JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 113, D17110, doi: /2007jd009107, 2008

Transcription

1 Click Here for Full Article JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 113,, doi: /2007jd009107, 2008 Variability in the teleconnection between the El Niño Southern Oscillation and West Antarctic climate deduced from West Antarctic ice core isotope records Scott Gregory 1 and David Noone 1 Received 28 June 2007; revised 5 March 2008; accepted 6 June 2008; published 5 September [1] The U.S. contribution to the International Trans-Antarctic Scientific Expedition (ITASE) program obtained several ice cores from the West Antarctic Ice Sheet. Because of proximity to the Pacific Ocean, the West Antarctic ice cores are expected to have an El Niño Southern Oscillation (ENSO) signature. The ITASE core d 18 O isotope was selected for detailed analysis here because its location, high annual accumulation, and record length make it an ideal candidate for capturing the effects of regional circulation anomalies in the isotopic composition. The core is compared to two other cores, the and cores, which are further west and therefore capture some spatial variability of the regional circulation on various time scales. Analysis shows that several phenomena, including ENSO, leave a signature in the ice cores. Evidence suggests that ENSO signals in the ice cores are significantly modulated by low-frequency variability. Correlation with the Southern Annular Mode (SAM), global temperature, Pacific Decadal Oscillation, and ENSO shows that temperature and ENSO generally appear to have the strongest influence on the isotope signal while there is no clearly dominant single influence in the other cores. Results suggest that the teleconnection between ENSO and the core is quite dependent on the state of the SAM. Specifically, when the Southern Oscillation Index (SOI) and SAM are in phase, there is an ENSO related pressure anomaly west of the Antarctic Peninsula, in the vicinity of the ice cores studied. This extends previous findings to span the entire 20th century. Citation: Gregory, S., and D. Noone (2008), Variability in the teleconnection between the El Niño Southern Oscillation and West Antarctic climate deduced from West Antarctic ice core isotope records, J. Geophys. Res., 113,, doi: /2007jd Introduction [2] The isotopic composition of the snow on West Antarctica is anticipated to show large variability on El Niño Southern Oscillation (ENSO) time scales because of its proximity to the Pacific Ocean and ENSO s influence on both the Amundsen Sea Low [Bertler et al., 2006] and the socalled Antarctic Dipole (ADP) circulation anomaly [Yuan, 2004; Harangozo, 2000]. ENSO has been shown to influence these features as well as characteristic southern hemisphere features such as the Southern Annular Mode (SAM) [Fogt and Bromwich, 2006; L Heureux and Thompson, 2006] and the zonal wave 3 [Raphael, 2004]. All of these atmospheric features will be accompanied by variations in the mean condensation and transport processes that will influence the isotopic content of the precipitation deposited at any given ice core site through differences in the temperature history and mixing characteristics as noted below. 1 Department of Atmospheric and Oceanic Sciences and Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, Colorado, USA. Copyright 2008 by the American Geophysical Union /08/2007JD009107$09.00 [3] The variations in the isotopic composition of ice cores reflect a history of climate and are often used to construct long-term temperature records [e.g., Aristarain et al., 1990; Jouzel et al., 2000]. While the temperature-isotope relationship is generally robust [e.g., Jouzel et al., 1997], there are many other influences that affect the isotopic composition of the ice core. Consider a parcel of air with water evaporated at midlatitudes; as the parcel cools, the water molecules with heavier isotopes of hydrogen and oxygen precipitate preferentially because their lower vapor pressure causes them to condense more readily. If the temperature gradient along the transport path to the ice core site is reduced, there will be less depletion relative to some mean isotopic content, and this will give rise to a positive anomaly. Confounding this simple view, locally evaporated water mixed into the poleward moving air will generally enrich the overlying vapor [Noone, 2008]. As such, a positive isotope anomaly may appear from an anomalously large contribution from local sources. Further complexity of an isotopic signal in an ice core is added by wind drifted snow, especially in West Antarctica where katabatic winds may transport isotopically depleted snow from the Antarctic interior, resulting in a negative isotope anomaly. The difference between local evaporative enrichment, reduced 1of17

2 depletion due to temperature, and variability due to blowing snow cannot be detected in a single isotope record, however these influences are all affected by regional circulation characteristics. Because of their role in regional climate, spatial patterns of atmospheric and oceanic variability must therefore be considered in developing a thorough understanding of the observed isotope records. Neglecting the temperature dependence of the isotopic fractionation, the most isotopically enriched snow in West Antarctica is associated with the features that transport vapor most directly from the source, the Southern Ocean, to the site of interest. [4] To understand the influence of ENSO on the isotopic content of the snow deposited in West Antarctica, the means by which the tropical signals propagate to high latitudes must be considered. Turner [2004] provided a review of the ENSO teleconnection to Antarctica and concludes that the clearest connection is through the Pacific South American (PSA) pattern but there is a great deal of variability in the Antarctic s response to individual events. Mo and Paegle [2001] note that the PSA modes exhibit a zonal wave 3 in the midlatitudes, and indeed are spatially coincident with the Pacific circulation associated with the ADP. They draw equivalence between their first PSA mode, the second EOF of geopotential height, and Karoly s [1989] southern hemisphere response to ENSO. Karoly [1989] showed that a Rossby wave train is excited in the tropics by the deep convection associated with Sea Surface Temperature (SST) anomalies during ENSO events. The Rossby wave train propagates to high latitudes yielding a