Measurement report: Characterizing O<sub>3</sub>-NOₓ-VOC sensitivity and O<sub>3</sub> formation in a heavily polluted central China megacity using multi-methods during 2019&ndash;2021 warm seasons

Yu, Shijie; Liu, Hongyu; Wang, Hui; Su, Fangcheng; Yuan, Minghao; Zhang, Ruiqin

doi:10.5194/egusphere-2025-4519

Preprints

https://doi.org/10.5194/egusphere-2025-4519

Preprints

23 Oct 2025

| 23 Oct 2025

Status: this preprint is open for discussion and under review for Atmospheric Chemistry and Physics (ACP).

Measurement report: Characterizing O₃-NOₓ-VOC sensitivity and O₃ formation in a heavily polluted central China megacity using multi-methods during 2019–2021 warm seasons

Shijie Yu, Hongyu Liu, Hui Wang, Fangcheng Su, Minghao Yuan, and Ruiqin Zhang

Abstract. This study investigated the high ozone pollution in Zhengzhou City from 2019 to 2021 using observational data and model simulations, focusing on volatile organic compound (VOC) pollution and its impact on ozone formation. Using online VOC data and statistical analyses, the results showed that VOC concentration increased with ozone pollution level, with average values of 84.7±51.0, 96.6±53.4 and 105.3±59.4 µg/m³ for non-pollution, mildly polluted and moderately polluted periods, respectively. Source apportionment of ozone and its precursor VOCs was performed using CMAQ and PMF models. The results demonstrated that reducing vehicle emissions should be prioritized to mitigate ozone pollution in Zhengzhou, as ransportation emissions accounted for 64 % and 31 % of ozone and VOC emissions, respectively. In addition, local ozone production rates and HOx base budgets were calculated using an observation-based model (OBM). The ozone production rates on non-pollution, mildly polluted, and moderately polluted days were respectively 2.0, 4.5, and 6.9 ppbv/h on average. The HOx radical concentration on polluted days was 1.5–6.4 times higher than that on non-pollution days, which is indicative of more efficient radical cycling during photochemical pollution. The O3-NOx-VOC sensitivity was analyzed using the OBM model, CMAQ model and ratio method. The results showed that ozone generation in Zhengzhou was mainly limited by VOCs, suggesting that the reduction of VOCs should be focused on aromatic hydrocarbons and olefins. The optimal reduction ratio of anthropogenic VOCs to NOx was about 2.9:1. This study will offer deeper insights for formulating effective ozone pollution prevention and control strategies.

Received: 14 Sep 2025 – Discussion started: 23 Oct 2025

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 2080 KB)

Supplement (1071 KB)

Download & links

Shijie Yu, Hongyu Liu, Hui Wang, Fangcheng Su, Minghao Yuan, and Ruiqin Zhang

Status: open (until 03 Jan 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on egusphere-2025-4519', Anonymous Referee #1, 27 Nov 2025 reply
This manuscript investigates O₃–NOx–VOC sensitivity and O₃formation mechanisms in Zhengzhou (2019–2021) using a combination of online VOC measurements, OBM, CMAQ-DDM, source apportionment, PMF, and machine-learning (ML)/SHAP interpretation. While the dataset is valuable and the research direction is meaningful, the manuscript suffers from inconsistent methodology, unclear descriptions of model configurations, uncertainties and machine learning, no comparisons between different analytical models for the O₃ formation mechanism, and very vague data interpretation. The manuscript is long and unreadable. Therefore, I suggest the manuscript be rejected.
Major Comment:
In Section 3, “Results and Discussion”, there is a substantial focus on O₃ source apportionment (CMAQ) and VOCs source apportionment (PMF). Unfortunately, how these contents are linked to the research question “O₃-NO_x-VOC sensitivity” is weak. For example, the authors claim that traffic and industry dominate both O₃ and VOCs. But there is no justification for “X% of O₃ from traffic” and “how O₃ responds if traffic NO_x/VOCs are reduced”.

