Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Sustainable Cities and Society 75 (2021) 103307

Contents lists available at ScienceDirect

Sustainable Cities and Society


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

CFD simulations of wind flow and pollutant dispersion in a street canyon


with traffic flow: Comparison between RANS and LES
Xing Zheng a, b, Jiachuan Yang a, *
a
Department of Civil and Environmental Engineering, The Hong Kong University of Science and Technology, Hong Kong, China
b
Unit Building Physics & Services, Department of the Built Environment, Eindhoven University of Technology, Eindhoven, The Netherlands

A R T I C L E I N F O A B S T R A C T

Keywords: The dispersion of traffic pollutants in street canyons takes place under the joint influence of the urban
Traffic emissions morphology and traffic-induced air motions. Computational fluid dynamics (CFD) are increasingly used to
Reynolds-averaged Navier-Stokes (RANS) explore air quality problems in cities, however, numerical setup guidance for studies considering moving traffics
Moving traffic
remains missing. This study presents a systematic evaluation and comparison of steady Reynolds-averaged
Pollutant dispersion
Large-eddy simulation (LES)
Navier-Stokes (RANS) and large-eddy simulation (LES) in reproducing the wind flow and pollutant dispersion
Vehicle-induced turbulence in a street canyon under traffic flow using the quasi-steady method. The RANS simulations are performed with
Urban ventilation five turbulence models, and the standard k-ε (SKE) turbulence model is found to yield the best agreement with
air quality the wind-tunnel data. The LES simulation with the wall-adapting local eddy-viscosity subgrid-scale model out­
performs all RANS models in simulating the mean velocity and mean pollutant concentration, and can accurately
predict the turbulent velocity. The counter-gradient turbulent mass flux captured by LES reveals that the
gradient-diffusion hypothesis used in steady RANS fails in the regions near moving traffics, which strongly affects
the predicted mean concentration. The findings of this study can support future CFD studies on pollutant
dispersion in urban environments.

List of acronyms 1. Introduction


ABL Atmospheric boundary layer
CFD Computational fluid dynamics Street canyons, which play important roles in regulating the near-
CFL Courant–Friedrichs–Lewy field pollutant dispersion, have been subject to intense study in recent
DMU Dynamic mesh updating method
FB Fractional bias
years (Hang et al., 2019; Miao et al., 2020; You & Ding, 2021). As the
LES Large-eddy simulation presence of canyon buildings prevents the wind from penetrating inside,
LKE Non-linear low-Reynolds k-ε model high concentrations of traffic-exhaust gases are usually observed in
MAE Mean absolute error street canyons with low wind velocity, especially under the perpendic­
NMSE Normalized mean square error
ular approaching wind (Kastner-Klein et al., 2004; Oke, 1988). The
Q-S Quasi-steady method
RANS Reynolds-averaged Navier-Stokes street-level pollutant does not only influence the outdoor air quality but
RNG Renormalization group k-ε model also enters buildings nearby and considerably influences the indoor air
RKE Realizable k–ε model quality (Shi & Zhao, 2016; Shi et al., 2015).
RSM Reynolds stress model Moving traffics add another mechanical process that modifies the
SF6 Sulfur hexafluoride
SGS Subgrid-scale model
flow field in street canyons, and the modification is found more evident
SST Shear stress transport k-ω model in the bottom areas of street canyons (Qin & Kot, 1993) and under low
SKE Standard k-ε model approaching wind speed (Solazzo et al., 2011; Zhao et al., 2021). The
WT Wind tunnel dispersion of traffic pollutants in street canyons takes place under the
WALE Wall-adapting local eddy-viscosity
joint influence of the background wind and vehicle-induced air motions
(Ahmad et al., 2005; Mazzeo & Venegas, 2005). Among the extensive
studies on the air quality inside street canyons, a significant portion of

* Corresponding author.
E-mail address: cejcyang@ust.hk (J. Yang).

https://doi.org/10.1016/j.scs.2021.103307
Received 24 May 2021; Received in revised form 24 August 2021; Accepted 25 August 2021
Available online 28 August 2021
2210-6707/© 2021 Elsevier Ltd. All rights reserved.
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

them did not consider the effect of moving vehicles (Chatzimichailidis

- Information not available, Sct = Turbulent Schmidt number, RKE = Realizable k–ε model, LKE = Non-linear low-Reynolds k-ε model, SKE = Standard k-ε model, RNG = Renormalization group k-ε model, V = Mean
Modeling method
et al., 2019; Han et al., 2018; Li et al., 2010; Scungio et al., 2018; Zhang
et al., 2011). A past study has shown the lack of accurate traffic-induced
turbulence parameterizations can lead to overestimated pollutant con­
centrations at the street level (Kastner-Klein et al., 2003). Therefore,

DMU

DMU

DMU
DMU

DMU
Q-S
Q-S

Q-S
Q-S

Q-S

Q-S

Q-S

Q-S
investigation on the traffic-induced airflow and turbulence is crucial for
-

understanding the pollutant dispersion characteristics in urban street


canyons.
Previous studies have explored the wind flow in street canyons using
Investigated parameter

controlled wind-tunnel testing (Baker and Hargreaves, 2001; Gromke


and Ruck, 2007; Kastner-Klein et al., 2001), full-scale on-site measure­
ments (DePaul & Sheih, 1986;Solazzo et al., 2007), and computational
fluid dynamics (CFD) simulations. Compared to wind-tunnel testing and
V, C, TP
TP, V
V, TP

V, TP

V, TP
C, TP

TP, C

on-site measurements, CFD simulations have the ability to provide


V, C

V, C
V, C
V, C
C

high-resolution data in the whole computational domain (Blocken et al.,


2011), therefore, have been widely used. Table 1 gives an overview of
CFD research on the wind flow for street canyons considering moving
Yes (Kastner-Klein et al., 2001; Niachou et al., 2008)

traffics. The table shows that the majority of the studies have focused on
generic canyon cases and analyzed the mean velocity, mean concen­
tration, and turbulent parameters. In modeling moving traffics, the dy­
namic mesh updating (DMU) method and quasi-steady (Q-S) method
have been used. The DMU method features dynamic changes of the
Yes (Kovar-Panskus et al., 2002) a

computational mesh with time, and consequently allows the simulation


Yes (Kastner-Klein et al., 2001)

Yes (Kastner-Klein et al., 2001)

of flow fields with temporally varying geometry (Wang et al., 2019). The
Yes (Gallagher et al., 2013) a
Yes (Gromke et al., 2008) a

Yes (Uehara et al., 2000) a

Q-S method (Jicha et al., 2002), also named the Eulerian-Lagrangian


velocity, C = Mean concentration, TP = Turbulence parameters, Q-S = Quasi-steady method, DMU = Dynamic mesh updating method.

method (Katolický & Jícha, 2005), models the effect of moving traffics
on airflow by calculating the drag force between moving traffics and
surrounding ambient air. Compared to the DMU method, the Q-S
method is less computationally demanding and more suitable for CFD
Validation

simulations with the focus on a relatively large scale, i.e. urban scale
(Zhao et al., 2021).
No
No

No
No

No

No

No

Table 1 clearly indicates the majority of the past studies have


employed the Reynolds-Averaged Navier-Stokes (RANS) approach,
while the use of the large-eddy simulation (LES) approach has been
1.3, 0.7

scarce and limited to Ref. (Zhang et al., 2017). For these RANS studies,
0.7
Sct

standard k-ε (SKE), non-linear low-Reynolds k-ε (LKE), realizable k–ε


-
-
-

-
-
-

-
-
-
-
-
-

turbulence model (RKE), and renormalization group (RNG) k-ε turbu­


lence models have been applied. However, comparative analysis on the
Turbulence model

performance of different turbulence models with experimental data has


not yet been conducted. Only a few studies using RANS have validated
Summary of previous CFD studies of wind flow in street canyons with moving traffics.

their results in street canyons under moving traffics. Moreover, valida­


tion is missing in the LES study. The lack of validation casts uncertainties
RNG

RNG
RNG
RNG
RNG
RKE
LKE
LKE

LKE
SKE
SKE

SKE

over the accuracy and reliability of the reported findings.


-

Pollutants emitted in the urban area are transported by turbulent


the validation only considered street canyons with standing vehicles.

flows around the buildings (Kubilay et al., 2017; Liu & Wong, 2014). The
CFD Approach

distribution of pollutant concentration is regulated by the combination


of molecular, convective, and turbulent diffusion, where the molecular
diffusion is often negligibly small compared with the two others. Pre­
RANS
RANS
RANS
RANS

RANS
RANS
RANS
RANS
RANS
RANS
RANS
RANS
RANS
LES

vious studies have shown that the prediction of the mean pollutant
concentration is considerably dependent on the turbulence modeling
approach (Tominaga & Stathopoulos, 2013, 2012, 2011). In simulations
with the RANS approach, the mean flow is resolved and the effect of
Generic or applied

turbulence (fluctuating components) is modeled using a turbulence


model. The turbulent mass transport is generally assumed with the
gradient-diffusion hypothesis to estimate the turbulent diffusion flux
Applied

Applied

Applied

Applied
Generic
Generic
Generic

Generic
Generic

Generic
Generic
Generic
Generic

Generic

<ui′ c′ >, i.e.:


∂C
(1)
′ ′
< ui c >= − Dt
∂xi
Kondo & Tomizuka, 2009
Zhang, Gu and Yu, 2017

where ui′ and c′ are the velocity and concentration fluctuations, i in­
Pospisil & Jicha, 2017

