Wind and turbulence evaluation of the ICON model (icon-2026.04) using Doppler lidar observations

Turbulence parameterization in numerical weather prediction (NWP) models remains a challenge, particularly as resolution continues to increase. Existing turbulence evaluation methods often rely on high-resolution benchmark simulations which commonly depend on idealized assumptions and boundary conditions. An alternative evaluation method is presented here, based on Doppler lidar (DL) retrievals of wind and turbulent properties in the lower atmospheric boundary layer providing a more realistic representation of atmospheric turbulence. Modern DL retrieval methods enable the derivation of spatio-temporal and physically consistent profiles for wind and turbulence variables, including the turbulent kinetic energy (TKE), eddy dissipation rate (EDR) and turbulent length scale, which are key parameters commonly used in turbulence parameterization. These can be directly compared to their model equivalents, yielding insights into model deficiencies. The evaluation method is demonstrated here by applying it to the TKE scheme “Turbdiff” used in the ICOsahedral Nonhydrostatic (ICON) model, run with 2.1 km horizontal mesh size in the regional NWP configuration at the German Weather Service. Diagnostics for an equitable comparison of observed and simulated TKE are applied, accounting for model contributions on scales equivalent to those of the DL retrieval. Results show that the model successfully represents the broad patterns of the boundary layer temporal evolution over a five-day summer period with convective days and stable nights. Discrepancies are mainly found at night for near-surface layers below the low level jets that form under stable conditions. Here, excessive mixing leads to an overestimate of wind speed, TKE and EDR. Errors in winds are smaller than in TKE and EDR, as expected given the higher uncertainty of parameterized turbulence. The turbulent length scale formulation, stability functions and prescribed minimum diffusivity are identified as potential candidates contributing to this model bias. The TKE diagnostics applied to the model to approximate the scales sampled by the DL retrieval also allow insights into the relative contribution from subgrid-scale and grid-scale processes and indicate that the model is able to flexibly re-partition these contributions according to the dominant scales of the flow. Lastly, the relevance of the demonstrated model performance is illustrated for two applications: the estimation of mixing layer height from EDR for dispersion modeling, and turbulence intensity derived from TKE for applications in the wind energy sector.