Impact of particle nonsphericity on the development and properties of aerosol models for East Asia

Atmospheric Environment 101 (2015) 246e256

Contents lists available at ScienceDirect

Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv

Impact of particle nonsphericity on the development and properties of aerosol models for East Asia Hao Chen, Tianhai Cheng*, Xingfa Gu, Yu Wu State Key Laboratory of Remote Sensing Science, Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, China

h i g h l i g h t s
Impact of particle nonsphericity on the aerosol models clustering is studied.
Aerosol models exhibit significant changes considering the shape information.
Difference of aerosol models due to shape affects the aerosol remote sensing.

a r t i c l e i n f o

a b s t r a c t

Article history: Received 19 May 2014 Received in revised form 14 November 2014 Accepted 17 November 2014 Available online 18 November 2014

In this paper, the effects of aerosol nonsphericity information on the classification of aerosol models and the associated radiative properties over East Asia are investigated. The radiance measurements and inversions of the Aerosol Robotic Network (AERONET) are used. Four aerosol models over East Asia are obtained by adding the shape information to the clustering analysis. These four aerosols are identified on the basis of their optical properties. Compared to the results without sphericity parameter, adding the sphericity parameter in the clustering process contributes to the extraction of a strongly absorbing aerosol. Furthermore, the effect of the physical and optical properties of the aerosol on the top of atmospheric (TOA) total reflectance and polarized reflectance are investigated. The results indicate that the addition of the sphericity parameter in the clustering process leads to a change in the total reflectance by up to 16% and a change in the polarized reflectance by up to 100%. © 2014 Elsevier Ltd. All rights reserved.

Keywords: Aerosol model Aerosol shape Radiative properties Cluster analysis East Asia

1. Introduction Aerosols play a significant role in climate change by directly interacting with atmospheric radiation and by indirectly modifying cloud optical properties and persistence. The uncertainty in quantifying the climatic impacts of aerosols continues to be greater than that of greenhouse gases (Solomon, 2007) due to the variety of sources, varying trends in aerosol loading and extreme heterogeneity in the spatial and temporal variability of their optical and microphysical properties (Morgan et al., 2006; Kaskaoutis et al., 2007). Reducing the uncertainties of aerosol radiative impacts requires integrated aerosol measurements and theoretical analysis to characterize various aerosol models and sources.

* Corresponding author. E-mail address: [email protected] (T. Cheng). http://dx.doi.org/10.1016/j.atmosenv.2014.11.036 1352-2310/© 2014 Elsevier Ltd. All rights reserved.

Various aerosol types have different effects on solar radiation (Kaskaoutis and Kambezidis, 2008) and on the sign and magnitude of the aerosol radiative forcing (Satheesh and Krishna Moorthy, 2005). For example, the presence of absorbing aerosols, such as black carbon, can change the sign of the forcing from negative to positive (Heintzenberg et al., 1997). Each aerosol model considers specific particle sizes and shapes and depends on sources, emission rates, transport, chemical reactions and removal mechanisms (Chin et al., 2002; Park, 2004). The scattering and absorption properties of different aerosol models are highly uncertain. To clarify the mechanisms of aerosol radiative forcing and to improve the accuracy of aerosol remote sensing retrieval, it is critical to use the proper aerosol models. In recent years, a number of studies have been performed to categorize the aerosol models into global or regional scales based on ground and satellite observations. Compared to satellite remote sensing, ground-based aerosol observations provide wide angular and spectral measurements of solar and sky radiation and are best suited to continuously derive the detailed aerosol optical

H. Chen et al. / Atmospheric Environment 101 (2015) 246e256

properties in key locations (Dubovik et al., 2002). Omar et al. (2005) developed global aerosol models via cluster analysis of approximately 250 Aerosol Robotic Network (AERONET) sites globally. Six significant clusters with distinct microphysical and optical properties are identified as desert dust, biomass burning, urban industrial pollution, rural background, polluted marine and dirty pollution. Qin et al. (2009) classified the Australian continental aerosol models via hierarchical cluster analysis of optical properties obtained from AERONET data at Australian aerosol ground stations over the last decade. Four classes are identified: aged smoke, fresh smoke, coarse dust and super-absorptive aerosols. Lee and Kim, 2010 developed type-specific aerosol models that include aerosol microphysics and optical properties. Although much attention has been paid to global and regional aerosol models and their corresponding microphysics and optical properties, information on particle shape effects in the development of aerosol models is still limited. Tropospheric aerosols have a large variety of shapes. Analyses of laboratory measurements and in situ data using scanning electron microscopes reveal that the mineral dust and carbonaceous soot particles have complex morphologies, rather than homogeneous spheres (Adachi et al., 2007; Borghese et al., 1984). It has become universally recognized that the nonsphericity of particles has a profound effect on their scattering and absorption properties (Mishchenko, 2009). Particle nonsphericity has been considered in both aerosol particle modeling and remote sensing applications during the last two decades (Bohren and Singham, 1991; Cheng et al., 2010; Derimian et al., 2012; Feng et al., 2009; Ginoux, 2003; Yang et al., 2007). Dubovik et al. (2006) utilized the spheroid model to reproduce mineral dust light scattering matrices and retrieved detailed aerosol properties measured by AERONET ground-based sun/sky radiometers. The nonsphericity of aerosol particles (%sphericity) can be obtained as part of AERONET retrievals (Dubovik et al., 2006). With the population increase, industrialization and demands for energy, the aerosol load in East Asia is gradually increasing and having significant impacts on the continuation of solar dimming (Badarinath et al., 2010). In East Asia, aerosols vary greatly in composition, shape and size and have a significant effect on the atmospheric radiation budget. Although much attention has been paid to this issue, information on the aerosol properties and their spatial and temporal variation is still limited. Thus, a systematic study is required to clarify the particle shape effect on the optical properties of different aerosol models and to improve our knowledge of East Asia aerosol radiation effects. In this paper, the effects of aerosol shape on the development of aerosol models over East Asia are studied via cluster analysis. The optical and microphysical properties of aerosols are obtained from radiance measurements and inversion data at 19 AERONET stations. To quantify the shape effect on the properties of various aerosol models, the corresponding optical properties of clustering results without the shape parameter are presented for comparison. In section 2, the AERONET site distribution and methodology of aerosol classification are briefly introduced. The clustering results and aerosol nonsphericity parameter in relation to the optical properties of various aerosol classes are presented in section 3. The aerosol particle shape impact on the TOA total/polarized reflectance is provided in section 4. In this section, the method used by Cheng et al. (2010) is referred to investigate the sensitivity of TOA total reflectance and polarized reflectance to different aerosol models.

