In 2011, experiments were undertaken from early March until April on a farming field (60∘24′ N, 24∘42′ E) near the town of Nurmijärvi, 30 km north of Helsinki, Finland. When the experiment commenced, the snowpack thickness was 50 cm and winter conditions with subzero temperatures prevailed in the area. The second experiment was conducted at the FMI observatory (60∘48′ N, 23∘30′ E) in Jokioinen, southern Finland, ∼100 km northwest of Helsinki, in February and March. At the start of the experiments, the snow depth at the site was 30 cm. The third experiment took place at the Sodankylä airfield (67∘23′ N, 26∘36′ E), located near the FMI Sodankylä observatory, in northern Finland, in April and May 2013. The snow depth at the experimental site was 65 cm before the soot deposition started. Hereafter the experiments in 2011, 2012 and 2013 will be referred to as SoS2011, SoS2012, and SoS2013, respectively.

Soot was deposited with different methods onto the snow surface during the experiments. In SoS2011, soot particles were produced by burning various organic materials (wood and rubber pellets from used tires) in a wood-burning stove. The smoke was lead through a pipe, cooled by snow surrounding the pipe, and lead into a rectangular chamber (3.3m×7.5m×2.8 m; width, length, height, respectively) situated on top of the snow. With this in situ production of soot, it was difficult to cool the air containing the particles before deposition and some particles were not deposited in the desired rectangular chamber. The temperature of the surface snow inside the chamber remained below freezing (visible in Fig. 1), hence, melting did not occur during the soot deposition. The deposited soot particles inside the chamber were deposited in a heterogeneous pattern (Fig. 1). An undisturbed reference site, with no impurities added, was established in close proximity (15 m) to the experimental chamber.

Because of the temperature gradient and the heterogeneous distribution pattern of the deposited soot particles in SoS2011, a different approach to deposit the soot was taken in SoS2012 and SoS2013. Soot was acquired beforehand from a chimney-sweeping company (Consti Talotekniikka) in Helsinki, which collected the soot from residential buildings with small-scale wood burning. The soot was blown into an in-house made cylindrical chamber (diameter of 4 m) carefully installed on top of the snow. The blowing system consisted of a blower, a tube blowing air into a barrel filled with the soot, and a cyclone removing particles larger than about 3 µm (the cyclone was changed between SoS2012 and SoS2013). Since the flow was not kept constant the removal of larger particles with the cyclone was only achieved with moderate success. Electron-microscopy images of particles sampled in SoS2013 after their deposition onto the snow indicate the presence of particles larger than 3 µm, most of which were some form of agglomeration consisting of carbon and some other elemental species. Several spots were made in SoS2012 and SoS2013 with varying amounts of soot. Following the deposition of soot in SoS2012, the spots were covered with 10 cm of new snow soon after the measurements began. After the snowfall (an event which included high winds as well) all of the spots had very similar albedos and the melting time of the snow depended mostly on the amount of snow in each spot. The soot analysis of the snow samples revealed that samples collected one month later contained significantly less soot (a factor of four) compared to the snow samples collected right after soot deposition. We hypothesize that the snow storm removed a large fraction of the top layer, which contained the deposited soot, and therefore no clear effects of soot on snow were observed. Any further results or analysis related to SoS2012 data will not be presented in much detail in this paper. In SoS2013 the reference spot at the airfield got contaminated during the soot deposition in the afternoon of 8 April. A new reference site, inside the airfield, was therefore created for the post depositional monitoring.

The soot used in SoS2012 and SoS2013 was black with a grey character to it and therefore cannot be considered to be only containing BC particles. In SoS2012 the size distribution inside the chamber was measured with a differential mobility particle sizer (DMPS, size range 0.015–0.92 µm) and an optical particle sizer (OPS, size range 0.3–10.1 µm). It is noted that the instruments measure diameters which are defined in a different way (due to the physically different principles used for particle sizing), leading to some inconsistency in the overlapping size range. The results, however, show the occurrence of several size modes, the highest ones around 0.5–1 µm diameter. The normalized (to Ntot) size distributions did not vary much, especially in the case of OPS data. The same peaks occurred in the median values for each deposition period, as well as in the 25th and 75th percentile values.

