Asian Emissions Explain Much of the Arctic Black Carbon Events

There is ample evidence that Black Carbon (BC) is harmful to the Arctic. BC can darken the otherwise highly reflective surfaces of snow and ice and increase atmospheric and ice surface temperatures. Because of the importance of BC to the Arctic, this work was designed to resolve the most significant source regions of Arctic BC as measured by monitoring stations in the Arctic. Using a bottom‐up approach, it is shown for the first time that there is one particular BC transport pathway from lower latitudes into the Arctic that registers at all but one of the Arctic surface monitoring stations included in this study. Through this pathway, pollutants are transported from the Indo‐Gangetic plane over Central Asia into the high Arctic in as little as 7 days. The measurement sites and BC pathways in this study are shown to be well representative of the Arctic as a whole.

Long term surface concentrations of BC in the Arctic have mostly decreased during the 1990s and 2000s Sharma et al., 2013). Recently, this trend seems to have leveled off with little to no trend observed for surface level BC in the Arctic during the last decade based on a trend analysis including all seasons (Coen et al., 2020). However, higher up in the Arctic atmosphere the situation might be different as East Asia is expected to contribute more substantially to the BC levels at higher altitudes (Ikeda et al., 2017;Sharma et al., 2013). It has been shown that emissions within the Arctic itself, such as those from gas flaring, have the potential to increase local BC surface levels substantially (Stohl et al., 2013), but this is not necessarily true for the whole Arctic (Winiger et al., 2017).
Research suggests that the seasonality of BC in the Arctic is less sensitive to annual variations of emissions than to annual changes in transport patterns of BC into the Arctic (Chen et al., 2020;Willis et al., 2018). Seasonal changes in removal processes are another known source of the seasonality of Arctic aerosols (Garrett et al., 2011). The day-to-day variability of BC in the Arctic is best modeled with high temporal resolution of BC emissions, while annual variability is best reproduced with static emissions (Schacht et al., 2019). Moreover, possible misallocations of emissions between biomass and fossil fuel sources in emissions inventories can lead to misleading results when modeling the relative importance of natural and anthropogenic sources of BC (Winiger et al., 2019). Emissions from domestic burning and wildfires are important for further constraining the impacts of BC on the Arctic (Winiger et al., 2019) and the inclusion of open biomass burning improves modeled Arctic BC (Winiger et al., 2017). Improved emissions inventories would also help to reproduce measured levels of BC that models tend to underestimate, particularly during winter and spring (Eckhardt et al., 2015).
In this work, the uncertainties associated with emissions inventories are circumvented with a focus on in-situ measurements of surface BC concentrations from multiple Arctic sites, standardized into one data set that is post processed to achieve a much better detection limit than what has been used in previous studies (Backman et al, 2017b). Using these measurements, we ask which source region is the largest contributor to Arctic BC concentrations; that is where does BC measured at a select six Arctic stations come from? The aim of this study is therefore to provide a bottom-up estimate of the annual Arctic BC source regions using the best available long-term data sets.
A widely used means of measuring BC is by filter-based absorption photometry, which yields light absorption coefficients (σ ap ) that are related with equivalent BC (eBC) mass concentration -a variable often used for modeling purposes (Petzold et al., 2013). In this work σ ap are reported, as this is what the absorption photometers actually measure. Given their importance in the Arctic, this study is focused on, but not restricted to, light absorbing aerosols. This work expands on previous work that involved data processing, data quality control, and interpretation of aerosol optical properties at six Arctic stations (Backman et al., 2017a;Schmeisser et al., 2018). The data set used here has been published previously (Backman et al., 2017b).

Materials and Methods
Lagrangian trajectory models are widely used to investigate source-receptor relationships of atmospheric constituents based on air-mass movement from meteorological data in grid form (Rastigejev et al., 2010). The method involves the calculation of an air parcel's movement from the receptor site, back in time, which yields the back trajectory of the virtual parcel of air (Stein et al., 2015).
The HYSPLIT model was run 7 days back in time using the ensemble method. The ensemble method will offset the meteorological grid by one grid point in the horizontal and 1% of the surface pressure level in the vertical. This will produce 27 back trajectories for all possible offsets in the horizontal and vertical, thus accounting for uncertainties in the meteorological data grid. The meteorological data used for the trajectories was the NCEP/GDAS data set with a 1° horizontal resolution with 23 pressure levels (Kanamitsu, 1989).
