Circum-Antarctic abundance and properties of CCN and INP

. Aerosol particles acting as cloud condensation nuclei (CCN) or ice nucleating particles (INP) play a major role in the formation and glaciation of clouds. Thereby they exert a strong impact on the radiation budget of the Earth. Data on abundance and properties of both types of particles are sparse, especially for remote areas of the world, such as the Southern Ocean (SO). In this work, we present unique results from ship-borne aerosol-particle-related in situ measurements and ﬁlter sampling in the SO region, carried out during the Antarctic Circumnavigation Expedition (ACE) in the Austral summer of 2016/17. An 5 overview of CCN and INP concentrations on the Southern Ocean is provided and using additional quantities, insights regarding possible CCN and INP sources and origins are presented. CCN number concentrations spanned 2 orders of magnitude, e.g., for a supersaturation of 0 . 3 % values ranged roughly from 3 to 590 cm − 3 . CCN showed variable contributions of organic and inorganic material (inter-quartile range of hygroscopicity parameter κ from 0 . 2 to 0 . 9 ). No distinct size-dependence of κ was apparent, indicating homogeneous composition across sizes (critical dry diameter on average between 37 and 123 nm ). The 10 contribution of sea spray aerosol (SSA) to the CCN number concentration was on average small. Ambient INP number concentrations were measured in the temperature range from − 5 to − 27 ◦ C . Concentrations spanned up to 3 orders of magnitude, e.g., at − 16 ◦ C from 0 . 2 to 100 m − 3 . Elevated values (above 10 m − 3 at − 16 ◦ C ) were measured when the research vessel was in the vicinity of land, with lower and more constant concentrations when at sea. This hints towards terrestrial and/or coastal INP sources being dominant close to land. In pristine marine areas INP may originate from both oceanic sources and/or long 15 range transport. Sampled aerosol particles (PM 10 ) were analysed for sodium and methanesulfonic acid (MSA). Resulting mass concentrations were used as tracers for primary marine and secondary aerosol particles, respectively. Sodium, with an average concentration around 2 . 8 µ g m − 3 , was found to dominate the sampled particle mass. MSA was highly variable over the SO, with concentrations up to 0 . 5 µ g m − 3 near the sea ice edge. A correlation analysis yielded strong correlations between sodium mass concentration and particle number concentration in the coarse mode, unsurprisingly indicating a signiﬁcant contribu- 20 tion of SSA to that mode. CCN number concentration was highly correlated with the number concentration of Aitken and accumulation mode particles. This, together with a lack of correlation between sodium mass and Aitken and accumulation mode number concentrations, underlines the important contribution of non-SSA, probably secondarily formed particles, to the CCN population. INP number concentrations did not signiﬁcantly correlate with any other measured aerosol physico-chemical parameter. are presenting ( T > − 30 ◦ C ). Analysis of the HV ﬁlters regarding mass concentrations of sodium and MSA was performed. Total ﬁlter load for each HV ﬁlter was determined using a micro-balance ( AT261 Delta Range , Mettler Toledo, Greifensee, Switzerland). Filter con- 225 tents were extracted and ion chromatography performed, following the procedures described in Müller et al. (2010) and van Pinxteren et al. (2017). Results of the analysis were corrected for standard conditions and are reported as atmospheric mass concentrations (in µ g m − 3 ). An inﬂuence from the RV’s exhaust stack on the measured sodium or MSA concentrations is not expected due to their respective source mechanism. Sodium is used as a conservative tracer for primary aerosol particles of marine origin, and MSA was solely over 1 order of magnitude lower maximum N INP observed in our study. Agreement of their values is highest with the subset of our observations in the proximity to land.

bursting was characterised for CAPRICORN II in McCluskey et al. (2018), using seawater samples. They found that INP were from oceanic sources, aerosolized by bubble bursting. Additionally, Uetake et al. (2020) show that bacteria sampled during CAPRICORN II are mostly of marine origin, suggesting a restricted meridional transport of continental aerosol towards the SO. In consequence, a dominance of sea spray on INP was concluded. 95 However, data on INP abundance, spatial distribution, properties, and sources over the SO region remain sparse. Regayre et al. (2020) pointed out that already a small number of observations from the SO can effectively reduce model uncertainty more than hundreds of measurements in the Northern hemisphere, as current simulations are based on very few observations in the Southern hemisphere. This demonstrates a need for further field measurements of CCN and INP in the SO region.
In this study, we present circum-Antarctic measurements from three months of continuous on-line CCN measurements and 100 filter sampling with subsequent INP, sodium and MSA analysis. Based on a correlation analysis, links between the measured properties are discussed.

