Evidence for tropical SST influence on Antarctic polar atmospheric dynamics



[1] A Singular Value Decomposition analysis is applied to climatological data to determine the modes of variability of monthly mean Sea Surface Temperature (SST) in the tropics coupled with the southern hemisphere stratospheric polar vortex intensity for the 1958 to 2006 period and to identify the tropical region strongly influencing the high latitude stratospheric dynamics. Two subsets of data have been considered by explicitly taking into account the phase of the Pacific Decadal Oscillation (PDO). The study evidences a shift of the tropical region affecting the polar vortex from the Western Pacific to the Indo-Pacific Ocean, driven by the modulation of the SST pattern, due to the phase of the PDO. The analysis shows in both cases (years with positive/negative PDO phases) a high degree of correlation between an SST related index, calculated from NOAA/ERSST and HadISST datasets, and an index of the polar vortex intensity based on ECMWF data.

1. Introduction

[2] El Niño-Southern Oscillation (ENSO) is known to be the leading mode of atmosphere-ocean coupled variability on interannual time scales in the tropical Pacific and can substantially affect the global climate. Early observational works [Van Loon and Labitzke, 1987; Baldwin and O'Sullivan, 1995] dealing with the influence of tropical Sea Surface Temperature (SST) variations on the stratospheric polar vortex provided evidence for an enhancement of waves propagating into the northern winter stratosphere during ENSO events. García-Herrera et al. [2006] studied the differences in the temperature response between El Niño and La Niña cases in two models and in the European Center's 40-year reanalysis (ERA-40). They evidenced that the ENSO signal affects the middle atmosphere by means of ultralong Rossby waves. Modeling works that investigate the influence of SST on the polar stratosphere showed a correlation between warm ENSO events and more disturbed winter polar stratosphere in the Northern Hemisphere (NH) [Sassi et al., 2004; Taguchi and Hartmann, 2006].

[3] However, several observational studies indicated that the most important role in forcing the El Niño teleconnection is not played by the strongest SST anomalies but by the SST anomalies investing the Tropical Western Pacific (TWP). Hamilton [1988] found that SSTs in the TWP are a major factor in determining the winter NH atmospheric response to El Niño, with the response being most pronounced when the far Western Pacific is anomalously warm. These results are supported by modeling experiments indicating that SST and diabatic heating anomalies in the TWP can excite Rossby wave trains propagating into the Northern Pacific [Hoerling and Kumar, 2003; Quan et al., 2006; Lau and Nath, 2003]. For the Southern Hemisphere (SH), Harangozo [2004] showed that the source of the Rossby waves reaching higher latitudes is not the area of the highest SST anomaly associated with an El Niño and that the efficiency of deep tropical convection in the Intertropical Convergence Zone (ITCZ) close to the dateline can play a key role on the SH extratropical circulation. Also, Lachlan-Cope and Connolley [2006] highlighted that SST anomalies produced by different El Niño events can force Rossby waves in very different ways, despite being of the same magnitude, depending on the location of the anomaly relative to the Hadley circulation.

[4] All the above discussed works stress the idea that the large scale perturbations induced by an El Niño event are not strictly related to the event itself but to the spreading of its anomalies in the Western Pacific Ocean, suggesting also a convective origin of the SST induced waves, being the convective heating in the tropical troposphere the main source of energy and momentum in the tropics [see, e.g., Ricciardulli and Garcia, 2000]. However, an El Niño event typically occurs irregularly 3 to 7 years apart persisting for 6 to 18 months and therefore it is not the best candidate to systematically (year by year) affect the polar vortex dynamics. Together with El Niño, the Pacific Decadal Oscillation (PDO) [Mantua and Hare, 2002] is a major climate cycle affecting the Pacific Ocean. While El Niño is not an oscillation but an intermittent event, the PDO can be seen as a long lived El Niño-like oscillating pattern of climate variability in the tropical Pacific (reversing its polarity every about 20–30 years), then able to produce persistent SST anomalies in the tropical Pacific Ocean, even if they represent only a secondary signature of the PDO, while the climatic fingerprints are most visible in the North Pacific. Several independent studies [see, e.g., Trenberth, 1990] find evidence, during last century, of a “cool” PDO regime that prevailed from 1947–1976 and a “warm” PDO regime that dominated from 1977 through at least the mid-1990's.

