DEVELOPMENT OF MULTISITE STREAMFLOW GENERATION MODELS

Forecasting of streamflow can have a significant economic impact, as this can help in water management and can be helpful tool to provide protection from water shortages and possible flood damage. In recent work the artificial neural networks different models with different training algorithms were examined to simulate Tigris River using the cross correlation between the flow of different sites or gauge stations. The challenging task in this work was to improve the forecasting models to generate a future series by ANNs models by using input parameters from nearby sites. Therefore the best conventional method to compare and judge the results was selected to be the 1st order autoregressive moving average Matalas which deals with multi variables as input parameters and generate future series for these variables at the same time. The traditional architecture of ANNs models were also a good comparison tools to decide the new multisite ANN models success. It was concluded from this study that consisting nearby sites monthly flow series as input parameters in ANN architecture after investigating the cross correlation between the series's may led to more successful forecasting models. This can also provide a good promise to predict ungauged flow values in some sites which are suffering from missed data by using the flow values from nearby stations.