Monthly average temperature and salinity profile data from 2001 to 2014 that were used to calculate the OHC were extracted from a data set provided by the Japan Agency for Marine-earth Science and Technology. It is produced by merging many types of in situ observational data, but most are Argo profiling data (Fig. 1). This data set was available on a 1°× 1° grid for 25 pressure levels spanning (0–2 000)×10⁴ Pa, with separate levels 10×10⁴ Pa apart above the 150×10⁴ Pa level and (20–100)×10⁴ Pa apart below the 150×10⁴ Pa level (Hosoda et al., 2008). The temperature and salinity profile data in the WPWP (20°S–30°N, 120°E–140°W, (0–120)×10⁴ Pa) region were used.

The OHC and the OSC can be integrated in a vertical profile as

where c_p is the specific heat of seawater at a constant pressure (c_p varies with the temperature and salinity); ρ is the seawater density; and dz is the vertical extent of the integration interval. The unit of the OHC is J/m² and that of the OSC is g/m²; we assume OHC/OSC is uniform across each grid.

We also integrated the OHC and the OSC in a horizontal profile at various depths:

where A is the horizontal extent of the integration interval. After the integration, we calculated the average of the OHC and the OSC at 10×10⁴ Pa, 50×10⁴ Pa and 120×10⁴ Pa, and researched yearly variations of their averages after standardization. In this situation, the unit of the average OHC is J/m³ and that of the average OSC is g/m³.

Given evidence that many signals in geophysical data are cyclostationary (Kim and Gerald, 1997; Kim and Wu, 1999; Kim and Cheng, 2001), we introduced the concept of a cyclostationary empirical orthogonal function (CSEOF) analysis to capture time-varying spatial patterns (LV component) and longer-time scale fluctuations (PC component) present in geophysical signals. The significant difference between the CSEOF and EOF analyses is LV time dependence, which allows the spatial pattern of each CSEOF mode to vary in time (Figs 5 and 7). The temporal evolution of the spatial pattern of the CSEOF LVs is constrained to be periodic with a selected “nested period”. That may be expressed as

where r and t represents spatial and temporal fields respectively, d is the nested period, an empirical number. Here we take d=12 because the period of the time series in the first mode of the EOF analysis is approximately 12 months. The advantages of the CSEOF analysis are that it can minimize mode mixing (a common problem in a EOF decomposition) and can account for both high and low frequency components of the annual cycle, and it does not require removal of the annual signal.

The WPWP is usually noted for its high temperature (SST>28°C) and low salinity (SSS<35). Figure 2 shows locations of the 28.5°C isotherm and 34.8 isohaline at the surface (10×10⁴ Pa) during the ENSO events. The eastern boundary of the isotherm expands substantially eastward during warm events and shrinks westward during cold events, with a double-tongue pattern. The north tongue is around 6°N with a longitudinal extent of 30°, and the south tongue is near 10°S with much smaller extent (about 15°). The trough of the double tongue was over the equator. Observational and modeling results indicate that such interannual displacements of the eastern edge of the warm pool are mainly attributable to the zonal advection associated with local wind forcing and its Kelvin and Rossby wave response (Picaut et al., 1996; Clarke et al., 2000). The northern and southern boundaries of the WPWP do not exhibit remarkable shifts during the ENSO cycle.

The isohaline does not change much in the ENSO events, especially east of the date line. However, in the southwest, there are complex changes that may be caused by heavy precipitation and transport from strong currents near the equatorial convergence zone (Delcroix et al., 1996; Reverdin et al., 1994).

Figure 3 shows yearly variations of area-average OHCs at standard levels. The OHC on the sea surface varies as a sine function with period approximately 1 a and tended to be uniform during 2001–2007. It fluctuated strongly after 2007, during which there were two intermediate cold events (2007/2008 and 2010/2011 La Niña) and one intermediate warm event (2009/2010 El Niño) (Zhang and Weng, 1999). The amplitudes of area-average OHC variations on the subsurface were larger than those in the surface layer, and the OHC in the upper layer near the thermocline in the equatorial Pacific Ocean showed the interannual variability that was positively correlated with the El Niño events; the former usually leads the latter. Such a result is consistent with earlier studies (Chao et al., 2003; Pu et al., 2003). That is, an increasing number of works found that the subsurface ocean frequently had much greater and more stable changes. Finally, we conclude that the OHC at each standard level during the ENSO years had very different variabilities than the neutral years. The greatest change was usually around the 120×10⁴ Pa layer, and the OHC change in the subsurface was opposite that on the surface layer. This may be related to the fact that abnormal signals in the ENSO cycle were first found in the subsurface ocean of the western Pacific and then moved eastward along the thermocline (Chao et al., 2003).

