GRAPES-TCM对登陆热带气旋降水的预报及其性能评估

Evaluation of GRAPES-TCM rainfall forecast for China landfalling tropical cyclone in 2006

基于GRAPES-TCM对2006年登陆热带气旋的降水预报结果,对该系统的24 h和6 h降水预报能力进行评估,并与基于卫星降水反演的外推预报(TRaP,Tropical Rainfall Potential)和相似台风降水预报技术(Analog Prediction Technique for Typhoon.Precipitation,TAPT)进行对比。各方法对登陆热带气旋降水的综合预报能力通过分析预报和观测降水散点图、预报平均绝对误差(MAE)及均方根误差得到,降水分布型态的预报能力通过计算预报和观测降水的相关系数估计。此外,还分析了BS、POD、FAR、ETS评分等常用降水预报评估指标。结果显示,GRAPES-TCM的24 h降水预报绝对误差和均方根误差比TRaP和TAPT都大。但是,GRAPES-TCM的24 h降水预报与观测降水的相关系数远比TRaP和TAPT高。对其他指标的分析表明,GRAPES-TCM的漏报率远低于TRaP和TAPT,但3种方法的空报率在同一水平;对任一强度的降水,GRAPES-TCM的ETS评分总是最高,TRaP和TAPT对于大暴雨以上的强降水则几乎没有预报能力。对24小时内每6 h的降水预报,3种方法相对性能与24 h总降水相似。通过对各强度降水造成的降水量在总降水量中的百分比的对比分析,发现GRAPES-TCM预报强降水占总降水量的比重与观测十分接近。总体上说,GRAPES-TCM能较好地预报出登陆热带气旋降水的分布型态,对强降水的预报能力强于外推和相似预报方法,但是预报的降水量绝对误差偏大,尤其对暴雨级别以上降水,其BS值明显偏大。

Up to the present, little verification has been performed for rainfall predictio ns from numerical forecasts of landfalling tropical cyclones. Using the output f rom the operational run of GRAPESTCM in 2006, this paper evaluated capability of the predicted rainfall for landfalling TC of this system. As comparison, the predicted rainfall from TRaP（Tropical Rainfall Potential） and a climatology met hod based on analogical tropical cyclone（Typhoon Analog Prediction Technique, TAPT） were analyzed simultaneously. Several measures of forecast quality were u sed to evaluate the predicted rainfall from these runs, using daily rain gauge d ata as ground truth. The overall quality was measured by the scattering plot bet ween observed and predicted rainfall, mean absolute errors and root mean square- errors. Additionally, more traditional precipitation verification scores were calculated including Bias Score, Probability of Detection （POD）, False Alarm Ratio （FAR）, and Equitable Threaten Score （ETS）. The MAE （mean absolute error） of pr edicted rainfall from GRAPES-TCM is 31 mm/d which is much larger than that f rom TRaP（20 mm/d） and TAPT（17 mm/d）. At the same time, GRAPES-TCM exhibited a higher pattern correlation with observations than TRaP and TAPT. The POD of predicted rainfall from GRAPES-TCM is also higher than that of TRaP and climatol ogy method while the FAR of these 3 methods are in the same grade. In addition, GRAPES-TCM got much higher ETS for every rainfall threshold than the other two methods. For the stronger sever storm （100 mm/d）, ETS of the predicted rainfal l from TRaP and climatology is 0. The correlation between 6-h forecasted precip itation and the observation also shows that GRAPESTCM has a better performance than the other two methods in each period （00-06, 06-12, 12-18, 18-24 hour）. Overall, GRAPES-TCM could forecast the precipitation pattern well while TC makes landfall and it has better performance on server precipitation forecast.