Theor Appl Climatol (2011) 103:197–211
DOI 10.1007/s00704-010-0293-8

ORIGINAL PAPER

Observing and modeling the nocturnal park cool island
of an arid city: horizontal and vertical impacts
Winston T. L. Chow & Ronald L. Pope &
Chris A. Martin & Anthony J. Brazel

Received: 27 July 2009 / Accepted: 4 May 2010 / Published online: 21 May 2010
# Springer-Verlag 2010

Abstract We examined the horizontal and vertical nocturnal cooling influence of a small park with irrigated lawn
and xeric surfaces (∼3 ha) within a university campus of a
hot arid city. Temperature data from 0.01- to 3-m heights
observed during a bicycle traverse of the campus were
combined with modeled spatial temperature data simulated
from a three-dimensional microclimate model (ENVI-met
3.1). A distinct park cool island, with mean observed
magnitudes of 0.7–3.6°C, was documented for both
traverse and model data with larger cooling intensities
measured closer to surface level. Modeled results possessed
varying but generally reasonable accuracy in simulating
both spatial and temporal temperature data, although some
systematic errors exist. A combination of several factors,
such as variations in surface thermal properties, urban
geometry, building orientation, and soil moisture, was
likely responsible for influencing differential urban and
non-urban near-surface temperatures. A strong inversion
layer up to 1 m over non-urban surfaces was detected,
contrasting with near-neutral lapse rates over urban surfaces. A key factor in the spatial expansion of the park cool
island was the advection of cooler park air to adjacent
W. T. L. Chow (*) : A. J. Brazel
School of Geographical Sciences and Urban Planning,
Arizona State University,
P.O. Box 875302, Tempe, AZ 85287-5302, USA
e-mail: wtchow@asu.edu
R. L. Pope
School of Life Sciences, Arizona State University,
Tempe, AZ, USA
C. A. Martin
Department of Applied Sciences and Mathematics,
Arizona State University,
Mesa, AZ, USA

urban surfaces, although this effect was mostly concentrated
from 0- to 1-m heights over urban surfaces that were more
exposed to the atmosphere.

1 Introduction
Parks and other green spaces located within cities generally
have lower surface and ambient air temperatures compared
to typical urban surfaces such as concrete, tarmac, and
asphalt (e.g., Jauregui 1990/1991; Saito et al. 1990/1991).
This is primarily due to alterations to the surface energy
balance, with relatively more radiant energy partitioned into
the latent heat and/or heat storage terms as opposed to
sensible heat (Oke 1987). Other reasons why parks/green
spaces might be cooler than the surrounding structures
include the greater direct shading from vegetation with
large canopies that reduces surface temperatures, a higher
sky view factor relative to urban surfaces allowing for
unimpeded radiative cooling to the sky, and a lack of heat
from combustion sources (Spronken-Smith and Oke 1999;
Shashua-Bar et al. 2009).
This phenomenon of patchy, cooler areas within the
urban mosaic is often termed the park cool island (PCI).
Increasingly, urban planners are aware that managed PCIs
through the use of urban green spaces could be a useful
applied tool towards mitigating detrimental impacts stemming from the urban heat island (UHI), especially within
tropical and sub-tropical cities of large spatial extent (e.g.,
Emmanuel 2005). Past research of PCI effects mostly relied
on observational studies focusing on the spatial distribution
of temperatures with respect to adjacent urban surroundings,
with much fewer studies either examining the vertical form
of PCI or modeling the impacts of green spaces on the urban
environment.

198

In this study, we examined both horizontal and vertical
nocturnal temperature differences at different heights
between a non-urban, largely vegetated surface and its
urban surroundings. We analyzed horizontal micro-scale
temperature fields and vertical (i.e., lapse rate) profiles
within a study area located in an arid desert climate. We
obtained PCI intensities through two methods: (a) direct,
relatively high frequency (1 s) temperature measurements
obtained from a bicycle traverse over different land use
types and (b) modeling study area temperatures through a
three-dimensional numerical micro-meteorological model
(ENVI-met 3.1) with inputs adapted to local soils and
vegetation. A secondary objective was to evaluate the
accuracy of modeled vs. observed temperatures from the
traverse and a nearby meteorological station. This study
aims to improve the understanding of PCI effects for a
desert city, of which research is relatively deficient
compared to other urban climates.

2 Literature review
PCI can be measured either through surface or air
temperature differences between urban surroundings with
the park or green space (ΔTu–p). Distinct differences of
surface temperatures between parks and adjacent urban
areas often result in dynamic variations of ΔTu–p(surface)
(Roth et al. 1989). Variations of ΔTu–p(air) are strongly
coupled with underlying surface types at micro-scales,
although greater turbulence and advection reduce magnitudes of surface air thermal coupling, especially at distinct
borders between surface types at micro-scale levels (Jansson
et al. 2007). Higher wind speeds and the resulting local-scale
turbulent mixing also dilutes the maximum magnitude of
ΔTu–p(air) relative to ΔTu–p(surface) and may also expand the
influence of the PCI beyond the physical boundaries of the
park (Oke 1989; Eliasson and Upmanis 2000). Magnitudes
of mean ΔTu–p(air) range from 1.5 to 4°C (Sham 1987;
Upmanis et al. 1998; Jonsson 2004), but maximum
intensities can be as large as ∼7°C during summer nocturnal
periods in Sacramento, CA, USA (Spronken-Smith and Oke
1998).
The size of the park or green space, topography, wind
speeds, and vegetation types are factors influencing PCI
magnitudes and development. Larger parks are generally
associated with greater PCI intensities (Barradas 1991;
Jauregui 1990/1991), and magnitudes of ΔTu–p(air) inversely
correlate to wind speeds (Spronken-Smith 1994). Topographic variations potentially affect regional airflow into
parks at different spatial scales, such as by large-scale
nocturnal katabatic drainage from mountains to a valley city
that reduces urban air temperatures adjacent to parks (Kirby
and Sellers 1987). Vegetation types in parks also affect the

W.T.L. Chow et al.

