| 19 Jan 2026
Wind and turbulence evaluation of the ICON model (icon-2024.01-1) at sub-kilometer scales using Doppler lidar observations
Abstract. Regional numerical weather prediction models are increasingly run at sub-kilometer scale horizontal resolutions, where turbulence can no longer be considered as an entirely sub-grid scale process (the ”turbulent gray zone”). Existing turbulence evaluation methods often rely on high-resolution benchmark simulations. We present an alternative evaluation method based on Doppler lidar retrievals of winds and turbulent properties in the atmospheric boundary layer.
Two configurations of the ICOsahedral Nonhydrostatic (ICON) model are compared for horizontal mesh sizes ranging from 2.1 km to 78 m: the operationally used 1D TKE scheme "Turbdiff" which includes scale-adaptive features and accounts for some 3D turbulent terms, and a 3D Smagorinsky scheme, commonly used in large eddy simulations. Winds, turbulent kinetic energy (TKE), eddy diffusivity rate (EDR) and turbulent length scale from the Doppler lidar retrieval are used in the evaluation. Diagnostics for an equitable comparison of observed and simulated TKE are applied, accounting for grid-scale and sub-grid scale model contributions.
Results show that winds remain very similar across all resolutions and between configurations, while pronounced differences are found in TKE and EDR. The 3D Smagorinsky configuration performs poorly at coarse resolutions, overestimating both TKE and EDR, but shows comparable performance to Turbdiff at 78 m. Turbdiff more successfully re-partitions TKE between grid-scale and subgrid-scale contributions, roughly preserving the total TKE across mesh sizes. Both configurations show systematic errors in the stable nocturnal boundary layer, coinciding with poor representations of the turbulent length scale.
This study demonstrates the value of the Doppler lidar retrieval as an evaluation tool, shows that the scale-adaptive Turbdiff scheme suffers no loss of performance into the turbulent gray zone and highlights specific aspects of the scheme that limit performance for stable night-time conditions. We illustrate the relevance of the demonstrated model performance 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.
This manuscript presents a detail-rich model validation sudy with a focus on wind structure and turbulence kinetic energy (TKE) with LIDAR observations. At a measurement site in Northeastern Germany, the authors employ the ICON model with several horizontal grid spacings (while a special focus is laid on ICON-D2, dx=2km and higher-resolution research setups in the hectometric range). Furthermore, two turbulence parameterizations are employed for all resolutions: the turbdiff scheme (roughly 1D, progrnosic TKE, based on Mellor-Yamada) and a 3D scheme based on the classical Smagorinsky closure. For validation, several innovative methods are employed, e.g., singular value decopmoisition, comparing cross-sections, integral length scales, etc.
The manuscript is very long (more than 30 pages) and rich in detail, while its purpose is not yet entirely clear. Right now, it rather reads like an internal "state of the art" report for the ICON modelling community, but it would profit from a more clear formulation of research questions, and a clear motivation statement for the methods used. Furthermore, I am very concerned on the "comparability" of the schemes themselves, since they are coupled to two different surface transfer schemes. Some passages are difficult to understand for readers who are not familiar with ICON's turbulence schemes. Most of my current concerns are on the methods chosen, and I think the authors have to analyse more closely the dependence of the turbulence schemes on the sensible heat flux and put the results into context to this challenge. Therefore, I have not yet provided detailed comments to the results themselves. I suggest major revisions for this manuscript.
Methods and their description
I have several concerns on the methods used in this manuscript, their description is not easy to follow, some of the comparison methods seem to be unfair towards the 3D scheme, and the averaging methods to calcuate the resolved TKE are questionable (at least based on the information provdided):
The two turbulence schemes and their behaviour are the centerpiece of this manuscript. However, their description in the methods part of the manuscript is severely lacking.
As a suggestion, I would first describe turbdiff scheme in a subsection, and then the Smagorinsky schem,e, and avoid the "jumping" inbetween schemes in the description.
