First direct observation of sea salt aerosol production from blowing
snow above sea ice

Abstract. Two consecutive cruises in the Weddell Sea, Antarctica, in winter 2013 provided the first direct observations of sea salt aerosol (SSA) production from blowing snow above sea ice, thereby validating a model hypothesis to account for winter time SSA maxima in polar regions not explained otherwise. Blowing or drifting snow always lead to increases in SSA during and after storms. Observed aerosol gradients suggest that net production of SSA takes place near the top of the blowing or drifting snow layer. The observed relative increase of SSA concentrations with wind speed suggests that on average the corresponding aerosol mass flux during storms was equal or larger above sea ice than above the open ocean, demonstrating the importance of the blowing snow source for SSA in winter and early spring. For the first time it is shown that snow on sea ice is depleted in sulphate relative to sodium with respect to sea water. Similar depletion observed in the aerosol suggests that most sea salt originated from snow on sea ice and not the open ocean or leads, e.g. on average 93 % during the 8 June and 12 August 2013 period. A mass budget calculation shows that sublimation of snow even with low salinity (


wind speed suggests that on average the corresponding aerosol mass flux during storms was equal or larger above sea ice than above the open ocean, demonstrating the importance of the blowing snow source for SSA in winter and early spring. For the first time it is shown that snow on sea ice is depleted in sulphate relative to sodium with respect to sea water. Similar depletion observed in the aerosol suggests that most sea salt originated from snow on sea ice and not the open ocean or leads, e.g. on average 93% during the 8 June and 12 August 2013 period. A 10 mass budget calculation shows that sublimation of snow even with low salinity (<1 psu) can account for observed increases of atmospheric sea salt from blowing snow. Furthermore, snow on sea ice and blowing snow showed no or small depletion of bromide relative to sodium with respect to sea water, whereas aerosol at 29 m was enriched suggesting that SSA from blowing snow is a source of atmospheric reactive bromine, an important ozone sink, with bromine loss taking place preferentially in the aerosol phase between 2 and 29 m above the sea ice surface. Evaluation 15 of the current model for SSA production from blowing snow showed that the parameterisations used can generally be applied to snow on sea ice. Snow salinity, a sensitive model parameter, depends to a first order on snowpack depth and therefore is higher above first-year than above multi-year sea ice. Shifts in the ratio of FYI and MYI over time are therefore expected to change the seasonal SSA source flux and contribute to the variability of SSA in ice cores, which both represents an opportunity and a challenge for the quantitative interpretation of the sea salt sea 20 ice proxy. It is expected that similar processes take place above Arctic sea ice.

Introduction
Atmospheric aerosol represents the largest source of uncertainty in global climate predictions (Boucher et al., 2013), and includes sea salt aerosol (SSA), which is the main background aerosol above the oceans. In general global climate and aerosol models strongly under-predict Aitken and accumulation mode particle concentrations compared 5 to observations (Mann et al., 2014) and do not capture the winter maximum of SSA observed at several locations in the Arctic and Antarctica (e.g. Huang and Jaeglé, 2017). A quantitative understanding of SSA sources is critical since SSA influences radiative forcing and therefore climate both directly by absorbing and scattering sunlight and indirectly by modifying the reflectivity, lifetime, and extent of clouds (O'Dowd et al., 1997;DeMott et al., 2016).
Uncertainties in cloud properties explain much of the spread in the modelled climate sensitivity (Flato et al., 2013) 10 and are due in large part to lacking knowledge about aerosol sources. SSA plays also an important role in polar tropospheric ozone and halogen chemistry through the release of active bromine in polar spring contributing to ozone depletion events (ODEs) (e.g. Yang et al., 2010;Kalnajs et al., 2013;Choi et al., 2018). Furthermore, SSA is easily measured in polar ice cores but its use as a quantitative proxy of past sea ice conditions is complicated by uncertainties related to SSA source contributions and processes as well as transport meteorology (Abram et al.,15 2013; Levine et al., 2014;Rhodes et al., 2017), and more recently in the case of bromide Br − (e.g. Spolaor et al., 2013) also to post-depositional processing associated with the bromine explosion chemistry of ODEs (Simpson et al., 2005;Pratt et al., 2013).
Globally most SSA originates from the open ocean, where sea spray is produced by wave breaking and bubble bursting generating film and jet drops . However, previous observations in Antarctica provide two-fold evidence of 20 the existence of a significant SSA source associated with sea ice. First, SSA exhibits maxima during winter/spring in the atmosphere (Rankin and Wolff, 2003;Jourdain et al., 2008;Legrand et al., 2017) and in seasonally resolved ice core records (e.g. Frey et al., 2006); in deep ice cores highest values are seen during glacial periods (Wolff et al., 2003). Thus the highest values are observed when sea ice is at its seasonal or long-term maximum. And second, SSA in the lower atmosphere above coastal Antarctica is strongly depleted in sulphate (SO 2− 4 ) compared to seawater 25  as are also brine and frost flowers in the sea ice nearby (Rankin et al., 2000). Fractionation of SO 2− 4 and to much lesser extent of sodium (Na + ) in sea ice occurs during the cooling of sea ice brine down to its eutectic point due to precipitation of the mineral mirabilite (Na 2 SO 4 · 10 H 2 O) at temperatures below -6.4 • C resulting in ion ratios in liquid and solid phase that are different compared to sea water (Butler et al., 2016). At first it had been thought that observed fractionated SSA originates mostly from highly saline frost flowers, which exhibit 30 a similar sulphate depletion (Rankin et al., 2000). However, recent laboratory (Roscoe et al., 2011;Yang et al., 2017) and model (Huang and Jaeglé, 2017) studies indicate that frost flowers likely play only a minor role in producing SSA because they do not get easily airborne and do not occur widely enough in space and time to produce large enough SSA quantities.
A more recent hypothesis based on a numerical model suggests that salty blowing snow that undergoes sublimation may be a significant source of SSA with a production per unit area equal or larger than that above the open ocean (Yang et al., 2008). Indeed, SSA winter maxima observed at a number of locations in the polar regions can only be 5 explained if the SSA source from blowing snow based on the parameterisation of Yang et al. (2008) is included in the model (Huang and Jaeglé, 2017). However, the blowing snow hypothesis and mechanism lack validation by direct observation. In particular, the applicability of the scheme used by all model studies to date (Huang and Jaeglé, 2017;Rhodes et al., 2017;Yang et al., 2018) for sea ice is not known, because model parameterisations are based on blowing snow measurements above ice sheets (Budd, 1966;Mann et al., 2000;Nishimura and Nemoto, 2005) and 10 the Canadian prairies (Déry and Yau, 2001, and ref. therein).
In this study we report the first direct observations of sea salt aerosol from within the Antarctic sea ice zone during winter, when the blowing snow source is expected to be most active. We discuss SSA variability during and after blowing snow events supported by a unique set of measurements of physical and chemical properties of atmospheric aerosol, snow particles and the snowpack on sea ice. We then critically evaluate the current model parameterisation 15 of sea salt aerosol production from blowing snow based on the in situ observations and discuss chemical fractionation of sulphate and bromide, aerosol size distribution and mass budget of sea salt aerosol above sea ice. A comparison between our observations and a global chemistry transport model is described in a companion study (Yang et al., 2018).

20
Two consecutive Antarctic expeditions were carried out in the Weddell Sea aboard the German icebreaker RV P olarstern in 2013, the Antarctic W inter Ecosystem Climate Study (AW ECS), ANT-XXIX/6 between 8 June and 12 August (Lemke, 2014), and the spring expedition ANT-XXIX/7 between 14 August and 16 October (Meyer, 2014) (Fig. 1). We report mainly on results from ANT-XXIX/6, which provided a significantly more extensive dataset.
Atmospheric measurements and sampling were carried out continuously from the crow's nest of RV P olarstern, and 25 on the sea ice during nine 3-104 hr long ice stations (Table 1). Ice station S3 was occupied for 3 hr (Table 1) allowing only for collection of snow samples. Instruments were set up on the sea ice typically at 0.8 to 1.5 km distance from the ship. Sea ice thickness measurements during ANT-XXIX/6 (Arndt and Paul, 2018) showed that ice stations S1-6 in the eastern sector of the Weddell Sea were on first-year sea ice, whereas ice stations S7-S9 near the Antarctic Peninsula were on multi-year sea ice (Fig. 1). This is in agreement with the observation that persistent multi-year sea 30 ice in the Weddell sea occurs only east of the Antarctic peninsula due to a circulating ocean current. The distance to the nearest open water was ∼600-1000 km for ice stations S1-6 and ∼200-400 km for ice stations S7-9 ( Fig. 1, Table 1). As RV P olarstern travelled south day lengths decreased until the sun remained entirely below the horizon between 23 June and 7 July providing only a few hours of twilight per day. All times are in UTC. A summary of instrumentation and temporal coverage is given in Table 2 and experimental details for each measurement are described below.

