During July and August 2016 samples were collected from the eastern Canadian Arctic on board the CCGS Amundsen as part of the NETCARE project (Fig. 1 and Table 1). Information recorded at each sampling station is provided in Table S1 in the Supplement.

In contrast to 2014, when we collected microlayer samples manually using a glass plate sampler (Irish et al., 2017), in 2016, microlayer samples were collected using rotating glass plates attached to a sampling catamaran (Shinki et al., 2012). At station 1, the sampling catamaran was deployed from a small boat at least 500 m away from the CCGS Amundsen. The sampling catamaran was remotely driven at least 20 m away from the small inflatable, rigid-hull boat before the rotating glass plates were activated remotely. A rotation rate of 10 revolutions per minute was used. From station 2 onwards, the remote control of the rotating glass plates on the sampling catamaran failed. Subsequently, the sampling catamaran was kept on the upwind side of the small inflatable, rigid-hull boat with its engine turned off, at least 500 m away from the CCGS Amundsen to avoid contamination, and the glass plates were rotated manually between 11 and 18 revolutions per minute. The microlayer that adhered to the plates from each rotation was scraped off with fixed Teflon wiper blades into a manifold and then pumped through Teflon tubing into high-density polyethylene (HDPE) Nalgene bottles (ranging from 250 mL to 2 L in volume). The thickness of the microlayer collected was approximately 80 µm based on the rotation rate (between 11 and 18 revolutions per minute), the average volume collected (3 L), and an average collection time (18 min). Bulk seawater samples were collected at the same times and locations through Teflon tubing suspended 0.2 m below the sampling catamaran. The bulk seawater was pumped, using peristaltic pumps, into HDPE Nalgene bottles (ranging from 250 mL to 2 L in volume). After collection, the Nalgene bottles containing the microlayer and bulk seawater samples were kept cool in an insulated container. Upon returning to the ship, the samples were subsampled into smaller bottles for subsequent analyses.

The glass plates, aluminium manifold, Teflon tubing, and all Nalgene bottles were sterilized first with bleach then cleaned with isopropanol and finally rinsed with ultrapure water. After cleaning, the sampler was further rinsed by collecting then discarding microlayer and bulk seawater for approximately 2 min, before samples were retained.

The abundances of heterotrophic bacteria and phytoplankton <20 µm (i.e., phycoerythrin-containing cyanobacteria, phycocyanin-containing cyanobacteria, and autotrophic eukaryotes) were measured by flow cytometry. Duplicate 4 mL subsamples were fixed with glutaraldehyde (Grade I; 0.12 % final concentration; Sigma-Aldrich G5882) in the dark at room temperature for 15 min, flash-frozen in liquid nitrogen, and then stored at -80 ∘C until analysis. Samples for heterotrophic bacteria enumeration were stained with SYBR Green I (Invitrogen) following Belzile et al. (2008) and counted with a BD Accuri C6 flow cytometer using the blue laser (488 nm). The green fluorescence of nucleic acid-bound SYBR Green I was measured at 525 nm. Archaea could not be discriminated from bacteria using this protocol; therefore, hereafter, we use the term bacteria to include both archaea and bacteria with high nucleic acid (HNA) content and low nucleic acid (LNA) content. SYBR Green I stains all DNA and RNA, but bacteria and archaea are easily discriminated from other organisms (or detritus or transparent exopolymeric particles) by their size (side scatter) and fluorescence intensity. In addition, autotrophs stained with SYBR Green I are discriminated from heterotrophic bacteria by their chlorophyll a fluorescence.

Samples for >20 µm phytoplankton abundances were analysed using a CytoFLEX flow cytometer (Beckman Coulter) fitted with a blue (488 nm) and red laser (638 nm), using CytoExpert v2 software. Using the blue laser, forward scatter, side scatter, orange fluorescence from phycoerythrin (582/42 nm BP), and red fluorescence from chlorophyll (690/50 nm BP) were measured. The red laser was used to measure the red fluorescence of phycocyanin (660/20 nm BP). Polystyrene microspheres of 2 µm diameter (Fluoresbrite YG, Polysciences) were added to each sample as an internal standard (Marie et al., 2005; Tremblay et al., 2009).

