Practical uses of the windstation computational fluid dynamics (CFD) model in air quality dispersion studies

Abstract

In this paper, we present several example applications and an evaluation scheme for an operational software tool that simulates wind flow over complex topography using Computational Fluid Dynamics (CFD) principles. This particular application software, named "WindStation ", is an example of a diagnostic simulator that solves the full 3D Navier-Stokes equations, including turbulence and thermal effects, using a boundary-fitted coordinate system. As currently implemented, it can be run as a Windows application based on a user-friendly graphical interface. This software is referred to below as one in a class of models that can be characterized as "rapid CFD models". The authors have adapted output from WindStation to provide for direct input into an air quality modeling system such as CALPUFF via the CALMET format. In that regard, this approach is an alternative to CALMET as well as the Mesoscale Model Interface Program (MMIF), and it is particularly useful for small-scale, very complex wind regimes. Because of the rapidity of solution to steady state, its potential for use in emergency response, short horizon forecasting, and lengthy records of hour-specific air quality simulation are discussed using specific examples. The rapid CFD modeling approach allows wind and turbulent kinetic energy (TKE) generation at resolutions of the scale of meters to hundreds of meters where there is a distinct lack of modeling tools for air quality and limited on-site meteorological observations that may not be representative. A discussion is conducted regarding its value to near-source air quality modeling in complex terrain and around nearby non-blunt obstacles. We discuss the limits to the modeling approach, but stress its improvement over other ad hoc parameterizations currently used in place of direct modeling.

Full text

1
Practical Uses of the WindStation Computational Fluid
Dynamics (CFD) Model in Air Quality Dispersion Studies
Control #51
Gary Moore, Robert J. Paine
AECOM, 250 Apollo Drive, Chelmsford, MA, 01824, USA
António Manuel Gameiro Lopes
ADAI-LAETA, Dept. Mechanical Engineering, University of Coimbra, 3030-788, Coimbra, Portugal
ABSTRACT
In this paper, we present several example applications and an evaluation scheme for an operational
software tool that simulates wind flow over complex topography using Computational Fluid Dynamics
(CFD) principles. This particular application software, named “WindStation1”, is an example of a
diagnostic simulator that solves the full 3D Navier-Stokes equations, including turbulence and thermal
effects, using a boundary-fitted coordinate system. As currently implemented, it can be run as a
Windows application based on a user-friendly graphical interface. This software is referred to below as
one in a class of models that can be characterized as “rapid CFD models”.
The authors have adapted output from WindStation to provide for direct input into an air quality
modeling system such as CALPUFF via the CALMET format. In that regard, this approach is an
alternative to CALMET as well as the Mesoscale Model Interface Program (MMIF), and it is
particularly useful for small-scale, very complex wind regimes. Because of the rapidity of solution to
steady state, its potential for use in emergency response, short horizon forecasting, and lengthy records
of hour-specific air quality simulation are discussed using specific examples. The rapid CFD modeling
approach allows wind and turbulent kinetic energy (TKE) generation at resolutions of the scale of
meters to hundreds of meters where there is a distinct lack of modeling tools for air quality and limited
on-site meteorological observations that may not be representative. A discussion is conducted regarding
its value to near-source air quality modeling in complex terrain and around nearby non-blunt obstacles.
We discuss the limits to the modeling approach, but stress its improvement over other ad hoc
parameterizations currently used in place of direct modeling.
INTRODUCTION
All CFD models have either grid or turbulence formulations that are limited in one form or another. It is
an imperfect world requiring the expert to make the best choices as to what formulations and grids are to
be applied to a specific problem. The air quality modeler is, because of near-source transport and
dispersion concerns, forced to depict wind flow regimes in what is often called the ‘terra incognita’ of
air science; a zone where the assumptions of meteorology give way to the local scale boundary layer
flow determined by obstacle driven flow features (e.g., 10 m to 1 km). This is often called the ‘human’
scale and requires developing atmospheric boundary layer models2 reaching up from typical CFD
applications and/or down from prognostic meteorological models. This is an area where diagnostic and
empirical models based on simple mass consistency (and potential flow) like CALMET described in
Scire3 et al. or the multi-grid model of Wang4 et al. and nearly a dozen others have been used for air
dispersion studies. However, many users of such models feel that more dynamical consistency, such as

