We use an extension of the classical 1.5-layer RGO model, introduced by Cane (1979) to study the ENSO phenomenon. The extension of the model, as in Chang (1994), includes thermodynamics of the upper ocean and allows for the prediction of the SST.

The model consists of a 50 m depth upper layer in which mass, heat and momentum obey the conservation laws and a lower layer of infinite depth in which the velocity must be null so that the kinetic energy is finite. The approximation is reasonable for the tropical ocean because of the existence of a sharp thermocline which inhibits the downward propagation of waves generated in the upper ocean (Zebiak, 1985). To better predict changes of the SST, a linear and homogeneous frictional layer (assumed to concentrate most of the induced Ekman transport) is added to the model. The subsurface temperature is parameterized in terms of the thermocline depth, the observed annual mean temperature at 50 m depth (from Levitus, 1982) and the annual mean thermocline depth when the model is forced by observed wind stress.

The resolution of the RGO model is 1∘ in latitude and 2∘ in longitude, applied in the 30∘ S–30∘ N tropical band. The model is run using an anomaly coupling strategy. In this strategy, the oceanic and atmospheric components of the model exchange momentum and heat flux anomalies computed relative to their own model annual mean. The modelled anomalies are then superimposed on the observed annual mean. Sponge layers of 5∘ width are introduced at the northern and southern boundaries to eliminate artificial coastal Kelvin waves.

A 70-year control simulation in which the AGCM is coupled to the RGO in the tropics and to the slab ocean model elsewhere is produced. The last 50 years of the control run are used for averaging and comparison with observational analogues. We use the NOAA Extended Reconstructed SST V3b (Smith et al., 2008) and the near-surface winds from the NCEP/NCAR Reanalysis (Kalnay et al., 2006), for the period 1979–2013.

With the imposed heat flux correction, the annual mean SST in the control simulation strongly resembles the observed pattern (not shown). In addition, the model reasonably captures the main characteristics of the seasonal cycle in the equatorial (2∘ S–2∘ N) Pacific and Atlantic oceans (Fig. 1), although in the equatorial Pacific the de-meaned (annual mean removed) simulated SST seasonal cycle is weaker than in the observations. Also, the Pacific cold tongue is not well developed during SH summer.

The control simulation also reproduces the main mode of variability in the tropical Pacific Ocean quite realistically both in the spatial and temporal domains (Fig. 2; please note that in all the latitude–longitude maps in the manuscript the land and sea mask used by the model is the one depicted). The first coupled pattern arising from a singular value decomposition (SVD) of the monthly SST and surface wind characterizes ENSO and explains 81 % (62 %) of the variability in the observations (simulation). The simulated pattern is weaker than the observed, with the SST anomaly maximum located too far eastward. The phase-locking to the seasonal cycle of the simulated pattern peaks during the end of the calendar year as it does in the observations, but its distribution is more uniform throughout the year. Both simulated and observed spectra show statistically significant peaks relative to a red-noise null hypothesis from 16 to 60 months.

For each model configuration two runs are produced: a control run (in which no forcing is applied) and a forced run (in which extratropical forcing is imposed). The applied forcing pattern consists of cooling in one hemisphere and warming in the other poleward of 40∘, applied only over ocean grid points, and with a resulting global average forcing equal to zero. This pattern is similar to the one used in Kang et al. (2008) and in Talento and Barreiro (2016), and it is intended to represent the asymmetric temperature changes associated with glacial–interglacial and millennial-scale climate variability as well as the asymmetric SST pattern characteristic of the global warming trend. The forcing pattern is superimposed on a background state and is obtained as explained in Talento and Barreiro (2016).

The forcing pattern is shown in Fig. 3, in which sign convention is positive out of sea and, therefore, positive values of the forcing could be considered to represent a situation where the atmosphere is dry and colder than the ocean below it so that there is a strong ocean-to-atmosphere net heat flux. This forcing generates a near-surface temperature (NSAT) anomaly response of up to 16 ∘C (-20 ∘C) over the north Atlantic Ocean at 70∘ N (over the Ross Sea), as shown in Fig. 4. For comparison, in a climate simulation of the last 21 000 years (TRACE2k experiment, He, 2011) anomalies of about -10 ∘C (6 ∘C) are obtained over the North Atlantic (Antarctica) during Heinrich stadial 1, 18 000 to 15 000 years ago.