The annual variations of the area-average OSCs at standard levels were simple. Almost all trends exhibited a short-term climatic change (Fig. 4). In contrast with the OHC, the OSC on the surface had greater change than that in the subsurface, i.e., the amplitudes of the OSC variations decreased from the surface to subsurface ocean. Further, the OSC grew substantially after 2007 and maintained large values between 2009 and 2013.

The first mode of the OHC variations in the WPWP had a zonal dipole pattern with see-saw anomalies in the western and eastern parts. The variance contributions of the first three modes are 46.57%, 10.91% and 8.26%. The sum of the first two modes’ feature vectors is 57.49%, i.e., there were two tongue-shaped, large-value variation areas. One was in the region 0°–12°N, 130°–155°E with axis at approximately 8°N and the other was east of the date line with its axis around the equator. Maximum absolute values in the two areas were basically the same. This zonal antiphase pattern of the variations indicates that when the OHC of the west part of the WPWP increased, the OHC of the east part decreased, and vice versa. Figure 4 shows the spatial distribution of the OHC in the WPWP in the depth range (0–120)×10⁴ Pa (January–December).

Seasonal OHC variations also had this zonal antiphase feature. The amplitude maximized in winter (November through February) and minimized in summer (May through August). In winter, the maximum absolute values in both variation areas increased (i.e., the amplitude increased) and the areas enlarged, with the west area expanding eastward and the east area expanding westward. In summer, the maximum absolute values in both variation areas decreased and the areas shrank, the west area toward the west and the east area toward the east.

Figure 5 shows the time series of the first mode obtained by the CSEOF analysis. As seen from that figure and Fig. 4 (spatial anomaly field), when the time series maximized, the spatial anomaly field was large in the east and small in the west. At the trough, the field was large in the west and small in the east. This indicates that the OHC of the WPWP region in the depth range (0–120)×10⁴ Pa had three phase adjustments after 2007. Because this mode resembles the spatial structure of the ENSO, it may have a relationship with it. In addition, from a maximum entropy spectral analysis, the time series of the first mode show a significant period of 3 a.

As revealed by the CSEOF analysis of the OSC, the variance contributions of the first three modes are 33.14%, 21.96%, and 10.43%. The sum of the first two modes’ feature vectors is 55.1%. Figure 6 shows the first mode’s distribution in the WPWP region at depths (0–120)×10⁴ Pa as revealed by the CSEOF method (January–December). It is seen from the figure that the OSC distribution in the WPWP roughly had a positive–negative–positive tripole pattern from south to north. Two positive variation areas were northwest and south of the study area, and a negative variation area was between 18°N and 8°S. The latter area expanded gradually from west to east and extended to the northeast and southeast, almost spanning the entire tropical Pacific. In addition, there were two extreme value centers in the negative variation area, one southwest of the equator and the other around 12°N. As seen from the upper OSC variations (January–December) in the WPWP region, the seasonal OSC variations also had this tripole pattern. The amplitudes of variation of the OSC were stronger in winter and spring (January through April), and weaker in summer and autumn (September through December). In winter and spring, the two maximum absolute values in the negative variation area maximized. The south extreme value center gradually extended southeast to 144°W, and its maximum absolute value and area of the center declined. The northeast extreme value center extended southeast and slightly east, oscillating east of the date line. The amplitudes in the north and south positive variation areas maximized in winter and spring, and minimized in summer and autumn. The coverage of the north positive variation area changed little with season, but the location of its extreme value center moved to the south in winter and spring, and to the north in summer and autumn. The south positive variation area expanded to the northwest in winter and spring, and shrank to the southeast in summer and autumn.

Figure 8 shows the time series of the first mode in the OSC variations. As seen from Figs 6 and 7 (spatial anomaly field), when the time series reaches its peak, the spatial anomaly of the OSC was large in both the north and south, and small in the middle. When it drops into the trough, the field is large in the middle and small in both the north and the south. The time series also show that OSC of the WPWP region only had one phase adjustment after 2007. In addition, an analysis via a maximum entropy spectral method showed that the OSC variations had significant period of 3 a.