A single particle soot photometer (SP2, Droplet Measurement Technologies, Boulder, Colorado) was also used during SoS2012. The SP2 uses a laser-incandescence method to measure the mass of individual refractory BC (rBC) from each particle (Schwarz et al., 2006; Stephens et al., 2003). It has a size range of 70–500 nm (mass-equivalent diameter assuming BC density of 1.8 gcm-3), and has the capability to measure the size distribution, as well as the coating thickness of the BC particles, through the leading edge optimization method (e.g. Gao et al., 2007). The majority of the particles were coated, with an average coated to core diameter ratio of 1.5. For comparison, coated to core diameter ratios larger than 2 have been observed in aged forest fire plumes (Dahlkötter et al., 2014).

The soot characterization from SoS2012 was used for the SoS2013. The size distribution might differ slightly, as the blowing system was somewhat modified for SoS2013, with a more efficient cyclone removing the large particles. The only expected change would be that the largest particles were not measured.

Our experimental data were compared with the SNICAR model (online version: http://snow.engin.umich.edu/; Flanner et al., 2007). The SNICAR model simulates the broadband albedo of snow for different combinations of impurities (including BC, dust and volcanic ash), snow physical properties, and incident solar fluxes. SNICAR assumes snow grains and impurities have spherical shape and are sparsely distributed, ignoring thus polarization, 3-D structure of the snow pack and shapes of the grains, giving expected accuracy of ±10 %. The model was originally developed and used by Flanner et al. (2007) to estimate the global radiative effect of BC on snow, and has since then been used numerous times in other different studies of BC in snow and ice (e.g. Hadley and Kirchstetter, 2012; Kaspari et al., 2011; McConnell et al., 2007; Meinander et al., 2013; Sterle et al., 2013).

The modeled SNICAR albedo was compared to our ambient measured albedo and corresponding EC concentrations measured following the soot deposition (within a couple of hours). This comparison provides an overview of the albedo reduction caused by the soot on the snowpack immediately following soot deposition before any snow processes (including metamorphism) have started to take place. Hence, it does not provide an evaluation of the SNICAR during conditions of melt.

For input values in the SNICAR model, we used direct incident radiation and an estimated solar zenith angle (SZA; for SoS2013 61.27). The surface spectral distribution was classified as mid-latitude winter, clear-sky. The input for snow grain effective radius was derived from the measured SSA, where the SSA was converted to optical equivalent grain radius (OEGR), using the theoretical relation between OEGR and SSA (Kokhanovsky and Zege, 2004; Legagneux et al., 2002): OEGR=6ρiSSA where ρi is the density of ice. The OEGR has been suggested to closely match the grain effective radius while the conversion is sufficient for studies of hemispheric reflectance (Grenfell and Warren, 1999; Grenfell et al., 2005; Neshyba et al., 2003). For SoS2013, the conversion yielded an OEGR of 270 µm for the surface layer and 320 µm for the average of the 0–5 cm snow layer. Further, we also ran the SNICAR-model with an effective grain radius of 750 µm, which was the visual maximum grain size (for the surface slayer), to have an upper estimate of the snow grain size measurements. The snowpack thickness was set to 65 cm, while the snow density was set to 200 kgm-3. The visible and near-infrared albedos of the underlying ground were set to 0.1 and 0.3, respectively. The BC concentrations (using both the uncoated and coated BC alternative) were varied over a wide range of concentrations to imitate the wide range of EC concentrations used in our experiments. Lastly, the MAC (mass absorption cross-section) scaling factor was varied between 1.0 (corresponding to a MAC of 7.5 m2g-1 at 550 nm) and 0.3 (equivalent to a MAC of 2.25 m2g-1 at 550 nm). The lower MAC is a hypothetical number that was based on electron-microscopy images of the soot particles extracted from the snow samples, revealing the presence of many large soot particles (>1 µm) in the snow. The size distribution from the soot characterization in SoS2012 also revealed the presence of larger sized particles which would result in a lower MAC (Bond and Bergstrom, 2005). Model runs were also made with input parameters from SoS2011, although accurate snow grain size measurements were not conducted immediately after soot deposition. Using effective grain sizes similar to those observed during SoS2013, the SoS2011 model runs did not differ much from SoS2013.

