Introduction

Recent decades have seen an increasing frequency of multi-year (MY) El Niño-Southern Oscillation (ENSO) events1,2,3, where consecutive ENSO events of the same phase persist beyond a single year. Notable cases include the 2018-19-20 El Niño, in which two El Niño events occurred in succession4, followed by a triple-dip La Niña spanning from 2020 to 20235,6,7. This prolonged persistence of ENSO in one phase challenges conventional ENSO theories, which primarily describe its oscillatory nature—where El Niño and La Niña events typically alternate, each lasting about a year—resulting in single-year (SY) ENSO events8,9. Those traditional ENSO theories, such as the delayed10,11 and recharge oscillator theories12,13, explain the alternation between El Niño and La Niña through negative feedback mechanisms in tropical Pacific ocean-atmosphere interactions, particularly the buildup and discharge of equatorial ocean heat content (OHC).

Efforts have been made to refine the traditional ENSO dynamic framework10,11,12,13 to account for the formation of MY ENSO events, by explaining how ENSO duration can be prolonged. For example, DiNezio and Deser14 demonstrated that incorporating nonlinear dynamics into delayed oscillators can extend ENSO duration beyond a single year. An and Kim15, along with Im et al.16, further emphasized the role of nonlinear ocean wave responses to wind in sustaining ENSO events. Iwakiri and Watanabe17 tuned the recharge oscillation framework with an observed weak recharge efficiency parameter, enabling the dynamic framework to reproduce the occurrence of MY ENSO. Using recharge ENSO dynamics, Wu et al.18,19,20 linked MY La Niña to the substantial OHC discharge caused by a preceding strong El Niño, which often prevents full recharge within a single cycle, leading to a second La Niña. Their work also discovered El Niño onset time related to its duration, with El Niño events that develop early tend to terminate quickly after the mature phase due to the early reflection of delayed negative oceanic feedback.

Efforts have also been made to explore atmosphere-ocean processes in the extratropical Pacific, particularly the North Pacific Meridional Mode (NPMM)21, as a mechanism for MY ENSO formation9,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36. The role of NPMM in ENSO dynamics was initially highlighted to explain the formation of Central Pacific (CP) ENSO27,28 whose sea surface temperature anomalies (SSTAs) center more around the International Dateline and influenced more by the variation of northeast trade winds originated from the northern subtropical Pacific but less by the variation of equatorial thermocline variations due to its relatively deeper mean thermocline depth29. The NPMM was further suggested to also play a role in the formation of MY ENSO24. Specifically, Yu and Fang24 proposed that El Niño (La Niña) events in the tropical central Pacific can trigger the positive (negative) phase of the NPMM during their decay phase. The SSTAs associated with this NPMM phase can then spread into the tropical central Pacific in the following seasons via the wind-evaporation-SST (WES) feedback mechanism30, potentially triggering another El Niño (La Niña) and resulting in MY ENSO events. Several additional studies have further substantiated the contribution of extratropical North Pacific processes to the development of MY ENSO events23,31,32.

An increasing number of studies have revealed that atmosphere-ocean processes originating from the extratropical South Pacific can also significantly impact ENSO dynamics33,34,35,36,37,38,39. The South Pacific Meridional Mode (SPMM)39, the Southern Hemisphere counterpart of the NPMM, was initially identified as a factor influencing the development of Eastern Pacific (EP) ENSO. Li et al.35 further discovered that ENSO can influence South Pacific sea level pressure (SLP) variability modes, leading to the development of the South Pacific Oscillation (SPO)38,40, which is a key driver of the SPMM. The SPO, a basin-scale climate mode over the South Pacific, can influence ENSO not only through the SPMM in the southeast Pacific but also via atmosphere-ocean coupling processes in the western Pacific34,37,41. These mutual interactions suggest that climate variability in the South Pacific may be initiated by an ENSO event and subsequently feedback to influence its further evolution. As a result, off-equatorial processes from the South Pacific may play a significant role in shaping the formation of MY ENSO events. This possibility has been tentatively examined in several recent case studies, with research showing that the 2020–2023 triple-dip La Niña was largely influenced by South Pacific SST and wind anomalies5,6,7,42,43.

Motivated by the proposed relay role of the South Pacific climate modes between ENSO events, this study investigates specifically the potential differentiating impacts of the SPO on the evolution of MY and SY ENSO events. Our research encompasses observational data from 1950 to 2023, Coupled Model Intercomparison Project Phase 6 (CMIP6) historical outputs, and forced atmospheric general circulation model (AGCM) experiments. Our findings reveal that the SPO significantly differentiates between MY and SY ENSO events, both El Niño and La Niña. These results underscore the crucial role of South Pacific climate variability in shaping ENSO evolution patterns.

