The Interdecadal Changes of South Pacific Sea Surface Temperature in the Mid-1990s and Their Connections with ENSO

Understanding the physical processes of global climate variability on decadal to multidecadal timescales is of great importance for climate predictability because of their influence on both the lives of people and socioeconomic development (Mehta et al., 2000), and indeed much existing literature is devoted to this topic. Previous studies on decadal climate variability have mainly focused on the North Pacific (e.g. Trenberth and Hurrell, 1994; Latif and Barnett, 1996; Nakamura et al., 1997; Wang and Liu, 2000; Li and Xian, 2003; Latif, 2006; Li et al., 2006; Zhou et al., 2007; Wang et al., 2009, 2010, 2012), North Atlantic (e.g. Deser and Blackman, 1993; Kushnir, 1994; Venzke et al., 1999; Eden and Willebrand, 2001; ´Alvarez-garc´ıa et al., 2008; Gu et al., 2009; Nathan and Kravtsov, 2010) and Indian Ocean (Wang et al., 1999). Although the essential dynamics of climate variability are still unclear, these studies nevertheless highlight the characteristics and physical processes of decadal to multidecadal variability of the coupled ocean-atmosphere system.

Among the aforementioned decadal climate variations, the 1976/77 climate regime shift is the most prominent in the global climate system (Nitta and Yamada, 1989; Latif and Barnett, 1994; Graham, 1994; Deser et al., 2004; Xiao and Li, 2007a, 2007b; Agosta and Compagnucci, 2008). Several studies have demonstrated that this climate regime shift is related to the Pacific Decadal Oscillation (PDO) (Mantua et al., 1997; Mantua and Hare, 2002), a phenomenon that reflects the decadal climate variability of the Pacific Ocean, and which changed from a cold phase to a warm phase during 1976/77 (Huang et al., 2005). However, recently, a few studies have reported that there may have been a new climate regime shift in the Pacific Ocean during the 1990s (Xiao and Li, 2007b; Chen et al., 2008; Shakun and Shaman, 2009), as evidenced by a significant change in the biosystem and ocean-atmosphere system, comparable to the 1976/77 climate regime shift, but in antiphase (Chen et al., 2008).

While many studies have been devoted to examining decadal climate variability in the oceans of the Northern Hemisphere, the oceans of the Southern Hemisphere have received far less attention because of poor observational data and the environment there being relatively more inhospitable. Exceptions are the important works of (Banks and Wood, 2002), (Wainer et al., 2004), (Hogg and Blundell, 2006), (Gille, 2008) and (Yuan and Yonekura, 2011). Moreover, some model studies have shown that climate change may be having a significant impact on the Southern Ocean (Banks andWood, 2002; Wainer et al., 2004; Hogg and Blundell, 2006; Yang et al., 2007a, b). (Gille, 2008) suggested that the Southern Ocean warmed during the 1990s, while (Yuan and Yonekura, 2011) suggested that decadal variability is more evident in the mid-latitude Southern Ocean than over Antarctica, and that the Southern Ocean plays an important role at quasi-decadal timescales.

The South Pacific is a vast region covering almost half of the Southern Ocean, and plays an important role in the climate system (Luo et al., 2003;Wang et al., 2007; Shakun and Shaman, 2009; Hsu and Chen, 2011). Recently, (Xiao and Li, 2007b) suggested that the South Pacific Ocean may be one of the most sensitive areas in terms of decadal abrupt changes in the global ocean. They further pointed out that increased decadal abrupt changes happened in SST in the mid-latitude South Pacific during the 1990s. However, the changes in South Pacific SST variability have not been studied in detail. Furthermore, the relationship between SST variability before and after the increased decadal abrupt changes during the 1990s and the El Niño-Southern Oscillation (ENSO) is unclear. (Zhang et al., 2005) suggested based on an idealized study that extratropical climate change may have an impact on the variability of ENSO. We investigated these two questions in the work reported in the present paper.

