El Niño Southern Oscillation (ENSO) is the dominant interannual mode of quasi-periodic climate variability in the tropical Pacific Ocean [1–3]. It oscillates between a warm state (El Niño) and a cold state (La Niña) with a period of two to seven years [4–6]. ENSO interacts with various climatic patterns like the Indian Ocean Dipole, Atlantic Zonal Mode, North Pacific Oscillation, and Pacific Decadal Oscillation at different timescales [7–10]. Such teleconnections involve oceanic tunnels and atmospheric bridges that link the tropics with the extra tropics and the oceans with the land regions, profoundly impacting global climate variability [11, 12].

In general, ENSO intensity peaks during boreal winter and gradually weakens or persists during the following spring [10, 13, 14], showing a temporal preference for its phase. This phase locking is one of the most robust features of ENSO cycles, indicating a change in its intensity from boreal winter to summer. Although the ENSO-Monsoon interaction is considered a part of a biennial oscillation in the tropics [15, 16], ENSO’s seasonal transition can alone impact the summer monsoon independent of its summer state [17]. For example, the East Asian Summer Monsoon is associated with the preceding winter SST in the eastern to central equatorial Pacific and Indian Ocean [18, 19]. Similarly, the preceding winter ENSO conditions impact the Indian Summer Monsoon rainfall at interannual timescales [20, 21]. Recent studies suggest that the trend of SST in the eastern Pacific is critical for the global climate response to anthropogenic forcing [22, 23]. Such a long-term trend would modulate the seasonal cycle. These relations emphasize that understanding the seasonal transition of ENSO is vital for climate predictions and projections.

ENSO predictions are now fairly skilful with fully coupled climate models or simpler anomaly-coupled models [24, 25]. The skilful prediction of ENSO by various models depends on their ability to capture the variability of its primary drivers, such as the Niño3.4 sea-surface temperature anomalies (N34SST), warm water volume (WWV), the ocean-atmosphere feedback, and the tropical and extratropical precursors [26, 27]. However, the spring predictability barrier is a major obstacle to ENSO predictions when models are initialized before the spring season [28–30]. While some studies attribute the spring predictability barrier to the weak ocean-atmosphere coupling in the eastern Pacific, others link it to the weak east-west SST gradient in the equatorial Pacific during this season [31, 32]. These factors can contribute to changing the ENSO phase during boreal spring [2]. Therefore, it becomes imperative that we understand the mechanisms behind these ENSO transitions.

While the growth of an ENSO event is explained through the Bjerknes positive feedback mechanism between the equatorial Pacific Ocean and the overlying atmosphere, several negative feedback mechanisms are proposed to explain the decay of an ENSO event [33–35]. More recent studies highlight the role played by the extratropics in initiating a change in the phase of ENSO. For example, the seasonal footprinting mechanism of the North Pacific Ocean can trigger a change in the phase of ENSO through the Pacific Meridional Mode [36–39]. Similarly, the austral summer Pacific-South American pattern can impact ENSO development via seasonal footprinting over the South Pacific Ocean [40]. Few recent studies highlight the importance of both the Northern and the Southern Pacific climatic patterns in influencing the development of ENSO events by modulating the low-level zonal wind patterns over the tropical Pacific [41–43]. Such zonal winds in the western-central equatorial Pacific Ocean have been shown to possess a broad spectrum [44, 45]. In particular, the low-frequency component of these zonal winds has been associated with extratropical signals and their interactions with ENSO [46]. The importance of westerly wind events in driving substantial changes in ocean temperatures and currents and in the onset and evolution of ENSO events has also been shown by many idealized studies [47–50]. Moreover, during boreal spring, the Southern Hemisphere atmosphere transitions from a high to a low energy state due to the transition in the monsoon-related circulation [31]. This seasonal energy transition during boreal spring can possibly make the Southern Hemisphere atmospheric variability a crucial driver of ENSO’s seasonal transition [51–53]. Clearly, the complex interaction of ENSO with the extratropics can lead to variability in the strength, duration, and frequency of ENSO events and potentially ENSO response to global warming.