247

2. Clustering classification 2.1. Data selection Measurements of sun and sky radiances observed in the solar almucantar at the AERONET network (Holben et al., 1998) are used to estimate detailed aerosol properties (e.g., the aerosol size distribution, the complex refractive index, the single scattering albedo (SSA), the sphericity, the phase function and absorption properties). The latest retrievals scheme (Dubovik et al., 2006) assumes that the aerosol is a mixture of spherical and non-spherical aerosol components and estimates the fraction that is non-spherical. The modeling is performed using kernel lookup tables of quadrature coefficients employed in the numerical integration of spheroid optical properties over size and shape (see the work of Dubovik et al., 2006). The AERONET data are provided as three categories; cloud-screened and quality-assured “Level 2 Inversion All Points” data are used in this paper. The locations of the AERONET sites are shown in Table 1. According to Omar et al. (2005), unique aerosol models are identified by parameters that represent aerosol size and absorption. However, as the aerosol particle shape is important in climate studies and in remote sensing of the environment (Gobbi et al., 2002; Kahnert et al., 2007), there is sufficient motivation to consider the degree of non-sphericity of the aerosols. Thus, the following parameters that comprise the optical and physical properties retrieved from AERONET inversion algorithms are chosen in the clustering analysis:

Table 1 AERONET site distribution. Country Site name China

Japan

Korea

Beijing Cheng-Kung_Univ EPA-NCU

Lon/Lat

Observation Period

Number of records

116.38/39.98 2001.3e2012.8 2280 120.22/23.00 2002.3e2012.3 1154 121.19/24.97 2006.7 798 e2011.12 Hefei 117.16/31.91 2005.11 175 e2008.11 Hong_Kong_Polyu 114.18/22.30 2005.11 240 e2011.10 Hong_Kong_Hok_Tsui 114.26/22.21 2007.11 338 e2010.7 Lulin 120.87/23.47 2007.3 175 e2011.11 NCU_Taiwan 121.19/24.97 1998.4e2012.5 704 Taihu 120.22/31.42 2005.9 1716 e2012.10 XiangHe 116.96/39.75 2001.3e2001.4; 2282 2004.9e2012.5 Xinglong 117.58/40.37 2006.2e2008.5; 363 2010.8e2012.5 Yulin 109.72/38.28 2001.4 173 e2002.10 Noto 137.14/37.33 2001.4 553 e2010.12 Osaka 135.59/34.65 2001.11 1051 e2011.11 Shirahama 135.36/33.69 2000.10 3807 e2010.11 Anmyon 126.33/36.54 1999.10 1541 e2007.11 Gosan_SNU 126.16/33.29 2001.4e2011.6 673 Gwangju_GIST 126.84/35.23 2004.2 818 e2011.11 Seoul_SNU 126.95/37.46 2000.11 252 e2003.2

248

H. Chen et al. / Atmospheric Environment 101 (2015) 246e256

(1) Complex refractive index (mr, mi) at 440, 676, 869 and 1020 nm; (2) Aerosol size distribution parameters: fine/coarse mean radius (Rfine, Rcoarse), fine/coarse standard deviation (Sfine, Scoarse), fine/coarse mode total volume (Cfine, Ccoarse); (3) Single scattering albedo (SSA) at 676, 869 and 1020 nm; (4) Asymmetry factor (ASY) at 440, 676 and 869 nm; (5) Sphericity (represents an estimate of the percentage of spherical particle scattering). Single scattering albedos at 440 nm and asymmetry parameters at 1020 nm are abandoned in the clustering because the dust aerosols exhibit an absorption feature at the near-ultraviolet 440 nm (Dubovik et al., 2002), and larger-sized aerosols are better separated by phase function asymmetry factors at short wavelengths. Before the cluster analysis, each of the parameters was normalized by the standard deviation. This step ensures that the relative contribution of the parameters to the distance calculation is balanced. The normalizing process is as follows:

pnormalized ¼

pM SD

(1)

Where pnormalized is the data used in the clustering, p a value of some selected parameter, M and SD the mean value and the standard deviation of the parameter respectively. Additionally, the fine mode fraction (FMF) is also obtained from AERONET measurements, but it is not used in the clustering. FMF is used to determine the dominant size mode and provides quantitative information for each fine- and coarse-mode aerosol. It is defined as the ratio of fine-mode AOT to total AOT.

hierarchy of clusters: each observation starts in its own cluster, and pairs of clusters are merged as one moves up the hierarchy. Hierarchical clustering has the distinct advantage that any valid measure of distance can be used. In fact, the observations themselves are not required: only a matrix of distances is used. A brief description of the algorithm is as follows. Initially, each single record is treated as a class. The distances between each pair of classes are calculated. The distance between two classes, A and B, is defined to be the maximum distance between any of class A records and any of class B records (the so called complete-link). The two closest classes are then merged as one. The last two steps are repeated until the minimum distance among the paired classes reaches a prescribed criterion. This criterion is experimental and controls the number of classes. Clusters which have more than 5% of the total records are kept. The remaining clusters with less than 5% of the total records in each are discarded for their fragmentation, which may be caused by the inversion error or due to short-lived variations of the aerosol properties. To illustrate the impact of aerosol particle shape information on aerosol classification, the selected AERONET records were classified in two runs with different input parameters. The sphericity parameter is added in the experiment group. Parameters in the experiment group are the same as above except for the sphericity parameter. The distance between different clusters are calculated to validate the clustering results, which is defined as the maximum distance between a pair of records from different clusters.

