Nice effort to simulate a realistic ABL evolution during a wake steering experiment. Capturing the interplay of the inflow with the wake dynamics is a notable challenge and the author made a good job addressing some of the most important couplings with high-fidelity and engineering models. 

Getting the inflow conditions right is the main challenge and it is there where I have focused my attention while reviewing. Different profile assimilation techniques are discussed but it is not justified why we need to add such complexity to a case study that seems to be dominated by surface-layer conditions and single-point assimilation may be a more effective solution to track mesoscale variability. 

Another topic of interest is to which extend we can extract some quantification of the impact of mesoscale inflow bias at the reference meteorological site and the bias due to horizontal heterogeneity not simulated explicitly by the models. By looking at the normalized power in T2, it is observed that these biases can lead to up to 20% differences in gross power. A fairer assessment of the wake models would be obtained if those biases could be quantified to gage their relative role in the assessment of wake steering.

In summary, I think it is a good case study to highlight the challenges of simulating realistic operational conditions departing from the traditional canonical microscale setup. Still, the large uncertainties in the inflow conditions make it extremely hard to isolate the assessment of wake steering effects. Maybe we need to change the orientation of profile assimilation techniques from the vertical to the horizontal direction. We need the inflow bias to be in the order of a few percent of normalized power to study wake steering effects that can stand out of this bias.      

Additional remarks

12: "demonstrating agreement with measurements": I'm not sure I would agree with such a general statement. Can you be more specific or also mention where there is disagreement?   

22: Nonstationarity is intrinsic to atmospheric flows so it really happens all the time although it is more notable in the presence of weather events and diurnal morning/evening transitions as the authors indicate. Maybe it would be worth speaking about “large-scale nonstationarity” in the context of this paper, to differentiate from the more traditional microscale nonstationarity that leads to quasi-steady canonical conditions, a.k.a. “microscale LES”.

92: Fleming et al (missing year)

115: Can you describe how you calibrated the turbulence intensity of the lidar from the cup anemometer measurements? How far off where the lidar measurements at the calibration levels?

121: Can you provide an estimate of the uncertainty of the yaw position achieved after applying the corrections to the SCADA signal?

220: Why using sounding data from 340 km instead of local mesoscale predictions? I wonder if the authors checked the accuracy of the mesoscale data at the sounding site. This could give some guidance as to whether the mesoscale data could be used reliably instead of relying on distant measurements. Would be nice to include the mesoscale profiles in Figure 4 to get a sense of the differences in this indirect validation.

336: There are notable differences of up to 20% in the power of T2 compared to the average from freestream turbines T1 and T5 in conditions without wake steering control. Is this not an indication of predominant heterogenous inflow? It is very challenging to assess potential net gains of a control strategy of a few percent when one of the hypothesis of the model can lead, by itself, to differences of up to 20%.  

Similarly, one could try to estimate the impact of the differences observed in the reference inflow conditions (Figure 9), especially regarding wind direction differences which result in an indirect yaw offset. Is it possible to infer from measurements or simulations the impact of these inflow biases in the freestream power?

These estimations, in the form of power ratios, would provide a more comprehensive framework to judge the relative importance of each phenomena individually so we can make more informed conclusions about their relative role when all factors are combined.

403: Concerning the data assimilation techniques it is not clear how the authors ended up choosing the DPA method over the much simpler single-level method. From Figure 8 it seems that the single-level method produces a more accurate representation of the friction velocity during the wake steering experiment while it shows similar performance than the other models in the other quantities. Maybe the case is not large enough for tall profile features to play a role and the flow is dominated by the bottom part of boundary layer, which can be calibrated with a single point to capture the mesoscale tendencies.

In this study, the author did a great job discussing and addressing some of the most challenging aspects of MMC couplings. Ability to accurately simulate conditions wind turbines are exposed to, especially in case of a complex terrain, persists as a challenge for the wind energy community. Better documented profiles, appropriate assimilation technique and accurate inflow conditions are essential component in addressing those challenges and reducing uncertainly that can lead to a significant bias in projected power production. I think study is a great example of challenges one would face when simulating more realistic ABL conditions, and those that wind energy community needs to overcome.  

