Increase in Elemental Carbon Values between 1970 and 2004 Observed in a 300-year Ice Core from Holtedahlfonna (svalbard)

Black carbon (BC) is a light-absorbing particle that warms the atmosphere–Earth system. The climate effects of BC are amplified in the Arctic, where its deposition on light surfaces decreases the albedo and causes earlier melt of snow and ice. Despite its suggested significant role in Arctic climate warming, there is little information on BC concentrations and deposition in the past. Here we present results on BC (here operationally defined as elemental carbon (EC)) concentrations and deposition on a Svalbard glacier between 1700 and 2004. The inner part of a 125 m deep ice core from Holtedahlfonna glacier (79 • 8 N, 13 • 16 E, 1150 m a.s.l.) was melted, filtered through a quartz fibre filter and analysed for EC using a thermal–optical method. The EC values started to increase after 1850 and peaked around 1910, similar to ice core records from Greenland. Strikingly, the EC values again increase rapidly between 1970 and 2004 after a temporary low point around 1970, reaching unprecedented values in the 1990s. This rise is not seen in Greenland ice cores, and it seems to contradict atmospheric BC measurements indicating generally decreasing atmospheric BC concentrations since 1989 in the Arctic. For example, changes in scav-enging efficiencies, post-depositional processes and differences in the vertical distribution of BC in the atmosphere are discussed for the differences between the Svalbard and Greenland ice core records, as well as the ice core and atmospheric measurements in Svalbard. In addition, the divergent BC trends between Greenland and Svalbard ice cores may be caused by differences in the analytical methods used, including the operational definitions of quantified particles, and detection efficiencies of different-sized BC particles. Regardless of the cause of the increasing EC values between 1970 and 2004, the results have significant implications for the past radiative energy balance at the coring site.


Introduction
During the last century the Arctic has warmed twice as fast as the rest of the world, which is likely partly explained by changes in albedo and related feedbacks in the Arctic, a region covered with high reflectivity snow and ice (ACIA, 2005).Black carbon (BC) particles are produced by incomplete combustion of biomass as well as fossil and biofuels.Due to its colour BC absorbs light and it is recognized as a strong warming agent in the atmosphere (e.g.Bond and Bergstrom, 2006;Ramanathan and Carmichael, 2008).The climate effects of BC are intensified in snow and ice, where it lowers their reflectivity, leading to earlier spring melt (e.g. Warren and Wiscombe, 1980;Hansen and Nazarenko, 2004;Flanner et al., 2007;Bond et al., 2013).BC has been estimated to be the second most important climate warming agent globally after carbon dioxide, and in the Arctic it is even more important than greenhouse gases (Jacobson, 2001;Hansen and Nazarenko, 2004;Bond et al., 2013).Furthermore, the BC-albedo effect has been suggested to have caused 20 % of the warming and snow-and ice-cover loss in the Arctic during the last century (Koch et al., 2011).However, the effects of BC remain one of the largest sources of uncertainty in climate change analysis and models (e.g.Bond et al., 2013).
Due to its importance in climate perturbation, BC distribution and concentrations have been increasingly studied around the world.The first continuous measurements on atmospheric BC concentrations in the Arctic started in the late 1980s (e.g.Hirdman et al., 2010a).The number and spatial coverage of BC measurements from Arctic snow are increasing (e.g.Clarke and Noone, 1985;Forsström et al., 2009;Doherty et al., 2010;Aamaas et al., 2011;Doherty et al., 2013;Forsström et al., 2013), but comparing the individual results and establishing regional trends remain challenging because most snow measurements only represent a snapshot of BC concentrations in time and space.Despite the significance of BC in Arctic climate warming and geophysical processes, knowledge on its long-term concentrations and climate effects in the area beyond the observational data is very limited at present.McConnell et al. (2007) and McConnell and Edwards (2008) were the first to report Arctic BC concentration and deposition trends, based on three ice cores from the Greenland Ice Sheet.McConnell et al. (2007) found a 7-fold increase in BC concentrations from 1850, peaking around 1910, followed by a decline to almost pre-industrial levels after 1950.This study was highly influential in introducing a new high-resolution method to measure BC concentrations with a single-particle soot photometer (SP2) in ice cores and was soon followed by other long-term studies on BC trends from Himalayan glaciers (e.g.Kaspari et al., 2011) and the Antarctic Ice Sheet (Bisiaux et al., 2012a, b), while European Alps glaciers had already been studied with other methods (e.g.Lavanchy et al., 1999;Jenk et al., 2006;Legrand et al., 2007;Thevenon et al., 2009).However, McConnell et al. (2007) recognized that the BC deposition at the highelevation sites in Greenland could not be taken as representative of overall BC trends in lower-elevation regions of the Arctic.The heterogeneity of the global distribution of BC is seen, for instance, in the different concentrations and trends between Himalayan, Arctic and European records, and even between individual records from these areas (e.g.Ming et al. 2008;Xu et al., 2009;Kaspari et al., 2011).To get a better overall view on its trends and effects, further investigations on BC concentrations and deposition are required from other locations in the Arctic.
Here we present results on BC (here operationally defined as elemental carbon (EC) due to the thermal-optical measurement technique used) concentrations and deposition from a Svalbard glacier, Holtedahlfonna, from 1700 to 2004.When discussing black carbon in general, we use the abbreviation BC, as is commonly used in the community.When discussing our specific measurements, or measurements made by others with similar thermal-optical methods, we use the abbreviation EC.For a recent review of BC terminology we refer the reader to Petzold et al. (2013).We find an unexpected increase in EC values between 1970 and 2004.Factors such as atmospheric BC concentrations, changes in BC scavenging efficiencies and post-depositional processes are discussed in order to explain the observed trend.In addition to BC records from Svalbard being of particular interest in light of the limited data available from the Arctic, Svalbard glaciers are valuable as they are expected to have different BC sources than Greenland (Hirdman et al., 2010b).Glaciers around the world are susceptible to melt and retreat caused by climate warming and possibly also BC deposition.As glaciers play an important part in climate feedbacks and the hydrological cycle, and as high-elevation glaciers at lower latitudes supply water to major human populations, it is crucial to learn more about the BC-glacier interactions.

