Formation mechanisms of the Central and Eastern Atlantic Niño

Abstract

Atlantic Niño is the dominant mode of interannual climate variability of the tropical Atlantic, prominently influencing climate conditions over local and remote regions. A recent study has identified two types of Atlantic Niño–central and eastern Atlantic Niño (CAN and EAN), with warm sea surface temperature (SST) anomalies centered in the central and eastern basins, respectively. Here we investigate their formation mechanisms by performing a mixed layer heat budget analysis and conducting numerical experiments. Results show that the development of both types is contributed by upper-ocean vertical processes caused by westerly wind anomalies. Furthermore, anomalous horizontal advection also plays an important role but is associated with distinct physical processes in the CAN and EAN. The difference is related to the climatological distribution of tropical Atlantic SST, exhibiting two warm centers located in the southwest and northeast tropical basins during boreal spring. Consequently, eastward current anomalies during Atlantic Niño cause warming only in the western-central equatorial Atlantic south of the equator, contributing to the formation of CAN. In contrast, Ekman convergence anomalies cause SST warming in the southwest and northeast equatorial Atlantic during CAN and EAN, respectively, favoring both types. We further analyze initiation mechanisms for the two Atlantic Niño types and find that CAN and EAN are triggered by the subtropical South Atlantic warming and oceanic Kelvin waves, respectively. These results suggest that the two Atlantic Niño types are associated with distinct physical drivers.

Introduction

The Atlantic Niño is the dominant mode of climate variability on interannual scales in the tropical Atlantic, characterized by warm sea surface temperature anomalies (SSTA) in the central-eastern equatorial Atlantic Ocean and westerly wind anomalies to the west^1,2,3,4,5. Atlantic Niño tends to peak during boreal summer, although some events also occur in boreal winter⁶. During Atlantic Niño, precipitation is typically enhanced in the tropical Atlantic and suppressed to the north^7,8, indicating a southward shift of the Inter-Tropical Convergence Zone (ITCZ)⁹. The westerly wind anomalies also drive anomalous Ekman convergence, leading to deepened thermocline and higher sea surface height in the basin⁶.

Atlantic Niño can prominently influence climate conditions over adjacent continents. For example, Atlantic Niño may cause increases in precipitation over West Africa^2,10,11,12, central Mediterranean¹³, and northeastern South America^12,14. Warm SSTA during Atlantic Niño affects the behaviors of different sardine species with different thermal preferences in West Africa¹⁵. Atlantic Niño can also modulate Atlantic hurricane activities^16,17. Furthermore, Atlantic Niño can influence the Pacific and Indian Oceans through atmospheric teleconnections^18,19,20. Atlantic Niño can trigger La Niña events by strengthening the Pacific Walker circulation^18,21,22,23. It can also weaken the Indian summer monsoon rainfall through the excitation of atmospheric equatorial Kelvin waves or extra-tropical atmospheric Rossby waves^{24,25,26,27,28}. Therefore, understanding the physical mechanisms for the formation of the Atlantic Niño is crucial for improving the seasonal climate forecast of various regions worldwide.

The development of the Atlantic Niño is associated with air-sea interaction processes in the region^1,5,29,30. The westerly wind anomalies during Atlantic Niño may induce downwelling oceanic Kelvin waves^31,32, which propagate eastward along the equator to the eastern basin, deepening the thermocline and inducing warm SSTA in the region. In addition, warming in the northern tropical Atlantic (NTA) can induce negative wind stress curl anomaly, causing downwelling oceanic Rossby waves^33,34. These waves may reflect at the western boundary as equatorial Kelvin waves and eventually lead to the Atlantic Niño^33,35,36. Furthermore, Richter et al.³⁴ suggested that Atlantic Niño can be also caused by the warm meridional advection from the NTA to the equatorial eastern Atlantic during the NTA warming. Hence, remote forcing such as ENSO and Atlantic Meridional Mode may cause the development of Atlantic Niño events through the associated NTA warming^37,38. In addition, during the positive Indian Ocean Dipole (IOD), rainfall is enhanced over the western tropical Indian Ocean, causing westerly wind anomalies over the tropical Atlantic that favor the formation of Atlantic Niño^39,40.

