ESTIMATING IMPERVIOUS SURFACE FOR THE URBAN AREA EXPANSION:
EXAMPLES FROM CHANGCHUN, NORTHEAST CHINA
Dongmei Lu1, 2
, Kaishan Song3, 2*
, Lihong Zeng3
, Dianwei Liu3
, Shahbaz Khan2
, Bai Zhang3
, Zongming Wang3
, Cui Jin3
1
Computer Science and Engineering College, Jilin Architectural and Civil Engineering Institute, Hongqi Street 1129,
Changchun 130021, Jilin Province, PR China.
2
Land and Water, CSIRO, Charles Sturt University, Building 24, Locked bag 558, Wagga Wagga, NSW Australia
3
Northeast Institute of Geography and Agricultural Ecology, Chinese Academy of Sciences, No.3195, Weishan Road,
Changchun 130012, Jilin Province, PR China.
Corresponding author: songks@neigae.ac.cn; Tel: +86-431-85542364
KEY WORDS: Changchun, remote sensing, estimation, impervious, land-use, classification.
ABSTRACT:
As a developing country, China is now undergoing a quick process of urbanization. Therefore, understanding and managing the
urban environment is a prerequisite for addressing sustainability, which is an increasingly important issue need a range of discipline
to cope with. This paper explored extraction of impervious surface information from Landsat ETM+ data with the integration of
fraction images from linear spectral mixture analysis based upon Ridd’s vegetation-impervious surface-soil (V-I-S) model. A new
approach for urban land-use classification, based on the combined use of impervious surface and spectral mixture analysis (SMA)
were applied in this paper. The minimum noise fraction transform (MNF) procedure was applied to transform the six reflective
bands into a new coordinate set to select the four endmebers, e.g. high-albedo surface, low-albedo surface, soil and vegetation..
Results showed that the integration of faction images improved urban impervious surface estimation. The impervious surface in the
urban area were derived from high-albedo surface and low-albedo surface. Accuracy assessment indicated that the root-mean-square
error is less than 10.2% for the impervious surface image. The main factors that affect the accuracy are the reflectance variation
caused by atmospheric factors, sun-sensor-target geometry. How to deal with these factors to minimize reflectance variation will be
the future study topics. Also water body and shade were not addressed in this paper, which also need to be considered in the future
study.
1. INTRODUCTION
Urbanization is perhaps one the most important human
activities, creating enormous impacts on the environment at the
local, regional and global scales (Turner et al., 1990). Although
urbanization, the form of land cover (built-up or impervious
surface) occupies less than 2% of the global land surface, many
evidence show that human disturbance due to urbanization has
significantly altered the natural landscape (Grubler, 1994).
Therefore, understanding and managing the urban environment
is a prerequisite for addressing sustainability, an increasingly
important issue across a range of disciplines (Newman &
Kenworthy, 1999). Improving urban land-use/cover
classification accuracy has been an important issue in remote
sensing literature (Liu and Wen, 2004). Different approaches
have been applied, which include incorporation of geographic
data, census data, texture feature and structure or contextual
information into remote sensing spectral data. Furthermore,
expert systems, fuzzy classification, and merged multi-sensor
data for improving spatial resolution have been applied.
However, urban land-use/cover classification is still a challenge
with medium or coarse spatial resolution remotely sensed data
due to the large number of mixed pixels and the spectral
confusions among different land-use/cover types.
Recent institutional changes, marketization and globalization
combined together have bought about new processes of ruralurban
interaction, giving rise to new forms of human
settlements in china, especially after Chinese reform and open
policies were carried out since 1978. The latest national
population census conducted in 2000, revealed an accelerated
rate of urbanization, characterized both by its scale and change
speed that have no parallel elsewhere in the world (Tian et al.,
2005). China owns the largest population (1.265 billion) in the
world, of which 456 million (36%) live in cities and towns
(State Council of China Office of Population Census 2001).
Urbanization in China has speeded up, urban population only
comprise 20% of the whole country population in 1982, while it
has reached 36% in year 2000. A net gain of 16% took place in
less than 20 years, which cause a sharp contrast o the prereform
era, when the urban population only increased 7.1%
over 30 year period (Zhang and Zhao, 1998; Tian et al., 2005).
Rapid urban growth was mainly attributed to migration form the
countryside to cities and town, the conversion from rural
administrative units to city units, and the natural urban
population increase, and all the driving force can be traced from
economic development.