To investigate the possible importance of sea-ice melt and meteoric water (terrestrial runoff plus precipitation) to INP concentrations, we determined δ18O in the samples. Measurements of δ18O have been used in the past to distinguish between sea-ice melt and meteoric water in the Arctic Ocean (Alkire et al., 2015; Macdonald et al., 1995; Östlund and Hut, 1984; Tan and Strain, 1980). δ18O, a measure of the ratio of oxygen-18 (18O) to oxygen-16 (16O) in water molecules, is expressed as per mil (‰) deviations from Vienna Standard Mean Ocean Water (V-SMOW): 2δ18O=18O16Osample18O16Ostandard-1×1000‰, where “standard” corresponds to V-SMOW. Samples were analysed at the GEOTOP-UQAM stable isotope laboratory at the Université du Québec à Montréal using the CO2 equilibration method (Ijiri et al., 2003), where 200 µL of sample was equilibrated with CO2 for 7 h at 408 ∘C. The CO2 was then analysed on a Micromass Isoprime^™ universal triple collector mass spectrometer in dual-inlet mode with an AquaPrep^™ system (Isoprime Ltd., Cheadle, UK). Two internal reference water samples (δ18O =-6.71 ‰ and -20.31 ‰) were used to normalize the sample data. Uncertainties in replicate measurements are ±0.05 ‰ (1σ). δ18O values were determined for all stations, except stations 1, 10, and 11.

From the measured δ18O values and measured salinities of the samples, the water volume fractions of sea-ice melt (fSIM), water volume fractions of meteoric water (fMW), and water volume fractions of seawater (fsw) were calculated using the following conservation equations (Yamamoto-Kawai et al., 2005): 3fSIMSSIM+fMWSMW+fSWSSW=Sobs4fSIMδ18OSIM+fMWδ18OMW+fSWδ18OSW=δ18Oobs5fSIM+fMW+fSW=1, where S represents salinity, and the subscripts obs, SIM, MW, and SW represent observed, sea-ice melt, meteoric water, and seawater, respectively. For SSIM, SMW, δ18OSIM, and δ18OMW we assumed 4 g kg-1, 0 g kg-1, 0.5 ‰, and -20 ‰, respectively, in Eqs. (3)–(4) (Burgers et al., 2017). The values of SSW and δ18OSW depend on the reference seawater chosen. In our studies the samples could have been influenced by either Arctic outflow waters (SSW=33.1 g kg-1 and δ18OSW=-1.53 ‰) or west Greenland current waters (SSW=33.5 g kg-1 and δ18OSW=-1.27 ‰) (Burgers et al., 2017). When calculating fSIM, fMW, and fsw values we used SSW=33.3±0.2 g kg-1 and δ18OSW=-1.40±0.13 ‰, which correspond to the average and limits for Arctic outflow waters and west Greenland current waters.

Chlorophyll a concentrations for case 1 waters (waters dominated by phytoplankton) were retrieved from the GlobColour project website (http://globcolour.info, last access: 19 September 2018, ACRI-ST, France). The GlobColour project provides a high-resolution, long time series of global ocean colour by merging data from several satellite systems. The data used here include retrievals from either or both the Moderate Imaging Spectrometer (MODIS) on the Aqua Earth Observing System (EOS) mission and the Visible Infrared Imager Radiometer Suite (VIIRS) aboard the Suomi National Polar-orbiting Partnership satellite. For this work we used data merged with weighted averaging, where weightings are based on the sensor and/or product. For more information regarding the weighted averaging refer to the GlobColour Product User Guide (http://www.globcolour.info/CDR_Docs/GlobCOLOUR_PUG.pdf, last access: 19 September 2018). In this study 8 d data were used to achieve the best balance between spatial coverage (1/24∘, ∼4 km) and high time resolution. For the chlorophyll a concentration at a given sampling location, we used the grid cell corresponding to the location of that station. We determined the chlorophyll a concentration at all stations except station 8.

