Real-time flood forecasting via combined use of conceptual and stochastic models

Stochastic modelling of the simulation errors resulting from the off-line application of conceptual rainfall-runoff models is often performed in the context of real-time flood forecasting, in order to improve the forecasting accuracy. Although widely applied in the operational practice, such approach has not been yet extensively investigated in the scientific literature. This analysis is aimed at evaluating the benefits in discharge forecast accuracy that can be gained by this kind of approach and to provide some insights into the identification and estimation procedures of the optimal stochastic model to be applied when updating the forecasts. Application of univariate linear ARIMA models, even in the fractionally differenced form, has been herein considered for a case study referred to the Sieve River basin, located in Central Italy. The results highlight the dependence of the benefits retrievable from the stochastic updating procedure on the lead time of the flood forecasting.