DðX; YÞ ¼ maxðdðx; yÞÞ

(2)

where 1) d(x,y) is the distance between elements x2X and y2Y 2) X and Y are two sets of records (clusters)

2.2. Methodology of aerosol classification Cluster analysis is a statistical tool used to group large datasets into several categories using predefined variables (Anderberg, 1973). This method was applied globally and regionally by Omar et al. (2005), Qin et al. (2009), Levy et al. (2007) and Lee and Kim, 2010. To classify the aerosol models, the microphysical, chemical or optical properties are considered. Two of the most commonly used cluster algorithms are K-means and the hierarchical clustering algorithm. K-means clustering is a commonly used data clustering method that performs unsupervised learning tasks. The greatest drawback of the k-means algorithm is the number of clusters that must be specified in advance. Furthermore, the algorithms prefer clusters of approximately similar sizes as they will always assign an object to the nearest centroid. This often leads to incorrectly cut borders in between clusters. The K-means clustering algorithm is used by Omar et al. (2005) and Levy et al. (2007). In this study, an agglomerative complete-link clustering algorithm of the hierarchical class (Kotsiantis and Pintelas, 2004), which was applied by Qin et al. (2009), is used. Hierarchical clustering is a method of cluster analysis which seeks to build a

Table 2 The distances between classes resulting from clustering adding sphericity parameter. Class

SS

MA

HA

Coarse-Dust

SS MA HA Coarse-Dust

58.1 73.7 73.6 65.1

73.7 56.0 67.7 77.4

73.6 67.7 51.1 83.1

65.1 77.4 83.1 48.8

If X ¼ Y, the distance is defined as the diameter of cluster X/Y (the maximum distance between a pair of records in a cluster). Table 2 shows the distances of 4 classes resulting from two experiments respectively. It is shown that these 4 classes are well separated from each other, demonstrated by much smaller cluster diameters compared to inter-cluster distances. The 4 classes resulting from the second clustering (without the sphericity parameter) also reveal a large distance with each other (results are not shown). 3. Results 3.1. Clustering results interpretation By applying the clustering analysis, four categories are obtained for the two groups. The size distribution, SSA, and ASY parameter of the four clustered aerosol models in the two groups are shown in Table 3. According to their absorptive properties and size parameter discrepancy, the four classes are identified: one coarse-size dominated aerosol model and three fine-sized dominated aerosol models that range from strong scattering to moderate and high absorbing aerosols. The mean sphericity and FMF of each class are investigated. The first four classes in Table 3 display mean sphericities of 0.70, 0.47, 0.57 and 0.18, respectively. The second four classes have mean sphericities of 0.61, 0.55, 0.72 and 0.14, respectively. A low value of sphericity indicates a low percentage of spherical particle scattering. The mean FMF of each class is also investigated. The mean FMF values of the former four classes are 0.87, 0.79, 0.79 and 0.32,

H. Chen et al. / Atmospheric Environment 101 (2015) 246e256

249

Table 3 Summary of single scattering albedo, refractive index and size distribution of the cluster analysis Results.a Cluster-with sphericity parameter

SSA676 g676 Cfine[um3/um2] Rfine[um] Sfine Ccoarse[um3/um2] Rcoarse[um] Scoarse Sphericity FMF Number a b

Cluster-without sphericity parameter

SSb

MA

HA

Coarse-dust

SS-1

SS-2

MA

Coarse-dust

0.95 (0.02) 0.675 (0.04) 0.127 (0.07) 0.219 (0.07) 0.51 (0.05) 0.08 (0.06) 2.731 (0.06) 0.602 (0.391) 0.70 0.87 3251

0.9 (0.03) 0.645 (0.03) 0.103 (0.05) 0.183 (0.06) 0.498 (0.04) 0.112 (0.06) 2.673 (0.06) 0.625 (0.373) 0.47 0.79 6457

0.85 (0.02) 0.638 (0.03) 0.084 (0.03) 0.179 (0.06) 0.516 (0.03) 0.083 (0.04) 2.757 (0.06) 0.643 (0.360) 0.57 0.79 752

0.94 (0.02) 0.674 (0.03) 0.083 (0.05) 0.161 (0.08) 0.515 (0.04) 0.3 (0.21) 2.402 (0.06) 0.599 (0.414) 0.18 0.32 1909

0.95 (0.02) 0.696 (0.03) 0.143 (0.08) 0.249 (0.05) 0.536 (0.04) 0.079 (0.06) 2.798 (0.06) 0.591 (0.347) 0.61 0.91 2123

0.94 (0.02) 0.648 (0.03) 0.101 (0.04) 0.172 (0.06) 0.465 (0.03) 0.114 (0.08) 2.532 (0.05) 0.629 (0.413) 0.55 0.76 3227

0.89 (0.03) 0.642 (0.03) 0.098 (0.04) 0.181 (0.06) 0.507 (0.04) 0.107 (0.06) 2.697 (0.06) 0.626 (0.371) 0.72 0.79 5193

