Introduction

The El Niño-Southern Oscillation (ENSO) is a naturally occurring fluctuation in the tropical Pacific with a typical 2-to-7-year cycle, which is characterized by irregular oscillations between abnormally cold (La Niña) and warm (El Niño) events owing to the Bjerknes feedback1,2,3. As one of the Earth’s most powerful interannual climate variability phenomena, ENSO has a far-reaching impact on the global weather patterns and climate system4,5. Extreme ENSO events, in particular, can cause severe climate disasters such as droughts, floods, and heatwaves on a global scale, seriously threatening agriculture and ecosystems. It is thus imperative to explore how ENSO will respond to the changing climate6,7.

The intensity of ENSO is known to be highly uncertain based on climate model projections8,9,10,11,12. This uncertainty stems not only from a balance of amplifying and damping feedbacks that govern ENSO and are influenced by climate change10, but also from the uncertainty of future anthropogenic aerosol and greenhouse gas (GHG) emissions13. On the other hand, anthropogenic aerosols and GHGs have been found to dramatically impact the mean state of climate10,14,15,16,17. Their effects on interannual climate variability such as ENSO, however, have remained unclear. In this context, understanding how historical anthropogenic aerosols and GHGs affect ENSO variability will shed light on future ENSO projection, as GHG and aerosol emissions and their atmospheric concentrations are better known during the historical period. Here, we leverage pre-industrial control runs and historical aerosol- and GHG-single-forcing simulations with a broad suite of climate models from the Coupled Model Intercomparison Project phases 6 (CMIP6) to investigate the mechanisms and effects of historical anthropogenic aerosols and well-mixed GHGs on modulating ENSO variability since 1850.

Results

We calculate the Niño 3.4 index by averaging detrended sea surface temperature (SST) anomalies in the Niño 3.4 region (5°N–5°S, 170°W–120°W) between 1850 and 2014 from CMIP6 pre-industrial, historical aerosol- and GHG-only simulations (Methods). The multi-model means of Niño-3.4 SST power spectrum show that ENSO variability peaks at 3.41, 3.74, and 3.48 years in the pre-industrial control run and historical aerosol- and GHG-only experiments, respectively, which indicates that anthropogenic aerosols slightly increase the ENSO lifetime while GHGs have a negligibly small effect on the ENSO period (Fig. 1a). Compared to the pre-industrial time, the multi-model mean amplitude of ENSO variability over 3 to 5 years increases by 5.6% in response to anthropogenic aerosols between 1850 and 2014 (Fig. 1a). This change in ENSO variability is closely linked to the mean-state change in the tropical Pacific18. In particular, anthropogenic aerosol-induced SST cooling generates an asymmetric temperature distribution in the equatorial Pacific, weakening cross-equatorial winds14 (Supplementary Fig. 1a). Such anomalous winds promote a southward shift of the Intertropical Convergence Zone, which prompts easterly wind anomalies in the eastern tropical Pacific and intensifies the upwelling there19,20. Accompanying thermocline shallowing in the eastern tropical Pacific leads to stronger feedback between temperature anomalies and surface winds within the ENSO system. Rising greenhouse gases, on the other hand, bring about a multi-model mean decrease by 8.6% in 3-to-5-year ENSO amplitude between 1850 and 2014 when compared to the pre-industrial time (Fig. 1a). SST warming due to GHG increases peaks in the eastern tropical Pacific, accompanied by an anomalous wind convergence from both hemispheres (Supplementary Fig. 1b). It is worth noting that, while these aerosol- and GHG-induced changes in 3-to-5-year ENSO magnitude appear to be relatively small based on percentages, they are statistically significant (Methods, Supplementary Table 1) and consistently opposite for each model (Fig. 1b and c), implying significant and distinct aerosol- and GHG effects on ENSO variability.

Fig. 1: Niño 3.4 SST power spectra and aerosol and GHG effects on the spectrum peak.
figure 1

a Power spectra of the Niño 3.4 indices of CMIP6 piControl (gray), HIST-AER (blue), and HIST-GHG (red) simulations as well as their 95% confidence limits (dashed/dotted curves). b The averaged energy spectra over a 3-to-5-year period for individual CMIP6 models to construct (a) for their piControl (gray), HIST-AER (blue), and HIST-GHG (red) simulations. c Differences of the averaged energy spectra over the 3-to-5-period between HIST-AER and piControl (HIST-AER minus piControl, green) and between HIST-GHG and piControl (HIST-GHG minus piControl, pink). All panels show the ensemble and multi-model mean results.

