Pedosphere 19(1): 96–103, 2009

ISSN 1002-0160/CN 32-1315/P

c 2009 Soil Science Society of China
Published by Elsevier Limited and Science Press

Urban Land Use Change Detection Using Multisensor

Satellite Images∗1

DENG Jin-Song1 , WANG Ke1,∗1 , LI Jun2 and DENG Yan-Hua3

1 Institute of Agricultural Remote Sensing and Information Technology, Zhejiang University, Hangzhou 310029 (China).

E-mail: jsong [email protected]

2 Zhejiang Environmental Scientific Institute, Hangzhou 310028 (China)
3 Changsha Environmental Protection College, Changsha 410004 (China)

(Received May 8, 2008; revised November 26, 2008)

ABSTRACT
Due to inappropriate planning and management, accelerated urban growth and tremendous loss in land, especially
cropland, have become a great challenge for sustainable urban development in China, especially in developed urban area
in the coastal regions; therefore, there is an urgent need to effectively detect and monitor the land use changes and provide
accurate and timely information for planning and management. In this study a method combining principal component
analysis (PCA) of multisensor satellite images from SPOT (systeme pour l’observation de la terre or earth observation
satellite)-5 multispectral (XS) and Landsat-7 enhanced thematic mapper (ETM) panchromatic (PAN) data, and supervised
classification was used to detect and analyze the dynamics of land use changes in the city proper of Hangzhou. The overall
accuracy of the land use change detection was 90.67% and Kappa index was 0.89. The results indicated that there was
a considerable land use change (10.03% of the total area) in the study area from 2001 to 2003, with three major types
of land use conversions: from cropland into built-up land, construction site, and water area (fish pond). Changes from
orchard land into built-up land were also detected. The method described in this study is feasible and useful for detecting
rapid land use change in the urban area.
Key Words: change detection, land use, multisensor satellite image, principal component analysis (PCA), urban area

Citation: Deng, J. S., Wang, K., Li, J. and Deng, Y. H. 2009. Urban land use change detection using multisensor satellite
images. Pedosphere. 19(1): 96–103.

INTRODUCTION

Concomitant with rapid economic development and urbanization, China has witnessed tremendous
land use changes, especially in developed urban area in the coastal regions. Unfortunately, because of
inappropriate planning and ineffective management, accelerated urban growth with tremendous loss in
cropland has become a great challenge for sustainable urban development. There is an urgent need to
effectively detect and monitor the land use changes and provide accurate and timely information for land
planning and management. Remote sensing has proved to be a powerful tool for monitoring rapid land
use changes. In the last three decades, the technologies and methods of remote sensing have dramatically
progressed to include operating a suite of sensors at a wide range of platforms with potential interests
and impacts on land planning and land management compared with the traditional manner (Rogan and
Chen, 2004).
In most urban land cover/land use change studies, Landsat data were utilized because the uniqueness
of the dataset as the only long-term digital archive with a medium spatial resolution and relatively
consistent spectral and radiometric resolution (Jensen et al., 1993; Ji et al., 2001; Yang and Lo, 2002;
Yang et al., 2003; Zhao et al., 2004; Christiane et al., 2005). However, because the urban environment is
generally characterized by highly heterogeneous surface covers with substantial inter-pixel and intra-pixel
∗1 Project supported by the National Natural Science Foundation of China (NSFC) (No. 30571112).
∗2 Corresponding author. E-mail: [email protected].
LAND USE CHANGE DETECTION USING SATELLITE IMAGES 97

changes, change detection capabilities are intrinsically limited by the spatial resolution for the digital
images in urban area (Lu et al., 2004). It is difficult to obtain detailed information of land use and land
use changes within the urban area using only the Landsat data. The demand for higher spatial resolution
data and effective processing methods has been important and urgent in urban land use analysis. The
application of multisensor data, however, may provide increased interpretation capabilities and more
reliable results and have the potential to more accurately detect land cover changes through integration
of different characteristics of single sensor data (Hack and Bechdol, 1999).
Hangzhou has witnessed notable economic and social development in recent years, which results in
significant land use changes. This study used a method combining principal component analysis (PCA)
of multisensor satellite images and supervised classification to detect and monitor land use changes in
the city proper of Hangzhou, to provide current, reliable land use information for management and
decision-making.

