Weakened impact of the Atlantic Niño on the future equatorial Atlantic and Guinean Coast rainfall

The Guinea Coast is the southern part of the West African continent. Its summer rainfall variability mostly occurs on interannual timescales and is highly influenced by the sea surface temperature (SST) variability in the eastern equatorial Atlantic, which is known as the Atlantic Niño (ATL3). Using historical simulations from 31 General Circulation Models (GCMs) participating in the sixth phase of the Coupled Model Intercomparison Project (CMIP6), we first show that these models are able to simulate reasonably well the rainfall annual cycle in the Guinea Coast, with, however, a wet bias during 5 boreal summer. This bias is associated with too high mean summer SSTs in the eastern equatorial and south Atlantic regions. Next, we analyze the near-term, mid-term and long-term changes of the Atlantic Niño mode relative to the present-day situation, in a climate with a high anthropogenic emission of greenhouse gases. We find a gradual decrease of the equatorial Atlantic SST anomalies associated with the Atlantic Niño in the three periods of the future. This result reflects a possible reduction of the Atlantic Niño variability in the future due to a weakening of the Bjerkness feedback over the equatorial Atlantic. In a 10 warmer climate, an oceanic extension of the Saharan Heat Low over the North Atlantic and an anomalous higher sea level pressure in the western equatorial Atlantic relative to the eastern equatorial Atlantic weaken the climatological trade winds over the equatorial Atlantic. As a result, the eastern equatorial Atlantic thermocline is deeper and responds less to Atlantic Niño events. Among the models that simulate a realistic rainfall pattern associated with ATL3 in the present-day climate, there are 15 GCMs which project a decrease of the Guinean Coast rainfall response related to ATL3, and 9 GCMs which show no 15 substantial change in the patterns associated with ATL3. In these 15 models, the zonal wind response to the ATL3 over the equatorial Atlantic is strongly attenuated in the future climate. Similar results are found when the analysis is focused on the rainfall response to ATL3 over the equatorial Atlantic. There is a higher confidence in the reduction of the rainfall associated with ATL3 over the Atlantic Ocean than over the Guinea Coast. We also found a decrease of the convection associated with ATL3 in the majority of the models. 20


Introduction
The West African Monsoon (WAM) generally begins in mid-June and is characterized by a rapid shift of the rain band from the coastal areas to the Sahel region (Hansen, 2002;Sultan et al., 2005). As the rain band has moved northward, an upper level subsidence appears over the Guinea Coast. This causes the so-called "little dry season" in that area (Adejuwon and Odekunle, Table 1. CMIP6 models and members of historical and SSP5-85 simulations. (*) indicates models for which the sea surface height variable was not yet available at the time of this study. Models for which, on the one hand, the sea level pressure and the 10 m horizontal wind variables, and, on the other, the mixed layer depth variable were not used for the mean state change analyses are indicated by (**) and (***), respectively. The other variables used are: rainfall, sea surface temperature, 850 hPa specific humidity, 850 hPa and 200 hPa horizontal wind components.

CMIP6 model
Historical member SSP5-85 member ACCESS-CM2 r1i1p1f1 r1i1p1f1 ACCESS-ESM1-5 r1i1p1f1 r1i1p1f1  3 Evaluation of the performance of the GCMs in simulating the rainfall in Guinea Coast and SST in ATL3 region

