Volcanic activity and gas emissions along the South Sandwich Arc

The South Sandwich Volcanic Arc is one of the most remote and enigmatic arcs on Earth. Sporadic observations from rare cloud-free satellite images—and even rarer in situ reports—provide glimpses into a dynamic arc system characterised by persistent gas emissions and frequent eruptive activity. Our understanding of the state of volcanic activity along this arc is incomplete compared to arcs globally. To fill this gap, we present here detailed geological and volcanological observations made during an expedition to the South Sandwich Islands in January 2020. We report the first in situ measurements of gas chemistry, emission rate and carbon isotope composition from along the arc. We show that Mt. Michael on Saunders Island is a persistent source of gas emissions, releasing 145 ± 59 t day−1 SO2 in a plume characterised by a CO2/SO2 molar ratio of 1.8 ± 0.2. Combining this CO2/SO2 ratio with our independent SO2 emission rate measured near simultaneously, we derive a CO2 flux of 179 ± 76 t day−1. Outgassing from low temperature (90–100 °C) fumaroles is pervasive at the active centres of Candlemas and Bellingshausen, with measured gas compositions indicative of interaction between magmatic fluids and hydrothermal systems. Carbon isotope measurements of dilute plume and fumarole gases from along the arc indicate a magmatic δ13C of − 4.5 ± 2.0‰. Interpreted most simply, this result suggests a carbon source dominated by mantle-derived carbon. However, based on a carbon mass balance from sediment core ODP 701, we show that mixing between depleted upper mantle and a subduction component composed of sediment and altered crust is also permissible. We conclude that, although remote, the South Sandwich Volcanic Arc is an ideal tectonic setting in which to explore geochemical processes in a young, developing arc.


Introduction
Arc volcanoes dominate the global volcanic emissions budget, collectively releasing tens of teragrams of volatiles into the atmosphere each year that have regional and global impacts on air quality and climate . For many arc systems, these fluxes are well studied by surface measurements and the subsurface processes controlling them are at least constrained, if not yet fully understood. However, the global inventory of volcanic emissions remains incomplete because we lack data from a few remote, but very active, arcs ). Many of these remote settings are also vulnerable ecological sites where the habitat, although often free from human interference, is impacted frequently by volcanic activity (Reynolds 1996;Roberts et al. 2017).
The volcanoes of the South Sandwich Volcanic Arc (SSVA) are a significant source of tephra to the South Atlantic (Ninkovich et al. 1964) and the Antarctic mainland (Smellie 1999;Basile et al. 2001) and yet their remote location in the Southern Ocean has meant that their eruptive processes Editorial responsibility: A. Cannata remain enigmatic compared to other arcs globally. The relatively recent development of the SSVA over the past 10 million years has been closely linked with the formation of the Drake Passage (Barker 1995;Vanneste and Larter 2002;Leat et al. 2013Leat et al. , 2016, making this one of the youngest known volcanic arcs on Earth and therefore one of the most critical for understanding the early stages of arc geochemical evolution. Records of historical activity are relatively limited, with in situ observations contributed by only a few expeditions (e.g. Holdgate and Baker 1979;Smellie et al. 1998;Derrien et al. 2019). Much of the monitoring data collected over the past two decades derive from satellite remote sensing observations (Lachlan-Cope et al. 2001;Patrick et al. 2005;Patrick and Smellie 2013;Gray et al. 2019), and yet even these are restricted to rare, cloud-free images. Satellite observations reveal persistent passive outgassing at many of the SSVA volcanoes, punctuated by sporadic explosive emissions, suggesting that the arc could be a significant source of volatiles to our atmosphere. Petrological constraints on magmatic volatile contents are lacking, as are data on the composition and flux of emitted gases. Here, we present geological and volcanological observations made during an expedition to the South Sandwich Islands in January 2020. We report measurements of gas chemistry, emission rate and carbon isotope signature from along the SSVA. Together, these observations contribute to a comprehensive view of the 'state of the arc', providing a foundation for future targeted studies.

Tectonic and ecological significance
The South Sandwich Volcanic Arc (SSVA) is situated on the leading edge of the South Sandwich microplate, bounded to the west by the East Scotia Ridge (ESR) back-arc spreading centre and to the east by the South Sandwich Trench (Fig. 1;Barker 1970;Larter et al. 2003). Seafloor magnetic anomalies record continuous W-E spreading at the ESR since at least~15 Ma, initially at a full spreading rate of~27 mm year −1 until~1.7 Ma when spreading accelerated to the current rate of 63-71 mm year −1 (Barker 1995;Livermore et al. 1997;Larter et al. 2003; Thomas et al. 2003). The SSVA is built largely on the oceanic crust that formed at the ESR~10 Ma (Barker 1995;Larter et al. 2003) and is consequently one of the youngest volcanic arcs active today. The crust beneath the SSVA is relatively thin at 16-20 km compared to arcs globally (~30 km on average), consistent with its young age and extensional regime due to back-arc spreading (e.g. Larter et al. 2003;Calvert 2011).
The southern end of the SSVA is bounded by the South Sandwich fracture zone, an active transform fault accommodating dextral motion between the Antarctic and South Sandwich plates (Barker 1995;Thomas et al. 2003;Giner-Robles et al. 2009). At the northern end of the arc, focal mechanisms associated with enhanced seismicity suggest that the underthrust plate is being torn at depth to accommodate subduction (Brett 1977;Giner-Robles et al. 2009). Upper mantle seismic anisotropy from shear wave splitting reveals a strain field that describes westward mantle flow into the back-arc region from the north and south (Livermore et al. 1997;Müller 2001;Müller et al. 2008), consistent with the enriched Atlantic mantle signatures observed in erupted basalts from the northern-and southern-most segments of the ESR (Leat et al. 2004).
Magmatism along the SSVA is the expression of westward oceanic-oceanic subduction of the South American Plate beneath the South Sandwich microplate. Plate convergence is accommodated by subduction rates of 69 ± 11 mm year −1 (north) to 71 ± 10.8 mm year −1 (south), associated with an eastward rollback of the trench (Thomas et al. 2003;Smalley et al. 2007). The age of subducted crust entering the trench from the South American plate varies from~83 Ma in the north tõ 27 Ma in the south (Barker 1995). Almost all of the sediment entering the trench is subducted, and a welldeveloped accretionary prism has not been established (Vanneste and Larter 2002). A sedimentary section from site 701 (Ocean Drilling Programme, ODP 701; Fig. 1) shows that the pelagic sediment north of the arc is carbon-poor overall (Ciesielski et al. 1988). From top to bottom, the pelagic sequence from ODP 701 exposes three main units:~245 m of diatomaceous ooze, with intervals of siliceous clay-rich muds and volcanic ash;~200 m of diatomrich clay/mud, with lesser volcanic ash;~30 m of indurated nannofossil chalk-of mid to late Eocene age-overlying amygdaloidal olivine basalt basement (Ciesielski et al. 1988;Vervoort et al. 2011). Sediment thicknesses are thinner immediately east of the arc (~200 m on average; Plank and Langmuir 1998). Dredge samples of oceanic crust from the South Sandwich Trench (Pearce et al. 2000) and the American-Antarctic Ridge to the east (Lawver and Dick 1983) are dominated by exposed ultramafic materialserpentinised peridotites, harzburgites, dunites and gabbros-typical of slow-spreading tectonic settings.
