A Control of ENSO Transition Complexity by Tropical Pacific Mean SSTs Through Tropical‐Subtropical Interaction

El Niño–Southern Oscillation (ENSO) transitions from one event to another in complex ways. Using observational analyses and forced atmospheric model experiments, we show that a preceding ENSO event can activate a subtropical Pacific forcing mechanism to trigger another ENSO event during the following year. These tropical‐subtropical Pacific interactions result in a cyclic ENSO transition if the two ENSO events are of opposite signs or a multiyear ENSO transition if they are of the same sign. The preceding ENSO event should excite deep convections in the tropical Pacific in order to activate the subtropical Pacific mechanism. This requirement enables mean temperatures in the cold tongue and warm pool to respectively control how easily the cyclic and multiyear transitions can occur. A future warmer tropical Pacific is projected to decrease the frequency of occurrence of multiyear ENSO transitions but increase the occurrence of cyclic ENSO transitions.


Introduction
El Niño-Southern Oscillation (ENSO) is the strongest year-to-year phenomenon in our global climate, causing regional climate extremes and massive ecosystem impacts (Coelho & Goddard, 2009;Donnelly & Woodruff, 2007;McPhaden et al., 2006;Power et al., 2013). ENSO has recently observed to exhibit a wide range of complex behaviors (Capotondi et al., 2015;Timmermann et al., 2018;Yu et al., 2017) that impede our understanding of its physical dynamics and results in huge uncertainties in projecting its future changes (Amaya, 2019;Cai et al., 2015;Collins et al., 2010;Fedorov et al., 2003;Sohn et al., 2016;Vecchi & Wittenberg, 2010). One aspect of ENSO complexity is related to the ways that ENSO transitions from one event to another (Wang et al., 2019;Yu & Fang, 2018). An El Niño (La Niña) event can be preceded (1) by a La Niña (El Niño) event to become a cyclic ENSO transition, (2) by a neutral event to become an episodic ENSO transition, or (3) by another El Niño (La Niña) event to become a multiyear ENSO transition. Unlike most other studies of ENSO complexity that focus on properties during the peak phase, ENSO transition complexity focuses on ENSO properties during the onset phase.
Whether an ENSO event goes through a cyclic, episodic, or multiyear transition is affected by its onset (or trigger) mechanism (Yu & Fang, 2018)-the process by which the initial ENSO sea surface temperature (SST) anomalies are produced in the equatorial Pacific. Two primary onset mechanisms have been identified by the research community over the past decades (see reviews in Capotondi et al., 2015;Wang et al., 2017;Yang et al., 2018;Yu et al., 2017). The tropical Pacific onset (TP onset) mechanism consists of the ©2020. American Geophysical Union. All Rights Reserved.

10.1029/2020GL087933
Key Points: • An ENSO event can activate a subtropical Pacific mechanism to onset another event resulting in a cyclic or multiyear ENSO transition • Mean SSTs in the cold tongue and warm pool respectively control how easily cyclic and multiyear transitions can occur via this process • A future warming of the tropical Pacific is projected to reduce (increase) the occurrence of the multiyear (cyclic) ENSO transitions Supporting Information: • Supporting Information S1 • Figure S1 • Figure S2 • Figure S3 • Figure S4 • Figure S5 • Figure S6 • Figure S7 • Figure S8 • Figure S9 Correspondence to: recharge oscillator (Jin, 1997;Wyrtki, 1975) and delayed oscillator (Battisti & Hirst, 1989;Zebiak & Cane, 1987) theories. Together, these theories describe how equatorial surface wind anomalies induced by an ENSO event can deepen or shallow the thermocline in the equatorial eastern Pacific over the following months, and onset another ENSO event with an opposite phase. The TP onset mechanism leads mostly to cyclic ENSO transitions and contributes to reduce ENSO transition complexity (Yu & Fang, 2018), although it should be noted that its asymmetric responses between El Niño and La Niña may allow for some complexity (Hu et al., 2017). In contrast, the subtropical Pacific onset (SP onset) mechanism describes how subtropical Pacific SST anomalies can intrude into the equatorial Pacific to trigger ENSO events (Alexander et al., 2010;Anderson & Perez, 2015;Kao & Yu, 2009;Vimont et al., 2003;Yu et al., 2010). This onset mechanism has been shown to produce all three transition patterns (Yu & Fang, 2018; Figure S1 in the supporting information). Therefore, the SP onset mechanism is a key contributor to ENSO transition complexity and needs to be better understood for projecting future changes in ENSO complexity.
