The observation site NUPT (32.09∘ N, 118.78∘ E) was located in the campus of Nanjing University of Posts and Telecommunications. It is in an urban area of eastern China, one of the China's economic centers and also one of the world's most advanced manufacturing bases. This site represents a severely anthropogenically influenced environment, with a complex pollution pattern contributed by vehicles, industries, biomass burning and atmospheric transport.

The sampling campaign was conducted from 14 to 30 January 2015. PM2.5 samples were collected at a flow rate of 1.05 m3 min-1 with a nominal sampling time of 3 h for each sample. Blank filters were kept in the sampler for 1 min, without air flow, at both the beginning and the end of this campaign. All filters used for sample collection were quartz fiber filters, which had been prebaked at 450 ∘C for 6 h to eliminate organic material.

OC and EC were detected using a Desert Research Institute (DRI) Model 2001 Thermal Optical Carbon Analyzer (Atmoslytic Inc., Calabasas, USA). A portion of sampled filter (0.53 cm2) was analyzed following the IMPROVE thermal optical reflectance (TOR) protocol (Bao et al., 2017). For the determination of WSOC, a portion of each filter with a size of 4.02 cm2 was removed from the parent filter and extracted with 10 mL ultrapure water in a sonication bath for 30 min. The water extracts were filtered through syringe filters (0.22 µm, ANPEL, Shanghai, China) for WSOC concentration analysis by a TOC analyzer (TOC-L, Shimadzu, Kyoto, Japan) following the WSTC-WSIC protocol. All the calculations related to the carbonaceous components were blank corrected and the system error between DRI and TOC-L analyzers was corrected by testing a sucrose solution of a specific concentration.

A portion of each sample filter (5.07 cm2) was cut off and extracted with 10 mL ultrapure water (>18.2Ω) via ultrasonic agitation (30 min). The extract solution was filtered to remove insoluble materials and then used for ion analysis by ion chromatography on a ThermoFisher Scientific ICS-5000+ system (US) equipped with a gradient pump, a conductivity detector/chromatography compartment and an automated sampler. The separation of cations was carried out on an IonPac CS12A analytical column and an IonPac CG12A guard column with aqueous methanesulfonic acid (MSA, 30 mM L-1) eluent at a flow rate of 1 mL min-1. While anions were separated on an IonPac AS11-HC analytical column and an IonPac AG11-HC guard column using sodium hydroxide (NaOH) gradient elution at a flow rate of 1.5 mL min-1: 0–3 min, 0.5 mM L-1; 3–5 min, 0.5–5 mM L-1; 5–15 min, 5–30 mM L-1; 15–20 min, 0.5 mM L-1.

The extraction procedure for carbohydrate analysis was basically same as that for ions, except that the water volume was changed from 10 to 2 mL. The equipment of the ion chromatography system used was largely the same, except that a conductivity detector was replaced by an electrochemical detector. The carbohydrates were separated by a CarboPac MA1 column and a matched guard column using NaOH gradient elution at a flow rate of 0.4 mL min-1. To determine levoglucosan and galactosan the elution was run at a time gradient: -15–34 min, 300 mM L-1; 34–45 min, 300–480 mM L-1; 45–60 min, 480 mM L-1. For mannosan, the elution was run at another time gradient: -15–5 min, 50 mM L-1; 5–25 min, 250 mM L-1; 25–35 min, 250–350 mM L-1; 35–40 min, 350–650 mM L-1; 40–70 min, 650 mM L-1. The volume of the sample loop was 200 µL. All ions and carbonaceous components amount calculations were corrected by the mean value of two field blanks.

