Hourly averaged carbon dioxide (CO₂) dry-air mole fractions used in this study result from the merging of several cross-evaluated measurements with cavity ring-down spectroscopes during MOSAiC. Measurements were performed in the University of Colorado (CU) container using a commercial Picarro instrument (model G2311-f); on the sea ice at Met City (a few hundred meters away from Polarstern), also using Picarro model G2311-f; and in the Swiss container using Picarro model G2401. Additional discrete whole-air samples were collected for post-cruise analysis at the National Oceanic and Atmospheric Administration (NOAA) Global Monitoring Laboratory (GML) and included in the data-merging procedure. Details regarding the instruments' operation, calibration, data processing, and the creation of the merged dataset can be found in Angot et al. (2022b​​​​​​​).

Measurements of CCN number concentrations were performed during MOSAiC using a cloud condensation nuclei counter (CCNC) model CCN-100 (Droplet Measurements Technologies, Boulder, USA) colocated with the Aethalometer and the AMS in the Swiss container. The supersaturation (SS) in the instrument's chamber was set to 0.15 %, 0.2 %, 0.3 %, 0.5 %, and 1 % SS throughout 1 h cycles (Heutte et al., 2023b). CCN number concentrations were averaged, for each SS level, to a 10 min time resolution (resulting in one 10 min average data point for each SS level per hour).

The aerosol total light scattering coefficients at the blue (450 nm), green (550 nm), and red (700 nm) wavelengths were measured during MOSAiC using an integrating nephelometer (TSI; model 3563) located in the AOS container. Scattering coefficients were measured at 1 min time resolution and corrected to account for the size-dependent truncation (incomplete collection) of strongly forward- or backward-scattered light (Koontz et al., 2022). Using an impactor at the inlet of the external sampling system, the aerodynamic diameter cutoff of sampled particles was alternated between 1 and 10 µm. In this study, we used the submicron measurements averaged at 10 min time resolution.

During MOSAiC, the particle number flux of airborne snow particles was measured, at 1 min time resolution, using two open-path snow particle counters (SPC-95; Niigata Electric Co., Ltd), and used to compute the snowdrift density (Gong et al., 2023). Blowing snow periods were identified (Gong et al., 2023) as times when airborne snow particles were detected and when the wind speed measured at 10 m above the snow surface exceeded a critical value, which was empirically estimated from the temperature-dependent parameterization proposed by Li and Pomeroy (1997). The two SPCs, Unit 1104 and Unit 1206, were located at 0.08 and 10 m above the snow surface, respectively. In this study, the SPC at 0.08 m was used to report the snowdrift density and derive the blowing snow flag, except for October and November 2019, where the instrument was not operational, and data from the SPC at 10 m were used instead.

In our study, we use a published dataset of lead fractions (von Albedyll et al., 2024). In brief, lead fractions, within a 50 km radius from Polarstern, were derived from a divergence-based method using satellite synthetic-aperture radar (SAR) data with a spatial resolution of 700 m. The daily available divergence and convergence fields were accumulated for up to 10 subsequent dates to account for leads continuously opening or closing. In this work, we used the divergence-derived lead fractions with no accumulation (LF_{no accu, div}), which represent newly opened leads, as well as the lead fractions accumulated five times (LF_{5× accu, div}), which account for leads opening, closing, or staying open within a 5 d period.

In early summer (1 June–10 July 2020; during MOSAiC) and late summer (1 August–15 September 2018; during MOCCHA), organics had the largest mass contribution to the total PM₁ (62 %–63 %; Fig. 2b and e), followed by SO42- (28 %–32 %), while eBC contributed 2 %–3 %. Nitrate accounted for 8 % of the total PM₁ mass in late summer during MOCCHA and was mostly below the detection limit (Table S1) in early summer during MOSAiC. However, it should be noted that NO3- measurements during MOCCHA were likely overestimated due to interferences between the NO⁺ and C¹⁸O⁺ fragments at m/z 30 in the AMS. Hence, NO3- will not be further discussed in this section. Chloride was below the detection limit during MOCCHA and contributed 1 % of the PM₁ mass in June during MOSAiC. Finally, NH4+ was either below the detection limit (during MOCCHA) or excluded due to instrumental issues (in June–July during MOSAiC; see Sect. 2.2). Overall, we observed very low mass concentrations for the different species at this time of the year, which is typical and characteristic of the high Arctic during this season (Leaitch et al., 2018; Massling et al., 2015; Schmale et al., 2022; Ström et al., 2003). This is likely the combined result of (1) limited long-range transport of aerosols from lower latitudes, as the extent of the Arctic dome is small (Stohl, 2006), and (2) the efficient wet and dry removal of locally emitted and transported aerosols, as previously observed and modeled (Browse et al., 2012; Freud et al., 2017; Pernov et al., 2022). In June, during MOSAiC, median (25th quantile, 75th quantile) concentrations of Org and SO42- were 0.14 (0.09 and 0.25) and 0.057 (0.012 and 0.122) µg m⁻³, respectively. These values are larger than the ones found in August during MOCCHA, namely 0.06 (0.05 and 0.08) and 0.016 (0.006 and 0.040) µg m⁻³ for Org and SO42-, respectively. The lower concentrations during MOCCHA can likely be explained by the position of Oden high up in the pack ice (> 88° N; see Fig. 1), far away from most open-ocean marine and terrestrial sources. Chang et al. (2011) reported similar mass concentrations of the organics (median Org = 0.055 µg m⁻³) at such high latitudes in August–September 2008, although with a higher sulfate fraction to the total mass of 45 % (median SO42- = 0.051 µg m⁻³). The very low mass concentrations for the different species in summer are also reflected in the low Vtot values (median = 0.131 µm³ cm⁻³ in June and 0.049 µm³ cm⁻³ in August). The large variability in Vtot in July (interquartile range = 0.719 µm³ cm⁻³) could be indicative of intermittent events of transport or the local release of organic material from melt ponds, the marginal ice zone, or nearby coastal and open-ocean areas (Chang et al., 2011). In general, it is likely that organic aerosols at this time of the year are dominated by local/regional natural marine and terrestrial biogenic sources (e.g., Baccarini et al., 2020; Chang et al., 2011; Hamacher-Barth et al., 2016; Heintzenberg et al., 2015; Siegel et al., 2021), with relevance towards cloud formation (Bulatovic et al., 2021; Duplessis et al., 2024; Karlsson et al., 2022). With the present observations, we further emphasize the major contribution of organics, which are likely naturally sourced, for the central Arctic submicron aerosol budget in summer. A follow-up source apportionment study will elucidate the sources associated with organic aerosols (including MSA) in the summertime central Arctic during MOSAiC.

