Research On Parameter Uncertainty Analysis Method For Water Quality Model With High-Dimensional Parameter Space

Water quality models with high-dimensional parameter space have been widely used in water environmental management.Parameter uncertainty analysis of these models is extremely significate since it can reduce artificial errors and the uncertainties,improving scientificity of decision making.However,the high-dimensionality,large amount of searches and numerous local optima of these models brought challenges to parameter uncertainty analysis.In this study,based on global sensitivity analysis and multi-chains MCMC,a method for parameter uncertainty analysis of water quality models with high dimensional parameter space was established.Using EFDC as typical water quality model with high-dimensional parameter space,the feasibility of this method was analysis though example test.The allowable pollutant load calculation in Miyun Reservoir was selected as a case study,to evaluate the effect of parameter uncertainty analysis of water quality models with high-dimensional parameter space.Then,suggestions for reducing uncertainty of parameter and allowable pollutant load were given.In the example test,we found that the results of global sensitivity analysis by Morris method,variance decomposition and standard regression analysis were different.For some variables like Bc,the sensitive parameter identified by these methods were not the same.For some variables like TOC,the sensitive parameter identified by these methods were the same,but the ranking of sensitivity were different.It is shown that using weighted average of these methods make more sense.Then using DREAM as the typical multi-chains MCMC for uncertainty analysis,the result showed the algorithm traversed the space and converged after 750 generations.The “true value” located in regions with high probability density of posteriori distributions or joint-posteriori distributions of the parameters.It is showed that DREAM is feasible for uncertainty analysis of water quality models with high-dimensional parameter spaces.Finally,the method was applied to evaluate the effect of parameter uncertainty analysis on allowable pollutant load of Miyun Reservoir.One hundred samples of parameter sets were generated from the 95% confidenceinterval of the joint-posteriori distribution of the parameters.Under single pollutant reduction scenario,the ranges of allowable loads were 551.6-977.6t·y-1,16.4-25.8 t·y-1,and 390.2-1560.9 t·y-1 for TOC,TP and TN,respectively.The relative ranges of TOC,TP,and TN were-36.5~12.6%,-24.6~18.9%,and-33.8~164.7%,respectively.The relative ranges exceed 10% obviously,which is used as margin of safety ratio in most allowable pollutant load calculation.And the results of joint pollutant reduction scenario are similar with the single pollutant reduction scenario.