Currently, coal plays a dominant role in China's energy consumption, making up about 70 percent of its total primary energy consumption (Tian et al., 2007, 2012b). Consequently, tons of hazardous HM pollutants can be released into the atmospheric environment, although the concentration of heavy metals in Chinese coals is normally at parts per million (ppm) levels.

Atmospheric emissions of varied HMs from coal combustion are calculated by combining the provincial average concentration of each heavy metal in feed coals, the detailed coal consumption data, and the specific emission factors, which are further classified into subcategories with respect to different boiler configurations and the application rates and removal efficiencies of various APCDs. The basic formulas can be expressed as follows: E(t)=∑i∑j∑kCi,j,k×Ai,j,k×Ri,j×1-ηPM(i,j)1-ηSO2(i,j)1-ηNOx(i,j), where E is the atmospheric emissions of Hg, As, Se, Pb, Cd, Cr, Ni, Sb, Mn, Co, Cu and Zn; C is the averaged concentration of each HM in feed coals in one province; A is the amount of annual coal consumption; R is the fraction of each heavy metal released with flue gas from varied coal combustion facilities; ηPM, ηSO2, and ηNOx represent the averaged fraction of one heavy metal which is removed from flue gas by the conventional PM / SO2 / NOx emission control devices, respectively; i represents the province (autonomous region or municipality); j represents the subcategory emission source which is classified by different sectors of the economy and combustion facilities, as well as the installed PM, SO2 and NOx control devices (the detailed source classification can be seen in the Supplement, Table S2); k represents the type of coal as consumed (raw coal, cleaned coal, briquette, and coke); and t represents the calendar year.

HM emissions from non-coal categories are calculated as a product of annual activity data (e.g., fuel consumption, industrial product yields) and specific emission factors of varied HMs. The basic calculation can be described by the following equation: E(t)=∑i∑j(Ai,j×EFj), where E is the atmospheric emissions of each heavy metal; A is the annual production yield of industrial producing processes, volume of municipal solid wastes incineration, or liquid fuel and biofuel consumption etc.; EF is the assumed average emission factors; and j is the emission source classified by source subcategories (see Supplement Table S1).

Notably, atmospheric emissions of Pb have significantly dropped in China, as a result of unleaded gasoline introduction since the early 2000s. The proportion of lead in leaded gasoline emitted to the air is estimated at about 77 % (Biggins and Harrison, 1979) or 75 % (Hassel et al., 1987), and thus this parameter is assumed to be at about 76 % for the period before 2000 in this study. Consequently, for leaded gasoline used by motor vehicles in China, the total Pb emitted to the atmosphere is calculated according to the following equation: E(t)=∑i(0.76×CPb×Ai), where E(t)is the emissions of Pb from motor vehicle gasoline combustion in calendar year t; CPb is the average content of lead in gasoline; and Ai is annual gasoline consumption in one province, autonomous region or municipality i.

For brake and tyre wear, the atmospheric emissions of several HMs are estimated by the following equation: E(t)=∑i∑j∑k(Pi,j,×Mj×EFj,k×Ck), where E(t) is the atmospheric emissions of As, Se, Pb, Cd, Cr, Ni, Sb, Mn, Co, Cu or Zn in calendar year t; Pi,j is the population of vehicles in category j (passenger car, bus and coach, light-duty truck, and heavy-duty vehicle) in province, autonomous region or municipality i; Mj is the average annual mileage driven by vehicle in category j; EFj,k is the emission factor of TSP (total suspended particles) for brake lining or tyre k by vehicle category j; and Ck is the averaged concentration of each heavy metal in brake lining or tyre k. Relevant parameters are summarized in the Supplement Tables S12–S13.

Coal and liquid fuels consumption data by sectors at the provincial level (e.g., power plant, coal-fired industrial boiler, coal-fired residential sector, other coal-fired sectors) are collected from China Energy Statistical Yearbooks. Industrial production data by provinces (e.g., the output of ferrous/nonferrous metal products, production of cement/glass/brick, amount of municipal waste incineration, number of vehicles) are compiled from relevant statistical yearbooks, such as China Statistical Yearbooks, the Yearbook of Nonferrous Metals Industry of China, China Steel Yearbook etc. The detailed data sources for the main sectors are listed in the Supplement, Table S19. Furthermore, trends of activity levels by different sectors in China between 2000 and 2012 are summarized in the Supplement, Figs. S1–S5.

