Low-dimensional nonlinearity of ENSO and its impact on predictability

Using a hybrid coupled model, we perform a bred vector (BV) analysis and retrospective ENSO (El Niño and the Southern Oscillation) forecast for the period from 1881 to 2000. The BV local dimension and BV-skewness inherent to the intensity of nonlinearity are analyzed. Emphasis is placed on exploring the nature of the low-dimensional nonlinearity of the ENSO system and the relationship between BV-skewness and model prediction skills. The results show that ENSO is a low-dimensional nonlinear system, and the BV-skewness is a good measure of its predictability at the decadal/interdecadal time scales. As the low-dimensional nonlinearity of ENSO is weakened, high predictability is attained, and vice versa. The low-dimensional nonlinearity of ENSO is also investigated and verified using observations.Another finding in this study is the relationship between the error growth rate (BV-rate) and actual prediction skill. While there is a good positive correlation between them in some decades, the BV-rate demonstrates a strong inverse correlation with the prediction skill in other decades. The BV-rate components contributed by the nonlinear process play a dominant role in quantifying ENSO predictability. The possible mechanism for the link between BV-rate, BV-skewness and ENSO predictability is discussed.