A novel study of the morphology and effective density of 1 externally mixed black carbon aerosols in ambient air using 2 a size-resolved single-particle soot photometer ( SP 2 ) 3

16 The morphology of externally mixed black carbon (extBC) aerosols, an important factor 17 affecting radiative forcing, was studied by a tandem method coupling a differential 18 mobility diameter (DMA) with a single-particle soot photometer (SP2). Ambient 19 particles were selected by the DMA, and the size-resolved extBC particles were 20 distinguished from those of a thick coating (internally mixed) and quantified by the SP2. 21 Time differences between the DMA size selection and the SP2 measurement were 22 processed previously, as well as the effects of multicharged particles. Based on the 23 mass-mobility relationship, the fractal dimension of the extBC particles was obtained, 24 with a value of 2.36±0.04. This value is comparable with those of diesel exhaust 25 particles, implying a predominant contribution of vehicle emissions to the ambient 26 extBC in urban Beijing. The effective densities (ρeff) of the extBC in the mobility 27 diameter range of 140–750 nm were also derived, with values gradually decreasing 28 Atmos. Meas. Tech. Discuss., https://doi.org/10.5194/amt-2018-408 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 10 December 2018 c © Author(s) 2018. CC BY 4.0 License.


Introduction
Black carbon (BC), a byproduct of incomplete combustion, is the main light-absorbing component in atmospheric aerosols.BC can lead to positive radiative forcing only second to CO2 and thus Earth warming (IPCC, 2013).However, there remains a large amount of uncertainty regarding the radiative forcing induced by BC due to its complexity and variability in the morphology, mixing state and hygroscopicity.Freshly emitted BC particles usually exhibit chain-like agglomerates built up by a number of primary carbon spherules (Park et al., 2004), which are generally hydrophobic.
Condensation of organic and/or inorganic components lead to a collapse of the chainlike agglomerates, and in turn, a compact structure of BC particles (Slowik et al., 2007;Zhang et al., 2008).Changes in the morphology of BC particles affect their optical properties.Meanwhile, encasement by organic and/or inorganic coatings also increases the absorption of BC particles through the lensing effect (Shiraiwa et al., 2010;Peng et al., 2016).In addition, water-soluble coatings increase the hydrophilic ability of the BC particles (Zhang et al., 2008;McMeeking et al., 2011), indirectly affecting the radiative forcing by affecting the cloud processes.
Laboratory studies indicate that freshly emitted BC particles can be thickly coated within a few hours in the atmosphere (Pagels et al., 2009;Peng et al., 2016).Thus, many studies have focused on the optical properties and radiative forcing of thicklycoated BC particles (Jacobson, 2001;Khalizov et al., 2009;Liu et al., 2017).However, in situ measurements have shown that a great number of uncoated and/or thinly-coated BC particles exist in the ambient atmosphere, with a fraction even higher than that of the aged BC particles (Schwarz et al., 2008).In general, thickly-coated BC particles account for <50% of the BC-containing particles in urban areas based on single-particle soot photometer (SP2) measurements (Huang et al., 2012;Wang et al., 2014;Wu et al., 2016).The existence of a large fraction of uncoated BC particles is likely due to continuous emission from combustion processes such as vehicle exhaust.Therefore, studies on the radiative forcing of BC particles without thick coatings are also essential, especially in urban areas.First, the morphologies and sizes of these quasi-bare BC particles should be investigated, as they greatly impact the optical properties of the particles.
The morphology of agglomerate BC particles is often studied directly using transmission electron microscopy (TEM) (Park et al., 2004).However, the timeliness and representation of filter-based TEM measurements is challenged because only a small fraction of particles on the filter collected during a considerable period are investigated (Wentzel et al. 2003).By comparison, online tandem techniques can provide much more efficient and representative measurements of particle properties, including morphology (Park et al., 2008).For example, based on the measurements of an electrical low-pressure impactor (ELPI) or an aerosol particle mass analyzer (APM) in tandem with a differential mobility analyzer (DMA), the fractal dimension, a crucial representative aspect of the morphology of agglomerate BC particles, has been studied (Park et al., 2003;Maricq and Xu, 2004;Olfert et al., 2007;Rissler et al., 2014).The effective density and the dynamic shape factor of these agglomerates can also be obtained from tandem measurements.Generally, a condensation particle counter (CPC) connected to an APM is used to obtain the typical mass of the size-selected particles from the DMA, which is determined as the mass corresponding to the peak APM voltage that results in the maximum particle number concentration.There exist large uncertainties from CPC detection when the particle concentration is low.Thus, the DMA-APM measurement is usually limited to relatively small particles that are abundant in the atmosphere (Geller et al., 2006).The upper size limit of the measured particles using the DMA-APM system is generally not larger than 400 nm in mobility diameter (Park et al., 2003;Maricq and Xu, 2004;Olfert et al., 2007;Rissler et al., 2014).The SP2 developed on the basis of the laser-induced incandescence technique was widely used to study BC aerosol properties.It determines the BC mass from particle to particle with very high precision and accuracy and distinguishes the BC aggregates with or without thin coatings from the thickly-coated BC particles (Schwardz et al., 2006;Moteki and Kondo, 2007).In this study, we developed a system consisting of an SP2 connected in tandem with the DMA.This system was used to study the morphology of the BC aggregates from ~100 nm to 750 nm in mobility diameter.An advantage of this system is that the SP2 measures the mass of pure BC instead of the entire mass of the BC-containing particles that was determined using the APM or ELPI in previous studies, leading to a more accurate detection of the effective density of the BC aggregates.
Meanwhile, the size limitation for BC aggregate detection is expanded to 750 nm in mobility diameter, providing more comprehensive knowledge on the morphology of the BC aggregates in ambient air.Using this novel tandem system, the fractal dimension, effective density and dynamic shape factor of ambient BC aggregates were investigated, providing important insights for better assessment of the regional radiative effects of agglomerate BC particles.

