Introduction

The El Niño-Southern Oscillation (ENSO) is the most dominant mode of interannual climate variability, exerting widespread influence on global climate patterns via atmospheric teleconnections1,2,3. These ENSO-induced teleconnections are crucial processes that drive regional extreme weather events, such as floods, droughts, and heavy rainfall worldwide4,5,6,7. Since El Niño’s characteristics and teleconnections depend on mean climate states, their impacts are expected to change significantly under global warming8,9.

In recent decades, numerous studies have pointed out that El Niño-like warming modifies atmospheric responses to the equatorial Pacific sea surface temperature anomalies (SSTA)10,11,12. The relatively enhanced sea surface temperature (SST) warming over the eastern Pacific promotes both the eastward migration and intensification of convection13,14,15. Additionally, precipitation responses will become highly sensitive to SSTA according to their non-linear relationship14,16,17, leading to the strengthening of convection, which in turn amplifies ENSO teleconnections. For instance, teleconnection patterns associated with the Pacific-North American (PNA) and Pacific-South American (PSA) patterns are projected to shift eastward and intensify with increased CO₂ forcing18,19,20. Moreover, the changes in position and strength of the westerly jet under greenhouse warming are expected to modulate these teleconnections by altering the midlatitude waveguide21,22. Consequently, these intensified ENSO-driven atmospheric teleconnections are likely to exacerbate regional climate impacts, leading to more frequent and severe extreme weather events23,24.

ENSO-driven atmospheric teleconnections influence regional climate directly and also indirectly through changes in regional SST, further amplifying their impacts on local climate patterns via complex air-sea interactions. ENSO-induced atmospheric circulation alters anomalies in near-surface temperature, humidity, and winds, which subsequently affect SST through ocean dynamics and surface energy fluxes25,26,27. Ocean dynamics include processes such as vertical mixing and horizontal advection, while surface energy fluxes refer to heat exchanges between the ocean and atmosphere, primarily through shortwave radiation and latent heat flux. These mechanisms illustrate how ENSO influences regional SST, subsequently shaping local climate patterns28.

Interannual SST variability driven by ENSO plays a pivotal role in not only shaping local climate patterns but also regulating marine ecosystems and biogeochemical cycles, influencing processes such as coral bleaching, fishery productivity, and oceanic CO₂ uptake29,30,31. These widespread impacts underscore ENSO’s ecological and environmental significance. Under global warming, the interaction between ENSO-induced SSTA and its subsequent effects on regional climate is expected to intensify. Elevated mean SST globally allows even minor SST variations to exert stronger influences on atmospheric circulation, thereby amplifying ENSO’s impacts on regional climate patterns14,16,32,33. Given these potential changes, it is essential to investigate how ENSO’s influence on regional SST may evolve in a warmer climate. However, despite the critical role of ENSO-induced SST in shaping climate and ecological systems, limited attention has been given to understanding how ENSO-induced SST changes in distant regions might respond to global warming.

To address this issue, we investigate changes in ENSO impacts on global SST and underlying mechanisms using the global warming scenario. We conduct a comprehensive regional analysis and demonstrate that the ENSO impacts on SST in most regions are intensified due to both enhanced ENSO teleconnections and changes in climatological mean states.

Results

Changes in El Niño-induced global SST

To examine the spatial pattern and magnitude of the El Niño impacts on global SST variability, we calculated the explainable variance (R2)34. We first calculated the lead-lag correlations between the 3-month running mean SSTA and the December-February (D(0)JF(1)) Niño3.4 SSTA (DJF Niño3.4). Then, we identified the maximum R2 at each grid point from June-August (JJA(0)) to the following June-August (JJA(1)) (see “Methods” for details). Figure 1a shows the result from the CESM1 present-day (PD) simulation, revealing the highest R2 values in the equatorial Pacific, attributable to autocorrelations in this region. Additionally, high R2 values are found in both the tropical Indian and Atlantic Ocean35, similar to the observational pattern, exhibiting pattern correlations ranging from 0.75 to 0.78 based on three observational datasets (Supplementary Fig. 1). In mid-to-high latitudes, the R2 distribution reflects SST responses driven by ENSO teleconnections, such as PNA and PSA patterns (Fig. 1a). The similarity to the observations suggests that the current model reasonably reproduces the impacts of El Niño on global SST.

