Elemental carbon measurements in European Arctic snow packs



[1] Black carbon (BC) and other light-absorbing particles deposited on snow and ice are known to perturb the surface radiative balance. There are few published observations of the concentration of these particles in the snow in Scandinavia and the European Arctic. We measured BC concentrations in snow samples collected in this region from 2007 to 2009, and we present the results here. The data set includes 484 surface samples and 24 column samples (covering the accumulation season) from snow on land, glaciers, and sea ice. Concentrations up to 88 ng of carbon per gram of snow (ng/g) were found in Scandinavia, while lower values were observed at higher latitudes: 11–14 ng/g in Svalbard, 7–42 ng/g in the Fram Strait, and 9 ng/g in Barrow. Values compare well with other observations but are generally found to be a factor of 2–3 higher than modeled BC concentrations in snow in the chemical transport model Oslo CTM2. This model underestimation comes in spite of potentially significant undercatch in the observations. The spring melt period enhanced BC levels in surface snow at the four sites where the BC concentrations were monitored from March to May in 2008 and 2009. A data set of replicate samples is used to establish a concentration-dependent estimate of the meter-scale variability of BC concentration in snow, found to be around ±30% of the average concentration.

1 Introduction

[2] Black carbon (BC) particles are emitted to the atmosphere through incomplete burning and have both natural (grassland and forest fires) and anthropogenic (agricultural fires, domestic fire places, flaring, and combustion engines) sources [Bond et al., 2004; Stohl et al., 2013]. BC is a very efficient light absorber, impacting radiation budgets both as an atmospheric aerosol and as an impurity in snow and ice [Bond et al., 2013]. The additional absorption of sunlight caused by BC in the snowpack enhances snow grain growth and triggers an earlier spring melt [Flanner et al., 2007]. The effect of the earlier melt of ice and snow is enhanced by the positive albedo feedback, adding to the importance of the light-absorbing pollutants. Recently published observations are leading to an ever clearer picture of the ambient concentrations of BC in the snow in polar regions [Doherty et al., 2010; Forsström et al., 2009; Hegg et al., 2009, 2010] and at lower latitudes [Ye et al., 2012; Huang et al., 2010; Wang et al., 2013a; Zhang et al., 2013].

[3] The most comprehensive effort to map the concentration of light-absorbing particles in snow across the Arctic was made by Clarke and Noone [1985] and Doherty et al. [2010]. However, they present few data points for northern Scandinavia. The motivation of the present study is to improve the mapping for this area, which is of special interest since model estimates show that this region is the area with the highest BC concentrations in the Arctic [Flanner et al., 2007; Koch et al., 2009]. We present measurements of elemental carbon (EC) concentration in snow samples collected from 2007 to 2009. EC (elemental carbon) concentration is often used to represent BC (black carbon) concentration [Andreae and Gelencsér, 2006; Bond et al., 2013]. The quantities and terms EC and BC have been defined based on the analytical method used (see Petzold et al. [2013] for a full discussion of EC/BC terminology). Thermal or refractory methods result in EC measurements, while optically based absorption measurements result in BC concentrations. Our data set includes EC concentration from 263 surface snow samples and 24 column profiles from Svalbard, Scandinavia, Barrow, and Arctic sea ice. At many of the 263 surface sampling sites, multiple samples were collected from within 1 m of each other and processed separately, providing a total of 484 individual EC measurements in surface snow.

[4] Measuring particles in snow is relatively challenging because of their low concentrations and small size. Most measurement techniques involve first melting a snow sample and passing it through a filter to collect the particles. The thermal optical method used in this study and in Forsström et al. [2009] and Aamaas et al. [2011] is a filter-based method in which particulate carbon is volatilized in two steps by exposing the substrate to high temperatures, first in an environment without oxygen, then in an environment with oxygen. The first step removes carbonate and organic carbon (OC), and the second step burns off EC. The volatilized carbon released during the two heating stages is measured to separately determine the mass of EC and other carbon on the filter. Correction for charring (conversion of OC to EC on the filter) during the first heating stage was done using the transmission method [Birch, 2003]. Some studies [e.g., Ming et al., 2008] have used similar thermal optical methods with additional preheating or acid soaking steps to remove carbonates. Warren and Clarke [1990], Grenfell et al. [2002], Warren et al. [2006], Doherty et al. [2010], and Wang et al. [2013a] used a filter-based optical method (ISSW: Integrating Sphere/Integrating Sandwich method) for determining concentrations of insoluble light-absorbing particulates (ILAP) by measuring light absorption by particles on the filter. They separate the ILAP concentrations into BC and other light-absorbing matter using standards for BC light absorption. An additional commonly used method that does not involve filtration is based on counting individual particles with a single-particle soot photometer [Schwarz et al., 2008]. This method has been used by McConnell et al. [2007] and Kaspari et al. [2011] for the detection of BC in ice cores from the Greenland ice cap and from a Himalayan glacier, respectively.

[5] Amounts of light-absorbing particles detected by different methods can vary considerably [Watson et al., 2005; Schwarz et al., 2012]. In addition to differences between the analysis methods, large spatial variability in snow impurity concentrations has been observed [Doherty et al., 2010; Aamaas et al., 2011; Svensson et al., 2013]. This study contributes to the estimation of the representativeness of snow samples analyzed for pollutants by presenting a large data set of replicate samples collected within a meter of each other at the same time and from the same snow layer.

[6] A recent multimodel study [Lee et al., 2013] showed that models severely underestimate the observed enhanced Arctic winter and spring near-surface atmospheric BC concentration. The same study also found that the modeled BC concentration in snow was, on average, within a factor of 2 to 3 of the measurements, except for the Arctic Ocean, where the concentration was significantly underestimated by the models. The episodic nature of some important pollution sources and the seasonality of the snow cover, aerosol deposition, and transport patterns are among the challenges for modeling BC in the Arctic atmosphere and its deposition in snow [Shindell et al., 2008; Koch et al., 2009; Skeie et al., 2011]. In this study the Oslo Chemical Transport Model 2 (CTM2) [Berntsen et al., 2006; Myhre et al., 2009; Rypdal et al., 2009; Skeie et al., 2011] is used to simulate the atmospheric transport and deposition of BC for the years 2007 through 2009, and the model results are compared to the measurements of EC in snow. Since emission factors used in the inventories are based on thermo-optical methods, comparing model results and measurements of EC should not include any methodological bias [Vignati et al., 2010]. When discussing model results or black carbon in general, we use the abbreviation BC, as is commonly used in the modeling community. When discussing our measurements we use the abbreviation EC, with [EC] referring specifically to the mass concentration of elemental carbon in our measurements.

