the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 12 Oct 2026
Hourly-Resolved temperature inflection points of ozone formation: Chemical pathways and precursor sensitivity in Zhengzhou, China
Abstract. Ground-level ozone (O3) is a key secondary pollutant. Its formation is nonlinearly modulated by temperature, an effect amplified under climate warming. However, the chemical mechanisms by which temperature influences the diurnal O3 cycle and its synergistic interactions with other parameters remain inadequately understood. By integrating observational data from a typical inland city in China with model simulations and machine learning, this study identifies the temperature inflection point for O3 formation and elucidates the underlying chemical mechanisms and dominant controlling factors. The net O3 production rate (Pnet(O3)) peaks at 31 °C. The rapid formation between 25 and 31 °C is driven by the HO2+NO and RO2+NO pathways, with enhanced nitrogen dioxide photolysis rate (J(NO2)) and elevated isoprene as the main drivers. From 31 to 36 °C, Pnet(O3) declines because the HO2+NO pathway is weakened and decreasing photolysis rates become the dominant limiting factor. Above 36 °C, nitrous acid (HONO) and formaldehyde (HCHO) concentrations decrease, weakening radical cycling and further suppressing O3 formation. Sensitivity and two-factor analyses support a temperature-dependent dynamic control strategy: in the low-temperature range (22–31 °C), priority should be given to controlling anthropogenic alkenes with nitrogen oxides (NOx) reductions to avoid diminishing returns from single-species cuts; in the high-temperature range (32–40 °C), NOx reductions should be intensified, and synergistic precursors such as HONO and oxygenated volatile organic compounds (OVOCs) should be reduced with emphasis, enhancing the O3 reduction benefit per unit emission and mitigating O3 pollution risks under a warming climate.
- Preprint
(1606 KB) - Metadata XML
-
Supplement
(2019 KB) - BibTeX
- EndNote
Status: open (until 29 Nov 2026)
- RC1: 'Comment on egusphere-2026-5563', Anonymous Referee #1, 26 Oct 2026 reply
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 292 | 122 | 31 | 445 | 60 | 35 | 35 |
- HTML: 292
- PDF: 122
- XML: 31
- Total: 445
- Supplement: 60
- BibTeX: 35
- EndNote: 35
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General comments
This manuscript investigates the nonlinear relationship between temperature and surface ozone in Zhengzhou, China, using hourly observations from June 2015-2024, an intensive June 2024 measurement campaign, an observation-based chemical box model (OBM), and machine learning. The topic is well aligned with Atmospheric Chemistry and Physics. Combining long-term observations with chemical modeling is a strength, and the effort to interpret ozone-temperature relationships at the hourly scale is potentially valuable.
In my view, the main contribution lies in linking the observed nonlinear relationship to modeled changes in ozone production pathways and precursor sensitivity. The modeled results suggest that radical chemistry, photolysis, and precursor availability play different roles across temperature regimes. However, nonlinear ozone-temperature relationships and temperature-dependent precursor sensitivity have been examined previously. The specific advance here is therefore more likely to lie in the proposed hourly-scale mechanistic interpretation than in the identification of temperature thresholds alone.
Several substantial issues need to be addressed before the mechanistic and quantitative conclusions can be evaluated reliably. The proposed thresholds are not yet supported by a sufficiently clear statistical framework, and temperature is strongly correlated with other changes in the diurnal photochemical environment. The interpretation of declining net ozone production above 31°C also needs to be reconciled with the reported gross production pathways and chemical losses. In addition, important aspects of the observational, machine learning, OBM, and sensitivity methods require clearer documentation and evaluation. These concerns affect the reliability of the mechanistic interpretation more than the presentation, which is generally clear and well organized.
I therefore recommend major revision. In particular, I encourage the authors to establish the reliability and meaning of the temperature thresholds, distinguish direct temperature effects from correlated diurnal changes, reconcile the chemical budget, and clarify the modeling and sensitivity frameworks. The novelty, generalizability, and emission-control implications should also be framed in line with the evidence. Addressing these issues would strengthen the study's scientific quality while preserving its potential contribution to atmospheric ozone chemistry.