Guinean coast rainfall: annual cycle, variability and JAS mean
This section is focused on the performance of CMIP6 models in representing the rainfall annual cycle and variability over the Guinea Coast (the GCB region) for the period 1985-2014. Much of the CMIP6 models overestimate the GCB rainfall magnitude throughout the year. The monthly biases of the rainfall averaged over the GCB area in the 31 GCMs lie between −3.4 and 160 5.5 mm · day −1 , relative to ERA5. The observed rainfalls show a bimodal structure, with two maxima in June and September respectively ( Fig. 1 (a)). The first (second) peak in the observations ranges from 7.3 (7.2) to 7.8 (8.4) mm · day −1 . Unlike the observations, the CMIP6 ensemble mean (EnsMean) depicts a plateau of 9 mm · day −1 in July and August, when much of the wet biases occur. This wet bias is present in 77 % and 61 % of the 31 CMIP6 models for July and August, respectively. In the other months, the CMIP6 EnsMean rainfall intensity is lower than in the reference (ERA5). The individual models hardly 165 simulate the bimodal structure of the Guinean Coast rainfall. GFDL-ESM4, MCM-UA-1-0 and MIROC6 for instance depict a bimodal structure with an incorrect timing and intensity of the two peaks.
The RMSE of the models averaged over the entire annual cycle ranges between 0.7 and 2.6 mm · day −1 . In addition, the spread of the modelled GCB rainfall annual cycle increases from April to October, when the interquartile range is between 1.5 and 3 mm · day −1 (Fig. A1 (a)). This spread is maximal in August and September. From November to March, the IQR of the 170 models is lower or equal to 1 mm · day −1 .
The CMIP5 models are also known to overestimate the rainfall in coastal areas of West Africa during the "little dry season" (Sow et al., 2020). Wainwright et al. (2019) have demonstrated that the misrepresentation of the July-August Guinean Coast rainfall in these models comes from the positive SST biases in the Atlantic Ocean, which strengthen the rising motions over the 175 Guinea Coast and increase the rainfall. Consistently, the CMIP6 ensemble mean exhibits a positive (negative) mean SST bias in the eastern (western) part of the tropical Atlantic, as depicted in Fig. 2. This figure also shows low-level anomalous northerlies that reinforce the convergence over the Guinea Coast. This result is also consistent with the conclusions obtained from Richter and Tokinaga (2020), who analyzed the CMIP6 pre-industrial control simulations. Beyond the biases in the magnitude of the where the correlation of the rainfall annual cycle in each CMIP6 model with that in ERA5 is above 0.9. In general, the JAS spatial structure of the West African rainfall is well reproduced by the models. The spatial correlation values with ERA5 lies between 0.7 and 0.95 (Fig. A2). However, the spatial variability of the rainfall intensity is underestimated in some of the models (e.g., GISS-E2-1-G and CAMS-CSM1-0), while it is exaggerated in others (e.g., GFDL-ESM4 and MIROC6).
3.2 SST in the ATL3 region: annual cycle, variability and JAS mean 200 The seasonal evolution of the SST in the Atlantic Niño region is similar in ERA5, HADISST, ERSST and COBE ( Fig. 3 (a)).
The annual cycle of the SST in this area shows a cold tongue which develops from April when the mean SST over the ATL3 region is about 29 • C and reaches its lowest value in August, around 24.5 • C. The CMIP6 EnsMean overestimates the SST in this region, with a bias that ranges from 0.7 • C (in October) to around 2 • C (in July). This warm bias in the seasonal cycle is present in more than 87 % of the 31 CMIP6 models whatever the season, and, is more pronounced in June-September. The 205 spread of the simulated ATL3 SST in the 31 GCMs ( Fig. A3 (a)) is particularly low (high) in December (September), with an IQR equals to 0.5 • C (0.9 • C). The correlation between the annual cycle values in each model with ERA5 ranges from 0.88 to 0.99 ( Fig. 3 (c)). This result indicates that the phasing of the SST in the ATL3 region is relatively well simulated by the models.
During the year, the observed variability of SST in the ATL3 area intensifies between April and September, with the maxi-210 mum of SST standard deviation occurring in June ( 0.7 • C for ERA5 and 0.6 • C for HADISST, ERSST and COBE, Fig. 3 (b)).
There is a second peak in November, of 0.4 • C in HADISST, ERSST and COBE, whereas it is observed one month later in ERA5. This corresponds to the winter Atlantic Niño which has greatly influenced the ENSO events and the rainfall in South America, mainly during decades in the mid-20 th century (Hounsou-Gbo et al., 2020). The first maximum of the multimodel mean of the SST standard deviation is delayed one month later compared to observations, whereas the second peak occurs at 215 the right time. Overall, the multimodel mean overestimates the SST STD in May-July and November-January, and this difference is at its maximum in June.
Throughout the year, the monthly SST standard deviation in the models covers a wide range of values, with different phasings ( Fig. 3 (d)). The spread of the SST STD annual cycle simulated by the 31 GCMs is particularly large in June, July and August, and the corresponding IQR is 0.2, 0.3 and 0.2 • C, respectively ( Fig. A3 (b)). In the other months, this spread is around 0.1 • C.   Note that the ocean surface warming associated with positive ATL3 phases is stronger in ERA5 than in HADISST, ERSST and COBE data.
cording to the sign of the pattern over the tropical Atlantic (the TAB1 region, see Table 2). In the first group, 14 models display a uniform sign of the SST regression coefficients over TAB1. In the other group, in addition to the positive regression coefficients in the eastern equatorial Atlantic, the models feature negative regression coefficients in the North Atlantic (e.g., ACCESS-ESM1-5), Southwest Atlantic (e.g., HadGEM3-GC31-LL), or both North and Southwest Atlantic (e.g., MIROC-ES2L). Except for GISS-E2-1-G, the spatial correlation between the ATL3 SST pattern in models and ERA5 over the TAB1 235 region ranges from 0.5 to 0.9 ( Fig. 4 (b)). The standard deviation of the spatial SST regression coefficients related to the ATL3 in the TAB1 region amounts 0.10 • C in ERA5, 0.09 • C in HADISST and 0.08 • C in ERSST and COBE. In 29 models, this spatial variability of the anomalous SST pattern is 0.01 to 0.19 • C higher than that in ERA5. In contrast, the spatial variability of MCM-UA-1-0 and GISS-E2-1-G SST regression coefficients is 0.02 • C lower than that in ERA5.  Table 2), 23 models depict a dipolar structure of the rainfall pattern, consistent with Mohino and Losada (2015). In these models, negative values of the rainfall regression coefficients are present north of the band of positive values, between 5 and 15 • N.

