Variability of the western Pacific warm pool structure associated with El Niño

Abstract

Sea surface temperature (SST) structure inside the western Pacific warm pool (WPWP) is usually overlooked because of its distinct homogeneity, but in fact it possesses a clear meridional high–low–high pattern. Here we show that the SST low in the WPWP is significantly intensified in July–October of El Niño years (especially extreme El Niño years) and splits the 28.5 °C-isotherm-defined WPWP (WPWP split for simplification). Composite analysis and heat budget analysis indicate that the enhanced upwelling due to positive wind stress curl anomaly and western propagating upwelling Rossby waves account for the WPWP split. Zonal advection at the eastern edge of split region plays a secondary role in the formation of the WPWP split. Composite analysis and results from a Matsuno–Gill model with an asymmetric cooling forcing imply that the WPWP split seems to give rise to significant anomalous westerly winds and intensify the following El Niño event. Lead-lag correlation shows that the WPWP split slightly leads the Niño 3.4 index.

1 Introduction

The warm pool in the western Pacific Ocean (WPWP) is a prominent feature in the world ocean. Since sea surface temperatures (SSTs) in excess of about 26–28 °C are required for deep convection over tropical oceans (Graham and Barnett 1987; Johnson and Xie 2010), WPWP is usually defined as a warm water body with SSTs warmer than 28–29 °C. In view of its essential role in global climate system, the WPWP is referred to as the heart of the oceanic-atmospheric system of the planet (e.g., Chen et al. 2004). In the past, the WPWP was studied in a number of aspects, such as its size (Yan et al. 1992; Picaut et al. 1996; Cravatte et al. 2009; Kim et al. 2012; Lin et al. 2012), centroid (Yan et al. 1997; Zhou et al. 2004; Chen and Fang 2005; Kim et al. 2012), heat center (Hu and Hu 2012), skewness of SST distribution (e.g., Clement et al. 2005) and so forth (e.g., Lukas et al. 1991; Guan et al. 2013). Though many papers have focused on the characteristic and variability of WPWP since the Tropical Ocean–Global Atmosphere (TOGA)–Coupled Ocean–Atmosphere Response Experiment (COARE) international program (Webster and Lukas 1992), the tropical bias remains in the Coupled Model Intercomparison Project 5 (CMIP5) models (Brown et al. 2013; Hu and Hu, 2016). As suggested in the first workshop of the WPWP took place in Hobart on 4–8th March 2013, the fundamental aspects of the WPWP should be revisited (Brown et al. 2012).

SSTs inside the WPWP are homogeneous relative to other oceans outside the WPWP, but they are also characterized by a meridional SST high–low–high spatial structure in boreal summer-autumn in the Advanced Very High Resolution Radiometer (AVHRR) infrared satellite SST data (Reynolds et al. 2007), as shown in Fig. 1. The SST low is climatologically located in the vicinity of 10°N inside the WPWP. As shown in the present paper, this SST low is dramatically intensified and splits the 28.5 °C-isotherm defined WPWP (we call it a WPWP split event hereafter for simplification) prior to the El Niño events. However, the spatial characteristics and variability of the SST structure in the WPWP are of particular significance in understanding the WPWP dynamics and need further study.

Fig. 1

Climatological summer-autumn (July–September) SST (°C, contour interval is 0.1 °C) averaged over 1982–2011 using the AVHRR SST data set. WPWP SST lows and highs are denoted in capitals (H SST high; L SST low) and the blue box marks the SST low zone inside the WPWP

Previous studies show that small SST anomalies in the far western Pacific Ocean are sufficient to influence the El Niño—Southern oscillation (ENSO) events (e.g., Wang et al. 1999a, b, c). Yan et al. (1992) suggested that the WPWP size is impacted by solar irradiance variabilities, ENSO events, volcanic activities, and global warming. Wang et al. (1999a, b, c) defined a Niño-6 region of 140°–160°E, 8°–16°N and mentioned an nearly out-of-phase relationship between Niño-3 and Niño-6 region. Hu and Hu (2012) and Kim et al. (2012) found that the longitudinal displacement of WPWP is closely correlated with ENSO. Particularly, Hu and Hu (2012) suggested that the longitude of WPWP heat center leads ENSO about 3–4 months. But the relationship between the spatial structure of warm pool SST and ENSO are not yet fully understood. As one of classical theories of ENSO dynamics, the western Pacific oscillator theory is proposed and suggests that the western Pacific Ocean is a causal region for the ENSO cycle (e.g., Weisberg and Wang 1997; Wang et al. 1999a, b, c). In detail, in the western Pacific oscillator mechanism, cold SST anomalies that are symmetrical about the equator in the off-equatorial western Pacific cause easterly wind anomalies over the far western Pacific Ocean, produce equatorial upwelling Kelvin waves that propagating eastward and thus provide a negative feedback (e.g., Weisberg and Wang 1997; Wang et al. 1999a, b, c). Equatorial Kelvin waves forced by western boundary and wind, especially the intraseasonal westerly, are suggested to significantly contribute to the strength and onset of El Niño events (e.g., McPhaden 1999; McPhaden and Yu 1999). The role of western Pacific Ocean in the ENSO cycle has been described in many documented studies (e.g., Wang 1992; Wang et al. 1999a, b, 2000). Warming and cooling events in the western North Pacific associated with local wind stress curl are suggested to play a critical role in the ENSO phase transition (Wang et al. 1999b). Sea surface wind anomalies in the western North Pacific Ocean associated with SST anomalies play an important part in linking the central Pacific and eastern Asia during the extreme phases of ENSO cycles (Wang et al. 2000). Here we revisit the SST anomalies in the western Pacific Ocean from the perspective of WPWP structure on the basis of advanced observations and further explore the relationship between the WPWP split and El Niño events, with emphasis on the effect of an asymmetrical cooling by the WPWP split on the development of El Niño events.

