J. Wind Eng. Ind. Aerodyn.

142 (2015) 198–216

Contents lists available at ScienceDirect

Journal of Wind Engineering

and Industrial Aerodynamics
journal homepage: www.elsevier.com/locate/jweia

Consistent inflow turbulence generator for LES evaluation of wind-

induced responses for tall buildings
Haitham Aboshosha a, Ahmed Elshaer b, Girma T. Bitsuamlak b,n, Ashraf El Damatty b
a
Boundary Layer Wind Tunnel Laboratory, University of Western Ontario, London, ON, Canada
b
Department of Civil and Environmental Engineering, WindEEE Research Institute, University of Western Ontario, London, ON, Canada

art ic l e i nf o a b s t r a c t

Article history: A new turbulent inflow generator technique that can be used as inflow boundary condition for LES of
Received 28 December 2014 lower atmospheric boundary flow (ABL) is proposed based on synthesizing random divergent-free tur-
Received in revised form bulent velocities. The proposed technique distinctively maintains both the turbulent spectra and coher-
3 April 2015
ency function similar with the target ABL flow, which is essential for proper simulation of interaction of
Accepted 6 April 2015
turbulent ABL flow with flexible structures prone to wind-induced dynamic excitation. The accuracy of
the proposed technique to produce turbulent velocities with proper spectra and coherency function is
Keywords: assessed in comparison with typical ABL flow characteristics obtained from the literature. Further, its
Inflow turbulence appropriateness to evaluate wind-induced response for tall building is assessed by employing the pro-
Wind spectra
posed technique as inlet boundary condition for LES of the ABL flow around a tall building that was
Coherency
previously tested in a boundary layer wind tunnel. Comparison between dynamic responses (accelera-
Turbulent intensity
Wind load tion) and base load spectra (along-wind, across-wind and torsional) acquired from the proposed LES with
Tall building those from the boundary layer wind tunnel shows a very good agreement. Thus, confirming the suit-
Large eddy simulation ability of the proposed technique for evaluating wind-induced responses. Moreover, the proposed
Atmospheric boundary layer technique is computationally effective and easily amendable for parallel computing making it useful for
generating consistent inflow turbulence while adopting LES for wider wind engineering applications.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction
With encouraging software and hardware technology develop-
ment trends, the cost of conducting large eddy simulation (LES) for
wind engineering applications is becoming computationally afford-
able. This is also reflected through an increased number of pub-
lication that uses LES for variety of wind engineering applications. To
give few examples, recently Dagnew and Bitsuamlak (2014, 2012,
2010), Daniels et al. (2013), applied LES to evaluate wind load on
standard tall buildings. Nozu et al. (2008), Tamura (2010a, b), Huang
and Li (2010), Lim et al. (2009) employed LES to study building
aerodynamics. Bitsuamlak et al. (2010) used LES to assess wind load
on ground mounted solar panel. Aboshosha et al. (2015) used LES to
characterize the turbulence structure of downburst. Abdi and Bit-
suamlak (2014) used LES among other turbulence models to char-
acterize flow over topography. Tominaga and Stathopoulos (2010,
2011) used LES to study the pollution dispersion around a building
and street canyon, respectively, Gousseau et al. (2011) used LES to
n
Corresponding author.
E-mail address: gbitsuam@uwo.ca (G.T. Bitsuamlak).
study pollution dispersion in a city center, and Jiang et al. (2003)
used LES to study natural ventilation.
The importance of defining proper inflow turbulence boundary
condition while using LES was extensively discussed by various
researchers (Sagaut et al. .2003; Tutar and Celik, 2007; Xie and
Castro, 2008; Tominaga et al., 2008; Dagnew and Bitsuamlak, 2013).
The inflow condition should satisfy specific spectra, correlations and
magnitudes. To this end, several techniques are available in the lit-
erature (Kondo et al., 1997; Smirnov et al., 2001; Jarrin et al., 2006;
Tamura, 2000). Keating et al. (2004) classified the techniques used to
generate inflow turbulence for LES into three categories, which are
(i) precursor database, (ii) recycling method and (iii) synthetic tur-
bulence. Liu and Pletcher (2006) provided a review on the precursor
database and recycling method. In the precursor database, simula-
tion of the flow around a targeted zone is conducted in two stages.
First, a parent simulation for the incoming wind, upstream to the
zone of interest, is conducted to produce temporal and spatial dis-
tribution of the incoming turbulent velocities. These turbulent
velocities are saved in a database and used for the second stage,
where the flow simulation is focused on the zone of interest.
Although this method is employed previously in wind engineering
application, it is computationally costly and not preferable unless the
first simulation stage already exists and extensive turbulent velocity

http://dx.doi.org/10.1016/j.jweia.2015.04.004
0167-6105/& 2015 Elsevier Ltd. All rights reserved.
 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216
database is available corresponding to various exposure conditions
(Bitsuamlak and Simiu, 2010). Lund et al. (1998) used the recycling
method to generate inflow velocities for smooth terrains. Nozawa
and Tamura (2002) extended Lund's method and employed it with
rough terrains. Similar to precursor database method, computational
domain is divided into two domains in the recycling method: (i) the
driver domain and (ii) the calculation domain. In the driver domain,
the flow is recycled over a short domain until the flow becomes
statistically stable. Flow characteristics on a mapping plane is stored
and used as the inflow condition for the calculation domain as
illustrated in Fig. 1.
The main drawback of the recycling method is the dependence of
the resulting inflow characteristics on the roughness elements used
at the floor of the driver domain. Unless shape and distribution of the
roughness elements leading to targeted flow characteristics (i.e. ter-
rain exposure) are known, this method cannot be used (Tamura,
2008). Aboshosha (2014) suggested a technique suitable for recycling
method that allows for simulating any targeted terrain exposure
through the usage of fractal surfaces. This technique has been utilized
by Aboshosha et al. (2015) while studying downburst flows for var-
ious terrain exposures. The drawback with all recycling methods is
associated with the high computational cost to run the driver domain
flow simulation compared to other methods such as synthesizing
inflow turbulence (Tamura, 2008). Synthesizing inflow turbulence
does not require costly prior flow simulations, making it a more
robust approach provided that the target flow statistics are met
satisfactorily.
According to Huang et al. (2010), synthesizing inflow turbulence
techniques can be classified into two main groups. The first group
include the work of Hoshiya (1972), Iwatani (1982), Maruyama and
Morikawa (1994), and Kondo et al. (1997). This group uses a weighted
Recycling
Auxiliary simulation Main computation
Mapping
Fig. 1. Recycling technique (Lund et al., 1998).
Table 1
Inflow generation methods for wind engineering applications.
Group/subgroup Study
Precursor database Liu and Pletcher (2006) and Bitsuamlak and Simiu (2010)
Recycling method Lund et al. (1998)
Nozawa and Tamura (2002)
Aboshosha (2014)
Synthetic WAWS Hoshiya (1972), Iwatani (1982), Maruyama and Morikawa
turbulence (1994), and Kondo et al. (1997)
Kondo et al. (1997) and Kim et al. (2013)
RFG Kraichnan (1970), Li et al. (1994), Bechara et al. (1994), Fung Gaussian spectra is used that is not compatible with the spectra in the ABL
et al. (1992), Smirnov et al. (2001), Klein et al. (2003), and
Batten et al. (2004)
Huang et al. (2010) and Castro et al. (2011)
199
amplitude wave superposition method (WAWS) which results in a
turbulent velocity field that satisfies both the targeted power- and
cross-spectra. The drawback of this method is that the resulting tur-
bulent field is not dependent on the computational grid used, thus,
does not satisfy the continuity condition of the flow (i.e. divergence-
free is not guaranteed). This would require enormous effort from the
solver to correct the assigned flow field and enforce the continuity
(Tamura, 2008). Kondo et al. (1997) employed the method originally
developed by Shirani et al. (1981) to make the generated inflow
divergent free. However, the step involved to maintain the diver-
gence-free criterion alters the targeted statistical characteristics. Kim
et al. (2013) suggested to introduce the turbulent field on a vertical
plane near (rather than at) the inlet and relied on the pressure-cor-
rection to maintain the divergence-free criterion. This reduced
degradation of the statistical characteristics compared to introducing
the field right at the inlet. Daniels et al. (2013) employed this method
to estimate peak pressures on a typical tall building and reported that
the method is rapid and led to encouraging results.
The second group includes the work of Kraichnan (1970), Li
et al. (1994), Bechara et al. (1994), Fung et al. (1992), Smirnov et al.
(2001), Klein et al. (2003), and Batten et al. (2004). This group
generates divergent-free velocity field with Gaussian spectra and
is usually referred as random flow generation (RFG) method. This
approach is widely implemented in many commercial CFD soft-
wares. Unfortunately, turbulent spectra encountered in the lower
atmospheric boundary layer (ABL) is different from the Gaussian
spectra (Lumley and Panofsky, 1964), thus making RFG method not
suitable for wind engineering application. Huang et al. (2010)
suggested the discrete random flow generation (DRFG) method to
produce turbulent velocity field that has turbulent spectra close to
the target ABL flow characteristics.
Castro et al. (2011) proposed a modification to the DRFG method
to obtain velocity field that had a better match with the target
spectra. Generally, the DRFG method is able to computationally
effectively generate turbulent spectra that is close to the target,
maintain the spatial velocity correlations, and it can easily be
implemented in parallel computing environment. As a result it
forms the base for current study. Table 1 summarizes the methods
available in the literature to generate the inflow condition.
However, there are other additional important conditions that
need to be satisfied by the generated inflow for wind engineering
applications such as maintaining the proper coherence among the
velocities (Davenport, 1993). This include maintaining proper cor-
relations among the turbulent velocities within different frequencies
Comments
Two steps (i) parent simulation for the incoming wind upstream and, (ii)
main simulation for the targeted zone
Generates inflow for smooth terrains
Generates inflow for rough terrains
Simulated any targeted terrain exposure by using of fractal surfaces
Turbulent field is not dependent on the computational grid, thus, does not
satisfy the continuity condition
Suggested methods to satisfy the divergence-free criterion but affects the
targeted statistical properties, coherency among the velocities is not
maintained
Turbulent spectra that is close to the target, maintain the spatial corre-
lation among the resulting velocities, easily be implemented in parallel
computing environment, coherency among the velocities is not
maintained
200 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216

