WRF-Chem model version 3.9.1.1 is used in this study. It is a fully coupled, online meteorological and chemical transport mesoscale model . Recent improvements to the WRF-Chem model over the Arctic are included in the version used here . The model setup, including meteorological and chemical schemes, is shown in Table 1. Briefly, Yonsei University (YSU – boundary layer), Model Version 5 similarity (MM5 – surface layer), and the Noah-Multiparameterization land surface model (Noah MP LSM) are used. More details about the Noah-MP scheme are given in Appendix A.

The well-known processes for aerosols in the atmosphere, like nucleation, evaporation, coagulation, condensation, dry deposition, aerosol–cloud interactions, and aqueous chemistry, are included in the Model for Simulating Aerosol Interactions and Chemistry (MOSAIC; ) aerosol scheme used here. MOSAIC treats all the major aerosol species, such as SO42-, NO3-, Cl-, ammonium (NH4+), Na+, calcium (Ca2+), BC, and OAs. Reactive inorganic species such as potassium (K+) and magnesium (Mg2+) are not modelled in MOSAIC. The size distribution of each aerosol species is represented by eight bins, from 0.0391 micrometres (µm) to 10 µm: [0.0391 to 0.0781], [0.0781 to 0.1562], [0.1562 to 0.3125], [0.3125 to 0.625], [0.625 to 1.25], [1.25 to 2.5], [2.5 to 5.0], and [5.0 to 10.0] in µm. Each bin is assumed to be internally mixed, and both mass and number are simulated. As a result, aerosols are aged when emitted (coagulated with other species). The MOSAIC version used in this study also includes 18 irreversible heterogeneous reactions (see Table 1 in ), such as the reaction of HNO3 on sodium chloride (NaCl) to form sodium nitrate (NaNO3), with depletion of Cl-. The reaction between NaCl and sulfuric acid (H2SO4) to produce sodium sulfate (Na2SO4), with associated Cl- depletion, is also included. Nighttime chemistry, notably heterogeneous hydrolysis of dinitrogen pentoxide leading to HNO3 formation, is also included . The applied MOSAIC version includes secondary organic aerosol (SOA) formation from the oxidation of anthropogenic and biogenic species and is combined with SAPRC-99 gas-phase chemistry. In the base model, OAs are the sum of SOA and anthropogenic emissions of organic matter (OM). Aqueous chemistry in grid-scale and subgrid-scale clouds is also included. Aerosol sedimentation in MOSAIC is calculated throughout the atmospheric column based on the Stokes velocity scheme, as described in . Wet removal of aerosols by grid-resolved stratiform clouds (precipitation) includes in-cloud and below-cloud removal by rain, snow, and graupel by Brownian diffusion, interception, and impaction mechanisms following and . Wet removal due to subgrid-scale convective clouds is also included in this MOSAIC version and is described in previous studies .

Anthropogenic emissions are from the Evaluating the Climate and Air Quality Impacts of Short-Lived Pollutants version 6 (ECLIPSE v6b) inventory, with a resolution of 0.5∘×0.5∘ . Emissions of dimethyl sulfide (DMS) and lightning nitrogen oxides (NOx) are calculated online in the model (see , and references therein). Dust emissions in MOSAIC are calculated following . Biogenic emissions for 2014 are calculated online using the Model of Emissions of Gases and Aerosol from Nature (MEGAN) model . Details about the treatment of SSA emissions and their improvement in the model are provided in Sect. 4 and are summarised in Table 1.

Two simulation domains on a polar stereo-graphic projection are used in this study, as shown in Fig. 1. The first (parent) domain (d01) covers a large part of the Northern Hemisphere with 100km×100 km horizontal resolution. The boundary and initial conditions are derived from the National Centers for Environmental Prediction Final meteorological reanalysis data (NCEP FNL) and the Model for OZone And Related chemical Tracers (MOZART; ) for atmospheric trace gases and aerosols. The nested domain (d02), run at a horizontal resolution of 20km×20 km, covers continental Alaska, a small area of northwest Canada, and the Chukchi Sea and Beaufort Sea (see Fig. 1). A total of 50 vertical levels and grid nudging are used for the 100 km resolution domain, while spectral nudging, following , is implemented in the nested domain. WRF-Chem temperatures and winds are nudged at each dynamical step toward the reanalysis and are updated every 6 h above the atmospheric boundary layer.

Two simulations at 100 km are performed, one using the base model version (CONTROL) and one using CONTROL plus improvements to the SSA emissions (HEM_NEW; see Sect. 4). They are run for 4 months from November 2013 until the end of February 2014, with the first 2 months considered to be a spin-up. The results are evaluated over the wider Arctic in Sect. 5. Results from HEM_NEW are then used as boundary conditions for regional runs over northern Alaska at 20 km for two different periods (23–28 January and 24–28 February 2014) corresponding to the Utqiaġvik campaign described earlier , referred to as KRP18 and KRP19 from now on (see also Sect. 3.2). In this case, the model is run for 4 d prior to the beginning of each campaign, and this time is considered to be a spin-up. Results from a series of sensitivity runs examining processes that affect SSAs on a regional scale are discussed in Sect. 6. In all runs, model results are output every 3 h.

Surface mass concentration data (for aerodynamic diameters (defined as da) < 10 µm) from EMEP (European Monitoring and Evaluation Programme), available via EBAS, for Zeppelin, Ny-Ålesund, Norway (78.9∘ N, 11.9∘ W), and for Alert, Canada (82.5∘ N, -63.3∘ W), are used to evaluate the 100 km model simulations together with total suspended particulate (TSP; cutoff at 20 µm) data from Villum Research Station, Station Nord, Greenland (81.6∘ N, -16.7∘ W), referred to as Villum from now on. The data are collected on a daily (Zeppelin) and weekly (Villum, Alert) basis. At Alert, observations of Na+, Cl-, NO3-, and total SO42- measured with ion chromatography are used . This is also the case at Zeppelin . At Villum, the same observations are collected using a filter pack over a week and are analysed using ion chromatography (Cl-, total SO42-) and cation ion chromatography (Na+). For all the EBAS stations, observed inorganic aerosols (total SO42-, NO3-) are converted to mass concentrations (µgm-3) for comparison to model results using the ratio of the molar weights of NO3- and SO42- to the molar weights of nitrogen and sulfur, respectively. With regard to measurement uncertainties, EBAS documentation notes that uncertainties range between 33 % and 36 % for Na+, total SO42-, NO3-, and Cl- at Alert. These high uncertainties may be related to uncertainties in the size cut-off of sub-micron filters. Uncertainties in coarse-particle observations (da<10 µm) are based on the difference between high-volume (TSP) filters collected outside and sub-micron filters collected inside.

