Introduction

The tropical Atlantic and Pacific Oceans exhibit similarities in both climate and its interannual variability. Their eastern equatorial regions are characterized by strong upwelling1,2 of cold subsurface water rich in dissolved inorganic carbon making the regions abundant sources of carbon to the atmosphere3,4,5. The two basins also experience similar equatorial pattern of SST variability modes that dominate at interannual timescale2,6,7,8,9: the Atlantic Niño and the El Niño/Southern Oscillation (ENSO) in the Pacific. These patterns of variability present two distinct phases that have impacts on climate and ecosystems both locally and globally2,10,11 - the Atlantic Niño and El Niño are characterized by reduced upwelling and warmer SST, while the opposite occurs during Atlantic Niña and La Niña.

In contrast to the Pacific3,4, knowledge is lacking on how this pattern of variability impacts the CO2 fluxes in the equatorial Atlantic. Since the equatorial Atlantic represents a major CO2 outgassing system and a region of early anthropogenic signal emergence in the global ocean12, understanding how Atlantic Niño variability modulates the sea-air CO2 fluxes is crucial to better constrain present and future carbon budget and anthropogenic climate change. Here, we investigate how the CO2 flux responds to the equatorial Atlantic interannual variability by identifying the CO2 flux anomalies associated with the Atlantic Niño and compare them with the respective anomalies during El Niño in the equatorial Pacific, using observational SST and sea-air CO2 flux data (see Methods).

Subsequently, we quantify the contribution of SST, sea surface salinity (SSS), dissolved inorganic carbon (DIC), and total alkalinity (ALK)13,14 to driving the CO2 partial pressure (pCO2), and consequently the CO2 flux anomalies in the equatorial Atlantic using observations and simulations of an IPCC-class model - the Norwegian Earth System Model version 2 (henceforth, NorESM215,16); this model was selected because of its good performance in the equatorial Atlantic and as it is one of the better performing CMIP617 ESMs in the equatorial Atlantic (see Method and Supplementary Table S1, Supplementary Figs. S1 and S2). For each driver of pCO2, we elucidate its mechanism using NorESM2 because the usage of observational data of ocean carbon parameters like dissolved inorganic carbon and total alkalinity is limited. A brief model-data comparison of NorESM2 SST and sea-air CO2 flux are given in Supplementary Fig. S1 and the Method section.

We show composite differences of governing parameters to emphasize the anomalies associated with the Atlantic Niño (i.e., the average difference between Atlantic Niño and Niña conditions; see Method). According to previous studies8,18,19,20,21, the Pacific and Atlantic Niño variability have a few different mechanisms, patterns and periodicities (e.g., ENSO Modoki, canonical Atlantic Niño, and early/late onset of Atlantic Niños). We focus on the climatological peak of the Pacific and Atlantic Niños in November-December and June-July, respectively and investigate the most common CO2 flux response to the mean climate variability modes without any detailed categorization of events.

Results

CO2 flux anomalies in the tropical Pacific and Atlantic

The eastern equatorial Pacific is predominantly characterized by climatological cold SST, known as the cold tongue, and CO2 outgassing, which are associated with the steady upwelling of cold and DIC-rich deep water3,4,5 (Fig. 1a, 0.43 Pg C yr−1 over the shown domain). The equatorial Atlantic is also a CO2 outgassing system, but its spatial distribution does not closely follow the cold tongue, which is strongest from June to July (Fig. 1c, 0.075 Pg C yr−1 over the shown domain). The spatial correlation coefficients between the observed climatological CO2 fluxes and SST (Fig. 1a, c) are −0.69 and −0.29 (both are significant at 95% level) in the equatorial Pacific and Atlantic, respectively.

Fig. 1: Observed sea-air CO2 flux and sea surface temperature in the tropical Pacific and Atlantic.
figure 1

a, c Observed climatology (1990–2015) and (b, d) composite anomalies of sea surface temperature (SST, black contour, degree C) and sea-air CO2 flux (color, mol m−2 yr−1) in the tropical Pacific (November-December) and Atlantic Ocean (June-July) between warm and cold events. Gray dots denote the air-sea carbon flux anomalies that are statistically significant at the 90% level estimated by the Student’s t test. Note that the sea-air carbon flux anomalies are scaled by NINO3.4 and Atlantic-3 indices anomalies (units in mol m−2 yr−1 degree C−1) between Niño and Niña events for the Pacific and Atlantic, respectively. The observed data of CO2 flux and SST are from MPI-SOM FFM23,42 and OISST41.

