Building and Environment 70 (2013) 318e333

Contents lists available at ScienceDirect

Building and Environment

journal homepage: www.elsevier.com/locate/buildenv

Natural ventilation assessment in typical open and semi-open urban

environments under various wind directions
Jian Hang a, *, Zhiwen Luo b, Mats Sandberg c, Jian Gong d
a
Department of Atmospheric Sciences, School of Environmental Science and Engineering, Sun Yat-Sen University, Guangzhou, Guangdong 510275, PR China
b
School of Construction Management and Engineering, University of Reading, Reading, UK
c
Laboratory of Ventilation and Air Quality, University of Gävle, SE-80176 Gävle, Sweden
d
School of Civil Engineering and Architecture, Nanchang Hangkong University, Nanchang, Jiangxi 330063, PR China

a r t i c l e i n f o a b s t r a c t

Article history: Semi-open street roofs protect pedestrians from intense sunshine and rains. Their effects on natural
Received 20 June 2013 ventilation of urban canopy layers (UCL) are less understood. This paper investigates two idealized urban
Received in revised form models consisting of 4(2
2) or 16(4
4) buildings under a neutral atmospheric condition with parallel
2 September 2013
(0 ) or non-parallel (15 , 30 , 45 ) approaching wind. The aspect ratio (building height (H)/street width
Accepted 5 September 2013
(W)) is 1 and building width is B ¼ 3H. Computational fluid dynamic (CFD) simulations were first vali-
dated by experimental data, confirming that standard keε model predicted airflow velocity better than
Keywords:
RNG keε model, realizable keε model and Reynolds stress model. Three ventilation indices were
Semi-open street roof
Natural ventilation
numerically analyzed for ventilation assessment, including flow rates across street roofs and openings to
Age of air show the mechanisms of air exchange, age of air to display how long external air reaches a place after
Purging flow rate entering UCL, and purging flow rate to quantify the net UCL ventilation capacity induced by mean flows
CFD simulations and turbulence.
Wind tunnel experiment Five semi-open roof types are studied: Walls being hung above street roofs (coverage ratio la ¼ 100%)
at z ¼ 1.5H, 1.2H, 1.1H (‘Hung1.5H’, ‘Hung1.2H’, ‘Hung1.1H’ types); Walls partly covering street roofs
(la ¼ 80%) at z ¼ H (‘Partly-covered’ type); Walls fully covering street roofs (la ¼ 100%) at z ¼ H (‘Fully-
covered’ type). They basically obtain worse UCL ventilation than open street roof type due to the
decreased roof ventilation. ‘Hung1.1H’, ‘Hung1.2H’, ‘Hung1.5H’ types are better designs than ‘Fully-
covered’ and ‘Partly-covered’ types. Greater urban size contains larger UCL volume and requires longer
time to ventilate. The methodologies and ventilation indices are confirmed effective to quantify UCL
ventilation.
Crown Copyright Ó 2013 Published by Elsevier Ltd. All rights reserved.

1. Introduction areas. Good reviews on this topic can be found in the literature [12e
15]. For two-dimensional (2D) street canyons [1,15e19], street
Wind from rural areas provides cleaner rural air into urban aspect ratio (building height/street width, H/W) is the first key
canopy layers (UCL) to help pollutant and heat dilution. Good UCL parameter to affect the flow regimes and pollutant dispersion. For
ventilation has been known as one of the possible mitigation so- three-dimensional (3D) urban canopy layers, total street length or
lutions to improve urban air environments [1e11], meanwhile urban size [8,11,30], building packing density and frontal area
ameliorate indoor air quality through building ventilation systems. density [8,10,20e23], ambient wind directions [23,24,32,37],
Complemented by wind tunnel/field experiments, computa- building layouts and height variations [8,21e23,25e26] etc, are
tional fluid dynamics (CFD) simulations have been widely used to significant parameters and have been widely investigated.
predict turbulent airflow, mass transports and energy budgets In addition to the widely studied urban models with open street
within, close to and above different UCLs [2,4e11,17e26,28e37], roofs, semi-open street roof is one of popular urban design ele-
ranging from street canyons, street intersections, cavities and ments existing in the realistic urban areas to protect pedestrians
courtyards, up to structured building arrays and realistic urban from strong sunshine and reduce the inconveniences in rainy or
snowy days. Such semi-open street roofs have been reported and
investigated by experiments and CFD simulations in the literature
* Corresponding author. Tel./fax: þ86 20 84110375. [5e7], including a large naturally ventilated semi-open market
E-mail addresses: [email protected], [email protected] (J. Hang). building [5], a semi-open shopping mall being located in Lisbon,

0360-1323/$ e see front matter Crown Copyright Ó 2013 Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.buildenv.2013.09.002
 J. Hang et al. / Building and Environment 70 (2013) 318e333 319

Portugal [6], enclosed-arcade (or semi-open) markets of Korea with with doors connected to the semi-open streets. These semi-open
eleven arcade-type designs (or semi-open street roof) [7]. Although outdoor environments are naturally ventilated to reduce energy
the requirements of design are different according to various consumption. Such semi-open street roof designs are used to pro-
climate conditions, sufficient natural UCL ventilation has been vide convenience for pedestrians, but they possibly deteriorate UCL
considered as an important environment design factor for more ventilation performance. This paper aims to quantitatively evaluate
healthy semi-open outdoor environments [5e7]. Fig. 1 shows two these effects. Although thermal buoyancy force induced by tem-
other kinds of semi-open street roof designs in the suburb of perature difference and atmospheric stability also influence urban
Guangzhou China, which are located in a subtropical region airflows and UCL ventilation [19,28,29], this paper takes the first
annually characterized by intense solar radiation and precipitation. step to consider a neutral atmospheric condition assuming that the
Fig. 1a shows walls being hung above street roofs of a food court, ambient wind velocity is sufficiently large and thermal effects are
and Fig. 1b displays walls partially covering street roofs of a retail negligible.
center. Each shop or restaurant has its own enclosed space with air In building ventilation, as reviewed by Chen [27], indoor
conditioners inside for cooling in summer (April to September) and ventilation indices have been widely used to evaluate how external

Fig. 1. Two urban configurations of semi-open street roof design: (a) Walls being hung above street roofs of food court, (b) Walls being partly covered at street roof height (z ¼ H) of
retail center.
320 J. Hang et al. / Building and Environment 70 (2013) 318e333

air enters a room and ventilates it. In recent years, researchers have
started to apply similar concepts to estimate UCL ventilation [2,4e
11,24,28e32,37], including ventilation flow rate and air change rate
per hour (ACH) [4,6,7,28e30], pollutant exchange rate [31],
pollutant retention time and purging flow rate [2,8,24], age of air
and air exchange efficiency [32], city breathability [10,11] etc. This
paper emphasizes the quantitative analysis of UCL ventilation
induced by rural wind assuming that rural air is relatively clean.
Flow rates across street openings and street roofs are first analyzed
to quantify the mechanisms of air exchange [37], moreover the
local mean age of air [32] is used to quantify how long the external
air can reach a place after it enters the UCL. Finally, the UCL purging
flow rate [2,8] is also applied to estimate the net UCL ventilation
capacity induced by both mean flows and turbulent diffusions.
Tracer gas techniques [27,44] are usually used to measure indoor
ventilation indices. However for both open or semi-open outdoor
spaces, ventilation indices such as age of air and purging flow rate
are difficult to be measured by tracer gas techniques, since outdoor
environment is not an enclosed space with more complicated
openings than indoor, moreover perfect mixing and uniform
pollutant generation rate in UCLs are difficult to experimentally
control. Thus the literature [5e11,24,28e32] usually use experi-
mental data to validate the reliability of CFD methods in predicting
concentration and airflow field, then analyze outdoor ventilation
indices by using CFD simulations. This paper also utilizes similar
methodologies.