Recently, Zhang et al.⁴¹ have identified two types of Atlantic Niño: Central Atlantic Niño (CAN) and Eastern Atlantic Niño (EAN). The SSTA warming is primarily concentrated in the central basin during the CAN, whereas warm SSTAs mainly occur in the eastern Atlantic cold tongue region during the EAN. Consequently, SSTA-induced westerly anomalies dominate the western Atlantic during the CAN, while the wind anomalies occupy the entire equatorial Atlantic during the EAN. The climatic impacts of the two types of Atlantic Niño also exhibit evident discrepancies^42,43. For example, atmospheric circulation anomalies in Europe induced by the CAN and EAN exhibit distinct patterns, further causing different rainfall and temperature anomalies in the region⁴⁴. Given the notably different patterns and impacts between the CAN and EAN, it is imperative to understand their respective formation mechanisms. In this study, we analyzed multiple sources of observational data sets and conducted numerical experiments to examine the contributions of various physical processes to the development of the CAN and EAN as well as their initial triggers.

Results

Distinct patterns of the two Atlantic Niño types

We first examine the seasonal evolutions of the two types of Atlantic Niño by analyzing the composites of their respective indices (see Methods, Fig. 1e, f). Both the central and EAN emerge in February and March (FM), develop rapidly in April and May (AM), and peak in June and July (JJ). Thus, FM is considered as the “initiation stage”, AM the “developing stage”, and JJ the “mature stage” for the CAN and EAN events. To further characterize distinct patterns of the two types, we perform composite analysis of SSTA and 850hPa wind anomalies during JJ for both types (Fig. 1a, b). During the CAN, positive SSTAs are located in the central equatorial Atlantic, accompanied by wind anomalies converging toward the warm center. By contrast, during the EAN, positive SSTAs are concentrated in the eastern basin, with westerly wind anomalies occupying the entire equatorial Atlantic. Interestingly, the CAN warming exhibits a large loading south of the equator, whereas the warm SSTAs associated with the EAN are primarily located in the north equatorial Atlantic except for the coastal region. Such a meridional asymmetry may have important implications for different climatic impacts, the cause for which will be further explored below.

Fig. 1: Characteristics of the central and eastern Atlantic Niño.

Composites of June–July mean sea surface temperature anomalies (SSTA) (shading and contours, °C) from the Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST) and 850hPa wind anomalies (vectors, m s⁻¹) from the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5) during (a) central Atlantic Niño (CAN) and (b) eastern Atlantic Niño (EAN) events. Blue contours indicate warm SSTA higher than 0.55°C with an interval of 0.1°C. Purple line indicates the equator. c, d Same as a, b, but for sea surface height anomalies (SSHA) (shading, cm) and ocean current anomalies averaged over the upper 30 m (vectors, cm s⁻¹) from ECMWF Ocean Reanalysis System 5 (ORAS5). Hatching and gray arrows represent anomalies that are not statistically significant at the 95% confidence level. e, f show composites of CAN index and EAN index during CAN and EAN events, respectively.

Oceanic changes also differ between the two types of Atlantic Niño (Fig. 1c, d). The prevailing westerly wind anomalies during Atlantic Niño induce opposite sea surface height anomalies (SSHA) in the western and eastern basins. Furthermore, centers of the high SSHAs are located at the central and eastern equatorial Atlantic during the CAN and EAN, respectively. Similarly, although eastward ocean current anomalies driven by wind changes appear in the equatorial Atlantic in both types, the CAN-associated changes are mainly concentrated in the central basin whereas those during EAN extend to the eastern equatorial Atlantic.

To further analyze developing processes of the two Atlantic Niño types, we examine their associated anomalies in different stages (Fig. 2). In the initial stage of the CAN, significant warming appears in the South Atlantic, accompanied by westerly wind anomalies in the South tropical Atlantic (Fig. 2a). The warming in the central Atlantic and the westerly wind anomalies in the western equatorial Atlantic further strengthen during the developing stage, and reach their peaks during the mature stage (Fig. 2b, c). During the EAN, the initial warming is negligible in the South Atlantic. Instead, there is significant warming in the eastern tropical Atlantic coastal region, along with northwesterly anomalies over the eastern tropical Atlantic (Fig. 2d). During the developing stage, the coastal warming intensifies and extends toward the equator, where the extended warming induces westerly wind anomalies (Fig. 2e). The warming and westerly wind anomalies continue to strengthen and develop into the EAN at its mature stage. (Fig. 2f).

