Mineral dust aerosol is the main ice nucleus (IN) in the Arctic. Observed dust concentrations at Alert, Canada, are lowest in winter and summer and highest in spring and autumn. In this study, we simulate transport and deposition of dust in a global chemical transport model. The model predicts the spring maximum caused by natural dust from desert sources in Asia and Sahara but underestimates the observations in autumn. Both natural and pollution sources contribute to the wintertime dust burden, as suggested by previous measurements of elemental compositions. Cloud parcel model simulations were carried out to study the impact of dust aerosol on the formation of mixed-phase and ice clouds in the Arctic lower troposphere. The liquid water path of low-level cloud is most sensitive to dust aerosol concentration from winter to early spring when air temperature is at its lowest in the annual cycle. The global and parcel models together suggest that low concentrations and acid coating of dust particles are favorable conditions for occurrence of mixed-phase clouds and that anthropogenic pollution can cause significant perturbations to Arctic IN and clouds in winter.
 Mineral dust and metal oxides contribute most to the particles found in ice crystal residues, followed by carbonaceous components, in the Arctic atmosphere [Prenni et al., 2009]. Mineral dust was also the main ice nucleus measured in cirrus clouds in the midlatitude Northern Hemisphere [Pratt et al., 2009; Kamphus et al., 2010; Ebert et al., 2011]. This is attributed to the abundance of dust aerosol in the atmosphere and its superior ice nucleating ability [Hoose et al., 2010]. In an analysis of global supercooled clouds and coincident dust aerosol data obtained from the same instrument on a satellite, it was found that the variation of supercooled cloud fraction is negatively correlated with the frequencies of dust aerosols at the −20°C isotherm [Choi et al., 2010]. This correlation suggests a possibility that dust particles are mainly responsible for the glaciation of supercooled clouds on a global scale [Choi et al., 2010], although other types of ice nuclei are also present in the atmosphere [Hoose et al., 2010].
 Mineral dust particles are variable in mineralogical composition [Claquin et al., 1999] and thus have variable ice nucleation properties [Zimmermann et al., 2008]. Mineral dust particles are also variably coated by soluble materials, including inorganic acid or salts and organic compounds [Falkovich et al., 2001; Sullivan et al., 2007; Shi et al., 2008; Kamphus et al., 2010; Ginoux et al., 2012a]. Changes in the ice nucleation properties of mineral dust particles due to soluble coatings are still not well understood [Tobo et al., 2012]. Nonreactive coatings were found to reduce ice nucleating ability below water saturation, while reactive coatings were found to impact ice nucleation both below and above water saturation [Tobo et al., 2012]. Being most relevant in the atmosphere, the ice nucleating ability of dust particles may be reduced by nitric and sulfuric acids, ammonium nitrate and sulfate [Möhler et al., 2008; Cziczo et al., 2009; Eastwood et al., 2009; Chernoff and Bertram, 2010; Sullivan et al., 2010; Niedermeier et al., 2011; Reitz et al., 2011].
 Field measurements show that Arctic haze aerosol had lower ice nucleus (IN)-to-particle ratios and slower ice nucleation rates than aerosol in unpolluted troposphere, which is hypothesized to be caused by sulfuric acid coating on existing IN in Arctic haze [Borys, 1989]. Several large-scale modeling studies have been conducted on potential effects of acid-coated IN in clouds; these include a reduced ice number promoting further ice crystal growth and precipitation [Girard and Stefanof, 2007], a less frequent glaciation of mixed-phase clouds leading to longer cloud lifetimes and higher cloud albedos [Hoose et al., 2008; Storelvmo et al., 2008; Du et al., 2011], and an increased occurrence of Arctic mixed-phase clouds leading to more atmospheric cooling in winter [Girard et al., 2013]. Arctic clouds and their radiative forcing in climate models are sensitive to parameterizations of IN, especially in winter [Du et al., 2011; Xie et al., 2013].
 Dust entrainment in deserts and semiarid regions is by far the most important source of mineral dust in the global atmosphere, followed by emissions from dry agricultural lands (wind-driven erosion and soil particle suspension during harvesting and tilling) [Ginoux et al., 2010; Ginoux et al., 2012b]. In the populated regions, anthropogenic activities may cause significant local air pollution by traffic (exhaust, road dust, tire wear, and break abrasion), power generation (coal fly ash), smelter and industry (metal oxides), and construction. Continuous emissions from these pollution sources have been found to be more important than natural sources in Europe, East Asia, and North America except on days when transport of desert dust is significant [e.g., Pakkanen et al., 2001; Schwab et al., 2004; Sun et al., 2004; Heal et al., 2005; Lough et al., 2005; Almeida et al., 2006; Zhao et al., 2006; Canepari et al., 2008; Viana et al., 2008; Amato et al., 2009; Karanasiou et al., 2009; Limbeck et al., 2009; Reff et al., 2009; Bukowiecki et al., 2010; Moreno et al., 2011].
