Content uploaded by Li Zhang
Author content
All content in this area was uploaded by Li Zhang on Jun 18, 2020
Content may be subject to copyright.
Poverty assessment using DMSP/OLS night-time light satellite
imagery at a provincial scale in China
Wen Wang
a,1
, Hui Cheng
a,1
, Li Zhang
b,⇑
a
School of Environmental and Natural Resources, Renmin University of China, Beijing 100872, China
b
Department of Geography, King’s College London, Strand, London WC2R 2LS, UK
Received 6 September 2011; received in revised form 29 January 2012; accepted 30 January 2012
Available online 8 February 2012
Abstract
All countries around the world and many international bodies, including the United Nations Development Program (UNDP), United
Nations Food and Agricultural Organization (FAO), the International Fund for Agricultural Development (IFAD) and the Interna-
tional Labor Organization (ILO), have to eliminate rural poverty. Estimation of regional poverty level is a key issue for making strategies
to eradicate poverty. Most of previous studies on regional poverty evaluations are based on statistics collected typically in administrative
units. This paper has discussed the deficiencies of traditional studies, and attempted to research regional poverty evaluation issues using
3-year DMSP/OLS night-time light satellite imagery. In this study, we adopted 17 socio-economic indexes to establish an integrated pov-
erty index (IPI) using principal component analysis (PCA), which was proven to provide a good descriptor of poverty levels in 31 regions
at a provincial scale in China. We also explored the relationship between DMSP/OLS night-time average light index and the poverty
index using regression analysis in SPSS and a good positive linear correlation was modelled, with R
2
equal to 0.854. We then looked
at provincial poverty problems in China based on this correlation. The research results indicated that the DMSP/OLS night-time light
data can assist analysing provincial poverty evaluation issues.
Ó2012 COSPAR. Published by Elsevier Ltd. All rights reserved.
Keywords: DMSP/OLS night-time light; Provincial scale; Socio-economic development; Principal component analysis; Poverty index
1. Introduction
Poverty is a general term describing living conditions
that are detrimental to health, comfort, and economic
development (Elvidge et al., 2009). After 30 years of eco-
nomic transformation, China has now become the second
largest economy and the second largest trading nation in
the world according to recent statistics of the World Bank
and the World Trade Organisation. China’s Gross Domes-
tic Product (GDP) has increased from 268.3 billion dollars
to 5.3 trillion dollars since 1978, meanwhile the gap
between Western China and other regions has been increas-
ing (Li et al., 2008). Not everyone has equally shared the
fruits of Chinese economic reform. Poverty is still a signif-
icant problem in China and it needs a long time and great
efforts to be solved. So accurate assessments of regional
poverty levels are essential for the central government
and local policy makers to obtain reliable up-to-date data
of the socio-economic situation and tackle regional
inequality problems.
Traditionally, regional socio-economic development
assessment is based on statistics collected by local govern-
ments. GDP is the most popular indicator of economic per-
formance (Sutton and Costanza, 2002) and has been used
in a wide range of socio-economic development studies in
China. For example Jian et al. (1996) adopted GDP data
to analyse the regional inequality trends. Li et al. (2004)
applied it to evaluate economic standards of 31 provinces
0273-1177/$36.00 Ó2012 COSPAR. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.asr.2012.01.025
⇑
Corresponding author. Tel.: +44 (0) 20 78482692; fax: +44 (0) 20
78482287.
E-mail addresses: wen_wang2000@hotmail.com (W. Wang), chenghui-
whu@163.com (H. Cheng), li.zhang@kcl.ac.uk (L. Zhang).
1
Tel.: +86 (0) 10 88893061; fax: +86 (0) 10 62511645.
www.elsevier.com/locate/asr
Available online at www.sciencedirect.com
Advances in Space Research 49 (2012) 1253–1264
(or municipalities). Jin (2007) used GDP as one of the
urban economic vitality indexes for quantitative economic
analysis of 50 Chinese cities. However, there are limits to
this type of data, as economic census is usually collected
once every five years in China and it takes substantial man-
power and generates huge amount of economic costs. It
also needs a long period to update existing data and some-
times may become impossible because of various reasons,
e.g. change of local administrative units. It cannot meet
special demands either due to the lack of spatial
information.
In comparison to traditional methods, satellite remote
sensing has an advantage to provide efficient and accuracy
spatial data for various physical and social science research
purposes due to its high temporal resolution and extensive
spatial coverage. Satellite imagery has been recognised to
be capable of mapping and analyzing socio-economic
related issues with high accuracy since the late 1960s (e.g.
Tobler 1969; Welch 1980; Foster 1983) and the night-time
radiance data has been proven to be capable of providing
strong estimation of population, GDP and electricity con-
sumption based on the strong correlation between lights
and human activities (Elvidge et al., 1997a,b). It shows a
good potential in regional poverty analysis. The night-time
light images are collected by the US Air Force Weather
Agency and processed at the National Geophysical Data
Centre (NGDC) of the National Ocean and Atmosphere
Administration (NOAA) using Defence Meteorological
Satellite Program (DMSP) Operational Linescan System
(OLS) data. NGDC combines the cloud-free portions of
night-time orbital segments over a full year to generate
annual night-time lights products (Elvidge et al., 1997a,b
Elvidge et al., 2001) that have been used in a range of stud-
ies, such as GDP estimation and energy consumption anal-
ysis (Elvidge et al., 1997a,b; Elvidge et al., 2001), mineral
in-use stocks (Takahashi et al., 2010) and income proxy
(e.g. Sutton and Costanza, 2002). Nakayama and Tanaka
(1983) explored the relationship between the light diameter
of a city and its economy. Elvidge et al. (1997b) Elvidge
et al. (1999) then found a close relationship between
night-time light and human activity such as energy con-
sumption and the important economic activity indicator
GDP. The strong relationship between economic activities
and CO
2
emissions with the total lit area were also revealed
and mapped by (Doll et al., 2000). Later, Doll (2003) used
the cumulative radiance value in the radiance-calibrated
night-time image to develop an area-GDP relationship at
a national scale for the United Kingdom. Sutton et al.
