In this study, we conducted one standard simulation and four sensitivity simulations for ground-level PM2.5 using the nested grid capability of GEOS-Chem for eastern Asia. The simulation scenarios are summarized in Table 2. In the standard simulation, we use the emissions for the year 2013 that are discussed in Sect. 2. To select the year of meteorology, we conducted standard simulation using the same emissions and different meteorology from the years 2010 to 2012 because the meteorological fields are not available for the whole year of 2013. We chose the year 2012 as our meteorological year, with which the simulation results best represented the mean PM2.5 concentration from 2010 to 2012.

In sensitivity scenarios, we removed emissions from coal combustion in different sectors. In sensitivity scenario 1, we removed emissions from coal burning from all energy sectors (scenario for total coal burning, TC). In sensitivity scenarios 2 to 4, we respectively shut down emissions from total coal burning in power plants, industries, and domestic sectors (TCP, TCI, and TCD). All the meteorology used in the sensitivity simulation was the same as the standard simulation. Used as spin-up were the 3 months before each simulation year. The differences between standard and sensitivity simulations are used to represent the contributions from coal combustion in each sector.

GEOS-Chem is a global chemical transport model that has been widely applied to study PM2.5 over China (e.g., Brauer et al., 2012, 2015; Jiang et al., 2015; Kharol et al., 2013; van Donkelaar et al., 2010, 2015; Wang et al., 2013; Y. Wang et., 2014; Xu et al., 2015; L. Zhang et al., 2015; Q. Q. Zhang et al., 2015). The model is driven by assimilated meteorological data from the NASA Goddard Earth Observing System (GEOS), including winds, temperature, clouds, precipitation, and other surface properties. GEOS-Chem (version 9-01-03) includes detailed HOX–NOx–VOC–ozone–BrOX tropospheric chemistry originally described by Bey et al. (2001), with the addition of BrOX chemistry by Parrella et al. (2012). Aerosol simulation is fully coupled with gas-phase chemistry, including sulfate (SO42-), nitrate (NO3-), ammonium (NH4+) (Park et al., 2004; Pye et al., 2009), OC, BC (Park et al., 2003), sea salt (Alexander et al., 2005), and mineral dust (Fairlie et al., 2007). The aerosol thermodynamic equilibriums use the ISORROPIA II model (Fountoukis and Nenes, 2007) to calculate the partitioning of nitric acid and ammonia between gas and aerosol phases. The formation of secondary organic aerosol (SOA) includes the oxidation of isoprene (Henze and Seinfeld, 2006), monoterpenes, aromatics (Henze et al., 2008), and other reactive VOCs (Liao et al., 2007). In addition, we corrected errors in the model representation of too-shallow nighttime mixing depth following Walker et al. (2012) and introduced the production mechanism of sulfate on aerosol surface described in Wang et al. (2013). Aerosols interact with gas-phase chemistry in GEOS-Chem through the effect of aerosol extinction on photolysis rates (Martin et al., 2003) and heterogeneous chemistry (Jacob, 2000).

In this study, we conducted simulations for ground-level PM2.5 using the nested grid capability of GEOS-Chem for eastern Asia, which was originally described by Wang et al. (2004) and Chen et al. (2009). The nested domain for eastern Asia covers an area spanning from 70∘ to 150∘ E and from 11∘ S to 55∘ N, with a horizontal resolution of 0.5∘ latitude by 0.667∘ longitude. The boundary fields are provided by the global GEOS-Chem simulation, with a resolution of four latitudes by five longitudes, and are updated every 3 h. We assume that the organic mass / organic carbon ratio is 1.8 and relative humidity is 50 % for PM2.5 in China.

The global simulations use emissions from the Global Emission Inventory Activity (GEIA) (Benkovitz et al., 1996), which is overwritten by the NEI05, EMEP, and INTEX-B inventories (Zhang et al., 2009) over the US, Europe, and eastern Asia, respectively. The CO emission we used in this study is from EDGAR v3, which is also overwritten by INTEX-B in the nested domain of eastern Asia. In the nested-grid simulation for eastern Asia, we use the emissions for the year 2013 (as discussed in Sect. 2) over China, with emissions over the rest of eastern Asia taken from the INTEX-B emission inventory. In addition, the simulation also includes open fire emissions from the GFED3 inventory (Giglio et al., 2010; van der Werf et al., 2010; Mu et al., 2011), lightning NOx emissions calculated with the algorithm of Price and Rind (1992), and volcanic SO2 emissions from the AEROCOM database (http://aerocom.met.no/download/emissions/AEROCOM_HC/volc/) implemented by Fisher et al. (2011).