The EC concentrations in the surface snow samples in SoS2011 varied between 120–29 000 ppb. Samples were gathered at several locations throughout the rectangular sooted snow area, and the wide range reflects the heterogeneous deposition pattern of soot inside the rectangular chamber used in the experiment. The clean reference spot had an EC surface concentration of 80±31 ppb. For SoS2013, the EC concentrations from the different spots also had a wide range, 230–6400 ppb. Within each spot, at least three snow samples were collected from different parts of the spot to observe the variation within the spot. For four of the spots, a coefficient of variation between 20 and 40 % (see Table 1) was measured. Some spots had a more homogenous deposition pattern, indicated by a variation of 8 % or lower (two spots, Table 1). The small-scale horizontal variation of EC particles within the spots is similar to what has been observed in natural conditions (Doherty et al., 2010; Forsström et al., 2013; Sterle et al., 2013; Svensson et al., 2013).

The results from the two spots that were sampled at a later stage (9 and 16 days later) are presented in Table 2. The initial average EC surface snow concentration at spot 7 was 1465±26 ppb. After 9 days, the surface snow (5 cm sampling interval) contained an EC concentration of 529 ppb, accounting for 36 % of the initial EC concentration. The surface layer contained 71 % of the total EC measured for the whole snow column at that time. The majority of the remaining soot particles had at that point percolated down to 15 cm or more below the snow surface. At the last sampling (16 days after the initial soot deposition) the surface snow contained 859 ppb of EC, while the bottom snow layer contained 475 ppb. Hence, about 64 % of the total EC concentration for the snow column at this time was in the surface layer. The EC in the surface layer accounted for 59 % of the originally deposited soot. The second spot (spot 5) was also sampled at later stages and the results were similar (see Table 2). Nine days after the initial soot deposition, about 43 % of the soot was still in the surface layer, accounting for 69 % of the total EC concentration measured in the whole snow column. Similarly, after 16 days, the EC concentration in the surface snow was 45 % of the original EC concentration.

The temporal progression of EC concentrations during SoS2013 shows that the BC particles scavenging efficiency was less than 100 % with meltwater during snow melt. Ambient measurements show that BC particles and other light-absorbing impurities, such as dust, have a tendency to stick at the snowpack surface during spring melt (e.g. Clarke and Noone, 1985; Conway et al., 1996; Doherty et al., 2010, 2013; Sterle et al., 2013; Svensson et al., 2013). Confined to the top centimeters of the snowpack, the BC concentrations in the snow during the melt have been measured to increase by a factor of 2–7 (Doherty et al., 2013; Sterle et al., 2013). Doherty et al. (2013) found an even higher amplification factor of ∼10–15, when melting snow BC concentrations were compared to the concentrations from earlier in the year at a site near Dye-2, on the Greenland ice sheet.

Due to the chosen set-ups for our experiments, where higher amounts of soot were deposited on the snow surface, it was not expected that the EC concentrations would increase during melt. However, 9 days after deposition, 36 and 43 % (at two different experimental spots) of the initial soot particles were observed at the snow surface. It is also worth noting that in one of the spots (spot N) there was a substantial increase in EC concentration between the 9th and the 16th day, with the EC concentration increasing from 529 to 859 ppb, while the other spot had a marginal increase from 730 to 756 ppb during that time. The reason for this increase is unknown. Conway et al. (1996) reported that the soot concentration (for both the hydrophobic and hydrophilic soot) in the surface snow layer decreased by 2 orders of magnitude over the course of 10 days. Noteworthy though, is that Conway et al. (1996) used amounts of soot that were much higher compared to our EC concentrations.

