Impacts of emission changes in China from 2010 to 2017 on domestic and intercontinental air quality and health effect

China has experienced dramatic changes in emissions since 2010, which accelerated following the implementation of the Clean Air Action in 2013. These changes have resulted in significant air quality improvements that are reflected in 20 observations from both surface networks and satellite observations. Air pollutants, such as PM2.5, surface ozone, and their precursors, have long enough lifetimes in the troposphere to be easily transported downwind. Emission changes in China will thus not only change the domestic air quality, but will also affect the air quality in other regions. In this study, we use a global chemistry transport model (CAM-chem) to simulate the influence of Chinese emission changes from 2010 to 2017 on both domestic and foreign air quality. We then quantify the changes in air pollution-associated (including both PM2.5 and O3) 25 premature mortality burdens at regional and global scales. Within our simulation period, the population-weighted annual PM2.5 concentration in China peaks in 2011 (94.1 μg m) and decreases to 69.8 μg m by 2017.. These estimated national PM2.5 concentration changes in China are comparable with previous studies using fine-resolution regional models, though our model tends to overestimate PM2.5 from 2013 to 2017 when evaluated with surface observations. Relative to 2010, emission changes in China increased the global PM2.5-associated premature mortality burdens through 2013, among which a 30 majority of the changes (~93%) occurred in China. The sharp emission decreases after 2013 generated significant benefits for human-health. By 2017, emission changes in China reduced premature deaths associated with PM2.5 by 108, 800 (92,800—124,800) deaths yr globally, relative to 2010, among which 92% were realized in China. In contrast, the population-weighted, annually averaged maximum daily 8-hr ozone concentration peaked in 2014 and did not reach 2010levels by 2017. As such, O3 generated nearly 8,500 (6,500—9,900) more premature deaths yr in 2017 compared to 2010. 35


Introduction 40
Fine particulate matter with an aerodynamic diameter of less than 2.5 μm (PM 2.5 ) is of acute interest to the atmospheric chemistry research community due to its environmental impacts, such as visibility impairment and material damages (Hand et al., 2013(Hand et al., , 2014Wu and Zhang, 2018), and effects on human health (e.g. Pope et al., 2002;Krewski et al., 2009).
Associations between short-term and long-term PM 2.5 exposure and various deleterious human health impacts have been widely and consistently reported (Krewski et al., 2009;Burnett et al., 2014;, including premature mortality 45 via several endpoints (e.g., cardiopulmonary and respiratory disease). Exposure to ozone (O 3 ) also impacts human health, with studies reporting an association between short-term O 3 exposure and hospital admissions, emergency room visits for respiratory causes, and school absences (Katsouyanni et al., 2009), and long-term O 3 exposure with premature mortality from respiratory disease (Jerrett et al., 2009;Turner et al., 2016). PM 2.5 and its precursors can travel long distances, affecting air quality and health in other receptor regions (Ewing et al. 50 2010;Pfister et al., 2011;Anenberg et al., 2014), despite its relatively short lifetime in the atmosphere (days to weeks).
Tropospheric ozone features a longer lifetime, with a global average of ~23 days (Young et al., 2013) and the research community has expressed particular interest in studying its intercontinental transport (e.g. Zhang et al., 2008Zhang et al., , 2014Cooper et al., 2015;Lin M. et al., 2012Lin M. et al., , 2017Parrish et al., 2014). Numerous studies have investigated the source-reception relationship on air quality and associated premature mortality burden from emission changes in one source region on others 55 (West et al., 2009a,b;Fry et al., 2013;Crippa et al., 2019). Liang et al. (2018) used the ensemble model outputs from the al., 2019; UN Environment 2019; Zheng B. et al., 2018a,b). The relative change of China's anthropogenic emissions for specific air pollutants during the 2010-2017 period include: −35 % for primary PM 2.5 , -62 % for SO 2 , −27 % for BC, −35 % for OC, −17 % for NO x , and −27 % for CO (Zheng B. et al., 2018b). Surface monitors indicate that these reductions have 70 resulted in significant decreases in ambient PM 2.5 concentrations, which are further reflected in observations from satellites and results from model simulations (e.g. Song et al., 2017;Huang et al., 2018;Zheng Y. et al., 2017;Zhang Q. et al., 2019). The rapid reductions of major air pollutants in China are also confirmed using a long-term, robust observational record at Fukue Island, Japan (Kanaya et al., 2020). A recent study using high-resolution regional air quality model showed that the estimated national population-weighted annual mean PM 2.5 75 concentrations decreased from 61.8 to 42.0 μg m -3 from 2013 to 2017 . Meanwhile, summertime daily maximum 8-h average (MDA8) O 3 in China has shown an increasing trend since 2013 . The increasing trend in surface O 3 may be partially explained by the slowing down of the aerosol sink of hydroperoxyl radicals , though this chemical pathway remains disputed (Tan et al., 2020).