Fig. 2: Evolution of the two types of Atlantic Niño.

Composites of (a) February-March (FM, initiation stage), (b) April-May (developing stage), (c) June-July (mature stage) mean SSTA (shading, °C) and 850hPa wind anomalies (vectors, m s⁻¹) during CAN events. d–f Same as a–c, but for EAN events. The purple line indicates the equator. Hatching and gray arrows represent anomalies that are not statistically significant at the 95% confidence level.

Heat budget analysis during the developing stage

To explore the relative contributions of various physical processes to the formation of the CAN and EAN, we conduct a mixed layer heat budget analysis for the two types during their developing phase (Fig. 3). Both types of Atlantic Niño primarily arise from two processes: anomalous horizontal advection and vertical processes in the upper ocean (Fig. 3). The surface net heat flux term is a cooling effect, and thus does not contribute to the SST warming, consistent with previous finding^29,30,32. This result is robust across different data sets (Figs. S1 and S2). Further analysis indicates that both Ekman feedback and thermocline feedback play significant roles in the vertical process term, although their relative importance is difficult to assess due to uncertainties in the reanalysis data and the use of monthly data (Figs. S1 and S2).

Fig. 3: Mixed layer heat budget for the two types of Atlantic Niño.

Composites of April-May mean (a) temperature tendency term, (b) surface net heat flux, (c) horizontal advection, and (d) vertical processes (°C month⁻¹) during CAN events. The heat flux data is from the ERA5 and other data is from ORAS5. e–h Same as a–d, but for the EAN years. Hatching represents anomalies that are not statistically significant at the 95% confidence level.

Variations in the horizontal advection and vertical processes have different contributions to the formations of the CAN and EAN. While the vertical process contributes to the SST warming in the equatorial Atlantic in both types, it is stronger in the central and eastern basins during the CAN and EAN, respectively (Fig. 3d, h). This effect is associated with the Ekman feedback and thermocline deepening (Figs. S1 and S2)^45,46,47, both of which contribute to the differences in the warming patterns between the CAN and EAN, which is further linked to the different wind anomalies (Fig. 2b, e). Westerly wind anomalies during the CAN primarily occur over the western-central basin, suppressing upwelling and favoring the deepening of the thermocline in the region, whereas during the EAN, the wind anomalies and associated anomalous downwelling and thermocline changes are stronger over the eastern equatorial Atlantic (Fig. S3). Note that the wind anomalies are, in turn, driven by the central and eastern Atlantic warming during the two types. Hence, the vertical processes term essentially describes the Ekman feedback and thermocline feedback that amplifies either the CAN or EAN, depending on the location of the SST warming in the initiation stage.

The horizontal advection term differs prominently between the CAN and EAN (Fig. 3c, g). During the CAN, this effect tends to cause warming in the central equatorial Atlantic and cooling in the eastern basin (Fig. 3c). In contrast, the anomalous horizontal advection during the EAN mainly contributes to the SST warming in the eastern equatorial Atlantic and the coastal region (Fig. 3g). Note that compared to other terms in the budget analysis, this process makes the most prominent contribution to the formation of distinct warming patterns during the two Atlantic Niño types (Fig. S4).

The horizontal advection term consists of three components: ${U}^{{\prime} }\bar{T}$ term, ${\bar{U}T}^{{\prime} }$ term, and the nonlinear advection term (Figs. S5 and S6). The first two terms represent temperature transport by ocean current anomalies and temperature anomalies advected by climatological current respectively. The nonlinear term is generally small and noisy and is therefore neglected here. Also, note that the second term primarily redistributes thermal anomalies initially induced by other processes. Hence, only the ${U}^{{\prime} }\bar{T}$ term actively modulates the growth of the two types of Atlantic Niño. Next, we examine the zonal and meridional components of ${U}^{{\prime} }\bar{T}$ term, i.e., ${u}^{{\prime} }\frac{\partial {\bar{T}}_{m}}{\partial x}$ and ${v}^{{\prime} }\frac{\partial {\bar{T}}_{m}}{\partial y}$, separately (Fig. 4).

