8/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
1/12
An inter-comparison of three urban wind models using Oklahoma City Joint
Urban 2003 wind field measurements
Marina Neophytou a,n, Akshay Gowardhan b, Michael Brown b
a Environmental Fluid Mechanics Laboratory, Department of Civil and Environmental Engineering, University of Cyprus, Cyprusb Los Alamos National Laboratory, New Mexico, USA
a r t i c l e i n f o
Available online 22 March 2011
Keywords:
Fast-response models
Model evaluation
Large eddy simulation
Computational fluid dynamics
Empirical-diagnostic model
Field measurements
a b s t r a c t
Three wind models are compared to near-surface time-averaged wind measurements obtained in
downtown Oklahoma City during the Joint Urban 2003 Field Campaign. The models cover severallevels of computational approximation and include in increasing order of computational demand: a
mass-consistent empirical-diagnostic code, a Reynolds-averaged NavierStokes (RANS) computational
fluid dynamics (CFD) model, and a Large Eddy Simulation (LES) CFD code. The models were run with
identical inlet and boundary conditions using the same grid resolution; the choice of the specific
computational set-up reflects demands for fast-response models, although it may be a sub-optimal
choice for the more complex models. A qualitative comparison of the model-computed flow fields with
the Joint Urban 2003 wind measurements shows that all three models compare favorably to the near-
surface wind measurements in many locations, although there are often instances of winds being
calculated poorly in specific locations. The CFD models, however, had clearly superior looking flow
fields, whereas the empirical-diagnostic code produced fields that were less smoothly varying. The
inter-comparison exercise was supported by point-by-point quantitative comparisons of the wind
speed and wind direction and with statistical measures. The RANS-CFD code, for example, was within
50% of the measured wind speed 62% of the time as compared to 53% for the LES model and 49% for the
empirical-diagnostic code. For wind direction, the RANS-CFD code was within 301 of the measured
wind direction 58% of the time as compared to 50% for the LES code and 43% for the empiricaldiagnostic code. It is noticeable that throughout the various IOP cases examined, and under the specific
computational set-up used in the simulations for fast-response needs, there is no clear superiority of
one model over another. In addition, for the LES model, which in theory should provide the most
realistic representation of the flow field, it appears that further to the sub-optimal computational set-
up, as well as the uncertainty and natural variability persistent in the real world, has resulted in
diminished performance.
& 2011 Elsevier Ltd. All rights reserved.
1. Introduction
Simulating urban wind flow for fast-response applications
poses significant challenges for computational modeling; on one
hand both the inhomogeneity of the urban geometry and thecomplexity of the resulting flows urge for more sophisticated
models, on the other, however, more advanced or detailed models
are associated with substantial increase in computational time
that nearly removes the model application from the context of
fast response. Expected or accepted run-up times range from a
few seconds to tens of minutes.
The complexity of the flows that develop in cities, around
and in-between buildings lies in the three-dimensionality of the
resulting flows (Boris et al., 2001). For example, vertically rotating
vortices can develop between buildings in so-called street can-
yons (e.g. DePaul and Sheih, 1986; Oke, 1988), horizontally
rotating vortices may exist near the canyon-intersection interface
(e.g., Hoydysh and Dabberdt, 1988; Pol and Brown, 2008),updrafts often occur on the backsides of tall buildings (e.g., Heist
et al., 2004), and strong downdrafts on the front faces of tall
buildings that stick up above others can lead to divergence at the
surface and street-level winds in all directions (e.g. Hanna et al.,
2006; Nelson et al., 2007; Princevac et al., 2010). In more open
areas, the prevailing wind can penetrate down into the city and
result in strong channel flow, bifurcating at 4-way intersections,
T-intersections, and along side streets (e.g., Belcher, 2005). The
channel winds interact, compete, and merge with the building-
induced vortices, updrafts and downdrafts and may produce highly
spatially variable winds at street level. As the prevailing wind
directions shifts, the vertically rotating vortices can turn into spiral
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/jweia
Journal of Wind Engineeringand Industrial Aerodynamics
0167-6105/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jweia.2011.01.010
n Corresponding author.
E-mail address: [email protected] (M. Neophytou).
J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368
http://-/?-http://www.elsevier.com/locate/jweiahttp://dx.doi.org/10.1016/j.jweia.2011.01.010mailto:[email protected]://dx.doi.org/10.1016/j.jweia.2011.01.010http://dx.doi.org/10.1016/j.jweia.2011.01.010mailto:[email protected]://dx.doi.org/10.1016/j.jweia.2011.01.010http://www.elsevier.com/locate/jweiahttp://-/?-8/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
2/12
vortices (e.g., Nakamura and Oke, 1988; Klein and Clark, 2007),
downdrafts can dissipate in intensity as the prevailing wind hits tall
buildings at oblique angles, and channel flow reduces in strength in
some streets and increases in others, resulting in complicated non-
linear responses to the changing wind direction. These complex
building-induced 3D flow patterns can significantly alter the trans-
port and dispersion of plumes, resulting in displaced plume cen-
trelines (e.g., Theurer et al., 1996; MacDonald and Ejim, 2002),
significant upwind transport (e.g., Hanna et al., 2006) and reducedstreet-level concentrations in some cases due to enhanced venting
(e.g., Heist et al., 2004, 2009) and increased street-level concentra-
tions in other cases due to trapping and sheltering effects
(e.g., Chang and Meroney, 2003).
Accurate transport and dispersion prediction at the city scale
requires models that account for these effects on the wind field.
Over the past several decades considerable progress has been
seen in the application of computational fluid dynamics (CFD)
models. Some of the first CFD simulations were performed around
a few idealized buildings (e.g., Patankar, 1975; Murakami et al.,
1992). With the enhancement in computer power, researchers
are currently performing simulations in real cities that include
hundreds to thousands of building obstacles (e.g., see Neophytou
and Britter, 2005; Pullen et al., 2005; Hanna et al., 2006).
A number of teams have successfully applied Reynolds-averaged
NavierStokes (RANS) codes and large-eddy simulation (LES)
codes to neighborhood- and city-scale transport and dispersion
problems (e.g., Coirier et al., 2006a, b). In addition, progress has
been made in highly parameterized (non-NavierStokes) numer-
ical wind field models for use in emergency response systems, or
even for assessments and planning where many cases must be
run over a short time period (Boris et al., 2001). These codes are
based on potential flow or diagnosticempirical approaches
(e.g. Rockle, 1990; Kaplan and Dinar, 1996). These types of models
attempt to approximate the impacts of large numbers of buildings
on the wind flow, providing a semi-realistic 3D flow field around
buildings, but at a fraction of the run time as compared to CFD
codes. The substantial decrease of run time makes such models
particularly useful; however, it is extremely important to deter-
mine: (i) what kind of wind flow features are missed using such
simplifications and assumptions and consequently (ii) how much
these predictions from simpler models differ from predictions of
higher-complexity models.