The rest of this paper is arranged as below. Section 2 gives an overview of the data and corresponding processing methods. Spatial features and variabilities of the WPWP SST structure are described in the Sect. 3. Possible mechanisms of the WPWP split based on heat budget analysis are elucidated in the Sect. 4. Implications of the WPWP split pattern are examined in the Sect. 5, and we summarize the major results in the Sect. 6.

2 Datasets

Daily SST data (0.25° × 0.25°, from 1982 to 2011) from the AVHRR infrared satellite SST (Reynolds et al. 2007) used in this paper are provided by the National Oceanic and Atmospheric Administration (NOAA).

We extract daily heat flux data from the third version of global ocean-surface heat flux (1985–2009) developed by the Objectively Analyzed Air-sea Heat Fluxes (OAFlux) project (Zhang et al. 2004; Yu and Weller 2007) at Woods Hole Oceanographic Institution (WHOI) to examine the role of sea surface heat flux in the WPWP split.

Sea surface height and surface absolute geostrophic current data (1993–2011) are produced by the Segment Sol multimissions d’ALTimetrie, d’Orbitographieet de localisation precise/Data Unification and Altimeter Combination System (SSALTO/DUACS) and distributed by Archiving Validation and Interpretation of Satellite Data in Oceanography (AVISO) (http://www.aviso.oceanobs.com/duacs/).

European Centre for Medium-Range Weather Forecasts (ECMWF) Ocean Analysis Reanalysis System 3 (ORA-S3) datasets including monthly temperature, zonal/meridional current and wind stress data are utilized (Balmaseda et al. 2008). The ECMWF ORA-S3 data covers the global ocean spanning 53 years from January 1959 to December 2011, with a horizontal resolution of 1° × 1°. Ekman pumping velocity w _E is calculated based on ECMWF ORA S3 wind stress data applying the follow equation: $w_{E} = {{curl\left( {{\tau \mathord{\left/ {\vphantom {\tau f}} \right. \kern-0pt} f}} \right)} \mathord{\left/ {\vphantom {{curl\left( {{\tau \mathord{\left/ {\vphantom {\tau f}} \right. \kern-0pt} f}} \right)} \rho }} \right. \kern-0pt} \rho }$, where $\tau$ is wind stress, f Coriolis factor, $\rho$ sea water density.

In addition, we also use the Simple Ocean Data Assimilation (SODA) 2.0.2 (Carton and Giese 2008; Schott et al. 2008) to illustrate the vertical thermal structure in the WPWP split zone. The SODA 2.0.2 is based on optimal interpolation with 0.5° × 0.5° × 40-level spatial resolution spans a total of 44 years from January 1958 until December 2001. The above data sets applied in the present study are summarized in Table 1 as below.

Table 1 Summary of data sets

3 Characteristics of SST structure and WPWP split

The WPWP is enclosed by an isotherm usually chosen to be between 28 and 29 °C in the western–central Pacific Ocean. In the present paper, we define the WPWP as the area embraced by 28.5 °C isotherm in the tropical western–central Pacific Ocean (e.g., Cravatte et al. 2009). Seasonal SST climatology over 1982–2011 are calculated using the AVHRR SST data (months are grouped to four seasons: January–March, April–June, July–September and October–December). Figure 1 shows the SST in the July–September season. As expected, the climatological SSTs are relatively homogeneous inside the WPWP. However, the WPWP in the July–September season is dominated by a meridional high-low–high spatial structure: two SST highs near the 16°N/142°E and 0°N/145°E and a SST low in the vicinity of 7°N/140°E (Fig. 1). SSTs in the relative cool core are about 0.5 °C colder than that in the warm cores.

To gain an insight into the variability of this high–low–high SST structure, we have checked the SSTs in the western-central tropical Pacific Ocean every 20-day mean (to exclude synoptic signals) from 1982 until 2011. It shows that the SST low is intensified before the El Niño events and weakened prior to the La Niña events. In July–October of the El Niño years, the WPWP SSTs are cooled and splits the 28.5 °C-isotherm defined WPWP. On this occasion it appears that the WPWP is separated into two parts by the relative SST low.

We plot in Fig. 2 the SSTs along 135°E (near the SST minimum) during 1982–2011. Seasonal variations and long-term trend of the SSTs in Fig. 2 are excluded to focus on the split events. One of the most striking features is the interannual occurring cooling events in the Fig. 2a. These cooling events continued for several months and most of them are before the mature phases of the El Niño events with SST minimum lower than 26.4 °C. We then calculate $\Delta SST$ that is the mean SST over 4.125°N–16.125°N along 135°E minus 28.5 °C. Since the $\Delta SST$ reflects the extent of the cooling and dividing the WPWP, here the $\Delta SST$ is defined as an index describing the intensity of the WPWP split events (Split Index, SI). As shown in Fig. 2b, negative values of SI correspond to the cooling events in Fig. 2a, and we choose the relative strong cooling events (large negative SI values) as the WPWP split events. The significance of the SI is estimated by computing the confidence interval of the split index using t test. The SI of 0.4973 °C is significant at 95% confidence level and significant at 98% confidence level when SI is 0.6210 °C, thus we define the WPWP split years as the years of SI less that −0.5 °C. As a result, years of 1982, 1986, 1990/1991, 1994, 1997, 2000, 2002, 2004, 2006 and 2009 are recognized as the WPWP split years during 1982–2011. As shown in Fig. 2, most the split events are significant at 98% confidence level. Among these events, 1982 and 1997 are much stronger than others, when the two strongest El Niño events in the history took place. On the contrary, the WPWP enhances during the periods with SI greater than 0.5 °C. The WPWP enhancing events during 1982–2011 include 1984, 1988/1989, 1992, 1994/1995, 1998, 2001, 2003 and 2010. Obviously, most of the WPWP enhancing events are found in La Niña periods.