Fine-mode (da≤2.5 µm) mass concentration data from the Interagency Monitoring for Protected Visual Environments (IMPROVE) database are also used for model evaluation for the sub-Arctic site Simeonof (55.3∘ N, -160.5∘ W) on the Aleutian Islands, south of Alaska, and for an inland site, Gates of the Arctic (66.9∘ N, -151.5∘ W), referred to as GoA from now on, which is located 391 km southeast of Utqiaġvik town in northern Alaska (see Fig. 1). The samples are collected on-site over the course of 24 h every 3 d . At these two sites, observations of Na+, Cl-, organic carbon (OC), NO3-, and total SO42- are used. To compare with the OC observations at the two Alaskan sites, modelled OA is divided by 1.8, the ratio of OM/OC reported in the documentation for these two stations . Sub-micron (da<1.0 µm) and super-micron (1.0<da<10 µm) surface mass concentration data from the National Oceanic and Atmospheric Administration (NOAA) Barrow Observatory (71.3∘ N, -156.8∘ W), near Utqiaġvik town (Utqiaġvik from now on), are also used in this study, with daily and weekly temporal coverage, respectively. The sampling site is located 8 km northeast of Utqiaġvik and 3 km southwest of the Arctic Ocean. It is covered with snow during winter and is 20 m above mean sea level (m.s.l.), with a prevailing, east–northeast wind off the Beaufort Sea. Na+, Cl-, NO3-, and total SO42- mass concentrations are determined by ion chromatography and are sampled only for wind directions between 0 and 130∘ (with 0∘ indicating north). According to , measurement uncertainties in sub-micron components of SSAs and nss SO42- are below 1.0 % ± 6.1 % µgm-3 (concentration ±95 % uncertainty). The uncertainties in sub-micron NH4+ are below 0.2 % ± 7.8 % µgm-3. The uncertainties in super-micron aerosols can be up to 7 times higher than for sub-micron aerosols, since 7-daily sub-micron samples are collected for every (weekly) super-micron sample. Measurement uncertainties are also due to sampling losses in the inlets. At Utqiaġvik, for particles with a diameter up to 10 µm, losses in the inlet system from all loss mechanisms are estimated to be less than 10 %, and losses of particles with diameters between 0.01 and 1 µm are below 5 % .

At each site, observed ss SO42- is calculated from observed Na+ concentrations and the mass ratio of SO42- to Na+ in seawater of 0.252 , and nss SO42- is the difference between total SO42- and ss SO42-. Note that, in some cases, observed nss SO42- has small negative concentrations due to the depletion of ss SO42- through fractionation processes . We note that, apart from the sub-micron observations at Utqiaġvik and fine-mode observations from the IMPROVE database, there are no other sub-micron or fine-mode observations collected routinely in the Arctic, as has also been reported recently by . Finally, data from a scanning mobility particle sizer (SMPS) located at Utqiaġvik are used, providing measurements of particle number distributions at high temporal resolution, with a size range from 8.6 to 985 nm (0.0086 to 0.985 µm). No wind speed criteria have been applied to exclude local (Utqiaġvik town) or regional (e.g. North Slope of Alaska oil fields) pollution. Here, only SMPS observations at Utqiaġvik are used, since observations at the other Arctic sites used in this study are not available for winter 2014, as discussed by .

The model Stokes aerosol diameters equivalent to da, estimated using the formula, are used. Thus, the diameter of modelled sub-micron particles is up to 0.73 µm (including the first four MOSAIC bins and a fraction of the fifth bin), and super-micron particle diameters are between 0.73 to 7.3 µm (a fraction of the fifth bin, the sixth and seventh bins, and a fraction of the eighth bin). Seven MOSAIC bins and a fraction of the eighth bin are used (modelled Stokes ra≤7.3 µm) to compare with Alert and Zeppelin observations (da<10 µm). All model aerosol bins are used to compare with observations at Villum, where the observations are reported as TSP. For each site, modelled aerosols are estimated for the same conditions (temperature, pressure) as the reported observations. Also, observed total OC is assumed to include SOAs, anthropogenic OA emissions, and marine organics. Thus, from now on, it will be referred to as tOC to distinguish it from OAs defined earlier and from OM.

Details about the field campaign (23–27 January and 24–28 February 2014) measurements near Utqiaġvik, Alaska, can be found in KRP18 and KRP19. Briefly, atmospheric particles were collected using a rotating micro-orifice uniform deposition impactor located 2 m above the snow surface at a site located 5 km across the tundra from the NOAA Barrow Observatory and inland from the Arctic Ocean. The sampled particles were analysed by computer-controlled scanning electron microscopy with energy-scattering X-ray spectroscopy (CCSEM-EDX) to determine the individual particle morphology and elemental composition. The analysed samples were collected either during the daytime or the nighttime and only when wind directions were between 75 and 225∘ to minimise local pollution influence. Data analysis provided information about the different chemical components as a fraction of the total number of particles sampled during the campaign.

SSA emission fluxes (F) in MOSAIC are calculated per particle radius r, with 1000 sub-bins per MOSAIC bin, assuming that sea salt is a simple mix of pure NaCl and using the density function dF/dr (in particlesm-2s-1µm-1) based on , referred to as G97 from now on. The G97 source function represents the rate that seawater droplets form per unit area (sea surface) and as a function of particle radius. The fraction of Na+ is calculated using the molar weight of Na+ and Cl-, and then the fraction of Cl- is estimated, with the total being equal to 1. The G97 density function derived from the source function is based on laboratory experiments described in , referred to as MO86 from now on. The following expression is used: 1dFdr=1.373×U103.41×r-3(1+0.057×r1.05)×101.19e-B2, where F is a function of U, the 10 m elevation wind speed, r is relative humidity (RH) at 80 %, and B=(0.380-log⁡r)/0.650. The source function is applied for particles with dry diameters of 0.45 µm or more (equivalent to model particle diameters). For particles with dry diameters less than 0.45 µm, a correction is applied based on reported data in , since G97 overestimates the production of small particles . G97 is based on the whitecap method, where the emission flux scales linearly with the fraction of the ocean area covered by whitecaps. Over open ocean, the whitecap fraction, W(U), is determined as a function of wind speed (; MO80 from now on): 2W(U)=3.84×10-6×U103.41.