0.94 (0.02) 0.674 (0.03) 0.089 (0.05) 0.169 (0.08) 0.534 (0.04) 0.303 (0.22) 2.55 (0.06) 0.589 (0.433) 0.14 0.35 1712

Values in the parentheses represent standard deviation of each cluster. “SS” represents strong scattering aerosols, “MA” moderately absorbing aerosols, “HA” highly absorbing aerosols, and “Dust” coarse-size dominated aerosols.

respectively, and the values of the latter four classes are 0.91, 0.76, 0.79 and 0.35, respectively. Large values of FMF represent aerosol types that are dominated by fine particles, whereas low values indicate aerosol types that are dominated by coarse particles. Considering the size distribution and absorptive properties, the first four classes (adding the sphericity parameter during the clustering process) are labeled S-SS, S-MA, S-HA and S-Dust respectively. In these labels, “S” represents clustering results with the sphericity parameter, “SS” strong scattering aerosols, “MA” moderately absorbing aerosols, “HA” highly absorbing aerosols, and “Dust” coarse-size dominated aerosols. The SSA (within the wavelengths of 440e1020 nm) of S-SS ranges from 0.94 to 0.95 and is close to non-absorbing. The fine mode volume concentration of S-SS (0.127 um3/um2) is far greater than that of coarse mode volume concentration (0.08 um3/um2), which reveals fine mode dominated aerosol particles. Meanwhile, the real parts of the refractive index for the S-SS are the lowest ones (Table 3). According to Dubovik et al. (2002), urbaneindustrial aerosols usually show higher SSA correlated with lower real part of the refractive index. This correlation is likely to be appearing due to geophysical reasons. The SSA of S-MA ranges from 0.89 to 0.90, indicating moderately absorbing. They are the dominant aerosols with 52% of records classified in this class. The fine mode volume concentration of S-MA is comparative to that of the coarse mode volume concentration, which reveals mixed aerosol particles. Considering the location of the aerosols (near or in urban centers), these aerosols are mainly urban pollution.

Table 4 The fraction of records in the “S” classes (obtained from the clustering process with the sphericity parameter) which are also classified into the corresponding “NS” classes (obtained from the clustering process without the sphericity parameter).

S-SS S-MA S-HA S-DU

NS-SS-1

NS-SS-2

NS-MA

NS-DU

40% 12% 0% 2%

47% 19% 1% 27%

10% 64% 85% 5%

3% 5% 0% 67%

The SSA of S-HA ranges from 0.80 to 0.85, which shows a strong absorbing feature. The real parts of the refractive index for S-HA aerosols are in the range of 1.47e1.50 over the wavelength range 440e120 nm. According to studies of Dubovik et al. (2002), higher real part of refractive index combined with lower SSA are possibly associated with high concentrations of black carbon in the atmosphere. The S-Dust shows relatively low FMFs (whose mean FMF is 0.32) and high asymmetry factors compared to the three fine classes. The mean SSA ranges from 0.90 to 0.95 and exhibits a distinct feature in the spectral distribution of single scattering albedo, with a depression at 440 nm followed by a gradual increase at longer wavelengths. This feature is typically observed in desert dust condition (Dubovik et al., 2002; Levy et al., 2007; Qin and Mitchell, 2009) and is probably due to absorption in the blue spectral region by the iron oxide hematite. Based on this method, the second four classes (without the sphericity parameter in the clustering process) are labeled NS-SS-1, NS-SS-2, NS-MA and NS-Dust. Here, “NS” represents no sphericity parameter in the clustering process. In the second group, The values of SSA for the SS1 and SS2 are very close and both represent the strong scattering aerosols. This is also an important reason for they both being labeled SS. The difference between these two aerosol types are mainly shown in the size distribution. SS1 aerosols are apparent fine mode dominated, with the fine mode volume concentration (0.143 um3/um2) 2 times larger than that of coarse mode volume concentration (0.079 um3/um2). The SS2 aerosols are mixed. The fine mode volume concentration (0.101 um3/um2) is close to that of coarse mode volume concentration (0.114 um3/ um2). The SSA (within the wavelengths of 440e1020 nm) of NS-MA ranges from 0.87 to 0.89, and that of NS-Dust ranges from 0.90 to 0.94.

3.2. Comparison of similar aerosol models from two clustering processes Changing the clustering parameters will result in different classes containing different record groups. To quantify the effects of particle nonsphericity on the optical properties of aerosol models, it

250

H. Chen et al. / Atmospheric Environment 101 (2015) 246e256

is necessary to compare the optical properties (refractive index, SSA and asymmetry) of different groups with duplicate names. Before comparing the results obtained from the two clustering processes (i.e., including or excluding the sphericity parameter), it is necessary to provide a detailed comparison of the data records between the two types of clustering results (e.g., the fraction of records in the “S-SS” class that remain the “NS-SS” classes) and explain how the change in the clustering parameters modify the derived classes. This type of comparison will reveal the actual similarities and differences between the aerosol models with duplicate names.

Table 4 compares the data records between the two types of clustering results. For the SS aerosol models, approximately half of the records (40%) in the S-SS class stem from the NS-SS-1 class and others mainly originate from the NS-SS-2 class. This is consistent with the physical and optical properties of S-SS shown in Table 3. The refractive index, size distribution and asymmetry of S-SS aerosols are between those of NS-SS-1 and NS-SS-2 aerosols. For the MA aerosol models, 64% of records in the S-MA class stem from the NS-MA class, another 19% come from NS-SS-2 and approximately 12% come from NS-SS-1. Based on Table 3, the

Fig. 1. The AERONET-retrieved refractive index of three aerosol classes at their cluster centers. The solid lines (labeled as “S”) represent clustering results that include the sphericity parameter, whereas the dashed lines (labeled as “NS”) represent results without the sphericity parameter in the clustering process. The SS and MA represent strong scattering and moderately absorbing fine-mode aerosols, respectively.

