Remote Sensing of Environment 174 (2016) 233–243

Contents lists available at ScienceDirect

Remote Sensing of Environment
journal homepage: www.elsevier.com/locate/rse

Remote sensing of the surface urban heat island and land architecture in
Phoenix, Arizona: Combined effects of land composition and
configuration and cadastral–demographic–economic factors
Xiaoxiao Li a,⁎, Wenwen Li b, A. Middel a,b, S.L. Harlan c, A.J. Brazel a,b, B.L. Turner II

a,b,d

a

Julie Ann Wrigley Global Institute of Sustainability, Arizona State University, Tempe, AZ 85287-5402, United States
School of Geographical Sciences & Urban Planning, Arizona State University, Tempe, AZ 85287-5302, United States
c
Department of Health Sciences and Department of Sociology, Northeastern University, Boston, Massachusetts, 02115, United States
d
School of Sustainability, Arizona State University, Tempe, AZ 85287-5502, United States
b

a r t i c l e

i n f o

Article history:
Received 20 October 2014
Received in revised form 5 November 2015
Accepted 15 December 2015
Available online 29 December 2015
Keywords:
Land architecture
Surface urban heat island
Land surface temperature
Land cover
Spatial pattern

a b s t r a c t
This study seeks to determine the role of land architecture—the composition and configuration of land cover—as
well as cadastral–demographic–economic factors on land surface temperature (LST) and the surface urban heat
island effect of Phoenix, Arizona. It employs 1 m National Agricultural Imagery Program data of land-cover with
120 m Landsat-derived land surface temperature, decomposed to 30 m, a new measure of configuration, the normalized moment of inertia, and U.S. Census data to address the question for two randomly selected samples comprising 523 and 545 residential neighborhoods (census blocks) in the city. The results indicate that, contrary to
most other studies, land configuration has a stronger influence on LST than land composition. In addition, both
land configuration and architecture combined with cadastral, demographic, and economic variables, capture a
significant amount of explained variance in LST. The results indicate that attention to land architecture in the development of or reshaping of neighborhoods may ameliorate the summer extremes in LST.
© 2015 Elsevier Inc. All rights reserved.

1. Introduction
The urban heat island (UHI) effect refers to the higher air and surface
temperature in urban areas compared to that of the surrounding rural
hinterland, generated by high levels of near surface energy emission,
solar radiation absorption of ground objects, and often, low levels of
evapotranspiration in cities (Buyantuyev & Wu, 2010; Oke, 1982,
1997; Rizwan, Dennis, & Liu, 2008). The UHI of large cities has increased
substantially since the middle of the 20th century (Akbari, Pomerantz, &
Taha, 2001; Oke, 1976; Stone, 2007), with urban conglomerations generating modeled and observed changes in regional temperatures
(Georgescu, Moustaoui, Mahalov, & Dudhia, 2011; He, Liu, Zhuang,
Zhang, & Liu, 2007; Kalnay & Cai, 2003; Li, Wang, Shen, & Song, 2004).
Extensively examined, the UHI draws increasing attention owing to its
effects on energy and water consumption, human health, environmental (ecosystem) services, especially in the context of global warming
(e.g. Gober, Kirkwood, Balling, Ellis, & Deitrick, 2009; Harlan, Brazel,
Prashad, Stefanov, & Larsen, 2006; Harlan, Declet-Barreto, Stefanov, &
Petitti, 2013; Hondula, Vanos, & Gosling, 2013; Hondula et al., 2012).
For these and other reasons, attention to the means to mitigate the

⁎ Corresponding author.
E-mail address: xiaoxia4@asu.edu (X. Li).

http://dx.doi.org/10.1016/j.rse.2015.12.022
0034-4257/© 2015 Elsevier Inc. All rights reserved.

UHI effect have garnered considerable attention, especially through
model simulations (e.g. Golany, 1996; Sailor, 1995), but increasingly
through the use or remote sensing data that permit novel assessments
(e.g., Zhou, Huang, & Cadensasso, 2011).
Remote sensing technology has been a boon to the study of the UHI
in at least two fundamental ways: [1] direct observation of land surface
thermal radiance converted to land surface temperature to address the
Surface UHI (SUHI) (e.g. Lo, Qiattrochi, & Luvall, 1997; Streutker, 2002);
and [2] direct spatial linkages of ground features, both their vertical dimensions and patterns, to land surface temperature at fine spatial resolutions (e.g. Arnfield, 2003; Buyantuyev & Wu, 2010; Chow, Chuang, &
Gober, 2012; Nichol, 1996; Unger, 2004). To date, this research has focused on the relationship between land surface temperature (LST) and
particular land-cover types (or land composition) (e.g. Chow & Brazel,
2012; Li, Song, Cao, Meng, & Wu, 2011; Middel, Häb, Brazel, Martin, &
Guhathakurta, 2014; Stone & Rodgers, 2001; Zheng, Myint, & Fan,
2014; Zhou et al., 2011), and between spatial thermal patterns and social economic factors (Buyantuyev & Wu, 2010; Harlan et al., 2006;
Hondula et al., 2013; Huang, Zhou, & Cadenasso, 2011; Jenerette,
Harlan, Stefanov, & Martin, 2011). Recently, however, attention to
land system architecture (Turner, Janetos, Verburg, & Murray,
2013)—the composition and configuration (e.g., size, shape, patterns,
and connectivity) of the urban land cover—on SUHI has been examined
in regard to possible UHI mitigation strategies (Chen, Zhao, Li, & Yin,

234

X. Li et al. / Remote Sensing of Environment 174 (2016) 233–243

2006; Connors, Galletti, & Chow, 2013; Li, Zhou, Quyang, & Zheng, 2012;
Li et al., 2013b; Zhou et al., 2011).
Briefly summarizing, this research informs us that increasing
greenspace, water, and skyview (open area ventilation) tend to ameliorate the UHI effect, while dark-colored impervious surfaces tend to amplify it, with larger impacts on nighttime temperatures (e.g. Chow &
Brazel, 2012; Li et al., 2011; Maimaitiyiming et al., 2014; Weng, Lu, &
Schubring, 2004; Xian & Crane, 2006; Zheng et al., 2014; Zhou, Qian,
Li, Li, & Han, 2014). These relationships tend to hold across different
urban areas and environments, but vary in magnitude diurnally and
seasonally. Specific linkages to human outcomes demonstrate that UHI
impacts tend to be registered most highly among those parts of city
that have dense occupation with low levels of shade, either from buildings or trees, and greenspaces. These conditions, at least in the U.S. cities
examined, tend to be related to lower levels of income, often linked to
neighborhoods dominated by certain ethnic groups (Buyantuyev &
Wu, 2010; Harlan et al., 2006; Hondula et al., 2012; Jenerette et al.,
2007, 2011). Finally, the land architecture of urban areas, from the parcel to larger levels of assessment, has been hypothesized to amplify or
ameliorate ecosystem services, such as those related to SUHI effects
(Turner et al., 2013). Nascent research suggests that, controlling for
land composition, edge and patch densities, landscape shape index,
and fractal dimensions (FRAGSTAT metrics) of land covers hold significant consequences for land surface temperatures (Buyantuyev & Wu,
2010; Connors et al., 2013; Li et al., 2011; Li et al., 2012; Middel,
Brazel, Kaplan, & Myint, 2012; Middel et al., 2014; Stone & Rodgers,
2001; Zhang, Odeh, & Ramadan, 2013; Zhang, Zhong, Feng, & Wang,
2009; Zhou et al., 2011).
Our study follows from but extends these lines of research. It seeks
to determine if LST among residential neighborhoods during the summer season, specifically June, is related to the land architecture and certain cadastral, demographic, and socioeconomic characteristics of the
neighborhoods in the metropolitan area of Phoenix, AZ. We employ
fine resolution spatial data and a compactness pattern measure—the
normalized moment of inertia—applied for the first time in land architecture assessments. We test the following hypotheses drawn from or
implied in the body of research reviewed above:
[1] [1] The composition and configuration (i.e., land architecture) of
land-cover types affect summer daytime LST and thus the SUHI. Research to date has yet to explicate adequately the configuration
dimensions of multiple land covers on SUHI.
[2] [2] Land composition is more strongly related to summer daytime
LST than configuration. Most of the research results to date, typically using FRAGSTATS metrics, indicate the stronger role of composition, or area of land covers, on the SUHI.
[3] [3] Land architecture has more impact on summer daytime LST than
do cadastral–demographic–socioeconomic factors. Some research
implies that these factors and land architecture may be linked,
but assessing their relative roles has yet to be examined fully.
In our study, cadastral data provide information on parcel size,
which are used as ancillary data for the economic dimension in
question.

