Soot on Snow experiment: bidirectional reflectance factor measurements of contaminated snow

In order to quantify the effects of absorbing contaminants on snow, a series of spectral reflectance measurements were conducted. Chimney soot, volcanic sand, and glaciogenic silt were deposited on a natural snow surface in a controlled way as a part of the Soot on Snow (SoS) campaign. The bidirectional reflectance factors of these soiled surfaces and untouched snow were measured using the Finnish Geodetic Institute’s Field Goniospectropolariradiometer, FIGIFIGO. A remarkable feature is the fact that the absorbing contaminants on snow enhanced the metamorphism of snow under strong sunlight in our experiments. Immediately after deposition, the contaminated snow surface appeared darker than the natural snow in all viewing directions, but the absorbing particles sank deep into the snow in minutes. The nadir measurement remained the darkest, but at larger zenith angles, the surface of the contaminated snow changed back to almost as white as clean snow. Thus, for a ground observer the darkening caused by impurities can be completely invisible, overestimating the albedo, but a nadir-observing satellite sees the darkest points, underestimating the albedo. Through a reciprocity argument, we predict that at noon, the albedo perturbation should be lower than in the morning or afternoon. When sunlight stimulates sinking more than melting, the albedo should be higher in the afternoon than in the morning, and vice versa when melting dominates. However, differences in the hydrophobic properties, porosity, clumping, or size of the impurities may cause different results than observed in these measurements.


Introduction
Snowpacks and ice sheets around the globe play a crucial role in the Earth's radiation budget.The albedo of snow depends, among other factors, on its physical properties, such as, for example, snow grain size, shape, packing, topography, and snow thickness.It is usually significantly higher compared to that of other natural surfaces (Peltoniemi et al., 2015).
With the rapidly-growing number of techniques for Earth observation, the accelerating shrinkage of snowpacks and glaciers over the past decades has been confirmed based on dedicated satellite and in situ measurements.In order to reliably interpret these observations and forecast further changes in snow cover, there is a need to increase existing knowledge on the processes affecting the state of snow.Deposition of light-absorbing impurities on the surface of snow decreases its reflectance (Warren and Wiscombe, 1980;Clarke and Noone, 1985;Dumont et al., 2014) and accelerates snowmelt (Bond et al., 2013).
The presence of light-absorbing impurities in the snow may cause dramatic effects even on fresh snow surfaces, lowering their reflectivity below the typical range of 0.7-0.9.Aerosol particles originating from both anthropogenic and natural sources can be transported over very long distances.
The reflectance of natural snow has been modelled by many, using various combinations of radiative transfer, raytracing, and electromagnetic techniques (Wiscombe and Warren, 1980;Aoki et al., 2000;Tanikawa et al., 2006;Peltoniemi, 2007;Lyapustin et al., 2010;Räisänen et al., 2015).Snow contaminated with impurities has been modelled for example by Warren and Wiscombe (1980); Flanner et al. (2007); Kokhanovsky (2013).These models can usually provide a first-order approximation for the distribution of the reflected radiation and effects of impurities.However, all the models contain uncertainties in modelling the shape of the snow grains, their size distributions, three-dimensional structure of the snowpack, and the distribution of impurities.These factors can change the albedo by several percent, and at certain viewing directions, the reflectance may change more than 10 % (Peltoniemi, 2007).None of the models have yet been able to explain the observed polarization features (Peltoniemi et al., 2010b).
The Finnish Meteorological Institute (FMI) organized a series of experiments, using a different approach than the cited studies (Meinander et al., 2014;Svensson et al., 2015).Experiments were conducted in different regions in Finland, in 2011 and 2013, with the general aim to monitor and quantify the effects of soot on the albedo and physical properties of the snowpack.To this end, absorbing aerosols of different origins were deposited on a natural snowpack in a controlled way and the changes in the snowpack properties were measured (Meinander et al., 2014;Svensson et al., 2015).In this Definition of the angles used in surface reflectance work: and ι are the zenith angles of the emergent (observer) and incident (solar) radiation respectively (shorthand µ = cos and µ 0 = cos ι are also used).φ and φ 0 are the corresponding azimuth angles.The phase angle α is the angle between the observer and the Sun.The principal plane is fixed by the solar direction and the surface normal, while the cross plane is a vertical plane perpendicular to the principal plane.
paper we describe the measurements of the bidirectional reflectance factor (BRF) used to quantify the effects of different absorbing materials on snow using the Finnish Geodetic Institute's Field Goniospectropolariradiometer, FIGIFIGO.