Full size image

During the peak of El Niño phase from November to December, the CO2 outflux is suppressed3,4 in most parts of the central to eastern Pacific mainly due to less DIC upwelled from the deeper layers3,4 (Fig. 1b, −0.270 Pg C yr−1 area-integrated over the shown domain). Interestingly the CO2 flux anomalies during Atlantic Niño differ remarkably from those in the Pacific: in the Atlantic, the strongest anomalies occur in the western basin away from the core of SST anomalies in the central to eastern basin (Fig. 1d, −2.557 Tg C yr−1 area-integrated over the shown domain), while the SST and CO2 flux anomalies nearly co-located in the Pacific (Fig. 1b). In the western equatorial Atlantic, the CO2 outgassing is suppressed during the Atlantic Niño, but the opposite pattern is observed in the central basin. The tendencies in the western Atlantic and in the central Pacific are similar, but they differ markedly in the east. The event selection of the Pacific and Atlantic Niño/Niña in this study is shown in Supplementary Fig. S3.

NorESM2 can reproduce the observed spatial peculiarities of sea-air CO2 flux associated with the Atlantic Niño (Figs. 1d and 2b). This indicates that the model reasonably reproduces the mechanisms giving rise to the anomalous CO2 fluxes, even though the Atlantic Niño tends to be overestimated (contours in Fig. 2 and Supplementary Fig. S1). In addition, the observation shows outgassing, while NorESM2 (as well as MRI-ESM2-0 and CESM2-WACCM in Supplementary Fig. S2b, c) shows ingassing, adjacent to the Amazon River mouth (2°N-6°N and 50°W-48°W). This might be because the observation-based CO2 flux estimate we use has insufficient resolution to properly capturing the effect of Amazon River plume on pCO222,23. According to previous studies24,25,26, the Amazon River plume is a large CO2 sink around the Amazon River mouth and the influence of the plume extends eastward along the North Equatorial Countercurrent. In the NorESM2 simulation there is an extension of weak CO2 sink and source to 40°W-35°W at 4°N to 6°N (Fig. 2a) and, to some extent, NorESM2 is able to capture the importance of the Amazon River plume in the western basin of the equatorial Atlantic. However, the CO2 sink at 10°N is not well reproduced as in NorESM2 as in the previous observational study25. This is one of the limitations of the NorESM2. Because there is no observational data with sufficient spatial and temporal coverage required to investigate the drivers of pCO2 change associated with the Atlantic Niño, the simulation of NorESM2 will be used to elucidate the driving mechanisms of the sea-air CO2 flux anomalies.

Fig. 2: Simulated sea-air CO2 flux and sea surface temperature (SST).
figure 2

Same as Fig. 1c, d, but for NorESM2 simulation, a climatology and b composite anomalies. Composite anomalies are scaled by the Atlantic-3 index composite anomaly between Niño and Niña events. The dashed red rectangles denote western and central basins defined in this study.

Full size image

Drivers of pCO2 anomalies

The pCO2 anomalies predominantly regulate the CO2 flux patterns during Atlantic Niño (see the close resemblance in Fig. 2b and Supplementary Fig. S4b). On the other hand, the surface wind anomalies seem to be less responsible for forming the sea-air CO2 flux patterns according to its climatology and anomaly distribution (Supplementary Fig. S5). To determine the governing mechanisms of the CO2 flux anomalies, we decompose the pCO2 anomalies into its main drivers: SST, SSS, ALK and DIC, applying the seawater inorganic carbonate system calculation (CO2SYS27) on the NorESM2 simulation. Supplementary Table S2 lists the combination of input variables for the CO2SYS calculation in order to decompose the relative contributions of each driver on the total pCO2 anomaly. Here, the total pCO2 anomaly (pCO2F, where F denotes four contributing variables) is compared with SST-, SSS-, ALK-, and DIC-driven pCO2 anomalies (denoted as pCO2T, pCO2S, pCO2A, and pCO2D, respectively). The climatology and anomalies of CO2SYS-estimated pCO2F are nearly identical with the respective prognostic NorESM2 pCO2 (Fig. 3a, b and Supplementary Fig. S5a, b). The climatological pCO2F has a maximum in the western basin between 4°S and the equator where the maximum CO2 outflux occurs (Figs. 3a and 2a). Furthermore, the pCO2F anomalies also exhibit a consistent zonal dipole structure as seen in the sea-air CO2 flux anomalies. This reaffirms that the spatial CO2 flux anomaly pattern is, to a first order, governed mainly by the pCO2 variability.

