Abstract
This paper proposes a technique to extract urban built-up
land features from Landsat Thematic Mapper (TM) and
Enhanced Thematic Mapper Plus (ETM) imagery taking two
cities in southeastern China as examples. The study selected
three indices, Normalized Difference Built-up Index (NDBI),
Modified Normalized Difference Water Index (MNDWI), and
Soil Adjusted Vegetation Index (SAVI) to represent three
major urban land-use classes, built-up land, open water
body, and vegetation, respectively. Consequently, the seven
bands of an original Landsat image were reduced into three
thematic-oriented bands derived from above indices. The
three new bands were then combined to compose a new
image. This considerably reduced data correlation and
redundancy between original multispectral bands, and thus
significantly avoided the spectral confusion of the above
three land-use classes. As a result, the spectral signatures of
the three urban land-use classes are more distinguishable in
the new composite image than in the original seven-band
image as the spectral clusters of the classes are well separated.
Through a supervised classification, a principal
components analysis, or a logic calculation on the new
image, the urban built-up lands were finally extracted with
overall accuracy ranging from 91.5 to 98.5 percent. Therefore,
the technique is effective and reliable. In addition, the
advantages of SAVI over NDVI and MNDWI over NDWI in the
urban study are also discussed in this paper.
Introduction
Urban spatial areas have expanded in an accelerated speed
during the last five decades, and rates of urban population
growth are higher than the overall growth in most countries
because urban areas are the locus of economic activity and
transportation nodes (Masek et al., 2000). Expanded urbanized
areas encroached on surrounding valuable natural lands
such as paddy fields, forestlands, or wetlands (Xu et al.,
2000). Urban areas are dominated by built-up lands with
impervious surfaces, and therefore the conversion of the
nature lands into these impervious built-up lands may have
significant impacts on the ecosystem, hydrologic system,
biodiversity, and local climate which can result in the negative
aspects such as the urban heat island phenomenon.
The study of urban spatial expansion and the resultant urban
Extraction of Urban Built-up Land Features
from Landsat Imagery Using a Thematicoriented
Index Combination Technique
Hanqiu Xu
heat island phenomenon always needs accurate data on urban
built-up areas such as the size, shape, and spatial context.
Therefore, a technique is required to quickly reveal the data.
Timely availability of the data is of great importance for
urban planners and decision makers. Fortunately, satellite
remote sensing technology offers considerable promise to
meet this requirement. With different spatial and spectral
resolutions, the satellite observations can provide globally
consistent and repetitive measurements of the Earth’s surface
conditions. The objective of this study is to develop a new
technique to extract urban built-up land features from Landsat
Thematic Mapper (TM) and Enhanced Thematic Mapper Plus
(ETM) imagery. This would allow urban planner and decision
makers to timely understand and evaluate urban growth with
related land-cover changes and be aware of the sustainable
usage of the invaluable nature lands.
Many researchers have made use of remote sensing
imagery to discriminate urban lands from non-urban lands.
A popular method for the definitions of urban areas started
with conventional multispectral classification. However, this
may not produce satisfactory accuracy, normally less than
80 percent, due to spectral confusion of the heterogeneous
urban built-up land class. Therefore, many studies have not
only used a single classification method to extract the urban
built-up lands but also combined different methods to
improve the extraction. Masek et al. (2000) identified urban
built-up areas of the Washington D.C. metropolitan area from
multi-date Landsat images based on an NDVI-differencing
approach with the assistance of an unsupervised classification
and achieved overall accuracy of 85 percent. Xu (2002)
extracted urban built-up lands of Fuqing City in southeastern
China by a combination of signature analysis and supervised
classification. Based on the analysis of spectral response
differences between built land and various non-built classes
within multispectral bands, urban land information was
extracted and then integrated with a classification layer to
generate a final product with improved accuracy. Zhang et al.
(2002) integrated a road density layer with spectral bands for
the post-classification change detection of Beijing, China.
This greatly reduced spectral confusion and increased accuracy
of the change detection. Zha et al. (2003) proposed the
Normalized Difference Built-up Index using TM4 and TM5 and
applied it in extracting urban areas of Najing City of China
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING December 2007 1381
College of Environment and Resources, Fuzhou University,
Key Laboratory of Data Mining and Information Sharing,
China’s Ministry of Education, Fuzhou, Fujian 350108,
China (fdy@public.fz.fj.cn).