Although the quality and quantity of oceanic observations in the South Pacific are lacking, satellite-derived SST data are available from the late 1970s onwards, meaning we now have more than four decades’ worth of SST observations available, which is long enough for us to study the decadal variability of South Pacific SST. Section 2 introduces the datasets and methodology used in the study. In section 3, we examine the ocean-atmosphere interactions associated with the increased decadal abrupt changes in the South Pacific. We then turn to the relationships between the increased decadal abrupt changes and ENSO in section 4. Section 5 discusses the possible physical processes of the impact of ENSO on South Pacific SST before and after the decadal abrupt changes. Finally, an overall discussion and conclusions are presented in section 6.

The monthly mean SST data were taken from the Hadley Centre global sea ice and sea surface temperature (HadISST) analysis datasets (Rayner et al., 2003), which have a horizontal resolution of 1°×1°and are available from January 1870 to present, and the National Oceanic and Atmospheric Administration (NOAA) Extended Reconstructed SST datasets Version 3b (ERSST V3b) (Smith et al., 2008), which have a horizontal resolution of 2°×2°and are available from January 1850 to present.

Monthly mean sea level pressure, 850-hPa wind, and wind at 10-m fields were obtained from the National Centers for Environmental Protection (NCEP) Reanalysis II (Kanamitsu et al., 2002). These data are available with a horizontal resolution of 2.5°×2.5 ° from January 1979.

Monthly mean latent heat flux data for the period 1958-2010 with a horizontal of 1°×1° were derived from the third version of the global ocean-surface heat flux products developed by the Objectively Analyzed air-sea Heat Fluxes (OAFlux) project at the Woods Hole Oceanographic Institution (WHOI) (Yu et al., 2008).

Monthly mean upper-ocean (0-700 m) heat content data with a horizontal resolution of 1°×1° were obtained from NOAA for the period January 1955 to present (Levitus et al., 2009).

Finally, Niño3.4 (5°S-5°N, 170°-120°W) SST anomalies (SSTAs) time series were used to indicate the variability of ENSO in the tropical Pacific.

We focused on the recent period of 1979 to present because the quality and quantity of SST observations are much better.

All the anomalies of the above fields were computed by removing the monthly mean climatology and long-term linear trends. Empirical Orthogonal Function (EOF) analysis was performed on the covariance matrix. Lead-lag correlation, regression, and composite analysis were also adopted. The moving t-test (MTT) method was used to detect the decadal abrupt changes of SST (Xiao and Li, 2007b). The global wavelet power spectrum was used to identify the periodicity of the EOF mode.

Figure 1 shows the time series of the area-averaged SSTA in the South Pacific (60°-20°S, 144°E-68°W) for the period January 1979 to December 2011 based on ERSST (Fig. 1a) and HadISST (Fig. 1b) data. It can clearly be seen that there was a prominent interdecadal change in the South Pacific SSTA in the mid-1990s. The South Pacific SSTA changed from being negative during the period 1979-94 to positive during the period 1996-2011. This interdecadal variation can be clearly seen from the 7-yr running averaged time series of the South Pacific SSTA, which indicates a significant interdecadal shift as having occurred in the mid-1990s.

Figure 1.  Time series of the monthly mean SSTA in the South Pacific (60°-20°S, 144°E-68°W) for the period Jan 1979 to Dec 2011 based on (a) ERSST and (b) HadISST data (units: °C). The solid thick red lines are the 7-yr moving average time series.

Figure 2.  Monthly mean SSTA time series blue line, Jan 1979 to Dec 2011) and the epoch average divided by the decadal abrupt changes (red line) in the South Pacific based on (a) ERSST and (b) HadISST data (units: °C). The decadal abrupt changes detected by the MTT method exceed the 99% confidence level.

Figure 3.  Composite of the SSTA (units: °C) during Jan 1979 to Dec 1994 (top row); composite of the SSTA during Jan 1996 to Dec 2011 (middle row); difference of the SSTA between Jan 1996 to Dec 2011 and Jan 1979 to Dec 1994 (bottom row). The shaded areas in the bottom row are significant at the 95% confidence level based on the Student’s t-test.

Figure 4.  The composite of global 0-700-m heat content anomalies (units: GJ m^-2) during (a) Jan 1979 to Dec 1994 and (b) Jan 1996 to Dec 2011. (c) The difference of the SSTA between Jan 1996 to Dec 2011 and Jan 1979 to Dec 1994. The black contoured areas in the bottom are significant at the 95% confidence level based on the Student’s t-test.