Ice core recovery and sampling
Svalbard is located in the Arctic Ocean (Fig. 1) at the southern edge of the permanent sea ice.Despite its location at high latitudes, the archipelago has a relatively mild climate due to an intrusion of the North Atlantic Current bordering western Svalbard and its location on the pathway of both Arctic and North Atlantic cyclones.About 60 % of Svalbard is covered by glaciers, of which the majority have had a negative total volume change during the last 15-40 years (Nuth et al., 2010).Even the highest elevation glaciers on Svalbard, such as Lomonosovfonna (Fig. 1), can experience frequent surface melt in the summer (e.g.Beaudon et al., 2013).
Holtedahlfonna is the largest ice field on the western island of Spitsbergen in Svalbard, covering ca.300 km 2 and situated 40 km northeast of the Ny-Ålesund research station (Fig. 1).The Holtedahlfonna ice core was drilled in April 2005 at 79 • 8 15 N, 13 • 16 20 E, 1150 m elevation, at a saddle point where the lateral ice flow velocity is expected to be minimal (Lefauconnier et al., 2001;Sjögren et al., 2007).The retrieved 125 m deep ice core did not reach the bedrock and thus the precise ice depth at the coring site is unknown, but radar measurements suggest it to be approximately 150 m (Sjögren et al., 2007;Beaudon et al., 2013).The ice core was retrieved in ca.50-60 cm sections and immediately packed into plastic bags, subsequently stored frozen and then transported to the cold-room facilities at the Norwegian Polar Institute (NPI), Tromsø, Norway.
The ice core was cut and processed in a cold room (−22 • C) using a cleaned thin-blade band saw.Each vertical ice core section was split into subsamples assigned to tritium (van der Wel et al. 2011), oxygen isotope (Divine et al., 2011), organic contaminant (Ruggirello et al., 2010), major ion (Beaudon et al., 2013) and EC measurements.EC measurements were performed on subsamples cut from the inner part of the core, i.e. the part best protected from contamination during drilling and handling of the ice.The samples were continuous except that a section representing the time period 1740-1755 was not available for EC analysis.After preliminary cutting, each surface of the subsamples was scraped with a clean stainless steel knife under a laminar bench, after which the samples were placed in plastic bags and stored frozen.After subsampling there were 739 ice pieces of 5-20 cm vertical length and an average horizontal cross section of 2.8 cm by 3.5 cm, equal to around 10 cm 2 (± 2 cm 2 ) surface area, available for EC analysis.
Dating of the ice core was performed using an age-depth scale based on the ice-thinning model by Nye (1963) constrained by the depth of the 1963 radioactivity fallout layer at 28.5 m depth (van der Wel et al., 2011), as well as counting of annual δ 18 O layers (Divine et al., 2011).In addition, a dating method based on statistical extraction of historically known volcanic eruptions (Moore et al., 2012) complemented the other dating methods.Volcanic signatures of five known eruptions, including the Laki eruption (1783) at 103.6 m depth, could be used as reference horizons in a refined Nye thinning model, and suggest that the core covers a period of 305 years with a mean accumulation rate of 0.38 m w.e.yr −1 (Moore et al., 2012).

Filtering and EC analysis
The 739 ice samples were grouped for filtering so that ice samples representing the 18th century were filtered in ca.10-year resolution, of the period 1800-1850 with a 5-year resolution, and samples from the industrial era with a resolution of approximately 2 years.The ice was melted and immediately filtered through pre-burned (at 800 • C for 4 hours) quartz fibre filters (Munktel) following the procedures described, for instance, in Forsström et al. (2009) and Svensson et al. (2013).This resulted in 88 EC filter samples.In order to check for possible contamination in the filtering system, blanks (nine) were prepared by placing a filter in the filtering system for a minute or by filtering distilled water through a filter (three).
The filters were analysed for EC using a thermal-optical method (TO, Sunset Laboratory Inc., Forest Grove, USA; Birch and Cary, 1996) at Stockholm University.The analysis was performed with the latest recommended thermal sequence EUSAAR_2 (Cavalli et al., 2010) in order to separate EC and other (carbonate and organic) carbon from the filters.Carbonates were not eliminated from the filters with acid treatment before the analysis, since this has been shown to cause some errors in the analysis, e.g.induction of intense charring phenomena (Jankowski et al., 2008), and a decrease in BC particle size (Kaspari et al., 2011;Schwarz et al., 2013).Moreover, natural calcite evolves in the helium mode with the EUSAAR_2 protocol, and will be detected as organic carbon (Cavalli et al., 2010).Therefore we concluded that natural carbonates would not interfere with the EC measurements.All blanks showed EC concentrations from below detection limit to a magnitude lower than the detection limit of the analysis method (0.2 EC µg cm −2 ).

