Condensation and immersion freezing Ice Nucleating Particle measurements at Ny-Ålesund (Svalbard) during 2018: evidence of multiple source contribution

The current inadequate understanding of ice nucleating particle (INP) sources in the Arctic region affects the 15 uncertainty in global radiative budgets and in regional climate predictions. In this study, we present atmospheric INP concentrations by offline analyses on samples collected at ground level in Ny-Ålesund (Svalbard), in spring and summer 2018. The ice nucleation properties of the samples were characterized by means of two offline instruments: the Dynamic Filter Processing Chamber (DFPC), detecting condensation freezing INPs, and the West Texas Cryogenic Refrigerator Applied to Freezing Test system (WT-CRAFT), measuring INPs by immersion freezing. 20 Both measurements agreed within an order of magnitude although with some notable offset. INP concentration measured by DFPC ranged 33-185 (median 88), 5-107 (50) and 3-66 (20) m-3, for T = -22, -18 and -15°C, respectively, while at the same activation temperatures WT-CRAFT measured 3-199 (26), 1-34 (6) and 1-4 (2) m-3, with an offset apparently dependent on the INP activation temperature. This observation may indicate a different sensitivity of Arctic INPs to different ice nucleation modes, even though a contribution from measurement and/or sampling uncertainties cannot be ruled out. 25 An increase in the coarse INP fraction was observed from spring to summer, particularly at the warmest temperature (up to ~70% at -15°C). This suggests a non-negligible contribution from local sources of biogenic aerosol particles. This conclusion is also supported by the INP temperature spectra, showing ice-forming activity at temperatures higher than -15°C. Contrary to recent works (e.g., INP measurements from Ny-Ålesund in 2012), our results do not show a sharp spring-to-summer increase of the INP concentration, with distinct behaviors for particles active in different temperature ranges. This likely indicates that 30 the inter-annual variability of conditions affecting the INP emission by local sources may be wider than previously considered and suggests a complex interplay between INP sources. This demonstrate the necessity of further data coverage. https://doi.org/10.5194/acp-2020-605 Preprint. Discussion started: 2 July 2020 c © Author(s) 2020. CC BY 4.0 License.


Sampling
The aerosol sampling was performed at the Gruvebadet observatory, located in proximity of the village of Ny-Ålesund (78°55' N, 11°56' E) on the Spitsbergen Island, Svalbard. The observatory is about 70 m above sea level, located about 1 Km south-100 west of the village. This position guarantees that the aerosol samples are not affected by local sources of pollution, being the main wind flow from southeast . Aerosol sampling for INP quantification analyses occurred independently for the two methods. Nevertheless, the inlets were all located at the same altitude, about 1 m above the building roof.
For the Dynamic Filter Processing Chamber (DFPC; see Par 2.2.1) aerosol samples were collected on nitrocellulose membrane filters (Millipore HABG04700, nominal porosity 0.45 μm) by deploying two parallel sampling systems, one equipped with a 105 PM1 inlet and the other with a PM10 one (cut-point-Standard EN 12341, TCR Tecora). The operative flow was 38.3 lpm in each sampling line and was generated by two independent pumps (Bravo H Plus, TCR Tecora). Sampling for DFPC occurred on an intensive campaign basis. The spring campaign occurred between 17 April and 2 May 2018, while the summer campaign covered the period between 11 and 27 July 2018. One couple of samples (PM1/PM10) was collected per day, with a sampling duration between 3 and 4 hours, to avoid filter overloading. The sampling generally started in the morning, during the spring 110 campaign, while it started typically in the afternoon during the summer campaign. Samples were stored at room temperature until analysis.
For the West Texas Cryogenic Refrigerator Applied to Freezing Test system (WT-CRAFT; see Sect. 2.2.2) analysis, a total of 28 aerosol filter samples were collected using 47 mm membrane filters (0.2 μm pore size). Briefly, particle-laden air was drawn through a central TSP inlet with a constant inlet flow of 150 lpm. From the inlet, an 8 mm OD stainless steel tube was 115 directly connected to the filter sampler. Below the filter sampler, the filtered-air was constantly pumped through a diaphragm pump. It is noteworthy that a critical orifice was installed upstream of the pump to ensure a constant volume flow rate and control the mass flow rate through the sampling line. A typical sampling interval was approximately of 4 days with only one exception (i.e., 8 days for the sample collected starting on 26 May 2018).

