Spatial distributions and temporal variations of atmospheric aerosols and the affecting factors: A case study for a region in central China

Abstract

Using Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol data, temporal variations and the spatial distribution of aerosol optical depth (AOD or τ) over the Hubei Province in China were investigated from 2003 to 2008. self-organizing maps (SOMs) and linear models were further used to analyse the relationships between AODs and elevation, normalized difference vegetation index (NDVI) and population density. The results were as follows: high AOD values were observed in south-central areas with lower elevations, lower NDVI and larger population densities, whereas low AOD values were observed in the western, northeastern and southeastern areas. The highest AOD values were observed in spring; summer was characterized by lower AOD values, but also the largest ratio of fine particles; in autumn, the coverage of AOD was only smaller than spring with most being fine particles; in winter, coarse particles were dominant when AOD values were the lowest. The AOD monthly average rose substantially in the winter–spring season and dropped sharply in the spring–winter season. Based on these data, both SOMs and linear models show that AOD distribution is influenced by the complex interactions that occur among various elements. The annual AODs are negatively related to ln(elevation) and NDVI and positively related to ln(population density). The ln(elevation) factor affects aerosol distribution more than do the other two factors. Compared to fine-particle aerosols, the selected three factors have a greater impact on the coarse particles.

Full text

This article was downloaded by: [Wuhan University]
On: 19 October 2014, At: 20:38
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
International Journal of Remote
Sensing
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/tres20
Spatial distributions and temporal
variations of atmospheric aerosols and
the affecting factors: a case study for a
region in central China
Yangjie Guo
a
, Song Hong
a
, Nan Feng
b
, Yanhua Zhuang
a
&
Liang Zhang
c
a
Department of Environmental Science, School of Resource and
Environmental Science , Wuhan University , Wuhan , 430072 , PR
China
b
Department of Atmospheric Sciences , University of Alabama in
Huntsville , AL , 35805 , USA
c
Institute of Geodesy and Geophysics, Chinese Academy of
Sciences , Wuhan , 430077 , PR China
Published online: 08 Dec 2011.
To cite this article: Yangjie Guo , Song Hong , Nan Feng , Yanhua Zhuang & Liang Zhang (2012)
Spatial distributions and temporal variations of atmospheric aerosols and the affecting factors: a
case study for a region in central China, International Journal of Remote Sensing, 33:12, 3672-3692,
DOI: 10.1080/01431161.2011.631951
To link to this article: http://dx.doi.org/10.1080/01431161.2011.631951
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
Downloaded by [Wuhan University] at 20:38 19 October 2014

International Journal of Remote Sensing
Vol. 33, No. 12, 20 June 2012, 3672–3692
Spatial distributions and temporal variations of atmospheric aerosols and
the affecting factors: a case study for a region in central China
YANGJIE GUO†, SONG HONG*†, NAN FENG‡, YANHUA ZHUANG†
and LIANG ZHANG§
†Department of Environmental Science, School of Resource and Environmental Science,
Wuhan University, Wuhan 430072, PR China
‡Department of Atmospheric Sciences, University of Alabama in Huntsville, Huntsville,
AL 35805, USA
§Institute of Geodesy and Geophysics, Chinese Academy of Sciences,
Wuhan 430077, PR China
(Received 15 November 2010; in final form 3 April 2011)
Using Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol data,
temporal variations and the spatial distribution of aerosol optical depth (AOD
or τ ) over the Hubei Province in China were investigated from 2003 to 2008.
self-organizing maps (SOMs) and linear models were further used to analyse the
relationships between AODs and elevation, normalized difference vegetation index
(NDVI) and population density. The results were as follows: high AOD values
were observed in south-central areas with lower elevations, lower NDVI and larger
population densities, whereas low AOD values were observed in the western, north-
eastern and southeastern areas. The highest AOD values were observed in spring;
summer was characterized by lower AOD values, but also the largest ratio of fine
particles; in autumn, the coverage of AOD was only smaller than spring with most
being fine particles; in winter, coarse particles were dominant when AOD values
were the lowest. The AOD monthly average rose substantially in the winter–spring
season and dropped sharply in the spring–winter season. Based on these data, both
SOMs and linear models show that AOD distribution is influenced by the complex
interactions that occur among various elements. The annual AODs are negatively
related to ln(elevation) and NDVI and positively related to ln(population density).
The ln(elevation) factor affects aerosol distribution more than do the other two
factors. Compared to fine-particle aerosols, the selected three factors have a greater
impact on the coarse particles.
1. Introduction
In a ground-breaking study, Ramanathan et al. (2001) suggested that a shift in
less than 1% of the earth’s energy balance during the last century has contributed
to an increase in the earth’s surface temperature of 0.6
◦
C. Atmospheric aerosols,
acting as agents of radiative forcing by scattering and/or absorbing incident electro-
magnetic radiation, can force a significant change in the earth’s radiative energy
balance (Crutzen and Andreae 1990, Kaufman et al. 2002, Prospero et al. 2002).
The influence of anthropogenic and natural aerosols has been suggested as being an
*Corresponding author. Email: environmentalanalytics@gmail.com
International Journal of Remote Sensing
ISSN 0143-1161 print/ISSN 1366-5901 online © 2012 Taylor & Francis
http://www.tandf.co.uk/journals
http://dx.doi.org/10.1080/01431161.2011.631951
Downloaded by [Wuhan University] at 20:38 19 October 2014

Spatial and temporal variations of aerosols in central China 3673
important reason for air pollution and climate change on both a global and regional
scale (Randerson et al. 2006, Dey and Tripathi 2008, Rosenfeld et al. 2008). However,
the role of these aerosols in altering regional climate is poorly understood. One of
the major sources of uncertainty in the estimation of aerosol radiative forcing is the
highly variable spatial and temporal distribution of tropospheric aerosols due to their
short lifetimes – a range of a couple of hours to about a week (Andreae et al. 1986).
Understanding the complexities of aerosol radiative forcing and reducing the associ-
ated uncertainties requires a daily monitoring of aerosol emissions and transport as
well as the evolution of their physical and chemical properties. State-of-the-art satellite
instruments, such as the Moderate Resolution Imaging Spectroradiometer (MODIS),
provide routine global measurements of aerosol properties, such as aerosol optical
depth (AOD or τ). Satellite measurements, a tool for studying aerosols, can make up
for the lack of ground-based observations (Bréon et al. 2002, Kaufman et al. 2002,
Kinne et al. 2006).
Over the past decade, aerosols and their effects have gained a lot of attention, and
several studies have been conducted on both a regional and a global scale. Most of
the studies are based on modelling estimates. However, recently, greater emphasis has
been placed on observational studies, which provide the best method of validating and
constraining models (Anderson et al. 2005, Yu et al. 2006). Observational results have
shown that temporal and spatial distributions of aerosols are closely related to ter-
rain, climate, population density and socio-economic activities (Kaufman et al. 2002,
Li et al. 2010). Scholars have also reported studies designed to measure AOD values
over China (Luo et al. 2001, Han et al. 2010, Li et al. 2010). It was found that both the
yearly and monthly mean AOD values over China have patterns related to geographi-
cal features, with the maximum values being located over basin areas. AOD increased
dramatically over the Chinese mainland from 1961 to 1990, particularly in the mid-
dle and lower areas of the Yangtze River and the eastern part of southwest China
(Luo et al. 2001). The yearly mean of PM2.5 concentrations exceed 80 µgm
–3
over
eastern China (Van Donkelaar et al. 2010). The distribution of fine AOD at 0.55 µm
(τ
0.55-fine
) in eastern China is significantly higher than that in western China, and the
distribution of coarse AOD at 0.55 µm(τ
0.55-coarse
) reflected the influence of spring
dust and urban/industrial pollution (Li et al. 2010). But with reference to the factors
affecting AOD distributions, studies consistently provide descriptive and qualitative
analyses, but lack quantitative results. The development of modern neural networks
and computer technology, such as self-organizing maps (SOMs), makes representing
the relevance between variables possible (Kohonen and Member 1990, Céréghino and
Park 2009).
Hubei Province (29
◦
05´ to 33
◦
20

