During the study period, less snow fell in 2014 than in 2010, and no seasonal snow was present in the western part of Inner Mongolia. Therefore, we collected 94 snow samples at 14 sites in January and February of 2014 across north China following the sampling route of Huang et al. (2011). The sites are numbered in chronological order from 90 to 103, following previous snow surveys (Ye et al., 2012; Wang et al., 2013). Figure 1 shows the locations of the snow field campaigns across northern China. The sampling locations were selected to be at least 50 km from any settlement and 1 km from the nearest road. Snow samples were kept frozen before being filtered. We set up a temporary laboratory along the sampling route. Owing to BC in snow often being hydrophobic, long-time melting could cause more BC loss to the container walls instead of being collected on the filter (Ogren et al., 1983). In order to minimize the loss of insoluble LAPs (ILAPs), we quickly melted the snow samples in a microwave within a very short time. Therefore, the loss of insoluble LAPs is very limited and can be neglectable. At present, this method is a widely performed snow-melting procedure (Dothery et al., 2010, 2014; Wang et al., 2013). Subsequently, we simultaneously filtered the snow samples using quartz-fiber filters with 1 µm pores and Nuclepore filters with 0.4 µm pores. Then, we refiltered the snow samples for the quartz-fiber filters using Nuclepore filters with 0.4 µm pores to account for the loss of BC mass in the 1 µm pore quartz-fiber filters. Finally, we stored the original and refiltered snow samples in clean high-density polyethylene bottles in a freezer at -30 ∘C for subsequent analysis. For details of the sampling and filtration procedures, see Wang et al. (2013).

To evaluate the accuracy of the TSI technique in measuring BC concentrations, the atmospheric samples were continuously collected on Nuclepore and quartz-fiber filters with high-volume samplers during the periods 09:00 to 17:00 (daytime; local time) and 23:00 to 07:00 (nighttime) at site 103 in Lanzhou from 5 to 25 August 2015. The pumps were operated at a flow rate of 10 L min-1. In total, 40 atmospheric samples were collected during this experiment and used to assess the accuracy of the atmospheric BC concentration measurements of the improved TSI technique.

Light transmission techniques are the most commonly used methods for determining light-absorbing impurities in aerosol filter samples of the atmosphere and snow–ice. Since the 1970s, a series of optical attenuation techniques have been developed for estimating BC concentrations using light transmission changes through filters based on Beer's law. An integrating sphere (IS) technique was first proposed for measuring BC by Fischer (1970). The integrating sphere was coated with diffusely reflecting white paint through a small hole, and the reduction in signal after measuring the sample filters represented the absorption of BC. Subsequently, a new integrating plate (IP) instrument was developed to measure scavenging BC on filters based on the IS technique, which uses a light-diffusing support to provide a nearly Lambertian light source for light transmission using 0.4 µm Nuclepore filters (Clarke et al., 1967; Horvath, 1993b). However, the multiple scattering of solar radiation affects the accuracy of the IP technique (Clarke et al., 1967; Hitzenberger, 1993; Petzold et al., 1997; Bond et al., 1999). A new integrating sandwich configuration of the ISSW instrument was designed to measure the absorption of light-absorbing impurities based on the ISSW principle of Grenfell et al. (2011). The ISSW instrument can isolate the absorption properties of light-absorbing impurities deposited on polycarbonate Nuclepore filters. By assuming the mass absorption efficiency and non-BC Ångström exponent at 550 nm, this technique is currently capable of reliably measuring BC and non-BC light absorption (Wang et al., 2013; Dang and Hegg, 2014; Doherty et al., 2014). However, Schwarz et al. (2012) found that the total instrumental uncertainty associated with ISSW BC concentration determinations for ambient snow is 11 %, and this uncertainty is partially due to the scattering effects of insoluble impurities deposited on the filters (Doherty et al., 2010; Grenfell et al., 2011).