Pospisil & Jicha, 2019

Pospisil et al., 2020


Solazzo et al., 2008

dicates the coordinate (ux, uy, uz = u, v, w), Dt is the turbulent mass


Wang et al., 2019
Jicha et al., 2000
Jicha et al., 2002

Zhao et al., 2021


Cai et al., 2020

Shi et al., 2020


Jin et al., 2017

diffusivity, <⋅> indicates an averaging operator in time, C = <c> is the


Li et al., 2017

mean (time-averaged) concentration. CFD simulations of mass transfer


Reference

using RANS are further complicated by the fact that the turbulent
Table 1

Schmidt number (Sct) is selected prior to the simulation (Blocken et al.,


a

2
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

2008). Sct is defined as the ratio of computed turbulent viscosity vt to Dt: 2. Methodology
vt
Sct = (2) 2.1. Wind-tunnel experiment
Dt
As recorded in Table 1, different values of Sct have been used in past The CFD simulations in this study are validated using data from a
RANS studies (Kondo & Tomizuka, 2009; Li et al., 2017). Nonetheless, previous wind-tunnel experiment (Gromke & Ruck, 2009a, 2007). The
the impact of Sct on the reproduction of pollutant dispersion by RANS in experiment was conducted in an atmospheric boundary layer (ABL)
a street canyon under traffic flow has not been evaluated. Compared to wind tunnel with a cross-section of 1 m × 2 m (height × width) (Gromke
RANS, LES is able to provide more accurate descriptions of unsteady and Ruck, 2009a, 2007). The measured data from the street canyon in a
flow fields (Liu et al., 2019; Murakami et al., 1987; Rodi, 1997; Shirzadi, 1:1 aspect ratio (i.e., the building height equals the street width) with
Mirzaei, & Tominaga, 2020; Tominaga & Stathopoulos, 2013; Zheng two-way traffics were used.
et al., 2020) and turbulent mass transport process (Gousseau et al., 2011; As illustrated in Fig. 1a and 1b, the street canyon model was
Zhou et al., 2021) in the built environment, although it is achieved with composed of two buildings with width × depth × height = 1.2 × 0.12 ×
a much larger computational cost (Blocken, 2018). For the flow field in a 0.12 m3 each (at the scale of 1:150). The approaching wind was
street canyon with moving traffics, the performance of RANS and LES perpendicular to the canyon length axis. Fig. 2 shows the profiles of
has not been systematically compared before. Given the increasing mean wind speed and turbulent kinetic energy in the WT. Note that the
number of CFD studies on urban air pollution in recent years, a turbulent kinetic energy was estimated based on the measured mean
comparative study is imperative to provide a guidance for future nu­ wind speed and turbulent intensity. The mean wind speed profile rep­
merical setups. Therefore, the objective of this study is to evaluate the resents power-law correlations with an exponent of 0.30 (Gromke,
performance of steady RANS and LES in modeling the wind flow and 2008). Two circulating belts were employed to model the turbulent air
pollutant dispersion in street canyons with moving traffics. motions induced by traffic movement along the two lanes: windward
In this study, the quasi-steady (Q-S) method (Katolický & Jícha, lane and leeward lane (Fig. 1b). Small plates with a distance of 0.1 m
2005) is applied to model the effect of moving traffics. A systematic were mounted on two circulating belts, representing a vehicle density of
validation and evaluation for 3D steady RANS and LES simulations of 6.67 cars per 100 m in the full scale (Fig. 1c). The moving speed of
the wind flow and pollutant concentration are conducted. The 3D steady circulating belts is 11.1 m/s (Gromke & Ruck, 2009a).
RANS simulations are performed with various turbulence models and To model the release of traffic exhausts, the tracer gas, i.e., sulfur
turbulent Schmidt numbers. hexafluoride (SF6) was emitted from four wirelike sources mounted at
the ground (red points in Fig. 1a) (Gromke & Ruck, 2009b). 49 con­
centration measurement taps along 7 vertical lines were used on each
near-street building wall (i.e., 98 taps in total). These taps were placed 5
mm in front of the walls (Buccolieri et al., 2009). The relative locations
of the measurement taps to the walls are detailed in Sections 3 and 4.
The non-dimensional mean SF6 concentration was calculated following:
CUref H
C+ = (3)
Q/l

where Uref (= 4.65 m/s) was the mean wind velocity of the approach-
flow at the building roof height H, C was the mean SF6 concentration,
and Q/l was the SF6 emission rate per unit length. The mean and tur­
bulent velocity components were measured in vertical planes across the
canyon using a Laser-doppler velocimetry.

2.2. CFD simulations

2.2.1. Computational domain and grid


All the CFD simulations are conducted at the WT scale. Fig. 3 displays
the computational domains and grids for RANS and LES, respectively.
Following the size of the WT test section, the domain height and width

Fig. 1. Model of the street canyon in the WT, (a) side view and locations of Fig. 2. Profiles of (a) mean wind speed (U) and (b) turbulent kinetic energy (k),
wirelike sources, (b) top view, and (c) small plates with circulating belt. modified from (Gromke, 2008).

3
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

Fig. 3. (a) Computational domain, (b) perspective view of the near-canyon grid, and (c) cross-section view of the near-canyon grid for RANS simulations. (d-f) Same
as (a-c) but for LES.

are 1 m and 2 m, respectively. The total streamwise domain length is 2.2.2. Boundary conditions
3.36 m (28H, H is the roof height), which consists of an approach flow The bottom, top, and lateral boundaries and the surfaces of the
distance of 1.2 m (10H), a model distance of 0.36 m (3H), and a canyon buildings use no-slip walls. This is to reproduce the ground,
downwind distance of 1.8 m (15H). ceiling, sidewalls of the WT test section, and the canyon model surfaces.
A grid for RANS simulations is generated according to the best The mean wind velocity and turbulent kinetic energy inlet boundary
practice guidelines (Franke et al., 2007), resulting in a total cell number conditions take the profiles in the WT (Fig 2a and b). The turbulence
of 6.03 million. In the near building region, cubic cells with Δx = Δy = dissipation rate ε (Richards and Hoxey, 1993) follows:
Δz = 0.003 m (H/40) are used, which extends to a distance of 1/4H
u*3
away from the canyon windward, leeward, and roof surfaces and 1H ε(z) = ABL
(4)
κ(z + z0 )
away from lateral surfaces (see Figs. 3d and 3e). In the outer region, i.e.,
outside the near-building region, hexahedral cells with stretching ratios
where u*ABL the ABL friction velocity and z0 the roughness length take
below 1.2 are used. An exception is the Δz of ground-adjacent cells, in
0.59 m⋅s− 1 and 0.004 m, respectively (obtained from logarithmic law
which 0.009 m (3H/40) is applied (Fig. 3c). The assignment of specific
curve fitting of the mean velocity profile in WT (Gromke, 2008)). The
height to wall-adjacent cells is to satisfy the requirements of using wall
von Karman constant κ is considered to be 0.42. Note that for the RANS
functions to model the WT ground roughness, which has been adopted in
with SST k-ω model, the specific dissipation rate (ω) takes:
previous studies (Blocken et al., 2007; van Druenen et al., 2019).
To reduce the total computational cost, a non-conformal grid is used ε(z)
ω(z) = (5)
for LES (Iousef et al., 2017). It includes three parts, an inner subdomain Cμ k(z)
Ω1, a middle subdomain Ω2, and an outer subdomain Ω3. Ω1 extends up
to 1/4H away from canyon windward, leeward, and roof surfaces (see where Cµ is an empirical constant equals 0.09, and k is the turbulent
Figs. 3c and 3d) and 5/4H away from the lateral side, which covers the kinetic energy. For the LES simulation, the vortex method (Sergent,
region with circulating belts. Cubic cells with Δx = Δy = Δz = H/40 are 2002) is used as the inflow fluctuating algorithm at the inlet plane of the
used in Ω1. Ω2 extends from Ω1 to 1H away from the canyon windward, domain, which introduces perturbations to the mean velocity profile by
leeward, and roof surfaces and 2H away from the lateral surfaces. It randomly generating a certain number of 2D vortices.
consists of cubic cells with Δx = Δy = Δz = H/20. Ω3 uses hexahedral For the RANS simulations with the k-ω model, an automated wall
cells and the stretching ratios are below 1.05. The refinement ratio of the treatment is applied at the no-slip wall surfaces. For other RANS models,
grids between adjacent subdomains is 1:2. The total number of cells of the standard wall functions (Cebeci & Bradshaw, 1977; Launder &
this LES grid is 3.94 million. Grid resolutions for both RANS and LES Spalding, 1974) are applied. The sand grain roughness height ks and the
simulations are adopted from a grid-sensitivity analysis, in which fine, roughness constant Cs are determined based on their consistent relation
basic, and coarse grids are generated based on the same grid topology for with the aerodynamic roughness length z0 (Eq. (6)) (Blocken, Statho­
RANS and LES. The details of the grids, computational settings, and poulos and Carmeliet, 2007):
results of the grid-sensitivity analysis can be found in the Supplementary 9.793z0
Materials. ks = (6)
Cs

