Introduction

The tropical Indo-Pacific climate exhibits predictability on seasonal timescales, largely due to the relatively wellunderstood coupled dynamics of the El Niño-Southern Oscillation (ENSO) (L'Heureux et al., 2017;McPhaden & Nagura, 2014;Stuecker et al., 2017). In contrast, the equatorial Atlantic region, in particular the Atlantic Niño phenomenon, largely evades skillful seasonal predictions. The Atlantic Niño, like its Pacific counterpart, is the primary driver of the interannual variability in the equatorial Atlantic (Merle et al., 1980;Zebiak, 1993) with farreaching consequences (Chenillat et al., 2021;Hastenrath, 1984;Janicot et al., 1998; N. S. Keenlyside et al., 2013;Kim et al., 2023). However, unlike ENSO, the Atlantic Niño consistently exhibits poor seasonal prediction skills (Exarchou et al., 2021;Stockdale et al., 2006), with skillful prediction limited to a short lead time of about one season using either statistical or dynamical models (Dippe et al., 2019;Hu & Huang, 2007;Repelli & Nobre, 2004;Stockdale et al., 2006;Wang et al., 2021). This limitation has been previously attributed to the pronounced biases for climate models to simulate coupled dynamics in this region related to local and remote forcing, as well as the lack of a comprehensive understanding of the mechanisms maintaining the climatology and its seasonal cycle (e.g., Counillon et al., 2021;Grodsky et al., 2012;Richter & Xie, 2008;Richter et al., 2014).

Despite three decades of debate about the causal relationship between Atlantic Niño and Pacific ENSO events, it is evident that these two climate phenomena exhibit high rates of co-occurrence (Chang et al., 2006;Chikamoto et al., 2020;Ding et al., 2012;Jiang et al., 2023;Keenlyside & Latif, 2007;Lübbecke & McPhaden, 2012;Prigent et al., 2023;Richter et al., 2023;Richter & Tokinaga, 2020;Rodríguez-Fonseca et al., 2009;Roy & Reason, 2001;Tokinaga et al., 2019). Notably, extreme winter El Niño events often follow Atlantic Niña events that occurred during the preceding summer (Keenlyside et al., 2013;Rodríguez-Fonseca et al., 2009). Also, ENSO events that persist from late spring through the following fall and winter tend to be accompanied by the oppositely signed equatorial Atlantic events in the intervening summer (Tokinaga et al., 2019). In particular, recent research suggests that the incipient onset of Pacific ENSO events may trigger the occurrence of oppositesigned equatorial Atlantic events during the boreal summer (Jiang et al., 2023). In light of these findings, we here propose a seasonal prediction scheme for the summer equatorial Atlantic events that leverages the longer-lasting air-sea preconditioning in the Pacific, in which the Atlantic Niño can be skillfully predicted up to three seasons in advance.

Materials and Methods

We used the global sea ice and SST analysis from the Met Office Hadley Centre (HadISST version 1.1) with a 1°× 1°horizontal resolution (Rayner et al., 2003). The horizontal winds at the 200-hPa level were obtained from the European Center for Medium-Range Weather Forecasts Reanalysis v5 (ERA5) with a 0.25°× 0.25°horizontal resolution. The surface wind stress and potential temperature data were derived from the Ocean ReAnalysis System-5 (ORAS-5) (Zuo et al., 2019). The observational and reanalysis data sets used in this study cover the period from 1980 to 2021.

For the model data, we used the SST outputs of seasonal forecasts from the North American Multi-model Ensemble project (NMME) project phase1 (Table S2 in Supporting Information S1) during 1982-2020, a multi-model forecast system that includes models from the operational forecast centers in the United States and Canada (Kirtman et al., 2014). We also analyzed the SST outputs from 10-member ensembles of tropical Pacific Pacemaker simulations based on the National Center for Atmospheric Research Community Earth System Model version 2 (NCAR CESM2) during 1980-2019, in which the SST anomalies in the eastern Pacific were nudged toward observations between 15°S and 15°N and the rest of the climate system is allowed to evolve freely (Danabasoglu et al., 2020). To avoid the potential influence of inaccurately simulated decadal variability, a 10year high-pass Fast Fourier Transform filter was applied to the CESM pacemaker results to focus on the ENSO's impact on interannual timescales.

