Historical (1850–2010) mercury stable isotope inventory from anthropogenic sources to the atmosphere

Mercury (Hg) is a toxic element. Transfer of Hg from Earth’s lithosphere to the atmosphere and other surface environments is continuous by natural processes, and has been accelerated by human activities dating back to antiquity (Goldwater, 1972; Nriagu, 1979). Natural atmospheric Hg emissions from volcanoes, crustal weathering and hydrothermal activity are thought to be 1–2 orders of magnitude smaller than modern anthropogenic Hg emissions (Amos et al., 2015; Bagnato et al., 2014; UNEP, 2013). Prior to the 1850s, anthropogenic Hg releases mainly came from primary Hg mining and use of Hg as amalgamation agent for silver extraction in the Spanish colonial Americas (Camargo, 2002;Hagan et al., 2011; Robins and Hagan, 2012). Following the industrial revolution, anthropogenic Hg releases further increased due to large-scale gold mining, non-ferrous metal production and combustion of fossil fuels (Streets et al., 2011). Streets et al. (2011) estimated that 215 Gg of cumulative anthropogenic Hg has been directly emitted into the atmosphere since the 1850s. This inventory was recently updated in Horowitz et al., (2014) by incorporating Hg releases from previously unquantified commercial Hg uses. The authors showed that an additional 540 Gg of Hg has been released to the atmosphere (20%), water (30%) and terrestrial (50%) reservoirs (Horowitz et al., 2014).

Released Hg will undergo complex redox reactions within and among atmospheric, terrestrial and aqueous reservoirs and is ultimately sequestered in the Earth’s lithosphere when it is incorporated in marine sediments (Amos et al., 2014; Andren and Nriagu, 1979; Mason et al., 1994). Substantial legacy Hg has accumulated in Earth’s surface reservoirs, significantly amplifying Hg loading to biogeochemically active pools (Amos et al., 2013). Natural Hg archives such as sediments, ice cores and peat cores (Faïn et al., 2009; Fitzgerald et al., 2004; Lamborg et al., 2002; Martínez-Cortizas et al., 1999; Schuster et al., 2002), and box models of global Hg cycling (Amos et al., 2013, 2015) suggest that present-day atmospheric Hg deposition has increased by a factor of 3–5 since the industrial period and by a factor of 17–27 since 3000 BC. Substantial debate exists on the historical Hg emission sources, e.g., Spanish-American mining in the 16^th–18^th century, North-American mining in the late 19^th century, and other global change factors that have led to the increase in atmospheric Hg deposition (Amos et al., 2015; Engstrom et al., 2014; Krabbenhoft and Sunderland, 2013).

Hg stable isotope signatures have the potential to differentiate natural and anthropogenic Hg sources, and identify and quantify Hg transformations (Blum et al., 2014; Sonke, 2011; Sonke and Blum, 2013; Yin et al., 2014b). More than 10‰ variations in mass dependent Hg isotope fractionation (MDF, indicated by δ²⁰²Hg) and mass independent fractionation (MIF, indicated by Δ¹⁹⁹Hg, Δ²⁰¹Hg) of odd Hg isotopes have been documented in natural samples such as crustal rocks (Smith et al., 2008), cinnabar (Gehrke et al., 2011; Gray et al., 2013; Hintelmann and Lu, 2003; Smith et al., 2008;Stetson et al., 2009; Wiederhold et al., 2013), coal (Biswas et al., 2008; Sun et al., 2016; Yin et al., 2014a), and non-ferrous metals (Smith, 2010; Sonke et al., 2010; Yin et al., 2016). Different physical, chemical and biological Hg transformation processes induce characteristic MDF and MIF signs and magnitudes (Bergquist and Blum, 2007; Chandan et al., 2015; Ghosh et al., 2013;Jiskra et al., 2012; Kritee et al., 2013; Perrot et al., 2015; Rodríguez-González et al., 2009; Rose et al., 2015; Smith et al., 2015; Wiederhold et al., 2010;Zheng and Hintelmann, 2010a, 2010b). Historical Hg isotope composition of anthropogenic emissions can aid our understanding of the Hg source-receptor relationships and the redox cycling of Hg after it is emitted from sources. Such an inventory is the first step in incorporating Hg isotopes into state-of-the-science global Hg cycling models. This would in turn help constrain current poorly-known Hg fluxes and transformations such as volcanic Hg emissions, Hg wet/dry deposition, atmospheric Hg⁰ oxidation and Hg^II reduction, marine and terrestrial Hg re-emissions (Engstrom et al., 2014; Holmes et al., 2010;Huang and Gustin, 2015; Lindberg et al., 2007; Obrist et al., 2014; Pongprueksa et al., 2008; Pyle and Mather, 2003; Streets et al., 2011).

In addition to Hg emissions from different sources, accurate estimates of Hg isotope composition of anthropogenic emissions rely on two key factors: Hg isotope variation ranges of source Hg materials, and Hg isotope shifts between source Hg and emitted Hg. Over one decade of research has generated a large set of Hg isotope data in primary source materials (e.g., coal, cinnabar), and has advanced our understanding of several important Hg isotope fractionation processes during processing/combustion of source materials (e.g., coal combustion, cinnabar roasting) (Blum et al., 2014; Gray et al., 2013; Hintelmann and Lu, 2003; Sun et al., 2014a). Nevertheless, not all Hg emissions sources are covered and Hg isotope MDF during industrial processing is complex.