The focus of this study is to examine urban LULC patterns
by SMA method based upon V-I-S model. Changchun (Jilin
provincial Capital city), the third largest city in northeast China,
has been chosen as the area of study. With over 3.58 million
populations live in the urban area, the city is the nation’s
automobile capital of the nation. 3 scenes Landsat TM and
ETM+ image since 1993 that covers the City were used in
conjunction with other types of spatial data for the analysis city
LULC for city expansion study. Specific objectives of this
research are: (1) employing V-I-S model to derive landscape
fractional components, and to apply them to characterize the
urban expansion; (2) exploring the urban area expansion
process from 1993 to 2004 based upon V-I-S model and
analyzed the possible driving forces.
385
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B8. Beijing 2008
2. STUDY AREA
Changchun is the capital and largest city of Jilin province,
located in the northeast of China. It is the largest centre for
China's automotive industry. As of 2007, Changchun has a
population of 7.45 million, including counties and county-level
cities. The urban districts have a total population of 3.58
million. Changchun has incredibly cold, long winters with the
temperature dropping as low as -30°C; and the temperature
sometimes goes up 35 °C in late July in summer. Changchun in
its present form is a new city with only about 200 years of
history. Figure.1 showed the urban expansion from 1905 to
1993. It expanded rapidly as the junction between the Japaneseowned
South Manchurian Railway and the Russian-owned
Chinese Eastern Railway from 1905-1935. In 1932, Changchun
became the capital of Manchukuo, which existed from 1931 to
1945. The city underwent rapid expansion in both its economy
and infrastructure. Changchun became the capital of Jilin
Province, P.R China in 1954. As Changchun's main industry,
the manufacturing of transportation facilities and machinery
such as those of automobile, passenger train, and tractor has
developed very well. Industries such as machinery, electronics,
optics, chemistry, medicine, textile, metallurgy, building
materials and foodstuffs all assume their own features and
advantages.
Figure.1 The expansion of Changchun urban area from 1905 to
1993
3. MATERIAL AND METHODOLOGIES
3.1 Landsat image processing
3 sub-scenes of Landsat 5 TM and Landsat 7 ETM+ image
(path 118/row 28, 29) of Changchun, China, which acquired on
the 23 August 1993, the 18 September 2000 and 23 August
2003, were used in this research. Images were processed by
Remote Sensing Ground Station, Chinese Academy of Sciences
(RSGS). Possible geo-position errors due to terrain effects were
corrected using the ortho-correction method based on DEM,
which derived from digitized topographic map with 1:5,000
scales. All images were rectified to a common Universal
Transverse Mercator coordinate system. Bands 1 through 5 and
band 7 were utilized at a spatial resolution of 30 m.
3.2 Determination of reflectance from TM and ETM+
The digital numbers (DNs) of the TM image were converted to
normalized exo-atmospheric reflectance measures following the
methods proposed by Markham and Barker (1987). The
calibration parameters for Landsat-7 ETM+ were obtained from
the ETM+ data header (Irish, 1998). We assumed homogeneous
atmospheric conditions within the image, so no atmospheric
corrections were performed.
3.3 Spectral mixture analysis
Spectral mixture analysis (SMA) was utilized for calculating
land cover fractions within a pixel and involves modeling a
mixed spectrum as a combination of spectra for pure land cover
types, called endmembers (Roberts et al., 1998). The linear
spectral mixture model describes the surface composition in
each pixel of an image using two to six endmembers (for an
ETM+ image). Each endmember represents a pure land cover
type. The linear mixture model is:
j
N
1j
ijij eRfR += ∑=
(1)
Where
∑=
=
N
1i
i 1f and 0fi ≥ (2)
where Rj is the reflectance for each band j in the ETM+ image,
N is the number of endmembers, fi is the fraction of endmember
i, Rij is the reflectance of endmember i in band j, and ej is the
unmodeled residual. Model fitness is normally assessed by the
residual term ej or the RMS over all image bands (M):
21
N
1j
j MeRMS /
)/( ∑=
= (3)
The fraction of each endmember can be obtained by applying a
least squares technique in order to minimize the unmodeled
residual error ej, given the constraints on fi. As stated by Small
(2001), the linear mixture model may not be appropriate for
applications in which only subtle spectral differences exist in all
sampled bands. In many applications, about three to four
endmembers are chosen for simple linear mixture models
(Roberts et al., 1993; Small, 2001).