timing of maximum PCI (Spronken-Smith and Oke 1998),
with (a) parks with substantial tree canopies having greatest
cooling impacts during the afternoon, (b) mixed use, garden,
or savannah parks having PCI maxima around sunset, and
(c) open grass parks having observed PCI maximums around
sunrise.
Of particular interest are geographical differences in
ΔTu–p within cities located in hot arid vs. other climates.
The scarcity of surface and atmospheric moisture, combined with intense diurnal surface radiative exchanges, is a
significant qualitative difference affecting PCI development
in desert cities (Pearlmutter et al. 2007). Latent heat fluxes
from surface evapotranspiration in desert cities are much
lower relative to urban areas in other climates, but this
strongly depends on irrigation availability for parks in these
cities. For instance, daytime summer latent heat fluxes in
Tucson, AZ, USA, where landscape irrigation of suburban
residential green spaces is prevalent, accounted for 25–35%
of the daytime surface energy balance (Grimmond and Oke
1995). In contrast, a similar contribution to the surface
energy balance of Mexicali, Mexico, which lacked water
availability for irrigation, was only ∼10% (Garcia-Cueto et
al. 2003). Irrigation of green spaces is therefore important
for daytime cooling through evapotranspiration within
desert cities (e.g., Pearlmutter et al. 2009), especially
combined with windy conditions that expand the “oasis
effect” of PCI (Spronken-Smith et al. 2000). Increases in
surface and soil moisture from irrigation also alter the
surface thermal admittance of parks (μ), an important
property measuring the ability of a surface to accept or
release heat (Oke 1987). Higher (lower) soil moisture
content results in higher (lower) μ, resulting in reduced
(increased) urban vs. green space near-surface temperature
differences (Chow and Roth 2006). As nocturnal park
cooling appears to be more dependent on surface conduction and radiation processes as opposed to evapotranspiration (Oke et al. 1991; Spronken-Smith and Oke 1999), the
timing of irrigation of parks post-sunset potentially affects
diurnal PCI development and nocturnal PCI intensities in
arid cities.
The study of PCI impacts within arid urban areas is
important, given the increasing population and urban
development seen in several Middle-Eastern and North
American desert cities in recent years, including Phoenix,
AZ, USA (Gober 2006). It is also essential to examine the
methods and techniques of cooling urban landscapes (e.g.,
Shashua-Bar et al., 2009), which include increasing and
maintaining urban green spaces. It is thus unsurprising that
urban parks have a great potential in mitigating detrimental
nocturnal UHI impacts in desert cities, as well as reducing
overall urban energy consumption (Kurn et al. 1994),
especially if urban vegetation are well maintained (Muellar
and Day 2004). In Phoenix, the placement, design and

Observing and modeling the nocturnal park cool island of an arid city: horizontal and vertical impacts

upkeep of urban green spaces are important in reducing
UHI intensities and in potentially improving human health
and quality of life without excessively consuming natural
resources. Thus, they are a key component in developing
sustainable urban landscapes for this large metropolitan
area (Stabler et al. 2005; Martin 2008).
Previous study methodologies investigating on the
thermal impacts of urban parks generally used data
obtained from station temperatures and/or vehicular
traverses (e.g., Sham 1987; Upmanis et al. 1998; Jonsson
2004) or through hardware scale modeling (SpronkenSmith and Oke 1999; Pearlmutter et al. 2009). These
methods usually focus on examining both spatial form and
intensities of ΔTu–p across different urban land use types.
Much fewer studies (e.g., Jansson et al. 2007), however,
examine the vertical impacts of PCI, i.e., analyze temperature profiles or lapse rates over different surface types
within the urban canopy layer (i.e., from surface to mean
roof height). Compared to these research methods, there has
also been hitherto less work into modeling ΔTu–p,
especially at micro-spatial scales (i.e., linear and areal extents
ranging from approximately 1 to 102 m or m2) (Oke, 2004).
Some research of micro-scale numerical modeling of urban
vegetation impacts exist, although these are limited to
simulations adding individual street trees in Colombo, Sri
Lanka, and in downtown Phoenix (Emmanuel and Fernando
2007) or via increasing rooftop or vertical vegetation density
in Singapore (Wong and Jusof 2008).

3 Materials and methods

199

height trees of similar height, e.g., blue palo verde
(Parkinsonia florida), southern live oak (Quercus virginiana), and desert fan palm (Washingtonina filifera), lined
several roads and boulevards, especially on the northern and
southern boundary areas. These vegetation species are also
commonly found in residential areas for landscaping.
Buildings in the study area varied between 5 and 30 m
above surface level (a.s.l.), with a mean height of ∼10 m. A
section of several hardcourt tennis courts and an Olympicsized swimming pool were located to the west and south of
the field.
Using ArcView 9.3 GIS, we created a three-dimensional
(3-d) model by digitizing an aerial photograph of the study
area as well as from fieldwork reconnaissance (Fig. 1). This
model was subsequently used in traverse route design
(Fig. 2), with the objective of extensive temperature
sampling coverage over different surface types, such as
urban (e.g., asphalt, concrete, and tennis court) and nonurban (e.g., xeric and lawn grass) surfaces categorized
within the study area. In addition, temperatures on surfaces
confined within a street or pavement canyon between
buildings (i.e., urban canyons) were classed in a separate
category. To quantify the aspect ratio of building height vs.
spacing in these urban canyons, sky view factors were
estimated using 3DSkyView, an ArcView 3.2 GIS program
extension written by Souza et al. (2003). This program has
been shown to produce more accurate results than other
geometric estimation techniques, with similar, if slightly
elevated, results when compared with actual field data.
Minimum (mean) sky view factors in these canyons were
0.5 (0.7), indicating low- to medium-density urban development in the study area.

3.1 Study area
3.2 Observed temperature data
The study was conducted in the Tempe campus of Arizona
State University, located within the Phoenix metropolitan
area (33.416° N, 111.931° W, 355.4 m above sea level). A
23-ha section of the campus centered on the Student
Recreation Complex (SRC) field was selected as our study
area (Fig. 1). The ∼3-ha-sized field is a level urban green
space mostly covered by lawn grass (0.05–0.1-m heights)
of either bermuda (summer) or rye (winter) species, which
are similar to grasses used for residential landscaping
throughout suburban Phoenix. Parts of the field were
exposed bare sandy–loam soil, especially in the center and
along three of its four corners. The field was bounded by a
wide variety of medium height trees (5–10 m) (e.g., Seville
sour orange, Citrus aurantium, and Chinese pistache,
Pistacia chinensis) with a wide canopy and dense foliage
along the north, south, and east field perimeters and by taller
trees (20–25 m) (e.g., coolibah, Eucalyptus microtheca, and
Mexican fan palm, Washingtonia robusta) along the west
field perimeter. Other smaller lawn grass plots with medium-

An instrument setup similar to Sun et al. (2009) was
utilized to obtain and geo-code spatial temperature data. We
used the TR-72U temperature/relative humidity sensor–
datalogger (T&D Corporation), which had a specified
resolution (accuracy) of 0.1°C (± 0.3°C), and the BTQ1000 GPS travel recorder (Qstarz) which noted geographical coordinates of latitude, longitude, and altitude with a
reacquisition rate of ∼1 s. Four TR-72U sensors were
attached to a metal pole before being vertically mounted to
a bicycle at 0.01-, 1-, 2-, and 3-m heights a.s.l. As these
sensors were not aspirated, it was also important that
measurements occurred when the air was not completely
stagnant during the study to minimize lags in sensor
response. The GPS system was also attached to the pole
to record geographical coordinates of every temperature
sampling point. Each TR-72U sensor and the Q1000
recorder were synchronized and programmed to sample
temperatures and geographical coordinates at 1-s intervals.