Fig. 3: Drivers of variability in partial pressure of CO2.
figure 3

a Climatology of partial pressure of CO2 (pCO2F) in June-July. b–f the composite difference of the different contributions to pCO2 calculated between the Atlantic Niños and Niñas: b four variables, c sea surface temperature, d sea surface salinity, e salinity-nomalized total alkalinity, and f salinity-normalized dissolved inorganic carbon, respectively. Gray dots denote 95% of significance level.

Full size image

The positive SST anomaly during the Atlantic Niño (Fig. 2), induced by the reduced upwelling of cold water, has a positive effect on the pCO2 anomaly through the reduced solubility associated with warmer surface waters (Fig. 3c). It reaches an amplitude of 85 μatm exceeding that of pCO2F, and is found mainly in the central to eastern basin, consistent with the SST changes presented in the previous section. In addition, a weak and negative pCO2T anomaly is found in the northwestern basin, in particular off the equator.

During the Atlantic Niño, negative pCO2S anomalies, related to negative salinity (i.e., freshening) anomalies (Supplementary Fig. S6e, f), dominate the western basin with an amplitude of −25 μatm similar to the pCO2F anomalies (Fig. 3d). These negative salinity anomalies are induced mainly by anomalously higher precipitation in the large part of western basin (Supplementary Fig. S6a–d) during the Atlantic Niño. The high precipitation extends to the eastern equatorial Atlantic area21 and NorESM2 simulations can capture its geographical distribution properly. Note that the precipitation anomaly associated with the Atlantic Niño is overestimated compared to the observation and its location shifts slightly southward (Supplementary Fig. S6b, d) partially because of the model’s SST bias. In addition, a reduced advection of tropical high-saline waters (this advection is a crude estimate computed with monthly SSS and ocean current output) by the westward equatorial current28 contributes to the lower salinity off the northern coast of the South American Continent. However, in NorESM2, the horizontal salinity advection anomalies are statistically less significant around the Amazon River mouth (Supplementary Fig. S6h). However, between 40°W-30°W at 2°N-4°N, the freshening due to the horizontal salinity advection is significant (Supplementary Fig. S6j) and therefore, there may contribute to the Amazon River plume in freshening the western basin until 30°W. This salinity advection anomaly is mainly due to the eastward anomalies of North Brazil Current and part of South Equatorial Current (Supplementary Fig. S6h, j). The North Equatorial Countercurrent anomalies are westward around 6°N (Supplementary Fig. S6h) in agreement with a previous study29 and are therefore not responsible for freshening the western basin during Atlantic Niño events in our simulation. On the other hand, Amazon River discharge is reduced during Atlantic Niño event in NorESM2 and this does not contribute to the freshening anomaly (Supplementary Fig. S7). The potential contribution of Amazon river discharge and plume to interannual variability of pCO2 anomalies requires further analysis with higher resolution ocean models and observations. In the western basin, the negative latent heat flux anomalies also tend to freshen the ocean surface (Supplementary Fig. S6k–n). However, the positive latent heat flux anomaly in the central basin does not result in any positive SSS anomaly. Therefore, the latent heat flux is not a main driver for the salinity change in this region.

