Properties of black carbon and other insoluble light-absorbing particles in seasonal snow of northwestern China

A large field campaign was conducted and 284 snow samples were collected at 38 sites in Xinjiang Province and 6 sites in Qinghai Province across northwestern China from January to February 2012. A spectrophotometer combined with chemical analysis was used to measure the insoluble light-absorbing particles (ILAPs) and chemical components in seasonal snow. The results indicate that the cleanest snow was found in northeastern Xinjiang along the border of China, and it presented an estimated black carbon (Cest BC) of approximately 5 ng g−1. The dirtiest snow presented a Cest BC of approximately 450 ngg−1 near industrial cities in Xinjiang. Overall, the Cest BC of most of the snow samples collected in this campaign was in the range of 10–150 ngg−1. Vertical variations in the snowpack ILAPs indicated a probable shift in emission sources with the progression of winter. An analysis of the fractional contributions to absorption implied that organic carbon (OC) dominated the 450 nm absorption in Qinghai, while the contributions from BC and OC were comparable in Xinjiang. Finally, a positive matrix factorization (PMF) model was run to explore the sources of particulate light absorption, and the results indicated an optimal three-factor/source solution that included industrial pollution, biomass burning, and soil dust.

pollution sources.

Introduction
The deposition of insoluble light-absorbing particles (ILAPs), primarily black carbon (BC), organic carbon (OC), and dust, on snow can reduce snow albedo (Warren and 5 Wiscombe , 1980;Chylek et al., 1983;Brandt et al., 2011;Hadley and Kirchstetter, 2012), which can significantly affect regional and global climate (Jacobson, 2002(Jacobson, , 2004Hansen and Nazarenko, 2004;Flanner et al., 2007Flanner et al., , 2009McConnell et al., 2007;Ramanathan and Carmichael, 2008;Bond et al., 2013). Warren and Wiscombe (1980) suggested that a mixing ratio of 10 ng g -1 of BC in snow may reduce the snow 10 albedo by approximately 1%. A modeling study indicated that soot could reduce snow and sea ice albedo by 0.4% from the global average and by 1% in the Northern Hemisphere (Jacobson, 2004). Previous studies found that the "efficacy" of BC/snow forcing in the Arctic is more than three times greater than that of forcing by CO2 (Hansen and Nazarenko, 2004;Flanner et al., 2007). However, radiative forcing is 15 highly uncertain. For example, Hansen and Nazarenko (2004) found that the effect of soot on snow and ice albedo yielded a climate forcing of +0.3 W m -2 in the Northern Hemisphere. Recently, the IPCC's AR5 (2013) reported that the radiative forcing from BC in snow and ice is 0.04 W m -2 of the global mean, although it presents a low confidence level. Bond et al. (2013) indicated that the best estimate of climate forcing 20 from BC deposition on snow and sea ice in the industrial era is +0.13 W m -2 with 90% uncertainty bounds of +0.04 to +0.33 W m -2 . The all-source present-day climate The Cryosphere Discuss., doi:10.5194/tc-2016-233, 2016 Manuscript under review for journal The Cryosphere Published: 21 November 2016 c Author(s) 2016. CC-BY 3.0 License.
forcing including preindustrial emissions is somewhat higher at +0.16 W m -2 . Many factors complicate the evaluation of climate forcing by BC in snow (Hansen and Nazarenko, 2004;Bond et al., 2013). Hence, abundant comprehensive field campaigns are required to measure ILAPs in snow to limit this uncertainty.

5
In addition to BC, which presented the most absorptive impurity per unit mass, OC and dust can significantly contribute to particulate light absorption in snow. OC in snow may be related to either combustion products that are deposited onto snow or soil that is mixed into snow. Xu et al. (2006) first quantified the OC content on the Tibetan Plateau and determined the effect of OC on surface snow melting. Wang et al.

10
(2013) suggested that OC dominates the particulate light absorption across the grasslands of Inner Mongolia in North China. Light absorption by dust is usually related to iron oxides. Although its ability to reduce snow albedo is less than that of BC by approximately a factor of 50 (Warren, 1984), dust is the dominant absorber in snow locations. For example, the increased radiative forcing by dust in snow has 15 affected the timing and magnitude of runoff from the Upper Colorado River Basin (Painter et al., 2007(Painter et al., , 2010. Understanding the sources of ILAPs in snow is necessary for examining the climatic effects of ILAPs in snow. Certain scientists have focused on exploring the potential sources of BC in snow Shindell et al., 2008;Forsstrom et al., 20 2009); however, these studies primarily relied on numerical transport modeling based on limited data from emission inventories or calculated back trajectories, and they  (Ye et al., 2012).

