Study uncertainty of parameters of hydrological model (SWAT) by Differential Evolution Adaptive Metropolis algorithm ‎(DREAM-ZS)‎

Background and Objectives: Quantifying the uncertainties of the parameters of hydrological models are the role of great importance in water resource management and is a challenge that due to the large number of parameters and lack of proper physical understanding of them, these models face problems in the calibration stage. Considering the importance of water resources in Iran and the need to investigate uncertainty in order to achieve reliable results, the purpose of this study is to investigate, identify and quantify parameters uncertainty of Soil and Water Assessment Tool (SWAT) and their performance to predict watershed runoff. The Kashfrood river is a semi-arid large-scale basin in northeastern Iran using a Monte Carlo chain-based Markov chain simulation method called the Differential Evolution Adaptive Metropolis algorithm (DREAM-ZS).
Material and Methods: With the purpose of assess the uncetainty in this study only 20 out of 29 available parameters were selected and evaluated based on Regional Sensitivity Analysis (RSA) as sensitive parameters. In order to optimize the model and quantify the parameters uncertainty, scenarios S1 (first scenario) and S2 (second scenario), which belong to the DREAM-ZS algorithm, have been defined. The prior parameter ranges of the S1 scenario were determined using the final calibration of parameter ranges in SWAT-Calibration and Uncertainty Program (SWAT-CUP) software and Sequential Uncertainty Fitting version 2 (SUFI 2) algorithm, and the prior ranges of the S2 scenario were determined using a compromising approach between the prior ranges of the SWAT-CUP and posterior ranges from S1 scenario. In this study, the parameters, uncertainties and statistical analysis have to be computed via an appropriate likelihood function. Therefore, the Differential Evolution Adaptive Metropolis (DREAM-ZS) combined with the standard least squares (SLS) as a simple formal likelihood function. Also, to evaluate the performance of the model uncertainty in the two mentioned scenarios, evaluation criteria including P-factor, d-factor, Nash–Sutcliffe (NS), Total Uncertainty Index (TUI) and Average Deviation Amplitude (ADA) were used.
Results: P-factor, d-factor, Nash–Sutcliffe (NS), Total Uncertainty Index (TUI) and Average Deviation Amplitude (ADA) showed that the S2 scenario has a better performance than scenario S1 in reducing forecast uncertainties. According to S1 simulation, the NS coefficient ranged from 0.54 to 0.72, while in S2 simulation, it ranged from 0.63 to 0.78. The TUI for total uncertainty was in a range of 0.2–0.6 and 0.22–0.66 for S1 and S2 scenarios, respectively. The S1 and S2 simulations led to the TUI of 0.63–0.94 and 0.74–1.22 for parameter uncertainty, respectively. Finally, ADA index for total uncertainty was 0.098 and 0.445 for S1 and S2 scenarios, while in accordance to S1 and S2 simulations, the ADA index for parameter uncertainty was 0.098 and 0.451, respectively.
Conclusion: The DREAM-ZS algorithm improved the calibration efficiency of the model and led to the presentation of more real values of runoff simulation parameters by the SWAT model in the Kashafrood River Basin.