The sampling station is located at the Gulou campus of Nanjing University, urban area of Nanjing (32.05∘ N, 118.78∘ E). It is built on the roof of a 79.3 m-tall building, around which there are no industrial pollution sources within a 30 km radius but there are several main roads with apparent traffic pollution, especially at rush hours. The layout of the site and the corresponding climatology have been described in Zhu et al. (2012).

The wavelength-dependent AAC and concentrations of BC were derived from the measurements using a seven-channel Aethalometer (model AE-31, Magee Scientific, USA). The wavelength-dependent aerosol SC and Bsp were measured using a three-wavelength-integrating Nephelometer (Aurora 3000, Australia). To make a brief comparison between surface and column aerosols, the wavelength-dependent columnar aerosol optical depth (AOD) was observed using a Cimel sunphotometer (CE-318). The AE-31 model measures light attenuation at seven wavelengths, including 370, 470, 520, 590, 660, 880 and 950 nm, with a desired flow rate of 5.0 L min-1 and a sampling interval of 5 min. Aurora 3000 measures aerosol light scattering, including SC and Bsp at 450, 525 and 635 nm, with a sampling interval of 1 min. CE-318 measures the AOD from 340 to 1640 nm in the daytime. Routine calibrations and maintenances were carried out for all these instruments during the sampling periods. R-134 was used as a span gas for Aurora 3000. The aerosol inlet is located about 1 m above the roof. Data to be analysed in this study were measured from March 2014 to February 2016 for AE-31 and CE-318 and from June 2014 to February 2016 for Aurora 3000. Meteorological data (such as relative humidity) during the sampling period are from the National Meteorological Station of Nanjing (no. 58238).

The wavelength-dependent aerosol absorption coefficient (AAC) and BC mass concentration can be calculated directly based on the measured light attenuations (ATN) through a quartz filter matrix (Petzold et al., 1997; Weingartner et al., 2003; Arnott et al., 2005; Schmid et al., 2006): σATN,t(λ)=(ATNt(λ)-ATNt-1(λ))Δt×AV, where A (in m2) is the area of the aerosol-laden filter spot, V is the volumetric sampling flow rate (in L min-1) and Δt is the time interval (= 5 min) between t and t-1. σATN is the AAC without any correction, which is generally larger than the actual one (σabs) because of the optical interactions of the filter substrate with the deposited aerosol. Generally, there are two key factors leading to the bias: (1) multiple scattering of light at the filter fibres (multiple-scattering effect) and (2) instrumental response with increased particle loading on the filter (shadowing effect). Thus, the correction is needed and the calibration factors C and R (shown in Eq. 2) are introduced against the scattering effect and shadowing effect, respectively: σabs,t(λ)=σATN,t(λ)C×R. Collaud Coen et al. (2010) suggested that AAC corrected from Weingartner et al. (2003) (WC2003 for short, hereinafter) and Schmid et al. (2006) (SC2006 for short, hereinafter) have good agreements with the one measured by a Multi-Angle Absorption Photometer. These two corrections are similar to each other and they use the same R(λ): Rt(λ)=1f-1×ln⁡(ATNt(λ))-ln⁡10ln⁡50-ln⁡10+1, where R=1 when ATN ≤ 10 and f=1.2. However, C is fixed in WC2003 but is wavelength-dependent in SC2006. According to Wu et al. (2013) and Zhuang et al. (2015), C in Nanjing is 3.48 in WC2003 while it is 2.95, 3.37, 3.56, 3.79, 3.99, 4.51 and 4.64 at 370, 470, 520, 590, 660, 880 and 950 nm in SC2006. Zhuang et al. (2015) further suggested that wavelength-dependent AACs corrected by SC2006 might be closer to the real ones than WC2003s in Nanjing, although 532 nm AACs from these two corrections are close to each other. In addition to the direct method, AAC can be calculated indirectly: σabs,t(λ)=[BC]×γ, where [BC] is the mass concentration of Aethalometer BC (in µg m-3) without any correction and γ is the conversion factor determined empirically from linear regression of the Aethalometer BC concentration versus the aerosol absorption measurement (Yan et al., 2008). Zhuang et al. (2015) indicated that γ from the linear regression of the Aethalometer BC concentrations (ng m3) at 880 nm against the light absorption coefficient (Mm-1) at 532 nm in Nanjing is about 11.05 m2 g-1. It is obvious that only 532 nm AAC can be addressed from this way. Thus, AACs corrected from SC2006 are used in this study.