This paper reports results from an inter-comparison between
three urban wind models, each of different level\of complexity,
using data from time-averaged wind field measurements obtai-
ned in downtown Oklahoma City during the Joint Urban 2003
Field Campaign; the exercise is performed for 3 specific test cases
(corresponding to three different Intensive Observation Periods)
using a specific computational set-up reflecting fast-response
standards. The paper is structured as follows: a brief summary
of the Joint Urban 2003 field campaign including the specific
Intensive Observation Periods (IOP) that were used in our ana-lyses is given in Section 2. The models in increasing order of
computational cost and complexity include: (i) a mass-consistent
empirical-diagnostic code, (ii) a RANS-CFD code, and (iii) a LES-CFD
code with a more detailed description of the models presented
in Section 3. It is important to note that the specific computa-
tional set-up used in this exercise was selected to reflect realistic
running conditions for fast-response applications but may be sub-
optimal for the more complex RANS and LES models. Finally, a
systematic comparison of the models with the wind field data is
presented and discussed in Section 4 while Section 5 concludes
with the overall findings of this work considering carefully the
difficulty in arguing for clear superiority of one model over
another based on a certain number of cases, as well as suggestions
for further work.
2. Oklahoma City Joint Urban 2003 (JU2003)
Wind measurements obtained from the Joint Urban 2003
(JU03) field experiment were used to evaluate the wind models.
The JU03 experiment was held in Oklahoma City in July 2003 with
the goal of providing information useful for testing and evaluation
of the next generation of urban transport and dispersion models.
The experiment consisted of a large number of tracer releases, a
network of concentration samplers and meteorological sensorsplaced in and around the city. An overview of the field campaign
is provided in Allwine et al. (2004).
Although the Central Business District (CBD) in Oklahoma
City is only a little over a kilometer on a side (Fig. 1), a number
of relatively tall buildings exist there, including the Bank One
Building (152 m), the FNC Building (123 m), the Oklahoma Tower
(117 m), the Kerr-McGee Building (115 m), and City Place
(107 m). The plan area fraction in the CBD is 0.4 and the average
building height in the southern half of the CBD is 27 m, while in
the northern half it is 65 m ( Burian et al., 2005). There are some
trees in the CBD along streets and in the Botanical Gardens Park in
the southwest sector of the CBD. During the July experimental
period, temperatures were generally warm, winds were strong,
and the prevailing wind direction consistently had a southerly
component.
About one hundred and fifty 2D and 3D sonic anemometers
were placed throughout the city on towers, tripods, light poles,
and building rooftops and nine sodars were placed in and around
the city to provide upper-air information. Although a majority of
the wind instrumentation was operating during the entire period
of the experiment, there were intensive operating periods (IOPs)
where more equipment was placed in the field. Within the central
business district (CBD), the instrument distribution below roof-
level included twenty four anemometers in the eastwest running
street canyons (e.g., Park Avenue) and twenty four anemometers
in northsouth running street channels (most in intersections,
Depicted domain in comparative
results (Figures 2,3, and4)
350m
Fig. 1. A Google Earth aerial view of the entire Central Business District in
Oklahoma City. The extent of the image corresponds to the size of the modeling
domain used for this study.
M. Neophytou et al. / J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368358
http://dx.doi.org/10.1016/j.jweia.2011.01.0108/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
3/12
however) during the IOPs. For this model evaluation study, we have
used thirty-minute average sonic measurements made near street
level on tripods (22.5 m agl), towers (2.53.5 m) and on light
poles (7.58 m agl) from IOPs 2, 8, and 9. Note that in several
locations two anemometers were placed at the same x, y location,
but a meter or so apart vertically. In the vector plots to follow, this
explains what appears to be two vectors at the same location in
several cases.
Towers and remote sensing sodars located upwind (i.e., south)of the CBD were used to create inflow wind profiles. The closest
upwind location was about 200 m due south of our modeling
domain. A Dugway Proving Ground (DPG) propeller anemometer
was located at that location on a tower 35 m above the ground
and 25 m above the roof of a post office building. The anem-
ometer was not operating, however, during IOP 2. Eight sonic
anemometers from Indiana University (IU) were located between
2 and 80 m above ground level on several towers about 5 km
south of our domain. For IOP2, only two sonics provided informa-
tion, however. A sodar from the Pacific Northwest National
Laboratory (PNNL) located about 1 km south-southwest of the
CBD was used for the winds above 100 m. The Argonne National
Laboratory (ANL) sodar located in the Botanical Gardens in the
southwest corner of our domain also provided reasonable inflow
information if the winds were from the southwest, but not if they
were from the southeast (probably due to larger buildings being
directly southeast of the sodar). In addition, we used two other
ANL sodars that were downwind of the city to cross correlate with
the upwind measurements.
3. Description of computational models and model set-up
In this study, we have used three different urban wind models,
each representing a different level of complexity: an empirical-
diagnostic code, a simple Reynolds-averaged NavierStokes (RANS)
code, and a large eddy simulation (LES) codeall implemented as an
option in the Quick Urban & Industrial Complex (QUIC) dispersion
modeling system (Nelson and Brown, 2006).
3.1. Empirical-diagnostic wind solver
The QUIC-URB (Updated Rockle-style Building-aware) wind
solver produces a mass-consistent flow field and uses empirical
parameterizations to produce a velocity field that maintains
important features of the time-averaged flow around buildings.
The wind solver is based on Rockle (1990) in which various
empirical relationships based on the building height, width, and
length and the spacing between buildings are used to initialize
the velocity fields in the regions around buildings (e.g., upwind
rotor, downwind cavity and wake, street canyon vortex, and
rooftop vortex). This initial flow field is then forced to satisfy
mass conservation. More information about the QUIC-URB windsolver and modifications made to the code for high density urban
areas can be found in Pardyjak and Brown (2003), Singh et al.
(2008), Brown et al. (2009), and Gowardhan et al. (2010). Version
5.4 of the QUIC-URB wind solver has been evaluated in this study.