blocking high in the area of the Amundsen-Bellinghausen Sea (ABS) during warm ENSO (El Niño) events, with anomalies of opposite sign for cold (La Niña) events, as shown by Turner [2004] for eight El Niños and six La Niñas from NCEP-NCAR reanalysis. The differences between the Rossby wave trains that arose from four individual El Niño events were studied by Lachlan-Cope and Connolley [2006]. They showed that there are peculiarities from event to event that depend on the spatial distribution of tropical SST anomalies relative to areas of ascent. These peculiarities change the Rossby wave source regions and as a result modulate the ABS pressure anomaly s amplitude and location. They explain a significant Faraday temperature difference between 1987 and 1997 as being a result of the positive pressure anomaly closer to the Peninsula in 1987, and therefore an anomalously cold temperature that year. [5] The ABS pressure anomaly forms one center of action of the Antarctic dipole and modifies the location of the Amundsen Sea Low [Bertler et al., 2006]. Turner [2004] notes that such high-latitude features will be modulated by variations in westerlies, which will result from, among other things, SAM variability, further complicating the interpretation of the tropical to west Antarctic teleconnection using in situ observations. Rather than basing the analysis on in situ proxies, Harangozo [2000] found a more clearly defined ADP in sea level pressure (SLP) correlations with interseasonal changes in SST, and deduced the stronger dependence of the regional atmospheric flow on the rate of change of ENSO state rather than the state of the SSTs. [6] There is no clearly defined phase relationship between ENSO and West Antarctic precipitation, as shown by Bromwich et al. [2000]. Using ECMWF operational analyses and reanalyses, they show a 1980s to 1990s phase reversal of the west Antarctic moisture flux convergence (MFC) versus the Southern Oscillation Index (SOI). In contrast to their findings, Genthon and Cosme [2003] find that there is not a phase reversal but point out that there is phase opposition between the Bellinghausen-Weddell Sea area versus the Amunsdsen-Ross area with respect to ENSO. Their study suggests a spatial sensitivity of the precipitation phase that is also highlighted in Bertler et al. [2006] who discuss the opposition of warming/cooling phase of the eastern and western Ross Sea correlated with ENSO. Specifically, the displacement of the Amundsen Sea Low during El Niño causes katabatic flow of cold air from the interior of the continent to the western Ross Sea resulting in more sea ice in that area despite warmer SSTs which reduce the sea ice in the eastern Ross Sea [Bertler et al., 2006]. [7] This study focuses on interpreting the circulation patterns associated with the isotopic record from the ITASE ice core compared with records from the and cores [Steig et al., 2005; Schneider et al., 2006]. These cores were extracted from West Antarctica, near the coast and in proximity to the Pacific Ocean (Figure 1), and the isotope records are resolved at monthly intervals. The core covers the entire 19th and 20th centuries and the and cores are dated for the late 19th and the entire 20th century. Given their location and high temporal resolution, it is expected that they reflect ENSO and other variability, and as such give an indication of how West Antarctic climate is influenced by the tropical Pacific. [8] In a study of surface mass balance (SMB) of West Antarctica, Genthon et al. [2005] compare these three cores plus three other ITASE cores further west to ERA 40 analysis/reanalysis as well as the SOI and SAM. Their results show that at the core sites of interest to this study, the surface mass balance correlates weakly with SOI and moderately strong with an undetrended index of the SAM. The isotopic record will not necessarily reflect a similar relationship to the SOI and SAM as the SMB because the isotopic composition is controlled by the conditions under which both evaporation and condensation occurs upstream, and not just the net deposition at the site. The results of Genthon et al. [2005] show that local meteorological influences and postdepositional processes are responsible for a significant portion of the SMB signal at any particular site, and therefore the site-to-site correlations are very small, where the ERA 40 site-to-site SMB correlations are relatively large. Importantly, they find that the variability in accumulation is dominated by local processes but we speculate that the isotopic composition of the accumulated snow studied here will be a function of the regional circulation characteristics. At the site, Genthon et al. s [2005] correlation with the ERA 40 analysis is strong and statistically significant, suggesting that it is influenced to a lesser extent by the small-scale perturbations and therefore the best candidate for determining teleconnection patterns to the tropical ENSO. Further support for using the core is given by Steig et al. [2005], who found that the core has detectable seasonal variations for its entire depth, where some cores lose seasonal variations to isotopic diffusion in the firn. 2of17

3 Figure 1. Location map of ice cores used in this study. Contour intervals of topography are every 200 m. [9] Section 2 discusses the temporal signal of the , and ice cores and the ENSO signals therein. Section 3 identifies the associated circulation anomalies by correlating the ice core signals with SLPs, showing that there is multidecadal variability in the phase of ENSO that is linked to the isotope signal in the cores. To explain this, we identify the differences in the spatial patterns of temporal correlations from three epochs of strong ENSO activity. Variability on ENSO time scales is separated from the lower-frequency variability and trends to identify the dominant features that shift the way that ENSO influences the ice core records. The analysis continues by considering decadal variation in the ice core data and the Pacific Decadal Oscillation (PDO), which prompts a discussion of the role of the PDO in modulating the southern hemisphere circulation response to ENSO. Concluding the analysis, the relative influences of ENSO, PDO, SAM, and global temperature on the isotope signals are estimated by correlating them all with the isotope signals at time scales from 1 year to 25 years, to provide guidance in understanding the role of circulation changes on the records. 