设为首页
加入收藏
事件概率回归估计与降水等级预报
Regression Estimate of Event Possibility and Precipitation Categorical Forecast
-
摘要: 该文对比分析概率回归降水等级预报和回归降水等级预报的差异, 2007年秋季至2008年夏季全国平均检验结果表明:概率回归降水等级预报效果好于回归降水等级预报, 尤其是小雨预报, TS评分明显高于回归降水等级预报, 同预报偏差过大情况也有很大改善。进一步分析表明:回归降水等级预报方法在建立小雨预报方程的样本中, 少数较大降水量的样本方差占总方差的百分比过大, 导致预报方程中反映的预报量与预报因子的关系以少数大降水量样本为主, 是造成小雨预报空报过大的原因。与模式降水预报的对比分析表明:概率回归降水等级预报效果好于模式直接降水预报, 模式降水空报较大情况得到改善。Abstract: Objective precipitation forecast is a difficult problem in NWP products interpretation.Because of itscharacteristics, objective precipitation forecast is a categorical forecast rather than precipitation amountforecast.The differences between two kinds of categorical precipitation forecast are analyzed.One categorical forecast is based on probability regression.Its method is processing original precipitation to 0 and 1corresponding categories, and then developing forecast equations of different categories to calculate the criterions.In real forecasting, the categorical precipitation will be determined through the criterion and theprobability forecast of that category.The other forecast is based on regression, the method of which ispreprocessing original samples with value smaller than the threshold to category of 0, and then developingforecast equations and criterions.The experimental result from autumn of 2007 to summer of 2008 indicates that probability regressionprecipitation categorical forecast is better than regression precipitation categorical forecast.Especiallywhen forecasting light rain, the TS score averaged over China using probability regression method is higher than that of regression precipitation categorical forecast, the false alarm ratio is obviously smaller, andalso the forecast bias is closer to 1.Through the analysis of predictors and variance contribution of singlesample, the cause of these differences becomes obvious.In regression categorical forecast, the variancecontribution of a few heavy rain samples is too large.It results in the relation of predictors and precipitation mainly reflected those minority heavy rain samples.That is why the false alarm ratio of regression categorical forecast is too high.It can be shown in comparing analysis that the probability regression categorical precipitation forecast is better than direct model precipitation forecast and the situation that false alarmratio is too high is improved.
-
图 1 2007—2008年度全国平均降水预报TS评分
Fig. 1 TS score of precipitation forecast averaged over China during 2007—2008
图 2 2007—2008年度全国平均降水预报偏差
Fig. 2 Precipitation forecast bias averaged over China during 2007—2008
图 3 2008年6—8月全国平均降水预报 (a) TS评分, (b) 预报偏差
Fig. 3 TS score (a) and forecast bias (b) averaged over China during June to August of 2008
图 4 2008年7月23日00:00 24 h降水量 (a) 和2008年7月21日DMO (b)、S2(c)、S1(d) 36 h降水预报
Fig. 4 24-hour rainfall on 23 July 2008 (a) and corresponding 36-hour forecast on 21 July 2008 of DMO (b), Scheme 2 (c) and Scheme 1 (d)
表 1 S2预报因子统计
Table 1 Predictors statistic of Scheme 2
表 2 S1预报因子统计
Table 2 Predictor statistic of Scheme 1
表 3 北京夏季小雨预报方程样本数百分比及方差贡献
Table 3 Sample percentage and corresponding variance contribution of Beijing summer light rain forecast equation
-
[1] 刘还珠, 赵声蓉, 陆志善, 等.国家气象中心气象要素的客观预报———MOS系统.应用气象学报, 2004, 15(2): 181-191. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=20040223&flag=1 [2] 胡江林, 涂松柏, 冯光柳.基于人工神经网络的暴雨预报方法探讨.热带气象学报, 2003, 19(4): 423-428. http://www.cnki.com.cn/Article/CJFDTOTAL-RDQX200304009.htm [3] 刘爱鸣, 潘宁, 邹燕, 等.福建前汛期区域暴雨客观预报模型研究.应用气象学报, 2003, 14(4): 420-429. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=20030452&flag=1 [4] 赵声蓉.多模式温度集成预报.应用气象学报, 2005, 17(1): 52-58. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=20060109&flag=1 [5] 魏凤英.全国夏季降水区域动态权重集成预报试验.应用气象学报, 1999, 10(4): 402-409. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=19990489&flag=1 [6] 周家斌, 张海福, 杨桂英, 等.制作汛期降水集成预报的分区权重法.应用气象学报, 1999, 10(4), 428-435. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=19990488&flag=1 [7] 张万诚, 解明恩.奇异值分解方法对降水的预测试验.高原气象, 2002, 21(1): 103-107. http://www.cnki.com.cn/Article/CJFDTOTAL-GYQX200201016.htm [8] 刘还珠, 郝为, 林孔元, 等.基于智能计算的多模型气象综合预报∥刘还珠, 汤桂生.暴雨落区预报实用方法.北京:气象出版社, 2000:30-37. [9] 宋海鸥, 王永红, 顾善齐, 等.应用K指数和TOT指数制作江苏中期降水预报的试验.气象科学, 2002, 22(2): 242-246. http://www.cnki.com.cn/Article/CJFDTOTAL-QXKX200202015.htm [10] 陈力强, 韩秀君, 张立群.基于MM5模式的站点降水预报释用方法研究.气象科技, 2003, 31(5): 268-272. http://www.cnki.com.cn/Article/CJFDTOTAL-QXKJ200305002.htm [11] 赵声蓉, 裴海瑛.客观定量预报中降水的预处理.应用气象学报, 2007, 18(1): 21-28. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=20070104&flag=1 [12] 龚佃利, 王以琳, 谢考宪.山东飞机增雨降水区分级预报方法研究.应用气象学报, 2001, 12(增刊): 139-145. [13] 曹晓岗.利用T106数值预报产品作江西暴雨动态落区预报.江西气象科技, 1998, 21(1): 2-5. http://www.cnki.com.cn/Article/CJFDTOTAL-HXQO199801001.htm [14] 黄嘉佑.气象统计分析与预报方法.北京:气象出版社, 1990. [15] Mark S A.An overview of the National Weather Service' s centralized statistical quantitative precipitation forecast.J Hydrol, 2000, 239(9): 306-337. https://www.cabdirect.org/cabdirect/abstract/20013000759 -
下载:
计量
- 摘要浏览量: 3630
- HTML全文浏览量: 790
- PDF下载量: 1964
- 被引次数: 0