Airborne snow particles
Size resolved number densities of airborne snow particles in the diameter range 36-490 µm were measured at a mean 5 height of ∼0.16 m (range 0.07-0.37 m) on the sea ice, and at 29 m from the crow's nest of RV P olarstern using an open-path snow particle counter (SPC-95, Niigata Electric Co., Ltd) described previously (Nishimura and Nemoto, 2005;Nishimura et al., 2014). In brief, the SPC is a single slit sensor with a laser diode and measures diameter and number of drifting snow particles by detecting their shadows. The SPC is mounted on a self-steering wind vane, and hence, the sampling area A (2×10 −3 m × 25×10 −3 m) and volume (A×0.5×10 −3 m) are maintained perpendicular 10 to the prevailing wind direction. Electric pulse signals resulting from snow particles passing through the sampling volume are sent to a transducer and an analysing data logging system. Assuming that the detected particle size is the equivalent diameter d p of a sphere each signal is classified into one of 64 mean particle diameter classes between 36 and 490 µm. The SPCs were calibrated by the manufacturer in Japan at -10 and -30 • C for particle diameters 137, 229, 314, 399 and 449 µm, respectively. The instrument output at the calibration temperature of -10 • C, close to the 15 ANT-XXIX/6 median of -11.8 • C, showed very good agreement with the expected representative particle diameter.
Therefore no temperature corrections were applied. Particle counts from the smallest diameter class (d p =36 µm ) have large uncertainties due to the instrument's detection limit, whereas particle counts from the largest diameter class (d p =490 µm) showed frequent spikes due to detection of precipitating snow. Particle counts from the smallest and largest diameter class were therefore discarded. 20 Particle counts h measured at a sampling rate SR of 1 Hz are integrated to 1-minute values and divided by sampling area A to obtain particle number flux F in units of m −2 minute −1 . Snow particle number densities for the diameter range 46-478 µm, N 46−478 , are then computed using N 46−478 = F U −1 with horizontal wind speed U and are reported at ambient temperature, pressure and relative humidity in units of m −3 . Wind speeds measured near the crow's nest at 39 m by the ship's meteorology observatory (see section 2.5) and on the sea ice at 2 m by a 25 sonic anemometer (see section 2.5) were extrapolated to the respective instrument heights. To do this a logarithmic wind profile U (z) is assumed given by U (z) = k u * ln(z/z 0 ), with measurement height z, the von Karman constant κ, friction velocity u * and the surface roughness length of momentum z 0 set to 5.6×10 − 5m as measured very consistently above snow at Halley (King and Anderson, 1994). Wind speed at instrument level is then derived using U (z 2 ) = U (z 1 ) ln(z 2 /z 0 ) / ln(z 1 /z 0 ). It should be borne in mind that the distortion of flow caused by the ship may 30 mean both that speed at 39 m is not representative of flow in the far field at that height, and further, the turbulent field strength, which governs the gradient of the logarithmic profile, may be a residual from a different, likely lower, height. Thus, we suggest care when interpreting the data, and estimate that the conversion from particle counts to number density be seen as an estimate suitable for comparison, rather than quantitative with a well behaved uncertainty.

Sea salt aerosol
Size-resolved number densities of SSA sized particles were measured at mean heights of 0.19 m (range 0.13-0.25 m) and 2.16 m (range 1.98-2.2 m) on the sea ice, and at 29 m from the crow's nest of RV P olarstern using a Compact 5 Lightweight Aerosol Spectrometer Probe (CLASP) (Hill et al., 2008;Norris et al., 2008). The CLASP is a closedpath optical particle spectrometer, which aspirates sample air at a nominal flow rate Q of ∼3 STP-L min −1 , which is actively controlled by onboard electronics, and recorded to allow subsequent correction of the particle spectra for any minor flow variations. An improved version of the CLASP instrument as used by Norris et al. (2012) was deployed, which measures a 16-channel size spectrum covering particle diameters d p in the range of 0.36-11.62 µm 10 at ambient humidity at a sampling rate SR of 1 Hz. CLASP pump and scatter cell were calibrated in the lab before and after the cruise. Particle losses to inlet walls are minimized by use of a short inlet of <0.3 m. As an upper limit for particle losses we adopt the estimates by Norris et al. (2012)

Aerosol chemical composition
Aerosol was collected on filters via continuous low-volume sampling using open-face filter holders protected by a wind shield. The method described below follows the approach from a previous study in coastal Antarctica . One sampling unit was deployed on the sea ice for the duration of each ice station, and another one was operated continuously from the crow's nest of RV P olarstern. 25 Aerosol filters were polytetrafluoroethylene (PTFE) 1.0-µm pore membrane filters of 37 mm diameter (Ze- Total sample air volume was measured upstream of the pump using a diaphragm gas meter (KG-2-G1.6, BES), 5 which was not temperature compensated. However, gas meter temperature continuously recorded with a TinyTag logger (RS, Gemini) and atmospheric pressure available from the ship's meteorology observatory (see section 2.5) allowed to determine the total air sample volume at IUPAC standard temperature and pressure. Rotameters with glass float (ColeParmer) installed upstream of the gas meter were used to manually check flow rates 1-2 times daily.
Post-season lab tests confirmed accuracy of the crow's nest gas meter, whereas the sea ice gas meter showed a 10 negative bias of 30%, which is accounted for in the computation of final atmospheric concentrations. Filter sampling intervals were on average 6 hr (range 1. 5-8.6 hr) at 2 m and 12 hr (range 3.5-31 hr) at 29 m, with more frequent filter changes applied during storms. The resulting sample air volumes were 3.3 (range 0.8-5.1) m 3 at 2 m and 6.4 (range 1.8-17.3) m 3 at 29 m.
Since isokinetic sampling could not be applied, sea salt aerosol concentrations are expected to be generally under-15 estimated, especially during high wind speeds. The cut-off diameter of the air intake system, which critically controls the observed aerosol load, had not been quantified. However, previously a cut-off diameter of 6 µm was empirically estimated for a similar sampling set up with filter face velocities of 1.1 m s −1 . In this study average filter face velocities were ∼0.15 m s −1 implying a cut-off diameter smaller than 6 µm. The filter low-bias for sea salt is found to be on average 40% compared to the CLASP measurements as discussed in section 3.4.4.

20
Filter units, consisting of filters, support pads and filter holders were cleaned and assembled prior to shipping to Antarctica in a Class 100 clean laboratory in Cambridge (United Kingdom). Filters and support pads were soaked over-night in methanol (ACS >99.9%, Fisher Scientific) in batches of 25, rinsed 4 times with ultra-high purity water (UHP, electric resistivity 18.2 MΩcm), micro-waved for 3 minutes then rinsed once more with UHP. Filters were dried on their support pads under vacuum in a clean dessiccator. Filter holders were UHP-rinsed, placed for 10 minutes 25 into an ultra-sonic bath, UHP-rinsed 3 times and micro-waved for 5 minutes, then soaked over-night in UHP, followed by 4 UHP rinses and 5 minute microwaving. Filter holders were then dried in the incoming, filtered airflow of the Class 100 laboratory. Filter units were assembled in the Class 100 laboratory, sealed air-tight with plugs and then bagged in 2 layers of polyethylene bags.
In Antarctica, filter units were removed from the bags, mounted onto the sampling head by removing one plug, soaking over-night in UHP, repeating the same set of UHP-rinse and microwaving before drying and double-bagging.
Filters were then extracted in 8 ml of UHP by repeated shaking and immersion into an ultra-sonic bath for >30 min.
Samples were analysed using ion chromatography (IC) as described in the following section. About 7% (10 out of 151) of all filters were visibly grey from exposure to ship exhaust and therefore not analysed in order to protect the IC columns.

5
The field blank, mostly due to the filter itself with contributions from field and extraction procedures, was significant for all ions (Table 3). A mean value has been subtracted from all concentrations leading in some cases to negative values. The combination of small air sample volumes and high field blanks led to relatively high limits of detection (LOD) defined here as 2 times the standard deviation of the field blank (Table 3).