20
In this study, we analyzed the spatial and vertical distributions of ILAPs in the seasonal snow in northwestern China, investigated the contributions from BC, OC and  Figure 1 shows the locations of the sampling sites, which were numbered in chronological order and followed the field campaign by Wang et al. (2013). Fresh snow was gathered from 13 sites where snow was falling at the time of sampling.

10
Forty-two sites were separated into 5 regions according to their geographical distribution to investigate the spatial variations in snow-containing contaminants and their potential sources, with one region (Region 1) located in Qinghai and the other regions (Regions 2-5) located in Xinjiang ( Figure 1). upwind of the approach road or railway to minimize the effect of pollution from local sources and achieve a representation of large areas.
The snow samples were filtered at four temporary laboratories to prevent the melting snow from influencing the ILAP content. The snow samples were quickly melted in a microwave oven and then immediately filtered through a 0.4-μm Nuclepore filter. The 5 samples "before" and "after" filtration were collected and refrozen for subsequent chemical analyses, and the filters were subjected to BC and OC analyses. Additional details on the snow collection and filtration processes have been previously reported (Doherty et al. , 2014Wang et al., 2013).

10
The chemical analysis performed here followed that of Wang et al. (2015) and was similar to the procedures described by Zhang et al. (2013a) and Doherty et al. (2014).
Details of these approaches have been previously reported (Yesubabu et al., 2014).
Briefly, the major ions (SO 4 2-, NO 3 -, Cl -, F -, Na + , K + , Ca 2+ , Mg 2+ and NH 4 + ) were where Na Ss + is sea salt Na + , 0.12, 0.038, 0.038, and 0.25 are the mass ratios in seawater of magnesium to sodium, calcium to sodium, potassium to sodium and sulfate to sodium, respectively. Na Ss + is calculated as follows (Hsu et al., 2009): Na Ss =Na Total -Al*(Na/Al) Crust (3) 10 where (Na/Al) Crust is the Na/Al ratio of representative dust materials (Wedepohl, 1995). Following Hsu et al. (2009), we estimated all three fractions (dust, sea salt, and biosmoke fractions) of K in snow, and K Biosmoke was determined as follows: where (K/Al) Crust is 0.37 and represents the K/Al ratio in the dust materials (Wedepohl, 1995) and Na Ss is estimated by Equation (3). Following Zhang et al.
(2013b), we calculated the contribution of TEO using the following equation:  (Hsu et al., 2010). (E/Al) Snow and (E/Al) Crust are the ratios of the elements to the Al concentration in the snow sample and crust (Wedepohl, 1995), respectively. A multiplicative factor of 1.3 was used to convert the element abundance to the oxide abundance, which is similar to the method of Landis

Spectrophotometric analysis
The filters were analyzed for the ILAP content in the snow using a modified integrating-sandwich spectrophotometer (ISSW), which was described by Grenfell et al. (2011) and used by Doherty et al. (2010Doherty et al. ( , 2013Doherty et al. ( , 2014Doherty et al. ( , 2015, Dang and Hegg (2014) 15 and Wang et al. (2013Wang et al. ( , 2015. This ISSW spectrophotometer measures the light attenuation spectrum from 400 to 700 nm where the signal-to-noise ratio is optimized. The total light attenuation spectrum is extended over the full spectral range by linear extrapolation from 400 to 300 nm and from 700 to 750 nm (Grenfell et al., 2011).
Light attenuation is nominally only sensitive to ILAPs on the filter because of the 20 diffuse radiation field and the sandwich structure of two integrated spheres in the ISSW (Doherty et al., 2014). By considering all of the light absorption that occurred were calculated by using the wavelength dependence of the measured spectral light absorption and by assuming that the MACs of the BC, OC and Fe were 6.3, 0.3, and 0.9 m 2 g -1 at 550 nm, respectively, and that the absorption Ångström exponents (Å) were 1.1, 6, and 3, respectively. The details of this analysis were interpreted according 10 to Grenfell et al. (2011). The OC mixing ratio was also determined according to Equation (2) in Wang et al. (2013), and the Fe concentration was determined according to the ICP-MS measurements.
Many studies (e.g., Jacobson, 2001;Hadley and Kirchstetter, 2012;Bond et al., 2013) have indicated that the MAC of BC is somewhat higher than the value used here, and 15 Bond and Bergstrom (2006) recommended a value of 7.5±1.2 m 2 g -1 at 550 nm.
However, we applied a value of 6.3 m 2 g -1 to provide a comparison with previous studies (Hegg et al., 2009(Hegg et al., , 2010Doherty et al., 2010;Wang et al., 2013). If the MAC of BC is actually close to 7.5 m 2 g -1 , then our measured mass mixing ratio will be too high by a factor of 1.19. If the radiation models are run with the BC mixing ratio 20 reported in this study, then the MAC of BC must be 6.3 m 2 g -1 ; otherwise, the BC mixing ratio should be scaled appropriately.

