Physically based stochastic perturbations improve a high‐resolution forecast of convection

Abstract A physically based stochastic perturbation (PSP) scheme has been implemented in the convection‐permitting ICON‐D2 ensemble prediction system at Deutscher Wetterdienst (DWD) and run for a three‐month trial experiment in summer 2021. The scheme mimics the impact of boundary‐layer turbulence on the smallest resolved scales and impacts convective precipitation in particular. A weather‐regime‐dependent systematic evaluation shows that PSP efficiently increases ensemble spread of precipitation in weak synoptic forcing, while producing realistic convective structures. During strong forcing, the effect of the scheme is negligible, as expected by design. A probabilistic verification shows improvements in the forecast skill of other variables as well, especially the spread‐to‐skill ratio, but identifies starting points for further improvements of the method.