Ice nucleating particles in the marine boundary layer in the Canadian Arctic during summer 2014

. Ice nucleating


Introduction
Ice nucleating particles (INPs) initiate the heterogeneous formation of ice in clouds at temperatures warmer than required for homogeneous freezing. INPs are important since they can change the frequency and properties of ice and mixed-phase clouds. The frequency and properties of clouds in the Arctic have been shown to be especially sensitive to concentrations of INPs, yet our understanding of the concentrations and sources of INPs in this region remains uncertain (Coluzza et al., 2017;Creamean et al., 2018;Harrington et al., 1999;Kanji et al., 2017;Korolev et al., 2017).
Examples of atmospherically relevant INPs include, but are not limited to, mineral dust particles and sea spray aerosol Després et al., 2012;Hoose and Möhler, 2012;Murray et al., 2012;Niemand et al., 2012;Szyrmer and Zawadzki, 1997). Sea spray aerosol is generated by a bubble bursting mechanism at the ocean surface (Blanchard, 1964) and varies widely in composition, depending on the mechanism of formation. For example, particles from film rupture are enriched in organics compared to particles from jet droplets . Recent work has shown that the sea surface microlayer and bulk seawater contain INPs (Irish et al., 2017;Rosinski et al., 1986;Schnell, 1977;Schnell andVali, 1975, 1976;Wilson et al., 2015). Modelling studies have also suggested that sea spray aerosol can be a significant contributor to the INP population in marine environments when the source of other INPs is small (Burrows et al., 2013;Huang et al., 2018;Vergara-Temprado et al., 2017;Wilson et al., 2015). However, our understanding of the flux of INPs from the ocean to the atmosphere is incomplete, and more studies are needed to understand when and where sea spray aerosol is a significant contributor to INP concentrations in the atmosphere.
When glaciers and permafrost in the Arctic melt, erodible soil is exposed. The increased areas of erodible soil can be a potential source of mineral dust in the Arctic (Huang et al., 2015). The ice nucleation properties of dry mineral dust from Qaanaaq (also known as Thule), Greenland, were measured by Fenn and Weickmann (1959) who found that dry mineral dust could nucleate ice at temperatures as warm as −5 • C. Groot Zwaaftink et al. (2016) also suggested through a modelling study that during the summer local mineral dust sources in the Arctic (mineral dust from latitudes north of 60 • N) can dominate the total mineral dust concentrations at the surface. However, previous studies have not yet shown that mineral dust from regional erodible soil could be a major source of INPs in the atmosphere in the Arctic.
In the past 30 years, warming in the Arctic has decreased sea ice and land snow by approximately 20 % and 13 %, respectively (Derksen and Brown, 2012). This may have led to an increase in sea spray particles and mineral dust particles from local sources in the region and, as a result, an increase in INPs. Because of the continuing warming trend in this region, the concentration of INPs from these local sources may continue to increase with important implications for the frequency and properties of ice and mixed-phase clouds as well as climate in the region. To evaluate the scale of this feedback mechanism, studies are needed to determine the concentrations and sources of INPs in the Arctic.
To help address the issues raised above, we (1) determined the concentration of INPs in the immersion mode in the Canadian Arctic marine boundary layer during summer 2014, (2) measured the ratio of surface areas of mineral dust particles to sea spray particles, and (3) investigated the source region of the INPs using a particle dispersion model. The specific goals of this study were to quantify the concentrations of INPs in the Canadian Arctic marine boundary layer and to provide insights into the source of INPs in this region during the summer.

Sampling locations
All measurements and sample collections were performed on board the CCGS Amundsen as part of the Network on Climate and Aerosols: Addressing Key Uncertainties in Remote Canadian Environments (NETCARE). The 28 sampling locations are shown in Fig. 1, and the sampling dates, times, and coordinates are detailed in Table S1 in the Supplement. The air temperature, relative humidity, and wind speed during sampling were measured with on-board sensors at approximately 15 m above sea level and are shown in Fig. S1 in the Supplement. INP concentrations from a subset of the sampling locations (indicated with blue rings around the red symbols in Fig. 1) were previously reported in  but are also included here as they were collected during the same expedition and with the same methodology. The data reported in DeMott et al. (2016)  the samples were assumed to have reduced exposure to ship smokestack contamination, i.e. when the apparent wind direction measured on the ship was between 0 and 90 • or between 270 and 360 • , where 0 or 360 corresponds to the bow of the ship. (The apparent wind direction is defined as the wind direction experienced by an observer on the moving ship as opposed to the true wind direction, which is defined as the wind direction experienced by an observer when the ship is stationary.)

