Response to Reviewer 1

Comments:

This manuscript analyzes the characteristics of summer BVOC emissions in the Pearl River Delta (PRD) region from 2001 to 2020 and uses WRF-CMAQ simulations to reveal that climate-driven increases in BVOC emissions lead to higher ozone concentrations in the region. The entire study appears technically sound, and the results are well interpreted. Thus, I recommend the publication by addressing the comments below.

Response to Reviewer 1: Thank you very much for your valuable comments and suggestions. Your positive comments have made us more confident and energetic in revising this paper. We have carefully revised (revisions made in red color) the manuscript based on all your questions/suggestions, and hope the correction will meet with approval. Below is the point-to-point response.

Point-to-point response:

1. For Figure.1, The labels C and D in the title of Figure 1 should be swapped.

Reply: Thanks for pointing our mistakes. We have swapped their labels. Please kindly see the revised Figure. 1.

1. For lines 179-183, meteorological conditions and vegetation distribution mutually influence each other, so is it reasonable to fix only one of them while keeping the other unchanged?

Reply: Yes, we fully acknowledge the reviewer's point that meteorological conditions and vegetation distribution are closely interconnected. For example, Adequate precipitation and sunlight promote vegetation growth, while increased vegetation, in turn, alters surface albedo and subsequently influences meteorological conditions.

It is crucial to emphasize that our study employs monthly dynamic vegetation and meteorological data to drive the MEGAN model. For instance, when meteorological conditions are held constant, we use vegetation data that vary dynamically based on satellite observations (updated every 8 days). This observation is an overall result of vegetation variation, which ensures that the impact of meteorology on vegetation is inherently captured. Conversely, when vegetation data are fixed, meteorological conditions are represented using reanalysis data updated every six hours, meaning the influence of vegetation on meteorology remains accounted for. Since our study examines the climatic-scale effects of meteorological conditions and vegetation cover variation, this parallel comparison allows us to isolate their respective contributions to BVOC emissions. By fixing one while varying the other, we can effectively disentangle the influences of vegetation distribution changes and climate-driven variations, offering valuable insights into their distinct roles. Similar methods could be found in previous paper (Wang et al., 2024; Wang et al., 2025)

1. For 2.3 section, please list the evaluation metrics for the Random Forest model to verify its accuracy.

Reply : Thanks for the reminder. The RF model used reproduces 78% of the BVOC emissions variance across the full dataset, demonstrating its effectiveness in identifying relationships and interactions between BVOC emissions and meteorological factors.

Please see our revisions “To ensure the robustness of the results, we performed a 10-fold cross-validation, achieving an R² (coefficient of determination) of 0.78 and an MAE (mean absolute error) of 0.73, respectively (Fig S1). These metrics indicate that the machine learning model effectively reproduces BVOC emissions.” in the manuscript.

Fig S1 Evaluation of Random Forest model using a 10-fold cross-validation

1. For lines 405-409, the authors reveled the rise of ozone levels from climate-driven BVOC emissions and vegetation-change BVOC emissions, respectively. What about the compound effect of the two effects?

Reply: Thanks for the question. In this study, we adopted a brute force method to highlight the individual impact of climate-driven BVOC emissions and vegetation-change BVOC emissions on O3 formation. The overall effect could be roughly estimated by summing the two contributions, that is 7.0 ~ 7.7ppb. We acknowledge that this simple approach may introduce uncertainties due to the nonlinearity of ozone formation. However, this method is widely used in similar studies (Wang et al., 2024; Wang et al., 2025). Given that our focus is on climate-scale assessments, we believe it is a reasonable approach.

1. For Figure 5D, the PRD region shows a pair of symmetrical patterns with opposite effects. According to the author's description, under the influence of climate effects, areas with significant positive effects are mainly driven by heatwave days, while areas with significant negative effects are primarily due to extreme precipitation. Are there any relevant data or materials to further substantiate this claim?

Reply: Thanks for your question. We have added additional data to support this claim. Please see our revisions, “By analyzing historical monthly average temperature and precipitation in the PRD from 2000 to 2020, we found that both summer temperature and precipitation exceed the climatic monthly average, ranking the highest among the four seasons (Fig. S4). This result aligns with our description: on one hand, hot days (i.e., the periphery of typhoons, and heatwaves) enhance photochemical O₃ formation, while on the other, extreme precipitation (i.e., squall lines) suppresses O₃ concentration.” in the manuscript

Fig S4 Climatic monthly anomaly of 2m-temperature and total precipitation during 2000-2020 in PRD

Reference:

Wang H, Liu X, Wu C, et al. Regional to global distributions, trends, and drivers of biogenic volatile organic compound emission from 2001 to 2020[J]. Atmospheric Chemistry and Physics, 2024, 24(5): 3309-3328.

