Aquifer parameter estimation under transient and steady-state conditions using genetic algorithms

A parameter estimation method is developed for both transient and steady-state flow models based on genetic algorithms (GAs), one of the increasingly popular global optimization methods. Since GA is independent of the nature of the simulation model and the form of the objective function, any flow simulation code can be used for the forward modelling. The US Geological Survey modular flow simulator MODFLOW is used in this study. The objective function in the GAbased method can be highly complex and nonlinear, and various types of constraints can be readily incorporated. A two-dimensional model is investigated in this study, and theoretical observation data are obtained at selected model nodes by running the forward model with true parameters. Gaussian random noises are added to the theoretical observation data to represent the measurement errors. The results from a series of numerical experiments and a comparison with the GaussNewton method indicate that the GA-based method works well for both noise-free and noisy data. However, a large number of forward simulation runs may be required for convergence of the GA solution.