Field observations were conducted at the Panyu station (23∘00′ N, 113∘21′ E), which is a monitoring site of the Chinese Meteorological Administration (CMA) Atmospheric Watch Network (CAWNET) that is located on the summit of Dazhengang Mountain (approximately 150 m above sea level) in Guangzhou, China. Figure 1 shows the location of the Panyu site, which is situated at the center of the PRD and is separated from residential areas by at least 500 m. Some agricultural fields can be found to the west of the site. Although there were no significant pollution sources nearby, this suburban site was strongly affected by pollutants transported from the urban area of Guangzhou and crop residual fires transported from the rural area of the PRD. The field campaign was conducted from 29 November 2014 to 2 January 2015. During the measurement period, aerosol light scattering and extinction, BC concentration, particle number size distribution (PNSD), OC concentration, and the water-soluble ion concentrations of PM2.5 were continuously monitored.

All instruments were housed inside the second-floor measurement room of a ∼5 m tall, two-story building. The ambient sample was taken on the roof by a 2 m long, 12.7 mm OD stainless steel inlet, and a PM2.5 cyclone sampler was also used. The metal tubing was thermally insulated and maintained at a constant temperature of ∼25 ∘C. A diffusion drier was also used in-line to dry the relative humidity (RH) of the air sample to below 30 % before further analysis.

The AAEBC is widely defined as the uniform representation of the wavelength dependence of a BC particle (Olson et al., 2015). In reality, AAEBC may vary significantly with BC-containing aerosols of different sizes, mixing states, and morphologies (Lack and Langridge, 2013; Scarnato et al., 2013). In fact, some studies showed that the AAE of a large size, pure BC core may be less than 1.0 (Liu et al., 2018) and that the AAE of BC coated with a non-absorbing shell may be larger than that under uniformity (Lack and Cappa, 2010).

It has been suggested that a significant fraction of smaller size particles is non-BC-containing (Cheung et al., 2016; Ma et al., 2017). BC and non-BC materials can also be externally or internally mixed. Although size-resolved BC measurements were not available during this work, we conducted size-resolved V-TDMA measurements at 300 ∘C for 40, 80, 110, 150, 200, and 300 nm during an earlier field campaign (February 2014) at the same site as in this work. At 300 ∘C, all non-BC particles will be completely vaporized (CV), and thus the portion of non-BC particles at such size, denoted as ΦN,CV, can be determined. The average ΦN,CV values were 0.384, 0.181, 0.180, 0.158, 0.143, and 0.137, corresponding to 40, 80, 110, 150, 200, and 300 nm (see Fig. S3), respectively (Cheung et al., 2016; Tan et al., 2016a). The ΦN,CV values for other sizes were interpolated linearly from these six diameters. For particle sizes larger than 300 nm and less than 40 nm, ΦN,CV values were set to 0.137 and 0.384, respectively. Accordingly, the complete distribution of ΦN,CV for the whole PNSD was obtained. The mixing states of BC particles were also estimated here, i.e., the mass portion of externally mixed BC with respect to total BC, denoted as rext. The value of rext was taken as 0.58, which was obtained using an optical closure method during a previous field experiment at this site (Tan et al., 2016a). During the following Mie theory calculation, a fixed refractive index (m̃core=1.80–0.54i, m̃non=1.55–10-7i) was adopted for the whole size range. Accordingly, the calculated BC absorption at 880 nm (Abs880) was 21.869 Mm-1, which is reasonably close to the measured mean value of 21.199 Mm-1. To further validate our calculation scheme (Base Case), we have considered several extreme cases.  

Case 1: BC is completely externally mixed with non-BC particles, i.e., ΦN,CV=0 and rext=1;

