Ice nucleating particle sources and transports between the Central and Southern Arctic regions during winter cold air outbreaks

The Arctic is warming at a rate faster than the rest of the Earth (ACIA, 2005; Serreze and Barry, 2011; Overland et al., 2017; Dutta et al., 2021), and faster than predicted by climate models (Solomon et al., 2007). Coupled Model Intercomparison Project phase 6 (CMIP6) simulations project an Arctic-mean, near-surface temperature change that is approximately 3 times that of the global mean by the mid-21st century (Davy and Outten, 2020), while another recent study estimates surface warming at nearly 4 times the global mean (Rantanen et al., 2022). This warming is conducive to a seasonally sea-ice-free Arctic (Notz and Stroeve, 2018; Kim et al., 2023). Significant uncertainties underlie Arctic climate projections, largely because the future behavior of Arctic clouds in response to such changes is uncertain (Sherwood et al., 2020).

Among the cloud systems of relevance, so-called Cold Air Outbreak (CAO; Pithan et al., 2018) cloud systems ensue due to the intense air mass transformations that occur when cold airmasses over the Arctic ice flow southward over North Atlantic open water in late Fall through Spring. These are evident as beautiful linear and cellular cloud structures in satellite imagery between the Fram Straight and northern Scandinavia (Geerts et al., 2022). CAO conditions are defined by sea ice areal coverage <10%, meeting a static stability criterion estimate (M-index) for unstable boundary layers (M = θ_SST − θ_850hPa > 0, where θ_SST and θ_850hPa are the potential temperatures of the ocean surface skin and at 850 hPa, respectively) (Fletcher et al., 2016), and often including an additional wind speed condition (Geerts et al., 2022; Lackner et al., 2023). Air-sea temperature differences can exceed 30 K at the point that air exits from the Arctic ice sheet and give rise to latent and sensible oceanic surface fluxes that are among the largest observed anywhere on Earth (Aemisegger and Papritz, 2018).

Changes in frequency and intensity of CAOs and changes in sea ice extent in a changing climate will likely exert profound feedback on the Arctic climate system due to alterations in air-sea heat fluxes and impacts on ocean mixing (Moore et al., 2015). Yet, understanding of the properties of the CAO shallow, mixed-phase cloud regime, its role in energy and water cycles and its treatment in climate models is arguably among the poorest of all cloud regimes (Rémillard and Tselioudis, 2015). Numerical weather prediction models are also challenged to predict the strong surface winds, waves, and heavy snowfall during intense CAOs and in embedded polar lows (Kristiansen et al., 2011; Stoll et al., 2020), making hazards to naval operations, shipping and coastal communities difficult to foresee. CAO clouds can vary dramatically in their morphology. Although the drivers of variations are poorly understood, it is known that aerosols can affect both dynamical and cloud microphysical processes, and thus may play a role in both cloud organization and composition. Understanding the sensitivity of mixed phase ice/liquid partitioning to aerosol perturbations is critical to Arctic cloud optical depth and lifetime feedback (Morrison et al., 2012).

Mixed-phase clouds such as those forming under CAOs respond to 2 distinct subsets of aerosols: cloud condensation nuclei and ice nucleating particles (INPs). Herein we will focus on the latter category that is responsible for ice initiation, triggering of secondary ice formation processes (Korolev and Leisner, 2020; Mages et al., 2023), and thereby snow formation in clouds until the point that they reach altitudes where temperatures fall below the homogeneous freezing limit for supercooled water at approximately –38°C. Many recent measurements of INPs have been made in the Arctic region. Most have occurred at surface sites in multiple locations (including ships) over all seasons, although mostly in spring through fall rather than winter (Conen et al., 2016; DeMott et al., 2016; Creamean et al., 2018; Creamean et al., 2019; Irish et al., 2019; Wendisch et al., 2019; Wex et al., 2019; Hartmann et al., 2020; Li et al., 2022; Yun et al., 2022; Li et al., 2023; Gjelsvik et al., 2025). Annual cycle measurements of INPs have most recently been obtained in Svalbard (Pereira Freitas et al., 2023; Tobo et al., 2024) and during the Multidisciplinary Drifting Observatory for the Study of Arctic Climate (MOSAiC) study (Creamean et al., 2022; Barry et al., 2025), including data sets that we will utilize in this study. Few measurements of INPs are available to date that are focused specifically within the CAO regime. These primarily have happened as measurements of opportunity within field campaigns (e.g., Lance et al., 2011; Jackson et al., 2012; Creamean et al., 2018). More recently, Inoue et al. (2021) focused ship-based measurements within CAOs over the Chukchi Sea region in late Fall, while Raif et al. (2024) conducted some of the first aircraft studies focused within CAOs over a 9-day period in the region between Sweden and Svalbard in 2022.

Ice nucleating particles in CAOs could be sourced from populations present over the pack ice at the point of air mass origin, emissions from open water along the air fetch, and particles mixed from the free troposphere during marine boundary layer (MBL) evolution (growth in depth) under the strong vertical mixing conditions that occur. Thus, vastly different INP types may influence the clouds formed at different times and locations during a CAO, depending on air mass exchange into and out of the central Arctic and mixing processes during transit of air. Emissions from the underlying ocean via sea spray production of biogenic/organic (microbial and recalcitrant organic) INPs (Burrows et al., 2013; Wilson et al., 2015; DeMott et al., 2016; McCluskey et al., 2018a; McCluskey et al., 2018b) have been shown to frequently dominate INP budgets over remote ocean regions such as the Southern Ocean (McCluskey et al., 2019), and have also been suggested to be important components of the INP budget in Arctic regions (Creamean et al., 2018; Hartmann et al., 2020, Hartmann et al., 2021). Sea spray INPs could have strong sources over the far North Atlantic region of open water in winter. This includes the area from the Fram Straight to the coast of Norway and Sweden, a region of high occurrence of CAO events (Fletcher et al., 2016; Papritz and Spengler, 2017; Geerts et al., 2022; Lackner et al., 2023). Over the ice pack, at the point of CAO origin, INPs may be present from both prior marine aerosol transports into the Arctic during warm air advection events (Dada et al., 2022), long range transport of particles into the Arctic from continental regions especially during the Arctic haze but including particles from both desert and boreal forest regions (Behrenfeldt et al., 2008; Schmale et al., 2021; Song et al., 2022; Yun et al., 2022; Ansmann et al., 2023; Boyer et al., 2023; Lapere et al., 2024) or windblown snow evaporating in a dry boundary layer (Gong et al., 2023). Among long range transported particles, mineral dusts are ubiquitous potential sources of inorganic INPs at certain times of the year, and arable, organic-containing soil dusts from closer land regions likely possess even higher nucleation efficiencies (Kanji et al., 2017). In winter, mineral and soil dust transports are less likely to occur over long distance within the boundary layer, while within-Arctic organic INP soil sources such as glacial and other high latitude dust are snow covered (Tobo et al., 2019; Sanchez-Marroquin et al., 2020; Shi et al., 2022) and so not usually active. Nevertheless, mineral and soil dust likely is transported into the Arctic vortex from lower latitudes and mixes into the boundary layer in winter. In modeling studies, Shi et al. (2022) found that lower latitude dust from Africa and Asia dominates the dust loading in the Arctic upper troposphere, peaking in boreal spring and winter. Lidar studies have also found dust to occur throughout the troposphere in winter and spring over far northern Alaska (Xie et al., 2022), and aircraft INP studies over this region and the adjacent oceans in spring support arable soil dust transported long distances as the primary source (Sanchez-Marroquin et al., 2023).

