We used the LMDz-OR-INCA global chemistry–aerosol–climate model, which couples the LMDz (Laboratoire de Météorologie Dynamique) GCM (Hourdin et al., 2006) and the INCA (INteraction with Chemistry and Aerosols) model (Hauglustaine et al., 2004). The interaction between the atmosphere and the land surface was ensured through the coupling of LMDz with the ORCHIDEE (ORganizing Carbon and Hydrology In Dynamic Ecosystems) dynamical vegetation model (Krinner et al., 2005). In the present configuration, the model included 39 hybrid vertical levels extending to the stratosphere and a horizontal resolution of 1.29∘ × 0.94∘ (280 grid cells in longitude, 192 in latitude). A more detailed description and an extended evaluation of the GCM can be found in Hourdin et al. (2006). The large-scale advection of tracers was calculated based on a monotonic finite-volume second-order scheme (Hourdin and Armengaud, 1999). Deep convection was parameterized according to the scheme of Emanuel (1991). The turbulent mixing in the planetary boundary layer (PBL) was based on a local second-order closure formalism.

A comparison made with inert tracers indicated an enhanced vertical transport as the horizontal resolution of the model was increased from 144 × 142 grid points to 280 × 192. We have studied the effect of the model resolution on the robustness of the predicted BC concentration distribution in Wang et al. (2014). The higher model resolution and an updated emission inventory used in this work allowed for substantially improving the correlation between predicted and measured BC concentrations. The vertical profiles of BC over the Arctic measured during the ARCTAS campaigns (Arctic Research of the Composition of the Troposphere from Aircraft and Satellites) were used to assess the ability of such a model to represent the vertical distribution of BC. The comparison showed that BC loads over the Arctic were slightly underestimated for the time of the ARCTAS flights. The model was also compared to other models in the recent AeroCom Phase II intercomparison (Myhre et al., 2013). The global mean anthropogenic all-sky and clear-sky aerosol radiative forcing is slightly more negative than the median for all the models (-0.36 vs. -0.27 W m-2) and the BC load from fossil fuel and biofuel emissions is very close to the corresponding median of the models (0.15 vs. 0.14 mg m-2).

The INCA model simulates the distribution of anthropogenic aerosols such as sulfates, nitrate (NO3), BC, and particulate organic matter (POM), as well as natural aerosols such as sea salt and dust. The aerosol model keeps track of both the number and the mass of aerosols using a modal approach to treat the size distribution, which is described by a superposition of 5 log-normal modes (Schulz, 2007), each with a fixed spread. To treat the optically relevant aerosol size diversity, particle modes were categorized in three ranges: submicronic (diameter < 1 µm), corresponding to the accumulation mode; micronic (diameter 1–10 µm), corresponding to coarse particles; and super-micronic or super coarse particles (diameter > 10 µm). Compared to a bin-scheme, the treatment of the size distribution with modes was computationally much more efficient (Schulz et al., 1998). Furthermore, to account for the diversity in chemical composition, hygroscopicity, and mixing state, we distinguished between soluble and insoluble modes. In both submicron and micron size ranges, soluble and insoluble aerosols were treated separately. Sea salt, SO4, and methane sulfonic acid (MSA) were treated as soluble components of the aerosol, dust was treated as insoluble, whereas nitrate, BC, and POM appeared both in the soluble and insoluble fractions. The aging of primary insoluble carbonaceous particles transfers insoluble aerosol number and mass to soluble ones with a half-life of 1.1 days (Chung and Seinfeld, 2002). The deposition scheme used in the model is described in detail in Evangeliou et al. (2013).

The simulations lasted from 1 January 2002 to 31 December 2013. The model ran in a nudged mode using 6-hourly ERA Interim Re-analysis data (ECMWF, 2014) with a relaxation time of 10 days (Hourdin and Issartel, 2000).

