Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
 A relationship between the timing of El Niño onset and the subsequent evolution is examined, using 130-year long time series of Niño-3.4 index from 1871 to 2000. It is found that El Niño events can be classified into two major types: one is the onset of which is from April to June (spring type) and the other is from July to October (summer-fall type). Here, the duration of El Niño is defined as the period when the 5-month running mean anomaly of Niño-3.4 index is exceeding 0.5°C. As a result, 24 El Niño events are identified, and classified into 10 spring type events and 14 summer-fall type events. In general, spring type events grow greater in magnitude, and take the mature phase around a boreal winter and the evolution is relatively regular. On the contrary, summer-fall type events are relatively weaker in magnitude, and have rather irregular aspects.
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 interaction between a seasonal cycle and El Niño/Southern Oscillation (ENSO) event is one of the key issues in El Niño dynamics. Although it is widely known that the mature phase (peak in SST anomalies in the central to eastern equatorial Pacific) of El Niño event is firmly locked to a boreal winter (so-called phase-locking, e.g., Rasmusson and Carpenter ), El Niño events also have the aspects of irregularities: timing of the onset, magnitude of SST anomalies, and duration of events [e.g., Neelin et al., 2000].
 The linearity of phase-locking has been investigated by many researchers [e.g., Xie, 1995; Tziperman et al., 1997, 1998], as well as the nonlinearity [e.g., Jin et al., 1994]. Using numerical model, Neelin et al.  found that El Niño event, the onset of which is earlier than normal, had more irregular aspects. Xu and Chan  (hereinafter referred to as XC) focused on the timing of onset, and classified El Niño events into two types: one is spring type and the other is summer type. The above two studies strongly suggest that the evolution of El Niño event differs considerably depending on the timing of onset.
 So far, several studies grouped past El Niño events into several categories, for instance, based on magnitude of SST anomalies. The authors believe that this kind of classification would be important to elucidate El Niño dynamics. The purpose of this study is to explore the relationship between the timing of El Niño onset and the subsequent evolution, using long-term observational data.
2. Data and Definition of El Niño
 The present study uses Niño-3.4 index (SST anomalies in the region of 5°S–5°N, 170°W–120°W), in order to define El Niño events properly [Trenberth, 1997]. Niño-3.4 region has relatively high SSTs, and Niño-3.4 index is highly correlated with the Southern Oscillation index (about 0.9 in case of 5-month running mean). Further, Niño-3.4 includes the region where equatorial west-east SST gradient is large through a year [Wang, 1995]. Therefore, it is considered that the Niño-3.4 index is a meaningful indicator for detection of an El Niño event.
 We use 130-year long time series of monthly Niño-3.4 index from 1871 to 2000, prepared by Trenberth and Stepaniak . They made the index using a combination of two SST datasets. The first (1981–2000) is based on the dataset using optimum interpolated SST field prepared by the NCEP-CPC (National Centers for Environmental Prediction/Climate Prediction Center [Reynolds and Smith, 1994]). Another one (1871–1980) is based on the HadISST (Hadley Center Sea Ice and Sea Surface Temperature) dataset. HadISST is the improved version of GISST dataset (Global Sea-Ice and Sea Surface Temperature) [Hurrell and Trenberth, 1999], in which missing values are interpolated by means of an EOF (Empirical Orthogonal Function) analysis. The Niño-3.4 index was extracted from the homepage of Climate and Global Dynamics Division/National Center for Atmospheric Research (http://www.cgd.ucar.edu/cas/catalog/climind/TNI_N34).
 We define an El Niño event, using the 5-month running mean of Niño-3.4 index demeaned using 40-year average from 1961 to 2000. The procedure of 5-month running mean is done in order to remove the intraseasonal variation due to the Madden-Julian Oscillation in the tropical Pacific [e.g., Madden and Julian, 1971]. El Niño is defined as events when low-passed Niño-3.4 index exceeds 0.5°C for six consecutive months or longer. The month when the Niño-3.4 index exceeds (falls) 0.5°C is defined as the onset (termination) month. Although the value of 0.5°C is different from that of Trenberth , it is identical to the threshold used by NOAA (National Oceanic and Atmospheric Administration) and that of XC. By this definition, 24 El Niño events are identified during the analysis period of 130 years.