PMF model
The Positive Matrix Factorization (PMF) model that was used here (US EPA PMF 5.0) is a receptor model that can quantify contributions from sources to samples based on the composition or fingerprints of the sources, and it has been widely applied (e.g., Amato et al., 2009;Amato and Hopke, 2012). The speciation or composition is 5 determined by using analytical methods appropriate for the media, and key species or combinations of species are required to distinguish the effects. The PMF model is a multivariate factor analysis tool that decomposes a matrix of speciated sample data into two matrices: factor contributions and factor profiles. These factor profiles must be interpreted by the analyst to identify the source types that may have contributed to 10 the sample by using available ancillary information, such as the measured source profile information and emission or discharge inventories. The characteristic factor profiles are completely dependent on the mathematical approaches of the PMF model; therefore, the number of factors is not known a priori and must be selected individually in terms of the analyst's understanding of the sources that affect the 15 samples as well as the number of samples and species' characteristics.
The PMF model uses two data sets as inputs to weigh individual points. One is the set of the concentrations of the input species, including the chemically analyzed constituents, along with C BC max , and the other is the associated uncertainty data set.  Doherty et al. (2010). C BC max was likely biased by errors in the assumed MAC of our fullerene standards, which was applied to convert the measured absorption to a maximum BC mass. However, the relative biases of C BC max were uniform across all of the data sets, and the results of the PMF analysis were dependent on the relative variance of the given species against the absolute concentration; therefore, the effects 5 can be appropriately ignored (Doherty et al., 2014).
Normally, the PMF model is applied to analyze a time series of species concentrations at a single observation site to estimate temporal variations. In this study, we used the PMF model to analyze the spatial variations in source contributions. Although atypical, previous studies have effectively employed this model and confirmed its 10 reliability in terms of factor analyses of spatial distributions (Paatero et al., 2003;Chen et al., 2007;Hegg et al., 2009Hegg et al., , 2010Zhang et al., 2013a;Doherty et al., 2014).  Doherty et al. (2014); therefore, we did not consider these values. Thus, the lowest C BC est values were approximately 5 ng g -1 , which were smaller than the minimum BC mixing ratio of approximately 40 ng g -1 measured in North China via the same spectrophotometric analysis (ISSW) (Huang et al., 2011;Wang et al., 2013) and

Results
comparable to the values of approximately 3 ng g -1 from the Greenland Ice Sheet 5 . The highest C BC est was found at sites 53, 60, 67, 83 and 84. At site 83, the C BC est reached 619 ng g -1 at the bottom layer; however, the underlying soil may have been responsible for this high value. Therefore, this value cannot represent the regional background level of BC. After excluding site 83, the highest C BC est value was approximately 450 ng g -1 , which was much lower than the values of >1000 ng 10 g -1 in snow in the industrial area of northeastern China (Wang et al., 2013). Overall, the C BC est of most of the snow samples ranged from 10-150 ng g -1 , which were similar to the visible values reported by Ye et al. (2012) and BC measurements of 4-120 ng g -1 recorded from glaciers in Tibet and Xinjiang by a previous field campaign that used a controlled combustion method (Xu et al., 2006(Xu et al., , 2009Ming et al., 2008Ming et al., , 2009.

