You are on page 1of 26

Environ Fluid Mech (2011) 11:439–464

DOI 10.1007/s10652-011-9211-6

ORIGINAL ARTICLE

A CFD-based wind solver for an urban fast response


transport and dispersion model

Akshay A. Gowardhan · Eric R. Pardyjak ·


Inanc Senocak · Michael J. Brown

Received: 31 March 2010 / Accepted: 24 January 2011 / Published online: 24 February 2011
© Springer Science+Business Media B.V. 2011

Abstract In many cities, ambient air quality is deteriorating leading to concerns about the
health of city inhabitants. In urban areas with narrow streets surrounded by clusters of tall
buildings, called street canyons, air pollution from traffic emissions and other sources may
accumulate resulting in high pollutant concentrations. For various situations, including the
evacuation of populated areas in the event of an accidental or deliberate release of chemi-
cal, biological and radiological agents, it is important that models should be developed that
produce urban flow fields quickly. Various computational techniques have been used to cal-
culate these flow fields, but these techniques are often computationally intensive. Most fast
response models currently in use are at a disadvantage in these cases as they are unable
to correlate highly heterogeneous urban structures with the diagnostic parameterizations on
which they are based. In this paper, a novel variant of the popular projection method for
solving the Navier–Stokes equations has been developed and applied to produce fast and rea-
sonably accurate parallel computational fluid dynamics (CFD) solutions for flow in complex
urban areas. This model, called QUIC-CFD represents an intermediate balance between fast
(on the order of minutes for a several block problem) and reasonably accurate solutions. This
paper details the solution procedure and validates this model for various simple and complex
urban geometries.

Keywords Fast-response · Urban dispersion modeling · Computational fluid dynamics ·


Reynolds-averaged Navier–Stokes equations · Flow around buildings

A. A. Gowardhan · E. R. Pardyjak (B)


Department of Mechanical Engineering, University of Utah, Salt Lake City, UT 84112, USA
e-mail: pardyjak@eng.utah.edu

A. A. Gowardhan · M. J. Brown
Los Alamos National Laboratory, Los Alamos, NM 87545, USA

I. Senocak
Department of Mechanical and Biomedical Engineering, Boise State University, Boise, ID 83725, USA

123
440 Environ Fluid Mech (2011) 11:439–464

1 Introduction

The transport and dispersion of airborne contaminants within an urban region can be strongly
influenced by buildings which perturb local winds leading to regions of channeling, updrafts,
downdrafts, and recirculating flow (e.g. [5,4,24,31]). In addition, urban regions generate
zones of high turbulence intensity in proximity to regions of low turbulence intensity. Com-
putational fluid dynamics (CFD) models have been used to compute the flow field in urban
topography. Comparison of these models with field measurements indicates that these models
work well in most regions [7,12,13,37]. These CFD models, however, are computationally
too expensive, for many applications related to toxic releases in cities or at industrial facilities
where turn around time is very important. Examples of such applications include:
1. Vulnerability assessments (where many simulations must be performed)
2. Training and table top exercises (where feedback or interaction is desired)
3. Emergency response (where simulations may need to be performed quickly in the field)
4. Sensor siting and source inversion (where many simulations must be performed)
Several fast-response flow and dispersion models have been developed to account for the
effects of a single building or group of buildings. Hotchkiss and Harlow [22] derived an
analytical solution for potential flow over a notch to describe the vortex associated with the
velocity field within a 2D urban street canyon. The EPA PRIME model (Schulman et al. [43])
accounts for flow patterns and plume concentration fields for isolated buildings. The APRAC
model (Dabberdt et al. [14]) accounts for spatial variability of concentration due to street
canyon vortex motion. Likewise, Yamartino et al. [53] developed the Canyon-Plume-Box
model for computing vehicle emission concentrations in street canyons by adapting a unique
segmented Gaussian plume model.
Other urban dispersion models have been developed to predict transport over greater dis-
tances. For example, Theurer et al. [48] modified the Gaussian plume model to account for
the plume centerline shift that may occur due to channeling in street canyons. Hall et al. [21]
developed a Gaussian puff model called the Urban Dispersion Model (UDM) for use from
neighborhood to city scales. In UDM, empirical parameterizations account for the interaction
of puffs with buildings to produce building wake-mixing and some street channeling. Röckle
[42] derived a diagnostic model that computes 3-D flow around buildings using empirical
equations and mass conservation.
With the exception of the Röckle approach, the models cited above do not compute the
velocity field around clusters of buildings. The Röckle approach has been applied to real cit-
ies with non-ideal building layouts, but the underlying flow parameterizations were derived
from idealized wind-tunnel experiments for isolated buildings and simple arrays of buildings
[44].
It is clear that most fast response models rely on empirical algorithms based on idealized
building configurations. This makes it difficult to generalize the accuracy of these models for
flow fields in highly heterogeneous urban terrain without many validation exercises. Due to
these issues, there is a need to develop relatively fast and reasonably accurate computational
fluid dynamics (CFD) techniques that solve the Navier–Stokes equations for complex urban
areas. Here, by “reasonably fast”, we mean models that run on the order of minutes for a
typical urban problem of several blocks so that they can be run multiples times in order
to understand the flow features for different boundary condition, which would aid decision
making. “Reasonably accurate” refers to the capability of the model to simulate the important
flow features in an urban environment and predicted wind speeds are within a factor of two of
the observed data. As a part of the Quick Urban & Industrial Complex (QUIC) fast response

123
Environ Fluid Mech (2011) 11:439–464 441

dispersion modeling system [35,44,20], a fast, CFD based model has been developed called
QUIC-CFD which represents an intermediate model-type that produces reasonably fast and
accurate solutions. In this paper, we describe the QUIC-CFD model and evaluate its per-
formance using wind measurements from an isolated cubical building experiment, a 7 × 11
cubical building array experiment, and the Joint Urban 2003 Oklahoma City field experiment.

2 Introduction to the QUIC-CFD model

2.1 Solution of the steady-state Reynolds-averaged Navier–Stokes equations

This model extends CFD techniques to an urban area by solving the 3D Reynolds-Aver-
aged Navier–Stokes (RANS) equations for incompressible flow using a simple zero equation
(algebraic) turbulence model based on Prandtl’s mixing length theory. The RANS equations
are solved explicitly in time until steady state is reached using a projection method [36]. At
each time step of the projection method, the divergence-free condition is not strictly satis-
fied to machine precision levels, and it was hypothesized that when steady state is reached,
the incompressibility would be recovered. It will be shown that this method shares similar
features with the artificial compressibility method [10].
The projection method is a popular approach for solving the incompressible Navier–Stokes
equations that involves the solution of a Poisson equation for pressure. In this method, an
elliptic pressure-Poisson equation is derived from the momentum and continuity equations.
The solution of the resulting Poisson equation is the greatest computational expense of this
class of methods [36].
The method of artificial compressibility [10] involves the addition of a pseudo-time deriv-
ative to the equation set. In this formulation, the continuity equation is modified by adding a
time-derivative of the pressure term resulting in Eq. 1:
1 ∂P ∂u i
=− , (1)
β ∂τ ∂ xi
where P is pressure, u i is the velocity component in the xi direction using standard index nota-
tion and β is an artificial compressibility or a pseudo-compressibility parameter. Together with
the unsteady momentum equations, this forms a hyperbolic–parabolic type system of time-
dependent equations. Equation 1 and the momentum equation are simultaneously marched
out in τ until a steady solution is obtained. The incompressibility condition ∂u ∂ xi = 0 is vio-
i

lated during the artificial transient state but is satisfied at steady state when ∂ P/∂τ = 0.
Note, that τ no longer represents a true physical time in this formulation.
Qualitatively, the addition of the transient pressure term leads to the pressure being dis-
tributed in the incompressible flow field with finite speed waves. Thus, instead of the pressure
field being affected instantaneously by any disturbances in the flow field, there is a time lag
between the flow disturbance and its effect on the pressure field. In the limit of incompress-
ible flow, the wave-speed approaches infinity, whereas the speed of propagation of these
pseudo-waves depends on the magnitude of the artificial compressibility parameter β. The
magnitude of the artificial compressibility parameter β is usually chosen to be a large value
to recover the incompressibility condition quickly.
Peyret and Taylor [38] note that the artificial compressibility method of Chorin [10] can be
considered an alternative technique to solve the steady-state Navier–Stokes equations using
a pressure Poisson equation (e.g., the fractional step method). To illustrate this, consider the
temporally discretized form of Eq. 1