2. Methodologies

2.1. Turbulence modeling in CFD simulations

Large eddy simulation (LES) models are known to perform

better in predicting turbulent flows than the Reynolds-Averaged
NaviereStokes (RANS) approaches, but the applicability of LES
models is more problematic due to its much longer computational
time required than RANS approaches and some issues regarding the
implementation of wall and inlet boundary conditions [33,34].
Considering that RANS turbulence models are more time-saving
and provide reasonable results for mean flows and the spatial
average flow properties [33], this paper adopted RANS turbulence
models for evaluating UCL ventilation.
UCL ventilation relies on both mean flows and turbulence
within the UCL [8,37]. According to the literature [35,36], the
modified keε models, for example RNG keε model, are able to Fig. 2. Model descriptions of experimental model: (a) The idealized urban model with
4 buildings and open street roof, (b) Vertical profiles of velocity and turbulence in-
correct the drawback of the standard keε model that severely over- tensity in the upstream free flow of wind tunnel experiment.
predicts turbulent kinetic energy in separated flows around front
corners of buildings, however, they fail to predict the sizes of
reattachment lengths behind buildings and under-predict the ve-
locity in weak wind regions. It is desirable to compare different secondary streets. The scale ratio between small-scale and full-
RANS turbulence models in predicting urban airflows and UCL scale models is 1:100. Thus in full-scale real conditions
ventilation to provide a sensitivity study, including standard keε H ¼ W z 7 m, B ¼ 3H z 21 m, L z 49 m. In small-scale models the
model, RNG keε model, realizable keε model and Reynolds stress height of 1.5 mm (0.22H) corresponds to the face level (1.5 m) in
model (RSM). full-scale conditions.
The measurements were performed in the closed-circuit type
2.2. Experimental and CFD set-ups in the validation case wind tunnel at the Laboratory of Ventilation and Air Quality, Uni-
versity of Gävle, Sweden, with the working section of 11 m long,
This paper aims to study UCL ventilation in low-rise idealized 3 m wide, 1.5 m tall. Thus the blockage ratio is about 0.6%, which
and typical urban models consisting of two-storey buildings (about represents the percentage of the small-scale urban model
7 m tall). Wind tunnel data was first used to evaluate the reliability obstructing the test section area (3 m
1.5 m) of the wind tunnel.
of CFD methodologies. As shown in Fig. 2a, Hang et al. [37] per- The stream-wise, lateral and vertical directions are represented by
formed some wind tunnel experiments to investigate the flow in a x, y, z. Hotwire anemometer was used to measure vertical profiles of
small-scale urban model with four square building blocks (building velocity (Um(z)) and turbulence intensity (I(z)) in the upstream free
height H ¼ 0.069 m, building width B ¼ 3H) and two crossing flow of wind tunnel (see Fig. 2b), horizontal profiles of velocity uðxÞ
streets (street width W ¼ H, urban size L ¼ 7 H). The approaching and turbulence intensity I(x) along the main street centerline (see
wind was parallel to the main street and perpendicular to the Fig. 3b) at z ¼ 0.11H (7.5 mm). The sampling frequency was 100 Hz.
 J. Hang et al. / Building and Environment 70 (2013) 318e333 321

Fig. 3. (a) Computational domain for cases with a parallel approaching wind (0 ) and
half domain size, (b) Grid arrangements in xey plane in the validation case.

The measurement time was 30 s for each point. It is worth

mentioning that, the hotwire is only sensitive to velocity compo-
nents perpendicular to it (i.e. the vertical velocity w and the stream-
wise velocity
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi u). So data measured by the hotwire were actually
u2 þ w2 . Here the hotwire was only located where the span-wise
(y) velocity v was zero, including in the upstream free flow and
along the main street centerline, so pthe measured ffi data were actu-
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ally the velocity magnitude ðU ¼ u2 þ v2 þ w2 Þ. Fig. 4. (a) Computational domain with oblique wind direction and full domain size.
Because there were no roughness elements in wind tunnel ex- Model descriptions of urban models with (b) 4 (2
2) buildings and (c) 16 (4
4)
buildings.
periments, a thin neutral atmospheric boundary layer (ABL) and a
sharp vertical profile of velocity was produced in the upstream free
flow (see Fig. 2b). We only used the measured profiles (Um(z) and
I(z)) in Fig. 2b to provide boundary conditions at domain inlet in the in wind tunnel and the similar spatial mean velocity (about 3.2 m/s)
CFD validation case. At domain inlet, turbulent kinetic energy is as that in Eq. (1a). According to Snyder [39], Reynolds-number in-
defined as k(z) ¼ 1.5(I Um)2 and its dissipation rate is ε(z) ¼ C3/4 3/2
m k / dependence can be satisfied if the Reynolds number is greater than
l, where Cm ¼ 0.09 and l is the turbulent characteristic length scale. 4000, i.e. the main structure of turbulence can be almost entirely
Note that, the maximum velocity in the upstream free flow of wind responsible for the bulk transport of momentum and heat or mass
tunnel experiments was 13.33 m/s, however in cases for ventilation transfer. If the velocity z ¼ H ¼ 0.069 m in the upstream free flow
analysis, we used a realistic approaching wind (see Eq. (1a)) with a (see Fig. 2b) is defined as the reference velocity Uref z 2.94 m/s, the
spatial mean velocity of about 3.2 m/s, so in the validation case we reference Reynolds number (ReH ¼ rUrefH/m z 13887) is much
actually utilized a smaller fitting velocity profile (maximum ve- larger than 4000, Thus the technique of using a smaller inflow
locity is 3.24 m/s, see Fig. 2b) with the same thickness of ABL as that velocity (i.e. 3.24 m/s) can ensure Reynolds number independence.
322 J. Hang et al. / Building and Environment 70 (2013) 318e333

Table 1
Model descriptions of 48 test cases.

2 Rows, 2 columns (2
2) 4 Rows, 4 columns (4
4)

Case namea Ambient Case name Ambient wind

wind direction q 
direction q 

[2e2, 0, Open] 0 [4e4, 0, Open] 0

[2e2, 0, Hung1.5H] [4e4, 0, Hung1.5H]
[2e2, 0, Hung1.2H] [4e4, 0, Hung1.2H]
[2e2, 0, Hung1.1H] [4e4, 0, Hung1.1H]
[2e2, 0,Partlyecovered] [4e4, 0,Partly-covered]
[2e2, 0, Fully-covered] [4e4, 0, Fully-covered]
[2e2, 15, Open] 15 [4e4, 15, Open] 15
[2e2, 15, Hung1.5H] [4e4, 15, Hung1.5H]
[2e2, 15, Hung1.2H] [4e4, 15, Hung1.2H]
[2e2, 15, Hung1.1H] [4e4, 15, Hung1.1H]
[2e2, 15,Partly-covered] [4e4, 15,Partly-covered]
[2e2, 15, Fully-covered] [4e4, 15, Fully-covered]
[2e2, 30, Open] 30 [4e4, 30, Open] 30
[2e2, 30, Hung1.5H] [4e4, 30, Hung1.5H]
[2e2, 30, Hung1.2H] [4e4, 30, Hung1.2H]
[2e2, 30, Hung1.1H] [4e4, 30, Hung1.1H]
[2e2, 30,Partly-covered] [4e4, 30,Partly-covered]
[2e2, 30, Fully-covered] [4e4, 30, Fully-covered]
[2e2, 45, Open] 45 [4e4, 45, Open] 45
[2e2, 45, Hung1.5H] [4e4, 45, Hung1.5H]
[2e2, 45, Hung1.2H] [4e4, 45, Hung1.2H]
[2e2, 45, Hung1.1H] [4e4, 45, Hung1.1H]
[2e2, 45,Partly-covered] [4e4, 45,Partly-covered]
[2e2, 45, Fully-covered] [4e4, 45, Fully-covered]
a
Case name is defined as [row number-column number, wind direction (q ), roof
type]. Open’ denotes open street roofs; ‘Fully-covered’ and ‘Partly-covered’ means
solid walls ‘fully’ or ‘partly cover’ street roofs at z ¼ H. ‘Hung1.5H, Hung1.2H and
Hung1.1H’ represent solid walls are ’Hung’ above street roofs at z ¼ 1.5H, 1.2H and
1.1H.