As is well-known, the main period of the ENSO is 3–7 a. During the El Niño, the water temperature in the middle and eastern equatorial Pacific increases, while that of the western Pacific decreases. The situation during the La Niña is opposite. The ENSO events have a major influence on large-scale climate changes around the world. The Niño3.4 index is one of the parameters reflecting the ENSO cycles. To discover possible relationships between the OHC interannual variations and ENSO events, we analyzed the relationship between the Niño3.4 index and the time series of the OHC in the WPWP region (Fig. 9).

It is shown in Fig. 9 that the two time series have very similar trends, and their correlation coefficient is 0.73 (passing the significance test with a confidence level of 0.05). After combining the time series with the spatial pattern of the first mode, we see that during the El Niño (as in 2010), the eastern part of the WPWP (overlapping with the Niño3.4 region, i.e., 5°S–5°N, 120°–170°W) had the positive OHC anomalies, indicating an increased OHC. The western part had the negative anomalies, indicating the decreased OHC. The situation during the La Niña (as in 2007/2008) is opposite. Studies have shown that the positive and negative SST anomalies during the ENSO originate from the subsurface layer in the area around the warm pool, then spread eastward and upward along the thermocline in the equatorial region under certain atmospheric conditions. This results in the El Niño/La Niña events in the conventional sense when they spread to the surface layer of the eastern equatorial Pacific (Chao, 2001).

The upper OHC variations are affected by ocean–atmosphere net heat flux and advection. By calculating the regional correlation coefficient between the net heat flux through the ocean surface and the time series of the first mode from the CSEOF analysis of the OHC, we determined that the large-value zone of the coefficient (R>0.4 and R<–0.4) between the two was mainly between 10°N and 10°S. This indicates that the area whose OHC is greatly affected by the net heat flux through the ocean surface is mainly in the equatorial zone. Many other studies (Cayan, 1992; Kelly, 2004; Moon and Song, 2013) have suggested that the impact of the net heat flux through the ocean surface on the OHC is mainly on a seasonal scale, whereas the advection has a more important effect on OHC interannual variations. Therefore, we focused on the contribution of the heat advection.

As studies have suggested, westerly wind anomalies excite motions in the equatorial wave guide that depress the thermocline in the east, reduce equatorial upwelling, and enhance eastward transport of warm surface waters from the western Pacific (Cane and Zebiak, 1985). The wind-forced zonal advection is important in the thermodynamics of the WPWP on interannual time scales (McPhaden and Picaut, 1990). At the initial stage of the El Niño, there is a positive temperature anomaly in the subsurface layer of the WPWP. With the burst of westerly winds, this anomaly spreads eastward along the thermocline in the equatorial region. Because the thermocline in the Pacific is oblique from east to west, the temperature anomaly spreads from the subsurface layer of the western Pacific (120–200 m) to the near-surface layer of the eastern Pacific (20–60 m), resulting in an ENSO event. Afterward, this temperature anomaly spreads from east to west along the channel between 10° and 20°N (Mu and Li, 2000), laying the foundation for the next El Niño/La Niña as it reaches to the western Pacific. That is, the temperature anomaly in the subsurface layer of the Pacific circulates between the eastern and western Pacific, and there is a strong correlation between this temperature anomaly spread and zonal wind anomalies in the western Pacific. Therefore, it is reasonable to state that the OHC variations are induced by the ENSO through the wind forcing and the ocean current advection. This also explains why the spatial pattern of the first mode of the OHC has a zonal antiphase pattern that resembles the spatial structure of the ENSO.

Figure 10 shows time series correlation between the first mode from the CSEOF analysis of the OHC and zonal wind anomalies. As shown, the large-value area of the correlation coefficient between the two is mainly within the equatorial zone, where the correlation is positive. The maximum correlation (R>0.6) is over 5°S–5°N and 177°E–165°W. The correlation between the two is weakly negative in subtropical areas.

Figure 11 shows the OHC transport by the zonal currents (${Q_{{\rm{OHC}}}} = \mathop \int {\rm{OHC}} \times \vec U{\rm{d}}z$; ${Q_{{\rm{OSC}}}} = \mathop \int {\rm{OSC}} \times \vec U{\rm{d}}z{\rm{}}$) and the time-varying condition. The OHC transport corresponded well with the NEC, the NECC, and the SEC (Wu et al., 2015). The time series shows that during the El Niño, the OHC of the WPWP was mainly transported eastward, with the opposite pattern during the La Niña. This is consistent with the aforementioned OHC distribution with a positive anomaly in the east during El Niño and in the west during the La Niña. It is thus concluded that the zonal current and the zonal wind may be influences on the interannual OHC variations.