Figure 5.  The leading EOF mode of the South Pacific SSTA (units: °C) for (a) Jan 1979 to Dec 1994 and (b) Jan 1996 to Dec 2011. The shaded areas are significant at the 90% confidence level based on the Student’s t-test.

Figure 6.  (a) The normalized PC time series of the leading EOF mode (blue line) and the Niño3.4 index (red line) for the period Jan 1979 to Dec 1994. (b) The same as (a), but for the period Jan 1996 to Dec 2011. The global wavelet spectra of EOF-P1 and EOF-P2 are shown in (c) and (d), respectively. The red line in (c) and (d) indicates significance at the 90% confidence level based on the Student’s t-test.

In order to further study this interdecadal shift, the MTT method (Xiao2007a, b) was employed to detect the decadal abrupt changes of the South Pacific SSTA. In Fig. 2, it is clearly shown that there was a significant increased decadal abrupt change in the South Pacific SSTA in 1996, with an enhancement (0.18°C) of the mean SSTA value (Fig. 2a). This increased decadal abrupt change during the mid-1990s can also be seen in the HadISST data (Fig. 2b); more specifically, in 1995, with an enhancement of 0.17°C.

The aforementioned decadal abrupt change of the South Pacific SSTA in the mid-1990s in the South Pacific was pointed out by Xiao and Li (2007a), which found that the South Pacific may be one of the most sensitive areas in which the decadal abrupt change of the SSTA may occur more easily. They further pointed out that a decadal abrupt change of the subtropical South Pacific (36°-24°S, 150°-136°W) SSTA occurred in 1994, based on ERSST V2 data.

To reveal the spatial distribution of the interdecadal variability of the SSTA in the South Pacific, we calculated the differences of the South Pacific SSTA between the two periods of January 1979 to December 1994 (Period 1, or P1) and January 1996 to December 2011 (Period 2, or P2). The differences derived from ERSST and HadISST data are shown in Fig. 3. As can be seen, the composite SSTA during P1 was negative in the mid-latitudes of the South Pacific, but positive in the high-latitudes (top right). However, by P2 (middle right), the SSTA in the mid-latitudes (high-latitudes) of the South Pacific had changed substantially from negative (positive) to positive (negative). It can also be seen that there were large and significant differences of the SSTA in the mid- and high-latitudes of the South Pacific (bottom right). The most significant differences appeared in the mid-latitudes of the South Pacific, bearing a resemblance to the results reported by (Xiao and Li, 2007a). In general, the SSTA increases of approximately 0.8°C in the mid-latitudes of the South Pacific were in sharp contrast to the SSTA decreases of about 0.4°C in the high-latitudes (bottom right). Besides, it should be noted that those regions with SSTA decreases were smaller than those with SSTA increases. Therefore, one can argue that the South Pacific SSTA changes in 1994/95 are characterized by basin-scale warming. HadISST data were found to reproduce the main characteristics of the South Pacific SSTA variation, except that the amplitude appears to be weaker (left panels).

It should be noted that the upper-ocean heat content can feed the variability of SST on the decadal timescale (Wang et al., 2003b). Therefore, it is necessary to study the relationship of SSTA variability with the upper-ocean heat content. Figure 4 presents the composition of the upper-ocean (0-700 m) heat content anomalies (HC700A) for P1 and P2 (Figs. 4a and b, respectively). Also shown are the differences of HC700A in the South Pacific between these two periods. It is clearly shown that the spatial distribution of HC700A bears a resemblance to that of the SSTA in Fig. 3. During P1, the HC700A was negative in the mid-latitudes and southwestern South Pacific. A positive HC700A was mainly observed in the high-latitudes and northeastern South Pacific. For P2, however, the HC700A showed an almost opposite pattern compared to that for P1. The significant variability of HC700A was found in the subtropical (positive) and high-latitude (negative) South Pacific (Fig. 4c), which was consistent with the variability of the SSTA. This indicates that the variability of South Pacific HC700A may have a significant impact on that of the SSTA in the South Pacific.

