Introduction

Summer precipitation variations associated with global monsoon systems have significant impacts on countries worldwide by profoundly impacting local living conditions as well as the global climate system1,2. For instance, rainfall variations of the West African monsoon can exert prominent influences on health and livelihoods of its inhabitants3,4,5 since West Africa is one of the most vulnerable areas concerning food security6,7. Countries in the region exhibit high sensitivity to climate risk while possess limited adaptive capacity due to the inadequate infrastructure8. Additionally, the South American monsoon region encompasses the Amazon Rainforest and eastern Brazil9. The former region serves as Earth’s largest terrestrial carbon sink10,11,12 and contributes to local and regional climate stability13,14,15,16. In eastern Brazil, a very densely populated area, the rainfall quantity holds crucial importance for agriculture, hydroelectric power generation and water management17. Extreme rainfall events associated with monsoon variability may also bring urban floods and landslides, resulting in economic loss and mortality17. Recognizing the societal and climate significance, there is an urgent need for better understanding of the physical drivers and accurate predictions of the interannual variations of the West African and South American monsoon rainfall.

Both West African and South American monsoons are strongly affected by the leading mode of sea surface temperature (SST) variability in the tropical Atlantic Ocean, known as Atlantic Niño4,18,19, characterized by warm SST anomalies (SSTAs) in the central and eastern equatorial Atlantic basin, accompanied by weakened easterly trade winds20,21,22,23. The Atlantic Niño usually peaks in boreal summer, which can strengthen the West African monsoon, bringing more rainfall to countries bordering the Gulf of Guinea24,25,26,27,28,29,30. The Atlantic Niño also exhibits a secondary peak during austral summer31, allowing it to influence the South American monsoon4,19,20,32,33. A warming equatorial Atlantic could delay the northward migration of the ITCZ, resulting in changes in rainfall over northern South America34. The Amazon region experienced severe droughts and floods during 2005–2012, which are partly linked to SSTAs in the tropical Atlantic4,35,36,37.

Furthermore, a recent study has identified two distinct types of Atlantic Niño, characterized by different locations of positive SSTAs in the tropical Atlantic38. The eastern Atlantic Niño (EAN) is primarily associated with sea surface warming centered in the eastern equatorial Atlantic Ocean and along the West African coasts, while the central Atlantic Niño (CAN) exhibits peak warming in the central basin with weak coastal warming. The distinct spatial patterns between the CAN and the EAN lead to large discrepancies in their respective influences on other remote regions. It is suggested that the CAN has a more pronounced influence on the subsequent evolution of the El Niño-Southern Oscillation (ENSO) compared to the eastern type, especially after ~200038. The impacts of the two types of the Atlantic Niño on the European climate also differ substantially due to the different atmospheric teleconnection patterns associated with them39.

The two Atlantic Niño types therefore have distinct climatic impacts, yet it remains unknown whether the monsoon rainfall variations over West Africa and South America exhibit different responses to these two types of tropical Atlantic Ocean warming events. Understanding individual influence of the CAN and EAN on the two monsoon systems may provide additional sources of predictability for summer rainfall variability, holding important implications for the countries surrounding the tropical Atlantic Ocean. The present analysis aimed to (1) elucidate the different responses of monsoon rainfall over West Africa and South America to the CAN and EAN and (2) examine if the seasonal prediction skill of the monsoon rainfall over the two regions can be improved by taking into account the two distinct types of the Atlantic Niño.

Results

West African rainfall responses to the two Atlantic Niño types

The seasonal evolution of climatological rainfall over the West African monsoon region is presented in Fig. 1 along with 850-hPa winds. The West African summer monsoon is characterized by low-level southwesterlies over the tropical northeastern Atlantic Ocean and the Inter-Tropical Convergence Zone (ITCZ) north of the equator. The precipitation distribution corresponds to the position of the ITCZ, exhibiting an east-west oriented belt-like pattern with stronger rainfall concentrated along the Guinea coast. Maximum monsoon rainfall occurs during June through September, contributing significantly to the annual precipitation in the region40,41,42,43 (Fig. 1b and S1).

