Implications of radar rainfall estimates uncertainty on distributed hydrological model predictions

Real time availability and high space–time resolution of radar based quantitative precipitation estimates (QPE) are appealing features for spatially detailed hydrological prediction and forecasting applications using distributed hydrological models. However, the data obtained remain an important source of uncertainty for hydrological predictions. Insight into the characteristics of this uncertainty is still limited and its quantification is a challenging task. This work studies the characteristics of radar QPE uncertainties and its implications on both hydrological modelling results and hydrological model parameter estimates. The uncertainty of a real time radar QPE product available in 10 minute intervals on a 1 km 2 grid is quantified by comparison to a reference precipitation field, which includes additional observations from rain gauge records. Based on this analysis a probabilistic model is proposed that describes the uncertainty structure of the radar QPE field. An ensemble of precipitation fields is generated that represents a quantitative estimate of radar QPE uncertainty from the sampling of the probabilistic model. On this basis the implications of radar QPE uncertainty on distributed hydrological model predictions are studied. The methodology proposed is applied to a real-world case study using the river basin of the Besòs in Spain as a test bed. The feasibility of the approach to condense the knowledge about radar QPE uncertainty in a probabilistic model and to map this uncertainty to the response of a hydrological model using an ensemble of precipitation fields is demonstrated. The results show that the probabilistic use of radar QPE may add valuable information to hydrological predictions and may reduce the bias of hydrological model parameter estimates.