基于误差权重的站点降水订正预报方法研究  

Study on Revised Forecast Method of Station PrecipitationBased on Error Weight

作　　者：庄晓宵 Zhuang Xiaoxiao(Xiushui Meteorological Bureau,Jiujiang 332400,China)

机构地区：[1]修水县气象局,江西九江332400

出　　处：《气象与减灾研究》2022年第3期196-206,共11页Meteorology and Disaster Reduction Research

基　　金：2020年江西省气象局面上项目“基于智能网格的乡镇预报研究”.

摘　　要：基于2018—2020年逐日修水国家基本气象站降水实况和模式降水预报数据,对ECWMF、NCEP、JMA、CMA-GFS、CMA-SH9等5种模式24 h及48 h降水预报进行了误差分析及检验。结果表明,总体上各模式24 h、48 h降水预报误差均以一个量级的正误差为主;各模式24 h降水预报误差小于48 h;CMA-SH9、JMA模式24 h、48 h预报误差均较小;总体上各模式24 h预报评分优于48 h;华东模式、日本模式24 h、48 h晴雨准确率均较高;CMA-SH9、ECMWF、NCEP模式24 h、48 h降水预报各量级风险评分均较高;JMA模式对24 h暴雨及以上量级降水可能具有一定的预报指示意义;总体上各模式24 h、48 h降水预报风险评分随量级增大而降低,空报率、漏报率随量级增大而增大;总体上各模式24 h、48 h小雨预报性能较优,中雨均易空报,大雨均易漏报,暴雨及以上降水均易空报与漏报。进一步通过引入滑动训练期,并基于平均绝对误差权重的多模式集成订正方案对降水预报进行订正,且对订正预报结果进行了整体及分季节评估检验。结果表明,相对多数模式,经订正后24 h、48 h降水预报性能总体上均有所提升,即提高了预报准确率,减小了预报误差,改进了晴雨准确率,提升了小雨—中雨的预报评分,提高了夏季48 h暴雨及以上降水的风险评分,并降低了其空报率;48 h降水订正预报效果总体优于24 h,春季及秋季降水订正预报效果总体优于夏季及冬季。Based on the daily precipitation observation data of the Xiushui national basic meteorological station and various forecast data of different models from 2018 to 2020,the error analysis and verification of the 24-hour and 48-hour precipitation forecasts of five models including ECWMF,NCEP,JMA,CMA-GFS,and CMA-SH9 were carried out.The results showed that the precipitation forecast error of each model was mainly positive error of one order of magnitude.The 24-hour precipitation forecast error of each model was smaller than that of 48-hour.The 24-hour and 48-hour precipitation forecast errors of CMA-SH9 and JMA were smaller than others.The precipitation forecast evaluation of 24-hour was better than 48-hour for each model in general.The 24-hour and 48-hour precipitation forecast evaluation for“rain or no rain”events of CMA-SH9 and JMA were better than others.The 24-hour and 48-hour precipitation forecast threat score for each precipitation grade of CMA-SH9,ECMWF and NCEP were greater than others.The 24-hour precipitation forecast of the JMA may have a certain indicative significance for forecast of rainstorm and above.The 24-hour and 48-hour precipitation forecast threat score of each model may decrease,the false alarm ratio and missing alarm ratio may increase when the precipitation grade increased.The 24-hour and 48-hour light rain forecast performance of each model were excellent,but there was a high false alarm ratio for moderate rain and a high missing alarm ratio for heavy rain.For rainstorm and above,each model presented a high false alarm ratio and a high missing alarm ratio.Furthermore,by introducing the sliding training period,the precipitation forecasts were revised using a multi-model ensemble revision scheme based on mean absolute error weights,the overall and seasonal revised forecast were evaluated.The results showed that the precipitation forecast performance of 24-hour and 48-hour revised forecast was improved in general compared with most models,indicating that the forecast accuracy was improved

关 键 词：降水 订正预报 误差权重 多模式集成 训练期 

分 类 号：P456.5[天文地球—大气科学及气象学]