To trigger an ENSO event, the SP onset mechanism is initiated by a decrease or increase in trade winds that extends from off the coast of Baja California to the equatorial central Pacific that respectively warms or cools SSTs underneath. The wind and SST anomalies can prolong for several months through a wind-evaporation-SST feedback (WES; Xie & Philander, 1994) (aka the Pacific meridional mode (PMM) in Chiang and Vimont, 2004, or seasonal footprinting mechanism in Vimont et al., 2003). The initial trade wind variations are often induced by subtropical atmospheric processes (such as the North Pacific oscillation, NPO) that are non-ENSO related (Vimont et al., 2003). Therefore, it is easy to understand why the SP onset mechanism can give rise to episodic ENSO transitions. Less attention has been paid in past studies to the possibility that the SP onset mechanism can also be activated by a preceding ENSO event resulting in either cyclic or multiyear ENSO transitions.
The limited studies on the ENSO-induced SP onset mechanism have found that extreme El Niño events can activate the SP onset mechanism to trigger La Niña events during the following year and result in a cyclic El Niño-to-La Niña transition (Yu & Fang, 2018;Yu & Kim, 2011). Interestingly, the opposite tendency of a cyclic La Niña-to-El Niño transition via the SP onset mechanism seems to occur less frequently (Yu & Fang, 2018). The SP onset mechanism appears to respond asymmetrically to El Niño and La Niña. What is the cause for this asymmetry and how will this asymmetry change in the future as the earth warms? Answers to these questions are important to project future changes in ENSO transition complexity and are the main focus of this study.

Data Sets and Methods
Monthly mean values of SST, surface wind, geopotential height, precipitation, and sea surface height (SSH) during the period of 1958-2014 were used after they were regrided to a common 1.5°× 1°longitude-latitude grid covering the tropical Pacific (20°S to 20°N, 122°E to 70°W). The SST data are the Hadley Center Sea Ice and Sea Surface Temperature data set (Rayner et al., 2003), the surface wind, geopotential height, and precipitation fields are from the National Centers for Environmental Prediction/National Center for Atmospheric Research (Kalnay et al., 1996), and the SSH data are produced by the German contribution of the Estimating the Circulation & Climate of the Ocean (Köhl, 2015). Anomalies are defined as the deviations from the seasonal cycle averaged over the analysis period after removing the linear trend. The same procedures were applied to the simulations (the last 100 years of the preindustrial simulations and the full periods of the historical, RCP4.5 and RCP8.5 simulations) produced by 28 models of the Coupled Model Intercomparison Project (CMIP5) (Taylor et al., 2012; see Extended Data Table S1). Similar results are obtained when a quadratic trend removal is applied (in place of the linear trend removal).
Several indices were used in the analyses. Indices of the SP and TP onset mechanisms were constructed by applying a multivariate empirical orthogonal function (MEOF) analysis to combined SST, wind, and SSH anomalies in the tropical Pacific (Text S1; Xue et al., 2000;Yu & Fang, 2018). These indices are referred to as the SP onset index and the TP onset index. A Niño3.4 index is used to represent the intensity of ENSO events and is defined as the SST anomalies averaged between 5°S to 5°N and 170-120°W. The equatorial eastern Pacific (EEP) region is defined as the area between 5°S to 5°N and 120-90°W, whereas the equatorial central Pacific (ECP) region is defined as the area between 5°S to 5°N and 160°E to 170°W. A series of forced experiments were conducted in this study with an atmospheric general circulation model (AGCM). The designs of these model experiments are described in the supporting information (Text S2).

Results
We first examine the anomaly pattern associated with the SP onset mechanism by regressing SST and surface wind anomalies onto the SP onset index (Figure 1a). A key feature in the pattern is the overlap of trade wind anomalies with SST anomalies in the northeastern subtropical Pacific. This overlap reflects the subtropical coupled nature of the SP onset mechanism, which is a key to extending the SST anomalies southwestward into the equatorial central Pacific for the ENSO onset. A negative phase of the SP onset mechanism (see Figure 1a) can onset a La Niña event, whereas a positive phase of the SP onset mechanism can onset an El Niño event. As mentioned, the NPO used to be considered to be a key source for the trade wind anomalies. However, Figure 1a indicates that subtropical Pacific wind anomalies may also be related to the large SST anomalies in the equatorial eastern Pacific (EEP) and the equatorial central Pacific (ECP).