Full size image

To understand how anthropogenic aerosols and GHGs regulate ENSO variability, we first look at and into the aerosol- and GHG-driven changes in the annual cycle of SST in the eastern equatorial Pacific. Between 1850 and 2014, anthropogenic aerosols and GHGs reduce and enhance the SST annual cycle, respectively, in comparison to pre-industrial times (Supplementary Fig. 2). These reduced and enhanced SST annual cycles in the eastern equatorial Pacific contribute to the stronger and weaker ENSO variability seen from the historical aerosol- and GHG-only simulations via the frequency entrainment mechanism21,22. This is because when ENSO frequency and SST annual cycle are in close or multiplicative relation, a resonance effect may occur, making it easier to synchronize the timing of ENSO events with SST annual cycle23. As the SST annual cycle in the eastern equatorial Pacific strengthens, it tends to dominate the natural ENSO oscillation, also known as frequency locking. This frequency locking effect limits ENSO freedom, subjecting ENSO frequency and intensity to variations in SST annual cycle. As such, the GHG-intensified annual cycle has the potential to limit ENSO’s free oscillation, enabling it to restrain ENSO variability24. By contrast, anthropogenic aerosols weaken the SST annual cycle, which frees ENSO’s natural oscillation from the constraints of seasonal SST variations.

We further examine the changes in the Bjerknes stability (BJ) index to investigate the physical mechanisms by which anthropogenic aerosols and GHG influence ENSO variability (Methods). The BJ index is a metric for ENSO growth rates by evaluating the strength of air-sea interactions and their mean state in the equatorial Pacific, which envelopes a variety of feedback mechanisms, including the positive feedback effects of zonal advection, Ekman pumping, and thermocline, as well as the negative feedback effects of mean advection and thermodynamic damping25,26,27,28,29. Compared to the pre-industrial time, anthropogenic aerosols on average enlarge the total BJ index by around 57% between 1850 and 2014 (Fig. 2a, blue bar), as consistent with the enhanced aerosol-driven ENSO variability. A further analysis reveals that this enlarged BJ is due to a combination of reduced negative feedback and enhanced positive feedback. Specifically, contributions from the zonal advection and thermocline feedbacks increase by 5% and 11%, respectively (Fig. 2a). The increase is owing primarily to amplified coupling between surface winds and zonal currents \(({\beta }_{u})\) and higher sensitivity of thermocline to subsurface temperature changes \(({\alpha }_{h})\) (Fig. 2b). Notably, these effects are associated with increased zonal thermocline gradients in the east-central equatorial Pacific (Supplementary Fig. 3a) and intensified upwelling driven by anomalous cross-equatorial winds (Supplementary Fig. 1a).

Fig. 2: Bjerknes index and coefficients changes due to aerosol and GHG forcing.
figure 2

a The BJ index and individual components of piControl (gray), HIST-AER (blue), and HIST-GHG (red) simulations. TD, MA, ZA, EK, and TH represent the mean advection, thermal damping, zonal advection, Ekman, and thermocline feedbacks, respectively. b Changes in the percentage of different regression coefficients and mean temperature gradients due to aerosol forcing (HIST-AER minus piControl, light blue) and GHG forcing (HIST-GHG minus piControl, pink). Both panels show the ensemble and multi-model mean results. Error bars denote one standard deviation among models.

Full size image

In contrast, anthropogenic GHGs dwindle the total BJ index between 1850 and 2014 compared to pre-industrial times (Fig. 2a, red bar), as consistent with reduced GHG-driven ENSO variability. The thermocline feedback mechanism accounts for approximately 55% of the GHG-induced BJ index reduction (Fig. 2a). This is primarily measured by the impact of abnormal zonal wind stress on thermocline changes \(({\beta }_{h})\) (Fig. 2b). Trade winds weaken with rising GHGs and the eastern tropical Pacific warms accordingly (Supplementary Fig. 1b), which suppress equatorial upwelling30, diminish thermocline tilt (Supplementary Fig. 3b), and further warm the upper ocean in the eastern Pacific. Furthermore, the effect of well-mixed GHGs on the BJ index is linked to enhanced thermodynamic damping effect30 (Fig. 2a). Warmed background ocean temperature diminishes heat loss from the atmosphere to ocean, resulting in abated ocean temperature fluctuations and hence ENSO variability.