4
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

In ANSYS Fluent, the near-wall grid size should satisfy zp > ks, where shear stress transport (SST) k-ω model (Menter, 1994), (3) standard k-ε
zp is the center point of the ground-adjacent cell (=Δz/2). In the outer model (SKE) (Launder & Spalding, 1974), (4) renormalization group
region where zp = 0.0045 m, z0 of the WT profile (= 0.004 m) is used (RNG) k-ε model (Yakhot et al., 1992), and (5) realizable k-ε model
together with ks = 0.004 m to maintain an constant Cs of 9.793 and to (RKE) (Shih et al., 1995). The turbulent Schmidt number (Sct) of 0.7 is
satisfy zp > ks. For the near building region where zp = 0.0015 m, z0 = used for all the models. Note that to investigate the influence of Sct,
0.0014 m and ks = 0.0014 m are applied. This is to represent the less additional SKE simulations are performed with different Sct, which will
rough ground near the buildings in the WT and to keep sufficient grid be detailed in subsection 3.2. In all the RANS simulations, the
resolution near the buildings. For LES simulations, the Werner-Wengle pressure-velocity coupling uses SIMPLE algorithm. Pressure interpola­
wall functions (Werner & Wengle, 1993) are employed. tion is second order. Both the convection terms and the viscous terms of
For both LES and RANS simulations, wirelike SF6 (dynamic viscosity the governing equations take second-order discretization schemes.
1.42 × 10-4 kg⋅m-1s-1; molecular weight 146.055 kg⋅kmol-1; Diffusivity Convergence is assumed to be achieved once all the scaled residuals
2.3 × 10-5 m-2s-1) sources with an area of 1.42 m × 0.003 m are mounted reach a minimum value. These minimum values vary for different tur­
on the ground, following their locations in the WT (Fig. 1a). Following bulence models, e.g., the minimum values for the simulation with SKE
previous CFD validation studies (Gromke et al., 2008; Zheng et al., are 10− 4 for continuity, 10− 7 for x, y, and z momentum, and 10− 6 for k,
2021), the emission rate (Q) of SF6 is 10 g/s. ε, and SF6. Note that simulations with RNG and RSM show minor
oscillatory convergence, which requires averaging from a sufficient
2.2.3. Modeling the effect of moving traffics number of iterations (Ramponi & Blocken, 2012).
The quasi-steady (Q-S) method (Katolický & Jícha, 2005) is used to The LES simulation launches from the results of a preceding steady
model the effect of moving traffics. This method is based on the RANS simulation with RKE and Sct of 0.7 (Tominaga & Stathopoulos,
assumption that the force of aerodynamic resistance acting against the 2007). The wall-adapting local eddy-viscosity (WALE) subgrid-scale
vehicle is transferred to the ambient air as a momentum source. The (SGS) model is applied (Ducros, Nicoud and Poinsot, 1998). This
aerodynamic resistance force acting upon a vehicle is given by (Kato­ model was successfully used in validation studies for urban environ­
lický & Jícha, 2005; Pospisil & Jicha, 2017; Zhang et al., 2017): ments in the past (Ai & Mak, 2015; Bazdidi-Tehrani et al., 2019). A
⃒ ⃒( ) second-order upwind scheme is used for the energy and SF6 concen­
1
Fd.i = ρCD Aveh ⃒Uveh,i − ui ⃒ Uveh,i − ui (7) tration equations. For the momentum equation, the bounded central
2
differencing scheme is adopted. For pressure interpolation and time
where Fd.i is the aerodynamic resistance force in i direction, Uveh.i is the discretization, second-order schemes are applied. The non-iterative
moving velocity of vehicles in i direction, ui is the wind velocity in i fractional step method, with the time step set to 1.9 × 10− 4 s, was
direction, ρ is the density of the gas mixture, Aveh is the cross-sectional used for time advancement. The maximum Courant-Friedrichs-Lewy
area in i direction and CD is the drag coefficient. (CFL) number is 0.97. Ahead of the data sampling, the LES simulation
In the present study, the vehicles move along the y-direction and of an initialization period for 7.3 s is conducted to remove the depen­
cause the aerodynamic resistance force in this direction, i.e., Fd.y. The dence of results on the initial state (Gousseau et al., 2012; Ong et al.,
moving velocity of vehicles (Uveh,y) in the windward lane and leeward 2020). After this period, the results are sampled and averaged for
lane takes 11.1 m/s and -11.1 m/s, respectively (Gromke and Ruck, another 22.8 s (the monitoring of statistical convergences of LES can be
2009a). CD takes 1.2 and the Aveh takes 50 mm2, following the WT setup found in the Supplementary Materials).
(Gromke & Ruck, 2007). Each vehicle, i.e., small plates in the WT, All simulations are conducted on the Chinese national supercom­
moving along the traffic path would dynamically occupy a number of puter Tianhe-2 (www.nscc-gz.cn). A node equipped with 24 Intel Xeon
control volumes. The control volumes along the trajectory of moving CPU E5-2692 v2 2.2 GHz cores is used for each simulation.
vehicles are the directly affected region (illustrated in Fig. S1 in Sup­
plementary Materials), where the additional momentum source Su.y is 3. RANS results
calculated:
3.1. Impact of turbulence model
nveh Fd.y
Su.y = (8)
Acel This section gives a detailed comparison of the results obtained by
different RANS turbulence models. To quantify the agreement of CFD
where nveh is the number of vehicles per unit length, Acel is the cross-
results to the experimental (WT) results, three validation metrics are
section of control volumes passed by vehicles. Here, the additional
used: mean absolute error (MAE), normalized mean square error
momentum source term Su.y is added to the y-momentum transport
(NMSE) and fractional bias (FB) (Schatzmann, Olesen and Franke,
equation for both RANS and LES. For RANS, the transport equation of
2010). The metrics are calculated using Eqs. (10), (11), and (12):
turbulent kinetic energy (k) is solved, therefore, additional turbulent
kinetic energy Sk is added to the k equation into the directly affected
region, following Refs. (Ahmad et al., 2005; Katolický & Jícha, 2005):

nveh ρCc (Uveh.i − ui )2 Qveh


Sk = (9) Table 2
Acel
Validation metrics for non-dimensional mean y-velocity component (V/Uref),
non-dimensional mean x-velocity component (U/Uref), and non-dimensional
where the volume of the vehicle Qveh (equals 150 mm3) is based on the
mean z-velocity component (W/Uref): MAE.
WT setup (Gromke & Ruck, 2009a). The constant Cc of 0.41 follows
Ref. (Katolický & Jícha, 2005). Approach Turbulence model Vertical corner plane Vertical center plane
(y/H = -5) (y/H = 0)
V/Uref U/Uref U/Uref W/Uref
2.2.4. Solver settings MAE MAE MAE MAE
The commercial CFD software ANSYS Fluent 19 is employed and
RANS RSM 0.14 0.13 0.41 0.02
additional user-defined functions are used to simulate the effect of SST 0.17 0.31 0.49 0.03
moving traffics. The 3D steady RANS equations are solved with different SKE 0.16 0.08 0.05 0.04
turbulence models to compare their performances. One second-order RKE 0.15 0.11 0.15 0.03
closure model and four two-equation eddy-viscosity models are RNG 0.17 0.23 0.29 0.03
LES WALE 0.07 0.07 0.08 0.02
applied: (1) Reynolds stress model (RSM) (Wilcox & others, 1998), (2)