H. Chen et al. / Atmospheric Environment 101 (2015) 246e256

physical and optical properties of S-MA are similar to those of NSMA. For the HA aerosol models, the majority of records (85%) in SHA originate from NS-MA. It is obvious that the SSAs of S-HA are much lower than those of NS-MA in the four wavelengths (Table 3). Because the majority of records in S-MA (64%) and S-HA (85%) stem from the NS-MA class, we concluded that adding the sphericity parameter to the clustering process contributed to extracting strong absorbing aerosol records from moderately absorbing aerosol records. For the dust aerosols, 67% of records in S-Dust come from NSDust, and another 27% come from the NS-SS-2 class. Those 27% of the records in the NS-SS-2 class have a mean sphericity of 14.3%, which represents a low percentage of spherical particle scattering (or a large percentage of nonspherical particle scattering). Because dust aerosols are the main nonspherical coarse mode aerosol particles, the scattering ability of dust is usually high. On the other hand, the SS aerosols are mainly spherical fine mode particles, and the scattering ability of SS is also high. Adding the sphericity parameter to the clustering process contributed to extracting records with low sphericity values from records with high sphericity values. 3.3. Comparison of optical properties of similar aerosol models using two clustering processes Atmospheric aerosols absorb and reflect solar radiation, causing surface cooling and heating of the atmosphere. The interaction between aerosols and radiation depends partly on their complex index of refraction, which is related to the particles' chemical compositions. Using Mie scattering calculations, the complex refractive indices (RI) of the samples were retrieved including the scattering (real part) and the absorption (imaginary part) indices. The refractive indices were further used to derive the particles' single scattering albedos and mass scattering and absorption extinction coefficients and to estimate the possible radiative impact of mixed aerosols on climate change. Fig. 1 compares the real and imaginary parts of the refractive index for four clustered aerosol models, which include or exclude the spherical parameter in the clustering process. The results indicate that both the real and imaginary parts of the refractive index for the four aerosol models reveal significant changes after the spherical parameters were added to the clustering, and the magnitudes of the decreases/increases in the imaginary portion are greater than those in the real portion. For SS, adding the sphericity parameter to the clustering results in lower real parts of the refractive index at two short wavelengths (440 nm and 676 nm), which indicates that the scattering ability declines at the short wavelengths. Compared to the real parts of the refractive indices, the imaginary parts of the refractive indices (absorption) of SS declined at two short wavelengths (440 nm and 676 nm). In other words, adding the sphericity parameter to the clustering declines the scattering and absorption capabilities of strong scattering aerosols at short wavelengths. It is shown that 13% of records in the SS come from the MA and Dust. Further investigation reveal that the mean SSAs of these records range from 0.94 to 0.95 at the wavelength 440 nme1020 nm. Moreover, the average size distribution of those records indicates a greater volume of fine particles than coarse ones. Therefore it is reasonable to identify those 13% of records as SS. It is evident that clustering with the sphericity parameter helps distinguish between SS, MA and Dust types. For MA, both the real and imaginary parts of the refractive indices at the four wavelengths decreased after the sphericity parameter is added in the clustering; the magnitudes of the decreases are 4% and 14%, respectively. The decreases reveal that the

251

scattering ability and (especially) the absorbing ability of MA are weakened after considering the sphericity parameter in the clustering. For Dust, both the real and imaginary parts of the refractive indices at the four wavelengths decrease after the sphericity parameter was introduced, and the magnitudes of the decreases are 0.5% and 17%, respectively. The decreases represent minor and significant weakening of the scattering and absorbing abilities of the dust model aerosols, respectively. To investigate and quantify the particle shape factor in relation to the aerosol radiative effect and remote sensing retrieval, the optical properties (single scattering albedo and asymmetry parameter) of each aerosol model retrieved from the two experiments are presented in Figs. 2 and 3. Based on Fig. 2, including the sphericity parameter in the clustering greatly impacts the aerosol models (less absorbing). For the strong scattering aerosols, the mean SSA of S-SS at the wavelength of 676 nm is slightly higher than those of NS-SS-1 and NS-SS-2, by 0.2% and 1% respectively, due to the introduction of the sphericity parameter into the clustering process. The moderate absorbing aerosols are less absorptive after including the sphericity parameter in the clustering as expressed by the mean single scattering albedo at 676 nm that increased by approximately 0.012. The dust aerosols indicate a similar effect (but very small) in which the magnitude of the single scattering albedo (676 nm) increases by 0.007. The asymmetry parameter is a measure of the preferred scattering direction (forward or backward) of the light that is encountering the aerosol particles. In radiative transfer studies, the asymmetry factor ‘g’ is equal to the mean value of m (the cosine of the scattering angle) and weighted by the angular scattering phase function P(m). Among the three aerosol models, the SS has median values of total, fine and coarse asymmetry parameters after including the sphericity parameter in the clustering process. NS-SS1 and NS-SS-2 have a significant discrepancy (a magnitude of 0.1) in the fine asymmetry parameters at the wavelength of 1020 nm. MA and Dust show non-significant changes in the asymmetry parameters after the sphericity parameter is added in the clustering. 4. Comparison of the TOA total/polarized reflectance of similar aerosol models from two clustering processes The TOA total and polarized reflectance at 865 nm is the sum of the contributions of different sources: scattering from molecules, aerosols, and reflections from the surface. The contribution of scattering by molecules to the TOA total reflectance can be calculated easily from the mean altitude of the pixels and the viewing geometry. To investigate the aerosol shape effect on the TOA total reflectance and polarized reflectance at 865 nm, sensitivity studies are presented. Upward radiances and polarized radiances are simulated using the RT3 model (Evans and Stephens, 1991), which decomposes the atmosphere into several plane-parallel, vertically inhomogeneous layers that contain randomly-oriented and arbitrary shape particles. The radiative properties are computed by the doubling and adding method. This approach computes the radiative properties of the medium rather than the radiance field itself so that radiances exiting the atmosphere may be easily found for many boundary conditions after the solution is computed (Evans and Stephens, 1991). Given the Stokes parameters of the emergent light at the TOA, the reflectance and polarized reflectance at a given wavelength are calculated as follows:

R ¼ pI=m0 F0

(3)

252

H. Chen et al. / Atmospheric Environment 101 (2015) 246e256

Fig. 2. Same as in Fig. 1, but for the single scattering albedo of three aerosol classes at their cluster centers.

