Accepted Manuscript Performance analysis of an air quality CFD model in complex environments: Numerical simulation and experimental validation with EMU observations Pramod Kumar, Amir-Ali Feiz PII:
S0360-1323(16)30308-0
DOI:
10.1016/j.buildenv.2016.08.013
Reference:
BAE 4597
To appear in:
Building and Environment
Received Date: 3 May 2016 Revised Date:
10 August 2016
Accepted Date: 11 August 2016
Please cite this article as: Kumar P, Feiz A-A, Performance analysis of an air quality CFD model in complex environments: Numerical simulation and experimental validation with EMU observations, Building and Environment (2016), doi: 10.1016/j.buildenv.2016.08.013. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Pramod Kumara,∗, Amir-Ali Feiza
´ LMEE, Universit´e d’Evry Val-d’Essonne, 40 Rue Du Pelvoux 91080 Courcouronnes, France.
SC
a
RI PT
Performance Analysis of an Air Quality CFD Model in Complex Environments: Numerical Simulation and Experimental Validation with EMU Observations
M AN U
Abstract
Assessment of air quality in complex industrial and urban regions is essential in setting effective policies for sustainable development of society and assisting to improve the environment and thereby support health and safety. In this study, a CFD model fluidyn-PANACHE, dedicated to the dispersion
D
of hazardous gases in complex industrial and urban sites, is critically vali-
TE
dated with tracer observations from EMU experiment for two distinct types of sources in two geometrically different environments: (i) Case-A1 (release from an open door in courtyard area of a simple L-shaped building), and
EP
(ii) Case-C1 (continuous release over the larger distances around an industrial site featuring numerous buildings and complex local topography). The
AC C
steady RANS solution is used first to get quick results and comparison with the experimental data shows a good agreement between the modelled and measured concentrations. Nevertheless, in order to examine the performance of the solver, unsteady RANS simulations are performed. A detailed statisti∗
Corresponding author Email address:
[email protected] (Pramod Kumar )
Preprint submitted to Building and Environment
August 10, 2016
ACCEPTED MANUSCRIPT
cal analysis shows that the performance of fluidyn-PANACHE against both
RI PT
Cases of EMU observations is well within the acceptable bounds of statistical measures for air quality applications. It is observed that in a simple
geometry with lesser complexity in Case-A1, steady and unsteady simulations give similar performance against the tracer observations and predicts
SC
≈ 68% points within a factor of two. In a more complex industrial site in Case-C1, unsteady RANS is performing slightly better than the steady sim-
M AN U
ulation and it predicts ≈ 80.0% points within factor of two, which is higher than ≈ 67.0% points from the steady simulation.
Keywords: CFD modeling, EMU experiment, fluidyn-PANACHE, Model evaluation, RANS and URANS, Urban dispersion modelling.
1
1. Introduction
Air quality assessment in complex urban or industrial regions with new
3
and existing techniques is imperative to provide the information needed to
4
estimate the population exposure to hazardous airborne matter and to assist
5
regulators, urban planners or emergency authorities to outline the evacuation
6
plans in a case of the natural disasters, accidental or deliberated release. In
7
industrial safety and environment programmes, near-field (< 10 km) disper-
8
sion of toxic and hazardous gases in complex chemical sites and surrounding
10
11
TE
EP
AC C
9
D
2
the buildings and other structures are often predicted with the Computational Fluid Dynamic (CFD) codes. A CFD model solves the Navier-Stokes (N-S) fluid dynamics equations using a small grid size (of order 1 m or even
12
less) [1] over the complex terrains. With rapid advances in computer re-
13
sources and methods, CFD models provide accurate flow fields and pollutant
2
ACCEPTED MANUSCRIPT
dispersion modelling around buildings and other structures in complex urban
15
or industrial regions for various kind of release scenarios.
RI PT
14
The unsteady flow effect triggered by the presence of buildings and other
17
geometries is an influencing factor on the dispersion in urban or industrial
18
regions. Compared with simple Gaussian dispersion models or other analyti-
19
cal or empirical approximations, CFD models efficiently predict the obstacles
20
influence on wind patterns and pollutant cloud shapes in complex urban envi-
21
ronments [2]. Nevertheless, the CFD model validation against experimental
22
datasets is one critical point to examine its capability to provide reliable and
23
valuable information in emergency planing or chronic impact assessment.
24
Many CFD codes were successfully validated against experimental data for
25
various types of releases in diverse geometric environments and atmospheric
26
conditions [1, 3, 4, 5, 2, 6, 7, 8, etc.]. Some recent comprehensive review
27
studies, e.g., [9, 10, 11, 12, 13, 14, 15, etc.], highlight the applicability, ad-
28
vantages, disadvantages, limitations, and problems in the CFD modelling for
29
atmospheric dispersion in urban or industrial environments. However, to ob-
30
tain the reliable results for emergency preparedness and air quality analysis,
31
CFD models are required to be set up, tested, and evaluated correctly with
32
the experimental observations in different geometric environments and atmo-
33
spheric conditions [16, 9, 17, 18]. The validation of a CFD model is also an
35
36
M AN U
D
TE
EP
AC C
34
SC
16
essential part to utilize it with an inversion method for reconstruction of the unknown atmospheric tracer sources in urban or complex industrial regions [19, 20, 21].
37
Complexities in flow fields and pollutant transport and dispersion in ur-
38
ban or industrial regions make difficult to assess the contaminant plume
3
ACCEPTED MANUSCRIPT
concentrations due to local topography, terrain conditions, buildings and
40
other geometrical structures. Buildings in an urban region alter the flow-field
41
and deflect the wind, causing updrafts and downdrafts, channelling between
42
buildings, areas of calm winds adjacent to strong winds, and horizontally
43
and vertically rotating-eddies between buildings, at street corners, and other
44
places within the urban canopy [22]. The physics in urban regions is often
45
characterized by unsteadiness due to the presence of a complex assembly
46
of bluff bodies and to the variability of approaching winds [23, 12]. Often,
47
a question also arises as to whether the steady Reynolds-averaged Navier-
48
Stokes (RANS) solution would perform equally well with the unsteady RANS
49
(URANS) in various geometrically different environments. The basic equa-
50
tions of the URANS are formally derived by applying ensemble averaging.
51
Only with ensemble averaging the resulting equations comply with the steady
52
RANS equations now containing the partial time derivatives in URANS codes
53
[24]. URANS models are driven by time-dependent boundary conditions and
54
account, e.g., for different land uses with different radiation budgets and pro-
55
vide time dependent predictions [24]. This type of model is used in standard
56
meteorological meso- and macro-scale applications (weather forecasts etc.).
57
However, selection of proper solution strategies for complex urban disper-
58
sion applications includes choices between the steady RANS and URANS
60
61
SC
M AN U
D
TE
EP
AC C
59
RI PT
39
approaches [12]. Thus, one of the objective of this study is to evaluate the comparative performance of the steady and unsteady RANS solutions for pollutant dispersion in two geometrically different regions.
62
This study is a step towards the development of a methodology that will
63
utilize a CFD model fluidyn-PANACHE to retrieve an unknown atmospheric
4
ACCEPTED MANUSCRIPT
tracer source or leak in an industrial region from a finite set of the concen-
65
tration measurements. However, due to uncertainties in the synoptic wind,
66
boundary conditions, model errors or parametrization uncertainties, a pre-
67
cise validation of the CFD models is a necessity to obtain the reliable results
68
for emergency preparedness. For this, fluidyn-PANACHE required to be first
69
validated for forward dispersion in an industrial region so that we can utilize
70
the information based on this in a inversion process for source reconstruction
71
in an industrial or urban region. Thus, this study concerns a comprehensive
72
validation of the fluidyn-PANACHE against the experimental observations
73
in two geometrically different environments.
M AN U
SC
RI PT
64
In order to evaluate the CFD simulations of urban flows and pollutant
75
dispersion, completeness and reliability of an experimental data is a neces-
76
sity. CFD validation in turn requires high-quality experimental data to be
77
compared with the simulation results. This study utilizes the extensive field
78
observations involving tracer gas releases in a wind tunnel from the Evalua-
79
tion of Modelling Uncertainty (EMU) project. The EMU project provides an
80
unique and useful dataset for validation of dispersion models around build-
81
ings, industrial regions and over complex topography [25]. To demonstrate
82
the fluidyn-PANACHE model’s capabilities to simulate the flow and disper-
83
sion patterns in the near field but also at larger distances, two cases (i) the
85
86
TE
EP
AC C
84
D
74
single L-shaped building case and (ii) the real industrial site case are selected. These EMU cases are also useful to analyse the CFD model performance from two different types of releases, such as from area and point sources.
87
In this study, three-dimensional numerical simulations are performed to
88
evaluate a CFD model fluidyn-PANACHE with tracer observations from an
5
ACCEPTED MANUSCRIPT
experiment, for dispersion of toxic and hazardous gaseous pollutants around
90
buildings and in geometrically complex chemical sites. The steady RANS
91
solution is used first to get quick results and comparison with experimental
92
data. Nevertheless, in order to improve performance of the solver, unsteady
93
RANS solution is tried and compared with results from steady simulations.
94
2. Description of the CFD model
SC
RI PT
89
Various CFD approaches, e.g. large-eddy simulation (LES), direct numer-
96
ical simulation (DNS), steady RANS, URANS, and hybrid URANS/LES, etc.
97
have been used for basic research on urban flow and dispersion. However,
98
deciding which CFD approach is most appropriate for a given problem is not
99
always straightforward, as each approach has specific advantages and disad-
100
vantages and up to now, RANS has been the most commonly used approach
101
in CFD for urban physics [11, 12]. Thus, a CFD model fluidyn-PANACHE,
102
which includes both steady and unsteady RANS solutions, is applied here.