Experimental setup
The tandem system comprises a size selection unit and a measurement section.As shown in Fig. 1, a sample of air was drawn through the size selection unit (a DMA, model 3081, TSI Inc.) to generate quasi-monodisperse particles with a certain electrical mobility diameter.The size-selected particles were delivered to the measurement section for analysis by various methods, including an SP2 (Droplet Measurement Technologies), a CPC (model 3776, TSI Inc.) and two microaethalometers (model AE51, Aethlab).The operational flow rates were set to 0.1, 0.3 and 0.15 LPM (STP) for the SP2, CPC and AE51, respectively.The sheath flow ratio was set to 3 LPM, resulting in a ratio of sheath to sample flow rate of 4.3:1 for the DMA.Particles in the range of 15-750 nm in mobility diameter can be selected.The flow rate for each instrument was calibrated using a soap film flowmeter (model Gilian Gilibrator-2 February 2018 (19 days in total).The site is located in an urban area of Beijing.It is a few hundred meters away from two main roads and thus may be significantly affected by vehicle emissions.More information about the measurement site is described in previous studies (e.g., Wu et al., 2016Wu et al., , 2017)).The ambient air was dried by passing through a nafion dryer (model MD-700, Perma Pure LLC) before entering the system.
The DMA was connected to an external computer on which a program was run to control the voltage of the DMA, i.e., the particle mobility diameter.Thirty-three mobility diameters were set in the program to cyclically control the particles selected by the DMA gradually increasing from 20 nm to 750 nm.Stepwise size selection repeated until the operator stopped the program.A short cycle with each of the 33 diameters lasting for 18 s and a long cycle with each size lasting for 36 s were set to alternately operate in this experiment (Fig. S2).The purpose of this setting was to identify the time difference between the size selection and the subsequent measurement, as described in the following sections.

