Regional Characteristics of Atmospheric Sulfate Formation in East Antarctica Imprinted on 17O‐Excess Signature

17O‐excess (Δ17O = δ17O − 0.52 × δ18O) of sulfate trapped in Antarctic ice cores has been proposed as a potential tool for assessing past oxidant chemistry, while insufficient understanding of atmospheric sulfate formation around Antarctica hampers its interpretation. To probe influences of regional specific chemistry, we compared year‐round observations of Δ17O of non‐sea‐salt sulfate in aerosols (Δ17O(SO42−)nss) at Dome C and Dumont d'Urville, inland and coastal sites in East Antarctica, throughout the year 2011. Although Δ17O(SO42−)nss at both sites showed consistent seasonality with summer minima (∼1.0‰) and winter maxima (∼2.5‰) owing to sunlight‐driven changes in the relative importance of O3 oxidation to OH and H2O2 oxidation, significant intersite differences were observed in austral spring–summer and autumn. The cooccurrence of higher Δ17O(SO42−)nss at inland (2.0‰ ± 0.1‰) than the coastal site (1.2‰ ± 0.1‰) and chemical destruction of methanesulfonate (MS–) in aerosols at inland during spring–summer (October–December), combined with the first estimated Δ17O(MS–) of ∼16‰, implies that MS– destruction produces sulfate with high Δ17O(SO42−)nss of ∼12‰. If contributing to the known postdepositional decrease of MS– in snow, this process should also cause a significant postdepositional increase in Δ17O(SO42−)nss over 1‰, that can reconcile the discrepancy between Δ17O(SO42−)nss in the atmosphere and ice. The higher Δ17O(SO42−)nss at the coastal site than inland during autumn (March–May) may be associated with oxidation process involving reactive bromine and/or sea‐salt particles around the coastal region.


Introduction
Sulfate is a major component of impurities trapped in Antarctic ice cores and widely used for reconstruction of paleoclimate conditions (e.g., Legrand & Mayewski, 1997). For instance, since the main source of sulfate in Antarctica is oxidation of dimethyl sulfide (DMS) emitted by marine biota (Cosme et al., 2005;Minikin et al., 1998), sulfate in the Antarctic ice cores are often discussed in light of the past bioproductivity of the Southern Ocean (Goto-Azuma et al., 2019;Legrand et al., 1988;Wolff et al., 2006). In addition to the use of sulfate content in ice cores, 17  ) nss at the coastal site than inland during autumn (March-May) may be associated with oxidation process involving reactive bromine and/or seasalt particles around the coastal region.
Plain Language Summary It has been proposed that the past variations of atmospheric oxidants (e.g., ozone) might be estimated using 17 O-excess, a unique isotopic signature, of sulfate trapped in polar ice cores. However, chemical processes altering 17 O-excess of sulfate in the atmosphere and also in snow after deposition have not been fully understood, limiting the practicality of the signature. We investigated regional differences in 17 O-excess of sulfate in aerosol particles at inland and coastal sites in East Antarctica. Our results suggest that the chemical destruction of atmospheric methanesulfonate, the second abundant sulfur compound in Antarctic aerosols, produces sulfate with significantly high to be a potential tool assessing past oxidant chemistry involving ozone (O 3 ) and hydroxyl radicals (OH), those playing central roles in tropospheric chemistry but not directly preserved in ice cores (Alexander & Mickley, 2015;Kunasek et al., 2010;Murray et al., 2014;Sofen et al., 2014). Δ 17 O(SO 4 2− ) is generally assumed to reflect relative importance of different sulfate formation pathways, since the sulfate produced via gas-phase oxidation of SO 2 by OH possesses Δ 17 O = 0‰ (Barkan & Luz, 2005;Dubey et al., 1997), whereas those produced via aqueous-phase oxidations of dissolved SO 2 (S(IV) = SO 2 ·H 2 O + HSO 3 -+ SO 3 2− ) by O 3 or hydrogen peroxide (H 2 O 2 ) possess Δ 17 O > 0‰ (Savarino & Thiemens, 1999;Savarino et al., 2000;Vicars & Savarino, 2014). Sofen et al. (2014) observed a 1.1‰ increase of Δ 17 O(SO 4 2− ) within the early nineteenth century in West Antarctic Ice Sheet Divide ice cores, probably suggesting an increase of O 3 oxidation relative to OH and H 2 O 2 oxidation in sulfate formation in the mid-southern to high-southern latitude region. However, they estimated by their box model that a 1.1‰ increase of Δ 17 O(SO 4 2− ) requires a 260% increase of relative abundance of O 3 /OH, which they concluded was highly implausible given a 26% increase of O 3 / OH from a chemistry transport model estimate for the Southern Hemisphere extratropics. Based on such results, they pointed out deficiencies in the understanding of sulfate formation other than the recognized SO 2 oxidation by OH, H 2 O 2 , and O 3 . Furthermore, there is increasing evidence for a significant difference between the Δ 17 O(SO 4 2− ) in aerosol samples (ca., 1.5‰; Hill-Falkenthal et al., 2013;Ishino et al., 2017;Walters et al., 2019) and those in ice cores corresponding to the present-day climate conditions (ca., 3‰; Alexander et al., 2002Alexander et al., , 2003Kunasek et al., 2010;Sofen et al., 2014). Despite the significance of this ∼1.5‰ shift in Δ 17 O(SO 4 2− ) compared to the observed variability in ice cores (1.3‰-4.8‰ for glacial-interglacial time scale), there has been no study pointing it out so far. Thus, the interpretation of ice core Δ 17 O(SO 4 2− ) records requires a better understanding of atmospheric sulfate formation in Antarctica.