The goal of this work is to include Hg isotopes into the historically anthropogenic Hg emission inventory. We first compile the Hg isotope composition of source materials used by main human activities. Where possible, we estimate Hg isotope MDF during processing/combustion of source materials, following the framework of Sun et al. (2014a). A Monte Carlo approach is used to quantify the uncertainties of Hg isotope composition of anthropogenic emissions, following Streets et al. (2011). Finally, we provide new Hg isotope observations on bulk volcanic emissions that help better discriminate the natural and anthropogenic Hg emission end-members.

The vast majority of published Hg isotope data use delta notation with¹⁹⁸Hg as the denominator relative to Hg standard solution NIST 3133:

MIF signatures are defined using capital delta notation as:

where XXX is Hg isotope mass number, (²⁰²Hg/¹⁹⁸Hg) sample is the measured isotope ratio of the sample, and (²⁰²Hg/¹⁹⁸Hg) _{NIST 3133} is the average isotope ratio of the bracketing NIST 3133 Hg standard solution during measurement. The mass dependent scaling factor β_xxx is 0.2520 for¹⁹⁹Hg, 0.5024 for ²⁰⁰Hg, 0.7520 for ²⁰¹Hg and 1.4930 for ²⁰⁴Hg (Blum and Bergquist, 2007). In this study, we use δ²⁰²Hg, Δ¹⁹⁹Hg, and Δ²⁰⁰Hg as the tracers of Hg isotope MDF, odd Hg isotope MIF, and even Hg isotope MIF, respectively.

Based on historical (1850s–2010s, with decadal resolution) Hg inventories (Horowitz et al., 2014; Streets et al., 2011), emission sectors are broadly divided into two categories: ‘by-product’ and ‘intentional Hg use’, which respectively contribute 112 Gt and 213 Gt cumulative anthropogenic emissions to the atmosphere since 1850s. For all the individual emission sectors in each category, their atmospheric total Hg (THg) emissions are partitioned as gaseous elemental Hg (GEM), gaseous oxidized Hg (GOM), and particulate bound Hg (PBM). δ²⁰²Hgⁱ_t and Δ¹⁹⁹Hgⁱ_t values of atmospheric speciated Hg emissions at a decadal year (t) are estimated as:

in which, the superscript i, s andr represents Hg species (i.e., GEM, GOM and PBM), sectors and regions, respectively. The Hg emissions in ‘by-product’ sectors (copper, zinc and lead smelting; iron and steel manufacturing; liquid Hg⁰ production; cement manufacturing; combustion of coal and oil; large-scale gold mining without Hg amalgamation) (Figure S1A) are divided into 17 world regions aggregated into 5 technological groupings based on levels of regional development. Emissions from the ‘intentional Hg use’ sectors (artisanal and small-scale gold mining (ASGM); silver mining; large-scale gold mining with Hg amalgamation; chlor-alkali production; and other 12 sectors that use liquid Hg⁰ in processes and products) (Figure S1B) only discriminate between developed and developing world regions. For all the ‘by-product’ sectors and selected ‘intentional Hg use’ sectors (large-scale gold mining, silver mining, ASGM and chlor-alkali production), their Hg species emission profiles are taken from Streets et al. (2011). For the remaining ‘intentional Hg use’ sectors of Horowitz et al. (2014), their Hg species emission profiles are calculated by weight-averaging emitted Hg species ratios at each emission stage of a complete life cycle. Mercury isotope composition in various source materials (δ²⁰²Hg^s,r_t and Δ¹⁹⁹Hg^s,r_t) are summarized from published literature (Figure 1), and the shifts of MDF and MIF in emitted Hg species (ε²⁰²Hg^i,s,r_t and E¹⁹⁹Hg^i,s,r_t) are estimated using the best-available information on Hg isotope fractionation during processing/combustion of source materials.

Figure 1 summarizes the reported δ²⁰²Hg and Δ¹⁹⁹Hg in source materials (e.g., fossil fuels, non-ferrous metal ores and crustal rocks). Δ²⁰⁰Hg in all source materials is insignificant (<±0.1‰). There is approximately 5.5‰ (−3.9‰ to 1.6‰) variation in δ²⁰²Hg in all source materials, with 80% of the values distributed between −1.84‰ and −0.06‰ (Figure 1). This large variation in δ²⁰²Hg of source materials reflects the wide variability of Hg sources and/or large MDF during biogeochemical processes including hydrothermalism, volcanism, diagenesis and metamorphosis. Δ¹⁹⁹Hg in all source materials varies by 1.0‰ (−0.6 to 0.4‰) with 80% of the values distributed between −0.25‰ and 0.11‰. Most of the Hg in ores and rocks is derived from Earth’s crust and mantle, and has insignificant Δ¹⁹⁹Hg (<±0.1‰) (Figure 1).