Results

Significant SPO differences between MY and SY ENSOs

The SPO is characterized by a north-south dipole structure in the South Pacific SLP field, with its positive phase featuring an anomalous low-pressure center over the subtropics and a high-pressure center at higher latitudes (Supplementary Fig. 1)35,38,40. We first compared the SPO index (SPOI; see definition in Methods) between MY and SY ENSO events during the second-year boreal spring (February–April, FMA1) season (Fig. 1a, b) to examine whether contrasting SPO conditions exist between these two duration groups of ENSO events. In this study, we define Year 0 as the year when the first ENSO event develops and matures, and Year 1 as the following year.

Fig. 1: Comparison of South Pacific Oscillation between multi-year and single-year ENSO events.
figure 1

a Composite SPO indices for MY and SY El Niño events in FMA1 from 1950 to 2023, and b for MY and SY La Niña events, with error bars indicating the 95% confidence intervals. c Composite monthly SPO indices for MY (red line) and SY (pink line) El Niño events, and d for MY (blue line) and SY (azure line) La Niña events. c, d The evolution of the SPO indices from the first year to the second year of the ENSO lifecycle. All SPO indices are smoothed using a 3-month window and standardized for each calendar month. Dots over the line of MY ENSO represent months where the differences in SPO indices between MY and SY ENSO events are significant at the 95% confidence level. Composite SPO spatial patterns (shading; hPa) in FMA1 for e MY El Niño, f SY El Niño, g MY La Niña, and h SY La Niña. White dots represent the SLP anomalies that are significant at the 95% confidence level. Spatial correlation coefficients (R) between SLP anomalies over the South Pacific (75°S–0°, 180°–90°W) and the typical SPO pattern (Supplementary Fig. 1) are shown in the upper left corner of each panel. Three asterisks indicate that the correlation coefficients have passed the 99% confidence test.

Full size image

We find that the composite SPOI is strongly positive for MY El Niño, reaching 0.89 standard deviations (s.d.), but remains near zero for SY El Niño at −0.26 s.d., indicating a clear difference (Fig. 1a). During La Niña, this contrast becomes even more pronounced, with the SPOI strongly negative at −0.68 s.d. during MY La Niña, while notably positive at 0.63 s.d. during SY La Niña (Fig. 1b). The differences in the SPOI between MY and SY ENSO events are statistically significant at the 95% confidence level for both El Niño and La Niña. The seasonal evolutions of the composite SPOI (Fig. 1c, d) show that these differences persist through the boreal winters of Year 0 and Year 1. However, no significant differences are detected during the spring of Year 0.

We also analyzed the composite SLP anomalies during FMA1 for MY and SY El Niño and La Niña events (Fig. 1e–h) and found that all MY ENSO composites exhibit a clear resemblance to the SPO pattern (Supplementary Fig. 1). The composite SLP anomaly structure for MY El Niño closely aligns with the positive SPO phase, showing a strong pattern correlation of 0.80 (p < 0.01; Fig. 1e). In contrast, the SLP anomaly pattern for SY El Niño does not display a distinct SPO signature (Fig. 1f), with only a weak spatial correlation (R = −0.13, p > 0.2). Similarly, the SLP pattern associated with MY La Niña is characterized by a negative SPO phase (Fig. 1g), with a high spatial correlation of −0.84 (p < 0.01), whereas SY La Niña lacks this negative SPO pattern entirely (Fig. 1h). These results highlight the presence of a robust SPO signal in MY ENSO events, while such a signal is notably absent in SY ENSO events.

In this study, MY ENSO events were defined using a relatively lenient second-year criterion, requiring only that the Niño3.4 index maintain the same sign as the first year, without needing to exceed the ENSO threshold (see Methods and Supplementary Table 1)2,44,45. For validation, a stricter criterion (second-year Niño3.4 > 0.5 s.d.) was also tested, and the SPO differences between MY and SY ENSO events remained robust (Supplementary Table 2). However, this stricter definition reduced MY El Niño samples to just five, limiting statistical confidence. The more flexible criterion ultimately provided a larger sample size and improved statistical robustness.