123
442 Environ Fluid Mech (2011) 11:439–464

1  n+1  ∂u n+1
p − p n = −β i (2)
τ ∂ xi

and the temporally discretized momentum equation


 
1 ∂ pn
u in+1 = u in + t Din − Ain − . (3)
ρ ∂ xi

where Di and Ai are the diffusion and the advection in the ith-direction, respectively, and n
is the nth time step. Taking the divergence of Eq. 3,

∂u in+1 t ∂ 2 p n ∂  n 
=− + u i + t (Din − Ain ) (4)
∂ xi ρ ∂ xi 2 ∂ xi

and substituting Eq. 4 into Eq. 2 yields

1  n+1  βt ∂ 2 p n ∂  n 
p − pn = −β u + t (Din − Ain ) .
τ ρ ∂ xi2 ∂ xi i

From the predictor stage for solving the momentum equations, we have,

u i∗ = u in + t (Din − Ain )

and therefore,

1  n+1  βt ∂ 2 p n ∂u ∗
p − pn − = −β i . (5)
τ ρ ∂ xi 2 ∂ xi

At steady state p n+1 = p n ; the above equation reduces to the pressure Poisson equation used
in the fractional step method, namely:

∂ 2 pn ρ ∂u i∗
= . (6)
∂ xi2 t ∂ xi

Note that if the projection method is used to compute a steady state solution of the Na-
vier–Stokes equation and if, at each time step, the Poisson equation for pressure is not solved
exactly, but rather only one Jacobi iteration is executed, the method becomes identical to
the artificial compressibility method (Eq. 5) for βτ = 0.25ρ2 /t (for a 2D case, see
appendix), where  = x = y. To show this, we start with the discretized form of Eq. 5
for the 2D case. Here, we use the 2D case for clarity, but this can be easily extended to the
3D case:
βtτ n ∗
pi,n+1
j − pi, j −
n
( pi+1, j + pi−1,
n
j + pi, j+1 + pi, j−1 − 4 pi, j ) = −βτ ∇ · u
n n n
ρ2
4βtτ n βtτ n ∗
pi,n+1
j − pi, j +
n
pi, j = ( pi+1, j + pi−1,
n
j + pi, j+1 + pi, j−1 )−βτ ∇ · u .
n n
ρ2 ρ2

For the special case of βτ = 0.25ρ2 /t, Eq. 5 can reduce to the pressure Poisson
equation stencil Eq. 7:

123
Environ Fluid Mech (2011) 11:439–464 443

4(0.25ρ2 /t)t n
pi,n+1
j − pi, j +
n
pi, j =
ρ2
(0.25ρ2 /t)t n ρ2
( p + p n
+ p n
+ p n
− ∇ · u∗)
ρ2 i+1, j i−1, j i, j+1 i, j−1
t
 
n+1 ρ2 ∗
pi, j = 0.25 pi+1, j + pi−1, j + pi, j+1 + pi, j−1 −
n n n n
∇ ·u . (7)
t
Detailed analysis of the maximum time step and required pressure iteration for the pressure
Poisson solver has been given in the appendix.

2.2 Turbulence model

The Reynolds-Averaged Navier–Stokes (RANS) modeling approach has been widely used
for modeling turbulence in practical flows. Two-equation RANS models (k − ε, k − ω) have
been a very popular approach for such flows. However, these turbulence models have been
shown to give poor predictions in very complex geometry, unsteady flows, and separated
flows. For these reasons, and in attempt to minimize computational time, we hypothesize
that sufficiently accurate results can be obtained by using simple algebraic models.
In QUIC-CFD, a simple eddy viscosity approach is used to close the following set of
RANS equations for the motion of incompressible air without body forces in a non-rotating
coordinate system:
Continuity equation:
∂u i
=0 (8)
∂ xi
Momentum equation:
  
∂u i ∂u i u j 1 ∂p ∂ ∂u i ∂u j ∂u i u j
=− − + ν + − (9)
∂t ∂x j ρ ∂ xi ∂x j ∂x j ∂ xi ∂x j
where u i is the mean velocity in the ith direction (where i = 1, 2 and 3), u i is the turbulent
fluctuating velocity, p̄ is the mean pressure, ρ is the average density, u i u j are the turbulent
Reynolds stresses, and ν is the kinematic viscosity of the fluid.
For this work, the Reynolds stresses have been modeled using an eddy viscosity that is
evaluated using a zero equation algebraic turbulence model. Algebraic models are relatively
easy to implement and have been known to perform reasonably well for mixing layer prob-
lems. It has also been observed that algebraic models help ensure computational stability
[51]. However, it is generally accepted that mixing length models do not work well for sep-
arated flows ([51], Anderson [2]; Pope [40]). Despite this, results from the Baldwin-Lomax
algebraic model (a simple mixing length model) and a two-equation k-ω model [51] appear
acceptable. Driver [16] evaluated k-ω model for a separated flow and found that there was
reasonable agreement between the model and experimental data for the skin friction coeffi-
cient (C f ∞ ) and the coefficient of pressure (C p ). Hence, an algebraic model was chosen for
the present study because it provides a balance between stability, accuracy and efficiency.
Based on the assumption that there exists an analogy between the action of viscous stresses
and Reynolds stresses on the mean flow, a simplified zero equation algebraic turbulence model
based on Prandtl’s mixing length theory was used [41]:
 
∂u i ∂u j 2
τi j = u  i u  j = −νt + + δi j κ (10)
∂x j ∂ xi 3

123
444 Environ Fluid Mech (2011) 11:439–464

2 0.2
1.8 0.18
1.6 0.16
1.4 0.14
0.16
1.2 0.12 0.2 0.12
0.2 0.08
z/H

1 0.1
0.8 0.08
0.16
0.6 0.12 0.16 0.06
0.08
0.12
0.4 0.2 0.04
0.08 0.16
0.2 0.12 0.02
0.08
0 0
−1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5
x/H

Fig. 1 Contours of non-dimension mixing length in the x–z plane for a wall mounted cube

where κ is the turbulent kinetic energy and νt is the turbulent kinematic viscosity

νt = lmi
2
x Si j Si j . (11)
Here, the mixing length lmi x = kymin , where ymin is the shortest distance from any building
wall or the ground, k is the von Karman constant and
 
1 ∂u i ∂u j
Si j = +
2 ∂x j ∂ xi
is the rate of strain tensor.
Figure 1 shows the variation of non-dimensional mixing length (lmi x /H ) in x–z plane for
a wall-mounted cube. The value ymin is capped at a value of half the building height (or half
the average building height in case of flow over multiple buildings) under the assumption
that the maximum length scale of the turbulent eddies produced in such flows will be of the
order of building height.