3.2. RANS computational fluid dynamics model
The Q-CFD(RANS) solver is based on the 3D Reynolds-Aver-
aged NavierStokes (RANS) equations for incompressible flow
using a zero equation (algebraic) turbulence model based on
Prandtls mixing length theory (Gowardan et al., in press). The
selection of zero-equation turbulence model was made so as to
reduce the run time of the CFD simulation, and therefore making
it more closely adapted for a fast-response application (Chen and
Xu, 1998). Computational time using a zero-equation model can
be reduced by 28 times from that using a more complex
turbulence modelkeeping all other computational settings such
as discretization schemes and mesh size the same (REF). It is
accepted however, that more complex turbulence models could
also be considered, however, given that there has been no
evidence of clear superiority or appropriateness of one model
over another for all wind flow applications in real urban geome-
tries (REF), using a zero-equation model was considered accep-table in the context of this exercise.
The governing RANS equations are solved explicitly in time
until steady state is reached using a projection method. At each
time step of the projection method, the divergence-free condition
is not strictly satisfied to machine precision levels, but rather
when steady state is reached incompressibility is recovered. This
makes the method comparable to the artificial compressibility
method (Chorin, 1967). The RANS equations are solved on a
staggered mesh using a finite volume discretization scheme that
is second-order accurate in space (central difference) and time
(AdamsBashforth). The law-of-the-wall was imposed at all of the
solid surfaces. The pressure Poisson equation was solved using
the successive over-relaxation method (SOR). A free slip condition
was imposed at the top boundary and the side boundaries, while
an outflow boundary condition is used at the outlet. More
information on the numerical scheme and parameterizations
can be found in Gowardhan et al. (in press).
3.3. LES computational fluid dynamics model
In large eddy simulation (LES) application of a spatial filter to
the NavierStokes equations is generally used to partition the
solution space into resolved and subgrid parts. The large and
energetic scales of turbulence are calculated explicitly, while the
small scales are represented by a model. The QUIC-LES code uses the
Smagorinsky subgrid-scale eddy viscosity model. The NavierStokes
equations are discretized using the second-order QUICK scheme
in space and a second-order AdamsBashforth scheme in time ona staggered grid using a finite volume technique. The algorithm is
based on the fractional step method (Kim and Moin, 1985). The
Poisson equation for pressure p is solved by a multigrid method.
Inflow velocity profile is specified at the inlet and outflow boun-
dary condition is specified at the outlet, while the boundary
condition at the top of the domain is free slip. At all solid surfaces,
the local profile of the tangential velocity component is taken to
be logarithmic and the normal velocity component is zero. The
LES code used in this exercise is based on the development
by Gowardhan et al. (2007) and has been validated for various
turbulent flow problems.
3.4. Model set-up
The modeling domain covered most all of the Oklahoma City
central business district (CBD) and was 1.2 km 1.2 km in size for
all model calculations (see Fig. 1); the horizontal grid size was
set to 5 m and the vertical grid size was 3 m resulting in a
236 242 64 grid cell domain (3.66 million cells in total) for all
models, with a blockage ratio of 15%. Such a computational set-up
(e.g. of this mesh and cell size) is considered typicalreflect-
ing realistic scenarios of fast-response applications (Boris et al.,
2001; Patnaik and Boris, 2007). It should be noted that this may
not necessarily be optimal for a standard application of CFD, e.g.
LES or RANS, as the focus of this study is rather the application of
these models in a realistic scenario of a fast-response situation for
a corresponding typical computational set-up for such situation
(Boris et al., 2001; Patnaik and Boris, 2007). Sensitivity studies
M. Neophytou et al. / J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368 359
8/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
4/12
were not performed due to the extremely high computational
cost. The 3D building data were obtained from the Defense Threat
Reduction Agency and the University of Oklahoma. Parking
garages were set as solid buildings and although there were some
trees in the downtown area, they were ignored in the simulations.
The inflow wind speed profile was created by fitting a log-law
to thirty-minute average wind measurements from the IU sonics,
the DPG prop-vane and the PNNL sodar measurements (see
Table 1). The specification of the inflow wind direction profile
was somewhat more difficult as the different instrumentation
often showed up to 301 differences in wind direction, with little
consistent bias between instrument location and time periods. To
determine the inflow wind direction profile, we plotted thirty-
minute average measurements from the IU sonics, the DPG prop-
vane, the PNNL sodar, and the three ANL sodars and specified the
wind direction based on where the majority of measurements fell,
ignoring outliers (similar to a modal average). Note that the
inflow profile for the LES calculation did not vary with time, so
that the inflow turbulence will be underestimated (e.g. Xie and
Castro, 2006, 2009; DeCroix and Brown, 2002). The QUIC-LES code
was run for 1800 s (physical time) for the flow to develop and the
results were then averaged over the next 1800 s (actual duration
of an IOP) so that it can be compared with the field data.
In order to test the models under different inflow conditions,
simulations were performed for three distinct time periods when
the ambient winds were from the south-southwest, the south, and
the south-southeast. These time periods corresponded to IOP-2,
Release 1 (10:0010:30 CST, July 2), IOP-8, Release 2 (00:0000:30
CST, July 25), and IOP-9, Release 2 (01:0001:30 CST, July 27). All
simulations were performed assuming neutral stability. Within the
urban canopy this assumption is likely valid, but above the canopy
there may be a stable layer during IOP8 and an unstable layer during
IOP2. Winds were fairly brisk during all three IOPs; hence, it is
likely that the stability was not too far from neutral. Additionally,
numerous studies have shown that a several hundred meter well-
mixed neutrally stratified layer often exists above larger-sized city
centers due to enhanced thermal and mechanical mixing effects. Thedowntown district of Oklahoma City is rather small, however, and
thus it is not clear if the well-mixed layer is deep or shallow.
4. Results and discussion
4.1. Overview
The results in this section are focused on the central part of the
domain, which corresponds to the core of the Central Business
District with dimensions roughly 400 m 500 m. This area con-
tains a good number of the near-surface wind measurements in
the most built-up region of the city. Additionally, from the
computational modeling point of view, it avoids regions nearer
the edges of the computational domain, which may be influenced
by the application of specific boundary conditions. Fig. 1 shows an
aerial view of the computational domain. An important feature of
the flow field determining pollutant transport is the resulting
wind flow directions, which subsequently determine flow struc-
tures (e.g. channeling, recirculation, etc). Therefore, we have focused
on several different types of flow regimes within the CBD: (i) flow
channeling regions, (ii) street intersections, (iii) sheltered regions,
and (iv) open areas. In the subsequent discussion and the modelinter-comparison, emphasis is placed on the ability of the model to
capture such flow features discerned from the wind measurements.
