Dynamical-Statistical Hybrid Seasonal Prediction of Tropical Cyclone Track Density Over Western North Pacific

Abstract Compared with total account of basin-wide tropical cyclones (TC) genesis, the prevailing tracks of TC activity and its potential of landfalling is more important for disaster prevention. Despite its relatively lower predictability, a statistical-dynamical hybrid prediction model was developed based on the knowledge of the physical mechanism between western North Pacific (WNP) TC activity and related large-scale environmental fields from July to September. The leading modes of spatial-temporal variation of WNP TC tracks density its climatological peak season (July to September) was extracted using empirical orthogonal function (EOF) decomposition. The interannual variation of leading EOF modes of WNP TC track density was predicted using multiple linear regressions (MLR) method based on predictors selected by correlation analysis of both observational and Beijing Climate Center climate system model version 1.1 (BCC_CSM1.1) hindcast data. The predicted spatial distribution of WNP TC tracks density was obtained through weighted composite of forecasting EOF modes according to its variance explained respectively. Results of one-year-out cross validation indicates that forecast model well captures the interannual variation of WNP TC prevailing moving tracks, especially in South China Sea (SCS) and southeastern quadrant of WNP. The prediction skill enhanced with decreased forecast lead time, with anomaly correlation coefficient (ACC) of northern SCS and southeast quadrant of WNP reaches 0.6 for the period 1991-2020 with one month forecast lead time. Forecast assessment based on different ENSO phases indicate that source of predictability of WNP TC tracks was mainly originate from ENSO events, especially strong El Niño events.