A Comparison of Network Types in Artificial Neural Network-Based Rainfall-Runoff Modelling

Abstract

As a consequence of increasing applications of artificial neural networks (ANNs) in rainfall-runoff modelling, concern over the use of network type came into existence. The most widely used ANN is multilayer perceptron (MLP) ANN with a sigmoid transfer function, trained with back-propagation algorithm. The alternative to the MLPANN is application of radial basis function (RBF) network. This paper presents application and comparison of MLP and RBF-type neural network models developed for rainfall-runoff modelling of Upper Kharun Catchment in Chhattisgarh, India. In total, six neural networks’ architecture has been developed, three each on MLPANN and RBFANN utilizing daily, weekly and monthly flow. Performance of the models was evaluated in terms of their generalisation properties and simulated hydrograph characteristics. Comparison of the network type was carried out by calculating performance evaluation criteria namely mean absolute deviation, root mean square error, coefficient of correlation, Nash-Sutcliffe coefficient efficiency and volumetric error. Results of the study indicate that choice of the network type certainly has an impact on the model performance. It was concluded that the performance of RBFANN models was better than respective MLPANN models for the study area.