Comparing the accuracy of runoff prediction and estimation using SWAT and artificial intelligence models in Minab River

In arid regions,like most of the Iran, human is suffering fromwater shortage. Water harvesting can be effective, especially in correct exploitation of existing waters in arid regions. With an average rainfall of less than one-third of the world, there are different climates in Iran, even in southern parts like Minab and the areas around Estaghlal Dam. In current situations, rainfall pattern has been changed and the length of drought periods has been increased in Minab. Last designed standard operating systems for estimating the amount of water entering to reservoirs like Esteghlal Dam are not sufficient. So, it is necessary to use new methods with higher accuracy in estimating and predicting watershed surface runoff. To achieve this objective, the use of numerical models for estimating and predicting is inevitable. In this research, SWAT and artificial intelligence models are used to estimate and forecast surface runoff. Calibration, validation and prediction of surface runoff were computed using soil, land use, topography and hydro-climatic data layers in the yearly and monthly basis. The annual values of evaluation criteria such as Mean Square Error (RMSE) and mean absolute error (MSE) in the calibration of the SWAT model were 6.89, 8.37 and for FTDNN were 5.35, 7.76, respectively, while, the monthly calibration results were 16.29, 32.02 for the SWAT and 9.46, 22.86 for FTDNN models. Linear regression coefficients in monthly calibration of models were 0.96 and 0.60 and in annual calibration of models were 0.94 and 0.98, respectively. Comparing criteria of evaluation of two models concluded that artificial intelligent model (FTDNN) has more accuracy and superior performance compared to SWAT model.