Atmosphere 10 00683

atmosphere
Article
Numerical Simulation of Turbulent Flow and
Pollutant Dispersion in Urban Street Canyons
Van Thinh Nguyen 1, *, Thanh Chuyen Nguyen 1 and John Nguyen 2
1 Department of Civil and Environmental Engineering, Seoul National University, Seoul 151-744, Korea;
thanh@snu.ac.kr
2 John H. Daniels Faculty of Architecture, Landscape and Design, University of Toronto, Toronto, ON M5S 2J5,
Canada; johnnie.nguyen@mail.utoronto.ca
* Correspondence: vnguyen@snu.ac.kr

Received: 1 October 2019; Accepted: 31 October 2019; Published: 7 November 2019
Abstract: In this study, we have developed a numerical model based on an open source Computational
Fluid Dynamics (CFD) package OpenFOAM, in order to investigate the flow pattern and pollutant
dispersion in urban street canyons with different geometry configurations. In the new model, the
pollutant transport driven by airflow is modeled by the scalar transport equation coupling with the
momentum equations for airflow, which are deduced from the Reynolds Averaged Navier-Stokes
(RANS) equations. The turbulent flow calculation has been calibrated by various two-equation
turbulence closure models to select a practical and efficient turbulence model to reasonably capture
the flow pattern. Particularly, an appropriate value of the turbulent Schmidt number has been
selected for the pollutant dispersion in urban street canyons, based upon previous studies and careful
calibrations against experimental measurements. Eventually, the numerical model has been validated
against different well-known laboratory experiments in regard to various aspect ratios (a relationship
between the building height and the width of the street canyon), and different building roof shapes
(flat, shed, gable and round). The comparisons between the numerical simulations and experimental
measurements show a good agreement on the flow pattern and pollutant distribution. This indicates
the ability of the new numerical model, which can be applied to investigate the wind flow and
pollutant dispersion in urban street canyons.
Keywords: street canyon; roof shape; aspect ratio; numerical simulation; turbulent flow;
pollution transport
1. Introduction
Pollution from industrial activities, vehicle exhaust, heating and cooling systems, etc. can cause
fatal harms to humans in urban street canyons; therefore the investigation of flow characteristics and
pollution transports in urban street canyons is a vital task in the urban environment. The most important
characteristics of the flow in street canyons are the wind-induced flow patterns characterized by
internal flow, flow separation and reattachment, which effect on the local air quality and consequently
human health in urban areas. The study on wind flow and pollutant transport inside and over urban
street canyons has attracted great concern during the last three decades due to speedy urbanization
and city enlargement. Field measurements and laboratory-scale physical modeling are not only very
expensive, but also difficult, and somehow impossible due to the temporal and spatial scales and the
complex geometry configurations of urban street canyons. Advantaged from an increase in computer
technology (HPC facility), Computational Fluid Dynamics (CFD) becomes the most efficient tool for
the simulation of wind flows and pollutant transports in urban street canyons.
Atmosphere 2019, 10, 683; doi:10.3390/atmos10110683 www.mdpi.com/journal/atmosphere

Atmosphere 2019, 10, 683 2 of 30
In general, the three approaches applied in CFD are: Direct Numerical Simulation (DNS), Large
Eddy Simulation (LES) and Reynolds Averaged Navier-Stokes (RANS), which are all used to calculate
turbulent flows. The DNS is employed to numerically solve directly the Navier–Stokes equation in
order to calculate more accurately the mean flow and all turbulent velocity fluctuations for entire
ranges of spatial and temporal scales. Hence, the spatial grid sizes have to be sufficiently fine to capture
the smallest scales within the Kolmogorov microscales ( Reη ∼ 1). Correspondingly, the temporal
steps have to be sufficiently small to resolve the period of the fastest fluctuations. Consequently, these
calculations require a very strong capability of computer resources, and may exceed the available
capacity of most powerful high performance computers in solving the three dimensional problems of
wind flow and pollutant transport in urban street canyons with large Reynolds numbers. On the other
hand, LES uses a spatial filter to screen out the eddy scales, whereby the large scales are resolved by the
DNS method, and the small scales are resolved by a sub-grid-scale (SGS) model. Proper solving by LES
also requires very fine grids (near-wall grid sizes y+ ≤ 1), and this requirement is again tackled with a
high computing cost. Particularly in wind engineering, we usually need to simulate the flows in very
large domains in kilometers and complex geometry configurations, including a number of building
shapes, streets, trees, etc. Even, the results obtained from LES implemented by Liu et al. [1] for some
reasons did not show a good agreement with the experiments in comparison with the standard k-ε
model (see Figures 3 and 4 in Li et al. [2]). Moreover, it would not be an easy task to calibrate and
validate the turbulence characteristics obtained from the DNS or LES models with the data observed
from real urban street canyons, since such data are not usually available.
Therefore, the numerical simulation based on the Reynold Averaged Navier-Stokes (RANS)
equations is still applied to calculate the turbulent flow and pollutant dispersion in urban street
canyons, due to its practical and efficient applications. Many authors have applied RANS equations
with the standard k-ε turbulence closure model and its variants (Extended, Re-Normalization Group
(RNG), realizable) because of their robustness and efficiency. Sini et al. [3], Johnson and Hunter [4], Baik
and Kim [5,6], Chan et al. [7,8], Jeong and Andrews [9], Takano and Moonen [10], Yassin [11], etc., have
applied the standard k-ε turbulence. In addition, in order to calculate turbulent flow and pollutant
transport in urban street canyons, most of authors used a commercial CFD software, such as Fluent (Leitl
and Moroney [12]; Chan et al. [8]; Sagrado et al. [13]; Li et al. [14], Yassin [11]; etc.), CFX (Raw et al. [15];
Walton et al. [16]; etc.), PHONENICS (Hassan and Crowther [17]; Koutsourakis et al. [18]), etc. Recently,
Takano and Moonen [10] used the OpenFOAM package to study the influence of roof shapes on flow
and pollutant dispersion in an urban street canyon; however, they still applied the convection-diffusion
equation for passive scalar transport for the flow over regular arrangements of buildings with
slanted roofs.
In this study, based on an open source CFD package OpenFOAM (http://www.openfoam.com/) we
modified the source code to develop a new solver in order to investigate the flow patterns and pollutant
dispersions in urban street canyons. The OpenFOAM package is a general CFD tool box written in C++
and designed as a numerical library of solvers for Partial Differential Equations, which can provide
professional users an opportunity to build their own specific solvers, then immerse them into the
package. Based on this advantage, we developed a new solver combing the wind flow calculation with
a transport process together to facilitate the pollution transport simulation driven by the turbulent
flows. In the standard library of OpenFOAM, the numerical solution is designated only for a passive
scalar transport; i.e., the concentration field is solved for a given stationary velocity field, and it can deal
with only the constant diffusion coefficient, so that it cannot take into account the turbulent diffusion
caused by turbulent flows. In the new solver, the scalar transport equation is solved together with the
RANS equations with two-equation turbulence closure models, such as the standard k-ε turbulence
closure model and its variants (RNG k-ε, realizable k-ε), and the k-ω turbulence closure model and its
variant (k-ω SST). At each time step, after updating the wind flow field and turbulent parameters, the
advection-diffusion equation can be solved. A difference between the original and customized solvers
is shown in Figure 1 below.
Atmosphere 2019, 10, 683 3 of 30
Atmosphere 2019, 10, x FOR PEER REVIEW 3 of 30
Figure 1. A comparison between the original and customized solvers.

Figure 1. A comparison between the original and customized solvers.
2. Governing Equations
2. Governing Equations
2.1. Flow Equations
2.1. Flow Equations
For wind flow calculation, the RANS equations are modeled by various two-equation turbulence
closureFormodels,
wind flow suchcalculation,
as variousthe k-εRANS
models equations
(standard, are RNG,
modeled by various
realizable), and two-equation
k-ω modelsturbulence
(k-ω, k-ω
closure
SST). models, such
However, as various k-ε
the simulation models (standard,
of turbulent flow andRNG, realizable),
pollutant and k-𝜔
dispersion in anmodels
urban(k-𝜔,streetk-𝜔 SST).
canyon
However, the simulation of turbulent flow and pollutant dispersion
has to deal with a wide range of spatial and temporal scales and complex geometry configurations, in an urban street canyon has to deal
withso
and a wide
a robustrange andofefficient
spatial and temporal
turbulence scalesneeds
model and complex geometry
to be selected. configurations,
Therefore, we triedand so a robust
to validate the
and efficient turbulence model needs to be selected. Therefore, we
results obtained from different turbulence models against the different experimental measurements tried to validate the results obtained
from
of different
Li et al. [2] andturbulence
Brown models against
et al. [19]. Figurethe2different
shows aexperimental
validation ofmeasurements
dimensionlessofvertical Li et al.velocity
[2] and
Brown etobtained
profiles al. [19]. Figure 2 showsturbulence
from different a validationclosureof dimensionless
models against vertical
thevelocity profiles obtained
measurements of Li et al.from
[2].
different turbulence closure models against the measurements of
Figure 3 shows the comparisons of velocity profiles and turbulence kinetic energy between numerical Li et al. [2]. Figure 3 shows the
comparisons
results obtained of velocity profiles turbulence
from different and turbulence models kinetic energy between
(standard, realizable,numerical
RNG, k-ω results
and obtained
k-ω SST)from and
different
the turbulenceofmodels
measurements Brown(standard,
et al. [19] realizable,
for flat roofs.RNG, Fromk-𝜔Figures
and k-𝜔 SST)3,and
2 and the measurements
it shows that the results of
Brown et al. [19] for flat roofs. From Figures 2 and 3, it shows that the
obtained from various k-ε and k-ω models do not have noticeable differences in comparison with the results obtained from various k-ε
and k-𝜔 models
experimental do However,
data. not have noticeable
the standard differences
and RNG in comparison
k-ε modelswith are in the experimental
favor of practical data. However,
applications
the standard and RNG k-ε models are in favor of practical applications
for large scale simulations due to their robustness and efficiency. Furthermore, it is well-known that for large scale simulations duetheto
their robustness
standard k-ε model and under-predicts
efficiency. Furthermore,
reattachmentit is well-known
length (Demirdzicthat the standard
[20]; Thangamk-ε model andunder-predicts
Speziale [21]),
reattachment
and the velocity length
close to (Demirdzic
wall boundaries[20]; Thangam
within theand cavity Speziale
(Sahm et [21]), and Itthe
al. [22]). alsovelocity
has some close to wall
deficiencies
boundaries within the cavity (Sahm et al. [22]). It also has some
when the standard k-ε model is applied to simulate flow impingement and separation (Apsley deficiencies when the standard k-ε model
and
is applied
Castro to simulate
[23]). The flowflow impingement
in urban and separation
street canyons, (Apsleyisand
by its nature, Castrothe
inherent [23]). The flow in urban
characteristics of thestreet
flow
canyons,
over by its nature, is inherent
the backward-facing step, and the
thecharacteristics
flows including of thetheflow over the backward-facing
reattachment, step, and the
separation and recirculation
flows including the reattachment, separation and recirculation
zone. A modification of the strain-dependent correction term in the ε-equation of RNG k-ε model zone. A modification of the strain-
has
dependentthe
overcome correction
limitation term of in
thethe ε-equation
standard of RNG k-ε Yakhot
k-ε turbulence. model has et al.overcome
[24] report thethat
limitation
the RNG of k-ε
the
standard k-ε turbulence. Yakhot et al. [24] report that the RNG k-ε model
model shows very good predictions of the flow over the backward-facing step. Particularly, Rotach [25], shows very good predictions of
the flow over the backward-facing step. Particularly, Rotach [25], Chan
Chan et al. [7], Li et al. [14], Memon and Leung [26], and Koutsourakis et al. [18] recommend that theet al. [7], Li et al. [14], Memon and
Leungk-ε
RNG [26], and Koutsourakis
turbulence closure model et al. [18] recommend
has the best overall thatperformance
the RNG k-εamong turbulence closurevariants
k-ε model model has the
for the
best overall performance among k-ε model variants
simulation of turbulent flow and pollutant dispersion in urban street canyons.for the simulation of turbulent flow and pollutant
dispersion in urban
Therefore, street k-ε
the RNG canyons.
model (Yakhot et al. [24]) was selected for this numerical model, and its
equations read as follows:
Therefore, the RNG k-ε model (Yakhot et∂(al. Ui[24])
) was selected for this numerical model, and its
equations read as follows: =0 (1)
∂xi
∂ ( Ui ) ∂ ( Ui ) 1 ∂P 𝜕(𝑈∂ ) = µe f f ∂Ui ∂U j
" ! #
0 2 (1)
+ Uj =− +𝜕𝑥 + − kδij + gi (2)
∂t ∂x j ρ ∂xi ∂xi ρ ∂x j ∂xi 3
𝜕(𝑈 ) ∂k ) ∂k1 𝜕𝑃 ∂ µ𝜕e f f 𝜇 ∂k 𝜕𝑈1 𝜕𝑈