103
2.1. fluidyn-PANACHE
TE
D
M AN U
95
A CFD model fluidyn-PANACHE was developed for the simulation of
105
atmospheric flows and pollutant dispersion from various types of single and
106
multiple sources in complex environments. The fluidyn-PANACHE is a self-
108
109
110
AC C
107
EP
104
contained fully 3-dimensional (3-D) fluid dynamics, commercial CFD code, designed to simulate accidental and industrial pollutant dispersion into the complex terrains in the presence of obstacles. A description of the model and its applications are given in [26, 27, 28, 2]. The fluidyn-PANACHE
111
solves the N-S equations along with the equations describing conservation
112
of species concentration, mass, heat transfer and energy for a mixture of 6
ACCEPTED MANUSCRIPT
ideal gases using finite volume numerical techniques. The CFD model solves
114
the Reynolds averaged forms of these governing differential equations in 3-D
115
space and time for turbulent flow and dispersion in a computational domain.
116
It includes solutions of both steady and unsteady forms of the RANS equa-
117
tions. The steady RANS refers to time-averaging of the N-S equations and
118
yields statistically steady descriptions of turbulent flow. However, flow in the
119
atmospheric boundary layer (ABL) is inherently unsteady, and therefore, an
120
unsteady approach may required. URANS refers to ensemble-averaging of
121
the N-S equations and resolves only the unsteady mean-flow structures, while
122
it model the turbulence. URANS can be a good option when the unsteadiness
123
is pronounced and deterministic.
M AN U
SC
RI PT
113
The Reynolds stresses are modelled using a linear eddy viscosity model
125
[29]. Ideal gas law is used for the thermodynamic model of mixture of gases.
126
Air is modelled as the moist air with effective properties of the mixture of dry
127
air and water vapour. The ABL processes are built-in the CFD code with
128
different numerical models. Dispersion of gases is modelled by solving the full
129
conservation equations governing the transport of species concentration. It
130
includes a built-in automatic 3-D mesh generator that can create the finite-
131
volume mesh around obstacles and body-fitting the terrain undulations. A
132
detailed description of the fluidyn-PANACHE’s features is given in section
134
135
TE
EP
AC C
133
D
124
S1 of the Supplementary Information (SI). 2.2. Turbulence model Turbulent structure in the computational domain is resolved using a mod-
136
ified standard k − turbulence model in fluidyn-PANACHE. The k − model
137
is a 2-equation linear eddy viscosity model and it’s implementation is de7
ACCEPTED MANUSCRIPT
rived from the standard high-Reynolds number (Re) form with corrections
139
for buoyancy and compressibility [30, 31]. It solves the transport equations
140
for turbulent kinetic energy, k, and its dissipation rate, for adapted con-
141
stants. The k − model computes the length and time scales from the local
142
turbulence characteristics. Thus, it can model the turbulent flows subjected
143
to both mechanical shear (obstacles, terrain undulations, canopy) as well as
144
buoyancy (stability and buoyant/heavy gas plumes). The k − model is an
145
isotropic model of turbulence. Thus, it results in turbulent diffusivities that
146
are same in both horizontal and vertical directions at a location.
147
2.3. Boundary conditions
M AN U
SC
RI PT
138
Appropriate boundary conditions are required on the main computational
149
domain boundary, the ground, and on the obstacles. The top boundary is
150
treated as an outflow boundary. The lateral boundaries of the domain are
151
treated as inflow and outflow boundaries based on the direction of the wind
152
with respect to the domain boundary. At the inflow boundary, velocity, tem-
153
perature, species concentrations and turbulence variables are specified. Pres-
154
sure is extrapolated from inside the domain. Species concentrations are set
155
according to the specified background concentrations. The inflow boundary
156
conditions for wind, temperature, and turbulence profiles and wall functions
158
159
TE
EP
AC C
157
D
148
are briefly described in following subsections. 2.3.1. Wind and Temperature profiles The vertical wind profile is an important factor defining the structure of
160
the ABL in CFD simulations. In this study, the log-law profile in neutral
161
stability condition has been used to parametrize the inflow boundary condi8
ACCEPTED MANUSCRIPT
163
tion for the EMU Cases A1 and C1. The vertical wind profile u(z) is given by the Monin-Obukhov similarity theory, z u∗ ln − ψ(ζ) u(z) = κ z0
RI PT
162
(1)
where u∗ is the friction velocity, κ(= 0.41) is the von Karman constant, z
165
is the vertical height above the ground surface, z0 is the surface roughness
166
length, ψ is the similarity function depends on the atmospheric stability, and
167
ζ = z/L is the stability parameter in which L is the Obukhov length. For
168
neutral stability condition, ψ can be assumed to zero, that transformed the
169
wind profile u(z) in Eq. (1) to the standard log-law profile.
170
2.3.2. Turbulence profile
M AN U
SC
164
Several parametrizations in the fluidyn-PANACHE are implemented for
172
the inflow turbulence profiles. The profiles of k and used in this study is a
173
semi-empirical model based on the similarity theory and turbulence measure-
174
ments [32]. In neutral and stable regimes (z/L ≥ 0), the turbulence profiles
TE
according to [32] are determined as follows: 6u2∗ , for z/zi ≤ 0.1 k(z) = 6u2 (1 − z/zi )1.75 , for z/zi > 0.1 ∗
176
AC C
EP
175
D
171
(z) =
u3∗ (1.24 + 4.3z/L) , κz
for z/zi ≤ 0.1
(2)
(3)
u3∗ (1.24 + 4.3z/L) (1 − 0.85z/zi )1.5 , for z/zi > 0.1 κz
177
178
where zi is height of the ABL. In neutral stability condition, z/L → 0 and accordingly these profiles are transformed to apply in neutral regimes.
9
ACCEPTED MANUSCRIPT
179
2.3.3. Wall functions Standard wall functions in fluidyn-PANACHE [31] are used to compute
181
the drag forces on solid walls in a turbulent boundary layer. The wall func-
182
tions result from a solution of the N-S equations for the turbulent boundary
183
layer in equilibrium. A no-slip boundary condition, where velocity compo-
184
nents are set to zero at the ground surface, is defined for the bottom boundary
SC
condition. Following log law of the wall function for momentum is used: 1 ln(Ey + ), for y + > 11.63 κ + u = (4) + y + , for y < 11.63
M AN U
185
RI PT
180
where u+ = Up /u∗ is the non-dimensional velocity, in which Up is the fluid
187
velocity parallel and relative to wall, y + = ρu∗ y/µ is the non-dimensional
188
wall to cell-centre distance, in which y is the cell-centre to wall distance and
189
µ is the dynamic viscosity of fluid and E is a function of wall roughness [33].
190
2.4. CFD solver
TE
D
186
For both scenarios (Cases A1 & C1) of the EMU experiment, a fully im-
192
plicit NT CFD solver in fuidyn-PANACHE is used for flow-field computations
193
in the domains with unstructured mesh. The NT solver is a pressure-based
194
fully implicit segregated method on unstructured meshes that solves all gov-
196
197
AC C
195
EP
191
erning equations separately and uses an iterative procedure for both steady state and transient cases. The steady RANS solution is used first to get quick results and comparison with the EMU experimental data in both the
198
Cases. The selected parameters have been chosen for a fast convergence of
199
the simulations. Nevertheless, in order to improve performance of the solver,
10
ACCEPTED MANUSCRIPT
unsteady RANS solver is used for simulations of both EMU Cases. The solver
201
options have been chosen similar except the unsteady RANS solution.
RI PT
200
To couple the pressure and momentum equations in the numerical compu-
203
tations, the Semi-Implicit Method for Pressure Linked Equations-Consistent
204
(SIMPLEC or SIMPLE-Consistent) algorithm [34] is utilized. The buoy-
205
ancy model is used to parametrize the body force term in the N-S equations.
206
Dispersion of the gaseous pollutants is modelled by solving the standard Eu-
207
lerian advection-diffusion equation governing the transport and diffusion in
208
a computational domain. Residuals were used to check the convergence of
209
the solution during a computation and also to stop the simulations. For a
210
CFD simulation, the scaled residuals for all variables were considered to be
211
equal or less than O(10−4 ).
212
3. Description of the EMU experiment
D
M AN U
SC
202
The Evaluation of Modelling Uncertainty (EMU) project, funded by the
214
European Commission, involved a comprehensive evaluation of the models
215
to simulate flow and dispersion patterns in different realistic geometric envi-
216
ronments. The objective of this tracer experiment was to evaluate the spread
217
in results due to the way that CFD codes are applied and the accuracy of
218
such codes in complex gas dispersion situations [35, 25]. It records a com-
220
221
EP
AC C
219
TE
213
prehensive set of the concentration and other observations and is useful to setting up and validate the CFD models for dispersion in various realistic environments for different release scenarios. The EMU project consisted in
222
14 test cases of industrial scenarios, which ranged from single building on flat
223
terrain right through to the cases associated with a specific, complex topog11
ACCEPTED MANUSCRIPT
raphy industrial site. Stage A in the EMU project comprised three cases, Al
225
to A3, involving a simple building on flat ground, neutral atmosphere and
226
isothermal conditions. Stage B incorporated increases in complexity of the
227
geometry (i.e. terrain, obstacles and number of buildings), release conditions
228
(i.e. two-phase and non-isothermal releases) and meteorology (i.e. stabil-
229
ity and wind speed). Stage C concerned an actual industrial site, featuring
230
numerous buildings and complex local topography. Experiments were per-
231
formed at the University of Surrey [35] in a large stratified wind tunnel (20 m
232
× 3.5 m × 1.5 m) at a model scales between 1/133 and 1/250. Continuous jet
233
releases of dense, buoyant and neutrally-buoyant gases have been simulated
234
in neutral or stable atmosphere.
M AN U
SC
RI PT
224
In the present work, two test cases A1 and C1 of the EMU project are
236
simulated using fluidyn-PANACHE. Case A1 involves a release from an open
237
door in the courtyard area of a simple L-shaped building on a flat ground.
