Evaluating environmental influences of zoning in urban ecosystems
with remote sensing
Jeffrey S. Wilsona,*, Michaun Clayb,1
, Emily Martina,2
, Denise Stuckeyc,3
, Kim Vedder-Rischd,4
a
Department of Geography, Indiana University-Purdue University, 425 University Boulevard, Indianapolis, IN 46202-5143, USA
b
City of Indianapolis, USA
c
The Schneider Corporation, USA
d
Indiana Department of Environmental Management, USA
Received 26 March 2002; received in revised form 11 July 2002; accepted 2 March 2003
Abstract
The influence of zoning on Normalized Difference Vegetation Index (NDVI) and radiant surface temperature (Ts) measurements is
investigated in the City of Indianapolis, IN, USA using data collected by the Enhanced Thematic Mapper Plus (ETM+) remote sensing
system. Analysis of variance indicates statistically significant differences in mean Ts and NDVI values associated with different types of
zoning. Multiple comparisons of mean Ts and NDVI values associated with specific pairings of individual zoning categories are also shown
to be significantly different. An inverse relationship between Ts and NDVI was observed across the city as a whole and within all but one
zoning category. A range of environmental influences on sensible heat flux and urban vegetation was detected both within and between
individual zoning categories. Examples for implementing these findings in urban planning applications to find examples of high and low
impact development are demonstrated.
D 2003 Elsevier Science Inc. All rights reserved.
Keywords: Zoning; Urban ecosystems; Remote sensing
1. Introduction
Zoning is one of several tools urban planners use to
control physical characteristics of developing landscapes by
imposing restrictions on variables such as maximum building
height and density, extent of impervious surface and
open space, and land use types and activities. These variables,
in turn, influence environmental processes of atmosphere/surface
energy exchange (Grimmond & Oke, 1995;
Quattrochi & Ridd, 1994), surface and subsurface hydrologic
systems (Grimmond & Oke, 1991; Hammer, 1972),
and micro- to mesoscale weather and climate regimes
(Changnon, 1992; Oke, 1987; Roth, Oke, & Emery,
1989). The intrinsic relationships between zoning and the
physical structure of urban environments suggest that mechanisms
should be in place through which planners can
evaluate the environmental consequences of existing zoning
ordinances and improve the scientific basis of future decision
making in order to mitigate negative effects of development
in urban ecosystems. Remote sensing science and
technology offers the potential to contribute to this informed
decision-making process by providing synoptic data collection
capabilities and analysis techniques that generate pertinent
information about environmental implications of
planning decisions.
The purpose of the research presented in this paper is to
contribute to further validation of the applicability of
relatively low cost, moderate spatial resolution satellite
imagery in urban planning applications and provide some
examples of how these data can be used by planners to aid
in evaluating environmental impacts of zoning. First, the
question of whether or not physical manifestations of
0034-4257/03/$ - see front matter D 2003 Elsevier Science Inc. All rights reserved.
doi:10.1016/S0034-4257(03)00084-1
* Corresponding author. Fax: +1-317-274-2347.
E-mail addresses: jeswilso@iupui.edu (J.S. Wilson),
c1238@indygov.org (M. Clay), eskempf@iupui.edu (E. Martin),
dstuckey@theschneidercorp.com (D. Stuckey), kvedder@dem.state.in.us
(K. Vedder-Risch).
1
Fax: +1-317-327-3192.
2
Fax: +1-317-274-2347.
3
Fax: +1-317-826-7200.
4
Fax: +1-317-232-8714.
www.elsevier.com/locate/rse
Remote Sensing of Environment 86 (2003) 303–321
different types of zoning can be differentiated using two
objective measurements collected by a satellite remote
sensing system is examined. The two measurements examined
are radiant surface temperature (Ts) and the Normalized
Difference Vegetation Index (NDVI). Second, an analysis of
relationships between Ts and NDVI for the entire city and
within different zoning categories is presented. Finally, it is
shown that zoning results in a range of environmental
influences on sensible heat flux and urban vegetation (both
within and between different zoning categories). Examples
of how these results can be used by urban planners to find
instances of high and low impact implementations of different
types of zoning are then presented. In the current
research, zoning instances with high environmental impact
are assumed to be those with reduced vegetation biomass
(low NDVI) and/or greater sensible heat flux (high Ts) as
compared to other areas that fall within the same zoning
category.
It is important to point out that zoning ordinances, while
regulatory in nature, are only a guide for development of a
landscape. An individual zoning polygon, for example,
may contain nonconforming land uses and variances from
the characteristics prescribed by a specific zoning ordinance,
if it has been developed at all. While there may be a
strong correlation between zoning, land use, and land
cover, that relationship is governed, in part, by the specific
ordinances for that zoning category and how rigorously the
ordinances are applied. The intention of this paper is not to
reiterate the well-established fact that land use and land
cover can have a significant impact on measurements
collected by a satellite remote sensing system, but rather
to show how remote sensing technology can be used by
planners to obtain objective measurements of environmental
consequences of existing zoning policies. By examining
the spatial variation of Ts and NDVI as a function of
zoning, this research demonstrates how planners can construct
a comparative context in which to evaluate the
implications of existing zoning policies and potentially
shape future strategies that lessen the negative impacts of
development.
Consider, as a hypothetical example, two different portions
of landscape that have both been developed as highdensity
residential and assume that both of these areas
conform to existing zoning ordinances. A planner concerned
with environmental consequences of development might ask
the questions, ‘‘does one of these areas have a less detrimental
impact on the physical environment than the
other?’’, ‘‘If so, what are the reasons for these differences
and can we use this information to modify existing policies
to decrease the environmental consequences of future development?’’.
The research presented in this paper seeks to
provide a quantitative basis for developing such a comparative
context. The City of Indianapolis, IN, USA is used as
a case study, but the data and techniques presented should
be available and applicable in many moderate- to large-size
cities in temperate and tropical climates.
2. Background
Urban planning departments commonly use remote sensing
imagery in the development of geographic information
system (GIS) databases. For example, vector digitizing from
aerial photography has become a common method used in
the creation and updating of maps depicting the spatial
distribution of basic planning variables such as transportation
networks, cadastre, buildings and pavement, land use,
and utility infrastructures. The current and coming generations
of high spatial resolution earth observation satellites,
such as IKONOS and QuickBird, offer the potential for
application of spaceborne remote sensing data to some
planning applications that were previously limited to aerial
photography (Barr & Barnsley, 2000; Jensen & Cowen,
1999). However, while commonly used in certain capacities,
the full potential of remote sensing data, particularly quantitative
analysis of electromagnetic energy measurements,
has not been widely exploited by the planning community
(Nichol, 1996; Ryznar & Wagner, 2001) despite the fact that
such applications have been suggested at least since the
1960s and 1970s (e.g., Carlson, Augustine, & Boland, 1977;
Estes, 1966; Pease, Lewin, & Outcalt, 1976; Wellar & Estes,
1968) and have the potential to improve the quality of
planning processes.
As the world’s cities continue to grow in both population
and physical size, greater consideration of the manner in
which rural lands are developed for urban and suburban land
use will become progressively more important (Vitousek,
Mooney, Lubchenco, & Melillo, 1997). Planners are
increasingly pressured with the challenge of balancing the
demands for urban and suburban growth while mitigating
the negative environmental impacts of development.
Removal of rural land cover types such as soil, water, and
vegetation and their replacement with common urban materials
such as asphalt, concrete, and metal have significant
environmental implications including reduction in evapotranspiration,
promotion of more rapid surface runoff,
increased storage and transfer of sensible heat, and reduction
of air and water quality (Carlson, Dodd, Benjamin, &
Cooper, 1981; Goward, 1981; Owen, Carlson, & Gillies,
1998). These changes, in turn, can have negative effects on
landscape aesthetics, energy efficiency, human health, and
quality of living in urban environments (McPherson et al.,
1997; Rosenfeld et al., 1995).
