Despite the tremendous progress in the theory, observation and prediction of El Niño over the past three decades, the classification of El Niño diversity and the genesis of such diversity are still debated. This uncertainty renders El Niño prediction a continuously challenging task, as manifested by the absence of the large warm event in 2014 that was expected by many. We propose a unified perspective on El Niño diversity as well as its causes, and support our view with a fuzzy clustering analysis and model experiments. Specifically, the interannual variability of sea surface temperatures in the tropical Pacific Ocean can generally be classified into three warm patterns and one cold pattern, which together constitute a canonical cycle of El Niño/La Niña and its different flavours. Although the genesis of the canonical cycle can be readily explained by classic theories, we suggest that the asymmetry, irregularity and extremes of El Niño result from westerly wind bursts, a type of state-dependent atmospheric perturbation in the equatorial Pacific. Westerly wind bursts strongly affect El Niño but not La Niña because of their unidirectional nature. We conclude that properly accounting for the interplay between the canonical cycle and westerly wind bursts may improve El Niño prediction.

Our understanding of El Niño dynamics started from the recognition that it is a coupled variability of the tropical Pacific ocean–atmosphere system1. Present theories can be generally grouped into two different frameworks, one considering El Niño/La Niña as a regular and self-sustaining oscillation with its timescale determined by the recharge and discharge of the equatorial upper-ocean heat content2, 3, 4, 5, and the other regarding it as a highly damped oscillation with each event triggered by atmospheric noise, especially the westerly wind bursts (WWBs) in the tropical western Pacific6, 7, 8, 9. The former framework can be readily applied to the basic El Niño/La Niña cycle and is consistent with the high potential predictability of El Niño10, 11, 12, whereas the latter seems to explain the irregularity of El Niño yet suggests that it is virtually unpredictable at long lead times. To better describe El Niño diversity and to provide a dynamically consistent interpretation for such diversity, we need a unified perspective that considers the different views on the classification of El Niño diversity, and reconciles the present theories to account for the interaction between the low-frequency recharge–discharge oscillation and the stochastic atmospheric forcing13, 14.

Every El Niño event is different from others, but it is often useful to classify different events into a few distinctive types according to the common manifestation, mechanism and impact of each type. In early years, El Niño was mostly studied in a composite form, such as the 'canonical El Niño' constructed by Rasmusson and Carpenter15 based on seven events in the 1960–1970s, which has the largest variance in the central-eastern equatorial Pacific and remains the average pattern when all known events over the past 150 years are taken into account10. It was then suggested that El Niño could be classified into two or three basic types16. In the same spirit, a series of recent studies have emphasized the different flavours of El Niño, with particular attention to a type that consists of warm events centred in the central-western equatorial Pacific17, 18, 19, 20, 21. In contrast to the strong El Niño that occurs in the eastern Pacific cold tongue, weak warm events of this type have been named “El Niño Modoki”18, “warm-pool El Niño”19 or “central Pacific El Niño”21. It has been suggested that the spatial pattern of El Niño Modoki could be an artefact of the orthogonality requirement of empirical orthogonal function (EOF) analysis22. It was further argued that this mode is not distinctively different from the canonical El Niño, and that El Niño should instead be considered primarily as a broad central Pacific phenomenon plus a few extremely strong eastern Pacific events23.

As an alternative to EOF and composite analyses, we apply the fuzzy clustering method24 (see Methods) to the tropical Pacific sea surface temperature (SST) anomaly data from HadISST25 over the past 50 years. This naturally and consistently reveals three warm patterns and essentially only one cold pattern (Fig. 1). The first warm pattern (Fig. 1a) consists of extremely strong El Niño events that had the largest warming near the South American coast. The second warm pattern (Fig. 1b) is a cluster of weak warm events centred near the dateline, very similar to the 'warm-pool El Niño' that has recently aroused a great deal of interest. The third (Fig. 1c) is basically the canonical El Niño with moderate warming in the central-eastern equatorial Pacific, which is quite symmetric to the only cold pattern identified (Fig. 1d–f). Thus, there seems to be a symmetric, canonical cycle in the central-eastern equatorial Pacific that represents a large portion of El Niño and La Niña events. Superimposed on this basic cycle are rare extreme El Niño events in the eastern Pacific and weak but more frequent warm events near the dateline, which collectively give El Niño its different flavours.

http://www.nature.com/ngeo/journal/vaop/ncurrent/full/ngeo2399.html