wall surfaces, building corners, street openings. The grid size near
the ground is 0.036H(dz ¼ 2.5 mm). There are 6 cells vertically from
z ¼ 0 to the pedestrian height (z ¼ 20 mm ¼ 0.29H). The grid size
near building roofs at z ¼ H is 0.022H (dz ¼ 1.5 mm). The horizontal
grid size (dx and dy) near building surfaces varies from 0.022H to
0.043H. The maximum expansion ratio from building surfaces to
the surrounding is 1.15 and the total number of hexahedral cells is
about 0.82 million.
In the CFD validation case, all CFD set-ups including computa-
tional domain size, boundary conditions and grid arrangements
fulfilled the major CFD guidelines recommended by Tominaga et al.
[40].

Fig. 5. (a) ‘Fully-covered’ roof type: walls fully cover street roofs at z ¼ H (b) ‘Partly- 2.3. CFD set-ups for flow modeling
covered’ roof type: walls partly cover street roofs at z ¼ H, (c) Types of ‘Hung1.5H’,
‘Hung1.2H’, ‘Hung1.1H’: walls are hung above street roofs at z ¼ 1.1H, 1.2H, 15H. After the CFD validation case, more urban configurations with or
without semi-open street roofs and various ambient wind di-
rections were investigated. To better illustrate idealized urban
The CFD code FLUENT 6.3 [38] was used to solve the steady-state models, all test cases were defined as Case [number of rows-
isothermal turbulent flows. For CFD simulations, we used the same number of columns, wind direction, roof type]. ‘Open’ roof type
small-scale urban geometries (H ¼ 0.069 m) as those in wind tunnel denotes open street roofs; as shown in Fig. 4aec, four wind di-
experiments. Only half computational domain was used to reduce rections of 0 , 15 , 30 , 45 were included. So the name of validation
the calculation time. Fig. 3a displays the computational domain and case is Case [2e2, 0, Open] with four buildings (2 rows, 2 columns),
boundary conditions in the CFD validation case. The computational a parallel approaching wind (0 ) and open street roof (‘Open’ roof
domain is 14.5H wide (1 m) in the lateral (y) direction and 11H tall type). As displayed in Fig. 4c, a bigger urban model with 16 build-
(0.75 m) in the vertical (z) direction. Thus the blockage ratio is ings (4 columns, 4 rows, urban size L ¼ 15H z 105 m in full scale)
about 1.9% (less than 3%) satisfying the requirement of the litera- was also investigated in CFD simulations. Besides the ‘Open’ roof
ture [40]. No-slip wall boundary condition was utilized at wall type, Fig. 5 shows the other five types studied in CFD simulations.
surfaces, and zero normal gradient boundary condition was used at ‘Fully-covered’ roof type (see Fig. 5a) means walls entirely covering
domain outlet, domain roof, domain lateral boundary, domain street roofs with a coverage ratio (la) of 100% at z ¼ H, and ‘Partly-
symmetry boundary. covered’ roof type (see Fig. 5b) represents street roofs being partly
Fig. 3b displays the grid arrangements in xey plane of the covered (la ¼ 80%) by walls at z ¼ H. Roof types of ‘Hung1.5H’,
validation case. Finer grids are produced within the UCL and near ‘Hung1.2H’ and ‘Hung1.1H’ (see Fig. 5c) represent walls being hung
 J. Hang et al. / Building and Environment 70 (2013) 318e333 323

qffiffiffiffiffiffi
k0 ðzÞ ¼ u2* = Cm (1b)

3=4
ε0 ðzÞ ¼ Cm k0 ðzÞ3=2 =ðkv zÞ (1c)

where the friction velocity u* ¼ 0.24 m/s, kv ¼ 0.41 is von Karman’s

constant, UH ¼ 2.66 m/s is the reference velocity at z ¼ H ¼ 0.069 m
of domain inlet.
For test cases with a non-parallel approaching wind (15 , 30 ,
45 ), there are two domain inlets and two domain outlets (see
Fig. 4a). At domain inlets, the power-law velocity profiles (stream-
wise velocity u ¼ U0 ðzÞ cos⁡q, span-wise velocity v ¼ U0(z)sinq
and vertical velocity wðzÞ ¼ 0) and profiles of turbulent quantities
in Eqs. (1b)e(1c) were used to provide boundary conditions. Zero
normal gradient conditions were still used at two domain outlets
and domain roof.
Fig. 6a and b show two examples of the grid arrangements in
test cases with four (2
2) buildings and semi-open street roofs.
Note that, the thickness of hung walls to produce semi-open street
roofs was zero in CFD models. The grid arrangements were similar
with those in the CFD validation case except three points: The first
is that the grids near semi-open street roofs (i.e. at z ¼ 1.1H, 1.2H,

Fig. 6. Two examples of grid arrangements for urban geometries with 4 buildings: (a)
In xey plane, (b) in xez plane. (c) Definition of uniform pollutant source in UCL
volume.

above street roofs (la ¼ 100%) at z ¼ 1.5H, 1.2H and 1.1H, respec-
tively. As summarized in Table 1, total 48 test cases were numeri-
cally investigated.
For test cases with a parallel approaching wind (0 ), the
computational domain and boundary conditions were similar as
the CFD validation case. A power-law velocity profile was applied at
domain inlet with a power-law exponent of 0.16 (see Eq. (1a)). As
reported by Lien and Yee [41], it represents a neutral atmospheric
boundary layer (ABL) with a depth of 1.8 m created in the wind
tunnel by using spires and floor roughness with a roughness length
of approximately z0 ¼ 0.001 m. In full-scale real conditions, it
corresponds to a neutrally-stratified ABL with a surface roughness
of z0 ¼ 0.1 m [42] (i.e. a neutral ABL above open rural area with a
regular cover of low crop and occasional large obstacles [43]) The
spatial mean velocity at domain inlet calculated from Eq. (1a)
approximately equals to that calculated from the inflow velocity
profile of the CFD validation case (see Fig. 2b). The inlet profiles of
turbulent kinetic energy and its dissipation rate were calculated
by Eqs. (1b)e(1c)) [30,41].
Fig. 7. Validation profiles of (a) Velocity and (b) Turbulence intensity along the street
centerline at z ¼ 0.11H by using different turbulence models. (c) Horizontal profiles of
uðzÞ ¼ U0 ðzÞ ¼ UH ðz=HÞ0:16 ; vðzÞ ¼ wðzÞ ¼ 0 (1a) velocity for a grid independence study.
324 J. Hang et al. / Building and Environment 70 (2013) 318e333

Fig. 8. (a) 3D streamline, (b) s*p in z ¼ 0.22H and Q* in Case [2e2, 0, Open], Case [2e2, 0, Hung1.2H], Case [2e2, 0, Partly-covered], Case [2e2, 0, Fully-covered].

