Study On The Uncertainty Of Mass Transport In The Groundwater Based On Bayesian Theory

The study on uncertainty of groundwater mass transport modeling is a researchhot spot in the groundwater system modeling. Bayesian inference theory andstochastic simulation techniques were combined to respectively study the impact ofthe marcodispersion and the different prior distributions on the uncertainty ofgroundwater flow and solute transport modeling. MH-MCMC Algorithm in Bayesianinference method was used for parameters uncertainty identification in groundwatersystem modeling; stochastic simulation techniques were used to generate the spatialdistribution field of hydraulic conductivity. The research results showed as follows:The uncertainty of hydraulic conductivity can cause the changes of other inputparameters. The macrodispersion, one of the important parameter of groundwatermass transport model, can be viewed as a parameter reflecting the spatial variabilityof hydraulic conductivity. The results from the impact of the heterogeneity ofhydraulic conductivity and marcodispersion on groundwater flow and mass transportmodeling showed that the inferred parametric posterior probability distribution wasupdated by Bayesian approach. The posterior distributions were not obeyed touniform distributions. Parametric initial iterative concussion did not affect theconvergence of the posterior distribution of the marcodispersion. As to the relativecontribution of stochastic uncertainty and parametric uncertainty to the overallpredictive uncertainty of hydraulic head and mass concentration distribution, thenumerical experiment in this paper indicated that parametric uncertainty was a littlemore important than stochastic uncertainty for the predictive uncertainty of hydraulichead. However, when the uncertainty of hydraulic head as well as macrodispersionwas transported to mass transport model, a much bigger contribution of stochasticuncertainty was observed. Thus, it was difficult to draw conclusion on the importanceof stochastic uncertainty and parametric uncertainty, as it depends on the number andthe location of measurements, the prior specified and etc.The results from the impact of two kinds of prior distributins (uniform priordistribution and Gaussian prior dstribution) on groundwater flow and mass transportmodeling indicated that the prior knowledge could affect the posterior distributions ofparameters. In the two prior knowledge scenarios, all parametric posterior probabilitydistributions were updated. When the Gaussian prior distribution was adopted, there was a better convergence. The selection of the prior distribution could impact the headand the mass concentration. The uncertainty of the head and the mass concentrationfields was smaller when the Gaussian distribution was adopted. The aboveconsideration may lead to the conclusion that the Gaussian prior distribution waspreferred. However, it was difficult to draw conclusions about the relative influenceof parameter prior distribution, as it depended on the location, number of themeasurements, and the methods to reflect the heterogeneity of hydraulic conductivity.