DFPC
INP concentrations were quantified in the lab, after completion of the campaigns, by the membrane filter technique (Bigg et al., 1963;Vali, 1975) following the procedure presented in Santachiara et al. (2010) and described in Rinaldi et al.(2017). A replica of the Langer dynamic filter processing chamber housed in a refrigerator was used to determine the concentration of aerosol particles active as INP at different temperatures. Measurements were performed at activation temperatures (T) of -125 15°C, −18°C and −22°C and at water saturation ratio (Sw) = 1.02. Uncertainties for temperature and Sw are about 0.1 °C and 0.02, respectively. Consequently, the estimated, INP measurement uncertainty of the DFPC is ±30% (DeMott et al., 2018). https://doi.org/10.5194/acp-2020-605 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License. DeMott et al. (2018), McCluskey et al. (2018b) and Hiranuma et al. (2019).

WT-CRAFT 130
To complement the DFPC results, we also used an offline droplet-freezing assay instrument, the WT-CRAFT, to measure temperature-resolved INP concentrations at T > -25 °C, with a detection capability of >1 INP per m -3 of air. While WT-CRAFT is originally a replica of NIPR-CRAFT (Tobo, 2016), the two CRAFT systems possess different sensitivities to artifact and detectable T ranges as described in Hiranuma et al. (2019). As shown in Hiranuma et al. (2019, i.e., Table S2), the uncertainties of temperature as well as ice nucleation efficiency in WT-CRAFT are ± 0.5 °C and ±23.5%, respectively. Other detailed 135 descriptions of WT-CRAFT are provided in Hiranuma et al. (2019), Cory et al. (2019), and Whiteside et al. (2019). Therefore, we only give a brief method description of WT-CRAFT specific to this study. For each experiment, 70 solution droplets (3 µL each) placed on a hydrophobic Vaseline layer were analyzed. With a cooling rate of 1 °C min -1 , we manually counted cumulative number of frozen droplets based on the color contrast shift in the off-the-shelf video recording camera. Afterwards, INP concentration of super-microliter-sized droplets containing particles from the samples were estimated as a function of T 140 for every 0.5 °C. Prior to each WT-CRAFT experiment, we suspended particles on an individual filter sample in a known volume of ultrapure water (HPLC grade), in which the first frozen droplet corresponded to 1 INP per m -3 . More specifically, our suspension-generating protocol followed (1) cutting the filter in two and soaking one filter half in ultrapure water in a sterilized falcon tube, (2) applying a mechanical vibration to the suspension tube to scrub particles on the filter in suspension, (3) applying an idle time of 5 min to have the quasi-steady state suspension, and (4) preparation of droplets out of the 145 suspension. If necessary, the suspension sample was diluted until we observe their freezing spectrum collapsed onto the water background curve. It is noteworthy that our diluted spectra and original freezing spectrum reasonably agreed in their overlapped T region (within a factor of three at the most) without any notable artifacts at T above -25 °C. Due to the absence of failure, we simply stitched all spectra in the way that the diluted spectrum followed up and took over the cold temperature data points immediately after the last data point of less diluted spectrum. 150

Ice Nucleation Parameterizations
The atmospheric concentration of ice nucleating particles (nINP), expressed hereafter in units of m -3 , was calculated, for each technique, by dividing the number of INP quantified for each sample by the total volume of air passed through the corresponding filter. The ice nucleating active site density (ns) was derived as in Niemand et al. (2012), by normalizing the INP number concentration for the total aerosol surface in the range between 10 nm and 10 μm (see next paragraph for details), 155 calculated under the assumption of spherical particles. https://doi.org/10.5194/acp-2020-605 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.