2 Methods

2.1 Sampling and Analyses

[7] Snow samples were collected when the opportunity arose during various sea ice or glaciological field campaigns in 2007, 2008, and 2009. An overview of the sampling is given in Figure 1 and Tables 1 to 3. Samples were collected in Scandinavia, the Arctic archipelago of Svalbard, and Barrow, Alaska. In Svalbard (Figure 1c) annual visits by the Norwegian Polar Institute to the field sites of Brøggerhalvøya, Kongsvegen, Holtedahlfonna, Lomonosovfonna, and Austfonna were used for sampling, in addition to opportunistic sampling in Agardbukta, Inglefieldbukta, and Vestfonna. In addition samples were obtained from snow-covered sea ice during research cruises in Fram Strait, from Jan Mayen, and from the drifting station Tara.

Figure 1.

(a) Map of all sampling sites, (b) map of sampling sites in northern Scandinavia, and (c) map of sampling sites in Svalbard together with measured apparent elemental carbon concentrations in ng/g for March to April (all spring samples in parenthesis) 2007, 2008, and 2009. Values are medians at each site.

Table 1. Median Elemental Carbon Concentrations ([EC]), With (25th and 75th Percentiles in Parentheses) for Each Sitea
IDSiteDate[EC] (25–75% Percentile) (ng/g)SitesSubsites (Samples)DescriptionLatitudeLongitudeAltitude (m asl)hSWEb (mm)Commentc
  1. a

    An average of measured or estimated snow water equivalent (hSWE) in millimeters, together with the average of latitude, longitude, and altitude as meters above sea level, are included.

  2. b

    For hSWE, snow density ρs or/and snow depth hs was not measured at site but estimated from the following: A, ρs from Forsström et al. [2011]; B, ρs from Rasmus [2005] and hs from Norwegian Water Resources and Energy Directorate; C, ρs from same type of snow pack (C2) and hs from Norwegian Water Resources and Energy Directorate; D, ρs from Rasmus [2005] and hs from Finnish Meteorological Institute; E, ρs from same type of snow pack (C2).

  3. c

    P, pollution from local sources; M, pollution from local sources might have occurred.

S1Fram Strait21–27 Apr 200742.1 (13.3–88.2)15 (5)Sea ice77.48−7.34093 bA 
S2Fram Strait15–26 Sep 20078.2 (0.0–18.7)15 (5)Sea ice78.980.15030 bA 
S3Fram Strait26 Apr to 27 May 200811.4 (5.6–15.7)18 (14)Sea ice78.62−7.570108 bA 
S4Fram Strait02–08 Sep 20086.8 (3.8–7.4)13 (4)Sea ice79.41−9.97015 bA 
S5Tara25 Apr 200712.311 (2)Sea ice87.92130.050112 bA 
S6Barrow14–19 Apr 20088.7 (5.9–11.4)112 (34)Sea ice, tundra71.33−156.43058 
S7Svalbard25 Feb to 22 Apr 200713.8 (8.1–33.9)613 (45)Glacier, tundra, sea ice78.7616.00622401 
S8Svalbard11 Mar to 14 May 200813.0 (8.7–25.2)843 (98)Glacier, tundra, sea ice78.8117.21482588 
S9Svalbard27 Mar to 24 May 200911.4 (5.9–14.4)633 (69)Glacier, tundra, sea ice78.8816.39600450 
S10Pallas18 Jan to 23 May 200845.6 (30.5–89.4)117 (21)Treeline67.9724.12510335 
S11Pallas06 Mar to 05 May 200978.4 (50.5–150.9)111 (11)Treeline67.9724.12510312 
S12Tarfala12–14 Jul 200842.9 (42.5–60.5)13 (9)Tundra, glacier67.9318.471613305P
S13Tarfala05 Dec 2008 to 16 Apr 200914.5 (4.0–32.5)16 (15)Tundra67.9118.581300-P
S14Abisko24 Jan to 24 Apr 200851.4 (27.6–90.7)115 (19)Forrest68.3518.8236058 bB 
S15Abisko19 Nov 2008 to 23 Apr 200932.2 (17.5–41.8)116 (16)Forrest68.3518.8236064 bB 
S16Tromsø10 Jan to 20 May 200853.3 (31.3–95.1)124 (86)Town69.6518.9494178P
S17Jan Mayen31 Mar 200921.311 (1)Glacier71.01−8.201000684 
S18Tran620 Feb 200816.811 (3)Forrest, field69.2619.928382 bCM
S19Tran520 Feb 200813.711 (3)Forrest69.2920.46150188 bBM
S20Tran420 Feb 200810.611 (3)Tundra69.0920.76548263 bBM
S21Tran320 Feb 200819.911 (3)Tundra68.8921.0547161 bDM
S22Tran220 Feb 200815.811 (3)Forrest68.4422.6432044 bDM
S23Tran120 Feb 200832.511 (3)Forrest67.9724.1256069 bDM
S24Svanhovd14 Feb 200863.611 (1)Forrest69.4730.0530114 bEM
S25Valdres and Nordmarka19–20 Feb 200787.6 (44.5–130.6)12 (7)Forrest60.4210.0576081P
S26Tromsø Ramfjorden06 Mar 20075.411 (1)Sea ice69.5219.23099 bEM
Table 2. Svalbard [EC] (Median Concentration Together With 25th and 75th Percentiles) in Surface Snow Samples at Different Sitesa
IDSiteDate[EC] (kng/g)Subsites (Samples)DescriptionLatitudeLongitudeAltitude (m asl)hSWEb (mm)Commentc
  1. a

    These samples, except Linnébreen, are used to calculate concentrations for Svalbard in Table 1. An average of measured or estimated snow water equivalent (hSWE) in millimeters, together with the average of latitude, longitude, and altitude for eventual subsites, are given.

  2. b

    For hSWE, snow density was not measured at site but estimated from A, same type of snow pack C24.

