From the journal Environmental Science: Atmospheres Peer review history

05-Jun-2024

Dear Dr Pichelstorfer:

Manuscript ID: EA-ART-04-2024-000054
TITLE: Towards automated inclusion of autoxidation chemistry in models: from precursors to atmospheric implications

Thank you for your submission to Environmental Science: Atmospheres, published by the Royal Society of Chemistry. I sent your manuscript to reviewers and I have now received their reports which are copied below.

I have carefully evaluated your manuscript and the reviewers’ reports, and the reports indicate that major revisions are necessary.

Please submit a revised manuscript which addresses all of the reviewers’ comments. Further peer review of your revised manuscript may be needed. When you submit your revised manuscript please include a point by point response to the reviewers’ comments and highlight the changes you have made. Full details of the files you need to submit are listed at the end of this email.

Please submit your revised manuscript as soon as possible using this link:

*** PLEASE NOTE: This is a two-step process. After clicking on the link, you will be directed to a webpage to confirm. ***

https://mc.manuscriptcentral.com/esatmos?link_removed

(This link goes straight to your account, without the need to log on to the system. For your account security you should not share this link with others.)

Alternatively, you can login to your account (https://mc.manuscriptcentral.com/esatmos) where you will need your case-sensitive USER ID and password.

You should submit your revised manuscript as soon as possible; please note you will receive a series of automatic reminders. If your revisions will take a significant length of time, please contact me. If I do not hear from you, I may withdraw your manuscript from consideration and you will have to resubmit. Any resubmission will receive a new submission date.

The Royal Society of Chemistry requires all submitting authors to provide their ORCID iD when they submit a revised manuscript. This is quick and easy to do as part of the revised manuscript submission process. We will publish this information with the article, and you may choose to have your ORCID record updated automatically with details of the publication.

Please also encourage your co-authors to sign up for their own ORCID account and associate it with their account on our manuscript submission system. For further information see: https://www.rsc.org/journals-books-databases/journal-authors-reviewers/processes-policies/#attribution-id

Environmental Science: Atmospheres strongly encourages authors of research articles to include an ‘Author contributions’ section in their manuscript, for publication in the final article. This should appear immediately above the ‘Conflict of interest’ and ‘Acknowledgement’ sections. I strongly recommend you use CRediT (the Contributor Roles Taxonomy, https://credit.niso.org/) for standardised contribution descriptions. All authors should have agreed to their individual contributions ahead of submission and these should accurately reflect contributions to the work. Please refer to our general author guidelines https://www.rsc.org/journals-books-databases/author-and-reviewer-hub/authors-information/responsibilities/ for more information.

I look forward to receiving your revised manuscript.

Yours sincerely,
Dr. Stephen Klippenstein
Associate Editor, Environmental Science: Atmospheres

************

Reviewer 1

This manuscript represents an investigation of benzene chemistry and is broadly useful to the atmospheric science community. The idea that a generated mechanism could be fine-tuned with observations is novel and could be very useful. The article would benefit from streamlining and clarifying some text, especially regarding the methods, and by summarizing the resulting advancement in knowledge gained.

Major comments:
1. In the end, what is the improved understanding of the benzene system? What do you see as the current gaps/model sensitivities that need further work? The list of reactions in the supplement provides a comprehensive extension of the chemistry, but not all reactions are equally important. Xu et al. 2020 proposed that autoxidation of the bicyclic RO2 would largely lead to RO species that would decompose. Molteni et al. 2018 observed a 0.2% HOM yield from benzene. How do your predictions of the chemistry either support this previous work or provide a different conceptual picture? This could be accomplished by better demonstrating how autoPRAM-fw “allows to reproduce observed CIMS-data evolution as well as SOA formation while proposing realistic (according to the literature) reaction parameters.”

2. Consider streamlining some of the method presentation. There are a lot of acronyms (autoAPRAM-fw, ADCHAM, ADCHEM). Some specific suggestions:
• In the first paragraph of Section 2 for Methods, remove the section titles and eliminate acronyms. Focus on a brief overview that follows a logical trajectory. For example, how information feeds from one method to another.
• Clarify this on Page 4 “RO2 composition of species formed by autoxidation chemistry which are not described in MCM are read in.” Is this a function done by autoSMILES or are these RO2 user generated by hand? Does autoSMILES make all possible structures? How is their abundance weighted and matched with autoReactions?
• Consider reducing the amount of words spent on vapor pressure estimation technique portions. Specifically, the major conclusions related to saturation vapor pressure methods didn't come through in the manuscript. I recommend reframing text to highlight one method and indicating the sensitivity of predictions to vapor pressure estimation method (1-3 sentences) in the main text. Eliminate section 4.1.3 in main text and refer the reader to the SI for additional information.
• Clarify in Figure 4b: are the APRAM CHON species which include a nitrocatechol product part of standard MCM or new?

