Factors controlling black carbon distribution in the Arctic

We investigate the sensitivity of black carbon (BC) in the Arctic, including BC concentration in snow (BCsnow, ng g−1) and surface air (BCair, ng m−3), as well as emissions, dry deposition, and wet scavenging using the global threedimensional (3-D) chemical transport model (CTM) GEOSChem. We find that the model underestimates BCsnow in the Arctic by 40 % on average (median= 11.8 ng g−1). Natural gas flaring substantially increases total BC emissions in the Arctic (by ∼ 70 %). The flaring emissions lead to up to 49 % increases (0.1–8.5 ng g−1) in Arctic BCsnow, dramatically improving model comparison with observations (50 % reduction in discrepancy) near flaring source regions (the western side of the extreme north of Russia). Ample observations suggest that BC dry deposition velocities over snow and ice in current CTMs (0.03 cm s−1 in the GEOS-Chem) are too small. We apply the resistance-in-series method to compute a dry deposition velocity (vd) that varies with local meteorological and surface conditions. The resulting velocity is significantly larger and varies by a factor of 8 in the Arctic (0.03–0.24 cm s−1), which increases the fraction of dry to total BC deposition (16 to 25 %) yet leaves the total BC deposition and BCsnow in the Arctic unchanged. This is largely explained by the offsetting higher dry and lower wet deposition fluxes. Additionally, we account for the effect of the Wegener–Bergeron–Findeisen (WBF) process in mixed-phase clouds, which releases BC particles from condensed phases (water drops and ice crystals) back to the interstitial air and thereby substantially reduces the scavenging efficiency of clouds for BC (by 43–76 % in the Arctic). The resulting BCsnow is up to 80 % higher, BC loading is considerably larger (from 0.25 to 0.43 mg m−2), and BC lifetime is markedly prolonged (from 9 to 16 days) in the Arctic. Overall, flaring emissions increase BCair in the Arctic (by∼ 20 ng m−3), the updated vd more than halves BCair (by ∼ 20 ng m−3), and the WBF effect increases BCair by 25– 70 % during winter and early spring. The resulting model simulation of BCsnow is substantially improved (within 10 % of the observations) and the discrepancies of BCair are much smaller during the snow season at Barrow, Alert, and Summit (from−67–−47 % to−46–3 %). Our results point toward an urgent need for better characterization of flaring emissions of BC (e.g., the emission factors, temporal, and spatial distribution), extensive measurements of both the dry deposition of BC over snow and ice, and the scavenging efficiency of BC in mixed-phase clouds. In addition, we find that the poorly constrained precipitation in the Arctic may introduce large uncertainties in estimating BCsnow. Doubling precipitation introduces a positive bias approximately as large as the overall effects of flaring emissions and the WBF effect; halving precipitation produces a similarly large negative bias.

Abstract. We investigate the sensitivity of black carbon (BC) in the Arctic, including BC concentration in snow (BC snow , ng g −1 ) and surface air (BC air , ng m −3 ), as well as emissions, dry deposition, and wet scavenging using the global threedimensional (3-D) chemical transport model (CTM) GEOS-Chem. We find that the model underestimates BC snow in the Arctic by 40 % on average (median = 11.8 ng g −1 ). Natural gas flaring substantially increases total BC emissions in the Arctic (by ∼ 70 %). The flaring emissions lead to up to 49 % increases (0.1-8.5 ng g −1 ) in Arctic BC snow , dramatically improving model comparison with observations (50 % reduction in discrepancy) near flaring source regions (the western side of the extreme north of Russia). Ample observations suggest that BC dry deposition velocities over snow and ice in current CTMs (0.03 cm s −1 in the GEOS-Chem) are too small. We apply the resistance-in-series method to compute a dry deposition velocity (v d ) that varies with local meteorological and surface conditions. The resulting velocity is significantly larger and varies by a factor of 8 in the Arctic (0.03-0.24 cm s −1 ), which increases the fraction of dry to total BC deposition (16 to 25 %) yet leaves the total BC deposition and BC snow in the Arctic unchanged. This is largely explained by the offsetting higher dry and lower wet deposition fluxes. Additionally, we account for the effect of the Wegener-Bergeron-Findeisen (WBF) process in mixed-phase clouds, which releases BC particles from condensed phases (water drops and ice crystals) back to the interstitial air and thereby substantially reduces the scavenging efficiency of clouds for BC (by 43-76 % in the Arctic). The resulting BC snow is up to 80 % higher, BC loading is considerably larger (from 0.25 to 0.43 mg m −2 ), and BC lifetime is markedly prolonged (from 9 to 16 days) in the Arc-tic. Overall, flaring emissions increase BC air in the Arctic (by ∼ 20 ng m −3 ), the updated v d more than halves BC air (by ∼ 20 ng m −3 ), and the WBF effect increases BC air by 25-70 % during winter and early spring. The resulting model simulation of BC snow is substantially improved (within 10 % of the observations) and the discrepancies of BC air are much smaller during the snow season at Barrow, Alert, and Summit (from −67-−47 % to −46-3 %). Our results point toward an urgent need for better characterization of flaring emissions of BC (e.g., the emission factors, temporal, and spatial distribution), extensive measurements of both the dry deposition of BC over snow and ice, and the scavenging efficiency of BC in mixed-phase clouds. In addition, we find that the poorly constrained precipitation in the Arctic may introduce large uncertainties in estimating BC snow . Doubling precipitation introduces a positive bias approximately as large as the overall effects of flaring emissions and the WBF effect; halving precipitation produces a similarly large negative bias.
ing (Bond et al., 2013, and references therein). Warren and Wiscombe (1985) highlighted the climate effect of fallen soot from "smokes" for a nuclear war scenario, which reduced the surface reflectivity of snow and sea ice in the Arctic. Measurements by Clarke and Noone (1985) showed that there was ample amount of BC in the Arctic snow to exert climate impacts in the region. Using observations of BC snow , Hansen and Nazarenko (2004) quantified, for the first time, the albedo reduction due to BC deposition on snow and ice (2.5 % on average) across the Arctic. The snow-albedo effect of BC in the Arctic has since received wide attention. Numerous studies have examined the snow-albedo change in this region due to BC deposition (Jacobson, 2004;Marks and King, 2013;Namazi et al., 2015;Tedesco et al., 2016) and estimated the associated surface BC snow-albedo radiative forcing to be substantial (0.024-0.39 W m −2 ) in the Arctic (Bond et al., 2013, and references therein;Flanner, 2013;Jiao et al., 2014;Namazi et al., 2015), comparable to the forcing of tropospheric ozone in springtime Arctic (0.34 W m −2 , Quinn et al., 2008). BC deposited on snow and ice is likely to be an important reason for unexpectedly rapid sea-ice shrinkage in the Arctic (Koch et al., 2009;Goldenson et al., 2012;Stroeve et al., 2012). Widespread surface melting of the Greenland Ice Sheet was attributed to rising temperatures and reductions in surface albedo resulting from deposition of BC from Northern Hemisphere forest fires (Keegan et al., 2014;Tedesco et al., 2016).
To better constrain the radiative forcing and the associated uncertainties of the BC snow-albedo effect in the Arctic, it is imperative to improve the diagnosis and prediction of BC snow in the region. Previous studies found large discrepancies between modeled and observed BC snow (up to a factor of 6) in the Arctic (e.g., Flanner et al., 2007;Koch et al., 2009). A comprehensive survey of BC snow observations across the Arctic (∼ 1000 snow samples) by Doherty et al. (2010) provided a unique opportunity to constrain BC snow in the region. Bond et al. (2013) compared results of BC snow from the Community Atmospheric Model version 3.1 (CAM3.1) (Flanner et al., 2009) and the Goddard Institute of Space Studies (GISS) model (Koch et al., 2009) with the observations from Doherty et al. (2010), averaged over the eight Arctic sub-regions ( Fig. 1) as defined by Doherty et al. (2010). The resulting ratio of modeled to observed BC snow (sub-regional means) was 0.6-3.4 for CAM3.1 and 0.3-1.6 for GISS. Jiao et al. (2014) found large discrepancies in BC snow (up to a factor of 6) between results from the Aerosol Comparisons between Observations and Models (AeroCom; http://aerocom.met.no/) and the Doherty et al. (2010) observations. They also found large variations in BC deposition fluxes among the AeroCom models. Jiao et al. (2014) further pointed out that BC transport and deposition processes are more important for differences in simulated BC snow than differences in snow meltwater scavenging rates or emissions in models.
Studies have shown that Arctic atmospheric BC on average cools the surface due to surface dimming, while BC in the lower troposphere warms the surface with a climate sensitivity (surface temperature change per unit forcing) of 2.8 ± 0.5 K W −1 m 2 due to low clouds and sea-ice feedbacks that amplify the warming (e.g., Flanner, 2013). This sensitivity is a factor of 2 larger than that of the BC snowalbedo feedback (1.4 ± 0.7 K W −1 m 2 , Flanner, 2013), a factor of 4 larger than that of CO 2 (0.69 K W −1 m 2 , Bond et al., 2013), and much larger than that of tropospheric ozone (0.2 K W −1 m 2 ; Shindell and Faluvegi, 2009). However, estimates of BC air in the Arctic are associated with large uncertainties (Textor et al., 2006(Textor et al., , 2007Koch et al., 2009;Liu et al., 2011;Browse et al., 2012;Sharma et al., 2013). In general, current models successfully reproduced the decadal declining trends observed at the surface sites Barrow, Alert, and Zeppelin (Sharma et al., 2004(Sharma et al., , 2006Eleftheriadis et al., 2009), but failed to reproduce the seasonal cycles of BC air observed at the aforementioned sites, with large underestimates during the Arctic haze season (Textor et al., 2006(Textor et al., , 2007Koch et al., 2009;Liu et al., 2011;Browse et al., 2012;Sharma et al., 2013;Eckhardt et al., 2015). Specifically, mean BC air during January to March was underestimated by about a factor of 2 for the mean of all models, although the discrepancy is up to a factor of 27 for individual models (Eckhardt et al., 2015). The low biases are likely due to uncertainties associated with estimates of BC emissions in Russia , treatments of BC aging in the models He et al., 2016), excessive dry deposition of BC (Huang et al., 2010;Liu et al., 2011), wet scavenging of BC (Koch et al., 2009;Huang et al., 2010;Bourgeois and Bey, 2011;Liu et al., 2011), or overly efficient vertical mixing (Koch et al., 2009). Studies Huang et al., 2015) have pointed out that the low biases of BC air during the Arctic haze season are partially due to uncertainties in the estimates of BC emissions in Russia, resulted from biases in both BC emission rates and spatial distributions. A likely missing source of BC emissions in Russia is natural-gas-flaring emissions, most of which cluster in the western side of the extreme north of Russia (Stohl et al., 2013). Although in totality gas-flaring emissions are a rather small fraction of global BC emissions, their proximity to the Arctic can conceivably result in a disproportionately large impact. Dry deposition of BC on snow and ice is yet another poorly understood and quantified process. Observations show that v d over snow-and ice-covered surfaces vary by orders of magnitude (0.01-1.52 cm s −1 ; Hillamo et al., 1993;Bergin et al., 1995;Nilsson and Rannik, 2001;Gronlund et al., 2002;Held et al., 2011;Wang et al., 2014). Current chemical transport models (CTMs) tend to assume uniform and low dry deposition velocities over such surfaces to capture the high surface BC air during the Arctic haze season Sharma et al., 2013). For instance, Wang et al. (2011) used a uniform v d of 0.03 cm s −1 over snow and ice and found a better comparison with BC air mea- surements during the Arctic haze season. However, this value is probably too low for snow-covered land surfaces with larger roughness length. Additionally, observations show that BC scavenging efficiency in clouds varies from 0.06 to 0.7 depending on liquid water contents, temperature, and ice mass fraction because of the Wegener-Bergeron-Findeisen (WBF) process in mixed-phase clouds Verheggen et al., 2007). However, in most of the current Ae-roCom models, BC scavenging is poorly treated Bourgeois and Bey, 2011) or entirely missing  in mixed-phase clouds, which cover the Arctic in ∼ 40 % of the time through a whole year (Zhang et al., 2010). For example, BC scavenging in mixed-phase clouds was treated the same as that in warm clouds in the GEOS-Chem . In ECHAM5-HAM2, BC scavenging efficiency in mixed-phase clouds was set up as 0.06, the lowest observed value in those clouds (Bourgeois and Bey, 2011).
Constraining individual processes of BC is often challenging. Therefore, our focus is more geared toward highlighting missing processes or ones that were previously unaccounted for in governing BC in the Arctic, particularly BC deposition in the region. We first examine and incorporate gas-flaring emissions of BC, which was missing in previous emission estimates yet account for a large fraction of BC emissions in the Arctic as suggested by Stohl et al. (2013) (Sect. 4.1). We then discuss and improve the simulation of v d for BC over snow and ice, which varies by orders of magnitude but was treated as a uniform value by previous studies (Sect. 4.2). We then analyze BC wet scavenging efficiency in mixed-phase clouds accounting for effects of WBF (Sect. 4.3). Finally, we estimated the sensitivity of BC snow to precipitation in the Arctic (Sect. 4.4). We also use BC air as an additional constraint of these simulations.

