Evaluation of a Developed Discrete Time-Series Method in Flow Forecasting Models

Forecasting flow in rivers has special significance in surface water management, especially in agricultural planning and risk reduction of floods and droughts. In recent years, studies have shown the superiority of forecasting models based on artificial intelligence, using artificial neural networks (ANN) and genetic programming (GP), over time-series models. In this paper, continuous and discrete historical flow records are used for monthly river flow forecasting of the Saeed-Abad river in East Azarbaijan province, Iran. Auto regressive moving average with exogenous inputs (ARMAX), ANN, and GP models are used in both continuous and discrete flow series. For both flow series, results of the ARMAX, ANN, and GP models are then compared and results of each method are evaluated relative to each other. Two quantitative standard statistical performance evaluation measures, coefficient of determination (R 2 ) and root mean square error (RMSE), are employed to evaluate the performance of the aforementioned models. Results show that for the two methods, the GP model is more effective with respect to accuracy than ARMAX and ANN. For continuous time-series forecasting, GP is a more precise model (R 2 = 0.7 and RMSE = 0.172) than either ANN (R 2 = 0.627 and RMSE = 0.193) or ARMAX (R 2 = 0.595 and RMSE = 0.243). For discrete time-series forecasting, the superiority of the GP model is evident in most months. For monthly flow forecasting, results indicate that the discrete time-series forecasting method is superior to the continuous time-series forecasting method. Copyright Springer Science+Business Media Dordrecht 2015