|Reviewer comments on acp-2019-927|
The authors present here an investigation of how FIGAERO-CIMS thermal desorption data can be used to understand volatility distributions and evaporation kinetics, specifically through comparison to isothermal evaporation experiments. Overall, I think this is useful work of interest to the ACP community, and I think the authors have explored and considered in detail many of the potential areas of uncertainty in the approach. My opinion is that is generally suitable for publication in this journal. I have two main comments below that I think need to be discussed and addressed prior to publication, and list a number of specific comments below (in some cases, specific comments may just be specific examples/cases of the general comments and can be responded to as such).
1) Presentation. The authors are very detailed, but I fear this may contributed to the fact that I find this paper and these figures hard to get through. With all the different comparisons and model paramters, it takes a lot of re-reading sentences and mentally refreshing myself on the modeling frameworks to keep on top of what I am looking at. This is particularly true in considering all the variables in the evapogram modeling. Unfortunately I'm not sure I have a lot of concrete suggestions for how to fix this issue, it may just stem out of all the modeling details. I think one clear thing that might help would be to make the "takehome points" of the figures more clear within the figures themselves instead of relying entirely on captions and legends, for example: in all figures, including in each subplot what data set (e.g. "Fresh, medium O:C SOA") is being shown, in Figure 2 move most of the legend into the plot by just labeling "VD_evap" in gray and "Factors" in colored text, etc. Other possible ways to increase clarity might be: refer to and label the volatility classes as "low", "semi-volatile", and "volatile" instead of 1, 2, and 3 so the read doesn't have to keep track; spend less time discussing (and labeling in figures) the issues around min, mean, and max evaporation times since it doesn't really impact the conclusions and move most of that to the SI. A minor but still important issue is awkward phrasing and grammar - on top of general complexity of the presentation, there are a lots of grammatically questionable phrases and/or typos that should be fixed (some, but not all, are listed in the specific comments).
2) Benefit of PMF. A core component of this work is the PMF analysis of the FIGAERO data, and the authors do a detailed investigation to understand thermogram-derived volatility. A major conclusion of this work is that PMF factors can seemingly be used to describe the volatility/evaporation, but only if you optimize the T --> C* conversion by fitting to evapogram data (i.e., account for the "uncertainty in the desorption temperature"). However, it's totally unclear to me that the PMF step is at all necessary, given the scale of that uncertainty, and the typical absence of evapogram data to provide that constraint. The C* that describes each factor is uncertain to around an order of magnitude, bounded by the desorption temperature range of the factor. Given this range of uncertainty (which is typical for C* estimates), what do you gain by having some specific T_max or T range associated with each factor? Why not just cut the thermogram into VBS bins based on temperatures (either by cutoff temperatures, or by fitting peaks constrained to the temperatures defining each bin)? Would this approach do any worse a job in comparison to the VD_evap or evaporation kinetics? This might even do better - M4, for instance, is very broad with T_max near T_25, so forcing all mass in the factor to this range might drive some bias. There are reasons beyond volatility you might want PMF, but this paper does not convince me that the effort of PMF provides any benefit in estimating volatility, and if not, why is it being used at all? Something needs to be added to this manuscript to provide support or context for this decision, e.g., a discussion of other peoples use of PMF for volatility; a discussion of why PMF might reasonably be expected to do a better job than a simple VBS approach; or a comparison of the present PMF approach to a VBS-only approach. Ideally, I'd love to see the VBS-only approach applied because if it works it informs how one can use the thermogram without the more complex need for PMF, but I would understand if the editor and/or authors feel that is beyond the scope. If it is beyond the scope, I do think it should be considered and discussed as a possibility - one major takehome for me is that it seems like it should work at least within similar unceratinty, so I think it would broaden the audience and potential impact of this paper to explore the possibility.
Line 40-41. This whole sentence is awkward English and difficult to understand, re-phrase
Line 41. Should be phrased "the phase state...has also..."
Line 45. Mass Spectrometer should be capitalized
Line 77. "conduction a" is a typo
Line 82. For "How to interpret..." is not a grammatically correct question, should be "How shoud....be interpreted?"
Line 104-107. Do I understand correctly that these 80 nm particles sit in the 100 L chamber for 4-10 hours? Even a monodisperse population will have some size distribution - how do you account for particle-dependent wall losses that might change the apparent size distribution? These might be negligible, but if so, the authors should provide evidence to support such an assumption.
Line 111. I don't think you need to include Aerodyne Research Inc. here, you already reference it in the paranthetical at the end of the sentence.
Line 114. It might be worth noting that in Isaacman-VanWertz et al. cited here, the volatility distribution of the FIGAERO-CIMS based on thermograms was similar to that of a TD-AMS. This suggests that any overall conclusions from the present work likely extend to other thermal desorption based estimates of volatility.
