Emergence of the enhanced equatorial Atlantic warming as a fingerprint of global warming

Abstract

The response of tropical sea surface temperature (SST) to global warming plays a crucial role in shaping both global and regional climates, so it receives immense attention and remains being debated. Here, we demonstrate that enhanced equatorial warming (EEW) is a more robust response to global warming than the commonly examined changes in the zonal SST gradient across the tropical Pacific, which is marked by discrepancies between observations and models. EEW is defined as the annual-mean SST warming averaged over 5°S-5°N, relative to the tropical SST warming averaged over 20°S-20°N. By combining observations and climate models, we identify the emergence of EEW in the Atlantic since the 1950s, primarily attributed to greenhouse gas forcing. The formation of EEW is driven by weakened equatorial upwelling, resulting from the slowdown of equatorial zonal winds. The identification of Atlantic EEW as a fingerprint of global warming has important implications for understanding changes in the tropical oceans in a warming climate and the associated impacts.

Observations and climate model simulations suggest the global signal of enhanced equatorial warming has emerged in the Atlantic since the 1950s due to a slowdown of equatorial easterly primarily driven by greenhouse gas forcing.

Introduction

Tropical ocean warming has garnered attention due to its impact on the rate of global warming^1,2, regional precipitation changes³, tropical cyclone genesis⁴, global climate feedbacks⁵ and many others. Most previous studies have focused on changes in zonal sea surface temperature (SST) gradient in the tropical Pacific^6–19, which is a key that underpins the dominant modes of variability in the Pacific Ocean, namely, the El Niño Southern Oscillation (ENSO), Interdecadal Pacific Oscillation (IPO) and Pacific Decadal Oscillation (PDO), that modulate the Walker Circulation on interannual and decadal timescales. Climate models tend to project a weakening of the zonal SST gradient (i.e., an El Niño-like change) and Walker Circulation under greenhouse gas forcing²⁰. However, a discrepancy in the change between observations and models during the past several decades has undermined the future projection, and the reason behind the disparity remains being debated^21–24. On the other hand, a more prominent and robust feature of SST response to future warming exhibited by climate models is an enhanced equatorial warming (EEW), characterized by intensified warming near the equator relative to the broader tropical oceans²⁵.

The global EEW pattern drives the contraction of tropical convection toward the equator, manifesting as reduced meridional migration of the intertropical convergence zone (ITCZ), which is generally positioned over the highest SSTs^26–30. This contraction of tropical convection shortens tropical cyclone seasons³¹. Regionally, global EEW is associated with an equatorward shift in the East Asian summer monsoon, subtropical circulation such as the subtropical jet, and the Meiyu–Baiu rainband³⁰. In particular, such a meridional SST gradient change in the tropical Atlantic can significantly influence the Walker Circulation and Hadley Circulation^32,33, the movement of the ITCZ^32,34,35 and precipitation patterns in northern and northeastern Brazil^32,35–37. Given the critical role of meridional SST gradient, especially the EEW in global and regional climates, it is essential to elucidate its drivers for understanding and improving predictions of climate change at multiple scales.

Progress has been made in understanding the formation of EEW in response to transient CO₂ experiments. For example, the mechanism for global EEW has been linked to the effect of ocean heat uptake³⁸, off-equatorial radiative forcing³⁹, and the EEW in the Pacific has been influenced by reduced evaporative damping²⁵ and the weakening of surface ocean currents at the equator driven by the slowdown of tropical circulation^40,41. However, observations show an increase of the equatorial Pacific near-surface currents over the last few decades^42–44. Furthermore, the slowdown of global atmospheric circulation and the narrowing of the ITCZ have already been detected in recent decades^39,45. Hence, it is urgent to examine whether EEW has already emerged and where it has emerged in the observational records, to determine the relative importance of the different underpinning physical processes. Here, it is found that the EEW has already been detectable in the Atlantic, primarily driven by reduced equatorial upwelling associated with the weakened atmospheric circulation. This warming pattern is expected to exert significant influence on the tropical climate and global atmospheric circulation.

