Multivariate distributed ensemble generator: A new scheme for ensemble radar precipitation estimation over temperate maritime climate

Summary It is broadly recognized that large uncertainties are associated with radar rainfall (RR) estimates, which could propagate in the hydrologic forecast system and contaminate its final outcomes. Ensemble generation of probable true rainfall is an elegant and practical solution to characterize the uncertainty of RR estimates and behavior in the hydrologic forecast system. In this study, we have proposed a fully formulated uncertainty model that can statistically quantify the characteristics of the RR errors and their spatial and temporal structure, which is a novel method of its kind in the radar data uncertainty field. The error model is established based on the distribution of gauge rainfall conditioned on radar rainfall (GR|RR). It’s spatial and temporal dependencies are simulated based on the t-copula function. With this proposed error model, a Multivariate Distributed Ensemble Generator (MDEG) driven by the copula and autoregressive filter is designed and applied in the Brue catchment (135 km 2 ), an extensively gauged site in the United Kingdom. The products from MDEG include a time series of ensemble rainfall fields with each of them representing a probable true rainfall. A series of tests show that the ensemble fields generated by MDEG have realistically maintained the spatial and temporal structure of the random error in RR as they have relatively low mean absolute errors (MAEs) of spatio-temporal correlation towards the observed ones. In addition, the results show that the simulated uncertainty bands derived by the 500 realizations of ensemble rainfall encompass most of the reference rain gauge measurements, indicating that the proposed scheme is statistically reliable.