Decadal variability in the North Atlantic has been associated in the literature with a tripolar pattern of sea surface temperature (SST) anomalies that show one sign in the western midlatitudinal North Atlantic and the opposite in the subpolar and tropical North Atlantic. The present analysis of observed SST from 1870 to 2009 leads to the dissection of the SST tripole into two components, each with a different time scale in the decadal band and different teleconnections in the Atlantic basin; while the subpolar and tropical poles present quasi-decadal variations with a period of about 9 years, essentially uncorrelated with other parts of the basin, the center of action in the western midlatitudes is characterized by a longer time scale of about 14 years and significant correlations with the tropical South Atlantic and the Norwegian and North Sea(s). The 9-yr period variations are associated with an atmospheric configuration resembling the east Atlantic pattern, whereas the 14-yr period fluctuations seem to be related to the North Atlantic Oscillation pattern. Each component also bears a different relationship with the decadal variability in the Pacific Ocean.
The sea surface temperature (SST) of the North Atlantic Ocean displays large variability on multidecadal and decadal time scales, which has received profuse attention in recent years, on account of its relevance to the long-range potential predictability of climate variations in the North Atlantic sector (Latif et al. 2006). Distinct spatial structures have been associated to each of those time scales in analyses of the observed North Atlantic SST (Deser and Blackmon 1993; Kushnir 1994): a monopolar pattern with anomalies of the same sign that cover the entire basin is peculiar to the multidecadal variability, whereas the decadal fluctuations feature a tripolar configuration with SST anomalies of one sign in the midlatitudes, extending from the coast of North America into the center of the basin, surrounded by anomalies of the opposite sign in the tropical band and in the subpolar North Atlantic. This pattern is similar to the SST tripole forced by the North Atlantic Oscillation (NAO) (Cayan 1992; Visbeck et al. 2003).
The North Atlantic SST multidecadal variability appears to be related to changes in the thermohaline meridional overturning circulation, which affect the heat exchange between the North and South Atlantic, yielding an out-of-phase relationship between their respective SST anomalies on the multidecadal time scale (Latif et al. 2004). On the other hand, the mechanisms of the decadal North Atlantic variability as well as its expression in the South Atlantic seem to be much less clear. Concerning the former, a number of proposals have been put forward, emphasizing different processes, from the ocean’s response to stochastic atmospheric forcing to coupled ocean–atmosphere modes, which may lead to decadal fluctuations (a review can be found in Marshall et al. 2001b). As to the connection with the South Atlantic, while Houghton and Tourre (1992) found no significant correlation between the SST anomalies north and south of the ITCZ on the decadal time scale, Xie and Tanimoto (1998) suggest that decadal variations in the North and South Atlantic are part of a coherent pan-Atlantic decadal oscillation. Additionally, connections between North Atlantic and Pacific decadal variability have been recently stressed by Müller et al. (2008), who find a link between the North Atlantic SST tripole and the decadal El Niño–Southern Oscillation (ENSO)-like variations in the Pacific.
In this study, we focus on North Atlantic variability on the decadal time scale and try to gain insight into its features by analyzing separately the SST anomalies in two of its three characteristic poles: the one in the subpolar North Atlantic and the one off the eastern coast of North America. Our results suggest that their decadal fluctuations are mostly uncorrelated with each other and present different teleconnections with the South Atlantic and with the Pacific oceans.
2. Data and methodology
Our analysis considers the observed SST fields from 1870 to 2009, taken from the merged Hadley Centre Sea Ice and SST dataset version 1 (HadISST1)–optimum interpolation version 2 (OI.v2) dataset (Hurrell et al. 2008). We compute annual anomalies with respect to the mean of the entire 1870–2009 period and construct two indices from these data: the area-averaged SST anomaly over 40°–60°N, 50°–20°W and over 25°–40°N, 75°–45°W, respectively corresponding to the centers of action of the SST tripole in the subpolar (SUBP) and in the western midlatitudinal (WESTM) North Atlantic. We isolate the decadal variability in these areas by applying a 5–20-yr-period bandpass filter (Kaylor 1977) to these indices. Filtering is limited to these two time series throughout this study but for a preliminary empirical orthogonal function (EOF) analysis that is conducted on bandpass-filtered (5–20 yr) North Atlantic SST anomalies.
