Size-resolved refractive index of scattering aerosols in urban Beijing: A seasonal comparison

Abstract

The refractive index (RI) of atmospheric aerosols is a critical parameter in aerosol direct radiative forcing (DRF) calculation that exhibits considerable uncertainties owing partly to substantial spatiotemporal changes in the RI. A tandem system comprising a single-particle soot photometer connected to a differential mobility analyzer is developed. Two field campaigns are conducted to derive the size-resolved real part of the RI of scattering (i.e., BC-free) particles at 1064 nm (RRI_sp^1064nm) in urban Beijing. Measurements of the RRI_sp^1064nm within a mobility diameter (D_mob) range of approximately 200 nm to 400 nm in summer and winter clearly show seasonal variations in the RRI_sp^1064nm, including magnitude and size dependence. A decreasing RRI_sp^1064nm trend is observed in summer as the D_mob enlarges from 237 nm to 422 nm, whereas the RRI_sp^1064nm increases in winter as the D_mob grows. This finding likely indicates varying aerosol physicochemical properties with particle size in summer and winter. Size-resolved RRI_sp^1064nm values within the D_mob range are generally larger in summer than in winter, thereby resulting in a larger bulk RRI_sp^1064nm of 1.50 ± 0.01 in summer compared with winter (1.47 ± 0.02). The low RRI_sp^1064nm in winter is likely attributable to the high contribution of organic aerosols (OA), particularly less oxidized OA, which generally have lower RRI values than secondary inorganic aerosols. Pronounced low RRI_sp^1064nm values are observed during polluted episodes in winter, implying significant OA contribution. Photooxidation is likely an important regulator of the RRI_sp^1064nm in summer. This inference is supported by the fact that RRI_sp^1064nm values in the afternoon are larger than those at night, which corresponds well to diurnal variations in O_x (O_x=O₃+NO₂) concentrations. Results indicate that seasonal variations in the RRI_sp^1064nm should be considered carefully in research, such as in DRF assessments, aerosol property parameterization in chemical models, and aerosol optical property and composition remote sensing.

Copyright © 2021 American Association for Aerosol Research

EDITOR:

• Hans Moosmüller

1. Introduction

Aerosols are crucial factors that influence the energy budget of Earth’s systems by scattering and absorbing solar radiation and interacting with clouds (IPCC, 2013). The direct radiative forcing (DRF) of atmospheric aerosols is dependent predominantly on aerosol optical properties, including aerosol optical depth, single scattering albedo, and asymmetry factors (Seinfeld and Pandis 2006). Owing to heterogeneity in the spatial and temporal distribution of aerosol optical properties, considerable uncertainties abound in the assessment of aerosol DRF (IPCC, 2013). The refractive index (RI) is one of the most important parameters that determine aerosol optical properties. The RI is essential in aerosol DRF evaluation in aerosol-involved climate model simulations (Liao et al. 2004; Han et al. 2012; J. W. Li, Cheng, et al. 2019). Small variations in the RI can lead to non-negligible alterations in aerosol DRF (Zarzana, Cappa, and Tolbert 2014; Moise, Flores, and Rudich 2015; Valenzuela et al. 2018). The RI is also a key parameter in aerosol optical property and composition remote sensing from satellite- or ground-based observations (Dubovik and King 2000; Li et al. 2018).

The RI of an aerosol particle is determined largely by its chemical composition. The components comprising aerosol particles are generally recognized by their RI values. For example, the theoretical real part of the RI (RRI) of ammonia sulfate at a wavelength of approximately 1000 nm is 1.52. However, special cases exist in which the RI varies in specific components. For instance, secondary organic aerosols (SOA) have a wide RRI range (1.35 to 1.66 at various wavelengths in a range of 220–1200 nm; Moise, Flores, and Rudich 2015) owing to the complexity and mutability of their molecular composition. Meanwhile, atmospheric aerosols are a mixture of various internal or external species and thus differ from the ideal case of one pure component. A volume-weight mixing scheme is widely employed in aerosol-involved climate models to calculate the RI value of an internal mixture, which is used in the subsequent optical estimation (Liao et al. 2004; Han et al. 2012; J. W. Li, Cheng, et al. 2019). Sometimes, direct RI measurements cannot be reconstructed effectively with simultaneously measured chemical composition with known RI values for individual species (Zhao, Tan, et al. 2019). This finding implies that further development of RI measurement techniques is necessary. Specifically, the measurement of size-resolved RI may address this issue to a certain extent. The direct measurement of aerosol RI is necessary to validate the reliability of estimations from substitutive parameters and approaches. From another perspective, measured RI can provide insights into aerosol physicochemical properties, including effective density (ρ_eff) and chemical composition. A relationship exists between RRI and ρ_eff in monocomponent particles (Liu and Daum 2008), which was extended by Zhao, Tan, et al. (2019) to examine ambient aerosols, as follows: (1) $$\frac{{\mathit{RRI}}^2-1}{{\mathit{RRI}}^2+2}\propto {\rho }_{\mathit{eff}}.$$(1)

The RI can be measured directly using ellipsometry for the real part and ultraviolet-visible spectroscopy for the imaginary part (Wang and Rood 2008; Liu et al. 2015). The scattering and/or absorption intensity of airborne aerosol particles is determined by particle size distribution (PSD) and the RI; thus, a combination of PSD and aerosol scattering and/or absorption intensity measurements is used widely to obtain the bulk aerosol RI (Liu and Daum 2000; Hand and Kreidenweis 2002; Sorooshian et al. 2008; Cai, Montague, and Deshler 2011; Vratolis et al. 2018; Wu et al. 2018). The RI and PSD can also be retrieved simultaneously by applying a secondary optimization technique based on the measurement of a double integrating sphere system (Ren et al. 2015). In addition, the RI and PSD of ambient aerosols can be retrieved from solar spectral extinction and scattering measurements using an optimize retrieval algorithm (Dubovik and King 2000). The aerosol scattering phase function is moderately dependent on the RRI; thus, methods based on scattering intensity measurements at different angles are also utilized to retrieve the RRI (Espinosa et al. 2019; Eidhammer, Montague, and Deshler 2008; Barkey, Paulson, and Chung 2007). Owing to the development of single-particle measurement technology, aerosol RRI can be obtained using an aerosol time-of-flight mass spectrometer (Moffet and Prather 2005; Moffet et al. 2008) or single-particle mass spectrometry (Zhang et al. 2016). Moteki, Kondo, and Nakamura (2010) presented a method to infer RI of small nonspherical black carbon particles, based on a tandem system comprised by an aerosol particle mass analyzer (APM) and a single-particle soot photometer (SP2), which is developed based on single-particle measurement technology. These methods can generally measure the RI of bulk aerosol particles within a certain size range. However, the chemical composition and other properties (e.g., ρ_eff) of atmospheric aerosols are dependent on their size (M. Hu et al. 2012; W. W. Hu et al. 2016; W. Hu et al. 2017), which can result in potential variation in the RI with particle size. Therefore, investigations into the size dependence of the RI are important to improve aerosol DRF assessment accuracy and better understand size-resolved physicochemical properties.

