The Minamata Convention for mercury (Hg) has been signed by more than 120 nations and is now being ratified. The primary objective of the convention is to “protect human health and the environment from anthropogenic emissions and releases of mercury and mercury compounds” (UNEP Minamata Convention, 2014). A key challenge for Hg researchers is developing linkages between Hg in the atmosphere, deposition, and ecosystem contamination (Pirrone et al., 2013). Here we review where the science on measuring and modeling atmospheric Hg currently stands and offer suggestions for future research directions that will both advance understanding of Hg cycling in and between environmental reservoirs and better serve the needs of the convention.

Although the atmosphere is a relatively minor reservoir of Hg compared to oceans or soils, it is an important pathway by which Hg is distributed globally over short timescales (≤ 1 year). Atmospheric deposition represents the major pathway of Hg input to terrestrial and aquatic ecosystems outside areas of direct contamination. A variety of environmental archives, including remote lake sediments, ombrotrophic peat bogs, glacial ice, and tree rings, suggests Hg inputs to the atmosphere have increased several fold in the last 150 years (cf. Engstrom et al., 2014; Schuster et al., 2002; Wright et al., 2014a). Measured concentrations of atmospheric Hg have been declining over the last ∼ 15 years (Slemr et al., 2011; Cole and Steffen, 2010; Soerensen et al., 2012; Cole and Steffen, 2010; Cole et al., 2014), despite inventories suggesting global anthropogenic emissions have been relatively flat or increasing (AMAP/UNEP, 2013). This conundrum has challenged our understanding of Hg cycling and emissions, and underscores the need for continued atmospheric Hg monitoring.

Measuring the forms of Hg in the atmosphere is difficult. Mixing ratios are at low parts per quadrillion by volume (ng m-3 and pg m-3). Atmospheric Hg is operationally defined as gaseous elemental Hg (GEM), gaseous oxidized Hg (GOM), and particulate bound Hg (PBM) less than 2.5 µm in diameter (Lindberg et al., 2007; Schroeder and Munthe, 1998; Landis et al., 2002). GOM can be present as different forms (Huang et al., 2013, 2015). GOM and PBM have complex fundamental physiochemical properties. Because of the complexity, recent work has combined GOM and PBM concentrations as measured by the Tekran^® system and defined this as reactive Hg (RM = GOM + PBM) (cf. Rutter and Schauer, 2007a, b; Gustin et al., 2013; Weiss-Penzias et al., 2015). Previously it was thought that GEM was 95–99 % of Hg in the atmosphere (cf. Schroeder and Munthe, 1998). Recent work is pointing towards GOM being 25 % of total Hg in the boundary layer (see the discussion below). In the Arctic, up to 100 % conversion of GEM to GOM has been observed (Steffen et al., 2014, 2015). In addition, it has been demonstrated that there are different GOM compounds in the air (Huang et al., 2013, 2015).

Here we review current methods for measuring the forms of Hg in the atmosphere and models used to interpret these data. The advantages and limitations of each measurement method are discussed, and a narrative is provided on how we have arrived at our current understanding of the limitations. The number of models that have developed the capacity to simulate atmospheric Hg has multiplied in the last decade. We review major gains in Hg science gleaned from the use of measurements and models together, as well as key open questions. We conclude with a discussion of outstanding problems facing measurement and modeling communities.

In light of the new information about interferences affecting GOM measurements, we may begin to go back and re-examine features of past data that previously could not be explained. Here we explore Weiss-Penzias et al. (2003) as a case study. They measured GEM, GOM, and PBM at Cheeka Peak Observatory (Fig. 3), Washington, USA, in the marine boundary layer and found “air of continental origin containing anthropogenic pollutants contained on average 5.3 % lower GEM levels as compared with the marine boundary”. GOM and PBM concentrations in continental air were very low, 0–20 and 1–4 pg m-3, respectively. At the time, the results were “difficult to reconcile”. Now we see that the change in GEM concentrations during local anthropogenic pollution events relative to the mean of monthly marine air (-60 to -270 pg m-3) in Weiss-Penzias et al. (2003) are similar to the disparity in concentrations measured during RAMIX between the DOHGS and Tekran^® RM measurement.

