用统计集合方法制作全国汛期降水滚动预测试验  

Experiment of Running Prediction of Precipitation in Flood Period in China with Ensemble Analysis of Physical-Statistic Prediction

作　　者：毛炜峄[1,2] 陈颖[1] 白素琴[1] 李维京[2] 

机构地区：[1]新疆气候中心,乌鲁木齐830002 [2]中国气象局气候研究开放实验室,北京100081

出　　处：《气象》2011年第5期547-554,共8页Meteorological Monthly

基　　金：国家自然科学基金项目(41065006);国家科技支撑计划(2007BAC29B02)联合资助

摘　　要：以全国160站汛期(6—8月)降水量为预测量,以最新得到的74项环流特征量指数为因子,尝试制作全国160站汛期降水滚动预测。建立模型时考虑了预测量与环流特征量因子序列的显著线性变化趋势,以及预测量与环流因子之间的相关不稳定性,用"滑动相关—逐步回归—集合分析"预测方法,分别建立了2009年全国160站汛期降水量的物理统计集合分析预测模型,并进行了近10年独立样本预测试验分析。结果显示:(1)用物理统计集合分析预测方法,以最新得到的74项环流特征量指数为因子,实现了全国160站汛期降水逐月滚动预测,2009年以在5月份制作的滚动预测效果最好。(2)近10年预测试验的空间距平相关系数ACC、业务评分PS和异常级评分TS均高于国家气候中心近年汛期预测业务平均水平。经过不断改进思路和优化具体建模方案,该方法具有较高的业务应用潜力。Using precipitation data in flood period（from June to August）at 160 meteorological stations in China as predictands and 74 circulation characteristic indexes as predictors,a running prediction in flood period was conducted.The significant linear trend between predictands and circulation characteristic indexes was considered,as well as their correlation instability.Using running correlation,stepwise regression and ensemble analysis method,the multiple regression assembling prediction models of precipitation in flood period of 2009 at 160 stations were founded,and the independent swatch forecastings for the last 10 years were tested.It shows that the scores for ACC,PS and TS of the test all were better than those of operation of NCC in the recent years.The outputs through running correlation,stepwise regression and ensemble analysis method all have potential operational applications.

关 键 词：汛期降水 滚动预报 物理统计 预测试验 集合分析 

分 类 号：P333[天文地球—水文科学]