Particle size distribution measurements
The aerosol particle number size distribution is continuously monitored at Gruvebadet since 2010 using a Scanning Mobility Particle Sizer (SMPS) model TSI 3034 for the diameter range between 10 and 500 nm (54 channels) and an Aerodynamic 160 Particle Sizer (APS) model TSI 3321 for the diameters above 500 nm (same number of channels as the SMPS). Both instruments are connected to a common multiple inlet with laminar flow (the same where the sampling for the WT-CRAFT analysis was performed) and record data averaged over 10 minutes Lupi et al., 2016). The aerodynamic diameters reported by the APS were corrected to real physical diameters using a particle mass density equal to 1.95 g cm -3 and the number concentration in the resulting overlapping range was taken equal to that from the SMPS. 165

Meteorology
Meteorological parameters (air temperature, T; pressure, P; relative humidity, RH; wind speed, WS) were taken from those continuously provided by the Amundsen-Nobile Climate Change Tower, positioned less than 1 Km N-E of Gruvebadet , while precipitation data (type and amount) from the eKlima database, provided by the Norwegian Meteorological Institute (https://seklima.met.no/observations/). 170

Offline Chemical Analysis
The chemical analysis of major and trace ion species, used in this work as aerosol source tracers, was accomplished on filters collected at GVB. The filters were handled in conditions of minimal contamination (working under a class 100 laminar flow hood) during all the phases of the analytical procedure. The measurements were carried out by a triple Dionex ThermoFisher Ion Chromatography system equipped with electrochemical-suppressed conductivity detectors. In particular, a Dionex AS4A-175 4 mm analytical column with a 1.8 mM Na2CO3/1.7 mM NaHCO3 eluent, was used for the determination of most of inorganic anions (Cl -, NO3 -, SO4 -2 , C2O4 -2 ) while a Dionex AS11 separation column with a gradient elution (0.075-2.5 mM Na2B4O7 eluent) was used for the measurement of Fand some organic anions (acetate, glycolate, formate and methanesulfonate).

Back trajectories and satellite ground type maps
The ground types over which air masses travelled in the 5 days before arrival at GVB station were identified, for both DFPC and WT-CRAFT samples, following . The considered ground types were seawater, sea-ice, land, and snow (over land). For this analysis, five-day back-trajectory air masses (HYSPLIT4 with GDAS data: https://ready.arl.noaa.gov/) from the National Oceanic and Atmospheric Administration (NOAA) HYSPLIT model (Rolph et al., 2017;Stein et al., 2015) 185 arriving at an altitude of 100 m (amsl) over GVB station were calculated. For DFPC samples, the back-trajectories arrival time https://doi.org/10.5194/acp-2020-605 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.
was considered simultaneous to INP samples, while for WT-CRAFT, the trajectories were calculated 4 times (00, 06, 12 and 18 UTC) a day covering the INP sampling period from April to August.
Ground condition maps were obtained from the National Ice Center's Interactive Multisensor Snow and Ice Mapping System (IMS) (Helfrich et al., 2007;National Ice-Center, 2008), National Snow & Ice Data center (NISDC; https://nsidc.org/). IMS 190 maps are a composite product produced by NOAA/NESDIS (National Environmental Satellite Data and Information Service) combining information on both sea ice and snow cover. Information from 15 different sources of input is included in the production of these maps (Helfrich et al., 2007). We used the daily Northern Hemisphere maps with a resolution of 4 km. For each back-trajectory time step, we applied nearest-neighbour interpolation in space and time to find the corresponding satellite coordinate along the back trajectory. Consequently, the ground type conditions during air mass passage were determined. It is 195 worth highlighting that only low crossing air masses, up to an altitude of 500 m amsl were considered for this analysis.

Satellite chlorophyll-a data and correlation analysis
The best estimate "Cloud Free" (Level-4) daily sea surface chlorophyll-a concentration (CHL; mg m -3 ) data were downloaded from the EU Copernicus Marine Environment Monitoring Service (CMEMS; http://marine.copernicus.eu/) based on a multisensor approach (i.e., SeaWiFS, MODIS-Aqua, MERIS, VIIRS and OLCI-S3A). The Level-4 product is available globally at 200 ~4 km spatial resolution. From this global dataset, CHL fields were extracted in the Arctic Ocean during summer 2018 to be merged with INPs data.
The relationship between INPs and phytoplankton biomass, in terms of CHL concentration, were investigated excluding the samples influenced by land inputs. The DFPC dataset was chosen to run this analysis because it provides a higher timeresolution than the WT-CRAFT one, and because it allows to differentiate between fine and coarse INPs. Each DFPC sample 205 collected at a certain day has been considered as representative for that day, in order to be compared with the daily CHL timeseries. The Pearson correlation coefficients between INPs and satellite-derived ocean color data, obtained by standard least squares regression, were computed at each grid point of the Arctic domain, and for different time-lags, to obtain the correlation maps presented in the results section.

