Amplification of black carbon light absorption induced by atmospheric aging: temporal variation at seasonal and diel scales in urban Guangzhou

Abstract

Black carbon (BC) aerosols have been widely recognized as a vital climate forcer in the atmosphere. Amplification of light absorption can occur due to coatings on BC during atmospheric aging, an effect that remains uncertain in accessing the radiative forcing of BC. Existing studies on the absorption enhancement factor ($E_{\mathrm{abs}}$) have poor coverage on both seasonal and diurnal scales. In this study, we applied a recently developed minimum $R$ squared (MRS) method, which can cover both seasonal and diurnal scales, for $E_{\mathrm{abs}}$ quantification. Using field measurement data in Guangzhou, the aims of this study are to explore (1) the temporal dynamics of BC optical properties at seasonal (wet season, 31 July–10 September; dry season, 15 November 2017–15 January 2018) and diel scales (1 h time resolution) in the typical urban environment and (2) the influencing factors on$E_{\mathrm{abs}}$ temporal variability. Mass absorption efficiency at 520 nm by primary aerosols (MAE${}_{\mathrm{p}\mathrm{520}}$) determined by the MRS method exhibited a strong seasonality (8.6 m${}^{\mathrm{2}}$ g${}^{-\mathrm{1}}$ in the wet season and 16.8 m${}^{\mathrm{2}}$ g${}^{-\mathrm{1}}$ in the dry season). $E_{\mathrm{abs}\mathrm{520}}$ was higher in the wet season ($\mathrm{1.51}\pm \mathrm{0.50}$) and lower in the dry season ($\mathrm{1.29}\pm \mathrm{0.28}$). Absorption Ångström exponent (AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$) in the dry season ($\mathrm{1.46}\pm \mathrm{0.12}$) was higher than that in the wet season ($\mathrm{1.37}\pm \mathrm{0.10}$). Collective evidence showed that the active biomass burning (BB) in the dry season effectively altered the optical properties of BC, leading to elevated MAE, MAE${}_{\mathrm{p}}$ and AAE in the dry season compared to those in the wet season. Diurnal $E_{\mathrm{abs}\mathrm{520}}$ was positively correlated with AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ ($R^{\mathrm{2}}=\mathrm{0.71}$) and negatively correlated with the AE33 aerosol loading compensation parameter ($k$) ($R^{\mathrm{2}}=\mathrm{0.74}$) in the wet season, but these correlations were significantly weaker in the dry season, which may be related to the impact of BB. This result suggests that during the wet season, the lensing effect was more likely dominating the AAE diurnal variability rather than the contribution from brown carbon (BrC). Secondary processing can affect $E_{\mathrm{abs}}$ diurnal dynamics. The $E_{\mathrm{abs}\mathrm{520}}$ exhibited a clear dependency on the ratio of secondary organic carbon to organic carbon ($\mathrm{SOC}/\mathrm{OC}$), confirming the contribution of secondary organic aerosols to$E_{\mathrm{abs}}$; $E_{\mathrm{abs}\mathrm{520}}$ correlated well with nitrate and showed a clear dependence on temperature. This new finding implies that gas–particle partitioning of semivolatile compounds may potentially play an important role in steering the diurnal fluctuation of $E_{\mathrm{abs}\mathrm{520}}$. In the dry season, the diurnal variability in $E_{\mathrm{abs}\mathrm{520}}$ was associated with photochemical aging as evidenced by the good correlation ($R^{\mathrm{2}}=$0.69) between oxidant concentrations (${\mathrm{O}}_x={\mathrm{O}}_{\mathrm{3}}+{\mathrm{NO}}_{\mathrm{2}}$) and $E_{\mathrm{abs}\mathrm{520}}$.

1 Introduction

Atmospheric aerosols have received great attention in recent years due to their global climatic effects and environmental effects (Anderson et al., 2003). Carbonaceous aerosols account for a large fraction of the global aerosol mass as the main light-absorbing materials in aerosols (Kanakidou et al., 2005; Bond and Bergstrom, 2006). Black carbon (BC), which originates from incomplete combustion of hydrocarbon fuels (Johansson et al., 2018), is the dominating fraction of light-absorbing carbonaceous aerosols. BC has been widely recognized not only as an air pollutant that poses a threat to public health (Grahame et al., 2014; Apte et al., 2015) but also as an essential climate forcer (Chung and Seinfeld, 2002). The BC burden in the atmosphere increased substantially in recent years as evidenced by the ice core samples (Ruppel et al., 2014) and sediment cores from the eastern China marginal seas (Fang et al., 2018). The growing abundance of BC in the atmosphere leads to elevated environmental impacts. BC has been regarded as the third most important climate forcer after carbon dioxide and methane (IPCC, 2013). On a global scale, BC can heat the atmospheric directly owing to its strong light absorption across the solar spectrum (Bond and Bergstrom, 2006), thus contributing to the warming effect (Bond et al., 2013). On a regional scale, BC deposition on ice and snow can reduce surface albedo, leading to glacier melting (Gertler et al., 2016; Kopacz et al., 2011; Flanner et al., 2007; He et al., 2018; Hansen and Nazarenko, 2004), especially at high-altitude regions such as the Tibetan Plateau (Ming et al., 2008). On a local scale, BC can modify planetary boundary layer meteorology that leads to the “dome effect”, and thus enhances local pollution indirectly (Ding et al., 2016; Wilcox et al., 2016). At the microscale, BC was found to play a key role in the photochemical aging of soot by initiating the oxidation of OC (M. Li et al., 2018). In addition, BC can indirectly affect the climate by altering cloud formation and cloud cover (Nenes et al., 2002; Koch and Del Genio, 2010; Kaufman and Koren, 2006; Albrecht, 1989).

However, large uncertainties still exist in estimating the radiative forcing of BC (Bond et al., 2013). The gap largely arises from the limited characterization of BC mixing state and optical properties in the atmosphere (Fuller et al., 1999; Jacobson, 2001; Nordmann et al., 2014). BC is chemically inert, but morphology transformation is unavoidable once emitted into the atmosphere. A recent study suggested that BC restructuring during aging can be divided into two steps (Pei et al., 2018). First, the void of the BC particles will be filled by the aging-induced materials. Once filled, further accumulation of organic and inorganic coating materials leads to the growth of particle size. Ma et al. (2013) reported soot restructuring during water evaporation in a laboratory study. Such morphology transformation leads to alternation of BC optical properties, as evidenced by a number of numerical studies (Fuller et al., 1999; Bond et al., 2006; F. Liu et al., 2016; Zhang et al., 2017; Lefevre et al., 2019), laboratory experiments (Schnaiter et al., 2005; Zhang et al., 2008; Xue et al., 2009; Shiraiwa et al., 2010; Metcalf et al., 2013; Wei et al., 2013; Chen et al., 2015) and field studies (Knox et al., 2009; Cappa et al., 2012; Lack et al., 2012b; Liu et al., 2015, 2019b; D. Liu et al., 2017). The presence of coating materials on BC leads to the increase in mass absorption efficiency (MAE) through the lensing effect (Schwarz et al., 2008b). Besides coating thickness, the magnitude of light absorption enhancement by the lensing effect also depends on the optical properties of the coating materials. A coating of brown carbon (BrC) can further amplify the light absorption compared to a transparent coatings (Lack and Cappa, 2010). Recent studies suggested that BC mixing state diversity also affects the bulk $E_{\mathrm{abs}}$ (Fierce et al., 2016; Matsui et al., 2018; Cappa et al., 2019).

The total BC light absorption (${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}$) after aging can be segregated into primary absorption (${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{pri}})$ by the BC core and the additional absorption (${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}$) due to the presence of a coating:

1$${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}={\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{pri}}+{\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}.$$ The key parameter for light absorption enhancement, $E_{\mathrm{abs}}$, can be calculated from 2$$E_{\mathrm{abs}}={\displaystyle \frac{{\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}}{{\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{pri}}}}={\displaystyle \frac{{\mathrm{MAE}}_{\mathrm{t}}}{{\mathrm{MAE}}_{\mathrm{p}}}},$$ where MAE${}_{\mathrm{t}}$ is the MAE of coated BC, 3$${\mathrm{MAE}}_{\mathrm{t}}={\displaystyle \frac{{\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}}{\mathrm{EC}}},$$ and MAE${}_{\mathrm{p}}$ represents the MAE of BC when fleshly emitted, 4$${\mathrm{MAE}}_{\mathrm{p}}={\displaystyle \frac{{\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{pri}}}{\mathrm{EC}}}.$$ EC in Eqs. (3) and (4) represents elemental carbon (EC) mass concentration determined by the thermo-optical analysis method (Wu et al., 2012), which can be considered a surrogate of BC mass concentration. The atmospheric aging process can lead to BC $E_{\mathrm{abs}}$ values larger than 1 due to the increase in MAE${}_{\mathrm{t}}$.

Three technical approaches have been applied for $E_{\mathrm{abs}}$ quantification as summarized in Table 1. The first approach is to use a thermal denuder (TD) upstream of the instrument that measures ${\mathit{\sigma }}_{\mathrm{abs}}$ (e.g., PAS, photoacoustic spectrometer). By measuring the denuded and ambient sample in rotation with a desired interval (e.g., 5 min), ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}$ and ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{pri}}$ can be obtained to determine $E_{\mathrm{abs}}$ following Eq. (2). Particle loss in the TD is unavoidable and needs to be accounted for (Burtscher et al., 2001). The advantage of the TD is its ability to obtain high-time-resolution data (Cappa et al., 2012; Lack et al., 2012b; Liu et al., 2015; D. Liu et al., 2017). But the TD has its limitations. First, the TD is not suitable for long-term measurements (e.g., most studies last for a few weeks as shown in Table 2). Second, the selection of working temperature depends on the sample and varies by sampling site, which can largely affect the $E_{\mathrm{abs}}$ measurement results. As a result, a universal optimal TD working temperature does not exist. If the temperature is too low, the coating materials cannot be fully vaporized. The study by Ma et al. (2020) showed that coating materials still account for 60 % of particle mass after thermally denuding at a temperature of 280 ${}^{\circ }$C, implying the incomplete removal of coating materials. On the other hand, if the temperature is too high, pyrolysis would occur (Irwin et al., 2013), leading to a biased $E_{\mathrm{abs}}$ measurement. For example, G.-L. Li et al. (2018) explore the variability of TD temperature on $E_{\mathrm{abs}}$ determination in Hong Kong. For a TD temperature of 50 to 200 ${}^{\circ }$C, $E_{\mathrm{abs}}$ ranges from 1.02 to 1.20. $E_{\mathrm{abs}}$ reaches 1.6 for a TD temperature of 280 ${}^{\circ }$C. Third, the TD approach cannot perfectly reverse the morphology transformation of BC from the aged state back to the freshly emitted state. Previous studies have shown that the chain-like aggregate morphology of nascent BC cannot be restored after thermally denuding the coatings on the reconstructed BC core, which tends to be more compact and spherical (Bambha et al., 2013; Ghazi and Olfert, 2013). In addition, the high cost of a TD-PAS system thwarts its wider applications in field studies.

Table 1

Comparisons of three $E_{\mathrm{abs}}$ determination approaches.