It is necessary to examine the potential uncertainty in emissions by sources and regions to quantify the reliability, and identify the direction and degree of improvement, of emission inventories in the future. A detailed uncertainty analysis is conducted by combining uncertainties of both activity levels and emission factors, through adopting a Monte Carlo simulation (Zhao et al., 2011; Tian et al., 2014a, b). Streets et al. (2003) indicate that there is no way to judge the accuracy of activity data estimates. Furthermore, uncertainties are still inevitable when representative values are selected for specific emission sources, countries, and regions in spite of emission factors adopted from detailed experiments.

Most of the input parameters of specific activity levels and emission factors, with corresponding statistical distributions, are specified on the basis of the data fitting, or referred to the related published references (Wu et al., 2010; Zhao et al., 2011; Tian et al., 2012a, b). Additionally, for parameters with limited observation data, the probability distributions such as normal distribution and triangular distribution are assumed by the authors for corresponding sources. Further details about the probability distribution for each source discussed in this study are listed in Table S20. Finally, all of the input parameters are placed in a Monte Carlo framework, and simulations are run 10 000 times to estimate the uncertainty ranges of varied HM emissions with a 95 % confidence interval.

The historical trend of atmospheric emissions of Hg, As, Se, Pb, Cd, Cr, Ni, Sb, Mn, Co, Cu, and Zn by different source categories from 1949 to 2012 are illustrated in Fig. 2. The total emissions of HMs from primary anthropogenic sources since 1949 have shown substantial shifts among varied source categories that reflect technological and economic trends and transition during this period of over 60 years. Within the year after the establishment of the People's Republic of China in 1949, the total emissions of Hg, As, Se, Pb, Cd, Cr, Ni, Sb, Mn, Co, Cu, and Zn from anthropogenic sources are estimated at about 11.5–312.6 t (see Table 2). The discharges of HMs on a national scale increased by 3–20 times from 1949 to 1960 due to the increasing demands for energy consumption and industrial production (especially for the period of the Great Leap Forward from 1958 to 1960, resulting in a remarkable increasing output of industrial products), then, a substantial decrease in 1961 and 1962 of 27.6–55.7 % compared to 1960 on account of the serious imbalance of economic structure and the Great Leap Forward famine caused by policy mistakes together with natural disasters (Kung and Lin, 2003). In spite of the negative growth of heavy metal emissions in individual years such as 1967, 1974, and 1976, the annually averaged growth rates of national emissions of HMs from primary anthropogenic sources were still as high as 0.2–8.4 % during the period of 1963 to 1977.

Subsequently, the policy of openness and reformation was issued by the Chinese central government. With the implementation of this policy from 1978 to 2012, China's GDP has been growing at an average annual growth rate of about 9.8 %, resulting in tremendous energy consumption and enormous output of industrial products. As can be seen from Fig. 2, historically there have been two periods during which the total emissions of HMs (except Pb) increased rapidly after 1978. The first one is the period of 1978 to 2000, except for one remarkable fluctuation from 1998 to 1999, which reflects a decrease in input of raw materials and output of industrial products mainly owing to the influence of the Asian financial crisis (Hao et al., 2002). The second one is the period of the 10th Five-Year Plan (10th FYP, from 2001 to 2005) in which a sharp increase in emissions of Hg, As, Se, Cd, Cr, Ni, Sb, Mn, Co, Cu, and Zn occurred, with the emissions increasing from about 268.0–11 308.6 t in 2001 to about 378.9–15 987.9 t in 2005, at an annual average growth rate of 4.8–12.0 % (see Table 2).