Methods and instruments
The bidirectional reflectance factor (BRF for short, or R in equations) is defined as the ratio of the reflected light intensity of a given target to an ideal Lambertian reflector with a spherical albedo of 1.0 under the same incident irradiation (Nicodemus et al., 1977;Hapke, 1993).The BRF can be presented as where F 0 is the incident collimated flux and I the reflected radiance; ι and φ 0 are the zenith and azimuth angles of incidence, and φ are the zenith and azimuth angles of emergence, α is the phase angle that is defined as the angle between the source and observer equalling the complement of the scattering angle (cos α = cos ι cos + sin ι sin cos(φ − φ 0 )), and µ = cos , µ 0 = cos ι (Fig. 1).A related quantity is the bidirectional reflectance distribution function (BRDF), here denoted as R = R/π.Experimentalists and practical users prefer using BRF for its more intuitive magnitude which is normalized to a perfect Lambertian reflector R = 1.Modellers prefer BRDF for its more natural mathematical interpretation and simpler equations.For example reflected ra-diation can be written with R as where I Diff represents diffuse skylight.
For the detailed multi-angular reflectance measurements conducted within the experiment we have used the FIGI-FIGO, Finnish Geodetic Institute's Field Goniospectropolariradiometer (Peltoniemi et al., 2009(Peltoniemi et al., , 2010a(Peltoniemi et al., , b, 2014;;Suomalainen et al., 2009;Hakala et al., 2010Hakala et al., , 2014)), shown in Fig. 2. The primary sensor of the FIGIFIGO is an ASD Field-Spec Pro FR Spectroradiometer (Analytical Spectral Devices Inc.) with a spectral range of 350-2500 nm, full width at half maximum of 3 nm from 350 to 1000 nm, and 10 nm from 1000 to 2500 nm.The spectrometer is connected to the front optics with an optical cable of a length of 3 m.The optics is interchangeable, including normal optics with a 3 • field of view, and optics with a rotatable Glan-Thompson linear polarizer.
The optics are placed on top of a 1.5-2.5 m long telescope arm.At an arm length of 2.5 m the optics' footprint diameter is 20 cm.The arm is equipped with an inclinometer and a motor, which allows changing of the view zenith direction by tilting the arm.The view azimuth angle is selected by turning the whole device manually.In the field conditions the orientation relative to the Sun is measured with a fish-eye camera fixed to the FIGIFIGO frame, and the direction of the Sun is calculated based on GPS positioning and time.The fish-eye camera also saves images of the sky periodically for postprocessing sky monitoring purposes.Solar irradiance is additionally monitored by a silicon pyranometer (SP Lite, Kipp and Zonen) during the measurement.
The FIGIFIGO field measurement process was carried out as follows.
1.The target was selected and prepared (see next section for details).
2. The system was mounted and started.At least 20 min was allowed for the spectrometer to warm up, in most cases more.
3. The sensor was optimized and a reference measurement taken using a white 25 cm Spectralon ® panel by Labsphere with a nominal reflectance of 99 %.The panel was carefully levelled, and its cleanness checked.Any loose dust or water drops were removed by compressed air.
4. The contribution of diffuse light from the sky, clouds, and the environment (M D in Eq. (3) below) was measured by shadowing the target from direct sunlight using a small screen.5.Then, the target reflectance was measured, by automatically turning the zenith arm and manual azimuthal rotations.
6.The Spectralon ® white reference was remeasured several times during the measurements and at the end of the sequence, depending on illumination stability.
7. The documentation was completed, the data file was closed, and the system moved to the next target or dismounted.
In laboratory the diffuse part was not needed, but one more step to set-up the illumination with a 1000 W QTH (quartz tungsten halogen) lamp by Oriel was required.The data were checked and erratic data removed.Unstable data were marked as unstable.The spectral measurement was normalized with white reference data and the diffuse part subtracted.
For unpolarized measurements, the data renormalization is www.the-cryosphere.net/9/2323/2015/The Cryosphere, 9, 2323-2337, 2015 performed as follows: where M is the measurement of the target, M D the estimate for the diffuse part, S the measurement of the Spectralon ® standard, S D the diffuse part of that, and R ref the laboratorymeasured value of the reflectance of the Spectralon ® (Peltoniemi et al., 2014).Because the reference values are not measured simultaneously for each measurement, but only at the start and end (if possible), the reference values were interpolated from the nearest points based on time.The diffuse light estimate is based on nadir measurement only, yielding a 1-5 % uncertainty concentrated in the blue end.With polarization the procedure involves more phases (see, e.g.Peltoniemi et al., 2015), but here polarization data are not shown.

