Continuation of tropical Pacific Ocean temperature trend may weaken extreme El Niño and its linkage to the Southern Annular Mode

Observational records show that occurrences of the negative polarity of the Southern Annular Mode (low SAM) is significantly linked to El Niño during austral spring and summer, potentially providing long-lead predictability of the SAM and its associated surface climate conditions. In this study, we explore how this linkage may change under a scenario of a continuation of the ocean temperature trends that have been observed over the past 60 years, which are plausibly forced by increasing greenhouse gas concentrations. We generated coupled model seasonal forecasts for three recent extreme El Niño events by initialising the forecasts with observed ocean anomalies of 1 September 1982, 1997 and 2015 added into (1) the current ocean mean state and into (2) the ocean mean state updated to include double the recent ocean temperature trends. We show that the strength of extreme El Niño is reduced with the warmer ocean mean state as a result of reduced thermocline feedback and weakened rainfall-wind-sea surface temperature coupling over the tropical eastern Pacific. The El Niño-low SAM relationship also weakens, implying the possibility of reduced long-lead predictability of the SAM and associated surface climate impacts in the future.

The Southern Annular Mode (SAM) is the leading mode of variability of the Southern Hemisphere (SH) extratropical circulation on weekly and longer timescales that describes a meridional vacillation of the eddy-driven jet and associated storm track [1][2][3][4][5] . The spatial pattern of the positive polarity of SAM (high SAM) is characterised by a nearly zonally symmetric annular pattern of positive anomalies of pressure/geopotential height in the SH midlatitudes and negative anomalies in the Antarctic region, associated with a poleward shift of the eddy-driven jet and storm track 3 . The negative polarity of SAM (low SAM) is characterised by an equatorward shift of the eddy-driven jet and associated storm track. SAM, which is intrinsic to the troposphere, exhibits a decorrelation time of about two weeks 6 and can be realistically simulated in an atmospheric general circulation model forced with climatological sea surface temperatures (SSTs) at the lower boundary 7,8 . In addition to the well established trend toward the higher polarity phase of SAM in response to the Antarctic ozone depletion, year-to-year variations of seasonal mean SAM during spring and summer can be promoted by the El Niño-Southern Oscillation (ENSO) 4,[9][10][11][12][13][14][15] and by downward coupling of anomalous conditions in the Antarctic stratospheric polar vortex that develops as early as austral winter [16][17][18] .
The relationship between ENSO and SAM is particularly important for a long-lead prediction of SAM and its surface climate anomalies because ENSO can be skillfully predictable at lead times of 2-3 seasons and beyond 4,19 . During austral spring and summer, between 10-36% of the variance of SAM is explained by its relationship with eastern Pacific-type (EP) ENSO 4,9,10,13 . The warm phase of ENSO (El Niño) is associated with low SAM and the cold phase of ENSO (La Niña) is associated with high SAM. Seasonal mean SAM during austral spring and early summer is predictable with an atmosphere-ocean coupled forecast system at up to 6 month lead time, and the long-lead predictive skill is likely to be attributed to the relationship of SAM with eastern Pacific ENSO 4 . . The stippling indicates statistical significance of the trends at the 5% level (i.e. p < 0. 05), which was assessed by a two-sided Student t-test with 55 samples. Calculation of the SST and subsurface temperature trends over 54 years 55 times, leaving a year out each time (i.e. crossvalidation) confirms the robustness of the trend patterns (not shown).
Niño-like mean state change patterns 42 . Consequently, a significant increase in the frequency of extreme El Niño is projected in the 21 st century compared to the 20 th century 28 . However, the fidelity of this El Niño-like simulated mean state change in response to greenhouse gas forcing and its potential impact on ENSO characteristics have been vigorously debated 36,39,[47][48][49][50][51] . The El Niño-like mean state change in some models has been attributed to systematic model biases such as a too regular and symmetric El Niño-La Niña cycle together with ocean mixed layers that are too deep in the tropical Pacific Ocean 51,52 ; too weak inter-basin teleconnections in the tropical oceans 53 ; and a too-cold cold tongue in the equatorial Pacific with too high relative humidity and too low wind speed 39 . On the other hand, the ocean dynamical thermostat mechanism 40,54 , the non-linear ENSO warming suppression mechanism 55 , and/or the inter-basin warming contrast mechanisms 56,57 suggest that equatorial eastern Pacific SST would warm more slowly than the equatorial western Pacific SST as the Earth warms. Therefore, the possibility cannot be ruled out that the observed La Niña-like ocean temperature trend has been in part forced by global warming 39,56,58 , and it may continue into the future. Thus, we suggest that it is a valid and valuable question to address how the characteristics and teleconnection of extreme El Niño will change in the future if the observed La Niña-like long-term mean state warming continues, a scenario which has not yet been given much attention.
