3.1 Instrument and platforms

During SIPEX II the albedo measurements were made using a portable scanning spectral radiometer manufactured by Analytical Spectral Devices, Inc. (Reference Kindel, Qu and GoetzKindel and others, 2001). The instrument (ASD) employs three separate grating radiometers that simultaneously measure radiance across three wavelength regions covering the range 350-2500 nm, with spectral resolution 3-10nm, and can average 100 scans in 10s. The ASD is designed specifically to cover the solar spectrum, and has order-blocking and cut-off filters to ensure uncontaminated spectra. A ‘cosine collector’ consisting of a Spectralon diffuser-plate receptor is used to measure down-welling and upwelling incident flux. The ASD and its computer are both battery-powered and housed in separate insulated boxes. The electronics are kept warm by self-heating supplemented by battery-powered heaters.

The ASD is contained in a sled and is connected by a fiberoptic cable to the cosine collector located on a 3 m metal rod, allowing the cosine collector to be projected out over thin ice some distance from the ASD. The rod contains a counterweight on the end closest to the ASD, and a tripod is used to steady the rod. Bubble levels on both inboard and outboard ends of the rod facilitate leveling of the diffuser plate. The diffuser plate is rotated nine times, alternately up toward the sky and down toward the surface, to measure five downwelling and four upwelling spectra. Because the downwelling radiation flux may change slightly with time as cloud transmittance changes, two upwelling spectra bracketing a downwelling spectrum are interpolated to the time the downwelling spectrum was made; then the interpolated upwelling flux is divided by the downwelling flux to obtain the albedo. Four albedo spectra are thus obtained, which usually would be nearly identical; they are then averaged to form a represent albedo spectrum for that surface type. Figure 2 shows photographs of the ASD in use.

Fig. 2. Photographs showing ASD spectral radiometer. The ASD is contained in the sled (seen in left photograph) and is connected to a cosine collector on the far end of the metal rod by a fiber-optic cable. The rod contains a counterweight on the end closest to the ASD, and a tripod is used to steady the rod. The rod can be rotated to position the diffuser plate downward towards the ice surface and upwards towards the sky. The ASD is used here to measure the albedo of a dense frost-flower field covering nilas (left, 4 October 2012) and sparse frost-flower field covering nilas (right, 28 September 2012).

For albedo measurements of snow-covered first-year sea ice, the tripod legs were pushed into the snow so that the diffuser plate was suspended 70 cm above the surface. All other ice types were too thin to support our weight, so we operated either from an inflatable rubber boat (IRB) or from the edge of a thick ice floe. To measure thin sea ice in a narrow lead, we dug away the snow covering the thick sea ice and then placed the sled and tripod down onto the bare ice surface adjacent to the lead; the cosine collector was then extended out over the thin ice. If the tripod had instead rested on the snow, the cosine collector would have been considerably higher than the desired 70 cm height above the surface to be measured. When using the IRB, the sled and tripod rested on the IRB base while the rod and cosine collector extended over the side of the IRB.

The choice of 70 cm height for the diffuser plate is a compromise between two conflicting considerations. The diffuser plate receives radiation from all angles in a hemisphere, so when looking down at a narrow lead it receives radiation reflected not only from the target ice, but also from the snow walls of the lead. Moving the diffuser plate down closer to the surface reduces this source of error but also blocks some of the incident sunlight that would otherwise reach the surface. For diffuse incidence (under cloud), geometric analysis by Reference Brandt, Warren and ClarkeBrandt and others (2011) showed that a shadowing correction of 1.7% was needed if the cosine collector was positioned 70 cm above the surface; a larger and more uncertain correction would be applicable for a cosine collector closer to the surface. The height of 70 cm was used for all our measurements, so our raw albedos measured under overcast cloud were all multiplied by the factor 1.017 for the plots of albedo shown in this paper.