In Antarctica, where the impact of anthropogenic emissions is still insignificant, there exist unique oxidative conditions associated with natural emissions of reactive trace gases from snow and sea-ice surfaces (Grannas et al., 2007;Simpson et al., 2007). One characteristic is drastic enhancements of photochemical oxidants, represented by OH and O 3 , over the Antarctic Plateau after polar sunrise to the austral midsummer (Crawford et al., 2001;Grannas et al., 2007;Mauldin et al., 2001), which is mainly triggered by nitrate photolysis within snowpack Frey et al., 2009;Noro et al., 2018) emitting reactive nitrogen species (e.g., NO x ) to the atmosphere (Davis et al., 2008). It has been recently found that the concentration of methanesulfonate (MS -), a second abundant product of DMS oxidation after sulfate, suddenly decreased in the highly oxidative atmosphere in midsummer at Dome C, inland Antarctica . It was hypothesized that this may be due to a chemical destruction of MS -, possibly into sulfate, but the hypothesis needs confirmation and quantification. Since it is also known that MSis partially lost in snow after its deposition Wagnon et al., 1999;Weller et al., 2004), it is important to examine the impact of MSdestruction on sulfate formation and Δ 17 O(SO 4 2− ) nss values in snow and ice. Another characteristic is the elevated reactive bromine over coastal Antarctica during austral spring, as indicated from satellite observations of tropospheric BrO columns (Theys et al., 2011). Hypobromous acid (HOBr), which is produced from the BrO + HO 2 reaction, was proposed to represent up to 50% of total sulfate production in the summertime marine boundary layer (MBL) over the Southern Ocean (Chen et al., 2016). The contribution of this reaction is thus expected to be also significant in the Antarctic troposphere.  (Ishino et al., 2017), show similar seasonality with minima in the austral summer and higher values in the autumn to spring, which likely reflects a seasonal shift from OH-and H 2 O 2 -to O 3 -dominated chemistry. In addition, Ishino et al. (2017) and Walters et al. (2019)  appears most significant on the Antarctic Plateau  while reactive bromine is more abundant in coastal regions (Theys et al., 2011 ) observations newly obtained for the inland site Dome C in this study and those previously obtained for coastal site DDU (Ishino et al., 2017) in the same year 2011. We also compare the observations with the Δ 17 O(SO 4 2− ) values estimated using a global chemical-transport model GEOS-Chem, which includes reactive bromine production from sea-salt aerosols that originate from both the open ocean and blowing snow sublimation over sea-ice (Huang et al., 2020).

Aerosol Sampling and Measurement of Soluble Species
Aerosol samples were collected at Dome C (75°10′S, 123°30′E; 3,233 m above sea level), located on the East Antarctic Plateau 1,100 km from the nearest coast. The aerosol sampling is performed continuously at Dome C from the year 2010 as a part of Sulfate and Nitrate Evolution at Dome C (SUNITEDC) program (e.g., Erbland et al., 2013). Bulk aerosol was collected on glass fiber filters using high-volume air sampler (HVAS; General Metal Works GL 2000H Hi Vol TSP; Tisch Environmental, Cleves, OH) at the flow rate of 1.7 m 3 min −1 with time resolution of 1-2 weeks. The HVAS was placed at ∼1 km distant from the main building of research activity at Dome C. A field blank was checked once per month by mounting filters onto the filter holder and running for 1 min. After each collection run, filters were removed from the HVAS and wrapped in aluminum foil, which were sealed in plastic bags, stored at −20°C, and shipped to Institut des Géosciences de l'Environnement (Grenoble, France) for chemical analyses. Samples collected during January-December 2011 were used in this study.