Minor comments:
1. Page 3: “The metric to quantify the hazard is, albeit being studied intensely, not fully illuminated.6 “ Can you be more specific regarding what pollutant? ultrafines?
2. Abstract: “the calculated aerosol mass continuously increases.” Is that unique to your trajectory? Does mass continuously increase in a given airshed? What is the robust result?
3. Page 8: Where Psat of 1 Pa is mentioned, add a typical C* for broader audience.
4. Figure 2 caption: Clarify “APRAM monomer” indicates monomers predicted by autoAPRAM-fw. In the current form, it wasn’t immediately obvious if the colors were different simulations or different subsets of species from one simulation.
5. Figure 2: Double check panel b and the caption. Cumulative distributions usually increase left to right in a continuous way. The HOM fractions shows a peak near 175.
6. Figure 3: Could put titles on panels to facilitate reader.
7. Figure 3: why does the APRAM fraction decrease in Fig 3 b,d,f at high reacted HC but the APRAM lines grow in relation to the MCM line?
8. Figure 4d: Yield of what? Total products? SOA?
9. How were rates in SI reactions set? Were those fit to reproduce CIMS data or via SAR?
10. Were all simulations of experiments for low-NOx conditions?
11. Shorten the Section 4.1.1 title.
12. Some unnecessary words could be eliminated to make the article more concise.

Reviewer 2

Review attached.

We want to acknowledge the reviewer’s time and effort to review the submitted manuscript. Further, we are thankful for the constructive criticism which was considered carefully to improve the manuscript. Find below the answers to the issues raised and changes made in the text. Changes in modified sentences and added text are highlighted in the "markup"-version of the updated files.

Referee: 1
Comments to the Author
This manuscript represents an investigation of benzene chemistry and is broadly useful to the atmospheric science community. The idea that a generated mechanism could be fine-tuned with observations is novel and could be very useful. The article would benefit from streamlining and clarifying some text, especially regarding the methods, and by summarizing the resulting advancement in knowledge gained.

Comment:
Major comments:
1. In the end, what is the improved understanding of the benzene system?
Answer: The improved understanding of the benzene system mainly covers:
1) The proposed set of reaction equations describing autoxidation chemistry and competing reactions (R1 to R9) allow reproducing the experimentally observed CIMS spectrum for low NOx conditions. This is outlined in section 4.4, from 2nd paragraph starting with “In the present work we show that ...”
2) A potential relation between atomic composition and adduct formation rate coefficient was found (see equ. 1). To highlight this, we added a statement in the conclusions, section 3rd paragraph (see “changes made in the text” below).
3) The benzene autoxidation reaction scheme (derived from the pure gas-phase experiment) allows the reproduction of SOA mass formation in simulations of JPAC (steady state chemistry) as well as in CALTECH (evolving chemistry). This is outlined in section 4.4 Conclusions, 3rd paragraph, starting with “Predicting the species vapor pressure by group…”
4) Not a robust finding, but still interesting:
Our simulations highlight the potential importance of secondary or tertiary oxidation of benzene in the atmosphere. The oxidant, in this case, may be O3 or NO3 (note that benzene is oxidised by OH only under atmospheric conditions). → This is outlined in section 4.4, starting from “Apparent mass yields from daytime autoxidation chemistry”
Changes made in the text:
- 4.4. Conclusions section, 3rd paragraph: addition of “Thereby, we found a potential relationship (see equ. 1) between the rate coefficient and the molecular mass for the ROOR formation via reaction R5b.”