𝜕(𝑈 2
" #
+ 𝑈 +U =j − = + + + τij ·Sij −−ε 𝑘𝛿 + 𝑔 (3)
(2)
𝜕𝑡 ∂t
𝜕𝑥 ∂x j𝜌 𝜕𝑥∂x j 𝜕𝑥ρ σk ∂x𝜌 j 𝜕𝑥 ρ 𝜕𝑥 3
∂ε ∂ε ∂ µe f f 𝜇∂ε ε ε2
" #
1
+ 𝜕𝑘
Uj =𝜕𝑘 𝜕 𝜕𝑘 C 1 τij ·Sij − C∗
+ ε1 ε2 (4)
∂t +
∂x j𝑈 ∂x= j ρ σω ∂x j ρ + k𝜏 ∙ 𝑆 − 𝜀 k (3)
𝜕𝑡 𝜕𝑥 𝜕𝑥 𝜌 𝜎 𝜕𝑥 𝜌
𝜕𝜀 𝜕𝜀 𝜕 𝜇 𝜕𝜀 1 𝜀 𝜀
+ 𝑈 = + 𝐶 𝜏 ∙ 𝑆 − 𝐶∗ (4)
𝜕𝑡 𝜕𝑥 𝜕𝑥 𝜌 𝜎 𝜕𝑥 𝜌 𝑘 𝑘
Atmosphere 2019,10,
Atmosphere2019, 10,683
x FOR PEER REVIEW 4 of
of30
30
where: 𝜏 = −𝜌𝑢 𝑢 = 2𝜇 𝑆 − 𝜌𝑘𝛿 ; 𝑆 = +

1 ∂Ui ∂U j𝜇
= 𝜇 + 𝜇 ; 𝜇 = 𝜌𝐶 𝐶2∗ = 𝐶 +
where: τij = −ρu0i u0j = 2µt Sij − 23 ρkδij ; Sij = 2 ∂x j + ∂xi
µe f f = µ + µt ; µt = ρCµ kε ; C∗ε2 =
( )
Cµ η3 ;1−𝜂η = ; 𝑆 = 2𝑆 p𝑆 and the constants are in the following table (Table 1).
η

Cε2 + 1+βη3
0
; η = Sk
ε ;S= 2Sij Sij and the constants are in the following table (Table 1).
Table 1. Constants used in the Re-Normalization Group (RNG) k-ε model.
Table 1. Constants used in the Re-Normalization Group (RNG) k-ε model.
𝑪𝝁 𝝈𝒌 𝝈𝜺 𝑪𝜺𝟏 𝑪𝜺𝟐 𝜼𝟎 𝜷
Cµ 0.0845σk 0.7194σε 0.7194 Cε11.42 Cε2
1.68 4.377 0.012 β
η0
0.0845 0.7194 0.7194 1.42 1.68 4.377 0.012
(a) (b)
(c) (d)
Figure2.2.A A
Figure comparison
comparison of dimensionless vertical
of dimensionless velocity
vertical profilesprofiles
velocity (U/Uref (a,c)
(U/Uand W/Uref
ref (a,c) and(b,d))
W/U between
ref (b,d))
numerical results obtained

between numerical results from different
obtained from turbulence models and
different turbulence experimental
models data (Li et
and experimental al. (Li
data [2]).et(U
al.
and
[2]).W
(Uare mean
and stream-wise
W are and vertical
mean stream-wise andvelocity
verticalcomponents, respectively;
velocity components, Uref is freeUstream
respectively; velocity
ref is free stream
atvelocity
the inlet)
at the inlet)
Atmosphere 2019, 10, 683 5 of 30
(a) (b)
(c) (d)
(e) (f)
Figure
Figure 3. 3.AAcomparison
comparison of ofvertical
verticalvelocity profiles
velocity profiles ref) (a,c,e)
(U/U(U/U and turbulence
ref ) (a,c,e) kinetic energy
and turbulence (k/U
kinetic )
energy
2
(b,d,f) between numerical resultsresults
obtained from different turbulence modelsmodels
and experimental data
(k/U ref ) (b,d,f) between numerical obtained from different turbulence and experimental
(Brown
data (Brown et al., [19]).
et al., [19]).
2.2.2.2.
Scalar Transport
Scalar Equation
Transport Equation
AsAswith thethe
with same manner
same toto
manner obtain
obtainRANS
RANSequations,
equations,the
thetime-averaged
time-averaged transport
transport equation for a
equation for
scalar C is obtained:
a scalar C is obtained:
∂C ∂C ∂ ∂C ∂ 0 0
!
+ Ui = D − u c +S (5)
∂t ∂xi ∂xi ∂xi ∂xi i
Atmosphere 2019, 10, 683 6 of 30
where the prime denotes a fluctuating value, D is molecular diffusion coefficient and S is a pollutant
source. The simplest model for turbulent scalar fluxes follows from the standard gradient-diffusion
hypothesis (SGHD), where the turbulent scalar flux is assumed proportional to mean scalar gradient
as follows:
∂C
u0i c0 = −Dt
𝜕𝐶 𝜕𝐶 𝜕 𝜕𝐶∂xi 𝜕
+ 𝑈 = 𝐷 − (𝑢 𝑐 ) + 𝑆 (5)
So we can rewrite the𝜕𝑡 Equation𝜕𝑥
(5) as 𝜕𝑥 𝜕𝑥 𝜕𝑥
where the prime denotes a fluctuating∂C value,
∂C D ∂is molecular ∂C diffusion
! coefficient and S is a pollutant
source. The simplest model for ∂t + U
turbulent = ( D + D
i scalar fluxes follows t ) + S the standard gradient-diffusion
from (6)
∂xi ∂xi ∂xi
hypothesis (SGHD), where the turbulent scalar flux is assumed proportional to mean scalar gradient
where Dt is turbulent diffusion coefficient which is assumed to be isotropic and homogeneous.
as follows:
The turbulent diffusion coefficient Dt in Equation (6) is modeled by the relationship eddy-viscosity
2 𝜕𝐶
νt
(νt = Cµ kε ) and the turbulent Schmidt number 𝑢 𝑐′ Sc
=t , −𝐷
Dt = Sc , or in other words the Schmidt number is
𝜕𝑥 t
dependent upon the ratio of turbulent eddy viscosity and the turbulent mass diffusivity (Sct = Dνtt );
thus, So
it iswe
driven by the turbulent
can rewrite the Equationflow(5)andas has no universal value. Therefore, it needs to be calibrated
to select an appropriate value for the pollutant transport in urban street canyons as mentioned by
𝜕𝐶 𝜕𝐶 𝜕 𝜕𝐶
Tominaga and Stathopoulos [27]. + 𝑈
Spalding = confirmed
[28] (𝐷 +that 𝐷 Sc ) t = 0.7+gave
𝑆 close agreement with (6)the
𝜕𝑡 𝜕𝑥 𝜕𝑥 𝜕𝑥
experimental data, while Launder [29] found that Sct = 0.9 is a better value. So that, the value Sct of 0.7
or 0.9 isDmore
where commondiffusion
t is turbulent in applying to the calculation
coefficient of pollutant
which is assumed to bedispersion in urban
isotropic and street canyons.
homogeneous.
However, as shown in Table 2, the Schmidt number of 0.9 is more
The turbulent diffusion coefficient Dt in Equation (6) is modeled by the relationship in favor of use by various
eddy-
authors than its value of 0.7. To make sure the selection of an appropriate value for the Schmidt
viscosity (𝜈 = 𝐶 ) and the turbulent Schmidt number 𝑆𝑐 , 𝐷 = , or in other words the Schmidt
number, we also recalibrated this value against Kastner-Klein’s experiment [30]. As shown in Figure 4,
number
the valueisofdependent
0.9 deliversupon the fit
the best ratio of turbulent
of the eddy viscosity and the turbulent mass diffusivity
measurements.
(𝑆𝑐 = ); thus, it is driven by the turbulent flow and has no universal value. Therefore, it needs to
Tableto
be calibrated 2. select of Sc
Values an t used in previous
appropriate valuestudies
for theonpollutant
building and urban diffusion
transport in urban simulation
street canyons as
mentioned
Value of Sctby Tominaga and Stathopoulos [27]. Spalding
Authors [28] confirmed that Sc t = 0.7 gave close
agreement with the experimental data, while Launder [29] found that Sct = 0.9 is a better value. So
0.5 Huang et al. [31]
that, the value Sct of 0.7 or 0.9 is more common in applying to the calculation of pollutant dispersion
0.63 Lien et al. [32]
in urban street canyons.
0.7
However, as shown Li in and Stathopoulos
Table [33], Wang
2, the Schmidt numberand McNamara
of 0.9 is more[34], Solazzo
in favoretof al. use
[35] by various
authors0.8than its value of 0.7. To make sure the selection Brzoska et of
al. an
[36]appropriate value for the Schmidt
number, we also Launder
recalibrated thisetvalue
[29], Sini al. [3],against Kastner-Klein’s
Delaunay [37], Baik et al. experiment
[38], Kim et al.[40].
[39],As shownetin
Santiago al.Figure
[40],
0.9
4, the value of 0.9 delivers the best fit of the measurements. Yassin [11]
Figure4.4. A
Figure A validation
validation of
of the
the Schmidt
Schmidt number
numberagainst
againstKastner-Klein’s
Kastner-Klein’sexperiment
experiment[30].
[40].
Table 2. Values of Sct used in previous studies on building and urban diffusion simulation.
Value of Sct Authors

0.5 Huang et al. [30]
0.63 Lien et al. [31]
0.7 Li and Stathopoulos [32], Wang and McNamara [33], Solazzo et al. [34]
0.8 Brzoska et al. [35]
Launder [29], Sini et al. [3], Delaunay [36], Baik et al. [37], Kim et al. [38], Santiago et al.
Atmosphere 2019, 10, 683 7 of 30
3. Initial and Boundary Conditions

In order to initialize an analogous condition to the actual initial conditions of pollutant dispersion
in urban street canyons, we first ran the simulation for airflow only to obtain the flow field,
including its parameters. Thereafter, we restarted the new run initialized by the previous run
with a turned-on-pollutant source released from a location close to the road surface.
At the inlet, a specific velocity profile is given at the inlet boundary, which is dependent on the
characteristics of incoming boundary layers of airflow.
In general, the turbulence parameters of the RNG k-ε turbulence model are set at the inlet boundary,
as follows:
3 k1.5
k = (UI )2 , ε = C0.75
µ
2 l
where U is mean velocity at the inlet; I is the turbulent intensity; l is the turbulent characteristic length.
At the building walls and road surfaces, a no-slip condition was set for velocity. The other
turbulence parameters such as turbulence kinetic energy and the specific dissipation rate were set to
wall functions:
u2∗ u3
k = 1/2 and ε = ∗
cµ κy
where u∗ = τo /ρ is the friction velocity at the nearest wall.

p
The dimensionless velocity profile (u+ = uU∗ ) is followed by the wall function, as follows (Kundu
and Cohen [41]):
y+ i f y+ < 5
(
+
u =
κ ln(Ey ) i f 30 < y < 500
1 + +
where y+ = yu∗ /ν is the dimensionless wall distance, y is the distance to the nearest wall and E is an
empirical constant related to the wall roughness; for a smooth wall E = 9.54.
The dimensionless concentration at the road surface is represented as follows:
K = CUHL/Q (7)
where: C [ppm] is the actual trace-gas concentration; U [m/s] is free stream velocity; H [m] is building
height; L [m] is length of line source; and Q [m3 /s] is volume rate of trace-gas.
The top of the computational domain is considered as a symmetry plane, whereby the velocity,
pressure and turbulence parameters were set to the zero gradient boundary condition.
The outlet boundary has been set to the cyclic condition.
4. Validations
First step, the new model is validated against the data from Li et al.’s experiment to study the
effects of different aspect ratios (ARs) on flow pattern and dispersion process in urban street canyons.
Thereafter, the model is validated against the data from Rafailidis and Schatzmann’s [42], Kastner-Klein
and Plate’s [30] and Llaguno-Munitxa et al.’s [43] experiments to study the effects of various roof
shape configurations.
4.1. Validation against the Measurement Data from Li et al.’s Experiment

