Sampling was conducted at the Dr. Neil Trivett Global Atmosphere Watch Observatory (82.5∘ N, 62.5∘ W) at Alert, Nunavut (Fig. 1), a research laboratory operated by Environment and Climate Change Canada. The research laboratory is on a plateau that is approximately 185 m a.s.l. (above sea level) and is powered by electricity generated by the Canadian Forces Station Alert, which is 6 km to the north of the research laboratory. The population at Alert is about 75 on a regular basis. The closest community is Grise Fiord (population 129) located approximately 800 km to the south of Alert.

Concentrations of INPs were determined by first collecting particles on hydrophobic glass slides with an inertia impactor followed by determining the freezing temperatures of the collected aerosol particles with the droplet freezing technique (DeMott et al., 2016; Irish et al., 2019). Details are given below.

Local ground-level meteorological conditions were monitored at the site by Environment and Climate Change Canada. The March 2016 data were retrieved from http://climate.weather.gc.ca/ (last access: 28 August 2018) (climate IDs 2400305 and 2400306). The ambient temperature, ambient RH, wind speed, and wind direction were measured hourly. The sum of the total rainfall and the water equivalent of the total snowfall in millimetres was measured daily as total precipitation.

The elements Al, Fe, Ti, and Mn were used as tracers of mineral dust, as done previously (Balasubramanian et al., 2003; Barrie and Barrie, 1990; Formenti et al., 2003; Malm et al., 1994; Quinn et al., 2004). These elements are components of the Earth's crust (Usher et al., 2003; Wedepohl, 1995). The species Na+ and Cl-, which are the major inorganic components of seawater (Holland, 1978), were used as tracers of sea spray, as done previously (Balasubramanian et al., 2003; Malm et al., 1994; Quinn et al., 2002, 2004). For tracers of anthropogenic aerosol, we used Zn, Pb, NO3-, NH4+, and non-sea-salt SO42- (nss-SO42-). Pb and Zn are almost exclusively from anthropogenic sources (Macdonald et al., 2000; Nriagu and Pacyna, 1988; Pacyna, 1995). The major anthropogenic sources for Pb are gasoline combustion and, to a lesser extent, non-ferrous metal industry and fossil fuel combustion, and the major anthropogenic sources for Zn are non-ferrous metal industry followed by fossil fuel combustion (Barrie et al., 1992; Nriagu and Pacyna, 1988; Pacyna, 1995). NO3-, NH4+, and nss-SO42- can come from both anthropogenic and natural sources, but mostly from anthropogenic sources. NO3- is mainly formed from NOx, which is emitted from combustion processes (Seinfeld and Pandis, 2006). NH4+ originates mainly from agricultural activities (Follett and Hatfield, 2001). The main anthropogenic source of nss-SO42- is fossil fuel combustion (Schwikowski et al., 1999; Ward, 2009).

To determine the concentrations of the tracers discussed above, aerosol particulate samples were collected on 20×25 cm Whatman-41 quartz filters daily using a high-volume sampler (Barrie et al., 1981, 1989), which was located approximately 500 m away from the laboratory on the ground. The face velocity of sampling (50 cm s-1) and typical filter loadings ensured collection efficiencies better than 95 % (Watts et al., 1987). The average sampling time was 23.5 h except for the first sample, which was collected over 2 days due to a storm making it difficult to change the filter. Hence, collection times for tracer measurements (∼24 h) were different than the collection times for INP measurements (∼2 h). The implications of the different collection times are discussed in Sect. 3.3. For the filter samples, the average total air volume sampled was roughly 2300 m3 at standard conditions of 1 atm pressure and 0 ∘C. The precision of volume sampled was estimated to be ±5 % (Sirois and Barrie, 1999). The quartz filter samples were stored at room temperature before analysis.

The concentrations of Al, Fe, Ti, Mn, Zn, and Pb were determined using inductively coupled plasma atomic emission spectroscopy (ICP-AES). These experiments were carried out at Chester LabNet, Oregon, USA. Punches from the quartz filters were submerged in a solution containing ultrapure HNO3 and HCl. The solution was then heated and sonicated, and then further diluted and filtered before being nebulized and analyzed by the ICP-AES (Perkin-Elmer Optima 8300). Sample duplicates were also analyzed to estimate the method precision. The accuracy and precision of the technique were estimated to be ±10 %.