To support the IPCC-AR5 (Intergovernmental Panel for Climate Change Assessment Report 5) and the ACCMIP (Atmospheric Chemistry and Climate Model Intercomparison Project), a historical emissions dataset was developed (Lamarque et al., 2010) on a decadal basis (from 1850 to 2000 for the historical dataset), as well as RCP (Representation Concentration Pathway) emission scenarios for the period after the year 2000. As part of a project funded by the European Commission, MACC (Monitoring Atmospheric Composition and Climate) and CityZen (megacity Zoom for the Environment), the ACCMIP and the RCP emissions datasets were adapted and extended to the year 2013 on a yearly basis. For anthropogenic emissions, emission data were interpolated on a yearly basis. For BB emissions, the ACCMIP dataset was extended to yearly and monthly mean calculated from modified RETRO (REanalysis of the TROposhperic chemical composition) BC emission data for the years 1980 to 1996, and from GFEDv3 carbon emission data for the years 1997 to 2013 by applying a single set of vegetation-type-specific emission factors, and the predominant vegetation map used in the GFEDv3 inventory. This extension of the ACCMIP and RCP emission dataset for the MACC and CityZEN projects is referred to as MACCity (MACC/CityZen) emission dataset (Granier et al., 2011). As it is explained below, emissions of MACCity (anthropogenic and BB) were used as the input source of the model worldwide, in which BB sources were derived from GFEDv3. In addition, we adopted BC emissions from BB in northern Eurasia from 2002 to 2013 (FEI-NE) described in the companion paper (Hao et al., 2016), while MACCity emissions were used for all the sources in other regions and for anthropogenic sources in northern Eurasia.

Injection height is a key factor that controls transport and in turn deposition of BC emitted from fires. It is generally accepted that only explosive volcanic eruptions and strong crown fires (more common in North America than in northern Eurasia) have the energy to inject pollutants from the surface into the stratosphere (Jost et al., 2004; Fromm et al., 2005). In a modelling study of midlatitude supercell thunderstorms (Wang, 2003), it was reported that these plumes could induce important transport into the lowermost stratosphere. These findings suggest that extreme convection, even unassociated with energetic forest fires, may represent an important pathway for rapid, efficient redistribution of gases and particles from the lowest levels of the atmosphere to the lower stratosphere or, more commonly, to the upper troposphere. Nedelec et al. (2005) described such a case for a fire happening over Siberia. However, injection of emissions in the lower troposphere is more common. Recently, Sofiev et al. (2013) published global maps of emission heights of wildfires that occurred between 2000 and 2012, reporting that about 80 % of the smoke is generally injected within the PBL, while the rest is injected at higher altitudes. Here we follow the same pattern as Sofiev et al. (2013) for northern Eurasia, where 90 % of the emissions were injected below 1.1 km, with the rest in heights up to 1.5 km. Outside Eurasia, a similar injection profile is followed with 100 % of the emissions occurring up to 1 km.

For the 12-year global simulations, anthropogenic sources of BC were adopted from the MACCity emission database. With regards to BB emissions, MACCity BB emissions from GFED3 were applied for all the regions outside northern Eurasia, while within northern Eurasia anthropogenic emissions from MACCity and BB from FEI-NE (Hao et al., 2016) were adopted. In summary, BC emissions from BB in northern Eurasia, excluding agricultural fires, were estimated based on the area burned, fuel loading, percentage of the fuel burned, and emission factors of BC from different vegetation types (Hao et al., 2016).

This combined simulation is hereafter referred to as FEI-NE+MACCity. For comparison, we carried out the same simulation but using MACCity emissions alone (i.e. BB emissions in northern Eurasia were also taken from MACCity) for the same period (2002–2013) (from now on referred to it as MACCity simulation). The different simulations are shown in Table 1.

Several definitions for atmospheric lifetime exist. In any domain of the Earth's atmosphere the mass balance can be expressed as follows: dB(t)dt=St-B(t)τ(t), where B(t) is the atmospheric burden, St is the mass entering or exiting the domain and τ(t) is the removal time over a given time step. If one assumes equilibrium between St and deposition (steady state conditions), the mean steady state lifetime will be the following: τss=B‾D‾, where B‾ and D‾ are the mean atmospheric burden and deposition over a specific period (Croft et al., 2014).

With different types of instruments, we collected measurements of BC, which may not always be directly comparable. Following the nomenclature of Petzold et al. (2013), we referred to measurements based on light absorption as equivalent BC (eBC) and measurements based on thermal-optical methods as elemental carbon (EC).