 Transport of pollution to the Arctic has been observed and analyzed extensively in the last four decades [e.g., Rahn et al., 1977; Rahn, 1981; Schnell, 1984; Barrie, 1986; Schnell et al., 1989; Heidam et al., 1999; Sharma et al., 2004; Quinn et al., 2007; Eleftheriadis et al., 2009; Jacob et al., 2010]. In particular, long-term aerosol measurements have been made at four Arctic stations: Alert, Ellesmere Island, Canada (82.39°N, 62.3°W); Nord, Northeast Greenland (81.6°N, 16.67°W); Point Barrow, Alaska (71°N, 156.6°W); and Zeppelin, Spitsbergen, Norway (78.9°N, 11.9°E). These observations reveal frequent occurrences of Arctic haze in winter and early spring. Transport into the Arctic of air pollution from northern Eurasia occurs primarily in the lower troposphere, while that from lower latitudes occurs at higher altitudes [e.g., Klonecki et al., 2003; Sharma et al., 2006; Stohl, 2006; Shindell et al., 2008; Huang et al., 2010]. The occurrences of long-lived (~10 days) “blocking” anticyclone over Europe are most frequent during winter/spring [Lejenäs and Økland, 1983]. Blocking prevents the normal eastward propagation of cyclones, and the northernmost branch of the air current will then carry air toward the pole over large meridional distances, creating meridional transport “channels” from western Eurasia into the Arctic [Iversen and Joranger, 1985]. Injection of air pollutants into the Arctic through these channels is followed by rapid transport across the Arctic, for instance, from the Barents Sea into the Chukchi/Beaufort Sea area [Raatz and Shaw, 1984; Raatz, 1985].
 Pollutants from East Asia and North America are transported into the Arctic at high altitudes following surfaces of constant potential temperature which ascend northward [e.g., Klonecki et al., 2003]. However, clouds often form as air parcels ascend and soluble species and hygroscopic aerosols may be scavenged by precipitation. Transport efficiency (i.e., fraction remaining from source to receptor) of aerosols from East Asia and North America is much lower than from northern Eurasia [Matsui et al., 2011].
 More recent observational and modeling studies suggest that a small in-cloud scavenging of aerosols by snow is a significant factor for the long-range transport of pollution in winter [Koch et al., 2009; Vignati et al., 2010; Bourgeois and Bey, 2011; Garrett et al., 2011; Liu et al., 2011; Fan et al., 2012; Browse et al., 2012; Zhou et al., 2012]. This is attributed to two causes: (1) homogeneous ice nucleation removes only small fractions of deliquesced aerosols and liquid cloud droplets, and (2) growth of ice crystals can lower relative humidity and cause evaporation of droplets in mixed-phase clouds and release of aerosols. Radiative cooling causes strong stratification in the surface layer in winter, which dampens dry deposition of particulates over smooth surfaces of snow and sea ice [Liu et al., 2011]. As a result of seasonal variations in wet and dry deposition during the long-range transport, tracers of air pollution, such as sulfate and lead, show high concentrations from December to April and low concentrations from June to October [Barrie and Barrie, 1990; Heidam et al., 2004; Nguyen et al., 2013].
 In contrast, tracers of mineral dust, such as Al, Ca, Fe, and Si, show high concentrations in spring (April and May) and in autumn (September and October) [Barrie and Barrie, 1990; Heidam et al., 2004]. Dust storms in the Gobi and Taklimakan deserts are most frequent in spring [Luo et al., 2003]. The spring peak in dust aerosol is most likely associated with these sources [Winchester et al., 1984; Barrie and Barrie, 1990; Li and Winchester, 1990; VanCuren et al., 2012]. The cause of the autumn peak in dust aerosol is not clear; it may be due to local sources [Barrie and Barrie, 1990].
 Dust aerosols observed in the Arctic from December to March have been attributed to both pollution (dust tracers mixed with anthropogenic pollution tracers) and natural (not mixed with pollution) sources, with about 50% contribution from each [Maenhaut et al., 1989; Li and Winchester, 1990; Nguyen et al., 2013]. The source apportionment of particulate matter was based on statistical analysis of chemical/elemental compositions of aerosols measured in the Arctic. The pollution sources of dust aerosol become relatively important in winter as natural sources become small. Dust entrainment in the deserts of central Asia and East Asia is at its lowest during winter in the annual cycle [Luo et al., 2003]. Transport of Saharan dust aerosol to the high-latitude Northern Hemisphere also occurs at the lowest frequency in winter [Meloni et al., 2007], contributing on the average less than 10% of the particulate matter (PM10) mass in Europe north of 45°N [Pey et al., 2013].
 It is challenging to interpret these results using observations alone. In this study, we simulate mineral dust aerosols in the Geophysical Fluid Dynamics Laboratory global chemical transport model (GFDL GCTM). The goal of this study is to estimate the basin-wide apportionment of Arctic dust aerosol between pollution and natural sources, including spatial and temporal variations, and the impact of dust particles on wintertime low-level clouds which are important for atmospheric transmission of longwave radiation and growth of sea ice. Mineral dust from volcanic eruptions and farmlands is not considered. We also conduct a series of cloud parcel model simulations to study the impact of dust aerosol and acid coating on the liquid water path in boundary layer clouds. The models and simulations are described in section 2. Transport model results are presented in section 3. Cloud parcel model results are presented in section 4. Further discussion of the model results and implications is presented in section 5. A summary of results and conclusions is presented in section 6.