Li et al. [2] carried out their experiment in a water-flume with the water depth of 40 cm for the
water flow over six to ten identical flat roof model buildings, whose length, width and height are
29.8 cm, 10 cm and 10 cm, respectively. Freestream velocity U is taken at z = 30 cm from the flume
bottom. Our Reynolds number is 12,000, based on the freestream velocity and the building height.
Experiments were implemented for three aspect ratios: AR = 2.0, 1.0 and 0.5, which all correspond to
the canyon widths of 5, 10 and 20 cm. The pollutant transport was not observed in this experiment.
The following sections are to present the hydrodynamic validations of the numerical model against
Atmosphere 2019, 10, 683 8 of 30
Liet
Li etal.’s
al.’sexperiments
experiments[2]
[2]for
for different
differentaspect
aspectratios
ratios(AR),
(AR),which
whichisisaarelationship
relationshipbetween
betweenthe
thebuilding
building
heightH
height Hand
andthe
thewidth
widthofofstreet
streetcanyon (AR==H/B).
canyonBB(AR H/B).
4.1.1.
4.1.1. Street
StreetCanyon
Canyonof
ofAspect
AspectRatio AR== 2.0
RatioAR 2.0
According
AccordingtotoOke’s Oke’s classification
classification [44], the flow
[44], in this
the flow inconfiguration
this configuration belongs to a skimming
belongs regime
to a skimming
whose
regime whose flow pattern is more complex than the flow of the aspect ratio AR = 1.0. Similarresults
flow pattern is more complex than the flow of the aspect ratio AR = 1.0. Similar to the to the
obtained from Liu
results obtained et al.Liu
from [1],etthe
al. flow
[1], thegenerates two main
flow generates twovortices in the street
main vortices in thecanyon, which are
street canyon, whichin
opposite directions;
are in opposite an anticlockwise
directions; vortex below
an anticlockwise vortexthebelow
half building
the halfheight, building andheight,
another andclockwise
another
vortex above
clockwise the half
vortex building
above height,
the half as shown
building in Figure
height, as shown5. Thisin isFigure
a phenomenon
5. This istoa explain
phenomenon why the to
streamwise
explain whyvelocities change velocities
the streamwise direction change
twice from bottom
direction to the
twice fromroof level as
bottom to shown
the roofinlevel
Figure 6a,c,e.
as shown
Figure 6 shows
in Figure 6a,c,e.aFigure
comparison6 shows between the numerical
a comparison between results and measurements
the numerical results andofmeasurements
the normalized of
stream-wise
the normalized (U/Ustream-wise
ref —left column) (U/Uand vertical
ref—left (W/Uref
column) and—right
verticalcolumn)
(W/Uref velocity
—rightprofilescolumn) along the
velocity
leeward
profiles (x/B
along= 0.25), center (x/B
the leeward (x/B == 0.5)
0.25),and windward
center (x/Band
(x/B = 0.5) = 0.75) lines in the
windward (x/Bstreet
= 0.75) canyon.
lines inItthe
shows a
street
very good agreement between our numerical results and Li et al.’s measurements.
canyon. It shows a very good agreement between our numerical results and Li et al.’s measurements. Within the street
canyon,
Within our the results are similar
street canyon, ourtoresults
the results of Li et to
are similar al. the
[14]results
using k-ε turbulent
of Li closure
et al. [14] using model, and the
k-ε turbulent
results
closureofmodel,
Liu et al.
and[1]theusing an LES
results model,
of Liu et al.whereas
[1] usingour anresults
LES model,show awhereas
bit better our agreement
results showwithathe bit
experimental
better agreement datawith
than theLi et al.’s and Liudata
experimental et al.’s
thanresults
Li et [1,2] beyond
al.’s and Liu the roofresults
et al.’s level. [1,2]
Figure 7 shows
beyond the
well-matched
roof level. Figurecomparisons
7 shows between
well-matched our results and the experimental
comparisons between our data results(Liand
et al.the
[2])experimental
of the horizontal
data
velocity
(Li et al.profiles at the
[2]) of the roof level
horizontal (z/H =profiles
velocity 1.0) and at at
thehalf
roofbuilding
level (z/H height and=at0.5).
= 1.0)(z/H half In general,
building our
height
results
(z/H = are
0.5).closer to the experimental
In general, our results are data thantothe
closer theresults obtaineddata
experimental fromthanLi etthe
al. results
[14] andobtained
Liu et al.from
[1].
Figure
Li et al.8 [14]
shows andthe
Liucomparisons of the
et al. [1]. Figure turbulence
8 shows kinetic energy
the comparisons (TKE)
of the betweenkinetic
turbulence our numerical
energy (TKE)and
experimental
between our results.
numerical and experimental results.
(a) (b)
(1)
Figure5.5.Streamlines
Figure Streamlines(a) (a)and
andvelocity
velocityvector
vector(b)
(b) (1) of
of wind
wind flow
flow inside
insidethe
thestreet
streetcanyons
canyonswith
withaspect
aspect
(AR)== 2.2. (( Due to the velocity inside the
ratio (1)
ratio(AR) (1) the canyons
canyons being
being too
too small
small in
in comparison
comparison with
withthe
the
ambient
ambientvelocity,
velocity,the themagnitude
magnitudeof ofthe
thevelocity
velocityvector
vectorpresented
presentedin inthe
thegraphs
graphs(from
(fromnow
nowon
onin
inthe
the
manuscript)
manuscript)isisscaled
scaledby bythe
thecolor
colorbar
barrather
ratherthan
thanby bythe
thelength
lengthofofthe
thevector).
vector).
4.1.2. Street Canyon of Aspect Ratio AR = 1

This case is similar to the flat roof shape experiments of Rafailidis and Schatzmann [42]. As
classified by Oke [44], the flow regime in this configuration (AR = 1) is skimming, whereby the bulk
flow does not enter the canyon, and this is characterized by a stable and isolated vortex at the street
canyon center due to an ambient flow outside on top of the street canyon, as shown in Figure 18a in
Section 4.2.1 below.
Atmosphere 2019, 10, 683 9 of 30
(a) (b)
(c) (d)
(e) (f)
Figure Thevertical
Figure 6. The vertical profile
profile of normalized
of normalized streamwise
streamwise (a,c,e)(a,c,e) and vertical
and vertical (b,d,f) velocities
(b,d,f) velocities with AR =with
2.0.
AR = 2.0.
x FOR PEER REVIEW
Atmosphere 2019, 10, 683 10
10of
of 30
(a) (b)
(a) (b)
(c) (d)
(c) (d)
Figure 7. The horizontal profile of the normalized velocities U/Uref (a,c) and W/Uref (b,d) with AR = 2.
Figure 7. The horizontal
Figure 7. horizontal profile
profileof
ofthe
thenormalized
normalizedvelocities
velocitiesU/Uref
U/Uref(a,c)
(a,c)and
andW/Uref
W/Uref(b,d)
(b,d) with
with ARAR = 2.
= 2.
(a) (b) (c)

(a) (b) (c)
Turbulencekinetic
Figure8.8.Turbulence
Figure kineticenergy x/H= =
energyatatx/H 0.25
0.25 (a),
(a), x/H= =
x/H 0.50.5
(b)(b)
andand x/H
x/H = 0.75
= 0.75 (c)(c) with
with ARAR = 2.
= 2.
Figure 8. Turbulence kinetic energy at x/H = 0.25 (a), x/H = 0.5 (b) and x/H = 0.75 (c) with AR = 2.
4.1.2.Figure
Street 9Canyon
shows the normalized,
of Aspect Ratio vertical
AR = 1 profiles of the streamwise (left column) and vertical (right
4.1.2. Street
column) Canyon
velocities of Aspect
along leeward Ratio
(x/BAR = 1 center (x/B = 0.5) and windward (x/B = 0.75) locations in
= 0.25),
This
the street case is similar to the flat roof
ARshape experiments
= 1. The numericalof Rafailidis and Schatzmann
velocity [42]. As)
This canyon
case is with
similaran to
aspect ratioroof
the flat shape experiments results
of of streamwise
Rafailidis and Schatzmann (U/U
[42]. ref
As
classified
show a by Oke
very good [44], the flowwith
agreement regimethe inexperimental
this configuration
data (AR
as = 1) ison
shown skimming,
the left wherebywhile
column, the bulk
the
classified by Oke [44], the flow regime in this configuration (AR = 1) is skimming, whereby the bulk
flow does not enter
numerical of the canyon, and (W/U
this is characterized by a stable anddataisolated
at x/B =vortex at the street
flow does results
not enter vertical velocity
the canyon, ) capture the observation
and this isrefcharacterized by a stable and isolated 0.25 and
vortex at 0.75 very
the street
canyon center due to an ambient flow outside on top of the street canyon, as shown in Figure 18a in
well.
canyonSimilar
centertodue
thetonumerical
an ambient results
flowobtained
outside onfrom
topthe simulation
of the of Rafailidis
street canyon, as shownandinSchatzmann’s
Figure 18a in
Section 4.2.1 below.
experiment, the result
Section 4.2.1 below. in this case does not capture well the vertical velocity profile (W/U ref ) at x/B =
Figure 9 shows the normalized, vertical profiles of the streamwise (left column) and vertical (right
Figure 9 shows the normalized, vertical profiles of the streamwise (left column) and vertical (right
column) velocities along leeward (x/B = 0.25), center (x/B = 0.5) and windward (x/B = 0.75) locations in the
column) velocities along leeward (x/B = 0.25), center (x/B = 0.5) and windward (x/B = 0.75) locations in the
street canyon with an aspect ratio AR = 1. The numerical results of streamwise velocity (U/Uref) show a
very good agreement with the experimental data as shown on the left column, while the numerical results
of vertical velocity (W/Uref) capture the observation data at x/B = 0.25 and 0.75 very well. Similar to the
Atmosphere 2019, 10, 683 11 of 30
numerical results obtained from the simulation of Rafailidis and Schatzmann’s experiment, the result in
this case does not capture well the vertical velocity profile (W/Uref) at x/B = 0.5. However, our results are
0.5.overall very similar
However, to theare
our results results obtained
overall very from Li et
similar toal.
the[14] usingobtained
results a k-ε model, andLibetter
from et al.than
[14]the results
using a k-ε
obtained
model, from Liu
and better et the
than al. [1] usingobtained
results the LESfrom
model.
LiuActually,
et al. [1]they also
using thefailed to reproduce
LES model. the vertical
Actually, they also
velocity
failed at the center
to reproduce theofvertical
the street canyonat(x/B
velocity the=center
0.5). of the street canyon (x/B = 0.5).
(a) (b)
(c) (d)
(e) (f)
Figure 9. The vertical profile of normalized stream-wise U/Uref (a,c,e) and vertical W/Uref (b,d,f)
velocities with AR = 1.
The vertical velocity profiles clearly show the flow moves up at upstream (near leeward) and
moves down at downstream (near windward), producing a stable clockwise vortex at the center of
the street canyon. Due to the clockwise vortex at the street canyon center, the stream-wise velocity
normalized stream-wise and vertical velocities along a roof level and half-height line. The vertical
velocity at roof level (Figure 10d) is very small due to the suppression of the upper motion by the
ambient wind, which makes the flow in the street canyon similar to the flow in a lid-driven cavity.
The vertical velocity at the half-height level (Figure 10b) changes direction at about the center of
Atmosphere
canyon 2019, 10,
width; 683it takes a positive value on the upstream half-width of the canyon, and a negative
i.e., 12 of 30
value on downstream half-width of the canyon. The stream-wise velocity at the half-height level
(Figure 10a)below
is negative is almost negative.
half-height Alllevel,
roof of these features over
and positive demonstrate the characteristics
the half-height of an isolated
level. The vertical velocity
circulation at the center of the street canyon as a cavity center.
approaches to zero at the roof-top level (z/H = 1) because the ambient wind above the roof-top level
playsFigure
a role11asshows the normalized
a driven-lid turbulence
for the street canyon. kinetic
Figure energy (𝑘⁄𝑈 the
10 shows ) at three vertical
horizontal lines;
profiles of the
the
leeward (x/B = 0.25), center (x/B = 0.5) and windward (x/B = 0.75) locations,
normalized stream-wise and vertical velocities along a roof level and half-height line. The vertical in the street canyon.
Similar
velocitytoatthe case
roof of AR
level = 2, it10d)
(Figure shows our model
is very small duecan topicktheup the turbulent
suppression kinetic
of the upperenergy inside
motion by the
the
street canyon (below the roof height); particularly the numerical results are
ambient wind, which makes the flow in the street canyon similar to the flow in a lid-driven cavity. The matching well with
experimental
vertical velocity data at at
thethe upstreamlevel
half-height and (Figure
center of thechanges
10b) street canyon,
directionbutatitabout
is notthe
well in agreement
center of canyon
beyond
width; i.e., it takes a positive value on the upstream half-width of the canyon, and a negative valuethe
the roof level, where occurs the interaction of the ambient flow and the flow from inside on
canyon.
downstream half-width of the canyon. The stream-wise velocity at the half-height level (Figure 10a) is
However,
almost negative. our numerical
All results of
of these features turbulence kinetic
demonstrate energy are still
the characteristics of anclosed to the
isolated experimental
circulation at the
data than the numerical results obtained
center of the street canyon as a cavity center. from Li et al. [14] and Liu et al. [1].

