Building and Environment 117 (2017) 11e23

Contents lists available at ScienceDirect

Building and Environment

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

CFD simulations of wind distribution in an urban community with a

full-scale geometrical model
Sumei Liu a, Wuxuan Pan a, Hao Zhang a, Xionglei Cheng a, Zhengwei Long a, *,
Qingyan Chen a, b
a
Tianjin Key Laboratory of Indoor Air Environmental Quality Control, School of Environmental Science and Engineering, Tianjin University, Tianjin, China
b
School of Mechanical Engineering, Purdue University, West Lafayette, IN, USA

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

Article history: A CFD study explored the use of wind information from a meteorological station to simulate wind dis-
Received 22 January 2017 tribution in an urban community, where the station may be located far away from the community. The
Received in revised form study constructed a full-scale urban model with building details for the area from the meteorological
23 February 2017
station to the community. The full-scale model was typically 2 to 20 km long, which was between the
Accepted 25 February 2017
Available online 28 February 2017
micro-scale and the meso-scale model. The three-dimensional, steady Reynolds-averaged Navier-Stokes
(RANS) equations was solved to simulate the urban wind flows. The investigation used two different
roughness setting methods for the ground surface to approximate building structures. The two methods
Keywords:
Computational fluid dynamics
produced similar results, but only one of them was able to provide wind information close to the ground.
Wind flow The wind distribution computed by the full-scale model was compared with that computed by a micro-
Full-scale model scale local model, where the computational domain included only building structures in the community
Surface roughness and limited spaces in the vicinity of the community. The wind velocity computed by the full-scale model
Buildings was 20% higher than the experimental data obtained by the local weather stations on a building rooftop
in the community, and that computed by the micro-scale model was almost twice as much. The full-scale
model is recommended for predicting wind distribution in an urban community, although more time is
required for construction of the geometrical model.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction nearby meteorological station as the inflow boundary conditions.

Because of the diversity of city sizes and terrains, the variety of
The investigation of wind around buildings in an urban com- building heights, and the complexity of building structures, wind
munity in the lower part of the atmospheric boundary layer (0 to flow from a nearby meteorological station to a community is
200 m) [1] is crucial in many wind environmental problems, extremely complicated. The inhomogeneity of a city's underlying
including natural ventilation design, pedestrian comfort, and air surfaces and the high roughness arising from architectural com-
pollutant dispersion [2,3]. Solving wind environmental problems plexes make the urban boundary layer very different from the at-
requires the study of wind flow into a community [4,5]. With the mospheric boundary layer on homogeneous and flat terrain,
development of computing resources and grid generation tech- especially for the urban canopy layer (UCP) that extends from the
niques, an increasing number of researchers have adopted ground to the roofs of buildings [9e11]. Using different assump-
computational fluid dynamics (CFD) in their investigation of urban tions for the urban boundary layer, researchers have developed
wind environments [6e8]. Appropriate boundary conditions are different simulation methods [12]. These methods can be classified
prerequisite for obtaining rational wind flow fields. CFD simula- as meso-scale CFD models (10 to 200 km in the horizontal direc-
tions generally use wind velocity and direction measured at a tion) and micro-scale model CFD models (100 m to 2 km in the
horizontal direction) [13,14].
Meso-scale CFD models generally treat urban architectures as
* Corresponding author. Tianjin Key Laboratory of Indoor Air Environmental increased surface roughness to emphasize the impact of
Quality Control, School of Environmental Science and Engineering, Tianjin Uni- geographical and meteorological conditions on urban wind envi-
versity, Tianjin, China. ronments [15e17]. In order to describe the heterogeneity of ground
E-mail address: longzw@tju.edu.cn (Z. Long).

http://dx.doi.org/10.1016/j.buildenv.2017.02.021
0360-1323/© 2017 Elsevier Ltd. All rights reserved.
12 S. Liu et al. / Building and Environment 117 (2017) 11e23

cover structures reasonably well, vast studies have aimed at performed well in simulations of urban wind flows [38,39]. In the
developing roughness parametrization methods to describe the RNG k-ε model, the transport equations of turbulence kinetic en-
influence of roughness elements under the UCP on the wind flow ergy k and its dissipation rate ε are:
field. The parametrization methods can be classified into three " #

