Introduction

Mercury (Hg) is a global pollutant that bioaccumulates in aquatic food webs and leads to health issues for humans and wildlife1,2. Human activities have greatly increased Hg inputs to the global environment mainly through mining and industrial activities3. Anthropogenic Hg emissions to the atmosphere are estimated to be around 2500 Mg y−1 4, exceeding natural Hg emissions of 340 Mg y−1 by sevenfold5. Gaseous elemental Hg0 emissions disperse globally, due to the relatively long atmospheric Hg0 lifetime (~5 months6), and can reach the Arctic by long-range atmospheric transport7. Deposition of atmospheric Hg to Arctic marine ecosystems, microbial conversion to methylmercury8,9 and subsequent biomagnification along marine food webs expose indigenous populations to Hg through their traditional diet of high trophic level seafood10.

Atmospheric Hg is distributed primarily in three different chemical forms: gaseous elemental Hg0, gaseous oxidized HgII and particulate bound HgII. Redox reactions between Hg0 and HgII in the atmosphere are mostly photochemically driven6,11, and both forms can deposit to terrestrial and marine ecosystems12. Many efforts have been made to characterize and understand the atmospheric transport, delivery and fate of Hg to the Arctic10. In 1998 Schroeder and collaborators reported on atmospheric mercury depletion events (AMDEs) observed in springtime in the Arctic at Alert, Canada13. AMDEs are driven by sea-salt derived reactive halogen oxidants in the atmosphere following polar sunrise14,15,16,17. During AMDEs, Hg0 is near-quantitatively removed from the atmosphere by oxidation to HgII forms which subsequently deposit rapidly to snow (on ice and on land) or sea ice18,19,20. Oxidized HgII species deposited during AMDEs can undergo photoreduction in snow where a large part (on average ~80%, observed at coastal sites10) is re-emitted as Hg0 to the atmosphere. The re-emitted Hg fraction is lower from snow on sea ice than from snow on coastal land21, due to the higher marine-derived concentrations of Cl- which inhibit HgII photoreduction22. The integrated summertime rebound in Hg0 at Alert represents 62% of the springtime drop in Hg0 over the period 1995–200220. Thus, a significant fraction of AMDE deposited Hg therefore remains in the snowpack and runs off with snow melt to impact freshwater and marine ecosystems23,24.

The springtime atmospheric Hg0 depletion (mean of 1.35 ng m−3 for April–May, 2000–2009 period) observed at different monitoring stations across the Arctic including Alert, Villum, Zeppelin, Utqiagvik, and Amderma is generally followed by a summertime Hg0 maximum of 1.80 ng m−3 (July mean25,26,27). This unique Arctic Hg0 seasonality suggests net Hg deposition in spring, followed by net Hg emission during summer28. The origin of the summertime Hg0 peak is less well understood, and has been attributed to AMDE re-emissions14, evasion from the AO29, and long-range transport of Asian air27. In 2012 the GEOS-Chem atmospheric Hg chemistry and transport model was used to assess Arctic Hg0 seasonality25. Using a number of sensitivity runs, it was found that neither cryosphere and ocean re-emissions, nor transport from mid-latitudes, could explain the summertime Hg0 maximum. The authors suggested that the missing source was 95 Mg y−1 of terrestrial Hg inputs to the AO from rivers and coastal erosion. In the model a large fraction of this terrestrial Hg is photoreduced in the surface AO and emitted to the atmosphere as Hg0 during the early summer (corresponding to the onset of the sea ice melt season) resulting in the Arctic summertime Hg0 maximum. Subsequent model improvement refined this number to 62–97 Mg y−1, divided between rivers (46–50 Mg y−1)28,30 and coastal erosion (16–47 Mg y−1)30,31,32. Pan-Arctic seasonal river Hg observations have confirmed a large river contribution of 41 ± 4 Mg y−1, delivered mostly during the spring flood in May-June33,34. The coastal erosion Hg flux remains uncertain, largely due to variation in the assumed glacial sediment Hg concentration and will be updated in this work.

In order to reproduce Arctic atmospheric Hg0 seasonality, coupled Arctic air-sea Hg models require an unusually large fraction (80%) of terrestrial Hg inputs to be photoreduced in the AO compared to 8% in other ocean basins30,34. In addition, coastal erosion inputs lag river inputs by one and a half months, peaking late August and September, when sea ice cover is minimal and wave action on coast lines maximal35. These two caveats put into question the role of terrestrial Hg in driving the summertime atmospheric Hg0 peak. Potential re-emission of terrestrial Hg to air from the surface AO has led to concerns that the large permafrost soil Hg pool (72,000 Mg in the upper 30 cm36) will be released to the global atmosphere with Arctic warming37,38. On the contrary, if terrestrial Hg is predominantly buried with sediments over the large AO shelf, then the predicted global atmospheric impact from river Hg and AO coastal erosion would be less. However, enhanced burial of terrestrial Hg in AO shelf sediments could possibly lead to increased in situ production of MeHg that can impact both benthic and pelagic marine food webs. As there is no observational evidence for re-emission of terrestrial Hg from the AO, its fate remains uncertain as well as its contribution to the higher summer Hg0 in the Arctic atmosphere.

In this study, we explore the seasonal variability of Hg stable isotope signatures of atmospheric Hg0 at Alert (ALT), Villum (VRS) and Zeppelin (ZEP) research stations (Fig. 1) to assess if the origin of the summertime Hg0 maximum can be understood. Mass-dependent (δ202Hg) and mass-independent Hg isotope signatures (Δ199Hg, Δ200Hg, Δ201Hg, Δ204Hg) provide a wealth of information on Hg sources and transformations39,40. Previous Hg isotope studies in the Arctic have shown uniquely large Δ199Hg, Δ201Hg fractionation during AMDE snow Hg re-emission41, Δ200Hg evidence for important tundra vegetation and soil Hg0 uptake42,43, a dominant atmospheric Hg0 source in snowmelt Hg runoff23, and biota Δ199Hg variability controlled by sea ice44,45. We compliment new seasonal atmospheric Hg0 isotope observations with additional isotope data on atmospheric HgII, snow HgII, Yenisei River dissolved HgII, and surface AO particulate HgII. The ensemble of new Hg isotope observations suggests that the Arctic summertime atmospheric Hg0 maximum is caused by AMDE re-emissions and not by emission of terrestrial Hg inputs to the AO via river run-off and coastal erosion.

