Weng et al. (2004) suggested that the operational scale for the urban thermal landscape
analysis yielded around 120 m in Indianapolis, Indiana. Remote sensing of urban heat
islands has traditionally used the normalized difference vegetation index (NDVI) as the
indicator of vegetation abundance to estimate the LST–vegetation relationship. Weng
et al. (2004) investigated the applicability of vegetation fraction derived from LSMA as an
alternative indicator of vegetation abundance. An experiment was conducted with a
Landsat ETM+ image of Indianapolis City, Indiana, acquired on June 22, 2002. They
found that LST possessed a slightly stronger negative correlation with the unmixed
vegetation fraction than with NDVI for all land cover types across the spatial resolution
from 30 to 960 m. Correlations reached their strongest at 120 m resolution. Fractal
analysis of image texture further showed that the complexity of these images increased
initially with pixel aggregation and peaked around 120 m, but decreased with further
aggregation. It was, therefore, suggested that the operational scale for examining the
relationship between LST and NDVI or vegetation fraction was around 120 m.
In another study, Liu and Weng (2009) examined the scaling effect between LST
and LULC in Indianapolis and found that the optimal spatial resolution for assessing
this unique relationship was 90 m. Four Terra ASTER images were used to derive
LULC maps and LST patterns in four seasons. Each LULC and LST image was
resampled to eight aggregation levels: 15, 30, 60, 90, 120, 250, 500, and 1000 m. The
scaling-up effect on the spatial and ecological characteristics of landscape patterns
and LSTs were examined by the use of landscape metrics. Optimal spatial scale was
determined on the basis of the minimum distance in the landscape metric spaces. Their
results showed that the patch percentages of LULC and LST patches were not strongly
affected by the scaling-up process. The patch densities and landscape shape indices
and LST patches kept decreasing across the scales without distinct seasonal differences. Ninety meters was found to be the optimal spatial resolution for assessing the
landscape-level relationship between LULC and LST patterns.
4.5 SCALE DEPENDENCY OF URBAN PHENOMENA
The majority of urban phenomena are scale dependent, meaning that the urban
patterns change with scale of observation. In reality, very few geographical phenomena are scale independent, in which the patterns do not change across scales (Cao
and Lam, 1997). Geographic studies are frequently conducted on the basis of areal
units such as states, counties, census tracts, block groups, and blocks. Across-scale
analyses with such areal units may produce scale-related problems, most notably, the
modifiable areal unit problem (MAUP). Fotheringham and Wong (1991) defined
MAUP as “the sensitivity of analytical results to the definition of units for which data
are collected.” Specifically, the results may vary with the aggregation level (the “scale
effect”) and with the aggregation schemes (the “zoning effect”). A researcher who
attempts to extrapolate the result of a study at one scale to other scales may meet three
kinds of erroneous inferences: individualistic fallacy, cross-level fallacy, and ecological fallacy (Alker, 1969). There is no ideal solution to solve MAUP; it is, however,
mainly examined by conducting statistical correlation and regression analysis
68
ON THE ISSUE OF SCALE IN URBAN REMOTE SENSING
analysis yielded around 120 m in Indianapolis, Indiana. Remote sensing of urban heat
islands has traditionally used the normalized difference vegetation index (NDVI) as the
indicator of vegetation abundance to estimate the LST–vegetation relationship. Weng
et al. (2004) investigated the applicability of vegetation fraction derived from LSMA as an
alternative indicator of vegetation abundance. An experiment was conducted with a
Landsat ETM+ image of Indianapolis City, Indiana, acquired on June 22, 2002. They
found that LST possessed a slightly stronger negative correlation with the unmixed
vegetation fraction than with NDVI for all land cover types across the spatial resolution
from 30 to 960 m. Correlations reached their strongest at 120 m resolution. Fractal
analysis of image texture further showed that the complexity of these images increased
initially with pixel aggregation and peaked around 120 m, but decreased with further
aggregation. It was, therefore, suggested that the operational scale for examining the
relationship between LST and NDVI or vegetation fraction was around 120 m.
In another study, Liu and Weng (2009) examined the scaling effect between LST
and LULC in Indianapolis and found that the optimal spatial resolution for assessing
this unique relationship was 90 m. Four Terra ASTER images were used to derive
LULC maps and LST patterns in four seasons. Each LULC and LST image was
resampled to eight aggregation levels: 15, 30, 60, 90, 120, 250, 500, and 1000 m. The
scaling-up effect on the spatial and ecological characteristics of landscape patterns
and LSTs were examined by the use of landscape metrics. Optimal spatial scale was
determined on the basis of the minimum distance in the landscape metric spaces. Their
results showed that the patch percentages of LULC and LST patches were not strongly
affected by the scaling-up process. The patch densities and landscape shape indices
and LST patches kept decreasing across the scales without distinct seasonal differences. Ninety meters was found to be the optimal spatial resolution for assessing the
landscape-level relationship between LULC and LST patterns.
4.5 SCALE DEPENDENCY OF URBAN PHENOMENA
The majority of urban phenomena are scale dependent, meaning that the urban
patterns change with scale of observation. In reality, very few geographical phenomena are scale independent, in which the patterns do not change across scales (Cao
and Lam, 1997). Geographic studies are frequently conducted on the basis of areal
units such as states, counties, census tracts, block groups, and blocks. Across-scale
analyses with such areal units may produce scale-related problems, most notably, the
modifiable areal unit problem (MAUP). Fotheringham and Wong (1991) defined
MAUP as “the sensitivity of analytical results to the definition of units for which data
are collected.” Specifically, the results may vary with the aggregation level (the “scale
effect”) and with the aggregation schemes (the “zoning effect”). A researcher who
attempts to extrapolate the result of a study at one scale to other scales may meet three
kinds of erroneous inferences: individualistic fallacy, cross-level fallacy, and ecological fallacy (Alker, 1969). There is no ideal solution to solve MAUP; it is, however,
mainly examined by conducting statistical correlation and regression analysis
68
ON THE ISSUE OF SCALE IN URBAN REMOTE SENSING