The back trajectory analysis was executed by combining the measured aerosol property (c) at one of the sites (s) at time (t) with back-trajectories as given by the HYSPLIT model for that point in time. Back-trajectories were calculated for all sites and all times when there were measurement data available. For a given t (and s), the calculated back trajectory path was matched with grid points (g) in a geodesic grid that had been closest to that back trajectory. Those grid cells were associated with c(s, t = i) so that a weighted arithmetic mean ( g c ) could be calculated for each grid cell. While matching back-trajectories with grid points (g) and measurements (c), the distance traveled (d) from the site (s) was also calculated and used for weighing (w) the arithmetic means for each of the grid points. Given that there were N overpasses of a grid point g then g c is calculated as This equation describes how the footprint maps were calculated. The weights in Equation 1 were the inverse distance traveled from the receptor point along the trajectory path. Thus, the receptor location closest (in traveled distance) to a grid cell would weigh the grid cell highest. This follows the rationale that the station closest to a grid cell is the most representative since removal processes and dilution reduce the pollutants as they are advected away from the source. Furthermore, uncertainties in the trajectory path increase with distance traveled. Trajectory points less than 500 m above ground level were considered in the analysis. This corresponds to the two lowermost pressure levels in the meteorological input data.
The aerosol properties (c) investigated in this work are light absorption coefficients σ ap that can be converted into BC mass concentrations, light scattering coefficients σ sp , and single scattering albedo ω 0 .

Footprints of Aerosol Optical Properties
The advection analysis was conducted using the Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT) version 4.9 (Stein et al., 2015) and the National Center for Environmental Prediction Global Data Assimilation System (NCEP GDAS) meteorological data (Kanamitsu, 1989). The advection model output, in the form of back trajectories, was used to calculate footprints based on measurements of σ ap at the six Arctic stations ( Figure 1). The back trajectories were matched with aerosol optical properties measured at the receptor sites and those properties were assigned to traversed grid cells according to the back trajectories to construct a map of origin of the measured properties (referred to here as a footprint map). In addition to σ ap , footprints of light scattering coefficients (σ sp ) and single-scattering albedo (ω 0 ) are also calculated; ω 0 was calculated as σ sp /(σ sp + σ ap )). The aerosol optical properties data set is comprised of three years of standardized data from 2012 to 2014. Detailed site descriptions, data processing methods, and the data have all been published elsewhere previously (Backman et al., 2017a(Backman et al., , 2017bSchmeisser et al., 2018). The measurement data are reported at a wavelength of 700 nm.
The source region density map of the back trajectories and the footprints of the aerosol properties are depicted in Figure 1. Figure  the Arctic and the northern Eurasian and American continents, with decreasing influence from the midand low latitudes. Figure 1a also shows the underlying geodesic grid to which the measured properties were averaged to avoid poor statistics for grid cells close to the pole. Figure 1b shows the σ ap footprint map following the rationale of the methods section. In general, for σ ap , it can be said that the footprint map shows very low concentrations for the whole Arctic region. In the east, namely Central Asia, elevated concentrations of light absorbing aerosols are clearly visible compared to higher latitudes with low concentrations over the Arctic ocean and the Norwegian sea. This is expected since there are few sources of light absorbing aerosols over the oceans in general and especially in the Arctic since little human activity is expected on sea ice. When air masses are advected over the Arctic measurement stations from polluted areas in Asia in under 7 days, the probability of measuring high levels of light absorbing aerosols (i.e., above station mean values) clearly increases. This suggests a substantial contribution of Asian emissions to BC levels in the Arctic, and thus also σ ap .
The more densely populated areas of Europe and North America do not stand out on the map as one might expect. This could be due to a couple of factors. First, no deposition or scavenging were accounted for in producing Figure 1, so an increased distance to the source areas would therefore decrease the importance of distant sources. The proximity of a source to the high Arctic has been shown to increase surface concentrations substantially, which is the case for gas flaring from locations close to or within the Arctic (Stohl et al., 2013). Second, the geographical distance itself does not constitute the only variable for a source's ability to affect the Arctic. Atmospheric circulation patterns also play a significant role (Chen et al., 2020;Stohl, 2006). Some stations, namely SUM in this work, do not seem to receive the same air-masses as the other stations. This is most likely due to air-mass movement and the unique topography of the station's surroundings Schmeisser et al., 2018). Moreover, precipitation patterns, and therefore aerosol wet-scavenging, can affect the amount of aerosols transported into and within the Arctic. The fact that the Asian mid-latitudes stand out as a source region of BC suggests that these aerosols are indeed transported into the Arctic.
The continental land areas also stand out in the footprint for σ sp (Figure 1c). For the Arctic ocean and the Norwegian sea, a similar homogeneity is observed as was the case for σ ap . There is no clear difference between the high Arctic ocean and oceans at lower latitudes. What is apparent, though, is that more scattering aerosols seem to originate from over land than from over presumably open water. This is true for the areas over Russia, and to some extent, North America. Also visible, but to a lesser extent, is the area also associated with elevated light absorbing aerosols (southern and south of Siberia). The seas that do not freeze over during winter do not stand out as a source of scattering sea salt aerosols. The Arctic ocean should only be a source of sea salt aerosols to the extent that the sea ice melts during summer.
