The relative importance of chaotic dynamics versus stochastic processes in the evolution of El Nino-Southern Oscillation (ENSO) is examined in three different types of coupled models - an intermediate coupled model (ICM), a hybrid general circulation model (HGCM) and a fully coupled ocean-atmosphere general circulation model (CGCM). It is shown that in both the ICM and HGCM whose atmospheric components contain no internal high-frequency variability, irregularity of ENSO can be described by a low-order chaotic process generated by nonlinear interaction between the seasonal cycle and interannual oscillation. Numerical experiments reveal that the behavior of the ENSO cycle in this class of coupled models is sensitive to changes in coupling strength. By increasing the coupling strength, the model ENSO cycles evolve from nonoscillatory (stable) to time-periodic (unstable) and eventually to chaotic regimes, Although these models give reasonable ENSO frequency and spatial structure compared with observations, the phase-locking with the annual cycle is apparently too strong. Inclusion of stochastic forcing in these models can have two effects on the ENSO cycles. It can break up strong annual phase-locking in the unstable regime and it can also excite ENSO-like variability in the stable regime, where the coupling strength is so weak that no self-sustaining oscillations can exist in the coupled models. In contrast, the ENSO cycle in the CGCM, where internal high-frequency fluctuations are included, does not appear to be driven by a low-order chaos. By comparing the invariant properties of the dynamics derived from long-term sea surface temperature time series obtained from the different coupled models, it was found that the dynamical characteristics of the CGCM response were similar to those of the ICM forced with stochastic forcing in the stable regime, suggesting that the ENSO cycles in the CGCM may be dominated by stable dynamics driven by stochastic processes. Testing for nonlinearity gives further support to this result. The nonlinear time series analysis also implies that stochastic processes rather than chaotic dynamics are likely to be a major source of ENSO irregularity in reality.