Fig. 1: Seasonal evolution of West African and South American rainfall.
figure 1

Climatological mean precipitation rate (color shading in units of mm day−1) and 850-hPa winds (arrows in units of m s−1) averaged for (a) March-April-May (MAM), (b) June-July-August (JJA), (c) September-October-November (SON) and (d) December-January-February (DJF) from 1979 to 2022.

Full size image

As noted in previous studies, the Atlantic Niño exhibit two peaks during boreal and austral summer, respectively31. Indeed, the annual cycle of the variance for the two Atlantic Niño types both show a bimodal pattern with the most active period in June-July and a secondary peak in December-January (Fig. 2a). Previous study38 primarily focused on the two Atlantic Niño types during boreal summer. To examine whether the two types are seperable in austral summer, here we performed an Empirical Orthogonal Function (EOF) analysis of tropical Atlantic SSTA during December-January-February (DJF). The EOF patterns resemble previous findings38, based on which we further obtain the two Atlantic Niño patterns (Fig. S2). The central and eastern Atlantic Niño are indeed characterized by SSTA warming over the central and eastern tropical Atlantic, respectively (Fig. S2a, b). The corresponding indices for the central and eastern Atlantic Niño in DJF also exhibit large discrepancies (Fig. S2c), suggesting that the two types of Atlantic Niño are also prominent and largely independent during austral summer. Here, we focus on the influence of boreal summer Atlantic Niño on the West African monsoon rainfall.

Fig. 2: Two types of Atlantic Niño events.
figure 2

a Monthly variance of central Atlantic Niño (CAN) index (CANI, blue) and eastern Atlantic Niño (EAN) index (EANI, orange). b, c Regression of JJA mean sea surface temperature anomalies (SSTAs, °C) on the normalized JJA CANI and EANI, respectively. d, e Similar with (b, c) but for DJF mean. Dotted areas denote values that are statistically significant at 95% confidence level.

Full size image

The SSTAs in the tropical Atlantic associated with the CAN and EAN in summer exhibit evident differences. The CAN is characterized by warm SSTAs centered at about 20°W–15°W while the significant positive SSTAs of the EAN are situated in the eastern basin (Fig. 2b, c). To elucidate the different impacts of the two types of Atlantic Niño, we compare regressed June-July-August (JJA) mean rainfall anomalies over the tropical Atlantic and adjacent monsoon regions on the simultaneous CAN index (CANI) and EAN index (EANI, Fig. 3a, d). Note that ENSO and Atlantic Niño events are closely linked to each other, and they both can influence the rainfall variations of the West African summer monsoon44,45,46,47,48. In order to exclude the influence of ENSO and isolate the impacts of Atlantic Niño, a partial linear regression was applied by removing the regressed rainfall field with reference of simultaneous Niño 3.4 index.

Fig. 3: West African climate anomalies associated with Atlantic Niño.
figure 3

Regression of JJA mean (a) precipitation (mm day−1), (c) sea level pressure (SLP, shading, hPa) and 850-hPa wind (vector, m s−1) anomalies on the normalized JJA CANI during 1979–2022. Dotted areas and black vectors denote values that are statistically significant at 95% confidence level. b The variance ratio of rainfall anomalies induced by JJA CAN and the total rainfall anomalies during 1979–2022. (d–f) Similar with (a–c) but for EANI. g Differences of regression coefficients between (a) and (d). h Area-averaged regression coefficients of each term in Eq. (2) on normalized CANI (blue bar) and EANI (orange bar) during JJA. The area (0o-15oE, 0o-10oN) is denoted by the red rectangle in (a–g). P, E and R in the x axis represent total precipitation, surface evaporation and residual terms respectively. -\(\bar{u}{q}^{{\prime} }\) and -\(\bar{v}{q}^{{\prime} }\) represent changes in convergence of the zonal and meridional moisture flux anomalies due to moisture anomalies. -\(u{\prime} \bar{q}\) and -\(v{\prime} \bar{q}\) denote changes in convergence of the moisture flux anomalies caused by zonal and meridional advections.