  3. c

    P, pollution from local sources; M, pollution from local sources might have occurred.

A1Brøggerhalvøya25 Mar to 01 Apr 20078.0 (7.1–9.1)4 (15)Glacier, tundra78.8811.98158197 
A2Brøggerhalvøya08 Jan to 27 May 20089.8 (6.0–17.1)19 (23)Glacier, tundra, sea ice78.8811.94415727 
A3Brøggerhalvøya20 Feb to 18Jun 200911.1 (4.7–32.7)26 (55)Glacier, tundra78.8711.92273431 
A4Linnébreen10 Apr 2007634.01 (3)Glacier77.9613.91350657P
A5Linnébreen15 Apr 2008223.51 (2)Glacier77.9613.90340979P
A6Vestfonna28 Apr 20086.4 (5.4–9.8)10 (30)Glacier80.0019.54323553 
A7Austfonna18 Apr 200713.81 (4)Glacier79.8523.80749659 
A8Austfonna23 Apr 200822.21 (3)Glacier79.8324.02750766 
A9Austfonna30 Apr 200914.41 (1)Glacier79.8324.00600426 
A10Lomonosovfonna27 Mar 200764.4 (9.5–119.3)2 (8)Glacier78.8617.431250430 
A11Lomonosovfonna07–09 Apr 200813.0 (7.3–20.5)6 (18)Glacier78.7517.28709721 
A12Lomonosovfonna27 Mar 20095.9 (5.3–6.5)1 (2)Glacier78.8617.431250334 
A13Holtedahlfonna17 Apr 20071.41 (3)Glacier79.1413.271124286 
A14Holtedahlfonna30 Apr 20089.41 (4)Glacier79.1413.391250800 
A15Holtedahlfonna27 Apr 200911.6 ()2 (8)Glacier79.1413.391124629 
A16Kongsvegen22 Apr 20071.41 (3)Glacier78.7613.34640699 
A17Kongsvegen24 Apr 20084.21 (3)Glacier78.7617.43640956 
A18Kongsvegen23 Apr 20093.8 (2.9–4.8)2 (2)Glacier78.7713.25604820 
A19Inglefieldbukta25 Feb to 25 Mar 200725.7 (19.1–36.3)4 (12)Tundra, sea ice77.8818.293669 bAM
A20Inglefieldbukta11–12 Mar 200834.0 (23.0–38.9)4 (12)Tundra, sea ice77.8818.293657M
A21Inglefieldbukta16 Apr 200982.91 (1)Sea ice77.9018.31056M
A22Agardbukta13 Mar 200816.21 (5)Tundra78.0718.275127 bA 
Table 3. Column Samples Analyzed for Elemental Carbona
IDAreaSiteDateEC Column Load (mg m− 2)hSWE (mm)DescriptionLatitudeLongitudeAltitude (m asl)Commentb
  1. a

    Sample ID, sampling site, date, EC column load (mg m− 2), measured snow water equivalent (hSWE) in mm for the sampled snow column, latitude, longitude, and altitude in meters above sea level are shown.

  2. b

    P, pollution from local sources.

C1ScandinaviaNordmarka19 Feb 20074.358Forest60.5909.53495P
C2Tromsø07 Mar 20085.1240Town69.6518.93100P
C3Tarfala Storglaciären16 Apr 200914.61610Glacier67.9218.581240 
C4SvalbardCorbel W25 Mar 20073.0153Tundra78.90 10 
C5Nedre Brøggerbreen01 Apr 20071.5225Tundra78.9111.8320 
C6Linnébreen10 Apr 2007171.2657Glacier77.9613.90350P
C7Linnébreen15 Apr 2008165.0979Glacier77.9613.90340P
C8Vestfonna BC 628 Apr 20086.6486Glacier79.9319.19340 
C9Vestfonna BC 730 Apr 20082.4210Glacier79.9419.13195 
C10Vestfonna BC 905 May 200812.2798Glacier79.9421.28613 
C11Austfonna Base 0718 Apr 20074.2659Glacier79.8523.80749 
C12Austfonna Base 0423 Apr 200812.7766Glacier79.8324.02750 
C13Austfonna Base 0430 Apr 20099.3426Glacier79.8324.00750 
C14Lomonosovfonna summit27 Mar 200710.2431Glacier78.8617.431250 
C15Lomonosovfonna BC107 Apr 20084.2328Glacier78.6317.12200 
C16Lomonosovfonna BC207 Apr 20085.6269Glacier78.7217.28400 
C17Lomonosovfonna BC409 Apr 20088.6668Glacier78.7417.36600 
C18Lomonosovfonna summit09 Apr 200816.81089Glacier78.8617.431255 
C19Lomonosovfonna summit29 Mar 20094.3334Glacier78.8617.431250 
C20Holtedahlfonna30 Apr 20083.7796Glacier79.1413.391124 
C21Kongsvegen Stake 822 Apr 20071.8876Glacier78.7613.34640 
C22Kongsvegen Stake 824 Apr 20085.2983Glacier78.7613.34640 
C23Kongsvegen Stake 623 Apr 20094.0572Glacier78.7813.15534 
C24Inglefieldbukta16 Apr 20093.056Sea ice77.9018.310 

[8] Weekly monitoring of surface snow [EC] was conducted throughout the spring at four sites: Austre Brøggerbreen, a glacier next to the atmospheric monitoring station Zeppelin on Brøggerhalvøya in Svalbard (2008, 2009); the atmospheric monitoring station in Pallas in northern Finland (2008 and 2009); Abisko research station in northern Sweden (2008 and 2009); and the meteorological office in Tromsø in northern Norway (2008). At the three latter sites the snow depth was also measured throughout the sampling seasons. In February 2008 a transect in northern Scandinavia was sampled (S18–S23 in Table 1), ranging in elevation from sea level at the Norwegian coast to an elevation of 560 m above sea level (asl) in northern Finland. In addition, individual samples were collected at the Norwegian sites of Svanhovd, Valdres, Nordmarka, and Tromsø.