5
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

[ ]
|PExp − PCFD | lane (Gromke & Ruck, 2009a), e.g., higher measured C+ along lines y/H
MAE = (10) = -4.92 and -3.75 than lines y/H = 4.92 and 3.75, respectively. Very
n
different results are given by the five RANS turbulence models. The
[( )2 ]
PExp − PCFD spanwise concentration decays and traffic-induced concentration
NMSE = [ ] (11) skewness are only reproduced by SKE and RKE. For all the three metrics
PExp [PCFD ]
for the leeward wall, SKE yields the best performance, and RNG yields
([ ] ) the worst performance (Table 3). The good performance of SKE is in line
2 PExp − [PCFD ]
FB = [ ] (12) with a previous study on the street canyon without moving vehicles
PExp + [PCFD ]
(Salim et al., 2011b). For the windward wall, RKE and SKE show rela­
where PExp is the experimental (measured in the WT) value, PCFD is the tively good agreements with the experiment, (see Table 3). The other
CFD value, n is the number of data points, and the square brackets turbulence models considerably overestimate the C+ along the seven
indicate averaging over the data points. Table 2 presents the results of lines, and the worst performance is found for SST in terms of all the three
velocity components between CFD with different turbulence models and metrics. Considering the results of the three metrics for both windward
the WT experiment. Note that FB and NMSE could not be used for var­ and leeward walls, SKE has the overall best performance.
iables that have both positive and negative values within the same set.
Therefore, they are not depicted for velocity components. Table 3 pre­ 3.2. Impact of turbulent Schmidt number
sents the results of the mean concentration between CFD with different
turbulence models and the WT experiment. To explore the impact of turbulent Schmidt number (Sct), simula­
Fig. 4a-d show a comparison of the non-dimensional mean y-velocity tions with Sct of 0.3, 0.5, 0.7, and 0.9 are conducted using SKE, which
component (V/Uref) obtained from five RANS turbulence models against better predicts the C+ than the other investigated RANS models. Fig. 7
the measured data along horizontal lines in the vertical corner plane (y/ illustrates the effect of Sct on C+ along seven vertical lines for the
H = -5). For V/Uref below the roof height (z/H = 0.25, 0.5, and 0.75), all leeward and windward walls. Simulations with large Sct show an over­
the five RANS models show clear deviations from the experimental data, predicted traffic-induced skewness, e.g., C+ along lines y/H = -4.92 and
which significantly underestimate the positive V/Uref that entering the -3.75 are much higher than lines y/H = 4.92 and 3.75, respectively. A
canyon through the lateral boundary (see Fig. 4a-c). Comparison results decrease of Sct would increase the turbulent mass diffusivity Dt (= νt/
of the non-dimensional mean x-velocity component (U/Uref) are shown Sct), therefore resulting in a decrease of near field pollutant concentra­
in Fig. 4e-h. A large discrepancy between the simulated U/Uref from five tion and also the skewness. By changing Sct from 0.9 to 0.3, the MAE
RANS models is observed. Note that in RANS results, U/Uref is forced to reduces from 28.9 to 8.6 for the leeward wall and from 4.6 to 2.8 for the
be zero on the wall. This yields jump declines at the positions about H/ windward wall. This sensitivity analysis indicates that 0.3 is a reason­
40 (Δx) to the walls, especially along the line z/H = 1 (Fig. 4h). SST and able value for Sct in this case.
RNG yield large deviations from the experiment or even predict an
opposite direction to the experimental data (Fig. 4e and 4f). SKE pro­ 4. LES results
vides a relatively good agreement for U/Uref with the experiment
(Table 2). The general flow pattern in the canyon corner below the roof Fig. 8 shows the non-dimensional mean velocity components in the
height is better reproduced by SKE (better prediction of V and U). vertical corner plane from the LES simulation with the WALE SGS
Fig. 5a-c show a comparison of the non-dimensional mean x-velocity model. RANS results with SKE and RKE turbulence models are added as
component (U/Uref) obtained from the five RANS turbulence models the best-performing cases for a direct comparison with LES results. For
with the experimental data along three horizontal lines in the vertical V/Uref, LES results show a good agreement with the experiment.
center plane (y/H = 0). SKE has the smallest MAE of 0.05 from the Compared to the best-performing RANS results, LES yields a smaller
experiment, followed by RKE of 0.15 (see Table 2). Other RANS models deviation, especially in the lower region (z/H = 0.25 and 0.5) close to
substantially underpredict U/Uref or even predict an opposite direction the traffic trajectory. For U/Uref, LES accurately reproduces the value
to the experimental data (negative U/Uref by RSM and SST). Fig. 5d-f along the lines with a smaller deviation compared to the best-
show that all the five RANS turbulence models give fairly good agree­ performing RANS models (Table 2).
ments to the non-dimensional mean z-velocity component (W/Uref) with One important advantage of LES over RANS is its capability in
small MAE (see Table 2). simulating turbulent velocity, i.e., the rmse velocity component (ui.rmse,
Fig. 6 displays the profiles of the non-dimensional mean concentra­ defined in Eq. (13), where t is the time step after sampling, ui is the
tion (C+) from the five RANS turbulence models and the experiment velocity component and Ui is the mean velocity component). Fig. 9
along seven vertical lines for the leeward and windward walls. Due to shows the comparison of non-dimensional y- and x- rmse velocity
ventilation through the lateral boundary of the canyon, the measured C+ components (vrmse/Uref and urmse/Uref) in the vertical corner plane be­
decays along the spanwise direction (y-direction) towards the canyon tween LES results and experimental data. A good agreement of the vrmse/
corners. The measured C+ distribution also shows the traffic-induced Uref is observed along the four lines (Fig. 9a-d). For the urmse/Uref, LES
skewness, which shifted towards the traffic direction of the leeward reproduces the value inside the canyon with a fairly good agreement (z/
H = 0.25, 0.5, and 0.75) and slightly overpredicts the value at the roof
Table 3 height (z/H = 1).
Validation metrics for non-dimensional mean concentration (C+): MAE, RMSE, √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
and FB. 1∑ n
[ ]
ui.rmse = (ui − Ui )2 (13)
Approach Turbulence C+ for the leeward wall C+ for the windward
n t=1
model a wall
MAE RMSE FB MAE RMSE FB Fig. 10 compares the non-dimensional mean velocity components in
the vertical center plane from LES, RANS, and the WT experiment. LES
RANS RSM 28.1 1.55 -0.84 21.7 16.49 -1.51
SST 34.1 1.67 -0.94 39.0 17.16 -1.69
has comparable performance in simulating U/Uref with the best-
SKE 23.1 0.87 -0.75 3.9 0.68 -0.71 performing RANS model. For W/Uref, LES results agree with the exper­
RKE 26.5 1.18 -0.81 3.7 0.77 -0.69 imental data better than the two RANS models. At the roof height (z/H
RNG 34.8 1.94 -0.95 6.0 2.57 -0.92 = 1), LES captures the slope of W/Uref from the leeward side to the
LES WALE 3.0 0.03 -0.13 1.6 0.20 -0.37
windward side, while RANS models significantly underestimate the
a
The metrics of RANS models are obtained from simulations with Sct = 0.7. slope.

6
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

Fig. 4. Comparison of V/Uref obtained from five different RANS models against the experiment along four horizontal lines: (a) z/H = 0.25; (b) z/H = 0.5; (c) z/H =
0.75; (d) z/H = 1 in the vertical corner plane (y/H = -5). (e-h) Same as (a-d) but for U/Uref.

Fig. 5. Comparison of U/Uref obtained from five different RANS models with the experiment along three horizontal lines: (a) z/H = 1; (b) z/H = 1.08; (c) z/h = 1.17
in the vertical center plane (y/H = 0). (d-f) Same as (a-c) for W/Uref.

Fig. 11 shows the comparison of non-dimensional rmse x- and z- MAE of LES along the seven lines is 3, which is clearly smaller than that
velocity components (urmse/Uref and wrmse/Uref) in the vertical center of SKE with Sct = 0.3, i.e. 8.6. For the windward wall, a good agreement
plane between LES and the WT experiment. It indicates that LES accu­ between LES and the experiment is also observed. LES better captures
rately predicts urmse/Uref at the roof height (z/H = 1) and slightly the C+ than RANS in the near-ground region.
overpredicts urmse/Uref above the canyon (z/H = 1.08 and 1.17). For
wrmse/Uref, good agreements between LES results and experimental data 5. Comparison between RANS and LES results
are observed along the three lines (Fig. 11d-f).
Fig. 12 shows the non-dimensional mean concentration (C+) from Section 3 (RANS) and Section 4 (LES) indicate obvious discrepancies
the LES simulation, the best-performing RANS simulation (SKE with Sct in the simulated velocity components and pollutant concentration with
= 0.3), and the WT experiment. For the leeward wall, LES shows a good experimental measurements. This section provides a more detailed
agreement with the experimental data along the studied seven lines. LES analysis to better understand the performance of steady RANS and LES
clearly outperforms the best-performing RANS model in the middle re­ on the flow field and pollutant concentration.
gion, e.g., along y/H = -3.75, -1.25, 0, and 1.25 (see Fig. 12b-e). The Fig. 13 illustrates the spatial distributions of the non-dimensional

7
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

Fig. 6. Comparison of C+ obtained from RANS simulations with different turbulence models along seven vertical lines for the leeward wall: (a) y/H = -4.92; (b) y/H
= -3.75; (c) y/H = -1.25; (d) y/H = 0; (e) y/H = 1.25; (f) y/H = 3.75; (g) y/H = 4.92. (h-n) Same as (a-g) but for the windward wall.

Fig. 7. Same as Fig. 6 but for different turbulent Schmidt numbers (Sct).

mean wind velocity magnitude (U3D/Uref) and 2D velocity vectors in the and k). For the vertical center plane, the recirculation core is located at
horizontal plane (z/H = 0.25). It exhibits two corner eddies and LES the height of about 0.6H, which is consistent with a previous study using
results show a slight asymmetry between them, which is due to the effect LES for a street canyon without moving traffics (Merlier, Jacob and
of the moving traffics along the leeward lane. Although the locations of Sagaut, 2018). Compared to LES, SKE and RKE predict the core of the
the cores of the corner eddies by SKE and RKE generally match with canyon vortex closer to the windward wall. SKE and RKE slightly
those by LES, SKE and RKE predict weaker corner eddies with lower overpredict the absolute values of mean y-velocity (shown as V/Uref in
mean velocity magnitudes and excessive asymmetry of the flow field Fig. 14b, f, and j), while significantly underestimate the velocity
than LES. magnitude inside the canyon (shown as U3D/Uref in Fig. 14g and k). This
Fig. 14 illustrates the spatial distribution of the non-dimensional weaker predicted strength of circulation by SKE or RKE is in line with
mean y-velocity (V/Uref) and 2D velocity vectors in three vertical previous CFD studies for street canyons (Moonen et al., 2011; Salim
planes (y/H = -5, 0, and 5). The mean wind velocity magnitude (U3D/ et al., 2011a, 2011b). For the vertical plane y/H = -5, SKE and RKE
Uref) is also shown for the vertical center plane (y/H = 0, Figs. 14c, g, overestimate the negative mean y-velocity that spread from the leeward

8
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

Fig. 8. V/Uref obtained from LES, RANS (SKE and RKE), and the experiment along four horizontal lines: (a) z/H = 0.25; (b) z/H = 0.5; (c) z/H = 0.75; (d) z/H = 1 in
the vertical corner plane (y/H = -5). (e-h) Same as (a-d) but for U/Uref.