Snow chemical composition 10
In order to determine the snow chemical composition a total of 24 snow pits was sampled at 9 ice stations during ANT-XXIX/6, between 1-2 snow pits at shorter ice stations and up to 8 at the multi-day ice stations S6 and S8 (Table 1). Snow pit profiles were sampled at 2 cm depth resolution with a custom-built, cylindrical, stainless-steel sampling tool yielding samples of ∼60 cm 3 . Blowing or drifting snow was collected at approximately 0.3, 0.9 and 1.6 m above the snow surface using modified Mellor gages with a clear thermoplastic body (Schmidt et al., 1984) referred to 15 as rocket traps. All samples were transferred into 50 ml polypropylene tubes with screw-caps (Corning CentriStar), which prior to field deployment had been UHP-rinsed and dried in a Class 100 clean laboratory in Cambridge. One set of snow samples was melted onboard RV P olarstern to measure aqueous conductivity using a conductivity meter Conductivity values were converted into practical salinity S p using the Gibbs-SeaWater (GSW) Oceanographic Toolbox (McDougall and Barker, 2011), which applies the algorithm of the Practical Salinity Scale of 1978 (PSS-78) (Unesco, 1981(Unesco, , 1983 with an extension to salinities S p < 2 psu (Hill and Woods, 1986). S p is reported in psu 25 (practical salinity unit), approximately equivalent to the weight of dissolved inorganic matter in grams per kilogram of seawater, and has an accuracy as stated by the manufacturer of ±0.001 psu at low salinities (<1 psu).
The other snow samples were shipped frozen back to Cambridge and only melted prior to analysis in spring 2014 and December 2016. Elevated salinities required dilution of samples with UHP water, typically by a factor 100 for most snow samples and a factor 10000 for sea ice and frost flower samples. Samples were analysed for major ions using 30 Dionex ICS2000 ion chromatography systems with reagent free eluent generation. Cation analysis was performed with a CS12A separator column with isocratic methylsulfonic acid elution and a 250 µl sample loop. Anion analysis was performed using an AS17 separator column, gradient elution with potassium hydroxide and a 250 µl sample loop. Measurement accuracies were evaluated using European reference materials ERM A408 (simulated rain water) and CA616 (groundwater) and were all within 5%. Here we only report and discuss concentrations of ions relevant to this study: Na + , Cl − , SO 2− 4 and Br − . Chemical fractionation of ion x (=SO 2− 4 , Br − , Na + ) with respect to ion y (=Na + or Cl − ) in aerosol and snow compared to sea water is evaluated based on the depletion factor DF x = (R RSW -R spl )/R RSW , with R being the x: y mass ratio in reference sea water (RSW) after Millero et al. (2008) and sample, respectively. Throughout 5 DF SO 2− 4 (Na + ) and DF Br − (Na + ) refer to sulphate and bromide depletion with respect to Na + , and DF Na + (Cl − ) refers to sodium depletion with respect to Cl − . DF x between 0 and 1 indicates depletion, whereas DF x <0 indicates enrichment. DF SO 2− 4 values below that of pure mirabilite (= -7.3) are attributed either due to measurement error or sulphate contamination from the ship's engine emissions, and were therefore removed. DF Br − and DF Na + smaller than -3 were also considered outliers and removed. Propagation of the analytical error yields mean uncertainties in DF x of 0.03-0.04.

Ancillary measurements
General meteorology measurements were taken from RV P olarstern onboard sensors described in detail at https: //spaces.awi.de/confluence/display/PSdevices/Bordwetterwarte, and include ambient temperature T a and relative humidity with respect to water RH aq at 29 m (HMT337, Vaisala, Finland), wind speed U and direction at 39 m (Sonic 15 2D a, Thies, Germany) and global radiation (Pyranometer CM11, Kipp&Zonen, Netherlands). Still images recorded every 5 minutes by a webcam in the crows nest provided on occasion when light conditions were adequate further qualitative information on the presence of airborne snow particles. During ice stations the three-dimensional wind components (u, v, w) were measured above the sea ice at 25 Hz with an unheated sonic anemometer (Metek USA-1) that was mounted on a mast at ∼2 m. Processing of raw sonic data in 1-min blocks included temperature cross-wind 20 correction and a double coordinate rotation to force mean w to zero (Kaimal and Finnigan, 1994;Van Dijk et al., 2006), followed by computation of friction velocity u * . T a and RH aq were also measured above the sea ice on a 2-m mast at approximately 0.6, 1.2 and 2.0 m, respectively, using temperature-humidity probes (HMP45, VAISALA).
RH aq was converted to relative humidity with respect to ice RH ice . RH ice from the HMP45 was further corrected after Anderson (1994) to extend the calibrated temperature range to T a <-20 • C. Based on the available data the 25 calibration of RH ice is most accurate for the -40 to -20 • C range, and has greater uncertainty in RH ice in near freezing conditions. Accurate correction of the HMT337 output was not possible since available RH aq values are biased as they had been post-processed by accepting only values up to 105% and setting any values >100% to 100%.
In general, measuring RH ice during blizzards is very difficult and its estimated non-systematic error of ∼5% limits its use for blowing snow calculations. We therefore discuss below only RH ice trends. 30 3 Results and discussion The presentation and discussion of results in the sections below are organised as follows. An overview of observed meteorology, particle concentrations and aerosol chemistry (section 3.1) is followed by the description of blowing snow events during two time periods (section 3.2). We then evaluate the current model mechanism proposed by Yang et al. (2008,2018) for SSA production from blowing snow (section 3.3) and finally discuss chemical composition, 5 aerosol size spectra and air-snow budget of sea salt aerosol above sea ice (section 3.4).

Overview of atmospheric observations
Near-zero or positive ambient temperatures T a occurred when RV P olarstern was in the open ocean at the start and end of ANT-XXIX/6, as well as from the 22 July 2013 onwards, when the ship had moved into the marginal sea ice zone (MIZ) closer to open water (marker B in Fig. 1,2b). After entry into the sea ice zone on 17 June 2013 10 (marker A in Fig. 1) T a decreased to below -20 • C but showed thereafter frequent increases associated with storms ( Fig. 1,2a -b). The correlation between T a and U 39m is weak but significant (R = 0.37, p<0.05).
Winter storms occurred frequently with wind speeds ranging between 10 and 20 m s −1 , occasionally exceeding 20 m s −1 (Fig. 2a), and coincided almost always with snowfall inferred from direct observation and presence of clouds. Bearing in mind that wind direction and origin of air mass depend on the ship's position relative to the 15 path of a low or high pressure system the following is observed: winds were most of the time from W to SSW advecting cold (-15 to -30 • C) air of a low water load with specific humidities q v < 3 g kg −1 (Fig. 3). Second most frequent were winds from SSE to SE advecting relatively warm air often between -5 and 0 • C with higher q v values often between 3-5 g kg −1 (Fig. 3) indicating origin from lower latitudes. And third most frequent were winds from the N coinciding with the highest wind speeds (Fig. 3). The air was (super)saturated with respect to ice most of 20 the time (RH ice >100%), with increased frequency of subsaturation (RH ice <100%) when the winds were from SE ( Fig. 3d). Horizontal wind speed at 2 and 39 m correlated well with friction velocity u * (R=0.89, p<0.05) indicating a well-mixed near-neutral turbulent boundary layer above the sea ice during blowing snow.
Airborne snow particle concentrations at 29 m showed strong variability and a weak but significant correlation with wind speed (R=0.24, p<0.05) (Fig. 2a,c). During most storms measured snow particle concentrations have 25 contributions from both blowing snow from the sea ice and precipitation from above as further discussed in section 3.3.2. At 29 m mean total number densities N 46−478 were 8.7×10 3 m −3 during ANT-XXIX/6 and very similar 7.2×10 3 m −3 during ANT-XXIX/7. Near the sea ice surface snow particles were measured for a total sampling time of 6 days and showed a mean N 46−478 of 2.6×10 5 m −3 , on average ∼60 times the number density observed during the same time at 29 m. 30 Aerosol concentrations at 29 m showed strong variability with many but not all increases associated with storms and blowing snow (Fig. 2d). At 29 m mean total number densities N 0.4−12 were 2.1×10 6 m −3 during ANT-XXIX/6 ( Fig. 2d). N 0.4−12 mean values at 2.0 and 0.2 m during ice stations were 1.4×10 6 and 1.7×10 6 m −3 , respectively, about the same as the number densities observed during the same time at 29 m. The median aerosol particle diameters d p at the measurement heights 0.2, 2.0m and 29 m ranged between 0.60 and 0.66 µm showing dominance of sub-micron sized particles in atmospheric aerosol below the instrument particle size cut-off (>11 µm).
Descriptive statistics of the aerosol chemistry are summarised in Table 4. The Na + concentrations of bulk aerosol showed strong variability with most increases coinciding with storms and aerosol number density N 0.4−12 peaks 5 (Fig. 2e). Na + and sea salt concentrations were in general higher at 29 m than at 2 m (Table 4) (Table 4). Conversely, when in open water at the beginning and end of ANT-XXIX/6 Na + concentrations of bulk aerosol were relatively large but with small or no sulphate depletion consistent with the open ocean as the main SSA source, except for the very first and last sample (Fig. 2e).