categories, roughness length approaches (considering urban ar- vðrkÞ vðrkui Þ v n vk
þ ¼ nþ t þGε (1)
chitectures on wind speeds and turbulence by an enhanced vt vxi vxj sk vxj
roughness length) [18e20], drag force parametrization (using an
additional drag in the momentum equation to describe the effects "  #
of building structures on the wind flow field) [21e23], and porosity vðrεÞ vðrεui Þ v n vε ε ε2
þ ¼ nþ t þ Cε1 G  Cε2
concepts (treating buildings below the UCP as porous media vt vxj vxj sε vxj k k
through which wind can pass) [12,24,25]. The parametrization
1  h=h0 ε2
formulas or parameters have generally been obtained by means of  Cm h3 (2)
wind tunnel experiments in which roughness elements are simple 1 þ bh3 k
cubes with certain distributions, whereas actual urban construc-
tions are highly spatially inhomogeneous. It is essential to verify the where r is air density (kg/m3), t time (s), ui and uj the Reynolds
methods for actual urban environments. time-averaged velocity component in the xi and xj (i, j ¼ 1, 2, 3)
Micro-scale CFD models adopt large eddy simulation (LES) or directions, respectively, n the dynamic viscosity of air (m2/s),
nt ¼ Cm kε the turbulence kinematic viscosity (m2/s), sk ¼ 0:7194 the
2
Reynolds-averaged Navier-Stokes (RANS) turbulence models to
study wind flow around isolated building or architectural com- turbulence effective Prandtl number for k, G the source term, sε ¼
plexes. Those models explicitly construct building details on a local 0:7194 the turbulence effective Prandtl number for ε,
scale [26e28]. The wind velocity and direction from the nearest Cε1 ¼ 1:42 Cε2 ¼ 1:68, Cm ¼ 0:085, h¼ ð2Eij $Eij Þ1=2 kε,
meteorological station are used as boundary conditions. The impact !
vui vuj
of city buildings on upstream flow is modified by roughness length. Eij ¼ 12 vxj þ vxi , b ¼ 0:012, and h0 ¼ 4:38.
In the vertical direction, the wind profile is generally described as a
logarithmic law or power law [29e31]. As this kind of method
neglects the effect of buildings structures on the upstream flow, it is
applicable only where Monin-Obukhov similarity holds, such as in 2.2. Computational domain size and mesh generation
a suburban district with flat terrain [32,33].
Previous studies of meso-scale or micro-scale models have not The computational domain for a micro-scale model typically
determined how to correctly consider the influence of inhomoge- includes the community site concerned and its immediate sur-
neous building complexes in intermediate terrain between a roundings. The micro-scale model has a mature theory about the
community site in a city and a meteorological station that could be best way to determine the vertical, lateral, and flow direction do-
several kilometers away. Our investigation examined methods of mains. For the height of the computational domain, the blockage
simulating urban terrains with the use of roughness. Our goal was ratio should be below 3% [35,36]. The boundary in the four hori-
to find a suitable geometrical model with appropriate boundary zontal directions should be a distance of at least 5Hmax [35,40] to
conditions that could calculate wind distribution at a community allow for the establishment of a realistic flow behind the wake
site with reasonable accuracy. The accuracy was verified with region [35,36,41]. To eliminate as far as possible the errors caused
measured data from the field. This paper reports the results that by the size of the computational domain, for the micro-scale model
were obtained. we used a distance of 15Hmax beyond the community site in four
directions and blockage ratio should be below 3% for the height of
2. Research method the computational domain, as shown in Fig. 1(b). The micro-scale
model was not the focus of this investigation, but rather it was
2.1. CFD model used as a reference.
Our full-scale model used the nearest weather station as the
This investigation used a geometrical model that combines el- inflow boundary of the computational domain. The boundary in the
ements of both meso-scale and micro-scale models. The geomet- downstream region and the lateral boundaries were at a distance of
rical model is the same as a micro-scale model with explicitly 15Hmax from the community site, and blockage ratio should be
constructed building details, but the computational domain ex- below 3% for the height of the computational domain. Fig. 1(a) is a
tends from the community site to a weather station several kilo- sketch of the computational domain. Note that Hmax may differ
meters away. The domain is much larger than that used by a micro- between the full-scale and micro-scale models, because the tallest
scale model, which incorporates a small computational domain building in the full-scale model could be higher than that in the
around the community site. The full-scale model was typically 2 to micro-scale model. In the central computational domain, inho-
20 km long, which was between the micro-scale and the meso- mogeneous building complexes were elaborately constructed. This
scale model. Thus, we call this a full-scale model that differs from central domain included not only the community site concerned,
the micro-scale model only in the computational domain size. but also at least one additional street block in each lateral direction
Our study was based on recommendations from the overview of around the community [36]. In the outer regions in the lateral di-
CFD use in simulating the outdoor environment by Blocken et al. rections and on the downstream side of the central domain, the
[34], from “Best practice guideline for the CFD simulation of flows building geometry can be simplified as roughness, which will be
in the urban environment” [35], and from “Architectural Institute of discussed in Section 2.3.
Japan (AIJ) guidelines for practical applications of CFD to pedestrian Gambit 2.4.6 was used to generate a discrete grid in this study.
wind environment around buildings” [36]. These simulations were Because of the complexity of the geometrical model, this study
conducted with the use of a commercial CFD program, ANSYS used a hybrid grid scheme with a tetrahedral grid, which can adapt
Fluent 14.0 [37]. The CFD model used in this program solves the to the geometric structures very well. In the central region, building
Reynolds-averaged Navier-Stokes (RANS) equations with the re- structures were explicitly constructed, while on the two sides, no
normalization group (RNG) k-ε turbulence model. The model geometric configurations were established.
 S. Liu et al. / Building and Environment 117 (2017) 11e23 13

Fig. 1. Computational domain and boundary conditions for (a) the full-scale model and (b) the micro-scale model.

2.3. Boundary conditions For a fully developed atmospheric boundary layer profile, the
vertical profile for the mean wind speed has the following form
The mean velocity profile for the inflow boundary was modeled [42]:
as a power law, as shown in Eq. (3). For the full-scale model, the
 
