1 Introduction

Pacific decadal variability or so-called climate shifts are accompanied by changes in regional and also global temperatures and rainfall that have substantial socio-economic impacts. The mid 1970s climate shift is an example of such a relatively abrupt change that has been analyzed extensively and is related to the Pacific Decadal Oscillation or Interdecadal Pacific Oscillation (e.g. Zhang et al. 1997; Trenberth and Hurrell 1994; Miller et al. 1994; Meehl et al. 2009; Parker et al. 2007; Power et al. 1999). Meehl et al. (2009) showed, using observational data and numerical modelling that the 1970s climate shift may be understood as a combination of external forcing (greenhouse gases) and inherent decadal fluctuation of the Pacific climate system.

In the more recent period a climate shift that occurred at the end of the 20th century of perhaps equal importance has been observed with influences on the Walker circulation (Dong and Lu 2013), monsoons (Xiang and Wang 2012), ENSO variability (Xiang et al. 2012), and may have been responsible for the hiatus in global warming (England et al. 2014). Both of the climate shifts described above have been accompanied by changes in equatorial Pacific sea surface temperatures (SSTs). The Pacific changes are not isolated phenomena, and changes in the Indian Ocean and Atlantic Ocean region have also been documented (Chikamoto et al. 2012).

Recently, the influence of Atlantic El Nino-like warming events on the Pacific ENSO have been analyzed, and it has been found that a tropical Atlantic warming may contribute to an eastern Pacific cold event a few months later (Rodriguez-Fonseca et al. 2009; Ding et al. 2012; Frauen and Dommenget 2012; Jansen et al. 2009; Polo et al. 2014; Martin-Rey et al. 2012; Keenlyside et al. 2013). On multi-decadal to centennial time scales Kucharski et al. (2011) showed that the 20th century Atlantic Ocean warming caused an increase of the zonal SST gradient in the tropical Pacific. This is consistent with other recent studies which show that once ENSO variations are removed, long-term SST trends indicate that the eastern Pacific warms less than the western Pacific (Compo and Sardeshmukh 2010; Solomon and Newman 2012; L’Heureux et al. 2013). The physical mechanism proposed in these works for the Atlantic influence on the tropical Pacific is a modification of the Walker circulation that leads to low-level wind anomalies in the central-western Pacific, which, in turn, force oceanic Kelvin waves that propagate into the eastern Pacific. The response is then amplified through the Bjerknes feedback.

Zhang et al. (2010) showed that the 20th century eastern Pacific SST cooling trend is a robust feature among different observational datasets and that it can also be identified as the second Empirical Orthogonal Function of the Pacific SSTs. Yang et al. (2014) identified such a cooling trend also in the ocean subsurface by analysing reanalysis products. The Atlantic influence on Pacific interannual variability and long-term changes have also been identified in a subset of models of the World Climate Research Program’s Coupled Model Intercomparision Project Phase 5 dataset (Kucharski et al. 2014).

Influences of the Atlantic Multidecadal Oscillation-type variability on the Pacific and ENSO have also been documented (Zhang and Delworth 2007; Dong and Sutton 2007; Lu et al. 2008; Timmermann et al. 2007).

Regarding the end-of-century climate shift, a recent work by Chikamoto et al. (2012) shows that the shift toward equatorial cold conditions is reproduced in hindcast experiments only if also the Atlantic Ocean is initialized, pointing to a potential role of the abovementioned atmospheric bridge mechanism. Also McGregor et al. (2014) support the importance of the Atlantic warming in amplifying the recent eastern Pacific cooling.

Kang et al. (2014) suggested that the recent climate shift may be partly related to a shift of the AMO from its negative to its positive phase and showed that the observed reduction in ENSO amplitude after the shift was due to cooling in the central-eastern Pacific.

On the other hand, Farneti et al. (2014a, b) demonstrated the importance of internal Pacific dynamics in the decadal variations of tropical Pacific SSTs, and emphasized the role of extratropical wind variations and their influence on sub-tropical cell variability.

