Real Time Forecasting Using Regression and Artificial Neural Network Models

Knowledge of the time and the amount of flood occurrence has a significant impact in reducing human and financial damages caused by the flood. Traditional flood forecasting methods have usually been in forms of classical methods such as rainfall-runoff, routing, and regression. Recently the use of artificial neural networks (ANN) has been proposed. In this study the ability of ANN in site floods forecasting under limited data conditions has been investigated and compared with previously studied methods. Hence, two models have been prepared. The first model is a multiple regression model and the second an ANN model. In this study, the models were prepared using statistics and data from 10 simultaneous floods in four hydrometric stations, upstream of the case study site. Seven floods were used for calibrating and three for testing the models. In each model, multiple parameters were investigated. Comparison of the results of two models indicated that ANN showed a convenient and promising operation in flow forecasting and presented more precise forecasts when compared to the other model.