Calibration of Conceptual Stormwater Runoff Models Using the Genetic Algorithm: A Preliminary Study

In this paper, a publicly available genetic algorithm (GA) is used to calibrate conceptual runoff models of varying degrees of complexity, using storm hydrograph data from the 200-km² Bushkill Creek watershed, Northampton County, Pennsylvania. The simplest model tested is a linear lumped storage model, with two surface runoff parameters, one subsurface runoff parameter, and no routing. The most complex model uses coupled storage reservoirs (for soil moisture and groundwater storage), includes routing, contains nonlinear surface and subsurface runoff, and has seven parameters. Other model formulations tested fall between these two extremes. The coefficient of determination between the modeled and observed time series is used as the objective function for calibration. The GA calibrated all of the models tested in less than 15 sec using a PentiumIII computer with a 500 MHz processor. In terms of goodness of fit, the addition of routing gave the most significant improvement over the simplest model (r² of 0.92 vs. 0.42), with nonlinear runoff and baseflow generation providing modest improvement (r² of 0.96). The results suggest that a low-dimensional modeling approach will be sufficient to simulate streamflow in the watershed.