200

W.T.L. Chow et al.

Fig. 1 March 2007 aerial
photograph of the 23-ha study
area (demarcated by bold line)
within the Arizona State
University—Tempe campus
(top) and 3-d model of the study
area generated by Arcview
9.3 (bottom)

The TR-72U sensors were subject to pre- and post-fieldwork
calibration to ensure that measurements were within the
manufacturer’s specifications.
The traverse occurred during the early morning (0530–
0630 hours) on October 28, 2007, as large magnitudes of
ΔTu–p are expected a priori during minimum temperature
conditions given the open grass park that constitutes most
of the SRC field. Cloud-free skies and generally low E
winds of 0.4–0.8 ms-1 at 10 m above surface level were
recorded during the traverse, which minimized the meteorological impacts on both the PCI and UHI (SpronkenSmith and Oke 1998). These ambient weather conditions

near the study area were recorded at the Tempe meteorological station of the Maricopa County Air Quality
Department (MCAQD). The station was directly adjacent
to an ∼8-ha park mostly consisting of irrigated lawn grass
and was located within a low-density suburban residential
neighborhood with predominant mesic landscaping ∼670 m
SSW of the SRC field. We therefore assumed the station
data to be representative of the general near-surface
meteorological conditions of our study area. Hourly
temperatures from the MCAQD Tempe station were used
to obtain time-series data for the study area. The study took
place during a night when the SRC field was not irrigated,

201

Observing and modeling the nocturnal park cool island of an arid city: horizontal and vertical impacts

Fig. 2 Map of traverse route and classification of different surface types within the study area

although we could not ascertain if other smaller green
spaces within the study area were watered.
We recorded ∼3,100 temperature datum points per sensor
during the traverse, resulting in a total of ∼12,400 points for all
four sensors. These data were consequently geo-referenced
Table 1 Surface characteristics
for classification of traverse data

Land use type

into the previously created GIS model for data analysis. Each
datum point could therefore be categorized with its respective
surface classification (Table 1). Mean temperatures for each
sampling point were derived, together with mean vertical
lapse rates for each urban and non-urban surface type.

Description of surface over which sampling occurred

No. of traverse
sampling points

Dark-colored road surfaces, mainly used by vehicular traffic
Lighter-colored pavement used by pedestrians
Hardcourt green colored surface composed of asphalt with rubber
compound for cushioning
Urban surface confined within buildings with restricted exposure
to sky; minimum (mean) sky view factor is 0.5 (0.7)

1,110
1,021
243

Urban surfaces
Asphalt
Concrete
Tennis courts
Urban canyons
Rural surfaces
Xeric
Lawn

Surface with exposed sandy–loam soil without lawn cover
Surface planted with lawn grass (bermuda or ryegrass species)

281

74
612

202

W.T.L. Chow et al.

Fig. 3 Part of the area input file
for ENVI-met simulations
which was based on Figs. 1
and 2. Individual grid cell size
was 2.5 × 2.5 × 5 m

3.3 Microclimate model simulation
We selected the ENVI-met urban microclimate model
(version 3.1, available as freeware for MS Windows at
http://www.envi-met.com) to simulate soil, surface, and
vegetation interactions within the urban canopy layer (i.e.,
below mean roof height). It is a 3-d, non-hydrostatic,
prognostic numerical model based on fundamental fluid
dynamics and thermodynamic laws (Bruse and Fleer 1998).
Typical horizontal resolutions are 0.5–10 m, and simulations are normally run for 24–48 h, with a maximum time
step of 10 s (Bruse, 2010). Analysis of small-scale physical
interactions between individual buildings, surfaces, soils,

and vegetation is thus possible at these modeled spatial and
temporal resolutions.
The model requires a user-specified area input file that
defines the 3-d geometry of the model environment such as
individual buildings and building heights, vegetation, soil
and surface types, etc. (Fig. 3), which were derived from
Figs. 1 and 2. Although simulated concrete and asphalt
surfaces in Fig. 2 could be input into the model, a modeled
surface type similar to the tennis court surface could not be
found; instead, grid cells with this surface were classed as
concrete surfaces. We also selected a relatively high
resolution for individual grid cells (2.5 × 2.5×5 m) to
complement the observed temperature data resolution. The

Table 2 Observed LAD data for selected trees within the study area
Common name

Species

LADmax (standard error) (m2 m−3)

LADmax height (m)

Mean tree height (m)

Bottle tree
Seville sour orange
Coolibah
Shamel ash
White mulberry
Thornless mesquite
Blue palo verde
Southern live oak
African sumac
European olive
Chinese pistache
Aleppo pine
Chinese elm

Brachychiton populneus
Citrus aurantium
Eucalyptus microtheca
Fraxinus uhdei
Morus alba
Prosopis hybrid
Parkinsonia florida
Quercus virginiana
Rhus lancea
Olea europaea
Pistacia chinensis
Pinus halepensis
Ulmus parvifolia

3.23
4.83
2.59
2.46
3.77
2.92
1.85
3.02
4.39
3.39
2.79
2.68
1.68

2.25
0.5
2.75
3
2.75
2.5
2
3
2.25
2.25
1.75
2
2.75

7.75
5
8.75
12
9
9
7.5
9.25
8.25
8.25
4.5
4.75
8.5

(0.12)
(0.18)
(0.13)
(0.35)
(0.14)
(0.11)
(0.13)
(0.09)
(0.24)
(0.14)
(0.11)
(0.18)
(0.12)

203

Observing and modeling the nocturnal park cool island of an arid city: horizontal and vertical impacts

area input file was nested within four nesting grids (with
the predominant soil type) for numerical stability during
model runs. We initially used the vegetation database
included in the model defaults when configuring the area
input file; however, these database characteristics, such as
for normalized vegetation leaf area density data (LAD),
were based on temperate forest and crop stands (M. Bruse,
personal communication), which have little relevance to the
vegetation types found within an arid city. We therefore
surveyed vegetation within the study area to empirically
estimate leaf area density based on field observations with a
Li-Cor LAI-2000 plant canopy analyzer (http://www.licor.
com/) and derived vertical leaf area density for each
vegetation species based on the method proposed by Lalic
and Mihailovic (2004) (Table 2). These database alterations
subsequently allowed for input of local vegetation into
ENVI-met.
Apart from these vegetation data, the model also
required local soil, meteorological, and building input
parameters (Table 3) for model initialization. Temperature
and relative humidity (RH) at 2 ma.s.l., 10 ma.s.l. wind
speed, and direction data from the MCAQD Tempe station
were used as initial model inputs, together with soil
temperature and soil RH data from the Mesa station of the
Arizona Meteorological network (hourly data are available
at http://ag.arizona.edu/AZMET/). This station was
∼6.5 km away from the SRC field and assumed to have
similar soil profiles to the study area (i.e., sandy–loam
soils). Specific humidity at 2,500 ma.s.l. was estimated via
data from the NCEP/NCAR reanalysis project at the
Table 3 Selected input
parameters for ENVI-met
simulations