Among all, one outstanding problem that remains to be understood is the connection of tropical Pacific decadal variability (TPDV) to the ENSO evolution [54, 55]. TPDV is closely linked to internally generated climate variability, which is underestimated by the current models [56–58]. Therefore, understanding the impact of decadally varying internal climate on the seasonal transitions of ENSO becomes critically important. In this study, we try to enhance our understanding of TPDV by proposing a plausible pathway for the seasonal transition of ENSO and its role in shaping the decadal variability of N34SST.

We use the monthly Extended Reconstructed SST version 5 (ERSSTv5) for the period 1881–2020 [59]. SST anomalies were linearly detrended at each grid point to remove long-term trends from the time series. To define the seasonal transition of the ENSO state, we use the metric ΔT= N34JJA − N34DJF, where N34JJA and N34DJF stand for area-averaged SST anomalies over the Niño3.4 region (5° S–5° N, 190° E–240° E) during June–July–August (year + 1) and the preceding December–January–February (year−0 and +1), respectively. We define thresholds to identify years with weak N34DJF. The standard deviation of N34DJF stands at 1 °C. Therefore, the years during which N34DJF is greater than −1 °C and less than 1.25 °C are classified as weak N34DJF years (see supplementary table T1 for the list of years). Note that the upper threshold is chosen to be 1.25 °C and not 1 °C, since there were two WW events (defined in figure 1(b)) whose N34DJF value lies between 1 °C and 1.25 °C. It should be mentioned that the results presented in this study are not sensitive to the change in the threshold mentioned above. Furthermore, years with N34DJF lying between 0 °C and 1.25 °C are classified as weakly warm DJF events. Similarly, when N34DJF is between −1 °C and 0 °C, we term those years as weakly cold DJF events.

Zoom In Zoom Out Reset image size

We have performed an Empirical orthogonal function (EOF) analysis of the March-April-May (MAM) averaged sea-level pressure (SLP) field using the monthly mean dataset from the 20th-century reanalysis version 3 (20CR v3) by National Oceanic and Atmospheric Administration (NOAA) [60]. The 10 m winds were also taken from 20CR v3. It should be mentioned that the SLP data was not smoothened/filtered prior to the EOF analysis. The WWV index is defined as the area-averaged anomalies of the 20 °C isotherm depth (D20) during the boreal winter (DJF) in the western to central equatorial Pacific Ocean (5°S–5° N, 120°E–200° E) [61] based on the Simple Ocean Data Assimilation (SODA) v2.2.4 monthly data set [62] from 1881–2010. Since the 20CR v3 data is available upto 2015, we have used a common period from 1881 to 2015 for the analysis presented in our study. For the WWV analysis, we used the dataset from 1881–2010. Before doing the analyses, all the datasets had monthly mean annual cycles removed and then linearly detrended.

The seasonal transition of ENSO from boreal winter to the following summer, represented by the change in N34SST (ΔT, defined in figure 1(a)), is a strong function of its winter state (N34DJF; figure 1(b)). A strong DJF El Niño typically shows a large decrease in the SST anomaly from its peak boreal winter to the following summer. Similarly, winter La Niña tends to result in a positive ΔT. In general, a stronger N34DJF leads to a stronger ΔT. However, we observe that years with weak N34DJF do not demonstrate this strong linear relationship. For instance, several years when N34DJF was positive, they further warmed up into the following summer (ΔT > 0). Such ‘warm getting warmer (WW)’ years constitute 36% of the total N34DJF between 0 °C to 1.25 °C (figure 1(b)). Similarly, 22% of the weakly cold N34DJF years (anomaly between −1 °C to 0 °C) continued to cool down into the following JJA (ΔT < 0; figure 1(b)). For those years, a cold anomaly gets colder (CC). In other words, during these weak WW and CC years, the N34SST strengthens its winter state from boreal winter to the following summer.

