In their manuscript, Benas et al. use a long record of geostationary satellite measurements to investigate the effect of aerosol pollution on low clouds within an isolated shipping corridor in the southeastern Atlantic. Their methods are similar to those employed by my group previously in terms of comparing observations (presumably with shipping effects) with a counterfactual based on cloud properties outside the shipping corridor, and our results are similar in terms of their broad strokes, although there are notable and intriguing differences. Their method is cruder in terms of estimating the counterfactual using a cubic function fit on a reduced-dimension profile centered on the shipping corridor, but more sophisticated in terms of its spatial and temporal resolution. My major concern about the paper is the lack of uncertainty quantification for the counterfactual; although the authors do an admirable job quantifying the uncertainty of the observations, I would argue that we should expect the uncertainty due to estimating the “no ship” counterfactual to be even larger, and more difficult to constrain. Otherwise, my comments are relatively minor. I look forward to seeing an adequately revised manuscript published in ACP. -Michael Diamond

Major comments

A. Quantifying uncertainty of the counterfactual: The fundamental challenge of quantifying aerosol-cloud interactions in observations is we can’t just re-run reality while excluding the aerosol, like we can in a model. The SE Atlantic is such a nice “natural experiment” because the constrained nature of the ship pollution offers us the tantalizing prospect of really being to compare clouds under the same large-scale meteorology differing only with an exogenous aerosol perturbation. As nice as the setup is, however, estimating the “clean” (or at least, non-shipping) cloud field is non-trivial. The cubic fit is a reasonable choice, but I know from experience (albeit with coarser data) that various reasonable-seeming fitting strategies can result in very different answers in terms of the liquid water path (W) and cloud fraction (fc) results. The authors seem to have seen the effect of small variations in methodology as well, in their discussion of shifting the assumption that the non-shipping background starts 150 km instead of 250 km in changing their fc results. I would encourage the authors to do a similar exercise with W; my sense is that moving to 150 km or even 200 km will dramatically shrink the estimated effect magnitude (but not the sign). One suggestion I have in trying to quantify some of the uncertainty in the counterfactual method would be to run the analysis with counterfactual curves fit at 150, 200, 250 (current), and 300 km distances and take the error as the standard deviation from the different fits.

B. Missing discussion of previous shipping corridor work in introduction: The authors reference Diamond et al. (2020) in their methods and multiple times in comparing results, but do not engage much with the other literature attempting to glean information from shipping corridors instead of individual tracks, including some relevant papers focused on the SE Atlantic as well. Somewhere in the introduction, probably just before current line 69, I would recommend adding a section about the difference between studying individual ship tracks “bottom-up” and shipping corridors “top-down”. You should also mention here why the SE Atlantic region is chosen — and why other corridors, such as those investigated by Karsten Peters and colleagues, did not prove conducive to investigating the shipping effect on clouds — and give a summary of what is already known about the corridor.

Suggested references:

Hu, S., Zhu, Y., Rosenfeld, D., Mao, F., Lu, X., Pan, Z., Zang, L., and Gong, W.: The Dependence of Ship‐Polluted Marine Cloud Properties and Radiative Forcing on Background Drop Concentrations, Journal of Geophysical Research: Atmospheres, 126, e2020JD033852, 10.1029/2020jd033852, 2021.

Peters, K., Quaas, J., and Graßl, H.: A search for large-scale effects of ship emissions on clouds and radiation in satellite data, Journal of Geophysical Research: Atmospheres, 116, D24205, 10.1029/2011jd016531, 2011.

Peters, K., Quaas, J., Stier, P., and Graßl, H.: Processes limiting the emergence of detectable aerosol indirect effects on tropical warm clouds in global aerosol-climate model and satellite data, Tellus B: Chemical and Physical Meteorology, 66, 24054, 10.3402/tellusb.v66.24054, 2014.

Specific comments:

1. Line 20: Given the uniqueness of the SE Atlantic setup, you might want to soften the statement of generalizability of “for similar analyses” to something more like “studying aerosol-cloud interactions”, etc.
2. Line 45: I’m not sure what “typical” means here? I would say they are particularly good examples!
3. Line 183: Related to major comment A above, I would recommend the authors check out Tippett et al. (2024) and update their discussion of Manshausen et al. (2022) accordingly. This reflects the importance and difficulty of estimating a proper counterfactual! I’d also note that the global studies of Wall et al. (2022, 2023) do show globally negative LWP susceptibilities to aerosol.
   
   
   
   Tippett, A., Gryspeerdt, E., Manshausen, P., Stier, P., and Smith, T. W. P.: Weak liquid water path response in ship tracks, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2024-1479, 2024.
   
   
   
   Wall, C. J., Norris, J. R., Possner, A., McCoy, D. T., McCoy, I. L., and Lutsko, N. J.: Assessing effective radiative forcing from aerosol-cloud interactions over the global ocean, Proc Natl Acad Sci U S A, 119, e2210481119, 10.1073/pnas.2210481119, 2022.
   