Because the cosine collector receives radiation from all directions within a hemisphere, we attempted to position the cosine collector horizontally as far as possible from the edge of the snow-covered first-year sea ice, and the edge of the IRB, thus minimizing the influence of these surfaces on our reported albedos (see Section 3.2.2 for a computation of this error). Our most accurate measurements are made under cloudy-sky conditions, because ‘tilt errors’ due to inaccurate leveling of the cosine collector were minimized. Under clear-sky conditions, we made measurements within 2 hours of solar noon to experience the maximum solar altitude, which likewise minimized the tilt errors. During clear-sky conditions, we also oriented the rod toward the azimuth of the Sun so that the surface we were observing would not be shadowed by the tripod, rod and diffuser plate.

Table 2 lists, for all our albedo measurements, the date, location, weather conditions, surface types, and position of the ASD with respect to the surface being measured. Multiple albedo spectra were measured for each surface type. If the spectra looked similar they were averaged together to create one albedo spectrum and one set of broadband albedos. Sometimes it was necessary to remove an albedo spectrum if the sky conditions were changing too rapidly or if there was an operational error.

Table 2. Description of the ice stations where the albedos of surface types were measured during SIPEX II. The date, location, weather conditions, surface types, and position of the ASD radiometer with respect to the sea ice are provided. Measurements were made either from the edge of a floe or from an IRB. Ship time is coordinated universal time (UTC)+10; Sun time is UTC+8

3.2. Spectral albedos

The albedos of ice and snow are the result of scattering of light at air/ice and ice/water interfaces, and limited by absorption of photons passing through ice between the scattering interfaces. The scattering is nearly independent of wavelength because snow grains, air bubbles and brine inclusions are all much larger than solar wavelengths, so the spectral variation of albedo is determined by the absorption spectrum of ice. The absorption coefficient varies by 7 orders of magnitude across the solar spectrum, from ∼10–³ m^–1 at A = 390 nm to 10⁴ m–¹ at A = 2 m (Reference Warren and BrandtWarren and Brandt, 2008). The absorption length (the reciprocal of the linear absorption coefficient, and equal to the mean free path of a photon traveling through ice before absorption) exceeds 1.5 km in the near-UV but is as small as 0.1 mm at A = 2m in the near-infrared (nIR). Correspondingly, the albedo of deep snow approaches 1.0 in the UV and visible wavelengths, and -0.0 at the centers of vibrational absorption bands at 1.5 and 2.0 m.

Because scattering scales with the interfacial areas, and absorption scales with absorber mass, a good predictor of albedo is the ratio of air/ice interface area to ice mass, given the name ‘specific surface area’ (SSA; e.g. Reference Matzl and SchneebeliMatzl and Schneebeli, 2006, Reference Matzl and Schneebeli2010). The SSA can be converted to an effective radius r _e (e.g. Reference Grenfell and WarrenGrenfell and Warren, 1999):

(1)

where _ρice is the density of pure ice, 917 kg m–³. Larger SSA (smaller r _e) causes more scattering of radiation, decreasing the probability of absorption and increasing the albedo.

An example albedo spectrum is shown in Figure 3. The data measured in wavelength bands 1800-2100 nm and 2300-2500 nm are not plotted since they are so noisy as to be meaningless. At these wavelengths, both the upwelling and downwelling radiation fluxes are very small, close to the noise level of the instrument, particularly under cloud. Since clouds absorb nIR radiation at wavelengths where snow and ice are also absorptive (all three media consist of water and therefore exhibit the vibrational absorption bands of the water molecule), most of the radiation at those wavelengths is prevented from reaching the snow surface when clouds are present. We therefore smoothed the albedo values across these data gaps, paralleling the known behavior of snow albedo across these two wavelength domains (Reference Grenfell, Warren and MullenGrenfell and others, 1994), as indicated by the dashed blue lines in Figure 3. To determine whether the uncertainty in our smoothing procedure alters broadband albedos significantly, we calculated broadband albedos for every albedo spectrum, using the extreme upper and lower bounds shown in Figure 3. The largest change was in the broadband nIR albedo (700-3000 nm wavelength), which was altered by at most 2.3% when these limits were applied; the typical broadband nIR albedo change was ˂1%, and the total solar broadband albedo change was ˂1%. This small sensitivity to extreme assumptions is a consequence of the small fraction of incident solar energy in these two wavelength bands for which the smoothing was applied.