The contribution of carbonate species composition, ALK and DIC, to pCO2 is stronger than the SSS- and SST-induced solubility changes30. Similar to the pCO2S, the pCO2A and pCO2D anomalies dominate in the western basin (Supplementary Fig. S8a, b), indicating that upwelling- and freshwater-induced changes (through dilution and concentration) in surface ALK and DIC dominate the pCO2A and pCO2D. As the ALK and DIC contributions to pCO2 change is almost identical with opposite sign30, their associated contributions almost cancel out in the western basin. The pCO2 anomalies associated with salinity-normalized ALK (pCO2A_S) and DIC (pCO2D_S) more clearly show their contributions associated with the upwelling modification (Fig. 3e, f). The salinity-normalized anomalies (see Methods) exclude the role of surface freshwater fluxes and reflect the role of physical and biological activities on surface pCO230. In the central basin, there is a weak positive pCO2A_S anomaly with an amplitude of 10 μatm that is statistically detectable, and a negative pCO2D_S anomaly with an amplitude of 70 μatm. The net effect of these anomalies is a reduction in the surface pCO2 during the Atlantic Niño.

The anomalies associated with pCO2A_S and pCO2D_S reflect the thermocline modulation associated with equatorial upwelling. In the subsurface between 60 and 80 m depth, there are ALK and DIC anomalies with a see-saw pattern from western to central/eastern basin, with the DIC anomaly being more remarkable (Supplementary Fig. S9). Because the vertical gradient of the DIC is much steeper than that of ALK (Supplementary Fig. 10a, c), the DIC is more sensitive to vertical displacement of the thermocline. The pCO2A_S and pCO2D_S show large anomalies in the western basin with opposite signs (Fig. 3e, f). As for the central basin, these anomalies can also be associated with the thermocline displacement. For example, during the Atlantic Niño, the climatological thermocline zonal tilt (shallower in the east) is reduced. Under this condition, the western basin experiences an upwelling anomaly and an eastward ocean surface current anomaly as indicated in Supplementary Figs. S9 and S6h, i. Consequently, the combined dynamical and biological impacts on the pCO2A_S and pCO2D_S result in a di-pole pattern (Fig. 3e, f). However, we note that the contribution of biological processes (photosynthesis and respiration) is acting in opposition to the upwelling-induced DIC change31 and therefore, the pCO2D_S is dominated by the dynamical process. The anomalies in the western basin due to ALK and DIC approximately offset each other as indicated by pCO2AD (Supplementary Fig. S8c).

Discussions

Figure 4 summarizes the mechanism of sea-air CO2 flux anomalies due to the Atlantic Niño deduced in this study (for the Atlantic Niña, the opposite pattern occurs). In the central equatorial Atlantic, the upwelling of DIC is reduced during an Atlantic Niño, similar to the equatorial Pacific. However, the combined ALK- and SST-induced pCO2 anomalies that also accompany the upwelling of the cold ALK-rich deep waters is larger than the DIC-induced pCO2 anomalies mainly due to the higher carbon buffer capacity in the Atlantic (as compared to the Pacific). Consequently, the prevailing pCO2 anomaly is positive during the Atlantic Niños. In the western Atlantic basin, the CO2 flux and surface ocean pCO2 anomalies are opposite in sign to those in the central basin and are induced mainly by SSS changes. Additionally, the SST anomalies in the western basin during the Atlantic Niño tend to be opposite to that in the central basin even though its magnitude is considerably small. These opposing SST anomalies induce opposite surface pCO2 anomalies in the western basin to the central basin. The large-scale salinity-induced pCO2 anomaly is primarily attributed to the freshwater anomalies associated with the precipitation around the Brazilian coast. While it is less significant and limited in NorESM2, there is an indication that the horizontal advection of salinity modulated by the eastward anomaly of ocean current around the Amazon River mouth and in the western basin25. Further investigation on the role of the Amazon plume will be necessary.

Fig. 4: Schematic of the proposed mechanism of sea-air CO2 flux response to Atlantic Niño.
figure 4

Air-sea carbon flux of (a) climatology and (b) response to the Atlantic Niño suggested by this study.

Full size image

The dominant role of SST and SSS in driving the sea-air CO2 flux anomaly have been reported in a case study of warm SST anomaly event, observed from February to May in 2010, in the north tropical Atlantic, slightly to the north of the region in this study32,33,34. This warm SST anomaly event, however, is more related to the Atlantic Meridional Mode during spring while our investigated SST anomaly represents the Atlantic Zonal Mode during summer. Nevertheless a connection between these two climate modes through ocean dynamics is suggested35.