In terms of the lead content requirement in gasoline, the past 64 years since the foundation of the PR China (1949 to 2012) can be divided into two phases: the leaded gasoline period (1949 to 1990: gasoline with high lead content (0.64 g L-1); 1991–2000: gasoline with low lead content (0.35 g L-1)), and the unleaded gasoline period (2001 to 2012, 0.005 g L-1). As a result, the discharge of Pb from primary anthropogenic sources experienced two fluctuations over the 64 year period. The first sharp emission decline occurred in 1991, and the total emissions decreased by 26.2 % from 17 644.0 t in 1990 to 13 029.6 t in 1991, mainly because the average Pb content in leaded gasoline regulated by GB 484–89 decreased about 45.3 % compared to that in GB 484–64. The other sharp decline occurred in 2001, and the total Pb emissions from primary anthropogenic sources reduced abruptly by about 61.6 % in 2001. Subsequently, along with the rapid increase of vehicle volume and oil consumption, a substantial increase was once again experienced from 7747.2 t in 2001 to 14 397.6 t in 2012, at an annual average growth rate of about 5.8 %.

Due to technological progress resulting in relatively low emission factors of HMs, and economic development bringing about high coal consumption and industrial products output, the trends of total atmospheric emissions for different HMs in China are diverse during the period of 2006 to 2012 (Cheng et al., 2015). Generally speaking, the national atmospheric emissions of Hg, Pb, Cd, Cr, Sb, Cu, and Zn increased at an annual average growth rate of 1.5–7.2 % from 2006 to 2012. In spite of the remarkable growth in coal consumption and gross industrial production, the national As, Se, Ni, Mn, and Co emissions are well restrained in this period. These are mainly due to the different volatility of these 12 elements during the high temperature process resulting in diverse release rates of furnaces and synergistic removal efficiencies of control measures (Xu et al., 2004).

Due to limited information about historical ground-level concentrations of 12 HMs in different cities in China, the temporal characteristics of atmospheric concentrations of four HMs (As, Pb, Cr, and Cu) in Beijing during 2000 to 2012 are used as valid index to verify whether or not the trend of historical HM emissions is reasonable (see Supplement Fig. S6). The data sources and specific values about atmospheric concentrations of As, Pb, Cr, and Cu in Beijing during 2000 to 2012 are listed in the Supplement, Table S21. It should be acknowledged that the verification method applied in this study has certain limitations on account of sampling discrepancies, including sampling time, sampling site, and detection method etc. Therefore, the historical variation trends of HM emissions may be inconsistent with those of ambient concentrations of HMs in some years.

As can be seen from Fig. S6, minimum values of the atmospheric concentrations of As, Pb, Cr, and Cu occur in 2008. This is mainly because most of the aerosol samples compiled from published papers are collected during August in that year, the time during which China hosted the Beijing Olympics under which a series of strict measures about energy-saving and pollution reduction were implemented, such as suspending production of high polluting industries in the Beijing and neighboring municipalities, restricting the driving of vehicles on alternate days under an even–odd license plate system, limiting pollutant emissions from coal combustion facilities in Beijing and the surrounding provinces etc. Consequently, the variation trends of atmospheric concentrations of As, Pb, Cr, and Cu have some discrepancies with those of historical emissions of the above four HMs in Beijing in 2008. However, the historical emission trends of As, Pb, Cr, and Cu are consistent with those of atmospheric concentrations of the above four HMs during 2000 to 2012 in general (see Fig. S6), which indicates that the historical trend of HM emissions estimated by this study is reasonable.

Until now, the comprehensive and special studies on various HM (except Hg) emissions in China are quite limited. Therefore, only a detailed comparison with Hg emission estimates from other studies is discussed in this study (see Fig. 3). Specifically, limited data on China's Hg emissions can be cited directly from the global Hg inventories estimated by Pacyna and Pacyna (2001), Pacyna et al. (2006, 2010) and Streets et al. (2011). Consequently, here, we mainly focus on comparing our estimations with the results from China's specialized Hg emission inventories estimated by Streets et al. (2005) and Wu et al. (2006).