When the AAEBC was assumed to be uniform, the campaign-averaged σBrC values were 17.6±13.7 Mm-1 at 370 nm, 9.7±7.9 Mm-1 at 470 nm, 5.8±5.1 Mm-1 at 520 nm, 4.0±3.5 Mm-1 at 590 nm, and 2.3±2.1 Mm-1 at 660 nm. At the corresponding wavelengths, BrC absorption contributed 26.2 % ± 8.5 %, 20.0 % ± 7.3 %, 14.3 % ± 6.5 %, 11.7 % ± 5.3 %, and 7.8 % ± 4.1 % to the total aerosol absorptions. When the AAEBC was applied as the result of the Mie model calculation, the corrected campaign-averaged σabs,BrC values were 23.5±17.7 Mm-1 at 370 nm, 11.8±9.5 Mm-1 at 470 nm, 6.7±5.7 Mm-1 at 520 nm, 4.6±3.9 Mm-1 at 590 nm, and 2.6±2.3 Mm-1 at 660 nm. At the corresponding wavelengths, BrC absorption contributed 34.1 % ± 8.0 %, 23.7 % ± 7.3 %, 16.0 % ± 6.7 %, 13.0 % ± 5.4 %, and 8.7 % ± 4.3 % to the total aerosol absorption (see Fig. 2). Evidently, aerosol light absorption was predominantly due to BC; however, BrC also played a significant role, especially at shorter wavelengths. Table 2 shows the intercomparison of BrC light absorption in the near UV range between this work and other studies in the East Asian region. Clearly, the reported values vary substantially, and our result is toward the lower end of values. Figure S4 displays the time series of particle AAE measured by the Aethalometer, and AAEBC was derived from Mie model calculation. The AAEBC was almost always lower than AAE, indicating appreciable BrC light absorption at the Panyu site.

Theoretically, the magnitude of BC absorptions can be affected by both parts of the complex RIs; thus, AAEBC may also vary with the RIs of both the BC core and coating shell. In fact, RI was also one of the least known properties of BC and other coating materials with negligible absorbing capabilities. The refractive index of the BC core (m̃core) displays a wide range of variations (Liu et al., 2018). Typically, the real and imaginary parts of the RI can vary from 1.5 to 2.0 and 0.5 to 1.1, respectively. In addition, the shell was assumed to consist of non-absorbing material in the core–shell model; i.e., its imaginary RI was set to be close to zero (10-7). The real part of the non-absorbing material RI (m̃non) may vary from 1.35 to 1.6 due to the presence of OA (Redmond and Thompson, 2011; Zhang et al., 2018) and inorganic salts (Erlick et al., 2011). Hence, it is necessary to investigate the uncertainties associated with the variations in AAEBC by varying the RIs of both the BC core and the non-absorbing materials.

Figure 3 shows the impacts of RI on the evaluations of AAEBC based on core–shell and external configuration, where the RI of the BC core was set to be constant, i.e., m̃core=1.80–0.54i, and the real part of m̃non varied from 1.35 to 1.6 at an interval of 0.05, with the imaginary part of m̃non set at 10-7. As shown in Fig. 3a, the calculated AAEBC for the core–shell model was higher than 1.0 at longer wavelengths (520 to 880 nm) and lower than 1.0 at shorter wavelengths (370 to 520 nm) (the red line in Fig. 3 denotes AAEBC=1). The averaged AAEBC,370-520nm ranged from 0.84 to 0.87, and the AAEBC,520-880nm ranged from 1.07 to 1.15, indicating that the AAEBC,520-880nm appeared to be more sensitive to the shell's real part than AAEBC,370-520nm. Even if the shell material was assumed to be non-absorbing, the variation in the real RI of the shell, which was referred to as the real part of m̃non, still led to changes in the shell's refractivity and correspondingly altered its lensing effect, causing a change in AAEBC. Meanwhile, AAEBC,370-520nm and AAEBC,520-880nm generally increased with an increasing real part of the shell. In Fig. 3b, under the externally mixed conditions, AAEBC,370-520nm and AAEBC,520-880nm were both less than 1.0. The average AAEBC,370-520nm was 0.33, and the average AAEBC,520-880nm was 0.63. These values were far less than the values under core–shell mixture conditions. In the external mixture model, the BC core and non-light-absorbing materials were assumed to exist dependently, and then the optical properties of these two components were considered separately. Therefore, altering the real part of the externally mixed non-absorbing material would not affect the light absorption property of the BC core or AAEBC.