Most anthropogenic Hg emissions to the atmosphere are associated with high-temperature processing of source materials, e.g., combustion of fossil fuels, roasting/heating of gold/silver-Hg amalgam and cinnabar, roasting and smelting of ferrous and non-ferrous ores and concentrates. During the thermal processing of source materials, Hg in source materials is transformed into GEM, GOM and PBM in generated flue gases by a series of physicochemical reactions (oxidation, reduction, adsorption, diffusion etc.). These phase and redox transformations of Hg in high-temperature facilities are expected to induce significant Hg isotope MDF, but no photochemical MIF (Sherman et al., 2012; Sonke et al., 2010; Sun et al., 2013; Wiederhold et al., 2013). Therefore, we only address the Hg isotope MDF shifts of emitted Hg species relative to source materials. MDF and MIF of Hg isotopes during post-emission of volcanic and industrial plumes are also not further discussed here. Below we provide a detailed overview of Hg isotope composition in various source materials and associated Hg isotope MDF of emitted Hg species.

Combustion of fossil fuels. More than 200 coal samples from historical and modern world coal-producing/consuming regions covering nearly all coal-forming periods and coal ranks have been analyzed for their Hg isotope composition (Biswas et al., 2008; Lefticariu et al., 2011; Sherman et al., 2012; Sun et al., 2013, 2014a, 2014b; Yin et al., 2014a). In contrast, only a limited number of oil sand samples from Athabasca, Canada have been reported (Blum et al., 2012). World coal deposits show large variations in both δ²⁰²Hg (−3.90‰ to 0.77‰) and Δ¹⁹⁹Hg (−0.63‰ to 0.34‰), with an average value of −1.16±0.79‰ (1SD, n=216) and −0.11±0.18‰ (1SD, n=216), respectively. These coal samples are mainly from coal basins in China, former USSR, South Africa, Europe, India, Indonesia and Mongolia. Most of them are statistically distinguishable based on the means of δ²⁰²Hg, Δ¹⁹⁹Hg or both (Sun et al., 2014a). The very limited Athabasca oil samples show moderate variation of δ²⁰²Hg (−1.68‰ to −1.22‰, n=5) and large variation of Δ¹⁹⁹Hg (−0.39‰ to −0.02‰, n=5) (Blum et al., 2012) (Figure 1). We consider that more isotope data on oil are needed before this sector can be included in the isotope emission inventory.

Sun et al. (2014a) developed a speciated Hg isotope fractionation model for pulverized coal-fired boilers (the worldwide typical boiler type) by treating Hg isotope fractionation associated with Hg transformations in flue gases as two independent and consecutive Rayleigh fractionation systems: homogeneous gaseous oxidation of GEM to GOM, and heterogeneous gas-particle conversion of GOM to PBM. This model assumes Hg in coal is completely thermally reduced into GEM in the combustion zone of a boiler and Hg isotopes are only fractionated between the furnace outlet and inlet of air pollution control devices (APCD). Using a generalized GEM:GOM:PBM ratio of 0.15:0.52:0.33 in the flue gases before APCD, this model predicts a shift of 1.0±0.2‰ (1SD) for GEM, 0.05±0.2‰ (1SD) for GOM and −0.5±0.3‰ (1SD) for PBM, relative to δ²⁰²Hg of feed coal. However, the generalized pre-APCD GEM:GOM:PBM ratio of Sun et al. (2014a) does not differentiate the various sector/fuel/technology Hg emission scenarios that are characterized by different Hg species emission profiles. According toStreets et al. (2011), 69 out of 144 sector/fuel/technology combinations for combustion sources are designated to coal combustion, and all these combinations have their corresponding Hg species emission characteristics. Here, we improve the previous estimates on isotope composition of speciated Hg from coal combustion as follows: 1) APCDs are divided into five types: cyclone, scrubber, fiber filter (FF), electrostatic precipitator (ESP), flue gas desulfurization (FGD). It is assumed the former four types can only remove PBM, while the last type only removes GOM (Sun et al., 2014a). Each type of APCD is assumed to have a constant Hg removal efficiency (cyclone=1%, scrubber=6.5%, FF/ESP=30.6%, FGD=40%, Streets et al., 2011). 2) For the coal combustion sectors installed with APCDs, Hg species ratios in flue gases before APCDs are calculated based on emitted Hg species ratios and Hg removal efficiencies of APCDs. For the coal combustion sectors without APCDs, the emitted Hg species ratios are used directly to represent Hg species ratios before emission.

Ferrous (iron/steel) and non-ferrous metals (Cu, Pb, Zn, Au) production. In the present study, Hg emissions from large-scale Au production are quantified separately as ‘by-product’ (without Hg amalgamation) and ‘intentional Hg use’ (with Hg amalgamation) sectors, with the former from the trace Hg impurity in gold ores and the latter from the liquid Hg⁰ used for amalgamation. Up until now, only Hg isotope composition in zinc ores (sphalerite, from North America, China, South Africa, Spain, Australia and Congo, n=124) (Smith, 2010; Sonke et al., 2010; Yin et al., 2016) and gold ores (Carlin type, from USA, n=9) (Smith, 2010) have been reported. Due to lack of data, we assume Cu and Pb ores have the same Hg isotope composition as Zn ores based on their close geological association, commonly occurring together as hydrothermal sulfide ores. They possibly have common Hg sources. We further justify this by the similar median δ²⁰²Hg (and Δ¹⁹⁹Hg) of well documented cinnabar (−0.61‰) and Zn ores (−0.47‰) (Table S1). No data are available for iron ore processing by the iron/steel sector, and we therefore omit this relatively small (<1% historically) sector in the Hg isotope inventory.