Wang Y, Li Q, Wang Y, et al. Increasing soil nitrous acid emissions driven by climate and fertilization change aggravate global ozone pollution[J]. Nature communications, 2025, 16(1): 1-13.

Response to Reviewer 2

Comments:

This paper focuses on the long-term changes in ozone concentration in the Pearl River Delta (PRD) region during 2001–2020, as well as the contribution of climate-driven biogenic volatile organic compounds (BVOCs) to ozone formation. By using the WRF-CMAQ model and the MEGAN model, in combination with machine learning analysis, the study reveals the key influence of BVOCs on ozone formation potential. Climate-driven BVOC emissions enhance the atmospheric oxidation capacity and accelerate ozone formation, thereby weakening or even offsetting the effects of anthropogenic emission reductions. These emissions contribute 6.2 ppb to ozone production, leading to an unexpected rise in ozone levels. This research deepens the understanding the complex interactions between natural source emissions and anthropogenic control strategies, and it provides practical reference value for the formulation of regional air pollution control policies. However, some clarifications and changes are necessary. 

Response to Reviewer 2: We sincerely appreciate your time and effort in reviewing this manuscript. Your insightful comments and suggestions have greatly helped us improve the quality of our work. We have carefully revised the manuscript with revisions made in red color. We believe this revised version is better organized, and our point-by-point responses and revisions are detailed below.

Point-to-point response

1. The study utilizes a random forest (RF) model to analyze the primary driving factors of BVOC emissions and employs SHAP values to explain the contributions of various meteorological factors. However, the validation process of the RF model is insufficiently detailed, with no mention of cross-validation methods. It is recommended that the authors supplement the model validation.

Reply: Thanks for the suggestion. We have added a 10-fold cross-validation in this manuscript. Please see our revision “To ensure the robustness of the results, we performed a 10-fold cross-validation, achieving an R² (coefficient of determination) of 0.78 and an MAE (mean absolute error) of 0.73 (Fig S1). These metrics indicate that the machine learning model effectively reproduces BVOC emissions.” in the manuscript.

Fig S1 Evaluation of Random Forest model using a 10-fold cross-validation

1. Figure 4 presents the spatial distribution of sensitivity coefficients for ozone formation with respect to BVOCs. It is suggested that the authors include a percentage difference plot to more intuitively display the influence of BVOC emissions on the spatial distribution of sensitivity coefficients. Additionally, incorporating comparative charts for different years would better illustrate the long-term trends.

Reply: Thanks for the suggestions. Please note that the O₃ sensitivity coefficient could be zero in some grids in both the AVOC_ONLY and Add_BVOC scenarios, making percentage calculations undefined in these cases, which means it is impossible to provide a percentage difference plot. However, as suggested by the reviewer, we have added a difference plot between the AVOC_ONLY scenario and the Add_BVOC scenario. Hopefully, the revised Figure 4 could provide a more intuitive representation of the influence of BVOC emissions on the spatial distribution of sensitivity coefficients.

Figure 4. (A) Spatial distribution of O₃ sensitivity coefficients to NO_x emissions under AVOC_ONLY and Add_BVOC scenario. (B) Difference of O₃ sensitivity coefficients to NO_x emissions between Add_BVOC and AVOC_ONLY scenario. (C) Same as (A) but for sensitivity coefficients to VOCs emissions. (D) Same as (B) but for O₃ sensitivity coefficients to VOCs emissions (E) difference of production rate of NO₂ (via chemical pathway of RO₂+NO and HO₂+NO) and net production rate of O₃ at 14:00 between Add_BVOC and AVOC_ONLY scenario

We agree that incorporating comparative charts for different years would better illustrate long-term trends. However, due to the non-linear nature of these responses, the calculations are inherently complex. As you may know, HDDM is a highly computationally intensive model, as it computes O₃ sensitivity coefficients—including first-order, second-order, and higher-order terms—for each precursor (e.g., NO, NO₂, and various VOC species). Running long-term simulations would require extensive computational resources, exceeding the capacity of available supercomputing infrastructure. Moreover, our primary objective is to highlight the significant impact of BVOC emissions on O₃ concentrations. To achieve this, we designed two scenarios: AVOC_ONLY (considers only anthropogenic VOCs), and ADD_BVOC (accounts for both anthropogenic and biogenic emissions). Both simulations were driven by 2020 meteorological conditions. While comparative charts across multiple years would provide deeper insight into long-term trends, the conclusion regarding the importance of BVOCs emissions on O₃ formation holds true.