Fig. 3. Frost-flower albedo spectrum measured under overcast cloud on 28 September 2012 (blue line) along with the smoothing profile used during the calculation of broadband albedo (blue dashed line) and extreme upper and lower bounds (red and green dashed lines). See Section 3.2 for more detail.

3.2.1. Snow

Snow-covered first-year ice was the most common ice type encountered in the East Antarctic sea-ice zone. Figure 4 shows albedo spectra for both thin (2–4 cm) and thick (-45 cm) snow cover over first-year ice. The thick-snow albedo spectrum is similar to plots shown for clean snow on the Antarctic plateau at the South Pole by Reference Grenfell, Warren and MullenGrenfell and others (1994, Fig. 4) and at Dome C by Reference Hudson, Warren, Brandt, Grenfell and SixHudson and others (2006, Fig. 6). The two albedo spectra we measured on 8 October were both of ∽ 4 5 cm of snow on top of sea ice that was at least 1 m thick, but the snow surface in places was capped by a 0.2 cm crust, presumably from a melt-freeze cycle which would increase the grain sizes. We measured the albedo of both crusted and crust-free snow. In the UV and visible wavelength regions, the two spectra are identical because ice is so weakly absorptive here that the albedo is insensitive to grain size. In the nIR the crusted snow has lower albedo than the crust-free snow, as expected for larger r _e.

Fig. 4. Albedo spectra for thin (2-4 cm) and thick (-45 cm) snow over first-year ice measured on 14 and 8 October, respectively. Data for wavelengths 1350-1400nm, 1800-2100nm and 23502500 nm (here and in Fig. 6 and 9-11) are smoothed because of spectral noise resulting from low incident flux under cloudy sky conditions.

The spectrum measured on 14 October was of 2-4 cm of snow over 50 cm of ice. It has lower albedo than the deep snow at visible wavelengths where the albedo is mainly determined by snow thickness, but in the nIR its albedo is higher than that of the deep snow. In the nIR, ice is more absorptive, just 2 cm of snow being enough to hide the underlying surface, so the albedo is determined just by the SSA of the surface snow layers, which apparently was larger on 14 October than on 8 October. Comparison to model calculations of Reference Wiscombe and WarrenWiscombe and Warren (1980, Fig. 8) suggests r _e 100, 140 and 350 m for the thin snow, thick snow, and crust, respectively. A similar spectral pattern is seen when comparing the spectral albedo of snow (as in Fig. 4) to that of a cloud over a dark surface (Fig. 7a of Reference Wiscombe, Welch and HallWiscombe and others, 1984). The cloud has lower albedo than snow at visible wavelengths because it is optically thinner, but higher albedo than snow in the nIR because its particle sizes are smaller (r _e 10 m).

3.2.2. Nilas and open water

Thin nilas (Fig. 5) is rather smooth and dark, barely distinguishable from open water. Figure 6 shows the albedo spectra for open water and for several nilas cases with different thickness. Some nilas albedo spectra from Reference Brandt, Warren, Worby and GrenfellBrandt and others (2005) are also plotted in Figure 6 for comparison; they are the curves that terminate at 1060 nm because of the limited spectral range of the radiometer used in the earlier expeditions. The albedo of open water, close to 0.07 at all wavelengths, is due almost entirely to the Fresnel reflection expected for diffuse incident radiation on water with a refractive index of 1.33 (Reference Mironova and Kondrat’evMironova, 1973). The albedo spectrum of open water is in agreement with that reported by Reference Allison, Brandt and WarrenAllison and others (1993) and Reference Brandt, Warren, Worby and GrenfellBrandt and others (2005), and with broadband measurements from other studies (Reference Briegleb and RamanathanBriegleb and Ramanathan, 1982; Reference Mironova and Kondrat’evPegau and Paulson, 2001).

