Dear authors,

Your paper tries to assess the sensitivity of the Fitch scheme in wake simulations by implementing it on a newly developed PBL scheme. I am certain that this manuscript is interesting and has some value but I am not sure whether in its actual state displays correctly this value, since I think the reader cannot gain much from all the numbers you reach with your simulations. So in the following, you find a list of major and minor/specific comments.

Major comments:

1. My main comment is that the reader does not gain much when reading your work (this sounds harsh but I will try to explain). You compare results from Fitch using two PBL schemes. You do not conclude which PBL scheme is better. You do not conclude which PBL scheme is better with Fitch either. You basically simulate wakes and report the results and basically claim that since the results can be quite different then you suggest to use many different PBL schemes when simulating mesoscale wakes to get an idea of the uncertainty. This is of course valid but then what? One could then say you need to try all wind farm parametrization to get an idea of the uncertainty (and all PBL schemes available). I would think this would be a waste of resources. What I think it would have been nice to see here is some arguments/analysis in which given the differences between the wake simulations one could already say something about the ability of the Fitch scheme to model wakes. Or, better, that the authors have developed a methodology which could use the results of the simulations to gain knowledge about the accuracy of the Fitch scheme. All the quantitative results with regards to the differences under the idealized and the mid-Atlantic case are not really used to anything (I mean the actual numbers) and will not really help anybody to discern anything about the Fitch scheme and/or PBL schemes used. Maybe, one could also simply say: what should be the maximum differences in wind speeds and turbulence deficits when simulating wakes with different PBL schemes when using the same farm parametrization?
2. Your introduction is way too long. Particularly Section 1.2 is not really needed and it falls very much into kind of the same comment I just made about many numbers (from a number of previous wake works) without much meaning. The last paragraph of that subsection can be kept and will be sufficient. Also all the Fitch studies before Archer et al. (2020) are wrong (due to the bug in the model) and so they should not be mentioned. Lastly, what the introduction really lacks is why trying now Fitch with the 3DTKE PBL scheme? Would it be better? More realistic? Totally wrong? I guess the reader would tend to think that a 3D PBL scheme is better than a “2D” one such as MYNN
3. Second paragraph in Sect. 2.1: these lines should be complemented with the formulations so that the reader can get an idea of the advantages/extensions of the 3D TKE PBL.
4. Text between lines 198 and 208: due to the use of a new PBL scheme, it would be interesting to see the development with time of the turbine-free simulations and find out why 3 days are indeed needed to develop the ABL and reach quasi-steady state (it just reads as a quite extremely long spin up period). By the way, you do not mention (I think) the type of boundary conditions during spin up and wake simulations.
5. Idealized simulations: to compare fairly the Fitch scheme with the two PBL schemes, you should aim to get the same wind speed and direction at hub height. Therefore you should not use the same geostrophic wind for all simulations The problem is clear for the unstable simulation where it does not make sense at all to compare the wake results given the large differences in wind speed
6. Conclusions: the manuscript is already quite long and so such a long conclusion (which is not really concluding statements) is not needed. Can the reader get some nearly like bullet-points from your work? Also, and in relation to my comment 1, there are way too many sentences with a number of values that do not mean much if you do not have a reference or measurements. This is very clear between lines 657 and 674

Specific comments:

1. Line 4: “were only compatible with one PBL scheme” this is a general comment but I guess you mean the specific case of the WRF model, which is not mentioned at that point.
2. Line 8: “internal” the reader does not know what you mean by internal so maybe drop the work and be specific in the abstract
3. Line 10: add “atmospheric” before “stability”
4. Line 27 “their impacts in numerical” I guess you mean “their impacts on atmospheric variables when implemented in numerical weather…” or so
5. Line 31: you provide some low and high losses… what are the cases for this? I mean these are because of the size of wind farms?
6. Line 57: maybe delete “generation”
7. Line 64: you have some ? signs when making a particular reference… this is not the first time
8. Line 155: remove “to” before “behave”
9. Line 164: again a ? in a reference
10. Line 165: not sure whether you define “Sq”
11. Line 180: why not using the value suggested in Archer et al. (2020)?
12. Line 198: why not using the roughness of the sea? These parameterizations are mostly used offshore and it will be more straightforward to compare to the mid-Atlantic case
13. Line 199: the sentence kind of suggests that the hub height wind speed is very close to the geostrophic wind speed but that is not necessarily the case. So why 10 m/s?
14. Figure 1. This figure can be changed to show the model domains and maybe an inset with a zoom of the vineyard wind 1 with the turbine arrangement would be nice
15. Line 238 and similar: all these references to manuscripts in preparation are not useful. Particularly the one at this line is not needed (also that in line 252)
16. Sentences in lines 264-265 and 286 are redundant given each of the sentences before them
17. Figure 2: do you say somewhere whether these profiles are instantaneous output at some time? Are they spatial averages over the whole domain? Are they time-averaged? Also the profiles should be somehow smoother; however they show some weird peaks, e.g., the highest wind speed of the stable MYNN or those below the lowest farm boundary in the stable TKE
18. Also about the result in Fig 2 for the unstable TKE: why is MYNN 3 times lower than 3DTKE? You mention this is related to the empirical constants but the stable and neutral ones seem fine
19. Line 300 maybe you can add after “values” whether these are from instantaneous values at a given time
20. Figure 3 and related results: why not aligning the wind with x so that when you make the cross (side) analyses the plots are easy to digest
21. Line 327 “wakes erode” not sure how general is the erode term in wakes, could you replace it by recovery? I think you have different instances with this
22. Line 384 the ref. to your not published work can be changed. Some others have seen this as well
23. Line 432 delete one “of the”
24. Line 480 Not really true as TKE was quite different for the unstable case
25. Line 491: I think you need to add “on land” after “near-neutral stratification”
26. Line 491-494: a figure with the frequency distributions of these surface heat fluxes would be nice
27. Figure 9 how are the winds above the boundary layer? Are they close? Maybe an inset showing the full profile would be nice
28. Line 516 replace “off” by “on”
29. Line 600 “amplification” I am not sure what you actually mean
30. There are way too many avoidable references and also many references to your work (where any of the coauthors are involved)