ENSO extremes and asymmetry

We probe the impacts of anthropogenic aerosols and GHGs on extreme ENSO events because, as per Ref. 9., higher GHG concentrations in future would increase the likelihood of more frequent and intense ENSO events. Furthermore, it has been established that the tropics have a convection temperature threshold31. Above this point, the atmosphere is more susceptible to minute variations in SST (such as precipitation and wind speed), which can hasten the emergence of warm anomalous events and possibly start severe El Niño events. And, since severe El Niño events frequently precede extreme La Niña events, extreme La Niña events as well. As in the pre-industrial simulation (Fig. 3a), the critical temperature is determined by the average daily precipitation exceeding a 2-mm level30. When SST exceeds this threshold, each increase in SST causes an acceleration in precipitation, implying a nonlinear feedback relationship between ocean and the atmosphere. In the winter of the Northern Hemisphere, daily precipitation exceeding 5 mm/day serves as a vital criterion for detecting extreme El Niño events.

Fig. 3: SST and precipitation in the eastern equatorial Pacific.
figure 3

Scatterplots of monthly SST versus monthly precipitation in the Niño-3 region for a piControl, b HIST-AER, and c HIST-GHG simulations. Orange dots represent cases when precipitation is larger than 5 mm/day. The red vertical dashed line represents the critical temperature for convection. The black vertical and horizontal dashed lines denote the climatological mean values of SST and precipitation, respectively.

Full size image

Relative to pre-industrial times, anthropogenic aerosols cool the mean SST in the Niño-3 region (5°N–5°S, 150°W–90°W) by 0.1 °C, making it well below critical temperature and hence result in less intense El Niño events between 1850 and 2014 (Fig. 3b). Anthropogenic GHGs, on the other hand, can bring about more extreme El Niño events during this period (Fig. 3c). The underlying rationale is that GHGs-induced warmer mean-state SST diminishes the vertical oceanic temperature gradient and atmospheric stability, intensifying atmospheric response to SST anomalies and strengthening the feedback mechanism for El Niño events.

We also delve into ENSO asymmetry—the asymmetry in interannual SST anomalies between El Niño and La Niña phases—which is typically measured by skewness32. Relative to pre-industrial times (Supplementary Fig. 4a), anthropogenic aerosols abate the interannual SST skewness in the tropical Pacific (Supplementary Fig. 4b), which means more extreme La Niña events between 1850 and 2014 (229 compared to 197 in pre-industrial times). On the other hand, anthropogenic GHGs cause an increase in the interannual SST skewness in the tropical Pacific during this period (Supplementary Fig. 4c), accompanied with more extreme El Niño events between 1850 and 2014 (240 compared to 198 in pre-industrial times). This finding suggests that SST skewness helps to regulate the frequency of extreme ENSO events, with positive skewness favoring extreme warm events and negative skewness favoring extreme cold events.

ENSO phase-locking

ENSO phase-locking, which is critical for ENSO prediction, refers to the tendency of ENSO events to start in spring in the Northern Hemisphere and peak in winter33. Here, we compare ENSO event histograms from the Niño 3.4 index34 between historical aerosol-only/GHG-only and pre-industrial simulations to elucidate the effects of anthropogenic aerosols and GHGs on ENSO phase-locking. ENSO events normally peak in November to January of the following year in the pre-industrial time (Fig. 4). The probability of ENSO events in the historical aerosol- and GHG-only simulations, however, is different from the pre-industrial case. The peaks of ENSO events remain concentrated in the winter months in the Northern Hemisphere in the historical aerosol-only simulation, the difference is negligible when compared to the natural variability of the pre-industrial time. Notably, anthropogenic aerosols have a conspicuous influence on the likelihood of ENSO events in April and June (Fig. 4a). In contrast, the historical GHG-only simulation shows an increase in the probability of ENSO peaking in November relative to the pre-industrial time (Fig. 4b), suggesting enhanced phase-locking of ENSO in response to well-mixed GHGs.

Fig. 4: Peak phase histograms depicting ENSO phase-locking.
figure 4

Probabilities of peak months during ENSO of a HIST-AER (blue) and piControl (gray), and b HIST-GHG (red) and piControl (gray) simulations. Both panels illustrate the ensemble and multi-model mean results. Error bars in (a) and (b) denote one standard deviation among models for HIST-AER and HIST-GHG simulations, respectively.