The pair of anomalous cyclones straddling the equator to the west of the EEP SST anomalies appears to be a Gill-type response (Gill, 1980) to the SST anomalies. The trade wind anomalies over the northeastern subtropical Pacific are associated with the anomalous cyclone to the north. This anomaly structure suggests that the subtropical Pacific trade wind anomalies can be the result of a Gill-type response to the EEP SST anomalies (as illustrated by the red vectors in Figure 1a) (Wang et al., 2000). An El Niño (La Niña) event in the EEP can activate a negative (positive) phase of the SP onset mechanism to trigger a subsequent La Niña (El Niño) event, giving rise to a cyclic ENSO transition. On the other hand, the same trade wind anomalies over the northeastern Pacific can also be linked to the ECP SST anomalies through a wave train pattern connecting the two regions (as illustrated by the blue circles in Figure 1a) (Lyu et al., 2017;Stuecker, 2018). A La Niña (El Niño) event in the ECP can activate a negative (positive) phase of the SP onset mechanism to trigger another La Niña (El Niño) event, giving rise to a multiyear ENSO transition. Therefore, the Gill-type response and the basin-wide wave train enable the SP onset mechanism to produce cyclic ENSO transitions through the EEP region and multiyear ENSO transitions through the ECP region.
In order to excite the Gill-type response and the wave train, ENSO events in the EEP and ECP should induce convective heating anomalies. The threshold value for deep convection is around 28°C (Sud et al., 1999;Zhang, 1993), which is confirmed for both the EEP and ECP regions by examining the relationships between precipitation anomalies and SSTs over the regions (Figures 2a and 2b). A composite of all El Niño months when the EEP SSTs are greater than 28°C (the pink dots in Figure 2a) reveals strong evidence of a Gill-type response, which is characterized by a pair of anomalous cyclones in the lower troposphere (Figures 2c-2g) and a pair of anticyclones in the upper troposphere (Figures 2h-2l). An analysis of the time series of the EEP and SP onset indices confirms that a negative SP onset index occurs after the EEP SST index passes the 28°C threshold ( Figure S2a). The El Niño-induced anomalous trade winds (and the negative SP onset mechanism) bring anomalous cold water over the subtropical Pacific toward the tropical central Pacific and switch the composite SST anomalies from El Niño to La Niña condition (Figures 2c-2g). In contrast, the Gill-type response is either very weak or absent when we repeat the composite for all El Niño months where the EEP SSTs are lower than 28°C (the red dots in Figure 2a) or with all La Niña months (the blue dots in Figure 2a) (see Figure S3).
Our composite results confirm that only El Niño events that are strong enough to raise EEP SSTs higher than 28°C can activate the SP onset mechanism to produce the cyclic ENSO transition. Since the climatological SST in the EEP (25.19°C) is lower than the 28°C threshold (Figure 1b), it is impossible for La Niña events to induce anomalous convective heating and activate this mechanism. This explains why the SP onset mechanism produces more El Niño-to-La Niña than La Niña-to-El Niño transitions (Yu & Fang, 2018). There are only two such very strong El Niños during the analysis period: the 1982-1983and 1997-1998 We notice that the SP onset index only becomes strongly negative after the EEP SSTs exceed the 28°C during these two extreme events ( Figure S2b). In addition, the SP onset indices are stronger than the TP onset indices during these events (lag-0 in Figure S5a). This confirms that only extreme warming in the EEP can enable the SP onset mechanism to dominate the TP onset mechanism and to initiate cooling in the ECP resulting in cyclic transitions (Yu & Kim, 2011). Therefore, the climatological SST in the EEP region determines how easily and what type of cyclic ENSO evolution (i.e., El Niño-to-La Niña or La Niña-to-El Niño) can be produced via the SP onset mechanism. It should be noted that the TP onset mechanism also contributes to the cyclic transitions in these extreme cases.
Over the ECP region, the climatological SST (28.94°C) is just slightly above the 28°C threshold. Therefore, even moderate La Niña events here can shut off deep convections, resulting in large negative convective heating anomalies (Figure 1b). A composite analysis was performed with the La Niña months that lower the local SSTs to below 28°C. The results indicate that a La Niña event that shuts off deep connection in the ECP region can excite wave train responses that extend into the extratropical lower troposphere (Figures 2r-2v). The wave train includes an anomalous anticyclone off the North American coast that can further induce anomalous descent over the northeastern subtropical Pacific via geostrophic processes (Lyu et al., 2017). The subsidence then enhances the trade winds and initiates a negative phase of the SP onset mechanism (Figure 2b; Figure S7 in Yu & Fang, 2018) to onset another La Niña condition (Figures 2m-2q). The negative values of the SP onset index are confirmed during these events, while the TP onset index follows its own cycle leading to positive values ( Figure S5b). Therefore, even moderate La Niña events in the ECP region can result in a multiyear La Niña transition through the SP onset mechanism.
