Atmos. Meas. Tech.

, 12, 3789–3803, 2019

https://doi.org/10.5194/amt-12-3789-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

Aerosol-type classification based on AERONET version 3

inversion products
Sung-Kyun Shin1,a , Matthias Tesche1,b , Youngmin Noh2 , and Detlef Müller1
1 School of Physics, Astronomy and Mathematics, University of Hertfordshire, Hatfield, UK
2 Department of Environmental Engineering, Pukyong National University, Busan, Republic of Korea
a now at: Department of Atmospheric Particulate Matter Research, Seoul Institute of Technology,

Seoul, Republic of Korea

b now at: Leipzig Institute for Meteorology, Leipzig University, Leipzig, Germany

Correspondence: Youngmin Noh (nym@pknu.ac.kr)

Received: 26 March 2019 – Discussion started: 18 April 2019

Revised: 25 June 2019 – Accepted: 26 June 2019 – Published: 11 July 2019

Abstract. This study proposes an aerosol-type classifica- 1 Introduction

tion based on the particle linear depolarization ratio (PLDR)
and single-scattering albedo (SSA) provided in the AErosol Atmospheric aerosol particles play an essential role in the
RObotic NETwork (AERONET) version 3 level 2.0 inver- global climate system by affecting the Earth’s radiation bud-
sion product. We compare our aerosol-type classification get (Stocker et al., 2013). Aerosol particles directly interact
with an earlier method that uses fine-mode fraction (FMF) with solar and terrestrial radiation through absorption and
and SSA. Our new method allows for a refined classification scattering of radiation. Aerosols also can act as cloud con-
of mineral dust that occurs as a mixture with other absorb- densation nuclei and ice-nucleating particles whereby they
ing aerosols: pure dust (PD), dust-dominated mixed plume alter cloud properties (i.e. cloud albedo, cloud lifetime) and
(DDM), and pollutant-dominated mixed plume (PDM). We lead to a change in precipitation efficiency (Twomey, 1974).
test the aerosol classification at AERONET sites in East Radiative forcing is a useful parameter in quantifying the
Asia that are frequently affected by mixtures of Asian dust radiative effect of aerosols on climate change. Estimates of
and biomass-burning smoke or anthropogenic pollution. We aerosol radiative forcing require information on the amount
find that East Asia is strongly affected by pollution parti- of aerosol loading as well as aerosol properties (Bellouin
cles with high occurrence frequencies of 50 % to 67 %. The et al., 2013). The sign and the magnitude of the aerosol ra-
distribution and types of pollution particles vary with lo- diative forcing depends on aerosol characteristics (Stocker
cation and season. The frequency of PD and dusty aerosol et al., 2013). Despite an unprecedented global coverage of
mixture (DDM+PDM) is slightly lower (34 % to 49 %) than atmospheric aerosol information, it is still a challenging task
pollution-dominated mixtures. Pure dust particles have been to assess the aerosol radiative effect accurately. Atmospheric
detected in only 1 % of observations. This suggests that East aerosols are difficult to characterize in time and space due
Asian dust plumes generally exist in a mixture with pollu- to their short lifetime and geographically diverse sources and
tion aerosols rather than in pure form. In this study, we have production mechanisms. Moreover, atmospheric aerosol par-
also considered data from selected AERONET sites that are ticles from different natural and anthropogenic sources also
representative of anthropogenic pollution, biomass-burning frequently mix and undergo ageing processes during trans-
smoke, and mineral dust. We find that average aerosol prop- port, which influences optical and microphysical properties
erties obtained for aerosol types in our PLDR–SSA-based of the bulk aerosol. An improved discrimination of differ-
classification agree reasonably well with those obtained at ent aerosol types by observations, particularly from space,
AERONET sites representative for different aerosol types. increases the accuracy of estimates of the aerosol radiative
impact and is important for the assessment of aerosol disper-
sion modelling (Satheesh and Moorthy, 2005).

Published by Copernicus Publications on behalf of the European Geosciences Union.

3790 S.-K. Shin et al.: AERONET aerosol typing

Aerosol properties inferred from remote-sensing measure- tions (Burton et al., 2012; Russell et al., 2014). Although
ments can be used to classify different aerosol types. Aerosol the combination of polarization lidar and AERONET obser-
optical depth (AOD) and its spectral dependence, which vations might be an ideal option for aerosol classification,
can be characterized by the Ångström exponent, are typ- there are few AERONET stations that are equipped with this
ically used for aerosol classification from passive remote- additional instrumentation, i.e. well-calibrated polarization-
sensing observations. These parameters, together with the sensitive micro-pulse lidar or aerosol polarization lidar.
fine- or coarse-mode fraction of the aerosol size distribu- Version 3 of the AERONET retrieval methodology pro-
tion, allow for a quantification of aerosol loading and aerosol vides us with spectral PLDR as a standard inversion product.
size (Schuster et al., 2006). Dominant aerosol types can In this study, we use the information on the shape (PLDR)
then be inferred by using additional information on possi- and light-absorbing properties (SSA) of particles for refined
ble source regions or transport pathways (Boselli et al., aerosol typing that can be obtained from using standard
2012; Shin et al., 2019). The use of spectral absorbing char- AERONET measurements. We retrieve the dominant aerosol
acteristics of aerosols adds additional information on parti- type at selected AERONET sites in East Asia and investigate
cle type (Giles et al., 2012). For instance, Kim et al. (2007) their seasonal variation and optical properties.
used the fine-mode fraction (FMF) from the Moderate Res- We describe our methodology in Sect. 2. In Sect. 3, we
olution Imaging Spectrometer (MODIS) to determine par- present and discuss our results. A summary and conclusions
ticle size and then used the aerosol index (AI) from the of this work are presented in Sect. 4.
Ozone Monitoring Instrument (OMI) to determine the light-
absorbing characteristics of aerosols. Other studies have pro-
posed using properties related to particle size and absorption 2 Data and methodology
to determine the dominant aerosol type, based on AErosol
2.1 Parameters
RObotic Network (AERONET; Holben et al., 1998) obser-
vations (Lee et al., 2010; Russell et al., 2010; Giles et al., Polarization lidar allows us to infer PLDR from measure-
2012). AERONET provides a global, long-term aerosol data p
ments of the particle backscatter coefficient (βλ ) in different
set that includes spectral AOD, complex refractive index, planes of polarization compared to the plane of polarization
particle size distribution, single-scattering albedo (SSA), and of the emitted linearly polarized laser light:
absorption aerosol optical depth. The information on particle
size and light-absorbing properties enables the identification β p,⊥
of major aerosol types such as mineral dust, anthropogenic δp = . (1)
β p,k
pollution, biomass-burning smoke, and mixtures of aerosol
particles. Lee et al. (2010) used SSA at 440 nm and FMF of AERONET instruments measure direct solar radiation and
AOD at 550 nm from AERONET to determine the absorbing sky radiation. The measurements are automatically anal-
properties and dominant size mode of particles (i.e. fine or ysed using the AERONET inversion algorithm (Dubovik
coarse mode) for aerosol classification. Russell et al. (2010) and King, 2000; Dubovik et al., 2000). The retrieved
classified aerosol types based on the absorption Ångström aerosol products are stored in the AERONET database
exponent. Giles et al. (2012) used SSA, absorption Ångström (https://aeronet.gsfc.nasa.gov/, AERONET, 2019). In the
exponent, Ångström exponent, and FMF from AERONET to AERONET retrieval, the elements F11 (λ) and F22 (λ) of the
infer dominant aerosol types. More detailed and quantitative Müller scattering matrix (Bohren and Huffman, 1983) are
information useful for aerosol classification can be obtained computed from the particle size distribution and the complex
from active aerosol remote sensing with lidar, particularly refractive index that can be inferred from AERONET version
when we include information on the particle linear depolar- 3 level 2.0 data products. The element F11 (λ) is proportional
ization ratio (PLDR or δ p ), which is a very sensitive param- to the flux of scattered light in the case of unpolarized inci-
eter with respect to particle shape (Shin et al., 2019). Values dent light while F22 (λ) strongly depends on the angular and
of PLDR between 0.30 and 0.35 represent nonspherical par- spectral distribution of the radiative intensity (Bohren and
ticles (pure mineral dust, volcanic ash), while values close to Huffman, 1983). From the elements F11 (λ) and F22 (λ) at the
zero indicate the presence of spherical particles. Values in be- scattering angle 180◦ , δ p is computed as
tween these two extremes characterize mixtures of spherical 1 − F22 (λ, 180◦ )/F11 (λ, 180◦ )
p
and nonspherical particles. Measurements of PLDR can be δλ = . (2)
1 + F22 (λ, 180◦ )/F11 (λ, 180◦ )
used to estimate the contribution of these two fundamentally
different particle types (Freudenthaler et al., 2009; Tesche AERONET version 3 level 2.0 inversion products include
et al., 2009; Burton et al., 2012, 2014, 2015; Shin et al., 2015, PLDRs and lidar ratios at 440, 670, 870, and 1020 nm. The
2019). parameter sphericity, which was provided in the AERONET
Several studies have determined aerosol types in which version 2 inversion product, has been discontinued in ver-
lidar-derived PLDR was combined with aerosol optical and sion 3. We hence base this study on PLDR. This approach
microphysical properties inferred from AERONET observa- has the additional advantage that PLDR can also be obtained