238
Whereas, Case C1 comprises a continuous point release over the larger dis-
239
tances around an industrial site featuring numerous buildings and complex
240
local topography. The complete description of each case is described in fol-
241
lowing sections for the numerical simulations.
242
3.1. Description of Case A1, computational domain, and grid structure
244
245
246
TE
EP
AC C
243
D
235
Case A1 of the EMU project entailed the numerical simulation of a passive
release from an L-shaped building (Figure 1) located on a flat surface into the ABL flow. The EnFlo wind-tunnel was modelled at the University of Surrey and neutral ambient conditions were assumed during the experiment
247
[35]. In Case A1, a continuous steady release of a neutrally buoyant gas took
248
place from a relatively large “courtyard” door in the side of building (Figure 12
ACCEPTED MANUSCRIPT
1). The source in this case is an areal source of 20 m2 area that releases the
250
neutrally buoyant gas into horizontal direction. In this case, a mixture of
251
2.96% ethylene (C2 H4 ) in a nitrogen (N2 ) balance was used for the source
252
gas, and was essentially neutrally stable. Properties of the source are listed
253
in Table 1. Velocity UH at building height was 5 ms−1 , with wind direction
254
θ = 0o . The ground roughness length (z0 ) of the experimental site in Case
255
A1 is 0.12 m [35]. In this case, the concentrations at crosswind locations
256
on the cross-section at five distances downwind (x) of the lee edge of the
257
L-shaped building were measured at x/H = 0.5, 1, 2, 5, 10; where H = 10
258
m is height of the building. At each sampling line, sensors were deployed
259
in crosswind and vertical directions (Figure 1). The height (z/H) of the
260
sensors were varying (z/H = from 0.11 to 3.0) at all crosswind arc (y/H) of
261
the downwind distances (x/H) from the source. Receptors information for
262
the available concentration measurements in the EMU Case A1 is given in
263
SI section S2 & Table S1.
TE
D
M AN U
SC
RI PT
249
Figure 1 also represents the computational domain considered for the
265
EMU Case A1. The dimensions of the computational domain have been set
266
as follows: 300 m long, 180 m wide and 120 m high. The distance between
267
the source and the inlet flow boundary condition is 85 m and the source
268
is located in the middle of the width of the domain. The computational
270
271
AC C
269
EP
264
domain is discretized in 39 irregular vertical levels. In order to select the appropriate mesh, a grid sensitivity analysis is performed with simulation test-cases using six different mesh sizes (SI section S3, Table S3, Figure S1).
272
The unstructured grid cells numbers in these six test-cases of the investigated
273
meshes were varying from 1461876 to 2740725 (Table S3). The grid mesh
13
ACCEPTED MANUSCRIPT
in the simulation domain is more refined at close to the buildings, source
275
and receptor locations. Based on this grid sensitivity analysis, a mesh size
276
consists a total of 2134080 unstructured grid cells in the 3-D computational
277
domain with 54720 cells on each horizontal plane is adapted for the current
278
study. For Case A1, min/max of dx (or dy) are 0.2 m/2.5 m; min/max of dz
279
are 0.3 m/20 m.
280
3.2. Description of Case C1, computational domain, and mesh structure
SC
RI PT
274
Case C1 of the EMU project entails a continuous, passive jet release from
282
the side of a building within a real chemical site. The surrounding terrain
283
in this Case C1 is complex, with steep hills, trench-like features and cliffs at
284
the edge of the sea. The site comprises a large number of irregularly shaped
285
buildings, most of which however are conveniently aligned with each other
286
(Figure 2). Cowan [36] reported four categories of the surface roughness in
287
the domain: sea, land, village/town and industrial site. The atmospheric
288
stability classes encountered are neutral and stable. In this case, we take our
289
origin to be at sea-level, directly below the source position. Three types of
290
concentration data (126 observations) were recorded: (i) cross-stream profiles
291
at ground-level, (ii) at (z−zg )/H = ∼ 2−3 (where zg is the height of ground-
292
level), and (iii) vertical profiles through the ground-level maxima. A number
294
295
296
D
TE
EP
AC C
293
M AN U
281
of other ground-level concentration measurements were also made during the experiment [35]. The source height in the Case C1 is 2.0 m and the tracer was released from a point source tilted in horizontal direction with 327.5◦ (Table 1). The product released of Case C1 is a mixture of 77.3% by volume C2 H4 in
297
N2 , giving a mixture density ratio α = 1.0. Mass flux, release temperature,
298
release duration, exit velocity and emission direction have been considered 14
ACCEPTED MANUSCRIPT
as inputs for a point source continuous release. The source characteristics of
300
this EMU Case C1 for a passive tracer are tabulated in Table 1. Receptors
301
information for the available concentration measurements in the EMU Case
302
C1 is given in SI section S2 & Table S2.
RI PT
299
Figure 2 also represents the computational domain for Case C1. The
304
domain dimensions have been set as 800 m long, 1200 m wide and 200 m
305
high. 3-D unstructured mesh was generated in domain for the Case C1
306
with 2041182 cells, which contains 52338 cells on each horizontal plane. 39
307
irregular vertical planes were created in the vertical direction. A grid sensi-
308
tivity analysis with more refined and coarse meshes was also carried out to
309
adapt the present mesh for further CFD simulations. The min/max of dx
310
(or dy) are 0.7 m/12 m; min/max of dz are 0.8 m/40 m for the EMU Case
311
C1. For comparison with the observations (volume fractions of the source
312
fluid), the predicted concentrations are normalized with Cn = UH H 2 /Cs Qs
313
and converted into volume fractions [36]. In this computation of the normal-
314
ization conversion in Case C1: UH = 5 ms−1 , H = 10 m, Cs = 0.773 and
315
Qs = πDs2 Us /4 = 17.8 m3 s−1 , in which source diameter (Ds ) = 1.74 m, and
316
exit velocity Us = 7.5 ms−1 .
317
4. Model performance validation methods
319
320
M AN U
D
TE
EP
AC C
318
SC
303
The variables to evaluate for an air quality analysis depend on the in-
tended goals of a model validation, e.g., the highest and second highest hourly-averaged ground level concentrations are of interest for regulatory
321
applications to standard pollutants and for military applications, it is typi-
322
cally the hazard areas of certain concentration or dosage thresholds [37]. The 15
ACCEPTED MANUSCRIPT
quantities such as the maximum concentration over the entire receptor net-
324
work, and the maximum concentration across a given sampling arc or along
325
a sampling line are the potential outputs of an air quality model that should
326
be validated. The mean concentration at some desired locations in a region
327
is also required to analyze the extent of exposure from a release. These are
328
the potential outputs of an air quality model that should be validated. The
329
cumulative distribution functions (cdf) where the prediction of probability
330
to have a particular health effect above a threshold have also been utilized
331
for air quality analysis [38, 6, 39, etc.].
M AN U
SC
RI PT
323
The performance validation of the CFD model is produced both via the
333
result analysis and direct comparison of numerical results versus experimental
334
observations. The validation is performed both qualitatively and quantita-
335
tively by the scatter plots, quantile-quantile (Q − Q) plots, and computing
336
the statistical performance measures. In a scatter plot, the paired observa-
337
tions and predicted concentrations are plotted against each other and the
338
magnitude of the model’s over- or under-predictions can be visible by visual
339
inspection. On the other hand, Q − Q plot begins with the same paired data
340
as the scatter plots, but removes the pairing and instead ranks each of the
341
observed and predicted data separately from lowest to highest [37]. A Q − Q
342
plot is often useful to find out whether a model can generate a concentration
344
345
TE
EP
AC C
343
D
332
distribution that is similar to the observed distribution. Biases at the low or high concentration values are quickly visible in the Q − Q plots. For model validation, standard statistical performance measures: Nor-
346
malized Mean Square Error (NMSE), Fractional Bias (FB), Factor of Two
347
(FAC2), Pearson correlation coefficient (COR), Geometric Mean bias (MG),
16
ACCEPTED MANUSCRIPT
Geometric Variance (VG), and Index of Agreement (IA) are computed and
349
analyzed. These measures define the agreement between the model pre-
350
dicted concentrations with the experimental observations. NMSE defines
351
the degree of scattering in a sample. The positive and negative values of FB
352
respectively indicate the overall under- or over-prediction of the simulated
353
concentration from the observations. FAC2 is the most robust measure, be-
354
cause it is not overly influenced by the outliers in observed and predicted
355
concentrations. Since the distribution is close to lognormal for most atmo-
356
spheric pollutant concentrations, the linear measures FB and NMSE may be
357
overly influenced by infrequently occurring high observed and/or predicted
358
concentrations, whereas the logarithmic measures MG and VG may provide
359
a more balanced treatment of extreme high and low values. However, MG
360
and VG are overly influenced by the very small and zero values, which are not
361
uncommon in dispersion modelling [37]. COR reflects the linear relationship
362
between two variables and sensitive to a few aberrant data pairs.
TE
D
M AN U
SC
RI PT
348
A perfect model would present the ideal values of NMSE, FB = 0; and
364
FAC2, COR, MG, VG, IA = 1 [37]. However, due to influence of random
365
atmospheric processes, these values are not attainable and minimum perfor-
366
mance measures for a model to be defined as acceptable [37]. For an acceptable
367
model performance, [37] provided following bounds of these measures:
369
370
AC C
368
EP
363
NMSE < 4.0, −0.3 < FB < 0.3, FAC2 > 0.5, VG < 1.6, 0.7 < MG < 1.3. Computation of these performance measures assumes that the validation
dataset contains pairs of predicted and observed concentrations. For a contin-
371
uous and constant release during an experimental period, distribution of the
372
modelled concentrations reaches steady-state in a computational domain and
17
ACCEPTED MANUSCRIPT
thus, the averaged values and end-time simulation concentrations are simi-
374
lar. Accordingly, the result of both RANS and URANS is the time-averaged
375
value of the concentrations in a duration at the ending of simulations. In
376
both EMU Cases, available concentration observations are representative of
377
average values over the duration of the experiment in steady-state flows that
378
means the pairing used for the validation is in space only for all the sensors.