Initial reports on INPs during the year-long MOSAiC campaign (on the MOSAiC ship, the RV Polarstern) or in Svalbard have emphasized the enhanced INP concentrations of likely biogenic origin occurring at modest supercooling to −15°C following sea ice melting and loss of snow cover from nearby lands in the late Arctic spring through fall (Creamean et al., 2022; Li et al., 2023; Pereira Freitas et al., 2023; Tobo et al., 2024), also associated with the peak in solar radiation and primary biological productivity. There has been less reported or discussed about INPs and their dominant sources during winter, and especially during CAOs. From an aerosol standpoint, the central Arctic around the RV Polarstern received frequent input of anthropogenic particles from northern Russia and Siberia (Boyer et al., 2023), suggesting that other inorganic and organic terrestrial particles may have entered with that same air. Barry et al. (2025) indicated the occurrence of airborne fungi throughout the year in MOSAiC, also supporting long-range transport of terrestrial air in winter. Winter was also the time when INP concentrations were lowest, proteinaceous biological INP contributions at −15°C activation temperature were weakest, and other organic and inorganic contributions to INPs were the highest (Creamean et al., 2022; Barry et al., 2025). A similar situation prevailed in Svalbard, where wintertime aerosols were dominated by sea salt, mineral dust, and discernable contributions from blowing snow, the latter characterized by its bromine content and mixtures of sea salt and dust (Song et al., 2022). Again, blowing snow was also a ubiquitous aerosol source in MOSAiC (Gong et al., 2023). Dust and sea salt particles were also found to dominate cloud residual particles sampled by a counterflow virtual impactor inlet in winter on Zeppelin Observatory in Svalbard (Adachi et al., 2022), and an increase in contributions from mixed dust-sea salt particles in residuals, as compared with those in ambient air, supported a possible role of blowing snow impacting mixed-phase clouds. Like MOSAiC, the Svalbard site near to the winter ice edge had lowest INP concentrations in winter and the lowest contribution of biological INPs in that season (Pereira Freitas et al., 2023; Tobo et al., 2024). Lastly, Carlsen and David (2022) discuss inferences from satellite observations of the seasonal and spatial variability of INPs and their influence on cloud phase (ice vs. liquid), noting a much lower phase transition temperature for air entering clouds over the pack ice than for air entering clouds over the open water regions. The implication is that additional sources of INPs become active at warmer temperatures as air masses advect over open ocean in this season.

These past results primarily paint a picture of lowest INP concentrations in winter occurring at the source location of CAOs over the central Arctic ice pack, where weaker inputs of dust and sea spray particles of both long-range and local (e.g., open leads) origin are the INP sources. At this time of year, marine INP emissions may be a stronger source of INPs once air moves over open water, although CAO cloud formation also stimulates mixing down of free-tropospheric air that could contain additional INPs of terrestrial origin. It is important to note in this regard that marine-sourced INPs in the Arctic region could be quite distinct from those from other ocean regions, due to riverine and coastal inputs that fuel primary biological production in the ocean year-round (Terhaar et al., 2021), the fact that there are rich soil and permafrost sources of INPs that feed into the Arctic Ocean via terrestrial water sources and air deposition (Tobo et al., 2019; Creamean et al., 2020; Barry et al., 2023a; Barry et al., 2023b), and the turbid conditions of the shallower oceans that have been implicated in the higher activity of INPs emitted in sea spray (Inoue et al., 2021) during deeper ocean mixing (can extend to 1,000 m under high wind stress) that occurs during CAOs (Condron and Renfrew, 2013).

In contrast to the studies discussed above, Raif et al. (2024) have most recently reported on INPs in focused aircraft measurements around CAO cloud systems during a 9-day period in March 2022. During this period, higher INP concentrations were attributed to mineral dust mixed with a biogenic source that had accumulated within long range transported Arctic haze. Concentrations in these cases were more consistently higher above clouds. Thus, short-term, annual, and interannual variability of INP sources may occur during the time of CAOs.

Lastly, long range transported continental particles entering the Arctic in winter may also include biomass burning particles and fossil fuel combustion particles. Biomass burning particles have been detected persistently in winter at higher altitudes impacting cirrus clouds over the Arctic (Engelmann et al., 2021; Ohneiser et al., 2021; Ansmann et al., 2023), while combustion particles are the source of persistent lower-altitude Arctic haze in winter (Shaw, 1995). While these aged combustion particle types are not expected to be especially active as INPs (e.g., Schill et al., 2020), their influence could be expressed when other sources are limited, especially in winter. Mixing of free tropospheric air into the MBL once air passes from over the pack ice to the open ocean can be expected to mix these same aged, transported aerosols into the deepening boundary layer.

The Cold-air Outbreaks in the Marine Boundary Layer Experiment (COMBLE) was conducted by the United States Department of Energy (DOE) Atmospheric Radiation Measurement program (ARM) from approximately December 1, 2019, to May 31, 2020, on the coast of northern Norway (Andenes, Andøya). This deployment included a suite of instruments as part of the ARM aerosol observing system (AOS), and INP measurements. This record, which included determinations of periods defining CAO conditions, provided observations to evaluate the number concentrations and compositions of INPs where CAOs made landfall. Through use of trajectory analyses during these CAO events, additional upstream data over the ice pack and near to the entry of CAO air to the open ocean were sought to characterize the transformation of CAO INP characteristics during the transect of air southward. Sites used as the source of higher latitude data included the central Arctic deployment of the RV Polarstern into the ice pack during the MOSAiC, and the intermediate site at Zeppelin Observatory in Ny-Ålesund, Svalbard, that has been a site for long-term aerosol monitoring. Aerosol and INP measurements were obtained over the same time frame at both higher latitude sites, facilitating this trajectory investigation.

The first step for this study involved isolating CAO periods based on thermodynamic criteria, identifying CAO periods that aligned with INP data collections, and accepting only cases where air mass trajectory analyses indicated that air passed within proximity of the 2 sites that were upstream of the COMBLE measurement site (Figure 1), as discussed in the next subsection. In this section we will describe the selection of CAO events, the INP and aerosol data used from the respective sites, and the processing of that data to infer the INP types present at various times along the passage of air arriving at the COMBLE site.

The locations of the 3 sites, namely the Polarstern during MOSAiC (hereafter, MOS, which had a variable position), the site in Ny-Ålesund (hereafter, ZEP), and the COMBLE site (hereafter, ANX) are indicated in Figure 1. During the CAOs in this study, these sites respectively generally represented pack ice, edge of pack ice (but with potential marine influence), and open water sampling environments. Positional data are available for MOS (Shupe et al., 2022) but will not be specifically referenced herein. The site at Ny-Ålesund, on the western coast of Svalbard, is the mountaintop Zeppelin (ZEP) observatory (78.9067°N, 11.8883°E, 474 m a.s.l). The ANX site was located at a small harbor port (Nordmela) site (69.1415°N, 15.6838°E, 4 m a.s.l.) on Andøya Island near Andenes, approximately 1,075 km by direct line from ZEP. These sites and the measurements used at each are listed in Table 1, with details of measurements given in subsequent sections.

Definition of applicable CAOs began with the CAO periods at Andenes that were listed by Geerts et al. (2022) and Lackner et al. (2023) and proceeded to a subset that met additional criteria for passing near or representing the upstream sites during times when INP data were available. As documented in those publications, CAO periods for ANX were identified based on the CAO M-index, having values >0, as calculated offshore of the site. Additionally, wind at ANX needed to exceed 10 knots in speed and be directionally oriented between 250° to 30°, representing approaches to the site over open water (Geerts et al., 2022). A total of 49 CAO episodes of varying length were identified at ANX based on these objective CAO criteria (Geerts et al., 2022, c.f., figure S10; Lackner et al., 2023) between December 1, 2019, to May 31, 2020. These events lasted as short as just 200 min and as long as 3,600 min, with an average time of 1,005 min (Geerts et al., 2022). In the field, filter sample timing at ANX could not be controlled to align with a criterion that could only be computed in post-analysis. Rather, decisions on when to initiate INP sampling used forecast products (including for M) based on bespoke regional UK Met Office simulations that included a double moment microphysics scheme (Field et al., 2023), and consideration of the working schedules of the ARM technical team. The pulsing nature of some of the CAOs, that led to many of the shorter events reaching ANX, did not often allow perfect and continuous alignment of INP samples that were collected over 6- to 72-h periods (weighted mean of 33 h, or approximately 1.5 times the mean length of CAOs).

