1. Introduction
El Niño–Southern Oscillation (ENSO) is the strongest climate signal on interannual time scales and has global impacts. One particularly robust impact is the warming in the northern tropical Atlantic (NTA) that occurs during the decay phase of El Niño in boreal spring [March–May (MAM); hereafter seasons refer to the Northern Hemisphere]. Despite this robust influence on the NTA, ENSO’s influence on the equatorial Atlantic is notoriously inconsistent (Chang et al. 2006; Lübbecke and McPhaden 2012), with El Niño events sometimes followed by warming in the equatorial Atlantic and sometimes by cooling. This is illustrated by the fact that the major El Niño events of 1982/83 and 1997/98 were followed by the negative and positive phases of the equatorial Atlantic zonal mode (AZM), respectively (Figs. 1a,b). Likewise, the response to La Niña events is a robust cooling in the NTA but inconsistent in the equatorial Atlantic.
OISST anomalies (K; shading) and ERA5 surface wind anomalies (vectors; reference 1 m s−1) averaged over May–July (MJJ) for (a) 1983 and (b) 1998. (c) The Niño-3.4 SST (K; green lines) and ATL4 surface zonal wind (m s−1; blue lines) for 1982/83 (solid lines) and 1997/98 (dashed lines).
Citation: Journal of Climate 38, 2; 10.1175/JCLI-D-24-0182.1
ENSO events can influence the tropical Atlantic through several pathways. 1) Warm sea surface temperature (SST) anomalies in the central and eastern tropical Pacific displace the Walker circulation, which leads to anomalous descent and easterly winds over the equatorial Atlantic (Kidson 1975; Saravanan and Chang 2000; Wang 2002; Huang 2004). These easterly winds, in turn, induce upwelling and cooling in the eastern equatorial Atlantic. 2) Convective anomalies over the equatorial Pacific initiate a quasi-stationary Rossby wave train, known as the Pacific–North America (PNA) pattern, that extends across North America and into the NTA, where it causes a weakening of the trade winds in the case of El Niño and a strengthening in the case of La Niña (Wallace and Gutzler 1981; Lee et al. 2008). The associated latent heat flux anomalies cause NTA SST anomalies of the same sign as those in the tropical Pacific. These subtropical impacts may also influence the equatorial Atlantic through changes in the Hadley circulation. 3) Equatorial atmospheric waves associated with El Niño events quickly spread El Niño–related warming through the upper tropical troposphere (Chiang and Sobel 2002), thereby suppressing convection over other tropical regions, including the tropical Atlantic. This reduces surface cooling through latent heat flux and increases surface warming through downward shortwave radiation, resulting in a warming of the tropical Atlantic SST.
Several hypotheses have been put forward to explain the inconsistent influence of ENSO on the AZM. Chang et al. (2006) realized that, in the equatorial Atlantic region, the dynamic (1) and thermodynamic (3) ENSO impacts are of opposite sign and thus partially cancel. Tokinaga et al. (2019) pointed out that multiyear ENSO events feature a relatively robust, opposite-signed AZM response because in those events, the ENSO SST anomalies persist into spring. This leads to a stronger dynamic impact because the intertropical convergence zone (ITCZ) is close to the equator in this season (e.g., Richter et al. 2017) and thus enables a robust surface wind response.
Lübbecke and McPhaden (2012) highlighted the potential role of oceanic Rossby waves. In the spring following an El Niño event, surface wind anomalies are westerly over the NTA (due to the PNA mechanism) but easterly on the equator (due to the dynamic ENSO impact). This results in negative wind stress curl anomalies north of the equator that force downwelling Rossby waves. Upon reaching the western boundary, these waves are reflected into downwelling Kelvin waves that counteract the initial cooling induced by the equatorial easterly wind anomalies.
There is also the potential role of intrinsic tropical Atlantic variability. The study by Giannini et al. (2004) was one of the first studies to point to an important role for internal Atlantic variability. Kido et al. (2023) used a linear inverse model (LIM) framework to show that preexisting SST anomalies in the tropical Atlantic may have played a role in the different outcomes of the 1982/83 and 1997/98 El Niños.
Finally, it should be noted that the inconsistent influence of ENSO on the AZM may be partly attributed to the different seasonal preferences of these two phenomena: while ENSO peaks in winter [December–February (DJF)], the AZM develops in MAM and peaks in summer [June–August (JJA); see, e.g., reviews by Lübbecke et al. 2018 and Richter and Tokinaga 2021). Thus, the most pronounced SST anomalies in the tropical Pacific during fall and winter occur during a time when the AZM is not responsive due to its seasonal phase locking (Richter et al. 2017; Nnamchi et al. 2021).
While many studies have investigated the reasons for the inconsistent ENSO influence on the AZM, several aspects merit further study. Even though the competition of dynamic and thermodynamic influences can explain a weak impact on the AZM, it cannot explain the fact that pronounced AZM events of either sign can follow an ENSO event, as was the case in 1982/83 and 1997/98. The same argument also applies to the delayed feedback from off-equatorial Rossby waves. It is therefore of interest to ask how the various proposed mechanisms conspire to determine the response of the equatorial Atlantic to a given ENSO event and what their relative importance is. A better understanding of the remote ENSO influence may allow more skillful predictions of the AZM and its regional climate impacts. Current seasonal prediction systems can produce skillful AZM predictions about 3 months into the future (Richter et al. 2018; Exarchou et al. 2021), corresponding to initializations in spring. Can this be extended to 6-month lead time, i.e., winter initializations?
Here, we address these questions by focusing on preindustrial control (piControl) simulations in the archive of phase 6 of the Coupled Model Intercomparison Project (CMIP6; Eyring et al. 2016). The advantage of using these general circulation model (GCM) simulations is that they offer long time series under steady radiative forcing, thus facilitating robust results. The obvious disadvantage of GCMs is that they are subject to biases, particularly in the tropical Atlantic (Richter and Xie 2008; Tozuka et al. 2011; Richter et al. 2014). Recent studies, however, suggest that despite mean state biases, some GCMs can simulate tropical variability relatively realistically, including in the tropical Atlantic (Richter and Tokinaga 2020). In addition, disparate model behavior may also offer insights into the mechanisms that govern the ENSO–AZM relation.
In the next section, we introduce the reanalysis and CMIP6 datasets used in this study and describe our methodology. Section 3 first examines how the ENSO–AZM relation is simulated in each model (section 3a) and then analyzes some of the hypotheses that have been put forward to explain the inconsistent influence of ENSO on the AZM, namely, the competition of dynamic and thermodynamic impacts (section 3b), the Rossby wave mechanism (section 3c), the timing of ENSO decay (section 3d), and the role of intrinsic Atlantic variability (section 3e). In section 4, we build a simple prediction model based on the insights gained in section 3 and examine its skill in predicting the JJA AZM from conditions in the preceding DJF. Conclusions are given in section 5.