as indicated by Davenport (1993) and Kijewski and Kareem (1998).
Another important condition is modeling the turbulent spectra to be
exactly similar as the target flow. Unfortunately, these conditions are
not met by the DRFG method, as will be illustrated in the following
sections. The current study focuses on modifying the DRFG method
to maintain proper coherency among the resulting turbulent velo-
cities. The modified method is named consistent DRFG (or CDRFG)
method. In Section 2 a brief discussion on the original DRFG method
as suggested by Huang et al. (2010) is presented. Moreover the
rational that led to the need to improve the spectra and coherency
function of inflow turbulence to better fit the target flow char-
acteristics is highlighted. In Section 3, the proposed modifications to
the DRFG technique that enabled robust modeling of the spectra and
the coherency function are presented. In Section 4, both the new
CDRFG and the original DRFG techniques are applied as LES inflow
boundary conditions to evaluate wind-induced responses of a typi-
cal tall building. The numerical results are then compared with
aerodynamic data obtained from a boundary layer wind tunnel test
to comparatively assess their respective performance.
2. Discrete random inflow generation
As mentioned earlier, Huang et al. (2010) proposed the discrete
random flow generation (DRFG) technique to generate turbulent
velocity field that satisfies the targeted turbulent spectra and spatial
correlations. The technique is based on discretizing power spectra of
velocities into M number of segments and generate wind field within
each of these segments using the original random flow generation
(RFG) technique (Kraichnan, 1970; Smirnov, 2001), but with some
modifications to allow modeling of a spectra with an arbitrary dis-
tribution. According to Huang et al. (2010), turbulent velocity field,
ui(xj,t) can be generated using Eq. (1).
⎡ p m,n p m,n p m,n ⎤ ⎧ m,n ⎫
⎢ x y z ⎥ kx ⎧ ⎫
⎪
⎪ ⎪ ⎪0⎪
⎪
⎢q m,n q m,n q m,n ⎥ ⎨ k m,n ⎬ = ⎨ 0⎬
⎢ x y z ⎥⎪ y ⎪ ⎪ ⎪
⎢⎣k xm,n k ym,n k zm,n ⎥⎦ ⎪ ⎩1 ⎭
⎩k z ⎪
m,n
⎭ (3)
∼ m xj
xj = m
Lj (4)
The parameter L m j in Eq. (4) characterizes the spatial correlations
between the generated velocity fields. Huang et al. (2010) suggested
relating the parameter L m j to the integral length scale of turbulence
CL Luj , where CL is a factor ranging between 1 and 2, with an average
value of 1.5. They compared the spatial correlation of the generated
velocity vectors with the target and found that a value of 1.5Luj led to
a good agreement. It should be mentioned that Huang et al. (2010)
used a frequency independent parameter L m j , which is expected to
result in a frequency independent correlation (i.e. same correlation
for all frequencies). This contradicts with the fact that large eddies
(with low frequencies) have higher correlations than small eddies
(with high frequencies) (Davenport, 1967, 1993). Maintaining proper
frequency-dependent correlations is very important while estimat-
ing wind-induced responses of flexible structures such as tall
buildings and long span bridges (Davenport, 1993). Another dis-
advantage of DRFG technique is that spectra of the resulting tur-
bulent flow deviates from the target ABL flow statistics (Castro et al.,
2011). To explain these limitations in more detail, DRFG technique
(Eqs. (1)–(4)) is used to generate turbulent velocity field for an urban
terrain by using L m j = 1.5 Lui and parameters summarized in Table 2.
These parameters are chosen to match the urban exposure used in
the boundary layer wind tunnel experiments reported by Kijewski
and Kareem (1998) and Zhou et al. (2003). Mean velocity, turbulence
intensity and longitudinal integral scale of turbulence were adopted
from Zhou et al. (2003). The target coherency function (expression
given in Table 2) is adopted from Davenport (1993). Other

M N Table 2
ui (x j , t) = ∑ ∑ pim,n cos (k mj ,n∼x jm + 2πfn, m t) + qim,n Parameters used to generate velocity field for urban terrain exposure.
m= 1 n= 1
Parameter Definition/value
,n ∼
(
sin k m
j
m
x j + 2πfn, m t ) (1)
Exposure Urban
Mean velocity Uav ⎛ z ⎞α
where ui represents longitudinal u, transverse v, and vertical w Uav = Uavref ⎜ ⎟
⎝ zref ⎠
velocities, for i¼ 1, 2, and 3 respectively; j¼1, 2 and 3 represents x-,
Uavref = 10 m/s , zref ¼0.364 m, α¼ 0.326
y- and z-directions, respectively; M is the number of spectral seg-
Turbulent inten- ⎛ z ⎞−dj
ments; N is the number of random frequencies within each seg-
sity I Ij = Irefj ⎜ ⎟
ment; pim,n and qim,n are parameters defined in Eq. (2); fn, m is a ⎝ zref ⎠
normally distributed random number with zero mean and fm where Irefj ¼ 0.208, 0.182, 0.152 and dj ¼0.191, 0.123, 0.005
in the u, v and w directions, respectively (Zhou et al.,
standard deviation; k m j
,n are coordinates of a uniformly distributed
2003; ESDU, 2001)
points on a sphere with a unit radius that satisfy Eq. (3) to maintain von Karman tur- 4(Iu Uav )2 (Lu / Uav )
Su =
the divergence-free condition; ∼
m bulent spectra
x j is a non-dimensional location 5/6
Su, Sv, Sw (
1 + 70.8(fLu / Uav )2 )
coordinate where the velocity is being generated and is defined by
4(Iv Uav )2 (Lv / Uav ) (1 + 188.4(2fLv / Uav )2)
Eq. (4), where xj is the location coordinate in the j direction. Sv =
11/6
(1 + 70.8(2fLv / Uav )2)
2
(rim,n ) 4(Iw Uav )2 (Lw / Uav ) (1 + 188.4(2fLw / Uav )2)
1
pim,n = sign (rim,n ) 2 Sui (fm ) Δf Sw =
11/6
N 1 + (ri m, n)
2
(1 + 70.8(2fLw / Uav )2)
⎛ z ⎞εj
1 1 where Lj = Lrefj ⎜ ⎟ , LrefJ = 0.302, 0.0815 and 0.0326 m,
qim,n = sign (rim,n ) 2 Sui (fm ) Δf ⎝ zrefL ⎠
N 1 + (rim,n )
2
εj = 0.473, 0.881, 1.539, in the x, y, z directions respec-
(2)
tively; zrefL = 0.254 m

where Sui (fm ) is the spectra in the direction i at the frequency fm , Coherency func- ⎛ Cj fm dxj ⎞
tion other Coh (fm ) = exp ⎜ − ⎟ (Davenport, 1993) where Cj is
⎝ Uav ⎠
rim,n is a normally distributed random number with zero mean and parameters coherency decay constant.
unit standard deviation, and Δfm is bandwidth defining the fmmin ¼ 1.0 Hz, fmmax ¼100 Hz, Δf ¼1.0 Hz, M¼ 100, N ¼ 50
spectra segment.
 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216 201