379
5. Results and discussion
380
5.1. EMU project simulation, Case A, Phase I (Case A1)
M AN U
SC
RI PT
373
The simulations with both steady and unsteady RANS solutions were
382
performed on 20 processors (Intel(R) Xeon(R) CPU E52695
[email protected] GHz)
383
on a cluster CentOS release 6.5. The steady RANS solution took ≈ 5 hours
384
to complete the simulation for the Case A1, whereas, the computational time
385
in URANS simulation was ≈ 8 hours.
386
5.1.1. Overall performance
TE
D
381
The predicted concentrations at all the sensor positions are compared
388
with the experimental observations. Also, the approximate dimensions of
389
the predicted recirculation zones are compared with the measurements. The
390
predicted concentrations from both (i) steady and (ii) unsteady RANS simu-
392
393
AC C
391
EP
387
lations are analysed with the EMU observations. A comparative performance analysis with both types of the simulations is performed by computing and comparing the statistical evaluation measures.
394
Figure 3 shows isopleths of the ground level concentration from steady
395
RANS simulations in the computational domain. The contour plot shows
18
ACCEPTED MANUSCRIPT
influence of the elongated building on plume deflection. It shows that the
397
concentrations from the source exit are advected by the recirculation flow
398
behind the L-shaped building, and a low concentration region appears in
399
wake of the building in downwind direction. A high concentration region
400
appears close to the source, even in upwind direction of the source exit. The
401
high concentration in upwind of the source exit is due to the back reflection
402
of the plume concentration from the L-shaped building at close to the source.
403
The concentration contour in Figure 3 exhibits the ability of CFD model to
404
capture the effects of an building geometry on plume dispersion. Similar
405
characteristics of plume behavior are also observed from URANS simulation.
406
For steady RANS simulation, Figure 4 shows the zoom on velocity vectors
407
in x and z cross-section, behind the L-shaped building in Case A1. The wind
408
tunnel observations and empirical formulas in [40] suggested that the length
409
of the re-circulating wake or cavity in the edge of the L-shaped building is
410
about 1.0 to 1.5 times H. From the FLACS CFD model prediction, Hanna et
411
al. [1] predicted this length about 1.5-2.0 times H, whereas this length from
412
the fluidyn-PANACHE prediction in Figure 4 is observed about 1.0-2.0 times
413
H. The unsteady RANS simulations also observed the similar characteristics
414
of velocity vectors in this Case A1.
416
417
418
SC
M AN U
D
TE
EP
For comparison with the observations (volume fractions of the source
AC C
415
RI PT
396
fluid), the CFD model simulated concentrations are normalized by a source condition parameter, Cn = UH H 2 /Cs Qs , where Qs is the release rate. In this computation for Case A1: UH = 5 ms−1 , H = 10 m, Qs = 20 m3 s−1 and Cs =
419
0.0296. Normalized values of the concentrations are used for computing the
420
statistical performance measures and for further analysis of the results. Fig-
19
ACCEPTED MANUSCRIPT
ures 5(a1)&(b1) show the scatter diagrams between the normalized predicted
422
and observed concentrations for steady and unsteady RANS simulations, re-
423
spectively, for all 256 measurement points in Case A1. These figures show
424
a good agreement between the wind tunnel observations and the simulated
425
concentrations in both type of simulations by fluidyn-PANACHE. For un-
426
paired analysis, Q − Q plots in Figures 5(a2)&(b2)) show that the predicted
427
concentrations are close to one-to-one line in both steady and unsteady sim-
428
ulations. It was observed that the simulated concentrations in both steady
429
and unsteady RANS simulations are approximately similar in this Case A1.
430
The statistical performance measures between the predicted and observed
431
normalized concentrations are given in Table 2 for both steady and unsteady
432
RANS simulations. Table 2 also shows the statistical performance measures
433
computed from the normalized concentrations at each sampling line (x/H =
434
0.5, 1.0, 2.0, 5.0, 10.0) downwind from the source. As NMSE characterizes the
435
extent of scattering in a sample, NMSE value for the concentrations at all
436
receptor points is 0.95 for steady RANS simulation (Table 2), which is smaller
437
than its acceptable bound (NMSE < 4). A slightly smaller, but similar
438
value of NMSE (= 0.92) is observed for the concentrations from unsteady
439
RANS simulation (Table 2). Negative values of FB (= -0.13 (steady), -0.12
440
(unsteady)) show that the CFD model exhibits an overall overprediction
442
443
SC
M AN U
D
TE
EP
AC C
441
RI PT
421
with the observations in both type of simulations. Higher values of COR and IA in both the simulations show a good one-to-one correlation between the simulated and observed concentrations. The values of COR, MG, and
444
VG are 0.87, 1.31, and 1.93, respectively for steady simulation (Table 2)
445
and it predicts ≈ 68% of points within a factor of two of the observations.
20
ACCEPTED MANUSCRIPT
Approximately similar values of these statistical indices (COR = 0.87, FAC2
447
= 67.97, MG = 1.31, VG = 1.88) are observed for URANS in this Case A1.
448
5.1.2. Performance at each sampling line
RI PT
446
One can observe a slightly overprediction tendency of higher concentra-
450
tions at near-field of the release at some locations (Figures 5(a)&6, Table 2).
451
This trend of the overprediction can be exhibited by the negative values of
452
FB at close to the source at x/H = 0.5, 1.0, 2.0 (Table 2, Figure 6), for both
453
steady and unsteady RANS simulated concentrations. However, the extent
454
of overprediction decreases with increasing downwind distance from the re-
455
lease and the CFD model slightly underpredicts at far from the source at
456
x/H = 5.0, 10.0 (Table 2). At all sampling lines downwind from the release,
457
both CFD solutions predict more than ≈ 64% points within a factor of two of
458
the observations. A good one-to-one correlation between the predicted and
459
measured concentrations is also observed at all sampling lines (Figure 6).
460
NMSE values decrease with increasing downwind distances from the release
461
in both steady and unsteady RANS simulations (Figure 6, Table 2).
462
5.1.3. Performance validation of the crosswind concentrations
EP
TE
D
M AN U
SC
449
The prediction of the normalized concentrations in crosswind direction at
464
two different vertical levels can be assessed from Figures 7(a)&(b) at several
465
466
467
AC C
463
downwind distance from the release. These figures show the simulated crosswind results against the observed profiles at x/H = 1.0, 2.0, 5.0, 10.0 for (a) steady and (b) unsteady RANS simulations. It was observed that close to the
468
building (i.e. x/H ≤ 2, z/H ≤ 0.34), the crosswind CFD plumes have high
469
peaks in comparison to its experimental counterpart. In fact, the simulated 21
ACCEPTED MANUSCRIPT
crosswind plumes exhibit the normal distributions which is slightly different
471
from the observed long right tailed plume in the experimental concentrations
472
at x/H = 1.0 and z/H = 0.16 (Figure 7(a1)&(b1)). It is noticed that both
473
simulated and observed crosswind plumes have approximately similar char-
474
acteristics with increasing downwind distances from the release and also at
475
higher vertical levels. However, agreement between the simulated and exper-
476
imental data at sensor positions is good at a given distance from the source
477
in both steady and unsteady simulations (Figure 7(a)&(b)).
478
5.1.4. Performance validation of the mean arc-maximum concentrations
M AN U
SC
RI PT
470
The methodology has been traditionally used to study whether the air
480
quality models can correctly predict quantities such as the maximum con-
481
centration over the entire receptor network, and the maximum concentration
482
across a given sampling arc or along a sampling line [37]. Thus, the nor-
483
malized predicted and observed mean arc-maximum concentrations across
484
a given crosswind (y/H) sampling arc at a given downwind sampling line
485
(x/H) and vertical level (z/H) are compared and discussed. The analysis of
486
these mean arc-maximum concentration at several vertical levels at a given
487
downwind distance is advantageous to quantifying the maximum exposure at
488
the different vertical levels which can be equivalent to the different storeys
490
491
492
TE
EP
AC C
489
D
479
of the buildings in an urban environment. Figures 8(a)&(b) show the scatter plots between the normalized predicted
and observed mean arc-maximum concentrations at all vertical levels and at each downwind sampling lines in the EMU Case A1. These figures show
493
the normalized arc-maximum concentrations predicted from both (a) steady
494
and (b) unsteady RANS simulations. The ratio of normalized predicted and 22
ACCEPTED MANUSCRIPT
observed mean arc-maximum concentrations across a given crosswind (y/H)
496
sampling arc at all downwind distances and vertical levels are also presented
497
in Table 3. In general, both steady and unsteady RANS simulated results of
498
the mean arc-maximum predicted concentrations at each sampling line are
499
in good agreement with the wind-tunnel observations (Figure 8(a)&(b)). In
500
most of the cases, the concentrations are slightly overpredicted where the
501
heights are less than 10 m, but slightly underpredicted at the heights more
502
than 10 m (Table 3). In most of the cases (≈ 88%), the predicted mean
503
arc-maximum concentrations are within a factor of two of the observations
504
in both steady and unsteady RANS solutions (Figure 8, Table 3).