Based on wavelength-dependent AAC and SC, Ångström exponent of scattering (SAE) and absorbing (AAE) aerosols are estimated by the following: AAE470/660nm=-log⁡(AAC470nm/AAC660nm)/log⁡(470/660)SAE450/635nm=-log⁡(SC450nm/SC635nm)/log⁡(450/635). For purposes of comparison, AAC at 450, 525, 532, 550 and 635 nm, SC at 532 and 550 nm as well as Bsp at 532 and 550 nm were further calculated by the given coefficients and corresponding Ångström exponents: σλ=σλ0×λλ0-α, where σλ is the coefficient at wavelength λ and α is the corresponding Ångström exponents.

Based on wavelength-dependent SC, Bsp, AAC, aerosol asymmetry parameter (ASP), single-scattering albedo (SSA) and extinction coefficient (EC) are further estimated: ASPλ=-7.143889βλ3+7.46443βλ2-3.9356βλ+0.9893SSAλ=SCλSCλ+AACλECλ=SCλ+AACλ, where βλ is the ratio of Bsp to SC at wavelength λ. Equation (8) is derived from Andrews et al. (2006).

The aerosol absorption coefficient (AAC) was directly obtained from the measurement of AE-31 and the scattering and back-scattering coefficients (SC and Bsp), which were directly measured from Aurora 3000. Based on wavelength-dependent AAC and SC, the Ångström exponent of absorbing (AAE at 470/660 nm) and scattering (SAE at 450/635 nm) aerosols were estimated according to Eqs. (5) and (6). Based on AAC, SC and Bsp, wavelength-dependent aerosol asymmetry parameter (ASP), single-scattering albedo (SSA) and extinction coefficient (EC) are further estimated using Eqs. 8–10 and analysed. Table 1 lists the statistical summary of surface aerosol optical properties in the urban area of Nanjing during the sampling period. The annual mean AAC, SC, Bsp, EC, SSA and ASP at 550 nm, AAE at 470/660 nm and SAE at 450/635 nm are 29.615, 338.275, 44.257, 381.958 Mm-1, 0.901, 0.571, 1.583 and 1.320, with a standard deviation of 20.454, 228.078, 27.396 and 252.271 Mm-1, 0.049, 0.088, 0.228 and 0.407, respectively.