Some specific comments:

Line 12: "agreement with measurements" based on presented results, I feel that author should rephrase this a bit.

Line 92: “Fleming et al” here is missing a year.

Line 220: Author used sounding that is 340 km upstream from the study location. How far is second available sounding for the same time period and would it be useful to do some sort of spatial interpolation or comparison between those location for a possible better informed profiles of wind speed and virtual potential temperature? I understand that WRF was not able to properly simulate specific local conditions, but was WRF able to properly simulate wind and temperature profiles at some location closer than 340km (location of the closes sounding), and would it be better to use some of those closer to domain location profiles?

Line 326: Maybe I missed it, but what was a rational of using partial-wind-profile DPA compared to full-wind-profile DPA?

I highly agree with the statement that “the extent to which field conditions are reproducible with MMC depends on the nature of the background physical phenomena and their observability” and think that this work is a very good example of how tools and techniques presented in this study can improve prediction of local features and conditions only for a limited extend.

The submitted paper investigates an approach to simulate a published wake steering field campaign in large eddy simulations (LES) that derive forcing from the limited field measurements available. Overall, this paper is very valuable to the community, and an impressive and detailed effort to simulate more realistic atmospheric conditions in microscale LES. 

The paper is overall well written, and includes a lot of detail on certain aspects, but more limited details elsewhere, given the nature of applying existing methods and approaches to a given case study. My overall perspective is that this paper would be both easier to follow, and in my opinion more impactful to the community beyond users of NREL codes, with a restructure. Specifically, I’d like the methods section to detail meso-microscale coupling (MMC) methods that could be used in general, and for the results to intercompare the chosen methods. As is, Section 4 introduces one MMC method that has been seemingly preselected as the best, but then four different MMC methods are evaluated in results, with a 5th method being WRF-LES MMC, which was simulated but is not shown apparently because of poor performance (would be better to show). I think if the author approaches the paper as an intercomparison, with a discussion of benefits and limitations of each method, it would be easier to follow, and also useful to the community.

Further, there are a lot of choices in the simulation setup. This is not a criticism, in fact, this is part of what makes this paper useful and impressive. But somehow, it would be nice to very clearly and concisely summarize the choices made (not a complete list: guessed roughness length, potential temperature B.C., initialization with sounding data from far away ad hoc combined with in situ temperature measurements, spin-up period, power-law fitting, actuator disk model, Boussinesq approx., horizontally homogeneous flow, MMC method, etc.). Many of these choices were selected based on intuition, some were justified based on empirical results, and some were tested in sensitivity tests. It would be really helpful to just clearly list this, discuss why each decision was made in a concise summary, and to highlight the need for future work. If someone else wanted to start where this paper ended, why should they make similar or different decisions on the setup?

Finally, I feel that the ‘final’ wake steering results in Figure 10 are somewhat unconvincing, due to the multitude of choices and uncertainties discussed above, and due to the short term simulation window. It seems hard to make any conclusions from these data, as I listed below in the point comments.

Point comments

1. Line 73: Can you clarify what is intended by this statement“Not only is NWP not able to resolve local conditions, the available remote-sensing observations are limited.”
   
   It seems to me from the detailed literature review in the preceding paragraphs that the application of interest in this study would be well-suited for MMC using NWP for the large-scales, if in situ measurements are limited.
2. Line 100: Please clarify the filtering/time-averaging associated with the wind speed and direction that feed into the lookup table.
3. Line 115: “Lidar-measured turbulence intensity above the 60-m-tall met mast was corrected based on lidar and cup-anemometer measurements.”
   