The improved TSI spectrophotometer developed in this study is small, lightweight, and portable, and it can accurately quantify BC concentrations using a technique based on the integrating sphere and integrating plate transmission techniques (Fig. 2). The major improvement of this spectrophotometer is that we replaced the integrating sandwich of the ISSW instrument developed by Grenfell et al. (2011) with a new integrating sphere. In addition, an iron hoop is applied to the top integrating sphere surrounding the sapphire windows to reduce light scattering due to insoluble particles on the filters. Therefore, the total relative light absorption due to all insoluble impurities on the filter can be estimated from the visible to near-infrared wavelengths. The total light attenuation can be calculated from the light transmitted by a snow or atmospheric sample, S(λ), compared with that transmitted by a blank filter, S0(λ). Then, the relative attenuation (Atn) through the filter can be expressed as follows: 1Atn=ln⁡[S(λ)/S0(λ)]. The total absorption Ångström exponent Åtotλ0 of all the ILAPs on the filters can be calculated from the following formula: 2Åtotλ0=-ln[τtot(λ1)/τtot(λ2)]ln(λ1/λ2). Ånon-BC is calculated as a linear combination of the contributions to light absorption made by OC and Fe: 3Ånon-BC=ÅOC×fOC+ÅFe×fFe. The total absorption Ångström exponent of all ILAPs on a filter (Åtot) can be described as a linear combination of ÅBC and Ånon-BC weighted by the light absorption fraction: 4Åtotλ0=ÅBC×fBCλ0+Ånon-BC×fnon-BCλ0. Using the mass absorption efficiency and absorption Ångström exponents for BC, OC, and Fe described by Wang et al. (2013), we can further estimate the following parameters: equivalent BC (CBCequiv), maximum BC (CBCmax), estimated BC (CBCest), fraction of light absorption by non-BC ILAPs (insoluble light-absorbing particles) (fnon-BCest), absorption Ångström exponent of non-BC ILAPs  (Ånon-BC), and total absorption Ångström exponent (Åtot). These parameters are defined as follows.

CBCequiv (ng g-1): equivalent BC is the amount of BC that would be needed to produce the total light absorption by all insoluble particles in snow for wavelengths of 300–750 nm.

In this study, a series of 15 Nuclepore filters with a pore size of 0.2 µm (lot no. 7012284, 25 mm, Whatman) loaded with fullerene soot (stock no. 40971, lot no. L20W054, Alfa Aesar, Ward Hill, MA, USA) is used to calibrate the spectrophotometer over the range 0.63–38.6 µg, which typically covers > 75 % of ambient accumulation-mode mass (left panel in Table 1; Schwarz et al., 2012). Fullerene soot is commonly used for calibrating the light transmission and thermal–optical techniques for measuring BC concentrations (Baumgardner et al., 2012). Standard fullerene soot particles are fractal-like aggregates of spherical primary particles with a diameter of ∼50 nm, with a mean density of 1.05 g cm-3 (Moteki et al., 2009). Multiple filters with various loadings are required, as the system response deviates from Beer's law exponential behavior; related equations can be found in Grenfell et al. (2011). Note that uncertainties in mass absorption efficiencies, which range from 2 to 25 m2 g-1, can lead to uncertainty in this technique. Here, we use a mass absorption efficiency of 6.22 m2 g-1 at 525 nm, which is consistent with Doherty et al. (2010) and Grenfell et al. (2011). Figure 3 shows the best-fit curve (solid line) of the loading of the filters at 550 nm. When the filter loading was 0–40 µg cm-2, all measured results were close to the best-fit curve, indicating that the TSI spectrophotometer is stable and accurate in terms of BC mass measurements.

There are several types of thermal–optical methods that can be used to quantify EC and OC concentrations, including two-step temperatures in oxidizing and non-oxidizing atmospheres (Cachier et al., 1989; Xu et al., 2006, 2009b), thermal–optical reflectance (Chow et al., 1993, 2001; Chen et al., 2004), and thermal–optical transmittance (Sharma et al., 2002; Yang and Yu, 2002; Chow et al., 2004). Using an optimized two-step method, Cachier et al. (1989) first confirmed that soot carbon not only comprises EC, but is also mixed with highly condensed organic material. An optimized two-step thermal–optical system has been developed to detect EC and OC concentrations in ice cores (Xu et al., 2006). Here, we use the optimized two-step method based on the thermal–optical technique to measure EC concentrations. In this experiment, quartz-fiber filters were first preheated in a muffle furnace at 350 ∘C to remove organic carbon prior to sampling. All filters were punched to yield appropriately sized samples for analysis. Snow samples were analyzed for EC and OC concentrations using a thermal–optical carbon analyzer (Desert Research Institute, model 2001A), following the thermal–optical reflectance (TOR) protocol of the Interagency Monitoring of Protected Visual Environments (IMPROVE_A). We developed a new method, referred to as the two-step method, to measure the concentrations of BC collected by the quartz-fiber filters. The two-step method is an updated measurement procedure that first extracts an OC fraction below 550 ∘C in a He atmosphere. The volatilized OC is oxidized to CO2, reduced to CH4, and detected by a flame ionization system. Next, two EC fractions (EC1 and EC2) are extracted above 550 ∘C in an atmosphere of 2 % O2 and 98 % He. Detailed procedures can be found in Xu et al. (2006) and Chow et al. (2004). The analytical uncertainty of this method is 15 % for BC and 16 % for OC measured via four parallel ice samples cut lengthways in an ice core with high dust loading (Xu et al., 2009a).