Fig. 1: Arctic sampling locations and marine Hg isotope variability.
figure 1

Location of the three atmospheric research stations Alert (83°N, ALT), Villum (82°N, VRS), and Zeppelin (79°N, ZEP), and the city of Igarka, where the Yenisei River (YEN) was sampled (triangle symbols). Also shown are the surface Arctic Ocean (a) particulate Hg δ202Hg, (b) particulate Hg Δ199Hg and (c) particulate Hg Δ200Hg signatures (pHg, in ‰, round symbols) from the central basin and Barents Sea (this study) and from Fram Strait60. The figure was created using Ocean Data View80 with permission from Alfred Wegener Institute, Helmholtz Center for Polar and Marine Research (AWI) (https://odv.awi.de/).

Full size image

Results and discussion

Atmospheric Hg0 and HgII, and snow THg concentrations

Average Hg0 concentrations at ZEP, VRS and ALT stations during the 2018–2019 campaigns were 1.35 ± 0.24 (mean ± 1 SD; range 0.69–1.75) ng m−3, 1.11 ± 0.19 (0.51–1.45) ng m−3, and 1.33 ± 0.25 (0.58–1.68) ng m−3 respectively (Fig. 2a and Supplementary Data File). These concentrations are lower than the northern hemisphere (NH) mean of 1.51 ng m−3 for the year 2014 observed at 13 GMOS sites46, and lower than at ZEP (1.50 ± 0.13 ng m−3), VRS (1.40 ± 0.21 ng m−3), and ALT (1.40 ± 0.21 ng m−3) from 2011 to 201547. The low 2018–2019 Arctic Hg0 levels are compatible with continued declines in N-American and W-European Hg0 of −2% y−1 observed between 1990 and 2012, but contrast with stable Arctic Hg0 levels over the same period48. The global decline in atmospheric Hg0 has been attributed to decreases in Hg emissions from oceans49,50 and to decreases in anthropogenic emissions48. The lower Hg0 levels we observe for 2018–2019 therefore suggest that, despite a decade-long delay, the Arctic atmospheric Hg now follows the globally observed decline.

Fig. 2: Atmospheric Hg and Hg isotope seasonality in the Arctic.
figure 2

Time series of (a) Hg0 concentrations, b Hg0 and HgII Δ199Hg, c Hg0 Δ200Hg and d Hg0 δ202Hg at Alert (ALT), Villum (VRS) and Zeppelin (ZEP) research stations. The panel a also shows the monthly mean long-term (2000–2009) Arctic Hg0 seasonality for ALT, ZEP and Amderma (All Arctic) from25, with the summertime Hg0 maximum highlighted by the yellow shaded areas in all panels (vertical bars). The panel b also shows the predicted daily Eurasian river total Hg flux to the Arctic Ocean, based on34. Horizontal shaded areas (red) are the interquartile range of NH remote atmospheric Hg0 isotope variability. Dotted horizontal black lines indicate “0 per mil”. The error bars in b, c, d represent the analytical precision determined as SD (HgII) and 2 SD (Hg0), respectively from multiple measurements of an in-house standard.

Full size image

AMDEs are defined here as Hg0 levels below the 5th percentile of weekly mean Hg0, which are 0.92, 0.66 and 0.81 ng m−3 at ZEP, VRS and ALT, respectively. ZEP, at 474 m a.s.l., registered AMDEs during only four, short <12 h periods in April–May 2018, two times in Aug 2018, and on numerous prolonged occasions between November 2018 and May 2019. The ZEP polar winter dynamics are unusual, and did not occur over the 2000–2009 period51; they will not be further discussed here as they are beyond the scope of study. VRS, at 24 m a.s.l. in the planetary boundary layer, registered persistent AMDE events from April-June 2018, on three short occasions in Jan-Feb 2019, and again persistently from mid-March to May 2019. Numerous AMDEs at all sites were followed by short-lived increases in Hg0 above the 95th percentile of the data variability (Fig. 2a). These post-AMDE rebounds in Hg0 have been previously interpreted as photochemical Hg0 re-emission from snow HgII 20,52,53. Mean weekly Hg0 concentrations at ZEP and VRS show minor increases and a more pronounced rebound at ALT during July-August 2018 and June 2019 (Fig. 2a). Historically, the frequency of AMDEs, the springtime Hg0 minima and the summertime Hg0 maximum is strongest at ALT > VRS > Amderma (AMD) > ZEP and we observe the same trend here.

Weekly median atmospheric HgII concentrations (collected on CEMs at VRS and ZEP, and with a Tekran® 1130–1135 system at ALT) during AMDE weeks were higher at VRS, 243 pg m−3 (IQR 82 to 412) and ALT, 118 ± 46 pg m−3 than at ZEP, 38 pg m−3 (IQR 21 to 53) due to the frequent arrival of air masses at ZEP from the free troposphere. Hg concentrations in 2019 snow samples from Ny-Ålesund ranged from 0.14 to 21.3 (5.2 ± 7.0) ng L−1, which is similar to previous observations in the area54,55, but lower than snow Hg levels up to 373 ng L−1 during AMDE events53.

