1 The Impact of Power Generation Emissions on Ambient PM2.5 Pollution and Human Health in China and India Meng Gao a,*, Gufran Beig b, Shaojie Song a, Hongliang Zhang c, Jianlin Hu d, Qi Ying e, Fengchao Liang f,g, Yang Liu g, Haikun Wang h, Xiao Lu a,i, Tong Zhu j, Gregory R. Carmichael k, Chris P. Nielsen a,*, Michael B. McElroy a a Harvard John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA 02138, USA b Indian Institute of Tropical Meteorology, Pune, Maharashtra 411008, India c Department of Civil and Environmental Engineering, Louisiana State University, Baton Rouge, LA 70803, USA d School of Environmental Science and Engineering, Nanjing University of Information Science & Technology, 219 Ningliu Road, Nanjing 210044, China e Zachry Department of Civil Engineering, Texas A&M University, College Station, TX 77843, USA f Department of Occupational and Environmental Health, School of Public Heath, Peking University, Beijing 100191, China g Department of Environmental Health, Rollins School of Public Health, Emory University, Atlanta, Georgia 30322, United States h State Key Laboratory of Pollution Control and Resource Reuse, School of the Environment, Nanjing University, Nanjing 210023, PR China i Laboratory for Climate and Ocean-Atmosphere Sciences, Department of Atmospheric and Oceanic Sciences, School of Physics, Peking University, Beijing 100871, China j State Key Lab for Environmental Simulation and Pollution Control, College of Environmental Science and Engineering, Peking University, Beijing 100871, China k Center for the Global and Regional Environmental Research, the University of Iowa, Iowa City, IA 52242, USA * Corresponding Authors. mgao2@seas.harvard.edu (M. Gao) and nielsen2@fas.harvard.edu (C.P. Nielsen) mailto:mgao2@seas.harvard.edu mailto:nielsen2@fas.harvard.edu 2 Abstract Emissions from power plants in China and India contain a myriad of fine particulate matter (PM2.5, PM≤2.5 micrometers in diameter) precursors, posing significant health risks among large, densely settled populations. Studies isolating the contributions of various source classes and geographic regions are limited in China and India, but such information could be helpful for policy makers attempting to identify efficient mitigation strategies. We quantified the impact of power generation emissions on annual mean PM2.5 concentrations using the state-of-the-art atmospheric chemistry model WRF-Chem (Weather Research Forecasting model coupled with Chemistry) in China and India. Evaluations using nationwide surface measurements show the model performs reasonably well. We calculated province-specific annual changes in mortality and life expectancy due to power generation emissions generated PM2.5 using the Integrated Exposure Response (IER) model, recently updated IER parameters from Global Burden of Disease (GBD) 2015, population data, and the World Health Organization (WHO) life tables for China and India. We estimate that 15 million (95% Confidence Interval (CI): 10 to 21 million) years of life lost can be avoided in China each year and 11 million (95% CI: 7 to 15 million) in India by eliminating power generation emissions. Priorities in upgrading existing power generating technologies should be given to Shandong, Henan, and Sichuan provinces in China, and Uttar Pradesh state in India due to their dominant contributions to the current health risks. Keywords: Air quality modeling, Power generation, China, India, WRF-Chem 3 1. Introduction Exposure to fine particulate matter (PM2.5) has been linked to mortality from a variety of causes in both adults (ischemic heart disease, stroke, chronic obstructive pulmonary disease, lung cancer) and children (acute lower respiratory infections) (Dockery et al., 1993; Hoek et al., 2013). In Asia, particularly China and India, PM2.5 pollution has been an increasingly important research topic, and has attracted worldwide attention. A large fraction of the world’s population lives in these two countries where they are exposed to extremely unhealthy air. Lim et al. (2012) estimated that ambient PM2.5 pollution is the 4th largest contributor to deaths in China and the 5th in India. Anthropogenic activities, including industry, power generation, transportation, and residential energy usage (heating and cooking), contribute to the total ambient concentrations of PM2.5 directly and indirectly through gas-to-particle conversions. In China and India, secondary inorganic aerosols account for a large portion of the ambient PM2.5 mass concentration (Huang et al., 2014, Singh et al., 2017), which is mainly formed from sulfur dioxide (SO2) and nitrogen oxides (NOx). A significant use of coal in China and India generates large amounts of SO2 and NOx (Lu et al., 2011). In both China and India, power and industrial sectors are the largest sector consumers of coal (Lu et al., 2011). According to statistics in Li et al. (2017), the power generation sector contributes 28.5% and 32.5% to SO2 and NOx emissions in China, and 59.1% and 25.0% in India. For decades, the influence of power generation emissions on health damages in the United States has been of interest (Buonocore et al., 2014; Fann et al., 2013; Levy and Spengler, 2002; Levy et al., 2002, 2009). With growing attention to serious air pollution in China and India, the resulting health risks have become the focus of many studies. The 2010 Global Burden of Disease report (Lim et al., 2012) analyzed the worldwide impacts of PM2.5 and estimated that 1.2 million lives (corresponding to 25 million DALYs (Disability-Adjusted Life Year)) were lost in China and 0.6 million lives (corresponding to 17.7 million DALYs) in India each year due to ambient PM2.5 exposure, but results were not disaggregated to identify the impacts of various emission sources. Lelieveld et al. (2015) provided a similar assessment of the worldwide mortality impacts of PM2.5; they estimated that 1.3 million deaths each year in China and 0.6 million in India were attributable to ambient PM2.5 exposure, suggesting additionally that 18% and 14% of deaths attributable to PM2.5 exposure, respectively, were linked to power generation. More recently, HEI (2017) analyzed the mortality impacts of PM2.5 in China and estimated that 0.9 million lives were lost in 4 2013 due to ambient PM2.5 exposures. HEI (2017) also provided a breakdown of the contributions of multiple major source classes to this impact and the geographic distribution of the contributions. These studies present quite different pictures of the importance of utility coal combustion. In any study of the impacts of PM2.5 on mortality, the analyst faces several important choices: • What emissions inventory to use; • What approach to employ for atmospheric modeling, at what spatial resolution; • Which exposure-response coefficients to use; • Whether to focus on marginal or average impacts; and • Whether to report results as ‘attributable deaths,’ ‘years of life lost,’ and/or ‘DALYs.’ The choices underlying several previous studies are summarized in Table 1. Some of these studies relied on relatively simple global chemical transport models, in which important mesoscale information (i.e. boundary layer processes, cloud physics, etc.) might be missing and/or oversimplified. In addition, some are driven by older emissions inventories, which now have been replaced. Finally, many of the studies use older estimates (2010 or 2013) of the IER parameters, and fail to show convincing evidence of that their atmospheric fate and transport models agree with ground based PM2.5 measurements. This study fills these gaps by using a regional scale chemistry-meteorology model WRF-Chem (Weather Research Forecasting-Chemistry, Gao et al., 2015, 2016a, 2016b, 2016c, 2017; Marrapu et al., 2014); nationwide surface PM2.5 measurements in both China and India; a state-of-the-art emission inventory; and the newly updated IER parameters from GBD 2015 (Cohen et al. 2017). We present our results in terms of both the number of deaths attributable to PM2.5 exposure and the number of years of life lost (YLL) due to PM2.5 exposure. 5 Table 1. Summary of choices of estimation inputs in previous studies Study Emissions Inventory Atmospheric Model & Resolution Exposure- Response Coefficients Marginal or Average Impacts Results Reported Lim et al., 2012 1 n/a n/a 2010 Marginal Attributable Deaths & DALYs Lelieveld et al. 2015 EDGAR EMAC, 1.1°×1.1° degree 2010 Marginal Attributable Deaths HEI (2017) MIX GEOS-Chem, 0.5×0.667 degree 2013 Marginal Attributable Deaths GBD 2015 2 n/a n/a 2015 Marginal Attributable Deaths & DALYs 1 Global estimates of PM2.5 at 0.1°×0.1° scale: combination of TM5 global chemical transport model simulated PM2.5 (at 1°×1° resolution), satellite aerosol optical depth (AOD) derived PM2.5 (the relationship between AOD and PM2.5 is calculated using GEOS-Chem at 2°×2.5° resolution), and surface PM measurements (Brauer et al., 2012) 2 Global estimates of PM2.5 at 0.1°×0.1° scale: combined estimates from satellite AOD, chemical transport models (GEOS-Chem) and ground-level measurements (Cohen et al. 2017; van Donkelaar et al., 2015) 2. Materials and Methods 2.1 Atmospheric Modeling In this study, the WRF-Chem model version 3.6.1 was implemented to cover both China and India. WRF-Chem is a fully online coupled regional scale meteorology-chemistry model that enables aerosol-radiation-cloud interactions (Grell et al., 2005), and includes multiple options for physical and chemical parameterizations. The main chosen options for physical parameterizations of the Planetary Boundary Layer (PBL), cloud microphysics, and land surface are listed in Table S1, which include the Yonsei University PBL scheme (Hong et al., 2006), the Noah land surface scheme, the Goddard shortwave radiation scheme (Chou et al., 1998), the RRTM (Rapid Radiative 6 Transfer Model) long wave radiation scheme (Mlawer et al., 1997), and the Lin cloud microphysics scheme (Lin et al., 1983). We use the Carbon Bond Mechanism version Z (CBMZ, Zaveri and Peters, 1999) for gas-phase chemistry, and the Model for Simulating Aerosol Interactions and Chemistry (MOSAIC, Zaveri et al., 2008), which calculates size resolved sulfate, nitrate, ammonium, black carbon, organic carbon, and secondary organic aerosol (SOA). Dust and sea-salt are also considered in the model configuration. Simulations are conducted for the entire year of 2013, and the model was initialized at the beginning of each month, with the last 5 days of the previous month employed as model spin-up. The model is configured with a horizontal resolution of 60 km and with 27 vertical levels up to 50 hPa. Meteorological initial and boundary conditions are obtained from the National Centers for Environmental Prediction final analysis (NCEP FNL) 6-hourly 1° ×1° data, and analyses of wind, temperature and water vapor are nudged to correct model meteorology fields using the four-dimensional data assimilation (FDDA) method. Chemical initial and boundary conditions are taken from the Model for Ozone and Related chemical Tracers, version 4 (MOZART-4) global simulations (Emmons et al., 2010). 2.2 Emissions Anthropogenic emissions are adopted from the MIX emission inventory (Li et al., 2017), which combines five emission inventories for Asia and is considered as the most advanced inventory for Asia to date. Among them, the Multi-resolution Emission Inventory for China (MEIC) developed by Tsinghua University is used over China, and the ANL emission inventory developed at the Argonne National Laboratory is used over India (ANL-India). Power plant emissions in MEIC are taken from the China coal-fired Power Plant Emission Database (CPED), which includes estimates of emissions for each generation unit considering fuel consumption rates, fuel quality, combustion technology and emission control technology (Li et al., 2017). In ANL-India, power plant emissions are calculated also for each generation unit based on the reports of the Central Electricity Authority (CEA), which includes detailed information on geographical location, capacity, fuel type, electricity generation, time the plant was commissioned/decommissioned, etc. (Li et al., 2017). The ANL-India emission inventory covers only some MIX species (SO2, BC, and OC for all sectors, and NOx for power plants), and emissions of other species are taken from the REAS2 (Regional Emission Inventory in Asia version 2, Kurokawa et al., 2013) inventory (Li et al., 2017). 7 The MIX inventory includes 10 species, namely SO2, NOx, CO, non-methane volatile organic compounds (NMVOCs), ammonia (NH3), PM2.5, PM10, black carbon (BC), organic carbon (OC), and carbon dioxide (CO2). In this study, the MEIC data for 2010 are replaced in MIX with the revised results from MEIC 2013, but emissions for India are not changed. Biogenic emissions are calculated online using the MEGAN model (Model of Emissions of Gases and Aerosols from Nature, Guenther et al., 2012), and the driving variables of this model include land cover, weather, and atmospheric chemical composition. GFEDv4 (Global Fire Emissions Database, Version 4) emissions are used for open biomass burning, based on satellite information on fire activity and vegetation productivity (Randerson et al., 2015). Other emissions including online dust emissions and online sea-salt emissions are also considered. 2.3 Study Design We explore the impact of each emission sector on annual mean PM2.5 concentrations for the year 2013 through a series of simulations. Descriptions of these simulations are listed in Table S2. The BASE case includes all anthropogenic emission sectors, biogenic emissions, and biomass burning emissions. In the noIND case, the anthropogenic emissions from the industrial sector are excluded, and all the other settings are the same as in the BASE case. The remaining cases are similar. The noPOW case excludes power plant emissions, the noAGR case excludes agriculture emissions, the noTRA case excludes transportation emissions, the noRES case excludes residential emissions, and the noBB case excludes biomass burning emissions. The differences between BASE and noIND, BASE and noPOW, BASE and noAGR, BASE and noTRA, BASE and noRES, and BASE and noBB, are considered estimates of the impact of eliminating industrial emissions, power plant emissions, transportation emissions, residential emissions, and biomass burning emissions respectively. The contributions of source sectors outside the domain are not further separately quantified in this study. 