Approach		Time resolution	Temporal coverage	$E_{\mathrm{abs}}$ determination	Instruments	Advantages	Limitations
TD		minutes	weeks	$E_{\mathrm{abs}}=\frac{{\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}}{{\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{pri}}}$	TD $+$ PAS	very high time resolution	TD temperature selection is tricky; denuded particle morphology different from emission
AFD		daily	years		filter sampler $+$ offline OC/EC analyzer	can be applied on archived filter samples	labor intensive; only remove soluble coating; low time resolution
MAE	MAE $+$ MRS	hourly	years	$E_{\mathrm{abs}}=\frac{{\mathrm{MAE}}_{\mathrm{t}}}{{\mathrm{MAE}}_{\mathrm{p}}}$	Aeth $+$ online OC/EC analyzer	high time resolution	MRS has minimum data points requirement, not suitable for a small dataset
	MAE $+$ empirical	hourly	months		Aeth $+$ online OC/EC analyzer		empirical MAE${}_{\mathrm{p}}$ could be unrealistic for the sampling site
	MAE $+$ SP	hourly	weeks		PAS/Aeth $+$SP2/SP-AMS		expensive instruments; limited sampling duration

The second approach for $E_{\mathrm{abs}}$ determination is aerosol filter filtration–dissolution (AFD). AFD removes coatings on BC using water and organic solvents (X. Cui et al., 2016). The advantage of AFD is that this method can be applied on historical filters archived by long-term large-scale speciation sampling networks. It opens up a new path to retrieve the historical $E_{\mathrm{abs}}$ from datasets with large temporal and spatial coverage. The limitation mainly arises from the AFD treatment process, which only removes the soluble part of the coating. The AFD treatment process is also labor intensive. The time resolution of $E_{\mathrm{abs}}$ by AFD depends on the interval of filter sampling, which has a typical sampling time of 24 h, making it difficult to study the diurnal pattern of $E_{\mathrm{abs}}$.

The third approach is the MAE method. $E_{\mathrm{abs}}$ is quantified from the ratio of MAE${}_{\mathrm{t}}$ to MAE${}_{\mathrm{p}}$ as shown in Eq. (2). Since MAE${}_{\mathrm{t}}$ can be obtained from ambient measurements, the determination of MAE${}_{\mathrm{p}}$ is the key to this approach. One way is to adopt empirical MAE${}_{\mathrm{p}}$ in the literature (F. Cui et al., 2016) (abbreviated as MAE $+$ empirical hereafter). Since the real-world MAE${}_{\mathrm{p}}$ could be highly diverse by different sources and varies temporally and spatially (Roden et al., 2006; Adler et al., 2010; Shen et al., 2013; McMeeking et al., 2014; Healy et al., 2015; Cheng et al., 2016; Weyant et al., 2016; Dastanpour et al., 2017; Radney et al., 2017; Conrad and Johnson, 2019), an empirical MAE${}_{\mathrm{p}}$ at one site might not be applicable at other sites.

Another method to determine MAE${}_{\mathrm{p}}$ is to combine ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}$ measurements with single-particle measurements to provide the mixing state of BC, e.g., single-particle soot photometer (SP2) or soot particle aerosol mass spectrometer (SP-AMS). This method is abbreviated as MAE $+$ SP hereafter for easy reference. The lag time between the incandescent signal and scattering in SP2 can be used to differentiate thickly coated BC and bare BC. The intercept of the linear regression between MAE ($y$ axis) against the number fraction of aged BC ($r_{\mathrm{aged}}$, $x$ axis) represents MAE${}_{\mathrm{p}}$ (Lan et al., 2013; Wang et al., 2014). This method only considers $E_{\mathrm{abs}}$ dependency on the number fraction of aged particles and ignores the coating thickness of the aged particles; thus, it is only valid for a limited period when coating thickness and size distribution are relatively stable. An improved method for MAE${}_{\mathrm{p}}$ determination by SP2 is utilizing the rBC size distribution to calculate the MAE${}_{\mathrm{p}}$ by the Mie model (D. Liu et al., 2017; Wang et al., 2018a, b).

A recently developed approach, the minimum $R$ squared method (MRS), can be applied to MAE${}_{\mathrm{p}}$ determination using elemental carbon as a tracer (Wu et al., 2018). MRS is a statistical approach, and MAE${}_{\mathrm{p}}$ can be determined in a quantitative manner that minimizes the arbitrariness in MAE${}_{\mathrm{p}}$ estimation by the traditional approach. Application of MRS for $E_{\mathrm{abs}}$ determination is abbreviated as MAE $+$ MRS for easy reference. As summarized in Table 1, the TD approach has high time resolution but limited sampling duration, while the AFD approach has a long sampling duration but low time resolution. As a result, studies of $E_{\mathrm{abs}}$ with both high time resolution and long sampling duration are limited, leading to a lack of knowledge on $E_{\mathrm{abs}}$ variability on both seasonal and diel scales. To fill this knowledge gap, the aims of this study are (1) to explore the temporal dynamics of $E_{\mathrm{abs}}$ on both seasonal and diel scales using the recently developed MRS approach and (2) to investigate the influencing factors on $E_{\mathrm{abs}}$ temporal variability, including photochemical aging, biomass burning (BB) and BC mixing state. In this study, field measurements with 1 h time resolution were conducted in urban Guangzhou, a typical megacity in southern China in both wet (31 July–10 September 2017) and dry (15 November 2017–15 January 2018) seasons. Abbreviations used in this paper are listed in Table A1 in Appendix A for easy reference.

2 Field measurements and data analysis methods

2.1 Characteristics of the observation site

As shown in Fig. 1, sampling for this study was conducted at Jinan University (JNU) atmospheric supersite (23.13${}^{\circ }$ N, 113.35${}^{\circ }$ E; 40 m above sea level), which is located in Tianhe District, downtown Guangzhou. The site is on top of the library building and surrounded by teaching and residential areas. The campus is surrounded by the three busiest roads of the city (Fig. S1 in the Supplement), and traffic emissions are a major source of primary emissions. Guangzhou is located in southern China and is also the geographical center of Guangdong Province. There are limited industrial pollution sources around the sampling site, and thus this site can represent the typical urban environment in the Pearl River Delta (PRD) region.

Figure 1

The location of the observation site (JNU). Panels (a), (b), (c) and (d) show the box plots of EC, OC, $\mathrm{OC}/\mathrm{EC}$ and AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ during wet (July–September) and dry (November–January) seasons, respectively. Red circles represent the seasonal average. The line inside the box indicates the median. Upper and lower boundaries of the box represent the 75th and the 25th percentiles; the whiskers above and below each box represent the 95th and 5th percentiles.

[Figure omitted. See PDF]

The subtropical climate of the PRD is strongly affected by two monsoon systems: South China Sea (SCS) monsoon and northeast monsoon. April to May is the transition period for the northeast monsoon to the SCS monsoon. June to September is the SCS-monsoon-dominated period (wet season). The southern prevailing wind brings clean and humid air masses from the vast ocean. October is the transition period for the SCS monsoon to the northeast monsoon. November to March is the northeast-monsoon-dominated period (dry season). The northeastern prevailing wind brings polluted air masses from the more economically developed regions in eastern Asia. This study included two sampling periods: 31 July–10 September 2017 and 15 November 2017–7 January 2018, corresponding to the wet and dry seasons, respectively.

2.2 Light absorption measurements

A dual-spot Aethalometer (model AE33, Magee Scientific, Berkeley, CA, USA) was used for ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}$ determination. Aethalometer sampling was performed at a flow rate of 5 L min${}^{-\mathrm{1}}$ with a 2.5 $\mathrm{\mu }$m cyclone inlet. A Nafion dryer was used to maintain the RH $<$ 40 %. The data logging time resolution is 1 min. The AE33 reports results in the form of equivalent BC mass (eBC), which can be used to back-calculate the ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}$. MAE values from the study by Drinovec et al. (2015) were adopted for the ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}$ back-calculation from eBC at different wavelengths as shown in Table S1 in the Supplement. A multiple scattering correction factor, $C_{\mathrm{ref}}=\mathrm{3.29}$, was used according to a recent study in this region (Qin et al., 2018). As a filter-based method, the deposition of light-absorbing particles on filters leads to the attenuation of the filter transmittance signal, which is proportional to the BC mass concentration. However, as the particle deposition layer gradually increases, the light was blocked at the upper particle layer before reaching the underlying particle layer, resulting in a well-known artifact: the loading effect. Since the lower-layer particles did not contribute to the light attenuation, the linear relationship between BC mass concentration and light attenuation signal is distorted.

The AE33 adopted the “dual spot” design to minimize the loading effect (Drinovec et al., 2015), which is an improvement on the traditional “single spot” correction (Virkkula et al., 2007). Sampling of the two spots was performed simultaneously. The correction can be implemented for each wavelength by the following two equations: $$\begin{array}{}\text{5}& {\displaystyle }\mathrm{eBC}{\mathrm{1}}_{\mathrm{raw}}={\mathrm{eBC}}_{\mathrm{compensated}}\cdot \left(\mathrm{1}-k\cdot \mathrm{ATN}\mathrm{1}\right),\text{6}& {\displaystyle }\mathrm{eBC}{\mathrm{2}}_{\mathrm{raw}}={\mathrm{eBC}}_{\mathrm{compensated}}\cdot \left(\mathrm{1}-k\cdot \mathrm{ATN}\mathrm{2}\right),\end{array}$$ where $\mathrm{eBC}{\mathrm{1}}_{\mathrm{raw}}$ and $\mathrm{eBC}{\mathrm{2}}_{\mathrm{raw}}$ are the uncorrected eBC mass determined by the two spots. ${\mathrm{eBC}}_{\mathrm{compensated}}$ is the corrected eBC concentration to be determined. $k$ is the empirical compensation parameter. ATN1 and ATN2 are the light attenuations measured at the two spots. The flows of the two spots were maintained at a ratio of 2 : 1 to achieve a differential increase in ATN in a set window of time (e.g., 1 min). Since $\mathrm{BC}{\mathrm{1}}_{\mathrm{raw}}$, $\mathrm{BC}{\mathrm{2}}_{\mathrm{raw}}$, ATN1 and ATN2 are all known variables, ${\mathrm{eBC}}_{\mathrm{compensated}}$ and $k$ can be calculated for each measurement following Eqs. (5) and (6). As shown in Fig. S2 in the Supplement, $\mathrm{eBC}{\mathrm{1}}_{\mathrm{raw}}$ and $\mathrm{eBC}{\mathrm{2}}_{\mathrm{raw}}$ exhibit a discontinuity once the filter was moved to the next position, which implies biases induced by the loading effect. After the dual-spot correction, the discontinuity is minimized substantially as shown in Fig. S2.

It is worth noting that in the single-spot correction, $k$ is a constant in each spot cycle, which means all ${\mathrm{eBC}}_{\mathrm{compensated}}$ values within the same cycle (e.g., a cycle lasts for several hours) have to share the same $k$. In contrast, time-resolved $k$ values can be determined for individual ${\mathrm{eBC}}_{\mathrm{compensated}}$ in the dual-spot correction, which is a useful indicator for the mixing state (Drinovec et al., 2017). A zero test was conducted monthly for data quality control purposes.