October marked the beginning of the dark season in the central Arctic, associated with a decrease in surface temperatures (Shupe et al., 2022). In the transition from summer to autumn, we observed a drastic change in the aerosol chemical composition, whereby SO42- became the dominant measured species by mass (47 %, Fig. 2c), followed by Org (31 %), Chl (8 %), eBC (7 %), and NO3- (6 %). NH4+ was mostly below the detection limit. The combined increase in the fraction of SO42- and eBC, and the decrease in the fraction of Org, compared to summer is indicative of changes in the aerosol sources, sinks, and processing. The ongoing ocean freeze-up, coupled with the dark conditions, gradually decreases the influence of local marine aerosol sources (Leck and Persson, 1996; Moschos et al., 2022b; Schmale et al., 2022), while long-range transport of anthropogenic pollutants from lower latitudes occurs more readily towards the winter (Boyer et al., 2023; Moschos et al., 2022b; Quinn et al., 2009). During MOSAiC, November and December experienced several storms (see the annual wind speed measurements from the 2D ultrasonic anemometer on board Polarstern (Schmithüsen, 2021a, b, c, d, e) in Fig. S2), which have been shown to greatly increase the number of fine-sea-salt (NaCl) aerosols associated with the sublimation of salty blowing snow and/or sea spray aerosol (SSA) emissions from open leads (Chen et al., 2022; Gong et al., 2023). The elevated chloride fraction seen here may be related to NaCl, but the interpretation remains difficult because of the limited ability of the AMS to measure it (see Sect. 2.2). Another possibility is non-refractory chloride that partitioned into the particles (Hara et al., 2002). Using instead the NaCl⁺ fragment at m/z 58 (Ovadnevaite et al., 2012), we discuss the contribution of sea salt in the aerosol budget in autumn in Sect. 3.2 and 3.3. Similarly, for NO3-, part of the signal could be associated with sodium nitrate following the chloride displacement in aged sea salt particles (Gard et al., 1998), while long-range transport of NO3- in haze particles is also likely playing a role (Quinn et al., 2007). Due to the logistical complexities associated with in situ measurements in the central Arctic Ocean at this time of the year, reports of aerosol chemical composition are scarce, which challenges the comparison of our dataset with others. However, observations of aerosol number concentrations and size distributions from lower-latitude land-based stations commonly reported October as the yearly minimum in PNCs as a result of enhanced wet removal and limited long-range transport of pollutants (e.g., Croft et al., 2016b; Freud et al., 2017; Nguyen et al., 2016; Pernov et al., 2022; Tunved et al., 2013). In line with these studies, the yearly total aerosol number concentration reached a minimum in October during MOSAiC (Boyer et al., 2023) and the same for Vtot (median = 0.067 µm³ cm⁻³, Fig. 3a). In November, the median mass concentrations of Org and SO42- reached their seasonal maximum (i.e., for autumn), with medians (25th quantile and 75th quantile) of 0.12 (0.09 and 0.17) and 0.136 (0.094 and 0.223) µg m⁻³, respectively. Although limited, we show in Sect. 3.2.2, based on our high-time-resolution dataset, that the long-range transport of anthropogenic pollutants is an important and climate-relevant contribution to the central Arctic aerosol budget in autumn and likely represents the start of the well-known haze period.

Wintertime during MOSAiC (here defined as January and February 2020) marked the peak of the year 2020 haze season, with the highest yearly median aerosol total volume and eBC mass concentration observed in January (1.008 µm³ cm⁻³ and 0.11 µg m⁻³, respectively). This unusually early timing for the occurrence and intensity of Arctic haze has been discussed by Boyer et al. (2023) and was attributed to a record-breaking positive phase of the Arctic Oscillation (AO) between January and March 2020 (Lawrence et al., 2020), leading to enhanced air mass transport from lower latitudes to the central Arctic. The authors demonstrated the importance of Russia/Siberia as a pollution source for eBC and accumulation mode aerosol number concentrations (N100–500) during these 2 months, with eBC and N100–500 reaching their annual maxima in January. As already mentioned, the AMS was not measuring at this time of the year due to instrument malfunctions. Hence, it was not possible to derive any chemical composition information except for BC; however, given the similar fractional mass chemical composition for the neighboring months (see Fig. 2c–d), it is likely that SO42- was the dominant non-refractory species by mass. Nonetheless, due to a low abundance of photochemically produced oxidants in the dark winter conditions and the limited cloud liquid water for aqueous-phase reactions in the high Arctic, we could also expect that primary emissions such as BC or primary anthropogenic organics (Moschos et al., 2022b) dominated over secondary processes that would produce particulate sulfate (Schmale et al., 2022) and secondary organics. The two latter species have their peak contribution in March and April across Arctic observatories (Moschos et al., 2022a; Schmale et al., 2022). Wintertime oxidation pathways could have, however, still resulted in SO42- being a dominant species in the dark months of January and February in the context of anomalously high positive AO. Such pathways involve, for instance, the metal-catalyzed in-cloud oxidation of SO₂ by O₂ (Alexander et al., 2009; McCabe et al., 2006) and the poleward transport of SO42- formed at lower latitudes, where sunlight is available for photo-oxidation. This is also supported by recent findings from Boyer et al. (2024), who found a close agreement between measured high-SO₂ mixing ratios and simulated SO42- mass concentrations using the ECLIPSE v6b emission inventory coupled with back-trajectories in January and February 2020 at the location of Polarstern. Primary SO42- (i.e., emitted as fully oxidized from coal- and oil-burning stacks), which was found by Moon et al. (2024) to be the dominant source of SO42- in a polluted city of the Alaskan sub-Arctic, could also have contributed to the central Arctic winter SO42- budget, since the process likely applies to other locations around the Arctic, including Siberia. Without additional observational evidence, this will not be discussed further, and the focus will be turned towards the spring Arctic haze chemical characterization (see Sect. 3.1.4).