To overcome the limited sample sizes of observed MY and SY ENSO events, we examined 165-year historical simulations (1850–2014) from 45 CMIP6 models (Supplementary Table 3) to evaluate the robustness of SPO differences. Most models consistently simulate a positive SPOI for MY El Niño (82%, 37 out of 45 models; Fig. 2a) and a negative SPOI for MY La Niña (89%, 40 out of 45 models; Fig. 2c). In contrast, no consistent SPOI signal emerges for SY ENSO events, with values fluctuating around zero (Fig. 2b, d). The multi-model ensemble mean (MME) further highlights this contrast: MY El Niño and La Niña show significant positive and negative SPOI (0.16 and −0.19 s.d.), respectively, while SY events show near-zero SPOI (−0.04 and 0.01 s.d.). These model results confirm the key observational finding that the SPO is a distinguishing feature of MY ENSO events. They also demonstrate that CMIP6 models can reasonably capture this important difference between MY and SY ENSO.

Fig. 2: Composite SPO indices for multi-year and single-year ENSO events across CMIP6 models.
figure 2

The mean SPO indices for each CMIP6 model (hollow bars) are presented for a MY El Niño, b SY El Niño, c MY La Niña, and d SY La Niña during the second-year boreal spring. The solid bars on the far right of each subplot represent the ensemble mean of 45 models. Error bars indicate the 95% confidence interval.

Full size image

Mechanisms for different SPO conditions under MY and SY ENSOs

Previous studies have revealed that SPO can be forced by tropical Pacific SST variations35. Therefore, the differing SPO conditions between MY and SY ENSO events may be linked to differences in their tropical Pacific SSTA patterns. Fig. 3a–d show the composite SSTA patterns for different types of ENSO during FMA1 based on observations. For El Niño (Fig. 3a, c), the SSTAs in the MY El Niño have a more westward central location than SY El Niño (159.4°W vs. 151.4°W; see Methods for definition), while SSTA intensities are comparable with similar Cold Tongue Index (CTI) value (0.58 °C vs. 0.57 °C). For La Niña (Fig. 3b, d), the MY La Niña displays a much stronger magnitude (−0.65 °C) than the SY La Niña (−0.22 °C) and a more westward central location (162.6°W vs. 155.9°W).

Fig. 3: Linkages between the SPO and simultaneous ENSO properties.
figure 3

Composite SST anomalies (shading; °C) are shown for a MY El Niño, b SY El Niño, c MY La Niña, and d SY La Niña during FMA1. The CTI strength and the equatorial longitudinal location of maximum (minimum) SST anomalies are indicated in each panel. Scatter plots of the SPOI versus e CTI for El Niño, f CTI for La Niña, g SSTA center for El Niño, and h SSTA center for La Niña. In (e–h), red (pink) dots represent MY (SY) El Niño, and blue (azure) dots represent MY (SY) La Niña. Correlation coefficients (R) between the CTI (or SSTA center) and SPOI are indicated in each panel, with three asterisks denoting that the correlation coefficients have passed the 99% confidence test.

Full size image

To examine the possible impact of ENSO intensity on the occurrence of the SPO, we examine in Fig. 3e, f the linear linkage between the ENSO intensity (CTI) and SPOI. A notable correlation is found for La Niña, where stronger La Niña events are associated with more pronounced negative SPO, while moderate La Niña events are linked to neutral or even positive SPO (R = 0.63, p < 0.01; Fig. 3f). In contrast, this correlation is weak for El Niño, showing no clear trend that stronger El Niño events may correspond to stronger positive SPO (R = −0.06; Fig. 3e). This contrasting strong influence of La Niña intensity to SPO and weak influence of El Niño intensity to SPO remain consistent across various ENSO intensity indices (see results of Niño3.4, Niño3, and Niño4 indices in Supplementary Table 4).

Next, we assess the potential influence of ENSO location on SPO occurrence by examining the relationship between ENSO zonal location (see Methods for definition) and SPOI (Fig. 3g, h). For El Niño, we find a moderate negative correlation (R = −0.30), suggesting that more westward events are more likely to trigger a positive SPO, though this correlation does not reach the 90% confidence level (p < 0.2). In contrast, La Niña shows almost no relationship between zonal location and SPO (R = −0.07). Notably, all MY El Niño events occur when the central SSTA is located west of 125°W.

In general, La Niña events tend to be centered over the central equatorial Pacific, while El Niño events can occur across both the central and eastern equatorial Pacific (Fig. 3g, h). Consequently, El Niño events are typically categorized into EP and CP types27,28. The reason the SPO responds to El Niño varies with its type, specifically the central location of the El Niño’s SST anomalies, is that El Niño events must trigger deep convection to produce a teleconnection to the Southern Hemisphere, which gives rise to the SPO. Background SSTs in the central equatorial Pacific often exceed 27.5 °C (Supplementary Fig. 2), the threshold necessary to initiate deep convection. Therefore, SST anomalies associated with CP El Niño events can easily excite anomalous deep convection, facilitating the development of the SPO. In contrast, the background SSTs in the eastern equatorial Pacific are generally well below the 27.5 °C threshold. As a result, even strong warming associated with EP El Niño events is less effective at triggering deep convection, and thus less capable of inducing the SPO teleconnection.