 In order to confirm the robustness of the present result, we compared the result using the Cold Tongue Index (CTI: SST anomalies in the central-east equatorial region of 6°S–6°N, 180°–90°W: Deser and Wallace ). Monthly value of CTI (1849–2000) is extracted from the homepage of the JISAO (Joint Institute for the Study of the Atmosphere and the Ocean), University of Washington. It is found that Niño-3.4 index and CTI are highly correlated (about 0.9, in case of 5-month running mean), when CTI leads Niño-3.4 by 1–2 months. Naturally, these 24 El Niño events were also confirmed by the CTI. Further, we compared Niño-3.4 index and CTI from the viewpoint of the relative magnitude of peak, timing of warming, and duration of warming event and got the similar result (not shown here).
Figure 1a shows the frequency distribution of month when Niño-3.4 index takes a peak, i.e., month of the mature phase of each El Niño event. This diagram clearly shows that all 24 El Niño events in the past 130 years are phase-locked to a boreal winter. This result is consistent with the finding of previous studies [e.g., Rasmusson and Carpenter, 1982]. On the other hand, the months of El Niño onset widely distribute ranging from February to December, although most of them are mainly from April to October (Figure 1b). These features appear in not only the whole time series of 130 years, but also recent 50 years (not shown here).
 In order to explore the relationship between El Niño onset and subsequent evolution, we start with simple classification of El Niño events into two types depending on onset timing. Here we use XC's classification with some modification. Considering the distribution shown in Figure 1b, we eventually divide El Niño events as follows: spring type (10 events) and summer-fall type (14 events), the onsets of which are from February to June and from July to December, respectively. It is found that this classification can represent significantly different characteristics in subsequent evolution of each El Niño event, which will be described below.
Figure 2 shows time series of Niño-3.4 index for each type of El Niño events. It is seen that spring type El Niño events have large peak values at the mature phase, and have relatively regular evolution from the onset to the termination (Figure 2a). In contrast, summer-fall type events do not have such large peaks, and have rather irregular features in those evolutions, particularly after the first peak (Figure 2b). That is, several events (7 of 14) persist throughout the next year often having double or triple peaks, and others (7 of 14) terminate after the mature phase. This irregularity in summer-fall type El Niño events is consistent with “early-onset El Niño” mentioned by Neelin et al. . Their “early-onset” events mean their onset is earlier than regular El Niño (spring type in our definition). Therefore, “early-onset” type coincides with summer-fall type in our definition. Note that phase-locking of the peak in boreal winter is the same feature found in both types of El Niño events.
Figure 3a shows a relationship between the onset month and the Niño-3.4 index value at the peak. Spring type El Niño has a relatively large peak value (1.5–2.7°C), whereas most summer-fall type (13 of 14) has less than 1.5°C. It is interesting to note that all major El Niño events, such as 1972/73, 1982/83 and 1997/98 El Niño events, belong to the spring type.
Figure 3b shows a relationship between the onset month and the duration of El Niño event. The duration of the spring type is from 10 to 17 months. On the other hand, that of the summer-fall type widely scatters from 7 to 28 months. From the aspect of whether or not El Niño event terminates within the year next to El Niño onset year (Yr (+1); see also Figure 2b), we can further classify summer-fall type into two categories. One is the longer type, the duration of which is longer than 12 months, and often has double or triple peaks. The other is the shorter type, the duration of which is shorter than 12 months. Here, even if Niño-3.4 index falls below 0.5°C for a while (shorter than 6 months), when index exceeds again 0.5°C, then we dealt with the one consecutive El Niño event. Eventually, 7 events among 14 summer-fall type events belong to longer type and 7 shorter type. Our longer type events are generally in agreement with that of Tomita and Yasunari , that classified El Niño events focused on the duration.
 To provide further confidence for this classification, we examined the CTI by the same way taking account of the time lag between two indices. The result showed that, although a few events showed the change of timing of onset, the CTI provides essentially the same characteristics with those of using Niño-3.4 index. In addition to this examination, we also checked another threshold value of 0.75°C for Niño-3.4 index to define an El Niño event. This threshold resulted in similar results; the results are insensitive to minor modification of the threshold.
4.1. Development of El Niño
 Present classification of El Niño events is almost the same as that done in XC. However, it should be noted here that we treated longer period of the past 130 years, while XC did only recent 50 years. In addition, we put our principal interest on the relationship between the timing of El Niño onset and the subsequent evolution, while XC mainly studied the atmospheric and oceanic conditions before the El Niño onset.