The paper inserts for the first time the Fitch wind farm parameterization in the newly developed 3DPBL scheme in the WRF model. This is important and innovative because, as far as I know, the Fitch parameterization has only been coupled with the MYNN PBL scheme. It is therefore valuable to see how it would work with a different PBL scheme. However, the paper consists of pages and pages and pages of detailed and rather pointless differences between the results obtained with the two schemes, first with idealized cases, then with a series of simulations of a few offshore wind energy areas in the US Northeast, leaving however no useful information on which would be better for which cases and why. I am afraid that, in order for the paper to be acceptable in its current format, too much additional work would be required (i.e., redo all idealized runs and simulate a different real farm), as discussed next.

An alternative would be to remove Section 4 entirely. Adding a real case would be valuable if it allowed the authors to validate the 3DPBL+Fitch coupling, but it has no value in this manuscript unfortunately, it just adds pages and pages of minutia and repetition.  

Major points

1. Although some of the co-authors have access and/or have participated in field campaigns that have collected plenty of data on wind farm wakes, inland and offshore, and on observed wind farm power production (e.g., Siedersleben et al. 2020 just to mention one), no comparison against any type of observations is offered in this study. Why did the authors choose to simulate the Vineyard Wind and the other U.S. wind energy areas, for which no data are available yet, when so many other farms with data are available? At a very minimum, high-resolution simulations (like HRRR) could have been used for the wind speed profiles for August 2020 for Figure 9. But, better yet, a different farm with actual wake observations should have been simulated instead.
2. The authors state that TKE advection is turned on (see l. 243), but it does not seem to be true. Figure 6 shows without doubt that all the added TKE is confined within the boundaries of the wind farm and above it, but not advected downwind at all. With the MYNN scheme in particular (top two rows), one can even see the individual positions of the turbines, one every other grid cell, with the added TKE at their grid cells and above, but none in the next adjacent cells downwind. This proves that no advection is actually operating. The authors need to double check that bl_mynn_tkeadvect is indeed set to true in the inner domain. Since TKE advection appears to be wrongfully turned off in all the simulations, all the conclusions of the paper are potentially invalid.

Minor points

1. 55: There is another wind farm parameterization for WRF in the literature: the hybrid model by Pan and Archer (2018).
2. 64 and 133 and 164: Missing citation “?”
3. 101: I think I know what you are trying to say, but it needs to be defined better because an external wake cannot be defined as a “distance”. Also, here you suse 0.2 m/s as the threshold, but in the rest of the paper it seems to be 0.5 m/s (e.g., Figure 3 and 11, dashed blue line).
4. Table 1: the same label here is used to indicate three different runs. Please use unique labels for each run, like “S-NWF” for stable, “U-NWF” for unstable etc.
5. 203: not OK to cite a manuscript in preparation, please remove Rosencrans et al.
6. 206: type for “pseudo”
7. 209: How many turbines are there in total? 25 perhaps?
8. 322: Why 0.5 m/s deficit if 0.2 m/s was stated earlier?
9. 322: I cannot understand what the e-folding distance is. Please include an equation. To be honest, I do not even understand why this variable is even introduced, it does not add much, it is overly sensitive to the stability and choice of the scheme, and it is no longer used in the real simulations later. Consider dropping it since it does not add much.
10. Figure 3: I am surprised that the maximum deficit possible is 1 m/s (note that the maximum deficit is 4 m/s in Figure 11). This must be the most efficient ideal wind farm ever designed. Why is the flow from the west-southwest? I would recommend using white for the range -0.25 – 0.25 m/s.
11. 345-350: I find it very difficult to believe that the addition of TKE causes a longer wake. Also very confusing that the weird decrease in TKE in one specific case (Figure 6g) can be used to explain this general and counter-intuitive finding. To me this is another flag that suggests that advection of TKE was not turned on.
12. 384: not OK to cite a manuscript under review. Please remove Bodini et al.
13. 409: the authors themselves note that there is no advection of TKE! This is not a realistic result. Flag bl_mynn_tkeadvect must be true for TKE to be advected, at least with the MYNN scheme.
14. Figure 8: please use one color scheme! You can intervals that are variable to better emphasize features, but using two colorbars like that is not OK.
15. 476: Are these results with 0% TKE or 100% TKE? Why not 25% TKE as recommended?
16. 484: define “centroid”
17. Figure 12: as in #14, not OK to have 4 colorbars.
18. 28: by this point, I could not force myself to read the manuscript anymore. Too boring and pointless. This section on the real cases is rather useless without observations and does not add anything to the discussion of the idealized cases. The paper would be better off without Section 4.