The procedures for extraction and quantification of soluble species concentrations in aerosols are detailed in Ishino et al. (2019). Some anions (NO 3 − and SO 4 2− ) and cations (K + , Mg 2+ , and Ca 2+ ) were measured at Tokyo Tech using ion chromatograph ICS2100, DIONEX with a guard column (Dionex IonPac AG19) and a separation column (Dionex IonPac AS19) for anions, and 881 Compact IC Pro, Metrohm with a guard column (Metrosep C4 S-Guard/4.0) and a separation column (Metrosep C 4-150/4.0) for cations. Considering high blank loading on filters or the lack of ion standard materials at Tokyo Tech, the concentration of other ions (MS − , Cl − , Br − , oxalate (C 2 O 4 2− ), and Na + ) were obtained from other aerosol samples as described in Legrand, Preunkert, Wolff, et al. (2017) and . The measured ion concentrations were corrected for blank values and reported as atmospheric concentration in standard temperature and pressure (T = 273.15 K, p = 101,325 Pa) based on meteorological data of Dome C provided by IPEV/PNRA (www.climantartide.it). The uncertainties were estimated based on the typical uncertainty of the ion chromatography analysis (5%). ) values were measured at Tokyo Tech with an isotope ratio mass spectrometer (IRMS) (MAT253; Thermo Fisher Scientific, Bremen, Germany), coupled with an in-house measurement system built following original setup by Savarino et al. (2001) and a series of improvements (Geng et al., 2013;Schauer et al., 2012). The detailed method is described in Ishino et al. (2017). Briefly, 1 μmol of SO 4 2− was separated from other anions using ion chromatography and chemically converted to silver sulfate (Ag 2 SO 4 ) using ion exchange resin. O 2 produced via thermal decomposition of the Ag 2 SO 4 at 1,000°C within a high temperature conversion elemental analyzer (TC/EA; Thermo Fisher Scientific) was analyzed for isotopic compositions with the IRMS system. The interlaboratory calibrated standards (Sulf-α, β, and ε; Schauer et al., 2012) were used to assess the accuracy of our measurements of our working standards. Measured Δ 17 O(SO 4 2− ) was corrected for oxygen isotope exchange with quartz (Δ 17 O = 0‰; Matsuhisa et al., 1979) by estimating the magnitude of isotopic exchange based on the set of working standard measurements along with the sample measurement runs as described in Schauer et al. (2012) ) content, using the following Equations 1 and 2: (2)

Oxygen Isotope
where "total" is the sum of ss-and nss-SO 4 2− components; k is the mass ratio of [SO 4 2− ] ss /[Na + ] in seawater (0.25; Holland et al., 1986). To take into account sea-salt chemical fractionation processes that occur in the Antarctic region in winter, when temperatures drop below −8°C in the presence of sea-ice , k value of 0.16 ± 0.05 (Legrand, Preunkert, Wolff, et al., 2017)  ] total is minimum.

Complementary Data
The Δ 17 O(SO 4 2− ) nss at Dome C were compared to those previously obtained at DDU (Ishino et al., 2017) in the same year 2011. The intersite difference was evaluated at weekly resolution by subtracting the Δ 17 O(-SO 4 2− ) nss of each DDU sample from that of each Dome C sample collected in the closest time period, hereafter denoted Δ DC-DDU . Note that, at DDU, aerosol samples were collected separately for coarse (>1 μm) and fine (<1 μm) mode particles and the Δ 17 O(SO 4 2− ) nss was measured for fine mode (Ishino et al., 2017). The results were also compared to data sets of oxidants available year round at both Dome C and DDU, O 3 mixing ratios    (Legrand, Yang, et al., 2016), as indicators of regional characteristic processes. Furthermore, [MS -]/[SO 4 2− ] nss mass ratios at Dome C (Legrand,  and DDU (Ishino et al., 2017) for the same year were used to examine the influence of chemical destruction of MSon sulfate formation in the midsummer. ] total , cf., Equation 1), and sulfur isotopic composition of non-sea-salt sulfate (δ 34 S nss ) (Ishino et al., 2019). [H + ], estimated by using the following equation, was used to consider the contribution of aqueous-phase S(IV) + O 3 pathway, which is highly dependent on pH of liquid water in the atmosphere (Seinfeld & Pandis, 2006): ] total was used to consider the potential importance of aqueous-phase S(IV) + O 3 reaction, since it has been recognized that S(IV) + O 3 proceeds rapidly in alkaline (pH ∼ 8) deliquescent solution in fresh sea-salt particles, subsequently shutting off due to acidification of sea-salt aerosol by the produced sulfate as well as by uptake of acidic gases such as SO 2 , HNO 3 , and H 2 SO 4 (Alexander et al., 2005). δ 34 S nss was used to test the impact of sulfur sources on Δ 17 O(SO 4 2− ) nss , since it reflects the relative contributions of marine biogenic (i.e., DMS-sourced) sulfate and nonmarine sulfate (nmb-SO 4 2− ) including volcanic and continental sulfur sources (Ishino et al., 2019;Patris et al., 2000;Pruett et al., 2004). Additionally, 210 Pb was used to trace the contribution of long-range transport of continental submicron aerosols ISHINO ET AL.

10.1029/2020JD033583
4 of 20 (Elsasser et al., 2011;. The relationships were examined separately for each season: December-January-February (DJF), March-April-May (MAM), June-July-August (JJA), September-October-November (SON), as well as October 29 to December 23 (OND), where the last one was defined based on the specifically positive Δ DC-DDU values as explained later in Section 3.1. The obtained correlation coefficients (r and p values) are summarized in Table S1.

Model Description and Limitations
We used v11-02d of the GEOS-Chem global chemical-transport model of coupled aerosol-oxidant chemistry (Park et al., 2004; http://www.geos-chem.org/) to estimate the relative importance of sulfate formation processes and Δ 17 O(SO 4 2− ) nss in Antarctica. The model was run at 4° × 5° (latitude × longitude) horizontal resolution and 47 vertical levels up to 0.01 hPa, using the MERRA-2 assimilated meteorological data developed by the Global Modeling and Assimilation Office (GMAO) at NASA Goddard Space Flight Center. Simulations were performed for January-December 2011 after spinning up the model for 6 months prior to January 2011.