Concentration weighted trajectory 210
The allocation of regional source areas potentially affecting INP concentrations sampled at Ny-Ålesund was achieved by applying the concentration weighted trajectory (CWT) model (Bycenkiene et al., 2014;Jeong et al., 2011;Hsu et al., 2003). In this procedure, each grid cell within the studied domain is associated to a weighted concentration, which is a measure of the source strength of a grid cell with respect to concentrations observed at the sampling site. The average weighted concentration in the grid cell (i,j) is determined as follows: Where t is the index of the trajectory (arrival time simultaneous to INP samples), L is the total number of trajectories (5 days -hourly time step), Ct is the INP concentration observed at sampling location (receptor site) on arrival of trajectory t, and Dijt is the residence time (time spent) of trajectory t in the grid cell (i,j). Given Ct for INP, Dijt can be determined by counting the number of hourly trajectory segment endpoints in each grid cell for each trajectory. This was repeated for all the back 220 trajectories L. A high value for CWTij means that air parcels traveling over the grid cell (i,j) would be, on average, associated with elevated concentrations at the receptor site.
In this study, five-day low (< 500 m) air mass back-trajectory corresponding to DFPC INP samples were utilized to produce the CWT spatial distribution. Similarly to the correlation analysis, the INP samples with a clear influence from land were excluded to consider only marine sources. The domain extends up to the limits of the area covered by the above described low 225 back-trajectories (60° W -30° E; 50° -85° N) and was divided into 1°×3° latitude/longitude grid cells. The average number of endpoints over the grid cells with at least one endpoint (D * ) was 5.6. In order to avoid uncertainties that can occur due to grid cells containing a low number of endpoints, the CWT values were multiplied by a weighting factor (Wij) as follows.
3 Results an offset was observed between the two techniques. Specifically, nINP measured in condensation mode (DFPC) resulted generally higher than those measured in immersion mode (WT-CRAFT) and the difference increased with the activation temperature. On average, nINPDFPC was 3 times higher than nIPNWT-CRAFT at T = -22°C and 8 times higher at T= -15°C. As a result, WT-CRAFT data presented a sharper ∆nINP/∆T slope than the DFPC ones. 240

Comparison of DFPC and WT-CRAFT measurements
The observed offset may derive from the different time resolutions of the sampling for INP analyses, as well as from uncertainties in sampling activities and/or measurement uncertainties (Hiranuma et al., 2015;. Conversely, it is also valid to assume a different sensitivity of Arctic INPs to different ice nucleation modes. Some previous studies presumed that condensation and immersion freezing are equivalent, but this hypothesis is questionable. Briefly, condensation freezing occurs if ice is formed immediately after water vapour condensation on the solid particle, followed by an additional ice growth by 245 deposition. Unlike in condensation, a droplet must be formed at higher temperatures and necessarily undergoes supercooling https://doi.org/10.5194/acp-2020-605 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License. before freezing in immersion freezing (DeMott, 2002;Kanji et al., 2017). These different modalities could influence the obtained INP concentrations with the DFPC and the WT-CRAFT devices, particularly in case of mixed soluble/insoluble particles. Previous results, which evidenced a similar ice nucleating efficiency for immersion and condensation freezing, were indeed obtained with insoluble aerosol particles (Hiranuma et al., 2015;Wex et al., 2014). A detailed intercomparison of 250 techniques is not the scope of this study. However, it is noteworthy that our past attempts to intercompare DFPC and WT-CRAFT measurements with different aerosol types yielded different results. For instance, microcrystalline and fibrous cellulose samples tended to form more ice crystals in the WT-CRAFT (Hiranuma et al., 2019), while ambient continental aerosol particles from the Po Valley resulted in equivalent or higher ice crystal numbers in the DFPC (unpublished). This suggests some sensitivity of the aerosol type to the different ice activation modalities deployed by the two instruments. WT-CRAFT measurements probed the immersion freezing ice nucleation ability of aerosol particles between 0 and -25°C (Figure1). Above -9 °C, no ice nucleation activity was observed in GVB samples. Between -9 and -14°C only a fraction of the samples (<50%) presented INP concentrations above the detection limit, with concentrations never above 3 m -3 . In the rest of the temperature spectrum, nINP ranged 1-3 (median 2) m -3 , at T = -14°C and 24-9082 (166) m -3 , at T = -25°C.
The first ground level INP data reported in literature for Ny-Ålesund are those by Borys (1983). Measurements were performed 265 with T between -28 and -16°C and the observed INP concentrations ranged between less than 10 to ca. 500 m -3 . Later on, Bigg (1996) measured INP active at -15°C in a static chamber and at humidity just above 100%, during an icebreaker cruise to the North Pole (1 August -6 October 1991). The overall geometric mean was 8.2±2.9 m -3 and the highest measured concentration was 250 m -3 . The Ocean was the prevalent source of INPs. Similar measurements were performed by Bigg and Leck (2001)  continental particles, which we will show play an important role in the Arctic atmosphere (Sect. 3.7.2). If we compare with recent measurements performed at lower latitudes by DFPC nINP over the Arctic was lower than those observed in continental European sites Rinaldi et al., 2017), but comparable or even higher with respect to those observed at high altitudes  or at a Mediterranean coastal location (Rinaldi et al., 2019).
WT-CRAFT immersion freezing spectra (nINP as a function of T) measured in the present study show a unique feature of ice 300 nucleation behaviour at relatively high temperatures (T > -20 °C) in comparison to the freezing temperatures of other typical INPs (e.g., dust). Some spectra show initial nucleation at above -15 °C and follow the previously reported ice nucleation spectral feature of marine biogenic aerosol particles (Wilson et al., 2015;Irish et al., 2017). For instance, the August #2 sample (highlighted in Figure 2) shows the bimodal activation with a hump feature at T above -15 °C. The reason for early ice nucleation may be due to marine biogenic aerosols (Wilson et al., 2015;Irish et al., 2017). This aspect will be discussed in the 305 next Sections.