Major comments
1. Scientific contribution and identification of temperature thresholds (Sections 1 and 3.1; Figures 1 and 2; Figure S4)
The characteristic temperatures of 25, 31, and 36°C are central to the manuscript, but the statistical support for these thresholds and the additional insights they provide beyond previous studies need clearer justification. Earlier work has documented nonlinear ozone-temperature responses, temperature-dependent chemistry, and changes in VOC-NOx sensitivity. The authors should specify what the hourly-scale analysis establishes beyond these findings.
The three temperatures appear to be inferred from changes in ozone concentration, its rate of change, and modeled net ozone production. These are distinct quantities: the onset of faster ozone accumulation, a production maximum, and a weaker concentration response do not necessarily represent equivalent thresholds. In particular, Figure 1 shows ozone concentrations rising toward a plateau, whereas Figure 2 shows modeled net production peaking near 31°C. This distinction should be maintained throughout.
The statistical robustness of these thresholds also remains uncertain. Modeled net production is 15.05 ppb h-1 at 31°C and 13.80 ppb h-1 at 36°C, but Figure 2 shows substantial variability. Given the reported variability (15.05 ± 7.09 and 13.80 ± 6.12 ppb h-1, respectively), it is unclear whether the maximum can be reliably assigned to 31°C. The authors should explain the criteria used to identify each threshold, report sample sizes and associated uncertainties (for example, using resampling methods that account for temporal dependence), and assess the stability of these thresholds across the individual June periods from 2015 to 2024.
2. Direct temperature effects versus correlated diurnal changes (Sections 3.1 and 3.2; Figures 1-4; Figures S3 and S6)
A major interpretive issue is the covariance of temperature with the diurnal photochemical environment. The lower-temperature, approximately 31°C, and higher-temperature groups largely correspond to morning, late morning/noon, and afternoon conditions. Photolysis frequencies, precursor concentrations, and atmospheric mixing change over the same periods. Differences between these groups therefore cannot be attributed to temperature alone; analyses within the same hours of the day or comparisons under similar photochemical and meteorological conditions would help assess the influence of these concurrent changes.
This matters particularly for the decline in net ozone production above 31°C. The decrease in photolysis frequencies between 31 and 36°C may reflect post-noon solar conditions rather than direct thermal inhibition. Similarly, changes in HONO, HCHO, and NOx may reflect their diurnal chemical budgets. Figure 4 is informative in this respect: the modeled direct temperature contribution remains positive over several intervals, even as net production falls because other factors change. This distinction deserves greater emphasis.
Transport and mixing should also be discussed more cautiously. Increasing wind speed and boundary layer height in Figure S6 do not establish that observed high ozone is mainly produced locally. Vertical entrainment and other transport processes may contribute but are not explicitly resolved by the zero-dimensional OBM. Overall, the manuscript should distinguish temperature-driven chemical effects from temperature-associated changes in atmospheric conditions before presenting the three temperatures as chemical transition points.
3. Observational and modeling methods, evaluation, and uncertainty (Section 2; Figures S1, S2, and S5; Tables S1 and S3)
The relationship between the 2015-2024 observations and the intensive June 2024 measurements requires clarification. The long-term dataset establishes the statistical ozone-temperature relationship, whereas the one-month campaign provides the basis for the mechanistic interpretation. The authors should discuss how representative June 2024 is of the broader period and document data completeness, quality control, temporal matching, and the sampling of each temperature group.
The machine learning evaluation also raises an important concern. Section 2.3 lists ozone among the inputs used to predict ozone. The authors should clarify whether this refers to ozone measured at the same time (which would constitute target leakage), ozone measured at earlier time points, or a different prediction target. The stated 80:20 training-validation split also requires clarification. If observations were assigned randomly, temporal autocorrelation may inflate the apparent validation performance. A temporally separated validation or other independence check is needed. SHAP values describe contributions to model predictions rather than independent causal effects. Therefore, the high importance of temperature in Figure S5 and the zero crossing near 31°C in Figure 2b should not be presented as proof of a temperature-driven chemical threshold.