505
5.1.5. Comparison with a previous validation study
M AN U
SC
RI PT
495
In order to analyze the relative performance of the present CFD model–
507
fluidyn-PANACHE with other CFD models, a comparison of the present
508
simulation results is carried out with the previous simulation results from a
509
CFD model FLACS [1] at 36 locations for the EMU Case A1. Table 4 presents
510
a comparison of FLACS predicted concentrations (F Lro/p = Cobs /Cpred ) [1]
511
with present fluidyn-PANACHE model predicted concentrations (F P ro/p =
512
Cobs /Cpred ) from the unsteady RANS solution at x/H = 1.0 (i.e., one building
513
height downwind of the lee of the building) for six different heights (z/H =
515
516
517
TE
EP
AC C
514
D
506
0.16, 0.37, 0.67, 1.02, 1.47, 1.96), and for six different lateral positions (y/H = -2.0 -1.5, -1.0, -0.5, 0.0, 0.5) in this Case A1. It shows that the present model is simulating slightly better than the FLACS. At y/H = -1.5, -1.0, FLACS predicts 33% of points within a factor of two, whereas FAC2 points
518
predicted by the present model are 100% and 67%, respectively (Table 4).
519
Overall, 72% of the FLACS predictions are within a factor of two of the 23
ACCEPTED MANUSCRIPT
observations for cross-wind and vertical profiles [1]. Whereas, present model
521
predicts 83% points within a factor of two of all these observations, which
522
is more than the FAC2 value predicted by the FLACS. This comparison
523
shows that the crosswind and vertical profiles were well and slightly better
524
simulated by the fluidyn-PANACHE.
RI PT
520
In summary, based on the above statistical performance analysis, we can
526
conclude that the fluidyn-PANACHE exhibits good performance in both
527
steady and unsteady RANS simulations. It is also observed that in a simple
528
geometry with lesser complexity in the Case A1, steady and unsteady RANS
529
simulations give approximately similar performance against the observations.
530
5.2. EMU project simulation, Case C, Phase I (Case C1)
531
5.2.1. Overall performance
M AN U
SC
525
Both steady and unsteady RANS simulations were performed for this
533
Case C1 of “passive” plume dispersion in a real chemical site of the EMU
534
project. Figure 9 presents isopleths of the steady RANS simulated tracer
535
concentrations at ground level within the site. The plume centreline for the
536
“passive” release is displaced by more than 5H across-stream, even though its
537
nominal emission velocity ratio is Us /UH = 1.5. This is probably due to the
538
shielding effect of the tall rectangular building upstream of the source, which
540
541
542
TE
EP
AC C
539
D
532
reduces the background flow velocity around the source, thereby increasing the actual velocity ratio of the emission. The predicted and observed concentrations at all receptors are presented
in form of the scatter plots (Figures 10(a1)&(b1)) and the Q − Q plots (Fig-
543
ures 10(a2)&(b2)), for both steady and unsteady RANS simulations. In
544
scatter plots (Figures 10(a1)&(b1)), it is observed that the simulated con24
ACCEPTED MANUSCRIPT
centrations by the CFD model in both type of simulations have good agree-
546
ment with the observations. However, in comparison to the steady solution
547
(Figure 10(a1)), small scatter is observed in unsteady RANS simulated con-
548
centrations (Figure 10(b1)). The simulated higher concentrations in steady
549
solution at the receptors near to the source are close to one-to-one line, how-
550
ever, comparably more scatter is observed for lower concentrations at far
551
away from the source. However, in unsteady solution, even the lower con-
552
centrations at the receptors far away from the release are well predicted and
553
have smaller scatteredness. These trends of predicted concentrations in both
554
solutions are more visible in Q − Q plots in Figures 10(a2)&(b2), which show
555
a comparison of the concentration distributions.
M AN U
SC
RI PT
545
In order to quantitatively analyze the CFD model performance, the sta-
557
tistical performance measures are separately calculated for (i) all measure-
558
ments, (ii) at each downwind distance (x/H) from the release, (iii) at each
559
cross-stream profile (y/H), (iv) each vertical profile ((z − zg )/H), and (v)
560
ground-level maximum (GL − M AX) concentrations (Table 5). The com-
561
puted statistical indices for both steady and unsteady RANS simulations
562
show that the CFD model is preforming good with the observations. Over-
563
all, smaller value of NMSE (=0.34) in unsteady solution shows the lesser
564
scatter in comparison to the steady simulation (NMSE = 0.43). The un-
566
567
TE
EP
AC C
565
D
556
steady RANS solution predicts 79.5% of points within a factor of two of the observations, which is higher than the number of points (FAC2 = 67.0%) simulated with in a factor of two from the steady simulation (Table 5). The
568
value of FB shows the similar degree of slightly overprediction (FB = -0.10)
569
in both solutions. Good one-to-one agreement are observed by the higher
25
ACCEPTED MANUSCRIPT
values of COR and IA from both solutions (Table 5). The smaller values
571
of MG (=1.83) and VG(=29.7) in unsteady solutions show a smaller scatter
572
in comparison to the steady solution (MG = 2.0, VG = 43.8) and it is also
573
visible from the scatter and Q − Q plots in Figures 10(a1)&(b1).
574
5.2.2. Performance at each horizontal line
SC
RI PT
570
The variation of the statistical indices with downwind distance (x/H)
576
from the release is analysed to study the near- and far-field dispersion from
577
both steady and unsteady RANS solutions (Figure 11). Table 5 and Fig-
578
ure 11 show that both CFD solutions overpredict at the receptors close to
579
the source (at x/H = 11.0, steady: FB = -0.16, unsteady: FB = -0.04).
580
However, the overprediction is smaller in unsteady simulation. Positive and
581
close to the ideal values of FB (≤ 0.10) at x/H = 30.0, 45.0, & 67.5 show
582
a good, however, slightly underpredictions at far from the source in both
583
simulations. Close to the release at x/H = 11.0, steady RANS predicts only
584
29.4% of points within a factor of two of the observations, whereas, compara-
585
bly large no. of points (FAC2 = 58.8%) were simulated within a factor of two
586
with the unsteady simulation (Table 5). It is also observed that more num-
587
ber of points are predicted within a factor of two with increasing downwind
588
distances from the source in both types of the solutions (Figure 11, Table
590
591
592
D
TE
EP
AC C
589
M AN U
575
5). The degree of scatter between the predicted and observed concentrations decreases with increasing downwind distances from the source in both CFD simulations. This fact can be clearly seen with the decreasing value of NMSE with increasing downwind distances from the release (Figure 11, Table
593
5). However, NMSE values in unsteady simulations were observed smaller
594
than the steady solution at each horizontal line downwind from the source. 26
ACCEPTED MANUSCRIPT
The values of COR and IA are also slightly better in unsteady simulation,
596
especially at far from the source (Figure 11, Table 5).
597
5.2.3. Performance of crosswind and vertical concentrations
RI PT
595
Observed and predicted crosswind concentration profiles are analysed for
599
both steady and unsteady RANS simulations. Figures 12(a)&(b) show the
600
experimental and numerical results for cross-stream profiles at (i) ground
601
level, i.e. at (z − zg )/H = 0.0, and (ii) (z − zg )/H = 3.6, for all four
602
downwind distances from the release. The statistical indicies are also sepa-
603
rately computed at all horizontal and crosswind points at two vertical levels
604
(z − zg )/H = 0.0 & 3.6. For both solutions, the crosswind concentrations are
605
slightly overpredicted at ground-level, while an underprediction is observed
606
at (z − zg )/H = 3.6 (Table 5). For (z − zg )/H = 0.0 & 3.6, steady solution
607
predicts respectively 65.6% and 55.6% points within a factor of two of the
608
observations. Whereas, FAC2 values for the unsteady RANS solutions are
609
62.5% and 80.6% at (z − zg )/H = 0.0 & 3.6, respectively. NMSE values for
610
the unsteady RANS are also smaller than the steady RANS (Table 5).
TE
D
M AN U
SC
598
The statistical measures are also separately computed for the concen-
612
trations at all downwind distances and verticals receptors at two crosswind
613
points y/H = 5.0 & 6.3 (Table 5). It shows that URANS is predicting compa-
615
616
617
AC C
614
EP
611
rable better than the steady solution. The unsteady solution predicts 100.0% and 83.3% points within a factor of two at y/H = 5.0 & 6.3, respectively, which are better than FAC2 = 75.0% & 66.7% from the steady solution. The vertical concentration profiles were measured and predicted at y/H =
618
5.0 for x/H = 30.0, and at y/H = 6.3 for x/H = 11.0, 45.0, 67.5 (sensors
619
were deployed on 8 vertical levels at each of these x/H corresponds to these 27
ACCEPTED MANUSCRIPT
respective crosswind points (SI Table S2)). The vertical concentration pro-
621
files at near- and far-field are shown in Figures 13(a)&(b) for y/H = 5.0 and
622
y/H = 6.3. In general, the simulated concentrations have overprediction
623
tendency at receptors near to the ground-level, however, they are underpre-
624
dicted at the higher height above the ground surface. The values of these
625
performance measures in Table 5 exhibits that the unsteady RANS solution
626
is performing better than the steady RANS simulation in complex urban
627
geometry of the EMU Case C1.
628
5.2.4. Performance of the ground-level maximum concentrations
M AN U
SC
RI PT
620
The ratios of observed and modelled ground level maximum (GL − M ax)
630
concentrations at each sampling line are presented in Table 6 for both steady
631
and unsteady RANS solutions. It shows a comparison of the modelled and
632
experimental results of ground-level maximum concentration at each down-
633
wind distance from the source. It exhibits a good agreement of the simulated
634
concentrations with the experimental results. The model shows an overpre-
635
diction (steady: FB = -0.27, unsteady: FB = -0.28) of the ground level
636
maximum concentrations at (z − zg )/H = 0.0. Both solutions predict 100%
637
points within a FAC2 for the GL − M ax concentrations (Table 5). All these
638
simulated concentrations in both steady and unsteady RANS solutions have
640
641
642
TE
EP
AC C
639
D
629
good one-to-one correlation with the observed concentrations. In summary, predictions from both steady and unsteady RANS solutions
in both EMU cases are in good agreement with observations. Nevertheless, it is not surprising that the differences between steady and unsteady RANS are
643
relatively small as at heart the methods are very similar as they only describe
644
mean statistics in terms of (parameterized) Reynolds stresses. However, a 28
ACCEPTED MANUSCRIPT
validation of this theoretical remark with an experimental dataset is always
646
advantageous to avoid the uncertainty in selecting among these solutions,
647
especially in emergencies scenarios, and also for air quality and atmospheric
648
dispersion studies. In fact, a conclusion from the present study to utilize the
649
unsteady RANS instead of steady RANS in more complex urban regions can
650
be useful.