Contribution of fine and coarse INPs
The sampling strategy adopted for DFPC measurements (parallel PM1 and PM10 sampling) allowed a basic investigation of the INP size distribution. Table 1 reports the number concentrations of INPs measured in the two different size ranges, together with the average contribution of super-micrometer (coarse) INPs, derived by difference. A small contribution from coarse (p<0.005) increase of the contribution of coarse INPs was observed (from ~50% at T = -22°C to ~70% at T = -15°C), likely resulting from the activation of local sources after snow and ice melting. Furthermore, the increase of coarse INP contribution, from spring to summer time, is progressively more pronounced with increasing activation temperature, which may evidence 315 the contribution of biological coarse particles during summer. This aspect agrees with the above considerations on the ice nucleation spectra from WT-CRAFT data. Similar results were reported by Mason et al. (2016) for Alert Arctic station, including the increasing coarse INPs contribution as a function of the activation temperature. Si et al. (2018) and Creamen et al. (2018) reported a significant higher ice nucleation efficiency for super-micrometer particles sampled at Arctic stations. Similarly, comparing the INP number concentrations measured by DFPC in the fine and coarse 320 modes with particle number concentration in the same size regimes (sub-micrometer: 3-61 cm -3 ; super-micrometer: 2-28 cm -3 ), a higher ice nucleation efficiency can be attributed to coarse particles during the summer campaign. In particular, the ice nucleation efficiency of coarse particle resulted from 1.7 (at T = -22°C) to 5.5 (at T = -15°C) times higher than that of fine particles. For comparison, a lower enhancement of the ice nucleation efficiency was observed for coarse particles with respect to fine ones in spring, with a maximum enhancement of 2.5 times at T= -22°C and negligible effect at lower temperatures. 325 The time trends reported in Figure 3 do not show such a sharp seasonal increase in the INP atmospheric concentration from spring to summer. For DFPC data, actually, a significant (p<0.01) nINP reduction (by a factor 1.5) was observed at T=-22°C, passing from the spring campaign (April) to the summer period (July), while no significant (p>0.05) differences were observed 335 for T=-15°C and T=-18°C. The time evolution of INP concentrations measured by WT-CRAFT agrees with the parallel dataset if we consider only the periods in which the two measurements were run in parallel: a significant (p<0.05) reduction by a factor 1.6 is observed at -22°C and no significant differences can be appreciated at -15 and -18°C. On the other hand, considering the whole WT-CRAFT data extent, a statistically significant (p<0.5) increasing nINP seasonal trend was observed only for temperatures within -17.5 and -21.5°C. Even in these cases, the spring-to-summer enhancement ratios never exceeded three-340 times. Conversely, it is worth noticing that a clear nINP peak was observed during June, at all temperatures lower than T = -