Overall, the estimated Hg emissions from fuel combustion (except for the subcategory of coal consumption by residential sectors) in this work are substantially consistent with those reported by Streets et al. (2005) and Wu et al. (2006), although the values calculated for the same year are somewhat different. This may be mainly attributed to the difference in the averaged provincial content of Hg in raw coal. In our study, according to a comprehensive investigation of published literature, we determine the national averaged Hg content in China to be 0.18 mg kg-1 by using a bootstrap simulation method, a little lower than that used by the above two studies (0.19 mg kg-1). Another important factor influencing the result is the difference of removal effectiveness of Hg through traditional APCDs. Nevertheless, the estimated Hg emissions from coal consumption by residential sectors by Streets et al. (2005) and Wu et al. (2006) are higher than our estimation in the same year. This is mainly because the emission factor of Hg from coal consumption by residential sectors is cited from Australia NPI in this paper, which is only approximately half of that which EPA adopted in the above two studies. In terms of Hg emissions from industrial processes, the estimated Hg emissions in this study are generally lower than those in other Hg emission inventories in the same year. This may be because we have adopted S-shaped curves to quantify the positive effects on emission reduction of pollutants by technology improvement, so that the emission factors adopted in this study are generally lower than those used in studies of Streets et al. (2005), Wu et al. (2006), and Wu et al. (2012) in the same year. In addition, some anthropogenic sources with high uncertainties are not taken into account in this work due to the lack of detailed activity data for the long period. Certain natural sources (e.g., forest burning, grassland burning) are also not included in this study. Consequently, our estimated total Hg emissions are lower than those in inventories estimated by Streets et al. (2005) and Wu et al. (2006).

The total emissions of Hg, As, Se, Pb, Cd, Cr, Ni, Sb, Mn, Co, Cu, and Zn from primary anthropogenic sources by provinces in China for the year 2010 are estimated at about 72 955.1 t. As can be seen in Fig. 4, coal combustion sources represent the major contributors of Hg, As, Se, Pb, Cr, Ni, Mn, Co, and Cu emissions and are responsible for about 50.6, 74.2, 64.6, 60.1, 90.4, 56.2, 80.9, 98.6, and 53.4 % of total emissions, respectively, while their contribution to the total Cd, Sb, and Zn emissions are relatively lower, at about 32.7, 39.3, and 39.8 %, respectively.

Among all the coal-consuming sub-sectors, coal consumption by industrial boilers ranks as the primary source of total national emissions of 12 HMs, with the average proportion about 57.7 % of the total emissions from coal combustion. This may be attributed to the significant coal consumption of industrial boilers (about 1117.3 Mt in 2010) and the relatively high share of boilers with inadequate APCDs (Cheng et al., 2015; NBS, 2013b).

As the largest coal consumer in China, coal consumption by power plants is identified as the second largest contributor and accounts for about 14.0 % in total national emissions of 12 HMs. In order to achieve the emission reduction of PM, SO2, and NOx for satisfying the national or local emission reduction goals for the year 2010 (the end year of 11th FYP) (NBS, 2011; Tian et al., 2014a), control policies have been implemented, including replacement of small coal-fired plant units with large and high-efficiency units and the continuously increasing installation rate of advanced APCDs systems (e.g., ESP, FFs, WFGD, SCR). Consequently, the final discharge rates of HM from power plants have decreased obviously, even though the volume of coal consumption has grown substantially (see Fig. 2 and Fig. S1).

China has been the world's largest producer of pig iron and steel by a rapidly growing margin. By the end of 2012, the output of steel has amounted to 723.9 Mt, accounting for about 46 % of worldwide steel production (CISA, 2013). Despite the enormous achievement by China's iron and steel industry, China is still featured as a steel producer with low energy efficiency and high pollutant emissions level, compared with other major steel-producing countries (Guo and Fu, 2010). Because of limited application of FGD and de-NOx devices and poor control of PM, the ferrous metal smelting sector ranks as the third largest contributor, responsible for about 13.2 % of the total national emissions of 12 HMs. In terms of Zn emissions, the share of this sector is dominant, accounting for about 32.2 % of the total.

Regarding nonferrous metal smelting emissions, the primary smelting processes resulting in HM emissions discharge far more pollution than those emitted from the secondary smelting processes. Nonferrous metal smelting, as the fourth largest contributor, accounting for about 11.0 % of total emissions, represents the primary contributor to the discharges of Hg and Cd. Therein, primary Cu smelting contributes the largest part of most of these elements, including 89.5 % for As, 37.3 % for Pb, 74.8 % for Cd, 38.7 % for Ni and 76.6 % for Cu; primary Pb smelting is the major source of Sb and Pb; primary Zn smelting accounts for the largest proportion of Hg and Zn emissions among the nonferrous metal smelting category. In addition, with respect to Hg emissions from the nonferrous metal smelting sector, the mercury smelting industry is the other dominant subcategory source, with a share of about 33.0 % of nonferrous metal smelting emission in 2010.

