基于深度学习与HEC-HMS模型的小流域暴雨洪水耦合预报  

Coupled forecasting of rainstorms and floods in small watershed based onDeep Learning and HEC-HMS Model

作　　者：刘万 谢帅 钟德钰[3] 王永强[2] 包淑萍 朱旭东 LIU Wan;XIE Shuai;ZHONG Deyu;WANG Yongqiang;BAO Shuping;ZHU Xudong(School of Civil and Hydraulic Engineering,Huazhong University of Science and Technology,Wuhan 430074,China;Changjiang River Scientific Research Institute,Wuhan 430010,China;State Key Laboratory of Hydroscience and Engineering,Tsinghua University,Beijing 100084,China;Ningxia Hydrological and Water Resources Monitoring and Early Warning Center,Yinchuan 750001,China)

机构地区：[1]华中科技大学土木与水利工程学院,湖北武汉430074 [2]长江科学院水资源综合利用研究所,湖北武汉430010 [3]清华大学水沙科学与水利水电工程国家重点实验室,北京100084 [4]宁夏水文水资源监测预警中心,宁夏银川750001

出　　处：《水利学报》2025年第3期364-374,共11页Journal of Hydraulic Engineering

基　　金：国家自然科学基金项目(U2040211,U2040212,52379011);湖北省自然科学基金(2024AFA011,2023AFB039);中央级基本科研业务费(CKS20241021?SZ);宁夏重点研发项目(2020BCF01002)。

摘　　要：小流域暴雨洪水产汇流过程快,常规洪水监测预报的预见期较短,难以为防汛救灾提供有效支撑。本研究中构建了基于深度学习降雨预报的暴雨洪水耦合预报框架,通过以短临降雨预报驱动水文模型的方式延长洪水预报预见期,对小流域防洪具有重要意义。在短临降雨预报中,本文提出一种基于多预报因子输入的深度学习短临降雨预报模型(RNMCW),选取能有效反映水汽含量和水汽输送的组合反射率和雷达反演风场作为输入,以深度学习方法提取时空特征,从而预报流域内降雨。结果表明,该方法在各站降雨预报的相关系数均高于0.75,相比传统光流法提高了5%左右。之后用RNMCW预测的降雨对HEC-HMS模型进行参数率定,并以降雨预报结果作为水文模型的输入构建暴雨洪水耦合预报模型。相比HEC-HMS模型,耦合模型对2016年洪水的洪峰预测误差降低了2.17%,对20180722洪水预测的纳什系数提升了0.002。模型能够有效预报出洪水过程,并最多延长有效预见期2 h,从而提升洪水预报效果,为实际应用提供更好的支撑。The rainfall runoff process in small watershed is so quick that the forecast lead time of conventional flood monitoring and forecasting is short,which is difficult to provide effective support for flood prevention and disaster relief.In this study,a coupled forecasting framework for rainstorm and flood forecasting is established.In this framework,the hydrological model is driven by the rainfall nowcasting to prolong the forecast lead time of flooding,which is of great importance for flood prevention in small watershed.In the rainfall nowcasting method,this paper proposes a Rainfall Nowcasting Method with Combined Reflectance and Wind Field as Inputs(RNMCW),the combined reflection and radar-retrieved wind field that can effectively reflect the water vapor content and water vapor transportation are selected as inputs,and the deep learning model is proposed to extract the temporal and spatial characteristics of inputs to forecast the rainfall in the basin.The results show that the correlation coefficient of the rainfall nowcasting method is higher than 0.75 at each station,which is about 5%higher than that of the conventional optical flow method.Then,the parameters of the HEC-HMS model are calibrated with the rainfall predicted by RNMCW,and the rainfall forecast results are used as the input of the hydrological model to build a storm and flood coupling forecast model.Compared with the HEC-HMS model,the coupled model reduces the flood peak prediction error of 2016 flood by 2.17%,and increases the Nash coefficient of 20180722 flood prediction by 0.002.The model can effectively forecast the flood process and extend the effective prediction period by up to 2 hours,thus significantly improving the flood prediction effect and providing better support for practical application.

关 键 词：短临降雨预报 深度学习 雷达反演风场 暴雨洪水耦合预报 

分 类 号：P338[天文地球—水文科学] P457.6[水利工程—水文学及水资源]