A combined generalized regression neural network wavelet model for monthly streamflow prediction

Abstract

The ability of a combined model, Wavelet-Generalized Regression Neural Network (WGRNN), is investigated in the current study for the prediction of monthly streamflows. The WGRNN model is obtained by combining two methods, Discrete Wavelet Transform (DWT) and Generalized Regression Neural Network (GRNN), for one-month-ahead streamflow forecasting. The monthly flow data of two stations, the Gerdelli Station on the Canakdere River and the Isakoy Station on the Goksudere River, in the Eastern Black Sea region of Turkey are used in the study. The forecasts of the WGRNN model are tested using the Root Mean Square Error (RMSE), Variance Account For (VAF) and correlation coefficient (R) statistics and the results are compared with those of the single GRNN and Feed Forward Neural Network (FFNN). The comparison results revealed that the WGRNN performs better than the GRNN and FFNN models in monthly streamflow prediction. For the Gerdelli and Isakoy stations, it is found that the WGRNN models with RMSE = 5.31 m3/s, VAF = 52.3%, R = 0.728 and RMSE = 3.36 m3/s, VAF = 55.1%, R = 0.742 in the test period are superior in forecasting monthly streamflows than the best accurate GRNN models with RMSE = 6.39 m3/s, VAF = 30.1%, R = 0.553 and RMSE = 4.19 m3/s, VAF = 30.1%, R = 0.549, respectively.