1. The simulation only compares the years 2012 and 2020. This design may not fully capture the dynamic changes in ozone and BVOC emissions from 2001 to 2020, especially since China implemented several significant emission reduction policies between 2012 and 2020, resulting in notable nonlinear changes in ozone concentrations and precursor emissions. Studies, for example, doi.org/10.1038/s41561-023-01284-2, have suggested that the anthropogenic emission reduction in different phases actually led to different impacts on ozone.  Did you find the similar results? It will be better to include some intermediate years to illustrate the effects.

Reply: Thanks for the insightful questions. Indeed, previous studies—including the paper mentioned by the reviewer and our own publications—have already demonstrated that anthropogenic emission reductions at different phases have led to varying impacts on O₃. For instance, our earlier study found that emission control measures during 2012–2017 (Phase One) resulted in increased O₃ levels across most eastern city clusters of China due to NOₓ reductions in a VOC-limited regime. In contrast, the reference cited by the reviewer showed that during 2018–2021 (Phase Two), further emission reductions helped mitigate O₃ pollution. These contrasting effects largely stem from shifts in O₃-NOₓ-VOCs sensitivity, which have been widely discussed (Wang et al., 2019; Huang et al., 2021; Wang et al., 2023). Our findings align with this pattern, suggesting that nearly a decade of anthropogenic emission control has contributed to a decline in summer O₃ concentrations in the PRD. We did not evaluate year-to-year O₃ responses to emission reductions, as this topic has been extensively studied. To avoid redundancy, our assessment adopts a climatic-scale perspective, examining the anthropogenic controlling influences between 2012 (when anthropogenic emissions reached the peak during the past decade) and the present (2020, almost after a decade of control measures).

Nonetheless, we have added a discussion on O₃ responses to anthropogenic emissions in the manuscript, citing these published studies, “These contrasting effects largely stem from shifts in O₃-NOₓ-VOCs sensitivity. Past studies suggested that O₃ levels would temporarily increase in the short term following NOₓ emission controls (Wang et al., 2019; Huang et al., 2021). However, after a long-term (nearly a decade) emission reductions, our finding reveals that, when considering only anthropogenic emissions (AVOC_ONLY scenario), emission reductions could lead to varying degrees of O₃ decline in southern China. This result was consistent with a recent study by Wang et al. (2023)”

Moreover, as our primary focus is on biogenic emissions, evaluating the year-by-year impact of anthropogenic emission reductions does not fully align with our research objectives. According to our study, we found BVOCs emissions were kept rising during the past two decades, therefore, we compared their contribution between 2001 and 2020 in order to maxima the contribution of BVOCs to O3 (Please be noted that we didn’t only compare the year of 2012 and 2020). Actually, the O₃ formation algorithm we proposed could maximally account for the influences of anthropogenic and biogenic sources to highlight their respective contributions in a climatic scale. (see our revision in lines 243-244)

1. The authors employed WRF 3.9 and CMAQ 5.3 for the simulations. However, the MCIP module in CMAQ versions 5.2+ does not directly support handling data outputs from WRF versions below 4.0, which may result in data incompatibility or preprocessing errors. Was the MCIP module modified, or was an intermediate data conversion tool used? It should be clearly stated in the manuscript.

Reply: Thanks for the very technical question. Yes, the MCIP module in CMAQ 5.3 does not directly support data outputs from WRF versions below 4.0. To resolve this issue, we used an earlier version of MCIP (v5.2) to process meteorological data from WRF 3.9. Notably, CMAQ 5.3 remains compatible with MCIP outputs from lower versions (like MCIP v5.2).

Reference

Huang X, Ding A, Gao J, et al. Enhanced secondary pollution offset reduction of primary emissions during COVID-19 lockdown in China[J]. National Science Review, 2021, 8(2): nwaa137.

Wang Y, Zhao Y, Liu Y, et al. Sustained emission reductions have restrained the ozone pollution over China[J]. Nature Geoscience, 2023, 16(11): 967-974.

Wang N, Lyu X, Deng X, et al. Aggravating O3 pollution due to NOx emission control in eastern China[J]. Science of the Total Environment, 2019, 677: 732-744.