2. Methods and data
The present study focuses on the CMIP6 piControl simulations, which are typically integrated over 500–1000 years under steady radiative forcing representative of the year 1850. The 44 models analyzed in this study are listed in supplemental Table 1 in the online supplemental material, together with their integration periods and other details. The climatological seasonal cycle of the monthly means is calculated for the whole integration period and subtracted from the total fields to obtain the anomalies. All model fields are linearly detrended, even though many do not feature any significant trends due to the steady radiative forcing.
As an observation-based reference, we use the fifth major global reanalysis produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) (ERA5; Hersbach et al. 2018) and the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis (hereafter NCEP reanalysis; Kalnay et al. 1996). The reanalysis fields are linearly detrended. The analysis period spans 42 years from the satellite observation period (1980–2021). Reanalyses are preferred because they offer a set of physically consistent variables.
An important tool for this study is composite analysis. Specifically, we are interested in comparing El Niño events that are followed by a positive AZM (AZM+) event against those that are followed by a negative AZM (AZM−) event. To do this, we first selected all years in which SST anomalies in the Niño-3.4 region (170°–120°W, 5°S–5°N) exceed one standard deviation during DJF, the peak of the ENSO event. From these years, we select the ones in which El Niño in DJF is followed by an AZM+ event, defined as SST anomalies in the Atlantic Niño-3 (ATL3) region (20°W–0°, 3°S–3°N) exceeding one standard deviation during the following JJA. The composite mean over these years is referred to as El Niño/AZM+. Analogously, we define the El Niño/AZM− composites as those El Niño years which are followed by an AZM− event in JJA. We denote the year beginning with the ENSO peak followed by AZM as year 1 and the preceding year as year 0. If not otherwise noted, the analysis presented here shows the multimodel average composites. The percentage of ENSO events followed by AZM+, AZM−, or neutral AZM is summarized in supplemental Fig. 1.
Our focus here is on El Niño composites, but analysis of La Niña composites shows very similar results, suggesting that nonlinearities are relatively weak (supplemental Figs. 4–6 and 9).
3. Mechanisms for ENSO’s influence on the equatorial Atlantic
a. ENSO’s relation with the equatorial Atlantic in CMIP6
We start by examining how the individual models represent the ENSO–AZM relation and how they compare to the reanalyses. A simple metric is the correlation between the Niño-3.4 SST anomalies in DJF and the ATL3 SST anomalies in the following JJA (Fig. 2a). Both reanalyses show correlation coefficients that are close to zero and not statistically significant, indicative of the inconsistent influence of ENSO on the AZM. The CMIP6 models, on the other hand, show a wide range of behaviors, with correlations ranging from −0.5 to 0.5. Many of these correlation coefficients are significant at the 95% confidence level.
Correlations of the Niño-3.4 SST index with Atlantic indices for ERA5 and NCEP reanalyses (leftmost bars), and 44 CMIP6 piControl simulations. Light blue and dark blue shading indicates values whose p value is greater than 0.05 and less than 0.05, respectively. The two panels show (a) correlation of the DJF Niño-3.4 SST with the JJA ATL3 SST (two-season lag) and (b) correlation of the MAM Niño-3.4 index with the MAM ATL4 surface zonal wind.
Citation: Journal of Climate 38, 2; 10.1175/JCLI-D-24-0182.1
A very different picture emerges when we correlate the Niño-3.4 SST anomalies in MAM with surface zonal wind anomalies in the ATL4 region (45°–20°W, 3°S–3°N) for the same season (Fig. 2b). All correlations are negative and, except for two models, significant at the 95% confidence level. Correlations range from about −0.1 to −0.8, with the reanalysis showing −0.7 (ERA5) and −0.5 (NCEP). This robust relation is consistent with the dynamic influence of ENSO on the equatorial Atlantic, i.e., an El Niño event drives easterly wind anomalies over the equatorial Atlantic. If this were the only mechanism by which ENSO influences the equatorial Atlantic, we would expect El Niño to be always followed by an AZM− event.
Seasonally stratified regressions show that the influence of ENSO on equatorial Atlantic surface zonal winds is highly seasonal (Fig. 3a), with the most negative value in May (−0.57 m s−1 K−1) and the least negative value in October (−0.04 m s−1 K−1). This is consistent with previous studies that suggested such a seasonality (Münnich and Neelin 2005; Sasaki et al. 2015; Tokinaga et al. 2019). These studies suggested that the proximity of the ITCZ to the equator is responsible for the strong impact of the equatorial Pacific on equatorial Atlantic surface winds. We note, however, that the impact on surface winds is also highly seasonal over the NTA, with the strongest impacts in winter and spring (supplemental Fig. 2).
Seasonally stratified regression of (a) ATL4 surface zonal wind on the contemporaneous Niño-3.4 SST (m s−1 K−1), and (b) net surface energy flux averaged over the equatorial Atlantic (50°W–10°E, 5°S–5°N) on the contemporaneous Niño-3.4 SST (W m−2 K−1). The lines in each panel show regression coefficients for the CMIP6 multimodel average (green; intermodel standard deviation indicated by light green shading), the ERA5 reanalysis (blue), and the NCEP reanalysis (orange). Regression coefficients that are significant at the 95% confidence level are marked by filled circles.
Citation: Journal of Climate 38, 2; 10.1175/JCLI-D-24-0182.1
The net surface energy flux (the sum of latent heat, sensible heat, net longwave radiation, and net shortwave radiation; positive upward), averaged over the equatorial Atlantic (50°W–10°E, 5°S–5°N), displays a less pronounced seasonality when regressed onto the Niño-3.4 SST index (Fig. 3b), with values in the CMIP6 ensemble mean ranging between −2.5 and −4 W m−2 K−1 from January through June. Based on these regression coefficients and a composite ENSO event, we estimate the thermodynamically induced SST changes in the equatorial Atlantic to be approximately 0.5 K (supplemental Table 2), enough to cause a weak-to-moderate AZM event.
We note that the seasonality of the energy flux regression coefficient is equally weak in the NCEP reanalysis but quite pronounced in the ERA5 reanalysis (though still weaker than that of the surface winds). The discrepancy between ERA5 and NCEP suggests that the surface heat fluxes are poorly constrained. The weak seasonality in the model average compared to ERA5 may point to systematic model biases, but due to the high uncertainty in the reanalysis products, we do not investigate this further in the present study.
The different seasonality of the thermodynamic and dynamic influences could explain why slowly decaying El Niño events have a robust association with AZM− events: The thermodynamic influence of El Niño consistently warms the equatorial Atlantic throughout the life cycle of the event, but the dynamic influence, while strong, only has a narrow window in spring. Thus, if the El Niño event decays quickly, the thermodynamic influence wins out. If, however, the El Niño event persists into spring, the dynamic influence gets the upper hand.