2. Study area, data, and methods
2.1. Study area
The City of Phoenix, AZ, is the center of an expansive metropolitan
area located on the northern edge of the Sonoran Desert (Fig. 1). June
maximum daily temperatures average 40 °C, with the highest recorded
temperature reaching 48.3 °C (Middel et al., 2012). Temperatures are
amplified by the UHI effect, especially in regard to an enlarged minimum night time temperature (Chow et al., 2012; Hawkins, Brazel,
Stefanov, Bigler, & Saffell, 2004; Stabler, Martin, & Brazel, 2005). This

effect has been triggered by the massive growth in the metropolis.
Since the middle of the 20th century, there have been major increases
in the area of impervious surface and numbers of residential parcels
and neighborhoods with different levels of vegetation and bare soil
(including rock and desert surfaces). Residential landscape
composition—variations of turf lawns to xeric- and desert-scapes—are
associated with the period of development, rules of Home Owner Associations (HOAs), and income levels, among other factors (Chow &
Brazel, 2012; Kane, Connors, & Galletti, 2014; Larson, White, Gober,
Harlan, & Wutich, 2009b; Sha & Tian, 2010; Shrestha, York, Boone, &
Zhang, 2012; Turner & Ibes, 2011). For the most part, the presence of
vegetation and open greenspaces beyond residential parcels
(e.g., parks and golf courses) varies across the city, with lower levels
of both apparently related to lower income and Hispanic neighborhoods
in Phoenix proper (Harlan et al., 2006; Jenerette et al., 2011).
2.2. Data and methods
2.2.1. Parcel and neighborhood selection
This study draws on residential census blocks (Fig. 1), our surrogate
for neighborhoods, composed primarily, but not exclusively, of single
family residential parcels and for which cadastral data provide landuse information for each parcel. Two random samples (A and B) were
employed in order to verify the results of the exercise, and each sample
was drawn by the same method (Fig. 1). The Create Random Points
function in the GIS software ArcMap selected 1000 single family residential parcels (0.2% of such parcels in the city) for both samples. To ensure spatial independence, both samples were reduced by applying a
distance filter in which each parcel had to be at least 1000 m from the
others. The census block (neighborhood) for each parcel was identified,
and another distance filter was applied: each census block had to be at
least 500 m from one another. This sampling and filtering procedure
yielded 21,505 and 24,167 single family parcels that are distributed
over 523 and 545 census blocks, respectively, across the city. The parcel
and census block selection outcomes are provided in Table 1 and the
distributions of the A and B sample census blocks are mapped in Fig. 1.
At least two issues are noteworthy in our sample. First, parts of the
metropolitan area with the highest SUHI were excluded from consideration because they are predominately nonresidential in composition.
Nevertheless, the sample included neighborhoods within the higher
SUHI zones adjacent to the central commercial district and to nativevegetation parklands, both of which tend to have high LST (Fig. 2). Second, census block data were employed because of our focus on the land
architecture–LST relationship, which is more accurately captured at the
fine-spatial grain of the census block rather than at the more coarsespatial grain of the census block group. In a few cases, our sampled census blocks had 10 or fewer households, and household income data provided by the census are derived at the census block group level and
applied to block level for our entire sample. These two issues may affect
our results.
2.2.2. Land-cover classification
To examine the heterogeneous land architecture of neighborhoods
requires detailed land-cover data. These data were derived through
the image classification of the National Agricultural Imagery Program
(NAIP) data, 1 m orthorized aerial photography taken from June 8–10,
2010. The NAIP dataset includes four spectral bands (red, green, blue,
and near infrared band) with their radiances converted to a Digital
Number ranging from 0–255, mosaicked at the county scale. The dataset
was cookie-cut to the extent of the Central Arizona-Phoenix Long-term
Ecological Research program (www.caplter.asu.edu), which covers
most of the metro-Phoenix area, and preprocessed with pixel-based
spectral transformations, which included red, green, and blue to intensity, hue, and saturation, principal components analysis, and Normalized
Difference Vegetation Index (NDVI). An Object-Based Image Analysis
(OBIA) was utilized to produce the land-cover classification (Li et al.,

X. Li et al. / Remote Sensing of Environment 174 (2016) 233–243

235

Fig. 1. The study area, random sample A and B census blocks, and NAIP-based land cover. The points indicate the selected census blocks, and the polygon is the boundary of the city of
Phoenix. The background image is the NAIP data with near-infrared band, red band, and green band displayed on the red, green, and blue channel respectively.

2014b). This approach was built on a hierarchical image object network,
with image segmentation algorithms (multi-resolution segmentation,
multi-threshold segmentation, chessboard segmentation, and Quadtree
segmentation) to divide the image into different levels of image objects.
Image objects that belong to specific land-cover types maintain unique
characteristics in terms of spectral, spatial, contextual, and geometric information. In addition, cadastral data in a GIS vector layer were used to

Table 1
Description of the random sample selection for sample A and sample B.
Sample Sample
A
B
1
2
3
4
5
6
7
8
9
10
11
12
13
14

Number of initial randomly selected parcels
Number of parcels reduced by distance filter (1000 m)
Number of census blocks with parcels in #2 with distance
filter (500 m)
Percent of selected census blocks for Phoenix
Number of parcels in selected census blocks
Number of residential parcels in selected census blocks
Percent of residential parcels in selected census blocks
Number of single family residential parcels (subset of #6)
Percent of single family residential parcels of total
residential parcels (#6)
Number of multi-family residential parcels (subset of #6)
Percent of multi-family residential parcels of total
residential parcels (#6)
Percent of residential parcels in selected census blocks
Number of non-residential parcels in selected census
blocks
Percent of non-residential parcels in selected census blocks

1000
723
523

1000
787
545

55%
27,470
26,614
97%
21,505
81%

57%
27,528
26,326
96%
24,167
92%

5109
19%

2159
8%

97%
856

96%
1202

3%

4%

regulate the boundary of each parcel. Expert knowledge decision rules
were created based on those characteristics to delineate the landcover types. All parcels and adjacent streets were analyzed for the selected neighborhoods. The land-cover mapping includes 12 classes
with a 91.86% overall accuracy (Li et al., 2014b). Five classes (building,
impervious surfaces, open soil, vegetation, and swimming pool) dominated our neighborhoods and were used in this analysis. The category
“impervious surfaces” refers to roads, parking lots, and related features,
but not to buildings.
2.2.3. Land surface temperature
Landsat Thematic Mapper was used to derive LST: cloud free TM sensor (row = 37, path = 37) taken about 10 AM, on June 10, 2010. This date
matched the three-day window in which the NAIP data (above) were derived. In addition, weather station data for that window (June 8–10), revealed highly similar temperatures and conditions, clear and calm, typical
mid-morning weather for Phoenix in early June. The Landsat Band 6 thermal data (120 m) were re-projected to the Universal Transverse Mercator
projection system Zone 12 N and converted to 30 m.
To estimate the LST from the Landsat TM thermal infrared band data,
the Digital Numbers of sensors were converted to meaningful radiance
using a spectral radiance scaling method (Eq. 1) (Chander &
Groeneveld, 2009):
Lλ ¼

LMAX λ −LMINλ
 ðQ cal −Q calmin Þ þ LMINλ
Q calmax −Q calmin

ð1Þ

where the Lλ is the cell value as radiance (W/(m2 sr μm)), Qcal is the
quantized calibrated digital number, Qcalmin is the minimum quantized

236

X. Li et al. / Remote Sensing of Environment 174 (2016) 233–243

Fig. 2. Derived land surface temperatures for the city of Phoenix using Landsat TM and selected census blocks in samples A and B. The points represent the selected census blocks.

calibrated pixel value, and Qcalmax is the maximum quantized calibrated
pixel value; LMINλ is the spectral radiance scales to Qcalmin, LMAXλ is the
spectral radiance scales to Qcalmax.
Spectral radiance was converted to brightness temperature by assuming the earth's surface is a black body (Eq. 2) (Chander, Markham,
& Helder, 2009; Coll, Galve, Sánchez, & Caselles, 2010):
Tb ¼

K

 2

K1
ln
þ1
Lλ

ð2Þ

where Tb is the brightness temperature in Kelvin, Lλ is the cell value as
radiance, K1 (607.76 W / (m2 sr μm)) and K2 (1260.56 K) are the constants of Landsat TM calibration.
The brightness temperature was converted to kinetic temperature,
or land surface temperature (Eq. 3) (Artis & Carnahan, 1982; Dakin,
Pratt, Bibby, & Ross, 1985):
Ts ¼