(a) (b)
(c) (d)
The vertical
Figure 10. The
Figure vertical profile
profile of
of stream-wise
stream-wise velocity
velocity (a,c)
(a,c) at
at center height (z/H =
center height 0.5), and
= 0.5), and vertical
vertical
velocity (b,d) roof
velocity (b,d) roof height (z/H = 1) with AR = 1.
= 1) with AR = 1.
Figure 11 shows the normalized turbulence kinetic energy (k/Ure 2 ) at three vertical lines; the
f
leeward (x/B = 0.25), center (x/B = 0.5) and windward (x/B = 0.75) locations, in the street canyon. Similar
to the case of AR = 2, it shows our model can pick up the turbulent kinetic energy inside the street
canyon (below the roof height); particularly the numerical results are matching well with experimental
data at the upstream and center of the street canyon, but it is not well in agreement beyond the roof
level, where occurs the interaction of the ambient flow and the flow from inside the canyon.
(a) (b) (c)

Figure 11. Turbulence kinetic energy at x/B = 0.25 (a), x/B = 0.5 (b) and x/B = 0.75 (c) with AR = 1.
(c) (d)
Figure 10. The vertical profile of stream-wise velocity (a,c) at center height (z/H = 0.5), and vertical
Atmosphere 2019, 10, 683 13 of 30
(c) (z/H = 1) with AR = 1.
velocity (b,d) roof height (d)
Figure 10. The vertical profile of stream-wise velocity (a,c) at center height (z/H = 0.5), and vertical
velocity (b,d) roof height (z/H = 1) with AR = 1.
(a) (b) (c)

Figure
Figure 11.11. Turbulence
Turbulence kinetic
kinetic energy x/B= =
energyatatx/B 0.25
0.25 (a),
(a), x/B
x/B = 0.5
= 0.5 (b)(b)
andand x/B
x/B = 0.75
= 0.75 (c)(c) with
with = 1.= 1.
ARAR
(a) (b) (c)
4.1.3.However,
Street Canyonour numerical
of Aspect results of turbulence kinetic energy are still closed to the experimental
Ratio 0.5
data Figure
than the11. Turbulence kinetic energy at x/B = 0.25 (a), x/B = 0.5 (b) and x/B = 0.75 (c) with AR = 1.
numerical results obtained from Li et al. [14] and Liu et al. [1].
Due to the aspect ratio of 0.5, the flow in this geometry configuration belongs to the wake
4.1.3. Street
Street Canyon
interference
4.1.3. flow regime
Canyon of
of Aspect Ratio
Ratio 0.5
in which
Aspect there are two horizontal interacting vortices; the downwind
0.5
building
Due disturbs the recirculation vortex before readjustmentconfiguration
can occur. Similar totothe results
Due to to the
the aspect
aspect ratio
ratio ofof 0.5,
0.5, the
the flow
flow in in this
this geometry
geometry configuration belongs belongs to the the wake
wake
obtained
interference fromflowLiu et
regimeal. [1],
in the
whichflow forms
there arethree
two vortices
horizontalinside the street
interacting canyon;
vortices; the
the largest
downwind one
interference flow regime in which there are two horizontal interacting vortices; the downwind building
occupies
building about
disturbs two thirds of
the recirculationthe area of
vortex the street canyon,
before readjustment and is located
can occur. on windward side; the
disturbs the recirculation vortex before readjustment can occur. Similar to theSimilar to the results
results obtained from
second large
obtained from one is et located
al. [1],on upstream nearthree
to the lower corner
insideon leeward; the littlethe
third vortex is
Liu et al. [1], theLiuflow forms the
three flow forms
vortices inside thevortices
street canyon; the street canyon;
the largest one occupies largest
about one
two
located
occupies at the
about right lower corner of the canyon, as shown in Figure 12.
thirds of the areatwo thirds
of the ofcanyon,
street the areaand of isthe streeton
located canyon,
windward and side;
is located on windward
the second large one is side; the
located
second large one
on upstream nearistolocated on upstream
the lower corner onnear to the the
leeward; lower corner
little thirdon leeward;
vortex the little
is located third
at the vortex
right loweris
located at the right lower corner
corner of the canyon, as shown in Figure 12. of the canyon, as shown in Figure 12.
(a) (b)
(a) (b)
Figure 12. Streamline (a) and velocity vector (b) of wind flow inside the street canyons with AR = 0.5.
Figures 13 and 14 show the comparisons between the numerical results and the experimental
measurements of the vertical profiles of the normalized streamwise and vertical velocities along the
vertical lines at x/B = 0.25, 0.5 and 0.75, and the horizontal lines at the roof level (z/H = 1) in the street
canyon. Although there are some discrepancies from the experiment data, the figures show that the
numerical model can capture the form of the velocity profiles.
Figure 15 shows the normalized turbulence kinetic energy (k/Ure 2 ) at three vertical lines; the
f
leeward (x/B = 0.25), center (x/B = 0.5) and windward (x/B = 0.75) locations inside the street canyon. It
shows that the numerical results are reasonably agreed with the experimental data at the leeward and
center locations, but not at the windward location. This trend is similar to the cases of AR = 1 and 2.
2, the turbulence kinetic energy (𝑘/𝑈 ) at z/H = 1 and z/H = 0.5 is poorly captured, which is also
indicated by a low value of R2 in combination with a high value of RMSE. Particularly, in the case of
AR = 0.5, at x/B = 0.75, the index R2 is below 0.5, however the value of RMSE is still smaller than those
in the same case (AR = 0.5). This different tendency of two indices comes from the different trends of
the numerical results and observation data as shown in Figure 13b,d,f, and even the deviation
Atmosphere 2019, 10, 683 14 of 30
between numerical results and observations is not large (shown by the value of RMSE).

(a) (b)
(c) (d)
(e) (f)
Figure 13. Thevertical
13.The vertical profile
profile of normalized
of normalized streamwise
streamwise (a,b,c)(a,b,c) and vertical
and vertical (d,e,f) velocities
(d,e,f) velocities with AR with
= 0.5.
AR = 0.5.
(e) (f)
Atmosphere 2019, 10, 683 15 of 30
Figure 13. The vertical profile of normalized streamwise (a,b,c) and vertical (d,e,f) velocities with AR = 0.5.
(a) (b)
14. The
Figure 14.
Figure The horizontal
horizontal profile
profile of
of the
thenormalized
normalizedvelocities
velocitiesU/U ref (a) and W/U
U/Uref ref (b) along a roof
W/Uref roof
level
level line
line with AR =
with AR Squares (Li et al. [9]), dash line (RNG).
= 0.5. Squares
(a) (b) (c)
Figure
Figure 15. Turbulence
15. Turbulence Intensity
Intensity at x/b =at0.25 = 0.25
x/b (a), x/b (a), (b)=and
= 0.5x/b 0.5x/b
(b) =and (c) =
0.75x/b 0.75AR
with 0.5. AR = 0.5.
(c)=with
Table 3 shows
Table the evaluation
3. Evaluation between the
of flow parameters numerical
between results and
the numerical experimental
results data ofdata.
and experimental velocity and
turbulence kinetic energy based on the efficiency criteria RMSE (root mean square 𝟐
error) and R2 (the
coefficient of determination) 𝑼/𝑼 𝑾/𝑼
𝒓𝒆𝒇= 0.5, 1.0 and 2.0. 𝒓𝒆𝒇 𝒌/𝑼 𝒓𝒆𝒇 good agreement
AR & Locationsfor all cases, AR It shows an overall
R 2 (1) RMSE (1) R2 RMSE
between the numerical results and the measurements. However, it should state that R 2 RMSE
there are still some
AR = 1, x/B = 0.25 0.9810 0.0475 0.9121 0.0372
specific locations where the numerical model still cannot well capture the observation 0.9581 0.0020 values. For
ARcase
instance, in the = 1, where
x/B = 0.5AR = 1,0.9875
the vertical 0.0393
velocity0.5942
(W/Uref )0.0696 0.9012
at x/B = 0.5 and the0.0030
turbulence kinetic
AR = 1, x/B = 0.75 0.9907 0.0349 0.8900 0.0195 0.9433 0.0044
energy (k/Ure f ) at z/H = 1 are not well captured by the numerical model, which fact is indicated by
2
AR = 1, z/H = 1 0.8959 0.0814 0.8023 0.0082 0.4133 0.0030
a low value of R2 in combination with a high value of RMSE; in the case of AR = 2, the turbulence
AR = 1, 2z/H = 0.5 0.7723 0.0107 0.9578 0.0560 0.9213 0.0029
kinetic energy (k/Ure ) at z/H = 1 and z/H = 0.5 is poorly captured, which is also indicated by a low
f = 0.25
AR = 2, x/B 0.9850 0.0410 0.8918 0.0233 0.9785 0.0011
value of R2 inARcombination
= 2, x/B = 0.5with a0.9776
high value of RMSE.
0.0404 Particularly,
0.6223 0.0271in the case of
0.9301 AR = 0.5, at x/B =
0.0015
0.75, the index 2
ARR= 2,is x/B
below 0.5, however
= 0.75 0.9899the value
0.0386of RMSE0.7510is still smaller
0.0464 than those
0.7368 in the same case
0.0027
(AR = 0.5). This different
AR = 2, z/H = 1 tendency of
0.9791 two indices
0.0417 comes
0.6918 0.0286 0.4618 0.0032the numerical
from the different trends of
results and observation data
AR = 2, z/H = 0.5 as shown
0.7605 in Figure
0.015713b,d,f, and even
0.9900 0.0236the deviation between numerical
0.3767 0.0036
results and observations
AR = 0.5, x/Bis= not0.25large (shown0.1661
0.9684 by the value of RMSE).
0.8871 0.0608 0.9786 0.0016
AR = 0.5, x/B = 0.5 0.9809 0.1221 0.8749
4.2. Validation against the Measurement Data from Rafailidis and Schatzmann’s 0.0252 0.8248 0.0060
Experiment
AR = 0.5, x/B = 0.75 0.9691 0.1743 0.4083 0.0212 0.6705 0.0102
In theseAR
experiments,
= 0.5, z/H =two-dimensional
1 0.6986 wind-tunnel
0.0817 models
0.8124 simulated
0.0394 from0.0104
0.7647 an urban boundary
layer corresponding to(1)the street-canyon configurations with eight flat
Definition of R and RMSE is described in the Appendix A.
2 (Figure 16a) or slanted roof-shape
buildings (Figure 16b) are studied.
4.2. Validation against the Measurement Data from Rafailidis and Schatzmann’s Experiment
In these experiments, two-dimensional wind-tunnel models simulated from an urban boundary
layer corresponding to the street-canyon configurations with eight flat (Figure 16a) or slanted roof-
shape buildings (Figure 16b) are studied.
The modeled street canyons have the building height of H, the street width of B, and the slanted
roof height of ZH. Velocity and turbulence intensity profiles were measured along vertical lines from
Atmosphere 2019, 10, 683 16 of 30
Table 3. Evaluation of flow parameters between the numerical results and experimental data.
U/Uref W/Uref k/U2ref

AR & Locations
R2 (1) RMSE (1) R2 RMSE R2 RMSE
AR = 1, x/B = 0.25 0.9810 0.0475 0.9121 0.0372 0.9581 0.0020
AR = 1, x/B = 0.5 0.9875 0.0393 0.5942 0.0696 0.9012 0.0030
AR = 1, x/B = 0.75 0.9907 0.0349 0.8900 0.0195 0.9433 0.0044
AR = 1, z/H = 1 0.8959 0.0814 0.8023 0.0082 0.4133 0.0030
AR = 1, z/H = 0.5 0.7723 0.0107 0.9578 0.0560 0.9213 0.0029
AR = 2, x/B = 0.25 0.9850 0.0410 0.8918 0.0233 0.9785 0.0011
AR = 2, x/B = 0.5 0.9776 0.0404 0.6223 0.0271 0.9301 0.0015
AR = 2, x/B = 0.75 0.9899 0.0386 0.7510 0.0464 0.7368 0.0027
AR = 2, z/H = 1 0.9791 0.0417 0.6918 0.0286 0.4618 0.0032
AR = 2, z/H = 0.5 0.7605 0.0157 0.9900 0.0236 0.3767 0.0036
AR = 0.5, x/B = 0.25 0.9684 0.1661 0.8871 0.0608 0.9786 0.0016
AR = 0.5, x/B = 0.5 0.9809 0.1221 0.8749 0.0252 0.8248 0.0060
AR = 0.5, x/B = 0.75 0.9691 0.1743 0.4083 0.0212 0.6705 0.0102
AR = 0.5, z/H = 1 0.6986 0.0817 0.8124 0.0394 0.7647 0.0104
(1) Definition of R2 and RMSE is described in the Appendix A.
(a) Flat roof shape
(b) Slanted roof shape

Figure
Figure 16. 16. Experimentalset-up
Experimental set-up by
by Rafailidis
Rafailidisand
andSchatzmann [42].[42].
Schatzmann (a) Flat
(a)roof
Flatshape, (b) Slanted
roof shape, roof
(b) Slanted
shape.
roof shape.
4.2.1. Flat Roof Shape

The modeled street canyons have the building height of H, the street width of B, and the slanted
roof height of Zexperiment,
In this H . Velocitytheand turbulence
modeled streetintensity profilesofwere
canyon consists eightmeasured along
idealized flat roofvertical lines
buildings withfrom
the road surface
the aspect over
ratio AR the
= H/Broofs
= 1.0at
(Hseveral
= B = 60locations. A tracer
mm). A steady sourcegas
S =S150
placed
ppmatis the bottom
released center
at the bottomof the
fourth street
center canyon
of the fourthwas a mixture
street canyon, ofas air andinethane
shown Figurewith the flowrate
5a above. Figure 17Qshows
air = 100
a L/h and
comparison Q ethane = 4
of the
L/h, normalized
respectively.vertical velocity profiles (w/U) at three positions inside the street canyon, being the
leeward (x/B =of0.25),
At the inlet this center (x/B = 0.5)
experiment, and windward
an inlet (x/B = 0.75)
velocity profile positions, between
was prescribed by the the numerical
power law, as in
results[45]:
Rafailidis and the measurements. The positive value of vertical velocity at the leeward side has shown
that the airflow moves upward from bottom α
U (z)to thezroof
− D height level, reaching the zero-value at the

roof top level due to the suppression of =
ambient wind, whereas it shows opposite direction at (8)
U (δ) δ−D
windward side. Consequently, a stable clockwise vortex is formed at the center of the street canyon.
The parameters
Figure aredistribution
18a shows the obtained from
of thethe measurement,
velocity where:
vector inside α =canyon.
the street 0.28, D = result
This 0.002 is
mthe
is the
displacement height above the ground, U(δ) = 5 m/s is free-stream velocity and δ = 0.5 m is the
same as those obtained from Yassin [11] using Fluent with the k-ε turbulence model, and Chan et al.
thickness ofusing
[8], also the boundary layer.
Fluent with the RNG k-ε turbulence model for flat roofs.
Figure 18b shows a very good agreement between the simulation results and measurements of
the vertical profile of pollutant concentration at the leeward (x/B = 0.25) and windward (x/B = 0.75)
locations in the street canyon, whereby the pollutant magnitude at the leeward side is significantly
reduced (about 50%) upward from the road surface to the building roof level, and almost double
Atmosphere 2019, 10, 683 17 of 30
4.2.1. Flat Roof Shape

