首页|基于聚类分区的中国夏季降水预测模型

基于聚类分区的中国夏季降水预测模型

扫码查看
原文链接
  • 国家科技期刊平台
  • NETL
  • NSTL
  • 万方数据
  • 维普
文章基于近邻传播客观聚类方法对中国夏季降水进行了气候分区，以中国不同分区的夏季降水为预测对象，使用前期的海温和海平面气压场为预测因子，利用图像标签算法提取高相关封闭区域的预测因子信息，结合最小二乘回归法建立预测模型。采用 Ps 评分、距平符号一致率和距平相关系数三种评分方法检验了该预测模型，比较了四种不同的因子配置方案的预测能力。研究结果表明，利用冬春季海温的演变特征结合海平面气压的年际变化为预测因子的分区预测模型效果较好，在1982—2009年期间的平均交叉检验平均 Ps 得分为81．4，距平符号一致率为63％，距平相关系数为0．35，2010—2014期间的独立样本预测检验的平均评分分别为77．1，58％和0．19，且逐年回报效果较为稳定，表明该方法对中国夏季降水有较好的预测效果。研究结果显示，该预测模型能较好地预测出2014年中国夏季降水南多北少的分布特征。
Summer Precipitation Prediction Models Based on the Clustering Regionalization in China
Based on affinity propagation clustering method,summer precipitation is regionalized over Chi-na.Summer precipitation in different regions serves as predicted objects,and preceding sea surface tem-perature and sea level pressure are selected to be predictors.By methods of image labeling algorithm,con-secutive high correlation areas are extracted to determine the predictors.Least squares regression method is used to construct a model to predict summer precipitation in different regions.Different scoring methods including Ps,the anomaly sign sameness rate and anomaly correlation coefficient are used to validate the skills of prediction model in four different factor combination schemes.The results show that the model performs best when sea surface temperature and sea level pressure in winter and spring are together consid-ered as factors.The averaged Ps score of cross validation is 81.4 from 1982 to 2009,with the anomaly sameness rate 63% and anomaly correlation coefficient 0.35.The retrospective forecast verification shows that the averaged scores from 2010 to 2014 are 77.1,58% and 0.19,respectively.The reforecast skills are relatively stable every year,which means that the method has a good ability to predict summer precipitati-on in China.Moreover,it succeeds to predict the spatial characteristic of southern flood and northern drought in China in summer 2014.

affinity propagation clusteringregionalizationsummer precipitation predictionimage labe-ling algorithm

杜良敏、柯宗建、刘长征、肖莺、刘绿柳

展开 >

武汉区域气候中心，武汉 430074

国家气候中心，北京 100081

近邻传播聚类 分区 夏季降水预测 图形标签算法

公益性行业(气象)科研专项公益性行业(气象)科研专项国家重大科学研究计划国家自然科学基金

GYHY201306024GYHY201306033和 GYHY2014060222012CB95590241005051和41205039

2016

10.7519/j.issn.1000-0526.2016.01.011

气象
国家气象中心

气象

CSTPCDCSCD北大核心
影响因子：2.337
ISSN：1000-0526
年,卷(期)：2016.(1)
  • 12
  • 27