Depolarization ratios retrieved by AERONET sun – sky radiometer data and comparison to depolarization ratios measured with lidar

The linear particle depolarization ratios at 440, 675, 870, and 1020 nm were derived using data taken with the AERONET sun–sky radiometer at Seoul (37.45 N, 126.95 E), Kongju (36.47 N, 127.14 E), Gosan (33.29 N, 126.16 E), and Osaka (34.65 N, 135.59 E). The results are compared to the linear particle depolarization ratio measured by lidar at 532 nm. The correlation coefficient R2 between the linear particle depolarization ratio derived by AERONET data at 1020 nm and the linear particle depolarization ratio measured with lidar at 532 nm is 0.90, 0.92, 0.79, and 0.89 at Seoul, Kongju, Gosan, and Osaka, respectively. The correlation coefficients between the lidar-measured depolarization ratio at 532 nm and that retrieved by AERONET at 870 nm are 0.89, 0.92, 0.76, and 0.88 at Seoul, Kongju, Gosan, and Osaka, respectively. The correlation coefficients for the data taken at 675 nm are lower than the correlation coefficients at 870 and 1020 nm, respectively. Values are 0.81, 0.90, 0.64, and 0.81 at Seoul, Kongju, Gosan, and Osaka, respectively. The lowest correlation values are found for the AERONET-derived linear particle depolarization ratio at 440 nm, i.e., 0.38, 0.62, 0.26, and 0.28 at Seoul, Kongju, Gosan, and Osaka, respectively. We should expect a higher correlation between lidar-measured linear particle depolarization ratios at 532 nm and the ones derived from AERONET at 675 and 440 nm as the lidar wavelength is between the two AERONET wavelengths. We cannot currently explain why we find better correlation between lidar and AERONET linear particle depolarization ratios for the case that the AERONET wavelengths (675, 870, and 1020 nm) are significantly larger than the lidar measurement wavelength (532 nm). The linear particle depolarization ratio can be used as a parameter to obtain insight into the variation of optical and microphysical properties of dust when it is mixed with anthropogenic pollution particles. The single-scattering albedo increases with increasing measurement wavelength for low linear particle depolarization ratios, which indicates a high share of fine-mode anthropogenic pollution. In contrast, single-scattering albedo increases with increasing wavelength for high linear particle depolarization ratios, which indicated a high share of coarse-mode mineral dust particles. The retrieved volume particle size distributions are dominated by the fine-mode fraction if linear particle depolarization ratios are less than 0.15 at 532 nm. The fine-mode fraction of the size distributions decreases and the coarsemode fraction of the size distribution increases for increasing linear particle depolarization ratio at 1020 nm. The dust ratio based on using the linear particle depolarization ratio derived from AERONET data is 0.12 to 0.17. These values are lower than the coarse-mode fraction derived from the volume conPublished by Copernicus Publications on behalf of the European Geosciences Union. 6272 Y. Noh et al.: Depolarization ratios retrieved by AERONET sun–sky radiometer data centrations of particle size distributions, in which case we can compute the coarse-mode fraction of dust.

tween the two AERONET wavelengths.We cannot currently explain why we find better correlation between lidar and AERONET linear particle depolarization ratios for the case that the AERONET wavelengths (675, 870, and 1020 nm) are significantly larger than the lidar measurement wavelength (532 nm).
The linear particle depolarization ratio can be used as a parameter to obtain insight into the variation of optical and microphysical properties of dust when it is mixed with anthropogenic pollution particles.The single-scattering albedo increases with increasing measurement wavelength for low linear particle depolarization ratios, which indicates a high share of fine-mode anthropogenic pollution.In contrast, single-scattering albedo increases with increasing wavelength for high linear particle depolarization ratios, which indicated a high share of coarse-mode mineral dust particles.The retrieved volume particle size distributions are dominated by the fine-mode fraction if linear particle depolarization ratios are less than 0.15 at 532 nm.The fine-mode fraction of the size distributions decreases and the coarsemode fraction of the size distribution increases for increasing linear particle depolarization ratio at 1020 nm.The dust ratio based on using the linear particle depolarization ratio derived from AERONET data is 0.12 to 0.17.These values are lower than the coarse-mode fraction derived from the volume con-centrations of particle size distributions, in which case we can compute the coarse-mode fraction of dust.