Among the most widely studied consequences of converting
rural land cover to urban and suburban development
is the urban heat island (UHI), which is manifested by
higher surface and air temperatures within urban environments
as compared to surrounding rural areas. Approaches
to measuring the UHI include sampling of air and surface
kinetic temperature at fixed locations, mobile sampling of
kinetic temperature from ground and airborne vehicles, and
remote measurement of radiant temperature from airborne
and spaceborne platforms (Henry, Dicks, Wetterquist, &
Roguski, 1989; Stone & Rodgers, 2001). Compromises are
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321304
associated with each of these methods, as they measure
different aspects of the thermal characteristics of cities
(Henry et al., 1989), but remote sensing offers the most
synoptic capabilities, at least for measurement of radiant
temperature, because of the ability to collect many samples
over a broad area almost instantaneously (Saaroni, Ben-Dor,
Bitan, & Potcher, 2000). While remote sensing is not a
panacea for analysis of urban thermal environments since it
only directly measures radiant temperature (Ts) from surfaces
exposed to the sensors view (Voogt & Oke, 1997, 1998),
it can still be a useful tool for gaining an overview of
thermal variations within an urban ecosystem. Direct measurements
of Ts can also be used in the derivation of hybrid
variables such as fractional vegetation cover, impervious
surface area, and kinetic surface temperature (Carlson &
Arthur, 2000; Carlson, Gillies, & Perry, 1994; Henry et al.,
1989; Ji & Jensen, 1999; Nemain, Pierce, Running, &
Goward, 1993). Measurements of Ts from the Enhanced
Thematic Mapper Plus (ETM+) sensor onboard the recently
launched (1999) Landsat 7 satellite are used in the current
research because this instrument collects the highest spatial
resolution thermal measurements currently available for
virtually all of earth’s urban areas on a temporally consistent
basis.
Vegetation is another important component of the urban
ecosystem that has been the subject of much basic and
applied research. Urban vegetation influences the physical
environment of cities through selective absorption and
reflection of incident radiation and regulation of latent and
sensible heat exchange (Carlson et al., 1994; Gillies, Cui,
Carlson, Kustas, & Humes, 1997; Goward, Cruikhanks, &
Hopes, 1985). Commonly cited benefits of urban vegetation
include aesthetic enhancement, reduction of air and water
pollution (Taha, 1997), increased latent heat exchange
(Goward, 1981; Heisler, 1986), increased energy efficiency
(Akbari & Taha, 1992; Huang, Akari, Taha, & Rosenfeld,
1987), and contribution to the psychological well-being of
urban dwellers (Ulrich, 1986). As in the case of the urban
heat island, remote sensing can be a useful tool for studying
urban vegetation. In addition to providing measurements of
radiant surface temperature, remote sensing instruments
collect measurements of reflected energy in the red and
near infrared (NIR) portions of the electromagnetic spectrum
that can be used to quantify the extent and changing
conditions of urban vegetation.
3. Data
3.1. Satellite imagery
A Landsat 7 ETM+ image acquired on June 6, 2000
under relatively clear atmospheric conditions served as the
primary data source for mapping of radiant surface temperature
and vegetation in the Indianapolis urban ecosystem
and as a basis for analysis of the relationship between these
variables and zoning categories. The ETM+ sensor is an
imaging radiometer that collects measurements of reflected
and emitted energy from earth surface features in eight
regions of the electromagnetic spectrum ranging from visible
to thermal infrared at an 8-bit (0–255) level of quantization.
The spatial and spectral characteristics of these eight
channels or bands are summarized in Table 1.
The Landsat 7 satellite orbits the earth at an altitude of
approximately 700 km following a sun-synchronous, nearpolar
path at an inclination of 98j from the equator. This
orbital pattern provides opportunity to collect imagery of
urban environments anywhere between 81j north and south
latitudes. The temporal resolution of the sensor (the time it
takes to acquire repeat imagery of the same region) is 16
days. On the descending (N to S) daytime portion of an
orbit, the satellite crosses the equator at approximately
10:00 AM local solar time, with increasingly later local
acquisition times north of the equator, and increasingly
earlier local acquisition times south of the equator. The
Indianapolis image used in this research (centered at about
39.75jN, 86.16jW) was acquired at approximately 11:15
AM local time (16:15 GMT).
Since urban air temperatures usually reach their maximum
sometime after solar noon and maximum thermal
contrast in air temperatures between urban and rural air
temperatures often occurs at night (Karl, Diaz, & Kukla,
1988; Oke, 1981), the acquisition time of ETM+ imagery
may not be considered the most ideal for analysis of the
UHI and related phenomena. However, if one takes the
urban temperature anomaly as the difference between surface
radiant temperatures in the city vs. that in the countryside,
the daytime anomaly is typically greater, at least
during the summer time. In the current research, it is
demonstrated that measurements of both reflective shortwave
radiation (in the form of NDVI) and emitted thermal
radiation (Ts) collected by the ETM+ sensor provide sufficient
contrast to differentiate between not only urban vs.
rural, but between different zoning categories within the
urban environment.
Landsat images are spatially indexed by a system of
paths and rows known as the World Referencing System
(WRS). Path numbers refer to the ground track of the
Table 1
Characteristics of Enhanced Thematic Mapper Plus (ETM+) bands
Band Spatial
resolution
(m)
Lower
limit
(Am)
Upper
limit
(Am)
Bandwidth
(nm)
Bits per
pixel
1 28.50 0.45 0.52 70 8
2 28.50 0.53 0.61 80 8
3 28.50 0.63 0.69 60 8
4 28.50 0.78 0.90 120 8
5 28.50 1.55 1.75 200 8
6 57.00 10.40 12.50 2100 8
7 28.50 2.09 2.35 260 8
8 14.25 0.52 0.90 380 8
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321 305
satellite along each orbit, and row numbers refer to the
latitudinal center of an image along a given path. The WRS
coordinate for the Indianapolis image used in this research is
path 021, row 032. The 185-km-wide continuous strip of
imagery collected along a given path is divided into 170-km
segments prior to distribution, resulting in total areal coverage
of approximately 31,500 km2
per scene. A GIS polygon
was used to create a subset of the June 6, 2000 ETM+ image
corresponding to the border of Marion County, IN, which
shares a common boundary with the City of Indianapolis.
The total area of the image analyzed in the research is
approximately 1200 km2
.
3.2. Radiometric enhancement
ETM+ imagery processed by the US Geological Survey’s
Earth Resource Observation Systems (EROS) Data Center
to correct for radiometric and geometric distortions are
referred to as a Level 1G product. Digital numbers (DNs)
in each band of the Level 1G ETM+ imagery used in this
research were converted to physical measurements of at
sensor radiance (LE) using a formula that accounts for the
transformation function used to convert the analog signal
received at the sensor to DNs stored in the resulting image
pixels (Eq. (1)):
LE ¼ gain  DN þ offset ð1Þ
where LE = at sensor radiance, gain = slope of the radiance/
DN conversion function, DN = digital number of a given
pixel, and offset = intercept of the radiance/DN conversion
function (Landsat Project Science Office, 2002). Gain and
offset values are supplied in metadata accompanying each
ETM+ image, and the DN to radiance formula (as with all
formulas presented in this paper) can easily be implemented
using map algebra functions of a raster-based GIS if
specialized image processing software was unavailable in
an urban planning department. ERDAS Imagine 8.5k was
used as the primary image processing and GIS tool in the
current project.
Radiance values were transformed to unitless planetary
reflectance for the reflective ETM+ bands (channels 1–5
and 7) and radiant temperature for the thermal band (channel
6). The Landsat Project Science Office (2002) suggests that
a reasonable transformation of radiance to reflectance can be
achieved for images acquired under relatively clear atmospheric
conditions by accounting for the combined reflectance
of the surface and atmosphere (Eq. (2)):
qp ¼ ðpLEd2
Þ=ðESUNEcoshsÞ ð2Þ
where qp = unitless planetary reflectance, LE = spectral radiance
at sensor aperture, d = earth–sun distance in astronomical
units, ESUNE = mean solar exoatmospheric
irradiance, and hs = solar zenith angle in degrees. Radiance
values from the ETM+ thermal band were transformed to
radiant surface temperature values using thermal calibration
constants supplied by the Landsat Project Science Office
(2002) (Eq. (3)):
Ts ¼ K2=ðlnðK1=LE þ 1ÞÞ ð3Þ
where Ts = radiant surface temperature (K), K2 = calibration
constant 2 (1282.71), K1 = calibration constant 1 (666.09),
and LE = spectral radiance of thermal band pixels.