4.2. Comparative qualitative results
4.2.1. IOP2southwesterly inflow
For the IOP2 southwesterly inflow case (2151), the three
models agree qualitatively with measurements in several loca-
tions, and disagree in several other places (Fig. 2). Qualitatively
the flows created by the Q-CFD(RANS) and Q-LES codes look
smoother and are less noisy than the Q-URB solution. All three
models capture extremely well the west to east channeling along
Sheridan just north of the Convention Center, as well as the
change in direction of the flow from west-northwesterly to
southeasterly as one goes from the western side of the Broad-
way-Main St. intersection to the northern side. As one travels up
Broadway just past the Bank One Building, the Q-LES and
Q-CFD(RANS) models show much stronger south-to-north chan-
neling flow as compared to Q-URB. It appears that the CFD models
overestimate the magnitude in this region, while the empirical
Q-URB model underestimates it (this is discussed further in
subsequent sections). All three models show a downwind cavity
on the north side of the Bank One building (just to the west of
the Broadway-Park Ave. intersection), however the CFD codes
better represent the strength and direction of the single wind
measurement there.
There are some distinct differences between the models
near the RobinsonSheridan intersection in the SW sector of
the domain. First, in the southwestern corner of Fig. 2 plots, in
the open region immediately west of the Convention Center, the
Q-LES model has produced near zero winds, the Q-URB model
shows the flow in this region being virtually identical to the
ambient SW inflow, and Q-CFD(RANS) shows flow in the same SW
direction, but it has been significantly retarded. Although there is
only one measurement in this region it appears that the Q-URB
and Q-CFD(RANS) codes have best matched the flow direction,
and Q-CFD(RANS) has best matched the wind speed. The stagna-
tion region produced by the Q-LES model appears to have resulted
from strong southerly backflow on the front faces of tall buildings
a few blocks north that are just outside Fig. 2 western plot
boundary.
Just to the north near the RobinsonMain St. intersection,the Q-CFD(RANS) model appears to be doing the best, capturing
the westerly flow along Main St. and the southerly flow along
Robinson. The Q-LES model also has the westerly flow penetrating
through the intersection and matching the measurement along
Main St., however, it also shows weak northerly winds on
Robinson at the measurement location, not the relatively strong
southerly winds measured there. Q-URB, on the other hand, is not
able to propagate the westerly flow through the Robinson-Main
St. intersection, but does correctly produce the southerly winds to
the immediate north of the intersection on Robinson.
Further north along Robinson at the Park Ave. cross street, the
Q-CFD(RANS) model is seen to compute the merger of the south-
erly flow along Robinson and the westerly flow along Park Avenue
exceptionally well. The Q-LES model also does well, but appears
Table 1
The inflow wind profiles used to describe the different IOP cases in the numerical
models.
Test case: IOP2
Wind inlet profile (logarithmic) Uref5 m/s, zref50 m, wind angle 215;
Surface roughness z00.6 m
Test case: IOP8
Wind inlet profile (logarithmic) Uref8.5 m/s, zref80 m, wind angle 165;
Surface roughness z01 m
Test Case: IOP9
Wind inlet profile (logarithmic) Uref7 m/s, zref50 m, wind angle 180;
Surface roughness z01 m
M. Neophytou et al. / J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368360
8/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
5/12
to overestimate the strength of the westerly flow to the west of
the intersection and underestimate the strength of the southerly
flow to the south of the intersection. Q-URB actually matchesthe wind directions fairly well point by point in the Robinson
Park Ave. intersection, but the flow field in this region is
unsatisfactorily noisy.
Within the western half of the Park Ave. street canyon
between Robinson and Broadway, the Q-URB model has the flow
going down the street the right way (westerly) at about the right
magnitude, but individual flow vectors can be off by ninety
degrees. Both the Q-CFD(RANS) and Q-LES models capture the
flow direction well on the western half of Park Ave., but the
strength of the flow is overestimated. The Q-CFD(RANS) code is
able to resolve the easterly reverse flow at the eastern end of the
street canyon, whereas the Q-LES model incorrectly shows the
westerly flow extending through the entire length of the street
canyon.
The Q-LES model appears to perform better on the northern
side of the domain. The backflow produced by Q-LES along McGee
Ave. matches the measurements at McGee & Broadway and atMcGee & Robinson (note that there are tall buildings immediately
north beyond the plot boundary). Both Q-URB and Q-CFD(RANS)
are off by 901 at these two locations. It appears that the models
slightly underestimated the spatial extent of the backflow as
only a few grid cells away the model-computed flow is in the
right direction. Another big difference between Q-URB and the
two CFD codes is apparent in the northeast quadrant downwind
of the wide building running along Kerr Avenue between Broad-
way and Gaylord. Here, both CFD models show a rather short
downwind cavity region, whereas Q-URB shows relatively strong
backflow. Similarly, the long rectangular building directly south
parallel to Gaylord shows southerly flow on the downwind side
near the back wall, whereas Q-URB shows a thin region of
northerly flow.
Fig. 2. Measured (in black) and predicted (in gray) velocity field in the Central Business Area using QUIC-URB (a), Q-CFD(RANS) (b) and Q-LES (c) for IOP2 (wind direction:
2151).
M. Neophytou et al. / J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368 361
8/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
6/12
4.2.2. IOP8south-southeasterly inflow
For the IOP8 south-southwesterly inflow case (1651) depicted
in Fig. 3, the three models do fairly well reproducing the flow
along Sheridan near the Convention Center. The flow is straigh-
tened to a pure southerly flow east of the Convention Center on
Gaylord Blvd., is close to background flow direction to the west of
the Convention Center on Robinson, and shows the easterly
backflow in the Park Ave.Sheridan intersection immediately
north of the Convention Center. The strength of the backflow offof the building immediately north of the Convention Center
(x750, y 575) is weakest for Q-URB and strongest for Q-LES.
All three models show southerly channeling along sections of
Broadway, but there are differences in the strength of the flow,
where the southerly flow is interrupted, and whether the south-
erly flow covers the entire width of the street. The Q-LES model,
for example, shows westerly flow penetrating through the Main
St.Broadway intersection (in contradiction to the measurements
along Main St. near the intersection), while the Q-CFD(RANS)
shows the entire intersection consisting of southerly flow and
Q-URB has the western half of the street with southerly flow and
the eastern half being northerly. All of the models have some level
of disagreement with the measurements along Main St., in some
cases being 901801 out of phase.