The footprint of ω 0 shows that the marine regions are associated with brighter aerosols (ω 0 = 0.93-0.96) while lower ω 0 values are observed over the continents (Figure 1d). The lower ω 0 values over land do not hold true, however, for the western part of North America. The ω 0 over Europe is lower than over the Norwegian sea and the Arctic ocean. This is expected as combustion aerosols with low ω 0 are found in continental Europe in much greater quantities than over the oceans. If the scavenging and dilution processes are similar for light absorbing and light scattering aerosols, which is likely for long-rage transported submicron aerosols, then the ω 0 should be a better reflection of differences in aerosol types than the extensive aerosol properties of σ ap and σ sp . At lower latitudes, below 60 °N, the ω 0 footprint becomes less uniform, which is likely due to fewer data points in the figure; see Figure 1a.
It has been shown that SUM receives different air-masses than the other stations (Schmeisser et al., 2018).
The ω 0 can also drop well below 0.9 in late summer which is the reason for the low ω 0 over northern Greenland Schmeisser et al., 2018). It seems unlikely that the source of the low ω 0 would be from Greenland; rather, it is more probable that the source is further away and beyond the transport times investigated in this study. The characteristically different air masses arriving at SUM are due to the fact that SUM is the only high altitude station and is located on top of the Greenland ice shelf.
The 7-day back trajectory was a compromise between having a time period long enough to be able to include distant sources in the analysis and short enough so that the trajectories are still trustworthy. The 7-day duration is also the time scale that could be expected for e-folding lifetime of typical BC-containing atmospheric aerosols (Raes et al., 2000). Some parts of the aerosol-laden areas in Asia could be beyond the reach of the 7-day transport time that has been modeled here. If the modeled long-range transport exceeds 7 days, then the edge of the model domain would be associated with elevated concentrations in that area. This is indeed the case for the earlier discussed elevated values of σ ap over Central Asia; i.e. east of the Caspian Sea. Evidence of this is discussed in the next section with the aid of emissions inventories.
The multiple trajectory approach of HYSPLIT was chosen to include the uncertainty in the meteorological data fields (Stein et al., 2015). In the chosen modeling approach, the footprint represents the areas that are relatively important as sources of aerosols in the Arctic rather than the absolute amount of aerosols released at the source. If interest lies in the absolute amount at the source, then accounting for deposition and dilution during transport would be essential. Several studies have shown that modeling these processes in the Arctic is challenging, which in turn makes source apportionment uncertain (Willis et al., 2018). Figure 2 shows the site specific footprints for σ ap . Results for all stations except SUM show elevated concentrations of light absorbing aerosols arriving from central Asia; although the absolute concentrations differ (Figure 2). A difference in absolute levels is expected because of the different paths (and therefore removal and dilution processes) that the aerosols have been subject to before arriving at the stations. This consistency in independent station-specific results gives confidence for the overall conclusions of this study. Moreover, it implies that the same source area is not only important for single stations but for a greater portion of the Arctic as well.

Source Areas
Emissions of the pollutants measured at the Arctic stations are plotted in Figure 3. These include anthropogenic BC and PM10 emissions from the ECLIPSE (Evaluating the CLimate and Air Quality ImPacts of ShortlivEd Pollutants) v5 emission inventory (Klimont et al., 2017) and BC emissions from Savanna, Boreal, Temperate forests, tropical deforestation, peat and agricultural fires from the Global Fire Emissions Database GFED4s (Randerson et al., 2017). The BC emissions in the ECLIPSE and GFED inventories should both increase σ ap levels if present in air-masses arriving at the respective stations, whereas PM10 emissions would show up in elevated σ sp levels. As wildfires not only produce light absorbing aerosols but also light scattering aerosols, forest fires also register as elevated σ sp .
BACKMAN ET AL.
10.1029/2020GL091913 5 of 9 The emission maps of Figure 3 show some of the features of the footprint maps in Figure 1. The BC emissions (ECLIPSE + GFED) show the same general pattern as the σ ap footprint depicts (Figure 1). Over Russia the σ ap footprint shows elevated concentrations corresponding to the GFED wildfire areas. The source area east of the Caspian Sea (Central Asia) that was discussed earlier is also visible on the emission inventory maps. However, those sources are further south in northern Pakistan and India, whereas the footprints of Figure 1 show the source area to be farther north. This supports our earlier hypothesis that source areas outside of the modeled 7-day travel time would show up at the edges of the footprint maps if the sources farther away are strong enough.