The OBM configuration, chemical constraints, and budget calculations also need more detailed documentation. Section 2.4 lists observed ozone as an OBM constraint, while Figure S2 uses observed ozone to assess model performance. It is therefore unclear whether this evaluation provides an independent assessment of model performance. The authors should explain this treatment, define the production and loss terms used to calculate net ozone production, and report the relevant model settings and assumptions. Agreement in ozone concentration alone does not validate the modeled radicals or reaction pathways. Uncertainty in these model outputs should also be considered. These questions do not necessarily invalidate the OBM results, but they are important for assessing the proposed mechanisms.
4. Consistency of the ozone budget and radical pathway interpretation (Section 3.2.1; Figure 3; Table S2)
The chemical budget is central to the manuscript, but its interpretation of declining net production above 31°C is not fully consistent with the reported reaction rates. Between 31 and 36°C, HO2 + NO decreases from 12.74 to 11.09 ppb h-1 (-1.65 ppb h-1), whereas RO2 + NO increases from 7.22 to 8.96 ppb h-1 (+1.74 ppb h-1). Their combined rate therefore rises slightly from 19.96 to 20.05 ppb h-1, while Pnet(O3) declines from 15.05 to 13.80 ppb h-1.
If the two reaction-rate sums are directly comparable to gross ozone production, the residual between their sum and Pnet(O3) increases by approximately 1.34 ppb h-1. From 36 to 40°C, the sum rises further (20.05 to 21.92 ppb h-1), while Pnet(O3) decreases slightly (13.80 to 13.50 ppb h-1). The residual may reflect chemical losses or differences in how the production terms are defined. The authors should explicitly reconcile gross production, relevant chemical losses, and net production rather than attributing the decline mainly to weaker HO2 + NO chemistry.
The mechanistic discussion should also distinguish reaction rates from radical abundance. A lower HO2 + NO rate does not by itself demonstrate thermal HO2 depletion, since it also depends on NO and competing reactions. The proposed competition between RO2 + NO and intramolecular reactions needs more direct support. Similarly, declining HONO and HCHO concentrations do not establish enhanced radical termination, reservoir formation, or increased PAN formation without corresponding budget or pathway evidence. A consistent accounting of production, loss, and radical cycling is important for the central mechanistic claim.
5. Attribution of individual factors and interpretation of precursor changes (Section 3.2.2; Figure 4; Figures S7 and S8; Tables S3 and S4)
The waterfall analysis in Figure 4 could provide a useful decomposition of changes in net ozone production, but the attribution method is not sufficiently clear. In a nonlinear chemical system, single-factor responses depend on the reference state and can interact with other variables. The manuscript should specify the baseline conditions, perturbation sequence, treatment of interactions, and how the assigned contributions reconcile with the total change in Pnet(O3). The reliability of the estimates is particularly relevant where the total net-production differences are small.
The interpretation of precursor changes should distinguish observed concentrations from the processes controlling them. Figure S8 supports an association between isoprene concentration and temperature, but does not independently establish temperature-enhanced biogenic emission fluxes; chemical loss, mixing, and transport also affect concentrations. Likewise, decreases in HONO and HCHO should not be attributed solely to photolytic depletion without examining other production and removal processes.
The single-factor curves in Figure S7 indicate modeled sensitivities under specified conditions, not necessarily quantitative contributions to observed atmospheric changes. This limitation should be stated when interpreting the relative importance of temperature, photolysis, and precursor changes.
6. Precursor sensitivity regimes and emission-control implications (Section 3.3; Figure 5; Figure S9)
The EKMA and relative incremental reactivity (RIR) analyses are potentially useful, but their calculation and interpretation need greater transparency. The manuscript should define the RIR perturbations and ozone response metric, explain how the EKMA surface was constructed, and clarify how MDA8 O3 was derived from the OBM.