250
The sign-dependent average of the 1985-2014 rainfall regression coefficients over the Guinea Coast box is positive in the observations and ERA5 ( Fig. A5 (a)). Two groups of GCMs are considered, according to their sign-dependent average over the GCB area. The first group is termed GC+, and defines models with a positive sign-dependent average. It includes 24 models which are able to reproduce an increased rainfall associated with the Atlantic Niño positive phases ( Fig. 6 (a)). In contrast, there 255 are 6 models in the second group, termed GC-, which show negative correlations between the rainfall over the Guinea Coast and the Atlantic Niño ( Fig. 6 (b)). The GC-ensemble mean rainfall response over the GCB is weak in magnitude, with less model agreement compared to the GC+ ensemble mean. In both groups, the dipolar structure of the rainfall pattern is similar over the tropical Atlantic (the TAB2 area). However, this dipolar structure is not present in ERA5 ( Fig. 6 (d)), nor in the observations.

260
The SST and sea surface height imprints related to the Atlantic Niño mode are similar in the GC+ and GC-ensemble means   flux over the equatorial and South Atlantic is weaker in both GC+ and GC-, compared to ERA5 (Fig. 6 (p)).
Finally, during positive phases of the Atlantic Niño, over the Guinea Coast and the equatorial Atlantic, the atmospheric convection is more enhanced in the GC+ group than in GC-. This is indicated by the more pronounced wind divergence difference between the 200 hPa and 850 hPa levels in Fig. 6 (q)-(r). Moreover, in GC+ and GC-, the oceanic areas where the atmosphere 275 is destabilized are located 5 • south of their positions in ERA5 ( Fig. 6 (t)). This explains why the positive rainfall anomalies associated with the positive phases of the Atlantic Niño in the GCMs are located south of the obtained position in ERA5. In conclusion, the combination of a large moisture flux from the equatorial Atlantic toward the Guinea Coast, and a more destabilized atmosphere over the Guinea Coast leads to an enhanced rainfall response to ATL3 in the GC+ models compared to GC-models.

280
Over the West African region, the latitudinal position of the maximum of the rainfall regression coefficients averaged between 20 • W and 10 • E varies in the different observational data and ERA5. This position is 4.5 • N in CMAP-HADISST, 5.5 • N in CRUTS403-COBE and 6.5 • N in GPCP-ERSST and ERA5. 75 % of the 31 GCMs depicts a maximum rainfall regression coefficient which is positioned between the West African coast and 7 • N, leading to a relatively good CMIP6 EnsMean position at 4.5 • N (Fig. A4)). Furthermore, the spatial distribution of the rainfall regressed coefficients onto the ATL3 index in 285 the models and ERA5 are compared. Over the West Africa (the WAB, see Table 2), the performance of the GCMs is poor to modest, which is in line with the various rainfall responses described above. The spatial correlation with ERA5 lies between −0.6 and 0.6. There are 18 GCMs which present a greater spatial variability of the rainfall pattern than the one observed in ERA5 (Fig. 5 (c)). In particular, the spatial variability in 10 GCMs is one time and half greater than that in ERA5.
Over the oceanic area (the TAB2), the position of the maximum of the rainfall regression coefficients averaged between 70 • W 290 and 10 • E ranges from 2.5 to 4.5 • N in 50 % of the 31 models (Fig. A4). The position in the CMIP6 EnsMean is located at 3.5 • N, which is two (three) degrees below the position in ERA5 (CMAP-HADISST and GPCP-ERSST). Relative to ERA5, the standard deviation of the rainfall regression coefficients onto ATL3 in the TAB2 region is generally too high in the CMIP6 models (27 models out of 31), as indicated in Fig. 5 (b). This spatial variability of the rainfall pattern in 17 GCMs is two times greater than that in ERA5. The models show a poor to modest spatial correlation with ERA5, which ranges from −0.4 to 295 0.6 . The 1985-2014 sign-dependent average of the rainfall pattern associated with ATL3 over the equatorial Atlantic (the EAB region, see Table 2) indicates that 30 GCMs out of 31 depict an overall positive rainfall correlation with the Atlantic Niño index ( Fig. A5 (b)), with only GISS-E2-1-G presenting insignificant correlations. We define a third group, OC+, which comprises these models. The aspects of the OC+ patterns are obviously very similar to the ones of the GC+ (Fig. 6 (c), (g), (k), (o), (s), (w)).