MATERIALS AND METHODS

Study area

The city proper of Hangzhou was taken as the study area. Hangzhou City is the capital of Zhejiang
Province, one of the seven ancient capitals and key scenic tourism and historical culture cities in China.
As a central city on the southern wing of the Yangtze delta, Hangzhou covers an area of 3 068 km2 with
a population of 3.9 million. Its GDP was RMB yuan 209.977 billion in 2003, an increase of 15.2% over
the previous year. It has also maintained an annual growth rate of more than 10% for successive 13
years. Its total economic strength ranked the eighth among all the large- and medium-sized cities in
China, second among all the provincial capitals, and third among all the sub-provincial cities. Its GDP
per capita reached RMB yuan 32 700 or US $ 3 950, as converted according to the foreign exchange rate
of that year (2003).

Data pre-processing

Two different source datasets were selected for detecting the land use changes in the study area.
One is the Landsat-7 enhanced thematic mapper (ETM) panchromatic (PAN) imagery, recorded on
March 13, 2001, with a 15-m spatial resolution. The other is SPOT (systeme pour l’observation de la
terre or earth observation satellite)-5 multispectral (XS) imagery collected on December 15, 2003. The
main characteristics of all the spectral bands of the sensors used in this study are shown in Table I.
TABLE I
Sensor spectral bands and resolution of SPOT (systeme pour l’observation de la terre or earth observation satellite)-5
multispectral (XS) and Landsat-7 enhanced thematic mapper (ETM) panchromatic (PAN) data used in this study

Sensor Band Spatial resolution Spectral coverage

m μm
SPOT-5 (XS) Green 10 0.49–0.61
Red 10 0.61–0.68
Near infrared (IR) 10 0.78–0.89
Short-wave infrared (SWIR) 20 1.58–1.75
ETM (PAN) PAN 15 0.52–0.90

Of the various pre-processing steps for change detection, multidate image registration and radiomet-
ric correction are the most important. The SPOT-5 image in 2003 was geometrically rectified using the
control points taken from a topographic map at 1:50 000 scale. The ETM image was then registered to
the rectified SPOT-5 image with the same projection. The two images were resampled into the same
10-m spatial resolution using the nearest-neighbor method. A root mean square (RMS) error of less
than 0.6 was obtained. The relative radiometric normalization method was employed because no in-situ
98 J. S. DENG et al.

atmospheric data were available at the time of satellite overpass.

Change detection procedure

The first step was to stack multitemporal images into one single vector (X). Suppose there are two
images dated time i and time j available for the detection of land use changes, these two images can be
referred to as two pixel vectors:

Xi = [Xi1 , Xi2 , . . . , Xin ] (1)

Xj = [Xj1 , Xj2 , . . . , Xjn ] (2)

X = [Xi1 , Xi2 , . . . , Xin , Xj1 , Xj2 , . . . , Xjn ] (3)

where Xi1 to Xin and Xj1 to Xjn are the brightness values of the pixel X in bands from 1 to n for time
i and j, respectively. The principal component analysis (PCA) is used to reduce the dimensionality
of a dataset consisting of a large number of interrelated variables, while still retaining the maximum
information of the variation present in the dataset. This is achieved by transforming the original dataset
to a new set of variables, named principal components (PCs), which are uncorrelated and ordered with
the first few components explaining most of the variation present in all of the original variables (Li and
Yeh, 1998; Jolliffe, 2002).
The vector consisting of a total of 2n bands is difficult to be processed because of the large size of
the dataset. The presence of too many bands can make the operation of supervised training difficult as
the training usually employs a color composite of three bands (Yeh and Li, 1997). The most commonly
used technique for compressing remote sensing data is PCA, which can be standardized using a corre-
lation matrix instead of a covariance matrix (Singh and Harrison, 1985; Fung and LeDrew, 1987). The
standardized PCA has a better alignment along the object of interest and appears to be more effective
than non-standardized PCA for detection of change (Fung and LeDrew, 1987).
The first principal component explains the maximum amount of the variance of the original data.
The second principal component explains the largest amount of the variance in the data that has not
yet been explained by the first principal component, and so forth (Taylor, 1977). Although n principal
components may be acquired in the analysis, only the first few principal components account for a large
proportion of the variance in the data. In some situations, almost 100 percent of the variance can be
represented by these few components. In this study, standardized PCA was applied to the stacked 2n
bands. Most variance was demonstrated in the first four components (Fig. 1).
Interactive supervised classification of land use changes was applied on the compressed image formed
by PCA. As the compressed PCA image contains most of the information of the original images, it should
be possible to use the image for the classification of land use changes. The signature of each class can
be created by interactive supervised training. In the classification, it is important that training samples
should be the representatives for each class. The samples for the supervised classification are selected
from the compressed PCA image by displaying the two original images and the compressed PCA image
on the screen and using the cursor to identify the training sites. The two original images, which are
linked to the PCA image, are simply used as supports for the identification of training samples for land
use change classes in the PCA image.
The land use change classes have to be discernable in the PCA image for the training. The original
images need to be shown for providing useful information in identifying change classes. Fortunately, all
the change classes of the study area can be identified in the PCA image. The next step is to perform
a maximum likelihood classification on this compressed components image based on the supervised
training. A thematic image containing the information of the land use and the land use changes is
derived from the classification.
LAND USE CHANGE DETECTION USING SATELLITE IMAGES 99