2. Temporal Signature of ENSO in ITASE , , and Cores [10] The ITASE , , and ice cores were extracted from sites close to the Antarctic Peninsula and the Pacific coast of Antarctica (Figure 1), and are dated at monthly intervals starting before the 20th century. The three core sites are in the Pine Island Thwaites drainage system while the core is near a topographic ridge between mainland Antarctica and the peninsula at 89.1 W 77.1 S. The and cores are on a slope facing the Pacific Ocean at W 77.8 S and 95.6 W 78.1 S, and are approximately 300 km and 500 km, respectively, from the coast. These locations are advantageous for collecting precipitation from coastal cyclonic activity, especially the core, because of its proximity to the Antarctic Peninsula and Drake Passage. Kaspari et al. [2004] compared the accumulation rates of the West Antarctic ITASE ice cores and found that of the cores in this vicinity, these sites have the greatest accumulation rates (43.6 cm for , 33.1 cm for , and 34.2 cm water equivalent per year for ), and they have the greatest standard deviation of accumulation, implying sensitivity to various seasonal influences. [11] Dating of the ITASE ice cores was performed by Dixon et al. [2004] and Steig et al. [2005]. The accuracy and precision of the dating is a critical issue when performing the correlative studies here. Their method of dating the core was done by annual layer counting, primarily by identifying the summer peaks in non-sea-salt sulfate concentration, and validated with stable isotopes and other seasonal chemical variations and by identifying volcanic markers. They conservatively estimate the accuracy to be better than ±2 years, and ±1 year in the more recent parts of the cores. Subannual dating is achieved by assuming the isotope maximum occurs in January, and dividing the remainder into 12 equal depth samples. Though it is unlikely that there is equal accumulation in each month, Steig et al. [2005] argue that the precision of the January peaks in non-sea-salt sulfate are within one month and the winter troughs are within 1 2 months of the nominal midpoint, 1 July, of each year. Given this accuracy and precision this data provides a powerful tool for understanding West Antarctica s sensitivity to seasonal ENSO. [12] Isotope ratios were obtained from the cores at the University of Washington using standard CO 2 equilibration for oxygen and reduction of chromium for hydrogen [Steig et al., 2005; Schneider et al., 2005, 2006]. The ITASE and cores time series of deseasonalized d 18 O anomalies relative to the annual mean d 18 O values of 29.8% in and 31.8% in are shown in Figure 2. They have standard deviations of 1.86% and 1.96% and mean seasonal amplitudes of 1.4% and 1.8% respectively. The core time series of dd has an annual mean of 253.5%, a standard deviation of anomalies of 11.8% and seasonal amplitude of 12.1%. The ice cores time series is accompanied in Figure 2 by the Southern Oscillation Index (SOI) provided by the Australian Bureau of Meteorology (available from For all time series, the annual, 5-year and 15-year moving average anomalies are shown to assist in identifying low-frequency interannual variability. While the SOI is considered a standard and robust measure of ENSO activity, values prior to 1935 should be viewed with caution as there are questions regarding the consistency and quality of the Tahiti pressure values prior to 1935 [Ropelewski and Jones, 1987]. 3of17

4 Figure 2. The 20th century time series of (a) SOI and ITASE (b) d 18 O, (c) dd, and (d) d 18 O isotope anomalies from average seasonal cycle, shown in % as annual average and 5-year and 15-year moving averages. [13] It is useful to consider some basic similarities and differences between these series. Counting the number of peaks in the annual average of the isotopes and SOI yields approximately 20 in the 20th century, suggesting a dominant periodicity in the vicinity of 5 years, which is within the 2 8 year range acknowledged to be the spectral signature of ENSO [e.g., Trenberth, 1997]. There appears to be no consistent coincidence of extrema in any of the averages over the entire record (although some of the peaks are aligned in the annual data) and indeed the absolute value of the correlation coefficient between the annually averaged data is less than 0.05 for each core and not statistically significant. This suggests that there may be other, stronger influences on the isotope records or, of specific interest in this study, the causal link between ENSO and the isotope can change with time. Finally, we note that in recent decades the SOI is distinctly lower (more El Niño like), while the overall trend in d 18 O of the core is both upward and not unlike the record of global average temperature [cf. Brohan et al., 2006], while the primary feature that distinguishes the core is the positive anomaly around 1940, which is also a feature in the core and global temperature records. The time series of the dd is qualitatively similar to the , without the distinct trend. The trend in the core is most clearly seen in the 15-year moving averages and its general similarity to global temperature suggests that it, unlike the other cores, responds to the global temperature variability. The monthly record was compared to ERA temperature for the closest grid point (not shown); the seasonal peaks generally agree as do some longer time scale features, while the differences highlight the fact that signals other than temperature appear in the isotopic records. [14] To better establish an association between the isotope anomalies from the three cores and the SOI, Figure 3 shows correlation between them in each season for each 20-year interval in the record. Ninety five percent confidence requires a correlation of 0.42 for a two sided t test, 90% requires correlation of Though many individual years fail these confidence tests, the analysis reveals both a multidecadal oscillation and unprecedented differences between seasons in recent decades for the core. 4of17

