The Shuffled Complex Evolution (SCE-UA)

method has been widely applied for calibration of rainfall-runoff models and has been shown to be robust and efficient search algorithm. In spite of its superiority, since many commonly used rainfall-runoff models have large simulation times, the use of model calibrating algorithms may become impractical due to the high computational time involved. This would necessitate the use of superior parallel computing technologies in the calibrating algorithms, with the aim of reducing the computational times. This study aims to parallelize the SCE-UA to accelerate the model calibrating process. In the field of evolutionary algorithms (EAs), the use of parallel computing within the standard Genetic Algorithm (GA) has been researched into for the past two to three decades. There are three main parallel paradigms in evolutionary algorithms: the master-slave model, the diffusion model and the multi-population (or island) model. In this study, the master-slave model (see Figure 1) is adopted, since it is the simplest potential parallelization strategy, which can be easily implemented on a cluster of PCs linked by a Local Area Network (LAN). Moreover, in calibrating rainfall-runoff models, since the model simulation times are significantly greater than the communication times (between master and slaves), very good speedups are possible using master-slave models