In this study, a SP2 is combined with stepwise particle size selection using a differential mobility analyzer (DMA) to obtain the RRI of scattering (i.e., BC-free) particles (RRI_sp). The advantage of this DMA-SP2 tandem system is that it can resolve the dependence of the RRI_sp on aerosol size, which is valuable for analyzing aerosol density and chemical components as well as size dependence (Liu and Daum 2008). The in-depth experimental setup and data processing of this tandem system are introduced in one of our previous studies (Wu et al. 2019), in which the morphology and effective density of externally mixed BC particles in the atmosphere are revealed. A similar experimental design was also presented by Zhao, Zhao, et al. (2019) and Zhao, Tan, et al. (2019). Compared with previous studies focusing on either a method description (Zhao, Zhao, et al. 2019) or summer RRI_sp estimation (Zhao, Tan, et al. 2019), the current study deploys the DMA-SP2 tandem system in two field campaigns (in summer and winter) during which aerosol physicochemical properties and sources differ significantly (Sun et al. 2015; W. W. Hu et al. 2016; W. Hu et al. 2017; Zhou et al. 2019). Seasonal discrepancies in the RRI_sp, including magnitude and size dependence, are obtained for the first time in urban Beijing. In the following sections, first, we provide a detailed description of our experimental setup and field measurements. Next, the retrieved RRI_sp values in summer and winter are presented, and the potential mechanisms leading to temporal variations in the RRI_sp are discussed in detail. Finally, the impact of the RRI_sp on coating thickness retrieval and light absorption enhancement in SP2-involved studies is determined. This study broadens knowledge on the size-resolved RRI of ambient aerosols, which is meaningful for model examinations and remote sensing retrievals.

2. Methodology

2.1. Tandem DMA-SP2/condensation particle counter (CPC) system

The tandem DMA-SP2/CPC system consists of a TSI model 3081 DMA, a TSI model 3776 CPC (TSI Inc., Shoreview, MN, USA), and a version C SP2 (Droplet Measurement Technologies, Boulder, CO, USA). This setup is built to obtain size-resolved aerosol physical properties, such as the morphology and effective density of externally mixed BC particles (Wu et al. 2019), BC mixing states, and the RRI_sp. Briefly, the DMA selects particles with a given mobility (a narrow mobility range owing to the transfer function of the DMA). Size-resolved particles are delivered to the CPC for counting and to the parallel SP2 for particle scattering intensity and refractory BC (rBC) mass determination (Figure S1 in the online supplementary information). The selection of particles with different mobilities is achieved by altering the voltage of the DMA by controlling a TSI model 3080 electrostatic classifier. A TSI model 3087 neutralizer is installed in front of the DMA to charge the particles according to a known charge probability before size selection. The electrostatic classifier communicates with an external computer through a series cable, on which a self-compiled size selection program is run to alter the voltage of the DMA using a stepwise mode. Essential settings, including stepwise voltage, the duration of each voltage, and the operated sheath flow of the DMA, are compiled in the size selection program before the operator alters each measurement based on specific needs.

Figure 1. Comparisons of scattering peak heights detected by the SP2 for BC-free particles at prescribed mobility diameters with those theoretically calculated on the basis of Mie theory by assuming an RI of 1.50-0i; blue squares represent results in the summer campaign, and red dots represent results in the winter campaign.

2.2. Aerosol scattering cross-section determination

Particles with a given mobility selected by the DMA are drawn into the SP2 for single-particle quantitative measurements. The SP2 determines the rBC mass particle by particle using a laser-induced incandescence technique (Schwarz et al. 2006). Scattering signals at the laser wavelength of 1064 nm are synchronously detected, which provides an advantage for evaluating the mixing state of rBC-containing particles by comparing the incandescence signals (Gao et al. 2007; Moteki and Kondo 2007). Owing to the light absorption and consequential heating of an rBC-containing particle when it passes through the intensive and Gaussian-distributed laser beam in the SP2 measurement cavity, the coating of the rBC core evaporates before the onset of incandescence, thereby resulting in a gradual decrease in the particle scattering cross-section (σ_sp). Thus, the actual σ_sp of the rBC-containing particle cannot be obtained directly from the SP2 measurement unless lead-edge-only (LEO) fitting is employed for the scattering signal (Gao et al. 2007). However, the σ_sp is stable for a BC-free particle owing to the absence of evaporation, which is proportional to the induced scattering peak height (H_sp) when the particle is passing through the laser beam. In practice, the SP2 only detects scattering signals over a certain solid angle (ΔΩ) of scattered light collection (i.e., 15°–75° and 105°–165° in the plane parallel to the laser beam and 60°–120° in the plane perpendicular to the laser beam; Moteki and Kondo 2007). Thus, to be exact, the H_sp is proportional to the differential σ_sp (dσ_sp/dΩ) integrated over the solid angle (denoted as Δσ_sp), expressed as (Moteki and Kondo 2008): (2) $$H_{sp}=C·{\int }_{\Delta \Omega }\frac{{\sigma }_{sp}}{\Omega }\Omega =C{·\Delta \sigma }_{sp},$$(2) where C is a scale factor which is dependent on the intensity of the laser beam. In the case of a stable laser intensity, C is a constant and can be obtained through calibration using pure scattering particles with known sizes, morphologies, and RRI, such as the polystyrene latex (PSL) microspheres with a 269 nm diameter used in the current study. PSL spheres have an RRI of 1.59 and a negligible RI imaginary part; thus, the Δσ_sp value of 269 nm PSL microspheres can be obtained by using Mie calculation. In practice, 269 nm PSL microspheres are generated by an aerosol generator (model AG-100, Droplet Measurement Technologies, Boulder, CO, USA) and delivered to the SP2 for scattering calibration. A diffusing dryer is installed in front of the SP2 to ensure the measurement of dry aerosols. To reduce the influence of the impurities of the PSL microspheres, they are further size selected by passing them through the DMA, given a fixed mobility diameter (D_mob) of 269 nm, before entering the SP2.