Although it is known that there was a pronounced interdecadal change in the South Pacific SSTA around 1995/96, its dominant mode on the interannual timescale is not clear. Therefore, to study this, EOF analysis was performed on the South Pacific SSTA after removing the linear trend for P1 and P2, respectively. Before the analysis, a band-pass filter was applied to the SSTA for retaining the variability between 11 and 85 months (i.e. the interannual timescale). Qualitatively similar analysis results were obtained from using both the HadISST and ERSST data; therefore, we only show the results from HadISST in this paper.

Figure 5 displays the spatial patterns of the EOF for P1 (Fig. 5a) and P2 (Fig. 5b). Note that the spatial patterns were obtained by regressing the South Pacific SSTA with the normalized principal components. As suggested by (Wang et al., 2003a), however, EOF analysis has a shortcoming that is relevant here. That is, the interannual component of the SSTA was estimated by the band-filter (between 11 and 85 months), and this process ought to reduce the degrees of freedom of the SSTA. Therefore, we needed to calculate the effective degrees of freedom based on the method used by (Davis, 1976). The effective degrees of freedom N is computed as n/T, where n is the number of sample observations, and T=∑^K_τ=0 C_xx(τ) C_yy(τ) between two different fields. C_xx(τ) and C_yy(τ) are the autocorrelation coefficients of x_i(i=1,…,n) and y_i(i=1,…,n), respectively, with a time lag of r. The maximum of the integer K corresponds to n/2.

We first examine the spatial characteristics of SSTA variability during P1. Note that we defined the SSTA relative to the mean SST from January 1979 to December 1994. Figure 5a displays the leading EOF mode of the South Pacific SSTA for P1 (EOF-P1). The percentage variance explained by this mode was 36.6%, and it was well separated from the lower EOF modes in terms of explained variance according to the criterion of (North et al., 1982). This suggests that this mode was stable and robust, and thus we only discuss this leading mode of SSTA variability here. In order to reduce the potential influence of ENSO, we chose the domain south of 20°S. However, it is noted that the EOF pattern did not change significantly when we chose the region south of 10°S, or even the equator (figure not shown).

The pattern of EOF-P1 was characterized by an anomalous high SSTA in the northeast and high-latitude region of the South Pacific, and an anomalous low SSTA in the subtropical South Pacific and east of New Zealand (Fig. 5a). In general, this mode showed a significant tripolar structure of the SSTA pattern in the South Pacific, with two negative SSTA centers east of New Zealand and in the mid-latitude South Pacific, and a positive center in the high-latitude South Pacific (45°-60°S, 160°-110°W). This pattern has a close relationship with ENSO (Shakun and Shaman, 2009; Li et al., 2012), with (Shakun and Shaman, 2009) suggesting that this mode may be viewed as a reddened response to ENSO. Besides, this pattern was quite similar to the structure of the South Pacific Decadal Oscillation (SPDO) mentioned by previous studies (Shakun and Shaman, 2009; Hsu and Chen, 2011). Therefore, it may be that the spatial features of the dominant mode in the South Pacific on both the interannual and interdecadal timescales are similar, which is not presented in the North Pacific (Yeh and Kirtman, 2003).

We next examine the spatial features of the South Pacific SSTA for P2 (i.e. January 1996 to December 2011). Note that we defined the SSTA relative to the mean SST of this period. Figure 5b presents the leading EOF mode of the South Pacific SSTA for P2 (EOF-P2), which describes 43.9% of the total variance. Although the position of the positive SSTA did not change significantly, the amplitude of the positive SSTA in the high-latitude South Pacific weakened remarkably. Besides, the negative SSTA in the mid-latitude South Pacific showed a more pronounced variability than the positive SSTA. Note that the anomalously low SSTA center around New Zealand almost vanished in P2. Another negative SSTA center in the subtropical South Pacific extended northwestward to the region northeast of New Zealand. In summary, the spatial features of the South Pacific SSTA during P2 were mainly characterized by a dipole-like structure in the meridional direction.