The GEOS-Chem model is driven by assimilated meteorological data from the NASA GEOS. Y. Wang et al. (2014) evaluated the important meteorological factors that are relevant to particle formation in the model, including temperature, relative humidity (RH), wind speed, and direction, using observation data from the National Meteorological Center (NMC) of China. It reported good spatial and temporal correlations with observed temperature, RH, and wind direction. The correlation of wind speed, however, was poorer as the model tends to overestimate in low speed conditions.

In this study, we conducted model evaluation using the surface PM2.5 observation network of the China National Environmental Monitoring Center (CNEMC, http://106.37.208.233:20035). This monitoring program was initiated in January 2013, covering 74 major cities in China. Figure 1 compares simulated annual mean PM2.5 concentrations with those observed in 74 major cities in China for the year 2013. As shown in Fig. 1a, the simulated ambient PM2.5 concentration has a clear regional distribution with high values in the Sichuan Basin (SCB), North China Plain (NC), and middle Yangtze River area (MYR). The highest concentration occurs in the Sichuan Basin with an average value of 73.5 µgm-3. Concentrations in the abovementioned severely polluted regions are generally above 60 µgm-3. The observation data are compared with the concentrations in the grids where the city centers are located. The comparison shows that the model reproduces the spatial distribution well with a normalized mean bias (NMB) of -16.3 %. The correlation coefficient for annual mean concentration is 0.68. The slight underestimate mainly appears in the heavily polluted area in the NC region where observations are largely influenced by local emissions; however, current simulation cannot capture it due to relatively coarse resolution (H. Zhang et al., 2012). Figure 2 shows comparisons between simulated and observed seasonal mean concentrations. PM2.5 concentration has an obvious seasonal variation, with the highest value in winter and the lowest in summer, which is correctly reproduced by the model. The largest bias occurred in winter with the value of -23.3 %. The inconsistency of meteorology also partly accounts for the underestimate because the meteorological condition was more unfavorable in January 2013. Y. Wang et al. (2014) conducted simulations for January in 2012 and 2013 using the same emissions and found that the ground PM2.5 concentration was 27 % higher in January 2013 than that in 2012. The model performs better in the other three seasons, with biases between -13.3 and -10.8 %. Correlation maps for each season are shown in Fig. 3. The PM2.5 concentration in winter is more spread out in coordinates as it varies substantially across China, which has a larger correlation coefficient of 0.71. In other seasons, the correlation coefficients are around 0.6.

We also evaluated the monthly variation using averaged monthly mean concentrations in cities in each key region since analyses and discussions mainly focused on these six areas. The six key regions are shown with frames in Fig. 1a, which includes Northeast China (NEC; 123–128∘ E, 41–47∘ N), North China (NC; 113–119∘ E, 33–40∘ N), Yangtze River Delta (YRD; 119–122∘ E, 29.5–32.5∘ N), Middle Yangtze River (MYR; 111–115∘ E, 27–32.5∘ N), Sichuan Basin (SCB; 103–107∘ E, 28–32∘ N), and Pearl River Delta (PRD; 112–114∘ E, 22–24∘ N). Cities in each region share the similar weather conditions, terrain, and pollution levels. As shown in Fig. 4, the model generally reproduces the monthly variation well. The NMB ranges from -45 to 1 %, and the correlation coefficient varies between 0.7 and 0.94. The model performance is better in the MYR, SCB, and PRD than in NC, NEC, and the YRD. The large discrepancy is mainly due to the failure to capture the extremely high concentration in wintertime. The normalized mean errors (NMEs) of simulated PM2.5 concentrations in NEC, NC, and the YRD regions are estimated to be 38, 45, and 36 %, which is the same as the values of the NMB since the model underestimated the PM2.5 concentration throughout the year. In the MYR, SCB, and PRD regions, the NMEs are estimated to be 18, 21, and 22 %, which are higher than the estimated the NMB, especially in the SCB. Overall, the model can reproduce the monthly variation of ambient PM2.5 concentration in these key regions.

