Our knowledge of the distribution of mercury in air over the Southern Hemisphere is mostly based on measurements made during ship cruises. According to the most comprehensive review of shipboard measurements made between 1990 and 2009 by Soerensen et al. (2012) mercury concentrations varied between 0.72 ng m-3 reported by Kuss et al. (2011) for the southern Atlantic Ocean and 2.20 ng m-3 observed by Xia et al. (2010) over the southeastern Indian Ocean. These data were collected in different areas during different seasons, typically over a period of 1 or 2 months. Only a few of these measurements were accompanied by measurements of tracers specific to anthropogenic pollution and the influence from the ship such as CO, nitrogen oxides, and particles. Consequently, the influence of sources such as from biomass burning, regional pollution, and pollution from the ship itself could not be properly filtered out from the data. Part of the reported variability may also be due to the use of frequently undeclared and non-uniform standard conditions under which these concentrations are reported. Mercury concentrations in ng m-3 are usually reported at a standard pressure of 1013 hPa and a standard temperature of 273.14 K. However, some researchers and organisations use 293.14 K or 298.14 K. Since the same concentrations reported at 273.14 and 298.14 K differ by almost 10 %, the non-uniform standard conditions alone would prevent the detection of the statistically significant decrease in annual median mercury concentrations at Cape Point from ∼ 1.3 ng m-3 in 1996 to below 1.2 ng m-3 in 2004 (Slemr et al., 2008). Lastly, averages and standard deviations are quite frequently quoted without the number of measurements on which they are based. This means that the averages or medians cannot be weighed by the number of the measurements. It also makes statistical tests for the differences of averages impossible. It is not surprising that, using such data, Soerensen et al. (2012) concluded that no significant trend in the Southern Hemisphere could be detected so far. While we agree with this conclusion, a qualification is required: the quality of the data used by Soerensen et al. (2012) does not allow detection of trends smaller than their variability, i.e. some 50 % or even more. Consequently, with trends of up to ∼ 2 % per year (Slemr et al., 2008; Ebinghaus et al., 2011), it would take several decades to detect trends from measurements onboard ships.

Recently, mercury has been measured continuously at several remote sites in the Southern Hemisphere over periods of a year or more. In this paper we will compare these measurements in terms of their monthly and annual statistics. We selected stations which are either baseline stations (Amsterdam Island, Troll research station in Antarctica) or where additional measurements (e.g. CO, 222Rn, wind direction, aerosol) allow us to filter out baseline conditions (Cape Point and Cape Grim). The results show that atmospheric mercury is more uniformly distributed over the Southern Hemisphere than the measurements onboard ships suggest. Stationary sites with continuous and reproducible measurements of higher quality over longer periods allow for the detection of smaller trends in shorter time periods.

Figure 1 shows the location of the sites whose data are used in this paper: Amsterdam Island, Cape Grim, Cape Point, Troll research station, and Galápagos Archipelago.

The Cape Point site (CPT, 34∘21′ S, 18∘29′ E) is operated as one of the Global Atmospheric Watch (GAW) baseline monitoring observatories of the World Meteorological Organization (WMO). The station is located on the southern tip of the Cape Peninsula within the Cape Point National Park on top of a peak 230 m a.s.l. and about 60 km south of Cape Town. The station has been in operation since the end of the 1970s and its current continuous measurement portfolio includes Hg, CO, O3, CH4, N2O, 222Rn, CO2, several halocarbons, particles, and meteorological parameters. The station receives clean marine air masses for most of the time. Occasional events with continental and polluted air can easily be filtered out using a combination of the CO and 222Rn measurements (Brunke et al., 2004). Gaseous elemental mercury (GEM) was measured by a manual amalgamation technique (Slemr et al., 2008) between September 1995 and December 2004 and has been measured by the automated Tekran 2537B instrument (Tekran Inc., Toronto, Canada) since March 2007. Only the Tekran data are reported here. These data were obtained in compliance with the standard operating procedures of the GMOS (Global Mercury Observation System, www.gmos.eu) project. The instrument has been run with a 15 min sampling frequency. For data analysis, 30 min averages were used. On average, 30 % of the data were classified as baseline using the 222Rn≤250 mBq m-3 criterion.

