Atmospheric mercury in the southern hemisphere tropics: seasonal and diurnal variations and influence of inter-hemispheric transport

Mercury is a toxic element of serious concern for human and environmental health. Understanding its natural cycling in the environment is an important goal towards assessing its impacts and the effectiveness of mitigation strategies. Due to the unique chemical and physical properties of mercury, the atmosphere is the dominant transport pathway for this heavy metal, with the consequence that regions far removed from sources can be impacted. However, there exists a dearth of long-term monitoring of atmospheric mercury, particularly in the tropics and southern hemisphere. This paper presents the first two years of 5 gaseous elemental mercury (GEM) measurements taken at the Australian Tropical Atmospheric Research Station (ATARS) in northern Australia, as part of the Global Mercury Observation System (GMOS). Annual mean GEM concentrations determined at ATARS (0.95 ± 0.12 ng m−3) are consistent with recent observations at other sites in the southern hemisphere. Comparison with GEM data from other Australian monitoring sites suggests a concentration gradient that decreases with increasing latitude. Seasonal analysis shows that GEM concentrations at ATARS are significantly lower in the distinct wet monsoon season 10 than in the dry season. This result provides insight into alterations of natural mercury cycling processes as a result of changes in atmospheric humidity, oceanic/terrestrial fetch and convective mixing, and invites future investigation using wet mercury deposition measurements. Due to its location relative to the atmospheric equator, ATARS intermittently samples air originating from the northern hemisphere, allowing an opportunity to gain greater understanding of inter-hemispheric transport of mercury and other atmospheric species. Diurnal cycles of GEM at ATARS show distinct nocturnal depletion events that are attributed to 15 dry deposition under stable boundary layer conditions. These cycles provide strong further evidence for the “multi-hop” model of global GEM cycling, whereby long-range transport is characterised by multiple surface depositions and re-emissions, rather than continuous transport over long distances. 1 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2017-307, 2017 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 21 April 2017 c © Author(s) 2017. CC-BY 3.0 License.

. Map of region surrounding ATARS. Composed in QGIS using Natural Earth dataset.
Primary calibration of this source took place twice each year using manual injections of mercury vapour. No change in the internal source permeation rate was detected over this period. Furthermore, standard additions of mercury are automatically introduced to the 2537X from the internal permeation source every 35 samples (∼3 hours) in order to verify GEM recovery performance.
Continuous hourly measurements of radon were sampled at 12 m using an ANSTO-designed and -built, 700 l dual-flow-5 loop two-filter radon detector (Whittlestone and Zahorowski, 1998;Chambers et al., 2011). This detector samples at 40 l min −1 through 25 mm high-density polyethylene agricultural pipe and has a lower limit of detection of 40-50 mBq m −3 . Calibrations are performed monthly by injecting radon from a 101.15 ± 4 % kBq 226 Ra source (delivering 12.745 Bq 222 Rn min −1 ), traceable to NIST standards. Instrumental background is checked every 3 months. Radon measurements were corrected for the response time of the instrument (Griffiths et al., 2016), although the main trends were not affected by this time correction. 10 Time-corrected radon data were then split into "fetch" and "diurnal" components by interpolating between minimum afternoon (12:00 to 17:00) values when atmospheric mixing is greatest and subtracting these interpolated values (fetch component) from the original signal, leaving the diurnal component (see Chambers et al., 2016a, for details).
Meteorological measurements are collected at ATARS using a standard automated weather station (AWS) operated by the Australian Bureau of Meteorology. Precipitation data were collected using a 203 mm tipping bucket rain gauge and daily totals 15 were summed to give cumulative season totals centred around a hydrologic year beginning 1st June. The temporal extents of what we define here as "wet seasons" were then determined using the method of Smith et al. (2008), whereby 15 % and 85 % of 5 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2017-307, 2017 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 21 April 2017 c Author(s) 2017. CC-BY 3.0 License. the total cumulative rainfall marked their onset and conclusion, respectively. The wet season of 2014-15 was further extended to include two 100+ mm rain events that took place in November and March.

