Comparison between disaggregation and Artificial Neural Network (ANN) approaches for seasonal streamflow forecasting

Water discharge is one of the main components for planning, management of waterresources as well as studies of reservoir operation. Therefore it is necessary to use somemethods for forecasting by disaggregation hydrological variables into finer time scales. In this research, artificial neural networks (ANN) have been used for disaggregation of annual into semi-annually and monthly series, disggregation. The models were basic, extended and feed forward back propagation (FFBP). Lake Uromiyeh river basin was selected as a case study for this study which is situated in North West of Iran. Streamflow data were taken from the rivers in the basin. Statistical measures including RMSE, MRE, and SE were calculated to investigate the accuracy of the results. The results indicated that the ANNs and extend models have better performance than basic model for seasonal streamflow forecasting but in smaller scales the extended model is recommended. Results indicated that the differences between observed and disaggregated streamflow series have been increased by presence trend in forecasted values because the effect of parameter estimation step. The statistical properties of variables like mean and standard deviation are also preserved using these models.