El Niño events in this region can also strengthen deep convection to produce positive convective heating anomalies. However, since most of the central Pacific El Niño (Kao & Yu, 2009;Yu & Kao, 2007) are of weaker SST anomalies (Lee & McPhaden, 2010), the anomalous heating they induce are relatively small. A composite of the El Niño months in the ECP region ( Figures S4p-S4t) indicates that the extratropical wave train pattern excited by warm ECP SST anomalies is relatively weaker than that excited by cold ECP SST anomalies (cf. Figures S4q and S2s). Also, the evolution of the composite SST anomalies does not show multiyear El Niño transitions ( Figures S4k-S4o). A similar result was obtained in the composite analysis of weak La Niña events that do not lower ECP SSTs to below 28°C ( Figures S4a-S4j).
Our composite analyses indicate that La Niña events in the ECP are more effective than El Niño events in giving rise to multiyear ENSO transitions through the SP onset mechanism. This explains why the SP onset mechanism produces more multiyear La Niña transitions than multiyear El Niño transitions (Yu & Fang, 2018). Therefore, the climatological SST in the ECP region determine how easily and what type of multiyear transitions (i.e., multiyear La Niña or multiyear El Niño) can be produced via the SP onset mechanism.
The processes described above were further confirmed by a series of forced AGCM experiments conducted with the Community Earth System Model version 2 (CESM2) (see Text S2). In the experiments, Gaussian shaped SST anomalies were prescribed and added onto the climatological SSTs in either the ECP or EEP region with an amplitude varied from −4°C to +4°C. The EEP experiments confirm that the SST anomalies have to be large enough (+4°C) to raise EEP SSTs above the 28°C threshold to excite deep convection ( Figure S6a), induce the Gill-type responses, and enhance the surface trade winds over the northeastern subtropical Pacific for activating a negative SP onset mechanism ( Figure S6c). No such response was found in the other EEP experiments (Figures S6a and S6b). The ECP experiments confirm that large and negative deep convection anomalies can be induced by moderate cold (−1°C SST) anomalies in the region, which can then excite a wave train response in the extratropical atmosphere and enhance the trade winds to activate a negative SP onset mechanism ( Figure S7). In contrast, the wave train response is weaker in the +1°C SST experiment ( Figure S7c). In addition, the weak wave train in this experiment emanates from a more eastern part of the ECP region (compared to the −1°C experiment) and results in a surface wind response that is away from the trade wind region.
Tropical Pacific mean SSTs are projected to increase in the future (Kirtman et al., 2013;Levitus et al., 2009;Meehl et al., 2006). The multimodel mean (MMM) of the simulations using 28 CMIP5 models (Table S1) show that mean SSTs in the EEP are projected to reach 28°C around 2100 (2060) under the RCP4.5 (RCP8.5) scenario (Figure 3a). This tendency should enable El Niño events of any intensity to induce anomalous deep convection in the EEP, activate the negative phase of the SP onset mechanism, and give rise to cyclic El Niño-to-La Niña transitions. Conversely, as the mean temperatures increase and exceed the threshold, La Niña events in the EEP will be able to shut off deep convection and lead to a subsequent El Niño event via a positive SP onset mechanism. The latter process should give rise to cyclic La Niña-to-El Niño transitions. As a result, cyclic ENSO transitions (i.e., both El Niño-to-La Niña and La Niña-to-El Niño) should occur more frequently via the SP onset mechanism in the future. On the other hand, the projected temperature increases in the ECP region ( Figure 2b) will make it harder for La Niña events to reduce ECP SSTs below 28°C and to activate a negative SP onset mechanism in the region. This should result in fewer multiyear La Niña transitions via the SP onset mechanism in the future. In other words, the warmer tropical Pacific in the future is expected to increase the dominance of the cyclic ENSO transition via the SP onset mechanism and to reduce the occurrence of the multiyear ENSO transition.
It has been suggested that the threshold value for tropical convection may change as the climate warms (Johnson & Xie, 2010), which can affect our argument above. To assess this possible impact, we examine the relationships between ENSO SST anomalies and precipitation anomalies over the EEP and ECP regions in the preindustrial, historical, RCP4.5, and RCP8.5 simulations ( Figure S8). We find that, in the EEP region ( Figures S8a-S8d), the same magnitudes of El Niño SST anomalies induce larger precipitation (i.e., deep convection) anomalies as the region warms (from preindustrial to RCP8.5 simulations). On the other hand, the La Niña events in the ECP region ( Figures S8e-S8h) require larger negative anomalies to turn off the deep convection (i.e., reduce the precipitation to 0) in the RCP8.5 simulation than in the preindustrial simulation. These results indicate that, even if the 28°C threshold may not remain unchanged or other factors may appear to moderate regional deep convection (Back & Bretherton, 2009;Sabin et al., 2013), a warmer mean state of the tropical Pacific enable El Niño events to turn on deep convections more easily over the EEP region but make La Niña events less easily to turn off deep convections over the ECP region.