T

 b 
T
1 þ λ  b lnε
ρ

ð3Þ

where λ (11.5 μm) is the emitted radiance wavelength (Markham &
Barker, 1985), the coefficient ρ (0.01438 mK) is generated from the equitation ρ = hc/b, in which h (6.626 × 10−34 Js) is the Planck's constant,
c (2.998 × 108 m/s) is the velocity of light, and b (1.38 × 10−23 J/K) is the
Boltzmann constant, and ε is the surface emissivity.
The estimation of land surface emissivity is one of the essential parameters to retrieve LST. Several methods can be used to calculate
land surface emissivity (Gillespie, Matsunaga, & Rokugawa, 1998;
Snyder, Wan, Zhang, & Feng, 1998). Our study utilized the simplified

Normalized Difference Vegetation Index (NDVI) thresholds method
(NDVITHM) to obtain the values from Landsat imagery (Sobrino,
Jimenez-Munoz, & Paolini, 2004; Sobrino et al., 2008). NDVI is calculated from the Landsat imagery with the following equation (Eq. 4):
NDVI ¼

Band 4–Band3
:
Band 4 þ Band3

ð4Þ

Different approaches have been utilized to calculate the LST from the
NDVI value. Like most urban areas, Phoenix is composed of heterogeneous and fragmented landscapes, requiring methods that account for
emissivity in these conditions (e.g., Liu & Zhang, 2011). We adopted
that of Sobrino et al. (2008) (Eq. 5):
ε ¼ ε v P v þ εs ð1−P v Þ þ dε:

ð5Þ

Eq. 5 is the simplified version of the land surface emissivity calculation, where εv is the vegetation emissivity and εs is the emissivity for
urban soil surface; Pv is the proportion of vegetation derived from an
NDVI based empirical model (Eq. 7) by Carlson and Ripley (1997) and
Sobrino et al. (2004, 2008); dε is the error in the model shown in
Eq. 5, including the land surface geometrical distribution effect and internal reflections (Sobrino et al., 2004; Sobrino et al., 2008). The estimation of dε is given by (Eq. 6):
dε ¼ ð1−ε s Þð1−P v ÞFεv

ð6Þ

where F (0.55) is a shape factor (Sobrino, Caselles, & Becker, 1990;
Sobrino et al., 2004). In order to obtain more accurate and consistence
values of Pv, we assign it 0.99 when NDVI N NDVIv and 0 when
NDVI b NDVIs (where v and s refer to vegetation and soil). The NDVITHM

X. Li et al. / Remote Sensing of Environment 174 (2016) 233–243

methods described here estimate the surface emissivity for bare soil
pixels (NDVI b NDVIs) using the reflectivity value of red band ρred
(Landsat TM band 3), and the equations are expressed as (Eqs. 7 & 8):

ε¼

8
<

a þ b ρred when NDVIbNDVIs
ε v P v þ ε s ð1−P v Þ þ dε when NDVIs≤NDVI ≤NDVIv
:
εv þ dε when NDVINNDVIv


Pv ¼

NDVI−NDVI min
NDVI max −NDVI min

ð7Þ

2
:

ð8Þ

Adapting the approximation of vegetation and urban soil surface
emissivity values from Sobrino et al. (2008), land surface emissivity is
expressed as (Eq. 9):
8
< 0:979−0:035 ρred when NDVIb0:2
ε ¼ 0:986 þ 0:004 P v when 0:2 ≤NDVI ≤0:5
:
0:99 when NDVIN0:5:

ð9Þ

The derived LST (Fig. 2) describes the surface thermal condition for
the city of Phoenix. Pixel values range from 29.9 °C to 40.6 °C, and the
mean LST is 33.5 °C. Only a 0.10 C difference in LST separates the mean
temperature for Sample A = 33.4 and Sample B = 33.5. Finally, the
LSTs reported for each census block includes all parcels and streets
within that block.
It is noteworthy that retrieving representative in-situ measurements
to validate LST data from satellite images in heterogeneous environments is challenging. LST varies significantly over space and time,
even at sub-grid resolution (Li et al., 2013). Vanos et al. (in press)
found surface temperature differences of up to 30 °C between shaded
and non-shaded surfaces on a playground in Phoenix at the touch
scale (1 cm). These differences were not captured by an infrared
MASTER super spectral image (6.8 m resolution) available to them.
Due to a lack of high-resolution data, we can only qualitatively assess
the accuracy of the LST product derived from Landsat TM.
The overall accuracy of LST from satellite-based thermal infrared
(TIR) data depends on atmospheric and emissivity corrections.
Jiménez-Muñoz and Sobrino (2006) assessed errors associated with
LST estimation from TIR data and found that atmospheric effects typically lead to an error of 0.2 K to 0.7 K. Uncertainties in emissivity retrieval
result in an error of 0.2 K to 0.4 K. The total minimum error to be expected without in-situ measurements is 0.8 K. Various other studies are in
line with this accuracy assessment. Yu, Guo, and Wu (2014) retrieved
LST from Landsat 8 TIRS and found that a radiative transfer equationbased method produced results with RMSE b 1.0 K. Rozenstein, Qin,
Derimian, and Karnieli (2014) used a split window algorithm to estimate LST from Landsat 8 at RMSE = 0.93 K. We expect the accuracy of
our LST calculation to be comparable to those results.
2.2.4. Land-architecture data
Neighborhood-scale (entire census block) land architecture measures were derived from the land-cover (2.2.3) and census block data.
Measures include the area and compactness pattern index of each
land-cover type, and the same index of the census block (Table 2). The
area of each land-cover type (area_class#) was determined by the number of pixels of each type within the census block unit. The spatial pattern indicator of the land-cover types was calculated using a new
compactness pattern measure, the Normalized Moment of Inertia
(NMI) (Li et al., 2013a). As described below, NMI has been demonstrated to be more effective and accurate than other measures of the compactness of a patch (land-cover unit) and the concentration of patches
(Miller, 1953; Schwartzberg, 1965; McGarigal & Marks, 1995; Li et al.,
2014a). Compactness measures the extent to which the shape is spread
out from its center; a circle constitutes the most compact shape. Concentration measures the compactness of multiple patches, a product of

237

the distances among patches and the compactness/moment of each single patch.
The NMI measure draws on the second moment of inertia (aka, area
moment of inertia) to measure the degree to which all elements on a
natural planar shape i are concentrated. Mathematically, the second
moment of a shape IGi i about an axis perpendicular to it and passing
through its centroid Gi can be computed from:
IGi i ¼

Z

2

ð10Þ

d dai

where
dai is the infinitely small area on the shape i, d is the distance from dai to
Gi. The normalized value NMI, can then be derived by comparing IGi i to
the second moment of a circle I0 which has the same area of the shape:
NMI ¼

I0
IGi i

:

ð11Þ

2

A
,
Given that the area moment of a circle with area A is known as 2π
NMI can then be represented as:

NMI ¼

A2
2πI Gi i

:

ð12Þ

The NMI has a range of 0, 1, with 0 representing an infinitely elongated shape, and 1 representing the most compact shape, a circle.
Several prominent features of NMI make it suitable for identifying
the spatial configuration pattern of land architecture. First, NMI is
scale-independent: patches with the same shape but different size will
be given the same value, providing a more consistent view of the land
configuration than a PARA (perimeter-area ratio) approach, as in
FRAGSTATS (McGarigal, Cushman, & Ene, 2012). Second, different
from other FRAGSTATS shape indices, such as the fractal dimension
index, the linearity index, and the related circumscribing circle index,
which focus more on measuring the edge complexity by considering perimeter as an important factor, the NMI pays more attention to the
shape itself rather than the roughness of the edge boundary. Moreover,
the NMI is not only capable of measuring shape compactness of a single
object, but it can also quantify the overall compactness pattern for target
objects as a whole. Therefore, it has the ability to evaluate accurately
how patches/sub-patches are concentrated or dispersed within an
area. This is an important consideration for neighborhood study, in
which the basic area, a neighborhood, always contains more than one
object of a given land type (e.g., buildings) dispersed at different locations within the neighborhood.
Fig. 3 demonstrates the computation of NMI for multi-objects. Supposing there are three objects within the study area, each has area moment I1 , I2, and I3, and areas of A1 , A2, and A3, the area moment Ientire of
the three objects composited together can be represented as:

Ientire ¼

n
n
X
X
2
Ii þ
Ai di
i¼1

ð13Þ

i¼1

where n = 3 in this case, di is the distance from shape i’s centroid to the
centroid (X,Y) of the new object. If (xi ,yi) is the coordinates of the centroid for shape i, then (X ,Y) can be computed as:
Xn
X¼