The absorption Ångström exponent (AAE) can be determined by the multiwavelength measurement of AE33. AAE is a useful parameter to quantify the wavelength dependency of BC light absorption, as defined by the following equation (Moosmüller et al., 2011):

7$${\displaystyle \frac{{\mathit{\sigma }}_{\mathrm{abs},{\mathit{\lambda }}_{\mathrm{1}}}}{{\mathit{\sigma }}_{\mathrm{abs},{\mathit{\lambda }}_{\mathrm{2}}}}}={\left({\displaystyle \frac{{\mathit{\lambda }}_{\mathrm{1}}}{{\mathit{\lambda }}_{\mathrm{2}}}}\right)}^{-\mathrm{AAE}},$$ where ${\mathit{\sigma }}_{\mathrm{abs},{\mathit{\lambda }}_{\mathrm{1}}}$ and ${\mathit{\sigma }}_{\mathrm{abs},{\mathit{\lambda }}_{\mathrm{2}}}$ are the light absorption coefficients at wavelengths of ${\mathit{\lambda }}_{\mathrm{1}}$ and ${\mathit{\lambda }}_{\mathrm{2}}$. The AAE of freshly emitted soot from vehicular emissions is close to 1 (Bond and Bergstrom, 2006; You et al., 2016). An increase in AAE could occur due to the coating of either BrC or non-absorbing materials. Samples that are strongly influenced by BB, which are generally rich in primary BrC, can inflate AAE larger than 2 (Reid et al., 2005; Lewis et al., 2008; McMeeking et al., 2009; Pokhrel et al., 2016). Besides BB influence, an increase in AAE up to 1.5 due to the coating of non-absorbing materials on the BC particles has also been observed in both model simulations (Lack and Langridge, 2013) and laboratory experiments (You et al., 2016).

2.3 OC and EC measurements

A field carbon analyzer (model RT-4, Sunset Laboratory Inc, Tigard, Oregon, USA) was used for OC and EC determination. Detailed sampling procedures can be found in our previous study (Wu et al., 2019), and only a brief description is given here. The sample was collected in the first 45 min of each hour at a flow rate of 8 L min${}^{-\mathrm{1}}$. The sample was analyzed in the next 15 min using thermo-optical analysis (Huntzicker et al., 1982). In the first stage, OC was vaporized by stepwise heating under helium (He), which provides an oxygen-free environment. In the second stage, carrier gas was shifted to oxygen (2 % ${\mathrm{O}}_{\mathrm{2}}$ in He) to oxidize EC on the filter. The decomposition products of these two stages were converted to carbon dioxide (${\mathrm{CO}}_{\mathrm{2}}$) by a manganese dioxide (${\mathrm{MnO}}_{\mathrm{2}}$) catalyst, then detected by a nondispersive infrared absorption (NDIR) detector. The instrument blank was analyzed on a daily basis. The filter was changed every 6 d to minimized the bias due to the accumulation of refractory materials on the filter (Jung et al., 2011).

2.4 Single-particle mass spectrometry measurements

In the wet season, a single-particle aerosol mass spectrometer (SPAMS; Hexin Analytical Instrument Co., Ltd., China) was deployed at Jinan university atmospheric super site from 11 to 18 August 2017. In the dry season, SPAMS data (15 November to 27 December 2017) from Guangdong Environmental Monitoring Center (GDEMC) were used to characterize the EC-containing particles. The GDEMC site was located south to the JNU site (4 km). The operation principle of SPAMS has been introduced previously (Li et al., 2011), and only a brief introduction is given here. The particles are introduced into the vacuum system through an 80 $\mathrm{\mu }$m critical orifice and then focused into a particle beam by the aerodynamic lens. As a result, the particles are accelerated to a size-dependent terminal velocity. The flight time of a known distance (6 cm) for individual particles is then detected by two orthogonally oriented continuous laser beams (Nd:YAG, 532 nm) for particle size determination. Sized particles are individually vaporized and ionized by a 266 nm pulsed laser (Nd:YAG, 0.6 mJ). The generated positive and negative ions are then detected by a $Z$-shaped bipolar time-of-flight mass spectrometer. SPAMS data analysis was performed by the Computational Continuation Core (COCO, V3.2) toolkit based on the MATLAB software. In total, 327 453 and 2 212 688 particles with both positive and negative mass spectra were determined by the SPAMS in the wet and dry seasons, respectively. Based on the ion marker criteria shown in Table S2, 120 351 and 595 180 EC-containing particles were identified in the wet and dry season, respectively. EC-containing particles accounting for 37 % and 27 % of the total detected particles in the wet and dry season, respectively, which is comparable with a previous SPAMS study in Guangzhou (Zhang et al., 2015). EC-containing particles were further grouped into two categories, EC-fresh and EC-aged particles. EC-aged particles were extracted from EC-containing particles using the ion markers with the relative peak area (RPA) threshold listed in Table S2, including $-\mathrm{97}$ [${\mathrm{HSO}}_{\mathrm{4}}$]${}^{-}$, $-\mathrm{62}$ [${\mathrm{HNO}}_{\mathrm{3}}$]${}^{-}$, $-\mathrm{46}$ [${\mathrm{NO}}_{\mathrm{2}}$]${}^{-}$, 43 [${\mathrm{C}}_{\mathrm{2}}{\mathrm{H}}_{\mathrm{3}}\mathrm{O}$]${}^{+}$, etc. Once EC-aged particles were defined, the remaining EC-containing particles are considered EC-fresh particles.

Despite the limitations in chemical composition quantification that is associated with the matrix effects induced by laser desorption/ionization, SPAMS is a unique technique that can provide chemical composition on a single particle level. The major advantage of single-particle analysis by SPAMS enables the characterization of coating materials exclusively on soot particles (K. Li et al., 2018), while bulk analytical techniques are incapable of distinguishing whether the non-EC materials are internally or externally mixed with EC. Relative peak area (RPA), which is defined as the peak area of each marker ion divided by the peak area of total ions, has been recognized as an indicator of the relative amount of a species on a particle (Gross et al., 2000; Jeong et al., 2011; Hatch et al., 2014; Zhou et al., 2016). Therefore, RPA is used in this study for SPAMS data analysis.

2.5 Auxiliary measurements

${\mathrm{NO}}_{\mathrm{2}}$ was determined by a chemiluminescence analyzer (model 42iTL, Thermo Scientific), while ${\mathrm{O}}_{\mathrm{3}}$ was measured by a UV photometric analyzer (model 49i, Thermo Fisher Scientific, Waltham, MA, USA). Span and zero calibrations for the gas analyzers were performed automatically on a weekly basis. Meteorological factors were measured by a multiparameter sensor (model WXT 520, Vaisala, Vantaa, Finland). The planetary boundary layer height (PBLH) measurements were conducted by a micro-pulse lidar (Sigma Space Co., USA) at the Guangzhou Meteorological Bureau (GMB; 23.00${}^{\circ }$ N, 113.32${}^{\circ }$ E; elevation: 43 m). Hourly backward trajectories for the past 72 h were calculated using NOAA's HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory, version 4) model (Draxier and Hess, 1998) for both dry and wet seasons. Backward trajectory cluster analysis was conducted using MeteoInfo (Wang, 2014, 2019). Fire count data from the Visible Infrared Imaging Radiometer Suite (VIIRS) aboard the Suomi NPP weather satellite (Csiszar et al., 2014) were downloaded from the NASA FIRMS website (https://firms.modaps.eosdis.nasa.gov/, last access: 26 February 2020) to generate the fire count map.

2.6

MAE${}_{\mathrm{p}}$ estimation by MRS method

MAE${}_{\mathrm{p}}$ is the key parameter in the $E_{\mathrm{abs}}$ calculation. In this study, MAE${}_{\mathrm{p}}$ was determined by the newly developed MRS method (Wu et al., 2018), using EC as a tracer. The atmospheric aging-induced additional light absorption (${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}$), which can be calculated by subtracting the absorption coefficient of primary aerosols, is shown in Eq. (8) (a combination of Eqs. 1 and 4): 8$${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}={\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}-{\mathrm{MAE}}_{\mathrm{p}}\times \mathrm{EC}.$$ In the MRS calculation, the correlation ($R^{\mathrm{2}}$) between measured EC and estimated hypothetical ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}$ is examined as a function of a series of hypothetical MAE${}_{\mathrm{p}}$ (MAE${}_{\mathrm{p}\mathit{\_}\mathrm{h}}$). Since ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}$ results from secondary processing and EC comes from primary emissions, a MAE${}_{\mathrm{p}\mathit{\_}\mathrm{h}}$ that leads to a minimum $R^{\mathrm{2}}$ (EC, ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}\mathit{\_}\mathrm{h}}$) can best represent the independent nature between EC and ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}$. As a result, MAE${}_{\mathrm{p}\mathit{\_}\mathrm{h}}$ at minimum $R^{\mathrm{2}}$ (EC, ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}\mathit{\_}\mathrm{h}}$) corresponds to the authentic MAE${}_{\mathrm{p}}$. The detailed method evaluation of MRS can be found in our previous paper (Wu et al., 2018). Only a brief description of the calculation steps is provided here. EC from the Sunset carbon analyzer and ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}$ from the AE33 are used as input variables. During the calculation of MAE${}_{\mathrm{p}}$ by MRS, MAE${}_{\mathrm{p}\mathit{\_}\mathrm{h}}$ is varied continuously in a reasonable range. At each MAE${}_{\mathrm{p}\mathit{\_}\mathrm{h}}$, corresponding hypothetical ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}$ (${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}\mathit{\_}\mathrm{h}}$) values are calculated for the dataset, and a correlation coefficient value ($R^{\mathrm{2}}$) of EC vs. ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}\mathit{\_}\mathrm{h}}$ (i.e., $R^{\mathrm{2}}$ (EC, ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}\mathit{\_}\mathrm{h}}$)) is obtained. By searching the MAE${}_{\mathrm{p}\mathit{\_}\mathrm{h}}$ in the desired range (e.g., from 0.1 to 50 with an interval of 0.1), a series of $R^{\mathrm{2}}$ (EC, ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}\mathit{\_}\mathrm{h}}$) values are then plotted against the MAE${}_{\mathrm{p}\mathit{\_}\mathrm{h}}$ values (Fig. 2), which only has a single minimum point.

Figure 2

$E_{\mathrm{abs}}$ determination by MRS. (a) Wet season MAE${}_{\mathrm{p}}$ determined by MRS at 520 nm. The red curve represents the correlation coefficient ($R^{\mathrm{2}}$) between hypothetical ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}$ (${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}-\mathrm{EC}\cdot {\mathrm{MAE}}_{\mathrm{p}}$) and EC mass as a function of ${\mathrm{MAE}}_{\mathrm{p}\mathit{\_}\mathrm{h}}$. The shaded area in light tan represents the frequency distribution of observed MAE. The dashed green line is the cumulative distribution of observed MAE. (b) Same as (a) but for the dry season. (c) Spectral $E_{\mathrm{abs}}$ determined by MRS in the wet season. Red circles represent the average values. The line inside the box indicates the median. Upper and lower boundaries of the box represent the 75th and the 25th percentiles; the whiskers above and below each box represent the 95th and 5th percentiles. (d) Spectral $E_{\mathrm{abs}}$ in the dry season.

