Calibration of a semi-distributed rainfall-runoff model using global optimization strategies

The physically based, semi-distributed Regionalized Rainfall Runoff Model `R<sup>3</sup>' has been designed to simulate watershed behavior using information derived from observable watershed characteristics. While some of this information can be directly measured in the field, in practice other aspects may not be easy to obtain due to the lack of measurement techniques or nonobservable model parameters. The R<sup>3</sup> model therefore relies on calibration for the accurate definition of some of its model parameters. Previous attempts to estimate the 7 model parameters and the initial wetness were performed with the global gradient search algorithm `Globex', but the results proved to be unsatisfying. A recently developed evolution-based global search algorithm `Shuffled Complex Evolution' (SCE-UA) has been reported to be capable of localizing the global optimum even for `difficult' response surfaces. In this study, a comparison between these two algorithms was performed with historical data. The results showed the SCE-UA algorithm to be significantly superior to the Globex method in terms of efficiency and effectiveness.