The concentrations of Cl-, Na+, NO3-, SO42-, and NH4+ were determined using ion chromatography (IC) (Macdonald et al., 2017; Toom-Sauntry and Barrie, 2002). These experiments were carried out at Environment and Climate Change Canada in Ontario, Canada. To quantify water-soluble cations and anions, punches taken from each quartz filter were extracted in deionized water, and the extraction solution was passed through an IC (Dionex IC: DX600). The concentration of nss-SO42- was calculated using the following equation based on the assumption that the chemical composition of sea salt particles is the same as that of seawater (Millero, 1974): nss-SO42-=SO42--0.14Cl-. To determine whether the calculated concentrations of nss-SO42- were sensitive to depletion of Cl- in the particles, we also calculated nss-SO42- using Na+ rather than Cl- (Balasubramanian, 2003). The difference between the concentrations of nss-SO42- based on Cl- and the concentrations based on Na+ was less than 5 %.

The source regions of measured air masses were investigated using the Lagrangian FLEXible PARTicle dispersion model (FLEXPART) (Stohl et al., 2005), which was driven using operational meteorological analyses from the European Centre for Medium-Range Weather Forecasts (ECMWF). FLEXPART was run at hourly intervals in the backward mode for each sample collected for INP analysis. In each run, 40 000 particles were released over 1 min in a 0.1∘ × 0.1∘ area. The particles were followed backward for 20 days with output generated at 1 h intervals. In backward mode, FLEXPART provides potential emission sensitivities as output. This is the response function of a source–receptor matrix (Seibert and Frank, 2004) and corresponds directly to the residence time of the released particles in a given volume of air. For each run, the hourly output was integrated over the 20 days to produce a potential emission sensitivity (PES) plot. For a given INP sample, the mean PES plot was generated by averaging all hourly PES plots. Since the focus of this study is INP sources close to the surface, near-surface PES plots (from the surface up to 100 m) were plotted as footprint PES plots.

To compute a correlation coefficient (R), Pearson correlation analysis was applied between INPs and the variables measured in this study. The P values were also calculated using a t test for Student's t distribution to determine if the correlations were statistically significant at the 95 % confidence level (P<0.05).

Concentrations of INPs in the immersion mode are plotted as a function of temperature in Fig. 2. Concentrations of INPs measured were higher than concentrations of the blanks. In addition, all of the samples have warmer onset temperatures than the blanks, with onset temperatures of the samples varying from approximately -11 to -22 ∘C. For the remainder of this document, we focus on INP concentrations at -15, -20, and -25 ∘C. INP concentrations at temperatures warmer than -15 ∘C are not discussed because freezing events at these temperatures were rare. INP concentrations at temperatures below -25 ∘C are not discussed since freezing of the blanks became significant at these temperatures. Since freezing was rarely observed at ≥-25 ∘C in the blank experiments, the INP concentrations determined in the blank experiments were not subtracted from the INP concentrations reported in this study.

In the current study, the mean number concentration of INPs was 0.005±0.002 L-1 at -15 ∘C, 0.020±0.004 L-1 at -20 ∘C, and 0.186±0.040 L-1 at -25 ∘C (Fig. 3). These concentrations are within the range of INP concentrations measured in previous studies at ground level or sea level in the Arctic (Fig. 3). The sampling platform, location, and dates of previous Arctic INP studies at ground level or sea level that are shown in Fig. 3 are summarized in Table 1 and Fig. 1 for comparison purposes.

The time series of INP concentrations measured in the current study at -15, -20, and -25 ∘C is plotted in Fig. 4. The difference between the highest INP concentration at -25 ∘C (measured on 28 March) and the lowest INP concentration at -25 ∘C (measured on 17 March) was roughly a factor of 50.

Shown in Fig. 5 is the time series of meteorological parameters measured at Alert during the field campaign. Precipitation was rare throughout the campaign, and the average ambient temperature was approximately -30 ∘C. The RH was above 70 % for most of the time. The wind speed was below 10 km h-1 except for two storm events during the first part of the campaign. During the field campaign, the wind came mainly from the SW (more than half of the time) with some contribution from the SE and NW.

In the current study, INP concentrations at freezing temperatures of -15, -20, and -25 ∘C were not correlated with any meteorological parameters (Table S1 in the Supplement). This result is consistent with the results from Fountain and Ohtake (1985), who also did not find any correlations between meteorological variables (including air temperature, precipitation, fog, and wind direction, etc.) and INP concentrations measured in Utqiaġvik (formerly Barrow), Alaska, from August 1978 to April 1979. In contrast, measurements by Radke et al. (1976) in Utqiaġvik, Alaska, during March found that the INP concentrations were affected by local weather conditions.