Fig. 5. Photograph of 0.5–0.8cm thick nilas in a narrow lead on 6 October 2012. The albedo spectrum for this scene is shown in Figure 6.

Fig. 6. Albedo spectra measured during SIPEX II for nilas and open water for wavelengths 350–2500 nm, along with albedo spectra for nilas from previous expeditions for wavelengths 350–1060 nm. The nilas of thickness 3–6cm was measured on 13 October 2012, the nilas of thickness 0.5–0.8cm was measured on 6 October 2012, and open water was measured on 24 October 2012. The 5 cm thick nilas, 2.2–3 cm nilas and 2.2 cm nilas were measured during earlier voyages and reported by Reference Brandt, Warren, Worby and GrenfellBrandt and others (2005).

In all nilas cases there is a decreasing trend in albedo toward the nIR where ice becomes more absorptive. As nilas thickens, the albedo increases in the UV and visible wavelength regions because increased scattering in the ice prevents some radiation from penetrating through the ice to be absorbed in the underlying water. In the nIR, nilas is an optically thick medium because ice is strongly absorbing at those wavelengths; the albedo is therefore determined not by ice thickness but rather by the SSA, which is probably similar for all the nilas cases examined.

The exception to this general pattern is the albedo spectrum for the thinnest nilas sample, just 0.5-0.8cm thick. It has an intermediate albedo in the UV and visible, but higher albedo than the other cases in the nIR. The nIR albedo spectrum for 3-6cm nilas is similar to that of open water, and it appears that the 2.2 cm and 2.2-3 cm nilas would also be similar; however, measurements of those spectra do not extend beyond λ=1060nm. A possible explanation for the unusual albedo spectrum of thin nilas is that the down-looking measurement could contain a significant contribution from the snow walls of the lead.

To determine the potential contribution of snow to the nilas spectrum, we produced hypothetical albedo spectra as linear combinations of the spectrum of snow and the spectrum of water, varying their relative contributions. The best match was obtained with snow occupying 9% of the downward-looking hemispheric field of view. The match was not perfect at all wavelengths, but this exercise does suggest that we should ignore the 0.5–0.8 cm spectrum shown in Figure 6 as not being representative of nilas because it was contaminated with snow at the margins of the scene.

3.2.3. Frost flowers

Frost flowers (Fig. 2) grow out of a brine skim on bare sea ice when there is a temperature difference of at least 12–16 K between the ice surface and the air directly above the sea ice (Reference Perovich and Richter-MengePerovich and Richter-Menge, 1994; Reference Rankin, Wolff and MartinRankin and others, 2002). This large temperature difference is easiest to achieve when the ice surface is warm, as is the case for thin nilas. Evaporation from the brine skim quickly deposits as frost onto small bumps on the nearby nilas surface because the air is cold. The frost flowers continue to grow as more deposition accumulates on the crystals. Frost flowers grow when wind speeds are low; otherwise the vapor released into the atmosphere can be transported away before condensing back onto the local ice surface (Reference Perovich and Richter-MengePerovich and Richter-Menge, 1994). Evidence that the brine skim is the major source of vapor for the frost flowers comes from comparing the nilas surface to nearby snow-covered first-year ice. Although frost crystals did grow on the snow surface during a cold night, they were only a few mm in height, a factor of ∼ 4 smaller than the massive frost crystals (1-1.5cm)onthenilas.