Figure 5 places characteristic temperature conditions on a common AVOC-NOx response surface. This illustrates changes in the chemical environment but does not isolate temperature as the cause of a transition from VOC-limited to NOx-limited conditions. Changes in precursor concentrations, photolysis, and other factors could contribute. Figure S9 suggests increasing NOx sensitivity at higher temperatures while alkenes remain important, but uncertainty in the RIR estimates and the boundaries between regimes should be assessed.
The emission-control implications also need qualification. Sensitivity to ambient precursor concentrations is not equivalent to sensitivity to emission reductions in a nonlinear atmosphere. In particular, the approximately 5-12 ppb NOx range associated with relatively high modeled ozone production (Figure S11) does not establish a desirable ambient NOx level or an emission-reduction target. I suggest presenting the results as temperature-associated changes in modeled precursor sensitivity, with control strategies treated as conditional implications rather than established quantitative recommendations.
7. Two-factor sensitivity and temperature interactions (Section 3.4; Figure 6; Figures S10 and S11)
The two-factor sensitivity analysis is a useful extension, but the interpretation of "synergistic" and "antagonistic" effects requires a more rigorous quantitative basis. Although the authors compare ozone responses across different temperatures and factor levels, the analysis should more clearly distinguish the effects of individual factors from their interactions. In particular, a weaker ozone response to increasing precursor concentrations at higher temperatures does not necessarily mean that the precursor directly suppresses ozone formation. For example, the reported ozone percentage changes associated with increasing alkene concentrations remain positive at both 293 and 314 K (23.86% and 22.29%, respectively). This indicates a weakening of the positive response rather than a direct ozone-suppressing effect. The authors should clarify the criteria used to identify synergistic and antagonistic interactions and explain how these are distinguished from the individual effects of the factors.
The authors should provide the equations and reference values used to calculate the percentage changes and interaction effects in Figure S10. They should also explain how these effects can be compared across panels with different parameter ranges and axis scales. In addition, the authors should discuss whether the simulated combinations of temperature and other factors are representative of conditions observed during the study period. The proposed alkane temperature threshold and alkene concentration threshold may depend on the selected parameter ranges and model assumptions. The robustness and uncertainty of these thresholds should therefore be discussed.
Finally, the reported positive interactions involving HONO and OVOCs do not necessarily demonstrate the effectiveness of emission reduction measures targeting their sources. Both species can be influenced by direct emissions and secondary atmospheric formation. The authors should therefore distinguish modeled chemical sensitivity from the expected effectiveness of practical emission control strategies.
8. Data and software availability and reproducibility
The Zenodo observation file linked in the manuscript is useful but is insufficient on its own to reproduce the main findings. "Observation data.xlsx" contains 720 hourly records covering June 2024, including meteorological variables, pollutants, photolysis parameters, and speciated VOCs. It does not contain the June 2015-2023 observations needed to reproduce the 2015-2024 statistical analysis.
The spreadsheet has 67 individual-compound columns, compared with the 115 species said to have been monitored in Section 2.2, and it does not have an identifiable NO concentration column despite NO being an OBM constraint. Several compound headers are ambiguous. Moreover, 35 hourly records are coded as 1206 across several photolysis-frequency columns, without a documented missing/invalid-value convention, and J(O1D) includes negative entries. The authors should document these codes and their treatment, the available species and missing constraints, the unit conversions (some archived gases are in mass-concentration units whereas model inputs are reported in ppb), and the derivation of processed inputs and representative temperature conditions.
The manuscript, supplement, and supplied observation file do not provide the scripts, model configuration, or derived products needed to reproduce the XGBoost, OBM, EKMA, RIR, or sensitivity analyses. If these materials or additional datasets are deposited separately, the authors should identify their locations. The Data Availability statement should make clear which long-term observations, processed model inputs, and analysis resources are openly accessible and how the remaining results can be reproduced.