Aerosol light absorption data were obtained from five sites in different parts of the Arctic: Alert, Canada (62.3∘ W, 82.5∘ N; 210 m a.s.l.), Barrow, Alaska (156.6∘ W, 71.3∘ N; 11 m a.s.l.), Zeppelin/Ny Ålesund, Spitsbergen, Norway (11.9∘ E, 78.9∘ N; 478 m a.s.l.), and Tiksi, Russia (128.9∘ E, 15 71.6∘ N; 1 m a.s.l.). Different types of particle soot absorption photometers (PSAPs) were used for the measurements at Barrow and Zeppelin, and an aethalometer was used at Alert and Tiksi. All these instruments measured the particle light absorption coefficient σap, each at their own specific wavelength (typically at around 530–550 nm), and for different size fractions of the aerosol (typically particles smaller than 1, 2.5, or 10 µm are sampled at different humidities). Conversion of σap to eBC mass concentrations is not straightforward and requires certain assumptions (Petzold et al., 2013). The mass absorption efficiency used for conversion can be specific to a site and is uncertain by at least a factor of two. For Tiksi, the conversion was done internally by the aethalometer. For the other sites, a mass absorption efficiency of 10 m2 g-1, typical of aged BC aerosol (Bond and Bergstrom, 2006), was used. Sharma et al. (2013) used an even higher value of 19 m2 g-1 for Barrow and 10 m2 g-1 for Alert data.

At Villum Research Station, Station Nord, Greenland, thermal-optical measurements were available. Weekly aerosol samples were analysed with a thermal-optical Lab OC / EC instrument from Sunset Laboratory Inc (Tigard, OR, USA). Punches of 2.5 cm2 were cut from the filters sampled at Villum and analysed according to the EUSAAR-2 protocol (Cavalli et al., 2010).

At Alert, eBC data were available for the years 2006–2013, at Barrow for 2002–2013, at Tiksi for 2009–2013, and at Zeppelin for 2002–2013. At Villum station, EC data were available for 2008–2011. The EC and eBC data were directly compared with modelled BC concentrations for the same locations and periods. Tiksi data were not filtered for a clean air sector and may have been affected by local pollution events. Barrow and Alert data were routinely subject to data cleaning, which removed the influence from local sources. Zeppelin generally was not strongly influenced by local emissions; however, summer values were enhanced by some 11 % due to local cruise ship emissions (Eckhardt et al., 2013).

Northern Eurasia encompasses diverse ecosystems including forest, shrubland, cropland, grassland, and savanna (Friedl et al., 2002). The total burned areas (excluding agricultural fires) during the period of 2002–2013 were estimated to be 250 000 km2 yr-1 (n=12) (Hao et al., 2016), which consisted of 61 % of grassland and 27 % of forest. Grassland fires occurred predominantly over central and western Asia, and forest fires over Siberia in Russia. The years 2003, 2006, and 2008 showed respectively 96, 40, and 30 % more burn scars than the annual mean from 2002 to 2013 (Fig. S1 in the Supplement). The unusual high fire activity in 2003, 2006, and 2008 was a result of extensive grassland fires over central and western Asia, and forest and grassland fires over Russia.

Table 2 presents the yearly mean atmospheric emissions of BC from anthropogenic and BB sources for the period 2002–2013 from FEI-NE+MACCity and MACCity. Annual global BC emissions varied in a range from 8.02 Tg yr-1 in 2007 to 9.48 Tg yr-1 in 2003 with an average: 8.42 ± 0.43 Tg yr-1 over the period 2002 to 2013 according to FEI-NE+MACCity. These values compared well with other published results. For instance, Wang et al. (2014) reported that, according to the PKU-BC-2007 (Peking University BC Inventory for 2007) inventory of global BC emissions (both anthropogenic and biomass burning sources), a total amount of 8.9 Tg yr-1 was emitted in 2007. The ECLIPSE inventory estimated for 2010 by Eckhardt et al. (2015) was 8.32 Tg yr-1, only slightly higher than our estimations of 8.02 Tg yr-1. In contrast, the ACCMIP BC emissions for 2005 (Lamarque et al., 2010) was 15 % lower (e.g. ∼ 7.82 Tg yr-1 compared to 8.13 Tg yr-1). BC emissions from vegetation fires in northern Eurasia ranged between 0.45 and 2.19 Tg yr-1 for the period 2002–2013, representing 20 % to 49 % of the total BC emissions over this region. In comparison, the emissions over the same region based on the MACCity inventory ranged between 0.13 and 0.43 Tg yr-1 and accounted for 5 to 16 % of total BC emissions. The BC emissions in the Northern Hemisphere (FEI-NE+MACCity) were classified as 70 % from anthropogenic sources and 30 % from BB. Northern Eurasian vegetation fires accounted for 56 % of the BB BC emissions in the Northern Hemisphere.