The channeling of the flow on Broadway north of the
BroadwayMain St. intersection, the reverse flow in the lee ofthe Bank One Building, and the penetration of the flow into Park
Ave. seems to be best captured by the Q-CFD(RANS) model,
although the Q-LES does a fair job as well. Q-URB completely
misses the reverse flow on the downwind side of the Bank One
Building, but other than that appears to get the wind directions
more or less correct in this region. On the western side of
Park Ave there is clearly divergent flow resulting from
Fig. 3. Measured (in black) and predicted (in gray) velocity field in the Central Business Area using QUIC-URB (a), Q-CFD(RANS) (b) and Q-LES (c) for IOP8 (wind direction:
1651).
M. Neophytou et al. / J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368362
8/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
7/12
south-southeasterly flow impinging on the front face of the
relatively tall City Place building. The CFD models capture this
feature best, although Q-URB also captures some aspects of
the flow.
The flow computed by all three models near the Robinson
Main St. intersection is in good agreement with the two measure-
ments there. Further north, at the RobinsonPark Ave. intersec-
tion, the southerly flow is also well predicted by all three models
and in the case of the two CFD models the easterly outflow fromPark Ave as well. All three models, however, incorrectly show a
southerly flow through the RobinsonKerr intersection, whereas
the wind measurement indicates an easterly flow there.
The southerly flow measured near the BroadwayKerr intersec-
tion is captured by the Q-URB and Q-LES models, but the Q-
CFD(RANS) code shows easterly flow there. Further north, at the
BroadwayMcGee intersection, the Q-LES model is in best
agreement to the weak westerly flow measured there, while
Q-URB and Q-CFD(RANS) are both off by 901 with a strong
southerly flow.
4.2.3. IOP9southerly inflow
This case with winds coming from 1801 proved to be fairly
difficult for Q-URB, while the Q-CFD(RANS) code appears to have
done fairly well and the Q-LES somewhere in between ( Fig. 4). Atthe BroadwaySheridan intersection just north of the Convention
Center, the Q-URB model has likely overestimated the length of
the downwind cavity reverse flow region (however, the measure-
ments are inconclusive as one of the measurements at this
location shows weak reverse flow, while the other does not).
Q-LES shows a combination of westerly channel flow, backflow off
the tall building to the north (x 750, y 575), and weak reverse
Fig. 4. Measured (in black) and predicted (in gray) velocity field in the Central Business Area using QUIC-URB (a), Q-CFD(RANS) (b) and Q-LES (c) for IOP9
(wind direction:1801).
M. Neophytou et al. / J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368 363
8/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
8/12
flow in the lee of the Convention Center in reasonable agreement
with the measurements. Q-CFD(RANS) shows similar features as
the Q-LES calculation, except that the westerly channeling is
stronger and no evidence of reverse flow is apparent immediately
downwind (north of) the Convention Center.
Both Q-URB and Q-LES incorrectly predict southerly flow
through the BroadwayMain St. intersection, while Q-CFD(RANS)
correctly computes the easterly flow seen there (resulting from
strong street-level backflow off the front face of the tall Bank Onebuilding). All three models are in agreement with the strong
southerly channeling north of the BroadwayMain intersection.
Both Q-CFD(RANS) and Q-LES predict backflow on the lee (north)
side of the Bank One building in agreement with measurements,
but the flow field circulation is very different in each case. Q-URB
computes some backflow in this region, but the locations of the
backflow do not agree with the measurements.
Both Q-LES and Q-CFD(RANS) capture the penetration of the
winds from Broadway into Park Ave as depicted in the measure-
ments, while Q-URB does not. Q-CFD(RANS) is the only model
that shows the end vortex in this region in agreement with the
easterly flow on the north side of the street and westerly flow on
the southern side. Both Q-LES and Q-CFD(RANS) agree with the
measurements showing westerly flow in the center of the street
canyon and both have produced a counter rotating vorticity in the
street canyon near the RobinsonPark Ave intersection (although
it is not clear from the measurements if such a vortex exists
there). Overall, Q-URB performed poorly in Park Ave.
The southerly winds along Robinson, however, were well
predicted by Q-URB, as well as Q-LES. Q-CFD(RANS) did well also,
but incorrectly computed a westerly wind at the RobinsonMain
St. intersection, whereas the measurement revealed a southerly
wind. All three models incorrectly predicted a southerly wind at
the RobinsonMcGee intersection, whereas the actual wind was
easterly. The direction of the wind one block to the east, also on
McGee, was correctly computed by Q-CFD(RANS), but not by
Q-URB and Q-LES. It is obvious that Q-URB has severely over-
estimated the southerly channel flow along Broadway resulting in
the incorrect wind flow at the McGeeBroadway intersection.
Overall, it could be noted that the models have reasonable
agreement with measurements in some locations and significant
differences in others. Although an attempt has been made to
group the zones of weak agreement and provide possible reasons
for the discrepancies, due to the relatively limited number of
points, areas and cases we avoided to deduce a more generalized
conclusion on such issue of stronger/weaker agreement and their
possible related explanation. For instance, the inherent transient
effects of urban aerodynamics are predominant in certain loca-
tions than others and that could explain perhaps some discre-
pancies especially for RANS.
4.3. Comparative quantitative results
A quantitative comparison has been made by examining
various wind scatter plots produced by the three models. An
important feature that is examined here is the wind direction, in
addition to wind speed, as this determines substantially the flow
pattern as well as features produced within the urban area, which
thereby determine the transport and dispersion. Figs. 5 and 6
show the scatter plots for the wind speed and wind direction for
each of the 3 IOP cases IOP2, IOP8, IOP9, respectively and
depict the bounds for the 710%, 25%, 50% and 100% percentage
error for wind speed as well the bounds within 7151, 301, 451and 901 for wind direction. From these plots, the percentage of the
model predictions within the specified error bounds was calcu-
lated. The results from each IOP case and each model are listed in
Table 2, while Table 3 lists the average results for all three IOP
cases for each model and Table 4 lists the correlation coefficients
(between observed and predicted) derived from each scatter plot
(each case and model).
As indicated in Tables 2 and 3, the RANS code was within 50%
of the measured wind speed 62% of the time (on average
considering all three IOP cases, ranging from 57% in IOP9 to 71%
IOP2) as compared to 53% for the LES model (on average
considering all three IOP cases, ranging from 41% in IOP2 to 60%in IOP8) and 49% for the empirical-diagnostic code (on average
considering all three IOP cases, ranging from 47% in IOP2 to 51%
IOP8). Considering all the cases individually, there does not seem
any one proving consistently more difficult than others for the
models to capture. In turn, considering the level of quantitative
differences in the computed errors, no model seems to be
performing clearly better than other in predicting the wind speed.
