Atmospheric Research 94 (2009) 3–9

Contents lists available at ScienceDirect

Atmospheric Research j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / a t m o s

Seasonal locking of the ENSO asymmetry and its influence on the seasonal cycle of the tropical eastern Pacific sea surface temperature Soon-Il An ⁎, Jung Choi Department of Atmospheric Sciences, Yonsei University, 134 Shinchon-dong, Seodaemun-gu, Seoul 120-749, Republic of Korea

a r t i c l e

i n f o

Article history: Received 14 March 2008 Received in revised form 19 August 2008 Accepted 26 September 2008 Keywords: El Nino Nonlinearity Asymmetry Seasonal cycle

a b s t r a c t Using the monthly extended reconstructed sea surface temperature (SST) dataset version 3 (ERSST.v3) spanning 1880–2007 and SODA ocean assimilation data spanning 1958–2001, we investigate the seasonality of the asymmetry of the El Niño-Southern Oscillation (ENSO) and its possible impacts on the tropical eastern Pacific annual cycle. Like the amplitude of ENSO, the skewness of ENSO (i.e., asymmetry of ENSO) is locked to the seasonal cycle, showing large amplitude during the winter and small amplitude during the spring except May. Furthermore, the seasonality of the asymmetry of ENSO is changing decade by decade, which is strong during 1930s, 1950s, and 1990s and weak during 1900s, 1910s, 1940s, and 1970s. These decadal changes are significantly correlated to those in the amplitude of the annual and semi-annual cycles of tropical eastern Pacific SST, suggesting that the changes in the seasonality of the asymmetry of ENSO may modify the amplitude of the annual and semi-annual cycles of the tropical eastern Pacific SST via a nonlinear process. Using the coupled general circulation model simulations, we also showed similar results to the aforementioned observed features, which overcame some deficiencies due to relatively short reliable records of the observation. © 2008 Elsevier B.V. All rights reserved.

1. Introduction The asymmetry in the amplitudes between the warm and cold events is a robust feature of El Niño-Southern Oscillation (ENSO), which nonetheless is not yet fully understood. The evidence on the asymmetry of ENSO was deduced from the skewness of ENSO indices in observational records (Burgers and Stephenson, 1999) as well as from the results of many coupled ocean-atmosphere models (An et al., 2005b). An et al. (2005a) identified the spatial and temporal behaviors of the El Niño–La Niña asymmetry by using nonlinear principal component analysis (NLPCA) that detects a low-dimensional nonlinear structure in multivariate datasets (Hsieh, 2001, 2004; Wu and Hsieh, 2003). They also showed that the

⁎ Corresponding author. Department of Atmospheric Sciences, Yonsei University, Seoul 120-749, Republic of Korea. Tel.: +82 2 2123 5684; fax: +82 2 365 5163. E-mail addresses: [email protected] (S.-I. An), [email protected] (J. Choi). 0169-8095/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.atmosres.2008.09.029

asymmetry of ENSO became stronger during recent decades (also in An, 2004), consistent with the changes in the general characteristics of ENSO, tracking the climate shifts of the late 1970s (Wang, 1995; An and Jin, 2000; Fedorov and Philander, 2000; An and Wang, 2000; Wang and An, 2001). As a possible cause for the ENSO asymmetry, the nonlinear dynamical heating (NDH) in the tropical Pacific SST budget has been stressed (Jin et al., 2003; An and Jin, 2004; An, 2009). The NDH, anomalous thermal advections by the anomalous three-dimensional currents in the ocean mixed layer, provides an overall warming tendency, and thus typically strengthens the warm events and weakens the cold events (An and Jin, 2004). As a result, long-term accumulation of the nonlinear thermal heating may result in a warming trend, which eventually modifies the climate state. Schopf and Burgman (2006) and Sun and Zhang (2006) actually verified that the residual of the sum of the El Niño and La Niña events was rectified into the climate state. Thus, the climate state can be modified by the asymmetric ENSO through such nonlinear rectification.