[9] To investigate the column load (mg/m2) of EC, samples were collected in vertical profiles through the snow pack, down to the ground, or down to previous summer surface on glaciers. EC column load was calculated using snow density measured at the time of sampling with a standard glaciological field method using a metal tube 20 cm in length to collect and weigh a known volume of snow. The column load was derived from [EC] measured in samples from layers between 2 and 20 cm in a vertical profile from top to bottom. The resulting vertical profile of EC mass was integrated to derive the column load.

[10] The surface samples were collected from the top 5 cm of snow, using small plastic shovels (4% of samples were from a different surface layer thickness, between 2 and 25 cm). Samples were collected in plastic bags or in glass jars. Snow was kept frozen until just before the filtering. Particles were collected on preheated quartz fiber filters (Munktel, 5.5 cm diameter) using electric or hand vacuum pumps and one of the following filtration setups:

  1. [11] The 2007 samples were melted at room temperature and filtered using a plastic funnel, cross-patterned filter holder, and small electrical pump. The part of the filter used for the analysis was taken in one of the quarters of the filter not influenced by the cross pattern on the filter holder.

  2. [12] An improved setup was used for samples collected in 2008 and 2009. The snow sample was placed into a glass funnel located inside a microwave oven that was then used for melting. Once the snow was melted, the whole sample was filtered through a filter sitting on a glass holder attached to the funnel under the microwave. The filter holder is made of sintered glass and yields a more even deposition of particles on the filter. The piece of the filter for analysis was taken from the middle of the filter.

[13] The filters were analyzed for elemental carbon using a thermal optical method (Sunset Laboratory Inc., Forest Grove, U.S., [Birch and Cary, 1996]) at Department of Applied Environmental Science, Stockholm University. The National Institute of Occupational Safety and Health (NIOSH)-5040 thermal sequence [Birch, 2003] was used to separate EC and other (carbonate and organic) carbon from the filter. The latest recommended temperature protocol European Supersites for Atmospheric Aerosol Research (EUSAAR2) [Cavalli et al., 2010] gives, on average, twice as large EC masses as NIOSH-5040, due to improved separation between the different types of carbon. In this paper the amounts of elemental carbon on each filter obtained using NIOSH-5040 are therefore multiplied by 2. The average EC mass on laboratory blanks, 29.5 ng/cm2, was subtracted from all analyzed EC masses. For typical samples used in this study, this value corresponds to less than 6% of the observed EC mass.

2.2 The Oslo Chemical Transport Model

[14] The Oslo CTM2 model is an off-line chemical transport model driven with meteorological input data, run in this study at horizontal resolution T42 (2.8 × 2.8°). The model has been updated from the model used in Skeie et al. [2011]. The main differences are that large-scale precipitation is now removed every hour, rather than every third hour in the previous version, a bug was fixed in the snow module, and meteorological data are generated from cycle 36, rather than cycle 29, of the Integrated Forecast System model at the European Centre for Medium-Range Weather Forecasts. The emissions used in this study are the 2010 fossil fuel and biofuel emissions from the Representative Concentration Pathway (RCP) 8.5 is a so-called ‘baseline’ scenario that does not include any specific climate mitigation target. The greenhouse gas emissions and concentrations in this scenario increase considerably over time, leading to a radiative forcing of 8.5 W/m2 at the end of the century [Riahi et al., 2011] and monthly biomass burning emissions from the Global Fire Emissions Database version 3 [Van der Werf et al., 2010]. The model was run, starting in August, for the years 2006–2007, 2007–2008, and 2008–2009. Meteorological input data and biomass burning emissions are specific for each year.

3 Results

3.1 Variability in the Snow Samples

[15] To investigate the meter-scale variability of [EC] in snow, 101 sets of replicate samples were collected during the surface snow sampling. For each set, two to five (most often three) samples were collected within 1 m of each other, at the same depth. The variability within a set was found to increase with concentration. A least squares fit indicates the standard deviation σ increases as

display math(1)

where the overbar indicates an average over all replicates in the set. Similarly, fits for the 25th and 75th percentiles as a function of median EC concentration ([EC]m) were obtained:

display math(2)
display math(3)

[16] Based on equations ((1)), ((2)), and ((3)), the meter-scale variability in the snow pack is on the order of ±30% of the average concentration. In addition to this meter-scale variability, the variability caused by the analysis method, discussed in section 5.4, is included in σ, [EC]25 %, and  [EC]75 %. Svensson [2011] investigated repeated analyses of EC mass from individual filters, with the same thermal optical method used here. Based on his results, we estimate that the analytical variability in [EC] in our study accounts for less than 20% of the total standard deviation among [EC] values from replicate sampling, with most of the variability represented in equation ((1)) coming from real meter-scale variability within the snow, though sampling and filtering variations also contribute.

[17] Figure 2 shows the variability observed between the replicate samples. While the ratio of [EC] determined from replicate samples ([EC]1/[EC]2, where the subscript indicates sample number, chosen such that [EC]1 > [EC]2) is between 1 and 1.4 for 47% of replicate pairs, 25% of pairs have a ratio between 1.5 and 2, and 5% have a ratio greater than 4. These large ratios illustrate the potential for large variability over small distances and suggest that sampling for BC concentration in snow should include several samples from each site to get a representative value.

Figure 2.

Variability in [EC] measured in replicate samples collected within a 1 m horizontal distance. At each site, two to five (usually three) replicates were collected. The histogram shows the ratio s1/s2 , where s1 > s2 , for all replicate pairs (for sites with more than two replicates, all possible combinations are included).

3.2 EC Distribution in Surface Snow 2007–2009

[18] The distribution of [EC] in surface samples is close to lognormal (Figure 3), with a median [EC] (25th–75th percentile) for the whole data set of 16.3 (11.4–46.9) ng/g. Concentrations measured in different sampling areas and seasons are presented in Table 1. Table 2 shows detailed results from the Svalbard sampling sites. Most of the sites were out of range of any local pollution sources; sites which we know have local pollution sources are indicated with “P.” Similarly, sites where local pollution might have occurred are marked with “M” in the tables. Where relevant, values reported here are medians over subsites and/or replicates.

Figure 3.

Histogram of all elemental carbon concentrations [EC], averaged over replicate samples and excluding two outliers from a site near a significant pollution source.