651
6. Conclusions
M AN U
SC
RI PT
645
This study presents the 3-D CFD simulations for near-field dispersion
653
of toxic and hazardous gases near buildings and in a geometrically complex
654
chemical site. A CFD model fluidyn-PANACHE is validated using two cases
655
of the EMU project: (i) Case A1 and (ii) Case C1. In both EMU cases,
656
the steady RANS solver is used first to get quick results and comparison
657
with the experimental data. This first comparison shows a good agreement
658
between the predicted and measured concentrations. Nevertheless, in order
659
to improve the performance of the solver, the unsteady mode has been tried.
660
A comprehensive statistical analysis of the results is performed to analyze
661
the performance of the CFD model in complex urban or industrial regions.
662
The performance in each EMU case is validated at (i) all receptors (overall)
663
(ii) each sampling line downwind from the release (iii) crosswind and vertical
665
666
TE
EP
AC C
664
D
652
concentrations, and (iv) maximum concentrations. In both cases of the EMU experiment, predicted concentrations from both
steady and unsteady RANS simulations by fluidyn-PANACHE are in good
667
agreement with observations. Based on the statistical analysis of results, it is
668
observed that in a simple geometry with lesser complexity in the EMU Case 29
ACCEPTED MANUSCRIPT
A1, both steady and unsteady RANS solutions give approximately similar
670
performance from the observations. In this Case A1, overall ≈ 68% of points
671
are predicted within a factor of two of observations from both solutions and
672
an overprediction tendency of the concentrations is observed at the heights
673
close to ground surface. The statistical measures are within the acceptable
674
range at each sampling line. In Case A1, at all sampling lines downwind from
675
the release, CFD model predicts more than ≈ 64% points within a factor of
676
two. Crosswind and mean arc-maximum concentrations from both solutions
677
are in good agreement with observations.
M AN U
SC
RI PT
669
In a comparably more complex and real urban geometry in the EMU
679
Case C1, CFD model with both steady and unsteady solvers shows a good
680
agreement between the simulated and measured concentrations. However, it
681
was observed that the unsteady RANS solution is performing slightly bet-
682
ter than the steady solution. In Case C1, unsteady RANS solution predicts
683
79.5% of points within a factor of two, which is higher than 67.0% points
684
simulated with in a factor of two from the steady simulation. All points
685
(100%) were predicted within a factor of two for the ground-level maximum
686
concentrations in both simulations. NMSE values in unsteady simulation
687
were observed smaller than the steady solution at each horizontal line down-
688
wind from the source. Close to the release at x/H = 11.0, steady solution
690
691
TE
EP
AC C
689
D
678
predicts only 29.4% of points within a factor of two, whereas, comparably large no. of points (FAC2 = 58.8%) were predicted with the unsteady simulation. For both solutions, crosswind concentrations are slightly overpredicted
692
at ground-level, while a slightly underprediction is observed at higher vertical
693
level above the ground surface. The crosswind and vertical profiles in the
30
ACCEPTED MANUSCRIPT
near- and far-field show the overprediction tendency at the receptors near to
695
the source but slightly underprediction at far away from the release.
RI PT
694
The statistical evaluation results with two cases of the EMU experiment
697
show an overall good performance of the CFD model fluidyn-PANACHE in
698
complex urban environments. It shows that the fluidyn-PANACHE is well
699
suited for the air pollution and emergency planning in industrial or urban
700
areas. This study critically examines the real predictive capability of fluidyn-
701
PANACHE in contexts of an accidental or deliberate airborne release in ur-
702
ban regions and strengthen the evidence that it is capable of dealing properly
703
with dispersion phenomena in complex urban or industrial environments.
704
Acknowledgment
M AN U
SC
696
A sincere thanks to Liying Chen, Dr. Malo Le Guellec and Dr. Claude
706
Souprayen from Fluidyn France, for discussions and helping in CFD simu-
707
lation. Authors gratefully acknowledge Fluidyn France for the use of CFD
708
model fluidyn-PANACHE. The EMU dataset was obtained from the CED-
709
VAL at Hamburg University, by the team of Prof. Michael Schatzmann.
710
References
712
713
714
TE
EP
AC C
711
D
705
[1] S. R. Hanna, O. R. Hansen, S. Dharmavaram, FLACS CFD air quality model performance evaluation with Kit Fox, MUST, Prairie Grass, and EMU observations, Atmospheric Environment 38 (28) (2004) 4675–4687. doi:http://dx.doi.org/10.1016/j.atmosenv.2004.05.041.
715
31
ACCEPTED MANUSCRIPT
[2] P. Kumar, A.-A. Feiz, P. Ngae, S. K. Singh, J.-P. Issartel, CFD sim-
717
ulation of short-range plume dispersion from a point release in an ur-
718
ban like environment, Atmospheric Environment 122 (2015) 645 – 656.
719
doi:http://dx.doi.org/10.1016/j.atmosenv.2015.10.027. [3] J.
Labovsky,
Jelemensky,
using
dynamic
CFD
simulations
boundary
of
conditions,
ammo-
722
nia
Process
723
Safety and Environmental Protection 88 (4) (2010) 243 – 252.
724
doi:http://dx.doi.org/10.1016/j.psep.2010.03.001.
M AN U
dispersion
L.
SC
720 721
RI PT
716
[4] P. Gousseau, B. Blocken, T. Stathopoulos, G. van Heijst, CFD sim-
727
ulation of near-field pollutant dispersion on a high-resolution grid:
728
A case study by LES and RANS for a building group in down-
729
town montreal, Atmospheric Environment 45 (2) (2011) 428 – 438.
730
doi:http://dx.doi.org/10.1016/j.atmosenv.2010.09.065.
TE
D
725 726
[5] C. Gromke, B. Blocken, Influence of avenue-trees on air qual-
733
ity at the urban neighborhood scale. Part I: Quality assur-
734
ance studies and turbulent schmidt number analysis for RANS
735
CFD simulations, Environmental Pollution 196 (2015) 214–223.
737 738
739
740
741
AC C
736
EP
731 732
doi:http://dx.doi.org/10.1016/j.envpol.2014.10.016.
[6] Prediction of high concentrations and concentration distribution of a continuous point source release in a semi-idealized urban canopy using CFD-RANS modeling, Atmospheric Environment 100 48–56, year =.
[7] J. Liu, J. Niu, CFD simulation of the wind environment around 32
ACCEPTED MANUSCRIPT
an isolated high-rise building:
An evaluation of SRANS, LES
743
and DES models, Building and Environment 96 (2016) 91 – 106.
744
doi:http://dx.doi.org/10.1016/j.buildenv.2015.11.007.
RI PT
742
[8] P.-Y. Cui, Z. Li, W.-Q. Tao, Buoyancy flows and pollutant dis-
747
persion through different scale urban areas: CFD simulations and
748
wind-tunnel measurements, Building and Environment (2016) –
749
doi:http://dx.doi.org/10.1016/j.buildenv.2016.04.028.
SC
745 746
[9] Y. Tominaga, T. Stathopoulos, CFD simulation of near-field pollu-
752
tant dispersion in the urban environment: A review of current mod-
753
eling techniques, Atmospheric Environment 79 (0) (2013) 716 – 730.
754
doi:http://dx.doi.org/10.1016/j.atmosenv.2013.07.028.
M AN U
750 751
[10] S. D. Sabatino, R. Buccolieri, P. Salizzoni, Recent advancements in nu-
757
merical modelling of flow and dispersion in urban areas: a short review,
758
International Journal of Environment and Pollution 52 (3-4) (2013) 172–
759
191. doi:10.1504/IJEP.2013.058454.
TE
D
755 756
[11] B. Blocken, 50 years of computational wind engineering: Past, present
762
and future, Journal of Wind Engineering and Industrial Aerodynamics
763
129 (2014) 69 – 102. doi:http://dx.doi.org/10.1016/j.jweia.2014.03.008.
766
767
AC C
764 765
EP
760 761
[12] B. Blocken, Computational fluid dynamics for urban physics: Importance, scales, possibilities, limitations and ten tips and tricks towards accurate and reliable simulations, Building and Environment 91
768
(2015) 219–245, fifty Year Anniversary for Building and Environment.
769
doi:http://dx.doi.org/10.1016/j.buildenv.2015.02.015.
770
33
ACCEPTED MANUSCRIPT
[13] R. Meroney, R. Ohba, B. Leitl, H. Kondo, D. Grawe, Y. Tominaga,
772
Review of CFD guidelines for dispersion modeling, Fluids 1 (2) (2016)
773
14. doi:10.3390/fluids1020014.
RI PT
771
[14] M. Lateb, R. Meroney, M. Yataghene, H. Fellouah, F. Saleh, M. Bo-
776
ufadel, On the use of numerical modelling for near-field pollutant dis-
777
persion in urban environments – a review, Environmental Pollution 208,
778
Part A (2016) 271 – 283, special Issue: Urban Health and Wellbeing.
779
doi:http://dx.doi.org/10.1016/j.envpol.2015.07.039. [15] Y.
Tominaga, of
T.
M AN U
780 781
SC
774 775
Stathopoulos,
near-field
questions
782
modeling
783
vironment,
784
doi:http://dx.doi.org/10.1016/j.buildenv.2016.06.027.
Building
pollutant
Ten
and
dispersion
Environment
in
concerning
the
(2016)
built –In
en-
press.