2.3 Grid generation

Another important feature of our QUIC-CFD model is the grid generation technique. High
quality grids are required by high fidelity CFD models and their grid generation techniques
are often time consuming especially when dealing with complex geometries like buildings
in urban areas. For the QUIC-CFD model, a 3D matrix of zeros and ones (zeros for solid
cells and ones for fluid cells) is used to define the buildings on a simple uniform structured
grid. The built-in pre-processor converts Environmental Systems Research Institute (ESRI)
shape files into the required format in seconds as shown in Fig. 2.

2.4 Model description and solution procedure

Using the approach to model turbulence described above, the momentum equation (Eq. 9)
becomes:
  
∂u i ∂u i u j 1 ∂p ∂ ∂u i ∂u j
=− − + (ν + νt ) + . (12)
∂t ∂x j ρ ∂ xi ∂x j ∂x j ∂ xi
The Navier–Stokes equations were solved on a staggered mesh using a finite volume tech-
nique. The discretization schemes were second-order accurate in space (central difference)

123
Environ Fluid Mech (2011) 11:439–464 445

(a)

(b)
Fig. 2 a 3D figure of a sample building dataset obtained in ESRI shape file format. b Grid generated for the
QUIC-CFD model with d x = dy = dz = 5 m

and time (Adams-Bashforth). The law of the wall was imposed at all of the solid surfaces.
The law of the wall boundary condition was imposed at the rigid surface by applying a free
slip boundary condition at the surface. The tangential shear stress was forced to be equal to
u2∗ . The value of u∗ was evaluated using a log-law (u ∗ = uk/ ln(0.5z/z o )), where u is the
magnitude of the tangential velocity and z 0 is the surface roughness.
The pressure Poisson equation was solved using the successive over-relaxation method
(SOR via the methodology described above). A free slip condition was used at the top and
side boundaries. The following outflow boundary condition was prescribed at the outlet:

∂φ ∂φ
= Ub
∂t ∂n

123
446 Environ Fluid Mech (2011) 11:439–464

where, Ub is a velocity that is independent of the location of the outflow surface and is
selected so that an overall mass balance is maintained [17]. This boundary condition allows
the convection of structures out of the domain and avoids problems with reflection of pressure
waves back to the interior of the domain.

3 Results and discussion

The QUIC-CFD results were compared with available wind-tunnel data for an isolated cube,
a more complex 7 × 11 array of cubes and field data obtained during the Joint Urban 2003,
Oklahoma City field experiment. For both wind-tunnel cases, the incident wind direction was
normal to the cubes.

3.1 Normal incident flow over an isolated cube

The experimental setup of Snyder and Lawson [46] consisted of a smooth cube of height 0.2 m
immersed in a simulated atmospheric boundary layer that was approximately 1.8 m deep. The
boundary-layer velocity profile was well characterized by a power-law profile with an expo-
nent of 0.16 and a free stream value of 4 m/s which produced a flow with the Reynolds
number (based on cube height and free stream velocity) approximately equal to 40,000. A
log-law profile was found in the surface layer with a friction velocity, u ∗ = 0.05 m/s and
a roughness length, z 0 = 1 mm. Measurements were obtained along the centerline in the
x–z plane and along ground-level in the x–y plane. Further details of this experiment can
be found in Snyder and Lawson [46]. This case was simulated using the QUIC-CFD model.
The grid cell sizes were uniform and set to H/10 (H = 0.2 m, building height) so that flow
features associated with the building were adequately resolved and the domain size was set
to Lx = 10 H, Ly = 5 H and Lz = 3 H. For this case, a grid resolution study was performed
using three levels of grid refinement having 5, 10 and 20 cells on the cube respectively. The
results did not dramatically change, and most of the important features associated with flow
over a bluff body were captured by all the three grids (see Fig. 3). However, the simulations
on the coarsest grid and medium grid (5 and 10 cells on the cube) were unable to capture the
recirculation on the rooftop, while the finest grids captured this effect.

3.1.1 Qualitative results

Figure 4a, b are velocity vector plots for an isolated cube in the vertical x–z plane along
the centerline and horizontal x–y plane near the ground respectively. The results from the
QUIC-CFD model are overlaid with wind-tunnel data. From both of these figures, it can be
observed that the QUIC-CFD model is able to predict the important flow features reasonably
well. The location of stagnation point on the front wall and the dimension of the upwind
cavity are very well predicted. On the back wall, the near wake vortex and the length of the
reattachment point are also predicted fairly well. On the rooftop, the experimental data shows
strong shear, which is not observed in the model results, whereas on the sidewalls, the model
over predicts the shear.

3.1.2 Quantitative results

Figure 5a, b are vertical line plots of stream-wise velocity for an isolated cube in x–z plane
and horizontal line plots of stream-wise velocity for a single isolated cube in x–y plane for

123
Environ Fluid Mech (2011) 11:439–464 447

1.2

0.8
z/H

0.6

0.4

0.2

0
−1 −0.5 0 0.5 1 1.5 2
x/H

Fig. 3 Effect of grid resolution on the streamwise component of velocity in a vertical xz plane (thick line
fine, solid line with filled circles medium, dot-dash line coarse and open circle wind-tunnel data)

1.5 u= 1m/s

1
z/H

0.5

0
−3 −2 −1 0 1 2 3
x/H
(a)
2

1.5
u= 1m/s

0.5
z/H

−0.5

−1

−1.5

−2
−3 −2 −1 0 1 2 3
x/H
(b)
Fig. 4 Velocity vectors from the QUIC-CFD computations (Gray arrow) overlayed with wind-tunnel data
(black arrow) for an isolated cube with inflow wind normal to the building face: a vertical slice (xz plane)
along the centreline (y = 0), b horizontal slice (xy plane) at the ground-level (z = 0.1 H)

123
448 Environ Fluid Mech (2011) 11:439–464

1.5 u= 1m/s

z/H
1

0.5

−3 −2 −1 0 1 2 3
x/H
(a)
2

1.5 u= 1m/s

0.5
y/H

−0.5

−1

−1.5

−2
−3 −2 −1 0 1 2 3
x/H

(b)
Fig. 5 Line plots of the stream-wise velocity at various x values comparing the Q-CFD results (thick line) and
wind-tunnel data ( f illed cir cle) for an isolated cube with inflow wind normal to the building face: a vertical
slice (xz plane) along the centreline (y = 0), b horizontal slice (xy plane) at the ground-level (z = 0.1 H)

the QUIC-CFD model overlaid with wind-tunnel data. Figure 5a clearly shows that the model
was unable to predict the strong shear on the rooftop as seen in the experimental data. It is also
observed that the stream-wise component of velocity is slightly under predicted in the wake
region. Figure 5b shows that the stream-wise component of velocity is over predicted near the
sidewalls. Overall, there is a reasonable agreement between the model and the wind-tunnel
data.
The turbulence model was also evaluated for flow over a wall-mounted cube. Since the
turbulence model calculates only the cross tensor term of Reynolds stress explicitly, only
these values (u  w  ) are compared with available DNS data [52] in Fig. 6. It can be observed
the Reynolds stress value in the wake is predicted reasonably well by the QUIC-CFD model.
The QUIC-CFD model under predicts the value on the rooftop. Large value of Reynolds
stress, similar to other eddy viscosity based turbulence model [47], can also be observed at
the edge of the rooftop because of the dominance of diagonal terms in the stress tensor in
these regions. This leads to overestimation of the eddy viscosity in this region, thus under
predicting the shear.
Despite the known deficiencies in the algebraic model, this model provides a very efficient
way of modeling turbulence, thus, is well-suited for the applications presented in this study.
A performance evaluation of the pressure solver for this test case is shown in Fig. 7. The
same simulations were performed varying the number of iterations in the pressure solver. It