4

S.-I. An, J. Choi / Atmospheric Research 94 (2009) 3–9

The amplitude of ENSO is locked to the seasonal cycle (Rasmusson and Carpenter, 1982; Galanti and Tziperman, 2000; An and Wang, 2001). Usually, El Niño (as well as La Niña) starts during the boreal spring and grows from the summer to the fall, reaching its maximum during the winter and ending its life during the next spring. Both the amplitude and the skewness of ENSO are expected to be locked to the seasonal cycle (see Fig. 1). Now suppose the residual of sum of El Niño and La Niña is rectified into the mean climate state, which is actually evident during the recent decades (Jin et al., 2003). Since the asymmetry of ENSO is locked to the seasonal cycle, the rectified amounts are different at each calendar month. Thus, the seasonal cycle would be modified via such season-dependent nonlinear rectification. On the other hand, if La Niña is just a mirror image of El Niño (i.e., symmetric ENSO), then such a symmetric ENSO could not modify the seasonal cycle because of the complete compensation between El Niño and La Niña (i.e., no residual). In this study, we examine the seasonality of the asymmetry of ENSO and the modification of the seasonal cycle due to the season-dependent nonlinear rectification. Section 2 introduces the data utilized herein. Section 3 depicts the seasonal locking of the ENSO asymmetry, while Section 4 addresses how the ENSO can modify the seasonal cycle of the tropical equatorial eastern Pacific SST. The same analysis as done in the previous section is applied to a coupled general circulation model output, and the results are documented in Section 5. Summary and discussion are presented in Section 6. 2. Data The data utilized in this study is the monthly extended reconstructed sea surface temperature dataset version 3 (ERSST. v3) from 1854–2007, improved by Smith and Reynolds (2004), with 2° by 2° horizontal resolution. These data were reconstructed by applying the optimal reanalysis method, in which marine surface observations and the satellite AVHRR data were used. Due to the relatively poor quality of the dataset prior to 1879, we focus on the period from 1880 to 2007 in this study. Data are available in the directory (ftp://eclipse.ncdc.noaa.gov/ pub/ersst/). The SODA v1.4.2 (Simple Ocean Data Assimilation) dataset is used to obtain the ocean subsurface information. SODA is a

Fig. 2. Annual cycle of skewness (solid line) and the SST tendencies by the linear (dotted line) and nonlinear (dash-dotted line) dynamical advections obtained from SODA ocean assimilation data spanning 1958 to 2001. Units for skewness and SST tendency are nondimensional and °C month− 1, respectively.

global ocean retrospective analysis (Carton et al., 2000; Carton and Giese, 2008). It is assimilated by the Parallel Ocean Program (POP) model with atmospheric forcing from the ERA-40 atmospheric analysis (Uppala et al., 2005) of the European Center for Medium-Range Weather Forecast. Observations include virtually all available hydrographic profile data, as well as ocean station data, moored temperature, salinity time series, surface temperature and salinity observations of various types, and nighttime infrared satellite SST data. Average resolution is 0.25° × 0.4°× 40-level. Actual horizontal and temporal resolutions used here are 2° ×2° and a month, respectively. Data collected span the period from 1958 to 2001. We also use simulations of a coupled general circulation model (CGCM). The atmospheric component is the third version of the ARPEGE (Action de Recherche Petite Echelle Grande Echelle)-Climat Atmospheric General Circulation Model (AGCM) developed at Me'te'o-France (Déqué et al., 1994), with T63 triangular horizontal truncation and 31 vertical levels. The ocean component is the global configuration ORCA2 of the OPA8.2 ocean general circulation model (OGCM) (Madec et al., 1998), with horizontal resolution of 2° in longitude, latitudinal resolution varying from 0.5° at the equator to 2° near the poles, and with 31 vertical levels with the highest resolution (10 m) in the upper 150 m. The concentrations of greenhouse gases are fixed at pre-industrial values and no flux corrections are applied, and thus no climate trend is appeared. Data from the last 200-year period, which is likely free of any systematic drift in the Pacific, are used for the analysis. Details of the model can be found at Cibot et al. (2005). 3. Seasonal locking of asymmetry of ENSO In order to examine the asymmetry of ENSO, we calculate the skewness of Niño-3 index (SST anomaly averaged over 5°S–5°N, 90–150°W) for each calendar month. Statistically, skewness is a way of measuring the ENSO nonlinearity (Burgers and Stephenson, 1999). The skewness is defined as the normalized third statistical moment

Fig. 1. Annual cycle of variance (dashed line; scales in the right y-axis) and skewness (solid line; scales in the left y-axis) of Niño-3 index obtained from ERSST data averaged over 1880 to 2007.

skewness =

m3 ; ðm2 Þ3 = 2

ð1Þ

S.-I. An, J. Choi / Atmospheric Research 94 (2009) 3–9

where mk is the kth moment: 

mk =

N X xi −X N i=1

k ;