Owing to the detection efficiency of the SP2, only particles within a certain size range can be characterized effectively by their scattering signals (i.e., low gain H_sp within a range of 300–60,000 A/D counts for the SP2 used in the current study). Particles beyond the effective size range are either too small to be distinguished from instrument noise or too large to be detected accurately. Generally, a commercial SP2 has an effective scattering detection capability within a 200 nm to 400 nm diameter range for BC-free particles (Zhang et al. 2018), which somewhat depends on laser intensity (see Section 2.4). Gaussian or LEO fitting can be used to estimate the scattering peak intensity of particles larger than the upper limit of the scattering detection range (approximately 400 nm). However, additional uncertainties could arise from the fitting process. Moreover, in the atmosphere, particles larger than ∼400 nm generally have much lower number concentrations than smaller particles. This fact may result in insufficient data for further statistical analysis and fitting in the subsequent data processing. Thus, in the current study, we focus on particles within the effective scattering detection range of ∼200–400 nm.

2.3. Ambient measurements

The DMA-SP2/CPC system was deployed to obtain ambient measurements at an urban site in Beijing from 27 July to 23 August 2017 for the summer campaign and from 23 January to 10 February 2018 for the winter campaign. The observation site (39°58′N, 116°22′E) encompasses a residential community and a few hundred meters beyond several main roads and thus is affected by local vehicle emissions. Air quality in Beijing is influenced substantially by neighboring regions with intensive heavy industry emissions. Thus, the site is likely affected by regionally transported emissions, particularly during polluted periods, which are carried by moderate southerly winds (Ma et al. 2017).

In the summer campaign, we set the size selection program as 9 stepwise voltages to control the DMA to select particles with D_mob gradually increasing from 75 nm to 750 nm on a logarithmic scale. The indicated D_mob values correspond to the mobility diameters of single-charged particles at the prescribed DMA voltages. Particles with multiple charges have larger mobility diameters at each voltage compared with those with a single charge. If not specified, the mobility diameters hereinafter represent diameters corresponding to the mobility of single-charged particles. Each voltage is kept invariant for 48 s. A periodic cycle of the 9 stepwise voltages is operated in the measurement. In addition, 48 s for system stability are appended at the end of the last selected size (i.e., 750 nm), thereby leading to a total time cost of 8 min per cycle. For technical purposes, to obtain the time bias between the size selection and subsequent measurements, which results from the computer system time difference and consumption time of particles transported from the DMA to the SP2 and CPC (Wu et al. 2019), an auxiliary cycle is set to operate between the main cycles. The voltage (in turn, D_mob) setting of the auxiliary cycle is consistent with that of the main cycle, but the duration of each voltage and the system stable time decrease to 12 s. The total time of the auxiliary cycle is 2 min. The main cycle and auxiliary cycle operate alternately. Figure S2 shows an example of a variation in particle number concentration detected by the CPC and SP2 during an auxiliary cycle and adjacent main cycle in the summer campaign. The total number concentrations of SP2-detectable rBC-containing particles and BC-free particles are shown.

Figure 2. (a) Campaign-averaged size-resolved RRI_sp^1064nm as a function of D_mob, with shadow areas representing corresponding standard deviations; (b) boxplot of bulk RRI_sp^1064nm with upper and lower whisks representing 90% and 10% percentiles, respectively; upper and lower ends of the box representing 75% and 25% percentiles, respectively; the horizontal line in the box representing the median; and the square representing the mean value; blue represents results in summer, and red represents results in winter.

A similar size selection program setting is adopted for the winter campaign, except for the use of a fine voltage step for the DMA. The fine voltage setting is for the purpose of particle number size distribution inversion from CPC measurements at discrete D_mob. We set 33 stepwise voltages corresponding to single-charged particles within the 20–750 nm D_mob range on a logarithmic scale, each with a duration of 36 s, followed by 12 s for system stabilization, thereby resulting in a total duration of 20 min per cycle. An auxiliary cycle with a half-time duration of 10 min is set to operate between two major cycles. Correspondingly, the duration of each voltage in the auxiliary cycle decreases to 18 s, and the system stable time decreases to 6 s. An example of a variation in particle number concentrations during an auxiliary cycle and adjacent main cycle in the winter campaign is shown in Figure S3. Both in the summer and winter campaigns, only data from the main cycles are used in the following RRI_sp determination.

Figure 3. Temporal variations in (a) hourly retrieved bulk RRI_sp^1064nm and RRI of column-integrated aerosols over an urban site (i.e., CAMS) in Beijing (RRI_aeronet^1020nm); (b) PM_2.5 and BC mass concentrations; (c) O_x (O_x=NO₂+O₃) and CO concentrations; (d) ambient T and RH at a height of 100 m; and (c) wind speed (V) at 100 m color scaled by wind direction.