The PM2.5 composition shows a great diversity across China. Sulfate–nitrate–ammonium (SNA), BC, organic matter (OM), and crustal material constituted 7.1–57 %, 1.3–12.8 %, 17.7–53 %, and 7.1–43 %, respectively, in PM2.5 mass in China, and the fractions of SNA in PM2.5 (40–57 %) are much higher in eastern China (Yang et al., 2011). OM and mineral dust also play significant roles in PM2.5 concentration. PM2.5 speciation in China simulated by GEOS-Chem has been evaluated in some previous studies. Wang et al. (2013) reported annual biases of -10, +31, and +35 % for sulfate, nitrate, and ammonia, respectively, compared with observations at 22 sites in eastern Asia. Fu et al. (2012) indicated that annual mean BC and OC concentrations in rural and background sites were underestimated by 56 and 75 %. PM2.5 speciation is also evaluated in this study using the observed concentration of aerosol compositions averaged from 2012 to 2013 in 12 cities across China (X. Y. Zhang et al., 2015), as shown in Fig. 5. The information of each site is described in detail in X. Y. Zhang et al. (2012). The underestimate of sulfate mainly occurs in the two cities of Zhengzhou and Xi'an, two orange spots in central and northern China, as these two sites are located in an urban area. Nitrate and ammonia are overestimated by around 20 %, which is a common issue in most chemistry transport models (CTMs.) OC is underestimated by 28.9 % due to the incomplete mechanism of SOA simulation. The NME is calculated between 30 and 41 %. The correlation coefficients range between 0.44 and 0.78.

Figure 6 shows the spatial distribution of annual mean source contributions from coal burning. As shown in Fig. 6a, the contribution from total coal burning has a similar spatial distribution with the annual mean PM2.5 concentration, which indicates the large influence of coal burning on air quality. Table 3 also shows a higher percentage contribution in areas with higher PM2.5 concentrations such as the NC, MYR, and SCB regions. The national average contribution from total coal burning, which is an average of concentrations in 74 major cities, is up to 22.5 µgm-3, accounting for almost 40 % of the total PM2.5 concentration. In the six key regions, coal burning contributes 34.5–50.2 % of the total ambient PM2.5 concentration. The largest contribution occurs in the SCB, which reaches 36.9 µgm-3 on average due to the dense population, large emissions, and unfavorable terrain that tends to trap the emissions and secondary pollutants in this area. The highest contribution is up to 56.9 µgm-3, occurring in the southwestern city of Chengdu. Following the SCB, coal-burning contributions in the MYR and NC are also above the national average, with average values of 30.8 µgm-3 (45.1 %) and 26 µgm-3 (40.5 %), respectively. Among the six key regions, coal combustion in the PRD shows the smallest contribution of 12.6 µgm-3, yet still accounting for 35 % of the local PM2.5 concentration. In addition to the key regions, coal burning contributes to around 25 µgm-3 (more than 50 %) of the local PM2.5 in cities like Baotou and Hohhot in Inner Mongolia, an autonomous region near the middle northern border, as it is one of the largest production areas of coal and a large amount of raw coal is burnt for energy supply. In the northwestern city of Ürümqi, coal burning is also a large contributor that accounts for around 40 % of the local PM2.5 concentration as there are no other large anthropogenic sources of air pollutants there.

Among all the subsectors in coal combustion, industrial coal burning is the most significant contributor, followed by coal burning in power plants and the domestic sector, which is shown in Fig. 6b–d and Table 3. The contribution from industrial coal burning is up to 9.6 µgm-3 (17 %) on national average (average of 74 major cities), while those from coal burning in power plants and the domestic sector are 5.6 µgm-3 (9.8 %) and 2.2 µgm-3 (4 %), respectively. The contribution from each sector differs in different regions. Contributions from coal burning in power plants and industry have similar spatial distributions to the annual mean PM2.5 concentration. As shown in Fig. 6b, coal burning in power plants has the largest contribution in NC, with the highest value of 13.1 µgm-3 (15 %) and an average of 7.7 µgm-3 (12 %), due to the large number of power plants in this area. The smallest contribution occurs in the PRD with the value of only 2.7 µgm-3 (7.5 %). In most key areas in China, coal burning in the power sector contributes to around 10 % of the local PM2.5 concentration, which is a relatively minor source compared with industry due to higher energy efficiency and more stringent emission control policies in power sectors. Industrial coal burning, as shown in Fig. 6c, has the largest contribution in the SCB, with an average value of 19 µgm-3 (25.9 %). The largest contribution occurs in the city of Chengdu, which is up to 35.8 µgm-3, accounting for around one-third of the local PM2.5. NC and the MYR are also significantly influenced by industrial coal burning, with contributions of 10.8 µgm-3 (16.8 %) and 14 µgm-3 (20.5 %), respectively. In other areas, including NEC, the YRD, and the PRD, the average contributions of coal burning in the industrial sector are generally less than 10 µgm-3, accounting for around 15 % of the local PM2.5 concentration. As shown in Fig. 6d, domestic coal burning has little contribution to ambient PM2.5 in most areas in the six key regions. However, in some individual regions in Guizhou Province in the southwest and Inner Mongolia in North China, domestic coal burning contributes more than 10 µgm-3, which accounts for more than 15 % in Guizhou and 25 % in Inner Mongolia where people tend to burn more raw coal for heating. In addition, the high sulfur content of coal in Guizhou Province also accounts for the large contribution.