5 Figure 3. (a) The 20th century seasonal 20 year moving correlation between Southern Oscillation Index and ITASE d 18 O. Same for (b) ITASE dd and(c)itase d 18 O. For the core, the correlations are weak and inconsistent between seasons but overall the multidecadal variation is of opposite sign to the core. The multidecadal oscillation shows a generally positive (albeit weak) correlation in the first few decades of the 20th century, a generally negative correlation in the middle of the century, and a stronger positive correlation toward the end of the century. Similar correlations performed on the core, which is located between the and sites, are similar to the core in the first half of the 20th century but more like the in the second half of the century. Figure 3 indicates that at the beginning and end of the century, the less depleted (2001-2) isotope values coincide with La Niña (El Niño) events, while in the middle of the century, they are associated with El Niño (La Niña). [15] The core seasonal correlations generally behave similarly to each other throughout the century until about 1980, at which time the austral summer and autumn diverge from the winter and spring, which continue to behave similarly to each other. Bromwich et al. [2000] found an ENSO-Antarctic precipitation phase reversal occurring abruptly around The 20 year moving correlations performed here do not allow detailed examination on the same time scale as Bromwich et al. [2000] but the recent spread in the seasonal correlation may be an artifact of the same mechanism. This change is within the portion of the core that Steig et al. [2005] demonstrate better than 1 year accuracy in the dating. [16] Thompson and Solomon [2002] studied the southern high-latitude variability in recent decades and concentrated on the SAM, which is a measure of the variability of the circumpolar vortex. They found that the seasonal breakdown of the polar vortex has migrated from early November in the 1970s to late December in the 1990s, in part because of significant cooling in the stratosphere. Their results are consistent with what one might expect from stratospheric ozone loss and may also explain the divergence of the seasonal correlations shown here in recent decades when ozone changes were observed. One tantalizing explanation for the present result is that the wintertime forcing on the flow has changed differently from the forcing in other seasons because of a link between ENSO and SAM via the stratosphere. Such a linkage is explored in the analysis below. [17] The ENSO phase difference between the core and the core may be explained by the Antarctic dipole and its phase relationship to ENSO. The dipole is represented by anomalies of surface variables such as sea ice, surface temperature, and sea level pressure that are of opposing phase about the eastern and western sides of the Antarctic Peninsula [Yuan, 2004]. The dipole is characterized by warmer temperatures and therefore reduced sea ice occurring west of the Peninsula during El Niño conditions, while the eastern side experiences cooler temperatures and increased sea ice. The boundary between the high and low anomalies is generally in the vicinity of the base of the Peninsula, west of the Peninsula itself [Yuan, 2004]. Yuan s [2004] study showed the boundary to be close to the three core sites and from the ENSO phase found here, we can assert that although the dipole need not be fixed in space, the core is generally sensitive to the eastern center of 5of17

6 the dipole s influence while the core is more influenced by the western center of the dipole. This may be expected from the location on a ridge dividing the east and west. It is also reinforced by the core s secular similarities to the and , suggesting eastern versus western influences wander among the core sites. Also, the opposing phase between the and suggests that the dipole is a distinct element of the teleconnection between the tropics and West Antarctica. Finally, the weaker correlation of the core with SOI suggests that it is either closer, in the mean, to the boundary between the centers of action of the dipole or that it is simply less sensitive to the ENSO related circulation features, and more prone to local influence as found by Genthon et al. [2005]. [18] A wavelet spectral analysis of the time series of the ice cores and SOI was performed using the method developed by Torrence and Compo [1998] and the resulting spectra for the core are shown in Figure 4. The anomaly time series input to the wavelet transform were normalized to their standard deviation, and the power is plotted in units of variance. The abscissa of the plots is years from 1900 to 2000, the ordinate is the period of the sampled signal ranging from zero to 25 years. The power distribution in the autospectra shows the evolution of dominant periods in the two records over the 20th century. The 2 8 year range of ENSO [Trenberth, 1997] is a prominent, yet intermittent, aspect of each of the auto spectra as well as for the cross spectra. Indeed this result confirms the correlations found in Figure 3 are associated, at least in part, with the 2 8 year ENSO band. Also, the spectrum of SOI has a trend in upper limit of the ENSO period range, discussed further in a later section. The annual accumulation signals were also analyzed using the wavelet analysis but did not reflect the ENSO variability seen in the spectra of the isotopes, which is further confirmation of the results of Genthon et al. [2005]. [19] There is also a year period SOI signal that appears toward the end of the century and, while exists only to a lesser extent in the d 18 O, appears also in the cross spectra. This year signal is the spectral representation of the positive deviations seen in Figure 2, most clearly in the 5-year moving average departures from the smoother 15-year moving average which shows the recent more negative SOI. This portion of the spectrum was also found by Hasegawa and Hanawa [2006] in the Niño 3.4 ENSO index. Their investigation focuses on the frequency of ENSO events and amplitude of temperature anomalies depending on the phase of this year oscillation, finding more El Niño events and larger amplitude anomalies during the low SST values indicated by the Niño 3.4 index. As such, while this tropical variability is known, the present finding suggests there are more tropical signals recorded in the ice core than simply the 2 8 year 6of17 Figure 4. (a) Wavelet power spectrum of the Southern Oscillation Index in dimensionless units relative to SOI variance. (b) Wavelet power spectrum of the ITASE d 18 O isotope in dimensionless units relative to d 18 O variance. (c) Wavelet cross spectrum of SOI with the ITASE d 18 O isotope units relative to the product SOI and d 18 O standard deviations. The time periods of apparent ENSO activity are as follows: epoch 1, ; epoch 2, ; and epoch 3, Morlet wavelet is used for all.