Fig. 2

(Upper) Time-latitude plot of SST along 135°E. (Bottom) SI defined as SSTs averaged over 4.125°N–16.125°N along 135°E minus 28.5 °C. Seasonal variations and long-term trend are removed. The black lines show the confidence levels of the SI using t test. WPWP split events and enhancements are highlighted in blue and red color, respectively

In Fig. 3, we examine the year 1997 to show the detailed evolution of a WPWP split case during the early phase (July to October) of the 1997/1998 El Niño event (other split events are also investigated and possess similar spatial pattern, but are not shown here). We find, from Fig. 3b–e, an area of low SSTs less than 28.5 °C approximately within the rectangular box of 4–16°N, 120–180°E. The split region is mainly located in between the North Equatorial Countercurrent (NECC) and North Equatorial Current (NEC) centered at about 7°N. This event initiated in later July (Fig. 3b), matured from the beginning of August to early September (Fig. 3c, d), and terminated gradually between later September and October (Fig. 3e–g). Thus, the whole WPWP split event persisted for about two months in 1997.

Fig. 3

SST/WPWP evolution from June 29 until December 5 1997 (20-day mean) demonstrating a WPWP split event in 1997. WPWP is surrounded by 28.5 °C isotherm. Rectangular boxes (4°–16°N, 120°–180°E) in b–e approximatively denote the WPWP split region

The minimum SSTs in the WPWP region during the onset phase (July–October) of El Niño years between 1982 and 2011 are of much interest to us. We first smooth the original SST with fine resolution (0.25° × 0.25°) to exclude high frequency and small scale processes, gain a 20-day-mean SST field with a coarse resolution of 1° × 1°, and then find the SST minimum in the split region (Table 2). The minimum SSTs in these events are in the range of 27.1–28.3 °C and show a general increase, which might be associated with the significant warming up of the WPWP in the last few decades (e.g., Cravatte et al. 2009).

Table 2 Split minimum SST in 1° × 1° corresponding to onset phase (i.e., July–October) of El Niño

To examine a common feature of the WPWP split events, composite SSTs during the split periods (between July 20 and October 17 in the split years including 1982, 1986, 1987, 1990, 1991, 1994, 1997, 2002, 2004, 2006 and 2009) are further calculated and compared with the climatological SST averaged over normal years without split events during 1982–2011 (Fig. 4). Though the 28/28.5 °C isotherms are closed, a lower SST area separating the WPWP is obvious and the high-low–high SST structure is very clear during the composite split event (Fig. 4a). But the SST pattern in normal years is quite different. As shown in Fig. 4b, the SST low is very weak and the WPWP appears as a homogenous warm water mass. Thus, the climatological high–low–high SST structure (Fig. 1) almost comes from the WPWP split events. Difference between the composite SSTs in WPWP split and non-split years are further calculated and tested with student-t method. Shaded area in Fig. 4c indicates SST anomalies at 99% confidence level. It shows that the WPWP during split events are about 0.3–0.6 °C colder than that in the non-split years and most of the SSTs in the split region are at 99% confidence level, implying that the WPWP split is statistical significant.

Fig. 4

Comparison of composited SST (averaged over July 20–October 17) in a split years and b non-split years over 1982–2011. c SST composited in split years minus that in non-split years. Shaded area indicates anomalies at 99% confidence level

4 Mechanisms for WPWP Split

In this section, we examine possible mechanisms in producing the WPWP split events. To do this, we first examine the related physical processes applying composite analysis, and then perform a heat budget analysis in the split region to investigate the quantitative contributions by different physical processes.

4.1 Composite analysis

The equation governing the heat of a water mass near the ocean surface is (Stevenson and Niiler 1983):

$$ \rho C_{p} \left( {\frac{\partial T}{\partial t} + {\mathbf{v}} \cdot \nabla T + w\frac{\partial T}{\partial z}} \right) + \frac{\partial q}{\partial z} = 0 $$

(1)

where T is sea surface temperature, t time, $C_{p}$ specific heat capacity, ${\mathbf{v}}$ horizontal velocity with zonal component u and meridional component v, w vertical velocity, q vertical heat flux (positive for downward), ${\mathbf{v}} \cdot \nabla T = u\frac{\partial T}{\partial x} + v\frac{\partial T}{\partial y}$ the horizontal advection, x, y, and z are the zonal, meridional and vertical components respectively. Hence, the SST change is determined by processes including horizontal advection, entrainment term (that is vertical advection for a fixed depth layer) and sea surface heat flux (sum of radiative and diffusive heat fluxes). To examine the contribution of these terms to the WPWP split, the surface geostrophic current anomaly (relative to 1993–2011 climatology), sea surface height anomaly (SSHA relative to 1993–2011 climatology) and sea surface net heat flux anomaly (relative to 1985–2009 climatology) in the north western Pacific Ocean are composited over the split periods.