123
Environ Fluid Mech (2011) 11:439–464 449

2
1.8 0.001

1.6
0.001 −0.01
1.4
−0.01 −0.02
1.2 −0.02 −0.03
0.001
−0.05
z/H

1 −0.03 −0.02
−0.01
−0.02 −0.01 −0.03
−0.02
0.8
0.001
0.6
0.001 0.001 −0.01
0.4
−0.01
0.2
0.001
0
−1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5
x/H
(a)
2
1.8
1.6
1.4
−0.05 −0.03 −0.01
1.2 −0.07
z/H

1
0.8 −0.02

0.6
0.4
0.2
0
−1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5
x/H
(b)
Fig. 6 Contours of u  w  in the x–z plane for a wall-mounted cube from the a QUIC-CFD model and b DNS
data (Yakhot et al. 2006)

was observed that the term ∂ P/∂τ converges quickly with increasing number of iterations.
However, after a certain number of iterations, the term ∂ P/∂τ converged for all the three
simulation, indicating that once the steady sate is reached, the incompressibility is recovered.

3.2 Cross flow over an array of cubical buildings

The second test case was based on a wind-tunnel experiment [3] of an array of 7 × 11 cubes
with H = 0.15 m and a stream-wise spacing S = H and is shown in Fig. 8. According to the
criteria of Oke [33], the 7×11 array should be in the skimming flow regime. In the experiment,
the reference velocity was 3 m/s at z = H, which produced a flow with a Reynolds number
(based on cube height and reference velocity) of ∼30,000, which is well above the critical
value required for Reynolds number independence [8]. Measurements were also obtained
along the centerline in the x–z plane and along the x–y plane at various heights. Further
details of this experiment can be found in Brown et al. [3].
For the simulations, an inflow profile was specified in the QUIC-CFD model using a
power law with exponent set to 0.16 to match the experimental data. The inflow winds were
perpendicular to the building face. The grid cell sizes were uniform and the value was set

123
450 Environ Fluid Mech (2011) 11:439–464

0
10 0
Pressure iterations=10 10
Pressure iterations=100 Pressure iterations=10
−1 Pressure iterations=1000 Pressure iterations=100
10

Maximum divergence
−1 Pressure iterations=1000
10
−2
10 −2
|| ∂P/∂τ ||2

10
−3 −3
10 10

−4 −4
10 10
−5 −5
10 10

−6 −6
10 0 1 2 3
10 0 1 2 3
10 10 10 10 10 10 10 10
Time iterations Time iterations
(a) (b)
Fig. 7 Variation of the a L 2 norm of the term ∂ P/∂t and b the maximum divergence as a function of time
iteration

z
Inflow
x H H

H
x
H H

Fig. 8 Schematic of the wind-tunnel configuration for a 3D array of 7 × 11 cubes. Top panel—side view,
bottom panel—plan view

123
Environ Fluid Mech (2011) 11:439–464 451

2
1.8
1.6 u = 1 m/s
1.4
1.2

z/H
1
0.8
0.6
0.4
0.2
0
−2 −1 0 1 2 3 4 5
x/H
(a)

1
u = 1 m/s
0.8
0.6
0.4
0.2
y/H

0
−0.2
−0.4
−0.6
−0.8
−1

0.5 1 1.5 2 2.5 3 3.5 4 4.5


x/H
(b)
Fig. 9 Velocity vectors from the QUIC-CFD computations (gray arrow) overlayed with wind-tunnel data
(black arrow) for a 7 × 11 array of cubes with inflow wind normal to the building face: a vertical slice (x–z
plane) along the centreline (y = 0), b horizontal slice (x–y plane) near the ground (z = 0.1 H)

to H/10 so that flow features associated with the building were adequately resolved and the
domain size was set to Lx = 25 H, Ly = 25 H and Lz = 3 H.

3.2.1 Qualitative results

Figure 9a, b are velocity vector plots in the x–z plane along the centerline (y = 0) of the array
and in the x–y plane along z = 0.5 H for the QUIC-CFD model overlaid with wind-tunnel
data. From both of these figures, it can be observed that the QUIC-CFD model is able to
predict the important flow features reasonably well. The location of stagnation point and the
dimension of the upwind cavity on the front wall of the first cube are very well predicted. The
model is also able to predict the central canyon vortex in the vertical plane and the presence
of lateral vortices in the horizontal plane quite closely. The model predicted location of these
vortices also compares well with the wind-tunnel data. However, it is again observed the
model is unable to predict the strong shear on the rooftop on the first cube as observed in the
experimental data.

3.2.2 Quantitative results

Figure 10a, b are vertical line plots of stream-wise velocity in the x–z plane along the cen-
terline (y = 0) of the array and horizontal line plots of stream-wise velocity along z = 0.5 H

123
452 Environ Fluid Mech (2011) 11:439–464

2
1.8
1.6
1.4 u = 1m/s
1.2

z/H
1
0.8
0.6
0.4
0.2
0
−2 −1 0 1 2 3 4 5
x/H
(a)

1
0.8 u = 1 m/s

0.6
0.4
0.2
y/H

0
−0.2
−0.4
−0.6
−0.8
−1

0.5 1 1.5 2 2.5 3 3.5 4 4.5


x/H
(b)
Fig. 10 Line plots of the stream-wise velocity at various x values (streamwise location) comparing the
Q-CFD results (thick line) and wind-tunnel data ( f illed cicle) for a 7 × 11 array of cubes with the inflow
wind normal to the building face: a vertical slice (x–z plane) along the centreline (y = 0), b horizontal slice
(xy plane) near the ground (z = 0.1 H)

in the x–y plane for the Q-CFD model overlaid with wind-tunnel data. As mentioned earlier,
the model is unable to predict the large shear on the rooftop of the first building. Also, in the
horizontal plane, the model seems to over predict the velocity in the channel. However, the
model predicts the velocity in the street-canyon and upwind region reasonably well.

3.3 Joint Urban 2003 field experiment

The Joint Urban 2003 field experiment was performed in July 2003 in Oklahoma City, U.S.A.
A large number of meteorological instruments and tracer samplers where deployed through-
out Oklahoma City. Meteorological measurements were taken at over 160 different locations
[1], while tracer (SF6 ) measurements were made at over 130 locations [11]. Ten intensive
observational periods (IOPs) were conducted for both daytime and nighttime periods in which
most meteorological and tracer sampling instrumentation were operational. During the IOPs,
the winds were predominantly from the south. Further details about the experiment, instru-
ment types and locations, and tracer release information can be found in Allwine et al. [1],
Clawson et al. [11], Flaherty et al. [18], Nelson et al. [31], and Brown et al. [6].
For this study, we have focused on the second trial of IOP 8 conducted during night-
time. The winds were predominantly from the south-south-west direction. The portable wind
detector at the Post Office (PWID 15), a propeller anemometer was used for determining a
scaling velocity and wind direction for the computation’s inflow wind profile (see Fig. 11).