N, 108
◦
21´ to 116
◦
07´ E) is located in central
China at the mid-stream point of the Yangtze River. Its capital, Wuhan, in eastern
Hubei, is a highly industrialized city with over nine million inhabitants. Wuhan is
an emerging megacity and is one of the largest automobile industry bases as well as
the third largest steel-manufacturing city in China. The high mean of PM10 daily
levels in this area mainly result from local emissions related to the steel industry,
power generation, mineral-cement manufacturing, road traffic and anthropogenic
regional background contributions (Querol et al. 2006). The western part of the
province is very mountainous, quite different from the eastern part, and is where
the Shennongjia Nature Reserve, one of the most famous nature reserves in China,
is located. Owing to forest fires and the absence of urban/industrial pollution, hilly
topographical areas with dense forests mainly distribute fine-particle aerosols (Li
et al. 2010). Thus, Hubei is characterized by a complex terrain and a diverse regional
Downloaded by [Wuhan University] at 20:38 19 October 2014

3674 Y. J. Guo et al.
climate (A Series of China’s Natural Resources: Hubei Volume; Editorial Committee
of Natural Resources Series of China 1995), offering rich water resources through
widespread rivers and lakes as well as intensive urbanization areas. This diversity
results in complex aerosol types as well as properties, making it an ideal region for
research concerning the temporal and spatial distribution of aerosols. This article
presents research in which Hubei acts as a case study of aerosol distribution by using
a long-term MODIS-retrieved aerosol data set during a 6-year period (from 2003 to
2008). Using SOM modelling and linear regression methods, quantitative analysis was
also done to simulate the correlation between AOD spatial distribution and elevation,
population density and vegetation index.
2. Methodology
2.1 MODIS aerosol data set
Granules of Aqua MODIS-retrieved aerosol level-2 C005 product over Hubei
were downloaded for the 6-year period (from 2003 to 2008) from the Level 1
and Atmosphere Archive and Distribution System (LAADS) (at http://modis-
atmos.gsfc.nasa.gov/MOD04_L2/index). After stitching and cutting to fit Hubei’s
borders, 2192 daily images for AOD at 0.55 µm(τ
0.55
) and fine-mode fraction (η)
were generated for this study. Thus, monthly average images were calculated for τ
0.55
and η for the 6-year period. Furthermore, monthly average images of AOD were cal-
culated for τ
0.55-fine
and τ
0.55-coarse
, individually, by multiplying τ
0.55
by η and 1 – η,
respectively.
Characteristic images of τ
0.55
, η, τ
0.55-fine
and τ
0.55-coarse
integrated from January
2003 to December 2008 were generated by averaging 72 monthly images. Based on
these integrated images, spatial distributions of τ
0.55
, η, τ
0.55-fine
and τ
0.55-coarse
were
addressed, taking into consideration major influences on the variability of AOD. In
this study, time-series profiles of τ
0.55
, τ
0.55-fine
and τ
0.55-coarse
were extracted over the
72-month sequence to illustrate temporal variation of AOD in Hubei.
2.2 The SOM model analysis
2.2.1 Data preprocessing. First, the Hubei region was divided into 86 research units
by the distribution of counties. Then, using a spatial statistics approach, the annual
average AODs including τ
0.55
, τ
0.55-coarse
and τ
0.55-fine
over every research unit in 2003
were obtained as three of the six variables to simulate. Based on the literature (Luo
et al. 2001, Han et al. 2010, Li et al. 2010), three other quantitative variables were
selected from a series of influencing factors of the τ
0.55
spatial distribution: the eleva-
tion, the vegetation index and the population density. The Shuttle Radar Topography
Mission (SRTM) data set (90 × 90 m) provided by the International Scientific Data
Service Platform (http://datamirror.csdb.cn) was used to acquire the regional ele-
vation. The vegetation index was represented by Global Inventory Modelling and
Mapping Studies (GIMMS) normalized difference vegetation index (NDVI) data
(Tucker et al. 2004), which was provided by the Environmental and Ecological Science
Data Centre for West China sponsored by National Natural Science Foundation of
China (http://westdc.westgis.ac.cn). Population density numbers for every region came
from the population calculated at the end of the year divided by the area of the
region and was obtained from the Hubei Statistical Yearbook (Hubei Statistics Bureau
2004–2007) and the Administrative Divisions of the People’s Republic of China 2009,
respectively.
Downloaded by [Wuhan University] at 20:38 19 October 2014

Spatial and temporal variations of aerosols in central China 3675
Annual values of the aforementioned six variables from 2003, acting as the input
vectors of the SOM model, needed to undergo normal verification to enhance the
performance of the model algorithm (Obu-Cann et al. 2001). If they did not conform
to the normal distribution, normal transformation was needed. After removing the
outliers through the Kolmogorov–Smirnov test, unit data of Hubei’s τ
0.55
, τ
0.55-coarse
,
τ
0.55-fine
, ln(elevation), NDVI and ln(population density) approximately conformed to
the normal distribution (SPSS 15.0; SPSS China, Shanghai, China).
2.2.2 SOM algorithm. The SOM is a competitive learning neural network based
on unsupervised learning (Vesanto 2002, Zhang et al. 2008). In this study, the
SOM simulation was based on Matlab 7.1, using the SOM toolbox developed by
the Laboratory of Computer and Information Science of Helsinki University of
Technology (Kohonen et al. 1996). The learning rate, neighbourhood function and
neighbourhood radius were adopted as the default values in the training process.
2.2.3 SOM simulation results. SOM visualization results of the simulation model
were obtained to analyse the relationship between the input variables. The unified dis-
tance matrix (U-matrix) indicates the distance between each of the SOM neuron nodes
and its adjacent node. The component plane shows the value of the variable in each
map unit. Six components reflect the data characteristics each variable learned after
training. The corresponding numerical size of each node is indicated with a differ-
ent colour. Relationships between variables can be seen as similar patterns in identical
places on the component planes. Whenever the value of one variable changes, the value
of other variables also change (Vesanto and Ahola 1999, Vesanto 2002).
2.3 Linear regression analysis
Using the linear regression method, a quantitative analysis of each factor that influ-
enced the AOD spatial distribution was carried out. The linear regression analysis
was addressed in SPSS 15.0 software (SPSS China). The regression equations between
AOD, elevation, NDVI and population density were computed and compared with
the SOM model results. The stability of the results has been evaluated based on
predictions of linear regression.
3. Results and discussion
3.1 Geographic distribution of aerosols
3.1.1 General distribution. Figures 1(a) and (b) show the location of Hubei in China
and its placement in relation to the rest of the world. In figure 1(c), the geographic
distribution of the 6-year mean τ
0.55
over Hubei shows remarkable characteristics. The
highest τ
0.55
values (τ
0.55
> 0.70) are mainly over the mid-south area of Hubei, which
is surrounded on three sides by lower value (0.50 <τ
0.55
< 0.70) areas, located to the
north, east and southeast, whereas the lowest value (τ
0.55
< 0.50) areas are located in
the western portion of the province. Maximum τ
0.55
occurs mainly in three regions:
Wuhan, Ezhou and Jingzhou. As Luo et al. (2000) reported, Wuhan city is one of the
highest value regions of AOD in China. The high-mean PM10 daily levels reach up to
156 mg m
–3
at the urban site and 197 mg m
–3
at the industrial hotspot, exceeding the
US-EPA or EU annual limit values by 3–4 times (Querol et al. 2006).
Downloaded by [Wuhan University] at 20:38 19 October 2014