Uncertainties
In the filter-based method used in this study, a 1.5 cm 2 filter punch was analysed from the middle part of each filter (total area 11.34 cm 2 ).Additional punches from the same filters showed non-significant variation in EC loadings between different parts of the filters (n = 8; average relative standard deviation = 3 %; range of relative standard deviation = 1.1-4.9%).However, some of the filters had a visible gradient of deposited matter, with one edge being lighter than the other, possibly due to problems keeping the filtering system completely in level.In the cases where the EC was not uniformly deposited, based on ocular inspection and the deviation from the three punches, the EC concentration in the middle of the filter was close to the average of the whole filter, whereas the light side contained lower, and the darker side higher, EC concentrations (n = 16, average relative standard deviation = 19.6 %; range of relative standard deviation = 9.1-38.3%).These deviation values are in accordance to previously reported average standard deviations of 21 % in Svensson et al. (2013).As expected, the relative standard deviations were highest for filters with low EC loadings.The analysed filter punch was taken from the middle of the filter in all cases.For the samples from which several filter punches were analysed, the reported concentrations are averages of the filter punches and the error bars are shown in Fig. 2.
A known error source of the filter approach relates to the efficiency of the used filter to capture EC particles in liquid samples.Previous studies have shown that significant amounts of EC particles in liquid samples may percolate through the filter, resulting in undercatch (e.g.Forsström et al., 2013;Doherty et al., 2013;Torres et al., 2014).These penetrating particles are most likely small in size, as they are not even captured by additional filters in series, as seen in our measurements and reported by Torres et al. (2014).This will result in the presented EC concentrations likely being an underestimate of real EC concentrations.Forsström et al. (2013) reported this undercatch to be 22 % on average by using the same filters as in this study and by analysing a second filter (Nuclepore 0.4 µm) in line with the ISSW (integrating sphere/integrating sandwich method) method used by Doherty et al. (2010Doherty et al. ( , 2013) ) at the University of Washington.However, the amount of undercatch samples in Forsström et al. (2013) was not enough (n = 6) to confidently develop a quantitative correction of undercatch with the used method.But as the undercatch values reported in Forsström et al. (2013) are from snow samples partly collected from Svalbard, we feel that they could provide an estimate for the related error in our data.Possible additional losses of EC particles through the filter were avoided by refraining from the nitric acid treatment to eliminate carbonates from the filters.If the two separate error sources of the method used (first, the general standard deviation due to heterogeneous loading of EC on the filters, 20 %, and second, the undercatch, 22 %) are added together in quadrature, as in Schwarz et al. (2012), we end up with an estimate of maximum 35-40 % total uncertainty related to our measurements.
However, Schwarz et al. (2013) suggested that atmospheric processes during the formation of snow and repeated thaw-freeze cycles in deposited snow may lead to agglomeration of BC particles and a subsequent shift to larger particles in snow samples compared to atmospheric samples.The Holtedahlfonna ice core samples have most likely experienced several thaw-freeze cycles as the glacier experiences annual summer melt (Beaudon et al., 2013).Therefore, the EC particles in the ice core samples may tend to be larger and be filtered with higher efficiency than estimated based on the results of Forsström et al. (2013).A possibility to increase the filtering efficiency of quartz fibre filters in future studies is the addition of salts and acids to the melted snow and ice, as it causes coagulation of EC (Torres et al., 2014).Unfortunately, we were unaware of this possibility during the filtering in our study.

Holtedahlfonna EC concentrations and comparison to other European data
EC concentrations in the Holtedahlfonna ice core have varied significantly between 1700 and 2004 (Fig. 2).Between 1700 and 1850, the concentrations are generally low, on average 23 µg L −1 (Table 1), showing no specific trend.In samples corresponding to the period between about 1850 and 2004, the concentrations are generally higher than for the oldest part of the ice core, on average 39 µg L −1 .Two local minima in the 1920s and 1970 are shown clearly in the EC record.Peaks in EC concentrations occur at about 1910, 1940 and 1998 at 80, 58 and 103 µg L −1 , respectively.A clear feature of the Holtedahlfonna EC record is the steady increase in EC concentrations starting around 1970 after a temporary minimum in the record, reaching unprecedented values in the 1990s.This increasing trend since 1970 is unexpected as it contradicts previous data from the Arctic.Atmospheric BC measurements from the Arctic only go back to 1989, but the stations at Alert (Ellesmere Island, Canada), Barrow (Alaska, USA) and Zeppelin (Ny-Ålesund, Svalbard, Norway) record an overall 40 % decline in atmospheric BC concentrations between 1990 and 2009 (Sharma et al., 2013).The atmospheric trend is explained by a general decrease in BC emissions in northern latitudes since the 1990s, in particular associated with the collapse of the Soviet Union (Hirdman et al., 2010a;Sharma et al., 2013).The BC ice core records from Greenland show decreasing BC values to almost pre-industrial levels during the 20th century after a clear peak around 1910 (McConnell et al., 2007, Fig. 3d;Mc-Connell and Edwards, 2008;McConnell, 2010).Therefore, either stable, and for this record comparably low, or moderately declining EC concentrations could also have been expected in the Holtedahlfonna ice core during the last decades of the record.
The magnitude of the Holtedahlfonna EC concentrations seems to be in accordance with EC values measured with other filter-based thermal, thermal-optical or other similar methods in ice cores from the European Alps, presented in Table 1.It may seem surprising that EC concentrations on Svalbard, which is more remote from emission sources, are comparable or even higher than on the glaciers in the European Alps.However, this is likely partly explained by differences in the specific methodologies used in the studies.
For instance, the EUSAAR_2 temperature protocol used in this study typically gives an EC content a factor of 2 higher than the previously commonly used NIOSH protocol, due to improved separation between the different types of carbon (Cavalli et al., 2010).
A clear similarity between the present study and studies from the European Alps, at Col du Dôme, France (Legrand et al., 2007); Colle Gnifetti, Switzerland (Lavanchy et al., 1999;Thevenon et al., 2009); and Fiescherhorn, Switzerland (Jenk et al., 2006), is that the EC (or BC) trends indicate an increase in concentrations from the end of the 19th century to the present (Table 1), while the timing of peak values in the 20th century varies between the studies.Unfortunately, the record of Jenk et al. (2006)  BC and EC studies on Himalayan ice cores, as in Europe, have repeatedly shown different and contradicting trends when measured with different analytical methods, even from the same glaciers (e.g.Ming et al., 2008;Xu et al., 2009;Kaspari et al., 2011).The SP2 method was originally de-signed to quantify BC mass concentrations and size distribution of atmospheric samples (e.g.Schwarz et al., 2006).Thus, when applying the SP2 method to liquid samples, such as melted snow or ice, the liquid and particulates need to be nebulized into a dry aerosol phase.This added step of nebulization has shown to cause additional uncertainties in the BC concentration measurements, as larger sized BC particles are not aerosolized as effectively as small ones (e.g.Schwarz et al., 2012;Schwarz et al., 2013).Consequently, typically only particles of smaller than ∼ 500 nm core diameter have been quantified with the method from ice core samples (e.g.Kaspari et al., 2011;Bisiaux et al., 2012a, b).However, Schwarz et al. (2012Schwarz et al. ( , 2013) ) showed the presence of significantly larger BC particles in snow than are typically observed in the atmosphere.Therefore, Schwarz et al. (2012) recommend extending the size range of particles quantified with the SP2 method to at least 1.5 µm.At the same time the filter-based methods may underestimate the occurrence of BC particles due to lowered filtering efficiency at smaller particle sizes (e.g.Lim et al., 2014;Torres et al., 2014).Due to these differences in the methods, they will most likely measure different concentrations, but could also measure different temporal trends if the size distribution of BC has varied through time, for instance due to changes in BC source types.Therefore, the European and Himalayan ice core results suggest that variations in observed EC and BC ice core trends can be expected also in the Arctic, especially when using different analytical methods.At dusty sites, such as the Himalayas, dust may interfere with the thermal-optical method (e.g.Wang et al., 2012) resulting in additional problems with comparing records, but dust should not be a problem when comparing records from the Arctic.

