Choosing various likelihood functions on uncertainty assessment in groundwater simulation-optimization model

Abstract

The main goal in this research is study of impacts of various likelihood functions on DREAM(zs) (Differential Evolution Adaptive Metropolis) method results in simulation-optimization model of aquifer. In this study, DREAM(zs) algorithm has been developed to study aquifer simulation-optimization model uncertainties. DREAM(zs) is used to investigate uncertainty of parameters of the simulation-optimization model in Isfahan-Barkhar aquifer, Isfehan province, Iran. This study is carried out on an aquifer simulation model of MODFLOW that is coupled with MOPSO (multi-objective particle swarm optimization) optimization. Three likelihood functions, L1, L2, and L3, are considered as informal and the remaining (L4 and L5) are represented as formal categories. Likelihood function L1 is Nash-Sutcliffe efficiency and L2 is based on minimum mean square error. L3 uses estimation of model error variance and L4 focuses on the relationship between the traditional least squares fitting and the Bayesian inference. In likelihood function L5 the serial dependence of residual errors is calculated using a first-order autoregressive model of the residuals. Results suggested that the parameters sensitivity depend on the likelihood function selection, and sensitivity of all parameters is not similar in different likelihood functions. MOPSO algorithm outputs indicated that likelihood function No. 5 has a higher speed in reaching convergence and this function also showed that objective functions had a better performance compared to the other likelihood functions.