2
that found in the RANS CFD models, might be a good option to have when developing flow fields for
modeling at the ‘human’ scale.
The use of CFD for atmospheric problems has been critically reviewed by Bitsuamlat5 et al. and
Cochran and Derikson6 among many others. How to apply CFD models in a dispersion modeling
program has been the topic of past AB-3 committees (e.g., Petersen and Huber7). A common theme
touched on is the issue of ‘real’ versus ‘apparent’ realism of CFD-generated flows. There are real
concerns about factors such as eddy detachment points and the location of flow separation streamlines
that have been examined in evaluations like that of Furbo8. When dealing with such issues, the modeler
has to grapple with multiple modeling options such as selecting the best TKE model for the modeling
scenario. In some cases, the limits as to how a mesh informs the model of a surface versus an obstacle
become a potential limiting factor.
Recognizing such issues and steering clear of factors such as urban canopies and arrays of ‘blunt’
obstacles, there are still quite a number of applications that can be done with CFD models that are not so
generalized as a model like FLUENT9 and which may be far preferable to the more ‘empirical’
diagnostic model methods. This paper discusses a highly efficient Reynolds Averaged Navier Stokes
(RANS) “rapid CFD model” that can be usefully applied to atmospheric flow fields in regions of
complex terrain where slope angles can be rather large, but not infinite. The combination of numerical
efficiency and current personal computer power is great enough that serial time series of steady-state
solutions can be developed (e.g., hour-by-hour) and the graphical user interfaces (GUIs) are sufficient to
be called an ‘app’ – commonly known as a user personal application.
The specific rapid CFD modeling tool discussed in this paper is the RANS CFD model WindStation
described by Lopes10 and documented in Lopes1. This paper has several objectives. One is to illustrate
a range of model applications where it can develop specific types of wind flow response that is of
significant use to dispersion modeling. Another is to illustrate how the model can be operated in
conjunction with a diagnostic model to improve the realism of the modeled flow. This may prove useful
when simple air quality models approaches run up against ‘straight face’ tests of realism in complex
flows.
OVERVIEW
WindStation is a 3-dimensional CFD model utilizing the SIMPLEC solver approach based on an
engineering formulation popularized by flow modeling experts like Bakker12. In this respect, it is a
highly simplified version of FLUENT. Among its physical modeling features are the following:
 A structured grid is used with telescoping in the vertical like weather models
 Turbulent kinetic energy (TKE) is simulated with versions of the ‘kappa-epsilon’ model
 Terrain roughness is taken into account through the ground shear stress modeling
 Atmospheric stability (domain constant lapse) is modeled by solving the energy equation
 Implied heat transfer at the ground by surface temperature imposition or heat flux imposition
(new).
The structured grid achieves an enhanced level of efficiency by the use of diastrophism, where the
terrain surface is initially ‘grown’ over a few initial iterations. The terrain-following grid introduces
slope terms, which in the case of perfectly blunt objects will become ill-defined due to the mesh
distortion, but (as will be shown) allows for a wide variety of application.

