Estimation of daily streamflow using different artificial intelligence methods-a case study of Haldizen Stream

Streamflow forecasting is an important task for planning and management of water resources and structers. To make especially long-term predictions, most of water resources applications, environmental protection, management of drought, the use of water resources, is also of great importance. Reservoir management of irrigation can provide a useful planning and management facilities for manufacturers and users in developing sustainable water supply and hydroelectric generation. Streamflow is the most important parameter used in water resources planning and design of water structures. Generally modelling of monthly or weekly flow data has better results due to their low variances however, the shorter duration with higher variances such as daily flow data are needed in most usual design cases. Therefore, daily flow data with shorter duration and higher variances is used in the modelling in this study to shed light on the precise data for short term planning. Data of Haldizen Stream in the East Balack Sea Basin is used and attempted to develop a model to use in water structure management in the region. In the study, during the period of 1998-2000 years of daily stream flow measured data of Haldizen Stream located in the Eastern Black Sea Basin has been used in the models. The following models are used in this study for streamflow forecasting: Feed forward back propagation multilayer artificial neural network (MLP-NN), Principal Component Neural Network (PC-NN), Time-Lagged Recurrent Neural Network (TL-NN). Performance of the models are compared with various criteria and best fit of the models are determined and presented with tables and figures. Key Words: Daily Flow Estimation, Haldizen Stream, Multi Layer Perceptron Neural Network, Principal Component Analysis Neural Network, Time Lagged Recurrent Neural Network.