The SHARK, shown in Fig. 1, comprises two cascade impactors and corresponding pumps, alongside an OPC (OPC-N2, Alphasense, UK) and radiosonde (S1H2-R, Windsond, Sweden), all mounted within a weatherproof enclosure with a tail fin to orient it into the wind. A photograph of the internal components of the SHARK are shown in Fig. 1b. The two cascade impactors were employed to collect particles across different size bins: Impactor 1 from 0.25 to 2.5 µm and Impactor 2 from 1 to 10 µm. Impactor 1 is a cascade impactor (U.S. patent no. 6786105, Sioutas, SKC Ltd., UK), which requires a flow rate of 9 L min-1 and operates with a portable pump (Leland Legacy, SKC Ltd., UK). Impactor 2 is also a cascade impactor (MSP Model 128, TSI, USA), which requires a flow rate of 100 L min-1 at a pressure drop of 0.6 kPa (Marple et al., 1991; Misra et al., 2002), and for which a radial flow impeller (Radial Blower U51, Micronel, UK) was used in reverse as a lightweight pump (∼120 g). These pumps maintain the volumetric flow rate through the impactors as temperature and atmospheric pressure change with altitude. The pump for Impactor 1 was calibrated to apply this adjustment to at least 2.3 km (SKC, 2020), although the presence of the after-filter may reduce the battery life at this altitude. The pump for Impactor 2 is supplied by a larger battery and should be able to maintain flow to at least the same altitude as the Impactor 1 pump and over a longer period of time. The SHARK records the volume of air sampled through Impactor 1 during the flight, and so if the pump battery was depleted or the pressure drop became too great before Impactor 2 had finished sampling, the Impactor 1 pump would shut down and store the recorded value for later analysis. Further testing of the SHARK would be required to define a maximum altitude limit that each SHARK component could operate at. In order to provide RH, temperature, GPS altitude, and pressure data in real time, the sensors and transmitter from a radiosonde were integrated into the system. The OPC measured aerosol size distributions, which were saved in the onboard memory. Servo-controlled caps covered the sample inlets and outlets to reduce contamination during ascent and descent, as well as to protect the components from cloud water. The operation of the SHARK components was controlled remotely via a radio link using an Arduino microcontroller board (16 km range); once the SHARK was at the desired altitude according to the constantly transmitting radiosonde, the inlet caps opened 10 s prior to the pumps and OPC starting in order to initiate aerosol sampling and monitoring. The payload components, including the servo inlet covers and Arduino control boards, were powered by a 5000 mAh battery (4S 14.8 V LiPo, Overlander, UK). The components were assembled into the SHARK payload with the static (i.e. no wind) weight budget of 10 kg for a 21 m3 balloon (Skyhook Helikite, Allsopp Helikites Ltd., UK) in mind; hence the SHARK weighs 9 kg when fully instrumented.

The cascade impactors allow for the collection of size-segregated aerosol (further details are provided in Sect. 2.2) onto thin films (0.25 mm thickness) for subsequent offline analysis, which can be used alongside information about the aerosol size distributions obtained via the OPC and atmospheric conditions from the radiosonde. Our initial focus concerns the analysis of the ice-nucleating properties of the collected aerosol, but an array of analytical techniques could be applied to characterize the size-selected aerosol, including mass spectrometry, DNA analysis, scanning electron microscopy (SEM), and transmission electron microscopy (TEM) (Ault and Axson, 2017; Garcia et al., 2012; Huffman et al., 2013; Laskin et al., 2018).

Two separate cascade impactors were installed, each operating over different size ranges. This enabled size-resolved aerosol sampling onto substrates across both the fine and coarse modes at high flow rates, while keeping power consumption low enough to be run from batteries. Single impactor systems designed to operate across the accumulation and coarse modes simultaneously require a relatively large pressure drop that would typically require a prohibitively large (and heavy) pump and battery for this application.