1.5H) are also fine with a grid size of dz ¼ 0.014H ¼ 1 mm (see diffusion. The SIMPLE scheme was used for the pressure and ve-
Fig. 6b); The second is that for test cases with 16 buildings the locity coupling. CFD simulations were run until all residuals became
maximum expansion ratio of grid size from wall surfaces to the constant. Overall, residual for the continuity equation was below
surrounding is 1.2 which is less than 1.3 and satisfies the CFD 104, residuals for the velocity components and k were below 107,
guideline [40]; The third is that the grid number in cases with residuals for pollutant concentration and ε were below 0.5
105
‘Partly-covered’ roof type (see Fig. 6a) is a little more than the other and 0.5
104 respectively.
roof types, because fine grids with grid size of dy ¼ 0.029H were
also generated near lateral boundaries of partly-covered street 2.4. Ventilation assessment indices
roofs. The maximum grid number is about 3.5 million in Case [4e
4,45, Partly-covered]. 2.4.1. Age of air
All transport equations were discretized by the second order The local mean age of air (sp) was originally defined in indoor
upwind scheme to increase the accuracy and reduce numerical ventilation and can be measured by tracer gas techniques [44]. The
 J. Hang et al. / Building and Environment 70 (2013) 318e333 325

Fig. 9. (a) 3D streamline, (b)s*p and Q* in Case [2e2, 0, Hung1.5H], Case [2e2, 15, Hung1.5H], Case [2e2, 30, Hung1.5H], Case [2e2, 45, Hung1.5H]. Note that in Fig. 9b, negative values
of Q* by mean flows denote air leaving UCL and positive ones represent air entering UCL.

Table 2
Effect of turbulence models and turbulent Schimdt number (Sct) on hs*p i, PFRa and QTa in the entire UCL, Qroof(turb)a and Qa across O3 in Case [2e2, 0, Open].

Turbulence models Sct hs*p i PFRa QTa Qaroof(out) Qaroof(in) Qaroof(turb) Qa(O3)

Standard keε 0.4 21.2 1.847 1.376 0.825 0.148 1.211 0.551
0.7 24.3 1.609
1.0 26.4 1.482
Realizable keε 0.7 27.2 1.439 1.401 0.844 0.145 1.066 0.536
RNG keε 0.7 28.8 1.358 1.378 1.127 0.181 0.919 0.274
a
Negative values denote air leaving UCL and positive ones represent air entering it.
326 J. Hang et al. / Building and Environment 70 (2013) 318e333

local age of air in UCLs represents the mean time required for the
external young air to reach a point since it enters UCLs. If the age of
air in rural areas is zero, the greater age of air in UCLs represents a
greater probability to be polluted. The UCL age of air depicts how
rural air is supplied and distributed within UCLs. Hang et al. [32]
first introduced the homogeneous emission method [44] to
numerically predict age of air in UCLs.
The governing equations of time-averaged pollutant concen-
tration (c, kg/m3) and the age of air (sp, s) are displayed as below:

 
vsp v vsp
uj  Kc ¼ 1 (2)
vxj vxk vxk

!
vc v vc
uj  Kc ¼ Sc (3)
vxj vxj vxj

where uj is the velocity components ðu; v; wÞ in the stream-wise (x),

span-wise (y) and vertical (z) directions, Kc ¼ nt/Sct is the turbulent
eddy diffusivity of pollutants, nt is the kinematic eddy viscosity, Sct
is the turbulent Schimdt number ( Sct ¼ 0.7) [8,10,20,45]. Sc is the
pollutant source term (kg m3 s1).
In the homogeneous emission method [44], a relation between
these two variables was mathematically derived. If a homogenous
pollutant release rate (Sc, kg m3 s1) is defined in the entire UCL,
the age of air (sp, s) can be calculated:

sp ¼ c=Sc (4)
Eq. (4) illustrates a relationship that, with a uniform pollutant
source in the entire UCL, higher pollutant concentration at a point
represents that it takes the external clean air a longer time to arrive.
Fig. 6c shows an example of defining uniform pollutant source
in the entire UCL. In this paper, the pollutant emission rate was
small (Sc ¼ 107 kg m3 s1) to ensure the source release producing
little disturbance to the flow field. The inflow concentration at
domain inlet was defined zero, and the zero normal flux condition
was used at wall surfaces. At all other boundaries zero normal
gradient condition was utilized.
Because the age of air in small-scale urban models is small (scale
ratio 1:100), the age of air was normalized in Eq. (5a). To compare
the age of air in the entire UCLs, this paper also analyzed the
normalized spatial mean age of air ðhs*p iÞ in Eq. (5b)

s*p ¼ sp
100 (5a)

D E Z
s*p ¼ s*p dxdydz=Vol (5b)
Vol

where Vol is the entire UCL volume.

2.4.2. Ventilation flow rates and UCL purging flow rates

Both mean flows and turbulent diffusions are significant factors
for UCL ventilation [37] and pollutant removal [8]. The purging flow
rate represents the net flow rate induced by both mean flows and
turbulent diffusions for a volume to be purged out by wind through
it. It has been used to quantify the ventilation in UCLs [2] and at the
pedestrian levels [8].
This paper mainly emphasizes the purging flow rate for the
entire UCL. If a passive contaminant source is generated within the
Fig. 10. Q* in urban models with 4 buildings and wind directions of (a) 0 , (b) 15 , (c)
30 , (d) 45 .
entire UCL (see Fig. 6c) with a uniform emission rate (here
Sc ¼ 107 kg m3 s1), the UCL purging flow rate (PFR, m3/s) is
calculated in Eq. (6).
 J. Hang et al. / Building and Environment 70 (2013) 318e333 327