   
   
   Wall, C. J., Storelvmo, T., and Possner, A.: Global observations of aerosol indirect effects from marine liquid clouds, Atmos. Chem. Phys., 23, 13125-13141, 10.5194/acp-23-13125-2023, 2023.
   
   
   
   
4. Lines 193-195: Another possibility is simply error in the method! Even if the true effect were zero, we still wouldn’t expect to get a result of precisely zero unless the method was absolutely perfect.
5. Line 196: You could also do a quick test to see if you should expect a noticeable perturbation in cloud optical thickness (COT) given your inferred changes in Nd and W. Just from eyeballing Figures 2-3, dln(COT) ~ 1/3 dln(Nd) + 5/6 dln(W) = 1/3(4%) + 5/6(-3%) = -1%. From Fig. S2, I would expect a decrease in COT of ~0.08 to be apparent. The difficulty in obtaining a COT result that fits with your other values could also be a reflection of potential methodological limitations.
6. Line 196: It might be worth trying to analyze log(COT) instead of COT
7. Lines 199-200: I don’t see why this should be true. Diurnally or seasonally opposing positive and negative effects would average to zero overall but would be discernible with your method.
8. Line 213: Is this comparison referring to Fig. 9 in Grosvenor et al. (2018)? I don’t believe Grosvenor & Wood (2014) provide a seasonal breakdown of the subtropics.
9. Lines 272-273: However, it should also be noted that geostationary retrievals suffer from diurnally varying biases related to scattering geometry that could be relevant here.
   
   
   
   Smalley, K. M., and Lebsock, M. D.: Corrections for Geostationary Cloud Liquid Water Path Using Microwave Imagery, Journal of Atmospheric and Oceanic Technology, 40, 1049-1061, 10.1175/jtech-d-23-0030.1, 2023.
   
   
   
   
10. Line 276: It’s worth noting that negative cloud adjustments at night and positive during the day would be the opposite of what we’d expect from the diurnal cycle of precipitation (maximizing at night) and evaporation (maximizing during the day). See, e.g., Sandu et al. (2008) Figure 7.
    
    
    
    Sandu, I., Brenguier, J.-L., Geoffroy, O., Thouron, O., and Masson, V.: Aerosol Impacts on the Diurnal Cycle of Marine Stratocumulus, Journal of the Atmospheric Sciences, 65, 2705-2718, 10.1175/2008jas2451.1, 2008.
    
    
    
    
11. Lines 281-282: I would not feel safe concluding this…
12. Lines 313-314: Similar conclusions about detectability without using a technique like ML-assisted ship track detection or statistically-generated counterfactual fields were reached by Watson-Parris et al. (2022) and Diamond (2023).
13. Data availability: I’d encourage the authors to consider making a repository with some processed data needed to reproduce the key figures as well.

The study uses geostationary satellite observations over the SE Atlantic to examine the influence of ship emissions on cloud properties. A shipping corridor is defined based on the density of shipping traffic, and cloud properties are sampled both inside the corridor and outside of it, at a safe distance from the influence of the busy shipping corridor. Clouds outside the shipping corridor are considered unperturbed, allowing the influence of ship emissions to be quantified with respect to the perturbed cloud inside the corridor using a cubic fit. The use of geostationary satellite data enables analysis across various time scales, from diurnal variations to seasonal and long-term changes. This complements previous research by Diamond et al. (2020), which used polar satellite data, providing an advancement in this area of study.

Major comments

A description of the criteria used to filter out clouds that are not liquid low-level clouds is missing. While marine clouds are the most common in the SE Atlantic region, other cloud types are also prevalent. Additionally, it is recommended to exclude cloudy pixels with uncertain retrievals based on lower thresholds of re​ and τ values (Sourdeval et al., 2016). Were such filters applied here? Given that marine clouds can also exist as broken cloud fields, combined with the coarser spatial resolution of SEVIRI, retrieval uncertainty is likely to be even higher than that of MODIS, for which such filters are frequently applied. Including biased retrievals can affect the averaged values, especially for Nd​, due to its high sensitivity to re​ (Grosvenor et al., 2018). This can introduce a bias in Nd​ that non-linearly depends on the observed re​, which varies between the corridor and the reference regions.

Sourdeval, O., C.‐Labonnote, L., Baran, A. J., Mülmenstädt, J., and Brogniez, G.: A methodology for simultaneous retrieval of ice and liquid water cloud properties. Part 2: Near‐global retrievals and evaluation against A‐Train products, Q. J. Roy. Meteorol. Soc., 142, 3063–3081, https://doi.org/10.1002/qj.2889, 2016.

The authors chose not to include an analysis of τ because they found no response between the corridor and the reference region, likely due to the cancellation effect between the decrease in W and the increase in Nd​. Given that τ is closely related to cloud albedo, does this imply there is no radiative effect from the shipping corridor? Including a radiative perspective could enhance the study's impact by providing further insights into the potential climate effects.

