STOCHASTIC PROCESS IN THE TIME SERIES MODEL OF PACIFIC DECADAL OSCILLATION (PDO)

Abstract

Pacific Decadal Oscillation (PDO) is long-lived El Niño-like pattern of Pacific climate variability generated by coupled ocean-atmosphere interaction in the Northern Pacific Ocean. The best way to acquire a signal of PDO evidence is by determining the index of PDO. In this study, the PDO indexes are accurately modeled with time series methods through exponential smoothing analysis (Single and Holts Double exponential smoothing model) and Box-Jenkins analysis (ARIMA {1,1,1}, {2,1,1}. {3,1,1} and {4,1,1}). Nicholas’s PDO Model (ARMA 9, 7) is also considered as comparative model in order to obtain the level of the reliability models that have been produced. The best selected prediction model that close to the real PDO index is ARIMA (2,1,1) Zt = 1.574* Zt-1 -0.427* Zt-2 -0.147* Zt-3 -0.976* at-1 which means the forecast of PDO in the future depending on three months earlier data and a month earlier error of PDO index. Mean absolute error (MAE) of this model is 0.5283 and with root mean square error (RMSE) 0.6661. The predicted and observed PDO indexes are significantly correlated with r =0.76.

Keywords: PDO, Box-Jenkins analysis, Exponential smoothing analysis