A fraction of sampled filter (5.02 cm2) was cut and extracted with 2 mL ultrapure water. The filter extract was tested for light absorption by an ultraviolet-visible absorption spectrophotometer (UV-2600, Shimadzu, Kyoto, Japan) with a scanning wavelength range of 200–800 nm. The absorption coefficient (babs, M m-1 or 10-6 m-1) at 365 nm was calculated by Eq. (1): 1babs=(A365-A700)×(Vwater×factor)×ln⁡(10)÷(Vaero×L), where A365 and A700 refer to absorbance (or light attenuation) at 365 and 700 nm, respectively, measured by the spectrometer; Vwater (mL) corresponds to the volume of the aqueous extract (water); the factor is set to 103, which is estimated from the absorption signal for the full filter; Vaero (m3) refers to the volume of air filtered; and L (mm) is the path length of the cell (10 mm). The choice of 365 nm as the reference wavelength to represent light absorption of WSOC is made according to the strong light-absorbing capacity and also is meant to avoid light-absorbing disturbance by other substances in the extract. Absorbance at 700 nm represents the baseline drift in the analysis (Bosch et al., 2014; Cheng et al., 2011; Hecobian et al., 2010). Mass absorption efficiency (MAE, m2 g-1) of BrC in WSOC was derived from Eq. (2): 2MAE=babs/WSOC, where WSOC refers to the concentration of WSOC (µg m-3). Absorption Ångström exponents (AAE) were computed by Eq. (3): 3babs≈K⋅λ-AAE, where K is a constant, AAE describes the wavelength-dependent absorption enhancement of BrC, associated with its origin, size and composition (Bikkina, 2014); and λ is wavelength in the range of 310–460 nm. AAE for WSOC was fitted within the range of 310–460 nm in which coefficients of determination (R2) for all samples are above 0.90. The solar energy absorbance of WSOC and EC at the ground level was estimated using Eqs. (4) and (5): 4EWSOC=∫Iλ⋅{1-e-(MAE⋅(365λ)AAE⋅WSOC⋅hABL)}dλ,5EEC=∫Iλ⋅{1-e-(MAEEC⋅(550λ)⋅EC⋅hABL)}dλ, where I(λ) is the clear-sky Air Mass 1 Global Horizontal (AM1GH) solar radiance spectrum at the surface; hABL refers to the vertical height of the boundary layer (set to 1000 m); MAEEC is set as 7.5±1.2 m2 g-1, which is the MAE of EC at 550 nm; and EC (µg m-3) is the EC concentration in each sample. The fraction EWSOC/EEC is the ratio of EWSOC and EEC, comparing the light-absorbing capacity of BrC in WSOC and EC. Here EWSOC, EEC and EWSOC/EEC were computed in a wavelength range of 300–400 nm (Bosch et al., 2014; Kirillova et al., 2014b; Bond and Bergstrom, 2006; Yan et al., 2015).

We also estimated the radiative forcing of WSOC and EC with a stand-alone radiative transfer model RRTMG_SW (Global climate model version of Short Wave Rapid Radiative Transfer Model) (Iacono Michael et al., 2000). The correlated-k approach was applied in RRTMG_SW to calculate radiative fluxes and heating rates in 14 contiguous bands in the shortwave (820–500 00 cm-1). The stand-alone RRTMG_SW can be driven by a given atmospheric profile. In this study, we prescribed the midlatitude winter atmospheric profile, which is default in the RRTMG package (Iacono Michael et al., 2008; Mlawer and Clough, 1998). We chose the average solar zenith angle of 60∘ in the midlatitude regions (Li, 2017) and ran RRTMG_SW for clear-sky conditions without the impact of cloud.

Aerosol radiative (absorptive) properties for WSOC and EC are parameterized in terms of their mass concentrations in RRTMG_SW. The optics per unit mass (mass absorption efficiency, MAE*; single scattering albedo, SSA*; and asymmetry factor, g*) of EC was provided from the Optical Properties of Aerosols and Clouds (OPAC) dataset (Hess et al., 1998). For WSOC, we calculated the optics using the Mie model (Bohren and Huffman, 1998; Pruppacher and Klett, 1997) with the input optics and prescribed size parameters. Refractive index (RI) was set to be 1.55–0.112i (Kirchstetter et al., 2004; Shamjad et al., 2016). The density of WSOC is 1.569 g cm-3 (Feng et al., 2013). The dry-mode radius and standard deviation of the WSOC particle size distribution were assumed to be 0.0212 µm and 2.24, respectively (Hess et al., 1998). Because WSOC is hydrophilic, we also calculated the wet particle radius and wet RIs according to Pruppacher and Klett (1997). The optics of WSOC obtained from the Mie model were provided to RRTMG_SW for radiative calculations.