Previous studies have evaluated the benefits of China's Air Pollution Prevention and Control Action Plan (APPCAP) on 80 improved air quality, including both PM 2.5 and O 3 pollution, and avoided premature mortalities (e.g. Huang et al., 2018;Lu et al., 2020). However, limited studies have investigated the benefits of these actions on global air quality and the air pollution related premature mortality burden. In this study, we use a global chemical transport model to simulate the global air quality changes from 2010 to 2017 that result from the historical emission changes in China. We then estimate the air pollution-related premature mortality burden, both within China and elsewhere. This time period was 85 selected because emissions in China slightly increased from 2010 to 2013, and substantially decreased thereafter, which allows for a comparison of different emission trends in China on global air quality and health impacts.

Model simulation using the CAM-chem
derived yields for monoterpenes, isoprene, and aromatic photooxidation (Heald et al., 2008;Times et al., 2016). Recent 100 research has suggested that anthropogenic SOA may be a dominant contributor of health impacts globally (Nault et al., 2021). As our simulations lack representation of important anthropogenic SOA precursors, such as Intermediate-Volatility Organic Compounds (IVOCs; Zhao et al., 2014;Lu et al., 2020;Pennington et al., 2021), our simulated PM 2.5 concentrations may be low biased. PM 2.5 is calculated as the sum of SO 4 +NO 3 +NH 4 +OC+BC+SOA+0.2*Dust+Seasalt (West et al., 2013;Silva et al., 2016). For dust and sea salt, only the size fractions relevant for PM 2.5 (size bins 1-3) are used. Dust in desert 105 regions was found to be too high in the model, so global dust concentrations were multiplied by 0.2 to achieve rough consistency with the PM 2.5 concentrations estimated with Brauer et al. (2012). A comprehensive evaluation of model performance in simulating temporal and spatial distribution of global ozone and aerosols were carried out in previous studies (Tilmes et al., 2015Zhang et al., 2016). Here, we evaluate the model using surface ozone and PM 2.5 concentrations in China from 2013 to 2017. The lowest modeled gridcell (~58 m above the surface) is taken to indicate ground-level 110 concentrations.
Our base case simulation (CEDS_MEIC in Table 1) (Table 1). Finally, we performed a sensitivity simulation with global anthropogenic emissions frozen at 2010 levels (CEDS_MEIC_GlobalFix, Table 1). This simulation isolates the influence of meteorological changes on PM 2.5 and O 3 changes. 125 It has been reported that the CEDS emissions inventory used here is high biased in the magnitude of the Chinese emissions, and underestimates the decreasing emissions trend in China since 2013 (Zheng B. et al., 2018b;Paulot et al., 2018). Indeed, we see that in 2014 (Fig. S1) the emissions in the CEDS inventory are at least 20% higher than the emissions from the MEIC inventory for most air pollutants. More specifically, the SO 2 , OC and BC emissions in CEDS are 84%, 81% and 58% higher, respectively, in 2014 than those reported in the MEIC inventory. In addition, the CEDS inventory estimated a continued 130 increasing trend for several pollutants, whereas the MEIC inventory often peaked prior to 2012 for pollutants (Liu et al., 2016;Zheng B. et al., 2018b). We also find emissions in the CEDS inventory to be higher in the western and south parts of China when compared to the MEIC inventory, and lower in the eastern China (Fig. S2). To test the influence of these differences in Chinese emissions, we performed another sensitivity simulation from 2010 to 2014 that applied the CEDS inventory globally (CEDS_Global). This enables an evaluation into the model's performance when using a variable 135 inventory and allows for a discussion on the relative air quality changes when applying a different emission inventory in China.