Atmos. Meas. Tech., 12, 3789–3803, 2019 www.atmos-meas-tech.net/12/3789/2019/

S.-K. Shin et al.: AERONET aerosol typing 3791

from independent measurements, e.g. with aerosol lidar. Noh 2.3 Correlation between PLDR and other properties
et al. (2017) report that PLDR from the AERONET inversion
products shows high correlation with lidar-derived values. To investigate whether PLDR can be used as the base param-
Shin et al. (2018) discuss AERONET-retrieved PLDR from eter for aerosol-type classification, we analysed how SSA
different source regions. The authors find that PLDRs at 870 and volume particle size distribution vary with respect to
and 1020 nm are likely to be the two most reliable quanti- different intervals of PLDR. These parameters have been
ties. The authors based this finding on comparison to litera- used for aerosol-type classification in previous studies (Kim
ture values obtained from lidar observations of pure dust par- et al., 2007; Lee et al., 2010). Figure 1 shows clearly distin-
ticles which show that AERONET-derived PLDRs at lower guishable patterns of spectral SSA for different intervals of
wavelengths are systematically lower than lidar observations PLDR. The increase in SSA with increasing wavelength for
at 532 nm. PLDR > 0.28 is characteristic for mineral dust (Giles et al.,
The SSA is the ratio of the aerosol scattering coefficient to 2012). In contrast, the decrease in SSA with increasing wave-
the aerosol extinction coefficient and is generally used to de- length for PLDR< 0.10 is in line with anthropogenic pollu-
fine the absorbing characteristics of the scattering particles. tion and biomass-burning smoke (Giles et al., 2012). In addi-
In this study, we follow Lee et al. (2010) and use SSA to dis- tion, SSA at 1020 nm shows the biggest range of values for
tinguish between non-dust particles that are non-absorbing different PLDR intervals. Values are in the range between
(NA), weakly absorbing (WA), moderately absorbing (MA), 0.91 and 0.95 for PLDR < 0.10 and between 0.96 and 0.99
and strongly absorbing (SA). for higher PLDRs. In comparison, these differences are very
small at 440 nm. The composition of mineral dust often in-
cludes iron oxides that lead to strong light absorption at short
2.2 Data selection wavelengths (Derimian et al., 2008). The main light absorber
of anthropogenic pollution is black carbon or soot, which ex-
hibits light absorption throughout the entire solar spectrum
We selected AERONET sites in East Asia to develop our (Bergstrom et al., 2002). Cases with considerable contribu-
methodology. These sites are in a region that is influenced tion of mineral dust as identified by high PLDR values there-
by both natural and anthropogenic aerosols such as an- fore show relatively low SSA at 440 nm compared to longer
thropogenic particles from fossil fuel combustion, biomass- wavelengths.
burning smoke, and mineral dust (Nakata et al., 2016). We Figure 2 shows that the variation in the fine-mode and
combine the stations Beijing and Xianghe into what we the coarse-mode parts of the size distribution with respect
denote as Greater Beijing. The stations Seoul and Yon- to PLDR is also clearly distinguishable. The fine-mode par-
sei_University are considered representative for Seoul. Os- ticles dominate at lower PLDR and their contribution de-
aka is represented by the stations Osaka and Shirahama. We creases with increasing PLDR. Conversely, the contribution
considered Gosan_SNU to represent regional background of the coarse mode to the total volume concentration in-
aerosols in East Asia as there are few anthropogenic sources creases as PLDR increases. It is less than 0.5 for PLDR <
near this site. The AERONET inversion is only performed for 0.06 and higher than 0.9 for PLDR > 0.30.
measurements with a 440 nm AOD larger than 0.4 (Dubovik We know from lidar measurements at 532 nm that values
et al., 2006). For the aerosol-type classification we use the of PLDR> 0.30 indicate nearly pure dust particles (Freuden-
AERONET level 2.0 version 3 inversion products of particle thaler et al., 2009; Shin et al., 2018), whereas PLDR < 0.08
size distribution, PLDR, and SSA available as of February is representative for marine aerosols, anthropogenic pollu-
2018. This means that the time series considered here cover tion, or fresh biomass-burning smoke (Burton et al., 2012).
the time frame from the beginning of measurements at the re- External mixtures of mineral dust and nonspherical particles
spective sites up to between the end of 2016 and the middle reveal 532 nm PLDRs in the range from 0.08 to 0.30 (Tesche
of 2017. et al., 2009, 2011; Gross et al., 2011). Consequently, PLDR
We considered other AERONET stations to evaluate the can be used to calculate the contribution of dust to the par-
results of our aerosol classification. We followed earlier ticle backscatter coefficient for an external aerosol mixture.
AERONET studies on aerosol classification (Dubovik et al., Shimizu et al. (2004) and Tesche et al. (2009) used polariza-
2001; Gobbi et al., 2007; Giles et al., 2012) to select stations tion lidar measurements to define the dust ratio (Rd ) as
that are generally considered to be representative for (1) an- p
p

thropogenic particles: Goddard Space Flight Centre (GSFC, δ p − δnd 1 + δd
Rd = p p
. (3)
USA), Ispra (Italy), and Mexico City (Mexico); (2) biomass- δd − δnd (1 + δ p )
burning smoke: Alta Floresta (Brazil), Mongu (Zambia), and
p p
Abracos Hill (Brazil); and (3) Saharan dust: Capo Verde Here, δnd and δd indicate the PLDR of non-dust and pure
(Cabo Verde), Banizoumbou (Niger), and Dakar (Senegal). dust particles, respectively. The corresponding values can be
Detailed information on the AERONET sites considered in determined from lidar or AERONET observations (Burton
this study is provided in Table 1. et al., 2014; Shin et al., 2018). We apply the Rd retrieval

www.atmos-meas-tech.net/12/3789/2019/ Atmos. Meas. Tech., 12, 3789–3803, 2019

3792 S.-K. Shin et al.: AERONET aerosol typing

Table 1. Geographical and retrieval information for each AERONET site used in this study, along with some references for each aerosol
type. The number of cases refers to observations for which SSA and PLDR are available in the AERONET database.

Site Latitude (◦ ) Longitude (◦ ) Elevation (m) Years Cases

Greater Beijing
Beijing 39.98 116.38 92 2001–2017 2713
Xianghe 39.75 116.96 36 2001–2017 4827
Seoul
Seoul_SNU 37.45 126.95 116 2002–2017 821
Yonsei_University 37.56 126.94 97 2011–2017 1367
Gosan_SNU 33.29 126.16 72 2001–2016 531
Osaka
Osaka 34.65 135.59 50 2000–2016 700
Shirahama 33.69 135.36 10 2000–2016 904
Dust
Banizoumbou 13.55 2.67 274 1995–2018 4217
Capo_Verde 16.73 −22.94 60 1994–2018 1612
Dakar 14.39 −16.96 21 1996–2018 3676
Biomass burning
Abracos_Hill −10.76 −62.36 259 1999–2015 392
Mongu −15.25 23.15 1047 1995–2010 1276
Alta_Floresta −9.87 −56.10 277 1993–2018 817
Anthropogenic
GSFC 38.99 −76.84 87 1993–2018 1237
Ispra 45.80 8.63 235 1997–2018 1114
Mexico_City 19.33 −99.18 2268 1999–2017 692

