2007 GSA Denver Annual Meeting (28–31 October 2007)

Paper No. 12
Presentation Time: 4:55 PM

INVERSE HYDROLOGIC MODELING WITH GENETIC ALGORITHMS


LIU, Ganming, School of Earth Sciences, The Ohio State University, 125 S. Oval Mall, Columbus, OH 43210, SCHWARTZ, Franklin W., School of Earth Sciences, The Ohio State University, Columbus, OH 43210 and CROWE, Allan S., Environment Canada, National Water Research Institute, Canada Centre for Inland Waters, 867 Lakeshore Road, Burlington, ON L7R 4A6, Canada, liu.669@osu.edu

Parameter estimation with inverse techniques provides a powerful approach to model development. Certain limitations, however, can exist because of a limited number of sampling, an inability to conduct field experiments, or uncertainty in observations. Genetic algorithms (GAs) are robust, global, search techniques that automatically search an optimum through generations. Each population of independent parameters within a generation is generated through selection, crossover and mutation operators.

In this research, genetic algorithm has been coupled to a lake-watershed model (GA-LAKE) to optimize parameter values. A hypothetical example, in which the optimum error-free set of lake-level values was assumed known, was used initially to investigate the capability and performance of the GA. Various GA scenarios with different fitness functions and GA parameters (e.g. population size, crossover and mutation probabilities) were tested. The improved GA-LAKE program was then applied to calibrate the hydrologic parameters with real observation data.

GA-LAKE provided a parameter set that approached the global minimum error between simulated and measured data, i.e. lake level and lake water chemistry, separately or both together simultaneously as multi-objective problem. A comparison of these results with a trial-and-error scheme shows that the GA-LAKE could optimize the parameters more quickly and precisely, especially with a large number of parameters with large parametric spaces. Furthermore, by examining the optimization process, we found that GA-LAKE is also a powerful tool for identifying parameter sensitivity. Sensitive parameters always contribute greater difference to the fitness and reach optimum values in earlier generations than insensitive ones.