inflow boundary was specified as the weather station location, uz 1 z
¼ ln (5)
whereas for the micro-scale model, the wind velocity profile was u*ABL k z0
modified to consider the impact of the terrain. The downstream
vertical boundary was modeled as outflow. The sky was treated as a where uz is the wind speed at a height of z (m/s), z is the height (m),
mirror plane. For the ground surface, the standard wall functions u*ABL the friction velocity of the atmospheric boundary layer (m/s),
with a roughness modification based on the equivalent sand-grain and z0 the aerodynamic roughness length (m) that is equivalent to
roughness height ks and the roughness constant Cs were used to the height at which the wind speed theoretically becomes zero.
reflect the influence of roughness elements, such as buildings and Table 1 shows the z0 values for different terrains [1,43].
trees, on the urban wind flow field [37]. The surfaces of the The combination of Eqs. (4) and (5) yields [1]:
buildings in the computational domain were non-slip.
The wind information provided by a weather station is the mean
ks Cs ¼ 9:793z0 (6)
velocity at a reference height over a period of 1 h, and therefore the
inlet boundary conditions can be determined by This equation can be used to describe z0 in order to take into
account the influence of roughness elements on the wind flow field.
 a
z As it is very time consuming to build the flow domain with full
Uz ¼ Ur (3) building details on the ground, the micro-scale and full-scale
zr
models divide the ground surface into two parts. One part is the
where Uz is the mean velocity profile (m/s), Ur the velocity (m/s) at central region with building details, apart from some relatively
reference height zr (m), and a a power law exponent determined by small buildings and trees that have been omitted. The other part is
the terrain category. the perimeter of the micro-scale model or the two sides of the full-
For the ground, the wall functions are based on the universal scale model that do not have building structures, as shown in Fig. 1.
near-wall velocity distribution (law of the wall) modified for a For the central region with building details, the roughness value
rough surface [37]: of the ground surface is z0 ¼ 0.03 m according to Table 1. However,
for the part of the ground surface without building structures, a
!
up u* 1 Eru* yp roughness value of z0 ¼ 1 m should be used. This investigation
¼ ln  DB (4) aimed to compare two kinds of roughness setting methods for the
ut2 k m
ground surface, (1) change in Cs while keeping ks ¼ 1 and (2) change

where up is the tangential wind speed in the center (point P) of the

1 1
wall-adjacent cell (m/s), u* ¼ Cm 4 k 2 is the wall function friction
= =
Table 1
velocity (m/s) with constant Cm ¼ 0.085, ut the wall function fric- Roughness values for different terrains [1,43].
tion velocity (m/s), k the von Karman constant (~0.42), E an Type z0 (m)
empirical constant for a smooth wall (z9.793), yp the distance from
Smooth wall 0
the center of the wall-adjacent cell to the wall (m), and m the dy- Grass land 0.03
namic viscosity (N$s/m2). The term DB hinges on the type and size Few isolated obstacles 0.05
of the roughness elements and has been found to be closely Low crops/Occasional large obstacles 0.1
correlated with the roughness Reynolds number, kþ s ¼ rksu /m,
* Parkland/Shrubs/Numerous obstacles 0.5
Densely distributed mid-rise and high-rise buildings/Clumps of forest 1
where ks is the physical roughness height.
14 S. Liu et al. / Building and Environment 117 (2017) 11e23