EC concentration vs. EC deposition
EC concentrations in snow are determined by numerous factors, such as BC concentration in the air; dry and wet deposition velocities; precipitation amounts; and post-depositional processes of wind drift, sublimation and melt (Doherty et al., 2010(Doherty et al., , 2013)).Ice core EC concentration trends may be particularly sensitive to changes in the snow and ice accumulation at the specific site through time.In addition, seasonal melt can lead to enhanced EC concentration at the surface of a glacier due to melt amplification (Doherty et al., 2013).These factors may lead to apparent fluctuations in recorded EC concentrations, although EC deposition may have been constant over time.
Consequently, to further illuminate the processes leading to the recorded EC concentrations in Holtedahlfonna, we calculated EC deposition (Fig. 3b).Generally, flux and deposition calculations will give the same results in the same unit (EC µg m −2 yr −1 ) even though they are calculated differently.Here EC deposition was calculated by dividing the total amount of EC in a (filter) sample by the cross section of the ice sample and the amount of years covered in one filtered ice sample.Flux, on the other hand, would be calculated by multiplying the measured EC concentration by the snow accumulation rate from the glacier.The yearly snow accumulation rate is the sum of all snow accumulating (precipitation) and reducing (ablation, runoff) processes.We chose to calculate deposition, rather than flux, to avoid incorporating snow accumulation rates which are another source of uncertainty.The available snow accumulation rate data from Holtedahlfonna (Fig. 3c) are detrended average accumulation rates between volcanic reference horizons (Beaudon et al., 2013) and are therefore insufficient in detail for meaningful flux calculations.Deposition calculations eliminate postdepositional glaciological factors, such as snow accumulation rate and redistribution of EC in the snowpack, affecting the EC concentrations in the ice.The summer surface melt and subsequent redistribution of EC in snowpacks will not affect deposition values as long as the redistribution is limited to the annual snowpack and no significant amounts of melt water are lost from the glacier during melt.From glaciers with summer melt, Doherty et al. (2013), Ming et al. (2009) and Xu et al. (2012) have shown that BC is enriched at the top of the melt layer, and only about 10-20 % of the total BC is elusive and can percolate into deeper layers in the snowpack (Doherty et al., 2013).Percolation of BC is prevented by ice layers formed in the snowpack by refreezing of the previous year's melt layer (Doherty et al., 2013).Xu et al. (2012) showed that, on Tibetan glaciers experiencing extensive summer melt, BC can also be significantly enriched above the previous year's melt layer.As summer melt is occurring at Holtedahlfonna, ice layers are likely formed annually in the snowpack and firn (Beaudon et al., 2013), which suggests that the deposition values that we find in the ice core should not significantly be affected by EC redistribution.Despite summer melt, no significant melt water runoff that would affect EC concentration and deposition values is expected to occur at the ice coring site, as discussed in more detail in Sect.3.5.
The EC deposition trend is shown in Fig. 3b.Similar to the EC concentration trend it indicates lower deposition before the industrial era and high deposition at the beginning of the 20th century.The most evident difference is that the wide pronounced peak in the EC concentrations between 1920 and 1970 is not clear in the deposition.If atmospheric and postdepositional processes stay constant, concentrations and deposition present similar trends in ice cores.If concentration and deposition trends differ from each other, the most likely explanation is in general a change in snow accumulation rate.
Temporal changes in snow accumulation may, under stable atmospheric processes, dilute or concentrate EC amounts in snow or ice, whereas the deposition is not affected.Here, the snow accumulation rate must be compared to accumulation rates before and after the discrepancies between the concentration and deposition trend, and not necessarily the whole snow accumulation record.Consequently, it seems that the peak and variation in EC concentrations occurring between 1920 and 1970 may be partly caused by changes in snow accumulation during this time period, rather than more EC being deposited on the glacier.Comparison of the EC concentrations with the measured snow accumulation of the ice core (Fig. 3c) indicates that snow accumulation at the ice core site was lowered during ca.1930-1960 compared to the adjacent time periods (1910-1930 and 1960-2000), which may have concentrated EC in the ice core, while EC deposition seems not to have increased in that period.
The EC deposition trend in Holtedahlfonna shows similarities to the BC trend recorded in the Greenland ice cores, indicating a peak in both BC concentrations and deposition in the early 1900s and a decrease afterwards (McConnell et al., 2007, (Fig. 3d);McConnell and Edwards, 2008;McConnell, 2010).This BC peak is simultaneous in the Greenland and Svalbard ice cores, although more pronounced in Greenland.
The rapid increase in the post-1970 EC concentrations in Holtedahlfonna after a temporary minimum is also apparent in the deposition (Fig. 3a and b).This indicates that increasing EC amounts have been deposited at the site from 1970 to 2004 despite the measured decrease in atmospheric concentrations since 1989 in the Arctic (Hirdman et al., 2010a;Sharma et al., 2013).Consequently, the increase in EC concentrations cannot be simply explained by changes in the snow accumulation rate at the glacier.This example from Holtedahlfonna illustrates the importance of also studying the absolute deposition rather than focusing on the concentrations per se.The observed differences in the temporal evolution of EC deposition and concentration in the ice core show that additional processes to changes in atmospheric BC concentration are important for the final EC concentration in the ice core.Therefore, when comparing different ice cores to each other, both deposition and concentration of BC or EC are important to the analysis.In the following we will focus on explanations, which we find most plausible, for the unexpected increase in EC values in Holtedahlfonna between 1970 and 2004.