The sheath flow rate of the DMA is set to 3 L min⁻¹ (hereinafter STP) in both campaigns, but the total sample flow through the DMA is 0.55 L min⁻¹ in summer, which is slightly lower than the flow rate of 0.70 L min⁻¹ in winter. Flow rate discrepancies arise from different mounted instruments following the DMA. In addition to the SP2 and CPC, which operate with 0.10 L min⁻¹ and 0.30 L min⁻¹ flow rates, respectively, two microaethalometers (model AE51, AethLabs, San Francisco, CA, USA) operating with a 0.15 L min⁻¹ flow rate are measured in parallel during the winter campaign (Figure S1). However, only one microaethalometer operating with a 0.15 L min⁻¹ flow rate is deployed in the summer campaign. A nozzle with a diameter of 0.071 cm is installed in the impactor to prevent large particles from entering the DMA. Affected by the sample flow rate, particles with an aerodynamic diameter larger than 908 nm and 797 nm in the summer and winter campaigns, respectively, are blocked by the impactor. A model MD-700-12F-3 Nafion dryer (Perma Pure LLC, Toms River, NJ, USA) is deployed to dry the sampling aerosols before they enter the DMA-SP2/CPC system.

The SP2 operates under satisfactory conditions, with stable laser intensity throughout both campaigns. H_sp (low gain H_sp hereinafter if not specified) of 269 nm PSL microspheres is approximately 5900 A/D counts in the summer campaign, and its variation is within 3% based on two calibrations performed before and after the campaign. A lower H_sp of 269 nm PSL microspheres (∼3600 A/D counts) is observed in the winter campaign, largely attributed to the decreasing SP2 laser intensity. The decrease in laser intensity is illustrated by the considerable decrease in the YAG power recorded by the SP2, from ∼4.0 V in summer to ∼3.0 V in winter. However, only a < 1% difference is observed in H_sp of 269 nm PSL microspheres obtained from the calibrations before and after the winter campaign (3580 vs. 3600 A/D counts), thereby indicating that laser intensity remains stable throughout the winter campaign, even its absolute value decreases compared with that in summer. The H_sp value of 5900 A/D counts combined with the theoretically calculated Δσ_sp of 269 nm PSL spheres is used in the summer campaign to obtain the C value based on Equation (2)(2) $$H_{sp}=C·{\int }_{\Delta \Omega }\frac{{\sigma }_{sp}}{\Omega }\Omega =C{·\Delta \sigma }_{sp},$$(2) , while the H_sp value of 3600 A/D counts is used in the winter campaign.

PM_2.5 mass concentrations are recorded using a synchronized hybrid ambient real-time particulate monitor (Model SHARP 5030, Thermo Scientific, Franklin, MA, USA), with an average 1-hour time resolution, for both campaigns. This PM_2.5 monitor is based on the principles of aerosol light scattering (nephelometry) and beta attenuation to measure precise and accurate ambient aerosol concentrations. The instrument has an hourly precision of ±2.0 μg m⁻³ for measured mass loading less than 80 μg m⁻³ and ±5.0 μg m⁻³ for measurements greater than 80 μg m⁻³ as the manufacturer states. The accuracy is ±5% compared to 24 h Federal Reference Method and the minimum detectable limit for the mass transducer is < 0.5 µg. Ambient air is sampled at a flow rate of 16.7 L min⁻¹ and passed through a PM_2.5 cyclone inlet to remove particles with diameters >2.5 μm before entering the instrument. A model AE31 aethalometer (Magee Scientific, Berkeley, CA, USA) is used to record hourly BC mass concentrations. Hourly O3, NO₂ and CO concentrations are obtained from the observations of the China National Environmental Monitoring Center at a site (Olympic center site) approximately 4 km away from our observation site. Meteorological variables, including air temperature (T), relative humidity (RH), and horizontal wind velocity and direction, are measured from 10 levels of a 325 m-high metrological tower approximately 50 m away from the aerosol measurement site. Hourly means at a 100-meter height are used in the analysis. In addition, the RI values (at 1020 nm) of column-integrated aerosols retrieved from ground-based sun photometer measurements at an urban site in Beijing (Chinese Academy of Meteorological Sciences [CAMS]) several kilometers away from our observation site are obtained from the dataset website of the Aerosol Robotic Network (AERONET, https://aeronet.gsfc.nasa.gov/) to compare the RRI_sp obtained in this study. The uncertainty in AERONET RRI is expected to be <0.04.

2.4. Retrieval of size-resolved RRI_sp

A local maximum method is utilized to identify the time bias between the size selection and subsequent measurements of the SP2 and CPC (Wu et al. 2019). Figures S2 and S3 show that obvious time lags are observed in the SP2 and CPC measurements, which could result in inaccuracies in the size-resolved analysis if not corrected. After the time bias is corrected, particle information at each prescribed mobility size, including H_sp, rBC mass, and mixing state, is well characterized by the SP2, and the total aerosol number concentrations are determined by the CPC. In practice, the data collected in the first and last 3 s of each size step are eliminated from the analysis to reduce measurement errors arising from the size alteration.

Figures S4 and S5 show that H_sp varies within a considerable range for each D_mob, owing to the effect of DMA transfer function. Multicharged particles also lead to evidently minor modes of H_sp probability distribution, particularly at sizes with a D_mob < 300 nm. The major mode of H_sp probability distribution, which is induced by singly charged particles, is fitted by a log-normal distribution function. The geometric mean value of H_sp of the log-normal distribution is considered the typical H_sp (H_sp0) corresponding to the given D_mob (Zhao, Zhao, et al. 2019; Zhao, Tan, et al. 2019). In practice, to ensure adequate data for the fitting, BC-free particles in a 3-hour moving time window are used to establish the probability distribution, as shown in Figures S4 and S5. Hourly Hsp0 values at each prescribed mobility are obtained. The campaign average H_sp0 values are shown in Figure 1, which gradually increase as D_mob enlarges. Notably, owing to a lower laser intensity, the H_sp0 value for a given D_mob is evidently lower in winter than summer. Furthermore, because the detectable H_sp is limited within the range of 300–60,000 A/D counts, H_sp0 values within the 178–422 nm D_mob range in summer whereas within the 200–450 nm D_mob range in winter are obtained effectively. Theoretical H_sp values calculated on the basis of Mie theory and Equation (2)(2) $$H_{sp}=C·{\int }_{\Delta \Omega }\frac{{\sigma }_{sp}}{\Omega }\Omega =C{·\Delta \sigma }_{sp},$$(2) at different D_mob by using an initially guessed RRI_sp value of 1.50 and a negligible RI imaginary part (Zhang et al. 2018) are comparably shown in Figure 1. In general, the theoretical H_sp agrees well with the measured H_sp0, indicating the practicability of the particle size estimation from the scattering measurements of the SP2 combined with Mie calculation. However, systematic overestimations in the theoretical H_sp are observed, particularly in winter (red circles in Figure 1). The scattering intensity increases gradually with increasing RRI_sp (Zhao, Zhao, et al. 2019), thus the overestimated theoretical H_sp implies that a high RRI_sp (1.50) is likely adopted in the calculation. Thus, we adjust the RRI_sp value to constrain the theoretical H_sp consistent with the measured H_sp0 at each prescribed D_mob. Subsequently, size-resolved RRI_sp values are obtained hourly within the 178–422 nm D_mob range in summer and within the 200–450 nm range in winter. Notably, morphology is also an essential factor influencing particle scattering intensity beyond size and RI. However, a recent study showed that the dynamic shape factors of BC-free (i.e., they called non-rBC) particles were close to 1 within the 300–500 nm aerodynamic diameter (D_ae) (equivalent to ∼240–400 nm physical diameter by assuming a density of 1.5 g cm⁻³; Cross et al. 2007), indicating a spherical structure of BC-free particles in urban Beijing (Liu et al. 2020). Thus, the morphology effect on particle scattering calculation is neglected in the current study. Furthermore, because the H_sp0 values are used in the retrieval, the scattering properties of the measured particle ensembles are studied instead of the particle-resolved properties. That is to say, the retrieved RRI_sp values represent an effective value of the measured particle ensembles for each given D_mob.

