Calibration of the soil moisture accounting model using the adaptive random search algorithm

Abstract

Random search techniques are being applied to a variety of non-linear parameter estimation problems. Random search for global optimization has the potential to overcome many of the problems associated with direct or pattern search techniques. In this research, an adaptive random search algorithm was applied to a conceptual rainfall-runoff model to study the efficiency of the algorithm in locating an optimum set of model parameters. The goal of the study was to determine how changes in internal algorithm control variables and objective functions affected the efficiency of the algorithm. Results indicated that the value of internal control variables did not have a strong influence on algorithm efficiency. Neither objective function gave demonstrably better results in calibration runs. Variability in results due to the random number seed was observed. Recommendations for further research are presented.