The spring season (March–May) was characterized by elevated PM₁ concentrations, where SO42- contributed by 50 % to the measured mass, followed by Org (36 %), eBC (6 %), NH4+ (3 %), NO3- (3 %), and Chl (2 %). This pattern is representative of the well-studied Arctic haze phenomenon (Nielsen et al., 2019; Quinn et al., 2007). A number of high mass concentration events were also observed, such as on 15 March when PM₁ mass concentration neared 2 µg m⁻³ and during two intense episodes of warm and moist air mass intrusions from northern Eurasia on 15 and 16 April, when pollution levels ([PM₁] ≥ 4 µg m⁻³) became comparable to central European urban pollution levels (Dada et al., 2022a). Dada et al. (2022a) showed that the sudden direct transport of pollution to the central Arctic can have important impacts on climate-relevant properties (i.e., acidity, oxidation state, and, hence, hygroscopicity). The highest monthly median mass concentrations in spring were found in April, with medians (25th quantile and 75th quantile) of 0.35 (0.31 and 045) and 0.50 (0.41 and 0.63) µg m⁻³ for Org and SO42-, respectively. At this time of the year, atmospheric conditions favored transport from lower latitudes compared to summer (Bozem et al., 2019), and Boyer et al. (2023) found that the surface aerosol population was largely influenced by transport from Siberia in spring during MOSAiC. The prevalence of SO42- observed here corroborates that Russia/Siberia is an important source of pollution to the central Arctic haze burden (Hirdman et al., 2010; Petäjä et al., 2020) as industrial activities in these regions (mainly metal smelters) are known to be important sources of atmospheric sulfur (Sipilä et al., 2021). We also measured relatively low NH4+ concentrations at the surface. Observational and modeling studies have shown strong vertical gradient of NH4+ / SO42- ratio in the springtime Arctic, with higher concentrations of NH4+ in the upper (free) troposphere than in the boundary layer resulting from a stronger contribution of east Asian anthropogenic (agricultural) NH4+ emissions at higher altitudes (Fisher et al., 2011; Willis et al., 2019). Together, these observations suggest that submicron aerosols measured in the springtime at the surface are very acidic, with potential implications for the partitioning of gaseous organic acids to particle phase, as observed for MSA during MOSAiC (Dada et al., 2022a). In May, SO42- concentrations remained high, especially at the beginning of the month, when large-scale vertical mixing associated with the collapse of the polar vortex could have introduced large quantities of aged particles into the troposphere from aloft (Ansmann et al., 2023). Natural sources of sulfur species from DMS oxidation had a growing contribution to gaseous sulfur compounds (MSA) during this month and towards summer (Boyer et al., 2024), with the initiation of the summer sea ice melt in late May (Shupe et al., 2022). The detailed aerosol chemical and geographical sources during haze, especially those of organics, will be presented in a follow-up source apportionment study.

To understand potential spatiotemporal variability, we compared our yearly chemical composition observations with measurements from six land-based stations around the Arctic (Fig. 4), including Alert, Canada (ALT); Baranova, Russia (BAR); Gruvebadet, Svalbard/Norway (GRU); Pallas, Finland (PAL); Villum, Greenland (VRS); and Zeppelin, Svalbard/Norway (ZEP), with measurements from 2015 to 2019, depending on the station (see Fig. 4 caption for details). Further information on the location of these stations, their sampling methods, and a description of their yearly cycles of chemical composition were presented and discussed by Moschos et al. (2022a). The results of this comparison need nonetheless to be interpreted with caution for several reasons: (1) the sampling method differed substantially as offline analysis of weekly to bi-weekly filter samples by ion chromatography for inorganic ions, and an OC / EC sunset analyzer for organics was employed for the pan-Arctic datasets, while an AMS was used during MOCCHA and MOSAiC; (2) the sampling sites for the pan-Arctic datasets are located at lower latitudes than those at which MOSAiC and MOCCHA took place; (3) the cutoff size for the sampling inlets was different, with 10 µm (PM₁₀) for the filter samples and 1 µm (PM₁) for the AMS; and (4), finally, for the anomaly calculations, the yearly mean value for MOCCHA and MOSAiC could be biased by the fact that data were missing in December, January, February, and July. Despite this, intercomparisons of the trends in the chemical species were still possible.

Sulfate, organics, and nitrate were, on average, within the same range of absolute mass concentrations in the central Arctic as at the land-based stations, although geographical variability was evident (e.g., high-SO42- concentration at ALT in spring or high-Org concentrations at PAL in summer) and expected (Schmale et al., 2021). These similarities are remarkable when considering the differences in the sampling conditions described above between the two datasets. An exception was for August and September during MOCCHA, where SO42- and Org (median SO42- = 0.016 µg m⁻³ and median Org = 0.050 µg m⁻³) were consistently lower than at the various land-based stations (median SO42- = 0.218 µg m⁻³ and median Org = 0.293 µg m⁻³ for the August–September station average). As discussed in Sect. 3.1.1, Oden was close to the North Pole and deep in the pack ice, which partly isolated it from most remote natural and anthropogenic sources. SO42- during springtime also exhibited lower concentrations in the central Arctic during MOSAiC (median = 0.446 µg m⁻³, for March–May) compared to other land-based stations (median = 0.697 µg m⁻³). Interestingly, at the same time, organics levels (median = 0.329 µg m⁻³) were relatively similar to the station measurements (median = 0.334 µg m⁻³). This could possibly suggest that SO42- and Org had different emission intensities or different sources. Alternatively, the fraction of sulfate in the coarse mode (PM₁₀) could have been larger than that of organics, which could explain the difference between the PM₁ MOSAiC observations and the PM₁₀ pan-Arctic observations. NO3- was generally low in the central Arctic (yearly median = 0.017 µg m⁻³) and at the stations (yearly median = 0.021 µg m⁻³). In the case of MOCCHA and MOSAiC, this was potentially furthered by the PM₁ limitation, since a large fraction of NO3- is expected to be found in supermicron-sized, and more alkaline, sea salt particles (Cavalli et al., 2004; Fenger et al., 2013; Mukherjee et al., 2021; Ricard et al., 2002; Saltzman, 2009).