Furthermore, the description of the turbdiff scheme in lines 135-150 is extremely difficult to follow. The reader does not know the equations of the scheme by heart. Therefore, I would strongly suggest to add the governing equations of this scheme to the manuscript (or to the appendix). Questions which pop up during reading this passage are, for example:
- what is a "classical 1D scheme with 2.5 order closure"? What does this mean? Does this stem from the classical turbulence closure assumptions or from the Mellor-Yamada framework?
- How are the horizontal shear terms implemented? Similar to Goger et al. (2018), or with other horizontal length scales? As far as I remember, the implementation was tricky on the icosahedral grid.
- Which "scale transfer terms" have been implemented for scale-awareness of the scheme? Is there a reference?
- How is "the turbulence length scale" mesh-size dependent? Horizontal or vertical? Do you mean the vertical length scale after Blackadar? (as in Buzzi et al., 2011 Goger et al., 2018)?
- Please elaborate of "turbulence generated by organized motions" versus "true isotropic turbulence". Why is this necessary? And how do you determine L_P_max? Is this a tuning parameter?
Therefore, I would suggest that the authors add the giverning equations of the two turbulence schemes to the manuscript instead of just introducing them in a very descriptive manner - which would also allow the interested reader to understand them better.
Different surface coupling schemes for turbdiff and 3D Smagorinsky
Given the curent data, it is not possible to compare the schemes in a "fair" way, because they are coupled to different surface coupling schemes, which strongly influence the surface sensible heat flux, and henceforth one of the major forcings of turbulence generation (buoyancy production). Therefore, it is challenging to attribute any deficiency in the turbulence properties to the turbulence scheme itself within the comparison of the two schemes. Therefore, I think it is impossible to say which scheme performs better or worse, because many issues, which are visible in the turbulence schemes (e.g., overstimated TKE), might actually be present due to the different surface transfer schemes. Previous studies suggest that the turbtrans scheme performs better during convective days than the classical (and much more less complex) Louis scheme (e.g., Goger & Dipankar, 2024).
I do not think that this a reason to reject the manuscript, though. However, I would stroingly suggest to the authors that they address the different sensible heat fluxes and the schemes in an additional analysis, for example with a correlation plot of the dependence of the relevant turbulence paramters on the surface sensible heat flux (possibly also with observations of the sensible heat flux, I assume you have them from the EC tower). Otherwise, the entire comparison between the schemes is inherently unfair, because the turbulence schemes do not have the same "starting point" (i.e., sensible heat flux input.)
In general, a fair comparison between the two schemes would only work with prescribed surface fluxes - or in a setting where both schemes are coupled to the same surface transfer scheme.
Appendix, lines 773- 783
The description here is again difficult to follow, and my remark here is mostly to invite the authors to double-check the analysis method.
Please describe first, how TKE is calculated from the observations - and then - in a separate subsection - how it is calculated from the model. However, my largest concern appears at line 765:
"Grid-scale TKE is calculated for the SKM simulation winds saved at 5 min intervals." - Are these intervals already averaged temporally?
From this description, it is not entirely clear how the temporal 5-minute data are derived. Is this instanteneous model output every 5 minutes? Or are the 5-minute intervals already averaged from high-frequency model output, for example at every time step (i.e., meteogram output)? Currently, I think the only (realistic and reliable way) to obtain high-frequency output in the ICON model is via the meteogram option for selected grid points. If the 5-minute intervals indeed stem from 5-minute instantaneous output (non-averaged), which I hope is not the case, it is impossible to derive the resolved TKE (for both schemes), because turbulent motions are relevant on time scales much shorter than 5 minutes (or even one minute) - you would need model output at every time step. This would detoriate all your findings in Section 5.1 (and their interpretation); and furthermore, this methodological flaw would be a reason for rejection of the MS, especially when you interpret the data from the sub-hectometric range.
Further questions on the methods:
- How do you deal with changing horizontal resolution in your spatial averages?
- What is is the difference between grid-scale averages and SGS time averages? Given the description, it sounds similar.
- line 773: are the 5-min 'sub-intervals' already averaged?