parameters (listed in Table 2) that are required for the inflow gen-
eration are adopted from ESDU (2001).
Fig. 2 shows the coherency function between two velocity vectors
at heights of 0.1 m and 0.3 m from ground. The coherency is com-
pared with the targeted coherency values suggested by Davenport
(1993) (given in Table 2) and generated by using a coherency decay
constant, Cj, of 10 (Davenport, 1993; Kijewski and Kareem, 1998). As
shown in Fig. 2, the coherency produced by adopting the DRFG
technique is frequency independent and fails to capture the decaying
distribution with the frequency increase. This leads, for example, to an
overestimation of the forces acting on structures that has funda-
mental frequency greater than fint, shown in Fig. 2.
Fig. 3 shows comparison of DRFG spectra with von Karman at
point 2 (located at 0.3 m from ground), in the longitudinal, trans-
verse and vertical directions. The same figure also includes the
smoothed spectra of the resulting velocities (i.e. after applying a
moving average) that allows for an easier comparison with the
Coherency(u1, u2)
1 velocities (i.e. having spectra and coherency function that match
0.9
0.8
0.7
Higher correlation
is modeled
0.6
Coherency
target spectra. As indicated in the figure, the resulting spectra from
DRFG do not match the target spectra at low frequencies. Similar
observation was also reported by Castro et al. (2011). Such a dis-
crepancy in the resulting spectra can lead to erroneous wind-
induced structural responses, especially if this discrepancy occurs
close to the natural frequencies of the structure. In the following
section, proposed solutions to address the discrepancies both in the
coherency and the spectra are presented.
3. Consistent discrete random inflow generation (CDRFG)
As illustrated in the previous section, turbulent velocities
generated using DRFG technique show some coherency and
spectra discrepancies compared to the target ABL flow statistics.
Proposed enhancements to DRFG technique are presented in this
section. The proposed solutions to correct the inflow spectra are
presented first, followed by the proposed enhancements for pro-
ducing consistent coherency in the velocity field. The modified
technique is referred as consistent discrete random flow genera-
tion (CDRFG) technique, as it generates consistent turbulent
the ABL flow statistics) as will be shown later in this section.
Target 3.1. Consistent wind spectra
DRFG
According to Huang et al. (2010), turbulent velocity resulting
from DRFG technique corresponding to a frequency fm, ui (xj , t , fm ),
can be generated using Eq. (5), where the frequency fn,m is a ran-
dom frequency with zero mean and fm standard deviation. Fig. 4
illustrates the velocity time traces resulting from Eq. (5) using

fm ¼20 Hz for the urban exposure parameters summarized in

0.5 Table 2.
N
,n ∼
0.4 ui (x j , t , fm ) = ∑ pim,n cos k m
j ( m
x j + 2πfn, m t + qim,n )
n= 1
,n ∼
0.3
(
sin k m
j
m
x j + 2πfn, m t ) (5)

As shown in Fig. 4, the resulting spectra have multiple peaks in

0.2 the frequency band ranging approximately between 0 and 3fm.
This means that the DRFG technique distributes the energy spectra
0.1 fint for the frequency fm over a band of frequencies 0–3fm, as opposed
to focusing the energy close to fm. This is believed to be the main
reason for the spectral discrepancy shown in Fig. 3.
0 In order to correct the discrepancy in the resulting spectra, it is
0 5 10 15 20 25
suggested to use random frequencies fn,m that are more focused
f(hz)
near the frequency fm. Thus, random frequencies fn,m are chosen to
Fig. 2. DRFG coherency between velocities at two different heights (points 1 and 2). have a mean value of fm and a standard deviation of Δf , where Δf

0 0
Spectra Su, Sv, Sw 0
10 10 10

discrepancy DRFG
DRFG (smoothed)
-1 -1 -1 von Karman
Su (m2s2/hz)

10 10 10
S (m2s2/hz)
Sv (m2s2/hz)

-2 -2 -2
10 DRFG 10 DRFG 10
DRFG (smoothed) DRFG (smoothed)
von Karman von Karman discrepancy
-3 -3 -3
10 10 10
0 1 2 0 1 2 0 1 2
10 10 10 10 10 10 10 10 10
f (hz) f (hz) f (hz)

Fig. 3. Sample DRFG velocity spectra.

202 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216

Velocities u ,v ,w (m/s)
fm fm fm
5
(m/s)

0
fm
u

-5
0 1 2 3 4 5 6
5
(m/s)

0
fm
v

-5
0 1 2 3 4 5 6
5
(m/s)

0
fm
w

-5
0 1 2 3 4 5 6
time (s)

0 0
Spectra Su, Sv, Sw 0
10 10 10

DRFG (f =20 hz)

m
von Karman
Su (m2s2/hz)

-1 -1 -1
10 10 10

S (m2s2/hz)
Sv (m2s2/hz)

DRFG (f =20 hz)

DRFG (f =20 hz)

w
-2 -2 m -2
10 m 10 10
von Karman
von Karman
f=f f=f
f=f m
m
m
-3 -3 -3
10 10 10
0 1 2 0 1 2 0 1 2
10 10 10 10 10 10 10 10 10
f (hz) f (hz) f (hz)

Fig. 4. DRFG velocity time trace using a single fm of 20 Hz and their spectral plots.

Velocities u ,v ,w (m/s)
fm fm fm
1
(m/s)

0
fm
u

-1
0 1 2 3 4 5 6
1
(m/s)

0
fm
v

-1
0 1 2 3 4 5 6
1
(m/s)

0
fm
w

-1
0 1 2 3 4 5 6
time (s)

0 0 Spectra Su, Sv, Sw 0

10 10 10
CDRFG (f =20 hz) CDRFG (f =20 hz)
m
m
von Karman von Karman
-1 -1 -1
10 10 10
Sv (m2s2/hz)
Su (m2s2/hz)

S (m2s2/hz)

CDRFG (f =20 hz)

m
-2 -2 -2
w

10 von Karman 10 10
f=f f=f
m m

f=f
m
-3 -3 -3
10 10 10
0 1 2 0 1 2 0 1 2
10 10 10 10 10 10 10 10 10
f (hz) f (hz) f (hz)

Fig. 5. CDRFG velocity time trace and their spectral plots using one fm ¼ 20 Hz.

is frequency step used to represent the target spectra. The mag- expressions for fn,m, pim,n and qim,n expressions, and employing
nitude of the factors pim,n and qim,n is halved according to Eq. (6) in Δf = 1.0 Hz, are shown in Fig. 5. As shown in Fig. 5, the resulting
order to compensate for the new utilized values of frequencies fn,m. spectra are more focused around the frequency fn,m and closer to
The resulting velocity and spectra obtained using the updated the targeted value.
 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216 203

0 0
Spectra Su, Sv, Sw 0
10 10 10
CDRFG
CDRFG (smoothed)
-1 -1 -1 von Karman
10 10 10

S (m 2 s 2 /hz)
S v (m 2 s 2 /hz)
S u (m 2 s 2 /hz)

-2

w
-2 -2
10 CDRFG 10 CDRFG 10
CDRFG (smoothed) CDRFG (smoothed)
von Karman von Karman

-3 -3 -3
10 10 10
0 1 2 0 1 2 0 1 2
10 10 10 10 10 10 10 10 10
f (hz) f (hz) f (hz)

Fig. 6. Sample CDRFG velocity spectral plots.

2 3.2.1. Step 1: coherence resulting from CDRFG

1 m (rim,n ) Coherency function based on the new definition of Ljm (Eq. (7)) can
pim,n = sign ( rim,n ) Sui Δf q m,n
N 2 i
1 + (rim,n ) be calculated as the cross-correlation between velocities generated by
Eq. (5) close to the frequency fm. Derivation for coherency function is
1 m 1
= sign (rim,n ) Sui Δf given in Eq. (8).
N 1 + (rim,n )
2
(6)
σu1u2
The new expressions for and fn,m, pim,n,
are used with Eq. (1) qim,n Cohu1, u2 (fm ) = u1 (fm )
σu1 σu2
to generate turbulent velocities that produce the entire turbulent
N
spectra range. The spectra is shown in Fig. 6 for a point located at
0.3 m from the ground. The comparison of the spectra generated
= ∑ pim1 ,n cos (k mj ,n k
x1mj + 2πfn, m t) + qim
1
,n

n= 1
using the new expressions for fn,m, pim,n and qim,n with von Karman
,n k
spectra reveals an improved agreement as shown in Fig. 6. (
sin k m
j x1mj + 2πfn, m t u2 (fm ) )
N

3.2. Correction for the coherency function

= ∑ pim2 ,n cos (k mj ,n k
x2mj + 2πfn, m t) + qim
2
,n

n= 1
,n k
As discussed earlier, the DRFG technique leads to unrealistic (
sin k m
j x2mj + 2πfn, m t σu1u2 )
coherency function that is frequency independent. To address this T
N
1
shortcoming, it is proposed to relate the parameter L m j , which =
T
∫0 ∑ (pim1 ,n pim2 ,n ) cos (k mj ,n k
x1mj + 2πfn, m t) cos
characterizes the correlations to the frequency, fm, in accordance n= 1
,n k
with Eq. (7). (km
j x2mj + 2πfn, m t )
Uav N
1 T
Lm
j = + ∫0 ∑ (qim1 ,n qim2 ,n )
γCj fm (7) T n= 1
,n k
where Uav is the mean velocity, fm is the frequency at segment m, γ is a sin ( km
j
,n x m
1j + 2πfn, m t sin k m
j ) (x2mj + 2πfn, m t σu1u2 )
tuning factor, Cj is the coherency decay constant, and j¼ 1, 2, and N
( pim ,n m,n
1 pi2 ) ,n k
3 represents longitudinal, transverse and vertical directions, respectively. = ∑ (
cos k m
j x1mj − k
( x2mj ))
The expression given by Eq. (7) requires the tuning factor γ to be n= 1 2
defined. This tuning factor is estimated from the non-dimensional N
(qim1 ,n qim2 ,n ) ,n k
length scale, β = CD/Lu (z), where Lu(z) is the longitudinal length + ∑ cos k m
j ( x1mj − k
(
x2mj ))σu u
1 2
scale of turbulence, D is a characteristic distance chosen to tune the n= 1 2
N
correlations, and C is the coherency decay constant. The character-
,n k
x1mj − k
istic distance D is function of the problem being solved. Estimating = Su1 (fm ) Su2 (fm ) Δf ∑
n= 1
cos k m
j ( ( x2mj ))σu 1

