Comparison between kinematic wave and artificial neural network models in event-based runoff simulation for an overland plane

The results of a study comparing the Kinematic Wave, coupled with the Phi-index toss model (KWM) with Artificial Neural Network (ANN) model in event-based rainfall-runoff modeling for an asphalt plane are reported in this paper. The rainfall and runoff data for ten natural storm events at 0.25-min interval were used in the analyses. Two categories of ANN models, based on the input configuration, were considered: (i) using measured rainfall only, and using measured rainfall with calculated discharge, and (ii) using both measured rainfall and measured discharge. In the first category, three cases were studied: (i) Type 1 - total rainfall up to four time tags, (ii) Type 2 - same as Type I plus a calculated flow by the KWM, and (iii) Type 3 - same as Type 1 plus a calculated flow by the ANN. In the second category, three other cases were studied: (i) Type 4 - total rainfall up to four time tags and measured discharge up to three time tags, (ii) Type 5 total rainfall up to four time tags and measured discharge up to one time tag, and (iii) Type 6 - total. rainfall up to one time tag and measured discharge up to one time tag. The ANN with Type 2 input performed better than the Type 1 and Type 3 ANN models. Thus, including a calculated discharge from the KWM was found to improve the predictions by the ANN. The ANN with Type 3 input fared worst among the ANN models considered and was found to be an unreliable method for runoff prediction. Type 1 ANNs generally fared worse than the KWM. In general, the ANN models in the second category performed better that the ANN models in the first category. In addition, the second category of ANN models out-performed the KWM in most cases. However, the predictions by the KWM were based on the calculated discharge and not the measured discharge. Finally, the ANN was found to be unable to make accurate predictions beyond the range of its training data. In this case, the KWM performed better that the ANN. (c) 2008 Elsevier B.V. All rights reserved.