Using Genetic Algorithms on Groundwater Modeling Problems in a Consulting Setting

This paper presents a practical application for writing and applying simple genetic algorithms (GAs) for the common groundwater flow model, MODFLOW. The method employed by GAs is derived from the driving forces of evolution in the natural world. They employ functions that mimic natural evolutionary processes including selection, mutation, and genetic crossover. A GA solves mathematical problems where a desired outcome to the problem is defined (for example, calibration targets or remediation goals), but the inputs needed to arrive at this outcome are unknown. Our paper includes an introduction to genetic algorithms, the pseudocode of our genetic algorithm for MODFLOW, and the results of an experiential application. Due to the lack of commercially available GAs for MODFLOW, we coded a simple algorithm in Visual Basic Script and applied it to an example model. In the example model, the GA was used to conduct parameter estimation on a MODFLOW model of a river basin in New England that we had previously developed and calibrated in our practice. The calibration target used was net groundwater flow into the river. Four model input parameters were selected as chromosomes for the GA to act on: recharge, river conductance, and two general head boundaries. An initial population of 100 models was developed by varying the value of the gene parameters. The GA ran a MODFLOW simulation for each member of the population, extracted each output file, and established the error of each model from the calibration target. It then evolved the entire population of models towards the calibration target. The GA converged on a single set of input parameter that established best-fit values for all of the chromosome parameters. Genetic algorithms provide a practical alternative to trial-and-error and automated statistical calibration procedures, and can also be used for optimization