A striking difference was observed for ammonium, which was found to be consistently less abundant throughout the year in the central Arctic (yearly median = 0.001 µg m⁻³) compared to the land-based stations (yearly median = 0.043 µg m⁻³), especially in spring. This results in generally more acidic aerosols in the central Arctic. Differences could be explained by a stronger contribution of ammonia emissions at Arctic coastal sites from migratory seabird colonies (Croft et al., 2016a), as well as different spatiotemporal NH4+ contributions from open biomass burning events in the Arctic or sub-Arctic regions (Gramlich et al., 2024). In light of the decreasing sulfate concentrations in the Arctic (Schmale et al., 2022), efforts should be maintained to rigorously monitor aerosol chemical composition in the future, as there is a range of aerosol physicochemical processes that depends on the particles' acidity (Pye et al., 2020); for example, the partitioning of nitrate into the particle phase tends to increase as the sulfate-to-ammonium ratio decreases (Sharma et al., 2019).

Regarding the seasonality of the anomaly values (Fig. 4b), the haze signal peaking in March/April appeared to be similar for all species between MOSAiC and the pan-Arctic station averages. As stated above, the summer peak for organics was not observed during MOCCHA, resulting in a lower summer anomaly. This comparative study shows that long-term observations at Arctic land-based stations are relevant to the central Arctic seasonal cycle of chemical composition and mass loading. Differences are nonetheless noticeable, in particular for ammonium, which seems to be far less abundant in the central Arctic throughout the year, as well as sulfate and organics in summer.

During the first storm, the wind speed, measured on board Polarstern, started to increase on 10 November, reaching a maximum on 11 November, with values above 16 m s⁻¹. Blowing or drifting snow was detected (see Sect. 2.4.4) without interruption between 11 November at 01:30 and 12 November at 12:00. Compared to the background period prior to the event from 12:00 to 22:00 on 10 November, a strong increase in N>1000nm was observed (by a factor of ∼ 54 from 0.2 to 10.7 cm⁻³), following, with a 3 h lag, the increase in wind speed (3 h lag Pearson correlation (ρpearson) = 0.87; p value < 0.001). Similarly, the NaCl signal correlated greatly with N>1000nm (ρpearson = 0.89; p value < 0.001) and increased by a factor of ∼ 87 during the storm (from 0.003 to 0.26 a.u.). Since supermicron particles are mainly related to primary particles formed by mechanical processes (Seinfeld and Pandis, 2016), in this case wind-generated, the comparison of NaCl signal was made with N>1000nm rather than with submicron PNCs (N10–500 nm) from the SMPS, which would be influenced by other sources such as long-range transport. Nonetheless, we show (in Fig. S4a) a comparison between N>1000nm and N10–500 nm  during the storm, where the two were highly covariant, especially during the blowing snow episode (ρpearson = 0.99; p value < 0.001). Despite the PM₁ limitation of the AMS, the fact that the submicron NaCl signal correlated with N>1000nm is an indication that sea salt was likely present in the blowing snow, as expected (Frey et al., 2020; Gong et al., 2023). It is also likely that a fraction of the observed increase in submicron NaCl signal originated from wind-driven SSA emissions from neighboring open leads in the sea ice, as has been observed elsewhere (e.g., Chen et al., 2022; Kirpes et al., 2019; Myers et al., 2021; Nilsson et al., 2001; Radke et al., 1976), especially since storms are associated with the mechanical deformation of the sea ice and leads opening (von Albedyll et al., 2024). However, as shown in Fig. 5a, the lead fraction (spatial resolution of 700 m) within a 50 km radius of Polarstern was less than 1 % during the storm. Hence, comparing the relative surface area of open leads to that of sea ice covered by salty snow (i.e., well above 95 %), submicron NaCl emissions from salty blowing snow can conceivably dominate over SSA emissions from leads. A recent modeling study suggested an anti-phased seasonal contribution of leads and blowing snow to sea spray fluxes in the high Arctic, with leads being the dominant source of sea salt in terms of mass in summertime and blowing snow being dominant in winter (Lapere et al., 2024). Furthermore, it cannot be entirely excluded that the observed wind-driven increase in N>1000nm and submicron NaCl came from longer-range-transported SSA from the ice-free Arctic Ocean. As a sea salt source apportionment is impossible here, the increase in its signal has to be seen as a mixed contribution from various wind-dependent emission sources, where blowing snow may be the dominant one. The exact strengths of these different SSA sources remain an open research question and cannot be fully answered here. Hence, for any further references to blowing snow, we implicitly include wind-generated SSA as a potential additional contribution to our observations.

The wind-driven increase in N>1000nm and NaCl mass concentrations resulted in a proportional increase in the CCN number concentrations (shown for SS levels from 0.15 % to 1 % in Fig. 5g). We found correlations (ρpearson) between the NaCl signal and CCN number concentrations between 0.84 and 0.88, depending on the SS level (all p values < 0.001), during blowing snow. Compared to the background period, CCN number concentrations during the storm peak increased by factors of ∼ 4 (from 27.0 to 119.3 cm⁻³), ∼ 5 (from 30.2 to 144.5 cm⁻³), ∼ 5 (from 32.1 to 161.3 cm⁻³), ∼ 6 (from 32.4 to 186.0 cm⁻³), and ∼ 7 (from 33.1 to 228.3 cm⁻³) at SS levels of 0.15 %, 0.2 %, 0.3 %, 0.5 %, and 1 %, respectively. The larger increase for higher SS levels is indicative of the presence of Aitken mode particles (as seen in Fig. S4a, with the stronger increase in 10-80 nm particles compared to the 80–200 or 200–500 nm ones) generated from blowing snow and SSA, which are only activated in the instrument when the SS is high enough to overcome the high curvature of these small particles (Kelvin effect). In ambient autumn conditions (i.e., not in the artificial conditions of the CCNC), high values of maximum cloud supersaturation (> 1 %) are likely to happen (Duplessis et al., 2024; Motos et al., 2023), making the Aitken mode fraction of blowing snow-related particles climate-relevant. The strong enhancement of CCN number concentrations from fine sea salt particles associated with blowing snow has been shown by Gong et al. (2023) for several blowing snow events during MOSAiC in autumn and winter. The authors further estimated, from model simulations including sea salt aerosol generation from blowing snow, that the increase in CCN number concentrations associated with blowing snow led to an increase in the downwelling longwave radiation of about +2.3 W m⁻² under cloudy-sky conditions from November to April. We also observed a blowing-snow-related increase in the total submicron aerosol light scattering coefficient (shown for the blue wavelength in Fig. 5g), tightly following the NaCl signal time series (ρpearson = 0.90; p value < 0.001). Compared to 10 November background, the scattering coefficient increased by a factor of ∼ 21 at the storm's peak (from 1.0 to 21.0 Mm⁻¹). The production of wavelength-dependent scattering particles during blowing snow episodes would be specifically relevant for radiative forcing at lower latitudes of the central Arctic and other times of the year, where sunlight is present (Bergner et al., 2025).