Chlorophyll a concentrations were also measured in collected samples of seawater. Samples were filtered onto Whatman GF/F glass-fibre filters, and chlorophyll a concentrations were measured using a Turner Designs AU-10 fluorometer, after 24 h extraction in 90 % acetone at 4 ∘C in the dark (acidification method: Parsons et al., 1984).

The fraction frozen curves for all microlayer and bulk seawater samples measured in 2016 are shown in Fig. 2. Also shown for comparison are the fraction frozen curves of the samples after filtration through a 0.02 µm Anotop 25 syringe filter, the fraction frozen curves for the laboratory blanks (ultrapure water passed through a filter with a 0.22 µm pore size), and fraction frozen curves for field blanks (ultrapure water passed through the sampling catamaran). The laboratory blanks are at similar or warmer temperatures than the samples passed through a 0.02 µm Anotop 25 syringe filter. Differences are most likely due to the difference in pore sizes of the filters used: the laboratory blanks were passed through filters with a 0.22 µm pore size, whereas the samples were passed through filters with a 0.02 µm pore size. Previous experiments in our laboratory have shown that ultrapure water passed through a filter with a 0.02 µm pore size can freeze at slightly colder temperatures than ultrapure water passed through a filter with a 0.22 µm pore size (Fig. S2).

For the bulk seawater, all untreated samples froze at temperatures warmer than the laboratory and field blanks. Freezing temperatures as warm as -6 ∘C were observed. These results indicate the presence of ice-active material in all bulk seawater samples. For the microlayer samples, all samples froze at temperatures warmer than laboratory blanks. In addition, most samples froze at temperatures warmer than the field blanks. These results also indicate that most microlayer samples contained ice-active material. For some of the samples, the freezing temperatures of the field blanks were warmer than the freezing temperatures of the samples. However, the freezing temperatures of the field blanks should be viewed as an upper limit to the background freezing temperatures, since prior to collecting the field blanks, the sampler had not been rinsed as thoroughly as before collecting the microlayer samples. For the remainder of this paper we will only show and discuss freezing data that were at warmer temperatures than the field blanks.

In Fig. 3 the concentrations of INPs, [INP(T)]vol,liq, measured in 2016 are compared with concentrations measured in 2014 (sample locations for both years shown in Fig. 1). In both 2016 and 2014, the concentrations of INPs vary by at least 2 orders of magnitude at a given temperature, but warmer freezing temperatures were observed in 2016 compared to 2014.

Figure S3 shows the correlation between T10 values (temperatures at which 10 % of the droplets froze) in the microlayer and bulk seawater samples from 2016. We focus on T10 values to be consistent with our previous studies and because T10 values of the samples were at warmer temperatures than the field blanks in almost all cases. Pearson correlation analysis was applied to compute correlation coefficients (r). P values were also calculated to determine the significance of the correlations at the 95 % confidence level (p<0.05). A strong positive correlation (r=0.89 and p<0.001) was observed between the T10 values of the microlayer and the T10 values of the bulk seawater, consistent with our previous observations (Irish et al., 2017).

In 2016, four out of nine samples had warmer T10 values in the microlayer compared to bulk seawater (Fig. S3). However, in the 2014 samples, only one out of eight samples had warmer T10 values in the microlayer compared to bulk seawater (Irish et al., 2017). The difference between 2016 and 2014 may simply be due to year-to-year variations in the properties of the microlayer relative to the bulk seawater related to variations in oceanic conditions. For example, Collins et al. (2017) documented differences in the activity of marine microbial communities between our 2016 and 2014 campaigns in the Canadian Arctic. In addition, the differences between 2016 and 2014 may be related to sampling techniques. In 2014 the glass plate technique collected a layer that was up to 220 µm thick. In contrast, in 2016 a thinner layer (approximately 80 µm thick) was collected. In the thicker layers collected in 2014, the microlayer INPs would have been diluted by bulk waters by roughly a factor of 2.8. Additional studies of how INP activity varies as a function of microlayer sample thickness are necessary to resolve this issue.