Blowing snow events
Drifting and blowing snow events were frequently observed during ANTXXIX-6 ( Fig. 2c), of which seven occurred 15 during the 14 days of total time spent at ice stations. Based on data coverage two 7-10 day long periods were chosen to discuss key features of observed blowing snow and associated SSA increases.

Period 23 June to 3 July 2013
The period 23 June to 3 July featured a major storm with 4 wind speed maxima between 15 and 25 m s −1 centred around midnight of the 24, 25, 27 and 28 June (Fig. 4a). Temperature T a increased during that time from -20 • C on 20 23 June to -3 • C on 26 June before dropping again to -20 • C on 30 June (Fig. 2b).
Wind speed U 39 increased from <5 m s −1 in the late evening of 23 June over a 24 hr period to >20 m s −1 (Fig. 4a).
First snow particles were detected at 29 m once U 39 exceeded a threshold of ∼9 m s −1 with spectral number densities N 46−478 reaching values on the order of 10 4 m −3 including large snow particles with d p >300 µm (Fig. 4b). Nearsurface N 46−478 measured at ice station S2 on 24 June (Table 1) showed very large number densities on the order 25 of 10 6 m −3 across the entire particle size spectrum confirming blowing snow from the sea ice surface as the main particle source (Fig. 4c). A decrease in wind speed U always concurred with a drop in snow particle number densities at 29 m and often also in RH ice (25 June, 26 and 28 June) indicative of air being subsaturated with respect to ice ( Fig. 4a-b).
Aerosol at 29 m showed background spectral number densities N 0.4−12 of 10 5 m −3 for particles with d p <2 µm 30 during the calm periods on 23 June, 30 June and 2 July (U 39m <5 m −1 ), when no blowing snow or precipitation was present (Fig. 4d). During or after the 4 wind speed maxima on 24, 25, 27 and 28 June large increases of aerosol spectral number densities were observed especially of sub-micron sized particles reaching up to 10 7 m −3 , often at RH ice <100% consistent with a source from sublimating snow particles (Fig. 4a,d). Near the surface spectral number densities N 0.4−12 for particles with d p <2 µm during the storm on 24 June remained with 10 5 m −3 below those seen at 29 m ( Fig. 4e) likely due to scavenging of aerosol by snow particles.
Na + concentrations in aerosol at 29 m show strong increases from background values of ∼0.1 µg m −3 to 1.4 µg m −3 on 25 June and to 1.1-1.7 µg m −3 during 28-29 June coinciding with wind speed maxima and increased number 5 densities of aerosol and snow particles (Fig. 4). Na + concentrations returned to background values after the storm on 30 June coinciding with low wind speeds U 39m <5 m s −1 , reduction in aerosol number concentrations and absence of any airborne snow particles (Fig. 4). Na + concentrations also dropped to background values during the storm on 26-27 June, consistent with the absence of blowing snow particles due to wind speed falling below the drift threshold (discussed in section 3.3.1) and a concurrent slight decrease in aerosol concentrations (Fig. 4). 10 During the calm periods on 24 June, 26-27 June and 29 June DF SO 2− 4 of aerosol was negative or near 0, while during the storm episodes aerosol became increasingly more depleted in sulphate with DF SO 2− 4 maxima of 0.53-0.59 coinciding with peaks in aerosol Na + concentration (Fig. 4f). In order to constrain the origin of the observed aerosol three snow pit profiles were sampled at ice station S2 on 24 June. The mean snow depth was 33 cm, and bulk mean Na + concentrations were 50, 6 and 17 µg g −1 and bulk median DF   (Table 1), three snow drift episodes on 11, 12 and 14 July 2013, and a major blowing snow event from 14 to 16 July 2013 concurrent with strong warming were observed (Fig. 5). Starting on 14 July ambient temperature T a rapidly increased over a 12-hr period from ∼-22 to -1 • C ( Fig. 2b). 25 During the snow drift episodes wind speed at 39 m U 39m peaked at 7-10 m s −1 (Fig. 5a), snow particle spectral number densities N 46−478 reached up to 10 6 m −3 near the surface, but remained relatively low at 29 m with 10 2 -10 3 m −3 with d p <100 µm except on 11 July (Fig. 5b-c). During the blowing snow event 14-16 July U 39m ranged between 15 and 20 m s −1 (Fig. 5a). Spectral number densities N 46−478 reached again up to 10 6 m −3 near the surface but 10 4 m −3 at 29 m for particle diameters d p 50-200 µm (Fig. 5b-c). The air within 2 m of the surface was supersaturated 30 with respect to ice during drifting or blowing snow, but then became undersaturated towards the end of the snow drift episodes on 11 and 12 July as wind speed and snow particle concentrations decreased (Fig. 5a-c).
Aerosol at 29 and 2 m above the sea ice showed during the calm periods on 10, 12 and 14 July background spectral Two hours after the onset of snow drift total aerosol number densities N 0.4−12 started to gradually increase at 2 and 29 m to reach peak values first at 29 m after an initial spike, and with a delay of 2 hr also at 2 m coinciding with a decrease in snow particle concentrations ( Fig. 6b-c). Despite RH ice >100% within 2 m of the sea ice surface during the blowing snow event on 15 July (Fig. 5a) sub-saturation cannot be ruled out because of the large error 25 in RH ice near 0 • C. Thus, the observed anti-correlation between aerosol and snow particle number densities at both measurement heights is consistent with (a) increased aerosol production from snow particle sublimation near the top or above the blowing snow layer, where undersaturated conditions are more likely, and (b) reduced aerosol scavenging by snow particles.
Na + concentrations in aerosol followed the observed total N 0.4−12 showing first a maximum at 29 m and then near 30 the surface at 2 m albeit of smaller magnitude (Fig. 6d) suggesting that much of the observed increase of aerosol particles consists of sea salt aerosol. Non-zero DF SO 2− 4 in aerosol suggest that the sea salt particles originate from a fractionated sea ice source.
In order to further constrain the origin of the aerosol Na + and DF SO 2− 4 of the snowpack were measured. Blowing snow was collected during the storm every two hours over a 12 hr period and snow pit profiles were sampled at various ice floe locations from 11 to 14 July prior to the storm. During the storm blowing snow Na + concentrations on average more than doubled from 4.5 to 12 µg g −1 and salinities S p increased from 0.02 to 0.04 psu, whereas DF SO 2− 4 remained fairly constant at a median of ∼0.54 (Fig. 6e-f). The large spatial variability in Na + concentrations and S p of the snowpack masked any temporal trend, yet the snowpack profiles could be grouped into two types, one showing a very steep S p (Na + ) decline with snow layer height from 1-10 psu (10 2 µg g −1 ) near the bottom to 10 −3 psu (10 −1 µg g −1 ) 5 near the surface, and the other showing a less steep decline from 30 psu (10 3 µg g −1 ) to 10 −2 psu (10 1 µg g −1 ) ( Fig. 7a showed large scatter, with the more saline snow being more depleted in sulphate (Fig. 7c). Negative was found at the snow-ice interface and in some snow layers indicating the presence of mirabilite (Fig. 7c).
Comparison shows that the range of Na + concentrations and S p values observed in blowing snow can be explained by mixing between the two snowpack types ( Fig. 7a-b). An increase over time of salinity or Na + concentration in In summary, observations show that blowing or drifting snow leads to increases in aerosol number densities during 15 and after a storm. Spectral N 0.4−12 increased during individual storms by 2-3 orders of magnitude above background levels for particle sizes d p <2 µm. Concurrent increases in aerosol Na + concentration show that the observed new particles consist mainly of sea salt aerosol. Observed aerosol gradients suggest that net production of SSA takes place near the top of the blowing or drifting snow layer. Furthermore, similar sulphate depletion in aerosol, blowing snow and the local snowpack is strong evidence that the bulk of the observed SSA originated from snow on sea ice 20 and not the open ocean, which is consistent with the independent model results of Yang et al. (2018). Advection of SSA from the open ocean cannot be ruled out during storms with warm and moist air, but plays only a minor role due to the large distance to the nearest open water, in particular at ice stations S1-S6 (Table 1). The scale-length for SSA removal over an ice shelf had been estimated previously to be ∼30% per 100 km, thus the reduction of the SSA burden during advection over 600 to 1000 km would be larger than a factor 6 to 20 (Wagenbach et al., 1998, and refs. 25 therein). Distance to the open ocean together with a temporally very close association of snow particle and aerosol number density dynamics (Fig. 6) further support the above finding that a local SSA sea ice source dominates over advection. Observations show also that snow drift is conceptually not different from a blowing snow event with the same physical processes leading to SSA production.
3.3 Evaluation of SSA production scheme from blowing snow 30 It has been proposed that the production or mass flux of SSA from blowing snow, Q SSA , is proportional to the bulk sublimation flux of suspended snow particles, Q s in units of kg m −2 s −1 , and snow salinity, S p in psu (Yang et al., 2008(Yang et al., , 2018. The model scheme however relies on blowing snow measurements above ice sheets (Budd, 1966;Mann et al., 2000;Nishimura and Nemoto, 2005), and parameterisations developed for in-land regions in the high Arctic (Déry and Yau, 2001, and ref. therein). Below we briefly summarise the model mechanism and then evaluate its applicability to snow on sea ice based on the Weddell Sea observations.
The mass flux of SSA from blowing snow Q SSA in units of kg m −2 s −1 is computed as where d p is the snow particle diameter (m), f (d p ) is the particle size distribution of blowing snow and ψ(S p ) the 5 snow salinity probability distribution. The quantities f (d p ) and ψ(S p ) are compared to direct observations from this study, whereas the blowing snow bulk sublimation flux Q s was not amenable to direct measurement under field conditions. However, the model parameterisation used for Q s (Déry and Yau, 2001) depends on observable quantities, which are validated against measurements from this study.
3.3.1 Blowing snow bulk sublimation flux Q s