The spatial SST structures related to the decadal fluctuations of SUBP and WESTM were determined by means of linear regression of the annual SST anomalies on the bandpass-filtered version of those two indices. Prior to the regression analysis, the SST anomalies were detrended by removing their projection on the 21-yr running mean of the globally averaged SST anomaly, following the procedure suggested by Ting et al. (2009). The statistical significance of the regression coefficient at each location is assessed by means of a moving block bootstrap procedure (Wilks 1997). The SST anomaly time series is randomly resampled in blocks (with replacement) to obtain a large number of synthetic time series from the original set of observations, which are then used to build confidence intervals for the regression coefficients on the bandpass-filtered SUBP and WESTM indices. We have taken 10 000 resampled time series and a block length of 5 yr. We have checked our results are robust under the choice of this parameter by trying block lengths of 3 and 10 yr as well. In the next sections, regression maps display results at a significance level of 90%, which we establish so as to illustrate the lack or marginality of significant features in some areas even at this level. It must be noted, though, that elements highlighted in the text as significant at the 90% level are also significant at the 95% level.
The associated sea level pressure (SLP) patterns were identified applying the same linear regression procedure described above, including the moving block bootstrap test, except that no trend is removed from the data. These correspond to the period from 1870 to 2009 and were taken from the Hadley Centre Sea Level Pressure (HadSLP2r) dataset (Allan and Ansell 2006). We considered annual anomalies with respect to the 1870–2009 mean.
Annual wind stress anomalies over the North Atlantic from 1949 to 2006, computed from monthly data belonging to the Common Ocean Reference Experiment (CORE.2) global air–sea flux dataset (Large and Yeager 2009), have also been subjected to a linear regression analysis following the lines detailed above.
Additionally, cross-correlations have been calculated between the bandpass-filtered SUBP and WESTM indices, respectively, and an annual, station-based (Ponta Delgada, Azores–Stykkisholmur/Reykjavic) NAO index provided by the Climate Analysis Section, National Center for Atmospheric Research (NCAR) (Hurrell 1995). Statistical significance has been evaluated again through a moving block bootstrap test with the parameters indicated above.
3. Deconstructing the tripole
a. EOF analysis of bandpass-filtered (5–20 yr) Atlantic SST anomalies
The first EOF (EOF1) of the bandpass-filtered North Atlantic SST anomalies (Fig. 1a) explains about 42% of the variance of the field and displays the characteristic tripolar configuration. If the fraction of variance explained by this EOF at each location is considered, it is found that it accounts for over 30% of the local variance in the SUBP region and in the tropical North Atlantic (TNA) but explains less than 15% in most of the WESTM area. This situation is reversed for EOF3 (Fig. 1b), which explains about 11% of the total variance, more than 30% of the variance in the greater part of WESTM, and less than 15% in practically all of SUBP and TNA. The power spectra of the corresponding principal components (PCs) reveal that EOF1 and EOF3 are associated with different time scales within the decadal band. While the power spectrum of PC1 (Fig. 1c) presents a pronounced peak at a period of ~9 yr, the spectrum of PC2 attains its maximum at ~14 yr and has secondary peaks at periods shorter than 8 yr. It seems then that, despite the tripolar configuration emerging in EOF1, most of the decadal variability in WESTM is actually uncorrelated with that in the other two poles.
To pursue this issue further, we compute the area-averaged SST anomaly in the WESTM and SUBP regions and analyze the resulting time series separately. Figure 2a depicts both the detrended and the bandpass-filtered versions of these indices. As in the case of PC1, the power spectrum of the detrended SUBP index (Fig. 2b) peaks at a period of ~9 yr, attaining significance above red noise at the 95% level. The spectrum of WESTM resembles that of EOF3, instead, its peaks occurring at periods of ~14 yr and slightly shorter than 8 yr, neither significant above red noise at the 95% level. A singular spectrum analysis (SSA) (Ghil et al. 2002) (not shown) of the annual SUBP and WESTM indices reveals that, in both cases, the eigenvalues ranking third and fourth in the variance spectrum form an oscillatory pair with decadal periodicity (the first and second eigenvalues correspond to the warming trend and to multidecadal variability). The reconstruction of SUBP with this decadal oscillatory pair isolates the 9-yr period fluctuations, leaving out the lower amplitude and shorter period variations that appear between the 1920s and 1970s in the bandpass-filtered version of SUBP (Fig. 2a). This SSA reconstruction of SUBP closely matches its counterpart in the multichannel SSA analysis of Álvarez-García et al. (2008). On the other hand, the reconstruction of WESTM with its decadal SSA oscillatory pair presents fluctuations with a period of ~14 yr, which occur in the second-half of the record, being practically absent before 1940. The longer time scale in WESTM after 1940 is noticeable also in its bandpass-filtered version (Fig. 2a) and consists of the appreciation of Deser and Blackmon (1993) that the period of the decadal variations of North Atlantic SST averaged 9 yr before 1945 and increased to 12 yr afterward.