Fig. 9. vrmse/Uref obtained from LES and the experiment along four horizontal lines: (a) z/H = 0.25; (b) z/H = 0.5; (c) z/H = 0.75; (d) z/H = 1 in the vertical corner
plane (y/H = -5). (e-h) Same as (a-d) but for urmse/Uref.

lane to the leeward wall, compared with LES. While they underpredict lateral edges. For the leeward wall, LES can accurately reproduce the
the positive mean y-velocity in a large area (Fig. 14a, e, and i). For the traffic-induced concentration skewness, while SKE predicts an exces­
vertical plane y/H = 5, SKE and RKE underpredict the negative mean sively skewed C+ distribution towards the right-hand side. LES also
y-velocity in a large area (Fig. 14d, h, and l), due to the weaker corner reasonably reproduces the distribution of C+ for the windward wall
eddies reproduced by the two RANS models (see Fig 13). (Fig. 15e), while SKE overpredicts the C+ especially in the near-ground
Fig. 15 compares the C+ distribution among the LES, and the best- region (Fig. 15f). The possible reason for this overestimated concen­
performing steady RANS (SKE with Sct = 0.3) for the leeward and tration at the lower region of the canyon is the underestimated flow
windward walls. Fig. 15a and d show that the measured C+ reaches the velocity and eddy intensity by SKE (Figs. 13 and 14).
maximum in the middle of the canyon, and decays towards the canyon The essentials of concentration transport can be represented by the

9
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

Fig. 10. U/Uref obtained from LES, RANS (SKE and RKE), and the WT experiment along three horizontal lines: (a) z/H = 1; (b) z/H = 1.08; (c) z/H = 1.17 in the
vertical center plane (y/H = 0). (d-f) Same as (a-c) but for W/Uref.

Fig. 11. urmse/Uref obtained from LES and the WT experiment along three horizontal lines: (a) z/H = 1; (b) z/H = 1.08; (c) z/H = 1.17 in the vertical center plane (y/
H = 0). (d-f) Same as (a-c) but for wrmse/Uref.

distribution of concentration fluxes (Tominaga & Stathopoulos, 2011). region of the canyon is mostly more than one order of magnitude larger
As stated before, scalar transport of concentration consists of convective than the convective diffusion flux. For the majority area in the three
and turbulent diffusion effects, which are expressed by convective planes, the turbulent diffusion fluxes are larger than 1/10 of the
diffusion fluxes <ui><c> and turbulent diffusion fluxes <ui′ c′ >, convective diffusion fluxes (|<v′ c′ >/(<v><c>)|>0.1), which illus­
respectively. As shown before, moving traffics strongly affect the ve­ trates the important role of turbulent transport in determining the
locity components and pollutant removal in the spanwise direction pollution concentration in canyons.
(y-direction). Fig. 16 shows the ratio of spanwise turbulent diffusion Turbulent diffusion fluxes are calculated directly in LES, but they are
fluxes to convective diffusion fluxes in three vertical planes by LES. The modeled by the gradient diffusion hypothesis in RANS. Therefore, the
paths of dark color in Fig. 16a and 16c indicate the existence of a non-dimensional spanwise turbulent diffusion fluxes (<v′ c′ >/Q0) are
coherent structure. As a result, turbulent transport is effective in used to compare the LES and RANS results, where the reference mass
removing vehicle pollution out of the canyon at spanwise outlets. For the flux magnitude Q0 (g⋅m-2⋅s-1) follows:
y/H = 0 plane (Fig. 16b), the local turbulent diffusion flux in the center

10
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

Fig. 12. C+ obtained from LES, RANS (SKE with Sct = 0.3) and the WT experiment along seven vertical lines for the leeward wall: (a) y/H = -4.92; (b) y/H = -3.75;
(c) y/H = -1.25; (d) y/H = 0; (e) y/H = 1.25; (f) y/H = 3.75; (g) y/H = 4.92. (h-n) Same as (a-g) but for the windward wall.

Fig. 13. Distribution of U3D/Uref and 2D velocity vector field in the horizontal plane at z/H = 0.25 from top view by (a) LES, (b) RANS with SKE, and (c) RANS
with RKE.

11
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

Fig. 14. Distributions of (a) V/Uref and 2D velocity vector field in the vertical plane y/H = -5; (b) V/Uref and 2D velocity vector field in the vertical center plane (y/H
= 0); (c) U3D/Uref and 2D velocity vector field in the vertical center plane (y/H = 0); (d) V/Uref and 2D velocity vector field in vertical plane y/H = 5 by LES. (e-h)
Same as (a-d) but by SKE. (i-l) Same as (a-d) but by RKE.

Fig. 15. Distribution of C+ near the leeward wall (x/H = 0.042) from (a) the experiment (redrawn based on Ref. (Gromke & Ruck, 2009a)), (b) the LES simulation,
and (c) the steady RANS simulation (SKE with Sct = 0.3). (d-f) Same as (a-c) but for the windward wall (x/H = 0.958).

12
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

Fig. 16. Ratio of the spanwise turbulent diffusion fluxes to convective diffusion fluxes (|<v′ c′ >/(<v><c>)|) in three vertical planes (a) y/H = -5, (b) y/H = 0 and
(c) y/H = 5 by LES.

Fig. 17. Contours of the <v′ c′ >/Q0 in vertical planes (a) y/H = -5, (b) y/H = 0 and (c) y/H = 5 by LES. (d-f) Same as (a)-(c) but by RANS with SKE with Sct = 0.3.
The dashed lines represent the isolines ∂C/∂y = 0. The sign of ∂C/∂y on each side of the isoline is marked in circles (+: positive; − : negative). The counter-gradient
mechanism of turbulent diffusion flux is characterized by <v′ c′ >/Q0 and ∂C/∂y of the same sign.

Q0 = (Q / l)/H (13) the LES results indicate that the turbulence drives the pollutant out of
the canyon space through a large area of this boundary, i.e. negative
where the emission rate of tracer gas source Q = 10 g/s, and the length of <v′ c′ >/Q0. In the region close to the two lanes in this plane, LES predicts
pollutant source l = 1.42 m. <v′ c′ > by RANS can be calculated as the direction of <v′ c′ >/Q0 the same as the moving direction of traffics,
follows: while SKE predicts the opposite. For the vertical plane y/H = 5, LES
results show positive <v′ c′ >/Q0 except for the region in the windward
νt ∂C
(14) corner and around the leeward lane that <v′ c′ >/Q0 follows the moving
′ ′
< v c >= −
Sct ∂y
direction of traffics. Compared to LES, SKE substantially underpredicted
Fig. 17 compares <v′ c′ >/Q0 by the LES and the best-performing the absolute values of <v′ c′ >/Q0 in the three planes, which is in general
steady RANS (SKE with Sct = 0.3) in three vertical planes (y/H = -5, one order of magnitude smaller. This underprediction by steady RANS is
0, and 5). In addition to contours of <v′ c′ >/Q0, the isoline ∂C/∂y = 0 is in line with a previous study (van Hooff et al., 2014).
displayed, and the areas with positive or negative concentration de­ The results from LES show a negative spanwise concentration
rivatives are marked with a ‘+’ or ‘− ’ sign, respectively. For the plane y/ gradient (∂C/∂y < 0) around the leeward lane in the vertical plane y/H
H = 0, the LES results indicate that the direction of <v′ c′ >/Q0 close to = 0, however, there is also a negative turbulent flux (<v′ c′ >/Q0 < 0) in
the two lanes clearly follows the moving direction of traffics. While SKE this region. This observation indicates the so-called counter-gradient
predicts differently for the leeward lane. For the vertical plane y/H = -5, mechanism, i.e., spanwise turbulent diffusion flux is directed from a