Fig. 1: Changes in ENSO impacts on global SST.
Fig. 1: Changes in ENSO impacts on global SST.The alternative text for this image may have been generated using AI.
Full size image

The maximum explainable variance between sea surface temperature (SST) and December(0)–February(1) (D(0)JF(1)) Niño3.4 SST anomaly from June–August(0) (JJA(0)) to JJA(1) during a Present-day (PD), b 2085–2115 (2100) in CESM1. c The change in maximum R2 from the PD to 2100. Stippling in c indicates regions that are statistically significant at the 95% confidence level using the bootstrap test. d–f Same as a–c but CMIP6 multi-model ensemble mean (MME) results for pre-industrial, 1pctCO2 scenario and their difference, respectively. Stippled areas indicate regions where more than 80% of the models agree with the sign of the MME result. The boxes in c and f delineate the equatorial Pacific (10°S–10°N, 150°E–80°W), which is masked out to capture only the impacts on remote SST.

In the greenhouse warming scenarios, the spatial pattern of R2 remains largely similar to that in the PD simulation. However, the R2 values exhibit significant increases over most of the global ocean (Fig. 1b), indicating an amplification of El Niño impacts on global SST. Figure 1c highlights the difference between the present and future climates. It is clear that the pronounced increase is especially apparent in the subtropical north and south Pacific, the western tropical Atlantic, the North Atlantic Ocean off the U.S. East Coast, the south Indian Ocean (SIO), the East China Sea (EastCS), and the southern region of the Australian continent. While the tropical Indian Ocean experiences a notable weakening of the ENSO impact (Fig. 1c), SST responses are generally strengthened globally, suggesting an enhanced potential for ENSO to influence regional climates.

To ensure the robustness of these findings and avoid a single model dependency, we analyzed 34 models from the Coupled Model Intercomparison Project Phase 6 (CMIP6) pre-industrial and 1pctCO2 scenarios (Fig. 1d–f). The CMIP6 Multi-Model Ensemble mean (MME) exhibits an overall significant increase in the R2, consistent with the CESM1 results, although the SST response in the CMIP6 MME is slightly weaker than in CESM1 (Fig. 1d), likely due to multi-model averaging. Notably, R2 associated with ENSO is amplified at every grid point in the CMIP6 MME results (Fig. 1f). This consistent amplification across all models underscores the robustness of the enhanced ENSO impacts on global SST, which might be driven by fundamental dynamical processes.

To quantify the changes in ENSO impacts, we calculated the global mean of the absolute values of regression coefficients between the maximum R2 period SSTA and DJF Niño3.4 for each grid (hereafter referred to as SSTTOT; see “Methods” for details). As shown in Fig. 2a, SSTTOT increases significantly under greenhouse warming by approximately 13% compared to the PD simulation. This result is consistent across all 34 CMIP6 models, with each model showing a statistically significant increase (Supplementary Figs. 2 and 3).

Fig. 2: Contributors of the changes in ENSO-induced SST.
Fig. 2: Contributors of the changes in ENSO-induced SST.The alternative text for this image may have been generated using AI.
Full size image

a Global mean of the absolute values of regression coefficients between the maximum R2 period sea surface temperature (SST) and December(0)-February(1) (D(0)JF(1)) Niño3.4 SST anomaly (SSTTOT, left) and global mean of the contribution of surface flux to El Niño-Southern Oscillation (ENSO)-induced SST (SSTSFC+MLD, right). b The changes from Present-day (PD) to 2085–2115 (2100) in ENSO-induced SST and surface flux (SFC)-induced SST (gray and yellow). The ∆SSTSFC+MLD is decomposed into changes in the contribution of the ENSO-induced surface flux and the climatological mixed layer depth (∆SSTSFC and ∆SSTMLD, green and purple). c ENSO-induced surface flux contribution decomposed radiation (SSTRAD), sensible heat flux (SSTSH), and latent heat flux (SSTLH) contributions. All error bars are calculated at the 95% confidence level.

