基于MODE方法的日本细网格模式降水预报的诊断分析  

Diagnostic Analysis of Precipitation Forecasting from Japan Thin-Grid Model based on MODE

作　　者：薛春芳[1,2] 潘留杰[3] 

机构地区：[1]兰州大学大气科学学院/甘肃省干旱气候变化与减灾重点实验室,兰州730000 [2]陕西省气象局,西安710014 [3]陕西省气象台,西安710014

出　　处：《高原气象》2016年第2期406-418,共13页Plateau Meteorology

基　　金：陕西省自然科学基金(2015JM4140);中国气象局预报员专项(CMAYBY2014-070);公益性行业(气象)科研专项(GYHY201306005)

摘　　要：利用CMORPH卫星与中国30000余个自动站逐时降水融合资料,基于面向对象模式诊断分析MODE方法研究了降水对象的客观表现以及日本细网格模式对2012-2013年中国34个降水个例的预报能力,并与传统TS、ETS检验方法进行了对比分析。结果表明:模式整体上能较好地捕获降水对象,但不同卷积半径下对象的个数及特征差异明显;模式对局地性或弱降水事件预报偏多(偏大),对区域性或强降水事件预报偏少(偏弱);模式较好地反映不同卷积半径下,降水对象在10 mm左右识别对象的面积、数量的变化特性,不足之处在于模式的固定阈值更为集中,识别对象的数量减少幅度偏快;采用较大半径进行卷积,有助于强降水区域的形态判定,在不同量级降水情况下存在提高收益函数的合适的卷积半径;中、低纬降水预报的总收益整体优于高纬,但不同属性呈现较大差异,具体而言,高纬地区降水强度预报更为合理,而低纬降水面积预报优于高纬;与传统TS、ETS检验相比,MODE方法受气候概率影响较小,其检验结果更加客观、多样。Using CMORPH（NOAA Climate Prediction Center Morphing Method）and fusion datas of hourly precipitation from more than 30 thousand automatic stations,based on MODE（Method for Object-Based Diagnostic Evaluation）object matching technology,the forecasting ability of Japan high resolution model of 34 precipitation cases in China from 2012 to 2013 was studied and compared with the traditional TS and ETS skill scores. The result shows that：first,the model can generally forecast the rainfall,but the number and characteristics of objects significant differences in different convolution radius. Second,the model gives much forecast of local or weak precipitation cases and limited forecast of regional or heavy precipitation. Third,the model can better tell variation characteristics of the area and the quantity when different convolution radii exist and the rainfall is about 10 mm. The disadvantage of the model is that the fixed threshold is much more centralized and the number of identified cases decrease quickly. If bigger radius are used to convolution,the model can identify the pattern better in heavy rainfall regions. There are possibilities to exist a proper convolution radius to improve total interest in different cases of different magnitude of rainfall. Fourth,the forecasting reward generally is bigger in the low-and mid-latitude areas than in high-latitude regions. However,the model shows no stable abilities to forecast different aspects of the rainfall. Specifically speaking,forecasting rainfall intensity in high-latitudes regions is more reasonable,while in low latitudes forecasting the rainfall area is superior to that in high latitudes regions.Compared with the traditional TS,ETS verification,MODE method is less affected by climate probability,the verify results are more objective and diversity.

关 键 词：细网格模式 MODE方法 卷积 技巧评分 

分 类 号：P456.7[天文地球—大气科学及气象学]