Reflectance values from the red (qred) and NIR (qnir)
ETM+ channels were used to compute Normalized Difference
Vegetation Index (NDVI) values for the Indianapolis
image subset using the formula NDVI=(qnir À qred)/
(qnir + qred). The NDVI is a quantitative index useful for
monitoring spatial and temporal vegetation dynamics (e.g.,
Fung & Siu, 2000; Schmidt & Gitelson, 2000; Verstrate &
Pinty, 1996). NDVI measures vegetation vigor as a function
of spectral reflectance attributed to chloroplast and mesophyll
cells in plants and has been shown to have significant
relationships with a variety of biophysical properties of
vegetation including leaf area index (LAI) (Qi, Moran, Cabot,
& Dedieu, 1995), fraction of vegetation ground cover and
green biomass (Gamon et al., 1995; Gamon, Field, Roberts,
Ustin, & Valentini, 1993), and photosynthetic activity (Reed
et al., 1994).
3.3. Geometric correction
NDVI and Ts images of the Indianapolis study region
were georeferenced to the Universal Transverse Mercator
coordinate system, Zone 16 North in order to facilitate
integration of these data with GIS data of the study region.
Georeferencing was accomplished using 26 ground control
points drawn from 1:24,000 topographic maps, a nearest
neighbor resampling algorithm, with a maximum root mean
square error (RMSE) of 0.5 pixels. Spatial resolution of the
reflective bands was resampled to 30 m and spatial resolution
of the thermal band to 60 m in the georeferencing
process. Spatial transformations were implemented in such a
way that extent of a pixel in the 60-m resolution thermal
channel matched exactly the extent of four corresponding
30-m pixels in the reflective bands. This later point was
important because subsequent analysis of relationships
between Ts and NDVI was achieved by computing the
mean NDVI of four pixels and pairing that measurement
with the single spatially corresponding value of Ts.
3.4. Geographic information system data
Additional spatial data sources used in the analysis of the
Indianapolis urban ecosystem included a GIS polygon layer
depicting the spatial distribution of zoning categories within
the city, a polygon layer depicting the digitized footprints
(plan view) of all structures greater than 10 m2
, and a
polygon coverage which delineated the extent of roads.
The building footprints and roads GIS layers were created
by the City of Indianapolis using on-screen digitizing from
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321306
high-resolution ( < 0.5 m) aerial photography. These data are
employed later in Section 4 of this paper to show how
density of development can be integrated with Ts and NDVI
to find comparative examples of high and low impact zoning.
The zoning GIS layer was obtained from the Indianapolis
Mapping and Geographic Information System (IMAGIS)
database and reflected zoning ordinances in place at the time
the ETM+ image was acquired. The zoning layer contained
6246 polygons categorized into 77 different zoning categories.
Zoning categories were aggregated to two different
levels of thematic detail: 13 general zoning categories and
16 detailed categories within the ‘‘dwellings’’ (residential)
zoning category (Table 2a and b). The ‘‘Dwelling Agricultural’’
(DA) zoning category was not included in our
analysis of the detailed dwellings categorizations because
of the heterogeneity of land cover types that occurred within
these comparatively large polygons; this category was
instead treated as an individual class in the general zoning
categorization. Spatial distributions of these zoning categories
are depicted in Fig. 1a and b.
Table 2
(a) Residential zoning categories
Residential
zoning
Typical density
(units/acre)
Primary use Minimum
open space
Typical
lot size
Comprehensive
plan classification
Number of
polygons
% Total
city area
DA* >0.5 agriculture and single-family 85% 3 acres very low density 699 21.92
DP varies planned unit development varies varies varies 89 3.53
DS 0.5 suburban single-family 85% 1 acre very low density 67 2.22
D1 0.9 suburban single-family 80% 24,000 ft2
very low density 104 2.16
D2 1.9 suburban single-family 75% 15,000 ft2
very low density 226 7.16
D3 2.6 low or medium intensity single-family 70% 10,000 ft2
low density 296 8.53
D4 4.2 low or medium intensity single-family 65% 7200 ft2
low density 195 4.98
D5 4.5 medium intensity single-family 65% 5000 ft2
low and medium
density
231 8.83
D5II 5.0 medium intensity single- or
two-family
65% 3200 ft2
low and medium
density
17 0.13
D6 6–9 low intensity multifamily (OSR) 3.85 varies medium density 81 0.83
D6II 9–12 medium intensity multifamily (OSR) 2.65 varies medium density 118 1.47
D7 12–15 medium intensity multifamily (OSR) 2.10 varies medium density 153 1.40
D8 5–26 urban multi-use residential (OSR) 2.65 varies high density 112 0.78
D9 12–120 suburban high-rise apartment (OSR) 0.29–1.45 varies high density 29 0.15
D10 20–140 central and inner-city high-rise
apartment
(OSR) 0.27–1.18 varies high density 9 0.02
D11 6.0 mobile dwellings varies varies medium density 23 0.42
D12 5.0 low-density two-family 65% 9000 ft2
low density 13 0.04
(b) General zoning categories
General zoning Basic description Number of
polygons
% Total
city area
Airport public airports municipally owned or operated, including all necessary
navigation and flight operation facilities, and accessory uses
3 1.39
CBD central business district—core activities of all types with a
variety of related land uses
22 0.30
Commercial includes office-buffer, high-intensity office/apartment, neighborhood
commercial, thoroughfare service, and corridor commercial districts
2013 8.14
Dwellings variety of residential categories summarized in table above (note: DA
treated as separate general zoning class in subsequent analysis)
1763 64.57
Historic historic preservation district including a variety of land uses
(mostly residential)
1 0.01
Hospital major hospital complexes and campuses 29 0.48
Industrial variety of industrial uses including urban and suburban and light,
medium, and heavy industry
471 9.86
Interstate interstate highway 4 1.94
Park permits all sizes and ranges of public park land and facilities, includes
park peripheral areas assuring compatibility of adjacent land use
164 4.76
Regional regional attractions close to CBD including museums, visitor centers,
and Indianapolis Zoo
1 0.09
Special use wide variety of uses such as schools, utility infrastructure, cemeteries, libraries,
community centers, charitable organizations, golf courses, and penal institutions
1065 8.12
University variety of land uses typical of higher education institutions including classroom,
office, dormitory, facility maintenance, and parking structures
11 0.36
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321 307
Fig. 1. Map of (a) general zoning categories (b) and detailed residential zoning categories.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321308
4. Analysis
4.1. Visual interpretation of NDVI and Ts imagery
Fig. 2a and b shows the transformed ETM+ data
depicting the spatial distribution of NDVI and Ts in the
study region. When printed at appropriate scales, such
maps provide an informative visual depiction of the
spatial distribution of radiant surface temperature and
vegetation components within a city, potentially applicable
in planning and public education contexts (Fig. 2). In the
NDVI image of Indianapolis, the relatively large Eagle
Creek and Geist Reservoirs, located in the northwestern
and northeastern portion of the county, respectively,
appear in dark tones due to the absence of vegetation.
Conversely, forested areas around the reservoirs appear in
bright tones as a function of relatively high levels of green
biomass. The central business district (roughly in the
center of image) appears in dark tones, as do some of
the commercial and industrial features such as the Indianapolis
International Airport (WSW of the CBD near the
western edge of the county) and the Indianapolis Motor
Speedway (3 km WNW of CBD). Many of the agricultural
fields in the study region were only recently plowed
or planted by the June 6 image acquisition date so
significant crop cover did not have time to develop. These
agricultural areas appear in relatively dark gray tones with
square or rectangular geometric shapes, particularly in the
southeast and southwest corners of the study region.