While we identify Atlantic Niño events based on the climatological peak of variability and assume canonical dynamics, there are other mechanisms causing the Atlantic Niño18,19,20,21. As one non-canonical example, equatorial warm events form due to the advection of off-equatorial subsurface temperature anomalies from the north19. In these events the upwelled sea water would contain the chemical properties of the off-equatorial north Atlantic Ocean. This may lead to different contributions of DIC and ALK to the pCO2 anomalies than that analyzed in our more generalized Atlantic Niño.

Our results indicate that the Atlantic Niño and Pacific El Niño induce very different sea-air CO2 flux anomalies (Fig. 1). This difference is most likely a result of different carbonate composition of the Pacific and Atlantic deep waters. Supplementary Fig. S10. The vertical structure of DIC and ALK in both the Pacific and Atlantic are to first order quite similar. This indicates that the upwelling-induced DIC and ALK contributions to pCO2 anomaly would be similar in both basins. Different responses in the pCO2 could result from differences in the carbon buffering capacity, which is proportional to the carbonate ion (CO3−2, which can be approximately ALK minus DIC36) concentration. The CO3−2 can be approximated from ALK minus DIC36. Supplementary Fig. S11 shows the surface carbonate ion concentration in observational data and as simulated by the NorESM2. The tropical Pacific has a lower buffer capacity than the tropical Atlantic, primarily due to the higher subsurface alkalinity and surface evaporation in the latter. Since the pCO2 change due to DIC (dpCO2/dDIC) is larger in the sea water with lower buffer capacity36, the pCO2 is more sensitive to the same DIC change in the tropical Pacific than in the tropical Atlantic. Supplementary Fig. S12 depicts the composite differences in CO2 flux and each contribution to pCO2 between Pacific El Niño and La Niña from NorESM2 simulation during November and December. In the central tropical Pacific, the salinity contribution is much less than the temperature (Supplementary Fig. S12b, c). Similar to the tropical Atlantic, the temperature anomaly enhances the upward CO2 flux. The DIC contribution is much more dominant than the ALK contribution and the pCO2D_S anomaly (−180 μatm) is much more pronounced than that in the tropical Atlantic (−70 μatm, Fig. 3f). In the Pacific, the DIC contribution overwhelms the temperature contribution resulting in the reduction in CO2 outflux at the equator (Supplementary Fig. S12a). On the other hand, the CO2 flux anomaly is positive off the equator (Supplementary Fig. S12a), which is induced by the temperature. This positive anomaly is not seen in the observations and is common across CMIP6 models that simulate the observed negative anomaly in the equator37. In short, the CO2 flux anomaly in the tropical Pacific can primarily be attributed to this DIC-driven pCO2 change at the equator3.

Here, we have revealed the unique pattern and underlying mechanism of CO2 flux anomalies triggered by the Atlantic Niño, applying an observational-validated simulation from the latest generation of the Norwegian Earth System Model (NorESM215,16). We showed that SSS and SST are the main drivers of the distinct anomaly pattern, and that upwelling of DIC only plays a secondary role. This pattern differs remarkably from that observed in the Pacific during ENSO events, where upwelling of DIC plays a major role. The smaller contribution of upwelling-induced DIC in the Atlantic compared to the Pacific can be attributed to the higher carbon buffer capacity of the Atlantic waters. These results indicate that, given the different governing mechanisms, the tropical Atlantic CO2 flux responses to future climate change will also be different from that in the tropical Pacific37.

This study focused on only Atlantic Niño events based on the geographically-fixed SST variability indices. However, as previous studies have shown, there are similar, but different types (also mechanisms) of the Niño events18,19,20,21. One of the desirable future studies is to investigate how different the CO2 flux response is to the distinct types of the Atlantic Niños events, particularly when long term observational-based CO2 flux estimates become available. Sustained long-term observations of salinity and water-column carbonate parameters would be crucial to reaffirm our understanding of how long-term climate change signals may affect the CO2 flux variability.