in ks while keeping Cs ¼ 1 for desirable values of z0. results. There were 3637 buildings scattered throughout the inner
region of the full-scale model, and the building heights ranged from
2.4. Field measurements for CFD validation 3 m to 117 m. As shown in Fig. 3(a), this study divided the inner
region into 69 sub-regions according to building height and den-
Validation of CFD simulations with experimental data is sity. Streets and community fences were usually used as partition
essential. This study calculated wind flow for an urban community lines. The mesh was first generated for each sub-region and then
by using wind information from a meteorological station. For combined for the whole domain. Because of our limited computing
validation purposes, several HOBO micro weather stations were capacity and the possible impact of grid size on accuracy, the
used to measure the wind velocity magnitude inside the commu- maximum grid size was 20 m near the Xiqing meteorological sta-
nity site from August 15, 2016 to October 31, 2016. The measure- tion area and was gradually reduced to 8 m at the community site in
ment locations were 2 m above the roof of one of the buildings. The the lengthwise direction. The maximum grid size in the vertical
frequency of the measured data was collected once per minute. The direction was 10 m. The grid resolution along the perimeter
measured data was averaged hourly for consistency with hourly without buildings was 20 m. The result was a total grid number of
data acquired from the meteorological station. The micro weather 7.0 million for the full-scale model of Fig. 2(c). Fig. 3(b) shows the
stations had a measuring accuracy of ±0.4% for wind velocity if it grid cells generated on the building surfaces and in the area
was greater than 0.5 m/s. without building details for the top right corner in Fig. 3(a), with
triangular and quadrilateral mesh. The mesh for the micro-scale
3. Case setup model was similar to that for the full-scale model, with the use of
a hybrid grid scheme with a tetrahedral grid in the central area
3.1. Geometrical models where building structures were explicitly constructed and a hex-
ahedral grid at the perimeter where there were no geometric de-
To compare the performance of the full-scale model with the tails. The maximum grid sizes of the inner region and in the
micro-scale model, this study selected Tianjin City in China as an perimeter area were 4 m and 6 m, respectively, and the total grid
example. We selected this city for our convenience. The city has an number was 6.4 million.
urban population of 5.21 million and urban structures that are
typical of large cities in China. Since Tianjin University is close to 3.3. Boundary conditions and case setup
the downtown area, this investigation chose a 1.05 km  1.05 km
section of the Tianjin University campus as the community site, as This investigation used the average wind velocity and direction
shown in Fig. 2(a). Tianjin has 12 meteorological stations scattered measured at the Xiqing meteorological station from 11:00 a.m. to
in and around the city [44] at distances of 10 kme100 km from the 2:00 p.m. on August 30, 2016, as the boundary conditions when the
university campus. The further a meteorological station is from the wind angles on the hour were 229 , 230 , 221, 232 , respectively.
campus, the less reliably does weather data at the station reflect The wind direction was southwest and the wind speed was 2.6 m/s.
that on the campus. This study selected the closest meteorological Thus, the western and southern boundaries were inflow ones.
station, Xiqing station, to provide hourly wind velocity and wind Because the Xiqing station was in a suburb, a power law with an
direction. This station is located 10 km from the campus in the exponent of 0.22 (as shown in Eq. (3)) was used to specify the
west-west-south (WWS) direction. vertical wind profile [45] at the two inlet boundaries in CFD sim-
Fig. 2(a) shows the geographical location of the Xiqing station ulations with the full-scale model, which corresponds z0 ¼ 1 m. For
(the large star at the lower left) in relation to the community site the micro-scale model, the exponent was changed to 0.3 to reflect
(the square at the upper right) on a Google satellite map. Fig. 2(b) the change from a suburb to a more built-up area, which corre-
illustrates the constructed three-dimensional geometrical model sponds z0 ¼ 0.3 m. The sky of the domain was modeled as a sym-
from the Xiqing station to the site. The model was 10.8 km long and metrical surface or as a mirror. The outlet boundary conditions on
1.8 km wide (encompassing at least one additional street block in the eastern and northern boundaries were set at zero static pres-
each direction around Tianjin University, as discussed in Section sure. The wall functions with the roughness values were used for
2.2). All buildings inside this area were explicitly resolved, while the ground surfaces.
other structures, such as trees, landscaping, roads, etc., were As discussed in the previous section, there were two roughness
neglected in the model. The corresponding computational domain setting methods. This investigation aimed to identify the better
was extended by 1.8 km or about 15Hmax on both sides and at the method. As shown in Table 2, Case 01 used a smooth wall boundary
right end, where the tallest building in the area was 117 m high. The for the ground surface as a benchmark case to be compared with
resulting computational domain was 12.6 km long, 5.4 km wide, the other cases with using wall functions to illustrate the influence
and 0.351 km high. The maximum blockage ratio was 1.5%, as of roughness elements on the wind flow field. Eight cases were
shown in Fig. 2(c). The use of the outer regions would reduce the then set up to compare the effects of changing Cs or ks according to
amount of labor-intensive work required for constructing the Eq. (6), which also led to different values of z0. In these cases, we
detailed building structures, if they could be approximated by changed only one parameter at a time in simulations of the wind
roughness. distribution with the full-scale model.
For comparison, this study also built a micro-scale model as Cases 02 to 05 assumed ks ¼ 1 and varied Cs for different z0 in the
shown in Fig. 2(d). The height of the computational domain was outer regions without building details, while z0 for the inner region
105 m or three times the height of the tallest building (35 m) inside remained at zero. This investigation focused more on the outer
this area. The other four directions were extended by a distance of regions because there were no building details, and the impact of
15  35 ¼ 525 m. Therefore, the computational domain for the the roughness setting on the wind flow could be easily studied.
micro-scale model was 2.1 km long, 2.1 km wide, and 0.105 km Cases 06 to 09 kept Cs at 1 and varied ks for different z0 in the outer
high. The maximum blockage ratio was 2.4%. regions. These cases were used to assess the two roughness setting
methods.
3.2. Grid arrangement and mesh generation Cases 10 and 11 changed z0 to 0.03 (compared to z0 ¼ 0 in the
reference case) with the two different methods, while the rough-
Mesh quality is very important for the accuracy of simulation ness in the outer regions was assumed to be zero. Finally, Cases 12
 S. Liu et al. / Building and Environment 117 (2017) 11e23 15

Fig. 2. Geometrical models used in this study: (a) plane view of the topographical information from Xiqing meteorological station (the star to the community site (the square), (b)
building details used in the full-scale model, (c) computational domain of the full-scale model, and (d) computational domain of the micro-scale model with building details in the
center.
16 S. Liu et al. / Building and Environment 117 (2017) 11e23

3.5. Field measurements

Field measurements were conducted at Tianjin University with

three HOBO micro weather stations positioned on the rooftop of a
building as shown in Fig. 4. The building height was 9 m, and the
sensor height was 11 m above the ground.

4. Results

4.1. Model validation

This investigation first validated the CFD models by comparing

our simulated results with wind tunnel data from the literature.
The wind tunnel data had been collected at an urban complex in
Niigata, Japan, with a scale of 1:250 [46]. Scalar wind velocities
were measured at a height of 8 mm above the wind tunnel floor
(corresponding to a height of 2 m above the ground in actual scale)
by multi-point thermister anemometers. Fig. 5(a) shows the
building model of the urban area in the actual scale of 500 m long,
500 m wide, and 300 m high. A hybrid grid type was used in this
study, with a tetrahedral grid in the inner area with building
structures and a hexahedral grid in the outer regions. The total grid
Fig. 3. Grid generation for the full-scale model: (a) sub-regions and (b) grid cells on number was 5.6 million, as shown in Fig. 5(b). Table 3 lists the
the building surfaces and in the perimeter zone without building details.
boundary conditions used in this investigation.
In the wind tunnel case, the micro-scale model is the same as
the full-scale model. Therefore, the experimental data from the
wind tunnel was used to validate aspects of the physical model
and 13 changed z0 to 1 for the outer regions.
such as the turbulence model, numerical algorithm, etc. The RNG k-
Evaluation of these cases allowed us to identify a suitable
ε model was used to simulate the urban wind flow, the SIMPLE
roughness setting method for the full-scale model. We could then
algorithm was used for pressure and velocity coupling, and second-
compare the performance of the full-scale model with that of the
order discretization schemes were used for the governing
micro-scale model.
equations.
Fig. 6(a) compares the wind speed ratio (defined as the ratio of
the wind speed magnitude at each measuring point and the wind
3.4. Other computational parameters speed at the same height at the inflow boundary) computed in
selected locations at a height of 8 mm above the wind tunnel floor
The governing equ ations were solved using the finite volume with the corresponding experimental data. The mean relative error
method. The SIMPLE algorithm was used for pressure and velocity of our calculation was 36.4%. As a comparison, Fig. 6(b) shows the
coupling, and the second-order discretization schemes were used computed results from Yoshihide et al. [48] with the same exper-
for solving all the variables. When the scaled residuals reached 104 imental data. Although Yoshihide et al. did not show the quanti-
for mass conservation, U, V, W, k, ε, the solution was considered to tative error, the two computed results were similar. The
be converged. Velocity magnitude on several specified points were comparisons indicate that the RNG k-ε model and numerical al-
also monitored to confirm that the solution did not change with the gorithm can predict the airflow in an urban area reasonably well,
iteration, which was another criterion for convergence. considering the complexity of the geometry and wind flow.