Among the most prominent features in the Ts image of
Indianapolis are the bright tones (warmer radiant temperatures)
associated with the central portion of the city,
some of the major transportation arteries, and adjacent
commercial and industrial areas. Cooler radiant surface
temperatures (dark tones) are indicated for the reservoirs
and surrounding forested areas as well as the White River
and adjacent riparian vegetation, which run north to south
through the central portion of the city near the center of
the image.
4.2. Significance of differences in mean Ts and NDVI by
zoning categories
A zonal summary GIS operation was used to derive
measurements of central tendency and dispersion of Ts
and NDVI associated with different types of zoning at both
the general and detailed residential levels of thematic detail.
The purpose of creating these statistical summaries was to
Fig. 2. (a) Normalized difference vegetation index (NDVI) and (b) radiant surface temperature (Ts) images of study region. White line represents extent of
Marion County/City of Indianapolis.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321 309
provide a database that could be used to address the
following research questions.
 How do mean Ts and NDVI values vary between zoning
categories and are these variations statistically signifi-
cant?
 What is the general relationship between NDVI and Ts
values in the study region and how does this relationship
change within and between different zoning categories?
 How can mean Ts and NDVI values associated with
zoning polygons be used by urban planners to evaluate
environmental implications of past planning decisions
and improve the basis of future decision making?
Fig. 3a and b depicts mean Ts and NDVI values
associated with general zoning categories with error bars
representing F 1 standard deviation from the mean. As
might be expected, the parks zoning category exhibited the
highest mean NDVI value because of the predominance of
trees and other vegetated land cover. Greater vegetation
cover in the park zoning class suggests comparatively
higher rates of evapotranspiration and favoring of latent
over sensible heat exchange between surface and atmosphere
as compared to more developed locations. This
general observation is supported by comparing Fig. 3a
and b, where the parks category displays the lowest mean
Ts and highest mean NDVI values. The CBD zoning
category mirrors the response of parks, having the lowest
mean NDVI and highest mean Ts values of the general
zoning categories. This observation is indicative of the lack
of vegetation and greater sensible heat exchange in the
CBD, which forms one of the bases of the UHI effect (Oke,
1982).
Comparable graphs of mean Ts and NDVI values associated
with detailed residential zoning categories are provided
in Fig. 4a and b. Like the parks category in the general
classification, the dwelling suburban (DS) residential zoning
category exhibited the lowest mean Ts and highest mean
NDVI. The DS category has the lowest housing unit density
of the residential zoning categories exclusive to singlefamily
dwellings, and is intended for use ‘‘in areas conducive
to estate development. . . or where it is desirable to
permit only low density development such as adjacent to
flood plains, aquifers, and urban conservation areas’’ (Metropolitan
Development Commission, 1998). Higher mean Ts
values were associated with residential zoning categories
situated close to the CBD such as the D8 and D5 categories,
as well as the D7 category, which is closely associated with
commercial shopping centers, industrial areas, and primary
transportation arteries.
Fig. 2 (continued).
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321310
The applicability of moderate spatial resolution measurements
of Ts and NDVI in urban planning depends partly on
the ability to use these measurements as a basis for differentiating
among different types of land use and land cover
of interest to planners. While the error bars presented in
Figs. 3 and 4 indicate overlap which occurs between most of
the zoning categories for both Ts and NDVI, we were
interested in determining whether or not different zoning
ordinances have, on average, a significantly different effect
on these variables. One-way analysis of variance (ANOVA)
Fig. 3. Mean values of (a) Ts and (b) NDVI associated with general zoning categories.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321 311
tests were used to evaluate the statistical significance of
differences in mean Ts and NDVI values associated with
different zoning types at the two levels of thematic detail
(general zoning and detailed residential zoning). Results of
ANOVA tests in all four cases [(1) Ts of general zoning
categories, (2) NDVI of general zoning categories, (3) Ts of
residential zoning categories, and (4) NDVI of residential
zoning categories] indicated that the between groups varFig.
4. Mean values of (a) Ts and (b) NDVI associated with residential zoning categories.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321312
iance component warranted a rejection of the null hypothesis
that there is no significant difference between group
means, and acceptance of the alternative hypothesis that
mean radiant surface temperature and NDVI are significantly
different. The implications of these results suggest
that on average, the physical manifestations of different
types of zoning have significantly different influences on
radiant temperature and vegetation as measured by the
ETM+ sensor.
4.3. Multiple comparisons of means
While the ANOVA results indicate that at least some
mean Ts and NDVI values are significantly different among
the possible pairings of zoning categories, these results do
not provide information on which specific pairing(s) differ.
To gain further insight, a series of post-hoc multiple
comparison tests using Tamhane’s T2 were conducted to
evaluate the significance of differences in mean Ts and
NDVI measurements associated with each possible pairing
of zoning categories at both the general and detailed
residential levels. Tamhane’s T2 post-hoc tests are similar
to the Student’s t-test, but decrease the chance of Type I
error for multiple tests by adjusting a as a function of
sample size and the number of tests completed. These tests
are appropriate in situations where the variance and/or
sample size of groups is unequal (Hochberg & Tamhane,
1987). Table 3a and b shows results of post-hoc tests
associated with bivariate combinations of general zoning
categories for mean Ts and NDVI, respectively. Values in
bold italic type indicate cases where the mean difference
exceeded the critical value at a = 0.0001, indicating a
statistically significant difference in mean values associated
with a given pairing.
Of the 78 possible combinations of general zoning
categories, 66 pairings exhibited significantly different
mean values of Ts. Seventy-three pairings were significantly
different in terms of mean NDVI. For example,
mean Ts of the agriculture zoning class was significantly
different from all other general zoning classes, while
pairings such as CBD and historic, and commercial and
hospital were not significantly different. Similarly, mean
NDVI associated with the historic zoning was not significantly
different from industrial, regional, and interstate
zoning, while agriculture, CBD, and dwelling categories
were significantly different from all other general zoning
categories. The same multiple comparison procedures were
used to test for significant differences in mean Ts and
NDVI values associated with detailed residential zoning
categories (Table 4a and b). Of the 120 possible combinations
of detailed residential zoning classes, 83 pairings
displayed significantly different mean values of Ts.
Table 3
Results of post-hoc tests to evaluate significance of differences in (a) mean Ts and (b) mean NDVI values associated with bivariate combinations of general
zoning categories
Agriculture Airport CBD Commercial Dwellings Historic Hospital Industrial Interstate Park Regional Special use
(a) Tamhane’s T2 multiple comparison table of mean differences in Ts for general zoning categories
Airport À 1.34
CBD À 6.48 À 5.14
Commercial À 3.00 À 1.66 3.48
Dwellings À 1.27 0.08 5.21 1.74
Historic À 6.12 À 4.78 0.36 À 3.12 À 4.86
Hospital À 3.00 À 1.66 3.48 0.01 À 1.73 3.13
Industrial À 1.81 À 0.47 4.67 1.19 À 0.54 4.31 1.19
Interstate À 1.40 À 0.06 5.08 1.60 À 0.13 4.72 1.60 0.41
Park 1.37 2.71 7.85 4.37 2.63 7.49 4.36 3.17 2.77
Regional À 1.95 À 0.61 4.53 1.05 À 0.69 4.17 1.05 À 0.14 À 0.55 À 3.32
Special use À 0.37 0.97 6.11 2.63 0.90 5.75 2.63 1.44 1.03 À 1.73 1.58
University À 1.80 À 0.45 4.68 1.21 À 0.53 4.33 1.20 0.01 À 0.40 À 3.16 0.16 À 1.43
(b) Tamhane’s T2 multiple comparison table of mean differences in NDVI for general zoning categories
Airport 0.15
CBD 0.38 0.23
Commercial 0.18 0.03 À 0.20
Dwellings À 0.04 À 0.19 À 0.41 À 0.22
Historic 0.19 0.04 À 0.18 0.01 0.23
Hospital 0.08 À 0.07 À 0.30 À 0.10 0.12 À 0.11
Industrial 0.18 0.02 À 0.20 0.00 0.21 À 0.02 0.10
Interstate 0.16 0.01 À 0.22 À 0.02 0.20 À 0.03 0.08 À 0.01
Park À 0.09 À 0.24 À 0.46 À 0.27 À 0.05 À 0.28 À 0.17 À 0.26 À 0.25
Regional 0.24 0.08 À 0.14 0.05 0.27 0.04 0.16 0.06 0.07 0.32
Special use À 0.01 À 0.16 À 0.39 À 0.19 0.03 À 0.20 À 0.09 À 0.19 À 0.17 0.08 À 0.25
University 0.12 À 0.03 À 0.25 À 0.06 0.16 À 0.07 0.04 À 0.05 À 0.04 0.21 À 0.11 0.13
Values in bold italic-type indicate pairings where the mean difference exceeded critical value at a = 0.0001.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321 313
Ninety-four pairings were significantly different in terms of
mean NDVI measurements. Overall, multiple comparisons
of means indicated that more zoning categories could be
differentiated on the basis of NDVI as compared to Ts at
both the general and detailed levels.