[Figure omitted. See PDF]

The ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{pri}}$ is the part of light absorption from primary emitted soot particles. As a result, ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{pri}}$ is well correlated with EC mass. In contrast, ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}$ is the part of light absorption gained during the aging processes after emission. The variability in ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}$ mainly depends on the coating thickness of the soot particles. Consequently, ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}$ is independent of EC mass, and the MAE${}_{\mathrm{p}\mathit{\_}\mathrm{h}}$ corresponding to the minimum $R^{\mathrm{2}}$ (EC, ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}\mathit{\_}\mathrm{h}}$) would then represent the authentic MAE${}_{\mathrm{p}}$.

It is worth noting that MAE${}_{\mathrm{p}}$ by MRS represents the MAE${}_{\mathrm{p}}$ at the emission source, which is conceptually different from the MAE${}_{\mathrm{p}}$ by the TD method. First, the morphology and optical properties of freshly emitted BC particles (chain-like aggregates) are different from that of thermally denuded BC particles (compact aggregates). Second, most of the coatings are removed for TD denuded BC particles, but freshly emitted BC particles usually come with a thin coating of OC formed from the condensation of organic vapors due to the temperature gradient from the flame to the ambient air. As a result, the MRS-derived MAE${}_{\mathrm{p}}$ is expected to be higher than the MAE${}_{\mathrm{p}}$ by the TD method.

2.7 Secondary organic carbon (SOC) estimation by the MRS method

OC can be separated into two categories based on the formation nature. Primary organic carbon (POC) can be emitted from traffic emission (Huang et al., 2014), biomass burning (Simoneit, 2002), trash burning and cooking (Mohr et al., 2009). Secondary organic carbon (SOC) can be formed through the oxidation of volatile organic compounds (VOCs) or semivolatile POC (Hallquist et al., 2009). The EC tracer method had been used extensively for SOC estimation (Turpin and Huntzicker, 1995): $$\begin{array}{}\text{9}& {\displaystyle }\mathrm{POC}=(\mathrm{OC}/\mathrm{EC}{)}_{\mathrm{pri}}\times \mathrm{EC}+{\mathrm{OC}}_{\text{non-comb}},\text{10}& {\displaystyle }\mathrm{SOC}={\mathrm{OC}}_{\mathrm{total}}-\mathrm{POC}.\end{array}$$ Combing Eqs. (9) and (10) gives the following:

11$$\mathrm{SOC}={\mathrm{OC}}_{\mathrm{total}}-(\mathrm{OC}/\mathrm{EC}{)}_{\mathrm{pri}}\times \mathrm{EC}-{\mathrm{OC}}_{\text{non-comb}},$$ where ($\mathrm{OC}/\mathrm{EC}$)${}_{\mathrm{pri}}$ represents the overall $\mathrm{OC}/\mathrm{EC}$ ratio of aerosols from primary emission sources, while OC${}_{\text{non-comb}}$ represents primary OC from the non-combustion process. OC${}_{\text{non-comb}}$ can be determined from the intercept of OC vs. EC linear regression. In this study, weighted orthogonal distance regression (WODR) was used to account for errors in both $x$ and $y$ variables (Wu and Yu, 2018). By grouping the data into percentile subsets using $\mathrm{OC}/\mathrm{EC}$ ratio from the lowest to the highest (1 %–100 %, with an interval of 1 %), a series of intercepts were obtained as a function of $\mathrm{OC}/\mathrm{EC}$ percentile (Fig. S3). The intercept term in the OC vs. EC WODR is very small ($-\mathrm{0.88}$ to $-\mathrm{0.05}$) throughout the percentile range (1 %–100 %). Since this term is small, OC${}_{\text{non-comb}}$ was set to zero for SOC estimation in this study.

($\mathrm{OC}/\mathrm{EC}$)${}_{\mathrm{pri}}$ is the key parameter for SOC calculation in the EC tracer method. In the MRS method, the correlation ($R^{\mathrm{2}}$) between measured EC and estimated SOC (from Eq. 10) was examined as a function of a series of hypothetical ($\mathrm{OC}/\mathrm{EC}$)${}_{\mathrm{pri}}$ (($\mathrm{OC}/\mathrm{EC}$)${}_{\mathrm{pri}\mathit{\_}\mathrm{h}}$). The $\mathrm{OC}/\mathrm{EC}$ ratio at the minimum $R^{\mathrm{2}}$ (EC vs. SOC) corresponds to the authentic primary $\mathrm{OC}/\mathrm{EC}$ ratio (Millet et al., 2005). The detailed calculation steps can be found in our previous paper (Wu and Yu, 2016). Only a brief description is given here. In the MRS calculation, ($\mathrm{OC}/\mathrm{EC}$)${}_{\mathrm{pri}\mathit{\_}\mathrm{h}}$ was varied continuously in a reasonable range (e.g., from 0.1 to 10 with an interval of 0.1). Hypothetical SOC (SOC${}_{\mathrm{h}}$) values were calculated at individual ($\mathrm{OC}/\mathrm{EC}$)${}_{\mathrm{pri}\mathit{\_}\mathrm{h}}$ for the whole dataset. A series of $R^{\mathrm{2}}$ values of EC vs. SOC${}_{\mathrm{h}}$ (i.e., $R^{\mathrm{2}}$ (EC, SOC${}_{\mathrm{h}}$)) were generated and then plotted against the ($\mathrm{OC}/\mathrm{EC}$)${}_{\mathrm{pri}\mathit{\_}\mathrm{h}}$ values. Based on the assumption that variations in EC and SOC are independent, the ($\mathrm{OC}/\mathrm{EC}$)${}_{\mathrm{pri}\mathit{\_}\mathrm{h}}$ corresponding to the minimum $R^{\mathrm{2}}$ (EC, SOC${}_{\mathrm{h}}$) would then indicate the authentic ($\mathrm{OC}/\mathrm{EC}$)${}_{\mathrm{pri}}$ ratio.

In our previous work, numerical studies were performed, and the results showed that the minimum $R$ squared method (MRS) is more robust in SOC estimation than the minimum $\mathrm{OC}/\mathrm{EC}$ and percentile $\mathrm{OC}/\mathrm{EC}$ methods (Wu and Yu, 2016). As a result, the MRS method had been gradually adopted for SOC estimation in recent studies (Xu et al., 2018b; Bian et al., 2018; Ji et al., 2018, 2019; Ying et al., 2018; Wu et al., 2019).

An Igor Pro-based computer program (WaveMetrics, Inc. Lake Oswego, OR, USA) (Wu and Yu, 2016) was used to implement the MRS calculation. Another two Igor Pro-based computer programs, Histbox (Wu et al., 2018) and Scatter Plot (Wu and Yu, 2018), were used for generating the box plots and scatter plots presented in this study. These computer programs (with operation manuals) can be downloaded freely from https://sites.google.com/site/wuchengust (last access: 26 February 2020).

3 Results and discussions

3.1 Seasonality of carbonaceous aerosol concentrations and optical properties

The time series of EC, OC, optical properties and supporting measurements during the wet and dry seasons are shown in Fig. S4. The hourly EC concentrations ranged from 0.43 to 7.40 and 0.54 to 12.04 $\mathrm{\mu }$gC m${}^{-\mathrm{3}}$ in the wet and dry seasons, respectively. As for OC, the hourly average ranged from 0.32 to 13.84 and 0.51 to 25.31 $\mathrm{\mu }$gC m${}^{-\mathrm{3}}$ in the wet and dry seasons, respectively. The hourly $\mathrm{OC}/\mathrm{EC}$ ratios ranged from 0.25 to 6.92 and 0.33 to 8.69 in the wet and dry seasons, respectively. In the wet season, the wind direction is southeasterly dominated, bringing the relatively clean background air masses from the vast ocean. In the dry season, the northeasterly wind prevails, which promotes the long-range transport of air pollutants from eastern and central China.

As shown in Fig. 1, EC, OC and $\mathrm{OC}/\mathrm{EC}$ AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ all exhibit clear seasonality. Average EC concentrations (with 1 standard deviation, hereafter) were $\mathrm{1.94}\pm \mathrm{0.93}$ and $\mathrm{2.81}\pm \mathrm{2.01}$ $\mathrm{\mu }$gC m${}^{-\mathrm{3}}$ in the wet and dry seasons, respectively. The EC level was comparable to the measurements made in 2012 at a Guangzhou suburban site ($\mathrm{1.67}\pm \mathrm{1.35}$ $\mathrm{\mu }$gC m${}^{-\mathrm{3}}$ in the wet season, $\mathrm{3.47}\pm \mathrm{2.75}$ $\mathrm{\mu }$gC m${}^{-\mathrm{3}}$ in the dry season) (Wu et al., 2019).

The average concentrations of OC doubled in the dry season ($\mathrm{7.02}\pm \mathrm{5.19}$ $\mathrm{\mu }$gC m${}^{-\mathrm{3}}$) compared to those in the wet season ($\mathrm{3.38}\pm \mathrm{1.93}$ $\mathrm{\mu }$gC m${}^{-\mathrm{3}}$), leading to elevated $\mathrm{OC}/\mathrm{EC}$ ratio in the dry season ($\mathrm{2.56}\pm \mathrm{0.94}$) in contrast to wet season ($\mathrm{1.78}\pm \mathrm{0.83}$). The hourly AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ ranged from 1.14 to 1.67 and 1.07 to 1.76 in the wet and dry seasons, respectively. As shown in Fig. S5a, AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ observed in the dry season ($\mathrm{1.46}\pm \mathrm{0.12}$) was significantly ($P<\mathrm{0.001}$) higher than that in the wet season ($\mathrm{1.37}\pm \mathrm{0.10}$).

Measured MAE${}_{\mathrm{520}}$ in the dry season ($\mathrm{18.47}\pm \mathrm{5.49}$ m${}^{\mathrm{2}}$ g${}^{-\mathrm{1}}$) is significantly ($P<\mathrm{0.001}$) higher than that in the wet season ($\mathrm{10.73}\pm \mathrm{4.96}$ m${}^{\mathrm{2}}$ g${}^{-\mathrm{1}}$), as shown in Fig. S5b. The elevated MAE during the dry season was likely a result of BB influences, which will be discussed in detail in Sect. 3.3.

The MAE${}_{\mathrm{p}\mathrm{520}}$ values determined by MRS were 8.6 and 16.8 m${}^{\mathrm{2}}$ g${}^{-\mathrm{1}}$, for wet and dry seasons, respectively (Fig. 2a and b). Similar to MAE${}_{\mathrm{520}}$, the increase in MAE${}_{\mathrm{p}\mathrm{520}}$ in the dry season was also likely a result of BB influence, which could lead to larger BC cores (Ditas et al., 2018) and with thicker primary coatings (Schwarz et al., 2008a; Kondo et al., 2011; Lack et al., 2012a; Liu et al., 2014). More details of the BB influences will be discussed in Sect. 3.3. Consequently, light absorption enhancement was found to be more pronounced in the wet season ($E_{\mathrm{abs}\mathrm{520}}=\mathrm{1.51}\pm \mathrm{0.50}$, Table 2) than in dry season ($\mathrm{1.29}\pm \mathrm{0.28}$), because $E_{\mathrm{abs}}$ depends on the ratio of MAE to MAE${}_{\mathrm{p}}$, not their absolute values.

Table 2

Comparisons of $E_{\mathrm{abs}}$ measurements in various field studies.