Results by region
As discussed above, the sample sites were separated into five regions. Table 2 lists the regional averages and standard deviations of C BC equiv , C BC max , C BC est , Å tot , and f nonBC est for the surface and subsurface layers. The spatial distributions of C BC est and Å tot for the surface snow samples are shown in Figure 2a and 2b. In addition, Figure 2c shows the determine the relative contributions from BC and non-BC ILAPs to the snow albedo reduction at each site. However, the BC mass deposited onto snow over a specified period is more useful when comparing models than the surface values because the average BC over many snowfall events across a typical month or season presents a more representative contribution to the background levels throughout the entire 5 accumulation period (Doherty et al., 2014). Thus, in Table 3, we list the integrated snow water equivalent (SWE) and the total BC mass for a 1-cm 2 column of snow (integrated BC). We also estimated the average BC mixing ratios in the snow column (C BC est ̅̅̅̅̅ ), which were calculated as the integrated BC divided by the SWE (Figure 2d and Table 3). Indeed, the C BC est ̅̅̅̅̅ value was more spatially uniform than the C BC est value.

10
In Region 1 (sites 47-52), which is located in the eastern Tibetan Plateau in Qinghai Province, the snow was thin and patchy and presented a sample snow depth of 2.5-10 cm. During windy periods, local soil can be lofted and deposited onto snow, and this deposition has been confirmed by previous reports (Ye et al., 2012), which observed yellowed filters because of the heavy loading of soil dust. Although the C BC est was 15 intermediate and presented typical values of 30-150 ng g -1 , both the C BC equiv and C BC equiv * f nonBC est values were highest in the surface (307±119, 213±89 ng g -1 ) and subsurface (332±201, 214±116 ng g -1 ) snow among all five regions (Table 2 and  Region 4 (sites 64-70) is located in northwestern Xinjiang along the border of China.
The C BC est generally ranged from 20-150 ng g -1 . The regional average Å tot was approximately 3. The f nonBC est in this region was the lowest and presented an average 5 value near 50%; therefore, the BC and non-BC particles presented almost identical contributions to light absorption, which was inconsistent with the other regions, where non-BCs played a dominant role.
The cleanest snow of this campaign was found in Region 5 (sites 72-78) in northeastern Xinjiang. Most of the snow samples had low C BC est values of 10-50 ng g -1 , 10 which were much smaller than the values of 50-150 ng g -1 in the cleanest snow in northeastern China. The Å tot value generally ranged from 2.5 to 3.5 and presented a regional average of approximately 3, which was consistent with other regions in Xinjiang. The f nonBC est varied obviously varied from < 50% to > 90%. This wide range indicates the spatial variance in the dominant emission sources of particulate 15 light absorption in this region.

Vertical variations in snowpack light-absorbing particulates
The vertical profiles of the C BC max , C BC est , and Å tot at each sample site are shown in  Therefore, we did not report the values at these layers. In addition, the sampling at sites 47-49 was conducted in drift snow as discussed above. Thus, the vertical profiles from these sites in Qinghai did not accurately represent the temporal variations in the deposition of snow, although apparent vertical differences were observed at could sites, such as site 47. In Regions 3, 4 and 5 in Xinjiang, the C BC est values were much larger in the surface snow (127±158, 126±124, and 74±56 ng g -1 , respectively) than in the subsurface snow (75±120, 82±56, and 37±31 ng g -1 , respectively), with the ratio 5 of the C BC est from the top layer to the average C BC est from all the subsurface layers presenting values of 1.7, 1.5, and 2, respectively, which indicates an increase in aerosol deposition later in winter. However, the C BC est values in the surface (81±102 ng g -1 ) and subsurface layers (89±69 ng g -1 ) in Region 2 were comparable. These Å nonBC , which are shown in Figure 4b. Overall, the Å nonBC values were almost in a narrow range of 5-6, which indicated that OC was the major component of non-BC  Doherty et al. (2010) analyzed the errors that originated from these assumptions and indicated a likelihood of uncertainty of < 50% based on liberal evaluations of these potential sources of errors. the errors of C BC max were the lowest among the studied variables. Three to seven factors and 7 or more random seeds were always applied in the PMF model. Thus, the optimal number of factors/sources was 4 based on the robust and theoretical Q values (Hegg et al., 2009(Hegg et al., , 2010. However, 3 factors provided more physically reasonable results and the most easily identifiable sources, which was consistent with studies of 10 snow in northeastern China (Zhang et al., 2013a) and North America (Doherty et al., 2014). The diagnostic regression R 2 value for C BC max with this 3-factor solution was considerably high (0.87). Rotational ambiguity was tested by varying the peak parameter, which also indicated stable results. Hence, the 3-factor solution was the best choice.