In this experiment, the modeled street canyon consists of eight idealized flat roof buildings with
the aspect ratio AR = H/B = 1.0 (H = B = 60 mm). A steady source S = 150 ppm is released at the
bottom center of the fourth street canyon, as shown in Figure 5a above. Figure 17 shows a comparison
of the normalized vertical velocity profiles (w/U) at three positions inside the street canyon, being the
leeward (x/B = 0.25), center (x/B = 0.5) and windward (x/B = 0.75) positions, between the numerical
results and the measurements. The positive value of vertical velocity at the leeward side has shown
that the airflow moves upward from bottom to the roof height level, reaching the zero-value at the roof
top level due to the suppression of ambient wind, whereas it shows opposite direction at windward
side. Consequently, a stable clockwise vortex is formed at the center of the street canyon. Figure 18a
shows the distribution of the velocity vector inside the street canyon. This result is the same as those
obtained from10,
Atmosphere 2019, Yassin
x FOR [11]
PEERusing Fluent with the k-ε turbulence model, and Chan et al. [8], also18using
REVIEW of 30
Fluent with the RNG k-ε turbulence model for flat roofs.
At x/B= 0.25 At x/b=0.5 At x/B=0.75
1.2 At x/B= 0.25 1.2 At x/b=0.5 At x/B=0.75
1.2
1.2 1.2 Exp. 1.2
Exp.
Height
1 1 1
Height
1 1 Sim. 1
Sim.
0.8 0.8 0.8
0.8 0.8 0.8
0.6 0.6 0.6
0.6 0.6 0.6
0.4 0.4 0.4
0.4 0.4 0.4
0.2 0.2 0.2
0.2 0.2 0.2
w/U
0 w/U 0 0
-0.10 0 0.1 0.2 0.3 -0.1 0 0 0.1 0.2 -0.4 -0.2 0 0
-0.1 0 0.1 0.2 0.3 -0.1 0 0.1 0.2 -0.4 -0.2 0
Figure 17.
Figure A comparison
A comparison
17. A between
comparison between the
between the measurements
the measurements
measurements andand numerical
and numerical results
numerical results of
results of the
of the normalized
the normalized vertical
normalized vertical
vertical
Figure 17.
velocity
velocity profile
profile (w/U)
(w/U) at
at the
the leeward
leeward x/B
x/B =
= 0.25,
0.25, center
center x/B
x/B =
= 0.5
0.5 and
and windward
windward x/B
x/B =
= 0.75.
0.75.
velocity profile (w/U) at the leeward x/B = 0.25, center x/B = 0.5 and windward x/B = 0.75.
1 ExpWindward
1 ExpWindward
ExpLeeward
ExpLeeward
SimWindward
0.8 SimWindward
SimLeeward
0.8 SimLeeward
0.6
0.6
z/H
z/H
0.4
0.4
0.2
0.2
K
0 K
0
0 50 100 150
(a) 0 50 100 150
(a) (b)
(b)
Figure 18. The flow velocity field (a), and a comparison
comparison of the dimensionless concentration K in the
Figure 18. The flow velocity field (a), and a comparison of the dimensionless concentration K in the
street canyon between simulation and experiments
experiments (b).
(b).
street canyon between simulation and experiments (b).
4.2.2.Figure 18b shows

Slanted-Roof a very good agreement between the simulation results and measurements of the
Shape
4.2.2. Slanted-Roof
vertical Shape concentration at the leeward (x/B = 0.25) and windward (x/B = 0.75) locations
profile of pollutant
Similar
in theSimilar to
street to the flat roof buildings,
canyon, there are eight different slanted roof buildings where the ratios
the flatwhereby the pollutant
roof buildings, magnitude
there are at theslanted
eight different leeward side
roof is significantly
buildings where thereduced
ratios
between
(about the upward
50%) vertical slant height
fromheight and the building
the road height (ZH/H) level,
are of 0.5, 0.33 and 0.17, respectively.
between the vertical slant andsurface to the height
the building building
(ZHroof
/H) are of and almost
0.5, 0.33 anddouble larger than
0.17, respectively.
The source magnitude and released location are the same as in the flat roof case. The geometry of the
The source magnitude and released location are the same as in the flat roof case. The geometry of[11],
that along the windward side. In comparison with the results obtained from the Yassin simulation the
slanted roof
our numerical shape is
results shown in Figure
of pollutant 16b.
distribution are matching better with the observations of Rafailidis
slanted roof shape is shown in Figure 16b.
Figure 19 shows
and Schatzman the velocity vector in the street canyons with slanted roofs obtained from the
[42]. the
Figure 19 shows velocity vector in the street canyons with slanted roofs obtained from the
numerical results. It shows that the center vortices of wind flow in the street canyon are lifted upward
once the slope of the slanted roof is increased, and this similar tendency is also obtained from the
simulation of Yassin [11]. The locations of the vortex center obtained from our simulations are also very
similar to the numerical results of Yassin [11], particularly when the slope of the slanted is still small (𝑍 ≤
similar to the numerical results of Yassin [11], particularly when the slope of the slanted is still small (𝑍 ≤
0.17), and the vortex center location is kept the same as in flat roof situation. The same rotation in the
0.17), and the vortex center location is kept the same as in flat roof situation. The same rotation in the
Atmosphere 2019, 10, 683 18 of 30
4.2.2. Slanted-Roof Shape

Similar to the flat roof buildings, there are eight different slanted roof buildings where the ratios
between the vertical slant height and the building height (ZH /H) are of 0.5, 0.33 and 0.17, respectively.
The source magnitude and released location are the same as in the flat roof case. The geometry of the
slanted roof shape is shown in Figure 16b.
Figure 19 shows the velocity vector in the street canyons with slanted roofs obtained from the
similar to the numerical results of Yassin [11], particularly when the slope of the slanted is still small
(ZH ≤ 0.17), and the vortex center location is kept the same as in flat roof situation. The same rotation
in the clockwise
Atmosphere 2019, 10, direction
x FOR PEERofREVIEW
the vortices is obtained for both cases; the flat and slanted roof shapes.
19 of 30
ZH/H = 0.5 ZH/H = 0.33
ZH/H = 0.17
Figure 19. Velocity
Figure 19. Velocity vector
vector of
of wind
wind flow
flow inside
inside the
thestreet
streetcanyons
canyonswith
withslanted
slantedroofs.
roofs.
Figure 20 shows a comparison between the numerical results and experimental observations of
the normalized vertical concentration K at the leeward side (x/B = 0.25) and windward side (x/B =
0.75) with the slanted roof slope ZH = 0.5. It shows our model can capture very well the pollutant
concentration in the street canyon with this slanted roof shape. In particular, the pollutant concentration
obtained from our numerical result at the windward side matches better with the experiment data
than that result at the leeward side.
In addition, the same tendency as in the flat roof case, the vertical pollutant distribution along the
leeward side is much larger than those distributed along the windward side. The vertical pollutant
Atmosphere 2019, 10, 683 19 of 30
ZH/H = 0.17
concentration drops significantly upward along the leeward side (more than 50%), while it is almost
Figurealong
very little changed 19. Velocity vector of wind
the windward side.flow inside the street canyons with slanted roofs.
Figure 20.AAcomparison
Figure20. comparisonof ofthe
thedimensionless
dimensionlessconcentration
concentrationKKbetween
betweenexperiment
experimentand andcomputation
computation
(withZZHH/H
(with /H== 0.5)
0.5) at
at the
the leeward
leeward side (x/B== 0.25)
side (x/B 0.25) and
and at
at the
the windward
windward side (x/B == 0.75).
side (x/B
Figure 21 shows
In addition, the the
samedistribution
tendency as of pollutant
in the flatinside the street
roof case, canyons
the vertical of slanted
pollutant roof buildings
distribution along
with different
Atmosphere
the roof
2019,side
leeward slopes;
10, x FOR Z /H
PEERHREVIEW
is much =
larger than those distributed along the windward side.the
0.50, 0.33 and 0.17, respectively. This again shows that The pollutant
20 of 30
vertical
concentration at the leeward
pollutant concentration dropsside is always larger
significantly upward than at the
along thewindward
leeward sideside(more
for thethan
slopes /H) ofit
(ZHwhile
50%),
0.33
0.33 and
and 0.17,
0.17, while
while the
the concentration
concentration is
is quite
is almost very little changed along the windward side. symmetrical in
in the
the street
streetcanyon
canyon with
with the
theroof
roofslope
slopeof
of0.5.
0.5.In general,
In general, the
the pollutant
pollutant concentration
concentration is
is upwardly
upwardly decreased
decreased when the slope
slope
Figure 21 shows the distribution of pollutant inside the street canyons of slanted roof buildings of
of the
the roof
roof isis
increased.
increased.
with These
These
different results
results
roof are
are
slopes; similartotothe
Zsimilar theresults
resultsobtained
obtainedfrom
fromYassin
Yassin [11].
[11].
H/H = 0.50, 0.33 and 0.17, respectively. This again shows that the pollutant
concentration at the leeward side is always larger than at the windward side for the slopes (ZH/H) of
ZH/H = 0.5 ZH/H = 0.33
ZH/H = 0.17
Figure 21.Distribution
Figure21. Distributionofofconcentration
concentrationKKininthe
thestreet
streetcanyons
canyonswith
withslanted
slantedroofs.
roofs.
4.3. Validation against the Measurement Data from Kastner-Klein’s Experiment

Kastner-Klein [40] carried out the experiments in a boundary layer wind tunnel. The modeled
building height is of H = 12 cm, which is a typical average building height Hn = 20 m in urban areas.
Line sources are designed following Meroney et al. [46], and located at position A, distancing 35 mm
Atmosphere 2019, 10, 683 20 of 30
4.3. Validation against the Measurement Data from Kastner-Klein’s Experiment

Kastner-Klein [30] carried out the experiments in a boundary layer wind tunnel. The modeled
building height is of H = 12 cm, which is a typical average building height Hn = 20 m in urban areas.
Line sources are designed following Meroney et al. [46], and located at position A, distancing 35 mm
from building I, and at position B, distancing 85 mm from building I, as shown in Figure 22. The
pollutant source was mixture of air and a tracer gas (SF6), which mixture was released at location A
or B.
Figure 22. Schematic diagrams of the street canyon arrangement.
The approaching
As shown in Figurevertical
23, invelocity profiletoofthe
comparison thereference
wind is given by the
case (flat power
roof), the law
flowaspatterns
follows:inside
the urban street canyons of the situations described in Figure 22 are very distinguished from each
u z α
=
other. Only the flow in situation 3 (stepdown-stepup) is quite similar to the flow in the reference case.
ur zr
Moreover, the flow in the reference case, the situations 3, 4 and 5 form only one main vortex, whose
direction
where zr =is 100
clockwise,
m (similar except the flowuin
to nature), r = u100 = 7.7
situation m/s, and α = 0.23.
1 (stepup-stepup). TheAflow in the situations
normalized concentration4 and
5 is quite
value similar,
is shown in the location
Equation (7).of the vortex center is quite the same and located close to building roof
levelTheand geometry
depressedconfiguration
by outer ambient air flows.roof
of building In situations
shapes and 1 and
the2,source
the flow forms two
locations are vortices
shown in
the street
Figure 22. canyon, one main larger vortex and one small. However, the order of the two vortices is
different in these
Reference case:situations.
Flat roofWhereas the larger1:vortex
shapes; Situation One-side is a pitched
bit downward to the(step-up);
roof shapes street surface,
Situationand
theStep-up
2: smaller and vortex stepis located
down roof at the building
shapes; roof level
Situation near theroof
3: Slanted windward
shapes;side in situation
Situation 1; the
4: Street larger
canyon
vortex is slanted
between a bit upward
and flat to roof
the building height level
shapes; Situation and the
5: Street smaller
canyon vortexflat
between is located near the
and slanted roofleeward
shapes.
close
A andtoB street surface
are source in situation 2.
positions.
Figures
As shown 24inand 25 show
Figure 23, inacomparison
good agreement to the between
referencethe casenumerical
(flat roof),results
the flow and experimental
patterns inside
measurements
the urban streetofcanyons the dimensionless concentration
of the situations described K in
along the 22
Figure leeward
are veryand windward lines.
distinguished fromIneach the
case ofOnly
other. a flatthe
roofflow(reference case),
in situation similar to the results
3 (stepdown-stepup) is obtained fromtoRafailidis’
quite similar the flow inexperiments
the reference [45],
case.it
again shows
Moreover, thethat
flow theinconcentration
the reference at thethe
case, windward
situations is smaller
3, 4 andthan
5 format the
onlyleeward
one main side.
vortex, whose
direction is clockwise, except the flow in situation 1 (stepup-stepup). The flow in the situations 4 and 5
At similar,
is quite the leeward, the largest
the location of the concentration
vortex center isisquitefound thenear
sametoandstreet surface,
located closeand it dropsroof
to building off
significantly upward. In addition, below the half building height
level and depressed by outer ambient air flows. In situations 1 and 2, the flow forms two vortices in (z/H < 0.5) the pollutant
concentration
the street canyon, at the oneleeward site isvortex
main larger almost andtwice
one as largeHowever,
small. when thethe source
orderlocation changes
of the two vortices from is
location A
different in to position
these B. Whereas,
situations. Whereas thethe
pollutant concentration
larger vortex was not changed,
is a bit downward and was
to the street keptand
surface, almost the
steady from
smaller vortex theis road
locatedsurface
at thetobuilding
the roofrooflevellevel
at the
nearleeward and windward
the windward side in sites when1;the
situation the source
larger
location
vortex is changed
a bit upward fromto A the
to B, as shown
building in Figure
height level 24.
and the smaller vortex is located near the leeward
close Figure
to street 25surface
shows in the comparisons
situation 2. of pollutant concentrations released from position source A
with different building roof configurations. In the reference case, situations 2 and 5, as shown in
Figure 25a,c,f, the concentration at the leeward side is much larger than the concentration at the
windward side; moreover the concentration is significantly upward dropped at the leeward site,
while it keeps steady at the windward site. Whereas, in the situations 1, 3 and 4, as shown in Figure
25b,d,e, it shows an opposite trend to the results obtained from the reference case, situations 2 and 4;
i.e., the concentration at the windward side is much larger than the concentration at the leeward side,
and the concentration is significantly upward dropped at the windward site, while it keeps steady at
the leeward site.
Figure 26 shows the distribution of the turbulence kinetic energy (TKE) in the street canyons
Atmosphere
Atmosphere 2019,
2019, 10,10,
683x FOR PEER REVIEW 2122ofof
30 30
(a) Reference (b) Situation 1
(c) Situation 2 (d) Situation 3
(e) Situation 4 (f) Situation 5