Fig. 1 Principal component analysis (PCA) of the stacked multisensor images: color composition image of Landsat-7
enhanced thematic mapper (ETM) in 2001 (a); color composition image of SPOT (systeme pour l’observation de la terre
or earth observation satellite)-5 in 2003 (b); and the first four components of the compressed PCA image (c–f).

Accuracy assessment of change detection

Accuracy assessment is an important part of classification and change detection processes (Lu et
al., 2005). A common method for accuracy assessment applies an error matrix, which can be used as
a starting point for a series of descriptive and analytical statistical analyses. Some analysts continue
to perform error evaluation based only on the training pixels used to train or seed the classification
algorithm. Unfortunately, the location of these training sites is usually not random. They are biased
by the analyst’s prior knowledge of where certain land-use/land-cover types exist in the scene (Jensen,
1996). For these reasons, stratified random sampling is usually recommended so that the sampling
points are fairly spread in each land use change category (Congalton, 1991). This is better than a purely
random sampling which may ignore many smaller categories.
To evaluate the performance of the proposed change detection method as compared with the conven-
tional one, post-classification comparison method was also performed using the Landsat-7 ETM (XS)
image in 2000 and SPOT-5 (XS) image in 2003 in this study. A total of 300 reference points were selected
to calculate the detection accuracy. Aerial photograph in 2000 and Landsat-7 ETM (XS) image in 2001
were used as ground truth data for 2001, and field survey using global positioning system (GPS) in 2003
was also employed to assist in the identification of changes on SPOT-5 image of 2003.

RESULTS

The land use change map derived from the compressed PCA images between 2001 and 2003 is given
in Fig. 2. Both land use and land use changes in the city proper of Hangzhou detected and identified
using the method of combining PCA with interactive supervised classification, were grouped into the
following ten classes: cropland, built-up land, orchard, water area, construction site, forest, built-up
land from cropland, construction site from cropland, water area (fishpond) from cropland, and built-up
land from orchard (Fig. 2).
The overall accuracy of the land use change detection was determined to be 90.67% and the Kappa
index was 0.89. Clearly, the results met the minimum standard of 85% stipulated by the USGS classifi-
cation scheme (Anderson et al., 1976). From Table II it can be found that there were misclassifications
especially between the built-up land and the cropland to built-up land and between the water area and
100 J. S. DENG et al.

Fig. 2 Land use and land use changes in the city proper of Hangzhou.

TABLE II

Confusion matrix of the principal component analysis (PCA) and the post-classification methods

Method Class Reference data Producer’s User’s

accuracy accuracy
1 2 3 4 5 6 7 8 9 10 Total
%
PCA Cropland 55 1 56 96.49 98.21
Built-up land 65 2 10 77 92.86 84.42
Orchard 1 11 1 13 100.00 84.62
Water area 32 32 86.49 100.00
Construction site 1 10 2 13 83.33 76.92
Forest 1 50 51 100.00 98.04
Built-up land from cropland 3 15 18 60.00 83.33
Water area from cropland 3 12 15 100.00 80.00
Construction site from cropland 2 13 15 81.25 86.67
Built-up land from orchard 1 9 10 90.00 90.00
Total 57 70 11 37 12 50 25 12 16 10 300
Post-classification Cropland 51 4 1 56 92.73 91.07
Built-up land 65 2 10 77 82.28 84.42
Orchard 1 1 10 1 13 83.33 76.92
Water area 1 31 32 86.11 96.88
Construction site 1 1 8 3 13 72.73 61.54
Forest 1 1 49 51 100.00 96.08
Built-up land from cropland 5 13 18 56.52 72.22
Water area from cropland 1 3 11 15 100.00 73.33
Construction site from cropland 1 1 3 10 15 71.43 66.67
Built-up land from orchard 1 9 10 90.00 90.00
Total 55 79 12 36 11 49 23 11 14 10 300