BC emissions from FEI-NE were relatively constant in time except for four (2003, 2006, 2008, 2012) intense years (Fig. S2). Similarly, the GFED3 of MACCity showed relatively high emissions in some of these years (2003 and 2006), but the relative emission increase was significantly lower for these years (Table 2). The fire episodes that occurred in northern Eurasia during these extreme fire years were particularly intense. In 2003, fire events in the Transbaikal region (Russian provinces Chita and Buryatia) caused severe smoke pollution in the Far East of the Russian Federation (IFFN, 2004), while in spring 2006, smoke from peat and forest fires in the western Russian Federation was noticed as far as the United Kingdom (Hao et al., 2009) and in the Arctic (Stohl et al., 2007). In summer 2006, smoke from vegetation fires in the Russian Federation persisted for weeks over Finland (GFMC, 2006).

Figure 1 depicts deposition anomalies of BC for the period 2002–2013. A remarkable feature was that the highest anomalies were observed in the northernmost part of Asia and in the Arctic in 2003, 2006, 2008, and 2012. Taking into account emissions fluxes from these years (Fig. S2), one can note that the largest amounts of BC were deposited over Arctic regions as a result of large fire events in Siberia, western Russia, and Kazakhstan. The annual amount of BC deposited over the Arctic from all possible global sources (including BB) during the 12-year period ranged from 65 to 152 kt yr-1 (average: 102 ± 29 kt yr-1) representing about 0.9–18 % of total global emissions. The total annual deposition of BC to the Arctic from vegetation fires in northern Eurasia (FEI-NE) during the same period ranged between 29 and 120 kt yr-1 (average: 65 ± 28 kt yr-1) or about 0.3–14 % of total global BB emissions (Table 2). Hence, more than half of the total BC that deposited over the Arctic originated from BB in northern Eurasia, which underlines the importance of the northern Eurasian vegetation fires on the Arctic BC budget compared with contributions of other sources.

Figure 2 illustrates BC deposition from FEI-NE detected vegetation fires only, excluding anthropogenic sources, while Fig. S3 shows the deposition from all sources (FEI-NE+MACCity). For the most intense fire years, 2003, 2006, 2008, and 2012, annual amounts of 142, 129, 117, and 152 kt yr-1 of BC were respectively deposited over the Arctic from northern Eurasian vegetation fires (Table 2). They amount to 3.2 to 4.1 times the 37 ± 5 kt yr-1 average (2002 to 2013) deposition flux for anthropogenic sources over northern Eurasia. Finally, when using MACCity emissions, both from anthropogenic and BB sources, the estimated average deposition of BC over the Arctic was 56 ± 8 kt yr-1 for the period 2002–2013. Consequently, Arctic deposition was lower by 45 % compared to FEI-NE, when BB emissions of BC from GFED (MACCity) were used.

Figure 3 depicts the global mean aerosol lifetimes for anthropogenic and BB BC from the combined FEI-NE+MACCity simulation. The results are given in box and whisker plots of daily lifetimes for the 12-year period. Mean aerosol lifetimes for anthropogenic BC were stable for all the studied years with an annual average value of 5.6 ± 0.2 d. Mean aerosol lifetimes of BC from FEI-NE+MACCity simulation exceeded these lifetimes by 1.2 days (6.8 ± 1.0 d). The pronounced variability of the mean lifetime of BB BC was attributed to the variability in regions and injection height. According to the injection scheme used (Sofiev et al., 2013), continuous injection close to the PBL (around 65 % of the mass of BC is emitted up to 0.8 km, 90 % up to 1.1 km) results in accumulation of BC in the troposphere, where longer lifetime occurs compared to the PBL. This was likely the main reason for transport of BC into the Arctic. For any soluble species emitted in the PBL, the lifetime is controlled by the removal, which happens within a few days if it is not transported to the free troposphere. Two weeks after being injected into the atmosphere, most of the tropospheric aerosol was scavenged through wet deposition, whereas aerosols that have been transported into the high troposphere/lower stratosphere persist, given the absence of wet scavenging at these heights. The total aerosol mass and the lifetimes are then dominated by the stratospheric loading (Cassiani et al., 2013).