Measurements of BC in snow
The most comprehensive measurements of BC snow were in eight sub-regions in the Arctic: Alaska, Arctic Ocean, Canadian Arctic, Canadian sub-Arctic, Greenland, Russia, Ny-Ålesund, and Tromsø, mostly from March to May during (Doherty et al., 2010; data available at http:// www.atmos.washington.edu/sootinsnow/). Samples were for full snowpack depth and the sampling sites are shown in Fig. 1 (color-coded by the sub-regions). These observations provide a reasonable constraint on Arctic-wide annual mean radiative effect from BC deposited in snow (Jiao et al., 2014). Doherty et al. (2010) measured the light absorption of impurity in snow samples using the integrating sphere/integrating sandwich optical method and derived equivalent, maximum, and estimated BC snow using the wavelength-dependent absorption of BC and non-BC fractions (Doherty et al., 2010). We use here the estimated BC snow . The largest sources of uncertainty stem from uncertainties of BC mass absorption cross section (MAC), BC absorption Ångstrom exponent (Å BC ), and non-BC absorption Ångstrom exponent (Å non−BC ) constituents. Doherty et al. (2010) used MAC = 6.0 mg 2 g −1 (at 500 nm), the MAC of their calibration filters. Using MAC = 7.5 mg 2 g −1 (at 500 nm) as recommended by Bond and Bergstrom (2006) would increase the estimated BC snow by ∼ 25 %. Doherty et al. (2010) used Å BC = 1.0 (range of 0.8-1.9) and Å non−BC = 5.0 (range of 3.5-7.0) in their derivation and estimated a 50 % error in the estimated BC snow . Additional uncertainties include instrumental uncertainty (≤ 11 %), under-catch correction (±15 %), and loss of aerosol to plastic flakes in the collection bags (−20 %) for samples from western Russia and the Canadian sub-Arctic. The overall uncertainty of the estimated BC snow is < 60 %.