The mechanisms of amplified SST variability

We adapted the mixed-layer heat budget equation to decompose the contributions to the enhanced El Niño impacts. Previous studies have identified surface flux (SFC; positive into the ocean) as a primary driver of SST responses to ENSO25,36,37. To quantify the contribution of the SFC, we first define SSTSFC+MLD as the time-integrated surface flux-related term of the equation, which is determined by (1) El Niño-induced surface flux anomaly and (2) climatological mixed layer depth (MLD, see “Methods” for details). Figure 2a shows that the magnitude of the SSTSFC+MLD is larger than SSTTOT in both PD and 2100, possibly due to the damping role of the ocean dynamics. These results suggest that the SFC is a primary driver inducing ENSO-driven global SSTA at least on a global average. Additionally, the changes in SSTSFC+MLD between PD and 2100 are comparable to the change in SSTTOT, indicating that the amplified El Niño impacts can be attributed to anomalies induced by SFC.

To separate the mechanism, we decompose SSTSFC+MLD into (1) SSTSFC, which holds MLD fixed at its PD climatology, and (2) SSTMLD, which holds the surface flux anomaly (SFC anomaly regressed onto DJF Niño3.4) fixed at its PD value (see “Methods” for details). Since the MLD tends to be shallower due to enhanced upper ocean stratification under global warming38 (Supplementary Fig. 4), the SST responses to a given SFC forcing can be amplified owing to the reduced mixed-layer heat capacity39. However, the ∆SSTMLD explains only ~10% of the ∆SSTSFC+MLD (yellow bar in Fig. 2b). As a result, ∆SSTSFC substantially contributes to the amplification of the impacts on SST, accounting for ~90% of the change (green bar in Fig. 2b).

Next, we further decomposed SSTSFC into three components: radiation (SSTRAD), sensible heat flux (SSTSH), and latent heat flux (SSTLH). Figure 2c illustrates the contributions of each component to SSTSFC. In the present climate, radiation and latent heat flux are the primary contributors to the ENSO-induced SSTA. Notably, ∆SSTSFC is predominantly due to changes in ENSO-induced latent heat flux (~50%) and radiation (~37%). The increase in ∆SSTRAD is largely attributed to ENSO-induced longwave radiation, which is largely compensated by shortwave radiation (Supplementary Fig. 5). The increase in longwave radiation is likely driven by enhanced ENSO-induced local SST anomalies, with higher atmospheric moisture content and temperature. Stronger SST anomalies amplify downward longwave radiation by increasing atmospheric absorption, thereby reinforcing the radiative feedback process.

Importantly, ΔSSTLH is the most significant contributor, underscoring the critical role of changes in the evaporation process in amplifying ENSO impacts on SST. Under global warming, the atmospheric water vapor response intensifies markedly due to the Clausius–Clapeyron relationship33. As the mean temperature rises, the atmosphere’s capacity to hold moisture increases by approximately 7% per 1 K of warming, leading to substantially larger mean moisture. This enhanced moisture capacity amplifies latent heat flux responses, making evaporation a key factor in understanding the intensified ENSO-driven climate variability in a warmer world. Building on previous research that highlights the importance of water vapor response33, our study focuses on latent heat flux as a crucial element in explaining the intensified ENSO-induced SST responses.

Why are ENSO-induced latent heat flux anomalies amplified in the future climate? We first investigated changes in wind speed associated with ENSO by calculating the lead-lag regression of the 3-month running mean surface wind speed anomaly with respect to DJF Niño3.4 SSTA from MAM(0) to the following JJA(1), identifying the maximum absolute value at each grid point for both the PD and 2100. As shown in Fig. 3a, surface wind speed increases over the extratropical Pacific, the Indo-Pacific warm pool region, and the subtropical Atlantic Ocean, while it decreases over the equatorial central Pacific, the western Indian Ocean, and the northern off-equatorial Atlantic, which closely resembles the observed pattern40. Under greenhouse warming, the spatial pattern remains generally similar to that of the PD, but the magnitudes are enhanced (Fig. 3b). This enhancement is clearly depicted in the difference map (Fig. 3c and Supplementary Fig. 6), indicating that the wind-evaporation processes41 will be intensified over most of the global oceans via atmospheric teleconnection.