Methods
Measurements were carried out in the framework of ACE (Walton and Thomas, 2018). The cruise took place between December 2016 and March 2017 on board the research vessel (RV) Akademik Tryoshnikov. Starting and ending in Cape Town (South 105 Africa), the cruise was divided into three legs: Cape Town to Hobart (Australia), Hobart to Punta Arenas (Chile) and Punta Arenas to Cape Town (Fig. 1). Several islands (Marion, Crozet, Kerguelen, Balleny, Scott, Peter 1 st , Diego Ramierez, South over all mobility diameters gave total aerosol particle number concentration (N total ) for each time step. Analogously, the concentration of particles with D p > 500 nm (N 500 ) was derived, which is later used in a commonly-used parameterization for INP concentration (see subsection 3.4).

145
A Cloud Condensation Nuclei counter (CCNc; CCN-100 instrument by DMT, Boulder, CO, USA) was used to measure the CCN concentration at various SS. The CCNc's main part is a continuous-flow thermal gradient diffusion chamber, in which a stream-wise temperature gradient is induced to achieve defined SS and corresponding particle activation to droplets. The aerosol flow rate inside the CCNc is 0.5 L min −1 . Activated particles are counted by an optical particle counter. Further documentation on the CCNc can be found in Roberts and Nenes (2005). Calibration of the CCNc was performed prior to 150 the cruise, following the standard operating procedure given in Gysel and Stratmann (2014) and recommendations in Schmale et al. (2017). During ACE, the CCNc was operated at SS of 0.1, 0.15, 0.2, 0.3, 0.5, and 1 % maintained for ten minutes each. To ensure stable thermal conditions within the instrument, data collected during the first five minutes of each SS setpoint were discarded. Furthermore it was ensured that (1) the instrument's internal thermal stability control reported thermally stable conditions, and (2) the absolute difference between set and read temperature of the optics was smaller than 2 K. The 155 remaining data were aggregated into one minute intervals and filtered for ship exhaust influences (same as for the SMPS and APS instruments). Based on the filtered values, averaged N CCN at a particular SS were calculated. This procedure results in one N CCN value per hour and supersaturation. During data analysis, CCN concentrations at 0.1 % were found to lack sufficient data quality, therefore measurements at this supersaturation were discarded.
For determining the critical dry diameters for particle activation (D crit ) and aerosol particle hygroscopicity parameters 160 (κ), we applied the procedure used in, e.g., Kristensen et al. (2016) and Petters and Kreidenweis (2007). D crit is implicitly defined as the lower boundary of the integral over the PNSD for which the integrated particle number concentration equals the measured CCN number concentration. In our case, the upper boundary of the integral was always 10 µm, due to using instruments operated on a PM 10 inlet. The κ value, an indirect measure of chemical composition of the CCN at given D crit , is derived from the SS applied in the CCNc and the corresponding D crit . Corresponding to N CCN , one κ value per hour 165 and supersaturation is determined. A Monte Carlo simulation (MCS) approach with an iterative solver was used, following the procedure described in Herenz et al. (2019), to model error propagation in both derivation of D crit and calculation of κ(D crit ). The calculation of D crit and thus κ is highly sensitive to the PNSD which, in our case, depends on the quality of the mode-fitting. To exclude unreasonable values, D crit values were filtered. For this, the range between 10 th and 90 th percentile of D crit was calculated for each SS separately. D crit values outside this range and associated κ values were excluded from 170 further analysis. Hence, the presented results are representative of the most frequently occurring κ values.