the tuning factor γ from the non-dimensional length scale β is con-

ducted in three steps. In the first step, an expression for coherency = Su1 (fm ) Δf , σu2
function resulting from the DRFG (Eq. (1)) technique using the new = Su2 (fm ) Δf Cohu1, u2 (fm )
definition for L m
j (Eq. (7)) is obtained. The coherence is a function of N
,n k
x1mj − k
the tuning factor γ . In the second step, target coherency function
suggested by Davenport (1993) (see Table 2) is fitted with the
= ∑
n= 1
cos k m
j ( x2mj ( ))Cohu ,u 1 2 (fm )

coherency function obtained from the first step using γ as the fitting N

parameter. It is observed that depending on the area under the = ∑ cos γk m

j (
,n C
f ) (8)
coherency curve (i.e. correlation in the wide frequency band, Ru1u2n), n= 1

different tuning factor γ values are obtained. This leads to a rela- where Cf ¼Cfd/Uav, C is coherency decay constant, and d is the
tionship between γ and Ru1u2n. In the third step, an expression for distance between points 1 and 2.
Ru1u2n is obtained as a function of the non-dimensional length
scale β, which is used to evaluate the tuning factor γ . All the 3.2.2. Step 2: relationship between γ and Ru1u2*
mathematical expressions employed at each steps to relate γ and β As indicated from Eq. (8), the coherency function from CDRFG
are given below. technique is dependent on tuning factor γ and non-dimensional
204 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216

frequency Cf. This coherency function needs to be equal to the tar- f max
∫0 Coh (f ) df
geted coherency (given in Table 2). By fitting the resulting coherency Ru1u2 ⁎ =
f max
function (Eq. (8)) with targeted coherency function (given in ∫0 df (9)
Table 2), factor γ is obtained as the fitting parameter. Depending on
the area under the coherency curve (i.e. cross-correlation in the Fig. 7(d) shows the relationship between γ and Ru1u2n. This rela-
wide frequency band, Ru1u2n), different values of γ are obtained as tionship allows for estimating γ provided that Ru1u2n is known. In the
shown in Fig. 7(a–c). The cross-correlation in the wide frequency next step, Ru1u2n is related to the non-dimensional length scale, β ,
band, Ru1u2n, can be expressed by Eq. (9) and is shown in Fig. 7(a–c). and then a relationship between γ and β is obtained.

1 1 1

0.8 0.8 0.8

Coherency

Coherency
Coherency
0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
0 0.2 0.4 0.6 0.8 0 0.5 1 1.5 0 1 2 3
C =c.f.D/U C =c.f.D/U C =c.f.D/U
f av f av f av

n,m *
Σ (cos( γ Kj Cf)) exp(-C f) Ru1u2

γ 4 γ =3.6
γ =2.1 γ =2.1
2

0
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2
R*u1u2

Fig. 7. Fitting process for coherency function resulting from CDRFG technique for different Ru1u2n values (a) to (c), and (d) relationship between Ru1u2n and γ.

40

30
β

20

10

0
1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
Ru1u2

5
Resulting relationship
4 12
Expression in Equation 10
γ

1
20 18 16 14 12 10 8 6 4 2 0
β
Fig. 8. Relationship between Ru1u2n, β and γ.
 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216 205

3.2.3. Step 3: relationship between γ and β As shown in Fig. 7(d), Ru1u2n is also a function of γ . By equating
Cross-correlation between velocities u1 and u2, Ru1u2n, is calcu- Ru1u2n from Figs. 7(d) and 8(a), a relationship between γ and β is
lated as the ratio between velocity covariance σu⁎1u2 and rms velo- obtained, as shown in Fig. 8(b). This relationship can be expressed
cities σu⁎1 and σu⁎2 , as expressed by Eq. (10). By using von Karman by Eq. (11), which is also plotted in Fig. 8(b).
spectra to model the distribution of the turbulent energy, Ru1u2n is ⎧ 3.7β −3 β < 6.0⎫
⎪ ⎪
obtained as a function of β = CD/Lu and plotted in Fig. 8(a). γ=⎨
⎪
⎬
⎪
⎩ 2.1 β ≥ 6.0⎭ (11)

cfD The flowchart shown in Fig. 9 summarizes all the steps

∞ −
σu⁎1u2 ∫0 Su1 Su2 e Uav df involved in the CDRFG technique. A MATLAB code is developed to
⁎
Ru1u2 = = Su1
σu⁎1 σu⁎2 ∞
∫0 Su1 df ∫0 Su2 df
∞
conduct the velocity turbulent generation using CDRFG and is
provided in Appendix A. Fig. 9 shows that the user needs to choose
σu⁎12 4(Lu /Uav ) σu⁎22 4(Lu /Uav )
= Su2 = n the distance D to tune the correlations. This distance shall be
5/6 5/6
(1 + 70.8n2) (1 + 70.8n2) related to the problem being solved. For instance, for tall building
∞ 4 D shall be taken in the order of 0.5–1.0h to maintain the proper
= fLu /Uav R u⁎1u2 = ∫0 5/6
e −βndn = f (β)
correlation along the building height, where h represents building
(1 + 70.8n2) (10) height. It is worth mentioning that values of D resulting with

Fig. 9. CDRFG technique flow chart.

206 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216

Coherency(u1, u2) β ¼CD/Lu values larger than 6 would result in a tuning factor γ
1 independent of D, as indicated in Fig. 8(b).
Target The accuracy of the CDRFG technique described in Fig. 9 to
Current model the proper coherency function is assessed by generating
0.9
velocity field for the urban boundary layer with the parameters
summarized in Table 2. A value of the distance D equal to 0.2 m
0.8 (β ¼ 6.7) is chosen to tune the correlation. The coherency of velo-
cities at point 1 (at 0.1 m from the ground) and point 2 (at 0.3 m
0.7 from the ground) are plotted in Fig. 10 and compared with the
targeted coherency (Eq. (5)). Fig. 11 shows coherency functions
between velocities with separation distances, d¼ 0.1 and 0.3 m,
0.6 respectively. As indicated from these figures, it is fair to conclude
Coherency

that CDRFG technique is able to maintain the proper coherency

0.5 among resulting turbulent velocities.
In the next section, CDRFG technique is employed to evaluate
the dynamic response of a tall building that was previously tested
0.4
in a boundary layer wind tunnel (Kijewski and Kareem, 1998; Zhou
et al., 2003). Its efficacy is examined through comparison of the
0.3 numerical aerodynamic data with those obtained from boundary
layer wind tunnel.
0.2
Table 3
Properties of the examined building.
0.1
Property Value

0 Height Hs, width Ws, depth Ds 182.2, 30.48, 30.48 m

0 5 10 15 20 25 Natural frequency 0.15 (along-wind), 0.15 (across-wind), 0.3
f(hz) (torsional)
Damping ratio 1% For all modes
Fig. 10. CDRFG velocity coherency between different points. Mass per unit volume ms 192 kg/m3
Air density 1.25 kg/m3

Coherency(u , u ) d=0.1 m Coherency(u , u ) d=0.3 m

1 2 1 2
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6
Coherency

Coherency

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0 5 10 15 20 25 0 5 10 15 20 25
f(hz) f(hz)

Target Current
Fig. 11. Target and CDRFG coherency comparison for different separation distances.
 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216 207

4. Application of CDRFG to evaluate wind load on a tall
building
4.1. Numerical model description
LES of wind flow around a tall building immersed in an urban
boundary layer is conducted to examine efficiency of the developed
technique. Properties of the urban boundary layer flow and the
building are summarized in Tables 2 and 3, respectively. Inflow field
generated by using both the CDRFG and DRFG techniques are
employed to comparatively test their applicability to evaluate the
building dynamic response. Further, the building's dynamic responses
using the two inflow techniques are compared with those obtained
from the boundary layer wind tunnel experiment reported in Kijewski
and Kareem (1998) and Zhou et al. (2003). The simulations are con-
ducted using a length scale of 1:500 and a reference velocity at the
building height equal to 10 m/s.
Fig. 12 shows the study building model and computational domain
dimensions, and the adopted boundary conditions, which follows the
recommendation by Franke (2006) and COST (2007). In the model,
x-axis represents longitudinal (i.e. the main flow) direction, while
y- and z-axes represent transverse and vertical directions, respectively.
Table 4 summarizes the employed dimensions compared with those
recommended by COST (2007) and Architectural Institute of Japan (AIJ)
recommendations (Tamura et al., 2008). Commercial CFD package
(STAR-CCMþ solver) is utilized to solve the LES represented by Eq. (13).
Dynamic sub-grid scale model by Smagorinsky (1963) and Germano
et al. (1991) is adopted. Parameters used to handle flow quantities as
well as the solution technique are summarized in Table 5. Inflow tur-
bulence velocity fields generated by DRFG and CDRFG techniques are
introduced into STAR CCM using space and time-dependent table
option (x, y, z, t) at the inflow boundary of the computational domain.
where i ¼1, 2, 3 correspond to the x-, y- and z-directions, respec-
tively, The over bar represents the filtered quantities, ui, p, t, τij and
ν represent fluid velocity, pressure, time, the SGS Reynolds stress
and molecular viscosity coefficient, respectively. Sij, νe , Δ, Cs
represent strain rate tensor, eddy viscosity, grid size, Smagorinsky
constant which is determined instantaneously based on the
dynamic model (Germano et al., 1991), respectively. δij represents
Kronecker delta.
The computational domain is discretized using polyhedral mesh
option available in Star CCMþ. Two grids sizes G1 and G2 are
employed to study the grid independency of the results. For both
grids, fine meshes are used near the building faces, the wake zone,
and the zone between the inflow and the building. Distribution of
the mesh size within computational domain is divided into three
zones as illustrated in Fig. 13 and summarized in Table 6. Fig. 14
shows details of the employed grids. COST (2007) and Tominaga
et al. (2008) suggested that the stretching ratio of the grids in
regions of high velocity gradients should be less than 1.3. The use of
a high stretching ratio with LES can cause numerical divergence due
to the sudden differences in the cut-off wave number of the energy
spectrum between resolved and sub-grid modeled scales. In the
current study, a number of 10 prism layers with 1.05 stretching is
Table 4
Computational domain dimensions.
Parameter Current Cost (2007) AIJ (2008)
X1 5H (30B) 5H 36B
X2 15H (90B) 15H
Y 10H (60B) 4.6H 21.6B
Z 4H (30B) 4H 40B

