El Niño in Five Guises: How Fuzzy Clustering Explained the Shift in Variability
Context
El Niño–Southern Oscillation (ENSO) is the main source of interannual climate variability. Its events differ in amplitude, evolution, and spatial pattern: classic eastern Pacific El Niños are strong, while central Pacific ones are weaker and shorter. These differences affect global impacts: droughts in Australia, rainfall in Brazil, and Atlantic temperatures. Decadal variations in ENSO characteristics are known, but how different event types contribute to these changes remained unclear. Existing classifications are often binary and depend on index choices, complicating analysis. This work fits within ENSO diversity research but differs from neighboring studies on El Niño prediction and gravity waves by focusing on the contribution of event types to variability rather than on forecasting or physical mechanisms.
Methods
The authors used monthly sea surface temperature anomalies (SSTA) in the tropical Pacific (130°E–70°W, 30°S–30°N) from eight reanalyses spanning 1901–2022. Data were projected onto the first two principal components (PC1-PC2) from empirical orthogonal function (EOF) analysis, reducing dimensionality. A Gaussian mixture model (GMM), a probabilistic fuzzy clustering method, was then applied to these two dimensions. Each winter (December–February) El Niño or La Niña event received membership probabilities for each cluster rather than a hard label. The number of clusters was determined by minimizing the Bayesian information criterion (BIC), which balances model complexity and fit. Robustness checks varied the number of EOFs, included sea level anomalies, and changed the latitude range. A nonlinear autoencoder was also compared.
Results
The analysis identified five categories: three El Niño types (extreme, strong, weak) and two La Niña types (weak, strong). Categories differ in intensity and the longitude of maximum anomalies. Extreme El Niños (1982-83, 1997-98, 2015-16) form a separate cluster distinct from strong ones. The 2015-16 event has nonzero membership probability in the strong type as well, confirming its mixed nature. Category contributions to the 20-year running variance of the Niño3.4 index showed that strong El Niños dominated before the 1930s, weak events from the 1930s through the 1960s, and after 1970 the contributions of strong La Niñas and extreme El Niños rose sharply. Covariance between categories accounts for 20–40% of decadal variance. Analysis of the CESM2 model revealed the absence of the extreme El Niño type, indicating an underestimation of nonlinearity.
Implications
This work is the first to quantitatively link ENSO diversity to low-frequency variability by decomposing variance into event-type contributions. Unlike neighboring studies that treat location and intensity separately or use binary classifications, this approach is fuzzy, allowing events to belong to multiple categories. This yields a more detailed picture and shows that the post-1970 variability increase is tied not simply to stronger El Niños but to specific types: extreme El Niños and strong La Niñas. The method also exposes shortcomings in climate models, which is important for their improvement.
Future development
The proposed approach can be used to assess how well climate models reproduce ENSO diversity. A preliminary CESM2 analysis showed that the model does not reproduce the extreme type, warranting further investigation. Future work could explore changes in ENSO types under climate change scenarios and their links to external forcings. It would also be valuable to study the dynamical mechanisms behind decadal shifts in category contributions, for example through heat budget analysis.
Impact
The results affect our understanding of ENSO predictability and its global impacts. They could help assess regional climate risks associated with extreme events.
Next steps
The next step is to systematically apply the method to an ensemble of climate models to evaluate their ability to reproduce ENSO diversity. Plans also include analyzing changes in ENSO types under future climate scenarios.
Key open problems
This work directly relates to the problem of ENSO prediction, as decadal modulation of diversity affects predictability. It also touches on the role of anthropogenic forcing on ENSO variability, since the rise in extreme events may be linked to warming. The dynamical causes of decadal shifts remain an open question: whether they result from external forcings or internal stochasticity.
The 2015-16 El Niño, one of the strongest on record, did not trigger the expected California drought nor lead to a strong La Niña, hinting at its mixed nature. This underscores the limitations of binary classifications.
Key numbers
- number of ENSO categories: 5 (3 El Niño types, 2 La Niña types)
- share of covariance in decadal variance: 20–40%
- Niño3.4 anomaly threshold for event selection: ±0.5 K
- analysis period: 1901–2022
- number of reanalyses: 8