[19] We found no link between snow water equivalent (hSWE) and surface [EC], but hSWE is included in Tables 1 and Table 2 as a measure of the snow accumulation at each site up to the time of the sampling. When snow depth (hs) and density (ρs) were not measured during sampling (indicated by superscripted “b” in the tables), hSWE was estimated using nearby measurements or values from the literature, as described in the footnotes to the tables.

[20] The site-median surface [EC] in Scandinavia varied between 5 and 88 ng/g and was generally higher than at the Arctic sampling locations (S1–S9 in Table 1). The highest concentrations within Scandinavia were measured near Oslo (S25) and close to the Russian border in Svanhovd (S24). Surface samples from Tromsø, a high-precipitation urban site, had a median [EC] of 53 ng/g, similar to remote but drier areas inland (Abisko, Pallas, and Tarfala, S10 through S15).

[21] Samples from Barrow and the drifting station Tara had lower concentrations than most of the Scandinavian mainland samples, with medians of 9 and 12 ng/g, respectively. Svalbard samples had intermediate median concentrations of 12–17 ng/g. The snow on drifting sea ice in the Fram Strait had low median concentrations (6.8 to 11.4 ng/g), except in spring 2007 when the median of five samples was 42 ng/g. A statistical test comparing Scandinavian with those from Svalbard and the Fram Strait (combined) confirms that the Scandinavian samples had significantly higher [EC]. The test holds for the whole data set, for springtime data only, and for individual years.

[22] Large variability between different sites and areas is expected since significant variability is found in side-by-side samples (section 4.1). Figure 1c shows the distribution of [EC] (as site medians) in surface snow in Svalbard, an area with systematic spring sampling. Despite the large variability, there does seem to be a pattern, with surface [EC] higher in the eastern part of Svalbard than in the western part in all years (ignoring locally polluted Linnébreen). The strongest gradient is seen in data from 2007, which were discussed by Forsström et al. [2009].

[23] The evolution of snow surface [EC] at four monitoring sites through spring 2008 and 2009 is presented in Figure 4, together with measured snow depths for three of the sites. At all sites, surface snow [EC] reached its maximum values in the months of March to May. The four monitoring sites have different pollution and precipitation levels, which are reflected in the measured concentrations. Brøggerhalvøya, a remote Arctic site, had low concentrations, ranging from 2.5 to 48.5 ng/g, with medians during the observation period of 9.8 in 2008 and 11.1 ng/g in 2009. Pallas and Abisko are remote subarctic sites with low precipitation rates (300 mm yr− 1). The median [EC] measured in the surface snow in Pallas during the observation period was 45.6 ng/g in 2008 and 78.4 ng/g in 2009. At Abisko, median observed values were 51.4 ng/g in 2008 and 32.2 ng/g in the 2008–2009 snow season. The urban sampling site in Tromsø receives large amounts of BC from the surrounding town, but the high precipitation rate (>1000 mm yr− 1) resulted in samples with a median [EC] of 53.3 ng/g, comparable to that in Abisko or Pallas. In Abisko and Brøggerhalvøya [EC] was generally higher in spring 2008 compared to spring 2009.

Figure 4.

Elemental carbon concentration [EC] measured in the surface snow (top 5 cm) at four different monitoring stations over the snow seasons 2007–2008 (black) and 2008–2009 (red). The cross shows median values for replicate samples, and the error bars show the 25th and 75th percentiles, estimated with equations ((2)) and ((3)). (a) The Brøggerhalvøya sampling site is located on a glacier (440 m asl) near the atmospheric monitoring station Zeppelin. The square markers in May 2009 show samples collected close to the same glacier but at a lower elevation site (60 m asl). (b) Samples collected near the Pallas atmospheric monitoring station (510 m asl), from a site close to the tree line and therefore somewhat sheltered from winds. (c) Samples from Abisko are collected in a wind-protected site in a birch forest close to the Abisko research station. Brøggerhalvøya, Pallas, and Abisko can be considered remote sites, with minimal local pollution. (d) Samples collected in an urban site near downtown Tromsø, at the instrument field of the Norwegian Meteorological Institute (100 m asl). At the three latter sites snow depth (plotted as black (2008) and red (2009) solid lines) was monitored throughout the sampling period, in Pallas using automatic acoustic sensor and in Abisko and Tromsø by manual measurements.

[24] The sampling site at Brøggerhalvøya is located on a glacier (440 m asl), with no monitoring of the snow depth during the sampling seasons. In both years, measurements in May have about 3 times higher concentrations than in the other months. Note that most of the May samples from 2009 were collected at a lower elevation site (60 m asl), somewhat closer to the settlement of Ny-Ålesund.

[25] The onset of snowmelt (as inferred from the snow depth data) occurred in Pallas during the last weeks of April. The highest measured surface EC concentrations followed a steep decrease in snow depth in both 2008 and 2009. The 2008 samples from Abisko show an order of magnitude increase in surface [EC] during April, but in 2009 there was no significant increase. In both years sampling was terminated around 10 days before the ground became snow free. In 2008, the snow pack in Tromsø started to decrease rapidly during the last days of April, leading to [EC] above 800 ng/g; lower values returned after a snow event in mid-May, and then another peak in concentrations came as melting resumed.

3.3 Model Comparison to Observations

[26] Column samples (Table 3) had column loads of EC ranging from 1.5 to 16.8 mg/m2. The highest column loads were found at the glaciated sites Lomonosovfonna (in Svalbard) and Storglaciären (in Tarfala, Sweden). Linnébreen (in Svalbard, C6 and C7) is affected by strong local pollution (column load over 169 mg/m2) and is thus not considered. Figure 5 compares the modeled BC column loads with the measured EC column loads, providing a comparison of the modeled and observed deposition over the whole snow season. The model underestimates the column load by up to a factor of 10. The sample in Nordmarka (in southern Norway) is an exception, where the model slightly overestimates the column load.

Figure 5.

Scatterplot of modeled BC column load and observed EC column load, in mg/m2, for the column samples in Table 3. Linnébreen C6 and C7 are excluded due to local pollution. The 1:1 and 10:1 (observed:modeled) lines are indicated with dashed lines.

