Five Faces of El Niño: How Fuzzy Clustering Explained the Shift in Variability

Original: "Contributions of El Niño Southern Oscillation (ENSO) Diversity to Low-Frequency Changes in ENSO Variance"
arXiv:2307.11552 · 2023-07-21 · CC BY 4.0 · 3 min
Fuzzy clustering revealed five types of El Niño and La Niña, showing that the rise in variability after 1970 is linked to extreme and strong events.

The Pacific Ocean is a vast pot of water that someone occasionally stirs, now at one edge, now at another. When warm water piles up off the coast of South America, fishermen call it El Niño, "the child," because it often arrives around Christmas. But today El Niño is not a single mask but an entire theater with five roles, and new research shows why some roles have become more frequent on stage while others have faded.

For a long time, scientists divided El Niño events into two types: eastern, with anomalies off Peru, and central, with a shift toward the middle of the ocean. That division is convenient, but it is like sorting apples only by color: size, sweetness, and ripeness remain out of view. The study's authors abandoned rigid labels and allowed each event to belong to several types at once. They took sea surface temperature data spanning 122 years and applied a method that breaks down each event into probabilities: for example, the 2015–2016 event is 80% extreme, 15% strong, and 5% weak. This approach is called fuzzy clustering, and it resembles the work of a forensic scientist who does not say "it is definitely this suspect" but instead assesses the odds for each.

Five faces: extreme El Niño (three times in 122 years), strong, weak, weak La Niña, and strong La Niña. The extreme type is separated from the strong type as a Category 5 hurricane is from a Category 3.

By decomposing variability into contributions from these five types, the authors saw a changing of the guard. Before the 1930s, strong El Niños played the leading role. From the 1930s through the 1960s, weak events dominated the stage, and the Pacific behaved calmly, like a pot simmering on low heat. After 1970, the heat was turned up: contributions from strong La Niñas and extreme El Niños surged. These types, not El Niño in general, are responsible for the increase in variability that meteorologists noticed in the late 20th century. Weak events, in contrast, have become rare guests.

The climate model CESM2, one of the most advanced, does not reproduce the extreme El Niño type. It is like a theater that has no role for King Lear: the play goes on, but the climax is missing.

Why does this matter? Because forecasting El Niño means forecasting droughts in Australia, floods in Peru, and warm winters in Siberia. If a model misses the extreme type, it may underestimate risks. The new method allows us to test models for their ability to play all five masks, not just two. And this is no longer classification for its own sake: it is about how accurately we can peer into the climate's kitchen and understand who is at the stove today.

The Pacific pot continues to boil, and now we have a recipe to separate the pepper from the bay leaf. The five faces of El Niño are not just a list but a tool that shows how the very character of variability is changing. If before we only saw that the soup had become spicier, now we know exactly which ingredient was added.

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.

p(z) = \sum_{k=1}^{K} \pi_k \mathcal{N}(z | \mu_k, \Sigma_k)
The data distribution in principal component space is described by a mixture of K Gaussian bells, each with its own mean and covariance; the weights π_k are the probabilities of belonging to a cluster.
y(t) = \sum_{k} p(c_k | z(t)) \cdot y(t) := \sum_{k} y_k(t)
Any quantity (e.g., the Niño3.4 index) can be expressed as the sum of its parts, weighted by the probability that each time point belongs to each category.
\langle y^2 \rangle = \sum_{k} \langle y_k^2 \rangle + \sum_{l \neq m} \langle y_l, y_m \rangle
Total variance is the sum of each category's variance plus the sum of covariances between different categories, allowing their relative contributions to be assessed.
Concepts
ENSO El Niño index Southern Oscillation El Niño-Southern Oscillation Multivariate ENSO Index La Niña spring predictability barrier ERSST Jin-Timmermann model recharge oscillator model Westerly wind burst NOAA CPC TAO/TRITON moorings equatorial wave propagation natural variability relative ONI North Equatorial Countercurrent Warm water volume interannual variability mixed mode oscillations