Research On Hydrological Model Uncertainty Based On Baysian Theory

Hydrological models are extensively used for simulating runoff dynamics and water balance at various scales and regions,which aggregate at some level of complexly hydrological cycle,spatially distributed vegetation and subsurface properties into much simpler mathematical equations.However,hydrological models often include significant uncertainties that are resulting from parameter,model structure and input data.Especially,hydrological parameters can merely be inferred by calibration to a historical record of runoff data.Baysesian theory has found widespread application in hydrological model because it is capable of assessing model uncertainty.Bayesian uncertainty analysis techniques assume that hydrological model parameters are random variables with probability distribution,which provides the theoretical basis for hydrologic simulation and prediction.In this study,the methods based on Bayesian theory have been developed for hydrological model calibration and uncertainty analysis.A series of uncertainty analysis method have been conducted:(1)A Bayesian-based multilevel factorial analysis(BMFA)method is developed to assess parameter uncertainties and their effects on hydrological model responses;(2)The uncertainty analysis method under Bayesian framework,which consider heteroscedasticity of model errors,is proposed for analyzing and comparing input data and parameter uncertainty of hydrological model.The proposed method has been applied to SWAT model of Jinghe River watershed that is located in the Loess Plateau,China.The results revealed that uncertainty analysis method based on Bayesian theory can well identify uncertainty of SWAT model.The proposed BMFA method not only provides an estimation of the parameter uncertainty,but also provides an evaluation of how much each parameter and their interaction impact on model responses.Besides,considering rainfall error model in a Bayesian framework can lead to more realistic parameter values,better representation of prediction uncertainty intervals.Finally,these results will facilitate the research of Beyesian theory in the hydrological model uncertainty analysis as well as the understanding of relationship between SWAT model and hydrological characteristics in Jinghe River watershed.