 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  ffi Rp ¼ p Q 2 þ U2 m0 F0

(4)

where I, Q and U are the Stokes parameters,m0 is the cosine of solar zenith angle, and F0 represents the incident solar flux density at the given wavelength; the p factor converts the dimension from flux density to radiance. V was neglected in these simulations because its magnitude is always approximately three orders of magnitude smaller than the other three Stokes parameters. The microphysical and chemical parameters at three class centers (SS, MA and Dust) are used as input for modeling. The solar zenith angle and relative azimuth angle are assumed to be 30 and 0 , respectively. To identify the aerosol information content in the TOA reflectance and polarized reflectance, the land surface contribution is ignored and assumed to be Lambertian. The atmosphere is assumed to be plane parallel and cloud free. Experiments are divided into two groups: for one group we calculated the TOA upward reflectance and polarized reflectance with parameters from clustering without sphericity (the NS-SS-1, NS-MA and NSDust class centers); the other group performs the same calculations with clustering results that included the sphericity parameter (the S-SS, S-MA and S-Dust class centers). The relative differences between the results of the two groups are calculated (Fig. 4). The solar zenith angle is assumed to be 30 , the relative azimuth angle is 0 , the aerosol optical depths (AOD) is 0.45, and the surface reflectance is assumed to be 0.0. In addition, for the second group of results (sphericity parameter added in the clustering), the discrepancy of the TOA

reflectance/polarized reflectance among the four aerosol models are investigated. Figs. 5 and 6 show the relative difference of the TOA reflectance/polarized reflectance for the four aerosol types with sphericity parameter added in the clustering. The AODs are assumed to be 0.1, 0.45 and 0.8, which represent fresh, moderately polluted and heavy polluted conditions respectively. To illustrate the discrepancy of the TOA reflectance among the four aerosol models, the reflectance simulated with MA records is used as a reference. The relative differences of the TOA reflectance/polarized reflectance among MA and the other three aerosol models are plotted. Based on Fig. 4, the TOA upward reflectance simulated from the two clustering results exhibits a large SS discrepancy. The relative differences reach 16% for the scattering angle at 150 . MA and Dust produce more similar values of TOA upward reflectances; the relative differences reach 3% and 5%, respectively, as the scattering angle is close to 150 . For the polarized reflectance, the Dust model represents the largest relative difference of almost 100% at a scattering angle of 150 . Followed by SS, the relative discrepancies reach 70% at a scattering angle of 150 . MA produces the minimum discrepancies of the TOA upward reflectance: the relative differences are lower than 3% when the scattering angle changes from 60 to 150 . In other words, consideration of the particle shape factor in developing aerosol models based on ground measurements will cause 16% of the relative difference to the TOA reflectance and up to 100% of the relative difference to the TOA polarized reflectance. The influence of the aerosol's shape on the TOA polarized reflectance is bigger than that of the TOA reflectance.

H. Chen et al. / Atmospheric Environment 101 (2015) 246e256

253

Fig. 3. Same as in Fig. 2, but for the total, fine mode and coarse mode asymmetry factors of three aerosol classes at their class centers.

Fig. 4. The relative differences of the TOA upward reflectance/polarized reflectance simulated from clustering results with or without the sphericity parameter for each aerosol model at a wavelength of 0.865 mm. The solar zenith angle is 30 , the relative azimuth angle is 0 , and the AOD and surface reflectance areassumed to be 0.45 and 0.0, respectively.

Large differences of the TOA reflectance/polarized reflectance simulated from the four aerosol models are apparent. Under fresh air conditions (AOD ¼ 0.1), the relative TOA reflectance differences between SS, HA, Dust and MA are small (approximately 10%). For

the polarized reflectance, the relative difference between dust and MA is large and can reach 60% when the scattering angle is close to 140 . Followed by SS, the relative TOA polarized reflectance difference is close to 8%. HA and MA exhibit a minimal difference of

254

H. Chen et al. / Atmospheric Environment 101 (2015) 246e256

Fig. 5. The relative difference of TOA upward reflectance/polarized reflectance simulated from the clustering results with the sphericity parameter for each aerosol model at a wavelength of 0.865 mm. The solar zenith angle is 30 , the relative azimuth angle is 0 , and the surface reflectance is 0.0. The solid, dashed and dash dotted lines represent the relative differences between SS, HA, Dust and MA, respectively; the AOD is assumed to be 0.1, 0.45 and 0.8, respectively.