Measurements of BC in surface air
In situ measurements of BC air from 2007 to 2009 are available at five sites within the Arctic Circle (  Table 1. Denali is part of the Interagency Monitoring of PROtected Visual Environment (IMPROVE) network (Malm et al., 1994; data available at http://vista.cira. colostate.edu/improve/). IMPROVE measurements are made every 3 days and 24 h averages are reported. Thermal Optical Reflectance (TOR) combustion method is used based on the preferential oxidation of organic carbon (OC) and BC at different temperatures (Chow et al., 2004). BC-like products of OC pyrolysis can lead to an overestimate of the BC mass. The uncertainties of the TOR method are difficult to quantify (Park et al., 2003;Chow et al., 1993).
Barrow is part of the NOAA Global Monitoring Division (GMD) network, where BC light absorption coefficients have been measured from a particle soot absorption photometer (PSAP) since 1997 (Bond et al., 1999;Delene and Ogren, 2002; data available at http://www.esrl.noaa.gov/gmd/aero/ net/). PSAP measures the change in light transmission at three wavelengths (467, 530 and 660 nm) through a filter on which particles are collected. We used the measurements at 530 nm in this study. Site Barrow is about 8 km northeast of the village of Barrow and is less than 3 km southeast of the Arctic Ocean. Given that the site has a prevailing eastnortheast wind off the Beaufort Sea, it receives minimal influence from local anthropogenic emissions and is strongly affected by weather in the central Arctic.
BC air at Alert were measured using an aethalometer model AE-6 with one wavelength operated by Environment Canada (Sharma et al., 2004(Sharma et al., , 2006; data available at http: //www.ec.gc.ca/). The instruments measure the attenuation of light transmitted through particles that accumulate on a quartz fiber filter at 880 nm. Alert, located the furthest north of the five sites on the northeastern tip of Ellesmere Island, is most isolated from continental sources (Hirdman et al., 2010).
The Zeppelin observatory is part of the European Supersites for Atmospheric Aerosol Research, where BC mass concentrations are also measured by an aethalometer and reported for seven wavelengths (370,470,520,590,660,880, and 950 nm) (Eleftheriadis et al., 2009; data available at http://ebas.nilu.no/). We use the 520 nm data. Measurements at site Zeppelin, on the mountain Zeppelin on the island archipelago of Svalbard, were generally considered to represent the free-troposphere conditions (Eleftheriadis et al., 2009).
BC mass concentrations were also measured by an aethalometer at Summit (von Schneidemesser et al., 2009; data available at http://www.esrl.noaa.gov/gmd/aero/net/), on the center of the Greenland glacial ice sheet. The Summit site is at high elevation (3.2 km a.s.l.) and surrounded by flat and homogeneous terrain (Hirdman et al., 2010).
The uncertainty of filter-based absorption measurements of BC (PSAP and aethalometer) lies in empirical corrections of the overestimated absorption if light transmission is also affected by particulate light scattering (Bond et al., 1999). Accuracy of this correction is 20-30 % (Delene and Ogren, 2002;Weingartner et al., 2003;Virkkula et al., 2005). Additional uncertainty results from the empirical conversion from optical response to BC mass using an assumed MAC, which depends on the composition and morphology of the particles used in the calibration of the instrument and on the specific technique used to quantify the BC mass (Clarke et al., 1987;Slowik et al., 2007). The MAC of BC varies by up to a factor of 4 (5-20 m 2 g −1 ) (Weingartner et al., 2003). We use 9.5 m 2 g −1 for the station Barrow at wavelength 530 nm as recommended for the ARCTAS period (McNaughton et al., 2011;Wang et al., 2011). The MAC used at Alert , Zeppelin (Eleftheriadis et al., 2009), and Summit (Hagler et al., 2007) are 19, 15.9, and 20 m 2 g −1 . The uncertainty of absorption enhancement by non-BC absorbers (organic carbon and mineral dust) is generally difficult to quantify unless the non-BC absorbers contribute more than 40 % of absorption (Petzold et al., 2013).
3 Model description and simulations 3.1 GEOS-Chem simulation of BC GEOS-Chem is a global three-dimensional (3-D) CTM driven with assimilated meteorology from the Goddard Earth Observing System (GEOS) of the NASA Global Modeling and Assimilation Office. The GEOS-5 meteorological data sets are used to drive model simulation at 2 • latitude × 2.5 • longitude resolution and 47 vertical layers from the surface to 0.01 hPa. The model averages over "polar caps" beyond ±84 • to compensate for artificial polar singularities. Tracer advection is computed every 15 min with a flux-form semi-Lagrangian method (Lin and Rood, 1996). Tracer moist convection is computed using GEOS convective, entrainment, and detrainment mass fluxes as described by Allen et al. (1996a, b). Deep convection is parameterized using the relaxed Arakawa-Schubert scheme (Moorthi and Suarez, 1992;Arakawa and Schubert, 1974), and the shallow convection treatment follows Hack (1994). BC aerosols are emitted by incomplete fossil fuel and biofuel combustion and biomass burning. We use global BC emissions from Bond et al. (2007) with updated emissions in Asia from Zhang et al. (2009). Biomass-burning emissions are from the Global Fire Emissions Database version 3 (GFEDv3) (van der Werf et al., 2010) with updates for small fires in Randerson et al. (2012). It is assumed that 80 % of the freshly emitted BC aerosols are hydrophobic (Park et al., 2003) and are converted to hydrophilic with an e-folding time of 1.15 days, which yields a good simulation of BC export efficiency in continental outflow (Park et al., 2005). Dry deposition in the model is computed using a resistance-in-series method (Wesely, 1989;Zhang et al., 2001), whereas it assumes a constant aerosol v d of 0.03 cm s −1 over snow and ice (see Sect. 3.3). Wet deposition follows Liu et al. (2001), with updates as described in Wang et al. (2011).

Gas-flaring emissions of BC
Gas flaring is the controlled burning of natural gas in petroleum producing areas, particularly in areas lacking gas transportation infrastructure (Elvidge et al., 2009(Elvidge et al., , 2011. It is estimated that 3.5 % of the world's natural gas is flared (Elvidge et al., 2016) and results in a large amount of greenhouse gas emissions (13 662.6 Gg of CO 2 , Bradbury et al., 2015). Stohl et al. (2013) derived BC emissions from gas flares by multiplying gas-flaring volumes by emission factors. The flaring volumes were estimated using low-light imaging data acquired by the Defense Meteorological Satel-lite Program (DMSP) (Elvidge et al., 2011). The DMSP estimates of flared gas volume are based on a calibration developed with a pooled set of reported national gas-flaring volumes and data from individual flares. Derived using labbased, pilot-based, and field-based approaches, currently available BC emission factors vary by orders of magnitudes (0.0-6.4 g m −3 ; Fawole et al., 2016). Stohl et al. (2013) derived BC emission factor (1.6 g m −3 ) based upon emission factors of particulate matter from flared gases. The resulting gas-flaring emissions (228 Gg yr −1 ) account for ∼ 5 % of global anthropogenic emissions (4.8 Tg yr −1 ; Bond et al., 2007) and ∼ 3 % of global total emissions (8.5 Tg yr −1 ; including anthropogenic emissions from Bond et al., 2007, andZhang et al., 2009, and biomass-burning emissions from Randerson et al., 2012). However, the largest contributor Russia, contributing ∼ 30 % to the global flaring volume, is located in the clean Arctic Circle. About 40 % of BC emissions in the Arctic (115 Gg yr −1 ) are from gas flaring (48 Gg yr −1 ), shown in Fig. 1. It is estimated that flaring emissions contribute 42 % to the annual mean BC air at surface in the Arctic (Stohl et al., 2013). However, to our knowledge, no study so far has investigated the contribution of flaring emissions to BC snow in the Arctic. Thus, we included flaring emissions from Stohl et al. (2013, data on flaring emissions are available at http://eclipse.nilu.no upon request) and investigated the contribution of flaring emissions to BC snow and BC air in the Arctic in experiment B (Table 2).