Erupted magma compositions range from basaltic to silicic and populate both tholeiitic and calk-alkaline trends; tholeiitic basalts are by far the most abundant (Baker 1978;Pearce et al. 1995). Overall, the SSVA is considered a type example of early subduction zone magmatism (Baker 1978). Trace element systematics suggest a variable, element-specific subduction component from altered oceanic crust (Pearce et al. 1995). Boron isotope compositions (δ 11 B) for South Sandwich lavas are among the highest reported for mantlederived lavas, suggesting substantial subduction erosion of serpentinised forearc material (Tonarini et al. 2011). Rates of terrigenous erosion from the South Sandwich volcanoes and forearc are high, supplying large amounts of volcaniclastic material eastwards towards the trench (Leat et al. 2010;Tonarini et al. 2011). Overall, the carbon-poor nature of the sediments entering the trench lead to low predicted CO 2 /SO 2 ratios~1.2 in the corresponding arc emissions based on trace element proxies .
The South Sandwich Islands (SSI) represent the emergent portions of deep-rooted volcanic edifices that comprise the arc front and extend to depths of 2000-3000 m below the seafloor (Leat et al. 2010(Leat et al. , 2013(Leat et al. , 2016. From north to south along the arc, the volcanic centres include Protector Seamounts (submerged), Zavodovski, Visokoi, Candlemas group (Candlemas and Vindication), Saunders, Montagu, Bristol, Southern Thule group (Bellingshausen, Cook and Thule) and Kemp and Adventure calderas (submerged). Since 1900, expressions of active volcanism (ranging from sporadic fumaroles to explosive eruptions) have been reported from all but Vindication, which represents the eroded remnants of an ancient stratovolcano. Volcanic edifices are constructed from layers of pyroclastic material and lavas (Baker 1978;Holdgate and Baker 1979;Smellie et al. 1998;Leat et al. 2003). Bathymetric surveys reveal numerous mass flow deposits associated with the submarine flanks of the SSI volcanoes (Leat et al. 2010(Leat et al. , 2013. Active seafloor hydrothermal venting, important for sustaining chemosynthetic communities, has been reported from Kemp and Adventure calderas in the southern section of the arc and from the Protector Seamount region in the north (Boschen et al. 2013;Linse et al. 2019). The SSI are located south of the Antarctic Circumpolar Current and the topographic profile of the arc strongly influences the path of oceanographic currents (Garabato et al. 2002).
The South Sandwich Islands have a polar climate, with heavy snowfall in winter and frequent strong winds and storm swells year-round. Terrestrial flora is scarce, dominated by mosses and lichens (Convey et al. 2000;Hart and Convey 2018). However, the islands are breeding grounds for globally significant populations of seabirds, notably penguins (Hart and Convey 2018). At the most recent census in 2011, the SSI contain almost half of the world's population of Chinstrap penguins (Lynch et al. 2016); 600,000 pairs are reported from Zavodovski Island alone, together with a further 95,000 pairs of Macaroni penguins, and > 125,000 pairs of Adelie penguins. The islands are administered as a UK Overseas Territory by the Government of South Georgia and the South Sandwich Islands. In response to calls for long-term management of the fishery and dependent wildlife, South Georgia and SSI were declared a Marine Protected Area in 2012, which was further expanded in 2019 (Marine Protected Areas Order 2013, 2019. Recent volcanic eruptions in the SSI are likely to have had significant and ongoing impacts on local terrestrial and marine ecosystems, through immediate disturbance, deposition of pyroclastic material (mostly as volcanic ash) and from persistent passive degassing. Historic volcanic eruptions elsewhere in the Antarctic have caused mass mortality in penguin colonies (Roberts et al. 2017).

Methods
We report on the state of current volcanic activity along the South Sandwich Volcanic Arc, with an emphasis on active outgassing. These observations were made during a 5-week   (Ciesielski et al. 1988). b Magnified view of the SSVA showing subaerial islands. Topographic data extracted from WorldDEM™ (Airbus Space and Defence) expedition to South Georgia and the South Sandwich Islands in January 2020 aboard the Pelagic Australis. We sailed from Stanley, Falkland Islands, on 30 December 2019 and returned to the harbour on 30 January 2020. We made shore landings by zodiac on Saunders Island on three consecutive days between 6 and 8 January; Thule Island on 10 January; Bellingshausen on 11 January and Candlemas on 12 and 13 January. Aerial images of Zavodovski, Candlemas, Saunders, Bristol and Bellingshausen were acquired using a small Unoccupied Aerial System (UAS; DJI Mavic 2 Pro) and used to aid geological interpretation. Aerial gas measurements (see instrument specification below; Liu et al. 2019;Pering et al. 2020) were acquired using a multi-rotor UAS (DJI Matrice 200 modified with custom mounting platform). Full permissions for beyond visual line of sight UAS flights were obtained from the Government of South Georgia and the South Sandwich Islands under Air Navigation (Overseas Territories) Order 2013.