Approach	Location	Sampling period	Sampling duration	Timeresolution	$\mathit{\lambda }$ (nm)	$E_{\mathrm{abs}}$	Reference
MAE $+$	Guangzhou, China	Jul–Sep 2017	6 months	1 h	520	$\mathrm{1.51}\pm \mathrm{0.50}$	This study
MRS	(urban)	Nov 2017–Jan 2018				$\mathrm{1.29}\pm \mathrm{0.28}$
	Guangzhou, China (suburban)	Feb 2012–Jan 2013	1 year	1 h	550	$\mathrm{1.50}\pm \mathrm{0.48}$	Wu et al. (2018)
MAE $+$ empirical	Nanjing, China (suburban)	Nov 2012	15 d	12 h	532	1.6	F. Cui et al. (2016)
	Beijing, China (suburban)	Nov 2014–Jan 2015	2 months	1 h	470	2.6–4.0	Xu et al. (2016)
MAE $+$ SP	Beijing, China (urban)	Nov 2014	14 d	1 h	880	1.66–1.91	Zhang et al. (2018b)
	Manchester, UK (urban)	Oct–Nov 2014	13 d	5 min	532	1.0–1.3	D. Liu et al. (2017)
	Paris, France (urban)	Mar 2014–Mar 2017	3 years	24 h	880	$\mathrm{1.53}\pm \mathrm{0.39}$	Zhang et al. (2018a)
	Kanpur, India (urban)	Jan–Feb 2015	2 months	1 h	781	1.8	Thamban et al. (2017)
	Xi'an, China (urban)	Dec 2012–Jan 2013	1 month	20 min	870	1.8	Wang et al. (2014)
TD	Shouxian, China (Rural)	Jun–Jul 2016	8 d	10 min	532	$\mathrm{2.3}\pm \mathrm{0.9}$	Xu et al. (2018)
	Beijing, China (urban)	Jun 2017	10 d	20 min	630	$\mathrm{1.59}\pm \mathrm{0.26}$	Xie et al. (2019)
	Sacramento, USA (urban)	Jun–Jul 2010	13 d	20 min	532	$\mathrm{1.06}\pm \mathrm{0.01}$	Cappa et al. (2012)
	Fresno, USA	Dec 2014–Jan 2015	19 d	10 min	532	$\mathrm{1.22}\pm \mathrm{0.15}$	Cappa et al. (2019)
	(urban)
	Fontana, USA	July 2015	23 d		532	$\mathrm{1.07}\pm \mathrm{0.22}$
	(urban)
	Suzu, Japan	Apr–May 2013	26 d	20 min	532	1.06	Ueda et al. (2016)
	Nanjing, China	Aug–Sep 2014	16 d	2 h	405	$\mathrm{1.41}\pm \mathrm{0.39}$	Ma et al. (2020)
	(suburban)				532	$\mathrm{1.42}\pm \mathrm{0.40}$
					781	$\mathrm{1.35}\pm \mathrm{0.38}$
AFD	Jinan, China (urban)	Feb 2014	13 d	8 h	678	$\mathrm{2.07}\pm \mathrm{0.72}$	Chen et al. (2017)
	Yuncheng, China (rural)	Jun–Jul 2014	18 d	12 h	678	$\mathrm{2.25}\pm \mathrm{0.55}$	X. Cui et al. (2016)
	Jinan, China	Jul–Jul 2016	30 d	24 h	678	$\mathrm{1.9}\pm \mathrm{0.7}$	Bai et al. (2018)
	(urban)
	Mt. Tai, China	Jul–Aug 2014	31 d	24 h	678	$\mathrm{2.0}\pm \mathrm{0.3}$
	(rural)

In summary, as evidenced by AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ and MAE results, carbonaceous aerosols exhibit strong seasonality in urban Guangzhou. This seasonality was associated with two seasonal factors, including the contrasted direction of the prevailing wind and diverse primary BC optical properties induced by seasonal BB influence.

3.2

Comparison of $E_{\mathrm{abs}}$ with previous studies

Field measurements of $E_{\mathrm{abs}}$ values around the world are summarized in Table 2. Studies using the TD approach can achieve a sub-hour time resolution, but studies using the TD approach had limited temporal coverage (normally less than a month). The AFD approach can potentially provide long-term $E_{\mathrm{abs}}$ results as long as filter samples are available. However, the measurement duration of existing AFD studies was less than 1 month as shown in Table 2. The limited temporal coverage of existing AFD studies was likely due to the intense labor involved in filter treatment. In addition, the time resolution of existing AFD studies (8–24 h) was not sufficient to fully resolve $E_{\mathrm{abs}}$ diurnal pattern. As a result, diurnal variations of $E_{\mathrm{abs}}$ values for different seasons were not covered in previous studies. In comparison, the MRS approach is a good alternative to explore the $E_{\mathrm{abs}}$ variations at both seasonal and diurnal scales. As shown in Table 2, low $E_{\mathrm{abs}}$ values were found in California (1.06 at 532 nm) (Cappa et al., 2012) and Japan (1.06 at 532 nm) (Ueda et al., 2016). D. Liu et al. (2017) observed moderate $E_{\mathrm{abs}}$ in the UK (1.0–1.3 at 532 nm) and suggested that the small $E_{\mathrm{abs}}$ observed by Cappa et al. (2012) was a result of mixing state diversity. A recent study in California (Cappa et al., 2019) found moderate $E_{\mathrm{abs}}$ at Fresno (1.22 at 532 nm) but low $E_{\mathrm{abs}}$ at Fontana (1.07 at 532 nm), which was partially associated with an unequal distribution of coating between different BC-containing particle types (Lee et al., 2019). In general, higher $E_{\mathrm{abs}}$ values have been observed in more polluted urban areas, such as France (Paris, 1.53 at 880 nm) (Zhang et al., 2018a) and India (Kanpur, 1.8 at 781 nm) (Thamban et al., 2017). High $E_{\mathrm{abs}}$ values have been reported in various locations in China. The $E_{\mathrm{abs}}$ value in the wet season in our study (1.51) is higher than that in Nanjing (1.42) (Ma et al., 2020) but lower than those in central China (Shouxian, 2.3) (Xu et al., 2018), eastern China (Jinan, 1.9) (Bai et al., 2018) and northern China (Yuncheng, 2.25) (X. Cui et al., 2016). The $E_{\mathrm{abs}}$ value in the dry season in our study (1.29) is lower than those in other locations in China, such as Beijing (1.66–4.0) (Xu et al., 2016; Zhang et al., 2018b), Nanjing (1.6) (F. Cui et al., 2016), Xi'an (1.8) (Wang et al., 2014) and Jinan (2.07) (Chen et al., 2017). Since the colocated comparison of the three $E_{\mathrm{abs}}$ methods does not exist, a direct comparison between the three methods remains difficult. Nevertheless, a few studies, which conducted in the same city but during different periods, yielded comparable $E_{\mathrm{abs}}$ values. For example, $E_{\mathrm{abs}}$ in Nanjing by the MAE method (1.6) (F. Cui et al., 2016) was higher than that by the TD method (1.42) (Ma et al., 2020). This difference in $E_{\mathrm{abs}}$ might not only be due to the different $E_{\mathrm{abs}}$ determination methods but also could be a result of seasonal variations in $E_{\mathrm{abs}}$.

3.3 Influence of biomass burning on the BC optical properties during the dry season

Evidence from particle chemical compositions showed that the BB influence was more intense in the dry season. Levoglucosan had been widely accepted as a tracer for BB in PM${}_{\mathrm{2.5}}$ (Engling et al., 2006; Bhattarai et al., 2019). As shown in Fig. S6, levoglucosan concentrations in Guangzhou were elevated by 1 order of magnitude during the dry season (159.33 ng m${}^{-\mathrm{3}}$) compared to those in the wet season (35.93 ng m${}^{-\mathrm{3}}$). Besides levoglucosan, the primary $\mathrm{OC}/\mathrm{EC}$ ratio can also be used as an indicator of BB influence since the BB-influenced samples have a higher $\mathrm{OC}/\mathrm{EC}$ ratio than those from traffic emissions (Schmidl et al., 2008; Pokhrel et al., 2016). In this study, ($\mathrm{OC}/\mathrm{EC}$)${}_{\mathrm{pri}}$ determined by MRS in the dry season (2.31) was higher than that in the wet season (1.49), as shown in Fig. S7. In addition, the northeasterly wind prevailed during the dry season, which favors long-range transport of aerosols from BB from central and eastern China to the PRD region. Remote sensing results also confirmed the more intense BB in the dry season, as shown by the gridded fire count map (Fig. S8) determined by VIIRS.

As a result, the optical properties of BC were largely affected by the intense BB influences during the dry season. First, as shown in Fig. S5b, a significantly ($P<\mathrm{0.001}$) higher MAE${}_{\mathrm{520}}$ was observed in the dry season ($\mathrm{18.47}\pm \mathrm{5.49}$) compared to that in the wet season ($\mathrm{11.28}\pm \mathrm{9.88}$). A previous field study at a suburban site in Guangzhou also reported the influence of BB on MAE, which observed a positive correlation between ${\mathrm{K}}^{+}$ and MAE (Wu et al., 2018). High MAE from BB had been reported in BB emission studies as well (Roden et al., 2006; Schmidl et al., 2008; Levin et al., 2010; Wang et al., 2018a). Single-particle soot photometer (SP2) studies have shown that BB-influenced BC particles are more likely to have larger BC cores (Ditas et al., 2018) and have thicker initial coatings than those from vehicular emissions (Schwarz et al., 2008a; Kondo et al., 2011; Lack et al., 2012a; Liu et al., 2014). This is in good agreement with the MAE${}_{\mathrm{p}\mathrm{520}}$ obtained in the present study, which was almost doubled in the dry season (16.8 m${}^{\mathrm{2}}$ g${}^{-\mathrm{1}}$) compared to that in the wet season (8.6 m${}^{\mathrm{2}}$ g${}^{-\mathrm{1}}$).

In the dry season the $E_{\mathrm{abs}}$ showed little wavelength dependence (Fig. 2d) despite the influence of BB. In this sense, the BB influence did substantially alter the optical properties of primary BC in the dry season, but the contribution of secondary BrC to $E_{\mathrm{abs}}$ was likely limited. The weak wavelength dependence of $E_{\mathrm{abs}}$ has also been observed in a previous study at a suburban site in Guangzhou (Wu et al., 2018). A previous study in Guangzhou also found that the seasonal difference in BrC light absorption contribution at 405 nm between dry season (15 %–19 %) and wet seasons (12 %–15 %) was small (S. Li et al., 2018). In addition, the small seasonal difference of AAE between wet ($\mathrm{1.37}\pm \mathrm{0.10}$) and dry ($\mathrm{1.46}\pm \mathrm{0.12}$) seasons observed in this study also implies that the secondary BrC contribution was not the dominating driver for AAE deviation from 1, which was the typical AAE for fresh soot without atmospheric aging. The results found in the PRD were in contrast to a study in Paris, which found systematically higher $E_{\mathrm{abs}\mathrm{370}}$ than $E_{\mathrm{abs}\mathrm{880}}$ at wintertime due to the influence of BB (Zhang et al., 2018a). This discrepancy implies the complex linkage between BB and BrC optical properties.