Samples
The major contributions to the absorption of light in nature originate from volcanic sand, glaciogenic silt, and natural and anthropogenic sources of black and brown carbon.For the purposes of our experiment, the components described below were primarily selected.See also Table 1 and Fig. 3.

Volcanic sand
Volcanic sand was collected from the Mýrdalssandur dust source in southern Iceland in November 2012.This material has been mixed by aeolian processes, resulting in an enriched sand proportion as the silt-sized material is lost as dust along the way.The sample is a near-black mixture of volcanic ash of glaciofluvial nature, originating from under the Mýrdalsjökull glacier, which may be mixed with the ash of the Eyjafjallajökull eruption in 2010 and the Grimsvötn eruption in 2011.It represents well the material re-suspended in the most active dust source in southern Iceland and deposited on glaciers or snow in south and south-east Iceland.Atmospheric dust in Iceland comes from two different dust sources.The first are the extensive sandy deserts.Many of these produce little dust per unit area (very variable because of many types of sandy surfaces), but due to their extent, they are an important dust source.The second main source of dust from Iceland are dust plume areas with high dust productivity per unit area, located mostly near glaciers and along glacial river beds.The volcanic dust is mostly made of basaltic to andesitic poorly crystallized glass particles, dark to black in colour.The density is about 0.9 g cm −3 .

Glaciogenic silt
The sample was collected from the glacial river Múlakvisl, about 10 km from the Mýrdalsjökull glacier in southern Ice- land, one of the main rivers draining the glacier, and the materials originates from the Katla volcanic system under the glacier.The glaciogenic silt is brighter in colour than the sand.It is light-brown to slightly yellowish in colour and it consists mainly of silt and some coarse clay-sized particles, which are easily re-suspended on a daily basis on dry days.This material can be transported and deposited on the local glaciers as well as being transported by wind several hundreds of kilometres towards Europe.The density is about 1.2 g cm −3 .

Chimney soot
Soot was acquired from a chimney-sweeping company in Helsinki, which collected the soot from residential buildings that use small-scale wood and oil burning for heating.The soot was blown onto the snow surface as explained by Svensson et al. (2015).Here the soot was used as a proxy of black carbon and also contained some brown carbon.