A summary of all sampling sites-including coordinates, distance from vent and access descriptions-are presented in Table S1 (supplementary information). All altitudes reported in the text are extracted from the WorldDEM™ topographic model, which has a relative vertical accuracy of 2 m (4 m absolute) in a 12 m × 12 m raster (© DLR e.V. 2019 and © Airbus Defence and Space GmbH 2019). Images and videos were taken both on-and offshore and used for geological interpretation. A landing on Zavodovski was unfortunately not possible due to inclement weather and ocean conditions, instead, we report large-scale stratigraphic observations made from offshore and at outcrop-scale using a UAS launched from the boat (see Video, S1-S2, supplementary information). On Candlemas, we landed ashore by zodiac in Kraken Cove (in the north of the island; Fig. 2) before hiking to the peninsula of Lucifer Hills (~1 h). Travel was slow, but not technical, over a'a lava flows. We acquired gas samples and measurements at the summit of Lucifer Hills (see methods below) and made outcrop-scale observations of the lavas as we traversed the peninsula, which were aided by aerial imagery acquired using a UAS launched from offshore. On Saunders, we landed ashore by zodiac at the southern end of Cordelia Bay (eastern side of the island; Fig. 2). We walked to the centre of the bay before ascending onto the glacier on the eastern flank of Mt. Michael, contouring around to reach sampling sites on the northeastern and southern sides of the volcano (2-3 h one way). Access involved technical glacier travel across heavily crevassed terrain. We made in situ observations of recent pyroclastic deposits exposed on and within the glacier, together with stratigraphic observations of the layered hydromagmatic deposits exposed on the eastern flank of Ashen Hills (Fig. 2). Large-scale observations of the topography and the plume were aided by a UAS launched from Natriss Point (Video S3, supplementary information). On Bellingshausen, access involved a challenging shore landing near Isaacson Point (southern side of the island; Fig. 2). There are few sheltered landing sites for a zodiac due to steeply eroded cliffs and none during a heavy swell. We swam ashore from a zodiac positioned several tens of metres offshore, wearing a full dry suit and buoyancy aid, and then hauled equipment ashore in sealed dry bags attached to a rope line. From the coast, the hike to the crater rim (30 min one way) was straightforward over relatively smooth terrain (eroded and vegetated lava flows). From the southern rim of the crater, we descended a steep but not technical scree slope to the base of the crater where several fumaroles are located. We made in situ observations of the large-scale stratigraphy of the crater from the crater rim, supplemented by aerial imagery from a UAS launched from the boat (Video S4, supplementary information). We made outcrop-scale observations of layered pyroclastic deposits and massive lava flows at Hardy Point. At Thule, we landed ashore by zodiac at the ice-free Hewison Point (southeastern tip of the island; Fig. 2), acquiring outcrop-scale observations of the exposed lava flows and layered pyroclastic deposits.
Fumarole temperatures were measured using a K-type thermocouple that was calibrated immediately prior to the field campaign. Gas compositions were measured using two versions of a multi-component gas analyser system (Multi-GAS; Aiuppa et al. 2005;Shinohara 2005): ground-based and dronemounted. The plume was sampled through two 1-μm particle filters exposed to ambient air, at a pump rate of 1.0 L/min. Concentrations of CO 2 , SO 2 and H 2 S were logged at 1 Hz. For the ground-based Multi-GAS, SO 2 , H 2 S and H 2 electrochemical sensors (3ST/F-TD2G-1A, T3H-TC4E-1A and T3HYT-TE1G-1A, City Technology) were calibrated for 0-200, 0-100 and 0-200 ppmv, respectively, each with an accuracy of ± 2% and a resolution of 0.1 ppmv. A nondispersive infrared (NDIR) spectrometer (GasCard NG, Edinburgh Instruments) was calibrated for 0-3000 ppmv CO 2 with an accuracy of ± 2% and a resolution of 0.8 ppmv. Pressure (± 1 hPa), temperature (± 0.5°C) and relative humidity (± 2%) were also measured at 1 Hz (Galltec sensors). For the aerial Multi-GAS, SO 2 and H 2 S electrochemical sensors (3ST/F-TD2G-1A and T3H-TC4E-1A, City Technology) were calibrated for 0-200 and 0-50 ppmv, respectively, each with an accuracy of ± 2% and a resolution of 0.1 ppmv. A NDIR spectrometer (Smartgas Modul Premium2, Microsensorik) was calibrated for 0-5000 ppmv CO 2 with an accuracy of ± 2% and a resolution of 1 ppmv. The unit was shielded from radio frequency interference from the UAS transmission system using a foil wrapping. Pressure (± 1 hPa), temperature (± 0.5°C) and relative humidity (± 3%) were also measured at 1 Hz using a combined sensor module (Bluedot BME280) exposed to ambient air. Both Multi-GAS instruments were calibrated with standard reference gases at the Università di Palermo (Palermo, Italy) immediately prior to the field campaign. All sensor data was logged to a micro-SD card. H 2 O concentrations were calculated from records of temperature, relative humidity and timeaveraged ambient pressure (according to the Arden Buck equations relating the pressure of vapour saturation to temperature for moist air; Buck 1981). Time series of gas concentrations were post-processed using Ratiocalc software (Tamburello 2015). Volcanogenic CO 2 and H 2 O were resolved from the atmospheric background by subtracting the CO 2 and H 2 O in ambient air (measured outside the plume where SO 2 = 0) from the raw CO 2 time series. H 2 S concentrations were corrected for the 13% cross-sensitivity of the sensor to SO 2 (determined during calibration with standard reference gases). For example, although the absolute detection limit of the H 2 S sensor is 0.1 ppm, the effective detection limit is 0.13 × SO 2 ppm. Molar gas ratios are derived from the gradient of the linear regression in two-species scatterplots, with the exception of H 2 O. Uncertainties are reported as the 95% confidence intervals on the regression. The relative humidity sensor reached saturation very quickly during fumarole measurements at Bellingshausen and Candlemas, and therefore, derived H 2 O concentrations are of lower quality. H 2 O/X ratios, where X is the gas species used as the plume marker, were estimated assuming a simple step change in Due to sensor saturation, derived H 2 O/X ratios underestimate the true ratio and will be highly uncertain; we estimate a 50% uncertainty on these values to be conservative, although it is difficult to quantify the error explicitly. The emission rate, or flux, of SO 2 was determined spectroscopically using a dual UV imaging camera developed at Università di Palermo (Palermo, Italy) and described in Delle Donne et al. (2017,2019). The instrument, powered by a lithium-polymer battery, was equipped with two JAI CM-140GE-UV cameras sensible to UV radiation and fitted with two distinct bandpass optical filters (both of 10 nm full width at half-maximum) with central wavelengths of 310 (strong SO 2 absorption) and 330 nm (weak SO 2 absorption). A separate computer was used to command and control operations and to save the acquired images. We deployed the UV camera on Saunders on 8 January 2020, operating (at 0.5 Hz rate) from 13:50 to 16:40 UTC. During this time interval, the plume was dispersing to the east with the prevailing winds. The plume was grounding down the eastern flank, before lifting off and being transported at an altitude of several hundred metres above sea level.
