River flood routing using artificial neural networks

Abstract

Flood routing through rivers is very important for the design of protection measures. Sensitivity analysis was performed using HEC-RAS 1D to show the effect of Manning’s coefficient, reach length, peak inflow hydrograph, base time of the inflow hydrograph, channel bed slope, channel side slope, and channel bed width on the peak discharge and the shape of the downstream hydrograph. About 1,600 synthetic realizations were generated using HEC-RAS 1D to train, validate, and test the artificial neural networks (ANNs). The network was trained to estimate the peak discharge from river reach and inflow hydrograph parameters. Testing this network shows a very good agreement between the predicted and actual peak discharges. Another network is trained to predict the whole outflow hydrograph from inflow hydrograph for a certain reach. Hence, flood routing through rivers can be performed using ANNs easily, accurately and quickly.