Fig. 4: Horizontal advection anomalies during the two types of Atlantic Niño.

a Composites of April-May mean zonal advection (shading, °C month⁻¹) and ocean current anomalies (vectors, cm s⁻¹) during CAN events. Purple line indicates the equator. b Same as a, but for meridional advection anomalies. c, d Same as a, b, but for EAN. Hatching and gray arrows represent anomalies that are not statistically significant at the 95% confidence level. e April–May mean climatology of mixed layer temperature (°C). Contours indicate the 28.5 °C isotherm. f Same as e, but for zonal (shading) and meridional (contours) gradients of mixed layer temperature (10⁻⁶ °C m⁻¹). The solid (dashed) lines represent positive (negative) values (contour interval = 3 × 10⁻⁶ °C m⁻¹), and the dotted line represents zero values.

Results show that although eastward ocean current anomalies occur in the equatorial Atlantic during both types of the Atlantic Niño, the ${u}^{{\prime} }\frac{\partial {\bar{T}}_{m}}{\partial x}$ term is positive only in the western-central basin south of the equator (Fig. 4a, c). Hence, while zonal advection anomalies contribute prominently to the SST warming of the CAN (Fig. 4a), they play a negligible role in the EAN (Fig. 4c). This result is related to the mean state distribution of mixed layer temperature in AM. Unlike the tropical Pacific, which exhibits a large warm pool in the western basin, two warm centers are present in the southwest and northeast equatorial Atlantic during boreal spring, respectively (Fig. 4e). Hence, the zonal gradient of the mixed layer temperature is negative in the southwestern tropical Atlantic and positive elsewhere in the basin (Fig. 4f). As a result, the eastward current anomalies contribute to SST warming only in the southwestern tropical Atlantic, favoring the formation of the CAN.

By contrast, the meridional advection term plays an important role in the development of both the CAN and EAN (Fig. 4b, d). Westerly wind anomalies during the Atlantic Niño induce Ekman convergence anomalies. Consequently, warm water from the two climatological warm centers is advected toward the equator. Given that the westerly wind anomalies are stronger over the central basin during the development of the CAN while dominating the eastern basin during the development of the EAN, the meridional advection term significantly contributes to the warming in the southwestern tropical Atlantic and the northeastern basin warming during the CAN and EAN, respectively. Moreover, northerly wind anomalies induced by coastal warming in the eastern basin further transport warm water from north of the equator to the southeastern Atlantic, causing coastal warming (Figs. 2e and 4d).

The results above suggest that both zonal and meridional advection anomalies contribute to SST warming in the south equatorial Atlantic during the CAN, whereas only the anomalous meridional advection favors warming in the northeastern equatorial Atlantic. This may contribute to the meridional asymmetry of the warming centers during the two Atlantic Niño types–higher warming south of the equator during the CAN and stronger warming in the northeastern Atlantic during the EAN (Fig. 1a, b).

Initiation mechanisms for the CAN and EAN

Employing a heat budget analysis, we have investigated different contributions of various physical processes to the development of the two types of Atlantic Niño. However, it is noted that both the horizontal advection and vertical processes may provide positive feedback mechanisms that amplify either the CAN or EAN. Hence, whether the Atlantic Niño develops into the central or eastern type depends on the initial atmosphere-ocean conditions prior to the developing stage. Therefore, we further explore the different triggering mechanisms for the two types during the initiation stage in FM.

Prior to the development of most (7/10) CAN events, significant warming is observed in the subtropical South Atlantic (Fig. 2a), inducing an interhemispheric thermal contrast. Consequently, prominent cross-equatorial northerly wind anomalies emerge over the tropical Atlantic. Under the influence of the Coriolis force, the anomalous northerlies turn into westerly anomalies over the equatorial southern Atlantic. This result is supported by atmospheric model experiments, wherein warming in the subtropical South Atlantic is superimposed on top of the monthly SST climatology as additional forcing in the sensitivity experiment (see Methods). Model results indeed exhibit cross-equatorial wind anomalies as well as anomalous westerlies over the tropical Atlantic (Fig. S7). The westerly wind anomalies may subsequently cause thermocline deepening, reduce upper-ocean upwelling, and induce eastward current anomalies. These changes favor central equatorial Atlantic warming, which triggers Bjerknes feedback and leads to further strengthening of wind and SST anomalies. Hence, the South Atlantic warming emerges as an important precursor for the CAN.