(2007) made a similar attempt to estimate sub-national
GDP for the United States, China, India and Turkey.
Elvidge et al. (2009) produced a global poverty map using
a poverty index calculated by dividing population count
(LandScan 2004) by the brightness of satellite observed
lighting (DMSP/OLS night-time lights). The main socio-
economic factors considered by most of these studies were
population, energy consumption, greenhouse gas emis-
sions, urban sprawl, forest fires monitoring and light pollu-
tion. There are no studies on poverty issues of China at a
provincial scale using remote sensing data so far.
This study combines the 3-year DMSP/OLS night-time
light data with other socio-economic statistical indicators
to establish DMSP/OLS night-time average light indexes
at a provincial scale in China and analyse the relationship
between them and an integrated poverty index to explore
the spatially irregular distribution of social wealth of
China. It may contribute to the effort of a more balanced
regional development in China.
2. Data and methods
2.1. Study area
31 provinces and municipalities in mainland China
(Fig. 1) have been selected to carry out this study. The
rapid economical growth of these 31 regions in the last
30 years has drawn worldwide attention and made China
the world’s second largest economy according to the World
Bank. Meanwhile, the uneven economic growth rate has
caused apparent economical inequality amongst different
regions and built up a big gap between the west and the
east (Li et al., 2008). The inequality is now recognised as
a great barrier for the future sustainable economical devel-
opment of the nation.
2.2. Socio-economic statistical data and fundamental
geographic data
The 3-year socio-economic statistical data (from 2007 to
2009) for the selected 31 provinces and municipalities is
obtained from the National Bureau of Statistics of China.
The fundamental geographic administrative boundaries in
a vector format (ESRI shapefiles) were downloaded from
the website of the National Fundamental Geographic
Information System and their projections were reprojected
into the China Lambert Conformal Conic Projection using
ESRI ArcGIS 9.3. Both GDP data and administrative data
were then integrated into a geospatial database for further
analysis.
2.3. DMSP/OLS night-time lights data
DMSP/OLS night-time data are annual night-time
cloud-free image composites of lights of the globe collected
by the DMSP/OLS sensors on a low-earth orbiting satellite
(at 833 km altitude above earth). DMSP operates satellites
in sun-synchronous orbits with night-time overpasses at 8–
10 pm local time. With a swath width of 3000 km and 14
orbits per day, each OLS instrument is capable of generat-
ing a complete coverage of night-time data in a 24-hour
period. The OLS is an oscillating scan radiometer with
two spectral bands. The visible band straddles the visible
and near-infrared (VNIR) portion of the spectrum (0.5–
0.9 lm) and the thermal band covers the 10.5–12.5 lm
spectrum range. At night, the visible band is intensified
1254 W. Wang et al./ Advances in Space Research 49 (2012) 1253–1264
using a photomultiplier tube (PMT) to permit detection of
clouds illuminated by moonlight. The light intensification
enables observation of faint sources of VNIR emissions
at night on the earth’s surface, including cities, towns, vil-
lages, gas flares, heavily lit fishing boats, and fires (Elvidge
et al., 1997a,b). The low-light-sensing capabilities of the
OLS at night permit the measurement of radiances down
to 10
9
W/cm
2
/sr. NGDC has developed algorithms to
remove areas contaminated by sunlight, moonlight, solar
glare and fires and produced high quality global cloud-free
composites of DMSP night-time light emissions with aver-
age intensity (digital number recorded at the sensor) since
1994 (Elvidge et al., 1997a,b). The spatial resolution of
the data is reasonably high, 2.8 km at full mode and
0.56 km at fine mode. The high contrast and spatial resolu-
tion of the data makes it a tool to identify regions of
intense human activity (Croft, 1978).
The version 4 DMSP/OLS night-time image products
from 2007 to 2009 (30 arc seconds spatial resolution),
released by NOAA-NGDC in 2010 at http://
www.ngdc.noaa.gov/dmsp/ downloadV4composites.html,
were used for this study. The data is derived by multiplying
the average visible band digital number (DN) of cloud-free
light detections with the percent frequency of light detec-
tion. The inclusion of the percent frequency of detection
term normalizes the resulting digital values for variations
in the persistence of lighting. The original global DMSP/
OLS night-time light image of 2009 is shown in Fig. 2.
Fig. 1. Administrative map of China.
W. Wang et al. / Advances in Space Research 49 (2012) 1253–1264 1255
2.4. Methods
2.4.1. Establishment of DMSP/OLS night-time average light
index (ALI)
The DMSP/OLS night-time light imageries of China
(Fig. 3) were extracted from the global DMSP/OLS
night-time light data using the Extraction Tool of Spatial
Analyst of ESRI ArcGIS 9.3 and the data was then repro-
jected to the China Lambert Conformal Conic Projection
from the original geographic projection (Lat/Lon) using
nearest neighbour resampling algorithm. Each pixel in
the imageries has a DN value ranging from 0 to 63. Higher
DN values associate with more intense lights.
The regional total luminance of night-time light can be
calculated using the follow Eq. (1) (Zhao et al., 2011):
B¼X
63
i¼1
BiNið1Þ
where Bis the regional total luminance of night-time light;
B
i
is the image DN value, ranging from 1 to 63; N
i
is the
number of pixels that have a DN value of B
i
.
