General comments:

The work „Molecular composition of clouds: a comparison between samples collected at tropical Réunion Island, France) and mid-north (puy de Dôme, France) latitudes.“ by Lucas Pailler et al,  represents an important and innovative study about organic compounds in cloud water samples. As the authors pointed out, most studies about cloud water chemical composition in literature are limited to targeted approaches, only covering a relatively small fraction of the organic matter. The authors used ESI-FTICR-MS for a comprehensive non-targeted analysis of DOM in cloud water samples from two different locations.

Many aspects regarding the technical quality of the experimental work and data analysis are excellent. The sample collection and handling process was carefully conducted in respect of the challenging nature and sensibility of environmental samples, which is crucial for a conclusive analysis. Regarding the FTICR-MS measurements, published and proven concepts for data filtering, assignments of elemental composition, graphical representation and interpretation were applied. However, there are several aspects that must be discussed or clarified (see also detailed comments below). I think it is very unfortunate that two different Instruments were used for the analysis of the samples (REU: 12 Tesla FTICR in Rouen; PUY: 9.4 Tesla FTICR at LCM lab). Most of the manuscript outline, scientific question, and discussion is centered on differences between REU and PUY samples. Therefore, the lack of a clear proof (e.g. control samples analyzed on both systems), that measurements between both systems are comparable, is somewhat problematic for such a complex analytical technique. Many experimental parameters can significantly influence obtained mass spectra (e.g. relative abundances of peaks), which could potentially hinder a direct comparison. This uncertainty is amplified by the lack of analytical replicates or an alternative demonstration of method robustness, which would help to understand the significance of the results.

The overall technical quality of presentation is on a high level and the precise language, style, and literature referencing is very appreciated. A common theme in this work is the classification of the samples based on the sampling location (PUY and REU) to make general conclusions about the difference in their chemical nature. However, the sample set seems not appropriate for this kind of discussion: The three samples for REU were taken within a week, showed comparable number of assigned MFs (2276-3098) and have a very similar organic fingerprint (evidenced by figures 1 b/c, 2 a/b, 3b and figure S6). In contrast, the group of PUY samples were taken over a period of 2.5 years, vary drastically in the number of assigned MFs (120-7436 !) and show very different molecular composition in the OSC vs #C plots (figure 2 c and S4). Therefore, the grouping and comparison of average values calculated over PUY vs REU to differentiate and draw conclusions on the sampling location, does often not seem meaningful (clearly visible in figure 2d, with large boxes for PUY samples due to the heterogeneity of the samples compared to very focused boxes for REU). Instead a more detailed discussion of mass spectral features in individual samples would be interesting, since e.g. the six PUY samples seem to show very different mass spectra.

Overall, this work has great potential to improve our understanding of DOM in cloud water and many aspects in this manuscript are excellent. However, some details in the experimental setup (FTICR-MS analysis at different instruments, small dataset, no replicates or control samples) hold back the significance of obtained results. This study would furthermore benefit from changing the focus of the and intesnify the discussion of observations in individual samples instead of mostly a PUY vs REU comparison, since some of the pronounced features in the data are currently ignored.

specific comments:

P1-L40: The assumption that the difference in cloud water chemical composition between REU and PUY samples is “mainly linked” to chemical processing is not sufficiently supported by the results in my opinion.

P4-L145: While the explanation for the number of assignments (higher R => more peaks, ion suppression => less peaks, + different chemical composition) might be theoretically correct it is not convincing for the presented data. The actual data does not really follow any trend. Specifically: 12T instrument: quite consistent for the three samples ranging between 2000-3000 MFs. 9.4T instrument: 120 (!) – 7000 (!) assigned peaks. Also, a comparison to DOC values (vs. ESI-FTICR-MS response) would be appreciated, as DOC is relatively similar for all samples, while the MS response seem to vary in more than one magnitude of order. A better explanation why the ion suppression effects are influencing specifically the three mentioned samples would also be interesting. And: Is it actually reasonable to include them in the analysis/comparison if their MS quality might not be sufficient?

P4-L49: The terms “relative number of occurrence” or “relative weighted occurrence” is not used in the cited publication Hawkes et al 2020. I can guess that the data was normalized and/or weighted to the total ion current to give the relative abundance instead of using absolute ion counts?

In any case, there is some confusion regarding the normalization & discussion of the data since the authors also use absolute counts for argumentation several times within this manuscript (which actually might not be viable for this specific analysis).

