scenes of Landsat MSS, ETM+, and ASTER data tended to be consistent (Table 12.4).
With the highest dissimilarity values of 0.142, 0.903, and 1.0, the original image
scenes were mostly different from its resample images at the first aggregation level for
both Landsat MSS (120 m) and ETM+ (60 m) data but at the fourth aggregation level
for the ASTER image (240 m). The zero dissimilarity values reported along the
original image row/column in Table 12.4 may suggest that at certain aggregation
levels the fractal complexity of the original scenes could be less dissimilar with those
found within their resampled images. These scales were found to be 240–960 m for
the MSS, 480 m for the ETM+, and 30–60 m and 960 m for the ASTER images.
When all three image types were considered, it should be noted that at the 480 m
aggregation level, the dissimilarity values were either the lowest (for the MSS and
ETM+ images) or the second lowest (for the ASTER image). This implies that the
textural complexity of all three image types was mostly alike after being resampled to
the 480 m pixel size. This corresponds to the finding from Figure 12.8a. The total
dissimilarity values for all original and all resampled images were also calculated for
each image type. When only the resampled images were taken into account, the
highest total dissimilarity values were found to be 2.043, 2.591, and 2.763 for the
MSS, ETM+, and ASTER data, respectively. Interestingly, these all corresponded to
the scale of 480 m, although the ones from the 60 m (3.349) were higher for the
ASTER image. This suggests that for all three image types this aggregation level
could generate images that were more distinct from those from other aggregation
levels (Figure 12.8a).
More discrepancies were observed from image subscenes derived from the ETM+,
ASTER, and IKONOS images (Table 12.5). Since FDs calculated from the 240 m
level were all higher than the theoretical limit of 3.0, only dissimilarity values reported
for 15–30 m were analyzed. In general, the resampled images mostly/least dissimilar
from the original image subscenes tended to vary by image types as well as the
dominant landscape (downtown and residential). With zero dissimilarity values
between the original and resampled images for both landscapes, the resampling
process did not seem to significantly change the fractal properties recorded in the
ETM+ images. For the ASTER subscenes, with the highest dissimilarity values of
0.994, and 1.0, the original images were mostly different from its resample images at
the second aggregation level (60 m) for the residential landscape but the third level
(120 m) for the downtown landscape. When the lowest dissimilarity values were
examined, it was found that the two ASTER subscenes were not similar across all
aggregation scales. The results from the IKONOS subscenes tended to be more
consistent. With the higher dissimilarity values of 0.868 and 0.966, the original scenes
were both observed to be greatly different from their resampled images at the third
aggregation level (30 m), although those from the four aggregation level (60 m) was
even higher for the downtown subscene (1.0). Besides, both original scenes were least
dissimilar to their resampled images at the 15 and 120 m levels, suggesting the
difference in landscape may have the least impact on the scaling change on the
textural complexity of the IKONOS image at the two scales. When all three image
types were considered, the dissimilarity values between the original and resampled
subscenes tended to be the lowest at the 120 m aggregation level for the two
RESULTS
247
With the highest dissimilarity values of 0.142, 0.903, and 1.0, the original image
scenes were mostly different from its resample images at the first aggregation level for
both Landsat MSS (120 m) and ETM+ (60 m) data but at the fourth aggregation level
for the ASTER image (240 m). The zero dissimilarity values reported along the
original image row/column in Table 12.4 may suggest that at certain aggregation
levels the fractal complexity of the original scenes could be less dissimilar with those
found within their resampled images. These scales were found to be 240–960 m for
the MSS, 480 m for the ETM+, and 30–60 m and 960 m for the ASTER images.
When all three image types were considered, it should be noted that at the 480 m
aggregation level, the dissimilarity values were either the lowest (for the MSS and
ETM+ images) or the second lowest (for the ASTER image). This implies that the
textural complexity of all three image types was mostly alike after being resampled to
the 480 m pixel size. This corresponds to the finding from Figure 12.8a. The total
dissimilarity values for all original and all resampled images were also calculated for
each image type. When only the resampled images were taken into account, the
highest total dissimilarity values were found to be 2.043, 2.591, and 2.763 for the
MSS, ETM+, and ASTER data, respectively. Interestingly, these all corresponded to
the scale of 480 m, although the ones from the 60 m (3.349) were higher for the
ASTER image. This suggests that for all three image types this aggregation level
could generate images that were more distinct from those from other aggregation
levels (Figure 12.8a).
More discrepancies were observed from image subscenes derived from the ETM+,
ASTER, and IKONOS images (Table 12.5). Since FDs calculated from the 240 m
level were all higher than the theoretical limit of 3.0, only dissimilarity values reported
for 15–30 m were analyzed. In general, the resampled images mostly/least dissimilar
from the original image subscenes tended to vary by image types as well as the
dominant landscape (downtown and residential). With zero dissimilarity values
between the original and resampled images for both landscapes, the resampling
process did not seem to significantly change the fractal properties recorded in the
ETM+ images. For the ASTER subscenes, with the highest dissimilarity values of
0.994, and 1.0, the original images were mostly different from its resample images at
the second aggregation level (60 m) for the residential landscape but the third level
(120 m) for the downtown landscape. When the lowest dissimilarity values were
examined, it was found that the two ASTER subscenes were not similar across all
aggregation scales. The results from the IKONOS subscenes tended to be more
consistent. With the higher dissimilarity values of 0.868 and 0.966, the original scenes
were both observed to be greatly different from their resampled images at the third
aggregation level (30 m), although those from the four aggregation level (60 m) was
even higher for the downtown subscene (1.0). Besides, both original scenes were least
dissimilar to their resampled images at the 15 and 120 m levels, suggesting the
difference in landscape may have the least impact on the scaling change on the
textural complexity of the IKONOS image at the two scales. When all three image
types were considered, the dissimilarity values between the original and resampled
subscenes tended to be the lowest at the 120 m aggregation level for the two
RESULTS
247