In the following regression analyses, the 5–20-yr-period bandpass-filtered versions of the SUBP and WESTM indices (hereafter SUBPDEC and WESTMDEC, respectively) are used to monitor the decadal variations in those areas. Over 1870–2009 (1940–2009), there exists a correlation of −0.36 (−0.30) between SUBPDEC and WESTMDEC. This correlation is eliminated in the regression analyses by removing from WESTMDEC its projection on SUBPDEC. Removal of this correlation, corresponding to 13% (9%) of the variance in the WESTMDEC index, does not affect our conclusions and ensures that the regression is performed on linearly independent time series. Our results also hold if SUBPDEC and WESTMDEC are obtained through the SSA data-adaptive filter instead of the specified 5–20-yr-period bandpass filter.
b. Regression analysis: Atlantic SST anomalies
The regression of the annual, detrended (unfiltered) Atlantic SST anomalies onto SUBPDEC yields the North Atlantic SST tripole (Fig. 3a), with significant (90% level) regression coefficients in SUBP, TNA, and part of WESTM. In the latter region, though, the variance explained by the regression (Fig. 3c) hardly reaches 20%. We compute next the regression of the Atlantic SST anomalies on WESTMDEC. Significant (90% level) regression coefficients are found now in WESTM and also in the Norwegian and North Sea(s) and in the tropical South Atlantic (Fig. 3b). The variance explained in WESTM by this regression rises well above 20% now and lies below 10% in SUBP and TNA.
This regression analysis was conducted separately for the second half of the record under consideration here, that is, from 1940 to 2009, because of the greater reliability of the data. The results are depicted in Fig. 4 and are essentially equivalent to those obtained from the full record. In summary, both EOF and linear regression analyses of the bandpass-filtered Atlantic SST anomalies indicate that most of the decadal variability in WESTM is not correlated with SUBP and TNA and seems to bear a stronger connection with the tropical South Atlantic and the Norwegian and North Sea(s), instead. To our knowledge, this fact had not been previously documented and could be relevant to our understanding of the mechanisms of North Atlantic decadal variability.
c. Atmospheric patterns over the North Atlantic
We consider next the atmospheric patterns appearing over the North Atlantic in connection with SUBPDEC and WESTMDEC. The regression of the annual SLP anomalies, from 1870 to 2009, on each of those indices is depicted in Figs. 5a and 5b. Both maps show a north–south contrast, but the SUBPDEC pattern is displaced southward with respect to that of WESTMDEC with significant (90%) SLP anomalies appearing in the North Atlantic south of about 20°N in the former case and not in the latter, and the maximum anomalous meridional SLP gradient across the North Atlantic occurring at a lower latitude in the first case as well. This southward displacement of the centers of action (as compared to WESTMDEC), together with the extension of the subtropical SLP anomalies north–eastward into Europe, reaching as far north as Scandinavia, makes the SUBPDEC SLP pattern very similar to the east Atlantic (EA) pattern. The configuration associated with WESTMDEC has a stronger similarity to the NAO pattern, instead. In fact, the annual NAO index presents a significant (95%) correlation with WESTMDEC, but not with SUBPDEC (Fig. 6).