3
Speed aside, perhaps one of the most useful features of a rapid CFD modeling application lies in the
ability for a user to get information in and out of the model in a transparent manner. This is possible via
input-output text formatting and simple text control files that can be manipulated by either a GUI or a
text editor. The software application also allows the user to visually monitor the major term residuals as
a solution evolves and to control numerical criteria like the level of residuals for convergence, the max
number of iterations, under-relaxation coefficients and so forth.
Besides a neutral lapse rate, WindStation can apply either a stable or unstable temperature lapse rate.
Because a full energy model was not implemented in older versions of the model, an open upper
boundary would lead to convergence issues. Therefore, the model is usually exercised with a closed lid
far enough away from the surface so as to not interfere significantly with features like separating
streamlines that propagate vertically. A down side is that the careful user generally has to make a
number of sensitivity simulations to see what effect the input choices have on the model solution.
Fortunately on today’s workstation platforms, these sensitivity runs usually get done within a few
minutes.
Example 1: An Open Pit Mine
As its wind solver name CANYON suggests, the WindStation model is suited to modeling frequent re-
circulating flows in deep open pit mines like those described in Collingwood13 et al. Open mines get
quite deep and their walls’ aspect ratio make it unwise to apply a diagnostic model to simulate the flow
within such areas that themselves may be bounded by 1-3 km of extent. The topography of our example
open pit mine is shown in Figure 1. The terrain is resolved to 90 m. In this example, the 10-m winds
are presented under neutral conditions with westerly winds specified over a 500-m planetary boundary
layer (PBL) with a 6 m/s 10-m wind and a 12-m/s wind from the west at 500 m above the surface. The
resulting vertical cross section of flow shown in Figure 2 on the line shown in Figure 1 exhibits a
pronounced and robust re-circulation eddy within the mine.
Another feature of the 10-m wind field in Figure 1 is the appearance of a significant northerly wind near
the bottom of the pit. Also notable is the speed of the wind over the plateau of tailing to the west of the
pit. The convergence on the western lip and the divergence on the easterly lip can be noted.
No explicit data observations exist for a within-pit evaluation of the winds. However, Figure 3 displays
an interesting satellite image of a dust plume within a large open pit mine in the Atacama Desert that has
approximately the same dimensions and orientation. This Google view was taken recently in the
morning with winds in the area out of the west northwest. While the older version of WindStation does
not simulate the possible thermally driven flow component due to differential shading and solar heating
within the pit, the view of apparent dust plume movements is intriguing as evidence of the within-pit
flow field.
This recirculation flow is present even at a model resolution of 500 m. CALMET was run with a 500-m
resolution as well and used to drive a refined WindStation field using the restart option. This process is
encapsulated in the CAL2WND and WND2CAL conversion software described in Moore14, which
allows a WindStation dynamically-augmented CALMET formatted meteorological fields to be
developed within the constraints of the grid interpolation required. In the case of the open pit mine, the
flow separation at the top of the mine (analogous to plastic wrap stretched over the top) results in a time
averaged (annual) area PM10 emission ejection pattern from the lips of the pit that resembles a
doughnut/horseshoe like that shown in Figure 4.

4
Figure 1. An open pit copper mine showing the topography (2100m – blue and 3100 m –red) at 90
m resolution. Wind vectors at 10m are shown for a west wind case under neutral conditions. The
line indicates where an X-Z slice of wind vectors was taken.

5
Figure 2. An example cross-section (X,Z) of winds generated by WindStation within an open pit
mine under neutral conditions (6 m/s at 10 m) with west winds. The model grid is illustrated.

6
Figure 3. A recent Google Earth view of dust within the Chiquicamata open copper pit on a day
with regional westerly winds around 10 AM.

7
Figure 4. An example of the annual pattern of within-pit PM10 emissions exiting as an area source
at the top of the mine pit.

8
Example 2: Mineral Piles
The previous example explored a horizontal cell size of 100-500 m. Can the same model be used to
describe the flow over and around a mineral pile where the cubic cell size is 1-2 m on a side? The topic
of CFD-predicted wind speeds over arrays of mineral piles of various shapes has been extensively
reported in the literature. The work of Turpin and Harion15, Cong16 et al., and Yeh17 et al. are notable.
Many of these studies are put forth as engineering solutions that examine the efficacy of porous barriers
or berms in reducing and controlling the winds within a mineral stock yard where piles of various
configurations are built. The issue is again that of windblown dust migrating to nearby areas. Currently,
several methods for barrier development are being explored, and the most general is expected to allow a
non-ziggurat barrier to be applied (beginnings of a blunt obstacle capability).
A simple grid generation program was developed that implements line conveyors to lay out piles of coal
in much the same fashion like loaves of unbaked bread. The elongated lozenges are constructed with
simple functions to make complex surfaces of slightly flat topped piles. A number of yard
configurations have been studied. A four-pile configuration is shown in Figure 5 with a cross-crest wind
flow at 1 m above the surface over the piles. From this figure, one can note from the slightly stable
wind flow the following:
 Wind diversion and speed-up around the pile ends
 Wind speed-up over the crest of the piles
 Lee horizontal wind shading wake with vortices behind piles
Figure 5. An example of cross crest wind flow 1 m above the surface in and around 25 m high flat
topped coal piles under slightly stable conditions. Wind speeds at 1 m at numbered points are
summarized in Table 1.
This wind pattern was produced with a less than 10-minute simulation time on a single CPU PC. This
should be compared versus a FLOW-3D18 commercial simulation that required the better part of a day
on a powerful multi-core workstation, but gave qualitatively the same result on viewing. The flow field
is qualitatively quite similar to those found in the literature.