2 Model Descriptions
2.1 The Global Chemical Transport Model
 The original GFDL GCTM [Mahlman and Moxim, 1978] was adapted to use National Centers for Environmental Prediction (NCEP) reanalysis winds [Kalnay et al., 1996], archived every 6 h, on the same 28 vertical (sigma) levels [Fan et al., 2004]. The u and v components of the horizontal NCEP winds were linearly interpolated to the GCTM equal area 265 km horizontal grid. The vertical velocities are calculated online from the horizontal mass divergence and the surface pressure tendency. The model has been used previously for simulations of global dust entrainment and transport and for simulations of soluble iron deposition [Fan et al., 2004; Moxim et al., 2011].
 Subgrid-scale vertical mixing is parameterized by a diffusion coefficient (Kz) [Levy et al., 1982]. In the surface mixed layer, Kz is calculated from the surface momentum flux and vertical wind shear, both of which are derived from the NCEP reanalysis. The boundary layer height is diagnosed from the bulk Richardson number. Vertical transport by deep convection is parameterized as lifting an ensemble of air masses from the boundary layer to the top of convection as determined from the moist Richardson number [Levy et al., 1982] and a layer-by-layer downward redistribution of air to represent subsidence and maintain mass balance. The amount of air lifted (per horizontal grid per 6 h) is calculated from the difference of saturation humidity between the bottom and top layers in the convective column and the grid-scale average precipitation. Grid-scale advection is calculated using a finite difference scheme [Mahlman and Moxim, 1978]. The advection time step is 26 min, and the vertical diffusion time step is 2.6 min.
 Natural dust entrainment flux (F; in kg m−2 s−1) is predicted in the model from surface friction velocity (u*; in m s−1) derived from the surface momentum flux as follows [Marticorena and Bergametti, 1995]:
where C is a scaling factor (equal to 0.003 kg m−5 s2), S(x,y) is a function of longitude (x) and latitude (y) that specifies the spatial distribution of dust sources taken from Ginoux et al. , and u*t is a threshold of u* below which there is no entrainment. The S(x,y) function has implicitly incorporated factors such as vegetation cover, soil texture, and erosion surface area. The magnitude of u*t depends on soil moisture [Fécan et al., 1999] and surface roughness [Laurent et al., 2008]. However, we choose to use u*t = 0.35 m s−1 everywhere because we do not have accurate spatial data on surface soil moisture. Our dust emission is suppressed in the 24 h after precipitation and when there is snow on the surface. The value of u*t is within the range of measurements (0.2–0.5 m s−1) in desert areas [Marticorena and Bergametti, 1995; Shao and Lu, 2000] but smaller than estimated (0.5–2 m s−1) in semiarid areas [Shao and Leslie, 1997]. The predicted frequency and magnitude of dust storms are in agreement with observations downwind of East Asia (unpublished results) and Sahara [Moxim et al., 2011].
 Pollution dust (insoluble road dust, metal oxides, fly ash, etc.) emissions are assumed in this study to be proportional to that of black carbon due to fossil fuel combustion (see section 4). Trace metals are found to be positively correlated with elemental carbon concentrations at rural sites as well as in urban areas in the Interagency Monitoring of Protected Visual Environments network [Malm et al., 1994] (unpublished results). The emissions inventory for black carbon was estimated by Bond et al.  and was used previously in a global aerosol models intercomparison study [Koch et al., 2009]. It should be noted that dust/black carbon (BC) ratios in western Europe are considerably smaller than those in more arid regions of eastern Europe and southern Europe. A more realistic emissions inventory for pollution dust is desirable for future studies.
 Dry deposition of dust aerosols is parameterized according to Giorgi  and is proportional to dust concentration and a deposition velocity. The dry deposition velocity is expressed in terms of u* and particle size and density. In-cloud wet deposition for hygroscopic aerosols is described by Kasibhatla et al.  and is proportional to dust concentration and precipitating cloud volume fraction during each time step. The volume fraction is calculated from grid-scale precipitation, cloud thickness, and cloud water content. Because the GCTM is an off-line model and cloud thickness and water content are not available from the archived meteorology, the cloud top is diagnosed from gradient Richardson number [Levy et al., 1982], and the cloud water content is specified to be 0.5 g m−3 for stable rain and 2.0 g m−3 for convective rain [Kasibhatla et al., 1991]. The removal rate is scaled by a factor = 50/(50 + D), where D is the dust mixing ratio in µg kg−1, to account for reduced activated fraction at high dust concentrations. The reduction of activated fraction is necessary in the model to obtain dust concentrations comparable to long-term observations in the North Atlantic (see supporting information) and is attributed to lack of soluble coating and competition at high concentrations of cloud condensation nuclei [Fan et al., 2005]. Below-cloud scavenging is parameterized based on Zender et al.  and is dependent on precipitation and particle size.