Hg0 stable isotopes and air mass back trajectories

Figures 2b, c, d and 3–5 show the Hg0 stable isotope variability of the dataset; NH background Hg0 isotope observations are indicated in Fig. 2b–d as horizontal shaded red bands, representing the IQR of published data. Hg0 at ZEP, VRS and ALT show overall similar positive median δ202Hg of 0.57‰ (IQR 0.45 to 0.75), 0.66‰ (IQR 0.47 to 0.85), 0.54‰ (IQR 0.39 to 0.64) and negative median Δ199Hg of −0.20‰ (IQR −0.28 to −0.08), −0.21‰ (IQR −0.31 to 0.10), and −0.31‰ (IQR −0.15 to −0.38) respectively (Figs. 2b–d and 4). Δ199Hg variability ranged from −0.50 to 1.32‰ at all three sites, which is uniquely larger than global Hg0 observations elsewhere (Figs. 2b and 4). Δ200Hg of Hg0 was persistently negative, and did not vary at the three sites with median values of −0.08‰ (IQR −0.10 to −0.07), −0.09‰ (IQR −0.13 to −0.03), and −0.07‰ (IQR −0.09 to −0.05) at ZEP, VRS and ALT respectively (Fig. 2c). Limited observations were made for Hg0 Δ204Hg in ZEP (2019 only) and ALT samples, and show median Δ204Hg of 0.16‰ (IQR 0.12 to 0.19) and 0.12‰ (IQR 0.10 to 0.14) respectively. Overall, we report a Δ200Hg/Δ204Hg slope of −0.45 (Supplementary Fig. 1), which is in agreement with published atmospheric Hg data40. Our Hg0 isotope observations broadly agree with the limited observations from Utqiagvik (formerly Barrow; n = 241), and ALT (n = 256), except for the case of Δ199Hg at Utqiagvik, discussed below.

Fig. 3: Comparison of the Δ200Hg isotope signature of different Arctic matrices.
figure 3

Toolik vegetation, soil, snow from42,43, Utqiagvik Snowfall and snow from41, Siberian tundra vegetation (lichen/moss) from81. The middle line in the box represents the median. The box limits indicate the upper and lower quartiles. The whiskers represent 1.5 times the interquartile range.

Full size image
Fig. 4: Stable isotope signatures of Hg in the global atmosphere and the Arctic environment.
figure 4

Mass-dependent (δ202Hg) and mass-independent (Δ199Hg) Hg0 isotope signatures (a) at different global locations and b for different Arctic pools. c Mass-independent (Δ201Hg) and mass-independent (Δ199Hg) isotope signatures for different Arctic pools. Arctic Hg0 Δ199Hg variability (−0.50 to 1.32‰) is much larger than the NH background Δ199Hg at remote sites. Reported literature data is summarized in the Supplementary Text.

Full size image
Fig. 5: Mean monthly variability in Arctic atmospheric Hg and air mass origins.
figure 5

a Mean monthly atmospheric Hg0 at Zeppelin (ZEP), Alert (ALT) and Amderma (Russia) stations from 2000 to 2009 (orange line), and mean monthly Hg0 (red line), Δ199Hg (black line) and Δ200Hg (yellow line) for ZEP,Villum (VRS) and ALT from this study for 2018–2019. Mean monthly air mass origins in the boundary layer (BL), in %, for 10-day HYSPLIT air mass trajectories ending at ZEP, VRS and ALT are shown for sea ice + snow-covered land (gray line) and open water (blue line). b Mean monthly δ202Hg (purple line), air mass origins over land (%, green line), in the free troposphere (FT, %, lightblue line), boundary layer (BL, %, darkblue line), and monthly pan-Arctic River and coastal erosion Hg inputs (Mg month−1). The vertical yellow shaded bar indicates the broad summertime Hg0 maximum. Error bars represent the 2 SD uncertainty for Δ199Hg and SD uncertainty for Hg0 concentrations.

Full size image

Figure 6 shows 10-day boundary layer air mass provenance maps for the months June, July and August at the three sites during both years of observation. June, July and August capture the start, peak and descent of the summertime atmospheric Hg0 maximum. Air mass provenance (hours per km2) is regional and does not originate over the Siberian shelf where the majority (88%34) of Arctic river and coastal erosion Hg inputs to the AO occur. Based on the HYSPLIT mixed layer depth (i.e., boundary layer height) and 10-day trajectory altitude profiles, we calculate that on average the air sampled at the three sites spent 21% of time in the boundary layer and 79% in the lower free troposphere (Fig. 5). The June air mass provenance within the boundary layer is associated by 62% with sea ice in the Lincoln Sea and snow-covered coastal land of Ellesmere Island and N-Greenland. The July Hg0 maximum and August descent remain associated with boundary layer sea ice and snow-covered land by 39%, but with an increasing proportion of boundary layer trajectories from Fram Strait, Nares Strait and Baffin Bay open waters (51%). Previous studies on the Arctic summertime Hg0 peak20,25,30,34,57 did not assess air mass origins. Seasonal air mass provenance was determined using 10-day HYSPLIT back trajectories in a pan-Arctic aerosol study58, finding similar lack of origins over Siberian coastal waters for ALT and VRS. A study on long-term Hg0 observations at VRS used 5-day back trajectories, finding no strong seasonal correlations between Hg0 and time spent over sea ice, open water, land or snow26. The study did not, however, produce seasonal air mass provenance maps for boundary layer air as we present here in Fig. 6.

Fig. 6: Sea ice extent and origin of air masses arriving at Zeppelin (ZEP), Villum (VRS) and Alert (ALT) research stations from June to August 2018.
figure 6

Sea ice extent (all panels, blue line) and air mass residence time maps (lower panels, hours per km2) for combined 10-day HYSPLIT back trajectories from ZEP, VRS and ALT. Residence time was calculated only for trajectories within the atmospheric boundary layer. The maps display sea ice extent and air mass residence time for a, d the 2018 June Δ199Hg maximum, b, e July summertime Hg0 maximum, and c, f August Δ199Hg minimum that were detected at all sites (Figs. 2 and 5). Trajectories in June 2018 and July 2018 originate for 62% and 39% over sea ice and snow-covered coastal land, with an increasing contribution of open water from the Fram Strait in July 2018 and August 2018 (52%). Sea ice extent data from June to August 2018 is obtained from the NSIDC NASA DAAC: National Snow and Ice Data Center82.