Table 2
Case setup for different terrains (z0) under different combination of ks and Cs.

Cases Outer regions without building details (EX) Inner region with building details (IN)

ks Cs z0 (m) ks Cs z0 (m)

Case 01 0 0.5 0 0 0.5 0

Case 02 0.5 0.05
Case 03 1 0.1
1
Case 04 4.897 0.5
Case 05 9.793 1
0 0.5 0
Case 06 0.5 0.05
Case 07 1 0.1
1
Case 08 4.897 0.5
Case 09 9.793 1

Case 10 0.59 0.5

0 0.5 0 0.03
Case 11 0.5 0.59

Case 12 1 9.793
1 0.59 0.5 0.03
Case 13 9.793 1
 S. Liu et al. / Building and Environment 117 (2017) 11e23 17

Fig. 4. Three measurement positions on the rooftop of a campus building: (a) sensor locations and (b) building location at the university site.

Fig. 5. (a) Geometric configuration of the urban area used for the CFD validation and (b) the grid built for the CFD simulations.

Table 3
Boundary conditions used in this study.

Inlet flow boundary Interpolated values of velocity and turbulence kinetic energy from experimental data for the inflow, ε ¼ C1/2
D $k$dU/dz(ε ¼ Pk)
Outflow boundary Outflow with zero pressure
Top surface Solid mirror wall
Ground surface Logarithmic law with roughness length z0 ¼ 0.024 m
Building surfaces Logarithmic law for smooth walls

4.2. Grid-independence analysis the university campus. As discussed in the case setup section, it is
important to study the impact of different z0 values on the wind
A mesh independence study was conducted for the micro-scale profile in the outer regions without building details. Case 01 was a
model and full-scale model. Fig. 7(a) shows the computed vertical reference case with a smooth ground surface (z0 ¼ 0). The z0 for
wind profiles with different grid number for the micro-scale model. Cases 02 to 05 increased in the outer regions, while smooth ground
The results show that the 1.4 million grid was sufficiently good for was assumed in the inner region. In the outer regions, different z0
the micro-scale model. This study used the 6.4 million grid for the values were achieved by varying Cs while keeping ks ¼ 1. Fig. 8
subsequent simulations. Fig. 7(b), (c) and (d) show the computed shows the computed vertical wind profile along the height in
vertical wind profiles with different grid numbers for the full-scale three different regions: the windward outer region, inner region,
model in the windward outer region, inner region, and leeward and leeward outer region. With an increase in Cs or z0, the wind
outer region. The 7.0 million grid can produce a grid-independent velocity near the ground in the two outer regions decreased
result. dramatically, while that in the inner region remained the same. The
results look plausible, as a larger roughness length would create a
4.3. Comparison of different methods in specifying roughness high resistance and reduce the wind velocity near the ground.
parameters These results have verified the possibility of achieving different z0
values by changing Cs.
With the validated turbulence model, this investigation used the This investigation also simulated the wind profiles by varying ks
full-scale geometrical model to determine the wind distribution on for different z0 while keeping Cs constant from Case 06 to Case 09 as
18 S. Liu et al. / Building and Environment 117 (2017) 11e23

Fig. 6. Comparison of the normalized wind speed computed by the RNG k-ε model with the measured data from the wind tunnel [46] at selected locations (a) between the present
study and the data, and (b) between the study from Yoshihide et al. [46] and the data.

Fig. 7. Computed wind profiles with different grid number: (a) surface averaged velocity for micro-scale model, (b) averaged in the windward outer region for the full-scale model,
(c) averaged in the inner region for the full-scale model, and (d) averaged in the leeward outer region for the full-scale model.

shown on Table 2. Fig. 9 compares the wind profiles for Cases 02 to According to Eq. (8), any combination of ks and Cs with the same
05 with those for Cases 06 to 09. When the roughness was small, product can be used. ANSYS Fluent requires ks < zp where zp is the
the two methods for the same z0, such as Cases 02 and 06, yielded distance from the center (point p) of the wall-adjacent cell to the
nearly the same results. With the largest z0 value of 1, the difference wall surface [47] and the default maximum Cs is 1. When z0 ¼ 1 and
between the two roughness setting methods was 11.6% for the air Cs ¼ 1 as in Case 09, then ks ¼ 9.793, which requires a minimum
velocity near the ground. first grid size near the ground of 2  9.793 ¼ 19.586 m. If
 S. Liu et al. / Building and Environment 117 (2017) 11e23 19

Fig. 8. Computed wind profiles with different z0 by varying Cs and keeping ks ¼ 1: (a) averaged in the windward outer region, (b) averaged in the inner region, and (c) averaged in
the leeward outer region.