Figure S4 shows that the fraction frozen curves were shifted to colder temperatures after the microlayer and bulk seawater samples were heated to 100 ∘C. These results are similar to what we observed for the 2014 samples (Irish et al., 2017). This suggests that the ice-active material we found in the microlayer and bulk seawater samples was likely heat-labile biological material (Christner et al., 2008).

Figure S5 shows that the temperature at which droplets froze in microlayer and bulk seawater samples significantly decreased after the samples were passed through a 0.02 µm filter but not through 10 µm or 0.22 µm filters. A similar result was observed in the 2014 samples (Irish et al., 2017), suggesting that the INPs in the microlayer and bulk seawater were between 0.22 and 0.02 µm in size.

The spatial distributions of T10 values for bulk seawater samples in both 2016 and 2014 are shown in Fig. 4. The spatial distributions are similar for microlayer samples (Fig. S6). In each panel the colour scales have been adjusted to easily compare the general pattern of T10 values between years. For both 2014 and 2016, the T10 values for samples taken from northern Baffin Bay and Nares Strait between Greenland and Canada, above 75∘ N, are generally lower than the T10 values elsewhere. To further investigate the similarities in spatial patterns between 2014 and 2016, we compared T10 values at sampling sites in close proximity for the two years (Fig. 5a). A strong positive correlation (r=0.93, p<0.001) was found between the T10 values measured in 2014 and T10 values measured in 2016 at those proximal locations (Fig. 5b), suggesting that the general spatial distributions of T10 values measured in 2014 and 2016 were similar even though warmer freezing temperatures were observed in 2016 compared to 2014 (Fig. 3).

In Table 2 and Fig. S7, we present correlations between T10 values for bulk seawater in 2016 and heterotrophic bacterial abundance, phytoplankton (including 0.2–20 µm photosynthetic eukaryotes and cyanobacteria) abundance, salinity, and temperature. The strongest correlation was with salinity (r=-0.83, p≤0.001). Similar correlations were observed for T50 values (Table S2). One possible explanation for the negative correlation between T10 values and salinity is a non-colligative effect of sea salt on the freezing temperature. For example, solutes can impact freezing temperature by blocking INP active sites (Kumar et al., 2018). To test this hypothesis, we varied the salinity in one of the microlayer samples (station 4) by adding commercial sea salt (Instant Ocean^™) while keeping the concentration of ice-nucleating material in the samples constant (see Supplement Sect. S1 for more details). As the salinity of the sample was increased from 29 to 55 g kg-1, the T10 values for the salinity-enhanced samples (after correcting for freezing point depression) varied by less than the uncertainty in the measurements (Fig. S8 in the Supplement). These results suggest that sea salt does not have a non-colligative effect on the freezing temperature of the samples, at least not for the microlayer sample tested (station 4). Consistent with these results, non-colligative effects have not been observed in previous studies of immersion freezing with seawater and sodium chloride solutions (Alpert et al., 2011a, b; Knopf et al., 2011; Wilson et al., 2015; Zobrist et al., 2008). Non-colligative effects have been observed in immersion freezing studies with ammonium-containing salts, but these results are not likely relevant for seawater solutions (Whale et al., 2018).

As suggested in our earlier study (Irish et al., 2017), another possible explanation for the negative correlation between salinity and freezing temperature is that the INPs are associated with either sea-ice melt or terrestrial runoff (including that from melting glaciers or permafrost). Melting sea ice and terrestrial runoff have lower salinities than seawater. In addition, sea-ice melt and terrestrial runoff often contain microorganisms and their exudates, which can be especially effective INPs (Assmy et al., 2013; Boetius et al., 2015; Christner et al., 2008; Ewert and Deming, 2013; Fernández-Méndez et al., 2014). Terrestrial runoff could also enhance the production of INPs in the ocean by providing additional nutrients for the growth of marine microorganisms.

