Evaluation of the Neuro-Fuzzy and Hybrid Wavelet-Neural Models Efficiency in River Flow Forecasting (Case Study: Mohmmad Abad Watershed)

  One of the most important issues in watersheds management is rainfall-runoff hydrological process forecasting. Using new models in this field can contribute to proper management and planning. In addition, river flow forecasting, especially in flood conditions, will allow authorities to reduce the risk of flood damage. Considering the importance of river flow forecasting in water resources management, various methods are used to model rivers flow in order to minimize their potential damage by using the model in drought and flood management. In this study, the neuro-fazzy and hybrid neural-wavelet models were used to forecast the daily flow time series of the Sarmo water meter station located on the Mohammad Abad River. For this purpose, the original time series has been translated to three sub-series for 28 years using wavelet transformation and type IV Daubechies mother wavelet. The correlation coefficient value was obtained 0.88 for neuro-fuzzy model and 0.95 for hybrid wavelet -neural model, and the RMSE, MSE, and NS evaluation parameters in neural-wavelet model were 0.004, 0.043, and 0.91, respectively; these parameters in neuro-fuzzy model were 0.32, 0.10, and 0.77, respectively. Finally, the results of wavelet-neural transformation were compared with the results of neuro-fuzzy model, and it was observed that the wavelet- neural method has a higher predictive accuracy than the neuro-fuzzy method.