The FLEXPART Lagrangian particle dispersion model was used to estimate the footprint of the site (Stohl et al., 1998, 2005; Grythe et al., 2017). FLEXPART version 10.1 was run in the backward model in which the potential emission sensitivity (PES) of the receptor point is provided (Seibert and Frank, 2004). The operational reanalyzed data from the European Centre for Medium-Range Weather Forecasts (ECMWF) at a spatial resolution of 1∘×1∘ with 61 vertical levels were used as meteorology. The simulation considered BC (corresponding to EC in this study) as the tracer species, and dry deposition and wet scavenging were accounted for. The output was set as the retention time(s) of particles in each grid during the simulation period. The HYSPLIT 4.8 model from NOAA was used to compute 48 h backward trajectories of air masses reaching Nanjing sites at a planetary boundary layer (PBL) height of 500 m. MODIS fire hotspot data were applied to evaluate open biomass burning intensity during the study period. The Potential Source Contribution Function (PSCF) model was usually applied to localize the potential sources of pollutants. The details about the setup of the model can be seen in the research of Bao et al. (2017). Here we introduce the light absorbance of WSOC into the model. The higher the PSCF value, the higher the possibility that the areas make potential contributions to the light absorbance of WSOC in the aerosols at the receptor site.

The temporal variations of OC/EC ratios and OC and EC concentrations in PM2.5 during the studied episode are plotted in Fig. 1. The average concentrations of OC and EC were 19.1±8.6 and 6.6±3.5 µg m-3, respectively (Table 1). Our results are comparable to the seasonal mean levels of OC and EC in PM2.5 in the winter of 2015 in a nearby site in Nanjing, which were 22.5±9.6 and 8.2±3.1 µg m-3, respectively (Li et al., 2015). OC shows a robust relationship with EC (R2=0.90, p<0.01), which underlines the significant contribution of fuel burning to OC, since EC is only released by inefficient combustion (Liu et al., 2014a). In this study, EC peaked at 17:00 LT on 21 January and 20:00 LT on 24 January, implying intensive fuel burning at those times. OC ranged from 5.5 to 45.8 µg m-3 and EC ranged from 0.8 to 20.1 µg m-3. The relative standard deviation was used to represent the variation of OC (44 %) and EC (53 %), suggesting that EC had a larger uncertainty than OC. It indicates that EC came from a variety of combustion sources which were dynamic and unstable. The variation of OC and EC over time led to the significant changes in PM2.5 levels.

The OC/EC ratios varied from 2.0 to 7.3, with a mean value of 3.2±0.8. The OC/EC ratio is usually used as the indicator of secondary organic carbon (Hou et al., 2011; Zeng and Wang, 2011). According to the temporal variation of the OC/EC ratio in Fig. 1, the impact of secondary organic carbon varied in different stages of the pollution episode. The OC/EC ratio variation range was 2.0–7.3 when the atmosphere was cleaner. Generally, an OC/EC ratio above two means a significant contribution of secondary organic aerosol (SOA) (Haque et al., 2019; Kunwar and Kawamura, 2014). As the pollution got more severe over time, it shrank to 2.0–4.0 after 16 January, indicating that the secondary sources had a relatively stable but still considerable impact on PM2.5 at the polluted stage. According to the previous studies, secondary organic aerosol usually played an important role in the wintertime air pollution (Huang et al., 2014; Sun et al., 2013).