In contrast, warming in the south tropical Atlantic is insignificant during the initiation stage of the EAN (Fig. 2d). Instead, the precursors of the EAN are similar to those of the canonical El Nino^48,49,50, with warm SSTAs first appearing in the eastern basin along the western coasts of Africa, which subsequently develops in the eastern tropical Atlantic in the following season. Previous studies suggest that oceanic equatorial Kelvin waves can cause the Atlantic Niño by deepening the thermocline in the eastern basin and resulting in warm SSTA in the region^31,32. Since the mean thermocline depth is shallower off the coasts of West Africa (Fig. 6b), surface warming tends to emerge in the coastal region, consistent with Fig. 2d. These results indicate that oceanic Kelvin waves could initiate the EAN. To examine this effect, we analyze the Hovmöller diagram of daily sea level anomalies in the equatorial Atlantic during selected EAN/Niña events. Results show that a majority (4/6) of observed EAN events are preceded by the eastward propagating Kelvin waves, inducing SST anomalies in the eastern basin (Fig. 5). In contrast, only one CAN event appears to be associated with the oceanic waves (Fig. S8). For the CAN, the southeast wind anomalies in the eastern Atlantic, induced by the warming in the central basin, weaken the downwelling Kelvin waves propagating to the eastern Atlantic and coastal region, eventually causing the warming to remain confined in the central Atlantic, as revealed by our linear ocean model experiments that separate the role of wind anomalies in the western and eastern basins in causing sea level anomalies during the CAN (Fig. S9). These results suggest the important role of equatorial Kelvin waves in triggering the EAN.

Fig. 5: Oceanic Kelvin waves during EAN events.

a Regions used for Hovmöller diagram calculations. b–g Hovmöller diagrams of SSHA (shading, cm) and SSTA (contours, °C) showing the propagation of equatorial Kelvin waves (2°S–2°N average) from 40°W to the eastern ocean boundary (indicated with EB), and then southward along the boundary (anomalies are averaged over the pink area along the coast in Fig. 5a) for the EAN years. The daily SSH data is from the Archiving, Validation, and Interpretation of Satellite Oceanographic (AVISO) dataset. Brown (purple) dashed line indicates downwelling (upwelling) Kelvin wave propagation during Atlantic Niño (Atlantic Niña) events.

Discussions

Atlantic Niño plays an important role in tropical basin interactions and thereby prominently influences climate conditions across various regions worldwide. Understanding its formation mechanisms is crucial for enhancing our capacity to predict the tropical Atlantic climate variability as well as the global climate system. In this study, we investigate the formation mechanisms of the CAN and EAN by analyzing observational data and conducting numerical model experiments. Results show that both types of Atlantic Niño arise from the Bjerknes feedback processes associated with changes in horizontal advection and vertical processes. The latter is primarily associated with the thermocline feedback and Ekman feedback.

Horizontal advection anomalies play a dominant role in causing the differences in the warming patterns observed between the two Atlantic Niño types. Specifically, due to the two climatological warm centers in the southwest and northeast equatorial Atlantic in April and May, the eastward current anomalies can only contribute to SST warming in the southwestern tropical Atlantic, promoting the formation of the CAN but not favoring the EAN. In addition, westerly wind anomalies dominate the central basin and eastern basin during the developing stage of the CAN and EAN, respectively. The wind anomalies induce anomalous Ekman convergence that significantly warms the southwestern tropical Atlantic during the CAN and the northeastern basin during the EAN, resulting in the different types of Atlantic Niño.

Whether the Atlantic Niño develops into the central or eastern type depends on the initial warming pattern, which is further amplified through positive feedback processes. Our results suggest that the eastward propagating equatorial Kelvin waves cause thermocline deepening and SST warming in the eastern equatorial Atlantic upon reaching the region, favoring the formation of the EAN. On the other hand, the CAN is typically triggered by a warming in the subtropical South Atlantic, which induces westerly wind anomalies over the western Atlantic that favors central basin warming.