We address this premise by conducting four forecast sensitivity experiments whereby the ocean initial conditions are altered in a manner following the approaches of earlier studies using POAMA 59,60 . These experiments are summarised in Table 1. The present climate El Niño experiment (pElNiño) was conducted by initialising the coupled forecast model POAMA with observed ocean initial conditions for 00 UTC 1 September of 1982, 1997 and 2015 that represent the developing stages of the three extreme El Niños observed in the modern instrumental record 43 . We also initialised POAMA with the mean ocean conditions for 00 UTC 1 September computed over 1981-2013 in order to produce a set of climatological forecasts for the present climate (the present climatology experiment; pClim). To explore changes to extreme El Niño events and their linkages to SAM in the climate warmed up by enhanced observed ocean temperature trends, we computed the trends in ocean temperatures and salinity at all available vertical levels, latitudes, and longitudes on 00 UTC 1 September for the period of 1960-2014. We then doubled their magnitudes in order to produce a stronger impact in our modeling framework ( Supplementary Fig. S2). We added these doubled trends to the observed ocean initial conditions used in pElNiño to generate forecasts of El Niño that occur with a warmer ocean mean state (the warmer climate El Niño experiment; wElNiño). Finally, we generated the climatological forecasts of the warmer climate by adding the doubled ocean temperature and salinity trends to the climatological ocean initial conditions used in pClim. We refer to this as the warmer climatology experiment, wClim. We denote present and warmer climate predicted El Niño with respect to their respective present and warmer climatologies as pElNiño' and wElNiño' , respectively.
The atmosphere and land component models of POAMA were initialised with 33 different conditions drawn from observed states during 1981-2013 in order to scramble atmosphere and land initial conditions within the observed range so to generate ensemble forecasts. The CO 2 concentration was fixed to 345 ppm, which is the default value of POAMA retrospective forecast set, and the ozone concentration was prescribed by the observed monthly climatology 34 . Therefore, differences in the predicted atmospheric circulation in our experiments can be interpreted as responses to the differences in ocean forecasts in the experiments. We initialised all experiments on 1 September and limited our interest to the three month mean forecasts for October to December (OND) because ENSO and SAM are better correlated on a seasonal time scale than on shorter time scales and the three month mean observed correlation is maximum in OND 4 .

Revisiting the observed Relationship Between enSo and SAM
Austral spring-summer SAM is influenced by both ENSO and Antarctic stratospheric vortex variation, the latter of which is considered to be a stronger driver 16 . SAM has also shown a strong positive trend from the 1980s to the late 1990s due to the anthropogenically-driven Antarctic ozone depletion 61 . To highlight the relationship of SAM with ENSO, which is largely an interannual variation, we de-trended the SAM index for the period of 1979-2016. Then, we regressed out the component of SAM related to the stratospheric polar vortex variation from the de-trended SAM index, using the SH stratosphere-troposphere coupled mode index 18 as a predictor representing anomalous stratospheric conditions (i.e. de-trended residual SAM; see Methods for the climate indices used in the study). Earlier studies 18,62 reported no significant relationship between the Niño3.4 SST index, which represents eastern Pacific ENSO variability, and the Antarctic stratospheric vortex variation during the last 40 years, but SAM is significantly better correlated with Niño3.4 SST after the removal of the influence of the SH stratospheric polar vortex variation from the SAM