Minor comments
line 24: Please cite relevant references for the turbulence gray zone, e.g. Honnert et al, 2020, and Wyngaard er tal. (2004)
line 33: What is the "standard ICON turbulence scheme"?
lines 39-46: Since the turbdiff scheme uses a prognostic TKE equations, it would be fair to mention TKE budget-based model validation methods as suggested by Goger et al, 2018, Nilsson et al (2016), and Rohanizadegan (...)?
line 82: Please describe the location and its surroundings a bit closer - not every reader knows the observatory and its location. Furthermore, where exactly is the instrument placed? After reading the entire manuscript, I would strongly suggest to add a figure showing the model domains and adding the location of the DL on the map.
line 121: Here - at latest - adding the figure with the model domains would be appropriate.
line 126: "SKM horizontal resolution" - maybe I've missed it, but what does this abbrevation mean?
line 155: "3D like extensions". This is very ambigous. What 3D effects are you talking about?
line 162: It should be clarified that the Blackadar length scale is vertical
line 175: at this point (at latest), the prognostic TKE equation should be introduced.
line 178: Could you please provide a reference on the equation to estimate EDR from the Smagorinsky scheme?
line 176: "Underlying the Smagorinsky scheme is the assumption that turbulence
generation via shear production and buoyancy production/destruction is balanced by dissipation" - this is not specific to the Smagorinsky scheme, but a general concept in boundary-layer meteorology (Stull, 1988).
line 194: "ICON turbulence scheme" - be consistent with the naming of the schemes. Both turbdiff and 3D-Smag are "ICON turbulence schemes".
line 208: You might inducate somewhere that cold pools are associated with convective systems, i.e., thunderstorms
line 214: I am aware that this is not the focus of the MS, but to double-check, you could look at RADAR composits to make sure...
line 380-81: Could it be that these unrealisically high TKE values are related to the higher sensible heat fluxes compared to the turbdiff scheme? So the scheme itself is likely physically consistent, but you just have a slightly different surface forcing.
References
Buzzi, Matteo; Rotach, Mathias W.; Holtslag, Matthias; Holtslag, Albert A.M., 2011: Evaluation of the COSMO-SC turbulence scheme in a shear-driven stable boundary layer, Meteorol Z., DOI: 10.1127/0941-2948/2011/0050
Goger, B., Rotach, M.W., Gohm, A. et al. The Impact of Three-Dimensional Effects on the Simulation of Turbulence Kinetic Energy in a Major Alpine Valley. Boundary-Layer Meteorol 168, 1–27 (2018). https://doi.org/10.1007/s10546-018-0341-y
Goger, B. & Dipankar, A. (2024) The impact of mesh size, turbulence parameterization, and land-surface-exchange scheme on simulations of the mountain boundary layer in the hectometric range. Quarterly Journal of the Royal Meteorological Society, 150(763), 3853–3873. Available from: https://doi.org/10.1002/qj.4799
Honnert, R., Efstathiou, G., Beare, R., Ito, J., Lock, A., Neggers, R., et al. (2020). The atmospheric boundary layer and the “gray zone” of turbulence: A critical review.Journal of Geophysical Research: Atmospheres, 125, e2019JD030317. https://doi.org/10.1029/2019JD030317
Nilsson, E., Lothon, M., Lohou, F., Pardyjak, E., Hartogensis, O., and Darbieu, C.: Turbulence kinetic energy budget during the afternoon transition – Part 2: A simple TKE model, Atmos. Chem. Phys., 16, 8873–8898, https://doi.org/10.5194/acp-16-8873-2016, 2016.
Rohanizadegan, M., Petrone, R. M., Pomeroy, J. W., & Kosovic, B. (2025). Analysis of turbulence and turbulence kinetic energy dynamics in complex terrain. Journal of Geophysical Research: Atmospheres, 130, e2023JD040558. https://doi.org/10.1029/2023JD040558
Wyngaard, J. C., 2004: Toward Numerical Modeling in the “Terra Incognita”. J. Atmos. Sci., 61, 1816–1826, https://doi.org/10.1175/1520-0469(2004)061<1816:TNMITT>2.0.CO;2.