Contaminated snow in Sodankylä
The main experiment series of the SoS 2013 campaign took place in Sodankylä Airport, northern Finland, where specific set-ups had been build already before snow fall.In order to control the absorption in snow, the above described contaminants were deposited on snow in various ways.The main experiment and the soot depositing system are described in Svensson et al. (2015).In short, the soot was fed to a cyclone separating smaller particles to be blown into a closed chamber around the target snow where the soot slowly deposited onto the snow surface.Only one preliminary test spot made this way could be measured by FIGIFIGO.However, because the sampling area of FIGIFIGO is much smaller than that of the albedometers, it was possible to use different techniques.
For most FIGIFIGO measurements, the particles were distributed manually over an area of 0.5 m 2 using a salt shaker filled with a measured amount of soot.
On the first measurement day of the campaign, 3 April 2013, a clean and smooth snow spot was selected, and its BRF was measured.After that, 8 g black volcanic sand was spread manually on the snow over about 0.5 m 2 , and the BRF measurements were acquired as quickly as possible.The measurement was repeated in the next spot similarly with 8 g of ash spread over 0.5 m 2 .The properties of the clean snow nearby are shown in Tables 2 and 3, with some photographs in Fig. 4.
During the same day (3 April), the primary experiment made its first large test spot using chimney soot.At night, a thin layer of new snow fell over the site, and all the target spots were also covered.In the morning the sky cleared, and a modest wind partially cleaned the fallen snow.The BRF of this sooted sample was measured four times on the 4 April, at different solar zenith angles, also with polarization, examining the metamorphism process.After soot was spread on the spot surface, clean untouched snow was measured nearby as a reference.Some profile pictures are shown in Fig. 5.
The measurement continued by depositing 10 g of glaciogenic silt over an area of 0.5 m 2 of clean snow.Again some sinking happened, but the measurement was fast and the results are considered useful.After some time, another 10 g was spread over the same spot to make it darker, and the target was measured again.
The sample remained untouched during the night, and the next morning it was measured again.This day (5 April) was already more cloudy than the preceding ones, adding too much uncertainty to perform low concentration measurements.Thus we deposited another 20 g silt over the same sample, and measured the reflectance, followed by a clean snow sample taken for reference, and again a new sample of snow with 40 g of silt deposited over an area of 0.5 m 2 .The properties of the clean snow nearby are shown in Tables 4  and 5, with some photographs in Fig. 6.
The air temperature varied between −20 • C at night and +5 • C at the warmest part of the day.The clean snow temperature was several degrees below zero, from −8 to −3 • C near the surface and around −2 • C near the bottom.Figure 7 shows the snow and air temperatures measured at a nearby swamp by an FMI weather station.The actual temperatures at the airport site may differ a small amount.

Results
The key results are shown in Figs. 9 to 12. Most of the features are shown in three different plots of selected measurements.On the left, the BRF is shown as a function of observer zenith and azimuth angles, integrated over the full measurement range weighed with a fixed solar spectrum; it gives a good overview of the reflected energy distribution.
In the middle, the principal plane BRF curve as a function of the observer zenith angle gives more details of the directional effects, here shown in six different wavelength bands: violet 443 nm, green 565 nm, red 670 nm, NIR (nearinfrared) 865 nm, SWIR (short-wave infrared) 1250 nm, and SWIR 1555 nm, each with a bandwidth of 20 nm in visual wavelengths and 50 nm in IR.On the right, the spectral albedo integrated over all directions using semi-polynomial inter/extrapolation is compared with the nadir spectrum.Full data with many different plots can be downloaded on our website 1 .The first set (Fig. 9) shows the 3 April 11:33 EET (Eastern European Time) measurement of clean snow, the 1 All the measurement data are stored in FGI's Reflectance Library and are available at https://webdisk.kotisivut.com/fgi/Reflectance_Library/2013SoS/.Ask the authors for the password to access the data.12:25 measurement of the snow contaminated by 8 g volcanic sand per 0.5 m 2 , then the 13:22 measurement of 8 g volcanic sand per 0.5 m 2 , and lastly, the pure volcanic sand sample measured in the laboratory.The second set (Fig. 10) shows the measurements from a sooted snow sample on 4 April at 09:56, and at 11:38, natural snow at 13:31, and  the laboratory measurement of pure soot.The 12:39 results are similar to the 11:18 measurements, and are not shown.During the 10:31 measurement, changes were too large to obtain a full data set.The third set (Fig. 11) shows the ef-fects of fine silt on snow, first 10 g of silt per 0.5 m 2 on the 4 April at 15:05, then 40 g of silt per 0.5 m 2 at 16:29, the third same sample on the 5th of April at 11:16, and lastly, the pure silt measured in the laboratory.In the laboratory, about 1 cm thick layer of the material was spread evenly on a black surface.
To demonstrate the fast metamorphism and sinking effects, Fig. 12 shows data from three measurements, first the principal plane BRF taken at the beginning of the measurements, second the principal plane BRF at the end of the measurements, and finally a time plot of all nadir BRF during one measurement's sequence.The measurements of clean snow on the 5 April started at 15:08, snow with 40 g of silt on the 5 April at 14:06, and the same 40 g of silt at a larger zenith angle started at 16:55.