NOAA/ESRL Physical Sciences Division (http://www.cdc.
noaa.gov/data/reanalysis/reanalysis.shtml). Lastly, building
interior temperatures, mean heat transmission (i.e., thermal
conductivity divided by mean wall or roof width), and roof/
wall albedo values were estimated from existing table-base
data (e.g., Oke 1987) and from field reconnaissance
observations.
We ran a model simulation for 24 h starting from sunrise
(0600-0600 hours; Oct 27–28, 2007) with updated surface
data every 60 s. We subsequently analyzed model output at
0530 hours, coinciding with the start of the bicycle traverse.
Modeled potential temperatures at 0 (surface), 1, 2, and 3 m
were extracted as geo-referenced (x–y coordinate reference)
text-based .dat files using the accompanying LEONARDO
program (V 3.75 Beta) in the ENVI-met software package
and were converted to absolute temperatures using
Poisson’s equation. Using the ASCII to raster tool in
ArcView 9.3 GIS software, these files were converted and
entered as raster files with 2.5-m x–y grid cells. The raster
files were geo-referenced with the other map layers; warm
features of the model output, i.e., buildings, were prominent
in the raster and created excellent spatial references. This
geo-referencing had an average root mean square error
(RMSE) of 1.45 m; this error was both unavoidable and
systematic and is the result of 2.5-m gridded layers being
geo-referenced to a discrete continuous map surface.
Following geo-referencing, the raster grid was converted
to a polygon feature class so that it could be spatially joined
with the observed sample point layer. In this manner, each
of the ∼3,100 datum points at the 0.01-, 1-, 2-, and 3-m

Category

User input during simulations

Meteorological inputs
Wind speed and direction 10 m above ground
Approximate roughness length of study area
Specific humidity at 2,500 ma.s.l.
Relative humidity at 2 ma.s.l.
Initial atmospheric temperature

0.8 ms−1/85°
0.5 m
2.75 g/kg
20%
290.3 K

Soil inputs
Initial soil temperature at upper layer (0–20 cm)
Initial soil temperature at middle layer (20–50 cm)
Initial soil temperature at lower layer (below 50 cm)
Relative humidity, upper layer
Relative humidity, middle layer
Relative humidity, lower layer
Building inputs
Building interior temperature
Mean heat transmissiona of walls
a
Estimated thermal conductivity
of building material (Wm−1 K−1 )
divided by mean wall thickness
(m)

Mean heat transmission of roofs
Mean wall albedo
Mean roof albedo

293 K
294 K
295 K
25%
30%
35%
303 K
1.7 Wm−2 K−1
6.0 Wm−2 K−1
0.3
0.4

204

elevations could be directly compared with the corresponding
ENVI-met model output.

4 Results
4.1 PCI effect—observations and modeling
Both data from the bicycle traverse and from the ENVI-met
simulations illustrated a distinct PCI that developed over
the SRC field. Observed temperatures from 0.01- and 2-m
heights showed the SRC field and the adjacent west tennis
court as having cooler temperatures compared to the
surrounding buildings, pavements, and urban canyons
(Fig. 4). Temperatures at 2-m heights were also generally
warmer when compared to the surface, especially over the
lawn and xeric surfaces. The approximate maximum
difference in temperature at 0.01 m (2 m) between the
Fig. 4 Temperatures observed
from traverse for 0.01- (top) and
2-m heights (bottom) from
0530–0630 hours

W.T.L. Chow et al.

SRC field and its urban surroundings was ∼5.8°C (3.1°C),
although there were relatively large variations in temperature over the urban surfaces adjacent to the field. We
observed notable temperature differences within the field,
with a cooler “core” in its center compared to its warmer
perimeter adjacent to concrete or asphalt surfaces. Slightly
cooler urban temperatures were observed W of the field and
tennis courts, possibly indicating the effect of advection
from easterly wind flow, with urban canyon temperatures in
the western section of the study generally being ∼1°C
cooler compared to the eastern section.
Simulated temperatures from ENVI-met also showed a
PCI, albeit with lower approximate maxima of 3.6°C and
2.9°C at surface and 2-m heights, respectively. Cooler
temperatures were observed over the field and tennis courts,
and a similar shift of cold air from advection from E to W
was present (Fig. 5). Temperatures at the surface were
generally cooler than at 2-m heights, with lower temper-

Observing and modeling the nocturnal park cool island of an arid city: horizontal and vertical impacts

205

Fig. 5 2-d profiles of ENVImet temperature simulations for
surface (a) and 2-m levels (b) at
0530 hours

atures at the center of the SRC lawn field as compared to its
xeric surfaces. Colder temperatures were present along the
eastern part of the study area, in contrast to observed
temperature data, but this was most likely a consequence of
the model’s nesting grids of sandy–loam soils along its
boundaries coupled with the constant E wind flow during
the simulation period. Pockets of warmer air were observed
within urban canyons in the western part of the study area,
especially within N–S-orientated canyons, which possibly
resulted as a lee effect from buildings acting as windbreaks
and impeding the advection of cooler air from the park.
4.2 Variations of PCI intensity and vertical lapse rates over
different surfaces
We examined variations in ΔTu–p over different heights for
both modeled and observed data from the surface up to
3 m. Magnitudes of mean PCI were defined as the

difference in temperatures between each mean urban (i.e.,
asphalt, concrete, tennis courts, and urban canyons) vs.
non-urban surface (i.e., lawn or xeric) for all heights
(Table 4). Mean observed PCI intensities ranged from 1.4
to 3.8°C with magnitudes decreasing significantly with
increasing observation height for all surface types. For
example, mean ΔTurban-non-urban, derived by the difference
between mean urban and non-urban surfaces, was 2.6°C
less at screen-level heights of 2 m vs. 0.01 m. Mean ENVImet PCI intensities were, however, generally lower in
magnitude at all comparative heights, especially between
the surface and 1-m heights, although qualitative differences (i.e., ranking of PCI intensities between surface
types) are largely similar between observed and modeled
ΔTu–p.
Different surface types influenced urban vs. non-urban
lapse rates. Vertical distribution of mean temperatures
measured over asphalt, concrete, and within urban canyons

206
Table 4 Mean PCI (ΔTu–p)
over different surface types from
observed and modeled data

W.T.L. Chow et al.
PCI by surface type (m)