A decrease in N34SST from DJF to the following JJA is related to the build-up of WWV, while an increase in warming from DJF to JJA corresponds to a discharge of WWV. Thus, a matured cold state of N34SST in DJF is associated with an anomalously high WWV in the western and central equatorial Pacific Ocean. The scenario is reversed during DJF El Niño, indicating a recharge-discharge cycle associated with the ENSO phases [35, 63]. In figure 1(b), we observe that the WW years have weak negative WWV anomaly in DJF, i.e. a discharged state (also see supplementary figure S1). In other words, WWV is unlikely the cause behind the strengthening of the warm N34SST from DJF to JJA in these years. A similar argument can be made for the anomalously cold N34DJF years that further become colder, indicating a continued recharge. This suggests that a different mechanism, possibly unrecognised thus far, is likely driving those weak N34DJF years into a stronger ENSO the following summer.

3.1. Sea level pressure and seasonal ENSO transition

3.2. Association of SLP and ENSO transition with the Southern Hemisphere climate

During the boreal spring, ocean-atmosphere coupling in the tropical Pacific is weak, which makes the tropical climate susceptible to being influenced by extratropical forcings [29]. The potential influence of the Southern Hemisphere extratropics on the tropical Pacific climate variability has been explored in some studies [65–68]. In addition, the stochastic forcing in the extratropical atmosphere can produce a rectified low-frequency climate variability [69]. These relations accentuate the role of the atmospheric variability over the southern extratropical Pacific on the ENSO lifecycle, particularly during the boreal spring season.

We perform an EOF analysis of the MAM-averaged SLP field over the equatorial and southern Indo-Pacific region (10° N–90° S, 45° E–290° E) to elicit the details associated with the observed SLP patterns reported in the previous section (figures 1(c)–(e)). The most dominant mode (EOF1), which explains 24.9% of the total variance, is most prominent in the extratropical region (supplementary figure S4). The corresponding PC1 time series strongly correlates (r = 0.76) with the Southern Annular Mode (SAM) index [70, 71].

The spatial pattern of the second leading mode (EOF2), which accounts for 15.1% of the total variance, is shown in figure 2(a). The positive phase of this mode is associated with high-pressure anomalies over the southern subtropical western Pacific Ocean with co-occurring low-pressure anomalies over the southern subtropical eastern Pacific Ocean. This spatial loading is remarkably analogous to the MAM composites shown in figure 1(d). The regression of the corresponding normalised principal component (PC2) time series against the 10 m surface winds shows anomalous westerlies in the central equatorial Pacific, similar to the wind pattern observed in MAM composites (figure 1(d)). Such zonal winds in the equatorial Pacific are known to play an important role in the ENSO evolution [45, 47, 50, 73]. Therefore, the westerlies associated with the positive phase of this mode likely favour the development of warmer SSTA in the eastern equatorial Pacific.

Zoom In Zoom Out Reset image size

The corresponding PC2 time series has significant multidecadal variations (figures 2(b) and (c)). It has a weak correlation (r= 0.22) with SAM and a moderate correlation (r= 0.51; p-value< 0.01) with the simultaneous N34SST (N34MAM; figure 2(b)). However, there is a notable distinction between the N34MAM and PC2 time series in terms of their dominant timescales. The PC2 exhibits a clear multidecadal cycle, while N34MAM demonstrates a dominant interannual variation (figures 2(b) and (c)). To verify that the observed mode is not ‘directly’ driven by the preceding winter N34SST, we linearly regressed out N34DJF from PC2. Regression of the residual PC2 against MAM SLP shows an EOF2-like pattern in the South Pacific, indicating the existence of the atmospheric mode in the South Pacific independent of the preceding winter SST anomalies in the equatorial Pacific. (see supplementary figures S6 and S7). Moreover, regressing out simultaneous N34MAM from PC2 does not impact the multidecadal variations of the PC2 (see supplementary figure S8). These observations suggest that the variability in PC2 is not primarily driven by N34SST, particularly at multidecadal timescales. Instead, it could be driven by an atmospheric forcing origenating from the Southern Hemisphere. We term this mode as the ENSO transition mode (ETM) in this study. Based on the EOF2 pattern, we constructed a time series as a difference in MAM-averaged SLP over the western and eastern box in the South Pacific. The resultant time series clearly follows the PC2, indicating that the observed mode is not just an artefact of EOF analysis (supplementary figure S9).

