Forecasting experiments of a dynamical-statistical model of the sea surface temperature anomaly field based on the improved self-memorization principle
Institutional Repository Document
Overview
Identity
Other
View All
Overview
abstract
Abstract. With the objective of tackling the problem of inaccurate long-term El Nio Southern Oscillation (ENSO) forecasts, this paper develops a new dynamical-statistical forecast model of sea surface temperature anomaly (SSTA) field. To avoid single initial prediction values, a self-memorization principle is introduced to improve the dynamic reconstruction model, thus making the model more appropriate for describing such chaotic systems as ENSO events. The improved dynamical-statistical model of the SSTA field is used to predict SSTA in the equatorial eastern Pacific and during El Nio and La Nia events. The long-term step-by-step forecast results and cross-validated retroactive hindcast results of time series T1 and T2 are found to be satisfactory, with a correlation coefficient of approximately 0.80 and a mean absolute percentage error of less than 15%. The corresponding forecast SSTA field is accurate in that not only is the forecast shape similar to the actual field, but the contour lines are essentially the same. This model can also be used to forecast the ENSO index. The correlation coefficient is 0.8062, and the MAPE value of 19.55% is small. The difference between forecast results in summer and those in winter is not high, indicating that the improved model can overcome the spring predictability barrier to some extent. Compared with six mature models published previously, the present model has an advantage in prediction precision and length, and is a novel exploration of the ENSO forecast method.