The acquired images (520 × 676 pixels at 10-bit resolution) were post-processed using standard techniques (Kern et al. 2015;Delle Donne et al. 2017, 2019. Sets of co-acquired images were first combined to obtain sequences of "absorbance" images and then converted into slant column densities using calibrations derived from SO 2 reference cells (uncertainty, ± 10%). Finally, a time series of integrated column densities were obtained for each dataset by integrating the sequences of slant column densities images along a crosssection perpendicular to plume transport. Multiplication of the integrated column density by the plume speed yielded the SO 2 flux time series. Plume speeds (uncertainty, ± 5%) were derived using the optical flow algorithm of Lucas and Kanade (1981) to track the motion of plume gas fronts in image sequences (Delle Donne et al. 2017, 2019. Light dilution was corrected for using the method of Campion et al. (2015) and found to be negligible. In-plume, multiple scattering effects associated with the highly condensed plume are significant and mean that our derived slant column densities almost certainly underestimate the true column density; this is not modelled explicitly, however.
Dilute gas samples for carbon isotope measurements were collected in foil bags using a pumped inlet system operating for 90 s. Two bag samples were collected in the downwind plume from Mt. Michael, Saunders, and a further two from strong fumarolic emissions within the crater on Bellingshausen. A clean ocean air sample was collected several miles offshore from the Falkland Islands, with the boat engine turned off. Gas samples were analysed for their CO 2 concentration and carbon isotope composition (expressed as δ 13 C [‰]) by Delta Ray Infrared Isotope Ratio Spectrometer (Fischer and Lopez 2016;Ilanko et al. 2019) at the University of New Mexico (Albuquerque, USA). One sample was collected at a temperature of 96°C from a vent on Lucifer Hills, Candlemas, using a preevacuated Giggenbach bottle. The bottles did not contain NaOH solution due to the challenges of transporting caustic solution to these pristine islands. The bottle was connected via silicone tubing to a titanium tube and inserted into the vent. The sample was analysed in the Volatiles Lab at the University of New Mexico utilizing a combination of gas chromatography with a discharge ionization detector and quadrupole mass spectrometry following the techniques described in Ilanko et al. (2019).

Protector Seamounts
Protector Seamounts represent the northernmost volcanic region of the SSVA. A submarine eruption of rhyolite lava took place at Protector Shoal-a shallow cone within 55 m of the sea surface-in 1962 (Gass et al. 1963). The eruption produced an extensive pumice raft that was encountered by HMS Protector in the vicinity of Zavodovski island soon after (Gass et al. 1963) and as far as New Zealand over the following years (Coombs and Landis 1966). No observations were made during this expedition.

Mt. Curry, Zavodovski
Zavodovski hosts a single volcanic edifice, Mt. Curry (also known as Mt. Asphyxia; 557 m asl), surrounded by an extensive lava platform to the east (Fig. 2). Volcanic activity is focused at a steeply inclined crater on the southwest flank of Mt. Curry (Patrick and Smellie 2013). Vigorous vapour emissions were first reported in 1819 and similar observations have been reported since, notably in 1830 when emplacement of a fresh "pumiceous" lava flow was also documented on the eastern flank (Holdgate and Baker 1979). Most recently, ash-rich plumes were observed emanating from the main crater between March and June 2016, covering the eastern flanks in fresh dark ash deposits; large bombs were reportedly ejected from the crater (GVP 2017a).
The broad, low-lying lava platform on the eastern side of the island has undulating topography with deeply incised water gulleys. Two pyroclastic cones are visible on the lower flanks of Mt. Curry, one of which appears to be constructed from two coalesced explosion craters (Fig. 3d). Columnar jointing is exposed at the base of the eastern margins of the lava platform (Fig. 3f). This quasi-hexagonal fracture geometry forms due to thermal contraction of lava driven by conductive cooling, often associated with water infiltration into fractures (e.g. Spry 1962; Phillips et al. 2013;Lamur et al. 2018;Li and Liu 2020). Here, straight and uniform columnar fractures-colonnade jointing-are present in the dense lower part of the lava flow, suggesting enhanced cooling gradients as the flow entered the sea. The lack of jointing in the overlying rubbly flow unit indicates that the upper part of the flow was insulated to some degree. A thin layer of pyroclastic deposits appears to mantle the top surface of the lava flows in many of the eroded cross-sections.
Intense outgassing was visible on the western side of the island, originating from the main crater ( Fig. 3a)-located on the inclined western flank of Mt. Curry-and also from fumaroles within the steeply eroded cliffs below. The base of the crater was obscured by dense, condensed emissions (Video S1, supplementary information), presumably rich in water vapour. The rim of the crater is precipitously thin, with vertical inside walls. The lower left rim of the crater is heavily fractured and encrusted in yellow sulfurous deposits from the pervasive fumaroles (Fig. 3a). Dispersed fumarolic emissions are visible in the exposed cliffs and are particularly prevalent at the base of the cliff directly beneath the main crater.
The local stratigraphy is well exposed in the cliffs between Acrid Point and Noxious Bluff, passing Stench Point ( Fig. 4; Video S2, supplementary information). The rocks are vibrant shades of red through yellow, indicative of intense oxidative and hydrothermal weathering. At the base of the sequence, a series of relatively thin lava flows are interbedded with thin pyroclastic units. Thick lava flow units with minor pyroclastic intervals characterise the middle part of the stratigraphy. Thick pyroclastic units cap the upper part of the sequence, with a sharp contact with the underlying flow units. Pyroclastic deposits are horizontally layered and mantle the topography of underlying units. The stratigraphic transition from basal lava flows with minor pyroclastic intervals to dominantly pyroclastic emission suggests a general evolution from effusive cone-building activity to more explosive activity through time. A layered pyroclastic deposit mantles the topography, suggesting deposition as ashfall and is particularly well preserved on Stench Point (Fig. 4c, f). Vertical dyke-like structures are visible in the stratigraphy, standing proud by resistance to erosion; however, it is unclear where they originate or terminate (Fig. 4b, e). The ubiquitous dense, angular bombs scattered over Stench Point (Video S2, supplementary information) represent ballistic projectiles emitted from the main crater during periods of enhanced explosive activity, most likely in 2016.

Lucifer Hills, Candlemas
Candlemas is dominated by the imposing forms of two steeply eroded stratovolcanoes, Mt. Perseus and Mt. Andromeda (557 m asl; Fig. 2), both extensively glaciated. Steep cliffs beneath the ice cap expose a thick stratigraphy of alternating lava flows and pyroclastic units (Fig. 5b). Historically, active volcanism is been focused at the Lucifer Hills Complex, a series of coalesced cinder cones in the northeast region of the island (140 m asl; Fig. 5d-e). Here, andesitic-dacitic extrusive activity has built a radial promontory composed of multiple overlapping lava The peninsula is composed of silicic lava flows (andesite to dacite ;Baker 1978;Pearce et al. 1995) that have built an extension to the main island landmass. The lava flows are arranged in overlapping lobes that protrude out into the sea. The flows originate from small scoria cones within the main Lucifer Hills complex and can generally be traced back to breaches of the cone wall (Fig. 5f). The flows are generally thick and brecciated and have not travelled far from their source vents. Aerial views show radiating surface ridges (Fig. 5f) resembling those reported in silicic lava flows at Cordon-Caulle (Farquharson et al. 2015). Surface ridges are compressional features formed in response to flow under a cross-sectional viscosity gradient (between the cooled crust and the hot, fluid lava beneath; Fink 1980).