Results

Emergence of enhanced equatorial warming in the Atlantic

We first compare the trends of global EEW and Pacific SST zonal gradient during 1951–2020 based on 36 CMIP6 models under external forcing, including greenhouse gas forcing (GHG), aerosol forcing (AER) and natural forcing (NAT). We select this time period for analysis because the observations are more accurate after 1951⁴⁶. The EEW index is defined as the difference in SST anomalies between 5 °S-5 °N and 20 °S-20 °N, while the Pacific SST zonal gradient is defined as the SST difference between the equatorial eastern Pacific (5 °S-5 °N, 140 °W-80 °W) and the equatorial western Pacific (5 °S-5 °N, 140 °E-170 °W) (See “Definition of SST metrics” in Methods). Although the two indices are highly correlated at the interannual timescale due to internal variability, there is no significant relationship between the long-term trend of the two indices (r = 0.02, Fig. 1a), suggesting that they are independent for longer timescales. The Pacific SST zonal gradient trends exhibit large uncertainty among models, with approximately half of the models exhibiting La Niña-like warming patterns (i.e., strengthening zonal SST gradient), while the other half showing El Niño-like warming patterns (i.e., weakening zonal SST gradient), with the multi-model ensemble mean (MME) close to zero. Instead, EEW trends exhibit positive values based on the MME, with inter-model agreement exceeding 77%. Specifically, there are 17 models simulating a statistically significant EEW trend, but only 3 models with a statistically significant zonal gradient trend (Fig. 1a). Further, the EEW pattern is evident regardless of whether models simulate El Niño-like or La Niña-like warming trends (Fig. 1b, c). In addition, despite the different changes in zonal SST gradient over the tropical Pacific, with La Niña-like warming in observations (Supplementary Fig. 1a) and El Niño-like warming in CMIP6 models (Supplementary Fig. 1b), the EEW trends across the entire tropical ocean are positive for both observations and CMIP6 models, statistically significant based on the MME of CMIP6 models (Fig. 2a). Therefore, EEW is a more robust response to global warming compared to the Pacific zonal SST gradient in the historical period. It’s worth noting that there is a peak of SST warming in the subtropical northeastern Pacific in both El Niño-like and La Niña-like warming patterns, which has been investigated in detail by Geng et al. (2023)⁴⁷.

Fig. 1. Higher model consistency of enhanced equatorial warming relative to zonal gradient of sea surface temperature in the tropical Pacific.

a Scatter plot between the trend of zonal gradient in tropical Pacific sea surface temperature (SST) and the trend of global enhanced equatorial warming (EEW) over 1951–2020 based on 36 CMIP6 models, multi-model ensemble mean (MME) and observations (OBS) based on the average of HadISST v1.1, ERSST v5 and COBE-SST 2. The correlation coefficient and p-value are noted in the upper left corner of figure (a). The markers with edges in red, blue, or gray denote statistically significant trends in EEW, the zonal gradient, or neither, respectively. Error bars denote one standard deviation from 36 CMIP6 models. The relative SST trend patterns with the warming trend averaged over 20 °S to 20 °N removed at each grid over 1951–2020 based on the mean of models simulating (b) La Niña-like warming pattern and (c) El Niño-like warming pattern. Stippling indicates the regions where the trend is statistically significant at the 95% level of confidence based on Student’s t test. Red lines are the zonal mean of the relative SST trend for all basins. Blue lines are the zonal mean of the relative SST trend for the Atlantic. Blue shading indicates 5 °S-5 °N. Unit: K per decade.

Fig. 2. Dominant role of external forcing in the observed enhanced equatorial warming trend in Atlantic Ocean.

a The trend of enhanced equatorial warming (EEW) over 1951–2020 from observations (OBS) (HadISST v1.1, ERSST v5 and COBE-SST 2), all forcing runs (ALL) based on 36 CMIP6 models, greenhouse gas (GHG) and aerosol (AER) only forcing runs based on 10 CMIP6 models in Global, Indian (30 °E-110 °E), Pacific (140 °E-80 °W) and Atlantic (60 °W-20 °E). Gray shading indicates the trend is not statistically significant at the 95% level of confidence based on Student’s t test. Unit: K per decade. b Scaling factors and their 5%–95% uncertainty ranges for ALL, GHG and AER in single-signal analyses for EEW of three tropical oceans. The dashed lines represent y-axis values of ± 1.