9
The WindStation program can visually magnify cross section simulations as illustrated in Figure 6,
which shows a vertical cross-section plot of the wind vectors near the surface under isothermal
conditions. The plot shows a vertically oriented rotor in the upwind pile lee as well as the crest speed-
up. Table 1 gives 1-m and 10-m wind speed ratios at the four points marked. In this stable case, the
separation zone remains just above the top of the pile, but as can be seen in Figure 5, extends several
pile widths downwind where it can interacts with piles further downwind. The lack of downwind
shielding on the second pile appears to be a consequence of both stability, strong vertical shear (a PBL
height of 150 m), and a tendency for WindStation to conservatively produce slightly less shielding than
a model like FLOW-3D at the same horizontal resolution. Using a realizable K-E TKE formulation
reduces this difference.
Figure 6. An uncluttered blow up of the wind vectors over one coal pile illustrating the upslope
speed-up to the crest and the corresponding lee wind shading and formation of a rotor in the lee of
the pile.
Table 1. A summary of wind speed-up factors and wind turning at various points in the wind
fields at 1 m above the surface.
Variable at 1 m
Ref (1)
Pt 2
Pt 3
Pt 4
Speed factor
(3.7) 1.0
2.6
1.683784
0.894595
direction
180
189
198
80
Speed factor
(3.7) 1.0
0.410811
1.897297
0.308108
direction
269
287
227
180
Speed factor
(3.7) 1.0
0.378378
1.927027
0.294595
direction
225
288
273
327
(U) speed in m/s - direction in degrees

10
The pile crest angle of approaching wind attack has been identified as the worst for dust, since pile self-
sheltering plays a far smaller role in reducing the wind over large areas of the stockyard. This is
confirmed in Figure 7a, which under isothermal conditions shows only a limited area of slow-down in
the lee and immediately upwind of the head of the pile. The four marked points suggest in Table 1 that
the wind variations are smaller than in the cross-crest case. In all cases, the crest speed-up is of the
order of a factor of 2 or greater similar to that reported in the literature while the lee reduction in speeds
are a factor of three or greater. Figure 7b show an intermediate angle of attack from out of the
southwest. The lee minimum rotates almost as if it were a coherent object. The pile approach slow-
down, the crest speed-up, and lee shadowing is similar in Table 1 to the westerly case; the major
difference being where the crest maximum speed-up occurs.
Figure 7a. An example of along crest wind flow 1 m above the surface in and around 25 m high
flat topped coal piles under slightly stable conditions. Wind speeds at 1 m at numbered points are
summarized in Table 1.
Figure 7b. An example of southwest diagonal to crest wind flow 1 m above the surface in and
around 25 m high flat topped coal piles under slightly stable conditions. Wind speeds at 1 m at
numbered points are summarized in Table 1.