Introduction
There are various aerosol types of natural (primarily desert dust and sea salt) and anthropogenic (primarily combustion of biomass and fossil fuels) origin.A precise understanding of the radiative forcing of these aerosol types is the key to quantifying the aerosol impact on regional and global climate change (IPCC, 2013).In order to better estimate the aerosol effect (direct and indirect radiative forcing) on global climate change many studies have been performed with the goal of classifying aerosol types (Burton et al., 2013;Eck et al., 1999Eck et al., , 2010;;Lee et al., 2010a;Dubovik et al., 2002).However, those studies do not separate aerosol types according to their contribution of mixed aerosols (respectively mixtures of different aerosol types) in a plume; they merely classify dominant aerosol types based on the optical properties of aerosols.
Dust is one of the major aerosol components in the atmosphere.Dust affects Earth's climate by interacting with solar as well as thermal infrared radiation.Dust also affects atmospheric dynamics, atmospheric chemistry, air quality, and ocean biogeochemistry over a wide range of spatial and temporal scales (e.g., Haywood et al., 2005;Jickells et al., 2005;Husar et al., 2001).On the global average, dust contributes to about one quarter of aerosol optical depth (τ ) in the midvisible wavelength range (Kinne et al., 2006).Dust is also light absorbing (Lafon et al., 2004(Lafon et al., , 2006;;Alfaro et al., 2004;Sokolik and Toon, 1999).It is estimated that more than half of aerosol absorption optical depth at 550 nm may come from dust (Chin et al., 2009).
The size distribution and absorption properties of desert dust and anthropogenic aerosols show properties that can be clearly distinguished (Russel et al., 2010;Dubovik et al., 2002).Desert dust predominately consists of coarse-mode particles (typical radius > ∼ 1 µm).In contrast, combustionproduced particles are predominately found in the finemode fraction of particle size distributions (typically radius < ∼ 1 µm).Aerosols in which fractions of fine-mode and coarse-mode particles are mixed are among the most challenging aerosol types to characterize.If we can separate desert dust from other aerosols in mixed dust plumes, we will improve our understanding of the effect those mixed aerosol plumes have on climate change.
The linear particle depolarization ratio (δ p ) strongly depends on particle shape.Since dust particles have a nonspherical shape, the δ p can be used to identify the presence of dust particles in the atmosphere.In that regard lidar is a particularly powerful measurement technique (Shimizu et al., 2016;Tesche et al., 2009;Noh et al., 2007Noh et al., , 2008;;Iwasaka et al., 2003;Cairo et al., 1999).The δ p has also been used to identify biogenic aerosols.Noh et al. (2013a, b) and Sassen et al. (2008) identified the vertical distribution of pollen in the atmosphere using the δ p measured by lidar.The possibility that dust particles are mixed with other, man-made pollution and/or biomass burning particles is very high, except in source regions of dust emissions where population density, and thus emissions caused by human activities are comparably low (Sun et al., 2005(Sun et al., , 2010;;Yu et al., 2006).Mixtures of dust particles and anthropogenic particles cause changes in the δ p of dust plumes.Shin et al. (2015) reported on lidar measurements and show that the δ p decreases when dust is mixed with pollution particles.The δ p is also a very useful parameter that allows us to separate dust from non-dust particles (anthropogenic, smoke, and sulfate particles) in mixed dust plumes by retrieving the dust ratio (R D ).Shimizu et al. (2004) estimated the contributions of dust and pollution particles in a mixed dust plume by the assumption that both aerosol types are externally mixed.The optical data of these mixed dust plumes were separated by the R D into pure dust content and the contribution by anthropogenic particles (Noh et al., 2012b(Noh et al., , 2016a;;Bravo-Aranda et al., 2015;Noh, 2014).Tesche et al. (2011) separated the optical properties of desert dust and biomass burning particles in mixed dust and smoke plumes observed over the tropical North Atlantic, west of the African continent.The authors used multi-wavelength aerosol Raman lidar in combination with polarization lidar.Burton et al. (2014) provided a generalized version of the separation methodology between two aerosol types, i.e., urban pollution plus dust, marine plus dust, and smoke plus marine, by modifying the methodology suggested by Shimizu et al. (2004) and Tesche et al. (2011).Noh (2014) and Tesche et al. (2009Tesche et al. ( , 2011) ) used the δ p to retrieve vertically resolved single-scattering albedo (SSA) of mixed dust plumes by separating the contribution of dust and non-dust particles.Ansmann et al. (2011) andNavas-Guzmán et al. (2013) separated the contribution of volcanic ash and sulfate particles to total backscatter and extinction coefficient by using the δ p .
However, the number of lidar measurement sites is limited and provides rather sparse spatially resolved and temporally resolved information on the global scale.Thus, we need another methods that could allow us to measure δ p .
AERONET (Aerosol Robotic NETwork) is an automated robotic sun-and-sky-scanning measurement network that took and still collects data at more than 797 measurement sites worldwide (http://aeronet.gsfc.nasa.gov/).These sites span everything from temporally limited observations at sites at which field campaigns were carried out to sites that carry out long-term observations since AERONET started with its observations.AERONET sun-sky radiometers provide globally distributed observations of spectrally resolved aerosol optical depth (τ ) and data inversion products such as particle size distributions and complex refractive indices of different aerosol types (Holben et al., 1998).Dubovik et al. (2006) suggested using AERONET sunsky radiometer data to retrieve δ p .Müller et al. (2010Müller et al. ( , 2012) ) calculated the δ p of Saharan dust using AERONET sun-sky radiometer data.Noh et al. (2016b) and Lee et al. (2010b) used AERONET sun-sky radiometer data to retrieve the δ p of Asian dust.However, only cases of nearly pure desert dust particles were analyzed in these studies.There exist no studies in which the δ p for various mixtures (mixing ratio) between desert dust and anthropogenic pollution particles using AERONET data has been determined.
In this contribution we attempt to verify the reliability of AERONET-derived values of δ p by comparing these values to values of δ p measured by lidar.
Section 2 presents the methods used in this study.Section 3 presents our results.We discuss our results and summarize our findings in Sect. 4.

Study sites
The AERONET sites considered in this study are all located along the pathway of the regime of prevailing westerly winds that serve as major transport routes of Asian dust, which is advected from the arid regions of China and Mongolia.Figure 1 shows the locations of the AERONET sun-sky radiometers and lidars used in this study.
Seoul (37.45  E) are coastal sites.The Gosan site faces the Yellow Sea and is considered an ideal location for monitoring regional background aerosols in East Asia because there are few local industrial sources in that region.The other three sites are located inside large cities.We also use data from the AERONET site at Dunhuang (40.49• N, 94.95 • E) in our study, and we analyzed the depolarization ratios and optical properties of pure Asian dust at this source region.
Lidar data are obtained from the lidar network of the National Institute of Environmental Research (NIES), Japan.The lidars operated in this network are two-wavelength (1064 and 532 nm) Mie-scattering lidars that also measure the linear particle depolarization ratio at 532 nm.The details of these lidar systems and data processing procedures are explained by Shimizu et al. (2016) and Sugimoto et al. (2014).
The locations of the lidar systems used for our research work are the same as the locations of the AERONET systems for the sites in Seoul, Gosan, and Osaka.The lidar used for the Kongju site is located approximately 32 km to the northwest of the AERONET site.