Although EEW is a robust signal in the CMIP6 models, the observed EEW during 1951–2020 is not statistically significant when averaged in the whole tropical ocean (Fig. 2a). Given the distinct spatial warming patterns among the three tropical oceans (Supplementary Fig. 1a), we further extend the EEW definition from global scale to individual ocean basins (Fig. 2). Both the Indian Ocean and Atlantic Ocean see statistically significant EEW trends (Fig. 2a). External forcing to the EEW trends is detectable using the optimal fingerprinting (OFP) method with the 5–95% confidence interval of scaling factor significantly above zero (Fig. 2b; see “Methods”). By contrast, EEW in the Pacific Ocean is neither statistically significant nor detectable, likely due to its substantial internal variability, such as the IPO and PDO^21,24,48. This is consistent with the discrepancy between the observed SST warming patterns and those driven by external forcing in the Pacific (Supplementary Fig. 1a, b). In contrast, the internal variability, represented as the noise based on the OFP method, is insufficient to mask the EEW trends under external forcing in the Indian and Atlantic Ocean (Fig. 2b). In addition, the EEW trends in the Atlantic are more evident than those across the entire tropical ocean in both the La Niña-like warming and El Niño-like warming in CMIP6 models (Fig. 1b,c). Among the 36 CMIP6 models, the Atlantic exhibits the highest inter-model agreement in positive EEW trends (86%), followed by the Indian Ocean (77%) and the Pacific Ocean (63%) (Supplementary Fig. 2).

Among the external forcings, greenhouse gases (GHGs) predominantly shape the spatial pattern of relative SST trends, with pattern correlation coefficient of 0.70 (Supplementary Fig. 1b, c). This pattern features stronger warming along the equator in the Pacific and Atlantic Oceans, in the subtropical northeastern Pacific, as well as in the western Indian Ocean. In contrast, aerosol forcing produces an opposite effect on the spatial patterns of SST trends, with stronger cooling along the equatorial Pacific and weaker relative warming along the equatorial Atlantic (Supplementary Fig. 1d). This may be due to the effects of reduced aerosol emissions from North America on the SST over North Atlantic⁴⁹. Analysis of EEW trends in each ocean reveals that GHG forcing dominates the positive trends in the Indian and Atlantic Oceans, though these effects are partially offset by aerosol forcing (Fig. 2a). Furthermore, the impact of GHGs on the EEW trend is detectable in the Indian and Atlantic Oceans (Fig. 2b). The responses of Atlantic EEW to external forcing and GHG-only forcing are closer to the observed trend with scaling factors closer to 1, while the EEW trends in the Indian Ocean are underestimated in models with scaling factors greater than 1. Moreover, the warming pattern in the Indian Ocean does not exhibit enhanced warming along the equator, but features an east-west dipole pattern under external forcing and GHG-only forcing (Supplementary Fig. 1b, c). Therefore, the EEW has emerged in the Atlantic, which can be regarded as a robust response to global warming. We will focus on the mechanism for the EEW in the Atlantic Ocean based on CMIP6 models in the following analysis.

Mechanisms for the enhanced Atlantic equatorial warming

To investigate the mechanism driving Atlantic EEW in recent decades, we conduct a mixed-layer heat budget analysis using 35 CMIP6 models, which have the complete heat flux variables (See “Mixed-layer heat budget analysis” in Methods). Among all atmospheric and oceanic processes, oceanic dynamics ($D_o$) stand out as the prominent driver of EEW trends, with a pronounced warming effect centered along the equator (Fig. 3a, b). By further estimating the contributions of different oceanic processes, both warm advection from weakening of the equatorial upwelling ($W_a{T}_c$) and changes in the zonal temperature gradient ($U_c{T}_a$) contribute to the EEW trend (Supplementary Fig. 3a, b, See “Mixed layer heat budget analysis” in Methods), with the former term dominating the stronger warming along the Atlantic equator (Supplementary Fig. 3d). Surface net heat flux acts as a counteracting response to EEW, primarily due to reduced shortwave radiation associated with enhanced equatorial convection (Supplementary Fig. 4). Climatologically, the poleward Ekman transport driven by the trade winds and Kelvin waves can both produce equatorial upwelling in the upper ocean⁵⁰. According to the ocean thermostat mechanism, equatorial upwelling has a significant influence on tropical SST variations⁵¹. In response to global warming, the weakening of ocean currents along the equator, driven by the slowdown of tropical circulation, is a key factor in EEW formation in future projections^40,41. In the historical period, the weakening of equatorial upwelling is evident in the upper Atlantic Ocean (0–150 m) based on CMIP6 models, with the most pronounced weakening occurring in the region (3 °S-3 °N) where the climatological upwelling is the strongest (Fig. 3c–e). Furthermore, the reanalysis data sets indicate that the equatorial upwelling in the Atlantic has weakened since the 1980s, consistent with externally forced experiments (Supplementary Fig. 5). These findings suggest that the weakened equatorial upwelling has contributed to the Atlantic EEW trend in recent decades.