Discussion
The BRF of snow has been well characterized by many authors, see e.g. a review by Peltoniemi et al. (2010b).The spectral shape is strongly influenced by the grain size, and weakly by the presence of liquid water, impurities, and the surface below.Directionally, snow is a forward scatterer, with a weak bowl shape in the visible band, and a much deeper bowl in the NIR bands.The directional dependence is most related to the grain shape and topography.Rough snow surfaces scatter more backwards and less sideways than smooth surfaces.Also the irregularity of grains increases the backscattering slightly.The present results do not contradict this.Below, the new effects of impurities are analysed in more detail.
Typically, contaminated snow darkens the most in the nadir, and the least in the forward direction.Especially, when the black contaminants sink deeper into the snow, the bowl shape of the BRF is enhanced, and from larger zenith angles the snow looks brighter.Thus, from the normal human "feeton-the-ground" perspective, the differences between contaminated and clean snow almost disappear, while from the nadir the contaminated snow can still be almost black.The directional effects of contaminants are clearly distinguishable from grain size, shape, and topography effects only when the concentration of impurities is high and the surface is visibly darker.At more natural < ppm concentrations the directional effect is probably in the limit of significance.The strong forward spike in the 4 April 09:56 measurement (Fig. 10, top) is typical for lower solar angles (here 68 • ), but may have been enhanced by a tiny amount of fresh snow.Another angular effect may be seen between the 4 April 16:30 measurement, with a starting solar zenith angle of 72 • , and the 5 April 11:56 measurement, starting at 61 • .At the solar zenith angle of 72 • , the albedo appears clearly brighter than at 61 • .Although this may well be only a result of metamorphism during the night, it also fits the pattern that high sun encounters more dark impurities than low sun, if the heating sinks dark material downwards.
Spectrally, the contaminants darken the snow mostly in the visual bands.The soot, silt, and volcanic sand used here have a smooth (grey) spectrum without significant features.The volcanic sand is darker than snow in all wavelengths (Fig. 9), while the silt may be even brighter in the deepest absorption bands of snow around 1500 and 2000 nm.As expected, at these dark bands the silt-contaminated snow is brighter than the cleaner snow (Fig. 11).Thus, a spectral signal for impurity inversion exists, but care is needed to separate it from the grain size effect.
From the data, especially Fig. 12, one can see that the clean snow also varies.The reflectance can change by ±10 % in a short time, due to metamorphism.It is not possible to say how clean the clean snow really is.In any case it is too clean to observe the impurities, but it is quite likely that some of the deposited dust may have landed on the clean snow samples, and all possible aerosol and human and animal traces can cause unknown perturbations.An important observation is, however, that adding absorbing impurities enhance melting and metamorphism remarkably.When heated by the sun, the small particles can melt the snow around them, letting them fall down or float up at least several centimetres, depending on their physical properties.This process happens in minutes, and it is thus impossible to say exactly what the state of a sample was upon measuring.