3.4 Endmember selection methods
One approach for choosing image endmembers is selecting
representative homogeneous pixels from satellite images
through visualizing spectral scatter plots of image band
combinations (Rashed et al., 2001). In this study, the maximum
noise fraction (MNF) transformation was applied to trace the
endmember (Wu & Murray, 2003). The first three components
were shown in Fig.2. Previous research has shown that use of
MNF transform can improve the quality of fraction images (Wu
& Murray, 2003). The first four components were retained for
use in the LSMA models, while the last two components
discarded due to the high proportion of noise content. Four
endmembers were selected in this study; they are bright albedo
surface, dark albedo surface, soil and vegetation (Fig.3). These
endmembers were initially identified from the images based on
ground truth and high-resolution aerial photographs, and then
the reflectances of these initial endmembers were compared
with the endmembers selected from the scatterplot of
combination MNF1-3. The endmembers with similar MNF
386
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B8. Beijing 2008
spectra at the extreme vertices of the scatterplots were selected
(see figure.3a-c).
Figure.2a MNF component1
Figure.2b MNF component2
Figure.2c MNF component3
Figure.3 Feature space represents the first three MNF
components. The extreme pixel clusters which bound almost all
other pixels in these three feature spaces were delineated as
endmembers. High albedo (e.g. concrete, clouds, and sand), low
albedo (e.g. water, shade and asphalt), vegetation (e.g. crop,
grass and trees), and soil (e.g. bare soil and quarry), were
identified according to these feature spaces and their
corresponding features obtained from original reflectance data.
Vegetation
Bright-imp
Dark-imp
Soil
Dark-imp
Bright-imp
Bright-imp
Dark-imp
Vegetation
387
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B8. Beijing 2008
4. RESULTS AND DISCUSSION
4.1 Urban fraction mapping
The endmember fractions were calculated by solving a fully
constrained four-endmember linear mixing model using the
Landsat TM or ETM+ reflectance data, and the final result were
shown in Fig.4~6 for 1993, 2000 and 2003, respectively. The
vegetation fraction image correlates with known vegetated areas
within the original TM or ETM+ image. From Fig.4a it can be
seen that bright albedo surface mainly distributed in urban
region, especially in the CBD region, and for there was some
haze over the urban region in 1993, which also reflected in the
un-mixing result of impervious surface. The low albedo surface
also mainly distributed in urban area (Fig. 4b), it included water
bodies in this study. But it is more reasonable to mask out water
bodies in the future study. The vegetation fraction is near zero
in the urban region, while increasing to 10–20% in high-density
residential areas, and 20–30% in low-density residential areas,
and it is about 70-90% in vegetated areas, nearly 100% only in
a small part of crop area (Fig. 4c). Moreover, the soil fraction
image is also consistent with the soil distribution in the study
area because the soil fraction in the CBD and residential areas
is 10–20% but higher than 60% in some parts of the urban
fringe (Fig. 4d), where new constructions sites located. Fraction
images of dark albedo and high albedo cannot be directly
interpreted from the image.
Figure.4a Fraction image of high albedo in 1993
Figure.4b Fraction image of low albedo in 1993
Figure.4c Fraction image of vegetation in 1993
Figure.4d Fraction image of soil in 1993
However, their relationship with impervious surfaces will be
built using a two-endmember linear mixture model in this
study. Fig. 4 also shows that this model performance is not as
good for modeling some high albedo materials, such as high
reflectance roofs, clouds, and sand
From Fig.5, it can be seen that the four endmember
fractions show the similar trend as that from 1993. It can be
easily noticed that high albedo fraction is more evenly
distributed in the urban area as no haze disturbance in the
original image. While the low albedo surface fraction
distributed in most parts of the city, especially in CBD and
residential area, also included some water bodies. Most part of
urban area is higher than 60%, some area even higher than 80%.
This indicates that the building roof or road materials are dark
color in Changchun. From Fig.5c, it can be seen that only some
parks and area along Yitong River have high values, but in most
of the urban area, it is only about 10%, but in the suburban area,
especially in the cropping area and forest area, the vegetation
fraction is about 70%, only a small vegetable field has high
value. This is because the image acquired in late September, so
most of the crop is nearly ready for harvest, only irrigated
vegetable at their highest cover with sufficient water supply and
no sign of chlorophyll lack. Soil fraction has high values in
some region, including construction sites, quarries and bare soil
regions. But in most of the region, it has a value about 30%, one
reason is that Changchun locates in black soil region, the soil
spectral reflectance is low compared with most of the other
soils, so its reflectance signature is quite close to some of the
low albedo surface, also most of the crop in their senescence
388
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B8. Beijing 2008
stage, also show dark color, that may explain why soil fraction
have no obvious distinctive pattern.