Filter sampling for INP, sodium and MSA analysis
Filter sampling of ambient air for off-line INP, sodium and MSA analysis at the laboratories of Leibniz Institute for Tropospheric Research (TROPOS) was carried out using a high-volume sampler (HV; DHA-80 filter sampler, DIGITEL, Volketswil, for INP analysis. LV sampling was performed at eight hours time resolution using track-etched polycarbonate membrane filters (Whatman Nuclepore, Cytiva, Little Chalfont, UK; 200 nm pore size, 47 mm in diameter) at a flow rate of roughly 25 L min −1 . The HV sampler used a flow rate of roughly 500 L min −1 , sampling air through quartz-fibre filters (MK 360, Munktell, Bärenstein, Germany) of 150 mm in diameter for up to 24 hours per filter. Here, each filter's individual sampling time (<1 to 1437 min) was dependent on the automatic shut-down mechanism. In total, 258 LV and 94 HV filters were col-180 lected throughout the cruise, including five (four) un-sampled reference filters for LV (HV) sampling, called field blank filters (FBF). FBF were handled in the same way as the sampled ones, enabling assessment of background concentrations due to both methodology and handling. After sampling, filters were stored in a freezer at −20 • C and shipped frozen to TROPOS for offline analysis after the cruise concluded. INP analysis was performed for both LV and HV filters. LV filters were used solely for the INP analysis, while the HV filters were split between INP, sodium and MSA analysis and reserve samples. HV filters with 185 a too small sampling volume (<100 m 3 ), due to the aforementioned automated shut-down mechanism, were not considered further to prevent unreasonably high conversion factors to infer atmospheric concentrations from filter analysis results. A total of 79 sampled HV filters were included in the following analysis.
The freezing behaviour of the aerosol particles collected on each LV and HV filter was investigated using the Ice Nucleation Droplet Array (INDA) at TROPOS. INDA is based on the freezing array method described in Conen et al. (2012) and a detailed 190 instrument description is given in the supporting information in Hartmann et al. (2019). As a first step of the analysis process, stored filters were acclimatised to roughly −3 • C in a fridge. LV filter contents were washed off by submerging the filter in 7.5 mL (V water ; 10 mL at later stages) ultra-pure water (milliQ, 18.2 MΩ cm −2 ). In contrast, 96 randomly punched-out pieces of 1 mm in diameter (D punchout ) per filter were used for the INP analysis of the HV filters. The 96 wells of a PCR (polymerase chain reaction) plate (BRAND, Wertheim, Germany) were either filled with 50 µL each (V droplet ) of the filter washing water 195 (LV) or with 50 µL of milliQ water and one punch-out (HV). The PCR plate was sealed and partially submerged in the ethanol bath of a cryostat (FP 40,Julabo,Seelbach,Germany). Cooled at a rate of roughly 1 K min −1 , the number of frozen droplets (n frozen ) and corresponding temperature (T ) was documented automatically every six seconds. Recommendations on sample handling and processing given in Polen et al. (2018) were followed.
The frozen fraction (f ice ) was calculated by dividing n frozen by the total number of droplets per PCR plate (n total = 96).

200
Obtained f ice at any T was used to derive the cumulative INP concentration N INP , using: according to Vali (1971). The reference volume V for the LV filters was calculated as: where V flow is the sampled air volume, V water is the volume of washing water and V droplet is the water volume per PCR plate 205 well. For the HV filters, V was calculated using: where D punchout is the diameter of the filter sub-sample per well and D filter,HV the diameter of a HV filter. V flow was logged by both (LV and HV) samplers.
Uncertainties arising from the methodology were assessed similarly to previous studies (e.g., Wex et al., 2019;Gong et al., 210 2020). Confidence intervals for f ice (T ) of each filter were determined using a method described in Agresti and Coull (1998 (2020) that ship exhaust is not ice-active in the temperature range we are presenting (T > −30 • C).
Analysis of the HV filters regarding mass concentrations of sodium and MSA was performed. Total filter load for each HV filter was determined using a micro-balance (AT261 Delta Range, Mettler Toledo, Greifensee, Switzerland). Filter con-225 tents were extracted and ion chromatography performed, following the procedures described in Müller et al. (2010) and van Pinxteren et al. (2017). Results of the analysis were corrected for standard conditions and are reported as atmospheric mass concentrations (in µg m −3 ). An influence from the RV's exhaust stack on the measured sodium or MSA concentrations is not expected due to their respective source mechanism. Sodium is used as a conservative tracer for primary aerosol particles of marine origin, and MSA was found to be solely a product of DMS oxidation (Legrand and Pasteur, 1998).