The impacts of the BC core on AAEBC are shown in Fig. 4, where the refractive index of non-light-absorbing materials was assumed to be m̃non=1.5510-7i and m̃non was wavelength-independent. Figure 4 was obtained with a core–shell mixture model (Fig. 4a and b) and an external mixture model (Fig. 4c and d) by varying the real part of m̃core from 1.5 to 2.0 with a step of 0.05 and varying the imaginary part of the m̃core from 0.4 to 1.0 with a step of 0.05. As shown in Fig. 4a and b, for the core–shell mixture, the averaged AAEBC,370-520nm ranged from 0.55 to 0.99 and the averaged AAEBC,520-880nm ranged from 0.84 to 1.27. The AAEBC at a certain wavelength generally increased when increasing the real part of m̃core but decreased when increasing the imaginary part of m̃core. The AAEBC appeared to be more sensitive to the imaginary part of m̃core than the real part of m̃core because the imaginary part of m̃core was directly related to the light-absorbing properties of particles. In Fig. 4c and d, for the external mixture, the averaged AAEBC,370-520nm ranged from 0.04 to 0.45 and the averaged AAEBC,520-880nm ranged from 0.28 to 0.79, while the averaged AAEBC,370-520nm and AAEBC,520-880nm were both less than 1.0. Similar to the core–shell mixture, the AAEBC,520-880nm increased when increasing the real part of m̃core but decreased when increasing the imaginary part of m̃core. However, the variation patterns of AAEBC,370-520nm were different from those of AAEBC,520-880nm. The AAEBC,370-520nm values were not changed by altering the real part of m̃core within the low imaginary part of m̃core, whereas the AAEBC,370-520nm values still increased when increasing the real part of m̃core within the high imaginary part of m̃core. A possible explanation was that the externally mixed BC core had weak light absorption within the low imaginary part of m̃core, causing the AAEBC,370-520nm values to be insensitive to the real part of m̃core. The AAEBC,520-880nm values were higher than the AAEBC,370-520nm values regardless of whether they were for the core–shell mixture or the external mixture. In addition, the AAEBC values conducted by the core–shell mixture were higher than those conducted by the external mixture.

Figure 4 demonstrated that the variation in the imaginary RI of the BC core had the most significant impact on the estimated AAEBC, indicating that the chemical component of BC emitted from different sources led to a large uncertainty in AAEBC estimation. At the same time, the influence arising from varying the real RI of the BC core was relatively moderate. Nevertheless, Fig. 3 shows that change in the real RI of the non-absorbing materials caused the least or no impact compared to the variations in the complex RI of the BC core.

It should be pointed out that most BC-containing particles are often observed as being fractal rather than spherical in shape (Katrinak et al., 1993). Because the Mie model assumes that all particles are spherical, it may lead to potential uncertainty for the estimation of AAEBC and BrC absorption contributions. Moreover, the externally mixed soot aggregates were “chain-like” or “puff-like” in the PRD dry season (Feng et al., 2010), in which the fractal dimension (Df) was between 1.5 and 2.0. Coating soot aggregates were likely spheres (Df approaches 3) from the high-resolution transmission electron microscopy (TEM) measurements taken in Hong Kong (Zhou et al., 2014). A soot aggregate sensitivity study with the superposition T-matrix method indicated that using the assumption of volume-equivalent spheres for the soot aggregates may result in an overestimation of approximately up to 15 % and an underestimation of approximately up to 50 % in the predicted 870 nm light absorption when the Df is between 1.5 and 3.0 (Liu et al., 2008). However, it should be recognized that the complex shapes or positions of the BC core inside the particle make it impractical to be numerically simulated in the exact details. By far the Mie model with a core–shell configuration would be the most practical and effective simulation scheme for BC particle optical property simulation.

