Main

The Southern Ocean (SO) plays an outsized role in absorbing excess heat from global warming1,2,3,4,5, accounting for some 70% of greenhouse warming-induced extra heating6,7 and contributing to 60–90% of heat absorption by the global oceans8,9. Observed changes in zonal average SO temperatures are dominated by a region of warming centred near 45° S and extending from the surface to below 1,000 m depth1,2,3,4,5. At mid-depths and within the latitudes of the Antarctic Circumpolar Current, the warming has proceeded at approximately twice the rate of global upper ocean warming1. Southern Ocean warming is projected to continue given the persistent increase in greenhouse gas emissions8,10,11 with expected far-reaching global impacts, for example, global sea level rise12,13, inter-hemispheric moisture transports and associated rainfall patterns14,15, enhanced warming in the Arctic16,17 and tropical climate18,19,20,21,22,23,24,25,26,27,28,29. Therefore, understanding the processes that influence SO heat uptake is vital to reducing uncertainty in the rate of future SO warming and associated impacts30.

The physical processes behind SO heat uptake involve equatorward Ekman transport and the associated Ekman convergence driven by high-latitude westerly winds4,5,31,32. These winds drive upward-inclined density surfaces towards the pole, with a compensating impact from mesoscale eddy fluxes. Strong circumpolar upwelling driven by the westerly winds brings deep water to the surface along sloped density surfaces south of the westerly wind maximum (~50° S), where the water becomes exposed to the atmosphere. The equatorward Ekman transports advect this high-latitude surface water to the north where subduction occurs along isopycnals within the mid-latitude deep-mixed layers centred around 45° S. Under global warming, to the south of the deep-mixed layers, warming is dispersed away primarily by the mean circulation, with a secondary role by upwelling of cooler subsurface waters;2,4,5,31,32 whereas to the north, mesoscale eddies play a primary role by flattening the density surface and driving a decrease in upward heat transport32,33.

Projected SO warming varies widely across models7,8,10,31,34 and is affected by multiple factors. For a given emissions scenario, these factors include model climate sensitivity35, response of high-latitude westerlies36 and the parameterization of mesoscale eddies and their response32,33. Further, SO warming is dependent on the relative importance of increasing concentrations of carbon dioxide in the atmosphere, offsetting ozone recovery37,38,39,40, and a reinforcing influence from a reduction in anthropogenic emissions of aerosols10. In addition to these factors, this study presents a systematic influence by the response of El Niño–Southern Oscillation (ENSO) amplitude to global warming.

Simulated impact of winds on SO warming

We use output from 27 climate models that participated in phase 6 of the Coupled Model Intercomparison Project (CMIP6). These models were forced with historical forcings until 2014 and with a high-emissions scenario from 2015 to 2100 (that is, shared socioeconomic pathway (SSP)5–8.5) (ref. 41) (Methods and Supplementary Table 1). Although there is noticeable difference in the warming centred around 65° S, simulated trends of zonal average ocean temperature resemble many of the observed features since 19482,3,4,5, including warming near 45° S that extends down to 1,000 m depth (Extended Data Fig. 1a versus Extended Data Fig. 1b).

For each model, we compare changes at and below the surface averaged over the twentieth century (1900–1999) and the twenty-first century (2000–2099). Under global warming, mid- to high-latitude westerly winds undergo a poleward intensification42,43,44 (Fig. 1a), with SO warming centred near 45° S (Fig. 1b). We use an index to measure SO warming, defined as the difference between the twentieth and twenty-first centuriesʼ mean temperatures spatially averaged over the upper 1,000 m of the ocean from 40° S–60° S (box in Fig. 1b). There is an inter-model relationship in which a greater climate sensitivity (Methods) is associated with greater SO warming (Extended Data Fig. 2a,b). The inter-model difference in climate sensitivity, which differs widely across models, accounts for about 50% (the square of the correlation coefficient) of the inter-model difference in the SO warming, with an inter-model correlation of 0.7 (Extended Data Fig. 2b). Because the value of climate sensitivity is not available for all models, we use the increase in global mean surface temperature (GMT) as its surrogate (Extended Data Fig. 2c). Under global warming, net heat flux into the ocean (positive downward) over the 50° S–70° S band is a heat source for the SO warming and increases with the GMT rise (Extended Data Fig. 3a–c), and the increase in atmospheric heat flux over the major heating latitude band 50° S–60° S is mainly contributed by the increasing sensible heat flux due to the warming atmosphere (Extended Data Fig. 4) (ref. 2).