Dry deposition over snow and ice
Nilsson and Rannik (2001) conducted eddy-covariance flux measurements of aerosol number dry deposition in the Arctic Ocean and found a mean v d of 0.19 cm s −1 over open sea, 0.03 cm s −1 over ice floes and 0.03-0.09 cm s −1 over leads (Table 3). Following Nilsson and Rannik (2001),  imposed v d = 0.03 cm s −1 for aerosols over snow and ice. They found improved agreements of simulated sulfate with in situ observations in spring and winter in the Arctic. Wang et al. (2011), also imposing v d = 0.03 cm s −1 for aerosols over snow and ice, found better agreements for BC at the same stations as used by Fisher et al. (2011). They thus recommended a uniform v d = 0.03 cm s −1 for sulfate and BC over snow and ice. To capture the winter and spring haze, other studies also used relatively low v d = 0.01-0.07 cm s −1 Sharma et al., 2013). These low values, however, are likely too small for snow-covered land surface, where larger roughness lengths reduce the aerodynamic resistance thereby increasing v d (Gallagher, 2002). The roughness length is 0.005 m for sea ice and 0.03-0.25 m for snow-covered land surface with grass and scattered obstacles (Wieringa, 1980). As a result, v d is larger over a snowcovered land surface than over sea ice. Observed values over snow and ice are 0.01-2.4 cm s −1 for aerosol particles in general and 0.01-1.52 cm s −1 for BC in particular (Table 3). Again, this suggests that a uniform value of v d = 0.03 cm s −1 BC aging e-folding time 1.15 days Deposition Dry 0.03 cm s −1 over snow/ice and resistance-in-series over all surfaces and resistance-in-series over other (Wesely, 1989;Zhang et al., 2001) surfaces  Wet Liu et al. (2001) with updates from Wang et al. (2011) riming: scavenging efficiency for hydrophilic account for both riming and WBF in BC is 100 % in warm and mixed-phase clouds mixed-phase clouds (Fukuta and Takahashi, 1999;Verheggen et al., 2007;Cozic et al., 2007) , 2011). The simulated dry deposition velocities are sampled at the same region during the same time period as observations for BC particles. b Observations are medians of dry deposition velocities derived from aerosol number fluxes measured by an eddy-covariance system over different surface types in late July and early August in 1996 in the Arctic Ocean for ultra-fine and Aitken-mode aerosol particles (Nilsson and Rannik, 2001). Simulations are sampled in the same region during the same months as observations in 2008 for BC particles. c Sulfate dry deposition velocities were derived based on particle mass using surrogate surfaces and impactor data at site Summit, Greenland in July 1993 (Bergin et al., 1995). Simulations are sampled at the same site during July 2008 for BC particles. d Sulfate dry deposition velocities were derived based on particle mass from Cascade impactor at Dye 3 on the south-central Greenland Ice Sheet in March 1989 (Hillamo et al., 1993). Simulations are sampled at the same site during March 2008 for BC particles. e The dry deposition velocities specific to BC particles were derived from measured surface enhancement of BC in snow between two snow events at Changbai Mountain in northern China in winter (December, January, and February) in (Wang et al., 2014. Simulations are sampled at the same site during the same time period for BC particles. is problematic. We apply the resistance-in-series method to calculate v d of BC over snow and ice, as a function of aerodynamic resistance, particle density and size, and surface types (experiment C, Table 2). We would like to note that most of these observations (Held et al., 2011;Nilsson and Rannik, 2001;Bergin et al., 1995) were from summertime Arctic (June-August) and clean regions (e.g., the Arctic Ocean and Greenland) far from anthropogenic pollution. In addition, most of the v d measurements are for general aerosol particles. The only available dry deposition velocities specific to BC particles are derived from the strong surface enhancement of BC snow between two snow events at Mt. Changbai (42.5 • N, 128.5 • E; 0.74 km) in northern China (Table 3). Wang et al. (2014) derived v d = 0.16-1.52 cm s −1 . Despite uncertainties from sublimation (Wang et al., 2014), these measurements suggest that the low v d used in previous studies Liu et al., 2011;Wang et al., 2011;Sharma et al., 2013) might underestimate the role of dry deposition during the snow season, particularly near source regions. Wang et al. (2014) concluded that dry deposition in the boundary layer may dominate over wet deposition (a factor of 5 larger) during the dry season in some regions, particularly near source regions with high BC air . It is thus imperative to obtain measurements of v d of BC in polluted regions in Russia and northern Europe in spring, when radiative forcing associated with a BC snowalbedo effect is at a maximum (Flanner, 2013).

Wegener-Bergeron-Findeisen (WBF) process in mixed-phase clouds
Most AeroCom models (Textor et al., 2006) parameterize rainout rate following Giorgi and Chameides (1986). The rainout ratio is proportional to the precipitation formation rate and mass mixing ratio of BC in a condensed phase in clouds, which is determined by the scavenging efficiency of BC (r scav ), where r scav is the scavenging efficiency and quantifies the partition of BC aerosols between condensed phase and the interstitial air; [BC] condensed is the mass mixing ratio of BC in condensed phase, including water drops and ice crystals in clouds, and [BC] interstitial is the mass mixing ratio of BC in the interstitial air. The hygroscopicity and size of BC-containing particles are determining factors for r scav (Sellegri et al., 2003;Hallberg et al., 1992Hallberg et al., , 1995. Internal mixing with soluble inorganic species enhances the r scav for aged BC particles (Sellegri et al., 2003). For instance, r scav is 0.39 ± 0.16 for BCcontaining particles with a diameter smaller than 0.3 µm and a small fraction (38 %) of soluble inorganic material. It increases to 0.97 ± 0.02 for particles with a diameter larger than 0.3 µm and a larger fraction (57 %) of soluble inorganic material (Sellegri et al., 2003). In addition to particle properties, cloud microphysics and dynamics play a significant role in determining r scav of BC in mixed-phase clouds (Hitzenberger et al., 2000(Hitzenberger et al., , 2001Cozic et al., 2007;Hegg et al., 2011). Measured r scav of BC decreased from 0.60 in liquid only clouds to 0.05-0.10 in mixed-phase clouds, a reduction of more than a factor of 5 Henning et al., 2004;Verheggen et al., 2007). Such reduction was attributed to the effect of the WBF process . In mixed-phase clouds, ice crystals grow at the expense of water drops when the environmental vapor pressure is higher than the saturation vapor pressure of ice crystals but lower than the saturation vapor pressure of water droplets (Wegener, 1911;Bergeron, 1935;Findeisen, 1938). Therefore, BC-containing aerosol particles in water drops, which evaporate to dryness, are released back to the interstitial air and consequently r scav is reduced. Another process, riming (Hegg et al., 2011), in mixed-phase clouds has an opposite effect on BC scavenging. When ice particles fall and collect the water drops along the pathway, the snow particles show rimed structure and the scavenging efficiency remains the same. The riming rate is determined by the terminal velocity of snowflakes, ice crystals, and liquid water contents (LWC) in clouds (Fukuta and Takahashi, 1999).
Previously, only the hygroscopicity of BC-containing particles is considered in BC r scav in models (Wang et al., 2011, and references therein). It is typically assumed that 100 % of hydrophilic BC particles are readily incorporated into cloud drops and all hydrophobic BC particles remain in the interstitial air in warm and mixed-phase clouds. This treatment of mixed-phase clouds as liquid phase is likely to overestimate r scav in mixed-phase clouds. In models that include mixedphase clouds, assumptions still need to be made about r scav . A uniform scavenging efficiency (0.4 or 0.06) for all mixedphase clouds has been imposed (Stier et al., 2005;Bourgeois and Bey, 2011), while observations show that BC scavenging efficiency varies dramatically with temperature and ice mass fraction Henning et al., 2004;Verheggen et al., 2007).
In experiment D (Table 2), we discriminate WBF-vs. riming-dominated conditions and parameterize BC scavenging efficiency under the two conditions separately in mixedphase clouds (248 K < T < 273 K; Garrett et al., 2010). We assume that riming dominates when temperature is around −10 • C (261 K < T < 265 K) and LWC is above 1.0 g m −3 , following Fukuta and Takahashi (1999). The WBF process dominates otherwise. Our parameterization of the effect of the WBF process on BC scavenging efficiency is based on the measurements at Mt. Jungfraujoch (46.4 • N, 8 • E; 3.85 km), an elevated mountainous site far from pollution sources and regularly engulfed in clouds (30 % of the time) . We evaluated the effects of WBF on global BC distribution and tested the sensitivity of the simulation to the switch temperature from warm clouds to mixed-phase clouds and from mixed-phase clouds to ice clouds in a companion study . In this study, we focus on the effects of WBF on BC distribution in the Arctic.

BC concentration in snow
In snow models, such as SNICAR, the initial surface BC snow is defined as the ratio of BC deposition to snow precipitation (Flanner et al., 2007). Here we approximate BC snow using BC deposition flux and snow precipitation rate, following Kopacz et al. (2011), and He et al. (2014a: where F BC,dep , F wet_dep , and F dry_dep are total, dry, and wet deposition flux of BC, and F snow is the snow precipitation. The top and bottom snow depth of each sample are provided in the observation data set (Doherty et al., 2010). We accumulate snow precipitation (GEOS-5) in the model from the collection date backward until the modeled snow depths, respectively, reach the observed top and bottom depths of the snow sample, then the two dates are stored. We use the average BC deposition fluxes and snow precipitation between the two dates to estimate the BC snow for the sample. The rate of snow accumulation at the surface is estimated as snow precipitation flux (kg m −2 s −1 ) over snow density (kg m −3 ). The observed annual average snow density is 300 kg m −3 over the Arctic basin, increasing from 250 kg m −3 in September to 320 kg m −3 in May with little geographical variation across the Arctic (Warren et al., 1999;Forsström et al., 2013). We use the annual average snow density in the estimate. The above estimate of BC snow ignores many processes that may alter the BC snow concentrations, such as wind redistribution of surface snow, sublimation, and meltwater flushing (Doherty et al., 2010Wang et al., 2014). Wind redistribution of surface snow is a subgrid-scale phenomenon. Except for turbulent-scale wind direction and strength, smallscale topography also plays an important role in surface snow distribution; this process is difficult to simulate in global models. Precipitation rate and relative humidity in much of the Arctic are low, so in some areas appreciable (up to 30-50 %) surface snow is lost to sublimation (Liston and Sturm, 2004). BC snow at surface can thus be underestimated by our method. We filtered out snow samples collected during the melting season, so the meltwater flushing has little effect on our estimate.
To reduce the biases in comparison of model results and observations, we organize the observations as follows: (1) observations from March to May in 2007-2009 are used while those from June to August are excluded because our estimate of BC snow does not resolve snow melting, (2) we exclude observations with obvious dust or local wood-burning contaminations as described in Doherty et al. (2010), and (3) we average the observations in the same model grid and snow layer and collected on the same day. Table 2 summarizes various model simulations in the present study. Experiment A is the standard case. We include gas-flaring emissions in experiment B (Sect. 3.2). Contrasting experiments B and A thus offer insights to the contribution of gas-flaring emissions on BC in the Arctic. Experiment C includes the updated v d (Sect. 3.3) as well as the gas-flaring emissions. The difference of experiment B and C denotes the effects of updated v d to BC distribution. Experiment D includes temperature-based WBF parameterization (Sect. 3.4) as well as the gas flaring and v d updates. The effects of WBF to BC in the Arctic are shown by the difference of experiment C and D. Additional simulations are described where appropriate. In our discussion of the results of the model runs, we assume that there is little or no interaction between each of the updates; i.e., the order in which the processes have been included does not affect the overall results.