We note that we have chosen a wider area for the regression analysis of the energy flux. This is motivated by the fact that the energy flux anomalies can influence the tropical Atlantic SST rather uniformly and that the resulting SST anomalies can subsequently be advected toward the ATL3 region, e.g., through subsurface meridional advection from north of the equator (Richter et al. 2013) or through surface westward advection. The impact of the surface winds, on the other hand, is more concentrated in the western equatorial Atlantic, where they can effectively excite equatorial Kelvin waves. If the regression analysis is limited to the ATL3 region, the seasonality is more pronounced but still weaker than that of the dynamic effect (not shown).
The strength of the ENSO–AZM relationship may be subject to modulation on interdecadal time scales, as has been argued for the influence of the AZM on ENSO (Rodríguez-Fonseca et al. 2009; Martín-Rey et al. 2014), although it has been argued that this could be just due to intrinsic Pacific variability (Zhang et al. 2021). We examine the potential interdecadal modulation by calculating running correlations with a 51-yr window (Fig. 4). The correlation between the DJF Niño-3.4 SST and the following JJA ATL3 SST displays a considerable range and crosses the zero line in many models (Fig. 4a). The correlation between MAM Niño-3.4 SST and MAM ATL4 surface zonal winds generally shows a smaller range (Fig. 4b), with most models featuring consistently negative correlations throughout the integration period. Nevertheless, there is considerable spread in this correlation as well. This may be related to interdecadal shifts of the Atlantic ITCZ, with periods of proximity to the equator marked by stronger correlations. The running average of the ITCZ latitude shows support for this in some models (supplemental Fig. 3) but more careful analysis will be needed to confirm this mechanism. Finally, the running correlation of DJF Niño-3.4 SST and DJF net surface energy flux in the equatorial Atlantic is also relatively robust (Fig. 4c). The majority of models maintain negative correlations throughout their integration periods and have a median below −0.4.
Box plots of 51-yr running correlations. The minimum and maximum correlations are shown by the lower and upper whiskers, respectively, while the lower and upper edge of the box denote the lower and upper quartiles, respectively. The horizontal bar in each box denotes the median. The running correlations shown are (a) DJF Niño-3.4 SST and following JJA ATL3 SST, (b) MAM Niño-3.4 SST and MAM ATL4 surface zonal wind, and (c) DJF Niño-3.4 SST and DJF net surface energy flux averaged over the equatorial Atlantic.
Citation: Journal of Climate 38, 2; 10.1175/JCLI-D-24-0182.1
b. Competition of thermodynamic and dynamic effects
We compare composite El Niño/AZM− events (Fig. 5a) with El Niño/AZM+ events (Fig. 5b). The Pacific SST evolution is similar in both, but in the former, El Niño persists longer into spring than in the latter. Consistent with the regression analysis in section 3a (Fig. 3a), the longer persisting El Niño in El Niño/AZM− is associated with substantial easterly wind anomalies over the equatorial Atlantic in spring. These easterly wind anomalies drive cooling in the eastern equatorial Atlantic (Fig. 5a). In El Niño/AZM+, on the other hand, only June shows some weak wind anomalies over the equatorial Atlantic (Fig. 5b). These anomalies are westerly, opposite to what would be expected from the El Niño dynamic forcing, and are likely a response to the SST warming in the eastern equatorial Atlantic.
Multimodel means of composite longitude–time sections, meridionally averaged over 3°S–3°N. (a) SST (shading; K) and surface zonal wind anomalies (contours; interval 0.5 m s−1; dashed contours indicate negative values; zero-contour omitted) over the Pacific and Atlantic for the El Niño/AZM− composite. Only values that are significant at the 95% confidence level are shown and land areas are masked out (South American landmass from about 80°–50°W). (b) As in (a), but for the El Niño/AZM+ composite. (c) SST (shading; K) and net surface energy flux (contours; interval 2 W m−2; positive upward; dashed contours indicate negative values) over the Atlantic for the El Niño/AZM− composite. (d) As in (c), but for the El Niño/AZM+ composite.
Citation: Journal of Climate 38, 2; 10.1175/JCLI-D-24-0182.1
If the equatorial Atlantic warming in El Niño/AZM+ is not primarily driven by wind anomalies, how is it generated? The composite net surface energy flux (Fig. 5d) suggests persistent downward anomalies (i.e., ocean warming) starting from late winter in the El Niño developing year and continuing until April of the El Niño decaying year. As the underlying SST anomalies exceed 0.5 K from May onward, the net surface energy flux becomes positive and thus damps the AZM+ event. This is expected due to the dependence of turbulent heat flux and longwave radiation on SST. Persistent warming by the net surface energy flux can also be seen in El Niño/AZM− (Fig. 5c). Here, the downward flux becomes even more pronounced as the AZM− event matures, due to the influence of the underlying SST. The composite analysis thus suggests that El Niño invariably leads to surface heat flux anomalies that warm the equatorial Atlantic, thus underscoring the role of the thermodynamic effect. In El Niño/AZM+, this leads to a gradual warming of the equatorial Atlantic SSTs that is roughly in synchrony with the equatorial Pacific. In El Niño/AZM−, however, the warming is counteracted by the development of easterly wind anomalies over the equatorial Atlantic in MAM, thus highlighting the role of the dynamic ENSO effect. The corresponding La Niña composites exhibit very similar behavior (supplemental Fig. 4).
c. Destructive interference from reflected Rossby waves
We examine the potential impact of Rossby waves excited north of the equator (Fig. 6). In the case of El Niño events, one would expect easterly wind anomalies on the equator and westerly anomalies to the north, resulting in negative wind stress curl anomalies north of the equator that excite downwelling Rossby waves. These waves should lead to a delayed warming on the equator, i.e., an AZM+ event. Thus, we would expect this effect to be prominent in the El Niño/AZM+ composites (Figs. 6c,d). Specifically, we would expect to see evidence for westward propagation north of the equator from late winter through late spring. Inspection of the composites, however, does not indicate an important role for off-equatorial processes. Anomalies of the 20°C isotherm depth (Z20; a proxy for thermocline depth), averaged between 3° and 6°N, show no sign of westward propagation (Fig. 6c), and wind stress curl anomalies are weak and inconsistent. Rather, Z20 anomalies develop simultaneously at 2°S–2°N and 3°–6°N. The El Niño/AZM− composites (Figs. 6a,b) show substantial negative wind stress curl anomalies in the center of the basin (Fig. 6a) but no westward propagation of Z20 anomalies. Here too, the Z20 anomalies develop simultaneously on the equator and to the north of it. The corresponding La Niña composites show very similar behavior (supplemental Fig. 5).