The purpose of the present study is to isolate the Atlantic role in the equatorial Pacific decadal variability for the whole 20th century and to examine the robustness of the above findings for different Pacific climate shift events in the past. For this goal, an ensemble of partially coupled ocean-atmosphere simulations is performed, where an atmospheric general circulation model (AGCM) is coupled to an ocean general circulation model in the Indo-Pacific region and forced with observed SSTs in the Atlantic region.

The paper is organised as follows: Sect. 2 introduces the data, model and experimental set-up. A brief analysis of the model performance is presented in Sect. 3. In Sect. 4 results are presented and the summary and conclusions given in Sect. 5.

2 Data, model and experimental set-up

The Hadley Centre HadISST (1870 to present; Rayner et al. 2003) is used for the observational SST dataset. The NOAA-CIRES 20th Century Reanalysis version 2 (Compo et al. 2011; 1900 to present) is used for surface winds, and the CMAP (Xi and Arkin 1997) dataset is used for rainfall verification of the model. For ocean mean state verification we use data from the ORA-S3 ECMWF ocean reanalysis (Balmaseda et al. 2008, 1960 to present).

The model used is the International Centre for Theoretical Physics AGCM (ICTPAGCM, version 41), formerly called “SPEEDY”, coupled to the Nucleus for European Modelling of the Ocean (NEMO) model (Madec 2008) using the OASIS3 coupler (Valcke 2006). NEMO also includes a dynamic sea-ice model component (LIM; Fichefet et al. 1997) which is activated in the simulations for the current study. This model is referred to as ICTPCGCM in the following. The ICTPAGCM has eight vertical levels, and the horizontal spectral truncation is T30. The model includes physically based parametrizations of large-scale condensation, shallow and deep convection, short-wave and long-wave radiation, surface fluxes of momentum, heat and moisture, and vertical diffusion. The model has been originally described in Molteni (2003) in the 5-layer version and in Kucharski et al. (2013) in the current 8-layer version. The ocean model is based on NEMO v.3.0 (Madec 2008) which is a primitive equation z level model making use of the hydrostatic and Boussinesq approximations. The version used has a tripolar ORCA2 configuration with horizontal resolution of 2° and a tropical refinement to 1/2°. The model has 31 vertical levels.

The ICTPAGCM is coupled to NEMO in the Indo-Pacific region, whereas in the Atlantic region, the AGCM is forced by observed SSTs from the HadISST dataset, apart from the regions with sea-ice where surface temperatures are used from the ice model (Atlantic pacemaker). A flux-correction is applied to avoid drifts in SSTs that may deteriorate interannual and longer term variability, particularly in tropical regions (Kroeger and Kucharski 2011). The flux-correction approach employs a one-way anomaly coupling from the ocean to the atmosphere. The heat fluxes used by the ocean component are based on full SST, while heat fluxes that force the atmosphere are based on anomalies of SST. The latter requires knowledge of the models climatological SST field, which is determind by a previous 60-year long run, in which the ocean model is forced with fluxes from the atmospheric component. The model integrations start in 1872 and run through 2013. The atmospheric CO2 absorptivity is kept constant at a values representative for the middle of the 20th century.

The ICTPCGCM is much faster than most CGCMs and its computational speed is practically that of the ocean component NEMO, which is because of the very fast ICTPAGCM. An ensemble of ten members is generated by restarting the ICTPCGCM from a previous 300-year long run and using small initial perturbations to the atmospheric state. The first 29 years of the integrations are considered as further spin-up and the analysis is therefore restricted to the period 1900–2013. This experiment is referred to as ATL_VAR in the following. Unless otherwise stated, results are presented from the ensemble mean of these simulations, while individual ensemble members are used to evaluate the statistical significance of the results. A second ensemble of ten members is performed (ATL_CLIM), similar to the first one, but with SSTs in the Atlantic region set to monthly varying climatological values (average monthly varying SSTs for the period 1979–2008) in order to better identify the role of the Atlantic SST anomalies in modifying the Pacific variability. Because of the Atlantic pacemaker strategy and since the CO2 concentration is fixed in the simulations the focus of this study is natural variability induced by the Atlantic. Therefore all observational datasets and outputs of model simulations are detrended before the analysis of decadal variability.

