Review of “Horizontal geometry of trade-wind cumuli” by Dorff et al

This paper contrasts 1D and 2D observations of cloud (size) during the NARVAL campaign. In general, I think the paper is great, and should be published quickly. Many of the questions I had while reading it were actually answered a section later. I do have a few suggestions for the authors:

1. Is AMT the best venue for this paper? Sure, it is nominally about comparing different observational techniques, but the results are much more broadly applicable to the cloud physics community. I feel ACP would be the more appropriate journal in the EGU stable. I therefore would suggest moving to a different journal, but otherwise minor revisions
2. In Section 4.1, I would like to see a few more statistics about the compatibility between radar and imager. For instance, what is the % agreement between the two on clear/cloudy pixels (false positive/negative rate, if you will). Is this a function of certain parameters and choices of thresholds? Do the 1D CSDs from both instruments pass a KS test for certain sets of parameters?
3. I noticed some choices of words where I am not sure I would have given those words the same meaning. Some suggestions for alternatives are below whenever I found them – not exhaustive, and maybe not always what you intended to say. It would be good to go through the paper with a non-native reader in mind with a somewhat limited English vocabulary, and when in doubt just use the simplest words possible. Otherwise the paper is written in a very clear language.

Minor comments and word suggestions:

L 14: While-> Since

L15: Do clouds become invisible, or simply gridpoint? The lower end of your CSDs is not much discussed, other than by the scale break. I can see several different mechanisms at play here

L42: Why is that a challenge? If anything, perhaps “2D  imagers are better equipped to address the challenge…” or so

L50 barely -> rarely.

L56 LES has been able to do this for a while, but now also for large domains (>100km)

L73: How collocated are the instruments? How many meters away in spanwise and streamwise direction? I doubt that at least the streamwise direction is going to matter much (after the correction you talk about later) , but good to mention here either way.

L99: How much is this FOV in practice in meters? and what is the typical resolution in meters? I’m not a fan of pixel# as a unit. Perhaps the spatial equivalents of Time and Pixel units can be put on secondary x/y axes in figs2,4,7,9?

L102: pronounce -> result

L105 non-zero reflectivity

L143: Is CTH the correct metric? Since you’re integrating over the entire depth of the cloud, mid-cloud level would be more precise, I guess. Again, shouldn’t matter much in practice for these shallow clouds

L143: “lower and further to the aircraft” not sure what that means exactly

L149: Emphasize 2D connectivity

L174: dammed -> limited

L183: This does introduce the bias that cloud size is artificially limited by the FOV size.

L240: This may be cloud misrepresentation, but it is the fair comparison between the two instruments. This is an important part, because it validates the radar for use in the (extremely common) situation that no imager is available.

L281: I would  be interested in a bit more discussion of the scale break, as it is located much sooner than often reported for shallow Cu (1km+). Are the authors sure that this is not an artifact of the observational strategies/instrument resolution?

L298: Could be interest to compare the overlap corrections from Sulak et al (JGR, 2020).

Excellent paper, great summary and illustration of the limitations of 1D transects / chords lengths for characterizing cloud size distributions. Good follow on from Barron (2020) which they cite multiple times. I have only minor comments, detailed below. I have a suggestion which is not necessary for the authors to perform, but I wonder why they did not show at the end a joint-frequency distribution of cloud size and eccentricity? Should we expect some trend in eccentricity with cloud size? Perhaps eliminate figure 9 whose purpose is unclear, and add such a joint-pdf plot? Or leave that for your next work. There are other possible follow-up studies, which just illustrates why this is a good peice of technical foundation.

Line 25 “limitedly understood” is strange phrasing, consider “only partially understood”

Line 50 “barely” should be “rarely”

Line 72 “precedent” doesn’t make sense here, maybe “The conclusions consider the abilities of a prospective flight campaign to answer new research questions.”

Line 74 “We consider” change to “we analyze”?

Line 78 “… in a 2D image.”

Line 85 “… the profiling radar enables”

Line 96 Why choose the SWIR band instead of VNIR? Just one short sentence to explain. Something about sun-glint perhaps?

Line 127 I wonder if the window freezing issue should be mentioned again at the end in the recommendations.

Line 164 “following aspects” should be “the following aspects”

Line 174 consider “limited” instead of “dammed” which sounds too much like “damned” 

Line 183-185 This is a very important and difficult aspect of this analysis, I have experience with this issue. I believe you made the best choice to reduce bias and error of counting a section of a large cloud as a small cloud. However, this does not eliminate bias, but shifts the bias to larger size clouds (which are now systematically undersampled), so that is ideal for this analysis focused on smaller scales, but maybe a note about this should be included in the discussion in the beginning of section 3.4 to warn future users of this analysis technique. The larger clouds are indeed more rare, but will also be undersampled as the cloud length scale approaches the typical image scale.

Line 214 A “scale-break” might also be a sign that a power law is the wrong choice, because, for example, a scatter-plot of frequency vs length scale data on a log-log axis plot that looks like two power-laws with a scale break in between could instead be considered as a single exponential distribution, with the “scale-break” location being the bend in the exponential on a log-log plot. Since 1 function is less complex than 2 functions, the principle of parsimony would suggest considering an exponential distribution instead of a power-law. I don’t expect you to change this for this paper, or change the power-law obsession everyone seems to have, but I do suggest that you consider the exponential instead of a “scale-break” in future work.

Line 227 Unclear “this prounounces”?

Line 276 “contrarily” is not clear, something more like “This affects the distributions in the opposite direction

Caption of Figure 5 Remove “exemplary” 

Line 315 The point of Figure 9 is unclear… are you trying to show clouds that don’t make it into the analysis at all? Maybe draw some lines on Figure 9 to indicate which clouds in that image are included (if any?)

Line 316 “constraint” should be “constrained” and “This pronounces” doesn’t make sense, maybe “This results”

Line 317 “regardless of the resolution” 

Line 352 Not clear, maybe change to “Using only the radar resolution and statistical methods, e.g., considering circular assumptions (Romps and Vogelmann 2017), or as in Barron et al. (2020), such methods will fail to reproduce the actual double power-laws (not shown). Cloud shapes being rather more elliptical than circular…”

Line 382 “arises” should be “raises”