A nice discussion / evaluation of wind park requirements, and an assessment of different microwave ground based observations, I have only some minor comments, shown below with Page Line:

• Figure 1: What is the difference between the top 2 plots (besides that they compare to different model data)? Are these different locations, thus the measurements are not the same? And might be good to include some general statistics in all the plots, as text (as done later in other plots, e.g. correlation). 
• Figure 2: the manuscript has top 1 plot, bottom 2, not left and right.
• P9L216: The presented results show that global NWP / re-analysis models can be used, but that is unlikely to be valid globally. I assume ERA5 is particularly good in areas where a lot of measurements are available. 
• Section 2.2: Is ECMWF actually assimilating these tower measurements? If yes, then they are not independent when doing validations.
• P9L226: Are these 2 models providing data exactly on the vertical levels of the tower? Or better/worse? Any averaging required to align the resolutions?
• Figure 5: Might be more instructive to use a range that is within the atmospheric conditions (I assume, -100K change over 1km is not really a valid range), and print the lengthy legend on the side of the plot?
  
  Also, are you doing some outlier detection and removal? Sometimes the model seems to be "stuck" at about -20K/km, while the tower detects values up to 150K/km. 
• Figure 6: (a) I assume the figures should somewhere have the a, b, c, d label?; (b) the map plot should have lat/lon labels; (c) the lower left plots have no y label units.
• Figure 7: Again, the a, b, c, d labels are missing, making it difficult to understand this already complex plot compilation. 
• P22L516: Just interpolation to the radiosonde data? The resolution of MW is much lower, thus you could also average the sonde data (or fold it with an averaging kernel of the MW). 
• Figure 9: Is there a possibility to split this into stable/unstable atmospheric conditions? 
• P23L540: Resolve a.g.l. somewhere. 
• P24L552: Sorry, I might have missed it above, but how are MW data sets averaged in time to the sonde times? And they provide different scanning modes, as indicated above. Does that influence the averaging? And are you considering the ascent times, or just an average time of the sonde? 
• Figure 11: Is there a correlation difference between stable and unstable conditions?
• P25L588: Any comment regarding the different seasons being used? And with 4 sondes, you'll also capture different diurnal variations, compared to the 2 sondes at GRA. 
• Figure 12: Maybe be consistent in using titles of the plots. Regarding panel a, the agreement is remarkable between 300m to 700m. Is there any use of sonde data in the MW retrieval? 
• Figure 13: Maybe also include titles? No, after further reading, please use titles, so that the campaign/site is easy identifiable. And regarding presentation, maybe make a blank plot c, write no data into it?  
• P29L622: While reading this: are these MW retrievals all consistent? Or each uses its own retrieval setup? If the setups are different, what is the impact on the assessed performance and uncertainty? 
• P30L647: I think this is the first time, a quality control is mentioned. What does it consist of, and is it used for other data sets too (see also comment above)? 
• P30L663: How robust is this improve correlation? There is one data point at about 30K/km for scanning. Do you still get good correlation if that data point is removed?  
• P30L668: Okay, here is finally some remark on the different retrieval methods. As these seem to vary quite a bit, I think they should be shortly mentioned, maybe in the section where the instruments are introduced. 
• P31L693: Not for this work, but it would be a very good follow on work, to assess these different instruments / locations / seasons with the same retrieval algorithm. -> saw you identified the need in the conclusion. Good.
• Table 3: Here AGL is used. Is that introduced somewhere? 
• P32L703: While reading these different gradients - is there actually a limit what gradient can be detected by which instrument, as they do differ in observation capability, e.g. such as the number of channels, polarisation, etc? 

Review for AMT-2024-186

Title: “Atmospheric stability from microwave radiometer observations for on/offshore wind energy applications”

Authors: Domenico Cimini, Rémi Gandoin, Stephanie Fiedler, Claudia Acquistapace, Andrea Balotti, Sabrina Gentile, Edoardo Geraldi, Christine Knist, Pauline Martinet, Saverio T. Nilo, Giandomenico Pace, Bernhard Pospichal, Filomena Romano

General comments:

The study presented in this manuscript first reviews stability metrics derived by NWP models and reanalysis, useful for wind energy. Second, it quantifies the performances of different microwave radiometers at estimating stability metrics on land and offshore. The manuscript is relevant, well written and easy to read. I think it fits well in the scope of AMT. I suggest only some minor modifications and to include some additional information. For instance, some more info on retrievals’ and calibration’s procedures, a-priori or radiosonde dataset used in the retrievals, radiosonde type, should be provided.

The title of the manuscript does not mention the first part of the study. The authors might want to consider if to mention this as well.

Specific comments:

Page 1, Abstract, line 32: Include units for RMS.

Page 5, line 138: “Measurement data came from the FINO1, FINO2 and FINO3 met masts”. How are these data used in the before mentioned NWP models? Are they assimilated?

Page 6, Figure 1 caption: Describe the pink line in the caption of the figure. Also, in figures with multiple panels, a), b) c)… would be useful.

Page 7, line 194: How id the boundary-layer height determined during stable vs unstable conditions?

Page 8, Figure 2 caption: Panels are ‘top’ and ‘bottom’, not ‘left’ and ‘right’. Also, in the caption you say ‘(8 and 12 m/s, 207 respectively)’, but in the title of the panels you say ‘4 +/- 0.5’ and 8 +/- 0.5’.

Page 11, Figure 4 caption: This is another figure with multiple panels, where a), b) c)… would be useful.

Page 11, Figure 5: Do the models have hard limits on e lower values of dT/dz?

Page 13, Figure 6: Please include a), b) c)… to the panels. Also, include x- and y-labels to the upper left panel, and y-labels to the lower two left panels.

Page 14, Figure 7: Please include a), b) c)… to the panels. This figure really needs them (same for all the figures in the Supplemental material). For the figures in the supplemental material, it is interesting to see the jumpy behavior of the zi time series, particularly for the FINO1 site. This is why it might be interesting to know how are these zi values obtained.

Page 16, Table 1: I think it’d be useful to list somewhere the retrieval techniques utilized for each of the MWRs? Also, what a-priori were used?

Page 17, after line 368: Could you describe what type/schedule of maintenance, calibration were performed on these MWRs?

Page 19, line 435: How many daily radiosondes per day and at what time?

Page 20, line 473: What type of MWRs?

Page 23, Section 4: Would it make more sense to add the info on Radiosonde/MWR matchups, radiosonde launch times, radiosonde type, and so on in a table, rater than in the text for each dataset? In this way, you could also remove some of these details from the individual dataset descriptions in Section 3.2.

Page 30, line 655: ‘mostly driven by only one point’. I agree.

Page 32, lines 710-711: No mention to the different method used for the retrievals (and the a-priori used for these) is given up to this point, nor about calibration performed, but I think it would be an useful information to add to the Section with the different dataset descriptions. Good calibrations and adequate a-priori datasets are crucial for MWRs.