Modelling
As the atmospheric equator changes its position relative to the geographic equator, we employed a system of passive tracers within the GEOS-Chem chemical transport model to help assess the impact of air originating from the northern hemisphere 5 (NH) on the site, based on the work of Holmes and Prather (in press). We use GEOS-Chem v10-01 driven by assimilated meteorology from the NASA Goddard Earth Observing System Forward Processing (GEOS-FP) data product, run at 2 • x 2.5 • horizontal resolution and 47 vertical levels from the surface to 0.01 hPa. Tracers with 90-day lifetimes were uniformly released from the surface in all model boxes poleward of 45 • latitude within each hemisphere. The atmospheric equator is then defined as the point where mixing ratios of tracers from the two hemispheres are equal. Tracer concentrations in surface air 10 over ATARS were saved as daily mean values in the model grid box containing the site (2 • latitude by 2.5 • longitude and an approximate atmospheric depth of 130 m). Increasing the number of grid squares over which tracer values were averaged did not significantly affect the results.
The NOAA Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) Model (Draxler, 1999;Draxler and Hess, 1998;Stein et al., 2015) was also employed to assess influences of air mass source regions. Global Data Assimilation System 15 (GDAS) 0.5 • meteorological reanalysis data were used to drive the model, and trajectories were initialised at 0.5 times the mixed layer height as determined by HYSPLIT. To reduce the influence of local daily variation in GEM concentrations on this analysis, back trajectories were calculated for each hour of the day rather than as a daily or part-daily mean. For each trajectory, air parcel coordinates were calculated every two hours and weighted per the corresponding GEM concentration.
These weighted values were then averaged over 0.5 • x 0.5 • grid cells. 20 3 Results and discussion

Overall means and seasonal trends
Measurements of GEM at ATARS began on 5th June, 2014 and were still ongoing at the time of writing. Instrument maintenance/downtime plus application of QC protocols, including calibration and standard additions, resulted in 68.1 % temporal measurement coverage during the first two years of operation (Fig. 2, Table 1). Concentrations are normally distributed across 25 this period with an overall mean of 0.95 ± 0.12 ng m −3 (1 standard deviation), which is within the range of long-term background GEM concentrations for the southern hemisphere as reported by Slemr et al. (2015).   , 2016). GEM measurements at ATARS were coincident with those reported by Sprovieri et al. (2016) for only the latter 6 months of 2014, a period spanning the late dry season and early wet season. Concentrations during this period were 1.02 ± 0.10 ng m −3 -higher than the overall mean at ATARS, though still lower than mean values reported for other tropical GMOS sites.
A seasonal trend is apparent in the GEM time series (Fig. 2), which shows higher concentrations during the dry season 5 compared to the wet. Wind sector analysis also shows distinctly different wind patterns between wet and dry seasons (Fig. 3).
During the wet season, ∼60 % of winds come to the site from a westerly direction, consistent with shifting of the ITCZ and associated low pressure systems towards northern Australia. In the dry season, south-easterly to north-easterly winds are more common (∼65 % between 30 • and 150 • ), although there is also a notable westerly element. Concentration distributions vary between seasons, with a larger fraction of values above 1 ng m −3 seen in the dry period. Within each season however, these 10 distributions do not change significantly with wind direction. Furthermore, the small percentage of winds arriving from the south-west show no change in GEM distribution, implying that the low mercury emissions from Darwin are not significantly impacting measurements and that overall trends are indicative of influences from the global atmospheric mercury pool rather than local sources.  An intensive study of these biomass burning events undertaken at ATARS during the early dry season in 2014 also confirmed 5 spikes in GEM concentration that were associated with biomass burning (Mallet et al., 2016;Desservettaz et al., in press). The distance to the fire and atmospheric dispersion, as well as vegetation type and associated mercury loading, were all identified as factors influencing the strength of these biomass burning signals. Desservettaz et al. (in press) calculated emission factors for GEM between 0.0035 and 0.032 g Hg per kg dry fuel, around 2 orders of magnitude higher than that determined by Andreae and Merlet (2001) over savannah grasslands. The fires observed by Desservettaz et al. were shown to be from scrubland fires rather 10 than grassland fires, excluding the possibility of direct comparison between the two results. With a full suite of greenhouse gas and aerosol measurements taking place at ATARS, further identification of smoke plumes and precise calculation of emission factors is possible in a manner that is comparable with previous studies.
Wet season GEM concentrations in 2014-15 were characterised by a steady, gradual decrease that reversed abruptly in early April shortly after the onset of the dry season (Fig. 2). GEM concentrations during the 2015-16 wet season saw a similar, though 15 much less distinct decrease over a shorter and drier season. Figure 2 also shows that fetch-component radon concentrations begin to drop in both years around September-October, which HYSPLIT trajectories show is coincident with air mass origin shifting away from the Australian continent and towards the northern Arafura and Timor Seas. Throughout the wet season fetch-component radon remains low, though not at baseline levels (Zahorowski et al., 2013;Chambers et al., 2016b), implying that there is still some terrestrial influence on incoming air masses from the Australian continent or surrounding islands to the 20 north. Wet season wind data (Fig. 3) confirm that the predominant fetch during this period is from the west, where the Timor Sea lies less than 2 km from ATARS. Air-sea exchange of GEM is complex, although the ocean is generally considered a sink for atmospheric mercury (Mason and Sheu, 2002;Song et al., 2015). The increase in GEM concentrations in the early 2015 dry season was coincident with a shift to largely terrestrial-influenced fetch, as evidenced by a coincident increase in fetch-component radon, however the timing offset between decreases in GEM and fetch-component radon in the early wet and late dry seasons suggests that air mass origin is not the only influence on wet season GEM decreases.
Within tropical regions, wet deposition has been shown to be a significant pathway for mercury from the atmosphere to 5 ecosystems, even in relatively low-mercury air and despite the low solubility of mercury in its elemental form (Fostier et al., 2000;Costa et al., 2012;Hansen and Gay, 2013;Shanley et al., 2015). mercury "rainout" -or the tendency for mercury rainwater loading to decrease with increasing precipitation -has also been demonstrated in Mercury Deposition Network (MDN) data in North America (Glass and Sorensen, 1999;Prestbo and Gay, 2009)  . Furthermore, increases in minimum relative humidity during the wet season may have an impact on aqueous oxidative chemistry, increasing the likelihood that GEM may be converted to the more reactive GOM or PBM and subsequently deposited to the surface. Mercury wet deposition is currently not being measured at ATARS; however, given the large differences in GEM trends between the wet and dry seasons, deposition measurements could help to highlight differing processes between these periods.