Fig. 9. Comparison of the computed wind profiles with different z0 by varying Cs and keeping ks ¼ 1 for Cases 02 to 06 or by varying ks and keeping Cs ¼ 1: (a) averaged in the
windward outer region, (b) averaged in the inner region, and (c) averaged in the leeward outer region.
20 S. Liu et al. / Building and Environment 117 (2017) 11e23

Fig. 10. Comparison of the computed wind profiles with z0 ¼ 0.03 for the inner region and z0 ¼ 1 for the outer region with the two specification methods (a) averaged in the
windward outer region, (b) averaged in the inner region, and (c) averaged in the leeward outer region.

information close to the ground was desired, such as the wind 11 km away at the meteorological station, while in the micro-scale
speed at pedestrian level, this method of setting roughness would model it was specified in the vicinity of the campus, and the power
not produce the necessary information. Thus, changing Cs is better law exponent was changed to 0.3 to reflect the change from a
than changing ks. suburb to a more built-up area.
We designed Cases 10 and 11 to study the impact of the two Fig. 11(a) depicts the wind flow field computed by the full-scale
roughness setting methods in the inner region with z0 ¼ 0.03. As model at a height of 11 m above the ground (in accordance with the
shown in Fig. 10, the difference was as small as 0.6%. Maintaining height of the measuring locations on the building roof). Fig. 11(b)
the z0 value in the inner region at 0.03, we studied the two different and (c) compare the wind flow fields on the university campus
roughness setting methods for z0 ¼ 1 in the outer regions (Cases 12 obtained by the two models. The velocity magnitude from the
and 13). The methods again differed by 11.6% as illustrated in Fig. 10. micro-scale model was much larger than that from the full-scale
This result implies that the impact of roughness in the inner region model. The wind would lose its momentum as it travels from the
on the velocity profile in the outer regions was small. The velocity meteorological station to the downtown area because of the
profiles in the outer regions depended only on the roughness in building structures in the inner region and the roughness used in
these regions. the outer regions. However, the micro-model used the weather
As the outer regions of the full-scale model did not have any data at a distance of 11 km away as the boundary conditions in the
building structures, z0 ¼ 1 should be used according to Table 1. The vicinity of the campus even though the power law exponent was
velocity near the ground using the roughness length of z0 ¼ 1 was changed. The results clearly indicate that the two models did not
similar to that in the inner region as shown in Figs. 8e10. Thus, perform similarly.
z0 ¼ 1 can represent the city terrains reasonably well for outer re- Fig. 12 compares the simulated wind velocity with the corre-
gions of Tianjin that do not have detailed building structures. sponding experimental data at the rooftop locations in Fig. 4.
Although the data points from the experiment were limited, the
4.4. Comparison of different geometric models trend is clear. The results show that the micro-scale model signif-
icantly overestimated the wind magnitude. This occurred because
In Section 4.2 we compared two kinds of roughness setting the model failed to consider the decay in velocity as the wind
methods for the ground surface in the full-scale model. Using the traveled through the city from the meteorological station. The re-
same boundary conditions, this study has also compared the per- sults obtained by the full-scale model were in reasonable agree-
formance of the full-scale model with that of the micro-scale model ment with the measured data. The simulations with the two
as shown in Fig. 2 for predicting the wind environment on the roughness setting methods for the full-scale model again generated
university campus. The two models used the same wind velocity similar accuracy at this height. Nevertheless, the full-scale model
profiles from the meteorological station as the inlet boundary overestimated the wind velocity by 20%. This could be due to our
conditions and the same roughness length on the ground surface. underestimation of the roughness in the inner region. We can
In the full-scale model, the wind velocity profile was specified conclude from the comparison that the full-scale model performed
 S. Liu et al. / Building and Environment 117 (2017) 11e23 21

Fig. 11. Comparison of the performance of the two geometrical models: (a) wind flow field computed by the full-scale model, (b) the wind on the university campus computed by
the full-scale model, and (c) the wind on the campus computed by the micro-scale model.

much better than the micro-scale model. case study. However, the model had some limitations:

 The full-scale geometrical model was established with the aid of

5. Discussion a satellite image because no public geometrical model was
available. The model used approximations that may have
The full-scale model performed quite well for such a complex

Fig. 12. Comparison of the results simulated by the full-scale model and micro-scale model with the wind velocity measured on the roof of a building on the campus.
22 S. Liu et al. / Building and Environment 117 (2017) 11e23