4.4. Relationships between Ts and NDVI
A significant inverse relationship between Ts and NDVI
has been well documented in the remote sensing literature
for both urban and rural environments (e.g., Gallo &
Tarpley, 1996; Goward, Xue, & Czajkowski, 2002). The
basis of this relationship is that higher levels of latent heat
exchange are more typical of areas characterized by significant
vegetation cover as compared to areas with little or
no vegetation cover and low surface moisture availability,
such as densely developed urban areas, where sensible heat
exchange is favored (Oke, 1982; Owen et al., 1998).
Sobrino and Raissouni (2000), for example, used the inverse
relationship between NDVI and radiant surface temperature
measurements collected in multitemporal Advanced Very
High Resolution Radiometer (AVHRR) imagery for characterizing
land cover dynamics in Morocco. Lo, Quattrochi,
and Luvall (1997) found significant inverse relationships
between NDVI and radiant temperature associated with
residential and agricultural land use in suburban areas of
Huntsville, AL using high-resolution measurements captured
by the airborne Advanced Thermal and Land Applications
Sensor (ATLAS). Gallo and Owen (1998) used the
inverse relationship between Ts and NDVI for categorizing
urban and rural climate monitoring stations as a function of
predominate surrounding land cover type. Carlson et al.
(1994) and Gillies and Carlson (1995) expanded on the
inverse relationship between Ts and NDVI to develop
methods for quantifying additional biophysical variables
including fractional vegetation cover and soil moisture
availability.
The inverse relationship between Ts and NDVI in the
Indianapolis urban ecosystem is depicted in Fig. 5, a scatter
plot in which the horizontal axis represents NDVI and the
vertical axis represents Ts. Each point represents the mean Ts
and NDVI value associated with one of the 6462 zoning
polygons in the study region. The relationship between
these variables was quantified for the entire city and within
Table 4
Results of post-hoc tests to evaluate significance of differences in (a) mean Ts and (b) mean NDVI values associated with bivariate combinations of residential
zoning categories
DS DP D1 D2 D3 D4 D5 D5II D6 D6II D7 D8 D9 D10 D11
(a) Tamhane’s T2 multiple comparison table of mean differences in Ts for residential zoning categories
DP À 3.67
D1 À 1.57 2.09
D2 À 2.86 0.81 À 1.29
D3 À 3.88 À 0.21 À 2.30 À 1.02
D4 À 4.21 À 0.54 À 2.64 À 1.35 À 0.33
D5 À 4.92 À 1.26 À 3.35 À 2.06 À 1.04 À 0.71
D5II À 4.71 À 1.05 À 3.14 À 1.85 À 0.83 À 0.50 0.21
D6 À 4.56 À 0.89 À 2.99 À 1.70 À 0.68 À 0.35 0.36 0.15
D6II À 4.69 À 1.02 À 3.11 À 1.83 À 0.81 À 0.47 0.24 0.03 À 0.12
D7 À 5.30 À 1.64 À 3.73 À 2.44 À 1.43 À 1.09 À 0.38 À 0.59 À 0.74 À 0.62
D8 À 5.67 À 2.01 À 4.10 À 2.81 À 1.79 À 1.46 À 0.75 À 0.96 À 1.11 À 0.99 À 0.37
D9 À 4.19 À 0.52 À 2.62 À 1.33 À 0.31 0.02 0.73 0.52 0.37 0.50 1.11 1.48
D10 À 4.57 À 0.91 À 3.00 À 1.71 À 0.69 À 0.36 0.35 0.14 À 0.01 0.11 0.73 1.10 À 0.38
D11 À 4.03 À 0.36 À 2.45 À 1.17 À 0.15 0.18 0.90 0.69 0.53 0.66 1.28 1.65 0.16 0.55
D12 À 5.19 À 21.52 À 3.62 À 2.33 À 1.31 À 0.98 À 0.27 À 0.48 À 0.63 À 0.50 0.12 0.48 À 1.00 À 0.62 À 1.16
(b) Tamhane’s T2 multiple comparison table of mean differences in NDVI for residential zoning categories
DP 0.27
D1 0.08 À 0.19
D2 0.13 À 0.14 0.06
D3 0.20 À 0.07 0.12 0.07
D4 0.23 À 0.04 0.15 0.10 0.03
D5 0.27 À 0.01 0.19 0.13 0.06 0.03
D5II 0.35 0.08 0.28 0.22 0.15 0.12 0.09
D6 0.28 0.01 0.20 0.15 0.08 0.05 0.01 À 0.07
D6II 0.29 0.02 0.21 0.15 0.09 0.06 0.02 À 0.07 0.01
D7 0.35 0.08 0.27 0.22 0.15 0.12 0.09 0.00 0.07 0.06
D8 0.33 0.06 0.25 0.20 0.13 0.10 0.07 À 0.02 0.05 0.04 À 0.02
D9 0.33 0.06 0.26 0.20 0.13 0.10 0.07 À 0.02 0.06 0.05 À 0.02 0.00
D10 0.27 0.00 0.19 0.13 0.07 0.04 0.00 À 0.09 À 0.01 À 0.02 À 0.08 À 0.06 À 0.07
D11 0.29 0.02 0.21 0.16 0.09 0.06 0.03 À 0.06 0.01 0.00 À 0.06 À 0.04 À 0.04 0.03
D12 0.26 À 0.01 0.18 0.13 0.06 0.03 À 0.01 À 0.09 À 0.02 À 0.03 À 0.09 À 0.07 À 0.07 À 0.01 À 0.03
Values in bold italic-type indicate pairings where the mean difference exceeded critical value at a = 0.0001.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321314
each of the general and detailed residential zoning categories
by computing a series of Pearson Product Moment
Correlation coefficients (Table 5). These tests were developed
by analyzing the relationship between each pixel value
from the Ts image and the mean of the four corresponding
pixels from the NDVI image. Aside from the correlation
computed for the entire city, each test included only the
pixels delineated in the respective zoning categories as
defined by the zoning GIS layer.
Correlation results show that relationships between Ts
and NDVI were significant for the city as a whole and
within all individual zoning categories except for the
regional category in the general zoning classification. At
the time the June 6, 2000 ETM+ image was acquired, there
was one polygon zoned as regional, which occurred just to
the west of the CBD. This area of the city includes regional
attractions such as a relatively new baseball stadium (Victory
Field), the Indianapolis Zoo, the White River State Park
Visitor’s Center, and proportionally, a relatively large segment
of the White River (Fig. 6a). The low correlation
coefficient between Ts and NDVI for this zoning polygon
can be attributed to the large area that is occupied by water,
which has very low NDVI and also low values of Ts.
Similarly, the low but statistically significant relationship
between Ts and NDVI for the parks zoning category is
primarily a function of the inclusion of Eagle Creek Reservoir
in the parks category (Fig. 6b).
