This work presents observational evidence of a change in Atlantic-Pacific Niños connection since the late 60's. Accordingly, summer Atlantic Niños (Niñas) alter the tropical circulation favoring the development of following-winter Pacific Niñas (Niños). The same change is obtained in an ensemble of AGCM integrations in which SSTs in the Atlantic are the observed in 1949–2002 and those in the tropical Indo-Pacific are from a coupled OGCM. The mechanism (for the positive Atlantic phase) involves the strengthening of the Walker circulation with ascending branch over the Atlantic and descending branch over the central Pacific. The enhanced surface divergence in the latter region shallows the equatorial thermocline triggering coupled processes, and favoring the development of a Pacific La Niña. Results could be linked to the reported 60's and 70's climate shifts; emphasizing the importance of tropical Atlantic for the success of seasonal forecast skill.
 The tropical Pacific and Atlantic oceans host their own “El Niño” events peaking in boreal winter and summer, respectively [Philander, 1990; Merle, 1980; Zebiak, 1993]. These phenomena result from ocean-atmosphere interactions in which the Bjerknes positive feedback and equatorial wave dynamics operate [Bjerknes, 1969; Zebiak, 1993; Keenlyside and Latif, 2007]. Accordingly, positive SST anomalies in the eastern part of the basins are associated with local enhancements of convective activity, weaker east-to-west (trade) winds, deeper ocean thermocline, and further increase of the SST anomalies. Negative SST anomalies are associated with features of the opposite sign.
 Links between those events of strong interannual variability have been sought with modest success. Inter-basin tropical connections are expected, but there is no consensus in their direction and strength [Chang et al., 2006], with some authors arguing for a significant Pacific leadership [Enfield and Mayer, 1997; Latif and Grötzner, 2000]. Although Wang's  observational analysis rejects the contemporaneous connection between Atlantic and Pacific Niños, it shows that the tropical Atlantic and the tropical Pacific Oceans can be contemporaneously connected via the inter-basin SST gradient. There, the Pacific and Atlantic Walker circulations would be disturbed by induced zonal wind anomalies across northern South America, reinforcing the initial inter-basin SST gradient and constituting a positive feedback for involved tropical climate variability.
 Notably, a few recent studies have shown significant associations between summer Atlantic Niños and following-winter Pacific Niñas after the 70's [Melice and Servain, 2003; Keenlyside and Latif, 2007; Polo et al., 2008]. Some of those studies give no special attention to these associations [Melice and Servain, 2003], and no one has discussed the dynamical mechanisms at work. Likewise, no detailed analysis of such changes in the stationarity of the Atlantic-Pacific Niños connection has been made so far. A recent work by Jansen et al. , however, emphasizes the importance of this subject, since the inclusion of the Atlantic Niño seems to improve the El Niño-Southern Oscillation (ENSO) forecast skill of their conceptual model.
 In a related context, the Atlantic Niño impact on the Indian monsoon variability has strengthened since the climate shift of the 70's whereas the impact from the Pacific ENSO has waned [Kucharski et al., 2007]. Also, a remarkable change in the lead-lag relationships between Indian Ocean SSTs and the Pacific El Niño evolution has been documented to hold after the shift [Terray and Dominiak, 2005]. Accordingly, SST anomalies in the southeastern Indian Ocean during the boreal winter may have contributed to trigger either ENSO events or transitions between ENSO phases in recent decades. Other works point to a change in El Niño (La Niña) properties as part of the climate shift in the central Pacific; and the eastward displacement of the westerly (easterly) wind anomalies at the equator [Fedorov and Philander, 2000; Terray and Dominiak, 2005; Wang, 1995; Wang and An, 2002].
 The present work examines the connections between Pacific and Atlantic Niños during the 1950–2002 period with two main objectives. First, we aim to further analyze the stationarity of this connection in the second half of the 20th century. Second, we attempt to give physical support of the recent change in connections between the tropical Pacific and Atlantic climates.
 The Niño3 and ATL3 indexes are used in this work. The former is defined as the SST anomaly in the eastern tropical pacific area [5°S–5°N; 150°W–90°W]. The latter index is defined as the SST anomaly in the central-eastern tropical Atlantic: [3°S–3°N; 20°W–0°E]. The observational atmospheric fields are taken from the ERA-40 Reanalysis [Uppala et al., 2005]. Thermocline depth data come from the NCEP Pacific Ocean Analysis [Behringer et al., 1998], and SST comes from HadISST1 [Rayner et al., 2003] dataset of the UK-Metoffice (http://hadobs.metoffice.com/hadisst/). We concentrate on the monthly-mean fields.