3676 Y. J. Guo et al.
3332313029
109 110 111 112 113
Longitude (°E)
Longitude (°E)
Latitude (°N)
Latitude (°N)
114 115 116
1.00+
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
N
116
115
114113112111110109
33
(c)
(a)(b)
32
31
30
29
Elevation (m)
<50
50–100
100–200
200–300
300–400
400–500
500–600
600–800
800–1000
1000–1200
1200–1500
1500–2000
2000–2500
> 2500
Figure 1. Research area and geographical distribution of 6-year mean τ
0.55
over Hubei
Province during 2003–2008. (a) Location of Hubei in China and on Earth. (b) Administrative
divisions and landform zoning map of Hubei. (c) Geographical distribution of 6-year mean τ
0.55
over Hubei.
Based on the Hubei landform zoning map (figure 1(b)), landform is a key factor
affecting τ
0.55
spatial distribution. With a higher terrain on three sides, and open
central and southern areas, the overall province appears to be an incomplete basin
(http://www.hubei.gov.cn/hbgk/index.shtml). Similarly, the τ
0.55
distribution shows a
strong ‘basin’ effect (Luo et al. 2001): high values (τ
0.55
> 0.70) primarily concen-
trated in the plain lake areas, which are located at a lower altitude; plain and hillock
areas, hilly and hillock areas, and east low mountain-hilly areas in the northeast and
southeast Hubei surround the high-value areas. Diffusion of these high-concentration
aerosol particles stops and forms lower τ
0.55
value areas near Wudang Mountain,
Jingshan Mountain, Da Hong Mountain and Dabie Mountain. The lowest τ
0.55
areas
are found in the Shennongjia forest region in the northwest, and the cities of Enshi
in the southwest and Shiyan in the northwest. Thus, τ
0.55
spatial distribution is signif-
icantly influenced by terrain. Li et al. (2003) also proved the significant influence of
terrain on the AOD distribution pattern.
In addition, high humidity over the plain lake areas, where a number of lakes are
located, also contributes to the high τ
0.55
, as reported by Che et al. (2009). A higher
Downloaded by [Wuhan University] at 20:38 19 October 2014

Spatial and temporal variations of aerosols in central China 3677
33
32
30
31
29
109 110 111 112 113 114 115 116
Latitude (°N)
Longitude (°E)
(a)
33
32
30
31
29
Latitude (°N)
109 110 111 112 113 114 115 116
Longitude (°E)
(b)
N
0.60+
0.50
0.40
0.30
0.20
0.10
0.00
Figure 2. Fine (a) and coarse (b) particle aerosol distributions over Hubei Province during
2003–2008.
relative humidity could clearly cause the particles’ hygroscopic increase, which could
result in greater extinction and a larger volume of fine particles.
3.1.2 Geographic distribution of fine and coarse aerosols. Coarse-particle aerosol
distribution and fine-particle aerosol distribution over Hubei are obviously different
(figure 2). Fine particles (size < 0.6 µm) are mainly distributed in the central and
eastern regions of Hubei (figure 2(a)), where cities are densely populated and fuel con-
sumption from human activities such as industrial production, heating, and daily life
exhaust large sums of waste gas, creating a great deal of fine-particle aerosol. The
western region of Hubei shows relatively low emissions of fine-particle aerosol.
The spatial distribution of τ
0.55-coarse
(figure 2(b)) over Hubei is similar to the
multi-year average of τ
0.55
. In the southern region of Hubei, which features high con-
centrations of aerosols, τ
0.55-coarse
is also significantly higher than in other regions
(0.5–0.87). The southern region is followed by the northeastern and southeastern
(0.2–0.5), with western Hubei having the lowest concentration of aerosols (<0.2).
Coarse-particle aerosols in south-central Hubei may be composed of organic com-
pounds from urban/industrial emissions. Li et al. (2009) mentioned that a relatively
high proportion of coarse particles (η<0.3) was observed in metropolitan areas w ith-
out the influence of spring dust in south China, such as the Pearl River Delta (PRD)
and Yangtze River Delta (YRD), which can be explained by heavy urban/industrial
pollution-related coarse particles with a radius of more than 0.6 mm. A lower popula-
tion density and fewer human activities in the western region mean that there are far
fewer sources of coarse-particle aerosols, resulting in lower τ
0.55-coarse
in this region.
3.2 Aerosol temporal variation
3.2.1 Monthly variation. Within 1 year, the monthly mean τ
0.55
shows an increas-
ing and then a decreasing trend, and all peaks appear in May–June except in 2004
(figure 3). Since 2004, the monthly mean maximum τ
0.55
has been increasing year by
year, from 0.64 to 0.92 in 2008, which indirectly shows the trend of increasing aerosol
concentrations.
Between spring and summer (in May–June or in June–July), the monthly mean τ
0.55
value usually went down, especially in June–July 2003 and 2007, showing a decline
by 64% and 51% from 0.84 and 0.89 to 0.30 and 0.44, respectively. This phenomenon
may be due to the monsoon effect. Hubei is located in the typical East Asian monsoon
Downloaded by [Wuhan University] at 20:38 19 October 2014

3678 Y. J. Guo et al.
Oct
Jul
Apr
Jan
Oct
Jul
Apr
Jan
Oct
Jul
Apr
Jan
Oct
Jul
Apr
Jan
Oct
Jul
Apr
Jan
Oct
Jul
Apr
Jan
1.2
1.0
0.8
0.6
MODIS aerosol optical depth at 550 nm
0.4
0.2
0.0
2003 2004 2005 2006
Time (year)
2007 2008
Hubei, China 2003–2008
τ
0.55
τ
0.55-fine
Figure 3. Time series of monthly average τ
0.55
and τ
0.55-fine
over Hubei Province during
2003–2008.
zone; thus, during the summer monsoon season, strong cross-equatorial flows bring
clean air to China from the southern hemisphere, thus diluting aerosol concentrations.
Also, rain belts associated with the summer monsoon season move from southeast to
northern China during the June–August months, leading to a large wet deposition of
aerosols (Zhang et al. 2010).
The monthly mean τ
0.55-fine
of Hubei aerosols shows significant cyclic variation. It
goes down month-by-month in the winter half year (September–February) and goes
up in the summer half year (March–August). Aerosols are primarily coarse particles in
December–February, which may be caused by the burning of coal in winter (Kaufman
et al. 2002).
Generally, monthly mean τ
0.55
and τ
0.55-fine
peak simultaneously around June, except
in 2006, indicating that fine particles occupy the predominant proportion of aerosols
at this time. This result is in agreement with the study of Kim et al. ( 2007), in which
heavy fine-mode pollution particle loading was reported in June. It is interpreted that
this loading was caused by the contributions of various processes such as stagnant
synoptic meteorological patterns, secondary aerosol formation, hygroscopic growth of
hydrophilic aerosols due to enhanced relative humidity and smoke aerosols produced
by regional biomass burning (Kim et al. 2007).
Figure 3 shows that both time periods exhibit non-stationarities, which may indi-
cate that both obey the power-law distribution (long-range correlations). For instance,
in Varotsos et al. (2005), persistent power-law correlations from about 4 hours to 9
months were found in PM10 fluctuations in Athens and at the University of Maryland
Downloaded by [Wuhan University] at 20:38 19 October 2014