As shown in Fig. 5a, in the duration of WPWP split, western Pacific currents including the NEC, NECC and Mindanao Current get reinforced, the anomalous cyclonic circulation features a surface current anomaly field, and the tropical gyre between the NEC and NECC are hence strengthened. This is in agreement with previous studies which suggest an enhancement of the NEC/NECC during the early phase of El Niño events (e.g., Kim et al. 2004; Hsin and Qiu 2012; Hu and Hu 2014; Hu et al. 2015). Negative SSHAs above 99% confidence level lie in the split region in accordance with the current anomaly (Fig. 5b). This suggests that anomalous divergence and upwelling are formed in the region between the NEC and NECC and might contribute to the generating of the WPWP split.

Fig. 5

a Surface geostrophic current anomaly (arrow) and its zonal component (color) composited over WPWP split periods during 1993–2011. b As in (a), but for sea surface height anomaly (cm, contours). Anomalies are relative to the climatology averaged over 1993–2011. Shaded area in the bottom panel indicates anomalies above 99% confidence level

To examine upwelling anomaly in the split region, we use the Ekman pumping velocity derived from ECMWF ORA S3 wind stress to show the w _E composited over the WPWP split events in between 1982 and 2009. It shows that w _E is enhanced by about 0.2–1.8 × 10⁻⁵ m s⁻¹ in the WPWP split region as presented in Fig. 6a. Note that the spatial structure in Fig. 6a is similar to the composite SST difference in Fig. 4c, suggesting the possible importance of anomalous upwelling in the formation of split events. The enhancement of w _E is a result of the wind stress curl anomaly (Fig. 6b). Composite wind stress curl in the split region exhibits positive anomaly above the regions of positive w _E anomaly and thus is in reasonable agreement with the latter.

Fig. 6

Upper and middle panels: Composited Ekman pumping velocity (a, w _E ) anomaly and wind stress curl anomaly (b) relative to 1982–2011 ECMWF ORA-S3 climatology) over WPWP split periods (July until October) in between 1982 and 2011. Bottom three panels: comparison between split index (c), longitude-time plot of sea level anomaly (d) and wind stress curl anomaly (e) averaged over 6°N–16°N

As suggested by documented studies, upwelling Rossby waves forced by the trade wind collapse is also suggested to be one of the major reasons that account for the thermocline shoals off the equatorial in the western Pacific (e.g., McPhaden and Yu 1999; Hu et al. 2016; Hu and Sprintall 2016). The Hovmoller diagrams of sea level and wind stress curl anomalies averaged over the split latitudes 6°N–16°N are contoured and compared with the split index. As shown in Fig. 6c–e, each WPWP split event corresponds to negative sea level anomaly in the western Pacific Ocean. Rossby waves originated in central to eastern Pacific Ocean are very clear, propagate westward and contribute to the sea level decrease (Fig. 6d). However, the propagation of these waves is either shut around the central Pacific Ocean (e.g., 2004 and 2009), or very weak (say 1994, 1997) relative to the sea level anomalies in the WPWP region marked by a black box in Fig. 6d, implying that the Rossby wave process is the background but not the major driver of upwelling, and that local process dominates the fluctuation of sea level in the split region. In contrast, variation of wind stress curl anomalies in the split region is in agreement with that of sea level anomalies (Fig. 6e). Positive/negative wind stress curl anomalies correspond to negative/positive sea level anomalies one by one, suggesting that the Ekman pumping is a primary process controlling the fluctuation of sea level in the split region.

The vertical thermal anomalies of the WPWP during the split event can be illustrated by a case study. To avoid data-dependent problems, we use two relative independent datasets. Figures 7 and 8 show the vertical sections of temperature along 140°E and 8°N in June, August and October 1982 from SODA and ECMWF ORA-S3 datasets. Result in SODA (Fig. 7) is slightly different from that in the ECMWF ORA-S3 (Fig. 8). Because the tropical gyre in the north-western Pacific is cyclonic, the upwelling seems to be a permanent feature in the underlying isotherms below the WPWP bottom and near the mean position of the NECC ridge. But during the split period (e.g., August to October 1982), the upwelling in both the datasets are enhanced. A striking feature is that the WPWP bottom (28.5 °C isotherm) outcrops to the sea surface, and as well as the WPWP split apparently (Figs. 7b, e, 8b, e). Therefore, vertical advections probably account for the split of the WPWP.

Fig. 7

Vertical sections of temperature (contours) and temperature anomales (colors) along 140°E (a–c) and 8°N (d–f) in June, August and October 1982 from SODA. The interval of the contours is 1 °C and seasonal variations in the temperature anomalies are removed

Fig. 8

As in Fig. 7 but for ECMWF ORA S3 temperature products

As discussed above, another important term in regulating the SST change is the sea surface heat flux. We composite the sea surface net heat flux anomaly from OAFlux data set over the split period. It shows that, during the split period, the net heat flux is increased in most of the split region and warms the ocean (Fig. 9). In the region east of the Mindanao Island, net heat flux is slightly decreased with magnitude less than 5 W m⁻² and less than 10% relative to the climatology value. Thus the sea surface heat flux might make very limited contribution to the WPWP split.