the cropland to water area. The confusions may mainly result from the spectral and temporal vari-
ances and differences in the spatial resolution from different sensors. The misregistration between the
multisensor images may also contribute to the confusions. Overall, the user’s and producer’s accuracies
were high. The results had reasonably high accuracy, and thus were feasible and satisfying for land use
LAND USE CHANGE DETECTION USING SATELLITE IMAGES 101

change detection in the urban area for this study. The overall accuracy and Kappa index of the change
detection using post-classification were only 85.67% and 0.83, respectively. At the same time, we can
find higher producer’s accuracy and user’s accuracy for the proposed PCA method and more land use
confusions for the post-classification method. This indicated that the proposed PCA method was better
than the conventional method.
The land use change matrix of the city proper of Hangzhou from 2001 to 2003 is shown in Table
III. From this table, it could be found that there was a considerable change (10.03% of the total area)
in the land use in the study area during the two-year period. There were three major types of land
use conversion: from cropland into built-up land, construction site, and fishpond. Fewer changes from
orchard land into built-up land were detected. Both built-up land and construction site increased in
area by 3 192.02 and 2 007.52 ha, respectively, while cropland and orchard decreased in area by 6 953.22
and 290.92 ha, respectively. During the same period, the increase of water area was determined to be
2 044.6 ha in this area.
TABLE III
Matrix of land use change in 2001–2003

2001 2003 2001

total
Cropland Built-up land Orchard Water Construction site Forest
ha
Cropland 16 288.83 2 901.10 2 044.60 2 007.52 23 242.05
(70.08%) (12.48%) (8.80%) (8.64%)
Built-up 23 656.28 23 656.28
land (100%)
Orchard 290.92 1 289.02 1 579.94
(18.41%) (81.59%)
Water area 7 954.66 7 954.66
(100%)
Construction 1 487.54 1 487.54
site (100%)
Forest 14 319.17 14 319.17
(100%)
2003 total 16 288.83 26 848.30 1 289.02 9 999.26 3 495.06 14 319.17 72 239.64

It can be seen that major changes occurred in cropland. Many of them was transformed into urban
land uses (built-up land and construction site), which was triggered by the booming economy and
land speculation. A large amount of cropland was encroached due to urban development, such as real
estate, economic and technological development zones, and large university campus centers (Figs. 1–2).
Table III further indicated that of the increase in urban land use (5 199.54 ha), 94.40% resulted from
cropland and 5.60% from orchard. As a result, these changes constitute permanent loss of agricultural
land because it is impossible to convert urban land use back to agricultural land.
A special phenomenon could be found in the city proper of Hangzhou. A vast amount of cropland
was no longer used to grow grain as more than 8% of the cropland was converted to fishpond (Table
III, Fig. 3). The change from cropland to fishpond was the result of agricultural restructuring triggered
by the market economy because of higher economic value of aquatic breeding compared with that of
grain growing. With the rise in standard of living, the consumption of aquatic products has a promising
market, which results in the emergence of new fishponds.

DISCUSSION

The use of SPOT-5 (XS) and Landsat-7 ETM (PAN) data to detect the land use and land use
changes proved successful in the city proper of Hangzhou. The method of combining PCA and digi-
tal image supervised classification demonstrated the ability to provide comprehensive information on the
102 J. S. DENG et al.

Fig. 3 Examples for the conversions from cropland to fishpond, 2001–2003.

direction, nature, rate, and location of land use and land use changes as a result of rapid industrialization
and urbanization. The overall accuracy and Kappa index of the land use change detection using the
proposed method were calculated to be 90.67% and 0.89 but were only 85.67% and 0.83 for the post-
classification method, respectively. The proposed PCA method showed better performance compared
with the conventional method for the detection of land use changes. However, misclassification of the
land use and confusion in the land use changes existed in this study. Although the change detection had a
relative high overall accuracy, the accuracy for different classes varied. On the whole, the result could be
considered to be feasible and satisfying. The spectral and temporal variances and difference in spatial
resolution from different sensors contributed to the land use confusions and misclassification, which
were common and intractable problems in multitemporal and multisensor change detection analysis.
The misregistration between the multisensor images may also exaggerate the problems because accurate
geometric fidelity is particularly important for the detection of land use changes. Besides, another issue
that must be considered is the land use classification process. An effective and suitable method could
decrease the misclassification and increase accuracy at the same time.
Concomitant with the rapid economic development and urbanization, land use in the city proper of
Hangzhou was rapidly changing: 10.03% of the total area underwent changes from 2001–2003. It was
found that the built-up land and construction site notably increased in this area, while the cropland and
orchard decreased. The extension of urban land constituted major loss of the cropland along with the
industrialization and urbanization. At the same time, a major change from cropland to fishpond due to
agricultural restructuring and market economic development was identified. As food supply is a main
component of sustainable development, it is important to ensure that unnecessary urban development on
valuable cropland is prevented to the maximum extent (Yeh and Li, 1999). Further studies should focus
on the land development process in the city proper of Hangzhou to analyze the spatial restructuring of
the land use changes and their impact on the environment.