There is also a possible slight contamination of the 3–6 cm thick nilas albedo spectrum. There are some bumps in the spectrum from 400 to 800 nm that are not seen in any of the other nilas spectra. We measured this spectrum from the side of an IRB, which was red. The cosine collector likely measured some of the reflected light from the side of the IRB. This contamination is small compared to that in the 0.5–0.8 cm thick nilas case, so we do not exclude the 3–6 cm thick nilas albedo spectra from this analysis.

We measured frost-flower albedo on two days; on both days the frost flowers were 1-1.5 cm in height, but the nilas on 4 October was more densely coated than on 28 September (Fig. 2). The albedos are shown in Figure 7; comparison to Figure 6 shows that the growth of frost flowers had caused the albedo of nilas to increase by 0.2-0.3 at visible wavelengths.

Fig. 7. Albedo spectra for frost flowers over nilas (the scenes photographed in Fig. 2). The frost flowers with dense coverage were measured on 4 October 2012 and those with sparse coverage were measured on 28 September 2012. The ‘Arctic’ spectrum was measured by D. Perovich (personal communication, 2014) for frost flowers with sparse coverage over 6 cm nilas on 3 April 1998 at the SHEBA (Surface Heat Budget of the Arctic Ocean program) site in the Beaufort Sea using a Spectron Engineering SE-590 radiometer.

The dense frost flowers have a higher albedo than the sparse frost flowers by a factor of 1.5 in the UV and visible and a factor of 1.7 in the nIR. Both of the frost spectra have a closer resemblance in shape to the albedo spectra of snow than to open water and nilas. The frost-flower and snow albedo spectra have distinctive peaks at 1050 nm, 1800 nm and 2250 nm, but these peaks are muted in the nilas and open-water spectra. Frost flowers also grow on Arctic sea ice; in Figure 7 we include a spectrum measured in the Arctic springtime by D. Perovich (personal communication, 2014).

In summary, the growth of frost flowers on nilas dramatically increases the albedo at all wavelengths, resembling a thin snow cover.

3.2.4. Slush

Slush (Fig. 8) is formed when snow is blown into a new lead of open water. For pure slush, Figure 9 shows that the UV and visible albedos generally increase with increasing slush thickness because increased scattering reduces the flux of radiation reaching the underlying ocean, but the shape of the albedo spectrum remains similar. In the nIR, slush is optically thick, so there the albedo is determined more by SSA than by thickness.

Fig. 8. Photograph of slush of thickness 18cm in a narrow lead on 23 October 2012. The albedo spectrum for this scene is shown in Figure 9.

Fig. 9. Albedo spectra for slush of several thicknesses. The slush of 18cm thickness was measured on 23 October, the slush of 10cm thickness was measured on 20 October, the slush of 3 cm thickness over 50cm ice was measured on 14 October and the slush of 5 cm thickness was measured on 9 October.

Figure 9 also displays a spectrum of 3 cm slush over 50cm ice. This slush was formed in a different manner than the ‘pure slush’ cases; ice of 50 cm thickness floats with its surface above sea level, so snow falling onto it would not be expected to mix with sea water. We accessed this slush-covered ice floe by IRB in the wake of the ship; it is likely that turbulence caused by the ship’s propellers had splashed sea water up onto the floe, wetting the snow. At UV and visible wavelengths the albedo of this surface closely resembles the spectrum of 10 cm slush, but in the nIR the slush-on-ice has higher albedo, indicating larger SSA. The likely cause of the larger SSA (smaller r _e) is that the snow perched on the ice platform was incompletely wetted.

3.2.5. Three-layer cases

Figure 10 shows two scenes of snow over slush over ice; their albedo spectra are plotted in Figure 11. The albedo spectra appear to be dominated by the snow: at visible wavelengths the visible albedo is much greater for the thicker snow cover but not as high as for deep snow (Fig. 4). In the nIR, where even just 0.5 cm of snow is nearly semi-infinite, its albedo is only slightly lower than for the 2 cm snow. The snow grain size is therefore probably the same in both cases, r _e 60m, by comparison to figure 8 of Reference Wiscombe and WarrenWiscombe and Warren (1980).