The dependence of the SPO response on the location of El Niño warming is clearly illustrated in Supplementary Fig. 3, which compares SLP anomalies in the South Pacific during CP and EP El Niño events. During CP El Niño events, strong negative SLP anomalies dominate the SPO region and extend westward to 180°W. In contrast, EP El Niño events exhibit weaker negative SLP anomalies, with the anomaly center confined east of 150°W. Supplementary Fig. 4 further demonstrates that even the three strongest EP El Niño events (1982/83, 1997/98, and 2015/16) fail to generate a significant SPO response. As for CP El Niño events, Fang and Yu24 found that once the SST exceeds the 27.5 °C threshold for deep convection, further increases in SST anomaly amplitude do not substantially enhance convective activity. Together, these results explain why the longitudinal location of El Niño warming is more critical than its intensity in driving the development of the SPO.

In contrast, La Niña events are almost exclusively centered in the central equatorial Pacific (Fig. 3h). In this region, background SSTs are typically just above the threshold required to initiate deep convection. Strong negative SST anomalies during intense CP La Niña events can lower local SSTs below this threshold, completely suppressing deep convection. This suppression leads to substantial changes in latent heat flux, which in turn generates significant atmospheric teleconnections to the South Pacific and gives rise to a strong SPO response. Conversely, weaker CP La Niña events do not reduce SSTs enough to inhibit deep convection, resulting in smaller heat flux changes and a weaker SPO response. Thus, for La Niña events, the magnitude of the SST anomalies plays a key role in determining the strength of the SPO response.

The above observational analyses indicate that the central location of El Niño and the intensity of La Niña are key factors in shaping SPO responses during boreal winter following the first-year ENSO peak. To further validate the controlling role of ENSO’s location and magnitude on SPO formation, we conducted a 30-member ensemble experiment using the Geophysical Fluid Dynamics Laboratory Atmospheric Model version 2.1 (GFDL AM2.1) model46. The model was forced by observed monthly SSTAs from 1951 to 2018 within the tropical Pacific domain (20°S–20°N, 140°E–90°W, with a 5° inner buffer zone), while maintaining climatological SSTs elsewhere (see Methods for details). We focused on evaluating the simulated SPO responses during prescribed MY and SY ENSO events (Supplementary Table 1). To isolate the ENSO forcing, we examined the 30-member ensemble mean of the simulated SPOI (Fig. 4a, b). Consistent with observations, MY El Niño events forced a stronger positive SPO than SY El Niño events, while MY La Niña events forced a stronger negative SPO than SY La Niña events.

Fig. 4: Simulated responses of the SPO to ENSO events in the forced AGCM experiments.
figure 4

The model experiments were conducted using prescribed monthly varying SST anomalies over the tropical Pacific (20°S–20°N, 140°E–90°W). a, b The 30-member ensemble mean SPOI during the second-year boreal spring for composite MY El Niño (red bar) and SY El Niño (pink bar), and for MY La Niña (blue bar) and SY La Niña (azure bar) events, respectively. Error bars indicate the 95% confidence interval. Box plots of simulated SPOI distributions for individual events, including c 10 MY El Niño events, d 14 SY El Niño events, e 18 MY La Niña events, and f 9 SY La Niña events. Each boxplot represents the distribution across the 30 ensemble members. Colored dots overlaid on the box plots indicate observed SPOI values for reference. The whiskers denote the maximum and minimum values, while the box boundaries correspond to the interquartile range (25th–75th percentiles) of the simulated SPOI. Years on the x-axis indicate the first developing year of each ENSO event.

Full size image

However, a notable discrepancy between observations and the forced atmospheric model experiment emerged. In the simulations, SPOI in SY ENSO events retains the same sign as in MY ENSO events and remains significantly different from zero (Fig. 4a, b). In contrast, observations show that SPOI in SY ENSO events differs substantially from that in MY ENSO events, even exhibiting opposite signs (Fig. 1a, b). This discrepancy suggests that internal atmospheric variability, largely reduced or removed in the 30-member ensemble mean, may be playing a role in modulating the ENSO-forced SPO in observations. This effect is particularly pronounced in SY El Niño and SY La Niña events, where the weaker ENSO-forced SPO allows internal variability to drive substantial deviations, bringing the values closer to those observed. It should be noted that the discrepancy between observations and the forced atmospheric model results may also be influenced by model dependence, and further studies using multiple AGCMs would help to assess the robustness of our findings.