to AERONET measurements at 1020 nm. We use values of straightforward and size-independent identification of non-
p p
δnd = 0.02 and δd = 0.30 for pure Asian dust (Shin et al., spherical dust particles in a mixture while it has to be as-
p p
2018). When PLDR is lower than δnd or higher than δd , Rd is sumed that dust is constrained to the coarse mode when us-
set to 0 or 1, respectively. ing FMF for aerosol classification. However, dust might also
be present in the fine mode (Mamouri and Ansmann, 2014,
2.4 Aerosol classification method 2017). Lee et al. (2010) used a threshold of FMF < 0.4 to
identify mineral dust. Jones and Christopher (2007) reported
Atmospheric aerosols are typically classified into four ma- annual mean values of MODIS FMF for mineral dust of
jor types according to their size and light-absorbing prop- 0.45 ± 0.05. Shin et al. (2019) compared coarse-mode AOD
erty. For instance, Lee et al. (2010) used FMF and SSA provided by AERONET to dust AOD retrieved with the use
from the AERONET version 2 level 2.0 inversion products to of PLDR and showed that the former tends to overestimate
classify aerosols into black carbon (BC), mineral dust, non- the contribution of mineral dust to AOD. As a consequence,
absorbing (NA) particles, and mixed particles. In our aerosol FMF or fine-mode AOD, when used as a proxy for identify-
classification method, we consider the contribution of non- ing non-dust aerosols, would lead to a systematic underesti-
spherical particles (i.e. mineral dust) to the aerosol mixture mation of the contribution of non-dust particles to the total
based on thresholds of the dust ratio Rd as obtained from aerosol plume. Further benefits of using PLDR rather than
the shape-dependent parameter PLDR and use SSA to distin- FMF for aerosol-type classification are discussed later.
guish absorbing from non-absorbing aerosols. The reason for Figure 3 shows a flowchart of our aerosol classification
using Rd instead of FMF, which has been used, in Schuster method. The contribution of dust in the aerosol mixture is
et al. (2006) and Lee et al. (2010), is that the former pro- determined based on thresholds of Rd . Previous studies re-
vides a clearer separation between dust and non-dust par- ported PLDR values for Asian dust of 0.08 to 0.35 at 532 nm
ticles. The sensitivity of Rd to particle shape allows for a (Murayama et al., 2004; Shimizu et al., 2004; Shin et al.,

Atmos. Meas. Tech., 12, 3789–3803, 2019 www.atmos-meas-tech.net/12/3789/2019/

S.-K. Shin et al.: AERONET aerosol typing 3793

Figure 1. The mean spectra of SSA for different ranges of PLDR at the East Asian AERONET sites considered in this study.

2015). Asian dust generally mixes with other aerosol types 1998). The SSA values for aerosol mixtures (e.g. BC with
during long-range transport, which leads to a wide range in sulfate) vary, depending on the relative humidity and mixing
PLDR. Shin et al. (2015) used 0.08 at 532 nm as the thresh- ratio (e.g. 0.8 at 70 % RH and 0.05 BC/sulfate mixing ra-
old value to identify a contribution of mineral dust in mixed tio for an internal mixture; see Fig. 3 in Wang and Martin,
aerosol plumes. Shimizu et al. (2004) defined 0.10 at 532 nm 2007). Dubovik et al. (2002) report SSA at 1020 nm of 0.83–
as the threshold value for the classification of mixed mineral 0.95, 0.78–0.91, 0.97–0.99, and 0.97 for urban/industrial pol-
dust. In this study, we define pollution particles to show Rd < lutants, biomass burning, desert dust, and marine aerosol, re-
0.17 (equivalent to 1020 nm PLDR< 0.06) and mineral dust spectively. We note that spherical non-dust particles which
to show Rd > 0.89 (equivalent to 1020 nm PLDR > 0.28). were classified by Rd could contain biomass-burning aerosol,
Values of Rd between 0.17 and 0.89 are considered to repre- anthropogenic aerosols, or both. For simplicity and because
sent mixtures of mineral dust and pollution aerosols. We used AODs during purely marine conditions are generally too low
a threshold of Rd = 0.53 (equivalent to 1020 nm PLDR = to yield the AERONET level 2.0 inversion products used in
0.16) to further separate between pollution-dominated mixed this study, we refer to these particles as pollution. We now
aerosol (PDM, Rd = 0.17–0.53) and dust-dominated mixed assume that pollution with higher SSA consists mostly of
aerosol (DDM, Rd = 0.53–0.89). The light-absorbing prop- NA, whereas particles with lower SSA indicate that NA is
erties and size of PDM are likely to be closer to the charac- mixed with BC. For this reason, an SSA threshold of 0.95
teristics of pollution particles than to DDM. In the same way, is used to identify NA and to mark the upper limit for pol-
the light-absorbing properties and size for DDM are more lution aerosols. Depending on 1020 nm SSA and adapting
likely to resemble the characteristics of pure dust. the threshold values of Lee et al. (2010), absorbing aerosols
SSA at 440 nm is usually used to identify dust particles are then further divided into weakly absorbing (WA, SSA
(Kim et al., 2007; Lee et al., 2010). Here, we use SSA at = 0.90–0.95), moderately absorbing (MA, SSA = 0.85–0.9),
1020 nm as we have already classified mineral dust and non- and strongly absorbing (SA, SSA < 0.85). For reference,
dust particles based on Rd . This means that we can make biomass-burning aerosol generally contains a larger fraction
use of the better contrast that we find between black carbon of BC (and often brown carbon as well) compared to anthro-
and dust particles when using SSA at 1020 nm, and we can pogenic aerosol (Washenfelder et al., 2015), and thus is likely
use this contrast for our threshold-based identification of dif- to fall into the more absorbing categories.
ferent absorbing particle types and their mixtures. The SSA
at 1020 nm of water-soluble aerosols, including sulfates, is
close to unity (Hess et al., 1998), whereas SSA is close to
zero for BC (0.07 at 1020 nm; Haywood and Ramaswamy,

www.atmos-meas-tech.net/12/3789/2019/ Atmos. Meas. Tech., 12, 3789–3803, 2019

3794 S.-K. Shin et al.: AERONET aerosol typing

(sector G). This leads to an overall more comprehensive

aerosol classification with respect to dusty mixtures.
Figure 5 shows the mean volume particle size distribution
and SSA of the seven sectors. The optical and microphysi-
cal properties of the aerosols in each sector are listed in Ta-
ble 2. The volume particle size distribution and SSA for the
two dust sectors A and B are markedly different. Sector A
shows a mean FMF of 0.31 ± 0.18 while this value increases
to 0.41 ± 0.17 in sector B. Additionally, the SSA at 1020 nm
in sector A is 0.99 ± 0.01 while SSA in sector B drop to
0.97 ± 0.01. Although both sectors C and D are classified as
mixed aerosols, particle size distribution of sector C is char-
acterized by a more dominant coarse mode than particle size
distribution in sector D.
The difference in the values of optical and microphysi-
cal properties of aerosol in sectors E, F, and G (pollution
dominated) is even more pronounced. Values of FMF are
0.71 ± 0.09, 0.80 ± 0.08, and 0.90 ± 0.05, respectively. SSA
is 0.96±0.01, 0.94±0.01, and 0.92±0.01, respectively. The
spectral dependence of SSA in sector E is also distinguish-
able from that of SSA in sectors F and G. SSA increases with
increasing wavelength in sector E. Such a spectral behaviour
is a characteristic feature of mineral dust. The SSA decreases
with increasing wavelength in sector G. We find little spec-
tral variation in SSA in sector F. Sectors F and G are more
likely to represent pollution particles, whereas sector E more
likely contains mixtures of pollution and mineral dust.
The volume particle size distribution and spectral SSA in
Figure 2. The mean volume particle size distributions for different sector A (pure dust based on PLDR) is clearly different from
ranges of PLDR for the East Asian AERONET sites considered in the ones in the other sectors. Sector A contains coarse-mode-
this study. dominated particle size distributions (FMF = 0.31 ± 0.18)
and shows a spectral SSA that resembles the one of mineral
dust. Similarly, the spectral characteristics of SSA of aerosols
in sectors B, C, and E, which are classified as dust-dominant
3 Results mixtures, are characteristic for dust particles. However, SSA
at 1020 nm is lower (0.96–0.97) compared to SSA at 1020 nm
3.1 FMF versus Rd for aerosol-type classification in sector A. SSA at 1020 nm varies around 0.94 in sectors D
and F. These sectors were classified as pollution-dominated
Figure 4 shows a 2-D histogram of the relation between mixtures (based on PLDR). The volume size distributions
FMF and Rd . Desert dust predominately consists of coarse- for sectors D and F show that both coarse- and fine-mode
mode particles, whereas combustion-produced particles are particles are present in these mixtures. Sector G resembles
predominantly found in the FMF of particle size distribu- the characteristics of pollution particles (high FMF and low
tions. Nevertheless, nonspherical dust particles might also SSA).
be present in the fine mode – in which case FMF would no As in our earlier study (Shin et al., 2019), we con-
longer be ideally suited to represent dust in the particle size clude that FMF with its relation to particle size might be
distribution. Rd is close to zero for spherical particles such too ambiguous a parameter to distinguish aerosol types in
as anthropogenic or smoke particles and increases with in- mixed dust-pollution plumes. In contrast, PLDR and Rd
creasing dust contribution. The dominant aerosol type in the are related to particle shape, and thus are likely to be bet-
aerosol mixture can be distinguished based on knowledge ter suited and physically more meaningful parameters for
of FMF and Rd as the parameters refer to particle size and aerosol-type classification in which mineral dust is part of
shape, respectively. We find a strong negative correlation of the aerosol mixture. Our method is particularly useful for ob-
R 2 = 0.77 for FMF versus Rd . The use of Rd and FMF al- servations in East Asia were mineral dust is almost always
lows us to separate between pure dust (PD, sector A), dust- mixed with pollution. We therefore propose a refined catego-
dominated mixture (DDM, sectors B, C, and E), pollution- rization into pure dust, dust-dominated mixtures, pollution-
dominated mixture (PDM, sectors D and E), and pollution dominated mixtures, and pollution.