300
Besides, we found no clear relation between the intensity of the warm SST biases and the JAS mean rainfall biases over the equatorial Atlantic region (the EAB) in the different models. It is also the case for the link between the warm biases and the biases of the rainfall regression coefficients related to the Atlantic Niño (Fig. A7). This conclusion is valid for the relationship between the SST biases in the EAB region and the Guinean Coast seasonal rainfall and rainfall associated with ATL3.   (Fig. 7 (a)). Relative to the 1985-2014 period, the 310 standard deviation of the Atlantic Niño index shows a change that ranges from -44 % to 49 % for the near-term period, -41 % to 21 % for the mid-term period and -49 % to 10 % for the long-term period ( Fig. 7 (b)). The average of the relative changes in the 31 GCMs amounts to -8 %, -13 % and -21 % for the near-term, mid-term and long-term periods, respectively. Among the 31 GCMs considered, 20, 22, and 26 agree on the reduction of the Atlantic Niño variability for the near-term, mid-term and long-term periods, respectively. Interestingly, these results are opposite to the findings of Brierley and Wainer (2018), who 315 evaluated the change of the ATL3 variability in a quadrupled CO 2 experiment with CMIP5 models. They found that 3 (12) GCMs out of 15 depict a decrease (an increase) of the Atlantic Niño variability. In the present-day climate, the multimodel mean of SST anomalies associated with ATL3 and averaged over the equatorial Atlantic (the EAB region) is 0.32 • C. This value is consistent with the observations and ERA5. For the periods 2015-2039, 320 2040-2069 and 2070-2099, the EnsMean value is reduced to 0.29, 0.28 and 0.26 • C, respectively. This result indicates that, relative to the present-day climate, the multimodel mean of the SST response to the Atlantic Niño has gradually decreased, with a percentage of change equals to -8 %, -13 % and -19 % for the near-term, mid-term and long-term periods, respectively ( Fig. 8 (a)). There are 81 % and 84 % of the 31 GCMs which agree on the sign of the change in the mid and long-term periods, against 55 % in the near-term period. In addition, the consistency of the equatorial Atlantic SST response to ATL3 among the GCMs Over the period 1985-2014, the CMIP6 EnsMean rainfall anomalies related to one standard deviation of the ATL3 index and averaged over the EAB is 0.59 mm · day −1 , which is greater than the observed value (0.2 mm · day −1 ). Subsequent to the 330 weakening of the equatorial SST anomalies in the future, the rainfall associated with ATL3 over the EAB area decreases ( Fig.   8 (b)). The CMIP6 EnsMean of the anomalous EAB rainfall values for the three consecutive future periods are 0.50, 0.44 and 0.37 mm · day −1 , respectively. The corresponding EnsMean rainfall reduction relative to the present-day period is about 14 %, 25 % and 37 %, with an agreement of 68 %, 77 % and 84 % of the models on the sign of the change in the near-term, mid-term and long-term periods, respectively.

