FLOOD PREDICTION FOR KLANG RIVER USING MUSKINGUM AND ANN MODELS

Abstract

Temporal and spatial variations of a flood hydrograph moving through a river reach can be simulated using flood routing tools such as hydrodynamic, hydrological and the ANN (Artificial Neural Networks) models. The ANN models have emerged as viable tools in flood routing and are widely adopted for this purpose. The aim of this study is to make an objective comparison of these two flood routing models to evaluate their individual performance. Four flood events recorded for Klang river at Kuala Lumpur in the period October 1973 to December 1974 for stations at Leboh Pasar and Sulaiman Bridge which are 950m apart were used for this study. The statistical performance of the models is assessed using criteria such as peak flow, root mean square error, mean absolute error and Nash –sutcliffe coefficient. Results from calibration runs for the 02/05/1974 flood event show that the MAE, RMSE and NAE for ANN and Muskingum models are 0.75,1.24,0.9917 and 1.1,1.3, 0.992 respectively. The performance of the two models was verified using three other different events. Results of simulation runs for the 10/12/1974 event gave 2.72, 3.24, 0.96 and 2.1,3.1,0.963 MAE, RMSE and NAE values for ANN and Muskingum. Graphical inspections and statistical tests show that the ANN and Muskingum methods performed equally well in flood prediction for this study, using the flood events of Klang river.