基于统计理论方法的水文模型参数敏感性分析

参数敏感性分析是模型不确定性量化的重要环节,有助于有效识别关键参数,减少参数的不确定性影响,进而提高参数优化效率。利用Morris筛选方法定性识别相对重要参数,耦合方差分解的Sobol方法和统计理论的响应曲面模型构建一种新的定量敏感性分析方法——RSMSobol方法。以长江支流沿渡河流域的日降雨径流过程模拟为例,系统分析4种不同目标函数响应条件下新安江模型的参数敏感性。结果表明Morris方法和RSMSobol方法的集成应用极大地提高了全局敏感性分析的效率,Morris定性筛选结果为定量评估减少了模型参数维数,采用代理模型技术的RSMSobol方法减少了模型的计算消耗。

Sensitivity analysis of hydrological model parameters using a statistical theory approach

The sensitivity analysis is a key step in model uncertainty quantification.And it can identify the dominant parameters,reduce the model uncertainty,and enhance the model optimization efficiency.In order to make quantitative global sensitivity analysis(GSA) more tractable,the Morris screening method is used to qualitatively assess a model first.Then,the response surface methodology(RSM) based on the statistical theory will be applied to construct a surrogate model,and to integrate with the variance-based Sobol’ method to establishing a new method,named as the RSMSobol method.The new method is tested on the Yanduhe basin using the Xin’anjiang model with daily precipitation data and hydrographs.The sensitivity analysis is conducted for four different objective functions.The results demonstrate that the new integrated qualitative and quantitative method can improve the efficiency of the sensitivity analysis,in which the Morris qualitative method can decrease the number of parameters by 50% for the next round of the quantitative analysis.The RSMSobol method can improve the computational cost.