Atmos. Meas. Tech., 12, 3789–3803, 2019 www.atmos-meas-tech.net/12/3789/2019/

S.-K. Shin et al.: AERONET aerosol typing 3795

Figure 3. Flow chart of the aerosol classification based on the dust ratio Rd derived using PLDR at 1020 nm and the 1020 nm SSAs that are
inferred from the inversion of AERONET observations.

Table 2. Average values of AERONET-derived aerosol parameters for the sectors identified in Fig. 5: FMF, AOD-related Ångström exponent
for the wavelength pair 440 and 870 nm (AE), PLDR at 1020 nm, SSA at 1020 nm, total volume concentration (VolC), effective radius (Reff),
and the volume median radius (VMR). For comparison, Lee et al. (2010) classify categories A and B as dust, categories C and D as mixed,
and categories E, F, and G as pollution.

A B C D E F G
Pure dust Dust domi- Dust domi- Pollutant domi- Dust domi- Pollutant domi- Pollution
(PD) nated (DDM) nated (DDM) nated (PDM) nated (DDM) nated (PDM)
N 90 355 520 207 512 3488 6691
FMF 0.31 ± 0.18 0.41 ± 0.17 0.59 ± 0.13 0.62 ± 0.09 0.71 ± 0.09 0.80 ± 0.08 0.90 ± 0.05
AE 0.17 ± 0.08 0.38 ± 0.15 0.70 ± 0.14 0.83 ± 0.11 0.88 ± 0.14 1.17 ± 0.15 1.36 ± 0.19
PLDR 0.28 ± 0.04 0.22 ± 0.03 0.17 ± 0.03 0.13 ± 0.02 0.16 ± 0.02 0.09 ± 0.02 0.02 ± 0.01
SSA 0.99 ± 0.01 0.97 ± 0.01 0.96 ± 0.01 0.94 ± 0.01 0.96 ± 0.01 0.94 ± 0.01 0.92 ± 0.01
VolC 0.52 ± 0.31 0.44 ± 0.27 0.38 ± 0.18 0.40 ± 0.15 0.38 ± 0.21 0.40 ± 0.23 0.50 ± 0.30
Reff 0.73 ± 0.36 0.59 ± 0.33 0.40 ± 0.20 0.43 ± 0.19 0.37 ± 0.15 0.30 ± 0.12 0.28 ± 0.11

3.2 Characteristics of aerosol types over East Asia sites. Aerosols classified as more light absorbing (MA+SA),
are detected most frequently at Greater Beijing (SA: 4 %;
We investigated the regional and seasonal characteris- MA: 15 %) and Osaka (SA: 5 %; MA: 15 %) followed by
tics of aerosol over East Asia. The rates of occurrences Seoul (SA: 2 %; MA: 9 %) and Gosan_SNU (SA: 3 %; MA:
of aerosol types at each AERONET observation site are 4 %). The most frequently detected aerosol types over Seoul
shown in Fig. 6. Regardless of season, pollution particles were NA (24 %), PDM (28 %), and WA (32 %). Differences
(NA+WA+MA+SA) are detected persistently over East Asia in the frequency of different types of pollution particles, i.e.
with a 51 %–66 % occurrence rate when considering the an- the BC-dominated types (MA and SA) and WA and NA,
nual average. Pollution particles are detected most frequently which contain little or no BC, can be explained by the sources
at Seoul and Osaka (66 %). The lowest occurrence rate of of the aerosols. The occurrence rate of MA and SA at Greater
pollution of 51 % was found at Gosan_SNU. The seasonal Beijing (19 %) is larger than at Seoul (11 %). China has be-
variation and specific type of pollution differ for the different

www.atmos-meas-tech.net/12/3789/2019/ Atmos. Meas. Tech., 12, 3789–3803, 2019

3796 S.-K. Shin et al.: AERONET aerosol typing

Xia et al. (2007), Guo et al. (2011), and Jung et al. (2009).
Overall, dust-containing aerosol types (PD, DDM, and PDM)
have the highest occurrence rates during spring with as much
as 82 % of cases for Gosan_SNU in April. This is in line
with the Asian dust season in spring (Guo et al., 2011). PD
is observed at all sites during spring but only with low oc-
currence rates of 1 % to 4 %. PD is also rarely observed dur-
ing the other seasons. The occurrence rates of pollutant par-
ticles are consistently high except during spring. The distri-
butions and types of pollutant particles are rather complex
because of the variety of their sources with respect to loca-
tion and season. Higher rates of SA and MA are detected at
most observation sites during winter, which is likely the re-
sult of increased combustion of coal for domestic heating.
In contrast, the Gosan_SNU site is less influenced by strong
light-absorbing particles during winter compared to the other
observation sites as it is located away from highly industrial-
ized areas. The higher SA and MA occurrence rates detected
during summer at Osaka might be due to regional climate
Figure 4. Two-dimensional histograms of Rd derived using PLDR effects: a high oxidant level from local emissions, high tem-
at 1020 nm and FMF at 550 nm for the considered AERONET sta- perature, and strong thermal insulation support the conver-
tions over East Asia. Seven sectors, A to G, are defined according to sion of volatile organic compounds into secondary organic
the ranges of PLDR and FMF: pure dust (sector A), dust-dominated aerosols (Sano et al., 2016). The aged secondary organic
mixture (sectors B, C, and E), pollution-dominated mixture (sectors aerosols could produce brown carbon, which is significantly
D and E), and pollution (sector G). The colour coding indicates the light absorbing (Liu et al., 2016).
number of observation data in log scale. Another feature to note is the occurrence rate of NA and
WA, which increases during summer at all sites. Sulfate and
nitrate are major contributors to the aerosol loading during
come a major source of BC because of fossil fuel combus- summer over East Asia. Strong solar radiation, high ambi-
tion stemming from industry, domestic heating, and cooking. ent temperature, and high relative humidity during summer
The contribution of BC in pollution thus has increased. In in East Asia enhance photochemical processes. These con-
contrast, water-soluble (hygroscopic) aerosol particles such ditions may have been responsible for the higher summer
as sulfate or nitrate might be dominant in the anthropogenic concentrations of ammonium sulfate and nitrate (Jung et al.,
pollution plumes over Seoul. 2009). The Asian monsoon is another possible reason for
The occurrence rate of pure and polluted dust (PD, DDM, the increased frequency of occurrence of NA and WA dur-
and PDM) over East Asia is slightly lower (34 %–49 %) than ing summer. Increased relative humidity and associated hy-
that of dust-free pollution. PDM is the most frequently de- groscopic particle growth during monsoon season can cause
tected aerosol type (over East Asia) of all types that include changes in aerosol scattering properties (Xia et al., 2007).
dust, i.e. PD, DDM, and PDM. The occurrence rates of PDM Regardless of season, significant amounts of NA and WA are
were 28 %, 28 %, 33 %, and 23 % for Greater Beijing, Seoul, detected at the coastal sites Gosan_SNU and Osaka. The high
Gosan_SNU, and Osaka, respectively. In contrast, the oc- values of NA and WA might be the result of large contribu-
currence rate for PD was around 1 % at most sites. Central tions of maritime aerosol, which is strongly light scattering
Asian dust has usually travelled over densely populated and as well as highly hygroscopic.
highly industrialized areas of China before it is detected. That
means, as the result of transport path and time dust has usu-
3.3 Dominant aerosol types at representative aerosol
ally mixed with pollutants before being detected (Sun et al.,
source regions
2005; Shin et al., 2015). The low occurrence rate for PD and
relatively higher occurrence rate for PDM and DDM over
East Asia corroborate this assumption, i.e. that dust particles In order to further test our method, we used it to determine
over East Asia are typically mixed with anthropogenic pol- the dominant aerosol types at well-characterized AERONET
lution or that any stratification of the different aerosol types sites representative for anthropogenic pollution, biomass-
will show the optical signature of a mixture in the column- burning particles, and mineral dust (Dubovik et al., 2002;
integrated AERONET measurements. Giles et al., 2012; Choi et al., 2016). Details on the selected
The seasonal variation in each aerosol type retrieved with sites together with the occurrence rates of different aerosol
our aerosol classification method corroborates the findings of types are given in Table 3.