Figure 1 shows the 10th, 25th, median, 75th and 90th percentile values of the 550 nm AAC, SC, Bsp, 470/660 nm-AAE and 450/635 nm-SAE in the urban area of Nanjing in each season from March 2014 to February 2016. Default values of aerosol scattering properties in spring 2014 are blank because the measurements of Aurora 3000 started from June 2014. The figure suggests that AAC, SC, Bsp, AAE and SAE have substantially seasonal variations. A high level of AAC appears in winter (DJF) while a lower level is found in summer (JJA) (Fig. 1a). The temporal variability of Bsp is similar to that of AAC (Fig. 1d). According to Zhang et al. (2009), emissions of the aerosols and trace gases in China are larger in winter than in other seasons, especially for carbonaceous aerosols (Fig. 1c in Zhuang et al., 2013b). Thus, the higher AAC values in winter than in summer might result from the higher aerosol emissions, lower boundary height (Guo et al., 2016b) and less rainfall. However, possibly due to the impacts of hygroscopic growth of aerosol caused by higher RH in summer and dust aerosol in spring (Zhuang et al., 2014a), SC is considerably large in these two seasons (Fig. 1c). Thus, the lowest SC is found in autumn in both 2014 and 2015. AAE has a similar seasonality to AAC. Due to relatively higher RH, a small value of AAE is found in JJA while the larger ones appear in the other seasons (Fig. 1b), which is different from the seasonality of SAE. SAE is larger in warmer seasons but is smaller in the other seasons. Scattering aerosols, including inorganic and partially organic components, mainly come from gas-to-particle transformation, so that they have smaller sizes (larger AE) compared to the primary aerosols (such as dust and BC). The efficiency of gas-to-particle transformation is higher in warmer seasons. The observations of the aerosol compositions at the site showed that seasonal mean inorganic aerosols, including sulfate, nitrate and ammonium, account for about 50 % of the total PM2.5 in spring and might be higher than 50 % in the other seasons (Zhuang et al., 2014b). Thus, SAE in summer and autumn is large (Fig. 1e). RH can impose substantial influences on scattering aerosols. SAE might be much larger than the current values in these two seasons if the hygroscopic growth were excluded. Seasonal mean RH is about 75.41 and 70.86 % in JJA and SON, respectively, to a certain degree leading to higher values of SAE in autumn than in summer. The figure also suggests that the aerosol absorption coefficient and scattering coefficient as well as their sizes in 2014 are higher than in 2015. The observed RH difference in these 2 years at least partly accounts for the variation of aerosol absorption coefficient and scattering coefficient as well as their sizes. A comparison of RH between 2014 and 2015 indicates that RH is 79.49 and 72.86 % in JJA and SON, respectively, in 2014, larger than in 2015 (71.33 % in JJA and 69.03 in SON).

Figure 2 plots the seasonal mean values with standard deviations of AAC, SC, Bsp, EC, SSA, ASP, AAE at 470/660 nm and SAE at 450/635 nm. AAC, SC, Bsp and EC increase with decreasing wavelength in four seasons. Changes in SSA and ASP with increasing wavelength are different in different seasons. SSA increases with increasing wavelength in colder seasons but much less in JJA and SON. ASP increases with wavelength in JJA, contrary to in other seasons. The figure also suggests that seasonal variation of EC is more consistent with SCs, with large values in JJA and DJF (370.236 and 422.569 Mm-1 at 550 nm). The largest values of SSA and ASP appear in JJA (0.933 and 0.638 at 550 nm), implying that aerosols in the urban area of Nanjing are more scattering and have a stronger forward-scattering ability in JJA than in other seasons. The urban aerosols are more absorbent in SON and DJF in Nanjing (550 nm SSA is no more than 0.89). The seasonal variation of SSA is determined by the variations of both AAC and SC. As mentioned above, AAC is highest in winter and lowest in summer, to a great degree due to the seasonality of the emissions. SC would have the same variation as AAC if only emissions were taken into account. Zhang et al. (2009) indicates that the emission seasonality of carbonaceous aerosols are much stronger than the trace gases (such SO2 and NOx), and they show that the anthropogenic emission rate in winter is about 1.87 times that in summer for black carbon but only about 1.2 for SO2 in China. This is also supported by Sun et al. (2015), who found that the concentration of black carbon aerosol in north China was much higher in winter due to enhanced emissions based on 1-year observations. Thus, the different emission seasonal variations between black carbon and trace gases alone would cause a lower SSA in winter compared to that in summer. What is more, both a higher efficiency of gas-to-particle transformation and higher level of RH in summer are in favour of a much larger SC, which to a certain degree could further enlarge SSA in summer. The smaller SSA in the colder season might mainly be caused by a higher emission of absorbing aerosol.