Fig. 3: Causes of intensified ENSO-induced latent heat flux variability.
Fig. 3: Causes of intensified ENSO-induced latent heat flux variability.The alternative text for this image may have been generated using AI.
Full size image

a Surface wind speed regressed onto December(0)–February(1) (D(0)JF(1)) Niño3.4 SSTA with considering time-lag effect from March–May(0) (MAM(0)) to June–August(1) (JJA(1)) in Present-day (PD) 400 years with CESM1. b Same as a but in 2085–2115 (2100). c Absolute difference between (b) and (a). d Regression coefficient between D(0)JF(1) precipitation and D(0)JF(1) Niño3.4 SST anomaly. Black contour and shading indicate the PD period and the change from PD to 2100, respectively. e, f Climatological air-sea specific humidity difference in the same data of (a) and (b). g Change in climatological air-sea specific humidity difference from PD to 2100. c, g show only statistically significant at the 95% confidence level using the bootstrap test, and hatches in g indicate sea ice region (not calculated).

The intensified El Niño-related wind responses are likely linked to enhanced convective responses driven by ENSO SST forcing. As shown in Fig. 3d, positive precipitation anomalies over the central and eastern Pacific become stronger and extend farther eastward under greenhouse warming. Such intensified precipitation responses have been documented in previous studies, which proposed plausible physical mechanisms42. First, atmospheric warming increases column water vapor at approximately the Clausius–Clapeyron rate, so that a given SST anomaly produces stronger precipitation and diabatic heating16,43. Second, El Niño–like mean state changes favor an eastward extension of ENSO-related convection anomalies13,14,18,44. Consequently, the same El Niño-related SST anomaly induces a larger atmospheric response—manifested as stronger convective heating and circulation changes under warming33. This intensified atmospheric circulation, in turn, leads to stronger anomalous surface wind speeds, thereby modulating evaporation through enhanced air–sea interaction.

Moreover, the latent heat flux depends on the mean specific humidity difference between the sea surface (qs) and the adjacent air (qa). Under global warming, this climatological air-sea specific humidity difference (δq = qs − qa) becomes larger, resulting in a greater latent heat flux response to a given wind speed anomaly. Figure 3e–g clearly shows that climatological δq increases over most global oceans, except for the subpolar North Atlantic. It is well known that the saturation specific humidity at the sea surface increases almost exponentially with rising SST following the Clausius-Clapeyron relationship. Although the qa increases, the qs increases more rapidly, leading to an overall increase in the climatological δq as SST rises. In this regard, the decreased mean δq in the subpolar North Atlantic (Fig. 3g) is also explained by the North Atlantic warming hole45,46. With the larger background δq and enhanced wind speed anomalies associated with ENSO, significantly larger latent heat flux anomalies will occur over most of the ocean surface.

While Figs. 2 and 3 provide a global overview of sensitivity changes, a regional perspective reveals that El Niño–induced SST anomalies intensify under warming across four representative basins—the East China Sea (EastCS), the U.S. East Coast, the South Indian Ocean (SIO), and the subtropical North Pacific (SNP)—with the amplification primarily driven by ENSO-induced latent heat flux changes (ΔQ’LH) (Fig. 4 and Supplementary Fig. 7). In the EastCS, an intensified western North Pacific anticyclone during D(0)JF(1)–MAM(1) weakens the climatological northwesterlies and reduces evaporation, yielding warmer SSTA (Fig. 4b and Supplementary Figs. 8, 9). Along the U.S. East Coast, stronger ENSO-induced westerlies from ASO(0) to JFM(1) increase wind speed and enhance local evaporation, producing pronounced cooling that peaks in FMA(1) (Fig. 4c and Supplementary Figs. 8, 9). In the SIO, by contrast, evaporation decreases, and warmer SSTAs develop (Fig. 4d), primarily governed by increases in climatological δq rather than by ENSO-induced wind speed (Supplementary Figs. 9 and 11). The SNP experiences cooler SST (Fig. 4e), as strengthened ENSO-induced winds align with the trade winds from ND(0)J(1) to AMJ(1), increasing wind speed and evaporation; here, ENSO-induced wind speed changes lead, whereas ENSO-induced δq acts as a damping (Supplementary Figs. 8–10). Across regions, the roles of MLD changes are secondary. Collectively, these results indicate that changes in the background δq—which modulate the efficacy of latent heat fluxes—together with stronger, region-dependent ENSO-induced wind anomalies, lead to an overall amplification of El Niño’s SST impacts (see Supplementary Note for quantitative shares and seasonal phasing).

