Statistical Prediction of the Accumulated Cyclone Energy in the Western North Pacific from July to August

Based on the causality of information flow, sea surface temperature (SST), and air–sea indices are used to determine the factors that can affect the interannual variation of the dominant modes of the accumulated cyclone energy (ACE) in the Western North Pacific (WNP) from July to August during 1979–2020. Then, multiple linear stepwise regression is used to select the most significant predictors. The first two modes of July–August ACE are the basin and dipole modes, as determined by empirical orthogonal function (EOF) analysis. The prediction factors for the principal component (PC) of the basin mode include the 3-month leading SST in Marine Continent and Central North Pacific, the 5-month leading quasi-biennial oscillation index (QBO), and the 11-month leading tropical Indian Ocean dipole mode index (TIOD). On the other hand, the predictive factors for the PC of the dipole mode include the 2-month leading SST in North Atlantic, the 12-month leading SST in the Japanese trench, the 7-month leading Atlantic meridional mode index (AMM), and the 8-month leading North Atlantic Oscillation index (NAO). The prediction equations are established based on these prediction factors. The correlation coefficients between the predicted PCs and the observed PCs of the first two modes are 0.75 and 0.77, respectively, both statistically significant at the level of 0.01. The cross-validation method indicates the prediction equations are stable and have good hindcasting ability. The temporal correlation coefficient skill of the WNP area averaged ACE anomaly is 0.76. The averaged pattern correlation coefficient skill of the ACE anomaly is 0.35 over the WNP basin during 1980–2020. The prediction model performs well in the years when the ACE can be reconstructed by the first two modes.