Performance Evaluation of Combined Multivariate Time Series, MPAR and MPAR-ARCH Models for Modeling River Flow Series Considering the Effective Meteorological Components (Case Study: Nazloochai River)

Applying multivariate models in describing and modeling complicated hydrological events has been recommended by hydrologists in the recent three decades. In fact, employing effective factors in the multivariate models can improve the results of describing, modeling and predicting different variables. Furthermore, nonlinear conditional heteroscedastic models can be used for modeling linear residual part of time series and it is expected that combining the linear and nonlinear models increases the accuracy of modeling and forecasting results. In this study the two multivariate periodic ARMA and multivariate combined with the conditional heteroscedasticity models were compared and used to model Nazloochai River discharge located at the West Azerbaijan Province by considering air temperature and precipitation variables during the period of 1962-2011.The results of the models evaluations and verifications showed that both of the models had acceptable accuracy in modeling of the river flow discharge. Also results indicated that the combined conditional heteroscedasticity multivariate models involving the effective parameters of river flow series had more accuracy than multivariate periodic ARMA model. The both models estimated the maximum and minimum points of discharge series correctly. Also the results showed that by combining two multivariate periodic ARMA and nonlinear autoregressive conditional heteroscedastic models the error was decreased about 16% in comparison with the error of the periodic ARMA model.