13
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

low-concentration place to a high-concentration place; The same holds vertical center plane above the canyon, the five RANS models show
for regions at the windward side in the vertical plane y/H = 0 and near fairly good agreement with the experimental data for the mean
the ground in the vertical planes y/H = -5 and 5, as a positive ∂C/∂y vertical velocity. But only SKE and RKE models reasonably reproduce
exists in combination with a positive <v′ c′ >/Q0. These observations the mean streamwise velocity.
suggest that the gradient-diffusion hypothesis is not always valid inside • The LES simulation can accurately reproduce the mean and rmse
the street canyon, especially in the region near moving traffics. The large velocity components in the lateral boundary and vertical center
difference between the obtained turbulent fluxes indicates the impor­ plane of the canyon. Compared to LES, the best-performing RANS
tance of accurate turbulent diffusion modeling for the predicting of the models, i.e. SKE and RKE, predict the weaker strength of the corner
mean concentration in street canyon considering moving vehicles. It eddies and canyon vortex.
supports the superiority of LES in predicting the flow field, which is • Among the five investigated RANS models, SKE yields the best per­
mainly caused by the reproduction of the unsteady nature of the flow in formance for predicting mean concentration, which is supported by
street canyons and has a significant influence on the predicted mean the results of the three validation metrics. It confirms that the value
concentration field. It is worth mentioning that the computational de­ of Sct has a significant influence on the prediction of pollutant
mand for the LES simulation is approximately 50 times larger than that transfer by RANS. Large Sct shows an overpredicted traffic-induced
for steady RANS simulations, according to the HPC system employed. skewness of the mean concentration distribution. SKE with a
The computational time for most steady RANS simulations was below 7 smaller value of Sct such as 0.3 tends to give better-predicated results
h, while the LES simulation took around 2 weeks. of mean pollutant concentration.
• The LES simulation better reproduces the mean concentration dis­
6. Limitations and future work tribution than the best-performing RANS, i.e. SKE with Sct = 0.3,
especially the skewed distribution of the mean concentration
Despite the extensive evaluation analysis, this study has some limi­ induced by moving traffics. LES results indicate the turbulent
tations: 1) The focus is on a single street canyon with two-way traffics transport plays a significant role in spanwise pollutant removal, and
under the perpendicular wind, which is susceptible to high pollutant counter-gradient mechanisms are presented inside the canyon. The
concentrations and has been widely investigated in the past. The pres­ gradient-diffusion hypothesis that RANS adopted is not always valid
ence of crossroads or street canyons in sequence may lead to more in the street canyon, especially in regions close to the moving traffics,
complex airflow (Nikolova et al., 2011). Comparison between steady which strongly affects the predicted mean concentration.
RANS and LES approaches can be extended further by considering
complex urban configurations and under oblique wind directions. 2) The Declaration of Competing Interest
focus is on a sample generic street canyon with a simplified configura­
tion. In real cities, trees are usually present inside street canyons, and None.
buildings have various structural elements such as balconies. The pres­
ence of such roughness elements can induce a high level of complexity in Acknowledgments
the airflow in the street canyon (Montazeri & Blocken, 2013; Buccolieri
et al., 2018; Chatzidimitriou & Yannas, 2017; Ming et al., 2021; Qin The authors would like to express sincere thanks to Dr. Christof
et al., 2020; Zheng, Montazeri, & Blocken, 2021). Future work can Bernhard Gromke for his advice. This work was supported by the Hong
extend this study to street canyons with trees and surface roughness Kong Research Grants Council funded project 26202319. The numerical
details. 3) The numerical setups in previous RANS simulations with the simulation for this article was supported by the National Tianhe-2 center
Q-S method were not always consistent with the best configuration in Guangzhou, P.R. China.
recognized in this study (see Table 1). While the present study only
evaluates the performance of steady RANS with five turbulent models. Supplementary materials
Future work can explorer the potential of using unsteady RANS to model
the unsteady motion of the flow, as the computational demand of un­ Supplementary material associated with this article can be found, in
steady RANS is less than that of LES. 4) The present study adopted the the online version, at doi:10.1016/j.scs.2021.103307.
Q-S method, in which the effect of moving traffics is represented by
additional source terms. The DMU method can explicitly model the References
shape of moving vehicles, although it is achieved with a larger compu­
tational cost. The performance of RANS and LES approaches for simu­ Ahmad, K., Khare, M., & Chaudhry, K. K. (2005). Wind tunnel simulation studies on
lations adopting the DMU method shall be compared in the future. dispersion at urban street canyons and intersections—a review. Journal of Wind
Engineering and Industrial Aerodynamics, 93, 697–717. https://doi.org/10.1016/J.
JWEIA.2005.04.002
7. Conclusions Ai, Z. T., & Mak, C. M. (2015). Large-eddy Simulation of flow and dispersion around an
isolated building: Analysis of influencing factors. Computers & Fluids, 118, 89–100.
https://doi.org/10.1016/j.compfluid.2015.06.006
The present paper provides a CFD validation study of wind flow and Baker, C. J., & Hargreaves, D. M. (2001). Wind evaluation of a vehicle pollution
pollutant concentration in a street canyon with moving traffics. Grid- dispersion model. Journal of Wind Engineering and Industrial Aerodynamics, 89,
sensitivity studies are conducted, after which the performance of 187–200. https://doi.org/10.1016/S0167-6105(00)00061-1
Bazdidi-Tehrani, F., Masoumi-Verki, S., Gholamalipour, P., & Kiamansouri, M. (2019).
steady RANS and LES approaches in simulating velocity and concen­ Large eddy simulation of pollutant dispersion in a naturally cross-ventilated model
tration is systematically evaluated against a wind-tunnel experiment. building: Comparison between sub-grid scale models. Building Simulation, 12,
Comparisons between different RANS turbulence models, between 921–941. https://doi.org/10.1007/S12273-019-0525-5
Blocken, B. (2018). LES over RANS in building simulation for outdoor and indoor
different turbulent Schmidt numbers (Sct), and between RANS and LES applications: A foregone conclusion? Building Simulation. https://doi.org/10.1007/
provide useful guidance to support future CFD simulations, especially s12273-018-0459-3
when deciding which numerical approach to use. The main conclusions Blocken, B., Stathopoulos, T., & Carmeliet, J. (2007). CFD simulation of the atmospheric
boundary layer: wall function problems. Atmospheric Environment, 41, 238–252.
can be made as follows:
https://doi.org/10.1016/j.atmosenv.2006.08.019
Blocken, B., Stathopoulos, T., Carmeliet, J., & Hensen, J. L. M. (2011). Application of
• At the lateral boundaries of the street canyon, all the five RANS computational fluid dynamics in building performance simulation for the outdoor
models substantially underestimate the mean spanwise velocity environment: an overview. Journal of Building Performance Simulation, 4, 157–184.
https://doi.org/10.1080/19401493.2010.513740
comparing with the wind-tunnel experiment, and only SKE and RKE Blocken, B., Stathopoulos, T., Saathoff, P., & Wang, X. (2008). Numerical evaluation of
models reasonably capture the mean streamwise velocity. In the pollutant dispersion in the built environment: Comparisons between models and