To assess the influence of internal variability, we examined the spread of SPOI across the 30-member ensemble for each historical ENSO event (Fig. 4c–f). For most events (45 out of 51), the observed SPOI falls within the ensemble spread, indicating that observed SPO variance can largely be explained by ENSO forcing combined with internal atmospheric variability. Notably, for SY El Niño events, observed SPOI tends to cluster near the lower edge of the ensemble spread (Fig. 4d), while for SY La Niña events, it tends to sit near the upper edge (Fig. 4f). In contrast, MY El Niño and La Niña events show no such systematic bias (Fig. 4c, e). Among 14 SY El Niño events, 8 have observed SPOI below the 75th percentile, suggesting internal variability may weaken the simulated positive SPO response (Fig. 4a), shifting it toward the observed negative SPOI (Fig. 1a). Similarly, for 9 SY La Niña events, 6 show observed SPOI above the 25th percentile, implying internal variability could shift the simulated negative SPOI (Fig. 4b) toward the observed positive SPOI (Fig. 1b). These results suggest internal atmospheric variability plays a larger role in shaping SPO for SY ENSO events compared to MY ENSO events.

Taken together, these findings suggest that ENSO characteristics initially determine the relative strength of the SPO. Subsequently, atmospheric internal variability plays a role. When ENSO forcing is weak, internal variability is more likely to diminish the ENSO-forced SPO to near zero or even reverse its phase. Conversely, when ENSO forcing is strong (due to high intensity or a more westward-located central ENSO SST anomalies), the ENSO-forced SPO response can maintain a relatively strong level even when affected by internal variability.

Feedback from the South Pacific on different ENSO evolutions

In this section, we further explore how the SPO, shaped by the combined effects of first-year ENSO forcing and atmospheric internal variability, feedbacks to the tropical Pacific and influences the evolution of second-year ENSO. To investigate the feedback processes, we examined the evolution of SST and 10 m horizontal wind anomalies, composited for MY and SY ENSO events, from FMA1 to ND1J2 (November–January) in observation (Supplementary Figs. 5 and 6).

During FMA1, the Southwest Pacific exhibits southerly surface wind anomalies associated with the strong positive SPO in MY El Niño (about 0.5 m/s; Fig. 5a). Given that the climatological 10 m wind over the Southwest Pacific is southeasterly (Supplementary Fig. 7), the positive SPO-induced southerly anomalies enhance the climatological winds, increasing local evaporation and cooling SSTs. This leads to a gradual cooling in the Southwest Pacific during MY El Niño (Fig. 5a). This cooling sustains from FMA1 through ND1J2 (Fig. 5a) and spreads toward the equatorial far-western Pacific via the WES mechanism, strengthening SSTA cooling there (Supplementary Fig. 5a). In SY El Niño, the weak anomalous southerly during FMA1 shifts to northerly during MJJ1 (March–July; −0.21 m/s) (Fig. 5a), causing warm SST anomalies to develop due to the weakening of the climatological southerlies (Fig. 5a and Supplementary Fig. 5b). These warm SST anomalies intensify through MJJ1, ASO1 (August–October), and ND1J2 (Fig. 5a).

Fig. 5: Quantitative examination of the feedback from the SPO to subsequent ENSO events.
figure 5

a–c Seasonal evolutions of regional mean values for 10 m zonal wind (m/s), meridional wind (m/s), and SST anomalies (°C) in MY El Niño (orange bars) and SY El Niño (gray bars). Values are averaged over a Southwest Pacific (15°S–35°S, 160°E–170°W), b Southeast Pacific (15°S–35°S, 76°W–103°W), and c Tropical Pacific (CTI) regions. Progressively darker shades correspond to seasons FMA1, MJJ1, ASO1, and ND1J2, respectively. d–f The same variables as in (a–c), but for La Niña events, comparing MY (blue bars) and SY (gray bars).

Full size image

In the Southeast Pacific, westerly wind anomalies appear during FMA1 in MY El Niño (about 0.26 m/s), accompanied by warm SST anomalies (Fig. 5b). Due to the background southeasterly (Supplementary Fig. 7), these westerly winds and warm SSTAs reinforce each other through the WES feedback, maintaining the warm SSTAs and driving them toward the tropics in the subsequent seasons (Fig. 5b and Supplementary Fig. 5a). In SY El Niño, negative SPO-related strong easterly and southerly winds take over in MJJ1, resulting in localized cold SSTAs that develop toward the equator (Fig. 5b and Supplementary Fig. 5b).