Fig. 1: Projected wind changes, Southern Ocean warming, and their relationship with ENSO response.
figure 1

a, Multi-model ensemble mean of changes in zonal wind stress between the twenty-first and twentieth centuries scaled by global warming (N m−2 °C−1 of global warming) over 27 CMIP6 models. Stippling indicates where the difference between the two periods is statistically significant above the 90% confidence level based on a Student’s t-test. b, the same as in a, but for the ensemble mean of changes in zonal mean ocean subsurface temperatures (°C °C−1 of global warming). c, Inter-model regression pattern of projected changes in zonal wind stress onto the SO warming indices (N m−2 °C−1), that is, a regional average in the upper 1,000 m between 40° S and 60° S as indicated by the black box in b. Stippling indicates where the correlation is statistically significant above the 90% confidence level based on a Student’s t-test. d, An inter-model relationship between the SO warming index and projected changes in variability of the Niño3.4 index, both scaled by global warming. The purple dashed ellipse indicates 5–95% ranges. Correlation coefficient (Corre. coeff.), slope and P value are also indicated.

To examine processes that contribute to SO warming beyond model climate sensitivity, projected changes are scaled by the corresponding GMT increase, that is, changes per °C of global warming (Methods). Inter-model regression finds that a further poleward wind intensification, along with the associated northward Ekman transport, Ekman upwelling and downwelling to its south and north, respectively, is conducive to greater SO warming (Fig. 1c and Supplementary Fig. 1). This wind change pattern promotes high-latitude equatorward Ekman transport of heated water towards the subduction zone.

Sourced in the equatorial Pacific Ocean, and as Earth’s most dominant source of year-to-year climate variability, ENSO exerts reverberating impacts across the globe, including the Southern Hemisphere high-latitude winds45. ENSO amplitude in the majority of climate models is projected to intensify to varying degrees as a response to the transient increase in carbon dioxide emissions46,47,48,49,50. Below we show that an increase or a decrease in ENSO amplitude is respectively tied to a slowdown or acceleration of the SO warming.

Strong link between SO warming and ENSO response

We use ENSO peak season (December, January, February, [DJF]) Niño3.4 (5° S–5° N, 170° W–120° W) sea surface temperature (SST) index to depict ENSO. Seasonal SST anomalies are constructed in reference to the climatology over the 1900–1999 period and quadratically detrended. Most models simulate a reasonable spatial pattern (Extended Data Fig. 5a) and nonlinear properties of ENSO, for example, positively skewed SST anomalies in the eastern equatorial Pacific, which extends to the Niño3 region (5° S–5° N, 150° W–90° W; Extended Data Fig. 5b,c). The positive skewness means that the amplitude of warm El Niño anomalies is larger than cold anomalies of La Niña, principally governed by the nonlinear Bjerknes feedback51 by which, once El Niño warm anomalies establish atmospheric deep convection, they trigger a nonlinear wind response to further warm anomalies52,53. This process is facilitated and amplified by thermocline and zonal advective feedbacks involved in the Bjerknes feedback. Due to the El Niño–La Niña asymmetry, wind anomalies of El Niño and La Niña do not completely offset but rectify onto the mean state47,54,55.

We calculate changes in ENSO amplitude as the difference in Niño3.4 standard deviation between the twentieth and twenty-first centuries. There is a large inter-model spread in the response of ENSO amplitude that has no direct relationship with inter-model spread in GMT change, but a majority of the models generate an increase in magnitude (Extended Data Fig. 5d). The increased magnitude is underpinned by enhanced vertical temperature gradients along the equatorial Pacific as the upper equatorial Pacific Ocean warms faster than the ocean below46,50,56.