2.4 Observational Networks 8 This study benefits from a wealth of nationwide surface PM2.5 measurements in China and India which allow us to evaluate the performance of the WRF-Chem simulated PM2.5 concentrations. Since January 2013, the China National Environmental Monitoring Center (CNEMC) has released monitored PM2.5 concentrations to the public. The CNEMC monitoring sites (shown as red dots in Figure S1) are located mostly in east China. Hourly average PM2.5 concentrations for 2013 were downloaded from the www.cnemc.cn website. This dataset has been used widely to statistically evaluate air quality models across China (Hu et al., 2016, 2017). Modeling of Atmospheric Pollution and Networking (MAPAN) was set up by the Indian Institute of Tropical Meteorology (IITM) under the project SAFAR (System of Air Quality and weather Forecasting And Research) (Beig et al., 2015, WMO report) across all of India to measure various pollutants, including ozone (O3), NOx, PM2.5, PM10, CO, hydrocarbon, BC and OC, as well as weather parameters. The measured PM2.5 at the MAPAN sites are used in this study to evaluate model performance for India. 2.5 Mortality Analysis The lack of cohort mortality evidence in developing countries, such as China and India, hinders research on health impacts attributable to PM2.5 exposure. Burnett et al. (2014) developed integrated exposure response (IER) functions to include data from western cohort studies of exposures to PM2.5 in ambient air and the smoke from active and second-hand tobacco smoking as well as from the burning of solid fuels for household cooking and heating. We rely on the 2015 GBD IER because it is the most widely accepted and employed synthesis of the epidemiological evidence on the mortality impacts of PM2.5. In this study, annual mean ambient PM2.5 concentrations derived from the WRF-Chem model are taken into the IER functions to examine mortality attributable to PM2.5 exposure. In GBD the mortality burden attributable to PM2.5 is calculated for four diseases among adults, namely ischemic heart disease (IHD), stroke (STK, including both ischemic and hemorrhagic stroke), lung cancer (LC), and chronic obstructive pulmonary disease (COPD), and for one disease among young children, acute lower respiratory infections (LRI). The RR for each disease is calculated as, http://www.cnemc.cn/ 9 𝑅𝑅𝑖,𝑗,𝑘(C𝑙) = { 1 + 𝛼𝑖,𝑗,𝑘 (1 − 𝑒−𝛽𝑖,𝑗,𝑘(𝐶𝑙−𝐶0) 𝛾𝑖,𝑗,𝑘 ) , 𝐶𝑙 ≥ 𝐶0 1, 𝐶𝑙 < 𝐶0 (1) where C𝑙 is the annual PM2.5 concentrations calculated from the WRF-Chem model in the lth geographic region, and 𝐶0 is the counterfactual concentration; 𝛼𝑖,𝑗,𝑘 , 𝛽𝑖,𝑗,𝑘 and 𝛾𝑖,𝑗,𝑘 are the parameters used to describe the shape of IER curves in the ith age and jth sex group for the kth disease. Our calculations rely on the GBD 2015 IER parameter estimates reported by Cohen et al. (2017) for 𝛼𝑖,𝑗,𝑘 , 𝛽𝑖,𝑗,𝑘 , 𝛾𝑖,𝑗,𝑘 and 𝐶0 ,. These new parameters reflect all cohort studies conducted on subjects living in the US and Western Europe published as of mid-2016. More detailed explanations for the revised methods are documented in the appendix for Cohen et al. (2017). The relative risk (RR) factors are used then to calculate population attributable fractions (PAF, equation 2), for each disease for each age and sex subgroup. 𝑃𝐴𝐹𝑖,𝑗,𝑘 = 𝑅𝑅𝑖,𝑗,𝑘(C𝑙) − 1 𝑅𝑅𝑖,𝑗,𝑘(C𝑙) (2) 𝛥𝑀𝑖,𝑗,𝑘,𝑙 = 𝑃𝐴𝐹𝑖,𝑗,𝑘,𝑙 × 𝑦0𝑖,𝑗,𝑘,𝑙 × 𝑃𝑜𝑝𝑖,𝑗,𝑙 (3) Equation (3) is used to calculate mortality, 𝑀, attributable to PM2.5 exposure for each disease. 𝑦0𝑖,𝑗,𝑘,𝑙 represents the current age-sex-specific mortality rate for the kth disease, and 𝑃𝑜𝑝𝑖,𝑗,𝑙 reflects the size of the exposed population in that age-sex-specific group in that grid cell. The United Nations (UN)-adjusted population distribution for years 2010 and 2015 from the Center for International Earth Science Information Network (CIESIN) are used to calculate the population exposure. To represent the population in 2013, we average data for years 2010 and 2015. The estimates for year 2013 are re-gridded to 0.5° × 0.5° horizontal resolution, which approximates the WRF-Chem model resolution. National baseline age-sex-disease-specific dependent mortality rates for IHD, STK, LC, COPD, and LRI for years 2010 and 2015 are obtained from the GHDx (Global Heath Data Exchange) database. We interpolate to year 2013 based on the trends observed from 2005 to 2015. For China, provincial level baseline rates are estimated using the relationships between provincial and national rates shown in Xie et al. (2016). 10 While it is common to report the number of ‘premature deaths attributable to air pollution’ calculated in this manner, it has long been known that estimates of ‘premature deaths’ based on the attributable fraction may be biased in either direction and misleading (Robins and Greenland, 1989; Greenland and Robins, 1991; Hammitt, 2017 (under review)). Fortunately, estimates of the impacts of PM on life expectancy are not affected by these issues and are therefore preferred. The impact of PM exposure on the life expectancy of the population is calculated by multiplying the number of deaths (𝑁𝑖,𝑗) in each age and sex group by the remaining life expectancy (𝐿𝑖,𝑗) for that age and sex group and summing across all age and sex groups: 𝑌𝐿𝐿 = ∑ 𝑁𝑖,𝑗 × 𝐿𝑖,𝑗 𝑖,𝑗 In this study, life tables for China and India for 2013, downloaded from the World Health Organization website, are used. In 2013, the life expectancy at birth for Chinese males and females were 74.1 and 77.2 years, respectively, and for Indian males and females 66.2 and 69.1 years, respectively (Table S3). It is important to note that the approach we use is different from that used in the GBD studies. The GBD uses the life tables from Japan (which has the highest life expectancy in the world – 80.3 years for men and 86.7 for women) in the calculation of disability adjusted years lost (DALYs) rather than using country-specific life tables. They do so in an effort to reflect the potential benefits of improvements in air quality in a world where other public health risks had already been mitigated. As a result of this difference in approach, our estimates of life expectancy impacts will be lower by 5 to10% than those given by studies which use DALYs and rely on Japanese life tables. 2.6 Source Sector Attribution The gridded annual surface PM2.5 concentrations from the BASE case and the noPOW case are used to calculate the fraction of PM2.5 health impacts attributable to power generation emissions using the following equation: 𝐹𝑝𝑜𝑤 = 𝑃𝑀𝐵𝐴𝑆𝐸 − 𝑃𝑀𝑛𝑜𝑃𝑂𝑊 𝑃𝑀𝐵𝐴𝑆𝐸 where 𝑃𝑀𝐵𝐴𝑆𝐸 and 𝑃𝑀𝑛𝑜𝑃𝑂𝑊 denote annual mean surface PM2.5 concentrations from the WRF- Chem BASE and noPOW cases, respectively. 11 This approach is similar to that used in the China MAPS (Major Air Pollution Sources project) study for apportioning mortality impacts to various source classes. It differs however from the approach used by Lelieveld et al. (2015) – in which the contribution of a source class was computed using: 𝐹𝑠𝑜𝑢𝑟𝑐𝑒 𝑐𝑙𝑎𝑠𝑠 = 𝑀(𝑃𝑀𝐵𝐴𝑆𝐸)−𝑀(𝑃𝑀𝑤𝑖𝑡ℎ𝑜𝑢𝑡 𝑠𝑜𝑢𝑟𝑐𝑒 𝑐𝑙𝑎𝑠𝑠) 𝑀(𝑃𝑀𝐵𝐴𝑆𝐸) , where 𝑀 denotes mortality. It is important to recognize that the approaches taken by China MAPS and by Lelieveld et al. (2015) differ in the question they seek to answer. The approach of Lelieveld et al. (2015) follows the tradition of forward-looking ‘consequential’ analysis. It seeks to determine how large of a reduction in mortality would be expected if emissions from a single source class were eliminated. In contrast the China MAPS approach is rooted in backward-looking ‘attributional analysis’ and seeks to determine what fraction of total PM2.5 related mortality is attributable to (caused by) emissions of a single source. In cases where the exposure-response function is linear, these two approaches will give the same answer. However, given the strong nonlinearities of the IER concentration-response function, they will not in the current case. At current levels of PM exposure in China, the consequential approach used by Lelieveld et al. (2015) will give estimates of source class impacts that are significantly smaller – perhaps by a factor of 2 or 3 – than those from the attributional approach. 