Studies on the change mechanism of the salt content at basin scale are very limited. However, studies of the salt content at smaller scales found that ocean current and the freshwater flux through the ocean surface are the two main causes of the OSC variations (Delcroix et al., 1998). Our OSC data were obtained by an integral calculation over depth, so the influence of the vertical currents is not addressed in this paper. One study (Gouriou and Delcroix, 2002) suggested that the upper OSC variations in the convergence belt of the southern Pacific are the strongest (that study also reveals that this is the main OSC variation area). Because the freshwater flux through the ocean surface in the study area is mainly dependent on precipitation, the following addresses possible mechanisms resulting in the OSC variations mainly from two aspects: zonal current and precipitation, which are controlled by the ENSO. This is done to explore possible relationships between the OSC and the ENSO cycle.

Because there have been few studies of the OSC, possible mechanisms resulting in upper OSC variations were explored by referring to previous studies. Delcroix et al. (1996, 1998) indicated that around the convergence belt of the southern Pacific, the salinity front of the WPWP divides low-salinity water in the tropical Pacific from high-salinity water, but the interannual displacement of the front is caused by precipitation changes induced by changes of the ascending branches of the Walker and Hadley circulations above the convergence belt of the southern Pacific during the ENSO. This results in the interannual variations of the sea surface salinity distribution (Gouriou and Delcroix, 2002). The interannual variations of the zonal currents may also influence that distribution by the convergence and the divergence. For example, in the latitudinal band of 17°S, the zonal currents are westward to the east of 175°E, and they are eastward west of that longitude. The convergence of the zonal currents as in this pattern could strengthen the salinity front. Therefore, the zonal current anomalies may also alter the sea surface salinity distribution. A westward anomaly of the zonal current speed would be 2–3 months earlier than westward movement of the salinity front, and an eastward anomaly of zonal current speed would be consistent with eastward movement of the salinity front. Under the influence of various factors, the salinity front in the southwest tropical Pacific would move westward and the salinity front in the equatorial Pacific would move eastward during the El Niño, The situation during the La Niña would be opposite. Therefore it is also reasonable to consider that the OSC variations are induced by the ENSO through precipitation and ocean current advection.

The OSC transport by the zonal currents was investigated (figure not shown). OSC was mainly transported westward from March through August by the NEC. There was also weak eastward transport around the NECC and the SEC, but the transport capacity was small. The westward transport by the NEC decreased substantially from September through January. The eastward transport by the NECC and the SEC increased slightly and maximized in December, but the transport capacity was still much smaller than the maximum capacity of westward transport in summer. During the El Niño, the OSC in the WPWP region was mainly transported westward, with the opposite situation during the La Niña. One study suggests that a westward geostrophic current in the western Pacific is intensified during the El Niño (Delcroix et al., 1998). As seen from Figs 6 and 7, the OSC distribution in the WPWP region (specifically, during both the 2009/2010 El Niño and 2010/2011 La Niña events) generally had a negative–positive–negative tripole pattern. That is, the OSC of the north and south parts of the WPWP region continued decreasing, while the OSC of the middle and west parts continued increasing. The amplitudes of the increases in the middle and west regions were different over the years. For example, the amplitude of the OSC increase in 2007/2008 was greatly reduced, while that amplitude during the El Niño in 2010 was relatively large. The above analysis indicates that though the influence of the zonal currents is limited, it still has a positive feedback on the OSC.

The interannual variations of the freshwater flux in the WPWP region were further analyzed (Fig. 12). The time series of the first mode from the EOF analysis for freshwater flux shows that it is 6 months ahead of the OSC, and its correlation coefficient is the maximum, at –0.71. As seen from Fig. 11, the largest variation area of freshwater flux was in the middle of the WPWP region, presenting a negative–positive–negative tripole pattern, similar to that of the OSC. A positive anomaly of the freshwater flux means that a sea area is losing water, and a negative anomaly means the opposite. By combining the pattern and time series of the fresh water flux, we found that OSC increase in the central part of the study area, and the decrease in the northwest and south parts that occurred 6 months later, lost freshwater and the latter gained freshwater. It can thus be inferred that the sea surface freshwater flux may also have a positive feedback effect on the OSC.