We further examine if the ENSO transition complexity related to the SP onset mechanism changes from the preindustrial to RCP simulations. Figure 3c shows the MMM of the strongest lead correlation of the Niño3.4 index with the SP onset index. A positive correlation implies a multiyear transition via the SP onset mechanism, since an El Niño (La Niña) event leads to a positive (negative) phase of the SP onset mechanism and should later develop into another El Niño (La Niña) event. Conversely, a negative correlation implies a cyclic transition. A weak negative value is found in the reanalysis product, which is consistent with the fact that both multiyear ENSO and cyclic ENSO events are observed (Yu & Fang, 2018). In the CMIP5 simulations, the MMM values change from weakly positive in the preindustrial and historical simulations to strong negatives in the RCP4.5 and RCP8.5 simulations. This tendency indicates that, as the Pacific Ocean warms, the SP onset mechanism tends to change from a preference for generating multiyear transitions to a preference for cyclic transitions. This future change of ENSO transition complexity is consistent with our projection based on the control of the tropical Pacific mean state on the SP onset mechanism.
Consistent results can be found by examining the transition patterns of SST anomalies composited for large values of the SP onset index (Figure 4). Both the positive and negative phases of the SP onset mechanism produce a transition pattern that changes from being dominated by multiyear ENSO transitions in present-day simulations to being dominated by cyclic ENSO transitions in the RCP simulations. Furthermore, in the preindustrial and historical simulations, the simulated mean ECP SSTs are colder  (a) and (b), the thick lines are the multimodel means and the thin lines are the spreads of the projected SSTs from preindustrial (blue) and historical (orange) to RCP4.5 (green) and RCP8.5 (red) calculated from the 28 models. The black solid lines are based on the reanalysis data. In (c), the tendency is represented by the strongest correlation coefficient when the Niño3.4 index leads the SP onset index by 1 to 6 months. The multimodel means of the preindustrial (blue), historical (orange), RCP4.5 (green), and RCP8.5 (red) simulations are shown, and the reanalysis is in black. Solid bars are the standard deviations of the model tendency.
(around 28°C; see Figure 3b) than the reanalysis SSTs (28.94°C). Therefore, both El Niño and La Niña can enhance or suppress deep convections to give rise to multiyear events. The models with colder ECP SSTs tend to simulate more multiyear El Niños, while the models with warmer ECP SSTs have more multiyear La Niñas as the observations ( Figure S9). This CMIP5 model deficiency in simulating too frequent multiyear El Niños in the present-day climate further supports the notion that mean SSTs in the ECP region control the frequency of occurrence of multiyear ENSO transitions.

Summary and Discussion
We have, in this study, shown that the tropical Pacific mean SSTs control the ENSO transition complexity produced by the SP onset mechanism. Since the SP onset mechanism is a key source of the transition complexity, our finding offers a way to use tropical Pacific mean states to assess future or past ENSO transition complexities. More interestingly, we find the mean SSTs in different parts of the equatorial Pacific matter for different ENSO transition patterns. The mean SSTs in the EEP region affects how frequently cyclic ENSO transitions can occur, whereas the mean SSTs in the ECP affect how frequently multiyear transitions can occur. We find these relationships are useful to link CMIP5 model deficiencies in simulating ENSO transition complexity and tropical Pacific mean state simulations. Finally, the control of the tropical mean state can also explain why the SP onset mechanism responds differently to El Niño and La Niña, which is a key source of El Niño-La Niña asymmetries that has not yet been fully explored nor understood. The values shown are the SST anomalies composited for strong events of the positive (upper panels) and negative (lower panels) phases of the SP onset mechanism, which were identified respectively as the months in which the SP onset index is larger than 0.7 times its standard deviation or lower than −0.7 times its standard deviation. The evolution extends from 12 months before to 12 months after the events and is calculated from the reanalysis (first column) and the multimodel means for the preindustrial (second column), historical (third column), RCP4.5 (fourth column), and RCP8.5 runs (fifth column column). The SSTs are normalized by the standard deviation of their modeled Niño3.4 index before the composites are formed and the color bar for the reanalysis panel is 2 times larger than those of the other panels (shown on the right in°C).