In the nested simulation of eastern Asia, the contributions from outside the nested domain are also accounted for. In order to quantify the background concentration, we conducted another sensitivity simulation with all sources outside the domain shut off. The standard and sensitivity simulation results are shown in Fig. 7a and b, and the difference between them is analyzed as the contribution from outside the domain, which is shown in Fig. 7c. The maximum contribution from outside is up to 13.8 µgm-3, which mainly occurs at the western and northwestern boundaries. The average contribution is 1.57 µgm-3 in the simulation domain of eastern Asia. Within the boundary of China, the largest contribution occurs in the northeast, which is 7.35 µgm-3. The average contribution from outside the nested domain is only 0.3 µgm-3 within China.

Figure 8 shows the simulated seasonal mean PM2.5 concentration (Fig. 8a and b) and source contributions from coal burning in winter (averaged from December to February) and in summer (averaged from June to August) (Fig. 8c–j), which is also summarized in Tables 4 and 5. As shown in Fig. 8a and b, the ambient PM2.5 concentration has obviously different distributions in winter and in summer. PM2.5 in winter has a similar distribution with the annual mean, but with much higher values. The highest value still occurs in the SCB with an average of 118.8 µgm-3 due to the large emission, unfavorable terrain, and weather conditions in winter. Following the SCB, the average concentrations in the MYR and NC regions are above 100 and 90 µgm-3, respectively. There are also several populated cities in NEC where PM2.5 are generally above 75 and up to 150 µgm-3. PM2.5 in summer has an obviously different distribution from winter with much lower concentrations and more even distribution throughout the country due to the stronger vertical mix, more wet deposition, and lower emissions. The largest concentration occurs in the NC region with 46.9 µgm-3 on average, followed by the SCB with an average of 44.1 µgm-3. In addition to the two regions above, PM2.5 concentrations in other key regions are generally around or below 35 µgm-3 on average.

In winter, coal burning contributes to 28.2 µgm-3 (35.4 %) of total PM2.5 concentration on the national level. Similar to the annual mean, coal-burning contribution in winter peaks in the SCB with an average of 50.3 µgm-3 (42.3 %) and reaches the lowest in the PRD with 16.1 µgm-3 (29 %). Among the coal-burning sectors, the contributions from power plants and industry also have similar spatial patterns to the annual mean distribution. Coal burning in industry, followed by that in power plants, is the largest contributor in both seasons. Domestic coal burning is a significant contributor in winter due to the large amount of emissions from heating supply. The high PM2.5 concentration from the domestic sector mainly occurs in some areas in Guizhou Province in the southwest and Inner Mongolia in the north, where a large amount of raw coal is burnt for heating. The largest contribution reaches as much as 37.6 µgm-3 in Inner Mongolia, which accounts for almost 40 % of the local PM2.5 concentration.

In summer, the national average contribution from coal burning is estimated to be 17.8 µgm-3 (46.2 %), which is less than two-thirds of the contribution in winter due to the favorable meteorological condition including stronger convection and more frequent wet deposition. Regional contribution ranges from 8.2 µgm-3 in the PRD to 26.3 µgm-3 in the SCB, which is approximately half of the contributions in winter. The seasonal variation of contributions in inland areas (NEC, MYR, SCB) is more significant than those in coastal areas (NC, YRD, PRD). In coal-burning sectors, the absolute contributions from power plants and industry do not show very noticeable reductions in summer compared with those in winter as emissions from these two sectors are in a relatively constant status throughout the year and the nitrate reduction due to the high temperature in summer is counteracted by the enhancement of sulfate formation (H. Zhang et al., 2012). In contrast, the domestic sector contributes 1 µgm-3 (2.5 %) on the national level in summer, which is 3–8 times less than that in winter.