For the determination of selected case studies that connected the sites, 5-day air mass back-trajectories were generated with the National Oceanic and Atmospheric Administration Air Resources Laboratory’s Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT, Stein et al., 2015; Rolph et al., 2017), using the 1° × 1° GDAS dataset. These trajectories were run at 3-h intervals, and from 50 m starting height (using model vertical velocity), starting from the beginning to the end of the sampling times for filters collected during CAO events at ANX. Even if trajectories did not pass directly over MOS and ZEP, an assumption was made that air within a few hundred km at proximal latitude was likely within the same CAO system airmass, and that its aerosol was likely representative. As for ANX, it was also not usually possible to achieve perfect alignment of INP samples with the midpoint time of source of air at the higher latitude sites, as is evident in Table S1. Hence, other subjective measures were used to define INP filter periods that met the criteria for use in this present analysis. For example, breaks in CAO periods or extensions of them during which winds persisted mostly off the ocean were accepted as valid possible cases of interest. The most restrictive consideration was nevertheless the constraint of having CAO events satisfy the need for trajectories to pass near all 3 sites during some portion of the filter collection periods. Through these analyses, 6 case studies were identified as having proximal trajectories connecting the sites, as listed in Table S1 and shown in Figure 2. A final height of 100 m was tested without finding differences in trajectory analysis results. It was also the case that trajectory altitudes remained below 950 mb along the path in most cases. For CAO1, the pressure level reached 900 mb over MOS and CAO 5 was the outlier with a pressure level of 850 mb over MOS. Of all the identified case studies, CAO 4 had the most variance of upstream source during filter collections, but is retained because the last 24 h of the ANX filter had a more direct source from ZEP and MOS. We may also note that while most of these CAOs align with ones listed in Geerts et al. (2022, table ES6), the filters for our events 5 and 6 span adjacent CAO couplets from Geerts et al. (2022) that were separated by a break period.

Here we describe the array of aerosol measurements used, which varied by site, as well as various treatments of these data and derived parameters. A compilation of fundamental measurement uncertainties, calibration protocols and quality control protocols for all instruments discussed in this section is provided in Table S2.

The DOE ARM Aerosol Observing System (AOS) was a key element of the analysis for this study. It was positioned approximately 5 m above sea level, with the AOS aerosol inlet extending to 10 m above ground level (Figure S1). The AOS inlet is a whole air inlet designed for transmission of particles up to at least 10 µm in diameter, and with a rain/snow guard that has a heated edge to prevent ice buildup (Bullard et al., 2017; Uin et al., 2019). An effective transmission size (50% cut diameter) of 6.4 µm was recently reported by Kuang and Singh (2024). Aerosol size distributions at ANX were determined through use of data streams from 3 separate AOS instruments. Submicron aerosol measurements were represented by a scanning mobility particle sizer (SMPS, Model 3936, TSI Inc.) and an ultra-high sensitivity aerosol spectrometer (UHSAS; Droplet Measurement Technologies, Inc.), both located in the AOS. The SMPS measured over a size range from 10 to 500 nm, at a 5-min resolution (Uin et al., 2019), while the UHSAS measured from 60 nm to 1 µm at a 10-s resolution. Both measurements may be considered as “dry” since the relative humidity (RH) at the point of measurement was quite low, always below 40% RH. In contrast to other sites, there were no explicit measurements of particle concentrations at sizes larger than 1 μm. As concerns estimating aerosol surface area, important with respect to relational analyses of INP data, this was a critical limitation. Hence, we exploited the use of optical property data from the 3-wavelength nephelometer (TSI Model 3563) that also sampled from the main AOS inlet to estimate supermicron aerosol surface area, as discussed in Section 2.2.4. These measurements were also “dry,” typically obtained at below 20% RH and never exceeding 30% in the record for this site.

The ZEP site aerosol measurements were considered representative of cold air layers moving south from over the ice pack that extended to envelop the laboratory. As described in detail by Platt et al. (2022), a variety of aerosol measurements have been in place at ZEP since the late 1980s, including continuous nephelometer measurements of aerosol light scattering. In this study we used total particle number concentrations from a condensation particle counter (CPC; TSI Model 3010) with a 10 nm lower limit of detection, and custom-made twin differential mobility particle sizers (DMPS) measuring submicron particle size distributions from approximately 5 to 800 nm. The air was sampled through a heated whole air inlet built according to ACTRIS recommendations for sites often engulfed in clouds, with an upper size cut off around 40 microns (https://www.actris-ecac.eu/aerosol-inlets-and-conditioning.html). The sample air RH at point of measurement was always below 40%. Supermicron aerosol size distributions were measured with an optical particle size spectrometer (FIDAS^® 200 E, Palas GmbH) that was situated on the measurement platform atop the Zeppelin Observatory. The FIDAS^® had its own heated inlet, controlled at approximately 40°C, such that the sample RH in winter will typically range from 5% to 15%. A nephelometer (EcoTech Aurora 3000) was operating at Zeppelin but did not have the 10 and 1 µm impactors that operate with the AOS nephelometer, so we did not use that data. Besides the semi-permanent particle sizing instruments, an additional optical particle spectrometer (OPS) was added during this study, from the National Institute of Polar Research (NIPR). The OPS (TSI Model 3330) measured over sizes 0.3 to 10 µm. The OPS sampled from a single line connected vertically to a PM10 inlet (DIGITEL Elektronik GmbH) through the roof of the facility. Since this latter arrangement likely removed any influences from falling or blowing snow, data from this OPS were emphasized for constraining aerosol surface area, although we used data from all systems to guide surface area estimates. Although the inlet was not heated for the OPS, the transit to the instrument within the laboratory at approximately 20°C was sufficient to dry the air stream to below 20% RH as noted in prior sampling at this site (Schmeisser et al., 2018). Thus, we considered all ZEP aerosol size observations as dry, unless otherwise noted.

The AOS was positioned on the bow of Polarstern, approximately 18 m above the sea ice or ocean surface. Aerosol size distributions for MOS were determined through use of data streams from the SMPS and UHSAS instruments. For MOS, supermicron aerosols were also measured using an aerodynamic particle sizer (APS, model 3321; TSI, Inc.) producing 20-s resolution data from 500 nm to 20 µm (Heutte et al., 2023). The APS was operated by École Polytechnique Fédérale de Lausanne (EPFL) and was mounted in a separate container adjacent to the AOS. The 2 containers had separate inlets, but of similar design that used precipitation guards and higher flow rates through large diameter tubes intended to maximize transmission efficiency, up to 20 µm in the case of the AOS (Jefferson, 2011; Bullard et al., 2017) and 40 µm for the EPFL container (Heutte et al., 2023). The AOS inlet contained a back-pressure purge system that was applied to minimize sampling of ship emissions from the aft stacks, such that during obvious periods of ship emission influences (visually), the purge system would cause the inlet to stop sampling ambient air and was switched to purified air while all instruments remained operational. Since not all periods of stack pollution were detected visually, a pollution “mask” was used on both data streams that was based on the derivative of the ultrafine particle concentration, the particle size distribution, CO₂, and black carbon measurements, and wind direction (Beck et al., 2022). This screening was especially important for removing local pollution contamination from the ship stack, vents, and snow machines used in the study. During the measured CAOs, additional data filtering was required for the MOS data streams post-study. The first filter removed any data corresponding to a CPC spike, likely representing ship stack pollution that was not already isolated by the pollution flag. The second removed any data in which the SMPS or UHSAS instruments were either completely turned off or displayed erroneously low concentrations. These may have been clean air “purge” events that were not clearly flagged.