Based on the above analyses, we can conclude that the spatial manifestations of the South Pacific SSTA between P1 and P2 showed significant differences, especially in the mid-latitude South Pacific. Besides, it was clearly shown that the negative SSTA appeared around New Zealand during P1, but not during P2. In accordance with the results of (Kushnir et al., 2002), this negative SSTA in the mid-latitude South Pacific may have induced atmospheric circulation anomalies with a poleward cyclone and an equatorward anticyclone, which may have had an important impact on the climate variability of Australia. However, investigating this aspect is beyond the scope of the present study.

In this section, we examine the temporal features of the leading mode of the SSTA in the South Pacific. Note that the principal components (PCs) of the leading mode were normalized by subtracting the long-term means and dividing by the long-term standard deviations of the respective periods.

Figure 6 displays the PC time series for P1 (PC-P1) and P2 (PC-P2). Note that both of the PCs were characterized by significant variations on the interannual timescale. By comparing the PCs with the Niño3.4 index (red line), we can see that both of the PC time series of the leading mode during P1 (Fig. 6a) and P2 (Fig. 6b) were significantly correlated with SST index in the Niño3.4 (the correlation coefficients were 0.78 for P1 and 0.88 for P2, both of which were statistically significant at the 99% confidence level). This suggests that the variability of the South Pacific SSTA has a close relationship with ENSO. We further discuss the relationships between the variability of SSTA in the South Pacific and ENSO in the next section.

To examine the exact interannual periodicity of the leading mode during P1 and P2, we present the global wavelet spectrum of the two leading modes in Figs. 6c and d. The blue line is the global wavelet spectrum and the red line indicates that the periodicity is significant at the 95% confidence level. Note that the EOF-P1 mode has only one spectral peak at 4.6 yr (significant at the 95% confidence level) (Fig. 6c). However, the EOF-P2 mode has a significant spectral peak at 2.7 yr (significant at the 95% confidence level) (Fig. 6d). Therefore, the main periodicity on the interannual timescale for P2 was shorter than that for P1.

In summary, we can see that there was a prominent difference between P1 and P2 not only in terms of the spatial features, but also the temporal features of the leading EOF modes of the South Pacific SSTA. Besides, the relationships between the leading EOF modes and ENSO also experienced remarkable variation between P1 and P2.

The aforementioned results suggest that the relationships between the leading EOF modes of the South Pacific SSTA for P1 and P2 and ENSO show some differences. To further examine their relationships with ENSO, we first discuss the lead-lag correlations between the PC time series of the two leading EOF modes and the Niño3.4 SST index. Figure 7 shows the lead-lag correlation coefficients between the PC time series of the leading EOF mode and the Niño3.4 SST index for P1 and P2, respectively.

It can clearly be seen from the results presented in Fig. 7 that PC-P1 had the largest correlation coefficient with the Niño3.4 SST index (0.83, significant at the 95% confidence level according to the Student’s t-test) when the Niño3.4 SST index led PC-P1 by two months (blue line). However, the correlation between PC-P2 and the Niño3.4 SST index showed remarkable change; PC-P2 had the largest correlation coefficient with the Niño3.4 SST index (0.9, significant at the 95% confidence level according to the Student’s t-test) when the Niño3.4 SST index led PC-P2 by one month (red line). Therefore, although ENSO had a pronounced impact on the variability of the South Pacific SSTA both for P1 and P2, we can conclude that ENSO had a larger and faster impact on EOF-P2 than on EOF-P1. The South Pacific SSTA response to ENSO for P1 was one month slower than the response to ENSO for P2. This means that, during the different time periods, the impact of ENSO on SST in the South Pacific had different temporal responses, since there were different oceanic/atmospheric basic states in periods P1 and P2.

In order to further study the spatial features of the South Pacific SSTA, we also examined the variability of the South Pacific SSTA associated with individual El Niño events.

Figure 7.  Lead-lag correlation coefficients between the PC of the leading EOF mode and the Niño3.4 index for the period Jan 1979 to Dec 1994 (blue line) and Jan 1996 to Dec 2011 (red line). The dashed line denotes the 95% confidence level.