Figure 4. Bulk RRI_sp^1064nm as function of PM_2.5 mass concentrations during summer (blue) and winter (red), with shadow areas representing corresponding standard deviations.

Figure 5. Diurnal variations in (a) bulk RRI_sp^1064nm and (b) O_x (O_x=NO₂+O₃) concentrations; the upper and lower whisks of each box represent 90% and 10% percentiles at each hour, respectively, and the upper and lower ends of each box represent 75% and 25% percentiles, respectively; the horizontal line in the box represents the median, and the dot represents the mean value; blue represents results in the summer campaign, and red represents results in the winter campaign; diurnal variations in bulk RRI_sp^1064nm on clean days in winter with maximum hourly PM2_.5 concentrations lower than 20 μg m⁻³ are shown as gray diamonds in panel (a).

RRI_sp at the SP2 laser wavelength 1064 nm (RRI_sp^1064nm) is retrieved in this study, which may be slightly discrepant to that at a short wavelength in the near-ultraviolet and visible spectra, particularly for particles dominated by organic matter. The clear spectral dependence of the RRI of SOA is observed, with the RRI increasing at short wavelengths (Moise, Flores, and Rudich 2015). The spectral dependence of the RRI of inorganic components, such as secondary inorganic aerosol (SIA; including sulfate, nitrate, and ammonium), is not obvious in the near-ultraviolet and visible spectra. Owing to the slight variation in laser intensity (<3%) in each campaign, the uncertainty in the retrieved RRI_sp^1064nm is estimated to be approximately 0.01 at a prescribed mobility based on Mie theory. Furthermore, because the SP2 is only efficient to detect rBC with mass larger than its lower detection limit (∼0.2 fg), the termed ‘BC-free’ particles in this study are likely to also have small BC fractions, thus have non-ignorable RI imaginary parts. According to the incandescence calibration factor and the criterion of incandescence peak heigh (>500 A/D at high gain) used to discriminate rBC-containing particles, the upper limit rBC mass in individual ‘BC-free’ particles is 0.19 fg. We assume all the termed ‘BC-free’ particles have rBC mass of 0.19 fg to examine the utmost effect of possibly existed RI imaginary parts on the retrieved RRI_sp^1064nm. A RI imaginary part value of 1.26 (at 1064 nm; Moteki, Kondo, and Nakamura 2010) and a density of 1.8 g cm⁻³ (Bond and Bergstrom 2006) is assumed for rBC in the calculation. The RI imaginary part value of a whole particle containing 0.19 fg rBC is calculated using the volume-weighted mixing scheme assuming a zero RI imaginary part value for remaining non-BC components of the particle. The absorption of brown carbon at the IR wavelength of 1064 nm is neglected. The retrieved RRI_sp^1064nm values are almost the same as those originally derived by assuming pure scattering particles without RI imaginary parts (Figure S6). The absolute biases of the retrieved RRI_sp^1064nm due to RI imaginary parts are within ±0.002 for all D_mob, which is much smaller compared to the uncertainty (∼0.01) resulted from the slight fluctuation of laser intensity. Considering the both effects of laser intensity fluctuation and possibly existed RI imaginary part, the retrieved RRI_sp^1064nm for ‘BC-free’ particle ensembles at each prescribed D_mob is expected to have a total uncertainty within 1%.

Figure 6. Size-resolved RRI_sp^1064nm as a function of the D_mob at local times of 14:00–16:00, 05:00–07:00, and 18:00–20:00; blue represents results in summer, and red represents results winter.