PMF results
15 Figure 5 shows the source profile, including the measured mass concentrations (lines) and the percent of species apportioned to each factor (dots) for the 3-factor solution.
The first factor (top panel) was obviously characterized by high loadings of SO 4 2-, combustion (Hailin et al., 2008). Furthermore, the molar ratio of Clto Na + was clearly larger than that of sea salt by a factor of > 2, which implied another source in addition to sea salt. Additionally, NH 4 + is a recognized marker of coal combustion.
However, the ratio of C BC max to SO 4 2was low (0.04) and close to the values in pollution sources as reported by Hegg et al. (2009). Therefore, we considered the first 5 factor an industrial pollution source. The second factor (middle panel) presented substantial loadings of Na + , K + , and KBiosmoke. However, chemical analyses are not available for certain organic matter, including levoglucosan, succinate, oxalate, and formate, which generally indicate biomass burning. K + and KBiosmoke are markers for biomass burning emissions and were highly loaded for this factor. In particular,

10
KBiosmoke, which was calculated as the biosmoke fraction of K, was representative (Zhang et al., 2013a). Compared with the first factor, the molar ratio of Clto Na + was smaller than 1, which indicates that Na + was also a potential product of emission sources in addition to sea salt, such as biomass burning (Oh et al., 2011). However, the ratio of C BC max to SO 4 2was relatively high compared with that of the previously 15 identified industrial pollution sources and close to the ratio for a biomass burning source as reported by Doherty et al. (2014). Therefore, we interpreted the second factor as a biomass burning source. The third factor (bottom panel) accounted for over 50% Al, Cr, Fe, Cu, and Ba in the samples. Al and Fe are well-known crustal constituents, and they are usually used to determine the mass of soil dust. The mass 20 ratio of Fe to Al (0.36) was close to the value in the continental crust (0.40) (Wedepohl, 1995). Furthermore, the enrichment factors (EFs) of the trace elements, including Cr, Cu, and Ba, were < 5 at many sample sites, which indicates that these elements may have originated from a crustal source (Zhang et al., 2013b). Hence, we can interpret the third factor as a soil dust source. with the highest C BC max (53%) were more geographically dispersed and relatively evenly distributed across Regions 1, 2, and 4. At sites 53 and 67, the contributions were significantly large, which were likely caused by the concentrations of KBiosmoke, which were over 500 ng g -1 and much higher than those at the other sites. Indeed, biomass burning, such as biofuel combustion for heating, in winter and early spring in 15 northwestern China is normally prevalent . Unsurprisingly, a soil dust source, which was characterized by the highest loadings of Al and Fe, was mainly associated with the sites in Qinghai (Region 1), although the contributions were obvious at certain sites in Xinjiang, especially the sites in Region 5, which was partially because these sites were located on hills with scarce plants, and wind may 20 have blown local soil dust onto the snow.

Source attribution of the particulate absorption
The  Figure 7. The average regional contributions are shown in Table 4. The most remarkable feature of the source attributions is the differences observed by region. Biomass burning was the primary source in Region 1 (in Qinghai) and in Regions 2 and 4 (in Xinjiang), and it 5 presented average regional contributions of 59%, 60%, and 67%, respectively.
Although high dust mass was present in the snow samples from Region 1, this source attribution was reasonable because the C BC max from the biomass burning sources was much larger than that from the soil dust sources by a factor of > 3 ( Figure 6). In particular, biomass burning in Qinghai is prevalent, especially during winter (Yan et 10 al., 2006). In Region 1, soil dust accounted for 29% of the particulate absorption, which was less than the contribution from the biomass burning sources but more significant compared with the contributions from soil dust in the other regions. In Regions 2 and 4, most of the sample sites were located on mountains and far from industrial areas; therefore, dominant absorption by biomass burning sources was not 15 anomalous. The only exception was site 58, which was dominated by industrial pollution sources, and this result was likely because of its shorter distance from cities and lower elevation. In Region 3, all of the sample sites were located near cities and suffered from anthropogenic emissions; therefore, industrial pollution was the primary source and presented a contribution of 58%. In Region 5, the primary source differed   Ye et al. (2012) performed a preliminary study on the same field campaign and found that the snow BC mixing ratios, which were based on visual estimates, were negatively correlated with the altitudes of the sample sites in Xinjiang. In our study, the C BC est and C BC max values from the ISSW presented a similar trend (Figure 8a and b).