Figure.5a Fraction image of high albedo in 2000
Figure.5b Fraction image of low albedo in 2000
Figure.5c Fraction image of vegetation in 2000
Figure.5d Fraction image of soil in 2000
From Fig.6a-b, it can be seen that high albedo and low
albedo fraction have the similar distribution pattern in most
parts of the city, especially in CBD and residential area as that
from previous years. From Fig.5c, it can be seen that vegetation
fraction value is about 80% in most of the cropping area and
forest area because this image acquired in late August 2003
when vegetation is just begin to senesce. Soil fraction has also
have similar trend, but with low value in the croppin area and
forest.
Figure.6a Fraction image of high albedo in 2003
Figure.6b Fraction image of low albedo in 2003
389
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B8. Beijing 2008
Figure.6c Fraction image of vegetation in 2003
Figure.6d Fraction image of soil in 2003
4.2 Assessment if unmixing model based up V-I-S
The RMS for every image pixel was calculated in order
to assess the performance of LSMA model (Table.1 and
Figure.7). The mean RMS over the image is 0.0042,
0.0028 and 0.0034, respectively, which suggests a
generally good fit (less than 0.015). The RMS images
show that this model represents residential, vegetation,
soil, and water cover types very well. However,
performance is not as good for modeling some high
albedo materials, such as high reflectance roofs, clouds,
and sand. Also some dark pine forest can not modeled
well for their spectral variation from that of deciduous
tree (not demonstrated in this paper). Figure.7 only the
distribution characteristics of RMS in 2003, it can be
seen that RMS value is generally less than 0.015.
Table.1 Unmixing modled RMSE values
year Min Max Mean Stdev
1993 0.0000 0.1100 0.0042 0.0022
2000 0.0000 0.0943 0.0028 0.0018
2003 0.0000 0.1238 0.0034 0.0025
0
1
2
3
4
5
6
7
8
9
10
0.01 0.02 0.03 0.04 0.06 0.07 0.08 0.09
Root Means Square Error of Spectral Unmixing
Frequency(%)
Figure.7 Distribution of Unmixing RMSE
4.3 Impervious surface mapping
According to Wu and Murray’s study (2003), impervious
surface is likely on or near the line connecting the low albedo
and high albedo endmembers in the feature spaces. In other
words, most impervious surfaces might be represented by low
and high albedo endmembers as follows.
bhighhighlowlowimp eRfRfR +×+×= (4)
Rimp,b is the reflectance spectra of impervious surfaces for band
b, and flow and fhigh are the fractions of low albedo and high
albedo, respectively. Rlow,b and Rhigh,b are the reflectance spectra
of low albedo and high albedo for band b, and eb is the
unmodeled residual. Associated with determining Rimp,b is the
requirement flow + fhigh = 1 and flow , fhigh great or equal to 0.
In this study, we also applied this methodology proposed by
Wu and Murray (2003) to map the impervious surface. We
applied this model to build a relationship between
impervious surfaces and high and low albedo materials
with formula (4) to derive the impervious surface in the
urban area o Changchun. Also water bodies were masked
out in the final result. The final impervious surface in
2000 and 2003 were shown in figure.8. It can be seen
that in the urban area, the impervious surface is quite
consistent, the fraction value is about 0.7 in most of the
urban area, while in the CBD area it can reach 0.9. This
indicates that V-I-S model with LSMA is suitable for
urban impervious surface study. As there was some haze
in the original image in 1993 that affected the modeling
result, impervious surface area was not shown in the final
result.
Figure.8a Impervious surface in 2000
390
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B8. Beijing 2008
Figure.8b Impervious surface in 2003
5. CONCLUSION AND FUTURE RESEARCH
In this paper, a LSMA method was applied to quantify urban
composition under the framework of V–I–S model. MNF
method was applied for the acquired image to select the
endmembers, and a linear spectral mixture analysis method was
developed to derive the fractions of green vegetation, high
albedo and low albedo, and soil. The impervious surface was
derived from high albedo and low albedo surface. It can be
concluded that the LSMA model is reasonable to retrieve urban
impervious surface in our study case; V-I-S model can be used
to monitoring urban expansion, especially for city impervious
area estimation. One future research direction could be the
accuracy assessment of the vegetation and soil fraction.