Temporal variations of WSOC concentration and WSOC/OC ratio are illustrated in Fig. 1. The average concentration of WSOC in the event was 9.7±4.3 µg m-3 and WSOC varied from 2.2 to 23.2 µg m-3. WSOC was at a higher level in this pollution episode than in the episodes reported by Du et al. (2014a). It was reported that WSOC was abundant in biomass burning emissions in previous studies (Jaffrezo et al., 2005; Park and Cho, 2011). In addition, a strong correlation was also found between OC and WSOC (R2>0.70, p<0.01), and the WSOC/OC mass ratio was averaged to 0.5±0.1, a value consistent with that of another study (Bikkina, 2014). If a major portion of OC was derived from combustion, WSOC could be very likely released from fuel burning. Moreover, WSOC is an important form of BrC, which is able to absorb light radiation. It seems that biomass burning can emit BrC, increasing the light-absorbing capacity of aerosols.

To explore the combustion sources, the roles of traffic and industrial emissions, which are main fossil fuel combustion sources in the urban area (Cao et al., 2005), were first investigated. Nitrogen dioxide (NO2) is the main chemical component in motor vehicle exhaust (Kendrick et al., 2015), and sulfur dioxide (SO2) is usually treated as the tracer for coal combustion widely existing in industrial activities (Akimoto and Narita, 1994). Therefore, the sum of NO2 and SO2 concentrations can represent the contribution of fossil fuel combustion to the atmospheric pollutants. As shown in Fig. 2, carbon monoxide (CO) and the sum of NO2 and SO2 concentrations have a positive correlation, but with a low correlation coefficient (R2=0.40, p<0.01). The same pattern is also shown between EC and sum of nitrate (NO3-) and sulfate (SO42-) in PM2.5 (R2=0.46, p<0.01). These patterns suggest that fossil fuel was not the only main type of fuel and that biomass fuel might have been burned during the episode.

Tracers of biomass burning have been introduced to describe the role of biomass burning in this event. As shown in Fig. 3, all tracers strongly correlate with each other (R2>0.60, p<0.01). But there is a more significant correlation between anhydrosugars (R2>0.75, p<0.01) than that between nss-K+ and anhydrosugars (0.61≤R2≤0.67, p<0.01). Because of the non-unique origin of potassium, nss-K+ seems a weak tracer of biomass burning, in agreement with other published results. K+ has additional significant sources other than biomass burning, such as seawater, soil resuspension and fertilizers (Urban et al., 2012). Furthermore, K+ is abundant in firework aerosols according to several recent studies (Cheng et al., 2013; Drewnick et al., 2006). Among all three types of anhydrosugars, levoglucosan was treated as the main tracer in this study because of its larger quantity in PM2.5 than the other two.

Time series plots of nss-K+, levoglucosan, mannosan and galactosan in PM2.5 are shown in Fig. 1. In this study, nss-K+ varied from 0.2 to 3.8 µg m-3, with a mean level of 1.2±0.7 µg m-3. A consistent trend between nss-K+ and EC can be seen in Fig. 1, especially with both peaks of EC and nss-K+ during 14:00–17:00 LT on 21 January (15.7 and 3.8 µg m-3) and during 17:00–20:00 LT on 24 January (20.1 and 3.5 µg m-3), suggesting that the intensities of total fuel combustion and biomass fuel combustion changed synchronously.

The average concentration of levoglucosan in PM2.5 was 373±268 ng m-3 in this study, which was significantly higher than those observed at the coastal sites in Europe (Puxbaum et al., 2007; von Schneidemesser et al., 2009), the forest sites in the eastern China (W. Wang et al., 2008) and some rural sites in Canada and Hong Kong (Leithead et al., 2006; Sang et al., 2011). But for the rural sites of Australia, levoglucosan was found to be much more enriched than in this study (Reisen et al., 2011). For the urban areas, levoglucosan levels in this study were higher than the level in Shanghai, Hong Kong, Singapore and the southeastern US (X. Li et al., 2016; Sang et al., 2011; Yang et al., 2013; X. Zhang et al., 2010) but was lower than those measured in Grenoble, France, and also Beijing during winter and a biomass burning episode in summer (Cheng et al., 2013; Favez et al., 2010) (Table 2).