It is noteworthy that both CAN and EAN exhibit pronounced asymmetries between Niño and Niña phases. In particular, the central Atlantic Niño appears weaker than Niña, while the EAN is stronger than its negative phase. These intriguing asymmetric patterns suggest differing formation mechanisms and distinct climate impacts on the surrounding regions, warranting further investigation in future studies.

As the dominant mode of interannual climate variability on the planet, El Niño-Southern Oscillation (ENSO) also exhibits two distinct types: the central Pacific (CP) and eastern Pacific (EP) ENSO^45,51,52,53 (Fig. 6a, d). The formation mechanisms of the two flavors of ENSO have been extensively studied^{45,49,51,54,55}. The dominant contributor to the formation of the EP ENSO is the thermocline feedback, whereas the zonal advective feedback is conducive to warming the central basin, favoring the CP ENSO^45,52 (Fig. S10). However, during the Atlantic Niño, the contribution of the thermocline deepening to the tropical Atlantic warming seems higher during the CAN compared to the EAN. This inter-basin discrepancy is associated with the different mean thermocline depth between the two basins (Fig. 6c). In particular, the thermocline in the CP is almost twice as deep as that in the central Atlantic. Consequently, while variations in the thermocline cannot exert substantial influences on SST in the CP, they play an important role in SST warming during the CAN. These results suggest that the thermocline feedback and zonal and meridional advection effects all contribute to the development of SST warming of the CAN. As a result, the amplitude of equatorial SSTA is higher during the CAN compared to the EAN, which also differs from ENSO (Fig. 6d). Furthermore, the relationship between the Atlantic Niño and ENSO, a subject of long-standing significant interest, has consistently attracted considerable attention^56,57. Given the mechanisms of two distinct types of Atlantic Niño, further research is warranted to ascertain whether they can provide new perspectives on this debated issue.

Fig. 6: Comparison between ENSO and Atlantic Niño.

Annual mean climatology of (a) SST (°C) and (b) depth of 20°C isotherm (z20, m). Purple boxes denote the centers of the CP ENSO and CAN, and red boxes denote the centers of the EP ENSO and EAN. c Annual mean climatology of SST (°C, orange line) and z20 (m, purple line) averaged between 3°S and 3°N. Purple shadings (red shadings) denote the centers of CP ENSO (EP ENSO) and CAN (EAN). d Composites of DJF SSTA (°C) averaged between 3°S and 3°N during the CP ENSO (purple line) and EP ENSO (red line) years. e Same as d, but for JJ-mean composites during the CAN and EAN years.

Previous studies have found that the EAN has weakened substantially since ~2000, whereas the amplitude of the CAN has not changed much in the past few decades^41,58. These interdecadal changes have been attributed to the more prominent deepening of the thermocline in the eastern basin weakening the thermocline feedback during the EAN. However, as revealed in this study, meridional advection anomalies associated with the mean SST distribution also play an important role in the formation of the EAN. Whether the low-frequency changes in the mean state SST pattern in the tropical Atlantic also contribute to the recent weakening of the EAN deserves further investigation. Additionally, how changes in the tropical Atlantic mean conditions under global warming may modulate the behaviors of the two Atlantic Niño types also warrants further discussions.

Methods

Observational data

The monthly SST data from 1979 to 2021 from the Hadley Centre Sea Ice and Sea Surface Temperature (HadISST)⁵⁹, daily SSH data from 1993 to 2021 from Copernicus Marine Environment Monitoring Service (CMEMS)⁶⁰, monthly 850hPa wind and surface heat flux data during 1979 to 2021 from European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis 5 (ERA5)⁶¹ are used to examine the oceanic and atmospheric variations associated with the two types of Atlantic Niño.