Poverty is caused by comprehensive aspects of the socio-
economic situations. Administrative areas can also affect
the poverty evaluation results in different regions. As the
above total luminance of night-time light can only repre-
sent an intuitive impression of socio-economic activities
at night for a region, we used an average light index
(ALI) that can better represent the average level of different
regions in this study, as shown in Eq. (2).
L¼B=Nð2Þ
where Lis the average light index (ALI); Bis the regional
total luminance of night-time light; Nis the sum of the num-
ber of all the pixels with DN value ranging from 1 to 63.
2.4.2. Establishment of integrated poverty index (IPI) using
principal component analysis (PCA)
Principal components analysis (PCA) is a type of factor
analysis method that can be used to reduce large dataset.
Based on statistics, PCA transfers a given number of vari-
ables to a set of uncorrelated variables, called principal
components (PC), each of which contains a linear combi-
nation of all original variables. The first few PCs account
for most of the variance of the variables. Regional poverty
levels are determined by a number of socio-economic vari-
ables. Cavatassi et al. (2004) revealed that the first PC, a
linear combination that captures the greatest variation
amongst the set of socio-economic variables, can be con-
verted into factor scores that can serve as weights for the
creation of the marginality index or poverty index. In this
study we used the following 17 socio-economic variables
to extract an integrated poverty index (IPI) that can be
used as a multidimensional community-level poverty
indicator:
(1) per capita GDP;
Fig. 2. Global DMSP/OLS night-time light image obtained in 2009.
Fig. 3. DMSP/OLS night-time light imageries of China (2007–2009).
1256 W. Wang et al. / Advances in Space Research 49 (2012) 1253–1264
(2) per capita labour compensation;
(3) consumption level of residents;
(4) urbanization rate;
(5) life expectancy;
(6) sex ratio;
(7) dependency ratio;
(8) illiteracy rate;
(9) employment rate;
(10) regional finance income per capita;
(11) the net income per peasant;
(12) per capita consumption expenditure of farmers;
(13) the living space per capita;
(14) electricity consumption per capita;
(15) the original value of productive fixed assets per rural
family;
(16) beds in health care institutions;
(17) tax income per capita.
Both KMO and Bartlett’s tests were applied to the 17
variables using SPSS 18.0. The test results show that the
Pvalue is approximately 0, indicating strong relationships
amongst the variables. The KMO sampling adequacy is
0.753, showing that the PCA method is suitable for this
study.
Table 1 demonstrates the PCA results of the 17 socio-
economic variables, where the first component accounted
for about 54.73%. The first four components together
accounted for about 85.56% of the total variance of the
dataset, therefore are used in this study.
The component score coefficient matrix of the PCA
results are given in Table 2. The eigenvectors of the 17 vari-
ables in the matrix can be used to express each of the com-
ponents. The first PC Z
1
can be expressed by the following
Eq. (3):
Z1¼0:110x1þ0:110x2þ0:105x3þ0:103x4þ0:057x5
0:058x60:093x70:010x80:014x9
0:034x10 þ0:096x11 þ0:093x12 þ0:032x13
þ0:102x14 þ0:044x15 þ0:114x16 þ0:110x17 ð3Þ
where x
i
are variables (see Table 2) used in this study and
the coefficients are the eigenvectors of the PCA result.
Similarly, the second, third and fourth components can
be expressed as follows:
Z2¼0:032x10:058x2þ0:003x3þ0:070x4
þ0:195x5þ0:308x6þ0:040x70:413x8
0:067x9þ0:015x10 0:004x11 0:015x12
þ0:043x13 0:103x14 0:417x15 0:082x16
0:037x17 ð4Þ
Z3¼0:003x1þ0:012x2þ0:082x30:091x4
0:050x5þ0:028x6þ0:260x7þ0:238x8
þ0:498x90:044x10 þ0:090x11 þ0:156x12
þ0:320x13 0:024x14 0:088x15 0:132x16
þ0:072x17 ð5Þ
Z4¼0:012x10:001x20:068x30:025x4þ0:144x5
0:309x60:129x70:087x80:062x9
þ0:612x10 þ0:044x11 0:012x12 þ0:097x13
0:378x14 0:010x15 0:112x16 0:107x17 ð6Þ
Combining the above 4 PCs, an integrated poverty index
(IPI) that can best represent the 17 socio-economic vari-
ables can be created using the following Eq. (7):
IPI ¼k1Z1=ðk1þk2þk3þk4Þþk2Z2=ðk1þk2
þk3þk4Þþk3Z3=ðk1þk2þk3þk4Þþk4
Z4=ðk1þk2þk3þk4Þð7Þ
where k
1
,k
2
,k
3
and k
4
are eigenvectors for the first 4 PCs,
and Z
1
,Z
2
,Z
3
and Z
4
are the values calculated by Eqs. (3)–
(6).
By introducing the k
1
,k
2
,k
3
and k
4
values from Table 1
(k
1
= 9.303, k
2
= 2.139, k
3
= 1.783, k
4
= 1.319), IPI can
then be simplified to the following Eq. (8):
IPI ¼0:6396Z1þ0:1471Z2þ0:1226Z3þ0:0907Z4ð8Þ
3. Results
3.1. IPIs of 31 regions in China
The IPIs of the 31 regions in China are shown in Table 3.
The lower the IPI value is, the poorer the region is. All rich
provinces and municipalities with positive poverty index
values are located in eastern China. The poorest 5 prov-
inces with poverty index less than –0.50, including Qinghai,
Yunnan, Gansu, Guizhou and Xizang, are all located in
Table 1
Total variance of the 17 socio-economic variables explained by the first 4 PCs.