Spatial and temporal variations of aerosols in central China 3679
‘Supersite’ in East Baltimore. Similar behaviour was also found by Varotsos et al .
(2006) by applying detrended fluctuation analysis to the zonal mean daily Aerosol
Index (AI) values derived from satellite observations from 1979 to 2003 to search for
self-similarity properties.
3.2.2 Seasonal variation. The spatial distribution characteristics of the different
seasons (figure 4) are generally similar to that of multi-year distribution patterns, but
the changes between seasons are more obvious.
In spring, high τ
0.55
values are the largest of the year; most areas have significantly
higher mean τ
0.55
than they have in the other three seasons. The central plains and hills
are covered with high τ
0.55
values over 0.7, which further reflects the idea that spring
aerosols are mainly affected by terrain. The regional mean τ
0.55
values in spring are
higher than the corresponding multi-yearly mean values by about 11–26%. For exam-
ple, Wuhan, Ezhou, Xiantao and Tianmen are all covered with high τ
0.55
values of
above 0.95 in spring, and Qianjiang even reaches 0.99. The lowest mean value appears
in the Shennongjia forest region with only 0.23, and the extreme low value is only 0.17.
The wind blows from the southeast over high τ
0.55
areas, where aerosol particles may
be trapped by the mountains in the northern and northwestern portions of these areas.
In summer, the lower τ
0.55
area is slightly increased, whereas the higher value areas
are sharply decreased, which is due to the summer monsoon effect. The high-value
centres concentrate in three places: near the Wuhan and Ezhou areas, the Qianjiang
and Xiantao areas and part of Xiangfan. In lower value areas, there is little change
between seasonal mean τ
0.55
in spring and summer, but in high-value areas, there are
big changes. For example, the τ
0.55
value in Jingzhou drops from 0.92 in spring to 0.71
in summer; in addition, although the high-value centre in Wuhan still exists, the mean
33
32
30
31
29
109 110 111 112 113 114 115 116
Latitude (°N)
Longitude (°E)
Wuhan
Xiantao
Jingzhou
33
32
30
31
29
109 110 111 112 113 114 115 116
Latitude (°N)
Longitude (°E)
Wuhan
Jingzhou
33
32
30
31
29
109 110 111 112 113 114 115 116
Latitude (°N)
Ezhou
Wuhan
Xiangfan
Honghu
Jingzhou
Longitude (°E)
Latitude (°N)
33
32
30
31
29
109 110 111 112 113 114 115
Tianmen
Qianjiang
Jingzhou
116
Longitude (°E)
N
0.70
0.80
0.90
1.00+
0.60
0.50
0.40
0.30
0.20
0.10
0.00
(a)(b)
(c)(d)
Figure 4. Seasonal distributions of mean τ
0.55
over Hubei Province during 2003–2008 and
wind directions. (a) spring (March–May), (b) summer (June–August), (c) autumn (September–
November), (d) winter (December–February of the following year).
Downloaded by [Wuhan University] at 20:38 19 October 2014

3680 Y. J. Guo et al.
Table 1. Precipitation and temperature during 2000–2006 in Hubei.
Annual Spring Summer Autumn Winter
Total precipitation (mm) 905 302 446 210 128
Average temperature (
◦
C) 17.1 17.3 27.3 17.8 5.7
Source: Li et al. (2009).
value drops from 0.95 to 0.81. The total precipitat ion and average temperature are
at the highest levels during summer (table 1). The high temperature could obviously
cause the gas-to-particle conversion process (Che et al. 2009), and the precipitation
could easily remove atmospheric aerosols (Lee et al. 2007).
In autumn, the lower τ
0.55
area is slightly increased compared with that in sum-
mer, and the higher value area is decreased in the north-central part of the province.
In addition, the Xiangfan high-value centre has disappeared. The reason may be
the northeastern wind effect, which diffuses the dense aerosols over these high-value
centres. In most regions of the province, mean τ
0.55
in autumn is lower than the
multi-yearly mean value. Comparing the mean τ
0.55
of autumn to summer shows a
little decrease near the Wuhan and Ezhou areas. Near another high-value centre, the
Qianjiang and Xiantao areas, it shows little change, and in most other areas, it is
decreased. The Jingzhou area has been a new high-value centre where mean τ
0.55
is
about 0.79, with parts of the area even reaching 0.92. The Shennongjia forest region
shows the lowest value of only 0.12.
In winter, the lower τ
0.55
value area is increased in southeastern Hubei compared
with that in autumn, and the higher value area near Wuhan is decreased to the
annual minimum. Compared with the mean τ
0.55
of winter to autumn, the value
near Wuhan dropped to 0.62, lower than its multi-yearly mean value by 20%. The
Jingzhou and Xiantao areas are still a high-value centre, with mean τ
0.55
of 0.76 and
0.80, respectively.
3.3 Model analysis on AOD affecting factors
3.3.1 SOM Analysis. The results shown in figure 5 were obtained using the SOM
model. The component planes help to visualize the relationships between AOD and
other variables. The darker grey unit indicates higher values of the variable. Lower
values of annual mean τ
0.55
(<0.505), τ
0.55-coarse
(<0.29) and τ
0.55-fine
(<0.22) are
approximately corresponding to lower values of ln(population density) (<5.51). In
particular, the lowest AODs occur at the top right corner (the dark grey area) corre-
sponding to the lowest ln(population density). Thus, it reveals that τ
0.55
, τ
0.55-coarse
and
τ
0.55-fine
are all positively correlated with ln(population density).
The higher value districts of τ
0.55
(>0.505), τ
0.55-coarse
(>0.29) and τ
0.55-fine
(>0.22)
match the lower value districts of ln(elevation) (<5.27) and NDVI (<402), and vice
versa, which indicates the negative relevance between AODs and these two factors.
Comparing the colour patterns of all the component planes, it was found
that the ln(elevation) pattern is more similar to the three AOD patterns than
NDVI or ln(population density). Thus, it can be inferred that for annual AODs,
τ
0.55
/τ
0.55-coarse
/τ
0.55-fine
is more highly related to ln(elevation) than the other two
factors.
Downloaded by [Wuhan University] at 20:38 19 October 2014