7 Figure 5. Station sites south of 30 S included in the HadSLP2 data: (a) 71 epoch 1 stations, (b) 97 epoch 2 stations, and (c) 171 epoch 3 stations. Notice in particular the absence of observations in the eastern sector of the South Pacific Ocean. ENSO cyclicity. Similar to the late twentieth century changes in the time series and correlations (Figures 2 and 3), this recent low-frequency signal is suggestive of a response to changing climate in recent decades, including a trend in the SAM. [20] Concentrating on the auto spectra of d 18 O in the core (Figure 4b), there is distinct intermittency in the strength of the signal in the 2 8 year period range, with altogether less power in this ENSO frequency range. The intermittency is coincident with that of the SOI which strongly suggests that ENSO is causing that portion of the d 18 O signal. Indeed, the cross spectrum (Figure 4c) confirms the correlation of the signals, seen in Figure 3, on a spectral level. As such, we are motivated to focus analysis on these three distinct epochs for examining the ENSO- Antarctic teleconnection. The three epochs identified are (1) , (2) , and (3) Spatial Signature of ENSO in ITASE Core [21] The comparisons of ice core and SOI time series of section 2 show periods of ENSO activity influence the ITASE core, albeit with different sign correlations found when comparing epoch 2 to epochs 1 and 3. In this section we aim to determine the atmospheric circulation patterns that link ENSO to West Antarctica. The spatial analysis is restricted to the core since it generally shows the most pronounced correlation to the SOI and therefore more likely dominated by circulation associated with tropical forcing and regional circulation anomalies, rather than local-scale perturbations identified by the Genthon et al. [2005]. The spatial analysis is also confined to the austral spring (September, October, November (SON)) season. Fogt and Bromwich [2006] suggest the austral spring (SON) and summer (December, January, February (DJF)) best capture ENSO s influence on Antarctica, since it is during these seasons that ENSO events are in their mature stage. They point out that since other studies have not constrained their work to the SON and DJF seasons, the decadal variability of ENSO and a relationship with the SAM index had been overlooked. Also, seasonal averaging reduces potential misinterpretations due to dating errors in the ice core and initial analysis of spatial patterns in HadSLP2 and HadSST showed the strongest ENSO pattern in the SON season. Therefore the spatial analysis shown in this study is confined to SON. [22] To determine the spatial patterns, monthly mean gridded sea level pressures (SLP) from the Hadley Center (HadSLP2 [Allan and Ansell, 2006]) are used. The HadSLP2 data are compiled from raw observations and interpolated to the final grid. The SLP data set is on a 5 degree grid and dates from 1850 to The interpolation technique used in producing the data is the Reduced Space Optimal Interpolation which uses a generally well observed period to allow filling of data-sparse regions in space and time. The observations used to generate the globally complete fields were passed through rigorous quality controls, yet locations with limited observations are subject to greater uncertainty. The data prior to the satellite era, especially in the Southern Hemisphere, are sparse and therefore the interpretation of the results is accompanied by a cautionary note of their reliability for epochs 1 and 2, and should be considered significantly more reliable for epoch 3. Figure 5 shows the location of the high southern latitude stations (south of 30 S) that were used to generate the HadSLP2 for each epoch. The stations are marked for the epochs if they were operational for at least half of the span of the particular epoch; 71 stations in epoch 1, 97 in epoch 2, and 171 in epoch 3. In the optimal interpolation approach employed in the construction of the HadSLP2 data, the well observed (recent) period of data is used to generate the leading Empirical Orthogonal Functions (EOFs) of the global sea level pressures. ENSO plays a major role in the leading EOFs [Allan and Ansell, 2006] and it is constrained in nonpolar regions, therefore global ENSO patterns are represented in HadSLP2. [23] As discussed earlier, the presence of atmospheric circulation anomalies, such as an ENSO-related Rossby wave train [Karoly, 1989; Turner, 2004], SAM [Fogt and Bromwich, 2006; L Heureux and Thompson, 2006], the ADP [Yuan, 2004; Harangozo, 2000] and wave 3 [Raphael, 2004], are expected to influence the circulation and thus the isotope signal through changes in regional moisture trans- 7of17

8 Figure 6. Correlation between ITASE ice core d18o isotope and sea level pressures for the three epochs identified in Figure 3: (a) epoch 1, ; (b) epoch 2, ; and (c) epoch 3, The contour interval is 0.1. Negative contours are dotted, and the zero and positive contours are solid. Significance at the 95% level requires correlation of 0.4. Northernmost latitude is 30 S, and grid lines marked are 40, 60, and 80 S. port and condensation history. As shown by Turner [2004], La Niña conditions are characterized by an anomalous stationary cyclone in the Southern Ocean in the region of the Bellingsausen Sea. The stationary cyclone is expected to lead to isotope enrichment in West Antarctica because the source of advection is in lower latitudes where the ambient vapor is less depleted, and also because the advection is warmer which allows reduced isotopic rainout. The role of such a cyclone on increasing the local evaporative source and boundary layer transport is unclear, but would only further enrich the vapor relative to the mean. [24] Figure 6 shows spatial correlations between d 18 O and SLP. As with the correlation between d 18 O and SOI, a sign reversal of the pressure anomaly in the ABS in the second epoch is found in Figure 6b. The significance of the correlations at much of the high southern latitudes is 90% but typically fails a one-sided t test at the 95% level at all but a few isolated locations. For the first and third epoch, the correlated circulation anomaly west of the Peninsula (Figures 6a and 6c) is in agreement with Turner s [2004] La Niña pattern and suggests northerly advection of isotopically enriched water from the Southern Ocean. Indeed, it is stronger in the third epoch than the first. In fact, in the first epoch there is a zero correlation contour in the area of the ice cores studied here, while in the third epoch the cyclonic