Ground-based microwave radiometers (MWRs) are used by many to provide measurements of the temperature profile and its evolution in the boundary layer.  This paper builds upon work by other authors, exploring some importat aspects that could lead to errors in the retrieved profiles.  In particular, explore the role of horizontal inhomogeneities near the instrument, tilt of the radiometer, physical obstructions, and radio frequency interference (RFI).

My main question to the authors is simply: how does this paper add to the already extensive number of papers that address many of these topics?  They reference the Meunier et al 2013 paper (but don't include any of those findings in their discussion here -- indeed, they ignore the possible uncertainties associated with beamwidth totally).  There were also good papers by Han and Westwater (2000) and Liljegren (2000) that should have been referenced, and from which this paper should build.  Regardless, the authors need to clearly state the new knowledge this paper is contributing to the field.

I also found that this paper has a very informal tone about it; i.e., the language is a bit "loose". Many of the statements are repeated multiple times and could be better organized.  There were multiple radiosonde (and retrieval coefficient) datasets used and at least two different MWRs; I suspect this does not impact the overall results at all, but it does add confusion.

Also, the focus of the paper (from the title and abstract) was on the impact to the retrieved temperature profiles, and thus the inclusion of the discussion on the K band channels is distracting from the main message.  And in many ways, it seemed that the K-band results were included as an afterthought, and not well organized.  I would recommend that either the title changed and the K-band results be separated into their own (sub)sections, or that the K-band results be removed.j

I think that a key point that is being made here, but not explicitly stated, is the importance of always collecting elevation scans on both sides of the MWR.  This allows the analyst to determine (from a sufficiently long dataset) the possible tilt of the instrument and the frequency of horizontal inhomogeneities.  I think this point should be stated strongly, as I've observed many groups who believe it is sufficient to only collect elevation angles along one side of the radiometer.

The retrievals performed in this paper were done using a statistical method.  Would have the results changed if a more accurate physical retrieval method was used instead?  Line 79 indicates that the method of retrieval could matter (it might be useful to reference the Maahn et al. BAMS 2020 paper here for context -- Loehnert is a coauthor of that paper).

Regarding obstructions: one of the more common setups that could affect these observations are power lines that are in the field of view of the radiometer.  These lines clearly don't fill the entire field-of-view.  Would the authors be able to provide any guidance on how far power lines would need to be away from the radiometer as to not impact the V-band observations?

Equations 2 and 3: why is the retrieved temperature a function of both frequency and height?  I think you can remove the nu from the left side of both of those equations.

Line 230: Aren't rapid changes in zenith radiance obserations also a measure of the horizontal inhomogeneity?  Is there a way to look at the variability of the zenith radiance observations over time (and the trend of the magnitude of these radiance obs over time) to estimate the possible level of horizontal inhomogeneity?

Lines 315-320: (if you decide to keep the K-band material and reorganize it): the magnitude of the impact on the K-band as described in the text does not seem to match with the magnitudes shown in Fig 4.

You make the point a few times in different places that the impact of obstacles depends on the vertical structure of the temperature profile; namely, that there is a smaller impact when there is an inversion.  This really should be discussed a bit more to explain why this is.

This paper would be markedly stronger if a plot showing the changes in Tb and the associated changes in total optical depth for small elevation differences was shown.  I've included such a figure here for a mid-day, midcontinental radiosonde with IWV=43 kg/m2, for small elevation angle changed around 5.4 deg.  In particular, note that the change in the optical depth is constant, in a fraction changed sense, as the elevation angle changes.  This also demonstrates why there is little impmact for channels 12, 13, and 14 -- the optical depth is already very large.

Line 440: Clear sky scenes also have very little temporal variation in the observed Tb values.  So not only should the mean LWP be small, but the standard deviation of the LWP should be small also.

LIne 452: Earlier, you made the point that obstacles needed to been many hundred of meters away; yet this lightning rod is only 5 m away and has no impact.  I suspect this is because the beamwidth, together with the 10-deg azimuth sampling, resulted in this rod not being in the V-band's field-of-view.  Is that correct?  Regardless, the way this section is worded conflicts with what you wrote earlier in the paper about obstructions.

Line 475: here is an example of the loose writing.  You've already stated that off-zenith observatios from channels 8 and 9 are not used in the temperature retrievals, so RFI disturbances in these channels at off-zenith angles should ZERO impact on the temperature retrievals (not negligible)

Line 485: I really don't understand how internal uncertainties / misalignments would cause these results.  This needs to be explained, or (as is my preference) the K-band results removed.

Line 500: this is a great place to emphasize the need to have matching elevation scans on both sides of the radiometer

Line 511: Are the errors in retrieved temperature profile small because the errors in the observed Tbs are somewhat offsetting because you are using data from both sides of the radiometer?  I think yes.