complexity than the downtown landscape. However, when the ETM+ and ASTER
images were taken into account, the same result was observed to be true only with
their NIR bands. Therefore, for ETM+ and ASTER images, their NIR bands may be
better than the G and R bands to characterize the residential landscape in the scene,
while their G and R bands will be a better option than the NIR bands to capture the
downtown landscape in the scene.
Obviously, when the spatial resolution increased from 30 to 15 m and then to 4 m,
all resultant FD values were different, illustrating the scale effect on fractal measurement due to the sensor’s spatial resolution. Besides, the three reflective bands did not
behave the same to the fractal measurement. With finer measurement scales, FDs from
the NIR bands increased consistently while those from the two visible bands tended to
decline instead. Ideally, more spatial details in terms of the level of variability within a
single geographic area should be recorded by images with smaller pixel sizes.
Potentially, this will allow the creation of images with rougher texture. Yet this
only seemed to be true for the NIR data, indicating this may be more applicable than
other bands for studying the scale effect on the overall spatial complexity resulting
directly from the sensor’s spatial resolution. For the two visible bands (G and R
bands), however, the recorded shapes of the image texture did not grow more complex
with increasing spatial resolution. They thus fail to characterize the spatial complexity
captured by better spatial resolutions.
12.4.2 Fractals Analysis Using LULC Maps
Six LULC maps were created from Landsat (MSS, TM, and ETM+) and ASTER
images acquired in 1975, 1985, 1991, 1995, 2000, and 2001 and their overall accuracy
of classification was 83.20, 90.00, 88.80, 87.30, 89.00, and 89.87%, respectively. The
two LULC maps were also derived from the two IKONOS images dated in 2001 and
2003 and they were found to be highly accurate with the overall accuracy of 93.33%
for IKN01 and 94% for IKN03. Figure 12.5 illustrates LULC maps derived from the
ETM+ 2000 image (a), the ASTER 2001 image (b), subscenes from the IKONO
2001 (c), and the 2003 image (d). The LULC change matrix from 1975 to 2000 in
Table 12.2 clearly shows the study area’s landscape had undergone a significant
change during the 25-year period. Urban built-up lands and grassland had each
increased in sizes by 42.68 and 67.28%, while the areas of cropland and pasture had
decreased by 76.70%. Table 12.2 also summarizes the city population change over the
time. Although the Indianapolis population reduced by .91% from 1975 to 1985, its
residents increased consistently after 1985, with the greatest change occurring
between 1990 and 1995 (3.46%).
12.4.2.1 Fractals Analysis Using Raw Red Bands Classified by LULC Classes In
Section 12.4.1 the red band was found to contain more spatial information content
than others for most images used in this research, and they were used frequently for
the analysis presented in this section and Sections 12.4.3 and 12.4.4. The five Landsat
LULC maps were first employed as filters to mask out specific areas covered by
various LULC types using their corresponding raw red bands. Table 12.3 presents
RESULTS
239
images were taken into account, the same result was observed to be true only with
their NIR bands. Therefore, for ETM+ and ASTER images, their NIR bands may be
better than the G and R bands to characterize the residential landscape in the scene,
while their G and R bands will be a better option than the NIR bands to capture the
downtown landscape in the scene.
Obviously, when the spatial resolution increased from 30 to 15 m and then to 4 m,
all resultant FD values were different, illustrating the scale effect on fractal measurement due to the sensor’s spatial resolution. Besides, the three reflective bands did not
behave the same to the fractal measurement. With finer measurement scales, FDs from
the NIR bands increased consistently while those from the two visible bands tended to
decline instead. Ideally, more spatial details in terms of the level of variability within a
single geographic area should be recorded by images with smaller pixel sizes.
Potentially, this will allow the creation of images with rougher texture. Yet this
only seemed to be true for the NIR data, indicating this may be more applicable than
other bands for studying the scale effect on the overall spatial complexity resulting
directly from the sensor’s spatial resolution. For the two visible bands (G and R
bands), however, the recorded shapes of the image texture did not grow more complex
with increasing spatial resolution. They thus fail to characterize the spatial complexity
captured by better spatial resolutions.
12.4.2 Fractals Analysis Using LULC Maps
Six LULC maps were created from Landsat (MSS, TM, and ETM+) and ASTER
images acquired in 1975, 1985, 1991, 1995, 2000, and 2001 and their overall accuracy
of classification was 83.20, 90.00, 88.80, 87.30, 89.00, and 89.87%, respectively. The
two LULC maps were also derived from the two IKONOS images dated in 2001 and
2003 and they were found to be highly accurate with the overall accuracy of 93.33%
for IKN01 and 94% for IKN03. Figure 12.5 illustrates LULC maps derived from the
ETM+ 2000 image (a), the ASTER 2001 image (b), subscenes from the IKONO
2001 (c), and the 2003 image (d). The LULC change matrix from 1975 to 2000 in
Table 12.2 clearly shows the study area’s landscape had undergone a significant
change during the 25-year period. Urban built-up lands and grassland had each
increased in sizes by 42.68 and 67.28%, while the areas of cropland and pasture had
decreased by 76.70%. Table 12.2 also summarizes the city population change over the
time. Although the Indianapolis population reduced by .91% from 1975 to 1985, its
residents increased consistently after 1985, with the greatest change occurring
between 1990 and 1995 (3.46%).
12.4.2.1 Fractals Analysis Using Raw Red Bands Classified by LULC Classes In
Section 12.4.1 the red band was found to contain more spatial information content
than others for most images used in this research, and they were used frequently for
the analysis presented in this section and Sections 12.4.3 and 12.4.4. The five Landsat
LULC maps were first employed as filters to mask out specific areas covered by
various LULC types using their corresponding raw red bands. Table 12.3 presents
RESULTS
239