The impact of downscaled initial condition perturbations on convective‐scale ensemble forecasts of precipitation

A major issue in convective‐scale ensemble prediction systems (EPSs) is the specification of effective initial condition perturbations (ICPs). The present work considers the suitability of downscaled ICPs from a multi‐model global EPS for short‐range regional ensemble forecasts of convective precipitation at convection‐permitting resolution. Previous studies indicated the importance of convective‐scale initial condition uncertainties, with the most pronounced impact in weather conditions characterised by weak versus strong synoptic‐scale forcing of convection. However, the downscaled ICPs do not explicitly represent small‐scale uncertainty, which questions their effectiveness in convective‐scale EPSs. To investigate the issue, the high‐resolution ensemble system of the Deutscher Wetterdienst, COSMO‐DE‐EPS, which includes physics perturbations and lateral boundary condition perturbations in addition to ICPs, is employed. Forecasts are compared with a second EPS, identical but without ICPs, for a period of 3.5 months in the central European warm season. Weakly forced conditions are considered separately from strongly forced conditions, using an objective classification based on the area‐averaged convective adjustment time‐scale. Generally for all EPSs, forecast quality measures show a distinct behaviour in strong versus weak forcing conditions. However, the impact of the ICPs is found to be similar in the two regimes. The impact of the ICPs is clearly largest and positive (consistently in terms of ensemble variance and probabilistic forecast quality, but negative for the equitable threat score) in the first six forecast hours when the ICPs dominate the physics perturbations and lateral boundary condition perturbations. The ICPs then decay relatively quickly with lead time as the physics perturbations and lateral boundary condition perturbations start to become important and later dominant. Probabilistic precipitation forecasts by the EPSs outperform the deterministic COSMO‐DE at the same convection‐permitting resolution, and this more strongly in the first nine forecast hours with the EPS applying the ICPs.