Figure 23. Velocity vector of wind flow inside the street canyons with different roof shape
Figure 23. Velocity vector of wind flow inside the street canyons with different roof shape configurations.
configurations. (a): Reference case, (b–f): Situation 1–5.
(a): Reference case, (b–f): Situation 1–5.
Figures 24 and 25 show a good agreement between the numerical results and experimental
measurements of the dimensionless concentration K along the leeward and windward lines. In the
case of a flat roof (reference case), similar to the results obtained from Rafailidis’ experiments [45], it
again shows that the concentration at the windward is smaller than at the leeward side.
Figure 23. Velocity vector of wind flow inside the street canyons with different roof shape
Atmosphere 2019, 10, 683 22 of 30
configurations. (a): Reference case, (b–f): Situation 1–5.
Figure 24. The dimensionless concentration K along the leeward and windward sites in the reference
case (flat roof) corresponding to position sources located at A (left) and B (right).
At the leeward, the largest concentration is found near to street surface, and it drops off significantly
upward. In addition, below the half building height (z/H < 0.5) the pollutant concentration at the
leeward site is almost twice as large when the source location changes from location A to position B.
Whereas, the pollutant concentration was not changed, and was kept almost steady from the road
surface to the roof level at the leeward and windward sites when the source location changed from A
to B, as shown in Figure 24.
Figure 25 shows the comparisons of pollutant concentrations released from position source A
with different building roof configurations. In the reference case, situations 2 and 5, as shown in
Figure 25a,c,f, the concentration at the leeward side is much larger than the concentration at the
windward side; moreover the concentration is significantly upward dropped at the leeward site, while
it keeps steady at the windward site. Whereas, in the situations 1, 3 and 4, as shown in Figure 25b,d,e,
it shows an opposite trend to the results obtained from the reference case, situations 2 and 4; i.e., the
concentration at the windward side is much larger than the concentration at the leeward side, and
the concentration is significantly upward dropped at the windward site, while it keeps steady at the
leeward site.
Figure 26 shows the distribution of the turbulence kinetic energy (TKE) in the street canyons
with different building roof shape configurations. It shows that the low value of TKE usually takes
place at the center of the vortex, the corners and near road surface, where the fluctuating velocity is
small. In addition, the high value of TKE can be found at the windward side near roof level due to the
interaction of the downwind from the ambient flow with the stagnant air inside the canyon, which is
similar as lid-driven flow in a square cavity.
Table 4 shows the evaluation between the numerical results and experimental data of the
dimensionless concentration K along the leeward and windward sites for all cases following
Kastner-Klein’s experiments [30]. It shows a very good agreement between the numerical results and
the measurements. Particularly, in the cases of Reference and Situation 5, it shows a higher value of
RMSE in comparison with other cases; however, the index R2 is located in a very good range (>0.9).
These situations are referred to in Figure 25 (left top and right bottom), which are similar to the case
AR = 0.5, at x/B = 0.75 shown in Table 3 for flow parameters, when the different trends between the
simulation results and observation data occurs.
Figure 24. The dimensionless concentration K along the leeward and windward sites in the reference
Atmosphere 2019, 10, 683 23 of 30
case (flat roof) corresponding to position sources located at A (left) and B (right).
(a). Reference case (b). Situation 1
(c). Situation 2 (d). Situation 3
(e). Situation 4 (f). Situation 5

Figure 25. The dimensionless
dimensionless concentration
concentrationKKalong
alongthe theleeward
leeward and
and windward
windward sites
sites in the
in the casecase of
of flat
flat
roofroof corresponding
corresponding to position
to position sources
sources located
located at A.at(a):
A. Reference
(a): Reference
case,case, (b–f):
(b–f): Situation
Situation 1–5.1–5.
Table 4 shows the evaluation between the numerical results and experimental data of the
dimensionless concentration K along the leeward and windward sites for all cases following Kastner-
Klein’s experiments [40]. It shows a very good agreement between the numerical results and the
measurements. Particularly, in the cases of Reference and Situation 5, it shows a higher value of RMSE
in comparison with other cases; however, the index R2 is located in a very good range (>0.9). These
situations are referred to in Figure 25 (left top and right bottom), which are similar to the case AR =
Atmosphere 2019, 10, 683 24 of 30
0.5, at x/B = 0.75 shown in Table 3 for flow parameters, when the different trends between the
simulation results and observation data occurs.
(a) Reference (b) Situation 1
(c) Situation 2 (d) Situation 3
Figure Turbulence
26.26.
Figure Turbulence kinetic energy
kinetic energyof of
wind
wind flow inside
flow insidethethe
street canyons
street with
canyons different
with roof
different shape
roof shape
configurations. (a):(a):
configurations. Reference
Referencecase, (b–f):
case, (b–f):Situation
Situation 1–5.
1–5.
Table 4. Evaluation of the dimensionless concentration K between the numerical results and
Table 4. Evaluation of the dimensionless concentration K between the numerical results and
experimental data.
experimental data.
Concentration K along Leeward Concentration K along Windward
Case Concentration K along Leeward Concentration K along Windward
Case R22 RMSE R2R2 RMSE
R RMSE RMSE
Reference
Reference 0.9249
0.9249 7.3128
7.3128 0.9516
0.9516 2.0042
2.0042
Situation
Situation1 1 0.8708
0.8708 3.6100
3.6100 0.8940
0.8940 5.0342
5.0342
Situation 2 0.9906 2.2443 0.7284 2.5450
Situation 2 0.9906 2.2443 0.7284 2.5450
Situation 3 0.7678 1.9154 0.9975 1.1540
Situation4 3
Situation 0.7678
0.7641 1.9154
2.1600 0.9975
0.9935 1.1540
2.0394
Situation5 4
Situation 0.7641
0.9611 2.1600
12.0912 0.9935
0.8941 2.0394
0.3091
Situation 5 0.9611 12.0912 0.8941 0.3091
Atmosphere 2019, 10, 683 25 of 30