3. Results 3.1 Evaluation of surface PM2.5 In this study, we compare the WRF-Chem outputs from the BASE case against surface PM2.5 observations in the CNEMC network in China and the MAPAN network in India. Figure 1 shows how our model performs in capturing temporal variations of PM2.5 concentrations in China and India. In general, the simulated monthly and seasonal trends of PM2.5 surface concentrations in China are consistent with surface measurements, with extremely high monthly PM2.5 during winter months and relatively low concentrations during summer months (Figure 1(a)). The calculated R2 value for China is as high as 0.87. In China, PM2.5 concentrations during summer are overestimated by our model, which is likely due to errors in model wet deposition. Summer is the season with a 12 large amount of precipitation, but the 60 km horizontal resolution in this study is insufficient to capture the variability in summer precipitation. The calculated R2 value for India is 0.54, and the simulated magnitudes of surface PM2.5 concentrations are close to the measurements across India. Detailed comparisons for each observation site in China and India are presented in Figure S2 and S3 in the supporting information. For estimates of PM2.5 exposure, spatial features are more important. We also evaluate the spatial distribution of simulated yearly mean PM2.5 surface concentrations by comparing the model results against observations at 58 cities in China and 9 sites in India. As shown in Figure 2(a), in China, high PM2.5 concentrations are located mainly in eastern China and southwestern China (Sichuan Basin) due to high local emissions of air pollutants. Relatively high PM2.5 concentrations in Xinjiang, in the northwest, result partly from windblown dust. In India, PM2.5 pollution hotspots are located mostly in the Indo-Gangetic Plain (IGP), not only because of high emissions of air pollutants. Reduced ventilation due to obstruction from the Tibetan Plateau may also play a role. The calculated mean bias, index of agreement, and normalized mean bias are -15.7 (-5.7), 0.8 (0.77), and 21.1% (-12.5%) for China (India), respectively. The model generally reproduces well the spatial patterns of observed PM2.5 concentrations in 2013, with high magnitudes of PM2.5 over South Hebei, Shandong and Henan provinces in China. Relatively low PM2.5 magnitudes in south China are also well reproduced well by the model. For India, high measured concentrations of PM2.5 over the IGP is simulated well by the model, and relatively low concentrations observed in southern regions are also consistent with the model results. The above evaluations show that the WRF-Chem model has reasonable success in simulating both the temporal and spatial features of PM2.5 concentrations in China and India, and results are consequently reliable for use in analysis of health exposures. 13 Figure 1. Simulated and observed monthly mean PM2.5 concentrations averaged across China (a) and India (b) 3.2 Impacts of Source Sectors on PM2.5 Concentrations PM2.5 in the atmosphere is emitted either directly or formed through gas-to-particle conversions from various emission source sectors. Understanding the contributions of source sectors on air quality and climate forcing is of great importance for policy makers, charged with design of emission control strategies. In this study, we examine the relative importance of individual source sectors on PM2.5 concentrations in China and India excluding them one-by-one from model emission inputs in simulations (listed in Table S2). Figure 2 (b-g) shows the individual impact of industrial, power plant, agriculture, residential, transportation, and biomass burning emissions on annual mean PM2.5 concentrations in China and India. In China, the industrial sector is the largest contributor, followed by power generation. In India, emissions from power generation significantly increase PM2.5 concentrations, and have a larger effect compared with residential and other source sectors. Transportation emissions play a minor role in both countries and PM2.5 increases induced by biomass burning are significant only in Southeast Asian countries. Over Laos and northern Thailand, PM2.5 increases resulting from biomass burning can reach as high as 50μg/m3. 14 In China and India, secondary inorganic aerosols (mostly sulfate and nitrate formed via oxidation of SO2 and NOx emissions) account for a large fraction of the PM2.5 mass concentration (Huang et al., 2014, Singh et al., 2017). In China, emissions from power generation contribute 28.5% of SO2 and 32.5% of NOx emissions. In India, the contribution of power generation of SO2 emissions is almost 60% (Figure S4). These significant emissions of SO2 and NOx by power generation in China and India lead to substantial increments in PM2.5 mass concentrations, as shown in Figure 2(c) and Figure S5. Figure 2. Spatial distributions of annual mean surface PM2.5 concentrations (a: observations are shown in dots) and sector contributions from industry, power plants, agriculture, residential, transportation and biomass burning (b-f) 3.3 Mortality and YLL Attributable to PM2.5 exposure Estimates of mortality attributable to PM2.5 exposure due to all emissions and power generation emissions in China and India are listed in Table 2. The 95% uncertainty confidence intervals (CIs) are calculated based only on the uncertainties of IER curve parameters. Other sources of uncertainty, such as the uncertainty in air quality modeling (emissions, model setup, etc.), 15 population datasets and the fundamental uncertainties inherent in applying a concentration- response function developed largely on the basis of epidemiology conducted in the US and Western Europe at much lower concentrations than those now prevalent in China and India to populations with a different genetic makeup, access to health care, diet and so forth to predict risks in China and India (Dockery and Evans, 2017), are not included. Total ambient PM2.5 concentrations resulting from all sources would have led to 1331.1 (95% Confidence Interval: 824.8-1914.6) thousand deaths attributable to PM2.5 exposure in China in 2013. Large fractions of the effect come from stoke, IHD and COPD diseases. Table 2 Estimated mortality attributable to PM2.5 concentrations due to all sources and power plant emissions over China in 2013 (95% uncertainty confidential intervals are based on IER parameters) (thousand) China Stoke IHD COPD LC LRI Total All sources 365.7 (203.7- 542.0) 388.1 (227.6- 606.0) 345.4 (225.1- 477.1) 150.3 (106.0- 190.5) 81.7 (62.5- 99.0) 1331.1 (824.8- 1914.6) Power Plant 144.4 (81.0- 213.6) 152.8 (90.2- 238.9) 130.9 (86.0- 180.5) 59.8 (42.4- 75.5) 32.1 (24.6- 38.7) 520.0 (324.3- 747.3) India Stoke IHD COPD LC LRI Total All sources 123.2 (64.4- 185.6) 191.4 (106.3- 295.5) 300.9 (184.5- 419.3) 18.0 (12.0- 23.8) 170.4 (126.3- 211.0) 803.8 (493.3- 1135.2) Power Plant 41.0 (21.6- 61.6) 63.4 (35.4- 98.1) 100.4 (62.0- 139.7) 6.1 (4.0-8.0) 57.0 (42.5- 70.3) 267.9 (165.6- 377.6) 16 Power plants are responsible for approximately 39% of ambient PM2.5 across China, and therefore are responsible for this share of the mortality attributable to PM2.5 exposure -- some 500 thousand annual deaths (CI: 320 to 750 thousand deaths). However, because of the non-linearity in the IER, replacing all traditional coal-fired power generation in China with clean energy sources would be expected to reduce the mortality attributable to PM2.5 by a fraction less than this value (Lelieveld et al., 2015). Of course, if such a strategy were coupled with other aggressive air pollution controls, the mortality benefits of replacing power plants with clean energy could be significantly larger. In India, ambient PM2.5 concentrations resulting from all sources are projected to be responsible for 803.8 (493.3-1135.2) thousand premature deaths. Specifically, COPD contributes about 37.4%, IHD is responsible for 23.8%, LRI contributes about 21.2%, stroke (both ischemic and hemorrhagic) contributes about 15.3%, with a negligible contribution from LC. Emissions from power generation account for about 60% of total SO2 emissions in India, but their influence on ambient PM2.5 concentrations (accounting for only 32% of PM2.5 mass) is lower than in China (where power plant emissions account for 39% of PM2.5 mass). Provincial health impacts attributable to PM2.5 exposure depend on both population and ambient PM2.5 concentrations. In China, the Beijing-Tianjin-Hebei (BTH), the Yangtze River Delta (YRD), the Pearl River Delta (PRD), and the Sichuan Basin (SCB) are densely populated (Figure S6). In India, the IGP and east India are densely populated. As shown in Figure 2(a), the BTH (and Shandong, Henan provinces), YRD and the SCB regions in China, and the IGP region in India exhibit the highest PM2.5 exposure. Figure 3 (a, d) shows the distributions of mortality attributable to PM2.5 exposure by province, in China and India. Shandong, Henan and Sichuan provinces in China, and Uttar Pradesh state in India exhibit the largest mortality impacts of PM2.5. When attention is focused on the impacts of emissions from power generation, Shandong and Henan provinces show the largest impacts in China (Figure 3(b, c)). In India, the largest impacts of power plants are evident in the state of Uttar Pradesh (Figure (e, f)). 17 Figure 3. Maps of provincial-level estimates of mortality attributable to PM2.5 exposure due to all emissions (a), and power plant emissions (b), and provincial rankings of power plants induced mortality (c) in China; (d-f) are for India. Estimates of YLL attributable to PM2.5 exposure due to all emissions and power generation emissions in China and India are summarized in Table 3. In China, the estimated YLL attributable to PM2.5 concentrations due to all sources are about 38.9 million years. In India, the total YLL number attributable to PM2.5 exposure is only slightly lower (32.3 million) than in China, in contrast to the large dissimilarity in mortality. 18 Table 3 Estimated YLL attributable to PM2.5 concentrations due to all sources and power plant emissions over China in 2013 (95% uncertainty intervals are based on IER parameters) (million) China Stoke IHD COPD LC LRI Total All sources 5.6 (3.1- 8.1) 6.0 (3.6- 9.2) 16.3 (10.6- 22.5) 7.1 (5.0- 9.0) 3.9 (3.9- 4.7) 38.9 (25.2- 53.5) POW 2.2 (1.2- 3.2) 2.4 (1.4- 3.6) 6.2 (4.1- 8.5) 2.8 (2.0- 3.6) 1.5 (1.2- 1.8) 15.1 (9.9- 20.7) India Stoke IHD COPD LC LRI Total All sources 1.7 (0.9- 2.5) 2.8 (1.6- 4.3) 17.1 (10.5- 23.9) 1.0 (0.7- 1.4) 9.7 (7.2- 12.0) 32.3 (20.9- 44.1) POW 0.6 (0.3- 0.8) 0.9 (0.5- 1.4) 5.7 (3.5- 7.9) 0.3 (0.2- 0.5) 3.2 (2.4- 4.0) 10.7 (6.9- 14.6) The calculations of YLL involve age specific life expectancy. In this study, we take the average based on the distributions of ages in the populations. Since the birth rate in India is higher than in China, and its population is younger, the remaining life expectancy (𝐿) value for India for all-age disease is higher than for China. This leads to high numbers of YLL attributable to PM2.5 exposure. In India, power generation emissions contribute about 33.1% of the YLL attributable to PM2.5 exposure. Figure 4 shows the spatial distributions of provincial-level estimates of YLL attributable to PM2.5 exposure due to all emissions and power generation emissions in China and India. Similar to the spatial distributions of mortality attributable to PM2.5 exposure, Shandong, Henan and Sichuan provinces in China and Uttar Pradesh state in India exhibit the largest YLL. 19 Figure 4. Maps of provincial-level estimates of YLL attributable to PM2.5 exposure due to all emissions (a), and power plant emissions (b), and provincial rankings of power plants induced YLL (c) in China; (d-f) are for India. 4. Discussion 4.1 Comparing Health Impacts with Other Studies Several previous studies have estimated the mortality attributable to PM2.5 exposure in China and India (GBD MAPS Working Group, 2016; Ghude et al., 2016; Lelieveld et al., 2015; Liu et al., 2016), and our estimates, 1.3 million deaths in China and 0.8 million deaths in India in 2013, are consistent with the findings of these previous studies. When viewed in the context of the likely uncertainty in any of these estimates, the differences among the many estimates are relatively minor – especially considering that the studies used a variety of emissions estimates, atmospheric fate and transport models, and (to some extent) different syntheses of the epidemiological literature linking PM2.5 exposure to mortality. It may be of interest that although the estimates of the total mortality impact of PM2.5 exposure are similar across all studies, there are differences in the relative importance of various causes of death. 20 The more recent studies, which rely on the 2015 IER, tend to find greater impacts on COPD disease and smaller impacts on stroke. This is a reflection of differences between the IER 2015 parameter estimates from the 2010 and 2013 versions of the IER. It is also worth noting that the role of ambient PM2.5 in the development of COPD is generally considered to be uncertain (Schikowski et al., 2013). Table 4 Summary of Health Impacts from This Study and Other Studies Author Year Analyzed Sources China India Deaths (million) YLL (million) Deaths (million) YLL (million) Lelieveld et al. (2015) 2010 All 1.4 0.6 Ghude et al. (2016) 2011 All -- 0.6 Liu et al. (2016) 2013 All 0.9 -- This Study 2013 All 1.3 38.9 0.8 32.3 HEI China MAPS, 2017 2015 All 1.1 Even using the same GBD framework (GBD 2015), estimates can differ. Cohen et al. (2017) estimated 1108.1 thousand deaths attributable to PM2.5 (LRI: 66.3, LC: 146.0, IHD: 291.8, COPD: 281.7, and Stroke: 322.3) in China and 1090.4 thousand deaths attributable to PM2.5 (LRI: 200.7, LC: 22.5, IHD: 365.6, COPD: 349.0, Stroke: 152.5) in India in 2015. In our study, the fractions of each disease in all-cause death are similar to Cohen et al. (2017), but the all-cause deaths are higher in China (1331.1 thousand), and lower in India (803.8 thousand). The differences of annual mean PM2.5 concentrations in 2015 and 2013 are likely to represent the major cause for these disparities. From 2013 to 2015, PM2.5 concentrations in China decreased significantly (Clean Air Alliance of China, 2016) because of the strict air pollution control measures taken since 2013. The emission inventory used for India is based on the year 2010 in this study, so it might have led to underestimations of all-cause deaths, and lower results for India than Cohen et al. (2017). 