Relation of nINP with particle number concentration and meteorological parameters
Analyzing the patterns of the main meteorological parameters (T, P, RH and WS) in relation to nINP, no clear relation emerges, 360 with the exception of precipitation events, which were often associated to a reduction of the INP concentration ( Figure S2).
The precipitation scavenging of aerosol particles (and consequently of INPs) by simple ice crystals and snowflakes (aggregate of ice crystals) was examined in the past both theoretically (Miller and Wang, 1989;Feng, 2009) and experimentally (Murakami et al., 1985;Zikova and Zdimal, 2016;Bell and Saunders, 1991). Kyrö et al. (2009) measured the snow scavenging coefficient of sub-micrometer aerosol particles in the clean background SMEAR II station (Hyttiala), using 4 years of particle number 365 concentration spectra and meteorological parameters measurements. The obtained experimental median scavenging coefficient was found to be 1.8 x 10 -5 s -1 , varying between 0.87 and 5.2 x 10 -5 s -1 in the 10 nm to 1 μm size range. Paramonov et al. (2011) reported an analysis of below-cloud snow scavenging of aerosols in the 0.01 to 1 μm size range for an urban environment, where the levels of air pollution were typically higher than at background sites. The calculated mean scavenging coefficients varied between 6.65 x 10 -6 and 5.14 x 10 -5 s -1 , in good agreement with those reported by Kyrö et al. (2009) for background 370 conditions.
Although nINP tends to covariate with particle number concentration (in the range 0.5-10 μm) during the spring campaign, no significant correlation was observed, for the DFPC dataset (with the only exception of INPPM10 at T = -15°C). During summer, the lack of correlation between nINP and particle number is even more accentuated. For WT-CRAFT significant correlations  et al., 1995). In other cases, a complete lack of correlation has been documented (Rogers et al., 1998;Richardson et al., 2007), which is not surprising considering that INPs are only a small fraction of total particles. Bigg et al. (1996) reported a good correlation between nINP and accumulation mode particles, for one day of measurements over the high Arctic, while a modest 380 but significant correlation (R = 0.25 -0.30) between nINP and particle number concentration in the 50-120 nm range was reported by Bigg et al. (2001), close to the North Pole. No other paper, to the best of our knowledge, addressed this issue in the Arctic environment. Figure 4 presents the ns distribution as a function of the activation temperature for the two datasets. DFPC showed a significant 385 (p<0.05) increase of the ice active site density (ns), passing from the spring campaign to the summer period for all the probed activation temperatures ( Figure 5). This result shows that the spring aerosol population, mainly related to long-range transport of anthropogenic aerosol particles from lower latitudes (Arctic haze), has a lower ability in nucleating ice than the summertime aerosol population, more related to local (Arctic) sources. This is in agreement with the findings by Hartmann et al.

Ice active site density (ns)
(2019), which showed a low impact of anthropogenic emissions over the INP concentration, with respect to the preindustrial 390 period, through the analysis of ice core records. The spring-to-summer ns increase is progressively more evident at T = -15°C than at T = -22°C, suggesting that local aerosol particles are particularly efficient in nucleating ice at warmer temperatures, which is typical of biological INPs (Murray et al., 2012;Wilson et al., 2015;DeMott et al., 2016;McCluskey et al., 2018b). The time series of ns values by WT-CRAFT reported in Figure 5 reflected the increase in INP concentration characterizing the month of June described above. This confirms that the enhancement in INP concentration (for T < -17°C) observed in June was due to enhanced ice activity of the particle population, rather than to an increase of aerosol particle concentration.
In Figure  A more detailed analysis of the relative contribution of mineral dust and marine aerosol sources during the study period will be presented in the following Sections.