The concentration range of levoglucosan in PM2.5 of our study was 22.4–1476 ng m-3. The maximum value approximately approached that observed in Shanghai in the autumn when biomass burning prevailed (1606 ng m-3, Table 2). The highest level of levoglucosan appeared during 05:00–08:00 LT on 25 January, implying that the strongest biomass fuel combustion occurred in this period. Remarkably, levoglucosan concentration presented a valley during 11:00–14:00 LT on 19 January, when EC presented a peak. It can be explained that the intense combustion at that time was not dominated by biomass burning. The discrepancy of the time distribution between nss-K+ and levoglucosan were again attributed to the diversity of origins of potassium.

The concentrations of mannosan and galactosan in PM2.5 were averaged to 18.5±12.5 and 9.9±6.7 ng m-3, with ranges of 2.1–56.2 and 1.4–32.2 ng m-3, respectively. The trend of mannosan, galactosan and levoglucosan were closely matched, as illustrated in Fig. 1. They reached their highest and second highest levels together, which highlighted the fact that biomass burning heavily impacted the compositions of aerosols in these two periods.

Figure 4 shows the correlations between carbonaceous components (OC, EC or WSOC) and biomass burning tracers (levoglucosan, mannosan, galactosan or nss-K+). All biomass burning tracers are associated with EC (0.40<R2<0.80, p<0.01), which demonstrates the conjecture that biomass burning was one of the main types of combustion. Moreover, biomass burning tracers exhibit strong correlations with OC and WSOC (0.40<R2<0.80, p<0.01), indicating that biomass burning might have contributed to organic carbon including WSOC. For each tracer, the higher coefficient of determination (R2) for OC than EC illustrates that the biomass burning impacts on OC were more significant than those on EC. Notably, the slopes and the intercepts of the trend line for WSOC are both about half of those for OC, in accordance with results shown in Sect. 3.1 that WSOC accounted for around 50 % of OC. As a consequence, biomass burning made a stable contribution to WSOC and the contribution was even larger than that to EC in this episode.

The contribution of biomass burning to OC in PM2.5 (BB-OC) was calculated using Eq. (6): 6BB-OC=(lev/(1000×OC))ambient0.082×100%,lev:ngm-3;OC:µgm-3, (Z. Zhang et al., 2010; Puxbaum et al., 2007; Zdráhal et al., 2002; Sang et al., 2011). In biomass burning source emission tests for three major types of cereal straw (corn, wheat and rice) in China, 0.082 was reported as the lev/OC ratio for PM2.5 (Zhang et al., 2007). This value can be used in combination with the lev/OC ratios of our PM2.5 samples to roughly estimate the contribution of biomass burning smoke to the ambient OC. The reason why we used the ratio lev/OC associated with cereal straw burning will be explained in Sect. 3.5. As shown in Table 1, BB-OC ranged from 0.2 % to 53.6 % in this event, with an average of 20.9±9.3 %. Our calculated biomass burning contribution has a greater span than that of the Beijing rural area (18 %–38 %), which was computed with the same lev/OC ratio (T. Zhang et al., 2008; Y. Zhang et al., 2008), even though January was not in a common biomass burning prevailing season of Nanjing.

The contribution from biomass burning to the WSOC in PM2.5 (BB-WSOC) trend in Fig. 5 makes the above analysis more convincing due to the consistency of the contributions computed by two different methods. BB-WSOC was estimated with Eq. (7): 7BB-WSOC=lev/WSOCambient0.17×100%,lev/WSOC:µgµgC-1, which used a lev/WSOC ratio of 0.17 µg µgC-1 from the test burns of rice straw and wheat straw (Yan et al., 2015). BB-WSOC ranged from 1.1 % to 55.4 %, with an average of 22.3±9.9 %. It was at an equivalent level with that investigated in Beijing wintertime, where BB-WSOC was averaged at 23±7 % (Yan et al., 2015) and had a pretty large span. BB-WSOC exhibits a similar trend with BB-OC (Fig. 5). The robust relationship between BB-OC and BB-WSOC (R2=0.81, p<0.01) with a regression slope close to 1 (0.96) confirms the reliability of our biomass burning contribution quantification. However, the biomass burning contribution here could be related to both local emissions and long-range transport due to the atmospheric motion and the subsequent mixing of aerosols. The origins of biomass burning will be discussed in the following section.