It can be concluded that the emissions of HMs from brake wear are associated with the vehicle population, vehicle mileage as well as the content of HMs in brake linings and tyres. Currently, numerous studies have reported that airborne HMs (e.g., Sb, Cu, Zn) in urban areas are associated with road traffic and more definitely with emissions from brake wear (Gómez et al., 2005; Hjortenkrans et al., 2007). As can be seen from Fig. 4h and k, the brake and tyre wear sector is the largest source of national Sb and Cu emissions – 39.9 and 26.3 %, respectively. Brake wear is the dominant sub-contributor, accounting for over 99.9 and 99.6 % for Sb and Cu emissions from this sector in 2010, respectively. This is mainly due to the high content of Sb and Cu in the brake linings (see Table S13, Hjortenkrans et al., 2007) and the explosive expansion of the vehicle population in China (see Fig. S5). Nevertheless, the adverse effects of airborne PM originating from brake wear on human health and the ecosystem have still not received sufficient attention from the policymakers as well as the public.

Although the non-metallic mineral manufacturing sector is not the dominant source of most HMs, the discharge of Se from this sector makes it the largest contributor to the total. Within this category, the glass production sector discharges about 92.9 % of the total Se emissions due to the widespread application of selenium powder as a decolorizing agent in the glass production process and the huge output of glass production (Kavlak and Graedel, 2013).

As can be seen from Fig. 4a–l, the source contributions on the provincial scale in 2010 vary substantially due to the difference of industrial conformations and energy structures (Cheng et al., 2015; NBS, 2013a, b). Among the provinces with high HM emissions, Shandong ranks as the largest province with As, Se, Cd, Ni, Mn, and Cu emissions; accounting for about 8.1–10.6 % of the national emissions; Hebei contributes the largest part of about 9.3 and 11.3 % to national Pb and Zn emissions respectively; Guizhou represents the primary province with Hg and Sb emissions; the key provinces with Cr and Co emissions are found in Yunnan and Shanxi, respectively. These can be mainly attributed to the follow reasons (NBS, 2013a, b; Wu et al., 2008): (1) the enormous coal consumption of industrial boilers, the considerable electric power generation, a substantial increase of vehicle population, and the huge output of industrial products in Shandong, (2) the flourishing pig iron and steel production in Hebei, (3) the dominant outputs of mercury and obviously high average concentration of Sb in feed coals in Guizhou (about 6.0 µg g-1, which is approximately 4 times higher than the national averaged concentration of Sb in coal as consumed in China, see Table S8), (4) the booming coke-making industry in Shanxi, and (5) the relatively high concentration of Cr in feed coals in Yunnan (about 71.7 µg g-1, which is 2 times higher than the national averaged concentration of Cr in coal as consumed in China, see Table S8).

The spatial distribution patterns of HM emissions from anthropogenic sources are illustrated in Fig. 5. In this study, 1796 power plants with capacity larger than 6000 kW, 566 copper/lead/zinc smelting plants, 33 large iron and steel plants, and 101 MSW incineration plants are identified as large point sources and their emissions are precisely allocated at their latitude/longitude coordinates (the geographical distribution of 2496 point sources in China is shown in the Supplement, Fig. S7). It should be noted here, the emissions from point sources of nonferrous metal smelting industry and ferrous metal smelting industry contain two parts: emissions originating from fuel combustion and emissions emitted from industrial production processes. Except for the emissions from point sources discussed above, the remaining anthropogenic sources in the provincial level are all treated as regional area sources. The specific method of geographical location for area sources has been discussed in our previous studies (Tian et al., 2012b, c).

