Analysis of observational estimates indicates that the El Niño–Southern Oscillation (ENSO) forced pattern in the extratropical Southern Hemisphere (SH), somewhat surprisingly, leads the peak phase of ENSO by one season. A Rossby wave source (RWS) analysis indicates that the tropical and extratropical RWS in the SH develops before the ENSO peak season and abruptly weakens thereafter. Further analysis shows that anomalous divergence/convergence and corresponding irrotational wind anomalies are sensitive to local seasonality. Numerical experiments in which the tropical Pacific is prescribed with perfectly periodic ENSO while all other oceans are simulated as a slab mixed layer model coupled to AGCM also show similar features. Additional numerical experiments in which ENSO forcing is shifted by 6 months (i.e., the ENSO peak in the southern winter season) indicate that the Northern Hemisphere atmosphere rather than the SH atmosphere precedes ENSO. This result supports the hypothesis that the ENSO forced pattern in the extratropics is strongly limited by local seasonality, rather than by the temporal phase of tropical remote forcing.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The impact of the El Niño–Southern Oscillation (ENSO) phenomenon is felt globally. In this study, we examine the impact of ENSO in the Southern Hemisphere (SH). Several recent studies have investigated various impacts of ENSO in the SH [e.g., Kwok and Comiso, 2002; Park et al., 2004; Ribera and Mann, 2003]. For instance, L'Heureux and Thompson  found a strong linear relationship between the Southern Annular Mode (SAM, also called the Antarctic Oscillation) and ENSO. The authors showed that the impact of ENSO includes not only thermally forced zonal wind anomalies but also eddy-driven zonal wind anomalies. Hobbs and Raphael  showed that zonal wave one of the subpolar atmospheric circulation has a link with ENSO as well as South Pacific sea surface temperature (SST). South Pacific sea temperatures are also affected by ENSO [Kidson and Renwick, 2002; Lachlan-Cope and Connolley, 2006]. These authors argued that the Rossby wave train induced by ENSO impacts sea temperature variations in the SH.
 The atmospheric variability on interannual time scales in the SH is mainly composed of the Antarctic Oscillation pattern (AAO) [Thompson and Wallace, 2000], and the Pacific–South America patterns (PSA) [Ghil and Mo, 1991]. Mo  displayed the details of these patterns based on an empirical orthogonal function (EOF) analysis of the SH winter season (June–August, JJA) 500 hPa geopotential height (GPH). The first EOF pattern is AAO, and the second and third EOF patterns correspond to PSA. Among these patterns, the first PSA pattern has been argued to be related to ENSO in several studies. For example, Karoly  examined the relationship between the first PSA pattern and ENSO based on a composite analysis of four El Niño events. When the author computed the southern winter (JJA) composite for the developing phase of ENSO, the first PSA pattern was dominant in SH. However, for the following southern summer season (December–February, DJF) when ENSO matures, the composite indicated a zonally symmetric pattern similar to AAO. On the other hand, Lau et al.  argued that the first and second PSAs, the phases of which are shifted by nearly one-quarter cycle, are independent of ENSO. The authors separated PSA patterns from pentad stream function data, and showed that the time series of those patterns have no simultaneous relationship with tropical convection. In summary, both Karoly  and Lau et al.  examined the same relationship between ENSO and PSA, but reached very different conclusions. Both studies examined the simultaneous relationship, but in different seasons, SH winter versus summer (i.e., developing phase versus peak phase of ENSO).
 In addition to the simultaneous relationship, lag-lead relationship between the SH atmosphere and ENSO has been examined. For example, Trenberth and Shea  argued that the South Pacific leads the Southern Oscillation (SO). The authors found this relationship by the lag-lead correlation between the SO and surface pressure station data. The authors also argued possible physical process for the South Pacific to effect the SO development, but this process remains unclear.
 In order to confirm the lag-lead relationship between ENSO and the SH atmosphere, a lag-lead maximum covariance analysis (LLMCA, details in section 2) of 500 hPa GPH in the SH and the tropical Pacific (TP) SST is presented here. The domain of 500 hPa GPH is defined as from 0° to 360°, 87.5°S to 25°S. The domain of SST is defined as from 160°E to 90°W, 15°S to 15°N, so no overlap is allowed. Figure 1a shows the lag-lead correlation coefficient and covariance. The maximum coefficient, 0.475, occurs when the SH GPH leads TP SST by two months. Around the peak month (−3 to −1 in Figure 1a), all of the first corresponding patterns are absolutely dominant (>90%), and spatial patterns are similar to each other. Figure 1b shows the corresponding spatial pattern at the time of maximum correlation. The TP SST clearly shows an ENSO pattern. The SH GPH pattern at 500 hPa is similar to the first PSA pattern shown in Figure 1b of Mo , but the location of midlatitude low is shifted 30° eastward.