In addition, there is no cross-method comparison. A combined table is highly recommended to show the section contributions across PMF, CMAQ-DDM, and OBM RIR/EKMA. As a result, the manuscript reads like a report by stacking results (sensitivity diagnostics + VOC and O₃ source apportionment + ML/SHAP), but with limited discussion.
The CMAQ simulation suggests that transportation and industry dominate MDA8 O₃ contributions, while the PMF result indicates that vehicle, solvent, and industrial sources contribute most VOCs. The current discussion is very vague: “transportation should be prioritized,” “more aggressive control is required,” etc. However, what is the take-home message for policy translation? There are no recommended emission-reduction scenarios (e.g. -30% traffic VOCs, -30% industrial NO_x, different VOC/NO_x ratios by sector) and no quantitative estimate of the expected O₃ under those scenarios.

Machine-learning setup and data leakage risk are not clearly addressed. The authors use XGBoost and RF with grid search and cross-validation (CV), and report training/test metrics in Table S4. However, essential methodological details are missing. How are training and test sets defined—random split or chronological split for this time-series dataset? What CV scheme is used (k-fold, blocked time-series CV)? Does CV account for temporal ordering?

What is the inherent relationship between the VOCs/NO_x ratio method for determining O₃sensitivity, modelling results from OBM and CMAQ, and radical budgets?

How do the results of XGBoost relate to those of other methods? Why is XGBoost analysis not mentioned at all in the Section 4 Summary and Conclusions? How do the authors come up with the claim that the ML method should be advised in the future in the subsection Limitations and future research directions in Line 986?

The role of biogenic VOCs is mentioned but not fully clarified in terms of O₃ formation and control. Biogenic VOCs are explicitly identified as a separate PMF factor with isoprene as tracer, and OBM RIR shows that biogenic VOCs have non-negligible reactivity, especially on polluted days. Emissions used in CMAQ also include biogenic VOCs via MEGAN. On the one hand, the authors claim that the contribution of biogenic sources to VOC is small (Line 724). On the other hand, they claim that the contribution of biogenic VOCs to local O₃ is high (Lines 840-850). So, I wonder what the overall role of biogenic VOCs in the O₃–NO_x–VOC sensitivity and O₃ formation mechanism is.

Minor Comment:
The period simulated by the CMAQ model was not mentioned in the methodology.

What is the rationale for studying warm seasons? Could you provide any justification and supporting references?

Lines 141-143: Could you provide any reference for the study city of Zhengzhou?

Lines 212-215: “PO₃^S(X) and PO₃^S (X-DX) refer to the simulated O₃ yields…” Are you sure “yield” is the right terminology here? Why does the author later state that “The net O₃ production rate (PO₃^S)” in Lines 215-216?

The CMAQ result shows transportation can contribute 64% to O₃(Fig.4), but the PMF result shows vehicles contribute only 31% to VOCs. This apparent discrepancy needs justification. Presumably, VOCs and NOₓ emissions from transportation are equally important for O₃ formation. Therefore, the role of NO_x emissions needs to be discussed.

Why does Table S5 show a negative correlation (Kendall’s, -0.305) between VOCs and O3, while the RIR indicates a positive correlation (Fig.9)? Shouldn’t this discrepancy require some explanations?

Table S4: The rationale for selecting XGBoost over RF is well justified.

A strong positive correlation between O₃ and alkenes is observed in SHAP (Fig.2), whereas Spearman's analysis reveals the opposite relationship (Table S6, -0.054**). Please standardise the terminology for “olefins” (Fig.2) or “Alkene” (Table S6)!

Line 427: “160 160 µg/m³ and 160 µg/m³ are categorised as light pollution and moderate pollution”? Could you please carefully check if this is CORRECT?

9: What are “AVOCs”, “ALKA”, “ALKE”, “ARO”? What is the relationship between them? Which subsets of compounds belong to AVOCs? Why is there no further discussion of the contribution and sensitivity of VOCs from different sources to O₃?

Technical Comment:
2 The font is too small to read.

Line 351: What is the full name of “XGBoost”?

Line 386: Should it be “SHAP” instead of “Shapley”?