Fig. 9

Composited sea surface net heat flux anomaly (relative to 1985–2009 OAFlux climatology) over WPWP split periods (July 20 until October 17) in between 1985 and 2009. Positive for downward

4.2 Heat budget analysis

To further quantitatively evaluate the contribution of various related physical processes to the WPWP split, we perform heat budget analysis in the mixed layer of a box (hereafter SBOX) enclosed by 125.1563°E, 170.1563°E, 15.3°N sections and the equator using ECMWF ORA-S3 temperature and current datasets. Since the mean mixed layer depth (MLD) in this region is about 75 m, we estimate the heat budget terms in the upper 75-m layer. Following Zhang and McPhaden (2010), we derive the integration of the temperature equation and gain the heat balance in the mixed layer of the split region as below:

$$ \begin{aligned} Q_{t} & = Q_{advB} + Q_{advW} + Q_{advE} + Q_{advS} + Q_{advN} + Q_{surf} + Q_{difB} + Q_{difW} + Q_{difE} + Q_{difS} + Q_{difN} \\ & = Q_{advB} + Q_{advW} + Q_{advE} + Q_{advS} + Q_{advN} + Q_{surf} + Q_{res} \\ \end{aligned} $$

(2)

where Q _t is temperature tendency term, Q _advB , Q _advW , Q _advE , Q _advS and Q _advN (Q _diffB , Q _diffW , Q _diffE , Q _diffS and Q _diffN ) are heat advection (diffusion) across the bottom, western (125.2°E), eastern (170.2°E), southern (0°N) and northern (15.3°N) faces of the SBOX, Q _surf the net surface heat flux absorbed by the mixed layer. Note that the short wave radiation penetrating the mixed layer bottom is neglected. Because we are limited to calculate the diffusion terms using ECMWF ORA S3 data set, all the diffusion terms are included in the residual term Q _res , which is estimated as (Cronin and McPhaden 1997)

$$ Q_{res} = Q_{t} - (Q_{advB} + Q_{advW} + Q_{advE} + Q_{advS} + Q_{advN} + Q_{surf} ). $$

(3)

As suggested by Zhang and McPhaden (2010), the horizontal heat transport is calculated using the scheme as below:

$$ \rho C_{p} \int\limits_{A} {({\mathbf{V}} \cdot {\mathbf{n}})(T - T_{ave} )dS} , $$

(4)

where T _ave is temperature averaged over the SBOX. For the heat advection across the bottom of the mixed layer, it can be expressed as:

$$ \rho C_{p} \iint_{B} {w_{75m} \left( {T_{75m} - T_{ave} } \right)dxdy}, $$

(5)

where w _75m (T _75m ) is vertical velocity (temperature) at the depth of 75 m (bottom of the mixed layer).

All the heat budget terms low-pass filtered by a 13-month running mean filter are presented in Fig. 10 and compared with the anomaly of temperature tendency term Q _t . It shows that Q _t anomaly is in between about −23 and 18 W m⁻², while all the horizontal advection anomalies vary in a range of −6 to 7 W m⁻². Among the heat budget terms, the vertical advection at the bottom, Q _advB , sea surface net heat flux term Q _surf and zonal advection at the eastern edge of the split region Q _advE are the major components that account for the variation of Q _t . The heat variation induced by sea surface net heat flux anomaly varies from −8 to 12 W m⁻². But in most cases, Q _surf is positive, heats the ocean, and shows no significant relationship with Q _t .

Fig. 10

a Interannual anomalies of tendency term Q _t , Q _advW , Q _advS , Q _advN , Q _surf , Q _advB and Q _advN . b Comparison between the residual term Q _res , Q _t , Q _advB and Q _advN . c As in (b) but for Q _t and bottom advection plus zonal advection term (Q _advB  + Q _advE ). All the budget terms are 13-month low-pass filtered

Zonal advection at the eastern section Q _advE is the strongest term of the horizontal advection terms. Focusing on the WPWP split periods, Q _advE is negative, implying that an aggregate contribution of the enhancement of the NEC and NECC is warming the split region. But the amplitude of Q _advE is much smaller than those of Q _t and Q _advB .

The residual term Q _res is compared with Q _advB and Q _advE in Fig. 10b. We find that Q _res varies approximately between −5 and 5 W m⁻² with a standard deviation of about 3 W m⁻². As we discussed above, Q _res represents turbulent diffusion processes and other unknown physical processes that are not captured by the model or errors in the model. Though the Q _res is relative large, it is very small or positive during WPWP split period except 2002 (Fig. 10b).

We also calculate the sum of Q _advB and Q _advE and compare it with Q _t . As shown in Fig. 10d, the temperature tendency Q _t is well balanced by the vertical advection at the bottom and the zonal advection at the east edge of SBOX. Correlation coefficient between Q _t and Q _advB (Q _advE ) is about 0.76 (0.74), which is significant at 99.9% (99.9%) confidence level. During the split events, Q _advB and Q _advE are significant negative and drive the decrease of Q _t and SST cooling. Therefore, we conclude that the enhancement of the vertical advection and zonal advection near the eastern edge of WPWP account for the formation of the WPWP split events.

5 Potential impacts on El Niño

The occurring of the WPWP split redistributes the tropical SSTs and hence probably influences the following El Niño event. Here we use the Matsuno–Gill model (Matsuno 1966; Gill 1980) to elucidate the response of the wind field to the WPWP split, then discuss the influences of these split events on the El Niño events.

5.1 Influence on surface wind

For a steady forcing (heating) with Rayleigh friction and Newtonian cooling, the shallow-water equations with beta-plane approximation and non-dimensional form have been discussed by Matsuno (1966) and Gill (1980). We suppose that the beta-plane approximation is effective around 7°N where the WPWP split zone is located. We also use a length scale s of 10° of latitude to gain a non-dimensional form (Gill 1980). As derived by Gill (1980), the shallow-water equations have the following form:

$$ \varepsilon u - \frac{1}{2}yv = - \frac{\partial p}{\partial x}, $$

(6)

$$ \varepsilon v + \frac{1}{2}yu = - \frac{\partial p}{\partial y}, $$

(7)

$$ \varepsilon p + \frac{\partial u}{\partial x} + \frac{\partial v}{\partial y} = - Q, $$

(8)

$$ w = \varepsilon p + Q, $$

(9)

where (x, y) are non-dimensional zonal and meridional distances, (u, v) non-dimensional variables being proportional to zonal and meridional wind velocities, epsilon is proportional to friction, p pressure perturbation, w vertical velocity, and Q heating (negative for cooling).