Black carbon emissions
The prevalent conception is that EC concentrations and deposition in the Arctic have decreased during the recent decades.This view is supported by atmospheric measurements (Hirdman et al., 2010a;Sharma et al., 2013) and ice core BC records from Greenland (McConnell et al. 2007;McConnell and Edwards, 2008).Model results also suggest that BC deposition in Svalbard was at a maximum around 1950 to 1960 and has decreased since then (Koch et al., 2011;Ruppel et al., 2013).Historical anthropogenic fossil fuel emission inventories similarly indicate decreasing BC emissions in North America and Europe in the latter half of the 20th century and since 1990 in the former Soviet Union (e.g.Novakov et al., 2003;Bond et al., 2007;Fig. 4).However, these BC emission inventories (e.g.Nonakov et al., 2003;Bond et al., 2007) may insufficiently portray some significant fossil fuel emission sources (such as flaring in northern Russia, discussed below) and do not include natural emissions, such as forest and wild fires, that are major sources of BC in the Arctic (e.g.Stohl et al., 2007;Bond et al., 2013).At the same time, the recorded EC deposition at Holtedahlfonna shows evident similarities with the historical global annual anthropogenic BC emission trend, peaking in the early 20th century and followed by a drastic rise since the 1970s (Fig. 4; Lamarque et al., 2010;Smith and Bond, 2014).The early 20th century global emission peak was mainly caused by European and American emissions, whereas the increase since the 1970s is dominated by Asian emissions (Bond et al., 2007).However, northern Eurasia surface emissions have shown to be the dominant sources for atmospheric BC in the Arctic (e.g.Hirdman et al., 2010a, b, and references therein), and therefore regional, as opposed to global, sources are expected to be mostly responsible for the recorded EC deposition trend in Holtedahlfonna.
Potential local emission sources on Svalbard are, among others, the coal mines in Longyearbyen, Barentsburg and Pyramiden, but these are not expected to have caused the observed trend in the Holtedahlfonna ice core.The nearest mine, located in Ny-Ålesund, closed down in 1963 and therefore cannot account for the recent trend.In addition, Forsström et al. (2013) showed that local EC pollution on Svalbard snow is focused around the sources, and, more specifically, snow around Ny-Ålesund presents EC concentrations similar to Arctic background levels (Aamaas et al., 2011).Apart from local terrestrial emissions, increased shipping in the Arctic has been proposed as a potential significant local BC source in the area (e.g.Eckhardt et al., 2013, and references therein).However, Eckhardt et al. (2013) showed that while cruise ships visiting Ny-Ålesund in the summer months cause equivalent BC concentration peaks at the atmospheric Zeppelin measurement station (474 m a.s.l.) located less than 1 km away from the village, their influence on the annual mean concentration is minimal.
As mentioned, air mass transport to Greenland and Svalbard are rather different.The greater part of atmospheric transport to Svalbard comes from Europe and northern Siberia (Stohl et al., 2006;Hirdman et al., 2010b), as opposed to North American emissions dominating Greenland ice core records (McConnell et al., 2007).In addition, the Greenland Ice Sheet and Svalbard glaciers receive divergent pollution deposition due to their different elevation.For instance, according to Hirdman et al. (2010b), the BC source regions for Summit on the Greenland Ice Sheet (3208 m a.s.l.) were specific for this high-elevation site and did not reflect the lowlevel atmospheric transport of pollution from high-latitude Eurasia that influences much of the rest of the Arctic which lies at lower elevations.
Of the Arctic monitoring stations recording atmospheric BC concentrations, the Zeppelin station in Ny-Ålesund has been shown to be particularly susceptible to BC emissions from northern Russia, especially western Siberia (Eleftheriadis et al., 2009;Hirdman et al., 2010b;Stohl et al., 2013;Tunved et al., 2013).Stohl et al. (2013) pointed out that this is an intensive gas flaring region (circled in Fig. 1) and a previously underestimated major Arctic BC emission source.In fact, based on simulations run for 3 years with a Lagrangian particle dispersion model, Stohl et al. (2013) suggested that flaring in northern Russia may contribute to 42 % of the annual mean atmospheric BC concentrations near the surface in the Arctic.
Since flaring emissions from northern Russia originate at high latitudes, their isentropic lifting in the polar dome is limited (Stohl, 2006;Stohl et al., 2013).The BC concentrations resulting from flaring are highest closer to the surface and decrease rapidly vertically in the atmosphere (Stohl et al., 2013).Furthermore, model results by Sand et al. (2013) suggest that a large fraction of BC emitted in the Arctic stays in the lowermost layers in the atmosphere and gets deposited at the surface.Therefore, despite their globally comparably small emission quantities, Arctic emissions may have a predominant impact on BC deposition at lower elevations of the Arctic (Sand et al., 2013).According to Stohl et al. (2013, Fig. 6 therein), emissions from flaring may contribute to 30-40 % of the simulated annual surface concentration of BC (ng m −3 ) on Svalbard in comparison to 5-10 % on Greenland.It seems that the Greenland ice cores (e.g. at 2713 and 2410 m a.s.l.) are likely missing most of the flaring emissions both because of their different source areas and their elevation, whereas the Holtedahlfonna ice core is likely to capture more of these emissions.Interestingly, the large gas and oil fields in northern Russia, especially around the Gulf of Ob and west of it (circled area in Fig. 1), were discovered and established at the end of the 1960s to the 1970s, and it can be expected that flaring commenced and increased soon after the extraction of oil and natural gas started there.Today, Russia is still the leading contributor to flaring globally, which is mostly a consequence of lacking infrastructure in the region to transport and utilize all extracted natural gas (Elvidge et al., 2009).According to satellite imaginary, flaring emissions from Russia continued to increase from 1994 to 2005 (Elvidge et al., 2009).Therefore, it seems that the concurrent peak in the Greenland and Svalbard BC ice core records around 1910 could have been caused by the sites receiving the same emissions from distant sources, most likely North America, since the BC peak in the Greenland ice cores was more pronounced than in Svalbard.The discrepancies between the sites since 1970 could partly be caused by Greenland not receiving major lower troposphere emissions that are recorded in Svalbard.However, it is unlikely that the post-1970 EC trend suggested by this Svalbard ice core was caused by Russian flaring emissions alone, as, for instance, it does not seem to indicate clear or even short-term signs of lowered emissions associated with the collapse of the Soviet Union.