Fig. 3. The weakened equatorial upwelling emerges as the primary driver of the enhanced equatorial warming trend in the Atlantic.

a The relative trends of latent heat (LH) flux, long wave (LW) radiation, sensible heat (SH) flux, short wave (SW) radiation, net surface heat flux ($Q_{net}$), oceanic heat transport effect ($D_o$, in W m⁻² per decade) and sea surface temperature (SST) (in K per decade) averaged over 5 °S-5 °N during 1951–2020 based on 35 CMIP6 models in Atlantic, with the average over 20 °S-20°N removed. We set the effects of warming SST as positive. The trends are all statistically significant at the 95% level of confidence based on Student’s t test. b The spatial patterns of relative trends of $D_o$ during 1951–2020 based on the average of 35 CMIP6 models in the Atlantic. Units: W m⁻² per decade. c Meridional mean (3°S-3°N) and (d) vertical mean (0–150 m) ocean vertical velocity trend (shading, in m day^-1 per decade) over 1951–2020 and climatological vertical velocity (black contours, contour interval 2 m day⁻¹; solid > 0 m day⁻¹, dashed < 0 m day⁻¹; thick 0 m day⁻¹) based on 21 CMIP6 models. Stippling indicates the regions where the trend is statistically significant at the 95% level of confidence based on Student’s t test. e The zonal mean of vertical velocity trend (red line, in m day⁻¹ per decade) and climatological vertical velocity (black line, in m day⁻¹) in (d). Blue shading indicates 3 °S-3 °N.

Inter-model relationship further supports this mechanism. Specifically, the linear trend in oceanic dynamics ($D_o$) is linked to the weakening of equatorial upwelling (r = − 0.56, Fig. 4a), which in turn is associated with a more pronounced EEW in the Atlantic (Fig. 4b). The weakening of equatorial upwelling occurs in most CMIP6 models, with 90% inter-model agreement. This highlights the critical role of oceanic processes, such as the weakening of equatorial upwelling, in shaping EEW trends for both the multi-model mean and inter-model spread. Since the equatorial upwelling is driven by zonal winds via poleward Ekman transport and equatorial Kelvin waves⁵⁰, we examine the inter-model relationship between trends of upwelling and zonal winds. We use the absolute values of zonal wind speed here, with negative values indicating the slowdown of easterlies along the equator. A statistically significant positive relationship (Fig. 4c) suggests that weaker zonal winds lead to reduced equatorial upwelling and vice versa. In addition, zonal wind changes are strongly correlated with EEW trends (Fig. 4d), and the weakened zonal wind has a high inter-model agreement of 75%. These findings confirm the dominant role of weakening equatorial upwelling, driven by the relaxation of trade winds, in the emergence of EEW trends over the Atlantic during recent decades.

Fig. 4. Weakened oceanic upwelling, driven by the slowdown of equatorial zonal winds, contributes to enhanced equatorial warming and dominates its inter-model spread.

a, b Scatter plot of the equatorial upwelling current (eqW: 3 °S-3 °N, 0–150 m mean of ocean vertical velocity) trend (in m day⁻¹ per decade) versus (a) relative trend of oceanic heat transport effect ($D_o$) (in W m⁻² per decade) and (b) enhanced equatorial warming (EEW) trend (in K per decade) based on 21 CMIP6 models over 1951–2020 in the Atlantic. c, d Scatter plot of the trend of absolute zonal wind speed ⁽in m s⁻¹ per decade) along the equator (3 °S-3 °N) versus (c) eqW trend based on 21 CMIP6 models and (d) EEW trend based on 36 CMIP6 models over 1951–2020 in the Atlantic. Regression lines are shown as black lines, correspondingly. The correlation coefficients and p-values are noted in each panel.