Fig. 4: The changes in SST responses to ENSO.
Fig. 4: The changes in SST responses to ENSO.The alternative text for this image may have been generated using AI.
Full size image

a Change in regression coefficients between December(0)–February(1) (D(0)JF(1)) Niño3.4 sea surface temperature (SST) anomaly and SST during maximum R2 period (τ) relative to March–May(0) (MAM(0)), based on Present-day (PD) simulations using CESM1 (shade; K K−1). Stippling indicates regions with statistically significant changes between PD and 2085–2115 (2100), where the size of the dots represents confidence levels (90%, 95%, and 99%) and the color of the dots indicates whether the regression coefficient increased (red) or decreased (blue). b–e Temporal evolution of regression coefficients of area-averaged SST anomaly and latent heat flux with respect to D(0)JF(1) Niño3.4 SST anomaly from MAM(0) to June–August(1) (JJA(1)), during PD (blue line) and the 2100 period (red line), for each region: East China Sea (125°E–145°E, 20°N–35°N), U.S. East Coast (60°W–80°W, 25°N–40°N), South Indian Ocean (50°E–90°E, 40°S–20°S), and the subtropical North Pacific (150°W–165°W, 12°N–22°N). Red dots indicate statistically significant at the 95% confidence level using the bootstrap test. Vertical line marks τ (see “Methods” for details). The hatches show the net difference between the 2100 and PD El Niño-Southern Oscillation (ENSO)-induced latent heat flux curves (starting at the last moment before τ when the two curves are equal and ending at τ). A hatched area indicates an enlarged net heat uptake/loss during that window.

Discussion

In this study, we showed compelling evidence of amplified El Niño impacts on SST on the entire globe. The R2 between an El Niño index and SST increases globally both in large-ensemble global warming experiments in CESM1 and 34 CMIP6 models. Considering the large multi-model uncertainty, this robustness indicates that changes in fundamental dynamics lead to reliable amplification of the global SST response under greenhouse warming. Our findings demonstrate the enhanced linkage between tropical variability and global consequences, which is a combined outcome of the intensified El Niño-induced surface wind response and enlarged air-sea humidity difference. Overall, given the critical role of local SST in the regional climate system7,47,48,49,50, these results indicate that global climate variability becomes more dominantly regulated by El Niño under global warming, potentially leading to more frequent and severe extreme events.

The influence of El Niño on regional climate is known to vary depending on the type of El Niño event, commonly categorized as Eastern Pacific (EP) and Central Pacific (CP) El Niño51. Additionally, under global warming, ENSO diversity is projected to shift toward more frequent CP El Niño events and stronger EP El Niño events52. While previous research highlights the distinct climate impacts of the two types of El Niño53, we found that the amplification of El Niño-induced global SST and surface wind speed variability under global warming is consistent across both EP and CP types (Supplementary Figs. 12 and 13). This indicates that the intensified influence of El Niño on SST is a robust feature, independent of the specific spatial pattern of ENSO. Additionally, we further examined El Niño and La Niña composites separately to check whether the intensified ENSO impacts are phase-dependent. In both cases, the SST responses increase under warming, though the regional patterns are asymmetric (Supplementary Fig. 14), indicating that the impacts of both El Niño and La Niña on global SSTs will be intensified under greenhouse warming.

Though our study focused on changes in the SST sensitivity to ENSO, the actual ENSO impacts will also depend on changes in ENSO amplitude itself. Importantly, changes in ENSO amplitude are not clearly consistent across CESM1 and CMIP6 models, with the sign and magnitude varying across models. Because ENSO teleconnections tend to increase with the strength of tropical forcing, an increase (decrease) in ENSO amplitude would be expected to amplify (dampen) the remote SST response. To quantify ENSO’s total SST influence, we decomposed it into a non-amplitude term and an amplitude term, following regression-based approaches (Supplementary Note). The results show that the increase is dominated by the changes in non-amplitude component—~64% in CESM1 and ~76% in the CMIP6 multi-model mean—while amplitude accounts for ~32% and ~20%, respectively; importantly, even models with reduced ENSO amplitude54,55,56 show a net increase because atmospheric and oceanic sensitivity strengthens (Supplementary Figs. 15 and 16). Thus, while amplitude changes are model-dependent and introduce uncertainty in the total response, the strengthening of background-state and teleconnection sensitivity provides a more robust and consistent contribution across experiments.