index (r = −0.53, statistically significant at the 0.1% level (i.e. p < 0.001); Fig. 2c). The three extreme El Niño events of 1982, 1997 and 2015, which were accompanied by a substantial spread in the magnitude of raw SAM (e.g. SAM was neutral during the late spring of the strong 2015 El Niño), all exhibited strong low SAM after the removal of the stratospheric influence. This result supports the connection between ENSO and SAM in OND. Figure 3 shows the SST anomalies of the three extreme El Niño events and de-trended residual MSLP anomalies of OND. Although there are differences in the spatial details and magnitudes of anomalies of SSTs, all three El Niño events are associated with low pressure anomalies equatorward of 60°S with a wavenumber 3 structure and high pressure anomalies poleward of 60°S, which typify low SAM. Strong stationary Rossby wave propagation from the Maritime Continent poleward and eastward to the Amundsen-Bellingshausen Seas is a distinctive teleconnection driven by El Niño, which is likely to further contribute to the amplitude of low SAM by placing a strong low pressure anomaly centre north of 60°S and a strong high pressure anomaly centre slightly south of 60°S. As noted in Fig. 2a, SAM was observed to be neutral despite the extreme strength of El Niño in OND 2015 because anomalous Antarctic stratospheric vortex strengthening 18 and associated ozone depletion were acting to produce high SAM (https://ozonewatch.gsfc.nasa.gov/meteorology/figures/merra/ozone/toms_areas_2015_ omi+merra.pdf). After removing this influence of the stratospheric polar vortex, the residual SAM is strongly negative with zonally symmetric pressure anomalies like those of 1982 and 1997 ( Fig. 3 right panels). This implies that in 2015 the forcing of high SAM from a strengthening of the polar stratospheric vortex and associated depletion of Antarctic ozone may have completely countered the forcing of low SAM from El Niño in the OND season, thereby lowering the predictability of SAM and the associated surface climate.
The promotion of low SAM by El Niño has previously been shown to result from intensification and equatorward contraction of the the Hadley circulation because the El Niño SST anomalies act to increase near equatorial diabatic heating (latent heat release as a result of moist atmospheric convection), which is flanked by enhanced diabatic cooling (longwave radiation to space as a result of enhanced subsidence and reduced latent heat release) in the subtropics (Fig. 4a). Thus, westerlies on the equatorward side of the climatological subtropical jet strengthen (Fig. 4b). This increase in westerly winds closer to the equator allows deeper penetration of extratropical baroclinic Rossby waves into the tropics 11,12 . As a result, eddy momentum flux divergence in the tropics shifts closer to the equator, and momentum flux convergence anomalously increases in the midlatitudes (30-40°S), while decreasing in the higher latitudes (50-70°S; Fig. 4c). Extratropical baroclinicity and the associated storm track thus shift equatorward, which is manifest as low SAM. These general features resulting from the eastern Pacific SST variations during El Niño, which were proposed in the literature, are confirmed by the composites of the three extreme El Niños in  www.nature.com/scientificreports www.nature.com/scientificreports/ eastward to the south eastern Pacific, which are consistent with the observed anomalies shown in Fig. 3. However, a strong low-pressure center observed over the southern Atlantic Ocean is missed in all three of the simulated El Niño cases ( Fig. 5 right panels). Reproducibility of low SAM in pElNiño' , which is initialised with realistic ocean conditions but random atmosphere and land conditions, confirms that strong El Niño can be an important forcing of low SAM, and the polarity of SAM would have been more negative during El Niño of 2015 if the Antarctic polar vortex were not significantly stronger than usual.