P and where xi is the ith observation, X the mean, and N the number of observations. The statistical significance of skewness can be estimated from the standard error of skewness (White, 1980) if the number of independent samples in an

5

analyzed variable is known. For example, January mean SST for any year can be considered to be independent from January mean SST for other years. Thus, the total numbers of years (=127) may be a good estimation for the number of independent samples. According to White (1980), the threshold for significant skewness at the 95% confidence level is about ±1.0. Fig. 1 shows the climatological skewness of the Niño-3 index, calculated with respect to each calendar month over

Fig. 3. (a) 20-year sliding skewness of Niño-3 index for each calendar month obtained from ERSST. (b) 20-year sliding annual range of the annual cycle of skewness that is obtained from time series of (a).

Fig. 4. Wavelet spectral density of Niño-3 index obtained from ERSST.

6

S.-I. An, J. Choi / Atmospheric Research 94 (2009) 3–9

Fig. 5. Two leading PCA modes of 20-year sliding skewness. The elements of eigenvector are calendar months, and expansion coefficients represent temporal variations of the corresponding PCA modes.

the period March 1880–February 2007. The consistent positive skewness implies that the warm-event amplitude is frequently larger than the cold-event amplitude regardless of calendar month. The evolution feature of the climatological skewness of ENSO is quite similar to that of the climatological variance of ENSO, which is known to be locked to the seasonal cycle, such that a large skewness is found during the winter and a small skewness occurs during the spring and early summer, except May.

The non-Gaussian behavior of ENSO, such as the skewed probabilistic distribution of ENSO indices (i.e., no zero skewness) just described, is understandable in the context of nonlinear dynamical processes. Jin et al. (2003) and An and Jin (2004) mentioned that the El Niño–La Niña asymmetry is attributed to anomalous thermal advections by anomalous three-dimensional currents, termed ‘Nonlinear dynamical heating (NDH).’ Both NDH and the similar asymmetric damping mechanism of ENSO by the tropical instability waves play a role in inducing the asymmetry of ENSO (An, 2008). On the whole, these nonlinear processes cause the positively skewed ENSO events by intensifying the warm events (i.e., El Niño) and suppressing the cold events (i.e., La Niña). If such nonlinear processes and their influence on ENSO tend to vary season by season, then they play a part in inducing the seasonality of the ENSO asymmetry. Here, we calculate the heating tendency by NDH and the linear dynamical heating tendency in the ocean mixed layer, and the skewness of SST for the most-recent 50 years using the monthly mean SODA ocean assimilation data. (The subsurface data of SODA are not available prior to 1958.) For each calendar month the domain average over the Niño-3 region and the climatological mean are taken. An and Jin (2004) provides details on the calculation method regarding dynamical heating. As shown in Fig. 2, the seasonality of the heating tendency by NDH is pronounced, while the annual range of the linear dynamical heating is minimal. The

Fig. 6. (a) Expansion coefficients of first PCA mode of 20-year sliding skewness (solid line) and the 20-year sliding spectral density for the semi-annual band (dashed line). (b) As in (a) but for the 20-year sliding spectral density for the annual band. (c) Expansion coefficients of third PCA mode of 20-year sliding skewness (solid line) and the 20-year sliding spectral density for the semi-annual band (dashed line). (d) As in (c) but for the 20-year sliding spectral density for the annual band. Units are nondimensional.

S.-I. An, J. Choi / Atmospheric Research 94 (2009) 3–9

seasonal variations in the SST tendency induced by NDH are more or less similar to those in skewness, with large values during the fall and winter and small values during the late spring. Since the asymmetric behavior of the interannual variability of tropical Pacific SST is not totally due to the NDH (An, 2009), the seasonality of NDH cannot be exactly matched to that of skewness. Even though the local maxima and minima for the two data sets do not occur in exactly the same month, it is likely that the thermal tendency induced by NDH is related to the seasonality of the ENSO asymmetry. To depict the decadal changes in the seasonality of the asymmetry of ENSO, the 20-year sliding skewness of the monthly-mean Niño-3 index with respect to each calendar month is computed. As shown in Fig. 3, twelve time series commonly represent a significant decadal fluctuation. To measure the seasonality of the ENSO asymmetry, the yearly variance of each time series is computed (here, a year indicates March to February of the following year). To filter out short time scale fluctuations, a 20-year moving average is applied to the yearly variance (Fig. 3b). Fig. 3b represents the decadal changes in the seasonality of the asymmetry of ENSO, which is strong during 1930s, 1950s, and 1990s and weak during 1900s, 1910s, 1940s, and 1970s. 4. ENSO impact on the seasonal cycle In the previous section, we showed that the asymmetry of ENSO is locked to the seasonal cycle, and it varies decade by