Surface observation for PM2.5 and O3 in China
Hourly surface observations for PM 2.5 and O 3 were retrieved from the China National Environmental Monitoring Center O 3 , as these two metrics have been reported to be associated with the health impacts in epidemiological analyses.

Health impact assessment for surface PM2.5 and O3
To calculate health impacts, we applied the relative risk associated with long-term air pollution exposure from various epidemiological studies, baseline mortality rates, population, and modelled exposure concentrations as follows: 145 Where ∆ is mortality burden attributed to long-term PM 2.5 and O 3 exposure, 0 is the baseline mortality rates for cause of specific disease, is the attribution fraction of mortality associated with air pollution exposure, which is calculated as 1 − 1 , and is the exposed population with ages greater than 25 years old. The (relative risk) for long-term PM 2.5 exposure is calculated using the integrated exposure response model (IER, Burnett et al., 2014) from the Global Burden of 150 Disease 2017 (GBD2017) study (Stanaway et al., 2018). The for long-term O 3 exposure is retrieved from Turner et al., (2016), with reports a of 1.12 (95 % confidence interval (CI): 1.08, 1.16) for respiratory disease. Country-age-specific baseline mortality rates (Y 0 ) in 2010 were retrieved from the GBD2017 project (Stanaway et al., 2018), and remapped to match the 10 th International Statistical Classification of Diseases and Related Health Problems codes used in the cohort study (Turner et al., 2016;Seltzer et al., 2020). The theoretical minimum risk exposure level for PM 2.5 exposure assessment is 155 drawn from a uniform distribution with a lower bound of 5.8 μg m -3 and an upper bound of 8.8 μg m -3 , and for O 3 exposure it is 26.7 ppbv. Previous studies have shown that coarse resolution global CTMs, e.g. 1.9°×2.5°, likely generate low biases in estimating health effects, especially in urban areas (Li et al., 2016;Punger and West, 2013;Silva et al., 2013Silva et al., , 2016. However, less is known how these underestimates would affect the relative contributions of downwind transportation (Liang et al., 2018). Jin et al. (2019) concluded that the uncertainties in estimating the ambient PM 2.5 -related mortality burden is 160 dominated by the uncertainties in the underlying exposure-response function and less influenced by the uncertainties associated with the PM 2.5 concentration estimates.

Model evaluation with surface observation in China
In the base CAM-chem simulation (CEDS_MEIC scenario), predicted annual PM 2.5 concentrations in China are 165 overestimated, with a mean bias (MB) of 19.3 µg m -3 and normalized mean bias (NMB) of 37.2%. The MB fluctuations around 20 µg m -3 from 2014 to 2017 and is lowest in 2013 (MB of 7.6 µg m -3 ). The lower MB and NMB in 2013 could be due to the fact that there is much less data available in 2013. The consistent positive NMBs indicate that the overestimates are systematic. This should not affect our main conclusions since we focus on the changes among years. The high modelling bias for CAM-chem predicted surface PM 2.5 has also been reported in other studies (e.g., He and Zhang, 2014). These studies 170 attribute CAM-chem modeling bias in surface PM 2.5 to predictions of SO 4 2-, NH 4 + , and organic aerosols, and missing major inorganic aerosol species such as nitrate and chloride (He and Zhang, 2014;Tilmes et al., 2016). By including advanced inorganic aerosol treatments, such as condensation of volatile species, explicit inorganic aerosol thermodynamics for sulfate, ammonium, nitrate, sodium, and chloride (He and Zhang, 2014), and more comprehensive representing secondary organic aerosol formation (e.g., using the Volatility Basis Set scheme, Times et al., 2019;, modeling performance 175 could significantly improve. We also find that the bias metrics exhibit small inter-annual variability, with NMB generally around 40% and NME around 50%. Exceptions include the years 2013 and 2014, which have smaller NMB values due to the limited number of the observations (Table 2). CAM-chem can generally reproduce the spatial patterns of the annual PM 2.5 distributions, with a correlation coefficient (R) greater than 0.7. Notably, simulations using only CEDS emissions (i.e., the CEDS_Global scenario) generate better performance in both 2013 and 2014. Part of the reason is that we have less available 180 data in these two years. However, we suspect the main reason for this variable bias is due to spatial changes in emissions between the inventories. While total emissions in China are higher in the CEDS inventory than in the MEIC inventory, CEDS tends to allocate more emissions in western and central China (Fig. S2). As such, predictions of annual PM 2.5 concentrations using the CEDS inventory are lower in eastern China and higher in western and northwestern China (Fig. S3).