Depolarization ratios derived from data taken with
AERONET sun-sky radiometer Dubovik et al. (2006) introduced kernel lookup tables that describe mixtures of spheroid particles.These kernel lookup tables were used to infer the δ p of mineral dust observed with sun-sky radiometer.The details of the AERONET inversion algorithm that processes data of mineral dust are given by Dubovik et al. (2006).
Briefly, the retrieval of the depolarization ratios works as follows.The elements F 11 (λ) and F 22 (λ) of the Müller scattering matrices (Bohren and Huffman, 1983) are computed from the retrieved complex refractive indices and particle size distributions.The term F 11 (λ) is proportional to the flux of the scattered light (Volten et al., 2001) in the case of unpolarized incident light.The F 22 (λ) term in turn follows from the angular and spectral distribution of the radiative intensity which is measured with the AERONET instrument (Dubovik et al., 2006).Another input parameter that is needed for the retrieval of δ p is the aspect ratio distribution.The aspect ratio indicates the ratio of a particle's longest axis to its shortest axis.In the case of prolate particles its polar diameter is greater than the equatorial diameter, in contrast to oblate particles where this ratio is vice versa.The aspect ratio distribution is kept to a fixed distribution in the AERONET model since scattering elements are nearly equivalent for all mixtures of spheroid particles (Dubovik et al., 2006).
From the ratio of the elements F 11 (λ) and F 22 (λ) at the scattering angle 180 • the δ p (λ) can be computed as The δ P derived from the sun-sky radiometer data is written as δ S P so that we can distinguish it from the lidar-derived δ P , i.e., δ L P .The contributions of dust and anthropogenic pollution particles to the total backscatter coefficients of mixed aerosol plumes were estimated from the δ P under the assumption that both types of aerosol particles are externally mixed.The dust 6274 Y. Noh et al.: Depolarization ratios retrieved by AERONET sun-sky radiometer data ratio (R D ) of the dust-related backscatter coefficient to the total backscatter coefficient was calculated using Eq. ( 1), based on the method suggested by Shimizu et al. (2004): where δ 1 and δ 2 denote the δ S P of pure dust and non-dust particles (i.e., the total aerosol plume without the contribution by dust), respectively, in the external mixture of aerosol particles.The values δ 1 and δ 2 can be empirically determined.In the present study, we used the value 0.34 for δ 1 .This value was derived from adding 0.01 to the maximum value observed at the Dunhuang site (Asian dust source region).The value of 0.02 was used for δ 2 , which is the minimum value used in this study.When δ S P was higher than δ 1 or lower than δ 2 , R D was set to 1 or 0, respectively.
Two kinds of coarse-mode fractions (CMFs) were calculated.The CMF of the aerosol optical depth (τ ) (CMFτ ) is calculated from the ratio of the coarse-mode τ to the total (coarse + fine mode) τ at the same wavelength at which the δ S P is available.The CMF is also calculated on the basis of the volume concentration (CMF vc ).

Column-integrated depolarization ratio measured by lidar
The lidar systems used in our study measure the linear volume depolarization ratio (aerosols + molecules; δ L ) from the linearly and perpendicularly polarized components of the Mie-Rayleigh backscatter signals at 532 nm wavelength (Shimizu et al., 2004).The value of δ L is defined as where P (z) is the backscatter signal with respect to height z.
The symbols and ⊥ denote the linearly and perpendicularly polarized components with respect to the plane of polarization of the emitted laser light, respectively.In order to obtain reliable depolarization ratios, the data of the lidar measurements must be calibrated before physical quantities such as the linear volume depolarization ratio can be retrieved.It is important to calibrate the signal intensities of the P and P ⊥ first, before the linear volume depolarization ratio is calculated.The calibration method of the lidar system is explained in detail by Shimizu et al. (2016) andNishizawa et al. (2017).
The difference of the sensitivity between two photomultiplier tubes (PMTs) that are used by the lidar system to detect these components is checked regularly by the following method.A sheet polarizer whose polarizing direction is set at 45 • to the polarizing plane of the emitted light is inserted in front of the beam splitter cube, and the backscatter signal from the sky is recorded as a reference signal.In this reference record, the light intensities of the two channels are equal after the sheet polarizer, so the calibration constant can be obtained by comparing the recorded values of P and P ⊥ .In the next step, the sheet polarizer is rotated by 90 • , which sets the polarizing angle at −45 • , and another reference signal is recorded.Then relative calibration of P and P ⊥ channels is performed using signals measured for the polarizing angles at ±45 • .This pair of reference signals reduces any error caused by a poor positioning of the sheet polarizer (Freudenthaler et al., 2009(Freudenthaler et al., , 2016)).The reference signals are usually recorded once per year for each lidar (Shimizu et al., 2016).
A different definition of the δ L (δ LT = P ⊥ /(P + P ⊥ )) is used in many other studies, e.g., Noh et al. (2016aNoh et al. ( , 2013b) ) and Sakai et al. (2000).The term δ LT is defined differently from the definition we use for δ L in our work.It is expressed by the following equation, derived from Eq. (3) (see Cairo et al., 1999).
The variable δ L P differs from δ L as it depends on the concentration of particles and it does not take account of the contribution (concentration) of air molecules.In this contribution, δ L P can be calculated on the basis of the definition provided by Sakai et al. (2003): The backscatter ratio R is the ratio of the sum of the aerosol backscatter coefficient (β P + β M ) to the pure molecular backscatter coefficient (β M ), which, according to Whiteman et al. (1992) can be expressed by β P and β M are the volume backscatter coefficients of aerosol particles and air molecules, respectively.The molecular depolarization ratio (δ M ) is assumed to be 0.014 (Shimizu et al., 2004).
The parameter δ L P can be derived by lidar measurements in terms of vertical profiles.In contrast, δ S P describes a columnintegrated value.For that reason, δ L P had to be changed to column-integrated values in our study so that we could carry out a direct comparison with δ S P .The column-integrated and weighted values of δ L P (δ CL P ) can be calculated by where the term W (z) is a weight factor that is calculated by using the measured aerosol backscatter coefficient (β P ) according to The aerosol backscatter coefficient is derived by the backward version of Fernald's method (Fernald, 1984).The data observed at 9 km height are used as reference height for the analysis of data taken under cloud-free conditions.The reference height is lowered if the signal-to-noise ratio at 9 km is not sufficient, which may be the case of high aerosol concentration.Molecular density profiles are taken from the COSPAR international reference atmosphere (CIRA-86) for computing the Rayleigh scattering component.A constant lidar ratio of 50 sr is applied in the calculation of the aerosol backscatter coefficients (Shimizu et al., 2016).Since the lidar ratio differs for different aerosol types, the selection of the lidar ratio is important to obtain exact values of extinction and backscatter coefficients.However, we only use the ratio of the backscatter coefficient in the calculation of W (z) in our study.For that reason, the value of the lidar ratio does not affect the calculation of W (z).
Figure 2 shows a retrieval example of δ CL P .Three cases, corresponding to lidar measurements carried out at 23:00 UTC (start time of measurement) on 13 March 2010 (Fig. 2a), at 06:00 UTC on 22 March 2010 (Fig. 2b), and at 23:15 UTC on 3 May 2010 (Fig. 2c), are shown.The measurement on 13 March 2010 describes an aerosol plume that has a high value of δ L P and a high value of β P (case 1, Fig. 2a).The measurement on 22 March 2010 describes an aerosol plume with high β P below and above the planetary boundary layer (PBL), respectively.However, a high value of δ L P is detected above the PBL (case 2, Fig. 2b) only.The measurement on 2 May 2010 describes an aerosol plume with low backscatter coefficients and a high value of δ L P (case 3, Fig. 2c).The averaged values of δ L P (δ ave ) which do not consider the backscatter intensity but just the averaged δ L P from the bottom to the top height of the profiles of δ L P are listed in Fig. 2, too.
The values of δ S P at 1020 nm are 0.250, 0.140, and 0.164 for the cases 1-3, respectively.The values of δ CL P are similar to the values of δ S P .We find 0.243, 0.129, and 0.157 for the cases 1-3, respectively.However, the values of δ ave for case 1 and case 3 are different compared to the values of δ S P .Since δ L P is not directly related to aerosol concentration but only to the non-sphericity of (an ensemble of) aerosol particles (inside a given volume of air) large values of δ L P can occur for high as well as for low aerosol backscatter coefficients.Lower values of δ ave compared to δ S P are linked to high values of δ L P for the case of high aerosol backscatter coefficients of a thin aerosol layer (see Fig. 2a).We find that the value of δ ave is higher than the value of δ S P for the situation in which a high value of δ L P is combined with a low aerosol backscatter coefficient (see Fig. 2c).The examples in Fig. 2 show that δ CL P has to be compared with δ S P .P is 1020 nm.The height is expressed as altitude above ground level (a.g.l.).