According to recent studies, the Atlantic Niño variability in the CMIP5 models tends to be reduced under global warming38,39,40. This might imply that the CO2 flux response to the Atlantic Niño variability can be modified in the warmer climate. As suggested in previous studies, the amplitude of SST anomaly associated with the Atlantic Niño variability is reduced mainly due to less active SST-thermocline feedback38 and suppressed surface wind variability40. In addition, the precipitation variability associated with the Atlantic Niño appears to be reduced39. All these changes would influence the pCO2 and CO2 flux and further investigation with inter-model comparison is needed.

Methods

Observational data and model

Optimum Interpolated Sea Surface Temperature41 (OISST, 1990–2019) data is used for observational SST. For the sea-air CO2 flux, a global observation-based gridded data23,42 (1990–2015) is used. The observed data of ocean dissolved carbon species are obtained from GLODAP version 243. The observed atmospheric data is from ERA5 reanalysis44 for momentum flux, precipitation from GPCP45, and latent heat flux from TropFlux46. The Norwegian Earth System Model version 2 (NorESM215,16) is one of the state-of-the-art Earth system models that participate in the CMIP6. The configuration used in this study is 2 degree for the atmospheric and land surface components and 1 degree for the ocean and the sea-ice components (NorESM2-LM). The marine biogeochemical component, iHAMOCC16, has the same resolution as the ocean component. In this study, we use the 3 ensemble-member data from 1990 to 2014 of the historical run. Because of large model spread and uncertainty, we need to assess the performance of NorESM2-LM comparing to other ESMs. Therefore, we additionally use 12 CMIP6 models: CESM247, CESM2-WACCM48, CMCC-ESM249, CNRM-ESM2-150, GFDL-CM451, GFDL-ESM452, IPSL-CM6A-LR53, MIROC-ES2L54, MPI-ESM1-2-HR55, MPI-ESM1-2-LR56, MRI-ESM2-057, and UKESM1-0-LL58 as summarized in Supplementary Table S1. The references for these 12 models are also given in Supplementary Table S1 for more details of model configurations.

A brief model evaluation

Challenges in accurately reproducing the observed SST pattern in the equatorial Atlantic SST are common across state-of-the-art Earth system models59,60,61. As summarized Supplementary Table S1, the NorESM2 and 12 CMIP6 models are assessed in terms of SST and sea-air CO2 flux. The statistical metrics for this assessment are root mean squared error and spatial correlation with respect to the observations. For climatology of SST and CO2 flux and CO2 flux anomaly, the statistical metrics are calculated over 6°S-2°N and 45°W-0 and 6°S-5°N and 45°W-10°E boxes, respectively. The box for the climatology covers the climatological peak of the Atlantic Cold Tongue and CO2 outgassing and the box for CO2 anomaly covers the entire di-pole anomaly. First we assess the capacity of 13 CMIP6 models in capturing the observed patterns of climatological SST, sea-air CO2 flux, and their responses to the Atlantic Niño variability. As one of recent studies shows60, the state-of-the-art ESMs still exhibit the large SST bias in the tropical Atlantic. However, there are also less biased models (i.e., low RMSE and high correlation with observations: NorESM2, CMCC-ESM2, MRI-ESM2-0, IPSL-CM6A-LR, and UKESM1-0-LL). These models with less SST bias models tend to better reproduce the climatology of sea-air CO2 flux distribution (e.g., NorESM2 and MRI-ESM2-0; Supplementary Table S1). Both NorESM2 and MRI-ESM2-0 can also capture the CO2 flux response to the Atlantic Niño variability. While UKESM1-0-LL is also able to capture the anomaly pattern well, its climatology of SST and CO2 is relatively poorly simulated (Supplementary Table S1). NorESM2 and MRI-ESM2-0 can simulate the climatological CO2 flux maximum in the western basin while reproducing more realistic Atlantic Cold Tongue (ACT, Fig. 2 and Supplementary Fig. S1). Contrastingly, the CO2 flux maximum in CESM2-WACCM locates more eastward (Supplementary Fig. S1) and the development of ACT is not well captured. These SST and CO2 biases are also seen in other poorly performing models (not shown). The CO2 flux responses in MRI-ESM2-0 is also realistic with a di-pole structure along the equator. CESM2-WACCM has a mono-pole anomaly of less CO2 outflux overlapping the warm SST anomaly along the equator (Fig. Sd). This assessment illustrates the high uncertainty in CO2 flux response to the Atlantic Niño variability simulated in CMIP6 models and indicates that models that better simulate the climatology of SST and CO2 flux tend to reproduce more realistic di-pole structure of CO2 response to Atlantic Niño events along the equator. NorESM2 is one of the best models and there are 3 ensemble members available for more robust results. This gives us in total 75 years, providing us the possibility of having more than 10 Atlantic Niño/Niña events, which is enough to construct canonical Atlantic Niño events.