SSTs in the WPWP split region are negative anomalous and play a cooling role in regulating the tropical atmosphere. But it should be noted that the cooling in the western Pacific Ocean during the split events is asymmetrical about the equator and different from that used by Gill (1980). We construct a form of Q fitting the WPWP split following the form constructed by Gill (1980):

$$ Q(x,y) = - A \cdot g(x) \cdot { \exp }\left( { - \frac{1}{4} \cdot (y + d)^{2} } \right), $$

(10)

$$ g(x) = \left\{ {\begin{array}{*{20}l} {{ \cos }\;kx\quad \left| x \right| < L} \hfill \\ {0\quad \left| x \right| > L} \hfill \\ \end{array} } \right., $$

(11)

$$ k = {\pi \mathord{\left/ {\vphantom {\pi {2L}}} \right. \kern-0pt} {2L}}, $$

(12)

where A is the intensity of the force (cooling term) due to WPWP split, d the meridional distance of the cooling center from the equator, 2L is the zonal length of the cooling. Figure 11 shows the horizontal distribution of the ideal Q.

Fig. 11

Contour of ideal Q which is proportional to heating rate (positive indicates heating) and represents the WPWP split (i.e., SST cooling) in the Matsuno–Gill model

Obviously, the maximum cooling occurs north of the equator centered at about 7°N, which is in well agreement with the composited WPWP split shown in Fig. 4c. Following the study by Xing et al. (2014), we can gain the analytical solution for the simple model forced by the asymmetric cooling Q (For details, please see the “Appendix”).

Figure 12a depicts the (u, v) field of the solution corresponding to the ideal cooling in Fig. 11. Obviously, the asymmetrical cooling induces anomalous eastward winds over the tropical Pacific Ocean east of the cooling center (i.e., about 135°E) but anomalous westward winds above the region west of the cooling center. The influence of the WPWP Split on the wind field can be exposed by a comparison between the observed wind stress change during the WPWP split event and the Matsuno–Gill model wind. Figure 12b shows the composited wind stress anomaly over October-December of all the split years during 1982–2011. As shown in Fig. 12, the major features of ideal solution and composite wind field in terms of their spatial patterns are quite similar to each other: anomalous eastward wind (or westerly) is significant above the tropical Pacific Ocean, while anomalous westward winds exist over the Philippines. We also investigated the evolution of wind stress field during a composited WPWP split period (June–November), and we find that from June to November in WPWP split years, the major change in wind stress field is the enhancement of westerly over the equatorial Pacific Ocean (east of New Guinea coast) and eastward wind stress anomaly over the north tropical Pacific Ocean near the western boundary (Figure not shown). The influence of the WPWP split on the wind stress field is also examined by calculating the wind stress difference between October–November composite (after WPWP split) and June–July composite (before WPWP split) and the result also supports the speculation that the WPWP split events generate significant anomalous westerly over the western equatorial Pacific Ocean (Figure not shown). This implies that the wind anomalies induced by WPWP split plays an important role in the formation of anomalous westerly in the development of El Niño events following the WPWP split events.

Fig. 12

a Distribution of (u, v) (nondimensional but proportional to the sea surface wind velocity) as a response to the SST cooling (WPWP split event) shown in Fig. 11 in the framework of the Matsuno–Gill model. b Wind stress anomaly composite over October to December (after split events) in split years during 1982–2011

5.2 Influence on the subsequent El Niño event

Westerly wind plays an important role in the development of El Niño events. During the onset phase of an El Niño event, anomalous westerly wind occurs in the equatorial region and modulates the strength of El Niño event in two aspects. On one hand, the anomalous westerly winds in the equatorial western Pacific Ocean can enhance the eastward ocean currents that transport warm water to the eastern Pacific Ocean and drive the eastward movement of WPWP. Meanwhile, it also generates eastward propagating downwelling equatorial Kelvin waves that deepens the thermocline and warms in the eastern equatorial Pacific Ocean. On the other hand, the anomalous westerly winds in the equatorial central-eastern Pacific Ocean can weaken the equatorial upwelling and cause positive SST anomalies there. For example, the absence of westerly wind events is suggested to be the main reason for the fade of predicted El Niño in 2014 (Menkes et al. 2014). As discussed in the subsection 5.1, the happening of the WPWP split events cools the atmosphere and drives significant anomalous westerly winds in the tropical Pacific Ocean. Thus one can speculate that a WPWP split event might strengthen the subsequent El Niño event.

To examine the role of WPWP split in El Niño events, we group the historical El Niño events into two types: strong and weak El Niño events. Figure 13 illustrates the evolution of the sea surface wind stress anomalies composited over strong El Niño events (1982/1983 and 1997/1998) and weak El Niño events (1986/1987, 1991/1992, 1994/1995, 2002/2003, 2004/2005, 2006/2007 and 2009/2010). Obviously, both the strong and weak El Niño events feature westerly wind stress anomalies in their onset phases (June to November). However, comparison between strong and weak El Niño events indicates that the eastward wind stress anomalies in a strong El Niño event are much stronger than that in a weak El Niño event (Fig. 13). Given that the WPWP split events in 1982 and 1997 are much intense than that in 2002 and 2004, we suggest that the WPWP split generating anomalous westerly winds might be important in intensifying the following El Niño event.