Seasonal mean 550 nm AAC, SC, Bsp, EC, SSA and ASP, 470/660 nm AAE and 450/635 nm SAE as well as corresponding standard deviations are listed in Table 2. It suggests that seasonal mean 550 nm AAC, SC, Bsp, EC, SSA and ASP vary from 19.65 to 37.96, 294.62 to 385.14, 36.99 to 54.79 and 341.3 to 422.57 Mm-1, 0.874 to 0.933 and 0.54 to 0.64, respectively. Seasonal mean AAE and SAE vary from 1.49 to 1.70 and 1.1 to 1.54, respectively. AAC and Bsp in DJF are about 2 and 1.5 times of those in JJA, respectively. SSA in JJA is about 6.75 % larger than in SON.

In addition to seasonality, the aerosol optical properties near the surface at urban Nanjing have substantial diurnal variations (Fig. 3), especially for the coefficients (AAC, SC, Bsp and EC). The diurnal variation of EC, which is consistent with SC, is not shown in the figure. AAC levels are usually high at the rush hours around 07:00–09:00 LT and around 09:00–11:00 LT but low in the afternoon (Fig. 3a). At 08:00 LT, mean 550 nm AAC is about as large as 34 Mm-1, while at 02:00, it is about 23 Mm-1. SC and Bsp (Fig. 3b, c), to some extent, have diurnal variations similar to AACs. Their lowest values also appear in the afternoon (about 280 Mm-1 for SC and 38 Mm-1 for Bsp). However, only one peak of the aerosol scattering coefficient is found in the early morning (about 379 Mm-1 for Sc and 48 Mm-1 for Bsp) and it is about 1–2 h earlier than its absorption coefficient, possibly owing to the different emissions between these two types of aerosols. Absorbing aerosols in urban Nanjing mainly come from the vehicle emissions because of the developed transportation network, resulting in two peaks of AAC within 1 day (Zhuang et al., 2015). Scattering aerosol loadings are somewhat less affected by traffic emissions, especially at night-time. Their precursors, such as SO2 and NOx, mostly come from coal combustion and industrial emissions in urban Nanjing based on source apportionment. Therefore, there is no peak for SC or Bsp before midnight, although their values are considerably large (about 350 and 46 Mm-1). Different diurnal cycles between AAC and SC were also observed in the suburban area of Nanjing (Yu et al., 2016). Diurnal variations of AAC, SC and Bsp might be highly affected by the diurnal cycles of the boundary layer. The small coefficients in the afternoon are mostly induced by a well-developed mixing layer (Zhuang et al., 2014b). Generally, the boundary layer becomes more and more stable after sunset and its height lowers, which is conducive to the accumulation of air pollutants at night-time, especially during the period from midnight to sunrise. Therefore, SC usually peaks in early morning and the peak appears at different times in different seasons (05:00 LT in JJA and 08:00–09:00 LT in DJF). The daytime peak of AAC appears at 07:00 LT in JJA and at 09:00 LT in DJF. The diurnal variation of SSA also reflects the difference between AAC and SC (Fig. 3d), implying that aerosols in urban Nanjing are more scattering after midnight (SSA is about 0.91) but more absorbing before noon and midnight (SSA is about 0.89). Scattering aerosols mainly come from strong chemical production (gas-to-particle transformation) in the daytime, which to some extent might offset the dilution effect of the boundary on SC, thus leading to a relatively larger SSA in the afternoon. The figure further shows that both AAE (Fig. 3e) and SAE (Fig. 3f) in the daytime are slightly larger than those after midnight because both absorbing and scattering aerosols are fresher in the daytime whereas they are more aged before sunrise. Diurnal variations of SAE and AAE are relatively weaker compared to the corresponding coefficients. In addition to aerosol loadings, the level of Bsp is also affected by the size of the aerosols, as suggested by Yu et al. (2016), and so is ASP (Fig. 3g). The diurnal cycle of ASP is similar to that of Bsp but is opposite to that of SAE. A large ASP appears in the early morning (0.587) and the lower ASP in the afternoon (0.552).