Sc
Vol Sc
Vol velocity in the downstream region of the main street. This finding
PFR ¼ ¼ Z (6) agrees with the literature [35,36] that non-standard keε models
hci
cdxdydz=Vol perform better in predicting separate flows but do worse in pre-
Vol dicting airflow velocity in weak wind regions. All RANS turbulence
models can only predict the shape of turbulence intensity profile,
Here hci is the spatially-averaged concentration in the entire
thus Q*roof(turb) calculated by CFD simulations were only used to
UCL volume (Vol). It is worth mentioning that PFR is independent of
provide a reference study and the relative values of Q*roof(turb)
pollutant sources, and illustrates the net UCL ventilation capacity
among different test cases were emphasized. Since the better
due to both mean flows and turbulent diffusion.
prediction of mean flows within UCL and along the streets is more
Because PFR is small for small-scale urban models (scale ratio
important, this paper hereby regards the standard keε model as the
1:100), PFR is normalized by the reference flow rate (QN).
default turbulence model in the following CFD simulations.
Sc
Vol PFR For the validation case (medium grid, 0.8 million), a finer grid
PFR* ¼ ¼ (7) arrangement with the minimum grid size of 0.014H and grid
hciQN QN
number of 1.3 million was used to perform a grid independence
study. As displayed in Fig. 7c, numerical results were not sensitive
ZH to the grid refinement, indicating present grid arrangements in
QN ¼ H
U0 ðzÞdz (8) Fig. 3b were sufficiently fine.
0
3.2. Ventilation assessment in cases with four buildings
where QN ¼ 0.01093 m3/s is the flow rate far upstream through the
same area with a windward street opening (area A ¼ H
H), U0(z) is In this subsection, the effects of semi-open street roofs and
defined in Eq. (1a). various wind directions in test cases with four buildings and two
Fig. 4b and c show the definition of street openings in test cases crossing streets (i.e. Case [2e2,wind direction, roof type], see
with 4 (2
2) and 16 (4
4) buildings. To quantify the ventilation Table 1) were investigated.
pattern, all flow rates entering and leaving UCL volumes were
normalized by the reference flow rate (QN), including Q* due to 3.2.1. Effect of semi-open street roofs in four example test cases
mean flows (see Eq. (9)) and Q*roof(turb) due to turbulence fluctu- Fig. 8a displays three-dimensional (3D) streamline in four test
ations across street roofs [37] (see Eq. (10)): cases (only half domain, 0 ), i.e. Case [2e2, 0, Open], Case [2e2, 0,
Z Hung1.2H], Case [2e2, 0, Partly-covered], Case [2e2, 0, Fully-
! !
Q* ¼ V $ n dA=QN (9) covered]. Channel flows are found in the main streets parallel to the
A
approaching wind and 3D helical flows exist in the secondary
Z streets. These channel and helical flows produce air exchange and
* turbulent diffusion through street openings and street roofs.
Qroof ðturbÞ ¼  0:5sw dA=QN (10)
Different semi-open street roofs may produce various flow pattern
! ! and ventilation capacity but this effect cannot be clearly displayed
where in Eq. (9), V is velocity vector, n is the normal direction of by only 3D streamlines in Fig. 8a. To quantify this effect, Fig. 8b
street openings or street roofs, A is surface area; In Eq. (10), shows the normalized age of air ðs*p ¼ sp
100Þ in z ¼ 0.22H (i.e.
pffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffi
sw ¼ w’ w’ ¼ 2k=3 is the fluctuation velocity on street roofs 1.5 m in full scale) and normalized flow rates (Q*) in these four test
based on the approximation of isotropic turbulence (k is the tur- cases. Positive values denote air entering UCLs and negative ones
bulent kinetic energy). represent air leaving UCLs. s*p along the main street (Street 1 and
Due to the flow balance by mean flows, the total flow rate Street 3) is relatively small (i.e. air is relatively young) because Q*
leaving UCL (Qout) through UCL boundaries equals to that entering through O1 and O3 are always large (Q*(O1) ¼ 1.048 to 0.848;
UCL (Qin). They are named as the total flow rates by mean flows QT Q*(O3) ¼ -0.551 to 0.813). In the secondary streets (Street 2 and
and are normalized by the reference flow rate QN. Street 4), Q* through O2 (O4) are small (only 0.086 to 0.019). Thus
the roof ventilations are more significant to the secondary streets.
QT* ¼ Qin
* *
¼ Qout (11) For example, in Case [2e2, 0, Open], s*p in Street 2 (or Street 4) is
similar with that in Street 3 because the flow rates across street
By applying the above concepts, this paper quantifies the effects roofs are comparable to those across O1 and O3, including the
of semi-open street roofs and various wind directions on the age upward and downward flow rates due to mean flows
distribution, the ventilation pattern and the entire UCL ventilation (Q*roof(out) ¼ 0.825 and Q*roof(in) ¼ 0.148), and the effective flow
capacity. rate induced by turbulence fluctuations (Q*roof(turb) ¼ 1.211). For
types of ‘Hung1.2H’ and ’Partly-covered’, roof ventilation capacity
3. Results and discussions significantly decreases, including Q*roof(out) ¼ 0.825 to 0.424
and 0.306, Q*roof(in) ¼ 0.148 to 0.116 and
3.1. Evaluation and validation of CFD results 0.008,Q*roof(turb) ¼ 1.211 to 1.059 and 0.258. Moreover Q* across
O1 decreases a little (1.048e0.999 and 0.950) due to the displace-
Fig. 7 shows the validation of CFD results by using the measured ment by semi-open street roofs, and Q* across O3 increases a little
horizontal profiles of velocity and turbulent intensity along street (0.551 to 0.684 and 0.685). These results show that semi-open
centerline at z ¼ 0.11H in Case [2e2.0, Open]. x/H ¼ 0 denotes the street roofs not only pose additional flow resistances and therefore
location of windward street opening (at O1). The velocity was reduce the ventilation by vertical mean flows and turbulence across
normalized by the inflow velocity at domain inlet at the same street roofs, but also influence the inflow rates and redistribution of
height (z ¼ 0.11H). In comparison to wind tunnel data, the standard airflows along the streets within UCL, especially driving more air
keε model and realizable keε model predicted the velocity profile across Street 3 (O3). Thus in contrast to Case [2e2, 0, Open], models
better than RNG keε model and RSM model. More importantly the with semi-open street roofs obtain much greater s*p and older air in
standard keε model performed the best in predicting airflow the secondary streets due to the weakened roof ventilation. An
328 J. Hang et al. / Building and Environment 70 (2013) 318e333

Fig. 11. Ventilation indices and their NVR for test cases with 4 buildings: (a) Q*roof (in) and Q*roof (out), (b) Q*roof (turb), (c) QT*, (d) PFR*, (e) hs*p i.
 J. Hang et al. / Building and Environment 70 (2013) 318e333 329

Fig. 12. s*p in z ¼ 0.22H in (a) Case [4e4, 0, Hung1.2H], (b) Case [4e4, 15, Hung1.2H], (c) Case [4e4, 30, Hung1.2H], (d) Case [4e4, 45, Hung1.2H].

extreme example is ’Fully-covered’ type, in which the flow rates both roof ventilation and overall UCL ventilation are improved
across street roofs are zero, and s*p in the secondary street (125e including Q*roof(out) varies from 0.547 (0 ) to 0.939
225) is much greater than that in the main street (0e45). The UCL (15 ), 0.919 (30 ) and 0.730 (45 ), Q*roof(in) changes from 0.106
spatial mean age of air hs*p i with ‘Open’ and ‘Hung1.2H’ types are (0 ) to 0.586 (15 ), 1.092 (30 ) and 1.041 (45 )), and hs*p i decreases
24.3 and 37.7, which is much smaller than hs*p i with ‘Partly-covered’ from 29.6 (0 ) to 22.6 (15 ), 18.9 (30 ) and 18.5 (45 ). These results
and ‘Fully-covered’ types (54.9 and 90.4), confirming that the confirm that 30 and 45 produce better UCL ventilation than
‘Hung1.2H’ type provide better overall UCL ventilation than ‘Partly- 0 and 15 .
covered’ and ‘Fully-covered’ types. As discussed and reported by the literature [2,8e11,18e
20,24,31,32,45],turbulent Schmidt numbers (Sct) may influence
3.2.2. Effect of ambient wind directions in four example test cases numerical results of pollutant dispersion. As displayed in Table 2,
Fig. 9 displays 3D streamline, s*p and Q* in Case [2e2, 0, the effects of different Sct and turbulence models are studied in Case
Hung1.5H], Case [2e2, 15, Hung1.5H], Case [2e2, 30, Hung1.5H] and [2e2, 0, Open] to quantify the sensitivity of turbulence models and
Case [2e2, 45, Hung1.5H]. The flow patterns are obviously different Sct on UCL ventilation: Sct ¼ 1.0, 0.7 and 0.4 are used in standard keε
and flow rates are redistributed. With a parallel approaching wind, model, Sct ¼ 0.7 in RNG keε model, and Sct ¼ 0.7 in Realizable keε
air enters UCL through O1, O2 and O4, then leaves through O3. model. With the same standard keε model and Sct of 1.0, 0.7 or 0.4,
Moreover 3D helical flows mainly exist in Street 2 and Street 4 hs*p i in the entire UCL are 26.4, 24.3 and 21.2, respectively, showing
where air is relatively old. With non-parallel approaching wind, air that smaller Sct may enhance pollutant dispersion by turbulent
enters UCLs across O1 and O2, then leaves through O3 and O4; diffusion and slightly reduce the age of air. With the same Sct of 0.7,
Recirculation flows exist in all four streets and s*p is relatively large realizable keε model and RNG keε model obtain different flow
in the downstream streets (Street 3 and Street 4) and in recircu- rates through O3 and street roofs which result in a little greater hs*p i
lation regions. If wind directions change from 0 to 15 , 30 , 45 , (27.2 and 28.2) than that by standard keε model (24.3). Especially
330 J. Hang et al. / Building and Environment 70 (2013) 318e333