[16] G. Antonioni, S. Burkhart, J. Burman, A. Dejoan, A. Fusco, R. Gaas-
787
beek, T. Gjesdal, A. Jppinen, K. Riikonen, P. Morra, O. Parmhed,
788
J. Santiago, Comparison of CFD and operational dispersion models in
789
an urban-like environment, Atmospheric Environment 47 (2012) 365 –
790
372. doi:http://dx.doi.org/10.1016/j.atmosenv.2011.10.053.
793
794
795
796 797
798
TE
EP
[17] C. Yuan, E. Ng, L. K. Norford, Improving air quality in high-density
AC C
791 792
D
785 786
cities by understanding the relationship between air pollutant dispersion and urban morphologies, Building and Environment 71 (2014) 245 – 258. doi:http://dx.doi.org/10.1016/j.buildenv.2013.10.008.
[18] COST ES1006, Best Practice Guidelines, COST Action ES1006, no. ISBN: 987-3-9817334-0-2, 2015.
34
ACCEPTED MANUSCRIPT
[19] I. V. Kovalets, S. Andronopoulos, A. G. Venetsanos, J. G. Bartzis, Iden-
800
tification of strength and location of stationary point source of atmo-
801
spheric pollutant in urban conditions using computational fluid dynam-
802
ics model, Mathematics and Computers in Simulation 82 (2) (2011) 244
803
– 257. doi:http://dx.doi.org/10.1016/j.matcom.2011.07.002.
RI PT
799
[20] P. Kumar, A.-A. Feiz, S. K. Singh, P. Ngae, G. Turbelin, Reconstruction
806
of an atmospheric tracer source in an urban-like environment, Journal
807
of Geophysical Research: Atmospheres 120 (24) (2015) 12589–12604.
808
doi:10.1002/2015JD024110.
M AN U
SC
804 805
[21] P. Kumar, S. K. Singh, A.-A. Feiz, P. Ngae, An urban scale inverse
811
modelling for retrieving unknown elevated emissions with building-
812
resolving simulations, Atmospheric Environment 140 (2016) 135 – 146.
813
doi:http://dx.doi.org/10.1016/j.atmosenv.2016.05.050.
816
[22] M. J. Brown, Urban dispersionchallenges for fast response modeling, in:
TE
814 815
D
809 810
Fifth AMS Symposium on the Urban Environment, 2004. [23] L. Margheri, P. Sagaut, An uncertainty quantification analysis in
818
a simplified problem of urban pollutant dispersion by means of
819
anova-pod/kriging-based response surfaces, in:
820
Joint US-European Fluids Engineering Division Summer Meeting and
822
823
ASME 2014 4th
AC C
821
EP
817
12th International Conference on Nanochannels, Microchannels, and Minichannels, American Society of Mechanical Engineers, 2014, pp. V01DT28A004–V01DT28A004.
824
[24] J. Franke, A. Hellsten, H. Schl¨ unzen, B. Carissimo, Best practice guide-
825
line for the CFD simulation of flows in the urban environment. COST 35
ACCEPTED MANUSCRIPT
Action 732: Quality Assurance and Improvement of Microscale Meteo-
827
rological Models (2007) , Hamburg, Germany.
RI PT
826
[25] R. C. Hall, Evaluation of Model Uncertainty (EMU)-CFD modelling
829
of near-field atmospheric dispersion. Project EMU Final Report, Tech.
830
Rep. WS Atkins Doc No. WSA/AM5017/R7, European Commission,
831
WS Atkins, Woodcote Grove, Ashley Road, Epsom, Surrey KT18 5BW,
832
UK (1997).
M AN U
SC
828
833
[26] A. Mazzoldi, T. Hill, J. J. Colls, CFD and gaussian atmospheric disper-
834
sion models: A comparison for leak from carbon dioxide transportation
835
and storage facilities, Atmospheric Environment 42 (34) (2008) 8046 –
836
8054. doi:http://dx.doi.org/10.1016/j.atmosenv.2008.06.038.
839
[27] Fluidyn-PANACHE, User Manual, FLUIDYN France / TRANSOFT International, version 4.0.7 Edition (2010).
D
837 838
[28] R. Hill, A. Arnott, P. Hayden, T. Lawton, A. Robins, T. Parker, Eval-
841
uation of cfd model predictions of local dispersion from an area source
842
on a complex industrial site, International Journal of Environment and
843
Pollution 44 (1-4) (2011) 173–181.
845
846
847
EP
[29] J. H. Ferziger, M. Peric, Computational Methods for Fluid Dynamics,
AC C
844
TE
840
Springer Berlin Heidelberg, 2002. doi:10.1007/978-3-642-56026-2.
[30] B. Launder, Turbulence modelling of buoyancy-affected flows, in: Singapore Turbulence Colloquium, 2004.
36
ACCEPTED MANUSCRIPT
[31] K. Hanjalic, Turbulence and transport phenomena: Modelling and simu-
849
lation, in: Turbulence Modeling and Simulation (TMS) Workshop, Tech-
850
nische Universitt Darmstadt, 2005.
RI PT
848
[32] J. Han, S. P. Arya, S. Shen, Y.-L. Lin, An estimation of turbulent
852
kinetic energy and energy dissipation rate based on atmospheric bound-
853
ary layer similarity theory, no. NASA/CR-2000-210298, National Aero-
854
nautics and Space Administration, Langley Research Center, Hampton,
855
Virginia 23681-2199, 2000.
M AN U
SC
851
856
[33] C. V. Jayatilleke, The influence of prandtl number and roughness on
857
the resistance of the laminar sublayer to momentum and heat transfer,
858
Progress in Heat and Mass Transfer 1 (1969) 193–329. [34] J. P. Van Doormaal, G. D. Raithby, Enhancements of the simple method
860
for predicting incompressible fluid flows, Numerical Heat Transfer 7 (2)
861
(1984) 147–163. doi:10.1080/01495728408961817.
TE
D
859
[35] I. R. Cowan, A. G. Robins, I. P. Castro, Project EMU Experimental
864
data, Case A, Release 1; Case B, Release 3 and Case C, Release 1, Tech.
865
rep., EnFlo Research Centre, Department of Mechanical Engineering,
866
University of Surrey, Guildford, Surrey GU2 5XH, UK (1995).
868
869
AC C
867
EP
862 863
[36] I. R. Cowan, Project EMU Simulations, Stage C, Tech. rep., EnFlo Research Centre, Department of Mechanical Engineering, University of Surrey, Guildford, Surrey GU2 5XH, UK (1996).
870
[37] J. C. Chang, S. R. Hanna, Air quality model performance evalua-
871
tion, Meteorology and Atmospheric Physics 87 (1-3) (2004) 167–196. 37
ACCEPTED MANUSCRIPT
doi:10.1007/s00703-003-0070-7.
872
[38] E. Yee, B.-C. Wang, F.-S. Lien, Probabilistic model for concentra-
875
tion fluctuations in compact-source plumes in an urban environment,
876
Boundary-layer meteorology 130 (2) (2009) 169–208.
RI PT
873 874
[39] G. C. Efthimiou, S. Andronopoulos, I. Tolias, A. Venetsanos, Prediction
878
of the upper tail of concentration distributions of a continuous point
879
source release in urban environments, Environmental Fluid Mechanics
880
(2016) 1–23doi:10.1007/s10652-016-9455-2.
M AN U
SC
877
[40] S. R. Hanna, G. A. Briggs, R. P. Hosker Jr, Handbook on atmospheric
882
diffusion, Tech. Rep. DOE/TIC-11223, National Oceanic and Atmo-
883
spheric Administration, Oak Ridge, TN (USA). Atmospheric Turbulence
884
and Diffusion Lab. (1982).
887
1 2
888 889 890
892 893 894
3
Source data for the EMU project: Case A1 and Case C1. . . . 41 Statistical performance measures of the concentrations from both steady and unsteady RANS simulations for all and at each downwind distance of the EMU project Case A1. FAC2 is given in percentage (%). . . . . . . . . . . . . . . . . . . . . 41 Comparison of the normalized observed and predicted mean arc-maximum concentrations (ro/p = Cobs /Cpred ) across a given crosswind (y/H) sampling arc at a given downwind sampling line (x/H) and vertical level (z/H). . . . . . . . . . . . . . . . 42
AC C
891
TE
886
List of Tables
EP
885
D
881
38
ACCEPTED MANUSCRIPT
899 900 901
5
903 904 905
6
907
908
909
List of Figures 1
910 911
2
912 913 914
3
915 916
4
917 918 919
5
921 922 923 924 925 926 927 928
AC C
920
Layout and the dimension of the computational domain considered for the EMU Case A1. . . . . . . . . . . . . . . . . The site features and dimension of the computational domain considered for the EMU project Case C1. The black lines are terrain height contours close to the site. . . . . . . . . . . . Ground level concentration contours from steady RANS solution for the EMU Case A1. . . . . . . . . . . . . . . . . . . Zoom on velocity vectors computed from steady RANS solution in vertical plane (x−z) behind the building in EMU Case A1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Scatter (first row) and Q-Q (second row) plots between the normalized predicted and observed concentrations at all vertical levels of each sampling line for the EMU Case A1. First and second columns are respectively for steady and unsteady RANS simulated concentrations. The middle solid line is oneto-one line between observed and simulated concentrations whereas the dotted lines correspond to factor of two. . . . . The statistical performance measures (a) NMSE, (b) FB, (c) COR, and (d) FAC2 with increasing downwind distances from the source (at each horizontal lines) in EMU Case A1 . . .
