Application of ANN for runoff forecasting : an analysis of the methodology

The rainfall-runoff process is the physical basis of ANN (Artificial Neural Network) runoff forecasting models. Flood forecasting models can be divided into two categories based on differences between the input and output factors. One is the multiple inputs and single output model, and the other is the multiple inputs and multiple outputs model. In this paper, the applicability and existing problems of these models are analysed. Then, improvements for sample selection, input and output data adoption, etc. are suggested, based on the physical processes of rainfall-runoff, to cope with the existing problems in flood forecasting.