∂u¯i
=0
∂xi
∂u¯i ∂u¯i 1 ∂P¯ ∂
+ u¯j =− + ( − τij + 2νS¯ij ) Table 5
∂t ∂x j ρ ∂xi ∂x j Parameters used in the LES.
τij = ui ¯uj − u¯i u¯j
Parameter Type
1 ⎛ ∂u¯i ∂u¯j ⎞
S¯ij = ⎜⎜ + ⎟⎟ Time discretization Second order implicit
2 ⎝ ∂x j ∂xi ⎠ Momentum discretization Bounded central difference
1 Pressure discretization Second order
τij − δij τkk = 2νe S¯ij Pressure–velocity coupling Coupled
3 Under relaxation factors A value of 0.7 for the momentum and 0.7 for the
2
νe = (Cs Δ)2 (2S¯ij S¯ij ) (12)
pressure

Fig. 12. Boundary conditions and computational domain.

208 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216
Fig. 13. Dimensions of different mesh zones.
Table 6
Properties of the employed grids
Grid Grid 1 (G1)
Zone 1 Zone 2
Grid size H/10 H/50
Total number of grids 990,000
utilized for both grids as indicated in Fig. 14, following the recom-
mendation by Murakami (1998).
Time step in the LES is chosen to be equal to 0.0002 s to
maintain Courant–Friedrichs–Lewy (CFL) number less than 1.0. A
number of 30,000 time steps are resolved which represents a 6 s.
The 6 s at the model scale is equivalent to 750 s at full-scale using
a reference velocity of 10 at the building height or 3000 s at full-
scale using a reference velocity of 40 m/s at the building height.
The SharcNet high performance computer (HPC) facility at the
University of Western Ontario has been used to conduct the
simulations, which employed 128 cores for each grid. Simulation
on grid G1 took 28 h and on grid G2 took around 19 h. Results for
the last 24,000 time steps are employed in calculating flow sta-
tistics and building responses.
4.2. Velocity field
Fig. 15 illustrates the instantaneous quasi-streamlines super-
imposed on the velocity field on a vertical section (passing
through mid-building width) and on a horizontal section (passing
through mid-building height) resulting from the CDRFG employ-
ing G1. The quasi-streamlines are generated in 2D plane assuming
zero velocities in the perpendicular direction to that plane. As
indicated in Fig. 15, instantaneous field depicts clearly large and
small scale turbulent structures at the inflow and near the building
walls. Fig. 16 shows instantaneous surfaces of equal vorticity
magnitudes where turbulent structure including various shear
layers and horseshoe vortex at the upstream side near the ground
is captured by the numerical simulation.
4.3. Building responses
In the current study, dynamic building responses are calculated
using wind-induced base moments, following the procedure for
force balance tests in the boundary layer wind tunnel (Warsido
and Bitsuamlak, 2013). Fig. 17 shows time traces of the base
moments around x-axis (due to across-wind force), y-axis (due to
along-wind force), and z-axis (torsional) obtained from LES using
the CDRFG technique and employing grid G1. The base moments
are normalized using reference base moments defined by Eq. (13).
1 1 1 the responses predicted using LES employing DRFG are in a good
Myref = 2 ρV h2 BH 2, Mxref = 2 ρV h2 DH 2, Mxref = 2 ρV h2 DBH
Grid 2 (G2)
Zone 3 Zone 1 Zone 2 Zone 3
H/90 H/10 H/36 H/60
670,000
where Vh is the mean velocity at the building height and ρ is the air
density which is taken equal to 1.25 kg/m3.
Power spectral density (PSD) for the three base moments that
illustrates the energy distribution with the frequencies are plot-
ted in Fig. 18. This figure shows PSD resulting from LES employing
CDRFG, LES employing DRFG, and the boundary layer wind tun-
nel test by Zhou et al. (2003). As can be seen from Fig. 18, PSD
from the LES using the CDRFG technique matches very well with
the boundary layer wind tunnel in the along-wind, across-wind
and torsional directions. Although PSD for the across-wind
moment resulting from LES employing DRFG technique is in a
good agreement with the boundary layer wind tunnel, PSD for
other moment directions (i.e. along-wind and torsional) deviates
from the boundary layer wind tunnel results. The main reason
behind these differences is attributed to the level of the coher-
ency function modeling consideration. As indicated in Figs. 2, 10
and 11, CDRFG well maintains the coherency function similar to
the target while it is not the case when DRFG is used. This leads
to unrealistically correlated fluctuations of pressure. These
unrealistic fluctuations act directly on the windward face of the
building which affects primarily the along-wind and torsional
base moments. The across-wind base moment is affected less as
it is primarily dominated by the wind/building interactions.
Dynamic responses of the building are evaluated using the base
moments' spectra shown in Fig. 18. Analysis of peak displacement
and acceleration at the top floor, and equivalent static base
moments are conducted following the method described by
Kijewski and Kareem (1998) and Chen and Kareem (2005). The
expressions utilized in the analysis are provided in Appendix B. The
analysis is conducted to cover a velocity range from 8 to 40 m/s at
the building height.
Figs. 19 and 20 show plots of top floor peak displacement and
accelerations, respectively. Fig. 21 shows plots of the peak
equivalent static moment at the base. In general, similar to what is
observed from Fig. 18, Figs. 19–21 show that the responses pre-
dicted using LES employing CDRFG technique are in a very good
agreement with those from the boundary layer wind tunnel. It is
to be noted that CDRFG using G1 and G2 produced comparable top
floor acceleration and base moment spectra attesting the grid
independency of the solution. Also the same figures also show that
(13) agreement for the across-wind responses, but the agreement
 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216 209

Grid G1 Grid G2

3D view

Sectional views at zones 2 and 3

Prism
layers Prism
layers

Sectional views at zones 3 close to the building

Fig. 14. Comparison between grids G1 and G2.

Vertical sectional view Plan sectional view at mid-height

Vertical sectional view at mid-width Plan sectional view at building

mid-height

Fig. 15. Flow field: Instantaneous velocity magnitude contour and quasi-streamlines.
210 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216

Shear layer

Wind direction

Inward wind
direction

Side view Front view

Shear layer

Wind direction

Wind Horse shoe vortex

direction
Horse shoe vortex

Plan view Isometric view

Fig. 16. Surfaces of equal vorticity magnitude.

Across Wind
0.4
0.2
Mx/Mxref

-0.2

-0.4
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)

Along Wind
1
0.8
My/Myref

0.6
0.4

0.2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)

Torsional
0.1
0.05
Mtor/Mtorref

0
-0.05

-0.1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)

Fig. 17. Base moment plots around the x-axis (across-wind), y-axis (along-wind) and z-axis (torsional) obtained from LES using the CDRFG technique.
 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216 211

Along Wind Moment My Across Wind Moment Mx Torsional Moment Mt

0
10 0
10

-1
10
-1
10 10
-1
SMy.f/σ 2My

Mx

SMt.f/σ2Mt
S .f/ σ2
Mx
-2
-2
-2
10
10 10

-3 -3
10 10 -3
-2 -1 -2 -1 10
10 10 10 10 -2 -1
10 10
f.B/Vh f.B/Vh f.B/V
h

NatHaz Wind Tunnel CDRFG Huang et al. (2010)

Fig. 18. Spectra of the base moments.