[27] Figure 6 shows the average concentration of modeled BC in surface snow (uppermost 5 cm) for the spring period in 2008. The medians of the observed [EC] values for the spring period (March–May) are also shown. The model captures the observed features with higher concentrations in Scandinavia than in the Arctic. Visual comparison shows good agreement in the Arctic and larger than observed values in northern Scandinavia in spring. In Figure 7 the measured surface concentrations are plotted against the modeled surface values for the corresponding day. In winter, the model underestimates the surface concentration by a factor of 3 in both Svalbard and Scandinavia. In spring, the model overestimates the concentration at a few Scandinavian sites, but generally the model underestimates the Scandinavian surface concentrations by a factor of 1.6. In Svalbard the model underpredicts the surface concentrations in spring by a factor of 2.6. No meltwater scavenging is included in the model, and it is assumed that BC in melting snow layers remains at the surface of the snowpack during melting. Thus, the BC content in the snow column is conserved until the whole snow column has melted [Skeie et al., 2011]. This will result in a positive bias in the springtime model results.

Figure 6.

Mean modeled concentration of BC in surface snow (uppermost 5 cm) for March, April, and May in 2008. The medians of the surface samples done in the spring period from Table 1 are shown in colored circles. The observations from 2007 are shifted 5° to the west and observations from 2009 shifted 5° to the east for better visualization. Locally polluted Linnébreen samples are excluded from the values shown for Svalbard.

Figure 7.

Observed surface snow EC concentrations against modeled BC concentrations. Winter (DJF, December-January-February) observations are in blue, spring (MAM, March-April-May) observations in green, summer (JJA, June-July-August) observations in red, and autumn (SON, September-October-November) observations in orange. The solid line shows a one-to-one correspondence and the dashed lines a factor of 10 difference between the model results and the observations.

4 Discussion

4.1 EC Distribution in Snow

[28] Due to reduced emissions in North America [Murphy et al., 2011] and Europe [Legrand et al., 2007], recent data from Arctic atmospheric monitoring stations show a decrease in atmospheric concentrations of BC [Eleftheriadis et al., 2009; Hirdman et al., 2010]. Clarke and Noone [1985] measured BC concentrations in Arctic snow in the early 1980s and found higher levels than the median concentrations we found at the Arctic sites of Svalbard, Fram Strait, and Barrow. Their samples from near Abisko contained similar or somewhat lower concentrations than those we found there (median of their eight samples was 31 ng/g, compared to our annual medians of 32 and 51 ng/g); however, they sampled only fresh snow, which may have caused a negative bias. There are, however, methodological differences between these two data sets, and the large temporal variability in surface snow [EC] (Figure 4) shows how important the timing of sampling can be.

[29] Snow [EC] was found to increase toward late spring at the sites where concentrations were monitored throughout the snow season. The springtime increase in surface snow [EC] could be due to a springtime peak in atmospheric transport of pollutants to the Arctic [e.g., Stohl, 2006; Quinn et al., 2007], which causes enhanced deposition, to the climatological precipitation minimum in spring which leaves the same snow surface exposed to dry deposition for a longer time, or to melting of the surface snow that leaves insoluble impurities at the surface [Meyer and Wania, 2008; Doherty et al., 2013].

[30] The observation that most of the high surface snow [EC] events occur just after the beginning of snowmelt (Figure 4), rather than developing gradually over the course of the spring, strongly suggests that particles being left at the surface during melt are the primary driver of our observed springtime [EC] maxima. Most large [EC] spikes seen in Figure 4 are clearly associated with snow depth decreases; the one clear exception is the March 2008 spike in Abisko. The increase on Brøggerhalvøya occurred in early May, at about the same time as the first observations of temperatures above freezing in Ny-Ålesund. The other proposed factors may play a lesser role in enhancing springtime [EC] but seem not to dominate. Atmospheric BC concentrations at Zeppelin and Pallas stations have been observed to peak around March [Eleftheriadis et al., 2009; Forsström et al., 2009; Hyvärinen et al., 2011], 1–2 months before the observed peak in snow [EC]. The climatologies for all of the sites shown in Figure 4 show minimum precipitation in spring, and the years in question were unexceptional. However, dry deposition on a stable snow pack would likely lead to a slow increase in [EC] with infrequent, episodic decreases when there was new snow, a pattern that does not dominate in Figure 4. More frequent sampling and better quantification of precipitation and surface melt would be needed to fully describe the roles that dry and wet deposition and snowmelt play in creating the seasonal variations in [EC] in surface snow.

[31] For a given concentration of BC in the atmosphere, one would expect lower concentrations, on average, in the snow pack at locations with more precipitation. This effect of greater precipitation amount diluting the BC in the snow should be especially strong if dry deposition dominates, though it would likely also appear even if wet deposition dominates. Figure 8 shows the column-average [EC] versus snow water equivalent, based on data from Table 3, excluding the locally polluted Linnébreen samples (column-average [EC] is the column load of EC divided by the mass of snow per unit area). As most of these sites experience limited wintertime snowmelt, snow water equivalent is a reasonable proxy for seasonal snowfall. On glaciers, the column was defined as the snow surface to the previous summer layer. The data in Figure 8 suggest that a negative correlation is apparent in our column samples, though accurately quantifying the relationship would require more systematic sampling in specific regions with spatially varying precipitation but spatially uniform atmospheric BC concentrations.

Figure 8.

Column-average EC concentration over the depth of the seasonal snow pack (column load divided by mass of snow per unit area) versus the snow water equivalent in the seasonal snow pack. The dashed line shows a least squares linear regression to the logarithm of the data.

4.2 Oslo CTM2 Model Comparison

[32] The model generally shows lower BC concentrations in snow than those observed, with column loads up to a factor of 10 lower than observed. Skeie et al. [2011] showed that the model underestimates BC in the atmosphere and snow during spring (the period in which most of the observations were conducted), compared with surface measurements and flight campaigns. The Oslo CTM2 model was included in the multimodel study by Lee et al. [2013], which showed that the Oslo CTM2, like other models, fails to reproduce the seasonality in the Arctic atmospheric BC concentration, with a significant underestimation in winter and spring. Since atmospheric concentrations in winter and spring are underestimated in the Arctic, the deposition will also be too low. This error will be most readily seen in the modeled column load of BC in snow, which is the accumulated BC over the whole snow season.

