A Conceptual and Neural Network Model for Real-Time Flood Forecasting of the Tiber River in Rome

B. Calvo, G. Napolitano, F. Savi (Italy), L. See, B. Irvine, and A. Heppenstall (UK)


real time forecasting, flooding of urban areas, conceptual models, artificial neural networks


Rome is subjected to a serious risk of inundation when extreme floods propagate along the Tiber river. The mathematical model TFF (Tevere Flood Forecasting) for real time forecasting of hourly discharge or water levels at Ripetta gauging station in Rome is proposed. This is a conceptual model, composed of a semi-distributed rainfall-runoff model applied on 41 ungauged sub-basins covering about 30% of the catchment area of the Tiber river downstream from Corbara dam to Tyrrhenian sea, and a flood routing model. The parameters of the flood routing model were calibrated offline and held constant during the forecasting computations. The parameters of the rainfall-runoff model are calibrated during the event at each time step via an adaptive procedure. A neural network model (Tevere Neural Network, TNN) has also been developed and the performance compared with the conceptual model.

Important Links:

Go Back