 To achieve a better understanding of the teleconnections being examined, we use a numerical model that allows for coupled atmosphere-ocean interactions in selected regions. The atmospheric component of the model is the ICTP General Circulation Model (AGCM) [Kucharski et al., 2007, 2008], version 40. This AGCM is a balanced complexity model based on a spectral primitive-equation core and a set of simplified parameterization schemes. The horizontal resolution is T30 and it has 8 levels in the vertical. The ocean model is an extended 1.5 layer reduced-gravity model [Zebiak and Cane, 1987; Chang, 1994; Chang et al., 2006] with a resolution of 2° in longitude by 1° in latitude. A flux-correction is applied to climatological surface heat-fluxes and wind stresses to reduce the model's drift [Chang et al., 2006]. The heat-flux and wind stress calculation is F = F_model + (F_obsclim − F_modelclim). The so-called flux-correction term is (F_obsclim − F_modelclim) in which the term F_obsclim has been derived from the ERA40 re-analysis for the period 1961 to 2000, the same period for which F_modelclim has been derived. The flux-correction is a monthly climatology, which means it varies from month to month throughout the year, but has no interannual variations. Control simulations in which the tropical Indo-Pacific basin is fully coupled and the SST's in all other regions are climatological, show the ENSO variability approximately in the right position and with a correct magnitude (not shown). In the configuration selected for this study, the AGCM and ocean model are run fully coupled in the tropical Indo-Pacific basin (between 30°S and 30°N). Elsewhere, climatological SSTs are prescribed except for the Atlantic sector, where observed, monthly varying SSTs are prescribed. Different members differ in the initial conditions. The results shown correspond to an ensemble of nine runs for the period 1949–2002. We show regression maps for the statistically significant areas at the 95% confidence level according to a Student's t-test for the effective number of degrees of freedom [Metz, 1991].
 We start by examining the inter-basin connections. Figure 1a presents the 20-year running mean of lead-lag correlations between Atl3 and Niño3 in the period 1950–2002. Figure 1a shows significant negative effective correlations at 6–8 month lag with Atl3 leading from the late 60's. There are also significant positive correlations at 12–18 month lags with Niño3 leading; which are due to the autocorrelation of this index [Wang and An, 2002; Joseph and Nigam, 2006] and confirm that the Atl3 signal is inherent in the Niño3 time series. To better illustrate the features highlighted in Figure 1a, we show in Figure 1b the Atl3 index in boreal summer (bars) and the Niño3 index in the next winter (continuous line) during 1950–2002. All Pacific ENSO events after the 1960's are preceded by Atlantic events with the opposite sign. In addition, regression maps of Indo-Pacific SST anomalies (Figure 1c) onto ATL3 reveal no apparent relationship with Atlantic Niños in the previous summer (June–September) before the 70's, while they show a pattern with opposite phases in the Atlantic and Pacific after that date (see also Animations S1–S4 of the auxiliary material). The presence of significant correlations at negative lags (Pacific leading) and at lag 0 in Figure 1a suggests the possible influence of the Pacific on the Atlantic (as has also been suggested by Chang et al. ). It is worth to notice that the Atlantic Niño is in its mature phase [Haarsma and Hazeleger, 2007], whereas the Pacific La Niña is in its onset/development phase [Wang, 2002]. An alternative explanation comes from the independence of both phenomena, as previously reported, meaning that those SST correlations are fortuitous. Nevertheless, it can also be seen how, for positive lags, these negative correlations increases (from 0.4 for negative lags to more than 0.6 for positive lags), suggesting that the Atlantic could have an active role in the enhancement of the Pacific ENSO phenomenon.
 Next, we search for the mechanism involved in connection change. Figure 1d shows atmospheric and oceanic fields projected onto the summer Atl3 index, for the period 1979–2001 (we choose this period to analyse the signal after the 60's, for homogeneity of data assimilation and following other works as Trenberth et al. , Polo et al. , and García-Serrano et al. ). The equatorial Atlantic warming has the typical Gill-type atmospheric response [Gill, 1980] to tropical heating anomalies, including two anticyclones straddling the equator at upper levels accompanied by surface wind convergence and divergence aloft. Thus, the global streamfunction and surface wind response show the picture of an anomalous Walker circulation, with rising air and heavy rainfall in the eastern equatorial Atlantic, and sinking air and drier conditions in the central equatorial Pacific, being the latter consistent with a Gill-type response to the zonally-compensated heating in the Pacific. The presence of SST anomalies in the Pacific for negative lags (not shown) and at lag 0, together with the increase of the above mentioned surface divergence, would help to shallow the Pacific thermocline. The thermocline shallowing propagates eastward at a speed consistent with an upwelling Kelvin wave (thermocline perturbations require approximately 3–4 months for travelling from the central to the eastern Pacific). Associated with the thermocline shallowing, surface SST anomalies and the related surface wind divergence propagates also to the east. During the whole process, the Bjerknes feedback operates enhancing the SST anomalies and the east–west SST gradient. By the following winter, full La Niña conditions prevail in the Pacific with anomalous east-to-west surface winds and increased thermocline slope across the equatorial basin.