Components Extraction sums of squared loadings Rotation sums of squared loadings
Total % of variance Cumulative% Total % of variance Cumulative%
1 9.303 54.725 54.725 9.099 53.521 53.521
2 2.139 12.581 67.306 2.079 12.229 65.750
3 1.783 10.491 77.796 1.875 11.027 76.776
4 1.319 7.759 85.555 1.492 8.779 85.555
W. Wang et al. / Advances in Space Research 49 (2012) 1253–1264 1257
western China, where regional economy is mainly based on
agriculture with less industry and poor transportation and
other public utilities. A large population in these provinces
is suffering from poverty.
The above IPIs have been grouped into the following
five classes by cluster analysis, as shown in Fig. 4: very high
IPIs (>=1.7), high IPIs (0–1.7), medium IPIs (0.3–0), low
IPIs (0.6-0.3), very low IPIs (<0.6). Provinces with
high IPIs are all located in eastern China. The two munic-
ipalities, Shanghai and Beijing have much higher values
than any other regions. Shanghai is a global city, having
influences over finance, commerce, fashion, technology
and culture in both China and the world. Beijing is the
political, educational, and cultural centre of China. Other
regions with high IPIs are all east-coast developed prov-
inces of China. Most of the provinces with medium IPIs
are in central China, and the ones with lower IPIs are
mainly western regions.
3.2. DMSP/OLS night-time ALIs of 31 regions in China
The DMSP/OLS night-time ALIs of 31 regions in China
were calculated using Eq. (2) as shown in Table 4. As most
of socio-economic activities during night time are centred
in developed areas, associated with bright patterns in the
DMSP/OLS night-time imagery, a higher DMSP/OLS
night-time ALI indicates more prosperous socio-economic
vigour in a region. The 3 municipalities, namely Shanghai,
Beijing and Tianjin, major cities in China since early 20th
century, occupying the top ranks of the table, have the
highest ALI values, on the contrary, Guizhou, a province
in the southwest mountainous area of China is at the bot-
tom of the table.
The ALIs have been separated into 5 classes (Fig. 5.),
they are: very high ALIs (>=20), high ALIs (10–20), med-
ium ALIs (8–10), low ALIs (6.5–8), very low ALIs (<6.5),
indicating different levels of socio-economic vigour. Similar
to the poverty index results, Shanghai, Beijing and Tianjin
are the most developed regions and other east-coast devel-
oped regions including Jiangsu, Guangdong, Zhejiang,
Shandong, and Fujian still share the second class. Prov-
inces with medium ALIs are mainly those in central China,
and the regions with low or very low ALIs are mainly dis-
tributed in Western China.
3.3. Relationship between ALI and IPI at a provincial scale
A regression analysis was carried out in SPSS to explore
the relationship between DMSP/OLS night-time ALI and
the statistical IPI. A positive linear relationship (Fig. 6.)
was found with coefficient of determination R
2
= 0.854.
The linear regression model can be expressed as Eq. (9):
Y¼0:091X0:975 ð9Þ
where Yis the regional IPI; Xis the regional DMSP/OLS
night-time ALI.
The above linear relationship indicates that DMSP/OLS
night-time average light data can provide a good estimate
of regional economic situation and poverty levels with bet-
ter efficiency than the expensive and time-consuming socio-
economic statistic data that traditional methods rely on.
4. Discussion
4.1. Comparison of IPI to GDP at a provincial scale
GDP refers to the market value of all final goods and
services produced in a given period within a country
Table 2
Component score coefficient matrix.
Variables Variable name PCs
1234
X1 Per capita GDP 0.110 0.032 0.003 0.012
X2 Per capita labour
compensation
0.110 0.058 0.012 0.001
X3 Consumption level of
residents
0.105 0.003 0.082 0.068
X4 Urbanization rate 0.103 0.070 0.091 0.025
X5 Life expectancy 0.057 0.195 0.050 0.144
X6 Sex ratio 0.058 0.308 0.028 0.309
X7 Dependency ratio 0.093 0.040 0.260 0.129
X8 Illiteracy rate 0.010 0.413 0.238 0.087
X9 Employment rate 0.014 0.067 0.498 0.062
X10 Regional finance
income per capita
0.034 0.015 0.044 0.612
X11 The net income per
peasant
0.096 0.004 0.090 0.044
X12 Per capita consumption
expenditure of farmers
0.093 0.015 0.156 0.012
X13 The living space per
capita
0.032 0.043 0.320 0.097
X14 Electricity
consumption per capita
0.102 0.103 0.024 0.378
X15 The original value of
productive fixed assets
per rural family
0.044 0.417 0.088 0.010
X16 Beds in health care
institutions
0.114 0.082 0.132 0.112
X17 Tax income per capita 0.110 0.037 0.072 0.107
Table 3
the poverty indexes at provincial scale.
Province Poverty index Rank Province Poverty index Rank
Shanghai 2.13 1 Shanxi 0.16 17
Beijing 1.73 2 Henan 0.22 18
Zhejiang 0.93 3 Sichuan 0.23 19
Jiangsu 0.72 4 Shaanxi 0.29 20
Tianjin 0.69 5 Jiangxi 0.32 21
Guangdong 0.46 6 Hainan 0.33 22
Fujian 0.30 7 Guangxi 0.33 23
Liaoning 0.24 8 Anhui 0.35 24
Shandong 0.19 9 Ningxia 0.37 25
Hebei 0.07 10 Xinjiang 0.46 26
Chongqing 0.07 11 Qinghai 0.55 27
Hunan 0.08 12 Yunnan 0.62 28
Nei Mongol 0.08 13 Gansu 0.69 29
Hubei 0.08 14 Guizhou 0.77 30
Jilin 0.11 15 Xizang 1.02 31
Heilongjiang 0.15 16
1258 W. Wang et al. / Advances in Space Research 49 (2012) 1253–1264
(Goossens et al., 2007). It is a standard indicator used to
measure a country’s economic performance and is often
seen as an indicator of well-being. However, GDP was
never intended to be used for measuring social well-being.