Along Wind Across Wind Torsional

2.5 2 0.2

1.8 0.18

2 1.6 0.16

1.4 0.14
Rot. disp. x Radius (rad)

1.5 1.2 0.12

Disp. (m)

Disp. (m)

1 0.1

1 0.8 0.08

0.6 0.06

0.5 0.4 0.04

0.2 0.02

0 0 0
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40
Vh (m/s) V (m/s) (m/s)
h
Vh
NatHaz Wind Tunnel Current G1 Current G2 Huang et al. 2010

Fig. 19. Top floor peak displacements.

deteriorates for the along-wind and torsional responses that are
dominantly affected by incoming turbulence. This illustrate the
advantage of the new CDRFG technique proposed in the current
study to analyze wind-induced responses of tall buildings and
other wind-sensitive structures.
5. Conclusions
The current study presented a literature review on inflow tur-
bulence generation approaches for LES that highlights the unique
advantages and some limitations of the discrete random flow
generation (DRFG) technique by Huang et al. (2010). Two mod-
ifications have been proposed to the DRFG technique to accurately
model the target spectra and the coherency function. The modified
technique is called consistent discrete random flow generation
(CDRFG) technique, owing to its consistent spectra and coherency
reproduction. Accuracy of the technique in generating proper
coherence and spectra is assessed in comparison with target ABL
flow statistics obtained from literature. This is followed by
assessment of the technique's applicability to evaluate wind-
induced responses of a tall building by comparing base moments
212 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216

Along Wind Across Wind Torsional

160 180 70

140 160
60

140
120
50

Rot. Acc x Radius (milli-G)

120
100
Acc (milli-G)

Acc (milli-G)
100 40
80
80 30
60
60
20
40
40

20 10
20

0 0 0
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40
Vh (m/s) V (m/s) V (m/s)
h h

NatHaz Wind Tunnel Current G1 Current G2 Huang et al. 2010

Fig. 20. Top floor peak accelerations.

Along Wind Across Wind Torsional

4000 3500 80

3500 70
3000

3000 60
2500

2500 50
Moment (MN.m)

Moment (MN.m)

Moment (MN.m)

2000
2000 40
1500
1500 30

1000
1000 20

500 500
10

0 0 0
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40
V (m/s) V (m/s) V (m/s)
h h h

NatHaz Wind Tunnel Current G1 Current G2 Huang et al. 2010

Fig. 21. Peak base moments.

and top floor acceleration with aerodynamic data collected from a
boundary layer wind tunnel. The results indicate that LES
employing CDRFG produces building responses that are in a very
good agreement with those obtained from the boundary layer
wind tunnel. The results also indicate that CDRFG produced more
accurate responses compared to the original DRFG, particularly in
the along-wind and torsional directions. The CDRFG technique is
also amendable for parallel implementation and robust compared
with other methods of generating inflows for LES, thus, it is
expected to be widely used for wind engineering applications
employing LES.
Acknowledgments
The authors would like to acknowledge the financial support from
the National Research Council of Canada (NSERC), the Ontario Center
of Excellence (OCE) and Canada Research Chair (for the third author).
The authors are grateful for the SharcNet for providing access to their
high performance computation facility and excellent support from
their technical staff. Findings and opinions expressed in this paper are
those of the authors alone, and do not necessarily reflect the views of
the sponsoring agency.
 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216
Appendix A. : MATLAB code for the CDRFG technique
% Example on using the CDRFG_2015 Function
h0u ¼ 0.3644;alphau ¼0.3264;Uh ¼10.0;
h0I ¼0.3364;
Iuh ¼0.2084;Ivh ¼ 0.1815;Iwh¼ 0.1523;
dIu ¼ -0.1914;dIv ¼ -0.1228;dIw ¼-0.0048;
h0L ¼0.254;
Luh ¼0.302;Lvh ¼ 0.0815;Lwh ¼0.0326;
dLu ¼0.473;dLv¼ 0.8813;dLw ¼ 1.5390;
Cxyz ¼[10 10 10];DGamma ¼ 0.3;
nf ¼100;nm ¼50;fmax ¼ 100;
dt ¼1/fmax/2/2.5;nt ¼1000;
M ¼[zeros(10,1) zeros(10,1) (0.05:0.1:1)’]; % Sample
coordinate matrix
CDRFG_2015(h0u,alphau,Uh,h0I,Iuh,Ivh,Iwh,dIu,dIv,dIw,
h0L,Luh,Lvh,Iwh,…
dLu,dLv,dLw,Cxyz,DGamma,nf,nm,fmax,dt,nt,M)
function CDRFG_2015(h0u,alphau,Uh,h0I,Iuh,Ivh,Iwh,
dIu,dIv,dIw,h0L,…
Luh,Lvh,Lwh,dLu,dLv,dLw,Cxyz,DGamma,nf,nm,fmax,dt,
nt,M)
% Consistent DRFG Function By Aboshosha et al. (2015)
% INPUT Parameters
% h0u Reference height for the mean velocity
% Uh Mean velocity at h0u
% alphau Power low exponent of the mean velocity
% h0I Reference height for the turbulent intensity
% Iuh Longitudinal turbulent intensity at h0I
% Ivh Transverse turbulent intensity at h0I
% Iwh Vertical turbulent intensity at h0I
% dIu Power low exponent of the longitudinal turbulent
intensity
% dIv Power low exponent of the longitudinal turbulent
intensity
% dIw Power low exponent of the longitudinal turbulent
intensity
% h0L Reference height for the length scale
% Luh Longitudinal length scale at h0L
% Lvh Transverse length scale at h0L
% Lwh Vertical length scale at h0L
% dLu Power low exponent of the longitudinal length
scale
% dLv Power low exponent of the longitudinal length
scale
% dLw Power low exponent of the longitudinal length
scale
% Cxyz Coherency decay constants in x, y and z direc-
tions [1  3] matrix
% DGamma Characteristic length used to maintain the
coherency
% nf Number of random frequencies in one segment
% nm Number of frequency segments
% fmax Maximum frequency
% dt Time step
% n Number of time steps
% M Matrix of the inflow coordinates [x y z]
% OUTPUT Parameters
% Three files (i.e. inletdata_U, inletdata_V and inletda-
ta_W) that have the
% generated velocity records compatible with STAR
CCM þ
213
for i¼ 1:nd
for j¼1:nm
fmjs¼fm(j) þ fms;if j¼ ¼1;fmjs ¼fm(j) þfms(nfsm:nfs);end
Su(j,i) ¼mean(4n(Iu(i)nUav(i))^2n(Lu(i)/Uav(i))./(1 þ70.8n(fmjsnLu(i)/…
Uav(i)).^2).^(5/6));
Sv(j,i) ¼mean(4n(Iv(i)nUav(i))^2n(Lv(i)/Uav(i))n(1 þ188.4n(2nfmjsnLv(i)/…
Uav(i)).^2)./(1 þ70.8n(2nfmjsnLv(i)/Uav(i)).^2).^(11/6));
Sw(j,i) ¼mean(4n(Iw(i)nUav(i))^2n(Lw(i)/Uav(i))n(1 þ188.4n(2nfmjsn…
Lw(i)/Uav(i)).^2)./(1 þ70.8n(2nfmjsnLw(i)/Uav(i)).^2).^(11/6));
end
end
UavLs ¼mean(Uav); % mean long. velocity used to identify the L scale
%% Generate of P,Q,K Matrices
K¼zeros(nf,3,nm);
r¼randn(nf,3,nm);
P¼r./abs(r).nsqrt(1/nfn(r).^2./(1þr.^2));
Q¼ r./abs(r).nsqrt(1/nfn(1).^2./(1 þr.^2));
Ls¼ zeros(nm,3,nd);
for nmi ¼1:nm
for nyi¼ 1:nd;
Beta ¼10nDGamma/Lu(nyi);if Beta 46;Gammai ¼2.1;
else Gammai ¼3.7nBeta^-.3;end
Ls(nmi,:,nyi) ¼Uav(nyi)/fm(nmi)./Cxyz/Gammai;
end
K(:,:,nmi) ¼ RandSampleSphere(nf);
for i¼ 1:nf
XX ¼K(i,:,nmi)’;
myfun ¼@(xx) mapp(xx,P(i,:,nmi),Q(i,:,nmi));
K(i,:,nmi) ¼(fsolve(myfun,XX))’;
end
end
%% Generate the Velocity Vectors
U¼ zeros(nd,nt);V ¼zeros(nd,nt);W ¼zeros(nd,nt);
for inxyi¼ 1:nd
for nmi ¼1:nm;
xjbar¼ 1./Ls(nmi,:,inxyi).n[X(inxyi) Y(inxyi) Z(inxyi)];
kjxj ¼(xjbar(1)nK(:,1,nmi) þxjbar(2)nK(:,2,nmi) þ xjbar(3)nK(:,3,nmi));
U(inxyi,:) ¼ U(inxyi,:) þsum(sqrt(Su(nmi,inxyi)ndfn2)n(P(:,1,nmi)n… ones(1,
nt)).ncos(wn(:,nmi)nttþ kjxjnones(1,nt)) þsqrt(Su(nmi,inxyi)ndfn2)…
n(Q(:,1,nmi)nones(1,nt)).nsin(wn(:,nmi)ntt þkjxjnones(1,nt)));
V(inxyi,:) ¼V(inxyi,:) þsum(sqrt(Sv(nmi,inxyi)ndfn2)n(P(:,2,nmi)n…
ones(1,nt)).ncos(wn(:,nmi)nttþ kjxjnones(1,nt)) þsqrt(Sv(nmi,inxyi)n…
dfn2)n(Q(:,2,nmi)nones(1,nt)).nsin(wn(:,nmi)ntt þkjxjnones(1,nt)));
W(inxyi,:) ¼W(inxyi,:) þsum(sqrt(Sw(nmi,inxyi)ndfn2)n(P(:,3,nmi)n… ones
(1,nt)).ncos(wn(:,nmi)ntt þkjxjnones(1,nt)) þ sqrt(Sw(nmi,inxyi)ndfn2)…
n(Q(:,3,nmi)nones(1,nt)).nsin(wn(:,nmi)nttþ kjxjnones(1,nt)));
end
U(inxyi,:) ¼ U(inxyi,:) þUav(inxyi); % Add the mean velocity
214 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216