Furthermore, we have performed Monte Carlo simulations to evaluate the uncertainties of the Mie calculation performed during this work. In the simulation, a sequence of random numbers or errors were applied to the input parameters, and then the corresponding uncertainties of particle light absorption and AAEBC were computed using the Mie model. Five hundred reiterations were conducted during the simulation such that the random errors will be normally distributed. The standard deviations (σ) of all input parameters are listed in Table S1. In order to cover the effect of extreme value, we used a range of ±3σ, or a confidence level of 99 %, in the Monte Carlo simulation. Table S2 lists the Monte Carlo simulation results, i.e., the average relative standard deviations (σMie) of the calculated BC light absorption at 880 nm (Abs880), AAEBC,370-520, and AAEBC,520-880. The uncertainties of the calculated Abs880, AAEBC,370-520, and AAEBC,520-880 at two times of σMie, i.e., at a confidence level of 95 %, were approximately ±31 %, ±16 %, and ±13 %, respectively. Figure S5a shows the time series of the uncertainties of Abs880, AAEBC,370-520, and AAEBC,520-880 from a Monte Carlo simulation for the campaign period. These uncertainties will certainly be propagated into the calculated BrC absorption contributions, too. Hence, we also estimated the corresponding uncertainties in the BrC absorption contribution results, as shown in Fig. S5b. Accordingly, the averaged lower limits of BrC absorption contributions were 26.8 % ± 9.1 % at 370 nm, 17.5 % ± 8.1 % at 470 nm, 10.1 % ± 7.3 % at 520 nm, 8.5 % ± 5.8 % at 590 nm, and 5.3 % ± 4.5 % at 660 nm, and the averaged upper limits of BrC absorption contribution ratios were 40.7 % ± 7.2 % at 370 nm, 29.5 % ± 6.7 % at 470 nm, 21.1 % ± 6.2 % at 520 nm, 17.3 % ± 5.2 % at 590 nm, and 12.0 % ± 4.1 % at 660 nm.

Globally, BrC has been observed to be highly correlated with biomass and biofuel burning emissions (Laskin et al., 2015). Since large quantities of sylvite are present in biomass burning particles, the K+ abundance has often been used as a biomass burning tracer (Levine, 1991). Figure 5 presents the time series of the OC mass concentration, K+ concentration, and BrC absorption from 29 November 2014 to 2 January 2015 at the Panyu site. The range of the OC concentrations obtained from the OC/EC online analyzer was from 1.5 to 65.2 µg cm-3, and the campaign average was 12.5±7.3 µg cm-3. The BrC absorption hourly mean data were between 0.2 and 123.2 Mm-1, and the campaign average was 23.5±17.7 Mm-1. On the other hand, the average K+ concentration was 1.0±0.7 µg cm-3 (ranging from 0 to 5.4 µg cm-3). Clearly, similar trends among OC, K+, and BrC absorption can be seen during this field campaign (Fig. 5).

To investigate the origins of these observed OC, K+, and BrC, wind rose plots (as shown in Fig. 6) were generated for OC, K+, and BrC absorption, respectively. All three panels of Fig. 6 consistently show that the three substances were associated with the same wind pattern. For the entire campaign period, the highest values of OC, K+, and σabs,BrC,370nm were mostly associated with southwesterly winds with a relatively low wind speed (∼2 m s-1). The relatively higher OC and K+ concentrations were highly related to the seasonal straw burning in the countryside of the PRD located to the west of the Panyu station. In contrast, OC and K+ concentrations during periods with easterly winds were substantially lower than those during periods with westerly winds. The wind rose plot of σabs,BrC,370nm is shown in Fig. 6c. Similar to OC and K+, σabs,BrC,370nm showed higher values under weak (<2 m s-1) westerly winds and lower values from the north and south, indicating that BrC absorption was likely attributed to local sources and was accumulated under calm wind conditions. Figure S6 shows the 3 d backward trajectory and the fire counts for 5 to 7 (Fig. S6a), 12 to 14 (Fig. S6b), and 24 to 26 (Fig. S6c) in November 2014, representing low-loading, moderate-loading, and high-loading period. Clearly, the high-loading period concurred with stagnant air movement and higher fire counts, indicating the contribution from open fire burning sources. However, there was a detectable difference among the three rose plots of Fig. 6 in the maximum concentration direction. A possible explanation was that although biomass burning emissions were believed to be the dominant and primary source of OC, K+, and BrC, their emission ratios were highly variable and may change with the type of biofuel and burning condition and may even vary during different stages of burning (Burling et al., 2012). Although biomass burning emissions contain substantial light-absorbing BrC, further atmospheric aging processes may significantly reduce its light-absorbing capability (Satish et al., 2017). Moreover, secondary formation may also lead to BrC formation inside these primary aerosols, such as humic-like substances formed through aqueous-phase reactions, which have been suggested to be an important component of BrC (Andreae and Gelencser, 2006).