Full size image

Arctic Δ199Hg variability

The Δ199Hg signature is known to orginate from HgII photoreduction in water59 and in snow, in particular during and after AMDEs41. Pronounced seasonal Hg0 Δ199Hg variability can be seen in Fig. 2b relative to the remote NH median Δ199Hg of −0.20‰ (IQR −0.13 to −0.25, horizontal red shaded bar), with increased Δ199Hg from April to June in both years, and decreased Δ199Hg in August and September. Figure 2b also shows that at the height of the summertime atmospheric Hg0 maximum, Δ199Hg plunges to its minimal with negative values down to −0.5‰. Sherman et al.41 previously observed strong odd-Hg isotope MIF during post-AMDE snow Hg0 re-emissions, with Hg0 Δ199Hg fractionated by +3.3‰ relative to snow HgII. Such large MIF during photoreduction in snow suggests that post-AMDE Hg0 emission is capable of generating enough Hg0 with positive Δ199Hg to reach the Δ199Hg values of up to 1.4‰ observed in weekly Hg0 during the AMDE spring months. Sherman et al.41 also observed that as snow re-emits Hg0 with positive Δ199Hg, the residual snow HgII fraction progressively attains Δ199Hg as low as −5.1‰. We observe low snow Δ199Hg down to −2.4‰ in Ny-Ålesund in 2011 and 2019, and so did others in Toolik, AK (USA) and in ALT43,56. Snow Δ199Hg/Δ199Hg regression slopes are near 1.0 in our Ny-Ålesund snow data (Fig. 4c), as well as in the Utqiagvik, Toolik and ALT data indicating that a common photochemical mechanism is involved.

Atmospheric reactive HgII compounds (i.e., aerosol HgII and gaseous HgII compounds) at ZEP and VRS during the springtime AMDEs show negative odd-MIF, similar to snow HgII, with a Δ199Hg/Δ201Hg ratio of 1.00 ± 0.02‰ (1 SD, Fig. 4c). Reactive HgII Δ199Hg decreases progressively from April to June, and the minima down to −2.15‰; coincide exactly, in the same week of sampling, with the maximum Δ199Hg up to 1.32‰ of Hg0. The similar magnitude and Δ199Hg/Δ201Hg slope of odd-MIF in atmospheric reactive HgII compared to snow HgII suggests that the photoreduction mechanism and associated odd-Hg magnetic isotope effect that is active in snow41 also operates in the boundary layer, most likely in aerosols or blowing snow and ice crystals over sea ice and land. The synchronous, yet opposite Δ199Hg signs in atmospheric both Hg0 and HgII strongly suggest these signals to be produced in the boundary layer over sea ice and over land and not in surface AO waters. Further, the repetitive seasonal Δ199Hg variability, detected simultaneously at all three sites, suggests that the snow Hg0 re-emission with positive Δ199Hg is imprinted regionally on the entire boundary layer Hg0 pool, and possibly on the lower free troposphere.

At the height of the summertime Hg0 maximum in July, Δ199Hg sharply drops from its positive peak value in June to a negative minimum in July and August (Fig. 2b). This behavior is expected from AMDE Hg re-emissions that undergo strong odd-MIF41, where early snow Hg0 re-emissions in May and June carry positive Δ199Hg, leading to a residual snow HgII pool with progressively more negative Δ199Hg as low as −2 to −5‰. Sherman et al. used a Rayleigh fractionation model to show that beyond 60% of snow HgII re-emission, the final fractions of re-emitted Hg0 also attain negative Δ199Hg, which potentially explains the observations we make in July and August. Alternatively, we will see below from the air mass back-trajectory analysis that July and August Hg0 emissions are predominantly from regional marine waters. Since there is no evidence for MIF during marine HgII photoreduction globally60, it is also possible that the late summer shift to negative Hg0 Δ199Hg is inherited from AMDE snowmelt runoff, which carries negative Δ199Hg to marine waters. To support this, we detected isotopic signatures consistent with AMDE runoff in pHg from the surface ocean near the North East Greenland shelf in mid-August, with median Δ199Hg of −0.20‰ (IQR, −0.16 to −0.27, Fig. 1), which is well below the global marine Δ199Hg baseline for total Hg of 0.06‰ or for marine sediments of 0.08‰60.

Finally, we note that the two 3-day integrated Hg0 isotope observations at Utqiagvik, from mid-June 200841, have similar δ202Hg and Δ200Hg as observed in this study, but distinctly lower Δ199Hg of −0.1 to −0.2‰ than the positive June 2018–2019 Δ199Hg observed at ZEP, VRS and ALT. HYSPLIT back trajectories for the short mid-June 2008 observations (Supplementary Fig. 2) indicate mostly open water air mass provenance in the Bering and Chukchi Seas. Longer-term atmospheric Hg0 isotope observations need to be made at Utqiagvik to see if sea ice and coastal snow-covered areas lead to the high Δ199Hg observed further east.

Arctic δ202Hg variability

The ZEP, VRS and ALT Hg0 δ202Hg time-series show a large variability, from −0.4 to 1.45‰, but also substantial overlap with the median NH background of 0.43‰ (IQR 0.09 to 0.77; Fig. 2d). Overall, δ202Hg drops during springtime to reach a minimum of 0.38 ± 0.12‰ that coincides with the Δ199Hg maximum in June (Fig. 5). δ202Hg then progressively increases during summer to reach its maximum of 0.82 ± 0.23‰ in October. These observations are again compatible with the Rayleigh fractionation behavior of AMDE Hg re-emissions from snow that are initially, in May and June, enriched in the light Hg isotopes, and then become progressively heavier as the residual snow (on sea ice and on land) HgII pool becomes enriched in the heavy Hg isotopes. In addition, plant and soil uptake of Hg0 during the summertime growth season preferentially removes light Hg isotopes, leading to potentially higher δ202Hg in residual atmospheric Hg0 61,62. The overall positive Hg0 δ202Hg contrasts with Arctic terrestrial Hg δ202Hg. Siberian tundra vegetation and river dissolved dHg runoff shows negative δ202Hg of −1.0‰ (median, IQR −0.2 to −2.8) and −2.6‰ (median, IQR −2.3 to −4.0) respectively. Similar negative δ202Hg was observed recently for N-American river run-off in the Mackenzie delta (−1.1 ± 0.6‰ in dHg, −1.5 ± 0.4‰ in pHg, Fig. 4b)63. The Siberian and N-American terrestrial Hg inputs can also be observed in marine surface water pHg, which has low δ202Hg of −1.7 ± 0.4‰ in the transpolar drift current in the central AO and −2.3 ± 0.2‰ in the shallow Barents Sea off Scandinavia (Figs. 1, 4b and Supplementary Fig. 3).