Atmos. Meas. Tech., 12, 3789–3803, 2019 www.atmos-meas-tech.net/12/3789/2019/

S.-K. Shin et al.: AERONET aerosol typing 3797

Figure 5. Mean volume size distributions (a) and spectral SSA (b) for the aerosol-type sectors identified in Fig. 4. Black lines relate to pure
dust (solid lines and circles) and dust-dominated mixtures (dashed lines, open circles, and triangles) while grey lines mark pollution (solid
lines and squares) and pollution-dominated mixtures (dashed lines, open squares, and diamonds).

Table 3. Occurrence rate (in percent) of the aerosol types identified with our method at the AERONET sites given in the lower half of Table 1.

Cases PD DDM PDM NA WA MA SA

Anthropogenic
GSFC 1237 0 0 1 34 47 16 2
Ispra 1114 0 0 6 32 37 19 5
Mexico_City 692 0 0 2 10 24 34 29
Biomass burning
Alta_Floresta 817 0 0 1 4 34 42 20
Mongu 1276 0 0 0 0 0 4 96
Abracos_Hill 392 0 0 0 4 34 41 21
Saharan dust
Capo_Verde 1612 81 16 2 0 0 0 0
Banizoumbou 4217 65 30 4 0 0 0 0
Dakar 3676 60 33 0 0 0 0 0

The dominant aerosol type at dust sites was PD with oc- Banizoumbou) and summer (JJA, 74 % and 72 %), decreases
currence rates of 64 % to 81 %. PD was most frequently de- in autumn (SON, 55 % and 63 %), and has minimum contri-
tected at Cabo Verde with an 81 % occurrence rate. Addi- butions in winter (DJF, 37 % and 49 %). The decrease in PD
tionally, notable occurrence rates for DDM (16 % to 35 %) comes with an increase in DDM from spring and summer to
and PDM (0 % to 4 %) are also found at the dust sites. This autumn and winter (Dakar: 25 % in MAM, 21 % in JJA, 40 %
suggests that AERONET sites in or close to source regions in SON, and 45 % in DJF; Banizoumbou: 11 % in MAM,
of mineral dust are frequently affected by local pollution 23 % in JJA, 36 % in SON, and 40 % in DJF). In addition,
or other aerosols from sources upwind. The highest occur- PDM has a maximum in winter (17 % at Dakar and 10 % at
rence rate of 35 % for mixed dust types (DDM+PDM) was Banizoumbou) with values between 0 % and 5 % during the
found at Dakar where local pollution has a much stronger ef- other seasons. Contributions of other aerosol types are negli-
fect on aerosol composition compared to other Saharan sites gible at the two sites throughout the year.
(Petzold et al., 2011). We have considered the AERONET The sites for anthropogenic pollution show an extremely
sites Dakar and Banizoumbou to investigate if our method- high occurrence rate of pollutants of 93 % to 99 %, though
ology can be used to resolve the seasonal cycles of dust and the occurrence rates of the individual subtypes vary from
biomass burning over West Africa (not shown). We find that site to site. Higher occurrence rates of NA and WA were
the two sites are dominated by PD, DDM, and PDM but that found at GSFC (34 % and 47 %) and Ispra (32 % and
the ratio of the three types varies with season. PD contributes 37 %). Conversely, a higher occurrence rate of MA (34 %)
most strongly in spring (MAM, 71 % at Dakar and 88 % at and SA (29 %) was found at Mexico City, which is fre-

www.atmos-meas-tech.net/12/3789/2019/ Atmos. Meas. Tech., 12, 3789–3803, 2019

3798 S.-K. Shin et al.: AERONET aerosol typing

by flaming fires, i.e. at high temperatures. This type of BC

production is prevalent at Mongu, whereas production of pri-
mary organic carbon at lower temperatures (smoldering fires)
is dominant at Alta Floresta (Reid et al., 2005; Choi et al.,
2016). The occurrence rates of NA (4 %) and WA (34) at
Alta Floresta and Abracos Hill are also distinguishable from
the rates at Mongu where less than 0.5 % was detected as NA
or WA. The photochemical formation of secondary inorganic
aerosols due to emissions from biomass burning seems to be
much more frequent at Alta Floresta and Abracos Hill com-
pared to Mongu (Choi et al., 2016).
Figure 7, analogous to Figs. 1 and 2, presents the mean
spectral SSA and volume particle size distributions for the
AERONET sites that are mainly affected by anthropogenic
pollution, biomass-burning smoke, and Saharan dust. The
spectral behaviour of SSA at these sites strongly resembles
that presented by Dubovik et al. (2002), Eck et al. (2009),
Giles et al. (2012), and Shin et al. (2018). Dust particles ex-
hibit strong light absorption at short wavelengths and lower
absorption in the visible and near-infrared wavelength re-
gions (Shin et al., 2018). Fine-mode particles and hygro-
scopic particles such as sulfate show nearly neutral spectral
dependence of SSA and overall stronger light scattering, i.e.
higher SSA (Dubovik et al., 2002). BC has the strongest
light-absorption properties in the near-infrared wavelength
region. Organic and brown carbon exhibit stronger light-
absorption properties at ultraviolet and visible wavelengths
(Eck et al., 2009).
The SSA at sites that represent anthropogenic pollution
likely reflects the spectral features of SSA of fine-mode par-
ticles and hygroscopic particles. The values of SSA at Mex-
ico City are lower than at other anthropogenic sites for the
entire range of wavelengths. Mexico City is frequently in-
fluenced by air pollution that likely consists of more car-
bonaceous aerosols, whereas GSFC and Ispra are affected by
Figure 6. Monthly and annual occurrence rate of the aerosol types
secondary inorganic particles (Choi et al., 2016). Addition-
classified according to this study for AERONET sites in East Asia.
The colour coding refers to the aerosol species introduced in Fig. 3. ally, the spectral dependence of SSA at the biomass-burning
The numbers above the plots refer to the number of observations for sites is similar to that of BC. We note that the values of
which SSA and PLDR are available in the AERONET database. SSA at the Mongu site are much lower than the values of
SSA at the other biomass-burning sites in the entire wave-
length range. Various absorbing aerosols (e.g. black, organic
quently affected by severe air pollution that mostly con- and brown carbon, or secondary inorganic aerosols) can be
sists of fine-mode and carbonaceous aerosols (Choi et al., emitted from biomass burning. The lower SSA at Mongu
2016). Choi et al. (2016) used cluster analysis in combina- might be the result of a higher fraction of BC compared
tion with AERONET data and found that secondary inor- to the other biomass-burning sites. Consequently, the spec-
ganic particles dominate at GSFC. Carbonaceous aerosol has tral dependence of SSA at Alta Floresta and Abracos Hill is
strong light-absorption properties, whereas secondary inor- more neutral with higher values of SSA compared to Mongu.
ganic aerosols such as sulfates predominantly scatter light BC has the strongest light-absorption properties in the near-
(Bond and Bergstrom, 2006). infrared wavelength region, while organic and brown carbon
The frequency distributions of SA and MA are higher at exhibit stronger light-absorption properties at ultraviolet and
the biomass-burning sites. Those sites are considered to be visible wavelengths (Eck et al., 2009). The spectral depen-
affected by mostly light-absorbing particles from combus- dence of SSA at the dust sites is representative of mineral
tion. A high occurrence rate of 96 % of SA was found at dust particles.
Mongu. MA is the most frequently detected type at Alta Figure 7 also compares the spectral SSA at the AERONET
Floresta (42 %) and Abracos Hill (41 %). BC is produced sites to the spectral SSA of the aerosol types defined in our

Atmos. Meas. Tech., 12, 3789–3803, 2019 www.atmos-meas-tech.net/12/3789/2019/

S.-K. Shin et al.: AERONET aerosol typing 3799

Figure 7. Mean spectral SSA (upper row) and mean particle size distributions (lower row) for AERONET sites that are considered to be
representative for different aerosol types as well as for the aerosol types defined in this study (see Fig. 3 for colour coding). Note that the size
distribution for pure mineral dust (red line) has been scaled to 50 % to fit the plot.