 Wet removal of pollution dust by snow is reduced to 10% of that for warm rain at temperature (T) < −15°C in order to increase aerosol transport efficiency during the winter season [Liu et al., 2011]. A larger fraction (20%) is removed at −15°C < T < −5°C to account for increased riming fraction (accretion of snow and droplets). The reduced wet removal may be attributed to the availability of ice nuclei more active than mineral dust particles in pollution plumes, which nucleate ice at temperatures warmer than −15°C such as biological ice nuclei [Szyrmer and Zawadzki, 1997; Christner et al., 2008; Garcia et al., 2012]. Furthermore, only a small fraction of coated dust particles nucleate ice during the transit from cloud base to cloud top based on cloud parcel model calculations (section 4). Pollution dust particles are more extensively coated by sulfate and other soluble materials than natural dust particles. The parameterization for wet removal of soluble aerosols is evaluated with long-term measurements of 210Pb at Point Barrow, Alaska, and aircraft measurements of 7Be over the Arctic in March–April and June–August 2008 (unpublished results).
 Natural dust particles are emitted and transported in four separate size bins (r = 0.1–1, 1–1.8, 1.8–3, and 3–6 µm), with a size distribution (mass fraction = 0.12, 0.26, 0.32, and 0.30, respectively) based on measurements in the Sahara on days with no active dust storms [D'Almeida, 1987]. Pollution dust is emitted and transported in a single size bin (mass median radius = 0.55 µm), and the model results are scaled to match observations of black carbon and trace metals in the Arctic (section 3). The scaling factor corrects for biases in emissions and wet deposition during the long-range transport from northern Eurasia to the Arctic stations.
2.2 The Parcel Model
 The adiabatic parcel model calculates changes in thermodynamic properties as a parcel of air rises, expands, and cools, and simulates the transformation of water among vapor, liquid, and ice phases. Input parameters include initial pressure, temperature, relative humidity, dry aerosol mass concentrations, dry aerosol size distributions, and updraft velocity. Four types of aerosol are considered in the model: sulfate, sea salt, dust, and black carbon (BC); they are mixed externally. However, soluble sulfate coatings are assumed for insoluble cores of dust and BC aerosols. The aerosol concentrations, size distributions, and soluble fractions are summarized in Table 1. Initial dry mass and number distributions in 20 logarithmically equal size bins between 0.01 and 10 µm in radius are calculated for each aerosol type. Concentration of BC is assumed to be much lower than observed, such that ice nucleation on BC particles is negligible. Presently, there is substantial uncertainty as to whether BC is an IN at temperatures warmer than 240 K [Friedman et al., 2011; Hoose and Möhler, 2012]. Organic aerosol is not considered in the present model. Biological ice nuclei, such as bacteria and leaf debris, may be abundant in polluted air mass but is not considered due to lack of observations. The model predicts water and ice supersaturation, droplet and ice crystal number concentrations, liquid and ice water content, and pressure and temperature.
Table 1. Prescribed Aerosol Parameters in the Parcel Modela
ad = mass median diameter, σ = geometrical standard deviation, ϵ = soluble mass fraction, ρ = density, and C = initial concentration in dry mass.
bFine mode contributes to 15% of the total dust mass, which is specified to be 0.02, 0.05, 0.1, 0.2, 0.5, 1.0, 2.0, and 5.0 µg m−3 in separate simulations.
 Ice nucleation rates are calculated based on Koop et al.  for homogeneous nucleation and on the classical nucleation theory for heterogeneous nucleation. For dust particles, the activation energy is −6.21 × 10-21 J for deposition freezing and 1.57 × 10-19 J for immersion freezing, and the contact angle is 12.7° for deposition and 30.98° for immersion [Hoose et al., 2010]. Sulfate coating is assumed to increase the contact angle to 27° for deposition nucleation on dust particles [Du et al., 2011]. Contact freezing is limited by collision between unnucleated particles with ice germs and liquid droplets, estimated to be small compared to immersion, and thus neglected in the model. The effect of mineralogy on the onset ice supersaturation for deposition nucleation is neglected. The ice crystals are assumed to be spherical in shape in standard simulations. The assumption of spherical ice particles instead of nonspheric habits leads to an underestimation of ice growth [Ervens et al., 2011; Sulia and Harrington, 2011]. In sensitivity simulations, ice growth is calculated with habit evolution based on the two-axis oblate or prolate spheroid method [Sulia and Harrington, 2011]. In addition to condensational growth, large ice crystals may collide with water droplets, growing quickly in mass, and precipitate out of the cloud layer. Presently, this process is not treated in the parcel model. Therefore, predicted ice water content and crystal number concentration are not comparable to observations. The sedimentation of large ice crystals leads to higher supersaturation in the parcel and slower glaciation of supercooled clouds [Ervens et al., 2011]. In sensitivity runs, these effects are simulated by turning off growth for large crystals (e.g., radius > 50 µm) as if they were removed from the air parcel.