3 Model performance

In this section we briefly present some basic evaluations of the model performance in the coupled Indo-Pacific region. Figure 1 shows the observed (Fig. 1a) and modeled (ATL_VAR; Fig. 1b) annual mean precipitation and 925 hpa wind climatologies for the period 1981–2010. Overall, the model reproduces the main features of the annual mean precipitation and low-level wind climatologies reasonably well. Some biases that are typical in GCMs are also present, for example the tilt of the South Pacific convergence zone is too weak in the model.

Fig. 1
figure 1

Annual mean precipitation and 925 hPa wind climatologies. a Observations, b ATL_VAR. Units are mm/day for precipitation and m/s for winds

Full size image
Fig. 2
figure 2

Temperature (shading) and zonal current (contours) in zonal-vertical section at the equator. a ECMWF ocean reanalysis, b ATL_VAR. Units are °C for temperature and cm/s for zonal current

Full size image

Figure 2 shows the thermal structure of the equatorial upper Pacific Ocean (shaded) and also the zonal velocity (contours) for observations (Fig. 2a) and the model (ATL_VAR; Fig. 2b). The thermocline (approximately 20 °C isoline) is quite realistically reproduced. However, the temperature in the upper 50 m is lower in the model, particularly in the central Pacific. The zonal equatorial jet at the thermocline is located in approximately in the right position, but is slightly underestimated, and reflects the slightly reduced upper ocean temperature gradients. On the other hand, the near-surface currents are overestimated.

The annual rainfall, near-surface wind and equatorial ocean temperature and zonal velocity climatologies in the ATL_CLIM ensemble are very close to the ones from the ATL_VAR ensemble and therefore not shown.

Fig. 3
figure 3

Standard deviation of SSTs. a Observed and b ATL_VAR. c Difference in standard deviation between ATL_VAR and ATL_CLIM. In c 95 % statistically significant differences are shown with contours. Units are K

Full size image
Fig. 4
figure 4

Central equatorial zonal near surface wind anomaly index (CPWI; averaged over area 160°E−190°E, 5°S–5°N). The anomaly time series have been filtered by a 10-year running mean. Shown are observations (black line), ATL_VAR ensemble mean (red line). The ATL_CLIM ensemble mean (dashed blue line). The AMO index is also plotted in the figure, multiplied by 4 as green dashed line. Units are m/s for all wind indexes, and K for the AMO index

Full size image

The standard deviation of SSTs for the period 1950–2010 is shown in Fig. 3a for HadISST and Fig. 3b for the model (ATL_VAR; the standard deviation is estimated by the variances of individual ensemble members). The distribution of the standard deviation (after removing the annual cycle) in the equatorial Pacific region is quite well reproduced with amplitudes close to observations (the standard deviations of the Nino3.4 index are about 0.83 K for the model compared to 0.86 K for the observations). The equatorial eastern Pacific standard deviation maximum is extending too far into the central Pacific, a feature common also in many state-of-the-art CGCMs (Guilyardi et al. 2009) . The standard deviation is overestimated in parts of the North Pacific. Also, in the eastern Indian Ocean the standard deviation is too large. The Nino3.4 index has a dominant period of about 6 years, and is thus on the low-frequency end of the observed variability (2–7 years). Figure 3c shows the difference in standard deviations between the experiments ATL_VAR and ATL_CLIM which clearly shows increased equatorial Pacific variability (note that here the periods 1979– 2008 are chosen because of identical mean SSTs in the two simulations in the Atlantic region; 95 % statistically significant differences according to an F-test are shown with contours). Such a result has also been reported by Sasaki et al. (2014) and indicates the Atlantic influence on equatorial Pacific interannual variability that has been analyzed in details in the recent literature (e.g. Rodriguez-Fonseca et al. 2009; Ding et al. 2012; Frauen and Dommenget 2012; Jansen et al. 2009; Polo et al. 2014; Martin-Rey et al. 2012; Keenlyside et al. 2013). However, it is not the focus of this paper to analyze this feature in further details.