Retrospectively, we suggest the observed differences between the two air masses reported can be explained by differences in the mix of oxidants and the resultant Hg(II) compounds formed. GOM and PBM were likely low due to lack of collection efficiency, interferences with O3, and loss in the sampling line (see the Supplement for details of sampling set up). Significantly lower GEM concentrations in the continental air are indicative of greater oxidation, which is supported by decreases in GEM concentrations coincident with O3 increases. Eastern Washington is covered by forests, which generate volatile organic compounds that could contribute to O3 and GOM formation. The marine air masses likely contained HgBr2 or HgCl2, and the continental air Hg-O, Hg-S, and Hg-N compounds associated with industry, agriculture, and mobile sources. The capture efficiency of HgBr2 and HgCl2 is greater than for O, S, and N compounds (Fig. 1; Table 2). The case study exemplifies how we can use the loss of GEM as a means of understanding the amount of GOM present or produced in air.

Here we discuss several key scientific advancements that have come from comparing models with speciated measurements, as well as the major questions left open by these studies. The number of atmospheric models capable of simulating speciated Hg has multiplied over the last decade (Table 3). Detailed discussion on model/measurement comparisons of RM can be found in Kos et al. (2013). Limitations and uncertainties of the models themselves have been written about at length in original research articles on model intercomparisons (Bullock et al., 2008; Pongprueksa et al., 2008; Lin et al., 2006). Fully acknowledging current limitations, there have still been huge strides made in our scientific understanding of the processes controlling GEM, GOM, and PBM cycling in the atmosphere including: marine boundary layer cycling, plume chemistry, source–receptor relationships, gas–particle partitioning, and vertical distribution.

Our understanding of speciated Hg cycling in the marine boundary layer (MBL) is one example of Hg science advancing as a result of using measurements and models in combination. GOM in the MBL has a diurnal pattern characterized by a midday peak and is depleted through deposition at night (Laurier and Mason, 2007; Laurier et al., 2003; Sprovieri et al., 2003). The use of observations and models together determined that the MBL has bromine photochemistry and was not affected by the hydroxyl (OH) radical. This drives the midday photochemical peak in GOM concentrations in the MBL and that scavenging by sea salt was driving rapid deposition at night (Holmes et al., 2009; Selin et al., 2007; Obrist et al., 2010; Hedgecock and Pirrone, 2001, 2004; Hedgecock et al., 2003; Jaffe et al., 2005; Laurier and Masson, 2007; Laurier et al., 2003; Sprovieri et al., 2003).

Model–observation comparisons consistently suggest models overestimate GOM surface concentrations, sometimes by as much as an order of magnitude (Amos et al., 2012; W. Zhang et al., 2012; Kos et al., 2013; Holloway et al., 2012; Bieser et al., 2014). The measurement–model mismatch is now understood as being partly explained by a low sampling bias (see Sect. 3), but this alone cannot reconcile the discrepancy. Reduction of GOM to GEM in coal-fired power plant plumes (Edgerton et al., 2006; Lohman et al., 2006) has been invoked as a possible explanation (Amos et al., 2012; W. Zhang et al., 2012; Kos et al., 2013; Holloway et al., 2012; Vijayaraghavan et al., 2008). The mechanism for in-plume reduction (IPR) remains speculative, hindering inference about how in-plume reduction may vary with coal type, control technology, or atmospheric composition. Results from recent field and laboratory data have been mixed, providing evidence for and against IPR (Tong et al., 2014; Landis et al., 2015) (Deeds et al., 2013). The speciation of anthropogenic emission inventories is also being revisited in order to reconcile model–measurement RM mismatches (Wang et al., 2014; Bieser et al., 2014). Improving our understanding of IPR and emission speciation has important implications for the efficacy of domestic regulation such as the US EPA Mercury Air Toxics Standard and for potentially attributing trends in Hg wet deposition over the USA (Y. Zhang et al., 2012).