 The key question of this study is why the PSA pattern in the SH atmosphere precedes the peak of tropical eastern Pacific SST anomaly (SSTA). The PSA patterns are wave train patterns from the subtropics to high latitudes, and therefore, they are thought of as the ENSO forced patterns [e.g., Karoly, 1989; Mo, 2000]. Karoly [1989, Figure 11a] shows schematic illustration how enhanced tropical convection is related to the PSA patterns in the early stage of ENSO (JJA). However, during the mature phase of ENSO (DJF), the PSA patterns in the SH are not dominant [Ciasto and Thompson, 2008; Garreaud and Battisti, 1999; Harangozo, 2004]. Does the SH really lead the ENSO as argued by Trenberth and Shea ? Or does the other mechanism exist to explain this phenomenon? We examine possible answers with observational estimates as well as a sequence of idealized numerical experiments. In section 2, the data and model experiments are described. The analysis of observational estimates and numerical experiments are discussed in section 3 and section 4, respectively, and the conclusions are given in section 5.
2. Data and Model
 The atmospheric variables used in this study are the 500 hPa GPH and 200 hPa winds from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis data set [Kalnay et al., 1996]. The horizontal resolution is 2.5° latitude by 2.5° longitude. The SST data is from the Hadley center SST data set [Rayner et al., 2000]. The original resolution of the SST data is 1° by 1°, but it is reduced to T42 resolution for numerical convenience. The period of both data sets used in this study is from January 1949 to December 2004. In this study, all anomalies are calculated by removing 20 year running mean annual cycle. This approach is used to remove very low frequency variability (i.e., trend).
 In order to investigate the lag-lead relationship between the SH atmosphere and the TP SST, we used LLMCA (lag-lead maximum covariance analysis). The LLMCA is based on a singular value decomposition (SVD) of the covariance matrix of two variables at a given lag month [see Czaja and Frankignoul, 2002]. Before applying LLMCA, monthly anomalies of GPH are weighted by the square root of the cosine of latitude. This ensures that equal areas are given equal weight in the analysis.
 The monthly evolution of the impact of ENSO is revealed by the ENSO composite analysis. The selection criterion for the composite is based on the NINO3.4 index (area average of SST for 5°S∼5°N, 170°W∼120°W) so that the top 9 January NINO3.4 values define the warm event composites and the bottom 9 January NINO3.4 values define the cold event composites. We have excluded the 1968–69, 1986–87, and 1987–88 events even though they meet our composite criterion. The reason for this is because these events have significantly different evolution (i.e., they do not peak in boreal winter). See Table 1 for a list of events contributing to the composites.
Table 1. Selected Years for ENSO Composite Analysisa
For the ENSO composite analysis, nine El Niño and La Niña events are selected based on Jan (1) NINO3.4. The years during which the peak is not located in boreal winter are excluded.
 For the numerical experiments we have used the National Center for Atmospheric Research (NCAR) Community Atmosphere Model Version 3 (CAM3) with a horizontal resolution of T42 and 26 vertical levels with a hybrid terrain-following coordinate. The dynamical core is Eulerian spectral and originated from NCAR CCM3 [Kiehl et al., 1998], and the changes are described by Collins et al. . The ocean model is coupled to the atmospheric general circulation model (AGCM) using a simple thermodynamic slab mixed-layer model, except for the tropical Pacific region as described below. The depth of the mixed layer is fixed to the climatological annual cycle with a 200 m cap, which is part of the CCSM modeling framework.