[33] There are several reasons for the underestimation of atmospheric BC concentration in the Arctic. As Skeie et al. [2011] point out, the model forcing might be missing emissions of biomass burning in midlatitudes to high latitudes during spring. There is also no seasonal variation in the fossil fuel and biofuel emissions used in the model simulations. Stohl et al. [2013] showed that including seasonal variation in domestic emissions, as well as improved emissions from gas flaring in northern Russia, enhanced the winter and spring BC concentration in the Arctic. In the modeling, there are also uncertainties related to the aging processes of black carbon, its atmospheric transport, and the removal processes of BC from the atmosphere [Skeie et al., 2011]. Wet removal is a major source of uncertainty in modeling of atmospheric BC [e.g. Shindell et al., 2008; Koch et al., 2009; Vignati et al., 2010; Schwarz et al., 2010], and several studies have pointed to wet removal as the key uncertainty in the modeling of the seasonal cycle of Arctic BC concentration [Wang et al., 2013b; Lee et al., 2013; Browse et al., 2012; Garret et al., 2010].

[34] We have compared the modeled snow column loads of BC (Figure 5) and the modeled surface snow BC concentrations (Figures 6 and 7) in coarse grid boxes of 2.8 × 2.8° with point measurements of [EC]. We expect significant variability within such a large grid box, and the point measurements may be made at locations that differ from the mean conditions for the region. The model-predicted snow-free conditions for many of the study sites that had shallow observed snowpacks (less than about 0.5 m) and the modeled seasonal snow depths were generally lower than the column samples. This may indicate a sampling bias or a precipitation bias in the model; in either case, having too little snow in the model compared to the column sample will give a modeled column load that is too small, even if the modeled concentrations are correct.

4.3 Variability in Snow

[35] Column loads of BC are less affected than near-surface concentrations by transient processes acting on a layer of snow, and they provide an integrated seasonal signal of accumulation and deposition. As a result they should show less small-scale spatial variability. The data in Table 3, excluding Linnébreen, show a positive correlation (coefficient 0.64) between hSWE, a proxy for precipitation, and column loads of EC. A similar correlation (coefficient 0.53) is found when considering only the Svalbard data. Assuming that dry deposition rates do not vary much across Svalbard, this positive correlation suggests that wet deposition is an important mechanism for depositing BC in Arctic snow. More samples in different precipitation and pollution regimes are needed to further quantify this observation.

[36] While column loads are useful for examining integrated signals, it is the BC concentration in near-surface snow that has the strongest impact on the absorption of solar radiation, and here we found large meter-scale variability in [EC] that cannot be explained by methodological uncertainties (section 4.1). Spatial variability in snow properties at different scales has been presented by earlier studies [Stenberg et al., 1999; Gusain et al., 2006; Karlöf et al., 2006; Svensson et al., 2013]. Wind-driven drifting causes relocation, densification, and ridging of snow, enhancing the spatial variability of BC concentrations in snow [Svensson et al., 2013]. Also, greater horizontal variability is observed in samples from areas with large vertical gradients in impurity concentrations [Doherty et al., 2010]. According to Aamaas et al. [2011], postdepositional processes alter BC concentrations in the snow pack and tend to increase the concentrations through sublimation.

[37] The side-by-side ratios of the replicate samples in this study (Figure 2) show a slightly larger spread than those reported by Doherty et al. [2010], whose results from samples collected less than 1 m apart were typically within 20–30% of each other. They observed the largest variability in samples collected closest to sources, which is supported by our observation that variability increases with concentration.

[38] Aamaas et al. [2011] measured vertical profiles of [EC] in snow using analytical methods identical to this study but performing vertical sampling based on observed stratigraphic layers rather than from constant depths. Their measurements include four snow pits, each 1 m from the previous, which show relatively low variability, with [EC] medians within 18% of each other. This low variability within stratigraphic layers suggests that the spatial variability seen in our data may be due, in part, to variations in which layers were nearest the surface at each sampling location.

[39] In a study from northern Scandinavia, Svensson [2011] found four times higher horizontal variability in [EC] in snow at a wind-exposed site compared to snow at a wind-protected site. Svensson's sampling design included a grid of 25 pits, 5 m apart, at two sites and reported median values within 22% of each other at the wind-protected site and within 81% of each other at the wind-exposed site. We find no such strong covariance between the spatial variability and the degree to which the site is exposed to wind.

[40] Some of our sites, such as the town of Tromsø, have local BC sources. We have indicated these sites in Tables 13. Vestreng et al. [2009] estimated that annual BC emissions in Svalbard total 61 t, mainly from the coal plants in the settlements of Longyearbyen and Barentsburg. The study by Aamaas et al. [2011] supports the assumption that none of the Svalbard sites in our study suffers from pollution from local settlements, except Linnébreen (samples A4–5 and C6–7). The relatively high values measured in Inglefieldbukta (samples A19–21) suggest local pollution. The closest settlement, Sveagruva, is a coal mining settlement with considerable emissions [Aamaas et al., 2011], located about 40 km west of the sampling site. However, because of prevailing easterly winds, the snow in Inglefieldbukta is not expected to be influenced by the settlement.

4.4 Method Uncertainties

[41] The EC concentrations presented in this study are likely underestimates of real [EC]. EC particles may become stuck to the sampling jar or plastic bag or to the funnel during filtration. There will also be some undercatch due to particles that are not trapped on the quartz filter. Additional analytical uncertainties include any sampling uncertainty related to uneven filter loading and the imperfect ability of the thermal optical method to differentiate EC from carbonate and organic carbon. As discussed in section 4.1, based on Svensson's [2011] study of the variability in EC mass derived from repeated analysis with the thermal optical method from individual filters, the variability due to uneven filter loading and method uncertainties accounts for less than 20% of the total variability of [EC] observed at a site (equation (1)). The remaining variability in the replicate data is primarily attributed to natural variability in snow, though variable loss of EC to the sampling container and funnel could also contribute.

[42] The undercatch of the quartz filters was studied with six samples which were mixed carefully and divided into two parts each. The first part was passed through a Nuclepore 0.4 µm filter, and the second part was first passed through a quartz filter then through a 0.4 µm Nuclepore filter. The Nuclepore filters were analyzed with the ISSW method at the University of Washington. Comparing the BC derived from the six pairs of Nuclepore filters indicates an average undercatch on the quartz filters of 22%. For the two samples from remote sites in Svalbard, the undercatch was 30%, and for four samples collected close to the town of Tromsø, it was 18%. These results have been corrected for the estimated undercatch of Nuclepore filters [Doherty et al., 2010]. We do not have enough undercatch samples to confidently develop a quantitative correction, so the undercatch values reported here provide an estimate for the related error in our data. Any variability in the undercatch would also contribute to the variability we report from the replicate samples.

