Bayesian Experimental Design For Parameter Estimation Of Groundwater Reactive Transport Models

Groundwater reactive transport modeling is a development tendency in groundwater numerical simulation.Some key parameters in groundwater reactive transport models are unable to be measured directly due to restrictions like techniques,budget or time.Thus,it is necessary to estimate unknown parameters by solving inference problem with easily obtained observations.On this occasion,selecting a monitoring scheme that could provide the biggest data worth for the estimation of reactive transport model parameters is critical to accurate model parameters estimation.In this study,from simple to complex,two synthetic reactive transport cases were used to illustrate the effectiveness of the proposed Bayesian experimental design and Markov chain-Monte Carlo(MCMC)method.Firstly,a reactive transport model describing the continuous degradation of chlorinated hydrocarbons in groundwater was constructed.The Bayesian experimental design method was used to identify the optimal monitoring scheme both in one-dimensional homogeneous and two-dimensional heterogeneous hydraulic conductivity field.Then Markov chain-Monte Carlo method was used with the observations obtained under the optimal scheme to estimate unknown parameters.The performance of Bayesian experimental design method was demonstrated by comparing the posterior marginal probability densities of unknown parameters and two dimensionless performance indicators RC and RP under the optimal monitoring scheme and other reference schemes.To further illustrate the applicability of Bayesian experimental design and parameter inversion methods in real complex reactive transport systems,the second case considered in this work is a groundwater tar oil biodegradation reactive transport at a filed site in Germany reported by Prommer et al(2009).In this case,in addition to simultaneous designing the optimal monitoring schemes in homogeneous and stratified heterogeneous hydraulic conductivity field,the influence of measurement types on monitoring network design and parameter estimation was also discussed.The results of this paper indicate that:(1)The Bayesian experimental design method is applicable in groundwater reactive transport models.With the proposed method,the optimal monitoring scheme with largest information can be designed.(2)In subsurface reactive transport models,the sensitivity of various types of observations to unknown parameters is different.When estimating parameters in reactive transport models,it is not always the case that the more observation types you use,the more accurate estimation results you could obtain.Adding the type of observations that provide little additional information could decrease the accuracy of parameter estimation.