relatively symmetrical, it has the possibility to expand in all directions. Like many
other cities in the United States, its urban area enlarges at the expense of the loss of
adjacent agricultural areas and forestlands. Indianapolis is now the most populous city
of the state and the second most populous capital in the United States.
Data sets used in this chapter consisted of eight satellite images that were acquired
by five types of sensors (Table 12.1). For these data, only MSS, TM, and ASTER
image scenes cover the whole study area. The two IKONOS scenes roughly cover the
downtown and residential areas of the city. Since the two images partially overlap
each other, to avoid confusion that may result from this overlap on the fractal analysis,
the two IKONOS data were used to create two subscenes (7 ´ 7 km
2
) to include only
the downtown and residential land use of the city (Figure 12.2). All satellite images
were first georectified to a common Universal Transverse Mercator (UTM) coordinate
system using the 1 : 24,000 scale topographic maps as reference. For each image, 25
ground control points were selected to generate coefficients for a first-order polynomial, and a nearest-neighbor method was applied to resample the image according to
their nominal spatial resolution (Table 12.1). The resultant values of the root meansquare error (RMSE) were all found to be less than 0.4 pixel.
Eight classified images were derived by performing the ISODATA unsupervised
classification on the spectral bands that were available from all sensors [green, red,
and near-infrared (NIR) bands]. Five LULC types were then identified: cropland and
pasture (CroPas), water, urban built-up lands (Build-up), forest, and grass. Accuracy
assessment was conducted for the resultant classified images. Aerial photographs
ranging from 1 : 80,000 (for early photos) to 1 : 12,000 (for recent photos) scales were
used as the reference data. For each class, 50 points were randomly selected and
compared with the reference points to generate the producer’s accuracy and user’s
accuracy. The overall accuracy for each classified images was also computed.
The effect of changes of spatial scale and resolution on detecting landscape
patterns and changes are central to geography and mapping science such as remote
sensing (Dell’Acqua and Gamba, 2006; Frohn, 1998; Lam and Quattrochi, 1992). In
order to provide some insights into the scaling effect on the analytical results, two
image groups were used to perform the fractal analysis. The first image group
consisted of resampled images by smoothing the image spatial resolution. All the
selected reflective bands were resampled to different levels of pixel sizes. Specifically,
starting from their own nominal spatial resolutions, all MSS, ETM+, and ASTER
images were resampled up to 960 m and the two IKONOS images were resampled up
to 240 m. A total of 33 resampled images were thus generated. Besides, the underlying
dominant landscape pattern (e.g., downtown vs. residential) can potentially change
the analytical results dealing with scale and resolution. To reduce this effect, the
second group of data sets comprised only images covering the downtown and
residential landscapes but collected by sensors that are inherently different in their
spatial resolutions. Only images acquired by ETM+, ASTER, and IKONOS sensors
were considered as they were collected about the same time, which could reduce any
confusion that may result from the time factor. An additional 30 subscenes from the
ETM+ and ASTER images were created to cover the same areas as the 2 IKONOS
image subscenes. This included 12 subscenes from the three reflective bands, 4 from
234
MULTISCALE FRACTAL CHARACTERISTICS OF URBAN
other cities in the United States, its urban area enlarges at the expense of the loss of
adjacent agricultural areas and forestlands. Indianapolis is now the most populous city
of the state and the second most populous capital in the United States.
Data sets used in this chapter consisted of eight satellite images that were acquired
by five types of sensors (Table 12.1). For these data, only MSS, TM, and ASTER
image scenes cover the whole study area. The two IKONOS scenes roughly cover the
downtown and residential areas of the city. Since the two images partially overlap
each other, to avoid confusion that may result from this overlap on the fractal analysis,
the two IKONOS data were used to create two subscenes (7 ´ 7 km
2
) to include only
the downtown and residential land use of the city (Figure 12.2). All satellite images
were first georectified to a common Universal Transverse Mercator (UTM) coordinate
system using the 1 : 24,000 scale topographic maps as reference. For each image, 25
ground control points were selected to generate coefficients for a first-order polynomial, and a nearest-neighbor method was applied to resample the image according to
their nominal spatial resolution (Table 12.1). The resultant values of the root meansquare error (RMSE) were all found to be less than 0.4 pixel.
Eight classified images were derived by performing the ISODATA unsupervised
classification on the spectral bands that were available from all sensors [green, red,
and near-infrared (NIR) bands]. Five LULC types were then identified: cropland and
pasture (CroPas), water, urban built-up lands (Build-up), forest, and grass. Accuracy
assessment was conducted for the resultant classified images. Aerial photographs
ranging from 1 : 80,000 (for early photos) to 1 : 12,000 (for recent photos) scales were
used as the reference data. For each class, 50 points were randomly selected and
compared with the reference points to generate the producer’s accuracy and user’s
accuracy. The overall accuracy for each classified images was also computed.
The effect of changes of spatial scale and resolution on detecting landscape
patterns and changes are central to geography and mapping science such as remote
sensing (Dell’Acqua and Gamba, 2006; Frohn, 1998; Lam and Quattrochi, 1992). In
order to provide some insights into the scaling effect on the analytical results, two
image groups were used to perform the fractal analysis. The first image group
consisted of resampled images by smoothing the image spatial resolution. All the
selected reflective bands were resampled to different levels of pixel sizes. Specifically,
starting from their own nominal spatial resolutions, all MSS, ETM+, and ASTER
images were resampled up to 960 m and the two IKONOS images were resampled up
to 240 m. A total of 33 resampled images were thus generated. Besides, the underlying
dominant landscape pattern (e.g., downtown vs. residential) can potentially change
the analytical results dealing with scale and resolution. To reduce this effect, the
second group of data sets comprised only images covering the downtown and
residential landscapes but collected by sensors that are inherently different in their
spatial resolutions. Only images acquired by ETM+, ASTER, and IKONOS sensors
were considered as they were collected about the same time, which could reduce any
confusion that may result from the time factor. An additional 30 subscenes from the
ETM+ and ASTER images were created to cover the same areas as the 2 IKONOS
image subscenes. This included 12 subscenes from the three reflective bands, 4 from
234
MULTISCALE FRACTAL CHARACTERISTICS OF URBAN