data of the study area with nearest neighbor 
resampling at 25-metre resolution, giving a RMSE 
of planimetry of + 0.33 pixel. The census block 
group boundaries were also converted into UTM 
coordinates and raster format. 
The biophysical data extracted from Landsat TM 
data were: (1) land use/cover, particularly the 
percentage of urban use, (2) normalized difference 
vegetation index (NDVI), and (3) apparent surface 
temperatures. 
The land use/cover was extracted using a 
supervised approach with the maximum likelihood 
classifier. Eight land use/cover types were 
extracted: (1) water, (2) forest, (3) commercial 
lindustrial, (4) transportation, (5) agriculture in 
pasture, (6) agriculture in crop, (7) low-density 
residential, and (8) high-density residential. The 
misclassifications were later manually corrected 
area by area with the aid of a parallelpiped 
classifier and ground truth from aerial 
photography, thus achieving an overall accuracy of 
99.1 per cent. The "commercial /industrial" cover 
class constituted "urban use". 
NDVI, which measures vegetation amount, was 
extracted using TM band 4 (0.76-0.90 uim) and TM 
band 3 (0.63-0.69 um) as follows: 
NDVI = (TM4-TM3)/(TM4+TM3) 
The value varies from -1 to +1 with an increase in 
vegetation amount. The raw NDVI exhibited a range 
of values from -0.4778761 to 0.9148936. The NDVI 
image very clearly delineated the built-up areas 
from the vegetated areas. 
The apparent surface temperatures were extracted 
from the thermal infrared TM band 6 (10.3-12.5 um) 
using a quadratic regression model in the form: 
T(K) = 209.831 * 0.834 DC - 0.00133 DC? 
where DC is the digital counts between 0 and 255, 
and T(K) is absolute temperature in Kelvin. The 
resulting image of surface temperatures revealed 
a range from 21?C to 38 C, with high surface 
temperatures clearly delineating the built-up areas. 
From the population census data, the following 
socio-economic variables were extracted or 
derived: (1) population density, (2) per capita 
income, (3) median home value, and (4) per cent of 
college graduates, by block groups. Each one of 
these variables produced a map which was then 
converted into raster format for analysis by the 
IDRISI software. 
3. INTEGRATION APPROACHES 
3.1 Principal Components Analysis (PCA) 
PCA was applied to the seven layers of bio. 
physical and socio-economic image data describeq 
above all aggregated at the census block group 
level. The correlation matrix of these variables 
indicated particularly strong negative correlation 
between (1) NDVI and surface temperatures (r=. 
0.88 at 0.0001 rejection level), and (2) NDVI and per 
cent urban (r=-0.85 at 0.0001 rejection level) (Table 
1). The implication is that NDVI by itself is a 
versatile environmental quality variable. NDVI also 
showed moderately positive correlation with per 
capita income, median home value, and per cent 
college graduates, as well as negative correlation 
with population density. The first of the three 
principal components extracted explained about 54 
per cent of the variance (Table 2). This component 
showed strong positive loadings on four variables, 
namely, NDVI, per capita income, median home 
value, and per capita income, and per cent of 
college graduates, but also strong negative 
loadings on three variables, namely, per cent of 
urban use, surface temperatures, and population 
density. The second principal component which 
explained 21 per cent of the variance, showed only 
one weak negative loading on NDVI, but positive 
loadings on the remaining six variables (Table 2). 
A plot of the component pattern indicates two 
clusters of variables: (1) population density, per 
cent of urban use, and surface temperatures verus 
(2) per capita income, median home value, per cent 
of college graduates, and NDVI (Figure 1). These 
two clusters of variables suggest that the two 
principal components represent two planes of data 
similar in nature to those defined by the Tasseled 
Cap features of Greenness and Brightness, which 
are related to the plane of vegetation and the plane 
of soils respectively (Crist and Cicone, 1984). The 
first cluster of variables is "environmental" in 
nature while the second cluster is "socio- 
economic" in nature. It is interesting that NDVI falls 
in the "socio-economic" cluster. These first plane 
can be labelled as "Greenness" because of its 
strong positive correlation with NDVI and the 
second plane can be called "Economic Well Being" 
because of its strong correlation with housing and 
education variables. Clearly, PCA has succeeded 
to integrate the biophysical environment with the 
socio-economic characteristics of the population. 
NDVI as an effective measure of "Greenness" can 
be regarded as an excellent quality of life indicator. 
A map of the first principal component scores by 
block group for the study area is therefore an 
excellent map of the quality of life of the population 
living in each block group (Figure 2). The map 
revealed high life quality areas in the southeast 
and southwest edges of the county and the low life 
quality areas in the middle of the county near the 
432 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996