Changes in scavenging efficiency?
While the increasing post-1970 EC trend in the Holtedahlfonna ice core may partly be explained by rising BC emissions from flaring (and possibly global BC emissions), these emission increases have not been captured by the atmospheric measurements at the Zeppelin station since 1998 (Eleftheriadis et al., 2009;Hirdman et al., 2010a) or any other atmospheric measurement station in the Arctic since 1989 (Hirdman et al., 2010a;Sharma et al., 2013).Although the overlapping time period of the ice core and atmospheric measurements is very short (three ice core data points), and the resolution of the ice core measurements too low for detailed temporal comparison with hourly atmospheric measurements, the recorded differences indicate further processes affecting the EC trends in the Holtedahlfonna ice core.
Variations in EC deposition that are not caused by changes in atmospheric concentrations may be caused by changes in the scavenging efficiency of BC.With constant or declining atmospheric EC concentrations, EC deposition may increase if BC scavenging efficiency increases.One possible pathway to influence the scavenging efficiency could be by changes in the temperature at which precipitation forms.In general, in-cloud scavenging of aerosols is less effective in ice clouds compared to liquid-phase clouds.This is because, at cold temperatures, precipitation is formed via activation of especially efficient, but relatively few, ice nuclei that then grow by vapour deposition.At warmer temperatures, liquid precipitation is formed via collision and coalescence, which is a more efficient way to remove aerosols from clouds.Observations by Cozic et al. (2007) illustrate this effect clearly for BC particles.Below about −20 • C the scavenged fraction of BC is about 10 %.The fraction increases with temperature to about 60 % at temperatures just below the freezing point (Cozic et al., 2007).In the case of mixedphase clouds, precipitation can form through the Wegener-Bergeron-Findeisen process or through riming.In the former process, ice crystals grow by vapour deposition at the expense of the water droplets, which will have the same effect on the scavenging efficiency as cold cloud precipitation, i.e. low efficiency.Riming, on the other hand, yields much higher efficiency because ice crystals that are formed in the cloud will collide with the super-cooled droplets and the cloud condensation nuclei that formed the droplet will be scavenged in the process.As suggested by Cozic et al. (2007), the observed increase in scavenging efficiency with increasing temperature is consistent with a shift from non-rimed to rimed hydrometeors in mixed-phase clouds.If there has been a shift over time in the temperature at which precipitation is formed above Holtedahlfonna, this could have resulted in a corresponding shift in the scavenging efficiency of BC.
Riming causes a preferential separation in snow chemistry as snowflakes scavenge nitrate (NO − 3 ) more efficiently than sulfate (SO 2− 4 ) (Raynor and Hayes, 1983).Consequently, the nitrate to sulfate ratio is different in rimed as opposed to nonrimed snow (Takahashi et al., 1996).This chemical signature was used by Hegg et al. (2011) to explain variation in BC concentration observed in fresh snow.The nitrate and sulfate records have been reported from Holtedahlfonna in Beaudon et al. (2013), but the hypothesis of riming-controlled variation in BC scavenging efficiencies causing discrepancies between atmospheric and ice core EC trends cannot be verified with the available ion data from Holtedahlfonna, due to relocation of the nitrate and sulfate from the original snow layer during summer surface melt (e.g.Moore et al., 2005;Beaudon et al., 2013).
Whereas warmer temperatures could lead to an increased scavenging efficiency of BC in the Arctic, Browse et al. (2012) came to the conclusion that global warming may actually reduce the aerosol concentration and deposition rates in the Arctic.This is because enhanced BC scavenging efficiency en route from the source regions to the Arctic will reduce the amount of BC reaching the Arctic.While precipitation and temperature measurements are missing from the ice core drilling site, observed annual precipitation and temperatures have, in general, increased in Svalbard in recent decades (Førland et al., 2011).Indirect observations of summer melt (see 3.5 below) are an indicator of strong changes at the Holtedahlfonna glacier between the 1970s and 2004.

