A COMBINED WAVELET AND NEURAL NETWORK MODEL FOR FORECASTING STREAMFLOW DATA

The modeling of streamflow is often needed for the sustainable management of water resources and for the protection against flooding. Over the years numerous streamflow forecasting models have been developed, black-box models, like Artificial Neural Networks (ANN), have became quite popular in the field of hydrologic engineering, because of their rapidity and less data requirements compared to physics-based models. In this study, a hybrid model, Wavelet-Neural Network (WNN), for the prediction of streamflow is developed. The model incorporates ANN and wavelet transform for the analysis of variations in streamflow time series. For demonstration, the model is applied to streamflow data from four flow observation stations (FOS), located in the West Mediterranean Basin of Turkey. Monthly mean streamflow data from the four FOS were used in the model. Original series were decomposed sub-series by wavelet transform. These sub-series were used for ANN model. In order to evaluate the performance of the WNN model, a multi regression (MR) model was also developed based on the same data set. Results show that WNN model forecasts the streamflow more accurately than the MR model with correlations between estimated and observed streamflow data ranging from 0.84-0.88.