The Natural Resources Defense Council (NRDC) launched the China Coal Consumption Cap Project in October 2013 and released the report “Coal Use's Contribution to Air Pollution in China” as part of the study results in October 2014 (NRDC, 2014). This study used the CAMx model with the Multi-resolution of Emission Inventory for China (MEIC) inventory and meteorology from the Weather Research and Forecasting (WRF) model to simulate coal contributions to ambient PM2.5 in January, February, April, and October in the year 2012 in 333 main cities in China. In order to compare with the NRDC study, we extracted the simulated contribution in the 333 main cities during the same periods from our study results. Figure 9 represents the comparison in each province and shows that our study underestimates the coal contribution by 22 % compared to that in the NRDC study. This discrepancy is mainly generated from the different amounts of emissions that originate from coal in the two studies. According to the report, the NRDC study included both emissions directly from coal burning and emissions from industries closely related to coal burning. For example, air pollutants from industries like coke, steel, cement, and nonferrous metal are generated two ways: directly from coal combustion and from technological processes. As coal is used as fuel in these industries and is not likely to be substituted for in the near future, the NRDC study includes both parts as emissions from coal use. In our study, we include only the first part of the emissions as the contribution from coal, which is actually generated from coal burning. According to the report by the NRDC, coal combustion is responsible for 79 % of SO2 emissions, 57 % of NOx emissions, and 44 % of primary PM emissions, and the coal-related sources are responsible for 15, 13, and 23 % of the SO2, NOx, and PM emissions, respectively. Despite the different definition of coal contribution to air pollutant emissions, the NRDC and our study both predicted a high contribution to PM2.5 concentration from coal, especially in the municipality of Chongqing and Sichuan Province in the SCB.

The uncertainties of the contribution estimates in this study may arise from the uncertainties of the emission inventory, model simulation, and non-linearity of the atmospheric chemistry. A Monte Carlo uncertainty analysis was performed on the emission inventory, as described in Zhao et al. (2013c) and S. X. Wang et al. (2014b). Table 6 shows the uncertainty analysis of the emissions in China. Among all the coal-consuming sectors analyzed in this study, the domestic sector is subject to the highest uncertainty, which may lead to more uncertainty in the PM2.5 simulation and contribution estimates. Other studies on major pollutant emissions in China are summarized in Table 7. Emissions from Liu et al. (2016), Xia et al. (2016), and Wu et al. (2016) are also estimated using the bottom-up method, while those from Zhao et al. (2014) are projected emissions for 2015 based on the year 2010. The results of this study fall into the range of previous studies except for China's Ministry of Environmental Protection (MEP, 2014a), which is at low end. One major reason for low NOx emission from MEP (2014a) is that it does not include the emissions from non-road vehicles.

Another important cause of uncertainty is the model simulation of the PM2.5 composition. The coal contribution to sulfate is larger than that to nitrate since the share of coal-burning emissions of SO2 is 79 % in this study, 25 % higher than that of NOx emissions. Therefore, the actual coal-burning contribution to PM2.5 is very likely to be larger than the estimates in this study due to the underestimation of sulfate concentration and overestimation of nitrate concentration by the model.

In addition, due to the nonlinear response of PM2.5 concentration to precursor emissions, contributions from coal burning in each sector add up to less than the contribution from the total coal burning, which indicates the probable underestimation of the contribution in subsectors. The impact of nonlinearity of the atmospheric chemistry on PM2.5 concentrations and their composition has been discussed in detail in previous studies (Zhao et al., 2013b; S. Wang et al., 2014a). There are some studies using different methods to study the source apportionment of ambient PM2.5. As this study only focuses on coal-burning emissions in each sector, the results are not directly comparable to most similar studies except for results for the power sector as coal combustion dominates the emissions in the power plant sector. Zhao et al. (2015) used the extended response surface modeling technique to access the nonlinear response of fine particles to precursor emissions in each sector in the PRD region, reporting that local PM2.5 concentration decreased less than 3 % (7.2 % in our study) in January and around 12 % in August (13.8 % in our study), when 90 % of emissions in power plants are reduced. Our results include the trans-boundary contributions as we shut off emissions across the country in the sensitivity simulation, which is one of the reasons causing the discrepancies. L. Zhang et al. (2015) took advantage of the adjoint capability of GEOS-Chem, reporting that power plants contributed 6 % to PM2.5 concentration in Beijing, which is consistent with our study (6.9 %).