AERONET-derived depolarization ratios and lidar-measured values: comparison
Figure 3 shows the temporal variation of τ at 500 nm and the values of δ S P at 1020 nm at the four AERONET sites.The combined data of the four sites are shown in the same figure.The number of measurement cases for the four sites is listed in Table 1.The total number of retrieved values of δ S P is 163, 44, 139, and 234 at Seoul, Kongju, Gosan, and Osaka, respectively.Since the measurement cases are limited it is difficult to analyze seasonal trends.However, we find high values of τ and δ S P in spring (March to May).We assume that these high values are caused by transport of dust from East Asia to the Pacific Ocean.
Figure 4 presents scatterplots of δ CL P and δ S P at the four AERONET sites.We compare the values of the δ S P at these four wavelengths (440, 675, 870, and 1020 nm) to the values of δ CL P at 532 nm.The correlation coefficients R 2 at 1020 nm are high.We find 0.90, 0.92, 0.79, and 0.89 at Seoul, Kongju, Gosan, and Osaka, respectively.
We find similarly high correlation between δ CL P and δ S P at 870 nm, i.e., numbers are 0.89, 0.92, 0.76, and 0.88 at Seoul, Kongju, Gosan, and Osaka, respectively.The correlation at 675 nm is lower than the values we find at 870 and 1020 nm.Values are 0.81, 0.90, 0.64, and 0.81 at Seoul, Kongju, Gosan, and Osaka, respectively.The correlation is significantly low at 440 nm.Values are 0.38, 0.62, 0.26, and 0.28 at Seoul, Kongju, Gosan, and Osaka, respectively.The correlation coefficient at 440 nm at Kongju is much higher than at the other sites.This higher correlation may be caused by the limited number of observational data and/or observation times.Only 44 cases were taken during a short period of 2 months, from April to May 2012 at Kongju.
Figure 4 shows that the differences between δ CL P and δ S P are high when the value of δ CL P is less than 0.10 at Seoul, Gosan and Osaka.However, the number of cases of low δ CL P (< 0.10) is comparably low at Kongju compared to the other sites.The number of cases with high δ CL P (> 0.25) is comparably high (with respect to all cases), i.e., 23 % (10 cases out of 44 in total), compared to what we find for the other sites: 0/163, 3/139, and 10/234 at Seoul, Gosan, and Osaka, respectively (see Table 1).We tried to find the reason for the comparably low correlation at 440 nm.For that reason, we retrieved the δ S P at 532 nm by interpolating the value of δ S P at 532 nm on the basis of the four AERONET wavelengths.In the next step the differences between δ S P and δ CL P at 532 nm (δ CL P,532 − δ S P,532 ) were calculated by deducting δ CL P (at 532 nm) from δ S P (at 532 nm) for all 580 cases for which we have δ S P at the four AERONET sites.
Figure 5 shows the variation of the averaged δ S P at the five wavelengths and the values of δ CL P at 532 nm divided by the differences of δ CL P,532 − δ S P,532 .The differences of δ S P between the wavelengths at 440 and 1020 nm are high.
We find that δ S P decreases with increasing wavelength if the value of δ CL P,532 − δ S P,532 is low.The differences between δ S P at 440 nm and δ S P at 1020 nm become less for increasing interval number, i.e., δ CL P,532 − δ S P,532 for the interval number 7 (0-0.02)(i.e., the yellow triangle pointing to the right in Fig. 5).The value of δ CL P at 532 nm shows lower values than δ S P at 1020 nm in those intervals.The differences between δ CL P at 532 nm and δ S P at 1020 nm are reduced as δ CL P,532 − δ S P,532 is decreased up to the interval number 7 (0-0.02).We find an increase of δ S P with increasing wavelength from the interval number 8 (0-(−0.02)).The value of δ CL P at 532 nm is larger than the value of δ S P at 1020 nm in the interval number 8 (0-(−0.02)).Also, the differences between δ CL P at 532 nm and δ S P at 1020 nm are increased as the interval number increased.
If we assume that the value of δ CL P at 532 nm is close to real value of δ P , the results in Fig. 5 indicate that the δ S P at 440 nm has been retrieved to be higher value than the δ CL P at 532 nm in the interval number from 1 (> 0.12) to 7 (0-0.02)when the δ CL P at 532 nm showed low values (less than 0.08).Conversely, when the high δ CL P at 532 nm was measured, the δ S P at 440 nm showed a lower value than the δ CL P at 532 nm. Figure 6 shows the average of volume particle size distributions of each interval data as separated in Fig. 5.We see that the volume size distributions change from fine-mode-dominated size distributions to coarse-modedominated size distributions when the interval number moves from 1 (> 0.12) to 12 (< (−0.08)).
The important point of Fig. 6 is the variation of the volume median radius (Rv).The volume median radius of the coarse (Rv c ) and the fine (Rv f ) mode shows a maximum value at the interval number 1 (> 0.12).Rv f clearly and progressively decreases as the interval number moves from 1 (> 0.12) to 12 (< (−0.08)).The Rv f of the interval number 1 (> 0.12) is 2 times larger than the interval number of 12 (< (−0.08)) as 0.28 ± 0.03 and 0.13 ± 0.01 µm, respectively.Rv c also shows a pattern of decreasing values with decreasing values of δ CL P,532 − δ S P,532 , but it does not show as progressively as Rv f .
Figures 5 and 6 show that the value of δ S P at 440 nm tends to be retrieved high for conditions where there is no dust at all or the dust concentration is low.Such conditions are usually dominated by a significant fine mode of the particle size distribution.When dust particles contribute the main share to the particle concentration, i.e., high values of δ CL P at 532 nm, the contribution of fine-mode particles is small.When particles in the fine mode are the main contribution of the particle size distribution, i.e., low values of δ CL P at 532 nm, the size of the particles in the fine-mode fraction are considered to have a large influence on the retrieval of the values of δ S P .This effect is considered to be more significant at 440 nm, i.e., at short wavelengths.Mamouri and Ansmann (2017) found that the value of δ L P is maximum at 532 nm and lower at 355 and 1064 nm because of the competing influence the fine-mode and coarsemode dust fraction have on the overall values (fine + coarse) of δ L P at the three wavelengths.Haarig et al. (2017) found that on average the values of δ L P for aged Saharan dust were 0.25 at 355 nm, 0.31 at 532 nm, and 0.225 at 1064 nm.Müller et al. (2010Müller et al. ( , 2012) ) and Freudenthaler et al. (2009) also found spectral slope of the depolarization ratio with the maximum at 532 nm and lower values at 355 and 1064 nm during the 2006 Saharan Mineral Dust Experiment (SAMUM).The results clearly show a different pattern of the spectral variations of δ P measured by lidar and retrieved from sun-sky radiometer observations of dust.
It is a striking result that δ S P at 1020 nm, unlike δ S P at 440 nm, is very similar to the values of δ CL P at 532 nm.Though we cannot identify the reason for this similarity and -even if the wavelengths (lidar at 532 nm and AERONET sun-sky radiometer at 1020 nm) are different -we may use the values of δ S P at 1020 nm as a qualitative indicator of the presence of mineral dust particles in the atmosphere.It remains to be seen whether we can use this parameter also as a qualitative measure of the mixing ratio of mineral dust and anthropogenic pollution particles compared to the more robust parameter δ CL P (at 532 nm).