TOA upward polarized reflectance (a relative difference of approximately 2%) when the scattering angle is lower than 150 .The discrepancy of the polarized reflectance between dust and the other two aerosol models is caused by the depolarization of the coarse mode aerosol. The relative difference of reflectance for the HA changes significantly as the aerosol optical depth increases. The relative difference is approximately 9% when the aerosol is relatively ‘clean’ (AOD equals 0.1 at 865 nm). It reaches 12% under moderately polluted conditions (AOD equals 0.45 at 865 nm) and

approximately 16% when the aerosol is heavily polluted (AOD equals 0.8 at 865 nm). For the other two aerosol models, the relative differences in the reflectance represent non-obvious changes as the AOD increases. For the polarized reflectance, the relative reflectance for the SS model has a decreasing trend when the AOD increases, whereas the other two models display non-significant changes as the AOD increases. In Figs. 4 and 5, the solar zenith angle is set as 30 . Considering that the solar zenith angle is not constant during the day, it is necessary to simulate the diurnal effect of the TOA change in the

H. Chen et al. / Atmospheric Environment 101 (2015) 246e256

255

Fig. 6. Same as in Fig. 5, but for the relative differences of the TOA reflectance/polarized reflectance as functions of the solar zenith angle. The relative azimuth angle and sensor zenith angle are assumed to be 0 and 30 , respectively.

total/polarized reflectance by including the full range of zenith angles. We employed the solar zenith angles ranging from 10 to approximately 80 . The relative azimuth angle and sensor zenith angle were 0 and 30 , respectively. The land surface is Lambertian with an albedo of 0. The relative differences of the TOA reflectance/ polarized reflectance as functions of the solar zenith angle under the high, moderate and low aerosol loadings are shown in Fig. 6. Based on Fig. 6, the HA exhibits an increasing trend in the relative reflectance difference with values ranging from 8% for fresh air to 16% under heavily polluted conditions. The Dust model exhibits a decreasing trend in the relative reflectance difference. The relative difference is approximately 9% when the AOD is assumed to

be 0.1 and declines to 4% when the AOD increases to 0.8. The SS shows non-obvious changes in the TOA reflectance differences as the AOD changes. For the polarized reflectance, the relative reflectance for the SS model exhibits a decreasing trend while the AOD increases, whereas the other two models show non-significant changes as the AOD increases. 5. Conclusion To quantify the effects of particle nonsphericity on the development and radiative properties of different aerosol models, four aerosol classes over East Asia are obtained by taking into account

256

H. Chen et al. / Atmospheric Environment 101 (2015) 246e256

the sphericity parameter based on the clustering analysis method. The four aerosol models mainly differ according to their absorption capabilities and size distributions: ‘strong scattering’ (SS), ‘moderately absorbing’ (MA), ‘highly absorbing’ (HA) and dust. Compared to the results without the sphericity parameter, adding the sphericity parameter in the clustering process helps to extract the highly absorbing aerosol records. The aerosol clustering results indicate a dramatic change when adding the sphericity parameter. By applying the sphericity parameter during the clustering method, both the real and imaginary parts of the refractive index for the three aerosol models (SS, MA and Dust) reveal significant changes compared to the results without the sphericity parameter. The magnitude of the decreases in the imaginary portion is greater than that in the real portion. The single scattering albedo (676 nm) of the moderately absorbing aerosols increased by approximately 0.012, and that of dust increased by 0.007. Furthermore, including the particle shape in aerosol model development based on the ground measurements results in significant discrepancies in the aerosol remote sensing retrieval. Within the various aerosol models, the effects of including or excluding the aerosol shape factor on the TOA total reflectance and polarized reflectance are studied using the RT3 methods. The results indicate that including particle nonsphericity in the aerosol model can change the TOA reflectance by 16% and can change the TOA polarized reflectance up to 100%. Dramatic differences exist among the TOA reflectance/polarized reflectance as simulated by the four aerosol models. The relative difference of reflectance/ polarized reflectance changes as the aerosol optical depth increases. Dust exhibits a significant discrepancy in TOA polarized reflectance with the other three aerosol models, which is caused by the depolarization of the coarse mode aerosols. Based on the results, it is necessary to considering the effects of aerosol particle nonsphericity on the development and radiative properties of aerosol model. In future work, we will apply the nonspherical aerosol models resulting from taking into account the sphericity parameter in clustering process in satellite inversion algorithms and the aerosol optical depths over East Asia will be retrieved. Acknowledgments This research was supported by the National Basic Research Program of China (No: 2010CB950800), Natural Science Foundation of China (No: 41371015, 41001207), Funds of Chinese Academy of Sciences for Key Topics in Innovation Engineering (No: KZCX2-EWQN311), Strategic Priority Research Program of Chinese Academy of Sciences, Climate Change: Carbon Budget (No: XDA05100203), CAS/ SAFEA International Partnership Program for Creative Research Teams (No: KZZD-EW-TZ-09). We thank the PIs and their staff for establishing and maintaining the AERONET sites used in this study. References Adachi, K., Chung, S.H., Friedrich, H., Buseck, P.R., 2007. Fractal parameters of individual soot particles determined using electron tomography: implications for optical properties. J. Geophys. Res. 112, D14202. Anderberg, M.R., 1973. Cluster Analysis for Applications. Office of the Assistant for Study Support Kirtland AFB N MEX. Badarinath, K.V.S., Kharol, S.K., Kaskaoutis, D.G., Sharma, A.R., Ramaswamy, V., Kambezidis, H.D., 2010. Long-range transport of dust aerosols over the Arabian Sea and Indian region d A case study using satellite data and ground-based measurements. Glob. Planet. Change 72, 164e181.