The effects of gas flares, dry deposition, WBF, and precipitation
We discuss the effects of gas-flaring emissions, dry deposition, WBF in mixed-phase clouds, and precipitation on BC distribution in the Arctic in this section. The probability density function of observed and GEOS-Chem simulated BC snow in the Arctic is approximately lognormal (Fig. 2a).
The arithmetic mean of observations is 17.4 ng g −1 , larger than the geometric mean of 12.7 ng g −1 and the median of 11.8 ng g −1 (see the vertical lines in Fig. 2 and Table 1). The model reproduces the observed distribution, but underestimates BC snow by 40 % (experiment A). By including flaring emissions (Sect. 4.1), updating v d (Sect. 4.2), and including WBF in mixed-phase clouds (Sect. 4.3), the discrepancy is reduced to −10 %. Gas-flaring emissions lower the discrepancy from −40 to −20 % (experiment B). The updated v d (experiment C) makes insignificant changes to BC snow in the Arctic. WBF (experiment D) further reduces the discrepancy from −20 to −10 %. The resulting BC snow in the eight subregions agree with observations within a factor of 2. This discrepancy is acceptable for global models because it has been suggested that the error due to different spatial sampling of global models (∼ 200 km) and point observations was up to 160 % (Schutgens et al., 2016). In addition, BC air at the surface and in the free troposphere is sensitive to the above three processes in the Arctic, particularly during winter and spring (see Sect. 4.1-4.3).

Gas-flaring emissions
Gas-flaring emissions increase total BC emissions by 67 % (from 0.068 to 0.115 Tg yr −1 ) in the Arctic Circle (60 • N and higher latitudes), resulting in a 19 % increase of the total BC deposition (from 0.32 to 0.38 Tg yr −1 ). Flaring emissions increase BC snow (by 0.1-8.5 ng g −1 ) in the eight Arctic  sub-regions. The higher BC snow leads to a significant reduction in the negative biases (by 20-100 %), except in the Arctic Ocean and in Tromsø, where BC snow is already overestimated without flaring emissions (Fig. 3). BC snow in Greenland is not affected by gas-flaring emissions. The reason is 2fold: first, snow samples in Greenland are far from the flares in western Russia, and second, the vertical transport of BC from surface to the upper troposphere is suppressed by the stable atmosphere in the Arctic (Stohl, 2006), resulting in a negligible effect of flaring emissions to BC snow over Greenland (above 1.5 km). The largest enhancement of BC snow from flaring emissions is in the western side of the extreme north of Russia within the Arctic Circle (by 5.0 ng g −1 on average, or, 50 %), which reduces model discrepancy substantially across Russia (from −50 to −30 %). However, simulated BC snow is now too high (by a factor of 2) near the flares (observed value ∼ 19.3 ng g −1 ). The overestimate is likely because of excessively large flaring emission estimates. Yet BC snow is too low (by a factor of 2) in far fields (observed value ∼ 30.7 ng g −1 ), despite a large increase (by 50 %, from 10.5 to 15.5 ng g −1 ) as a result of flaring emissions.
Flaring emissions are assumed to be proportional to flared gas volumes and emission factors. Errors in estimates of flared volumes in Russia are small (within ±5 %, Elvidge et al., 2009). Estimates of emission factors, on the other hand, are known to have several orders of magnitude uncertainties (Schwarz et al., 2015;Weyant et al., 2016). Given limited observations of BC emission factors from actual flares, Stohl et al. (2013) derived BC emission factor based upon emission factors of particulate matter from flared gases. They used a BC emission factor of 1.6 g m −3 , which is more than a factor of 3 higher than that (0.5 g m −3 ) from a lab experiment on fuel mixtures typical in the oil and gas industry (McEwen and Johnson, 2012). Recent field measurements have suggested an even lower emission factor (0.13 ± 0.36 g m −3 ) from ∼ 30 individual flares in North Dakota, with an upper bound of 0.57 g m −3 (Schwarz et al., 2015;Weyant et al., 2016). These studies found that average BC emission factors for individual flares varied by 2 orders of magnitude and, furthermore, two flares from the same flare stack that were resampled on different days showed different BC emission factors (Weyant et al., 2016). They also pointed out that emission factors are not correlated with ambient temperature, pressure, humidity, flared gas volumes, or gas composition. It is thus imperative that extensive measurements of BC emission factors be made in the flare regions.
Yet another source of uncertainty is flare stack height, which is not accounted for in current flaring emission estimates. Typical stack heights vary from 15 to 250 m, some-times above the nighttime boundary layer height of 150-300 m in the Arctic (Di Liberto et al., 2012). The stack height affects the ventilation, dispersion, deposition, and long-range transport of the emissions. For example, local deposition of BC may be suppressed and downwind long-range transport enhanced when the stacks emitted BC in the free troposphere . The lack of proper treatment of flare stack height in the model may partially explain the aforementioned discrepancies of modeled BC snow (biased high in western Russian and low in eastern Russia). Another factor for the underestimate of BC snow in eastern Russia is likely local sources, such as domestic wood burning in nearby villages and fishing camps, diesel trucks on the highway, and coal burning in a power plant, which are unaccounted for in the emission inventory (Doherty et al., 2010, Fig. 1). Although we filter out samples with strong local contamination, it is conceivable that local emissions still add to the background BC snow in eastern Russia. Jiao et al. (2014) have shown that most AeroCom models underestimated BC snow in Russia and pointed to the flaring emissions as a likely cause. Our model results show that even with flaring emissions, which are likely on the high side, BC snow is still too low (by 50 %) in eastern Russia. Therefore, there are likely other factors such as the lack of local emissions in eastern Russia, weak dry deposition fluxes (Sect. 4.2), and excessively low rates of sublimation of surface snow, which contribute to the large model discrepancy in BC snow . Figure 4 shows observed and GEOS-Chem simulated daily BC air from January to March at Zeppelin, a site that is closest to the gas flares in the western side of the extreme north of Russia. The inclusion of flaring emissions captures some of the large spikes in the observed BC air , such as those from late February to March in 2008 and in January 2009. Stohl et al. (2013 found that flaring emissions captured observed large spikes at Zeppelin during a transport event in February 2010 with a high BC / CO ratio, a signature of gas-flaring emissions (CAPP, 2007). The inclusion of flaring emissions results in enhanced BC air , for instance, in February 2007 and in January 2008, which are not seen in the observations. This is largely from the lack of temporal variation of flaring emissions (Weyant et al., 2016). The temporal variation is, however, difficult to characterize based on the current knowledge of flaring emissions on the western side of the extreme north of Russia (Stohl et al., 2013). Flaring emissions also increase BC air during the snow season (September to April) (by 16-19 ng m −3 ) at Barrow and Alert, resulting in substantial reductions of discrepancies (from −47 to −15 % at Barrow and −67 to −46 % at Alert) (Fig. 5). The effect of flaring emissions at Denali in the low Arctic is negligible, because the site is outside of the cold Arctic front (around 65-70 • N in Alaska) (Barrie, 1986;Ladd and Gajewski, 2010), which is a strong barrier for the meridional transport of BC (Stohl, 2006). BC air at Summit (3.22 km a.s.l.), which is mostly in the free troposphere, is not affected by flaring emissions ei-   Table 2 and text for details) daily BC concentrations in air (ng m −3 ) at Zeppelin from January to March in 2007-2009. ther. This is because the vertical transport of BC is suppressed by the stable atmosphere during the snow season in the Arctic (Stohl, 2006).

