Multi-objective automatic calibration of SWAT using NSGA-II

This paper presents a diagnostic study on muttiobjective, automatic calibration of a physically based, semi-distributed watershed model known as Soil and Water Assessment Too[ (SWAT). Unlike lumped models, distributed models involve large number of calibration parameters, representing the spatial heterogeneity of inputs and various physical processes within a watershed. An automatic calibration routine is developed using the Non-dominated Sorting Genetic Algorithm II (NSGA-II) that has been proved to be an effective and efficient muttiobjective search technique in various applications. The automatic routine is capable of incorporating multiple objectives into the calibration process and also employs parameterization to help reduce the number of calibration parameters. In this study, SWAT is calibrated for daily streamftow and sediment concentration. Two calibration scenarios have been considered; the first scenario uses specific objective functions to fit different portions of the time series whereas in the second scenario, the calibration is performed using data from multiple gauging stations, simultaneously. In addition, two cases of parameter distribution have been considered in the second scenario during parameterization. The application results show that the approach is consistent and effective in estimating parameters of the model. The use of multiple objectives during the calibration process resulted in improved model performance and the second scenario, in particular, provided better results partly due to the respective location of the gauging stations within the watershed. Further distribution of parameters during parameterization also resulted in better sediment simulation. (c) 2007 Elsevier B.V. AR rights reserved.