classification method (see Figure 3). The SSA for NA is The volume size distributions show a dominance of
higher than SSA at the anthropogenic sites at all wavelengths coarse-mode particles at the dust sites. Coarse-mode parti-
except for the observation at GSFC at 440 nm. The SSA for cles also contribute strongly to the total volume size distri-
WA is similar to values obtained at GSFC and Ispra. We bution for PD and DDM, whereas a lower contribution of
conclude from this that NA – in view of the definition of coarse-mode particles is found for PDM. This result is in
aerosol type – consists to a largest degree of light-scattering line with an increased concentration of anthropogenic pol-
particles. Similarly, WA represents aerosols that contain both lution or biomass-burning smoke which, in the PDM type,
light-scattering and light-absorbing particles with the light- are typically considered to be fine-mode particles (Eck et al.,
scattering contribution being predominant. 1999).
Biomass-burning sites observe light-scattering particles Fine-mode particles contribute most strongly to the to-
and light-absorbing particles, which is reflected in SSA. The tal volume size distribution at biomass-burning sites. Ad-
spectral SSA of MA is similar to the values found at Alta ditionally, the contributions of fine and coarse modes are
Floresta and Abracos Hill except at short wavelengths. Con- distributed evenly in the total volume size distribution for
sequently, MA is an aerosol mixture in which light-absorbing MA and SA. Reid et al. (1998) found from studies in rain-
particles have a stronger impact. The absorbing properties are forest regions of Brazil that fresh smoke particles are sig-
most likely related to the contribution of organic carbon. Fi- nificantly smaller than well-aged smoke particles. We be-
nally, the spectral SSA of SA is closest to the observations lieve that fresh smoke particles contribute significantly at the
at Mongu and suggests that the light- absorbing properties biomass-burning sites, whereas MA and SA detected over
of SA are related to a strong contribution of BC. We note East Asia are affected not only by fine-mode particles but
that the SSA of SA and Mongu are different. SA has been also a considerable amount of coarse-mode particles, in con-
defined based on observations in East Asia where absorbing trast to the source regions of biomass burning. The contri-
particles are likely present in complex mixtures. In contrast, butions of fine-mode and coarse-mode particles to the total
BC is considered to be the dominant aerosol at Mongu. volume size distributions are rather similar for the anthro-
The spectral SSA of DDM and PD is similar to the one pogenic sites. However the contribution of fine-mode parti-
found at the dust sites. However, the SSA of DDM is slightly cles to the total volume size distributions is dominant for NA.
lower than SSA of PD and of the AERONET dust sites
as a result of mixing with pollutant particles. Accordingly,
the SSAs of PDM are even lower than of PD, DDM, and 4 Summary and conclusions
AERONET dust sites as they are defined to feature a larger
contribution of pollutants. We present a methodology that is used for classifying the
dominant types of aerosol particles in mixed aerosol plumes.

www.atmos-meas-tech.net/12/3789/2019/ Atmos. Meas. Tech., 12, 3789–3803, 2019

3800 S.-K. Shin et al.: AERONET aerosol typing

Note that in the case of AERONET measurements, mix- All authors contributed to the discussion of the findings and the
tures might also refer to a vertical stratification of differ- preparation of the paper.
ent aerosol types that shows the column-integrated optical
signature of a mixed plume. We used PLDR and SSA that
are provided in the AERONET version 3 level 2.0 inver- Competing interests. The authors declare that they have no conflict
sion products. Dusty aerosol mixtures were separated ac- of interest.
cording to pure dust, dust-dominated mixtures, pollution-
dominated mixtures, and pollution. The pollutants were clas-
sified as non-absorbing, weakly absorbing, moderately ab- Acknowledgements. We thank the principal investigators and their
staff for establishing and maintaining the AERONET sites used in
sorbing, and strongly absorbing particles based on using SSA
this investigation.
at 1020 nm. The new aerosol typing method provides detailed
information on aerosol mixtures that contain varying levels
of mineral dust. Financial support. This work was supported by a National Re-
We tested our method at East Asian AERONET sites that search Foundation of Korea (NRF) grant funded by the Korean gov-
are frequently affected by various aerosol type. We found that ernment (grant no. NRF-2018R1D1A3B07048047).
aerosols categorized as pollutant (NA+WA+MA+SA) are
most frequently detected over East Asia. The detection rate
was 50 %–67 %. The distribution and types of pollutant vary Review statement. This paper was edited by Alexander
according to the location of sites and season. The frequency Kokhanovsky and reviewed by Patrick Hamill and one anonymous
distribution for PD or dust-containing aerosol (DDM+PDM) referee.
plumes is lower than that of pollutants. Moreover, PD was de-
tected in less than 1 % of the observations at most sites over
East Asia. This suggests that dust over East Asia is almost
always mixed with other types of aerosol.
We compared the results of our aerosol-type classification References
to the aerosol properties obtained from selected AERONET
Aerosol Robotic Network (AERONET): available at: https://
sites which are considered to be representative for different
aeronet.gsfc.nasa.gov/, 9 July 2019.
aerosol types, i.e. anthropogenic pollution, biomass-burning
Bellouin, N., Quaas, J., Morcrette, J.-J., and Boucher, O.: Estimates
smoke, and mineral dust. For anthropogenic pollution, we of aerosol radiative forcing from the MACC re-analysis, At-
found that MA and SA occur most frequently at Mexico mos. Chem. Phys., 13, 2045–2062, https://doi.org/10.5194/acp-
City, whereas GSFC and Ispra are affected most strongly by 13-2045-2013, 2013.
NA and WA. The frequency distribution of SA and MA is Bergstrom, R. W., Russell, P. B., and Hignett, P.: Wavelength de-
higher at the biomass-burning sites. Those sites are affected pendence of the absorption of black carbon particles: Predictions
by mostly light-absorbing particles from combustion. PD is and results from the TARFOX experiment and implications for
the dominant aerosol at dust sites. From the comparison to the aerosol single scattering albedo, J. Atmos. Sci., 59, 568–578,
representative AERONET sites, we conclude that the method 2002.
we use to identify dominant aerosol types over East Asia Bohren, C. and Huffman, D.: Absorbing and scattering of light by
small particles, Wiley, https://doi.org/10.1002/9783527618156,
also leads to reasonable results at other sites. The proposed
1983.
method has the potential to provide improved information on
Bond, T. C. and Bergstrom, R. W.: Light absorption by carbona-
aerosol type in regions where various types of aerosol are ceous particles: An investigative review, Aerosol Sci. Tech., 40,
frequently present in the form of complex mixtures – not 27–67, https://doi.org/10.1080/02786820500421521, 2006.
only over East Asia but also elsewhere on the globe. Appli- Boselli, A., Caggiano, R., Cornacchia, C., Madonna, F., Mona,
cation of our aerosol-typing method to the global AERONET L., Macchiato, M., Pappalardo, G., and Trippetta, S.: Multi
database will provide useful information for the validation of year sun-photometer measurements for aerosol characterization
chemical transport modelling as well as (active) spaceborne in a Central Mediterranean site, Atmos. Res., 104, 98–110,
sensors that provide PLDR observations. https://doi.org/10.1016/j.atmosres.2011.08.002, 2012.
Burton, S. P., Ferrare, R. A., Hostetler, C. A., Hair, J. W., Rogers, R.
R., Obland, M. D., Butler, C. F., Cook, A. L., Harper, D. B., and
Data availability. The data used in this work are freely avail- Froyd, K. D.: Aerosol classification using airborne High Spectral
able through the AERONET portal at https://aeronet.gsfc.nasa.gov/ Resolution Lidar measurements – methodology and examples,
(AERONET, 2019). Atmos. Meas. Tech., 5, 73–98, https://doi.org/10.5194/amt-5-73-
2012, 2012.
Burton, S. P., Vaughan, M. A., Ferrare, R. A., and Hostetler, C.
A.: Separating mixtures of aerosol types in airborne High Spec-
Author contributions. SKS, MT, and YN had the idea for this study.
tral Resolution Lidar data, Atmos. Meas. Tech., 7, 419–436,
SKS and MT performed the data analysis and prepared the figures.
https://doi.org/10.5194/amt-7-419-2014, 2014.

