Appropriate dispositions in the case of floods require longer lead times in both, meteorological and hydrological forecasting. This in turn implies an increased uncertainty, which cannot be accounted for by a deterministic forecast. An attractive possibility to address this issue is the use of probabilistic forecasts driven by meteorological ensemble prediction systems (EPS), as the meteorological input is considered to represent the main source of uncertainty in small to intermediate catchments.
Ensemble runoff forecasts with 51 ensemble members were conducted for three case studies (spring 1999, November 2002 and August 2005) in the upper Rhine catchment using as input ECMWF ensemble forecasts that were dynamically downscaled to a horizontal grid-spacing of 10 km. A comparison with deterministic forecasts reveals an improved forecast guidance with appropriate uncertainty intervals. Additionally, daily hindcasts for the year 2005 were carried out, using the operational COSMO-LEPS meteorological ensemble, which is a subsample of 10 downscaled ECMWF members, selected by a cluster analysis. Analysis of the results shows, that the clustering methodology does not overly reduce the ensemble spread. The analysis of the continuous runoff forecasts focuses on the evolution of the ensemble spread and skill scores over time, to evaluate the forecast performance.