335
Figures 9 (a) and (c) indicate that, over the Atlantic Ocean, in the four periods of the study, the position of the maximum rainfall anomalies associated with ATL3 remains close to 4 • N, but with a linear decrease of the rainfall intensity with time.
North of 5 • N, an upper level subsidence leads to negative rainfall anomalies in the tropical North Atlantic, which gradually weaken in the future periods. Over the EAB region, the decrease of the rainfall interquartile range in the CMIP6 models for the 340 mid and long-term periods indicates an increased consistency of the rainfall response among the models. For the present-day, the near-term, the mid-term and the long-term periods, the IQR values are 0.46, 0.49, 0.40 and 0.34 mm · day −1 , respectively.
Regarding the ATL3 induced rainfall over the Guinea Coast, an overall reduction of the multimodel mean is projected (Fig.   8 (c), Figs. 9 (b) and (d)). The CMIP6 EnsMean rainfall averaged over the Guinea Coast (the GCB area) decreases from a 345 value of 0.36 mm · day −1 in the period 1985-2014 to 0.22 mm · day −1 at the end of the 21st century. The corresponding values for the near-term and mid-term periods are 0.29 and 0.24 mm · day −1 , respectively. The amount of the GCB EnsMean rainfall reduction reaches thus 18 %, 33 % and 38 % in the near-term, mid-term and long-term periods, respectively. The percentages of the models that agree on a reduction of the rainfall magnitude associated with ATL3 over GCB for the three periods are 61 %, 65 % and 58 %, respectively. This means that there is a lesser agreement on the projected ATL3-rainfall changes over the 350 Guinea Coast than over the equatorial Atlantic. Moreover, the IQR of the rainfall associated with ATL3 over the Guinea Coast is more or less the same over the four periods (0.6 mm · day −1 ).
In the future climate, following the SSP5-8.5 scenario, the variability of the trade winds associated with the Atlantic Niño feedback, which links anomalous westerlies to the abnormal warming in the eastern equatorial Atlantic. The CMIP6 EnsMean of the 850 hPa zonal wind anomalies corresponding to one standard deviation of the ATL3 index and averaged over the EAB is equal to 0.22 m · s −1 in the 1985-2014 period, which is close to the value derived from ERA5. For the near-term, mid-term and long-term future, this value decreases to 0.14, 0.10 and 0.04 m · s −1 , respectively. This reduction corresponds to 36 %, 54 % and 82 % of the value obtained in the present-day climate, respectively, with a high agreement among the 31 CMIP6 models 360 (74 % for the near-term and mid-term periods, and 84 % for the long-term period).
The anomalous westerlies forcing on the eastern equatorial Atlantic thermocline is the second component of the Bjerkness from the ocean to the westernmost region of the Guinea Coast is shifted equatorward. and moisture flux divergence (in colors) (u-y) regression patterns associated with the standardized ATL3 index, relative to the present-day climate (2070-2099 minus 1985-2014). Stippling in (a)-(t) and contours in (u)-(y) indicate areas where the mean change (in colors) is significant at 95 % level according to a two-sided Welch t-test and where at least two thirds of the models agree on the sign of the change.
The number of models in each group is indicated in parentheses.

Change of the Atlantic Niño impact on the equatorial Atlantic rainfall
When considering the ATL3-rainfall pattern over the equatorial Atlantic, by construction, 21 GCMs belonging to the OC+group project a decreased rainfall signal in a warmer climate (Fig. 10 (c)). In contrast, 9 GCMs in the OC++ group show no change of the ATL3-rainfall magnitude (Fig. 10 (d)). The change patterns of the ATL3-SST anomalies are similar in the OC+and OC++ groups. They show a future decrease of the SST associated with ATL3 in the Atlantic Niño region (Fig. 10 (h)-(i)).
However, while there is no change in the low level zonal wind response to ATL3 in the OC++ group, the OC+-group projects an important weakened zonal wind speed related to ATL3 (Fig. 10 (m)-(n)).

445
The analysis of the SSH anomalies associated with one standard deviation of the ATL3 reveals a projected decrease of the SSH in the tropical Atlantic for the OC+-group (Fig. 10 (r)). In the OC++ group, this change is weaker (Fig. 10 (s)). These results suggest a projected weakened Bjerkness feedback in the OC+-group, while this feedback keeps the same intensity in the OC++ group. Furthermore, consistent with the change of the rainfall pattern, Figure 10   , for the near-term, mid-term and long-term periods, relative to the present-day conditions. These changes are computed for the JAS season and averaged over 23 GCMs (Table 1).
In a warmer climate, the deepening of the Saharan heat low and its extension over the North Atlantic weakens the mean equatorial easterlies in JAS. In addition, over the tropical Atlantic, the models project an anomalous increase (decrease) of the mean sea level pressure (SLP) in the western (eastern) part of the basin in the future. This favors anomalous westerlies over the equatorial Atlantic ( Fig. 11 (a)). As a consequence, the weakened trade winds lead to a deeper mean thermocline in the eastern equatorial Atlantic. This is indicated by the increased oceanic mixed layer depth, here used as a proxy for the top of 460 the thermocline depth. Figure 11 (b) shows that this deepening is gradual with time. As the thermocline becomes deeper, it responds less to the SST changes at the surface during the Atlantic Niño phases, and leads to a weakened Bjerkness feedback.