To further assess the accuracy of the TSI system, we use standard fullerene soot and quantify BC concentrations using theoretical calculations for comparison with BC values measured by a laboratory-based TSI spectrophotometer. To ensure the stability and accuracy of the improved TSI spectrophotometer, two individual sets of standard BC filters were used: 0.4 µm Nuclepore and 1 µm quartz-fiber filters. All filters were preheated in a muffle furnace at 350 ∘C to remove organic carbon prior to sampling. A measured amount of BC was mixed into a known volume of ultrapure water. The mixture was then agitated by ultrasound for ∼10 min, and the same volumes of liquid were then filtered through the two types of filters. Using the calculated BC mass, seven filters with gradually increasing BC concentrations were obtained for both the 0.4 µm Nuclepore and 1 µm quartz-fiber filters. Next, all the filters were placed in a dryer for 24 h and then measured using the TSI spectrophotometer. Using the BC mass and the volume of the ultrapure water used for filtration, we can estimate the theoretical BC concentration for each filter. The mass for each filter is listed in Table 1 (right panel).

Assuming a mass absorption cross section (MAC) of BC of 6.22 m2 g-1 at 525 nm, the BC concentrations measured using the TSI spectrophotometer were in good agreement with the theoretical BC values, with a slope of 1.07 (Fig. 4). The BC mass loaded on the Nuclepore filters was approximately equal to that measured by the improved TSI spectrometer, which indicates that the TSI system developed here can accurately measure BC concentrations with the assumed mass absorption efficiency. In contrast, the standard BC mass on the quartz-fiber filters was underestimated by 35 %–45 % using the two-step thermal–optical technique compared with the theoretical value. During the filtration process, we found that the time required to filter liquid snow samples on the 0.4 µm Nuclepore filters was much longer than was the case for the 1 µm quartz-fiber filters. Therefore, we first filtered the melted snow samples on the quartz-fiber filters and then refiltered the snow samples using the 0.4 µm Nuclepore filters. Using this process, BC mass losses can be obtained using the TSI technique, assuming that optical BC is equivalent to thermal EC.

As shown in Fig. 5, the fraction of BC mass collected during the second filtration (0.4 µm filter) ranges from 12 % to 21 % of the total collected mass (filter directly with 0.4 µm filters), as might be expected for the small particles of standard fullerene soot (< 50 nm). This under-sampled fraction decreases with increasing BC mass on the filters, possibly owing to blocking of the filter pores. As a result, the under-sampled fraction of the thermal–optical method was larger than that of the TSI technique, leading to a lower filtration efficiency. Note that these sampling efficiencies are strongly related to the BC size distribution. Therefore, the improved TSI technique developed here is more stable and accurate for measuring pure BC masses, and the data obtained using this method can be used as the standard BC mass. After correcting for systematic biases, the results of both methods were closer to the theoretical BC calculations. Note, however, that the size distribution of the laboratory BC standard was much smaller than those of the atmospheric and seasonal snow samples (Schwarz et al., 2012). Therefore, underestimates caused by the filtration efficiency for ambient BC should be lower than that for the standard BC.

Recent studies have indicated that mineral dust can affect the accurate detection of BC concentrations using the ISSW and thermal–optical methods (Wang et al., 2012; Zhou et al., 2017). To eliminate the large uncertainty and bias due to dust particles, we only used snow samples collected in industrial areas over northeastern China, where the light absorption was dominated by fine-mode ILAPs (e.g., BC and OC; Wang et al., 2013). Hence, most of the snow samples did not contain very large coarse-mode particles, such as mineral and local soil dust.