Full size image

The results reveal that the CAN events contribute to increased rainfall over the western Atlantic north of the equator with a zonal expansion to the eastern basin, with only weak positive rainfall anomalies over the West Africa monsoon region (Fig. 3a). Conversely, the EAN events bring more rainfall to the eastern equatorial Atlantic and West Africa coastal region (Fig. 3d), prominently strengthen the climatological monsoon rainfall in this region during JJA (Fig. 1b). The differences of rainfall anomalies associated with the CAN and EAN exhibit a clear zonal dipole pattern (Fig. 3g), further suggesting the more pronounced impact of the EAN on the West African rainfall.

To further quantify the relative importance of different Atlantic Niño types on the rainfall variability, we calculate the variance ratio of rainfall anomalies induced by the CAN/EAN and the total rainfall anomalies (refer to Fig. 3b for CAN and 3e for EAN). In the western tropical Atlantic region, the precipitation variance associated with CAN events accounts for 10%-30% of the total variance. For the West African summer monsoon zone, the precipitation variance resulted from EAN events constitutes a substantial portion of the total variance, exceeding 40% in certain coastal areas of the Gulf of Guinea, with negligible contribution from the CAN. The results are consistent with the patterns of regressed rainfall anomalies depicted in Fig. 3a, d.

To explore the physical mechanisms underlying the notable distinctions in rainfall anomalies associated with two types of Atlantic Niño events, we examine changes in large-scale circulations. Figures 3c and 3f show the partial regressions of the 850-hPa winds and sea level pressure (SLP) during JJA on the CANI and EANI respectively. Negative SLP anomalies appear in the tropical central (Fig. 3c) and eastern Atlantic (Fig. 3f), which are mainly induced by the SST warming associated with the two distinct types of the Atlantic Niño (Fig. 2). Furthermore, significant westerlies are observed to the west of the depressed pressure center and easterlies to the east, leading to an anomalous water vapor convergence over the central and eastern Atlantic. In the EAN, consequently, abundant moisture is converged in Gulf of Guinea, leading to enhanced summer rainfall over the West African summer monsoon region.

A moisture budget analysis is employed to further quantify the relative contributions of various process to the anomalous moisture flux convergence. Over the West African monsoon region (0°-15°E, 0°-10°N), where both the rainfall climatology is high and the EAN induced rainfall is most pronounced, the CAN impact is negligible (Fig. 3h and S3a). The EAN influence is primarily attributed to the anomalous zonal moisture flux convergence, which are associated with the prominent westerly wind anomalies over the eastern tropical Atlantic that transport the moist air toward the Gulf of Guinea (Fig. 3f and S3e). By contrast, the wind anomalies associated with the CAN are mainly located at the western basin, exerting negligible impacts on the moisture transport over West Africa (Fig. 3c, h). The distinct wind changes are associated with different locations of the warm SSTAs of the two types.

Given the complexity of the climate system, observational results based on partial regression analysis may be affected by factors other than the Atlantic Niño impacts. To further examine the influences of the CAN and EAN on rainfall anomalies during boreal summer, we compare the atmospheric general circulation model (AGCM) experiments forced with their respective SSTA patterns. Model results indeed show the distinct rainfall anomalies caused by the two Atlantic Niño types (Fig. 4a–c). While the CAN primarily tends to induce positive rainfall anomalies over the western basin, the EAN strengthens the West African monsoon rainfall prominently. The different wind changes over the tropical Atlantic Ocean are reasonably reproduced as well. Since the AGCM experiments are forced solely by tropical Atlantic SSTAs, the results isolate the Atlantic Niño impacts and support the observational analysis described above.

Fig. 4: Atmospheric model experiments.
figure 4

a Differences of boreal summer precipitation (shading, mm day−1) and 850-hPa winds (vector, m s−1) between the central Atlantic Niño experiment (JJA_CN) and the control run (CTRL). b Similar with (a) but for the eastern Atlantic Niño experiment (JJA_EN). c Differences between (a) and (b). d–f Similar with (a–c) but for the season of austral summer. Dotted areas and black vectors denote the differences that are statistically significant at 95% confidence level.