To further explore the possible sources of BrC optical absorption, the diurnal variations in OC, K+, σabs,BrC,370nm, and σabs,BrC,370nm/OC values are plotted in Fig. 7. The diurnal variation in OC at the Panyu site appeared to be dominated by the development of the planetary boundary layer (PBL) height; i.e., primary emissions accumulated at night and were swiftly diluted by vertical mixing in the morning. The slight increase in OC in the afternoon indicated that photochemistry may have still weakly contributed to SOA formation. Figure 7b shows the diurnal variation in K+. Unlike OC, K+ shows a small peak at approximately 06:00, which was consistent with breakfast time and was very likely due to cooking activities using biofuel. No lunch and dinnertime K+ peaks were observed. The most likely explanation is that the boundary layer height is much higher during lunch and dinnertime than in the early morning, providing a much better atmospheric diffusion condition for air pollutants. It is still a common practice to collect straw as biofuel in local rural areas, which can be visually spotted but is not heavily utilized in the region. However, the diurnal profile of σabs,BrC,370nm (see Fig. 7c) shows the combined features of OC and K+ since both primary and secondary processes affect its intensity. The nighttime increasing trend was most likely attributed to straw burning activities in early winter in nearby rural areas that continued to accumulate within the shallow PBL (Jiang et al., 2013). σabs,BrC,370nm/OC, i.e., the mass absorption coefficient of BrC (MACBrC) (Fig. 7d), showed a relatively flat pattern, with a pronounced dip in the afternoon and higher values at nighttime, which was likely due to enhanced primary emissions and stable stratification at nighttime. Declining trends during the late morning and afternoon hours indicated that the aging process and photochemical production may reduce the light-absorbing capacity of BrC (Qin et al., 2018).

Furthermore, Fig. 8 shows the linear regression analysis results used to evaluate the correlations of σabs,BrC,370nm with the OC, K+, Ca2+, Mg2+, Cl-, SO42-, NO3-, and NH4+ concentrations. The best correlations can be found between σabs,BrC,370nm and K+ (R2=0.6148), followed by those between σabs,BrC,370nm and OC (R2=0.4514), NO3- (R2=0.4224), and NH4+ (R2=0.4656). Source apportionment analysis of OA and BrC absorption in Beijing and Guangzhou illustrated that biomass burning organic aerosols (BBOAs) correlated well with BrC light absorption (Qin et al., 2018; Xie et al., 2019). Thus, the significant correlation between BrC absorption and K+ reaffirmed that biomass burning was the crucial emission source of BrC observed in this work. Although the geographic location of the observation site was situated in a coastal area and K+ could also be found in sea salt (Pio et al., 2008), it should be noted that the prevailing wind direction during winter was from the north (see Fig. S6), which drives maritime air parcels away from the site. Hence, the effect of sea salt and crustal materials to K+ was slight, which was demonstrated in the Supplement as shown in Fig. S7. Other earlier studies also suggested that the sea salt contribution to the K+ concentrations of PM2.5 was trivial in the PRD region during the winter (Lai et al., 2007). Another possible K+ source was coal combustion. The coal consumption in the PRD region was dominated by coal-fired power plants. The emission from power plants was usually very steady and was less likely to affect the diurnal correlation between K+ and BrC absorption. As shown in Fig. S8, the ratios of K+/PM2.5 vary approximately from 0.015 and 0.020 and the diurnal profile of K+/PM2.5 shows very little variation. Yu et al. (2018) have suggested that K+ usually accounted for 2.34 %–5.49 % of PM2.5 in the laboratory biomass burning study. However, K+ was normally lower than 1 % of coal combustion PM2.5. Therefore, the ratio range of K+ to PM2.5 observed in this work likely indicated aged biomass burning particles. Both nitrogen oxides (NOx) and ammonia (NH3) can be found in biomass burning plumes (Andreae and Merlet, 2001). For NO3- and NH4+, nitrate can be converted from NOx through atmospheric reactions, and ammonium may originate from NH3. However, similar to the diurnal variation in σabs,BrC,370nm, diurnal variations in NH4+ and NO3- also increased in the afternoon and appeared at nighttime in Fig. S8. However, NO3-/PM2.5 and NH4+/PM2.5 reached their peaks at noon, indicating that ammonium nitrate formed from the secondary reaction at this time. Along with the reduced boundary layer height and ambient temperature, NO3- was accumulated until the photochemical reaction stopped at night. The diurnal variation in NH4+ was similar to that in NO3- due to the acid–base neutralization reaction. The overlapping of the σabs,BrC,370nm, NH4+, and NO3- diurnal variations would lead to a significant correlation between BrC absorption and NO3- or NH4+. High concentrations of Ca2+ and Mg2+ are often found in dust-related aerosols (Lee et al., 1999). σabs,BrC,370nm showed poor correlations with both Ca2+ and Mg2+, indicating that dust-related aerosol components contribute insignificantly to the total aerosol mass loading and, thus, dust may not affect the AAE differentiation method used in this work. Although sulfur dioxide (SO2) may also be emitted by biomass burning, SO42- is often believed to be secondary in nature, and the presence of other intense SO2 sources (e.g., automobile and industrial emissions) further reduces the correlation between BrC and SO42-. Sources of Cl- include both combustion and sea salt spray (Waldman et al., 1991). Although the prevailing wintertime wind direction was from the north, sea salt can still be carried to the site by a weak sea breeze, and thus Cl- may not show considerable correlation with BrC.