The spreading of less dense river runoff over denser AO waters in estuaries and over the large AO shelf provides, in principle, the conditions where terrestrial HgII can be (photo-)chemically and biologically reduced and emitted to AO marine boundary layer air. The large amount of terrestrial HgII (80%) that needs to be photoreduced and emitted from AO surface waters in models30,34 should lead to minor further enrichment of emitted Hg0 in the light Hg isotopes to values below −1.1‰, and possibly lower. The ZEP, VRS and ALT Hg0 δ202Hg data however, do not show evidence of strong light isotope enrichment during the summertime Hg0 maximum (Figs. 2 and 4b). Rather, the observed June-August Hg0 δ202Hg of 0.58 ± 0.17‰ is more similar to δ202Hg observations of HgII in Arctic coastal snow associated with AMDEs of 0.43‰ (median, IQR 0.08 to 0.75, n = 58)41,43,56(this study). Similar to Δ199Hg, the atmospheric Hg0 δ202Hg observations are therefore compatible with a dominant contribution of AMDE deposited Hg, but not terrestrial Hg.

Arctic Δ200Hg variability

Δ200Hg, and the related Δ204Hg signature, are thought to be produced by photochemical Hg redox reactions at or above the tropopause64, and no relevant Hg transformations at the Earth’s surface have thus far been shown to further result in even-Hg MIF40. Hg redox transformations in the upper atmosphere lead to Hg0 and HgII pools with distinctly different median Δ200Hg0 of −0.05‰ (IQR −0.08 to −0.03) and Δ200HgII of 0.14‰ (IQR 0.09 to 0.1860). Δ200Hg is therefore considered a conservative tracer for atmospheric Hg deposition pathways61, and Δ200Hg of terrestrial surfaces and water bodies reflect the relative proportions of Hg0 and HgII deposition. For example, recent work has suggested that global soils, runoff and lake sediments, with Δ200Hg of 0.00‰, reflect ~75% vegetation and soil Hg0 uptake and ~25% HgII wet and dry deposition61,65. Similarly, pelagic marine waters, sediment and biota with average slightly positive Δ200Hg of 0.04‰ reflect 50% of ocean Hg0 uptake and 50% HgII wet and dry deposition60.

In Fig. 3, we compare the persistently negative Hg0 Δ200Hg of −0.08‰ at ZEP, VRS and ALT to observations of Δ200Hg in Russian and N-American tundra vegetation66, to Yenisei and Mackenzie River dHg and pHg, to surface AO pHg, and to atmospheric HgII and snow THg at ZEP, VRS and ALT. We observe that Russian vegetation has a positive median Δ200Hg of 0.08‰, which is directly reflected in the constant Δ200Hg of 0.04‰ in dHg of the Yenisei River, the largest Arctic river in terms of discharge and Hg input to the AO33,34. Mackenzie River dHg and pHg also carry median positive Δ200Hg of 0.06‰, and 0.04‰ respectively (Fig. 3 and Supplementary Fig. 4). Figure 3 illustrates that atmospheric Hg0 Δ200Hg at all three sites is significantly different from both Russian tundra vegetation and Yenisei and Mackenzie river run-off Δ200Hg (Wilcoxon two sample t test, p < 0.05). The seasonal Hg0 Δ200Hg flatlines of −0.08‰; at ZEP, VRS and ALT (Fig. 2c) therefore corroborate the back-trajectory, δ202Hg, and Δ199Hg observations and suggest that there is not a large Russian or N-American terrestrial Hg contribution (due to the positive Δ200Hg for Russian vegetation, Yenisei and Mackenzie River Hg), via AO Hg0 emission, to the summertime Hg0 maximum at these sites.

Interestingly, Alaskan tundra vegetation (Fig. 3, Vegetation, Toolik) has a lower median Δ200Hg of −0.03‰ than Eurasian vegetation (Lichen/Moss, Siberia) and runoff (River, Yenisei, Mackenzie). This is likely due to latitude bias in the data, with Alaskan tundra data from >67°N, and most of the Siberian tundra vegetation data and >90% of the Yenisei and Mackenzie River watersheds from <67°N. HgII wet deposition, with its elevated Δ200Hg of 0.14‰, is known to increase towards the mid-latitudes, explaining the relatively positive Δ200Hg of 0.04–0.06‰ in dHg of the Yenisei and Mackenzie Rivers. Surface AO pHg, which reflects internally and externally sorbed HgII on dead and living marine particulate matter, has a median Δ200Hg of 0.01‰ (IQR −0.02 to 0.05, Fig. 1, Supplementary Fig. 3), which potentially reflects multiple source contributions to HgII in marine particles: incoming Atlantic and Pacific waters, terrestrial Hg, and atmospheric Hg0 and HgII. We observe relatively little geographic surface AO pHg Δ200Hg variation across Fram Strait, the Barents Sea, and the central basin, except two samples (out of 32, Fig. 1, Supplementary Fig. 3) in the transpolar drift current, which merits further study in the future.