 Why is the timing of El Niño onset so different? Some sources of the irregularity shown in previous studies are deterministic chaos associated with the nonlinear dynamics of ENSO [e.g., Jin et al., 1994], changes in the background states [e.g., Wang, 1995], an atmospheric forcing due to Asian-Australian monsoon [XC], stochastic weather noise forcing [e.g., Blanke et al., 1997], and an atmospheric forcing due to an intraseasonal oscillation in the tropical Pacific [e.g., McPhaden, 1999; XC]. Our observation shows that, although they are not common characteristics to all events, spring (summer-fall) type events are likely to have large (small) accumulation of positive OHC anomaly and anomalous westerly around February (April) before the onset in the western equatorial Pacific (not shown here). These suggest that the different antecedent atmospheric and oceanic conditions may cause the different onset timing.
 The noticeable different evolutions of two types of El Niño events might be attributed to seasonally varying strength of ocean-atmosphere coupling in the equatorial Pacific. That is, as many studies have already pointed out, the strength of atmosphere-ocean coupling, i.e., magnitude of air-sea coupled instability, varies seasonally [e.g., Tziperman et al., 1998]. So far, several possible controlling factors which induce stronger air-sea coupled instability have been pointed out: more equator ward position of the Intertropical Convergence Zone [Philander, 1983; Hirst, 1986], large west-east horizontal SST gradients, shallower thermocline, stronger winds, higher SSTs [Hirst, 1986], and stronger upwelling [Battisti, 1988]. Each of these factors can enhance the atmosphere-ocean coupling depending on a season. As a result, it is considered that the spring from April to June is a favorable season to enhance air-sea coupled instability, while an other season, especially from November to December is not [Tziperman et al., 1998].
 Therefore, we can suggest that spring type El Niño events can experience the favorable condition by which air-sea coupled instability can rapidly grow, and eventually El Niño event can develop. In contrast, summer-fall type events have to experience an unfavorable condition for instability, and resultantly cannot develop into as large events as spring type events.
4.2. Termination of El Niño
 The regularity of evolution from the peak (or mature phase) to the termination in spring type events may attribute to the transition mechanism from the warm event (El Niño) to the cold event (La Niña). Kessler  and Hasegawa and Hanawa  found that the negative ocean heat content (OHC: or equivalently shallower thermocline) anomaly in the western equatorial Pacific appears at the mature phase of El Niño event. This negative OHC anomaly then propagates eastward along the equator, and forces to terminate El Niño event. This scenario is in line with equatorial wave dynamics (“delayed oscillator theory” by Suarez and Schopf ), and consistent with the works that pointed out the importance of direct atmospheric forcing in western equatorial Pacific [Weisberg and Wang, 1997; An and Wang, 2001].
Figure 4 shows composite maps of OHC anomaly distribution at the month when Niño-3.4 index takes a peak. Here, we used the upper ocean temperature dataset prepared by White  from 1955 to 2000. OHC is defined as vertically averaged temperature (°C) from the sea surface to the depth of 300m following Hasegawa and Hanawa . In the present study, considering the data reliability, we used only 40 years from 1961 to 2000. Composite events are 5 for spring type (Figure 4a), 3 for longer summer-fall type (Figure 4b), and 2 for shorter summer-fall type (Figure 4c).
Figure 4a shows strong negative OHC anomaly in the western equatorial Pacific, while Figures 4b and 4c show weak negative OHC anomaly. Positive eastern OHC anomalies are also different in magnitude and area from each other. It is expected that in spring type events, the above mentioned transition mechanism might effectively act to terminate the event in the following seasons. On the other hand, in summer-fall type, the negative OHC anomaly in the western equatorial Pacific is rather weak. Assuming that the eastward propagation of negative OHC anomaly plays an important role for termination of El Niño, OHC anomaly in summer-fall type is indecisive to terminate, or to make the transition to La Niña. It is likely that this is one of the reasons why the summer-fall type events show irregularity in their evolution. Since the above is just speculation, this point should be solved in future work.
 In the present study, a relationship between the timing of El Niño onset and subsequent evolution was explored, using 130-year time series of Niño-3.4 index from 1871 to 2000. It was found that 24 El Niño events identified in the period treated can be classified into two types, based on the onset timing: spring type (10 of 24 events) and summer-fall type (14). In general, spring type events can develop relatively greater in magnitude, and continue for one year or so. On the other hand, summer-fall type events are weaker in magnitude, and can further be classified into two types in duration of event: longer one (7 of 14 events) and shorter one (other 7 events). The character of the evolution is regular in spring type, while it is rather irregular in summer-fall type.
 The authors wish to express their sincere thanks to members of Physical Oceanography Group at Tohoku University for their useful discussion. The first author (TH) was financially supported by the 21st Century Center-Of-Excellence program, ‘Advanced Science and Technology Center for the Dynamic Earth’, of Tohoku University.