A fingerprint of global warming

The above analysis shows that the weakening of trade winds over the Atlantic Ocean has driven the EEW via reducing the upwelling. In the observations, the recent decades have witnessed a significant weakening trend of zonal wind along the equatorial Atlantic (black bar in Fig. 5a). Given that the zonal wind weakening is also a response to surface warming pattern, here we further demonstrate that the wind weakening occurs even without the SST pattern and is thus a driver of EEW. In response to global warming, the weakening of tropical atmospheric circulation, mainly Walker Circulation and Hadley Circulation, required by energetic and hydrological constraints, has been identified as a robust feature in future climate projections^40,52. The Walker Circulation slowdown over the Pacific is a clear response to the spatially uniform surface warming without any change in SST gradients⁵³. Thus, the weakening of zonal winds is a regional manifestation of this broader atmospheric circulation slowdown. To confirm this hypothesis, we further analyze three sets of prescribed-SST atmosphere-only experiments from CMIP6 models (See “Model simulations” in Methods). We use AMIPfuture4K − AMIP to isolate the effects of patterned SST warming and AMIPp4K − AMIP to estimate the response to spatially uniform SST warming^29,54.

Fig. 5. Uniform global warming dominates the weakened surface equatorial zonal wind speed.

a The absolute surface zonal wind speed along the equator (3°S-3°N) trends (in m s⁻¹ per decade) during 1951–2020 from ERA5, all forcing runs (ALL) based on 37 CMIP6 models, aerosol (AER) and greenhouse gas (GHG) only forcing runs based on 10 CMIP6 models in the Atlantic and the change of surface equatorial zonal wind speed (in m s⁻¹) from AMIPp4K minus AMIP and AMIPfuture4K minus AMIP in the Atlantic. Gray shading of the bar indicates the trend is not statistically significant at the 95% level of confidence based on Student’s t test. b, c The patterns of the absolute zonal wind speed changes (shading, in m s⁻¹) between (b) AMIPp4K and AMIP, (c) AMIPfuture4K and AMIP in the Atlantic and the climatological zonal wind (black contours, contour interval 1 m s⁻¹; solid > 0 m s⁻¹, dashed < 0 m s⁻¹; thick 0 m s⁻¹). The red vectors present the change of wind velocity (in m s⁻¹), correspondingly.

The experiments indicate that zonal wind speed weakens in response to both uniform and patterned SST warming (Fig. 5), consistent across all of the CMIP6 models (Supplementary Fig. 6). The uniform warming simulation without EEW shows zonal wind speed decreases over tropical Atlantic (Fig. 5b). While the SST warming pattern also leads to weakened zonal winds mainly along the equator (Fig. 5c), the magnitude of this weakening is smaller than that under uniform warming (Fig. 5a, b). This difference could be due to the effect of an El Niño-like warming pattern in the Pacific in the AMIPfuture4K experiments that tends to strengthen the trade winds along the equatorial Atlantic, partially offsetting the circulation slowdown that would occur without the SST warming pattern. Therefore, we conclude that EEW has emerged from observations in the Atlantic over recent decades driven by weakened zonal winds as a fingerprint of global warming.