This expression for W(U) is included implicitly in Eq. (1) following details provided in MO80. In the base version, SSA emissions are calculated for every grid cell that is open ocean or a salt-water lake. In this study, grid cells which are covered by sea ice are considered together with the fraction that is ice-free. In this way, SSA emissions from open leads are taken into account. It can be noted that this SSA scheme, based on , is still being used in global and regional models (e.g. Community Multiscale Air Quality (CMAQ) ), the Goddard Earth Observing System (GEOS)-Chem, or in other models (e.g. LOTUS-EUROS; ) to simulate SSAs despite being relatively old. However, modelling studies have shown that G97 overestimates super-micron SSAs (e.g. , referred to as JA11 from now on) or underestimates sub-micron SSAs (e.g. ).

Here, updates to the model treatment of SSA emissions are described. They are included in the run HEM_NEW, which is also used as boundary conditions for the higher-resolution runs over northern Alaska.

Average absolute differences in super-micron and sub-micron inorganic aerosols between the HEM_NEW and CONTROL are shown in Figs. 2 and 3, respectively. HEM_NEW simulates less super-micron Na+ by up to 20 µgm-3 and less Cl- by up to 30 µgm-3, especially south of Alaska and over the northern Atlantic Ocean. This is due to the combined effect of using a lower wind speed dependence and including the SST dependence (Fig. 2). Inclusion of the SST dependence leads to a larger decrease in locally produced super-micron Na+ and Cl- over the Arctic and sub-Arctic ice-free regions due to lower temperatures north of 50∘ N compared to using the lower wind speed dependence based on SAL14, which has a smaller effect. Overall, one-third of the super-micron reductions can be attributed to the lower wind speed dependence, and two-thirds can be attributed to the SST dependence. Super-micron NO3- is also lower (by up to 1.0 µgm-3) due to less formation of NO3- via heterogeneous uptake of HNO3 on SSAs. These reactions involving heterogeneous uptake of acid gases also produce HCl, thus depleting Cl- relative to Na+ . The presence of sea ice also plays a role. Smaller decreases in Na+ and Cl- are found north of Alaska (Beaufort Sea) compared to in ice-free regions such as the northern Atlantic Ocean. The local influence of sea-ice fraction and open leads on SSA production over northern Alaska is examined further in Sect. 6. Furthermore, due to the addition of marine organics and ss SO42- in HEM_NEW, there is more super-micron SO42-, by up to 2 µgm-3, and more super-micron OA, by up to 0.6 µgm-3, over marine regions. Super-micron NH4+ increases slightly, up to 0.15 µgm-3, over regions where NO3- increases.

There are smaller decreases in HEM_NEW sub-micron Na+ compared to in CONTROL by up to 0.25 µgm-3, south of Alaska and in the North Atlantic (Fig. 3). Again, this is due primarily to the introduction of the SST dependence. When using the SALI14 lower wind speed dependence alone, there is a small decrease in sub-micron Cl- and a small increase in sub-micron Na+ over the Arctic. Sub-micron Cl- also decreases over continental areas, where NO3- and HNO3 are higher due to anthropogenic sources (Fig. 3). Heterogeneous uptake on SSAs reduces Cl- and increases sub-micron NO3- by up to 6.0 µgm-3 in HEM_NEW over continental regions, while the increases over the Arctic Ocean are smaller. This is in contrast to super-micron NO3- decreases. These results are consistent with the study of , who also used WRF-Chem with MOSAIC and who noted that, since SSAs are primarily present in the coarse (super-micron) mode, this favours the formation of NaNO3, which is thermodynamically stable, and limits the formation of NH4NO3, which is semi-volatile . Therefore, lower super-micron SSAs in HEM_NEW result in less super-micron NO3- and more sub-micron NO3-. We also note that, for these reasons, sub-micron NH4+ also increases by up to 1.5 µgm-3, especially over continental areas, and displays similar regional patterns to sub-micron NO3-. The inclusion of marine organics linked to SSAs leads to increases in sub-micron OA, by up to 1.5 µgm-3, and total SO42- increases due to the addition of ss SO42-.

Model results are evaluated against available observations of aerosols at different sites, as shown in Figs. 4–6. These figures are grouped according to the size ranges of the measurements at the different sites, as discussed in Sect. 3.1. Average biases and root-mean-square errors (RMSEs) between the observations and the model results for January and February 2014 are given in Tables 2 and C.1 (Appendix C), respectively. In the following, the main findings are discussed by aerosol component.

SSAs (Na+ and Cl-). Updates to the treatment of SSA emissions in HEM_NEW greatly improve modelled SSAs over the Arctic, with notable reductions in Na+ and Cl- biases and RMSEs compared to observations at Alert, Zeppelin (da<10 µm), Villum (TSP), Gates of the Arctic (GoA; fine mode), and the sub-Arctic site Simeonof (fine mode). Overall, HEM_NEW captures the spatial variability between observed Na+ and Cl- at the different sites, particularly the lower observed concentrations at Villum, which is surrounded by sea ice at this time of year, and the higher concentrations at Simeonof and Zeppelin. The extent to which sea ice is present near different sites is an important factor. For example, the high variability in modelled SSAs at Villum at the end of January and in the middle of February 2014 is likely to be due to fluctuations in sea-ice fractions around the site (0.9–1.0 in the FNL analyses). At Utqiaġvik, the model captures better super-micron Na+, whereas Cl- is now underestimated due to Cl- depletion. Sub-micron Na+ and Cl- are still underestimated in HEM_NEW at this site, with average biases of about -0.5 µgm-3 for Na+ and -0.12 µgm-3 for Cl-, with higher biases during episodes with elevated observed SSAs. Here, sub-micron SSAs may have been transported to the Arctic from the Pacific Ocean ; thus, model underestimations may point to deficiencies in the SSA source function further south or to issues related either to long-range transport or to wet- and dry-deposition treatments in the model. However, the fact that the model agrees better with observations over the wider Arctic, as well as at the sub-Arctic Simeonof, provides confidence in the modelled long-range transport as a source of Arctic (sub-micron) SSAs. HEM_NEW Na+ also compares well with reported weekly averaged sub-micron Na+ mass concentrations collected during January and February 2014 at Alert (0.1 µgm-3 observed; ; up to 0.08 µgm-3 modelled). We also note that, at Utqiaġvik, while attributed sub-micron SSAs to long-range transport, KRP18 estimated that, in the sub-micron range, for their analysed samples, 42 % were fresh SSAs with chemical signatures similar to seawater, 18 % were classed as partially aged with enhanced anthropogenic components (S, N) and depleted Cl-, and the remainder included organics and sulfate particles. Thus, model discrepancies may also be due to local processes that influence SSAs over northern Alaska. This is investigated further in Sect. 6.

