ATARS is situated on the Gunn Point peninsula (12.2491∘ S, 131.0447∘ E; Fig. ), approximately 20 km northeast from the suburban edge of Darwin 2013 population 136 200; in Australia's Northern Territory. Between 2 and 9 km to the north and west of ATARS lies the edge of the peninsula that gives way to the Tiwi Islands and Timor Sea, whilst the land to the east and south is largely uninhabited and includes national parks and conservation areas.

The climate in the region is best described as tropical Köppen category Aw, as reported by with mean monthly maximum temperatures between 30 and 33 ∘C 1941 to 2016 means; and a distinct monsoon (wet) season that coincides generally with the austral summer (December–February). The build-up to these monsoon seasons is characterised by steadily increasing minimum temperatures (19 ∘C in July to 25 ∘C in December) and associated increases in relative humidity (daily ranges of 37–60 % in July to 72–83 % in February). Mean annual rainfall is 1728 mm, with an average of 1604 mm (> 90 %) of this falling in the period November–April. As the site is located on a peninsula, a sea–land breeze cycle is often experienced in the dry season, resulting in mostly southeasterly winds throughout the morning, tending northerly as the sea breeze circulation sets in from the nearby coast. In the wet season, shifting synoptic patterns result in an increased frequency of westerly winds.

The vegetation classification is savannah with coarse grasses and scattered tree growth immediately surrounding the site. Burning of the grassed areas occurs frequently, with a fire return interval of 1–2 years. Direct mercury analysis see for methodology of vegetation within 500 m of the station gave total mercury concentrations of 7.23 ± 0.37 µg kg-1 (n=18) for grass and 21.09 ± 3.79 µg kg-1 (n=9) for tree litter. Sampling of soils in the same locations gave total mercury concentrations of 9.14 ± 0.58 µg kg-1 (n=18) in grassed areas and 26.49 ± 3.31 µg kg-1 (n=9) under forest canopy, confirming that soils in the area are categorised as background > 100 µg kg-1;. Sampling was undertaken in the early dry season, approximately 10–12 months after the last grass fire.

Anthropogenic emissions of mercury and its compounds to the atmosphere in and around Darwin are generally quite low. Australian National Pollutant Inventory (NPI) data for 2014–2015 state that six sites situated between 20 and 40 km from ATARS in the direction of Darwin (wind directions 190 to 240∘) emitted a total of 0.12 kg Hg to the atmosphere . Other distributed anthropogenic mercury emissions in Darwin are estimated at less than 0.2 kg a-1, based on 25 km × 25 km gridded population data .

Continuous (5 min sample) GEM measurements were obtained using a Tekran 2537X Automated Ambient Air Analyser (2537X). This instrument is housed in an air-conditioned structure with internal temperature set at 25 ∘C. Air is sampled from a 10 m high tower through 7.95 mm I.D. perfluoroalkoxy tubing using a Thomas 2688 vacuum pump drawing approximately 50 L min-1 (residence time 0.6 s). The 2537X subsamples from this flow at 1 L min-1 through 6 m of heated polytetrafluoroethylene (PTFE) line maintained at 50 ∘C, and two 0.2 µm PTFE filters positioned before and after the heated line. The 2537X operates on the principle of cold vapour atomic fluorescence spectroscopy (CVAFS) following gold amalgamation pre-concentration see for example. This technique quantifies total gaseous mercury (TGM = GEM + GOM); however, experience from other researchers suggests that the fraction of GOM in the atmosphere is generally small and removed upstream of the 2537X. As such we present the results here as GEM and not TGM, in line with reporting standards employed by other GMOS secondary sites . Reference volumes are reported at 1 atm and 0 ∘C.

Quality assurance and quality control procedures were applied as per protocols derived for GMOS sites . Calibration of the 2537X took place every 23 h using an internal mercury permeation source maintained at 50 ∘C. 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 h) 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-loop two-filter radon detector . 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 226Ra source (delivering 12.745 Bq 222Rn min-1), traceable to NIST standards. Instrumental background is checked every 3 months. Radon measurements were corrected for the response time of the instrument , although the main trends were not affected by this time correction. 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 seefor 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 were summed to give cumulative season totals centred around a hydrologic year beginning 1 June. The temporal extents of what we define here as “wet seasons” were then determined using the method of , whereby 15 and 85 % of the total cumulative rainfall marked their onset and conclusion, respectively. The wet season of 2014–2015 was further extended to include two > 100 mm rain events that took place in November and March.