The eastern parts of China do not stand out in the σ ap footprint (Figure 1b). These areas are clearly heavily polluted according to the emission inventory maps in Figure 3a. This implies that light absorbing aerosols from those areas do not end up in the Arctic as easily as those that originate south of Central Asia, which holds true for all the six measurement stations and supports previous research (Sharma et al., 2013). TIK, the closest station to East Asia, does not show an abundance of light absorbing aerosols originating from the eastern China region. ALT, BRW and ZEP do show elevated σ ap levels from over Asia but not from the eastern China area which is associated with high emissions in the ECLIPSE emission inventory maps.
The footprint analysis does not show the time of the individual pollution events that knowingly registered at the different stations. Thus, it is unclear whether the stations observe the same or separate events. Therefore, back trajectories that traversed the Central Asian "hot spot" and entered the Arctic were investigated to see if the associated events occur concurrently at the respective stations. The hot spot area is shown in Figure 3d and the results confirm that several events are detected at all or several of the low altitude stations. Figure 4 shows the time series of σ ap with the periods when air masses originate from inside the square in Figure 3d marked with black dots.
Events occur so that ALT, BRW, PAL, TIK and ZEP concentrations peak for extended periods of time when receiving air-masses from the area shown in Figure 3d. PAL is the station with the clearest peaks and the longest duration of air masses coming from that area and is clearly visible in the site specific footprint (Figure 2). TIK is most frequently influenced by the hot spot area due to its geographical proximity. A carbon isotope source-apportionment study of BC containing aerosols at Tiksi concluded that biomass burning aerosols were dominant during low concentrations in summer and fossil fuel was dominant during the Arctic haze period (Winiger et al., 2017). The concentrations shown for TIK in Figure 4 follow the same annual cycle as the isotope study and during the Arctic haze period much of the air mass originates from within the box. Moreover, since forest fires do not occur during the Arctic winter, the BC sources within the box should be anthropogenic in origin, at least during the winter. Unsurprisingly, SUM is again the outlier, as the station experiences virtually no periods with air originating within the selected area.
BACKMAN ET AL.

Conclusions
There is ample evidence that black carbon (BC) aerosols are detrimental for the Arctic climate. Modeling BC in the Arctic has improved, but models still struggle to replicate the seasonality and concentration of measured levels of pollution (Eckhardt et al., 2015;Sato et al., 2016;Willis et al., 2018). This seasonality is also evident in our data set, which shows that the BC concentrations are the lowest during summer (Schmeisser et al., 2018).
This unique three year data set of Arctic aerosol optical properties, including light absorption and scattering, was used as a base in the bottom-up modeling approach designed to resolve the most significant Arctic aerosol source regions. The source-area analysis using back trajectories is a simple yet effective technique for pinpointing important regions that contribute to increased levels of light absorbing aerosols, such as BC, in the Arctic. The analysis was, in a general way, able to reproduce the spatial pattern of emissions in the emission datasets, providing confidence for the chosen analysis method (Figures 1 and 3). Figure 1 shows elevated concentrations of light absorbing aerosols over Central Asia, which frequently register at all low altitude monitoring stations and far exceed background levels Figure 4.
There are two very clear pollution hot spots in Asia-the Indo-Gangetic Plain and Eastern China (Gustafsson & Ramanathan, 2016). Out of these two, pollution from the Indo-Gangetic Plain is shown to be far more likely to end up in the Arctic than has previously been shown. The analysis shows that there is measurement-based evidence of this major pathway, which frequently contributes to elevated levels of light absorbing aerosols throughout the Arctic. This source area shows up in all low altitude Arctic monitoring stations included in this study. Furthermore, the pathway identified does not contribute to isolated events of pollution for the individual sites but shows up in multiple measurements stations at the same time, indicating that the pollution intrusions spread out over wide areas in the Arctic. For at least half of the stations, these bursts far exceed seasonal levels of increased light absorbing aerosols common in the Arctic during winter months.
BC measurements provide invaluable proof of whether or not mitigation strategies work and, furthermore, are paramount for showing how well models are able to represent aerosols in the Arctic. This work clearly indicates that one important transport pathway of light absorbing aerosols into the Arctic is from central BACKMAN ET AL.
10.1029/2020GL091913 7 of 9 Asia. In order to properly model Arctic aerosols, transport pathways like those found here will have to be represented in models. The Meteorological data for the model are provided by National Weather Service's National Centers for Environmental Prediction (NCEP) Global Data Assimilation System (GDAS) and is only available at the ftp link at ftp://arlftp.arlhq.noaa.gov/pub/archives/gdas1