Quantifications of INPs
To determine the concentration of INPs in the immersion mode, atmospheric particles were collected on hydrophobic glass slides using a micro-orifice single-stage impactor (MOSSI; MSP corp., Shoreview, MN, USA). The freezing properties of the collected particles were then determined with the droplet freezing technique (DFT). The combination (MOSSI-DFT) is similar to the micro-orifice uniform deposit impactor droplet freezing technique (MOUDI-DFT) recently used to determine the size-resolved concentrations of INPs (Mason et al., 2015a(Mason et al., , b, 2016. The main difference between the MOSSI-DFT technique and the MOUDI-DFT technique is the use of a single-stage impactor compared to a multistage impactor. The use of a single-stage impactor simplifies the analysis and reduces collection time but sacrifices size information. The MOSSI-DFT technique is also similar to the technique used by others to measure deposition freezing (Knopf et al., 2010(Knopf et al., , 2014Wang et al., 2012a, b).

Micro-orifice single-stage impactor
The MOSSI was located on the port side of the bridge on the ship, approximately 10 m in front of the ship's smokestack. During sampling the flow rate through the MOSSI was 10 L min −1 , resulting in particles with aerodynamic diameters > 0.18 µm being collected on the hydrophobic glass slides placed within the MOSSI. The collection time of samples with the MOSSI for INP analysis was approximately 20 min. The MOSSI sampled air through a louvred total suspended particulate (TSP) inlet, which was approximately 15 m above sea level. The transmission efficiency of 10 µm particles through the TSP inlet is ≥ 90 % (Kenny et al., 2005). The nozzle plate within the MOSSI contained 300 micro-orifices. As a result, particles collected on the hydrophobic glass slides beneath the nozzle plate were concentrated into 300 spots. After collection the hydrophobic glass slides containing the particles were stored at 4 • C for no longer than 3 months prior to analysis. Particle rebound can be an issue with an inertial impactor such as the MOSSI. Particle rebound occurs when particles impact the collection substrate but are not retained. Rebound has been shown to be reduced at relative humidity (RH) values above 70 %, although this depends on the chemical composition of the particles (Bateman et al., 2013;Chen et al., 2011;Fang et al., 1991;Lawson, 1980;Saukko et al., 2012;Vasiliou et al., 1999;Winkler, 1974). During collection with the MOSSI, the RH was always well above 70 % ( Fig. S1) with an average of 93 %. Furthermore, field measurements of INP concentrations using the MOUDI-DFT (a method similar to the MOSSI-DFT) have shown good agreement with INP concentrations measured with an instrument that is not susceptible to particle rebound (a continuous-flow diffusion chamber) when the RH of the sampled aerosol was as low as 40 % Mason et al., 2016). Nevertheless, particle rebound cannot be ruled out, and therefore the INP concentrations reported here should be considered lower limits.

