Analysis of parameter uncertainty in semi-distributed hydrological models using bootstrap method: a case study of SWAT model applied to Yingluoxia watershed in northwest China.
Analysis of parameter uncertainty in semi-distributed hydrological models using bootstrap method: a case study of SWAT model applied to Yingluoxia watershed in northwest China.
复制标题
DOI：
10.1016/j.jhydrol.2010.01.025
复制
发表时间：
2010-05
影响因子：
6.4
通讯作者：
Zhanling Li;Q. Shao;Zongxue Xu;Xitian Cai
中科院分区：
文献类型：
--
作者：
Zhanling Li;Q. Shao;Zongxue Xu;Xitian Cai
水文模拟中的不确定性问题因其对预测乃至决策的重要影响而备受关注。模型参数的不确定性是水文模拟中的主要不确定性来源之一。以黑河上游英落峡小流域为例，采用自助法对SWAT模型中参数的不确定性进行量化。Bootstrap方法是一种模拟参数分布的非参数方法。九个敏感的聚集参数进行了研究。Bootstrap方法的结果表明，9个边际分布中有6个非正态分布，每个参数都有自己的不确定范围。进一步研究了参数不确定性对仿真结果的影响，结果表明，参数不确定性虽然是不确定性的重要来源之一，但其对仿真不确定性的贡献相对较小。在校准和验证期间，只有12-13%的观测径流数据落在95%的模拟置信区间内。为了更好地理解自助法的适用性，常用的贝叶斯方法也进行了比较研究。结果表明，两种方法得到的模拟结果在95%置信区间内的观测值百分比和参数的不确定度范围上都是近似的，虽然贝叶斯方法得到的不确定度范围比自助法略窄，这可能是由于贝叶斯方法中采用的MCMC（马尔可夫链蒙特卡罗）模拟中的参数之间的相关性结构。这两种方法的计算效率也是相当的。
Much attention has been paid to uncertainty issues in hydrological modelling due to their great effects on prediction and further on decision-making. The uncertainty of model parameters is one of the major uncertainty sources in hydrological modelling. The aim of this study is to quantify the parameter uncertainty in Soil and Water Assessment Tool (SWAT) model using bootstrap method with application to Yingluoxia watershed located in the upper reaches of Heihe River basin. Bootstrap method is a nonparametric technique for simulating the parameter distribution. Nine sensitive aggregate parameters are investigated. The results from bootstrap method show that six of the nine marginal distributions are not normally distributed and each parameter has its own uncertainty range. Further investigation about the effects of parameter uncertainty on simulation results shows that although the parameter uncertainty is one of the important sources of uncertainties, its contribution to simulation uncertainty is relatively small. Only 12–13% of the observed runoff data fall inside the 95% simulation confidence intervals in the calibration and validation periods. For a better understanding of the applicability of bootstrap method, the commonly used Bayesian approach is also investigated for comparison. Results show that the approximate results are obtained from both methods, not only in the percentage of observations falling inside the 95% confidence interval of simulations, but also in the uncertainty range of parameters, although the range obtained from Bayesian method is slightly narrower than that from bootstrap method, possibly due to the correlation structure amongst parameters in the MCMC (Markov Chain Monte Carlo) simulation employed in Bayesian method. The computational efficiencies of both methods presented are comparable as well.