correlation covers the area. Comparing this to Figure 3, the 20 year moving correlation between SOI and d 18 O, this observation is in agreement with weak positive correlation in the beginning of the 20th century (the first epoch) and stronger positive correlations in the late 20th century (the third epoch). This result, while perhaps expected on the basis of the correlations between the SOI and d 18 O, provides confidence that the HadSLP2 captures physically consistent circulation features and that this data is appropriate for using in this study, even in the first two epochs. [25] Figure 6 shows not only that the variability within the second epoch is different than that of the first and third, but allows speculation on the circulation characteristics that are responsible for the differences in the isotopic anomalies. In the second epoch, the positive SLP-d 18 O correlation area near the Antarctic Peninsula suggests a blocking high, in agreement with Turner s [2004] El Niño pattern and forming one center of the ADP. This pattern provides approximate geostrophic moisture transport in a sense opposite to the first and third epochs, which have pressure anomalies that would advect isotopically enriched water from the ocean. Though this epoch 2 pattern seems to counter the expectation of the circulation anomaly that will produce a higher isotope signal, the correlation is also positive for the entire Antarctic continent, equivalent to a negative SAM index, which also implies warmer temperature on the continent correlating with positive isotope anomalies. Recall from Figure 5 that the third epoch of the HadSLP2 data is constrained on the Peninsula and at Scott Base in epoch 2, as well as many midlatitude stations and, as such, the broad characteristics of the anomalies may be considered robust. Adding to the argument of a positive polar pressure (negative SAM) influencing the isotopic composition of the core, Figure 7 shows mean SLP anomalies for each of the epochs. These patterns are not annular in their spatial distribution but would project onto SAM patterns with negative indices for the first ( 0.18 standard deviations) and second ( 0.35 standard deviations) epochs, positive for the third (+0.55 standard deviations). The epoch 2 spatial patterns are consistent with the observation of Fogt and Bromwich [2006], that strong ENSO teleconnection to Antarctica requires SOI and SAM to be in phase. This may also explain the weak and seasonally variable correlations between the isotope and SOI in the first epoch, since it appears that the weaker negative SAM index would be competing with the positive SOI s enriching circulation feature. In the third epoch, where there is a positive SAM, the SOI correlation with the isotope is strongly positive (Figure 3), whichisalsoconsistentwithfogt and Bromwich [2006]. This is explored further in the section 4. [26] All of the SLP patterns show an extension of the polar pressure anomaly off the western side of the Antarctic Peninsula which is in the region of ENSO-related pressure anomalies shown by Turner [2004]. Similar patterns were found by Fogt and Bromwich [2006] in the 500hPa geo- 8of17

9 Figure 7. Mean sea level pressure anomalies for the Austral spring (SON) relative to the mean for (a) epoch 1, ; (b) epoch 2, ; and (c) epoch 3, The contour interval is 0.4 hpa. Negative contours are dotted, and the zero and positive contours are solid. Northernmost latitude is 30 S, and grid lines marked are 40, 60, and 80 S. potential height of ERA-40 reanalysis, which they attributed to both ENSO and SAM. The SLP trend toward lower pressure shown in Figure 7 is also consistent with the findings of Thompson and Solomon [2002], who suggested the possibility that the trend between 1958 and 2001 deduced from the NCAR/NCEP reanalysis is associated with enhancement of the lower stratospheric wintertime polar vortex over that period. Since it appears that the West Antarctic region is influenced differently by ENSO and lower-frequency influences such as this SLP trend, an analysis that separates high- and low-frequency variations is warranted. 4. ENSO Frequency Versus Low-Frequency Circulation Anomalies [27] A band-pass filtering of the data was performed where the ENSO periods were separated from the data by selecting only those Fourier coefficients with periods from 2 to 8 years in the time series of the d 18 O and SLPs. As noted earlier, the periods that appear to represent ENSO are different for the three different epochs (with the upper bound having shorter period later in the record) but the filtering was performed with the 2 8 year band for consistency and because it captures most of the ENSO activity for all epochs. The lower-frequency portion of the data was determined by combining all periods greater than 8 years. The reader is reminded to view results with caution because of the limited number of high southern latitude observations in the first two epochs, while also noting that global SLP patterns related to ENSO are captured by the EOFs used in the optimal interpolation. [28] Figures 8a 8c show the regression of SLP on the core d 18 O for ENSO periods, and Figures 8d 8f show the regression of periods greater than 8 years. The first important feature to note from these spatial patterns is that the similarity in the spatial structure between the highfrequency regression maps (Figures 8a 8c) and the unfiltered correlations (Figures 6a 6c) suggests that a significant portion of the covariance of the two signals is due to the high (ENSO) frequencies. This is seen in the difference in the magnitudes of the high- and low-frequency regressions in Figure 8 (the contour interval differs from 0.1 hpa/% in the high frequencies to 0.6 hpa/% for the low frequencies). [29] Referring to the ENSO frequencies (Figures 8a 8c), again in epoch 2 (Figure 8b), the pressure center in the vicinity of the Antarctic Peninsula regresses positively with the d 18 O, characteristic of El Niño, while in the first and third epochs there is a low-pressure center in that region, characteristic of La Niña. Figure 8b also shows evidence of a wave train that links the epoch 2 pressure in the Peninsula region with the western Pacific, along a great circle trajectory, consistent with Karoly s [1989] description of an ENSO-related wave train. The cyclonic anomalies found in the ABS in epochs 1 and 3 suggest enhanced moisture advection to the ice core site on ENSO time scales. Also, while the negative pressure at the pole in epochs 1 and 3 is equivalent to a positive SAM index, the