affected the wind profiles. We were unable to estimate the er- References
rors that may have resulted.
 This investigation selected the Xiqing meteorological station, [1] B. Blocken, T. Stathopoulos, J. Carmeliet, CFD simulation of the atmospheric
boundary layer: wall function problems, Atmos. Environ. 41 (2007) 238e252.
which was the closest to the university campus. The simulations [2] B. Blocken, Computational Fluid Dynamics for urban physics: importance,
used wind data from the direction upwind from the meteoro- scales, possibilities, limitations and ten tips and tricks towards accurate and
logical station as the boundary conditions. If the wind came reliable simulations, Build. Environ. 91 (2015) 219e245.
[3] A.A. Razak, A. Hagishima, S.A.Z.S. Salim, Progress in wind environment and
from a different direction, such as from the university campus to outdoor air ventilation at pedestrian level in urban area, Appl. Mech. Mater.
the meteorological station, could the wind data from the station 819 (2016) 236e240.
be used as the inlet boundary conditions? To use wind data from [4] T. Yukio, Y. Ryuichiro, Advanced Environmental Wind Engineering. Japan,
2016.
the upwind direction, we would need to build a new full-scale [5] E. Willemsen, J.A. Wisse, Design for wind comfort in The Netherlands: pro-
geometrical model. The effort would be significant. cedures, criteria and open research issues, J. Wind Eng. Ind. Aerodyn. 95
 This study used the RNG k-ε model to simulate the urban wind (2007) 1541e1550.
[6] Q. Chen, Chapter 7: design of natural ventilation with CFD, in: L.R. Glicksman,
flow under steady-state conditions. However, wind constantly
J. Lin (Eds.), Sustainable Urban Housing in China, Springer, 2006, pp. 116e123.
changes in direction and magnitude. Should a transient simu- [7] R. Ramponi, B. Blocken, L.B.D. Coo, et al., CFD simulation of outdoor ventilation
lation method be used, it would significantly increase the of generic urban configurations with different urban densities and equal and
computing costs. unequal street widths, Build. Environ. 92 (2015) 152e166.
[8] M. Jin, W. Zuo, Q. Chen, Simulating natural ventilation in and around buildings
by fast fluid dynamics, Numer. Heat. Transf. Part A Appl. 64 (4) (2013)
273e289.
[9] T.R. Oke, Boundary Layer Climate, Methuan & Co. LTD, London, 1987, p. 274.
6. Conclusions [10] I.N. Harman, J.J. Finnigan, A simple unified theory for flow in the canopy and
roughness sublayer, Bound.-Layer Meteorol. 123 (2007) 339e363.
[11] R.B. Stull, An introduction to boundary layer meteorology. Canada, Kluwer
This investigation conducted CFD simulations to compare Acad. Publ. 105 (3) (1988) 515e520.
different geometrical models and roughness setting methods for [12] R.D. Crago, W. Okello, M.F. Jasinski, Equations for the drag force and aero-
dynamic roughness length of urban areas with random building heights,
simulating the wind environment in an urban community. The
Bound.-Layer Meteorol. 145 (2012) 423e437.
study led to the following conclusions: [13] J. Hang, Y.G. Li, Wind conditions in idealized building clusters: macroscopic
simulations using a porous turbulence model, Bound.-Layer Meteorol. 136
 The micro-scale model predicted the wind distribution for a (2010) 129e159.
[14] R.E. Britter, S.R. Hanna, Flow and dispersion in urban areas, Annu. Rev. Fluid
community with boundary conditions from a wind tunnel with Mech. 35 (2003) 469e496.
a relative error of 36.4% compared with experimental data from [15] H.P. Liu, B.Y. Zhang, J.G. Sang, et al., A laboratory simulation of plume
the literature. Our simulation accuracy appeared to be similar to dispersion in stratified atmospheres over complex terrain, J. Wind Eng. Ind.
Aerodyn. 89 (1) (2001) 1e15.
that obtained by the group who did the wind tunnel tests. [16] A.J. Ding, T. Wang, M. Zhao, et al., Simulation of sea-land breezes and a dis-
 This study tested two different roughness setting methods, and cussion of their implications on the transport of air pollution during a multi-
both produced satisfactory wind profiles for outer regions day ozone episode in the Pearl River Delta of China, Atmos. Environ. 38 (39)
(2004) 6737e6750.
without building details. It is better to vary Cs while keeping [17] H. Tong, A. Walton, J.G. Sang, et al., Numerical simulation of the urban
ks ¼ 1 because the method can be used to calculate wind dis- boundary layer over the complex terrain of Hong Kong, Atmos. Environ. 39
tribution close to the ground. (19) (2005) 3549e3563.
[18] E. Ng, C. Yuan, L. Chen, et al., Improving the wind environment in high-density
 The full-scale model was able to produce the wind velocity cities by understanding urban morphology and surface roughness: a study in
distribution in the community by using wind information from a Hong Kong, Landsc. Urban Plan. 101 (2011) 59e74.
meteorological station that was 11 km away. It overestimated [19] B.G. Kim, C.H. Lee, S.J. Joo, et al., Estimation of roughness parameters within
sparse urban-like obstacle arrays, Bound.-Layer Meteorol. 139 (2011)
the wind velocity by only 20% compared with the measured data
457e485.
obtained on the rooftop of a building in the community site. If [20] M.C. Cao, Z.H. Lin, Impact of urban surface roughness length parameterization
the wind information was used for the micro-scale model, the scheme on urban atmospheric environment simulation, J. Appl. Math. 12 (2)
model would overestimate the wind velocity by a factor of two. (2014) 155e175.
[21] O. Coceal, S.E. Belcher, A canopy model of mean winds through urban areas,
This is because the model did not consider the wind decay when Q. J. R. Meteorol. Soc. 130 (2004) 1349e1372.
it traveled through the city, even though the exponent on the [22] S.D. Sabatino, E. Solazzo, P. Paradisi, A simple model for spatially-averaged
wind profile was modified for the city terrain. wind profiles within and above an urban canopy, Bound.-Layer Meteorol.
127 (2008) 131e151.
[23] Z. Peng, J. Sun, Characteristics of the drag coefficient in the roughness sublayer
over a complex urban surface, Bound.-Layer Meteorol. 153 (2014) 569e580.
[24] M. Skote, M. Sandberg, U. Westerberg, et al., Numerical and experimental
Funding studies of wind environment in an urban morphology, Atmos. Environ. 39
(2005) 6147e6158.
[25] F. Kuwahara, I. Yamane, A. Nakayama, Large eddy simulation of turbulent flow
The research presented in this paper was partially supported by in porous media, Int. Commun. Heat Mass Transf. 33 (4) (2006) 411e418.
the National Natural Science Foundation of China through Grant [26] Y. Li, T. Stathopoulos, Numerical evaluation of wind-induced dispersion of
pollutants around a building, J. Wind Eng. Ind. Aerodyn. 67 (1997) 757e766.
No. 51678395 and by the national key project of the Ministry of
[27] A. Martilli, J.L. Santiago, CFD simulation of airflow over a regular array of
Science and Technology, China, on “Green Buildings and Building cubes. Part II: analysis of spatial average properties, Bound.-Layer Meteorol.
Industrialization” through Grant No. 2016YFC0700500. 122 (3) (2007) 635e654.
[28] J.L. Santiago, A. Martilli, F. Martin, 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. 122 (3) (2007)
Acknowledgement 609e634.
[29] H. Montazeri, B. Blocken, CFD simulation of wind-induced pressure co-
The authors would like to thank Prof. Yoshihide Tominaga of efficients on buildings with and without balconies: validation and sensitivity
analysis, Build. Environ. 60 (2013) 137e149.
Niigata Institute of Technology, Japan, for providing us with the
[30] S.M. Liu, J.J. Liu, Q.X. Yang, et al., Coupled simulation of natural ventilation and
wind tunnel data and further explanation of the data. The authors daylighting for a residential community design, Energy Build. 68 (2014)
are also in debt to Prof. Bert Blocken of Eindhoven University of 686e695.
[31] J.L. Santiago, O. Coceal, A. Martilli, et al., Variation of the sectional drag co-
Technology, The Netherlands, for his help with boundary-layer
efficient of a group of buildings with packing density, Bound.-Layer Meteorol.
theory.
 S. Liu et al. / Building and Environment 117 (2017) 11e23 23