Warmer climate mean state. For the results of the forecasts of extreme El Niño in the idealised future warmer climate (wElNiño) to be scientifically reliable and attributable to the change in the mean state of the initial conditions, key features of the observed ocean temperature trends added in the initial conditions and resultant ocean circulation changes should be faithfully maintained through the October to December verification period. This is demonstrated by comparing mean differences in SSTs, equatorial Pacific upper ocean temperatures and tropical Pacific mixed layer circulations between wClim and pClim to the observed trends in OND over 1960-2014 in Fig. 6. We note that although we added doubled the observed ocean temperature trends to the climatological ocean initial conditions for wClim, the magnitudes of the ocean surface and subsurface warming over the following four months are more or less comparable to the magnitudes of the observed trends of the 55 years of 1960-2014. This loss of amplitude in the mean state trend may be partly due to the scrambled atmosphere initial conditions and their damping effect until the atmosphere and the ocean are brought to a balance in the first month of forecasts 24 , or the lack of changing the greenhouse gas forcing in the warmer climate runs. Nevertheless, it is encouraging to see the overall similarities between the warmer minus present climate differences and the observed trends for the spatial structures of SSTs, equatorial Pacific subsurface temperatures and tropical Pacific mixed layer ocean circulations. For instance, the pattern correlation between Figs. 6a,b is 0.55 (p < 0.001 with 5760 grid values), and the enhanced zonal SST gradient between the tropical western Pacific (120-160°E, 5°S-5°N) 40 and eastern Pacific (Niño3.4 region) is maintained in OND forecasts in wClim minus pClim although the gradient is simulated with only about one third of the observed magnitude (0.3 °C compared to 0.9 °C). The observed reduction of upwelling into the mixed layer (averaged 0-45 m depth) in the equatorial eastern Pacific is also faithfully captured by wClim forecasts (Figs. 6e,f). On the other hand, forecasts of SSTs and subsurface temperatures are significantly different from the observed trends in the far eastern Pacific east of 120°W, which should be borne in mind when we diagnose below how the change in the mean state acts to change El Niño growth.
The atmospheric response to the warmer climate (Fig. 7a) shows more intense warming in the upper troposphere in the tropics than in the higher latitudes of the SH, steepening the meridional temperature gradient between the tropics and the south pole. This change results in a poleward shift of the eddy-driven jet with anomalously increased eddy momentum flux convergence around 60°S but decreased eddy momentum flux www.nature.com/scientificreports www.nature.com/scientificreports/ convergence around 45°S (Figs. 7b,c). The associated MSLP change projects onto high SAM (standardized SAM index = 0.9), although the high-pressure anomalies in the midlatitudes are not as annular as that of conventional SAM (Fig. 7d). These atmospheric changes are similar to those projected by climate models in response to increasing greenhouse gases 63 despite the contrasting tropical Pacific zonal SST gradients between wClim and climate change simulations. The similarity in the atmospheric responses despite differences in the degree of warming in the tropical eastern SSTs is because equatorial waves quickly spread anomalous temperature change across the entire tropics 59 .  www.nature.com/scientificreports www.nature.com/scientificreports/ el niño on the warmer ocean mean state. We now turn to the warmer climate El Niño compared to the present climate El Niño. Because forecasts of the SST and MSLP patterns of the three El Niño years in the present climate appear to be all very similar to one another, we will present results of a grand ensemble of 99 members by combining the anomalies for the 1982, 1997 and 2015 simulations.
When El Niño of the same extreme strength occurs on the "La Niña-like" warmer ocean mean state (wElNiño' = wElNiño-wClim), its strength is simulated to be weaker than that in the current climate (pEl-Niño' = pElNiño-pClim) over the tropical eastern Pacific (p < 0.05, see Methods for statistical significance calculation; Fig. 8a, Supplementary Fig. S4). Therefore, the maximum warming of El Niño in the warmer mean state appears to be more confined to the central Pacific, giving it some central Pacific flavour. Analysis of advective feedback terms in the ocean mixed layer heat budget (averaged over 0-45 m, 5°S-5°N; see Methods) suggests that the reduced mean upwelling in the eastern Pacific in the warmer ocean mean state (Fig. 6f) is partly responsible for the weaker growth of eastern Pacific SST anomalies in the warmer climate (Fig. 8b). Across the central to western Pacific, the reduced strength of mean westward surface currents in the warmer ocean mean state (Fig. 6f) also contributes to weaker growth of warm anomalies over and west of the dateline (Fig. 8c), while weaker anomalous eastward currents during El Niño in the warmer climate contributes to weaker warming over the central Pacific (Fig. 8d). The cause of the weaker anomalous eastward currents during El Niño in the warmer climate is addressed below. In contrast, the enhanced mean zonal temperature gradient of the warmer ocean mean state over the equatorial Pacific (e.g. Figs. 6b,d) positively contributes to the growth of positive SSTs during El Niño over the central to western Pacific (Fig. 8e), thereby compensating for some of the negative contributions caused by the weakening of the mean westward zonal currents and the weaker El Niño-generated anomalous eastward currents there.  Figures 9a-c show that El Niño-driven rainfall in the warmer climate is shifted westward relative to that in the present climate, and over the Niño3.4 region the rainfall anomaly is weaker (p < 0.05). This likely occurs due to the cooler equatorial mean SSTs (Fig. 6b) and weaker El Niño SST anomalies caused by the reduced thermocline feedback in the eastern Pacific (Figs. 8a,b), which makes a SST anomaly less efficient at driving a rainfall/wind response 60,64,65 . In contrast, there is an increased rainfall response over and west of the dateline during wElNiño' compared to pElNiño' likely boosted by the warmer SST mean state of the tropical central to western Pacific, shifting the location of the maximum rainfall associated with the Niño3.4 SSTs about ~10° westward (p < 0.05). Likewise, SST anomalies over the Niño3.4 region induce weaker westerly response over the eastern Pacific but stronger westerly response west of the dateline with the warmer mean state compared to the present mean state (p < 0.05; Fig. 9d-f). Together these westward shifts of the zonal wind and rainfall responses during El Niño on the warmer mean state feed back into producing a weaker El Niño in the eastern Pacific.