It has been well documented in the remote sensing
literature that relationship between Ts and NDVI is strongly
influenced by the presence of water, both in the form of
moisture contained in soil and vegetation, as well as surface
water features such as lakes, ponds, and rivers (e.g.,
Carlson et al., 1981; Gillies & Carlson, 1995). Excluding
pixels that correspond to major surface water features from
the analysis would undoubtedly increase the significance of
the relationship between Ts and NDVI and tighten up the
dispersion of points around the regression line depicted in
Fig. 5. This would be a desirable modification to the
methods presented here if the objective was to study only
land surfaces. However, in the context of the current
research, surface water pixels were not excluded for several
reasons.
First, the sighting of specific types of zoning around the
surface water features such as lakes and rivers often
represents a conscious decision on the part of planners.
Since the intention of this research is to analyze individual
zoning polygons as a unit of the urban landscape, excluding
water pixels would preclude comparative analysis of
specific instances of zoning within their respective environmental
contexts. Secondly, surface water in the form of
retention ponds located in many of the newer residential
developments within the Indianapolis/Marion County study
region is a direct result of planning policies to account for
increased storm water runoff created by the addition of
impervious surfaces and has been a point of debate among
planners, developers, and residents. Exclusion of surface
water pixels in this case would inhibit the ability of
planners to evaluate the effects of different forms of
Fig. 5. Scatter plot of mean NDVI (x-axis) and Ts ( y-axis) associated with each zoning polygon in the Indianapolis study region.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321 315
accounting for increased impervious surface area on Ts and
NDVI. An additional reason for not excluding pixels that
correspond to surface water features relates to the spatial
resolution of the imagery. While major surface water
features in the study region, such as the larger streams
and reservoirs, could be masked out of the 60-m resolution
thermal imagery, smaller surface water features would
still be included in the form of mixed pixels, in which
a measurement for a pixel is a function of both water
and land surface that occurs within a pixel instantaneous
field of view. There are many mixed water/land surface
pixels in the imagery of the Indianapolis study region
that result from, for example, smaller retention ponds
and streams that occur within developed areas, drainage
ditches in agricultural areas, and the Whitewater Canal,
which runs through the western portion of the Indianapolis
CBD. Thus, at some point, the removal of large surface
water features and the inclusion of smaller surface water
features in the form of mixed pixels may ultimately bias
the results.
4.5. Example applications of Ts and NDVI feature space in
urban planning
A potential application of the inverse relationship
between Ts and NDVI in urban planning is to use this
phenomenon as a comparative tool for finding examples
of low and high impact implementations within a particular
zoning category. Consider that all polygons making
up a particular zoning category provide examples of a
range of possible development options. By examining
variations of mean Ts and NDVI within this group,
planners can find examples of both high impact implementations,
in which development has led to low levels of
green biomass and/or high levels of radiant temperature
(i.e., low NDVI and high Ts), and low impact implementations,
where NDVI remains high and/or surface temperatures
are relatively low. Below, we use the DP (planned
unit development) residential zoning category to illustrate
this concept.
Fig. 7 shows a scatter plot of mean Ts and NDVI
associated with the 89 DP zoning polygons that occur in
the study region. High-resolution (5 m) near infrared aerial
imagery is used to illustrate the nature of land use and land
cover within selected polygons occurring in different locations
of the Ts/NDVI feature space. Points occurring at the
lower right of the scatter plot represent examples that could
be categorized as low impact implementations, while those
at the upper left represent high impact implementations.
The image at upper right shows a relatively new and
affluent single-family development zoned as DP, which
might be categorized as a low impact implementation in
terms of effect on Ts and NDVI. This subdivision occurs
along the southern portion of Geist Reservoir, and a
significant amount of tree cover was left intact during the
development. Tree cover appears as middle gray tones
located to the rear of individual houses, while grass appears
in brighter tones, located primarily in the front of the
houses. While the dark area around the development
represents the surface water of the reservoir, the lake itself
is not part of the same polygon, and thus has no direct
effect on the location of this zoning unit in the Ts/NDVI
feature space image. The image at upper left represents
what might be categorized as a high impact example,
displaying high mean Ts and low mean NDVI. This area
consists of multifamily apartment buildings, with little
vegetation left intact. Mean Ts remains high despite the
retention ponds that are included in the center of the
polygon, which have an effect of pulling the location of
this zoning polygon towards the origin the graph. The
image at lower left depicts what might be categorized as
an outlier in the sense that while the mean NDVI value is
relatively low, so is the mean Ts value. Like the regional
and park examples shown in Fig. 6, this polygon exhibits
low correlation between Ts and NDVI because of the
relatively large proportion covered by water. This development
occurs on the site of a former gravel quarry.
Table 5
Pearson product moment correlation coefficients and coefficients of
determination for the relationship between Ts and NDVI for (a) general
zoning classes and (b) residential zoning classes
r r2
sig.
(a) General zoning
Agriculture À 0.665 0.443 *
Airport À 0.658 0.433 *
CBD À 0.440 0.193 *
Commercial À 0.632 0.400 *
Dwellings À 0.661 0.436 *
Historic À 0.677 0.459 *
Hospital À 0.787 0.619 *
Industrial À 0.484 0.234 *
Interstate À 0.571 0.326 *
Park À 0.268 0.072 *
Regional À 0.009 0.000
Special use À 0.523 0.274 *
University À 0.669 0.447 *
Entire city À 0.563 0.316 *
(b) Residential zoning
DS À 0.616 0.380 *
DP À 0.520 0.271 *
D1 À 0.523 0.274 *
D2 À 0.593 0.351 *
D3 À 0.647 0.419 *
D4 À 0.649 0.421 *
D5 À 0.701 0.491 *
D5II À 0.682 0.465 *
D6 À 0.450 0.202 *
D6II À 0.588 0.346 *
D7 À 0.361 0.130 *
D8 À 0.627 0.394 *
D9 À 0.520 0.271 *
D10 À 0.913 0.834 *
D11 À 0.670 0.449 *
D12 À 0.647 0.419 *
Statistical significance at a = 0.01 for two-tailed test at corresponding df
signified by *.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321316
4.6. Extending the Ts/NDVI feature space model
A possible point of contention arising from the results
presented thus far is that most of the low environmental
impact (i.e., high NDVI and/or low Ts) examples of zoning
ordinances observed in the study region are characterized by
low-density development (e.g., the single family housing
development shown in the upper right of Fig. 7 and the DS
residential zoning category). These forms of development
may be considered the least desirable to planners who cite
low-density development as an undesirable trait in the
context of urban sprawl. A more compliant application of
the comparison of zoning in Ts/NDVI feature space from the
perspective of mitigating urban sprawl would be the inclusion
of third axis of development density—the idea being
that the most ideal examples of development would exhibit
some combination of low Ts, high NDVI, and high-density
development.
An example of extending the Ts/NDVI feature space by
adding a third axis of development density is presented in
Fig. 8a. The third axis represents the percent of the zoning
polygon covered by developed land cover as defined by
building footprint and edge of pavement GIS polygon data.
The same DP zoning polygons used in the 2D example
above are used here. A zoning polygon that occurs in the
‘‘low impact’’ corner of the 3D feature space is selected to
Fig. 6. Examples of zoning polygons exhibiting low correlation between Ts and NDVI: (a) 5-m panchromatic aerial image of regional zoning polygon, (b) 30-m
resolution near infrared ETM+ (Band 4) image of Eagle Creek Park Polygon.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321 317
illustrate the relationship between what occurs on the ground
and how this translates into the feature space graphic. This
polygon exhibited a development density value of 37%, a
mean NDVI of 0.3, and a mean Ts value of 297jK. This
zoning polygon is symbolized by the cube within the 3D
feature space graph and contains five multifamily housing
structures. Although this zoning unit occurs close to a busy
thoroughfare, a fair amount of tree cover has been retained as
indicated by the medium gray tones that occur near the edges
of the polygon. Instead of creating separate impervious areas
to accommodate parking, the developers have included
attached garages that also accommodate living space above.
The only additional impervious surfaces, aside from the
buildings, are the access roads. The remainder of the development
is occupied by grass as indicated by the bright tones
immediately surrounding the structures.