Atmos. Meas. Tech., 12, 3789–3803, 2019 www.atmos-meas-tech.net/12/3789/2019/

S.-K. Shin et al.: AERONET aerosol typing 3801

Burton, S. P., Hair, J. W., Kahnert, M., Ferrare, R. A., Hostetler, pure Saharan dust during SAMUM 2006, Tellus B, 61, 165–179,
C. A., Cook, A. L., Harper, D. B., Berkoff, T. A., Seaman, S. https://doi.org/10.1111/j.1600-0889.2008.00396.x, 2009.
T., Collins, J. E., Fenn, M. A., and Rogers, R. R.: Observa- Giles, D. M., Holben, B. N., Eck, T. F., Sinyuk, A., Smirnov, A.,
tions of the spectral dependence of linear particle depolariza- Slutsker, I., Dickerson, R., Thompson, A., and Schafer, J.: An
tion ratio of aerosols using NASA Langley airborne High Spec- analysis of AERONET aerosol absorption properties and clas-
tral Resolution Lidar, Atmos. Chem. Phys., 15, 13453–13473, sifications representative of aerosol source regions, J. Geophys.
https://doi.org/10.5194/acp-15-13453-2015, 2015. Res., 117, D17203, https://doi.org/10.1029/2012JD018127,
Choi, Y., Ghim, Y. S., and Holben, B. N.: Identification of columnar 2012.
aerosol types under high aerosol optical depth conditions for a Gobbi, G. P., Kaufman, Y. J., Koren, I., and Eck, T. F.: Classification
single AERONET site in Korea. J. Geophys. Res., 121, 1264– of aerosol properties derived from AERONET direct sun data,
1277, https://doi.org/10.1002/2015JD024115, 2016. Atmos. Chem. Phys., 7, 453–458, https://doi.org/10.5194/acp-7-
Derimian, Y., Karnieli, A., Kaufman, Y. J., Andreae, M. O., An- 453-2007, 2007.
dreae, T. W., Dubovik, O., Maenhaut, W., and Koren, I.: The Groß, S., Tesche, M., Freudenthaler, V., Toledano, C., Wiegner, M.,
role of iron and black carbon in aerosol light absorption, Atmos. Ansmann, A., Althausen, D., and Seefeldner, M.: Characteriza-
Chem. Phys., 8, 3623–3637, https://doi.org/10.5194/acp-8-3623- tion of Saharan dust, marine aerosols and mixtures of biomass-
2008, 2008. burning aerosols and dust by means of multi-wavelength
Dubovik, O. and King, M.: A flexible inversion algorithm for re- depolarization and Raman lidar measurements during SA-
trieval of aerosol optical properties from sun and sky radiance MUM 2, Tellus B, 63, 706–724, https://doi.org/10.1111/j.1600-
measurements, J. Geophys. Res., 105, 20673–20696, 2000. 0889.2011.00556.x, 2011.
Dubovik, O., Smirnov, A., Holben, B., King, M., Kaufman, Y., Guo, J.-P., Zhang, X.-Y., Wu, Y.-R., Zhaxi, Y., Che, H.-
Eck, T., and Slutsker, I.: Accuracy assessments of aerosol optical Z., La, B., Wang, W., and Li, X.-W.: Spatio-temporal
properties retrieved from Aerosol Robotic Network (AERONET) variation trends of satellite-based aerosol optical depth in
sun and sky radiance measurements, J. Geophys. Res., 105, China during 1980–2008, Atmos. Environ., 45, 6802–6811,
9791–9806, 2000. https://doi.org/10.1016/j.atmosenv.2011.03.068, 2011.
Dubovik, O., Holben, B., Eck, T. F., Smirnov, A., Kauf- Haywood, J. M. and Ramaswamy, V.: Global sensitivity studies
man, Y. J., King, M. D., Tanré, D., and Slutsker, I.: of the direct radiative forcing due to anthropogenic sulfate and
Variability of Absorption and Optical Properties of Key black carbon aerosols, J. Geophys. Res.-Atmos., 103, 6043–
Aerosol Types Observed in Worldwide Locations, J. 6058, https://doi.org/10.1029/97JD03426, 1998.
Atmos. Sci., 59, 590–608, https://doi.org/10.1175/1520- Hess, M., Koepke, P., and Schult, I.: Optical properties of
0469(2002)059<0590:VOAAOP>2.0.CO;2, 2002. aerosols and clouds: The software package OPAC, B. Am.
Dubovik, O., Holben, B., Eck, T. F., Smirnov, A., Kauf- Meteorol. Soc., 79, 831–844, https://doi.org/10.1175/1520-
man, Y. J., King, M. D., Tanré, D., and Slutsker, I.: 0477(1998)079<0831:OPOAAC>2.0.CO;2, 1998.
Variability of absorption and optical properties of key Holben, B. N., Eck, T. F., Slutsker, I., Tanré, D., Buis, J., Set-
aerosol types observed in worldwide locations, J. At- zer, A., Vermote, E., Reagan, J., Kaufman, Y. J., and Naka-
mos. Sci., 59, 590–608, https://doi.org/10.1175/1520- jima, T.: AERONET – A federated instrument network and data
0469(2002)059<0590:VOAAOP>2.0.CO;2, 2002. archive for aerosol characterization, Rem. Sens. Environ., 66, 1–
Dubovik, O., Sinyuk, A., Lapyonok, T., Holben, B. N., Mishchenko, 16, https://doi.org/10.1016/S0034-4257(98)00031-5, 1998.
M., Yang, P., Eck, T. F., Volten, H., Muñoz, O., and Veihelmann, Jones, T. A. and Christopher, S. A.: MODIS derived fine
B.: Application of spheroid models to account for aerosol particle mode fraction characteristics of marine, dust, and anthro-
nonsphericity in remote sensing of desert dust, J. Geophys. Res.- pogenic aerosols over the ocean, constrained by GOCART,
Atmos., 111, D11208, https://doi.org/10.1029/2005JD006619, MOPITT, and TOMS, J. Geophys. Res., 112, D22204,
2006. https://doi.org/10.1029/2007JD008974, 2007.
Eck, T., Holben, B., Reid, J., Dubovik, O., Smirnov, A., O’Neill, Jung, J., Lee, H., Kim, Y. J., Liu, X., Zhang, Y., Hu, M., and Sugi-
N., Slutsker, I., and Kinne, S.: Wavelength dependence of moto, N.: Optical properties of atmospheric aerosols obtained by
the optical depth of biomass burning, urban, and desert in situ and remote measurements during 2006 Campaign of Air
dust aerosols, J. Geophys. Res.-Atmos., 104, 31333–31349, Quality Research in Beijing (CAREBeijing-2006), J. Geophys.
https://doi.org/10.1029/1999JD900923, 1999. Res., 114, D00G02, https://doi.org/10.1029/2008JD010337,
Eck, T. F., Holben, B. N., Reid, J. S., Sinyuk, A., Hyer, E. 2009.
J., O’Neill, N. T., Shaw, G. E., Vande Castle, J. R., Chapin, Kim, J., Lee, J., Lee, H. C., Higurashi, A., Takemura, T., and Song,
F. S., Dubovik, O., Smirnov, A., Vermote, E., Schafer, J. S., C. H.: Consistency of the aerosol type classification from satellite
Giles, D., Slutsker, I., Sorokine, M., and Newcomb, W. W.: remote sensing during the Atmospheric Brown Cloud–East Asia
Optical properties of boreal region biomass burning aerosols Regional Experiment campaign, J. Geophys. Res., 112, D22S33,
in central Alaska and seasonal variation of aerosol optical https://doi.org/10.1029/2006JD008201, 2007.
depth at an Arctic coastal site, J. Geophys. Res., 114, D11201, Lee, J., Kim, J., Song, C., Kim, S., Chun, Y., Sohn, B., and Hol-
https://doi.org/10.1029/2008JD010870, 2009. ben, B.: Characteristics of aerosol types from AERONET sun-
Freudenthaler, V., Esselborn, M., Wiegner, M., Heese, B., Tesche, photometer measurements, Atmos. Environ., 44, 3110–3117,
M., Ansmann, A., Müller, D., Althausen, D., Wirth, M., and https://doi.org/10.1016/j.atmosenv.2010.05.035, 2010.
Fix, A.: Depolarization ratio profiling at several wavelengths in Liu, J., Lin, P., Laskin, A., Laskin, J., Kathmann, S. M.,
Wise, M., Caylor, R., Imholt, F., Selimovic, V., and Shilling,