The frequency of the aerosol optical properties is presented in Fig. 4 and Table 3. Similarly, the frequency of EC is not shown in the figure because it has similar pattern to SCs. Almost all of them follow a unimodal pattern. As listed in Table 3, the dominant ranges for all the aerosol optical properties are distributed around their annual mean values with different widths and they account for at least 60 % of the total samplings during the entire study period. The maximum frequencies of 32.9 % (AAC), 24.04 % (SC), 26.45 % (Bsp), 18.64 % (SSA), 20.9 % (AAE), 18.06 % (SAE) and 34 % (ASP) occur in the ranges from 9 to 21, 170 to 280 and 30 to 45 Mm-1, 0.91 to 93, 1.5 to 1.6, 1.32 to 1.5 and 0.55 to 0.62, respectively. Frequency distributions of the aerosol optical properties have substantially seasonal variations. The frequency peaks of the properties would be more concentrated at lower/higher ranges if their seasonal means are smaller/larger. As shown in Fig. 4a, c and e, relatively larger values or the peaks of frequencies for AAC, Bsp and AAE are concentrated in lower value ranges in JJA but in higher value ranges in the other seasons. Moisture absorption growth of absorbing aerosols leads to a leftward shift in an AAE frequency curve in JJA. Effects of dust aerosol also might result in a leftward shift in a SC frequency curve in spring (Fig. 4f). Furthermore, due to dust and RH, SC is considerably large in MAM and JJA, leading to relatively larger frequencies of SC distributed at larger SC ranges compared with the ones of AAC. As mentioned above, aerosols in urban Nanjing are more scattering and have a stronger forward-scattering ability in JJA than in the other seasons, thus larger frequencies occur more at higher-value ranges of SSA and ASP in JJA.

The east Asian monsoon is active in midlatitudes. Nanjing could be affected by the east Asian summer monsoon in JJA and by the winter monsoon in DJF. Airflows in these two seasons are significantly different (Fig. 5a, b), altering the aerosol optical properties in different seasons. Air masses mostly come from the oceans (about 77 %) in JJA and from continental regions in the north and north-west of China (57 %) in DJF. Only a few percentages of air masses are from the northern region of China in JJA. Additionally, considerable air masses arriving at the site are from the local areas (cluster 1 in JJA) or from places near Nanjing (cluster 1 in DJF). Therefore, the aerosol optical properties at the study site are characterized differently with different air masses in the two seasons.

As suggested by Zhuang et al. (2014b), high BC loadings in early June 2012 were observed at the site in Nanjing when the air masses were from a north-westerly direction, in which serious biomass burning was detected. Therefore, the aerosol optical properties are further analysed by their origins in both JJA and DJF (Fig 5c, d). In JJA, seasonal mean AAC, SC, Bsp, SSA, ASP, AAE and SAE are about 19.65, 340.87 and 36.99 Mm-1, 0.93, 0.64, 1.49 and 1.34, respectively (Table 2). The dominant air masses are from local areas (cluster 1 in Fig. 5a) and the East China Sea (passing through urban agglomeration regions (cluster 2) and less-developed regions (cluster 3) of the Yangtze River Delta YRD), accounting for 90 % of the total characteristics of the aerosol optical properties in urban Nanjing. All the values of the properties in the first three clusters are closer to their season means. Aerosol absorption and scattering coefficients from local emissions are larger than those in the other clusters. Although air masses in cluster 2 and cluster 3 come from the oceans and have the same level of relative humidity (RH), differences still exist between the clusters. The air masses have to cross the urban agglomeration (from Shanghai to Nanjing) of YRD when they arrive in Nanjing in cluster 2 but past less-developed regions (north Jiangsu Province) in cluster 3. In YRD, emissions of the aerosols and trace gases are much stronger in urban agglomeration regions than those in other areas as suggested in Zhang et al. (2009) and Zhuang et al. (2013b). Therefore, AAC and SC in cluster 2 are larger than in cluster 3 to some extent (Fig. 5a, c). Aerosols from these two clusters are more scattering than the local ones. There are two clusters (cluster 4 and 5 in Fig. 5a) from the remote areas in JJA. Aerosol loadings are relatively small when the air masses come from these two clusters. The size of the aerosols is finer (larger AAE in cluster 5 and SAE in cluster 4 and 5 in Fig. 5c). ASP varying with the clusters coincides with RH varying with the clusters (Fig. 5c), implying that RH might influence ASP significantly. In DJF, seasonal mean AAC, SC, Bsp, SSA, ASP, AAE and SAE are about 37.96, 385.14 and 54.79 Mm-1, 0.89, 0.54, 1.70 and 1.24, respectively (Table 2). Similarly to JJA, the aerosol absorption and scattering coefficients are the largest, of which AAC, SC and Bsp are about 1.3 times their seasonal means (Fig. 5d), when the air masses are local or from the regions (cluster 1 in Fig. 5b) near Nanjing in DJF. AAC, SC, Bsp, SSA and ASP are small but AAE and SAE are large if air masses are from remote areas. Aerosols are the smallest, most absorbing and finest when the air masses are from near Lake Baikal. ASP varying with the clusters also coincides with RH varying with the clusters in this season (Fig. 5d), further implying the effect of RH on ASP.