Interestingly, these 1) weakened thermocline feedback and 2)air-sea coupling strength and resultant weakened El Niño with its maximum SST warming and rainfall response concentrated in the central Pacific are consistent with what has been diagnosed to have occurred during the early 2000s as compared to the 1980s and 1990s and www.nature.com/scientificreports www.nature.com/scientificreports/ is thought to reflect the shift to the cold phase of the IPO 60,64-67 . The cold phase of the IPO is also characterised by an enhanced zonal SST gradient across the tropical Pacific and shallower thermocline in the equatorial eastern Pacific, but there are many differences between the cold phase of the IPO and our warmer ocean mean state ( Supplementary Fig. S1). Hence, the consistency in the changes in the ocean and atmosphere feedback and El Niño properties between our warmer ocean mean state and the cold phase of IPO highlight the role of the equatorial Pacific mean state in modulating the amplitude of El Niño and the longitude of maximum warming. el niño and SAM on the warmer ocean mean state. Although El Niño weakens on the warmer ocean mean state, it still promotes low SAM. However, the El Niño-low SAM connection is substantially weaker in the warmer climate compared to the current climate as judged by standardized SAM being −0.7 as derived from the pressure anomalies of Fig. 10d compared to standardized SAM being −2.1 as derived from those of Fig. 10c (this weakening of the SAM strength is statistically significant at the 1% level). The Rossby wave train emanating from the Maritime Continent is also weaker in the warmer climate. This weakening of teleconnection is likely due to both weakening of El Niño and weakening of the tropical convective heating response to El Niño on the warmer www.nature.com/scientificreports www.nature.com/scientificreports/ mean state. As shown in Figs. 10a,b, wElNiño' induces less intense warming over the tropical upper troposphere, and therefore, the meridional temperature gradient between the tropics and the SH midlatitudes is not as steep as that caused by pElNiño' , resulting in weakened westerly anomalies associated with wElNiño' compared to those with pElNiño's (Figs. 10e,f). This reduced change in the mean winds in the upper troposphere leads to reduced changes in eddy momentum flux convergence anomalies in the subtropics to the high latitudes of the SH (Fig. 10g,h); which feed less of anomalous momentum back to the mean flow, resulting in weaker dipole wind anomalies in the SH extratropics, leading to weaker low SAM. The strengths of the extratropical zonal-mean zonal wind dipole and the eddy momentum flux convergence dipole in the upper troposphere are statistically significantly different between pElNiño' and wElNiño' at the 5% level ( Supplementary Fig. S8). Consequently, predictability of SAM in OND would be substantially reduced during El Niño under our scenario of continuation of the observed ocean temperature trends to the future.

concluding Remarks
This study was motivated by earlier research findings that ENSO is an important source of predictability of SH extratropical climate through its connection to SAM during austral spring and summer; yet this ENSO-SAM relationship is not skillfully simulated by climate models, therefore making it hard to foresee any possible change in this relationship in a future warmer climate. Furthermore, climate models have not been able to reach a strong consensus on how ENSO amplitude will change 43 , which would be a key determinant in predictability of extratropical climate via its teleconnections. The strength of ENSO and its spatial characteristics influence and are influenced by the tropical ocean mean state, especially the tropical Pacific Ocean mean state 60,66 . There is still vigorous debate, though, about whether the response of the tropical Pacific to global warming will be more "La Niña-like" with greater warming over the tropical western Pacific than over the eastern Pacific, which is what has been observed over the past 60 years 39 , or more "El Niño-like" with the opposite pattern of warming, which is projected by the majority of the CMIP5 models 47 .