Ax
i¼1 i i
n

Xn
;Y ¼

Ay
i¼1 i i
n

:

ð14Þ

238

X. Li et al. / Remote Sensing of Environment 174 (2016) 233–243

Table 2
Land architecture metrics and their descriptive statistics. Sample A and B on top and bottom of each row-column numeric entry, respectively.
Factor abbreviation

Factor description

Land architecture
Land composition
area1

Area building cover (m2)

Range

458.42–16,752.41
319.43–13,859.25
361.18–29,100.47
480.67–22,051.06
119.17–53,963.01
940.31–21,272.91
138.38–25,034.44
171.30–26,995.83
0.00–224.79
0.00–274.32

2

area2

Area impervious surfaces(m )

area3

Area soil cover (m2)

area4

Area vegetation cover (m2)

area5

Area swimming pool cover (m2)

Land configuration
nmi1

Compactness pattern index building cover

nmi2

Compactness pattern index road

nmi3

Compactness pattern index soil cover

nmi4

Compactness pattern index vegetation cover

nmi5

Compactness pattern index swimming pool cover

nmi_block

Compactness pattern index census block

0.01–0.84
0.02–0.42
0.01–0.54
0.01–0.53
0.02–0.90
0.03–0.73
0.01–0.66
0.01–0.59
0.00–0.85
0.00–0.98
0.02–1.00
0.01–0.09

Cadastral–demographic–economic variables (CDE)
Households
Total number of households per census block
median_hh

Median of household income in $ of census blocks

pop_den

Total population density (per m2)of census blocks

hispnic_den

Hispanic population (of any race) density (per m2) of census blocks

white_den

White population density (per m2) of census blocks

afam_den

African American population density (per m2) of census blocks

amind_den

American Indian population density (per m2) of census blocks

asian_den

Asian population density (per m2) of census blocks

total_area_block

Total area (m2) of census blocks

total_edge_length

Total edge length (m) of census blocks

ave_area

Average area (m2) of parcels of census blocks

According to Eq.11, the NMI of the composited object can then
become:
Xn
2
A
I0 entire
i¼1 i
:
¼ Xn
NMI ¼
Xn
2
I entire
2π
I þ
Ad
i¼1 i
i¼1 i i

ð15Þ

Benefitting from this additive nature, the compactness index of a
composited object can be easily obtained. In a land-use classification
map, the pixels with the same type/value will be collectively summed
up (non-linearly) to compute the overall compactness. With this indicator, both the shape of a single object and the concentration of multiple
objects in a neighborhood can be detected. Using the above approach,
we calculated the area moment and the compactness pattern value for
each land-use type (e.g., building, soil) and the compactness of the
neighborhood boundary itself with the objective to study how well
the composition of these objects impacts the local surface temperature.

182–1378
220–1378
17,500–148,958
22,393–170,179
0.01–0.42
0.01–0.41
0.00–0.87
0.00–0.37
0.00–0.66
0.00–0.39
0.00–0.02
0.00–0.22
0.00–0.02
0.00–0.01
0.00–0.11
0.00–0.14
688–19,865
918–23,315
172.76–5535.11
100.58–8394.19
758–5369
439–6750

Mean

Std. deviation

Std. error mean

1245.84
902.00
4935.30
3090.00
7715.21
5215.50
3760.95
2695.50
125.66
92.37

887.00
660.52
4855.49
2965.75
7298.62
4157.45
3938.98
3483.21
38.50
72.47

38.78
28.19
212.31
126.57
319.14
177.43
172.24
148.66
1.68
4.26

0.17
0.17
0.13
0.11
0.26
0.27
0.13
0.13
0.02
0.11
0.66
0.71

0.00
0.07
0.09
0.06
0.13
0.12
0.10
0.09
0.08
0.27
0.19
0.20

0.00
0.00
0.00
0.00
0.01
0.00
0.00
0.00
0.00
0.01
0.01
0.01

569.27
568.84
64,086.84
65,229.23
0.23
0.13
0.28
0.31
0.11
0.12
0.01
0.01
0.00
0.00
0.01
0.01
6692.27
4483.24
1339.26
501.74
1911.38
1057.19

171.56
190.52
28,492.56
30,681.10
0.16
0.05
0.12
0.02
0.08
0.08
0.02
0.01
0.01
0.01
0.01
0.01
4782.32
3474.23
837.11
658.98
747.90
863.87

7.50
8.13
1245.89
1309.44
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
209.34
148.02
36.60
24.29
32.70
36.87

2.2.5. Cadastral–demographic–economic data
Cadastral–demographic–economic (CDE) variables were collected
or calculated from the City of Phoenix cadastral data, the 2010 U.S. Census, and our land-cover classification. These variables included the following for each census block: the median household income, number
of parcels and average parcel sizes, population density (total population/area), population densities of residents by ethnicity (i.e., Hispanic
of any race and non-Hispanic White, African-American, American
Indian-Alaskan Native, Asian, Native Hawaiian and other Pacific Islanders, and others), the total area of each neighborhood/census
block, and the total edge length of the census block (Table 2).
2.3. Statistical analysis
We joined all data spatially, including land architecture measures,
CDE variables, and the aggregated LST (average per census block) derived from the Landsat image with the point shape file of the spatial distribution of the sampled census blocks. Multiple ordinary least squares
regressions (OLS) were used to determine the effects of land

X. Li et al. / Remote Sensing of Environment 174 (2016) 233–243

239

Fig. 3. An example of multi-patch object and parameters (d = distance and l = moment of inertia) used to compute compactness of multiple patches in one neighborhood (see text). The
images show two neighborhoods with approximately the same area of vegetation but contrasting NMI values: the image on the left has a low NMI value (0.21) for vegetation and the
image on the right has a high NMI value (0.61) for vegetation.

architecture and demographic-economic variables (independent variables) on the LST (dependent variable). In order to process a valid OLS
and avoid type I errors that happen when the error terms of the OLS
are not independent (De Knegt et al., 2010; Li et al., 2012; Lichstein,
Simons, Shriner, & Franzreb, 2002), we further tested each variable for
spatial auto-correlation using global Moran's I. Each variable has a
Moran's I ≤ 0.1, and is not significant at the 0.1 level, indicating the
weak spatial autocorrelation within the independent variables and
thus, greater confidence in the validity of OLS parameter estimates.
Table 2 presents descriptive statistics on the independent variables
used in the regression models.
Initially joined in the regression, the following variables were removed because they were not statistically significant with LST or were
collinear with other variables in the models listed in Table 3: NMI per
census block, average parcel size, total edge of houses, total edge length
of census block, and shape index of the residential neighborhood, total
area of the neighborhoods/census block, compactness pattern index

Table 3
Correlation coefficients of seven models for samples A and B (p b 0.01).

(NMI) of the census block, total number of households in the census
block, and American Indian, Asian, African American, Hispanic and
White population density.
We explored the hypotheses through four scenarios involving seven
models (all the initial independent variables in each model are listed in
Table 2): (1) LST — land composition model; (2) LST — land configuration model; (3) LST — CDE factors model; (4) LST — land
composition + land configuration; (5) LST — land composition + CDE
factors model; (6) LST — land configuration model + CDE factors
model; and (7) LST — land architecture (composition and
configuration) + CDE factors model. We conducted the forward stepwise selection with the least Akaike information criterion with a correction and significance level at α = 0.01 to create the OLS models for each
scenario. The selection of the explanatory variables also involved the examination of the variance inflation factor, a factor to test the
multicollinearity among the explanatory variables. All the models
have variance inflation factor values less than two for the explanatory
variables. Finally, an incremental R2 test was employed to determine
the explained variance gained by adding variables to the base OLS
models.