The spatial variation is closely related with the unbalanced economic development and population density in the Chinese mainland, so that these 12 typical HM emissions are distributed very unevenly from one area to another, with the annual As emissions at the province level ranging from 0.009 kg km-2 in Qinghai to 1.6 kg km-2 in Shandong, for instance. One notable characteristic of the spatial distribution of China's HM emissions is that the HM emission intensities are much higher in central and eastern China than those in western China, and the coastal regions are classed as the most polluted areas of varied HMs. The emissions of HMs from Hebei, Shandong, Henan, Jiangsu, Shanxi, and Liaoning provinces almost account for about 39.4 % of the total emissions of these HMs. These six provinces above are characterized by extensive economy growth mode, a large volume of coal consumption and various industrial products output, as well as a high population density. Therefore, more energy consumption and higher travel demand characterize these six provinces, compared to other provinces and districts, resulting in higher HM emission intensity.

Moreover, several provinces in the southwestern and central-southern regions also play a prominent role for these 12 HM emissions, especially for Guizhou, Sichuan, Yunnan, Hubei, and Hunan provinces. In general, Guizhou province starts out with high emissions of HMs from coal consumption by other sectors, mainly owing to both the high HM contents in the feed coals and the large magnitude of coal consumption by this sector. In addition, the nonferrous industries of Hunan and Yunnan provinces are flourishing, especially the copper and zinc smelting industries. Consequently, the nonferrous metal smelting sector is seen as one of the major sources of Cu and Zn emissions in these two provinces.

The situations of atmospheric HM concentrations in the aerosols of 44 major cities in China during the last 10 years have been reviewed comprehensively by Duan and Tan (2013). Their results indicate that the ambient concentrations of HMs (As, Pb, Cd, Cr, Ni, Mn, Cu, and Zn) are high in some cities, including Beijing, Tianjin, Shijiazhuang, Shenyang, Harbin, Jinan, Zhengzhou, Hangzhou, Nanjing, Hefei, Xian, Yinchuan, Urumqi, Wuhan, Changsha, Chongqing, Guangzhou, Shenzhen, Foshan, Shaoguan, and others. For HM emissions on the urban scale in 2010, these 20 cities with high HMs concentrations also represent the cities with the highest HM emissions in China (see Fig. 5). In general, the spatial distribution characteristics of gridded HM emissions from primary anthropogenic sources for the year 2010 in this study are reasonable and representative of the real situation of this HM pollution.

Emissions of varied HMs from primary anthropogenic sources with uncertainties in 2010 are summarized in Fig. 6 and Supplement Table S22. As can be seen, the overall uncertainties of the total emissions in our inventories quantified by the Monte Carlo simulation are -39.1 to 50.8 %. Among all the coal combustion sectors, uncertainties for thermal power plants emissions are smallest, whereas those for coal-fired residential sectors and other coal-fired sectors are considerable. These are mainly attributed to the relatively poor resolution of coal burning technologies and emission control devices in these two subcategories. In contrast, relatively higher uncertainties are observed in the non-coal combustion categories, particular for non-metallic mineral manufacturing and brake and tyre wear emissions. These high uncertainties of HM emissions can be mainly attributed to imprecise statistics information, poor source understanding, as well as the absence of adequate field test data in China.

The earlier statistical data for activity levels are considered to have high uncertainty for developing countries (including China) with less developed statistical systems. Unfortunately, we have to acknowledge that it is quite difficult to accurately assess the specific uncertainty of activity data from China's earlier official statistics. Akimoto et al. (2006) argue that the data of energy consumption of China during 1996–2003 is not recommended for use in the study of emission inventories due to the probable underestimates. However, the discrepancies in coal consumption from the power sector are considered to be less than ±5 %, which do not dominate the emission uncertainties (Wu et al., 2010; Zhao, et al., 2008). In order to approximately quantify the uncertainty of activity data, we divide the whole period of 1949 to 2012 into three stages with respect to economic development and emission control: before reform and opening (1949–1978), intermediate stage (1979–2005), and the substantial control stage of atmospheric pollutants (2006–2012). For the activity level of anthropogenic sources obtained from official statistics after 2006, we assume normal distributions with sector-dependent uncertainties (see Supplement Table S20). On the basis of the discussion and considerations above, the uncertainty of activity data from official statistics during the two early periods of 1949–1978 and 1979–2005, is assumed to be about 2 and 1.5 times that in the period of 2006–2012.