GEOS-Chem v11-02d includes detailed bromine chemistry as described in Parrella et al. (2012), Schmidt et al. (2016), , Sherwen, Schmidt, et al. (2016), and Chen et al. (2017). We also applied the reactive bromine emission scheme from sea-salt aerosols produced by blowing snow sublimation over sea-ice as described in Huang et al. (2018Huang et al. ( , 2020, with surface snow salinity of 0.1 and 0.03 psu over the Arctic and the Antarctic, respectively, for both first-year-ice and multi-year-ice. We assumed an enrichment factor of 9 for Br -/Na + ratio in surface snow on sea-ice relative to seawater. Sea-salt emission from open ocean is simulated as a function of sea surface temperature and wind speed (Jaeglé et al., 2011), with updates from Huang and Jaeglé (2017) for cold ocean waters (SST < 5°C). The model is able to reproduce observed Br y * concentrations at DDU from winter to spring (June-November), within the range of standard deviations of the monthly mean observations (Legrand, Yang, et al., 2016; Figure S1). DMS emission is parameterized as a function of sea surface temperature, wind speed, and DMS concentration in seawater obtained from Lana et al. (2011). Note that the model does not include reactive nitrogen emissions from snow nitrate photolysis (Grannas et al., 2007). This would lead to underestimates in oxidants over large area of the Antarctic continents, as exhibited by the underestimates in surface O 3 concentrations at both Dome C and DDU by a factor of ∼1.5 ( Figure 1e). Zatko et al. (2016) previously simulated that the inclusion of their snow NO x emission scheme in GEOS-Chem model will increase surface O 3 concentrations over the Antarctic continent by factors of 1.1-1.8. It is also recently pointed out that the underestimates in O 3 over the Southern Ocean are improved by the inclusion of a new O 3 deposition scheme associated with chemical reaction of O 3 with iodide on the ocean surface (Pound et al., 2020), which is not implemented in this study. Thus, the model has limitations in reproducing oxidants abundances, even though the general seasonality in O 3 with increases in winter and decreases in summer was reproduced ( Figure 1e). We therefore note that our estimates in the relative contributions of sulfate formation processes bear an uncertainty associated with oxidant abundances despite good reproducibilities in Δ 17 O(SO 4 2− ) nss as shown later in Section 3.2 (Figure 2). The further precise prediction of Δ 17 O(SO 4 2− ) nss will require improvements in oxidants reproducibility in the future.
This version of the model includes gas-phase oxidation of SO 2 by OH, in-cloud aqueous-phase oxidations of S(IV) by H 2 O 2 , O 3 (Park et al., 2004), and HOBr (Chen et al., 2017), and oxidation of S(IV) by O 3 on sea-salt particles (Alexander et al., 2005). For in-cloud reactions, the cloud fraction and the liquid water content of cloud are obtained from MERRA-2 meteorological fields. For pH-dependent reactions such as in-cloud S(IV) + O 3 and S(IV) + HOBr, the effect of heterogeneity of cloud pH on S(IV) partitioning is accounted as described in Alexander et al. (2012). The model assumes in-cloud sulfate formations are prohibited at temperature <−15°C as originally designed by Park et al. (2004). We confirmed that this assumption limits the annual tropospheric sulfate production via aqueous-phase reactions to 13.8 Gg-S on the Antarctic Plateau (>68°S) in contrast to 137.0 Gg-S on the Southern Ocean (60°-68°S) within longitudes of 0°-180°E ( Figure S3). This result is consistent with the limited occurrence frequency of super-cooled-liquid-water containing cloud (0%-10%) over the Antarctic Plateau compared to over the Southern Ocean (20%-60%) (Listowski et al., 2019). S(IV) + O 3 in sea salt is assumed as a function of SO 2 transfer rate constant from the gas to the aerosol phase, because the rate limiting step is not the aqueous-phase reaction rate of S(IV) + O 3 ISHINO ET AL.