Surprisingly, there is a close inter-model relationship between changes in ENSO amplitude and SO warming indices beyond influences from inter-model spread in GMT change. A greater increase in ENSO amplitude is systematically associated with a lower SO warming (Fig. 1d), with a correlation coefficient of −0.74, statistically significant above the 99% confidence level. The inter-model differences in ENSO amplitude change are associated with ~50% of the inter-model differences in SO warming.

Although temperature changes over the SO region could impact tropical Pacific climate18,19,20,21,22,23,24,25,26,27,28,29, the relationship depicted in Fig. 1d instead suggests an impact of the tropical Pacific climate change on the SO. SO surface warming tends to correspond with increased SO heat uptake or Antarctic sea ice loss18,25,57. As the SO temperatures increase, poleward heat transport decreases, leading to a convergence of heat in the low latitudes and a faster surface warming in the eastern equatorial Pacific25. The faster warming in the eastern equatorial Pacific would lead to increased ENSO variability by facilitating atmospheric deep convection in the east, conducive to the development of El Niño events46,49. As such, a greater SO warming would be associated with an increase in ENSO variability, opposite to what we find in Fig. 1d. Below we show the mechanism by which ENSO response to global warming modulates SO heat uptake not through air–sea heat flux (Extended Data Fig. 3d,e), but through changes in southern high-latitude winds as a result of ENSO rectification.

Observed ENSO rectification on SO winds

ENSO influences extra-tropical and high-latitude atmospheric circulations in the Southern Hemisphere through two pathways. El Niño-induced anomalies in atmospheric deep convection excite equivalent barotropic Rossby wave trains that veer poleward, featuring a high-pressure anomaly over the Amundsen Sea region58, referred to as the Pacific–South American (PSA) pattern. The same equatorial convective heating also generates a response of the zonal mean atmosphere circulation59,60. These responses are reflected in regression patterns of surface zonal wind stress anomalies onto the Niño3.4 index, which shows the PSA teleconnection pattern (Fig. 2a) and anomalous easterlies south of approximately 45° S but anomalous westerlies to the north (Fig. 2b and Methods).

Fig. 2: Observed ENSO teleconnections to Southern Hemisphere atmospheric circulation.
figure 2

a, Regression pattern (in units of N m−2 °C−1) of surface zonal wind stress onto Niño3.4 index using (NCEP)/(NCAR) Reanalysis 1 (ref. 63) from 1948 to 2019 (Methods). b, Zonal averaged regression pattern of zonal wind stress as in a. c, the same as in a, but for anomalous meridional mass stream function (ψm) in colour (in units of kg s−1 °C−1), superimposed on its climatological distributions in black contours (in units of kg s−1) with 1.0×109, 5.0×109, 1.0×1010, 3.0×1010, 9.0×1010 and 13.0×1010 as contour levels for both negative (dashed lines) and positive (solid lines) values. Negative values indicate anticlockwise flows and vice versa. The surface wind direction due to Coriolis effect for the climatological Hadley cell, Ferrell cell and Polar cell is indicated by the grey circles. d, The sum of zonally averaged surface zonal wind stress anomalies over all El Niño (red curve) and all La Niña (blue curve) events. El Niño and La Niña events are defined as when the magnitude of the DJF averaged Niño3.4 index is greater than 0.75 °C. The black curve is the sum of the red and blue curves, that is, cumulative ENSO impact. To focus on inter-annual timescales, trends and decadal variability are removed first before analysis. Stippling in a and c indicates where the correlation is significant above the 90% confidence level based on a Student’s t-test.

