基于ECMWF产品福建省前汛期短时强降水预报方法

A Method of Short-time Strong Rainfall Forecasting During Pre-rainy Season in Fujian Based on ECMWF Productions

利用2014—2016年福建省1605个自动气象站逐时降水资料和ECMWF全球模式细网格预报产品,分析福建省前汛期短时强降水发生背景下模式预报物理量的分布特征,并基于阈值判定的方法建立短时强降水预报模型。结果表明:福建省内陆县市前汛期短时强降水发生频次较高,沿海县市发生频次低,且日变化特征表现出双峰结构。箱型图差异指数(I_(bd))在评估相关变量对于区分短时强降水发生与否的敏感程度有较好的作用,比湿、整层可降水量等水汽变量I_(bd)最为显著,K指数、对流有效位能等变量的I_(bd)仅次于水汽变量,说明模式预报变量对于预测短时强降水有较好的表征作用。针对短时强降水事件的物理量集合,采用剔除异常值后的最小值作为判定阈值,通过训练集分析结果客观订正对流有效位能和3 h降水量两个高I_(bd)变量的阈值,建立潜势预报模型。对于福建省西部的关键区,检验集白天时段12 h时间分辨率预报TS评分可达0.5,夜间时段约为0.3。对于福建省进行分区建模预报,检验集预报结果显示白天时段比夜间准确率高、内陆县市比沿海县市准确率高。

Distribution features of model variables accompanied with short-time strong rainfall events are investigated based on hourly precipitation data from 1605 automatic weather stations and ECMWF 0. 125° X 0. 125° fine grid model products, and a method of short-time strong rainfall forecasting based on threshold determination is established. Results show that short-time strong rainfall occur more frequently in Fujian inland area and less in Fujian coastal area during pre-rainy season, and the diurnal variation exhibits double peaks with the notable one at 1700 BT and the inapparent one at 0500 BT. The box difference index is useful to check whether a variable could differentiate short-time strong rainfall events well. The box difference index of humidity variables like specific humidity at 925 hPa and total column water vapor are most prominent followed by K index and convective available potential energy（CAPE） which shows these variables have good performances in distinguishing short-time strong rainfall events. Some variables like temperature difference between 850 hPa and 500 hPa and temperature change in 24 h at 500 hPa perform poorly in differentiating short-time strong rainfall events.The minimum threshold method based on the minimum values of variables after eliminating outliers works well in judging short-time strong rainfall events, which could decrease vacancy forecast rate effectively through increasing missing forecast rate appropriately compared with adopting real minimum values as threshold. In the key area（25. 9° —27. 1°N, 116. 4° —117. 4°E）, TS（threaten score） of validation set in2016 with 12 h interval reaches 0. 5 at daytime and 0. 3 at nighttime just based on the minimum threshold method. Revising threshold of variables with high box difference index values could improve the accuracy with the nighttime TS of validation set in 2016 increasing from 0. 3 to 0. 34. TS of 2016 is relatively lower compared with that of 2014—2015, and the cause may be that short-time strong rainfalls happen much more frequently in 2016 which is a very strong El Nino year.To establish a potential forecast model of short-time strong rainfall during pre-rainy season, Fujian is divided into grids of 1°X1°, and minimum threshold method is applied in each grid followed by threshold revise of variables with high box difference index values. This model could analyze all kinds of variables comprehensively besides those considered by weather forecasters. TS with 12 h interval at daytime mainly ranges from 0. 3 to 0. 5 while TS at nighttime is relatively lower. TS in inland area is much better than coastal area both at daytime and nighttime mainly because short-time strong rainfall occurs more frequently in inland area than coastal area during pre-rainy season climatologically.