%% Extract the coordinates and frequency information end

X ¼M(:,1);Y ¼ M(:,2); Z¼M(:,3); % x and y coordinates %% Print the velocity vectors
vector at the inflow plane
nd ¼size(X,1); % overall no of points TableU ¼[X Y Z U]; TableV ¼[X Y Z V]; TableW ¼[X Y Z W];
fmin ¼fmax/nm/2; % Min Frequency for i¼ 1:size(TableU,1)
df ¼ (fmax-fmin)/(nm-1); % Frequency step fprintf(fid2,’\r\n’); j¼1:size(TableU,2); fprintf(fid2,’%e,’,TableU(i,j));
fm ¼fmin:df:fmax; % Frequency vector end
tt ¼dtn(0:(nt-1)); % time vector for i¼ 1:size(TableV,1)
fprintf(fid3,’\r\n’); j¼1:size(TableV,2); fprintf(fid3,’%e,’,TableV(i,j));
%% Prepare the output file according to the format end
required by STAR CCM þ
fid2 ¼fopen(‘inletdata_U.csv’,’‘w’); for i¼ 1:size(TableW,1)
fid3 ¼fopen(‘inletdata_V.csv’,’‘w’); fprintf(fid4,’\r\n’); j¼1:size(TableW,2); fprintf(fid4,’%e,’,TableW(i,j));
fid4 ¼fopen(‘inletdata_W.csv’,’‘w’); end
csv ¼‘X,Y,Z,’; fclose(fid2);fclose(fid3);fclose(fid4);
fprintf(fid2,csv); fprintf(fid2,‘ux(m/s)[t ¼%es],’,tt); end
fprintf(fid3,csv); fprintf(fid3,‘vx(m/s)[t ¼%es],’,tt); function XYZ ¼RandSampleSphere(N)
fprintf(fid4,csv); fprintf(fid4,‘wx(m/s)[t ¼%es],’,tt); % Generate a random sampling XYZ N coordinates on a unit radius sphere
%% Calculate the average velocity, turbulent Intensity, t0z¼ 2npinrand(N,1);
and length scale profiles
Uav ¼Uhn(Z/h0u).^alphau; Iu ¼Iuhn(Z/h0I).^dIu; Iv¼ Ivhn z¼ sin(t0z);
(Z/h0I).^dIv;
Iw¼ Iwhn(Z/h0I).^dIw; Lu¼ Luhn(Z/h0L).^dLu; Lv¼ Lvhn t¼2npinrand(N,1);
(Z/h0L).^dLv;
Lw ¼Lwhn(Z/h0L).^dLw; %% Generate Wn (nf x nm) r¼sqrt(1-z.^2);
matrix, wn has 2.pi.fm mean and rms¼2.pi.df
wn ¼randn(nf,nm)n2npindf; x¼ r.ncos(t);
for nmi ¼ 1:nm y¼r.nsin(t);
wn(:,nmi) ¼wn(:,nmi)-mean(wn(:,nmi)); XYZ¼ [x,y,z];
wn(:,nmi) ¼wn(:,nmi)/std(wn(:,nmi))n2npindf; end
wn(:,nmi) ¼wn(:,nmi) þfm(nmi)n2npi; function F¼ mapp(X,q,p)
end % System of non-linear equations that maintains the divergence-free
condition
%% Calculate the spectrum matrices fx ¼[1;0;0];
Su ¼ zeros(nm,nd);Sv ¼zeros(nm,nd);Sw¼zeros(nm, K¼[X(1) X(2) X(3);q;p];
nd);
fms ¼(  0.5:0.05:0.5)ndf;nfsm¼ size(fms,2)/2 þ0.5; F¼fx-KnX;
nfs¼ size(fms,2);
end

Appendix B. : expressions for the dynamic response

B.1 Building dynamic responses

Modal forces Fi can be calculated from the base moments Mi using Eq. (B1).
⎧ Fx ⎫ ⎡1/h 0 0 ⎤ ⎧ My ⎫
⎪ ⎪ ⎢ ⎥⎪ ⎪
⎨ Fy ⎬ = ⎢0 1/h 0 ⎥ ⎨ Mx ⎬
⎪ ⎪ ⎢ ⎪ ⎪
⎩ Fθ ⎭ ⎣0 0 0.7⎥⎦ ⎩ Mt ⎭ (B1)

where h is the building height.

The rms displacement response in the generalized coordinate corresponding to a vibration mode i is calculated using the integral in Eq. (B2),
where Hi2 is called the mechanical admittance function and is expressed by Eq. (B3).
∞
∼ ⁎
xi = ∫0 H 2 SFi df
(B2)

1
Hi2 =
⎛⎛ ⎞
f2⎞
2
f2
Ki 2 ⎜⎜ ⎜1 − ⎟ + 4ζ 2 ⎟⎟
⎝⎝ fs 2 ⎠ fs 2 ⎠
(B3)

where ∼
⁎
x i is the rms generalized displacement; Hi 2 is the
mechanical admittance function for the mode i; SFi is the force
spectra for mode i.
 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216 215

x iBg , and resonant component, ∼

Mean, x¯ i⁎, background, ∼
⁎
xires , of the generalized displacement are calculated using to Eq. (B4).
F¯i ∼ ⁎ σFi ∼
, xires = ∼
xi − ∼
2 2
x¯ i⁎ = , xiBg = xiBg
Ki Ki (B4)

where σFi is the rms modal force of the mode i.

Peak displacement, x^itop , and acceleration x^¨itop at the building height are calculated as function of the generalized displacement
according to Eq. (B5).

x^itop = x¯ i⁎ + gf ∼
⁎
xi
^ 2 ⁎
x¨itop = gf (2πfi ) ∼
xires (B5)
^ jbi is the rms base
¯ bi is the mean base moment and M
Peak equivalent static base moments, Mbi , are calculated from Eq. (B6), where M
moment which can be calculated using Eq. (B7), where gf is the peak factor and it is taken here equal to 3.5.
^ ¯ bi + g Mjbi
Mbi = M f (B6)

jbi = (2πf )2 m Hs2 ∼

M
⁎
xi (along− and across− wind)
i
3
jbθ = (2πf )2 I Hs ∼
M
⁎
x θ (torsional direction)
θ
2 (B7)

where m, I are the mass and inertia per unit height which equals to
msWsDs and msWsDs (Ws2 þ Ds2)/12, respectively.

Appendix A. Supplementary material

Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.jweia.2015.04.004.

