The El Niño–Southern Oscillation (ENSO) remains the leading source of interannual climate variability, yet event strength fluctuates irregularly: the exceptionally powerful episodes of 1982–83, 1997–98, and 2015–16 were separated by 15–20 years. Forecasts issued in 2014 anticipated a strong El Niño a year before it actually occurred, and that miss underscored the limits of our understanding. This work joins efforts that seek sources of unpredictability in low-dimensional climate models, but it stands out by uncovering not just chaos but the coexistence of two qualitatively different regimes with sharp transitions between them. That links it to questions about early warning signals for climate shifts and to analyses of long paleoclimate records, which reveal epochs with differing ENSO variability.
The authors used the Jin–Timmermann (JT) conceptual model with three state variables: the east–west sea surface temperature difference across the tropical Pacific, the western temperature relative to the mean, and the thermocline depth in the west. They performed bifurcation analysis in the MATCONT software to locate parameter regions where two attractors coexist: a periodic mixed-mode oscillation (MMO) regime with strong El Niños and a chaotic attractor without them. They then added weak seasonal forcing with an amplitude of only 0.002 in the parameter a and followed a trajectory spanning 100,000 years, identifying epochs near each attractor based on threshold values of a specially chosen observable.
In the JT model with parameters [δ, ρ, c, k, a] = [0.225423, 0.3224, 2.3952, 0.4032, 7.3939], two attractors coexist. One is an almost periodic cycle with a period of about 12.1 years, featuring one strong El Niño per cycle; the other is a chaotic attractor where the variable x (the east–west temperature difference) stays far from zero, so no strong events occur. The basins of attraction for these attractors are interwoven in a fractal manner, and the attractors themselves lie very close to their basin boundaries. When weak annual forcing is added (amplitude 0.002), the system switches irregularly between regimes. In a 100,000-year simulation, the durations of chaotic epochs are roughly exponentially distributed, and successive durations show no correlation: knowing how long the current epoch lasts does not help predict the next. The longest epoch lasted about 540 years, but most were shorter than 400 years.
The study shows that the unpredictability of strong El Niños may stem not merely from chaotic sensitivity to initial conditions but from abrupt switches between two coexisting dynamical regimes. This differs from related studies, which typically consider a single attractor or gradual regime changes. Here, for the first time in a low-dimensional ENSO model, a mechanism is demonstrated whereby epochs of regular strong events can suddenly give way to epochs without them, and the timing of the switch is fundamentally unpredictable on a decadal scale.
This topic could evolve toward searching for analogous regimes and switches in full climate models and in paleodata. If similar almost-invariant sets are found in comprehensive models, that would change the approach to ENSO forecasting: instead of predicting a specific event, one would need to estimate the probability that the system resides in one regime or another. It may also be worthwhile to investigate the role of noise and the seasonal cycle in accelerating transitions and to seek observable precursors of regime shifts.
The results are relevant for seasonal and decadal climate prediction, for interpreting paleoclimate reconstructions of ENSO, and for assessing the risk of extreme weather events associated with strong El Niños.
Check whether similar coexisting attractors and abrupt switches occur in more realistic ENSO models; develop methods to detect almost-invariant sets in observational data.
This work directly addresses the unresolved problem of limited ENSO predictability: why strong events occur irregularly and whether their onset can be predicted beyond the seasonal predictability barrier. It also touches on the question of how well low-dimensional models can reproduce key features of climate variability.
In the JT model, a strong El Niño occurs when the east–west temperature difference across the tropical Pacific approaches zero: the ocean briefly becomes almost 'flat' in temperature.