Black carbon measurement
The individual particulate BC mass was measured by the SP2 according to the incandescence signal when the BC passed through the intense Nd:YAG intracavity continuous laser beam (Schwardz et al., 2006).If a BC particle is coated with nonrefractory components, the coating will evaporate before the BC core incandesces, leading to a time lag between the peaks of incandescence and scattering signals that are synchronously detected by the SP2 (Moteki and Kondo, 2007).According to the frequency of the time lag, there was a significant distinction between thickly-coated (i.e., internally mixed) BC particles and thinly-coated or uncoated (i.e., externally mixed) ones (Fig. S1).BC-containing particles with a time lag shorter than 2 μs were identified as externally mixed.The time delay threshold might vary slightly from one SP2 to another, e.g., Zhang et al. (2016) reported a short time lag of 1.6 μs.However, it should be constant for a given instrument.In previous measurements using the same SP2 employed in this study, the critical time lag was maintained at 2 μs regardless of the ambient conditions, such as the pollution level (Wu et al., 2016(Wu et al., , 2017)).Before the experiment, the incandescence signal was calibrated using DMA-selected monodisperse Aquadag particles.The effective density of Aquadag was reported by Gysel et al. (2011).Since the incandescence signal is more sensitive to Aquadag, the peak intensity of the incandescence signal was reduced by a factor of 25% when calculating the calibration coefficient (Laborde et al., 2012).The calibration was performed again after the experiment to ensure the stability of the measurement during the entire experiment.

Theoretical calculation of the morphology and effective density
The structure of the externally mixed BC (extBC), agglomerated by primary spherules with diameters of 20-60 nm (Alexander et al., 2008), can be characterized by a mathematical parameter, the fractal dimension (Df), which is approximately expressed as a power law relationship between the mass of the agglomerate particle (m) and its mobility diameter (dmob), expressed as where k is a constant.It should be noted that the approximation is applied only when the Df is greater than 2, the regime in which the extBC particles are generally located (Park et al., 2003).The Df value of a sphere is 3. Thus, a particle is closer and closer to a sphere if the Df increases gradually to 3.
The effective density (ρeff) of the extBC particles is calculated as the ratio of the BC mass (m) measured using the SP2 and its volume based on the mobility diameter selected by the DMA, expressed as Combining Eq. 1 and 2, ρeff can also be expressed as a function of dmob, The dynamic shape factor is also calculated to indicate the morphology of the extBC particles.It is derived from the ratio of the slip-corrected mass equivalent diameter (dme) and dmob, expressed as where dme is calculated from the BC mass (m) by assuming the BC particle to be a compact sphere with density 1.8 g cm -3 , and Cc is the Cunningham slip correction factor parameterized by particle diameter (d) where λ is the mean free path of the gas molecules, which is set to 65 nm in this study according to Zhang et al. (2016).The values of the three empirical parameters α, β and γ are 1.257, 0.4 and 1.1, respectively (Eq.9.34 at Page 407 in Seinfeld and Pandis, 2006).

Minimizing the multicharged particles
The typical mass of extBC at a given dmob was affected by multicharged particles, especially in the size range of 100-400 nm (Ning et al., 2013).If the simple average of the individual extBC masses at a given dmob was employed, the typical extBC mass would be overestimated, resulting in measurement bias in the morphology of the extBC particles, such as an overestimation in ρeff and an underestimation in χ.Thus, following the approach in the SP2 calibration process, Gaussian fitting was performed for the frequency of extBC mass at each dmob.The major peak mass was considered as the mass of singly charged extBC particles corresponding to the given dmob (Fig. 2).Due to the detection limit of the SP2 for small BC particles with masses lower than ~0.3 fg, only the frequency of the extBC mass with sizes larger than 125 nm in dmob could be fit by a Gaussian function.Thus, the typical extBC masses at sixteen dmob in the range of 140-750 nm were obtained, as shown in Fig. 2 and Fig. 3.