The regression of the annual wind stress anomalies on SUBPDEC and WESTMDEC (Fig. 7) also reveals remarkable differences between the patterns related to each of them. While significant easterly anomalies appear in the subtropical North Atlantic in connection with WESTMDEC (Fig. 7b), the wind stress anomalies associated with SUBPDEC between 10° and 40°N are basically meridional and confined east of about 30°W (Fig. 7a). These diverse structures in the wind stress anomalies suggest some conjectures by way of explaining the SST patterns peculiar to SUBPDEC and WESTMDEC. It has been shown by Foltz and McPhaden (2006) that horizontal advection, specifically meridional advection induced by the trade winds, has a damping effect on surface-flux–forced changes in the SST in the latitudinal band between 10° and 20°N. This could justify the low, nonsignificant SST anomalies occurring in this area in connection with WESTMDEC, given its wind stress pattern: increased (decreased) poleward advection due to the stronger (weaker) easterlies would offset the effects of the corresponding negative (positive) heat flux anomalies. Such an offsetting mechanism would not operate in the case of SUBPDEC so that comparatively large SST variability in TNA is associated to it. We hypothesize that the SST anomalies in this case would stem from the mean advection of the anomalous east–west gradient induced by the meridional wind stress anomalies blowing over the eastern tropical North Atlantic, through anomalous heat fluxes and/or upwelling anomalies along the African coast. Anomalous advection could also explain the comparatively low anomalies in SUBP associated with WESTMDEC. It has been proposed by Marshall et al. (2001a) that the “intergyre gyre” oceanic response to the NAO wind stress pattern could act as a delayed (due to baroclinic adjustment) negative feedback on the SST anomalies in SUBP. An anticyclonic (cyclonic) intergyre gyre following a positive (negative) NAO phase would lead to enhanced (reduced) northward heat transport, damping the warm (cold) SST anomalies in SUBP. This mechanism led to a damped decadal oscillation in the simple model of Czaja and Marshall (2001). More sophisticated modeling studies, though, suggest this intergyre gyre constitutes a fast barotropic response to the NAO forcing (Eden and Willebrand, 2001; Eden and Greatbatch 2003) rather than a delayed baroclinic adjustment. It could then act as an immediate damping on the surface-flux-forced anomalies in SUBP, resulting in the low SST variability in this area in connection with WESTMDEC. An argument against this hypothesis comes from the consideration that the intergyre gyre would also weaken the SST anomalies in the WESTM region. We surmise, however, that such an effect could be largely overcome in WESTM by anomalous advection related with the rapid response of the Gulf Stream to the NAO (Frankignoul et al. 2001). It must be cautioned as well that, in the simulation of Eden and Greatbatch (2003), the intergyre gyre has an effect opposite to that suggested by Marshall et al. (2001a) so that the role played by this element in reality requires clarification. Finally, we speculate that the SUBPDEC wind stress pattern, with its limited occurrence of significant anomalies south of 40°N (Fig. 7a), might be unable to excite these instantaneous responses in the North Atlantic currents, thereby allowing comparatively strong SST anomalies to develop in SUBP and at the same time preventing its formation over WESTM. Further analysis is needed to test the validity of our hypotheses concerning the decoupling of SUBP and TNA from WESTM on the decadal time scale.
4. Teleconnections with the Pacific Ocean
The regression of global SLP anomalies on SUBPDEC (Fig. 8) indicates that the (negative) EA-like pattern characteristic of SUBPDEC (Fig. 8f) is preceded by ~2 yr by negative SLP anomalies over the North Pacific (Fig. 8d). These anomalies are associated with warm SST anomalies in the central tropical Pacific, extending northward along the North American coast, and straddled by cold SST anomalies at about 40° latitude in both hemispheres (Fig. 8a). The significance (90% level) of this SST structure, though, is limited to the extratropics and even there it is quite marginal. At lag 0, when warmest SST anomalies occur in SUBP and TNA (Fig. 8c), the previous patterns have given way to significant warm SST anomalies in the western Pacific between 40°S and 40°N, cold SST anomalies along the central and eastern tropical Pacific, and positive SLP anomalies in the middle of the tropical South Pacific (Fig. 8f). These patterns attain their maximum strength one year later (not shown), lagging SUBPDEC. Essentially the same evolution is found in the Pacific if the regression analysis is restricted to the second-half of the record, from 1940 to 2009, but statistical significance is lost almost completely (Fig. 9).
Figure 10 illustrates the connections to the Pacific through the cross-correlations of SUBPDEC with the Niño-3.4 index (area-averaged SST anomaly over 5°S–5°N, 170°–120°W) and an additional index defined here on the basis of the SLP pattern associated to SUBPDEC over the Pacific. We define a Pacific high index (hereafter PHI) as the area-averaged SLP anomaly over 20°–40°N, 160°–120°W spanning the mean location of the Pacific high. Over both the full record and 1940–2009, the maximum correlation between PHI and SUBPDEC occurs when the former leads the latter by 3 yr, but statistical significance (at the 95% level) is only attained in the full record calculation, a weaker correlation (significant at the 90% level) being found during 1940–2009. As to the Niño-3.4, significant (at the 95% level) correlations over 1870–2009 occur when it leads (lags) SUBPDEC by 3 yr (1 yr). In the 1940–2009 subset, these correlations are weaker, particularly that lagging SUBPDEC by 1 yr, and do not attain significance even at the 90% level.