This anomalous zonal mean circulation (Fig. 2b) is consistent with anomalies of atmosphere meridional mass stream function (Methods), in the backdrop of a climatological state that features the Hadley, Ferrell and Polar cells (contours, Fig. 2c). Associated with the climatological equatorward surface flows of the Polar cells are climatological easterlies. Associated with poleward surface flows of the Ferrell cells are climatological westerlies. During El Niño, the Hadley cells intensify and contract equatorward, accompanied by an anomalous equatorward movement of the Ferrell and Polar cells59,60 (colour, Fig. 2c). The equatorward shift induces the anomalous easterlies in high latitudes. The reverse occurs during La Niña. An anomalous teleconnection pattern in the zonal mean also operates on inter-decadal timescales and long-term trends61.

Owing to ENSO asymmetry46,47,49,52,53,54,55, anomalous circulations due to El Niño and La Niña are not symmetric in terms of latitudinal structure or intensity, leading to a rectification onto the mean state. In particular, given that El Niños tend to be stronger than La Niñas, accumulating zonal mean zonal wind anomalies for all observed warm (red curve, Fig. 2d) and cold (blue curve, Fig. 2d) events with |Niño3.4 | > 0.75 °C over the 1948–2019 period produces a residual ENSO impact (black curve, Fig. 2d) featuring high-latitude easterlies and hence decreased Ekman transports. These changes are robust across several reanalysis products (Extended Data Fig. 6). Through these ENSO teleconnections to Southern Hemisphere high-latitude zonal winds, the response of ENSO to global warming affects the projected high-latitude westerly poleward intensification, modulating SO warming.

ENSO response affects SO warming via winds

For each model, we calculate ENSO-induced zonal wind anomalies for the twentieth and twenty-first centuries, separately. We obtain ENSO teleconnection changes between the two centuries in the 27 models, each scaled by the corresponding GMT rise. We then apply empirical orthogonal function analysis62 to the 27 fields of ENSO teleconnection changes in zonal wind. The two leading empirical orthogonal function patterns reflect inter-model differences in the PSA response (Extended Data Fig. 7a) and in the more zonally symmetric response (Fig. 3a), respectively.

Fig. 3: Inter-model differences in ENSO-induced zonally symmetric anomalies.
figure 3

An empirical orthogonal function analysis is applied to 27 patterns of ENSO-induced Southern Hemisphere zonal wind stress changes between the twenty-first and twentieth centuries over the domain of (80° S–20° S, 360° E–W) from 27 CMIP6 models. a, The principal pattern (N m−2 °C−1 of global warming) describes inter-model differences in zonally symmetric ENSO teleconnections, explaining 24% of the total variance in the inter-model differences. b, An inter-model relationship of the SO warming index scaled by global warming with the corresponding principal component (in standard deviation). The purple dashed ellipse indicates 5–95% ranges. The correlation coefficient (Corre. coeff.), slope and P value are indicated.

Unlike the ENSO-induced change in the PSA pattern, which shows little impact on the SO warming (Extended Data Fig. 7b), inter-model differences in the zonally symmetric pattern systematically affect the SO warming (Fig. 3b). This zonally symmetric pattern features greater ENSO-induced high-latitude (poleward of 50° S) easterlies (Fig. 3a) hence weaker high-latitude equatorward Ekman transports and convergence south of 45° S, opposite to the pattern favourable for the SO warming shown in Fig. 1c. A greater increase in ENSO amplitude is associated with a weaker poleward intensification of westerly winds, and hence with weaker SO warming (Figs. 3b and 1d). Similar results are obtained in terms of meridional mass stream function. The dominant mode pattern reflects an equatorward shift of the zonal mean atmosphere circulation and is associated with a high-latitude easterly response and weaker SO warming (Extended Data Fig. 8).

Consistently, the inter-model relationship between projected zonal wind changes and ENSO amplitude changes features a pattern that is unfavourable to SO warming (Fig. 4a and Supplementary Fig. 2 compared with Fig. 1c). Rectification onto the mean circulation through zonally symmetric ENSO teleconnection plays a key role. A greater increase in ENSO amplitude leads to cumulatively stronger high-latitude easterly anomalies and westerly anomalies north of ~50° S. This is reinforced by differences in cumulative ENSO impact between a group of five models that project the strongest increase in ENSO amplitude (top five) and another group of five models that project the weakest change (bottom five) (Fig. 4b). The high-latitude cumulative easterly anomalies are substantially greater in the top five group than the bottom five group (black line, Fig. 4b).