Model

Sample Sample A
adjust R2
A R2

Sample Sample B
B R2
adjust R2

3. Results

1
2
3

0.15
0.52
0.21

0.14
0.51
0.21

0.17
0.54
0.15

0.16
0.54
0.15

0.54
0.31
0.64
0.65

0.54
0.31
0.63
0.64

0.57
0.27
0.63
0.64

0.56
0.27
0.63
0.64

The seven OLS models are statistically significant (p b 0.001) as are 19
of the 24 added variables in the incremental R2 test (Tables 3 & 4). The
correlation results by model are very close for both samples in the OLS
and incremental R2 test. Significantly, land configuration as a lone variable
or with CDE variables provides stronger associations with mean neighborhood LST than does land composition in both samples of the OLS
models. This observation is further supported by the extremely small
(0.01) improvement of Model 7 (composition + configuration + CDE)

4
5
6
7

Land composition
Land configuration
Cadastral–demographic–economic
variables (CDE)
Composition + configuration
Composition + CDE
Configuration + CDE
Composition + configuration +
CDE

240

X. Li et al. / Remote Sensing of Environment 174 (2016) 233–243

Table 4
Incremental R2 test: the added R2 value to the base models of the models with additional independent variables.
Models with addition of independent variables
Composition +
configuration

Base models

a

Land composition
Land configuration
Cadastral–demographic–economic factors (CDE)
Composition + configuration
Composition + CDE
Configuration + CDE

Composition + CDE

Configuration + CDE
Sample A

Sample A

Sample B

Sample A

Sample B

0.269a
0.015

0.269a
0.029a

0.012

0.202a

0.127a

0.093a

0.257a
0.388a

Sample B
0.185a
0.316a

Composition +
configuration + CDE
Sample A

Sample B

0.521a
0.261a
0.635a
0.232a
0.514a
0.004

0.445a
0.203a
0.334a
0.162a
0.242a
0.018a

Denotes increments significant at α = 0.001 level (F test).

over Model 6 (configuration + CDE) (Table 3). It is additionally supported
by the small correlation increments that land configuration gains by
adding other variables and the much larger increments gained by those
variables when added to land configuration (Table 4). Equally significant,
strong correlations, R2 = 0.54 to 0.65, were generated for various combinations of land architecture and CDE variables (Table 3).
The significant variables in all models for both samples display the
same impact (positive or negative) on LST (Table 5). Similar to the results of Jenerette et al. (2007, 2011), vegetation and swimming pools
decrease LST, whereas buildings, soil, and impervious surfaces increase
it. Importantly to our case, increases in the compactness and concentration of bare soil increase LST, while those characteristics for vegetation

decrease it. In addition, a positive association exists between population
density and LST, whereas an inverse relationship exists between median
household income (Census data designation) of neighborhoods and LST.
4. Discussions
Although the relationships between LST and vegetation and CDE factors have been studied previously (Hanamean et al., 2003; Saunders,
Chen, Crow, & Brosofske, 1998, Smith and Smith & Johnson, 2004;
Jenerette et al., 2007; Martin, 2008; Jenerette et al., 2011; Myint,
Wentz, Brazel, & Quattrochi, 2013), examination of the kind undertaken
in this study has been restricted owing to the paucity of use of high-

Table 5
Impact of independent and/or joint land architecture and cadastral–demographic–economic factors on LST. Descriptions of the independent variables can be found in Table 2.
Sample A

Sample B

Model

Variables

Coefficient

Std. coefficient

Variables

1

area3
area4

0.0079
−0.0036

0.3099
−0.5381

2

NMI1
NMI3
NMI4

129.2561
45.4130
−270.3970

0.2402
0.1428
−0.6606

3
4

total_den
area5
NMI1
NMI3
NMI4

13,780.3044
−0.0027
112.5680
48.6427
−258.5293

0.4592
−0.1724
0.2092
0.1530
−0.6316

5

area3
area4
total_den

0.0010
−0.0030
13,189.1067

0.3947
−0.4460
0.4396

area3
0.0026
0.4307
area4
−0.0018
−0.5462
area5
−0.0001
−0.1478
NMI1
75.0498
0.1543
NMI3
48.3265
0.1741
NMI4
−248.4714
−0.6778
NMI5
16.4240
0.1122
median_hh
−0.0004
−0.3826
area1
−0.0032
−0.2235
area2
0.0021
0.1732
area5
−0.0010
−0.1211
NMI1
66.5219
0.1368
NMI3
48.4170
0.1744
NMI4
−246.3083
−0.6719
NMI5
15.1407
0.1034
area3
0.0025
0.4155
area4
−0.0016
−0.4860
area5
−0.0010
−0.1261
median_hh
−0.0004
−0.3342
NMI1
59.2853
0.1219
NMI3
59.6380
0.2148
NMI4
−236.7961
−0.6460
median_hh
−0.0004
−0.3323
area5
−0.0009
−0.1058
NMI1
52.6507
0.1082
NMI3
59.3241
0.2137
NMI4
−233.5761
−0.6372
median_hh
−0.0004
−0.3244
All independent variables in this table are significant at α = 0.001 level. The dependent variable is the mean
2
neighborhood (census block) LST. The R of each model is in Table 3.
Area and NMI #s refer to
1: building
2: impervious surfaces
3: soil
4: vegetation
5: swimming pool

6

NMI1
99.4659
0.1849
NMI3
64.1074
0.2016
NMI4
−252.1356
−0.6160
median_hh
−0.0004
−0.3601
7
area5
−0.0002
−0.0982
NMI1
91.7325
0.1705
NMI3
64.8338
0.2039
NMI4
−246.4618
−0.6022
median_hh
−0.0004
−0.3387
The following independent variables were removed owing to
multicollinearity:
•Total area of the census block
•Compactness pattern index (NMI) of the census block
•Total edge length of the census block
•Asian population density
•American Indian population density
•African American population density
•Hispanic population density
•Total population density of the census block
•Total number of households in the neighborhood

Coefficient

Std. coefficient

X. Li et al. / Remote Sensing of Environment 174 (2016) 233–243

resolution land-cover data and effective shape/configuration measures
of individual land covers (but see Zheng et al., 2014). Our results provide insights into testable hypotheses about land architecture, point to
characteristics of that architecture that may serve to reduce the SUHI effect during the summer day, and offer insights to studies of neighborhood inequalities with respect to SUHI intensity.
4.1. The implications of land architecture on LST

Hypothesis 1. that land architecture affects summer daytime LST, is supported. Land composition and configuration (or land architecture;
Model 4, Table 3) of the sample neighborhoods in Phoenix considerably
exceeded CDE variables (Model 3, Table 3) in terms of impacts on LST.
The role of configuration is elaborated below (4.2). Here we focus on
composition.
Table 5 lists the role of the independent variables on daytime LST at
the neighborhood level. Consistent results across the models include:
increasing area of vegetation and swimming pools lowers LST, owing
to evapotranspiration, whereas increasing area of bare soils has the opposite effect, owing to rapid reradiation of solar energy, especially in this
hot, arid environment (Table 5). These results are consistent with those
found elsewhere (Martin, 2008, Akbari 2009, Li et al., 2011; Weng & Fu,
2014) and in recent studies of the Phoenix area (Jenerette et al., 2007;
Jenerette et al., 2011; Jenerette et al., 2015; Myint et al., 2013). Buildings
also reradiate heat, but the area of them proved significant only in Sample B of Model 4, surprisingly with a negative LST relationship. Further
research is required to determine why the two samples differed and
the cause of the negative relationship, the latter perhaps influenced by
differences in the albedo of the predominant roof-tops in different
neighborhoods. Surprisingly, the area of impervious surfaces is significant only in the same sample and model as building, but increases LST
as expected. The absence of this variable in our other models requires
further investigation, given its prominence in other studies
(e.g., Zheng et al., 2014).
Hypothesis 2. that land composition affects summer daytime LST more
strongly than land configuration, is not supported. This hypothesis is implied in the various works that have found a strong and significant relationship between the area or area fraction of specific types of land
covers and LST. It is directly supported by at least one study that considered both land composition and configuration in the Baltimore area
(Zhou et al., 2011). In our models, however, land configuration has a
much stronger correlation with LST than does land composition as
noted above (Tables 3 & 4), although its positive or negative impact on
LST depends on the land-cover classes in question (Table 5). This distinction between configuration and composition remains when CDE factors
are added (Models 5 & 6; Table 3). It is noteworthy that our measure of
compactness and concentration (NMI) of building, soil, and vegetation
appear in all models, complete with the expected LST relationships
(building and soil increasing LST and vegetation cooling). The uniformity
in compactness and concentration of swimming pools likely explains the
omission of this land-cover class. As noted for composition, the absence of
impervious surfaces in the results requires investigation, perhaps
reflecting the uniformity among road networks in residential neighborhoods. Finally, consistent with work by Zhou et al. (2011), joining land
composition and configuration (land architecture) improves the correlation with LST, but only marginally (Model 4, Table 3).
4.2. The influence of cadastral–demographic–economic factors (CDE)

Hypothesis 3. that land architecture has a stronger relation to summer
daytime LST than do CDE factors, is supported. Land cover is the proximate

