Importance of hybrid models for forecasting of hydrological variable

In this study, a forecasting model for nonlinear and non-stationary hydrological data based on singular spectrum analysis (SSA) and artificial neural networks (ANN) is presented. The stream flow data were decomposed into its independent components using SSA. These sub-bands representing the trend and oscillatory behavior of hydrological data were forecasted 1 month ahead using ANN. The forecasted data were obtained with summation of each forecasted sub-bands. The mean square errors (MSE), mean absolute errors (MAE) and correlation coefficient (R) statistics were used for evaluating the performance of the proposed model. According to statistical parameters, the hybrid SSA–ANN model was a very promising approach for forecasting of hydrological data. The statistical performance parameters were obtained as MSE = 0.00088, MAE = 0.0217 and R = 0.986. Also, hydrological data were forecasted using single ANN model for the comparison. Results were compared with the SSA–ANN model and showed that the SSA–ANN model was much more accurate than the ANN model for the prediction of 1 month ahead stream flow data. To demonstrate the practical utility of the proposed method, SSA–ANN and ANN models were used from 1 to 6 months ahead for forecasting of hydrological data.