Conclusions
A field experiment was made to study the effects of different kinds of impurities on the reflectance of snow.Soot collected from chimneys, volcanic sand, and glaciogenic silt were deposited on natural snow surfaces.This simulates both anthropogenic and natural dust sources, both being of importance (Dagsson-Waldhauserova et al., 2015).The bidirectional reflectance factor (BRF) was measured using the Finnish Geodetic Institute field Goniospectropolariradiometer FIGIFIGO.
Impurities make snow darker.However, the concentrations must be rather significant or How the reflectance changes during the measurements.Left plots show the reflectance in the principal plane measured in the beginning of a sequence, and the plots in the middle show the same, but measured at the end of the sequence.On the right the reflectance in nadir is shown, taken at different time steps.The top shows untouched snow, the middle shows snow with fine silt 40 g/0.5 m 2 , taken at 62 • , and the bottom plots show snow samples taken at 74 • .Untouched snow darkens slightly at nadir and brightens at large zenith angles, while contaminated snow brightens clearly in all directions, especially at large zenith angles.ations, such as grain size, surface roughness, or snowpack structures, from reflectance data.Typical natural concentrations of black carbon are less than 1 ppm, which cannot be detected from optical satellite data without additional information, as already pointed out by Warren (2013).
Snow contaminated with impurities is unstable.When the Sun heats the absorbing particles, they melt or soften the ice around them, allowing the particles to move inside the snow.In this experiment, the particles sank down, leaving the topmost (on a millimetre to centimetre scale) surface whiter than expected.Snow grain size, shape, density, and surface roughness also changed visually.These metamorphism and sinking processes occur so fast that it is difficult to link successive measurements to each other, and to know what was really measured.
After the sinking, the difference between contaminated and clean snow is largest from nadir, where one can still see the dark contaminants through the sink holes, and smallest at large zenith angles, where one sees mostly pure snow.Also the solar zenith angle has a significant effect on what is seen.At small solar zenith angles light goes deeper inside and may interact more with sunk-in impurities than at large zenith angles.
However, different melting conditions, or different hydrophobic properties of the contaminants, may reverse the process, accumulating the dirt on the surface (Conway et al., 1996;Meinander et al., 2014).Such snow was not measured here, and must be studied more before wide conclusions are drawn.But in both cases, models assuming a homogeneous distribution of absorbers may over-or underestimate the effect of impurities on the albedo and climate quite significantly.
This experiment aims to fill the gap between laboratory experiments (where one can control the target and measurements optimally) and natural observations (where one measures the targets as they are).In all, we conclude that for modelling snowmelt and spectral albedo affected by lightabsorbing impurities, more experimental results are needed.

Figure 2 .
Figure 2. FIGIFIGO at the SoS test site at Sodankylä Airport.A clean snow sample has been measured, and the last reference measurement of the white Spectralon ® panel is being taken.Deep and soft snow complicated the usability of FIGIFIGO.While the movements in the snow left significant traces, the target area under the instrument footprint was carefully kept untouched during the measurement.

Figure 3 .
Figure 3. Photographs of the samples.On the left is an overview; the right shows some details.Starting from the top photographs, clean snow on 4 April 2013 is shown, snow deposited with volcanic ash on 3 April 2013, snow deposited with chimney soot on 4 April 2013, and on the bottom, snow deposited with glaciogenic silt on 4 April 2013 is shown.

Figure 4 .
Figure 4.More snow profiles taken on the 3 April of unprocessed snow near the FIGIFIGO spot.Because of near-zero temperature, the snow crystals melt fast on the black plate, and the smallest details and grains have already disappeared.

Figure 5 .
Figure 5.More snow profiles taken on the 4 April of unprocessed snow near the FIGIFIGO spot, plus one profile from the sample sooted on the 3 April, and measured on the 4 April.

Figure 6 .Figure 7 .
Figure 6.More snow profiles taken on the 5 April of unprocessed snow near the FIGIFIGO spot.

Figure 8 .
Figure 8.The SoS experiment system at Sodankylä Airport.Top photo by Rae Ellen Bichell.
Figure12.How the reflectance changes during the measurements.Left plots show the reflectance in the principal plane measured in the beginning of a sequence, and the plots in the middle show the same, but measured at the end of the sequence.On the right the reflectance in nadir is shown, taken at different time steps.The top shows untouched snow, the middle shows snow with fine silt 40 g/0.5 m 2 , taken at 62 • , and the bottom plots show snow samples taken at 74 • .Untouched snow darkens slightly at nadir and brightens at large zenith angles, while contaminated snow brightens clearly in all directions, especially at large zenith angles.

Table 1 .
List of the samples measured using FIGIFIGO during the SoS campaign at Sodankylä Airport.Numbers in parentheses are from unstable data; double parentheses denote very unstable data, due to increasing cloudiness.ID means the name by which the measurement is found in the data base, SZA is the solar zenith angle, and PAR alb is the albedo integrated over the photosynthetically active range.

Table 4 .
Snow depth profile measured from a snow pit on 5 April 2013.