CAM-chem NMB in simulating the annual MDA8 O 3 is lower than 20% for all evaluated years (Table 3). From Table 3, we  185 can also see that CAM-chem overestimates the annual MDA8 O 3 in China, which means our estimates for the O 3 -related mortality burden will likely be biased high, too. A high bias of about 10 ppb can be attributed to the coarse model resolution, which leads to an overestimate of ozone production due to diluted emissions of ozone precursors (Tilmes et al, 2015). In contrast to the PM 2.5 performance, the CEDS simulation (CEDS_Global scenario) generates poorer model performance for MDA8 O 3 . For both PM 2.5 and ozone, we also find that the NMBs are lower in the eastern China compared with other inland 190 regions (Figs. S5-S6). PM 2.5 concentration decreased by 15.9 µg m -3 , which is comparable to the reductions (19.8 µg m -3 ) reported by , who used a high-resolution regional air quality model. At the national-scale, the Pop-weighted PM 2.5 concentration features a similar trend as the area-weighted average trend, but is notably higher (Fig. 1a), indicating that higher PM 2.5 concentrations happen in regions with higher population density. The annual average of area-weighted PM 2.5 concentration decreased by 7.6 µg m -3 between 2010 and 2017, which is consistent with the estimates reported by Ding et al. 200 (2019a) at 9.0 µg m -3 . The surface PM 2.5 changes in China due to emission changes usually peak in the fall and winter (Fig.   S7a). Spatially, we see that significant PM 2.5 changes (increases before 2013 and decreases thereafter) occur in eastern China (Fig. 2), which was the focus region for China in the APPCAP (Ding et al., 2019a,b). We also find that the annual PM 2.5 decreases in China are mainly dominated by the changes in emissions and not due to meteorology, which is consistent with Meanwhile, the emission changes led to annual PM 2.5 decreases as high as 16.7 µg m -3 in 2017, and increases as high as 4.9 µg m -3 in 2015 (Fig. 4a). There were also isolated increases in PM 2.5 in northwest China from 2010 to 2013, which were mainly caused by the dust storms Luo et al., 2020;Zhao et al., 2020).

Air quality changes in
In contrast to the PM 2.5 trends, the annual average Pop-weighted MDA8 O 3 has increased relative to 2010 (Fig. 1b), with a 210 peak in 2014 (59.5 ppbv) and subsequent decreases to 2017 (57.1 ppbv). The area-weighted MDA8 O 3 was comparable to or larger than the Pop-weighted concentration due to the more uniform O 3 distribution in China or even higher ozone events in western China from stratosphere-troposphere exchange (Wang et al., 2011;Li et al., 2019). For ozone, the emission changes from 2010 to 2013 exacerbate summer ozone pollution in China, but alleviate ozone pollution in the other three regions (Fig.   S8a). After 2013, the emission decreases in China exacerbate the ozone pollution for all the seasons, especially in winter. 215 The spatial pattern of ozone trends mainly featured increases in the Beijing-Tianjin-Hebei and Yangtze-River-Delta regions, and slightly decreases in the south ( Fig. 3; Fig. S5). The anthropogenic emission reductions in China that led to ozone increases ( Fig. 4b) could partially be explained by the aerosol sink of hydroperoxyl radicals, which slowed in recent years due to PM 2.5 decreases . In addition, the effect of inter-annual meteorological conditions had a profound influence on the annual average MDA8 O 3 concentrations. Ozone increased as much as 8.1 ppbv due to meteorology. The 220 meteorology-induced ozone increases can be attributed to increasing temperature, which enhances the ozone production and biogenic NMVOCs emissions (Ding et al., 2019b;, and the increases solar radiation Ma et al., 2021).