Full size image

Influences of the two Atlantic Niño types on the South American monsoon

Unlike the West African monsoon, the South American monsoon experiences peak rainfall during December to March (Fig. 1d and S1), coinciding with the maximum solar radiation in the southern tropics49,50. During DJF, the Amazon Rainforest region experiences the most intense precipitation, gradually decreasing towards the surrounding areas. Unlike most monsoon systems with different wind directions between summer and winter, easterly winds dominate throughout the year over the northern South America (Fig. 1). However, evident reversal of lower-level winds still shows up when the annual mean is removed from the winter and summer mean circulations51.

The influence of austral summer CAN and EAN on January, February and March (JFM) mean rainfall anomalies over the South American summer monsoon region is explored in this section. Here we focus on the season of JFM mean instead of DJF mean since the austral summer Atlantic Niño events have closer associations with JFM mean rainfall compared with the season of DJF (Fig. S4).

The regression of JFM rainfall anomalies on normalized DJF mean CANI and EANI are shown in Fig. 5a, d. The CAN-induced rainfall changes exhibit a dipole pattern over South America. Enhanced rainfall is observed over eastern Brazil while depressed signals are located over Amazon, suggesting an eastward shift of South American summer rainfall belt compared to the climatology pattern (Fig. 1d). As shown in Fig. 5b, the rainfall variance contributed from the CAN accounts for more than 40% of the total rainfall variance over eastern Brazil. The influence over Amazon is relatively weaker and the variance ratio is about 10–30%. In contrast, there is no significant influence of the EAN events on South American summer rainfall (Fig. 5d, e).

Fig. 5: South American climate anomalies associated with Atlantic Niño.
figure 5

a–g Similar with Fig. 3a–g but for the season of austral summer. h, i Similar with Fig. 3h but for the season of austral summer and the region of eastern Brazil (55oW-35oW, 15oS-0o, h) and Amazon (78oW-60oW, 12oS-0o, i).

Full size image

The formation mechanisms of the rainfall dipole pattern over South America associated with CAN events are further examined. During the CAN events, the pronounced warm SSTAs in the central equatorial Atlantic excites ascending Rossby waves to the west, manifested as depressed SLP and cyclonic circulation anomalies over the southwestern tropical Atlantic and South America region (Fig. 2d, 5c). The moisture budget analysis reveals that the southerly wind anomalies at the western flank of the anomalous cyclone are the primary cause for the enhanced rainfall over Brazil (Figs. 5h and S5f). In the meantime, the zonal thermodynamic process (-\(\bar{u}q{\prime}\)) plays a secondary role in enhancing the rainfall (Figs. 5h and S5c), which are due to more moisture over the tropical Atlantic induced by the warm SSTAs (Fig. 2d) that are further transported toward the land region by the prevailing easterly winds (Fig. 1d).

The moisture budget analysis also suggests that the reduced rainfall over Amazon is mainly due to the weakened zonal moisture flux convergence (-\(u{\prime} \bar{q}\), Fig. 5i). In the target area (blue box), the CAN-induced wind anomalies over the western part are weaker compared with the eastern part. Thus, ∂u/∂x shows positive values within the blue box, suggesting anomalous divergence that favor drying anomalies over Amazon (Fig. 5c). In addition, the maximum moisture is located in the central Amazon, with a sharp decrease to its west (Fig. S6a, c). Consequently, the zonal gradient of moisture is positive in the western and central Amazon. The climatological moisture in the eastern Amazon is close to the maximum values in the central region (Fig. S6a, c), resulting in a small zonal gradient (Fig. S6b, d). Therefore, the westerly anomalies also transport more drier air to the east of the Amazon area. These two factors both favor a depressed zonal moisture flux convergence which further cause suppressed rainfall over Amazon.