Fractal dimension of the ambient extBC
A power law relationship was applied to the dmob-determined extBC mass values, delivering the fractal dimension (Df) of the ambient extBC (Fig. 3).In the mobility diameter range of 140-750 nm, the fitted Df is 2.36, with one standard deviation of 0.04.
It is comparable with the Df values of diesel exhaust particles.For example, depending on the fuel type, engine type and load, the Df of diesel exhaust particles measured by the DMA-APM or DMA-ELPI system ranged between 2.22 and 2.84 (Park et al., 2003;Maricq and Xu, 2004;Olfert et al., 2007).This similarity indicates the dominant contribution of diesel exhausts to the extBC in our measurement site in urban Beijing.
A high fraction of organic and/or inorganic (e.g., sulfate) components in the diesel exhausts increased the Df values by filling in the gap of the chain-like agglomerates or coatings outside the primary spherules (Park et al.,2003;Olfert et al., 2007).Thus, the relatively low Df value observed in urban Beijing is likely related to the high fuel quality (e.g., low sulfur content) and more efficient combustion in the vehicle engines, which decrease the organic and/or inorganic fractions in diesel exhausts.The Df value for the ambient soot agglomerates measured with a DMA-APM system near a diesel truckdominated highway was 2.41 (Geller et al., 2006), very close to the value in our study.et al., 2016, 2017) but they can also lead to a more compact structure of extBC particles, and in turn, a higher Df value.BC particles thinly-coated by organic/inorganic components were somewhat considered as extBC using the SP2 time delay approach (Moteki and Kondo, 2007;Laborde et al., 2012).However, less mass condensed onto the BC primary particles would lead to more compaction of the extBC particles even when the dmob remained constant (Slowik et al., 2007).
Fig. 4 The diurnal variation of the fractal dimension (Df) of ambient extBC particles.
The error bars represent one standard deviation of the fitted Df.

Size-resolved effective densities of the ambient extBC
In contrast to the mass of extBC (m), the effective density of the extBC particles (ρeff) showed a significant decreasing trend as dmob increased from 140 nm to 750 nm (Fig. 5).The highest ρeff of 0.34 g cm -3 was observed in the dmob range of 140-160 nm, likely due to the fewest primary spherules of the BC aggregates at the smallest size.When dmob increased to 700 nm, ρeff decreased to as low as 0.12 g cm -3 , approximately onethird of that at 140-160 nm.The very low ρeff values agree well with the fractal-like nature of the extBC particles.diesel exhausts (Park et al., 2003) and near-traffic urban environments (Rissler et al., 2014) are also presented as blue triangles and red squares, respectively.
The ρeff values measured with the DMA-SP2 measurement are considerably lower than those of diesel exhaust particles measured by the DMA-APM or DMA-ELPI systems.Park et al. (2003) reported a decrease in ρeff of diesel exhaust particles under a 50% engine load from 0.95 g cm -3 to 0.32 g cm -3 as the mobility diameter increased from 50 nm to 300 nm (Fig. 5).The ρeff values presented in Park et al. (2003) are approximately 1.62, 1.34 and 1.23 times those in our study at ~150 nm, 220 and 300 nm in mobility diameter, respectively.Some previous studies on the ρeff of diesel exhaust particles also commonly showed a larger ρeff than that measured in this study (Maricq and Xu, 2004;Olfert et al., 2007).Nevertheless, the ρeff values reported in the literature also differ.For to 0.31 g cm -3 in the near-traffic urban environment as dmob increased from 50 nm to 350 nm (Fig. 5), similar to that of the freshly emitted diesel exhaust particles presented in Park et al. (2003).However, based on the same method, the ρeff values of the ambient BC aggregates that mostly originated from diesel exhausts (Geller et al., 2006) are largely different from those presented in Rissler et al. (2014), especially in the large particle size range.The ρeff at ~350 nm was 0.17 g cm -3 in Geller et al. ( 2006), approximately half of the value presented in Rissler et al. (2014).The reason for the discrepancy might be related to the large measurement uncertainties of the DMA-APM system for large particles, e.g., with sizes greater than 300 nm in dmob, since these large particles are less abundant in the ambient atmosphere (Geller et al., 2006).
The DMA-APM system directly measures the mass of particles selected at a given mobility diameter by the DMA.The organic and/or inorganic components filling in the gap or coating outside the primary spherules will increase the mass of the BC agglomerates, although their content might not be significant in fresh exhausts.Olfert et al. (2007) showed that sulfates and water accounted for ~2% of the mass of the total particulate matter under a low engine load (15%) and increased to ~30% under a high engine load (40%).Rissler et al. (2014) found that the volatile mass fraction of the soot aggregate was ~10% when heated to 300 °C.Thus, in practice, the DMA-APM system measured the ρeff of soot particles comprising chain-like BC aggregates and a few volatile components instead of pure BC.However, the SP2 measured the mass of pure BC merely through its incandescence in the laser beam, and the DMA-SP2 system determined the ρeff of extBC exclusive of those volatile components coupled in the BC particles.Thus, a lower ρeff was obtained, but this value was likely closer to the value of the pure extBC particles.Although the ρeff of extBC at small sizes (dmob<140 nm) cannot be determined due to the limitation of the DMA-SP2 system, we extended the ρeff of extBC to a large size range (300<dmob<750), which was barely investigated in previous studies.A continuous decrease in ρeff with increasing dmob was observed even in the large size range between 300 nm and 750 nm (Fig. 5).It is reasonable that the structure of the extBC particles is looser when the agglomerate chain built up by the primary spherules increases.