As the primary analysis dataset, monthly oceanic data from ECMWF Ocean Reanalysis System5 (ORAS5)⁶² are used to calculate the ocean mixed layer heat budget and examine the oceanic changes in the study, including SSH, ocean temperature, mixed layer depth, horizontal velocity, and thermocline depth defined as the 20˚C isotherm (z20). The product has a horizontal resolution of 0.25°× 0.25°, covering the period from 1979 to 2021. To validate the accuracy of the heat budget analysis, oceanic data from the Simple Ocean Data Assimilation (SODA) version 3.12.2⁶³ and National Centers for Environmental Prediction (NCEP) Global Ocean Data Assimilation System (GODAS)⁶⁴ are also used to perform the ocean mixed layer heat budget, including monthly temperature, mixed layer depth, and horizontal velocity from 1980 to 2017 and 1980 to 2021, respectively. The net surface heat flux data exhibit considerable discrepancies in different data sets. Hence, monthly surface net heat flux data from 1984 to 2009 from the Objectively Analyzed air-sea Fluxes for the global oceans (OAFlux)⁶⁵ are analyzed for comparison. Results show that the ERA5 data exhibits a higher consistency with OAFlux compared with other reanalysis datasets (Fig. S11). Therefore, the net heat flux data used in this study are all from ERA5.

The monthly anomalies are calculated by subtracting the climatological mean annual cycle from 1979–2023. All anomaly fields have been linearly detrended. Since we focus on interannual timescales, a Butterworth filter is applied to all anomaly fields to remove low-frequency signals with periods longer than 10 years. The statistical significance of results is tested with the two-sided student’s t-test.

Climate indices

The indices of the two types of Atlantic Niño are obtained through an empirical orthogonal function (EOF) analysis of SSTA over 60°W-20°E, 10°S-10°N, following Zhang et al.⁴¹. The first EOF mode (EOF1) represents the whole Atlantic Niño pattern, with warming in the central and eastern Atlantic. The third EOF mode (EOF3) describes the zonal shift of the warming center between the CAN and EAN. The CAN index and EAN index are then defined as $({PC}1-{PC}3)/\sqrt{2}$ and $({PC}1+{PC}3)/\sqrt{2}$. The central Atlantic Niño (1972, 1974, 1996, 2008, 2010, and 2021) and central Atlantic Niña (1992, 1997, 2005, and 2012) events are defined as years when the absolute value of the CAN index exceeds one standard deviation. Similar to the EAN (1984, 1987, 1991, 1995, 1998, and 1999) and eastern Atlantic Niña (1972, 1980, 1982, 1983, 1994, 2011, 2013, and 2015) events.

A similar criterion was applied to identify the two types of ENSO events, except for a different region to conduct the EOF analysis (110°E-70°W, 30°S-30°N)⁶⁶. The EOF1 describes the canonical El Niño pattern, characterized by warming in the central and EP. The EOF2 depicts an east-west SSTA dipole between the equatorial central and EP. The EP-ENSO and CP-ENSO index are defined as $({PC}1-{PC}2)/\sqrt{2}$ and $({PC}1+{PC}2)/\sqrt{2}$. Five EP El Niño (1972, 1982, 1991, 1997, and 2015) and six EP La Niña (1970, 1980, 1985, 1995, 1996, and 2017) events are identified with a threshold of ±0.8 of the EP- ENSO index. Similar to the CP El Niño (1977, 1979, 1986, 1987, 1990, 1992, 1994, 2009, and 2014) and CP La Niña (1973, 1975, 1988, 1998, 1999, 2000, 2007, 2008, 2010, 2011, and 2020) events. 0.8 was chosen as the threshold value to obtain a larger sample size for ENSO.

Mixed layer heat budget

To investigate the formation mechanisms of the two types of Atlantic Niño, we calculate the mixed layer heat budget expressed below:

$$\left\langle \frac{\partial T}{\partial t}\right\rangle =\frac{{Q}_{0}-{Q}_{-h}}{{\rho }_{0}{c}_{p}h}-\left\langle u\frac{\partial T}{\partial x}+v\frac{\partial T}{\partial y}\right\rangle +{Vertical\; processes}$$

