Uncertainty assessment through a precipitation dependent hydrologic uncertainty processor: An application to a small catchment in southern Italy

SummaryAdequate assessment of uncertainty for prediction and simulation purposes is a current issue in hydrological research. This article describes the application of the Hydrologic Uncertainty Processor (HUP) proposed by Krzystofowicz in 1999 to a small semi-arid watershed in southern Italy. The version applied in this work is a precipitation-dependent HUP aimed at assessing the hydrologic uncertainty about actual streamflow at some future time, with lead times of a few hours, given the information available at the forecast time and assuming a perfectly known amount of precipitation. The processor is based on Bayes theorem and hence models the prior and likelihood functions to obtain the revised posterior distribution.A complete example of the modelling assumptions, estimation procedure and results is carried out in the present paper. In detail, we analysed a 26-km2 semi-arid basin, considering hourly forecasts over an almost continuous five-year period in 2000–2005. A distributed rainfall–runoff model suited to represent contributions of different runoff generation mechanisms to hydrologic response is used for deterministic predictions. Analysis of the resulting posterior distributions show that hydrologic uncertainty: (i) grows with the value of discharge predicted by the model; (ii) is higher when associated with high precipitation amounts; and (iii) increases with lead time of predictions. The predictive ability of the processor is investigated for several runoff events. The results indicate good processor performance for a lead time equal to the period covered by the precipitation forecast, and a significant deterioration for higher lead times that is heavily dominated by the presumption of null precipitation beyond the forecast period. Finally, the skill of the processor is assessed through a retrospective analysis in terms of the probability of detection and the false-alarm rate.