As for the ANX site, a 3-wavelength nephelometer (TSI Model 3563) also sampled from the main AOS inlet and its measurements were used as discussed in Section 2.2.4. These measurements were also “dry,” typically obtained at below 10% RH for this site.

To create continuous aerosol size distributions spanning the full range of diameters of interest for relating to ice nucleation measurements, data from the various instruments measuring over specific size ranges were merged using previously documented procedures (Hand and Kreidenweis, 2002; Khlystov et al., 2004; Kupc et al., 2018; Testa et al., 2021). As all sites did not have the same aerosol package (Table 1), observations from 2 or more instruments were combined to create continuous size distributions for MOS (SMPS, APS), ANX (SMPS, UHSAS), and ZEP (DMPS, OPS, FIDAS^®); we filled in missing data using the nephelometer data as a constraint, as described below. For all sites, the mobility diameters measured by the SMPS or DMPS were assumed to be equivalent to volume sphere diameters. For MOS and ANX, the size distributions from each of the available data streams were re-averaged into 5-min measurements and converted to volume distributions, assuming spherical particles, to maximize structure in the overlap region. For ANX, the UHSAS optical diameters were merged with the SMPS data by applying a Nelder-Mead algorithm (Nelder and Mead, 1965) to determine a size correction factor for each optical diameter bin that optimized alignment of the distributions. For MOS, although a UHSAS was deployed in the AOS, the data quality was flagged as “suspect” and thus we did not use UHSAS data. Instead, a similar approach as described for ANX was used to obtain a size correction factor that best aligned the aerodynamic diameters measured by the APS with the SMPS size distributions. In the alignment procedures, the first several reported bins in the UHSAS and APS data streams were not utilized in the merging process, due to uncertain detection limits that affected the alignment procedure. Our approach mirrored the results of Bergner et al. (2023) and the few distributions shown in Heutte et al. (2023), in that stronger nudging of APS diameters was required beyond the basic conversion of aerodynamic to physical diameter, possibly due to calibration offsets. Once these merged distributions were developed, they were projected onto a 500-point logarithmic grid and smoothed via a linear convolution function to fill in the sparse data points. Finally, the 5-min merged distributions and the corresponding nephelometer data were averaged over the same period as the INP sampling (Table S3), generally between 24 and 48 h at ANX and 72 h at MOS. For ZEP, alignment was already quite good in the overlap region of DMPS and optical data in all cases. Smoothing and interpolation of the ZEP size distributions occurred during subsequent fitting, as described below, and resolved occasional discrepancies among the optical data.

The merged distributions for MOS spanned submicron and supermicron diameter ranges, up to about 20 µm in diameter. In contrast, the ANX distributions were limited to smaller than about 1 µm by the upper size limit of the UHSAS. We used the nephelometer data from both sites to constrain estimates of ANX supermicron aerosol size distributions, and to check the accuracy of the MOS merged distributions. The nephelometers used in the MOS and ANX campaigns alternated between a 1 µm (PM₁ or submicron) and 10 µm (PM₁₀) cut-point impactor, which removed particles above those aerodynamic cut sizes. Nephelometer data obtained when sampling through the 1 µm impactor (b_sp (PM₁)) were assumed to represent the submicron aerosol, while data obtained when sampling through the 10 µm cut impactor (b_sp (PM₁₀)) represented both supermicron and submicron particles such that b_sp(supermicron) = b_sp (PM₁₀) − b_sp (PM₁). The scattering data for each inlet were interpolated in time to span the period when the alternate inlet was being used. We then used Mie theory to compute scattering coefficients from the merged size distributions, assuming spherical particles and an appropriate assumption for index of refraction, n (see next paragraph), and selecting a diameter that separated submicron and supermicron aerosols. For ANX, we developed an estimate of the supermicron size distribution by assuming a range of reasonable lognormal distributions (Dedrick et al., 2022; Williams et al., 2024), guided by structure in the UHSAS data as discussed further below, and selecting one that resulted in a match to observed averaged scattering coefficients. It is well-known that angular nonidealities in the nephelometer result in underestimates of the measured scattering coefficients, especially for supermicron aerosols (Anderson and Ogren, 1998). We therefore modified the Mie code to sum only over angles from 7° to 170°, corresponding to the collection angles in the TSI nephelometer (Uin, 2024), so that the calculations could be directly compared with the observations. While additional uncertainty may exist due to unresolved particle losses in sample lines, we matched aerosol distribution data and nephelometer scattering that were both uncorrected for such loss estimates. The results for a wavelength of 700 nm are shown in Figure 3 and Table S3, where 2 combinations of refractive indices and sub-/supermicron separation diameter were assumed; n = 1.45 and a sub-/supermicron separation diameter of 1 µm, and n = 1.53 and a sub-/supermicron separation diameter of 0.8 µm.

We note that this approach cannot fully constrain the aerosol size distribution, because other assumptions for refractive index, separation diameter, and, for ANX, supermicron size distributions could also satisfy the observational constraints. However, we chose assumptions that are likely to bracket the range applicable to a polar, ice-covered location (MOS) and one where sea spray particles have been shown to often dominate under CAO conditions due to both production and scavenging during transit over the vast open ocean (Williams et al., 2024). Thus, for Figure 3a, we chose a refractive index of 1.45, motivated by recent observational data for the Antarctic (Jurányi and Weller, 2019), AERONET data inversions for several sites in the Arctic (Xie et al., 2018), and compiled recommendations for polar aerosols from the review of Tomasi et al. (2015). For ANX, we consider that the database of Polyanskiy (2024) lists x = 1.53 for sodium chloride while Bi et al. (2018) suggested values of n approximately equal to 1.5 for dried sea salt particles. Thus, in Figure 3d we show data for n = 1.53 and a smaller (0.8 µm) sub-/supermicron separation diameter, to account for correction of the 1 µm aerodynamic diameter cut point to a volume equivalent diameter. These latter calculations were matched to the nephelometer data to estimate supermicron aerosol distributions at ANX, as shown for the example in Figure 4. The comparisons between the Mie calculations and nephelometer observations shown in Figure 3b and 3c are for completeness and show worse agreement of the 2 sets of assumptions at the sites for which selected parameters are less justified. Computed surface area changes between the assumptions are generally small (within about 20%) for ANX but around 50% for MOS. We note that direct observations or chemical composition data supporting choices for the various sites could offer better constraints for optical closure calculations. Aerosol surface area and volume concentrations are tabulated at below and above 1 µm (physical) and in total for all CAOs at ANX and MOS in Tables S2 and S3, respectively. We may note that while our approach is effectively like Williams et al. (2024), we could not use the exact CAO periods used in that study because of the need to expand the definition of CAO periods for aerosol merging to encompass INP sampling periods.

We note the “hump” in aerosol concentrations at 0.6 to 0.7 µm and drop in aerosol concentrations in the uppermost bins of the UHSAS instrument at ANX in Figure 4. These same features have been noted in UHSAS size distributions in other marine environments and are posited to be most likely a result of under-sizing of sea spray particles by this instrument in the 0.7 to 1 µm size range due to calibration and partial drying issues (Sanchez et al., 2021; Moore et al., 2022). Uin et al. (2024) also discuss such features as fundamental artifacts that are inherent in optical measurements and advise caution on interpreting them as true features of the aerosol distribution. Through ignoring the upper tail of the UHSAS distribution and constraining with calculated supermicron scattering, as discussed above, the extended sea spray aerosol mode to sizes well above 1 µm at ANX was discerned, as shown in Figure 4.

For ZEP, the INP data sets represent samples integrated over a full week, selected to contain the CAO period. Therefore, to create corresponding size distributions covering the full range of relevant diameters, we used all 3 data streams—DMPS, OPS, and FIDAS^®—and fit lognormal modes directly to the size distribution data that were averaged over the same period and converted into aerosol surface area distributions; the fitting process simultaneously smoothed the data. Examples for CAO3 are shown in Figure 4. The corresponding surface area and volume concentrations for all CAOs at ZEP are also shown in Tables S2 and S3. The OPS and FIDAS^® data showed good correspondence except above sizes of several microns; as shown. This may relate to the more direct path of air from the FIDAS^® inlet to the optical cell. The final fit was chosen to reside between these two estimates, though favoring the FIDAS^® usual upper bound. We did not attempt optical (PM₁₀ nephelometer) closure for the averaged ZEP data due to the long integration times for these samples and the observed high variability in aerosols and scattering over those periods.