 The tropical Pacific region is prescribed with perfectly periodic ENSO in this study. The prescribed region is defined as 120°E to 85°W in longitude and 10°S to 10°N (M10 hereafter) or 20°S to 20°N (M20 hereafter) in latitude. We apply the two different forcing regions because we have found that the global ENSO response is sensitive to the meridional extent of tropical forcing region [Jin and Kirtman, 2009]. The prescribed SST data is composed of the climatological annual cycle (the control experiment), and El Niño or La Niña composite anomalies (the ENSO experiment), which are derived from observational estimates. In the ENSO experiment, the El Niño and La Niña anomalies are added to the climatological annual cycle with fixed period. In this study, we set the ENSO evolution to a 4 year period. This experimental design is used to reduce the uncertainties of the ENSO impacts. Because the only difference between the control and ENSO experiment is the existence of ENSO SSTA, we can assert that the ENSO experiment composite anomaly from the control experiment climatological annual cycle is the response to the ENSO SSTA. Another merit of this perfectly periodic ENSO strategy is that we can cleanly construct a “seasonally shifted” ENSO. When the ENSO anomalies are added to the climatological annual cycle, if we shift the ENSO anomalies by 6 months, we can perform a similar numerical experiment but with ENSO events that peak in boreal summer. The experiments are summarized in Table 2. Finally, a notational remark: “JAN1” or “JAN(1)” means “January of year 1.”
A set of experiments is described. The longitudinal definition of prescribed region is commonly from 120°E to 85°W. The period of ENSO is set to four year (P4).
Annual cycle only (control)
3. Observational El Niño–Southern Oscillation (ENSO) Composite Analysis
 As mentioned in the introduction, the SH atmospheric pattern most related to ENSO precedes tropical eastern Pacific SSTA by two months. In order to examine the temporal evolution of GPH in the SH, and to compare it with that in the Northern Hemisphere (NH), time-latitude cross section of the standardized ENSO composite GPH at 500 hPa is shown in Figure 2. A zonal average is taken from 180 to 120°W. The choice for this regional average is based on the SH midlatitude negative anomaly pattern shown in Figure 1b (left), so this may not be the best for the NH GPH patterns. For comparison of the temporal phase of tropical Pacific SSTA, ENSO composites of NINO3.4 index are also shown in Figure 2. During the El Niño (Figure 2a) and La Niña (Figure 2b) events, most extratropical signals are in common and can be detected between 25° to 45° in the both hemispheres. However, the temporal phase when a strong signal appears is clearly different between the hemispheres. In the NH, the midlatitude signal is in phase with tropical atmosphere which directly responds to ENSO SSTA. In the SH, strong signals appear before the ENSO peak season in the both El Niño and La Niña composites. In the southern summer season when ENSO matures, nearly no signals are detected in the SH.
 The temporal evolution of the zonal mean atmosphere indicates that strong signals in the South Pacific appear in OCT(0), a few months before the El Niño and La Niña peak phase. A detailed spatial structure of the ENSO composite GPH in OCT(0) is compared to that in JAN(1), the ENSO peak month in Figure 3. Figure 3 shows the SH ENSO composite of GPH anomalies at 500 hPa (contoured) and Rossby wave source (RWS) anomalies at 200 hPa (shaded). The RWS is calculated following Sardeshmukh and Hoskins :
where η is the absolute vorticity and Vχ is the divergent component of the wind [see also Kirtman et al., 2001; Lachlan-Cope and Connolley, 2006]. It is known that the ENSO forced pattern in the extratropical atmosphere can be described as a forced Rossby wave train [Trenberth et al., 1998, and references therein]. Hence, if the SH atmospheric signal preceding ENSO is an ENSO forced pattern, RWS should be temporally and spatially consistent with the GPH at 500 hPa.
 In Figure 3, the RWS is also stronger before the ENSO peak. Particularly in the El Niño composite, the RWS anomalies as well as GPH anomalies are much stronger in OCT(0) than in JAN(1). In the La Niña composite, the differences between the top and bottom of Figure 3 are less pronounced than the El Niño composite. However, it is still clearly shown that midlatitude signals of GPH and RWS anomalies shrink during the La Niña peak. It is also interesting that the wave train in the El Niño events seems to start from the region 180° to 150°W, while, in the La Niña composite, the wave train seems to start from the region 150°W to 120°W.
 The RWS anomaly composite in the SH becomes weaker at the ENSO peak, consistent with GPH results. This suggests that the SH atmosphere ENSO composite pattern which matures before the ENSO peak season is likely a teleconnection pattern in the form of Rossby wave train. This idea leads us to hypothesize that local seasonality impacts the RWS. We investigate this possibility further below.