During high temperature smelting, roasting or sintering of ferrous and non-ferrous metal ores, most Hg will be volatilized into flue gases (Wu et al., 2012; Zhang et al., 2012). Sonke et al. (2010) suggested >99% of Hg in raw sphalerite ore is volatilized during smelting processes, and the refinery slags (δ²⁰²Hg=−0.24 ± 0.71‰, 1SD, n=4) are only slightly enriched in heavier Hg isotopes as compared to raw sphalerite ores (δ²⁰²Hg=−0.76 ± 1.25‰, 1SD, n=7). Based on mass balance, the volatilized bulk GEM fraction should therefore have nearly the same isotope composition as raw sphalerite ores. Subsequent partitioning of bulk GEM to GOM and PBM during cooling of flue gases and by-product removal by APCDs are expected to fractionate Hg isotopes, but no studies have been performed to determine the signs and magnitudes. At current stage, we assume all emitted Hg species conserve the Hg isotope composition of sphalerite ores.

Cement production. Following UNEP/AMAP (2013), we assume Hg emissions from cement production are exclusively from limestone, although other raw materials (e.g., clay, sandstone) contribute small amounts of Hg. We note that cement Hg emissions from combustion of fueled coals have been separately accounted for in industrial sectors of coal combustion. Here we measured Hg isotope composition in bioclastic limestone used in some Chinese areas for cement production. δ²⁰²Hg in these limestone samples varies from −2.47‰ to −0.37‰, with an average value of −1.64±0.61‰ (1SD, n=11) that is very similar to the one limestone sample (δ²⁰²Hg=−1.43±0.10‰, 2SD) reported by Wang et al. (2015). Limestone is significantly depleted in heavier Hg isotopes compare to other crustal rocks (Figure 1).

Cement production involves clinker formation at very high temperatures (up to 2000 °C in the cement kiln) by roasting a mixture of limestone and silicate rocks (Wang et al., 2014), which should theoretically induce minimal Hg isotope fractionation between flue gases and raw materials. In addition, for some cement plants with particulate matter recycling and raw material preheating, more than 90% of the input Hg is emitted into atmosphere (Wang et al., 2014). We therefore assume Hg emissions have the same Hg isotope composition as limestone in Figure 1. This assumption was also validated in a typical cement plant, Sichuan Province, China where the emitted THg was calculated to have nearly the same Hg isotope composition as raw materials (Wang et al., 2015).

Liquid Hg⁰ production. Hg-bearing ores (predominantly cinnabar, hexagonal α-HgS) have been extensively mined for liquid Hg⁰ production. It is estimated that nearly one million tons of liquid Hg⁰ has been produced over the last 500 years (Hylander and Meili, 2003;Streets et al., 2011). Hg isotope composition in cinnabar ores from primary Hg mining areas producing >90% world historical liquid Hg⁰ have been reported, including the world’s largest Spanish Almaden mine, the world’s second largest Slovenian Idrija mine, the largest three Hg production districts of the USA (California Coast Ranges, Nevada, and Texas), the Central-South America Hg mining districts, and the Hg production center in China (Foucher et al., 2009; Gehrke et al., 2011; Gray et al., 2013; Hintelmann and Lu, 2003; Smith et al., 2005, 2008; Stetson et al., 2009; Wiederhold et al., 2013; Yin et al., 2013).

During liquid Hg⁰ production, crushed cinnabar ores are sealed in retort or rotary furnaces and roasted at high temperatures (600–850 °C) to convert the solid ore HgS phases into Hg⁰ vapor, which is subsequently condensed as liquid Hg⁰. This conversion is not complete. Historically, <1% to 40% of Hg is estimated to have escaped into the ambient atmosphere as GEM, and <10% of Hg is retained in the residual calcine (Hylander and Meili, 2003, 2005; Streets et al., 2011). Significant MDF of Hg isotopes has been observed between cinnabar and its roasting products (calcine, escaped GEM, liquid Hg⁰) in the field and in controlled laboratory experiments (Stetson et al., 2009; Wiederhold et al., 2013; Yin et al., 2013). Following Gray et al, (2013), we use an isotope mass balance equation to constrain the δ²⁰²Hg values of emitted GEM during HgS roasting and Hg vapor condensation:

In this equation, f’s represent the Hg mass fractions (%) of roasting products of cinnabar, summing to unity. The <10% residual Hg pool in calcine is commonly enriched in heavier Hg isotopes relative to cinnabar (Figure S2) (Stetson et al., 2009; Wiederhold et al., 2013; Yin et al., 2013). A comparison of mean δ²⁰²Hg between cinnabar ores and roasted calcine wastes at five individual Hg mines from US, Spain, and China shows a 0.1-1.9‰ (mean=0.8‰) positive shift of δ²⁰²Hg from cinnabar to calcine (Figure S2). This amounts to a <0.05‰ negative shift, on average, of δ²⁰²Hg in the generated bulk GEM that is subsequently cooled and condensed into liquid Hg⁰ at the condenser column/coils. Hg isotope fractionation at high temperatures is theoretically predicted to be minimal (Schauble, 2007). We therefore assume, as have others (Wiederhold et al., 2013; Yin et al., 2013), that the bulk GEM produced by cinnabar reduction approximately conserves the Hg isotope composition of cinnabar. Therefore, equation 5 can be approximated as:

During condensation of bulk GEM into liquid Hg⁰, the fraction of GEM unintentionally lost to the atmosphere may carry a different isotope signature relative to cinnabar and condensed liquid Hg⁰. Through a laboratory cinnabar retorting experiment, Gray et al, (2013) showed that on average, the condensed liquid Hg⁰ is shifted by +0.52±0.13‰ (1SD, n=3) and the GEM fraction lost is shifted by −0.79±0.44‰ (1SD, n=3) in δ²⁰²Hg relative to cinnabar ore (−0.36±0.10‰, 1SD, n=3). A direct comparison of δ²⁰²Hg in world cinnabar (−0.66±0.73‰, 1SD, n=210) (Cooke et al., 2013; Foucher et al., 2009; Gehrke et al., 2011; Hintelmann and Lu, 2003; Smith et al., 2005, 2008; Stetson et al., 2009; Wiederhold et al., 2013; Yin et al., 2013) and commercial liquid Hg⁰ (−0.38±0.34‰, 1SD, n=13) (Blum and Bergquist, 2007; Estrade et al., 2009; Foucher et al., 2009; Laffont et al., 2011; Mead et al., 2013; Sonke et al., 2008) also shows a slight enrichment of heavier Hg isotopes in liquid Hg⁰. It appears that the emitted GEM and condensed liquid Hg⁰ are complementary Hg pools, and their difference in δ²⁰²Hg is controlled by the extent to which Hg isotopes fractionate between GEM and condensed liquid Hg⁰. Hg⁰ condensation takes place from a saturated Hg⁰ vapor on a cooled surface. We therefore assume that these conditions are more similar to Hg⁰ liquid/vapor equilibrium Hg isotope fractionation than to kinetic isotope fractionation into a vacuum (Estrade et al., 2009). Equilibrium Hg⁰ liquid/vapor isotope fractionation enriches the vapor in the lighter Hg isotopes relative to the liquid (Estrade et al., 2009; Ghosh et al., 2013), which can be expressed as:

The per mil fractionation factor ε²⁰²Hg_liq-vap has an uncertainty range of 0.9–1.4‰ (mean=1.1‰), as experimentally determined byEstrade et al. (2009) andGhosh et al. (2013). The observed δ²⁰²Hg difference (i.e., 1.31±0.31‰, 1SD) between GEM (reactant) and liquid Hg⁰ (product) during cinnabar retorting ofGray et al. (2013) is within the range of experimentally observed ε²⁰²Hg_liq-vap. As discussed above, the δ²⁰²Hg difference between global cinnabar and commercial liquid Hg⁰ is ∼0.28‰. To achieve this shift, the Equations 6 and 7 would give a f_GEM value of 20% when using ε²⁰²Hg_liq-vap of 1.4‰. This is in the range of previously estimated f_GEM (<1–40%) for the 20^th century (Hylander and Meili, 2003; Streets et al., 2011). The fraction of emitted GEM (f_GEM), can be calculated as a function of historical Hg emission factor, EF_Hg, of Hg production in each region:

Figure S3A shows EF_Hg in five world regions between 1900s and 2010s (Streets et al., 2011) and the calculated fractions of GEM emitted to the atmosphere (f_GEM). Before 1900,EF_Hg and f_GEM are identical to values in 1900 (Streets et al., 2011). Using a constant ε²⁰²Hg_liq-vap of 1.4‰ and the calculatedf_GEM, we simulate the evolution of δ²⁰²Hg in emitted GEM and condensed liquid Hg⁰ inFigure S3B. It shows that the variations of δ²⁰²Hg in both produced liquid Hg⁰ and emitted GEM vary over a very limited range (<0.3‰), depending on the f_GEM for each region. The predicted δ²⁰²Hg of liquid Hg⁰ is within the range measured for commercial liquid Hg⁰ (Blum and Bergquist, 2007; Estrade et al., 2009; Foucher et al., 2009; Laffont et al., 2011; Mead et al., 2013; Sonke et al., 2008).

Small Hg MIF (Δ¹⁹⁹Hg up to 0.14‰, and Δ²⁰¹Hg up to 0.09‰) due to the nuclear volume effect was observed during laboratory experiments of liquid Hg⁰ in equilibrium with its vapor (Estrade et al., 2009; Ghosh et al., 2013). This could potentially lead to small and opposite Hg MIF, < ±0.1‰, in emitted GEM and condensed liquid Hg⁰ during cinnabar retorting. Such small effects are close to measurement uncertainty, and were consequently not detected during the cinnabar retorting experiment of Gray et al. (2013). We therefore omit cinnabar retorting Hg MIF in the Hg isotope emission inventory, but suggest that further work in this direction is worthwhile.