10
The blowing snow bulk sublimation flux Q s is the local bulk sublimation rate integrated over the entire blowing snow column. Q s is parameterised after Yau (1999, 2001), where K (= 1.1574×10 −5 ) is a factor to convert Q s into units of kg m −2 s −1 , A is an empirical snow age factor, Q s is a normalised column-integrated sublimation rate (mm day −1 , SWE), q bsalt is the saltation layer blowing snow 15 mixing ratio (kg kg −1 ) under ambient conditions, and q b0 the value it would have when the 10-m threshold wind speed U t , below which no snow drift occurs, has its minimum value U t0 (= 6.975 m s −1 ) based on the empirical model of Li and Pomeroy (1997). The ratio q bsalt /q b0 comes from the required introduction of a lower boundary condition for particle number densities in the saltation layer and scales the normalised Q s accordingly. Values for q bsalt are computed with where U 10m is the 10-m wind speed (m s −1 ) and u * is the friction velocity (Eq.24 in Déry and Yau, 1999). For the 10-m threshold wind speed U t an empirical model for dry snow was previously derived based on observations in the prairies of Western Canada, which applies to dry snow conditions and is a function of ambient temperature T a (Li and Pomeroy, 1997) This expression is compared to observations above sea ice from this study as follows. The onset of drifting or blowing snow is defined similar to Mann et al. (2000) as the moment when snow drift density µ right above the snow surface exceeds a threshold value of µ c = 0.005 (0.001) kg m −3 . The snow drift density is calculated as with the density of ice ρ ice (= 917 kg m −3 ) and the measured spectral number density N s . U 10m is extrapolated from the sonic anemometer measurements at 2 m assuming a logarithmic wind profile and 5 T a comes from the uppermost sensor on the sea ice at ∼2 m. Using the lower µ threshold most observed U t values are still within the model range of uncertainty (Fig. 8) and show a mean U t of 7.1 (range 2.2-9.8) m s −1 corresponding to u * t of 0.37 (range 0.08 -0.58) m s −1 . For comparison, u * t for blowing snow above an Antarctic ice shelf during winter was observed to range between 0.18 and 0.38 m s −1 (Dover, 1993;Mann et al., 2000). Earlier work on land based snowpacks found u * t to range from 0.15 m s −1 for loose, fresh, dry snow to 0.4 m s −1 for old, wind-hardened 10 snow (Dover, 1993, and references therein). Thus, u * t observed for snow on sea ice are on average at the upper end of previous observations for snow on land or ice shelves. It is concluded that Eq. 4 provides robust estimates of U t above sea ice for the snowpack conditions encountered during this study exhibiting in general very low salinities (Table 5). However, temperature alone may not be a good predictor of U t if surface snow on sea ice exposed to wind is very saline with a relatively larger liquid fraction at a given temperature due to freezing point depression leading 15 to increased snow crystal cohesion.
In a next step blowing snow mixing ratios in the saltation layer q bsalt were computed by dividing observed drift density µ by the density of air and then compared to q bsalt predicted by the empirical parameterisation after Déry and Yau (1999) (Eq. 3). Considered were only observations when µ >0.001 kg m −3 and when the SPC was mounted below reported values of the saltation layer depth (=0.1 m). It is found that the model overpredicts observed q bsalt 20 on average by at least a factor ∼10 (Fig. 9) suggesting that the conditions during this study were likely different to those in the Canadian Prairie for which Eq. 3 was developed. One model uncertainty is the assumption that the saltation layer is saturated with respect to ice, which has not been confirmed yet for the dry and cold polar boundary layer due to the lack of sufficiently accurate humidity measurements.
However, relative changes of q bsalt are well captured by the model, in general showing an increase with wind speed 25 (Fig. 9a,c). Except during the blowing snow event on 14-15 July (ice station S6) q bsalt decreased even though wind speed remained constant at ∼12 m s −1 (Fig. 9b), as noted above for total N 46−478 near the snow surface (Fig. 6b).
Strong warming during the storm of ∼20 K increased snow particle cohesion and therefore threshold wind speed U t as predicted by the model, thereby reducing the uplift of snow particles (Fig. 9b). A similar case occurred during 26-27 June, when warming increased the theoretical drift threshold and wind speed dropped at the same time below 30 U t consistent with the absence of blowing snow particles at 29 m (Fig. 4b).
The model bias in q bsalt has only a minor impact on estimates of bulk sublimation rate Q s (Eq. 2) and therefore also of SSA production Q SSA (Eq. 1) because the calculation uses not absolute values but ratios of actual q bsalt and maximum q b0 .
3.3.2 Snow particle size distribution f (d p ) above sea ice Previous studies concluded that a 2-parameter gamma probability density function gives a reasonable fit to observed distributions of snow particle diameter f (d p ) (Budd, 1966;Schmidt, 1982;Dover, 1993): with shape and scale parameter α and β, and mean particle diameter d p = α β. Particle size distributions of blowing 5 snow have been found to vary with wind speed and height: at a given height, d p increases with wind speed, whereas at a given wind speed, preferential removal of large snow particles due to gravitational settling leads to a decrease in d p (e.g. Dover, 1993;Mann et al., 2000) and an increase in α with height (Nishimura and Nemoto, 2005). Increasing α is equivalent to a shift to more symmetrical size distributions. Model predictions of the mass flux of SSA from blowing snow Q SSA depend critically on f (d p ) (Eq. 1) and therefore on α and β (Eq. 5).