INP data utilized in this study were entirely from particles collected on filters and resuspended into water for immersion freezing measurements. The ANX INP procedures and data set are described in the report of DeMott and Hill (2021), as well as in Geerts et al. (2022), but selected details will be repeated herein. Pre-sterilized (and bagged until use) open-faced filter units were placed beneath a rain/snow shield at a height of 4 m above ground level on the AOS trailer (Figure S1) and connected to a vacuum pump. Units used clean, 47-mm Nuclepore polycarbonate filters (0.2-µm pore size, backed with 10-µm pore size filters), collected at 20–35 Lpm over 6 to 74 h (average of 34 h, 42 m³ of air filtered), with periods tailored to integrate sampling during expected CAO periods, but also collected during periods flanking these in time that were non-CAO air. These latter periods may have included trajectories passing from ocean or land regions. Four field blanks were collected during the study. The collected filters were removed from the disposable filter units and stored frozen at −20°C in sterile Petri dishes until shipping in a dry nitrogen shipper to Colorado State University (CSU).

Immersion freezing INP temperature spectra for ANX were obtained from suspensions of particles removed from filters by agitation in purified water (0.1 µm-filtered deionized [DI] water) and subsequent distribution for cooling in an array of wells in the CSU Ice Spectrometer (IS). This device is supported by well-established experimental protocols and has been applied for measurements on diverse aerosol samples (Hiranuma et al., 2015; DeMott et al., 2017; McCluskey et al., 2018b; Barry et al., 2021b). Droplet volumes (50 µL) were cooled at 0.33°C min⁻¹ and the freezing of wells was recorded from videos through a LabVIEW interface. Lowest freezing temperatures ranged from −27°C to −30°C. Cumulative INP concentrations mL⁻¹ were obtained from the proportion of frozen wells at each temperature following Vali (1971). Subtractions of DI water INPs and filter blank INPs per filter were then applied before conversion of these data to INPs per standard liter of air sampled. Considering sampled volumes of air, detection limits were approximately 0.0001 INPs sL⁻¹. Confidence intervals (95%) for binomial sampling were calculated based on Agresti and Coull (1998).

For MOS, an INP filter unit was set up atop the AOS trailer in an equivalent manner as for ANX, with the vacuum lines running to the flow meters and pumps in the container. Collection periods for MOS were standardized to 72 h. Immersion freezing measurements of these filters followed the same procedures as those for ANX, employing the IS instrument and using blank filter corrections. We used these filter samples in preference to the immersion freezing data produced from the cold plate analysis of suspensions from the time-resolved and size-resolved substrate collections with a Davis Rotating-drum Unit for Monitoring (DRUM) described by Creamean et al. (2022), following Barry et al. (2025), so that immersion freezing methods were similar/consistent for all sites considered here. We note that there was no isolation of the MOS filter unit from potential ship stack pollution. That is, the filter pump was not synched with other AOS pumps so that it would turn off when visual stack pollution was observed, and even so, there would be no way to apply additional correction to nonvisual pollution contamination. Ship stack contamination was deemed to have negligible influence on INP measurements in other deployments (e.g., McCluskey et al., 2018a; Welti et al., 2020), and inspection of processed MOS data again did not reveal a relation between stack contamination and INP concentrations. At the time of collection, every MOS filter sample was categorized with a 38-level gray-scale assignment, as shown in Figure S2. A fresh filter had a gray-scale value between 1 and 2 and a usual filter collecting ambient aerosol and no soot had a value between 2 and 3. A value of 5 indicated modest soot contamination and a value of 11 moderate influence. As a test of stack contamination influence on ice nucleation, 2 samples collected during a period of stable ambient aerosol conditions were compared. These filters, from the first 3 and last 3 days of a 9-day period during November 2019 are shown in Figure S3. Despite a 4.5 difference in gray-scale between the relatively clean first filter (2.5 gray-scale) and moderately influenced later filter (7 gray-scale), INP concentrations were nearly identical from –14°C to –27°C for the 2 samples. For MOS filter samples collected during CAOs 1 through 6, gray-scale values of 2, 1.5, 3, 5, 3.5, and 11 were determined. Thus, even the moderate influence detected in CAO6 should have had no influence on INP results.

Finally, we utilize INP data collected at the Zeppelin Observatory as described in Pereira Freitas et al. (2023) and Tobo et al. (2024). Filters from ZEP, which were the exact same type as used at other sites, were collected over 1-week intervals, using a 10-line Global Sampler (GS-10N, Tokyo Dylec Corp.) operated at a regulated flow rate of 3 L min⁻¹. This flow was also obtained from a vertical sample line connected to the PM10 inlet used for the OPS. Samples were processed using the Cryogenic Refrigerator Applied to Freezing Test (CRAFT) system (Tobo, 2016), upon return for processing at the NIPR. The CRAFT system is a cold plate device that has been demonstrated to give results in close correspondence to measurements with the IS for equivalent samples (Tobo, 2016; DeMott et al., 2017). In the CRAFT system, droplet volumes (5 µL) were cooled at 1°C min⁻¹, and lowest freezing temperatures were set to −30°C.

For ANX and MOS filter samples, 1 or 2 mL of the particle suspensions were subjected to additional freezing experiments to examine the contributions of heat labile INPs that have been attributed to “biological” INP sources (e.g., primarily proteins), and the contributions of other organic INPs (bio-organic), following methods described in McCluskey et al. (2018b) and Suski et al. (2018). For thermal testing, 2 mL of each suspension were heated to 95°C for 20 min and the sample reanalyzed to gauge the reduction in INPs. To quantify other organic INPs, typically on the same samples treated with heat, hydrogen peroxide digestions under 95°C heat were performed on additional 1 mL suspensions, followed by neutralization with catalase and reanalysis of freezing spectra for reductions in INPs. INPs remaining after peroxide digestion were classified as inorganics (likely mineral).

We also explore the relation of INPs to spherical-equivalent aerosol surface area and spherical-equivalent aerosol volume, as determined from aerosol data in different size ranges. From this, we calculate the surface-active site densities (n_s(T)) and volume-active densities (n_v (T)) as the ratio of INP concentrations to aerosol surface area and aerosol volume concentrations, respectively. In equation form,

where n_INP is INP concentration per liter of air at temperature T, S_tot is the total mean aerosol surface area for the filter period in μm² cm⁻³, and V_tot is the total mean aerosol volume in μm³ cm⁻³. n_s (T) and n_v (T) thus have units cm⁻² and cm⁻³, respectively. Since all INP and aerosol concentration data are calculated at standard conditions for this study, n_s (T) and n_v (T) are also at standard conditions. This allows comparison to existing parameterizations of n_s (T) and n_v (T) on the assumption of single INP types such as marine aerosols (McCluskey et al., 2018b; Mitts et al., 2021), as done by Moore et al. (2024) for the remote Southern Ocean.

Merged aerosol distribution measurements during the winter CAOs in 2019–2020 showed variability at the sites at different times but support a general picture that indicates aerosol surface area concentrations were dominated by submicron particles over the pack ice, transitioning to dominated by supermicron contributions to surface area concentrations by the time air reached the ANX site in Norway. As seen in the example of Figure 4 for CAO3, the different sites displayed similar aerosol distributions in the submicron regime, whereas the ANX site possessed a more pronounced mode above 1 µm diameter. Again, we deduced the presence of this enhanced sea spray aerosol mode to sizes well above 1 μm at ANX using auxiliary data, including light scattering, and estimated the contributions from supermicron particles as described in Section 2.2.4.