The changes in rainfall and winds due to the CAN and EAN during JFM are simulated by the AGCM experiments reasonably well (Fig. 4d–f). In the model, prominent cyclonic wind anomalies show up over the southwestern tropical Atlantic due to the CAN forcing, causing enhanced rainfall over eastern Brazil. By contrast, the EAN mainly induces positive rainfall anomalies over the eastern basin. Over Amazon, the negative rainfall anomalies during the CAN are underestimated in the model, which are likely due to the complex topography effects that cannot be properly resolved by the model. The overall agreement between the model and observations further supports the prominent influence of the CAN on the South American monsoon rainfall.

Seasonal prediction of summer rainfall

Since there are notably different responses of local summer rainfall over West Africa and South America to the CAN and EAN, whether or not the seasonal prediction skills of these monsoon rainfall systems can be improved by considering the two distinct Atlantic Niño types is explored. Studies pointed out that short-term predictions of the tropical Atlantic are still challenging due to the persistent biases in climate models that have seen little progress over the past two decades4. Thereby, a physically-based empirical model52,53,54,55 is employed in this study to make the seasonal forecast by establishing the prediction equation via multi-linear regression. In addition to the SSTA signals from tropical Atlantic, ENSO is also a dominant precursor for both West African and South American summer monsoons as documented by previous studies5,56,57. Indeed, the predictability of rainy season rainfall over northeast Brazil for the period 1912–98 was explored by previous study using dynamical and empirical techniques by taking SSTAs associated with ENSO and Atlantic Niño as predictors32.

Based on the previous studies, we choose Niño 3.4 index as the first predictor. As shown above, the CAN and EAN have prominent impacts on the two monsoon systems analyzed in this study. Hence, we select either the CANI or the EANI as the other predictor. For comparison, we also use the ATL3 defined as SSTA averaged over (20°W-0°, 3°S-3°N) as a potential predictor. Since the ATL3 mainly describes the Atlantic Niño as a whole, by comparing its contribution as a predictor allows us to examine whether or not classifying the Atlantic Niño into two types could improve the prediction skill. The definition of each predictand and the corresponding predictors are summarized in Table 1.

Table 1 Definition of each predictand and the corresponding predictors
Full size table

First, the leading correlation coefficients between predictand and each predictor are demonstrated in Fig. 6. Among the three factors characterizing the Atlantic Niño events, the correlation between precipitation over West Africa and EANI is the highest (0.61), significantly surpassing that of ATL3 (0.51) and CANI (0.06). This is consistent with our findings that the EAN events can have a more significant influence on the West African summer monsoon than the CAN events. Similarly, the South American summer monsoon is mainly affected by the CAN during austral summer, and the correlation between rainfall changes in eastern Brazil and the Amazon Rainforest region and the CANI is higher compared to ATL3, while correlations with the EANI are negligible.

Fig. 6: Correlations between predictand and predictors.
figure 6

a Leading correlation coefficients between area-averaged July-August mean rainfall index over West Africa (0o-15oE, 0o-10oN) and May-June mean CANI (yellow bar), EANI (orange bar), ATL3 (purple bar) and Niño 3.4 (blue bar). b, c Similar with (a) but for each predictor during December–January mean with reference of February–March mean rainfall index over eastern Brail (55oW-35oW, 15So-0o) and Amazon (78oW-60oW, 12oS-0o).

Full size image

We further examine the relative impacts of the tropical Atlantic and Pacific climate modes on the two monsoon systems. It is interesting to note that influences of ENSO on precipitation variations over West Africa and eastern Brazil are less pronounced compared to the variability in Atlantic SSTs (Fig. 6a, b). For the Amazon region, the influence of ENSO is indeed more robust, with the CAN playing a secondary role. The stronger impact from ENSO could be related to the geographical proximity of Amazon to the Pacific.

Then the predictive capability of various models derived from Niño 3.4 index and different Atlantic Niño indices are evaluated. Multi-linear regression is employed to establish the prediction equation, after which we apply the cross-validation method to make a retrospective forecast for the 44 years from 1979 to 2022. The red/blue/yellow line in Fig. 7a shows the predicted July–August mean West African rainfall by taking CANI/EANI/ATL3 as well as Niño 3.4 index as predictors. The equation of each model is summarized in Table S1. The retrospective forecast skill is quantitively evaluated from the correlation coefficient between the observations and predicted values, indicated by the numbers in the bottom legend of Fig. 7a. Results show that the prediction model taking EANI as the predictor has the highest prediction skill compared to the other two prediction models though Niño 3.4 index is included in all models. The prediction skills of different models for eastern Brazil and Amazon region are also assessed (Fig. 7b, c). It is not surprising that the models established based on CANI and Niño3.4 have the highest prediction skills.