Classification of AERONET-derived depolarization ratios based on values of δ S P at 1020 nm
The previous section showed that the depolarization ratio from AERONET does not agree with the lidar values at 532 nm.A quantitative interpretation of the AERONET values is not possible.However, even if the quantitative numbers are wrong there may still be a possibility to use the AERONET numbers as qualitative indicators, i.e., trend.We must keep in mind, however, that the spectral slope of the depolarization ratios from AERONET are not correct for cases of (high) dust load.We classified the observational data into six groups based on the values of δ S P at 1020 nm.Group 1 contains cases for which the values of δ S P are 0.05 at 1020 nm.Groups 2-5 include values in the range of 0.05-0.1,0.1-0.15,0.15-0.2,and 0.2-0.25 at 1020 nm, respectively.Values are above 0.25 in group 6.
Table 1 shows the number of data sets (observation cases) for each of the six data groups at the four observation sites.Table 1 also lists the averaged values of τ at 500 nm, the optical-depth-related Ångström exponent (Å) between 440 and 870 nm, and the light-absorption-related Ångström exponent (Å A ) between 440 and 870 nm of each group.The values of τ are similar in all six groups.The reason for that similarity is because τ is insensitive to the shape and size of the particles.The values of Å A increase.The values of Å decrease with increasing value of δ S P .Figure 7 shows the values δ S P of the six groups at the four measurement wavelengths of the AERONET sun-sky radiometers.There is a rather clear increase of δ S P with respect to increasing measurement wavelength in group 6.We see a similar pattern in groups 4 and 5, respectively.In contrast, groups 1-3 show the highest values of δ S P at 440 nm whereas the values of δ S P are similar at the other three measurement wavelengths.
Values of δ L P of pure mineral dust plumes were measured at three wavelengths (355, 532, and 1064 nm) with lidar (Freudenthaler et al., 2009) during SAMUM in 2006.Freudenthaler et al. (2009) found values of 0.31 for δ L P at 532 nm.Müller et al. (2010Müller et al. ( , 2012) compared those data with data derived from collocated AERONET sun-sky radiometer observations.Values of δ p from both instruments agree at 1064 nm wavelength (Müller et al., 2010(Müller et al., , 2012)).If the sun-sky radiometer results are extrapolated to the lidar wavelength of 355 nm, the value of δ p obtained from the sun-sky radiometer is 20 % lower than the value obtained from the lidar observations (see Fig. 3 in Müller et al., 2010, andFig. 7 in Müller et al., 2012).Müller et al. (2010) find that the values of δ p inferred from the sun-sky radiometer observations tend to be lower than the values measured with lidar for the case of pure mineral dust.Only group 6 (Kongju and Gosan) shows 20-30 % lower values of δ S P compared to δ CL P in the visible wavelength range, which is a similar trend reported by Müller et al. (2010).This feature, i.e., that δ CL P is higher than δ S P , is also found in group 5.However, the differences of the numbers are less compared to the differences we find for group 6.
The opposite situation, i.e., δ CL P is lower than δ S P , is found in groups 1-3. Figure 4 shows that the values of δ S P at 440 nm are significantly different from the values found at the other three wavelengths.
Figure 7 shows the values of δ CL P at 532 nm of each group.We find 0.27 ± 0.02 and 0.27 ± 0.03 at Kongju and Gosan, respectively, for group 6.The values at the Osaka site are lower; we find 0.21 ± 0.04.The highest values of δ CL P are 0.29 at the Kongju and Gosan sites, respectively; see Fig. 4.
The differences of the δ p at the observation sites are likely caused by the appearance of dust.The transport distance to the four observation sites may have influence on the values of δ p .Kongju and Gosan have similar transport distances from the source regions of East Asian dust.Osaka is located at a distance of 1 to 2 days of transport time from Kongju and Gosan.It means that more dust particles can be removed by gravity sedimentation during transport (Maring et al., 2003;Gong et al., 2003).Another reason may be that more anthropogenic pollution particles are mixed into these Asian dust plume because of the longer transport times (Kanayama et al., 2002;Noh et al., 2012a;Noh, 2014;Shin et al., 2015).
Figure 8 shows the averaged values of the vertically resolved β P , the values of δ L P , and the weighted values of δ L P (δ LW P ) measured by lidar for each group.The backscatter coefficients generally decrease with height.However, high values are found in the upper atmosphere in those cases in which δ L P is high.Values of δ L P are as high as 0.1 in group 1 and increase with increasing group number (from group 1 to group 6).The increase of the values of δ L P is more obviwww.atmos-chem-phys.net/17/6271/2017/Atmos.Chem.Phys., 17, 6271-6290, 2017 Figure 8.Average values of the vertical profiles of (a) the particle backscatter coefficient, (b) the linear particle depolarization ratios (δ L P ), and (c) the weighted linear particle depolarization ratios (δ LW P ) for group 1 (1), group 2 (2), group 3 (3), group 4 (4), group 5 ( 5), and group 6 (6).The sites are Seoul (red), Kongju (orange), Gosan (blue), and Osaka (black).ous above 2 km height than below 2 km height above ground.The higher values above 2 km above ground may be caused by the fact that Asian dust has a relatively lower chance of being mixed with other pollutants if it is transported in the upper parts of the atmosphere (Shin et al., 2015).In particular, groups 5 and 6 show high values of δ L P , i.e., larger than 0.3 above 2 km height.
We see that β P shows different trends in these two groups.Group 5 has low values of β P and high values of δ L P .In contrast, group 6 shows high values of β P and high values of δ L P .This different behavior in these two groups is clearly visible in the values of δ L P W . Values of δ LW P in group 5 are less than 2 throughout the whole altitude range.Values of δ L P W are larger than 2 in group 6.
Based on our discussion in the previous section we concluded that δ S P at 1020 nm may serve a qualitative indicator of the presence of mineral dust and that values of δ S P at 1020 nm are large for high dust concentrations and small for low dust concentrations (see Fig. 4, 7, and 8).This means that δ S P can be used to retrieve the dust ratio in mixed dust plumes even if vertically resolved information on the linear particle depolarization ratio is not available.