During AMDEs, Hg0 is thought to be efficiently and near-quantitatively oxidized into reactive HgII, deposited to snow, and then partially photoreduced and re-emitted as Hg0. Membrane-collected reactive HgII at ZEP and VRS during the springtime AMDE seasons have median Δ200Hg of −0.03‰ (IQR −0.07 to 0.02), and snow HgII at Ny-Ålesund and across the Arctic have median Δ200Hg of −0.11‰ (IQR −0.11 to −0.04) and −0.05‰ (IQR −0.08 to −0.01) respectively (Fig. 3). Hg in snowmelt at Utqiagvik also showed Δ200Hg of −0.08 and −0.09‰ (n = 223). Within the analytical uncertainty, the negative Hg0 Δ200Hg signatures are therefore conserved during the chain of AMDEs into snow HgII, and snowmelt HgII which has also been observed by others41,42. The Δ200Hg of −0.03‰ in reactive HgII forms sampled at ZEP and VRS are significantly higher than the simultaneously sampled Hg0 with Δ200Hg of −0.08‰ (p < 0.05). The absolute Δ200Hg difference between HgII and Hg0 is small however, and at the limit of atmospheric HgII isotope analytical uncertainty. In addition, while membranes collect more HgII than denuder-based methods in the Arctic67, there is a possibility that over a 1-week sampling period some gaseous or particulate HgII was lost from the membranes, leading to unpredictable minor bias in HgII Δ200Hg (but not Hg0 Δ200Hg).

Cause of the summertime Hg0 maximum

Air mass back-trajectory analysis (Figs. 5 and 6) shows that the summertime boundary layer air masses, with their distinct δ202Hg, Δ199Hg and Δ200Hg signatures typical of AMDE re-emission dynamics, have their origins over sea ice and snow-covered coastal land (62%) in the N-Greenland and Ellesmere Island region in June. In July, boundary layer air masses shift progressively from snow-covered sea ice and land (39%) to open waters (51%) of Baffin Bay and Fram Strait. In August, during the descent from the Hg0 peak, these proportions reach 25% for snow-covered ice and land and 53% for open waters. Boundary layer trajectories over land are 26% on average from June to August. At no time do boundary layer trajectories reach ice-free Siberian shelf waters where most terrestrial Hg is discharged. These findings therefore contradict previous suggestions that terrestrial Hg inputs from Russian rivers and coastal erosion to the AO, followed by 80% oceanic emissions to the atmosphere, make an important contribution to the summertime Hg0 maximum observed throughout the Arctic25,30,34. While there is broad overlap in the timing between estimated river discharge and Hg0 Δ199Hg maxima (Fig. 2b), Δ199Hg increases with the onset of AMDEs about 1 month before substantial river Hg discharge. In addition, the Siberian and Beaufort Sea shelves remain ice-covered until mid-June, 2 months after the first Δ199Hg increases, and during the steepest increase in atmospheric Hg0 (Supplementary Figs. 5 and 6). This suggests that while atmospheric Hg0 concentrations (and Hg0 Δ199Hg) and river Hg discharge partially co-vary, they are not causally related.

The dominant interaction of July and August air masses with open marine waters raises the question if marine Hg0 emissions contributing to the July Hg0 maximum represent recent AMDE Hg runoff from snowmelt over ice and coastal land (including glaciers) to marine waters or non-AMDE related marine Hg derived from background atmospheric deposition to the same regional marine waters throughout the year. Marine total Hg concentrations over the North East Greenland shelf and in Baffin Bay are indeed elevated in August, reaching up to 4 pM in surface waters due to meltwater inputs68,69. The August pHg isotope data over the North East Greenland shelf also support a meltwater Hg source to the atmosphere, because of its relatively elevated δ202Hg of −0.40‰ and low Δ199Hg of −0.20‰ (Fig. 1), which contrast with the terrestrial signatures observed in the transpolar drift current further north.

Previous Arctic Hg budgets of coastal erosion Hg flux are variable, 16–47 Mg y−1, depending on assumed, and highly variable, glacial sediment Hg concentrations. Here we propose a new, lower estimate of 9 Mg y−1 (Supplementary Table 2), taking benefit of improved erosional mass and carbon budgets and deep mineral soil and coastal glacial sediment Hg/C ratios. We estimate seasonal coastal erosion Hg fluxes (Fig. 2b, and Supplementary Data File) by scaling the annual flux of 9 Mg y−1 to monthly estimates of erosional C and N fluxes35. Coastal erosion Hg inputs are too small and arrive too late, peaking late August and September, to make a contribution to the summertime atmospheric Hg0 maximum. On the contrary, we document how river Hg inputs imprint their low δ202Hg, and positive Δ199Hg and Δ200Hg on surface AO pHg in August 2015 (Fig. 1). This suggests that terrestrial Hg from rivers and coastal erosion likely impacts AO marine ecosystems but not the atmosphere. We therefore conclude, based on air mass origins and Hg isotope signatures, that the summertime atmospheric Hg0 peak is most likely sustained by re-emission of Hg deposited during the previous spring season to the cryosphere. This re-emission takes place directly from the cryosphere but also from regional open marine waters that receive meltwater Hg inputs.

Recent studies have quantified a large permafrost soil Hg pool, approximately 72,000 Mg in the upper 30 cm36, which results in an important river and coastal erosion Hg flux to the AO. Ongoing Arctic warming has fueled concerns about enhanced mobilization of permafrost Hg to surface AO (0–200 m) ecosystems and to the global atmosphere, which contain relatively smaller amounts of 270 and 4000 Mg of Hg, respectively.6,69 Models incorporating this terrestrial Hg flux to the AO suggest its partial reduction in AO surface waters and emission to the atmosphere25,30,34,37, where it would then sustain the summertime Hg0 maximum. The model predictions of AO emission of terrestrial Hg inputs hinge on the key assumption that approximately 80% of the terrestrial Hg inputs are reduced with the remaining 20% depositing to shelf sediments. Our findings, based on the Hg isotope fingerprints of Arctic atmospheric Hg0, do not show evidence of a terrestrial origin for summertime Hg0. We speculate that terrestrial Hg is mostly deposited to AO shelf sediments, without dramatically evading to the global atmosphere. The large AO shelf area supports a rich and diverse ecosystem, and enhanced deposition of terrestrial Hg to shelf sediments may lead to enhanced microbial Hg methylation and increased MeHg exposure to the benthic and pelagic food webs and ultimately humans. More work is needed to assess the impact of an increased terrestrial Hg load on coastal Arctic ecosystems.