Outgassing activity is currently focused around the rim of the summit crater, with a further region of diffuse steaming ground located on the upper western flank of the edifice. We sampled the summit crater emissions in two places: first at a fumarole on the inside of the eastern rim and then at another on the southwestern rim (annotated on Fig. 5d and described in Table S1, supplementary information). The two measurement sites showed contrasting gas compositions, as measured by Multi-GAS ( Fig. 6; Table 1). We detected substantial excess CO 2 concentrations up to 2134 ppm (after subtraction of an atmospheric background concentration of 407 ppm) at the first fumarole site CM-1, but no sulfur species. Gas temperatures at the point of emission ranged from 88 to 97°C but varied by less than 2°C during repeat measurements at each fumarole. The second fumarole site CM-2, in contrast, was encrusted by conspicuous yellow sulfur deposits and its emissions contained detectable SO 2 and H 2 S (in the proportion H 2 S/SO 2 = 25.6 ± 2.2), together with excess CO 2 showing pervasive fumarolic outgassing. Gas sampling areas CM-1 and CM-2 are annotated; coordinates and location details are presented in Table S1, supplementary information. f Radiating flow textures on the surface of viscous lava flow lobes. For scale, the height above sea level of the cone is 140 m and the distance between the two sampling sites is~50 m concentrations up to 2027 ppm. Gas temperatures were not measured. We measured a CO 2 /SO 2 ratio of 430 ± 11, equivalent to a CO 2 /S T of 16.4 ± 3.3 when combined with H 2 S. H 2 O was calculated for both sites based on pressure, temperature and relative humidity (see Methods; Table 1). We collected a sample of gas from fumarole site CM-1 using a preevacuated Giggenbach flask; however, the composition of this sample is identical to the composition of air (Table S3; supplementary information)-this is likely due to the low gas flux from the vent. We also note that the flask sample was collected during a blizzard, where the reduced visibility and low temperatures made the procedure very challenging.

Mt. Michael and Ashen Hills, Saunders Island
Saunders is dominated by the conical form of Mt. Michael (890 m asl; Fig. 2). There are two active vents at the summit: a main crater~500 m diameter and a smaller subsidiary icefilled crater to the southeast that has been visibly emitting gas since 2006 (Patrick and Smellie 2013) despite having previously been considered inactive (Holdgate and Baker 1979). The edifice is extensively glaciated and the upper flanks are heavily crevassed. The only ice-free areas are the southeastern peninsula at Nattriss Point, which includes the Ashen Hills pyroclastic cone complex, and the northeastern tip of the island where basaltic lava flows are exposed at Blackstone Plain (Fig. 2). Recurrent thermal anomalies in multi-spectral satellite imagery suggest the presence of a persistent lava lake within the crater ( Strong and persistent gas emissions emanated from the summit of Mt. Michael during 5-8 January 2020. The plume varied from semi-transparent to condensed and could be traced downwind over the ocean for several tens of kilometres (Fig. 7a, b). On 7 January, we flew the aerial Multi-GAS  Fig. 5d) on Lucifer Hills, Candlemas. CO 2 concentration is reported with atmospheric background subtracted (418 ppm). The grey-shaded region shows the 95% confidence interval Table 1 Molar proportions (%) of gas species present in fumarolic emissions. S T refers to total sulfur (SO 2 + H 2 S). All gas molar ratios are presented in Table S2 instrument on-board a multi-rotor UAS (see Methods) from mid-way up the flanks of the volcano to intercept the plume at an altitude~300 m above ground level (~600 a.s.l). The launch site was~600 m northeast of the summit vent, and full details of the sampling site and access requirements are presented in Table S1 (supplementary information). Blizzard conditions combined with treacherous crevassed terrain prevented closer ascent to the summit in pursuit of higher gas concentrations. Linear regression of the SO 2 and CO 2 time series yields a molar CO 2 /SO 2 ratio of 1.8 ± 0.2 (where the uncertainty reported is the standard error of the regression; Fig. 7d). At the dilute plume concentrations encountered (< 1 ppm SO 2 ) and strong eddying winds, there is considerable noise in the data that translates to uncertainty in the gas ratio. We therefore calculated the ratio based on a subsection of the total time series corresponding to the abrupt step change in gas concentrations observed when the UAS first intercepted the plume (n = 72, where the sampling rate is 1 Hz), using a background value that represents an average of the preceding 30 s of measurements. We recognise that the strength of the Looking eastwards out to sea on 7 January 2020 shows the difference in visibility in and out of the gas plume. c Steaming ground, and a ground surface warm to touch, suggests widespread geothermal heating. d CO 2 -SO 2 scatterplot measured using aerial Multi-GAS on 7 January 2020, showing linear regression. The grey-shaded region shows the 95% confidence interval. e The condensed gas plume from Mt. Michael on 8 January 2020, showing the plume travelling at ground level down the eastern flank before lifting off. f SO 2 flux time series on 8 January 2020. Flux data are calculated from UV camera SO 2 absorbance images multiplied by a plume speed derived from optical flow (see Methods). For scale, the height of the snow line is~400-500 m above sea level correlation is not ideal and so should be interpreted with appropriate caution. H 2 S was not present above the detection limit (~13% cross-sensitivity to SO 2 ) in either the aerial or ground-based measurements. Westerly winds on 8 January caused the gas plume to fumigate the eastern flanks of the volcano, yet blizzard conditions and poor visibility prevailed at the summit. The plume travelled at ground level down the eastern flanks of Mt. Michael-resulting in a distinctive blueish haze at ground level-before lifting off over the ocean (Video S3, supplementary material). We encountered SO 2 concentrations up to 2.5 ppm~600 m downwind on the upper flanks of the volcano (~850 m to the east of the summit vent; Table S1): dilute, but enough to be felt in the nose and throat. CO 2 concentrations during this measurement were affected by radio interference, and so, despite attempted correction using signal processing methods, the dataset was unusable. Two bag samples of gas were collected for carbon isotope measurements; these data are presented in Table S4 and described in the "Carbon isotope composition" section.