Historically produced liquid Hg⁰ (cumulative 720 Gg in 1850–2010) was primarily used in Au/Ag mining in the late 19^th to early 20^th century, and then in commercial products and industrial processes (Horowitz et al., 2014;Streets et al., 2011). The Au/Ag mining uses liquid Hg⁰ to amalgamate Au/Ag ores. Mercury is evaporated by roasting the Au/Ag–Hg amalgam at temperatures >1000 °C to obtain Au/Ag (Robins and Hagan, 2012; Velásquez-López et al., 2010). This process involves a high-temperature, near-complete physical transformation of liquid Hg⁰ to GEM, and should induce insignificant Hg MDF (Laffont et al., 2011). We assume that the bulk Hg evaporated from Au/Ag-Hg amalgam conserves the Hg isotope composition of liquid Hg⁰ used for amalgamation. Liquid Hg⁰ and its derivatives are also widely used in commercial products (e.g., lamps, battery, medical devices, measuring devices), industrial processes (e.g., chlor-alkali production, vinyl chloride monomer, and pulp and paper) and agricultural fungicides and pesticides. Emissions of Hg to the atmosphere can occur during initial production, storage stages and final disposal processes (e.g., dumping, incineration) of Hg-containing products (Horowitz et al., 2014). Within one decade, most of the used liquid Hg⁰ in products and processes is thought to be volatilized to the atmosphere (Horowitz et al., 2014). Analogous to liquid Hg⁰ used in Au/Ag mining, the bulk Hg evaporated from these sectors also presumably conserve the Hg isotope composition of liquid Hg⁰. We use the measured isotope composition of commercial liquid Hg⁰ to represent those of intentional uses of liquid Hg⁰.

Following the previous uncertainty estimates of Hg emission inventories (Streets et al., 2011; Wu et al., 2010), we use the Monte Carlo stochastic simulator embedded in the Crystal Ball^TM software to quantify the uncertainties of Hg isotope inventory from anthropogenic emissions. Each set of the key input variables (isotope composition, emission-source isotope shifts) in individual decadal years and regions is best-fitted using probability distribution functions incorporated into the Crystal Ball^TM framework. When the number of data points of a variable is equal to or larger than 15, the distribution function that best describes the data characteristics of the variable is automatically fitted by Crystal Ball^TM. For the variables with limited data points, we use simple distribution functions (commonly triangle or normal) to describe these variables (Subramanyan et al., 2008; Wu et al., 2010). For example, both δ²⁰²Hg and Δ¹⁹⁹Hg values for world coals (n=216) are automatically fitted by Weibull distributions. Due to limited coal samples in India (n=12) and OECD Europe (n=3), normal and triangle distribution functions are used, respectively, to represent their data scatter. A list of assumed distribution functions of Hg isotope composition of source materials with their main descriptive parameters are shown in Table S1. These distributions are truncated at both sides using minimum and maximum values of corresponding variables, and then are combined with previously estimated distributions of Hg emission inventories (Horowitz et al., 2014; Streets et al., 2011) to simulate the isotope inventories of anthropogenic THg and speciated Hg emissions. The number of stochastic simulation trials is set at 10 000 to obtain the reliable forecasted results that are expressed in the forms of statistical probability distributions (e.g., median value, P50) bounded by 10% (P10)–90% (P90) confidence interval).

Some parameters are not quantified for uncertainties at present. Examples include Hg isotope shifts of emitted GEM relative to raw Hg ores during liquid Hg⁰ production, the historical variations of specific APCDs removal efficiencies during coal combustion, and the ratios of GEM:GOM:PBM in emission sources. Currently, Hg isotope data are lacking for the oil combustion and iron/steel metallurgy sectors, as well as for the roasting and smelting of Cu and Pb ores. These aspects were therefore omitted from the Hg isotope emission estimates. The uncertainty associated with these omissions is limited as the cumulative Hg emissions from oil combustion, Fe/steel and Cu/Pb/Zn production only represent <8% of total anthropogenic Hg emissions since 1850, compared to >92% from the coal combustion, liquid Hg⁰ production and the uses of liquid Hg⁰ in Au/Ag mining and industrial processes/products.

Figure 2 shows the estimated median values (P50) of δ²⁰²Hg and Δ¹⁹⁹Hg for THg emissions from ‘by-product’ sectors and all sectors (‘by-product’ + ‘intentional Hg use’) between 1850s and 2010s, which are bounded by 80% confidence intervals (P10–P90). The corresponding speciated Hg (GEM, GOM and PBM) emissions are shown in Figures 3A–C. Table 1 lists the estimated δ²⁰²Hg and Δ¹⁹⁹Hg for THg and speciated Hg emissions from ‘by-product’ sectors and all sectors in the 2010s. A full dataset of δ²⁰²Hg and Δ¹⁹⁹Hg for THg, GEM, GOM, PBM and oxidized Hg^II (weighted mean of GOM and PBM) is tabulated in Table S2. Δ²⁰⁰Hg of all historical THg and speciated Hg emissions are <±0.1‰.