 Ten sets of parcel model simulations are conducted (Table 2). Each set of simulations covers a range of initial temperature and a range of dust concentration. Dust particles are either assumed to be coated with sulfate or uncoated in terms of ice nucleation parameters (activation energy and contact angle), the latter nucleate ice in the deposition mode at an ice saturation >112%. Immersion freezing occurs for coated dust for the range of simulated temperatures. The updraft velocity is specified to be 10 cm s−1 for a period of 100 min or 20 cm s−1 for a period of 50 min. The initial relative humidity is specified at 80%, 85%, and 90% for different updraft velocities and freezing modes (Table 2), in order to homogenize runup time to cloud base (initiation of ice or droplet nucleation) and to obtain similar cloud layer thickness. The rate of glaciation is the main cause of differences in liquid water content in the 600 m column among these simulations. The simulations are designed for the Arctic lower troposphere, where cloud top cooling causes a negative buoyancy and drives a vertical cell of air motion (downdrafts of air mass balanced by updrafts) [Solomon et al., 2011; Morrison et al., 2012]. The vertical velocities measured inside Arctic mixed-phase clouds and predicted in large eddy simulations of these clouds have approximately a Gaussian distribution, with a mean near zero and a standard deviation of 0.3 m s−1 [Fan et al., 2011]. The simulations in this study do not consider clouds in downdrafts or in horizontal advection.
Table 2. Input Parameters Used in the Parcel Model Simulations
aIce crystals are assumed to settle out (stop growing) above the critical radius.
 Seven-day aerosol samples have been collected at Alert, Canada, since 1980 [Sirois and Barrie, 1999] and analyzed for trace metals using instrumental neutron activation analysis [Barrie et al., 1989]. Long-term mean Al concentrations for each month of the year are converted to mineral dust concentrations assuming a mass ratio of 7.1% [Guieu et al., 2002]. The annual cycle of dust aerosol observed at Alert is compared with the model result for natural emissions in Figure 1. Mean and standard deviations are calculated based on weekly measurements from 2000 to 2006 and for weekly GCTM model outputs from the same period. The observed dust at Alert is highest from September to November, followed by April and May, and lowest in winter and summer (Figure 1). In comparison, the model dust is high from March to May and is low from November to February. The model results show general agreement with the measurements from March to August, but significant underestimation from September to February. The model results for natural dust are also in agreement with long-term measurements in the North Atlantic and in the North Pacific (see supporting information). It is noted that dust simulation is sensitive to meteorology. Different simulation results were obtained over East Asia when different reanalysis meteorology data sets were used to drive dust emission, transport, and wet removal [Luo et al., 2003].
 Trace metals are enriched in pollution aerosols [Pacyna and Pacyna, 2001], and the elemental ratio can be used to assess the origin of the dust. The observed mass ratio of Mn/Fe in Alert aerosols is shown in Figure 2. The elemental ratio is close to that measured in Saharan dust (0.017) [Mendez et al., 2010], Asian dust (0.021) [Liu et al., 2002; Zhang et al., 2003], and the upper crust (0.017) [Wedepohl, 1995] from April to November but is higher from December to March. The higher ratio in winter is comparable to that measured in Europe (0.030 ± 0.012 for PM2.5 and 0.016 ± 0.006 for PM2.5–10 [Viksna et al., 2004; Heal et al., 2005; Sillanpää et al., 2005; Limbeck et al., 2009] and East Asia (0.027) [Liu et al., 2002; Kim et al., 2011]. Even higher Mn/Fe ratios (>0.030) have been reported from urban and suburban environments [Puxbaum et al., 2004; Sun et al., 2004; Heal et al., 2005]. Both Fe and Mn are emitted by various anthropogenic activities [Reff et al., 2009], although Mn is more enriched in PM2.5 relative to soil elemental compositions. It appears that dust aerosols observed at Alert originate largely from natural sources from April to November but mainly from anthropogenic sources from December to March. The measured Mn/Fe ratios also suggest that dust aerosols are present mostly in PM2.5 in winter to early spring. When combined with results from Figure 1, Alert Mn/Fe ratios (Figure 2) suggest that the model underestimates dust from September to November because it lacks soil emissions, perhaps from local sources [Barrie and Barrie, 1990], and that wintertime underestimation may be the result of missing sources of pollution.
 Pollution dust is then simulated in the model as a hygroscopic aerosol with a source proportional to black carbon emissions from fossil fuel combustion. Presently, the GCTM under-predicts BC at Alert by a factor of 5 (~20 versus ~100 ng m−3 for multiyear averages) in the winter season, like many other models [Vignati et al., 2010; Bourgeois and Bey, 2011; Liu et al., 2011; Browse et al., 2012; Zhou et al., 2012]. The cause of this is not clear; it is likely a combination of biases in emissions inventory and wet deposition. The model results for the hygroscopic tracer are converted to dust by a multiplication factor = (dust/Al) (Al/BC) (BCobs/BCmodel), where (dust/Al) = 14, (Al/BC) = 0.3, and (BCobs/BCmodel) = 5. The measurements of the Al/BC ratio range from 0.085 to 0.78 in Europe [Sillanpää et al., 2005], being lowest in western Europe and highest in relatively dry regions of northern Eurasia. The monthly concentrations of pollution dust at Alert show a broad peak (~0.09 µg m−3) from December to April and are low (~0.02 µg m−3) from June to October (Figure 3). Therefore, transport of pollution contributes about one third of the observed dust at Alert in the winter season, which is lower than the 50% contribution estimated based on factor analysis of trace metals in the Arctic [Maenhaut et al., 1989; Li and Winchester, 1990]. Recently, source apportionment of particles at Station Nord, Northeast Greenland, based on year-round elemental measurements from 2008 to 2010 also suggests a contribution of about 50% by natural soil [Nguyen et al., 2013]. It is likely that the model underestimated sources from the Former Soviet Union countries where fugitive dust is much higher than western Europe, as indicated by measurements in Finland [Pakkanen et al., 2001; Sillanpää et al., 2005], a gateway of transport to the Arctic.
