Periods of Linear Development of the ENSO Cycle and POP Forecast Experiments

Abstract

The results of the POP (principal oscillation pattern) analysis of the tropical Pacific wind stress data are presented. The wind stress data are smoothed and detrended in the same way as that used by Lamont's coupled ocean-atmosphere model to initialize EJ Niño forecasts. Thus, the present wind stress POP model serves as an indicator of prediction skill of the data alone, without the use of the coupled model. The POP results show that predictions of warm and cold events can be obtained at lead times of about two seasons, which is much shorter than the lead time of more than one year achieved by Lamont's coupled ocean-atmosphere model. It is shown that during the period of about two to three seasons before the peak of a warm/cold event, the ENSO system evolves in a linear, low-dimensional way. This properly allows a precursor of a warm/cold event to be identified around May of the event year. In other periods of the ENSO cycle, the POP model does not perform well. The author performed a cross-validation experiment, in which the data of two years following the month of the initial condition are withheld in both the EOF calculation and the POP model construction. It was found that the skill (measured by correlation coefficient) of the cross-validation model is 0.05–0.15 lower than that of the hindcast model. The ability of the cross-validation model to pick up the precursor of a warm/cold event is also slightly lower than that of the hindcast model.