In SoS2011, the broadband albedos of the contaminated and the reference snow, following the initial soot deposition, were 0.52 and 0.83, respectively. In Fig. 3a, the temporal evolution of the albedo from the contaminated and clean snow is presented as daily averages at solar noon ± one hour. As expected with natural snow aging, the reference site's albedo decreased overall from 0.83 to 0.65 over the first month of observations. During this period, there were a few snow fall events, leading to the fluctuations in the albedo. Following this period, the albedo decreased rapidly until the snowpack had completely melted on 12 April. The fluctuations in the soot-contaminated snow albedo were larger than those for the reference site, with the snow fall events increasing the albedo to similar values as the reference site. Following the snowfalls, the new snow melted faster over the contaminated snow than over the reference snow. After this melt, the contaminated snow layer would be visible again with its lower albedo. Noteworthy is also a large snowfall event on 20 March that covered the albedo measurement devices. The snow accumulated onto the upward-looking sensor hindering the instrument from proper collection of photons, resulting in irradiance values close to zero, and consequently albedo values exceeding the unity.

For SoS2013, the trend in the albedo was similar to that shown in Fig. 3 for SoS2011. Following soot deposition, the albedo was lowest at the spots with the highest amount of soot and vice versa, the albedo was highest at the spots with the lowest amount of soot (see Table 1).

The albedo time series in Fig. 3b show a sharp increase of 0.23 in albedo of the most contaminated spot during the first two measurements days. This increase could be explained by the fact that after deposition the soot particles sunk into the snow surface, within minutes of deposition, as visually observed. The soot particles sunk during the day at elevated solar radiation, and thereafter stopped sinking during the nights when the temperature was well below zero (indicated by the temperature minimum in Fig. 3b). The soot particles probably became stagnant after this initial sinking, and with the snow surface melting, but particles not being washed down with the meltwater, they became closer to the surface, affecting the albedo again. The albedo started to decrease again on the third day of measurements. On 14 April, a snow shower put few centimeters of fresh snow on the snowpack. A distinct increase in albedo was observed between 14 and 15 April. This increase was not as pronounced for the other spots. Similar to the snowfall events in SoS2011, the melting of the fresh snow and decrease in albedo occurred fastest on the spot where the soot concentration was highest. Another event with an increase in the albedo, visible in all spots, occurred on 18 and 19 April. At this time, the precipitation was in the form of rain. The increase in albedo can possibly be due to rain lowering the density of the BC-containing snowpack (Meinander et al., 2014), meaning lower water content in the sooted snow and resulting in optically smaller snow grains with higher albedo, and/or possibly that some soot particles were lost from the surface at this time. Both of these cases would lead to an enhanced albedo of the snowpack. On 22 April, the albedo began to decrease rapidly starting with the spot with the highest BC concentration changing its albedo from 0.55 to 0.15 during 72 h. The spots with lower BC concentrations followed this rapid decrease in albedo a few days later. The temporal variation of the albedo at Spot S9 (second highest EC concentration) and the contaminated reference spot (S9B) were nearly identical during this fast albedo reduction. The albedo decreased earlier at the spot with the lowest EC concentration (S10) than at the spot with the second lowest EC concentration (S8). However, since snow samples were not collected for determination of the EC concentrations for these spots at this time, their concentrations are unknown. Further, it is questionable how much of an impact the BC has on the snowmelt (especially with lower EC concentrations) during this time when the snowpack height was lower than ∼30 cm as one would expect the albedo of the underlying ground to have the strongest effect on snow melt.

During SoS2011, the melting rates of the snow at the reference site and at the soot contaminated spot were 3 and 7 cmday-1, respectively. From the snow pits dug and studied one month after soot deposition in SoS2011, it was evident that the snow containing the soot had experienced further changes than the reference snow (Fig. 2a–d). The soot doped snow had transformed to a more homogenous snowpack containing rounded polycrystal snow grains, whereas the grain shapes of the reference snow were more heterogeneous. Similarly, the hardness test revealed a more uniform pattern in the sooted snow, while the reference site was more diverse. The snow depth for the dirty snowpack was at that time 35 cm, while the clean snow had a depth of 50 cm. Both of the pits had a layer of freshly fallen snow (4 cm deep at the reference site, and 2 cm deep at the sooted snow), containing 0.5 mm snow grains. The remaining snow had a snow grain size of 2 mm, except for the bottom 5 cm at the reference site, where the snow grain size was 1 mm. The snow density was practically the same for the two snow pits. The density varied between 340–400 kgm-3 in the top part of snowpack and was about 460 kgm-3. Since the snowpack had brittle layers and also some very loose layers, it was difficult to conduct the density measurements. Therefore density data were obtained from only a few measurements.