Dry deposition velocity
It is known that v d of aerosol particles over snow and ice surfaces strongly depends on particle size, surface types and meteorological conditions and varies by orders of magnitude (Table 3). We estimate v d of BC particles as a function of particle properties, aerodynamic resistance, and surface types (Sect. 3.3). The results over the Arctic Ocean and Greenland are shown in Table 3, generally within the observed range. At Mt. Changbai, the model result of BC v d (0.09-0.14 cm s −1 ) is an order of magnitude lower than that derived by Wang et al. (2014) (0.16-1.52 cm s −1 ). The resulting dry deposition fluxes are lower than observations by a factor of 5. We attribute the large discrepancies to two factors. First, the point measurements were at a mountainous site with complex terrain and micro-meteorological conditions. Neither can be resolved in a global model (He et al., 2014a). Second, the values reported by Wang et al. (2014) were estimated from relative enhancements of surface BC snow between two snow events. These estimates are known to have large uncertainties (a factor of 2) from the measured sublimation fluxes and the assumption of snow density (Wang et al., 2014). Compared to the results of uniform v d of 0.03 cm s −1 over snow and ice, the updated v d leads to larger dry deposition fluxes, a larger fraction of dry over total deposition, and rel- atively unchanged total deposition fluxes. Simulated mean BC v d in the eight Arctic sub-regions ( Fig. 1) are 0.03-0.14 cm s −1 , which is considerably larger that the uniform value of 0.03 cm s −1 over snow and ice (Table 5). Correspondingly, the v d are 19-195 % larger in most sub-regions, with the largest increase in Greenland (by 195 %) and over Russia (by 87 %) (Table 5). We find that BC dry deposition flux is more sensitive to v d in source regions (e.g., Russia) than in remote regions, reflecting the high BC air in the former. A comparable increase in v d of BC (from 0.03 to 0.08 cm s −1 ) in Russia and Alaska results in vastly different increases in BC dry deposition flux (87 % in Russia vs. 30 % in Alaska). As expected, larger dry deposition flux depletes BC air thereby reducing wet deposition flux but offsets the reduction in wet deposition. As a result, both total deposition flux and BC snow remain relatively unchanged (< 5 %) in the eight sub-regions, except in Ny-Ålesund and Tromsø. In these latter two regions, the total deposition fluxes are 10-15 % smaller. The lower deposition fluxes reflect efficient removal of BC aerosols over source regions. BC in Ny-Ålesund and Tromsø are primarily from Europe and Russia, transported isentropically in the cold season (Stohl, 2006;Eleftheriadis et al., 2009). Rapid dry deposition in these source regions results in enhanced boundary layer removal hence lower BC loadings in air and a reduced boundary layer outflow . The change in the fraction of dry to total deposition has important implications for BC radiative forcing in the Arctic. The fraction increases from 19 % (7-33 %) to 26 % (14-41 %), by 14-73 %, with the largest increase in Russia (from 23 to 40 %) where BC deposition flux and BC snow are the largest in the Arctic (Tables 4 and 5). Typically, BC particles removed by dry deposition are externally mixed with snow particles, while those removed by wet deposition are internally mixed with snow particles (Flanner et al., 2009(Flanner et al., , 2012. Internal mixing of BC with snow/ice particles increases the absorption cross section of BC/snow composites by about a factor of 2 (Flanner et al., 2012). The enhanced absorption further increases the snow-albedo radiative forcing (He et al., 2014b). It is thus conceivable that the larger dry deposition fraction will lead to less internally mixed BC/snow composite and lower snow-albedo radiative forcing. This effect is critical before the melting season, because melting might quickly eliminate the differences in the mode of BC deposition. Other post-depositional processes include wind-driven drifting and sublimation . The former does not change the fraction of external and internal mixing of BC with snow. The latter might expose BC particles in the internally mixed BC/snow composite and reduce the fraction of internally mixed BC/snow composite. Yet this process occurs slowly in a relatively long time.
Unlike BC snow , BC air is a strong function of v d , particularly during the snow season. With updated v d , model results fail to capture the seasonal cycle of BC air with dramatic decreases during the snow season (by 20-23 ng m −3 , 27-68 %) at Barrow, Alert, and Zeppelin (Fig. 5). The decreases at Barrow and Alert are a direct result of larger dry deposition in the boundary layer because of substantially larger v d (0.07 cm s −1 , Table 5). At Zeppelin (in Ny-Ålesund), where v d is only marginally higher (17 %), the large reduction of BC air (∼ 40 %) is largely attributed to the suppressed transport from proximate source regions in Europe and Russia. 10.4 (0.8) 7.6 (0.7) 9.6 (1.4) 7.6 (0.6) 6.1 (0.7) 2.4 (0.9) 11.4 (1.0) 14.3 (0.7) 35.1 (1.9) C 10.1 (0.8) 7.9 (0.7) 9.3 (1.4) 7.3 (0.6) 6.3 (0.8) 2.8 (1.0) 9.7 (0.9) 13.9 (0.7) 31.6 (1.7) D 11.5 (0.9) 11.6 (1.0) 11.6 (1.7) 7.6 (0.6) 8.1 (  This dramatic decrease of BC air in winter with larger v d and the lack of winter and spring Arctic haze is one of the major reasons for using low v d in previous studies Sharma et al., 2013;Liu et al., 2011). However, this does not justify the use of a low v d over snow and ice. First, observations have shown very large variations of v d (Table 3), which suggest that a uniform representation might involve large uncertainties. Second, observations of v d over snow and ice show very large values in certain regions, which is still underestimated by the resistance-in-series method. Third, besides dry deposition in boundary layer, BC air is affected by many other factors, such as emissions, transport, and wet deposition (Sect. 4.3).