As mentioned before we use two ocean models. When the AGCM is coupled to a slab ocean model, the experiments are named Control_slab and Forced_slab. If an RGO is used in the tropical band while the slab ocean model is applied elsewhere, the corresponding experiments are named Control_slab + rgo and Forced_slab + rgo. In all the simulations the model was run for 70 years and the last 50 are used for averaging. Running the simulations for 70 years proved to be more than enough to reach the equilibrium; a timescale of 10 years was estimated to be the time span necessary for adjustment. In Table 1 we summarize the experiments. All the data are publicly available in Talento and Barreiro (2018).

In this subsection we compare the results obtained with the two implemented configurations in terms of annual means of different fields. The results are presented in the form of anomalies with respect to the corresponding control case.

Figure 4 shows the near-surface air temperature (NSAT) changes with respect to the corresponding control for the two configurations. In both experiments there is generalized warming (cooling) in the NH (SH), while in the southern tropics a strengthening of the zonal gradient is evident. The most pronounced differences between the two configurations are seen in the tropical region, in which the slab plus rgo configuration anomalies tend to be up to 1 ∘C weaker than the slab configuration anomalies. In particular, the equatorial Pacific cooling seen in the slab configuration is no longer present in the rgo + slab configuration, and the southeastern ocean basins do not cool as much. This suggests that, overall, tropical ocean dynamics tend to oppose changes in the annual mean conditions. In the extratropics the differences between the two configurations are barely noticeable, although regions of up to 2 ∘C are noted in the vicinity of the Antarctic Peninsula and Greenland.

As a consequence, tropical changes in precipitation are weaker when using the RGO: while in both experiments the most pronounced feature is a northward shift of the ITCZ, anomalies for the slab + rgo configuration are much weaker (Fig. 5). Also, in the slab configuration strong changes of tropical precipitation are found equally over the three ocean basins, but for the slab + rgo setup the most intense anomalies are seen over the Atlantic Ocean concurrent with a still relatively strong cross-equatorial SST gradient, suggesting a larger role for continental temperatures in controlling the position of the ITCZ (as in Talento and Barreiro, 2016). Changes in the subtropical convergence zones are also weaker in the slab + rgo configuration, differently from the southward shift seen in the slab configuration. In particular for the case of the South Atlantic convergence zone, this result is consistent with Talento and Barreiro (2016), which showed that during southern summer the weakening of the SACZ is related to the development of a region of strong rainfall in the tropical north Atlantic.

As expected from the above results, both experiments present similar patterns of near-surface (950 hPa) wind anomalies (Fig. 6): anomalous northward winds associated with the ITCZ northward displacement in the tropics, and anomalous westerly winds over the Southern Ocean. The Forced_slab + rgo response is weaker in the tropics but stronger over the Southern Ocean, compared to the Forced_slab response. Similar pictures of a weaker response in the case of slab + rgo configuration can be seen in upper-level winds, mean sea level pressure and mass stream-function (not shown).

To summarize, in Fig. 7 we present the northward atmospheric energy transport for the control and forced runs in the two configurations implemented. As can be seen, while the control runs display almost identical transport, the forced runs significantly disagree in magnitude in the tropical region, with the slab + rgo configuration producing the weakest changes with a decrease of the transport toward the southern high latitudes, representing damping by a factor of 1.9. In the perturbed runs, the energy flux equator is located around 12∘ N (8∘ N) for the slab (slab + RGO) configuration, equivalent to a damping in the shift by a factor of 1.5.

As the previous subsection showed, the most pronounced differences between the two implemented configurations are found in the tropical band. Therefore, for the analysis of variations in the seasonal cycle we will focus on the 30∘ S–30∘ N region.

Three-month means of SST and near-surface wind changes for the tropics are shown in Fig. 8. In the Pacific Ocean, for the Forced_slab experiment negative SST anomalies are seen reaching the Equator (or even more to the north) in all four seasons, September–November (SON) being the period of strongest cooling and June–August (JJA) being the period in which the negative anomalies have the weakest penetration into the NH. Meanwhile, for the Forced_slab + rgo experiment the negative SST anomalies barely reach the Equator and, in fact, positive anomalies are the ones penetrating into the SH for the seasons March–May (MAM) and JJA. Consistent with these changes in response, the equatorial anomalous winds in the slab configuration are mainly easterlies throughout the year, while in the slab + rgo configuration they have a marked northward component and are eastward during MAM season. Also, the equatorial Atlantic tends to warm up during most of the year when using the RGO model.