opposite is true in epoch 2. This is again consistent with the Fogt and Bromwich [2006] finding that strong teleconnections are associated with the SOI and SAM being in phase. The influence of SAM on the d 18 O thus opposes the influence associated with the advection from lower latitudes, where the positive SAM index implies a colder pole, and therefore a negative isotope anomaly due to temperature, but the positive SOI will advect isotopically enriched moisture at some meridians, such as over West Antarctica. This is also consistent with L Heureux and Thompson [2006] who find that ENSO projects strongly onto the SAM in the months of October through February and that it is La Niña events that positively enhance the westerlies at approximately 60S, equivalent to increasing the SAM index. In fact all three epochs of ENSO frequencies have polar pressure that is in phase with the ENSO-related pressure center in the South Pacific. This is effectively identical to the conclusion of Fogt and Bromwich [2006], although analyzed in a different temporal framework, and suggests that SAM is a critical element of ENSO s teleconnection to the isotopic content of the ice cores in West Antarctica. 9of17

10 Figure 8. Austral spring season regressions. (a c) High-frequency (2 8 year periods) regressions of SLP on the ITASE d 18 O isotope. (d f) Low-frequency (8+ year periods) plus trend regressions of SLP on the ITASE d 18 O isotope. Shown are results for epoch 1 (Figures 8a and 8d), epoch 2 (Figures 8b and 8e), and epoch 3 (Figures 8c and 8f). The contour interval is 0.1 hpa/% for Figures 8a 8c and 0.6 hpa/% for Figures 8d 8f. Negative contours are dotted, and the zero and positive contours are solid. Northernmost latitude is 30 S, and grid lines marked are 40, 60, and 80 S. [30] Figures 8d 8f show the SLP-d 18 O regressions associated with periods greater than 8 years. In the first epoch s low frequencies (Figure 8d) the feature of the regression map which appears very strong is the zonal wave 3, and a negative pressure (positive SAM index) over the pole. This spatial regression pattern does not provide clear rationale for enriching isotopes but the HadSLP2 is known to be poorly constrained in this epoch. The circulation centers of the wave 3 in Figure 8d are not aligned to advect enriched vapor from the ocean to the core site, and the low polar pressure regression would be indicative of anomalously cold polar temperature, thereby depleting the isotopic content of the ice core. Although the direct interpretation of this pattern does not support enrichment, the positive SAM index regression is coincident with the positive correlation between SOI and the d 18 O. This lowfrequency pattern is qualitatively consistent with the Fogt and Bromwich [2006] finding that the SOI and SAM must be in phase for strong teleconnection between the tropics and West Antarctica. [31] As with the ENSO frequencies, it appears that the low-frequency portion of epoch 2 (Figure 8e) has a persistent blocking feature in the Bellinghausen Sea and positive regression at the pole, not altogether different from the mean SLP anomaly for the epoch. This suggests that the d 18 O increases in conjunction with both the high mean SLP (Figure 7) and the positive low-frequency pressure anomaly in the region, despite the orientation of the circulation feature which would likely deplete the isotopes. Without other influences, both d 18 O and SLP will have a positive anomaly when there is anomalous polar warmth and negative anomalies during cool periods, as previously discussed in association with SAM. Investigating this further, a SAM index is constructed by taking the difference of the zonal mean pressure between 40 S and 70 S. Since this discussion is in reference to the second epoch, during which observations are quite limited, independent validation of the relationship between the epoch isotope and SAM is desirable. The HadSLP2 derived SAM index as well as one from Jones and Widmann [2003] (JWSAM) are correlated with the d 18 O isotope in Figure 9. The two SAM indices have similar 20 year moving correlation to the isotope after about 1935 and we see that the SAM index correlates negatively with the isotope during the second 10 of 17

11 Figure 9. Correlation between d 18 O and SAM indices from Jones and Widmann [2003] and HadSLP2. epoch, reinforcing that the positive (negative) isotope anomalies are accompanied by positive (negative) polar pressure which will be accompanied by warmer (cooler) polar temperature. While there is a lack of observations in the Southern Hemisphere which is a fundamental limitation in this analysis, the fact that reasonable explanations of the ice core data have emerged provides some measure of confidence and validation of the large-scale patterns in the HadSLP2 data. [32] The low-frequency regression map of the third epoch (Figure 8f) shows the negative center of action in a similar location to the positive one in the second epoch while there is weak positive pressure over the pole. The d 18 Oin this epoch generally increases with time as does the surface temperature at several coastal stations as shown by Thompson and Solomon [2002]. The positive low-frequency polar pressure regression may be argued, in a fashion similar to epoch 2, to be associated with a mean temperature change that gives rise to the link between the isotope signal and the pressure. The negative pressure regression region in the west represents an anomaly that may advect water from lower latitudes to the ice core site. Notice in this epoch that each of the mean SLP, the low-frequency ADP-like pattern, and the high-frequency variations all reinforce one another in the regions of the ABS and thus lead to a very strong positive d 18 O anomaly. A similar reinforcing superposition occurs in epoch 2, but with opposite sign, while there is some degree of cancellation among the contributing patterns in epoch 1. This provides additional, albeit qualitative, explanation of the lower d 18 O correlation with SOI in the first epoch. [33] As noted above, the reinforcement and decadal variability is intriguingly analogous to results of Fogt and Bromwich [2006]. Their analysis of the differences in ENSO to Antarctic teleconnection between the 1980s and 1990s shows that when the SAM and SOI were in phase during the 1990s, there was a much stronger, statistically significant, ENSO-related center of pressure in the vicinity of the ice cores studied here. With the possible exception of the very weak epoch 3 low-frequency pattern, all of the