Anomalies for all variables were derived relative to the monthly mean climatology and the linear trend was removed. All statistical significance tests were conducted using the two-tailed Student's t-test with n -2 degrees of freedom, where n represents the sample size. The velocity potential (VP) χ was calculated using the horizontal winds (V) at the 200-hPa level based on the definition of ∇ • V = -∇ 2 χ (Krishnamurti, 1971). The Niño3.4 index was calculated as the SST anomalies averaged over the region of 5°S-5°N, 120°-170°W, and the Atl3 index over the region of 3°S-3°N, 0°-20°W. The selection for El Niño/Niña events and the Atlantic Niño/Niña events has been detailed in Table S1 in Supporting Information S1. The depth of the 20°C isotherm (D20) is used to indicate the thermocline depth over the equatorial Pacific region (Meinen & McPhaden, 2000) while the depth of the 23°C isotherm (D23) is typically used for the equatorial Atlantic region (Florenchie et al., 2003;Lübbecke et al., 2010). Accordingly, the warm water volume (WWV) in the Pacific (WWV P ) was defined as the volume of water above the D20 in the region 5°S to 5°N, 120°E to 80°W, and the WWV in the Atlantic (WWV A ) was defined as the volume of water above the D23 in the region 5°S to 5°N, 50°W to 20°E. The Pacific thermocline slope was defined as the difference of D20 between the eastern equatorial Pacific (5°S-5°N, 160°W-80°W) and western equatorial Pacific (5°S-5°N, 120°E-180°), and the Atlantic thermocline slope was defined as the difference of D23 between the central-to-eastern equatorial Atlantic (5°S-5°N, 20°W-10°E) and western equatorial Atlantic (5°S-5°N, 50°W-20°W).

Results

The Intrinsic Predictability of Atlantic Niño and Its Co-Variation With the Pacific Counterpart

While the air-sea coupled processes in the equatorial Atlantic are similar to the Pacific ENSO (Figure 1a), the Atlantic Niño exhibits a distinct seasonality and timescale with much weaker variability (Keenlyside & Latif, 2007;Latif & Grötzner, 2000). A typical Atlantic Niño event begins to develop in boreal spring, peaks in the summer, and then decays rapidly (Figure 1b). In comparison, ENSO events exhibit a longer life cycle, typically beginning in early spring, peaking in winter, and decaying in the following summer (Figure S1 in Supporting Information S1), more than twice as long as those of the Atlantic Niño. It is also noteworthy that the strength of Bjerknes feedback is weaker in the equatorial Atlantic than in the Pacific (Burls et al., 2012;Lübbecke & McPhaden, 2013;Zebiak, 1993), as evidenced by the much weaker lead-lag correlations of SST, wind stress, and thermocline slope anomalies in the equatorial Atlantic (Figure 1b and Figure S1 in Supporting Information S1).

Based on the recharge oscillator model (Jin, 1997), the phase lag between changes in equatorial Pacific heat content and changes in the eastern Pacific SST is evidence of the ocean memory for ENSO, and therefore the

Geophysical Research Letters

10.1029/2024GL111494 heat content anomalies constitute a reliable predictor for the SST changes up to two-to-three seasons in advance (Figure S2 in Supporting Information S1). Here we apply an empirical orthogonal function (EOF) analysis to the anomalies of the equatorial Atlantic thermocline depth. Analogous to the Pacific (Figure S2 in Supporting Information S1), the first EOF mode (EOF1) in the Atlantic represents a zonal tilting mode that is roughly in phase with the western Atlantic zonal wind stress and the central-to-eastern SST changes associated with the Atlantic Niño (Figures S3a and S3c in Supporting Information S1). The positive (negative) phase of second EOF mode (EOF2) indicates an overall deepening (shoaling) of the thermocline in the equatorial Atlantic that is mainly associated with off-equatorial Sverdrup transport. This mode well represents changes in WWV in the equatorial Atlantic region (Figure S3d in Supporting Information S1). However, the maximum lagged correlation between these two modes reaches a modest value of about 0.26 with the first mode lagging the second mode by about one season (Figure S4a in Supporting Information S1). In particular, the lead-lag correlation exhibits a distinctly sharp pattern, in stark contrast to the broader correlation structure observed in the Pacific.

It is clear that Atlantic Niño and ENSO events often co-occur, despite ongoing debates on their causal relationship (Jiang et al., 2023; N. S. Keenlyside et al., 2013;Martín-Rey et al., 2012;Richter et al., 2023;Tokinaga et al., 2019). Analysis of the spectral characteristics of these two phenomena (Figure 1c) shows that ENSO exhibits a broad frequency peak on interannual timescales ranging from 2 to 7 years and a strong near-annual peak around 1.4 years arising from the nonlinear interaction of ENSO with the background annual cycle (Stein et al., 2014;Stuecker et al., 2013), while the Atlantic Niño displays less pronounced interannual power around 4 years (Latif & Grötzner, 2000) and is predominantly influenced by near-annual variability peaking at approximately 1.3 years. The co-variation of the equatorial Pacific and Atlantic variability is further supported by the cross-spectral analysis between the Niño3.4 and Atl3 indices. As shown in Figures 1d and 1e, there exhibits a strong inter-basin coherence with the Pacific leading the Atlantic in both the interannual and near-annual frequency bands.