Spatial and temporal variations of aerosols in central China 3681
6.42
6.97
0.288
0.220
0.153
In(population density)
In(elevation)
U-matrix
NDVI
5.27
3.57
285
402
519
0.0702
0.2900
0.5100
0.787
0.505
0.223
0.162
34
67.8
τ
0.55-fine
τ
0.55-coarse
τ
0.55
5.51
4.61
Figure 5. Visualization of simulating AOD and the three features: annual mean τ
0.55
, τ
0.55-coarse
and τ
0.55-fine
in 2003, elevation, NDVI and population density over Hubei Province by using the
SOM model. After removing the outliers and transforming elevation and population density to
natural logarithmic form, all the variables approximately conformed to the normal distribution.
3.3.2 Linear model analysis (unary linear regression and multiple linear regression).
Through Pearson correlation analysis, the annual mean τ
0.55
/τ
0.55-coarse
/τ
0.55-fine
of
Hubei is shown to be significantly negatively correlated with the ln(elevation) and
NDVI (p < 0.0001) and significantly positively correlated with ln(population density)
(p < 0.0001). These results coincide with SOM. The details of unary linear regression
(ULR) are summarized in figures 6–8 and tables 2–4.
For annual τ
0.55
, the three R
2
values reach up to 0.929, 0.680 and 0.656, respectively
(table 2), indicating that τ
0.55
is highly linear in relation to the three factors. With
the elevation and NDVI decreasing, or with population density increasing, the annual
average τ
0.55
rises. The three relevance ranks are as follows: ln(elevation) > NDVI >
ln(population density).
For aerosol modes, R
2
values also show that ln(elevation) affects τ
0.55-coarse
and
τ
0.55-fine
more than do the other two factors. Interestingly, R
2
values for coarse modes
(table 3) exhibit significantly higher values than that for fine modes (table 4), which
reveals that the three factors impact the coarse-particle aerosols more and may be an
indication of the different distribution characteristics of the two aerosol modes.
Moreover, multiple linear regressions (MLRs) on the three features have been pro-
posed (tables 2–4) to offer a more complete comparison with SOM. Each of the
regression coefficients for the specific dependent and independent variables presents
the same sign with ULR models, except for the coefficient between τ
0.55-fine
and NDVI,
Downloaded by [Wuhan University] at 20:38 19 October 2014

3682 Y. J. Guo et al.
1.00
0.80
0.60
0.40
0.20
0.00
200 300 400 500 600
NDVI
τ
0.55
(b)
1.00
0.80
0.60
0.40
0.20
0.00
2345
678
ln(elevation)
(a)
τ
0.55
45678
1.00
0.80
0.60
0.40
0.20
0.00
(c)
ln(population density)
τ
0.55
Figure 6. Relationships between τ
0.55
over Hubei Province in 2003 and (a) ln(elevation), (b)
NDVI and (c) ln(population density). The equations and determination coefficients (in brack-
ets) are (a) τ
0.55
= –0.153 ln(elevation) + 1.306 (R
2
= 0.929); (b) τ
0.55
= –0.003 NDVI + 1.594
(R
2
= 0.680) and (c) τ
0.55
= 0.265 ln(population density) – 0.980 (R
2
= 0.656).
which turns inversely from –0.000404 in ULR to 0.00021 in MLR. However, R
2
value
for τ
0.55-fine
here is only 0.447, far less than that for τ
0.55
and τ
0.55-coarse
, which rise up
to 0.930 and 0.911, respectively, meaning that regression coefficients for τ
0.55-fine
are
less credible than that for τ
0.55
and τ
0.55-coarse
. Consequently, MLR performances on
the correlations further prove the SOM results.
All the R
2
values are raised compared with ULR. It can be inferred that MLR is
more suitable than ULR for interpreting the relationships between annual aerosols
and distribution factors, indicating that aerosol distribution is influenced by the
complex interactions among various elements.
3.3.3 SOM and the linear model comparison analysis. As pointed out by Lampinen
and Kostiainen (2000), a simple use of the SOM may lead to an excessive number
of false hypotheses; thus, it is necessary to validate the hypotheses using other sta-
tistical methods. Comparing the analysis on SOM to the linear model, the relevance
Downloaded by [Wuhan University] at 20:38 19 October 2014

Spatial and temporal variations of aerosols in central China 3683
200 300 400 500 600
NDVI
0.40
0.50
0.60
0.70
0.30
0.20
0.10
0.00
τ
0.55-coarse
(b)
2345
678
ln(elevation)
0.40
0.50
0.60
0.70
0.30
0.20
0.10
0.00
τ
0.55-coarse
(a)
4.00
5.00 6.00 7.00
8.00
ln(population density)
0.40
0.50
0.60
0.70
0.30
0.20
0.10
0.00
τ
0.55-coarse
(c)
Figure 7. Relationships between τ
0.55-coarse
over Hubei Province in 2003 and (a) ln(elevation),
(b)NDVIand(c) ln(population density). The equations and determination coefficients (in
brackets) are (a) τ
0.55-coarse
= –0.126 ln(elevation) + 0.930 (R
2
= 0.903); (b) τ
0.55-coarse
= –0.002
NDVI + 1.187 (R
2
= 0.692) and (c) τ
0.55-coarse
= 0.223 ln(population density) – 0.976 (R
2
=
0.647).
results are completely consistent. So, it is confirmed that SOM can truly reflect the
relationships between the spatial distribution of AOD over Hubei and the affecting
factors.
Regarding the influence of elevation, the spatial distribution of the annual average
τ
0.55
in 2003, which is substantially similar to the 6-year mean τ
0.55
, also have quite
similar patterns with elevation. As mentioned above, figures 1(b) and (c) show the
remarkable qualitative relationship between τ
0.55
and landform, while results of SOM
strengthen this point of view, and further, the linear models make this connection
clear quantificationally. The higher surrounding terrain can prevent the horizontal
dispersion of concentrated air pollutants, resulting in significantly higher τ
0.55
over
lower areas than that over the surrounding regions. For example, the Huangpi district
of Wuhan City adjoins on the north side of the downtown Wuhan and on the south
side of the Dawu district of Xiaogan City. As the average elevations of these three
Downloaded by [Wuhan University] at 20:38 19 October 2014

3684 Y. J. Guo et al.
2345678
ln(elevation)
0.40
0.50
0.30
0.20
0.10
0.00
τ
0.55-fine
(a)
200 300 400 500 600
NDVI
0.40
0.50
0.30
0.20
0.10
0.00
τ
0.55-fine
(b)
(c)
0.40
0.50
0.30
0.20
0.10
0.00
τ
0.55-fine
4.00
5.00 6.00 7.00
8.00
ln(population densit
y
)
Figure 8. Relationships between τ
0.55-fine
over Hubei Province in 2003 and (a) ln(elevation), (b)
NDVI and (c) ln(population density). The equations and determination coefficients (in brack-
ets) are (a) τ
0.55-fine
= –0.027 ln(elevation) + 0.337 ( R
2
= 0.448); (b) τ
0.55-fine
= –0.000404 NDVI
+ 0.407 (R
2
= 0.260) and (c) τ
0.55-fine
= 0.043 ln(population density) – 0.004 (R
2
= 0.275).
regions rise from the south to the north (25.5, 72.9, 174.8 m), their annual average
τ
0.55
decreases sequentially and sharply from 0.86 and 0.73 to 0.53.
Regarding the influence of the vegetation index, the annual AOD is lower where
NDVI is higher, and vice versa. There are two possible reasons: on the one hand,
areas with denser vegetation are less affected by human activities, and atmospheric
pollution is not so severe, resulting in less emissions of urban aerosol; on the other
hand, more forest fires burning biomass occur in these areas, which generate more
fine particles of aerosol (Li et al. 2010). Combining the two aspects, the vegetation
index and AOD spatial distribution are shown to be negatively related, so the former
reason is more dominant. For example, regions covered with higher NDVI such as the
Shennongjia forest region (550), Hefeng (511) and WuFeng (523), have a relatively low
annual average τ
0.55
of 0.15, 0.20 and 0.22, respectively, whereas regions characterized
by lower NDVI, such as downtown Wuhan (249) and downtown Huangshi (275), have
average
0.555
values that are higher, being 0.86 and 0.76, respectively.
Downloaded by [Wuhan University] at 20:38 19 October 2014