Discussion and Conclusion
In this study, the seasonality of the rainfall in the Guinea Coast has been analyzed based on the output of 31 models participating in the CMIP6 project. First, the 1985-2014 period of the historical simulations has been considered. Results show that 465 the majority of the models fails to reproduce the bimodal structure of the rainfall annual cycle over this area, due to a warm surface bias in the eastern tropical Atlantic. The amplitude of the Guinean Coast rainfall annual cycle is also larger than in the observations, owing to a wet bias in this area in boreal summer. Overall, the Guinean Coast rainfall annual cycles in the CMIP6 models are reasonably well correlated with the one in the reanalysis, ERA5. The correlation values are above 0.9. Regarding the annual cycle of the rainfall standard deviation, models display a wide range of variability. Averaged over the entire annual 470 cycle, this variability is more larger in the models than in ERA5. The correlations between the models and ERA5 rainfall STD annual cycles range from 0.5 to 0.9.
The annual cycle of the ATL3 index and the spatial pattern of the associated SST anomalies in the GCMs show good correlations with the observations and ERA5. However, the maximum rainfall position associated with the Atlantic Niño in the 31 475 GCM ensemble mean is displaced south of the observed positions over the tropical Atlantic. Nearly all the models manifest an increased rainfall associated with positive ATL3 phases over the equatorial Atlantic. In the case of the Guinea Coast region, 24 models exhibit a realistic increased rainfall related to a warm phase of the Atlantic Niño, against 6 models which show a negative or weak rainfall response. In connection with ATL3, there is a weaker zonal wind response and a weaker zonal moisture flux toward the Guinea Coast, as well as a weaker deep convection in these 6 models (the GC-group), relative to the 480 24 models (the GC+ group).
Secondly, the change of the Atlantic Niño and its impact on the rainfall over the equatorial Atlantic and the Guinea Coast in a climate with a high anthropogenic emission of greenhouse gases has been evaluated. Analyses of the Shared Socioeconomic Atlantic. In these models, unlike the OC+-ones, the zonal wind response to ATL3 is more or less identical in the present-day and future periods. The Bjerkness feedback is projected to weaken in the OC+-group, while its intensity remains the same in the OC++ group.

510
In the GC+-and OC+-groups, the equatorial Atlantic convection associated with the Atlantic Niño is projected to decrease, as indicated by the weakened wind divergence difference between 200 hPa and 850 hPa. This decrease is weaker in GC++ and OC++ groups. This result adds support to Jia et al. (2019), who found a more stabilized troposphere over the tropical Atlantic in the future, which in turn reduces the convection during the positive phases of the Atlantic Niño. Their analysis was focused 515 on the link between the Atlantic Niño and La-Niña, while our analysis is focused on the Atlantic Niño teleconnection with the tropical Atlantic and Guinea Coast.
In the last part of this study, we linked the background JAS mean state change to the change of the Atlantic Niño variability. In a warmer climate, we found an oceanic extension of the Saharan Heat Low over the North Atlantic. This anomalous 520 low pressure weakens the equatorial trade winds. In addition, over the equatorial Atlantic, the mean sea level pressure change is positive in the western part of the basin, and decreases eastward. This contributes to decreasing the intensity of the mean easterlies, thus limiting the equatorial upwelling. Subsequently, the thermocline becomes deeper along the equator, and the deepening is more important in the eastern equatorial Atlantic. For this reason, the influence of the eastern equatorial Atlantic thermocline on the surface during the Atlantic Niño phases is weakened in the future. This explains the projected decreased 525 variability of the Atlantic Niño in the SSP5-8.5 emission scenario.
Future studies are needed to assess the impact of the warm biases in the eastern equatorial Atlantic on the teleconnection patterns of the Atlantic Niño, and could help to increase the reliability of the projected changes. Moreover, the reduction of these biases is crucial for the rainfall projections over West Africa. Further analyses are also needed to understand the 530 multidecadal modulation of the Guinean Coast rainfall and extreme rainfall by oceanic internal modes of variability in climate models for both present-day and future climate conditions.