| 蒙特卡罗方法在降水评估中的应用研究 |
| |
| 引用本文： | 赵滨,李子良,张博.蒙特卡罗方法在降水评估中的应用研究[J].南京气象学院学报,2016,8(6):553-559. |
| |
| 作者姓名： | 赵滨 李子良 张博 |
| |
| 作者单位： | 国家气象中心, 北京, 100081;黑龙江黑河市气象局, 黑河, 164300;国家气象中心, 北京, 100081 |
| |
| 基金项目： | 国家自然科学基金青年基金（41305091）；中国气象局成都高原气象研究所基金（LPM201401）；公益性行业专项课题（GYHY201506002） |
| |
| 摘 要： | 预报检验重点关注预报与观测间的综合统计特征用以探讨模式预报性能，而统计显著性检验方法是衡量评估结论的重要指标，是判断预报效果改进与否的有效手段.当前诸多重要检验指标如降水技巧评分等由于不满足正态分布特征均难以采用简单的计算方式获得置信区间以衡量检验指标的误差特征，因此难以正确判断通过统计检验所获得的评估差异是真实反映模式预报效果差异或是由检验样本的不确定性所造成.蒙特卡罗方法可通过样本重构获取正态分布的统计样本从而有效地解决这一问题.采用2015年8月的T639模式及GRAPES全球预报模式24 h降水预报产品，使用中国区域2400站日降水资料作为实况，重点研究蒙特卡罗方法在统计显著性检验中的应用特征，分析不同蒙特卡罗重构次数对检验结果的收敛性.结果表明10 000次蒙特卡罗重构后统计指标可满足正态分布，而通过显著性检验分析后可明显区分预报系统间降水评分差异的统计特征.
|
| 关 键 词： | 蒙特卡罗方法 降水技巧评分 显著性检验 T639模式 GRAPES全球模式 |
| 收稿时间： | 2015/12/25 0:00:00 |
Application of Monte Carlo significance test in precipitation skill score |
| |
| ZHAO Bin,LI Ziliang and ZHANG Bo.Application of Monte Carlo significance test in precipitation skill score[J].Journal of Nanjing Institute of Meteorology,2016,8(6):553-559. |
| |
| Authors: | ZHAO Bin LI Ziliang and ZHANG Bo |
| |
| Institution: | National Meteorological Center, Beijing 100081;Heihe Weather Office of Heilongjiang Province, Heihe 164300;National Meteorological Center, Beijing 100081 |
| |
| Abstract: | Forecast verification involves exploring and summarizing the relationship between sets of forecast and observation and making comparisons between the performances of different forecasting systems.Statistical significance is one important aspect of measuring the absolute quality of verification results.It is an effective way to judge whether the performance improvement is statistically significant or just arisen by chance.For general meteorological forecast verification,some verification scores,such as precipitation skill scores,can hardly use the standard procedure for confidence interval to measures the difference in performance between different forecast systems.It is not possible to be sure that the apparent differences in skill scores are real and not just due to random fluctuations because of the data uncertainty.The Monte Carlo method is a numerical way to account for this.By resampling process,we can provide an adequate representation of the full underlying population which satisfies normal distribution by the verification samples of random variables.In this paper,some precipitation skill scores of GRAPES global forecast system and T639 models such as Threat Score and Bias Score are calculated from 1Aug to 31 Aug of 2015.The daily precipitation observation data are taken from 2 400 Chinese rain gauges.The Monte Carlo method is used for a statistical significance test and the convergence characteristics with different resampling times are also analyzed.Results show that Monte Carlo test using 10 000 test samples looks sufficient and a real model performance with significant improvement is provided. |
| |
| Keywords: | Monte Carlo method precipitation skill score significance test T639 mode GRAPES global mode |
|
| 点击此处可从《南京气象学院学报》浏览原始摘要信息 |
| 点击此处可从《南京气象学院学报》下载免费的PDF全文 |