In summary, the connection of SUBPDEC to the Pacific Ocean seems to be weak, though it has stable features that appear consistently in the analysis of the full record and, more faintly, in the examination of the more recent data, and suggest that the Pacific may lead SUBPDEC by about 3 yr. Nevertheless, further analysis is needed to confirm this relation and to describe how it operates and also to explain the additional peak in the cross-correlations obtained when the Pacific variability lags SUBPDEC by 1 yr.
Turning to WESTMDEC, its connection to the Pacific undergoes a drastic change from one-half of the record to the other, displaying a complete reversal of its correlation with the Niño-3.4 and with the PHI indices, as shown in Fig. 11. This is related with a change in the characteristic time scale in WESTM, which was already noted above and is illustrated in the power spectra of Fig. 12, computed over the first and second halves of the record. A period shorter than 8 yr dominates the power spectrum of WESTM from 1870 to 1939, with little power appearing at periods between 10 and 20 yr. From 1940 to 2009, instead, the dominant time scale corresponds to the 14-yr period, which presents a large increase in power with respect to the earlier part of the record. We focus in the following on the 1940–2009 interval and, therefore, in the teleconnections related to the 14-yr time scale in WESTM.
The NAO pattern associated to the decadal variations of WESTMDEC is seen to develop in phase with SLP anomalies in the eastern midlatitudes of the North Pacific in Figs. 13d–13f. At lag 0, these positive SLP anomalies extend over the eastern Pacific within 40°S and 40°N (Fig. 13f) and appear in conjunction with cold SST anomalies in the tropical Pacific (Fig. 13c), followed by warm anomalies along 20°–40°, in both hemispheres and then again by cold anomalies poleward of 40°N. The cold anomalies in the tropical Pacific first appear in the central-western Pacific (Fig. 13b) and then expand eastward. A clear in-phase (out-of-phase if we consider the sign of the SST anomalies in the tropical Pacific) evolution, contrasting the lagged, weaker relationship with SUBPDEC, thus links WESTMDEC with the Pacific variability as summarized by the significant zero-lag correlations with the Niño-3.4 and the PHI indices (Fig. 11).
Last, significant SST anomalies appear in the tropical South Atlantic 2 years ahead of WESTMDEC (Fig. 13a) when no anomalies have yet formed in the North Atlantic, concurring with warm anomalies in the equatorial Pacific off the American coast. This suggests the South Atlantic plays a precursory role in the development of the WESTMDEC anomalies in the North Atlantic. This is illustrated in Fig. 14, which shows South Atlantic anomalies averaged over 5°S–35°S, 50°–10°W have a maximum significant (95%) correlation with WESTMDEC at 1-yr lead. Opposed to this, SUBPDEC seems to lack such a link with the South Atlantic, as reflected in the SST patterns in Figs. 8 and 9, and in the cross-correlation in, Fig. 14, confined within the limits of the 90% confidence interval.
5. Summary and discussion
To investigate the properties of North Atlantic SST decadal variability, we have built two indices representative of variations in two foci of its characteristic tripolar pattern. We have found that decadal SST anomalies in the subpolar and in the western midlatitudinal North Atlantic are essentially uncorrelated and show different periodicities of about 9 and 14 yr, respectively. Linear regression analysis shows that the anomalies in the tropical North Atlantic covary in phase with the 9-yr-period fluctuations in the subpolar North Atlantic, forming a horseshoe pattern confined to the North Atlantic basin, while the 14-yr-period variations in the western midlatitudinal North Atlantic are linked to anomalies of the same sign in the tropical South Atlantic (leading the North Atlantic by 1–2 yr) and in the Norwegian and North Sea(s), in a configuration reminiscent of the pan-Atlantic oscillation (PADO) of Xie and Tanimoto (1998) but with only weak anomalies in the tropical and subpolar North Atlantic. The different time scales of 9 and 14 yr respectively attributed here to the tropical North and South Atlantic match the estimations given by Mélice and Servain (2003).