Nitrate. Improved SSA treatments in HEM_NEW also lead to improved simulation of NO3- at some sites, notably at Simeonof, GoA, Alert, and Villum, and biases are reduced (see Table 2). While modelled super-micron NO3- at Utqiaġvik is improved, the model still underestimates certain periods when elevated sub-micron NO3- is observed, which is also the case at GoA and Simeonof. The improved behaviour of modelled NO3- is, in general, due to reductions in Na+ and Cl-, leading to less NO3- production in the coarse mode, especially close to or just downwind of major anthropogenic emission regions at mid-latitudes, and to a shift to more NO3- in the fine mode, as discussed previously. These effects are most evident at Utqiaġvik, where the model can be compared to sub- and super-micron data. Comparison with data from other sites is with either total, coarse- plus fine-mode, or fine-mode aerosol observations and therefore includes both increases and decreases in simulated NO3-. Overall, these results illustrate the importance of correctly simulating SSAs and their effects on anthropogenic aerosols. While observed NO3- concentrations are generally lower than other aerosol components, such as Na+, Cl-, or nss SO42-, during Arctic winter, a recent trend analysis study showed that NO3- is clearly increasing at Alert, especially during the winter months . Such increases in NO3- may be due to increased NO3- formation due to lower acidity following sulfur dioxide (SO2) reductions that outweigh reductions in NOx emissions at mid-latitudes . However, increases in SSAs over the Arctic Ocean due to reductions in ice-covered waters may also explain these changes (e.g. ), although no significant trends in Na+ have yet been detected .

Sulfate. Figures 4–6 show observed ss SO42- and total SO42- together with results from CONTROL and HEM_NEW. With regard to total simulated SO42-, the addition of ss SO42- improves the model results, for example, at Simeonof, where observed fine-mode ss SO42- makes a significant contribution (30 %–80 %, up to 0.3 µgm-3) to total SO42-; ss SO42- also contributes between 10 %–40 % of the total SO42- at Alert and Villum, and modelled ss SO42- agrees better with the observations. The remainder is nss SO42-, a dominant component of Arctic haze resulting from long-range transport from sources in Russia and Europe at these sites . Model results are at the lower end (up to 0.3 µgm-3) of reported sub-micron nss SO42- mass concentrations (0.3–1.1 µgm-3) at Alert during winter 2014 . On the other hand, HEM_NEW further overestimates total observed SO42- at Zeppelin due to the inclusion of ss SO42-, especially during certain episodes with elevated concentrations. We note that Zeppelin is a mountain site at 471 m, and thus discrepancies compared to the observations may also be due to issues simulating the vertical distribution and transport of nss SO42- from Eurasian source regions . At Utqiaġvik, on the northern coast of Alaska, most of the total observed super-micron SO42- is ss SO42- (up to 0.18 µgm-3, around 80 %), and the inclusion of ss SO42- in HEM_NEW improves agreement with the observations. With regard to total sub-micron SO42-, high mass concentrations are observed at Utqiaġvik compared to at other Arctic sites, consisting mostly of nss SO42- and peaking at 2.4 µgm-3, which is much higher than the total super-micron SO42- (peaking at 0.5 µgm-3), as also reported by . However, the model underestimates nss SO42- at this site. As noted by KRP18 and KRP19, there is a local influence from the North Slope of Alaska (NSA) oil fields to the east, and these emissions may be underestimated in the model. In a companion paper investigating BC at Utqiaġvik, it is estimated that up to 30 %–50 % of BC originates from these regional emissions . Indeed, at GoA, 391 km inland from the coast, and south of the NSA oil fields, the contribution of nss SO42- is more important, and ss SO42- is negligible. Here, the model captures total fine-mode SO42- (peaking at up to 0.64 µgm-3), and the addition of ss SO42- does not affect the results. In addition to local sources, difficulties simulating sub-micron nss SO42- at Utqiaġvik may be due to an underestimation in the transport of nss SO42- from mid-latitudes to the Arctic or issues related to deposition (as noted earlier for SSAs), as also discussed in previous studies such as that of . SO42- formation mechanisms under dark, cold winter conditions may also be lacking in the model. For example, high concentrations of hydroxymethane sulfonate (HMS) have been measured recently during winter in Fairbanks, Alaskan Interior , contributing to secondary SO42- formation during Arctic winter, although only a small contribution from HMS to SO42- was found in observations at Oliktok Point, situated within the NSA oil fields . Oxidation of SO2 by O3 in alkaline SSAs could also contribute up to 9 % to SO42- formation . However, the version of WRF-Chem used here does not include such reactions, in common with many chemistry–aerosol models run over the Arctic .