Correlation between δ S P and SSA
One main purpose of our study is to estimate the mixing ratio of dust particles with other pollutants in the atmosphere by analyzing the parameter δ S P .Another purpose of our study is to use δ S P as the basis parameters for estimating the variation of the optical and microphysical properties of dust when it mixes with anthropogenic pollution particles.Correlations between δ S P and other optical parameters allow us to gain insight into these variations.
Variations of aerosol absorption properties can be described by the SSA.The variability of light absorption of aerosol mixtures discussed in this contribution allows us to assess the direct forcing of mixed dust plumes.We can also investigate the semi-direct forcing that may occur from atmospheric heating by absorbing aerosol layers (Noh et al., 2012b(Noh et al., , 2016b;;Noh, 2014).We also investigate how SSA varies with the volume particle size distribution.For these investigations we use the values of δ S P .Figure 9 depicts the AERONET-derived SSA and the volume particle size distributions for each of the six groups.The SSA spectra vary with changing δ S P in clearly distinguishable patterns.The SSA spectra of group 1 (low δ S P ) show that SSA decreases with increasing wavelength.We find that SSA decreases with increasing measurement wavelength for particle plumes that are dominated by urban-industrial and biomass burning particles (Dubovik et al., 2002;Giles et al., 2012).Black carbon particles have the strongest light-absorption capacity in the near-infrared wavelength region.
In contrast, the SSA spectra of group 6 (high δ S P ) show an increase of SSA with increasing wavelength.The wave-Figure 9. Average value of the SSA and the volume particle size distributions derived from the AERONET sun-sky radiometer measurements for each of the six groups considered in this study: group 1 (black), group 2 (red), group 3 (blue), group 4 (pink), group 5 (gray), and group 6 (orange).length dependence (i.e., increasing, decreasing, or constant with wavelength) of SSA is an important property that is used in aerosol type classification because the spectral absorption characteristics depend on aerosol type (Giles et al., 2012;Russell et al., 2010;Eck et al., 2005Eck et al., , 2010;;Dubovik et al., 2002).
The increase of SSA with increasing wavelength is a characteristic optical feature of desert dust particles (Giles et al., 2012).Dust exhibits strong light absorption in the ultraviolet (UV) and at short visible wavelengths (e.g., 440 nm) and lower light absorption from mid-visible to near-infrared wavelengths (Sokolik and Toon, 1999).Kim et al. (2011) define particles with Å < 0.2 as "pure dust" based on observations of dust particles over North Africa and the Arabian Peninsula.The average value of SSA of the "pure dust" part of the aerosol plumes observed at the four observation sites is 0.91, 0.97, 0.97, and 0.97 at 440, 675, 870, and 1020 nm wavelength, respectively.
The SSA spectra of group 6 at Kongju, Gosan, and Osaka resemble the SSA spectra of "pure dust" described by Kim et al. (2011).The SSAs of group 6 at Kongju and Gosan show similar values reported by Kim et al. (2011), i.e., 0.94, 0.98, 0.98, and 0.99 at 440, 675, 870, and 1020 nm wavelength, respectively.Lower values of SSA are observed at Osaka; we find 0.88, 0.95, 0.96, and 0.95 at 440, 675, 870, and 1020 nm wavelength, respectively.The differences of SSA at Osaka may be caused by the mixing of pollution particles with dust.
Except for the SSA at 440 nm, the SSAs at 675, 870, and 1020 nm show higher values for high δ S P .This increase of SSA with increasing δ S P results from the mixing of fine-mode pollution particles with coarse-mode Asian dust.
Mixtures of desert dust and pollution aerosols contain two primary particulate light-absorbing species: black carbon in fine-mode particles (Bond and Bergstrom, 2006) and iron oxides in coarse-mode dust particles (Sokolik and Toon, 1999).Iron oxides cause strong light absorption in the UV and in the short-wavelength range of the visible spectral range of light (Derimian et al., 2008).In pollution particles the principal absorber is soot or black carbon, which exhibits light absorption throughout the entire solar spectrum, because the imaginary part of the complex refractive index shows comparably little wavelength dependence (Bergstrom et al., 2002).
The SSA as a function of δ S P at the Seoul and Gosan sites shows very little variation at 440 nm compared to the other wavelengths we considered in our study (675, 870, and 1020 nm).Eck et al. (2010) suggest that this low variation of SSA at 440 nm is induced when both coarse-modedominated aerosol mixtures (desert dust) and fine-modedominated aerosol mixtures (pollution) have relatively similar magnitudes of light absorption compared to their lightscattering capacity at that wavelength.
The variations of the fine-mode and the coarse-mode parts of the size distributions are clearly shown in Fig. 9. Group 1 contains fine-mode-dominated particle size distributions.The fine-mode part of the particle size distributions decreases and the coarse-mode part of the size distributions increases for increasing δ S P .The variations of Cv c / Cv f for each group listed in Table 2 show these tendencies more clearly.
The values of Cv c / Cv f increase as we move from group 1 to 6.The values of Cv c / Cv f in group 1 are similar at Seoul, Kongju, Gosan, and Osaka; we find 0.29, 0.32, 0.35, and 0.28, respectively.The values of Cv c / Cv f are similar for all three observation sites for each of the groups 1-5.
However, in the case of group 6, Cv c / Cv f is 7.6, 11.3, and 8.6 at Kongju, Gosan, and Osaka, respectively.The values of Cv c / Cv f in group 6 are smaller than the average value at Dunhuang.The average value of Cv c / Cv f at Dunhuang is 15.0 ± 2.6 (see Table 4).Cv c / Cv f decreases if finemode pollution particles are mixed into a dust plume and/or if coarse-mode dust particles are removed from the plume during long-range transport.3.4 δ S P , SSA, and particle size distribution at the dust source region We analyzed the AERONET sun-sky radiometer data taken in the source region of Asian dust and evaluated the optical properties and δ S P of "pure Asian dust".Figure 10 shows δ S P , SSA, and volume particle size distributions observed on 5 days in one of the source regions of Asian dust, i.e., Dunhuang in 2012.Values of δ S P at 440 nm are characteristic of pure dust particles; i.e., we find values larger than 0.25.SSA increases with increasing measurement wavelength, which is also characteristic of pure dust.We find this behavior on all days except for the data representing 8 April 2012.We find that the SSA at each wavelength (at the Dunhuang site) is higher than the SSA retrieved at the corresponding wavelengths for sites in North Africa and the Arabian Peninsula (Kim et al., 2011;Müller et al., 2010).
Table 3 lists aerosol optical depth (τ ) at 500 nm, the optical-depth-related Å for the wavelength range from 440 to 870 nm, the absorption-related Å A for that same wavelength range, the coarse-mode fraction of the volume concentration (CMF vc ), and the dust ratio (R D ) derived on the basis of δ S P at 1020 nm.The values of Å observed at Dunhuang on all measurement dates used in this study indicate the presence of pure dust, as suggested by Kim et al. (2011).The exception is the measurement on 8 April 2012.On that day, we find a value of 0.25 for δ S P at 1020 nm and a value of 1.76 for Å A .These values are comparably lower than the values retrieved for the other observation days.In addition, volume particle size distributions retrieved for 8 April (see Fig. 10) show a higher value of the modal volume radius for the coarse-mode size distribution compared to what is typically found for size distributions of desert dust (Dubovik et al., 2002;Müller et al., 2010Müller et al., , 2012)).We conclude that particles observed on 8 April 2012 describe a mixed dust plume rather than a pure dust plume.
We compare the average values of δ S P , SSA, and the volume particle size distributions observed at Dunhuang with the respective values of group 6 at Kongju, Gosan, and Osaka.We exclude the data taken on 8 April 2012 at Dunhuang in the calculation of the average values because those data likely do not represent pure Asian dust.
Figure 11 shows our comparison results.The highest values of δ S P at Dunhuang are 0.26, 0.28, 0.30, and 0.33 at 440, 675, 870, and 1020 nm wavelength, respectively.The spectral behavior of δ S P (at the four wavelengths) at Kongju, Gosan, and Osaka is similar to the spectral behavior of δ S P retrieved for the Dunhuang site.However, the values of δ S P at the   four measurement wavelengths at Kongju, Gosan, and Osaka are 0.04-0.05lower than the respective values at the Dunhuang site.This difference between Dunhuang and the other three sites may be caused by gravitational settling of coarsemode dust particles during transport and/or by a higher share of anthropogenic pollution particles that may enter the dust plume during long-range transport from the source region to the other three sites.
The volume particle size distributions shown in Fig. 11c corroborate our assumption.The volume concentration of the coarse-mode particles is as low as 0.36, 0.44, and 0.33 (µm 3 µm −2 ) in the far-field sites of Kongju, Gosan, and Osaka, respectively.We find that among all days during which we observed pure Asian dust at Dunhuang, the minimum value of 0.49 (µm 3 µm −2 ) was found on 9 April 2012 (see Fig. 11c and Table 4).
Figure 11b shows a comparison of the spectral SSA between Dunhuang and the other three sites.The average values of SSA at Dunhuang are 0.94, 0.98, 0.98, and 0.98 at 440, 675, 870, and 1020 nm, respectively.The value of 0.94 for SSA at 440 nm at the Dunhuang site is higher than the SSA of "pure dust" observed over North Africa and the Arabian Peninsula (Kim et al., 2011;Müller et al., 2010Müller et al., , 2011)).
As noticed previously, iron oxides cause the strongest light absorption in the ultraviolet and at visible wavelengths (Derimian et al., 2008).We assume that the differences of SSA at 440 nm between the Dunhuang site and observation sites in North Africa and the Arabian Peninsula are caused by dif-Y.Noh et al.: Depolarization ratios retrieved by AERONET sun-sky radiometer data ferences of the chemical composition of dust particles, as for example by the concentration of iron oxides in dust particles at these sites.
Our results show that nearly pure dust may be transported to Kongju and Gosan from long distances and that the CMF of the particle size distribution may not necessarily increase in that case.The spectral behavior of SSA and its values at each wavelength at the Kongju and Gosan sites, which are long-range transport sites, match the values of SSA at Dunhuang.This match suggests that the dust at these three sites may have a similar chemical composition and perhaps also a similar concentration of iron oxide.
SSA values at Osaka are lower than those at Kongju and Gosan.We find values of 0.88, 0.95, 0.96, and 0.95 at 440, 675, 870, and 1020 nm, respectively.These low SSAs can be caused by the presence of pollution particles.We first investigate the vertical distribution of the dust plumes of group 6 at Osaka.The vertical distribution of particles in the dust plumes of group 6 at Osaka can be clearly distinguished according to their observation date.
Figure 12 shows the separated values of β P , δ L P , and δ LW P (dust and non-dust contribution) in terms of cases 1 and 2 for group 6 at Osaka on the basis of the observation date.The data taken on 14 and 15 March 2010 (case 1) show that the dust plumes are distributed up to 3.5 km above the ground.
The main part of the dust plumes is located between 1 and 2 km height on both days.The values of δ L P are above 0.25 in that height range.The β P values of the dust plume are lower below 1 km height compared to what we find for β above 1 km above ground.The value of δ L P varies between 0.1 and 0.12 and thus is lower than what we find between 1 and 2 km height.
The dust plumes extend to 2 km height above ground on 2 May 2011 (measurement case 2).We find high values of β P near the surface.Values of 0.3-0.4 for δ L P are higher than depolarization ratios found for case 1.
Case 1 is quite different from case 2 if we look at δ LW P .The values of δ LW P of case 1 are mainly affected by Asian dust that is present above 1 km height above ground.Values for case 2 are mainly influenced by Asian dust near the surface.
The differences of main aerosol height may be one reason why the light-absorption capacity of the aerosols described by these two cases differs.With regard to case 2, there is a higher possibility that pollutants can be mixed during longrange transport.
Figure 13 corroborates our assumption.Figure 13 shows values of δ S P , SSA, and volume particle size distributions for two cases.The values of δ S P and SSA in case 1 show similar values compared to the case of pure dust observed at Dunhuang.Values of these parameters are different from those at Dunhuang for case 2. Particularly, the values of SSA of case 2 are significantly lower than SSA values at Dunhuang.Moreover, SSAs at 1020 nm are lower than those at 870 nm for case 2. This differences of the absolute values and the differences of the spectral behavior may be caused by the mixing of dust with light-absorbing pollutants, such as black carbon, when we take account of the vertical distribution of the dust plume (see Fig. 12).The differences of Å A between case 1 and case 2 (Table 5) corroborate this observation.The value of Å A in case 1 is 2.12, which points to pure dust.In contrast, the value of Å A of case 2 is 1.58, which is more likely representative of a mixture of dust with pollutants (Russell et al., 2010).
Figure 14 shows backward trajectories.The transport pattern is different for case 1 and case 2. The main portions of the dust plumes of case 1 passed at heights above 1.5-2 km over source regions of major pollution emissions in China and the Korean Peninsula, i.e., above the PBL, before the plumes arrived over Osaka.Since most of the pollution resides in the PBL the possibility is low that the main dust layer mixed with pollutants.
In contrast to case 1, case 2 indicates that pollutants were mixed into the Asian dust layers while they were transported near the surface and within the PBL over industrialized areas and before they arrived over Osaka.Shin et al. (2015) reported that more pollution particles can be mixed into dust plumes if these plumes are transported at low altitude above ground.

