Flow forecasting for a Hawaii stream using rating curves and neural networks

This paper illustrates the applications of feedforward back propagation and radial basis function neural networks for flow prediction of a Hawaii stream and one of its tributaries that render flash flood behavior. Stream flow is estimated from stream stage, width, cross-sectional area, and mean velocity at the gaging station. Since measurement of mean velocity is time-consuming, expensive, and dangerous during high flows, alternative methods of flow forecasting are needed. Traditionally, hydrologists use rating curves for flow estimates. The United States Geological Survey has been estimating stream discharge using conventional rating curves for Hawaii streams. The rating curves are developed from records of measured stream stage and discharge. Clearly, the rating curves are set up for cases in which stream stage is the only input and discharge is the output. However, major limitations of using a rating curve are that the effects of hysteresis (i.e. loop-rating) are not taken into account, and as a result the prediction accuracy is lost when the stream changes its flow behavior. As an alternative method, artificial neural networks are proposed. The performances of artificial neural networks are examined for two input data sets: one set with and the other set without mean velocity, but both including stream stage, width, and cross-sectional area for two gaging stations on the stream. The results show that for both input data sets, well-optimized neural networks can outperform rating curves for discharge forecasting. Additionally, it is worth noting that neural networks are capable of predicting the loop-rating curve, which is impossible to predict using conventional rating curves.