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?