Its key flaw is that it fails to differentiate costs from bene-
fits, identify productive activities from destructive ones,
and distinguish sustainable practices from unsustainable
actions. For example, GDP regards pollution and natural
resource depletion as an economic gain, whilst social activ-
ities such as care for the elderly and children gain just a
zero rating. Natural and “man-made”disasters, crime
and accidents, are seen as positive contributors to GDP
as they generate production, but they do not contribute
to social well-being. GDP does not account for harm
resulting from industrial, household and vehicle emissions,
or water disposal. Instead, it assumes that all monetary
transactions would add points to social well-being. It is
obvious that we cannot assume things are improving just
because more money has been spent. GDP is a total eco-
nomic indicator; it only expresses the economical develop-
ment for a region and is not capable of illustrating
inequalities in well-being.
As described in Section 2.4.2, the IPI is an integrated
poverty index established using a comprehensive evalua-
tion method that embraces many aspects of socio-eco-
nomic situation including per capita GDP as well as
factors that reflect people’s living standards. The following
Table 5 and Fig. 7 show IPIs and GDPs of the 31 regions in
Fig. 4. Poverty classification map for 31 regions in China.
W. Wang et al. / Advances in Space Research 49 (2012) 1253–1264 1259
China and demonstrate the class rank differences between
the two values in each region. As demonstrated in Sec-
tion 3.1, the IPI values were grouped into 5 classes (ranking
from 1 to 5, where lower class rank indicates poorer eco-
nomic state) using cluster analysis. Correspondently the
GDP values of these regions were also grouped into 5 clas-
ses using the same method with higher rank associated with
higher GDP value. Obvious differences were shown
between GDP and IPI class ranks in majority of the regions
due to different evaluation criterions. Only a third of them
have GDP class rank in accordance with IPI class rank.
High GDP does not necessarily produce better social well
being.
From Table 5 and Fig. 7, we can see that the poverty
indexes (IPIs) display different trends from GDP. Large
class rank differences (larger than 1 or smaller than 1)
Table 4
The DMSP/OLS night-time ALIs at a provincial scale.
Province Light index Rank Province Light index Rank
Shanghai 37.52 1 Shaanxi 8.46 17
Beijing 26.75 2 Hubei 8.19 18
Tianjin 20.35 3 Xinjiang 7.92 19
Jiangsu 16.14 4 Sichuan 7.78 20
Guangdong 15.18 5 Qinghai 7.43 21
Zhejiang 15 6 Hainan 7.28 22
Shandong 12.57 7 Gansu 6.95 23
Fujian 10.1 8 Jiangxi 6.93 24
Henan 10.06 9 Heilongjiang 6.43 25
Shanxi 9.79 10 Xizang 6.34 26
Anhui 9.71 11 Yunnan 6.09 27
Hebei 9.6 12 Hunan 6.08 28
Liaoning 9.48 13 Guangxi 5.88 29
Chongqing 9.29 14 Jilin 5.86 30
Ningxia 9.22 15 Guizhou 4.42 31
Nei Mongol 8.72 16
Fig. 5. DMSP/OLS night-time average light index classification map for 31 regions.
1260 W. Wang et al. / Advances in Space Research 49 (2012) 1253–1264
between GDP and IPIs are seen in 3 municipalities, i.e.
Shanghai, Beijing and Tianjin. Although their GDP class
ranks are only medium high or low due to their small
administrative areas, their good socio-economic environ-
ment and high social welfare result in the highest class rank
of IPIs. There are 4 major municipalities that are adminis-
tratively at the same level as provinces in China, including
Shanghai, Beijing, Tianjin and Chongqing. Unlike the
other three municipalities, Chongqing is much larger than
any other cities in China and even larger than some small
provinces. It is divided into 40 county-level subdivisions,
consisting of 19 districts, 17 counties and 4 autonomous
counties, and a large portion of its administrative area
(over 80,000 km
2
), is rural. It is more like a province other
than a municipality economically. Similar to most prov-
inces in western China, the overall economic performance
of Chongqing is lagging behind eastern coastal regions.
For instance, its per capita GDP was 22,909 yuan in
2009, below the national average. The surplus labour force
of Chongqing has to migrate to the east-coast areas to seek
employment opportunities. A third of the regions present a
good agreement between the GDP and IPI class ranks.
They are Zhejiang, Hunan, Hubei, Sichuan, Jiangxi, Guan-
gxi, Anhui, Gansu, Guizhou and Xizang. Amongst these
regions, Zhejiang is the most developed province at the
east-coast of China. Its strong economy provides a solid
basis to improve its socio-economic environment and pro-
duce adequate employment service, therefore, its IPI class
rank is as high as its GDP rank. Sichuan, Guangxi, Gansu,
Guizhou and Xizang are all underdeveloped landlocked
areas in Western China. Their transport facilities are not
facilitative and their secondary and tertiary industries can’t
well support local employment demands. Consequently,
these regions fail to establish a good foundation for solving
various poverty problems. Provinces in central China
including Hunan, Hubei, Jiangxi and Anhui are influenced
by east-coast developed regions. With their socio-economic
environment superior to western regions, they exert great
efforts in developing economy and ensure livelihood issues.
Some previous studies considered GDP as the main fac-
tor to analyse the regional poverty and inequality prob-
lems. For example Jian et al. (1996) have adopted GDP
as the evaluating indicator to analyse the regional inequal-
ity trends in China. In their study, they grouped provinces
into 3 regions: North Coastal, South Coastal and Interior.
