The authors present a model evaluation manuscript for evaluating WRF in LES mode above a forest in the vecinity of a double ridge. Modeled wind component velocitites are compared to observations at 100 m tall meteorological towers and LIDAR observations. The authors find that using a forest parameterization in the LES inmproves the agreement between model and observations. The applied forest parameterization is a canopy aware LES in which canopy elements are modeled as a drag coefficient for model layers that are assumed to be within the forest canopy (2-3 bottom most layers). While I don't think that there are technical errors in the manuscript, I have several comments and questions about their work that the authors should address before final publication.

General comments:

1) Canopy aware LES (also sometimes called canopy resolving LES) has become reasonably mainstream in the last 10 years, with many community  LES (i.e. WRF, DALES, PALM) having developed capabilities to represent canopy drag forces. It have been shown that such treatment, is capable of improved representation of turbulent statistics within and above the canopy (e.g. TKE, sigma_w, sweeps/ ejections). It has been shown that a sufficient number of within canopy model layers are needed to do so (order 10 layers) and that results are best when model resolutions are isotropic (i.e dx=dy=dz). Given the current setup of 2-3 model layers and dx/dy of 200 m in d03 (which would produce a grid aspect ratio of roghly 10:200), I am wondering to what extent the canopy drag formulation would be capable of improving representation of turbulent structures above the forest.

This brings me to my main question, which I could not find addressed in the manuscript: To what extent is the comparison between the forest parameterization and non-forest a true apples to apples comparison? The non-forest LES uses z0 from the land-use dataset (I assume this is the built in WRF one, that may not be very accurate at such high resolutions), while the forest parameterization uses a direct sink for momentum. A better comparison may be to apply a similar momentum sink to the first model layer only, or to first calibrate the model's z0 to observed values.

2) LLJ cases: The model evaluation against the LIDAR data is done as instantanious cross section plots. While I can see that there the forest parameterization exhibits more dampening of lee-waves and has somewhat lower near surface wind speeds, it is generally hard to make out finer differences in the plots. I would suggest that the authors think about how to better visualize these model results. For example, one might want to focus on the d04 domain and would also want to zoom in closer to the surface. Alternatively, difference plots would also be approciate. Lastly, I am wondering, given the 5 minute output interval, why comparisons are done for a single snapshot in time. Given the amount of output data, it would be interesting to see whehter there is a better quantiative comparison possible. Furthermore, Wagner et al 2019a presents a model evaluation paper with respect to d03 domain in non-forest mode. Compared to the model evaluation for d03 in the current manuscript, this evaluation appears to be much more detailed and I am wondering why this is the case. 

3) I am not familiar with the data from the field campaign that is available to the authors, but I am wondering whether there is a missed opportunity in not having any kind of turbulent quantities that the LES is evaluated against (e.g. from the towers or the LIDARs). Usually, the fact that LES are capable of partially resolving turbulence and to provide realistic profiles of turbulent statistics that can be compared against observations is seen as one of the crucial adavantages of the LES. In this manuscript, there is no consideration of turbulent quantities, which could be used to evaluate the actual LES.

4) Relationship to retracted Wager et al 2019b: The authors mention in the submission that this work builds on the retracted Wager et al 2019 b work. A cursory look at the discussion of this manuscript shows that two reviewers had issues with the WRF setup in the sense that they would have liked to see either a nudging or periodic restart for the long-term simulation. I would like to know whether these comments have been heeded in the present work. SImilarly, reviewers of the W19b manuscript would have liked to see the full namelist of the WRF for added transparency/ reproducibility. I feel that something like this would be particularly valuable given the objectives of GMD. Lastly, since W19b was retracted, it should not be referenced in this manuscript.

Specific comments:

P1L14: "The grids are now fine enough to fully resolve the atmosphere with techniques such as large eddy simulation (LES)" > I take issue with this statement, as it is incorrect. LES partially resolves turbulence, but this is nowhere near fully resolving (especially given the model resolutions of 200 and 40 m in this paper). This needs to be rephrased.