Fig. 13. Ventilation indices and their NVR: (a) Q*roof (in) and Q*roof (out) in 24 test cases with 16 buildings, in all 48 test cases: (b) Q*roof (turb), (c) QT*, (d) PFR*, (e) hs*p i.
 J. Hang et al. / Building and Environment 70 (2013) 318e333 331

Q* across O3 predicted by RNG keε model is much smaller than 3.3. Ventilation assessment in test cases with sixteen buildings
those by the other two, which can be explained by the fact that RNG
keε model significantly over-predicts Q*roof(out) (1.127) than the What happen if urban size enlarges? To quantify this effect, test
other two (0.825 and 0.844). To be consistent, standard keε cases with 16 buildings are investigated, as summarized in Table 1.
model with Sct of 0.7 was selected as the default settings in CFD Fig. 12 displays normalized age of air in four test cases, i.e. Case [4e
simulations. 4, 0, Hung1.2H], Case [4e4, 15, Hung1.2H], Case [4e4, 30,
Hung1.2H], Case [4e4, 45, Hung1.2H]. The ventilation patterns are
3.2.3. Overall ventilation assessment in cases with four (2
2) similar with those consisting of 4 buildings. For wind direction of
buildings 0 , air mainly enters UCL across windward street openings of O1a,
To quantify the effect of semi-open street roofs on UCL venti- O1b, O1c, and leaves UCL through leeward openings of O3a, O3b,
lation flow rates, Fig. 10 shows Q* through O1eO4 and Q*roof(out), O3c. For wind directions of 15 , 30 , and 45 , air enters UCL through
Q*roof(in), Q*roof(turb) in all test cases with 4 buildings and wind O1a to O1c and O2a to O2c, then leaves UCL across O2a to O2c and
directions of 0 e45 . Roof types change from ‘Open’, ‘Hung1.5H’, O4a to O4c. Age of air is relatively large and air is old in recirculation
‘Hung1.2H’, ‘Hung1.1H’, to ‘Partly-covered’ and ‘Fully-covered’ regions and downstream regions.
(reading figure from left to right). Roof ventilations for ‘Fully- UCL ventilation indices and their normalized ventilation ratios
covered’ type are all zero. For wind directions of 0 and 15 (see (NVR) in all 24 test cases with 16 buildings are quantitatively
Fig. 10a and b), roof type variations result in a slightly decreasing analyzed, including Q*roof (in) and Q*roof (out) in Fig.13a, Q*roof (turb)
flow rates across O1 and an increasing flow rates across O3. More in Fig. 13b, QT* in Fig. 13c, PFR* in Fig. 13d and hs*p i in the entire UCL in
importantly, the flow rates across street roofs are all significantly Fig. 13e. It is found that UCL ventilation indices basically become a
weakened, including Q*roof(out) from 0.825 (0 ) and 1.156 (15 ) little better if wind directions change from 0 to 15 to 30 and 45 .
to 0, Q*roof(in) from 0.148 (0 ) and 0.619 (15 ) to 0, and Q*roof(turb) More importantly, roof type variations from ‘Open’ to ‘Fully-covered’
from 1.211 (0 ) and 1.315 (15 ) to 0. Moreover, Q* across O2 and O4 produce a large decreasing rate of overall UCL ventilation and obtain
are relatively small for wind direction of 0 (see Fig. 10a), but they macroscopically older air, which can be represented by the below
become considerably large for wind direction of 15 (see Fig. 10b). data. For roof ventilation indices (see Fig. 13a and b), NVR for ‘Fully-
For wind directions of 30 and 45 (see Fig. 10c and d), similar covered’ type are all zero, and those for ‘Partly-covered’ type are
findings exist due to such roof type variations that all roof venti- 11%e23% for Q*roof (in), 28%e39% for Q*roof (out), and 16%e22% for
lation indices decrease quickly and Q* across street openings Q*roof (turb). For overall UCL ventilation, NVR of QT* (see Fig. 13c) are
decrease a little. 81%e96% for ‘Hung1.5H’ type, 78%e87% for ‘Hung1.2H’ type, 65%e
To quantify the reduction of UCL ventilation as roof types 86% for ‘Hung1.1H’ type, 52%e61% for ‘Partly-covered’ type and
varying from ‘Open’ type to ‘Fully-covered’ type, the normalized 28%e50% for ‘Fully-covered’ type, and NVR of PFR*(see Fig. 13d) for
ventilation ratio (NVR) is defined as the value of ventilation indices the above roof types are 84%e90%, 76%e87%, 65%e86%, 52%e68%,
in a case divided by those with ‘open street roofs’ and the same and 36%e45% respectively, moreover NVR of hs*p i increase from 111%
wind direction. Thus for cases with open street roofs, NVR ¼ 1, and to 120%, 115%e131%, 116%e154% to 148%e192%, 223%e279% (i.e. air
Q* across street roofs for ‘Fully-covered’ roof type are all zero becomes older). Results also confirm that, ‘Hung1.5H’, ‘Hung1.2H’
(NVR ¼ 0). Fig. 11 displays Q*roof (in) and Q*roof (out), Q*roof (turb), and ‘Hung1.1H’ types produce a little smaller but comparable UCL
total normalized flow rates by mean flows (QT*), normalized UCL ventilation in contrast to ‘Open’ type. Thus for cases with 16 build-
purging flow rate (PFR*), hs*p i in the entire UCL, and their NVR ings, the roof types of ‘Hung1.2H’ and ‘Hung1.1H’ are better choices
values for all 24 cases with 4 buildings. With the same roof type, considering they are more realistic designs.
wind direction of 30 and 45 obtain greater Q*roof (in) and Q*roof
(turb), larger QT* and PFR*, smaller hs*p i, showing that 30 and 45 3.4. Effect of urban size on UCL ventilation
produce better UCL ventilation than 0 and 15 . In addition,
Fig. 11a and b also confirm that, all roof ventilation indices To quantify how overall UCL ventilations change if building
decrease as roof type varies from ‘Open’ to ‘Partly-covered’, and number or urban size increases, Fig. 13bee also compares Q*roof
NVR for ‘Partly-covered’ type are as small as 5.6%e34% for Q*roof (turb), QT*, PFR* and hs*p i between urban models with 4 or 16
(in), 18.0%e37.1% for Q*roof (out), and 21.3%e22.6% for Q*roof (turb) buildings (the smaller or bigger model). By analyzing Fig. 13bed,
respectively. Fig. 11c and d displays that overall UCL ventilation Q*roof (turb), QT* and PFR* in the bigger model are found several
basically decreases from ‘Open’ type to ‘Fully-covered’ type, indi- times (about 3.2e4.7 for Q*roof, 1.2e2.6 for QT*, 0.8e3.5 for PFR*)
cated by the fact as below: the NVR of QT* are 87%e99% for larger than those in the smaller model. Larger urban model obtains
‘Hung1.5H’ type, 81%e92% for ‘Hung1.2H’ type, 67%e78% for greater ventilation capacity because their total area of street
‘Hung1.1H’ type, 57%e72% for ‘Partly-covered’ type and 41%e62% openings and street roofs are 2 and 5.2 times greater than the
for ‘Fully-covered’ type; the NVR of PFR* are from 82% to 110%, smaller one. However it does not represent larger urban model can
64%e110%, 52%e104% to 44%e87% and 27%e64%, and the NVR of produces better overall UCL ventilation. It can be confirmed by
hs*p i are from 90% to 122%, 91%e155%, 96%e190% to 115%e226% Fig. 13e that hs*p i in the bigger model is about 1.4e3.5 times as great
and 156e373%. Overall, Fig. 11d and e confirm that roof types of as that in the smaller model, showing that the bigger model obtains
‘Hung1.5H’, ‘Hung1.2H’ and ‘Hung1.1H’ may produce relatively macroscopically older air. It is because the bigger model has a UCL
considerable UCL ventilation in contrast to ’Open’ type (i.e. NVR volume of 5.2 times larger than that in the smaller model and re-
are 52%e110% for PFR* and 91%e190%for hs*p i). Considering quires longer time for wind to flow through.
‘Hung1.1H’ and ‘Hung1.2H’ types are more realistic, they are pro-
posed as better semi-open street roof configurations. Meanwhile, 3.5. Discussions and future outlooks
Fig. 11d and e also verify that, if roof types change from ‘Open’ to
‘Fully-covered’, overall UCL ventilation with 0 wind direction may Further investigations are still required before formulating a
decrease much more significantly (NVR are 100%e27% for PFR*, practical guidelines for these semi-open street roof designs, such as
and 100%e372% for hs*p i) than the other wind directions, because the effect of the surrounding building height, the effect of atmo-
the secondary streets with 0 wind direction and semi-open street spheric thermal stratification (not neutral) and buoyancy force due
roofs tend to be poorly ventilated. to solar shading, the analysis of rain-cover and shading capability
332 J. Hang et al. / Building and Environment 70 (2013) 318e333