241

source of LST, and similar architectures should produce similar LSTs, all
else held constant. In contrast, CDE variables are more distal to LST because the characteristics of households and neighborhoods are associated with land covers (and land uses) and operate only indirectly on LST
through land cover (Jenerette et al., 2007). For these reasons, land architecture is expected to be more strongly associated with LST. Similar architectures, however, may be related to economic wealth and the ability
to pay for the installation and maintenance of landscaping (e.g., Harlan
et al., 2009) as well as to perceptions of desirable yard-scapes, age of
neighborhood, rules of HOAs, or other such factors (e.g., Larson,
Casagrande, Harlan, & Yabiku, 2009a). In our models, CDE variables account for a much smaller amount of the explained variance (Model 3) in
LST than does land architecture (Model 4) in both samples, as well as
not adding significant R2 increments to composition and configuration
(Table 4).
The only significant CDE variables are population density and median
household income, which increase and decrease LST, respectively. These
results are consistent with findings from other research on Phoenix
(Buyantuyev & Wu, 2010; Harlan et al., 2006; Huang et al., 2011;
Jenerette et al., 2007; Martin, 2008) indicating lower levels of income
and vegetation and higher levels of occupation density concentrated toward the center of the metro-Phoenix SUHI, and increasing levels of incomes either at distance from this center or associated with the use of
turf grass (Connors et al., 2013). All other CDE variables were removed
from the analysis because they were collinear with other independent
variables.
Beyond the hypotheses it is noteworthy that the trends in the
model outcomes for both samples are consistent as are the correlation coefficients, with the exception of slightly lower correlations
for Sample B in Models 3 and 5 (Table 3), suggesting modest differences in CDE variables in the samples. This distinction signals the
need for future exploration of the role of income, preferences, and
HOA rules on land covers. In older parts of Phoenix, higher income
neighborhoods tend to maintain more vegetation, including turf
grass, than lower-income areas (Jenerette et al., 2007, 2011; Larson
et al., 2009a). Newer high-end neighborhoods, however, commonly
require at least some desert- or xeric-scaping (see Chow & Brazel,
2012), which, through the use of the NMI assessment, renders
composition characteristics similar to those displayed in lower
income (largely Hispanic) neighborhoods with large amounts of
bare soil, possibly resulting in similar LSTs. On-the-ground observation of higher income, desert-scaped and lower income, Hispanic
dominated parcels, however, indicates substantial differences in
landscaping.
It is also noteworthy that the results reported here using the more
recent Sobrino et al. (2008) approach are significantly improved relative
to those generated from their former approach (Sobrino et al., 2004).
The subdivision of the NDVIs and NDVIv and the use of the red band
provides a more accurate assessment of LST in the heterogeneous
urban environment, especially in regard to our bare soil class observed
during the morning.
Overall, our findings are consistent with the emerging literature
signaling that UHI effects are affected by the land-cover composition of the urban area (e.g.Georgescu et al., 2011; Zheng et al.,
2014; Zhou et al., 2014), but differs from most of this research by
demonstrating the strong role of land configuration as well. The
implications of corpus of these works is that urban areas, especially
those in warm desert climates, could ameliorate the SUHI effect
and the UHI effect at large by attention to the design of neighborhoods and, perhaps, parcels of new developments and to the
reshaping of existing neighborhoods. The essential elements of
this design focus on increasing the compactness and concentration
of vegetation covers and decreasing the same for buildings and impervious surfaces. Much more work is required, however, to provide metrics and elaboration (e.g., shape) for such configurations.

242

X. Li et al. / Remote Sensing of Environment 174 (2016) 233–243

5. Study limitations
This study is exploratory and the reported results should be
interpreted in terms of the study's limitations. Foremost, the spatial resolutions of the LST (120 m) and land cover (1 m) are incongruent. As a
consequence, the LST constitutes an average of multiple land covers applied to individual land-cover types. The smaller and less compact the
individual land-cover, the more likely its “real” LST deviates from that
applied here. In addition, we assume that the LST data generated from
a single Landsat image is representative of typical June mid-morning
conditions in Phoenix. Future research should examine other summer
days and time, foremost late summer afternoons when maximum temperature is reached, especially affecting impervious surfaces and buildings, and other images of the metropolitan region. Similar examinations
for different seasons are required, as are alternative means of establishing LSTs (e.g.Liu & Zhang, 2011; Mitraka et al., 2013; Sobrino et al.,
2009), including “unmixing” techniques reported by Mitraka et al.
(2013).
In addition, census blocks need not capture the neighborhood as defined by, for example, developers, HOAs, or other factors that may be
used to capture some coherent character of residential parcels, especially in regard to parcel-level land architecture. Significantly, our neighborhood selection was generated by a random sample of single family
residential parcels and did not control for any other parcel types, such
as multiple residential, recreational, or commercial parcels within the
identified residential area. The NMI measure of configuration employed
in this study is not compared to results from the use of other shape measures, such as FRAGSTATS, to determine their compatibility or the influence of different metric sets on the outcomes. Additionally, the
independent variables employed were selected by the research team,
possibly leading to an omitted variable bias. We did explore backward
correction steps in the regressions and found nothing significant. In
the future, however, we will explore quantile regressions. Finally, our
focus on linking as directly as possible land architecture and LST at the
finest spatial grain as possible led us to the use of demographic and socioeconomic data at the U.S. Census block level, rather than the larger
census block group. How this level assessment affected the model results will be examined in the future.
6. Conclusions
Employing fine-resolution (1 m) land-cover data and mesoresolution (120 m) LST data focused on U.S. Census blocks of Phoenix,
AZ, demonstrates the important role of land architecture on the neighborhood SUHI effect. Recognizing its exploratory nature, this study is
the first to indicate that land configuration is more important in its impacts on LST than land composition. That this finding reflects our use of
a new configuration metric, the NMI, deserves research attention and
comparison with other configuration metrics. Consistent with a set of
recent studies, ours demonstrates that the compactness and concentration of vegetation has the largest cooling impact on LST at the neighborhood scale (census block).
Overall, our results suggest that land architecture of neighborhoods
of Phoenix affect LST and the SUHI, and that this architecture is critical to
neighborhood LST. This result suggests that care in the development or
rearrangement in the composition and configuration of the land-cover
of neighborhoods, including at the parcel level, can be used to ameliorate the UHI effect.
Acknowledgments
This project was supported by the National Science Foundation
under Grant No. SES-0951366, NSF DMS Grant No. 1419593 and USDA
NIFA Grant No. 2015-67003-23508; Decision Center for a Desert City
II: Urban Climate Adaptation and Gant No. BCS-1026865, Central
Arizona–Phoenix Long-Term Ecological Research (CAP LTER) and

undertaken through the Environmental Remote Sensing and
Geoinformatics Lab (ERSG) of the CAP-LTER, the Julie Ann Wrigley Global Institute of Sustainability, and School of Geographical Sciences and
Urban Planning. The authors thank the reviewers of earlier versions of
this paper for their insights and comments, as well as input from John
Rogan and Arthur Elmes.