Table S3: What do “DISP” and “BS” mean? What is the meaning of BS mapping＜80%?

S3: A full description needs to be provided for “E/X” in the manuscript for the first time and figure captions.

Line 712: “Zhengzhou comprises vehicle emissions (31%), solvent use (24%), and industrial processes (21%)”. But in Table S8, the corresponding values are 32.4%, 24.8% and 18.3%.

Lines 1135-1137: Make sure the font style is consistent throughout.

Please be consistent in the use of subscripts and full names or abbreviations for proper nouns.

Reply
Citation: https://doi.org/10.5194/egusphere-2025-4519-RC1
RC2:
'Comment on egusphere-2025-4519', Anonymous Referee #2, 15 Dec 2025 reply
In “Measurement report: Characterizing O₃-NOₓ-VOC sensitivity and O₃ formation in a heavily polluted central China megacity using multi-methods during 2019–2021 warm seasons” Yu et al. investigate the sensitivity of O₃ formation in Zhengzhou City between 2019 and 2021 based on VOC in-situ observations, several models and machine learning tools. The authors find that O₃ formation in the Chinese megacity is limited by the availability of VOCs and recommend a focus on VOC emission reductions with simultaneous NO_x control.
While the investigation of O₃ pollution generally remains a highly important topic, I have major concerns regarding the implementation and results presented in this study. The large number of methods, that often seem redundant, the use of many abbreviations (often not defined) and changes between units makes it difficult to follow the line of argument. The lack of presenting NO_x measurements and a detailed discussion on the role of NO_x in O₃ formation makes it additionally challenging to understand how the authors reach their scientific conclusions. Unfortunately, I therefore cannot recommend this manuscript for publication in ACP.
Major Comments:
Manuscript type: The manuscript type “Measurement Report” does not seem appropriate to me. A measurement report should “present substantial new results from measurements of atmospheric properties and processes from field and laboratory experiments” (https://www.atmospheric-chemistry-and-physics.net/about/manuscript_types.html). While it can be accompanied by model results, the focus should be on the presentation of a unique dataset rather than a pool of different methods, including modeling and machine learning tools.

Number of methods: The authors use a large number of different methods to investigate O₃ formation sensitivity. It is not clearly established why all these methods are needed and what their added value is. The questions posed could be answered with a reliable set of in-situ observations and an appropriate model to simulate missing trace gases. Instead, the introduction of all these methods is confusing and makes it difficult to follow the central line of argument.

Measurements: From the results section, I understand that besides VOCs, measurements of NO_x and O₃ are available. However, there are no details presented in the methods section.

NOx: Besides VOCs, Nitrogen Oxides are important precursors to tropospheric O₃. However, the role of NO_x seems to be mostly neglected in this study. A detailed description and discussion of the role of NO_x are missing and I wonder how the authors reached their conclusions on O₃ formation sensitivity, without accounting for NO_x. E.g. the authors state that chemistry is VOC-limited in Zhengzhou, but suggest that VOC control is more important than NO_x. VOC-limited chemistry is characterized by a large excess of NO_x, which requires drastic emission cuts to improve long-term air quality.

Correlation analysis: The presentation of correlations between different parameters in Section 3.1.1 seems random and does not follow a clear strategy, e.g. hypothesis – method – result – discussion. The set of in-situ observations is much more powerful than this: I recommend presenting trace gas levels (and if possible a longer time series), the characteristics of each season, diurnal cycles and the weekend effect for a sensitivity analysis. The application of machine learning tools is not necessary here or needs to be better justified.

Abbreviations: Many abbreviations are used in this manuscript, and they are often not defined upon first use, which makes it difficult to follow. It is further concerning that the authors are in some cases not consistent with the abbreviations, e.g. “OBM” is an “observation-based model” in Line 41 and an “Ozone Box Model” in Line 128.

Units: Many different units for trace gases are used throughout the text, including ppbv, ug/m3 and molecules/cm3. This makes it difficult to compare trace gas levels and I recommend choosing one unit (preferably mixing ratios) and using it throughout the entire manuscript.