Fig. 13

Evolution of composite wind stress anomalies over (left) strong El Niños (1982/1983 and 1997/1998) and (right) weak El Niños (1986/1987, 1991/1992, 1994/1995, 2002/2003, 2004/2005, 2006/2007 and 2009/2010)

Figure 14 presents the evolution of SST anomalies composited over strong El Niño years (1982/1983 and 1997/1998, also strong WPWP split years) and weak El Niño years (1986/1987, 1991/1992, 1994/1995, 2002/2003, 2004/2005, 2006/2007 and 2009/2010). Before the WPWP split events, i.e. before June, the disparities between the strong and weak El Niño years are relative small (Fig. 14a, f). But from June until November, strong (weak) WPWP split event happens in the onset phase of strong (weak) El Niño events (Fig. 14). As mentioned above, by examining the SST field every 20 days from 1982 until 2011, we find that the WPWP split events in 1982 and 1997 are much stronger than others. This implies that the influence of the 1982 and 1997 WPWP split events on the following El Niño should be also much stronger than other years. For this reason, extreme El Niño events, say 1982/1983 and 1997/1998 El Niño events (Kim and Cai 2013), might be also partly induced by the extreme WPWP split events before these extreme El Niño events. Thus the WPWP split intensity is important in determining El Niño intensity.

Fig. 14

Evolution of composite SST anomalies over (left panels) strong El Niño events (1982/1983 and 1997/1998) and (right panels) weak El Niño events (1986/1987, 1991/1992, 1994/1995, 2002/2003, 2004/2005, 2006/2007 and 2009/2010)

The relationship between ENSO and WPWP split is further investigated by comparing the Niño 3.4 index and SI and examining the lead-lag relationship between them. As we can see in Fig. 15, Niño 3.4 index fluctuates out of phase with SI. This is because WPWP split is happened when SI is significantly negative, as we mentioned. The SI leads the Niño 3.4 index by about 0–1 month with lagged correlation coefficient of about −0.7, suggesting a quick response of the tropical air-sea system to the WPWP split events.

Fig. 15

a Normalized Niño 3.4 index and SI. b Lead-lag correlation coefficient between Niño 3.4 index and SI

6 Summary and discussion

The spatial structure of SSTs inside the WPWP is usually overlooked because of the strong SST homogeneity there. In the present paper, a significant high-low–high SST structure is exposed within the boreal summer-autumn WPWP. It suggests that the SST low is dramatically intensified in July–October of El Niño years and finally split the 28.5 °C-isotherm-defined WPWP, which we call a WPWP split event here. Composite and heat budget analysis among the WPWP split (or SST low) zone indicates that the enhanced upwelling associated with positive wind stress curl anomaly is the major reason for the SST cooling. Horizontal currents including NEC, NECC and Mindanao Current are strengthened during WPWP split event, lead to negative zonal advection and enhanced cooling at the edge of SBOX and play a secondary role in the formation of WPWP split. It seems that sea surface net heat flux has no significant contribution.

Based on the Matsuno–Gill model, the influences of the WPWP split on the wind field and the following El Niño event are addressed. We construct a cooling formula to represent the SST cooling during WPWP split events following Gill’s work (1980). The solution of the atmosphere response to the WPWP split shows significant anomalous westerly winds east of the WPWP split center. Anomalous westerly influences the El Niño by several oceanic processes including eastward oceanic current, equatorial Kelvin waves and anomalous equatorial downwelling.

Therefore, the cycle of the process is: positive wind curl anomaly in the western Pacific Ocean off the equator causes enhancement of upwelling and SST cooling (WPWP split). The cooling in the split region induces anomalous westerly in the equatorial Pacific Ocean, and then regulates the El Niño events by various oceanic processes, and this result is consistent with previous studies by Wang et al. (1999b) and Wang et al. (2000).

It should be noted that off-equatorial SST and SSH anomalies associated with ENSO occur in both hemispheres of the western Pacific, though the SST and SSH anomalies in the northern hemisphere are more intensive than the southern hemisphere (Figure not shown). More complicated response of the air-sea system to the SST cooling in the western Pacific Ocean is expected.

Kim and Cai (2013) suggested that the advection of mean temperature by anomalous eastward zonal current is important in producing extreme El Niño. Our results suggest that the extreme WPWP split events might be also important in generating extreme El Niño events by buildup westerly and preconditions the tropical atmosphere–ocean toward the occurrence of extreme El Niño events.

Issues remain. The existence of split events (or high-low–high SST pattern) suggests that the WPWP is not a persistent single homogenous water body but a variable water mass with various SST structures. The nature of the WPWP should be further examined carefully. Further study using fully coupled models is needed to examine the representation of the WPWP split in coupled models and the effect of WPWP split pattern on the climate system.

Appendix: Atmospheric responses to an asymmetric forcing

To study atmospheric responses to steady-state forcing, the advection term is neglected and the dissipative process is included in the model (Matsuno 1966; Gill 1980):

$$ \varepsilon u - \frac{1}{2}yv = - \frac{\partial p}{\partial x} $$

(13)

$$ \varepsilon v + \frac{1}{2}yu = - \frac{\partial p}{\partial y} $$

(14)

$$ \varepsilon p + \frac{\partial u}{\partial x} + \frac{\partial v}{\partial y} = - Q $$

(15)

$$ w = \varepsilon p + Q $$

(16)

In these equations (x, y) is non-dimensional distance with x eastwards and y measured northwards from the equator, (u, v) is proportional to horizontal velocity and p is proportional to the pressure perturbation. Q is proportional to the heating rate, $u{\raise0.7ex\hbox{${\partial T}$} \!\mathord{\left/ {\vphantom {{\partial T} {\partial x}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\partial x}$}}_{160E} + v{\raise0.7ex\hbox{${\partial T}$} \!\mathord{\left/ {\vphantom {{\partial T} {\partial y}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\partial y}$}}_{12N} - v{\raise0.7ex\hbox{${\partial T}$} \!\mathord{\left/ {\vphantom {{\partial T} {\partial y}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\partial y}$}}_{EQ}$ is the decay factor.