123
Environ Fluid Mech (2011) 11:439–464 453

6
x 10
3.9265

3.926
UTM NORTHING (m)

3.9255

DPG
ITT
UoU
OU
LANL
VOLPE
DSTL
PWID 15
3.925
6.34 6.342 6.344 6.346 6.348 6.35 6.352
5
UTM EASTING (m) x 10

Fig. 11 Map indicating the placement of various wind sensors in downtown Oklahoma City during Joint
Urban 2003. PWID 15 was used to specify the inflow boundary condition

It was located ∼1 km upstream (south) of the central business district (CBD) at 50 m above
ground on a 35 m rooftop tower free from building effects. The inflow boundary condition
for the model was specified using a log-law profile. Thirty-minute average wind speed and
direction data from PWID 15 were used as a reference velocity. A good fit was obtained
using z o = 0.1 m and a wind speed (PWID 15) of 7 m/s (u* = 0.45 m/s). The results from
the model shown below have been compared with 30 min averages from the portable wind
detectors (PWIDs), super portable wind detectors (SPWIDs) and sonic anemometers shown
in Fig. 11. These wind sensors were placed in the CBD areas as shown in Fig. 11 and were
at 8 m above ground level (AGL).
The QUIC-CFD modeling domain covers most all of the Oklahoma City CBD and is
1.3 km × 1.3 km in size. The horizontal grid size was set to 5 m, while the vertical resolution
was set to 3 m resulting in a 256 by 263 by 64 grid cell domain (4.3 million cells total) (see
Fig. 2). The 3D building data were obtained from the Defense Threat Reduction Agency and
the University of Oklahoma.

123
454 Environ Fluid Mech (2011) 11:439–464

900

800

700

600
Y (m)

500

400

u= 3 m/s

300

200
300 350 400 450 500 550 600 650 700 750 800 850
X (m)

Fig. 12 Velocity vectors from the QUIC-CFD results (gray arrow) overlayed with 30 min averaged field data
(black arrow) for IOP 8 during Joint urban 2003 field experiment: horizontal slice (xy plane) at 8 m AGL
(vectors are to scale)

3.3.1 Qualitative results

Figures 12 and 13 shows velocity vector plots for flow around Oklahoma City in the x–z
plane at 8 m AGL. The results from the QUIC-CFD model are overlaid with field data. From
this figure, it can be observed that the QUIC-CFD model is able to predict the important flow
features reasonably well. The model predicts a channeling effect along north-south running
streets and is able to predict high velocity in these regions. The model also predicts the
expected reversed flow in the street canyon and building wake regions of the domain. The
model produced velocity in the intersection areas are in good agreement with the field data.
However, one can note the poor agreement in the South-East corner of domain, especially
between Y = 600 m and Y = 650 m and X = 500 m and X = 650 m. This disagreement can
be attributed to inconsistency in the detailed building geometry file. The building north of
the above mentioned region is actually a parking lot, which allows the flow to pass through
it. Also, much of detailed information of the building geometry is not available, such as

123
Environ Fluid Mech (2011) 11:439–464 455

800

750

700
Y (m)

u= 3 m/s
650

600

550
450 500 550 600 650 700
X (m)
Fig. 13 Velocity vectors from the QUIC-CFD results (gray arrow) overlayed with 30 min averaged field data
(black arrow) for IOP 8 during the Joint Urban 2003 field experiment for qualitative comparison: horizontal
slice (xy plane) at 8 m AGL (vectors are to scale)

openings in parking garages, trees, sky bridges, etc. These are important features that can
locally affect the flow field.

3.3.2 Quantitative results

Figure 14a, b shows a scatter plot comparing observed wind speed and direction with
predicted wind speed and direction. Figure 14a clearly shows that for most of the sensors,
the model is able to predict the wind direction reasonably well. Approximately 87% of the
simulation results are in the same quadrant as that of the experimental value, while more than
60% are within 45 degrees of the experimental value. It is also observed that more than 40%
of the simulation results are with 10 degrees of the experimental value. Though there are
only a few sensors in the reverse or cross flow region, the model compares well with those
sensors. This indicates that the model is able to capture the bulk advection affect.
The wind speeds results shown in Fig. 14b indicate a fair degree of scatter. However, the
model appears to compare better for higher wind speeds. The model does not compare well
in the low wind speed region as in these regions usually the turbulent intensity is high. More
than 60% of the simulated wind speeds are within the factor of 2 criteria. About 50% of the
simulated results have an error of 50% or less, while only 20% of the simulation results have
an error of less than 10%.
Tables 1 and 2 gives further quantitative analysis of results in terms of fractional bias
(FB), Mean Error, the normalized absolute difference (NAD) and the bounded normalized
mean square error (BNMSE) for wind direction and wind speed respectively. These quantities
were calculated using the following equations following Warner et al. [50]. The predicted or
simulated value is denoted as COM and the observed or measured value is denoted as OBS.
(OBS − COM)
FB =
0.5(OBS + COM)

123
456 Environ Fluid Mech (2011) 11:439–464

Error <= 30o


Error <= 90o
Error <= 10o

Error <= 45o

(a)

Error <=10%

Error <=25%

Error <=50%

Error <=100%

(b)
Fig. 14 Half-hour averaged wind-speeds and wind-directions for IOP 8, Trial 2: a paired in time and space
scatter plot for predicted and observed wind direction (degrees) and b paired in time and space scatter plot for
predicted and observed wind speed (m/s)


n
MeanError = |OBS(i) − COM(i)|
i

n
|OBS(i) − COM(i)|
NAD =
ni
i [OBS(i) + COM(i)]

n
[OBS(i) − COM(i)]2
BNMSE =
in
i [OBS(i) + COM(i)]
2

It can be seen from Table 1, that the mean error in the prediction of the wind direction is
38.1◦ indicating that the model is able to predict most of the flow features fairly accurately.
Table 2 shows the performance of the model with respect to wind speed. A positive value of
fractional bias (FB) indicates that the model under predicts the wind speed. The mean error
∼0.8 m/s, thus the models is able to predict the wind speeds accurately within 1 m/s.