Droplet freezing technique
The droplet freezing technique (Koop et al., 1998;Mason et al., 2015a) was used to determine the concentration of INPs in the immersion mode collected on hydrophobic glass slides using the MOSSI. Briefly, the hydrophobic glass slides containing the collected particles were located in a temperature and RH-controlled flow cell, coupled to an optical microscope with a charge-coupled device (CCD) camera and a 1.25× objective (Axiolab; Zeiss, Oberkochen, Germany). Typically between 15 and 25 spots of particles (out of the 300 spots generated by the micro-orifices in the nozzle plate) could be monitored in the CCD field of view, which was approximately 12.25 mm 2 in area. Water was then condensed on the hydrophobic glass slides by decreasing the temperature to 2 • C and flowing a gas (pure helium) with a dew point of greater than 2 • C over the hydrophobic glass slides. This resulted in water droplets (with diameters between 100 and 500 µm) condensing on the spots (referred to here as spot droplets) as well as water droplets condensing on other areas of the slides (referred to here as non-spot droplets). After the droplets were condensed, the temperature of the flow cell was decreased at a rate of 10 • C min −1 . From videos recorded while the temperature was decreasing, the freezing temperature of each droplet was manually determined by observing the change in the droplet's optical properties. The droplets that contained spots of deposited particles were also identified from these videos. For comparison purposes, hydrophobic glass slides that were not exposed to atmospheric particles were also processed in the same way as described above and labelled as "blanks".
The number of INPs as a function of temperature, #INP(T ), was calculated for each experiment using the following equation: where N u is the number of unfrozen droplets covering the spots, and N s is the number of spots in the field of view. Equation (1) accounts for the possibility that each droplet covering a spot can contain multiple INPs (Vali, 1971). Equation (1) assumes that each spot was covered by only one droplet. For cases when more than one droplet formed on a spot, the first droplet that froze was considered in Eq. (1). This was expected to give a result equivalent to the case of only one droplet condensing on the spot. For cases when one droplet contained two spots (this occurred for 2 % of the total number of spots in all experiments), an upper limit to the number of INPs was calculated by assuming two droplets covered the two spots and both droplets froze at the observed freezing temperature. A lower limit was calculated by assuming the two spots were covered by two droplets with one droplet freezing at the observed freezing temperature, and the other droplet freezing at −37 • C (approximately the homogeneous freezing temperature). If one droplet contained three or more spots, a similar procedure to the above was used to calculate the upper and lower limits to #INP(T ).
The freezing of droplets that did not cover spots was a relatively rare occurrence at the temperature range we focused on in this manuscript (≥ −25 • C; see Sect. 3.1). For example, the ratios of frozen non-spot droplets to frozen spot droplets were 0.02 and 0.07 at −25 and −20 • C, respectively. We assumed these relatively rare occurrences were due to particles < 0.18 µm in diameter that were not focused into spots or due to rebound of a small fraction of the particles off the hydrophobic glass slides. To take into account the INPs not concentrated into the spots, we added the number of frozen nonspot droplets at each temperature to Eq. (1). Since the freezing of non-spot droplets was a relatively rare occurrence, we did not apply the Vali correction (Vali, 1971) to these freezing events.
Approximately 2 % of the freezing events in our experiments occurred by contact freezing between −16.2 and −34.8 • C. Contact freezing occurred when a frozen droplet grew in size, due to the Wegener-Bergeron-Findeisen process (Findeisen, 1938), and caused the freezing of a neighbouring unfrozen droplet. When calculating concentrations of INPs, contact freezing events were excluded.
The atmospheric concentration of INPs as a function of temperature, [INP(T )], was calculated with the following equation: where V is the volume of air sampled, and 300/N s is the ratio which takes into account that only a fraction of the total number of spots in the sample was observed in a freezing experiment.

Effect of ship emissions on measured INP
To determine if particles from the ship's smokestack affected the measured INP concentrations, we first investigated the relationship between INP concentrations measured on the ship and the gas-phase HONO concentrations, a product of the reaction between NO x from the ship's smokestack and water (von Glasow et al., 2003). HONO was measured by a chemical ionization mass spectrometer located on the bridge of the ship about 5 m in front of the smokestack and approximately 15 m above sea level. No correlation was observed between HONO and INP concentrations at −25 • C (R = 0.05, p = 0.403). Second, we separated our INP results into samples that were not exposed to smokestack emissions based on measured wind direction and wind speed, and samples that may have been exposed to smokestack emissions based on measured wind direction and wind speed. When the apparent wind direction measured on the ship was between 0 and 90 • or between 270 and 360 • (where 0 or 360 • marks the bow of the ship) and when the apparent wind speed (minute averaged) was higher than 2.5 m s −1 for the entire collection time, we assumed that the samples were not exposed to smokestack emissions. Within the uncertainty of the measurements, the INP concentrations measured when samples were not exposed to smokestack emissions (based on the apparent wind direction and speed) are the same as INP concentrations measured when samples may have been exposed to smokestack emissions (Fig. S2). Since the criteria discussed above and the results from the HONO analysis do not suggest measured INP concentrations were influenced by the smokestack emissions, all samples that were collected have been included in this study.