Ammonium. NH4+ observations are available at all sites except for Simeonof and GoA. Observed NH4+ concentrations are very low (below 0.2 µgm-3) at Alert and Villlum, with higher concentrations observed at Zeppelin. Overall, there is a good agreement between the model and measurements, with very low biases and RMSEs in both runs, apart from an underestimation of elevated NH4+ at Zeppelin. At Utqiaġvik, there is good agreement with super-micron NH4+, except for periods with higher observed NH4+ (up to 0.1 µgm-3). However, the model underestimates periods with elevated sub-micron NH4+ of up to 0.4 µgm-3, which is higher compared to the other sites. Temporal variations in NH4+ during January and February 2014 generally follow nss SO42-, as NH4+ preferentially forms ammonium bisulfate and, to a lesser extent, ammonium sulfate in the particle phase , and they have common anthropogenic origins. Previous studies also noted that NH4+ is 2 times higher at Utqiaġvik than at Alert, Zeppelin, and Villum, while SO42- is similar at all the sites , possibly suggesting differences in aerosol acidity at the different sites. This is also found in this study based on the observations and model results (HEM_NEW). It is therefore interesting to investigate the effect of the improved SSA emissions on modelled aerosol acidity. For this, we estimate the neutralisation factor f following . The results are discussed in Appendix D. CONTROL tends to predict more acidic aerosols than observed. Based on the observations, most acidic aerosols are found at Alert, Zeppelin, and Utqiaġvik (super-micron), with somewhat less-acidic aerosols at Villum and Utqiaġvik (sub-micron). This is improved to some degree in HEM_NEW, with aerosols becoming less acidic at some sites, notably at Alert and Villum, due to decreases in simulated NO3-. However, modelled sub-micron aerosols at Utqiaġvik are less acidic than the observations due to the underestimation of nss SO42-. Overall, the updates to SSA emissions lead to somewhat-less-acidic anthropogenic aerosols over the Arctic, again highlighting the importance of interactions between SSAs and other inorganic aerosols.

Organic aerosols. Only two sites provide tOC fine-mode observations ranging from 0.15 to 0.3 µgm-3 at Simeonof and from 0.15 to 0.5 µgm-3 at GoA during January and February 2014. The inclusion of marine organics in HEM_NEW improves modelled OAs, especially at the coastal Simeonof site. Since observations at other sites are not available for winter 2014, results are compared with other reported measurements. reported sub-micron OAs at Utqiaġvik of around 0.3 µgm-3 during winter 2008 (November to February). However, a more recent study by reported lower wintertime OA concentrations (da<10 µm) at this site (around 0.1 µgm-3), attributed mostly to primary anthropogenic or haze OAs originating from Eurasia. However, modelled OA for the same size range is only up to 0.05 µgm-3. At Villum, also reported higher sub-micron OA observations, peaking at 2.2 µgm-3 in February 2015, attributed mostly to Arctic haze influence (up to 1.1 µgm-3), with secondary influences from hydrocarbon-like organics (up to 1.0 µgm-3) and marine sources (up to 0.2 µgm-3). Modelled OA in HEM_NEW at this site does not exceed 0.1 µgm-3. Overall, the model underestimates Arctic OAs, in common with many other models . These discrepancies may be due to missing or underestimated anthropogenic or natural sources. For example, it is known that there are large uncertainties in anthropogenic OA emissions . The possibility of a wintertime marine OA source over northern Alaska is explored further in Sect. 6.

Lastly, we assess the contribution of SSAs to total inorganic aerosols in the Arctic during wintertime, since previous studies noted that they can make an important contribution to total sub-micron and super-micron mass fractions at this time of year . also showed that SSAs dominate wintertime PM10 (particulate matter with da≤10 µm) mass concentrations at remote Arctic sites, including Alert (56 %), Baranova (northern Russia; 41 %), Utqiaġvik (66 %), Villum (32 %), and Zeppelin (65 %). In contrast, at sites such as Tiksi (northern Russia) and Pallas (Finland), SO42- and OAs dominate (70 % and 55 %, respectively). To investigate the contribution of SSAs to total mass concentrations during the period of this study, observed and modelled fractions of SSAs to total (SSAs plus inorganic) aerosols are estimated (see Table 3). It should be noted that this fraction varies between sites, since not all components were measured. Taking into account the observations available at each site, the fraction of SSAs to total SSAs plus inorganics is higher at all the coastal sites (Utqiaġvik, Alert, Simeonof, and Villum) and at Zeppelin, ranging from 54 % to 93 %. Only at the GoA and Villum is the fraction of SSAs smaller (20 % and 32 %, respectively). SSA fractions, calculated using the HEM_NEW results, show similar patterns compared to the observations, with fractions ranging between 44 % and 84 %. An exception is Utqiaġvik, where the modelled fraction is lower than in the observations due to low simulated sub-micron SSA concentrations. When taking into account all aerosol components in the model, including OAs, BC, and dust, SSA is dominant at Simeonof, Utqiaġvik (super-micron), Zeppelin, and Villum (more than 54 %), whereas at Alert, SSAs contribute about 45 %. This analysis shows that SSAs are an important fraction of total inorganic aerosols at most Arctic coastal sites during wintertime.

Overall, the results presented here show that the simulation of Arctic SSAs and of other inorganic and organic aerosols is improved as a result of the updated SSA emission treatments. In particular, simulated aerosols, including the coarse-mode or super-micron fraction, are improved compared to the observations. The results also show that it is important to include natural SSA emissions of ss SO42- and marine organics, although the latter are highly uncertain. Missing anthropogenic sources could also contribute to the underestimation of OAs and nss SO42-. Many models in a recent Arctic Monitoring and Assessment Programme (AMAP) model evaluation of Arctic composition also showed similar discrepancies, attributed to issues with anthropogenic emissions or model transport, deposition, and aerosol formation . The results presented here also confirm the importance of interactions between SSAs and other inorganic aerosols via heterogeneous uptake, affecting mass concentrations and size distributions, notably NO3-, and thus model ability to capture wintertime Arctic haze.