 The RWS (equation (1)) can be decomposed into several physical processes. The linearized RWS anomaly is written as
where ζ is the relative vorticity, f is the Coriolis parameter, and (–) and (′) represent the climatological mean and perturbation, respectively [Sardeshmukh and Hoskins, 1988]. Qin and Robinson  categorized those terms into tropical and extratropical components. For example, the sum of first and fourth parts of right-hand side is of the tropical RWS, and the sum of second and third parts of right-hand side is of the extratropical RWS.
 On the basis of Figure 3, we expect that the temporal evolutions of tropical and extratropical RWS during ENSO events are also asymmetric between the NH and SH as is the GPH at 500 hPa (Figure 2). Figure 4 shows the ENSO composite zonal mean tropical and extratropical RWS anomalies in the time-latitude cross section, similar to Figure 2. It is important to note that the El Niño or La Niña composite of RWS anomalies are zonally root-mean-squared in the domain, 135°E∼120°W. This calculation is performed because, as shown in Figure 3, opposite sign of RWS anomalies exist zonally, so arithmetic average can cancel out the anomalies. The zonal domain of calculation is based on the results seen in Figure 3. In addition, the ENSO composite NINO3.4 anomalies are shown as with Figure 2.
 As argued by Qin and Robinson , tropical RWS is prominent feature in the lower latitudes (i.e., less than 30°; Figure 4, contour), and extratropical RWS is apparent in the higher latitudes (i.e., greater than 30°; Figure 4, shaded). During ENSO, the temporal phase of the tropical and extratropical RWS is clearly different in the NH and SH. In the SH, the RWS is strongest one season before the ENSO peak season, while in the NH, particularly in the El Niño composite, the RWS peak follows the El Niño peak. In the La Niña composite, the tropical and extratropical RWS in the NH does not form a clear peak like the El Niño composite. It is also interesting that, in the SH, both RWS anomalies are detected again after the ENSO peak season.
 Consistent with previous GPH results, Figure 4 indicates that the local peaks of the tropical and extratropical RWS in the NH and SH are in JAN(1) and OCT(0), respectively. Thus we examine the spatial structure of the tropical and extratropical RWS in these months in Figure 5. Here we focus on the El Niño composite because the temporal asymmetry was more evident in the El Niño composite rather than the La Niña composite.
 The spatial pattern of tropical and extratropical RWS in the NH and SH between in OCT(0) and JAN(1) can be compared in Figures 5a and 5c. Tropical Pacific SSTA is also shown for the reference (Figure 5b). Between OCT(0) and JAN(1), the SSTA pattern in the tropical Pacific is similar in terms of the pattern. In OCT(0), SSTA greater than 1.5°C appears although this is not as zonally consistent as that in JAN(1). However, differences of RWS are prominent between OCT(0) and JAN(1). For example, tropical RWS in the NH is barely detected in OCT(0), but forms several peaks in JAN(1). In the SH, both the tropical and extratropical RWS clearly appear in OCT(0), but are too weak to be detected in JAN(1).
 On the other hand, the sign of tropical RWS anomaly is generally opposite to the sign of extratropical RWS anomaly in nearly the same location. For example, in Figure 5a (right), the central and eastern negative anomaly of the tropical RWS (contoured) are paired with the positive anomaly of extratropical RWS (shaded) just north of tropical RWS. In the SH, Figure 5c (left) indicates that the western positive anomaly of the tropical RWS (contoured) is paired with the negative anomaly of the extratropical RWS (shaded) just south of tropical RWS. This meridionally adjacent but opposite sign of the RWS can result in reduced magnitude of total RWS by canceling each other.
Figure 5 indicates that the midlatitude RWS in the NH becomes abruptly stronger, and that in the SH it becomes abruptly weaker in JAN(1) compared to OCT(0). Qin and Robinson  argued that the tropical RWS is dominated by the first term on the right in equation (2). The physical interpretation of this term is the advection of climatological absolute vorticity by the perturbed divergent (irrotational) wind [Sardeshmukh and Hoskins, 1988]. Similarly, Qin and Robinson  also argued that the extratropical RWS is dominated by the second term on the right in equation (2): vortex stretching. Hence, we need to examine the mean absolute vorticity and irrotational wind in OCT(0) and JAN(1) for further diagnosis.
 The advection term dominating the tropical RWS is an inner product of irrotational wind anomaly and the gradient of climatological absolute vorticity. In Figures 6a and 6b, El Niño composite of irrotational wind anomaly and climatological gradient of absolute vorticity at 200 hPa are shown in vector form, respectively. Because of the Coriolis parameter, the absolute vorticity generally increase from the South Pole to the North Pole. Hence the meridional component of the gradient is much stronger than the zonal component of the gradient. This is why we have chosen the meridional magnitude of vector as a criterion to display wind anomaly and gradient.