In Figure 5a and 5b the derived aerosol surface areas at ANX are compared to those observed at MOS and ZEP, respectively, assuming the sub- and supermicron fractions were those summed below and above 1 μm diameter, respectively. In all but one case at MOS, the ANX supermicron surface area concentrations are at least a factor of 2 larger. Concentrations at ANX are quite typical of those found in remote ocean measurements with a long fetch of flow under a range of moderate to strong winds speeds (5–25 m s⁻¹) (e.g., Moore et al., 2022). In the submicron diameter range, ANX tends to also have more surface area than ZEP, excepting one outlier event, while MOS had much more submicron surface area than either site.

INP data represented long integral time periods, and we will normalize these data using aerosol surface area and volume averaged over the same times during CAOs. The reasonableness of this assumption is an inescapable issue for this study, especially for the longer INP sampling times at ZEP. To highlight this point, we evaluated the homogeneity of aerosol properties at ZEP using data from the OPS and FIDAS^® instruments. This site represents the most extreme example for this study because the INP data records were restricted to week-long periods. Time series of dN/dLogD_p values at 2 different geometric mean diameters for all CAOs are shown in Figure S4. At most times, these 2 optical instruments gave excellent agreement on the aerosol size distribution evaluated at sizes of 700 and 2,000 nm. Size distributions were also relatively stable during the times of CAO passage, except for CAO3 and CAO6, where dN/dLogD_p normalized number concentrations similarly increased up to several times (especially at 700 nm) as the CAO progressed through this site. Consequently, there is a strong likelihood that integrated surface area is biased low for CAO3 and CAO6 at ZEP. Conservative estimates of integrated surface area uncertainty for shorter INP filter periods in the other cases and for the other sites is estimated as a factor of two.

The temperature spectra of INPs in all CAOs observed at the end location at ANX, and treatment data to help infer their compositions, are shown in Figure 6. INP spectra for all filters overlapping with ANX CAOs are shown in Figure 6a. This graphically demonstrates the rather unique nature of the INP spectra in CAOs at ANX, with a convex elevation of INPs in the temperature range from −17°C to −23°C. This spectral feature or mode contrasted with the log-linear spectra associated with sea spray aerosols (SSA) as measured and parameterized from the North Atlantic at Mace Head, Ireland (McCluskey et al., 2018b; hereafter M18), and the log-linear spectra associated with INPs measured over a full year average at a boreal forest site in Finland (S21 in Figure 6a, Schneider et al., 2021), the latter indicating expected nearby land influence. For completeness, we show INP concentrations for all CAO periods at ANX in comparison to all filter periods, which includes those bookending CAOs in Figure S5. This makes it clear that highest INP concentrations overall, especially at the highest temperatures assessed, occurred outside of CAOs. We believe that the occurrence of many other spectra resembling those of CAOs is due to the common occurrence of air flow from ocean regions even outside of defined CAO periods. Note that the M18 prediction was pinned to Figure 6a through the dependency in that parameterization on aerosol surface area as noted in the legend, and 40 μm² cm⁻³ represented a value close to the maximum sum of submicron and supermicron surface area concentrations observed in CAOs at ANX (Table S4). Figure 6b demonstrates that biological INPs only modestly influenced this −17°C to −23°C INP concentration spectral feature in most cases, based on the relative lack of sensitivity of the freezing characteristics to thermal treatment of suspensions. In fact, biological INP contributions were especially evident only above about −12°C. The general lack of heat-lability in INPs in the −17°C to −23°C range is unusual and rules out the types of INPs associated with organic exudates in ocean microlayers that have been implicated in high marine organic aerosol and INP episodes in the North Atlantic (Wilson et al., 2015). This might be expected for a time of year when marine biological activity is at a minimum due to the lack of solar radiation, if the ocean is the primary source. Nevertheless, a key result shown in Figure 6b is that the INPs are dominantly organic in composition, based on peroxide digestion impacts on the freezing spectra of the bulk suspensions. Despite the lack of quantitative agreement with M18, the absence of biological/heat-labile biogenic influence (no response to 95°C heat) and the highly organic nature of INPs was shared with the typical marine INPs quantified by McCluskey et al. (2018b), which were attributed to smaller and long-lived organic matter elements from seawater that are emitted in SSA. Figure 6a also indicates the maintenance but weakening of the special spectral INP feature as the CAO season progressed, and that all INP concentrations decreased toward the M18 parameterization limit characterizing expectations for remote ocean INPs. Nevertheless, the highly organic nature of the observed INPs during CAOs at ANX was like that found for other categories of INPs, including arable and high latitude soil dusts (O’Sullivan et al., 2014; Tobo et al., 2014; Hill et al., 2016; Tobo et al., 2019), marine sediments (Inoue et al., 2021; Porter et al., 2022; Barry et al., 2023b), and biomass burning aerosols (Schill et al., 2020; Barry et al., 2021a) that could potentially be mixed into the MBL during cloud development in CAOs. That is, it was not possible from the spectra and treatments alone to tag the unique INPs at ANX as being solely from the ocean, nor as a combination of oceanic and free tropospheric sources.

The ANX INP spectra in the specific CAOs investigated in this article are repeated in Figure 6c, isolating the results after treatment. Although dates are not distinguished, comparison of Figure 6a and 6c indicates that spectra with the highest contribution of higher-temperature, biological INPs occurred during the early season (December to January) during polar night conditions at ANX. This was interesting considering recent laboratory studies of the impact of photochemical oxidation on SSA INPs, which demonstrated degradation in INP concentrations of up a factor of 5 at any one temperature and an equivalent 2°C reduction over the entire INP spectrum (DeMott et al., 2023). Biological INPs may be especially susceptible to the impact of atmospheric oxidation processes. The general decline in INP concentrations as the season progresses from winter into early spring was also consistent with a potential degradation under additional sunlight (i.e., consistent with laboratory studies) although there could be a variety of unrelated reasons for these changes.

Time series of INP concentrations at all 3 sites, during the overall project period at ANX and for 3 activation temperatures, are shown in Figure S6. We note that there was a large dynamic range in INP concentrations across sites, on the order of 100-fold at any temperature, and that (1) the sites were most similar in INP concentrations active at −25°C; (2) the sites were most different in INP concentrations active at −20°C, with ANX standing out as having a specially enhanced population at this temperature; and (3) ZEP and ANX were more similar at −15°C. All sites converged in relative INP concentrations at −15°C toward the start of spring. Barry et al. (2025) discusses the widespread similarities of INP spectra annually in the Arctic region using an extended set of these same data from MOS and ZEP. Figure S6 also includes data from Li et al. (2022), who sampled from an aerosol container located near sea level at a site about 600 m from the coast of Svalbard in Ny-Ålesund, 2.5 km from the Zeppelin site in Spring of 2020. While these data overlapped with the overall sampling period at Zeppelin, they were collected after the last CAO that was suitable for studies in this article. Concentration discrepancies are apparent in comparing the Li et al. (2022) data with the Zeppelin measurements, the former ranging higher, especially at −15°C. These differences mimic those reported in Pasquier et al. (2022) and Li et al. (2023) when comparing freezing data of aerosol collections using high-volume liquid impingement methods (also used in Li et al., 2022) to those of standard aerosol filter collections, as used in the present study. In Pasquier et al. (2022), the filter collection and processing by the CRAFT instrument are exactly the same as in our study. Li et al. (2023) supported that this difference was potentially due to the collection of INPs that exceeded the sizes entering the PM10 inlet used for filter collections (as in our study at Zeppelin). Direct comparisons of high-volume impinger method and filter collection methods have not been made without account for collection size biases, to our knowledge. In this study, we emphasize the use of nearly identical collection methods and freezing instruments that have been shown to display close correspondence on equivalent test and ambient samples.