Discussions

Our findings provide evidence for the emergence of EEW in the Atlantic from 1951 to 2020, establishing it as a robust fingerprint of tropical SST response to global warming. Unlike the uncertain changes in the Pacific zonal SST gradient, EEW exhibits greater consistency across climate models at the global scale. Among the three tropical oceans, the Atlantic displays the most significant and detectable EEW signal, which can be attributed to anthropogenic external forcing. Based on individual forcing simulations, GHG forcing stands out as the primary driver leading to the EEW trend, while anthropogenic aerosol forcing exerts an opposing influence. The reduced oceanic upwelling along the equator, driven by the relaxation of zonal wind speed, is a key mechanism behind EEW formation. Here, we emphasize the key role of tropical Atlantic atmospheric circulation in driving the Atlantic EEW under global warming. In contrast, some previous studies proposed that the changes in tropical atmospheric circulation were driven by the tropical Atlantic SST meridional gradient³². They are not conflicting, as the tropical ocean and atmosphere are closely coupled. Notably, weakened zonal winds over the Atlantic emerge as a robust response to global warming even under the spatially uniform SST warming experiments, supporting the EEW as a fingerprint of global warming. As EEW is a more robust signal than zonal SST gradient change under global warming, and has already emerged in the historical period, our study highlights the importance of further investigating the impacts of EEW on global and regional atmospheric circulation and hydrological cycle, rather than focusing solely on El Niño-like or La Niña-like warming patterns.

Methods

Observational datasets

In this study, we use three monthly mean sea surface temperature (SST) products to characterize the observed enhanced equatorial warming (EEW) from 1951 to 2020, including (1) HadISST v1.1 (Hadley Center Sea Ice and Sea Surface Temperature dataset version 1.1) with 1° × 1° horizontal resolution⁵⁵, (2) ERSST v5 (Extended Reconstructed Sea Surface Temperature version 5) with 2° × 2° horizontal resolution⁵⁶, and (3) COBE-SST2 (Centennial in situ Observation-Based Estimates version 2) with 1° × 1° horizontal resolution⁵⁷. While minor variations exist in EEW strength among these datasets, our main conclusions remain robust across all products. To analyze SST trend patterns, we interpolate the three datasets onto a common 90 × 180 global grid using bilinear interpolation. We have checked that this interpolation process does not affect the distribution of SST trends in the tropical oceans.

To examine the dominant factors influencing EEW, we use three monthly reanalysis datasets: (1) surface zonal wind from the ERA5 reanalysis (fifth generation ECMWF atmospheric reanalysis of the global climate) during 1940–2023 with 0.25° × 0.25° horizontal resolution⁵⁸, and (2) ocean vertical velocity from the Simple Ocean Data Assimilation (SODA) 3.15.2 during 1980–2020 with 0.5° × 0.5° horizontal resolution and 50 vertical levels extending to beyond 5000 m⁵⁹, and (3) ocean vertical velocity from NCEP Global Ocean Data Assimilation System (GODAS) during 1980–2020 with 1° × 0.5° horizontal resolution and 40 vertical levels extending to beyond 5000 m⁶⁰. The SODA datasets significantly enhance the accuracy, reliability, and applicability of ocean reanalysis through using higher-resolution models, improving assimilation methods, correcting flux bias and richer observational data⁵⁹. The GODAS datasets assimilates extensive in situ temperature and salinity observations, providing a more accurate description of the ocean’s internal structure, and is continuously updated⁶¹.

Model simulations

To investigate the physical mechanisms driving EEW under external forcing, we employ outputs from 38 Coupled Model Intercomparison Project 6 (CMIP6)⁶² models (Supplementary Data 1) under all-forcing (ALL) scenarios. External forcing refers to factors that can cause climate change outside the climate system, including natural (NAT) forcing, including solar radiation and volcanic eruption, and anthropogenic (ANT) forcing, such as greenhouse gas forcing (GHG) and aerosol forcing (AER).

We use monthly SST, surface heat flux, ocean horizontal and vertical velocity, mixed layer depth (MLD), ocean potential temperature (PT) and sea surface zonal wind data from both the historical and future scenario experiments. There are variations in the number of models used across figures because of the different availability of the required variables among all the models. The number of models and realizations used for each variable is detailed in Supplementary Data 1. To disentangle the contributions of different external forcings, mainly the greenhouse gases (GHGs) and anthropogenic aerosols (AERs), we utilize CMIP6 simulations from the Detection and Attribution Model Intercomparison Project (DAMIP), specifically hist-GHG and hist-AER, to isolate their respective impacts. To quantify the contributions of these forcings to EEW, we utilize monthly SST and sea surface zonal wind of historical (1850–2020) simulations across 10 CMIP6 models under GHG-only and AER-only forcing. The number of realizations per model is provided in Supplementary Data 1. All outputs from CMIP6 models are interpolated to a common 73 × 144 global grid using bilinear interpolation.