The linkage between summer melt and the EC concentration and deposition in Holtedahlfonna
In addition to the evident trend in EC values from 1970 to 2004, the ice core record suggests simultaneous changes in the summer melt of the glacier.Results from Beaudon et al. (2013) indicate that the Holtedahlfonna ice core has experienced summer melt throughout the last 300 years, but increasingly since the 1970s and in unprecedented values since 1980 (Fig. 5a).The indicator of summer melt in ice cores is called melt index, and in the ice core from Holtedahlfonna it is referred from log ([Na + ]/[Mg 2+ ]).The index is based on the fact that sodium (Na + ) and magnesium (Mg 2+ ) ions originate from same sources (sea salt) but percolate with different efficiencies in melting snowpacks (Iizuka et al., 2002).
The index has been defined as a good indicator of postdepositional processes of melt and percolation in Svalbard ice cores (Iizuka et al., 2002, Grinsted et al., 2006), with higher values reflecting more melt.Beaudon et al. (2013) suggested the melt index to correspond to observed June-August air temperatures measured at the Svalbard airport in Longyearbyen (29 m a.s.l.) since 1911 (Fig. 5b).Summer temperatures at Svalbard airport and summer melt on Holtedahlfonna were comparably high around 1920-1930 and reached temporary minima in the late 1960s.The summer temperatures started to increase in the late 1960s after the initial cool phase, which was accompanied by a distinct and steady increase in the ice core melt index, reaching unprecedented values during the 1980s.However, the increase in the melt index in 1970-1990 seems disproportionably strong when visually compared to the more variable and less pronounced temperature increase that does not exceed values of the early 1900s even by 2000 (Fig. 5b).
Hence, when including the EC concentration trend in the assessment (Fig. 5c), it seems that there is some evident correspondence in the intensifying summer melt and the EC concentration in the ice core.The records are not completely synchronous, as the melt index seems to lead the EC concentration by several years, but over a decadal perspective the records present similar trends.We compare the melt index to EC concentrations since EC concentrations in snow are more relevant for the melt process than EC deposition.The change in the melt index to a decrease in 1990 is not caused by a decrease in melting but rather the melt index failing in the most recent part of the ice core, where the summer melt was much more extensive, and thus the ions have been washed out from the annual snowpack (Beaudon et al., 2013).The ions have most likely been washed out of the ice core by slow lateral melt water flow at the bottom of the current years' snowpack, above the previous years' melt layer (cf.Pfeffer and Humphrey, 1996).The importance of BC in accelerated glacier melt has been previously discussed, for instance, by Xu et al. (2009) for the Himalayas, where glaciers are retreating faster than what could be expected due to temperature increases alone.The current data supports the suggested process of increased BC (or EC) concentrations accelerating melt in glaciers.Recently Keegan et al. (2014) suggested that particularly warm summers were not invariably sufficient to cause widespread melt in Greenland ice cores during the last two centuries.However, exceptionally warm temperatures combined with BC deposition from Northern Hemisphere forest fires reduced the coring site's albedo below a critical threshold, causing widespread summer melt in regions that do not normally experience summer melt (Keegan et al., 2014).Consequently, it seems that the recorded melt index trend in Holtedahlfonna can be better explained by the increasing measured summer temperatures and the increasing EC concentration trend together than by increasing temperatures alone.
At the same time, the increased melt itself might actually cause the EC concentrations in the ice core to increase, as melt is accompanied by water loss from the glacier surface to the atmosphere and possible runoff.During summer melt, water can be lost from the annual snowpack by evaporation or sublimation; lateral water flow within the snowpack above the previous years' impermeable ice layer (Pfeffer and Humphrey, 1996) after vertical percolation; or, in extreme melt cases, as surface runoff.As EC tends to be hydrophobic, it is left behind within the snowpack during melt, which leads to significantly higher EC concentrations at the snow surface (Conway et al., 1996;Doherty et al., 2010Doherty et al., , 2013;;Sterle et al., 2013), and in strong melt cases it may move downwards through the snowpack but be concentrated above the previous year's melt layer as observed by Xu et al. (2012).This melt amplification may have increased between 1970 and 2004 in the ice core, causing the increased EC concentrations.How-ever, even in the most recent layers of the ice core and its firn there are annual melt layers present which prevent EC penetrating into the snow/firn of the previous year (Beaudon et al., 2013, especially the density data in Figure 4 therein).After 1990 the increased summer melt caused most elusive ions to wash out laterally from the ice core above the previous years' melt layer at the same time as the less elusive EC was left behind.If runoff were to cause significant water losses from the coring site leaving EC behind, this in itself would cause the EC values to increase.However, while exact meteorological measurements from the coring site are missing, these summertime water losses are expected to have a rather small effect on the whole years' snow accumulation (Pfeffer and Humphrey, 1996), especially as the mean annual melt season on Holtedahlfonna is of the order of ca.80 days per year (Rotschky et al., 2011) and most of the melt water refreezes within the annual snow layer.Thus, increased melt is likely to have only a minor impact on the EC concentration trend.The fact that EC deposition, which is not affected by post-depositional processes unless significant water loss occurs, rises in tandem with the concentrations supports the inference that EC concentrations in the ice core are dominated by atmospheric processes, rather than post-depositional processes.
Although post-depositional processes do not seem to be overly significant for the ice core EC trends they may have a significant effect when comparing the results to snow measurements.For instance, at the Holtedahlfonna site, EC concentrations of 1.4, 9.4 and 11.6 µ g L −1 were measured in 2007, 2008and 2009, respectively (Forsström et al., 2013)), which are roughly a magnitude lower than the ice core concentrations.These EC measurements have been made in springtime on surface snow which had most likely not significantly aged, i.e. experienced sublimation and/or melt, and not received summertime dry deposition.The ice core results, on the other hand, are the annual sum of all annual deposition as well as EC enrichment due to water losses and other aging processes throughout the year, and may therefore reach far higher concentrations than springtime snow measurements.