The time series of concentrations of mineral dust tracers (Al, Fe, Ti, and Mn) is plotted in Fig. 6. Figure 6 shows that concentrations of different tracers of mineral dust are correlated with each other. Shown in Table 2 is a correlation coefficient matrix between each of the tracers based on a Pearson correlation analysis. Correlations that are both strong (R≥0.7) and statistically significant (P<0.05) are highlighted in bold. The correlation coefficients show that the mineral dust tracers are strongly correlated with each other (R≥0.85), which is expected since these elements are almost exclusively from the mineral dust sources.

In Fig. 7a, we compared the mean mass concentration of aluminum we measured at the site during the campaign with the concentrations reported in Sirois and Barrie (1999) for the same site and for the time period from 1980 to 1995. The mean mass concentration of aluminum measured during the current study was lower than the previous concentrations measured in March at Alert, indicating that the concentration of mineral dust during the current campaign was lower than many of the previous measurements at Alert during March.

Pearson regression analysis was also done between INP concentrations at -15, -20 and -25 ∘C, and tracers of mineral dust. The correlation coefficients and P values are summarized in Table 3. The linear regression plots between INP concentrations at -25 ∘C and mineral dust tracers are shown in Fig. 8. When calculating the correlation coefficients, a value of zero was used when tracer concentrations and INP concentrations were below the detection limits. It is important to keep in mind that the collection time was different for INPs (∼2 h) and the aerosol tracers (∼24 h), and that the analysis presented here is based on the assumption that the average concentrations of the aerosol tracers during the INP sampling time were the same as the average concentrations determined from the quartz filters. If this assumption is not correct, correlation coefficients between INPs and aerosol tracers will be smaller than expected. Shown in Fig. S1 in the Supplement are air mass back trajectories initiated every 2 h during each 24 h quartz filter sampling period. The back trajectories suggest that, for a large fraction of the samples, the source of the air masses during INP sampling was similar to the source of the air masses measured during quartz filter sampling.

As shown in Table 3, at -15 ∘C, the correlations between INP concentrations and tracers of mineral dust were not statistically significant (P>0.05). At -20 ∘C, the correlations between INPs and Ti and Mn were not statistically significant. Al and Fe were moderately correlated with INPs at -20 ∘C (R=0.53 and 0.60, respectively), and the correlations were statistically significant (P<0.05). At -25 ∘C, all four tracers were strongly correlated with INPs (R ranged from 0.70 to 0.76), and the correlations were statistically significant. This suggests that mineral dust was a component of the sampled INPs at a freezing temperature of -25 ∘C. This is consistent with previous field measurements that have identified mineral dust as a major component of the INP population at different locations (Boose et al., 2016; Cziczo et al., 2013; DeMott et al., 2003; Klein et al., 2010; Pratt et al., 2009; Prenni et al., 2009). A previous study of ice nucleation of Arctic aerosol found that the elements of crustal or natural dust were associated with high concentrations of INPs at -15 and -25 ∘C (Borys, 1989). Chemical analysis of the INPs during aircraft studies in the Arctic indicated that a significant fraction of the INPs (approximately 37 %) contained mineral dust (Rogers et al., 2001). Another study of snow crystals during summer on the Greenland ice cap also suggested that the natural snow crystals mainly formed on clay mineral particles by heterogeneous nucleation (Kumai and Francis, 1962).