Methods

Study area

Atmospheric Hg isotope observations were made at Zeppelin observatory (ZEP, Svalbard), Villum Research Station Nord (VRS, Greenland), and Alert station (ALT, Canada) (Fig. 1). All three stations are situated far from major air pollution point sources. ZEP is located on the top of Mt. Zeppelin, Svalbard (78.90°N, 11.89°E, 474 meters above sea level (m a.s.l.)), just outside the community of Ny-Ålesund. A steep downhill slope faces north towards the research village of Ny-Ålesund, a small settlement with 35 to 185 inhabitants at 2 km from the sampling site. VRS is located in the north-eastern corner of Greenland on the north-south oriented peninsula Princesse Ingeborgs, which is a 20 × 15 km Arctic lowland plain. Measurements were performed at the Air Observatory at Villum (81.6°N, 16.67°W, 24 m a.s.l.), which is located on the Danish military base Station Nord. The atmospheric observatory at Villum is located 2 km to the south of Station Nord and is upwind >95% of the time from local pollution sources at the military base. The ALT location (82.5°N, 62.3°W, 200 m a.s.l.) is at the Dr. Neil Trivett Global Atmosphere Watch Observatory at Alert, Nunavut, Canada and is located about 8 km south of the Lincoln Sea.

Atmospheric Hg0 sampling and processing

Activated carbon traps impregnated with sulfur (HGR-AC, Calgon Carbon Corp., Pittsburg, PA, USA) were used to collect atmospheric Hg0 isotopes at ZEP, VRS and ALT stations. Weekly and limited bi-weekly sampling was conducted from March, 2018 to June, 2019 (ZEP, VRS) and from May, 2018 to March, 2019 (ALT). The sampling train consisted of a 47 mm, 0.45 µm porosity polyethersulfone (PES) cation exchange membrane (Merck Millipore) in a 47 mm Savillex PFA Teflon filter holder, connected by 6 mm FEP Teflon tubing to HGR-AC traps (200 mg activated carbon powder, in 10 cm long, 4 mm internal, 7 mm external diameter Pyrex glass tubes), a ball flow meter (Fisher scientific), a digital volume meter (Siargo Ltd.), and a membrane vacuum pump (KNF). The PES membranes collect gaseous and aerosol HgII species, and the HGR-AC traps collect Hg0. At ALT, a calibrated MKS mass flow controller and Gast carbon vane pump were used. The sampling flow of 0.5–1.0 L min−1 was regularly checked with a calibration unit (Bios Defender) and considered stable throughout the campaigns. After weekly field sampling, HGR-AC traps were sealed with silicone stoppers, packed in a double-zipper bag, and stored frozen on site. Samples were transported frozen to Toulouse, France, at the end of campaigns for Hg isotope analysis. HGR-AC traps were combusted in a double-stage tube furnace in pure O2 at 80 mL min−1, and purged and pre-concentrated into 8 mL of 40% v/v HNO3/HCl in a 2:1 ratio (Sun et al.70). Purging impingers were rinsed with a total volume of 8 mL of MQ water, diluting the sample to 20% v/v acid. The final trapping solutions were kept in a refrigerator at 2–4 °C until Hg isotope analysis.

During the campaigns, automated Tekran® 2537 instruments (Tekran® Inc., Canada) measured Hg0 continuously, at 5 to 15 min resolution, at all three stations. Tekran® 2537 models pump and pre-concentrate ambient Hg0 over gold traps, and then Hg0 is thermally desorbed and detected by cold vapor atomic fluorescence spectrometry (CVAFS). The air inlets for all automated and manual air Hg sampling were installed 3–5 m above ground in close proximity to one another, facing downward and toward the predominant wind directions. Identical operating procedures were used at the sites and strict quality control criteria are documented elsewhere (Berg et al.51; Boyd Pernov et al.71; Steffen et al.20). HGR-AC trap Hg0 recoveries were estimated from Tekran® 2537 and trap solution Hg concentrations and were 90 ± 21% at ZEP and VRS (mean ± 2 SD).

Atmospheric reactive HgII sampling and processing

Atmospheric HgII was collected weekly during spring in 2018 and 2019 at ZEP and VRS, when HgII levels during AMDE events are high enough for Hg isotope analysis. The sampling train, validated elsewhere (Fu et al.64), consisted of a 90 mm, 0.45 µm porosity polyethersulfone (PES) cation exchange membrane (Merck Millipore) in a 90 mm Savillex PFA Teflon filter holder, connected by 6 mm FEP Teflon tubing to a ball flow meter set at 4.0 L min−1 (Fisher scientific), a digital volume meter (Siargo Ltd.), and a membrane vacuum pump (KNF). PES membranes were sealed in petri-dishes, packed in double-zipper bags, stored in a freezer (−20 °C) on site, and transported frozen to France for analysis. In the laboratory, PES membranes were leached in 16 mL of 2.5% v/v HNO3/HCl in a 2:1 ratio, in acid-cleaned 50 mm diameter PFA Teflon beakers (Savillex) on a hot plate (Analab) at 120 °C. The HgII concentration in the PES leachates was determined using a Brooks Rand Model III CV-AFS with a custom-made purge and trap system. The method detection limit (MDL) was 5 pg Hg72. When Hg concentrations were high enough, leachate solutions were combined and submitted to a purge and trap pre-concentration: leachates were diluted with MQ water to 0.5 L volume in a 1.0 L pre-cleaned Pyrex bottle with GL45 Savillex ¼” two-port cap. SnCl2 (2.5 mL of 3 wt% SnCl2 in 1 N HCl) was added to the bottle and purged with Hg-free argon for 6 h at 400 mL min−1 into a 8 mL, 40% v/v HNO3/HCl (2:1 ratio) trap42. Final trap solutions were diluted with MQ water to 20% v/v acid and kept refrigerated at 2–4 °C until Hg isotope analysis.

Snow, river water and marine particles sampling

Snow samples were collected in 2011 and 2019 from 40 cm deep pits dug outside of Ny-Ålesund close to the base of the Zeppelin mountain. All the samples were kept at −20 °C in the dark onsite, and transported to France frozen, where Hg-free BrCl was added upon thawing to convert all Hg species to labile HgII forms.

