Radar Data Assimilation in Idealized Experiments
Several weather services develop convection-permitting, limited-area ensemble prediction systems as, for example, COSMO-KENDA at DWD. In this context, spatially and temporally high resolved observations, such as radar data, to provide perturbed initial conditions, become increasingly important. We investigate practical predictability limits of convective precipitation in such a state-of-the-art system and their sensitivities to the type of assimilated information, observation errors and the time of assimilation in a perfect model experiment. The perfect model environment also allows for an investigation of the impact of initial and lateral boundary condition errors in this context.
For horizontally homogeneous, convectively unstable conditions favouring long-lived convection, we find skilful predictions for scales as small as 30km after 6h lead time. Experiments with larger observation errors or different observation types, like wind or reflectivity, depict decreased predictability, while scales above 100km still remain predictable. Experiments with realistic lateral boundary condition errors exhibit inherently less skill but the assimilation of storm-scale features is able to correct synoptic errors in the forecast. Additionally, we find that the presence of orography in the experiments increases the predictability, as it acts as a trigger for deep convection. This positive effect decreases with improving quality of the initial conditions for up to 6 forecast hours. Thereafter, the orography exhibits an increasing impact on the forecast skill again.
Mr. Kevin Bachmann DWD / LMU