Comparison of genetic algorithms and shuffled complex evolution approach for calibrating distributed rainfall-runoff model

Three methods, Shuffled Complex Evolution (SCE), Simple Genetic Algorithm (SGA) and Micro-Genetic Algorithm (μGA), are applied in parameter calibration of a grid-based distributed rainfall-runoff model (GBDM) and compared by their performances. Ten and four historical storm events in the Yan-Shui Creek catchment, Taiwan, provide the database for model calibration and verification, respectively. The study reveals that the SCE, SGA and μGA have close calibration results, and none of them are superior with respect to all the performance measures, i.e. the errors of time to peak, peak discharge and the total runoff volume, etc. The performances of the GBDM for the verification events are slightly worse than those in the calibration events, but still quite satisfactory. Among the three methods, the SCE seems to be more robust than the other two approaches because of the smallest influence of different initial random number seeds on calibrated model parameters, and has the best performance of verification with a relatively small number of calibration events.