 The irrotational wind anomalies are different between OCT(0) and JAN(1) (Figure 6a, left and right). In OCT(0), strong wind anomalies appear only in the SH: subtropical region from the eastern Australia to the central South Pacific. The strong wind anomalies in this region are multiplied by the large gradient of absolute vorticity in the region of western-to-central South Pacific (30°S to 15°S) to form a positive tropical RWS signal shown in Figure 5c (left, contoured). At the same time, wind anomalies are not detected in the NH although the gradient of climatological absolute vorticity is broadly detected there.
 In JAN(1), strong wind anomalies are much more detectable in the NH. Because this is in the ENSO peak season, strong divergence appears in the western to central equatorial Pacific. From this region, irrotational meridional wind anomalies are directed to both north and south. However, these meridional wind anomalies are not hemispherically symmetric. The wind flow going north is stronger than the flow going south, and there is corresponding stronger convergence in the NH than in the SH (Figure 6a, right, blue shading). The strong wind anomalies in the NH is multiplied by the strong gradient of absolute vorticity in the central North Pacific around 30°N (Figure 6b, right) to form a negative tropical RWS. The wind anomaly in the southern USA also contributes to another negative tropical RWS (Figure 5a, right). A few wind anomalies are detected in the western South Pacific region, but these anomalies only make a small contribution to the tropical RWS signals in the SH. This is because this region little overlaps with the large gradient of absolute vorticity in the western South Pacific.
 The differences of extratropical RWS between OCT(0) and JAN(1) is also mostly contributed by the irrotational wind component, similar to the tropical RWS. The vortex stretching term is composed of the climatological mean absolute vorticity multiplied by divergence or convergence of irrotational wind anomalies. In Figure 6b, the general pattern of the mean absolute vorticity which is positive in the NH and negative in the SH is unchanged from OCT(0) to JAN(1) although the details are different. In contrast, the divergence or convergence pattern of irrotational wind shown in Figure 6a is different between OCT(0) and JAN(1). For example, the strong convergence in the North Pacific midlatitude region abruptly appears in JAN(1) compared to in OCT(0). Weakened convergence (blue shading) east of Australia in JAN(1) is also apparent.
 In Figure 6a, divergence or convergence anomalies in the midlatitudes abruptly changed from OCT(0) to JAN(1). In contrast, tropical divergence or convergence anomalies gradually increased, in phase with ENSO. In other words, the phase relationships with ENSO in the midlatitudes changed, but remain constant in the tropics. This is consistent with the tropical GPH ENSO composite shown in Figure 2. In Figure 2, tropical GPH responses last longer, and evolve slowly relative to the extratropical responses. These suggest that anomalous divergence (convergence) in response to tropical Pacific El Niño (La Niña) SSTA is less sensitive to seasonality. The seasonality seems to have the largest effects on subtropical to midlatitude irrotational wind.
 The seasonal change of the global atmosphere is documented in several previous studies. For example, the seasonal change of atmospheric meridional circulation is described by Oort and Rasmusson , Oort and Yienger , and Trenberth et al. , etc. Oort and Yienger  showed the annual cycle of the zonal mean mass-stream function where a counterclockwise circulation expands from 20°S into tropics in the southern winter season. This circulation contributes to convergence in the southern subtropics, and persists until October. After November, this southern counterclockwise cell collapses, and in January, cyclonic circulation expands from the NH and settles in the southern subtropics. At that time, the contour density of mass-stream function in high-altitude southern subtropics becomes thin, so it is hard to expect large irrotational wind anomaly in this region on January. Furthermore, the seasonal change of upper atmosphere velocity potential is described by Mo and Rasmusson  and Tanaka et al. , and is consistent with the seasonal change of the Hadley cell. Tanaka et al.  showed horizontal patterns of 200 hPa velocity potential monthly climatology. In July, a strong gradient of velocity potential is detected in the southern subtropics between 90°E and 180°, and is associated with strong southward divergent wind. This pattern remains through October although somewhat weaker. The region of large southward divergent wind overlaps with the southward irrotational wind anomaly in Figure 6a (left). In contrast, in January, a strong velocity potential gradient exists mostly in the northern subtropics associated with strong northward divergent wind. In the southwest Pacific and Australia, climatological divergent wind in January is much less than in October. To summarize, the pattern of climatological global atmosphere looks similar between in July and October although the magnitude is weaker in October. After October, the pattern of global atmosphere changes, so in January, the circulation has a different pattern than either July or October.