Mean aerosol lifetimes from global models are typically in the range of 3–7 days (Benkovitz et al., 2004; Textor et al., 2006), very similar to our estimations for BC, but more variable in this study due to the higher injection heights of BC in smoke plumes. As stated, in the present case of wildfires in northern Eurasia, the lifetimes and the behaviour of BC was strongly affected by the fact that it was directly emitted aloft. Although 80–90 % of BC was emitted inside the PBL, nearly 65 % of the BC was emitted near the PBL height, while 10 % was injected above (according to Sofiev et al., 2013).

In this study, we analysed monthly values of both the Arctic BC burden and the mass deposited in the Arctic for all BC sources (FEI-NE+MACCity) and for BC produced from BB sources only over northern Eurasia (FEI-NE). A strong seasonal variation can be seen in both Arctic burden and deposition (Fig. 4a, b, c, and d). Relatively high BC burden from the combined emissions of the FEI-NE+MACCity run occurred in winter (December–January, Fig. 4b), while two peaks were found in late spring and in summer for intense fire years. The latest were clearly caused by spring and summer fire events over northern Eurasia (Fig. 4). The higher winter values corresponded to the meridional transport of BC, mostly emitted from anthropogenic sources (Fig. 4a). These large increases in the burden were not proportional among the different aerosol modes (hydrophilic and hydrophobic) and affected mostly hydrophilic BC aerosol from vegetation fires. This indicated that wet scavenging was less efficient than the rest of the year along the transport path between mid- and high-latitudes. A spring maximum was also simulated for BC deposition over the Arctic. It was caused by a combination of anthropogenic BC in the Northern Hemisphere and fires in northern Eurasia (Fig. 4a and c), the majority of which occurred between March and May. The annually deposited mass of BC over the Arctic exceeded the annual average for the period 2002–2013 by 52 % in 2003 and 86 % in 2012.

There were three noteworthy findings. First, the year 2012 appeared to be a year when BC transport to the Arctic was particularly efficient. Although the year itself (2012) was not an extreme fire year (Fig. S1) in terms of burn scars, it appeared that the prevailing winds and the lack of scavenging in mid-latitudes during June and July, when fires were most intense, favoured transport and subsequent deposition to high-latitudes (120 kt from vegetation fires only and 152 kt from all possible sources). Second, we simulated a higher relative contribution of wet to total deposition in the Arctic (90 %) than at mid-latitudes (69 %). Third, the annual mean lifetime of anthropogenic BC particles from BB was longer (6.8 d) than for BC from combustion (5.6 d). These values are within the range of other published results (e.g. 5.8 days from Park et al., 2005 and 7.3 days from Koch and Hansen, 2005).

In this section, we compare the contributions from different source regions and from different emission types to the deposition of BC in the Arctic region. Several simulations with BC tracers tagged by source region were carried out to isolate the different contributions. We selected the following regions (Fig. 5a): Europe, Asia, Siberia, Kazakhstan, and Mongolia, as well as from the latitude bands 35 to 40∘ N, 40 to 50∘ N, 50 to 60∘ N, above 60∘ N (60–90∘ N), and also the entirety of northern Eurasia. We discuss the results of these separate simulations and compare them to our FEI-NE+MACCity run (Fig. 5). Each simulation covered the whole period from 2002 to 2013.

Quinn et al. (2007) were the first to report that for BC over the Arctic, anthropogenic sources dominated during winter and early spring Arctic haze conditions. We estimated that anthropogenic emissions accounted for 70 % of the Arctic BC burden from all sources in the Northern Hemisphere during winter and fall and for 40 % during spring and summer when vegetation fires are most frequent.

Vegetation fires from northern Eurasia contributed 68 % of the annual BC deposition over the Arctic coming from that same region (Fig. 5d), whereas anthropogenic emissions constituted a lesser share (32 %). Northern Eurasian vegetation fires were the most numerous over the Northern Hemisphere; they contributed 81 % of the Arctic deposition of BC from Northern Hemispheric fires (Fig. 5d), while the rest came from other sources (e.g. BB in North America).