Equatorial (2∘ S–2∘ N) de-meaned seasonal cycles for near-surface zonal wind and SST anomalies in the Pacific basin for the two experimental configurations are shown in Figs. 9 and 10 respectively. In Forced_slab there are eastward (westward) near-surface wind anomalies from December to May (June to November) distributed along the basin. The positive wind anomalies in the Forced_slab are quite uniform along the basin, although there are maximums in the western and eastern ends during December–February (DJF). The negative anomalies during the second half of the year are maximal in the central-eastern basin. In the slab configuration, the equatorial SST response to these wind anomalies is, however, very weak (Fig. 10a). On the other hand, when the RGO is coupled (and although the annual mean anomalies were even weaker than for the slab configuration; Fig. 4) there are substantial changes occurring to the seasonal cycle of SST in the central-eastern basin: from April to August (October to December) the forced run produces warming (cooling) of up to 1 ∘C (-0.8 ∘C). The location and timing of these anomalies lead to a substantial strengthening of the SST seasonal cycle in the eastern Pacific Ocean (overlap Fig. 10b on the Control_slab + rgo SST seasonal cycle shown in Fig. 1c).

The thermocline depth shows consistent changes when the RGO is used (Fig. 11): a deepening in the east of the basin starting around March and finishing in July, consistent with the warmer SSTs seen in the region (with 1-month lag). Considering the wind anomalies of the slab setup (Fig. 9a) as the forcing pattern for the ocean dynamics derived from the extratropical signal, this thermocline-deepening pulse appears to be initiated during the SH summer in the west of the basin (due to a weakening of the trades) and is propagated eastward as a Kelvin wave, reaching the eastern boundary 2–3 months later. The deepening of the eastern Pacific thermocline is concurrent with a shallowing in the western Pacific particularly from May to July, and vice versa (but less obvious) in other seasons of the year. In the second half of the year the strengthening of the trades locally shallows the thermocline in the eastern Pacific and the western Pacific recovers its mean depth.

In summary, the equatorial near-surface zonal wind changes caused by the extratropical forcing seen in the slab configuration induce dynamical ocean–atmosphere coupling that generates seasonal changes in the SST field when the RGO is used. This results in late austral autumn warming and cooling in spring and summer in the equatorial eastern Pacific, leading to a strengthening of the SST seasonal cycle with consistent changes in the thermocline depth.

In this subsection we investigate how the interannual variability in the tropical Pacific is affected by the interhemispheric SST gradient induced by the imposed forcing.

The leading pattern of co-variability of SST and near-surface wind in the tropical Pacific basin when the extratropical forcing is applied is weaker than that obtained when no forcing is implemented (Figs. 12a and 2a), and it explains a smaller percentage of the total variability (46 % compared to 62 %). The phase-locking to the seasonal cycle (Figs. 12b and 2c) is also modified, being more uniformly distributed and with a peak season from July to the end of the calendar year. The frequency spectrum of the ENSO pattern under the effect of the extratropical forcing is characterized by shorter periods than in the absence of the forcing and has a peak at 24 months (Figs. 12c and 2e).

The weakening of the ENSO activity can be understood in relation to the changes produced by the extratropical forcing on the SST seasonal cycle in the eastern Pacific Ocean. According to the nonlinear frequency entrainment mechanism (Chang et al., 1994) ENSO amplitude is anticorrelated with the strength of the SST seasonal cycle. The frequency entrainment implies that a self-exciting oscillator (like ENSO) will give up its intrinsic mode of oscillation in the presence of strong external forcing (like a strong seasonal SST cycle) and acquire the frequency of the applied oscillating forcing. Therefore, in our case, as the extratropical forcing generates significant strengthening of the eastern Pacific SST seasonal cycle, a weakening of ENSO is expected according to this mechanism.

Assuming linear behaviour holds, our result of ENSO weakening is also in agreement with Timmermann et al. (2007). These authors analyse fully coupled GCMs in the context of an Atlantic meridional overturning circulation (AMOC) slowdown, producing generalized cooling of the NH and warming in the SH, and find that most of the models predict a ENSO intensification attributed to a seasonal cycle weakening. The weakening of ENSO activity in the presence of a northward ITG is also consistent with the work of Chiang et al. (2008), who use an AGCM coupled to an RGO model, a model configuration similar to ours. Although they do not attempt to explain the causes, they find that ENSO is sensitive to ITG with maximal activity when the ITG is close to zero and a weakened performance as the gradient increases in any direction.