Introduction: The authors present a literature review of canopy aware LES, that is somewhat odd and starts with Dutton et al 2008 (they do cite Shawman and Schuh, 2003 later), but there are plenty of papers that could/should be referenced including:

Shaw & Patton, 2003: https://doi.org/10.1016/S0168-1923(02)00165-X

Dwyer et al. 1997: 10.1023/A:1000301303543

Review of "Evaluation of a forest parameterization to improve boundary layer flow simulations over complex terrain" by J. Quimbayo-Duarte et al., submitted to Geoscientific Model Development

The paper makes a very interesting simulation over the complex terrain area of the Perdigao-2017 experiment, impressive in terms of computation power when looking at the horizontal resolution and the length of the integration. The sensitivity of the results to the introduction of a parameterization of the canopy in the lower 30 meters above the surface has a clear positive impact for the campaign-long simulation and shows improvement for the high-resolution cases. 

To my opinion, the paper is worth publishing in GMD, although some modifications would be needed. The paper is somewhat uncompensated, as the 49-day run is very succintly described while the changes are relevant, while the high resolution runs are discussed extensively but more superficially. Also the Hovmöller plots are hard to follow as the details explained in the text are difficult to be seen. Sensitivity tests would have been interesting, varying the height of the canopy, and some discussion on the effects of the changes in the temperature profiles (not shown) would have been very welcome. 

Some specific issues are:

1) A discussion on the choice of the vertical resolution, especially the location of the lowest point at 10 m above the surface. Why the lowest level is at this height?  Couldn’t the first level be located closer to the surface as other studies in complex terrain do? Is there a computationally-related reason, such as numerical instabilities on steep slopes, or is it because of some choice related to the applicability of the similarity theory?

2) In fact no explicit explanation is given about the use of the similarity theory near the surface, and what are the equations employed.

3) Justify the choice of the Lalic-Mihailovic shape for the leaf-area density. Is this the only available possibility or are there others and this one has shown to be the best option?

4) A more elaborated discussion of what it means to use a canopy-drag compared to surface roughness would be appreciated. In the latter case it is costumarily assumed that a logarithmic profile is imposed between the lowest point and z0 and that the similarity expressions hold (even if they were originally derived for flat terrain). Instead introducing the LAD profile and the corresponding drag breaks this conceptual model. I think the paper would benefit of a reflection on how the surface layer is different in both approaches, perhaps showing comparison profiles. Also, is similarity theory still used in the lowest level below the canopy profile?

5) The statistics with the modified run are much better than the standard one for the along-valley wind, while the cross-valley is not significally improved. The latter is attributed by the authors to the use of a too high canopy and to the heterogeneous distribution of forested areas. This aspect is superficially commented and could have been complemented with some sensitivity runs of short duration (only two levels with canopy during one day for instance), as it is an important point in the results.

6) I find the plots of the short-term simulations far from clear when differences between using forest or not are commented.  For instance, to my eye it is difficult to find the "southwest flow observed near the ground on the lee of the topography" (lines 195-196). Perhaps a circle on the figure would make the task easier for the reader. In general fig 6 is not good to follow the explanations given in the text, maybe vertical profiles at the points of interest comparing the two runs would be more informative. As it is now, I find myself believing what the authors say. I suggest that the whole discussion concerning figures 6 to 8 is remade with more clear grafic evidence of the results that the authors indicate.

7) when comenting figure 9 it is hard to see almost anything that is commented in text. I would suggest to show perhaps only the diferences between simulation and observation so that the relevant issues jump to the reader easily. Some graphical information on temperature -and the corresponding discussion- is missing too, not only here but all along the text. 

8) Instead I find Figure 10 a good summary representation of what is going on, it would be nicer if it was given together with average profiles for wind and temperature in order to see in what parts of the column the effect of the changes is more evident for wind and temperature.