The F10 parameterisation used in the 100 km HEM_NEW run is based on the C : Chl a ratio from a cruise at mid-latitudes. Whilst phytoplankton blooms may not be expected in the high Arctic winter, previous studies have shown evidence of sea-ice biological activity under low-light conditions coupled with decreased sea ice in the Arctic . Analysis of data collected over the Arctic and North Atlantic during winter and the winter–spring transition also showed that the majority of sub-micron OM is highly correlated with Na+ concentrations . More specifically, , referred to as RUS10 from now on, analysed samples from the International Chemistry Experiment in the Arctic Lower Troposphere (ICEALOT) cruise and found that most OM in the North Atlantic and the Arctic is composed of carbohydrate-like compounds containing organic hydroxyl groups from primary ocean emissions. , referred to as FRSS14 from now on, investigated the sources and composition of atmospheric marine aerosol particles based on the analysis of various samples, including ICEALOT, reporting that ocean-derived organic particles include primary marine OAs. In particular, they calculated the ratio of OC : Na+ as a metric for comparing the composition of model-generated primary marine aerosols and seawater and reported OC : Na+ ratios of 0.45 in atmospheric marine aerosol particles. KRP19 also reported that, during their campaign in 2014, almost all individual SSAs had thick organic coatings made up of marine saccharides, with average C:Na mole ratios of 0.5 and 0.3 for sub-micron and super-micron SSAs, respectively. They also identified open sea-ice leads enriched with exopolymeric substances as contributors to OAs in winter SSAs. Here, elemental fractions for sub- and super-micron aerosols, sampled during the KRP19 campaign, are used to better constrain modelled marine OC emissions. The ratio of sub- and super-micron OC : Na+ is calculated following FRSS14 and using the elemental fractions from KRP19 as an indicator of the presence of a local source of marine organics. The organic fraction of SSA emissions in WRF-Chem is increased from 0.2 to 0.4 for first and second MOSAIC bins, from 0.1 to 0.4 for the third MOSAIC bin, and from 0.01 to 0.11 for the remaining MOSAIC bins. Note again that no additional SSA mass is added.

Figure 7 shows the sensitivity of the model results to including a larger marine organic fraction over the regional domain. Sub-micron OA concentrations increase by a small amount, by up to 0.009 µgm-3, especially southwest of Alaska and along coastal areas, including around Utqiaġvik. There are only two available daily observations at GoA during the February simulation period to evaluate the model results. The model better captures observed tOC at the end of February in the run (LOC_ORG_FEB) with higher organic fractions (not shown here). However, it underestimates tOC on 25 February when the observed tOC reached 0.33 µgm-3. As mentioned previously, this discrepancy could also be due to missing local anthropogenic OA sources. Higher OA fractions in the super-micron leads to lower Na+ and, as a result, lower NO3-. As indicated above in Sect. 5, a decrease in super-micron NO3- results in an increase in sub-micron NO3-. It is probable that sub-micron Na+ increases due to the formation of NaNO3 in the model. In the following runs, higher organic fractions are used instead of those from F10.

In the 100 km HEM_NEW run, a lower wind speed dependence based on satellite data is used, since it improves modelled SSAs compared to observations at many sites over the Arctic, as discussed in Sect. 5. However, RUS10 found evidence for a higher wind speed dependence in the Arctic based on data collected during the Arctic leg of the ICEALOT cruise. They found that wind speed is a good predictor of a marine factor, calculated using positive matrix factorisation, for sub-micron organic matter (OM1sea). Their analysis showed a high correlation between OM1sea, sub-micron sodium (Na+1), and wind speed at 28 m (correlation r equal to 0.90 for the North Atlantic and the Arctic region; see Table S3 in the supplementary material in RUS10). Average OM1sea concentrations (0.2 µgm-3) reported by RUS10 for the eastern Arctic Ocean are about half those reported at Utqiaġvik by, for example, during wintertime.

In a sensitivity run, results from RUS10 are used to include a higher wind speed dependence for sub-micron SSAs. This linear dependence differs from the power dependencies included in G97, SAL14, and other studies but is based on empirical relationships determined from the analysis of data collected in the Arctic. Equations (5) and (6) from the RUS10 analysis for the Arctic legs of their cruise are applied to the model as a correction factor as follows: 4Na+1=0.022×U18-0.012,5OM1sea=0.025×U18-0.049, where U18 is wind speed at 18 m in ms-1 for wind speeds between 2 and 14 ms-1 (Fig. 2, RUS10). RUS10 used Na+1 as a proxy for sub-micron NaCl, and subsequently for SSAs, because Na+1 equalled sub-micron Cl-1 on a molar basis for the North Atlantic and Arctic sampling regions. Thus, Eq. (5) is also used to estimate a correction factor for Cl-. Here, wind speeds in the first model layer are used, i.e. around 26 m. Differences in U18m and U26m reach a maximum of 1 ms-1 (see Fig. E1 in Appendix E). Comparison with radiosonde data at Utqiaġvik shows that the model performs well in terms of winds and temperatures (see Appendix E), and the role of meteorology in relation to aerosols is not discussed further here. The correction factors are only applied to the simulated number and mass of the SSA emissions when modelled wind speeds are between 2 and 14 ms-1 and when RUS10-calculated sub-micron SSA emissions are greater than model-calculated SSA. In this way, SSA emissions are enhanced during periods with higher wind speeds.

To illustrate the sensitivity of the results to applying this correction, Fig. 8 shows the differences in sub-micron aerosol mass concentrations compared to the run including local marine organics and model SSA emission fluxes, the latter being the sum of dry mass emissions calculated in the model. The SSA emission flux is affected over ice-free model grids, leading to increased SSA production east and west of Utqiaġvik (by up to 0.015 µgm-2s-1), while the highest increases are southwest of Alaska (by up to 0.035 µgm-2). This results in an increase of 0.25, 0.19, and 0.11 µgm-3 in sub-micron Na+, NO3-, and OAs, respectively, over the Utqiaġvik region and over southwest Alaska during the February campaign. These results further illustrate the sensitivity of SSA emissions to wind speeds, which, in this case, affect fine-mode aerosols. These results are in contrast to previous studies that found stronger wind speed dependencies for larger SSA particles, such as , who analysed aircraft data, including over the Arctic. However, size-dependent source functions need to be developed for the Arctic region.

The sensitivity of modelled SSAs to prescribed sea-ice fractions during wintertime and the role of leads are also investigated, since KRP19 already pointed out the importance of using realistic sea-ice distributions to simulate marine aerosols. High-spatial-resolution images of sea-ice cover are available, including during the polar night, from a radar operating on top of a building in Utqiaġvik town (71∘17′13′′ N, 156∘47′17′′ W) 22.5 m above sea level, with a range of up to 11 km to the northwest . previously showed increased super-micron Na+ mass concentrations during periods of elevated wind speeds and lead presence in a multi-year study using the sea-ice radar data at Utqiaġvik. Between 23–28 January 2014, when the winds at Utqiaġvik were easterly, the radar showed that the coastal area east of Utqiaġvik featured open leads (KRP19). From 24–28 February 2014, the west coastal area also featured leads, as also shown by Moderate Resolution Imaging Spectroradiometer (MODIS) satellite images (KRP19). To examine the sensitivity of modelled SSA emissions to sea-ice cover, ERA5 sea-ice fractions with a resolution of 0.25∘×0.25∘ are used instead of the FNL fraction at 1.0∘×1.0∘ resolution. Note that only the sea-ice fraction field is different, while the rest of the meteorological fields are from FNL.