21 Large uncertainties are embodied in the calculations of source sector contributions because of the uncertainties in emission inventories, and different responses to emission perturbations in different atmospheric chemistry models. Lelieveld et al. (2015) concluded that residential energy was dominant in outdoor air pollution in 2010 in both China (32%) and India (50%), while industry accounted for only 8% in China and 7% in India. Both the GBD MAPS study and this study found that industrial sources are the largest sectoral contributor to air pollution in China. The differences result very likely from the dissimilarities in the emission inventories employed, and partially from the differences of atmospheric chemistry models. Lelieveld et al. (2015) used the Emission Database for Global Atmospheric Research (EDGAR), and this study used a state-of-the-art emission inventory for Asia (MIX, Li et al., 2017). Table S4 shows the comparison between these two datasets and shows that NOx non-methane volatile organic compounds (NMVOC) and NH3 emissions differ greatly, which might have led to the dominant role of the agriculture sector in the results of Lelieveld et al. (2015). EDGAR ignores primary PM2.5 emissions in the simulation, which could show an even larger contribution to PM2.5 than BC and OC, leading to reduced contributions from sectors with high primary PM2.5 emissions. Besides, the coarser global model resolution (~100km×100km) used in Lelieveld et al. (2015) might have missed mesoscale information important for assessments on a regional scale. 4.2 Limitations Although a complex regional meteorology-chemistry model, a state-of-the-art emission inventory in Asia, and more accurate IER parameters were used in this study compared to previous studies, there are still a number of limitations, involving each step in the calculation procedures. We used relatively high model resolution to represent regional pollution features (60 km) compared to ~100km in Lelieveld et al. (2015), but 60km is still not good for precipitation simulations, and detailed urban and sub-grid information might have been missed. In this study, we need to run six one-year simulations (Table S2), and used 60 km due to computational limitations. Besides, we did not include health impacts resulting from ozone exposure, mainly because the mortality due to ozone exposure is small compared to that from PM2.5 (Ghude et al., 2016). Second, although thorough and detailed energy use information was used in the development of the emission inventories, there is still a great deal of uncertainty related to emission estimates from power plants and other sectors. The uncertainties for different species differ greatly, and 22 uncertainties for particles are much larger than those for gases. For example, the uncertainties of BC and OC emissions in MIX are ±208% and ±258% in China, but uncertainties of SO2 and NOx are only ±12% and ±31% (Li et al., 2017). The uncertainties in emissions will be propagated into the meteorology-chemistry modeling. In terms of the source attribution to power plants, the uncertainties are relatively low because of low uncertainties in gases and the importance of gases in power plant emissions. In addition, SO2 and NOx emissions from power plants were constrained using satellite retrievals (Streets et al., 2013). Compared to the sector contributions in Lelieveld et al. (2015), we found that different inventories can lead to large differences. Thus we used the most advanced emission estimates (MIX, Li et al., 2017) in this study. Third, uncertainties also arise from chemistry-meteorology modeling, related to chemical reactions, atmospheric dispersion, and deposition. Although we use nationwide surface PM2.5 measurements to evaluate model performance, comparisons of other air pollutants, including SO2, NOx, VOCs, and ozone, were not presented due to inaccessibility of relevant measurements. Yet the WRF- Chem performance was comprehensively evaluated in China and India in many previous studies (Gao et al., 2015, 2016a, 2016b, 2016c, 2017; Marrapu et al., 2014), and the results were encouraging, increasing confidence to our findings. Atmospheric chemistry modeling is the only approach to examine sector-specific contributions to exposure. Errors can be reduced further with advances in modeling approaches. Fourth, the GBD 2015 IER parameters used in this study represent only the current best understanding, which will be enhanced with regular GBD updates in the future. For example, the IERs were developed based on observed cohort studies in the US, Canada, and west Europe (Cohen et al., 2017), implying large uncertainties when applied to China and India. In addition, the province-specific health benefits do not quantify the effect of inter-province transport due to computational complexity, which may be redressed in the future using source apportionment techniques embodied in atmospheric chemistry models. 5. Conclusion We estimate province-specific mortality and YLL attributable to ambient PM2.5 exposure, and examine the changes in ambient PM2.5 and the health benefits expected to flow from eliminating power plant emissions, by combining atmospheric modeling and health impact analyses. Several 23 previous studies have evaluated the global mortality attributable to PM2.5 exposure, but few have as yet explored the contributions of source classes. The mortality burdens attributable to PM2.5 are estimated to be 1.3 million in China and 0.8 million in India, which are consistent with previous studies. We further quantified the impact of power generation emissions and found that power generation emissions contribute to 0.5 million death in China and 0.3 million in India. We estimated also that 15 million (95% Confidence Interval (CI): 10 to 21 million) years of life lost can be avoided in China each year and 11 million (95% CI: 7 to 15 million) in India by eliminating power generation emissions. The spatial distribution of these results reveals that priorities in upgrading existing power generating technologies should be given to Shandong, Henan, and Sichuan provinces in China, and Uttar Pradesh state in India due to their dominant contributions to the current health risks. Acknowledgements We would like to acknowledge Professor John Evans at Harvard School of Public Health for insightful discussion and helpful comments; we thank Dr. Aaron J. Cohen and Dr. Richard T. Burnett for providing the latest version IER parameters. We are grateful also to the China National Environmental Monitoring Center (CNEMC) and the Modelling of Atmospheric Pollution and Networking (MAPAN) groups for maintaining the high quality measurements of air pollutants in China and India. This study was supported by the Harvard Global Institute (HGI). References Archer-Nicholls S, Carter E, Kumar R, Xiao Q, Liu Y, Frostad J, et al. 2016. The regional impacts of cooking and heating emissions on ambient air quality and disease burden in China. Environ. Sci. Technol. 