Overall, the model performace seems realistic enough to use the model to investigate changes in the Pacific region that have occurred in the the 20th century.

4 Results

4.1 Identification of the Atlantic influence

There are obviously many factors that may be responsible for Pacific decadal and multi-decadal changes. Zonal surface winds in the equatorial central Pacific have been identified as a crucial factor and may be interpreted as an indicator of variations in the Walker circulation (Kucharski et al. 2011; England et al. 2014; McGregor et al. 2014; Yang et al. 2014). These zonal surface wind changes may be potentially induced by Atlantic SSTs (Kucharski et al. 2011; McGregor et al. 2014; Chikamoto et al. 2012). Central Pacific zonal near-surface wind anomalies are part of the Bjerknes feedback and may therefore be the cause of, but also the response to eastern Pacific SST anomalies. We therefore consider the decadal variations of central Pacific low-level wind anomalies (at the 925 hPa level for observations and model; averaged over the region 160°E–190°E, 5°S–5°N after applying a 10-year running mean; this index is referred to as Central Pacific Wind Index (CPWI) in the following) in Fig. 4. The definition of the CPWI is motivated by the results from Kucharski et al. (2011; Fig. 3), who showed that in this region there is the strongest Atlantic impact on equatorial Pacific surface winds. Results presented are robust regarding modest changes of the averaging region, for example using the Nino4 region (160°E–210°E, 5°S–5°N) for the CPWI definition gives very similar results.

Fig. 5
figure 5

Correlation of central equatorial Pacific zonal near surface wind anomaly index (CPWI) with Atlantic SSTs. a Observations, b ATL_VAR ensemble mean

Full size image
Fig. 6
figure 6

1931–1950 minus 1910–1929 SST and near surface wind differences. a Observed, b ATL_VAR ensemble mean. In b anomalies that are 95 % statistically significant are indicated by the contours. Units are K for SSTs and m/s for wind

Full size image

There are clearly prominent decadal variations of the observed CPWI (black curve in Fig. 4). For example the 1970’s climate shift is clearly visible in the CPWI as well as the following downward trend in the 1990s. The wind index shows also low values in the 1930s and 1940s. The model ensemble mean (ATL_VAR; red line) shows overall similar variations at the multidecadal timescale to the observed one (particularly if the first 20 years are discarded), but with smaller amplitude. Indeed, the correlation between the observed and ensemble mean ATL_VAR CPWI is 0.33 for the whole period, but increases to 0.64 if the period after 1920 is taken into account. In Fig. 4, also the CPWI from the ensemble mean of integrations with climatological Atlantic SSTs (ATL_CLIM) is shown as dashed blue line. As can be seen the variations at multidecadal timescale (which can be considered as residual internal Pacific variability) are much smaller than that of ATL_VAR. The standard deviations of the observed, ATL_VAR, and ATL_CLIM ensemble mean CPWI are, respectively, 0.41, 0.21, 0.08 m/s. Individual members of ensembles ATL_VAR and ATL_CLIM (not shown) have average CPWI standard deviations of 0.37 and 0.27 m/s, which are significantly different according to a two-sample F test. This indicates that the Atlantic forcing provides a small, but not negligible contribution to the decadal CWPI variability (i.e. about 30 % forced variability can be derived from the ATL_VAR ensemble).

Fig. 7
figure 7

1970–1989 minus 1931–1950 SST and near surface wind differences. a Observed, b ATL_VAR ensemble mean. In b anomalies that are 95 % statistically significant are indicated by the contours. Units are K for SSTs and m/s for wind

Full size image
Fig. 8
figure 8

1994–2013 minus 1970–1989 SST and near surface wind differences. a Observed, b ATL_VAR ensemble mean. In b anomalies that are 95 % statistically significant are indicated by the contours. Units are K for SSTs and m/s for wind