The complex relationship between AAE and BrC can be affected by a variety of factors. First, the optical properties of primary BrC from BB exhibit a large diversity in previous studies (Martinsson et al., 2015; J. Tian et al., 2019), which can be affected by fuel type and combustion conditions (Reid et al., 2005; Roden et al., 2006). Second, atmospheric aging can lead to AAE elevation through the formation of secondary BrC from a variety of pathways (Moise et al., 2015; Laskin et al., 2015), including nitration of aromatic compounds (Jacobson, 1999), reaction of ammonia (Bones et al., 2010), bond-forming reactions between SOA constituents (Shapiro et al., 2009), reactions of BB products (Gilardoni et al., 2016; Kumar et al., 2018), photo-enhancement (Hems and Abbatt, 2018; J. Liu et al., 2016; Ye et al., 2019), and aqueous-phase reactions (Lin et al., 2015; Tang et al., 2016; Xu et al., 2018a). On the other hand, AAE decrease could also occur during atmospheric aging (Romonosky et al., 2019), either induced by photobleaching of BrC (Adler et al., 2011; Zhong and Jang, 2011, 2014; Lee et al., 2014; Canonaco et al., 2015; Lin et al., 2016; Sumlin et al., 2017; Bhattarai et al., 2018; Fortenberry et al., 2018; Hems and Abbatt, 2018; Browne et al., 2019; Dasari et al., 2019; C. Li et al., 2019; Wong et al., 2019; Cai et al., 2019; Liakakou et al., 2019; Ray et al., 2019) or aqueous-phase BrC degradation in the absence of light (Santos and Duarte, 2015; G. T. A. D Santos et al., 2016; P. S. M. Santos et al., 2016; Fan et al., 2019). The relative contribution of secondary BrC formation and BrC degradation to the total BrC light absorption budget is still poorly understood. BrC degradation could be one of the reasons for the small seasonal AAE difference observed in the PRD region. More studies are needed by incorporating both time-resolved optical measurements and time-resolved detailed chemical speciation measurements to better understand the balance of BrC formation and degradation.

3.4 Diurnal dynamics of carbonaceous aerosols concentrations and optical properties

The diurnal variations in EC, OC, $\mathrm{OC}/\mathrm{EC}$, SOC, AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ and $E_{\mathrm{abs}\mathrm{520}}$ in the wet and dry seasons are shown in Fig. 3. Two peaks can be observed for EC (Fig. 3a): one in the early morning (07:00 LT, UTC$+$8, all times in this paper are local time) and the other in the evening (19:00), which reflects local traffic emissions in two rush hours. The lowest EC was found in the afternoon (14:00), likely associated with two factors considering the nature of EC source exclusive from primary emissions. The first factor is the planetary boundary layer (PBL) height. As shown in Fig. S9, the diurnal maximum PBL height was at 14:00 and 15:00 for wet and dry seasons, respectively. The fully developed PBL would help dilute the concentrations of primary pollutants (Deng et al., 2016; Liu et al., 2019; Williams et al., 2019). The second factor is the diurnal variations of traffic volume. Previous studies (Yao et al., 2013; Xie et al., 2003) showed that traffic volume during 12:00–15:00 is lower than that in the morning and evening rush hours. To explore the effect of traffic volume, the weekday/weekend effect was investigated in Fig. S10. The evening EC peak reduced substantially during the weekend, implying that traffic volume has a strong influence on shaping the diurnal pattern of EC. The combination of these two factors leads to reduced EC concentrations in the afternoon. The diurnal pattern of EC is similar between wet and dry seasons, but the magnitude was greatly elevated in the dry season.

Figure 3

Diurnal pattern carbonaceous aerosols in the wet and dry seasons. The solid lines represent hourly averages and the shaded areas represent 25th and 75th percentiles. (a) EC. (b) OC. (c) $\mathrm{OC}/\mathrm{EC}$ ratio. (d) SOC. (e) AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$. (f) $E_{\mathrm{abs}\mathrm{520}}$.

[Figure omitted. See PDF]

OC exhibits a bimodal distribution (Fig. 3b), peaking at 13:00 and 19:00, respectively. OC can be both primary and secondary, making its diurnal pattern different from that of EC. $\mathrm{OC}/\mathrm{EC}$ also has two peaks as shown in Fig. 3c. The first peak appeared at 13:00 and the second peak showed up at 17:00. It is worth noting that in the wet season the afternoon $\mathrm{OC}/\mathrm{EC}$ peak was higher than that in the evening $\mathrm{OC}/\mathrm{EC}$ peak, while in the dry season the reverse is true. The difference in diurnal $\mathrm{OC}/\mathrm{EC}$ pattern between weekday and weekend is negligible (Fig. S10), suggesting that the proportion of different vehicle types (e.g., diesel vs. gasoline) is relatively constant between weekday and weekend.

As shown in Fig. 3d, two SOC peaks are observed in the wet season, with the first SOC peak at 13:00 and the second SOC peak at 19:00. In the dry season the afternoon SOC peak was merged into the broadened evening peak. Despite the higher SOC concentrations observed in the dry season, SOC formation was more active during the wet season as evidenced by the diurnal $\mathrm{SOC}/\mathrm{OC}$ ratios (Fig. 4a). The diurnal $\mathrm{SOC}/\mathrm{OC}$ ratio in the wet season was always higher than that in the dry season. It is worth noting that in the wet season, despite the fact that the SOC evening peak was comparable to the afternoon peak as shown in Fig. 3d, the $\mathrm{SOC}/\mathrm{OC}$ evening peak was smaller than the afternoon peak (Fig. 4a). This observation implies that the SOC evening peak in the wet season is a result of the combination of pollutant accumulation (e.g., PBL decrease after sunset) and SOC formation rather than the formation process alone. The small evening peak of $\mathrm{SOC}/\mathrm{OC}$ in summer (19:00–21:00) would also be likely a result of condensation of semivolatile organic compounds (Warren et al., 2009; Pathak et al., 2008; Liang et al., 1997) due to the temperature decrease after sunset.

Figure 4

The effect of the secondary process on $E_{\mathrm{abs}\mathrm{520}}$. (a) Diurnal pattern of $E_{\mathrm{abs}\mathrm{520}}$ and ${\mathrm{O}}_x$ in the wet season. The lines (${\mathrm{O}}_x$ in blue, $\mathrm{SOC}/\mathrm{OC}$ in green and $E_{\mathrm{abs}\mathrm{520}}$ in dotted black lines) represent hourly averages and the shaded areas represent 25th and 75th percentiles. (b) Same as (a) but in the dry season.

[Figure omitted. See PDF]

$\mathrm{SOC}/\mathrm{OC}$ dependence on RH was investigated (Fig. S11) to explore the effect of aqueous-phase secondary organic aerosol formation. During the wet season, $\mathrm{SOC}/\mathrm{OC}$ decreases as RH increases and the results were the same for both daytime and nighttime (Fig. S11a and b). During nighttime when no solar radiation was supplied, higher RH leads to a lower $\mathrm{SOC}/\mathrm{OC}$ (Fig. S11b). This piece of evidence suggests that aqueous-phase reactions were unlikely the dominating pathway for SOC formation during the wet season. In the dry season, $\mathrm{SOC}/\mathrm{OC}$ does not show a clear dependence on RH, suggesting that SOC formation is not sensitive to RH in the dry season.

The diurnal trend of AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ was similar between wet and dry seasons, which is higher in the evening and lower during midday, but the magnitude of AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ slightly increased during the dry season. The $E_{\mathrm{abs}\mathrm{520}}$ exhibits different diurnal patterns between the wet and dry seasons. As shown in Fig. 5, elevated $E_{\mathrm{abs}\mathrm{520}}$ was found during nighttime in the wet season, but in dry season inflated $E_{\mathrm{abs}\mathrm{520}}$ was observed in the afternoon. In addition, the degree of light absorption enhancement was more pronounced in the wet season. The influencing factors of dynamics are discussed in the following sections.

Figure 5

Diurnal patterns of $E_{\mathrm{abs}\mathrm{520}}$, AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ and $k_{\mathrm{3}}$. The lines ($k_{\mathrm{3}}$ in purple, AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ in red and $E_{\mathrm{abs}\mathrm{520}}$ in dotted black lines) represent hourly averages and the shaded areas represent 25th and 75th percentiles. (a) Wet season. (b) Dry season.

[Figure omitted. See PDF]

3.5

The diurnal correlations between AAE, $k$ and $E_{\mathrm{abs}}$

The loading effect correction factor used in AE33, $k$, has been found to be a useful indicator for the light absorption enhancement of BC (Drinovec et al., 2017). As shown in Fig. 5a, in the wet season a good anticorrelation was found between $k_{\mathrm{3}}$ ($k$ value for 520 nm) and $E_{\mathrm{abs}\mathrm{520}}$ with an $R^{\mathrm{2}}$ of 0.74. In the dry season, such an anticorrelation was substantially weakened ($R^{\mathrm{2}}=\mathrm{0.20}$) as shown in Fig. 5b, likely due to the influence of BB. These results agree well with the findings reported by Drinovec et al. (2017) that $k$ can be used as a BC particle coating indicator without the influence of BB. As shown in Fig. 5a, a good correlation was found for AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ with $E_{\mathrm{abs}\mathrm{520}}$ ($R^{\mathrm{2}}=\mathrm{0.71}$). Considering the weak BB influence in the wet season as discussed in Sect. 3.3, the atmospheric aging-induced coating on BC particles was more likely the dominating driver of AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ dynamics during the wet season in the PRD region. The presence of a coating of BC could also explain that despite the fact that the BB influence is small in the wet season, the observed average AAE ($\mathrm{1.37}\pm \mathrm{0.10}$) was significantly higher than the AAE of fresh BC ($\sim \mathrm{1}$). This result is also consistent with previous studies that found that a non-light-absorbing coating can lead to elevated AAE up to 1.5 (Lack and Cappa, 2010; Lack and Langridge, 2013). In the dry season, the variability in AAE was governed by both coating thickness and BB influence, thus leading to a degraded $R^{\mathrm{2}}$ (0.22) between AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$ and $E_{\mathrm{abs}\mathrm{520}}$ as shown in Fig. 5b. A recent study showed that the diurnal pattern of BrC was moderately correlated with a BB tracer, ${\mathrm{K}}^{+}$, in the PRD region during the dry season (Z. Li et al., 2019), implying that BB did have a considerable influence on AAE viability during the dry season.

The spectral fingerprints of $k$ were shown in Fig. S12. Observations in Europe showed that the presence of BrC could lead to increased $k$ at longer wavelengths (Drinovec et al., 2017). Our observations showed that the seasonal difference in spectral fingerprints of $k$ between wet and dry seasons is small. Considering the limited increase in AAE in the dry season and the similarity of seasonal spectral fingerprints of $k$, these results suggest that, in the PRD region, despite the fact that the BB influence in the dry season effectively altered the optical properties of BC aerosols, there was a likely limited secondary BrC contribution to $E_{\mathrm{abs}}$ during the dry season, which is in agreement with discussions in Sect. 3.3.