11
The modeling of barriers immediately illustrated the limitations of WindStation as currently
implemented. When the absolute values of the terrain slope become too large (mesh distortion becomes
too great), the solution becomes unstable or crashes. One can make ziggurat barriers by making the
steps small enough to keep the slopes small enough. Currently, even this is tricky since the grid
telescopes in the vertical – the higher the wall, the wider the steps have to become. Sensitivity analysis
suggests that problems can happen when the slopes grow beyond a 3-to-1 ratio. Model development is
currently underway to potentially mix terrain and blunt obstacles using techniques like the immersed
boundary method (ibm). This is, of course, important if one wants to deal with terrain with cliffs, true
canyons, or even the prior example of open pits mines with stepped rather than smooth (aerodynamic
roughness-allowed) walls.
Example 3: Islands and Promontories
There have been numerous evaluations of CFD wind models on isolated hills, with Askervein Hill19
being one of the more referenced experimental sites. Apart from the usual statistical evaluation
procedures, there exists the possibility of using wildfire scarring to get an estimate as to where winds
may have pushed a fire over a significant area and period of time. The thesis work of Forthofer20
evaluated this approach for several fires in really complex terrain. His findings suggest that CFD
models do better than mass consistent models and far better than large scale uniform winds.
Catalina Island with its well-known 2007 fire scar immediately comes to mind. Catalina Island, which
lies just off the coast from Los Angeles, literally lunges out of the Pacific Ocean with the terrain shown
in Figure 8. The island is pockmarked with valleys and ridges. The island has been instrumented by
Desert Research Institute with the Catalina Island Automated Climate Network21. There are 8 sites with
2-m meteorology, an airport ASOS site (KAVX), and a 10-m tower at the Avalon School. This database
is archived by MesoWest in the MADIS system. Data at some of the automated sites is collected every
15 min. The local area is also modeled down to 4-km resolution by numerical weather models. Also,
due to the fact that Avalon is a high-visitor destination area from Los Angeles, there are several on-line
galleries with considerable photography of the complex air flows.
Anecdotal evidence aside, Figure 8 shows the tremendous variation in the wind roses at the instrumented
sites. In such an environment, a straight line trajectory is hard to come by. In the middle of the day, the
air flow at 10 m displays considerable terrain and thermal forcing. Figure 9 presents a WindStation
simulation example of the terrain steering for a single hour of winds out of the northwest. The
WindStation model is driven by a coarser downscaled wind and TKE field obtained from the Rapid
Update Cycle (RUC) model at 20-km resolution.
Initially, an attempt was made to use the CALMET model as a downscaling tool; in effect providing
more local information. This effort backfired, since one must be careful to sufficiently smooth the field
before introducing it to WindStation or else instability arises when the mass consistent solution is really
quite different from the RANS. Instead, a simple (and fast) inverse distance scheme was used directly
on the RUC meteorology to interpolatively map the RUC winds and TKE to the WindStation coarse
grid. This approach appears to work much better as illustrated by Figure 9 and will allow a classic
‘take-away one’ exercise with the observation station data to be performed in order to determine the
degree of robustness the CFD field has for each of the observing sites.

12
Figure 8. Catalina Island terrain and wind roses from the MesoWest data base for Jan1 through
April 30, 2009. (Meteorological data was provided courtesy of MADIS.)

13
Figure 9. An example of a lee eddy forming off Avalon when a prognostic model specified
convergence zone makes its way across Catalina Island. Time of day is 1300 LST on July 1, 2008
and resolution is 120 m.
Example 4: Synthetic Wind Rose
A final example addresses a case of determining a wind rose in an area for which a tower-derived wind
rose is not available. The site is located off the coast of British Columbia where the terrain is extremely
complex. WindStation was used to derive numerous 22.5-degree sector centered wind fields for a year
driven by the archived 6-hourly 10- North American Mesoscale (NAM) model output at 12 km. These
were then used to build a synthetic wind rose based on the nearby NAM predictions. Figure 10
illustrates the level of detail in the wind flow in the modeling domain for a SE wind direction, one of the
most frequent in that area. Prince Edward lies at the mouth of a glacial moraine valley to the east.
Prince Rupert lies behind a high solid rock bluff to the immediate south. Digby Island is a flat, near
ocean, glacially sheared off island. The winds track in and out of the river valleys, bays, and sounds.

14
Figure 10. An example wind field generated from WindStation for a southeasterly wind flow
around Prince Rupert, British Columbia