Full size image

In addition, we employ the preferred strength (Methods) to assess the ENSO phase-locking properties35. The preferred peak month of ENSO remains stable in Northern Hemisphere winter no matter whether the simulations account for the effects of historical aerosols or GHGs. In the historical aerosol-only simulation, the preference strength of 0.20 is slightly smaller than that in pre-industrial times (0.23). While the historical GHG-only simulation suggests an increase of the preference strength to 0.28. Given that preference strength quantifies the likelihood for a particular month when ENSO events reach their peak intensity, the larger preference strength, the more likely an ENSO event occurs in a particular month or small range of months, indicative of a higher extent of phase locking. Thereby, anthropogenic aerosols have the potential to suppress ENSO phase-locking, whereas rising well-mixed GHGs tend to promote ENSO events in the Northern Hemisphere winter, revealing the different roles of anthropogenic forcings in regulating ENSO phase-locking properties.

ENSO diversity

We also explore the effects of anthropogenic aerosols and GHGs on ENSO diversity, with a focus on events in the Eastern Pacific (EP) and Central Pacific (CP) that can impact regional climates and prompt teleconnections within the global climate system36,37,38. Relative to pre-industrial times when the largest SST anomalies for EP and CP events appear in the eastern and central Pacific, respectively (Fig. 5a,b), anthropogenic aerosols can diminish the EP SST anomaly maximum, particularly in the vicinity of the Peruvian coast (Fig. 5e). Climbing GHG emissions, on the other hand, are more likely to affect a CP El Niño, characterized by warm SST anomalies spreading across the equatorial Pacific (Fig. 5d).

Fig. 5: SST patterns and statistics corresponding to EP and CP El Niño events.
figure 5

a,c,e Linear regression of SST anomalies (color shading, in °C) onto the E-index for a piControl, c HIST-GHG, and e HIST-AER, all for years 1850-2014. b,d,f As in a,c,e but for SST regression onto the C-index. g The fraction of EP and CP El Niño events in piControl (gray), HIST-GHG (red), and HIST-AER (blue) simulations. All panels show the ensemble and multi-model mean results. Error bars in g denote one standard deviation among models. The base map in a–f is from NCAR Command Language map outline databases.

Full size image

We adopt the EP and CP indices to quantify the frequencies of different types of El Niño events. We define an EP or CP warming event when the corresponding winter average index exceeds one standard deviation27,28. We find more CP than EP events during the pre-industrial time (Fig. 5g). Anthropogenic aerosols promote the zonal advection feedback39,40, resulting in more El Niño transitions to CP events rather than EP events over 1850–2014. GHG increases appear to inhibit the transition from EP events to CP, due likely to GHG-induced reduction of thermocline slope in the east-central tropical Pacific Ocean40,41 (Supplementary Fig. 3b), making it more difficult for upwelling to bring cold water to the surface. The diminished upwelling, combined with weaker trade winds, limits the cooling necessary for the transition from EP to CP events. As a result, the warm anomalies in the EP persist for longer periods, impeding the shift towards CP El Niño events. It merits attentions that changes are statistically significant for EP-type ENSO from pre-industrial to aerosol forcing scenario, as well as for CP-type ENSO from pre-industrial to either aerosol or GHG forcing scenario, but not for EP-type ENSO from pre-industrial to GHG forcing scenario (Methods).

Discussion

In summary, we discover distinct effects of anthropogenic aerosols and GHGs on ENSO variability between 1850 and 2014. Anthropogenic aerosols augment ENSO amplitude primarily by weakening the mean advection feedback while strengthening the zonal advection and thermocline feedbacks. Well-mixed GHGs, on the other hand, abate ENSO amplitude primarily by inhibiting the thermocline feedback. We also find that historical aerosol increases suppress the occurrence of strong El Niño events in Northern Hemisphere winters, leading to more CP events than EP events. GHG emissions have the potential to amplify the strength and frequency of ENSO events and hinder the transition of El Niño to CP events.

Additionally, we compare the effects of anthropogenic and natural forcings on ENSO variability. We examine the CMIP6 historical natural forcing-only experiment, which is solely driven by volcanic and solar forcings from 1850 to 2014 (Methods). The multi-model mean result shows that natural forcings slightly promote ENSO over the 3-to-5-year period (Supplementary Fig. 5), but to a lesser extent than anthropogenic aerosol and GHGs impacts, revealing that anthropogenic forcings play a major role in modulating ENSO variability during the historical period.