References
Akbari, H., Pomerantz, M., & Taha, H. (2001). Cool surfaces and shade trees to reduce energy use and improve air quality in urban areas. Solar Energy, 70, 295–310.
Arnfield, A.J. (2003). Two decades of urban climate research: a review of turbulence, exchanges of energy and water, and the urban heat island. International Journal of
Climatology, 23, 1–26.
Artis, D.A., & Carnahan, W.H. (1982). Survey of emissivity variability in thermography of
urban areas. Remote Sensing of Environment, 12, 313–329.
Buyantuyev, A., & Wu, J. (2010). Urban heat islands and landscape heterogeneity: Linking
spatiotemporal variations in surface temperatures to land-cover and socioeconomic
patterns. Landscape Ecology, 25, 17–33.
Carlson, T.N., & Ripley, D.A. (1997). On the relation between NDVI, fractional vegetation
cover, and leaf area index. Remote Sensing of Environment, 62, 241–252.
Chander, G., & Groeneveld, D.P. (2009). Intra-annual NDVI validation of the Landsat 5 TM
radiometric calibration. International Journal of Remote Sensing, 30, 1621–1628.
Chander, G., Markham, B.L., & Helder, D.L. (2009). Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM +, and EO-1 ALI sensors. Remote
Sensing of Environment, 113, 893–903.
Chen, X. -L., Zhao, H. -M., Li, P. -X., & Yin, Z. -Y. (2006). Remote sensing image-based analysis of the relationship between urban heat island and land use/cover changes.
Remote Sensing of Environment, 104, 133–146.
Chow, W.T.L., & Brazel, A.J. (2012). Assessing xeriscaping as a sustainable heat island mitigation approach for a desert city. Building and Environment, 47 (170–18).
Chow, W.T.L., Chuang, W., & Gober, P. (2012). Vulnerability to extreme heat in metropolitan phoenix: Spatial, temporal, and demographic dimensions. The Professional
Geographer, 64, 286–302.
Coll, C., Galve, J.M., Sánchez, J.M., & Caselles, V. (2010). Validation of landsat-7/ETM+
thermal-band calibration and atmospheric correction with ground-based measurements. IEEE Transactions on Geoscience and Remote Sensing, 48, 547–555.
Connors, J.P., Galletti, C.S., & Chow, W.T. (2013). Landscape configuration and urban heat
island effects: Assessing the relationship between landscape characteristics and land
surface temperature in Phoenix, Arizona. Landscape Ecology, 28, 271–283.
Dakin, J.P., Pratt, D.J., Bibby, G.W., & Ross, J.N. (1985). Distributed optical fiber Raman temperature sensor using a semiconductor light source and detectors. Electronics Letters,
21, 569–570.
De Knegt, H., van Langevelde, F., Coughenour, M., Skidmore, A., de Boer, W., Heitkönig, I.,
... Prins, H. (2010). Spatial autocorrelation and the scaling of species—environment
relationships. Ecology, 91, 2455–2465.
Georgescu, M., Moustaoui, M., Mahalov, A., & Dudhia, J. (2011). An alternative explanation
of the semiarid urban area “oasis effect”. Journal of Geophysical Research, 116, D24113.
http://dx.doi.org/10.1029/2011JD016720.
Gillespie, A.R., Matsunaga, T., & Rokugawa, S. (1998). Temperature and emissivity separation from advanced spaceborne thermal emission and reflection radiometer (ASTER)
images. IEEE Transactions on Geoscience and Remote Sensing, 36, 1113–1126.
Gober, P., Kirkwood, C.W., Balling, R.C., Jr., Ellis, A.W., & Deitrick, S. (2009). Water planning
under climatic uncertainty in Phoenix: Why we need a new paradigm. Annals of the
Association of American Geographers, 100, 356–372.
Golany, G.S. (1996). Urban design morphology and thermal performance. Atmospheric
Environment, 30, 455–465.
Hanamean, J.R., Jr., Pielke, R.A., Sr., Castro, C.L., Ojima, D.S., Reed, B.C., & Gao, Z. (2003).
Vegetation impacts on maximum and minimum temperatures in northeast Colorado.
Meteorological Applications, 10, 203–215.
Harlan, S.L., Brazel, A.J., Prashad, L., Stefanov, W.L., & Larsen, L. (2006). Neighborhood microclimates and vulnerability to heat stress. Social Science & Medicine, 63, 2847–2863.
Harlan, S. L., Yabiku, S. T., Larsen, L., & Brazel, A. J. (2009). Household water consumption
in an arid city: Affluence, affordance, and attitudes. Society and Natural Resources, 22,
691–709. http://dx.doi.org/10.1080/08941920802064679.
Harlan, S.L., Declet-Barreto, J.H., Stefanov, W.L., & Petitti, D.B. (2013). Neighborhood effects on heat deaths: Social and environmental predictors of vulnerability in
Maricopa County, Arizona. Environmental Health Perspectives, 121, 197–204.
Hawkins, T.W., Brazel, A.J., Stefanov, W.L., Bigler, W., & Saffell, E.M. (2004). The role of
rural variability in urban heat island determination for Phoenix, Arizona. Journal of
Applied Meteorology, 43, 476–486.
He, J.F., Liu, J.Y., Zhuang, D.F., Zhang, W., & Liu, M.L. (2007). Assessing the effect of land
use/land cover change on the change of urban heat island intensity. Theoretical and
Applied Climatology, 90, 217–226.
Hondula, D.M., Davis, R.E., Leisten, M.J., Saha, M.V., Veazay, L.M., & Wegner, C.R. (2012).
Fine-scale spatial variability of heat-related mortality in Philadelphia County, USA,
from 1983–2008: A case-series analysis. Environmental Health, 11, 1–11.
Hondula, D.M., Vanos, J.K., & Gosling, S.N. (2013). The SSC: A decade of climate–health research and future directions. International Journal of Biometeorology, 58, 1–12.
Huang, G., Zhou, W., & Cadenasso, M.L. (2011). Is everyone hot in the city? Spatial
pattern of land surface temperatures, land cover and neighborhood socioeconomic characteristics in Baltimore, MD. Journal of Environmental Management,
92, 1753–1759.

X. Li et al. / Remote Sensing of Environment 174 (2016) 233–243
Jenerette, G.D., Harlan, S.L., Brazel, A., Jones, N., Larsen, L., & Stefanov, W.L. (2007). Regional relationships between surface temperature, vegetation, and human settlement in a
rapidly urbanizing ecosystem. Landscape Ecology, 22, 353–365.
Jenerette, G.D., Harlan, S.L., Stefanov, W., Buyantuev, A., Stefanov, W.L., Declet-Barreto, J., ...
Li, X. (2015). Micro-scale urban surface temperatures are related to land-cover features and residential heat related health impacts in phoenix, AZ USA. Landscape
Ecology. http://dx.doi.org/10.1007/s10980-015-0284-3.
Jenerette, G.D., Harlan, S.L., Stefanov, W., & Martin, C. (2011). Ecosystem services and
urban heat riskscape moderation: Water, green spaces, and social inequality in Phoenix, USA. Ecological Applications, 21, 2637–2651.
Jiménez-Muñoz, J.C., & Sobrino, J.A. (2006). Error sources on the land surface temperature
retrieved from thermal infrared single channel remote sensing data. International
Journal of Remote Sensing, 27, 999–1014.
Kalnay, E., & Cai, M. (2003). Impact of urbanization and land-use change on climate.
Nature, 423, 528–553.
Kane, K., Connors, J.P., & Galletti, C.S. (2014). Beyond fragmentation at the fringe: A pathdependent, high-resolution analysis of urban land cover in Phoenix, Arizona. Applied
Geography, 52, 123–134.
Larson, K.L., Casagrande, D., Harlan, S.L., & Yabiku, S.T. (2009a). Residents' yard choices
and rationales in a desert city: Social priorities, ecological impacts, and decision
tradeoffs. Environmental Management, 44, 921–927.
Larson, K.L., White, D.D., Gober, P., Harlan, S., & Wutich, A. (2009b). Divergent perspectives on water resource sustainability in a public–policy–science context.
Environmental Science & Policy, 12, 1012–1023.
Li, J., Song, C., Cao, L., Meng, X., & Wu, J. (2011). Impacts of landscape structure on surface
urban heat islands: A case study of Shanghai, China. Remote Sensing of Environment,
115, 3249–3263.
Li, J., Wang, Y., Shen, X., & Song, Y. (2004). Landscape pattern analysis along an urban–
rural gradient in the Shanghai metropolitan region. Acta Ecologica Sinica, 24,
1973–1980.
Li, W., Goodchild, M.F., & Church, R. (2013a). An efficient measure of compactness for
two-dimensional shapes and its application in regionalization problems.
International Journal of Geographical Information Science, 27, 1227–1250.
Li, Z. -L., Tang, B. -H., Wu, H., Ren, H., Yan, G., Wan, Z., ... Sobrino, J.A. (2013b). Satellitederived land surface temperature: Current status and perspectives. Remote Sensing
of Environment, 131, 14–37.
Li, W., Church, R.L., & Goodchild, M.F. (2014a). The p-compact-regions problem.
Geographical Analysis, 46(3), 250–273.
Li, X., Myint, S.W., Zhang, Y., Galletti, C., Zhang, X., & Turner, B.L., II (2014b). Object-based
land-cover classification for metropolitan phoenix, Arizona, using aerial photography.
International Journal of Applied Earth Observation and Geoinformation, 33, 321–330.
Li, X., Zhou, W., Quyang, W., & Zheng, H. (2012). Spatial pattern of greenspace affects land
surface temperature: Evidence from the heavily urbanized Beijing metropolitan area
China. Landscape Ecology, 27, 887–898.
Lichstein, J.W., Simons, T.R., Shriner, S.A., & Franzreb, K.E. (2002). Spatial autocorrelation
and autoregressive models in ecology. Ecological Monographs, 72, 445–463.
Liu, L., & Zhang, Y. (2011). Urban heat island analysis using the Landsat TM data and
ASTER data: A case study in Hong Kong. Remote Sensing, 3(7), 1535–1552.
Lo, C.P., Qiattrochi, 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.
Maimaitiyiming, M., Ghulum, A., Tiyip, T., Pia, F., Latorre-Camona, P., Halik, U., ... Caetano,
M. (2014). Effects of green space spatial pattern on land surface temperature: Implications for sustainable urban planning and climate change adaptation. ISPRS Journal
of Photogrammetry and Remote Sensing, 89, 59–66.
Markham, B.L., & Barker, J.L. (1985). Spectral characterization of the LANDSAT thematic
mapper sensors. International Journal of Remote Sensing, 6, 697–716.
Martin, C.A. (2008). Landscape sustainability in a Sonoran desert city. Cities and the environment 1, article 5.
McGarigal, K., & Marks, B.J. (1995). Spatial pattern analysis program for quantifying landscape structure. Gen. Tech. Rep. PNW-GTR-351. US Department of Agriculture. Forest
Service, Pacific Northwest Research Station.
McGarigal, K., Cushman, S.A., & Ene, E. (2012). FRAGSTATS v4: Spatial pattern analysis program for categorical and continuous maps. Amherst, MA: University of Massachusettes
(URL http://www. umass. edu/landeco/research/fragstats/fragstats. html).
Middel, A., Brazel, A.J., Kaplan, S., & Myint, S. (2012). Daytime cooling efficiency and diurnal energy balance in Phoenix, Arizona, USA. Climate Research, 54, 21–34.
Middel, A., Häb, K., Brazel, A.J., Martin, C.A., & Guhathakurta, S. (2014). Impact of urban
form and design on mid-afternoon microclimate in Phoenix local climate zones.
Landscape and Urban Planning, 122, 16–28.
Miller, V.C. (1953). A quantitative geomorphic study of drainage basin characteristics in the
Clinch mountain area. New York: Department of Geology, Columbia University.
Mitraka, Z., Berger, M., Ruescas, A., Sobrino, J.A., Jiménez-Muñoz, J.C., Brockmann, C., &
Chrysoulakis, N. (2013). Estimation of land surface emissivity and temperature
based on spatial–spectral unmixing analysis. In 3rd ME- RIS/(A)ATSR & OCLI-SLSTR
preparatory workshop, Frascati, Italy (pp. 15–19) (October 2013).
Myint, S.W., Wentz, E.A., Brazel, A.J., & Quattrochi, D.A. (2013). The impact of distinct anthropogenic and vegetation features on urban warming. Landscape Ecology, 28,
959–978.
Nichol, J.E. (1996). High-resolution surface temperature patterns related to urban morphology in a tropical city: A satellite based study. Journal of Applied Meteorology, 35,
135–146.
Oke, T.R. (1976). The distinction between canopy and boundary layer heat islands.
Atmosphere, 14, 268–277.