9. Broader applicability, previous work, and conclusions (Sections 1 and 4; Figure S12)
The broader implications should be distinguished from findings established at Zhengzhou. Figure S12 shows similar statistical ozone-temperature patterns in Beijing, Shijiazhuang, Tianjin, and Taiyuan, but does not demonstrate equivalent maxima in chemical production, radical pathway transitions, or precursor sensitivity at the same temperatures. The three characteristic temperatures should therefore not be presented as generally applicable chemical thresholds without further evidence.
The manuscript cites relevant previous research, but the Discussion should identify more explicitly which mechanisms were already established and what the present hourly analysis adds. This is particularly important because the study focuses on one urban site, with most chemical diagnostics derived from a single month of measurements.
The conclusions should distinguish well-supported findings from model-dependent interpretations and conditional policy implications. In particular, the finding that the direct modeled temperature effect can remain positive while net production decreases because of concurrent environmental changes may offer a broader contribution than the exact numerical thresholds. A concise synthesis of this distinction, the main uncertainties, and the study's scope would strengthen the final message.
Minor and technical comments
10. Title, abstract, and terminology
The title phrase "temperature inflection points of ozone formation" implies more precisely established chemical thresholds than the evidence currently supports. The title and abstract should distinguish observed ozone concentration responses from modeled net chemical production and avoid overstating emission-control implications. The abstract should also clearly distinguish the long-term observations from the June 2024 modeling period.
The manuscript is generally understandable and well organized. However, ozone concentration, accumulation, formation, gross production, and net production should not be used interchangeably. Likewise, temperature-associated changes should be distinguished from direct temperature effects. The paragraph on positive interactions involving HONO, aromatics, halocarbons, and OVOCs is duplicated almost verbatim in Section 3.4 (lines 473-488) and should be removed once. A final check of terminology and grammar would improve precision.
11. Equations, numerical consistency, and units (Equation 1; Figure 1; Tables S1-S4)
Several technical points should be checked. Equation (1) appears to omit the squared term in the denominator of the conventional index of agreement (IOA), making the printed expression dimensionally inconsistent. The authors should confirm which equation was used and clarify the definitions of the observed and simulated variables. The calculation and units of dO3/dT also need clarification: the reported units include an inverse-time component, which is inconsistent with a derivative of concentration with respect to temperature. The distinction between dO3/dT and dO3/dt should be maintained.
In Table S1, "detection limit" may be used where instrument accuracy or measurement uncertainty is intended for meteorological parameters, including temperature. Section 3.2.1 describes ISOPBO2 + NO as both increasing and decreasing from 31 to 36°C. Table S2 shows an increase in ISOPBO2 + NO and a decrease in ISOPDO2 + NO.
The maximum OVOC concentration is 7.43 ppb in Table S3 but 8.09 ppb at 31°C in Table S4; the archived OVOC column reaches 36.72 ppb. In addition, archived pressure spans 980.6-1002.3 hPa, whereas Table S3 gives 995.7-1016.8 hPa. These may reflect different subsets, preprocessing, instrument corrections, or variable definitions, but a reconciliation is needed. RIR notation and units should also be consistent where decimal values appear with percentage labels.
12. Figures, Supporting Information, and references
The overall presentation is clear, and the supplementary analyses provide useful detail. However, methods essential to the main conclusions—particularly the sensitivity decomposition, EKMA/RIR analysis, and interaction calculations—should be sufficiently defined in the main text. Figure captions should specify the relevant response metrics, model conditions, and, for validation figures, which observations were used as constraints versus independent evaluation data. Axis and color-scale differences should be noted where panels are compared.
Some main-text and supplementary figures repeat similar ozone-temperature and sensitivity results. Reducing this overlap would make the presentation more concise without removing evidence needed to support the conclusions.
The references are generally appropriate, but the Discussion should make clearer how the present mechanism-based claims extend prior studies of nonlinear ozone-temperature responses and precursor sensitivity.