It is noteworthy that, between 1850 and 2014, both anthropogenic aerosols and GHGs grew relative to pre-industrial times, potentially counteracting each other’s effects on ENSO during this period42. While for future climate, anthropogenic aerosols are projected to peak around the beginning of the twenty-first century and then decline, whilst well-mixed GHGs will continue to grow (Supplementary Fig. 5a). The responses of ENSO variability to future aerosol and GHG changes in the 3-to-5-year band seem to be consistent with those during the historical period, with enhanced ENSO variability under aerosol forcing compared to GHG forcing when both forcing agents follow the SSP2-4.5 scenario during 2020-2100 (Supplementary Fig. 5b–d). Under this scenario, the trajectory of anthropogenic aerosols and GHGs may determine whether their effects on ENSO will counterbalance each other or act in concert. Our findings highlight the complexity of ENSO responses to different climate forcings and unravel the complex interactions between natural climate variability and human-induced changes, which are of practical importance especially when considering the possible shift in the driving mechanisms of future tropical Pacific surface warming patterns43.

Methods

Observations and model simulation

To examine observed ENSO variability, we leverage 5 reconstructed monthly SST datasets: COBE-SST, COBE-SSTv2, ERSSTv4, ERSSTv5 and HadISST, which are generally of a resolution of one or two degrees and cover more than a century44,45,46,47,48,49. For each dataset, we calculate the spectrum of the Niño 3.4 index from 1891 to 2014, during which all observational data are available.

We employ 10 CMIP6 models (Supplementary Table 1) that are forced only by historical anthropogenic aerosols (HIST-AER), well-mixed GHGs (HIST-GHG), or volcanic and solar forcings (HIST-NAT) during the 1850-2014 period50. For each model, there are 3 to 10 ensemble members for its HIST-AER or HIST-GHG simulation. We calculate the ensemble mean for each model and then the multi-model mean based on the ensemble means of the 10 models. Also, we adopt the pre-industrial control run (piControl) with the 10 CMIP6 models (Supplementary Table 1). To further investigate the impacts of future aerosol and GHG changes on ENSO variability, we employ three CMIP6 models (CanESM5, GISS-E2-1-G, and MIROC6) with simulations available that are forced only by anthropogenic aerosols (SSP245-AER) or well-mixed GHGs (SSP245-GHG) during the 2020-2100 period following the SSP2-4.5 scenario50 (Supplementary Table 2). In addition, we harness ensemble historical (HIST) simulations with the 10 CMIP6 models (Supplementary Table 2) in which all historical forcings are included. Comparing the power spectra of the Niño 3.4 indices in the multi-model mean of historical simulations and the average of 5 observations from 1891 to 2014 (Supplementary Fig. 7a), we find that the CMIP6 models generally well simulate the observed ENSO characteristics, justifying their usage.

Bjerknes (BJ) stability index

The BJ index determines the ENSO growth rate by assessing the strength and mean state of the feedback process between ocean and the atmosphere in the equatorial Pacific, which in turn measures the linear stability of the ENSO coupling pattern. This process affects changes in the strength of ENSO and its SST25. The BJ index can be expressed as

$${BJ}=-{\alpha }_{s}-{\alpha }_{{MA}}+{\mu }_{a}{\beta }_{u}\left\langle -\overline{{T}_{x}}\right\rangle +{\mu }_{a}{\beta }_{w}\left\langle -\overline{{T}_{z}}\right\rangle +{\mu }_{a}{\beta }_{h}\left\langle \frac{\overline{w}}{{H}_{1}}\right\rangle {\alpha }_{h}$$
(1)

which can be obtained from a linear equation based on temperature anomalies within the mixed layer (upper 50 m) in the east-central equatorial Pacific (5°S–5°N, 180°–80°W). In Eq. (1), \(-{\alpha }_{s}\) denotes the thermodynamic damping (TD), \(-{\alpha }_{{MA}}\) denotes the mean advection feedback (MA), \({\mu }_{a}{\beta }_{u}\left\langle -\overline{{T}_{x}}\right\rangle\) denotes the zonal advection feedback (ZA), \({\mu }_{a}{\beta }_{w}\left\langle -\overline{{T}_{z}}\right\rangle\) denotes the Ekman upwelling feedback (EK) and \({\mu }_{a}{\beta }_{h}\left\langle \frac{\overline{w}}{{H}_{1}}\right\rangle {\alpha }_{h}\) denotes the thermocline feedback (TH). These terms represent the effects of background state, which include mean zonal and vertical ocean temperature gradients and vertical ocean velocity (\(\overline{{T}_{x}}\), \(\overline{{T}_{z}}\), \(\overline{w}\)), wind stress response to SST anomalies (\({\mu }_{a}\)), and oceanic response to equatorial wind stress anomalies (\({\beta }_{u}\), \({\beta }_{w}\), and \({\beta }_{h}\)) for zonal currents, upwelling and thermocline slope on the growth of ENSO SST anomalies. \({\alpha }_{h}\) represents the sensitivity between ocean subsurface temperature and sea level anomalies. \({H}_{1}\) represents an effective depth for vertical advection. Here, we adopt the European Centre for Medium-Range Weather Forecasts Ocean Reanalysis System 5 (ORAS5) reanalysis51 to estimate model’s effectiveness of the BJ index during the historical period, in accordance with Refs. 26,52. In the calculation of ORAS5 BJ index, since vertical current velocities in the upper ocean are not provided in the reanalysis, we compute them using the mass continuity equation. Between 1980 and 2014, the multi-model mean of CMIP6 historical simulations shows that thermocline and zonal advection feedbacks are the two main positive feedbacks, while among the negative feedbacks, thermal damping has a slightly larger contribution than mean advection (Supplementary Fig. 7b), all of which are consistent with the ORAS5 reanalysis and thus demonstrate the effectiveness of the BJ index in the CMIP6 models.

