Flood estimation at ungauged sites using a new hybrid model

As flood forecasting in ungauged basins has been an area of extensive research, new techniques have been introduced to minimize the forecast errors and to issue more accurate forecasts. The use of Artificial Neural Networks (ANNs) in flood forecasting is new and still in the evolution stage. In this study, MLP and Elman networks and also a new nonlinear regression model are applied and combined with each other for T-year flood estimation in western basins of Urmia Lake. At first, these networks used physiographic and climatic data selected from the regression model, to train. Finally, the best structure of the networks is chosen based on correlation coefficient between observed and estimated discharges. In order to train the models well, the return period variable is considered as one of the input variables of them. The obtained results have proved the ability of the hybrid model to predict T-year flood events and the effect of networks types on prediction precision. © 2008 Asian Network for Scientific Information.