Comment:
What do you see as the current gaps/model sensitivities that need further work?
Answer:
Applying the autoAPRAM-fw to construct a reaction scheme and fitting the rate coefficients to experimentally derived CIMS-data needs to be done with caution:
1) In the autoAPRAM-fw several reaction pathways typically lead to the formation of a CIMS-detected peak. Accordingly, a dominant reaction channel can be equipped with a rate, while rate coefficients of the other, minor channels are not sensitive. To overcome, a set of experiments is needed covering different chemical regimes (high/low NOx; high/low RO2; high/low HO2).
→ This is outlined in section 4.1.2
2) Fitting the autoAPRAM-fw derived scheme to the experimental data takes considerable time. We are currently working on automatising this process (a manuscript on the methodology for doing so is in preparation), which is a necessary step to make the method applicable.
→ This is explained in the second paragraph of the conclusions section starting with “Currently, the fitting of reaction …”
3) Partly, the data processed in this method is subject to considerable uncertainty (e.g., wall partitioning of the species of varying volatility; CIMS spectra providing data in concentration per volume).
→ This is outlined in section 2.3 and in more detail in the SI-section “Interpreting ion count data from CIMS measurements”
An overview on the challenges and limitations can be found in sections 4.1 and 4.2

Comment:
The list of reactions in the supplement provides a comprehensive extension of the chemistry, but not all reactions are equally important. (Xu et al., 2020) proposed that autoxidation of the bicyclic RO2 would largely lead to RO species that would decompose. Molteni et al. 2018 observed a 0.2% HOM yield from benzene. How do your predictions of the chemistry either support this previous work or provide a different conceptual picture? This could be accomplished by better demonstrating how autoPRAM-fw “allows to reproduce observed CIMS-data evolution as well as SOA formation while proposing realistic (according to the literature) reaction parameters.”
Answer:
We compare our findings regarding rate coefficients to findings by Xu et al. (2020) in section 2.3 (last paragraph). Further, in the “Conclusions” section p. 10, left column. For high NOx experiments, the suggested decomposition of RO as the dominant reaction pathway does not seem plausible: it cannot explain the peroxy radical concentrations observed as well as the adducts formed.
The model results compare very well to the Molteni data (Molteni et al., 2018), as the rate coefficients of the scheme were fitted to this data (see Fig. 1, where we compare in depth the experimental observations and simulation results: the mass-spec – panel a; the distribution of oxidised species – panel b; the ratio monomer:dimer – panel c; and the RO2 distribution – panel d). The overall HOM fraction derived for CIMS data does not serve well to be compared to model data: the nitrate CIMS has low sensitivity to low oxidised species. Consequently, these species are strongly under-represented in the experimentally determined spectrum. To overcome, the present work focuses on comparison of higher oxidised products (see Fig. 1).
The present approach does not allow to pinpoint the most important pathway to autoxidation (e.g., autoxidation of the bicyclic RO2; autoxidation of some RO; criegee intermediates forming some species that can undergo autoxidation) from a single experiment. However, it allows to suggest a potential pathway that can explain the observations made with the CIMS.

Comment:
2. Consider streamlining some of the method presentation. There are a lot of acronyms (autoAPRAM-fw, ADCHAM, ADCHEM). Some specific suggestions:
• In the first paragraph of Section 2 for Methods, remove the section titles and eliminate acronyms. Focus on a brief overview that follows a logical trajectory. For example, how information feeds from one method to another.
Answer:
We agree and adapted the first paragraph of the methods section accordingly: we removed the section titles and all acronyms (except for the autoAPRAM-fw which is a key concept of the work and is introduced in the introduction section already). Further, we did some minor changes to improve clarity of the paragraph (last three sentences were edited).

Comment:
• Clarify this on Page 4 “RO2 composition of species formed by autoxidation chemistry which are not described in MCM are read in.” Is this a function done by autoSMILES or are these RO2 user generated by hand? Does autoSMILES make all possible structures? How is their abundance weighted and matched with autoReactions?
Answer:
The composition and structure of peroxy radical that are not part of MCM must to be provided. They are not generated by autoSMILES. autoSMILES produces the RO and the closed shell species structures (including adducts) for all RO2 species specified in the input. In the results shown in Fig 2 to Fig 4 of the main text , an autoAPRAM product is assigned a single structure (i.e., there is no weighting regarding the effects of an unknown variety of possible structures). Accordingly, it is a lumped approach with regard to peroxy radical isomers.
Changes made in the text:
“Further, RO2 composition of species formed by autoxidation chemistry which are not described in MCM, need to be provided by the user. They are not created by autoSMILES.”

