Prediction of tropical cyclone precipitation over China by the DSAEF_LTP model

Abstract

AbstractThe Dynamical–Statistical–Analog Ensemble Forecast for Landfalling Typhoon Precipitation (DSAEF_LTP) model was used to forecast the accumulated precipitation of 92 tropical cyclone cases from 2004 to 2018 in China. Through the basic experiment of 76 training samples from 2004 to 2016, the best forecasting scheme of the DSAEF_LTP model in China was selected. The forecasting performance of the DSAEF_LTP model in 16 independent samples (2017–2018) with this scheme was found to be comparable to four numerical weather prediction (NWP) models (European Centre for Medium‐Range Weather Forecasts, Global Forecast System, Global/Regional Assimilation and Prediction Enhanced System, and Shanghai Meteorological Service WRF ADAS Real‐Time Modeling System). Specifically, the prediction ability of the DSAEF_LTP model ranked third among the five models. Meanwhile, the prediction performance of the DSAEF_LTP model in different areas demonstrated clear regional characteristics with the single best forecasting scheme for the whole of China, providing a skillful forecast in the coastal areas of southeast and south China. To improve the performance of the DSAEF_LTP model, China was divided into three subregions (south, southeast, and northern China), and subregion experiments were conducted. After regionalization, the forecasting performance of the DSAEF_LTP model improved to some extent. In general, the TSsum (sum of the threat score with precipitation over 250 mm and 100 mm) of the DSAEF_LTP model in the subregion experiments was 0.236, which represented an increase of 19.8% from the value of 0.197 in the basic experiment. Compared with the four NWP models, the TSsum of the DSAEF_LTP model was higher than that of the Global Forecast System (0.225), ranking it second only to the Shanghai Meteorological Service WRF ADAS Real‐Time Modeling System (0.248). In addition, the bias scores of different models showed that the DSAEF_LTP model had a tendency of false alarms, whereas the NWP models tended to misses.