 Previously we noted that the tropical and extratropical RWS exist in similar locations with opposite sign. Figure 6 suggests why this phenomenon occurs. For example, around the region 120°W∼90°W and 30°N, northward meridional wind anomalies are directed toward the convergence zone (Figure 6a, right). In the same region, a large meridional gradient of absolute vorticity is seen in Figure 6b (right). As a result, both northward wind anomalies and vorticity gradient form a negative tropical RWS, and convergence and positive vorticity form a positive extratropical RWS. The same phenomenon, but a positive tropical RWS and negative extratropical RWS is also seen in the SH in OCT(0). Briefly, (1) if convergence (or divergence) exists in the midlatitudes, (2) if corresponding poleward (or equatorward) irrotational wind anomalies exist slightly equatorward of the convergence (or divergence) zone, and (3) if an absolute vorticity gradient exists in the same region as the wind anomalies, then the tropical and extratropical RWS occurs with opposite sign.
 Summarizing all, the analysis of observational estimates is consistent with the hypothesis: (1) the pattern preceding ENSO in the SH is the ENSO forced pattern (i.e., the SH variability does not cause ENSO), and (2) the ENSO forced pattern abruptly weakens during the ENSO peak season because of local seasonality. The RWS component analysis in the El Niño composite reveals that the irrotational wind anomalies are largely responsible for the relatively large change from OCT(0) to JAN(1). In the next section, this hypothesis is tested with numerical experiments.
4. Numerical Experiment to Test the Hypothesis
 In this section, a series of numerical experiments are analyzed to test the hypothesis that the ENSO forced pattern in the SH abruptly weakens in the ENSO peak season because of local seasonality. The experimental design is akin to the so-called tropical ocean–global atmosphere (TOGA) approach as described in section 2. In this framework, the prescribed SST in the tropical Pacific region forces global climate. We note again that all ENSO composite anomalies in this section are defined by subtracting the control experiment climatological annual cycle from the ENSO experiment composite (e.g., M10 DJFpk composite – M10C annual cycle).
4.1. Boreal Winter ENSO Peak Experiment
 Here the numerical experiment for DJF peak ENSO forcing is analyzed. This experimental design simulates the impact of ENSO which matures in the boreal winter season, as in nature. Because this experimental design does not allow any forcing of ENSO, if a similar atmospheric pattern in the SH precedes ENSO in this simulation, we argue that the atmospheric pattern in the SH is an ENSO forced pattern, and does not cause ENSO.
Figure 7 shows the temporal evolution of the zonally root-mean-squared tropical and extratropical RWS in the M10 and M20 DJFpk experiments in the same domain (i.e., 135°E∼120°W). The ENSO composite of the NINO3.4 anomaly is also shown, similar to Figure 4. In Figure 7, both the M10 and M20 DJFpk experiments simulate the RWS in the South Pacific basin preceding ENSO. In the both experiments, the tropical and extratropical RWS signals in the SH are detected from JUL(0) to OCT(0), which are one or two season(s) earlier than those in the NH. In the NH, particularly the extratropical RWS is in phase with ENSO. In the case of La Niña, both RWS in the SH also precedes ENSO, which is clearer in the M20 DJFpk experiment.
 In addition, it is worth noting that the ENSO composite RWS signals are also detected between ENSO events, e.g., around JAN2 in the NH (Figure 7b). Including this RWS signal, most of the RWS signals peak in the same month, particularly in the NH. For example, every December is a critical month in the M10 DJFpk experiment, and February is in the M20 DJFpk experiment. This result suggests that the tropical remote forcing enhances the extratropical local variability, but hardly shifts the temporal phase of local variability.