Derived source–receptor relationships will also be sensitive to uncertainties in IPR and emission speciation. On the whole, Hg models simulate wet deposition fluxes better than surface GOM concentrations, contributing to the relatively high degree of consensus among source–receptor studies. A comparison of source–receptor studies found models agreed within 10 % in terms of the attribution of total wet Hg deposition to a given continental region (e.g., Europe, Asia) (AMAP/UNEP, 2013; Travnikov et al., 2010). Several source–receptor studies have concluded domestic US emissions contribute ∼ 20 % to total Hg deposition over the contiguous USA (Selin and Jacob, 2008; Corbitt et al., 2011). W. Zhang et al. (2012) found that including IPR in a model decreased the domestic contribution to wet deposition over the USA from 22 to 10 %.

An additional area of measurement–model study has been gas–particle partitioning of GOM and PBM. Understanding gas–particle partitioning is important because gases and particles are removed from the atmosphere by different physical processes. There is observational and laboratory evidence that gas–particle partitioning between GOM and PBM is driven by air temperature and aerosol concentrations (Rutter and Schauer, 2007a, b; Steffen et al., 2014; Rutter et al., 2008; Amos et al., 2012; Chen et al., 2014). Implementing temperature-dependent gas–particle partitioning in a global model increased simulated annual Hg deposition at higher latitudes (Amos et al., 2012). Aircraft observations suggest gas–particle partitioning also plays a major role in influencing the vertical profile of Hg, especially in the upper troposphere/lower stratosphere (Swartzendruber et al., 2009; Lyman and Jaffe, 2012; Murphy et al., 2006). Current gas–particle partitioning relationships are derived from surface data. PBM measurements from the summit of Mt. Bachelor suggest these relationships do not capture PBM dynamics aloft (Timonen et al., 2013). Effects of aerosol composition (Rutter and Schauer, 2007b), relative humidity, or even repartitioning of RM within the Tekran^® (see Sect. 3.3) could potentially contribute to this deficiency.

Oxidation also plays a central role in Hg cycling at the upper troposphere/lower stratosphere boundary. Comparisons against vertical aircraft profiles of TGM consistently suggest there is too little oxidation in models in the lower stratosphere (W. Zhang et al., 2012; Holmes et al., 2010). Observations show that total Hg is depleted in the lower stratosphere (Holmes et al., 2010; Lyman and Jaffe, 2012; Slemr et al., 2014), which is thought to be the result of rapid oxidation of Hg(0) to Hg(II), partitioning of Hg(II) to sulfate aerosol, and subsequent sedimentation of PBM (Lyman and Jaffe, 2012). Aircraft measurements over Washington and Tennessee, USA, found summertime GOM peaks between 2 and 4 km (Swartzendruber et al., 2009; Brooks et al., 2014). Modeled GOM vertical profiles over the USA have a less pronounced peak and generally place it higher (4–6 km) (Bullock et al., 2008). Correctly modeling the vertical distribution of Hg, particularly GOM and PBM, is essential for simulating deposition and hence Hg loading to surface ecosystems.

Chemistry remains one of the greatest uncertainties in Hg models. Improving measurements to determine the chemistry can help determine the mechanism(s) at play. There is still a general lack of rate coefficients and corresponding step-by-step reaction mechanisms available. The estimated tropospheric lifetime of RM against deposition and reduction is 40 days (Holmes et al., 2010), but the reduction pathway is highly uncertain (Subir et al., 2011; Pongprueska et al., 2008), and the burden of RM in the free troposphere is uncertain by at least a factor of 2 (Selin et al., 2008; De Simone et al., 2014). Improving our knowledge of the reduction and oxidation rates in the atmosphere will allow models to better capture the vertical distribution of Hg and in turn better simulate Hg deposition. The recent AMAP/UNEP (2013) assessment identified this as the highest priority for Hg models due to the importance in the Hg exposure pathway.