Photogrammetric Engineering & Remote Sensing
Vol. 73, No. 12, December 2007, pp. 1381–1391.
0099-1112/07/7312–1381/$3.00/0
© 2007 American Society for Photogrammetry
and Remote Sensing
1382 December 2007 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
Figure 1. Landsat TM image of Quanzhou area (a) with its derived SAVI
(b), MNDWI (c), and NDBI (d) images, and ETM image of Fuzhou area
(e) with its derived SAVI (f), MNDWI (g), and NDBI (h) images (North to
the top; sizes of Quanzhou and Fuzhou images is 7.71 km 7.71 km
and 16.7 km 16.7 km, respectively). A color version of this figure
is available at the ASPRS website: www.asprs.org.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING December 2007 1383
from a Landsat TM image. The index-derived map was
further filtered using the NDVI to remove the vegetation
noise, as the vegetation information was mixed with the
extracted built-up lands. Guindon et al. (2004) mapped
urban land with a combination of spectral and spatial information.
This started with an unsupervised classification and
a segment-based classification, respectively. The two classifications
were then merged using rules to generate a final
product with enhanced land-use classes and accuracy. More
recently, Xian and Crane (2005) measured urban land expansion
of the Tampa Bay watershed of Florida by using a
regression tree algorithm to map urban impervious surfaces
and an unsupervised classification to reveal related landcover
classes which achieved an accuracy greater than
85 percent.
A new technique is proposed in this paper for the
extraction of urban build-up land information. The extraction
is mainly based on a new image derived from three
thematic indices, Soil Adjusted Vegetation Index (SAVI),
Modified Normalized Difference Water Index (MNDWI), and
Normalized Difference Built-up Index (NDBI). The technique
is demonstrated through the extraction of urban built-up
lands of Quanzhou and Fuzhou Cities in southeastern China
from Landsat TM/ETM images.
Methods
Study Area and Remote Sensing Data Source
The remotely sensed data used in this test are a Landsat TM
image (path 119, row 43, covering Quanzhou City) acquired
on 17 May 1996 and an ETM image (path 119, row 42,
covering Fuzhou City) acquired on 29 May 2003. The images
are cloud-free and have excellent quality. Sub-scenes covering
the test cities were further extracted from the two images
(Figure 1a and 1e). No preprocessing of the images was
carried out except a simple atmospheric scattering correction
procedure using the Dark Object Subtraction (DOS) method.
However, to maintain data objectivity and avoid introducing
uncertainty, the data provided in following Tables 1 through
5 are based on the original raw images. Nevertheless, the
extracted results from both raw and DOS-corrected images
were provided in Table 6 to examine whether there are any
differences between them. Also, the means mentioned in
following tables and figures are all calculated based on
related classified images.
Two cities are all located in Fujian Province. Fuzhou
City is the capital of the province, geographically ranging
from 119°13 to 119°25 East and 25°59 to 26°08 North,
and had a total urban area of approximately 119 km2 in the
study year of 2003. While Quanzhou City, about 200 km
south to the Fuzhou, ranges from 118°32 to 118°37 East
and 24°52 to 24°58 North and had a total urban area of
approximately 24 km2 in the study year of 1996.
Production of Index-derived Images
An urban area is a complex ecosystem composed of heterogeneous
materials. Nevertheless, there are still some generalizing
components among these materials. Ridd (1995) divided the
urban ecosystem into three components, i.e., impervious
surface material, green vegetation, and exposed soil while
ignoring water surfaces. However, the open water is an important
component of the urban surface and has to be taken into
consideration in this study. Accordingly, the urban land-use
was grouped into the other three generalized categories, i.e.,
built-up land, vegetation, and open water (Figure 2). Based on
these three elements, three indices, NDBI, SAVI, and MNDWI,
were selected in this study to represent above three major
land-use classes, respectively.
SAVI-derived Vegetation Image
There are various vegetation indices to enhance vegetation
information in remote sensing imagery usually by ratioing a
near-infrared (NIR) band to a red band. This takes advantage
of the high vegetation reflectance in NIR spectral range such
as TM band 4 and high pigment absorption of red light, such
as TM band 3 (Jensen, 2000). Although nearly everyone
working with the remote sensing of vegetation knows the
Normalized Difference Index (NDVI), this study employed
SAVI to highl