 Figure 4 illustrates predicted dust distributions in the Arctic for natural and anthropogenic sources, averaged over the lowest 1 km for the months between December and February, 2000–2006. Dust concentrations, both natural and anthropogenic, are higher in the East Arctic than in the West Arctic, with the highest concentrations (~1 µg m−3) in the region of Barents Sea and Kara Sea, and the lowest concentrations (~0.1 µg m−3) in the Beaufort Sea and Canada Basin. Multiyear regional averages of PM10 in Europe range from 24.0 ± 9.2 µg m−3 at rural sites to 30.6 ± 8.4 µg m−3 at urban sites [Wang et al., 2012], with a 17 ± 12% contribution from European crustal or mineral and a 19 ± 11% contribution from European traffic sources [Belis et al., 2013]. Based on these data, dust aerosol is estimated to be on the order of 5 µg m−3 in Europe, comparable to model results (2–5 µg m−3) shown in Figure 4. The concentration of dust in the Arctic could range from 0.01 to 10 µg m−3 in space and time. Natural and anthropogenic dust aerosols could be transported into the Arctic in the same air mass as well as in separate air masses; they may act together at times or separately at other times to influence clouds in the polar region.
4 Parcel Model Results
 To assess the influence of dust on Arctic clouds, we conducted cloud parcel model simulations for “coated” and “uncoated” dust using concentration derived from the transport model. The coated dust particles have lower ice nucleation temperature (greater contact angle) than the uncoated ones in the deposition mode, although they are assumed to have identical size distributions and soluble mass fractions (5%). Growth of large ice crystals (r > 50 µm) is assumed to continue in one set of simulations but to stop in another, in order to study the sensitivity of model results to whether large ice crystals settle out of the cloud. Figures 5-8 show the variation of total ice crystals, small ice crystals (r < 50 µm), water supersaturation (SS), and ice SS with height from the launching level, for a vertical velocity of 0.1 m s−1, a dust concentration of 0.5 µg m−3, and different initial temperatures. Deposition freezing is more likely on uncoated dust particles at an ice SS of ~12%, with subsequent growth of ice crystals keeping the water vapor pressure below water saturation. By contrast, immersion freezing occurs more likely on coated dust particles at higher ice SS (>12%) for the range of simulated temperatures, although deposition freezing on coated particles can occur at colder temperatures.
 When large ice crystals settle out of cloud, growth of total ice mass is slowed on fewer ice crystals remaining in the parcel, resulting in higher ice SS and continued ice nucleation (Figure 5 versus Figure 6 for uncoated dust and Figure 7 versus Figure 8 for coated dust). However, the depositional growth of individual ice crystals increases due to higher ice SS at a reduced competition. The total number of ice crystals produced is about three times higher with sedimentation than without (e.g., 0.6 versus 0.2 cm−3 for uncoated dust, and 0.12 versus 0.04 cm−3 for coated dust). The number of small ice crystals is also higher with sedimentation than without, especially near the cloud top (600 m above the launching level). The ice number produced by deposition freezing on uncoated dust is five times that by immersion freezing on coated dust. For the size distribution specified in the model, a dust concentration of 0.5 µg m−3 can provide ice nuclei (IN) totaling 1.4 cm−3 when available for activation at 100%. It is noted that not all dust particles can serve as ice nuclei [e.g., Field et al., 2006]. Nearly half of the IN is utilized in one cloud cycle (ascending from cloud base to cloud top) for uncoated dust, but less than 10% is utilized for coated dust. Furthermore, the low temperatures required for immersion freezing on coated dust occurs less frequently than the warmer temperatures for deposition freezing on uncoated dust in the lower troposphere. These model results suggest that in-cloud wet removal of coated dust is less efficient than uncoated dust by snow, because the ice nucleating ability of insoluble dust aerosol has been altered by coating. Long-range transport into the Arctic may be more efficient for coated dust with reduced ice nucleating ability than uncoated dust when air temperature is in the range from −10 to −30°C.