10
In this study we derived d p and α by fitting a 2-parameter gamma distribution to observed particle size spectra.
Confidence in retrieved values of α and β is affected by two limitations of the SPC measurements. Firstly, SPC size spectra are clipped due to instrument configuration and exclude very small snow particles (d p <46 µm). And secondly, SPC data include noise in the particle number count due to natural variability. A Monte Carlo model, which generates multiple noisy gamma distributions of known α and β shows that retrieving these parameters is robust, 15 even when the data are limited to d p > d p , that is capture only the tail of the data. Furthermore, the uncertainty from under-sampling and data noise in α and β can be retrieved from the relative scale of the smallest size bin and the residual noise.
During blowing snow episodes when U 10m was above the mean observed threshold wind speed U t of 7.1 m s −1 d p decreased with height, whereas α increased with height as shown by the respective median values (Table 6) consistent 20 with previous observations. Increase of α with height during blowing snow was found previously at Mizuho station on the Antarctic Plateau where α was ∼3 near the surface and increased to values >10 further aloft (Nishimura and Nemoto, 2005). Another study at Halley in coastal Antarctica found no gradient in α within the lower 4 m (Dover, 1993), which was set to 2 following the analysis of King et al. (1996).
However, the observed decrease in mean particle size d p with height was much smaller than that inferred from 25 previous studies. For example, median d p at 29 m was 43% larger than d p observed between 1.0 and 9.6 m (=80 µm regardless of wind speed) at Mizuho station (Nishimura and Nemoto, 2005) and ∼4  (131 µm) and the size distribution more skewed (α= 3.2) with a long tail showing significant contributions from particles with d p >200 µm (Fig. 10a).
It is suggested that small or no decrease of d p with height and the long tail in f (d p ) at 29 m can be explained by contributions from falling snow particles since blowing snow frequently coincided with precipitation. In order to analyse f (d p ) of snow precipitation only, a calm 17 h period (3 July 14:30 to 4 July 7:00) period) was chosen during 5 which light snowfall (N 46−478 <800 m −3 at both levels) but no snow drift (U 10m =2.7 m s −1 ) occurred. It is found that the size distribution (d p , α) of precipitation at 29 m (Fig. 10c) was very similar to that during a blowing snow event ( Fig. 10a), and also showed no preferential large size mode as observed at Mizuho station (Nishimura and Nemoto, 2005) preventing a correction of blowing snow particle spectra for snow precipitation based on particle size. The spectral shift to smaller particles near the surface during the snowfall event ( Fig. 10d) was likely due to a combination Plotting d p and α against wind speed allows to evaluate the parameter range characteristic for blowing snow: d p shows the expected increase near the surface but large scatter and little change at 29 m ( Fig. 11a-b). Values of α 20 show at both heights some dependence on wind speed ( Fig. 11c-d) and cluster between 3 and 6 when U 10m > U t .
Median, lower and upper quartiles of d p and α for blowing snow, i.e. when windspeed U 10m is above the observed mean threshold U t (=7.1 m s −1 ), are summarised in Table 6 as guidance for model sensitivity studies.

Salinity probability distribution ψ(S p ) of snow on sea ice
Correct modelling of SSA production from blowing snow (Eq. 1) requires choosing a snow salinity probability 25 distribution ψ(S p ) representative of the region and season under consideration (Massom et al., 2001). High salinities near the snow-sea ice interface originate from incorporation of frost flowers or upwards migration of brine due to capillary action. Brine is found at the top of sea ice as a result of ion exclusion during the freezing of seawater or of flooding due to negative freeboard. Top snow layers in deeper snowpacks not affected by brine migration receive sea salt aerosol through dry or wet deposition of background aerosol or sea spray. 30 Vertical profiles of snow salinity S p measured during this study (Frey, 2017) showed a marked decrease with height above the sea ice surface (Fig. 12a) in agreement with previous sea ice surveys in Antarctica (Massom et al., 2001). High salinities S p (1-43 psu) were found in basal snow, the sea ice surface and frost flowers, whereas very low salinities (<0.2 psu) were observed in snow layers >20 cm above the snow-ice interface and in blowing snow (range 0.01-0.12 psu) (Fig. 12a). The inferred mean migration distance of ∼20 cm above sea ice that brine can affect the snowpack falls within the range of those reported by previous studies, up to 17 cm in the Arctic (Domine et al.,
From the above follows that to a first order the distance of individual snow layers to the sea ice surface determines their salinity, and therefore snowpack depth determines to a first order mean volume-integrated salinity and salinity 5 probability distributions ψ(S p ) in snow on sea ice. Average snowpack depth based on snow pits from this study was 21 cm above first-year sea ice (FYI) and 50 cm above multi-year sea ice (MYI) (  (Table 5), and corresponding salinity probability distributions ψ(S p ) 10 for snow on FYI are shifted to higher salinities when compared to ψ(S p ) above MYI (Fig. 12b). Secondary factors, which contribute to higher salinities of snow on FYI in comparison to snow on MYI are flooding of thin FYI with sea water due to negative freeboard and brine rejection and drainage, which decreases MYI salinity over time.
Previous modelling of SSA production from blowing snow based ψ(S p ) on much higher mean snow salinities S p than observed during this study: 8.5 psu from observations in the Indian ocean sector in August 1995 (Yang et al.,

15
2008) and half that value in a later study (Levine et al., 2014). Bearing in mind that only surface snow, which often has lower density and crystal cohesion, is likely to get airborne, it is sensible to include only the top snow layers in the calculation of average snow salinity. For example, in this study the mean S p of snow layers within the top 10 cm of the snow surface was 0.31 psu (Table 5), a factor 28 smaller than that used for ψ(S p ) by Yang et al. (2008). More recent studies updated the model scheme accordingly using the very low salinities found here in surface and blowing 20 snow (Rhodes et al., 2017;Yang et al., 2018). It should be noted that snow on sea ice even at the lower end of the S p spectrum contains amounts of sea salt large enough to produce significant SSA especially in the sub-micron range (e.g. Yang et al., 2018), and is on average significantly saltier than snow in inner Antarctica. For example, mean ice core Cl − or Na + concentrations on the West Antarctic ice sheet are ∼100 ng g −1 or less (Frey et al., 2006), equivalent to S p of ∼10 −3 psu, thus up to 4 orders of magnitude smaller than in snow on sea ice (Fig. 12a). 25 In summary, model parameterisations for SSA production from blowing snow require a snow salinity probability distribution ψ(S p ), which takes into account regional and seasonal variabilities. Snow depth on sea ice, retrievable from satellite measurements, may be a good predictor of mean snow salinity as more field measurements become available.
3.4 The sea ice source of sea salt aerosol 30 Below we discuss chemical fractionation in snow and aerosol, aerosol spectra and the air-snow sea salt budget characteristic for sea ice. In non-summer months when biogenic non-ss SO 2− 4 is at a minimum Antarctic aerosol was observed to be depleted in SO 2− 4 (with respect to Na + ) and in Na + (with respect to Cl − ) Legrand et al., 2017). Similar fractionation in brine and frost flowers in the sea ice nearby attributed to precipitation of the mineral mirabilite (Na 2 SO 4 · 10 H 2 O) suggested that the observed SSA must originate from the sea ice and not from the open ocean 5 Rankin et al., 2000).
Here we show for the first time that snow on sea ice is depleted in sulphate (Fig. 13a). DF SO 2− 4 of most snow samples was positive, showing large scatter and no particular dependence on snowpack depth or sea ice age (Fig. 13a). Negative suggests presence of mirabilite precipitated during freezing. The median volume-integrated DF SO 2− 4 in snow ranged between 0.24 and 0.35 (Table 5) matching very well the sulphate depletion found in aerosol during the period 10 the snow was sampled (ANT6) with median DF SO 2− 4 ranging between 0.29 and 0.48 (Table 4). The similarity of sulphate depletion in both snow and aerosol strongly suggests that snow sea ice in the Weddell Sea is the dominant source of regional sea salt aerosol.
The sulphate depletion observed in aerosol and snow (Figs. 2f, 13a) is consistent with mirabilite precipitation. Mean temperatures during all ice stations except S3 were below the -6.4 • C threshold of mirabilite precipitation (Table 1).

15
And a recent lab study found that mirabilite precipitation results in DF SO 2− 4 of 0.93 in sea water brine cooled down to -20.6 • C (Butler et al., 2016). Conversely, DF SO 2− 4 <0 in snow or aerosol suggests contributions from non-sea salt SO 2− 4 or crystallized mirabilite, which has a DF SO 2− 4 of -7.3. The latter is supported by the observation that most DF SO 2− 4 <0 occur in or near the sea ice surface, where most of the partitioning between brine and mineral is expected to have occurred during freezing (Fig. 13a). 20 If the partitioning of crystallised Na 2 SO 4 · 10 H 2 O between brine and the snow-ice matrix is the dominating SO 2− 4 fractionation process then Na + should be depleted as well in the brine. Following the analysis of Wagenbach et al. (1998) we take Cl − as the reference species for bulk sea water and find from a mass balance calculation that the Cl − to Na + mass ratio would then be linearly related to the SO 2− 4 to Na + ratio with a slope of -0.98 and an intercept of 2.04. From that a theoretical relationship between DF Na + (with respect to Cl − ) and DF SO 2− 4 (with respect to Na + ) 25 is derived and compared to observations (Fig. 14). The observed fractionation is largely in agreement with the model prediction. Snow on sea ice follows closely the theoretical mirabilite fractionation line, whereas aerosol shows large scatter and a tendency to apparent Na + enrichment (Fig. 14). The latter was observed previously and attributed to acid-induced chloride loss, which had occurred from aerosol in the atmosphere or as a sampling artefact from sea salt already accumulated on the filter surface Legrand et al., 2017). It is noted that if all 30 sulphate is removed by mirabilite precipitation (DF SO 2− 4 =1) then sodium depletion reaches a theoretical maximum of ∼12% (DF N a + =1.1204) (Fig. 14).