We derive an SO 2 flux of 145 ± 59 [σ] t day −1 from~3 h of UV camera measurements acquired between 13:40 and 16:40 UTC on 8 January ( Fig. 7f; data presented in Table S2 and Fig.  S2, supplementary information) approximately perpendicular to the direction of plume transport (camera location annotated on Fig. 2c). Plume and ambient sunlight conditions remained relatively constant during this time interval. The condensed nature of the plume introduces additional uncertainty related to multiple scattering effects and leads us to underestimate the total flux, although we do not model this explicitly.
Stratigraphic sequences of tephra (ash and scoria lapilli) layers are preserved in the glacial ice and exposed in crevasse walls (Fig. S3, supplementary information). A layer of well-sorted coarse scoria lapilli (42 cm thickness) preserved in the upper parts of a crevasse was found to be laterally continuous over tens of metres, and similar but slightly thinner deposits were observed elsewhere. The upper part of this deposit has a distinctive band of red oxidation, suggesting a period of subaerial exposure before snow burial. Given the sharp lower boundary of the deposit against the underlying ice and the well-sorted nature of the clasts, we propose that this deposit is stratigraphically in situ and represents a single explosive phase of uncertain duration. Although we lack specific age constraints, given the regional snow accumulation rate, we expect this layer to have been deposited within the last 5 years. The ground surface on the eastern flank is covered by 10-30 cm of loose, unconsolidated ash and lapilli deposits that have likely been remobilised by wind or water. Widespread steaming ground, warm to the touch just below the surface, suggests that geothermal heating is pervasive across the island (Fig. 7c).
Ashen Hills comprises a cluster of overlapping hydromagmatic tuff cones and their associated pyroclastic deposits; the cone complex forms a striking topographic feature of the southeast peninsula of Saunders. Tuff cones (and rings) are built by explosive magma-water interaction involving the interaction of magma with either ground-or surface water (Wohletz and Sheridan 1983;Sohn and Chough 1989). Bombs of dense lava are widespread (Fig. 8d), suggesting considerable syn-eruptive fragmentation and ejection of country-rock material. The ages of the Ashen Hills eruptions are currently unconstrained, as is the geochemical relationship between the Ashen Hills magma and that erupting presently at the summit of Mt. Michael.
Fluvial run-off channels in the Ashen Hills expose sequences of stratified tephra deposits several tens of metres thick (Fig. 8). Large-scale sedimentary features include cross-stratification and undulatory bedding that resemble similar features described from "dry" base surges more generally (Waters and Fisher 1971;Sohn and Chough 1989). Base surges are turbulent, low-particle concentration density currents that are associated with explosive hydromagmatic activity and can be variably water saturated (Moore 1967;Waters and Fisher 1971;Nemeth et al. 2006). Tephra sequences comprise unconsolidated clast-supported beds of ash-coated lapilli and coarse ash alternating with laminated beds of fine to moderately coarse ash. Accretionary lapilli are common, supporting the involvement of magma-water interaction. Beds vary in thickness laterally and erode to produce characteristic circular formations, similar to those observed in eroded base surge deposits elsewhere (e.g. Hverfjall, Iceland; Liu et al. 2017). Unlike many tuff cones, the Ashen Hills deposits are not visibly palagonised.

Montagu
Montagu is a steep-sided eroded stratovolcano (Fig. 2). The structure of the summit region is dominated by a~6-km-wide ice-filled summit caldera-remnant from pre-historical explosive activity-that hosts an active intra-caldera cone, Mt. Belinda (1125 m asl). A prolonged summit eruption from Mt. Belinda began in 2001 and continued until 2007, during which eruptive activity included low-intensity explosive activity, a persistent steam/ash-plume, and several effusive events (Patrick et al. 2005;Patrick and Smellie 2013). A particularly voluminous lava flow in 2005 reached the ocean and built up a sizeable lava delta at the point of entry. Prior to this eruption, there has been no documented Holocene activity at Montagu (LeMasurier et al. 1990). Average annual SO 2 emission rates, measured by satellite OMI, declined abruptly from 180 t day −1 to < 50 t day −1 following the end of the effusive eruption in 2007 (Carn et al. 2017). No observations were made during this expedition.

Bristol
Bristol has been one of the most frequently active subaerial volcanic centres in the SSI over the last century. Historically, volcanic activity has been focused at Mt. Sourabaya (1155 m asl;  1935, 1956, 1962 (unconfirmed) and most recently April-July 2016 (Holdgate and Baker 1979;GVP, 2017b). A summit overflight in 1997 indicated that the summit of Mt. Sourabaya is built of three coalesced vents and steaming pyroclastic cones (Patrick and Smellie 2013). Two lava flows were emplaced concurrently during the 2016 eruption, alongside frequent low-intensity ash plumes from the summit cones that were visible in satellite imagery (GVP, 2017b). We did not observe any expressions of active volcanic activity or degassing as we sailed close in January 2020, although the summit was obscured.
Freezeland Rock is the largest of three small islets (inc. Grindle Rock and Wilson Rock) located on a linear trajectory immediately west of Bristol Island (Fig. S4, supplementary information). The edifice is elongated in an E-W direction and constructed from stratified pyroclastic deposits, mainly tuffs and agglomerates, crosscut by dyke structures of andesitic composition (Holdgate and Baker 1979). The three islets are considered the remnant deposits of one or more offshore volcanic centres.

Southern Thule
The islands of Bellingshausen, Cook and Thule are the subaerial exposures of a much larger submarine landform, collectively known as Southern Thule (Fig. 2). Thule and Cook represent the steeply eroded western and eastern flanks, respectively, of a large submerged caldera (Kemp and Nelson 1931;Smellie et al. 1998). Cook is extensively glaciated and bounded by steeply eroded cliffs. Interbedded lavas and red, oxidised scoriaceous deposits are exposed in the cliff walls, especially at Resolution Point (Fig. 9a, b). There is no record of historical volcanic activity. Thule is entirely glaciated, notably except for an ice-free peninsula at Hewison Point where pyroclastic sequences and dense lavas are exposed (Fig. 9c-e). A 1.5-2-km-wide ice-filled caldera is visible at the summit of Mt. Larsen (746 m asl), the highest point on Thule (Smellie et al. 1998). Steam was observed emanating from a small intra-caldera summit crater in 1962 (LeMasurier et al. 1990), but no emissions have been reported since (Patrick and Smellie 2013).