P4-L191: I am wondering how “not-found” MFs were treated. I guess the matrix of 9251 MFs represents all MFs, which were found in the whole dataset and for samples a given MF was not detected, a value of “0” was defined? I could imagine the statistical analysis can be very distorted for samples which have a low number of found MFs. E.g. for sample 22/10/2019 there are 120 variables with an actual numerical value between 0-1, potentially even dominated by very few signals, while more than 9000 (equally treated) scores are “0”. Also, especially for the statistical analysis I am suspecting a large influence of using different MS-instruments.

P9-L295: The use of ANOVA in this dataset for a strict PUY vs REU differentiation seems forced and not meaningful. According to figure 2a: two PUY samples (02/03/2019 and 15/03/2019) have a lower average OSC than REU samples, two PUY samples (02/10/2019, 17/07/2020) have a higher OSC and the rest of the PUY samples have comparable OSC values (as also mentioned later by the authors). Have you tried to check the significance of other groupings (seasons, air mass history)?

Actually, the whole significance calculation for the OSC might not be meaningful in the way it seems to be conducted: the uncertainty of the OSC appear to be calculated by taking the variation over all individual MFs within a sample (similar to how the boxplot 2a presents the data). That’s why they are so high (e.g. -0.97±0.56 for winter). Basically, this value describes the range of how different the OSC of individual analytes in the sample is, instead of how different the average OSC value of samples within a group is. Maybe some clarification on the significance calculation would be helpful to better understand the approach.

P9-L322: The interpretation of Figure 2b/c is missing some important observations. E.g. the fact that PUY samples are often dominated by few very abundant signals (e.g. for 02/10/2019: very large bubbles at 15-20 carbon atoms and 5-10 carbon atoms). Also visible for figure S4 with huge bubbles for 02/03/2019 which are not discussed and are significantly different from the other “summer” sample 15/03/2019.

Also, the scaling in figure 2 (and S4) is hindering a proper interpretation. Bubble sizes are proportional to the absolute intensity (while in other instances in the manuscript relative intensities are used). Its not possible to get an idea of the chemical space for samples 03/11/2020 and 22/10/2019. This is relevant to better compare the other figures and statistical analysis (which are normalized to relative intensities).

Bubbles for 2b should also be transparent. In its current form its not possible to see and evaluate the chemical composition of R8 and R9. But it seems MFs are relatively homogenously distributed over the chemical space for all three samples, in contrast to PUY samples.

P9-L325: I agree that one of the most significant differences between PUY and REU is the lower right region from OSC -1.5 to -4 and 0 to 10 carbon atoms. However, I don’t understand the conclusion “compounds in this region can be recalcitrant to oxidation in cloud water” (means hard to oxidize? Is there a reference for this statement?) and “These results clearly assess that DOM in REU samples is less oxidized than DOM in PUY samples”.  Are these two statements directly connected? Or is the second statement a summary of the whole paragraph? In any case, a more detailed interpretation of the region in the lower right would be important, since it seems very significant for PUY samples.

P12-L415: The Interpretation for the lack of organosulfates (L415-424) is very hard to follow and appears to be mostly speculation. Most of the paragraph is not connected to observations from this study. For the reader its hard to follow why literature values for isoprene (Dominutti et al., 2022 and Wang et al., 2020) are discussed and how they relate to the samples from this study (especially since table 1 lists isoprene as a target compound for this present study).

P14 -L479: Regarding typical MFs for PUY and REU: PC1 and PC2 seem to clearly differentiate between PUY and REU samples. There are also several MFs on the scores plot that align well with the R8, R9, and R10B loading vectors. What are the MFs mostly related to PC1 and PC2? It would be interesting to identify & extract these MFs and present them e.g. in an OSC vs #C space since they seem to be the relevant observations for the discrimination? This could also give a chemical meaning to the statistical analysis. At the moment it mainly serves to show that mass spectral fingerprints (also keeping in mind that these were measured on different instruments) are different.

Molecular composition of clouds: a comparison between samples collected at tropical (Réunion Island, France) and mid-north (puy de Dôme, France) latitudes.

Pailler et al.

General Comments

The purpose of this paper is to describe the molecular composition of dissolved organic matter in cloud water at a novel site in Reunion Island and compare it to Puy de Dome in France using primarily FT-ICR MS analysis. The samples are also compared to previous studies an use various metrics to evaluate the composition of the samples for comparison.

Overall, I feel this is a good paper that lays good groundwork for the analysis of cloud water in remote areas that have not previously been investigated with this type of analysis. There are some things that I am interested in and things that should be addressed before full publication, but they are relatively minor and should not hinder its publication in my view.