Multimodel means of composite longitude–time sections. (a) Anomalous Z20 (shading; m) and wind stress curl (contours; interval 0.5 × 10−8 N m−3; dashed contours indicate negative values; zero-contour omitted) over the Atlantic for El Niño/AZM−. The fields are meridionally averaged from 3° to 6°N. Only values that are significant at the 95% confidence level are shown. (b) As in (a), but averaged from 2°S to 2°N and only showing Z20. (c) As in (a), but for El Niño/AZM+. (d) As in (b), but for El Niño/AZM+. Note that the direction of the x axis is reversed in the left panels in order to facilitate visual inspection of the wave reflection at the western boundary.
Citation: Journal of Climate 38, 2; 10.1175/JCLI-D-24-0182.1
The timing of Rossby wave excitation may differ from year to year and across models. Averaging over all these events may therefore dilute the Rossby wave signal. To investigate this, we look at one particular model, the CESM2, in which El Niño/AZM+ events occur relatively frequently (Fig. S1a). Out of the 28 El Niño/AZM+ events that occurred during the integration period, we categorized eight as showing clear signs of north equatorial Rossby waves based on visual inspection. The timing of these waves is quite similar (not shown), and we therefore only show the composite mean over these eight events (Fig. 7). Positive Z20 anomalies are propagating westward from the eastern boundary into the central and western basin, consistent with Rossby wave propagation. West of 20°W, this is consistent with the negative wind stress curl anomalies that develop in January of year 1. The Z20 anomalies are reflected at the western boundary in March–April and propagate eastward along the equator to about 30°W in May. East of 30°W, the propagation is not obvious, but positive anomalies in the eastern equatorial Atlantic emerge in June and July. Overall, the analysis provides some support for the Rossby wave mechanism in selected years.
As in Figs. 6c and 6d, but for CESM2 only and averaged over a subset of El Niño/AZM+ years. Note that the direction of the x axis is reversed in the left panels in order to facilitate visual inspection of the wave reflection at the western boundary.
Citation: Journal of Climate 38, 2; 10.1175/JCLI-D-24-0182.1
d. Importance of the timing of ENSO decay
The composite evolution of the two types of events (Fig. 8) suggests that, on average, El Niño events that persist longer into spring are followed by AZM− events, consistent with the analysis shown in sections 3a and 3b. Differences in the Niño-3.4 index between El Niño/AZM− and El Niño/AZM+ events start to appear in March, with warmer anomalies in the former. As this SST difference grows, so does the difference in ATL4 surface zonal wind anomalies, with pronounced negative anomalies in El Niño/AZM− but near neutral winds in El Niño/AZM+. The easterly wind anomalies in the former are closely followed by negative Z20 anomalies (shoaling thermocline) and, eventually, cool SST anomalies in the ATL3 from June onward. In the La Niña composites (supplemental Fig. 6), the evolution is rather similar, but the distinction between La Niña/AZM+ and La Niña/AZM− evolution is somewhat less pronounced.
Multimodel means of composite anomalies for El Niño/AZM− (solid lines) and El Niño/AZM+ (dashed lines). The line colors denote Niño-3.4 SST (K; green), ATL3 SST (K; blue), ATL4 surface zonal wind (m s−1; orange), and ATL3 Z20 (m; brown). Filled circles denote values for which the difference between the two composites is significant at the 95% confidence level.
Citation: Journal of Climate 38, 2; 10.1175/JCLI-D-24-0182.1
It should be noted that surface zonal wind anomalies in the ATL4 start to diverge in December, 2 months before the difference in Niño-3.4 SST develops. This may be due to the Niño-3.4 region not representing the SST anomalies that are most relevant to the impact on ATL4 surface winds. It could also hint, however, to other influences, such as intrinsic tropical Atlantic variability. This is examined in section 3e.
The ATL3 SST anomalies in El Niño/AZM+ develop gradually from November of year 0 (dashed blue line in Fig. 8), consistent with the analysis in section 3b. The ATL4 wind anomalies do not appear to respond to these SST anomalies until May (dashed orange line in Fig. 8). This may be partly due to air–sea coupling in the equatorial Atlantic being stronger in spring, and partly due to competing influences, with the equatorial Atlantic SST gradient driving westerly wind anomalies while the remote Pacific influence drives easterly anomalies. A similar cancellation of effects may occur for the ATL3 SST anomalies in El Niño/AZM−, where SST cooling by the easterly wind anomalies and thermocline shoaling are counteracted by the remotely driven warming from the tropical Pacific.
In section 1, we discussed the conundrum of the major El Niño events in 1982/83 and 1997/98 being followed by negative and positive AZM events, respectively. Can the timing of the decay explain this observation? The evolution of the Niño-3.4 index in these 2 years (Fig. 1c) shows similar behavior as in Fig. 8, with the faster decay of the 1997/98 event evident from March and consistent differences in the ATL4 surface zonal winds. The Niño-3.4 differences, however, are subtle, ranging from about 0.1 to 0.3 K during spring. This indicates that other factors, such as intrinsic Atlantic variability, likely played a role.
One may ask to what extent ENSO persistence can explain intermodel differences in the ENSO–AZM correlation (Fig. 2a). Do models with a more persistent ENSO have a more negative correlation? An intermodel analysis suggests that there is no simple relationship (Fig. S7a). Instead, there appears to be a bimodal distribution, with one group of models following the expected anticorrelation. Likewise, one may ask whether the competition between dynamic and thermodynamic influences can explain the intermodel spread in ENSO–AZM correlations. Analysis suggests a weak-to-moderate correlation with a few outliers (Fig. S7b).
Model biases are one reason why it is difficult to explain the intermodel spread of the ENSO–AZM correlation. In some models, westerly wind anomalies over the equatorial Atlantic lead to subsequent cooling, as is the case for the three GISS models (Fig. S8). A comprehensive analysis of the reasons for this unrealistic behavior is beyond the scope of the present work, but it is clear that it can fundamentally alter the way a given model responds to ENSO forcing. We therefore focus on the ensemble means, where model errors partially cancel each other out.
e. Role of preconditioning
We follow the evolution of the differences between the two composites from JJA0 through MAM1 (Fig. 9). Weak positive SST anomalies can be seen in the Atlantic between the equator and approximately 30°S in JJA0 (Fig. 9a). These are accompanied by negative SST anomalies further south, roughly consistent with the negative phase of the South Atlantic subtropical dipole (SASD) mode (Venegas et al. 1996, 1997; Morioka et al. 2011). The negative SLP anomalies covering the South Atlantic are also consistent with a negative SASD. These SLP anomalies appear to be linked to a wave train emanating south of Australia. The prominent warm surface temperature anomalies in the Pacific sector of the Southern Ocean are likely related to sea ice anomalies. It is not clear whether these surface temperature anomalies contribute to the wave train and the negative SLP anomalies over the South Atlantic. SST differences in the tropical Pacific appear to be too weak to explain the differences in the Southern Ocean and South Atlantic.