Substantial studies on aerosol optical properties have been carried out in China from monthly to annual scales. Table 4 lists some annual and seasonal statistics of measured surface aerosol optical properties from literature. Annual and seasonal means listed in the table are comparable to some extent, although the observational periods and instruments are different. It suggests that AACs and SCs in urban areas are much higher than those in rural and remote areas. In Beijing (the centre of the Beijing–Tianjin–Hebei region), annual mean AAC and SC were 56 and 288 Mm-1 in urban site during the period from 2005 to 2006 (He et al., 2009), which were much larger than the ones (17.5 and 174.6 Mm-1) in rural area (Yan et al., 2008). In Chengdu (Tao et al., 2014), Xi'an (Cao et al., 2012) and Wuhan (Gong et al., 2015), which are in the centre from south-west to central China, the annual mean scattering coefficients in these cities exceeded 450, 520 and 370 Mm-1, respectively. In the Pearl River Delta (PRD) region, seasonal mean AAC at 532 nm was about 84 and 188 Mm-1 at an urban site (Panyu), about 47 and 95 Mm-1 at a suburban site (Dongguan), about 26 and 28 Mm-1 at a rural site, and only 7.21 and 8.37 Mm-1 at a remote site (Yongxing Island), in spring and winter, respectively (Wu et al., 2013). Additionally, aerosols in urban areas are more absorbing. The aerosol absorptions in urban areas have stronger seasonality than those in rural areas (Table 4). Urban aerosols in Nanjing on an annual scale are somewhat lower but more scattering than in most cities in China. In addition to annual and seasonal means, there are considerable studies on monthly mean aerosol optical properties (e.g. Bergin et al., 2001; Xu et al., 2002, 2004; Li et al., 2007, 2015a, b; Andreae et al., 2008). A few studies on the aerosol optical properties in Nanjing have been carried out previously (Zhuang et al., 2014a, 2015; Yu et al., 2016) based on observations. They were more focused on the columnar aerosols (Zhuang et al., 2014a), single optical property (Zhuang et al., 2015), or shortening observations (two months in Yu et al., 2016). Here, substantial analyses of the key optical properties of the surface aerosol, to a certain degree, fill the gaps in the studies of aerosols in Nanjing, even in YRD.

