not necessarily lead to more fragmented fractal patterns as suggested by higher FDs. It
is still uncertain how effectively FD values can reveal the true change over time, since
the discrepancy in FD values may also be caused by other factors, such as the changes
in spectral reflectance of an individual band in case of raw images and the spatial
composition of LULC classes in LULC images. More future research efforts are
needed to examine this uncertainty.
12.5 DISCUSSION AND CONCLUSION
This chapter sought to explore the potential of fractal measurement in characterizing
the dynamic changes of urban landscape of Indianapolis, Indiana, at multiple spatial
and temporal scales. Eight satellite images collected by various sensors (MSS, TM,
ETM+, ASTER, and IKONOS) in multiple years (from 1975 to 2003) and presented
in different formats (raw, classified, and resampled images) were analyzed extensively using the TP algorithm in ICAMS. Overall, the fractal analysis based on the TP
method was found to be useful in discriminating the geometric complexity of land
surface features in both the unclassified and classified images collected by the five
sensors. The examination of the original raw images shows that their overall spatial
complexity will vary by the spectral bands, the sensor/image types, as well as the
application being considered. For images collected by MSS, ETM+, and ASTER
covering the whole city, the reflective bands, the red and NIR bands in particular,
tended to create more spatially complex surfaces than other spectral bands for all
image types being examined. Hence, these two bands may be better than others to
characterize the texture of the overall landscape for the study area. However, for
images dominated by a given LULC type such as downtown and residential land use,
their textural complexity will depend on both sensor/image types and individual
spectral bands. Among the three sensors—ETM+, ASTER, and IKONOS—the G, R,
and NIR images collected by the IKONOS sensor are more suitable than those by the
ETM+ and ASTER sensors to reveal the rougher textural pattern in the residential
landscape as compared to the downtown landscape. Nevertheless, when only the
ETM+ and ASTER images are taken into account, their G and R bands may be better
than the NIR bands to capture the geometric complexity in the downtown land use,
while their NIR bands will be more useful than the G and R bands to characterize the
residential land use. On the other hand, the results based on the LULC maps indicated
that fractal analysis was useful in identifying the spatial complexity of individual
LULC classes recorded in the original image. The use of fragmented raw red bands
filtered by LULC types showed that grassland, forest, and urban built-up land were
more fragmented than other LULCs. Additionally, with only the LULC maps, the
calculated FD results could illustrate that the region’s urban landscape had experienced changes over time. Nevertheless, the fractal analysis alone was somewhat
ineffective to detect the actual temporal change. Detailed quantitative change
information is needed to assist in the interpretation of resultant FD values.
The potential of fractal measurement for illustrating the scale effect due to a
sensor’s nominal spatial resolution was examined. Based on image subscenes
DISCUSSION AND CONCLUSION
249
is still uncertain how effectively FD values can reveal the true change over time, since
the discrepancy in FD values may also be caused by other factors, such as the changes
in spectral reflectance of an individual band in case of raw images and the spatial
composition of LULC classes in LULC images. More future research efforts are
needed to examine this uncertainty.
12.5 DISCUSSION AND CONCLUSION
This chapter sought to explore the potential of fractal measurement in characterizing
the dynamic changes of urban landscape of Indianapolis, Indiana, at multiple spatial
and temporal scales. Eight satellite images collected by various sensors (MSS, TM,
ETM+, ASTER, and IKONOS) in multiple years (from 1975 to 2003) and presented
in different formats (raw, classified, and resampled images) were analyzed extensively using the TP algorithm in ICAMS. Overall, the fractal analysis based on the TP
method was found to be useful in discriminating the geometric complexity of land
surface features in both the unclassified and classified images collected by the five
sensors. The examination of the original raw images shows that their overall spatial
complexity will vary by the spectral bands, the sensor/image types, as well as the
application being considered. For images collected by MSS, ETM+, and ASTER
covering the whole city, the reflective bands, the red and NIR bands in particular,
tended to create more spatially complex surfaces than other spectral bands for all
image types being examined. Hence, these two bands may be better than others to
characterize the texture of the overall landscape for the study area. However, for
images dominated by a given LULC type such as downtown and residential land use,
their textural complexity will depend on both sensor/image types and individual
spectral bands. Among the three sensors—ETM+, ASTER, and IKONOS—the G, R,
and NIR images collected by the IKONOS sensor are more suitable than those by the
ETM+ and ASTER sensors to reveal the rougher textural pattern in the residential
landscape as compared to the downtown landscape. Nevertheless, when only the
ETM+ and ASTER images are taken into account, their G and R bands may be better
than the NIR bands to capture the geometric complexity in the downtown land use,
while their NIR bands will be more useful than the G and R bands to characterize the
residential land use. On the other hand, the results based on the LULC maps indicated
that fractal analysis was useful in identifying the spatial complexity of individual
LULC classes recorded in the original image. The use of fragmented raw red bands
filtered by LULC types showed that grassland, forest, and urban built-up land were
more fragmented than other LULCs. Additionally, with only the LULC maps, the
calculated FD results could illustrate that the region’s urban landscape had experienced changes over time. Nevertheless, the fractal analysis alone was somewhat
ineffective to detect the actual temporal change. Detailed quantitative change
information is needed to assist in the interpretation of resultant FD values.
The potential of fractal measurement for illustrating the scale effect due to a
sensor’s nominal spatial resolution was examined. Based on image subscenes
DISCUSSION AND CONCLUSION
249