Combining the SPO-induced impacts over the western and eastern Pacific, we find that during MY El Niño, colder SSTs in the far-western Pacific and warmer SSTs in the eastern Pacific—both driven by the positive SPO—create a positive SST gradient across the tropical Pacific, enhancing equatorial westerly winds (Supplementary Fig. 5a). These westerly winds cause a lowering of the thermocline in the eastern equatorial Pacific through eastward-propagating Kelvin waves, which help further develop warm SSTs in the east-central Pacific. The CTI remains positive and continues to strengthen from MJJ1 through ND1J2 (Fig. 5c), contributing to the first-year El Niño becoming an MY event. In contrast, for SY El Niño, warm SSTs in the western Pacific and cold SSTs in the eastern Pacific generate a negative SST gradient, weakening the residual equatorial westerly winds from the previous year’s El Niño. This reversal shifts the first-year CTI from a warm phase in FMA1 to a cold phase in MJJ1, ASO1, and ND1J2 (Fig. 5c). Thus, the absence of a positive SPO prevents the warm SSTs of the first-year SY El Niño from sustaining into the second year, leading to a failure to produce consecutive El Niño events.

A similar but opposite evolution occurs for SY and MY La Niña events (Fig. 5d–f and Supplementary Fig. 6). The bars shown in Fig. 5d–f for MY and SY La Niña events are generally reversed compared to the bars shown in Fig. 5a–c for MY and SY El Niño events. In brief, the negative SPO during MY La Niña leads to warm SSTAs in the tropical western Pacific and cold SSTAs in the tropical eastern Pacific, forming a negative zonal SST gradient that sustains equatorial easterly winds, maintaining the event as a MY La Niña (Supplementary Fig. 6a). In SY La Niña, the opposite occurs (Supplementary Fig. 6b). As a result, the absence of a negative SPO in SY La Niña prevents the cold SSTs from persisting into the second year, thus inhibiting the formation of consecutive La Niña events.

It is worth noting that several previous studies have shown that a positive SPO tends to induce SST warming in the tropical Pacific, while a negative SPO often leads to cooling35,37,38. Those studies conducted coupled climate model experiments to investigate extratropical–tropical interactions by prescribing SPO-related SST anomalies and analyzing the resulting response in the tropical Pacific. Therefore, we do not repeat similar numerical experiments in this study to reconfirm these interaction mechanisms. However, it is important to highlight that these modeling studies did not address the potential influence of the SPO on the evolution of SY and MY ENSO events. This aspect is the central focus of the present study, in which we examine the feedback role of the SPO following the first year of ENSO development.

Enhanced predictability of MY ENSO via the SPO

ENSO’s prediction skills exhibit seasonal variations, with a strong spring predictability barrier (SPB)47,48,49,50, which means the lowest prediction skill for the prediction implemented starts from the boreal spring season. The SPO precursor we employed in this study is in the boreal spring season, which may potentially reduce SPB and enhance ENSO predictability.

Therefore, we conducted a predictability assessment by comparing the SPO-based forecast approach with the operational ENSO forecast provided by the International Research Institute for Climate and Society (IRI). Since the earliest available IRI forecast begins in 2002, we limited our comparison to ENSO events occurring after that year. For the IRI forecasts, we focused on those issued in April and targeting the NDJ season of the decaying year of ENSO events. After an El Niño, if the IRI forecast assigned more than a 50% probability to another El Niño occurring, we classified it as a forecast of a MY El Niño event; otherwise, it was considered a forecast of a SY El Niño event. The same criteria were applied to La Niña cases. For the SPO-based forecast, we used the SPO index from the FMA season. If the SPO index was greater than zero after an El Niño, we forecast a MY El Niño event; if less than zero, a SY El Niño. Similarly, after a La Niña, a negative SPO index forecasted a MY La Niña event, while a positive index indicated a SY La Niña.

The forecast accuracies are summarized in Supplementary Tables 5–8. The IRI forecast correctly predicted: 2 out of 3 MY El Niño events (67%), 1 out of 6 MY La Niña events (17%), 3 out of 5 SY El Niño events (60%), and 3 out of 4 SY La Niña (75%). In summary, SY La Niña events had the highest predictability (75%), followed by MY El Niño (67%), SY El Niño (60%), and the most challenging one, MY La Niña (17%). Comparing across ENSO phases, MY and SY El Niño events exhibited comparable predictability (67% vs. 60%), but MY La Niña events were significantly harder to forecast than SY La Niña (17% vs. 75%). Overall, MY ENSO events were forecast with much lower accuracy than SY ENSO events (33% vs. 67%). Given the increasing occurrence of MY La Niña events in recent years and projected into the future1,2,3, improving the prediction skill for MY ENSO, especially the MY La Niña, is particularly urgent.