Although our analysis focuses on ENSO, the mechanisms we identify are not exclusive to ENSO and are likely relevant to tropical SST forcing more generally. To test this, we performed additional linear regression analyses using indices for the Indian Ocean and the tropical Atlantic. The results show that these modes also display a strengthened influence on remote SSTs under warming (Supplementary Fig. 17). This consistency supports our interpretation that the amplification of ENSO impacts mainly results from (1) stronger diabatic heating and the associated atmospheric teleconnections, and (2) a larger air–sea specific humidity contrast—processes that can likewise operate for other tropical SST forcings. A more comprehensive assessment is, however, needed to fully quantify the enhanced impacts of non-ENSO tropical modes, which we leave for future work.

Global warming is also expected to modify the statistics of the tropical modes themselves. For ENSO in particular, El Niño-related extreme events may intensify, and these heightened impacts can persist even after emissions are reduced. El Niño’s influence on local climate is governed not only by the sensitivity of local SST to equatorial Pacific SST but also by shifts in the frequency and intensity of El Niño. Previous studies indicate that rising CO₂ levels can boost both the frequency and severity of extreme El Niño events13,57, implying that the overall effect of El Niño teleconnections could increase in severity under future climate conditions. Moreover, a recent study suggested that the release of heat previously stored in the deep ocean can sustain El Niño-like warming under CO₂ mitigation scenarios58. This persistent El Niño-like warming is likely to maintain robust precipitation anomalies59, thereby prolonging the enhanced impact of El Niño on SST variability. While more detailed analyses are necessary to fully understand these processes, our findings suggest the possibility that El Niño’s impacts on SST could remain significant even under future CO₂ mitigation.

Methods

Model data and experiment design

This study employed the Community Earth System Model version 1.2.2 (CESM1) for ensemble analysis. The atmosphere and land models have a horizontal resolution of approximately 1° and 30 vertical levels, whereas the sea-ice and ocean models operate with a nominal horizontal resolution of 1° and 60 vertical ocean levels. Climate simulations were conducted for both present and future conditions. The control experiment was run with a fixed atmospheric CO2 concentration of 367 ppm over 900 years, using the last 400 years for the present-day (PD) simulation. The future simulation (1pctCO2) utilized an escalating atmospheric CO2 concentration with a rate of 1% increase per year over 140 years, until it quadrupled the initial CO2 concentration (1468 ppm). We used model years from 2085 to 2115 referred to as 2100 in this study (after 85 years and 115 years from 367 ppm). Across the period, the average CO2 concentration is approximately 997 ppm, which corresponds to the high-emission Shared Socioeconomic Pathways 5-8.5 scenario. There are 28 ensembles that start from a different initial condition from the PD experiment.

In addition, 34 models in the CMIP6 pre-industrial and 1pctCO2 scenarios are employed, using one ensemble member per model for each experiment. These scenarios are reasonably similar to the previously mentioned PD and 1pctCO2 simulation in CESM1, but not in the initial CO2 concentration (284.7 ppm). Used models are summarized in Supplementary Table 1.

Observation data

We utilized observational SST data from the Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST), NOAA Optimum Interpolated Sea Surface Temperature (OISST), and Extended Reconstructed Sea Surface Temperature version 5 (ERSSTv5) for the years 1981–2023. All the observed SST dataset is used after interpolating to 1° × 1° horizontal resolution.

Definition of ENSO and teleconnection

The DJF SSTA, averaged over the Niño3.4 region (170°E–120°W, 5°S–5°N), is used as an indicator of ENSO. The SSTA is calculated by subtracting the monthly linear trend and climatological mean from the SST. To isolate ENSO teleconnection impacts on SST, we excluded the equatorial Pacific region (150°E–80°W, 10°S–10°N) from the global mean calculation, allowing for a more precise evaluation of the SST response to ENSO-induced teleconnections.