Amsterdam Island (AMS, 37∘48′ S, 77∘33′ E) is a small isolated island (55 km2) located in the Indian Ocean 3400 km east of Madagascar. AMS is a GAW global station established in 1967. The climate of Amsterdam Island is mild oceanic, with frequent presence of clouds. Measurements are performed at Pointe Bénédicte station, which is located 2 km west of the Saint Martin de Viviès base on the edge of a cliff 55 m a.s.l. (GPS coordinates: 37∘48′ S, 77∘33′ E). GEM has been measured using a Tekran 2537B connected to a Tekran 1130/1135 speciation unit since January 2012 with a 5 min sampling frequency. For data analysis, 1 h averages were used. Details on operation and calibration procedures are given in Angot et al. (2014) and follow GMOS standard operating procedures. The station receives clean marine air masses almost all the time.

The Norwegian Antarctic Troll research station (TRS) is located in Queen Maud Land at 72∘01′ S and 2∘32′ E at an elevation of 1275 m and about 220 km from the Antarctic coast. The station has been in operation since January/February 2007 and its current continuous measurements include mercury, CO, O3, particles, greenhouse gases, hydrocarbons, persistent organic compounds (POPs) and meteorological parameters (Hansen et al., 2009; Pfaffhuber et al., 2012). Mercury has been measured using the Tekran 2537B instrument since February 2007 with a 5 min sampling frequency. For data analysis, 1 h averages were used. The original mercury concentrations were reported at a standard temperature of 293.14 K and were converted to the standard temperature of 273.14 K to be comparable with all other data reported here.

The Cape Grim Baseline Air Pollution Station is located on the north-western coast of Tasmania, Australia (40∘41′ S, 144∘41′ E, Fig. 2). The Cape Grim Baseline Air Pollution Station was established in 1976 to monitor and study global atmospheric composition and is part of the WMO GAW programme. Measurements at Cape Grim include greenhouse gases such as CO2, CH4, N2O, O3, reactive nitrogen oxides, stratospheric ozone depleting chemicals such as chlorofluorocarbons (CFCs), radon, and GEM. The Tekran 2537A instrument was run with 5 min sampling time. For data analysis, 15 min averages were used. Additionally, meteorological parameters are measured, such as wind speed and direction, rainfall, temperature, humidity, air pressure, solar radiation, along with condensation nuclei (CN) concentration (particles greater than 10 nm), ultrafine condensation nuclei concentration (greater than 3 nm), aerosol absorption, aerosol scattering, cloud condensation nuclei concentration and rainfall chemical composition. Baseline conditions are defined as those with wind directions at 50 m altitude lying between 190 and 280∘. In addition, CN should be less than a threshold concentration determined from 5 years of CN data for the current month based on the 90th percentile of CN hourly medians for this period, interpolated using cubic splines to give daily values (Fig. 2). During 2011–2013, the station received baseline marine air for 33 % of the time.

All mercury measurements reported here were made by an automated dual channel, single amalgamation, cold vapour atomic fluorescence analyser (Tekran-Analyzer model 2537 A or B, Tekran Inc., Toronto, Canada). The instrument features two gold cartridges. While one is adsorbing mercury during a sampling period, the other is being thermally desorbed using argon as a carrier gas. Mercury is detected using cold vapour atomic fluorescence spectroscopy (CVAFS). The functions of the cartridges are then interchanged, allowing continuous sampling of the incoming air stream. The instrument can be combined with a speciation unit (Tekran 1130/1135) consisting of a denuder, aerosol filter and pyrolyser that enables a determination of GEM, gaseous oxidised mercury (GOM), and particle-bound mercury (PM, < 2.5 µm) typically every 2–3 h (Landis et al., 2002). Operation and calibration of the instruments follow established and standardised procedures (e.g. Steffen and Schroeder, 1999). All mercury concentrations reported here are given in ng m-3 at 273.14 K and 1013 hPa.