Mineral dust at Alert can come from both local sources and long-range transport, but during the spring local sources are not likely important since land is covered with snow during this time of the year. To investigate the origins of mineral dust measured in this study, the FLEXPART model was used to generate the footprint PES plots, which show the residence time of aerosol particles in the layer from 0 to 100 m in altitude during the 20 days prior to sampling. Results are shown in Fig. 9. Figure 9a shows the surface coverage type on the first day of sampling (12 March). The sampling location at Alert was surrounded by ice and snow during the campaign. Figure 9b shows the average footprint PES plot for all samples, and Fig. 9c shows the average footprint PES plot for the four samples with the highest mineral dust concentrations, which were collected on 21, 22, 27, and 28 March. Note that since average footprint PES plots are shown in Fig. 9b and c, Fig. 9c is not a subset of Fig. 9b. Shown in Fig. 9d is a ratio plot of the sum of footprint PES plots of the four samples with the highest mineral dust concentrations to the sum of footprint PES plots of all samples. These types of ratio plots are often used as a sensitive method to identify the source regions of a component under investigation (Hirdman et al., 2010). In these types of plots, a value close to 1 indicates a more likely source region. Figure 9d suggests that the north Pacific Ocean, Alaska, and the Gobi Desert were possible source regions of the mineral dust sampled. Alaska was covered in snow during the sampling period, so this was not likely the source of the mineral dust, although mineral dust resuspended from blowing snow cannot be ruled out. Mineral dust from shorelines also cannot be ruled out. The north Pacific Ocean is not likely the source of mineral dust either since the Pacific Ocean is not considered as a traditional source of mineral dust. Based on the surface coverage types (Fig. 9a) and the ratio plot (Fig. 9d), a possible source of the mineral dust was long-range transport of dust from the Gobi Desert. This conclusion is consistent with previous studies that have shown that a substantial fraction of the dust reaching Alert in spring months comes from long-range transport of Asian dust (Drab et al., 2002; Franzén et al., 1994; Pacyna and Ottar, 1989; Sirois and Barrie, 1999; Welch et al., 1991).

The time series of sea spray aerosol tracers (Na+ and Cl-) is shown in Fig. 10, and the results of the Pearson correlation analysis between these tracers are listed in Table 2. The correlation coefficients show that sea spray tracers are strongly correlated with each other (R=0.95), which is expected since these species are almost exclusively from sea spray.

A comparison between the mean mass concentrations of Na+ and Cl- measured in the current study and the values measured by Sirois and Barrie (1999) from 1980 to 1995 is shown in Fig. 7b and c. Figure 7b and c illustrate that the concentrations of Na+ and Cl- measured in the current study were consistent with many of previous measurements at Alert during March.

The correlation coefficients between INP concentrations at -15, -20, and -25 ∘C, and tracers of sea spray aerosol are listed in Table 3. The linear regression plots between INP concentrations at -25 ∘C and the sea spray tracers are shown in Fig. 11. At -15 and -20 ∘C, the correlations between INP concentrations and tracers of sea spray aerosol were not statistically significant (P>0.05). At -25 ∘C, the sea spray tracers were negatively correlated with INPs with moderate correlation coefficients (R=-0.56) and statistical significance (P<0.05). Previous field studies and modelling studies have suggested sea spray aerosol as an important source of ambient INPs in marine environments when other sources of INPs, such as mineral dust, are low (Burrows et al., 2013; DeMott et al., 2016; Rosinski et al., 1986, 1988; Schnell, 1982; Vergara-Temprado et al., 2017; Wilson et al., 2015). Our results suggest that mineral dust is a more important source of INPs at -25 ∘C than sea spray aerosol for the time and location studied.

Due to the method of particle collection and the method of forming liquid droplets in the freezing experiments, soluble material, such as salts including NaCl, was mixed with mineral dust particles within the same droplets during the freezing experiments. Salts decrease the freezing temperature of droplets by decreasing the water activity in the solution. However, the decrease in freezing temperature by this mechanism is not expected to be important in our experiments since the concentration of salts within a droplet is small. For example, the maximum concentration of NaCl in the water droplets formed during the droplet freezing experiments was estimated to be ∼0.03 M based on the mass concentrations of Na+ and Cl- sampled. This concentration of NaCl in the droplets would only cause a freezing point depression of ∼0.1 ∘C, which is within the uncertainty of the measured freezing temperatures and is too small to explain the negative correlation between INP concentrations and tracers of sea spray aerosol shown in Fig. 11.