We further present the temporal evolution of oceanic and atmospheric conditions related to equatorial Atlantic events. Summer equatorial Atlantic Niño events (Figure 2g) tend to coincide with the emergence of negative subsurface temperature anomalies in the Pacific during the previous winter to early spring (Figure 2h). These Pacific subsurface cold anomalies are accompanied by significant easterly wind stress anomalies in the western-to-central Pacific (Figures 2a and 2b). As the subsurface cold signal progresses toward the eastern Pacific (Figure 2c), a pronounced Walker circulation response is soon established in the spring as manifested in the VP field at the 200-hPa level, with a descending branch in the tropical eastern Pacific and an ascending branch in the eastern Atlantic (Figure 2d). This consequently favors western equatorial Atlantic surface wind stress anomalies in spring (Figure 2d), setting the stage for the subsequent development of Atlantic Niño events.

Prolonged Skillful Prediction of Atlantic Niño by Pacific Variations

In light of the above physical analyses, the slowly-evolving oceanic processes in the tropical Pacific and the early manifestation of atmospheric precursor signals in the western tropical Pacific (Figure 2h) have the potential to improve the prediction of Atlantic Niño events. Here we first quantify the predictive skill of the Atlantic Niño from the NMME based on the Anomaly Correlation Coefficient (ACC) and Root-Mean-Squared Error (RMSE) metrics. In Figure 3a, we present the ACC values of the Atl3 index for the summer target season (JJA) from different initial months for each model in NMME. Consistent with previous studies (Stockdale et al., 2006;Wang et al., 2021), the damped persistence of SST anomalies over the Atl3 region is barely outperformed by the NMME seasonal forecast models within short lead times. The NMME mean demonstrates superior prediction skill compared to most individual models, maintaining skillful predictions for up to three (two) months before losing skill with longer time leads, when utilizing a threshold of 0.5 (0.6) for ACC. In terms of RMSE, the NMME models demonstrate a wide inter-model spread, with the NMME mean showing higher RMSE values compared to damped persistence prediction at around one season lead times and comparable values beyond this lead-time range.

Inspired by the basic knowledge of the Pacific ENSO which emphasizes the precursory role of the western-tocentral Pacific winds and the equatorial heat content (McPhaden et al., 2006), we here take a similar approach Geophysical Research Letters 10.1029/2024GL111494 to predicting the summer Atlantic Niño variability by considering the local damped persistence and the ocean response to wind stress forcing in the western equatorial Atlantic. A modest improvement in forecast skill is achieved by including the western Atlantic wind anomalies (Wind A ) as a local precursor compared to the damped SST persistence prediction (Equations S1 and S2 in Text S1 in Supporting Information S1) (Figure 3a). Incorporation of the local equatorial Atlantic WWV (WWV A ), which is closely related to the relatively slower oceanic processes (i.e., discharge/recharge), leads to a minor improvement in the forecast skill of Atlantic Niño (Figure S5 in Supporting Information S1). We further incorporate the longer-lasting air-sea preconditioning in the Pacific associated with ENSO, including Pacific WWV (WWV P ), western-to-central Pacific wind anomalies (Wind P ), and Niño3.4 SST anomalies as additional precursors (Equation 1): (g) Evolutions of 3-month running mean normalized Atl3 index (red line), normalized western Atlantic surface wind stress (black line), normalized equatorial Atlantic thermocline slope anomalies (green line) and normalized Atlantic warm water volume for equatorial Atlantic events (half the difference between composited Atlantic Niño and Atlantic Niña events). (h) Similar to (g) but for the corresponding variables in the Pacific for equatorial Atlantic events. Large (small) dots mark values exceeding the 95% (90%) confidence level.

in which t indicates the year, τ indicates the lead time (with τ = 0 indicating prediction initialized in July). Full details of the statistical models are provided in Text S1 in Supporting Information S1, and the regression coefficients for Equation 1 are listed in Table S2 in Supporting Information S1. By using this full statistical model, the skillful prediction can be extended to three seasons when utilizing a threshold of 0.5 for ACC (Figure 3a). At a 9-month lead time, the skill and RMSE of this model are comparable to the NMME mean, which is in line with our dynamical analysis (Figure 2h) showing a substantial reduction of significant anomalies in the Pacific prior to 9 months.