The isotope inventory of THg emissions from by-product sectors show an increasing trend in δ²⁰²Hg from −2.1‰ in 1850s to −0.9‰ in 2010s (Figure 2). This curve can be broadly divided into three phases: 1) 1850s-1930s with δ²⁰²Hg slightly increasing from −2.1‰ to −1.3‰, representing a transition of dominant Hg emission sectors from liquid Hg⁰ production and use to coal combustion and non-ferrous metals smelting; 2) 1930s-1960s with δ²⁰²Hg stabilized between −1.3‰ and −1.5‰, representing an insignificant variation in Hg emission amounts from different sectors; 3) 1960s–2010s with δ²⁰²Hg increasing from −1.5‰ and −0.9‰, representing a rapid increase of Hg emissions from coal combustion (Figure S1A and Table S1). δ²⁰²Hg values of GEM (−2.1‰ to −0.8%) (Figure 3A), GOM (−2.1‰ to −0.8‰) (Figure 3B) and PBM (−2.0‰ to −1.4‰) (Figure 3C) of by-product sectors from 1850s to 2010s generally follow the same increasing trends as THg. However, as compared to THg, PBM in recent decades is as large as 0.5‰ lower in δ²⁰²Hg. This is because δ²⁰²Hg of PBM predicted by the coal combustion MDF model is significantly lower than THg (Sun et al., 2014a). In contrast to δ²⁰²Hg, Δ¹⁹⁹Hg of THg (Figure 2) decreases from −0.03‰ to −0.10‰ in the first 100 years of the industrial period (1850s-1950s) due to dominant Hg emissions from the combustion of European/North American coals that have significantly negative Δ¹⁹⁹Hg (−0.4‰ to −0.1‰, Table S1). Δ¹⁹⁹Hg of THg then gradually rebounds to −0.06‰ in 2010s due to increasing Hg emissions from the combustion of Asian coals (e.g., China, India) and the smelting of non-ferrous metal ores which both are characterized by near-zero Δ¹⁹⁹Hg (Figure S1A and Table S1). The temporal (1850s to 2010s) variation of Δ¹⁹⁹Hg in GEM (−0.02‰ to −0.04% for 1850s-1950s, then stabilizes between −0.04% and −0.03% until 2010s) (Figure 3A), GOM (−0.01‰ to −0.13‰ for 1850s–1950s, then increases to −0.07% in 2010s) (Figure 3B) and PBM of by-product sectors (−0.08‰ to −0.16‰ for 1850s–1950s, then increases to −0.05% in 2010s) (Figure 3C) are basically similar to that of THg.

Adding Hg emissions from intentional uses of commercial liquid Hg⁰ (δ²⁰²Hg=−0.4‰; Δ¹⁹⁹Hg=−0.02‰, Table S1) positively shifts the δ²⁰²Hg and Δ¹⁹⁹Hg of by-product THg emissions by 0.2–1.2‰ and 0.00-0.05‰, respectively. From the 1850s to 2010s, this results in a variation of −1.1‰ to −0.7‰ in δ²⁰²Hg, and −0.02‰ to −0.04‰ in Δ¹⁹⁹Hg for THg emissions from all sectors (Figure 2). The distinctly increasing δ²⁰²Hg and decreasing Δ¹⁹⁹Hg trends for by-product sectors are not clearly seen for all sectors (Figure 2), due to large proportions (80%–50% for 1850s–2010s) of THg emissions from intentional uses of commercial liquid Hg⁰ (Figure S1). The δ²⁰²Hg (−1.1 to −0.6‰) and Δ¹⁹⁹Hg (−0.02 to −0.01‰) curves of GEM (Figure 3A) from all sectors closely follow those of THg as most (∼90%) of Hg in the ‘intentional Hg use’ sectors is emitted as GEM. δ²⁰²Hg of GOM varies within a rather limited range of −1.1‰ to −0.8‰ without clear trend (Figure 3B), while δ²⁰²Hg of PBM shows a decreasing trend from −1.2‰ in the 1850s to −1.5‰ in the 1980s before rising to −1.1‰ in the 2010s (Figure 3C). Δ¹⁹⁹Hg trends for both GOM (−0.02‰ to −0.10‰ for 1850s-1950s, then increases to −0.06% in 2010s) and PBM (−0.05‰ to −0.13‰ for 1850s-1950s, then increases to −0.04% in 2010s) from all sectors are similar to by-product sectors, as only ∼10% fraction of THg in ‘intentional Hg use’ sectors is emitted as GOM and PBM.

In addition to the uncertainties of previous Hg emission inventories (Horowitz et al., 2014; Streets et al., 2011), the expanded uncertainties of the Hg isotope inventories include variability in Hg isotope composition of source materials, and uncertainty in Hg isotope MDF between speciated Hg emissions and source materials. Sensitivity analysis of Hg isotope inventory shows that the largest contributors to the variance of δ²⁰²Hg of THg emissions from all sectors are the historical Hg emission inventory from 1850s to 1920s, δ²⁰²Hg of liquid Hg⁰ used in industrial processes/products between 1930s and 1970s, and δ²⁰²Hg of coal after 1980s (Figure S4A). In addition, Hg isotope MDF during coal combustion (ε²⁰²Hg_Coal) accounts for as much as 10-25% of the variance of δ²⁰²Hg since the 1980s. For the variance of Δ¹⁹⁹Hg of THg emissions from all sectors, the largest contributors are Δ¹⁹⁹Hg of liquid Hg⁰ uses before the 1940s, and Δ¹⁹⁹Hg of coal after the 1950s (Figure S4B). Hg emissions from Cu/Zn/Pb smelting; cement production, and large-scale gold mining without Hg amalgamation only contribute ∼10% of variances in δ²⁰²Hg and Δ¹⁹⁹Hg.

