Predicting streamflows based on neural networks

A number of approaches has been proposed in the literature for predicting and forecasting monthly streamflows. Neural networks (NN) is a fairly recent technique which has been suggested and applied for many computational problems in water resources. NN and periodic transfer function models (PTF) are compared for forecasting monthly flows of the Rio Grande Basin. Forecast biases and root mean square errors (RMSE) obtained from both models are calculated. The results show that forecast biases are about the same for both methods. On the other hand, smaller RMSE's are obtained for forecasts based on neural networks models. The differences are specially significant when forecasts are made based on independent data sets.