[5] On the base of model simulation results, Grassi et al. [2006] described an atmospheric teleconnection mechanism between the tropical and the southern polar latitudes that is activated during June by the propagation toward the winter hemisphere of tropical convectively induced waves, able to produce a preconditioning of the polar vortex and strongly affecting the following evolution of the polar dynamics.

[6] Also, Grassi et al. [2008] suggested the westward shift of the El Niño SST anomalies [McPhaden, 2004] as the possible cause of the increased wave generation leading to the particularly unusual evolution of the southern polar vortex [Newman and Nash, 2005]. That work points again the attention on the convective processes activated by the SST anomalies at the beginning of the winter due to their ability to affect the whole evolution of the polar vortex by perturbing its early phase.

[7] The main goals of the present study are: 1) to seek evidence, in atmospheric climatological data, of the oceanic driving on the SH polar atmospheric dynamics exerted by the SST anomalies of June, as suggested in our previous modeling works; 2) to gain insights into the tropical region playing a key role in the tropical/polar latitude connection. To this end, a Singular Value Decomposition (SVD) analysis [Bretherton et al., 1992] is applied to diagnose the impact of year by year variations of the tropical SST of June on the SH polar vortex during the first half of September (i.e., in a phase when its intensity is supposed not yet strongly affected by radiative effects related to the ozone depletion) in terms of pattern of co-variability and degree of correlation.

2. Methods and Data

[8] Data from different sources are used in the analysis. Wind fields are from ERA-40 reanalysis [Uppala et al., 2005] that includes winds at standard levels up to 1 hPa for the 1958 to 2002 period. These data are derived from a sophisticated three-dimensional assimilation of conventional and satellite data using a 60 level T159 medium range global weather forecast model. Although ERA-40 data are available starting in 1958, all important satellite data only became fully available in 1978 and doubts have been raised on the reliability of data for the Antarctic polar vortex in the pre-satellite period. However, ERA-40 reanalysis data are generally skilful at representing the most important aspects of the polar dynamics, at least in the stratosphere, even if the possibility of the presence of an early cold bias in the Antarctic stratosphere has been suggested [Renwick, 2004]. Operational analyses from ECMWF (European Centre for Medium-range Weather Forecasts) are used for the 2002–2006 period. Overall, the dynamical data used in the study consist of zonal component of the wind field from 1958 to 2006 at 30 hPa. Two different datasets for SST are considered. HadISST1.1 dataset (Hadley Centre Global Sea Ice and Sea Surface Temperature) [Rayner et al., 2003] provides monthly mean SST climatology from 1870 to present at a resolution of 1° area grid. HadISST1.1 analysed data include both sea surface observations and satellite derived estimates at the sea surface. SST data from NOAA/ERSST.v3 (National Oceanic and Atmospheric Administration's/Extended Reconstruction SST version 3) [Smith et al., 2008] are also used. These data are given for the period from 1880 to 2006 on a 2° grid. The first version of the ERSST [Smith and Reynolds, 2003] was produced based on Comprehensive Ocean-Atmosphere Data Set (COADS) observations from 1854 to 1997 period, together with quality control procedures and improved statistical methods to produce stable reconstruction using sparse data. Even in the period with satellite data, the ERSST.v1 analysis is always computed by a fit to in situ only data. The new analysis, ERSTT.v2 [Smith and Reynolds, 2004] is an improvement over version 1 because of its stronger variance in the tropical Western Pacific, its inclusion of ice concentration information, and its improved error estimates. A systematic comparison of HadISST and ERSST.v2 evidences a good agreement between the two datasets with an exception occurring in the 60°–23°S annual averages. The biases in satellite data, incorporated in the HadISST to help the analysis where data are sparse, have been suggested as a possible cause of the cool bias shown by HadISST relative to the ERSST.v2 in the Southern Ocean, in the period after 1980 [Smith and Reynolds, 2004]. Beginning in 1985, ERSST.v3 is further improved by also explicitly including bias-adjusted satellite infrared data from AVHRR (Advance Very High Resolution Radiometer) on board of NOAA satellites. Only the SST data for the period covered by ERA-40 analysis (i.e., from 1958) are used in this study.

[9] A SVD analysis is performed on the cross-covariance matrix between the SST anomalies and the anomalies of the overall polar vortex intensity (diagnosed by means of a chosen index calculated as described below), to identify modes of behaviour in which the variations of those variables are strongly coupled [Bjornsson and Venegas, 1997].

