Tropical cyclone full track simulation in the western North Pacific based on random forests

The tropical cyclone (TC) full track simulation method pioneered by Vickery has been widely used in the estimation of hurricane/typhoon wind hazards. The linear auto-regression used in the Vickery model may be inadequate to capture the complex patterns of storm tracks in the western North Pacific Ocean, which accommodates almost one-third of the world's annual TCs. In this study, a novel TC simulation method is developed based on the random forests method (RF) in conjunction with cluster analysis of TC tracks. Specifically, TC movement models with probabilistic termination classifiers, and the TC intensity model are newly developed based on the RF technique for typhoon simulation in the western North Pacific. Historical typhoon tracks are classified into four clusters in light of their geographical characteristics for facilitating the development of the TC movement model corresponding to each cluster. Thousand-year typhoon simulations for 47 coastal sites of China are performed by the proposed RF-based full track simulation method, which achieves more accurate results in reproducing historical statistics of key typhoon parameters than those of the original Vickery full track simulation approach. •A novel typhoon full track simulation method is developed by using random forests (RF) algorithms.•With the aid of cluster analysis, four RF-based movement models with termination classifiers, are developed and validated.•One RF-based tropical cyclone intensity model is also developed and evaluated by several intensity metrics.•The proposed RF-based full track simulation shows a well accuracy in reproducing the historical statistical characteristics.