On the other hand, other studies have illustrated that trace amounts of NaCl can lower the freezing temperature of mineral dust by surface-specific interactions between NaCl and the mineral dust surface, and the decrease in freezing temperature by this mechanism is more than expected based on the traditional freezing point depression mechanism (Reischel and Vali, 1975; Whale et al., 2018). For example, Reischel and Vali (1975) studied the effects of 0.01, 0.1, and 1 M solutions of NaCl on the nucleating ability of kaolin and found that the presence of NaCl led to lower freezing temperatures, by as much as 4 ∘C, for kaolin. A recent study by Whale et al. (2018) also found that a 0.015 M NaCl solution caused a decrease in freezing temperature of approximately 2 to 4 ∘C for BCS376 microcline, Eifel sanidine, quartz, and Arizona test dust, but had no effect on silica. To investigate the influence of NaCl on the ice-nucleating ability of mineral dust at -25 ∘C in the current study, INP concentrations at a freezing temperature of -25 ∘C ([INPs]-25∘C) were divided by the mass concentration of aluminum ([Al]), and this ratio was then plotted as a function of Cl- and Na+ (Fig. 12). The variable [INPs]-25∘C/[Al] is used as an estimation of the ice-nucleating ability of mineral dust, and it was negatively correlated with Cl- and Na+ with moderate correlation coefficients (R=-0.51 and -0.55, respectively) and statistical significance (P<0.05), which suggests that NaCl suppressed the ice-nucleating ability of mineral dust in the current study. This result is consistent with the two previous laboratory studies mentioned above and is the first field study that we are aware of that suggests that NaCl can suppress the ice-nucleating ability of mineral dust particles in the immersion freezing mode even in dilute solution droplets.

Another possible explanation for the negative correlation between INP concentrations at -25 ∘C and sea spray tracers is that the air masses containing sea spray aerosol have relatively few mineral dust particles. In this case, the sea spray aerosol does not influence the ice-nucleating ability of mineral dust but rather appears as a negative correlation by coincidence. However, Table 2 shows that the correlations between tracers of mineral dust and tracers of sea spray aerosol are not statistically significant, suggesting that this possible explanation is not the major reason for the negative correlation between INP concentrations at -25 ∘C and sea spray tracers.

The time series of anthropogenic aerosol tracers (Zn, Pb, NO3-, NH4+, and nss-SO42-) is plotted in Fig. 13. The results of the Pearson correlation analysis between these different tracers are listed in Table 2. Not all the anthropogenic aerosol tracers are correlated with each other. This is not surprising since these tracers come from different sources. However, there is a correlation between Zn and Pb, and between nss-SO42- and NH4+. Zn and Pb are strongly correlated with each other (R=0.84, P<0.01), and nss-SO42- and NH4+ are strongly correlated with each other (R=0.93, P<0.01).

The comparison between the mean mass concentrations of Zn, Pb, NO3-, SO42-, and NH4+ measured in the current study and the values measured by Sirois and Barrie (1999) from 1980 to 1995 is shown in Fig. 7d–h. The concentrations of Zn, Pb, SO42-, and NH4+ were lower than previous concentrations measured in March at Alert, which might be due to the reduced emissions from Eurasia after the dissolution of the former USSR in 1991 (Christensen, 1997). The concentration of NO3- was higher than previous concentrations measured in March. An increasing trend of annual concentrations of NO3- has been observed by others in the Arctic (Hole et al., 2006; Neftel et al., 1985).

The correlation coefficients between INP concentrations at -15, -20, and -25 ∘C, and tracers of anthropogenic aerosol are listed in Table 3. None of the correlations were statistically significant, consistent with a recent study that showed no effect of atmospheric pollution on INP concentrations in Beijing at freezing temperatures down to -25 ∘C (Chen et al., 2018). In contrast, Borys (1989) found that pollution-derived Arctic haze aerosol had lower INP concentrations than unpolluted troposphere aerosol.

There have been a few studies that investigated the effect of sulfate coating on the ice-nucleating properties of mineral dust particles. Most laboratory studies have shown decreased freezing abilities for sulfate-coated mineral dust particles compared to uncoated particles (Chernoff and Bertram, 2010; Eastwood et al., 2008; Gallavardin et al., 2008; Sullivan et al., 2010). However, other studies have shown an increase in the freezing temperature of droplets containing mineral dust particles in the presence of ammonium sulfate (Reischel and Vali, 1975; Whale et al., 2018). To investigate the effect of ammonium, sulfate, and the ammonium-to-sulfate ratio on the ice-nucleating ability of mineral dust at -25 ∘C, the parameter [INPs]-25∘C/[Al] was plotted as a function of NH4+, nss-SO42-, and the NH4+ / nss-SO42- ratio as shown in Fig. 14. The NH4+ / nss-SO42- ratio was between 0 and 1. The correlations between the ice-nucleating ability of mineral dust at -25 ∘C and NH4+, nss-SO42-, and the NH4+ / nss-SO42- ratio were not statistically significant.