In this study we explore the potential benefits of applying a cloud-resolving ensemble prediction system (EPS; 2.2 km grid spacing) over its driving synoptic-scale limited-area EPS (COSMO-LEPS, 10 km grid spacing, driven by the ECMWF EPS) for a case of heavy precipitation over the Alpine region. The selected event is the devastating August 2005 flood that affected the northern Alpine slopes. The cloud-resolving EPS includes an explicit treatment of deep convection and dynamically downscales the COSMO-LEPS information. Results are compared against rain-gauge and radar data. Furthermore, the sensitivity of the results to initial versus lateral boundary uncertainties are analyzed using a series of additional simulations.
Comparison of the cloud-resolving and its driving limited-area EPS pinpoints the high skill of both ensembles in simulating the major phase of heavy precipitation. The high-resolution EPS yields more realistic rain amounts, in particular in areas of active convection, but in general the resolution-induced differences tend to be smaller than typical member-to-member variability. The differences between the two ensembles can be tied to the synoptic situation (stratiform or convective precipitation, location of the cyclone), to the mesoscale interaction of the flow with the topography (flow over the Alpine ridge), and to the experimental set-up (lead time and computational domain).
For the considered event and set-up, the growth of initial perturbations dominates over lateral boundary uncertainties during the first ∼12 integration hours. Afterwards, the ensemble spread is controlled by large-scale error growth advected from the lateral boundaries into the domain. Copyright © 2008 Royal Meteorological Society