To estimate the noise term ($\epsilon$) used in the optimal fingerprinting (OFP) method for detecting anthropogenic influence on EEW in observation, we use 386 realizations of 10 large ensembles under all-forcing scenarios and 23 pre-industrial control simulations, with a total simulation length of 14,420 years (Supplementary Data 2). All outputs are interpolated to a common 73 × 144 global grid using bilinear interpolation.

To examine the response of equatorial zonal wind speed to uniform and patterned SST warming, we analyze monthly sea surface wind data from three sets of Atmospheric Model Intercomparison Project (AMIP), prescribed-SST atmosphere-only experiments based on CMIP6 models (Supplementary Data 1): (1) the standard AMIP simulation, which is driven by observed SSTs, sea ice, and prescribed anthropogenic forcing; (2) AMIP experiments with a globally uniform SST warming of 4 K (AMIPp4K); and (3) AMIP experiments with spatially patterned SST warming, derived from the ensemble mean of the coupled CMIP3 model experiments under a 1% per year CO₂ increase scenario at the time of CO₂ quadrupling, scaled to achieve a global mean SST increase of 4K (AMIPfuture4K). The analysis covers the period from 1979 to 2014. All model outputs are interpolated to a common 73 × 144 global grid using bilinear interpolation. In this study, the differences between AMIPfuture4K and AMIP represent the response to patterned SST warming, and the differences between AMIPp4K and AMIP represent the response to spatially uniform SST warming. This methodology is widely used in previous studies^29,54,63.

Definition of SST metrics

Enhanced equatorial warming (EEW) is measured as the annual-mean SST warming averaged over 5 °S-5 °N, relative to the tropical SST warming averaged over 20 °S-20 °N³⁰:

$$EEW={SST}_{equator}-{SST}_{tropical}$$ 1

The zonal SST gradient across the tropical Pacific is defined as the SST difference between the equatorial eastern Pacific (5 °S-5 °N, 140 °W-80 °W) and the equatorial western Pacific (5 °S-5 °N, 140 °E-170 °W)¹⁸:

$$SS{T}_{gradient}={SST}_{EP}-{SST}_{WP}$$ 2

A strengthened zonal gradient of the tropical Pacific SST is described as La Niña-like warming, and the weakened gradient is described as El Niño-like warming.

Optimal fingerprinting detection and attribution

To detect anthropogenic influence on EEW in observation, we apply an optimal fingerprinting (OFP) method using a generalized multivariate linear regression model^64,65:

$$y=\beta \left(X-\gamma \right)+\epsilon$$ 3

where $y$ represents observed variations, expressed as the sum of scaled responses to external forcings $X$ and observational noise $\epsilon$. The externally forced signal $X$ is estimated using a multi-model ensemble mean, with sampling noise $\gamma$ arising from the limited number of ensemble realizations. The scaling factor $\beta$ adjusts the forced responses to best match observations. A scaling factor with a 5–95% confidence interval significantly above zero indicates a detectable influence of external forcing. A scaling factor greater than 1 suggests an underestimated response to a specific forcing, and vice versa. To estimate the noise $\epsilon$, we use two independent internal variability estimates: (1) pre-industrial control simulations, providing 206 samples and (2) the inter-member spread of large ensembles from single-forcing runs, yielding 386 samples. The ensemble spread is derived by subtracting the multi-realization mean from each realization for a specific forcing in each model to eliminate the influence of external forcing. Each sample of a single model is a 70-year EEW sequence reflecting the influence of internal variability. Only large ensembles with at least 10 realizations are considered to ensure reliable internal variability estimates (Supplementary Data 2). In total, 592 samples of noise estimation are obtained, with half used for optimization and the other half for testing.

To enhance the signal-to-noise ratio, we apply a 5-year non-overlapping mean in the detection⁶⁶. Our study employs a regularized OPF algorithm that does not rely on Empirical Orthogonal Function (EOF) decomposition⁶⁷ and uses total least squares regression. We conduct one-signal analyses to detect the effects of individual external forcings by regressing observed anomalies onto simulated anomalies under all external forcings (ALL), greenhouse gas forcing (GHG), and anthropogenic aerosol forcing (AER).