Retrieval of dust ratio using depolarization ratio derived by sun-sky radiometer
The concept of separating lidar backscatter signals resulting from the contribution of dust particles in mixed dust plumes to the total backscatter signals using δ p has already been applied to lidar measurements (Shimizu et al., 2004;Noh, 2014;Tesche et al., 2009).In the case of AERONET sun-sky radiometer data R D is calculated from Eq. (2) (see Sect. 2) and the use of δ S P at 1020 nm.The retrieved R D is compared with CMF vc .The comparison between R D retrieved from AERONET data and CMF vc allows us to distinguish between non-dust coarse-mode particles and dust.
Figure 15 shows the correlation between R D and CMF vc in terms of R 2 .We find a comparably high correlation be-  tween R D and CMF vc .Values are 0.72, 0.95, 0.77, and 0.93 at Seoul, Kongju, Gosan, and Osaka, respectively.R D describes the ratio of dust particles to other types of non-spherical aerosols in the atmosphere.Unlike R D , CMF vc considers the size of particles and is not correlated to the shape of the particles.Since most of the dust particles belong to the CMF, the value of CMF vc increases with the dust ratio.However, the coarse mode of a particle size distribution includes not only dust but also large particles that are generated by physical and chemical reactions, e.g., coagulation, condensation processes, and hygroscopic growth.
Figure 15 shows that CMF vc is on average 0.12-0.17higher than R D at the four observation sites, which implies that non-dust particles are present; i.e., even though dust particles for the most part belong to the CMF of particle size distributions, not all coarse-mode particles are dust particles.The average values of CMF vc and the differences between R D and CMF vc for each of the six groups are listed in Table 1.
R D of groups 1-5 is 0.11-0.19lower than CMF vc at the four sites.The differences between R D and CMF vc are lower for group 6.We find 0.4, 0.4, and 0.3 at Kongju, Gosan, and Osaka, respectively.
We find a similar difference for the Dunhuang site (see Table 3).The average values of R D and CMF vc are 0.97 ± 0.02 and 0.94 ± 0.01, respectively, except for the case of 8 April.This means that most of the coarse-mode particles are composed of dust particles.However, pollution and/or biomass burning particles can be injected into the dust plume during transport from the source region.These particles may contribute to the coarse mode of the volume particle size distribution.If the dust plume is transported at low altitude above ground, there is an increased possibility that dust mixes with other aerosols (Shin et al., 2015).