123
Environ Fluid Mech (2011) 11:439–464 457

Table 1 Quantitative error analysis of wind direction computations

Error ≤90◦ Error ≤45◦ Error ≤30◦ Error ≤15◦ Mean error (deg.) NAD BNMSE

87.5% 62.5% 57.1% 42.9% 38.1 0.1416 0.0316

Table 2 Quantitative analysis of wind speed

Error ≤100% Error ≤50% Error ≤25% Error ≤10% FB Mean error (m/s) NAD BNMSE

96.3% 56.3% 29.1% 20.0% 0.156 0.816 0.22 0.053

4 Comparison of QUIC-CFD model with a fast response 3D diagnostic wind model

In this section, we compare the 3D velocity field from QUIC-CFD to the simple very fast
running 3D diagnostic wind model QUIC-URB [44]. Winds produced from the two models
as well as simulation times are briefly compared. A more detailed comparison of the two
models may be found in Neophytou et al. [32]. QUIC-URB is based on the dissertation of
Röckle [42] in which a mass consistent diagnostic wind model for computing the 3D flow
field around buildings was developed. In this approach, an initial wind field u o is prescribed.
For isolated buildings, QUIC-URB utilizes isolated building empirical algorithms for deter-
mining the initial wind fields of the vortex regions associated with urban flow. Once the flow
field modifications have been made, the flow field is forced to be mass consistent with the
weak constraint that the differences between the initial and final velocity fields be minimized
(e.g., Kaplan and Dinar [25]).
Flow in Oklahoma City (Sect. 3.3) was again simulated using both the QUIC-CFD and
QUIC-URB models. Both the models were run under exact same conditions using 4.3 million
grid points. The QUIC-URB simulation took 56 s while the QUIC-CFD simulation took 726 s
on two cores (both the models were optimized on multi-core processors using OpenMP).
While QUIC-URB was approximately ten times faster than QUIC-CFD, the QUIC-CFD
model appeared to perform much better than the diagnostic model in terms of capturing
key flow features in an urban area. Figure 15 shows velocity vector plots for flow within
Oklahoma City in the x–y plane at 8 m agl overlaid with contours of horizontal velocity
magnitude. The results from the model are also overlaid with the field data. It can be clearly
seen that QUIC-CFD is able to predict important effects like channeling along the north-
south running streets while QUIC-URB is not able to predict the strong acceleration in these
regions. In the Park Avenue Street Canyon the flow coming from the south-east impinges on
the buildings aligned on the north side of the street and forms a divergence zone (Fig. 16).
QUIC-CFD predicts this phenonmenon reasonably well while the QUIC-URB model is not
able to parameterize this effect correctly.
Figure 17 shows the contours of vertical velocity for both of the models. QUIC-URB
does not have any parameterizations for vertical velocity in the wake region on the building.
The vertical velocity is produced via the mass consistency solver. It has been observed that
vertical velocities calculated in QUIC-URB are usually under predicted [35]. However, it
has been observed in field experiments [31] and wind-tunnel experiments [27] that strong
updrafts and downdrafts are produced in the vicinity of tall buildings. In Fig. 17, building A
is a tall building (∼ 150 m), and it can be seen that QUIC-URB produces an updraft which
is three times smaller than the updraft produced by QUIC-CFD. It should be noted that

123
458 Environ Fluid Mech (2011) 11:439–464

900 900

800 800

700 700

600 600
Y (m)

Y (m)
500 500

400 400

u= 3 m/s u= 3 m/s
300 300

200 200
300 350 400 450 500 550 600 650 700 750 800 850 300 350 400 450 500 550 600 650 700 750 800 850
X (m) X (m)

Fig. 15 Velocity vectors (gray arrow) and contours of horizontal velocity magnitude (m/s) from a QUIC-CFD
and b QUIC-URB overlaid with 30 min averaged field data (black arrow) for IOP 8 during the Joint Urban
2003 field experiment: horizontal slice (xy plane) at 8 m AGL (vectors are to scale)

800
800

750
750

700
700 PARK AVE. PARK AVE.
Y (m)
Y (m)

u= 3 m/s u= 3 m/s
650 A 650 A

600 600

550 550
450 500 550 600 650 700 450 500 550 600 650 700
X (m) X (m)

Fig. 16 Velocity vectors (gray arrow) and contours of horizontal velocity magnitude (m/s) from a QUIC-CFD
and b QUIC-URB overlaid with 30 min averaged field data (black arrow) for IOP 8 during Joint Urban 2003
(vectors are to scale)

while sonic anemometer field data are not available in this updraft region. During the Joint
Urban 2003 experiment, neutrally buoyant balloons were released and the vertical rise of
balloons in the wake of this building were crudely timed and found to be much greater than
1 m/s.

5 Computational effort

Figure 18 shows the comparison of computational effort required by QUIC-CFD for various
problems. Three cases used for the validation study were also compared as a part of this
work: flow over a single isolated cube with 0.12 and 0.5 million nodes, flow over an array
of 7 × 11 cubes with 2 million nodes and a field experiment in Oklahoma City during Joint

123
Environ Fluid Mech (2011) 11:439–464 459

900 1 900 1

0.8 0.8
800 800
0.6 0.6

700 0.4 700 0.4

A 0.2
A 0.2
600 600
Y (m)

Y (m)
0 0

500 500
–0.2 –0.2

400 –0.4 –0.4


400

–0.6 –0.6
u= 3 m/s u= 3 m/s
300 300
–0.8 –0.8

200 –1 200 –1
300 350 400 450 500 550 600 650 700 750 800 850 300 350 400 450 500 550 600 650 700 750 800 850
X (m)
X (m)

Fig. 17 Velocity vectors (gray arrow) and contours of vertical velocity (m/s) from a QUIC-CFD and b
QUIC-URB overlaid with 30 min averaged field data (black arrow) for IOP 8 during Joint urban 2003 field
experiment: horizontal slice (xy plane) at 8 m AGL (vectors are to scale)

Fig. 18 Comparison of the computational effort (in seconds) required by QUIC-CFD for various cases

urban 2003 (1.3 × 1.3 × 0.18 km) with 4.3 and 6.25 million nodes. These simulations were
performed using a CFL number of 0.3, and the simulations were stopped when steady-state
solutions were achieved. The computational study was performed on a 2.66 Ghz Intel Core
2 Duo (Processor number E8200) with 4 Gb of 1066 MHz DDR3 SDRAM and was com-
piled using the Gfortran compiler. Current technology has enabled even the most common
computer to have multiple processors. For this reason, the QUIC-CFD code was parallel-
ized using OpenMP (OpenMP Standard) to take advantage of multiple processors/cores on
a shared memory platform. Since this model is intended for use for laptops, currently the
model is only parallelized with OpenMP. However, further efforts are being made to develop
a MPI (message passing interface) version to harness the capabilities of massive clusters.

123
460 Environ Fluid Mech (2011) 11:439–464

Figure 18 shows that simulations using both cores from the above-mentioned processor were
nearly twice as fast as the single processor simulations. As an additional comparison, the
Realistic Urban Spread and Transport of Intrusive Contaminants (RUSTIC) model is another
fast running urban flow model which quickly converges to a numerical solution of a modified
set of the compressible Navier–Stokes equations. RUSTIC uses the k–ω turbulence model
(Diehl et al. [15]). The model took ∼7000 s to produce a converged solution on a 2.2 GHz
Pentium 4 desktop for a domain with 0.5 million nodes.
Based on these results, it appears that the computational effort required by the QUIC-CFD
model is orders of magnitude less than that required by typical high fidelity CFD codes and
about an order of magnitude slower than most simple building-aware fast response wind
models. Unlike diagnostic wind models, QUIC-CFD model also produces turbulence and
pressure fields which can be used by dispersion models to compute the concentration of
pollutants inside and around the buildings. The projection method [10] used in this solver
enables the code to produce a mass consistent velocity filed when the solution is converged.
Since this model does not use prohibitively small time step to account for the acoustic wave
as in compressible flow solver, nor does it solve the pressure Poisson equation to machine
accuracy as in incompressible flow solver, the algorithm produces the velocity field very
efficiently.

6 Conclusions

The QUIC-CFD model described in this paper adopts CFD techniques to urban areas by
solving the 3D RANS equations. The pressure Poisson equation is not solved for complete
convergence and the turbulence closure problem is addressed by using a zero equation (alge-
braic) model based on Prandtl’s mixing length theory. Although still in validation stage, the
QUIC-CFD model shows potential as a valuable wind model for fast response modeling
purposes in an urban environment. In addition, the implementation of multi-core paralleliza-
tion enables the model to take advantage of multiple cores now commonplace in personal
computers.
From the above sections, it can be concluded that wind, pressure and turbulence fields in
urban domains can be obtained with substantially reduced computational effort and results
are fairly accurate in comparison to measurements. Unlike most fast response models (e.g.
diagnostic models), the QUIC-CFD model also computes turbulence and pressure fields that
can be used by dispersion models. Finally, for the wind-tunnel and complex urban tests dis-
cussed here, the model predicted velocity field is in fair agreement with the experimental
data.

