Crumeyrolle et al. are interested in quantifying the role of aerosol perturbations for cloud microphysics. They aim to do so by a statistical analysis of observational data, combining ground-based observations with passive microwave ground-based remote sensing and cloud retrievals from a geostationary satellite. The method is meant as a proof-of-concept study.

With this aim, the study may be ok for Atmos Chem Phys. The study itself, however, does not provide a substantial new scientific insights. There is a lot of data pre-processing, where some of the applied filters are poorly motivated and seem partly contradictory. Ultimately, a mere 188 data points remain for analysis. This severe lack of data precludes meaningful statistical analysis. The uncertainty quantification consequently shows ranges of almost ±100%.

If the editor decides the topic still fits Atmos Chem Phys, I would have a number of concerns, partly major, partly minor, to be addressed before further consideration.

L14 What exactly is meant by “spatially imprecise”?

L48 This could also refer to a new paper by Jia et al Sci Adv 2026

L92 What are “total..coefficients”?

L129 It would be useful to briefly report the method to retrieve LWC here, and discuss its uncertainties.

L140 This line reads like VIS channels would have 3 km resolution and NIR, 1 km, but that is not true. Only the HRV channel has 1km.

L155 say the “idealised” involves the assumption of adiabaticity

L159 Sentence unclear. The comparison with CALIPSO reveals biases of 30%. But isn’t that a function of the instrument sensitivity? See the GEWEX assessment, Stubenrauch et al. Surveys in Geophysics 2024. The sentence continues with “is less than 10% above the ATOLL station”. What is meant? The bias vs. CALIPSO? Or the liquid fraction itself?

L162 AMSR2 and CLAAS LWP are not directly comparable. Microwave LWP is more like an all-sky retrieval, NIR one, in-cloud. See Seethala and Horváth JGR 2010.

L163 Why jump between r and r²?

L161-164 Please report confidence intervals for the CLAAS retrievals, from the comparison to the reference data, rather than correlation coefficients.

L165 Typically, the statistical error is considered as uncertainty, while if a bias is known, it should be corrected for.

L173 (Table 1) Why not LWP from CLAAS?

L200 This is an awkward definition of stratiform clouds. It should rather be defined by some homogeneity criterion, such as temporally constant LWP and cloud base altitude over some time from the ground-based measurement (or distance from the satellite).

L201 How does a cloud base height threshold filter out rain? There should be a better way to define precipitation from the ground-based measurements.

L214 This is now a second, redundant, definition of stratiform clouds. Why would that criterion be “effective”?

L215 Why do the authors want to minimise convective mixing? Just a few lines earlier (L201) the aim was, on the contrary, to ascertain the aerosol in the boundary layer mixes into the clouds.

L216 Why this lower bound on the BLH? Just a little earlier (L208) the authors explained they want to limit themselves to stable conditions. Here it now seems that on the contrary, convectice situations are sought.

L216 Didn’t the earlier filtering criteria already reduce the data to 18% of its original size (L203)?

L222 I find this criterion puzzling. How is the 60% motivated? This doesn’t seem to me similar at all, if stratiform, and therefore spatially homogeneous clouds, are considered. If the point is to avoid overlying clouds, aren’t there better criteria to filter from the satellite retrievals, e.g. selecting scenes with liquid clouds only?

L223 Now another number for how much data is left is provided, 45%. Why this? What with the 18% (L203) and the 40% (L216)?

L228 This is true for well-mixed boundary layers, but weren’t criteria applied that tried to avoid these?

L232 How are these constrained if no information about CCN is available?

L239 (Table 2) Why this very low number of observations in the initial database? The upper bound would be 4 years * 365 days/year * 12 hours daylight / day * 4 15-min steps / hour = 70.000 time steps. Why only an order of magnitude fewer data points?

L245 My own experience with analysing clouds and sensitivities of microphysical quantities is that one needs a few thousands to obtain robust results. The remaining 188 data points are too few to yield meaningful insights.

L252 The authors should provide some evidence for this statement. For example, they could assess what happens to the conclusions if they further reduce the number of data points. If the results remain unchanged for a reduction by a further quarter or third, that would be convincing.

L263 I agree that Panel (f) is useful, to demonstrate there is very little data from Aug – Oct but otherwise an approximately equal seasonal distribution. What about the distribution between the years? For (a) – (e) it would be more useful to show the overall histrograms, it seems not to make sense to show the seasonal cycle. There is simply not enough data for meaningful conclusions about this.

L271 It is not understandable why the anyway very small dataset is further split into two. What is the motivation?