10.1029/2020JD033583 5 of 20 in alkaline solution on fresh sea salt but is gas-phase diffusion of SO 2 to the aerosol surface (Alexander et al., 2005). This reaction is calculated only within MBL column, where the newly emitted sea salt is available, assuming that sea-salt alkalinity is rapidly consumed by this reaction in addition to the uptake of gas form H 2 SO 4 and HNO 3 (Alexander et al., 2005). We modified the model to tag sulfate produced via each oxidation pathway as different tracers which are transported, as originally described in Alexander et al. (2005)   ) nss values between Dome C and DDU (Δ DC-DDU , see Section 3.1), (e) ozone mixing ratios ] nss ratios (Ishino et al., 2017;, and (g) total gaseous reactive bromine species (Br y *) (Legrand, Yang, et al., 2016) at Dome C (red) and DDU (blue). Error bars in (b)  via each formation pathway is determined by Δ 17 O value of corresponding oxidant and their transferring factors (Savarino et al., 2000). Since OH also efficiently exchanges its oxygen isotopes with water vapor (Dubey et al., 1997), Δ 17 O(OH) is also 0‰ under regions where water vapor is abundant (e.g., throughout most of the troposphere) (Lyons, 2001;Morin et al., 2007). Therefore, gas-phase SO 2 + OH produces sulfate with Δ 17 O(SO 4 2− ) is assumed to be 0‰. Note that it has been previously suggested that Δ 17 O(OH) can be 1‰-3‰ at Dome C  due to the limited availability of water vapor in inland Antarctica. We also conducted the calculation with Δ 17 O(OH) = 3‰ as a maximum case, which is expected to produce sulfate with Δ 17 O(SO 4 2− ) = 0.75‰ with assuming oxygen atom transferring factor of 0.25. We note that there remains possibility that Δ 17 O(OH) is higher than 3‰  ) value of sulfate produced via aqueous-phase S(IV) + H 2 O 2 is assumed to be 0.8‰ ± 0.2‰ based on Δ 17 O(H 2 O 2 ) of 1.6‰ ± 0.3‰ (Savarino & Thiemens, 1999) with a transferring factor of 0.5 (Savarino et al., 2000). Note that the Δ 17 O(H 2 O 2 ) is derived from only one set of observations at La Jolla, CA, and thus needs further verification in various environment in the future. For Δ 17 O(O 3 ), among the whole sets of observations to the present, the two early studies using cryogenic technique had shown large variabilities (24.7‰ ± 11.4‰ and 26.5‰ ± 5.0‰; Johnston & Thiemens, 1997;Krankowsky et al., 1995). Such variabilities were much greater than those expected from the experimentally determined pressure and temperature dependency of only ∼5‰ for an temperature increase from 260 to 320 K Morton et al., 1990). Based on these experimental data, it has been pointed out that these observations would have random errors associated with sampling artifacts . Therefore, we exclude the data of these two studies from the consideration. Given the consistency of the Δ 17 O(O 3 ) observations using nitrite-coated method among various locations and seasons including at Dome C and DDU (Ishino et al., 2017;Savarino et al., 2016;Vicars & Savarino, 2014), we decided to use the average value of the Δ 17 O(O 3 ) observations, which comes to 25.6‰ ± 1.3‰. The Δ 17 O(SO 4 2− ) for S(IV) + O 3 , both in cloud and in sea salt, is assumed to be 6.4‰ ± 0.3‰, by a transferring factor of 0.25 (Savarino et al., 2000). Since aqueous-phase S(IV) oxidation by HOBr gives an oxygen atom from liquid water to produce sulfate (Fogelman et al., 1989;Liu et al., 2001;Troy & Margerum, 1991)   ) nss at Dome C throughout 2011 in comparison to those previously reported for DDU (Ishino et al., 2017). Note that the concentration data are presented in Ishino et al. (2019). It is well established that [SO 4 2− ] nss is enhanced during austral summer and reduced during winter, a seasonality that is driven by marine biogenic emission of DMS Preunkert et al., 2007Preunkert et al., , 2008. This seasonal cycle in [SO 4 2− ] nss is partially intensified by atmospheric dynamics due to enhanced efficiency of meridional long-range transport and the weakened inversion layer on the Antarctic Plateau during summer, as previously suggested by the similar seasonal cycle with 210 Pb (Elsässer et al., 2011;. Δ 17 O(SO 4 2− ) nss show lower values in austral summer and higher values in winter, with monthly mean values ranging from 1.1‰ ± 0.1‰ in February to 2.5‰ ± 0.2‰ in August, with a mass-weighted annual average of 1.7‰ ± 0.1‰ (Table 1). These trends and values are generally consistent with previous observations at Dome C in 2010 (Hill-Falkenthal et al., 2013) as well as at DDU in 2011 (Ishino et al., 2017). ) nss values between the two sites in the year 2011. Throughout most of the year, Δ DC-DDU is 0‰ within the range of the estimated uncertainty. However, there are two specific time periods exhibiting Δ DC-DDU values different from 0‰. One is from October to December, a transition from the austral spring to summer, when a group of positive Δ DC-DDU values ranging 0.4‰ ± 0.3‰ to 1.4‰ ± 0.3‰ are observed. The second is found from March to May, the austral autumn, when negative Δ DC-DDU values ranging −1.4‰ ± 0.6‰ to −0.5‰ ± 0.3‰ are observed. These Δ DC-DDU values different from 0 suggest that sulfate at Dome C and DDU experienced different oxidation processes during their transport from source regions. The possible processes corresponding to these Δ DC-DDU values are discussed in Section 4. ] nss observations for summer (DJF) and winter (JJA) by a factor of 2 and 4 at Dome C, and 2 and 3 at DDU, respectively. Chen et al. (2018) reported that GEOS-Chem model run with DMS concentration in seawater from Lana et al. (2011) and without DMS oxidation by BrO, the condition used in this study, overestimates mixing ratio of DMS by a factor of 5 and 21 during summer and winter at DDU, respectively. This is likely a main reason for the overestimate of [SO 4

Modeled Sulfate Formation Processes and
2− ] nss at DDU and Dome C, as DMS oxidation is thought to be the main source of sulfate in these locations (e.g., Ishino et al., 2019;Minikin et al., 1998) ) nss in the model as we discuss in Section 4.
The model also reproduces the seasonality of Δ 17 O(SO 4 2− ) nss in the observations with austral summer minima and austral winter maxima, ranging from 0.9‰ (0.8‰-1.3‰, November and January) to 2.4‰ (2.2‰-2.6‰, June) and from 1.0‰ (0.9‰-1.2‰, November) to 2.4‰ (2.3‰-2.6‰, June-July) at Dome C and DDU, respectively. The modeled seasonality results from changes in the relative fractions of sulfate formed via different processes, F i in Equation 5 (Figures 2e and 2f  is significantly higher at Dome C (34% in summer) than DDU (16% in summer) (Table 1). This is because the model prohibits aqueous-phase sulfate production at temperatures lower than −15°C and therefore most sulfate formation along with the transport of precursors toward inland Antarctica occurs through gas-phase SO 2 + OH pathway.