Contemporaneous high-resolution images, such as Quickbird or
IKONOS imagery, should be helpful to obtain ground truthing
information of vegetation and soil. Also, how to deal with
spectral variation caused by sun-object-sensor geometry is not
delivered in this paper, which needs to be coped with in future
study, especially with consideration of urban dynamics study.
Acknowledgements
We would like to thank Dr. Yuanzhi Zhang from the Chinese
Unversity of Hongkong and Dr. Mohsin Hafeez from Charles
Sturt University, Australia for their help and valuable comments
on this work. Financial Support from NEIGAE frontier research
project (k08y44) is also acknowledged.
References
Grubler, A., 1994. Technology. In changes in land use and land
cover: A global perspective. pp. 287-328. Cambridge University
Press, Cambridge.
Guangjin, Tian., Jiyuan, Liu., Yichun, Xie., Zhifeng, Yang.,
Dafang, Zhuang., Zheng, Niu., 2005. Analysis of spatiotemporal
dynamic pattern and driving factors of urban land in
China in 1990s using TM images and GIS. Cities, 22(6), pp.
400-410.
Irish, R., 1998. Chapter 11: Data product. Landsat 7 science
data users handbook.
Lu, D., & Weng, Q., 2004. Spectral mixture analysis of the
urban landscape in Indianapolis city with Landsat ETM+
imagery. Photogrammetric Engineering & Remote Sensing,
70(9), pp. 1053– 1062.
Markham, B. L., Barker, J. L., 1987. Thematic Mapper
bandpass solar exoatmospheric irradiances. International
Journal of Remote Sensing, 8, pp. 517– 523.
Newman, P., Kenworthy, J., 1999. Sustainability and cities:
overcoming automobile dependence. Washington, DC: Island
Press.
Rashed, T., Weeks, J. R., & Gadalla, M. S., 2001. Revealing the
anatomy of cities through spectral mixture analysis of
multispectral satellite imagery: a case study of the greater Cairo
region, Egypt. Geocarto International, 16(4), pp. 5 – 15.
Ridd, M. K., 1995. Exploring a V–I–S (vegetation–impervious
surface–soil) model for urban ecosystem analysis through
remote sensing: comparative anatomy for cities. International
Journal of Remote Sensing, 16, pp. 2165– 2185.
Roberts, D. A., Gardner, M., Church, R., Ustin, S., Scheer, G.,
Green, R. O., 1998. Mapping chaparral in the Santa Monica
Mountains using multiple endmember spectral mixture models.
Remote Sensing of Environment, 65, pp. 267–279.
Small, C., 2002. Multitemporal analysis of urban reflectance.
Remote Sensing of Environment, 81, pp. 427–442.
State Council of China Office of Population Census., 2001.
China Population Census. Beijing: Statistical Press of China.
Stefanov, W. L., Ramsey, M. S., Christensen, P. R., 2001.
Monitoring urban land cover change: an expert system
approach to land cover classification of semiarid to arid urban
centers. Remote Sensing of Environment, 77, pp. 173– 185.
Stuckens, J., Coppin, P. R., Bauer, M. E., 2000. Integrating
contextual information with per-pixel classification for
improved land cover classification. Remote Sensing of
Environment, 71, pp. 282– 296.
Thomas, N., Hendrix, C., Congalton, R. G., 2003. A
comparison of urban mapping methods using high-resolution
digital imagery. Photogrammetric Engineering and Remote
Sensing, 69, pp. 963–972.
Turner, B L II, Clark, W C and Kates, R W, et al., 1990. The
earth as transformed by human action: global and regional
change in the biosphere over the past 300 years. Cambridge
Unversity Press, Cambridge.
Wu, C., & Murray, A. T., 2003. Estimating impervious surface
distribution by spectral mixture analysis. Remote Sensing of
Environment, 84, pp. 493– 505.
Zhang, L and Zhao, S., 1998. Re-examining China’s ‘Urban’
concept and the level of urbanization. The China Quarterly,
154, pp. 330-381.
391
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B8. Beijing 2008
392