Turbulence evaluation of the ICON model at sub-kilometer scales using Doppler lidar observations

Regional numerical weather prediction models are increasingly run at sub-kilometer scale horizontal resolutions, approaching the turbulent gray zone where turbulence can no longer be considered as an entirely subgrid-scale process. This requires scale-adaptive parameterizations that respond consistently to changing resolution, reducing the parameterized contribution as more of the turbulent transport is explicitly represented by the model dynamics.

The ICOsahedral Nonhydrostatic (ICON) model used for operational weather forecasting at the German Weather Service (DWD) uses the TURBDIFF turbulence parameterization, which includes some scale adaptive features. In order to assess the performance of the scheme for horizontal mesh sizes ranging from 2.1km to 78m, we present an evaluation method based on Doppler lidar retrievals of winds and turbulent properties, including the turbulent kinetic energy (TKE), eddy diffusivity rate (EDR) and turbulent length scale within the lowest 600m of the atmospheric boundary layer. We use observations from the Lindenberg observatory over a five-day period in June 2023 with typical daytime convective boundary layers, and stable conditions with low level jets observed at night.

To facilitate a fair comparison, grid-scale and parameterized, subgrid-scale contributions to the simulated TKE are considered consistent with the spatio-temporal scales of the Doppler lidar scan configuration.

Results show that predicted winds remain very similar across all resolutions, while subtle differences are evident in TKE and EDR. This suggests the scale-adaptive features of TURBDIFF turbulence scheme work reasonably well and result in similar validity of the turbulent properties across all scales, though the uncertainties in the simulated turbulence properties vary with time of day. While it is gratifying that the scheme shows no deteriorating performance at higher resolutions, it is somewhat disappointing that we do not see a clear benefit of the increased resolution reflected in the predicted winds either. This may point towards potential limitations in how the turbulence scheme interacts with the resolved dynamics. In addition, systematic errors in the stable night-time boundary layer are evident in all simulations, which coincide with poor representations of the turbulent length scale compared to observations.

Thus we demonstrate the usefulness of the Doppler lidar retrieval as an evaluation tool and highlight specific aspects of the scheme that limit performance for stable night-time conditions.