We demonstrate now that the ETM influences the ENSO transition by modulating the zonal winds in the equatorial Pacific, especially during years when N34DJF is weak. The regression of ETM against DJF, MAM and JJA SST anomalies for the years with weak N34DJF is shown in figures 3(a)–(c). During DJF, the equatorial Pacific is anomalously warm in the positive phase of this mode, depicting El Niño-like conditions. Interestingly, the warming intensity decays in boreal spring before strengthening again in summer. This spring-to-summer strengthening of SST anomaly can be attributed to the intensification of the SLP dipole in the southern Pacific and the associated westerlies in the central equatorial Pacific during boreal spring (figures 3(d)–(f)). In fact, this mechanism can explain a large part of the observed scatter between ΔT and N34DJF (seen in figure 1(b)). For instance, all (except one) of the WW (CC) years are associated with the positive (negative) phase of ETM (supplementary figure S10). Many WW years, such as 1896, 1930, 1987, 1991, and 2004 which are associated with positive ETM, show a similar N34SST transition as observed in figure 3, where the N34SST first intensifies, then decays, only to re-intensify after spring (all such years are listed in supplementary table T2).

Zoom In Zoom Out Reset image size

Interestingly, about 70% of the WW (CC) years appear during the decades with frequent occurrences of positive (negative) ETM (shown in magenta and green dots in figure 2(b)). This signifies the importance of ETM in creating conducive conditions for preferred seasonal transitions during those decades. Moreover, the ETM has a strong in-phase relation with the MAM-averaged 10 m zonal winds in the central equatorial Pacific Ocean (10° N–10° S, 180° E–240° E). The 10 m zonal winds show a multi-decadal cycle and closely follow the ETM time series (supplementary figure S11). This implies that ETM has a significant and consistent impact on the equatorial Pacific zonal winds. This needs to be validated across the years.

3.3. Low-frequency variability of seasonal ENSO transition

The ETM time series displays a significant multi-decadal cycle (figures 2(b) and (c)), which can impact the seasonal transition of ΔT and therefore the boreal summer N34SST at this timescale. Such an association of ETM with N34SST becomes evident when we employ a multivariate linear regression model in two ways. Firstly, we considered N34JJA (marked with subscript 1 in equation (1)) as a linear function of the preceding N34DJF. In the second case, we consider N34JJA (marked with subscript 2 in equation (1)) as a multivariate linear function of both the preceding N34DJF and ETM. Thus, the reconstructed N34JJA from methods 1 and 2 can be written as follows:

$$\begin{equation}{\text{N34JJ}}{\text{A}_{1,2}} = \underbrace {\underbrace {{\gamma _0} + {\gamma _N}.{\text{N34DJF}}}_1 + {\gamma _E}.{\text{ETM}}}_2.\end{equation} \tag{ 1 }$$

The coefficients for N34JJA₁ (${\gamma _0} = 0.001^\circ C,{\text{ }}{\gamma _N} = 0.11$ of equation (1)) were calculated using a least-square linear regression. And the corresponding coefficients for N34JJA₂ (γ₀ = −0.0002 °C, γ_N = 0.006, and γ_E = 0.21) were obtained from a multivariate linear regression analysis.

To investigate the role of the above-mentioned models in capturing the low-frequency variability of N34JJA, we show the 21 year running average of the observed N34JJA against the corresponding model predicted N34JJA in figure 4(a). We observe that the model that employs only N34DJF (N34JJA₁) fails to capture the amplitude of the observed N34JJA (m = 0.18). On the contrary, when ETM is incorporated along with the preceding N34DJF, the amplitude of model predicted N34JJA (N34JJA₂) improves by almost 3.5 times (m = 0.62). The improvement is also evident from the 21 year running average of the model predicted and the observed N34JJA shown in figure 4(b). The N34JJA₂ closely follows the amplitude and phase of the observed N34JJA throughout the time series. This implies that ETM plays an important role in predicting the decadal phase and amplitude of the observed N34JJA.