REFERENCES
Anderson, J. R., Hardy, E. E., Roach, J. T. and Witmer, R. E. 1976. A Land Use and Land Cover Classification System
for Use with Remote Sensor Data. Geological Survey Professional Paper 964. United States Government Printing
Office, Washington.
Christiane, W., Chryssanthi, P. and Jachy, H. 2005. Urban development in the Athens metropolitan area using remote
sensing data with supervised analysis and GIS. Int. J. Remote Sens. 26(4): 785–796.
Congalton, R. G. 1991. A review of assessing the accuracy of classification of remotely sensed data. Remote Sens. Environ.
37: 35–46.
Fung, T. and LeDrew, E. 1987. Application of principal components analysis change detection. Photogram. Eng. Remote
Sens. 59: 991–996.
LAND USE CHANGE DETECTION USING SATELLITE IMAGES 103

Hack, B. and Bechdol, M. 1999. Multisensor remote sensing data for land use/cover mapping. Comput. Environ. Urban
Syst. 23: 53–69.
Jensen, J. R. 1996. Introductory Digital Image Processing: A Remote Sensing Perspective. Prentice Hall, Upper Saddle
River.
Jensen, J. R., Cowen, D. J., Althausen, J. D., Narumalani, S. and Weatherbee, O. 1993. An evaluation of the coastwatch
change detection protocol in South Carolina. Photogram. Eng. Remote Sens. 59(9): 1 039–1 046.
Ji, C. Y., Liu, Q. H., Sun, D. F., Wang, S., Lin, P. and Li, X. W. 2001. Monitoring urban expansion with remote sensing
in China. Int. J. Remote Sens. 22(8): 1 441–1 455.
Jolliffe, I. T. 2002. Principal Component Analysis. Springer Press, New York.
Li, X. and Yeh, A. G. O. 1998. Principal component analysis of stacked multi-temporal images for the monitoring of rapid
urban expansion in the Pearl River Delta. Int. J. Remote Sens. 19(8): 1 501–1 518.
Lu, D., Mausel, P., Brondizio, E. and Moran, E. 2004. Change detection techniques. Int. J. Remote Sens. 25(2):
2 365–2 407.
Rogan, J. and Chen, D. M. 2004. Remote sensing technology for mapping and monitoring land-cover and land-use change.
Prog. Plan. 61: 301–325.
Singh, A. and Harrison, A. 1985. Standardized principal component. Int. J. Remote Sens. 6: 883–896.
Taylor, P. 1977. Quantitative Methods in Geography: An Introduction to Spatial Analysis. Houghton Mifflin Company,
Boston.
Yang, L. M., George, X., Jacqueline, M. K. and Brian, D. 2003. Urban land-cover change detection through sub-pixel
imperviousness mapping using remotely sensed data. Photogram. Eng. Remote Sens. 59(9): 1 039–1 046.
Yang, X. and Lo, C. P. 2002. Using a time series of satellite imagery to detect land use and land cover changes in the
Atlanta, Georgia metropolitan area. Int. J. Remote Sens. 23(9): 1 775–1 798.
Yeh, A. G. O. and Li, X. 1997. An integrated remote sensing and GIS approach in the monitoring and evaluation of rapid
urban growth for sustainable development in the Pearl River Delta, China. Int. Plan. Stud. 2(2): 193–210.
Yeh, A. G. O. and Li, X. 1999. Economic development and agricultural land loss in the Pearl River Delta, China. Habitat
Int. 23(3): 373–390.
Zhao, G. X., Lin, G. and Warner, T. 2004. Using thematic mapper data for change detection and sustainable use of
cultivated land: a case study in the Yellow River delta, China. Int. J. Remote Sens. 25(13): 2 509–2 522.