Full size image

Figure 5 shows the correlations of the observed (Fig. 5a) and ensemble mean of ATL_VAR (Fig. 5b) CPWI with the Atlantic SSTs to identify the likely forcing of the central equatorial zonal wind changes. Clearly, in both correlation maps there is a diplole pattern with negative values in the northern Atlantic and positive values in the southern Atlantic with the negative correlations in the North Atlantic dominating. Such a pattern clearly resembles the AMO pattern (e.g. Parker et al. 2007). Indeed the observed AMO time-series, defined as in Kang et al. (2014) as 10-year running mean of the averaged detrended SSTs in the North Atlantic region (70°W–0°E, 0°–60°N), is plotted in Fig. 4 (green dashed line, multiplied by four for better comparison), and clearly shows anticorrelation with the observed and ensemble mean ATL_VAR CPWI (correlation coefficients are −0.56 and −0.81, respectively). For individual members of ATL_VAR, the CPWI correlations with the AMO index are all negative (between −0.05 and −0.71) with an average of −0.46, which is broadly consistent with the observed AMO-CPWI correlation.

4.2 Individual events of Atlantic-induced Pacific decadal variability

Observed climate shift events may be identified by the CPWI introduced in Sect. 4.1 (e.g. we have seen that the 1970s and the 1990s shifts are reproduced by such an index). It is also clear from Sect. 4.1 that there is a likely AMO forcing of central Pacific wind variability in ATL_VAR and observations. Assuming that the AMO is partly forcing both observed and, to a larger extend, ATL_VAR central equatorial Pacific winds, it is useful to assess changes in the surface temperature and winds in different phases of the AMO, which, in parts also correspond to the well-known climate shift events of the 1970s and 1990s. Therefore, we select three events represented by 20-years periods that are characterized by large changes of the AMO before and afterwards, namely (1) 1931–1950 minus 1910–1929, (2) 1970–1989 minus 1931–1950 and (3) 1994–2013 minus 1970–1989 (note that AMO has been overall positive in the periods 1994 to 2013 and 1931 to 1950, and negative in 1970 to 1989 and 1910 to 1929). Event (2) corresponds clearly also a large increase in the zonal winds, and thus to the 1970’s climate shift, whereas event (3) corresponds roughly to the latest hiatus is global warming that was presumably induced by strengthening of equatorial Pacific winds (e.g., England et al. 2014).

Fig. 9
figure 9

200 hPa velocity potential change for ATL_VAR ensemble mean. a 1931–1950 minus 1910–1929, b 1970–1989 minus 1931–1950, c 1994–2013 minus 1970–1989. Only changes that are significant at the 95 % significance level are shown. Units are 106 m2/s. Vectors show the near-surface winds changes as in Figs. 6, 7 and 8

Full size image
Fig. 10
figure 10

Temperature (shading) and zonal current (contours) in changes in the zonal-vertical section near the equator (averaged from 5°S to 5°N) for ATL_VAR ensemble mean. a 1931–1950 minus 1910–1929, b 1970–1989 minus 1931–1950, c 1994–2013 minus 1970–1989. Only changes that are significant at the 95 % significance level are shown. Units are K for temperature and cm/s for zonal current

Full size image

Figure 6 shows the SST and surface wind changes for the early twentieth century event (1). The observations (Fig. 6a) clearly show an eastern Pacific cooling event with easterly wind anomalies in the central western Pacific, the Atlantic SSTs show a positive change everywhere. The ensemble mean of ATL_VAR (Fig. 6b) is reproducing this event with larger amplitude, and negative SST anomalies shifted towards the central Pacific. This feature may be due to the bias already identified in Fig. 3, where it was shown that ENSO-related variability is extending too far west in the model. The easterly surface wind anomalies are also simulated stronger than observed.

The well documented mid 70s climate shift (2) is analyzed in Fig. 7. Clearly, this event is characterized by a change towards more positive SSTs in the eastern Pacific region, and by westerly wind anomalies in the central-eastern Pacific (Fig. 7a), but also by a substantial cooling in the extratropical north and south Pacific. ATL_VAR reproduces also this event relatively well with increased amplitude in the equatorial regions (Fig. 7b). By definition, SST anomalies in the North Atlantic are negative during event (2).