spatial patterns shown in Figure 8 represent a SAM that is in phase with the SOI. In the isotopic composition of the ice cores, the ENSO related ABS circulation anomaly and the SAM oppose each other as the positive SAM will represent a colder, isotopically depleted, continental interior, but the advection of water from lower latitudes will be isotopically enriched. The Fogt and Bromwich [2006] study looks very closely at the year-to-year and the 1980s to 1990s variations in the SOI, SAM, and spatial patterns from ERA 40 reanalysis. While their time scale of study is different from the time scales studied here, it seems quite likely that a similar relationship between SAM and ENSO on a multidecadal time scale emerges from the present study. Comparing the 20 year moving correlation of the isotope record with SOI (Figure 3c) and with JWSAM (Figure 9), they are similar correlation time series. The correlation of the d 18 O with SOI and with JWSAM are nominally in phase which is another strong indicator of interaction between SAM and SOI, specifically that they must be in phase to show strong teleconnection to Antarctica [Fogt and Bromwich, 2006]. 5. Role of the Pacific Decadal Oscillation [34] The observations thus far suggest a decadal or multidecadal oscillation of the relationship between ENSO and isotopes in the ice cores, and therefore some consideration of the Pacific Decadal Oscillation (PDO) is needed. The PDO index represents mean SST anomalies of the North Pacific (north of 20N) and the PDO has a signature that bears some spatial resemblance to ENSO but behaves quite differently over time. Zhang et al. [1997] studied the ENSO-like interdecadal variability, focusing on the Northern Hemisphere. They determined that there are ENSO-like SST features that propagate to high (northern) latitudes, but the time scales tend more toward decadal scales with increasing latitude. Zhang et al. [1997] performed highand low-pass filtering with a cutoff at 6 years, and found that there was ENSO-like spatial structure in the low frequencies. This Northern Hemisphere result is consistent with that for the Southern Hemisphere found here for the second and third epochs. Zhang et al. [1997] also discussed pronounced shifts that occur around 1942/1943 (to a cool PDO phase) and around 1976/1977 (to a warm PDO phase), times similar to the epoch 2 and 3 onset times discussed in this work. [35] Figure 10 shows the PDO index, obtained from the Joint Institute for Study of the Atmosphere and Ocean at the University of Washington, for the 20th century with a 20 year moving average performed on each season. The PDO index shows a downward trend during the second epoch, but the onset of the second epoch appears to precede the shift to the negative phase of the PDO, which begins in approximately 1945 in the 20 year moving average. The downward trend in the PDO index is also preceded by the beginning of the negative correlation of the d 18 O with SOI which begins around 1930 (Figure 3). Much of the second epoch is in the negative phase of the PDO and we speculate the peculiarity of the second epoch s d 18 O signal may be linked to the shift of the PDO index to negative phase. The timing does not coincide with the timing of correlation between the d 18 O and SAM (Figure 9), 11 of 17

12 Figure 10. The 20th century Pacific Decadal Oscillation index. Shown are the seasonal 20 year moving averages. which appears to be closer in phase to the correlation between d 18 O and SOI (Figure 3). In fact the timing of the correlations suggests that the decadal teleconnection between the SOI or SAM and the ice core, if dependent on the PDO, is related to the rate of change of the PDO. [36] Figures 11a and 11b show the wavelet analysis on the PDO and the cross spectra of d 18 O with PDO. These reveal that during the second epoch, the PDO index captures variability at frequencies in the 2 8 year (ENSO) period range while it appears to do so to a much lesser extent during the other two epochs. The cross spectrum with d 18 O similarly implies that during that second epoch the two are coherent at ENSO frequencies but not during the other two epochs. Since the PDO captures a higher-latitude signal (poleward of 20 N), as does the ice core (but for the southern hemisphere), it appears that ENSO signals propagate to high latitudes of both hemispheres during the second epoch, but propagate strongly only to the southern high latitudes in the first and third epochs. [37] To investigate the high-latitude circulation anomalies associated with PDO, the regression of SLP on the PDO index is shown in Figure 12. The regression was performed with the 5-year moving average of each the SLP and the PDO for the entire 20th century. The resulting pattern is distinctly similar to the epoch 2 low-frequency SLP- d 18 O regression in Figure 8e and similar but of opposite sign to epoch 3 (Figure 8f), which also bear some resemblance to their high-frequency ENSO portion. This shows that the low-frequency d 18 O signals in the second epoch are associated with circulation anomalies that are similar to the anomalies typically associated with the PDO. The regressions presented here do not explain the phase reversal of the second epoch of d 18 O relative to ENSO, but tell us that the PDO related SLP pattern represents the primary SLP pattern associated with the d 18 O isotope variability in epoch 2 when the SOI and d 18 O are negatively correlated. 6. Relative Influence of Four Climate Variables on Isotopic Signals [38] Several climate variables and time scales have been discussed in the interpretation of ENSO signals in the isotopic composition of the West Antarctic ice cores. To compare the influences of global temperature, SAM, PDO, and ENSO on the ITASE , , and ice cores they are correlated at varying time scales up to 25 years. The correlations are performed with monthly centered averaging for each time scale, and the significance of the correlations is based on degrees of freedom equal to the number of years total divided by the time scale in years. The global temperature record from the Climatic Research Unit (available from is used for the present analysis and its accuracy is also limited by sparseness of measurement in the southern hemisphere before Both versions of the SAM index, HadSLP2 and Figure 11. (a) Wavelet power spectrum of the Pacific Decadal Oscillation index in units relative to PDO variance. (b) Wavelet cross spectrum of PDO with the d 18 O isotope in units relative to the product PDO and d 18 O standard deviations. 12 of 17