Comparisons of the INP concentration temperature spectra at 2°C intervals at all 3 sites for the CAOs identified as connecting the 3 sites through transport are shown in Figure 7 and Figure S7. Figure 7 separates data by CAO while Figure S7 separates data by site. All data are tabulated in Table S6. The similarity of INP spectral shape at MOS and ZEP was evident for CAOs overall, as might be expected for a common origin of air from over the pack ice. These spectra at MOS and ZEP increased exponentially with decreasing temperatures, similar to other freezing spectra based around filter collections that are reported by Li et al. (2023). Nevertheless, the INP spectra could only be considered nearly identical within confidence bounds at most temperatures at MOS and ZEP for CAO 1 and 2. INP concentrations are a few times higher at MOS compared to ZEP for CAOs 3 and 5, while the opposite order is apparent for CAO4. These differences likely reflect the different temporal integration times of the filter samples at the 2 sites that was unavoidable in attempting to define specific CAO event air in the INP data. For example, in the time series of Figure S6 it is apparent that the only filter for comparison of INPs at MOS for CAO3 came from after the time that the CAO period would have been present there. Also evident is that the unique spectral signature at ANX between −17°C and −23°C that we attributed to certain organic INPs was absent in the MOS and ZEP spectra. We also note from Figure 7 and Figure S7 that CAO2 was an outlier among the events at ANX. Rather than having the “hump” feature only at −17°C and −23°C, high INP concentrations extended to higher temperatures. This event had the lowest M-index (i.e., lowest CAO strength), and an inspection of local winds over time during the event showed weak values that sometimes came from over nearby land regions that included Boreal Forest regions. For this reason, we show the spectrum for comparison to other events but exclude it in INP normalization analyses. Finally, CAO6 showed the least distinct spectra across all sites, for reasons that are unclear.

Sampling at a nearby site to ANX, conducted in March of 2021 by Gjelsvik et al. (2025) also appears to support the presence of the “hump” feature of INP concentrations at −17°C and −23°C seen in this study. Limiting comparison of these studies (besides the timing) is the fact that Gjelsvik et al. (2025) did not focus only on CAOs, had a lower upper limit of detection, and used the same high-volume impinger method as Li et al. (2022). Nevertheless, for completeness, we include the comparison of our CAO data and those of Gjelsvik et al. (2025) in Figure S8a to show similarities and differences of INP concentration temperature spectra. In Figure S8b we repeat the comparison of the ZEP data during the time of overlap with Li et al. (2022), but as temperature spectra.

Relational analyses of INP concentrations to total aerosol surface area and volume concentrations, for specific INP activation temperatures and for each site during CAO periods, are shown in Figure 8. Data for these comparisons are in Table S4, Table S5, and Table S6. Correlations shown are exponential fits, being slightly better than linear fits. Relations between INP concentrations and total aerosol surface area or total volume concentrations were poor at the MOS (Figure 8a) and ZEP (Figure 8b) sites, based on the low correlation coefficients. In contrast, correlations at ANX were moderate to strong for all temperatures (Figure 8c). For MOS (Figure 8a), this finding may have reflected a complex aggregation of aerosol influences on INPs over the winter period that included free tropospheric aerosol of unknown sources (dust of various source types, biomass burning, continental pollution), sea salt particles accompanying occasional pollution events (Dada et al., 2022) transported northward in cyclonic systems passing over the ice pack, sea salt from brine emissions at ice ridges and leads, blowing snow, and perhaps the anthropogenic biomass burning from ship activities that motivated segregation of such effects from the aerosol record (Beck et al., 2022). We did not attempt to parse out special periods of influence of any of these factors in this study, although this may be possible with the MOS aerosol data sets. We found no clear relations between INPs and aerosol properties in these CAO events at MOS when isolating analysis to either submicron or supermicron aerosol size ranges (Figure S9). This is perhaps expected for a region that has few wintertime sources and otherwise contains an aged aerosol dominated during winter 2019–2020 by non-INP types. Similar size-dependent correlations for the ZEP data suggested weak correlations with submicron aerosol surface area and volume concentrations and stronger correlations with supermicron aerosol surface area and volume concentrations at all temperatures (Figure S10). This might be an expected result if the open water regions near ZEP were contributing moderately to overall aerosol properties, but substantially to INPs. Alternatively, free tropospheric influences occurred at the ZEP site at times. The results of size-dependent correlation analyses at ANX demonstrated, with few exceptions, weakest correlation with submicron aerosol surface area and volume concentrations but strongest correlations with supermicron and total aerosol surface area and volume concentrations (Figure S11). Thus, increasing concentrations of INPs seemed related to the increase in larger particles observed at both ZEP and ANX. For ANX, and likely for ZEP as well, this increase is presumed to result from sea spray generated in the flow over the large open ocean region between Svalbard and Norway. However, the deep mixing of air in convection during CAOs will simultaneously bring air from the free troposphere into the boundary layer, complicating a clear assessment of the source of the unique INP spectra observed at ANX, presumably characteristic also of air feeding clouds as they passed over the coast in this region. Unknown, until measurements can be made upstream over open ocean, is whether wave breaking at the coast itself, or even offshore over the coastal shelf where sediments may be suspended, exerts any influence on INP populations.

Figure 9 summarizes pertinent data demonstrating how INP concentrations and derived supermicron aerosol surface and volume INP site densities varied across the evolution of the CAOs investigated in this study, for all sites together. While the INP concentration versus temperature data at MOS and ZEP were similar and log-linear, representing pack ice and near-ice-edge regions during CAO flow from the central Arctic occurs, INP concentrations after air passed over the extensive open water region of the Barents Sea to the ANX region were elevated uniquely in the temperature range between about −17°C and −23°C (Figure 7 and Figure S7). This elevation occurred in association with a strong increase in the supermicron aerosol surface area and volume, as might be expected for a sea spray aerosol source. The dominance of sea spray aerosol is consistent with the long fetch over water in a growing boundary layer and the likely precipitation scavenging of many INPs that entered the air from higher latitudes, following from the results of aerosol distributions during the 2019–2020 CAO season at ANX documented in Williams et al. (2024). While normalization by supermicron surface area and volume drew some of the site data together, especially in the noted −17°C to −23°C temperature range, unification by these physical aerosol parameters was not achieved overall. Correspondence of INP site densities with volume concentration at most sites may reflect the contributions of a common INP type, although the poor correlations at MOS complicate such a conclusion. Future in-depth studies of aerosol and INP compositions would be helpful. Nevertheless, better agreements at lower temperatures could have reflected the influence of a broader (e.g., long range) aerosol source active at those temperatures. It was surprising that INP populations at ANX during episodes of long air fetch over the ocean were not at all described by existing parameterizations of SSA INPs, from M18 using aerosol surface area or from Mitts et al. (2021) using aerosol volume as the inferred best normalization variable. Both parameterizations fell low by at least one order of magnitude on average in describing the INP populations at ANX (Figure 9c and 9d).