Global and regional air quality 225
The simulated global tropospheric ozone burden (total ozone below the chemical tropopause of 150 ppbv) from 2010 to 2017 from the CEDS_MEIC simulation is 327.5 ± 5.2 Tg, which agrees well with the present tropospheric ozone burden estimated from previous ensemble models (ACCENT: 336 ± 27 Tg; ACCMIP: 337 ± 23 Tg; TOAR: 340 ± 34 Tg, and CMIP6: 348 ± 15 Tg; Griffiths et al., 2021). From Fig. 5, we find that the change in global tropospheric ozone burden from the emission changes in China ranges from 0.6 Tg (2011 and 2012) to -1.9 Tg (2017). The tropospheric ozone burden changes are not 230 only seen in China, but also in downwind regions, including throughout the Northern Hemisphere (Fig. 6).
Due to the prevailing western wind, air pollution in China could be easily transported to downwind regions, especially during springtime (Lin M. et al., 2012;Liang et al., 2018). From Fig. 6, we see Chinese emissions generate the largest impacts in South Korea, which experienced the largest pop-weighted PM 2.5 changes from 2010 through 2017 (ranging from 0.7 µg m -3 in 2012 to -2.63 µg m -3 in 2017; Fig. 6a). The influences are largest in spring than the other seasons for all the 235 downwind regions (Fig. S7b-d). For ozone, we find that the emission changes in China have increased surface ozone in South Korea since 2010, mainly form the increased export of ozone. Both Japan and U.S. first experience increases in ambient pollution due to changes in Chinese emissions, but subsequent decreases. However, the increases and the decreases are not temporally aligned and typically modest in magnitude. For the downwind regions, the season with peak ozone changes also varies ( Fig. S8b-d). 240

Global and regional air pollution-related mortality burden changes
In our simulations, the global ambient PM 2.5 -related mortality burden in 2010 is 4.08 million (95%CI: 2.15-6.0 million), which is consistent with previous estimates using the same year and applying the same IER method (3.6±1.0 million in 2010, Shindell et al., 2018). Relative to 2010, the emission changes in China lead to premature mortality increases of 27,700 deaths yr -1 (95%CI: 23,900-31,400 deaths yr -1 ) in 2011, with 93% occurring in China (25,800,95%CI: 22, For ozone, the global premature mortality burden in 2010 is 1.02 million (95% CI: 0.73-1.28 million), which is consistent with prior estimates using other global CTMs, such as GEOS-Chem (1.04-1.23 million) and GISS (0.8-1.3 million), while applying the same relative risk value (Mally et al., 2017;Shindell et al., 2018). The emission changes in China increased the global ozone-related mortality by 4,900 (95%CI, 3,700-5,900) premature deaths yr -1 in 2011 (Table 5), among which 73% occurs in China (3600 premature deaths yr -1 , 95%CI: 2,700-4,300). For the three downwind regions considered here, South 260 Korea, Japan, and U.S., the added O 3 -related premature mortality burdens in each country are 23, 172, and 131 premature deaths yr -1 , respectively. By 2017, the anthropogenic emission reductions in China increased the ozone-related mortality burden within China by 8,500 premature deaths yr -1 and generated mixes impacts elsewhere. In the downwind regions, the O 3 -related premature mortality burden decreased in some locations (-65 and -289 premature deaths yr -1 for Japan and U.S. respectively), and increased it in others (e.g., South Korea). 265

Conclusions