Dynamic shape factors of the ambient extBC
Due to the chain-like structures, the extBC particles generally have large dynamic shape factors (χ) with values in the range of 2.53 to 3.18, much larger than that of intBC with an average value of ~1.2 (Zhang et al., 2016).In general, the χ values are significantly higher for larger particle sizes.The extBC particles 700 nm in mobility diameter have a mean χ value of 3.18, approximately 1.26 times that for 160 nm particles.The larger particles have longer chains and looser structures, resulting in higher χ values.However, the χ values varied slightly, fluctuating in a narrow range between 2.53 to 2.64 in the size range of 140 nm to 350 nm.This is attributed to the large drag force for smaller particles.
Fig. 6 The dynamic shape factor of the extBC particles as a function of the mobility diameter.

Conclusion
The DMA-SP2 system was established to study the morphology of the ambient extBC particles.The quasi-monodisperse particles in the dmob range of 20-750 nm were stepwise selected using the DMA and then delivered to the SP2 for BC mass measurement and mixing state discrimination.The time difference between the size selection and the SP2 measurement was previously processed using the local peak approach.Gaussian fitting was performed to the extBC mass frequency distribution to minimize the effect of multicharged particles and obtain the typical extBC mass at a given dmob.On this basis, the Df was estimated to be 2.36±0.04 by fitting the derived extBC masses against dmob in the range of 140-750 nm, close to the Df values of diesel exhaust particles.A relatively lower Df value was observed in the morning traffic rush hours, implying a significant contribution of vehicle emissions to the agglomerate BC particles.A decrease in ρeff with increasing dmob was observed, with the ρeff value decreasing from 0.34 g cm -3 at 140-160 nm to 0.12 g cm -3 at 700 nm.The ρeff values derived using the DMA-SP2 measurement were generally lower than those based on the DMA-APM measurement.This was most likely due to the bias in the extBC mass determination between the SP2 and APM.The pure refractory BC mass determined using the SP2 in this study was generally lower than the mass of the BC aggregate, which comprises both refractory BC and nonrefractory organic and/or inorganic components.The large χ values varying in the range of 2.53 to 3.18 were derived at different dmob, further implying the high fractal structure of extBC particles.
Identifying the time difference between size selection and the SP2 measurement Atmos.Meas.Tech.Discuss., https://doi.org/10.5194/amt-2018-408Manuscript under review for journal Atmos.Meas.Tech.Discussion started: 10 December 2018 c Author(s) 2018.CC BY 4.0 License.There exists a considerable difference between the time recorded by the size selection program and that recorded by the SP2, due to the time cost of the particles transmitting from the DMA to the SP2, as well as the system clock difference between the computer on which the size selection program runs and that for the SP2 data acquisition.As shown in Fig.S2, the SP2 measurement is significantly later than the size selection.There are two methods to identify the time difference.The first method is to find the time difference between the local peak in the particle number concentration (including both scattering