15
Thus, altitude is an important influencing factor for BC mixing ratios and particulate absorption. Additionally, the C BC max value simulated by the PMF model decreased steadily with altitude, although this trend was not as obvious as that for the measured C BC est and C BC max . Therefore, we can explore the cause of altitude gradients for the BC mass based on the PMF 3-factor solution. Figure 9 shows the contributions of each 20 source as a function of altitude. The sample snow from site 53 was dirty (Ye et al., 2012); therefore, we did not consider the results of that site in the trend analysis.

Mass contribution of the chemical components
In addition to performing a PMF analysis, the chemical components must be evaluated to examine the potential emission sources. As shown in Figure 10a biomass burning source. BC is regarded as an important light-absorbing particle, and it ranged from 0.2 to 4.8% in mass at all sites, with an average of 1.3%, which is smaller than that (2.4-5.1%) in urban areas in China (Huang et al., 2014). KBiosmoke is a good indicator of biomass burning, and it ranged from 0.4 to 7.8% and presented the largest regional average fractional contribution (3.0%) in Region 2. Wang et al. (2016) 5 reported comparable values of 1.3-5.1% in the snow in northeastern China. Bond et al. (2004) indicated that the OC:BC ratio of the emissions from fossil fuel burning is lower than that from biomass or biofuel burning; therefore, we may qualitatively examine the primary emission sources based on this theory. In this work, 10 the regional average ratios of OC to BC were 20.9, 6.12, 3.99, 6.71, and 7.28 ( Figure   11a). The smallest value in Region 3 was similar to those observed in Beijing (Zhang et al., 2013b). The similar ratios in Regions 2, 4, and 5 were close to that of savanna and grassland regions as reported by Andreae and Merlet (2001). The results suggested the relative dominance of an industrial pollution source in Region 3 and a 15 biomass burning source in other regions in Xinjiang. This pattern was similar to that of the source apportionment analysis by the PMF model. The largest value in Region 1 in Qinghai implied a primary contribution to OC from soil dust.

Comparative analysis of chemical components
Sources of nitrate are considerably more varied than the sources of sulfate (Arimoto et al., 1996). For NO 3 -, the largest source is fossil fuel combustion. Biomass burning is 20 regarded as another main source, which was determined according to the analysis of Logan. Additionally, microbial activity in soil is a potential source of nitrate. However, SO 4 2is mainly a product of burning coal. Thus, we can compare the correlation between NO 3 and SO 4 2to explore the variety of emission sources. As shown in Figure 11b, the average regional ratio of NO 3 to SO 4 2in Region 1 was 0.49, which was lower than that in the other regions. The concentrations and correlation coefficients of NO 3 and SO 4 2were both the lowest in Region 1, which indicates 5 limited emissions from an industrial pollution source. In Xinjiang, the ratios mostly ranged from 1 to 1.5, and the high correlation coefficient in Region 5 was associated with similar industrial pollution sources, whereas the low correlation coefficients in Regions 2-4 may have been related to complicated industrial pollution sources.
A comparison between cations and anions is shown in Figure 11c. Generally, the 10 correlation coefficients were all significant at the 5% level. A large anion charge deficit was observed in Region 1, which presented an average regional ratio of 2.93, which was likely caused by the absence of detected CO 3 2and HCO 3 -. Carbonates (e.g., CaCO 3 and MgCO 3 ) are often abundant in dust, which was observed in the Central Himalayan Glacier (Xu et al., 2013), and account for the largest contribution 15 from soil dust to the mass concentration. However, the average ratios in the sample snow in Xinjiang were generally uniform and ranged from 0.7-1.2, which suggests an adequate charge balance. Overall, the concentrations of inorganic ions in the snow samples were lower than those in the rainwater in urban sites in China (Wang and Han, 2011) but larger than previous measurements in the Himalayas (Thompson et al., 2000;20 Xu et al., 2013).

Discussion and conclusions
The     according to region (see Figure 1). (c) Estimated BC-equivalent mixing ratio (ng g -1 ) required to explain the spectrally integrated (300-750 nm) absorption of sunlight by non-BC components in snow. (d) Estimated average snow BC mixing ratio, C BC est ̅̅̅̅̅ , which was calculated by integrating the snow water content and BC mass over the entire snowpack (see Table 3).