2.5. Estimation of bulk RRI_sp

Based on the size-resolved RRI_sp^1064nm, the bulk RRI_sp^1064nm within the detectable size range (i.e., 178–422 nm in summer and 200–450 nm in winter) is further estimated using a volume-weighting mixing approach. Volume size distribution is inverted from the CPC-measured stepwise particle number concentration beforehand. The inversion approach for the particle number size distribution (dN/dlogD_mob) is based on Hagen and Alofs (1983). The charging probability of the neutralizer and transfer function of the DMA are calculated from the parameters and equations provided in the instruction manual of the TSI model 3080 electrostatic classifier. An assumption of the size-independent number fraction of BC-free particles in the total number of particles is employed in the estimation of the bulk RRI_sp^1064nm. The estimation of the bulk RRI_sp^1064nm is expressed as (3) $${\mathit{RRI}}_{\mathit{bulk}}=\sum {\mathit{RRI}}_i·{\left(\frac{dV}{\mathit{dlog}{D}_{\mathit{mob}}}\right)}_i·\Delta \mathit{log}{D}_{\mathit{mob},i}/\sum {\left(\frac{dV}{\mathit{dlog}{D}_{\mathit{mob}}}\right)}_i·\Delta \mathit{log}{D}_{\mathit{mob},i},$$(3) where the RRI and dV/dlogD_mob with subscript i represent the size-resolved RRI_sp^1064nm and volume size distribution at a specific D_mob, respectively. In addition, ΔlogD_mob is the interval between two consecutive D_mob set on a logarithmic scale. The value of dV/dlogD_mob is calculated from the dN/dlogD_mob using a spherical assumption. The size-resolved RRI_sp^1064nm and dV/dlogD_mob at four D_mob values of 178 nm, 237 nm, 316 nm, and 422 nm are deployed in the bulk RRI_sp^1064nm calculation in the summer campaign, and eight D_mob values of 200 nm, 225 nm, 250 nm, 280 nm, 315 nm, 350 nm, 400 nm, and 450 nm are used in the winter campaign. The mass fraction of particles located within the size range, which is employed to calculate the bulk RRI_sp^1064nm, accounts for approximately 22 ± 9% and 33 ± 16% of the PM_2.5 mass concentrations in summer and winter, respectively, by assuming an average particle density of 1.5 g cm⁻³ (Cross et al. 2007). The low integrated mass in summer is likely attributable to coarse size resolution and high mass fraction beyond the concerned PM_2.5 size range (i.e., 178–422 nm).

3. Results and discussion

3.1. Seasonal variation in size-resolved RRI_sp

Seasonal average RRI_sp^1064nm varying with D_mob is shown in Figure 2. Size-resolved RRI_sp^1064nm values are generally high in summer, thereby leading to a higher bulk RRI_sp^1064nm, with a mean ± standard deviation of 1.50 ± 0.01, compared with that with a mean rate of 1.47 ± 0.02 in winter. Seasonal discrepancies in the RRI_sp^1064nm indicate the pronounced differences in aerosol chemical composition in summer and winter. SIA and organic aerosols (OA) are the predominant components of ambient aerosols in Beijing, which contribute more than 80% of the total PM₁ mass throughout the year (Sun et al. 2015; W. Hu et al. 2017). Thus, the RRI_sp^1064nm in this study is expected to be determined predominantly by SIA and OA, as only submicron scattering particles are measured. The RRI values of the main components of SIA are well known, 1.52 for ammonium sulfate (at approximately 1000 nm) and a high value of 1.55 for ammonium nitrate, which are generally higher than the observed RRI_sp^1064nm values in summer (1.50) and winter (1.47). This finding implies that OA should have a low RRI on average, such that their mixture with SIA has an RRI_sp^1064nm as low as the observed values. The possible influence of particle shape (nonsphere vs. sphere) on the RRI_sp^1064nm is neglected because BC-free particles within the studied diameter range are generally spherical in urban Beijing (Liu et al. 2020). Owing to the differing molecular composition, structure and mutability of OA in the atmosphere, obtaining a definite RRI value for OA is difficult. A wide range of RRIs for SOA (1.35–1.66 at various wavelengths in a range of 220–1200 nm) are given in the literature review of Moise, Flores, and Rudich (2015). Nevertheless, compared with the well-known RRI values for SIA, attributing the low RRI_sp^1064nm in winter to an increase in the contribution of OA submicron particles would be reasonable. Although simultaneous measurements of aerosol chemical composition are absent in this study, previous studies show a significant increase in the mass fraction of OA, particularly less oxidized OA, in PM_2.5 or PM₁ in winter compared with in summer owing to the increase in coal combustion and/or possible biomass burning for residential heating (Zhang et al. 2013; Sun et al. 2015; W. W. Hu et al. 2016; Zhou et al. 2019). Seasonal standard deviation in the bulk RRI_sp^1064nm (0.02) is higher in winter than in summer (0.01), thereby indicating a larger variation in aerosol composition owing to more diverse emissions sources in winter.

In addition to seasonal discrepancies in the magnitude of the RRI_sp^1064nm, a clear difference is observed in the size dependence of the RRI_sp^1064nm in summer and winter. By neglecting the RRI_sp^1064nm retrieved at the low limit diameter (i.e., 178 nm in summer and 200 nm in winter), which may have considerable uncertainty owing to an imperfect lognormal distribution fitting (Figures S4 and S5), we observe a decreasing RRI_sp^1064nm trend with an increasing D_mob in summer and a slightly increasing RRI_sp^1064nm trend in winter (Figure 2). Despite the ∼1% uncertainty in the retrieved RRI_sp^1064nm, the size dependence of the RRI_sp^1064nm seems reliable. The maximum discrepancy in the size-resolved RRI_sp^1064nm is approximately 0.03 both in summer (1.50 ± 0.01 at 237 nm vs. 1.47 ± 0.01 at 422 nm) and winter (1.46 ± 0.03 at 225 nm vs. 1.49 ± 0.02 at 450 nm). The dependence of RRI_sp^1064nm on D_mob is likely related to the variation in chemical composition with particle size. Although synchronous measurements of size-resolved aerosol chemical composition are absent, previous studies have showed a relatively stable mass ratio of SIA to OA throughout a D_ae range of ∼100–2000 nm in winter (W. W. Hu et al. 2016; W. Hu et al. 2017). The slightly increasing RRI_sp^1064nm with D_mob in winter is likely related to more aged (i.e., oxidized) OA in large particles. An increase in the RRI of SOA from a certain precursor (e.g., toluene, α-pinene, or limonene) is observed as diameter changes from ∼180 nm to 360 nm (Moise, Flores, and Rudich 2015). However, the size dependence of RRI_sp^1064nm cannot be simply interpreted by the chemical composition variation with particle size in summer. An evident increase in SIA mass fraction whereas decrease in OA mass fraction with increasing particle size is generally observed in Beijing in summer (W. W. Hu et al. 2016; W. Hu et al. 2017). Considering the generally higher RRI of SIA than OA, an increasing RRI_sp^1064nm with D_mob is expected according to the volume-weighting scheme, which contradicts to the observed decrease in RRI_sp^1064nm as particle size grows in summer. The observed size dependence of RRI_sp^1064nm in summer in this study is also contrary to that obtained at a suburban site in East China, where the increase in RRI_sp^1064nm with increasing particle size was attributed to ρ_eff increase (Zhao, Tan, et al. 2019). Thus, we suspect that our observed decrease in RRI_sp^1064nm with particle size in summer is likely related to a decrease in ρ_eff. More internal voids may exist in larger particles, resulting in lower ρ_eff and, in turn, lower RRI_sp^1064nm. More evidences relevant to size-resolved chemical composition and ρ_eff are necessary in future studies. A slight increase in ρ_eff with increasing particle size in Beijing was observed in winter (M. Hu et al. 2012), corresponding to the increasing RRI_sp^1064nm with D_mob.