Daily variation
Short, significant troughs in GEM values can be seen in Fig. 2, down to a minimum value of 0.28 ng m −3 . These are more pronounced in the dry season, though still common during the wet. GEM recoveries from standard additions during these periods were investigated and remained within 10 % of expected values with no evident pattern throughout the day, implying the drops in observed GEM were due to natural phenomena and not a change in instrument GEM recovery. Atmospheric 20 mercury depletion events (AMDEs) and the mechanisms behind them have been well-documented in polar regions (Steffen et al., 2008), though other similar events have been observed within the mid-latitudes (Mao et al., 2008;Brunke et al., 2010;Engle et al., 2010;Moore et al., 2013;Morrison et al., 2015;Howard et al.). The mechanisms behind these mid-latitude depletion events are less clear and likely varied, with hypotheses such as chemical conversion of GEM to RM and subsequent deposition; transport of GEM-depleted air masses; or deposition of GEM from isolated atmospheric pools, being offered.

25
Closer inspection of the dips in GEM observed at ATARS reveals that they occur overnight and are particularly pronounced in the early hours of the morning, with a marked rebound following sunrise. Previous flux studies have shown that surface GEM fluxes over soils with mercury concentration at background levels (such as at ATARS, with mean soil mercury concentration of 18 µg kg −1 ) are generally bi-directional, with little controlling influence from soil mercury concentration (Agnan et al., 2016, and references within). Correlations with solar radiation and air temperature tend to lead to emission fluxes throughout the day to first-order approximation, nocturnal build-up of radon is indicative of atmospheric stability, with highest radon values indicating the most stable atmospheres. This follows the radon-based stability categorisation method described by Chambers et al. (2016a) and Williams et al. (2016). In the dry season (left), it can clearly be seen that the magnitude of nocturnal GEM depletion increases with increasing stability, and conversely, little-to-no depletion occurs under well-mixed boundary layers.
Wind directions for the well-mixed category shift from coastal (westerly) in the early evening to terrestrial during the night.