Comment:
• Consider reducing the amount of words spent on vapor pressure estimation technique portions. Specifically, the major conclusions related to saturation vapor pressure methods didn't come through in the manuscript. I recommend reframing text to highlight one method and indicating the sensitivity of predictions to vapor pressure estimation method (1-3 sentences) in the main text. Eliminate section 4.1.3 in main text and refer the reader to the SI for additional information.
Answer:
We shortened the description on “Deriving the saturation vapor pressure” and moved the details to the SI. Evaporation is now highlighted among the group contribution methods (“In case a single SOA result is reported (i.e. for atmospheric simulations), the psat is based on EVAPORATION” – in 2.5) and the variation of SOA formed is given by NANNOOLAL and MYRDAL/YALKOWSKI.
Further, we added a summarising sentence at the end of the section highlighting that there is no evidence on which method best computs vapor pressures for the unknown variety of autoxidation products.
Changes made to the text:
1) Section 2.5 was rewritten and shortened.
2) Limiteations section (former 4.1.3) on the psat methods was moved to the SI. A regarding reference was added (“Limitations regarding the saturation vapor pressures assigned can be found in the SI.”)

Comment:
• Clarify in Figure 4b: are the APRAM CHON species which include a nitrocatechol product part of standard MCM or new?
Answer:
“APRAM CHON” species include all APRAM species containing a nitrogen atom. They either form in reaction R3a (RO2+NO→ RONO2) or in reaction R5b (RO2 + RO2 → ROOR). Products from the latter pathway do include (as reactants) peroxy radicals formed in MCM that include a nitrogen atom. We added some information to Fig. 4 caption.
Changes made in the text:
Panel b shows the modelled benzene gas-phase concentrations (Model [BZ]) and the modelled benzene SOA mass concentrations of non-nitrate APRAM species (APRAM CHO – species of atomic composition CxHyOz), APRAM organonitrates (APRAM CHON: nitrogen-containing species) and MCM species.

Comment:
Minor comments:
1. Page 3: “The metric to quantify the hazard is, albeit being studied intensely, not fully illuminated.6 “ Can you be more specific regarding what pollutant? Ultrafines?
Answer:
Here, we refer to the health effects of inhalable aerosol.
Changes made in the text:
“The metric to quantify the hazard of inhalable aerosol is, albeit being studied intensely, not fully illuminated”

Comment:
2. Abstract: “the calculated aerosol mass continuously increases.” Is that unique to your trajectory? Does mass continuously increase in a given airshed? What is the robust result?
Answer:
The main reason why the aerosol mass continuously increases for the selected air mass trajectory case studies is due to the air mass origin which is the clean Arctic Ocean between Greenland and Svalbard, 7 days before it reaches Malmö in Southern Sweden (see section 3.3). Continuous increase in SOA is not generally observed with ADCHEM trajectory simulations, which considers vertical mixing, as well as dry and wet deposition of aerosol particles along the air mass trajectory. However, to demonstrate the benzene SOA formation at variable NOx conditions we selected air mass trajectories with little anthropogenic influence upwind an urban region in Europe (Copenhagen+Malmö), and which after passing Copenhagen and Malmö continued over the less polluted (low VOC and NOx emissions) ocean regions.
However, we agree that the statement “...the calculated aerosol mass continuously increases...” may be misleading. Accordingly, we changed the statement. Further we added a statement on the choice of airmass trajectories to explain the accumulation in SOA mass.
Changes made in the text:
“Under atmospheric-like day-time conditions, the calculated benzene-aerosol mass continuously forms, as expected based on prior work.”
and we added the following statement at the end of paragraph 1 of section “3.3 Atmospheric implications”.
“To demonstrate this process, we selected airmass trajectories a) originating from the clean air above the Arctic Ocean between Greenland and Svalbard; b) travelling over an urban European region (Malmö/Copenhagen); and c) are characterized by changing NOx conditions during benzene SOA formation.”

Comment:
3. Page 8: Where Psat of 1 Pa is mentioned, add a typical C* for broader audience.
Answer:
We added the referring C* value.
Changes made in the text:
“To ease computation, only organic species with a psat lower than 1 Pa (which corresponds to a saturation mass concentration C* or roughly 0.1 g/m3) are considered to potentially partition to the particle phase.”

Comment:
4. Figure 2 caption: Clarify “APRAM monomer” indicates monomers predicted by autoAPRAM-fw. In the current form, it wasn’t immediately obvious if the colors were different simulations or different subsets of species from one simulation.
Answer:
The coloured bars represent results of subsets of species derived from one simulation. We changed the Figure 2 caption to clarify.
Changes made to the text:
“… and the simulation results (coloured bars show different subsets of species from one simulation...”