These signals differ in their respective atmospheric patterns over the North Atlantic. While the WESTMDEC PADO-like fluctuations seem to be related with the NAO, the SUBPDEC horseshoe SST pattern appears to be connected with the EA SLP pattern, showing a southward shift of the centers of action over the North Atlantic, as compared to the NAO, as well as a northeastward extension of the subtropical anomalies into western Europe. The disparate atmospheric configurations intimate some ideas that could explain the decoupling of SUBP and TNA from WESTM on the decadal time scale: the NAO-related wind stresses associated to the WESTMDEC variations could induce anomalous meridional advection in TNA that would inhibit the influence of the anomalous heat fluxes, in agreement with the findings of Foltz and McPhaden (2006). In SUBP, the immediate barotropic response to the NAO forcing (Eden and Willebrand 2001; Eden and Greatbatch 2003) in the form of an intergyre gyre (Marshall et al. 2001a) could also lead to an offset between the effects of anomalous northward advection and the heat fluxes. These mechanisms would not operate in the case of the SUBPDEC horseshoe pattern, given the different oceanic response that might be expected from the limited presence of significant wind stress anomalies south of 40° in its associated atmospheric pattern. The validation of these hypotheses demands further investigation, beginning with elucidation of the contradictory effects of the intergyre gyre in the works of Marshall et al. (2001a) and Eden and Greatbatch (2003).
Our WESTM and SUBP indices are somewhat similar to the TS and TN indices introduced by Czaja and Marshall (2001) to measure the meridional SST gradient (ΔT) across the Gulf Stream in the western North Atlantic. The power spectrum of WESTM (Fig. 2c) matches that of its counterpart in Czaja and Marshall (2001, their Fig. 4). As to SUBP and TN, much of the power dominating the spectrum of the latter at interdecadal time scales must have been eliminated from SUBP with the removal of the global SST trend; in shorter time scales, a peak between the 9- and 10-yr periods agrees with that in the power spectrum of the detrended SUBP index (Fig. 2b). Czaja and Marshall link the variations in ΔT to the tripolar SST pattern and to the NAO; we think that use of the ΔT index and the inclusion of the interannual variability in their compositing technique probably favor this outcome and hinder the detection of the two different patterns described here. Nevertheless, they notice indications of a southward shift of the SLP pattern at decadal time scales that might reflect the association between the SUBPDEC horseshoe SST pattern and the EA SLP pattern reported here.
The WESTMDEC PADO-like variations are not only pan-Atlantic but form part of a decadal teleconnection pattern involving the Atlantic and Pacific Oceans. SST anomalies in the tropical Pacific evolve out of phase with those in the WESTM region, and therefore with the NAO, in agreement with the results of Müller et al. (2008). This connection appears only in our analysis of the 1940–2009 data, having opposite sign in the earlier part of the record, when the WESTM region displayed variability on a shorter time scale of 7–8 yr. A change of sign in the phase relationship between the tropical Pacific and the NAO is also reported in the study of Müller et al. The structure of our WESTMDEC fluctuations closely corresponds with the 14-yr cycle described by Mo and Hakkinen (2001) including an eastward shift of the SST anomalies in the tropical Pacific, which is followed by the appearance of the anomalous SST in the South Atlantic (Fig. 13a). Mo and Hakkinen indicate that the South Atlantic SST anomalies result from a Pacific–South American wave train (Mo 2000) induced by the tropical Pacific anomalies. The SST changes in the tropical South Atlantic would then lead to the NAO pattern over the North Atlantic (Okumura et al. 2001) and consequently to the WESTMDEC SST anomalies. Thus, these decadal changes in the North Atlantic seem to constitute a South Atlantic mediated response to variations in the tropical Pacific, where different possible sources of variability on the decadal time scales have been proposed (Wang and Picaut 2004; Sun and Yu 2009).