Results for February are shown in Fig. 9. The SSA emission flux (Fig. 9a) increases over a small region west of Utqiaġvik and across the North Slope of Alaska due to decreased sea-ice fractions, but it decreases just to the east of Utqiaġvik and southwest of Alaska due to increased sea-ice fractions. Sub-micron Na+ increases slightly along the north coast of Alaska and around Utqiaġvik by up to 0.1 µgm-3 (see Fig. 9b). Larger super-micron Na+ is simulated by up to 0.4 µgm-3 around Utqiaġvik and decreases by up to 0.4 µgm-3 southwest of Alaska (Fig. 9c). SSA emission fluxes increased more during February (0.035 µgm-2s-1) compared to during January (0.015 µgm-2s-1), since there is more sea ice in the region east of Utqiaġvik and south west of Alaska in the January simulation (not shown here). Two further simulations are performed to explore model sensitivity to sea-ice fractions. First, ERA5 sea-ice fractions are set to be equal to zero to the north, west, and east of Utqiaġvik to examine the effect of having ice-free conditions and the presence of open leads locally (as seen by the radar). Second, ERA5 sea-ice fractions are set to be equal to 0.75 north, west, and east of Utqiaġvik and northwest of Alaska. In both cases, the model is run on a windy day (28 February 2014). The first sensitivity test leads to an increase in SSA emission fluxes of up to 0.2 µgm-2s-1 when sea-ice fractions equal zero and to an increase of up to 1.2 and 0.05 µgm-3 in super-micron and sub-micron Na+, respectively. The second sensitivity test yields similar results. This is because ERA5 sea-ice fractions are higher (more sea ice) than those of the test case (0.75), leading to increases in the SSA emission fluxes, especially east of Utqiaġvik. Again, super-micron SSAs (increases of up to 1.5 µgm-3) are affected more than sub-micron SSAs. These results illustrate the regional sensitivity of super-micron SSAs rather than of sub-micron SSAs to prescribed sea-ice fractions. Missing mechanisms influencing sub-micron SSA emissions may need to be included in the model, such as SSA production of, in particular, ultrafine particles from breaking waves in the surf zone . However, information about wave-breaking activity in the surf zone during winter along the northern Alaskan coast is needed to address this.

Results from runs at 20 km with and without the main changes included in the sensitivity tests (local source of marine organics, higher wind speed dependence, and ERA-5 sea-ice fractions) are compared to sub-micron aerosol observations at Utqiaġvik for both the January and February 2014 periods (see Fig. 10). We note that there are no super-micron observations available due to the weekly sampling frequency. It is interesting to compare to these periods, since the observations show different behaviours. While observed sub-micron Na+ and Cl- concentrations reached up to 2.5 µgm-3 in February, they did not exceed 1 µgm-3 in January. As noted earlier, such high concentrations were not observed at Alert and Villum during January and February 2014. The January run, including all the updates (ALASKA_NEW_JAN), better captures sub-micron Na+ and Cl- (reduced biases and RMSEs – see Table F1 in Appendix F), although it underestimates observations by up to 0.6 and 0.8 µgm-3 (Fig. 10a), respectively, while sub-micron NO3- is slightly overestimated. Biases and RMSEs (see Table F2 in Appendix F) are also slightly improved for February, although sub-micron Na+ and Cl- are still underestimated by up to 1.5 µgm-3. Slightly more NO3- is simulated in January, even if the model still underestimates the elevated observations. Up to 6 times more OA is simulated during both periods, which is in better agreement with reported observations , as discussed in Sect. 5.2. The sensitivity tests discussed in this section do not directly address the model underestimation of elevated episodes of sub-micron total SO42-, since it is mostly nss SO42-, and thus the changes are small (during both simulation periods). We note that these runs at 20 km are for periods including elevated aerosols and are thus more challenging for the model to reproduce. Runs at a higher resolution may be needed to better resolve, for example, sea-ice distributions.

Observations of particle number concentration are also used to validate the regional model results at Utqiaġvik (see Fig. 11). High number concentrations are observed during both periods, up to 103 particlespercm3, especially for particle sizes less than 20 nm. reported similar wintertime magnitudes in the accumulation mode (diameter range 100–200 nm) at Utqiaġvik, averaging between 1×102 and 2×102 particlespercm3, whereas smaller number concentrations were reported for particles less than 50 nm compared to that shown here. These differences can be explained by the fact that the SMPS data used here do not exclude local and/or regional pollution based on wind speed criteria, unlike . Local or regional pollution influenced by Utqiaġvik town or by NSA oil fields could lead to new particle growth events (small particles) in the absence of sunlight, as discussed by . also noted that particle number concentrations are higher at Utqiaġvik and Tiksi (sites in proximity to local sources) compared to at other Arctic sites (Alert, Villum, and Zeppelin). The model tends to underestimate observed number concentrations, especially in the fourth (312.5 to 625.0 nm) and fifth (625.0 to 1250 nm) MOSAIC bins, even if the model compares better in January when measured number concentrations are lower. This is consistent with the evaluation of sub-micron SSAs and other aerosol components, particularly for episodes when observed aerosols were enhanced (Fig. 10). Note that the model results cannot be compared to measurements smaller than 39 nm because MOSAIC does not represent these aerosols explicitly and because nucleation is parameterised. Inclusion of a source function to account for ultrafine SSA emissions, for example, from breaking waves at the surf zone, could also lead to improved model results .

Overall, the results presented in this section so far show that modelled sub-micron SSAs (Na+, Cl-) and, as a consequence, NO3- are more sensitive to using a higher wind speed dependence than sea-ice fractions over northern Alaska based on estimated biases and RMSEs for each test simulation (not shown here). Sea-ice fractions have a greater effect on super-micron SSA mass concentrations. Modelled sub-micron OAs are more sensitive to a higher wind speed dependence and, to a lesser extent, the introduction of an additional source of local marine organics. However, the latter is highly uncertain.