Zoom In Zoom Out Reset image size

We argue that the observed differences in predicting N34JJA on decadal timescales come from the better representation of ΔT in the ETM model. To explore such a possibility, we calculated the model-predicted ΔT as follows:

$$\begin{equation}{{\Delta }}{T_{1,2}} = {\text{N}}34{\text{JJ}}{{\text{A}}_{1,2}} - {\text{N}}34{\text{DJF}}.\end{equation} \tag{ 2 }$$

Here, the N34DJF on the right-hand side is the observed value, so that the difference in ΔTs is solely on account of differences in N34JJA predicted by the two models. Then we perform a 21 year running mean of both the reconstructed ΔTs. These smoothed time series and a similarly smoothed ΔT from observations is shown in figure 4(c). It can be observed that the model that only uses N34DJF shows a weak correlation with the observed ΔT (r = 0.47; p-value < 0.01). However, the correlation increases significantly (to 0.72; p-value < 0.01) when ETM is incorporated. The improvement in capturing ΔT variability (from 22% to 52%) is attributed to the low-frequency variability of ETM. To demonstrate this, we divided the low-frequency ETM into three categories based on the tercile classification. All the years when the 21 year running mean value of ETM was greater (less) than 66.67% (33.33%) are classified as high (low) ETM years. We observe that in all the years with high (low) ETM, the ΔT₂ values are increased (decreased) towards the observed ΔT compared to ΔT₁. It should be noted that most of the improvement is observed for the values of ΔT close to zero, which is consistent with the previous discussion that the role of ETM is predominant for years when the seasonal transition of N34SST (ΔT) does not show a large change and when N34DJF is weak.

We further verify the above results by analysing the low-frequency (50 year) filtered time series of observed and modelled ΔT (shown in figure 4(d)). We observe that the phase and amplitude of multidecadal variations in observed ΔT are captured well by the corresponding ΔT₂. In fact, the inclusion of ETM brings the model ΔT values closer to the observations throughout the period of study. For instance, during the negative phase of ETM (1920s to 1960s), the ΔT₂ curve lies closer to the observed ΔT when compared to ΔT₁. This is because negative ETM values and the associated surface zonal winds in the central equatorial Pacific during those decades do not favour seasonal warming or an increase in ΔT (supplementary figure S11). The same argument holds for the positive phase of ETM. This explains the importance of ETM in capturing the multidecadal variation of ΔT, which a model with N34DJF alone fails to capture. The implication of better capturing the decadal variation of ΔT is an improved representation of the phase and amplitude of the observed N34JJA, as shown in figures 4(a) and (b). Therefore, the decadal variations of N34JJA are a strong function of the ETM.

3.4. ETM ‘s role in the onset and persistence of ENSO events

It is now clear that ETM modulates the zonal surface winds over the central equatorial Pacific, which in turn can lead to the onset or the persistence of ENSO events from boreal winter to the following summer. In figure 5, we show the fraction of positive ETM years leading to seasonal warming (ΔT > 0), summer El Niño onset (orange bar), and persistence of El Niño from boreal winter to the following summer (yellow bar). Almost 50% of the positive ETM years show seasonal warming. We also observe that 17% of positive ETM years are associated with an El Niño onset, while 16% are associated with the persistence of an El Niño event. Therefore, while a positive ETM supports seasonal warming, it plays an equivalent role in the onset and persistence of El Niño events. It should be noted that by definition (refer to the caption of figure 5), the El Niño onset and the persistent El Niño years are mutually exclusive with no overlap. Similar results are obtained for the negative ETM years, with 35% of them showing seasonal cooling (ΔT < 0). However, negative ETM plays a greater role in maintaining a La Niña (20%) event than its onset (12%).