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 NH on the site, based on the work of . 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∘ × 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 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 was also employed to assess influences of air mass source regions. Global Data Assimilation System (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 2 h and weighted per the corresponding GEM concentration. These weighted values were then averaged over 0.5∘ × 0.5∘ grid cells.

Measurements of GEM at ATARS began on 5 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 2 years of operation (Fig. , Table ). Concentrations are normally distributed across this period with an overall mean of 0.95 ± 0.12 ng m-3 (n= 130 312), which is within the range of long-term background GEM concentrations for the SH as reported by . Mean GEM concentrations reported by over 2012–2013 at Cape Grim, Tasmania (40.6832∘ S, 144.6899∘ E), and by over 2014–2015 at Singleton, NSW (32.4777∘ S, 151.1018∘ E), were both 0.86 ng m-3 (9 % lower), suggesting a slight latitudinal gradient in GEM across the Australian continent. These differences are statistically significant (Student's t test, p < 0.0001), though differences in the sampling periods introduces additional uncertainty due to seasonal variation at the sites. Further, an analysis of systematic instrument uncertainty for the Tekran 2537 by showed this to be ∼ 10 %. A latitudinal gradient within the SH was more generally seen in median annual GEM concentrations for GMOS sites in 2013–2014, based on data from five sites . GEM measurements at ATARS were coincident with those reported by 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. ), which shows higher concentrations during the dry season compared to the wet. Wind sector analysis also shows distinctly different wind patterns between wet and dry seasons (Fig. ). 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, southeasterly to northeasterly 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 distributions do not change significantly with wind direction. Furthermore, the small percentage of winds arriving from the southwest 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.

Figure  shows that the highest GEM values are concentrated into short peaks, clustered more heavily around the mid- to late dry season. In the absence of local anthropogenic sources, this is considered consistent with biomass-burning events and the associated release of mercury from volatilisation and thermal desorption from vegetation and soils . These biomass-burning events occur extensively in northern Australia throughout the dry season as the result of natural and accidental lighting, as well as part of local land management practices . concluded that burning in the northernmost part of Australia can contribute up to around 2 kg Hg km-2 a-1 to the atmosphere (2006 data, 25 km × 25 km grid resolution).

An intensive study of these biomass-burning events undertaken at ATARS during the early dry season in 2014 also confirmed spikes in GEM concentration that were associated with biomass burning . 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. 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 reported by and references within for savannah grasslands. The fires observed by were shown to be from scrubland fires rather 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 2014–2015 were characterised by a steady, gradual decrease that reversed abruptly in early April shortly after the onset of the dry season (Fig. ). GEM concentrations during the 2015–2016 wet season saw a similar, though much less distinct decrease over a shorter and drier season. Figure  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 , implying that there is still some terrestrial influence on incoming air masses from the Australian continent or surrounding islands to the north. Wet-season wind data (Fig. ) 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 mercury is complex, with the ocean generally considered a net sink for atmospheric mercury . Reduction of mercury within the photolytic zone can give rise to increased concentrations of elemental mercury and hence evasion of GEM to the atmosphere . Terrestrial surfaces are also commonly sources of GEM; modelled terrestrial mercury emission fluxes over Australia that were generally between 8 and 44 µg m-2 a-1 from soil and vegetation. Figure  does not show a strong difference in concentration distributions between the two source regions.

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, as well as with the conclusion of the monsoon season. 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 both oceanic and terrestrial ecosystems, even in relatively low-mercury air and despite the low solubility of mercury in its elemental form . 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 and positive correlations between GEM (TGM) and rainwater mercury have been reported in MDN data GEM; and at Cape Point, South Africa TGM;. Re-emission of any deposited mercury is likely to be inhibited throughout the wet season, as it has been shown that GEM emission from background mercury soils is suppressed when the soils are saturated . Mercury wet deposition is currently not being measured at ATARS; however, given the large differences in GEM trends between the wet and dry seasons, these measurements could help to highlight differing processes between these periods.

Short, significant troughs in GEM values can be seen in Fig. , 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 mercury depletion events (AMDEs) and the mechanisms behind them have been well documented in polar regions , though other similar events have been observed within the mid-latitudes . 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. 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.