50:9416–9423; doi:10.1021/acs.est.6b02533. Bhalla, K., M. Shotten, A. Cohen, M. Brauer, S. Shahraz, R. Burnett, K. Leach-Kemon, G. Freedman, and C. J. Murray. Transport for health: the global burden of disease from motorized road transport. 2014. Buonocore JJ, Dong X, Spengler JD, Fu JS, Levy JI. 2014. Using the Community Multiscale Air Quality (CMAQ) model to estimate public health impacts of PM2.5 from individual power plants. Environ. Int. 68:200–208; doi:10.1016/j.envint.2014.03.031. 24 Burnett RT, Pope CA, Ezzati M, Olives C, Lim SS, Mehta S. 2014. An Integrated Risk Function for Estimating the Global Burden of Disease Attributable to Ambient Fine Particulate Matter Exposure. Environ. Health Perspect. 3: 397–404. Chafe ZA, Brauer M, Klimont Z, Van Dingenen R, Mehta S, Rao S, et al. 2015. Household cooking with solid fuels contributes to ambient PM2.5air pollution and the burden of disease. Environ. Health Perspect. 122:1314–1320; doi:10.1289/ehp.1206340. Chou M-D, Suarez MJ, Ho C-H, Yan MM-H, Lee K-T. 1998. Parameterizations for Cloud Overlapping and Shortwave Single-Scattering Properties for Use in General Circulation and Cloud Ensemble Models. J. Clim. 11: 202–214. Clean Air Alliance of China. 2016. CAAC Clean Air Management Report (2016). Cohen AJ, Brauer M, Burnett R, Anderson HR, Frostad J, Estep K, et al. 2017. Estimates and 25- year trends of the global burden of disease attributable to ambient air pollution: An analysis of data from the Global Burden of Diseases Study 2015. Lancet 389:1907–1918; doi:10.1016/S0140-6736(17)30505-6. Dockery, D.W., Pope, C.A., Xu, X., Spengler, J.D., Ware, J.H., Fay, M.E., Ferris Jr, B.G. and Speizer, F.E., 1993. An association between air pollution and mortality in six US cities. New England journal of medicine, 329(24), pp.1753-1759. Dockery, D.W. and Evans, J.S., 2017. Tallying the bills of mortality from air pollution. The Lancet, 389(10082), pp.1862-1864. Emmons LK, Walters S, Hess PG, Lamarque J-F, Pfister GG, Fillmore D, et al. 2010. Description and evaluation of the Model for Ozone and Related chemical Tracers, version 4 (MOZART-4). Geosci. Model Dev. 3:43–67; doi:10.5194/gmd-3-43-2010. Fann N, Fulcher CM, Baker K. 2013. The recent and future health burden of air pollution apportioned across U.S. sectors. Environ. Sci. Technol. 47:3580–3589; doi:10.1021/es304831q. Gao M, Carmichael GR, Wang Y, Saide PE, Yu M, Xin J, et al. 2016. Modeling study of the 2010 regional haze event in the North China Plain. Atmos. Chem. Phys. 16:1673–1691; doi:10.5194/acp-16-1673-2016. Gao M, Carmichael GR, Saide PE, Lu Z, Yu M, Streets DG, et al. 2016. Response of winter fine particulate matter concentrations to emission and meteorology changes in North China. Atmos. Chem. Phys. Discuss. 1:1–38; doi:10.5194/acp-2016-429. Gao M, Carmichael GR, Wang Y, Ji D, Liu Z, Wang Z. 2016. Improving simulations of sulfate aerosols during winter haze over Northern China: the impacts of heterogeneous oxidation by NO2. Front. Environ. Sci. Eng. 10:1–11; doi:10.1007/s11783-016-0878-2. 25 Gao M, Guttikunda SK, Carmichael GR, Wang Y, Liu Z, Stanier CO, et al. 2015. Health impacts and economic losses assessment of the 2013 severe haze event in Beijing area. Sci. Total Environ. 511C:553–561; doi:10.1016/j.scitotenv.2015.01.005. Gao M, Saide PE, Xin J, Wang Y, Liu Z, Wang Y, et al. 2017. Estimates of Health Impacts and Radiative Forcing in Winter Haze in Eastern China through Constraints of Surface PM 2.5 Predictions. Environ. Sci. Technol. 51:2178–2185; doi:10.1021/acs.est.6b03745. GBD MAPS Working Group. 2016. Burden of Disease Attributable to Coal-Burning and Other Air Pollution Sources in China | Health Effects Institute. Ghude SD, Chate DM, Jena C, Beig G, Kumar R, Barth MC, et al. 2016. Premature mortality in India due to PM 2.5 and ozone exposure. Geophys. Res. Lett. 43:4650–4658; doi:10.1002/2016GL068949. Grell G a., Peckham SE, Schmitz R, McKeen S a., Frost G, Skamarock WC, et al. 2005. Fully coupled “online” chemistry within the WRF model. Atmos. Environ. 39:6957–6975; doi:10.1016/j.atmosenv.2005.04.027. Guenther AB, Jiang X, Heald CL, Sakulyanontvittaya T, Duhl T, Emmons LK, et al. 2012. The model of emissions of gases and aerosols from nature version 2.1 (MEGAN2.1): An extended and updated framework for modeling biogenic emissions. Geosci. Model Dev. 5:1471–1492; doi:10.5194/gmd-5-1471-2012. Gufran Beig, D.M. Chate, S. K. Sahu, N. S. Parkhi, R. Srinivas, K. Ali, S. D. Ghude, S. Yadav and H. K. Trimbake, System of Air Quality Forecasting and Research (SAFAR-India), GAW Report No. 217, World Meteorological Organization, Global Atmosphere Watch, Geneva, Switzerland, 2015. Hong, Song-You; Noh, Yign; Dudhia J. 2006. A New Vertical Diffusion Package with an Explicit Treatment of. Mon. Weather Rev. 134: 2318–2341. Hu, J., Huang, L., Chen, M., Zhang, H., Wang, S., and Ying, Q.: Premature Mortality Attributable to Particulate Matter in China: Source Contributions and Responses to Reductions, Environmental Science & Technology, Sumbitted, 2017a. Hu, J., Huang, L., Chen, M., Liao, H., Zhang, H., Wang, S., Zhang, Q. and Ying, Q., 2017. Premature Mortality Attributable to Particulate Matter in China: Source Contributions and Responses to Reductions. Environmental Science & Technology. Huang, L., Hu, J., Chen, M. and Zhang, H., 2017. Impacts of power generation on air quality in China—part I: an overview. Resources, Conservation and Recycling, 121, pp.103-114. Huang R-J, Zhang Y, Bozzetti C, Ho K-F, Cao J-J, Han Y, et al. 2014. High secondary aerosol contribution to particulate pollution during haze events in China. Nature; doi:10.1038/nature13774. 26 Huang R-J, Zhang Y, Bozzetti C, Ho K-F, Cao J-J, Han Y, et al. 2014. High secondary aerosol contribution to particulate pollution during haze events in China. Nature; doi:10.1038/nature13774. Hoek, G., Krishnan, R.M., Beelen, R., Peters, A., Ostro, B., Brunekreef, B. and Kaufman, J.D., 2013. Long-term air pollution exposure and cardio-respiratory mortality: a review. Environmental Health, 12(1), p.43. Jacobson, D. and High, C., 2008. Wind energy and air emission reduction benefits: A primer (No. NREL/SR-500-42616). National Renewable Energy Laboratory (NREL), Golden, CO.. Kurokawa J, Ohara T, Morikawa T, Hanayama S, Janssens-Maenhout G, Fukui T, et al. 2013. Emissions of air pollutants and greenhouse gases over Asian regions during 2000-2008: Regional Emission inventory in ASia (REAS) version 2. Atmos. Chem. Phys. 13:11019– 11058; doi:10.5194/acp-13-11019-2013. Lelieveld J, Evans JS, Fnais M, Giannadaki D, Pozzer A. 2015. The contribution of outdoor air pollution sources to premature mortality on a global scale. Nature; doi:10.1038/nature15371. Levy JI, Baxter LK, Schwartz J. 2009. Uncertainty and variability in health-related damages from coal-fired power plants in the United States. Risk Anal. 29:1000–1014; doi:10.1111/j.1539-6924.2009.01227.x. Levy JI, Spengler JD. 2002. Modeling the Benefits of Power Plant Emission Controls in Massachusetts. J. Air Waste Manage. Assoc. 