基于降水极端预报指数的福建台风极端降水预报研究

应用1961—2017年中国气象局热带气旋最佳路径数据集、国家地面气象观测站日降水观测资料和2015年8月—2017年12月欧洲中期天气预报中心(ECMWF)集合预报系统降水极端预报指数(EFI)数据,根据百分位法定义台风影响期间福建省各站点的台风极端降水阈值,采用最小阈值法剔除台风极端降水时EFI箱线图中的异常值,保...

A study of typhoon extreme precipitation forecast in Fujian based on precipitation extreme forecast index

The Extreme Forecast Index (EFI) is an indication of potential extreme weather event based on measurement of the differences between the ensemble prediction and the model climate. By using 24 h total precipitation EFI product from the ECMWF ensemble prediction system, an objective method of EFI threshold is developed to predict extreme precipitation events in Fujian province caused by tropical cyclones (TCs). For individual national synoptic weather station, its daily TC rainfall observation exceeding the 95th percentile of its climatological ones from 1961 to 2017, used as the extreme threshold, is defined as TC extreme precipitation event. The EFI threshold at each lead time up to 5 d for TC extreme precipitation at every station is defined by a minimum threshold method, which takes the minimum value as the threshold after removing the anomalous values (outliers and negative values of EFI forecast in this study) from the EFI box plot drawn with the EFI forecasts when the TC extreme precipitation events occurred during the period from August 2015 to December 2018. Experiments have been done by applying the EFI thresholds to reforecast and forecast TC extreme precipitation events in Fujian province from August 2015 to December 2017 and in 2018, respectively, and the performances are evaluated based on the threat score (TS), false alarm ratio and miss rate. The results are as follows. The TC extreme precipitation threshold gradually reduces from coastal areas to inland areas in the northwest of Fujian province. The coastal areas in north central Fujian province have the largest threshold (greater than 100 mm) while the inland area in the northwest of Fujian province has the smallest threshold (less than 50 mm). Daily precipitation caused by TCs, especially extreme precipitation, has significant positive correlation with 24 h precipitation EFI. The box difference indexes of EFI also show great capability to distinguish TC extreme and non-extreme precipitation events. TS for the forecast experiments are 0.26, 0.22, 0.20 and 0.19 with lead times of 12—36, 36—60, 60—84 and 84—108 h initialized at 20:00 BT. Forecast experiments have better forecast skills than reforecast experiments in general, and TC cases with significant extreme precipitation tend to have better forecast results. The disadvantages of both experiments are high false alarm rates which are more likely to occur in TC cases with no extreme precipitation. In summary, precipitation EFI is a good indicator to forecast TC extreme precipitation. This objective forecast method based on precipitation EFI provides a reference for TC extreme precipitation forecast.