In this paper we compare measurements at different sites in terms of monthly and annual average and median concentrations. Random uncertainties of individual measurements will average out and all we have to discuss are thus the systematic uncertainties, i.e. biases. The Tekran analyser is a complex instrument and the systematic uncertainties of its measurements depend on the operation procedure, the performance of the instrument, and the experience of its operators. All instruments used in this study are equipped with an internal mercury permeation source that is used to check and adjust periodically the instrument span and zero, typically every 25–72 h depending on the standard operating procedures that are used. This periodical internal calibration removes drifts both in span and zero that are caused mostly by the temperature and ageing of the fluorimeter lamp. The permeation rate of ∼ 1 pg Hg s-1 is, however, too low to allow a gravimetric determination of the permeation rate within a reasonable time period, as is usually done when certifying permeation devices for other gases (Barratt, 1981). Consequently, the permeation rate is calibrated every 6–12 months by repeated injection (at least 10 injections) of known volumes of gas saturated with mercury vapour at a known temperature. A skilled operator can achieve an individual injection precision of ∼ 3 %, resulting in an uncertainty of ∼ 1 % for 10 injections. The flow rate uncertainty of ∼ 1 % represents the second major contribution to the overall systematic uncertainty (Widmer et al., 1982). Adding smaller contributions from uncertainties associated with the injected volume and the temperature of the mercury vapour saturating device yields an overall systematic uncertainty of ∼ 3 %. We consider this to be the lower limit of the overall systematic uncertainty because this estimate assumes ideal performances of the instrument, its internal permeation device, the calibration mercury vapour saturating device, the injection syringes, as well as of the instrument operators.

A comprehensive analysis of all random and systematic uncertainties involved in a single manual determination of mercury concentration in air is given by Brown et al. (2008), who estimated the combined relative uncertainty to be 16.7 % at the concentration of 1.2 ng m-3. This uncertainty includes the uncertainty from different published mercury vapour pressure curves and can be reduced to 12.6 % when one vapour pressure curve is accepted to be correct, as is the case here. This uncertainty analysis, however, is not directly applicable to measurements with the Tekran instrument because most items in the uncertainty budget are random rather than systematic. The combined systematic uncertainty (square root of the sum of uncertainties in quadrature) from uncertainties in flow calibration (2 %) and detector calibration (7 %) would be ∼ 7 %. Since one vapour pressure curve was used, the 5.5 % uncertainty in the saturated mercury concentration can be neglected. The overall systematic uncertainty would then be ∼ 3 % and is comparable to our estimate.

Contributions of deviations from an ideal performance, such as slow deactivation of the traps, difference between the concentrations from the two traps, contamination of the switching valves and traps, and leaks (Steffen et al., 2012), are difficult to quantify. Thus we take published results of Tekran instrument intercomparisons as a measure of practically achievable systematic uncertainty. In an intercomparison described by Ebinghaus et al. (1999) three Tekran instruments that were operated side by side at Mace Head were biased by 0.02–0.11 ng m-3 (median 0.01–0.13 ng m-3) against each other. With an average concentration of 1.75 ng m-3, this represents the highest systematic uncertainty of ∼ 6 %. Two Tekran instruments were run side by side for 4 days at a site in Tuscany in June 1998 (Munthe et al., 2001) with an average bias of 9 %. Mercury was measured by five Tekran instruments for 28 days within a 6-week period in May and June 2006 at German EMEP station Waldhof (Aas et al., 2006). The median concentrations were 2.02, 1.88, 1.77, 1.70, and 1.69 ng m-3, and their average was 1.81 ± 0.14 ng m-3. The average bias was thus ∼ 8 % and the bias between the instruments with the lowest and highest readings was ∼ 18 % (related to the average concentrations). In summary, based on experimental evidence, we can expect an average systematic uncertainty of ∼ 10 %, in extreme cases up to 20 %.