7

decade. Since the residual of the sum of El Niño and La Niña could be rectified into the climate state (e.g., Jin et al., 2003; An, 2009), the seasonally varying rectified effects from the asymmetry of ENSO may modify the annual and semi-annual cycles. In that case, the decadal changes in the amplitudes of annual and semi-annual cycle are assumed to be correlated to the decadal changes in the seasonality of the asymmetry of ENSO. In order to investigate the aforementioned points, we compute the decadal changes in the amplitudes of the annual and semi-annual cycles of the Niño-3 index by using wavelet analysis. After applying wavelet analysis to the Niño-3 index, we computed the local variances for the annual band (0.8– 1.2 years) and those for the semi-annual band (0.4–0.6 years). By taking the 20-year sliding average of local variances, shorter time scale fluctuations are removed. As shown in Fig. 4, most powers of the seasonal cycle are confined to the annual band, and the powers of the semi-annual band are relatively weak. Both time series commonly show strong decadal fluctuation (see Fig. 6). In order to identify the decadal changes in the seasonality of the asymmetry of ENSO, we applied Principal Component (a.k. a. empirical orthogonal function: EOF) Analysis (PCA) to the 20year sliding skewness of the Niño-3 index to each calendar month that is shown in Fig. 2. In the calculation, the elements of the eigenvector are calendar months (i.e., January, February, March, …, and December), and the principal component elements (or expansion coefficients) are temporal variations

Fig. 7. As in Fig. 3 except for a coupled GCM.

8

S.-I. An, J. Choi / Atmospheric Research 94 (2009) 3–9

of the corresponding PCA mode. Here, the leading PCA modes represent the seasonal-cycle pattern of the ENSO asymmetry. Two leading PCA modes are shown in Fig. 5. The first and third PCAs explain 41.6% and 19.4% of the total variance, respectively. The second PCA, which explains 21% of the total variance, is not considered in this study because of its low correlation with the decadal changes in amplitude of the annual and semi-annual cycles of the tropical eastern Pacific SST. The large loading in the first PCA mode (PCA1) is found at March and April and also at June and July with an opposite sign, indicating that PCA1 represents the semi-annual variation of the ENSO asymmetry with emphasis on the spring and summer. On the other hand, the third mode (PCA3) shows strong contrast between January–February and April–May, indicating a semi-annual cycle. In the third mode, an annual-cycle (i.e., April-toSeptember vs. November-to-February) is also observed. Thus, PCA3 is a kind of mixture of annual and semi-annual cycles. The main causes for these two leading PCA modes will not be discussed here, as it is beyond our scope. The PC time series associated with the two leading PCAs are shown in Fig. 6 together with the 20-year sliding amplitude of the annual and semi-annual cycles (i.e., time series shown in Fig. 4). Fig. 6a depicts that the PC1 (i.e., the first mode of principal component) time series is well correlated with the decadal changes in the semi-annual amplitude with the correlation coefficient of 0.58, while the correlation between PC1 and the decadal changes in the annual amplitude is − 0.25. The former is statistically significant but the latter is not. Since PCA1 represents the decadal changes in the semi-annual cycle of the ENSO asymmetry, it is expected that PC1 is well correlated to the changes in the amplitude of the semi-annual cycle but not to those in the amplitude of the annual cycle. In a same manner, it is expected that PCA3 is well correlated to either the decadal changes in the annual amplitude or those in the semiannual amplitude. Fig. 6c and d show the obvious similarity between the two curves. The correlation coefficient between PC3 and the changes in the semi-annual amplitude is 0.63, and that between PC3 and the changes in the annual amplitude is 0.58. As a whole, these results indicate that the decadal changes in the amplitude of the annual and semiannual cycles may be related to the changes in the seasonality of ENSO asymmetry.