The pattern of overnight GEM depletion is shown in diurnal composite data in Fig. , along with diurnal-component radon and wind direction. Days have been defined from midday to midday, then sorted into groups according to quartiles of the diurnal-component radon value at sunrise (marked in the top figures). As radon fluxes are, across daily timescales, constant 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 and . 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. In contrast, wind directions for moderately mixed to stable boundary layer categories are very similar to each other, shifting from a northeasterly to southeasterly direction shortly after sunset. Terrestrial fetches encompass this range of directions and the abrupt shift in wind direction at around 20:00 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.

Wet-season diurnal-component radon values (right) are lower than in the dry season, which fits with wind profile and fetch-component radon data showing greater influence of oceanic fetch. Additionally, during the wet season rates of radon emission may be reduced in saturated soils, as reduction of pore space inhibits upward mobility to the point where radon within the soil will undergo radioactive decay before reaching the surface . During the wet season, well-mixed and 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 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, at a similar rate to that seen in the dry season under moderately mixed to stable categories.

We suggest from these observations during the wet and dry seasons that the observed depletion results from deposition of GEM over terrestrial surfaces. Under increasingly lower capping inversions associated with more stable boundary layers, a near-constant rate of surface deposition would result in greater concentration drops within the boundary layer, consistent with the observations at ATARS. Turbulent break-up of the nocturnal boundary layer at sunrise is also consistent with the rebound of GEM concentrations and drop in diurnal-component radon observed at this time. The rebound of GEM, however, begins the hour before diurnal-component radon signals the break-up of the nocturnal boundary layer. In the absence of changes to advection or entrainment, this suggests emission of GEM from the surface. Furthermore, for stability categories where GEM depletion has taken place, daytime GEM concentration peaks at around 10:00 before decreasing to a minimum at around 15:00, where low radon values indicate the strongest turbulent mixing with free-tropospheric air. This “overshoot” of GEM in the early morning also cannot be explained by entrainment and, at least in the dry season, by changes to fetch. Early-morning GEM emission would likely be from the most readily volatile surface mercury, released under low-light conditions (shading denotes the period between geometric sunset/sunrise and so astronomical twilight will begin up to 75 min prior to the shaded edge). We propose that this initial release of GEM is volatilised from the reduction of mercury deposited overnight, as it has been shown that the most recently deposited mercury during AMDEs is preferentially released due to photochemical reactions .

Previous studies have shown that surface GEM fluxes over soils with mercury concentration at background levels are generally bidirectional, with little controlling influence from soil mercury concentration and references within. Correlations with solar radiation and air temperature tend to lead to emission fluxes throughout the day and deposition or near-zero flux overnight. , whilst undertaking micrometeorological measurements of surface GEM fluxes over a background mercury substrate grassland, observed nocturnal atmospheric mercury depletion events (NAMDEs) similar to the ones seen at ATARS. They attributed these events to enhanced nocturnal deposition of GEM under shallow, stable boundary layers. Enhancements in morning GEM emission were seen in days following the depletion events, similarly providing evidence for volatilisation of recently deposited mercury. Further, cumulative GEM exchange over the 20-day study was near zero, highlighting the short-lived nature of this nocturnal GEM sink. This result, and the radon-based analyses presented earlier, provides strong evidence for a “multi-hop” process of atmospheric transport.

NAMDEs have also been observed in the NH, in a range of ecosystems ranging from coastal to forested . attributed 70 % of their observed depletion to surface deposition and provided modelling evidence showing that stable boundary layers of height 100 m can be completely depleted of GEM due to deposition processes. The pervasiveness of NAMDEs across multiple ecosystems, and their pervasiveness throughout the ATARS time series across all seasons, suggests that this multi-hop process is widespread. It is important to note that, due to inhibited mixing at the top of the nocturnal boundary layer, the extent of any nocturnal depletion is limited to within tens to hundreds of metres above the surface. Beyond this, movement of free-tropospheric air continues to enable long-range transport of GEM. Nevertheless, extensive and rapid bidirectional exchange with the surface would have a significant impact on our understanding of atmospheric mercury transport, impacting the relative importance of intermediate and regional-scale sources, as well as expected timescales for observed decreases in environmental mercury following actions proposed under the Minamata Convention .