Spatial and temporal variations of aerosols in central China 3685
Table 2. Statistical models for annual mean τ
0.55
andpredictionaccuracy.
Model Prediction accuracy (RMSE)
No. Explanatory variable Regression model Pairs R
2
2004 2005 2006 Total
ULR 1 Elevation y = –0.153 lnx + 1.306 86 0.929 – – – –
2NDVI y = –0.003 x + 1.594 86 0.680 0.26 0.33 0.33 0.31
3 Population density y = 0.265 lnx – 0.980 79 0.656 0.13 0.15 0.16 0.15
MLR4Allthree y = –0.138 lnx
1
– 0.00033 x
2
+ 0.004 lnx
3
+ 1.346 79 0.930 0.05 0.09 0.12 0.09
Notes: y denotes annual mean τ
0.55
for each region. x
(1,2,3)
denote the three explanatory variables: elevation, NDVI and population density, respectively.
ULR, unary linear regression; MLR, multiple linear regression.
R
2
, determination coefficient. All of the corresponding p values are <0.0001.
Downloaded by [Wuhan University] at 20:38 19 October 2014

3686 Y. J. Guo et al.
Table 3. Statistical models for annual mean τ
0.55-coarse
andpredictionaccuracy.
Model Prediction accuracy (RMSE)
No. Explanatory variable Regression model Pairs R
2
2004 2005 2006 Total
ULR 5 Elevation y = –0.126 lnx + 0.930 86 0.903 – – – –
6NDVI y = –0.002 x + 1.187 86 0.692 0.09 0.11 0.09 0.10
7 Population density y = 0.223 lnx – 0.976 79 0.647 0.11 0.11 0.11 0.11
MLR 8 All three y = –0.105 lnx
1
– 0.00054 x
2
+ 0.003 lnx
3
+ 1.025 79 0.911 0.05 0.05 0.05 0.05
Note: y denotes annual mean τ
0.55-coarse
for each region. All of the corresponding p values are <0.0001.
Downloaded by [Wuhan University] at 20:38 19 October 2014

Spatial and temporal variations of aerosols in central China 3687
Table 4. Statistical models for annual mean τ
0.55-fine
andpredictionaccuracy.
Model Prediction accuracy (RMSE)
No. Explanatory variable Regression model Pairs R
2
2004 2005 2006 Total
ULR 9 Elevation y = –0.027 lnx + 0.337 86 0.448 – – – –
10 NDVI y = –0.000404 x+ 0.407 86 0.260 0.09 0.05 0.10 0.10
11 Population density y = 0.043 lnx – 0.004 79 0.275 0.11 0.05 0.09 0.09
MLR12Allthree y = –0.034 lnx
1
+ 0.00021 x
2
+ 0.00050 lnx
3
+ 0.321 79 0.447 0.05 0.04 0.08 0.09
Note: y denotes annual mean τ
0.55-fine
for each region. All of the corresponding p values are <0.0001.
Downloaded by [Wuhan University] at 20:38 19 October 2014

3688 Y. J. Guo et al.
Regarding the influence of population density, the spatial distribution patterns of
the annual AOD over Hubei generally appear higher in the east and lower in the west,
which aligns with the population distribution characteristics of the areas with heavy
distribution in the east and sparse distribution in the west. Also, the high and low τ
0.55
value centres are consistent with the central area of population density, as reported
by Kaufman et al. (2002), stating, ‘aerosol emission from human activities is approx-
imately positive related with population density’. For example, the heavily populated
area of downtown Wuhan (2813 persons/km
2
) is covered by high τ
0.55
values up to
0.86 in 2003, whereas there are lower annual average τ
0.55
values in the sparsely popu-
lated regions such as the Shennongjia forest region (24 persons/km
2
) and Hefeng (75
persons/km
2
), measuring only 0.15 and 0.20, respectively.
3.3.4 Model predictions. The stability of the results has been evaluated based on
the predictions of linear regression. Predicted annual τ
0.55
/τ
0.55-coarse
/τ
0.55-fine
from
2004 to 2006 was computed by ULR and MLR models. The root mean square error
(RMSE) between actual and predicted values was calculated to indicate the prediction
accuracy.
The RMSE values for ULR in tables 2–4 are found to be poor, and for each year
(from 2004 to 2006) range from 0.05 to 0.33, suggesting that, although the ULR
could describe the correlation between aerosols and the factors well, more compli-
cated relationships may exist and more work is imperative to improve the accuracy of
the models. Interestingly, the RMSE values for τ
0.55-coarse
/τ
0.55-fine
ranging from 0.05 to
0.11 are lower than that for τ
0.55
, indicating that, due to different contributions to this
region’s AOD between coarse and fine modes, it could be better to probe the factors’
complex effects on aerosols by separating the two modes.
In contrast, predictions based on the MLR modelling exhibit relatively high accu-
racy, with the RMSE being mostly below 0.10 (tables 2–4, also see figure 9), suggesting
that MLR is more suitable than ULR for interpreting the relationships between AOD
and distribution factors. It is inferred that aerosol distribution is influenced by the
complex interactions among various elements.
The MLR model for τ
0.55-coarse
exhibits a comparatively lower and more stationary
RMSE than that for τ
0.55-fine
, indicating that the selected factors are more effective
for explaining coarse aerosol modes distribution. Moreover, MLR shows relatively
high-prediction accuracy compared with ULR models. To some degree, explaining
the linear correlations between aerosols and the distribution factors offers addi-
tional proof for the aforementioned inference that MLR is a more reasonable method
than ULR.
4. Conclusions
The 6-year mean τ
0.55
and aerosol size over Hubei show obvious spatial distribution
characteristics. Higher τ
0.55
value areas are mainly located in the south-central area
of the province, with lower value areas in the southeast and northeast and the lowest
value areas in the west, which reveals the significant influence of landform. The high-
value centres of the 6-year average τ
0.55
are found in Wuhan, Ezhou and Jingzhou.
Higher value areas of τ
0.55-fine
are mainly distributed in the central and eastern portions
of Hubei, whereas τ
0.55-coarse
is distributed in central and southern Hubei.
The temporal variation of τ
0.55
over Hubei presents obvious regularity. For yearly
variation, average τ
0.55
shows an ascendant trend. For seasonal variation, in spring,
Downloaded by [Wuhan University] at 20:38 19 October 2014