Line 115. "a-" instead of "alpha-"
Line 143. Why not just call them "VD bins" in all cases, and avoid the confusion with "compounds"? Having read the paper in detail, it's not clear to me that the use of the term VD compound is at all necessary - each bin has average properties (e.g., T_max) and I don't see any need to refer to them explicitly as psuedo-individual compounds.
Line 144. Are these really the properties of each VD compound? It looks like just basic assumptions about the properties of all (not each) bin. This sentence makes me think the Table is going to contain a list of many different properties for each of the bins.
Table 1. Use column and row dividers to make clear that the rows with only one value are for all columns
Line 219. "and conducting a" should read "and conduct a"
Line 223-224. I don't understand the purpose of this interpolation. Isn't the mass loading profile basically a represenation of mass as a function of time/temperature? What do the authors mean a temperature "step". Usually, the CIMS collects data at ~1 Hz - how large a step in temperature occurs in one second? Do the authors mean they interpolate 100 spectra per degree C? Or 100 spectra per Hz? I would guess that if you are just interpolating from your existing data, this would not actually increase your statistical power, since the amount of "real" information is not increasing, but I'm not a statistician so I'm not sure.
Line 243. Missing close parenthesis
Line 251. Out of curiosity, how was 0.3 nm chosen? This is approximately the length of 2 carbon-carbon bonds, so effectively a monolayer or thinner.
Line 256-258. This assertion is made a few times, but it's not clear to me if this is simply an assertion, or if it is also observed that the VD_evap model suggests this to be true as well. This should be clarified/discussed
Line 271. The mass spectra contains ions, not compounds. In any case, do I understand correctly that DeRieux et al provide a way to estimate Tg as a function of molecular formula? This isn't quite clear.
Line 285. What is the difference between Figures 1 and S2? I see that the contaminant and blank factors have been removed, and MD1 and has been split, but this is not all of the differences. For instance, in Figure S2 factor M1 coes to ~5000 signal, but only 2000 signal in Figure 1. The caption seem to imply these are the same data, but they don't look like it. Also, the desorption temperatures in Figure 1 go down to 0 C, which I don't think is correct.
Line 288. I think it's fine to point the reader to Buchholz for details, but a one or two sentence summary explanation is still necessary, I think, so the reader does not have to go to Buchholz unless they want the details. In other words, its fine to have a companion paper, but to be a separate paper, this paper still needs to be readable and understandable on its own.
Line 291. How was it determined that they were "clearly an artifact"?
Line 295. This is a run-on sentence and should be split apart, probably at the first comma.
Line 299-301. I see why the authors split M/LD into two peaks, but this is a somewhat false dichomomty between the M/L peaks and the M/LD peaks - in reality they probably all contain some decomposition, it's just that for the other peaks it is a smooth enough transition as to appear monomodal.
Line 323. It would helpful to note here (based on the SI) that the selection of min, mean, or max evap time does not significantly impact the conclusions of this work.
Figure 3. It might be helpful to include error bars. Sources of uncertainty on VD_evap presumabely include min, mean, max evap time, and possible uncertainties in the LLEVAP model. Sources of uncertainty in VD_PMF are harder to assess - definitely uncertainty in the T --> C* conversion as discussed below, also theoretically PMF uncertainties but those are harder to understand in this context, and maybe other things?
Similarly, what is the benefit of grouping VD_PMF in this way? It requires all the mass in one factor to be assigned into a bin based on T_max. Why note just slice the thermogram by the relevant temperatures, and bin mass just based on evaporation temperature (without the need for PMF)? I guess that's actually an overall question - does PMF really improve modeling or understanding of volatility, or would slicing the thermogram into C* bins just based on temperature yield basically the same conclusions?
Lines 391-392. It is reassuring that the desorption profile is usable in this way, but it is notable that uncertainty in the C* of each factor seems to be on the order of 1 log unit based on Table 2. Such issues are typical in volatility experiments, but really suggest that this assumption of using T_max to describe a PMF factor is highly uncertain. Unless the operator has some evapogram data to validate against (which is of course uncommon), it is not clear that such an assumption should be used, nor is it clear that uncertainty in the desorption temperature can feasibily be considered. This does not mean the present work is not valuable, because the assumptions being tested are being used by the community, but I think it actually should present more doubt or caution in applying the assumptions being tested.
Line 506-507. Qualitatively I agree, but it's not clear that is quantitatively true - attempting to model evaporation using the PMF results alone (without optimization by comparison to an evapogram) yields moderately but not wholely successful results (Fig. 4). This is addressed below, but I think maybe this starting sentence needs to be tempered.