According to this assessment, NorESM2 and MRI-ESM2-0 are the best models to reproduce the CO2 flux climatology and its response to the Atlantic Niño variability. The inter-annual variability in the equatorial Atlantic is evaluated (Supplementary Fig. S1g). The observed variability is largest in June-to-July with a secondary peak in November to December. The NorESM2 and MRI-ESM2-0 are able to simulate successfully the summer peak in June-to-July, even though its amplitude is overestimated. The imperfect seasonality of variability is also a common issue among the state-of-the-art ESMs62 and we consider the NorESM2 to be a good representative of CMIP6 models in terms of simulating the inter-annual climate variability in the tropical Atlantic.

Definition of Atlantic Niños and Niña events

The June-July mean Atlantic-3 index is defined as the SST averaged over 20°W to 0° and 3°S and 3°N. The Atlantic Niño (Niña) event is defined when the detrended Atlantic-3 index is more (less) than plus (minus) one standard deviation of the Atlantic-3 index. For the model, the Atlantic Niño and Niña events have been defined by each ensemble member. All the composite plots are based on this selected Atlantic Niño and Niña events and their statistical significance are estimated by Student’s t test at the 90% of level. The anomalies of each variable are defined as the difference between the anomaly composite of Atlantic Niño and Atlantic Niña.

For the composite of pCO2, the detrended data is also used to remove the long-term trend. The gray dots in Fig. 3 denote 95% of significance level estimated by Student’s t-test.

Decomposition of pCO2 into drives

The sea-air CO2 flux is governed by surface wind, the solubility parameter, and the CO2 partial pressure (pCO2) difference between ocean-atmosphere interface. The surface ocean pCO2, in turn, is a function of the SST, SSS, DIC, and ALK. To quantify their contributions to the change in surface pCO2, we utilized an open source MATLAB script of CO2SYS27 that allows us to estimate the state of the carbonate system of oceanographic water samples and data. For the pCO2 calculation, we used monthly SST, SSS, and surface concentrations of total alkalinity and dissolved inorganic carbon and climatology of silicate and phosphate as inputs. Note that, due to the nonlinearlity of the carbonate system, pCO2F is not the simple sum of pCO2T, pCO2S, pCO2A, and pCO2D.

Definition of salinity-normalized total alkalinity and dissolved inorganic carbon

Normalization of total alkalinity (ALK) and dissolved inorganic carbon (DIC) with respect to salinity at the model surface layer is given by

$${{{{{{\rm{ALK}}}}}}}_{{{{{{\rm{s}}}}}}}(x,y,t)={{{{{\rm{ALK}}}}}}(x,y,t)\times {{{{{{\rm{SSS}}}}}}}_{{{{{{\rm{c}}}}}}}(x,y)/{{{{{\rm{SSS}}}}}}(x,y,t)$$
(1)
$${{{{{{\rm{DIC}}}}}}}_{{{{{{\rm{s}}}}}}}(x,y,t)={{{{{\rm{DIC}}}}}}(x,y,t)\times {{{{{{\rm{SSS}}}}}}}_{{{{{{\rm{c}}}}}}}(x,y)/{{{{{\rm{SSS}}}}}}(x,y,t)$$
(2)

Here, subscripts s and c denote salinity-normalized quantity and June-July climatology, respectively. x, y and t denote zonal and meridional grid point and time (June-July mean of each year), respectively.

Calculation of salinity horizontal advection

The salinity horizontal advection in Supplementary Fig. S6 is estimated as follows,

$${{\mbox{HADV}}}=-u\frac{\partial S}{\partial x}-v\frac{\partial S}{\partial y}$$
(3)

here, HADV denotes the salinity horizontal advection. u, v, and S are monthly outputs of NorESM2 for surface zonal current, meridional surface current, and sea surface salinity, respectively.