Calibration of a conceptual rainfall–runoff model using a genetic algorithm integrated with runoff estimation sensitivity to parameters

Abstract

This study proposes a parameter-calibration method (SA_GA) for the conceptual rainfall–runoff model using a real-value coding genetic algorithm (GA) which takes into account runoff estimation sensitivity to model parameters; this process is carried out using the standardized regression equation. The proposed SA_GA method treats the standardized values of model parameters as the real-value code and adopts a multinomial trial process with a probability of selecting genes for the crossover and mutation resulting from the runoff estimation sensitivity to the model parameters. A 19-parameter conceptual rainfall–runoff model, Sacramento Soil Moisture Accounting (SAC-SMA) model, and seven rainstorm events recorded in the Baj-Hang River watershed of South Taiwan are applied in the model development and validation. The results indicate that SA_GA is superior to a simple genetic algorithm (SGA) as regards the calculation of fitness values associated with the optimal parameters under various GA operators. In addition, by comparing the performance indices of estimated runoff with the calibrated optimal parameters by SA_GA and SGA with the different number of calibration rainstorm events, SA_GA can provide efficient and robust optimal parameters. These parameters not only estimate reliable and accurate runoff, but also capture the varying trends of discharge in time.