 Deposition ice nucleation and crystal growth can prevent the formation of supercooled clouds, while immersion freezing can cause glaciation of supercooled clouds. The rates of ice nucleation and glaciation depend on temperature and available IN concentration. In our parcel model simulations, insoluble mineral dust is the only ice nucleus. For a given temperature, liquid water path (LWP) in the 600 m column depends on the ice nucleating ability as well as the concentration of dust particles. The LWP is calculated assuming a steady state vertical distribution of air parcels ascending in the 600 m column. Figure 9 shows model-calculated LWP as a function of temperature and uncoated dust concentration. At warm temperatures (midheight T > 258 K in Figure 9a and T > 248 K in Figure 9c), ice crystals are too sparse to prevent water SS, and liquid clouds are formed in the column. At colder temperatures, formation and growth of ice crystals often prevent the formation of liquid clouds even at low dust concentrations (<0.1 µg m−3). Without ice sedimentation (Figures 9a, 9b, and 9d), temperature is the main control on LWP. With ice sedimentation (Figure 9c), temperature and dust concentration together control the thickness of liquid cloud, because dust concentration limits new ice nucleation under a higher SS environment. In this case (Figure 9c), significant LWP (>10 g m−2) may be maintained at temperatures as low as 250 K when dust concentration is low, but liquid cloud does not form at T > 255 K when dust concentration is high (>0.1 g m-3). For deposition freezing on uncoated dust particles (Figure 9), vertical air velocity and habit evolution have relatively minor influences on the LWP (compare Figures 9a, 9b, and 9c).
 The glaciation time of a mixed-phase cloud due to the Wegener-Bergeron-Findeisen mechanism is dependent on ice crystal concentration and relative humidity which themselves depend on temperature and vertical velocity [Korolev and Isaac, 2003]. Figure 10 shows variation of LWP with temperature and coated dust concentration. High dust concentrations often cause complete glaciation of supercooled clouds, which would reduce the cloud optical depth by about a factor of 5 assuming an effective radius of 10 µm for liquid droplets and 50 µm for ice crystals, respectively. However, the glaciation is slower and more often incomplete when vertical velocity is larger (Figure 10a versus Figure 10b) or ice sedimentation is allowed (Figure 10a versus Figure 10c), even at temperatures below 243 K (−30°C). Allowing habit evolution increases the rate of glaciation and decreases LWP (Figure 10b versus Figure 10d), but the impact is small at low dust concentrations (<1 µg m−3). Comparing results in Figure 9 and Figure 10, it is evident that the ice nucleating ability of dust particles and its alteration by surface coating have a large influence on the occurrence of mixed-phase clouds. Acid coating of dust particles occurs in polluted air with high concentrations of SO2, NO, and NO2 gases through heterogeneous reactions or uptake of H2SO4 and HNO3 molecules and could extend the temperature range for liquid and mixed-phase clouds from 260 to 243 K. This conclusion is in agreement with previous studies using large-scale models [Girard and Stefanof, 2007; Hoose et al., 2008; Storelvmo et al., 2008; Du et al., 2011; Girard et al., 2013]. In theory, however, the effects of IN and acid coating are more accurately quantified using a cloud parcel model that represents the rapid evolution of temperature and relative humidity, ice number and crystal size distribution, and aerosol and droplet size distributions in the ascending air (e.g., as shown in Figures 5-8). By contrast, certain assumptions have been made in the large-scale models to simplify representation of cloud microphysics over long time steps and large grid sizes.
 Previously, it was hypothesized that transport of IN into the Arctic and precipitation of ice crystals can cause atmospheric dehydration (and further positive greenhouse feedback) [e.g., Girard and Stefanof, 2007]. The parcel model results suggest that transport of uncoated dust into the Arctic can cause more dehydration (and active at warmer temperatures) than acid-coated dust particles can. Higher relative humidity can be maintained when acid coating inhibits deposition freezing on dust particles at warm temperatures or causes immersion freezing and slow glaciation at colder temperatures (see Figures 6 and 8 for simulations with large ice crystals removed from the air parcels). However, coated dust may be more efficiently removed by warm precipitation (T > −10°C) while uncoated dust by settling of ice crystals during transport. Measurements of coated and uncoated dust aerosols are needed to estimate their effects on atmospheric dehydration in the wintertime Arctic.
 Both observations and transport model simulations show dust concentrations in the range of 0.02–5 µg m−3 in the Arctic (Figures 1 and 4). Variations of dust in this range are shown to cause large variations in cloud liquid water path, either by preventing the formation of liquid clouds (Figure 9, for the case of uncoated dust) or causing glaciation of supercooled clouds (Figure 10, for the case of coated dust). The combination of global and parcel model results implies that low concentrations and acid coating of dust aerosols are essential conditions for winter season mixed-phase clouds in the lower Arctic troposphere.
5.1 Influence of Acid Coating of Dust on Cloud Phases in the Arctic
 Observations at Barrow, Surface Heat Budget of the Arctic Ocean (SHEBA), and Eureka stations in the Arctic show that the mixed-phase clouds are the most common cloud type roughly when the lowest in-cloud temperature (Tmin) is between −10°C and −25°C [de Boer et al., 2011]. Liquid clouds are more common than ice containing ones at Tmin > −10°C, and ice clouds are more common than liquid containing ones at Tmin < −30°C [de Boer et al., 2011]. Our model results for coated dust (Figure 10) are in general agreement with these observations, while our results for uncoated dust (Figure 9) have a warm bias in temperature for occurrence of ice clouds. The observed phase fractions for Arctic clouds appear to be consistent with significant influence of air pollution and acid coating of dust aerosols.