During the snow field campaign, two series of snow samples were filtered through the Nuclepore and quartz-fiber filters and measured using the TSI and two-step thermal–optical methods (Fig. 6). Result shows that most of the BC values measured by the TSI and two-step thermal–optical methods are close to the 1:1 line in a comparison plot and are generally in good agreement (slope of 1.11, R2=0.93, n=22). However, some BC values in seasonal snow measured by the two-step thermal–optical method are much larger than those measured by the TSI technique. Consequently, for each sample the mean ratio of BC concentrations measured by the two-step method and the TSI spectrophotometer varies from 0.64 to 3.97, with an overall mean of 1.57. This discrepancy arises from two factors. First, Wang et al. (2017) found that snow grain sizes varied considerably (from 0.07 to 1.3 mm) during this snow field campaign. This range is much larger than that recorded in previous studies, owing to snow melting by solar radiation and ILAPs (Hadley and Kirchstetter, 2012; Painter et al., 2013; Yasunari et al., 2013; Pedersen et al., 2015). These results agree well with those of Schwarz et al. (2012), who found that the sizes of BC particles in snow are much larger than those in typical ambient air. Therefore, the sampling efficiency of the quartz-fiber filters could have been significantly higher than expected. The other factor is that the insoluble light-absorbing impurities in seasonal snow over northeast China contained not only BC, but also insoluble organic carbon. This result is consistent with a previous study by Chow et al. (2004), who reported that the charring observed when employing the two-step thermal–optical method at higher temperatures (> 550 ∘C) was incomplete and that certain organic compounds are not completely pyrolyzed below 550 ∘C. Therefore, incomplete charring of absorbed organic compounds by the two-step processes may lead to incompletely pyrolyzed OC on the filters, artificially contributing to the BC concentration. This may explain why the BC concentration measured using the thermal–optical method was higher than that measured using the TSI spectrophotometer.

A comparison of BC concentrations in the atmosphere measured by the ISSW and thermal–optical methods reveals that they are vastly different than that for the snow samples (Fig. 7). Results are in excellent agreement for BC concentrations of < 3 µg m-3. However, biases increased gradually with increasing BC concentrations, leading to two-step-to-TSI ratios as low as 0.5. The BC concentrations of > 3 µg m-3 obtained using the two-step thermal–optical method are much lower than those measured using the improved TSI technique, possibly due to the small particle sizes in the atmosphere, which lead to a lower filtration efficiency. Overall, we conclude that the improved TSI method is more stable and suitable for measuring BC concentrations in both the atmosphere and snow samples compared with the two-step thermal–optical method.

The above results show that the improved TSI method measures BC concentrations in the atmosphere and snow–ice with higher accuracy than two-step thermal optical methods. In this section we investigate the spatial distribution of BC concentrations and their relative light absorption due to BC and non-BC snow impurities in seasonal snow over northeast China during January–February 2014. All BC mass concentrations in surface snow measured by the TSI and thermal–optical methods during the snow field campaigns are listed in Table 2. There was less snowfall in January 2014 than in 2010, and seasonal snow did not cover all of central Inner Mongolia during this time. Thus, we only collected snow samples at site 90. Given that this region is windy, the surface snow collected included drifted and aged snow. The surface BC concentration was 350 ng g-1 in the central Inner Mongolia region. The lowest BC concentrations in surface snow, 55 and 28 ng g-1, were found on the border of northeast China (sites 91–97). We note that there were considerable variations in BC concentrations in these regions. The median BC concentration was 1100 ng g-1 with a range of 520–3900 ng g-1 for surface snow in northeast industrial regions. On 10 February 2014, fresh snow samples were collected in Lanzhou at a mean snow depth of 6–8 cm. The mean BC concentration in these fresh snow samples from Lanzhou was ∼170 ng g-1.

The relative light absorption due to BC and non-BC fractions in seasonal snow measured using the improved TSI technique across northern China is shown in Fig. 8. A similar pattern for the light absorption of BC (∼70 %) and non-BC (∼30 %) from insoluble light-absorbing impurities in surface snow indicates a similar pollution emission source over northeast China. However, the light absorption due to BC in seasonal snow plays a dominant role (43.11 %–88.56 %, with a mean of 73.10 %). The largest BC light absorption was at site 102. This site is located in the central part of Jilin province, which is polluted by heavy industrial activity. For one sample, the light absorption of non-BC impurities in seasonal snow reached 56.89 %, which is the only time it exceeded BC light absorption. Biomass burning and fossil fuel are likely the major emission sources during the winter in Lanzhou, unlike the case over northeast China. These results are consistent with those of Wang et al. (2013), who found that snow particle light absorption was dominated by BC in northeast China in 2010.

Finally, we investigate atmospheric BC mass concentrations and their relative light absorption measured by the TSI spectrophotometer in Lanzhou during 5–25 August 2015. During this experiment, there were no noticeable trends of BC concentrations in Lanzhou. However, a notable feature in Fig. 9 is that the BC mass concentrations at night are generally much higher than during the day (Table 3). The unique topography of Lanzhou likely plays an important role in this phenomenon. Lanzhou is situated in a valley basin with low rainfall, high evaporation, low wind speeds, and a high calm-wind frequency, which often leads to a thick inversion layer in which air pollutants accumulate during the night. The light absorption due to BC in the atmosphere ranges from 62.76 % to 91.84 %, with a mean of 75.43 %.