Correlation with chemical tracers
In order to investigate the main sources of the INPs measured at GVB, a correlation analysis was performed between both 415 nINP datasets and the atmospheric concentration of chemical tracers routinely measured at the station. During the spring campaign by DFPC, nINPDFPC correlated, almost independently on the size fraction or the activation temperature, with tracers of long range transported anthropogenic aerosol particles as nitrates, non-sea-salt-sulfate and non-sea-salt-potassium (Table   2). Indeed, Udisti et al. (2016) associated spring-time non-sea-salt-sulfate at GVB to anthropogenic sources, showing that the production of biogenic non-sea-salt-sulfate from the sea is relevant only in summer-time. The results of the correlation analysis 420 are in line with the above considerations about long-range transport of anthropogenic aerosol during springtime over the Arctic.
A general tendency to anticorrelation with sodium and chlorine was also observed, even though significant values (p<0.05) were observed only in the PM1 size fraction. Less clear indications resulted from the analysis of the summer DFPC data. The only significant relations were observed for T = -15°C: an anticorrelation was observed between nINPPM10 and particulate mass, sea spray (sodium and chlorine) and mineral dust (calcium, magnesium and lithium) particles. Recently glacial soils 425 have been indicated as potentially important INP sources in the Arctic region during summertime (Tobo et al., 2019). If this was true also during the DFPC measurement period, calcium, magnesium and lithium may be not the best chemical tracers for this type of soils. Similarly, no clear source indications (no significant correlations) were derived from the correlation analysis of the WT-CRAFT data (not shown).

Influence of ground conditions 430
The influence of ground conditions (sea-ice, snow, seawater and land) on low-travelling back-trajectories (<500m) corresponding to the collected samples was evaluated by merging back trajectories and satellite ground type data . Figure 6 shows that the contribution of the four considered ground types varies with the season. In spring, the majority of contacts occurred with sea-ice or snow-covered land, while in summer low air masses were more influenced by ice-free seawaters. The (snow-free) land contribution was the lowest in every season. Nevertheless, the influence of land sources on 435 the INP concentrations emerges clearly from Table 3 and Figure S3: air masses with a higher terrestrial influence were always associated with nINP peaks. This is likely due to the higher ice nucleation efficiency of mineral dust and soil particles compared to marine biological particles (Wilson et al., 2015;McCluskey et al., 2018a;McCluskey et al., 2018b). In summer, contacts with snow-free land occurred mainly within the Svalbard archipelago (local sources) or over Greenland and Iceland (regional sources), as shown by Figures S1. This outcome is in agreement with recent works pointing to both local and regional soils as 440 important INP sources over the Arctic (Tobo et al., 2019). https://doi.org/10.5194/acp-2020-605 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.