Definition and metrics of ENSO phase-locking

Phase-locking refers to the tendency of ENSO events, such as El Niño and La Niña, to occur or reach their maximum intensity during specific months of the year. The peak phase histogram is essentially a probability histogram where each of the 12 calendar months is normalized so that their total sum equals 1. Two metrics are used to quantify the properties of ENSO phase-locking from the phase histogram. One is preferred peak month (\({\varphi }_{p}\)) defined as the calendar month with the largest value in a 3-month averaged SST anomaly peak phase histogram. The other is preference strength (\({\varphi }_{s}\)) of phase-locking35, which is estimated as follows:

$${\varphi }_{s}=\frac{4}{3}\left({\varphi }_{\max }-\frac{1}{4}\right)$$
(2)

where \({\varphi }_{\max }\) is the sum of the 3-month values of the histogram centered on the peak month.

Eastern and Central Pacific events

To evaluate the different ENSO types, we first calculate the empirical orthogonal function of the detrended monthly SST anomalies in the tropical Pacific (10°S–10°N) during the period of 1850-2014. Then we obtain the principal components (PCs) for the corresponding period and smooth the normalized PCs using a 1-2-1 filter. We calculate the E and C indices from the first two principal modes (PC1 and PC2) of monthly SST anomalies:

$$E=\frac{{PC}1-{PC}2}{\sqrt{2}}$$
(3)
$$C=\frac{{PC}1+{PC}2}{\sqrt{2}}$$
(4)

where the E and C indices describe ENSO events with Eastern Pacific (EP) and Central Pacific (CP) patterns, respectively38.

Statistical significance test

We examine the statistical significance of historical aerosol and GHG impacts on ENSO variability. For each model, we divide its pre-industrial simulation into 3-10 (equal to the number of ensembles of HIST-AER/HIST-GHG simulation) 165-year truncations and treat each truncation as a single ensemble member (Supplementary Table 3). The pre-industrial ensembles have non-overlapping 165-year periods for models (ACCESS-CM2, ACCESS-ESM1-5, BCC-CSM2-MR, CESM2, FGOALS-g3, and IPSL-CM6A-LR) with sufficient simulations and fewer ensembles. The 165-year periods for CanESM5, GISS-E2-1-G and MIROC6 overlap by 100 years. The 165-year periods for HadGEM3-GC31-LL overlap by 65 years. We use the Student’s t-test with two pairs of ensemble simulations: HIST-AER versus pre-industrial and HIST-GHG versus pre-industrial, to estimate the statistical significance of historical aerosol and GHG effects on ENSO variability in terms of the averaged Niño 3.4 SST power spectra over the 3-to-5-year period for each model (Supplementary Table 1). We find significant aerosol-induced increase and GHG-induced decrease of ENSO variability among all the models at the 95% confidence level (Supplementary Table 1). We also conduct the Student’s t-test on the percentages of EP- and CP-type ENSO in the multi-model mean of changes between the pre-industrial and HIST-AER/HIST-GHG simulations, and find that changes are significant for EP-type ENSO between pre-industrial and HIST-AER, as well as for CP-type ENSO between pre-industrial and HIST-AER, and between pre-industrial and HIST-GHG at the 95% confidence level, respectively. However, the difference in EP-type ENSO between pre-industrial and HIST-GHG is not significant at the 95% confidence level (p = 0.17).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.