0

Mean temperature from traverse observations (°C)
Concrete—xeric
Concrete—lawn
Asphalt—xeric
Asphalt—lawn
Urban canyon—xeric
Urban canyon—lawn
Tennis court—xeric
Tennis court—lawn
Mean urban − mean non-urban
Mean temperature from model simulations (°C)
Concrete—xeric
0.43
Concrete—lawn
0.83
Asphalt—xeric
0.49
Asphalt—lawn
0.89

Heights are above surface level

Urban canyon—xeric
Urban canyon—lawn
Tennis court—xeric
Tennis court—lawn
Mean urban − mean non-urban

showed insignificant changes at all heights (Fig. 6);
however, similar profiles over tennis court surfaces were
generally ∼1°C less compared to other urban surface types.
This influences PCI intensity, with ΔTtennis courts−xeric equal
to zero at both 2 and 3 m (Table 4). There were also
important differences of PCI intensity with respect to nonurban surfaces, as ΔTu–p having consistently greater
magnitudes when measured over lawn as opposed to xeric
surfaces, with higher intensities documented closer to the
ground. An analysis of the observed vertical lapse rates
showed significant differences of stability between urban
and non-urban surfaces. Asphalt, concrete, tennis court, and
urban canyon surfaces generally had near-neutral or slightly
stable conditions from 0 to 2 m, but non-urban xeric and
lawn surfaces had a strong inversion observed at those
heights, especially from 0 to 1 m (Fig. 6). Above 2 m,
however, all surfaces had neutral to slightly unstable air
capping the near-surface inversion.
Modeled lapse rates did not show the stable stratification
of observed temperatures; instead, lapse rates were neutral
for all surface types from 0 to 3 m, with a much smaller
range especially at lower heights. The order of warmer to
colder temperature profiles was largely similar to the
observed vertical profiles, with lawn surfaces being coldest
and urban canyons being warmest. Of note, however, is that
the modeled vertical temperature profile within urban
canyons had higher temperatures (0.4°C) when compared
to asphalt or concrete surfaces, whereas such differences in
temperature were negligible during the traverse observations.

0.78
1.18
0.15
0.55
0.82

0.01

1

2

3

2.4
3.8
2.3
3.7
2.3
3.7
1.4
2.8
3.6

1
1.3
1.1
1.4
1.1
1.4
0.2
0.5
1.4

0.7
1.1
0.7
1.1
0.7
1.1
0
0.4
1.0

0.5
0.9
0.5
0.9
0.4
0.8
0
0.4
0.7

0.37
0.75
0.45
0.83

0.34
0.70
0.41
0.77

0.32
0.65
0.38
0.71

0.75
1.13
0.11
0.49
0.75

0.72
1.08
0.08
0.44
0.70

0.70
1.03
0.05
0.38
0.65

4.3 Evaluation of ENVI-met simulations
We evaluated the model’s performance in terms of accuracy
with observed data in two ways. First, modeled spatial
temperatures at 0530 hours were compared with observed
traverse data across similar heights. Second, time-series
data categorized under different land use categories at 2 m
were averaged and subsequently compared with similar
data from the nearby MCAQD Tempe station. In this study,
we used several absolute quantitative difference measures
of mean bias error (MBE), mean average error (MAE),
RMSE, and its derived systematic (RMSES) and unsystematic (RMSEU) components (Willmott 1982). These error
indices, together with the relative, dimensionless difference
measure of the index of agreement (d), are used in
conjunction to evaluate the accuracy of ENVI-met temperatures, where accuracy is defined as the degree to which
predicted observations (P) approach magnitudes of actual
observed data (O). These measures are preferred to simple
correlation measures (e.g., r, r2) which are completely
insensitive to additive and proportional differences that
potentially exist between P and O and thus may not be
consistently related to model prediction accuracy. For
instance, it is possible for low or even negative measures
of r simultaneously occurring with relatively small differences between magnitudes of P and O, if the mean of either
one of the variables is stationary and its deviations about
that mean are unsystematic (Willmott 1984). Further,
utilizing difference or error measures allows for the

Observing and modeling the nocturnal park cool island of an arid city: horizontal and vertical impacts

207

Fig. 6 Mean vertical temperature profiles for each surface
from traverse data taken at 0.01,
1, 2, and 3 m above surface
level (top) and corresponding
ENVI-met data at 0, 1, 2, and
3 m above surface level
(bottom). Urban surface
temperatures are the mean
profiles of temperatures taken
over asphalt, concrete, tennis
court, and urban canyon
surfaces. Non-urban surface
temperatures comprise of the
mean xeric and lawn profiles

evaluation of systematic and unsystematic errors within the
model, which are not readily available with simple
correlation plots of P vs. O.
Four model levels corresponding to the nearest traverse
sensor height were selected for comparison of spatial
temperature fields: 0 m (P) with 0.01 m (O), 1 m (P) with
1 m (O), 2 m (P) with 2 m (O), and 3 m (P) with 3 m (O)
data. The comparison between surface and 0.01-m data may
be complicated by observed raised temperature minima
over a variety of surfaces (Oke 1970; Geiger et al. 2003),
but we assume that these differences are relatively
insignificant between 0 and 0.01 m, especially over
vegetated areas (e.g., short grass lawns) that have raised
active surfaces>0.01 m generally documented under clear
and windless weather conditions (Oke 1987).
The MBE for all comparisons show that the model
generally under-predicted temperatures by 0.74–1.38°C

(Table 5). MAE ranged from 1.16 to 1.47°C, with a larger
error occurring closer to the surface. The magnitudes of
RMSE for all heights were relatively low (<1.7°C), but a
decomposition of RMSE indicates that a systematic error
predominates. A probable source of this would be the
cooler temperatures consistently simulated along the eastern,
urban part of the study area (Fig. 5), which is in contrast to
the warmer temperatures observed during the traverse
(Fig. 4). A possible explanation could lie in how the model
parameterizes lateral boundary conditions for both surfaces
and turbulent inflow. The use of open lateral boundaries,
coupled with the study area being nested within grids of
sandy–loam soils, allows for the constant flow of easterly
winds to advect cooler air into the model’s eastern boundary.
Further, the observed temperature data may be subject to
sensor lag in some areas of stagnant turbulence, which
affects a key assumption that observed data are error-free in