Impactor 1 sorts aerosol into five size categories: < 0.25 µm (this size bin is defined by the impactor after-filter and is hereafter referred to as 1a), 0.25–0.5 µm (from stage 1b), 0.5–1.0 µm (from stage 1c), 1.0–2.5 µm (from stage 1d), and > 2.5 µm (from stage 1e). The size categories b to e correspond to the impactor stages where the 50 % collection cut-off diameter (d50) is the lower bound of each bin. The size bins and collection efficiencies for each impactor were digitized from data provided by the manufacturers, (Misra et al., 2002; Product Information Sheet – MSP, 2019) and are shown in Fig. 2. Several collection substrates were tested by Misra et al. (2002), and the dataset from the Teflon substrates was chosen to represent Impactor 1 here as that substrate most closely resembled those used in this study. For Impactor 1, the particles were collected on 25 mm diameter filters of pore size 0.05 µm (Nuclepore track-etched membrane polycarbonate filters, Whatman, UK). Filters were used as impactor substrates rather than films since they have very low background contamination and are easier to obtain. Size category 1a corresponds to an after-filter situated after Impactor 1, which comprised a 47 mm diameter polycarbonate filter with a pore size of 5 µm (Nuclepore track-etched membrane) to maintain the flow rate. The collection efficiency of the after-filter was estimated to be 50 %–100 % at 0.25 µm and below (Soo et al., 2016). Impactor 2 collected aerosol particles into three size categories: 1.0–2.5 (2d), 2.5–10 (2e), and > 10 µm (2f), also illustrated in Fig. 2. The 75 mm diameter filters of pore size 0.05 µm (Nuclepore track-etched membrane polycarbonate filters) were used in Impactor 2. An after-filter could not be used with this impactor since its inclusion increased the required pressure drop to beyond what the pump could supply at 100 L min-1.

A further benefit of using these two impactors in tandem is that, in the size ranges where they overlap of 1.0–2.5 (stage d) and 2.5–10 µm (stage e), the impaction efficiencies are very similar, allowing a direct comparison between the two impactors in this size range. The stages are labelled a through f for the smallest to largest impactor stage sizes (including the after-filter), such that 1d and 2d refer to stage d (1.0–2.5 µm) on Impactors 1 and 2, respectively (see Fig. 2). Background runs were produced by placing the substrates in the SHARK as if setting up to sample, before removing and analysing them as normal to determine the contamination introduced through the installation and recovery of the substrates.

Particle bounce, the bouncing of particles off the impaction substrate and the collection of these particles on the lower stages, has previously been identified as a factor that can cause biases when aerosol is collected by cascade impactors (Cheng and Yeh, 1979; Dzubay et al., 1976). The collection efficiency curves shown in Fig. 2 for Impactor 1 already account for some degree of particle bounce, having been determined experimentally by Misra et al. (2002) using monodispersed polymer particles on a variety of substrates. However, the efficiency curves for Impactor 2 are based on theoretical predictions (Rader and Marple, 1985) and so do not account for any bounce effects. Since two of the stages of Impactors 1 and 2 overlap (stages d and e), it is possible to comment on the possible effects, or lack thereof, of particle bounce, based on the results obtained using each of the comparable stages. This is briefly addressed in Sect. 3.4 where we show good agreement between these two impactors.

The OPC produced binned particle size distributions from 0.38 to 17 µm every 1.38 s at a typical flow rate of 1.2 L min-1. The OPC was remotely operated through the use of its serial link via an Arduino microcontroller board. Particle size, surface area, and mass concentration data were produced from the raw OPC data, and these were then used to calculate the fraction of the aerosol that acts as an INP – activated fraction, nn(T) – and to weight the INP data to particle surface area, generating the ice-active site density per surface area, ns(T), of aerosol. The particle density used was 1.65 g cm-3, as assumed by the OPC software, and they were assumed to be spherical. No correction was made for the hygroscopic growth of aerosol particles as this required assumptions about the chemical nature of the particles, and hygroscopic growth effects were minimized by avoiding sampling when the RH was above about 80 % (see next section).

Utilizing the radio control built into the payload, real-time data informed decisions of when to turn the pumps on and off to sample. Continuous monitoring of the radiosonde data allows the user to avoid sampling under conditions where RH approached 100 %, at which point aerosol particles become excessively swollen with water or activated to cloud droplets. Hence, the influence of hygroscopic growth or cloud droplets on the collected aerosol could be minimized. The temperature and pressure measurements allowed the volume of air sampled by the impactors and OPC to be corrected to standard conditions (1 atm at 0 ∘C).

