基于聚类分析法区域集合预报模式的初边值试验研究  

Validation of strategies using clustering analysis for initial perturbations in limited area model ensemble prediction system

作　　者：王太微[1] 王勇 陈德辉[3] 梁寒[1] 罗聪[4] 

机构地区：[1]辽宁省气象台,辽宁沈阳110166 [2]Central Institute for Meteorology and Geodynamics [3]中国气象局数值预报中心,北京100081 [4]辽宁省防雷技术服务中心,辽宁沈阳110166

出　　处：《气象与环境学报》2015年第6期18-26,共9页Journal of Meteorology and Environment

基　　金：国家重点基础研究发展计划(973项目)"无缝隙强对流天气临近短时预报关键技术和预报系统"(2013CB430106);中国气象局预报员专项"制约辽宁局地长历时暴雨的环境特征分析"(CMYAYBY2015-015);中国气象局小型业务建设项目"辽宁集合预报应用业务建设(一期)"共同资助

摘　　要：选取2008年3月21—29日奥地利地区的降水天气过程,从欧洲中期天气预报中心全球集合数值预报模式生成的50个全球集合预报成员出发,通过聚类方法选取有代表性的16个成员,为中欧有限区域集合预报系统区域集合预报系统提供初边值扰动,进行初边条件不确定性数值试验,揭示初边条件不确定性对区域集合预报的影响。结果表明:经过聚类后选取的16个成员可以较好地代表50个全球集合预报成员大尺度背景的不确定性;集合平均结果在一定程度上修正了奥地利中部地区降水量级偏大和北部的降水空报情况;通过聚类后的集合系统得到的预报结果与控制预报相比表明,某些集合成员对降水预报有更好的指示意义;通过连续试验的对比,两个经聚类后得到的集合系统的降水预报评分均比业务预报的评分高,且集合离散度也较大;而成对和不成对的集合预报系统的预报结果基本相似。说明直接从全球大尺度集合预报模式成员中"降尺度"构造区域集合预报模式成员是可行有效的。Based on the ECMWF global Ensemble Prediction System （EPS）, three strategies were investigated using a clustering method which downscaled 50 ECMWF-EPS （ECMWF global Ensemble Prediction System） members to 16 representative members. The 16 representative members provide initial and boundary conditions for the ECMWF/ALADIN system. A case of rainfall over a 9-day during March 21-29,2008 was studied to validate the effect of different downscaling strategies on initial and boundary conditions in the regional ensemble forecast system. The results indicate that the selected 16 members by a cluster method have good representative in the 50 global ECMWF ensemble members. The ensemble mean improves the intensity and location of rainfall in the mid- dle area of Austria. The poststamp results could provide more effective information for the forecasters and the TS score for precipitation gets better records after the clustering process. The spread distribution is certainly improved. The symmetric positive-negative paired and unpaired ECMWF/ALADIN systems are more skillful than operational ensemble forecast system which doesn＇t include the clustering process. These also imply that the clustering method of downscaling from global ensemble forecast system is feasible.

关 键 词：集合预报 初边界不确定性 聚类法 

分 类 号：P456.7[天文地球—大气科学及气象学]