Forecasting Annual Streamflow Using Autoregressive Integrated Moving Average Model and Fuzzy Regression

By Increasing population and the limitation of water resources, forecasting of streamflow using new methods in planning and management of water resources such as reservoirs operation has an outstanding importance. For this aim, over the past years time series models have been considered by hydrologists. The goal of this research was to investigate stochastic model (ARIMA) and fuzzy regression performance for the annual streamflow forecasting. The parameter estimation methods of ARIMA model were conditional and unconditional likelihoods. In fuzzy regression model the symmetric and non-symmetric triangular memberships were used regarding the uncertainties of the real systems. For the comparison of ARIMA and fuzzy regression performance, streamflow data of some tributaries of Urmia lake basin were employed. Results indicated that the unconditional likelihood was the best method among the parameter estimation methods. Comparison between the forecasted and observed streamflow series using the two models revealed that the fuzzy regression had the best fit to the observed streamflow data. The performance of the symmetric triangular membership was better than that of the non- symmetric.