In the top panel and inset of Figure 4, we show hindcasted summer Atl3 index for all years using the full statistical model at 3-month, 6-month and 9-month lead times along with their respective cross-validation error estimates (shadings; see details in Text S1 and Table S3 in Supporting Information S1) and the respective contributions of remote Pacific and local Atlantic air-sea conditionings to the summer Atlantic Niño on interannual timescales. The model demonstrates a robust skill in predicting the Atl3 variability across these different time leads, and notably, it clearly reveals that the remote Pacific air-sea conditioning, especially the Pacific WWV, stands out as the primary predictor with its impact markedly outweighing that of local predictors as the lead time increases.

To validate the impact of Pacific in the equatorial Atlantic on interannual timescales, we further show in Figure 4 the Atl3 index simulated by the 10-member ensemble mean of the tropical Pacific pacemaker experiments (bottom panel). Note that observed data sets in Figure 4a were similarly filtered before constructing the full statistical model for comparison. Since only the observed tropical Pacific forcing was prescribed, the 10-member ensemble mean of the simulated equatorial Atlantic variability can be considered as an approximation of the Pacific-driven Atlantic Niño variability. As shown in Figure 4, the Pacific-driven Atlantic Niño variability captures a considerable portion of the observed Atlantic Niño variability with correlation coefficient being around 0.45. The explained variance of Atl3 (∼20%) is similar to the increased explained variance by contrasting the full statistical model (Equation 1) and the model with only local precursors (Equation S2 in Supporting Information S1) in 6-9 months lead time. The similar relationship between the Pacific-driven Atlantic Niño variability and the hindcast values, especially at 6-month and 9-month lead times further supports the feasibility of using the Pacific air-sea conditions as a predictive source for the equatorial Atlantic SST variability.

Conclusions

While the existence of the Bjerknes feedback and the recharge oscillator dynamics in the tropical Atlantic region is evident, we in this study show that their influence on the Atlantic Niño variability is much less pronounced than on Pacific ENSO variability. This greatly limits the inherent predictability of the Atlantic Niño. Building upon previous research highlighting the impacts of Pacific Ocean on the equatorial Atlantic (Jiang et al., 2023;Latif & Grötzner, 2000;Tokinaga et al., 2019), we propose an approach to hindcast the Atlantic Niño with skill up to three seasons in advance by incorporating the Pacific air-sea conditions as additional precursors. Our statistical model for Atlantic Niño prediction is constructed under the assumption that opposite-signed equatorial Atlantic and Pacific events are usually concurrent, as the early-onset ENSO events in spring can trigger opposite-signed equatorial Atlantic events in the following summer (Jiang et al., 2023). However, the occurrence of equatorial Atlantic events has also been linked to decaying ENSO events (Tokinaga et al., 2019), other climate modes like the Madden-Julian Oscillation and the Atlantic Meridional Mode (Foltz & McPhaden, 2010;Lee et al., 2023), or local atmospheric or oceanic signal independent from ENSO (Grimm & Reason, 2011;Illig et al., 2020;Li et al., 2023;Lübbecke et al., 2010Lübbecke et al., , 2018;;Richter et al., 2013;Vallès-Casanova et al., 2020;Zhao & Fu, 2022).

Extending the predictive time horizon of such Atlantic Niño/Niña events remains challenging, as the long-lead precursor WWV P does not appear to influence these equatorial Atlantic events.

Furthermore, the frequent co-occurrence of Pacific SST anomalies and equatorial Atlantic SST anomalies, a relationship assumed throughout this paper, may not have held true before the satellite era (Ding et al., 2010;Martín-Rey et al., 2012, 2015;Rodríguez-Fonseca et al., 2009). Even in the past four decades, both climate modes have experienced significant changes due to radiative forcing and/or decadal background variability (Capotondi et al., 2015;Crespo et al., 2022;Dong et al., 2006;Jiang et al., 2020;Martín-Rey et al., 2018;Polo et al., 2013;Timmermann et al., 2018). Notably, the precursory Pacific WWV signal for ENSO events over the past two decades emerges much later than previously observed (McPhaden, 2012), and correspondingly ENSO predictability has decreased from two-to-three seasons to only one season (Hendon et al., 2009;Zhang et al., 2019). Interestingly, the hindcast skill for the equatorial Atlantic events during the post-2000 period using the statistical model is notably lower than in the pre-2000 period, although it still surpasses that of the MME (Figure S6 in Supporting Information S1). This reduced skill in Atlantic Niño could be directly related to changes in ENSO's ocean memory (i.e., shorter Pacific WWV-ENSO SST lead time) during the post-2000 period. Another possibility is that the Atlantic Niño is partly thermodynamically driven (Nnamchi et al., 2015(Nnamchi et al., , 2021)), whereby, in a warming climate, atmospheric stochastic forcing might exert an increasingly important impact via thermodynamic feedbacks. Further studies are needed to clarify the precise causes of these changes.