基于Morris和Sobol的水文模型参数敏感性分析

敏感性分析有助于识别模型相对敏感的重要参数，降低参数维度，为模型参数优化与不确定性分析提供支撑。该文以三水源新安江模型为研究模型，汉江上游高滩流域为研究对象，选用Nash-Sutcliffe效率系数DC和水量平衡系数WB及其线性组合作为目标函数，联合运用Morris筛选法与Sobol方法分析了新安江模型所选13个参数的敏感性。结果表明：流域蒸散发折算系数KC、自由水蓄水容量SM、壤中流出流系数KI/地下水出流系数KG、壤中流消退系数CI、地下水消退系数CG和地表径流消退系数CS属于敏感参数，其中KC为高敏感参数；不同目标函数下模型参数的敏感性分析结果有所差异，水量平衡系数对于KC更加敏感，Nash-Sutcliffe效率系数对于SM、KI/KG、CI、CG、CS更加敏感；部分模型参数（如SM）的交互效应超过其自身敏感度，主要以两两参数组合的二阶敏感性为主，证明参数之间的相关性也是造成参数敏感性的重要原因。Morris筛选法减少敏感性分析的参数数量，Sobol方法准确刻画了模型参数的敏感性特征，两者结合提高了新安江模型参数敏感性分析的可靠性与准确性。

Global Sensitivity Analysis of Hydrological Model Parameters Based on Morris and Sobol Methods

The emerging and popularity of distributed hydrological models promote the development of hydrology sciences meanwhile massive model parameters are introduced, emphasizing the necessity of screening out sensitive parameters. Sensitivity analysis（SA) helps recognize sensitive parameters, reduce the number of model parameters to be analyzed and facilitates parametric optimization and uncertainty estimation. In this study the Xin’anjiang model was applied to Gaotan catchment located in the upstream of Han River basin and Nash-Sutcliffe efficiency（DC), water balance coefficient（WB) and their combination served as objective functions to evaluate the accuracy of daily streamflow simulation. A qualitative SA approach, Morris method was first used to screen out sensitive parameters and reduce the dimensionality of model parameters. The Sobol method was then adopted to quantify degrees of sensitivity of sensitive parameters quantitatively. Both methods are implemented for three different objective functions. The result showed that among 13 selected model parameters, runoff generation parameter KC, runoff separation parameters SM and KI/ KG, runoff routing parameters CI,CG and CS are identified as sensitive parameters, among which KC is most sensitive. Furthermore, the selection of objective functions to be evaluated has significant impact on the determination of sensitive parameters: water balance coefficient is most sensitive to KC while Nash-Sutcliffe efficiency more sensitive to other sensitive parameters, especially CS. Further investigation into sensitive parameters revealed that some parameters like SM have higher interaction sensitivities compared to first-order sensitivities, indicating that the interactions between model parameters should be regarded as an unnegligible factor of parameters sensitivity. The Morris method facilitates the reduction of dimensionalities of model parameters while the Sobol method precisely characterizes the features of parametric sensitivities and provides more information on the sensitivities. The integration of both methods is demonstrated to enhance the reliability and accuracy of sensitivity analysis.