Overall, the same strong relations between wind speed, N>1000nm, NaCl signal, scattering coefficient, and CCN number concentrations were observed for the second storm case (Fig. 5b, d, f, h), where blowing snow was identified from 23 November at 10:00 to 25 November at 01:00, with an intermittent break in the storm on 24 November from 11:30 to 13:30. Defining the background period from 01:00 to 10:00 on 23 November (i.e., just before the blowing snow event) and the storm's peak from 01:00 to 03:00 on 24 November, we find an increase in the NaCl signal by a factor of ∼ 2 (from 0.14 to 0.25), for N>1000nm by a factor of ∼ 4 (from 2.4 to 8.5 cm⁻³), for the scattering coefficient by a factor of ∼ 2 (from 10.7 to 19.4 Mm⁻¹), and for the CCN number concentration by a factor of ∼ 2 at all SS levels (e.g., from 26.9 to 203.3 cm⁻³ at 1 % SS). The background concentration seemed, nonetheless, to be elevated already before the blowing snow event. If we consider the background period from 15:00 to 22:00 on 25 November (i.e., after the blowing snow event, when the influence of wind speed on the considered variables seemed minimized), we found relative increases by factors of ∼ 25, 11, and 15 for the NaCl signal, N>1000nm, and the scattering coefficient, respectively, and between ∼ 5 and ∼ 8 for CCN number concentrations (the increase being larger at higher SS levels). Although the relative increases differed between the two storms due to the different background conditions, the absolute values reached during the storms were very similar. In agreement with Gong et al. (2023), the large deviations from the relatively pristine background suggest that blowing snow episodes are an important, but intermittent, source of scattering particles and CCN in autumn in the central Arctic. As discussed by Bergner et al. (2025), the vertical extent of the blowing snow layer made these particles directly relevant at cloud level during MOSAiC. Further analysis is needed to better quantify these impacts. In spring, we observed similar relative wind-dependent increases in the variables discussed above, although with a smaller magnitude (Sect. S3.1.1).

Another major observation during both storms was the strong correlation between NaCl and eBC during blowing snow (ρpearson = 0.74, with p value < 0.001, and ρpearson = 0.59, with p value < 0.001, for the first and second blowing snow events, respectively). This indicates that eBC was possibly contained in the sublimated particles from blowing snow. It should be noted here that the eBC measurements could be slightly overestimated due to the enhancement of light absorption in the filter matrix under the presence of strongly scattering particles (associated with a high single-scattering albedo (SSAlb); Drinovec et al., 2022). However, during both storms, the SSAlb was below 0.94 (not shown), which is below the 0.99 threshold where strong bias emerges, as experimentally determined by Drinovec et al. (2022). Compared to background conditions, the mass concentration of eBC increased by factors of ∼ 12 (from 0.008 to 0.092 µg m⁻³) and ∼ 4 (from 0.016 to 0.066 µg m⁻³) during the first and second storms, respectively, reaching levels comparable to Arctic haze conditions (see Sect. 3.1.3–3.1.4 and Fig. 3a). To our knowledge, this is the first time that such an observation has been made. The source of deposited eBC on the snowpack is uncertain but could be explained by one, or a combination, of the following two hypotheses. On the one hand, eBC could have been transported long range from lower latitudes and subsequently dry- or wet-deposited on the snowpack. On the other hand, it is also conceivable that eBC on the snowpack originated from Polarstern's stack indirectly. Continuous in situ observations of eBC in snow, as well as measurements in collected snow, would be needed to further examine the hypotheses presented above. In any case, the re-emission of previously deposited eBC could represent an important and overlooked source of atmospheric eBC in the central Arctic during a period when long-range transport is still limited by the extent of the polar dome. Due to its hydrophobic properties, eBC could influence the CCN activation potential of the sublimated blowing snow particles, depending on the mixing state of the particles (Motos et al., 2019; Zieger et al., 2023). Additionally, eBC contributes to atmospheric warming and stratification through the absorption of incoming shortwave radiation (Flanner, 2013). The latter effect is irrelevant during the dark autumn months but could become important with the return of solar radiation in spring and at lower latitudes of the Arctic, where sunlight is present for longer during autumn. In spring, we also found a storm peak increase in eBC by a factor of ∼ 2, but the data availability was insufficient to draw robust conclusions on the source of the eBC during the storm (see Sect. S3.1.1). Overall, future studies and observational campaigns should focus on the characterization of this process, especially in the likely scenario where shipping becomes more important in the Arctic (Gilgen et al., 2018; Smith and Stephenson, 2013) and where this potential eBC-cycling process becomes increasingly relevant.

