PDO indices (Figure 1d). This is associated with an atmospheric teleconnection in the South Pacific (Figure 2, bottom left) that resembles the circulation patterns identified by Schneider et al. [2012a] and Ding et al. [2011]. Thus, while Antarctic climate is responsive to forcing by stratospheric ozone depletion and greenhouse gas increases [e.g., Gillett et al., 2008], natural variability has likely played a role in recent trends.

7. Conclusion

[20] While there are several possible reasons why the Antarctic has not warmed as strongly as the Arctic in recent decades, our analysis eliminates the bipolar seesaw as one of them. Apart from astronomically forced variations, there are few phenomena in the climate system that are truly oscillatory. It would be surprising if a periodic seesaw existed between the polar regions. An anti-correlation is evident in the GISTEMP Arctic and Antarctic data going back to 1904, but it is an artifact of changing Antarctic data coverage. The GISTEMP observational temperature dataset is only adequate for Antarctic climate studies for the period from ∼1958 to present. It is recommended that users of Antarctic observational data sets investigate the original source data and the methods used to create gridded fields. Furthermore, supporting evidence from related data sets and physical variables should be sought; we have done this and found a lack of support for the seesaw hypothesis. Nonetheless, this study underscores that natural decadal variability is evident in the records and is imprinted on recent trends. This is illustrated by the projection of Pacific decadal variability onto recent SST trends, and in turn by the association of tropical SSTs with Antarctic temperatures. Still, tropical variability does not explain all of the variance or trends in Antarctic climate. Studies that systematically account for the roles of multiple factors, both internal and external, are keenly needed.


[21] Support of this work for D.N. and D.S. at the University of Colorado came from NOAA's Climate Program Office's Paleoclimate program (grant NA17RJ1229) and the National Science Foundation (NSF) (grant ARC-1107795). D.S. acknowledges additional support while working at the National Center for Atmospheric Research (NCAR), under National Science Foundation (NSF) grant ANT-0838871. NCAR is sponsored by the NSF.