The weatherproof housing consisted of an acrylonitrile butadiene styrene (ABS) polymer box with dimensions of 560mm×380mm×180mm (IP67, Fibox). Holes to mount the impactors and OPCs were drilled so that Impactor 2 sat vertically upright, and Impactor 1 was oriented 180∘ to Impactor 2 so that it faced downwards, ensuring that both impactors were always oriented 90∘ to the wind. The OPC was at 90∘ to both impactors and facing towards the front of the box, into the wind (see Fig. 3a–c). See Sect. 2.6 for the rationale of the positioning of the OPC and impactor inlets. The tail fin, which is mounted to the lid of the box, was designed to keep the SHARK orientated into the wind and was fabricated from rigid polyvinyl chloride (PVC) sheet. Impactor 1 had its own mounting screws by which it was attached to the box, whilst for Impactor 2 a custom mount was built. Securing ropes were threaded through reinforced holes in the box and connected via a carabiner for quick and easy attachment to the balloon instrument line, as seen in Fig. 1a. Modular foam was used to keep all components in place during flight.

Calculation of the particle losses associated with the instrument inlets due to excessive wind speeds in various configurations were used to inform the design of the SHARK and to minimize sampling biases in higher wind conditions. The calculations were done using an open-source particle loss calculator program in Igor Pro, the details and assumptions for which are presented in Von Der Weiden et al. (2009). The particle loss characteristics of the impactor and OPC inlets at their required flow rates were calculated for a wind speed of 0 and 24 km h-1, the latter used as a maximum representative wind speed for operation. The wind speeds required for optimum performance are < 8 km h-1 for the impactors and OPC, but the system may experience higher wind speeds. Hence, we use this modelling to guide our choice of positioning of the instrument relative to wind direction in order to minimize sampling biases at the inlets. The modelling also allows us to better understand which impactor stages (and OPC size bins) will be most affected by such biases. We make no attempt to correct the measurements for sampling biases, since this correction itself would carry substantial uncertainty but used the calculations to inform us of the best configuration for the various inlets.

The inlet sampling efficiencies in the orientations chosen for the final design of the SHARK are shown in Fig. 3. It is important to note that, due to their dissimilar inlet dimensions and operational flow rates, Impactors 1 and 2 are affected differently by the wind. The particle losses for the largest stages of each impactor are the most affected. Stages a to d on both impactors are only minimally affected by losses. The losses are more significant in stage e on both impactors, but the losses on 1e are greater than on 2e with a 50 % cut-off at around 5.5 µm and a negligible sampling efficiency above about 8 µm on 1e. These calculations also demonstrate that the losses are wind-speed-dependent but that in situations where there is significant wind, the results from Impactor 2 will be less influenced by losses than Impactor 1 at sizes above 2.5 µm.

The OPC suffers up to 1.6 times oversampling for 10 µm particles when sampling into 24 km h-1 wind, but when oriented at 90∘ to the wind the collection efficiency of > 6 µm particles approaches 0 % (see Fig. 3c). Therefore, the OPC has been positioned in the SHARK to be oriented into the wind to ensure data are collected for the whole size range, with the caveat of a sub-isokinetic oversampling of larger particles.

An example of the radiosonde and OPC data that were collected during a SHARK flight are shown in Fig. 4. The data were from a sampling event in the High Arctic in the summer of 2018, during which the meteorological data from the radiosonde and aerosol particle data from the OPC were collected alongside impactor films for INP analysis (the INP results will be published elsewhere). Throughout the 4.5 h flight the altitude, humidity, and temperature were closely monitored to inform decisions on sampling. The sampling start and end times are indicated as solid lines in Fig. 4. The SHARK reached 450 m a.m.s.l. and in the last hour of flight lowered to 350 m due to ice formation on the balloon, instrument, and tether. The RH during the flight was monitored to ensure the SHARK did not sample in humidity approaching saturation; the impactor and OPC manufacturers' specified threshold for the components is 95 % RH, but we aim to only sample with the RH below this value (∼80 %) in order to reduce the influence of hygroscopic growth on aerosol size. After sampling was stopped, the SHARK was brought down to ground level, resulting in the humidity rising. The ability to stop the sampling during the flight meant the impactors were covered, and the pumps turned off during the descent and so did not sample the more humid environment. The ambient temperature was monitored alongside the dewpoint temperature to follow the surface inversions. The temperature inversion was used to determine where to stabilize the SHARK and begin sampling, as sampling was desired above the surface inversion for this run.