Bellingshausen
Bellingshausen Island is a small, simply structured, volcanic cone composed of stratified pyroclastic deposits and lava flows (Video S4, supplementary information). It is thought to be one of the youngest subaerial islands in the SSI (Holdgate and Baker 1979). The highest point of the island is on the western rim of the cone at Basilisk Peak (182 m asl; Fig. 2) and the cone tapers to a broad lava plateau to the south. The margins of the island are deeply incised by coastal erosion, exposing vertical cliffs of layered lava flow units (Fig. 10a, b). The northeast and eastern flanks of the volcanic cone are truncated by a sector collapses, exposing the internal stratigraphy (Figs. 10 and 11). A thick layer of welded agglutinate is found wherever the base of the sequence is exposed, overlain by several hundred metres of decimetre-scale massive lava flows interbedded with stratified pyroclastic layers of ash and scoria that dip away radially away from the cone centre in all directions and resemble fall deposits (Fig. 11). Cross-cutting discontinuities between pyroclastic units suggest multiple phases of fall deposition, each separated by periods of erosion. The sequence is capped by a dense lava flow unit.
Fumarolic outgassing inside the crater and along the seaward side of the eastern crater rim was reported by The cone is bounded by a broad lava platform, with steeply incised wave-cut cliffs. c Stratigraphy exposed inside the crater shows thick, massive lava flow units capped by stratified pyroclastics. The fumaroles sampled are visible in the base of the image. d Extensive fumarolic outgassing from the western rim of the crater. e Audible gas emissions emanating from fumarole BH-1 at~99°C. For scale, the height of the cone is 182 m above sea level at its highest point Holdgate and Baker (1979) and again in 1997, together with low-temperature thermal anomalies in satellite imagery between 2000 and 2010 (Patrick and Smellie 2013). We observed a similar distribution of emission sources from numerous fumarolic areas on the walls and base of the crater, together with ubiquitous diffuse emissions from the hydrothermally altered clay substrate. These diffuse emissions were composed only of water vapour as no measurable CO 2 or sulfur species were detected by the Multi-GAS. We measured and sampled two fumaroles in the centre of the base of the crater, which contributed the densest point source emissions (partially shown in Fig. 10c; full details of the sampling sites and access requirements are presented in Table S1, supplementary information). Gas temperatures range from 96 to 98°C. The gas composition as measured by Multi-GAS is dominated by H 2 O (75-84 mol%) and CO 2 (12-24 mol%), with minor H 2 S (0.2-0.3 mol%) and H 2 (< 0.2 mol%; Table 1). No SO 2 was present above the detection limit of 0.1 ppm. CO 2 and H 2 S exhibit a strong positive correlation (r 2 > 0.99), with distinct gradients for the two different fumaroles (Fig. 12). CO 2 /S T ratios (where S T refers to total sulfur) range from 40.6 ± 2.1 at the strong and audibly degassing fumarole BH-1 (gas temperature 99°C; Fig. 10e) to 100.1 ± 5.7 in a region of more diffuse and non-audible emissions downwind from multiple distributed gas sources (fumarole BH-2), about 10 m away from fumarole BH-1 (Fig. 12). Two bag samples of gas were collected from fumarole BH-1 for carbon isotope measurements; these data are presented  Table S4 and described in the "Carbon isotope composition" section. Thick lava flow units of basaltic andesite composition (Leat et al. 2003) dominate the stratigraphy in the south of the island. Layered pyroclastic fall deposits of ash/fine lapilli are exposed on the southwest tip of the island at Hardy Point, however (Fig. 11a). Where the lava flows are exposed on the beaches at Hardy and Isaacson Points, they exhibit a bulbous, brecciated structure with curved brittle fracture surfaces (Fig. 11d-f). The interiors of the flows are moderately vesicular with elongated tear-shaped vesicles.
Volcanic gas compositions (CO 2 /S) and SO 2 flux Molar CO 2 /S T gas ratios from the SSVA vary between 1.8 and 101, reflecting variable interaction of magmatic fluids with hydrothermal systems. Scrubbing of sulfur species by gasrock-water reactions under hydrothermal conditionsespecially at weak, low-temperature fumaroles (< 250°C)produce highly variable CO 2 /S T ratios in fumarolic emissions (Giggenbach 1996;Symonds et al. 2001). Di Napoli et al. (2016) simulate the influence of scrubbing on gas chemistry at near-surface conditions, producing a model that reproduces the near-exponential decrease in CO 2 /S T with increasing temperature observed in global compilations of volcanic gas measurements (Fig. S5, supplementary information; Aiuppa et al. 2017). Low-temperature fumarolic emissions measured at Bellingshausen and Candlemas (T~100°C; Fig. S5, supplementary information) are characterised by CO 2 /S T ratios that are elevated compared to the range of values associated with high-temperature magmatic gases globally (0.2-14.7; Aiuppa et al. 2017). A difference in gas temperature from 99 (strong, audible vent) to 87°C (weaker emission) between two closely spaced fumaroles within Bellingshausen crater manifests as an increase in the CO 2 /S T ratio from 38 ± 2 to 101 ± 6, respectively (Table 1). Furthermore, H 2 S is the dominant sulfur species in the emissions from Candlemas and Bellingshausen, with SO 2 only detected one of the four fumaroles measured. SO 2 is the more prevalent sulfur species in high-temperature magmatic gases, being thermodynamically favourable under oxidising conditions, reducing pressures and increasing temperatures (Gerlach and Nordlie 1975;Giggenbach 1996;Moretti and Papale 2004). Low SO 2 /H 2 S ratios (i.e. a change in speciation to favour H 2 S) reflect chemical equilibration of magmatic gases with colder, more reducing hydrothermal conditions. In the context of interpreting fundamental insights in volatile sources in the SSVA, these low-temperature fumaroles provide little insight, as any magmatic signature has been long since overprinted by hydrothermal processes. Monitoring changes in the gas ratios-particularly SO 2 /H 2 S-would be informative in a hazard context to identify changes in the balance of magmatic and hydrothermal contributions that may precede future phreatic or magmatic activity (e.g. Aiuppa et al. 2007Aiuppa et al. , 2017de Moor et al. 2019), but the logistical effort required to sustain such measurements in these remote sub-polarenvironments would be immense. Mt.