Comment:
5. Figure 2: Double check panel b and the caption. Cumulative distributions usually increase left to right in a continuous way. The HOM fractions shows a peak near 175.
Answer:
We intentionally start the cumulative representation from the right hand side for the following reasons: a) The sensitivity of the CIMS applied in the Molteni work has higher sensitivity for higher oxidised species. b) The concentration of dimers is much smaller than that of monomers. c) The APRAM-fw species mix with the MCM species at the lower end of the atomic mass spectrum. As a consequence, the information shown in the plot (the distribution of mass peaks formed fit the experimental distribution beyond individual peaks) would not be visible if species were cumulated from the lower mass end.

Comment:
6. Figure 3: Could put titles on panels to facilitate reader.
Answer:
We agree and added titles to Figure 3.
Changes in the manuscript:
We updated Figure 3 accordingly: the figures now clearly highlight the different experimental setups, which are “JPAC – low NOx”, “CALTECH – low NOx” and “CALTECH – high NOx”.

Comment:
7. Figure 3: why does the APRAM fraction decrease in Fig 3 b,d,f at high reacted HC but the APRAM lines grow in relation to the MCM line?
Answer:
For the benzene system, the APRAM species typically have a much lower vapor pressure than the MCM products. Accordingly, the APRAM species start the condensation (i.e., APRAM fraction is 1). The more organic mass there is, the more the higher volatility MCM products can partition to the particle phase. As a result, the share of the APRAM species gradually decreases.

Comment:
8. Figure 4d: Yield of what? Total products? SOA?
Answer:
Figure 4, panel d is on SOA mass yield.
Changes made in the text:
“Panel d illustrates the computed SOA mass yield from OH oxidation of benzene in the presence of different levels of NOx and VOC. “

Comment:
9. How were rates in SI reactions set? Were those fit to reproduce CIMS data or via SAR?
Answer:
Reaction rate coefficients in the autoAPRAM-fw generated benzene reaction scheme as given in the SI, are fitted to the CIMS data as outlined in section 2.3 of the main text. The SAR-like behaviour of the ROOR formation (equ. 1 in the main text) was discovered by chance.

Comment:
10. Were all simulations of experiments for low-NOx conditions?
Answer:
There are also simulations of experiments under high NOx conditions:
1) In section 3.1, we discuss the results of JPAC high NOx gas phase simulations.
2) In section 3.2, simulations of SOA formation under high NOx conditions are discussed.

Comment:
11. Shorten the Section 4.1.1 title.
Answer:
We have shortened the section title.
Changes in the manuscript:
“4.1.1 Main assumtptions: the autoAPRAM-fw, the chemical reaction types and structures formed”

Comment:
12. Some unnecessary words could be eliminated to make the article more concise.
Answer:
We shortened section 2.5 and moved section 4.1.3 to the SI. Further, we removed the last statement of the conclusions section.
The first paragraph of the method section was shortened as well.

Referee: 2

Comments to the Author
Review of Pichelstorfer et al for Environmental Science: Atmospheres

Pichelstorfer et al. presents the autoAPRAM-fw framework as an automated approach to model autoxidation schemes. The developed autoAPRAM-fw is based on MCM with additional peroxy radical chemistry. The autoAPRAM-fw method is applied to benzene to improve the known mechanism with autoxidation reactions. In addition, the model also considers and groups molecules relevant for aerosol processes.
The applied rate coefficients in the model are constrained on flow tube and chamber data. As a validation of the model, it is used to predict the product distribution measured by nitrate CI-APi-TOF and is found to reproduce the peaks well. Subsequently, the model is used with ADCHAM to model benzene flowtube and chamber experiments. It is found that the newly implemented autoxidation schemes can explain most of the formed SOA.
Finally, the authors apply autoAPRAM-fw in the chemical transport model ADCHEM. Again, the autoAPRAM-fw is found to produce the highest amount of SOA compared to the simple MCM model. Overall, the study illustrates the importance of HOMs in benzene oxidation. The derived model has many limitations. However, the authors are quite transparent in this regard in their discussion. In addition, it can be further augmented in the future, when new reaction paths or improved rate constants become available. The developed model is general in nature and could be applied to a variety of compounds. However, some input parameters are needed for modelling new systems.
The paper is well-written and structured. It is a bit tough to read due to the shear amount of text and information, but on the other hand I do not have any good suggestions for how to condense it further. Overall, this is very nice contribution to the field and the paper should be accepted after the following minor comments.