The radiative effects of aerosol scattering, BrC absorption, and BC absorption were investigated by the SBDART model. For each investigated variable under cloud-free conditions, we run the model twice to calculate the DRF at the TOA with and without the investigated variable. Accordingly, the difference of ΔF between the two simulations was considered as the radiative effect of the investigated variable. The results showed that the average radiative forcings at the TOA by scattering, BrC absorption, and BC absorption were -21.4±5.5 W m-2, 2.3±1.8 W m-2, and 10.9±5.1 W m-2, respectively. Furthermore, BrC absorption was attributed to 15.8 % ± 4.4 % of the warming effect caused by aerosol light absorption, demonstrating the nonnegligible role of BrC in radiative forcing evaluation.

We also calculated the BrC radiative forcing efficiency (RFE) under various SSAs (ranging from 0.70 to 0.99) at three wavelengths, i.e., 440, 675, and 870 nm. The RFE was denoted as the radiative forcing normalized by the AOD. The average AOD and ASY at the three wavelengths were 0.365 and 0.691 at 440 nm, 0.212 and 0.632 at 675 nm, and 0.154 and 0.619 at 870 nm. A solar zenith angle of 55∘ and an average shortwave broadband surface albedo (0.119) were used in the calculation. The results were plotted as a set of RFE lookup charts as a function of the surface BrC absorption contribution (see Fig. 9).

In general, for any wavelength, the RFE increased with increasing BrC absorption contribution for a certain SSA, indicating that BrC was a more efficient radiative forcing agent due to the preferential absorbance of BrC in a shorter wavelength range. However, for a certain BrC absorption contribution, the RFE increased with decreasing SSA; i.e., a higher portion of light-absorbing aerosol components can lead to more efficient radiative forcing. The trend among panels (a), (b), and (c) in Fig. 9 demonstrated that the effect of BrC absorption contribution on RFE was wavelength-dependent; i.e., BrC was a weaker radiative forcing agent at longer wavelengths, which is also consistent with the wavelength-dependent light-absorbing property of BrC. The red stars in Fig. 9 denote the average SSA and BrC absorption contribution conditions during this campaign, i.e., 0.029 W m-2 per unit AOD at 440 nm (Fig. 9a), 0.007 W m-2 per unit AOD at 675 nm (Fig. 9b), and 0.0002 W m-2 per unit AOD at 870 nm (Fig. 9c). These results suggested that the average value of RFE decreased distinctly from 440 to 870 nm not only because of the lower BrC absorption contribution but also because of the wavelength dependence of the BrC RFE. It should also be noted that the simulations were based on SSA measured under dry conditions. Under the typical ambient conditions of the PRD, the SSA might be markedly enhanced by aerosol water uptake (Jung et al., 2009), and then the BrC radiative forcing efficiency might be less. Moreover, Fig. 9 also serves as a lookup table to conveniently assess the BrC radiative forcing efficiency at different wavelengths with different BrC absorption contributions for a certain SSA.