Figure 6 shows the T10 values of bulk seawater as a function of the water volume fraction of meteoric water (fMW) and water volume fraction of sea-ice melt (fSIM) calculated using Eqs. (3)–(5). A strong positive correlation (r=0.91, p<0.001) was observed between T10 and fMW in the samples. In contrast, the correlation between T10 and fSIM in the samples was weaker, and the p value was close to 0.05 (r=0.63, p=0.048). These combined results suggest that meteoric water (terrestrial runoff plus precipitation) may be a major source of INPs in this area, or alternatively meteoric water may be enhancing INPs in this area by providing additional nutrients for the production of marine microorganisms. Terrestrial runoff has also been identified as a major source of INPs in temperate rivers and lakes (Knackstedt et al., 2018; Larsen et al., 2017; Moffett et al., 2018).

Figure 7 shows correlations between the chlorophyll data retrieved from GlobColour and the T10 values for the microlayer and bulk seawater. The correlations between T10 values in the microlayer or bulk seawater and chlorophyll a are not statistically significant. Figure S9 shows the relationship between the measured chlorophyll a concentrations and the T10 values for the microlayer and bulk seawater. Again, the correlations are not statistically significant. Our results from satellite and our measured chlorophyll a data are consistent with recent work by Wang et al. (2015), who showed that INP concentrations in sea spray aerosol emitted during a mesocosm tank experiment were not simply coupled to chlorophyll a concentrations.

In the following, we provide an initial estimate of the concentration of INPs in the Arctic marine boundary layer based on our freezing results. First, we calculated the concentration of INPs in the liquid per unit mass of sea salt, [INP(T)]mass,salt, using the following equation: 6[INPT]mass,salt=-ln⁡NuTNoNo1VρS, where ρ is the density of water, and S is the salinity of the seawater. Plots of [INP(T)]mass,salt as a function of temperature calculated from our freezing results are shown in Fig. S10. From [INP(T)]mass,salt, the concentration of INPs per unit volume of air in the Arctic marine boundary layer, [INP(T)]vol,air was estimated with following equation: 7[INPT]vol,air=[INPT]mass,saltC, where C is the concentration of sea salt in the Arctic marine boundary layer. Average concentrations of sea salt at Utqiaġvik (formerly Barrow), Alaska (71.3∘ N, 156.6∘ W), Alert, Nunavut, Canada (82.5∘ N, 62.5∘ W), and Zeppelin, Svalbard, Norway (78.9∘ N, 11.9∘ E) are 1.5, 0.1, and 0.6 µgm-3 in July and 1.4, 0.1, and 0.5 µgm-3 in August, respectively (Huang et al., 2018a). For these exploratory calculations we used a value of 1 µgm-3, which is within the range of the concentrations mentioned above.

Shown in Fig. 8 are estimated values for [INP(T)]vol,air based on our freezing data and a concentration of sea salt in the Arctic marine boundary layer of 1 µgm-3. Based on our freezing data, [INP(T)]vol,air ranges from ∼10-4 to <10-6 L-1 for freezing temperatures ranging from -5 to -10 ∘C. Over this temperature range, the highest estimated values for [INP(T)]vol,air were associated with two microlayer samples, and only these two microlayer samples resulted in [INP(T)]vol,air values as high as observed in direct atmospheric measurements of [INP(T)]vol,air in the marine boundary layer (Fig. 8) (DeMott et al., 2016; Irish et al., 2019). For freezing temperatures ranging from -15 to -25 ∘C, our estimated [INP(T)]vol,air values range from >10-4 to <10-6 L-1. Over this temperature range, many of our samples result in [INP(T)]vol,air values much less than observed in direct atmospheric measurements (Fig. 8). However, since our estimated [INP(T)]vol,air values are limited to 2×10-6 L-1, we cannot determine if our most active samples give [INP(T)]vol,air values similar to direct atmospheric measurements for freezing temperatures of -15 to -25 ∘C. The following caveats should be kept in mind when interpreting Fig. 8: first, we did not consider the possible enrichment of INPs in sea salt aerosols compared to the microlayer or bulk seawater samples, which can result from the bubble bursting mechanism. Second, the concentrations of sea salt used to estimate [INP(T)]vol,air were likely an upper limit based on the previous measurements at Utqiaġvik (formerly Barrow), Alert, and Zeppelin.