As a result of these estimates in sulfate formation pathways, the model roughly reproduces the observed seasonality and magnitude of Δ 17 O(SO 4 2− ) nss (Figures 2c and 2d). However, the model largely underestimates observed Δ 17 O(SO 4 2− ) nss from August to December at Dome C by 0.5‰-1.1‰ (Figure 2c)  ) nss from September to October by 0.6‰-0.8‰, but rather slightly overestimates in January by 0.5‰ (Figure 2d). This result is further discussed in the following sections within the focus of the observed intersite differences.

Discussion
The Δ DC-DDU values different from 0‰ during the austral spring to summer (October-December) and during the austral autumn (March-May) (Figure 1d) suggest that some fractions of sulfate existing at these inland and coastal sites experienced different oxidation processes. The possible processes may include long-range transport of sulfate produced above the other continents or in the stratosphere (i.e., nmb-SO 4 2− ), in addition to DMS-sourced sulfate produced within the troposphere above the Antarctic continents and the Southern Ocean. In the former case, it is expected that δ 34 S nss values in the same aerosol samples would specifically decrease during the corresponding periods, since nmb-SO 4 2− has lower δ 34 S nss val-  Table S1). Therefore, this nmb-SO 4 2is not likely the main factor causing the positive Δ DC-DDU , while it may dilute or perturb the high Δ 17 O(SO 4 2− ) nss signature in that period. Since δ 34 S nss values were homogeneous between Dome C and DDU for the rest period of the year (Ishino et al., 2019), the negative Δ DC-DDU values during the autumn would neither be associated with the contribution of nmb-SO 4 2-. Additionally, whereas the deposition of polar stratospheric clouds is thought to be a potential source of tropospheric sulfate in Antarctica, it is likely to occur during midwinter (July-August) , not coinciding with the positive and negative Δ DC-DDU values. Furthermore, the relative abundance of 35 S, a radioactive tracer often used as an indicator of stratospheric sulfate, relative to total sulfate is maximized in June (Hill-Falkenthal et al., 2013). Therefore, the intrusion of stratospheric sulfate is not the likely reason for the Δ DC-DDU values discussed here. Thus, below we discuss possible influences of regional characteristic chemistry taking place at the scale of the Antarctic continent on Δ 17 O(SO 4 2− ) nss during these periods.

Positive Δ DC-DDU Values in Austral Spring-Summer
Positive  summer (January; 0.05 ± 0.02) and then increases to maximum values in autumn (March; 0.25 ± 0.09) at Dome C .  found that the decline of [MS -]/[SO 4 2− ] nss during summer coincides with periods of high photochemical activity as indicated by high O 3 levels, suggesting the occurrence of chemical destruction of MS -.  also showed that the decrease in [MS -]/[SO 4 2− ] nss at DDU is less significant than Dome C, while MSdestruction may also occur at DDU where is frequently exposed to the highly oxidative atmosphere from the interior Antarctica during summer due to katabatic wind . The cooccurrence of MSdestruction at Dome C and positive Δ DC-DDU (Figures 1d and 1f)  This hypothesis requires that MSpossesses significantly high Δ 17 O signature or MSdestruction occurs via its oxidation by O 3 to produce sulfate. It is known that MSformation in the MBL involves O 3 as well as BrO, which is produced via Br + O 3 (Zhang et al., 1997), as important oxidants (Hoffmann et al., 2016;von Glasow & Crutzen, 2004). These oxidants would imprint a high Δ 17 O value on MS -. To date, however, there are no observations or estimates of Δ 17 O(MS -). The mechanism and the subsequent products of MSdestruction in inland Antarctica remain unclear. While MSoxidation by OH, SO 4 -, Cl, and Cl 2 has been proposed so far (Zhu, 2004;Zhu et al., 2003aZhu et al., , 2003b, there is no evidence for a reaction with O 3 . A previous box model study simulating multiphase sulfur chemistry (Hoffmann et al., 2016) indicated that aqueous-phase oxidation of MSby OH to produce sulfate (Zhu et al., 2003a) is the dominant pathway under typical pristine MBL conditions (Bräuer et al., 2013). Furthermore, it is also shown by a flow tube chamber experiment that MScan be oxidized on deliquesced aerosols to form sulfate, which may lead to shorter lifetime of MSthan in condensed aqueous phase in MBL (Mungall et al., 2018).  mentioned that, although the chance of aerosol experiencing aqueous-phase chemistry is far lower than in the MBL, far more acidic conditions on the Antarctic Plateau compared to the MBL would favor the production of OH via the reaction of O 3 with O 2 - (Ervens et al., 2003). Thus, here, we assume MSoxidation by OH in aqueous phase or on aerosols as the mechanism for MSdestruction, and for the first time estimate Δ 17 O transferred from DMS oxidation to MSand then to sulfate. , and then to MS -, as well as production of SO 2 from each species, whose importance in the MBL is recognized (Barnes et al., 2006;Chen et al., 2018;Hoffmann et al., 2016;von Glasow & Crutzen, 2004). Note that gas-phase and aqueous-phase reactions using the same oxidants are considered as one pathway for simplification, since they will result in the same Δ 17 O(SO 4 2− ). We also summarize the formula for calculating the Δ 17 O value of each sulfur species X (DMSO, MSIA, MS -, and SO 4 2− ) produced by each reaction j (Δ 17 O(X) j ) in Table 2, which is determined based on mechanisms of respective oxidation pathways as follows.