Fig. 10. Photographs of three-layer surface types taken on 2 November 2012. Left: 0.5 cm of snow covering 10 cm of slush covering 1 6 cm of ice. Right: 2 cm of snow covering 3 cm of slush covering 12 cm of ice. Albedos for these scenes are shown in Figure 11.

Fig. 11. Albedo spectra for the three-layer system: snow over slush over ice (scenes photographed in Fig. 10). Both spectra were measured on 2 November.

3.3. Broadband albedos

Following the procedures of Reference Brandt, Warren, Worby and GrenfellBrandt and others (2005), we integrate the measured spectral albedos over wavelength to obtain albedos for two broad wavelength bands used in climate models (290-700 nm, 700-3000 nm), as well as the average broadband albedo for the entire solar spectrum (290-3000 nm), α:

(2)

where aλ is the measured spectral albedo and Fλ is the downward solar spectral irradiance at wavelength ∆. The broadband albedo depends on the incident solar radiation spectrum, which differs depending on whether the sky is cloudy. Here two solar spectra, one for clear-sky conditions and one for cloudy-sky conditions, characteristic of high-latitude climate during sunlit seasons are used to compute broadband albedos. We use clear and cloudy spectra that were measured at the surface of the Arctic Ocean during summer by Reference Grenfell and PerovichGrenfell and Perovich (2008). The Grenfell-Perovich cloudy case can be matched by specifying a cloud of visible optical depth _T=11 in the atmospheric radiation model SBDART (Reference Ricchiazzi, Yang, Gautier and SowleRicchiazzi and others, 1998); this optical depth is typical for Arctic summer stratus clouds (Reference Tsay and JayaweeraTsay and Jayaweera, 1984).

For the broadband solar albedo, the limits of integration in Eqn (2) are 290 nm and 3000 nm, which exceed the spectral range of our measurements, so (following Reference Brandt, Warren, Worby and GrenfellBrandt and others, 2005) we assume that the UV albedo for 290-350 nm is the same as the albedo measured at 350 nm, and that the nIR albedo for 2500-3000nm is the same as the albedo measured at 2500 nm. The UV assumption is reasonable because ice is nearly non-absorptive in the near-UV and the scattering is independent of wavelength in the visible and UV. The albedo at 2500 nm is near zero for all ice types, and there is little solar flux for λ˃2500 nm, so little error can result from assuming a constant albedo for λ˃2500nm. The amount of radiation in the wavelength bands that we have extrapolated over is small: ˂4% of the total solar radiation is contained in the 290-350 nm wavelength band and ˂0.1% of the total solar radiation is contained in the 2500-3000 nm wavelength band.

For the broadband solar albedo, the limits of integration in Eqn (2) are 290 nm and 3000 nm, which exceed the spectral range of our measurements, so (following Reference Brandt, Warren, Worby and GrenfellBrandt and others, 2005) we assume that the UV albedo for 290-350 nm is the same as the albedo measured at 350 nm, and that the nIR albedo for 2500-3000nm is the same as the albedo measured at 2500 nm. The UV assumption is reasonable because ice is nearly non-absorptive in the near-UV and the scattering is independent of wavelength in the visible and UV. The albedo at 2500 nm is near zero for all ice types, and there is little solar flux for λ˃2500 nm, so little error can result from assuming a constant albedo for λ˃2500nm. The amount of radiation in the wavelength bands that we have extrapolated over is small: ˂4% of the total solar radiation is contained in the 290-350 nm wavelength band and ˂0.1% of the total solar radiation is contained in the 2500-3000 nm wavelength band.