Specific Comments

1. Line 177: MFAssignR also incorporates H2O, CH2O, and O homologous series for formula extension. A citation of the package on GitHub, or the manuscript itself (Schum et al. Env. Res. 2020) would be a good addition to this section as well.
2. Line 274-275: Is there an explanation for why 22/10/2019 has so few MF compared to 8/10/2019? Or maybe why 8/10/2021 has so many more than the rest of the samples? It seems like the DOC is pretty similar between them, with the main differences coming from the inorganic ions. Do you think it is related to the actual sample itself, or to the blank subtraction method? Conservative blank subtraction is a good choice, but I am curious what the formula numbers looked like prior to blank subtraction and whether they were more similar at that point.
3. Lines 303-304: You mention that the average OSC is similar between PUY and REU autumn samples, while this can definitely just be a coincidence (considering the different sources and conditions) I was curious if you looked into the molecular formulas to see what sort of differences occurred in them. For example is the OSC heavily influenced in both cases by a common set of molecular formulas (even if they are different molecules) or are there really no similarities at all, they just happen to average out to the same OSC?
4. Lines 362-364: If I am understanding correctly, the general percentage of formulas in each classification is similar between REU and PUY, which seems reasonable, I am still curious about the specific differences between the molecules in one sample or another in a more comprehensive view. Do the formulas in each classification match each other between the different sites or are they largely different? For example, for the LipidC classification, are the formulas found at REU and PUY 90% common, 70%, 50%, less? I think it could be interesting to see if the detailed composition of these samples is very different or the same, since it may say something about the cloud processing results. The “averages” are very useful, but as you have mentioned, even the same formula doesn’t necessarily mean the same molecule, so if a set of molecular formulas are in a particular classification, they may not be similar in any other way, or they could be very similar and highlight that cloud processing brings organic matter to a similar specific result.
5. Lines 370: While the FT-ICR is very well suited and effective for this work, the lack of structural information is a shortcoming as noted here, is there any interest in doing LC or fragmentation analysis in the future for these samples or others?
6. Lines 378-384: You are taking appropriate caution in classifying these molecules as one specific class or another with the database, but I was curious whether if you took a few of the formulas that you have classified as “prenol lipids” for example and just looked for any molecule matching that formula (in other databases or the search engine of your choice) if you could get any other classification?
7. Lines 426-428: I do not quite understand this sentence. Are the measured concentrations for alpha pinene 0.5, 71.5, and 2 for R8, R9, and R10B, while the beta pinene concentrations were 39.9 and 1.3 for R8, R9, and R10B, or are the detection limits for alpha pinene 39.9 and for beta pinene they are 1.3? I think the sentence could be restructured for clarity.
8. Lines 457: Does this mean that the organosulfate intensity was low in all samples (REU and PUY) with the exception of PUY 8/10/2021, or are you just comparing PUY 08/10/2021 to other PUY samples? Additionally, you explain the higher occurrence of limonene organosulfates at REU by the increased emission of limonene at the site, which makes sense, but does that imply that the organosulfate formation from limonene is a faster process than the oxidation of pinene? My understanding of the reason given for the relative lack of pinene oxidation products is that the emissions were too fresh to have oxidized yet. Is the organosulfate a primary oxidation product like C8H12O5? Or is the explanation that there is more limonene emissions relative to the pinenes?
9. Lines 465: What were the N and S beta caryophalene formulas? Is there any way to know that the formulas are N or S caryophyllene molecules other than matching the formulas? While presence of their emission sources on the coast may explain the N and S beta caryophyllene, why would there be no CHO oxidation products? Are the N and S reactions that much more favorable than the O oxidation? Or is the concentration of N and S so overwhelming that the O oxidation doesn’t really occur, relative to N and S?
10. Line 524: According to the classification you say that 50% of the molecules observed are reduced, is the explanation that the organic matter in the clouds is fairly fresh and hasn’t had a chance to oxidize more completely yet?

Technical Corrections

1. Line 25: Somewhat contradictory statements, can consider changing the language a bit to get to the assumed intended meaning.
2. Line 179: It may be more consistent and precise to say “same mass” instead of “same peak”, since the parenthetical on line 180 says “unique mass”.
3. Lines 323: Should probably change “is” to “are”
4. Lines 520: Instead of “emitted” you should probably say something like “developed” or “produced”. Overall the language in this manuscript is very good, but there are few minor things, like this and the comments for lines 426-428 that could be adjusted.