The light absorption coefficient of WSOC at 365 nm is used to represent water-soluble BrC, which is organic carbon with a capacity of absorbing radiation (Yan et al., 2015; Hecobian et al., 2010; J. Wang et al., 2018). The mean babs value was 9.4±4.8 M m-1, higher than the summertime value but lower than the wintertime value in Beijing and Guangzhou (Yan et al., 2015; Qin et al., 2018). Meanwhile, it had a large span from 1.6 to 30.0 M m-1, partially attributed to its dependence on biomass burning intensity. As shown in Fig. 9, babs is significantly correlated with biomass burning tracers (R2>0.50, p<0.01), indicating that a considerable quantity of BrC was associated with biomass burning emissions in this episode. According to the analysis in Sect. 3.4, the robust correlations between WSOC and biomass burning tracers (levoglucosan, mannosan, galactosan or nss-K+) imply that biomass burning had contributed to WSOC. With respect to this point, the correlation between babs and WSOC (R2=0.76, P<0.01) and that between babs and WSOC from biomass burning, calculated with WSOC multiplied by the BB-WSOC fraction (R2=0.55, P<0.01), are presented in Fig. 10. Both correlations are significantly positive at the 1 % possibility level. But babs has a relatively higher R2 with WSOC than WSOC from BB, indicating that other sources also contributed to water-soluble BrC in the episode. Other researchers pointed out that secondary photochemical reactions in the atmosphere can produce water-soluble BrC as well (Sareen et al., 2013; Liu et al., 2016).

It is evident that the absorption coefficient of WSOC shows a relationship with wavelength, which can be described by the absorption Ångström exponent. AAE is also used to distinguish the BrC emission types due to its dependence on particle size and composition. Generally, the Ångström exponent for compounds from biomass burning and biofuel emissions is larger than 6 (Hoffer et al., 2006; Chen and Bond, 2010; Hecobian et al., 2010; Bikkina, 2014). The AAE mean in this study was 6.6±1.3, ranging from 4.1 to 10.3, and more than 68 % of samples (71 out of 104) had an AAE above 6 (Fig. 13). Our result is consistent with those for the Indo-Gangetic Plain, where biomass burning emission has been proven to be the predominant aerosol source (Bikkina, 2014). If light absorption is dominated by elemental carbon, AAE is reported to be ∼1 (Bond, 2001; Kirchstetter et al., 2004). For fresh and aged SOA it should be ∼7 and ∼4.7, respectively (Bones et al., 2010).

Mass absorption efficiency characterizes the efficiency of light absorption by WSOC. As shown in Fig. 11, the MAE mean in this study was 1.0±0.2 m2 g-1, with a range of 0.5–1.6, higher than that of some US cities. Beijing (summer) and the Indo-Gangetic Plain were effected by biomass burning, but to a lesser extent than some Indian areas and Beijing, China, during wintertime. The MAE was lower than 1.79 m2 g-1 in Beijing during winter. As shown in Fig. 11, MAE has poor correlation with AAE (Yan et al., 2015; Kirillova et al., 2014b; Kim et al., 2016; Du et al., 2014b; Hecobian et al., 2010; Yan et al., 2017; Cheng et al., 2011, 2016; Zhang et al., 2013; Bikkina, 2014; Srinivas et al., 2016).