The total particle counts per 1.38 s interval from the OPC are shown in Fig. 4d. Processing of the OPC data yielded the results shown in Fig. 5 for the particle number (dN/dlog⁡Dp), particle surface area (dS/dlog⁡Dp), and particle mass (dM/dlog⁡Dp) size distribution data for the sampling period, where Dp is the particle diameter. We present this data to demonstrate that the OPC produces reasonable data when used facing into wind while suspended from a balloon at altitude. Unfortunately, there is no direct comparison with other aerosol size distribution measurements at the sampling location. While the particle number concentration decreases roughly linearly with size, the surface and mass concentration curves have a mode at around 4 µm in Fig. 5b and c. This is consistent with previous studies conducted within the boundary layer in the Arctic (Freud et al., 2017; Hegg et al., 1996; Seinfeld and Pandis, 2016).

The ability to measure INP concentrations and properties using samples collected via the SHARK was tested by performing immersion mode droplet freezing assays on the sampled aerosols. Following a flight, impactor films were removed from both cascade impactors of the SHARK, then each immersed in 5 mL of water and mixed on a vortex mixer for 5 min to wash the collected particles into suspension (O'Sullivan et al., 2018). This suspension was then analysed via a droplet freezing assay using the microlitre Nucleation by Immersed Particle Instrument (µL-NIPI) (Whale et al., 2015), in which 40–50 droplets of 1 µL volume were pipetted onto a hydrophobic glass slide atop a cold plate. A Perspex shield was placed over the cold stage and N2 gas introduced to purge the chamber of moisture as the cold plate was cooled to -40 ∘C at 1 ∘C min-1. The temperatures at which droplets froze were recorded using video analysis until the entire population had frozen. This allowed the fraction of droplets frozen as a function of temperature, fice(T), to be calculated (O'Sullivan et al., 2018; Whale et al., 2015) using the equation fice(T)=Nf/Nt, where Nf is the number of frozen droplets at temperature T, and Nt is the total number of droplets. The INP concentration per volume of sampled air as a function of temperature, [INP]T, was then calculated for each film using fice(T), according to Eq. (1) adapted from (Vali, 1971) to include weighting to the volume of air sampled: 1[INP]T=-ln(1-fice(T))Vdroplet⋅VwashVair, where Vdroplet is the droplet volume (i.e. 1 µL), Vwash is the amount of water into which the filter is immersed to produce the suspension for analysis (i.e. 5 mL), and Vair is the volume of air sampled.

In order to test whether the SHARK impactors were sampling in a representative manner, the SHARK was run concurrently with a filter-based particle sampler (BGI PQ100, Mesa Labs) and which is used as an EPA Federal Reference Method for PM10 (designation no. RFPS-1298-124). This sampler was equipped with a PM10 head and an optional cyclone impactor which provided a size cut at 2.5 µm. Aerosol was collected onto 0.4 µm pore size Nuclepore track-etched membrane polycarbonate filters at a flow rate of 16.7 L min-1 (i.e. 1 m3 h-1). This type of filter collects particles across the full range of available aerosol sizes, even at sizes smaller than the pore diameter, with high collection efficiencies (Lindsley, 2016; Soo et al., 2016). These polycarbonate filters have also been successfully employed in other ice nucleation field measurements (DeMott et al., 2016; Harrison et al., 2018; Huffman et al., 2013; McCluskey et al., 2016; Reicher et al., 2019; Tarn et al., 2018). These substrates are known to have a low ice-nucleating ability and allow the collected particles to be released into suspension for subsequent INP analysis (O'Sullivan et al., 2018). The filters were analysed using the µL-NIPI in the same manner as for the impactor films collected using the SHARK. The PQ100 filter sampler was deployed alongside the SHARK in Cardington (UK) and in Hyytiälä (Finland).