L276 “higher scattering coefficients are often associated with increased aerosol concentrations” why is that a meaningful statement? Are there situations where the opposite could be true?

L278 “enhanced scattering is indicative of larger particle sizes or higher particle number concentrations” same question

L298 It may be useful to clarify that a hard upper bound for S_n is 1 (every aerosol activates into a droplet), unless mistakes in the definition are made (e.g. 1.59 is physically nonsense).

L300 Schmidt et al. ACP 2015 (doi:10.5194/acp-15-10687-2015) compiled a list of S_n, maybe it is useful to refer to this here.

L302 / Section on r_e: Is this not entirely redundant with the analysis of n_d? i.e., isn’t n_d computed from the retrieved r_e?

L331 A “case study” would be for one case only (but in-depth), here there are 188 cases statistically analysed.

General comment

This study combines ground-based observations of aerosols and satellite-based observations of clouds to evaluate cloud droplet number concentration and effective radius susceptibilities to the aerosol scattering coefficient.  While it is an interesting study, I have two major points of concern:

• The study does not appear to be innovative. In fact, the authors compare their results with similar previous studies, but don’t highlight enough what their analysis offers additionally in the field.
• The authors apply strict criteria in selecting their cases. As a result, however, the analysis suffers from a very small sample that leads to very high susceptibility uncertainties and a very coarse LWP binning. The latter does not produce meaningfully distinctive results (the authors should justify if this is expected), while both these shortcomings prevent them from drawing robust conclusions.

Given these points, I would recommend that this study is revised substantially, especially in the “results and discussion” part, before being reconsidered. Specifically, a larger dataset (e.g. from other ACTRIS sites), after applying similarly strict selection criteria, could increase the robustness of the results and offer more insights on the studied processes. The authors already recognize this potential in their conclusions. Given the limitations of this study, and that the increase in the dataset size is a matter of scaling, not of expanding the analysis, this should be done in the present study, and not just discussed as a future plan in the concluding remarks.

Major Comments

Line 22: The authors use 2 bins of LWP values to study the susceptibility of N_d and r_e to the aerosol scattering coefficient. The values they find are very similar for the two bins. Then what is the value of this distinction?

Introduction: The authors highlight the disadvantages of satellite-based retrievals being used to estimate aerosol effects on clouds, and their main advantage being global coverage (and temporal frequency for geostationary platforms). They also describe the advantages of local, detailed measurements, with their main disadvantage being locality. The approach they propose still requires station measurements (it’s not satellite-based only), so the spatial limitation remains, even if expanded to other stations. How is then this study taking advantage of “the best of both worlds”? It appears that satellites are used here only to supplement the dataset with cloud properties missing from the ground-based measurements.

Line 56: While combining station-based and satellite measurements to study aerosol effects on clouds is indeed “an alternative path” compared to the methods described before, it also has a long history. What is innovative about this study?

Lines 180-181: It is not clear how the space- and ground-based data were collocated spatially (nearest neighbor? something else?) and temporally. It takes ~12 minutes for SEVIRI to scan the Earth’s disk from south to north. This means that the scanning time at the latitude of the ATOLL observatory occurs ~9 minutes after the nominal CLAAS-3 acquisition time. Did the authors correct for this time difference when collocating the data temporally?

Section 4.1: The authors compare their findings with those from previous studies. They should also include here possible explanations of the differences found, and highlight the value that this study adds to the field.

Section 4.2: Similarly to my previous comment, part of the discussion regarding explanation of differences when comparing with other studies is missing.

Line 326 (see also my comment on line 22): The authors attribute the only 2 LWP bins that they use to the small size of their sample. In fact, it is not clear what the LWP binning offers here: results are very similar in both bins, for both n_d and r_e.

Minor comments

Line 79: it is mentioned that the instruments are positioned “at least one meter above the rooftop”, but what is their height relative to the ground?

Line 104: Is there a reference for the neural network mentioned here?

Line 110: “and 0.4 g m^-3 RMS.” Consider adding “for liquid water content” for clarity. Please also add a reference for these numbers.

Line 155: “… is also retrieved from the cloud optical depth, r_e considering …”. I assume the authors mean “… is also retrieved from the cloud optical depth and r_e considering …” ?

Table 1 contains useful information but is not mentioned anywhere in the text.

Lines 197-198 should read “to determine the height of the boundary layer (e.g. Sinclair et al., 2022)”.

Line 328: “A similar tendency appears to be present here”. I don’t think this claim can be made here, even if the authors are cautious to state that “the large uncertainties prevent this behaviour from being statistically confirmed”. Please consider removing it.

Line 331 should read “… how cloud droplet concentration and effective radius susceptibilities …”