Around 11 November, a shift from anomalously low to high surface temperature, associated with two consecutive cyclones, triggered the storms presented above (Rinke et al., 2021). These synoptic-scale events were associated with air mass transport from lower latitudes, especially from northern Siberia (see Fig. S6). During the first storm, the gradual increase in CO₂ on 11 November, peaking on 13 November (an increase of ∼ 6 ppm), was evidence of the air mass change associated with the cyclonic conditions. Likewise, for the second storm, CO₂ started increasing on 24 November and peaked on 25 November (an increase of ∼ 4 ppm). For both cyclones, the perturbed CO₂ signal was highly correlated with that of SO42- (ρpearson = 0.69, with p value < 0.001, and ρpearson = 0.89, with p value < 0.001, for the first and second storms, respectively), following a distinct temporal evolution from that of the wind-speed-related variables discussed before. This decoupling is evident during the first storm, on 13 November, when SO42- and CO₂ peaked when the wind speed was continuously decreasing and N>1000nm and NaCl signals were low. In contrast with the blowing snow period when N>1000nm and N10–500 nm were highly correlated, the correlations dropped to ρpearson = 0.42 and ρpearson = 0.40 (p values < 0.001) outside the blowing snow events for the first and second storms, respectively (see Fig. S4a, b), highlighting that the sources of these particles were different. The peak SO42- during the first (∼ 0.32 µg m⁻³) and second (∼ 0.44 µg m⁻³) storms was, respectively, ∼ 2 and ∼ 3 times larger than the monthly SO42- median concentration in November (see Sect. 3.1.2). Organics behave similarly to SO42-, reaching about 0.26 µg m⁻³ during both storms, or ∼ 2 times that of the November median Org concentration. The relative abundance of SO42-, the increase in CO₂, and the related emission source region (i.e., Siberia) indicate that the pollution brought to the central Arctic under these cyclonic conditions was anthropogenic in origin. In spring, we observed an increase in SO42- and Org during a storm on 15 March, associated with air masses traveling from eastern Siberia (Sect. S3.1.2). The influence from long-range transport during spring storms was, however, partly masked by the high haze background concentrations.

We also observed a temporal co-variation between the mass concentrations of SO42- and Org and the number concentrations of CCN, particularly when the influence of blowing snow was ruled out (i.e., outside the blowing snow flag). As such, the increase in CCN number concentrations on 13 November, reaching 87.2, 95.4, 101.9, 107.7, and 119.6 cm⁻³ at the respective SS levels of 0.15 %, 0.2 %, 0.3 %, 0.5 %, and 1 %, was likely related to the increase in SO42- and Org mass concentrations from long-range transport. The smaller spread in CCN number concentrations with increasing SS compared to that of the blowing snow-related increase, as discussed before, indicates that the size distributions of the long-range-transported material contained fewer Aitken mode particles, as seen in Fig. S4a. For the second storm, the sharp increase in the CO₂ mixing ratio (and SO42- and Org mass concentrations) on 27 November, was also associated with an increase in CCN number concentration at all SS levels, although with a larger spread. The larger CCN concentration with increasing SS (∼ 95 cm⁻³ at 1 % SS versus 45 cm⁻³ at 0.15 % SS) could here be explained by the fact that organics became the dominant species and that because of their lower hygroscopicity compared to SO42- (Siegel et al., 2022) they required higher SS levels for droplet activation. Alternatively, a larger fraction of Aitken mode particles could also explain this behavior. However, as seen in Fig. S4b, PNCs in the size range 80–200 nm dominated over Aitken mode particles in the size range 10–80 nm.

Overall, under these cyclonic conditions, CCN number concentrations are influenced by both wind-driven local aerosol production from blowing snow and SSA, as well as from long-range-transported aerosols. Yet, the latter process plays a smaller role. Together, these large increases in CCN number concentration can increase cloud emissivity and longevity and could be in direct relation with the measured anomalously high downward longwave radiation in November during MOSAiC (Rinke et al., 2021).

We found that local wind-dependent aerosol generation and long-range transport were important sources of aerosols during the autumn and spring storms during MOSAiC. We extended this approach to the entire periods of October to November 2019 and March to April 2020 by clustering the PNSDs into distinct clusters (see Sect. S4.1 for a description). Figure 6a shows the resulting median size distributions in October–November associated with the clusters, where each SMPS time step was uniquely assigned to time periods classified by blowing-snow-related (BLSN), long-range-transport-related (LRT), long-range-transport-related to larger but fewer particles (LRT-aged), or low-concentration background conditions (BG). For March–April, the PNSDs were clustered into BLSN, haze-related (Haze), or bimodal haze-related (Haze bimodal), as shown in Fig. 6b. For a direct comparison of the shape of the clustered PNSDs, the normalized size distributions clusters are provided in Fig. S12. The fitting modal parameters (location and amplitude of the modes) are given in Table S2. Median values (and 25th–75th quantiles) of N>1000nm, N10–500 nm, SO42-, eBC, NaCl, and CCN concentrations associated with each cluster are given in Table 1. Importantly, the k-means clustering algorithm is a statistical dimensionality reduction method and cannot be used to separate the contribution of various aerosol sources to each single PNSD measured. In other words, the names given to the clusters do not indicate that the PNSDs included in each cluster are the result of a single contributing emission source, as multiple sources contribute to every aerosol size distribution. Names hence indicate the likely dominating process and source.

