Performance of DREAM-ZS and SUFI-2 algorithms to determine the optimum parameters of uncertainty in SWAT model

To estimate the hydrologic components of river basins, the semi-distributed watershed models are widely used. Although these models play an essential role in future management and planning of water resources, quantify the different sources of uncertainties in a hydrological model is a challenging task. Therefore, these models should be associated with a robust uncertainty analysis methods. Generally, parameters of hydrologic models involved with runoff cannot be measured directly. Therefore, inverse modeling helps to solve this problem and estimates these parameters. In this study, the performance of Sequential Uncertainty Fitting (SUFI-2) and DiffeRential Evolution Adaptive Metropolis (DREAM-ZS) algorithms in Soil and Water Assessment Tools (SWAT) model with Multisite flow gauging stations to predict runoff in the Kashafrood watershed (16870 km2) located in Khorasan Razavi Province, Iran was investigated. To select the best algorithm five criteria evaluation; P-factor, d-factor, NS, TUI and ADA was used. All statistical indicators and performance criteria at five runoff gauge stations, showed that the DREAM-ZS algorithm performed better than the SUFI-2 algorithm for reducing the prediction uncertainties (e.g. values of NS, TUI and ADA for sub-basins ranged from 0.6 to 0.71, 0.45 to 0.8 and 0.11 to 0.54, respectively for SUFI-2 and ranged from 0.64 to 0.78, 0.74 to 1.22 and 0.09 to 0.45, respectively for DREAM-ZS algorithm). Consequently, the use of DREAM-ZS algorithm which improved the model calibration and the fitted value of the parameters in this algorithm is recommended for further applications in same area.