Lastly, we consider whether enhanced SSAs, particularly in the sub-micron size range, at Utqiaġvik could be due to blowing-snow or frost flower sources. As noted earlier, KRP19 found no evidence of blowing snow or frost flowers at this site but found that SSAs originated from open leads during wintertime. Their findings are supported by the earlier laboratory study of , who reported that frost flowers are not an efficient source of SSAs. However, found that, during winter at Utqiaġvik, surface frost flowers forming on sea and lake ice are a source of marine-derived OM. Modelling studies including a source of blowing snow and frost flowers suggest that they are contributors to SSAs at this time of year at Utqiaġvik, Alert, and Zeppelin .

To investigate whether blowing snow or frost flowers could also be a source of SSAs during the campaign at Utqiaġvik, depletion factors are estimated following , referred to as FR20 from now on. FR20 reported that blowing snow was the main source of SSAs rather than frost flowers and open leads in Antarctic wintertime based on SO42- and bromide (Br-) depletion in SSAs being indicative of a blowing-snow origin and not of a seawater origin. Other studies also suggested that blowing snow and frost flowers near Utqiaġvik are more characterised by SO42- depletion compared to seawater .

Here, depletion factors are calculated using modelled and observed sub-micron aerosol mass concentrations during the campaign periods. More specifically, total SO42- depletion relative to Na+ (DFSO42-), Na+ depletion relative to Cl- (DFNa+), and Br- depletion relative to Na+ (DFBr- only for observations in this case) are calculated using the following equation: 6DFx=1-RRRSW, where R is the mass ratio (x:y) of species x and y in the model or in the sample and in reference seawater (RSW) . A depletion factor (DFx) between zero (small) and 1 (strong) indicates 0 %–100 % depletion, whereas DFx less than zero indicates enrichment. FR20 suggested, based on the depletion of SO42- relative to Na+, that most SSA originates from blowing snow on sea ice, with minor contributions from frost flowers, and not from open leads.

The average values of modelled and observed DFs are shown in Table 5. Total SO42- is enriched relative to Na+ in both the observations and the model results during both campaign periods; this is in contrast to FR20, who reported substantial depletion. In February, observed and modelled results both indicate SO42- enrichment relative to seawater, whereas in January, model results are less enriched compared to the observations, possibly due to underestimation of nss SO42-. In January, observed total SO42- concentrations are 7.56 times higher than in reference seawater, possibly due to internal mixing with anthropogenic nss SO42-, as noted by KRP18. Modelled total SO42- is less enriched than the observations (0.94 times higher than in reference seawater), likely due the model underestimation of nss SO42-. FR20 did report a case of enrichment due to possible contamination from the ship, an anthropogenic source. The Na+ depletion factor also shows enrichment during both campaigns, albeit more negligible in the observations than in the model. Observed Na+ depletion relative to Cl- is 1.09 or 1.19 times more than in reference seawater during January and February, respectively. Our analysis suggests that blowing snow and frost flowers are not a significant source of SSAs, at least during this campaign in winter 2014.

SSAs can also play an important role in polar tropospheric O3 and halogen chemistry through the release of active bromine during spring . Reactions involving bromine are an important sink of O3 (e.g. ). Br- depletion relative to Na+ is calculated only during February, since observed Br- was zero during the January campaign period. The results for February show a small enrichment, indicating a seawater origin. The observed mass ratio of Br- to Na+ ranges between 0.0057 and 0.0059, while the mass ratio of Br- to Na+ in reference seawater is equal to 0.006. FR20 reported no or little Br- depletion relative to Na+ due to Br- losses at the surface and small depletion further aloft (in Antarctica). For a more comprehensive analysis, observations are required at different locations and altitudes across northern Alaska. We note that the version of WRF-Chem used in this study does not include halogen chemistry. It has since been implemented in a later version by to examine springtime O3 depletion events at Utqiaġvik. Heterogeneous reactions on SSAs from the sublimation of lofted blowing snow were also included. Their results suggested that blowing snow could be a source of SSAs during spring, although it should be noted that this model version (including blowing snow as a source of SSAs) overestimated SSAs (da<10 µm) at Arctic sites such as Alert and Villum during spring, while wintertime conditions were not examined.

Finally, modelled and observed molar ratios of sub-micron Cl-:Na+ and SO42-:Na+ are estimated to further examine the origins of SSAs and to compare our findings with KPR19 (see Table 6). The averaged molar ratios of sub-micron Cl-:Na+ and SO42-:Na+ derived here for the campaign periods (Table 6) agree with KRP19 (Cl-:Na- equal to 1.08; see Table S3 in the Supplement of KRP19, as well as the text and references within). They indicate a seawater origin and confirm the findings of KRP19 that there was no evidence for blowing snow and frost flowers as a source of SSAs during the Utqiaġvik campaign, which is also in agreement with previous studies . Model averaged molar ratios are smaller in magnitude than the observations. These discrepancies could be due to the fact that the model underestimates sub-micron SSAs and SO42- for the reasons noted earlier. Differences could also be due to issues with modelled SSA lifetime and chemical processing during long-range transport. used Cl-:Na+ molar ratio enrichment factors as an indicator of the influence of long-range transport on SSAs at Utqiaġvik. They reported that Cl- depletion was larger for aged sub-micron than for aged super-micron SSAs due to a longer lifetime, which is in line with other studies . The regional results (ALASKA_NEW_JAN and ALASKA_NEW_FEB) indicate that modelled sub-micron Cl- has undergone significant atmospheric processing. Thus, the regional model results, influenced by the simulation at 100 km, overestimate aged sub-micron SSAs, while the contribution from locally produced sub-micron SSAs may be too low in the model (modelled enrichment factors equal to 0.5 and 0.4 during January and February simulation periods, respectively, which is lower than the threshold (0.75) defined by ). On the other hand, modelled enrichment factors for super-micron are equal to 0.6 and 0.85 during January and February, respectively, indicating that there is a possible background influence on super-micron SSAs during January, while they are locally produced during February. KRP18 reported the presence of both fresh (locally produced) SSAs and aged (partially Cl--depleted) SSAs for sub-micron SSAs, while super-micron were mostly fresh (KRP18, Fig. 2). Based on the analysis above (including observations), there is little evidence suggesting that blowing snow or frost flowers are a significant source of SSAs at Utqiaġvik during winter, and open leads are an important primary source, in agreement with KRP19.