In this study, we limited our focus to El Niño and attempted to address this question: How will extreme El Niño and its relationship with SAM change in a warmer climate if the "La Niña-like" ocean temperature trends that have occurred over the past 60 years continue into the future? To do so, we conducted a series of forecast sensitivity experiments for three extreme El Niños (1982, 1997 and 2015). The forecast simulations were initialised on 1 September using the present ocean mean state and using a hypothetical future ocean mean state that was created by addition of the doubled observed ocean temperature trends over the past 60 years (we have referred to this as the warmer ocean mean state). Present and warmer climatological forecasts were also produced with the climatological conditions of 1 September of 1981-2013 without and with the addition of the doubled observed ocean temperature trends, respectively, as reference forecast sets. www.nature.com/scientificreports www.nature.com/scientificreports/ Statistical analysis of observations and the forecast experiments confirmed that extreme El Niño is a key driver of strong low SAM in the October to December season. The observed El Niño-low SAM relationship is robust once we remove the influence of the Antarctic stratospheric vortex on the SAM by regressing it out in the observational analysis. This conclusion is also supported by the forecast experiments that use scrambled atmospheric initial conditions so that the promotion of low SAM during the El Niño forecasts can be attributed to the El Niño forcing.
The El Niño experiments with the present and the hypothetical warmer ocean mean states revealed that extreme El Niño is likely to lose some strength particularly over the tropical eastern Pacific as a result of reduced mean upper ocean upwelling on the warmer ocean mean state which causes a reduction in the thermocline feedback. In our experiments, the increased zonal temperature gradient of the observed La Niña-like upper ocean mean state change contributes to anomalous SST warming associated with extreme El Niño, as highlighted by the recent study of Wang et al. 68 However, this enhanced SST growth appears to be limited to the central and western Pacific 78 and is offset by weaker anomaly growth caused by the reduced strength of mean westward surface currents in the same region. The air-sea coupling strength appears to also significantly weaken in the eastern Pacific due to the cooler equatorial mean SST and weaker SST anomalies associated El Niño and shift westward in the warmer climate, further contributing to the weakening of extreme El Niño in the eastern Pacific. This delicate balance of processes in our model framework tips the scale in favor of weaker extreme El Niño events in a warmer world with the continuation of the observed long-term ocean temperature trends.
The weakened El Niño in the warmer climate appears to result in a significant weakening of the low SAM response. Our present climate El Niño experiment confirmed the findings of earlier studies 10-12 that tropical upper tropospheric warming caused by El Niño increases westerlies on the equatorward side of the SH subtropical jet, shifting the critical latitude equatorward. This shift induces an equatorward shift of the momentum flux convergence-divergence dipole, resulting in low SAM. This chain of processes appears to continue to operate in the warmer climate but with significantly weaker strength of zonal wind, eddy momentum transport and pressure anomalies because El Niño itself becomes weaker and the tropical convective heating response to El Niño becomes weaker as well. Therefore, if the ocean temperature trends that have been observed over the past 60 years continue into the future, predictability of low SAM and its associated impacts on SH surface climate during extreme El Niño is likely to be substantially reduced. A previous study 60 indicated that La Niña may also be expected to weaken in response to a "La Niña-like" temperature trend, but a further study is required to assess whether there will be a similar reduction in predictability of high SAM. An idealised experiment like ours but to examine the ENSO-SAM relationship with El Nino-like warming trends in the tropical Pacific, as suggested by most CMIP models, would also be valuable in understanding the mechanisms of the resultant changes in ENSO and SAM.