Table 1 lists the broadband albedos for each surface type observed during SIPEX II. The values, particularly for the nIR, are higher under cloudy sky. As mentioned above, clouds absorb nIR radiation at wavelengths where snow and ice are also absorptive. Radiation that has passed through a cloud is therefore depleted in the wavelengths for which snow and ice have low albedo, so the spectrally averaged albedo is higher under cloudy-sky conditions than under clear-sky conditions (Reference WarrenWarren, 1982). In general, the cloudy-sky albedo is more relevant in the Antarctic sea-ice zone because the average cloud cover in spring and summer exceeds 80% (Reference WarrenWarren and others, 1988).

The only surface types measured during both SIPEX II and previous ANARE expeditions were open water, nilas and snow-covered thick first-year ice, and their broadband albedos are close to the broadband albedos for those ice types reported by Reference Brandt, Warren, Worby and GrenfellBrandt and others (2005). For this reason, and because on SIPEX II we did not have the opportunity to measure the other ice types reported in Table 1 and 3 of Reference Brandt, Warren, Worby and GrenfellBrandt and others (2005), we see no need to revise the albedos reported in Reference Brandt, Warren, Worby and GrenfellBrandt and others (2005).

Table 3. The location, date (local time) and snow properties for each ice station where snow on first-year sea ice was collected for impurity analysis. Median values are given for the absorption A˚ ngstro¨m exponent (A˚ ), estimated fraction of absorption due to non-black carbon (nonBC) material for wavelengths 350–700 nm, maximum possible black carbon mixing ratio, if all the absorption at 650–700nm is due to black carbon (maxBC), and our best estimate of the black carbon mixing ratio (estBC)

4.1 Method

Vertical profiles of LAI and density in the snowpack were obtained at five ice stations. Established methods described by Reference Doherty, Warren, Grenfell, Clarke and BrandtDoherty and others (2010) were used to quantify the LAI, and some of the following description of the method is taken from that paper. Snow pits were dug upwind of the ship. Facing upwind and wearing clean, dust-free disposable over-gloves, we used a stainless-steel spatula to put snow into plastic bags which could each hold ∼1 kg of snow. Sample sizes of 1-11 kg were collected to obtain filter loadings in the range suitable for accurate spectrophotometric measurements. Duplicate samples were typically collected at vertical intervals of 5-10cm. The top 2 cm of snow was also collected at each snow pit. Additionally, if there was obvious layering (e.g. a thin top layer of newly fallen snow or drift snow), that layer was collected separately, however thin.

The snow was kept frozen until ready for processing in a laboratory aboard the Aurora Australis. The snow from each bag was spooned into a clean glass beaker and melted in a microwave oven. The meltwater was passed through a 0.4 m Nuclepore filter, using a handpump to create a partial vacuum, and the volume of filtrate was measured. The melting of one bag of snow typically required 5-10min, and the filtration another 10-20 min. This procedure is designed to minimize the time that meltwater is in contact with glass or plastic, because soot is often hydrophobic and some could be lost to the container walls instead of collected on the filter. Another reason for keeping the snow frozen until it is ready for processing is to avoid algal growth, since algae can change the water chemistry as well as absorbing light themselves.

The transmittance spectrum of each filter was measured in an integrating-sandwich spectrophotometer that was built for this purpose in our laboratory at the University of Washington. This incorporates an integrating sphere as one side of the sandwich (Reference Grenfell, Doherty, Clarke and WarrenGrenfell and others, 2011). The integrating-sandwich configuration is designed to minimize the effect of scattering by the aerosols on the filter, so that the measured signal is a function only of the losses due to light absorption. A set of standard filters containing known (weighed) amounts of BC in the form of a commercial soot is used to calibrate the system. The spectral shape of the absorption is used to separate BC from non-BC absorbers by the method of Reference Grenfell, Doherty, Clarke and WarrenGrenfell and others (2011), making use of the enhanced absorption at blue and UV wavelengths by brown carbon and mineral dust.