Comment:
Page 3: ”This increase has not yet been explored experimentally and questions the applicability of the widely accepted concept of OH-based SOA mass yield in the atmosphere.” This is perhaps a bit of a bold statement. Perhaps tone it down slightly.
Answer:
We agree and changed the statement to: “This increase has not yet been explored experimentally and stresses the potential for atmospheric SOA formation via secondary oxidation of benzene by O3 and NO3.”

Comment:
Figures: All the figures are a bit grainy. Could you please improve the resolution? In addition, the [1] for unitless quantities in the figures should be removed.
Answer:
The figures have now been uploaded separately with high resolution.
The term “[1]” to highlight unitless quantities has been removed from figures 2 b,d, 3 a-f and 4 d. An indicator “unitless” was added to the figure caption.

Comment:
Page 3: “Occurring symptoms are systematic in nature and can be severe.”
What do you mean with “occurring symptoms are systematic”?
Answer:
For clarity, we changed the sentence and added a reference to provide some context.
Changes made in the text:
“Occurring symptoms include systemic inflammation and can be severe.(Frampton, 2001)”

Comment:
Page 4: “(applying default rate coefficients)”
How are the default rate constants obtained?
Answer:
We agree, “applying default rate coefficients” is not very descriptive. We changed the text to increase clearity of the statement.
Changes made in the text:
“… capable to set up autoxidation chemistry schemes for any VOC system. Rate coefficients, if available, can be provided as an input. Otherwise they need to be derived from data.”

Comment:
Page 4: “Sect. 2.4 The molecular structures structures”
In general, when referring to the sections, it would be easier to remove the title of the section. Also remove the repeated word here.
Answer:
We removed the repeated word as well as the section titles.

Comment:
Page 4: “Potential atmospheric implications are illustrated in phase three”
I guess you mean section three here?
Answer:
We agree the term “phase three” may be misleading as we do not refer to phases before. We changed the text to “Finally, potential atmospheric implications are illustrated. To do so, a parameter study on SOA yield...”

Comment:
Page 4: “The framework itself is written in a way to describe any type of VOC undergoing autoxidation.”
From Figure 1, it is not completely clear to me whether the model then assign different rate constants depending on the functionalization R. For instance, if I give the model two “new” compounds, will it then end up producing the exact same product types in the same yields or is the chemical structure read in as a SMILES string and explicitly considered when assigning the rate constants?
Answer:
autoREACTIONS does not connect to autoSMILES with regard to the molecular structures – both tools can be run separately. As a result, functionalization of a structure does not affect the rate coefficients assigned by autoREACTIONS. The present approach is designed that way, as a single RO2 species in autoREACTIONS may represent a variety of species (note that the autoxidation chemistry is lumped in a way that all RO2 isomers are represented by a single autoAPRAM species). Further, rate coefficients as well as the dominant structures amongst the RO2 isomers are typically unknown. This is very unlikely to change in the near future (see section 2.4 – last paragraph): mass spectrometers used to detect the autoxidation products can only report atomic composition (SI - “Interpreting ion count data from CIMS measurements”).
On “… will it then end up producing the exact same product types …?”: If two (isomeric) sets of RO2 species are provided to the autoREACTIONS with the same names, it will produce the same reaction scheme for both species, irrespective of the SMILES strings. In case some of the rate coefficients are known, they can be considered by autoREACTIONS. In the present work, neither structures nor rates are known. As a result, we start with dummy rates and fit them to reproduce the experimental CIMS data (see section 2.3).

Comment:
Page 4: “Potentially, it can be run in an automated fashion allowing to apply machine learning techniques to tune the chemistry or investigate large numbers of RO2 isomers.”
Please elaborate on exactly how machine learning techniques are envisioned to be applied in this context. In the end of the manuscript, sit is further specified that random forest will be used, but some further information would be beneficial.
Answer:
We added some additional information on the ongoing work (see below).
Changes made in the text:
“We are currently exploring the possibility of applying machine learning in the following way: A chemical scheme created by the autoAPRAM-fw is used together with MCM chemistry to represent a VOC system. Next, we optimise the newly developed chemistry schemes by performing an in-depth Bayesian analysis of the VOC system. To sample all model parameters, we use methods like Hamilton Monte Carlo with automatic differentiation and Markov Chain Monte Carlo (MCMC).(Haario et al., 2006; Heikki Haario et al., 2001) Both techniques efficiently provide solutions to high-dimensional problems. The goal is to sample and quantify model parameters to increase accuracy and reliability in chemical schemes.”