Specific comments

Throughout the manuscript, the authors propose hypotheses for their findings that at times seem to be drawn too quickly or are overly speculative, such as in lines 220-223, 270–272 and 279-281. The authors should be more cautious when discussing findings that the current study was not specifically designed to address.

The differences found in the cloud properties are small, yet the axis range set in the plots make them appear more significant. This could be somewhat misleading. Why not present the results also as relative changes?

Line 172: re should be r_e.

Line 252-256: Can you provide a reference for why cloud thinning would lead to a smaller re​? It depends on homogeneous versus inhomogeneous mixing.

Line 256: This sentence in not clear. Nd can be calculated from Terra, why do you say “since no Terra Nd product is available”?

Line 274: Cloud fraction depends on the thresholds used to distinguish between cloudy and clear pixels. Perhaps the different threshold at nigh and day is related to the corridor effect changing between night and day?

Line 311: I don’t see a statistical significance test (S5 that you refer to shows maps of uncertainties). For the trend analysis related to IMO regulations, performing a significance test between the two time periods would be useful for quantifying the differences.

Line 313: What is CFC?

Lines 360-363: This should be included earlier, in the methodology section.

The manuscript documents a SEVIRI based analysis of the potential effects of ship emissions on low-cloud properties. The uniqueness of this analysis is that the authors make use of geostationary retrievals. However, it is hard to discern what is new, especially from the abstract, as the findings are comparable to those described in Diamond et al. (2020). While some differences between Benas et al and other studies are attributed to the use of Aqua MODIS and Terra MODIS and plausible calibration issues (which is far from being certain), the main findings of Benas et al. are a corroboration of Diamond et al and other studies. The novel result of this study is the analysis of the diurnal cycle, but the ship track signal is weak, except for cloud fraction. In addition, the trend analysis for the entire SEVIRI record is not conclusive, and the following sentence in the abstract is not supported by the evidence “The long-term analysis reveals a weakening of the shipping corridor effect on Nd and re presumably following the International Maritime Organization's 2020 stricter regulations on sulfur emissions…” I appreciate the effort but, again, the authors need to more clearly state what is new of their study.

Other comments

• Difference between SEVIRI and MODIS: While I agree with the authors about issues with some channels for Terra MODIS, I am not aware of any quality degradation of the retrievals for Terra MODIS relative to Aqua (line 270-271 should be rephrased because the results in Benas et al. do not indicate significant uncertainties between the 2 MODIS instruments). Moreover, sensor issues can also be invoked for SEVIRI, as more than 1 SEVIRI instrument provided the data for this study. While I would agree with the authors that SEVIRI is a stable sensor, I am more interested in learning about pixel resolution differences between MODIS and SEVIRI and how they would affect the findings. More specifically, what would be the impact of the 3-km (nadir) pixel resolution of SEVIRI versus MODIS (1 km). This could be the single most important difference between this study and Yuan et al./Diamond et al. A coarser resolution would certainly impact the cloud mask identification, and cloud retrievals (I would expect larger droplet effective radii and smaller optical depth as the pixel resolution is degraded). Is the 3-km pixel resolution sufficient for detecting ship tracks? Possibly, the pixel degradation with viewing zenith angle is minimized for the study region (which is good); however, there is no quantitative description of a solar zenith angle threshold.

• Statistical analysis: Considering the tiny changes in microphysical properties, a robust statistical method for testing the hypothesis is essential and should be highlighted throughout the article. I am also somewhat concerned about the methodology to construct Fig 3c-f. Particularly for cloud fraction (f_c, Fig 3f), I don’t understand why the pattern is undulating. This makes me speculate that the methodology is not ideal for identifying the effect of ship tracks because cloud spatial variability is likely dominating the signal. The same comment applies to optical depth (tau) in the supplement; the tau pattern does not make sense. If tau were unperturbed by ship emission, deltatau as a function of the corridor distance should be a flat curve, right?

• Last paragraph of page 7 should be discussed in section 2.3

• Equation 1: I am not sure why the combination of 2 dissimilar variables (standard deviation and retrieval uncertainty) can be used to produce the uncertainty of averaged data. At least, it does not seem to be mathematically correct.

• Line 171 You mean “determine” instead of “simulate”

• Page 9 “This result suggests that two opposite tendencies, namely an increase in τ due to the Twomey effect and a decrease due to the decreasing W cancel each other out.” This is an overinterpretation of the satellite data because from a remote sensing perspective droplet effective radius is the key variable, not LWP (LWP is indirectly estimated from effective radius and optical depth).

• Line 238: “it is difficult to draw any conclusion on the shipping corridor effect on fc,

day.” Is there a sentence missing here?

Introduction: Lines 22-43. Future readers already know this. For brevity, the authors should