128 (3) (2008) 445e457. 93 (12) (2005) 891e904.

[32] E. Solazzo, S.D. Sabatino, N. Aquilina, et al., Coupling mesoscale modelling [40] A. Scaperdas, S. Gilham, Thematic Area 4: best practice advice for civil con-
with a simple urban model: the Lisbon case study, Bound.-Layer Meteorol. struction and HVAC, QNET-CFD Netw. Newsl. 2 (4) (2004) 28e33.
137 (2010) 441e457. [41] J.G. Bartzis, D. Vlachogiannis, A. Sfetsos, Thematic Area 5: best practice advice
[33] J.F. Barlow, G.G. Rooney, S.V. Hünerbein, et al., Relating urban surface-layer for environmental flow, QNET-CFD Netw. Newsl. 2 (4) (2004) 34e39.
structure to upwind terrain for the Salford experiment (Salfex), Bound.- [42] P.A. Durbin, B.A. Petterson Reif, Statistical Theory and Modelling for Turbulent
Layer Meteorol. 127 (2008) 173e191. Flows, Wiley, Chichester, UK, 2001.
[34] B. Blocken, T. Stathopoulos, J. Carmeliet, et al., Application of CFD in building [43] J. Wieringa, Updating the Davenport roughness classification, J. Wind Eng. Ind.
performance simulation for the outdoor environment: an overview, J. Build Aerodyn. 41 (1e3) (1992) 357e368.
Perform. Simul. 4 (2) (2011) 157e184. [44] http://data.cma.cn/.
€rg, B. Alexander, Best Practice Guideline for the CFD Simulation of Flows in
[35] F. Jo [45] J.X. Gu, Dictionary of Atmospheric Science, China Meteorological Press, Bei-
the Urban Environment: COST Action 732 Quality Assurance and Improve- jing, 1994, p. 156.
ment of Microscale Meteorological Models, COST Office, Brussels, 2007. [46] Y. Tominaga, R. Yoshie, A. Mochida, et al., Cross comparisons of CFD prediction
[36] Y. Tominaga, A. Mochida, R. Yoshie, et al., AIJ guidelines for practical appli- for wind environment at pedestrian level around buildings, Part 2: compar-
cations of CFD to pedestrian wind environment around buildings, J. Wind Eng. ison of results for flowfield around building complex in actual urban area,
Ind. Aerodyn. 96 (2008) 1749e1761. September 12e14, in: The Sixth Asia-Pacific Conference on Wind Engineering
[37] ANSYS Inc, ANSYS Fluent 14.0 User's Guide, ANSYS Inc, Southpointe, 2011. 633 (APCWE-VI) Seoul, Korea, 2005, pp. 2661e2670.
[38] A.D. Ferreira, A.C.M. Sousa, D.X. Viegas, Prediction of building interference [47] B. Blocken, J. Carmeliet, T. Stathopoulos, CFD evaluation of wind speed con-
effects on pedestrian level comfort, Orig. Res. Article J. Wind Eng. Ind. Aero- ditions in passages between parallel buildingsdeffect of wall-function
dyn. 90 (4e5) (2002) 305e319. roughness modifications for the atmospheric boundary layer flow, J. Wind
[39] A.X. Zhang, C.L. Gao, L. Zhang, Numerical simulation of the windfield around Eng. Ind. Aerodyn. 95 (2007) 941e962.
different building arrangements, Orig. Res. Article J. Wind Eng. Ind. Aerodyn.