Multimodel mean of the difference between the El Niño/AZM+ and El Niño/AZM− composites, showing SST (shading; K), surface wind (vectors; reference 1 m s−1), and sea level pressure (contours; interval 0.25 hPa; negative contours dashed; zero-contour omitted). Only values significant at the 95% confidence level are shown. The individual panels show the following seasons: (a) JJA0, (b) SON0, (c) DJF1, and (d) MAM1.
Citation: Journal of Climate 38, 2; 10.1175/JCLI-D-24-0182.1
The patterns seen in JJA0 persist into SON0 (Fig. 9b), with the SASD-like pattern slightly strengthening. In DJF1 (Fig. 9c), the positive SST difference strengthens substantially and is accompanied by a westerly wind difference. In MAM1 (Fig. 9d), finally, eastern tropical Pacific SSTs become substantially cooler in El Niño/AZM+ than in El Niño/AZM−, which is associated with more pronounced westerlies over the equatorial Atlantic. The SASD-like pattern is still present, but it seems that the SLP pattern in the Southern Hemisphere is starting to shift. With the pronounced difference in equatorial Atlantic surface winds, it is obvious that the SST difference is set to further increase in JJA1 (see Fig. 8).
In the La Niña composites (supplemental Fig. 9), the evolution is very similar but the SLP anomalies over the Southern Ocean are less distinct, as are the surface temperature anomalies off Antarctica.
It has been suggested that the SASD is active on multiannual to interdecadal time scales (Venegas et al. 1997). Furthermore, some studies have suggested that the SASD can be a precursor of the AZM (Grimm and Reason 2011; Nnamchi et al. 2016; Ham et al. 2021). Furthermore, previous studies have indicated that the eastern Pacific ENSO is not strongly related to the SASD (Venegas et al. 1997; Kayano et al. 2013), although the central Pacific ENSO may be (Rodrigues et al. 2015). Thus, there is a possibility that variability intrinsic to the Atlantic Ocean modulates the response of the AZM to ENSO forcing. More detailed analysis and model sensitivity experiments will be needed to verify this.
4. Predicting the AZM based on a simple linear model
In section 3, we have discussed several factors that may determine how the AZM responds to ENSO forcing. Here, we would like to test whether we can construct a simple model that allows prediction of ATL3 SST anomalies in JJA based on SST anomalies in the preceding DJF. In particular, we are interested in whether the skill of this simple model improves if only ENSO years are considered. In other words, we would like to know whether the AZM becomes more predictable in ENSO years.
We use five SST-based indices for our prediction as explained in the following. The Niño-3.4 index is an obvious choice because it relates to the dynamic and thermodynamic ENSO impacts (section 3b). As a measure of ENSO persistence (section 3d), we use the Niño-3.4 SST tendency, calculated as the difference between February and December. The MAM Niño-3.4 would allow a clearer differentiation between slowly and fast decaying ENSO events but this season cannot be used for our two-season lead predictions. The NTA index (40°–10°W, 10°–20°N) indirectly relates to wind stress curl and Z20 anomalies in the northern tropical Atlantic and thus incorporates information on the Rossby wave effect (section 3c). Finally, the ATL3 and southern tropical Atlantic (STA; 20°W–20°E, 25°–5°S) indices relate to intrinsic equatorial Atlantic variability and the influence of the SASD, respectively.
To evaluate the relative importance of our five predictors, we start with a single predictor and fit it to the entire time period. The anomaly correlation coefficient (ACC) of this fit with the reference data is our metric for predictor selection. After testing each predictor individually, we select the predictor with the highest ACC. In the next step, we successively add the four unselected predictors and evaluate the fit of the two-predictor model, again selecting the one with the highest ACC. This is repeated until all five predictors are included. The ACCs of the dominant predictor in each model are shown in Fig. 10, along with the skill gains from the additional predictors. We note that this analysis only addresses the relative importance of the predictors, rather than expected prediction skill, because there is no separation between training and test periods.
Predictor analysis for each model. The largest segment of each pie chart shows the ACC of the dominant predictor, while the smaller segments show the gain in ACC obtained by adding the indicated predictor. The colors denote Niño-3.4 SST (red), NTA SST (orange), STA SST (blue), Niño-3.4 SST tendency (green), ATL3 SST (magenta), and the residual (gray).
Citation: Journal of Climate 38, 2; 10.1175/JCLI-D-24-0182.1
In 34 out of 44 models, as well as in the reanalyses, the STA emerges as the dominant predictor. This is consistent with the precursor role of the SASD. Among the remaining models, eight feature the Niño-3.4 as the dominant predictor and two feature the ATL3. In most models, the dominant predictor provides the bulk of the skill, with only incremental changes from the other predictors. Thus, in the models that are dominated by the STA, the Niño-3.4 typically only makes a small contribution. Some of the auxiliary predictors may explain similar variance as the dominant predictor due to multicollinearity. To examine this, we have calculated the variance inflation factor (Stine 1995) for all datasets and predictors. Typical values are around 1.5 and the maximum is 3.7, well below the typical threshold of 5. This suggests that multicollinearity is not a major issue and that the STA provides predictive power that is partially independent of ENSO. Systematic model errors likely influence the relative importance of predictors. The CMCC-ESM2, e.g., has an excessively strong ENSO amplitude (not shown) and, consistently, the Niño-3.4 index is the dominant predictor in that model. A detailed analysis of how individual model errors influence the importance of AZM predictors is beyond the scope of the present study. We note, however, that models with a relatively realistic representation of the AZM and its linkage to the tropical Pacific (according to the multimodel intercomparison by Richter and Tokinaga 2020), such as HadGEM3-GC31-MM, IPSL-CM6A-LR and MRI-CGCM3, support the dominant role of the STA at 6 months’ lead.
To predict the JJA ATL3 index, we next use multilinear regression using all five predictors introduced in Fig. 10. For each dataset, we evenly divide the data into a training period (first half) and a test period (second half). Thus, the training is 21 years in the reanalysis datasets (1980–2000) and 83–601 years in the piControl simulations, where the integration period ranges from 166 to 1202 years (see supplemental Table 1). For a strict test, detrending and anomaly calculation should be performed separately for the training and testing periods to avoid data leakage. Here, however, we are mostly interested in gaining a rough impression of the skill of this simple scheme and how it depends on the presence of ENSO. We therefore do not perform this strict separation.
Two prediction experiments are performed. In the first, prediction-CTRL (P-CTRL), all years are considered, as described above. In the second, prediction-ENSO (P-ENSO), we select only ENSO years (absolute Niño-3.4 > 1.0 standard deviation) for both the training and the testing.
To evaluate the performance of the scheme, we calculate the ACC with the target data, i.e., the test period in each dataset. In P-CTRL (Fig. 11a), 16 models show an ACC above 0.5, a value that is sometimes taken as the threshold for useful prediction skill. A further 16 models show ACCs above 0.4, while the reanalysis products show a slightly negative ACC. In four models, the skill exceeds 0.6, with the skill of the CMCC-ESM2 close to 0.7.