Bohren, C.F., Singham, S.B., 1991. Backscattering by nonspherical particles: a review of methods and suggested new approaches. J. Geophys. Res. 96, 5269. Borghese, F., Denti, P., Saija, R., Toscano, G., Sindoni, O.I., 1984. Multiple electromagnetic scattering from a cluster of spheres. I. Theory. Aerosol Sci. Technol. 3, 227e235. Cheng, T.H., Gu, X.F., Yu, T., Tian, G.L., 2010. The reflection and polarization properties of non-spherical aerosol particles. J. Quant. Spectrosc. Radiat. Transf. 111, 895e906. Chin, M., Ginoux, P., Kinne, S., Torres, O., Holben, B.N., Duncan, B.N., Martin, R.V., Logan, J.A., Higurashi, A., Nakajima, T., 2002. Tropospheric aerosol optical thickness from the GOCART model and comparisons with satellite and sun photometer measurements. J. Atmos. Sci. 59, 461e483. Derimian, Y., Dubovik, O., Tanre, D., Goloub, P., Lapyonok, T., Mortier, A., 2012. €kull volcanic ash layer Optical properties and radiative forcing of the Eyjafjallajo observed over Lille, France, in 2010. J. Geophys. Res. Atmos. 117 (D20). , D., Dubovik, O., Holben, B., Eck, T.F., Smirnov, A., Kaufman, Y.J., King, M.D., Tanre Slutsker, I., 2002. Variability of absorption and optical properties of key aerosol types observed in Worldwide locations. J. Atmos. Sci. 59, 590e608. Dubovik, O., Sinyuk, A., Lapyonok, T., Holben, B.N., Mishchenko, M., Yang, P., Eck, T.F., ~ oz, O., Veihelmann, B., van der Zande, W.J., Leon, J.-F., Volten, H., Mun Sorokin, M., Slutsker, I., 2006. Application of spheroid models to account for aerosol particle nonsphericity in remote sensing of desert dust. J. Geophys. Res. 111, D11208. Evans, K.F., Stephens, G.L., 1991. A new polarized atmospheric radiative transfer model. J. Quant. Spectrosc. Radiat. Transf. 46, 413e423. Feng, Q., Yang, P., Kattawar, G.W., Hsu, C.N., Tsay, S.-C., Laszlo, I., 2009. Effects of particle nonsphericity and radiation polarization on retrieving dust properties from MODIS observations. J. Aerosol Sci. 40, 776e789. Ginoux, P., 2003. Effects of nonsphericity on mineral dust modeling. J. Geophys. Res. 108, 4052. Gobbi, G., Barnaba, F., Blumthaler, M., Labow, G., Herman, J., 2002. Observed effects of particles nonsphericity on the retrieval of marine and desert dust aerosol optical depth by lidar. Atmos. Res. 61, 1e14. Heintzenberg, J., Charlson, R.J., Clarke, A.D., et al., 1997. Measurements and modelling of aerosol single-scattering albedo: progress, problems and prospects. Contrib. Atmos. Phys. 70, 249e263. , D., Buis, J.P., Setzer, A., Vermote, E., Holben, B.N., Eck, T.F., Slutsker, I., Tanre Reagan, J.A., Kaufman, Y.J., Nakajima, T., Lavenu, F., Jankowiak, I., Smirnov, A., 1998. AERONETdA federated instrument network and data archive for aerosol characterization. Remote Sens. Environ. 66, 1e16. €isa €nen, P., 2007. Mie simulations as an error source in Kahnert, M., Nousiainen, T., Ra mineral aerosol radiative forcing calculations. Q. J. R. Meteorol. Soc. 133, 299e307. Kaskaoutis, D.G., Kambezidis, H.D., 2008. The role of aerosol models of the SMARTS code in predicting the spectral direct-beam irradiance in an urban area. Renew. Energy 33, 1532e1543. Kaskaoutis, D.G., Kambezidis, H.D., Hatzianastassiou, N., Kosmopoulos, P.G., Badarinath, K.V.S., 2007. Aerosol climatology: on the discrimination of aerosol types over four AERONET sites. Atmos. Chem. Phys. Discuss. 7, 6357e6411. Kotsiantis, S., Pintelas, P., 2004. Recent advances in clustering: a brief survey. WSEAS Trans. Inf. Sci. Appl. 1, 73e81. Lee, K.H., Kim, Y.J., 2010. Satellite remote sensing of Asian aerosols: a case study of clean, polluted, and Asian dust storm days. Atmos. Meas. Tech. 3, 1771e1784. Levy, R.C., Remer, L.A., Dubovik, O., 2007. Global aerosol optical properties and application to moderate resolution imaging spectro radiometer aerosol retrieval over land. J. Geophys. Res. 112, D13. Mishchenko, M.I., 2009. Electromagnetic scattering by nonspherical particles: a tutorial review. J. Quant. Spectrosc. Radiat. Transf. 110, 808e832. Morgan, M.G., Adams, P.J., Keith, D.W., 2006. Elicitation of expert judgments of aerosol forcing. Clim. Change 75, 195e214. Omar, Ali H., Won, Jae-Gwang, Winker, David M., Yoon, Soon-Chang, Oleg Dubovik, M.P.M., 2005. Development of global aerosol models using cluster analysis of aerosol robotic network (AERONET) measurements. J. Geophys. Res. 110. Park, R.J., 2004. Natural and transboundary pollution influences on sulfate-nitrateammonium aerosols in the United States: implications for policy. J. Geophys. Res. 109, D15204. Qin, Y., Mitchell, R.M., 2009. Characterisation of episodic aerosol types over the Australian continent. Atmos. Chem. Phys. 9, 1943e1956. Satheesh, S.K., Krishna Moorthy, K., 2005. Radiative effects of natural aerosols: a review. Atmos. Environ. 39, 2089e2110. Solomon, S. (Ed.), 2007. Climate Change 2007-The Physical Science Basis: Working Group I Contribution to the Fourth Assessment Report of the IPCC. Yang, P., Feng, Q., Hong, G., Kattawar, G.W., Wiscombe, W.J., Mishchenko, M.I., Dubovik, O., Laszlo, I., Sokolik, I.N., 2007. Modeling of the scattering and radiative properties of nonspherical dust-like aerosols. J. Aerosol Sci. 38, 995e1014.