In order to compare the SHARK-derived, size-resolved INP data with the results of the PM10 or PM2.5 PQ100 filter sampler, the INP concentrations determined across the appropriate SHARK size categories were summed. In Fig. 6a, data are presented from Cardington, where the sum of 2d and 2e from SHARK is compared with the filter sampler fitted with a PM10 head (Impactor 1 was not available during this test). The SHARK was suspended from a tethered balloon roughly 20 m from the ground, whereas the filter sampler was on the ground (inlet ∼150  cm above the surface), where both samplers were within the well-mixed boundary layer. The agreement is very good apart from two highest temperature points from the filter sampler, but note that the Poisson uncertainties for these points are substantial and also that the two samplers were separated vertically by 20 m.

We then show data from Hyytiälä in Fig. 6b where we compare the INP spectrum from the filter sampler, with a PM2.5 cut-off installed, with the sum of stages 1b, 1c, and 1d (the after-filter; stage 1a was not used on Impactor 1 in this case). Here, both samplers were positioned within a few metres above the ground. Again, the agreement between the SHARK and the filter sampler was very good. For both Cardington and Hyytiälä, the smallest particles (< 0.25 µm) were not sampled using the SHARK, but the agreement between the filter sampler and the SHARK implies that, in these cases, the smallest particles made a minor contribution to the overall INP population, which is what we would generally anticipate from the literature (Berezinski et al., 1988; Huffman et al., 2013; Mason et al., 2016; Santachiara et al., 2010; Si et al., 2018; Welti et al., 2009). The consistency between the SHARK and the filter sampler indicates that there are no major losses of aerosol in the SHARK sampler, at least relative to the PQ100 filter sampler.

An example of data from the size-resolved collection and analysis of INPs is shown in Fig. 7, from a sampling run performed in Leeds (UK). The fice(T) curves for each impactor stage are illustrated in Fig. 7a. As discussed in Sect. 2.2, there are two stages, d and e, which have similar size cuts on both stages. Using stage e as an example, it can be seen that while the fraction frozen curves for the two samplers are shifted by about 3 ∘C (Fig. 7a), normalizing to the volume of air sampled to yield [INP]T in Fig. 7b shows that the INP spectra derived from stages 1e and 2e are consistent with one another. Stage 2e covers a lower range of INP concentrations than stage 1e by about 1 order of magnitude, because the flow rate through this impactor was more than a factor of 11.1 (100 L min-1/9 L min-1) higher, and the probability of collecting rarer INP was increased by this factor. The agreement between the two impactors indicates that aerosol was collected with no significant losses/enhancements due to factors like particle bounce or wind observed. Based on the inlet particle loss calculations in Fig. 3, higher losses may have been expected in impactor stage 1e, but these are not apparent here.

The derived size-resolved INP (srINP) concentrations for all four test sites are shown in Figs. 8 and 9. The fice(T) curves for these test sites can be found in Figs. S1 to S4 in the Supplement. Figure 8 shows the INP concentration spectra in the classic form, wherein INP concentrations are plotted against temperature for each size bin, whereas Fig. 9 shows the same data in novel srINP plots to allow more intuitive comparison of the INP concentration contribution from each stage with respect to temperature. In Fig. 9, where there were measurements from two impactors for the same stage (e.g. d and e), the INP concentrations were merged by taking an average at temperature intervals of 0.5 ∘C (also for Fig. 6). The colour gradient in Fig. 9 represents the temperature-dependant concentration for each size bin and the overall steepness of the d[INP]T/dT curve. The steepness of the INP spectra can be useful in discriminating between different INP species. On inspection of Figs. 8 and 9, it can be seen that the spectra in the four locations have very different characteristics. Not only does the general shape of the spectra vary, but the size-dependence is also very different in the four locations. Due to the sample size, these variations could be attributed to the different aerosol population in each location, the time of year, and meteorology, which could affect the INP concentrations and spectra (Kanji et al., 2017; Šantl-Temkiv et al., 2019; Tobo et al., 2019; Wex et al., 2019). We now discuss the size-resolved INP concentration spectra from these tests, bearing in mind that these four tests were one-offs and should not be regarded as characteristic of those sampling sites but rather illustrative of the importance of making size-resolved measurements.