Comment:
Page 5-6: Please fix the formatting of reaction R1-R9
Answer: we are sorry for the bad formatting. This, likely, is caused by incomplete communication between the open office formula editor (which is used by the authors to create the equations) and Microsoft word. Changes: the equations have been uploaded as a high-resolution figure to make sure they appear correctly.

Comment:
Page 6: “Under high NO conditions, the formation of RO2 via RO seems to be an important path to reproduce the observed peroxy radical levels.”
I guess this should be the other way around. RO is formed from ROO+NO
Answer:
We do agree, the reactions RO2 + NO/RO2/HO2 may dominantly form RO for the benzene system (see SI scheme where RO is formed from RO2+NO [59-99%]; RO2+HO2 [40 – 99%]; RO2+RO2,pool [~ 60%]).
However, in the mentioned section we discuss the RO2 levels observed. The high NO concentration effectively removes the RO2 by forming closed-shell products and alkoxy radicals. Accordingly, we either overestimate the loss of RO2 via reaction with NO or a considerable source of RO2 is required to compensate for some of the loss.
Additionally, autoxidation seems to take place under high NOx conditions as well. This is indicated by closed shell species detected. Since RO2 levels are too low to explain the autoxidation products, another pathway seems required.
A potential solution is the autoxidation of RO species: it’s suggested to happen fast enough(Vereecken and Peeters, 2010) and raises the general RO2 level.

Comment:
Page 7: “For decades already a specific structure, the bicyclic peroxy radical (see Supplementary Fig S11), is suggested to be a key RO2 structure that forms upon a single OH attack on benzene followed by O2 additions”
It has recently been suggested that the formed bicyclic peroxy radicals can lead to epoxide formation in aromatics, where the ring structure is still intact (10.1039/D3SC03638C). Such products have a large effect of SOA formation. It would perhaps be beneficial to include this mechanism in the model. No need to do the full implementation here, but it could be mentioned as an outlook for further work.
Answer:
We included a statement highlighting the potential impact of epoxides on SOA formation via reactive uptake.
Changes made in the text:
“Another species group potentially contributing to reactive uptake represents epoxides: they may form via secondary oxidation of aromatic species. At the aerosol surface, they may transform into less volatile species via acid-catalyzed ring-opening reactions.(Fu et al., 2023)”

Comment:
Page 7: “… , EVAPORATION shows best agreement with the COSMO-RS results for all species groups” I would perhaps briefly mention in the text that the “MYRDAL/YALKOWSKI” method seems to be quite off compared to the two other models.
In addition, as the psat has a huge effect on the SOA yield, it would be worth further stating how much EVAPORATION and COSMO-RS differs. Is it a consistent over/underestimation and by how large a magnitude?
Answer:
We did some quantification of the agreement/deviation of the methods applied. Following a suggestion by the other referee, the description of psat determination methods was largely moved to the SI.
Changes made in the text:
The SI section “Distribution of saturation vapor pressures from group contribution methods” was rewritten. Some quantification of the deviation from COSMO-RS has been added together with some qualitative analysis on the consistency of results.

Comment:
Page 10, Figure 1: In the methods section it was mentioned that the rate constants are constrained to the experiments. It is not clear to me whether the experiments you are comparing to in Figure 1 the same as you used for constraining the model? If that is the case, this is simply a case of data fitting, and the agreement should be good. On the other hand underprediction of products could also allude to that the model is potentially missing some reaction mechanisms.
Answer:
Yes, the experimental data shown in Figure 1 is the data the model was fitted to. We fully agree that deviations between model and experiment may highlight some missing pathways: see section 3.1, text following Figure 1, where we discuss potential reasons for the observed discrepancies in adduct formation between model and experimental data. We particularly mention the oxidation of closed shell APRAM species as a potential (omitted) pathway to explain observed discrepancies.
Further, we added a respective statement to section 2.2 (at the very end).
Changes made in the text:
“Note that the reaction equations (R1 to R9) proposed are considered to be relevant in autoxidation chemistry. However, they are not believed to be a closed list. Other reaction types may show important and may be added in future.”

Comment:
Page 13: “(see “APRAM CHO”).”
Perhaps specify that you mean Figure 4, b here.
Answer:
We agree and added the information.
Changes made in the text:
“… drive the SOA formation during daytime via autoxidation chemistry (see “APRAM CHO” in Fig. 4b).”