[43] Some studies have tried to estimate the loss of particles to containers. Once the sample is in liquid form, hydrophobic soot might be left behind in the glass jar or the funnel [Ogren and Charlson, 1983; Clarke and Noone, 1985]. Ming et al. [2008] found the loss of particles in a similar filtering process to be less than 5%. For the samples collected in plastic bags, particles can attach to surfactant from the plastic, which then remains in the sampling jar [Hegg et al., 2010; Doherty et al., 2013; Wang et al., 2013a]. Wang et al. [2013a] introduced a correction factor dependent on BC concentration, which is at its maximum (about 1.5) for smallest concentrations.

[44] Figure 9 shows an intercomparison between the ISSW method used by [Doherty et al., 2010] and the thermal optical method of this study. Fourteen samples were well mixed after melting and divided for filtration using the two methods. [EC] measured by the thermal optical and ILAP (insoluble light-absorbing particulates) concentrations measured by ISSW were well correlated and of comparable magnitude, with ILAP concentrations averaging 85% of [EC]. The ISSW method overestimates BC for samples containing mineral dust [Schwarz et al., 2012], and the quartz filter used in the thermal optical method has greater undercatch, especially of aerosol transported over long distances. This combination of different error sources seems to make the comparison sensitive to the pollution regime: the two samples collected in remote but mineral-dust-rich Svalbard resulted in [EC] that averaged half the ILAP concentrations, while samples collected downtown or close to Tromsø resulted in [EC] that averaged 1.3 times the ILAP concentration. The two methods are based on different physical principles, with no perfect agreement expected, but the conversion from the thermal optical method using the NIOSH-5040 protocol to EUSAAR2 has improved the agreement between the two methods substantially. Using NIOSH-5040, a factor of 2 difference to the ISSW method was commonly seen [Doherty et al., 2010].

Figure 9.

Intercomparison of measurements from the same samples of concentrations of insoluble light-absorbing particles (ILAP) from the ISSW method and of concentrations of EC from the thermal optical method used in the present study. The dashed line indicates the 1-to-1 line, while the solid line shows a linear fit to the data (ILAP = 0.869[EC] − 0.364).

[45] BC concentrations measured in snow in 2007 by Doherty et al. [2010] on Brøggerhalvøya (7–16 ng/g, ISSW method) compare well with those in this study (median 8.3 ng/g, thermal optical). Their measurements in Tromsø in 2008 (~19 ng/g) were lower than the median of 51 ng/g in this study, which can be explained by the fact that their sampling site was on a mountain plateau above town, while ours were mostly from downtown. Samples collected during a joint field campaign in spring 2008 in Barrow are in good agreement.

5 Conclusion

[46] A large number of snow samples were collected in Scandinavia, Svalbard, Alaska, and on Arctic sea ice, both from surface snow and from vertical profiles through the snow column. The samples were analyzed for elemental carbon (EC) concentration using a thermal optical method.

[47] Systematically higher values of snow EC concentration were observed in Scandinavia than in the Arctic sites, a feature that is also seen in the chemical transport model Oslo CTM2. The Oslo CTM2 typically underestimates snow black carbon concentrations (both surface concentrations and overall column loads). As the measured EC levels in snow are likely underestimated, the gap between real and modeled concentrations is likely higher than presented in this study.

[48] A rapid increase in EC concentrations in surface snow, of up to an order of magnitude, was observed at monitoring sites at the onset of snowmelt in April or May, 1–2 months after the annual peak in atmospheric concentrations. These increases typically coincide with the onset of rapid snowmelt and likely result from insoluble particles being left at the surface as meltwater runs off.

[49] Meter-scale variability in EC concentrations in snow was found to increase with concentration. The standard deviation of multiple samples from a given site was about 37% of the mean concentration and seems to be driven by real variability in the snow pack, rather than only methodological uncertainties. This variability needs to be taken into consideration when planning future sampling routines and when interpreting existing results.

[50] The comparison between the thermal optical method and ISSW, another common method for estimating black carbon in snow, has improved substantially since the EUSAAR2 temperature came into use in place of the older NIOSH-5040 sequence. After the conversion the two methods are in good agreement, but more intercomparison samples are needed to determine whether their covariance depends on the type of aerosol.


[51] We express our gratitude to the following persons who helped in collecting the samples: G. Rotchky, M. Nicolaus, A. Nicolaus, O-M. Olsen, J. Kohler, M. Björkman, O-P. Mattila, E. Beaudon, T. Martma, C. Vega, B. Møller, F. Pinczon du Sel, and E. Brossier. We thank the personnel at the Sverdrup, Abisko, Pallas, and Tarfala stations and on Tara for helping us to collect samples; personnel at ITM, Stockholm University for filter analysis; and the meteorological office at Tromsø for allowing us to use their measurement field for sampling. J-C. Gallet and D. Divine provided helpful discussions. S. Warren and S. Doherty shared their expertise and worked with us on the ISSW intercomparison and the undercatch study. Snow depth data for Abisko were provided by Abisko Scientific Research Station and the Swedish Polar Research Secretariat and in Pallas by the Finnish Meteorological Institute. J. Kohler provided some of the snow density measurements from Svalbard. The Norwegian Polar Institute's (NPI) Mapping Section provided the cartography used in the figures. Funding for this work was provided by the Research Council of Norway (Norklima and PolRes programs) through the projects “Climate effects of reducing black carbon emissions” (165064), “Measurements of BC in Arctic snow” (178912), VAUUAV (184724), and “Svalbard ice cores and climate variability” (178764), in addition to NPI internal funding. This publication is a contribution to Cryosphere-Atmosphere Interactions in a Changing Arctic Climate (CRAICC) a Top-level Research Initiative (TRI). Additional support was provided by Formas-The Swedish Research Council for Environment through the project “Black and White,” and through the projects ECLIPSE, funded by the European Union, and SLAC, funded by the Research Council of Norway.