El Niño–Southern Oscillation (ENSO) is the irregular alternation of warm (El Niño) and cold (La Niña) phases of ocean surface temperature in the tropical Pacific. It recurs every 3–7 years and disrupts normal wind and rainfall patterns across the globe. A hierarchy of ENSO models exists, from simple conceptual 'box' models with a few equations to complex general circulation models. Conceptual models are valued for their mathematical transparency and low computational cost, but they are usually built on physical intuition rather than derived rigorously from more complete systems. This work fills that gap: it shows how low-dimensional conceptual models can be rigorously obtained from a spatially distributed stochastic dynamical system with a linear stable core and multiplicative noise.
The original spatially distributed system describes the interaction of the atmosphere, ocean, and sea surface temperature (SST) on interannual timescales. Its dynamical core is linear, deterministic, and stable, while intraseasonal wind bursts (including the influence of the Madden–Julian oscillation) are parameterized as a simple stochastic process with multiplicative noise. The noise amplitude depends on western Pacific SST through a two-level Markov jump process. To reduce dimensionality, principal component analysis is applied based on the eigen-decomposition of the linearized system matrix. Two leading complex-conjugate modes emerge with a period of about 4.6 years and a decay time of 1.5 years, explaining the vast majority of variability. Projecting the stochastic forcing onto these modes yields a two-dimensional system of ordinary differential equations.
The projection of the full solution onto the two leading modes is nearly indistinguishable from the original solution in terms of spatiotemporal patterns and key statistics: the power spectrum has a peak in the 3–7 year range, and the probability density function of eastern Pacific SST is strongly non-Gaussian with positive skewness and a heavy tail corresponding to extreme El Niño events. The correlation between reconstructed and observed eastern Pacific SST exceeds 0.85 at all longitudes. Within the general framework, stochastic versions of the recharge-discharge oscillator and the delayed oscillator are rigorously derived. For the recharge-discharge oscillator, coupling coefficients are positive for the Bjerknes positive feedback and the recharge mechanism, while a negative coefficient describes discharge. The phase shift between eastern SST and western thermocline is 122 degrees, not 90, consistent with observations. In the delayed oscillator, an exponential damping of the delayed term appears, along with two types of noise contribution: instantaneous and memory-dependent.
This work stands apart from neighboring studies: it does not merely propose another conceptual model but provides a rigorous bridge between spatially distributed and low-dimensional descriptions of ENSO. Unlike many existing models that focus on a small set of variables or averaged quantities, here the low-dimensional model retains full spatiotemporal patterns for the atmosphere, ocean, and SST through the eigenvectors. This allows direct linking of different system components and testing hypotheses about causal relationships. The rigorous derivation also clarifies the role of instantaneous and delayed effects of stochastic noise in large-scale ENSO dynamics.
The current framework targets eastern Pacific El Niño events. To account for ENSO diversity, including central Pacific El Niño, it is necessary to add a mechanism for zonal ocean advection and possibly nonlinearity to the original spatially distributed model. The derived conceptual models can serve as testbeds for strategies to incorporate stochastic parameterizations into more complex intermediate-complexity models and general circulation models. It is also possible to build a hierarchy of models with varying complexity using data-driven methods.
The results may improve understanding of ENSO mechanisms and provide simple yet physically based models for forecasting El Niño indices and studying predictability from an ensemble perspective.
Incorporate the mechanism for central Pacific events and nonlinearity into the original model, then test how the low-dimensional models reproduce ENSO diversity. Use the conceptual models as testbeds for developing stochastic parameterizations in more complex climate models.
This work directly addresses the problem of El Niño diversity: why events occur in different regions of the Pacific and have different amplitudes and durations. The rigorous derivation of low-dimensional models provides a basis for studying how nonlinearity and advection generate central Pacific events, which are not yet reproduced by the linear framework.
The phase shift between sea surface temperature in the east and thermocline depth in the west is 122 degrees, not 90, as one might expect for a simple harmonic oscillator. This detail is important for correctly reproducing the time delay in the delayed oscillator.