The paper deals macroscopically with the overall econ-
omy’s performance at regional level and has not considered
indicators that are closely related to livelihood issues. Some
Fig. 6. Relationship between DMSP/OLS night-time average light index
and poverty index.
Table 5
IPI class rank versus GDP class rank.
Province IPI IPI class IPI
class
rank
GDP
(billion
yuan)
GDP
class
GDP
class
rank
Difference
Shanghai 2.13 Very
high
5 1364.45 Medium 3 2
Beijing 1.73 Very
high
5 1066.48 Medium 3 2
Zhejiang 0.93 High 4 2108.59 High 4 0
Jiangsu 0.72 High 4 3017.04 Very
high
51
Tianjin 0.69 High 4 630.89 Low 2 2
Guangdong 0.46 High 4 3542.11 Very
high
51
Fujian 0.3 High 4 1076.96 Medium 3 1
Liaoning 0.24 High 4 1323.25 Medium 3 1
Shandong 0.19 High 4 3031.15 Very
high
51
Hebei 0.07 Medium 3 1571.12 High 4 1
Chongqing 0.07 Medium 3 524.97 Low 2 1
Hunan 0.08 Medium 3 1113.88 Medium 3 0
Nei Mongol 0.08 Medium 3 786.44 Low 2 1
Hubei 0.08 Medium 3 1117.41 Medium 3 0
Jilin 0.11 Medium 3 632.92 Low 2 1
Heilongjiang 0.15 Medium 3 798.73 Low 2 1
Shanxi 0.16 Medium 3 667.68 Low 2 1
Henan 0.22 Medium 3 1763.36 High 4 1
Sichuan 0.23 Medium 3 1238.76 Medium 3 0
Shaanxi 0.29 Medium 3 682.9 Low 2 1
Jiangxi 0.32 Low 2 654.53 Low 2 0
Hainan 0.33 Low 2 144.56 Very
Low
11
Guangxi 0.33 Low 2 696.21 Low 2 0
Anhui 0.35 Low 2 876.71 Low 2 0
Ningxia 0.37 Low 2 111.37 Very
low
11
Xinjiang 0.46 Low 2 400.12 Very
low
11
Qinghai 0.55 Low 2 94.21 Very
low
11
Yunnan 0.62 Very
low
1 553.71 Low 2 1
Gansu 0.69 Very
low
1 308.87 Very
low
10
Guizhou 0.77 Very
low
1 332.93 Very
low
10
Xizang 1.02 Very
low
1 39.32 Very
low
10
W. Wang et al. / Advances in Space Research 49 (2012) 1253–1264 1261
other studies carried out household surveys whilst consid-
ering GDP. For example Yao et al. (2004) adopted the
urban household survey data obtained in 1998 that con-
tains 17,000 households in 31 provinces and regions.
Although the household survey data are detailed enough
to identify the poverty lines for different regions, it takes
substantial manpower and casts huge economic costs. It
is also weak in terms of timely assessment. Apart from that,
inconsistencies in the sampling structures, the nature and
timing of the surveys, and different definitions of poverty
makes the assembly of a consistent spatially disaggregated
poverty map impossible with the survey data alone
(Elvidge et al., 2009). Our study tries to overcome the
shortcomings of the above previous studies on socio-eco-
nomic situation and poverty indices by using PCA method,
only considering GDP as one important aspect of the inte-
grated poverty index (IPIs) and introducing the DMSP/
OLS night-time light data as a good measure of economic
activities.
4.2. Comparison of IPI to DMSP/OLS night-time ALI at a
provincial scale
Table 6 and Fig. 8 compare the IPI, i.e. the integrated
poverty indexes, and ALI, i.e. regional DMSP/OLS
night-time average light indexes class ranks. The class
ranks of the two indexes show similar trends in Fig. 8. They
are identical in the majority of the regions (20 provinces
and municipalities), in other words, different DMSP/OLS
night-time average light index levels can well reflect differ-
ent poverty levels of these regions. Only 3 provinces display
large differences (larger than 1). They are Hunan, Jilin and
Heilongjiang. Amongst them, 3 provinces, namely Hunan
Jilin and Heilongjiang, have their secondary industries
account for 43.5%, 48.7% and 47.3% of their own regional
GDP in 2009, however, their tertiary industries are not very
active which results in relatively low ALI class ranks in
these regions. Minor class rank differences (equal to 1 or
1) between IPIs and ALIs are seen in 8 regions, including
Tianjin, Liaoning, Henan, Sichuan, Guangxi, Anhui, Ning-
xia and Gansu. Amongst these regions, Tianjin has a very
Fig. 7. Comparison of the class ranks between regional poverty indexes (IPI) and GDP in China.
Table 6
IPI class rank versus ALI class rank.