Similarly, for wind direction, the RANS-CFD code was within 301of the measured wind direction 58% of the time (on average
considering all IOP cases, ranging from 54% in IOP8 to 63% in
IOP2) as compared to 50% for the LES code (on average consider-
ing all IOP cases, ranging from 50% in IOP9 to 57% in IOP2 and
IOP8) and 43% for the empirical diagnostic code (see Tables 2
and 3). If stricter bounds are considered, for example wind speed
errors within 25%, the RANS-CFD was within this error threshold
33% of the time, the LES model 29% and the empirical-diagnostic
code 28%, considering all three IOP cases on average. For the average
wind direction error, the model predictions were within 151 of the
measured wind direction 39% of the time for the RANS-CFD code, as
compared to 34% for the LES code and 27% for the empirical
diagnostic code. For cases where the plume is being channeled down
a street one direction or 1801 in the other direction, wind errors at
street-level of up to7601 can in many cases actually be considered a
success for street-level plume transport and dispersion since the
plume will be transported in the right direction within a street
canyon if the winds are within this error bound. That is, for
successful plume model transport one may consider whether the
model computations have the same easterly or westerly component
as the wind measurements in an eastwest running street or the
same northerly or southerly component in a northsouth running
street. It is interesting to note that although some of the vector plots
may show a fairly good agreement (in the flow patterns), it appears
that a relatively good agreement in vector plots does not necessarily
mean a good scatter plot for wind speeds. For example, although
the values of the correlation coefficients for wind direction may
reach as high as 0.860 (that is for Q-URB in the IOP 2 case), for
exactly the same case, the same model predictions produce a
correlation coefficient for the wind speeds as low as 0.068
(see Table 4). Given the range of correlation coefficient values across
the various cases, it appears that wind directions are overall better
captured than the wind speeds. All the correlation coefficients
between the observed and simulated values for each presented
scatter plot (of Figs. 5 and 6) are listed in Table 4.It is important to consider that given the natural variability
and uncertainties in the field, these levels of quantitative differ-
ences in the various computed statistical measures do not support
clearly a better performance of one model over others. It is also
worth noting the computational times required to run the
corresponding code. For example, the computational time for
the simulations using the Q-URB code for the three IOP cases were
of the order of 1 min, for the Q-CFD(RANS) of the order of 30-min
and for the Q-LES 30 h using a standard PC for Q-URB and Q-
CFD(RANS) and a parallel cluster of 8 nodes for the Q-LES.
Although the accuracy of Q-URB may not appear as good as
that of the Q-CFD(RANS) and Q-LES codes, it is only by a rela-
tively small percentage fraction when compared to the two
CFD codes.
M. Neophytou et al. / J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368364
8/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
9/12
Moreover, accounting for the inherent uncertainty and natural
variability in the wind field, it is difficult to argue in some cases
(especially when differences in percentage fraction of successes
are of the order of a few percentage points) for clear superiority of
one model over another. Especially, with regard to evaluationof urban wind models, it is necessary to specify the purpose of
model application and thereby judge its appropriateness through
specific performance evaluation criteria and metrics (COST732,
2007). For example, given that the empirical diagnostic code runs
23 orders of magnitude faster and its performance in the
quantitative comparisons, the empirical diagnostic code appears
to be a viable option for applications where fast turnaround time
is required or where many cases must be run.
It is important to note that for the purpose of consistent
comparisons, the LES code is most likely run under sub-optimal
conditions, e.g., the grid cell size near the walls should be smaller
and the inlet boundary conditions should be fully turbulent and
match the atmospheric conditions during the experiment. It is the
case, however, that conditions and parameters representative of
the field and necessary as input to the LES code are often not
available from the available field measurements and in some
cases the needed grid resolution is not operationally plausible.
5. Concluding remarks
Three computational wind models of different level of com-
plexity were systematically tested and compared with wind
measurements from the Oklahoma City Joint Urban 2003 field
experiment. The comparative exercise included only near-street
level data collected within the Central Business District (between
2 and 7.5 m) using 30 min averages. The size of the input domain,
grid resolution, building dimensions, wind inflow profiles and
other relevant parameters within the different codes were
selected to reflect fast-response modeling needs and were matched
to allow for consistent comparisons.
Overall, the results show that qualitatively all three models
compare favorably to the near-surface wind measurements in
IOP2 QUIC-URB
IOP2 QUIC-RANS
IOP2 QUIC-LES
IOP9 QUIC-URBIOP8 QUIC-URB
IOP9 QUIC-RANSIOP8 QUIC-RANS
IOP9 QUIC-LESIOP8 QUIC-LES
Fig. 5. Scatter plots for wind speed for measurements and corresponding predictions in the Central Business Area using QUIC-URB (top), Q-CFD(RANS) (middle) and Q-LES
(bottom) for IOP2, IOP8, IOP9 (left, center, right, respectively). The plotted lines show the bound lines for percentage error within 710%, 725%, 750%, 7100%.
M. Neophytou et al. / J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368 365
8/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
10/12
IOP2 QUIC-URB
IOP2 QUIC-RANS
IOP2 QUIC-LES
IOP9 QUIC-URBIOP8 QUIC-URB
IOP8 QUIC-RANS IOP9 QUIC-RANS
IOP9 QUIC-LESIOP8 QUIC-LES
Fig. 6. Scatter plots for wind velocity direction for measurements and corresponding predictions in the Central Business Area using QUIC-URB (top), Q-CFD(RANS) (middle)
and Q-LES (bottom) for IOP2, IOP8, IOP9 (left, center, right, respectively). The plotted lines show the bound lines for wind direction error within 7151, 7301, 7451 and
7901.
Table 2
Calculated percentage error for each model in each IOP case, for wind speed and wind direction; the symbols OBS and SIM represent the observed and simulated/predicted
values, respectively.