3. Results

[10] The SVD analysis has been performed by taking into account tropical SST and zonal wind anomalies. We use maps of the monthly mean SST anomalies for June over the tropical region (i.e., from −30°S to 30°N), from 1958 to 2006 and for both NOAA/ERSST and HadISST datasets. Two subsets of data have been considered by selecting years characterized by positive or negative PDO phase (thereafter PDO+ and PDO-, respectively), on the base of the PDO index provided by the Joint Institute for the Study of the Atmosphere and Ocean (JISAO) (http://jisao.washington.edu/). To identify the overall variability of the polar vortex, following Newman and Nash [2005], an index is calculated by space averaging, over 20°–90°S, the zonal mean of the zonal wind anomalies at 30 hPa. This value, time averaged over the first half of September, is assumed as an index of the Polar Vortex Intensity (hereafter PVI). The chosen index does not show a sharp discontinuity related to the advent of satellite observations. This could follow from the time period used to calculate the PVI index, selected in such a way that the ozone depletion does not yet affect the vortex intensity. Such choice could probably mitigate the effect on the index of the Antarctic stratosphere cold bias affecting ERA-40 data, which has instead a stronger impact during spring, when the ozone depletion is activated, as a consequence of mechanisms of feedback starting from an anomalous growth of PSCs (Polar Stratospheric Clouds) [Karpetchko et al., 2005]. Figure 1 shows the heterogeneous correlation maps (generally defined as the vectors of correlation values between the expansion coefficient of one field expansion mode and the grid point values of the other field we want to correlate to it) between the time series of the bidimensional map of the tropical SST and the time series of the PVI index. Calculations are performed for two subsets of data characterized by opposite PDO phases (PDO+, top level; PDO-, bottom level) and for both SST datasets (HadISST, right column, NOAA/ERSST left column). In our case, being the PVI index a mono-dimensional time series, the heterogeneous maps at each grid point directly represent the correlation coefficient between the time series of the SST anomalies at each grid point and the PVI index. Colour shades in Figure 1 indicate the value of the heterogeneous vector at each grid point where the correlation is significant at the 95% level. The significance of the correlation is computed using a Monte Carlo approach, as described by Wallace et al. [1992]. The SVD analysis is repeated 100 times linking the original SST dataset with randomly “scrambled” ensemble of the PVI index time series. The desired significance level is assessed, for each grid point, from the distribution of the 100 resulting correlation coefficients.

Figure 1.

Heterogeneous maps from SVD analysis between tropical SST for June and PVI index (see text for details), calculated with both (left) NOAA/ERSST and (right) HadISST datasets for the 1958 to 2006 period. Top plots refer to PDO+ case, while bottom plots are for PDO- case. Color shades indicate regions where the value of the heterogeneous vector shows a correlation significant at the 95% level.

[11] Heterogeneous maps show maximum values between −0.8 and −0.6, indicating a good coupling in all cases. However, the analysis shows different patterns for cases PDO+ and PDO-. In the first case (PDO+, upper plots), similar results are found for both NOAA/ERSST and HadISST, with positive values in a small region of the Central Pacific Ocean at equatorial latitudes and negative values over the Western Pacific Ocean and the Southern latitudes of the Indian Ocean. The only exception occurs in the Southern Pacific Ocean around 100°W where the map for NOAA/ERSST evidences a large significant region with negative values not found when HadISST data are used. In the second case (PDO-, bottom plots), the analysis highlights a large region with statistically significant negative values, showing a tendency to invest mostly the zones corresponding to positive SST anomalies when the PDO is in a negative phase. This region covers the Indian and the Northwestern Pacific Ocean and also some sectors of the Southern Pacific and Atlantic Ocean. Results for the two SST datasets are again quite similar even if, also in this case, a stronger correlation is found in the Southern Pacific Ocean (with a maximum around 120°W) when NOAA/ERSST are used. The recalled cool bias affecting HadISST in the Southern Ocean south of 20° of latitude, in the period after 1980, could probably explain this discrepancy between HadISST and NOAA/ERSST results.