The combined uncertainty bounds for the national emissions of 12 HMs during the historical period are shown in Fig. S8. In general, the range of uncertainty has gradually diminished over time. For example, we calculate an uncertainty level of -90.1 to 125.7 % (95 % confidence interval) in the estimate of national Hg emissions in 1949, which is higher than that of the other 11 HM emissions (between -90.0 and 119.3 %). This is mainly attributed to remarkable emissions from several Hg sources that have the largest uncertainty in both activity levels and emission factors, such as gold smelting and mercury mining. Since then, the relative uncertainties have gradually decreased from the beginning to the end of the period. This is primarily because more reliable activity data with a smaller coefficient of variation (CV) from related yearbooks and reports became available. The uncertainty range of national Hg emissions is estimated to be -40.6 to 55.8 % by 2003, which compares well with estimates of ±44 % for China's Hg emissions by Wu et al. (2006). By the end of 2012, the overall uncertainty level was reduced to -39.0 to 47.0 % for the national HM emissions. Particularly, it should be acknowledged that emission trends of HMs are probably more uncertain than indicated. This is mainly because dynamic emission factor curves of certain sectors are set between two very distant points (e.g., 1890 and 1990 in Fig. 1), which means that the timing and rate of the decreased emission factors are really unconstrained. Nevertheless, no data are available to improve the confidence presently, which merits further investigation in the future.

Generally speaking, emission inventories are never complete and perfect, and most emissions estimates possess a significant associated uncertainty, mainly owing to the lack of representativeness of specific emission factors and the reliability of the source-specific activity data. In this study, we have made great efforts to evaluate the historical trend of these HM emissions by collecting detailed activity levels for various source categories, adopting the best available dynamic emission factors for various anthropogenic sources in China today, and integrating publication literature and reports from developed countries and districts. Nevertheless, considerable uncertainties are still present, and this may lead to under- or overestimation of HM emissions from some source categories. Consequently, more detailed investigations and long-term field tests for all kinds of coal-fired facilities and industrial production processes are in great demand.

Presently, control of atmospheric HM emissions still has not received sufficient attention by the government and public in spite of the frequent occurrence of HM pollution in China (especially for provinces with high point sources of HM emissions). The implementation of more rigorous emission standards on primary anthropogenic sources (thermal power plant, coal-fired boiler, nonferrous metallurgy, pig iron and steel production etc.) and national ambient air quality standards (NAAQS) are regarded as important triggers to promote enterprises with diminished HM emissions. Therefore, the MEP should speed up the revision of the system of hazardous air pollutant (including HM) emission standards, and strengthen the amendment of NAAQS. Especially brake wear has been confirmed to be the main source of HM emissions from traffic, particularly in urban areas. However, there is no related emission standard of air pollutants for brake wear. In the near future, the promulgation of emission standards of brake wear should be expected, which will further strengthen the control of atmospheric HM emissions in China.

In addition, some specific actions are suggested as follows: (1) lower or stop mining and burning coal with high HM concentrations in certain provinces where the coals are mainly mined from small coal mines such as Zhejiang and Guangxi (or lower or stop using high-sulfur coal in corresponding provinces due to the high affinity between HMs and pyrite in coal) (Yuan et al., 2013; Zhu et al., 2015); (2) promote coal washing before combustion (the removal efficiencies of coal preparation to lower heavy metals can reach as high as approximately 30.0–60.0 %, see Supplement Table S9); (3) increase the application rate of advanced APCDs configurations in newly built or retrofitted coal-fired boilers; (4) initiate pilot tests or demonstration projects for specified mercury control (SMC) technologies in some sectors with high Hg emissions and develop comprehensive HM control technologies capable of simultaneously removing multiple heavy metals; (5) strengthen energy conservation and boost electricity and/or heat generation using cleaned energy and renewable energy, such as nuclear, wind and solar energy; (6) suspend small-scale coal-fired boilers and industrial production plants with backward emission control technologies (e.g., cement plants, ferrous smelting plants, nonferrous smelting plants); (7) eliminate outdated production technology, such as VR pyrometallurgy and ISP pyrometallurgy; and (8) improve cyclic utilization rate of nonferrous metals and ferrous metals during the period of 12th FYP etc.