Further resources
During the ACE cruise sea water was sampled every four hours using the RV's underway water supply system and during CTD (conductivity, temperature and depth) rosette deployments, at specific depths up to 200 m (Walton and Thomas, 2018).
Glass fibre filters (25 mm in diameter, 700 nm pore size) were sampled with up to 2 L of sampled sea water under low vacuum pressure and stored at −80 • C prior to analysis on-board the RV. After extraction in 90 % acetone for 24 h, chlorophyll a (Chl-a) Continuous data of wind speed and direction during the cruise were obtained from two ultrasonic anemometers (part of MAWS 420 system, Vaisala, Vantaa, Finland) located on the port-and starboard side of the RV on the observation deck (∼30 m a. s. l.) above the bridge of the RV (Walton and Thomas, 2018). Observed wind speeds were corrected by Landwehr et al. (2020). To estimate the wind speed at 10 m a. s. l. (U 10 ), measurement height and atmospheric stability were considered using 245 a logarithmic wind speed profile, including the drag coefficient. Quantification of air-flow distortion bias generated by the RV's structures was performed using the data from the operational ERA-interim weather model as a free stream reference. The resulting correction was applied to the observed wind speed, leading to a data set of wind speed at 10 m a. s. l. for the cruise with a five minute time resolution.
The distance between the RV's position and the nearest land for ACE was calculated by Volpi et al. (2020) using the cruise 250 track and additional information on islands in the SO inside a geographic information system application.

Correlation analysis
The collected data were used in a correlation analysis. The goal was to characterise the aerosol population on the SO by finding possible connections between their associated quantities, with the strength/lack of correlation as a first hint for potential sources.
Input variables were the MSA and sodium concentrations from the HV filters, INP concentrations at five temperatures (−8,

255
−12, −16, −20, and −24 • C) from the LV filters, N total , N 500 , particle concentrations of individual PNSD modes (N mode1 , N mode2 , and N mode3 ), N CCN at all measured SS and respective κ values, wind speed at 10 m a. s. l. (U 10 ) and in-water Chla and DMS concentrations. Correlation analysis was performed by calculating Spearman's rank correlation coefficients and associated p values between input variables. As data of diverse temporal resolution were used, the coarsest resolution (24 h, HV filter sampling) was chosen and variables with finer resolution were averaged over 24 hour periods, using arithmetic mean 260 values. For each variable, 79 data points were used for the correlation analysis. This corresponds to the number of HV filters which sampled a sufficient (>100 m 3 ) volume (see subsection 2.3).

Aerosol Particles and Cloud Condensation Nuclei
In Fig. 2a, smoothed PNSD for legs 1-3 of ACE are presented. For the most parts of the cruise, a bi-modal particle number size 265 distribution was present. Potential sources for the pronounced accumulation mode, which is causing the bi-modality, are either entrainment of aerosol particles from the free troposphere (FT) into the MBL or in-cloud processing, according to Hoppel et al. (1986). A general characterisation of the aerosol particles sampled during ACE is given in Schmale et al. (2019), including median values for the diameter of the Hoppel minimum, which are 48, 74, and 68 nm for Leg 1, Leg 2, and Leg 3, respectively (see Fig. 2a).

270
Time series of N total and N CCN (SS) for the ACE cruise are given in Fig. 2b. Here, days for which the average distance to land is lower than 200 km are highlighted with grey shading. Additionally, the starts and ends of the different cruise legs are given as dashed lines. Filtering by stack exhaust contamination caused concurrent data unavailability, while differences in temporal resolution and availability of the instruments create times with no overlap between N total and N CCN (SS). Fig. 2b shows that N CCN at a particular SS varied over 2 orders of magnitude throughout the cruise, e.g., at a SS of 0.2 % (N CCN,0.2 ) 275 from 4 to 309 cm −3 , which is consistent with the frequency distribution in Fig. 4a. In the vicinity of the ports, higher N total and N CCN (SS) are observed compared to the open ocean sections (stack exhaust contamination filtering being performed for both as described in subsection 2.2). This suggests aerosol particle abundance to be influenced by terrestrial and anthropogenic sources and is in line with Schmale et al. (2019) showing pristine conditions during ACE being encountered only south of 55 • S.