The first site testing of a prototype of the SHARK in which all of the components were installed was conducted in Cardington (UK) on the 15 May 2018, but only Impactor 2 was used (see Figs. 6a and 8a). The Cardington site is an airfield, with large areas of grassy land near a main road, and the sampling was conducted during spring. In order to demonstrate the utility of the SHARK to make balloon-borne INP measurements whilst providing a comparison with a commercial ground-based sampler, the SHARK was sampling whilst suspended from a tethered balloon, flying roughly 20 m above the ground. The INP spectra (Figs. 8a and 9a) in this location are steep, increasing by 2 orders of magnitude within 2.5 ∘C, and are centred around -18 to -20 ∘C; the [INP]T for 2f and 2e increases by an order of magnitude in just ∼1 ∘C. The INPs in this location were dominated by particles greater than 2.5 µm, whereas particles between 1 and 2.5 µm made a smaller contribution and show a shallower d[INP]T/dT, seen in Fig. 9a as a larger spread of data. We speculate that the coarse-mode INPs at this site were of biological origin, such as fungal material, pollen, or bacteria with a steep INP spectrum (Kanji et al., 2017). Some fertile and agricultural soil samples have also been shown to be very active (Hill et al., 2016; O'Sullivan et al., 2015; Steinke et al., 2016; Tobo et al., 2014), and a mechanism for emissions of soil material into the atmosphere has been proposed (Wang et al., 2016). However, the steep portion of the INP spectrum for fertile soils tends to be at temperatures above ∼-10 ∘C, warmer than observed in the Cardington sample. The steepness of the curve and the temperature are consistent with ice nucleation by pollen (O'Sullivan et al., 2015; Pummer et al., 2012; Tarn et al., 2018). Although the size of whole pollen grains are often larger than 10 µm, pollen is known to release nanoscale materials that nucleate ice, which might be internally mixed with aerosol in this size bin.

In Hyytiälä (Finland), a field site in the boreal forest, the INP spectra contrast quite strongly with those in Cardington (see Figs. 6b and 8b). Sampling took place on the 11 March 2018, when the Hyytiälä site was snow covered, and sampling was performed at the surface (inlet ∼150 cm above surface). In this case only Impactor 1 was used without the after-filter installed. The complex nature of the size-dependence of INP is clear here. Intriguingly, in this location, the INP concentration was greatest for the smallest stage used (1b; 0.25–0.5 µm) and accounted for the majority of the INPs between -17 and -22 ∘C. The fewest INPs came from the next smallest stage 1c (0.5–1 µm), while at temperatures below -23 ∘C, stage 1e contained the majority of the INPs. These results indicate that the INP spectra are complex and that concentrations of INPs do not always increase with increasing size as might be expected. Huffman et al. (2013) reported INP concentration measurements in a forest ecosystem, where the particles between 1.8 and 5.6 µm enhanced during rain. Hence, as in the present study, Huffman et al. (2013) showed that INP activity does not always increase with size. The highest INP concentrations in Hyytiälä were measured for aerosol sizes of 0.25–0.5 µm, and we note that these accumulation mode INPs would have lifetimes of many days to weeks in the atmosphere and could therefore be transported to locations and altitudes where they may influence clouds. Clearly, this would be an interesting location for more measurements with the full SHARK payload to gain further information on the long-term INP concentration variations and the aerosol sizes responsible for them.

The testing in Leeds (UK) used both impactors at ground level with the SHARK suspended from a frame to allow orientation into wind. The Leeds sampling was conducted within the University of Leeds campus on a patch of grass on the 7 June 2018 in close proximity to the School of Earth and Environment. In this test the full suite was deployed, including the impactors, after-filter, and OPC. The particle number, surface, and mass size distribution data for this test can be found in Figs. S5 to S7. It can be seen in Fig. 8c that generally, the larger bins contained more active INP. The only exception to this occurred with the after-filter (< 0.25 µm), which had slightly higher INP concentrations below about -25 ∘C than the next two size bins (0.25–1.0 µm). As with the measurements in Hyytiälä, clearly more measurements illuminating the contribution of the smaller particles in similar environments would be beneficial since the atmospheric lifetime of these fine particles is relatively long. We note that a substantial proportion of INPs quantified just outside of Leeds in a previous study were heat sensitive and therefore most likely of biological origin (O'Sullivan et al., 2018). In the future, conducting heat tests, as well as using mass spectrometry, SEM, and DNA analysis with the size-resolved INP samples may help to identify the INP types in the various size fractions and highlight any differences between size ranges.