The Yenisei River was sampled at Igarka (67.4°N, 86.4°E), 300 km from the river mouth, using a boat or via holes drilled in the ice cover. Hg samples were filtered in the field using pre-burnt 47 mm, 2.0 µm porosity quartz filters (QMA Millipore) and a 47 mm Savillex PFA Teflon filter holder into acid-cleaned 500 mL FEP Teflon bottles, acidified to 0.36 M with bi-distilled HCl, stored at 4 °C in the dark until transport to France. Sample aliquots were analyzed by CVAFS and dissolved Hg (dHg) concentrations published elsewhere34. Remaining samples were stored cold at 4 °C until BrCl addition and pre-concentration by the same purge and trap procedure mentioned above for HgII leachates and adapting purge and trap bottle volume where needed (1, 5, 20 L). Pre-concentration recoveries were found to be 85 ± 20% (2 SD) for the rivers and snow samples.

Marine particles were sampled on QMA filters with in situ pumps (McLane) in the Barents Sea and the central Arctic Ocean were sampled during the GEOTRACES TransArc II (GN04) cruise (17th August to 15th October 2015) and in Fram Strait during the 2016 GEOTRACES GRIFF (GN05) cruise (18th July to 6th September 2016) aboard the FS Polarstern. Sampling details and concentration of particulate Hg (pHg) are given in refs. 69, 73. QMA filters were combusted in a double-stage tube furnace in pure O2 at 80 mL min−1, and purged and pre-concentrated into 8 mL of 40% v/v HNO3/HCl in a 2:1 ratio70.

Hg isotope analysis

Hg isotope ratios were measured on a Neptune Plus multi-collector inductively coupled plasma mass spectrometer (MC-ICPMS, Thermo-Finnigan) at the Observatoire Midi-Pyrénées, Toulouse, France, and at the University of Toronto, Canada, following the methods described in previous studies74,75. In Toulouse, we used a CETAC ASX-520 autosampler and HGX-200 CV system coupled with the MC-ICPMS, equipped with a 1012 Ω amplifier on the 198Hg isotope in order to improve isotope ratio precision in the 50 mV range. Sample and standard signals at 0.3 to 1.0 ng g−1 Hg concentrations were generally 150–500 mV on the 202Hg isotope, at a sample introduction flow rate of 0.5 mL min−1. Thallium was not used as an internal standard, and the 203 and 205 masses were occasionally monitored to survey Hg-hydride interferences (i.e., 202Hg1H, and 202Hg1H1H), which were found to be negligible when using standard H-cones. In Toronto, Hg was introduced to the MC-ICPMS by reducing HgII in liquid samples with 3% SnCl2 to Hg0 vapor, which was separated using a gas-liquid separator. Instrumental mass bias of MC-ICPMS was corrected by standard-sample-standard bracketing using NIST3133 Hg at matching concentrations. In addition to sample-standard bracketing, mass bias at University of Toronto was also corrected using Tl as an internal standard (NIST SRM 997) that was introduced using an Aridus II desolvating nebulizer. Hg isotopic composition is reported in delta notation (δ) in units of per mil (‰) referenced to the bracketed NIST 3133 Hg standard76:

$${\delta }^{{xxx}}{Hg}({‰})\,=\,\left[\frac{{\left(\frac{{\,\!}^{xxx}{Hg}}{{\,\!}^{198}{Hg}}\right)}_{{sample}}}{{\left(\frac{{\,\!}^{xxx}{Hg}}{{\,\!}^{198}{Hg}}\right)}_{{NIST}3133}}-1\right]\times {10}^{3}$$
(1)

where xxx represent Hg isotopes 199, 200, 201, 202, 204. Mass-independent isotope fractionation (MIF) is expressed in “capital delta (Δ)” notation (‰), which is the difference between the measured values of δ199Hg, δ200Hg, δ201Hg, and δ204Hg and those predicted from δ202Hg using the kinetic mass-dependent fractionation law:

$${\triangle }^{{xxx}}{Hg}\left({‰}\right)\,=\,{\delta }^{{xxx}}{Hg}\, - \,({\beta }^{{xxx}}\,\times \, {\delta }^{202}{Hg})$$
(2)

where βxxx is 0.2520, 0.5024, 0.7520, 1.493 for the 199, 200, 201, and 204 Hg isotopes respectively. Analytical uncertainty of isotopic analysis was assessed by repeated measurement of the samples, of in-house standards UM-Almaden, ETH Fluka and JT Baker, and of procedural standards NIST SRM 1632d and 1632e (see Supplementary data file). The results obtained were in agreement with published values61,62,77,78. The 2 SD uncertainty reported for samples is the largest of the 2 SD’s of sample replicates, procedural standards, or in-house standards.

Back-trajectory analysis

The Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT, v. 4.2.0; https://www.arl.noaa.gov/hysplit/) developed by NOAA79. was driven with 3 hourly meteorological input data from the Global Data Analysis System (GDAS; 1° latitude-longitude 360 by 181 grid) to identify the potential Hg source regions. The model was run in backward mode for 240 h every 2 h throughout the sampling periods at ZEP, VRS and ALT. In total, ~84 backward trajectories were calculated for each sampling period. An analysis of the spatial (horizontal and vertical) residence time of the air mass history has been accomplished. The back-trajectory model results have been analyzed with respect to the air residence time above five surface types (land without snow cover, open water, permanent ice/snow, sea ice, and land based snow). The surface type maps “land without snow cover”, “open water” and “permanent ice/snow” were extracted from Climate Change Initiative (CCI) land cover data, the “sea ice” coverage from the AMSR-2 sea ice concentration product (sea ice coverage >50% sea ice concentration) and the “land based snow” from MODIS/Terra and MODIS/Aqua data sets. Data from MYD10C1 (Aqua) were used to fill gaps in MYD10C1 (Terra) (see Supplementary Table 1 for details). We used ESRI ArcGIS Pro (v. 10.6) to provide the percent surface exposure of particles along trajectories for each surface type. In addition, the percent of particles along trajectories that reside within the boundary layer (BL) and in the free troposphere (FT) has been calculated.