280
Periodic differences between N total and N CCN,1.0 throughout the cruise were observed (Fig. 2b). Periods of larger differences coincide with PNSD in Fig. 2a featuring a pronounced Aitken mode (D p = 10-100 nm) with elevated numbers in the size range below 40 nm. During these periods even SS = 1 % was not sufficient to activate the smaller Aitken mode particles.
Consequently, quantities presented later in this manuscript, that are derived from N CCN,1.0 , are representative for the larger Aitken mode particles, as D crit at this SS (∼37 nm, see Tab. S1) still falls within the Aitken mode. during Leg 2. Moreover, the average Hoppel minimum diameter was found to be the largest for Leg 2, when compared to the (relative and absolute) of SSA to CCN during Leg 2, suggests a significant fraction of CCN originating from secondary aerosol production. However, differences in N CCN between legs are within the ranges given by the respective geometric standard deviations. Therefore, the impact of longitudinal differences on CCN abundance is either small against the overall variability of the data, or a variety of effects cancel each other out so that no clear latitudinal trend can be observed. A similar conclusion can be drawn in terms of latitudinal trends, because the majority of the cruise track during Leg 2 was south of 60 • S, compared to legs 1 and 3 being solely north of 60 • S. With this, the CCN concentrations given in Tab. S1 can be considered representative for the whole SO region.
In addition to our data, N CCN from a selection of other studies performed on Antarctica or over the SO are given in Fig. 3a.  For the whole period, a median of ∼230 cm −3 is reported (triangle left in Fig. 3a). This is above our average (mean ± SD) of 189 ± 76 cm −3 for Leg 1 and differences could be due to continental air masses reaching CGBS. Conditions at CGBS are only representative for the SO when the wind direction is between 190 • and 280 • , the so-called "baseline" conditions (Gras and Keywood, 2017). The ambient radon concentration is used as a proxy for terrestrial influence (e.g., McCluskey et al., 2018) and a threshold of 100 mBq m −3 is used in Humphries et al. (2021). By excluding measurements at CGBS that are not from 315 baseline conditions, a median of ∼130 cm −3 (triangle right in Fig. 3) can be found. This is in good agreement with our results for Leg 1 and we conclude that the terrestrial influence on our average values is small. As for ship-based CCN measurements, comparison between our findings and the PEGASO cruise in the SO's Atlantic sector during January-February 2015 (Fossum et al., 2018) can only be done semi-quantitatively, since SS are not identical. Further, a comparison is only reasonable for Leg 3, the part of ACE on the Atlantic sector of the SO. The result of visual interpolation between our N CCN,0.5 and N CCN,1.0 320 for Leg 3 lies in the ranges of 217 ± 31 cm −3 reported for modified Antarctic air encountered during PEGASO (Fig. 3a) and an average of 178 ± 99 cm −3 (mean ± SD). Besides the difference in measurement height (SOCRATES: 50 m a. s. l. until height of inversion; ACE: ∼15 m a. s. l., see section 2), another factor is that measurements are from successive years, with the An overview on the aerosol particle hygroscopicity parameter κ values observed during legs 1-3, individually, is given in  Geometric mean values (and respective geometric standard deviation) of N CCN and D crit , and median values (and respective IQR) of κ for the entire cruise and its three legs are summarized in Tab. S1.
Probability density functions (PDF) of normalized frequencies for N CCN (SS) and κ(SS) during legs 1-3 are given in Fig. 4.