The spectral shape is characterized by the slope of absorption on a log-log plot across the wavelength domain 450-600 nm, where the spectrophotometer has good accuracy:

(3)

where T _i is the absorption optical depth on the filter at wavelength ∆ _i and A is called the absorption Angstrom exponent. The separation of contributions by BC and non-BC absorbers is done by assuming that the absorption Angstrom exponent for BC is 1.0 and for non-BC light-absorbing aerosol is 5.0. The absorption spectra are then integrated over wavelength, weighted by the incident solar spectrum for A ˂ 700 nm (the wavelength region where snow albedo is sensitive to impurities) to obtain the fraction of solar absorption by BC and by non-BC impurities (Reference Doherty, Warren, Grenfell, Clarke and BrandtDoherty and others, 2010; Reference Grenfell, Doherty, Clarke and WarrenGrenfell and others, 2011).

4.2. Results

Table 3 describes the snow properties at each site. Snow was collected at ice stations 1–4 and 7. We did not sample at ice stations 5, 6 and 8 because these stations were only a short distance from others we had already sampled. Samples of size ∼1 kg were collected at ice station 1. The filters were visually blank, so at ice stations 2 and 3 larger samples of size 2-4 kg were collected. The filters were still nearly blank, so at ice stations 4 and 7 larger samples of size 8-11 kg were collected; these do show a slight darkening visible by eye.

Table 3 gives the absorption Angstrom exponent, the estimated BC mixing ratio, and the fraction of absorption due to non-BC constituents. Because the Angstrom exponent for non-BC is variable, causing uncertainty in the BC/non-BC apportionment, we also list a quantity called ‘max BC. This is the maximum possible BC, under the assumption that all the absorption at 650-700 nm is caused by BC. The choice of 650-700 nm for this assessment is due to the fact that non-BC constituents absorb only weakly in this spectral region.

There is little difference among the sites. The BC amounts are very low, 0.1-0.5 ngg–¹similar to values we found previously on the East Antarctic plateau (Reference Warren and ClarkeWarren and Clarke, 1990; Reference Grenfell, Warren and MullenGrenfell and others, 1994; Reference Warren, Brandt and GrenfellWarren and others, 2006). A median vertical profile of BC, combining results from five sites, is shown in Figure 12. The top (springtime) layers tend to have higher BC mixing ratios than the lower (wintertime) layers. The fraction of particulate absorption due to non-BC is -40%, similar to values found in the Arctic (Reference Doherty, Warren, Grenfell, Clarke and BrandtDoherty and others, 2010). In the Arctic the non-BC could be organic matter from combustion, vegetation, or mineral dust containing iron oxides. In the Antarctic Ocean it may largely be particulate organic matter from the ocean water and from animals (seals and penguins) occupying the surface of first-year ice.

Fig. 12. Median vertical profile of estimated black carbon mixing ratio (estBC), combining results from five ice stations (when available). The legend indicates the number of ice stations used to calculate the median estBC value.

These values are at the extreme low end of the range of BC mixing ratios found in natural snow, which span four orders of magnitude: 100-1200ngg–¹ in northeast China (Reference Wang, Doherty and HuangWang and others, 2013), 30-40 ngg–¹ in Arctic Russia, 10-15ngg-¹ in Canada, 2-5 ngg–¹ in Greenland, 4-16 ngg–¹ in snow covering Arctic sea ice (Reference Doherty, Warren, Grenfell, Clarke and BrandtDoherty and others, 2010, Table 2 ) and 0.1-0.7 ngg–¹ in Antarctica. Of course it is not surprising that the Antarctic region has the cleanest snow on Earth.

Table 3 . The location, date (local time) and snow properties for each ice station where snow on first-year sea ice was collected for impurity analysis. Median values are given for the absorption Angstrom exponent (A), estimated fraction of absorption due to non-black carbon (nonBC) material for wavelengths 350–700 nm, maximum possible black carbon mixing ratio, if all the absorption at 650–700 nm is due to black carbon (maxBC), and our best estimate of the black carbon mixing ratio (estBC)