A flood forecasting system through the use of water levels

A real time system can provide flood forecasting with a reasonable lead time through the use of data gathered from a telemonitoring network. Hydrological models usually need discharge data to simulate or to forecast the behaviour of flood hydrographs because the flow in a river is a relatively conservative variable, but rating curves at the measuring cross sections are often unknown or unreliable. On the other hand, the use of hydrometers for measuring the level in water courses is widespread and, in spite of the river level is more directly affected by local characteristics of the river, its value is pertinent to flood warning. In this paper a stochastic flood forecasting model is proposed. It is based on level data and its structure is linear and quite robust: the non linearity involved in the physical phenomenon is taken in account by real time parameter adjustment. The accuracy and the reliability of the model are discussed by a study basin. The downstream station is located at Ponte Coperto (Pavia - Italy) and the upstream stations are distributed along the Ticino, Po and Tanaro Rivers. The model may work including data from three, two or one upstream stations, depending on the number of working stations during each of the selected flood events. Forecasting errors are due to model errors, because it is a well–established fact that no physical system is truly linear, and to errors contained in the experimental data. The error in magnitude for a 6-hours lead time is typically less then 20 cm. In this paper the possibility of improving the reliability of the forecasting system is investigated and a methodology to regularize level data is applied for this purpose.