etc. This paper is one of the first attempts to quantify and address a k, ε turbulent kinetic energy and its dissipation rate
!
relationship between semi-open street roof configurations and UCL n normal direction of street openings or canopy roofs
ventilation indices. The methodologies and techniques utilized in NVR normalized ventilation ratio in contrast to models with
this paper are promising, and possibly provide a valid tool to ‘open’ street roofs
investigate UCL ventilation in other types of idealized or realistic PFR, PFR* purging flow rate and its normalized value (PFR* ¼ PFR/)
urban configurations. Q* normalized flow rate through street openings or street
roofs
4. Conclusions * , Q*
Qin out normalized total inflow and outflow rate for entire UCL
QT* total ventilation flow rate by mean flows (m3 s1)
The arrangements of semi-open street roofs in urban space are QN reference flow rate in upstream free flow to normalize
effective to protect pedestrians from strong sunshine and heavy flow rates
rains or snows. Their effects on urban canopy layer (UCL) ventila- * (turb) normalized effective flow rate across street roofs by
Qroof
tion are still not fully understood. This paper numerically quanti- turbulence
fied how five types of semi-open street roofs influence isothermal *
Qroof (in) normalized inflow rate across street roofs by downward
turbulent airflows and UCL ventilation performance under a neutral flows
atmospheric condition with various ambient wind directions (0 , *
Qroof (out) normalized outflow rate across street roofs by upward
15 , 30 , 45 ). Two small-scale idealized urban models were outflows
investigated consisting of 4 (2
2) or 16 (4
4) buildings with Sc pollutant release rate
uniform building height of H ¼ 0.069 m, and street aspect ratio of H/ Sct turbulent Schmidt number
W ¼ 1, corresponding to full-scale urban models of about 7 m tall, sw fluctuation velocity on street roofs
49 m and 105 m long as the scale ratio is 1:100. In contrast to ‘Open’ sp, s*p age of air (s) and its normalized value
roof type (open street roof), five kinds of semi-open street roofs hs*p i normalized spatial mean age of air
were included: Walls are hung above open street roofs (coverage Um, Im velocity, turbulence intensity measured in upstream free
ratio la ¼ 100%) at z ¼ 1.1H, 1.2H, 1.5H, i.e. types of ‘Hung1.1H’, flow
‘Hung1.2H’, ‘Hung1.5H’; Walls partly cover street roofs at z ¼ H U0(z) velocity profiles used at CFD domain inlet for ventilation
(la ¼ 80%), i.e. ‘Partly-covered’ type; Walls are set up to cover the cases
entire street roof at z ¼ H (la ¼ 100%), i.e. ‘Fully-covered’ type. The UH reference velocity (2.66 m/s) at z ¼ H
age of air and its spatial mean value, flow rates across street uj , xj velocity and coordinate components
openings and street roofs, the UCL purging flow rate were numer- !
V velocity vector
ically analyzed to quantify UCL ventilation.
Vol control volume
Results show that the prediction of airflow velocity by using
x, y, z stream-wise, span-wise, vertical directions
standard keε model agreed better with wind tunnel data than other
three RANS turbulence models. Semi-open street roofs significantly
influence UCL ventilation patterns and redistribute flow rates across
street openings and street roofs. As roof types vary from ‘Open’ to References
‘Hung1.5H’, ‘Hung1.2H’, ‘Hung1.1H’ then to ‘Partly-covered’ and
‘Fully-covered’, both roof ventilation and overall UCL ventilation [1] Oke TR. Street design and urban canopy layer climate. Energy Build
1988;11(1e3):103e13.
performance are basically weakened. The net UCL ventilation is the [2] Bady M, Kato S, Huang H. Towards the application of indoor ventilation effi-
worst for the ‘Fully-covered’ type, followed by the ‘Partly-covered’ ciency indices to evaluate the air quality of urban areas. Build Environ
type. The roof types of ‘Hung1.2H’ and ‘Hung1.1H’ are proposed 2008;43(12):1991e2004.
[3] Deng Q, He G, Lu C, Liu W. Urban ventilation e a new concept and lumped
because they produce comparable UCL ventilation, meanwhile are
model. Int J Vent 2012;11:131e40.
more realistic roof designs. Oblique ambient wind directions of 30 [4] Yang XY, Li YG, Yang LN. Predicting and understanding temporal 3D exterior surface
and 45 obtain better UCL ventilation than 15 and 0 . If the building temperature distribution in an ideal courtyard. Build Environ 2012;57:38e48.
[5] Kato S, Murakami S, Takahashi T, Gyobu T. Chained analysis of wind tunnel
number increases from 4 (2
2) to 16 (4
4), air in the entire UCL
test and CFD on cross ventilation of large-scale market building. J Wind Eng
becomes macroscopically older because the greater UCL volume Ind Aerodyn 1997;67e68:573e87.
requires longer time for rural wind to flow through. [6] da Graça GC, Martins NR, Horta CS. Thermal and airflow simulation of a
naturally ventilated shopping mall. Energy Build 2012;50:177e88.
[7] Kim T, Kim K, Kim BS. A wind tunnel experiment and CFD analysis on airflow
Acknowledgments performance of enclosed-arcade markets in Korea. Build Environ 2010;45:
1329e38.
This study was financially supported by the National Natural [8] Hang J, Li Y, Sandberg M, Buccolieri R, Di Sabatino S. The influence of building
height variability on pollutant dispersion and pedestrian ventilation in
Science Foundation of China (No. 51108102) and Guangdong Nat- idealized high-rise urban areas. Build Environ 2012;56:346e60.
ural Science Foundation (Code S2011040004149). as well as sup- [9] Hu T, Yoshie R. Indices to evaluate ventilation efficiency in newly-built urban
ported by the Fundamental Research Funds for the Central area at pedestrian level. J Wind Eng Ind Aerodyn 2013;112:39e51.
[10] Buccolieri R, Sandberg M, Di Sabatino S. City breathability and its link to
Universities (No. 2013390003165002). The two anonymous re- pollutant concentration distribution within urban-like geometries. Atmos
viewers who provided constructive suggestions and comments are Environ 2010;44(15):1894e903.
also gratefully acknowledged. [11] Hang J, Li Y, Buccolieri R, Sandberg M, Di Sabatino S. On the contribution of
mean flow and turbulence to city breathability: the case of long streets with
tall buildings. Sci Total Environ 2012;416:363e73.
Nomenclature [12] Britter RE, Hanna SR. Flow and dispersion in urban areas. Annu Rev Fluid
Mech 2003;35:469e96.
[13] Arnfield AJ. Two decades of urban climate research: a review of turbulence,
A area of a surface (m2)
exchanges of energy and water, and the urban heat island. Int J Climato
B,H, L,W building width, building height, total length, street width 2003;23:1e26.
c, hci time-averaged pollutant concentration(kg m3) and its [14] Grimmond CSB, Blackett M, Best MJ, Barlow J, Baik JJ, Belcher SE, et al. The
spatial mean value international urban energy balance models comparison project: first results
from phase 1. J Appl Meteorol Clim 2010;49:1268e92.
Kc, nt turbulent eddy diffusivity of pollutant and [15] Li XX, Liu CH, Leung DYC, Lam KM. Recent progress in CFD modelling of wind field
momentumKc ¼ nt/Sct and pollutant transport in street canyons. Atmos Environ 2006;40(29):5640e58.
 J. Hang et al. / Building and Environment 70 (2013) 318e333 333