WBF in mixed-phase clouds
Our model results show that WBF increases BC snow by 20-80 % in the eight Arctic sub-regions, except Canadian sub-Arctic, and increases BC air during the snow season by 25-70 % (Figs. 2 and 7). Inclusion of a parameterization of the WBF process in the model suppresses the scavenging of BC in mixed-phase clouds and consequently enhances poleward transport. We validate the simulation of WBF and the associated effects on global BC distribution in a companion study . The parameterized WBF process not only increases BC snow in the model Arctic but also changes the partition of dry and wet deposition of BC snow . Intuitively, WBF slows down wet scavenging, thus allowing for more BC particles available for dry deposition. Our model results show that the fraction of dry to total deposition increases from 26 % (12-41 %) to 35 % (19-59 %) on average in the eight Arctic subregions, thereby lowering the absorption of solar radiation due to less internally mixed BC-snow composite (Sect. 4.2). In Alaska, Canadian Arctic, and Russia, BC removed by dry deposition increases to more than 50 %. However, averaged globally, this fraction increases only slightly (from 19 to 20 %), indicating that the fraction in the Arctic is more sensitive to the WBF parameterization in our model.
The scavenging efficiency of BC, heretofore defined as the fraction of BC incorporated in cloud water drops or ice crystals in mixed-phase clouds, is strongly affected by the WBF parameterization and as a result varies temporally and spatially in response to varying temperature (Sect. 3.3). Thus, improved treatment of mixed-phase cloud processes, such as WBF and riming, is essential to improve the simulation of spatial and temporal distribution of BC. BC in Alaska and the Canadian Arctic are most sensitive to the WBF effect in the Arctic in our model. WBF increases BC snow by 55 % in Alaska and 43 % in the Canadian Arctic and reduces the model discrepancies to within 10 % (Table 4 and Fig. 3). BC air at Barrow in Alaska and at Alert in Canadian Arctic are higher by 20-30 ng m −3 in winter, reducing the model discrepancies significantly (from −54 to −18 % at Barrow and from −72 to −46 % at Alert) and enhancing the seasonal variation (Fig. 5). Similar improvements are also seen at Summit in Greenland, where BC air increases by 12 ng m −3 and the model discrepancy lowers significantly (from −48 to 3 %). This modeling result is consistent with recent observations, which showed that a high riming rate was rare (12 %) in the North American sector of the Arctic and that WBF dominated in-cloud scavenging in mixed-phase clouds .
At Zeppelin where snow samples show rimed structures (Hegg et al., 2011), model discrepancy of BC air increases to 63 from −10 % with the WBF effect included. Model results do not capture the magnitude of BC air in winter at Barrow, Alert, and Zeppelin (Fig. 5). BC air is well simulated at Zeppelin but underestimated at Barrow and Alert in experiment A. BC air is well simulated at Barrow and Alert but overestimated at Zeppelin in experiment D (Fig. 5) -similar results were shown in Sharma et al. (2013). Such apparent discrepancy can be partly attributed to the fact that models do not properly distinguish WBF-dominated in-cloud scavenging at Barrow  and riming-dominated scavenging at Zeppelin (Hegg et al., 2011). Here we separate WBF-and riming-dominated conditions based on temperature and LWC (Sect. 3.3; Fukuta and Takahashi, 1999) in experiment D. However, model results still fail to capture the difference among the three sites. There are a number of reasons. First, LWC from GEOS-5 biased high compared to CloudSat observations (Barahona et al., 2014). In addition, the spatial distribution of LWC from GEOS-5 also has a large discrepancy (Li et al., 2012;Barahona et al., 2014). Second, this separation is based on a laboratory experiment, while conditions in the real atmosphere are much more complex. Therefore, more field measurements are required to better separate the two conditions and better parameterize BC scavenging efficiency.
Our model results show that the WBF parameterization exaggerates the positive bias of BC air in summer and delays the transition from the late-spring haze to the clean summer boundary layer (experiment D). Previous studies found that the dominant process controlling low summertime aerosol at Barrow is the onset of local wet scavenging by warmer clouds (Garrett et al., 2010(Garrett et al., , 2011. The WBF parameterization has the effect of suppressing scavenging in mixed-phase clouds and thus slows down the onset of strong scavenging by warmer clouds during the transition from winter to summer. However, the strong scavenging of warm drizzling clouds in late spring and summer boundary layer (Browse et al., 2012), which enhances the winter-summer transition, is not considered in the present study. At high latitudes in summer, low stratocumulus cloud decks in the boundary and lower troposphere produce frequent drizzle (90 % of the time) and remove aerosol effectively (Browse et al., 2012).

Precipitation
We compute BC snow as the ratio of BC deposition flux to precipitation rate (Sect. 3.5). It has been pointed out that this estimate is very sensitive to uncertainties in precipitation (He et al., 2014a). Climatological precipitation across the Arctic is 14.3 g cm −2 yr −1 for 1965-1989(Overland and Turet, 1994 and is 16.3 g cm −2 yr −1 for 1971−1 for -1991−1 for (Serreze et al., 1995 as constrained from an observed hydrologic budget (Warren et al., 1999). The annual precipitation, averaged for 2007-2009, is 15.5 g cm −2 yr −1 in GEOS-5, within the range of the observations. There are considerable uncertainties, spatially and temporally, in precipitation in the Arctic (Warren et al., 1999;Serreze and Hurst, 2000). Figure 6 compares monthly precipitation from the Global Precipitation Climatology Project (GPCP; Huffman et al., 2001), NOAA Climate Prediction Center Merged Analysis of Precipitation (CMAP; Xie and Arkin, 1997), and GEOS-5. The discrepancies can be as large as a factor of 10 and the seasonal cycles are largely out of phase between the three data sets. Specifically, GPCP precipitation is much stronger than CMAP, particularly during summer. GEOS-5 precipitation is within the range of GPCP and CMAP data. The exception is Greenland, Ny-Ålesund, and Tromsø, where GEOS-5 precipitation is substantially (a factor of 2-10) larger than GPCP and CMAP data during the snow season. Snow precipitation in the Arctic is difficult to constrain for two reasons. First, accurate measurements of snowfall in the Arctic have proven nearly impossible, because snow gauges strongly under-catch snowfall (by 55-75 %) depending on the gauge type and wind condition (Liston and Sturm, 2004). Second, a more fundamental problem is that the sparse observational network in the Arctic is vastly inadequate to accurately estimate the monthly mean precipitation (Serreze and Hurst, 2000) -10-40 stations are required in 2.5 • grid cells (WCRP, 1997).
To probe the sensitivity of BC deposition and BC snow to precipitation, we conduct two additional model simulations, where we halve and double the precipitation rate in the Arctic, with other processes configured as in experiment D. We find that, in GEOS-5, during the snow season, nearly all precipitation is in the form of snow in the Arctic. Halving precipitation leads to increases in BC snow by 15-136 %, with the largest enhancements in Greenland (136 %) and Ny-Ålesund (92 %) (Fig. 7). With precipitation halved, it takes a longer accumulation time for a given snow depth, which results in larger dry deposition (up to 153 % increases). Therefore, the ratio of BC dry deposition to snow precipitation increases as well. On the other hand, the ratio of BC wet deposition to snow precipitation, determined mainly by in-cloud scaveng-ing of BC, remains largely unchanged. Overall, BC snow increases with halved precipitation. Doubled precipitation has the opposite effect. Indeed, BC snow decreases by 14-43 % in the eight Arctic sub-regions. In addition, dry deposition decreases by 35-62 % and the fraction of dry to total deposition decreases by 23-43 %. Although BC snow as computed here is sensitive to precipitation, the resulting medians of BC snow in the eight sub-regions are in agreement with observations within a factor of 2, except over Greenland (a factor of 5 too high) and Tromsø (a factor of 3 too high). Further analysis of the results at Greenland and Tromsø is in Sect. 4.5. The strong sensitivity of BC snow calls for a better constraining of precipitation in the Arctic.
In contrast, annual BC loading and deposition are much less sensitive to precipitation (Table 6). Halving Arctic precipitation increases annual BC loading by 12 % and de-  Samset et al. (2013). d Lifetime is approximated by dividing the annual Arctic BC column burden by the annual Arctic deposition flux. e BC snow concentrations were calculated using CLM4 and CICE4 models with monthly deposition field from AeroCom models (Jiao et al., 2014). f Paticipating models are DlR, GISS, LOA, LSCE, MATCH, MPI-HAM, TM5, UIO-CTM, UIO-GCM,UIO-GCM-V2, ULAQ, UMI, CAM-Oslo (Jiao et al., 2014). g This range is for the AeroCom Phase I models except for ULAQ, which is the only one to produce a positive bias of +10.7 ng g −1 . h The regression is significant at α = 0.05.  Figure 7. Same as Fig. 3, but for Exp. D with standard precipitation (red symbols), 50 % precipitation (green symbols), and 200 % precipitation (blue symbols). See text for details. creases annual BC deposition by 16 % in the Arctic. This is because less precipitation removes fewer BC particles. BC lifetime in the Arctic, as determined by the BC loading and deposition, increases by 27 %. When precipitation is doubled, annual BC loading decreases by 14 %, while BC deposition increases by 8 %, resulting in a 23 % reduction of BC lifetime in the Arctic. BC air is more sensitive to precipitation at Barrow, Alert, and Zeppelin than at Denali and Summit (Fig. 8). When precipitation is halved, annual BC air increases by 20-70 % at Alert, by 10-40 % at Barrow and Zeppelin, and by 1-20 % at Denali and Summit. When precipitation is doubled, annual BC air decreases by 20-50 % at Alert, by 10-40 % at Barrow and Zeppelin, and by 2-20 % at Denali and Summit. Additionally, BC air is more sensitive to precipitation in summer than in winter. This is because the summer clean boundary layer in the Arctic is controlled by strong local scavenging (Garrett et al., 2010(Garrett et al., , 2011Browse et al., 2012).