While graphical comparison of individual zoning units
within the 3D feature space provides a visual tool for
comparing the impact of landscape units in terms of effect
on Ts, NDVI, and development density, this can be extended
to a more quantitative index by computing the Euclidean
distance from a zoning units location in the 3D feature space
to a benchmark location in the ‘‘low impact’’ corner of the
graphic. Such an environmental impact index could be
computed as (Eq. (4)):
Index Value ¼ SQRTðXi À XjÞ2
þ ðYi À YjÞ2
þ ðZi À ZjÞ2
ð4Þ
where Xi, Yi, and Zi = the respective values of Ts, NDVI, and
development density for a particular zoning unit, and Xj, Yj,
and Zj = the respective values of Ts, NDVI, and development
density for the benchmark location in the low impact
portion of feature space. The logic of applying this environmental
impact index to the 89 DP zoning polygons in the
study region is presented graphically in Fig. 8b. The lines
associated with each point represent the Euclidean distance
from each DP polygon location in the 3D feature space to
the benchmark location. The DP zoning polygon with the
lowest index value corresponds to the white cube depicted
in Fig. 8a and b. The benchmark value from which
Euclidean distance measurements are based in this example
was set at Ts = 295, NDVI = 0.5, and development density
= 65%.
The general idea suggested by these examples is that
the integration of city’s GIS data with remote sensing
measurements of Ts and NDVI provides the opportunity to
gain further insight into the environmental consequences of
planning decisions. A range of possible development
options can be represented in both spatial and quantitative
forms, providing a comparative context that relates to
the environmental effects imposed by development. This
provides planners with the ability to evaluate the effects
of existing development and, by exploring the specific
reasons for the occurrence of low impact examples, to
shape policies that may lessen the impact of future devel-
opment.
Fig. 7. Scatter plot of mean Ts and NDVI associated with the 89 DP (planned unit development) zoning polygons. Five-meter resolution panchromatic aerial
images illustrate land use–land cover of different forms of development within DP ordinance corresponding to selected locations in Ts/NDVI feature space.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321318
Fig. 8. (a) 3D scatter plot of Ts, NDVI, and % development density associated with the 89 DP (planned unit development) zoning polygons; (b) conceptual
diagram of environmental impact index using Ts, NDVI, and % development density.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321 319
5. Summary and conclusions
The results of this research indicate that different types of
zoning have significantly different effects on radiant surface
temperature and NDVI as measured by the ETM+ sensor
within the Indianapolis urban ecosystem. Comparisons of
mean Ts and NDVI values associated with individual pairings
of zoning categories also suggest that the majority can
be differentiated on the basis of both Ts and NDVI. A
significant inverse relationship between Ts and NDVI was
observed for the entire city of Indianapolis and within all but
one of the individual zoning categories. Examples of how
this relationship can be used by planners to find low and
high impact examples within a particular zoning category
are provided. An extension of this model that includes the
additional variable of development density provides a
quantitative index that should be useful to planners that
seek examples of development that increase density in the
context of slowing urban sprawl, while lessening the
impacts of development on radiant temperatures and vege-
tation.
While zoning polygons were used as the primary unit of
analysis in this research, the same methods and concepts
could be extended to other forms of urban landscape
delineation. Many of the specific decisions that influence
radiant surface temperature and vegetation in an urban
ecosystem, such as building materials, landscaping, and,
to some extent, building form, are often implemented at the
development, tract, or individual parcel level (Rosenfeld et
al., 1995). This study has shown that moderate spatial
resolution remote sensing data, such as that collected by
the ETM+ sensor, are useful for regional scale analyses, and
the methods presented should be applicable to other forms
of urban landscape delineations such as individual developments
and neighborhoods. A few examples of the application
of higher spatial resolution (e.g., < 5 m) multispectral
remote sensing imagery for analyzing the impacts of planning
on the interactions between radiant temperature and
vegetation within urban ecosystems are in development
(e.g., Quattrochi et al., 2000; Stone & Rodgers, 2001), but
are the exceptions. As higher resolution remote sensing
imagery becomes more commonly available, the methods
presented in this research could be extended to more
specific urban design considerations that impact the thermal
and vegetation components of urban ecosystems at finer
spatial scales.
Overall, planners have the opportunity to gain significant
insight into the physical manifestations of planning policies
within cities by integrating quantitative analysis of electromagnetic
energy measurements collected by remote sensing
systems. Many planning departments already have extensive
GIS systems in place that can be adapted to develop
analyses such as the one presented in this research. Personnel
familiar with principles of map algebra in the raster data
model should not have a significant difficulty in extending
those skills to analysis of remote sensing imagery. The
recent launch of the Landsat 7 satellite equipped with the
ETM+ sensor and the control of distribution and pricing of
ETM+ imagery by US Federal Government agencies have
also had the effect of driving down prices of moderate
resolution imagery; ETM+ imagery can currently be purchased
for about $600.00 (US) per scene. Thus, with a
moderate investment in time and at minimal cost, planners
have the potential to add a powerful tool to the planner’s
toolbox.
References
Akbari, H., & Taha, H. (1992). The impact of trees and white surfaces on
residential heating and cooling energy use in four Canadian cities. Energy,
17, 141–149.
Barr, S., & Barnsley, M. (2000). Reducing structural clutter in land cover
classifications of high spatial resolution remotely-sensed images for
urban land use mapping. Computers and Geosciences, 26, 433–449.
Carlson, T. N., & Arthur, S. T. (2000). The impact of land use–land cover
changes due to urbanization on surface microclimate and hydrology: As
satellite perspective. Global and Planetary Change, 25, 49–65.
Carlson, T. N., Augustine, J. N., & Boland, F. E. (1977). Potential application
of satellite temperature measurements in analysis of land use
over urban area. Bulletin of the American Meteorological Society, 58,
1301–1303.
Carlson, T. N., Dodd, J. K., Benjamin, S. G., & Cooper, J. N. (1981).
Satellite estimation of the surface energy balance, moisture availability
and thermal inertia. Journal of Applied Meteorology, 20, 67–87.
Carlson, T. N., Gillies, R. R., & Perry, E. M. (1994). A method to make use
of thermal infrared temperature and NDVI measurement to infer surface
soil water content and factional vegetation cover. Remote Sensing Reviews,
9, 161–173.
Changnon, S. A. (1992). Inadvertent weather modification in urban areas:
Lessons for global climate change. Bulletin of the American Meteorological
Society, 73, 619–627.
Estes, J. E. (1966). Some applications of aerial infrared imagery. Annals of
the Association of American Geographers, 56, 673–682.
Fung, T., & Siu, W. (2000). Environmental quality and its changes, and
analysis using NDVI. International Journal of Remote Sensing, 21,
1011–1024.
Gallo, K. P., & Owen, T. W. (1998). Assessment of urban heat islands: A
multi-sensor approach for the Dallas–Ft. Worth USA region. Geocarto
International, 13, 35–41.
Gallo, K. P., & Tarpley, J. D. (1996). The comparison of vegetation index
and surface temperature composites of urban heat-island analysis. International
Journal of Remote Sensing, 17, 3071–3076.
Gamon, J. A., Field, C. B., Goulden, M., Griffin, K., Hartley, A., Joel, G.,
Pen˜uelas, J., & Valentini, R. (1995). Relationships between NDVI,
canopy structure, and photosynthetic activity in three Californian vegetation
types. Ecological Applications, 5, 28–41.
Gamon, J. A., Field, C. B., Roberts, D. A., Ustin, S. L., & Valentini, R.
(1993). Functional patterns in an annual grassland during an AVIRIS
overflight. Remote Sensing of Environment, 44, 1–15.
Gillies, R. R., & Carlson, T. N. (1995). Thermal remote sensing of surface
soil water content with partial vegetation cover for incorporation into
climate models. Journal of Applied Meteorology, 34, 745–756.
Gillies, R. R., Cui, J., Carlson, T. N., Kustas, W. P., & Humes, K. S. (1997).
Verification of the ‘triangle’ method for obtaining surface soil water
content and energy fluxes from remote measurements of NDVI and
surface radiant temperature. International Journal of Remote Sensing,
18, 3145–3166.