A different teleconnection with the Pacific is displayed by the SUBPDEC 9-yr-period variations. They show highest correlations with the North Pacific SLP anomalies leading the Atlantic by about 3 yr. A similar phase relationship on decadal time scales was detected between the North Pacific index [NPI; area-averaged SLP anomaly in 30°–60°N, 160°E–140°W, as defined by Trenberth and Hurrell (1994)] and the NAO by Müller et al. (2008). In their study, this lag between the NPI and the NAO dissapears after 1940. With the indices used here, the lagged correlation still shows up after 1940 but is weaker. The Atlantic–Pacific correlation in our analysis also peaks with SUBPDEC leading the Pacific variability by 1 yr. This correlation is particularly weakened in the 1940–2009 data, as compared to the 3-yr lead of the Pacific, which suggests the latter is the actual phase relationship between SUBPDEC and the Pacific. Note that, if the 14-yr period Pacific variability were exciting a 9–10-yr-period damped oscillation in the North Atlantic with a lag of 3 yr, weaker correlations would be expected with the Atlantic leading the Pacific by 1–2 yr in addition to the Pacific 3-yr lead. A Pacific-induced enhancement of internal decadal variability in the North Atlantic was noticed by Wu and Liu (2005) in their coupled model simulations. We thus find some suggestion in our results that, while the 14-yr time scale is apparently set by the Pacific, the 9-yr period could have its origin in the North Atlantic itself. Several mechanisms have been proposed as causes of decadal variability in the North Atlantic. The stationarity of the horseshoe pattern found here, consistent with the results of Álvarez-García et al. (2008), and the weakness of the SST anomalies in the WESTM area might rule out a preponderant role of advection, either by the oceanic mean flow (Saravanan and McWilliams 1998) or due to a delayed adjustment of the ocean gyres to the wind stress forcing (Grötzner et al. 1998; Czaja and Marshall 2001; Wu and Liu 2005) in the SUBPDEC decadal variations. The relevance of the SUBP region might point instead to changes in the thermohaline circulation (THC) being the key element in these decadal fluctuations, in line with ideas put forward by Marshall et al. (2001a). The hampering (enhancement) of the THC in response to a warmer (colder) subpolar North Atlantic would provide a negative feedback on the SST anomalies. Eden and Greatbatch (2003) showed that this mechanism led to a damped decadal oscillation in a simulation with a realistic model of the North Atlantic Ocean coupled to a simple stochastic atmosphere model. Álvarez-García et al. (2008) found the mechanism described by Eden and Greatbatch (2003) was also operative in an ocean simulation forced with observed atmospheric fluxes. The relevance of changes in the THC in connection with the SUBPDEC fluctuations could also be hinted at by their association with the EA atmospheric pattern, as the latter has been identified as a primary driver of the surface freshwater flux in the eastern subpolar North Atlantic (Josey and Marsh 2005). Interestingly, although the horseshoe SST pattern was related to the NAO in the analysis of Frankignoul and Kestenare (2005), Losada et al. (2007) obtained an atmospheric response resembling the EA pattern when they forced their atmospheric model with a North Atlantic SST structure essentially similar to the horseshoe pattern, the anomalies in WESTM being notably weaker than those in SUBP and TNA. Pohlmann and Latif (2005) also identified an EA-like atmospheric response to SST anomalies in the Atlantic in their simulations. These results intimate that the SUBPDEC variability might involve atmosphere–ocean coupling. A small coupling was necessary in the simple model of Czaja and Marshall (2001) so as to reproduce observed spectral features. Coupling was also effective in reducing the damping of the decadal oscillation in the study of Eden and Greatbatch (2003).
One additional point demanding clarification concerns the reason why the 14-yr period WESTMDEC variability is absent before 1940. Sun and Yu (2009) have shown ENSO displays a modulation of its intensity with a period close to 14 yr (see their Fig. 2). This 14-yr modulation appears to be stronger after 1940 (their Fig. 1a), consistently with the behavior of WESTMDEC. Finally, there seems to exist a likeness between the multidecadal modulation of the SUBPDEC index (Fig. 2a) and that of ENSO (Fig. 1 in Sun and Yu 2009), with lower amplitude from the 1920s to the 1970s. This is particularly noticeable in the SSA version of SUBPDEC (not shown), which filters out shorter time scale variations in the index. This could corroborate the the link of SUBPDEC to Pacific variability, but might also be a manifestation of the dependence of both elements on multidecadal changes of the THC, hypothesized to cause the multidecadal modulation of ENSO in previous works (Dong and Sutton 2007).
The Merged HadISST-OI.v2 SST data and the CORE2 wind stress data for this study are from the Research Data Archive (RDA), which is maintained by the Computational and Information Systems Laboratory (CISL) at the National Center for Atmospheric Research (NCAR). NCAR is sponsored by the National Science Foundation (NSF). The original data are available from the RDA (http://dss.ucar.edu). The HadSLP2r data were provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, from their Web site at http://www.esrl.noaa.gov/psd/. The valuable comments of the two anonymous reviewers are gratefully acknowledged. This work was supported by the Spanish Ministry of Science and Innovation under Project CGL2006–092068.