15
CONCLUSIONS AND RECOMMENDATIONS
When limited to situations where there are no ‘blunt’ obstacles such as vertical cliffs and canyons or
windbreaks, an efficient CFD model such as WindStation appears able to provide useful information on
‘human’ scale wind flows ranging from meter scales up to several kilometers. Several of the example
applications are relevant to the mining industry, where dust generation and dispersion is a primary
concern. As shown from the examples, a rapid CFD model can simulate
 Deep pit mechanically driven vertical rotors,
 Obstacle lee wind shadowing and horizontal rotors,
 Upwind speed reductions and crest wind speed-ups, and
 Steered flow around steep obstacles and upslope and downslope valley flow.
A rapid CFD model can be combined with meteorological model downscaling software, allowing
linkage from prognostic models and CALMET into a model restart file. The model output has been
reformatted in the CALMET and AERMET formats for use in air quality modeling, and its output has
been used to create formatted emission files for CALPUFF inputs. The batch processing allows for
rapid generation of hourly files and for the generation of a large number of meteorological states for
time series modeling.
Currently, this type of rapid CFD model can be used by mining industries and for applications dealing
with complex terrain flow modeling in support of dispersion modeling. Research is currently underway
to allow objects requiring internal boundary conditions to be modeled, paving the way for near-source
rapid CFD modeling of building complexes, allowing a rapid emergency response and on-site rapid
evaluation of air concentrations for planning purposes.
REFERENCES
1. Lopes, A.M.G. WindStation Version 3.0 – Users Manual, http://www.easycfd.net/windstation/ (2011)
2. Allen, T. Flow over hills with variable roughness, Boundary Layer Meteorology 121:475-490.
(2006)
3. Scire, J. S., F.R. Robe, M.E. Fernau, and R. J. Yamartino. A User’s Guide for the CALMET
Meteorological Model, Earth Tech, Concord, MA 01742. (2000)
4. Wang, Y., C. Williamson, D. Garvey, S. Chang and J. Cogan. Application of a multi-grid method to
a mass-consistent diagnostic wind model, J. of Applied Meteorology, 44:1078-1089. (2005)
5. Bitsuamlak, G. T., T. Stathopoulos, F. ASCE and C. Bedard. Numerical evaluation of wind flow
over complex terrain: Review, J. of Aerospace Engineering, 17:135:-145. (2004)
6. Cochran, L. and R. Derickson. A physical modeler’s view of computational Wind Engineering,
Fifth International Symposium on computational Wind Engineering (CWE2010) Chapel Hill, NC
USA. (2010)
7. Petersen, R. and A. Huber. A&WMA AB3 Comments on Nonstandard Modeling Approaches, 8th
Conference on Air Quality Modeling, Air Waste Management Association. (2005)
8. Furbo, E. Evaluation of RANS turbulence models for flow problems with significant impact of
boundary layers, UPTEC F10061, University of Uppsala, ISSN: 1401-5757. (2010)
9. Lopes, A.M.G., Sousa, A.C.M., Viegas, D.X., “Numerical simulation of turbulent flow and fire
propagation in complex terrain”, Numerical Heat Transfer, Part A, N. 27, pp. 229-253. (1995)

16
10. ANSYS FLUENT Flow Modeling Software. Available at
http://www.ansys.com/Products/Simulation+Technology/Fluid+Dynamics/Fluid+Dynamics+Product
s/ANSYS+Fluent.
11. Lopes, A.M.G. - "WindStation - A software for the simulation of atmospheric flows over complex
topography", Environmental Modeling & Software, Vol.18, N.1, pp. 81-86. (2003)
12. Bakker A. The Colorful Fluid Mixing Gallery, http://www.bakker.org/cfm. (2012)
13. Collingwood, W, K.V. Raj, A Choudhury, and S. Bandopadhyay. CFD modeling of air flow,
Mining Engineering, 64(2):44-57. (2012)
14. Moore, G. E. A Users Guide to WindStation Pre- and Post-Processors with Illustrative Examples,
AeroKinetics Document 2013-02. Seabrook, NH 03874. (2013)
15. Turpin, C. and J.-L. Harion. Numerical Modeling of flow structures over various flat-topped
stockpiles height: Implications on dust emissions Atmospheric Environment, 43:5579-5587. (2009)
16. Cong, X.C., S.Q. Cao, Z.L.Chen, S.T. Peng, and S.L.Yang. Impact of the installation scenario of
pourous fences on wind-blown particle emission in open coal yards, Atmospheric Environment,
45(30):5247-5253. (2011)
17. Yeh, C. P., H. Tsai. And R.J. Yang. An investigation into the sheltering performance of porous
windbreaks under various wind directions, J. of Wind Engineering and Industrial Aerodynamics,
DOI: 10.1016/j.jweia.2010.04.002. (2010)
18. FLOW-3D software documentation. Available at http://www.flow3d.com/.
19. Taylor, P.A. and H. W. Teunissen. The Askervein Hill project: Overview and background data.
Boundary-Layer Meteorology, 39: 15-39. (1987)
20. Forthofer, J.M. Modeling Wind in Complex Terrain for Use in Fire Spread Prediction, Masters
Thesis, Colorado State University, Ft. Collins, CO, pp 1-123. (2007)
21. Catalina Island Automated Climate Network. Available at
http://www.dri.edu/component/content/article/180-dees-extended-research/3064-catalina-island-
automated-climate-network.
KEYWORDS
Computational Fluid Dynamics, Rapid CFD models, complex winds, WindStation, model evaluation