Influence of the Initialization ofMultilayer Perceptron for Flash FloodForecasting: Design of a Robust Model

During the last few decades neural networks have been increasingly used in hydrological modeling for their fundamental property of parsimonious universal approximator of non-linear functions. For the purpose of flash flood forecasting, feed-forward and recurrent multilayer perceptrons appear to be efficient tools. Nevertheless, their forecasting performances are sensitive to the initialization of the network parameters. The cross-validation efficiency to select initialization providing the best forecasts in real time situation is assessed. Sensitivity to initialization of feed-forward and recurrent models is compared for one hour lead time forecast. This study shows that cross-validation is unable to select the best initialization. A more robust model is designed using the median of several models outputs; in this context, this paper analyses the design of the ensemble model for both recurrent and feed-forward models.