208
Table 5 Difference measures of
ENVI-met model temperatures
with observed temperature data
at 0530 hours

a
Surface model data are compared
with observed 0.01 m data

W.T.L. Chow et al.
Height of ENVI-met model simulation
Difference measure (unit)
MBE (°C)
MAE (°C)
RMSE (°C)
RMSES (°C)
RMSEU (°C)
d (dimensionless)

the statistical analysis via difference measures. Another
potential error in the observed data could lie with the
systematic errors from the spatial join process described in
“Section 3.3”, which could propagate additional unavoidable
systematic errors. Despite these factors, the relatively low
error indices, combined with acceptable magnitudes of d
(0.4–0.57) given the systematic errors present, suggest that
the model reasonably approximates the spatial distribution of
temperatures and derived ΔTu–p.
The time-series data for mean modeled 2-m temperatures
over all surfaces compared favorably with 2-m temperature
data obtained from the MCAQD Tempe weather station
(Fig. 7). The model generally under-predicted (overpredicted) nocturnal (daytime) temperatures, which are
similar patterns documented in previous applications of
ENVI-met (e.g., Emmanuel and Fernando 2007). During
the day, mean temperatures over asphalt surfaces were
systematically and significantly higher vs. that of other
surfaces, probably due to its larger (lower) heat capacity
(heat conductivity) relative to other surface materials. It is
also notable that these modeled mean near-surface temperatures over asphalt cooled more than other urban surfaces
after sunset. RMSES (2.23°C) was slightly greater than
Fig. 7 Mean 2-m temperatures
from ENVI-met simulations
over different land use types
(see Table 1) and from MCAQD
Tempe meteorological station
over a 24-h period from Oct
27 to 28, 2007. Difference
measures are for mean modeled
2-m temperatures across all
surfaces vs. MCAQD 2-m
temperatures

Surfacea
−0.74
1.47
1.63
1.49
0.67
0.57

1 m
−1.33
1.33
1.50
1.32
0.71
0.47

2 m
−1.38
1.38
1.53
1.37
0.69
0.41

3 m
−1.16
1.16
1.31
1.24
0.69
0.40

RMSEU (1.95°C), which could be a consequence of several
limitations of ENVI-met, such as its inability to (a)
dynamically simulate heat storage for building walls and
roofs by having constant building indoor temperatures with
no thermal mass and (b) simulate regional-scale thermal
and turbulence exchanges that may directly influence
micro-scale climates. The latter point can be illustrated
with the sudden increase in temperature in the MCAQD
data from 0300 to 0400 hours, which could have resulted
from the advection of warmer air from adjacent urban areas
due to the nocturnal near-surface transition (Brazel et al.
2005). Despite these limitations, the relatively low RMSE
and MAE, coupled with the high d, suggest that the model
is accurate in simulating time-series temperature data for
this study.

5 Discussion
5.1 Surface controls on PCI intensity
Compared to results from other studies (e.g., SpronkenSmith and Oke 1998; Jonsson 2004), mean PCI (ΔTu–p)

Observing and modeling the nocturnal park cool island of an arid city: horizontal and vertical impacts

intensities documented in this present study are similar in
magnitude, i.e., between 0.7 and 3.6°C. There are large
variations in ΔTu–p between heights, with both the
observed and modeled PCI intensities being larger at
heights closer to the surface (Table 3; Fig. 6). Possible
explanations for these results are now discussed.
As observations were conducted under mostly ideal
meteorological conditions (i.e., clear and mostly calm
weather), which largely preclude the turbulent transfer of
sensible and latent heat fluxes, the large differences of
observed temperature magnitudes at 0.01-m heights between
urban and non-urban surfaces are probably due to variations
of net long-wave radiation losses. These relatively strong
upward radiative fluxes occur close to the surface, as seen
from the strong temperature gradient between 0 and 1 m,
especially over xeric and lawn surfaces (Fig. 6). This surface
flux divergence probably causes the strong non-urban
surface inversions compared to the near-neutral lapse rates
observed over urban surfaces.
The radiative losses are controlled by the varying
thermal responses of different surface materials, which
can be quantified by μ and/or moisture availability.
Generally, surfaces with high μ, such as concrete or
asphalt, release stored heat at slower rates and would
remain warmer at sunrise compared to soil or lawn surfaces
(e.g., Oke 1987). Lower temperatures measured over tennis
courts could thus be explained by the lower μ on this
surface. Likewise, differences in 0.01-m temperatures
between xeric and lawn surfaces probably arise from μ
variations, with lawn surfaces having lower μ values
corresponding to lower surface temperatures. Another
possible influence, however, would be deviations in nonurban soil moisture content. Two possible impacts on
temperature may arise. First, higher (lower) soil moisture
results in higher (lower) μ and has been shown to reduce
urban vs. non-urban temperature differences (Chow and
Roth 2006). Second, given the study area’s arid climate and
the strong surface-atmosphere vapor pressure deficit,
greater soil moisture would likely result in greater cooling
due to increased evapotranspiration, which a priori is more
likely to occur over lawns than xeric surfaces. The relative
importance of these two impacts is unknown and would
require a further process-based study, although SpronkenSmith and Oke (1999) suggested that cooling from
evapotranspiration of soil moisture is more significant at
pre-sunset, while the additional soil wetness would inhibit
nocturnal cooling due to increased μ based on scale and
numerical modeling evaluations.
Another possibility explaining the variance in urban
temperatures could be the relatively larger sky view factor
associated with the exposed tennis courts. The geometry of
urban canyons have been shown to influence temperatures
and cooling rates with respect to urban parks (Upmanis et

209

al. 1998), and the lack of urban canyons on tennis courts
results in unhindered radiative cooling and cooler temperatures. The small but significant variation in modeled
ENVI-met lapse rates between “tennis courts” (which are
simulated concrete surfaces) and concrete and urban canyon
surfaces (Fig. 6) further suggests that changes in urban
geometry could be a likely factor in explaining intra-urban
temperature differences. However, the lack of significant
differences in observed lapse rates within urban canyons vs.
exposed concrete and asphalt surfaces implies that the
impact of urban geometry on temperatures may be
superseded by other larger-scale influences not modeled
by ENVI-met.
5.2 Surface/lapse rates relationships and advection
Results from observed lapse rate data in this study are
qualitatively similar with the only known previous PCI
study that examined vertical temperature differences for
urban green spaces in Stockholm, Sweden (Jansson et al.
2007), with generally neutral/near-neutral (stable) urban
(non-urban) profiles observed in both cities. A significant
difference lies in the magnitude of profile stability for nonurban surfaces between studies. In Stockholm, the maximum increase in vertical temperatures within a park was
∼1.5°C from 0.47 to 2.47 m at 2200 hours LT, compared to
2.2°C (1.4°C) between 0 and 3 m for lawn (xeric) surfaces
in this study (Fig. 6). Variations in green space type (i.e.,
lawn/xeric vs. trees), timing of observations (i.e., early
evening vs. pre-dawn), and site micro-climates (i.e.,
different wind speeds and climate types) may help to
explain these differences in lapse rate magnitudes. Of more
interest is that, although the near-neutral vertical lapse rate
simulations in ENVI-met from 1 to 4 m were qualitatively
similar to observed data, the strong stability over lawn and
xeric surfaces between 0 and 1 m was not well modeled,
leading to a large under-estimation of simulated ΔTu–p
within these heights. Whether this is due to inadequate
model surface and soil parameterization or from limitations
of modeled surface–atmospheric physics is still unknown
and requires further investigation.
The Jansson et al. (2007) study also postulated that
advection, possibly generated by pressure gradient from
temperature differences between park and non-park surfaces, likely factored in decreases in PCI intensity with
height. These local-scale nocturnal “park breezes” of low
wind speed (∼0.5 ms-1) were observed in Göteborg under
clear and calm weather (Eliasson and Upmanis 2000) and
could have occurred near the SRC field in this study.
Although the lower 0.01-m temperatures at tennis courts
relative to other urban surfaces were probably due to its
different surface characteristics, its low 1- to 3-m temperatures were likely affected by the advection of the cooler air