In October–November, the BG size distribution was characterized by very low SO42- (median = 0.078 µg m⁻³) and eBC (median = 3 ng m⁻³) mass concentrations and by low number concentrations of sub- and supermicron particles (median N10–500 nm = 15.04 cm⁻³; median N>1000nm = 0.12 cm⁻³), with a weak (fitted) Aitken mode at 38 ± 2 nm and a dominant accumulation mode at 148 ± 1 nm. Overall, a quarter (25.2 %) of all available SMPS PNSDs during this period were in the BG cluster, mostly in October (see Fig. S9). In contrast, the BLSN cluster was associated with high concentrations of submicron particles (median N10–500 nm = 179.20 cm⁻³), with strong contributions of Aitken mode particles (mode at 36 ± 1 nm) and, predominantly, accumulation mode particles (mode at 165 ± 2 nm). Spikes in N10–500 nm, contributing to the high 75th quantile value (255.6 cm⁻³), were observed when blowing/drifting snow was detected (see Fig. S9b), making blowing snow a potential important contribution to the BLSN PNSDs, in line with findings from Gong et al. (2023). The accumulation mode particles could also be related to the long-range-transported particles, which co-occurred with wind-generated aerosols during the storms (see Sect. 3.2). The Aitken shoulder is also consistent with recent work from Xu et al. (2022), showing Aitken mode sea spray aerosol, and also consistent with ultrafine aerosols observed during spring in Utqiaġvik, Alaska, related to lead-based sea spray aerosol (Myers et al., 2021). This highlights that blowing snow as a source of aerosol is likely not the only process that contributes to the shape of the BLSN cluster size distributions and that other local wind-sourced aerosols (e.g., sea spray from open leads) should be considered. As such, it is important to mention that the BLSN cluster refers to periods when blowing snow was observed, but it is not exclusively associated with blowing snow particles. The BLSN cluster, representing 20.7 % of all available PNSD observations in October–November, was also associated with higher NaCl mass concentrations (a factor of ∼ 37), N>1000nm (a factor of ∼ 26), and eBC (a factor of ∼ 14) compared to the BG cluster. Furthermore, CCN number concentrations were greatly enhanced within the BLSN cluster periods, with medians of 68.5 cm⁻³ at 0.15 %, 102.6 cm⁻³ at 0.3 % SS, and 138.8 cm⁻³ at 1 % SS, which are ∼ 23, ∼ 15, and ∼ 17 times larger than BG median CCN concentrations at these SS levels, respectively. The LRT cluster's median size distribution is monomodal, dominated by an accumulation mode at 155 ± 1 nm, and only a weak Aitken mode at 41 ± 1 nm associated with a median N10–500 nm value of 83.97 cm⁻³ (∼ 6 times higher than the BG value). CCN number concentrations and SO42- mass concentrations were also enhanced for LRT compared to BG (Table 1). The second LRT cluster (i.e., LRT-aged) was characterized by a lower median N10–500 nm value of 29.61 cm⁻³ and an accumulation mode at 192 ± 1 nm. This could be indicative of longer atmospheric residence times of the particles, thus yielding lower concentrations through dilution and larger particles through aging and coagulation processes. Similar PNSD clusters were found at VRS and other Arctic stations, namely “Haze” and “Aged”, with strong contributions in November and with the main accumulation mode diameters similar to the ones found for our two LRT clusters (Dall'Osto et al., 2019; Lange et al., 2018, 2019; Pernov et al., 2022). Overall, LRT and LRT-aged contained, respectively, 29 % and 25.1 % of all available PNSD measurements in October–November. Since the BG median size distribution shape closely resembles that of the LRT one (i.e., dominant accumulation mode around 150 nm and weak Aitken mode), the background autumnal aerosol population could be interpreted as diluted long-range-transported aerosols. To estimate the contribution of each “main source” (cluster) to the CCN population, we divided the summed CCN number concentrations associated with a given cluster throughout the October–November period by the total (all clusters) summed CCN number concentrations for the same period. We used CCN concentrations at 0.3 % SS, which is assumed to be a representative SS level in autumn in the Arctic (Motos et al., 2023). This approach yielded contributions of 80 %, 17 %, and 3 % to the October–November CCN number concentrations (at 0.3 % SS) for BLSN, LRT + LRT-aged, and BG, respectively.

In March–April, under the high-aerosol background concentration characteristic of Arctic haze (see Sect. 3.1.4), we refer to background conditions with what we call the Haze cluster. The size distributions of the Haze cluster were strongly stable (i.e., low interquartile range) and dominated by an accumulation mode at 176 ± 1 nm and a small Aitken shoulder at 57 ± 1 nm. Overall, the Haze cluster size distribution was characteristic of haze conditions in the Arctic (Boyer et al., 2023; Croft et al., 2016b; Tunved et al., 2013). The association of this cluster with background conditions mostly stems from the fact that this cluster comprised 68.1 % of all PNSD observations in March–April. In contrast with the October–November period, the background Haze concentrations in March–April were high (median N10–500 nm = 138.73 cm⁻³; median SO42- = 0.439 µg m⁻³; similar to the overall March–April SO42- median in Sect. 3.1.4) as a result of more intense long-range transport pollution events and reduced sinks. These more intense long-range transport events are partly the ones that made up the Haze bimodal cluster, where a higher number of concentrations were reached (median N10–500 nm = 234.25 cm⁻³), along with high SO42- and eBC mass concentrations (median SO42- = 0.509 µg m⁻³; median eBC = 74 ng m⁻³). The median Haze bimodal size distribution had roughly equal magnitudes of the Aitken mode (at 38 ± 0.2 nm) and accumulation mode (at 163 ± 1 nm), although with a large interquartile range. Aerosol cloud processing could explain the distinct bimodal distribution shape with a Hoppel minimum (indicative of aerosol cloud processing; Hoppel et al., 1986) at about 80 nm, which is a similar value to what was found elsewhere in the Arctic (Boyer et al., 2023; Freud et al., 2017; Gramlich et al., 2023; Karlsson et al., 2022). Fast transport, differences in the contributing source regions, and contributions from newly formed particles are all possible explanations for the higher contribution of Aitken mode particles to the Haze bimodal cluster PNSDs. Freud et al. (2017) demonstrated the importance of cloud processing as a source of accumulation mode particles in the Arctic, suggesting that, overall, our Haze cluster could have been associated with more cloud processing than the Haze bimodal cluster. Haze bimodal comprised 13.9 % of the March–April PNSD observations but mainly occurred around the mid-April warm-air-mass intrusions (Fig. S11), where the changes in aerosol physicochemical properties associated with these fast-transport events were discussed by Dada et al. (2022a). Similar to October–November, the BLSN cluster in March–April was associated with high submicron PNCs (median N10–500 nm = 198.68 cm⁻³), contributing 18 % of all available PNSD observations, primarily during times when blowing/drifting snow was detected (see Fig. S11b). While the accumulation mode (at 156 ± 3 nm) contribution to the median BLSN PNSD was higher in March–April compared to October–November (see Table S2), the Aitken mode (at 43 ± 2 nm) amplitude was lower. This is also visible when comparing the shapes of the normalized autumn and spring BLSN PNSDs (see Fig. S12). Part of these PNSD differences could be due to differences in the surface snow salinity, size of the blowing snow particles, or snow age, which are all parameters that have been shown through modeling to influence the size of the dry sea salt particles produced from blowing snow (Yang et al., 2008, 2019). Regarding the relation to the measured CCN concentrations and following the same approach as for the October–November period, we find that the BLSN cluster contributed to 20 % of the March–April CCN number concentrations at 0.3 % SS, while Haze contributed 61 % and Haze bimodal 19 %. Although the fraction of blowing-snow-related CCN (remembering that other sources are likely contributing to the BLSN cluster) is smaller compared to autumn, blowing snow episodes remain an important (local) source of CCN, whose importance could increase in the future as the contribution of anthropogenic emissions to the haze burden decreases.