PM2.5: Why is PM2.5 relevant to this study? I recommend focusing on O₃ and its precursors to avoid overloading this study.

Minor comments:
33 f.: This sounds like VOCs increase in response to O₃ increases, while VOCs are precursors to O₃.

34 f.: Do these values refer to VOC or O₃ concentrations?

37: Please define abbreviations upon first use.

39: What is meant by “ozone emissions”? Ozone is not emitted but formed photochemically.

46: What’s the “ratio method”?

47: If ozone generation is limited by VOCs, it is highly important to control NO_x. Of course, it remains important to reduce VOCs simultaneously, but long-term air quality improvements can only be reached through NOx reductions in that case.

55 ff.: Several things are missing in the introduction, i.a. how O₃ is formed from its precursors and particularly what the role of NO_x is.

65 ff.: Are the authors saying that they are the first to investigate O₃ formation from increasing anthropogenic sources?

76 f.: This sounds like the range of VOC mixing ratios in China is 27 – 92 ppbv.

81 ff.: This section is difficult to follow due to the jumps between countries and continents.

83: Is BB the major VOC source throughout the entire year?

93 ff.: Are the authors talking about concentrations, emissions, formation rates or sensitivities?

115 ff.: Why not use a set of in-situ observations?

128: In the Abstract the authors state the OBM to be an observation-based model.

188: How were other trace gases and meteo parameters measured?

215: More details on the OBM are required.

223: The reaction of OH and NO₂ does not destroy O₃ but limits its formation. It should therefore be accounted for in Equation (3), rather than (4).

242: More details are needed on the WRF/CMAQ model.

248 ff.: What are all these abbreviations: FNL, SAPRC-99, AERO6, IC/BC, MEIC, REAS2?

262 f.: What are first- and second-order sensitivities of O₃?

265 ff.: What exactly do these equations show?

277: Why is PM2.5 needed in this study?

283: Why would the model be better at simulating emitted species?

284: Only a small fraction of NO2 is emitted directly, most is formed photochemically from NO.

319: What’s MDL?

348: Why exactly is machine learning needed in this study?

425 / Fig. 1: Why is a smoothing applied? What exactly does it involve? Why is the time series not just averaged to the desired resolution?

426 ff.: All three numbers are the same. How exactly are the pollution levels defined?

432: What exactly do the percentage values relate to? If it’s years, the time period is too short for a trend analysis.

437: Why is O₃ positively correlated with wind speed? Usually, higher wind speeds lead to less accumulation?

438: Because H₂O contributes to O₃ loss? These correlations need to be discussed in more detail.

476 / Fig. 2: Why is half the figure upside down?

482 f.: It should be specified what is meant by “affecting boundary layer structure” – the current term is very generic.

494: HO₂ can be lost on aerosol surfaces, which inhibits O₃ formation (opposite effect!)

522 ff.: I cannot follow this logic. NO is lower for high O₃ days because (a) O₃ generation is VOC-limited or because (b) of the titration effect close to NO_x sources (NO + O₃ --> NO₂ + O₂)

556 ff.: It is important to control NO_x when chemistry is VOC-limited.

559 ff: What exactly are these different phases? Is titration meant by suppression?

571: Is there a specific reason to investigate midnight concentrations? Maybe the analysis should be limited to daylight values.

652: These sources emit both VOCs and NO_x.

656 ff.: How exactly were these factors identified? Why are there six factors? Additional explanations are needed here.

696 ff. / Fig. 6: There does not seem to be a relevant difference between the three cases. What’s the uncertainty? Are the differences even significant?

706 ff.: Please provide an explanation for speculations.

799 ff.: Sillman et al. suggested the HCHO to NO₂ ratio for O₃ sensitivity analysis.

803: What is MEM?

827: What is RIR?

944: It is not clear why the slope of the ridge could indicate the ratio at which VOCs and NO_x need to decline. Wouldn’t it be important to reduce NO_x as quickly as possible to move towards NO_x limited chemistry?

961: What is meant by “high-resolution observations” – the hourly measurements?