Wind speed error, Error9OBSSIM9
0:5 OBS SIM
Wind direction error, Error 9OBSSIM9
o100% o50% o25% o 10% o901 o451 o301 o151
IOP 2
Q-URB 71% 47% 27% 14% 77% 65% 48% 32%
Q-CFD 88% 71% 37% 14% 84% 75% 63% 38%
Q-LES 82% 41% 27% 10% 77% 68% 57% 38%
IOP 8
Q-URB 85% 51% 35% 13% 77% 52% 43% 30%
Q-CFD 85% 58% 33% 16% 86% 59% 54% 38%
Q-LES 87% 60% 18% 5% 77% 68% 57% 38%
IOP 9
Q-URB 79% 48% 23% 4% 77% 57% 39% 20%
Q-CFD 90% 57% 30% 13% 91% 68% 57% 41%
Q-LES 84% 57% 43% 23% 91% 70% 50% 34%
M. Neophytou et al. / J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368366
8/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
11/12
many locations, although there are several instances of winds
being calculated poorly in specific locations. This qualitative
assessment was bolstered by the point-by-point quantitative
comparisons of the wind speed and wind direction. The RANS-
CFD code, for example, was within 50% of the measured wind
speed 62% of the time as compared to 53% for the LES model and
49% for the empirical-diagnostic code. For wind direction, the
RANS-CFD code was within 301 of the measured wind direction
58% of the time as compared to 50% for the LES code and 43% for
the empirical diagnostic code. It should be pointed out that there
were noticable differences between the wind fields produced by
the LES and RANS-CFD codes in specific regions of the domain,
especially in intersections, in the strength of backflow resulting
from flow impingement on the front faces of buildings, and the
strength of channel flow. It is important to consider that given the
variability and uncertainties in the field, the level of quantitative
differences in the computed errors do not support a clearlysuperior performance of one model over others. However, it is
interesting to note that the simpler model seems to perform
almost nearly as well as the more computationally demanding
models. It should also be noted that the computational set-up
selected for all the models reflects typical realistic settings and
requirements for fast-response modeling and may not necessarily
reflect optimal settings and performance, particularly for RANS
and LES in more general applications. Moreover, accounting for
the inherent uncertainty and natural variability in the wind field,
it is difficult to argue in some cases for clear superiority of one
model over another. Given that the empirical diagnostic code is
23 orders of magnitude faster than the RANS and LES-CFD codes,
it appears that in this context it can be considered a practical
option for studies where time is of concern. Future work on the
comparison of model-computed and measured concentrations
from the Joint Urban 2003 tracer field study using the three
different types of wind models to drive a plume transport and
dispersion model will help to answer whether or not the differ-
ences in the wind fields translates into significant differences in
the concentration fields.
Acknowledgements
Dr. M. Neophytou wishes to acknowledge financial supportgranted by the United Nations Educational, Science and Cultural
Organization (UNESCO) for a 6-month exchange visit to US during
which this research has been conducted. Dr. Neophytou wishes
also to acknowledge the Systems Engineering and Integration
Group of Los Alamos National Laboratory (LANL) for hosting and
facilitating her 3-month visit at LANL.
References
Allwine, K.J., Leach, M.J., Stockham, L.W., Shinn, J.S., Hosker, R.P., Bowers, J.F.,Pace, J.C., 2004. Overview of Joint Urban 2003an atmospheric dispersionstudy in Oklahoma City. In: Proceedings of the AMS Symposium on Planning inNowcasting and Forecasting in Urban Zones, Seattle, WA.
Belcher, S., 2005. Mixing and transport in urban areas. Philos. Trans. R. Soc. A 363,29472968.
Boris, J., et al., 2001. Simulation of fluid dynamics around complex urbangeometries. In: Proceedings of the 39th Aerospace Sciences Meeting, AmericanInstitute Aeronautics and Astronautics, Reston, VA.
Brown, M., Gowardhan, A., Nelson, M., Williams, M., Pardyjak, E., 2009. Evaluationof the QUIC wind and dispersion models using the Joint Urban 2003 fieldexperiment. In: Proceedings of the AMS eighth Symposium Urban Environ-ment, Phoenix, AZ.
Burian, S., McKinnon, A., Hartman, J., Han, W., 2005. National Building StatisticsDatabase: Oklahoma City, Los Alamos National Laboratory Report LA-UR-05-8153,17 pp.
Chang, C., Meroney, R., 2003. Concentrations and flow distributions in urban streetcanyons: wind tunnel and computational data. J. Wind Eng. Ind. Aerodyn. 91,11411154.
Chen, Q., Xu, W., 1998. A zero-equation turbulence model for indoor airflowsimulation. Energy Buildings 28 (2), 137144.
Chorin, A.J., 1967. A numerical method for solving incompressible viscous flow
problems. J. Comput. Phys. 2, 1226.Coirier, W.J., Kim, S., Chen, F., Tuwari, M., 2006b. Evaluation of urban scalecontaminant transport and dispersion modeling using loosely coupled CFDand mesoscale models. In: Proceedings of the sixth Symposium on the UrbanEnvironment, American Meteorological Society, Atlanta, GA.
COST Action 732, 2007. Quality Assurance of Microscale Meteorological Models.Background and Justification Document to support the model evaluationguidance and protocol. 1 May 2007, Rex Britter, Michael Schatzmann (Eds.).
DePaul, F., Sheih, C., 1986. Measurements of wind velocities in a street canyon.Atmos. Environ. 20, 455459.
DeCroix, D., Brown, M., 2002. A large-eddy simulation of urban-scale dispersionand comparison to URBAN 2000 Field Program measurements, LA-UR-02-4755,120 pp.
Gowardhan, A., Brown, M., Pardyjak, E.R., 2010. Evaluation of a fast responsepressure solver for flow around an isolated cube. J. Environ. Fluid Mech. 10,311328.
Gowardhan, A., Pardyjak, E., Senocak, I., Brown, M.J., 2007. Investigation ofReynolds stresses in a 3D idealized urban area using large eddy simulation.In: Proceedings of the AMS Seventh Symposium Urban Environment, San
Diego, CA, paper J12.2, 8 pp.
Table 3
Calculated overall average (based on the 3 IOP cases) percentage error for each model, for wind speed and wind direction; the symbols OBS and SIM represent the observed
and simulated/predicted values, respectively.
Average wind spee d error, ErrorOBSSIMj j
0:5 OBSSIM
Average wind direction Error, Error OBSSIMj j
o 100% o50% o25% o10% o901 o451 o301 o151
Q-URB 78% 49% 28% 10% 77% 58% 43% 27%
Q-CFD 88% 62% 33% 14% 87% 67% 58% 39%Q-LES 84% 53% 29% 13% 82% 64% 50% 34%
Table 4
Calculated correlation coefficient between the predicted and observed values of
wind speed and wind direction for each model in the 3 IOP cases; the symbols OBS
and SIM represent the observed and simulated/predicted values, respectively,
while s represents the standard deviation of the corresponding values.