 The spatial pattern of the GPH and the RWS in the SH which precedes ENSO is displayed in Figure 8. Here, the El Niño composites of both M10 and M20 DJFpk experiments are compared between OCT(0) and JAN(1) as in Figure 3a. Consistent with Figure 7, the GPH anomaly composites in the both experiments show teleconnection patterns in OCT(0) rather than in JAN(1). However, the teleconnection patterns are different between the M10 and M20 DJFpk experiments. In the M10 DJFpk experiment, the wave train in 150°E∼180° is dominant, but, in the M20 DJFpk experiment, the wave train is dominant in 150°W∼120°W. It is also interesting that the teleconnection pattern in the M10 DJFpk experiment is much stronger than that in the M20 DJFpk experiment in the SH. However, the relatively stronger signal in the M10 DJFpk experiment is actually much weaker than the observational estimates (the contour level of GPH in Figure 8 is one-third of that in Figure 3 (i.e., 5 m versus 15 m). In the NH, the GPH ENSO response in the M20 DJFpk is stronger than that in the M10 DJFpk experiment (not shown), which is consistent with the RWS results shown in Figure 7, and the anomaly magnitude in the M20 DJFpk experiment is similar to the observational estimates [Jin and Kirtman, 2009]. The reason for the weak ENSO responses in the SH remains unclear. In summary, the both M10 and M20 DJFpk experiments indicate that the SH extratropical ENSO responses leading ENSO although the details of ENSO responses are different from each other.
4.2. Numerical Experiment of Seasonally Shifted ENSO
 We have performed numerical experiments where the prescribed tropical Pacific ENSO cycle mimics observations in the sense that the ENSO events peak in DJF (i.e., DJFpk ENSO experiment). The results indicate that the SH atmospheric patterns also abruptly disappear in the southern summer season when ENSO matures, as with the observational estimates. We suggest that local seasonality is most likely the reason for different responses in the NH and SH. In order to test the effect of local seasonality on the ENSO forced pattern, we performed a numerical experiment where ENSO matures in the boreal summer (or southern winter) season. If local seasonality plays an important role to constructively or destructively interfere with the ENSO forced pattern, we expect that the NH atmosphere rather than the SH will precede ENSO.
Figure 9 shows the temporal evolution of the zonally root-mean-squared tropical and extratropical RWS ENSO composite, in the same format as Figure 7, for the M10 and M20 JJApk experiments. As mentioned previously, the ENSO SSTA evolution is same between the DJFpk and JJApk experiments except for the 6 month shifted temporal phase. The shifted temporal phase of ENSO results in the NH atmosphere leading ENSO. In Figure 9, the both JJApk experiments simulate the NH RWS signals preceding the ENSO peak by up to 6 months. Contrary to the NH, most RWS signals in the SH are in phase with NINO3.4 index in both El Niño and La Niña. The temporal relationship between the SH and ENSO in the JJApk experiments is nearly same to the relationship between the NH and ENSO in the DJFpk experiments (Figure 9 versus Figure 7).
 On the other hand, there are a few similarities between the DJFpk and JJApk experiments. For example, the RWS in JAN(2) around 30°N is detected in the M20 experiments (Figures 7b and 9b). It is also common that the tropical RWS signal is detected further poleward in the NH than in the SH. These phenomena are viewed as independent of local seasonality.
 The mid to high-latitude NH atmospheric response preceding ENSO is most easily detected in the M20 JJApk experiment in the case of El Niño (Figure 9, right) around JAN(1). In order to compare the spatial pattern in the NH and SH when the ENSO forced pattern is the strongest, we displayed the M20 JJApk experiment El Niño composite in the NH and SH, FEB(1) versus JUL(1) in Figure 10. The format of Figure 10 is similar to that of Figure 5, but, in the NH and SH, the contour indicates GPH anomalies at 500 hPa, and shading indicates the total RWS anomalies.
 Comparing the left column with the right column in Figure 10, the tropical Pacific SSTA in FEB(1) is not as developed as that in JUL(1). In the SH, it is hard to find a robust teleconnection pattern from the tropical Pacific to the extratropics in FEB(1). However, in the NH, the magnitudes of the RWS anomalies are large and GPH signals indicate clear wave train patterns in FEB(1). In JUL(1), ENSO matures and the SH atmospheric signals become stronger, but the wave train in the NH disappears.
 In the SH, compared to the strong ENSO signals in the DJFpk experiments (Figure 8, top), the wave train pattern in the M20 JJApk experiment (Figure 10c, right) is stronger (midlatitudes negative peaks are <25 m in Figure 8a (top) versus <40 m in Figure 10c (right). The details of the pattern are also different. For example, the high-latitude positive anomalies in the M20 JJApk experiment are not circular in structure. The location of the midlatitude negative anomaly is also different between Figure 8 and Figure 10. The pattern of the RWS is also different. Between the (M10 or M20) DJFpk and JJApk experiments, the tropical ENSO forcing (i.e., ENSO SSTA) is same except for the 6 month shift, but the temporal evolution and details of the response patterns are different.