The relationships between SC and AAC, and SC and Bsp are presented by season in Fig. 6. As shown in Figs. 3 and 4, these three types of coefficients have similar diurnal and frequency distributions. The linear correlation coefficient varies from 0.93 to 0.97 for SC and Bsp and from 0.66 to 0.87 for SC and AAC in urban Nanjing. It is obvious that relations between SC and Bsp are much better than those between SC and AAC in all seasons. The correlation between AAC and SC becomes poorer in MAM (0.66) and JJA (0.78) because the scattering aerosols are more affected by dust in spring and SC is more affected by RH in summer. The linear correlation coefficients between SC and AAC and between SC and Bsp in MAM at the site were a little smaller than in suburban Nanjing (Yu et al., 2016) in the same season in 2011. The slope of the fitting between Bsp and SC represents the levels of ASP. Analysis (not shown) suggests that ASP has a significant anti-correlation with the ratio of Bsp to SC (linear R=-0.98). Thus, a greater slope of curve represents a smaller ASP, thus less forward-scattering of the aerosols.

The correlations between ASP and SC under different RH conditions are illustrated in Fig. 7, showing that ASP has a quasi-log-normal distribution with SC, especially in lower RH conditions. ASP increases monotonically with increasing SC in low-RH ranges (Fig. 7a, b, RH < 60 %) and ASP mostly concentrates on small SC regions when RH is less than 40 % (Fig. 7a), implying that fine particles dominate the most in low-RH conditions, as also suggested by Andrews et al. (2006) and Babu et al. (2012). The correlation between ASP and SC becomes poorer with increasing RH (Fig. 7c), indicating that both fine and coarse aerosols might be equally important to the total SC.

Figure 8 shows the relationships between the SSA at 491 nm and extinction Ångström exponent (EAE) at 491/863 nm (Fig. 8a) as well as between SSA (863–491 nm) (short for dSSA) and EAE at 491/863 nm (Fig. 8b). Overall, SSA or dSSA to a certain degree have an anti-correlation with EAE in the urban area of Nanjing, especially the latter. Linear correlation coefficient is about -0.13 between SSA and EAE and about -0.75 between dSSA and EAE. Relationships between the SSA (or dSSA) and EAE to some extent reflect the aerosol types and sources as indicated by Russell et al. (2014), who proposed a method to identify the aerosol types based on the columnar aerosol optical properties (including SSA, EAE and the real refractive index) from the Aerosol Robotic Network (AERONET) retrievals. They suggested the following.

The polluted dust aerosol had smaller EAE (near 1.0) and SSA ranged from 0.85 to 0.95.

Atmospheric humidity has significant influences on the growth of particulate matter, subsequently affecting the sizes and absorbing/scattering abilities of the aerosols. As shown in Fig. 7a, c, high levels of SC are likely found in high RH ranges. Seasonal mean RH is largest in summer but lowest in winter (Fig. 9a). In summer, both trace gases and particulate matters have lower emission rates as suggested by Zhang et al. (2009). Furthermore, PBL height and precipitation mostly have larger values in this season than in other seasons. Thus, these three factors would result in smaller aerosol loadings in summer (such as Bsp is the smallest in this season). However, contrary to Bsp, SC in summer is larger than in spring and autumn, which might mainly result from the effects of high RH (Fig. 1c, d), although gas-to-particle transformation also contribute to SC to a certain degree in this season. Zhang et al. (2015) indicated that SC and Bsp in YRD would increase by 50 and 25 % as the RH increased from 40 to 85 % and the increment would become larger if there were considerable amounts of nitrate in fine particles. Nitrate in the urban area of Nanjing accounts for more than 20 % (as much as sulfate) of the total PM2.1 (Zhuang et al., 2014a). RH might also affect the size of the aerosols. The smallest AAE in JJA always corresponds to the highest RH, and vice versa (Figs. 1b and 9a). These results are consistent with Zhuang et al. (2014a), in which characteristics of columnar aerosol optical properties were investigated. Figure 9b further shows that AAE and SAE decrease monotonically with increasing RH. The correlation between ASP and RH is opposite to that between aerosol Ångström exponent and RH. The linear correlation coefficients are -0.36, -0.15 and 0.6 between AAE and RH, SAE and RH, and ASP and RH, respectively, in the urban area of Nanjing. ASP and RH are highly correlated with each other, which is also reflected in Figs. 2f, 5c, d and 9a, implying that RH might have considerable influence on the aerosol forward-scattering coefficient, hence on SC. These results could be used to correct the aerosol optical parameters in numerical models for estimating the aerosol radiative forcing in east China as suggested by Andrews et al. (2006), in the hope of reducing uncertainties in such estimations.