[12] The regions characterized by statistically significant negative values in the heterogeneous maps have been assumed to be active in terms of convective wave generation, and the value of the SST anomalies, averaged over the grid points of those areas, has been used as an index of Convective Processes Activation (hereafter SST_CPA). These assumptions are based on the strong anti-correlation shown in Figure 1, suggesting an active role of the highlighted region tropical SST in the high latitude dynamics, driven, as suggested by previous observational and modeling studies [see, e.g., Salby and Garcia, 1987; Harangozo, 2004; Lachlan-Cope and Connolley, 2006; Grassi et al., 2006], by the SST anomalies through the activation of convective processes. Our hypothesis seems also consistent with tropical maps of the convective heating, showing maximum values in the Indian and Western Pacific Ocean [see, e.g., Ricciardulli and Garcia, 2000].

[13] Figure 2 shows the scatter diagrams of SST_CPA (HadISST, right column, NOAA/ERSST left column; PDO+, top level, PDO−, bottom level) and PVI index. All the plots show similar values for the correlation coefficient (around −0.8) and about the same equation for the linear fit that indicates a similar efficiency of the SST forcing. Each point in the scatter plot is identified by the number representing the corresponding year. The linear regression fit and its corresponding equation give indications on the forcing exerted by the tropical SST on the polar vortex and may be used as a prognostic tool for the overall SH polar vortex intensity at the end of winter/beginning of spring, once the SSTs in June are known. In particular, it could help in predicting unusual polar vortex condition, as in the 2002 case. The maximization of the 2002 SST_CPA index, produced by NOAA/ERSST (left bottom plot), seems to confirm the relationship, suggested by Grassi et al. [2008], between the highly weak polar night jet and the unusual tropical oceanic conditions. The SST_CPA value for 2002, even bigger than SST_CPA value for 1997, an year characterized by an El Niño of stronger intensity, indicates a maximization of the tropical forcing that could have followed from the anomalous pattern of 2002 El Niño, characterized by westward shifted SST anomalies reaching the Pacific sector that our SVD analysis identified as particularly “sensitive” to convective process activation. The high SST_CPA value puts the point representing the 2002 in a position in the scatter diagram that suggests a behaviour consistent with the overall behaviour of other years, even in presence of strongly unusual polar vortex conditions. For HadISST data, the value of the SST_CPA coefficient has not a maximum in 2002 and this seems again related to the Southern Pacific cool bias. The uncertainty in the PDO phase, that seems to have shifted in a next cool phase from the end of last century, could probably explain the tendency of the latest years to spread from the linear fit in the graphics.

Figure 2.

Graphics of the correlation between SST_CPA index (K) and PVI index (m s−1) (see text for details). Data are for the 1958 to 2006 period (top, PDO+; bottom, PDO-) and for (left) NOAA/ERSST and (right) HadISST. The linear regression fit and its corresponding equation, and the value of the correlation coefficient between the two index time series are also shown.

4. Conclusions

[14] The analysis of the coupled year by year variability since 1958 of the tropical SST in June and the high latitude stratospheric zonal wind anomalies in September has been performed using data from both NOAA/ERSST and HadISST. Results evidence coupled modes of variability driven by the tropical SST anomalies of June in the Western Pacific or the Indo-Pacific Ocean, respectively, when years with positive or negative phase of PDO are considered. However, NOAA/ERSST produces, with respect to HadISST, a stronger contribution of the Southern Pacific Ocean to the coupled mode of variability. The tropical regions characterized by statistically significant negative values of the SVD heterogeneous maps have been assumed as the most able to affect the polar vortex dynamics. Based on the results from previous observational and modeling studies, averaged SST anomaly, calculated over the selected regions, has been used as an index of the convective process activation. The study of the co-variability between this index and an index of the overall anomaly of the polar vortex intensity, calculated for PDO+ and PDO- subsets of data, shows a very good correlation for both HadISST and NOAA/ERSST, with a correlation coefficient equal to about −0.8. This result indicates that the SST_CPA index can be a good alias for the convective process activation. At the same time, it gives experimental evidence in support of the teleconnection mechanism between the Tropical Oceans and the high latitude atmospheric dynamics, suggested by modeling studies and supposed to have driven also the particularly unusual evolution of the polar vortex in 2002.


[15] NOAA/ERSST.V3 data are provided by NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at http://www.cdc.noaa.gov/. UK Meteorological Office, Hadley Centre HadISST 1.1 data are available from the British Atmospheric Data Centre Web site at http://badc.nerc.ac.uk/data/hadisst/. PDO index is provided by the Joint Institute for the Study of the Atmosphere and Ocean (JISAO) (http://jisao.washington.edu/). We also thank anonymous referees for their constructive suggestions and comments. This work has been partly supported by Italian Space Agency.