Fig. 4: ENSO impacts through rectification of high-latitude zonal winds.
figure 4

a, Regression pattern of projected changes in zonal wind stress to changes in ENSO amplitude (in units of N m−2 °C−1). Stippling indicates where the correlation is statistically significant above the 90% confidence level based on a Student’s t-test. The projected changes are both scaled by global warming. b, Projected changes in cumulative annual-average (June to May) zonal-mean zonal wind stress (in units of N m−2 °C−1 of global warming) over all ENSO events defined as when |Niño3.4| is greater than 0.75 °C. The red curve is the average over the top five models with the largest ENSO increases, and the blue curve is the average over the bottom five models with the smallest ENSO increases. The black curve represents the difference between the two groups, that is, red minus blue.

The strong association between the projected change in ENSO amplitude and SO warming across models also holds under lower-emissions scenarios, that is, the SSP3–7.0, SSP2–4.5 and SSP1–2.6 scenarios (Extended Data Fig. 9). This is anticipated as influence by inter-model differences in GMT change have been excluded from the analysis. The conclusion is insensitive to choices of different observed datasets, thresholds for definition of ENSO events, ENSO index and length-of-time window to calculate projected changes (Methods).

Conclusions

We found a strong negative relationship between projected SO warming and ENSO response to greenhouse warming. Our finding of a slower SO warming associated with a stronger increase in ENSO amplitude through cumulative easterly wind anomalies over high latitudes suggests that uncertainty in projected SO heat uptake is linked to projected change in ENSO amplitude. If the projected increase in ENSO amplitude occurs, it would lead to reduced efficacy in SO heat uptake under global warming, and more heat would be retained in the atmosphere. On the flip side, projected weaker ENSO amplitude would lead to more SO warming. Thus, understanding the extent of future SO warming and its extended impacts requires an improved projection of ENSO. Our results highlight the far-reaching implications of ENSO’s response to global warming on the global climate system, beyond ENSO’s typical regional impacts such as drought, floods, tropical cyclones, heat waves and marine extremes45.

Methods

CMIP6 data and processing

To assess Southern Ocean heat uptake and influence from projected changes in ENSO teleconnections in the Southern Hemisphere, we take outputs from 27 CMIP6 models (Supplementary Table 1) in which data are available in all fields including ocean temperatures, SST, surface temperature and surface zonal wind stress. These models are forced under historical conditions (1850–2014) and the SSP 5–8.5 (2015–2100) emissions scenario41. Before data analysis, the horizontal grids of each model are regridded to 1° × 1°; the oceanic vertical level is interpolated to follow [10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220 250 280 300 340 400 500 600 700 800 900 1,000 1,300 1,500 1,800 2,000 2,300 2,500 2,800 3,000 3,300 3,500 3,800 4,000 4,300 4,500 4,800 5,000] in m. Monthly data from 1900 to 2099 are utilized. Outputs forced under the SSP3–7.0, SSP2–4.5 and SSP1–2.6 (2015–2100) emissions scenarios are also used to test the sensitivity of our results to future global warming.

The projected changes are measured by the difference between averages over the twenty-first century (2000–2099) and the twentieth centuries (1900–1999). Specifically, projected change in climatological fields uses annual data averaged from June (0) to May (1), and projected change in ENSO amplitude is based on December (0)–January (1)–February (1), the season in which ENSO peaks, with ‘(1)’ indicating year following the previous year ‘(0)’. Three-dimensional meridional winds are also used to calculate the atmospheric meridional mass stream function, and the atmospheric vertical level is interpolated from 0 mb to 1,000 mb with 10 mb as the interval. Net air–sea heat flux terms are analysed for examination of mechanism. Details of models and outputs availability are listed in Supplementary Table 1.