 On the basis of the observational evidence presented in the previous paragraphs, we posit that a strengthened ascending branch of the Walker circulation over the equatorial Atlantic during the peak phase of recent Atlantic Niños (Niñas) induces anomalous surface divergence over the equatorial Pacific, enhancing the development of La Niña (El Niño) in the next winter. This scenario would represent a direct impact of the atmospheric response to Atlantic Niño on ENSO variability, which could be triggering or strengthening the inter-basin positive feedback proposed by Wang . Our hypothesis is presented in the context of the tropical Atlantic warming [Hansen et al., 2006] since interannual SST anomalies in this ocean are superimposed on a warmer background state. Indeed, a simple test of equal population for the tropical Atlantic SST before and after the 70's is rejected (not shown), evidencing this change in the background state of the tropical Atlantic between both periods.
 We can demonstrate the soundness of our hypothesis by using our partially-coupled model. In the selected configuration (see methods), both the realistic evolutions of boundary conditions over the Atlantic and the coupled atmosphere-ocean dynamics in the Pacific are fully represented. According to Figure 2a and in agreement with the observation (Figure 1a), simulated Pacific La Niña (El Niño) events appear in association with Atlantic Niños (Niñas) in the previous summer from the 1970's, but not earlier. An analysis of the simulated Niño3 index in Figure 2b (continuous line) reveals a significant anticorrelation with the Atl3 index since late 60's. The pattern of simulated SST anomalies regressed onto the Atl3 index resembles the observed during a La Niña event for the last decades on the 20th century (Figure 2c). Hence, the simulations reproduce the change in the inter-basin SST relationship from the 1970s by just changing the prescribed SSTs over the Atlantic. The model also reproduces how the anomalous heating over the tropical Atlantic is able to alter the Walker circulations and to drive the anomalous upper-levels convergence and descent over the central equatorial Pacific (Figure 2d, June–September), although with weaker amplitude. The ocean model reproduces the thermocline shallowing in the central Pacific and its eastward propagation.
 Weaker responses are expected in the model, as the experiment was designed in order to check whether or not the Atlantic Niño could determine the phase of the Pacific ENSO events; and it does not intend to reproduce the full amplitudes. This lack in the reliability of the SST amplitude in the model could also be due to the absence of the observed cold (warm) anomalies in the central Pacific at negative lags and at lag 0 (for the Atlantic Niño (Niña), which would highlight the potential role of the inter-basin SST gradient variability suggested by Wang .
 Together with the above-mentioned atmospheric bridge mechanism for ENSO development/enhancement, the surface wind forcing in the southeast Pacific proposed by Toniazzo  could be also related to the ATL3 response in the Pacific region (Figure 1d). This is consistent with model simulations (see Figure 2d, right) and will be analyzed in a future paper.
 The Atlantic-Pacific connection presented here was established in the late 60's (1968), i.e., almost one decade before the climate shift in the 70's. Baines and Folland  mention this change in timing in their discussion of the rapid global climate change that took place at the end of the 60's.
 The strong relationship between Atlantic and Pacific Niños since the late 60's is an important component in the understanding of the climate shift and in the success of the seasonal forecast. The former has been attributed partly to the global warming trend [Fedorov and Philander, 2000; Wang and An, 2002] and partly to natural climate variability [Cane et al., 1997]. Our findings on the inter-basin connection, and their robustness, highlight the importance of proper consideration of the tropical Atlantic Ocean for the success of seasonal-to-decadal climate predictions.
 The study was supported by the Global Change and Ecosystems Programme (EU integrated project: African Monsoon Multidisciplinary Analysis (AMMA)). AMMA has been the beneficiary of a major financial contribution from the European Community's 6th Framework Research Programme. Detailed information on scientific coordination and funding is available on the AMMA International web site (http://www.amma-international.org). Also, the study has been supported through the Spanish MEC project CGL2006-04471. The model component of this article is as contribution to the ENSEMBLES project funded by the European Commission's 6th Framework Research Programme, contract GOCE-CT-2003-505539. The work at UCLA was supported by NOAA grant NA07OAR4310236.