Particle dispersion modelling
FLEXPART-WRF (Brioude et al., 2013), a version of the Lagrangian particle dispersion model FLEXPART (Stohl et (Skamarock et al., 2008), and was run in backward mode. The simulation domain for FLEXPART-WRF is shown in Fig. S3. WRF 3.5.1 was run for the 2014 Amundsen Campaign using initial and boundary conditions from the European Centre for Medium-Range Weather Forecasts (ECMWF) operational analysis (a grid resolution of 0.25 • ). The ECMWF wind, temperature, and RH were used to nudge the WRF run every 6 h above the atmospheric boundary layer. A full list of the parameterizations and options used for the WRF simulations is given in Table 1 of Wentworth et al. (2016).
FLEXPART-WRF was run in backward mode at 20 min intervals along the ship track. At the start of a simulation 100 000 particles were released from the ship's location in an initial box of volume of 100 m × 100 m in the horizontal, and from 0 to 60 m above mean sea level in the vertical. After the particles were released from the initial box, they were followed backward for 7 days with output generated hourly.
As mentioned above, each INP sample was collected over a period of approximately 20 min. As a result, one FLEXPART-WRF run overlapped with each INP sampling period. The FLEXPART-WRF runs that overlapped in time with the INP sampling periods were used to produce potential emission sensitivity (PES) plots for the INP samples. A PES plot was produced by integrating the FLEXPART-WRF output over the 7 days prior to the release of particles. The value of the PES in a particular grid cell is proportional to the particles' residence time in that cell. Since this study is concerned with INP sources from the surface, only particles within the footprint layer (0 to 300 m altitude) are considered when calculating PES values, and we report PES values for the footprint layer.

Statistical analyses
Pearson correlation analysis was used to compute correlation coefficients (R). P values were also calculated to determine if the correlations were statistically significant at the 95 % confidence level (p < 0.05).

Computer controlled scanning electron microscopy
with energy dispersive X-ray spectroscopy (CCSEM-EDX) Immediately following the collection of each sample for INP analysis, additional particle samples were collected to determine the ratio of mineral dust surface area to sea salt surface area by CCSEM-EDX. Particles were collected on transmission electron microscopy (TEM) grids (carbon 200 mesh; Ted Pella) using the same MOSSI used to collect INP samples. Collection time of samples for CCSEM-EDX was approximately 20 min. The samples were kept at room temper-ature for a maximum of 38 months before analysis. Since a long collection time was used (20 min), particles in the spots directly below the micro-orifices of the nozzle plate in the MOSSI were too close together to identify individual mineral dust and sea salt particles using CCSEM-EDX. To overcome this issue, we only analysed particles on the edge of the spots directly below the micro-orifices of the nozzle plate with CCSEM-EDX. Since the edge was still directly under the nozzle, we assumed that the composition at the edge of the spot was the same as the composition of the whole spot, although we are unable to confirm this assumption. Due to time constraints, only three samples (two with a high [INP(T )] and one with a low [INP(T )]) were analysed by CCSEM-EDX. The method of using CCSEM-EDX to study atmospheric particles is described by Laskin et al. (2006). Particles with sizes between 0.15 and 5 µm (area equivalent diameters) were analysed. First, the atomic percentages of each particle were quantitatively determined from EDX spectra. The EDX data for an individual particle represent an average signal from a raster scan. Then, based on the atomic percentages of each particle, particles were classified as sea salt, mineral dust, or other using the scheme shown in Fig. S4, which is based on the work by Laskin et al. (2012). This classification scheme does not include a mixed mineral dust/sea salt particle category. Rather, any particles that had both sea salt and mineral dust were classified as either mineral dust or sea salt depending on the largest atomic percentage contribution. If we used the classification scheme by Derimian et al. (2017), which explicitly classifies particles as sea salt, mineral dust, and mixed sea salt and mineral dust, no particles would be classified as mixed sea salt and mineral dust particles.
After each particle was classified, the surface areas of mineral dust particles and sea salt particles were determined using 2-D projected images recorded by SEM. Note that the actual surface area of mineral dust is underestimated using this method due to surface irregularities and complex topology. The ratio of mineral dust surface area to sea salt surface area was then determined by dividing the surface area of mineral dust for each sample by the surface area of sea salt for the same sample.