The final test was in Longyearbyen (Svalbard) from the seventh deck of the icebreaker Oden, 25 m above the surface, when moored ∼200 m from the shore, overnight from 23 to 24 September 2018. The full SHARK payload was used in this case, with the OPC, both impactors, and the after-filter on Impactor 1. The particle number, surface, and mass size distribution data for this test can be found in Figs. S8 to S10. The INP spectra in this location, shown in Fig. 8d were quite distinct from the other three locations in that all size fractions contributed similarly to the INP population, and there is a very shallow slope of dln⁡[INP]T/dT (Fig. 9d). We detected INPs at temperatures of up to -10 ∘C with concentrations of around 0.01 INP L-1. These high-temperature INP concentrations are consistent with the summertime measurements reported at other Arctic locations, including Ny-Ålesund (Svalbard) (Wex et al., 2019). The INP in this region potentially originate from a range of sources. Tobo et al. (2019) recently reported that dust and biological material from glacial valleys in Svalbard may be an important source of INPs in the region. We also note that we sampled while the Oden was moored in the port of Longyearbyen where local pollution sources may have been significant (Zhao et al., 2019).

The addition of size distribution information to the INP concentration spectra allowed the calculation of the number of active sites per unit surface area, ns(T), and the activated fraction, nn(T), of the size-resolved samples. These quantities are determined by weighting the srINP concentrations to the total surface area and the aerosol number in each size bin, respectively, as shown in Eqs. (2) and (3). 2ns(T)=-ln(1-ficeT)As, where As is the total surface area of the particles per droplet in a µL-NIPI droplet freezing assay. This was calculated for each impactor size range, using data from the relevant size bins of the OPC data. 3nn(T)=-ln(1-ficeT)N, where N is the total number of particles sampled by the impactor in each size bin, calculated using the number concentration in each size category as measured by the OPC and the volume of air sampled by the impactor. The size bins from the OPC which have been included in the calculations were matched to those in the impactors. The bin boundaries for the OPC calculations were within tens of nanometres of the impactor bin boundaries.

Calculating the ns(T) and nn(T) values from the INP data was only possible for some of the size ranges due to the sampling ranges of the instrumentation employed. The smallest particle diameter measured by the OPC is 0.38 µm, i.e. above the lower limit of impactor stage 1b, while the largest impactor stage, 2f (> 10 µm), has no defined upper bound. Therefore, the three bins (i.e. impactor stages) that were used to produce ns(T) and nn(T) were c (0.5–1.0 µm), d (1.0–2.5 µm), and e (2.5–10 µm). The ns(T) and nn(T) data were calculated for the field tests in Leeds and Longyearbyen; data from Cardington and Hyytiälä are not provided as the OPC was not in use at these sites.

The plots of activated fraction shown in Fig. 10 are addressed first. For the Leeds sample, there is a difference in the nn(T) values between bins c and e (Fig. 10a), where the smallest bin is 1–3 orders of magnitude lower than the largest bin, with the middle bin in the centre of the two. In Longyearbyen (Fig. 10b), the nn(T) for bin e is about a factor of 10 larger than bin c, but bins c and d produce very similar values of nn(T). Overall, these nn(T) plots show that the coarse-mode aerosol generally have a higher fraction of aerosol that serve as INPs than the fine mode, but there is variability in the dependence on size between the two samples. In contrast to the nn(T) values, the size-resolved ns(T) data for both Leeds and Longyearbyen show that the data from the three size categories are all within a factor of 2–10 (close to our uncertainty estimates). Given the activity of aerosol across these bins scales with surface area, this data might indicate the same INP species is active across each bin at these sites.