210

W.T.L. Chow et al.

mass from the adjacent SRC field, which can be discerned
from Figs. 4 and 5. It is very likely that this advection
extended the PCI effect westward, possibly by the width of
the park itself (∼ 200 m) as suggested by Spronken-Smith
and Oke (1998). However, the prevailing easterly winds
observed here were similar in direction to the expected E–
W evening transition, which affects most of the Phoenix
metropolitan area by sunrise (Brazel et al. 2005). It was
thus possible that the shift of cooler air over the SRC field
during the early morning was driven by this larger-scale
synoptic turbulent process and could have masked potential
park breezes. Further PCI observations, especially during
the early evening hours during the incipient stages of the
evening transition in Phoenix, are needed to determine if
park breezes develop in the study area.

6 Conclusion
This study investigated the micro-scale impacts of an urban
green space on temperatures located within a hot, arid city.
We obtained pre-sunrise temperature data of high spatial
and temporal resolution from a bicycle traverse as well as
from simulations from a numerical micro-scale climate
model (ENVI-met 3.1). Examination and discussion of
horizontal and vertical temperature distributions revealed
the following results:
&

&

&

A strong PCI effect was evident over a large park
consisting of irrigated lawn grass and bare soil (xeric)
areas. Mean (maximum) observed PCI intensities (i.e.,
ΔTu–p) were ∼3.5°C (∼6°C) depending on the urban and
non-urban surface category, with larger PCI magnitudes
documented closer to the surface, especially between 0and 1-m heights. Surface type affected PCI intensities,
with observed larger ΔTu–p occurring over irrigated lawn
as opposed to xeric surfaces, and it is likely that soil
moisture and surface thermal properties are likely factors
that explain this variance. Urban surfaces more exposed
to the atmosphere (i.e., with higher sky view factors) also
had observed lower PCI intensities when compared to
surfaces within urban canyons.
A strong inversion layer was detected over non-urban
surfaces, especially between 0- and 1-m heights, which
probably resulted from intense surface radiative cooling
from non-urban surfaces. Magnitudes of this nearsurface vertical cooling (2.2°C for lawn surfaces from
surface to 1 m) are much greater than that reported for a
PCI in a high-latitude city (Jansson et al. 2007).
Spatial and time-series data from the ENVI-met model
were evaluated using a suite of difference measure indices
of predicted vs. observed temperatures, which are considered a more preferable statistical method for determining

&

model accuracy as opposed to simple correlation measures. The model reasonably simulated mean spatial
temperature fields, especially in areas not directly adjacent
to model boundaries that increased systematic errors.
However, the model did not simulate the strong nearsurface inversion over non-urban surfaces. A comparison
with mean 2-m temperature time-series data at a nearby
meteorological station demonstrated a much higher
accuracy of mean modeled data, even after accounting
for the lack of regional exchange processes that were not
simulated in the micro-scale ENVI-met model.
Advection of colder park air towards urban surfaces,
combined with an increased surface exposure to the
atmosphere, may be important factors in explaining
cooling at different heights over adjacent urban surfaces. Both the modeled and observed results showed
that the PCI effect at 2- and 3-m heights over several
tennis courts next to the lawn/xeric field was either
negligible or non-existent depending on the park
surface type. Steady E winds, combined with high sky
view factors at the tennis courts, are probable causes for
this result. Urban canyons perpendicular to the prevailing
wind had lee effects that manifest as higher observed and
modeled temperatures.

Results from this study show that, for arid cities, the PCI
developed over irrigated lawn and xeric surfaces can be of
sufficient magnitude to mitigate the strong UHI in Phoenix,
especially within the urban canopy layer (e.g., Brazel et al.
2007). Advection of cooler park air and building orientation
are also important factors in variations of PCI over space
within the study area; urban planners and builders should be
cognizant of these issues when examining how green spaces
can mitigate the UHI. Lastly, although the ENVI-met model
does have relatively larger systematic vs. unsystematic errors,
this study shows that its accuracy of observed spatial and
time-series data is relatively reasonable, and it has the potential to be utilized as an effective planning tool for modeling
micro-scale climates in Phoenix, especially after adapting
several vegetation and soil parameters to local conditions

Acknowledgements The authors would like to thank Thomas K.
Swoveland, Abeer Hamdan and Paul Iñiguez (Arizona State University)
for their assistance during fieldwork reconnaissance and data collection.
Dr. Chen-Yi Sun (National Chin-Yi University, Taiwan) and Dr. Nancy
Selover (Office of the Arizona State Climatologist) advised on the
instrument and measurement techniques. Dr. Jianguo Wu (Arizona State
University) and Dr. Michael Bruse (Johannes Gutenberg-Universität
Mainz) provided critical feedback and comments. We also thank the
Maricopa County Air Quality Department for providing secondary climate
data from the Tempe meteorological station. This material is based upon
work supported by the National Science Foundation under Grant No.
DEB-0423704, Central Arizona - Phoenix Long-Term Ecological
Research (CAP LTER). The lead author was also supported by the
Department of Geography, National University of Singapore.