Province IPI IPI class IPI
class
rank
ALI ALI
class
ALI
class
rank
Difference
Shanghai 2.13 Very
high
5 37.52 Very
high
50
Beijing 1.73 Very
high
5 26.75 Very
high
50
Zhejiang 0.93 High 4 15.00 High 4 0
Jiangsu 0.72 High 4 16.14 High 4 0
Tianjin 0.69 High 4 20.35 Very
high
51
Guangdong 0.46 High 4 15.18 High 4 0
Fujian 0.3 High 4 10.10 High 4 0
Liaoning 0.24 High 4 9.48 Medium 3 1
Shandong 0.19 High 4 12.57 High 4 0
Hebei 0.07 Medium 3 9.60 Medium 3 0
Chongqing 0.07 Medium 3 9.29 Medium 3 0
Hunan 0.08 Medium 3 6.08 Very
low
12
Nei Mongol 0.08 Medium 3 8.72 Medium 3 0
Hubei 0.08 Medium 3 8.19 Medium 3 0
Jilin 0.11 Medium 3 5.86 Very
low
12
Heilongjiang 0.15 Medium 3 6.43 Very
low
12
Shanxi 0.16 Medium 3 9.79 Medium 3 0
Henan 0.22 Medium 3 10.06 High 4 1
Sichuan 0.23 Medium 3 7.78 Low 2 1
Shaanxi 0.29 Medium 3 8.46 Medium 3 0
Jiangxi 0.32 Low 2 6.93 Low 2 0
Hainan 0.33 Low 2 7.28 Low 2 0
Guangxi 0.33 Low 2 5.88 Very
low
11
Anhui 0.35 Low 2 9.71 Medium 3 1
Ningxia 0.37 Low 2 9.22 Medium 3 1
Xinjiang 0.46 Low 2 7.92 Low 2 0
Qinghai 0.55 Low 2 7.43 Low 2 0
Yunnan 0.62 Very
low
1 6.09 Very
low
10
Gansu 0.69 Very
low
1 6.95 Low 2 1
Guizhou 0.77 Very
low
1 4.42 Very
low
10
Xizang 1.02 Very
low
1 6.34 Very
low
10
1262 W. Wang et al. / Advances in Space Research 49 (2012) 1253–1264
high ALI class rank. Located at the west coast of Bohai
Gulf in north China, Tianjin is a dual-core modem interna-
tional metropolitan, composed of the old city and Binhai
New Area. The Binhai New Area is a new growth pole in
China, which maintains an annual growth rate of nearly
30% of its GDP. It has been seen as a base of China’s
advanced industry, a base for financial reform, and a base
of innovation in China. By the end of 2010, 285 Fortune
Global 500 companies have established their branch offices
in this area. On the contrary, Sichuan, Guangxi, Ningxia
and Gansu are all underdeveloped provinces in western
China. Due to the limit of natural conditions and transport
facilities, their economic performances are much less active
than the eastern coastal regions. Their lower ALIs can well
demonstrate the regional economic disparity. For example,
Henan has its value-added tertiary industry accounts for
570 billion yuan of its GDP in 2009 that achieved it the
ninth position among the 31 provinces and municipalities.
Its growing economic activities in the tertiary industries
have also contributed to its high ALI class rank. Mean-
while, Henan is a very populous province with a popula-
tion of 94.87 million, ranking the third place in China. A
lot of surplus labour force has to go to the east-coast areas
to seek employment opportunities. As a result, its ALI
class rank is not corresponding to its IPI class rank. Being
one of the central regions, Anhui has a lot in common with
Henan. It is also a populous province with a population of
61.31 million, of which the rural population accounts for
57.9%. The main social and economic development target
for Anhui is to provide adequate employment service for
its population. Being influenced by east-coast developed
regions, Anhui makes full use of its advantageous geo-
graphical location to develop its commercial economy. Its
economic performance is superior to western regions, and
its ALI class rank is medium. Liaoning is an industrial
province, with only 39.65% of its population living in the
rural area. It has the largest economy of Northeast China.
Its secondary industry accounted for 52.0% of its GDP in
2009 and its nominal GDP for 2010 was 1.83 trillion yuan,
making it the 7th largest economy in China. Its good
economic performance results in a high IPI class rank,
however, the inadequate development of its tertiary indus-
try only achieved it a rather lower ALI class rank.
The DMSP/OLS night-time light data have been used in
some previous studies. Elvidge et al. (2009) produced a glo-
bal poverty map using a poverty index calculated by divid-
ing population count (LandScan 2004) by the brightness of
satellite observed lighting (DMSP/OLS night-time lights).
The study is based on global scale, not capable of reflecting
the regional poverty details. Moreover, as it only considers
the population factor other than considering a comprehen-
sive mix of a few socio-economic factors that reflecting bet-
ter social wellbeing, the accuracy of its evaluations towards
poverty is reduced. In our study, 17 main socio-economic
indicators have been adopted to establish an integrated
poverty index for every region and a better comprehensive
evaluation of poverty situation for each region is produced.
The good correlation between the ALI and IPI values
revealed in Section 3.3 has also proven that remote sensing
technique can advance poverty evaluation at a regional
scale more efficiently and accurately.
5. Conclusion
It is an important goal for governments and local policy
makers to eradicate poverty in China and other countries.
In order to tackle the excessively wide gap of socio-eco-
nomic development levels in different regions, the measure-
ment of the overall poverty situation at a regional scale is
the primary task. To estimate poverty levels of different
regions and analyse their spatial and temporal characteris-
tics is the first step to research the regional disparity of
social wealth.
GDP as an indicator on its own is not capable of reflect-
ing regional poverty level. Household surveys contain
detailed information for poverty evaluation, however, it
takes substantial manpower and huge economic costs and
it is time consuming. Based on 17 socio-economic indica-
tors (including GDP) the IPI in this study is capable of
demonstrating the socio-economic situations of the 31
Fig. 8. Comparison of the class ranks between regional poverty indexes (IPI) and ALI in China.
W. Wang et al. / Advances in Space Research 49 (2012) 1253–1264 1263
study regions in China. Satellite remote sensing has the
advantage to provide efficient and accuracy spatial data
for various physical and social science research purposes
due to its high temporal resolution and extensive spatial
coverage. DMSP/OLS night-time light satellite imagery
as a new data source is proven in this study that it can pro-
vide a practical, efficient and reliable approach to explore
poverty issues at a regional scale. A good correlation
between IPI and ALI is revealed in this study with a coef-
ficient of determination (R
2
) of 0.854. Therefore, we sug-
gest that government administrators and policymakers
may refer DMSP/OLS night-time light data to a valid data
source for estimating regional poverty issues.