3.2. Variation in RRI_sp with pollution level

Temporal variations in the RRI_sp^1064nm at a high time resolution of 1 h are presented in Figure 3a. The hourly variations in the size-resolved RRI_sp^1064nm are generally similar and consistent with those in the bulk RRI_sp^1064nm (Figure S7); thus, only the hourly bulk RRI_sp^1064nm is presented. A more dramatic fluctuation in the hourly RRIsp1064nm is observed in winter than in summer, corresponding well to the large standard deviation of the seasonal mean. Generally, wintertime pollution episodes with high PM_2.5 concentrations are associated with a dramatic decrease in the RRI_sp^1064nm (Figure 3b). The bulk RRI_sp^1064nm values averaged in different PM_2.5 segments are presented in Figure 4. An evident decrease in RRI_sp^1064nm is observed in winter as PM_2.5 mass concentration increases from <15 μg m⁻³ to >75 μg m⁻³. During the extremely clean period (with PM_2.5<15 μg m⁻³), the bulk RRI_sp^1064nm has the largest mean of 1.48 ± 0.01. The lowest RRI_sp^1064nm with a mean of 1.45 ± 0.04 is observed during polluted period (with PM_2.5>75 μg m⁻³). By contrast, the RRI_sp^1064nm varies slightly with pollution level in summer, with means of 1.49–1.50 throughout different PM_2.5 segments, indicating relatively stable aerosol sources. The decreasing RRI_sp^1064nm as PM_2.5 increases in winter is likely related to the increasing contribution of less oxidized OA, the RRI of which is generally lower than SIA. Northerly gales prevail in extremely period in winter (Figure 3e), bringing continental background aerosols which are generally aged and thus have high RRI_sp^1064nm values. With the aggravation of air pollution, the contribution of local and/or regional emissions to PM_2.5 is likely to increase under the calm/weak southerly conditions (Figure 3e), thereby increasing the fraction of primary and/or less aged aerosols. This is likely to contribute to the evident decrease in RRI_sp^1064nm. Although numerous previous studies show that an increased SIA contribution is generally observed during heavily polluted periods (e.g., PM_2.5>150 μg m⁻³) in winter, this increase is typically accompanied by high ambient RH, which favors the formation of SIA via a heterogeneous reaction (Wang et al. 2016; Ma et al. 2017; H. Li, Cheng, et al. 2019). However, PM_2.5 pollution is generally light during the studied winter period, with the max hourly PM2_.5 concentration of 131.5 μg m⁻³ and a campaign average of 23.5 ± 27.5 μg m⁻³ (Figure 3b). Furthermore, wintertime RH is generally less than 60%, even during the polluted episodes in this study (Figure 3d), which is unfavorable to the formation of SIA. By comparing aerosol chemical composition during polluted periods under dry and humid conditions, W. W. Hu et al. (2016) found that OA dominates PM₁ during the former period, whereas the contribution of SIA increases remarkably during the latter period. Thus, an increase in the contribution of OA (particularly less oxidized OA) is expected under dry conditions, thereby resulting in low RRI_sp^1064nm values. Notably, large standard deviations are observed when pollution elevates (Figure 4), indicating more variant emission sources and complicated chemical composition during polluted periods in winter. The decrease in the RRI_sp^1064nm during polluted periods in winter can also be reproduced by the low RRI (at 1020 nm) of the column-integrated aerosols over Beijing obtained from AERONET (denoted as RRI_aeronet^1020nm), though RRI_aeronet^1020nm values are generally high (Figure 3a). Differences in the magnitude of the RRI_sp^1064nm and RRI_aeronet^1020nm may result from differences in physicochemical properties between near-surface and column-integrated aerosols as well as from discrepancies arising from the retrieval approaches, which are beyond the scope of this study. We focus mainly on relative variations in the RRI_sp^1064nm and RRI_aeronet^1020nm. Owing to considerable water content in aerosol particles resulted from hygroscopic growth under high ambient RH conditions (Figure 3d), the RRI_aeronet^1020nm is generally lower than the dry aerosol RRI_sp^1064nm in summer.

Variations in the dependence of RRI_sp^1064nm on particle size during polluted periods in winter are investigated further. Four polluted episodes are defined with a maximum hourly PM2_.5 value exceeding 50 μg m⁻³, which are denoted as EP1 to EP4 (Figure 3b). The average values of the bulk RRI_sp^1064nm in EP2, EP3, and EP4 are 1.45 ± 0.02, 1.43 ± 0.03, and 1.44 ± 0.03, respectively, which are obviously lower than the average RRI_sp^1064nm (1.49 ± 0.01) during the clean period between EP1 and EP2. A slightly lower bulk RRI_sp^1064nm is observed in EP1, with a mean of 1.48 ± 0.01, compared with the clean period. The generally increasing RRI_sp^1064nm trend with increasing particle size in winter when the D_mob is larger than ∼250 nm does not change with pollution and is extremely evident in EP2 to EP4. The RRI_sp^1064nm values in polluted episodes appear to decrease substantially at a D_mob of ∼250 nm (Figure S8). This finding may indicate that the increased contribution of OA (particularly less oxidized OA) with a low RRI is apparent at a D_mob of ∼250 nm in the polluted episodes. Further research accompanied with size-resolved chemical composition during different polluted episodes is necessary.