3.6

The influence of secondary processing on E${}_{\mathrm{abs}}$ diurnal dynamics

Photochemical reactions play an important role in the aging process of black carbon, leading to the modification of BC morphology and optical properties as evidenced by laboratory studies (Saathoff et al., 2003; Schnaiter et al., 2005; Martinsson et al., 2015; Pei et al., 2018) and quasi-atmospheric chamber studies (Peng et al., 2017, 2016). Field studies at various locations have also showed that photochemical processing can promote the light absorption enhancement of BC, including in Beijing (Liu et al., 2019a), the Yangtze River Delta (Xu et al., 2018), Xi'an (Wang et al., 2017), Los Angeles (Krasowsky et al., 2016) and Toronto (Knox et al., 2009). The concentration of odd oxygen (${\mathrm{O}}_x={\mathrm{O}}_{\mathrm{3}}+{\mathrm{NO}}_{\mathrm{2}}$) proposed by Liu (1977) and (Levy II et al., 1985) has been widely used as the indicator of photochemical aging. In this study, the diurnal correlations between ${\mathrm{O}}_x$ and $E_{\mathrm{abs}\mathrm{520}}$ were investigated to explore the effect of photochemical processing. As shown in Fig. 4a, in the wet season ${\mathrm{O}}_x$ and $E_{\mathrm{abs}\mathrm{520}}$ peaked at 15:00 and 00:00, respectively. The ${\mathrm{O}}_x$ experienced a continuous decline from 15:00 until sunrise of the next day, but the growth of $E_{\mathrm{abs}}$ extended to midnight. The nighttime $E_{\mathrm{abs}\mathrm{520}}$ peak suggests that the increase in coating can be achieved without the presence of solar radiation. These differences in the diurnal patterns led to a low correlation between ${\mathrm{O}}_x$ and $E_{\mathrm{abs}}$ ($R^{\mathrm{2}}=\mathrm{0.01}$). This result implies that in the wet season the diurnal variability in $E_{\mathrm{abs}\mathrm{520}}$ was unlikely dominated by photochemical reactions, despite the fact that ${\mathrm{O}}_x$ was more pronounced in the wet season. As for the dry season (Fig. 4b), both ${\mathrm{O}}_x$ and $E_{\mathrm{abs}\mathrm{520}}$ peaked at 17:00, leading to a good correlation with a $R^{\mathrm{2}}$ of 0.69, suggesting that photochemical reactions could be one of the main drivers for $E_{\mathrm{abs}}$ diurnal variations. This result strongly indicates that BC light absorption can be markedly amplified through photochemical reactions. Our dry-season results are consistent with a previous study in northern China (Wang et al., 2017), which also showed the dependence of light absorption enhancement on ${\mathrm{O}}_x$ during the wintertime.

In the meantime, the formation of SOC also contributes to light absorption enhancement of BC, which had been observed in both field studies (Moffet and Prather, 2009; Wang et al., 2017a; Zhang et al., 2018a) and laboratory studies (Schnaiter et al., 2005; Lambe et al., 2013; Saliba et al., 2016). In this study, the effect of SOC formation on $E_{\mathrm{abs}}$ was investigated using $\mathrm{SOC}/\mathrm{OC}$ ratio as the indicator rather than using SOC alone. The advantage of using $\mathrm{SOC}/\mathrm{OC}$ is that the SOC variations induced by the non-secondary-formation process (e.g., PBLH shallowing) can be minimized, thereby focusing the analysis on the effect of secondary formation processes. A good diurnal correlation between $\mathrm{SOC}/\mathrm{OC}$ and $E_{\mathrm{abs}\mathrm{520}}$ was observed in the dry season ($R^{\mathrm{2}}=\mathrm{0.53}$), but no correlation was found in the wet season ($R^{\mathrm{2}}=\mathrm{0.01}$). The $E_{\mathrm{abs}}$ dependence on $\mathrm{SOC}/\mathrm{OC}$ was examined in Fig. 6. $E_{\mathrm{abs}}$ dependence on $\mathrm{SOC}/\mathrm{OC}$ was found in both wet and dry seasons, but a clearer dependence was observed in the dry season. It should be noted that a good $E_{\mathrm{abs}}$ dependence on $\mathrm{SOC}/\mathrm{OC}$ observed in Fig. 6 does not necessarily lead to a good diurnal correlation between $E_{\mathrm{abs}}$ and $\mathrm{SOC}/\mathrm{OC}$ (e.g., Fig. 4a). In other words, the dependency of $E_{\mathrm{abs}}$ on $\mathrm{SOC}/\mathrm{OC}$ might not necessarily be reflected in the form of correlation on a diurnal scale. Thus, the poor diurnal correlation between $\mathrm{SOC}/\mathrm{OC}$ and $E_{\mathrm{abs}\mathrm{520}}$ observed in the wet season (Fig. 4a) cannot rule out the contribution of SOC to $E_{\mathrm{abs}\mathrm{520}}$. A study in Paris (Zhang et al., 2018a) found that more oxidized oxygenated organic aerosols (MO-OOA) and less oxidized OOA (LO-OOA), which are surrogates of SOA, were the dominating contributors for $E_{\mathrm{abs}}$, especially in the summertime. In the present study, due to the lack of quantitative chemical speciation data, quantification of contributions from different chemical species to $E_{\mathrm{abs}}$ is not possible. A recent study in Guangzhou (Wu et al., 2019) found that traffic-derived SOC could be a significant source of SOC in the urban area, which can account for half of the total SOC. In that sense, traffic emissions are expected to have a considerable contribution to BC light absorption enhancement in both wet and dry seasons.

Figure 6

$E_{\mathrm{abs}\mathrm{520}}$ dependency on $\mathrm{SOC}/\mathrm{OC}$ ratio and temperature during wet and dry seasons. Red circles represent the average values. The line inside the box indicates the median. Upper and lower boundaries of the box represent the 75th and the 25th percentiles; the whiskers above and below each box represent the 95th and 5th percentiles. The pink lines represent normalized frequency of data points in each bin.

[Figure omitted. See PDF]

The temperature effect on $E_{\mathrm{abs}}$ was examined in Fig. 6c and d for wet and dry seasons, respectively. In the wet season, a positive response of $E_{\mathrm{abs}}$ on temperature was observed for the temperature range of 24–30 ${}^{\circ }$C, implying a favorable condition for coating formation on BC particles. However, further temperature increment beyond 30 ${}^{\circ }$C led to a decline of $E_{\mathrm{abs}}$, which might be associated with the evaporation of the coating materials on BC. In the dry season, $E_{\mathrm{abs}}$ was not sensitive to temperature for the range of 12–24 ${}^{\circ }$C. The volatility effect of coating materials on $E_{\mathrm{abs}}$ will be discussed in more detail in the next section.

3.7

The influence of semivolatile compounds on $E_{\mathrm{abs}}$ diurnal variations

The SPAMS data from both wet (11–18 August 2017) and dry seasons (15 November to 27 December 2017) were analyzed to explore the mixing state of EC-containing particles from a single-particle perspective. The average EC-fresh and EC-aged mass spectra are shown in Fig. S13 for both wet and dry seasons. The domination of EC-aged particles in EC-containing particle number fraction suggests that most of the EC particles are internally mixed with other species (Table S3). This result agrees with previous studies in this region (Zhang et al., 2013, 2014).

Figure 7

Diurnal variations in coating (including organics, sulfate and nitrate) RPA to EC RPA ratios of EC-aged particles measured by SPAMS in the wet and dry seasons. The solid lines represent hourly averages and the shaded areas represent 25th and 75th percentiles. (a–c) In the wet season, organics, nitrate and sulfate RPA to EC RPA ratios. The scatter plots show the corresponding correlations with $E_{\mathrm{abs}\mathrm{520}}$. The scatter plots share the same $y$-axis scale with the diurnal plots. Panels (d)–(f) are the same as (a)–(c) but for the dry season. The following ions are used: EC ($m/z+\mathrm{12}[\mathrm{C}{]}^{+}$, $+\mathrm{24}[{\mathrm{C}}_{\mathrm{2}}{]}^{+}$, $+\mathrm{36}[{\mathrm{C}}_{\mathrm{3}}{]}^{+}$, $+\mathrm{48}[{\mathrm{C}}_{\mathrm{4}}{]}^{+}$), organics ($m/z+\mathrm{43}[{\mathrm{C}}_{\mathrm{2}}{\mathrm{H}}_{\mathrm{3}}\mathrm{O}{]}^{+}$), sulfate ($m/z-\mathrm{97}[{\mathrm{HSO}}_{\mathrm{4}}{]}^{-}$, $-\mathrm{80}[{\mathrm{SO}}_{\mathrm{3}}{]}^{-}$) and nitrate ($m/z-\mathrm{62}[{\mathrm{NO}}_{\mathrm{3}}{]}^{-}$, $-\mathrm{46}[{\mathrm{NO}}_{\mathrm{2}}{]}^{-}$).

[Figure omitted. See PDF]

To study the relative abundance of coating materials on EC particles, we investigate the ratios of RPA by different species (organics, sulfate and nitrate) to RPA by EC in both wet and dry seasons (Fig. 7). In the wet season, the effect of organics and sulfate on EC-containing particles demonstrated similar diurnal trends that both peaked at 13:00, implying an association with photochemical reactions. The timing of the organic peak by SPAMS shown in Fig. 7a also agrees well with the bulk measurements of $\mathrm{SOC}/\mathrm{OC}$ (Fig. 4a). However, the diurnal variations in organics and sulfate were poorly correlated with $E_{\mathrm{abs}\mathrm{520}}$ as shown by the low $R^{\mathrm{2}}$ in the scatter plot in Fig. 7a and b. There are two possibilities for the lower $R^{\mathrm{2}}$ (organics, $E_{\mathrm{abs}}$) in Fig. 7d compared to $R^{\mathrm{2}}$ ($\mathrm{SOC}/\mathrm{OC}$, $E_{\mathrm{abs}}$) in Fig. 4b. First, organics shown in Fig. 7d contain both primary and secondary organics, while $\mathrm{SOC}/\mathrm{OC}$ ratio shown in Fig. 4b represents the secondary portion only. Second, poor diurnal correlations do not necessarily rule out the contribution of organics and sulfate to $E_{\mathrm{abs}}$ by analogy with the SOC correlation with $E_{\mathrm{abs}}$ as discussed in Sect. 3.6. Although the quantitative contribution estimation of sulfate and SOA to $E_{\mathrm{abs}}$ is not possible in this study, a rough estimation can be projected. Considering the typical annual average SOC concentration (3 $\mathrm{\mu }$gC m${}^{-\mathrm{3}}$) (Wu et al., 2019), typical $\mathrm{SOA}/\mathrm{SOC}$ mass ratio (1.8) (Li et al., 2017) and sulfate concentration (8 $\mathrm{\mu }$g m${}^{-\mathrm{3}}$) (J. Liu et al., 2017) in the PRD region, SOA and sulfate would likely have comparable contributions to the $E_{\mathrm{abs}}$, according to the $E_{\mathrm{abs}}$ dependency on sulfate-to-SOA mass ratio results by Zhang et al. (2018a). Summertime nitrate was low in daytime and high in the nighttime (Fig. 7c), which agrees with measurements at the roadside site in Hong Kong (Lee et al., 2015) and Shanghai (K. Li et al., 2018). Temperature-dependent gas–particle partitioning would be one of the possible reasons for the observed nitrate diurnal pattern (Appel et al., 1981; Xue et al., 2014; Griffith et al., 2015). A higher temperature during the daytime (Fig. S14) favors ${\mathrm{HNO}}_{\mathrm{3}}$ partitioning into the gas-phase in the wet season. The diurnal pattern of nitrate correlates well with that of $E_{\mathrm{abs}\mathrm{520}}$ ($R^{\mathrm{2}}=\mathrm{0.59}$) as shown in Fig. 7c, suggesting that $E_{\mathrm{abs}\mathrm{520}}$ was likely affected by temperature-induced gas–particle partitioning during the wet season. Studies in Nanjing (Ma et al., 2020) and Beijing (Xie et al., 2019) also observed a nitrate-evaporation-induced $E_{\mathrm{abs}}$ decrease, which is in agreement with our study. A previous chamber study has shown a decrease in $E_{\mathrm{abs}}$ due to SOA evaporation (Metcalf et al., 2013). By analogy with nitrate, organic compounds with volatilities similar to nitrate might potentially be involved in shaping the diurnal pattern of $E_{\mathrm{abs}}$ in the wet season.