Of the northern Eurasian BB BC deposition in the Arctic, 95 % was from Asia, while only 5 % came from Europe. On a more regional basis, Siberia contributed 46 % of the northern Eurasian BB BC deposited in the Arctic, whereas Kazakhstan contributed 6 %, and Mongolia only 1 %. The rest was shared between fire events in Europe (5 %) and elsewhere in Asia (42 %) from areas that were not masked.

The relative contributions of fires at different latitudes to the Arctic BB BC deposition were distributed as follows: fires from the 35 to 40∘ N latitudinal band over northern Eurasia contributed only 7 %, from 40 to 50∘ N the contribution was 21 %, 40 % came from fires at 50–60∘ N, and 32 % from fires above 60∘ N (Fig. 5c).

Moreover, we examined the vertical distribution of BC over the Arctic for the different source regions (Fig. 6). When emitted from Europe, BC over the Arctic was found mostly below 5 km (either in the PBL or the low free troposphere), while BC emitted from Asia was found in higher layers that extended up to the mid- to high- free troposphere (Fig. 6). Similar vertical distribution for aerosols have been reported by Stohl et al. (2002), who estimated that aerosol originating from Asia was mixed throughout the entire troposphere within a few days.

These findings can be discussed in light of the ones reported by other authors. For instance, our results agree well with those of Hirdman et al. (2010a), who reported that the northern Eurasian region (Europe and Russia) was the main contributor to the Arctic surface concentrations of BC and sulfate. The present study shows the importance of all the regions north of 50∘ N. Stohl (2006) reported that Asia contributed 10 times less than Europe to the Arctic BC surface concentrations, which was not supported by our findings for total BC. Our results agreed with the conclusions of Koch and Hansen (2005), who reported that Europe contributed 10 % to the Arctic deposition of BC (anthropogenic and BB). In any case, all surface measurements of BC over the Arctic indicated that the main contributors to the Arctic BC during the summer were high-latitude sources (Hirdman et al., 2010a; Sharma et al., 2013).

There are contradictory results in the literature about where geographically the Arctic BC pollution originates. For example, Koch and Hansen (2005) estimated that south-east Asia, Europe, and Russia each contributed about 20–25 % of the Arctic pollution during the period January to March, whereas Stohl (2006) reported that Europe was the main contributor for the Arctic BC concentration at the surface. In addition, Huang et al. (2010) stated that Russia contributed 67 % to the surface Arctic pollution, whereas Europe and North America were 18 and 15 % respectively (16-year observations from Alert). Northern Eurasia appeared to be the main contributor in terms of Arctic BC during most seasons (Hirdman et al., 2010a). More recently, Stohl et al. (2007) and Warneke et al. (2009) reported that boreal and agricultural fires in eastern Europe and in Siberia might be strong contributors to Arctic BC, especially in the spring. Furthermore, Stohl et al. (2013) highlighted gas-flaring emissions in high latitudes as a major contributor to the Arctic BC.

Figure 8 compares the simulated surface BC concentrations from this study with in situ eBC and EC measurements from monitoring stations at Alert (Canada), Barrow (Alaska, USA), Villum (Greenland, Denmark), Zeppelin (Ny-Ålesund, Svalbard, Norway), and Tiksi (Russian Federation). The observations, when available, were represented for the entire period of our simulations (2002–2013). At Zeppelin and Barrow stations, the measurements were available between 2002 and 2013, at Alert from 2005 to 2013, at Villum from 2008 to 2011, and at Tiksi station from 2009 to 2013. Figure 7 compares the simulated vs. observed daily surface concentrations by a box and whisker plot at the five Arctic stations (Alert, Barrow, Villum, Tiksi, Zeppelin) for the period 2002 to 2013.

At Alert, there was a systematic underestimation in winter and early spring and an overestimation in late spring and summer, when the station did not record any enhanced eBC concentrations that would be attributed to vegetation fires in northern Eurasia (Figs. 7 and S2). At Barrow, the model not only accurately estimated surface concentrations during winter, when the Arctic haze is important, but also in spring and summer, when vegetation fires persisted (Figs. 7 and S2). At Villum, the measured seasonality was reproduced quite well by the model and individual episodes of elevated BC surface concentrations in spring and summer were also captured and attributed to vegetation fires. At the Tiksi station, the comparison showed a notable deviation from the measurements. Although some peaks in spring and summer were captured, the model completely missed the high concentrations of eBC in winter and early spring. Finally, at Zeppelin, it appeared that the anthropogenic BC contribution was slightly underestimated (spring), while vegetation fires in northern Eurasia elevated modelled surface concentration of BC.