A persistent issue is the ambiguity in comparing modeled Hg(II) compounds to GOM and PBM, which are operationally defined. Models either have a lumped Hg(II) tracer or explicitly resolve individual Hg(II) compounds (Table 3). Since different Hg(II) compounds have different collection efficiencies by the KCl coated denuder (Fig. 1), this further confounds how to best construct a GOM-like model quantity to compare against observations. An active dialogue between experimentalists and modelers is encouraged as the community moves forward, so modelers may implement Hg tracers that emulate the Hg compounds measured.

Recent papers have used a three-fold correction factor to adjust the GOM concentrations measured by the Tekran^® system to calculate dry deposition using models in the western USA and Florida (cf. Huang and Gustin, 2015a; Huang et al., 2015). Use of this correction factor is based on the discrepancy between denuder measurements in the field and cation exchange membranes dry deposition measurements and concentrations collected using the UNR active system. Weiss-Penzias et al. (2015) found the GEOS-Chem model overestimated RM/GEM by a factor of 2.8 compared to Tekran^® RM/GEM, which is roughly in line with this correction factor. These field observations were collected in dry and humid conditions and at O3 concentrations typically observed in the atmosphere. Additional consideration could be based on the RH and O3 concentrations and the potential GOM compounds in the air.

Mercury is present in the atmosphere at pg m-3 to ng m-3, and the capability to measure it is a substantial analytical accomplishment. Ongoing measurements of atmospheric Hg will be key in evaluating the environmental benefit of regulation on behalf of the Minimata Convention.

Here we reviewed the current state of the science for measuring and modeling atmospheric Hg concentrations. Recent laboratory and field investigations have shown numerous artifacts and environmental interferences can affect measurement methods. Some environments such as those with low humidity and O3 may be less susceptible to sampling interferences than others. In light of new information about the limitations of sampling methods, we may revisit and better explain certain features of previous data sets and measurement–model comparison.

Fundamental research is needed on measurement methods and the atmospheric chemistry of Hg. We need to obtain agreement between several methods for understanding the chemical forms and compounds in the air. Only through comparison of multiple calibrated measurements can results be determined to be accurate.

Identifying the chemical compounds of RM in the atmosphere is a top priority. Understanding the final oxidation products are key for resolving questions regarding Hg chemistry. Knowing the dominant compounds would help with the design of measurement methods and determination of deposition velocities. Thermal desorption shows promise and mass spectrometry may be a way to verify compounds.

Development of a standard, field-deployable calibration system is needed. This system should provide spikes into ambient air and allow for studying sampling efficiencies and artifacts associated with ambient air. Lack of calibration is currently a major shortcoming.

A pyrolyzer should be used at the inlet of the 2537 if the goal is to measure TAM. The way the Tekran^® 1130/1135 system is configured to capture GOM first and then PBM is the best method to measure these two compounds. However, given the difficulty of separating GOM from PBM, we recommend interpreting the sum of RM instead of PBM alone until separation is improved.

A measurement system that collects GOM on a denuder material demonstrated to work for all compounds of GOM, and a separate measurement on a filter using a cation-exchange membrane could be used for measurement of GOM and RM. Then PBM could be determined by difference. Due to negative artifacts during long sampling times measurements should be done for < 24 h.

A new passive sampler design is needed that quantitatively determines concentrations and is calibrated. Use of a computational fluid dynamics model to help design the sampler could be one successful way forward. Passive samplers and surrogate surfaces have longer time resolution (1 day to weeks), but are relatively inexpensive and easy to operate and could provide an alternative measure of GOM concentrations and dry deposition fluxes in large-scale sampling networks once the above issues are resolved.