and incandescence) detected by the SP2 and the beginning of the corresponding size selection cycle.During the experiment, stepwise size selection was cyclically operated to produce quasi-monodisperse particles with sizes gradually increasing from 20 nm to 750 nm.Thus, at the beginning of each new cycle, the voltage of the DMA should first drop drastically from a high value to a low one to make the particle size decrease from 750 nm to 20 nm.As a result, some particles with sizes in the SP2 efficiently detectable range (~100-500 nm) would be measured during the descent period, producing a local peak in the number concentration.Since it takes only several seconds for the descent, identifying the occurrence time of the local peak position in the SP2 clock and the beginning time of the size selection in the external computer clock will provide the time difference of each cycle.This method was utilized in this study.The other method is to check the consistency of the number and/or mass size distributions between the short-duration cycle and long-duration cycle.Although the durations of each size in the short cycle and long cycle are different (18 s vs. 36 s), the time difference between size selection and the SP2 measurement should be uniform for adjacent short and long cycles.Setting an initial time difference and calculating the mean number and/or mass concentration of each particle size, the number and/or mass size distributions are obtained.Then, the correlation coefficients between the size distributions during short and long cycles are calculated.Changing the time difference gradually, we can obtain a set of correlation coefficients as a function of the time differences.The time difference resulting in a maximum correlation coefficient was considered as the difference between the size selection and the SP2 measurement.Atmos.Meas.Tech.Discuss., https://doi.org/10.5194/amt-2018-408Manuscript under review for journal Atmos.Meas.Tech.Discussion started: 10 December 2018 c Author(s) 2018.CC BY 4.0 License.

Fig. 2
Fig. 2 Frequency distribution of the mass of the extBC particles selected at different

Fig. 3
Fig.3The mass of extBC particles as a function of the mobility diameter (gray squares), Fig. 5 The effective density (ρeff) of the extBC particles as a function of mobility diameter (dmob) (gray cycles).The black line represents the power law fitting of ρeff against dmob.The variations of ρeff with dmob measured for the soot agglomerates from Atmos.Meas.Tech.Discuss., https://doi.org/10.5194/amt-2018-408Manuscript under review for journal Atmos.Meas.Tech.Discussion started: 10 December 2018 c Author(s) 2018.CC BY 4.0 License.example,Olfert et al. (2007) found that the ρeff of diesel exhaust particles coated with little sulfate and water content was ~0.4 g cm -3 at 299 nm, larger than the value of 0.32 g cm -3 inPark et al. (2003) and 0.26 g cm -3 in our study for the same mobility size.The ρeff of ambient soot aggregates also showed a similar decreasing trend with increasing dmob based on the DMA-APM system(Geller et al., 2006;Rissler et al., 2014).Rissler et al. (2014) showed a decrease in the average ρeff of BC aggregates from 0.94 g cm -3 Atmos.Meas.Tech.Discuss., https://doi.org/10.5194/amt-2018-408Manuscript under review for journal Atmos.Meas.Tech.Discussion started: 10 December 2018 c Author(s) 2018.CC BY 4.0 License.