At the first step, DMS oxidation into DMSO includes four different oxidation pathways reacting with OH, BrO, O 3 , and Cl. These reactions generally occur through adduct of the oxidant to the sulfur atom of DMS (Barnes et al., 2006;Gershenzon et al., 2001;Ingham et al., 1999) (Zhang et al., 1997). DMS + O 3 and DMS + BrO are thus expected to produce DMSO with Δ 17 O of 39‰. Since DMS + Cl pathway is expected to form CH 3 S(Cl) CH 3 , which is followed by subsequent oxidation by O 2 (Barnes et al., 2006), Δ 17 O DMS+Cl is assumed to be equal to Δ 17 O(O 2 ) (=−0.3‰; Barkan & Luz, 2005). During DMSO oxidation by OH into MSIA, one of the two oxygen atoms of produced MSIA is from DMSO while another comes from OH (Bardouki et al., 2002), suggesting that Δ 17 O(MSIA) DMSO+OH is determined as the sum of 1/2Δ 17 O(DMSO) and 1/2Δ 17 O(OH). MSIA is oxidized into MSby OH or O 3 . In MSIA + OH, O-atom added to MSis assumed to come from OH, since ISHINO ET AL.
With the above assumptions on Δ 17 O transferring processes, Δ 17 O value of species X is determined by the isotopic mass balance as the following equation: ISHINO ET AL.

DMS
where f j is relative contribution of reaction j for production of X (P(X)). To obtain Δ 17 O(MS -), we here used P(X) j and thus f j estimated by the previous simulation using GEOS-Chem by Chen et al. (2018), which incorporated whole sulfur chemistry shown in Figure 4. We used the mean P(X) j within the troposphere in 60°-90°S during October-December. For production of DMSO, f DMS+OH , f DMS+BrO , , and f DMS+Cl are estimated to be 36%, 49%, 8%, and 7%, respectively. By applying these estimated f j with Δ 17 O(X) j defined in Table 2  ] nss decreases from 0.14 ± 0.07 to 0), which is consistent with Δ 17 O(SO 4 2− ) MS+OH × 0.14 ± 0.07 = 1.7‰ ± 0.9‰. These consistencies imply that the positive Δ DC-DDU observed in the austral spring-summer is mainly caused by MSdestruction. We therefore conclude that MSdestruction and subsequent sulfate production along with transport over the Antarctic Plateau is the most likely process responsible for the positive Δ DC-DDU in the austral spring-summer at inland Antarctica. ) nss underestimate for DDU during November-December is expected because MSdestruction in the midsummer is less significant at DDU than Dome C  ) nss records. Previous studies revealed that, in Antarctica, MSin snow is largely lost after deposition Wagnon et al., 1999;Weller et al., 2004). While the mechanism of this MSloss in snow is under debated, there are two proposed ideas: physical migration of MSwithin firn layers and possibly by MSoxidation by OH in quasi-brine layer of snow grain. If being viable, the latter MSoxidation should increase ) nss in Antarctic ice corresponding to the present-day warm climate period (Holocene) average 2.8‰ ± 0.4‰ (Alexander et al., 2002(Alexander et al., , 2003Kunasek et al., 2010;Sofen et al., 2014), which is significantly higher than the annual mass-weighted average of Δ 17 O(SO 4 2− ) nss of 1.7‰ ± 0.1‰ in aerosols at Dome C (Table 1) ) nss in snow after deposition. The degree of postdepositional loss of MStends to be higher at sites with lower snow accumulation rates  and reaches 80%-90% at Vostok where the present-day snow accumulation rate is 2.2 g cm -2 year -1 (Wagnon et al., 1999). Assuming 90% of MSin snow is converted into sulfate at Dome C where the accumulation rate (2.7 g cm -2 year -1 ) is similar to the one in Vostok, combined with the annual mean [MS -]/[SO 4 2− ] nss in aerosols of 0.11  2− ] nss in snow at various sites with different snow accumulation rates over Antarctica will be required as a future step. Additionally, observations of Δ 17 O(MS -) in aerosols, snow, and ice will provide useful information to prove the proposed mechanisms as well as to constrain the sulfur chemistry in atmospheric chemical-transport models.