Michael is an open-vent volcano emitting a modest gas flux of 145 ± 59 t day −1 SO 2 from a large summit crater. We measured a molar CO 2 /S T gas ratio of 1.8 ± 0.2 in the downwind plume, which lies close to the median arc composition of 2 (Shinohara 2013) or 1.6 ) based on high-temperature emissions. Together with the lack of detectable H 2 S, this low CO 2 /S T is consistent with a magmatic gas source unmodified by hydrothermal reactions. Globally, the CO 2 /S T is well correlated with the carbon content of the subducted sediments . Our data from Mt. Michael support predicted CO 2 /S T ratios of~1.2 for the South Sandwich Arc based on Ba/La trace element proxies and the carbon-poor nature of regional sediments (Plank 2005;Aiuppa et al. 2019). Considering total volatile fluxes, our calculated SO 2 flux is of a similar order of magnitude to the time-averaged decadal SO 2 flux of 263 ± 63 [σ] t day −1 from the Ozone Monitoring Instrument (OMI) satellite measurements (2005-2015Carn et al. 2017). Combining the CO 2 /SO 2 molar ratio with our independent SO 2 emission rate measured near simultaneously, we derive a CO 2 flux of 179 ± 76 t day −1 . If our CO 2 /SO 2 data are representative of the persistent outgassing behaviour of Mt. Michael, extrapolating our daily CO 2 emission rate yields an annual carbon output of 65 ± 28 kt CO 2 . However, we emphasise that we present here only a snapshot of the gas chemistry; further measurements at both Mt. Michael and Mt. Curry (Zavodovski)-the two most strongly degassing volcanoes of the SSI at present-would help to constrain both alongarc and temporal variability in gas chemistry and flux.
Although Multi-GAS and remote sensing techniques measure a smaller subset of gas species in more dilute plumes compared to a Giggenbach flask sample (arguably the most effective way to sample a "pure" gas), the ability to measure the plume composition and flux remotely proved essential during this campaign, as it does at many other volcanoes globally where high-temperature fumaroles or vent gases are inaccessible or the conditions hostile. Each technique has its merits for the study of volcanic gas chemistry, and we encourage a holistic approach wherever possible.

Carbon isotope composition
Fumarole gases and dilute plume emissions from two locations along the SSVA (the dilute summit plume from Mt. Michael, Saunders Island, and the fumarole in Bellingshausen crater) populate a linear mixing line from clean ocean air background towards that of the volcanic CO 2 composition ( Fig. 12; Table S4, supplementary information). All gas samples have CO 2 concentrations in excess of ambient air (419 ppm) between 467 and 789 ppm, with δ 13 C values between − 7.8 and − 11.3‰. Although unconstrained at higher CO 2 concentrations, extrapolation of the mixing line to 100% volcanic CO 2 indicates a δ 13 C composition of − 4.5 ± 2.0‰ (Fig. 13, where δ 13 C is the deviation of the ratio 13 C/ 12 C relative to that of Pee Dee belemnite and the reported error is the standard error of the regression. We note that our clean ocean air background sample is isotopically light (δ 13 C = − 10‰) compared to the global average suggesting that it may have been fractionated to lighter values during collection. If this is the case, given that all samples were collected and stored in the same way, we assume that all would be fractionated equally. If we normalise our background to the global average δ 13 C = − 8.5‰ (Fischer and Lopez 2016) and apply this offset to all samples, then the extrapolated magmatic carbon isotope composition shifts to a slightly heavier value of − 2.9 ± 2.0‰.
The two sets of plume and fumarole samples were collected from different positions along the arc: Saunders in the central arc and Bellingshausen in the southern arc. That all samples populate a single linear trend (regardless of any vertical offset) suggests a common carbon source. From these data-albeit from a small sample size-we propose no fundamental along-arc variation in the carbon isotope composition along the SSVA. Variations within 4‰ are indistinguishable given the uncertainties of our dataset and the ambient air regression; more samples with higher concentration CO 2 contents would improve these constraints.
Our best-extrapolated value of δ 13 C = − 4.5 ± 2.0 ‰ (or similarly the background-normalised value of − 2.9‰) lies at the upper range of the isotopic signature of the depleted upper mantle (δ 13 C = − 6.5 ± 2.5‰; Sano and Marty 1995) and well below the signature of marine limestones (δ 13 C≈0‰; Fig. 13). Interpreted most simply, this result suggests that most of the emitted carbon is derived from the mantle wedge with only minor addition of a subducted component. However, Th-Sm-La trace element systematics in SSI arc lavas populate a mixing array between depleted mantle and subducted sediment end-members, indicating a variable (but non-negligible) subduction contribution; the Th-REE compositions of Saunders lavas are consistent with~50% sediment and~50% mantle, for example (Plank 2005(Plank , 2014; Fig. S6, supplementary information). In ODP 701, carbonate and organic carbon comprise 75% and 25% of the total carbon by mass, respectively, although each is distributed differently through the core. Carbonate is concentrated in the lowermost units of the core (~70 m), with the bulk composition reaching 70-90% CaCO 3 in the basal indurated chalk. Organic carbon, in contrast, is evenly distributed at low concentrations0 .5 wt% throughout the section, apart from a thin carbonrich ooze at 150-m depth (Ciesielski et al. 1988). If we assign each age-specific δ 13 C signatures to these proportions of carbonate and organic carbon (from Hayes et al. 1999), we derive an approximate weighted average δ 13 C signature for the sediment pile of − 4.8‰ (Table S5, supplementary information). Assuming the lithologies exposed in ODP 701 are representative of those entering the South Sandwich trench, this sedimentary carbon contribution would be indistinguishable from depleted mantle using carbon isotopes alone.
Boron isotope enrichments in SSVA lavas point to an additional subduction contribution from altered oceanic crust (Tonarini et al. 2011). Similarly, Th-Sm-La (Plank 2005) are modelled more effectively by three-component mixing between depleted mantle, sediment and altered oceanic crust (Fig. S6, supplementary information) as are other trace elements (Pearce et al. 1995). Indeed, subduction of carbonated (serpentinised) peridotite-derived either from the downgoing oceanic crust or eroded forearc-is consistent with the composition of local dredge samples and those from other slow-to ultra-slow-spreading settings globally (Lawver and Dick 1983;Vanneste and Larter 2002;Tonarini et al. 2011;Plank and Manning 2019). Globally, the carbon isotope signatures of serpentinites exposed in ophiolite sequences and at oceanic ridges are highly variable-reflecting mixing between seawater carbonate and organic carbon-with bulk δ 13 C values ranging from − 4.6 ± 5.1‰ in the Northern Appenines ophiolites to − 13.7 ± 9.6‰ in the Lost City hydrothermal field (Alt et al. 2013). From the data presented in this study, we cannot evaluate quantitatively the relative contributions from each of the proposed carbon sources. Rather, we intend this discussion to highlight the contrasting scenarios in which both simple and complex mixing of distinct carbon sources may generate the carbon isotopic signature observed in volcanic gas emissions and thus motivate further research into the efficiency of volatile and trace element transfer through this young arc system.  Fig. 13 Carbon isotope composition of volcanic gas emissions. Samples include fumarole gases (Bellingshausen), dilute plume emissions (Saunders) and clean ocean air background (at sea). All data points are included in the regression. The intersection of the regression line with the y-axis represents the magmatic carbon isotope composition extrapolated to 100% CO 2 . Data presented in Table S4 (supplementary material)