Comparing Combined ARMA-PARCH and ARMA-ARCH Models for Modeling Peak Flow Discharge (Case Study: Siminehrood River in the West Azarbaijan Province)

Accurate prediction of river flow discharge is very important for optimization and management of surface water resources. One of the applicable ways for simulating and predicting hydrological data such as river flow is using time series models. But one of the major problems for prediction of hydrological data by time series models is assignment of the generation approach of the random and stochastic series. In this study, it is attempted to define a systematic framework for generating random component of a time series model. In this case, after initial tests on the Siminehrood river peak flow discharge time series during 1967-2011, the ARMA models were evaluated and the ARMA (1,0) model was selected as the best model. Then residual series of the selected ARMA model was extracted and fitted by ARCH and PARCH nonlinear models approaches to obtain the ARMA-PARCH and ARMA-ARCH models. Results showed that using combined ARMA-PARCH and ARMA-ARCH models reduced the error of models up to 92.22 and 92.16 percent, respectively. It was observed that between the two combined models, the ARMAPARCH model had more accuracy and less error than the ARMA-ARCH model. Also with the selected combined model, the maximum and minimum points of the peak flow discharges were modeled well for the studied station.