RBFNN Versus Empirical Models for Lag Time Prediction in Tropical Humid Rivers

Abstract

Lag time (Lt) reflects the speed at which a river basin responds to rainfall (RF) events and is influenced by many hydrological parameters such as RF and stream flow (SF). These two parameters are represented by four variables, namely peak RF intensity, previous 48-h rainfall, peak SF and previous 48-h SF. In fact, lag time is highly stochastic in nature and its relation with the mentioned four variables is highly nonlinear interrelationship. The main objective of this study is to develop a model to estimate the Lt between upstream and downstream stations in tropical humid rivers. The graphical hydrological approach (HGA) has been used to estimate the Lt based on 95 RF-SF and considered as the references value for the proposed model evaluation. Linear, non-linear and Radial Basis Function Neural Network (RBFNN) methods have been developed successfully for Selangor River basin. The results show that the RBFNN outperformed the linear and the non-linear model and could achieve correlation coefficient (r) between the observed Lt and predicted Lt equal to 0.979 while r for the linear and the non-linear model equal to 0.519 and 0.631, receptively. Furthermore, the RBFNN model could attain minimum root mean square error (RMSE) between the observed Lt and predicted Lt equal to 1.23 while RMSE for the linear and the non-linear model equal to 1.9 and 2.02, receptively. The proposed RBFNN model significantly abridges the estimation of Lt values and avoids the essential need for comprehensive description of all parameters affecting on its value.