Comparison of particle swarm optimization and shuffle complex evolution for auto-calibration of hourly tank model's parameters

The famous Hydrological Tank Model is always preferred for runoff forecasting. This main reason is Tank Model not only simple in term of its structures, but able to forecast runoff accurately using only rainfall and runoff data. However, much time and effort are required to calibrate a large numbers of parameters in the model for obtaining better results through trial-and-error procedure. Therefore, there is an urgent need to develop an auto-calibration method. Two types of global optimization methods (GOMs), named as Particle Swarm Optimization (PSO) and Shuffle Complex Evolution (SCE) are selected. The selected study area is Bedup basin, Samarahan, Sarawak, Malaysia. Input data used for model calibration are hourly rainfall and runoff only. The accuracy of the simulation results are measured using Coefficient of Correlation (R) and Nash-Sutcliffe Coefficient (E2). The robustness of the model parameters obtained are further analyzed with boxplots analysis. Peak errors are also evaluated to determine the difference between the observed and simulated peaks. Results revealed that the performance of simple PSO method is slightly better than the famous and complicated SCE method. PSO is able to obtain optimal values for 10 parameters fast and accurate within a multidimensional parameter space that could provide the best fit between the observed and simulated runoff