Fuzzy modelling of basin saturation state and neural networks for flood forecasting

Over the last decade, neural networks-based flood forecast systems have been increasingly used in hydrological studies. Usually, input data of the network are composed by past measurements of flows and rainfalls, without providing a description of the saturation state of the basin, which in contrast plays a key role in the rainfall-runoff process. Here, we adopt a fuzzy approach in order to provide a description of the basin saturation state; the basin state is classified as belonging with different degrees of membership to different saturation classes, starting from the analysis of the cumulated rainfall information. A different neural predictor is specialized to mimick the rainfall-runoff relationship which pertains to each different saturation class. The forecast is obtained weighting the outputs of the specialized neural predictors, the weights being given by the current memberships of the basin state to the different saturation classes. The framework has been tested on an Italian catchment where it overperforms a classical neural networks.