Chemical fractionation of Br −
The importance of sea salt as a source of atmospheric bromine species in the mid to high southern latitudes is now well established, and SSA from blowing snow is expected to release bromine (Yang et al., 2008(Yang et al., , 2010 driving ozone depletion events observed during or after snow storms (Jones et al., 2009). One of a number of processes identified 10 to cause halogen release from aerosol or ice surfaces involves the reaction of HOBr with halides (Br − and Cl − ) on acidic ice surfaces forming BrCl and Br 2 that are subsequently photolyzed to form reactive halogen atoms (Abbatt et al., 2012, and references therein). This is a multiphase process which, depending upon the substrate, may or may not proceed faster under acidic conditions (Abbatt et al., 2012), and associated uncertainties are currently still significantly limiting the modelling of impacts on tropospheric halogen chemistry (Sander et al., 2003;Long et al., 15 2014). Multi-phase bromine chemistry models use observed sea salt aerosol depletion of bromide relative to sodium (or chloride) with respect to seawater composition, i.e. bromide depletion factors DF Br − , to estimate the bromine flux from sea-salt aerosol (e.g. Yang et al., 2008;Long et al., 2014). However, to date observations of bromide aerosol depletion are only available from north of 55 • S (Sander et al., 2003, and references therein), with the exception of two recent pioneering studies in Antarctica (Legrand et al., 2016;Hara et al., 2018). Below we discuss bromine 20 chemistry observations above the Weddell Sea during this study.
Bulk aerosol concentrations of bromide ranged between <1 and 19 ng m −3 (Fig. 15a), showing occasional large increases to levels seen elsewhere in coastal Antarctica only during summer (Legrand et al., 2016). However, median bulk aerosol concentrations of Br − (Table 4) were similar to coastal observations at Dumont d'Urville during winter (Legrand et al., 2016). It should be borne in mind that the bromide aerosol concentrations were often below the 25 estimated LOD due to the short filter exposure times employed (Fig. 15a). Median bromide concentrations in snow ranged between 70 and 180 ng m −3 (Table 5).
It is found that aerosol at 29 m was on average strongly depleted in bromide relative to sodium with respect to seawater, whereas aerosol near the sea ice surface at 2 m showed mostly strong enrichment (Table 4, Fig. 15b).
Median DF Br − in the snowpack was 0.00-0.06 suggesting overall no or small depletion in bromide (Table 5). However, 30 examination of individual snow layers as well as blowing snow samples shows large scatter with both positive and negative DF Br − , suggesting both bromide depletion and enrichment (Fig. 13b). The sea ice surface below the snow was with the exception of one sample always enriched in bromide (Fig. 13b). During the storm on 14-15 July (see 20 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2019-259 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 3 April 2019 c Author(s) 2019. CC BY 4.0 License. section 3.2.2) it is found that the median DF Br − in blowing snow collected on 15 July at <1 m was close to zero but 0.39 in aerosol at 29 m sampled over the same time interval.
These observations are in agreement with previous studies in coastal Antarctica: Legrand et al. (2016) were the first to report bromide depletion in aerosol at Dumont D'Urville. And Hara et al. (2018) observed at Syowa Station bromide depletion in aerosol but not in blowing snow, which was enriched, and concluded that heterogeneous 5 chemistry driving bromine activation occurs on sea salt aerosol and not on blowing snow. Lieb-Lappen and Obbard (2015) observed bromide depletion in blowing snow but only at >5 m above the surface.
From this study it is concluded that air-ice recycling of bromine occurs on surface snow, airborne snow and aerosol particles. However, the vertical gradient in DF Br − (Na + ) observed in aerosol suggests that the bulk of the net bromine release must take place in the aerosol phase between 2 and 29 m above the sea ice surface. The bromine net release 10 from large snow particles is expected to be reduced compared to aerosol due a combination of shorter atmospheric lifetime and smaller specific surface area limiting diffusional outgassing. The bromine release from SSA produced by blowing snow may be more efficient because it has a large fraction of sub-micron sized particles (see section 3.4.3), resides at the well ventilated top of the blowing snow layer, and may also be more acidic. Thus sea salt aerosol from blowing snow provides an additional bromine reservoir, which is readily depleted.

15
Models assume a dependence of DF Br − on season and aerosol diameter (Yang et al., 2008;Sander et al., 2003).
DF Br − of aerosol showed a weak positive trend from winter into spring largely due to less frequent negative values during August and September (Fig. 15b, Table 4). Increasing DF Br − is consistent with enhanced bromine loss from aerosol as incident radiation increases concurrent with activation of the reactive bromine (BrO x ) cycle (Fig. 15b).
This becomes more evident considering monthly median DF Br − values of aerosol at 29 m, which increased from June  (Legrand et al., 2016). DF Br − values from this study would be even higher had they been referenced to a sea water ratio adjusted for the maximum possible removal of sodium due to precipitation of mirabilite as done by Legrand et al. (2016). Assuming that observed DF Br − is 25 representative of SSA with a median particle diameters d p of ∼0.60-0.66 µm, it is found that observed bromide depletion is larger than that reported by Sander et al. (2003) with a maximum of 0.4 for a d p of 1.3 µm, dropping to 0.1 at 10 µm.

Aerosol size distribution
Average aerosol number and volume distributions observed in the Weddell sea show that during calm conditions 30 (U 10m 3±1 m s −1 ) concentrations across most of the size spectrum were smaller above sea ice than above the open ocean (Fig. 16). The wind speed chosen for calm conditions is well below the mean snowdrift threshold U t of 7.1 m s −1 observed during this study and at the lower end of the range when breaking of waves commences (O'Dowd et al., 1997). A low sea salt aerosol background above compact sea ice during calm conditions is consistent with the absence shows two modes, one at 1-2 µm and the other at > 7µm (Fig. 16b), indicating that most of the aerosol particle mass resides in the super-micron range as expected for sea salt aerosol.
During stormy conditions (U 10m 10±1 m s −1 ) average aerosol number and volume concentrations above the sea ice increased significantly for all particle diameters d p below 10 µm, and slightly more so at smaller particle sizes 5 (d p <2 µm), albeit remained below those above the open ocean observed over two days during this study (Fig. 16).
However, the relative increase of aerosol number and volume concentrations during storms above background levels over the sea ice was equal or larger than above the open ocean. This implies that corresponding aerosol number and mass fluxes during storms were also equal or larger above sea ice than above the open ocean consistent with Yang et al. (2008) and underlining the importance of a sea ice source for atmospheric sea salt aerosol in winter and early 10 spring (Fig. 16).
The main observational constraint for the origin of the observed sea salt aerosol comes from its chemical source signature as discussed above. In addition, the relatively larger increase of small particle number densities provides further support for the blowing snow SSA source based on physical arguments: snow particles of low salinity as observed during this study will generate very small sea salt particles after complete loss of water ice by sublimation.

15
Taking the dry particle diameter d dry = d p (S p /1000 ρ ice /ρ NaCl ) 1/3 , where ρ ice and ρ NaCl are densities of ice and NaCl, respectively, we would expect based on observed snow salinity and snow particle size (Table 5, 6) potential d dry on the order of 10 −3 µm, or larger if not all water is lost. Instruments with a lower size detection limit than available in this study are needed to further investigate very small particle formation from blowing snow.
It is important to note that the use of sea salt measured in ice cores as a proxy of past sea ice conditions requires 20 that average emissions above sea ice exceed those above an equivalent area of open ocean and consequently lead to a comparatively larger burden of atmospheric SSA. Observations from this study support the former requirement, but not the latter. The low SSA background concentrations observed above sea ice may have been due to a combination of low storm frequency and low snow salinities in the area. Indeed, model calculations suggest that size distribution and associated flux of SSA originating from blowing snow are very sensitive to snow salinity S p , e.g. decreasing S p 25 from 0.92 to 0.06 psu causes SSA spectral number densities to decrease by about one order of magnitude (Yang et al., 2018). Further atmospheric modelling is needed to address this issue and will be subject of future research.