14
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

experiments. Journal of Wind Engineering and Industrial Aerodynamics, 96, against laboratory and full-scale concentration measurements in street canyons.
1817–1831. https://doi.org/10.1016/j.jweia.2008.02.049 Environmental Fluid Mechanics, 3, 145–172. https://doi.org/10.1023/A:
Buccolieri, R., Gromke, C., Di Sabatino, S., & Ruck, B. (2009). Aerodynamic effects of 1022049224166
trees on pollutant concentration in street canyons. The Science of the Total Kastner-Klein, P., Fedorovich, E., & Rotach, M. W. (2001). A wind tunnel study of
Environment, 407, 5247–5256. https://doi.org/10.1016/j.scitotenv.2009.06.016 organised and turbulent air motions in urban street canyons. Journal of Wind
Buccolieri, R., Jeanjean, A. P. R., Gatto, E., & Leigh, R. J. (2018). The impact of trees on Engineering and Industrial Aerodynamics, 89, 849–861. https://doi.org/10.1016/
street ventilation, NOx and PM2.5 concentrations across heights in Marylebone Rd S0167-6105(01)00074-5
street canyon, central London. Sustain. Cities Society, 41, 227–241. https://doi.org/ Katolický, J., & Jícha, M. (2005). Eulerian-Lagrangian model for traffic dynamics and its
10.1016/J.SCS.2018.05.030 impact on operational ventilation of road tunnels. Journal of Wind Engineering and
Cai, C., Ming, T., Fang, W., de Richter, R., & Peng, C. (2020). The effect of turbulence Industrial Aerodynamics, 93, 61–77. https://doi.org/10.1016/j.jweia.2004.09.002
induced by different kinds of moving vehicles in street canyons. Sustainable Cities Kondo, H., & Tomizuka, T. (2009). A numerical experiment of roadside diffusion under
Society, 54, Article 102015. https://doi.org/10.1016/j.scs.2020.102015 traffic-produced flow and turbulence. Atmospheric Environment, 43, 4137–4147.
Cebeci, T., & Bradshaw, P. (1977). Momentum transfer in boundary layers. Washington, D. https://doi.org/10.1016/j.atmosenv.2009.05.047
C.New York: Hemisphere Publishing Corp. Kovar-Panskus, A., Louka, P., Sini, J.-F., Savory, E., Czech, M., Abdelqari, A.,
Chatzidimitriou, A., & Yannas, S. (2017). Street canyon design and improvement Mestayer, P. G., & Toy, N. (2002). Influence of geometry on the mean flow within
potential for urban open spaces; the influence of canyon aspect ratio and orientation urban street canyons–a comparison of wind tunnel experiments and numerical
on microclimate and outdoor comfort. Sustainable Cities Society, 33, 85–101. https:// simulations. Water, Air Soil Pollution Focus, 2, 365–380.
doi.org/10.1016/J.SCS.2017.05.019 Kubilay, A., Neophytou, M. K. A., Matsentides, S., Loizou, M., & Carmeliet, J. (2017). The
Chatzimichailidis, A. E., Argyropoulos, C. D., Assael, M. J., & Kakosimos, K. E. (2019). pollutant removal capacity of urban street canyons as quantified by the pollutant
Qualitative and quantitative investigation of multiple large eddy simulation aspects exchange velocity. Urban Climate, 21, 136–153. https://doi.org/10.1016/j.
for pollutant dispersion in street canyons using OpenFOAM. Atmosphere (Basel), 10. uclim.2017.06.003
https://doi.org/10.3390/atmos10010017 Launder, B. E., & Spalding, D. B. (1974). The numerical computation of turbulent flows.
DePaul, F. T., & Sheih, C. M. (1986). Measurements of wind velocities in a street canyon. Computer Methods Applied Mechnical Engineering, 3, 269–289. https://doi.org/
Atmospheric Environment, 20, 455–459. https://doi.org/10.1016/0004-6981(86) 10.1016/0045-7825(74)90029-2
90085-5 Li, X.-X., Britter, R. E., Yong Koh, T., Norford, L. K., Liu,, C.-H., Entekhabi, D., & C
Ducros, F., Nicoud, F., & Poinsot, T. (1998). Wall-adapting local eddy-viscosity models Leung, D. Y. (2010). Large-eddy simulation of flow and pollutant transport in urban
for simulations in complex geometries. In Oxford University Computing Laboratory. street canyons with ground heating. Boundary-Layer Meteorology, 137, 187–204.
16th Conference on Numerical Methods in Fluid Dynamics (pp. 293–299). France: https://doi.org/10.1007/s10546-010-9534-8
Arcachon. Li, Z., Xu, J., Ming, T., Peng, C., Huang, J., & Gong, T. (2017). Numerical simulation on
Franke, J., Hellsten, A., Schlünzen, H., & Carissimo, B. (2007). Best Practice Guideline for the effect of vehicle movement on pollutant dispersion in urban street. Procedia
the CFD Simulation of Flows in the Urban Environment. Hamburg, Germany: Engineering, 205, 2303–2310. https://doi.org/10.1016/j.proeng.2017.10.104
Meteorological Inst., COST. Liu, C.-H., & Wong, C. C. C. (2014). On the pollutant removal, dispersion, and
Gallagher, J., Gill, L. W., & McNabola, A. (2013). The passive control of air pollution entrainment over two-dimensional idealized street canyons. Atmospheric Research,
exposure in Dublin, Ireland: A combined measurement and modelling case study. 135-136, 128–142. https://doi.org/10.1016/j.atmosres.2013.08.006
Science Total Environment, 331–343. https://doi.org/10.1016/J. Liu, J., Niu, J., Du, Y., Mak, C. M., & Zhang, Y. (2019). LES for pedestrian level wind
SCITOTENV.2013.03.079, 458-460. around an idealized building array—Assessment of sensitivity to influencing
Gousseau, P., Blocken, B., Stathopoulos, T., & van Heijst, G. J. F (2011). CFD simulation parameters. Sustainable Cities Society, 44, 406–415. https://doi.org/10.1016/j.
of near-field pollutant dispersion on a high-resolution grid: A case study by LES and scs.2018.10.034
RANS for a building group in downtown Montreal. Atmospheric Environment, 45, Mazzeo, N. A., & Venegas, L. E. (2005). Evaluation of turbulence from traffic using
428–438. https://doi.org/10.1016/j.atmosenv.2010.09.065 experimental data obtained in a street canyon. International Journal of Environment
Gousseau, P., Blocken, B., & van Heijst, G. J. F (2012). Large-Eddy Simulation of and Pollution, 25, 164–176. https://doi.org/10.1504/ijep.2005.007664
pollutant dispersion around a cubical building: Analysis of the turbulent mass Menter, F. R. (1994). Two-equation eddy-viscosity turbulence models for engineering
transport mechanism by unsteady concentration and velocity statistics. applications. AIAA Journal, 32, 1598–1605. https://doi.org/10.2514/3.12149
Environmental Pollution, 167, 47–57. https://doi.org/10.1016/j.envpol.2012.03.021 Merlier, L., Jacob, J., & Sagaut, P. (2018). Lattice-boltzmann large-eddy simulation of
Gromke, C. (2008). Database, 2008. Concentration data of street canyon. Internet pollutant dispersion in street canyons including tree planting effects. Atmospheric
database. [WWW Document]. URL http://www.codasc.de. Environment, 195, 89–103. https://doi.org/10.1016/j.atmosenv.2018.09.040
Gromke, C., Buccolieri, R., Di Sabatino, S., & Ruck, B. (2008). Dispersion study in a street Miao, C., Yu, S., Hu, Y., Bu, R., Qi, L., He, X., & Chen, W. (2020). How the morphology of
canyon with tree planting by means of wind tunnel and numerical investigations – urban street canyons affects suspended particulate matter concentration at the
Evaluation of CFD data with experimental data. Atmospheric Environment, 42, pedestrian level: An in-situ investigation. Sustainable Cities Society, 55, Article
8640–8650. https://doi.org/10.1016/j.atmosenv.2008.08.019 102042. https://doi.org/10.1016/J.SCS.2020.102042
Gromke, C., & Ruck, B. (2009a). Effects of trees on the dilution of vehicle exhaust Ming, T., Han, H., Zhao, Z., de Richter, R., Wu, Y., Li, W., & Wong, N. H. (2021). Field
emissions in urban street canyons. International Journal of Environment Waste synergy analysis of pollutant dispersion in street canyons and its optimization by
Management, 4, 225–242. https://doi.org/10.1504/IJEWM.2009.026894 adding wind catchers. Building Simulation, 142(14), 391–405. https://doi.org/
Gromke, C., & Ruck, B. (2009b). On the impact of trees on dispersion processes of traffic 10.1007/S12273-020-0720-4, 2020.
emissions in street canyons. Boundary-Layer Meteorology, 131, 19–34. https://doi. Montazeri, H., & Blocken, B. (2013). CFD simulation of wind-induced pressure
org/10.1007/s10546-008-9301-2 coefficients on buildings with and without balconies: Validation and sensitivity
Gromke, C., & Ruck, B. (2007). Influence of trees on the dispersion of pollutants in an analysis. Building and Environment, 60, 137–149. https://doi.org/10.1016/j.
urban street canyon—Experimental investigation of the flow and concentration field. buildenv.2012.11.012
Atmospheric Environment, 41, 3287–3302. https://doi.org/10.1016/j. Moonen, P., Dorer, V., & Carmeliet, J. (2011). Evaluation of the ventilation potential of
atmosenv.2006.12.043 courtyards and urban street canyons using RANS and LES. Journal of Wind
Han, B.-S., Baik, J.-J., Kwak, K.-H., & Park, S.-B. (2018). Large-eddy simulation of Engineering and Industrial Aerodynamic, 99, 414–423. https://doi.org/10.1016/j.
reactive pollutant exchange at the top of a street canyon. Atmospheric Environment, jweia.2010.12.012
187, 381–389. https://doi.org/10.1016/j.atmosenv.2018.06.012 Murakami, S., Mochida, A., & Hibi, K. (1987). Three-dimensional numerical simulation
Hang, J., Buccolieri, R., Yang, X., Yang, H., Quarta, F., & Wang, B. (2019). Impact of of air flow around a cubic model by means of large eddy simulation. Journal of Wind
indoor-outdoor temperature differences on dispersion of gaseous pollutant and Engineering and Industrial Aerodynamics, 25, 291–305. https://doi.org/10.1016/
particles in idealized street canyons with and without viaduct settings. Building 0167-6105(87)90023-7
Simulation, 12, 285–297. https://doi.org/10.1007/S12273-018-0476-2 Niachou, K., Livada, I., & Santamouris, M. (2008). Experimental study of temperature
Iousef, S., Montazeri, H., Blocken, B., & van Wesemael, P. J. V. (2017). On the use of non- and airflow distribution inside an urban street canyon during hot summer weather
conformal grids for economic LES of wind flow and convective heat transfer for a conditions-Part I: Air and surface temperatures. Building and Environment, 43,
wall-mounted cube. Building and Environment, 119, 44–61. https://doi.org/10.1016/ 1383–1392. https://doi.org/10.1016/j.buildenv.2007.01.039
j.buildenv.2017.04.004 Nikolova, I., Janssen, S., Vos, P., Vrancken, K., Mishra, V., & Berghmans, P. (2011).
Jicha, M., Katolicky, J., & Pospisil, J. (2002). Dispersion of pollutants in a street canyon Dispersion modelling of traffic induced ultrafine particles in a street canyon in
and street intersection under traffic-induced flow and turbulence using a iow Re k-ε Antwerp, Belgium and comparison with observations. The Science of the Total
model. International Journal of Environment and Pollution, 18, 160–170. https://doi. Environment, 412-413, 336–343. https://doi.org/10.1016/j.scitotenv.2011.09.081
org/10.1504/IJEP.2002.012126 Oke, T. R. (1988). Street design and urban canopy layer climate. Energy Building, 11,
Jicha, M., Pospisil, J., & Katolicky, J. (2000). Dispersion of pollutants in street canyon 103–113. https://doi.org/10.1016/0378-7788(88)90026-6
under traffic induced flow and turbulence. Environmental Monitoring and Assessment, Ong, R. H., Patruno, L., Yeo, D., He, Y., & Kwok, K. C. S. (2020). Numerical simulation of
65, 343–351. https://doi.org/10.1023/A:1006452422885 wind-induced mean and peak pressures around a low-rise structure. Engineering
Jin, X., Yang, L., Du, X., & Yang, Y. (2017). Transport characteristics of PM2.5 inside Structures, 214, Article 110583. https://doi.org/10.1016/J.
urban street canyons: The effects of trees and vehicles. Building Simulation, 10, ENGSTRUCT.2020.110583
337–350. https://doi.org/10.1007/S12273-016-0324-1 Pospisil, J., Huzlik, J., Licbinsky, R., & Spilacek, M. (2020). Dispersion Characteristics of
Kastner-Klein, P., Berkowicz, R., & Britter, R. (2004). The influence of street architecture PM10 particles identified by numerical simulation in the vicinity of roads passing
on flow and dispersion in street canyons. Meteorology and Atmospheric Physics, 87, through various types of urban areas. Atmosphere 2020, 11. https://doi.org/
121–131. https://doi.org/10.1007/s00703-003-0065-4 10.3390/ATMOS11050454. Page 454 11, 454.
Kastner-Klein, P., Fedorovich, E., Ketzel, M., Berkowicz, R., & Britter, R. (2003). The Pospisil, J., & Jicha, M. (2019). Numerical modelling of transient dispersion of air
modelling of turbulence from traffic in urban dispersion models - Part II: Evaluation pollution in perpendicular urban street intersection with detail inclusion of traffic