At the time of the soot deposition during SoS2013, the surface layer of the clean reference snow had a depth of 1 cm containing irregular precipitation particles with thereunder several layers with rounded and faceted crystals produced by snow metamorphism (Fig. 2e–g). The average visual grain size varied between 0.5 mm at the surface, while the bottom contained 2.8 mm grain sizes (this was a depth hoar). Four days later, the snow pit work in two of the sooted spots (Spot 5 and Spot 7) revealed that a hard melt-freeze layer had developed near the snow surface, followed by a layer of faceted snow crystals and another melt-freeze layer (Fig. 2f and g). The average surface grain sizes at these spots were estimated to be 0.5 and 1 mm, respectively. In addition, larger aggregates produced by snow grains melting and refreezing together were observed near the surface. However, snow pit work done two days earlier in the nearby mire and forested areas, revealed the same kind of snow stratification indicating that the melt-freeze layers were not only produced by the presence of light absorbing impurities but were caused by the changes in the weather conditions as well. It is, however, hypothesized that the impurities may enhance the development of surface crusts by enhancing the snow melt during the sunny hours, while the air temperature still drops below zero at night time. Eleven days after the first measurements, the snow characteristics of these two sooted spots plus a clean reference snow were measured again. At that time, the snow melt had already started as the temperature of the whole snowpack was 0 ∘C, and snow grain types of rounded melt forms were recorded. The average surface grain size in all three spots was estimated to be 1 mm.

During SoS2013, we observed that the soot containing snow lowered the density of melting snow (Meinander et al., 2014). We also observed that light-absorbing impurities deposited on snow enhance the immediate metamorphosis under strong sunlight (Peltoniemi et al., 2015). After soot deposition, the contaminated snow surface is darker than the pure snow in all viewing directions, but as stated above, we observed the soot particles sinking into the snow, thus increasing its surface roughness.

The SoS experimental data points are compared to SNICAR model simulations of the effect of BC concentrations on the snow albedo as shown in Fig. 4. In addition, the SoS experimental data was fitted with two different logarithmic fits, one for EC concentrations lower than 1 ppm and one for all of the EC concentrations used in the SoS campaigns. For EC concentrations below 1 ppm, the relation could be described as: albedo=-0.046ln⁡(EC)+0.71, whereas for the whole EC range used in SoS could be described as: albedo=-0.087ln⁡(EC)+0.66. This parameterization describes the reduction of the broadband snow albedo in response to increased EC loading, immediately following soot deposition. After 72 h of soot deposition, with no precipitation meanwhile, the parameterization changed only marginally, when applying either EC concentrations below 1 ppm or the entire EC range. The coefficient (and intercept) values were -0.045 (0.71) and -0.073 (0.67), respectively. Thus, without precipitation our parameterization was still valid after 72 h.

The comparison between the SNICAR model results and the SoS data shows that the results agree best when an effective snow grain radius of 270 µm is used in the model. With an effective radius of 750 µm, SNICAR overestimates the albedo reduction (with the exception of the most contaminated spot from SoS2013). Similarly, when the BC-coated option is chosen, the SNICAR model overestimates the albedo reduction.

During SoS2011 the albedo at the reference site (0.83) could not be reproduced with the SNICAR model (with SoS2011 data as input parameters). The SNICAR-simulated albedo varies from 0.78 (MAC-factor 0.3 and grain effective radius 270 µm) to 0.71 (MAC-factor 1 and grain effective radius 750 µm). However, tuning the grain effective radius to 100 µm and a MAC-factor of 0.3 produces an albedo of 0.83. Since accurate snow grain measurements were not conducted in SoS2011, as well as an accurate soot characterization, this is a realistic possibility. Thus, accurate knowledge of the snow grain size is crucial to properly model the snow albedo. Similarly, also in SoS2012, the measured albedo (0.92) could not be reproduced with SNICAR with our input parameters. The reason for this discrepancy is unknown.

The reductions of modelled albedo with the SNICAR model reflect the conditions that exist right after soot deposition in these experiments. The temporal variation, and the associated snow processes affected by the presence of BC in the snow are not supported by these simulations. Here the focus was on dry deposition of impurities on cold snow surfaces.