52:5–18; doi:10.1080/10473289.2002.10470753. Levy JI, Spengler JD, Hlinka D, Sullivan D, Moon D. 2002. Using CALPUFF to evaluate the impacts of power plant emissions in Illinois: Model sensitivity and implications. Atmos. Environ. 36:1063–1075; doi:10.1016/S1352-2310(01)00493-9. Li M, Zhang Q, Kurokawa JI, Woo JH, He K, Lu Z, et al. 2017. MIX: A mosaic Asian anthropogenic emission inventory under the international collaboration framework of the MICS-Asia and HTAP. Atmos. Chem. Phys. 17:935–963; doi:10.5194/acp-17-935-2017. Lim SS, Vos T, Flaxman AD, Danaei G, Shibuya K, Adair-Rohani H, et al. 2012. A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990-2010: A systematic analysis for the Global Burden of Disease Study 2010. Lancet 380:2224–2260; doi:10.1016/S0140-6736(12)61766-8. Liu J, Han Y, Tang X, Zhu J, Zhu T. 2016. Estimating adult mortality attributable to PM2.5 exposure in China with assimilated PM2.5 concentrations based on a ground monitoring network. Sci. Total Environ. 568:1253–1262; doi:10.1016/j.scitotenv.2016.05.165. 27 Lin Yuh-Lang;Farley, Richard D.;Orville HD. 1983. Bulk Parameterization of the Snow Field in a Cloud Model. J. Clim. Appl. Meteorol. 22: 1065–1092. Lu Z, Zhang Q, Streets DG. 2011. Sulfur dioxide and primary carbonaceous aerosol emissions in China and India, 1996-2010. Atmos. Chem. Phys. 11:9839–9864; doi:10.5194/acp-11-9839- 2011. Marrapu P, Cheng Y, Beig G, Sahu S, Srinivas R, Carmichael GR. 2014. Air quality in Delhi during the Commonwealth Games. Atmos. Chem. Phys. 14:10619–10630; doi:10.5194/acp- 14-10619-2014. Mlawer EJ, Taubman SJ, Brown PD, Iacono MJ, Clough S a. 1997. Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res. 102:16663; doi:10.1029/97JD00237. Murray CJL, Vos T, Lozano R, Naghavi M, Flaxman AD, Michaud C, et al. 2012. Disability- adjusted life years (DALYs) for 291 diseases and injuries in 21 regions, 1990-2010: A systematic analysis for the Global Burden of Disease Study 2010. Lancet 380:2197–2223; doi:10.1016/S0140-6736(12)61689-4. Pope III CA, Burnett RT, Thun MJ, Calle EE, Krewski D, Thurston GD. 2002. to Fine Particulate Air Pollution. J. Am. Med. Assoc. 287:1132–1141; doi:10.1001/jama.287.9.1132. Pope CA, Ezzati M, Dockery DW. 2009. Fine-particulate air pollution and life expectancy in the United States. N. Engl. J. Med. 360:376–386; doi:10.1056/NEJMsa0805646. Randerson, J.T., G.R. van der Werf, L. Giglio, G.J. Collatz, and P.S. Kasibhatla. 2015. Global Fire Emissions Database, Version 4, (GFEDv4). ORNL DAAC, Oak Ridge, Tennessee, USA. https://doi.org/10.3334/ORNLDAAC/1293 Schikowski, T., Mills, I.C., Anderson, H.R., Cohen, A., Hansell, A., Kauffmann, F., Krämer, U., Marcon, A., Perez, L., Sunyer, J. and Probst-Hensch, N., 2013. Ambient air pollution-a cause for COPD?. European Respiratory Journal, pp.erj01001-2012. Singh N, Murari V, Kumar M, Barman SC, Banerjee T. 2017. Fine particulates over South Asia: Review and meta-analysis of PM2.5 source apportionment through receptor model. Environ. Pollut. 223:121–136; doi:10.1016/j.envpol.2016.12.071. Streets DG, Canty T, Carmichael GR, De Foy B, Dickerson RR, Duncan BN, et al. 2013. Emissions estimation from satellite retrievals: A review of current capability. Atmos. Environ. 77:1011–1042; doi:10.1016/j.atmosenv.2013.05.051 Review. Van Aardenne, J., Dentener, F., Van Dingenen, R., Janssens-Maenhout, G., Marmer, E., Vignati, E., Russ, H.P., Szabo, L. and Raes, F., 2010. Climate and air quality impacts of combined climate change and air pollution policy scenarios(No. JRC61281). Joint Research Centre https://doi.org/10.3334/ORNLDAAC/1293 28 (Seville site). Xie R, Sabel CE, Lu X, Zhu W, Kan H, Nielsen CP, et al. 2016. Long-term trend and spatial pattern of PM2.5 induced premature mortality in China. Environ. Int. 97:180–186; doi:10.1016/j.envint.2016.09.003. Zaveri R a., Easter RC, Fast JD, Peters LK. 2008. Model for Simulating Aerosol Interactions and Chemistry (MOSAIC). J. Geophys. Res. 113:D13204; doi:10.1029/2007JD008782. Zaveri R a., Peters LK. 1999. A new lumped structure photochemical mechanism for large-scale applications. J. Geophys. Res. 104:30387; doi:10.1029/1999JD900876. 29 Supporting Information Table S1 Model Configuration Options Configurations Descriptions Horizontal Resolutions 60km Vertical layers 27 Cloud Microphysics Lin scheme Longwave Radiation Rapid Radiative Transfer Model (RRTM) Shortwave Radiation Goddard shortwave Land Surface Model Noah Planetary Boundary Layer Yonsei University MOSAIC Aerosol Bins 0.039–0.078µm, 0.078-0.156µm, 0.156–0.312µm, 0.312– 0.625µm, 0.625–1.25µm, 1.25–2.5µm, 2.5– 5.0µm, 5.0– 10µm Table S2 Simulations and descriptions Cases Descriptions BASE Anthropogenic emissions from all sectors, biogenic emissions, biomass burning emissions are all included noIND Same as BASE except anthropogenic emissions from industry are excluded noPOW Same as BASE except anthropogenic emissions from power plant are excluded noAGR Same as BASE except agriculture emissions are excluded noRES Same as BASE except anthropogenic emissions from residential sector are excluded noTRA Same as BASE except anthropogenic emissions from transportation are excluded noBB Same as BASE except biomass burning emissions are excluded 30 Table S3 Life Expectation at Different Ages in China and India China Male Female India Male Female <1 year 74.1 77.2 <1 year 66.2 69.1 1-4 years 74 76.9 1-4 years 68 71.1 5-9 years 70.1 73.1 5-9 years 64.7 68 10-14 years 65.2 68.2 10-14 years 60 63.4 15-19 years 60.3 63.3 15-19 years 55.2 58.6 20-24 years 55.4 58.4 20-24 years 50.5 54 25-29 years 50.6 53.5 25-29 years 45.9 49.4 30-34 years 45.8 48.6 30-34 years 41.4 44.8 35-39 years 41 43.8 35-39 years 37 40.2 40-44 years 36.3 39 40-44 years 32.7 35.6 45-49 years 31.6 34.2 45-49 years 28.5 31.1 50-54 years 27 29.6 50-54 years 24.4 26.7 55-59 years 22.5 25 55-59 years 20.6 22.5 60-64 years 18.3 20.6 60-64 years 17 18.5 65-69 years 14.6 16.6 65-69 years 13.8 15 70-74 years 11.3 13 70-74 years 11 11.9 75-79 years 8.6 9.9 75-79 years 8.7 9.4 80-84 years 6.6 7.4 80-84 years 6.9 7.3 85-89 years 4.9 5.5 85-89 years 5.3 5.6 90-94 years 3.7 4.2 90-94 years 4.1 4.2 95-99 years 2.9 3.2 95-99 years 3.2 3.2 100+ years 2.4 2.8 100+ years 2.5 2.5 31 Table S4 Comparison between the EDGAR 2010 inventory and the MIX 2010 inventory for China and India Species (Tg/year) EDGAR China EDGAR India MIX China MIX India SO2 29.19 9.31 28.66 9.26 BC 1.56 0.71 1.77 1.02 OC 3.87 2.24 3.39 2.53 NOx 15.20 5.35 29.07 9.57 NH3 13.97 6.85 9.80 9.87 NMVOC 14.49 3.26 23.62 16.89 CO 160.85 76.04 170.87 67.42 PM2.5 NA NA 12.20 5.22 PM10 NA NA 16.62 7.09 32 Figure S1. WRF-Chem modeling domain, terrain height, and locations of CNEMC and MAPAN measurement sites 33 Figure S2. Simulated and observed monthly PM2.5 concentrations in the individual sites in China Figure S3. Simulated and observed monthly PM2.5 concentrations in the individual sites in India 34 Figure S4. Sector contributions to SO2, NOx and primary PM2.5 emissions in China and India Figure S5. sectoral contributions to WRF-Chem simulated national mean PM2.5 concentrations in China (a) and India (b) 35 Figure S6. Spatial distributions of population in China and India