   Could you describe the technical details of this more specifically?
4. Section 2.2: I wonder if this section and associated discussion elsewhere in the paper, can be reorganized in three parts: 1) what input data are required by the MMC approach; 2) what data are available during the field campaign; 3) what methods were used to fill gaps/missing/imperfect information. To be clear, I thought point 1 (what data are required for your specific MMC approach) was most unclear in the study.
5. Section 2.3: What is the maximum time period that could be simulated given computational constraints? The reason I am asking, is because given all of the filter constraints, only one hour of data is quite limited.
6. Line 156: This sentence self-references the section that the sentence appears in
7. Section 3.1: I suggest including an Appendix that contains the mesoscale modeled results and associated comparisons with measurements to support this discussion.
8. Section 3.2: It seems to me that the filtering approach discussed in Section 2.3 has resulted in a set of complex meteorological conditions. Not a positive or negative per se, but maybe it would have been useful to start with a simpler case before moving to a more complex one.
9. Line 183: “The third and final assumption for this case study [...]”
   
   It may be easier to read for most if you state this assumption in terms of the standard Boussinesq approximation.
10. Line 219: Is this a typo? The sounding data are 340 km away? Why were these data used instead of the pressure, temperature, and humidity sensors mentioned in Table 1?
11. Line 227: “To adapt the nearest sounding to local conditions, the lowest 200 m of the wind and virtual potential temperature profiles were replaced with local site measurements”
    
    Did this sharp discontinuity in the field cause any challenges or instabilities? Are the results sensitivity to this choice, rather than the alternative approaches of using only the sounding or using only the local measurements? Is the capping inversion in Figure 4(c) only the result of the jump between local and sounding measurements?
12. Line 243: I didn’t quite understand this last line in the context of the paragraph. If I’m reading correctly, it says to use surface temperature instead of heat flux (Basu 2008), but then heat flux is used as the boundary condition for a reason I didn’t quite follow
13. Line 250: Couldn’t this discrepancy also be associated with a mis-specification of the geostrophic pressure gradient?
14. Section 4.3: Why was the power law used when it is well-known to deviate in conditions when stratification is present?
15. Figures 7 and 8: I don’t quite follow how the relatively large errors in Figure 7 translate to relatively low errors (“mean wind speed and direction trends from experiment are generally reproduced.”) in Figure 8. Could you explain this further? For example, from the single-level assimilation approach, I was expecting to see errors in the hub-height wind speed based on Figure 7 and associated discussion, but the spread between DA approaches in Figure 8(a) appears to be very small.
16. Figure 8: Can this figure be made more clear as to which ‘measurements’ are assimilated into each approach (interpolation), and which measurements are not (extrapolation/prediction).
17. Line 314: What do you make of the friction velocity being predicted well even with the guessed value of z_0 and the fact that the surface heat flux is prescribed?
18. Appendix A: More detail on the actuator disk model (ADM) are required. Numerical implementation. Thrust and power coefficients. Power-yaw relationship, etc.
19. Does FLORIS take inputs from the measurements or from LES?
20. Figure 10: Why is the impact of yaw misalignment on T2 power so small in FLORIS?
21. Line 363: Turbulence intensity: the discussion presumes that TI estimates from LiDAR are true, but there is a high degree of uncertainty estimating TI from the variance of averaged LiDAR measurements. Worth mentioning.
22. Line 374: Numerical oscillations from turbine forcing implementations in LES can also be addressed through improved ADM numerical methods [1]
23. Line 383: Where exactly did FLORIS overpredict the effect of wake steering in Figure 10? I am not sure I see an effect that is significant (statistically?). If you mean at 830, FLORIS seems to already disagree with LES at ~807, so it’s not clear what is from wake steering misprediction vs. dynamic errors.
24. After finishing reading the paper, I don’t think I agree with this statement in the Abstract “demonstrating agreement with measurements under partially and nearly waked conditions.” It would be better if you reported the error associated with the background wind conditions (freestream), and the error associated with wake steering predictions (wake effects).

References

[1] Shapiro, Carl R., Dennice F. Gayme, and Charles Meneveau. "Filtered actuator disks: Theory and application to wind turbine models in large eddy simulation." Wind Energy 22, no. 10 (2019): 1414-1420.