The sea salt atmosphere-snow mass budget
Assessing the sea salt mass distribution between atmosphere and snow during calm and blowing snow conditions provides further evidence for the importance of snow on sea ice as a SSA source (Fig. 17). To do this atmospheric 30 sea salt concentrations were estimated in two ways for the time period when RV P olarstern was well within the FYI zone (18 June to 21 July 2013; Fig. 1): one by multiplying the Na + concentration measured on aerosol filters by 3.262 based on the Na + mass fraction in reference seawater (Millero et al., 2008). And the other by converting the observed spectral particle number densities N 0.4−12 into total particle mass concentration assuming spherical particles with the density of NaCl (= 2160 kg m −3 ). Comparison shows that the median sea salt concentrations derived from N 0.4−12 during filter sampling intervals are on average 40% larger than filter-based values (data not shown). A low bias of the filter samples is expected because the smaller cut-off diameter (<6 µm) compared to the optical particle counter (>11 µm) limits capture of coarse sea salt aerosol, where much of the particle mass is located (Fig. 16b). 5 During storms median atmospheric sea salt concentrations from both estimates showed a significant increase above background values (Fig. 17a). For example, median sea salt concentration based on N 0.4−12 increased by a factor 3, from 390 to 1215 ng m −3 when comparing calm (3±1 m s −1 ) with windy (10±1 m s −1 ) conditions (Fig. 17a).
A potential atmospheric sea salt concentration if surface snow released its sea salt content by blowing snow sublimation is estimated as follows: taking a typical column total blowing snow sublimation rate of 0.1 mm day −1 10 observed at Halley during winter (King et al., 2001) and assuming a mean storm duration of 1 day on average the sea salt within the top 0.1 mm of snow can be released. Total sea salt mass observed in the top 0.1 mm of surface snow was converted into atmospheric concentration assuming a snow density ρ snow of 300 kg m −3 and complete mixing into a winter atmospheric boundary layer with an estimated mean depth of 100 m (Fig. 17). Further assuming that the atmospheric concentration measured at 29 m is constant throughout the 100 m atmospheric column it is found 15 that the potential snow reservoir is a factor 10 larger in comparison to the total atmospheric burden and therefore easily accounts for the sea salt aerosol increase observed during blowing snow (Fig. 17a). (with respect to Na + ) observed in aerosol and snow show that most of the aerosol observed during storms had its origin in the regional snow on sea ice (Fig. 17b). 25

Conclusions
Two consecutive sea ice cruises in the Weddell sea, Antarctica, during winter/spring 2013 provided the first direct observations of sea salt aerosol production from blowing snow above sea ice, thereby validating a model hypothesis to account for winter time sea salt maxima in polar regions not explained otherwise. Blowing or drifting snow always lead to increases in aerosol during and after storms consisting mostly of sea salt inferred from the concurrent increase snow layer SSA net production is suppressed due to saturated conditions and scavenging by snow particles which is efficient at the warmer temperatures encountered during storms.
During storms average number and volume spectra of SSA over the sea ice increased significantly above background concentrations but remained below those over the open ocean. However, the observed relative increase of SSA concentrations with wind speed suggests that on average the corresponding aerosol mass flux during storms was 5 equal or larger above sea ice than above the open ocean, demonstrating the importance of the blowing snow source for atmospheric sea salt aerosol in winter and early spring. Lower SSA concentrations above sea ice relative to the open ocean may have been due to a combination of low storm frequency and low snow salinities in the area sampled.
Upscaling of the SSA source flux of Weddell sea ice and atmospheric modelling similar to the study of Levine et al. (2014) are needed to address this issue as well as implications for the sea ice proxy interpretation. 10 The main evidence for the sea ice origin of the sea salt aerosol observed comes from its chemical fractionation: for the first time we show that snow on sea ice is depleted in sulphate relative to sodium with respect to sea water.
Similar depletion observed in the aerosol above the sea ice suggests that most sea salt originated from snow on sea ice with possibly minor contributions from frost flowers, and not the open ocean or leads, e.g. on average 93% during the 8 June and 12 August 2013 period. A temporally very close association of snow particle and aerosol number 15 density dynamics together with the far distance to the open ocean further support that the local blowing snow source of SSA dominates over advection. A mass budget calculation shows that even snow with low salinity (<1 psu) can account for observed increases of sea salt from blowing snow.
It is found that SSA originating from blowing snow is an important source of reactive bromine to the atmosphere, which then can contribute to ozone depletion events. On average snow on sea ice and blowing snow showed no 20 or small depletion of bromide relative to sodium with respect to sea water, whereas aerosol at 29 m was enriched suggesting that bromine loss takes place preferentially in the aerosol phase between 2 and 29 m above the sea ice surface.
Evaluation of the current model mechanism for sea salt aerosol production from blowing snow (Yang et al., 2008(Yang et al., , 2018 showed that the model parameterisations used can generally be applied to snow on sea ice:

25
-estimates of drift threshold wind speed based on the empirical model of (Li and Pomeroy, 1997) that depends on temperature only agreed well with observations above the low-salinity snow encountered during this study. However, temperature alone may not be a good predictor if the surface snow is very saline, implying a relatively larger liquid fraction at a given temperature and therefore increased snow crystal cohesion and drift threshold wind speed. Further measurements above a variety of sea ice surfaces will be needed to address this issue.

30
-modelled saltation layer snow mixing ratios exceed absolute values by a factor of 10, but the relative changes match the observations, and so the impact on calculating the SSA mass flux from blowing snow Q SSA is minor -retrieval of mean particle diameter d p and shape parameter α from fitting a 2-parameter gamma distribution to observed snow particle size spectra is robust, even when data exclude very small snow particles (d p <46 µm); a recommend range of d p and α values is given for blowing snow episodes. Relative contributions from precipitating Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2019-259 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 3 April 2019 c Author(s) 2019. CC BY 4.0 License. snow to total suspended snow particles become more important a few tens of meters above the sea ice surface and need to be accounted for when interpreting snow particle size spectra. Future work needs to fill the observational gap (d p 12-46 µm)between very small snow particles and aerosol due to the importance for the formation of sub-micron aerosol and associated climate impacts.
-to a first order it is the distance to the sea ice surface, i.e. snowpack depth, that determines the salinity 5 probability distribution of snow on sea ice. FYI can therefore be distinguished from MYI based on snow salinity, because snow on FYI is in general more shallow than on MYI. Secondary factors potentially increasing the difference in salinity between FYI and MYI are more frequent flooding of FYI with seawater due to negative freeboard and MYI desalination due to brine drainage. Snow depth retrieved from satellites may allow estimating snow salinity in the absence of ground-based measurements. 10 Snow salinity was shown to be a sensitive model parameter (Yang et al., 2018), which implies that SSA production above FYI should be larger than above MYI. Therefore, shifts in the ratio of FYI and MYI over time are expected to contribute to the variability of SSA in ice cores, which both represents an opportunity and a challenge for the quantitative interpretation of the sea salt sea ice proxy. It is expected that similar processes take place in the Arctic regions.   (Hill et al., 2008;Norris et al., 2008)       d. e.
f. horizontal wind speed U at 39 m; time periods at ice stations S1-9 (Table 1)  c. a.
d. Figure 10. Average snow particle size spectra f (dp) measured at 29 m and <0.2 m during blowing snow on 14 July 23:15-23:25 (a,b) and during light snowfall coinciding with very low winds on 3 July 14:30 -4 July 7:00 (c,d). Solid lines indicate the fit obtained with a 2-parameter gamma probability density function (Eq. 5). Note that mean particle diameters dp computed from observations show a positive bias compared to dp (=α β) retrieved from the fitting procedure, since the SPC instrument does not detect small particles (dp<46 µm). c.
d. Figure 11. Properties of blowing snow particle size distributions above sea ice as a function of the 10-m wind speed U10m retrieved from fitting a 2-parameter gamma probability density function (Eq. 5) to observations. Mean snow particle diameters dp   (with respect to Na + ) of snow on first-year sea ice (FYI, yellow symbols) at ice stations S1-6 and multi-year sea ice (MYI, blue symbols) at ice stations S7-9 in the Weddell Sea as a function of snow layer height above the sea ice surface. For comparison DF SO 2− 4 of the sea ice surface (triangles) and blowing snow at 1-17 cm above the snowpack (squares) are shown as well. Panel (b) shows the same but for bromide depletion factors DF Br − (with respect to Na + ). The sample statistics are summarised in Table 5.    (with respect to Na + ) in aerosol and surface snow. Symbols and errorbars represent median and lower and upper quartiles, respectively.