水文模型两种不确定性研究方法的比较

水文模型结构本身的缺陷、模型输入输出误差、水文模型参数冗余及其复杂的非线性联系等,导致了流域水文模型的不确定性.基于贝叶斯理论的MCMC方法及GLUE方法近年来被广泛应用于流域水文模型的不确定性研究工作中.为比较上述2种模型不确定性分析方法的分析效果及其优劣,以位于汉江流域的牧马河流域作为研究对象,采用集总式概念性水文模型SMAR模型作为实验模型,推求其模型参数的不确定性及参数的后验分布.采用基于实测流量资料估计的置信区间可靠性作为评判标准,实验结果表明:就SMAR模型而言,MCMC方法能够更好地推求模型参数的后验分布.

Comparison between GLUE and MCMC methods for estimating hydrological model uncertainty

The uncertainties of hydrological model are caused by deficiencies in model structure, errors associated with input and output data, poorly defined boundary conditions and the complexity of non-linear relationship between model parameters. Both the MCMC (Markov Chain Monte Carlo) approach and GLUE (Generalized Likelihood Uncertainty Estimation) methodology based on Bayesian framework have been widely used over the past ten years to analyze and estimate predictive uncertainty in hydrological applications. The two methodologies mentioned above are compared for the capability to estimate the model uncertainty, with the SMAR, a lumped hydrologic model, chosen as the test model. The results for the Muma river, watershed are assessed by judging the dependability prediction confidence interval. It has been found that the MCMC approaches can estimate and derive the posterior joint probability distribution of the model parameters properly.