Reply
Citation: https://doi.org/10.5194/egusphere-2025-4519-RC2
RC3: 'Comment on egusphere-2025-4519', Anonymous Referee #3, 23 Dec 2025 reply

This work discussed concerns ozone formation within ZhengZhou, China, a mega city with considerable anthropogenic activity. The authors use observations (2019-2021), a box model, CMAQ, and machine learning tools to investigate the emissions sectors and species contributing to ozone during the period.
Unfortunately, this manuscript is rambling and does not present a clear message. “Conclusions” discussed are not traceable, and basic concepts well established in the community about ozone formation are clearly not well understood by the authors. Inadequate discussion of key factors to the analysis, along with generic findings does not lend well to this work, in its current form, being useful to the community and therefore I do not recommend this manuscript for publication in ACP.

Section comments:
Methods: The manuscript uses a variety of models, and methods, many of which are confusing for the reader to understand and follow. The authors need to shorten the observations and methods section (10 pages is excessive), and more succinctly describe the methodology used for this analysis. Make use of supplemental information for details that are not as critical. The paper should convey the main points, the big concepts of this work.
Results: Many of the findings presented in the results section are not groundbreaking, or new, and rely on machine learning techniques which not well described and therefore are questionable at best for accurately accessing relationships between ozone and precursor species. Additionally, the authors mention PM2.5, like a buzz word – the scope of this paper needs to be made smaller and more impactful. The units should be mixing ratios when discussing gaseous species, not the variety of units that are currently mentioned in the paper. What are the criteria to classify non-polluted, lightly polluted, and moderately polluted days? Discussion of the ozone diurnal cycle are not fully accurate, basic knowledge of ozone formation is clearly lacking by the authors. No clear policy relevant message is conveyed – it seems the authors do not have a great handle on the differing results from the different methods used.
Conclusions: Nothing of consequence discussed here. No policies recommended, again, nothing novel presented.

Reply

Citation: https://doi.org/10.5194/egusphere-2025-4519-RC3

Shijie Yu, Hongyu Liu, Hui Wang, Fangcheng Su, Minghao Yuan, and Ruiqin Zhang

Supplement

https://doi.org/10.5194/egusphere-2025-4519-supplement

Data sets

VOCs Dataset S. Yu et al. https://doi.org/10.5281/zenodo.17214861

Shijie Yu, Hongyu Liu, Hui Wang, Fangcheng Su, Minghao Yuan, and Ruiqin Zhang

Viewed

Total article views: 434 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
251	152	31	434	26	22	24

HTML: 251
PDF: 152
XML: 31
Total: 434
Supplement: 26
BibTeX: 22
EndNote: 24

Views and downloads (calculated since 23 Oct 2025)

Month	HTML	PDF	XML	Total
Oct 2025	116	21	9	146
Nov 2025	71	63	10	144
Dec 2025	64	68	12	144

Cumulative views and downloads (calculated since 23 Oct 2025)

Month	HTML	PDF	XML	Total
Oct 2025	116	21	9	146
Nov 2025	71	63	10	144
Dec 2025	64	68	12	144

Viewed (geographical distribution)

Total article views: 417 (including HTML, PDF, and XML) Thereof 417 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 27 Dec 2025

Short summary

As ozone (O₃) pollution remains a major concern, this study draws on long-term observational data to thoroughly analyze O₃ and its precursor substances. This study examines VOC pollution characteristics, explores local O₃ formation mechanisms, and assesses O₃ sensitivity. Our key finding is that motor vehicles are main sources of VOCs and O₃; local O₃ is more sensitive to VOCs. These guide targeted pollution cuts and protect the public environment.


Total:	0
HTML:	0
PDF:	0
XML:	0

Measurement report: Characterizing O3-NOₓ-VOC sensitivity and O3 formation in a heavily polluted central China megacity using multi-methods during 2019–2021 warm seasons

Supplement

Data sets

Viewed

Viewed (geographical distribution)

Measurement report: Characterizing O₃-NOₓ-VOC sensitivity and O₃ formation in a heavily polluted central China megacity using multi-methods during 2019–2021 warm seasons