[16] Meroney RN, Pavegeau M, Rafailidis S, Schatzmann M. Study of line source [31] Liu CH, Cheng WC, Leung TCY, Leung DYC. On the mechanism of air pollutant
characteristics for 2-D physical modelling of pollutant dispersion in street re-entrainment in two-dimensional idealized street canyons. Atmos Environ
canyons. J Wind Eng Ind Aerodyn 1996;62(1):37e56. 2011;45(27):4763e9.
[17] Li XX, Liu CH, Leung DYC. Numerical investigation of pollutant transport charac- [32] Hang J, Sandberg M, Li YG. Age of air and air exchange efficiency in idealized
teristics inside deep urban street canyons. Atmos Environ 2009;43(15):2410e8. city models. Build Environ 2009;44(8):1714e23.
[18] Salim SM, Cheah SC, Chan A. Numerical simulation of dispersion in urban [33] Santiago JL, Dejoan A, Martilli A, Martin F, Pinelli A. Comparison between
street canyons with avenue-like tree plantings: comparison between RANS large-eddy simulation and Reynolds-Averaged NaviereStokes computations
and LES. Build Environ 2011;46(9):1735e46. for the MUST field experiment. Part I: study of the flow for an incident wind
[19] Cai XM. Effects of differential wall heating in street canyons on dispersion and directed perpendicularly to the front array of containers. Boundary-Layer
ventilation characteristics of a passive scalar. Atmos Environ 2012;51:268e77. Meteorol 2010;135:109e32.
[20] Di Sabatino S, Buccolieri R, Pulvirenti B, Bitter R. Simulations of pollutant [34] Salim SM, Buccolieri R, Chan A, Di Sabatino S. Numerical simulation of at-
dispersion within idealised urban-type geometries with CFD and integral mospheric pollutant dispersion in an urban street canyon: comparison be-
models. Atmos Environ 2007;41(37):8316e29. tween RANS and LES. J Wind Eng Ind Aerodyn 2011;99(2e3):103e13.
[21] Kanda M. Large-eddy simulations on the effects of surface geometry of [35] Yoshie R, Mochida A, Tominaga Y, Kataoka H, Harimoto K, Nozu T, et al.
building arrays on turbulent organized structures. Boundary-Layer Meteorol Cooperative project for CFD prediction of pedestrian wind environment in the
2006;18(1):151e68. Architectural Institute of Japan. J Wind Eng Ind Aerodyn 2007;95:1551e78.
[22] Zaki SA, Hagishima A, Tanimoto J, Ikegaya N. Aerodynamic parameters of [36] Mochida A, Lun IYF. Prediction of wind environment and thermal comfort at
urban building arrays with random geometries. Boundary-Layer Meteorol pedestrian level in urban area. J Wind Eng Ind Aerodyn 2008;96(10e11):
2011;138:99e120. 1498e527.
[23] Abd Razak A, Hagishima A, Ikegaya N, Tanimoto J. Analysis of airflow over [37] Hang J, Sandberg M, Li Y. Effect of urban morphology on wind condition in
building arrays for assessment of urban wind environment. Build Environ idealized city models. Atmos Environ 2009;43(4):869e78.
2013;59:56e65. [38] FLUENT V6.3. User’s manualhttp://www.fluent.com; 2006.
[24] Yim SHL, Fung JCH, Lau AKH, Kot SC. Air ventilation impacts of the “wall ef- [39] Snyder WH. Similarity criteria for the application of fluid models to the study
fect” resulting from the alignment of high-rise buildings. Atmos Environ of air pollution meteorology. Boundary-Layer Meteorol 1972;3:113e34.
2009;43(32):4982e94. [40] Tominaga Y, Mochida A, Yoshie R, Kataoka H, Nozu T, Yoshikawa M, et al. AIJ
[25] Hagishima A, Tanimoto J, Nagayama K, Meno S. Aerodynamic parameters of guidelines for practical applications of CFD to pedestrian wind environment
regular arrays of rectangular blocks with various geometries. Boundary-Layer around buildings. J Wind Eng Ind Aerodyn 2008;96(10e11):1749e61.
Meteorol 2009;132(2):315e37. [41] Lien FS, Yee E. Numerical modelling of the turbulent flow developing
[26] Gu ZL, Zhang YW, Cheng Y, Lee SC. Effect of uneven building layout on air flow within and over a 3-D building array, part I: a high-resolution Reynolds-
and pollutant dispersion in non-uniform street canyons. Build Environ Averaged NaviereStokes approach. Boundary-Layer Meteorol
2011;46(12):2657e65. 2004;112(3):427e66.
[27] Chen Q. Ventilation performance prediction for buildings: a method overview [42] Irwin JS. A theoretical variation of the wind profile power-law exponent as
and recent applications. Build Environ 2009;44:848e58. a function of surface roughness and stability. Atmos Environ 1979;13(1):
[28] Luo ZW, Li YG. Passive urban ventilation by combined buoyancy-driven slope 191e4.
flow and wall flow: parametric CFD studies on idealized city models. Atmos [43] WMO guide to meteorological instruments and methods of observation
Environ 2011;45(32):5946e56. WMO-No. 8, page I.5-I12.
[29] Yang LN, Li YG. Thermal conditions and ventilation in an ideal city model of [44] Etheridge D, Sandberg M. Building ventilation: theory and measurement.
Hong Kong. Energy Build 2011;43(5):1139e48. Chichester: John Wiley & Sons; 1996p.573e633.
[30] Hang J, Li YG. Wind conditions in idealized building clustersemacroscopic [45] Tominaga Y, Stathopoulos T. Turbulent Schmidt numbers for CFD analysis
simulations by a porous turbulence model. Boundary-Layer Meteorol with various types of flow field. Atmos Environ 2007;41(37):8091e9.
2010;136(1):129e59.