243

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. (1997). Urban climates and global change. In A. Perry, & R. Thompson (Eds.), Applied climatology: Principles and practices (pp. 273–287) (London, Routledge, London).
Rizwan, A.M., Dennis, L.Y.C., & Liu, C. (2008). A review on the generation, determination
and mitigation of Urban Heat Island. Journal of Environmental Sciences, 20, 120–128.
Rozenstein, O., Qin, Z., Derimian, Y., & Karnieli, A. (2014). Derivation of land surface temperature for Landsat-8 TIRS using a split window algorithm. Sensors, 14, 5768–5780
(Basel, Switzerland).
Sailor, D.J. (1995). Simulated urban climate response to modifications in surface albedo
and vegetative cover. Journal of Applied Meteorology, 34, 1694–1704.
Saunders, S.C., Chen, J.Q., Crow, T.R., & Brosofske, K.D. (1998). Hierarchical relationships
between landscape structure and temperature in a managed forest landscape.
Landscape Ecology, 13, 381–395.
Schwartzberg, J.E. (1965). Reapportionment, gerrymanders, and the notion of compactness. Minnesota Law Review, 50, 443.
Sha, M., & Tian, G. (2010). An analysis of spatiotemporal changes of urban landscape pattern in Phoenix metropolitan region. Procedia Environmental Sciences, 2, 600–604.
Shrestha, M.K., York, A.M., Boone, C.G., & Zhang, S. (2012). Land fragmentation due to
rapid urbanization in the phoenix metropolitan area: analyzing the spatiotemporal
patterns and drivers. Applied Geography, 32, 522–531.
Smith, D.L., & Johnson, L. (2004). Vegetation-mediated changes in microclimate reduce
soil respiration as woodlands expand into grasslands. Ecology, 85, 3348–3361.
Snyder, W.C., Wan, Z., Zhang, Y., & Feng, Y.Z. (1998). Classification-based emissivity for
land surface temperature measurement from space. International Journal of Remote
Sensing, 19, 2753–2774.
Sobrino, J.A., Bianchi, R., Paganini, M., Soria, G., Jimenez-Munzos, J., Oltra-Cario, R., et al.
(2009). Dual-use European security IR experiment 2008 (DESIREX 2008). Frascati
(Italy), European Space Agency (ESA): Final Report.
Sobrino, J.A., Caselles, V., & Becker, F. (1990). Significance of the remotely sensed thermal
infrared measurements obtained over a citrus orchard ISPRS. Photogrammetric
Engineering and Remote Sensing, 44, 343–354.
Sobrino, J.A., Jiménez-Muñoz, J.C., Sòria, G., Romaguera, M., Guanter, L., Moreno, J., ...
Martínez, P. (2008). Land surface emissivity retrieval from different VNIR and TIR
sensors. IEEE Transactions on Geoscience and Remote Sensing, 46, 316–327.
Sobrino, J.A., Jimenez-Munoz, J.C., & Paolini, L. (2004). Land surface temperature retrieval
from LANDSAT TM 5. Remote Sensing of Environment, 90, 434–440.
Stabler, L.B., Martin, C.A., & Brazel, A.J. (2005). Microclimates in a desert city were related
to land use and vegetation index. Urban Forestry & Urban Greening, 3, 137–147.
Stone, B., Jr. (2007). Urban sprawl and air quality in large US cities. Journal of
Environmental Management, 86, 688–698.
Stone, B., Jr., & 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.
Streutker, D.R. (2002). A remote sensing study of the Urban Heat Island of Houston, Texas.
International Journal of Remote Sensing, 23, 2592–2608.
Turner, B.L., II, Janetos, A.C., Verburg, P.H., & Murray, A.T. (2013). Land system architecture: Using land systems to adapt and mitigate global environmental change. Global
Environmental Change, 23, 395–397.
Turner, V.K., & Ibes, D.C. (2011). The impact of homeowners associations on residential
water demand management in Phoenix, Arizona. Urban Geography, 32, 1167–1188.
Unger, J. (2004). Intra-urban relationship between surface geometry and urban heat island: Review and new approach. Climate Research, 27, 253–264.
Vanos, J.K., Middel, A., McKercher, G.R., Kuras, E.R., & Ruddell, B.L. (2015). Multiscale surface temperature analysis of urban playgrounds in a hot, dry city. Landscape and urban
planning. (inpress).
Weng, Q., & Fu, P. (2014). Modeling annual parameters of clear-sky land surface temperature variations and evaluating the impact of cloud cover using time series of Landsat
TIR data. Remote Sensing of Environment, 140, 267–278.
Weng, Q., Lu, D., & Schubring, J. (2004). Estimation of land surface temperature–vegetation abundance relationship for Urban Heat Island studies. Remote Sensing of
Environment, 89, 467–483.
Xian, G., & Crane, M. (2006). An analysis of urban thermal characteristics and associated
land cover in Tampa Bay and Las Vegas using Landsat satellite data. Remote Sensing
of Environment, 104, 147–156.
Yu, X., Guo, X., & Wu, Z. (2014). Land surface temperature retrieval from Landsat 8
tirs—comparison between radiative transfer equation-based method, split window
algorithm and single channel method. Remote Sensing, 6, 9829–9852.
Zhang, X., Zhong, T., Feng, X., & Wang, K. (2009). Estimation of the relationship between
vegetation patches and urban land surface temperature with remote sensing.
International Journal of Remote Sensing, 30, 2105–2118.
Zhang, Y., Odeh, I.O.A., & Ramadan, E. (2013). Assessment of land surface temperature in
relation to landscape metrics and fractal vegetation cover in an urban/peri-urban region using Landsat data. International Journal of Remote Sensing, 34, 168–189.
Zheng, B., Myint, S.W., & Fan, C. (2014). Spatial configuration of anthropogenic land cover
impacts on urban warming. Landscape and Urban Planning, 130, 104–111.
Zhou, W., Huang, G., & Cadensasso, M.L. (2011). Does spatial configuration matter? Understanding the effects of land cover pattern on land surface temperature in urban
landscapes. Landscape and Urban Planning, 102, 54–63.
Zhou, W., Qian, Y., Li, X., Li, W., & Han, L. (2014). Relationships between land cover and the
surface urban heat island: seasonal variability and effects of spatial and thematic resolution of land cover data on predicting land surface temperatures. Landscape
Ecology, 29, 153–167.