Selection of optimum model structure based on reliability of aquifer parameters using genetic algorithm

This study deals with the identification of a suitable aquifer model structure from a range of alternate models using the zonation method of parameterization. A heterogeneous confined aquifer system involving aquitard recharge and pumping wells with a set of boundary conditions is chosen for the present analysis. Galerkin's finite element approach is used for computing the groundwater head distribution in the aquifer system. Genetic algorithm (G A) optimization technique is used to compute the system parameters (transmissivity distribution) within the nine zones of the aquifer. A number of alternate initial models starting from homogeneous to more complex model structures are considered. Inverse problem is solved for the different initial model structures using weighted least square performance criterion. The sum of squared head residual is used for evaluating the model performance, whereas the covariance of estimated parameters (which is a measure of parameter uncertainty) is used for measuring the degree of instability. All the initial models are ranked according to model performance and parameter uncertainty. The best model is selected based upon reliability analysis performed on the two most suitable model structures. It is observed that with the limited field data, an increase in the number of zones (to represent aquifer heterogeneity) may seriously decrease the reliability of identified parameters. The study suggests that reliability analysis of estimated parameters should be a part of an inverse aquifer modeling. Finally, the G A approach is applied to a real aquifer problem and inverse model results are compared with the field values.