5
In contrast, wind directions for moderately-mixed to stable boundary layer categories are very similar to each other, shifting from a north-easterly to south-easterly direction shortly after sunset. Terrestrial fetches encompass this range of directions and the abrupt shift in wind direction has little impact on the rates of GEM depletion or radon accumulation under these stability categories. This shows that changes in advection of GEM from local source/sink regions are not responsible for observed depletion. Rather, we suggest that the observed depletion results from dry deposition of GEM over terrestrial surfaces. Under moderately-mixed categories are more indicative of the influence of ocean fetch than stability, as evidenced by wind directions of 273 ± 8 • for these two categories. For weakly-mixed and stable categories, wind direction shifts southerly and easterly 20 throughout the evening, from an oceanic fetch to a terrestrial fetch. It is not until this shift in wind direction occurs that GEM depletion is observed, once again suggesting that these depletion events are due to terrestrial deposition fluxes under nocturnal capping inversion layers. Such a phenomenon would have a significant impact on our understanding of long-range transport of mercury, implying that this transport is due to a "multi-hop" or prompt recycling process of surface deposition and subsequent reemission, rather than continuous transport over long distances (Selin, 2009). Future research into these depletion events will 25 be undertaken by using measurements of radon fluxes from soils in the region to infer nocturnal GEM fluxes in a manner similar to Obrist et al. (2006).

Long-range transport
With seasonal changes in the latitudinal position of the ITCZ, ATARS is periodically located north of the atmospheric equator (Hamilton et al., 2008) and so the possibility of interhemispheric transport to the site was also of interest. Figure 2 shows 30 the GEOS-Chem output for NH-released tracer concentrations at ATARS. Throughout most of the year -and consistently through the dry season -this value remains low, indicating that the site is far enough below the atmospheric equator to not be affected by transport of NH air. However, there are notable periods when this tracer value increases, along with coincident GEM increases. We arbitrarily defined air masses at the site to be significantly influenced by northern hemisphere air (herein termed "NH wet season") when the ratio of NH tracers to SH tracers was greater than 0.5 (ratio not shown). Under this definition, ATARS saw 13 NH-influenced days over three distinct periods, all during the wet season and indicated in the lower panel of Fig. 2. Hereafter, wet season data that exclude these periods of NH influence are termed "SH wet season".
The normalised frequency distribution of NH wet season GEM data is compared against those of dry season and SH wet season data in Fig. 5. Mean values for each were 1.08 ng m −3 , 0.97 ng m −3 and 0.90 ng m −3 , respectively. The differences 5 between these means were small but significant; Student's t-tests showed the minimum differences between the 95 % confidence interval of each mean to be 0.10 ng m −3 (NH wet -Dry) and 0.07 ng m −3 (Dry -SH wet). Comparison with log-normal probability density functions for other GMOS sites over the years 2013-14 (Fig. 4, Sprovieri et al., 2016) shows that GEM data sampled at ATARS are more closely related to those from other southern hemisphere sites, rather than tropical or northern hemisphere sites. This is likely due to the location of ATARS within the Maritime Continent -a region of high variabil-10 ity in the latitudinal position of the ITCZ -and its southerly latitude that places it outside this range and hence within the atmospheric southern hemisphere for most of the year.

11
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2017-307, 2017 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 21 April 2017 c Author(s) 2017. CC-BY 3.0 License. Air mass source transport to ATARS across seasons was further investigated using 5-day HYSPLIT back trajectories. For NH-influenced air masses, use of 5-day trajectories and the geographic equator was found to be a poor predictor of NH influence at this site, with only 1.2 % of these trajectories originating from within the geographical NH. This is likely due to the significant disconnect between the geographical and meteorological equators over the Maritime Continent during the wet season. As such, 10-day back trajectories were calculated for these periods. Figure 6 shows median, 10th and 90th percentile 5 GEM-weighted trajectory coordinates for 0.5 • x 0.5 • grid cells. During the dry season (top row), the influence of persistent high pressure cells across the Australian continent can be seen, with most air parcels flowing over central and north-eastern Australia. Changes to air mass source regions are seen with the southward movement of the ITCZ and associated low pressure cells that characterise the SH wet season (centre row). The differing GEM concentration distributions between the two seasons outlined earlier are further apparent in these two figures. For NH-influenced air masses (bottom row), this analysis shows that 10 most air masses -particularly those with the highest GEM concentrations -passed over the Indonesian archipelago. North of this, air masses moved over the South China Sea or Western Pacific Ocean, with little influence from terrestrial South East Asia. Given that Indonesia's population is greater than 250 million and its biomass burning season coincides with the Australian monsoon, it is likely that the observed increases in GEM concentrations in NH-influenced air masses are more indicative of anthropogenic or biomass GEM source influence from the Indonesian archipelago than the northern hemisphere background 15 source pool. Further investigation using chemical transport and mercury emission modelling is needed. Regardless, the current analysis shows that ATARS does observe air masses of northern hemisphere origin and that measurements of GEM and other 12 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2017-307, 2017 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 21 April 2017 c Author(s) 2017. CC-BY 3.0 License. Figure 6. 10th percentile (left), median (centre) and 90th percentile (right) of hourly GEM-weighted HYSPLIT trajectories for 0.5 • x 0.5 • grid squares. a-c are for dry season data, d-f for SH wet season data and g-i for NH wet season data. NH wet season map created using 10-day back trajectories, all others using 5-day trajctories.
atmospheric species during these periods may help to assess the effectiveness of transport models investigating hemispheric air exchange associated with movement of the atmospheric equator.