Figure 8.  Austral summer (DJF) SSTA for (a) 1982/83, (b) 1986/87, (c) 1987/88, (d) 1991/92, and (e) the composite of the four events. Pattern correlation coefficients with the leading EOF mode (Fig. 4a) during January 1979 to December 1994 are also shown. The black contoured areas in Fig. 7e are significant at the 90% confidence level or better according to the Student’s t-test.

Figure 9.  The same as in Fig. 8, but for (a) 1997/98, (b) 2002/03, (c) 2004/05, (d) 2006/07, (e) 2009/10, and (f) the composite of the five events.

Figure 10.  The SSTA regressed with the Niño3.4 index from lag -10 months to lag 10 months for the period Jan 1979 to Dec 1994. Positive lags denote that the Niño3.4 index leads the South Pacific SSTA.

Figure 11.  The same as Fig. 10, but for the period Jan 1996 to Dec 2011.

Figure 12.  The same as Fig. 10, but for the 850-hPa wind and SLP.

Figure 13.  The same as Fig. 11, but for the 850-hPa wind and SLP.

Figure 14.  The same as Fig. 10, but for the wind at 10 m and the latent heat flux.

Figure 15.  The same as Fig. 11, but for the wind at 10 m and the latent heat flux.

Based on the Ocean Niño Index from the NOAA, we ascertained that four El Niño events (i.e., 1982/83, 1986/87, 1987/88, 1991/92) took place during P1 and five (i.e., 1997/98, 2002/03, 2004/052006/07, 2009/2010) during P2.

Figure 8 displays the austral summer (December-January; DJF) South Pacific SSTA associated with each of the four El Niño events during P1 and their pattern correlation coefficients with EOF-P1. Note that the pattern correlation coefficients were defined as the correlation coefficients for all overlapping grid points between the two spatial patterns (Wang, et al,. 2011; Yeh, et al,. 2011). It can clearly be seen that the spatial patterns of the South Pacific SSTA for 1982/83 and 1991/92 resembled EOF-P1 significantly (the correlation coefficients were 0.94 and 0.85, respectively), except for 1986/87 (-0.13) and 1987/88 (-0.22). The correlation coefficient between the composited South Pacific SSTA based on the four El Niño events and EOF-P1 was 0.92 (Fig. 8e).

Figure 9 displays the South Pacific SSTA (DJF) associated with each of the five El Niño events during P2 and their pattern correlation coefficients with EOF-P2. It shows that, except for 2004/05 (0.28) (Fig. 9c), the South Pacific SSTA in austral summer bore a significant resemblance to EOF-P2 (the correlation coefficient ranging from 0.59 to 0.83), especially for the composited SSTA (0.94) (Fig. 9f).

Based on the above case analysis, it is further confirmed that ENSO has an important influence on the variability of the South Pacific. Moreover, comparing the spatial correlation coefficient during P1 with that during P2, we may infer that ENSO had a stronger impact on the South Pacific SSTA for P2 than that for P1, the reason for which will be studied in the next section. Besides, it should be noted that most El Niño events selected during P1 were typical canonical El Niño events (Rasmusson and Carpenter, 1982), while all events selected during P2 were El Niño Modoki events (Ashok et al., 2007), except that of 1997/98. Is the changing variability of the SSTA in the South Pacific during the two time periods attributed to the frequency of El Niño Modoki? Future work to answer this question may be pivotal for understanding the decadal variability of the South Pacific SST, but was beyond the scope of this study.

In view of the lead-lag correlations between the leading EOF modes and ENSO described above, it was necessary to investigate the impact of ENSO on the variation of the South Pacific SSTA in the different periods (i.e. P1 and P2). To address this question, the spatial patterns of the regressed lead-lag South Pacific SSTA with the Niño3.4 SST index during P1 and P2 were established, and the results are shown in Figs. 10 and 11, respectively.