BC in snow in Greenland, Tromsø, and Canadian sub-Arctic
BC snow is associated with much larger uncertainties over short (hence shallower snow depth) than longer (hence larger snow depth) time periods. Because snow samples over Greenland were collected at the very surface (∼ 0 cm), the computed BC snow thus represents BC deposition only through the duration of a day for direct comparisons. The short time duration thus largely explains the larger uncertainties in the estimated BC snow . In Tromsø, observed BC snow were considerably lower (19.1 ng g −1 ) from samples collected over a clean mountain plateau upwind of the town Tromsø (Doherty et al., 2010) and much higher (  from samples collected in the town . We use the former for comparisons. Thus, the factor of 2 overestimate of BC snow in this region is because GEOS-Chem does not resolve subgrid variability. In the Canadian sub-Arctic, BC snow is underestimated by 50 % with all the improvements discussed above (experiment D). This large low bias is mainly from the low BC snow in the subsurface samples (1-20 cm, 11.7 ng g −1 , ∼ 60 % of all samples) accumulated through the snow season. BC snow in this region of the model increases by 33 % from flaring emissions and by 43 % from halving precipitation. Yet the resulting BC snow is still 25 % lower than observations (12.8 ng g −1 ). However, GEOS-5 precipitation is at the lower end among the three precipitation data sets (Fig. 6). The large discrepancy in BC snow warrants further studies.

Discussions
Global BC emissions in this study are within the range of previous studies, but emissions in the Arctic (0.115 Tg yr −1 ) exceed the higher end of those used in previous studies (0.037-0.077 Tg yr −1 , Table 6). The large Arctic emissions in this study result from gas flares, which have been missing in most previous estimates. It has been suggested that gas flares are a dominant BC source in the Arctic -it is 42 % of the total BC emissions in the Arctic, but a rather small fraction (3 %) of the global BC emissions (Stohl et al., 2013). Although this estimate is probably biased high because of the large emission factor (Sect. 4.1), including gas-flaring emissions in modeling Arctic BC appears to be justified from our results.
BC deposition in the Arctic (0.38 Tg yr −1 ) exceeds the higher end of those used in previous studies (0.13-0.34 Tg yr −1 ), with flaring emissions included (Table 6). Our model results suggest that annual BC deposition in the Arctic is more sensitive to the BC emissions and precipitation rate in the region than to v d and WBF. Flaring emissions increases BC deposition flux in the Arctic by 19 % in the model. In the model, doubling precipitation in the Arctic increases BC deposition by 8 %; halving precipitation decreases BC deposition by 18 %. Total modeled BC emissions in the Arctic are a factor of 2-5 lower than total modeled BC deposition, suggesting that a large fraction of BC deposited in the Arctic is from long-range transport.
Simulation of BC snow in this study is much better than most of the AeroCom models in the perspective of mean model bias across the Arctic (experiment D in this study: −0.8 ng g −1 ; AeroCom models: −13.2-+21.4 ng g −1 ; Table 6) and the biases for the eight sub-regions (experiment D in this study: a factor of 2; AeroCom estimates: a factor of 5-6; Jiao et al., 2014). In addition, the correlation coefficient of modeled and simulated BC snow in this study (0.21) is located at the higher end of previous AeroCom estimates (0.12-0.24). We find that flaring emissions improve the agreement of BC snow with observations significantly, with a 50 % reduction to the negative bias of modeled BC snow across the Arctic and a substantially stronger correlation (0.15 to 0.24) between simulated and observed BC snow in the region (Table 6). WBF further reduces the average bias across the Arctic by 70 %. Overall, modeled BC snow is poorly correlated with observations (r = 0.15 to 0.24) for all AeroCom models and GEOS-Chem. This disagreement is probably resulted from a common problem in the Arctic, which is the poorly constrained meteorological fields including precipitation in the Arctic due to the scarcity of observations in the region (Sect. 4.4). Our model results show that doubling precipitation introduces a much larger positive bias, similar to the magnitude of the overall effects of flaring emissions and the WBF effect; halving precipitation produces a similarly large negative bias (Sect. 4.4).
Modeled atmospheric BC loading in the Arctic in this study exceeds the high end of the previous AeroCom estimates (0.02-0.34 mg m −2 ) by including the WBF effect (Table 6). We find that BC scavenging efficiency plays a more important role in determining BC loading in the Arctic than emissions, v d , and precipitation. BC loading in this region increases by 13 % from flaring emissions, which represents a ∼ 70 % enhancement to previous emission estimates, and by 7 % from updated dry deposition velocity, which is, in some cases, a factor of 2-3 larger than the default value of 0.03 cm s −1 . In addition, Arctic BC loading in the atmosphere increases by 12 % when precipitation is halved and decreases by 14 % when precipitation is doubled. Modeled WBF reduces BC scavenging efficiency in mixed-phase clouds by 20-80 % and increases annual BC loading by 70 % in the Arctic. This large sensitivity of BC loading in the Arctic to treatments of BC scavenging efficiency in mixed-phase clouds and in ice clouds is also shown by previous studies. For example, Bourgeois and Bey (2011) reduced the scavenging efficiency in mixed-phase clouds from 0.10-0.75 to a uniform value of 0.06 in the ECHAM5-HAMMOZ model (Pozzoli et al., 2008) and found that the resulting BC air in the Arctic increased by up to a factor of 10 and were in improved agreement with aircraft observations. In addition, their model results of BC burden in the Arctic were 5 times higher. We note here that a scavenging efficiency of 0.06 is on the low end of observed values in mixed-phase clouds Verheggen et al., 2007), which leads to a considerably larger WBF effect. Liu et al. (2011) found that lowering BC scavenging efficiency in ice clouds (from 0.2 to 0.01) in the AM3 model (Anderson et al., 2004) dramatically enhanced BC transport to the Arctic (nearly 10 times higher) and improved model comparison with aircraft observations. Browse et al. (2012) suppressed the scavenging of soluble BC in ice clouds in the GLOMAP model (Mann et al., 2010) and found that the resulting BC air in the Arctic were 6 times higher. Better characterization of scavenging efficiency in all could types globally is thus critical for accurately reproducing BC distribution and the associated climatic effects in the Arctic.

Summary and conclusions
This study sought to understand the capability of GEOS-Chem in simulating BC distribution both in air and in snow in the Arctic and the controlling factors. We evaluated the model simulation against BC snow measurements across the Arctic and in situ measurements of surface BC air at Denali in the low Arctic, Barrow, Alert, and Zeppelin in the high Arctic, and Summit in the free troposphere. We also examined the role of gas-flaring emissions, v d , the WBF effect, and precipitation on BC distribution in the Arctic. We first included BC emissions from a missing source in the current emission inventories: natural gas flares. We then used the resistancein-series method to estimate v d of BC over snow and ice to replace the uniform constant v d of 0.03 cm s −1 over snow and ice. We also parameterized the effects of the WBF process on BC scavenging efficiency in mixed-phase clouds. WBF was stronger at lower temperature.
With all these changes, the discrepancy of BC snow across the whole Arctic decreased substantially (from −40 to −10 %). In the eight sub-regions, the simulated BC snow agreed with observations within a factor of 2. We also found that including flaring emissions significantly improves the simulation of BC snow with a strong reduction of discrepancy (from −40 to −20 %) and an increase of correlation coefficient with observations (from 0.15 to 0.24). WBF further reduced the discrepancy of BC snow to within −10 %, with the largest improvement in the North American section in the Arctic. Simulation of BC snow with the abovementioned improvements was among the best AeroCom models evaluated by Jiao et al. (2014). The resulting BC air agreed with observations within a factor of 2, also among the best simulations in Eckhardt et al. (2015).
In addition to these physical processes, we also tested the sensitivity of BC snow to precipitation in the Arctic, which is poorly constrained due to the sparse observation network. The difference of precipitation rate in the region among GEOS-5, GPCP, and CMAP was up to a factor of 10. Our model results suggested that the negative bias introduced by doubling the precipitation rate in the Arctic and the positive bias introduced by halving the precipitation rate was similar to the combined effects of flaring emissions and WBF. Although this effect (exaggerated because our method of estimating BC snow ) strongly depends on precipitation flux, it is worthwhile to notice the importance of precipitation on BC snow simulation.
There remains large uncertainties in flaring emission factors, spatial and temporal variation of flaring emissions, dry deposition velocities of BC, and BC scavenging efficiencies in clouds. Process-specific measurements, particularly in the Arctic, are useful to better constrain the simulation of BC distribution in the region. For example, we need direct measurements of emission factors of gas flares on the western side of the extreme north of Russia, including their spatial and temporal variations. In addition, v d measurements specific to BC particles over snow-and ice-covered land surfaces should be made in winter. Measurements of BC scavenging efficiency in clouds, particularly in mixed-phase and ice clouds in the Arctic, are also needed to constrain BC wet deposition.

Data availability
The data used in this study are available from the corresponding author upon request (qiling@atmos.ucla.edu).