The manuscript by Zhang and colleagues models the change in ozone and PM2.5 and related health impacts in China and downwind countries due to emission changes in China after the implementation of APPCAP. Overall, most of the results presented in this paper on decrease in PM2.5 in China post APPCAP is not new and has been established in multiple studies involving models/measurements. The new aspect is assessing its impact in details on PM2.5 and ozone exposure and related mortality burden in downwind regions, though the TF-HTAP initiative partly addresses this.

I list a few major issues which the authors may want to address-

• It might be interesting to check if there are any seasonal differences in change in PM2.5 and ozone in China and also in downwind regions after the implementation of APPCAP.
• China has a distinct east to west gradient in air pollution exposure. Rather than speculating this east to west gradient in changes in PM2.5 after the APPCAP, the authors may consider representing this information (on changes in population weighted PM2.5, ozone and averted premature death) by provinces in China, which might be more policy relevant. The authors may also plan to inspect changes in emissions in which provinces led to maximum benefit in terms of averted death in a downwind country, if feasible. They might also plan to estimate the averted mortality/exposure in a downwind country by province. Eg- In USA the maximum benefit is expected to be realized in the western states of CA, OR and WA.
• The authors may want to build few relevant emission scenarios and estimate their impact on PM2.5/O3 exposure in China and in downwind countries (eg. APPCAP is twice as effective in curbing emissions). This might inform the decision makers about the benefits of further curbing emissions in China.

Minor comments

• Line 190- please reconstruct the sentence
• Line 198- please add the changes due to emission changes in the corresponding years

The work by Zhang et al. examines health impacts via air quality changes stemming from emissions changes in China from 2010-2017, expanding the role of PM2.5 and O3 and estimating the domestic vs international impacts.  Overall the study is well posed. While other studies have examined this question specifically in China, here the authors focus on global-scale analysis, although in the end their findings support that a China-focused study would be sufficient, as >90% of the health impacts occur domestically.  That aside, it’s still likely sufficiently novel and interesting to ultimately warrant publication, however the paper itself needs some additional work in a few areas.  These are described in detail in the comments below. 

Major comments: 

Section 2.1: Please provide more details on which species are included in this model’s estimate of PM2.5.  List primary and secondary species, both inorganic and organic.  Describe how PM2.5 itself is defined / calculated, i.e. is H2O included, at what RH, and at what temperature and pressure are all values calculated.  Also, what is the height of the top of the first model layer? Are O3 concentrations adjusted from this height to the surface-level (typically 2m)? 

Somewhere this paper needs to address the significant issue of estimating PM2.5 health impacts at such coarse spatial resolutions.  Several previous studies, which are easy to find, have shown that biases can be up to  20-40% in estimates at these scales in global models. 

A recent study (Nault et al., ACP, 2021) showed that ~15% of PM2.5 associated deaths may come from anthropogenically influenced SOA.  This component of PM2.5 would be highly sensitive to the emissions impacted by APPCAP.  Was this accounted for in the simulated changes in total PM2.5? 

Other studies have shown that ammonium nitrate is a significant portion of PM2.5 in this region. By not including nitrate, the simulated response of PM2.5 to emissions will be rather muted.  Can the authors estimate the magnitude of the uncertainty associated with this omission?  I see they recognize this omission and others (lines 166-169), but it would be nice to see such approximations taken into account more quantitatively as prat of their final results. 

The abstract (and elsewhere) present premature mortality estimates with CI levels. These, I suspect, only reflect the uncertainty in the IERs.  Given the substantial model errors and biases of up to 50%, how do these uncertainties compare to the uncertainties associated with model error? It would have been nice to see an attempt at incorporating the results of the model evaluation (section 3.1) into the subsequent analysis, rather than just touching on it in passing.  They do tough on this on lines 269-275, but the text here is confusing.  Why would the impact of bias be mitigated by high concentrations in China? Also, they point out here that the IER functions are non-linear.  This is a reason why biases in the simulated PM2.5 would make a difference, rather than be negligible, not this other way around. 

Table 2 and 3 are useful. It would also be nice to see maps of the station measurements overlaid on top of model estimated surfaces.  If biases / errors are particularly large in regions most impacting export (NE), that would be useful to know, given that the main stated novelty of this work (line 87) is examining the impact of these changes on global air quality, not just domestically.  Also it would be good to evaluate the biases by season, since export is stronger in the spring. 

The discussion is a bit hard to parse — it seems like the 2nd paragraph contains a lot of ideas, some which aren’t well explained, and all of it mashed together in one final push of text that mixes sources of uncertainty with explanation of results and highlights of their findings.  I’d recommend spending more time on this section. 

According to Fig 6, the PM2.5 concentration impacts are significantly larger in South Korea than Japan and the US by 2017.  However, in Table 4, for 2017 the mortality impacts are much bigger for Japan, and similar for South Korea and the US.  Please check to make sure there wasn’t  a mixup.  If not, please explain how this is the case (possibly given different mortality rates and populations). I see they note this point on line 282 but offer zero explanation. 

Writing: the paper is in pretty rough shape.  I doubt that many of the co-authors have contributed in detail or paid much attention to the final draft, given the prevalence of basic issues.  I started making a note of corrections but tired of this after the first ten lines.  Please proof-read and polish the writing throughout, it was quite distracting and at times confusing.  For example:

19: of —> of the 

20: As a statistical aspect, it seems like human health benefits are hard to observe. Maybe rephrase?

20: PM2.5 and surface —> PM2.5, surface 

21: enough lifetime —> lifetimes

21: which can —> to

21: So emission —> Emission

22: will —> will thus 

22: region air quality domestically —> domestic air quality 

22: but also —> but will also 

24: from the emission change —> from emissions changes 

24: the health —> health 

Minor comments:

Abstract: state which model is used for this study? 

32: Not sure what is meant by “at least” in this context.  Why is this a lower bound? 

63: Are these numbers in response to the (arbitrary) 20% reduction in emissions used in the TF HTAP modeling tests?  

147-149: GBD methods evolve annually, so I wouldn’t say “latest” here.

Section 2.3: Please also report: what counterfactual values were used, if any, for PM2.5 and O3? What metric was used for the O3 concentration (annual average? 6 month? 1 hr or 8 hr max? etc).

194: What drives the isolated increase in PM2.5 in NW China shown in Fig 2? 

244: This was confusing, until I figured out they are referring specifically to 2011.  Please confirm/ clarify. 

247: This statement doesn’t make sense.  How could the changes in China alone be 43% higher than the global total change? Do they mean just the international change (excluding China) for the latter?