This manuscript proposes a three-stage modeling framework to estimate the current and future global distributions of wildfire-sourced ozone under various SSP scenarios. By combining GEOS-Chem simulations with XGBoost, the authors (1) calibrate modeled ozone using historical observations, (2) project future ozone using CMIP6 data, and (3) estimate fire-sourced ozone through a ratio-based approach. The methodology is rigorous and the study addresses an important but underexplored topic, namely the future concentrations of wildfire-induced ozone and their impacts on crop yields. However, several methodological assumptions and data consistency issues require further clarification or revision.

Major Comments

• Data Consistency Between Model Stages

While the authors avoid using GEOS-Chem for future total ozone due to uncertainties in emission inventories, they still apply the fire-to-total ozone ratio derived from GEOS-Chem to CMIP6-based projections. This raises two issues: (1) the manuscript does not evaluate whether this ratio is stable across different models or emission scenarios, and whether it can be reliably transferred from GEOS-Chem outputs to CMIP6-driven ozone fields; and (2) it does not explain why GEOS-Chem total ozone is considered unreliable, while the fire contribution ratio, derived from the same emission inputs, is assumed to be trustworthy.

• Estimation of Fire-Induced Ozone in GEOS-Chem

The manuscript states that fire-related ozone was estimated by comparing GEOS-Chem simulations with and without fire emissions. However, key details of this implementation are missing. The authors should provide more information on how the simulations were configured in order to ensure transparency and reproducibility.

• Attribution without internal evidence

The manuscript attributes regional ozone differences to known processes by citing external literature. For example, it attributes high wildfire-induced ozone in the SS region to higher fuel consumption and burned area, yet presents no corresponding model outputs. Similarly, the discussion of low ozone enhancement over the US refers to temperature and NOₓ concentrations but does not show any temperature or NOₓ fields from the model. The authors are encouraged to support such interpretations using variables directly derived from their simulation.

Minor Comments

• Figures 1 and 2 show wildfire-induced MDA8 O₃ for the 2010s and 2040s, but the temporal averaging method is unclear. Please specify whether values represent multi-year means, seasonal averages, or other metrics.
• The AOT40 formula is provided, but key details are missing, such as the daily 8-hour window used, how growing seasons were defined, and whether values were averaged annually or over multiple years.
• Line 77: The text refers to Figure S1 as showing the monitoring site distribution, but it only presents the modeling workflow.

Summary

This study aims to explore the future impact that fires will have on ozone, and how that fire-sourced ozone will change future crop yields across different emissions scenarios. This is an important and scientifically relevant question, particularly as wildfires have been shown to increase during the hotter and drier fire seasons brought about by climate change. The authors use a “multi-stage model” to answer this question, combining output from GEOS-Chem and CMIP models with reanalysis data and ground (MDA8 O₃) observations. I appreciate the effort it took to combine disparate data sources and attempt to extract meaningful results. The topic of the study is increasingly relevant, but some of the methods need clarification, and a few conclusions require additional evidence.

Major Comments

1. Estimation of fire-induced ozone in GEOS-Chem (bumping this from initial evaluation)

The GEOS-Chem chemical transport model (CTM) is used to simulate total ozone concentrations and wildfire-induced ozone concentrations. It is unclear to me exactly how the wildfire-induced ozone concentrations are calculated. Is this done by subtracting the output of simulations which include wildfire emissions from those which don’t? If so, this introduces some uncertainty to the issue: ozone chemistry is highly nonlinear and hence regionally specific. Shutting off wildfire emissions entirely will change the chemical environment (i.e. global oxidative potential) and makes a direct comparison to the base case more complex. In your response, you mention the “air pollutant tracing method” as another option, but it is unclear how this would be implemented. This discussion would benefit from another sentence for clarification or an illustrative reference.

Since the model is not evaluated against observations, also be clear that this conclusion is based off the results of a single model – using different CTMs could potentially lead to different conclusions. What are the implications if the surface ozone background is high in GEOS-Chem, as reported over the U.S. in Guo et al 2018?

1. Fire attribution with K+

I appreciate the authors efforts to address my earlier comments, and I am sympathetic to the challenge of limited data. However, I am still skeptical of the fit shown in Fig. S2b. Would it be possible to include levoglucosan measurements, like you mentioned in your response, from the same observational sites for validation of the fire influence? Are there anthropogenic tracers that are also observed from those sites, so that the possibility anthropogenic K+ could be counted out?

One thing that adds to my skepticism is that the predicted fire O₃ enhancements seem very small (1-2 ug/m3) in Figure S2b compared to the complete distribution shown in Figure S2a. Even if the potassium is fire-derived, do the authors feel that these enhancements are convincing enough to say that fires are contributing excess ozone?

1. XGBoost model training and evaluation

It is unclear what predictors are used. Li et al. 2024b, which is mentioned in L121 seems to incorporate satellite data, which doesn’t align with this study.

In the author’s evaluation of their model, they cite the R² but do not comment on the slope of predicted vs observed, which is 0.67 (Figure S2a). Does this indicate an underestimate in predicted ozone of about 33%?

1. Questionable attribution

The link to temperature is mentioned multiple times but there is not adequate evidence that this plays a role in governing wildfire-ozone production. Do the temperature fields reflect the conclusions that the study presents, such as in L224?

Minor Comments

L54 MDA8our typo

L56: Provide more details on this method, and be clear that this is a modeling study -- I believe Xu 2023 also uses GEOS-Chem.

L58: I wouldn't say the focus is historical, but instead that observational studies have sought to understand the exact impact of wildfires on ozone production, while modeling studies have sought to evaluate the performance of models against observations. There is a lot of uncertainty here that should be mentioned.

L71: What sort of control measures? To comment on this would need to know how ozone formation regime is changing and what anthropogenic emissions could be controlled to move in the right direction.

L90: It would be nice to see the spatial distribution of these observations.

L167: Does the crop growing season differ by region? How is it defined across the world?

L182: Would be interesting to see how much variance there is between different crop yield models, and how that impacts the spread of CPL estimated with your model.

L221: U.S. showed higher PM than the other regions discussed above? Or higher PM than reported in previous studies? Language is a bit unclear.

L224: “The lower air temperature…” This is an unsubstantiated hypothesis. The latitudinal distribution is driven by many factors and most of them are likely fire-related, not meteorological. This is even mentioned in the next sentence.