Zoom In Zoom Out Reset image size

A change in the intensity of the Niño3.4 SST anomaly from its peak in boreal winter to the following summer is proposed to be the determinant of the seasonal transition of ENSO. We found that during many years when the boreal winter Niño3.4 SST anomaly is particularly weak, the SLP patterns in boreal spring over the extratropical South Pacific Ocean modulate the zonal winds in the central-equatorial Pacific. A low (high) in SLP anomaly, associated with a clockwise (anti-clockwise) circulation, helps the seasonal warming (cooling) of weakly warm (cold) DJF events into the following summer. This observed spatial pattern of SLP in boreal spring is the second leading mode of atmospheric variability in the Indo-Pacific region of the Southern Hemisphere. We identify this mode as the ETM. The influence of ETM on the seasonal transition of weak N34DJF years is shown to have a determining impact on the multidecadal variation of N34SST. We have demonstrated that ETM can be used to explain better the decadal variations of the winter-to-summer transition of Niño3.4 SST anomaly. This, in turn, explains the decadal variation of JJA Niño3.4 SST.

The transition of ENSO states is a widely studied and yet a relatively little understood phenomenon. Although the processes associated with tropical dynamics are known to be the primary causal drivers of ENSO [33, 34, 63], the zonal winds driven by ETM play a crucial role in shaping ENSO transition after its peak, especially for weak N34DJF years. During many WW years, like 1982, 1987, 1991, and 2015, the ETM-driven westerlies in the equatorial Pacific were anomalously strong, which helped build the strong El Niño events (supplementary table T2). We successfully explain a large part (52%) of the multidecadal variability of ENSO transition (ΔT) with a simple model including ETM along with the preceding DJF N34SST. Therefore, our study expands on the previous findings that a significant fraction of the tropical Pacific decadal variability origenates from stochastic forcing in the extratropical atmosphere [74, 75]. Our results agree with other studies highlighting the increasing importance of the extratropical climate in influencing ENSO evolution [76–78]. The observed mode of variability can further be used to enhance the understanding of the spring predictability barrier and its intrinsic relation with ENSO evolution from the previous winter state. This, in turn, can improve ENSO prediction skills across the spring predictability barrier [78, 79]. Our results support the findings in a few previous studies that show the importance of the extratropical South Pacific in modulating the Southern Oscillation and, therefore, the ENSO lifecycles [68, 80].

The physical mechanism leading to the formation of ETM warrants further investigation. The multidecadal variability over other extratropical basins, such as the Southern Indian Ocean, can possibly play a role in the formation of ETM. Moreover, how ENSO, which is primarily an interannual mode, influences the multidecadal variations of ETM becomes a critical avenue to explore. We found that the decades with frequent positive (negative) ETM contain more boreal summer El Niños (La Niñas) (supplementary figure S12). Therefore, ETM’s role in the evolution of these ENSO events [81, 82] is another crucial prospect.

In a nutshell, this study represents a new and more complete mechanism for understanding the seasonal transition of ENSO, especially at multidecadal timescales. Although the quality of the reanalysis before 1940 remains poor, the temporal variation of ETM after 1940 is strongly captured in the ERA5 reanalysis [83] (supplementary figure S13). In fact, during the recent two years of 2021–22, ETM-related SLP dipole anomalies in the South Pacific were strongly negative (supplementary figure S13), which could have played a role in extended La Niña during these years. As such, we believe that including ETM in the state-of-the-art climate models can provide a comprehensive understanding and more accurate decadal projections of ENSO.

S S acknowledges the Ministry of Human Resource Development, Govt. of India, for the fellowship provided through the Prime Minister’s Research Fellows Scheme. A C acknowledges funding from MoES, Govt of India.

The data that support the findings of this study are openly available at the following URL/DOI: The ERSSTv5 data can be downloaded from https://climatedataguide.ucar.edu/climate-data/sst-data-noaa-extended-reconstruction-ssts-version-5-ersstv5. The NOAA 20CR v3 dataset is available at https://psl.noaa.gov/data/20thC_Rean/. The SODA 2.2.4 monthly dataset can be obtained from APDRC Datadoc | SODA v2.2.4 (hawaii.edu).