We then evaluated the prediction skill of the MY and SY ENSO events using the SPO precursor. The SPO-based method correctly forecasted: 2 out of 3 MY El Niño events (67%), 3 out of 6 MY La Niña events (50%), 3 out of 5 SY El Niño events (60%), and 4 out of 4 SY La Niña events (100%). Notably, the SPO precursor substantially improved the forecast accuracy for MY La Niña events (50% vs. 17% from IRI), while maintaining similar performance in the other categories. This indicates that the SPO index offers a simple and effective tool for enhancing ENSO prediction skill where it is most needed. We further extended the SPO forecast analysis to the full 1950–2023 period. The results were as follows: 8 out of 10 MY El Niño (80%), 14 out of 18 MY La Niña (78%), 9 out of 14 SY El Niño (64%), and 8 out of 9 SY La Niña (89%) were correctly forecasted using the SPO-based method. With this larger sample, the SPO-based forecasts exhibited even higher skill, especially for MY events: (El Niño: 80%; La Niña: 78%). This suggests possible interdecadal variations in the SPO–ENSO relationship. While this warrants further investigation, such analysis is beyond the scope of the present study. Overall, the SPO precursor achieved an impressive 79% accuracy for forecasting MY ENSO events, underscoring its strong potential in operational seasonal prediction.

Summary and discussions

The SPO plays a crucial role in distinguishing between SY and MY ENSO events, with significant positive (negative) SPO anomalies linked to MY El Niño (La Niña) events, while SY El Niño (La Niña) events are associated with the absence of such anomalies. This relationship, observed in both reanalysis data and CMIP6 simulations, is driven by first-year ENSO properties, including the longitudinal location of El Niño and the intensity of La Niña, with South Pacific atmospheric internal variability further modulating the SPO response. Unlike previous studies that primarily focus on one phase of ENSO, explaining why La Niña tends to persist over multiple years51,52, the SPO proxy used in this study is applicable to both El Niño and La Niña phases and can be regarded as a unified mechanism for MY ENSO. Compared to the forecast skill of dynamical and statistical models from IRI, our SPO-based approach shows enhanced ENSO forecast skill, especially in MY La Niña. These findings have significant implications for our understanding of MY ENSO dynamics and prediction.

Besides the SPO, other climate modes in the South Pacific, like the Southern Annular Mode53,54, the Pacific-South American Mode55, the South Pacific Quadrupole Mode33, and the SPMM, may also play roles in differentiating MY and SY ENSO, and could be further investigated in future research. In addition, given the significant contribution of South Pacific air-sea variability to the recent 2020–2023 triple-dip La Niña event5,6,7, it is possible that the SPO–ENSO relationship has undergone decadal variations. Therefore, it is worthwhile to examine the decadal evolution of the SPO–ENSO relationship, its potential connection to the South Pacific Decadal Oscillation56,57 or the Pacific Decadal Oscillation58, and its future projection under climate change. While this is an important direction for future research, such an analysis is beyond the scope of the present study.

This study highlights the important role of extratropical Southern Hemisphere processes in shaping MY ENSO evolution, a factor that has been less emphasized compared to influences from other basins. It is important to note that the highlighted role of South Pacific climate variability in distinguishing MY and SY ENSO does not preclude contributions from other regions, including the tropical Pacific, North Pacific, Atlantic, and Indian Ocean23,59,60,61,62,63,64,65. In fact, certain ENSO events, such as those in 1968/69, 1975/76, 2004/05, and 2016/17, cannot be fully explained by South Pacific processes alone (Fig. 4c–f). Future studies should examine the combined and potentially interactive roles of these basins in modulating ENSO evolution.

Methods

Reanalysis data

This study utilized global monthly SST data from the Hadley Centre Global Sea Ice and Sea Surface Temperature dataset (HadISST), which has a horizontal resolution of 1° × 1° 66. SLP data and horizontal 10 m wind data were sourced from the National Centers for Environmental Prediction-National Center for Atmospheric Research reanalysis 1 (NCEP/NCAR R1) dataset with a resolution of 2.5° × 2.5° 67. The analysis spans the period from 1950 to 2023. To mitigate the influence of global warming, long-term linear trends were removed from the data prior to analysis. Monthly anomalies were computed by subtracting the climatological monthly means calculated over the entire analysis period.