4.4. Validation against the Measurement Data from Llaguno-Munitxa et al.’s Experiment
4.4. Validation against the Measurement Data from Llaguno-Munitxa et al.’s Experiment
In this section, we apply the numerical model to study the flow over buildings with round roof
4.4. Validation against wetheapply
Measurement Data from Llaguno-Munitxa et al.’s Experiment
shapes.this
In Thesection,
numerical resultsthe havenumerical model
been validated to studythe
against thedataflow over
obtained buildings
from the with round roof
experiment of
shapes. The
Llaguno-Munitxa numerical
In this section,etwe results
al.apply
[43]. Thehave been
the numerical
experimentsvalidated
model against
weretocarried the
study the data
in theflow obtained
overtunnel
wind from
buildings the experiment
with round
to observe the flow of
roof
Llaguno-Munitxa
shapes.
over seven Thebuilding
numerical et al. [43]. The
results
models byhave experiments
the been validated
box size werem
(0.07 carried
against
× 0.07the in data
m) the
andwind obtainedtunnel
the canyon fromto the
width observe the m,
experiment
B = 0.07 flow of
as
over seven
Llaguno-Munitxa
shown in Figure 27.building models
et al. by the box
[43]. The experiments
Llaguno-Munitxa size (0.07
et al. were m × 0.07
carried out
[43] carried m)
in the and
wind
their the canyon width
tunnel to observe
experiments B
with three = 0.07
the flow m,
differentoveras
shown
seven in Figure
building
inlet boundary 27. Llaguno-Munitxa
models
conditions; by the box sizewe
however, etvalidate
(0.07 al.
m[43]× 0.07carried
only m)for out
and thethetheir
canyon
case experiments
with width
uniform B =with
0.07 three
inlet different
m, as shown
velocity U = 10 in
inlet
Figure boundary
27. conditions;
Llaguno-Munitxa however,
et al. [43] we validate
carried out only
their for the
experiments
m/s in this study. Figure 28 shows the comparison of the velocity profile of normalized streamwise case with
with uniform
three inlet
different velocity
inlet U
boundary = 10
m/s
(left)inand
thisthe
conditions; study.
however, Figure
turbulence 28 shows
we validate
kinetic thefor
only
energy comparison
the case
(right) with
at the ofcenter
the velocity
uniform profile
inlet
of canyon (x/Bof= normalized
velocity U = 10
0.5), m/s in
between streamwise
this study.
numerical
(left)
Figure and
results 28 the turbulence
shows from
obtained kinetic
the comparison
our model of energy
andthethe(right)
velocity at
resultsprofilethe center
obtained of
of normalizedcanyon
from the experiment (x/B =
streamwise 0.5),andbetween
(Figure numerical
28a)
simulation and (usingthe
results
turbulenceobtained kinetic from our
energy model
(Figure and28b)theatresults
the obtained
center of from
canyon
LES by Fluent software) of Llaguno-Munitxa et al. [43]. It shows that the simulation by the LES the
(x/B experiment
= 0.5), betweenand simulation
numerical (using
results
LES
method by Fluent
obtained from our
captures software)
model
the of
andLlaguno-Munitxa
streamwise the results
velocityobtained
profileet al.better
[43].the
from It shows
than experiment that the
our model, and simulation
simulation
whereas ourby the LES
(using
model LES
can
method
by Fluent
capture captures
the software)
turbulence the streamwise
of Llaguno-Munitxa velocity
kinetic energy better profile
et al.than[43]. better
theItLES
showsthan
method our
that the model,
by simulation whereas
Llaguno-Munitxa our
by the LES model
et al.method
[43]. can In
capture
captures
addition,thethe turbulence
streamwise to
in comparison kinetic
velocityenergy better
profile case
the reference than
better(flat the
than LES
ourand
roof method
model,
AR =whereasby Llaguno-Munitxa
1) in theour model can
experiment et al. [43].
of capture
Li et al. the In
[2],
addition,
turbulence
the tendency inkinetic
comparison energytobetter
of turbulence the reference
kineticthan thecase
energy LESat (flat
method
the roofby
street and AR =center
Llaguno-Munitxa
canyon 1) in the
(x/Bexperiment
=et0.5)
al. [43]. of addition,
In
is similar Li to
et al.
that[2], in
the
thistendency
comparison
case; i.e.,totheoftheturbulence
reference
value kinetic
case (flat
of turbulence energy
roof and
kinetic atenergy
the
AR street
= is incanyon
1)small thebelow center
experiment (x/B
4/5 height =roof-level,
of Li 0.5)
et al.is[2],
similar
the toitthat
tendency
while jumps in
of
this case;
turbulence i.e.,
significantlykinetic the value
near to energyof turbulence
the roofat the kinetic
street
level. canyon
Figure energy is small
center (x/B
29 shows below
a comparison 4/5
= 0.5) is similar height
of theto roof-level,
that in this
pressure while it jumps
case; i.e.,and
distribution the
significantly
value
velocity vector near
of turbulence fieldtokinetic
the roof
(background) level.isinside
energy Figure
smallthe 29street
below shows a comparison
4/5canyon
height roof-level,
and ambient ofwhile
the
flowpressure
itover
jumps topsdistribution
significantly
of round roofs, and
near
velocity
to the roof
which vector
are level. field
obtained (background)
Figurefrom 29 our
shows inside the street
a comparison
model (left) and of canyon
thetheLES and ambient
pressure
simulation offlow
distribution over tops of round
and velocity
Llaguno-Munitxa vector
et al. roofs,
field
[43]
which are
(background)
(right). It shows obtained
inside from
thatthe our model
street canyon
a similar trend isand(left) and
ambient
obtained the
from LES
flow
both simulation
over of Llaguno-Munitxa
tops of round roofs, which are obtained
models. et al. [43]
(right).
from our It shows
modelthat (Figurea similar
29a) andtrend theis LES
obtained from both
simulation models.
of Llaguno-Munitxa et al. [43] (Figure 29b). It
shows that a similar trend is obtained from both models.
H
H
5H B=H 15H
5H B=H 15H
Figure 27. Geometry configuration of Llaguno-Munitxa et al.’s experiment [43].
(a) (b)
Figure 28. The velocity profile of normalized streamwise (left) and turbulence kinetic energy (right)
Figure 28.
Figure The velocity
28. The velocity profile of normalized streamwise (a) and turbulence kinetic energy (b) at the
at the center of canyon.profile of normalized streamwise (left) and turbulence kinetic energy (right)
center
at of canyon.
the center of canyon.
Atmosphere 2019, 10, 683 26 of 30
(a) (b)
Figure 29. A comparison of the pressure distribution between numerical results obtained from our
Figure 29. A comparison of the pressure distribution between numerical results obtained from our
model (left) and Llaguno-Munitxa et al. [43] (right).
model (a) and Llaguno-Munitxa et al. [43] (b).
5.
5. Discussion
Discussion
From
Fromthe thevalidation
validation of of
thethe
numerical
numerical model modelagainst various
against experimental
various data obtained
experimental from various
data obtained from
experiments, such as Lisuch
various experiments, et al.’s
as Liexperiment [2] for different
et al.’s experiment aspect ratios
[2] for different aspect(H/B = 2.0,
ratios 1.0, =and
(H/B 2.0,0.5);
1.0, and
and
Rafailidis
0.5); and and Schatzmann’s
Rafailidis [42], Kastner-Klein’s
and Schatzmann’s [40] and Llaguno-Munitxa
[42], Kastner-Klein’s et al.’s [43] experiments
[30] and Llaguno-Munitxa for
et al.’s [43]
various roof shape combinations, it shows that the flow pattern
experiments for various roof shape combinations, it shows that the flow pattern and pollutant and pollutant distribution are strongly
dependent
distribution uponare the roof geometries
strongly dependentand upon their configurations
the roof geometries of theandurban
theirstreet canyons. of the urban
configurations
street canyons.
5.1. Effect of Aspect Ratios on Flow Patterns and Pollutant Transport
5.1. Effect of Aspect Ratios on Flow Patterns and Pollutant Transport
The numerical simulations for different aspect ratios (AR) following Li et al.’s experiment [2] are
shown Theat numerical
Section 4.2.simulations
The flows over for different
the street aspect
canyons ratios
with(AR) thefollowing Li etof
aspect ratios al.’s
2.0experiment
and 1.0 belong [2] are
to
shown
the at Section
skimming 4.2. The
regime flows over
following the street canyons
the classification by Oke with[44];the aspect ratios
whereby the flowof 2.0 and 1.0
inside thebelong
canyons to
the
is skimming regime
characterized by thefollowing
lid-driventhe classification
cavity flow because by Oke [44]; whereby
the ambient the flow
wind above theinside
roof-topthelevel
canyonsplays is
characterized
as a driven-lidby forthethelid-driven
street canyon. cavityInflow because
the case of AR the= ambient
2, the flow wind above the
generates tworoof-top
vorticeslevelin the plays
streetas
a driven-lid
canyon, which foraretheinstreet
opposite canyon. In the case
directions; of AR = 2, the
an anticlockwise flow below
vortex generates two building
the half vortices in the street
height, and
acanyon,
clockwise which are in
vortex opposite
above the halfdirections;
building anheight.
anticlockwise vortex below the half building height, and a
clockwise
Whereas,vortextheabove the half
flow forms building
a stable height.vortex at the center of street canyon in the case of AR
clockwise
= 1, andWhereas,
the vortexthe flow forms aisstable
circulation clockwise
strongest in thisvortex
case asatmentioned
the center of bystreet
Oke [44]canyon in the
as well. The case of AR
number
= 1,
of theand the vortex
vortices circulation
generated insideisthe strongest
canyonin this case astomentioned
correspond the number byofOke
times[44]when
as well.
the The number
direction of
of the
the vortices
vertical generated
profile inside the
of streamwise canyon
velocity correspond
is changed from to negative
the number of timesand
to positive when vicethe direction
versa.; e.g.,
of the
the vertical profile
stream-wise velocity ofisstreamwise
negative below velocity is changedroof
the half-height fromlevel,negative to positive
and positive overandthisvice
sameversa.;
half-
e.g., thelevel
height stream-wise velocitycase
in the reference is negative
AR = 1, below
and the thestreamwise
half-heightvelocities
roof level,change
and positive
direction over this from
twice same
half-height
bottom to thelevel
roofinlevel
the reference
in the case of AR == 2.
case 1, Inand thethe streamwise
case of AR = 0.5, velocities
the flow change
belongs direction twice
to the wake
from bottomflow
interference to the roof level
regime, in thetwo
in which case of AR
main = 2. In the
interacting case ofand
vortices ARa=small0.5, the flowform
vortex belongsinside to the
the
wake interference flow regime, in which two main interacting
canyon; the downwind building disturbs the recirculation vortex before readjustment can occur; thevortices and a small vortex form inside
the canyon;
largest vortex thetakes
downwind
place on building disturbsside,
the windward the recirculation
and occupiesvortex aboutbefore
two thirdsreadjustment
of the area can ofoccur;
the
the largest
street canyon; vortex takes place
the second largeon the windward
vortex takes place side, andleeward
on the occupiesward aboutsidetwonearthirds of the
to the lower area of the
corner;
street
and thecanyon; the second
little third vortex is large vortex
located takes
at the rightplaceloweron the leeward
corner of theward
canyon side(Figure
near to18a).
the lower corner;
and the little third
Moreover, vortexofisturbulence
the value located atkinetic
the right lower
energy corner ofsmall
is generally the canyon (Figure
below about 4/518a).
roof-level height,
whileMoreover,
it significantlythe increases
value of turbulence
upward from kinetic
there energy
to near theis generally
roof level.small below
This trend is about
also found4/5 roof-level
the same
height,
in the roundwhile it significantly
roofs in Llaguno-Munitxa increases et upward from there
al.’s experiment [43], to near the
whereby theroof
AR level.
(H/B) isThisalsotrend
of 1. is also
found the same in the round roofs in Llaguno-Munitxa et al.’s experiment [43], whereby the AR (H/B)
5.2. Effect
is also of Roof Shapes on the Flow Patterns and Pollutant Distribution
of 1.
In thisofstudy,
5.2. Effect the flat
Roof Shapes onroof withPatterns
the Flow the aspect
andratio AR = Distribution
Pollutant 1.0 was used as a reference case to compare
the results. From the simulations following Rafailidis and Schatzmann’s [42], Kastner-Klein’s [40] and
In this study, et
Llaguno-Munitxa theal.’s
flat [43]
roof experiments,
with the aspect
it ratio
shows = 1.0
ARthat thewas used
flow as a reference
patterns caseurban
inside the to compare
street
the results. From the simulations following Rafailidis and Schatzmann’s [42], Kastner-Klein’s
canyons under the effects of roof geometries are very distinguished from each other. Only the flow [30]
and aLlaguno-Munitxa
over street canyon formedet al.’s
by [43] experiments,
the slanted it shows
roof with that the
the slanted flow
slope patterns
ZH/H inside
= 0.17 (of the urban
Rafailidis and
Schatzmann’s experiment [42]) or stepdown-stepup in the situation 3 (of Kastner-Klein’s experiment
[40]) or round roofs (of Llaguno-Munitxa et al.’s experiments [43]) is quite similar to the flow in the
Atmosphere 2019, 10, 683 27 of 30
street canyons under the effects of roof geometries are very distinguished from each other. Only
the flow over a street canyon formed by the slanted roof with the slanted slope ZH /H = 0.17 (of
Rafailidis and Schatzmann’s experiment [42]) or stepdown-stepup in the situation 3 (of Kastner-Klein’s
experiment [30]) or round roofs (of Llaguno-Munitxa et al.’s experiments [43]) is quite similar to the
flow in the reference case, i.e., the flow forms a stable clockwise vortex at the center of the street canyon.
However, when the slope of a slanted roof (ZH /H) is increased over 0.33 (as shown in the Rafailidis and
Schatzmann experiment [42]) or the roof shapes change from flat roofs (reference case) to stepdown-flat
(situation 4) or to flat-stepup (situation 5) configurations (in Kastner-Klein’s experiment [30]), the
center of the vortex is lifted upward to the roof level.
In the stepup-stepup configuration (situations 1), the flow inside the canyon interfered with by
the outside ambient flow forms two vortices, a main larger one below and a smaller one located at the
building roof level. In the stepup-stepdown configuration (situation 2), this geometry increases the
aspect ratio, and consequently the flow forms two vortices in the street canyon, where the larger vortex
is a bit upward above mid-roof level, and the smaller vortex is located close to the street surface.
In comparison to the reference case, the vertical distributions of pollutant concentration at
the leeward side in the situations 2 (stepup-stepdown) and 5 (flat-stepup) are the same as in the
reference case; i.e., the pollutant concentrations at leeward is much larger than those at the windward
side, particularly below the mid-height roof level. In addition, the vertical pollutant concentration
significantly decreases upward at the leeward site, while it keeps quite steady at the windward site.
Whereas, in the situations 1 (stepup-stepup), 3 (stepdown-stepup) and 4 (stepdown-flat), it shows an
opposite trend to the results obtained from the reference case, that is situations 2 and 5. This is to say
that the vertical concentrations at the windward side are much larger than those at the leeward site,
and the concentration significantly drops upward at windward site, while it keeps steady at leeward
site. Moreover, the vertical pollutant concentration gradually decreases upward at windward site, and
is kept steady at the leeward site in those cases (situations 1, 3 and 4).
From the results obtained from the reference case, it shows that the pollutant sources regardless
are released at the location A or B, the vertical distribution of pollutant concentrations at windward
site has similar features in magnitude and tendency; i.e., it has the same small magnitude, and is kept
steadily upward (Figure 24). Since the location B is located at the windward site distancing from the
leeward site, the vertical concentrations consequently at the leeward are smaller when the source is
released from location B in comparison with those released at location A.
6. Conclusions
As shown above, we developed a numerical model based upon an open source CFD package
OpenFOAM in order to investigate the flow pattern and pollutant dispersion in urban street canyons.
Before we can apply the model to investigate the effects of various geometry configurations, such as
different roof shapes and aspect ratios, the model was carefully validated for different two-equation
turbulence closure models and various Schmidt numbers. Finally, the RNG k-ε turbulence model and
the Schmidt numbers Sct = 0.9 are selected.
The numerical model was validated against the experimental data obtained from a number of
well-known experiments, such as Li et al.’s experiment [2] for different aspect ratios (H/B = 2.0, 1.0, and
0.5), as well as Rafailidis and Schatzmann’s [42], Kastner-Klein’s [30] and Llaguno-Munitxa et al.’s [43]
experiments for various roof shapes and their combinations. Overall, the numerical results show very
good agreements with the measurements. It shows the ability of the numerical model, which can be
used to investigate the flows and pollutant dispersions in urban street canyons. However, it should be
noted that there still exist some differences between the numerical results and the observations at certain
locations, as mentioned in Tables 3 and 4 above. Therefore, the numerical model will be continued
to implement more intensively the quantitative validations against the observation data in various
street canyon configurations. The numerical model is actually capable of simulating three-dimensional
geometries; nevertheless, due to the limitation of available observation data in three-dimensional
Atmosphere 2019, 10, 683 28 of 30
regions, our validations are still limited with 2D available data sets. We are currently collecting the
data from typical real urban street canyons in order to further validate against a real case study. We
also plan to take into account the effect of temperature and planted trees in urban street canyons.
Author Contributions: Conceptualization, V.T.N. and J.N.; Methodology, V.T.N. and T.C.N.; Software, T.C.N.
and V.T.N.; Validation, T.C.N. and J.N.; Formal Analysis, V.T.N. and T.C.N.; Investigation, T.C.N. and J.N.; Data
Curation, V.T.N. and T.C.N.; Writing—Original Draft Preparation, V.T.N.; Writing—Review & Editing, V.T.N.;
Visualization, T.C.N. and J.N.; Supervision, V.T.N.; Project Administration, V.T.N.
Funding: This research was funded by the National Research Foundation of Korea under the grant
(NRF-2018R1D1A1A09083747).
Acknowledgments: The authors would like to thank the support from the National Research Foundation of Korea
under the grant (NRF-2018R1D1A1A09083747). The authors also would like to thank the anonymous reviewers
for their valuable and constructive comments to improve our manuscript.
Conflicts of Interest: The authors declare no conflict of interest.
Appendix A
Table A1. List of evaluation parameters used in the manuscript.
Evaluation Parameter Equation Author

n (O −P )2 0.5
" #
1 Root Mean Square Error, (RMSE) i i
Moriasi et al. [47]
P
RMSE = n
i=1
 2

 n
P − − 

 (Oi −O)(Pi −P) 