Observing and modeling the nocturnal park cool island of an arid city: horizontal and vertical impacts

References
Barradas VL (1991) Air temperature and humidity and human comfort
index of some city parks of Mexico City. Int J Biometeoro
35:24–28
Brazel AJ, Fernando HJS, Hunt JCR, Selover N, Hedquist BC,
Pardyjak E (2005) Evening transition observations in Phoenix,
Arizona. J Appl Meteor 44:99–112
Brazel AJ, Gober PA, Lee SJ, Grossman-Clarke S, Zehnder J,
Hedquist B, Comparri E (2007) Determinants of changes in the
regional urban heat island in metropolitan Phoenix (Arizona,
USA) between 1990 and 2004. Clim Res 33:171–182
Bruse M (2010) ENVI-met 3.1: model overview. University of
Bochum. Available at: www.envi-met.com
Bruse M, Fleer H (1998) Simulating surface–plant–air interactions
inside urban environments with a three dimensional numerical
model. Environ Model Softw 13:373–384
Chow WTL, Roth M (2006) Temporal dynamics of the urban heat
island of Singapore. Int J Climatol 26:2243–2260
Eliasson I, Upmanis H (2000) Nocturnal airflow from urban parks—
implications for city ventilation. Theor Appl Climatol 66:95–107
Emmanuel R (2005) An urban approach to climate-sensitive design:
strategies for the tropics. Spon, New York
Emmanuel R, Fernando HJS (2007) Urban heat islands in humid and
arid climates: role of urban form and thermal properties in
Colombo, Sri Lanka and Phoenix, USA. Clim Res 34:241–251
Garcia-Cueto R, Jauregui E, Tejeda A (2003) Urban/rural energy
balance observations in a desert city in northern Mexico. In:
Proceedings of Fifth International Conference on Urban Climate,
Lodz, Poland, 177–180
Geiger R, Aron RH, Todhunter P (2003) The climate near the ground,
6th edn. Rowman & Littlefield, Lanham, Maryland
Gober P (2006) Metropolitan Phoenix: place making and community
building in the desert. University of Pennsylvania Press,
Philadelphia
Grimmond CSB, Oke TR (1995) Comparison of heat fluxes from
summertime observations in the suburbs of four North American
cities. J Appl Meteorol 34:873–889
Jansson C, Jansson P-E, Gustafsson D (2007) Near surface climate in
an urban vegetated park and its surroundings. Theor Appl
Climatol 89:185–193
Jauregui E (1990/1991) Influence of a large park on temperature and
convective precipitation in a tropical city. Energ Buildings 15–
16:457–463
Jonsson P (2004) Vegetation as an urban climate control in the subtropical
city of Gaborone, Botswana. Int J Climatol 24:1307–1322
Kirby SF, Sellers WD (1987) Cold air drainage and urban heating in
Tucson, Arizona. J Arizona Nevada Acad Sci 22:123–128
Kurn D, Bretz S, Huang B, Akbari H (1994) The potential for reducing
urban air temperatures and energy consumption through vegetative
cooling. ACEEE Summer Study on Energy Efficiency in Buildings,
American Council for an Energy Efficient Economy. Pacific Grove,
California. Available at http://www.osti.gov/bridge/servlets/purl/
10180633-hLSlld/native/10180633.PDF. Accessed 23 Jul 2009
Lalic B, Mihailovic DT (2004) An empirical relation describing leafarea density inside the forest for environmental modeling. J Appl
Meteorol 43:641–645
Martin CA (2008) Landscape sustainability in a Sonoran desert city.
Cities Envir 1(2):1–16
Muellar EC, Day TA (2004) The effect of urban groundcover on
microclimate, growth and leaf gas exchange in oleander in
Phoenix, Arizona. Int J Biometeor 49:244–255

211

Oke TR (1970) The temperature profile near the ground on calm clear
nights. Q J Roy Meteor Soc 96:14–23
Oke TR (1987) Boundary layer climates. Methuen, New York, p 372
Oke TR (1989) The micrometeorology of the urban forest. Phil Trans
R Soc Lond B324:335–349
Oke TR (2004) Initial guidance to obtain representative meteorological
observations at urban sites. Instruments and methods of observation program. IOM report no. 81, WMO/TD 1250, World
Meteorological Organization, Geneva
Oke TR, Johnson GT, Steyn DG, Watson ID (1991) Simulation of
surface urban heat islands under “ideal” conditions at night. Part
2: diagnosis of causation. Bound-Layer Meteor 56:339–359
Pearlmutter D, Berlinera P, Shaviv E (2007) Urban climatology in arid
regions: current research in the Negev desert. Int J Climatol
27:1875–1885
Pearlmutter D, Krüger EL, Berlinera P (2009) The role of evaporation
in the energy balance of an open-air scaled urban surface. Int J
Climatol 29:911–920
Roth M, Oke TR, Emery WJ (1989) Satellite-derived urban heat
islands from three coastal cities and the utilization of such data in
urban climatology. Int J Remote Sens 10:1699–1720
Saito I, Ishihara O, Katayama T (1990/1991) Study of the effect of
green areas on the thermal environment in an urban area. Energ
Buildings 15–16:493–498
Sham S (1987) Urbanization and the atmospheric environment in the
low tropics: experiences from the Kelang Valley Region.
Malaysia Universiti Kebangsaan Malaysia, Bangi
Shashua-Bar L, Pearlmutter D, Erell E (2009) The cooling efficiency
of urban landscape strategies in a hot dry climate. Landscape
Urban Plan 92:179–186
Souza LCL, Rodrigues DS, Mendes JFG (2003) Sky view factors
estimation using a 3D-GIS extension. Eighth International
IBPSA Conference, Eindhoven, Netherlands. August 11–14,
2003. Available at www.ibpsa.org/proceedings/BS2003/
BS03_1227_1234.pdf (accessed 23 Jul 2009)
Spronken-Smith RA (1994) Energetics and cooling in urban parks.
Unpublished PhD thesis. The University of British Columbia,
Vancouver
Spronken-Smith RA, Oke TR (1998) The thermal regime of urban
parks in two cities with different summer climates. Int J Remote
Sens 19:2085–2104
Spronken-Smith RA, Oke TR (1999) Scale modelling of nocturnal
cooling in urban parks. Bound-Layer Meteor 93:287–312
Spronken-Smith RA, Oke TR, Lowry WP (2000) Advection and the
surface energy balance of an irrigated urban park. Int J Climatol
20:1033–1047
Stabler LB, Martin CA, Brazel AJ (2005) Microclimates in a desert
city were related to land use and vegetation index. Urban For
Urban Gree 3:137–147
Sun CY, Brazel AJ, Chow WTL, Hedquist BC, Prashad L (2009)
Desert heat island study in winter by mobile transect and remote
sensing techniques. Theor Appl Climatol 98(3–4):323–335.
doi:10.1007/s00704-009-0120-2
Upmanis H, Eliasson I, Lindqvist S (1998) The influence of green
areas on nocturnal temperatures in a high latitude city (Göteborg,
Sweden). Int J Climatol 18:681–700
Willmott CJ (1982) Some comments on the evaluation of model
performance. Bull Am Meteor Soc 63(11):1309–1313
Willmott CJ (1984) On the evaluation of model performance in
physical geography. In: Gaile GL, Willmott CJ (eds) Spatial
statistics and models. D Reidel, Dordrecht, pp 443–460
Wong NH, Jusof SK (2008) GIS-based greenery evaluation on campus
master plan. Landsc Urban Plan 84(2):166–182