Integrating remote sensing and GIS for environmental justice research - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 219

minorities were clustered nearby hazardous facilities, but not
in immediate proximity (i.e. within 500 m), as such facilities
tended to be sited on sparsely populated industrial land. Such
information would be masked by the use of spatially coarser
demographic data.
Because most remotely sensed imagery used in environmental
justice analyses is used to capture environmental hazards and
amenities at a scale that canbematched todemographic character,
fine and medium resolution image data are typically used. Thus,
remotely sensed imagery used in environmental justice analyses
are typically at a finer resolution than population data encoded
in census spatial units. When remotely sensed imagery is used for
quantifying characteristics of environmental hazards or amenity,
say, for vegetation concentration, the remotely sensed image data
may be aggregated to the level of the census unit to support
statistical analysis. For example, the mean NDVI value of all the
pixels contained within each Census unit may be calculated and
attached as an attribute of that census unit.

NDVI value in the data set was assigned a value of ''1,'' the next
highest a value of ''2,'' and so on. Pixels with equal NDVI values
are tied, and the highest rank is equal to the number of pixels
in the image (not including pixels with no data values). Thus, a
low NDVI rank indicates a low NDVI value - and thus indicates
a relatively low concentration of healthy green vegetation. The
mean NDVI rank was then computed for each tract and assigned
as an attribute of that tract. Descriptive statistics for each of the
variables used in the analysis are provided in Table 16.1.
Correlation and ordinary least squares regression were
employed to identify relationships among the explanatory
variables representing socioeconomic status and the dependent
variable representing vegetation character (mean NDVI rank).
The well-known Moran's I statistic (Lloyd, 2007) was used
to test for spatial autocorrelation in the residuals. Where
spatial dependency in the regression residuals is found, spatial
econometric modeling is employed to account for spatial effects.
The Pearson correlation between each of the explanatory vari-
ables and the meanNDVI rank is presented in Table 16.2.Clearly,
higher NDVI is associatedwith the absence of bothAfricanAmer-
icans and Hispanics, as well as socioeconomic advantage, i.e. low
poverty rate and high educational attainment. Not surprisingly,
higher vegetation concentration is also associated with sparser
population.
A four-class, quantile-classified choroplethmap of each of the
variables are presented in Fig. 16.1. The lowest values of mean
NDVI rank (Fig. 16.1, top left) are clearly concentrated in the city
of Philadelphia and the industrial and urban areas which lie to its
south along the Delaware River. These areas are colored orange in
the southeasternpart of themap. The highest values of NDVI lie at
the exurban and rural areas of Chester and Bucks Counties in the
southwest and northwest regions of the study area, respectively.
A high percentage of African Americans is strongly concentrated

16.5 Case study: vegetation
and socioeconomic
character in Philadelphia,
Pennsylvania

As an example of how remote sensing is used in environmental
justice research, an environmental equity study is demonstrated
for the Philadelphia, Pennsylvania metropolitan area. Here,
remotely sensed vegetation concentrationdata are integratedwith
data characterizing race, poverty, educational attainment, and
population density to assess the relationships of socioeconomic
status with the amenities associated with higher concentrations
of vegetation. The study area comprises the five-county Philadel-
phia metropolitan area in southeast Pennsylvania, including
the city/county of Philadelphia,aswellasMontgomery,Bucks,
Delaware, and Chester Counties.
Vegetation data are derived from an orthorectified, cloud-
free, leaf-on, 30 meter resolution Landsat ETM

TABLE 16.1 Descriptive statistics of explanatory variables.

Mean Standard deviation

% African American 22.68 31.86

%Hispanic 4.36 9.68

image (path
14, row 32) dated 29 September 1999, acquired from the Landsat
GeoCover collection as distributed by the University of Mary-
land's Global Land Cover Facility (GLCF) web site. A NDVI
image was derived from this Landsat image. Socioeconomic
data were acquired from the 2000 US Census at the tract level,
including the following variables: percent of the total population
self-identifying as African American (percent African American),
percent of the total population self-identifying as Hispanic (per-
cent Hispanic), percent of the total population whose income
falls below the poverty line (of the population for whom poverty
status has been recorded) (poverty rate), percent of the popula-
tion at least 25 years of age who have a high school diploma or
equivalent (percent high school), and population density (total
population per square kilometer).
Both the NDVI and Census data were reprojected to the
Universal Transverse Mercator projection and coordinate system
for overlay. The NDVI image was then clipped using the tract
data so that pixels outside the study area boundary were assigned
a value of 'no data.' The NDVI data were then converted to ranks,
as opposed to NDVI values, where the pixel with the very lowest

+

% High School 81.04 13.36

Poverty Rate 12.32 13.33

Population Density 3428 3753

Mean Rank NDVI 76 019 46 639

TABLE 16.2 Correlation of explanatory variables with mean NDVI
rank.

Pearson's r

% African American −0.45***

%Hispanic −0.28***

% High School 0.65***

Poverty Rate −0.61**

Population Density − 0.65***

*** p < 0 . 005.