Dramatic changes in anthropogenic emissions within China have occurred since 2010, with most air pollutants, such as NO X , SO 2 , peaking around 2012 and 2013, and decreasing significantly thereafter. In this study, we use a global chemistry transport model (CAM-chem) to simulate the effects of emission changes in China on domestic and international air quality, as well as the subsequent air pollution-related mortality burden, from 2010 to 2017. An evaluation of model performance 270 indicated that our model tends to overestimate the annual PM 2.5 concentrations in China, with a normalized mean bias (NMB) of 37.2% and normalized mean error (NME) of 52.0%. The PM 2.5 overestimation is likely caused by uncertainties in the bottom-up emission inventories Zhang Q. et al., 2019) and missing pollution pathways for PM 2.5 components (Tilmes et al., 2015. However, our biases are similar in scale to the biases reported in other high-resolution regional models (25%-30% in Shen et al., 2019;~20% in Zhang Q. et al., 2019). We also evaluated 275 model performance by applying two sets of emission inventories: a regional emission inventory (MEIC) and a global emission inventory (CEDS), which was extensively in the CMIP6 experiments. For surface PM 2.5 , we find that model performance with the CEDS inventory tends to predict lower bias metrics, which we attribute to the spatial allocation differences in the two inventories. For surface O 3 , the simulation using the MEIC inventory generates a lower NMB (13.7%) and NME (21.9%) than a comparable simulation using the CEDS inventory (NMB of 15.2 and NME of 36.6%). 280 Our simulations suggest that the annual average, population-weighted (Pop-weighted) PM 2.5 concentration in China peaked in 2011 (94.1 µg m -3 ) and has decreased sharply thereafter. The annual average, Pop-weighted PM 2.5 concentration in 2017 was 17.6% (-14.9 µg m -3 ) smaller than the concentration in 2010 (84.7 µg m -3 ). Though CAM-chem overestimates PM 2.5 concentrations in China, the simulated decreasing trend for annual PM 2.5 we report here (-15.9 µg m -3 for Pop-weighted and 7.6 µg m -3 for area-weighted from 2013 to 2017) is comparable with prior studies using a higher resolution regional air 285 quality model (-19.8  premature deaths yr -1 (95% confidence interval (CI): 23,900-31,400) in 2011 to 13,300 (95%CI: 11,110) premature deaths yr -1 in 2013, most of which occurred within China (~93%). The sharp emission decline following 2013 brought about significant health benefits by 2017. Relative to 2010, premature deaths declined by 108,800 (92,800-124, 290 800) yr -1 due to emission changes in China, most of which (92%) were realized within China. Downwind regions, such as South Korea, Japan, and U.S., experienced similar PM 2.5 trends due to Chinese emissions, but far smaller in scale. Our simulations indicate that the transport of PM 2.5 and its precursors changed the annual Pop-weighted PM 2.5 concentration in South Korea by 0.7 µg m -3 in 2011 and -2.6 µg m -3 in 2017. This led to 98 additional premature deaths in 2011, and 386 avoided premature deaths in 2017. Annual average Pop-weighted PM 2.5 concentrations in Japan were influenced less by 295 Chinese emissions, but generated much larger changes in PM 2.5 -associated premature mortality burden changes due to the age structure of the population and higher population (https://countryeconomy.com/countries/compare/japan/southkorea?sc=XE23, last accessed Sep 3 rd , 2021). The influence of Chinese emission on U.S. air quality led to 44 additional premature deaths in 2011, and 381 avoided premature deaths in 2017. In contrast to the PM 2.5 trends, Chinese emission changes increased the annual average maximum daily 8-hr ozone concentration in China, which subsequently increased 300 ozone-related premature deaths. These changes ranged from 3,600 additional premature deaths yr -1 in 2011 to 8,500 additional premature deaths yr -1 in 2017. Downwind regions, such as South Korea, Japan, and the U.S., also experienced health impacts (both benefits and dis-benefits, depending on the location). In general, we conclude that the sharp emission