Summary and conclusions
Unexpected increasing EC concentrations and deposition between 1970 and 2004 after a temporary minimum have been recorded from a Svalbard ice core.This trend differs from previous ice core records from Greenland and atmospheric measurements since 1989, which indicate decreasing BC concentrations in the Arctic during the last decades.However, the recorded trend agrees with estimated past global annual anthropogenic BC emissions and increased 20th century BC concentrations recorded in the European Alps glaciers.
The post-1970 EC trend in Holtedahlfonna is most likely explained by atmospheric processes.Of these processes, increased BC scavenging efficiency induced by rising temperatures is the most probable, as it is the only process that can simultaneously increase EC concentrations and deposition.This process may be responsible for the increasing EC deposition trend in the ice core record despite observed decreasing atmospheric concentrations in the Arctic.
Possible explanations for the differences in the recorded ice core BC trends from Greenland and Svalbard are partly different source areas, their different distances to sources, and the different elevations of the sites.Specifically, the post-1970 increase in Svalbard ice core EC values may be affected by northern Russian flaring emissions which do not reach the Greenland ice core sites as effectively.The observed differences between the Greenland and Svalbard ice core records may also be caused by differences between the two analytical methods used.These differences may result from both the operational definition of which BC particle type is quantified by each method and the size-dependent efficiency in the detection of particles.The SP2 method may not incorporate the largest BC particles and the filter-based method may not include the smallest ones.This can result in different concentrations, and, if the size distribution of the particles has varied temporally, in different trends being measured from the Arctic with different methods.
The results indicate that BC trends recorded in different ice cores may be quite different for various reasons.Results from a single spot glacier observation may not be easily extrapolated to a larger area.Therefore, more ice core and other sediment records from the Arctic are needed to better map the BC deposition in the region.Furthermore, comparison with atmospheric measurements and model estimations from the same area indicates that ice core measurements cannot a priori be converted to atmospheric concentrations, as several factors independent of atmospheric concentrations can affect the concentrations and the deposition of BC to an ice core.In general, this may complicate model validation in conjunction with ice core data.
The increasing EC values in Holtedahlfonna between 1970 and 2004 have major implications for the radiative transfer and forcing at the coring site.According to a calculated melt index from Holtedahlfonna, the glacier experienced increasing summer melt from 1970 to 2004.This trend is better explained by the increasing measured summer temperatures and the increasing EC concentration trend together, rather than by increasing temperatures alone.Whether our observed EC concentration and associated glacier summer melt trend is representative beyond the local scale remains to be determined.Consequently, further investigations of past BC values and BC's environmental impacts in the Arctic are necessary.

Figure 1 .
Figure 1.Map indicating the location of the Holtedahlfonna glacier on Svalbard and the geographical location of Svalbard.The circled area in the inset approximately indicates an area with substantial flaring activity in northern Russia.

Figure 2 .
Figure 2. EC concentrations in the Holtedahlfonna ice core during the last 300 years.The black curve represents the concentrations at sample resolution and the blue line the running 10-year averages of samples made with approximately 2-year resolution.The red dots and error bars indicate average EC concentration and the absolute errors of samples from which multiple analyses were performed.
(Fiescherhorn) only extends to the 1940s, making it impossible to compare to our results regarding the most recent trends.The Legrand et al. (2007) (Col du Dôme) results reach 1990 and show a decrease from the 1960s until 1990.On the other hand, the Thevenon et al. (2009) (Colle Gnifetti) results reach 1980 and show increasing BC concentrations towards the end of the record.Interestingly, Lavancy et al. (1999) present results from the same glacier (Colle Gnifetti) and the same elevation until 1975, but as opposed to Thevenon et al. (2007) they indicate a decrease at the end of the record.Generally, these two studies show somewhat varying trends and non-synchronized peaks in the records.

Figure 3 .
Figure 3. EC concentration and deposition compared to the snow accumulation rate in the Holtedahlfonna ice core, and BC concentrations in the Greenland D4 ice core (McConnell et al., 2007).(a) EC concentration (µg L −1 ) and (b) EC deposition (mg m −2 yr −1 ) with 10-year running averages (red).(c) Snow accumulation rate (m w.e.yr −1 ) of Holtedahlfonna.Note that the snow accumulation rate incorporates all processes increasing (precipitation) and decreasing (ablation and runoff) it.(d) Annual BC concentration (ng g −1 ) data (black) from the Greenland D4 ice core (McConnell et al., 2007) with 10-year running averages (red).

Figure 4 .
Figure 4. Global, North American, OECD European and former USSR annual anthropogenic fossil fuel BC emissions (emissions from Bond et al., 2007) compared to measured EC deposition in the Holtedahlfonna ice core (dashed line) between 1850 and 2000.The EC deposition is shown in 10-year averages.

Figure 5 .
Figure 5. Holtedahlfonna melt index (log ([Na + ]/[Mg 2+ ]) (Beaudon et al., 2013) compared to the measured June-August air temperatures at the Svalbard airport between 1911 and 2000 (source: http://climexp.knmi.nl/),and the EC concentration.The melt index (a) and EC concentration (c) are shown in 10-year running averages using linear interpolations in the resolution achieved by the ice samples.The temperatures (b) are shown in yearly summer resolution as well as 10-year running averages (thick red line).

Table 1 .
EC or BC concentrations measured with filter-based thermal, thermal-optical or similar methods from the European Alps and Holtedahlfonna, Svalbard.