370
PDF of N CCN (Fig. 4a) show mono-modal distributions for all SS, with the PDF maxima shifting towards higher N CCN with increasing SS, e.g., ∼90 cm −3 at 0.15 % to ∼210 cm −3 at 1 %. Comparing the distribution for N CCN,0.2 (blue line in Fig. 4a) with yearly-averaged PDF from measurement sites around the globe in Schmale et al. (2018), our values show lower number concentrations with a PDF maximum at ∼100 cm −3 and share resemblance in terms of number of modes and maximum location with the distribution reported for clean marine conditions (mono-modal, maximum at ∼200 cm −3 ). For the MBL legs 375 of SOCRATES, the PDF for N CCN,0.3 is bi-modal, with peaks at 100 and 150 cm −3 (Sanchez et al., 2021).
A change in distribution shape with increasing SS can be seen for PDF of κ(SS) in Fig. 4b. All five shown distributions have their maximum between 0.4 and 0.6, while a mono-modal distribution is only found for κ 0.15 (red line). PDF for SS of 0.3, 0.5, and 1 % (green, purple, and orange line, respectively) feature a tail towards smaller values of κ, which could be interpreted as an additional mode. This occurrence of small particles (activated at high SS) consisting of mainly organic material forms 380 a strong case for the sampled Aitken-mode CCN originating from secondary organic aerosol formation and growth processes.
The accumulation mode, probed with the measurement at SS = 0.15 %, shows similar κ values as the Aitken mode (Fig. 4b), while its N CCN values are on average over 33 % smaller (Fig. 4a). Additionally, PDF for all SS other than 0.15 % feature a tail towards higher κ values. Such high κ values at high SS seem counter-intuitive and are indicative of highly hygroscopic Aitken mode particles being sampled. A sensitivity study of our methodology with respect to (1) modelling the measurement 385 uncertainty via Monte Carlo simulations (Fig. S3a), (2) consideration of error propagation, and (3) quality of the fitted modes to the PNSD was performed. As κ values were robust against these variations, we conclude that this tail (yet counter-intuitive) is not an artefact that stems from our methodology. However, to avoid speculation on the reason, we take a conservative approach in keeping the focus of the interpretation on the median values presented in Fig. 3b.  (Cornwell et al., 2020) has shown that re-emission of dust particle from sea water into the atmosphere is possible and that the re-emitted particles retained their ability to act as INP. However, quantifying the contribution of this potential source is not possible with our data set.

Ice Nucleating Particles
The PDF of N INP,−24 is mono-modal with a tail towards lower concentrations (Fig. 7a). We attribute this mono-modality  Bigg et al. (1963) and Mossop and Thorndike (1966), but do not reflect the trend found for our two sampling techniques. There could be an averaging effect from the longer sampling interval for the HV filters, when sampling from an unevenly distributed INP population. However, such effects require further investigation. The differences in filter material could be another factor, but are in contrast to Wex et al. (2020) finding good agreement between quartz fibre and poly-carbonate filters for identical sampling intervals. Contamination from ship exhaust should not effect INP analysis results (see Appendix C in Welti et al.,490 2020) as exhaust particles are not ice-active in the investigated temperature range. However, deactivation of some INP due to exhaust contamination can not be ruled out. For completeness, we report N INP for HV sampling in Tab. S4. Due to a higher data coverage (LV: n filter = 253; HV: n filter = 79) allowing for more robust statistics, we decided to focus on LV samples for the in-depth analysis presented in the framework of the paper.

495
Information on the aerosol chemical composition is widely used to infer the origin of the sampled population. The abundance of specific compounds is used as indication of source strength. The inevitable removal of aerosol particles from the atmosphere is taken into account by considering the atmospheric lifetime. To aid the characterisation of CCN and INP sources over the SO, sampled HV filters were analysed regarding the aerosol load and the atmospheric particle mass concentrations of sodium and MSA, two compounds known to be unaffected by stack exhaust. Averaging sodium mass concentrations for the whole cruise gives a median value of 2.8 µg m −3 , with an IQR from 1.8-505 3.9 µg m −3 (Tab. S4). Higher median values for legs 1 and 3 compared to Leg 2 are found, similar to what is observed for PM 10 . This is consistent with Blanchard and Woodcock (1957) showing SSA production to be driven by wave breaking and 0.8 ± 0.8 µg m −3 (mean ± SD). During Leg 2 of ACE, the part of the cruise that has the largest geographical overlap with the region covered during CHINARE, the median sodium mass concentration was 1.8 µg m −3 , i.e., more than two times higher than that observed during CHINARE.
MSA mass concentrations were generally 2 and 1 order of magnitude lower than the ones for PM 10 and sodium, respectively.
Consequently, values are reported in ng m −3 in the following. A median mass concentration of 102 ng m −3 for the entire ACE two decades, the difference of up to 1 order of magnitude might be due to the difference in season, with higher concentrations for ACE due to increased marine biological activity in early fall compared to late fall for Davison et al. (1996). Another factor is the large degree of variability in MSA abundance across the SO, depending on season and location as illustrated in Castebrunet et al. (2009)