15
X. Zheng and J. Yang Sustainable Cities and Society 75 (2021) 103307

dynamics. International Journal of Environment and Pollution (pp. 71–83). Tominaga, Y., & Stathopoulos, T. (2013). CFD simulation of near-field pollutant
Inderscience Publishers. https://doi.org/10.1504/IJEP.2019.101834 dispersion in the urban environment: A review of current modeling techniques.
Pospisil, J., & Jicha, M. (2017). Influence of vehicle-induced turbulence on pollutant Atmospheric Environment, 79, 716–730. https://doi.org/10.1016/j.
dispersion in street canyon and adjacent urban area. International Journal of atmosenv.2013.07.028
Environment and Pollution (pp. 89–101). Inderscience Publishers. https://doi.org/ Tominaga, Y., & Stathopoulos, T. (2012). CFD modeling of pollution dispersion in
10.1504/IJEP.2017.089400 building array: Evaluation of turbulent scalar flux modeling in RANS model using
Qin, H., Hong, B., Huang, B., Cui, X., & Zhang, T. (2020). How dynamic growth of avenue LES results. Journal of Wind Engineering and Industrial Aerodynamics, 484–491.
trees affects particulate matter dispersion: CFD simulations in street canyons. https://doi.org/10.1016/j.jweia.2012.02.004, 104-106.
Sustainable Cities and Society, 61, Article 102331. https://doi.org/10.1016/J. Tominaga, Y., & Stathopoulos, T. (2011). CFD modeling of pollution dispersion in a street
SCS.2020.102331 canyon: Comparison between LES and RANS. Journal of Wind Engineering and
Qin, Y., & Kot, S. C. (1993). Dispersion of vehicular emission in street canyons, Industrial Aerodynamics, 99, 340–348. https://doi.org/10.1016/j.jweia.2010.12.005
Guangzhou City, South China (P.R.C.). Atmospheric Environment Part B. Urban Atmos, Tominaga, Y., & Stathopoulos, T. (2007). Turbulent Schmidt numbers for CFD analysis
27, 283–291. https://doi.org/10.1016/0957-1272(93)90023-Y with various types of flowfield. Atmospheric Environment, 41, 8091–8099. https://
Ramponi, R., & Blocken, B. (2012). CFD simulation of cross-ventilation for a generic doi.org/10.1016/j.atmosenv.2007.06.054
isolated building: Impact of computational parameters. Building and Environment, 53, Uehara, K., Murakami, S., Oikawa, S., & Wakamatsu, S. (2000). Wind tunnel experiments
34–48. https://doi.org/10.1016/j.buildenv.2012.01.004 on how thermal stratification affects flow in and above urban street canyons.
Richards, P. J., & Hoxey, R. P. (1993). Appropriate boundary conditions for Atmospheric Environment, 34, 1553–1562. https://doi.org/10.1016/S1352-2310(99)
computational wind engineering models using the k-ε turbulence model. Computer 00410-0
Wind Engineering, 145–153. https://doi.org/10.1016/B978-0-444-81688-7.50018-8, van Druenen, T., van Hooff, T., Montazeri, H., & Blocken, B. (2019). CFD evaluation of
1. building geometry modifications to reduce pedestrian-level wind speed. Building and
Rodi, W. (1997). Comparison of LES and RANS calculations of the flow around bluff Environment, 163, Article 106293. https://doi.org/10.1016/j.buildenv.2019.106293
bodies. Journal of Wind Engineering and Industrial Aerodynamics, 69-71, 55–75. van Hooff, T., Blocken, B., Gousseau, P., & van Heijst, G. J. F (2014). Counter-gradient
https://doi.org/10.1016/S0167-6105(97)00147-5 diffusion in a slot-ventilated enclosure assessed by LES and RANS. Computers &
Salim, S. M., Buccolieri, R., Chan, A., & Sabatino, S. Di (2011a). Numerical simulation of Fluids, 96, 63–75. https://doi.org/10.1016/j.compfluid.2014.02.020
atmospheric pollutant dispersion in an urban street canyon: Comparison between Wang, Q., Fang, W., de Richter, R., Peng, C., & Ming, T. (2019). Effect of moving vehicles
RANS and LES. Journal Wind Engineering and Industrial Aerodynamic, 99, 103–113. on pollutant dispersion in street canyon by using dynamic mesh updating method.
https://doi.org/10.1016/j.jweia.2010.12.002 Journal of Wind Engineering and Industrial Aerodynamics, 187, 15–25. https://doi.org/
Salim, S. M., Cheah, S. C., & Chan, A. (2011b). Numerical simulation of dispersion in 10.1016/j.jweia.2019.01.014
urban street canyons with avenue-like tree plantings: Comparison between RANS Werner, H., & Wengle, H. (1993). Large-eddy simulation of turbulent flow over and
and LES. Building and Environment, 46, 1735–1746. https://doi.org/10.1016/j. around a cube in a plate channel. Turbulent Shear Flows 8 (pp. 155–168). Berlin,
buildenv.2011.01.032032 Heidelberg: Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-642-77674-
Schatzmann, M., Olesen, H., & Franke, J. (2010). COST 732 model evaluation case 8_12
studies: approach and results. Cost Action. Wilcox, D. C. (1998). Turbulence modeling for CFD. La Canada, CA: DCW industries.
Scungio, M., Stabile, L., Rizza, V., Pacitto, A., Russi, A., & Buonanno, G. (2018). Lung Yakhot, V., Orszag, S. A., Thangam, S., Gatski, T. B., & Speziale, C. G. (1992).
cancer risk assessment due to traffic-generated particles exposure in urban street Development of turbulence models for shear flows by a double expansion technique.
canyons: A numerical modelling approach. The Science of the Total Environment, 631- Physics of Fluids A: Fluid Dynamics, 4, 1510–1520. https://doi.org/10.1063/1.858424
632, 1109–1116. https://doi.org/10.1016/j.scitotenv.2018.03.093 You, W., & Ding, W. (2021). Effects of urban square entry layouts on spatial ventilation
Sergent, E. (2002). Vers une méthodologie de couplage entre la simulation des grandes échelles under different surrounding building conditions. Building Simulation, (14), 377–390.
et les modèles statistiques. . Ecole centrale de Lyon. Ph.D. thesis. https://doi.org/10.1007/S12273-020-0656-8, 2020 142.
Shi, S., Chen, C., & Zhao, B. (2015). Air infiltration rate distributions of residences in Zhang, Y., Gu, Z., & Yu, C. W. (2017). Large eddy simulation of vehicle induced
Beijing. Building and Environment, 92, 528–537. https://doi.org/10.1016/J. turbulence in an urban street canyon with a new dynamically vehicle-tracking
BUILDENV.2015.05.027 scheme. Aerosol Air Quality Research, 17, 865–874. https://doi.org/10.4209/
Shi, S., & Zhao, B. (2016). Occupants’ interactions with windows in 8 residential aaqr.2016.05.0204
apartments in Beijing and Nanjing. China Building Simulation, 9, 221–231. Zhang, Y.-W., Gu, Z.-L., Cheng, Y., & Lee, S.-C. (2011). Effect of real-time boundary wind
Shi, T., Ming, T., Wu, Y., Peng, C., Fang, Y., & de_Richter, R. (2020). The effect of exhaust conditions on the air flow and pollutant dispersion in an urban street canyon—Large
emissions from a group of moving vehicles on pollutant dispersion in the street eddy simulations. Atmospheric Environment, 45, 3352–3359. https://doi.org/
canyons. Building and Environment, 181, Article 107120. https://doi.org/10.1016/j. 10.1016/j.atmosenv.2011.03.055
buildenv.2020.107120 Zhao, Y., Jiang, C., & Song, X. (2021). Numerical evaluation of turbulence induced by
Shih, T., Liou, W., Shabbir, A., Yang, Z., & Zhu, J. (1995). A new k-ϵ eddy viscosity model wind and traffic, and its impact on pollutant dispersion in street canyons. Sustainable
for high reynolds number turbulent flows. Computers & Fluids, 24, 227–238. https:// Cities Society, 103142. https://doi.org/10.1016/J.SCS.2021.103142
doi.org/10.1016/0045-7930(94)00032-T Zheng, X., Montazeri, H., & Blocken, B. (2020). CFD simulations of wind flow and mean
Shirzadi, M., Mirzaei, P. A., & Tominaga, Y. (2020). CFD analysis of cross-ventilation surface pressure for buildings with balconies: Comparison of RANS and LES. Building
flow in a group of generic buildings: Comparison between steady RANS, LES and and Environment, 173, Article 106747. https://doi.org/10.1016/j.
wind tunnel experiments. Building Simulation, 13, 1353–1372. https://doi.org/ buildenv.2020.106747
10.1007/S12273-020-0657-7, 2020 136 13. Zheng, X., Montazeri, M., & Blocken, B. (2021). Large-eddy simulation of pollutant
Solazzo, E., Cai, X., & Vardoulakis, S. (2008). Modelling wind flow and vehicle-induced dispersion in generic urban street canyons: guidelines for domain size. Journal of
turbulence in urban streets. Atmospheric Environment, 42, 4918–4931. https://doi. Wind Engineering and Industrial Aerodynamics, 211, Article 104527.
org/10.1016/j.atmosenv.2008.02.032 Zheng, X., Montazeri, H., & Blocken, B. (2021). CFD analysis of the impact of geometrical
Solazzo, E., Vardoulakis, S., & Cai, X. (2011). A novel methodology for interpreting air characteristics of building balconies on near-façade wind flow and surface pressure.
quality measurements from urban streets using CFD modelling. Atmospheric Building and Environment, 200, Article 107904. https://doi.org/10.1016/j.
Environment, 45, 5230–5239. https://doi.org/10.1016/J.ATMOSENV.2011.05.022 buildenv.2021.107904
Solazzo, E., Vardoulakis, S., & Cai, X. (2007). Evaluation of traffic-producing turbulence Zhou, X., Ying, A., Cong, B., Kikumoto, H., Ooka, R., Kang, L., & Hu, H. (2021). Large
schemes within operational street pollution models using roadside measurements. eddy simulation of the effect of unstable thermal stratification on airflow and
Atmospheric Environment, 41, 5357–5370. https://doi.org/10.1016/j. pollutant dispersion around a rectangular building. Journal of Wind Engineering and
atmosenv.2007.02.017 Industrial Aerodynamics, 211, Article 104526. https://doi.org/10.1016/j.
jweia.2021.104526

16

You might also like