Primary natural Hg emissions to the atmosphere result from soil degassing and volcanic activity. The Hg flux from passively degassing volcanic activity is relatively well constrained at 76 ± 31 Mg y^-1 (Bagnato et al., 2014). No Hg flux measurements exist for explosive volcanism. The Hg flux from explosive volcanic activity has been suggested to be larger, 600 Mg y^-1 (Pyle and Mather, 2003) based on a crude extrapolation of registered Hg deposition in a single ice core archive (Schuster et al., 2002). Global box model constraints on primary natural Hg emissions, including both volcanic emissions and soil degassing, suggest that the sum of these is unlikely to be >300 Mg y^-1 (Amos et al., 2015). We therefore suggest that the passive degassing volcanic flux of 76 ± 31 Mg y^-1 is a realistic estimate of annual total volcanic Hg emissions.

A single study examined the Hg isotope composition of fumaroles at the passively degassing Vulcano, Italy (Zambardi et al., 2009). A mean δ²⁰²Hg for bulk fumarole THg emissions of −0.74 ± 0.18‰ (2SD, n=4), together with insignificant MIF was observed. Here we include additional observations from five fumaroles at the passively degassing Merapi and Papandayan volcanoes, Indonesia. Bulk fumarole sample collection, processing and Hg isotope analysis were done in 2007–2008 following Zambardi et al. (2009) and results are summarized in Table 2. We find overall similar Hg isotope composition, based on nine fumaroles at the three studied volcanoes, with a mean δ²⁰²Hg and Δ¹⁹⁹Hg of −0.76 ± 0.22 ‰ (2SD, n=9) and 0.05 ± 0.06‰ (2SD, n=9), respectively. These values are very similar to those of geological cinnabar and silicate rocks (Figure 1). Compared to volcanic processes, hydrothermal processes can cause as large as 6‰ variation of δ²⁰²Hg in fossil and active hydrothermal systems due to vitalization of Hg⁰ vapor and active Hg redox reactions (Smith et al., 2005, 2008). The limited observations of δ²⁰²Hg and Δ¹⁹⁹Hg of bulk volcanic THg emissions overlap our estimated present-day anthropogenic THg emissions (Table 1). This may limit discerning natural from anthropogenic Hg emissions on a global scale, depending on whether bulk volcanic Hg emissions are further fractionated within volcanic plumes. A single observation on Hg speciation in an aged fumarole plume at Vulcano (Italy) suggests that GEM (δ²⁰²Hg, −1.7‰) and PBM (δ²⁰²Hg, −0.1‰) can indeed be fractionated mass dependently relative to bulk THg emitted (δ²⁰²Hg, −0.7‰) (Zambardi et al., 2009). Hg isotope fractionation in volcanic plumes is however not sufficiently understood to predict the δ²⁰²Hg of volcanic GEM emissions at this point.

In this study, we have constructed isotope inventories of total Hg and speciated Hg (GEM, GOM and PBM) emissions for both ‘by-product’ and ‘intentional Hg use’ sectors with a decadal-resolution from the 1850s to 2010s. More work is needed to constrain Hg isotope composition of primary source materials and Hg isotope shifts in MDF and MIF between emitted Hg species and source materials. In particular, data on Fe, Cu and Pb ore concentrates are lacking. Further experimental and observational work on Hg isotope MDF during coal combustion, metal refining and liquid Hg⁰ use is needed to confirm the presented conceptual models.

High-precision Hg isotope measurement of total gaseous Hg (TGM, GEM+GOM) and precipitation Hg has been reported for several N-American and European regions (Chen et al., 2012; Demers et al., 2013, 2015; Donovan et al., 2013; Fu et al., 2014; Gratz et al., 2010; Sherman et al., 2010, 2012, 2015; Wang et al., 2015). Global TGM is enriched in the heavy isotopes (δ²⁰²Hg=0.51 ± 0.41 ‰, 1SD) (Demers et al., 2013, 2015; Fu et al., 2014; Gratz et al., 2010; Sherman et al., 2010) compared to our model estimated GEM emissions (Table 1). It has been shown that foliar uptake of GEM enriches foliage in the light Hg isotopes by −2.9‰ in δ²⁰²Hg, driving residual GEM to higher δ²⁰²Hg values (Demers et al., 2013). This process is possibly responsible for the δ²⁰²Hg shift between GEM emissions and global TGM observations. As we start to understand Hg isotope fractionation during atmospheric Hg transport and deposition, our high-resolution Hg isotope MDF and MIF inventories may help explain the observed isotope composition of atmospheric Hg and natural receptors and archives that directly receive atmospheric Hg. This would bridge the gap of our understanding on Hg isotope differences between sources and receptors. Furthermore, our inventories are expected to be embedded into state-of-the-science biogeochemical cycle models of Hg to understand the complex interplay between primary natural and anthropogenic Hg emissions and between primary and re-emitted natural and anthropogenic Hg.

The Hg emission inventories of ‘by-product’ and ‘intentional Hg use’ sectors are publicly available from research group of Biogeochemistry of Global Contaminants in Harvard University: http://bgc.seas.harvard.edu/models.html. The stable Hg isotope (δ²⁰²Hg and Δ¹⁹⁹Hg) emission inventories of ‘by-product’ and ‘intentional Hg use’ sectors for THg, GEM, GOM, PBM and oxidized Hg^II (weighted mean of GOM and PBM) are available in Table S2 of the supplemental material.

© 2016 Sun et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.