Teleconnection impacts on SST

A 3-month running mean was applied to the SST, SFC, and MLD data, covering the period from MAM(0) to the following JJA(1). The maximum R2 value was determined by analyzing the lead-lag correlation between the DJF Niño3.4 SSTA and the global SSTA. Establishing the MAM(0) period as the baseline, we evaluate the correlation changes within the timeframe between JJA(0) and JJA(1) (excluding MAM(0)-MJJ(0)). Within this interval, we select the period with the most significant deviation in correlation from the MAM(0) baseline, denoted as τ. The explainable variance corresponding to τ is then extracted and designated as the maximum R2 value. By concentrating on the interval with the most pronounced correlation change, the maximum R2 value serves as a critical indicator of the strength of the teleconnection between the ENSO and global SST variabilities.

We calculated the surface flux-induced SSTA in diagnostic framework, based on the following mixed layer heat budget equation:

$$\frac{\partial {T}_{m}}{\partial t}=\frac{{Q}_{{{\mathrm{net}}}}}{{\rho }_{0}{C}_{p}h}+{{\mathrm{Res}}},$$
(1)

where Tm is the mixed layer temperature, Qnet is net surface flux (positive into the ocean), \({\rho }_{0}\) (1027 kg m−3) is the reference density of seawater, Cp (3985 J kg−1 K−1) is the specific heat of seawater at constant pressure. We only used the surface flux term and calculated the surface flux contribution to ENSO-driven SSTA. The Res mainly represents contributions from ocean dynamical processes such as horizontal/vertical advection, entrainment, and diffusion.

SSTSFC+MLD, SSTSFC, SSTMLD is calculated as follows:

$${{{\mathrm{SST}}}}_{{{{\mathrm{SFC}}}}+{{{\mathrm{MLD}}}}}={{{\mathrm{sign}}}}\left({{{\mathrm{SST}}}}_{{{\mathrm{TOT}}}}\right)\times {\int }_{\!{{{\mathrm{MAM}}}}\left(0\right)}^{\tau }\frac{{{Q}^{{\prime} }}_{{{\mathrm{net}}}}}{{\rho }_{0}{C}_{p}\bar{h}}{dt},$$
(2)
$${{{\mathrm{SST}}}}_{{{\mathrm{SFC}}}}={{{\mathrm{sign}}}}\left({{{\mathrm{SST}}}}_{{{\mathrm{TOT}}}}\right)\times {\int }_{\!{{\mathrm{MAM}}}\left(0\right)}^{\tau }\frac{{{Q}^{{\prime} }}_{{{\mathrm{net}}}}}{{\rho }_{0}{C}_{p}{\bar{h}}_{{PD}}}{dt},$$
(3)
$${{{\mathrm{SST}}}}_{{{\mathrm{MLD}}}}={{{\mathrm{sign}}}}\left({{{\mathrm{SST}}}}_{{{\mathrm{TOT}}}}\right)\times {\int }_{\!{{\mathrm{MAM}}}\left(0\right)}^{\tau }\frac{{{Q}^{{\prime} }}_{{{\mathrm{net}}},{PD}}}{{\rho }_{0}{C}_{p}\bar{h}}{dt},$$
(4)

where, SSTTOT is the difference between the regression coefficient of SSTA with respect to DJF Niño3.4 SSTA at time τ and MAM(0), Q’net is the regression coefficient of SFC anomaly with respect to DJF Niño3.4 SSTA during MAM(0) to following JJA(1), and \(\bar{h}\) is climatological MLD. We perform the time integration of the surface flux term of Eq. (1) to measure the SST anomaly generated by the ENSO-induced SFC. SSTSFC and SSTMLD were calculated the same as SSTSFC+MLD but \(\bar{h}\) and Q’net are fixed at their PD simulation, respectively. To capture the absolute magnitude of each contribution, the integrated terms were multiplied by the sign of SSTTOT. SSTRAD, SSTSH, and SSTLH are computed in the same manner of SSTSFC, replacing the radiative (Q’SW + Q’LW), sensible (Q’SH) and latent (Q’LH) flux anomalies, respectively.

For the maximum wind speed forcing, we identify the month with the largest absolute regression onto DJF Niño3.4 SSTA within MAM(0)-JJA(1) (including MAM(0)-MJJ(0)). We allow this to differ from τ because surface flux forcing generally precedes the SST response, so its timing may lead to τ.