Looking at other inventories and results from the Lagrangian particle dispersion model FLEXPART, we are convinced that local anthropogenic sources play an important role in the apparition of these peaks (Eckhardt et al., 2015). It was already noted that our model underestimated surface concentrations at Alert and Villum stations during the Arctic haze period. This is likely the consequence of an underestimation of Arctic transport in the model or a misleading emission inventory used for anthropogenic sources (MACCity). To verify it, we estimated surface concentrations of BC in the same longitude as the five Arctic stations, but five and ten degrees south in latitude (Fig. S4). In all stations except Tiksi, surface concentrations increased to the south confirming that the underestimation by our model over the Arctic could be attributed to a too-weak transport simulated towards the Arctic.

It has been reported that most models underestimated BC in the Arctic during winter and early spring (Eckhardt et al., 2015), likely due to an improper representation of the scavenging processes (lack of below-cloud scavenging for solid phase water), the different emission profiles used for BC, and underestimated emission inventories used as input to the models (e.g. Koch and Hansen, 2005; Liu et al., 2011; Jiao et al., 2014). Using the FEI-NE+MACCity inventory, BC concentrations over the Arctic are reasonable compared to observations, with a tendency to underestimate winter concentrations but capturing some summer peaks. Figure 7 also shows that using MACCity emissions globally, spring and summer concentrations are rather underestimated (e.g. in Villum, Tiksi, and Zeppelin). In both cases, misleading anthropogenic emissions appear to be the main problem during winter. The largest deviations occur at the station Tiksi, Russia. The station is located only a few kilometres away from Tiksi town and local pollution is likely to affect the measurements, as the town has both a small airport and a harbour. Despite these drawbacks at Tiksi station, the data from the station has been used to estimate the sources in northern Eurasia (Cheng, 2014). Eckhardt et al. (2015) compared both surface- and aircraft measurements of sulfate and BC in the Arctic to model output from eleven different models. They found that the models generally underestimated the surface concentrations of BC and sulfate in winter and spring, whereas concentrations in summer were overestimated. They also found a strong correlation between surface measured sulfate and BC concentrations in winter and spring (anthropogenic impact), which indicated that the sources contributing to sulfate and BC were similar throughout the Arctic and that the aerosols were internally mixed and undergo similar removal. Neither Eckhardt et al. (2015) nor Samset et al. (2014) could isolate the reason to explain why some models performed better than the others.

In the present model configuration, we included the emission inventory from FEI-NE and MACCity's anthropogenic BC inside northern Eurasia and MACCity (BB and anthropogenic) outside northern Eurasia (FEI-NE+MACCity) respectively to evaluate if it produced reliable results with respect to observations. The question that stemmed from this comparison was whether or not existing datasets included all possible sources of BC emission. This was examined by comparing the FEI-NE+MACCity inventory with MACCity, which included lower BC global emissions. Figure 8 depicts the difference of the average atmospheric burden of BC between FEI-NE+MACCity and MACCity runs, while Figs. S5 and S6 show the same comparison for emissions and Arctic deposition of BC. It is apparent that the difference in average atmospheric burden between the FEI-NE+MACCity and MACCity simulations (Fig. 8) is positive over northern Eurasia and over the Arctic, showing that vegetation fires in northern Eurasia have a direct impact on the Arctic budget, especially during the most intense fire years (2003, 2006, 2008, and 2012). The aforementioned impact extends up to North America and may affect the BC concentrations there as well. Subsequently, the deviation of the deposition of BC from northern Eurasian vegetation fires relative to FEI-NE is shown to be large over the Arctic (Fig. S6).