For a given equatorial asymmetric forcing Q, it could expand in the form of parabolic cylinder functions (Xing et al. 2014)

$$ Q\left( {x,y} \right) = Ag\left( x \right)\exp \left( { - \frac{1}{4}\left( {y + d} \right)^{2} } \right), $$

(17)

$$ g\left( x \right) = \left\{ {\begin{array}{*{20}l} {\cos kx\quad \left| x \right| \le L} \hfill \\ {0\quad \left| x \right| > L} \hfill \\ \end{array} ,\quad k = \frac{\pi }{2L}} \right., $$

(18)

$$ Q = Q_{n} \left( {x,y} \right) = AF_{n} \left( x \right)D_{n} \left( y \right), $$

(19)

where

$$ F_{n} \left( x \right) = c_{n} \left( x \right) = \left( { - 1} \right)^{n} \frac{1}{n!}\left( {\frac{d}{2}} \right)^{n} \exp \left( { - \frac{1}{8}d^{2} } \right)g\left( x \right). $$

(20)

Parabolic cylinder functions $u{\raise0.7ex\hbox{${\partial T}$} \!\mathord{\left/ {\vphantom {{\partial T} {\partial x}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\partial x}$}}_{160E} + v{\raise0.7ex\hbox{${\partial T}$} \!\mathord{\left/ {\vphantom {{\partial T} {\partial y}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\partial y}$}}_{12N} - v{\raise0.7ex\hbox{${\partial T}$} \!\mathord{\left/ {\vphantom {{\partial T} {\partial y}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\partial y}$}}_{EQ}$ are given by

$$ \left\{ \begin{aligned} D_{0} \left( y \right) = \exp \left( { - \frac{1}{4}y^{2} } \right) \hfill \\ D_{1} \left( y \right) = y\exp \left( { - \frac{1}{4}y^{2} } \right) \hfill \\ D_{2} \left( y \right) = \left( {y^{2} - 1} \right)\exp \left( { - \frac{1}{4}y^{2} } \right) \hfill \\ D_{3} \left( y \right) = \left( {y^{3} - 3y} \right)\exp \left( { - \frac{1}{4}y^{2} } \right) \hfill \\ \cdots \cdots \hfill \\ \end{aligned} \right.. $$

(21)

To solve the above equations, Gill (1980) first introduced two new variables q and r to replace p and u, where q = p+u, r = p−u, and assumed that solutions of the equations can be expanded in the form of parabolic cylinder functions $D_{n} \left( y \right)$, where $q = \sum\nolimits_{n = 0}^{\infty } {q_{n} \left( x \right)} D_{n} \left( y \right)$. The Rossby wave and Kelvin wave parts of the detailed solutions for the first two terms of expansion forcing source Q are given by Eq. (4.3), Eq. (4.8), Eq. (5.2) and Eq. (5.6) (Gill 1980). General solutions for the rest expansion terms (n ≥ 2) are as follows (Xing et al. 2014):

$$ I = \exp \left( {\frac{1}{8}d^{2} } \right) \cdot \left( {A\left( { - 1} \right)^{n} \frac{1}{n!}\left( {\frac{d}{2}} \right)^{n} } \right)^{ - 1} , $$

(22)

$$ \left\{ {\begin{array}{*{20}l} {\left( {\left( {2n + 3} \right)^{2} \varepsilon^{2} + k^{2} } \right)Iq_{n + 2} = - k\left( {1 + \exp \left( { - 2\left( {2n + 3} \right)\varepsilon L} \right)} \right)\exp \left( {\left( {2n + 3} \right)\varepsilon \left( {x + L} \right)} \right),} \hfill & {x < - L} \hfill \\ {\left( {\left( {2n + 3} \right)^{2} \varepsilon^{2} + k^{2} } \right)Iq_{n + 2} = - \left( {2n + 3} \right)\varepsilon \cos kx + k\left( {\sin kx - \exp \left( {\left( {2n + 3} \right)\varepsilon \left( {x - L} \right)} \right)} \right),} \hfill & {\left| x \right| \le L} \hfill \\ {\left( {\left( {2n + 3} \right)^{2} \varepsilon^{2} + k^{2} } \right)Iq_{n + 2} = 0,} \hfill & {x > L} \hfill \\ \end{array} ,} \right. $$

(23)

$$ \left\{ {\begin{array}{*{20}l} {p = \frac{1}{2}q_{n + 2} \left( x \right)D_{n + 2} \left( y \right) + \frac{1}{2}\left( {n + 2} \right)q_{n + 2} \left( x \right)D_{n} \left( y \right)} \hfill \\ {u = \frac{1}{2}q_{n + 2} \left( x \right)D_{n + 2} \left( y \right) - \frac{1}{2}\left( {n + 2} \right)q_{n + 2} \left( x \right)D_{n} \left( y \right)} \hfill \\ {v = 2\left( {n + 2} \right)\varepsilon q_{n + 2} \left( x \right)D_{n + 1} \left( y \right) + Q_{n} \left( {x,y} \right)} \hfill \\ {w = \varepsilon p + Q_{n} \left( {x,y} \right)} \hfill \\ \end{array} } \right.. $$

(24)