5.2 Influence of Dust on Cloud Liquid Water Content and Radiation
 Cloud top temperatures observed at Barrow between 1999–2007 average to −12°C under large-scale upward motions and −18°C under large-scale downward motions from December to March [Zhao and Wang, 2010]. The average in-cloud LWP is 33 g m−2 for a total cloud fraction of 68% in the winters between 1998 and 2008 [Dong et al., 2010]. For typical dust concentrations between 0.1 and 0.5 µg m−3 near Barrow (Figure 4) and temperatures from 245–261 K, the parcel-model-predicted LWPs (see Figures 10a and 10c, for w = 10 cm/s) are generally lower than 30 g m−2 without ice settling and mostly higher than 30 g m−2 with ice settling. A larger threshold size for ice precipitation would have produced results in better agreement with the observed LWP.
 The mean longwave cloud radiative forcing observed at the Barrow site is 9.6 W m−2 (equal to net longwave up for clear sky (39.6) and net longwave up for all sky (30), based on surface measurements) in the winter months [Dong et al., 2010]. The wintertime cloud radiative forcing is twice as large (~20 W m−2) at the SHEBA site farther north [Intrieri et al., 2002]. The observations in SHEBA show that the longwave cloud radiative forcing of the Arctic surface increases with LWP until LWP = 30 g m−2, after which point clouds emit radiation as blackbodies, and increasing LWP has no further impact on downward longwave radiation [Shupe and Intrieri, 2004]. Transport of midlatitude pollution has increased Arctic dust concentration by 0.1–0.5 µg m−3 in winter (Figure 4), which can decrease LWP to values significantly below 30 g m−2 (Figure 10) and increase the rate of surface cooling by longwave radiation (changes up to 20 W m−2). In contrast, low concentrations of uncoated dust and other aerosols which are able to nucleate ice through deposition freezing at warm temperatures (T > −10°C) would effectively prevent the formation of liquid containing clouds (Figure 9), dehydrate the Arctic atmosphere, and significantly increase surface cooling in winter.
5.3 Trends in Pollution Sources of Aerosol
 Measurements of PM10 have been made routinely at many locations in Eurasia and North America. PM10 has decreased by 44% in Europe from 1992 to 2009, 33% in the U.S. from 1993 to 2010, 10% in Canada from 1994 to 2009, and 26% in China from 2000 to 2011 [Wang et al., 2012]. It is not known how much of the decreases are associated with insoluble dust. The decrease in China occurred while the country has seen a marked growth in automobiles and road traffic. The decrease in PM10 is probably more related to decreasing emissions of SO2, NO and hydrocarbon gases, coal fly ash, and carbonaceous aerosols, and less related to traffic and fugitive dust. Whether the changes in midlatitude pollution have any impact on Arctic IN and clouds should be investigated in global models.
 In this study, we simulated contributions of natural and anthropogenic pollution sources to the observed dust (crustal, mineral, and metal oxides) at Alert, Canada, using a global chemical transport model. The observations suggest a significant contribution of pollution dust emitted in the midlatitudes in winter and early spring. The transport model predicts a peak of natural dust in spring and low concentrations from summer to winter. Pollution dust (e.g., road resuspension and coal fly ash) in the Arctic is predicted to be highest in winter to early spring and lowest from late spring to autumn, in agreement with observations. Modeling of transport of the pollution dust into the Arctic basin is a challenging task with multiple uncertainties, including a lack of accurate emissions inventory, parameterization of wet deposition (in-cloud scavenging by snow), precipitation, and transpolar winds.
 We need to face this challenge in order to improve climate and sea ice simulations in the Arctic. Dust particles may serve as the main ice nucleus in the Arctic in winter. Cloud parcel model simulations indicate that variation of dust concentration in the range of observations can either prevent the formation of liquid clouds or cause variable degree of glaciation of mixed-phase clouds, especially in winter to early spring when air temperature in the lower troposphere is frequently below −15°C. Acidic coating of dust can shift ice nucleation from a regime of deposition freezing to a regime of immersion freezing, with major consequences on the partitioning of condensed water between liquid and ice phases at temperatures typical of the lower troposphere during the polar night. Therefore, dust particles may influence the liquid water path and cloud optical thickness and can potentially affect Arctic seasonal climate [Prenni et al., 2007]. Longwave cloud radiative forcing in winter to early spring may be modulated by long-range transport of natural and anthropogenic dust, which is a major control on growth of sea ice during the polar night. In conclusion, humans can modify Arctic climate through perturbations to ice nuclei as well as greenhouse gases and light reflecting/absorbing aerosols.
 The Canadian Arctic Aerosol Chemistry Program (CAACP) data were obtained from the National Atmospheric Chemistry (NAtChem) Database and Analysis Facility of Environment Canada (www.msc-smc.ec.gc.ca/natchem). The authors gratefully acknowledge CAACP for its data and the NAtChem Database and Analysis Facility for access to the standardized data files and metadata. Bud Moxim helped maintain the global chemical transport model. Paul Ginoux provided useful suggestions. Catherine Raphael helped with graphics. The author is most grateful to three anonymous reviewers whose comments led to significant improvements of the paper.