Conclusions
We present here the first two years of ongoing measurements of GEM taken in tropical Australia. Comparison with other Australian datasets suggests that a latitudinal gradient of GEM exists across the continent, with higher values towards the 5 equator. Air masses from the northern hemisphere were shown to intermittently impact the tropical site ATARS, with associated increases in GEM. Generally, the concentrations seen at ATARS were indicative of southern hemisphere rather than tropical air, as determined by comparison with other GMOS monitoring stations around the globe.
Seasonal variation in GEM was observed, with higher values observed in the tropical dry season compared to the wet. Spikes in GEM associated with biomass burning in the region were measured, taking place during the mid-to late-dry season. Wet season GEM showed a decreasing trend throughout 2014-15; this was apparent though not as pronounced in the drier 2015-16 season. The cessation of this downward trend coincides with shifts of air mass source regions from oceanic to terrestrial, however the reverse is not the case for the onset of this trend. It is likely that precipitation rainout or aqueous-phase oxidation 5 of GEM is responsible for this observed downward trend. Continued monitoring and wet deposition data may help to explain these seasonal features.
Daily cycles in GEM were observed at the site, characterised by nocturnal decreases in concentration followed by rapid increases after sunrise, then further decreases throughout the day. Differences in these daily trends between wet and dry seasons, along with associated changes in wind direction and stability, suggest that these nocturnal depletions are related to 10 dry deposition of GEM over terrestrial surfaces under increasingly stable boundary layers. Such a phenomenon would have a significant impact on our understanding of long-range transport of mercury, implying that this transport is due to a prompt recycling process of surface deposition and subsequent re-emission, rather than continuous transport over long distances.
Currently, multi-annual atmospheric mercury datasets for tropical and SH sites are rare, impacting the skill of regional and global models designed to further our understanding of the natural mercury cycle and its potential impacts on human and 15 environmental health. The value of measurements such as these is in comparisons with other similar measurements around the globe. As such, the addition of this site to monitoring networks such as the Global Mercury Observation System (GMOS) or the Asia Pacific Mercury Monitoring Network (APMMN) is important in achieving greater understanding of the mercury cycle, as it is currently only one of two monitoring sites located in the tropical eastern hemisphere.
Article 19 of the Minamata Convention commits parties to develop and improve anthropogenic mercury inventories; efforts 20 to monitor mercury and mercury compounds in environmental media; and modelling of mercury transport (including longrange transport and deposition), transformation and fate in a range of ecosystems. ATARS is uniquely positioned to enhance the information required for these monitoring and modelling activities.

Data availability
GEM data used for this publication are available from the GMOS data repository (http://gmos.eu/sdi/). Weather data are 25 collected and supplied by the Australian Bureau of Meteorology (http://www.bom.gov.au/climate/data-services/). Competing interests. The authors declare that they have no conflict of interest.
Acknowledgements. The authors would like to thank Mark Cohen for his assistance with HYSPLIT modelling and for Chris Holmes for supplying code and assistance for GEOS-Chem tracer modelling. This research was undertaken with the assistance of resources provided at the NCI National Facility systems at the Australian National University through the National Computational Merit Allocation Scheme Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2017-307, 2017 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 21 April 2017 c Author(s) 2017. CC-BY 3.0 License.