Measured INP concentrations
The measured concentrations of INPs in the immersion mode, [INP(T )], sampled in the Arctic are shown in Fig. 2b and c. The measured [INP(T )] on new hydrophobic glass slides taken straight from the package, cleaned with ultra pure water, but not exposed to atmospheric particles (referred to as blanks) are shown in red in Fig. 2a and c. Freezing of the blanks occurred over the range of −25.9 to −38.4 • C. For the droplet sizes and cooling rates used here, homoge- . Each blank was performed on a separate hydrophobic glass substrate. Error bars are not shown to improve the visibility of the data in the graph. Error bars in the x direction are ±0.3 • C for each datapoint. Error bars in the y direction were calculated using nucleation statistics following Koop et al. (1997); the errors for our measured [INP] (L −1 ) can be seen in Fig. S5. neous freezing occurs at approximately −37 • C. Therefore, the freezing that occurred in the blanks at temperatures above approximately −37 • C was due to heterogeneous freezing likely caused by the hydrophobic glass slides. In the following we will focus on [INP(T )] at temperatures of −25 • C and warmer since no freezing from the blanks was observed in this temperature range. A full time series of [INP(T )] at −15, −20, and −25 • C are reported in Fig. S5.
In Fig. 3 we compare recent measurements of [INP(T )] from several field campaigns in marine environments with the average concentrations measured in the current study. Figure 3 illustrates that the average INP concentrations measured in the current study fall within the range of INP concentrations measured in other marine environments. This observation, however, does not confirm that sea spray aerosol was the major source of INPs during the studies reported here. The sample corresponding to a low [INP(T )] was collected on 29 July. This day was chosen randomly out of the samples that did not display any freezing activity above −25 • C.
In Table S2 we report the total number of particles analysed by CCSEM-EDX for each sample, the fraction of particles classified as mineral dust and sea salt particles, and the surface area corresponding to mineral dust and sea salt particles. Shown in Fig. 4a are the calculated ratios of mineral dust surface area to sea salt surface area using surface area measurements from CCSEM-EDX. For the three samples, this ratio ranged from 0.03 to 0.09. Using this ratio we esti- where n s (SS) is the ice active surface site density for sea spray aerosol, n s (MD) is the ice active surface site density for mineral dust, and S SS and S MD are the total surface ar-   Fig. 4b, c, and d, respectively. These ratios suggest that for the three samples when CCSEM-EDX measurements were performed, the [INP(T )] MD are higher than the [INP(T )] SS (ratios were between 10 and 10 3 , inclusive of errors, at −15, −20, and −25 • C), assuming the n s values used are applicable for the field studies reported here. These results also suggest that mineral dust is a more important contributor to the INP population than sea spray aerosol for the times and locations corresponding to the CCSEM-EDX measurements.
Above we assumed that the ice active surface site density for mineral dust, n s (MD), in our studies can be calculated with the exponential function reported by Niemand et al. (2012) that was determined from freezing data of Asian dust, Saharan dust, Canary Islands dust, and Israel dust. To test this assumption, we calculated n s (MD) values using the CCSEM-EDX measurements on 21 and 25 July. Details on how n s (MD) values were calculated is given in Sect. S2, and the results are shown in Fig. S6. In short, within the uncertainty of the measurements, our calculated n s (MD) values are consistent with the results from Niemand et al. (2012). Figure 5a shows the averaged PES for the footprint layer for all samples combined and suggests that the source of INPs sampled during the campaign was local (i.e. the Canadian Arctic Archipelago, Baffin Bay, and eastern Greenland). Figure 5b shows the averaged PES for the footprint layer for samples that had the highest INP concentrations (top 36 % of the samples) at −25 • C. Figure 5c shows the average PES for the footprint layer for samples that had the lowest INP concentrations (bottom 36 % of the samples) at −25 • C. A cut-off of 36 % was selected since no freezing was observed in 36 % of the samples at −25 • C. Figure 5d shows the ratio of the average PES for the highest INP concentrations to the average PES for all samples. Figure 5e shows the ratio of the average PES for the lowest INP concentrations to the average PES for all samples. Previous work has shown that ratios of average PES are useful for identifying likely sources (Hirdman et al., 2010). Considering all figures Bare land, open water, sea ice, and snow cover are shown as beige, dark blue, light blue, and white, respectively. Note that for this study, lakes are included in the bare land category. together, the highest INP concentrations are associated with lower-latitude regions such as the Hudson Bay area, eastern Greenland, or north-western continental Canada. On the other hand, the lowest concentrations (below the detection limit at −25 • C) were often associated with regions further north and over Baffin Bay. Figure 5f and g shows maps of surface cover type (i.e. bare land, open water, sea ice, and snow cover) from the first and last days of the campaign, respectively, based on data from the National Snow and Ice Data Center (NSIDC; National Snow and Ice Data Center, 2008). The maps of surface coverage were combined with the PES values in the footprint layer from FLEXPART to determine the total residence time over each surface type for a given INP sample. Specifically, surface coverage data from the NSIDC were downloaded in GEOtiff format and converted to vector shape files. The fraction of each FLEXPART grid cell that was over each surface type category (e.g. bare land, open water) was then calculated using these vector shape files. The residence time in a grid cell was then multiplied by the fraction of the cell in each surface type category, and then the results were summed over all grid cells to determine the relative time spent over each surface type.