Fig. 7: Seasonal prediction skills.
figure 7

a The hindcast index in cross-validated predictions during 1979–2022 compared with the observed time series (black line) of July–August mean precipitation over West Africa (0o-15oE, 0o-10oN). The lines in different color represents the results taking CANI (red), EANI (blue), ATL3 (yellow) as well as Niño 3.4 as predictors. The numbers in the bottom legend indicate the correlation coefficients of hindcast rainfall index using different Atlantic Niño indices as predictors with reference to the observation. b, c Similar with (a) but for February-March mean rainfall index over eastern Brail (55oW-35oW, 15So-0o) and Amazon (78oW-60oW, 12oS-0o).

Full size image

Note that we focus on impacts of boreal/austral summer Atlantic Niños on rainfall during JJA/JFM in previous sections. However, the seasonal prediction typically relies on leading impacts of predictors on the predictand. Hence, we selecte rainfall anomalies during JA/FM as predictands and MJ/DJ SSTA as the predictors. These selections fully represent the lead-lag relationship between the predictors and the predictands, and meanwhile ensure that they are both still primarily in boreal/austral summer seasons. For comparison, we also established similar models to predict the JJA/JFM rainfall for consistency, using May-June-July/DJF mean predictors. The conclusions remain the same, although the prediction skills may exhibit some minor differences (not shown). Nevertheless, the findings above suggest that it is necessary to separate the Atlantic Niño into EAN and CAN events, which can help improve seasonal climate predictions.

Discussion

Better understanding the physical drivers and accurate predictions of the interannual variations of the West African and South American monsoon rainfall has prominent socioeconomic benefits to the local inhabitants. Both monsoon systems are prominently affected by the Atlantic Niño, and this study mainly focuses on investigating the individual impact of CAN and EAN events.

The two types of the Atlantic Niño exhibit distinct influences on the rainfall variability over the West African and South American summer monsoon regions during their local summer. Observational analysis reveals that boreal summer EAN events generally favor enhanced rainfall across the eastern equatorial Atlantic Ocean and coastal West Africa regions, suggesting a strengthened West African summer monsoon. More than 40% of the total rainfall variance over the coastal Guinea region can be explained by the EAN events. By contrast, no significant rainfall anomalies are observed over West Africa during CAN events. The abundant rainfall related to the EAN events can be attributed to the direct forcing of the anomalous warm SSTAs in the eastern tropical Atlantic Ocean, resulting in decreased SLP along the coast of West Africa and anomalous westerlies, bringing plenty of moisture to Guinea coast and thereby leading to strengthened summer rainfall. This process is further supported by a suite of atmospheric model simulations.

For the South American summer monsoon region, the rainfall variance during local summer is mainly controlled by the CAN events rather than EAN. Rainfall anomalies associated with the CAN exhibit increased rainfall over eastern Brazil and drying signals located over the Amazon Rainforest. This dipole pattern corresponds to an eastward shift of summer rainfall belt which may be attributed to the anomalous westerlies along equatorial South America and the cyclonic circulation anomalies over eastern coast of South America. More moisture is transported to eastern Brazil from (1) southwestern Atlantic Ocean by the western branch of the anomalous cyclone and (2) tropical Atlantic by the prevailing easterly winds. In contrast, the westerly anomalies appearing over the tropical South America may suppress the rainfall over Amazon by weakening the zonal moisture flux convergence.

Previous studies have defined the domains of the West African and South American summer monsoons which describe the entire monsoon systems as a whole1,2. This work emphasizes the distinct impacts of the two Atlantic Niño types on the monsoon system in a regional scale. Hence, we focused on different domains compared to previous studies. The regression analysis used in this study mainly reveals the overall impact of Atlantic Niño/Niña. However, the warm and cold phases may affect the monsoon systems differently. Full investigation of the asymmetric impacts between Atlantic Niño and Niña is warranted in a future study.