Prediction of JJA ATL3 SST from SST indices during the preceding DJF using multilinear regression. The indices used are the same as in the correlations shown in Fig. 10. The skill metric shown is the ACC. The dashed horizontal line indicates an ACC of 0.5, which is sometimes seen as the threshold for useful prediction skill. The individual panels show predictions based on (a) all years and (b) ENSO years only.
Citation: Journal of Climate 38, 2; 10.1175/JCLI-D-24-0182.1
In P-ENSO, the skill increases in some models, with 18 models achieving an ACC greater than 0.5 (Fig. 11b) and three models exceeding 0.7. There are, however, also models in which the skill does not change much (e.g., CanESM5 and CESM2) or deteriorates (e.g., EC-Earth3-Veg and HadGEM3-GC31-LL). It is possible that the restriction to pronounced ENSO years limits the available training data to such an extent that it negatively affects the skill of the scheme. The impact of the length of the training period is supported by restricting the training data to 21-yr segments (the length of the shortest training data), so that all predictions are on an equal footing (supplemental Fig. 10). The average skill of these predictions notably decreases.
In the reanalysis products, P-ENSO has higher skill than P-CTRL. The increase is particularly striking in NCEP, where the ACC exceeds 0.6. The predictions with reduced training data (supplemental Fig. 10) suggest that higher skill could be achieved in the reanalyses if more training data were available. There is, however, substantial uncertainty in the reanalyses, due to the short data record.
5. Conclusions
We have analyzed the inconsistent influence of ENSO on the equatorial Atlantic with a focus on piControl simulations from the CMIP6 archive. The correlation between the DJF Niño-3.4 and following JJA ATL3 SST indices is close to zero in the observations. In the models, this correlation shows a wide spread with both positive and negative signs represented equally. A detailed analysis demonstrates that the thermodynamic (tropics-wide warming) and dynamic (easterly winds, upwelling, and cooling) influences of El Niño oppose each other over the equatorial Atlantic, which confirms the hypothesis of Chang et al. (2006). There is, however, an important difference between the thermodynamic and dynamic influences. While the former is active throughout the life cycle of El Niño, the latter is predominantly active in spring. This means that El Niño events persisting into spring tend to produce cooling in the equatorial Atlantic, while El Niño events that decay early will produce warming. Thus, the persistence of ENSO events is crucial to their impact on the equatorial Atlantic, which confirms the results of Tokinaga et al. (2019).
Previous studies have suggested that wind stress curl anomalies in the northern equatorial Atlantic may lead to a delayed feedback on the ATL3 via Rossby wave propagation. We find that this mechanism does not play a major role on average but that it may be important in some models in individual years. More detailed analysis would be required to quantify the role of this mechanism.
Our analysis suggests that winds over the equatorial Atlantic determine the wind stress curl differences over the northern equatorial Atlantic, rather than winds to the north. In other words, El Niño events that are followed by AZM+ versus those that are followed by AZM− show pronounced surface wind differences only over the equatorial Atlantic. This limits the explanatory power of the Rossby wave mechanism because the subsequent evolution of the ATL3 SST is, on average, consistent with the equatorial wind differences, i.e., in the El Niño/AZM− case, the SSTs keep cooling relative to El Niño/AZM+, rather than showing a delayed reversal due to the arrival of Rossby waves.
We also find an important role for the South Atlantic as a precursor of AZM events, which confirms the results of previous studies (Grimm and Reason 2011; Nnamchi et al. 2016; Ham et al. 2021). This influence may act at least partially independently of ENSO and thus complicate the ENSO–AZM relation. The SASD is closely related to changes in the South Atlantic subtropical high, which have been linked to coastal SST anomalies off Namibia and Angola and subsequent AZM events (Lübbecke et al. 2010; Richter et al. 2010; Illig et al. 2020), highlighting the coupling between the South Atlantic and the equatorial Atlantic.
Prediction experiments with a simple multilinear regression scheme suggest predictive potential for the South Atlantic. Overall, the scheme shows relatively high skill for predicting JJA ATL3 based on the preceding DJF SST indices in the tropical Atlantic and Pacific. About 40% of the models achieve an ACC of 0.5 or higher. This skill is substantially higher than what current prediction systems achieve in reforecast experiments. Unfortunately, ACCs above 0.5 are generally not achieved for the reanalysis datasets, except for the NCEP reanalysis when only ENSO years are considered (ACC ∼ 0.6). The relatively poor skill for reanalyses may partly be related to the limited availability of training data. Conversely, the misrepresentation of the ENSO–AZM relation in CMIP6 models likely contributes to the higher AZM predictability in those models. Both reanalyses and about 80% of the models agree on the STA being the most important predictor. This is a strong suggestion that consideration of the South Atlantic is important for AZM prediction at a lead time of 6 months. The reanalyses and the model ensemble mean also agree on the importance of ENSO decay speed as a predictor for the AZM behavior. This, however, does not manifest until spring and may therefore be more relevant at shorter lead times.
An important caveat of this study is that we exclusively focused on the one-way influence of ENSO on the AZM. In reality, however, the AZM must also exert an influence on ENSO, though exactly how strong this influence is, remains under debate (Rodríguez-Fonseca et al. 2009; Jansen et al. 2009; Exarchou et al. 2021; Richter et al. 2021, 2023; Jiang et al. 2023; Richter et al. 2024). If the influence of the AZM on ENSO is pronounced, it complicates the interpretation of our results. In that case, El Niño may be decaying faster because the AZM+ develops, rather than the other way around (cf. Fig. 8). Furthermore, it is possible that an external factor influences both the tropical Pacific and Atlantic. More detailed analysis and dedicated model experiments will be needed to address this issue.
Acknowledgments.
The authors thank the three anonymous reviewers whose constructive comments helped to improve the manuscript. I. R., T. T., S. K., Y. K., and H. T. were supported by the Japan Society for the Promotion of Science through Grant-in-Aid for Scientific Research (JSPS KAKENHI), Grant JP23K25946, and the Kyushu University Program for Collaborative Research, Grant 2023CR-AO-9. H. T. and Y. K. were supported by JSPS KAKENHI Grant JP24H00261. H. T. was also supported by JSPS KAKENHI Grant JP24H02229. YK was also supported by the Japan Ministry of Education, Culture, Sports, Science, and Technology through the advanced studies of climate change projection program (SENTAN, JPMXD0722680395). P. C. acknowledges support from the U.S. Department of Commerce/National Oceanic and Atmospheric Administration Grant NA20OAR4310408. The authors acknowledge the World Climate Research Programme, which, through its Working Group on Coupled Modelling, coordinated and promoted CMIP6. The authors thank the climate modeling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies who support CMIP6 and ESGF. The authors thank the ECMWF for providing the ERA5 reanalysis through the Copernicus Climate Change Service (C3S) Climate Data Store and the National Oceanic and Atmospheric Administration (NOAA) Physical Sciences Laboratory (PSL) for providing the NCEP reanalysis.