Goward, S. N. (1981). Thermal behavior of urban landscapes and the urban
heat island. Physical Geography, 2, 19–33.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321320
Goward, S. N., Cruikhanks, G. D., & Hopes, A. S. (1985). Observed
relation between thermal emission and reflected spectral radiance of a
complex vegetated landscape. Remote Sensing of Environment, 18,
137–146.
Goward, S. N., Xue, Y., & Czajkowski, K. P. (2002). Evaluating land
surface moisture conditions from the remotely sensed temperature/vegetation
index measurements: An exploration with the simplified simple
biosphere model. Remote Sensing of Environment, 79, 225–242.
Grimmond, C. S. B., & Oke, T. R. (1991). An evapotranspiration–
interception model for urban areas. Water Resources Research, 27,
1739–1755.
Grimmond, C. S. B., & Oke, T. R. (1995). Comparison of heat fluxes from
summertime observations in the suburbs of four North American cities.
Journal of Applied Meteorology, 34, 873–889.
Hammer, T. R. (1972). Stream channel enlargement due to urbanization.
Water Resources Research, 8, 1530–1540.
Heisler, G. M. (1986). Energy savings with trees. Journal of Arboriculture,
12, 113–125.
Henry, J. A., Dicks, S. E., Wetterquist, O. E., & Roguski, S. J. (1989).
Comparison of satellite, ground-based, and modeling techniques of
analyzing the urban heat island. Photogrammetric Engineering and
Remote Sensing, 55, 69–76.
Hochberg, Y., & Tamhane, A. (1987). Multiple comparison procedures.
New York: Wiley. 450 pp.
Huang, J., Akari, H., Taha, H., & Rosenfeld, A. (1987). The potential of
vegetation in reducing summer cooling loads in residential buildings.
Journal of Climate and Applied Meteorology, 26, 1103–1106.
Jensen, J. R., & Cowen, D. C. (1999). Remote sensing of urban/suburban
infrastructure and socio-economic attributes. Photogrammetric Engineering
and Remote Sensing, 65, 611–622.
Ji, M., & Jensen, J. R. (1999). Effectiveness of subpixel analysis in detecting
and quantifying urban imperviousness from Landsat Thematic Mapper
imagery. Geocarto International, 14, 31–39.
Karl, T. R., Diaz, H. F., & Kukla, G. (1988). Urbanization: It detection and
effect in the United States climate record. Journal of Climate and
Applied Meteorology, 1, 1099–1123.
Landsat Project Science Office (2002). Landsat 7 science data user’s
handbook. Goddard Space Flight Center. www address: http://ltpwww.
gsfc.nasa.gov/IAS/handbook/handbook_toc.html.
Lo, C. P., Quattrochi, D. A., & Luvall, J. C. (1997). Application of
high-resolution thermal infrared remote sensing and GIS to assess the
urban heat island effect. International Journal of Remote Sensing, 18,
287–304.
McPherson, E. G., Nowak, D., Heisler, G., Grimmond, S., Souch, C.,
Grant, R., & Rowntree, R. (1997). Quantifying urban forest structure,
function, and value: The Chicago Urban Forest Climate Project. Urban
Ecosystems, 1, 49–61.
Metropolitan Development Commission (1998). Dwelling zoning districts
for Marion County, Indiana. Indianapolis: Metropolitan Development
Commission. www address: http://www.indygov.org/cp/zonord.htm.
Nemain, R., Pierce, L., Running, S., & Goward, S. (1993). Developing
satellite-derived estimates of surface moisture status. Journal of Applied
Meteorology, 32, 548–557.
Nichol, J. E. (1996). Analysis of the urban thermal environment with
LANDSAT data. Environment and Planning B: Planning and Design,
23, 733–747.
Oke, T. R. (1981). Canyon geometry and the nocturnal urban heat island:
Comparison on scale model and field observation. International Journal
of Climatology, 1, 237–254.
Oke, T. R. (1982). The energetic basis of the urban heat island. Quarterly
Journal of the Royal Meteorological Society, 108, 1–24.
Oke, T. R. (1987). Boundary layer climates. New York: Methuen Press.
435 pp.
Owen, T. W., Carlson, T. N., & Gillies, R. R. (1998). An assessment of
satellite remotely-sensed land cover parameters in quantitatively describing
the climatic effect of urbanization. International Journal of
Remote Sensing, 19, 1663–1681.
Pease, R. W., Lewis, J. E., & Outcalt, S. I. (1976). Urban terrain climatology
and remote sensing. Annals of the Association of American Geographers,
66, 557–569.
Qi, J., Moran, M. S., Cabot, F., & Dedieu, G. (1995). Normalization of sun/
view angle effects using spectral albedo-based vegetation indices. Remote
Sensing of Environment, 52, 207–217.
Quattrochi, D. A., Luvall, J. C., Rickman, D. L., Estes Jr., M., Laymon,
C. A., & Howell, B. F. (2000). A decision support information system
for urban landscape management using thermal infrared data. Photogrammetric
Engineering and Remote Sensing, 66, 1195–1207.
Quattrochi, D. A., & Ridd, M. K. (1994). Measurement and analysis of
thermal energy responses from discrete urban surfaces using remote
sensing data. International Journal of Remote Sensing, 15, 1991–2022.
Reed, B. C., Brown, J. F., VanderZee, D., Loveland, T. R., Merchant, J. W.,
& Ohlen, D. O. (1994). Measuring phenological variability from satellite
imagery. Journal of Vegetation Science, 5, 703–714.
Rosenfeld, A. H., Akbari, H., Bretz, S., Fishman, B. L., Kurn, D. M.,
Sailro, D., & Taha, H. (1995). Mitigation of urban heat island: Materials,
utility programs, updates. Energy Building, 22, 255–265.
Roth, M., Oke, T. R., & Emery, W. J. (1989). Satellite-derived urban heat
islands from three coastal cities and the utilization of such data in urban
climatology. International Journal of Remote Sensing, 10, 1699–1720.
Ryznar, R. M., & Wagner, T. W. (2001). Using remotely sensed imagery to
detect urban change: Viewing Detroit from space. Journal of the American
Planning Association, 67, 327–336.
Saaroni, H., Ben-Dor, E., Bitan, A., & Potcher, O. (2000). Spatial distribution
and microscale characteristics of the urban heat island in TelAviv,
Israel. Landscape and Urban Planning, 48, 1–18.
Schmidt, H., & Gitelson, A. (2000). Temporal and spatial vegetation cover
changes in Israeli transition zone: AVHRR-based assessment of rainfall
impact. International Journal of Remote Sensing, 21, 997–1010.
Sobrino, J. A., & Raissouni, N. (2000). Toward remote sensing methods of
land cover dynamic monitoring: Application to Morocco. International
Journal of Remote Sensing, 21, 353–366.
Stone, B., & Rodgers, M. O. (2001). Urban form and thermal efficiency:
How the design of cities influences the urban heat island effect. Journal
of the American Planning Association, 67, 186–198.
Taha, H. (1997). Modeling impacts of increased urban vegetation on ozone
air quality in the South Coast Air Basin. Atmospheric Environment, 30,
3430–3432.
Ulrich, R. S. (1986). Human responses to vegetation and landscapes. Landscape
and Urban Planning, 13, 29–44.
Verstrate, M. M., & Pinty, B. (1996). Designing optimal spectral indexes
for remote sensing applications. IEEE Transactions in Geosciences and
Remote Sensing, 34, 1254–1265.
Vitousek, P. M., Mooney, H. A., Lubchenco, J., & Melillo, J. M. (1997).
Human domination of earth’s ecosystems. Science, 277, 494–499.
Voogt, J. A., & Oke, T. R. (1997). Complete urban surface temperatures.
Journal of Applied Meteorology, 36, 1117–1132.
Voogt, J. A., & Oke, T. R. (1998). Effects of urban geometry on remotelysensed
surface temperature. International Journal of Remote Sensing,
19, 895–920.
Wellar, B. S., & Estes, J. E. (1968). On applications of aerial infrared
imagery. Annals of the Association of American Geographers, 58,
411–415.
J.S. Wilson et al. / Remote Sensing of Environment 86 (2003) 303–321 321