To include all the possible lead–lag influences from ENSO, impacts are obtained by regressing averaged anomalies from June (0) to May (1) onto a December (0)–January (1)–February (1) averaged Niño3.4 index. The monthly anomalies are calculated with reference to the monthly climatology over the twentieth century and then quadratically detrended over 200 years.

Equilibrium climate sensitivity (ECS) measures global mean temperature change in response to a doubled CO2 concentration after the climate system reaches a steady state64,65,66. In the present study, projected changes in global mean surface temperature are used to represent the ECS because not all models provide an ECS value. Indeed, they are highly correlated with a correlation coefficient of 0.9 (Extended Data Fig. 2c) based on ECS values from a previous study64. To examine influence beyond that due to differences associated with ECS, projected changes in models are scaled by the corresponding increase in global mean surface temperature, that is, changes per °C of global warming. Unless stated otherwise, projected changes are all scaled in the same manner.

Observations

We use ocean temperature data from the Institute of Atmospheric Physics/Chinese Academy of Sciences (IAP/CAS)67, which covers data from 1948 to 2019, to calculate observed trends in Southern Ocean temperature (Extended Data Fig. 1a). The observed monthly surface zonal wind stress and three-dimensional meridional wind are from the National Centre for Environmental Prediction (NCEP)/National Centre for Atmospheric Research (NCAR) reanalysis 1 (1948–2019)63. To test the sensitivity of our main result shown in Fig. 2 to different datasets and different time periods, we also used the two fields from Twentieth Century Reanalysis Version 2 (20CRv2)68 and European Centre for Medium-Range Weather Forecasts–the fifth generation of European Reanalysis (ECMWF-ERA5) (ref. 69), covering 1948–2010 and 1979–2019, respectively (Extended Data Fig. 6). To identify ENSO years and calculate associated ENSO teleconnection, SSTs from the Hadley Centre Sea Ice and Sea Surface Temperature dataset (HadISST)70 are used. To include all the possible lead–lag influences from ENSO, impacts are obtained by regressing averaged anomalies from June (0) to May (1) onto a December (0)–January (1)–February (1)-averaged Niño3.4 index. The monthly anomalies are calculated with reference to the monthly climatology over the full period and then quadratically detrended.

Atmosphere meridional mass stream function

To demonstrate influence from ENSO to Southern Hemisphere atmospheric circulation, the three-dimensional meridional wind is utilized to calculate the meridional mass stream function (ψm) (ref. 71), defined as follows:

$$\psi _{\mathrm{m}} = \frac{{2\uppi{a}\cos (\phi )}}g\;\mathop {\smallint }\limits_0^{P{\mathrm{s}}} v {\mathrm{d}}p$$

v represents zonal mean meridional wind; a is Earth radius (6,378 km); ϕ is latitude; g is the acceleration of gravity (9.8 m s−2); p is the pressure and Ps is the surface pressure. Therefore, the equation represents an integration from the top of the atmosphere to the surface.

Sensitivity test

Sensitivity tests to choices of different observed datasets, emission scenarios, thresholds for definition of ENSO events, ENSO index and length-of-time window to calculate projected changes are constructed. First, multiple observed datasets are used to confirm the ENSO teleconnections to Southern Hemisphere atmospheric circulation (Fig. 2 versus Extended Data Fig. 6). Next, the projected slowdown in Southern Ocean warming by increased ENSO amplitude and vice versa also holds under the three lower-emissions scenarios (Extended Data Fig. 9). Furthermore, the cumulative ENSO zonal-mean zonal wind anomalies which feature easterlies over the Southern Hemisphere high latitudes are insensitive to definition of ENSO events using different thresholds, for example, |Niño3.4 | >0.5 °C. Moreover, repeating all analysis using Niño3 instead of Niño3.4 shows similar results. Finally, the relationship in Fig. 1d is also significant using 70 year, 50 year and 30 year time windows, although the strength of the link expectedly decreases with a shorter time window due to stronger influence of internal variability.