Data availability statement.
All the datasets used in this study are publicly available. The ERA5 reanalysis can be obtained from the C3S Climate Data Store at https://doi.org/10.24381/cds.143582cf. The NCEP reanalysis can be obtained from https://downloads.psl.noaa.gov/Datasets/ncep/. The CMIP6 model datasets are available from the Earth System Grid Federation (ESGF) at https://esgf-node.llnl.gov/search/cmip6/.
REFERENCES
Chang, P., Y. Fang, R. Saravanan, L. Ji, and H. Seidel, 2006: The cause of the fragile relationship between the Pacific El Niño and the Atlantic Niño. Nature, 443, 324–328, https://doi.org/10.1038/nature05053.
Chiang, J. C. H., and A. H. Sobel, 2002: Tropical tropospheric temperature variations caused by ENSO and their influence on the remote tropical climate. J. Climate, 15, 2616–2631, https://doi.org/10.1175/1520-0442(2002)015<2616:TTTVCB>2.0.CO;2.
Exarchou, E., P. Ortega, B. Rodríguez-Fonseca, T. Losada, I. Polo, and C. Prodhomme, 2021: Impact of equatorial Atlantic variability on ENSO predictive skill. Nat. Commun., 12, 1612, https://doi.org/10.1038/s41467-021-21857-2.
Eyring, V., S. Bony, G. A. Meehl, C. A. Senior, B. Stevens, R. J. Stouffer, and K. E. Taylor, 2016: Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev., 9, 1937–1958, https://doi.org/10.5194/gmd-9-1937-2016.
Giannini, A., R. Saravanan, and P. Chang, 2004: The preconditioning role of Tropical Atlantic Variability in the development of the ENSO teleconnection: Implications for the prediction of Nordeste rainfall. Climate Dyn., 22, 839–855, https://doi.org/10.1007/s00382-004-0420-2.
Grimm, A. M., and C. J. C. Reason, 2011: Does the South American monsoon influence African rainfall? J. Climate, 24, 1226–1238, https://doi.org/10.1175/2010JCLI3722.1.
Ham, Y.-G., H.-J. Lee, H.-S. Jo, S.-G. Lee, W. Cai, and R. R. Rodrigues, 2021: Inter-basin interaction between variability in the South Atlantic Ocean and the El Niño/Southern Oscillation. Geophys. Res. Lett., 48, e2021GL093338, https://doi.org/10.1029/2021GL093338.
Hersbach, H., and Coauthors, 2018: Operational global reanalysis: Progress, future directions and synergies with NWP. ECMWF ERA Rep. Series 27, 65 pp., https://www.ecmwf.int/en/elibrary/80922-operational-global-reanalysis-progress-future-directions-and-synergies-nwp.
Huang, B., 2004: Remotely forced variability in the tropical Atlantic Ocean. Climate Dyn., 23, 133–152, https://doi.org/10.1007/s00382-004-0443-8.
Illig, S., M.-L. Bachèlery, and J. F. Lübbecke, 2020: Why do Benguela Niños lead Atlantic Niños? J. Geophys. Res. Oceans, 125, e2019JC016003, https://doi.org/10.1029/2019JC016003.
Jansen, M. F., D. Dommenget, and N. Keenlyside, 2009: Tropical atmosphere–ocean interactions in a conceptual framework. J. Climate, 22, 550–567, https://doi.org/10.1175/2008JCLI2243.1.
Jiang, F., W. Zhang, F.-F. Jin, M. F. Stuecker, A. Timmermann, M. J. McPhaden, J. Boucharel, and A. T. Wittenberg, 2023: Resolving the tropical Pacific/Atlantic interaction conundrum. Geophys. Res. Lett., 50, e2023GL103777, https://doi.org/10.1029/2023GL103777.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471, https://doi.org/10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.
Kayano, M. T., R. V. Andreoli, and R. A. Ferreira de Souza, 2013: Relations between ENSO and the South Atlantic SST modes and their effects on the South American rainfall. Int. J. Climatol., 33, 2008–2023, https://doi.org/10.1002/joc.3569.
Kido, S., I. Richter, T. Tozuka, and P. Chang, 2023: Understanding the interplay between ENSO and related tropical SST variability using linear inverse models. Climate Dyn., 61, 1029–1048, https://doi.org/10.1007/s00382-022-06484-x.
Kidson, J. W., 1975: Tropical eigenvector analysis and the Southern Oscillation. Mon. Wea. Rev., 103, 187–196, https://doi.org/10.1175/1520-0493(1975)103<0187:TEAATS>2.0.CO;2.
Lee, S.-K., D. B. Enfield, and C. Wang, 2008: Why do some El Niños have no impact on tropical North Atlantic SST? Geophys. Res. Lett., 35, L16705, https://doi.org/10.1029/2008GL034734.
Lübbecke, J. F., and M. J. McPhaden, 2012: On the inconsistent relationship between Pacific and Atlantic Niños. J. Climate, 25, 4294–4303, https://doi.org/10.1175/JCLI-D-11-00553.1.
Lübbecke, J. F., C. W. Böning, N. S. Keenlyside, and S.-P. Xie, 2010: On the connection between Benguela and equatorial Atlantic Niños and the role of the South Atlantic Anticyclone. J. Geophys. Res., 115, C09015, https://doi.org/10.1029/2009JC005964.
Lübbecke, J. F., B. Rodríguez-Fonseca, I. Richter, M. Martín-Rey, T. Losada, I. Polo, and N. S. Keenlyside, 2018: Equatorial Atlantic variability—Modes, mechanisms, and global teleconnections. Wiley Interdiscip. Rev.: Climate Change, 9, e527, https://doi.org/10.1002/wcc.527.
Martín-Rey, M., B. Rodríguez-Fonseca, I. Polo, and F. Kucharski, 2014: On the Atlantic–Pacific Niños connection: A multidecadal modulated mode. Climate Dyn., 43, 3163–3178, https://doi.org/10.1007/s00382-014-2305-3.
Morioka, Y., T. Tozuka, and T. Yamagata, 2011: On the growth and decay of the subtropical dipole mode in the South Atlantic. J. Climate, 24, 5538–5554, https://doi.org/10.1175/2011JCLI4010.1.
Münnich, M., and J. D. Neelin, 2005: Seasonal influence of ENSO on the Atlantic ITCZ and equatorial South America. Geophys. Res. Lett., 32, L21709, https://doi.org/10.1029/2005GL023900.