Comment:
Page 17: “We are convinced that air quality forecasts, knowledge of the impacts on climate, and the basis for legislation can be significantly strengthened by following the approach introduced in this work.”
This is quite speculative and overexaggerating the importance a bit. Perhaps tone it down slightly.
Answer:
We agree and removed the statement.

Comment:
Technical comments
Page 8: Here the section title “2.6.1 Flow tube” should be bold.
Answer:
We changed the font style to bold.

Comment:
Page 8: benzene eaction -> benzene reaction
Answer:
We changed to “benzene reaction”

Comment:
Page 14: to visulaize the potential -> to visualize the potential
Answer:
We corrected the spelling to “visualize”.

Comment:
Page 14: maybe -> may be
Answer:
We applied the suggested change

Comment:
Page 16: by a recent work -> by recent work
Answer:
We applied the suggested change

Comment:
Page 17: NO(x) -> NOx
Answer:
We applied the suggested change


References:

Frampton, M. W.: Systemic and cardiovascular effects of airway injury and inflammation: ultrafine particle exposure in humans., Environ Health Perspect, 109, 529–532, https://doi.org/10.1289/ehp.01109s4529, 2001.
Fu, Z., Ma, F., Liu, Y., Yan, C., Huang, D., Chen, J., Elm, J., Li, Y., Ding, A., Pichelstorfer, L., Xie, H.-B., Nie, W., Francisco, J. S., and Zhou, P.: An overlooked oxidation mechanism of toluene: computational predictions and experimental validations, Chem. Sci., 14, 13050–13059, https://doi.org/10.1039/D3SC03638C, 2023.
Haario, H., Laine, M., Mira, A., and Saksman, E.: DRAM: Efficient adaptive MCMC, Stat Comput, 16, 339–354, https://doi.org/10.1007/s11222-006-9438-0, 2006.
Heikki Haario, Eero Saksman, and Johanna Tamminen: An adaptive Metropolis algorithm, Bernoulli, 7, 223–242, 2001.
Molteni, U., Bianchi, F., Klein, F., El Haddad, I., Frege, C., Rossi, M. J., Dommen, J., and Baltensperger, U.: Formation of highly oxygenated organic molecules from aromatic compounds, Atmos. Chem. Phys., 18, 1909–1921, https://doi.org/10.5194/acp-18-1909-2018, 2018.
Vereecken, L. and Peeters, J.: A structure–activity relationship for the rate coefficient of H-migration in substituted alkoxy radicals, Phys. Chem. Chem. Phys., 12, 12608, https://doi.org/10.1039/c0cp00387e, 2010.
Xu, L., Møller, K. H., Crounse, J. D., Kjaergaard, H. G., and Wennberg, P. O.: New Insights into the Radical Chemistry and Product Distribution in the OH-Initiated Oxidation of Benzene, Environ. Sci. Technol., 54, 13467–13477, https://doi.org/10.1021/acs.est.0c04780, 2020.

08-Jul-2024

Dear Dr Pichelstorfer:

Manuscript ID: EA-ART-04-2024-000054.R1
TITLE: Towards automated inclusion of autoxidation chemistry in models: from precursors to atmospheric implications

Thank you for submitting your revised manuscript to Environmental Science: Atmospheres. I am pleased to accept your manuscript for publication in its current form. I have copied any final comments from the reviewer(s) below.

You will shortly receive a separate email from us requesting you to submit a licence to publish for your article, so that we can proceed with the preparation and publication of your manuscript.

You can highlight your article and the work of your group on the back cover of Environmental Science: Atmospheres. If you are interested in this opportunity please contact the editorial office for more information.

Promote your research, accelerate its impact – find out more about our article promotion services here: https://rsc.li/promoteyourresearch.

We will publicise your paper on our Twitter account @EnvSciRSC – to aid our publicity of your work please fill out this form: https://form.jotform.com/211263048265047

How was your experience with us? Let us know your feedback by completing our short 5 minute survey: https://www.smartsurvey.co.uk/s/RSC-author-satisfaction-energyenvironment/

By publishing your article in Environmental Science: Atmospheres, you are supporting the Royal Society of Chemistry to help the chemical science community make the world a better place.

With best wishes,

Dr. Stephen Klippenstein
Associate Editor, Environmental Science: Atmospheres

Reviewer 2

I believe the authors have done an excellent job incorporating the comments from both reviewers. Please publish as is.