Finally, event (3) is characterized by easterly wind anomalies in the observations, however, the SST changes in the eastern Pacific are relatively small, whereas a warming is found in the western Pacific and Indian Ocean and extratropical Pacific (Fig. 8a). Also, the Atlantic shows considerable warming, particularly north of the Equator. The model (Fig. 8b) reproduces some features of this event, such as the warming in the Indian Ocean and in the extratropical Pacific. The model also shows a slight cooling in the central Pacific and easterly wind anomalies there.

Fig. 11
figure 11

Regression of a observed SSTs and near surface winds, b ATL_VAR ensemble mean SSTs and near surface winds, c ATL_VAR ensemble mean 200 hPa velocity potential and near surface winds onto the AMO index for the period 1910–2008. In b and c only values that are significant at the 95 % significance level are shown. Units are K for SSTs, m/s for wind and 106 m2/s for velocity potential

Full size image
Fig. 12
figure 12

Regression of temperature (shading) and zonal current (contours) near the equator (averaged from 5°S to 5°N) onto the AMO index for the period 1910–2008. Only values that are significant at the 95 % significance level are shown. Units are K for temperature and cm/s for zonal current

Full size image

As far as the model results from ATL_VAR are concerned, the only statistically significant near surface wind responses in the Pacific region are seen in the central tropical Pacific, supporting the modification of the Walker circulation as likely physical mechanism which has also been made responsible for the tropical Atlantic impact on the tropical Pacific variability on interannual timescale and for the long-term trend (e.g. Rodriguez-Fonseca et al. 2009; Kucharski et al. 2011). Even though the AMO pattern in the North Atlantic that are clearly seen in Figs. 6, 7 and 8 show a north-south dipole pattern, there is an overall warming of the tropical Atlantic if the AMO is in the positive phase and a cooling if the AMO is in its negative phase. Indeed, if we define a tropical Atlantic index as 10-year running mean SST averaged in the region 60°W–10°E, 20°S–20°N, the correlation of such an index with the AMO is 0.81. The warm (cold) tropical Atlantic during the positive (negative) phase of the AMO modifies the Walker circulation, leading to rising (sinking) and upper-level divergence (convergence) motion in the Atlantic region and sinking (rising) motion and upper-level convergence (divergence) in the central Pacific region. The sinking (rising) motion in the central Pacific induces easterly (westerly) near-surface anomalies near the western edge of it. Fig. 9 shows the 95 % statistically significant upper-level velocity potential responses for the three selected events for ATL_VAR, indicating clearly upper-level divergence (velocity potential minimum) in the tropical Atlantic region and upper-level convergence (velocity potential maximum) in the Pacific region in events (1) and (3), and the opposite in event (2). Also, the near-surface wind response is shown in Fig. 9, and is consistent with the velocity potential responses and the dynamics outlined above.

Figure 10 shows a longitude-depth section of the equatorial Pacific (averaged from 5°S to 5°N ) for the change in temperature and zonal velocity in experiment ATL_VAR. Note that in Fig. 10, only changes that are different from zero at the 95 % significance level are shown. All three events are characterized by consistent subsurface signatures that indicate increases of the east-west temperature gradient and equatorial undercurrent in the events (1) and (3) and opposite change in event (2). The changes in event (3) are weaker than in the other two events, but still statistically significant.

From Fig. 4 it can be seen that the first positive AMO period lasts from about 1930 to 1960, and that the subsequent negative AMO period lasts from about 1960 to 1994. It is therefore possible to define multiple 20-year periods within these two AMO phases. We have verified that the results are robust against shifts of periods within these AMO phases.

It should be noted that if in ATL_VAR the anomalies of 200 hPa velocity potential and low-level winds in individual periods with respect to the long-term climatology are considered then from the differences (a) 1910–1929 minus 1910–2013, (b) 1931–1950 minus 1910 minus 2013, (c) 1970–1989 minus 1910–2013, (d) 1994–2013 minus 1910–2013, only (b) and (c) give substantial and statistically significant contributions (not shown). This is because the AMO amplitudes in these two periods are much larger compared to the other two periods. This also suggests that, for example, the difference (3) 1994–2013 minus 1970–1989 is dominated by the anomaly induced in the earlier period. Thus the weaker response for this period shown in Figs. 8, 9c and 10c may be interpreted as relaxation from the strongly perturbed conditions (negative AMO and warm eastern Pacific) in the period 1970–1989 towards normal conditions. Following similar arguments, event (1) may be interpreted as a change from near normal to strongly positive AMO and cold eastern Pacific conditions.