CMIP6 historical outputs

To evaluate the distinct influence of the SPO on MY and SY ENSO, we employed CMIP6 models68. Monthly historical simulations, specifically SST and SLP variables, were utilized from 45 different CMIP6 models (Supplementary Table 3). The analysis covered the same period across all 45 models, spanning 165 years from 1850 to 2014. This comprehensive dataset allows for a robust assessment of the SPO’s role in different ENSO types across a range of climate models.

Definitions of climate indices

The SPO index was defined as the reversed average of SLP anomalies over the region 46°S–27°S and 165°E–82°W, both in observation and modeling35,38. Therefore, the positive phase of the SPO corresponds to negative SLP anomalies over the subtropical South Pacific (Supplementary Fig. 1). To quantify the strength of ENSO, several Niño indices were employed, each defined as the SSTAs averaged across various specific regions of the tropical Pacific: CTI (6°S–6°N and 180°–90°W), Niño3.4 index (5°S–5°N and 170°–120°W), Niño3 index (5°S–5°N and 150°W–90°W), and Niño4 index (5°S–5°N and 160°E–150°W). Longitudinal location of El Niño (La Niña) was determined as the longitude of the maximum (minimum) SST anomalies, after computing the 5°S–5°N meridional mean and applying a 10-degree zonally running mean.

Identified MY and SY ENSO events

We employed a commonly used definition to identify ENSO events and their different evolution types2,44,45. An El Niño (La Niña) event was identified when the NDJ running-mean Niño3.4 index exceeded 0.5 (−0.5) s.d. Applying this criterion, we identified a total of 24 El Niño and 27 La Niña events (Supplementary Table 1) from 1950 to 2023. Further classification into MY or SY ENSO types was based on the Niño3.4 index during the second-year winter. ENSO events were categorized as MY El Niño (La Niña) if the second-year NDJ1 Niño3.4 index was greater (less) than zero, and SY El Niño (La Niña) if it was less (greater) than zero2,44,45. We adopted this relatively loose criterion for two primary reasons. Firstly, this criterion serves to augment the pool of samples available for analysis, a crucial factor given the scarcity of MY ENSO events. Secondly, our study primarily aims to unveil the diverse underlying physical processes influencing the persistence or diminution of SST warming/cooling during the second year of ENSO events. Notably, there is no inherent expectation that the physical mechanisms sustaining SSTA in the second year should necessarily lead to warming exceeding a specific threshold (e.g., 0.5 s.d.). The classification patterns for all 51 identified El Niño and La Niña events are provided in Supplementary Table 1. The same criteria were applied for CMIP6 data to identify ENSO events and their different evolution types.

Identified CP and EP ENSO events

ENSO events are classified as CP or EP types based on the relative strength of the Niño4 and Niño1+2 indices. Th Niño4 index is defined as the SSTAs averaged over 5°S–5°N, 160°E–150°W, while the Niño1+2 index is defined over 10°S–0, 90°–80°W. An ENSO event is categorized as a CP (EP) type if the absolute value of the ND0J1 Niño4 index exceeds (is less than) that of the Niño1+2 index.

Climate model experiments

The atmospheric model employed in this study is the GFDL AM2.146. This model features a horizontal resolution of 2.5° in longitude by 2° in latitude and consists of 24 vertical layers. For the experiment, we added historical monthly SSTAs (1951–2018) within the tropical Pacific domain (20°S–20°N and 140°E–90°W) to the modeled monthly varying climatological SSTs. Outside the tropical Pacific region, the model was forced solely by climatological SSTs. The experiment included 30 ensemble members, each initialized with slightly different disturbances and integrated for 68 years. This experiment design allows us to assess whether the SPO differences between MY and SY ENSO can be reproduced using ENSO-related tropical Pacific SST forcing alone.

Significance tests

The statistical significance of the composites and correlations was assessed using a two-tailed Student’s t-test. The effective degrees of freedom (\({N}_{{\rm{eff}}}\)) were calculated based on the method proposed by Bretherton et al.69:

$${N}_{{\rm{eff}}}\approx N\frac{1-{r}_{x}{r}_{y}}{1+{r}_{x}{r}_{y}}$$
(1)

where \({N}\) is the total number of samples, and \({r}_{{x}}\) and \({r}_{{y}}\) represent the lag-1 autocorrelations of variables \(x\) and \(y\), respectively.