During the development of ENSO (from lag -10 to -4) in P1 (Fig. 10), although the negative SSTA in the southeast South Pacific showed a remarkable decay, the negative SSTA in the region north of New Zealand showed a significant eastward expansion along with an enhancement of its intensity. Besides, the positive SSTA around New Zealand also showed eastward propagation. About two months before the peak of ENSO (i.e. at lag -2), the negative SSTA in the region northeast of New Zealand began to extend southwestwardly and the positive SSTA in the high-latitude South Pacific intensified significantly. Note that the tripolar structure of the South Pacific SSTA described in the aforementioned analysis occurred at the peak of ENSO (i.e. lag 0) and reached its peak about two months after the peak of ENSO (i.e. lag 2). During the decay period of ENSO (from lag 4 to 10) we can see that, although the intensity of the tripolar pattern of the South Pacific SSTA weakened significantly, it did not show any kind of remarkable propagation. In addition, we can see that the negative SSTA in the subtropical South Pacific showed a faster decay than that around New Zealand.

Compared with P1, the variability of the South Pacific SSTA during P2 showed some differences (Fig. 11). The negative SSTA over the Tasman Sea showed a significant eastward propagation in the mid-latitudes of the South Pacific during the whole ENSO period, and its intensity strengthened significantly with eastward expansion during the development of ENSO (from lag -4 to 0) and weakened remarkably during the decay of ENSO (from lag 2 to 8). The main characteristic of the evolution of the positive SSTA in the region southeast of New Zealand was the resemblance to the variation of the negative SSTA, in terms of not only the propagation but also the intensity. A meridional dipole-like pattern of the SSTA began to appear two months before the peak of ENSO (i.e. lag -2) and reached its peak at about two months after the peak of ENSO (i.e. lag 2). Besides, we can see that the positive SSTA, which appeared at the peak of ENSO in the region southeast of Australia, expanded significantly eastward with the enhancement of intensity during the decay of ENSO.

Based on the above analysis, we can conclude that the intensity of the South Pacific SSTA during both P1 and P2 strengthened (weakened) along with the development (decay) of ENSO. In addition, during the development of ENSO, the SSTA in both P1 and P2 showed significant eastward expansion. However, during the decay of ENSO, the SSTA for P1 showed strong persistence, but for P2 it still showed a significant eastward propagation.

In the aforementioned analysis, we examined the impact of ENSO evolution on the variability of the South Pacific SSTA during the different periods (i.e. P1 and P2). However, the process by which ENSO forces the South Pacific SSTA is not clear (Ciasto and Thompson, 2008). We addressed this knowledge gap by investigating the possible process involved in the variability of the periods assessed in the present study. Some similar analyses to those in the study by (Terray, 2011) were completed here. The temporal evolutions of 850-hPa wind and SLP during P1 and P2 are shown in Figs. 12 and 13, respectively. Moreover, in order to examine how surface heat fluxes drive the variability of SSTA, latent heat flux regression patterns on the Niño3.4 SST index for the period P1 and P2 are shown in Figs. 14 and 15.

Figure 12 shows that both the subtropical cyclone and anticyclone over the middle-high latitudes of the South Pacific during the development and mature period of ENSO (from lag -10 to 0) for P1 were strengthened significantly and expanded eastwardly by stronger anomalous westerlies and easterlies, respectively. Besides, accompanying the development of an anticyclone over Australia, the dominant longitude dipole mode over the middle-high latitudes of the South Pacific transformed a wave-train mode with a northwest-southeast direction. During the decay period of ENSO (from lag 0 to 10), both the cyclone and anticyclone were significantly weakened, accompanying the weaker anomalous westerlies over the subtropical South Pacific and easterlies over the high-latitude South Pacific. Furthermore, the wave-train mode with a northwest-southeast direction was weakened continuously in the mature period of ENSO. However, the cyclone over southeastern New Zealand showed a slower decay.

For P2, we can see that the variability of 850-hPa wind and SLP during the development of ENSO (Fig. 13) showed a remarkable resemblance to that for P1. However, after the peak of ENSO (from lag 0 to 10), the cyclonic circulation over southeastern New Zealand began to strengthen and expand southeastwardly toward the high-latitudes of the South Pacific and a south-northward dipole circulation pattern existed continually. At the same time, stronger anomalous easterlies (westerlies) appeared in the mid-latitudes (high-latitudes) of the South Pacific.

