Decadal Variability in ENSO Predictability and Prediction

A simple coupled model is used to examine decadal variations in El Niño–Southern Oscillation (ENSO) prediction skill and predictability. Without any external forcing, the coupled model produces regular ENSO-like variability with a 5-yr period. Superimposed on the 5-yr oscillation is a relatively weak decadal amplitude modulation with a 20-yr period. External uncoupled atmospheric ‘‘weather noise’’ that is determined from observations is introduced into the coupled model. Including the weather noise leads to irregularity in the ENSO events, shifts the dominant period to 4 yr, and amplifies the decadal signal. The decadal signal results without any external prescribed changes to the mean climate of the model. Using the coupled simulation with weather noise as initial conditions and for verification, a large ensemble of prediction experiments were made. The forecast skill and predictability were examined and shown to have a strong decadal dependence. During decades when the amplitude of the interannual variability is large, the forecast skill is relatively high and the limit of predictability is relatively long. Conversely, during decades when the amplitude of the interannual variability is low, the forecast skill is relatively low and the limit of predictability is relatively short. During decades when the predictability is high, the delayed oscillator mechanism drives the sea surface temperature anomaly (SSTA), and during decades when the predictability is low, the atmospheric noise strongly influences the SSTA. Additional experiments indicate that the relative effectiveness of the delayed oscillator mechanism versus the external noise forcing in determining interannual SSTA variability is strongly influenced by much slower timescale (decadal) variations in the state of the coupled model.