Appendix

A Von Neumann analysis of Eq. 5 for a 2D case yields the stability restriction τ ≤
0.25ρ2 /βt [36], indicating that performing a single time advancement of the artificial
compressibility equation at the maximum allowable time step is tantamount to a single Jacobi
type iteration on the pressure Poisson equation. This 2D analysis can be extrapolated to 3D
and is done for clarity.
It is also known that the explicit solution of the compressible Navier–Stokes equations
at low Mach numbers is very inefficient due to the increasing stiffness of the equations as
the Mach number is decreased [39,9]. To retain stability, increasingly small time steps are

123
Environ Fluid Mech (2011) 11:439–464 461

required in order to capture the acoustic waves (i.e. temporal pressure disturbances), the
speed of which increases relative to the convection speed as the Mach number decreases.
Therefore, in order to be computationally efficient in reaching a steady state, a time step
should be chosen such that the convective terms (which have a substantially larger time scale
than the acoustic waves at low Mach numbers) alone are accurately represented. However,
this time step would be too large to accurately resolve the acoustic waves and would not
satisfy the Von Neumann stability restriction for Eq. 5 and the solution would be unstable.
Hence, for each physical time step (convective t), multiple time advancements (with time
step τ , which satisfies the Von Neumann stability criterion) of the time dependent pressure
equation or multiple iterations of the pressure Poisson equation should be done to ensure
better stability and convergence. This approach to rapidly achieving a steady state solution is
similar to the dual time step approach used in the artificial compressibility method [30,49].
The number of time iterations of the time dependent pressure equation or the number of
Jacobi type iterations of the pressure Poisson equation performed to ensure better stability
and convergence should be such that the total time advancement of the pressure equation is
equal to the physical time step of the flow (convective t). Thus, the number of iterations
can be evaluated as t/τ . Additional iterations can be performed so that the approach to
the incompressibility is much quicker. Substituting the formulae for t and τ from the
convectiveCourant–Friedrichs–Lewy (CFL) number criterion and Von Neumann analysis
of Eq. 5 for the 2D case yields an estimate for the minimum number of iterations. That is,
substituting

t = CFL
u max
into
0.25ρ2
τ ≤
βt
gives
0.25ρu max 
τ ≤
βCFL
Solving for t/τ gives the minimum number of Jacobi iterations (I termin ) :
 
t 4 × CFL 2 CFL 2
I termin ≥ = =4 (13)
τ ρu 2max /β M∞
According to Ferziger and Peric [17], the value of β is between 0.1 and 10 for a range of
problems. Kwak et al. [28] also suggested that a value of β in the range 0.1–10.0 will work
well for most problems. On the high side, the problem is one of stiffness, which retards
the convergence rate. On the low side, the value of βτ should be large enough to permit
pressure waves (which actually should move at infinite speed in the incompressible limit) to
propagate far enough to reasonably balance viscous effects during the artificial transient, or
the pseudo-time iterations will tend not to converge.
The minimum number
√ of iterations for most flows in general may be estimated using
scaling arguments. β can be thought of as the speed at which the pressure waves are prop-
agated in the media. In truly incompressible flow, this speed is close to infinity. However,
in the artificial compressibility method, it can be assumed that the pressure wave travels
at speeds O(10) times faster than the fluid parcel moves (Ferziger and Peric [17]). Thus,

123
462 Environ Fluid Mech (2011) 11:439–464

2 ≈ O(10−2 ) and CFL ≈ O(10−1 ). Substituting these estimates into Eq. 13,
ρu 2 /β = M∞
the minimum number of Jacobi iterations is:
I termin ≈ O(10)
Figure 15a shows the variation of change in pressure between two consecutive time step
(∂ P/∂t) as a function of time iterations on a log-log scale. It can be observed that after
sufficient number of time iterations, the term ∂ P/∂t reaches a steady state regardless of the
number of pressure iterations. However, the increase in the number of pressure iterations
makes the approach to incompressibility faster. Figure 15b shows the variation of maximum
divergence as a function of time iterations. It can again be observed that when the term ∂ P/∂t
reaches steady state, the maximum divergence also becomes small, irrespective of number of
pressure iterations. Thus, both of these figures confirm the hypothesis of the artificial com-
pressibility method. That is, although the incompressibility condition is not strictly satisfied
during the transient phase, the flow becomes incompressible when steady state is achieved.
The method described in this work provides a trivial way of converting a pressure projec-
tion method into an artificial compressibility method that is capable of producing efficient
steady state solutions of the Navier–Stokes equations for incompressible flows. Thus, this
method provides a nice balance between accuracy and efficiency which is important for many
urban dispersion scenarios as mentioned earlier. The simplicity of this implementation makes
the parallelization of the computer algorithms straight forward.

References

1. Allwine KJ, Leach MJ, Stockham LW, Shinn JS, Hosker RP, Bowers JF, Pace JC (2004) Overview of
Joint Urban 2003—an atmospheric dispersion study in Oklahoma City. In: Symposium on planning,
nowcasting and forecasting in the urban zone, AMS, Seattle, WA
2. Anderson JD (1995) Computational fluid dynamics: the basics with applications. McGraw Hill Publishing
Company, New York
3. Brown MJ, Lawson R, Decroix D, Lee R (2001) Mean flow and turbulence measurements around an array
of buildings in a wind tunnel. In: 11th AMS conference on applications of air pollution meteorology, Long
Beach, CA
4. Belcher S.E. (2005) Mixing and transport in urban areas. Phil Trans R Soc A 363:2947–2968
5. Britter RE, Hanna SR (2003) Flow and dispersion in urban areas. Ann Rev Fluid Mech 35:469–496
6. Brown MJ, Boswell D, Streit G, Nelson M, McPherson T, Hilton T, Pardyjak ER, Pol S, Ramamurthy
P, Hansen B, Kastner-Klein P, Clark J, Moore A, Felton N, Strickland D, Brook D, Princevac M, Zajic
D, Wayson R, MacDonald J, Fleming G, Storwold D (2004) Joint urban 2003 street canyon experiment.
In: Symposium on planning, nowcasting and forecasting in the urban zone, AMS, Seattle, WA
7. Camelli F, Coirier WJ, Hansen OR, Huber A, Kim S, Hanna S, Brown M (2006) An intercomparison
of four computational fluid dynamics models: transport and dispersion around Madison square garden.
American Meteorological Society. In: 6th Symposium on the urban environment, Atlanta, GA
8. Castro IP, Robins AG (1977) The flow around a surface-mounted cube in uniform and turbulent streams.
J Fluid Mech 79:307–335
9. Choi D, Merkle CL (1985) Application of time-iterative schemes to incompressible flow. AIAA J
23:1518–1524
10. Chorin AJ (1968) Numerical solution of the Navier–Stokes equations. Math Comput 22:745–762
11. Clawson KL, Carter RG, Lacroix DJ, Biltoft CA, Hukari NF, Johnson RC, Rich JD (2005) Joint urban
2003 (JU03) SF6 atmospheric tracer field tests. NOAA Technical Memorandum OARARL-254, Air
Resources Lab
12. Coirier WJ, Kim S (2006) Summary of CFD urban results in support of the Madison square garden and
urban dispersion program field tests. In: 6th symposium on the urban environment. American Meteoro-
logical Society, Atlanta, GA
13. Coirier WJ, Kim S, Chen F, Tuwari M (2006) Evaluation of urban scale contaminant transport and dis-
persion modeling using loosely coupled CFD and mesoscale models. In: 6th symposium on the urban
environment. American Meteorological Society, Atlanta, GA

