عنوان مقاله [English]
نویسندگان [English]چکیده [English]
With the popularity of complex hydrologic models, the time taken to run these models is increasing substantially. Also, in order to apply a precise runoff modeling in a watershed, the efficient parameter calibration is of importance while reviewing and producing runoff data. Comparing and evaluating the efficiency of different optimization algorithms for calibrating these hydrologic models is now becoming a nontrivial issue. In the present research, particle swarm optimization (PSO) algorithm is utilized for parameter calibration of the Soil and Water Assessment Tool (SWAT) Version 2009, in Sanjabi watershed. SWAT is a semi distribution hydrological model. This model is a river basin scale model developed to quantify the impact of land management practices in large and complex watersheds. In fact, SWAT model is capable to predict the impact of land management practice on water, sediment and agricultural chemical yields in large complex watersheds with varying soils, land use and management conditions over long periods of time. SWAT is widely used in assessing soil erosion prevention and control, non-point source pollution control and regional management in watersheds. Sanjabi watershed is located in Kermanshah province in west of Iran. For this purpose, SWAT model was run by applying required information layers such as land use, Dem layers and rain and temperature data in Sanjabi basin in the period of 1995-2004.
The hydrological model's performance entirely depends on the optimality of the calibration of the model's parameters which their real values are not available. To calibrate SWAT parameters, SWAT-CUP (SWAT Calibration and Uncertainty Procedures) program is used herein. The SWAT-CUP package is a program designed to integrate the various calibration/uncertainty analysis programs for SWAT using the same interface. This program links SUFI2, PSO, GLUE, ParaSol, and MCMC procedures to SWAT. Since the number of SWAT parameters is relatively large, the procedure of sensitivity analysis has been done firstly and thereafter most of sensitive parameters have been calibrated. Sensitivity analysis was performed herein for 22 more important flow parameters (in 1995-2004 period) by connecting the output of SWAT model to the PSO algorithm in SWAT-CUP package. The sensitivity of flow parameters was determined in PSO procedure by P-Value and t-State. The most sensitive parameters were, ranked by importance degree, are as follows: SOL_BD (Soil bulk density), GW_Dealy (Delay time for aquifer recharge), ESCO (Soil evaporation compensation coefficient), SOL_AWC (Available soil water capacity), CN2 (Moisture condition II curve number), RCHRG_DP (Deep aquifer percolation fraction), ALPHA_BNK (Base flow alpha factor for bank storage) and CH_K2 (Effective hydraulic conductivity in main channel alluvium). As a result of sensitivity analysis, the most eight sensitive parameters were selected for optimization purposes. So, the best value of these most sensitive parameters has been calibrated utilizing the PSO algorithm in SWAT-CUP. 4000 simulations of PSO have been performed (4 iterations with 1000 simulations) to obtain the best value of these SWAT parameters. The objective function considered in this research was Nash-Sutcliffe (NS) simulation efficiency. After calibrating the SWAT model in a specified period, the optimum values of calibrated parameters have been fixed in the model and the model was validated for a period extending from 2005 to 2007.
In order to assess SWAT simulation performance, in addition to NS, two more criteria, including R2 (coefficient of determination) and RMSE (root mean square error) have been used. The value of Nash-Sutcliffe objective function (NS) for calibration and validation period was obtained 0.58 and 0.60, respectively. The evaluation results of the model show that the values of R2 and RMSE for calibration period are 0.65 and 6.74 m3/s, respectively, and for validation period are 0.67 and 3.66 m3/s, respectively. If the obtained NS value of SWAT simulations was equal or higher than 0.75, the model considered to be accurate in predicting flood flows, the results of the model is satisfactory when NS value is between 0.36 and 0.75 and the results of the model is not satisfactory when NS value obtained less than 0.36. So the obtained results of SWAT simulation herein showed the desired accuracy of SWAT for runoff simulation was calibrated by PSO. Results show the high accuracy of SWAT model for runoff simulation in Sanjabi watershed. Therefore, SWAT model can be used to predict future surface runoff in Sanjabi watershed and other watersheds in Iran (after calibrating model for watersheds). Having knowledge of the results of the SWAT model in each watershed can play an effective role in water resource planning for local and regional planners.