In the dry season, organics were moderately correlated with $E_{\mathrm{abs}\mathrm{520}}$ ($R^{\mathrm{2}}=\mathrm{0.38}$) as shown in Fig. 7d. The improved correlation of organics in the dry season was in agreement with ${\mathrm{O}}_x$ and $\mathrm{SOC}/\mathrm{OC}$ results as shown in Fig. 4b. Sulfate was still poorly correlated with $E_{\mathrm{abs}\mathrm{520}}$. Since the contribution of sulfate on $E_{\mathrm{abs}\mathrm{520}}$ cannot be ruled out, one possible explanation is that the contribution of sulfate on $E_{\mathrm{abs}\mathrm{520}}$ was not reflected on the diurnal timescale.

4 Conclusions and implications

This study explored the temporal dynamics of the optical properties of carbonaceous aerosols in urban Guangzhou, a typical megacity in southern China, focusing on the atmospheric aging-induced light absorption enhancement of BC. Field measurements were conducted at an urban site during the wet season (31 July–10 September 2017) and dry season (15 November 2017–15 January 2018). A newly developed approach, the minimum $R$ squared (MRS) method (Wu et al., 2018), was successfully applied to determine the light absorption enhancement factor, $E_{\mathrm{abs}}$, using data from an Aethalometer and a field-deployable semicontinuous carbon analyzer. The MRS approach avoids a specialized instrument setup (e.g., thermal denuder and photo-acoustic spectrometer) for $E_{\mathrm{abs}}$ determination, hence it has a great potential for expending the data pool of $E_{\mathrm{abs}}$, considering the fact that colocated Aethalometer and field carbon analyzer measurements have been widely deployed around the world.

Strong seasonality of BC was observed. The average concentration of EC was $\mathrm{1.94}\pm \mathrm{0.93}$ and $\mathrm{2.81}\pm \mathrm{2.01}$ $\mathrm{\mu }$gC m${}^{-\mathrm{3}}$ in the wet and dry seasons, respectively. Collective evidence from remote sensing fire counts and ground measurements of levoglucosan showed that biomass burning (BB) was more active in the dry season. Consequently, optical properties of BC were effectively altered, leading to elevated MAE (dry season: $\mathrm{18.47}\pm \mathrm{5.49}$ m${}^{\mathrm{2}}$ g${}^{-\mathrm{1}}$, wet season: $\mathrm{10.73}\pm \mathrm{4.96}$ m${}^{\mathrm{2}}$ g${}^{-\mathrm{1}}$), MAE${}_{\mathrm{p}}$ (dry season: 15.8 m${}^{\mathrm{2}}$ g${}^{-\mathrm{1}}$, wet season: 8.1 m${}^{\mathrm{2}}$ g${}^{-\mathrm{1}}$) and AAE (dry season: $\mathrm{1.46}\pm \mathrm{0.12}$, wet season: $\mathrm{1.37}\pm \mathrm{0.10}$) in dry season compared to those in the wet season. However, little dependence of $E_{\mathrm{abs}}$ on wavelength was observed in the dry season despite the influence of BB.

The diurnal correlation analysis between AAE, $k$ and $E_{\mathrm{abs}}$ revealed different results between wet and dry seasons. During the wet season when the BB influence was small, AAE was well correlated with $E_{\mathrm{abs}}$, implying that the coating was likely the main driver for AAE $>$ 1. In other words, the two-component AAE model might not be suitable for BrC absorption estimation under such circumstances. The Aethalometer loading effect correction factor, $k$, was confirmed to be a useful $E_{\mathrm{abs}}$ indicator owing to its good correlation with $E_{\mathrm{abs}}$ during the wet season. In the dry season, the weak correlation between AAE and $E_{\mathrm{abs}}$ implies the contribution from BB to AAE. In the dry season, the BB influence leads to poor correlation between $k$ and $E_{\mathrm{abs}}$, confirming that $k$ can only be used as the coating indicator when BB influence is small.

The effect of atmospheric aging on the $E_{\mathrm{abs}}$ diurnal pattern was examined. ${\mathrm{O}}_x$ and $\mathrm{SOC}/\mathrm{OC}$ were found to be well correlated with $E_{\mathrm{abs}}$ during the dry season, but no correlation was observed in the wet season. However, further analysis showed an $E_{\mathrm{abs}}$ dependence on $\mathrm{SOC}/\mathrm{OC}$ in both wet and dry seasons. This observation implies that a poor diurnal correlation in the wet season does not necessarily rule out the $E_{\mathrm{abs}}$ contribution from SOC. In other words, the SOC contribution to $E_{\mathrm{abs}}$ in the wet season was not necessarily reflected in just the diurnal correlation.

In the wet season, a high-temperature-induced ($>\mathrm{32}$ ${}^{\circ }$C) $E_{\mathrm{abs}}$ decline was observed. In addition, a good diurnal correlation between nitrate and $E_{\mathrm{abs}}$ was found. These pieces of evidence imply the potential role of semivolatile coatings on BC in regulating the diurnal dynamics of $E_{\mathrm{abs}}$. In China, the sulfate problem has been effectively mitigated by the reduction measures implemented in recent years (Xia et al., 2016; Wang et al., 2017b). In contrast, nitrate increased substantially in recent years (Xu et al., 2019; M. Tian et al., 2019; H. Li et al., 2019; Wang et al., 2019). If the nitrate fraction in the coating materials on BC increases, the diurnal pattern of $E_{\mathrm{abs}}$ for BC may be affected by the fluctuation of nitrate content in aerosol particles. As a result, the increasing concentration of nitrate might potentially affect radiative forcing by BC in China.

Appendix A Abbreviations

Abbreviation	Definition
AAE${}_{\mathrm{470}\text{–}\mathrm{660}}$	Ångström absorption exponent between 470 and 660 nm
AFD	aerosol filter filtration–dissolution
Aeth	Aethalometer
ATN	attenuation
BB	biomass burning
BC	black carbon
BrC	brown carbon
$E_{\mathrm{abs}\mathrm{520}}$	light absorption enhancement factor at 520 nm
${\mathit{\sigma }}_{\mathrm{abs}\mathrm{520}}$	light absorption coefficient at 520 nm
${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{total}}$	total light absorption coefficient of a coated particle
${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{pri}}$	primary light absorption coefficient attributed to the soot core alone of a coated particle
${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{aging}}$	extra light absorption other than ${\mathit{\sigma }}_{\mathrm{abs}\mathit{\_}\mathrm{pri}}$ (including those from the lensing effect that arises from a non-absorbing coating on the soot core and secondary brown carbon during atmospheric aging)
eBC	equivalent BC mass concentration determined by optical methods (e.g., Aethalometer)
EC	elemental carbon
GDEMC	Guangdong Environmental Monitoring Center
GMB	Guangzhou Meteorological Bureau
$k_{\mathrm{1}}$, $k_{\mathrm{2}}$…$k_{\mathrm{7}}$	compensation factors (Eqs. 5 and 6) at seven wavelengths (370, 470, 520, 590, 660, 880 and 950 nm)
LO-OOA	less oxidized oxygenated organic aerosol
MAE${}_{\mathrm{520}}$	mass absorption efficiency at 520 nm, also known as mass absorption cross-section (MAC)
MAE${}_{\mathrm{p}}$	primary MAE of freshly emitted soot particles
MAE${}_{\mathrm{p}\mathit{\_}\mathrm{h}}$	a series of hypothetical MAE${}_{\mathrm{p}}$ tested in the MRS calculation
MAE $+$ empirical	MAE approach for $E_{\mathrm{abs}}$ quantification using empirical MAE${}_{\mathrm{p}}$
MAE $+$ SP	MAE approach for $E_{\mathrm{abs}}$ quantification using single-particle measurements for MAE${}_{\mathrm{p}}$ determination
MAE $+$ MRS	MAE approach for $E_{\mathrm{abs}}$ quantification using MRS for MAE${}_{\mathrm{p}}$ determination
MO-OOA	more oxidized oxygenated organic aerosol
MRS	minimum $R$ squared method
OC	organic carbon
OOA	oxygenated organic aerosol
PAS	photoacoustic spectrometer
PBL	planetary boundary layer
PBLH	planetary boundary layer height
POC	primary organic carbon
PRD	Pearl River Delta region, China
rBC	refectory black carbon (commonly used for reporting BC detected by SP2)
$r_{\mathrm{aged}}$	the ratio of aged particles to fresh particles determined by SP2
RPA	relative peak area
SOA	secondary organic aerosol
SP2	single-particle soot photometer
SPAMS	single-particle aerosol mass spectrometer
SP-AMS	soot particle aerosol mass spectrometer
SSA	single-scattering albedo
Suomi NPP	Suomi National Polar-orbiting Partnership
TD	thermal denuder
VIIRS	Visible Infrared Imaging Radiometer Suite
VOCs	volatile organic compounds
WODR	weighted orthogonal distance regression

Code and data availability

OC, EC and ${\mathit{\sigma }}_{\mathrm{abs}}$ data used in this study are available from the corresponding authors upon request. Data analysis and visualization toolkits (Histbox, MRS and Scatter Plot) used in this study are available at 10.5281/zenodo.832405 (Wu, 2020a), 10.5281/zenodo.832395 (Wu, 2020b) and 10.5281/zenodo.832416 (Wu, 2020c), respectively.

The supplement related to this article is available online at: https://doi.org/10.5194/acp-20-2445-2020-supplement.

Author contributions

CW designed the study. JYS and CW performed the experiments. JYS, CW, CC, QZ and YL conducted the data analysis. JYS and CW wrote the paper with the inputs from all authors.

Competing interests

Yanxiang Cen and Huiqing Nian both work for Guangzhou Hexin Analytical Instrument Co., Ltd.

Acknowledgements

The authors gratefully acknowledge the NOAA Air Resources Laboratory (ARL) for the provision of the HYSPLIT transport and dispersion model used in this publication. We acknowledge the use of data from the NASA FIRMS application (https://firms.modaps.eosdis.nasa.gov/, last access: 26 February 2020) operated by the NASA/Goddard Space Flight Center Earth Science Data and Information System (ESDIS) project.

Financial support

This research has been supported by the National Key Research and Development Program of China (grant no. 2016YFC0208503), the National Natural Science Foundation of China (grant nos. 41605002 and 41475004), the Guangzhou Science and Technology Project (grant no. 201604016053), the Major Project of Industry-University-Research Collaborative Innovation in Guangzhou (grant no. 2016201604030082), and the Pearl River Nova Program of Guangzhou (grant no. 201610010149).

Review statement

This paper was edited by Annele Virtanen and reviewed by two anonymous referees.