Performance of Neural Networks in Daily Streamflow Forecasting

Abstract

Feed-forward multilayer neural networks are widely used as predictors in several fields of applications. The purpose of this study is to investigate the performance of neural networks as potential models capable of forecasting daily streamflows. Once an appropriate network has been identified, a comparison approach is used to evaluate it against a conceptual model presently in use by the Alcan Company. The Mistassibi River, located in northeastern Quebec, serves as the case study, and results based on mean square errors and Nash coefficients show that artificial neural networks outperform the deterministic model PREVIS for up to 5-day-ahead forecasts. Moreover, the results obtained with the neural network are also superior to the ones obtained with a classic autoregressive model coupled with a Kalman filter.