In order to further increase confidence in the presented results we also perform a linear regression analysis of observed SSTs and low-level winds (Fig. 11a), ATL_VAR ensemble mean SSTs, low-level winds (Fig. 11b) and 200 hPa velocity (Fig. 11c) as well as ATL_VAR ensemble mean equatorial Pacific ocean temperature and zonal currents (Fig. 12) onto the AMO index. These regressions are based on the covariance between the normalized AMO index and the respective fields for the period 1910–2008. The results from the regression analysis are consistent with those of Figs. 6, 7, 8, 9 to 10, giving confidence in the robustness of the AMO impact on the equatorial Pacific.

5 Summary and conclusions

We have analyzed the simulation of decadal tropical Pacific variability in an Atlantic pacemaker CGCM experiment ensemble. Relating the decadal variability of central Pacific low-level winds to Atlantic SSTs we have found that the AMO plays a major role in modulating central equatorial Pacific winds, with largest impacts from the tropical North Atlantic. According to AMO phases we have identified three periods with strong Atlantic forcing of equatorial Pacific changes, namely (1) 1931–1950 minus 1910–1929, (2) 1970–1989 minus 1931–1950 and (3) 1994–2013 minus 1970–1989. Both observations and the model show easterly surface wind anomalies in the central Pacific, cooling in the central-eastern Pacific and warming in the western Pacific/Indian Ocean region in events (1) and (3) [although weaker in (3)] and the oppostite signals in event (2). The model responses are overall shifted to the west compared to the observed responses, which could be related to a bias also present in ENSO variability.

Nearly all CGCMs suffer from large biases in the Atlantic region (e.g. Richter et al. 2014), and their simulation of climate variability can be compromised because of this. Therefore, by removing these biases, our pacemaker simulations enable us to explore the effects of the observed Atlantic decadal variability on the other ocean basins.

The above results suggest that overall the east-west Pacific SST gradient is strengthened during the positive and weakened during the negative phase of the AMO. This finding is consistent with Kang et al. (2014). The physical mechanism is related to a modification of the Walker circulation because a positive (negative) AMO leads to an overall warmer (cooler) tropical Atlantic. The warmer (cooler) tropical Atlantic, consistent with the trans-basin variability described by McGregor et al. (2014), modifies the Walker circulation, leading to rising (sinking) and upper-level divergence (convergence) motion in the Atlantic Region and sinking (rising) motion and upper-level convergence (divergence) in the central Pacific region. We also find an increased (decreased) equatorial east-west gradient of sub-surface temperature in the equatorial Pacific during the positive (negative) AMO phase in the model.

Of course, Atlantic decadal variability is not the only driver of Pacific decadal variability. AMO forcing explains a modest correlation of 0.33 (0.64 if data from 1920 is considered) between the observed and modeled CPWI. On the other hand, it is important to take the Atlantic-induced variability into account when analysing Pacific decadal variability. The presented results are particularly relevant in light of a recent study by Chikamoto et al. (2015), who showed that realistic ocean state initialization lead to trans-basin climate variability and to predictability up to 3 years ahead.

Although the model used here is able to reproduce the main features of the observed rainfall and low-level wind climatologies (e.g. Fig. 1), it also presents some biases that may be relevant in particular regarding the mechanism of an AMO influence on the Pacific through a modified Walker circulation. For example, the model overestimates rainfall over South America. On the other hand, all CGCMs suffer from biases of different magnitudes, signs, and locations. A coordinated multi-model ensemble pacemaker experiment could be a way to assess with more confidence the Atlantic influence on Pacific decadal variability, and it could also provide information regarding the role of model biases in this teleconnection.