Image Processing Reference
In-Depth Information
thermal properties even at the relatively coarse 1 km scale. Relatively strong
negative correlations of daytime surface temperature with urban IJI for both Asphalt
and Built classes are also observed. This could reflect the high degree of association
of these two classes with vegetated regions (hence cooler surface temperatures).
Relatively strong positive correlations of night surface temperature with nonurban
CA and MPS are noted for the Soil and Bedrock class. This result is problematic as
the Phoenix peri-urban area typically has cooler nighttime temperatures than the urban
core and we would expect to see a negative correlation with these landscape metrics
(Brazel et al.
2000
; Hawkins et al.
2004
). This result could be an artifact of the loca-
tion of the nonurban grid cells in that they may include extensive regions of dark
bedrock and soil. Further analysis of the entire Phoenix metropolitan region and
comparison with other urban areas is needed to investigate this further. Negative
correlation of night surface temperature for the Agriculture class in both urban and
nonurban regions is obtained for the CA, ED, and IJI metrics. The distribution and
configuration of agricultural lands in urban and peri-urban regions influences urban
heat island and oases effects as discussed by Voogt and Oke (
2003
). The combined use
of ASTER-derived landscape metrics and MODIS temperature data may facilitate
further investigation of urban heat island/oasis effects and regional climatology. Initial
modifications of land cover input to a regional climate model for Phoenix have already
been accomplished using Landsat TM data (Grossman-Clarke et al.
2005
; Zehnder
2002
), and further improvements are possible using ASTER and MODIS data.
The strongest positive correlations are noted with NDVI and the nonurban
Agriculture class for the CA, ED, and IJI metrics. Strong correlations are also observed
with the urban and nonurban Built class for CA and IJI, and urban ED for the
Undifferentiated Vegetation class. All of these aggregate land cover types include
significant vegetation, and the correlations with the metrics reflect their complex dis-
tribution across the Phoenix urban and peri-urban landscape (particularly with regard
to the Built and Undifferentiated Vegetation classes (Figs.
12.2
-
12.5
). Relatively
strong negative correlations of NDVI with the urban and nonurban Soil and Bedrock
class in CA and MPS are the result of the generally low vegetation cover in both
Phoenix urban mountain parks and the surrounding Sonoran desert (Whitford
2002
).
Greenhill et al. (
2003
) calculate weighted mean patch size and lacunarity (a mea-
sure of the distribution of patches of pixels in a scene) for NDVI values obtained
from the IKONOS sensor for a suburban area of southwest London. Their work
demonstrates the usefulness of the landscape metric approach for urban planning
applications and urban ecological research using high spatial resolution data. The
availability of high spatial resolution data from sensors such as IKONOS and
Quickbird is limited however both temporally and spatially compared to ASTER and
MODIS (and similar) data. This is especially important with regard to comparative
studies that include numerous urban centers and their surrounding regions.
The results presented here are of course scale-dependant (refer to Chapters 4
and 5), and could be expected to vary if different spatial scales or land cover classifi-
cation schemes were used (Woodcock and Strahler
1987
; Wu et al.
2000
; Small
2003
).
The variance due to spatial scale was investigated in an allied study by Stefanov and
Netzband (
2005
) using the same classified ASTER and MODIS data described
here. Their study focused on gridded landscape metric analysis using 250 m/pixel,