Spatial and temporal variations of aerosols in central China 3689
1.00
0.80
0.60
0.60
0.50
0.40
0.30
0.20
0.10
0.00
0.40
0.20
0.00
0.70
0.50
0.30
0.10
–0.10
0
20 40 60 80 100 120 140
Serial number
160 180 200 220 240 260
0
20 40 60 80 100
Actual values
Predicted values
Actual values
Predicted values
Actual values
Predicted values
120 140
Serial number
160 180 200 220 240 260
0
20 40 60 80 100 120 140
Series number
160 180 200 220 240 260
τ
0.55
τ
0.55-coarse
τ
0.55-fine
(a)
(b)
(c)
Figure 9. Comparison of the actual and predicted (a) τ
0.55
,(b) τ
0.55-coarse
and (c) τ
0.55-fine
for 86
zonal annual values from 2004 to 2006 by MLR models.
the high τ
0.55
value area is the largest of the year; most areas have significantly higher
mean τ
0.55
in spring than that in the other three seasons. In summer, the lower τ
0.55
area is slightly increased with the value changing little, while higher value areas sharply
decrease with the value also being greatly reduced. In autumn, the lower value τ
0.55
area is slightly increased, and the higher value area is decreased in the north-central
region of the province. In winter, the lower τ
0.55
value area is increased in southeastern
Hubei, and the higher value area near Wuhan is decreased to the annual minimum. For
monthly variation, mean τ
0.55
shows approximately increasing and then decreasing
trends during 1 year, and all peaks of a year appear during May–June except in 2004;
the mean τ
0.55-fine
goes down month by month in the winter-half year and goes up in
the summer-half year.
Downloaded by [Wuhan University] at 20:38 19 October 2014

3690 Y. J. Guo et al.
SOM and linear model analysis both show that the spatial distribution of the
annual average AOD in Hubei is influenced by the complex interactions among vari-
ous elements. The AODs, which include τ
0.55
, τ
0.55-coarse
and τ
0.55-fine
, are significantly
negatively related to ln(elevation) and NDVI and positively related to ln(population
density). The ln(elevation) factor affects AODs more than the other two factors.
Compared to fine-particle aerosols, the selected three factors have more impact on
the coarse particles.
Acknowledgements
This research was supported by ‘the Fundamental Research Funds for the Central
Universities’ (201120502020002). The SRTM data set is provided by the International
Scientific & Technical Data Mirror Site, Computer Network Information Centre of
Chinese Academy of Sciences (http://datamirror.csdb.cn). The GIMMS NDVI data
set is provided by the Environmental and Ecological Science Data Centre for West
China, which is sponsored by the National Natural Science Foundation of China
(http://westdc.westgis.ac.cn). Wind direction data are from the Navy Operational
Global Atmospheric Prediction System (NOGAPS) Surface Winds.
References
ANDERSON, T.L., CHARLSON, R.J., BELLOUIN, N., BOUCHER, O., CHIN, M., CHRISTOPHER,
S.A., H
AYWOOD, J., KAUFMA N, Y.J., KINNE, S., OGREN, J.A., REMER, L .A.,
T
AKEMURA, T., TANRÉ, D., TORRES, O., TREPTE, C.R., WIELICKI, B.A., WINKER,
D.M. and Y
U, H.B., 2005, An “A-Train” strategy for quantifying direct climate forc-
ing by anthropogenic aerosols. Bulletin of the American Meteorological Society, 86,
pp. 1795–1809.
A
NDREAE, M.O., CHARLSON, R.J., BRUYNSEELS, F., STORMS, H., G RIEKEN,R.V.and
M
AENHAUT, W., 1986, Internal mixture of sea salt, silicates, and excess sulfate in marine
aerosols. Science, 232, pp. 1620–1623.
B
RÉON, F.M., TANRÉ,D.andGENEROSO, S., 2002, Aerosol effect on cloud droplet size
monitored from satellite. Science, 295, pp. 834–838.
C
HE, H.Z., ZHANG, X.Y., ALFRARO, S., CHATENET, B., GOMES,L.andZHAO, J.Q., 2009,
Aerosol optical properties and its radiative forcing over Yulin, China in 2001 and 2002.
Advances in Atmospheric Sciences, 26, pp. 564–576.
C
ÉRÉGHINO,R.andPARK, Y.S., 2009, Review of the self-organizing map (SOM) approach in
water resources: commentary. Environmental Modelling & Software, 24, pp. 945–947.
C
RUTZEN, P.J. and ANDREAE, M.O., 1990, Biomass burning in the tropics: impact on atmo-
spheric chemistry and b iogeochemical cycles. Science, 250, pp. 1669–1678.
D
EY,S.andTRIPATHI, S.N., 2008, Aerosol direct radiative effects over Kanpur in the
Indo-Gangetic basin, northern India: long-term (2001–2005) observations and impli-
cations to regional climate. Journal of Geophysical Research, 113, D04212, doi:
10.1029/2007JD009029.
E
DITORIAL COMMITTEE OF NATURAL RESOURCES SERIES OF CHINA (Ed.), 1995, A Series of
China’s Natural Resources: Hubei Volume (Beijing: China Environmental Science Press).
H
AN, X., ZHANG, M.G., HAN, Z.W., XIN, J.Y., WANG, L.L., QIU, J.H. and LIU, Y.J., 2010,
Model analysis of aerosol optical depth distributions over East Asia. Science China
Earth Sciences, 53, pp. 1079–1090.
H
UBEI STATISTICS BUREAU (Ed.), 2004–2007, Hubei Statistical Yearbook 2004–2007 (Beijing:
China Statistics Press).
K
AUFMAN, Y.J., TANRÉ,D.andBOUCHER, O., 2002, A satellite view of aerosols in the climate
system. Nature, 419, pp. 215–223.
Downloaded by [Wuhan University] at 20:38 19 October 2014

