基于贝叶斯模型平均法的分布式水文模型集合模拟

Distributed Hydrological Model Ensemble Simulation Based on Bayesian Model Average Method

基于分布式水文模型的研究现状,分析了分布式水文模型不确定性的影响,以暴雨洪水管理模型(SWMM)为例,介绍了过程集合运用于分布式水文模型以提高模型模拟可靠度的方法,并利用这一方法模拟深圳市龙岗区的两场降雨过程。结果表明,采用纳什效率系数(NS)衡量模型的模拟结果准确度时,选择不同的产流模型会有较大差异,在2018年8月11日场次降雨中,采用Horton模型、Green-Ampt模型、基于贝叶斯模型平均法的集合模拟时NS值分别为0.849、0.834、0.855;在2018年8月15日场次降雨中,采用上述3种模拟方法时NS值分别为0.875、0.879、0.891。集合模拟结果明显优于单一模型,集合模拟结果与实测值的“契合程度”较单一模型有明显提升。

This paper analyzed the influence of uncertainty of distributed hydrological model based on its research advances.The storm water management model(SWMM)was exemplified to illustrate the method of applying process ensemble to distributed hydrological model to improve the simulation reliability,and two rainfall processes in Longgang District of Shenzhen City were simulated by using this method.When Nash-Sutcliffe efficiency coefficient(NS)was used to measure the accuracy of simulation results,there were great differences between different runoff yield models.In the rainfall event on August 11,2018,the NS values were 0.849,0.834 and 0.855 when Horton model,Green-Ampt model and the ensemble simulation based on Bayesian model average method were adopted,respectively.In the rainfall event on August 15,2018,the NS values of the above three simulation methods were 0.875,0.879 and 0.891,respectively.The simulation results of the ensemble model were significantly better than that of the single model,and the former had significantly improved the“degree of fit”compared with the latter.