We also analysed the influence of anthropogenic emissions, as well as BB emissions from the regions (defined in Table 1) to the average surface concentration of the Arctic stations (Fig. 9). As expected, the predominant contributor to the surface concentrations of the Arctic stations was Northern Hemisphere anthropogenic emissions (29–55 %) (e.g. Shaw et al., 2010). The explanation is twofold. On one hand, transport of BC from the southern latitudes to the Arctic takes place as the air masses follow the trajectories of potential difference in temperature (a dome effect); they are lifted up, leaving the surface, especially from North America and Asia (China). On the other hand, transport from Russia/Siberia during the winter and spring is closer to the surface due to large anthropogenic emissions that are also effectively transferred from Europe. In addition, transport of BC from Russia/Siberia is less efficient during summer due to pressure systems that block the BC transport. Fires from northern Eurasia contributed less BC to Barrow, Zeppelin, and Villum, while this pattern changed for Alert and Tiksi stations (Fig. 9). This shows that emissions from northern Eurasia may extend up to the American Arctic (Barrie, 1986). The region marked as “other” in Fig. 9 stands for all BB emissions occurring over the Northern Hemisphere, excluding northern Eurasia, and shows that 4–42 % of the surface concentrations may be due to fires in North America and other BB sources. In all cases, fires in regions north of 50∘ N contributed the most, especially to Tiksi station. Fires in the Asian part of northern Eurasia contributed to the surface concentrations of the stations by 13 to 57 %, with the maximum at Tiksi station. This was expected since this station is in the middle of the northern Eurasian Arctic region and receives a lot of BC emitted from BB in Siberia. The respective portion for fires occurring in Europe was estimated to be only 1–2 %. Similarly, BC emitted from Siberia contributed 9–43 % to the simulated surface concentrations at the stations (with a maximum in Tiksi station). We estimated that the total of all vegetation fires in northern Eurasia contributed around 56 ng m-3, on average, to the Alert station (during spring and summer months), which is close to the 89 ng m-3 that Huang et al. (2010) estimated for the USSR and European Union, although with different geographic definitions than those assumed here, and also including anthropogenic emissions. Gong et al. (2010) and Shindell et al. (2008) estimated that the northern Eurasian contribution to Alert varied between 80–90 %, while our runs suggested a lower BB contribution (29 %). Nevertheless, considering our comparison of modelling results with surface observations for BC from the five Arctic stations, it should be noted that the calculated contribution constituted an upper bound, especially when considering Asian and Siberian regions.

In this section, we examine the relative roles of different regions to emitting BC that will ultimately be deposited over the Arctic. To do so, we computed the probability of BC emitted from different regions to reach the Arctic (based on FEI-NE). We defined the transport efficiency to the Arctic as the ratio between the mass of BC deposited in the Arctic and the total mass of BC emitted from a given region. These estimates were obtained by masking the same geographical regions as in Sect. 3.3 (anthropogenic sources in the Northern Hemisphere, vegetation fires in Europe, Asia, between 35 and 40∘ N, 40 and 50∘ N, 50–60∘ N and above 60∘ N) and simulating fires that occurred inside the masked areas for the period 2012–2013 (Table 1).

Figure 10 depicts the transport efficiency with which BC from vegetation fires reached the Arctic for the different geographic regions. The results clearly show that anthropogenic BC from large emitting regions in south-eastern Asia and other Asian regions were not transported efficiently to the Arctic. The main source contributing to Arctic deposition was BB in northern Eurasia with a transport efficiency of 10–32 % during spring and summer and 1–6 % in autumn and winter. Overall, the transport and subsequent deposition of BC from Asia was more effective than from Europe (12 % of the BB emissions from Asia were deposited in the Arctic, whereas only 5 % of the European BB emissions reach the Arctic), which was attributed to the fact that European BC tends to remain close to the PBL (Fig. 10), whereas the Asian BC mixes up rapidly into the free troposphere (Stohl et al., 2002). Therefore, European BC was much more affected by removal processes since its transport to the Arctic is much less efficient.

In contrast, Siberian BC was deposited very efficiently to the Arctic in summer and autumn similar to fires above 60∘ N (besides, Siberia covers the Asian part of the 60–90∘ N area). However, it was evident from our results that in our model the most efficient regional transport of BC to the Arctic occurred in the summer months and was attributed to vegetation fires in Kazakhstan and Mongolia (apart from Siberia, Fig. 10). To summarize these results, the highest transport efficiencies in our model occurred in the spring and summer for all of northern Eurasia. This may be a result of (i) extreme fire events, (ii) the relatively weak removal processes occurring in mid- and high-latitudes which favour transport without removal of BC, and (iii) the imposed fixed injection profiles used in these simulations.