Correlation Analysis
The results of a correlation analysis performed with a selection of variables gathered during ACE is given as a Spearman rank 545 correlation matrix in Fig. 9.
With regards to the results of our in situ aerosol particle measurement, N total was found to be correlated with N mode1 (correlation coefficient ρ = 0.9, p < .001) and N CCN,1.0 (ρ = 0.8, p < .001). This mirrors the behaviour these quantities show in Fig. 2a, and is indicative for the importance of Aitken mode particles for the total particle and CCN number concentrations at high SS.

550
Correlations between sodium and mode 3 (ρ = 0.7, p < .001) as well as PM 10 (ρ = 0.7, p < .001) concentrations were found. As sodium is used as a conservative tracer for primary aerosol particles of marine origin (Legrand and Pasteur, 1998), especially sea salt, the correlations suggest that SSA significantly contributes to both, PM 10 and the coarse mode. However, we do not find a significant correlation between wind speed (U 10 ) and sodium mass concentration. Bates et al. (1998) attributed this kind of observation to the fact that the instantaneous wind speed at the RV is not representative for the conditions an air 555 parcel experienced prior to its measurement. It is worth mentioning, that other studies on the SO found correlation between wind speed and sodium concentrations (e.g., Schmale et al., 2013;Yan et al., 2020a;Landwehr et al., 2021). Another factor might be, that the wind speed was averaged over 24 h, in order to match the temporal resolution of the filter sampling. Possible short term effects might be lost due to the averaging process. Note, that the relative variability of the daily U 10 averages is between 0.7 and 6 %.

560
The particle concentration of mode 2 shows a positive correlation (ρ ≈ 0.7) with N CCN at SS ≤ 0.5 %, pointing at the importance of accumulation mode particles for the CCN population at atmospheric relevant SS. No correlation was found between N CCN and mode 3 number concentrations, suggesting little influence of SSA on the CCN population probed with our

SS.
No correlation between N CCN and MSA concentration was found. This seemingly contradicts findings of previous studies 565 (e.g., Ayers and Gras, 1991) and our observations of the highest N CCN (subsection 3.1) and the highest MSA concentration (subsection 3.3) occurring near the coast of Antarctica, and might be a smearing effect due to averaging. However, finding no correlation with our method does not imply that there could not be a connection under specific conditions and shorter time scales.
Furthermore, no correlation between N CCN and in-water Chl-a or DMS concentration could be found, which is in line with 570 Ayers et al. (1997). Considering the long process chain from in-water DMS to particles of CCN size this is not surprising and the argument that conditions at measurement point must not be representative for the conditions encountered by the air-parcel during transport in Bates et al. (1998) are very likely applicable.
Looking at κ values, high correlation between different levels of supersaturation (except SS = 1 %) could be found, mirroring the lack of size-dependent composition presented in subsection 3.1. Further, no correlation between sodium concentration 575 and κ values was found, showing that the chemical information approximated by κ is not connected to the mass-dominated results of the analysis of sodium and MSA. This again supports the observation of SSA particles not significantly contributing to the CCN population, as SSA dominates the sampled particle mass but not the particle number.
No correlations with any other variable was found for the MSA concentration. This includes the absence of the correlation between MSA and in-water DMS concentration. Although MSA is known to form exclusively from oxidation of DMS in the 580 atmosphere (Sorooshian et al., 2007), a direct correlation is not expected. In-water DMS concentrations are not representative of DMS concentrations in the atmosphere (Ayers et al., 1997) and DMS has an atmospheric lifetime of several days over the SO (Chen et al., 2018).
INP concentrations measured at a temperature difference (∆T ) of 4 K showed positive correlation (ρ > 0.6). This is expected, since these concentrations are cumulative along the temperature axis (see Fig. 8) and could indicate a common source This correlation between N INP at −12 and −20 • C points at the importance of long-range transport and mixing influencing the INP population in the same way at both temperatures (Welti et al., 2018). The in-water Chl-a concentrations were also included in the correlation analysis, as it can be used as a proxy for biological activity (e.g., McCluskey et al., 2018

