Thank you for your submission. Microscale data assimilation is an important tool to have in the simulation toolbox but can be challenging to apply in practice, especially for LES. The authors demonstrate a recently developed method for data assimilation that appears attractive because when applied to a precursor LES flowfield, it can preserve the preexisting resolved turbulence. Reasonable steps have been taken to demonstrate the approach given different starting precursor simulations and in the end, results are shown for a wind turbine immersed in a near-neutral flow.

I think an advantage of using this approach is that it does not assume horizontal homogeneity like the work of Allaerts et al 2020, which allows for more general data assimilation scenarios—for example, assimilating simultaneous measurements or applicability in complex terrain. However, the effectiveness or applicability of this approach for nonstationary conditions is not clear to me. The vibration assimilation approach is essentially an integral controller, which has known limitations. My concern is about the time lag associated with this forcing strategy. Perhaps an assimilation strategy that replicates a proportional–integral controller would be more appropriate. 

It would be useful for the authors to show how the turbulent flow statistics downstream of the nudging zone evolve in time and space to inform the application of this assimilation technique. Recommendations for choosing the vibration frequency would also be helpful, given the sensitivity of downstream turbulence to this parameter. If you start from one LES and nudge toward another LES solution, do you recover the same turbulence as the target LES?

My biggest concern about this work is how Newtonian relaxation has been written off because the assimilated flow has reduced turbulence. If I understand the implementation correctly, the instantaneous velocity at each point within the nudging zone is relaxed from the turbulent flow field towards a mean profile. Therefore it is not surprising to me that the precursor turbulence is reduced or eliminated. A more reasonable approach and fair comparison would be to relax the horizontal mean within the nudging zone towards the target mean profile. This would constitute a localized version of the "direct" profile assimilation from Allaerts et al 2020. 

Please see the attached annotated PDF for more specific comments. I have elected not to review a revised mansucript not because I am not interested but because I will be on family leave in the near future. 

This article investigates the use of data assimilation to provide forcing for LES modeling of ABLs with the intent of modeling wind-turbine impacts. The paper presents comparisons between Newtonian relaxation (nudging) and another, more sophisticated approach proposed by Nakayama and Takemi (2020) based on the vibration equation with an imposed frequency. The authors compare the two approaches and highly idealized conditions and then include an example where the approach is applied to a combination of vertical lidar profile blended with mesoscale WRF data. Generating turbulence consistent with a specified forcing conditions is a topic of relevance to wind energy studies. However, there are a number of critical fundamental issues that need to be addressed and that make the study in its current (and very preliminary state) non suitable for publication in Wind Energy Science. These major concerns are outlined here below.

Major Comments

• The manuscript (starting from the title) is full of quotes and claims of this data assimilation method to be able to generate “realistic inflow fields”. There are a number of reasons because of which this is actually not the case, and it is in fact the opposite. Firstly, the method does not consider data assimilation for buoyancy and moisture effects, as only considers forcing terms for the momentum equations (Eq. 5). Secondly, it relies on a single vertical profile and evolves conditions from an idealized, flat, laterally periodic ABL, leading to homogeneous forcing. Even if the authors use a profile from observations to assimilate mean wind speed forcing, a single local observation is typically not in equilibrium neither represents spatially averaged conditions properly, which does not imply any realism as heterogeneous effects are not accounted for (both in space and time). These crucial aspects make the method exclusively suitable for highly idealized conditions. The tone of the paper comes across in the current form as excessively overselling of the approach, not outlining any of the limitations, and needs to be significantly altered to provide a fair view of what the method brings to the table and what the limitations are.
• A significant portion of the manuscript is devoted to repeat the results from Nakayama and Takemi (2020). All these results and discussion do not provide any new insights besides showing that the implementation is correct. Nakayama and Takemi (2020) already demonstrated that the “nudging” approach is not good in order to produce reasonable turbulence quantities, so all that part of the manuscript is redundant. I would recommend the authors to remove the majority of that content, and perhaps move a minimal part of these results to an appendix if at all.
• It is not surprising that assimilating a wind speed profile would lead to a matching velocity field within the area of the domain where the assimilation is applied, as the influence from the governing equations is being overpowered by that forcing term. The emphasis, which the authors attempt to provide in this study, is the quality of the resulting velocity fluctuations (i.e., turbulence). First, the authors should refrain from using turbulence intensity (TI) as a metric for comparison. TI is a very misleading derived quantity. I understand engineers like it, but you can have the right TI with properly offset wind speed and TKE. Please use TKE for the analyses instead (same for Reynolds stress). Secondly, to that end, run a precursor case with equivalent forcing so you can properly assess the skill of the approach, otherwise it is impossible to judge the adequacy of the results.
• Averaging over a spatial area is not a good idea. While that approach will inevitably lead to smoother results, it does implicitly hide any spatial variability. In the end, this method will be used in a limited area domain, as shown in Fig. 11. The authors need to quantify the spatial variability as the flow moves out of the nudging region. This evolution of the wind field is evident from Fig. 11 when examining x-direction evolution for abs(y/D) > 1.0. Please perform proper spatial analyses to understand this key practical aspect.
• The turbulence analyses need to be more rigorous and comprehensive. Again, the first-order mean may be captured somehow, but that does not guarantee proper balanced turbulence. The authors need to include energy spectra computed over time and show how the data assimilation approach alters the energy distribution across scales due to the oscillatory, single frequency nature of the assimilated forcing. Also, length scales are important. The authors need to show instantaneous flow fields to get started with, and then dig deeper into more careful and systematic comparisons of turbulence quantities.
• An aspect that appears to be essential to the method and that should be explored in the manuscript is how the disparity between the reference LES data and the forced profile influences the required area where the assimilation is applied, as well as how the amplitude of the forcing needs to be adjusted. Please explore different reference LES and target profiles to elucidate this aspect. Otherwise, practical applicability of the method cannot be guaranteed.