Nnamchi, H. C., J. Li, F. Kucharski, I.-S. Kang, N. S. Keenlyside, P. Chang, and R. Farneti, 2016: An equatorial–extratropical dipole structure of the Atlantic Niño. J. Climate, 29, 7295–7311, https://doi.org/10.1175/JCLI-D-15-0894.1.
Nnamchi, H. C., M. Latif, N. S. Keenlyside, J. Kjellsson, and I. Richter, 2021: Diabatic heating governs the seasonality of the Atlantic Niño. Nat. Commun., 12, 376, https://doi.org/10.1038/s41467-020-20452-1.
Richter, I., and S.-P. Xie, 2008: On the origin of equatorial Atlantic biases in coupled general circulation models. Climate Dyn., 31, 587–598, https://doi.org/10.1007/s00382-008-0364-z.
Richter, I., and H. Tokinaga, 2020: An overview of the performance of CMIP6 models in the tropical Atlantic: Mean state, variability, and remote impacts. Climate Dyn., 55, 2579–2601, https://doi.org/10.1007/s00382-020-05409-w.
Richter, I., and H. Tokinaga, 2021: The Atlantic zonal mode: Dynamics, thermodynamics, and teleconnections. Tropical and ExtraTropical Air–Sea Interactions, S. K. Behera, Ed., Elsevier, 171–206.
Richter, I., S. K. Behera, Y. Masumoto, B. Taguchi, N. Komori, and T. Yamagata, 2010: On the triggering of Benguela Niños: Remote equatorial versus local influences. Geophys. Res. Lett., 37, L20604, https://doi.org/10.1029/2010GL044461.
Richter, I., S. K. Behera, Y. Masumoto, B. Taguchi, H. Sasaki, and T. Yamagata, 2013: Multiple causes of interannual sea surface temperature variability in the equatorial Atlantic Ocean. Nat. Geosci., 6, 43–47, https://doi.org/10.1038/ngeo1660.
Richter, I., S.-P. Xie, S. K. Behera, T. Doi, and Y. Masumoto, 2014: Equatorial Atlantic variability and its relation to mean state biases in CMIP5. Climate Dyn., 42, 171–188, https://doi.org/10.1007/s00382-012-1624-5.
Richter, I., S.-P. Xie, Y. Morioka, T. Doi, B. Taguchi, and S. Behera, 2017: Phase locking of equatorial Atlantic variability through the seasonal migration of the ITCZ. Climate Dyn., 48, 3615–3629, https://doi.org/10.1007/s00382-016-3289-y.
Richter, I., T. Doi, S. K. Behera, and N. Keenlyside, 2018: On the link between mean state biases and prediction skill in the tropics: An atmospheric perspective. Climate Dyn., 50, 3355–3374, https://doi.org/10.1007/s00382-017-3809-4.
Richter, I., H. Tokinaga, Y. Kosaka, T. Doi, and T. Kataoka, 2021: Revisiting the tropical Atlantic influence on El Niño–Southern Oscillation. J. Climate, 34, 8533–8548, https://doi.org/10.1175/JCLI-D-21-0088.1.
Richter, I., Y. Kosaka, S. Kido, and H. Tokinaga, 2023: The tropical Atlantic as a negative feedback on ENSO. Climate Dyn., 61, 309–327, https://doi.org/10.1007/s00382-022-06582-w.
Richter, I., N. Keenlyside, T. Tozuka, Y. Okumura, C. Wang, P. Chang, S. Kido, and H. Tokinaga, 2024: Comment on “Resolving the tropical Pacific/Atlantic interaction conundrum” by Feng Jiang et al. (2023). Geophys. Res. Lett., 51, e2024GL111563, https://doi.org/10.1029/2024GL111563.
Rodrigues, R. R., E. J. D. Campos, and R. Haarsma, 2015: The impact of ENSO on the South Atlantic subtropical dipole mode. J. Climate, 28, 2691–2705, https://doi.org/10.1175/JCLI-D-14-00483.1.
Rodríguez-Fonseca, B., I. Polo, J. García-Serrano, T. Losada, E. Mohino, C. R. Mechoso, and F. Kucharski, 2009: Are Atlantic Niños enhancing Pacific ENSO events in recent decades? Geophys. Res. Lett., 36, L20705, https://doi.org/10.1029/2009GL040048.
Saravanan, R., and P. Chang, 2000: Interaction between tropical Atlantic variability and El Niño–Southern Oscillation. J. Climate, 13, 2177–2194, https://doi.org/10.1175/1520-0442(2000)013<2177:IBTAVA>2.0.CO;2.
Sasaki, W., T. Doi, K. J. Richards, and Y. Masumoto, 2015: The influence of ENSO on the equatorial Atlantic precipitation through the Walker circulation in a CGCM. Climate Dyn., 44, 191–202, https://doi.org/10.1007/s00382-014-2133-5.
Stine, R. A., 1995: Graphical interpretation of variance inflation factors. Amer. Stat., 49, 53–56, https://doi.org/10.1080/00031305.1995.10476113.
Tokinaga, H., I. Richter, and Y. Kosaka, 2019: ENSO influence on the Atlantic Niño, revisited: Multi-year versus single-year ENSO events. J. Climate, 32, 4585–4600, https://doi.org/10.1175/JCLI-D-18-0683.1.
Tozuka, T., T. Doi, T. Miyasaka, N. Keenlyside, and T. Yamagata, 2011: Key factors in simulating the equatorial Atlantic zonal sea surface temperature gradient in a coupled general circulation model. J. Geophys. Res., 116, C06010, https://doi.org/10.1029/2010JC006717.
Venegas, S. A., L. A. Mysak, and D. N. Straub, 1996: Evidence for interannual and interdecadal climate variability in the South Atlantic. Geophys. Res. Lett., 23, 2673–2676, https://doi.org/10.1029/96GL02373.
Venegas, S. A., L. A. Mysak, and D. N. Straub, 1997: Atmosphere–ocean coupled variability in the South Atlantic. J. Climate, 10, 2904–2920, https://doi.org/10.1175/1520-0442(1997)010<2904:AOCVIT>2.0.CO;2.
Wallace, J. M., and D. S. Gutzler, 1981: Teleconnections in the geopotential height field during the Northern Hemisphere winter. Mon. Wea. Rev., 109, 784–812, https://doi.org/10.1175/1520-0493(1981)109<0784:TITGHF>2.0.CO;2.
Wang, C., 2002: Atmospheric circulation cells associated with the El Niño–Southern Oscillation. J. Climate, 15, 399–419, https://doi.org/10.1175/1520-0442(2002)015<0399:ACCAWT>2.0.CO;2.
Zhang, W., F. Jiang, M. F. Stuecker, F.-F. Jin, and A. Timmermann, 2021: Spurious North Tropical Atlantic precursors to El Niño. Nat. Commun., 12, 3096, https://doi.org/10.1038/s41467-021-23411-6.