Despite using the same instrumentation, the measurements may target different mercury species at different sites, depending on their configuration and/or local conditions. At Amsterdam Island the instrument was operated with the Tekran 1130/1135 speciation unit. It showed GOM concentrations of less than 5 pg m-3 representing less than 1 % of the total gaseous mercury (TGM) concentrations of ∼ 1 ng m-3 (Angot et al., 2014). The data for Amsterdam Island presented here are stated explicitly as GEM. The instruments at Cape Point, Cape Grim, and Troll research station are operated without speciation units but with PTFE (Teflon) filters to protect the instrument from sea salt and other particles. Although not proven, we assume that the surface active GOM in the humid air of the marine boundary layer at Cape Point and Cape Grim will be filtered out together with PM, partly by the salt particle loaded PTFE filter (denuders coated with KCl are used to adsorb GOM (Landis et al., 2002)) and partly on the walls of the inlet tubing. Consequently, we assume that measurements at Cape Point and Cape Grim represent GEM only and are thus directly comparable to those at Amsterdam Island. Although at Troll research station the same configuration with a PTFE filter is used, measurements by Temme et al. (2003) showed that at the low temperature and humidity prevailing at this site, GOM passed the inlet tubing and the PTFE filter. The measurements at Troll research station are thus assumed to represent TGM. As the GOM concentrations at Amsterdam Island in particular and in the marine boundary layer in general are below 10 pg m-3 (Soerensen et al., 2010; Angot et al., 2014), the difference between TGM and GEM at Amsterdam Island, Cape Grim and Cape Point is usually less than 1 %, which is insignificant when compared with the uncertainties discussed above. Consequently, GEM measurements at Cape Point, Cape Grim and Amsterdam Island are comparable to TGM measured at Troll research station. We caution, however, that recent studies have shown that the KCl-coated denuder in the Tekran speciation technique does not efficiently collect all GOM (Gustin et al., 2013; Huang et al., 2013; Ambrose et al., 2013). The bias between the TGM measurements at Troll research station and GEM measurements at all other stations can thus be larger.

The pair data difference tests were done using a t test (Kaiser and Gottschalk, 1972). A Mann–Kendal test for trend detection and the estimate of Sen's slope were made using the program by Salmi et al. (2002).

We compared mercury concentrations measured at Cape Point, Amsterdam Island, Cape Grim, and Troll research station in Antarctica. Amsterdam Island and Troll research station are background stations per se, and at Cape Point and Cape Grim the influence of local and regional pollution can be eliminated by using filters such as CO and 222Rn or wind direction and aerosol concentrations. No systematic difference was found between the unfiltered and filtered monthly median mercury concentrations at Cape Point and Cape Grim. We find that in terms of annual averages and medians the gradients of background mercury concentrations within the Southern Hemisphere are small and do not exceed 0.2 ng m-3. Taking into account a systematic measurement uncertainty of ∼ 0.1 ng m-3, the real variability could be as low as 0.1 ng m-3. This is much lower than the variability of shipboard mercury measurements on which the discussions of secular trends of mercury concentrations have relied so far. Consequently, smaller trends at shorter time periods can be detected by increasingly available long-term measurements at background sites in the Southern Hemisphere. The preliminary threshold of ∼ 0.1 ng m-3 for trend detection will further decrease when the comparability of the data sets improves.

The discussion of the measurement uncertainties shows a large difference between a small theoretical uncertainty and the much larger uncertainty achieved experimentally during several intercomparisons. Sampling flow rate can be precisely calibrated, and thus we believe that most of the “surplus” uncertainty comes from the behaviour and calibration of the Tekran internal permeation source. The issues related to the injection of known amounts of mercury are relatively well known (for example, not all syringes and replacement needles are suitable) and the uncertainty caused by them can be reduced by meticulous work. To the best of our knowledge we could not find any information about the dynamical behaviour of the internal permeation source that would enable one to calculate how much time is needed to stabilise the permeation rate (Barratt, 1981). Working practice, however, suggests that the time needed to stabilise the permeation rate increases with the decreasing permeation rate. We surmise that the very small permeation rate of the device in the Tekran instrument needs days rather than hours to stabilise within a 1 % margin required for precision measurements (Barratt, 1981). We thus conclude that the limited time of the cruises and the field conditions onboard ships are at least partly responsible for the large spread of the data from shipborne measurements.

We also report here an increasing trend for mercury concentrations at Cape Point for the period 2007–2013. No significant trend could be detected in mercury concentrations measured at Troll research station in Antarctica over the same period, but this is at least partly due to pronounced seasonal variations at Troll. As mercury concentrations at Cape Point decreased over the period 1996–2004, we conclude that the trend must thus have changed in direction between 2004 and 2007. Such change is qualitatively consistent with the trend changes observed at Mace Head in the Northern Hemisphere (Weigelt et al., 2014).