123
Environ Fluid Mech (2011) 11:439–464 463

14. Dabberdt W, Ludwig F, Johnson W (1973) Validation and applications of an urban diffusion model for
vehicular pollutants. Atmos Environ 7:603–618
15. Diehl SR, Burrows DA, Hendricks EA, Keith R (2006) Urban dispersion modeling: comparison with
single-building measurements. J Appl Meteorol Clim 46:2180–2191
16. Driver DM (1991) Reynolds shear stress measurements in a separated boundary layer. In: 22nd AIAA
fluid dynamics, plasma dynamics and lasers conference, Honolulu, HI
17. Ferziger JH, Peric M (2002) Computational methods for fluid dynamics, 3rd edn. Springer, Berlin
18. Flaherty J, Allwine KJ, Allwine E (2007) Vertical tracer concentration profiles measured during the joint
urban 2003 dispersion study. J Appl Meteorol Clim 46:2019–2037
19. Fortin M, Peyret R, Temam R (1971) Résolution numérique des équations de Navier–Stokes pour un
fluide incompressible. J Mécanique 10:357–390
20. Gowardhan AA, Brown MJ, Pardyjak ER (2010) Evaluation of a fast response pressure solver for flow
around an isolated cube. Environ Fluid Mech 10: 311–328. doi:10.1007/s10652-009-9152-5
21. Hall DJ, Sharples H, Walker S, Kukadia V (2000) Attribution of pollutant concentrations on buildings
from local traffic—effects on ventilation requirements. BRE Client Report 81461
22. Hotchkiss RS, Harlow FH (1973) Air pollution in street canyons. EPA-R4-73-029
23. Irwin JS (1978) A theoretical variation of the wind profile power law exponent as a function of surface
roughness and stability. Atmos Environ 13:191–194
24. Kanda M (2007) Progress in urban meteorology: a review. J Meteorol Soc Jpn 85:363–383
25. Kaplan H, Dinar N (1996) A Lagrangian dispersion model for calculating concentration distribution
within a built-up domain. Atmos Environ 30:4197–4207
26. Kim J, Moin P (1986) Application of a fractional-step method to incompressible Navier–Stokes. J Comput
Phys 59:308–323
27. Klein P, Leitl B, Schatzmann M (2007) Driving physical mechanisms of flow and dispersion in urban
canopies. Int J Clim 27:1887–1907
28. Kwak D, Chang JLC, Shanks SP, Chakravarthy SR (1986) A three-dimensional incompressible Navier–
Stokes flow solver using primitive variables. AIAA J 24:390–396
29. Ladevéze J, Peyret R (1974) Calcul numérique d’une solution avec singularité des équations de Navier–
Stokes: écoulement dans un canal avec variation brusque de section. J Mécanique 13:367–396
30. Merkle CL, Athavale M (1987) Time-accurate unsteady incompressible flow algorithms based on artificial
compressibility. In: 8th AIAA computational fluid dynamics conference, AIAA-1987-1137, Honolulu,
Hawaii
31. Nelson MA, Pardyjak ER, Klewicki JC, Pol SU, Brown MJ (2007) Properties of the wind field within the
Oklahoma City Park Avenue street canyon. Part I: Mean flow and turbulence statistics. J Appl Meteorol
46:2038–2054
32. Neophytou M, Gowardhan A, Brown M (2010) An inter-comparison of three urban wind models using
Oklahoma City Joint Urban 2003 wind field measurements, submitted to the J Wind Eng Indus Aerodyn.
doi:10.1016/j.jweia.2011.01.010
33. Oke, TR (1987) Street design and urban canopy layer climate. Energy and Buildings 11:103–113
34. OpenMP Fortran/C Application Program Interface, http://www.openmp.org/
35. Pardyjak ER, Brown MJ (2001) Evaluation of a fast-response urban wind model-comparison to single-
building wind-tunnel data. Los Alamos National Laboratory report LA-UR-01-4028
36. Patankar SV (1980) Numerical heat transfer and fluid flow. McGraw Hill, New York
37. Patnaik G, Boris JP, Grinstein FF, Iselin JP (2003) Large scale urban simulations with the MILES approach.
In: Annual AIAA CFD conference, AIAA paper 2003-4104, Orlando, FL
38. Peyret R, Taylor TD (1983) Computational methods for fluid flow. Springer, Berlin
39. Pletcher RH, Chen K (1993) On solving the compressible Navier–Stokes equations for unsteady flows at
very low Mach numbers. AIAA Paper 93-3368
40. Pope SB (2000) Turbulent flows. Cambridge University Press, Cambridge
41. Prandtl L (1926) Application of the “Magnus effect” to the wind propulsion of ships. NACA TM 387
42. Röckle R (1990) Bestimmung der stomungsverhaltnisse im Bereich Komplexer Bebauugsstrukturen.
Ph.D. thesis, Vom Fachbereich Mechanik, der Technischen Hochschule Darmstadt, Germany
43. Schulman L, Strimaitis D, Scire J (2000) Development and evaluation of the PRIME plume rise and
building downwash model. J Air & Waste Manage Assoc 50:378–390
44. Singh B, Hansen B, Brown MJ, Pardyjak ER (2008) Evaluation of the QUIC-URB fast response urban
wind Model for a cubical building array and wide building street canyon. Environ Fluid Mech 8:281–312
45. Snyder WH (1979) The EPA meteorological wind tunnel: its design, construction, and operating charac-
teristics. EPA-600/4-79-051, Environmental Protection Agency, Res Tri Pk, NC

123
464 Environ Fluid Mech (2011) 11:439–464

46. Snyder WH, Lawson RE (1994) Wind-tunnel measurements of flow fields in the vicinity of buildings. In:
8th joint conference on application of air pollution meteorology with AWMA. American Meteorological
Society, Nashville, TN
47. Taulbee DB, Tran L (1988) Stagnation streamline turbulence. AIAA J 26:1011–1013
48. Theurer W, Baechlin W, Plate E (1992) Model study of the development of boundary layers above urban
areas. In: 8th joint conference on application of air pollution Meteorology with AWMA. American Mete-
orological Society, Nashville, TN
49. Viecelli JA (1971) A computing method for incompressible flows bounded by moving walls. J Comput
Phys 9:119–143
50. Warner S, Platt N, Heagy JF, Jordan JE, Bieberbach G (2006) Comparisons of transport and dispersion
model predictions of the mock urban setting test field experiment. J Appl Meteorol 45:1414–1428
51. Wilcox DC (1993) Turbulence modeling for CFD. DCW Industries Inc, La Canada, CA
52. Yakhot A, Liu HP, Nikitin N (2006) Turbulent flow around a wall-mounted cube: a direct numerical
simulation. Intl J Heat Fluid Flow 27:994–1009
53. Yamartino R, Strimaitis D, Messier T (1989) Modification of highway air pollution models for complex
site geometries. Data analyses and development of the CPB-3 model, vol 1. Rep No FHWA-RD-89-112

123

You might also like