Baranduz-Chay River Flow Modeling Using the K-Nearest Neighbor and Intelligent Methods

River flow accurate forecasting is so important in design, operation and planning of water resources systems. In this study, the performances of the non-parametric nearest neighbor method, adaptive neuro- fuzzy (ANFIS) and support vector regression (SVR) approaches were evaluated for streamflow forecasting. In order to derive the model, monthly streamflow observations of 36 years time period at Dizaj hydrometric station located in the Baranduz-Chay River (in monthly time scale) were used. Different combinations of the recorded data were used as the input pattern of streamflow forecasting. The results indicated that all of the applied models had reasonable performances in prediction of the monthly river flow. By adding the seasonality coefficient of streamflow to input pattern, performances of the intelligent models increased considerably. In general, the SVR model using the suitable input pattern was selected as the best method. The values of the three different evaluation criteria, namely correlation coefficient (R), root mean square error (RMSE), mean absolute relative error (MARE) were equal to 0.88, 3.63 (m3/s) and 78.45, respectively. Furthermore, evaluation of the performances of the models for streamflow forecasting revealed that in the cases of high discharges all of the models underestimated the streamflow comparing with the observed values.