Particle dispersion modelling
Correlations between the total residence time over each surface type and the concentration of INPs for each sample at −15, −20, and −25 • C were then investigated (Table 1 and Fig. S7). This correlation analysis showed statistically significant (p < 0.05) positive correlations between the total residence time over bare land in the footprint layer, and both    Bigg (1996) observed a correlation between INP concentrations measured in the high Arctic and the time since the sampled air mass was last in contact with open ocean (R = −0.54, p < 10 −4 ). In contrast, Bigg and Leck (2001) observed no correlation between INP concentrations and the time since the sampled air mass was last in contact with open ocean.

Conclusions
Concentrations of INPs in the marine boundary layer were measured at 28 different locations in the Canadian Arctic. Results showed that the concentrations of INPs are similar to concentrations measured in other marine environments. For three collected samples the ratio of mineral dust surface area to sea spray surface area ranged from 0.03 to 0.09. Based on these ratios, and the ice active surface site densities of mineral dust and sea spray aerosol determined in previous laboratory studies, mineral dust is a more important contributor to the INP population than sea spray aerosol for the samples analysed (ratios were between 10 and 10 3 , inclusive of errors, at −15, −20 and −25 • C). This result suggests that INPs from mineral dust are more important contributors to the INP population than sea spray aerosol for the times and locations during sampling.
Particle dispersion modelling suggested that the INPs sampled in this study were likely not from long-range transport. This correlation analysis together with the particle dispersion modelling provides further evidence that mineral dust is a more important contributor to the INP population than sea spray aerosol, at least when INP concentrations were high. Sea spray aerosol may still have played a role when the INP concentrations were low during sampling.
Previous studies have shown that INPs are ubiquitous in Arctic seawater (Irish et al., 2017;Wilson et al., 2015). Our results show that INPs in the Arctic seawater are not emitted at a high enough rate to compete with mineral dust, at least for the samples with high INP concentrations. The flux of sea spray aerosol to the atmosphere is a function of the wind speed. The minute averaged wind speed during sampling (5.5 m s −1 with 10th and 90th percentiles of 2.1 and 10.0 m s −1 , respectively) was similar to the minute averaged wind speed during the whole campaign (5.2 m s −1 with 10th and 90th percentiles of 1.5 and 9.3 m s −1 , respectively). In addition, our average wind speeds were similar to historical monthly averaged wind speed data from Alert, NU, Canada (3.6 m s −1 in July and 3.3 m s −1 in August; http://climate.weather.gc.ca; climate ID: 2400300, last access: 19 November 2018). The influence of sea spray aerosol may be more important during periods of higher wind speeds. On the other hand, high wind speeds are also likely to increase the flux of mineral dust from local sources. As warming increases in the Arctic, more erodible soil will be exposed for longer periods of time (Huang et al., 2015). These results, together with our freezing results, suggest that warming in the Arctic will increase concentrations of INPs from mineral dust in the region, with possible implications for cloud properties and climate. Additional studies, including modelling and field studies are needed to quantify the importance of this feedback process for climate in the region. Data availability. Underlying material and related items for this manuscript are located in the Supplement.
Author contributions. AKB, JPDA, and VEI conceptualised the research. VEI, SJH, and SC performed the research. JLT, MDW, JJBW, AC, WRL, JGM, AL, and EG provided additional data. VEI, SJH, and SC analysed the data. VEI, AKB, and SJH wrote the publication. All co-authors reviewed the paper. EG died before submission; we regard his approval for inclusion of his name on this paper as implicit.
Competing interests. The authors declare that they have no conflict of interest.
Special issue statement. This article is part of the special issue "NETCARE (Network on Aerosols and Climate: Addressing Key Uncertainties in Remote Canadian Environments) (ACP/AMT/BG inter-journal SI)". It is not associated with a conference.