Contribution of marine biological INP sources
Considering that, during summer, the sampled air masses had ground contacts mainly over seawater, one can hypothesize that marine biological sources may dominate the INP concentration at GVB, outside the periods of elevated terrestrial influence.
To check this hypothesis, we investigated the spatio-temporal correlation of the INP datasets with satellite retrieved surface 445 chlorophyll concentration, used as a tracer for marine biological activity, following the time-lag approach first introduced by Rinaldi et al. (2013). The DFPC dataset was selected to run this analysis because it provides a higher time resolution than the WT-CRAFT one and, most of all, because it allows to distinguish between fine and coarse INPs. In fact, McCluskey et al.
(2018b) and Mansour et al. (2020a) showed that fine INPs tend to correlate better with chlorophyll in clean marine air masses.
To exclude interferences from land sources, we removed from the dataset the samples corresponding to back trajectories that 450 have been in contact with land for more than 10% of the time (3 samples). Furthermore, we focused on INP data obtained at T = -15°C, which are the most representative of ice nucleation by biological particles and the less subject to influences from mineral particles.
The results of the correlation analysis are reported in Figure 7, in the form of correlation maps. In the maps, the colour of each pixel represents the correlation coefficient (R) resulting from the linear regression between the CHL concentration in that pixel 455 and nINPPM1 measured at GVB. Different maps were obtained by considering different lag times between the two correlated time series, i.e., by considering CHL concentration values shifted back in times of 1 to 27 days with respect to the INP filter sampling times (the maps are shown in Figure S3). The lag time approach has been demonstrated to maximize the correlation between in situ coastal measurements of aerosol properties and CHL concentration fields (Rinaldi et al., 2013;Mansour et al., 2020a;Mansour et al., 2020b); it likely reflects the time scale of the biochemical processes responsible for the production of 460 transferable organic matter in the seawater after the phytoplankton growing phase that is tracked by CHL patterns. Sea regions characterized by high correlation (red dots in the maps) are likely related to the emission of biological particles acting as INPs in our samples. Figure 7 reports two examples of correlation maps, with lag time 6 and 16 days. These maps were selected because they clearly show high correlation regions in the seawaters surrounding the Svalbard archipelago (lag time 6 days), close to the Greenland coast (lag time 16 days) and to the northeast of Iceland (lag time 16 days). These regions were all 465 consistently located upwind GVB during the sampling period ( Figure S1). All the obtained maps are available in the Supplementary Material, including those obtained with PM10 INP data, which as expected, do not evidence any significant correlation with CHL ( Figure S4). In our interpretation, the lack of a correlation between surface CHL concentration and coarse INPs does not imply that coarse INPs are not emitted from the ocean surface, it simply evidences that CHL is not the appropriate proxy to track the emission of large biological INP from the oceans. Indeed, CHL has been previously observed 470 to correlate with the enrichment of organic matter in sub-micron sea spray (Rinaldi et al., 2013;O'Dowd et al., 2015) but no investigation was ever attempted with super-micrometer particles. McCluskey et al. (2017) clearly evidenced the production of both sub-and super-micrometer INPs during laboratory experiments with controlled algal blooms, pointing out that different particle type and production mechanisms are involved.
Aware that the evidenced correlations alone cannot imply unambiguously a cause-effect relation, on the same INP dataset 475 (DFPC; PM1; T = -15°; no land influenced samples) we run also the CWT spatial source attribution model. The resulting map ( Figure 7c) evidence that potential sub-micron INP sources at GVB, during the study period, were broadly located in the same sea regions previously evidence by the spatio-temporal correlation with CHL. Furthermore, the CWT approach provides the same results also with PM10 data and independently on the considered activation temperature ( Figure S5). The consistency of the CWT source identification with those of the spatiotemporal correlation with CHL, two totally independent approaches, 480 suggests strongly that marine sources located between the Svalbard archipelago and Greenland/Iceland may have contributed to the INP population measured at GVB during summer 2018, outside the major evidenced episodes of terrestrial influence.

Conclusions
Concentrations of INPs measured by two independent techniques at Ny-Alesund, during spring-summer 2018, were presented in this work. The INP concentration trends obtained by the two techniques were qualitatively in good agreement, even though 485 with presence of a notable offset. However, the observed difference never exceeded one order of magnitude and we notice that it increased with the activation temperature. This is presumably attributable to the different ice nucleation mechanisms probed by the two techniques (condensation freezing, for DFPC, and immersion freezing, for WT-CRAFT), even though differences in the sampling resolution and overall measurement uncertainties may also have contributed. Understanding whether the INP data from Ny-Ålesund would be representing the pan-Arctic conditions or a local situation indeed requires long-term measurements. Our study presented in this paper might be an important step towards this broad objective. Overall, our new study motivates and warrants the necessity for more frequent measurements on the long-term in 500 order to understand INP production processes in the Arctic environment.
Analysis of INP concentrations, active site density, low-travelling back-trajectories and ground conditions during the passage of the air mass suggest that the summertime INP population may be contributed both by terrestrial and marine sources. When the sampled air masses were influenced by contact with snow-free land, the INP concentration tended to peak, likely reflecting the higher nucleation ability of terrestrial particles. Outside the major terrestrial inputs, the INP population was likely regulated 505 by marine INPs emitted from the sea surface. A prove of the relation between INP concentration (outside the major terrestrial https://doi.org/10.5194/acp-2020-605 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License. inputs) and the patterns of marine biological activity have been provided, suggesting that sourced of INP may be located both in the seawaters surrounding the Svalbard archipelago and/or as far as Greenland and Iceland.

Data availability
Data discussed in this work are available at http://dx.doi.org/10.17632/zf4wdcc3bw.1 510 Satellite Chlorophyll data are available for download at http://marine.copernicus.eu/ (product identifier: OCEANCOLOUR_GLO_CHL_L4_REP_OBSERVATIONS_009_082). The authors declare that they have no conflict of interest.

Acknowledgements
The authors thank DSSTTA-CNR and its staff for the logistical support that allowed the realization of the experimental activity. https://doi.org/10.5194/acp-2020-605 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.