Negative Δ DC-DDU Values in Austral Autumn
The during autumn compared to the other period is larger at DDU (4.7 ± 1.7 ng m -3 for MAM compared to 8.9 ± 1.8 ng m -3 for JJA) than Dome C (1.0 ± 0.4 ng m -3 for MAM compared to 1.9 ± 0.8 ng m -3 for JJA). Given the larger decrease of [Br y *] at DDU than Dome C during autumn, it seems that F S(IV)+HOBr might decrease and Δ 17 O(SO 4 2− ) nss might increase at larger degree at DDU than Dome C, resulting in the negative Δ DC-DDU values. However, although the model also shows the larger decrease in [Br y *] during autumn at DDU (3.5 ± 1.4 ng m -3 for MAM compared to 12.8 ± 2.7 ng m -3 for JJA) than Dome C (5.1 ± 1.4 ng m -3 for MAM compared to 2.2 ± 0.5 ng m -3 for JJA) ( Figure S1), the change in the modeled F S(IV)+HOBr was smaller at DDU (23% for MAM compared to 33% for JJA) than Dome C (18% for MAM compared to 33% for JJA) (  Figure S2), it could be a reason for overestimate of Δ 17 O(SO 4 2− ) nss such that seen for DDU in January ( Figure 2d). We note that, based on the kinetics of HSO 3 -+ HOCl investigated by a flow tube experiment, Liu and Abbatt (2020)  ) nss needs to be increased along with the transport of sulfate and its precursors toward inland Antarctica. It is reasonable that, since liquid water content becomes zero at −40°C (Jeffery & Austin, 1997;Pruppacher, 1995), aqueous-phase S(IV) oxidation is strictly limited and sulfate produced via gas-phase SO 2 + OH pathway with Δ 17 O(SO 4 2− ) nss = 0‰ is more important at inland Antarctica, represented by Dome C where the annual mean temperature is −50°C (Argentini et al., 2014). Additionally, given that [Na + ] is 2 orders of magnitude lower at Dome C (ca., 5 ng m -3 ; Legrand, Preunkert, Wolff, et al., 2017) than DDU (ca., 300 ng m -3 ; Jourdain & Legrand, 2002), it is expected that the sea-salt particles preferentially deposit during transport toward inland Antarctica together with sulfate produced in sea salt, possibly leading to sulfate with lower Δ 17 O(SO 4 2− ) nss values at Dome C. Thus, it seems qualitatively reasonable to observe lower Δ 17 O(SO 4 2− ) nss at Dome C than DDU.
Thus, the changes in sulfate formation processes responsible for the negative Δ DC-DDU values during March-May are uncertain. Considering the small but significant underestimate in Δ 17 O(SO 4 2− ) nss at DDU in the model despite of the good reproducibility at Dome C in that period, we looked into the processes that would be more important around coastal regions than inland such as the decrease in F S(IV)+HOBr and the increase in   S IV O 3 F in sea-salt particles at DDU. However, both of them are not decisive. Given the larger abundances of reactive bromine at coastal regions of West Antarctica than those of East Antarctica (Grilli et al., 2013;Saiz-Lopez et al., 2007;Theys et al., 2011)

Conclusions
We investigated the consequences of characteristic oxidation chemistry in Antarctica for sulfate formation processes and Δ 17 O(SO 4 2− ) nss by comparing weekly Δ 17 O(SO 4 2− ) nss observations at inland site Dome C and those previously obtained at coastal site DDU in the same year 2011. The Δ 17 O(SO 4 2− ) nss at Dome C showed lower values in austral summer (1.1‰ ± 0.1‰ in February) and higher values in winter (2.5‰ ± 0.2‰ in August), with a mass-weighted annual average of 1.7‰ ± 0.1‰, which are generally consistent with previous observations at DDU. This seasonality in Δ 17 O(SO 4 2− ) nss at Dome C is roughly reproduced by the GEOS-Chem atmospheric chemistry transport model, reflecting the increased relative fraction of sulfate produced via SO 2 + OH (34%) and S(IV) + H 2 O 2 (48%) in summer in contrast to the increased fraction of S(IV) + O 3 (30%) and S(IV) + HOBr (34%) in winter. The model also reproduces the Δ 17 O(SO 4 2− ) nss at DDU with estimated sulfate formation processes similar to Dome C but with lower fraction of gas-phase OH oxidation for DDU (16% in summer).
Aside from those general seasonal trends, we found that there are significant differences in Δ 17 O(SO 4 2− ) nss at Dome C and DDU during the austral spring-summer (October-December) and the austral autumn (March-May), indicating the contribution of specific oxidation chemistry to sulfate at each site. For springsummer, the higher Δ 17 O(SO 4 2− ) nss at Dome C than DDU was observed, which coincides with the period when chemical MSdestruction is enhanced under the high photochemical activity at Dome C. Combined with the first estimate of Δ 17 O(MS -) based on the isotopic mass balance calculations, we conclude that MSdestruction producing sulfate with Δ 17 O(SO 4 2− ) nss as high as 12‰ is the most likely process responsible for the observed high Δ 17 O(SO 4 2− ) nss at Dome C and suggests that MSdestruction is responsible for about 10% of total sulfate during spring-summer at this inland Antarctic location. This finding has important implications for the interpretation of ice core Δ 17 O(SO 4 2− ) nss records, since it is known that MScan be also chemically destroyed in snow. This process may lead to a ) nss at DDU than Dome C during autumn may be associated with decreased contribution of S(IV) + HOBr due to the limited reactive bromine availability and/or the increased contribution of S(IV) + O 3 due to the insufficient acidification of sea salt at DDU. Further observations of Δ 17 O(SO 4 2− ) nss at various coastal sites over Antarctica will help to constrain the impact of these processes.