As shown in Fig. 13, the solar energy absorption of WSOC (EWSOC) in short wavelengths (300–400 nm) had a mean value of 0.8±0.4 W m-2, ranging from 0.2 to 2.3 W m-2. The counterpart contributed by biomass burning (EWSOC_BB), derived from multiplying EWSOC by BB-WSOC, had a mean value of 0.2±0.1 W m-2, ranging from 0.0 to 0.9 W m-2. Solar energy absorption of EC (EEC) in the wavelength range of 300–400 nm had a mean value of 3.4±1.7 W m-2, ranging from 0.4 to 9.9 W m-2. The WSOC to EC absorption percentage (EWSOC/EEC) was 23.8±8.7 % and could be up to 73.0 % at the short wavelengths (300–400 nm), which is much higher than that for Beijing winter (42 %) (Yan et al., 2015). The map in Fig. 12 specifies the level of PSCF values related to the carbonaceous compounds (levoglucosan, WSOC and EC) and their radiative absorption properties (babs, EWSOC, EEC, EWSOC/EEC) in terms of a color bar. The areas with high PSCF values were interpreted as the potential areas loading the carbonaceous compounds or better light absorption. As shown in Fig. 12, the water-soluble carbonaceous aerosols transported to Nanjing over a long distance had a stronger light absorption capacity, compared to those aerosols from local emissions. It is evident that WSOC would enhance its light absorption capacity after aging along the long-range air mass transport, and the enhancement is much more than that for EC (Kirillova et al., 2014a). It means that the carbonaceous emissions over long-range transport can affect the local atmospheric solar radiation balance more. In addition, the levoglucosan has a similar PSCF distribution with WSOC and EC, illustrating that long-range-transported biomass burning emissions profoundly impacted the carbonaceous aerosols in Nanjing during this episode. The similar PSCF distributions between WSOC and babs or EWSOC can be explained by the fact that light absorption of WSOC is usually related to WSOC levels. The similarity of PSCF distributions between EC and EEC can be explained in the same way. However, the PSCF distribution of EWSOC/EEC is different from EEC or EWSOC in Fig. 12, indicating that southeastern China is potentially the area loading the aerosols with a higher light-absorbing contribution by WSOC. It is consistent with the fire spot distribution in Fig. 7, which implies that the long-range transported biomass burning emission significantly impacted the solar energy absorption of carbonaceous aerosols in Nanjing during this episode.

In Table 3, we show the optical properties (MAE*, SSA* and g*) for EC and WSOC (at RH =0.7) in the band 26 (22 650–29 000 cm-1, 345–442 nm) as provided in the RRTMG_SW model. Based on the surface mass concentration of EC and WSOC measured in this study, RRTMG_SW simulated the respective absorptive properties in the first layer of the model (the lowest layer above the ground), which accounts for the aerosol absorption in the radiative transfer calculation. To obtain the absorptive radiative forcing (ARF) of EC or WSOC, RRTMG_SW was run twice by including and excluding EC or WSOC in the model. The differences in the radiative flux between the two runs indicate the absorptive radiative effect of EC or WSOC in the atmosphere.

In the wavelength range of 345–442 nm, the change in the net flux at the surface caused by absorption of EC and WSOC was -3.4±1.6 and -1.4±0.6 W m-2, respectively. The ARF at the top of the atmosphere (TOA) was 0.8±0.4 W m-2 for EC and -0.2±0.1 W m-2 for WSOC. Therefore, the ARF in the atmosphere caused by absorption of EC (EEC*) and WSOC (EWSOC*) was 4.2±2.0 and 1.2±0.5 W m-2, respectively, which are slightly higher than the previously estimated EEC and EWSOC, indicating that the absorbing aerosols maintain a part of the solar energy in the atmosphere. Our estimates from the model indicate that the absorption ability of WSOC was equivalent to about 30 % of EC absorption in this study, which is comparable to the ratio from previous studies. For example, Lin et al. (2014) simulated the global OC radiative forcing and indicated that the absorption due to OC (both primary and secondary OC) aerosols is about 27 %–70 % of the EC warming effect in the global atmosphere. Liu et al. (2014b, 2015) investigated the absorption of OC aerosols at different altitudes over the North American continent and found that OC absorption at TOA is around 20 % of EC direct effect in the background troposphere. Our study provides additional evidence and stresses the necessity to account for the absorption capacity of WSOC in model simulations of the global energy budget.