Comparison of seasonal time series models and nonlinear BL models and SETAR models in forecasting the monthly in flow of mourn dam

In the present research, the SARIMA seasonal time series, Holt-Winters, bi-linear and Self-Exciting Threshold Auto-Regressive models were used to predict the monthly inflow to the Maroun dam reservoir. To this end, the 34-year data of the Idanak hydrometric station located in Khuzestan province of Iran between the years 1982 and 2015 have been used. The logarithmic transformation was used to normalize the monthly discharge data of the Idanak hydrometric station. differencing technique was used to eliminate the seasonal component of the monthly data. The independence test of the model residuals (Ljung-Box or porte-manteau), the autocorrelation and partial autocorrelation functions were used to check the validity (quality of fitting) of these models. SARIMA models (1.0.1) * (2.0.2) 12, BL (2.1.1.1) and SETAR (2; 7.3) were chosen as the best models with the minimum values of Akaike and Schwartz criteria. The results of the evaluation of fitted models showed that the BL model with the values of the coefficient of determination and root mean square error which are 0.81 and 14.80 m3/s, respectively, has an acceptable accuracy to predict the monthly flow to the Maroun River. It was also found that by increasing the non-seasonal rank degree in SARIMA models, the model validity and performance are weakened to predict monthly flow. The results showed that the Holt-Winters model with the values of the coefficient of determination and root mean square error which are 0.56 and 10 m3/s, respectively, has the weakest performance to predict the monthly flow to the Maroun dam reservoir.