Summary
During the Austral summer of 2016/17, we performed in situ measurements and filter sampling of PM 10 aerosol particles 600 for characterizing the physical and chemical properties of aerosol particles over the Southern Ocean during the Antarctic Circumnavigation Expedition. We focused on the abundance and properties of CCN and INP. A correlation analysis was performed to identify and interpret possible links between different aerosol physico-chemical parameters.
For the in situ measured aerosol particles, bi-modal aerosol particle number size distributions (PNSD) with a distinct Hoppel minimum between 50 and 80 nm were found (Fig. 2a). When the RV was close to continental land-masses, increased total 605 particle (N total ) and CCN number concentrations (N CCN ) were observed (Fig. 2b). The absolute difference between N total and N CCN varied during the cruise and was associated with particle activation in the Aitken mode size range. This indicates an importance of the Aitken mode for the CCN population and cloud-formation. Generally, N CCN spanned 2 orders of magnitude (e.g., at SS = 0.3 % from roughly 3 to 590 cm −3 ), with the respective probability density functions (PDF) sharing resemblance with distributions in Schmale et al. (2018) for clean marine conditions of other locations around the globe. Averages 610 of N CCN per cruise leg (Fig. 3a) showed little difference between the legs and compare well (<30 % percentage difference) with measurements of previous studies in the SO region. Values of the aerosol hygroscopicity parameter κ were found to be in the range between 0.2 and 0.9, corresponding to mixtures with different amounts of organic and inorganic materials. Our κ are about a factor of two lower than what was measured, e.g., over continental Antarctica or modelled for the SO region.
Average values of κ were found to be independent of SS and thus particle size (Fig. 3b), indicating in first approximation an 615 internally mixed CCN population in the Aitken and accumulation modes. The PDF of κ values was found to be mono-modal for SS = 0.15 % (Fig. 4c), while for higher SS tails towards smaller κ values were found, hinting at an increasing amount of organics in the smaller Aitken mode particles. In addition, tails towards higher κ at SS > 0.15 % indicate the occurrence of highly hygroscopic Aitken mode particles. The correlation analysis showed little-to-no connection between the CCN number concentration and quantities from the offline filter analysis, e.g., the concentrations of sodium and MSA (Fig. 9). This is due to the fact that the in situ measured aerosol properties considered here are governed by particle number, while the quantities determined from the filter samples (except for INP) are governed by particle mass. This often implies a focus on different size ranges. However, a connection was found through a positive correlation between total particle number concentration of the coarse mode and sodium mass concentration (Fig. 9). In addition, the absence of correlation between the sodium mass concentration and CCN number concentration clearly implies that SSA is not an important source of CCN. This agrees well 625 with previous findings, e.g., in Schmale et al. (2019).
Analysis of filter-collected atmospheric aerosol samples for N INP yielded temperature dependent concentrations between −4 and −27 • C (Fig. 8). Typically, the N INP from one filter sample increased by 3 orders of magnitude within steps of −10 • C. The results for the analysis of sodium and MSA in the sampled PM 10 show that during ACE we encountered (mass-wise) a marine aerosol environment with typical SSA signals. Sodium concentrations showed a median of 2.8 µg m −3 (Tab. S5). A moderate positive correlation between sodium and PM 10 ( Fig. 9) underlined the importance of SSA for the sampled mass.

640
During ACE, MSA concentrations were found to be highly variable, with a median of 102 ng m −3 (Tab. S5). Values were up to 1 order of magnitude higher than in comparable studies and seasonal variation seems to be one reason. The location of peak MSA concentrations near the sea ice edge is consistent with other studies. Similar patterns in the occurrence of maximum MSA concentrations and the hypothesised κ value for MSA were found. However, a clear connection between MSA and CCN concentrations or κ values did not show in our correlation analysis. With our data covering all sectors of the SO and the rich 645 variety of atmospheric conditions encountered during the cruise, we conclude that such a connection might only be event-based.
The presented data set gives an unique, circum-Antarctic view on CCN and INP abundance, their properties and indications towards aerosol particle origin. Our data give insights into the conditions on the SO regarding cloud-relevant aerosol particles, compare well with previous studies and found already use in climate modelling (Regayre et al., 2020) and remote sensing applications (Efraim et al., 2020).