Since there are different responses of local summer rainfall over West Africa and South America to the two Atlantic Niño types, whether and to what extent the seasonal prediction skill can be improved by distinguishing Atlantic Niño into two distinct types is further examined through physically-based empirical model method. The retrospective forecast skills of summer rainfall over West Africa, eastern Brazil and Amazon are evaluated respectively. It is revealed that combining EANI and ENSO for prediction of the West African summer precipitation clearly outperforms prediction models using ATL3 or CANI to represent the Atlantic Niño. Similarly, predicting the local summer precipitation in the eastern Brazil and Amazon using CANI instead of ATL3 or EANI as the predictors also leads to higher prediction skills. These findings suggest the necessity and advantage of separating the Atlantic Niño into EAN and CAN events since the distinguishment between them can help improve seasonal climate predictions.

The relative impacts of the tropical Atlantic and Pacific on rainfall variability over West Africa and South America are also assessed. The Atlantic Niño events seem to have a more important contribution to rainfall changes over West Africa and eastern Brazil while the influence of ENSO on rainfall over Amazon is more significant. These results are obtained by simply calculating the correlation coefficients between rainfall and each index representing ENSO and Atlantic Niño. However, ENSO and Atlantic Niño, both of which may exert influence on the West African and South American monsoon systems, have complicated interactions with each other. Further studies are required to quantitatively differentiate the relative contributions of the tropical Atlantic and Pacific Oceans. In addition, since rainfall prediction is not the primary focus of this study, the predictors selected for the physically-based empirical model may not be the best choices in practice. There is still room to further improve the prediction skill, although more efforts are needed to detect other predictable sources.

Methods

Observational data

The SST used in this study is the monthly Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST) data set on a 1° latitude‐longitude grid58. The observational rainfall data is from the monthly precipitation analysis from Global Precipitation Climatology Project (GPCP) Version 2.359, archived on a global 2.5° × 2.5° grid. The monthly atmospheric data including 850-hPa winds and SLP is obtained from the fifth generation of the European Center for Medium-Range Weather Forecasts atmospheric reanalysis dataset (ERA5)60 with a spatial resolution of 0.25° × 0.25°. The linear trends were removed from all the seasonal mean fields to exclude the effect of the anthropogenic global warming.

Definitions of the CAN and EAN

In order to identify and characterize the two types of Atlantic Niño events, an empirical orthogonal function (EOF) analysis was performed on the monthly SSTA over the tropical Atlantic region (40°W-20°E, 10°S-10°N). Then the EAN events are defined as (EOF1 + EOF3) \(/\sqrt{2}\) and the associated index referred to as EANI can be obtained by combining the corresponding principal components (PCs) as (PC1 + PC3) \(/\sqrt{2}\). Similarly, the CAN events and the CANI are defined as (EOF1-EOF3) \(/\sqrt{2}\) and (PC1-PC3) \(/\sqrt{2}\) respectively38.

Moisture budget

Moisture budget analysis was conducted in this study to examine the contributions of different moisture transport processes for precipitation changes under the influence of different Atlantic Niño types. The moisture budget equation can be written as follows:

$$P=-\frac{1}{{\rho }_{w}g}{\int}_{{\!P}_{t}}^{{P}_{s}}\left[\nabla \cdot \left(\bar{V}q^{\prime} \right)\right]{dp}-\frac{1}{{\rho }_{w}g}{\int}_{{\!P}_{t}}^{{P}_{s}}\left[\nabla \cdot \left(V{\prime} \bar{q}\right)\right]{dp}+E+R$$
(1)

where \({\rho }_{w}\) is the density of liquid water, g is the gravity acceleration speed, V = (u, v) is the horizontal wind field, q is the specific humidity, Ps is the surface pressure, and Pt is the pressure at the top of the atmosphere (which is taken as 100-hPa in this study). P and E are total precipitation and surface evaporation, respectively61,62,63, R is the residual term.