Here, $T$ is the temperature of seawater. $\left\langle \cdot \right\rangle =\frac{1}{h}{\int }_{-h}^{0}\cdot {dz}$ denotes the vertical mean within the mixed layer. ${\rho }_{0}$, ${c}_{p}$, and $h$ are the density (assumed to be a constant, 1025 kg m⁻³), specific heat capacity (3940 J kg⁻¹ °C ⁻¹), and the mixed layer depth. The mixed layer depth is defined as the depth where density exceeds surface density by 0.03 kg m⁻³. $u$ and $v$ donate speed of zonal and meridional currents. The surface net heat flux includes two parts: ${Q}_{0}$ denotes the net heat flux at the ocean surface, and ${Q}_{-h}$ denote the penetrative loss of shortwave radiation through the mixed layer, ${Q}_{-h}={Q}_{{sw}}[R{e}^{-\frac{h}{{\gamma }_{1}}}+(1-R){e}^{-\frac{h}{{\gamma }_{2}}}]$⁶⁷. $R$ = 0.62, ${\gamma }_{1}$ = 1.5 m and ${\gamma }_{1}$ = 20 m are attenuation depths. ${Q}_{{sw}}$ is shortwave radiation. The vertical process term is the residual of the heat budget, including vertical advection, entrainment, vertical and horizontal diffusion, and other unresolved processes. $\left\langle u\frac{\partial T}{\partial x}+v\frac{\partial T}{\partial y}\right\rangle$ represents the horizontal advection term, which can be further decomposed into three components:

$$u\frac{\partial T}{\partial x}={u}^{{\prime} }\frac{\partial \bar{T}}{\partial x}+\bar{u}\frac{\partial {T}^{{\prime} }}{\partial x}+{u}^{{\prime} }\frac{\partial {T}^{{\prime} }}{\partial x}$$

$$v\frac{\partial T}{\partial y}={v}^{{\prime} }\frac{\partial \bar{T}}{\partial y}+\bar{v}\frac{\partial {T}^{{\prime} }}{\partial y}+{v}^{{\prime} }\frac{\partial {T}^{{\prime} }}{\partial y}\,$$

These three components denote ${U}^{{\prime} }\bar{T}$ term, ${\bar{U}T}^{{\prime} }$ term, and the nonlinear advection term, respectively. The overbar and prime denote climatological mean values and interannual variability, respectively.

The contribution of the vertical processes is also more explicitly quantified from the SODA and GODAS in Figs. S1 and S2, and the calculation is as follows:

$${Vertical\; process}=-{w}_{-h}^{{\prime} }\frac{\partial {\bar{T}}_{-h}}{\partial z}-{\bar{w}}_{-h}\frac{\partial {T}_{-h}^{{\prime} }}{\partial z}+\frac{\partial {T}_{-h}^{{\prime} }}{\partial z}\left(\frac{\partial h}{\partial t}+{u}_{-h}\frac{\partial h}{\partial x}+{v}_{-h}\frac{\partial h}{\partial y}\right)$$

Here, $w$ denotes the vertical velocity of seawater. ${w}_{-h}$ and $\frac{\partial {T}_{-h}}{\partial z}$ denote the vertical velocity and vertical temperature gradient at the bottom of the mixed layer. The first term on the right-hand side of the equation represents Ekman feedback, while the remaining two terms represent the thermocline feedback (including the entrainment).

Model experiments

To explore the atmospheric response to the initial warming of the two Atlantic Niño types, we perform atmospheric general circulation model experiments using ECHAM4.6 from the Max Planck Institute for Meteorology in Hamburg⁶⁸. The horizontal resolution of the model is ~2.8°, with 19 vertical levels. We performed three sets of numerical model experiments. The control run is forced by climatological mean SST from HadISST from 1970 to 2021. In the two sensitivity experiments, composites of 12-month SSTAs in the tropical Atlantic Ocean between 22.5°S and 7.5°N during the CAN and EAN are added to the monthly SST climatology. Buffer zones were also applied in the northern and southern boundaries. Each numerical experiment is integrated for 42 years, and the first 4 years of each experiment are discarded to ensure that the model has achieved its statistical equilibrium state.

To assess the contributions of wind anomalies in the eastern Atlantic during the CAN, a linear ocean model with a horizontal resolution of 0.25° is used in this study⁶⁹. Two sets of experiments are conducted, each forced by surface wind stress anomalies during AM and JJ of the CAN events (Fig. 2b, c). Each set consists of three experiments driven by the whole equatorial Atlantic (100°W to 20°E, 7.5°S to 7.5°N), equatorial western Atlantic (100°W to 7.5°E, 7.5°S to 7.5°N), and eastern Atlantic (7.5°E to 20°E, 7.5°S to 7.5°N) wind stress anomalies, respectively. The numerical experiments are integrated for 10 years, with the last 5 years used for analysis in this study.