Spatial and temporal variations of aerosols in central China 3691
KIM, S.W., YOON, S.C., KIM,J.andKIM, S.Y., 2007, Seasonal and monthly variations
of columnar aerosol optical properties over east Asia determined from multi-year
MODIS, LIDAR, and AERONET Sun/sky radiometer measurements. Atmospheric
Environment, 41, pp. 1634–1651.
K
INNE, S., SCHULZ, M., TEXTOR, C., GUIBERT, S., BALKANSKI, Y., BAUER, S.E., BERNTSEN,
T., B
ERGLEN, T.F., BOUCHER, O., CHIN, M., COLLINS, W., DENTENER, F., DIEHL,T.,
E
ASTER, R., FEICHTER, J., FILLMORE, D., GHAN, S., GINOUX, P., GONG, S., GRINI,
A., H
ENDRICKS, J., HERZOG, M., HOROWITZ, L., ISAKSEN, I., IVERSEN, T., KIRKAVÅG,
A., K
LOSTER, S., KOCH, D., KRISTJANSSON, J.E., KROL, M., LAUER, A., LAMARQUE,
J.F., L
ESINS, G., LIU, X., LOHMANN, U., MONTANARO, V., MYHRE, G., PENNER, J.E.,
P
ITARI, G., REDDY, S., SELAND, O., STIER, P., TAKEMURA,T.andTIE, X., 2006, An
AeroCom initial assessment – optical properties in aerosol component modules of
global models. Atmospheric Chemistry and Physics, 6, pp. 1815–1834.
K
OHONEN,T.andMEMBER, S., 1990, The self-organizing map. Proceedings of the IEEE, 78,
pp. 1464–1480.
K
OHONEN, T., OJA, E., SIMULA, O., VISA,A. and KANGAS, J., 1996, Engineering applications
of the self-organizing map. Proceedings of the IEEE, 84, pp. 1358–1384.
L
AMPINEN,J.andKOSTIAINEN, T., 2000, Self-organizing map in data-analysis – notes on
overfitting and overinterpretation. In 8th E uropean Symposium on Artificial Neural
Networks, 26–28 April 2000, Bruges, Belgium (Brussels: D-Facto Publication), pp.
239–244.
L
EE, K.H., KIM, Y.J., VON HOYNINGEN-HUENE,W.andBURROW, J.P., 2007, Spatio-temporal
variability of satellite-derived aerosol optical thickness over Northeast Asia in 2004.
Atmospheric Environment, 41, pp. 3959–3973.
L
I, B.G., YUAN, H.S., FENG,N.andTAO, S., 2010, Spatial and temporal variations of aerosol
optical depth in China during the period from 2003 to 2006. International Journal of
Remote Sensing, 31, pp. 1801–1817.
L
I, C.C., MAO, J.T., LAU, K., CHEN, J.C., YUAN, Z.B., LIU, X.Y., ZHU, A.H. and LIU, G.Q.,
2003, Characteristics of distribution and seasonal variation of aerosol optical depth in
eastern China with MODIS products. Chinese Science Bulletin, 48, pp. 2488–2495.
L
I, S., TANG,Z.Q.andKUANG, Y.H., 2009, On the changing trend of temperature and
precipitation in Hubei in recent 50 years. Journal of Anhui Agriculture Science, 37,
pp. 1652–1655.
L
UO, Y.F., LÜ, D.R., HE, Q., LI,W.L.andZHOU, X.J., 2000, Characteristics of atmospheric
aerosol optical depth variation over China in recent 30 years. Chinese Science Bulletin,
45, pp. 1328–1334.
L
UO, Y.F., LÜ, D.R., ZHOU, X.J., LI,W.L.andHE, Q., 2001, Characteristics of the spatial
distribution and yearly variation of aerosol optical depth over China in last 30 years.
Journal of Geophysical Research, 106, pp. 14501–14513.
M
INISTRY OF CIVIL AFFAIRS OF THE PEOPLE’S REPUBLIC OF CHINA (Ed.), 2009, The
Administrative Divisions of the People’s Republic of China 2009 (Beijing: China Social
Sciences Press).
O
BU-CANN, K., MORITA, Y., FUJIMURA, K.., TOKUTAKA, H., OHKITA,M.andINUI,M.,
2001, Data mining of power transformer database using Self-Organizing Maps. In
International Conference on Info-tech & Info-net, 29 October–1 November 2001, Beijing,
China (Piscataway, NJ: IEEE Institute of Electrical and Electronics Engineers, Inc. and
People’s Posts & Telecommunications Publishing House), pp. 44–49.
P
ROSPERO, J.M., GINOUX, P., TORRES, O., NICHOLSON, S.E. and GILL, T.E., 2002,
Environmental characterization of global sources of atmospheric soil dust identified
with the Nimbus 7 Total Ozone Mapping Spectrometer (TOMS) absorbing aerosol
product. Reviews of Geophysics, 40, pp. 1002.
Downloaded by [Wuhan University] at 20:38 19 October 2014

3692 Y. J. Guo et al.
QUEROL, X., ZHUANG, X.G., ALASTUEY, A., VIANA, M., LV, W.W., WANG, Y.X., LÓPEZ,A.,
Z
HU, Z .C., WEI,H.M.andXU, S.Q., 2006, Speciation and sources of atmospheric
aerosols in a highly industrialized emerging mega-city in Central China. Journal of
Environmental Monitoring, 8, pp. 1049–1059.
R
AMANATHAN, V., CRUTZEN, P.J., KIEHL, J.T. and ROSENFELD, D., 2001, Aerosols, climate, and
the hydrological cycle. Science, 294, pp. 2119–2124.
R
ANDERSON, J.T., LIU, H., FLANNER, M.G., CHAMBERS, S.D., JIN, Y., HESS, P.G., PFISTER,G.,
M
AC K , M.C., TRESEDER, K.K., WELP, L.R., CHAPIN, F.S., HARDEN, J.W., GOULDEN,
M.L., L
YONS , E., NEFF, J.C., SCHUUR, E.A.G. and ZENDER, C.S., 2006, The impact of
boreal forest fire on climate warming. Science, 314, pp. 1130–1132.
R
OSENFELD, D., LOHMANN, U., RAGA, G.B., O’DOW D, C.D., KULMALA, M., FUZZI,S.,
R
EISSELL,A.andANDREAE, M.O., 2008, Flood or drought: How do aerosols affect
precipitation? Science, 321, pp. 1309–1313.
T
UCKER, C.J., PINZON,J.E.andBROWN, M.E., 2004, Global Inventory Modeling And Mapping
Studies, NA94apr15b.n11-VIg, 2.0 (College Park, MD: Global Land Cover Facility,
University of Maryland).
V
AN DONKELAAR, A., MARTIN, R.V., B RAUER, M., KAHN, R., LEVY, R., VERDUZCO,C.
and V
ILLENEUVE, P.J., 2010, Global estimates of ambient fine particulate matter con-
centrations from satellite-based aerosol optical depth: development and application.
Environmental Health Perspectives, 118, pp. 847–855.
V
AROTSOS, C., O NDOV,J.andEFSTATHIOU, M., 2005, Scaling properties of air pollution in
Athens, Greece and Baltimore, Maryland. Atmospheric Environment, 39, pp. 4041–4047.
V
AROTSOS, C.A., ONDOV, J.M., CRACKNELL, A.P., EFSTATHIOU, M.N. and ASSIMAKOPOULOS,
M.-N., 2006, Long-range persistence in global Aerosol Index dynamics. International
Journal of Remote Sensing, 27, pp. 3593–3603.
V
ESANTO, J., 2002, Data exploration process based on the Self-Organizing Map. PhD thesis,
Helsinki University of Technology, Espoo, Finland.
V
ESANTO,J.andAHOLA, J., 1999, Hunting for correlations in data using the Self-Organizing
Map. In International ICSC Congress on Computational Intelligence Methods and
Applications (CIMA ’99), 22–25 June 1999, New York, NY (Canada/Switzerland:
ICSC Academic Press), pp. 279–285.
Y
U, H., KAUFMAN, Y.J., CHIN, M., FEINGOLD, G., REMER, L.A., ANDERSON, T.L.,
B
ALKANSKI, Y., BELLOUIN, N., BOUCHER, O., CHRISTOPHER, S., DECOLA, P., KAHN,
R., K
OCH, D., LOEB, N., R EDDY, M.S., SCHULZ, M., TAKEMURA,T.andZHOU,M.,
2006, A review of measurement-based assessments of the aerosol direct radiative effect
and forcing. Atmospheric Chemistry and Physics, 6, pp. 613–666.
Z
HANG, L., LIAO,H.andLI, J.P., 2010, Impacts of Asian summer monsoon on seasonal and
interannual variations of aerosols over eastern China. Journal of Geophysical Research,
115, D00K05, doi: 10.1029/2009JD012299.
Z
HANG, L., SCHOLZ, M., MUSTAFA,A.andHARRINGTON, R., 2008, Assessment of the nutrient
removal performance in integrated constructed wetlands with the self-organizing map.
Water Research, 42, pp. 3519–3527.
Downloaded by [Wuhan University] at 20:38 19 October 2014