www.atmos-meas-tech.net/12/3789/2019/ Atmos. Meas. Tech., 12, 3789–3803, 2019

3802 S.-K. Shin et al.: AERONET aerosol typing

J. E.: Optical properties and aging of light-absorbing sec- Satheesh, S. and Moorthy, K. K.: Radiative effects of nat-
ondary organic aerosol, Atmos. Chem. Phys., 16, 12815–12827, ural aerosols: A review, Atmos. Environ., 39, 2089–2110,
https://doi.org/10.5194/acp-16-12815-2016, 2016. https://doi.org/10.1016/j.atmosenv.2004.12.029, 2005.
Mamouri, R. E. and Ansmann, A.: Fine and coarse dust separa- Schuster, G. L., Dubovik, O., and Holben, B. N.: Angstrom expo-
tion with polarization lidar, Atmos. Meas. Tech., 7, 3717–3735, nent and bimodal aerosol size distributions, J. Geophys. Res.-
https://doi.org/10.5194/amt-7-3717-2014, 2014. Atmos., 111, D07207, https://doi.org/10.1029/2005JD006328,
Mamouri, R.-E. and Ansmann, A.: Potential of polarization/Raman 2006.
lidar to separate fine dust, coarse dust, maritime, and anthro- Shimizu, A., Sugimoto, N., Matsui, I., Arao, K., Uno, I.,
pogenic aerosol profiles, Atmos. Meas. Tech., 10, 3403–3427, Murayama, T., Kagawa, N., Aoki, K., Uchiyama, A., and
https://doi.org/10.5194/amt-10-3403-2017, 2017. Yamazaki, A.: Continuous observations of Asian dust and
Murayama, T., Müller, D., Wada, K., Shimizu, A., Sekiguchi, other aerosols by polarization lidars in China and Japan
M., and Tsukamoto, T.: Characterization of Asian dust and during ACE-Asia, J. Geophys. Res.-Atmos., 109, D19S17,
Siberian smoke with multi-wavelength Raman lidar over Tokyo, https://doi.org/10.1029/2002JD003253, 2004.
Japan in spring 2003, Geophys. Res. Lett., 31, L23103, Shin, S.-K., Müller, D., Lee, C., Lee, K. H., Shin, D., Kim, Y. J.,
https://doi.org/10.1029/2004GL021105, 2004. and Noh, Y. M.: Vertical variation of optical properties of mixed
Sano, I., Mukai, S., Nakata, M., and Holben, B. N.: Regional and Asian dust/pollution plumes according to pathway of air mass
local variations in atmospheric aerosols using ground-based sun transport over East Asia, Atmos. Chem. Phys., 15, 6707–6720,
photometry during Distributed Regional Aerosol Gridded Ob- https://doi.org/10.5194/acp-15-6707-2015, 2015.
servation Networks (DRAGON) in 2012, Atmos. Chem. Phys., Shin, S.-K., Tesche, M., Kim, K., Kezoudi, M., Tatarov, B.,
16, 14795–14803, https://doi.org/10.5194/acp-16-14795-2016, Müller, D., and Noh, Y.: On the spectral depolarisation and li-
2016. dar ratio of mineral dust provided in the AERONET version
Noh, Y., Müller, D., Lee, K., Kim, K., Lee, K., Shimizu, A., Sano, 3 inversion product, Atmos. Chem. Phys., 18, 12735–12746,
I., and Park, C. B.: Depolarization ratios retrieved by AERONET https://doi.org/10.5194/acp-18-12735-2018, 2018.
sun–sky radiometer data and comparison to depolarization ra- Shin, S.-K., Tesche, M., Müller, D., and Noh, Y.: Technical note:
tios measured with lidar, Atmos. Chem. Phys., 17, 6271–6290, Absorption aerosol optical depth components from AERONET
https://doi.org/10.5194/acp-17-6271-2017, 2017. observations of mixed dust plumes, Atmos. Meas. Tech., 12,
Petzold, A., Veira, A., Mund, S., Esselborn, M., Kiemle, C., 607–618, https://doi.org/10.5194/amt-12-607-2019, 2019.
Weinzierl, B., Hamburger, T., Ehret, G., Lieke, K., and Stocker, T. F., Qin, D., Plattner, G.-K., Tignor, M., Allen, S. K.,
Kandler, K.: Mixing of mineral dust with urban pollu- Boschung, J., Nauels, A., Xia, Y., Bex, V., and Midgley, P. M.:
tion aerosol over Dakar (Senegal): Impact on dust physico- Climate Change 2013: The Physical Science Basis. Contribution
chemical and radiative properties, Tellus B, 63, 619–634, of Working Group I to the Fifth Assessment Report of the In-
https://doi.org/10.1111/j.1600-0889.2011.00547.x, 2011. tergovernmental Panel on Climate Change, 1535 pp., Cambridge
Reid, J. S., Hobbs, P. V., Ferek, R. J., Blake, D. R., Mar- Univ. Press, Cambridge, UK, and New York, 2013.
tins, J. V., Dunlap, M. R., and Liousse, C.: Physical, Sun, Y., Zhuang, G., Wang, Y., Zhao, X., Li, J., Wang, Z., and
chemical and optical properties of regional hazes dominated An, Z.: Chemical composition of dust storms in Beijing and
by smoke in Brazil, J. Geophys. Res., 103, 32059–32080, implications for the mixing of mineral aerosol with pollu-
https://doi.org/10.1029/98JD00458, 1998. tion aerosol on the pathway, J. Geophys. Res., 110, D24209,
Reid, J. S., Koppmann, R., Eck, T. F., and Eleuterio, D. P.: A review https://doi.org/10.1029/2005JD006054, 2005.
of biomass burning emissions part II: intensive physical proper- Tesche, M., Ansmann, A., Müller, D., Althausen, D., Engelmann,
ties of biomass burning particles, Atmos. Chem. Phys., 5, 799– R., Freudenthaler, V., and Groß, S.: Vertically resolved sepa-
825, https://doi.org/10.5194/acp-5-799-2005, 2005. ration of dust and smoke over Cape Verde using multiwave-
Russell, P. B., Bergstrom, R. W., Shinozuka, Y., Clarke, A. length Raman and polarization lidars during Saharan Mineral
D., DeCarlo, P. F., Jimenez, J. L., Livingston, J. M., Rede- Dust Experiment 2008, J. Geophys. Res.-Atmos., 114, D13202,
mann, J., Dubovik, O., and Strawa, A.: Absorption Angstrom https://doi.org/10.1029/2009JD011862, 2009.
Exponent in AERONET and related data as an indicator of Tesche, M., Groß, S., Ansmann, A., Müller, D., Althausen, D.,
aerosol composition, Atmos. Chem. Phys., 10, 1155–1169, Freudenthaler, V., and Esselborn, M.: Profiling of Saharan dust
https://doi.org/10.5194/acp-10-1155-2010, 2010. and biomass-burning smoke with multiwavelength polarization
Russell, P. B., Kacenelenbogen, M., Livingston, J. M., Hasekamp, Raman lidar at Cape Verde, Tellus B, 63, 649–676, 2011.
O. P., Burton, S. P., Schuster, G. L., Johnson, M. S., Knobel- Twomey, S.: Pollution and the planetary albedo, Atmos. Environ.,
spiesse, K. D., Redemann, J., and Ramachandran, S.: A mul- 8, 1251–1256, https://doi.org/10.1016/0004-6981(74)90004-3,
tiparameter aerosol classification method and its application to 1974.
retrievals from spaceborne polarimetry, J. Geophys. Res., 119, Wang, J. and Martin, S. T.: Satellite characterization of urban
9838–9863, https://doi.org/10.1002/2013JD021411, 2014. aerosols: Importance of including hygroscopicity and mixing
Sano, I., Mukai, S., Nakata, M., and Holben, B. N.: Regional and state in the retrieval algorithms, J. Geophys. Res., 112, D17203,
local variations in atmospheric aerosols using ground-based sun https://doi.org/10.1029/2006JD008078, 2007.
photometry during Distributed Regional Aerosol Gridded Ob- Washenfelder, R. A., Attwood, A. R., Brock, C. A., Guo, H., Xu,
servation Networks (DRAGON) in 2012, Atmos. Chem. Phys., L., Weber, R. J., Ng, N. L., Allen, H. M., Ayres, B. R., Bau-
16, 14795–14803, https://doi.org/10.5194/acp-16-14795-2016, mann, K., Cohen, R. C., Draper, D. C., Duffey, K. C., Edger-
2016. ton, E., Fry, J. L., Hu, W. W., Jimenez, J. L., Palm, B. B.,

Atmos. Meas. Tech., 12, 3789–3803, 2019 www.atmos-meas-tech.net/12/3789/2019/

S.-K. Shin et al.: AERONET aerosol typing 3803

Romer, P., Stone, E. A., Wooldridge, P. J., and Brown, S. S.: Xia, X. G., Li, Z. Q., Holben, B., Wang, P. C., Eck, T., Chen, H. B.,
Biomass burning dominates brown carbon absorption in the ru- Cribb, M., and Zhao, Y. X.: Aerosol optical properties and radia-
ral southeastern United States, Geophys. Res. Lett., 42, 653–664, tive effects in the Yangtze Delta region of China, J. Geophys.
https://doi.org/10.1002/2014GL062444, 2015. Res., 112, D22S12, https://doi.org/10.1029/2007JD008859,
2007.

www.atmos-meas-tech.net/12/3789/2019/ Atmos. Meas. Tech., 12, 3789–3803, 2019