The variability of 850-hPa wind and SLP anomalies were associated with the SSTA. The cyclone (anticyclone) was located over the negative (positive) SSTA. This may be evidence that the atmosphere forces the ocean, such that the slower decay of the atmosphere after the peak of ENSO during P1 seemed to result in a stronger persistence of the SSTA (see Fig. 10). However, after the peak of ENSO during P2, we can see that there was a reversal phenomenon, which seems to indicate that the ocean forces the atmosphere.

The regression patterns of latent heat flux shown in Figs. 14 and 15 suggest that the latent heat flux is an important contributor to the variability of the SSTA, which is consistent with many previous studies (Ciasto and Thompson, 2008; Ciasto and England, 2011; Terray, 2011). The stronger anomalous westerlies (easterlies) in the mid-latitudes (high-latitudes) of the South Pacific resulted in the enhancement of upward (downward) latent heat flux during the development of ENSO both for P1 and P2, which generated the ocean cooling (warming). After the peak of ENSO, the positive latent heat flux over eastern Australia during P1 (Fig. 14) and the mid-latitude South Pacific during P2 (Fig. 15) caused the negative SSTA to weaken significantly. Note that latent heat flux is associated with both surface wind and humidity. Therefore, it does not correspond entirely to SSTA variability.

The SST in the South Pacific experienced an obvious interdecadal abrupt change around 1995/96. The SSTA was mainly negative during the period 1979-1994 (i.e. P1) and positive during 1996-2011 (i.e. P2). The difference of spatial SSTA pattern between P2 and P1 was that the positive (negative) SSTA in the mid-latitude (high-latitude) South Pacific was more significant in the mid-latitude South Pacific than that in the high-latitude South Pacific. The variability of the SSTA was characterized by a basin-scale-like warming. This variability of the South Pacific SST may have been a response to the Pacific pan-decadal variability (PDV) interdecadal regime shifts (Chen et al., 2008).

We analyzed the leading EOF mode of the South Pacific SSTA in both P1 and P2. Both the spatial and temporal features of the two leading EOF modes during P1 and P2 showed significant differences. The EOF-P1 that showed a significant periodicity of 4.6 yr was characterized by a tripolar structure, with two negative centers in the subtropical South Pacific and around New Zealand, and a positive center in the high-latitude South Pacific. However, EOF-P2 showed a dipole-like structure with a negative center in the region northeast of New Zealand and a positive center in the high-latitude South Pacific. The main periodicity of EOF-P2 was about 2.7 yr.

ENSO has an important impact on the variability of South Pacific SST. Therefore, the relationships of ENSO with EOF-P1 and EOF-P2 were also examined. EOF-P2 showed a closer and faster response to ENSO than EOF-P1 based on the lead-lag correlations. The changes of EOF-P1 and EOF-P2 may be attributed to the variability of ENSO forcing. During the development of ENSO, the ENSO forcing would have triggered a cyclone (anticyclone) accompanied by stronger than normal westerlies (easterlies) in the mid-latitude (high-latitude) South Pacific in P1 and P2, causing the SSTA in both P1 and P2 to propagate eastwardly and strengthen significantly. However, after the peak of ENSO, the SSTA in P1 that showed stronger persistence was mainly forced by the atmosphere triggered by ENSO. During P2, the ENSO forcing seemed to be weak and the atmosphere began to show the characteristic of the response to SST forcing after the peak of ENSO. Besides, the variability of the SSTA over the South Pacific can partly be attributed to the latent heat flux associated with surface wind and humidity. In fact, a previous study has indicated that the decadal variability of South Pacific SST has a strong background of air-sea interaction (Yang et al., 2007b). Therefore, to study the air-sea interaction in the South Pacific, we need to further investigate the air-sea fluxes quantitatively. Future work is required to determine to what extent air-sea fluxes contribute to the variability of SST based on model simulations.

As we know, the atmospheric response to external forcing depends on not only the intensity and location of the external forcing but also the atmospheric basic state. For the variation of SST in the South Pacific, its response to ENSO forcing will depend on the intensity of ENSO, but the basic states in the ocean and atmosphere are also important. The present results highlight the fact that ENSO has a different impact on the South Pacific SSTA based on observations. However, these results need to be further studied by coupled atmosphere-ocean models, which can produce a more reasonably realistic mechanism.