Development of a rating curve using artificial neural networks

Abstract

The establishment of a rating curve is an important problem in hydrology. A relationship between stage and corresponding measured discharge is usually derived using various graphical and analytical methods. As the relationship between stage and discharge is not linear, conventional methods based on least squares regression analysis are unable to model the relationship of non-linearity. They are also prone to fail in fitting a relationship when the hysteresis is present in the data. The aim of this study is to investigate the potential of employing neural networks for constructing rating curves at gauging stations in the Han River, Korea, and to compare backpropagation (BP) type and radial basis function (RBF) type neural networks. The results show that neural network approaches are highly viable than the conventional approaches. A comparison of the BP and RBF models reveals that the RBF modeling approach is superior and can model the hysteresis effect more accurately than the BP based approach. This provides an impetus to use RBF to establish a stage-discharge relationship existing hysteresis.