References
Abdi, D., Bitsuamlak, G.T., 2014. Wind flow simulations on idealized and real
complex terrain using various turbulence models. Adv. Eng. Softw. 75, 30–41.
Aboshosha, H., 2014. Response of Transmission Line Conductors Under Downburst
Wind (Ph.D. thesis). University of Western Ontario.
Aboshosha, H., Bitsuamlak, G., El Damatty, A., 2015. Turbulence characterization of
downbursts using LES. J. Wind Eng. Ind. Aerodyn. 136, 44–61.
Batten, P., Goldberg, U., Chakravarthy, S., 2004. Interfacing statistical turbulence
closures with large-eddy simulation. Am. Inst. Aeronaut. Astronaut. J. 42,
485–492.
Bechara, W., Bailly, C., Lafon, P., 1994. Stochastic approach to noise modeling for free
turbulent flows. Am. Inst. Aeronaut. Astronaut. J. 32 (3), 455–463.
Bitsuamlak, G.T., Dagnew, A.K., Erwin, J., 2010. Evaluation of wind loads on solar
panel modules using CFD. In: Proceedings of the Fifth International Symposium
on Computational Wind Engineering, 23–27 May. Chapel Hill, NC.
Bitsuamlak, G.T., Simiu, E., 2010. CFD's potential applications: wind engineering
perspective. In: Proceedings of the Fifth International Symposium on Compu-
tational Wind Engineering, 23–27 May. Chapel Hill, NC.
Castro, G.H., Paz, R.R., Sonzogni, V.E., 2011. Generation of turbulent inlet velocity
conditions for large eddy simulations. Mec. Comput. 2275–2288.
Chen, X., Kareem, A., 2005. J. Eng. Mech. ASCE 131, 1115–1125.
COST, 2007. Best Practice Guideline for the CFD Simulation of Flows in the Urban
Environment COST Action 732.
Dagnew, A., Bitsuamlak, G.T., 2013. Computational evaluation of wind loads on
buildings: a review. Wind Struct. 16 (6), 629–660.
Dagnew, A., Bitsuamlak, G.T., 2014. Computational evaluation of wind loads on
standard tall building using a large eddy simulation. Wind Struct. 18 (5),
567–598.
Dagnew, A., Bitsuamlak, G.T., 2010. LES evaluation of external wind pressures on a
standard tall building with and without a neighboring building. In: Proceedings
of the Fifth International Symposium on Computational Wind Engineering, 23–
27 May. Chapel Hill, NC.
Dagnew, A., Bitsuamlak, G.T., 2012. Large eddy simulation for wind-induced
response of tall buildings located in a city center. In: Proceedings of the 2012
Engineering Mechanics Institute and 11th ASCE Joint Specialty Conference on
Probabilistic Mechanics and Structural Reliability (EMI/PMC 2012), 17–20 June.
Notre Dame, IN.
Daniels, S.J., Castro, I.P., Xie, Z.T., 2013. Peak loading and surface pressure fluctua-
tions of a tall model building. J. Wind Eng. Ind. Aerodyn. 120, 19–28.
Davenport, A.G., 1967. Gust loading factors. J. Struct. Eng. ASCE 93, 11–34.
Davenport, A.G., 1993. How can we simplify and generalize wind loads? In: Pro-
ceedings of the Third Asia-Pacific Symposium on Wind Engineering. Keynote
Lecture, 13–15 December. Hong Kong.
Engineering Sciences Data Unit (ESDU) 85020, 2001. Characteristics of atmospheric
turbulence near the ground. Part II: Single Point Data for Strong Winds.
Franke, J., 2006. Recommendations of the COST action C14 on the use of CFD in
predicting pedestrian wind environment. In: Proceedings of the Fourth Inter-
national Symposium on Computational Wind Engineering, 16–19 July. Yoko-
hama, Japan, pp. 529–523.
Fung, J., Hunt, J., Malik, N., Perkins, R., 1992. Kinematic simulation of homogeneous
turbulence by unsteady random Fourier modes. J. Fluid Mech. 236, 281–318.
Germano, M., Piomelli, U., Moin, P., Cabot, W.H., 1991. A dynamic subgrid-scale
eddy viscosity model. Phys. Fluids 3 (7), 1760–1765.
Gousseau, P., Blocken, B., Stathopoulos, T., van Heijst, G.J.F., 2011. CFD simulation of
near-field pollutant dispersion on a high-resolution grid: a case study by LES
and RANS for a building group in downtown Montreal. Atmos. Environ. 45 (2),
428–438.
Hoshiya, M., 1972. Simulation of multi-correlated random processes and applica-
tion to structural vibration problems. Proc. JSCE 204, 121–128.
Huang, S., Li, Q., Wu, J., 2010. A general inflow turbulence generator for large eddy
simulation. J. Wind Eng. Ind. Aerodyn. 98, 600–617.
Huang, S.H., Li, Q.S., 2010. Large eddy simulation of wind effects on a super-tall
building. Wind Struct. 13 (6), 557–580.
Iwatani, Y., 1982. Simulation of multidimensional wind fluctuations having any
arbitrary power spectra and cross spectra. J. Wind Eng. Jpn. 11, 5–18.
Jarrin, N., Benhamadouche, S., Laurence, D., Prosser, R., 2006. A synthetic-eddy-
method for generating inflow conditions for large-eddy simulations. Int. J. Heat
Fluid Flow 27 (4), 585–593.
Jiang, Y., Alexander, A., Jenkins, H., Arthur, R., Chen, Q., 2003. Natural ventilation in
buildings: measurement in a wind tunnel and numerical simulation with large-
eddy simulation. J. Wind Eng. Ind. Aerodyn. 91 (3), 331–353.
Keating, A., Piomelli, U., Balaras, E., Kaltenbach, H.J., 2004. A priori and a posteriori
tests of inflow conditions for large-eddy simulation. Phys. Fluids 16, 4696.
Kijewski, T., Kareem, A., 1998. Dynamic wind effects: a comparative study of pro-
visions in codes and standards with wind tunnel data. Wind Struct. 1 (1),
77–109.
Kim, Y., Castro, I.P., Xie, Z.T., 2013. Inflow conditions for large-eddy simulations with
incompressible flow solvers. Comput. Fluids 84, 56–68.
Klein, M., Sadiki, A., Janicka, J., 2003. A digital filter based generation of inflow data
for spatially developing direct numerical or large eddy simulations.
J. Comput. Phys. 186, 652–665.
Kondo, K., Murakami, S., Mochida, A., 1997. Generation of velocity fluctuations for
inflow boundary condition of LES. J. Wind Eng. Ind. Aerodyn. 67–68, 51–64.
Kraichnan, R., 1970. Diffusion by a random velocity field. Phys. Fluids 13, 22–31.
Li, A., Ahmadi, G., Bayer, R., Gaynes, M., 1994. Aerosol particle deposition in an
obstructed turbulent duct flow. J. Aerosol Sci. 25 (1), 91–112.
216 H. Aboshosha et al. / J. Wind Eng. Ind. Aerodyn. 142 (2015) 198–216
Lim, H.C., Thomas, T.G., Castro, I.P., 2009. Flow around a cube in a turbulent
boundary layer: LES and experiment. J. Wind Eng. Ind. Aerodyn. 97 (2), 96–109.
Liu, K.L., Pletcher, R.H., 2006. Inflow conditions for the large eddy simulation of
turbulent boundary layers: a dynamic recycling procedure. J. Comput. Phys. 219
(1), 1–6.
Lumley, J.L., Panofsky, H.A., 1964. The Structure of Atmospheric Turbulence. Wiley-
Interscience, New York, p. 239.
Lund, T.S., Wu, X., Squires, K.D., 1998. Generation of turbulent inflow data for
spatially developing boundary layer simulations. J. Comput. Phys. 140, 233–258.
Maruyama, T., Morikawa, H., 1994. Numerical simulation of wind fluctuation con-
ditioned by experimental data in turbulent boundary layer. In: Proceedings of
the 13th Symposium on Wind Engineering, pp. 573–578.
Murakami, S., 1998. Overview of turbulence models applied in CWE-1997. J. Wind
Eng. Ind. Aerodyn. 74–76, 1–24.
Nozawa, K., Tamura, T., 2002. Large eddy simulation of the flow around a low-rise
building in a rough-wall turbulent boundary layer. J. Wind Eng. Ind. Aerodyn.
90, 1151–1162.
Nozu, T., Tamura, T., Okuda, Y., Sanada, S., 2008. LES of the flow around building
wall pressures in the center of Tokyo. J. Wind Eng. Ind. Aerodyn. 96, 1762–1773.
Sagaut, P., Garnier, E., Tromeur, E., Larchevêque, L., Labourasse, E., 2003. Turbulent
inflow conditions for LES of subsonic and supersonic wall-bounded flows.
Am. Inst. Aeronaut. Astronaut. J. 42, 469–478.
E. Shirani, J.H. Ferziger, W.C. Reynolds, 1981. Mixing of a Passive Scalar in Isotropic
and Sheared Homogeneous Turbulence. Report TF-15. Mechanical of Engi-
neering Department, Stanford University.
Smagorinsky, J., 1963. General circulation experiments with the primitive equa-
tions, I. The basic experiment. Mon. Weather Rev. 91, 99–164.
Smirnov, R., Shi, S., Celik, I., 2001. Random flow generation technique for large eddy
simulations and particle-dynamics modeling. J. Fluids Eng. 123, 359–371.
Tamura, T., 2000. Towards practical use of LES in wind engineering. J. Wind Eng.
Ind. Aerodyn. 96 (10–11), 1451–1471.
Tamura, T., 2008. Towards practical use of LES in wind engineering. J. Wind Eng.
Ind. Aerodyn. 96, 1451–1471.
Tamura, T., Nozawa, K., Kondo, K., 2008. AIJ guide for numerical prediction of wind
loads on buildings. J. Wind Eng. Ind. Aerodyn. 96, 1974–1984.
Tamura, T., 2010a. Application of LES-based model to wind engineering – imple-
mentation of meteorological effects. In: Proceedings of the Fifth International
Symposium on Computational Wind Engineering, 23–27 May. Chapel Hill, NC.
Tamura, T., 2010b. LES for aerodynamic characteristics of a tall building inside a
dense city district. In: Proceedings of the Fifth International Symposium on
Computational Wind Engineering, 23–27 May. Chapel Hill, NC.
Tominaga, Y., Stathopoulos, T., 2010. Numerical simulation of dispersion around an
isolated cubic building: model evaluation of RANS and LES. Build. Environ. 45
(10), 2231–2239.
Tominaga, Y., Stathopoulos, T., 2011. CFD modeling of pollution dispersion in a
street canyon: comparison between LES and RANS. J. Wind Eng. Ind. Aerodyn.
99 (4), 340–348.
Tominaga, Y., Mochida, A., Yoshiec, R., Kataokad, H., Nozu, T., Masaru Yoshikawa, M.,
Shirasawa, T., 2008. AIJ guidelines for practical applications of CFD to pedestrian
wind environment around buildings. J. Wind Eng. Ind. Aerodyn. 96 (10–11),
1749–1761.
Tutar, M., Celik, I., 2007. Large eddy simulation of a square cylinder flow: modelling
of inflow turbulence. Wind Struct. 10 (6), 511–532.
Warsido, W., Bitsuamlak, G.T., 2013. A dual aerodynamic data analysis framework
for tall buildings. J. Wind Eng. 10 (2), 18–40.
Xie, X.T., Castro, I.P., 2008. Efficient generation of inflow conditions for large eddy
simulation of street-scale flow. Flow Turbul. Combust. 81 (3), 449–470.
Zhou, Y., Kijewski, T., Kareem, A., 2003. Aerodynamic loads on tall buildings:
interactive database. J. Struct. Eng. – ASCE 3, 394–404.