Figure 1 .
Figure 1.Map of the observation sites.Measurements with AERONET sun-sky radiometer and lidar were performed at Seoul, Kongju, Gosan, and Osaka.AERONET sun-sky radiometer measurements were made at Dunhuang.

Figure 3 .
Figure 3. Aerosol optical depth (τ ) at 500 nm retrieved from AERONET sun-sky radiometer measurements (black squares) and linear particle depolarization ratio derived at 1020 nm from sunsky radiometer data (δ S P ; blue open circles).(a) Data of the four sites taken from 2010 to 2014; (b) data taken at Seoul for two years (2012 and 2013); (c) data taken at Kongju in spring 2012; (d) data taken at Gosan during four years (2011-2014), mostly during spring; (e) data taken from 2010 to 2014 at Osaka.

Figure 5 .
Figure 5.The average values of δ S P at five wavelengths (440, 532, 670, 875, and 1020 nm) and δ CL P at 532 nm divided by the range of the differences between δ S P and δ CL P at 532 nm (δ CL P,532 −δ S P,532 ).The symbol of δ CL P at 532 nm is shown as open figure and in larger size.

Figure 6 .
Figure 6.The average value of the volume size distributions derived by AERONET sun-sky radiometer measurements divided by the range of the differences between δ S P and δ CL P at 532 nm (δ CL P,532 − δ S P,532 ).

Figure 7 .
Figure 7.The average value of the δ S P at 440, 675, 870, and 1020 nm for each group.Each group is distinguished by color: black (group 1), red (group 2), blue (group 3), pink (group 4), gray (group 5), and orange (group 6).The average values of the δ CL P at 532 nm are shown as open circles and the same color as the values of δ S P .

Figure 12 .
Figure12.Separation of data of group 6 at Osaka on the basis of case 1 (black) and case 2 (red).Shown are (a) backscatter coefficients, (b) linear particle depolarization ratios (δ L P ), and (c) weighted linear particle depolarization ratios (δ LW P ).

Table 2 .
Averaged volume concentration of the fine (Cv f ) and the coarse (Cv c ) modes and the ratio (Cv c / Cv f ) for the six groups.

Table 4 .
Volume concentration of the fine (Cv f ) and the coarse (Cv c ) modes and the ratio (Cv c / Cv f ) at Dunhuang.
1 Averaged for all data; 2 averaged except 8 April data.