Methods
Data for observational analyses. We used reanalysis data of the European Centre for Medium-Range Weather Forecasts Interim project (ERA-Interim) 69 and SST analyses of Hurrell et al. 70  climate indices. The SAM index was obtained by following Gong and Wang's definition 73 , which is the difference between normalised zonally averaged MSLP anomalies at 40°S and 65°S. The strength of extreme El Niños was determined by the Niño3.4 index, which was obtained by averaging SSTs over the domain of 5°S-5°N, 190-240°E. The stratosphere-troposphere (S-T) coupled mode index was obtained following the method of Lim et al. 18 by applying height-time domain EOF analysis to anomalies (seasonal cycle removed) of monthly mean zonal-mean zonal winds averaged over 55-65°S. The input data to the EOF was ordered from April to March each year for pressure level data extending from 1000 to 1 hPa. The resultant 1 st principal component time series consists of one value each year and depicts the year-to-year variations of the SH spring polar vortex strength and its downward coupling. To obtain the de-trended residual SAM independent of the influence of the Antarctic stratospheric polar vortex, we removed from the SAM index the components linearly related to time and the S-T coupled mode index variability.

Model and initial conditions for forecast sensitivity experiments.
For the model experiments, we used the Bureau of Meteorology's atmosphere-ocean fully coupled dynamical seasonal climate forecast system, POAMA version a 34 . Its atmosphere and ocean component models are the Bureau's Atmosphere Model version 3 (T47/L17) 74 , and the Australian Community Ocean Model version 2 (2° longitude by 0.5-1° latitude from the tropics to the pole) 75 , which are coupled by OASIS 76 coupler.
High quality observed atmosphere, land and ocean conditions are generated from the Bureau of Meteorology's atmosphere and land initialisation scheme (ALI) 77 and PEODAS, respectively.
Statistical significance tests. Statistical significance of the observed trends in Figs 1 and 6 was assessed by a two-sided Student t-test with 55 samples (i.e. data of 1960-2014), using the incomplete beta function available in NCAR Command Language. In Fig. 4, 1 standard deviation threshold was used to indicate if air temperature, zonal wind and eddy momentum flux convergence anomalies associated with extreme El Niño are significantly different from the climatological conditions in the ERA-Interim reanalysis set. A two-sided Student t-test was used to estimate the statistical significance on the difference of the two means (Figs 8, 9a,b,d,e, 10) and of the two regression coefficients (Fig. 9c,f) with 99 samples for p'ElNiño vs. w'ElNiño and and the difference of the two means with 33 samples for pClim vs wClim (Figs 6b,d,f and 7). (2019) 9:17044 | https://doi.org/10.1038/s41598-019-53371-3 www.nature.com/scientificreports www.nature.com/scientificreports/ ocean mixed layer heat advection analysis. Based on the ocean mixed layer heat budget and heat advection analyses of Zhao et al. 60 and Abellan et al. 78 , we computed contributions of advective feedback terms to the change of El Niño growth as follows: where u denotes zonal, meridional and vertical velocities, and T x , T y and T z indicate zonal (x), meridional (y) and vertical (z) temperature gradients, respectively. Overbar and prime signs denote a temporal mean and a departure from the mean, respectively. From Eq. (2), we further limited our interest to linear terms to tease out the contributions from the mean state changes vs the anomalous changes to changes in the growth of El Niño. To estimate the contributions of the mean state changes, we used the mean state terms u, v, w , T x , T y , and T z from the two different climatologies, pClim and wClim, together with the anomalies from the present climate u′, v′, w′, T′ x , T′ y and T′ z . Then, to estimate the contributions of the anomalous changes in El Niño, we used the anomalies u′, v′, w′, T′ x , T′ y and T′ z of El Niño for the two different mean states, wElNiño' and pElNiño' , together with the mean state ′ u , ′ v , ′ w , T x ′ , ′ T y , and T z ′ from the present climatology (pClim). Figure 8 and Supplementary Figs. S5-7 show that the most important term that explains the reduction of El Niño strength in the tropical eastern Pacific is the thermocline feedback change by the reduced mean upwelling on the warmer ocean, wT z ′ , in Eq. (2). The reduced mean westward currents of the warmer mean state in ′ uT x and the reduced anomalous eastward current of warmer climate El Niño in ′ u T x also suggest some contributions to the reduction of SST warming associated with warmer climate El Niño over the tropical central Pacific. However, as the enhanced zonal temperature gradient of the warmer mean state in ′ u T x somewhat compensates the temperature reduction, an overall contribution from the zonal advective feedback change to the change in the SSTs of warmer climate El Niño seems small.