High levels of aerosol loadings would directly affect the visibility (VIS), which is one of the factors to be concerned about in current air quality forecasting in China. The forecast accuracy of visibility or haze pollution would be increased significantly if the effects of aerosols on visibility can be figured out. Instead of the loadings of the particulate matter, the aerosol optical properties here are used when investigating the aerosol effects on VIS.

Figure 10 shows the relations between extinction coefficient (EC) and VIS and between SC and VIS by season under different RH levels. Atmospheric VIS is found to decrease exponentially with increasing EC or SC in all seasons. The lapse rate of VIS with EC or SC is much larger in spring and summer than in autumn and winter. The lower VIS always appears at higher RH ranges and vice versa. In small VIS regions (such as < 4 km), VIS values are much smaller in JJA than those in the other seasons under the same SC level, implying the strong effects of RH on VIS. The effect of AAC on VIS has substantial seasonality and it is strong in SON but weak in MAM and JJA as illustrated in the fitting lines in the figure. A study on the effects of PM on VIS might be more reasonable if it used the aerosol optical properties rather than its mass concentrations. The linear correlation coefficient between EC and VIS varies from -0.69 (in JJA) to -0.87 (in DJF), and between SC and VIS, it varies from -0.71 (in JJA) to -0.87 (in DJF) in the urban area of Nanjing.

In addition to the SC or EC, the aerosol SSA and ASP also have good relationships with VIS as shown in Fig. 11, in which the effects of RH and SAE are also included (larger markers represent smaller SAE, but larger size of the aerosols). The aerosols would become coarser, less absorbing and more forward scattering to some extent with increasing RH, which subsequently further exacerbates the deterioration of visibility in all seasons. The linear correlation coefficients vary from -0.48 (in JJA) to -0.73 (in SON) between SSA and VIS and -0.47 (in JJA) to -0.80 (in MAM) between ASP and VIS in urban Nanjing. These results additionally illustrate that the scattering aerosols are still key factors affecting the atmospheric visibility, although the absorbing aerosols might have considerable influence on VIS in some seasons (Fig. 10c). The results in this study further indicate that effects of aerosols on air quality are complex.

A comparison between surface aerosol extinction coefficient and columnar AOD is performed (Fig. 12). Differences exist between EC and AOD, although they are well correlated with each other in each season. AOD to some extent is less affected by the development of boundary layer and more affected by the transport of aerosols compared to EC at the surface. The seasonal mean EC is large, both in JJA and in DJF, while the largest AOD is only found in JJA, which is possibly related to higher boundary layer height in JJA. A lower boundary layer would lead to more aerosol accumulation at the surface thus result in its smaller column burden. These differences (high surface aerosol loadings but low AOD) have also been simulated by a regional climate chemistry model in Zhuang et al. (2011, 2013). Overall, high AOD level corresponds to large EC value in each season, implying that aerosols in the upper layers mostly come from surface emissions in urban Nanjing. In some cases, long-distance transport of aerosols might contribute significantly to the AOD as shown in Fig. 12a, in which AOD exceeds 2; meanwhile EC is found to appear in low-value ranges. The slope of the linear fitting is larger in JJA (about 0.0016) than in the other seasons (all about 0.001), indicating that for a given value of EC, AOD would be higher in JJA, possibly because of higher humidity in summer. The columnar water vapour in summer is about 2 to 5 times of that in the other seasons.