 Last, we examined the evolution of tropical divergence/convergence in response to ENSO forcing in the numerical model results. Previously we noted that tropical divergence/convergence in observational estimates seems to be less sensitive to seasonality. Here we test this idea by comparing the DJFpk and JJApk experiments. Figure 11 shows divergence at 193 hPa and pressure velocity at 510 hPa in the numerical experiments. For the latitude-time cross section, each variable is averaged meriodionally in 165°E to 165°W1. Between the M10 and M20 DJFpk experiments (Figures 11a and 11b), ENSO composite patterns are different in the details, but the timing of tropical response is similar. For example, during El Niño, strong upward and divergent flow on the equator commonly occurs just after the El Niño peak (around FEB(1)). The M10 and M20 JJApk experiments (Figures 11c and 11d) also show similar timing of ENSO response, e.g., strong divergence is detected just before the El Niño peak (around JUN(1)).
 Comparing the DJFpk and JJApk experiments, focusing on the time of peak magnitude of ENSO, we argue that tropical upward and divergent flows are less affected by seasonality, and follow the phase of the SST forcing (FEB(1) versus JUN(1)). However, we also can argue that the tropical upward and divergent flows are, in contrast to the previous argument, sensitive to seasonality because red (blue) shading during El Niño (La Niña) commonly appears in the boreal spring season in the both DJFpk and JJApk experiments. In addition, in the JJApk experiments (Figures 11c and 11d), tropical divergence, which is actually located a few degrees north of equator, abruptly ends after JUL(1) at the same time as northern subtropical convergence disappears. This is more evidence supporting the effect of seasonality. Tropical anomalous divergence/convergence is not entirely determined by seasonality as with subtropical to midlatitudes RWS. Nevertheless, there is some impact of seasonality that can be detected in the tropics.
5. Summary and Conclusion
 The atmosphere in the SH leads tropical Pacific SSTA by one season. This phenomenon is supported by both LLMCA and ENSO composite analysis of observational estimates. The GPH pattern preceding ENSO is a Rossby wave train response, and the RWS analysis also supports that the SH atmosphere precedes ENSO. The RWS decomposition indicates that both the tropical and extratropical RWS in the SH abruptly weakens during the ENSO peak season. Further analysis reveals that anomalous divergence/convergence and corresponding irrotational wind anomaly, which strongly depend on local seasonality, are largely responsible for the abrupt weakening of RWS in the SH summer season when ENSO matures.
 During ENSO, tropical Pacific SSTA smoothly develops and decays over almost 1 year. The prominent ENSO forced pattern in the NH appears in the ENSO peak season. However, this pattern does not evolve gradually like ENSO. The midlatitude wave train and RWS signals are difficult to detect one season before the ENSO peak season. The abruptly developing atmospheric signals in the NH and abruptly decaying signals in the SH suggest the hypothesis that the ENSO forced pattern in the extratropics is modulated by local seasonality.
 A series of numerical experiments were performed to examine the impact of local seasonality. First, we performed a TOGA-style ENSO experiment where the El Niño/La Niña life cycle mimics observations. This experiments reproduces the observational result that the SH atmosphere leads ENSO. Second, using the same experimental design but with the ENSO peak shifted by 6 months (i.e., JJApk experiment) we found that the NH atmosphere leads ENSO. Here, the strong SH atmospheric signals are in phase with tropical ENSO forcing, which supports our hypothesis that local seasonality strongly modulates the extratropical ENSO responses. Further analysis suggests that seasonality also affects tropical atmosphere although the modulation effect is weaker than in the extratropics.
 The numerical experiments used in this study do not perfectly reproduce nature. For example, the simulated wave train patterns in the NH and SH are different from those in nature in terms of magnitude, location, and pattern. Why the simulation of M10 experiment is different from that in the M20 experiments in the SH is also not resolved and remains for further research. However, the results of these experiments do support the conclusion that the tropical remote forcing can amplify the extratropical local variability, but does not change the temporal phase of local variability. The ENSO forced pattern is mostly limited by local seasonality.
 D.J. would like to thank David Straus for providing analysis tools. B.P.K. acknowledges support from NOAA grant NA17RJ1226 and NSF grants OCI0749165 and ATM0754341.