Using daily averaged PNSDs, Boyer et al. (2023) performed a similar PNSD clustering analysis for the entire MOSAiC year (October 2019–October 2020). The authors reported a bimodal cluster, with modal diameters of 46.1 and 135.8 nm, and the highest occurrence in November and April. While this cluster was not attributed to any particular emission process in their study, our results suggest that the bimodal nature of the PNSDs in November and April could have originated from different processes (i.e., locally produced blowing snow and SSA particles in November and long-range-transported cloud-processed particles in April). This highlights the importance of concurrent high-time-resolution observations of aerosol size distributions and chemical composition to understand short-term aerosol variability in the central Arctic and the emission processes related to it.

We obtained size-resolved chemical information for sulfate and organics using the AMS, which can provide critical information, such as the inference of the aerosol mixing state, to complement the PNSD analysis from Sect. 3.3.1. The monthly median of the size distributions of sulfate (in autumn and spring) and organics (in spring only) is shown in Fig. 7, and Fig. S13 shows the same distributions with interquartile range (25th and 75th quantiles). Each mass size distribution was normalized to its maximum to compare their shapes and mode diameters (in the vacuum aerodynamic diameter). The locations of the fitted modes of the monthly mass size distributions for sulfate and organics are given in Table S3.

Sulfate mass in autumn was characterized by a monomodal distribution with mode diameters at 486 ± 3 and 494 ± 2 nm for October and November, respectively. Our observations in Fig. 7a and S14 suggest that autumn was characterized by a smaller number of particles for which most of the (SO42-) mass was contained in the larger range of the size distribution (i.e., above ∼ 400 nm in the vacuum aerodynamic diameter or above 267 nm in the mobility diameter, assuming an average particle density of 1.5 g cm⁻³ (Hegg et al., 1996) and spherically shaped particles). It is possible that the SO42- mass could be related to the LRT-aged cluster in Fig. 6, which was associated with slightly higher (p value = 0.03) SO42- median mass concentration (0.150 µg m⁻³) compared to the LRT cluster (0.141 µg m⁻³), as reported in Table 1, and larger particles overall. In spring, the monomodal distributions had a mode diameter at around 300 nm (seasonal average from March–June = 295 nm), consistent with the characteristics of the Haze PNSD cluster in Sect. 3.3.1, which represented more than two-thirds of the March–April observations. This mode diameter is in line with past studies that have used stage impactor aerosol collection and chemical characterization with ion chromatography in the Arctic (Hillamo et al., 2001; Leck and Persson, 1996; Mukherjee et al., 2021; Ricard et al., 2002). Overall, we observed a non-negligible decrease in the SO42- mode diameter from October (486 ± 3 nm) to June (257 ± 2 nm). This decrease in mode diameter was also observed for the submicron particle volume size distribution from the SMPS (see Fig. S14b). Using ion chromatography measurements of non-sea-salt sulfate, Quinn et al. (2002) found smaller submicron sulfate mass scattering efficiency in summer (July–September) compared to spring (March–June) at Utqiaġvik, Alaska. The authors, however, argued that these differences were negligible and concluded that the sulfate size distribution was unchanged throughout the year. Our observations show that this is not the case, at least in the central Arctic, and that further measurements of size-resolved chemistry in the central Arctic are needed to resolve the influence of sulfate particle size on the climate-relevant scattering efficiency property.

Organics in spring exhibited a bimodal size distribution. The main mode diameter was found at a similar diameter as for sulfate (i.e., around 300 nm; seasonal average March–May = 321 nm), which indicates that, at these sizes, the two species were probably internally mixed. Apart from some variability around the main mode, a second mode was observed above 650 nm, with an unknown mode diameter above 1 µm. We could not estimate the location of this mode as the decrease in the signal above 900 nm was likely related to a decrease in the transmission efficiency at the upper cutoff of the lens. The second mode was also likely missed in the PNSD analysis presented in Sect. 3.3.1 due to the size limitation of the SMPS (10-500 nm). By integrating the monthly median mass size distributions of organics for March–May, we found reasonable agreements with the monthly medians obtained from the mass spectral quantification (see Fig. 3). Indeed, we obtained median organics mass concentrations from the integrated size distributions (mass spectra) of 0.420 (0.334), 0.465 (0.357), and 0.360 (0.283) µg m⁻³ for March, April, and May, respectively. Integrating the second organic mode only (i.e., from 650 nm onward, which corresponds to the approximated location of the minimum between the first and second modes), we found that about 15 %, 17 %, and 19 % of the submicron organic mass is in this second mode for March, April, and May. These percentage contributions are, however, lower estimates, due to the decreased lens transmission efficiency near 1 µm. We hypothesize that this mode corresponded to organic-coated sea salt particles, where only the organics are detected by the AMS, and the sea salt core is left (mostly) undetected due to its refractory nature. Sea spray particles with organic coatings have been observed in previous Arctic studies (Hawkins and Russell, 2010; Kirpes et al., 2019), as well as lab-generated nascent SSA (Ault et al., 2013; Kaluarachchi et al., 2022; Mirrielees et al., 2022). The organic coating is obtained during bubble bursting at the sea surface microlayer (Blanchard, 1975). The classes of organic compounds identified in individual sea spray particles collected in the Arctic include saccharides, fatty acids, and amino acids (Hawkins and Russell, 2010; Kirpes et al., 2019). Such an organic coating can have an impact on the particle CCN activation potential that is two-fold. (1) Through the presence of non-soluble surfactants on the outer shell of the particle, the surface tension can decrease, thus increasing the particle's activation potential (Giddings and Baker, 1977; Ovadnevaite et al., 2017); (2) the lower hygroscopicity of organics can, on the other hand, decrease the particles' activation potential (Ovadnevaite et al., 2011b). Both effects would play a more important role in particular for smaller particle sizes. Both effects could also offset each other, leading to small changes in CCN activity, as shown in controlled laboratory experiments (Moore et al., 2011). It remains to be elucidated what the impact of such organic coatings on CCN activation could be for these large particles in our study.