Coefficient of determination, R2 R2 = 
 i=1

3 Moriasi et al. [47]
 
" 0.5 " n
_ 2 0.5
 n _ 2


 P P 
 i = 1 −O )
( O ( Pi − P )

 i


i=1
_ _
Where: Oi : Observe, Pi : Predict, O: average of the observe value, P: average of the predicted value.
References
1. Liu, C.-H.; Barth, M.C.; Leung, D.Y.C. Large-Eddy Simulation of flow and pollutant transport in street
canyons of different building-height-to-street-width ratios. J. Appl. Meteorol. 2004, 43, 1410–1424. [CrossRef]
2. Li, X.X.; Leung, D.Y.C.; Liu, C.H.; Lam, K.M. Physical Modeling of flow field inside urban street canyons.
J. Appl. Meteorol. Climatol. 2008, 47, 2058–2067. [CrossRef]
3. Sini, J.F.; Anquetin, S.; Mestayer, P.G. Pollutant dispersion and thermal effects in urban street canyons. Atmos.
Environ. 1996, 30, 2659–2677. [CrossRef]
4. Johnson, G.T.; Hunter, L.J. Urban wind flows: Wind tunnel and numerical simulations—A preliminary
comparison. Environ. Model. Softw. 1998, 13, 279–286. [CrossRef]
5. Baik, J.J.; Kim, J.J. A numerical study of flow and pollutant dispersion characteristics in urban street canyons.
J. Appl. Meteorol. 1999, 38, 1576–1589. [CrossRef]
6. Baik, J.J.; Kim, J.J. On the escape of pollutants from urban street canyons. Atmos. Environ. 2002, 36, 527–536.
[CrossRef]
7. Chan, A.T.; Au, W.T.W.; So, E.S.P. Strategic guidelines for street canyon geometry to achieve sustainable
street air quality—Part II: Multiple canopies and canyons. Atmos. Environ. 2003, 37, 2761–2772. [CrossRef]
8. Chan, T.L.; Dong, G.; Leung, C.W.; Cheung, C.S.; Hung, W.T. Validation of a two-dimensional pollutant
dispersion model in an isolated street canyon. Atmos. Environ. 2002, 36, 861–872. [CrossRef]
9. Jeong, S.J.; Andrews, M.J. Application of the k-epsilon turbulence model to the high Reynolds number
skimming flow field of an urban street canyon. Atmos. Environ. 2002, 36, 1137–1145. [CrossRef]
10. Takano, Y.; Moonen, P. On the influence of roof shape on flow and dispersion in urban street canyon. J. Wind
Eng. Ind. Aerodyn. 2013, 123, 107–120. [CrossRef]
11. Yassin, M.F. Impact of height and shape of building roof on air quality in urban street canyons. Atmos.
Environ. 2011, 45, 5220–5229. [CrossRef]
12. Leitl, B.M.; Meroney, R.N. Car exhaust dispersion in a street canyon. Numerical critique of a wind tunnel
experiment. J. Wind Eng. Ind. Aerodyn. 1997, 67–68, 293–304. [CrossRef]
Atmosphere 2019, 10, 683 29 of 30
13. Sagrado, A.P.G.; van Beeck, J.; Rambaud, P.; Olivari, D. Numerical and experimental modelling of pollutant
dispersion in a street canyon. J. Wind Eng. Ind. Aerodyn. 2002, 90, 321–339. [CrossRef]
14. Li, X.-X.; Liu, C.-H.; Leung, D.Y.C. Development of a k-ε model for the determination of air exchange rates
for street canyons. Atmos. Environ. 2005, 398, 7285–7296. [CrossRef]
15. Raw, M.J.; Galpin, P.F.; Hutchinson, B.R. A collocated finite-volume method for solving the Navier-Stokes
equations for incompressible and compressible flows in Turbomachinery: Results and applications. Can.
Aeronaut. Space J. 1989, 35, 189–196.
16. Walton, A.; Cheng, A.Y.S. Large eddy simulation of pollution dispersion in an urban street canyon-part II:
Idealized canyon simulation. Atmos. Environ. 2002, 36, 3615–3627. [CrossRef]
17. Hassan, A.A.; Crowther, J.M. Modeling of fluid flow and pollutant dispersion in a street canyon. Environ.
Monit. Assess. 1998, 53, 281–297. [CrossRef]
18. Koutsourakis, N.; Bartzis, J.G.; Markatos, C.N. Evaluation of Reynolds stress, k-ε turbulence models in street
canyon flows using various experimental datasets. Environ. Fluid Mech. 2012, 12, 379–403. [CrossRef]
19. Brown, M.J.; Lawson, R.E.; DeCroix, D.S.; Lee, R.L. Mean flow and turbulence measurements around a 2-D
array of building in a wind tunnel. In Proceedings of the 11th AMS Joint Conference on the Applications of
Air Pollution Meteorology, Long Beach, CA, USA, 10 January 2000.
20. Demirdzic, I.A. A Finite Volume Method for Computation of Fluid Flow in Complex Geometries; Imperial College
London (University of London): London, UK, 1982.
21. Thangam, S.; Speziale, C.G. Turbulent flow past a backward facing step: A critical evaluation of two-equation
models. AIAA J. 1992, 30, 1314. [CrossRef]
22. Sahm, P.; Louka, P.; Ketzel, M.; Guilloteau, E.; Sini, J.F. Intercomparison of Numerical Urban Dispersion
Models—Part I: Street Canyon and Single Building Configurations. Water Air Soil Pollut. Focus 2002, 2,
587–601. [CrossRef]
23. Apsley, D.; Castro, I.P. Flow and dispersion over hills: Comparison between numerical predictions and
experimental data. J. Wind Eng. Ind. Aerodyn. 1997, 67, 375–386. [CrossRef]
24. Yakhot, V.; Orszag, S.A.; Thangam, S.; Gatski, T.B.; Speziale, C.G. Development of Turbulence Models for
Shear Flows by a Double Expansion technique. Phys. Fluids Fluid Dyn. 1992, 4. [CrossRef]
25. Rotach, M.W. Profiles of Turbulence Statistics and Above an Urban Street Canyon. Atmos. Environ. 1995, 29,
1473–1486. [CrossRef]
26. Memon, R.; Leung, D.Y.C. On the heating environment in street canyon. Environ. Fluid Mech. 2011, 11,
465–480. [CrossRef]
27. Tominaga, Y.; Stathopoulos, T. Turbulent Schmidt numbers of CFD analysis with various types of flowfield.
Atmos. Environ. 2007, 41, 8091–8099. [CrossRef]
28. Spalding, D.B. Mixing and chemical reaction in steady confined turbulent flames. Symp. Int. Combust. 1971,
13, 649–657. [CrossRef]
29. Launder, B.E. Heat and Mass Transport; Turbulence: Topics in Applied Physics; Springer: Berlin, Germany, 1978;
Volume 12, pp. 231–287.
30. Kastner-Klein, P.; Plate, E.J. Wind-tunnel study of concentration fields in street canyons. Atmos. Environ.
1999, 33, 3973–3979. [CrossRef]
31. Huang, H.; Akutsu, Y.; Arai, M.; Tamura, M. A two-dimensional air quality model in an urban street canyon:
Evaluation and sensitivity analysis. Atmos. Environ. 2000, 34, 689–698. [CrossRef]
32. Lien, F.S.; Yee, E.; Ji, H.; Keats, A.; Hsieh, K.J. Progress and challenges in the development of physically-based
numerical models for prediction of flow and contaminant dispersion in the urban environment. Int. J. Comp.
Fluid Dyn. 2006, 20, 323–337. [CrossRef]
33. Li, Y.; Stathopolous, T. Numerical evaluation of wind induced dispersion of pollutants around a building. J.
Wind. Eng. Indus. Aerodyn. 1997, 67&68, 757–766. [CrossRef]
34. Wang, X.; McNamara, K.F. Evaluation of CFD simulation using RANS turbulence models for building effects
on pollutant dispersion. Environ. Fluid Mech. 2006, 6, 181–202. [CrossRef]
35. Solazzo, E.; Cai, X.; Vardoulakis, S. Improved parameterization for the numerical modelling of air pollution
within an urban street canyon. Environ. Modell. Softw. 2009, 24, 381–388. [CrossRef]
36. Brzoska, M.A.; Stock, D.; Lamb, B. Determination of plume capture by the building wake. J. Wind. Eng.
Indus. Aerodyn. 1997, 67&68, 909–922. [CrossRef]
Atmosphere 2019, 10, 683 30 of 30
37. Delaunay, D. Numerical simulation of atmospheric dispersion in an urban site: Comparison with field data.
J. Wind. Eng. Indus. Aerodyn. 1996, 64, 221–231. [CrossRef]
38. Baik, J.J.; Kim, J.J.; Fernando, H.J.S. A CFD model for simulating urban flow and dispersion. J. Appl. Meteorol.
2003, 42, 1636–1648. [CrossRef]
39. Kim, J.-J.; Baik, J.-J. Effects of inflow turbulence intensity on flow and pollutant dispersion in an urban street
canyon. J. Wind. Eng. Indus. Aerodyn. 2003, 91, 309–329. [CrossRef]
40. Santiago, J.; Martilli, A.; Martin, F. CFD simulation of airflow over a regular array of cubes. Part I:
Three-dimensional simulation of the flow and validation with wind-tunnel measurements. Bound.-Layer
Meteorol. 2007, 122, 609–634. [CrossRef]
41. Kundu, P.K.; Cohen, I.M. Fluid Mechanics, 3rd ed.; Academic Press: Cambridge, MA, USA, 2005.
42. Rafailidis, S.; Schatzmann, M. Concentration Measurements with Different Roof Patterns in Street Canyon with
Aspect Ratios B/H=1/2 and B/H=1; Universitaet Hamburg, Meterologisches Institute: Hamburg, Germany,
1995.
43. Llaguno-Munitxa, M.; Bou-Zeid, E.; Hultmark, M. The influence of building geometry on street canyon air
flow: Validation of large eddy simulations against wind tunnel experiments. J. Wind Eng. Ind. Aerodyn. 2017,
165, 115–130. [CrossRef]
44. Oke, T.R. Street design and urban Canopy layer climate. Energy Build. 1988, 11, 103–113. [CrossRef]
45. Rafailidis, S. Influence of building area density and roof shape on the wind characteristics above a town.
Bound. Layer Meteorol. 1997, 85, 255–271. [CrossRef]
46. Meroney, R.N.; Pavageau, M.; Rafailidis, S. Study of line source characteristics for 2-D physical modelling of
pollutant dispersion in street canyons. J. Wind End. Ind. Aerodyn. 1996, 62, 37–56. [CrossRef]
47. Moriasi, D.N.; Gitau, M.W.; Daggupati, P. Hydrologic and Water Quality Models: Performance Measures
and Evaluation. Trans. ASBE 2015, 58, 1763–1785.
© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

Atmosphere 10 00683

Uploaded by

Copyright:

Available Formats

Atmosphere 10 00683

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Atmosphere 10 00683

Uploaded by

Copyright:

Available Formats

atmosphere

Atmosphere 2019, 10, 683; doi:10.3390/atmos10110683 www.mdpi.com/journal/atmosphere

Figure 1. A comparison between the original and customized solvers.

𝜕(𝑈 ) ∂k ) ∂k1 𝜕𝑃 ∂ µ𝜕e f f 𝜇 ∂k 𝜕𝑈1 𝜕𝑈

where: 𝜏 = −𝜌𝑢 𝑢 = 2𝜇 𝑆 − 𝜌𝑘𝛿 ; 𝑆 =  +

0.0845 0.7194 0.7194 1.42 1.68 4.377 0.012

numerical results obtained

Value of Sct Authors

3. Initial and Boundary Conditions

where u∗ = τo /ρ is the friction velocity at the nearest wall.

4.1. Validation against the Measurement Data from Li et al.’s Experiment

4.1.2. Street Canyon of Aspect Ratio AR = 1

(a) (b) (c)

Atmosphere 2019, 10, x FOR PEER REVIEW 13 of 30

(a) (b) (c)

(a) (b) (c)

Atmosphere 2019, 10, x FOR PEER REVIEW 15 of 30

(a) (b) (c)

U/Uref W/Uref k/U2ref

(a) Flat roof shape

(b) Slanted roof shape

4.2.1. Flat Roof Shape

4.2.1. Flat Roof Shape

4.2.2.Figure 18b shows

4.2.2. Slanted-Roof Shape

ZH/H = 0.5 ZH/H = 0.33

ZH/H = 0.5 ZH/H = 0.33

4.3. Validation against the Measurement Data from Kastner-Klein’s Experiment

4.3. Validation against the Measurement Data from Kastner-Klein’s Experiment

Figure 22. Schematic diagrams of the street canyon arrangement.

(a) Reference (b) Situation 1

(c) Situation 2 (d) Situation 3

(e) Situation 4 (f) Situation 5

(a). Reference case (b). Situation 1

(c). Situation 2 (d). Situation 3

(e). Situation 4 (f). Situation 5

(a) Reference (b) Situation 1

(c) Situation 2 (d) Situation 3

(e) Situation 4 (f) Situation 5

Atmosphere 2019, 10, 683 25 of 30

Table A1. List of evaluation parameters used in the manuscript.

Evaluation Parameter Equation Author

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

where: 𝜏 = −𝜌𝑢 𝑢 = 2𝜇 𝑆 − 𝜌𝑘𝛿 ; 𝑆 = +