Wind speed correlation
coefficient,
R OBSOBSSIMSIM
sOBSsSIM
Wind direction correlation
coefficient,
R OBSOBSSIMSIM
sOBSsSIM
IOP 2
Q-URB 0.068 0.860
Q-CFD 0.463 0.776Q-LES 0.160 0.676
IOP 8
Q-URB 0.352 0.425
Q-CFD 0.644 0.618
Q-LES 0.653 0.755
IOP 9
Q-URB 0.356 0.402
Q-CFD 0.561 0.802
Q-LES 0.513 0.784
M. Neophytou et al. / J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368 367
8/2/2019 An Inter-comparison of Three Urban Wind Models Using Oklahoma City Joint Urban 2003 Wind Field Measurements
12/12
Gowardhan, A., Pardyjak, E., Senocak, I., Brown, M (in press). A CFD-based wind
solver for an urban fast response transport and dispersion model, Environ
Fluid Mech, doi:10.1007/s10652-011-9211-6.Hanna, S.R., Brown, M.J., Camelli, F.E., Chan, S.T., Coirier, W.J., Hansen, O.R.,
Huber, A.H., Kim, S., Reynolds, R.M., 2006. Detailed simulations of atmospheric
flow and dispersion in downtown Manhattan. Bull. Am. Meteorol. Soc. 87 (12),17131726.
Heist, D.K., Brixey, L.A., Richmond-Bryant, J., Bowker, G.E., Perry, S.G., Wiener, R.W.,
2009. The effect of a tall tower on flowand dispersion through a modelurban neighborhood, Part 1. Flow characteristics. J. Environ. Monit. 11,
21632170.Hoydysh, W., Dabberdt, W., 1988. Kinematics and dispersion characteristics offlows in asymmetric street canyons. Atmos. Environ. 22, 26772689.
Heist, D.K., Perry, S.G., Bowker, G.E., 2004. Evidence of enhanced vertical disper-
sion in the wakes of tall buildings in wind tunnel simulations of LowerManhattan. In: Proceedings of the AMS Fifth Symposium on the Urban
Environment, Vancouver, BC Canada.Kaplan, H., Dinar, N., 1996. A Lagrangian dispersion model for calculating
concentration distribution within a built-up domain. Atmos. Environ. 30
(24), 41974207.Kim, J., Moin, P., 1985. Application of a fractional-step method to incompressible
NavierStokes. J. Comput. Phys. 59, 308323.Klein, P., Clark, J., 2007. Flow variability in a North American downtown street
canyon. J. Appl. Met. Clim. 46, 851877.Macdonald, R.W., Ejim, C.E., 2002. Hydraulic flume measurements of mean flow
and turbulence in a scale model array of 4:1 aspect ratio obstacles. In:Proceedings of the Fifth UK Conference on Wind Engineering (WES 02),
University of Nottingham, England, 46 September 2002.Murakami, S., Mochida, A., Hayashi, Y., Sakamoto, S., 1992. Numerical study of
velocity-pressure field and wind forces for bluff bodies by ke, ASM and LES.
J. Wind Eng. Ind. Aerodyn. 4144, 28412852.Nakamura, Oke, T., 1988. Wind, temperature, and stability conditions in an
eastwest oriented urban canyon. Atmos. Environ. 22, 26912700.Nelson, M.A., Pardyjak, E.R., Klewicki, J.C., Pol, S.U., Brown, M.J., 2007. Properties of
the Wind Field within the Oklahoma City Park Avenue Street Canyon. Part I:mean flow and turbulence statistics. J. Appl. Meteorol. Climatol. 46 (12),
20382054.
Neophytou, Britter, 2005. Modelling of atmospheric dispersion in complex urbantopographies: a computational-fluid-dynamics simulation of the central Lon-don area. In: Proceedings of the Fifth International Congress on ComputationalMechanics, Limassol, Cyprus.
Nelson, M.A., Brown, M.J., 2006. The QUIC Start Guide (v. 5.6). LA-UR-10-01062(available from /http://www.lanl.govS).
Oke, T.,1988. Boundary Layer Climates, Routledge.Pardyjak, E.R., Brown, M., 2003. QUIC-URB v. 1.1: Theory and Users Guide, LA-UR-
07-3181, 22 pp (available from /http://www.lanl.govS).Patankar, S.V., 1975. Numerical prediction of 3D flow, in Studies in Convection. In:
Launder (Ed.), Academic Press, London, pp. 175.
Patnaik, G., Boris, J., 2007. Fast and Accurate CBR Defense for Homeland Security:Bringing HPC to the First Responder and Warfighter. HPCMP-UGC pp.120-126,DoD High Performance Computing Modernization Program Users GroupConference, 2007, IEEE Computer Society, /http://doi.ieeecomputersociety.org/10.1109/HPCMP-UGC.2007.33 S.
Pol, S., Brown, M., 2008. Flow patterns at the ends of a street canyon: measure-ments from the Joint Urban 2003 field experiment. J. Appl. Met. Clim. 47,14131426.
Princevac, M., Baik, J.J., Xiangyi, L., Pan, H., Park, S.B., 2010. Lateral channelingwithin rectangular arrays of cubical obstacles. J. Wind Eng. Ind. Aerodyn. 98,377385.
Pullen, J., Boris, J.P., Young, T.R., Patnaik, G., Iselin, J.P., 2005. A comparison ofcontaminant plume statistics from a Gaussian puff and urban CFD model fortwo large cities. Atmos. Environ. 39, 10491068.
Rockle, R., 1990. Bestimmung der stomungsver-haltnisse im Bereich KomplexerBebauugsstruk-turen. Ph.D. Thesis, Vom Fachbereich Mechanik, der Tech-nischen Hochschule Darmstadt, Germany.
Singh, B., Hansen, B., Brown, M., Pardyjak, E., 2008. Evaluation of the QUIC-URB
fast response urban wind model for a cubical building array and wide buildingstreet canyon. Environ. Fluid Mech. 8, 281312.Theurer, W., Plate, E., Hoeschele, K., 1996. Semi-empirical models as a combination
of wind tunnel and numerical dispersion modeling. Atmos. Environ. 30,35833597.
Xie, Z., Castro, I., 2009. Large-eddy simulation for flow and dispersion in urbanstreets. Atmos. Environ. 43, 21742185.
Xie, Z., Castro, I.P., 2006. Large-eddy simulation for urban micro-meteorology.J. Hydrodyn. B, 18, 259264 doi:10.1016/S1001-6058(06)60062-0.
M. Neophytou et al. / J. Wind Eng. Ind. Aerodyn. 99 (2011) 357368368
http://www.lanl.gov/http://www.lanl.gov/http://doi.ieeecomputersociety.org/10.1109/HPCMP-UGC.2007.33http://doi.ieeecomputersociety.org/10.1109/HPCMP-UGC.2007.33http://doi.ieeecomputersociety.org/10.1109/HPCMP-UGC.2007.33http://doi.ieeecomputersociety.org/10.1109/HPCMP-UGC.2007.33http://www.lanl.gov/http://www.lanl.gov/