This study is currently preliminary. Some scholars have
tested the combination of the LandScan population data
and the global DMSP/OLS night-time light data. For
example Elvidge et al. (2009) produced a global poverty
map using a poverty index calculated by dividing popula-
tion count (LandScan 2004) by the brightness of satellite
observed lighting (DMSP/OLS night-time lights). The
LandScan population data produced by the US Depart-
ment of Energy, Oak Ridge National Laboratory is used
as it can help disaggregate estimated data from regional
scale to pixel scale. However, the demographic data change
over time and the LandScan population data from different
years are not compatible, thus the use of the older versions
is not recommended (Oak ridge National Laboratory,
2010). Due to this fact, we believe that the LandScan pop-
ulation data has limitations for studying regional poverty
problems in China. The Chinese government is currently
developing a native population grid data that is based on
the 2010 nationwide population census. This data will have
much higher accuracy and better reliability in comparison
with old data collected by traditional methods as GIS tech-
nique has been introduced into census. We plan to bring
the native population grid data into our future studies on
Chinese poverty issues once it is publicly published.
Acknowledgements
This study was supported by the Fundamental Research
Funds for the Central Universities, and the Research
Funds of Renmin University of China (10XNI008).
References
Cavatassi, R., Davis, B., Lipper, L. Estimating poverty over time and
space: construction of a time-variant poverty index for Costa Rica.
Working Papers, 2004.
Croft, T.A. Night-time images of the Earth from space. Scientific
American 239, pp. 68–79, 1978.
Doll, C. Estimating non-population activities from night-time satellite
imagery. Remotely Sensed Cities. London: Taylor and Francis, pp.
335–353, 2003.
Doll, C., Muller, J.P., Elvidge, C.D. Night-time imagery as a tool for
global mapping of socioeconomic parameters and greenhouse gas
emissions. Ambio 29 (3), 157–162, 2000.
Elvidge, C.D., Baugh, K.E., Kihn, E.A., et al. Mapping city lights with
nighttime data from the DMSP Operational Linescan System. Pho-
togram. Eng. Remote Sensing 63 (6), 727–734, 1997a.
Elvidge, C.D., Baugh, K.E., Kihn, E.A., et al. Relation between satellite
observed visible-near infrared emissions, population, economic activity
and electric power consumption. Int. J. Remote Sensing 18 (6), 1373–
1379, 1997b.
Elvidge, C.D., Baugh, K.E., Dietz, J.B., et al. Radiance calibration of
DMSP-OLS low-light imaging data of human settlements. Remote
Sens. Environ. 68, 77–88, 1999.
Elvidge, C.D., Imhoff, M.L., Baugh, K.E., et al. Night-time lights of the
world: 1994–1995. ISPRS Journal of Photogrammetry and Remote
Sensing 56 (2), 81–99, 2001.
Elvidge, C.D., Sutton, P.C., Ghosh, T., et al. A global poverty map
derived from satellite data. Computers & Geosciences 35 (8), 1652–
1660, 2009.
Foster, J.L. Observations of the Earth using night-time visible imagery.
International Journal of Remote Sensing 4, 785–791, 1983.
Goossens, Yanne et al. Alternative progress indicators to Gross Domestic
Product (GDP) as a means towards sustainable development. IP/A/
ENVI/ST/2007-10. Study provided for the European Parliament’s
Committee on the Environment, Public Health and Food Safety, 2007.
Jian, T., Sachs, J.D., Warner, A.M. Trends in regional inequality in
China. China Economic Review 7 (1), 1–21, 1996.
Jin, Y. Study on Urban Economic Vitality Index in China. SCIENTIA
GEOGRAPHICA SINICA 27 (1), 2007.
Li, G., Ji, S., Yang, X. Evaluation of Economic Development for Thirty-
one Provinces (or Municipalities) in China. Journal of Shijiazhuang
University of Economics 27 (002), 187–189, 2004.
Li, Z., Yang, Y., Liu, Y. Difference among the Growth of GDP and
Urbanization of the Provinces and the Cities in West China since the
Reform and Opening-up. China Population, Resources and Environ-
ment 18 (5), 19–26, 2008.
Nakayama, Y., Tanaka, S. World economic tendency watched by satellite
nighttime light distribution. Geography 28 (8), 94–103, 1983a.
Oak ridge National Laboratory. LandScan frequently asked questions,
why can’t I get the previous versions of the LandScan global
population datasets? <http://www.ornl.gov/sci/landscan/land-
scan_faq.shtml#07>. (Last accessed October 30, 2011), 2010.
Sutton, P.C., Costanza, R. Global estimates of market and non-market
values derived from nighttime satellite imagery, land cover, and
ecosystem service valuation. Ecological Economics 41 (3), 509–527,
2002.
Sutton, P.C., Elvidge, C.D., Ghosh, T. Estimation of gross domestic
product at sub-national scales using nighttime satellite imagery.
International Journal of Ecological Economics & Statistics 8 (S07),
5–21, 2007.
Takahashi, K.I., Terakado, R., Nakamura, J., et al. In-use stock analysis
using satellite nighttime light observation data. Resources, Conserva-
tion and Recycling 55 (2), 196–200, 2010.
Tobler, W.R. Satellite confirmation of settlement size coefficients. Area 1,
31–34, 1969.
Welch, R. Monitoring urban population and energy utilization
patterns from satellite data. Remote Sensing of Environment 9,
1–9, 1980.
Yao, S., Zhang, Z., Hanmer, L. Growing inequality and poverty in China.
China Economic Review 15 (2), 145–163, 2004.
Zhao, N., Currit, N., Samson, E. Net primary production and gross
domestic product in China derived from satellite imagery. Ecological
Economics 70 (5), 921–928, 2011.
1264 W. Wang et al. / Advances in Space Research 49 (2012) 1253–1264