Mixed layer heat budget analysis

The surface heat flux consists of four physical components: shortwave radiation $Q_S$, longwave radiation $Q_L$, and sensible heat flux $Q_H$ and latent heat flux $Q_E$. The net surface heat flux into the ocean $Q_{net}$, which is positive downward, is expressed as:

$$Q_{net}=Q_S+Q_L+Q_H+Q_E$$ 4

Following Xie et al. (2010)³, the SST tendency is assumed to approach zero when considering interdecadal and longer timescales. Compared to slowly increased SST, the variations in $Q_{net}$ are one order of magnitude larger, so the ocean transport effect balances the net surface flux to first order. Therefore, the oceanic process effects can be inferred from the net surface heat flux³:

$${D_o=-Q}_{net}$$ 5

where $D_o$ represents the ocean heat transport effect resulting from three-dimensional advection (ADV) and mixing. Since the sub-monthly and sub-grid processes, such as turbulent mixing and mesoscale oceanic eddies, cannot be well resolved by monthly heat budget analysis, we mainly focus on the ADV from horizontal and vertical temperature advection in this study. More importantly, the results clearly suggest the dominant role of the equatorial upwelling change in the total oceanic dynamic effects (Fig. 4a), indicating the minor role of mixing in oceanic dynamic effects on the EEW trend during 1951–2020. The ADV components of heat budget for the mixed layer can be expressed as follows, based on Jiang et al. (2024)⁶⁸:

$$ADV={\underbrace{-u_a\frac{\partial T}{\partial x_c}-}}_{U_a{T}_c}{\underbrace{u_c\frac{\partial T}{\partial x_a}-}}_{U_c{T}_a}{\underbrace{u_a\frac{\partial T}{\partial x_a}-}}_{U_a{T}_a}{\underbrace{v_a\frac{\partial T}{\partial y_c}-}}_{V_a{T}_c}{\underbrace{v_c\frac{\partial T}{\partial y_a}-}}_{V_c{T}_a}{\underbrace{v_a\frac{\partial T}{\partial y_a}-}}_{V_a{T}_a}{\underbrace{w_a\frac{\partial T}{\partial z_c}-}}_{W_a{T}_c}{\underbrace{w_c\frac{\partial T}{\partial z_a}-}}_{W_c{T}_a}{\underbrace{w_a\frac{\partial T}{\partial z_a}}}_{W_a{T}_a}$$ 6

in which a denotes linear trends during 1951–2020, and c denotes climatology during 1951–2020. The heat budget terms include changes in the temperature gradient ($U_c{T}_a$, $V_c{T}_a$, $W_c{T}_a$), changes in the oceanic current ($U_a{T}_c$, $V_a{T}_c$, $W_a{T}_c$), and their nonlinear interaction ($U_a{T}_a$, $V_a{T}_a$, $W_a{T}_a$). The horizontal advection terms are averaged over the mixed-layer depth. The vertical velocity is calculated at the bottom of the mixed layer.

Author contributions

L.D. designed the study and wrote the initial manuscript. Z.W. performed the analysis and generated all figures. L.W., F.S., A.S., W.Z., and T.Z. contributed to the interpretation of the results and improvement of the paper.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The data in this article can be downloaded from the following website: HadISST v1.1 is available at: https://climatedataguide.ucar.edu/climate-data/sst-data-hadisst-v11. ERSST v5 is available at: https://downloads.psl.noaa.gov/Datasets/noaa.ersst.v5/. COBE-SST2 is available at: https://psl.noaa.gov/data/gridded/data.cobe2.html. SODA is available at: https://soda.umd.edu/soda3_readme.htm. GODAS is available at: https://www.cpc.ncep.noaa.gov/products/GODAS/index.shtml. CMIP6 models are available at: https://esgf-data.dkrz.de/search/cmip6-dkrz/. CMIP5 models are available at: http://www.ipcc-data.org/sim/gcm_monthly/AR5/Reference-Archive.html.

Code availability

The codes used to generate the figures are based on MATLAB Language (Version: R2022b; https://ww2.mathworks.cn/products/matlab.html) and available at https://zenodo.org/records/17851140.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-68015-6.