As in Fig. 5, the major modes of the decadal changes in the seasonality of the asymmetry of the model ENSO are identified by applying the PCA method. The first two leading PCA modes are shown in Fig. 8a. The first and second PCAs explain 51.0% and 17.5% of the total variance, respectively, which is also comparable to the observation. The major loading in PCA1 is found at April-to-June, and the minor loading occurs at October-to-February with the opposite sign, which seems to modify either annual or semi-annual cycles. In the second mode, a strong contrast between March-to-May (except April) and June-to-September is noted, which likely modifies the annual cycle, not the semi-annual cycle. The relationship between the decadal changes in the seasonality of the asymmetry of ENSO and those in the

5. Additional results from a coupled GCM Although we used the observed SST in the previous sections to investigate the asymmetry of ENSO and its possible impacts on the annual and semi-annual cycles, deficiencies due to relatively less reliable records in the early years of the observation might lead to concerns over the overall validity of the conclusions reached. In order to alleviate this possible doubt, in this section we analyze the coupled GCM outputs using the same approach of the previous sections. As shown in Fig. 7, the significant decadal variations in the seasonality of the asymmetry of ENSO with the comparable range of fluctuations to the observation stand up. For example, the seasonality of the asymmetry of ENSO for 150s and 170s in model year is more than eight times larger than that for 40s and 90s in model year.

Fig. 8. Fig. 8 (a) As in Fig. 5 but for a coupled GCM. (b) Expansion coefficients of first PCA mode of 20-year sliding skewness (solid line) and the 20-year sliding spectral density for the annual (dashed line) and semi-annual bands (dash-dotted line) obtained from a coupled GCM. (c) As in (b) but for the second PCA mode.

S.-I. An, J. Choi / Atmospheric Research 94 (2009) 3–9

amplitude of the annual and semi-annual cycle of the tropical eastern Pacific SST is quantified by calculating the correlation as done previously. As shown in Fig. 8b and c, PC1 is correlated to the decadal changes in the semi-annual cycle with the correlation coefficient of 0.45, while PC2 is well correlated to the decadal changes in the annual cycle with the correlation coefficient of 0.70. As ascertained in the previous sections using the observed data, it is clear even in the coupled GCM output that the seasonality of the asymmetry of ENSO is highly related to the amplitude of the annual and semi-annual cycle of the tropical eastern Pacific SST. 6. Summary and discussion In this study, we investigate the seasonality of the asymmetry of ENSO and its possible impacts on the tropical eastern Pacific seasonal cycle. As the amplitude of ENSO is locked to the seasonal cycle, the skewness of ENSO is also found to be locked to the seasonal cycle. Moreover, the seasonality of the asymmetry of ENSO has been changing decade by decade, and has intensified during the recent decades. Here we also suggest a possibility that the decadal changes in the seasonality of the asymmetry of ENSO may influence the decadal changes in the amplitude of the annual and semi-annual cycles. This is because the season-dependent nonlinear rectification can modify the annual and semi-annual cycles. The residual from the compensation between El Niño and La Niña depends on the season, and thus, the asymmetry of ENSO also varies seasonally. For example, if the SST warming during the fall of El Niño year is larger than the SST cooling during the fall of the subsequent La Niña year, the residual after the compensation between the warming and the cooling becomes the additional warming for the fall season, which results in the reduction of the annual cycle. On the other hand, when the seasonal phase in the evolution of El Niño is different from that in the evolution of La Niña, the compensation between El Niño and La Niño can be uneven. Thus, the resultant residual can be rectified into the annual or semi-annual cycle. For example, when the maximum amplitude of El Niño (warming) is recorded during the winter and the maximum amplitude of La Niña (cooling) is recorded during the fall, it will result in the intensification of the annual cycle by decreasing the fall climatology, and the weakening of the semi-annual cycle by increasing the winter climatology. In this study, however these two effects were not separately analyzed. This will be left to a further study. Acknowledgements This work is supported by the SRC program of Korea Science and Engineering Foundation, Brain Korea 21 Project, and by the Korea Research Foundation Grant funded by the Korean Government (MOEHRD, Basic Research Promotion Fund) (KRF-2007-313-C00784).