The first and second terms on the right-hand side of Eq. (1) represent changes in the convergence of moisture flux anomalies due to moisture anomalies and changes in the winds, respectively. The mean components indicate the monthly climatology values of the variables, while the anomalies are deviations from the climatology of each month. These two terms thus can be regarded as thermodynamic and dynamic contributions to the interannual variability of moisture flux convergence. The last term R denotes the residual due to the vertical motions at surface and the tropopause, the moisture storage term, the high-frequency eddy moisture fluxes that are not considered in Eq. (1), which is assumed to be negligible. Since the horizontal wind field V can be further partitioned into zonal and meridional components (i.e., V = (u, v)), Eq. (1) can be expressed as

$$\begin{array}{l}P=-\frac{1}{{\rho }_{w}g}{\int }_{{P}_{t}}^{{P}_{s}}\left[\nabla \cdot \left(\bar{u}{q}^{{\prime} }\right)\right]{dp}-\frac{1}{{\rho }_{w}g}{\int }_{{P}_{t}}^{{P}_{s}}\left[\nabla \cdot \left(\bar{v}{q}^{{\prime} }\right)\right]{dp}\\ \qquad-\frac{1}{{\rho }_{w}g}{\int }_{{P}_{t}}^{{P}_{s}}\left[\nabla \cdot \left(u{\prime} \bar{q}\right)\right]{dp}-\frac{1}{{\rho }_{w}g}{\int }_{{P}_{t}}^{{P}_{s}}\left[\nabla \cdot \left(v^{\prime} \bar{q}\right)\right]{dp}+E+R\end{array}$$
(2)

The six terms on the right-hand side are referred to as -\(\bar{u}{q}^{{\prime} }\), -\(\bar{v}{q}^{{\prime} }\), -\(u^{\prime} \bar{q}\), -\(v^{\prime} \bar{q}\), E and R, respectively, in Figs. 3h, 5h and i.

Atmospheric model experiments

In order to explore the distinct impacts of the two Atlantic Niño types on monsoon rainfall over West Africa and South America during local summer season, a suite of AGCM experiments was conducted in this study. The model is Community Atmosphere Model 5 (CAM5) from National Center for Atmospheric Research (NCAR) with a horizontal resolution of approximately 1° × 1° and 30 vertical levels64.

First, a control simulation (CTRL) was conducted by employing monthly SST climatology as the forcing and the results were derived as a reference state. Then, a series of sensitive experiments were performed to compare the climate impacts of different SSTA patterns associated with the two types of Atlantic Niño. In the first experiment (JJA_CA), the regressed SSTA patterns associated with the boreal summer CANI was imposed on the monthly climatological SST from June to August in the tropical Atlantic (50°W–20°E, 17.5°S–10°N). A sponge layer was implemented at the boundaries of the SSTA forcing regions. In May and September, SSTAs with half of the amplitude were incorporated into the forcing field as temporal buffers. The second experiment (JJA_EA) is designed in the same way as the JJA_CA experiment, expect that the SSTA is associated with EAN events during boreal summer. The third and fourth experiments (DJF_CA and DJF_EA) were designed similar to the first two experiments but for the season of December to February mean. Similarly, half values of SSTAs in November and March were also incorporated into the forcing field. Each experiment was integrated for 30 years. Considering that the model typically takes several years to achieve statistical equilibrium, the initial 5 years were discarded and the subsequent 25 years of outputs were analyzed.

Statistical methods

Given the dominant role of ENSO in the interannual variability of the global climate system, partial regression analyses were employed throughout this study by removing the regressed fields on Niño 3.4 index from the total to exclude the possible influence of ENSO on summer rainfall variations. The significance test of the regression results was made based on the two-tailed Student’s t-test.

Multi-linear regression was used to establish the prediction model. Prior to regression, all variables were normalized by removing their means and dividing by their corresponding standard deviations. Cross-validation method65 was applied for a retrospective forecast. We left out three years of data progressively centered on a forecast target year for the period of 1979–2022, then trained the model using the data of the remaining years and finally applied the model to forecast the three target years.