Explanations for these results must necessarily be speculative. Nevertheless, as suggested in the introduction, differences associated with SSA INPs in comparison to prior observations in other global marine regions could be related to the fact that the Barents Sea is much shallower than other regions where INP measurements have occurred (Loeng, 1991), such as the North Atlantic and the Southern Ocean. Waters can become turbid due to the deep mixing initiated by high winds and due to the proximity of land soil matter sources feeding these shallower ocean regions, for example from river runoff (Dankers and Middelkoop, 2008; Shiklomanov et al., 2021). In that case, highly active organic INP components from soil sediments known to feed waters in this region (Tobo et al., 2019; Creamean et al., 2022; Barry et al., 2023a; Barry et al., 2023b) could be emitted along with SSA and distributed across the aerosol distribution in line with surface area or volume. Such a process has been implicated for dust resourcing as INPs over oceans (Cornwell et al., 2020). The factors responsible for producing the specific activity feature at –17°C to –23°C observed at ANX remain unresolved; the feature is unexpected based on INP temperature spectra reported for known terrestrial sources. In the CAO study over the Chukchi Sea by Inoue et al. (2021), a more prominent feature near −20°C was found to build in as winds increased (cf., their figure 3) and INPs were more generally heat labile at all temperatures. That finding may simply relate to the regional nature of sediments in turbid waters. No special INP spectral feature was found by Hartmann et al. (2020) in flights over the Arctic in winter, but their filters sampled entire flight periods, integrating over all vertical levels. No such feature is present as well in the recent data of Raif et al. (2024), although we have already noted that the INP concentrations they detected were far higher than observed in our study and so clearly delineate a different aerosol regime during their period of measurements. Interesting, nonetheless, is the alignment of the INP activity feature found in our study with the ice phase transition temperature of −17°C inferred over open water in winter based on satellite studies (Carlsen and David, 2022). As well, the 4°C to 6°C decrease in temperature needed to achieve similar INP concentrations (as found at −17°C at ANX during CAO periods) for particles at MOS or ZEP apparent in most cases in Figure 7 was consistent with the several degree lower ice phase transition temperatures found over Arctic ice and land regions in winter by Carlsen and David (2022).

Another possible factor affecting the endpoint INP spectra at ANX during CAOs was the mixing of free tropospheric air into the boundary layer during CAOs. While it might be expected that INP concentrations would be lower in air above clouds, this was not necessarily true during episodic transport events into the Arctic free troposphere, such as the biomass burning transports into this region noted by Ansmann et al. (2023) during MOSAiC. As mentioned previously, these INP types and arable soil types from the same distant locales that are lofted and transported into the Arctic vortex would also be expected to show the highly organic character evident in Figure 5. However, again, a specific source with exaggerated INP activity in the −17°C to −23°C range has not been demonstrated for these INP types. Furthermore, as the boundary layer deepens with fetch over water in a CAO, the impact of mixing of free tropospheric air into the growing boundary layer at higher altitudes on INPs measured at the surface must necessarily be diluted. We may note the apparent contrasting situation presented in Raif et al. (2024), when the entire lower and middle troposphere appeared to have been dominated by mineral or arable soil dust INPs over a 9-day period in March 2022 that appeared to be strongly influenced by Arctic haze. To emphasize the drastic differences between our study of 6 CAOs in 2019–2020 and those analyzed in the aircraft study of Raif et al. (2024) from March 2022, we show the surface site density parameterization of arable soil dust from O’Sullivan et al. (2014) in Figure 9c. Site densities from Raif et al. (2024) met or even exceeded the O’Sullivan et al. surface site densities in many cases, 2 orders of magnitude higher on average than observed in the present study.

One might also wonder about a potential influence of anthropogenic or biomass burning aerosols due to their presence in Arctic haze. However, Barry et al. (2021a) noted that n_s values for both black carbon and wildfire smoke most resemble those of sea spray aerosol n_s values (e.g., McCluskey et al., 2018b) shown in Figure 9, lower than values observed during the periods used in our study. Hence, it seems unlikely that anthropogenic sources, including biomass burning, could explain the elevated INP concentrations and n_s in CAOs, without alluding to an aging enhancement of INPs that has thus far been documented only in the laboratory (Jahn et al., 2020; Jahl et al., 2021).

Additional surface and airborne observations of INPs in this region over extended seasonal periods and in different years would help tremendously in mapping the spatial properties of INPs and discerning relative contributions from the ocean and other sources to observed INP concentrations, factors that may be critical for cloud phase and precipitation.

The data that support the findings of this study as represented in figures are available in tabular form within this manuscript and supplemental information provided, and in referenced data publications. Additional raw data used for creating merged aerosol distributions and all INP data are available via the U.S. Department of Energy data archive (https://www.arm.gov/data/). Raw ANX UHSAS data have DOI: 10.5439/1409033 and raw ANX and MOS SMPS data have DOI: 10.5439/1476898. COMBLE nephelometer data has DOI: 10.5439/1984401 and MOS nephelometer data have DOI: 10.5439/1228051. The MOSAiC INP data has DOI: 10.5439/1804484 and the COMBLE INP data have DOI: 10.5439/1755091.

The supplemental files for this article can be found as follows:

Tables S1–S6. Figures S1–S11. docx

The authors acknowledge that part of the work carried out, and data used in this manuscript, were produced as part of the international Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC20192020). The authors would like to thank all persons involved in the expedition of the Research Vessel Polarstern during MOSAiC in 2019–2020 (AWI_PS122_00) (Nixdorf et al., 2021). A subset of the data used in this study was obtained from the Atmospheric Radiation Measurement (ARM) User Facility, a U.S. Department of Energy (DOE) Office of Science User Facility Managed by the Office of Biological and Environmental Research. The authors wish to acknowledge the hard work of the Field Instrument Deployments and Operations team at the DOE Los Alamos National Lab for the successful deployment of the AMF suites and our INP collection instruments during COMBLE and MOSAiC. Special thanks to COMBLE lead technician, David Oaks, for assuring the quality of the ANX INP data collections. We thank the staff of the Norwegian Polar Institute for assistance on data provided by YT, PZ, RK, and DH-R at the Zeppelin Observatory, we thank the staff of the Norwegian Polar Institute. Guangyu Li is acknowledged for kindly providing formatted data from his Svalbard studies.

This research was supported primarily by the U.S. Department of Energy’s Atmospheric System Research (ASR), an Office of Science Biological and Environmental Research program, under Grant No. DE-SC0021116. Partial support related to MOS data used was from ASR Grant Nos. DE-SC0022046, DE-SC0019745, and Atmospheric Radiation Measurement program Grant No. DE-AC05-76RL01830. PZ and RK acknowledge the Swedish Research Council (grant no. 2018-05045, PZ), the Knut och Alice Wallenbergs Stiftelse (ACAS project grant no. 2016.0024, RK), the Swedish Environmental Agency (Naturvårdsverket), funding agency FORMAS, and support by ACTRIS-Sweden. IFB and JS received funding from the Swiss National Science Foundation (grant no. 200021_188478) and the Swiss Polar Institute (grant no. DIRCR-2018-004). Further funding was received from the European Union’s Horizon 2020 research and innovation program under grant agreement no. 101003826 via project CRiceS (Climate Relevant interactions and feedbacks: the key role of sea ice and Snow in the polar and global climate system) and JS holds the Ingvar Kamprad chair for extreme environments research, sponsored by Ferring Pharmaceuticals. This work was partly supported by JSPS KAKENHI (JP19H01972 to YT), the Arctic Challenge for Sustainability II (ArCS II) Project (JPMXD1420318865 to YT), and the Environment Research and Technology Development Fund (JPMEERF20172003, JPMEERF20202003, and JPMEERF20232001 to YT) of the Environmental Restoration and Conservation Agency of Japan.

The authors have no competing interests to declare.

PJD conceived the study. BES and SMK contributed to the conception and design of the study. PJD, BES, TCJH, JMC, YT, IFB, DH-R, GPF, and SMK contributed to analysis and interpretation of data. All authors contributed to the revision of the written content of the manuscript and all authors approve submission.

How to cite this article: DeMott, PJ, Swanson, BE, Creamean, JM, Tobo, Y, Hill, TCJ, Barry, KR, Beck, IF, Frietas, GP, Heslin-Rees, D, Lackner, CP, Schmale, J, Krejci, R, Zieger, P, Geerts, B, Kreidenweis, SM. 2025. Ice nucleating particle sources and transports between the Central and Southern Arctic regions during winter cold air outbreaks. Elementa: Science of the Anthropocene 13(1). DOI: https://doi.org/10.1525/elementa.2024.00063

Domain Editor-in-Chief: Detlev Helmig, Boulder AIR LLC, Boulder, CO, USA

Knowledge Domain: Atmospheric Science

Part of an Elementa Special Feature: The Multidisciplinary Drifting Observatory for the Study of Arctic Climate (MOSAiC)