9

References An, S.-I., 2004. Interdecadal changes in the El Nino–La Nina asymmetry. Geophys. Res. Lett. 31, L23210. doi:10.1029/2004GL021699. An, S.-I., 2008. Interannual variations of the tropical ocean instability wave and ENSO. J. Clim. 21, 3680–3686. An, S.-I., 2009. A review on interdecadal changes in the nonlinearity of the El Nino-Southern Oscillation. Theor. Appl. Climatol. 97, 29–40. An, S.-I., Jin, F.-F., 2000. An Eigen analysis of the interdecadal changes in the structure and frequency of ENSO mode. Geophys. Res. Lett. 27, 2573–2576. An, S.-I., Jin, F.-F., 2004. Nonlinearity and asymmetry of ENSO. J. Clim. 17, 2399–2412. An, S.-I., Wang, B., 2000. Interdecadal change of the structure of the ENSO mode and its impact on the ENSO frequency. J. Clim. 13, 2044–2055. An, S.-I., Wang, B., 2001. Mechanisms of locking the El Nino and La Nina mature phases to boreal winter. J. Clim. 27, 2164–2176. An, S.-I., Hsieh, W.W., Jin, F.-F., 2005a. A nonlinear analysis of ENSO cycle and its interdecadal changes. J. Clim. 18, 3229–3239. An, S.-I., Ham, Y.-G., Kug, J.-S., Jin, F.-F., Kang, I.-S., 2005b. El Nino–La Nina asymmetry in the coupled model intercomparison project simulations. J. Clim. 18, 2617–2627. Burgers, G., Stephenson, D.B., 1999. The ‘normality’ of El Nino. Geophys. Res. Lett. 26, 1027–1030. Carton, J.A., Giese, B.S., 2008. A reanalysis of ocean climate using simple ocean data assimilation (SODA). Mon. Weather Rev. 136, 2999–3017. Carton, J.A., Chepurin, G., Cao, X., Giese, B., 2000. A simple ocean data assimilation analysis of the global upper ocean 1950–95. Part I: methodology. J. Phys. Oceanogr. 30, 294–309. Cibot, C., Maisonnave, E., Terray, L., Dewitte, B., 2005. Mechanisms of tropical Pacific interannual-to-decadal variability in the ARPEGE/ORCA global coupled model. Clim. Dyn. 24, 823–842. Déqué, M., Dreveton, C., Braun, A., Cariolle, D., 1994. The climate version of ARPEGE/IFS: a contribution to the French community climate modeling. Clim. Dyn. 10, 249–266. Fedorov, A.V., Philander, S.G., 2000. Is El Nino changing? Science 228, 1997–2000. Galanti, E., Tziperman, E., 2000. ENSO's phase locking to the seasonal cycle in the fast-SST, fast-wave, and mixed-mode regimes. J. Atmos. Sci. 57, 2936–2950. Hsieh, W.W., 2001. Nonlinear principal component analysis by neural networks. Tellus 53A, 599–615. Hsieh, W.W., 2004. Nonlinear multivariate and time series analysis by neural network methods. Rev. Geophys. 42, RG1003. doi:10.1029/2002RG000112. Jin, F.-F., An, S.-I., Timmermann, A., Zhao, J., 2003. Strong El Nino events and nonlinear dynamical heating. Geophys. Res. Lett. 30. doi:10.1029/ 2002GL016356. Madec, G., Delecluse, P., Imbard, M., Levy, C., 1998. OPA 8.1 ocean general circulation model reference manual. Note du Pole de modelisation. Institu Pierre-Simon Laplace. No. 11 (91 pp). Rasmusson, E.M., Carpenter, T.H., 1982. Variations in tropical sea surface temperature and surface wind fields associated with the Southern Oscillation/El Nino. Mon. Weather Rev. 110, 354–384. Schopf, P.S., Burgman, R.J., 2006. A simple mechanism for ENSO residuals and asymmetry. J. Clim. 19, 3167–3179. Smith, T.M., Reynolds, R.W., 2004. Improved extended reconstruction of SST (1854–1997). J. Clim. 17, 2466–2477. Sun, D.-Z., Zhang, T., 2006. A regulatory effect of ENSO on the time-mean thermal stratification of the equatorial upper ocean. Geophys. Res. Lett. 33, L07710. doi:10.1029/2005GL025296. Uppala, S.M., et al., 2005. The ERA-40 re-analysis. Q. J. R. Meteorol. Soc. 131, 2961–3012. Wang, B., 1995. Interdecadal changes in El Nino onset in the last four decades. J. Clim. 8, 267–285. Wang, B., An, S.-I., 2001. Why the properties of El Nino changed during the late 1970s. Geophys. Res. Lett. 26, 1027–1030. White, H.G., 1980. Skewness, kurtosis and extreme values of Northern Hemisphere geopotential heights. Mon. Weather Rev. 108, 1446–1455. Wu, A., Hsieh, W.W., 2003. Nonlinear interdecadal changes of the El NinoSouthern Oscillation. Clim. Dyn. 21, 719–730.