3.3. Diurnal variations in RRI_sp

The chemical composition of aerosols is influenced significantly by aging processes in the atmosphere; thus, diurnal variations in the RRI_sp^1064nm are investigated further. As shown in Figure 5a, high RRI_sp^1064nm values are clearly observed in the afternoon, peaking at around 15:00 (hereinafter the local time) in summer. The diurnal variations in the RRI_sp^1064nm are highly consistent with those in O_x (O_x=O₃+NO₂) concentrations, which can be considered as a proxy for atmospheric aging caused by photochemical reactions (Canonaco et al. 2015), thereby implying the important impact of aging processes on the RRI_sp^1064nm (Figure 5b). A significant positive correlation is found between hourly bulk RRI_sp^1064nm and O_x concentrations in summer with an R value of 0.42 (p < 0.001). Aging processes result in the formation of secondary aerosol components, including SIA and SOA, which likely have larger RRI_sp^1064nm values than primary components. In addition to the increase in the RRI_sp^1064nm resulting from an increase in SIA, SOA from a certain anthropogenic precursor (e.g., toluene) displays high RRI values at high NO_x concentrations (Moise, Flores, and Rudich 2015). The diurnal variations in the size-resolved RRI_sp^1064nm are generally consistent with those in the bulk RRI_sp^1064nm (not shown). Similar diurnal variations in size-resolved RRI_sp^1064nm were observed by Zhao, Tan, et al. (2019) in summer. However, no distinct diurnal peak is observed in winter, which has only a minimum value in the evening (around 19:00 local time). The weak diurnal variations in the RRI_sp^1064nm during daytime in winter are attributable to low O_x concentrations and weak diurnal variations (Figures 3c and 5b), which correspond to low atmospheric aging caused by photochemical reactions. High primary aerosol emissions from traffic and residential heating may explain the low RRI_sp^1064nm values during evenings and at night (M. Hu et al. 2012). If only RRIsp1064nm values on clear days (with maximum hourly PM2_.5 concentrations lower than 20 μg m⁻³) are investigated, then a clear double-valley diurnal pattern can be presented (gray diamonds in Figure 5a). This high RRI_sp^1064nm value may be attributed to relatively high atmospheric photooxidation in the afternoon, whereas the low RRI_sp^1064nm values in the evenings and early mornings are likely related to high primary emissions.

The size dependence of the RRI_sp^1064nm is compared at three typical times, that is, 14:00–16:00, when the RRI_sp^1064nm reaches its maximum value; in the early morning at 05:00–07:00; and in the evening at 18:00–20:00, when RRI_sp is low. As shown in Figure 6, a decreasing RRI_sp^1064nm trend as D_mob increases is commonly observed in the three time periods in summer, whereas an increasing RRI_sp^1064nm trend is observed in winter. The increase in summertime RRI_sp^1064nm in the afternoon occurs mainly at a D_mob of < 400 nm, likely indicating that secondary aerosols formed by atmospheric photooxidation processes are located mainly in small particle sizes. The more evident increase in the RRI_sp^1064nm at a D_mob of < 400 nm is also likely related to the higher fraction of BC-free particles in small particle sizes. As shown in Figure S9, the number fraction of non-BC particles decreases from ∼70% at a D_mob of < 400 nm to ∼50% at a D_mob of 422 nm in summer, indicating large particles are more likely to contain BC. This finding implies SIA and/or SOA are more likely to exist solely with BC in small particle sizes, leading to a more evident increase in the RRI_sp^1064nm compared with large particles. The low RRI_sp^1064nm during evenings in winter is generally observed throughout the studied D_mob range (200–450 nm), thereby implying the uniform increase in the contribution of less oxidized aerosols at different particle sizes.

4. Conclusions

A DMA-SP2/CPC tandem system was designed and deployed into two field observations to continuously measure size-resolved RRI_sp^1064nm over a D_mob range of approximately 200 nm to 400 nm. Results are concluded as:

1. A decreasing RRI_sp^1064nm trend as D_mob increased from 237 nm to 422 nm was observed in summer, whereas an increasing RRI_sp^1064nm trend within the D_mob range of 225–400 nm was observed in winter, which likely indicate different variations in aerosol physicochemical properties with particle size in different seasons.

2. The size-resolved RRI_sp^1064nm values were generally higher in summer than in winter throughout the studied D_mob range, thereby resulting in a large seasonal average bulk RRI_sp^1064nm of 1.50 ± 0.01 in summer and a small seasonal bulk RRI_sp^1064nm of 1.47 ± 0.02 in winter. Submicron particles are generally dominated by SIA and OA (Sun et al. 2015; W. Hu et al. 2017); thus, the low RRI_sp^1064nm values in winter were likely attributable to the increased contribution of OA, particularly less oxidized OA, which may have lower RRI values than SIA.

3. Variations in the RRI_sp^1064nm were more dramatic in winter than in summer, thereby indicating increased diverse sources of atmospheric aerosols. Owing to the increasing contributions of primary aerosol locally emitted and/or transported from southern polluted regions, pronounced low RRI_sp^1064nm values were generally observed during polluted episodes in winter. By contrast, the RRI_sp^1064nm in summer mainly exhibited regular diurnal variations, with high values in the afternoon and low values at night, which corresponded well to diurnal variations in O_x concentrations. High O_x in the afternoon indicated a strong photooxidation, which is favorable to the formation of SIA and/or SOA and resulted in an increase in the RRI_sp^1064nm. However, chemical composition cannot be employed solely to interpret the decreasing RRI_sp^1064nm trend with an increasing D_mob in summer, because increased SIA contributions and decreased OA were generally observed as particle size enlarged, which should result in the expected increase in the RRI_sp^1064nm. Further studies on size-resolved chemical composition and physical properties, such as ρ_eff, are necessary to obtain an in-depth understanding of the size dependence of the RRI_sp^1064nm.

Acknowledgments

We would like to sincerely thank Prof. Aiguo Li and Dr. Yifan Tao from the Institute of Atmospheric Physics for providing the observational meteorological data from the Beijing 325 m-high meteorological tower.

Additional information

Funding

This research was supported by the National Key Research and Development Program of China (2019YFA0606801), the Natural Science Foundation of China (41775155 and 41575150), and the Jiangsu Collaborative Innovation Center for Climate Change.