基于分位点回归模型的洪水概率预报方法  

Flood probabilistic forecasting based on quantile regression method

作　　者：胡义明[1] 罗序义 梁忠民[1] 黄一昕 蒋晓蕾[1] HU Yiming;LUO Xuyi;LIANG Zhongmin;HUANG Yixin;JIANG Xiaolei(College of Hydrology and Water Resources,Hohai Uuniversity,Nanjing 210098,China)

机构地区：[1]河海大学水文水资源学院,南京210098

出　　处：《南水北调与水利科技（中英文）》2020年第5期1-12,共12页South-to-North Water Transfers and Water Science & Technology

基　　金：国家重点研发计划(2016YFC0402709,2016YFC0402707);国家自然科学基金(41730750,51709073)。

摘　　要：采用分位点回归模型分析洪水预报的不确定性,提供洪水预报倾向值(预报概率分布的中位数)和90%置信度的预报区间成果,实现了洪水概率预报。基于"精度-可靠性"联合评价指标对分位点回归模型计算的预报倾向值和预报区间成果进行了评估。在信江流域梅港站的应用结果表明:基于分位点回归模型提供的倾向值定值预报结果可进一步提升洪水预报的精度;同时该模型提供的90%预报区间结果具有较高的覆盖率(约90%)且离散度较小(小于0.20),表明预报区间以较窄的宽度包含了绝大多数的实测值,预报可靠性较强。The quantile regression model is used to analyze the uncertainty of flood forecasting.The results of preferred value(median of predicted probability distribution function)and 90%confidence interval of flood forecasting are provided to realize the forecast of flood probability.The performance of probabilistic forecasting obtained by the quantile regression model is evaluated using"accuracy-reliability"joint evaluation index.The application results of Meigang Station in the Xinjiang River Basin show that in term of prediction preferred value,the quantile regression model can further improve the accuracy of the flood forecasting.Simultaneously,the prediction interval results with 90%confidence level provided by the model have higher coverage(about 90%)and less dispersion(less than 0.20),which means that the narrow prediction interval contains most of the observation,and the reliability of the forecast interval is strong.

关 键 词：洪水概率预报 分位点回归模型 预报倾向值 预报区间 精度-可靠性评价 

分 类 号：TV213[水利工程—水文学及水资源]