D
906
TE
902
RI PT
898
SC
896 897
A Comparison of FLACS predicted concentrations (F Lro/p = Cobs /Cpred ) [1] with present fluidyn-PANACHE predicted concentrations (F P ro/p = Cobs /Cpred ) from the unsteady RANS solution at x/H = 1.0 for six different heights (z/H), and for six different lateral positions (y/H) in the EMU Case A1. y/H = 0.0 is along the center of the lee building wall and y/H is positive towards the longer side of the building. . . . . . . 43 Statistical performance measures of the concentrations from both steady and unsteady RANS simulations for all and at each downwind distance, and also for other profiles of the EMU project Case C1. FAC2 is given in percentage (%). . . . . . . 44 Comparison of modeled and experimental results (ro/p = Cobs /Cpred ) of ground-level maximum concentrations (Case C1). . . . . . 44
M AN U
4
EP
895
6
39
. 45
. 46 . 47
. 47
. 48
. 49
ACCEPTED MANUSCRIPT
933 934
8
935 936 937
9
938 939
10
940 941 942 943 944
11
945 946 947
12
D
948 949 950 951 952
955
13
EP
954
AC C
953
RI PT
932
SC
930 931
Comparison of the measured and simulated normalized concentrations for cross-stream profiles at two vertical levels for x/H = 1.0, 2.0, 5.0, 10.0 in Case A1. First and second columns are respectively for steady and unsteady RANS simulated concentrations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 Scatter plot between the normalized predicted and observed mean arc-maximum concentrations at all vertical levels of each sampling line for the EMU Case A1. . . . . . . . . . . . . . . 51 Ground level concentration contours computed from steady RANS solution for the EMU Case C1. . . . . . . . . . . . . . 52 Scatter (first row) and Q-Q (second row) plots between the normalized predicted and observed concentrations at all vertical levels of each sampling line for the EMU Case C1. First and second columns are respectively for steady and unsteady CFD simulated concentrations. . . . . . . . . . . . . . . . . . 53 The statistical performance measures (a) NMSE, (b) FB, (c) COR, and (d) FAC2 with increasing downwind distances from the source (at each horizontal lines) in EMU Case C1 . . . . 54 Comparison of measured and simulated normalized concentrations for cross-stream profiles at two vertical levels ((z − zg )/H = 0.0, 3.6) for four downwind distances (x/H = 11.0, 30.0, 45.0, 67.5) from the releases in Case C1. First and second columns are respectively for steady and unsteady RANS simulated concentrations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 Comparison of measured and simulated concentrations for vertical profiles in the near-field (left) and in the far-field (right) (Case C1). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
M AN U
7
TE
929
40
ACCEPTED MANUSCRIPT
C2 H4 2.96%
N2 97.04% 20 2.5 23.68 25 Continuous
Type of source Exit velocity (m/s) Chemical species Composition
Height of source (m) Mass flux (Kg/s) Temperature (o C) Release direction
Point 7.5
C2 H4 77.3%
N2 22.7%
2.0 5.46 15 Horizontal 327.5o
SC
Area 1.0
M AN U
Type of source Exit velocity (m/s) Chemical species Fraction volume Source surface (As ) (m2 ) Height of source’s center (m) Mass flux (Kg/s) Temperature (o C) Release duration (s)
RI PT
Table 1: Source data for the EMU project: Case A1 and Case C1. Case A1 Case C1
Table 2: Statistical performance measures of the concentrations from both steady and unsteady RANS simulations for all and at each downwind distance of the EMU project Case A1. FAC2 is given in percentage (%).
All x/H = 0.5 x/H = 1.0 x/H = 2.0 x/H = 5.0 x/H = 10.0 All x/H = 0.5 x/H = 1.0 x/H = 2.0 x/H = 5.0 x/H = 10.0
NMSE
FB
COR
FAC2
MG
VG
IA
0.95 0.95 0.83 0.52 0.17 0.16 0.92 0.91 0.81 0.51 0.17 0.16
-0.13 -0.21 -0.18 -0.09 0.09 0.17 -0.12 -0.20 -0.17 -0.08 0.10 0.18
0.87 0.85 0.84 0.87 0.93 0.95 0.87 0.86 0.84 0.87 0.93 0.95
67.58 63.89 72.92 70.83 66.67 65.00 67.97 65.28 70.83 70.83 68.75 65.00
1.31 1.17 1.22 1.29 1.39 1.64 1.31 1.17 1.22 1.28 1.39 1.62
1.93 2.54 1.68 1.59 1.74 1.96 1.88 2.45 1.65 1.57 1.71 1.89
0.85 0.83 0.81 0.85 0.95 0.97 0.86 0.84 0.81 0.85 0.95 0.97
EP
TE
steady
D
Sim. type
AC C
unsteady
41
RI PT
ACCEPTED MANUSCRIPT
SC
Table 3: Comparison of the normalized observed and predicted mean arc-maximum concentrations (ro/p = Cobs /Cpred ) across a given crosswind (y/H) sampling arc at a given
z/H steady (ro/p ) unsteady (ro/p )
0.13 0.45 0.46
0.33 0.48 0.48
0.50 0.54 0.55
0.66 0.50 0.51
1.03 0.69 0.68
1.37 1.23 1.19
x/H = 1.0
z/H steady (ro/p ) unsteady (ro/p )
0.16 0.44 0.45
0.37 0.48 0.48
0.67 0.68 0.68
1.02 0.84 0.84
1.47 1.62 1.58
1.99 1.78 1.72
x/H = 2.0
z/H steady (ro/p ) unsteady (ro/p )
0.11 0.50 0.50
0.34 0.57 0.58
0.66 0.67 0.68
1.03 1.03 1.02
1.51 2.55 2.50
1.99 1.84 1.79
x/H = 5.0
z/H steady (ro/p ) unsteady (ro/p )
0.11 0.69 0.70
0.34 0.74 0.75
0.67 0.88 0.88
0.98 1.02 1.03
1.49 1.48 1.47
1.97 1.71 1.69
0.16 0.74 0.76
0.66 0.87 0.88
0.98 1.03 1.05
1.83 1.30 1.30
3.00 1.30 1.26
TE
D
x/H = 0.5
EP
M AN U
downwind sampling line (x/H) and vertical level (z/H).
z/H steady (ro/p ) unsteady (ro/p )
AC C
x/H = 10.0
42
1.45 1.62 1.56
1.69 1.56 1.50
1.93 1.64 1.58
SC
RI PT
ACCEPTED MANUSCRIPT
Table 4: A Comparison of FLACS predicted concentrations (F Lro/p = Cobs /Cpred ) [1] with present fluidyn-PANACHE predicted concentrations (F P ro/p = Cobs /Cpred ) from
M AN U
the unsteady RANS solution at x/H = 1.0 for six different heights (z/H), and for six different lateral positions (y/H) in the EMU Case A1. y/H = 0.0 is along the center of
0.16
0.37
0.67
1.02
1.47
1.99
FAC2
y/H = −2.0
F Lro/p F P ro/p
1.40 1.25
1.82 1.48
2.08 1.31
1.47 0.83
1.00 0.49
1.25 0.69
0.83 1.00
y/H = −1.5
F Lro/p F P ro/p
3.06 0.51
3.17 0.82
2.27 1.10
3.16 1.39
1.43 1.12
1.11 0.51
0.33 1.00
y/H = −1.0
F Lro/p F P ro/p
2.44 0.34
2.59 0.40
2.17 0.77
1.91 1.06
2.17 1.40
0.65 0.70
0.33 0.67
F Lro/p F P ro/p
1.66 0.55
1.66 0.61
1.75 0.64
1.71 0.78
1.88 1.37
0.63 1.24
1.00 1.00
F Lro/p F P ro/p
0.67 1.22
0.94 1.13
1.00 0.77
0.93 0.85
0.85 1.60
0.83 1.88
1.00 1.00
F Lro/p F P ro/p
0.50 2.74
0.60 2.24
0.61 1.38
0.59 1.74
0.57 2.67
0.61 2.78
1.00 0.33
y/H = −0.5
EP
y/H = 0.0
TE
z/H
D
the lee building wall and y/H is positive towards the longer side of the building.
AC C
y/H = 0.5
43
ACCEPTED MANUSCRIPT
Table 5: Statistical performance measures of the concentrations from both steady and
RI PT
unsteady RANS simulations for all and at each downwind distance, and also for other profiles of the EMU project Case C1. FAC2 is given in percentage (%). COR
FAC2
0.43 0.77 0.42 0.36 0.16 0.59 0.62 0.21 0.10 0.21 0.34 0.34 0.22 0.20 0.07 0.29 0.22 0.14 0.08 0.21
-0.10 -0.16 0.06 0.04 0.04 -0.24 0.46 -0.14 0.12 -0.27 -0.10 -0.04 0.06 0.10 0.06 -0.08 0.25 -0.15 -0.04 -0.28
0.97 0.97 0.88 0.82 0.84 0.97 0.77 0.95 0.97 0.98 0.99 0.99 0.93 0.90 0.93 0.99 0.91 0.96 0.99 0.99
67.0 29.4 64.7 64.7 82.4 65.6 55.6 75.0 66.7 100.0 79.5 58.8 70.6 64.7 94.1 62.5 80.6 100.0 83.3 100.0
EP
TE
D
unsteady
All x/H = 11.0 x/H = 30.0 x/H = 45.0 x/H = 67.5 (z − zg )/H = 0.0 (z − zg )/H = 3.6 y/H = 5.0 y/H = 6.3 GL − M ax All x/H = 11.0 x/H = 30.0 x/H = 45.0 x/H = 67.5 (z − zg )/H = 0.0 (z − zg )/H = 3.6 y/H = 5.0 y/H = 6.3 GL − M ax
FB
MG
VG
IA
2.00 7.15 4.00 1.25 1.17 3.11 2.12 1.32 1.81 0.85 1.83 6.18 3.45 1.36 1.18 1.53 1.52 1.04 1.48 0.84
43.8 62781 69340 1.74 1.25 15411 9.75 1.48 3.34 1.07 29.7 24442 24566 2.34 1.22 3.75 3.75 1.09 2.01 1.06
0.96 0.94 0.91 0.88 0.91 0.94 0.75 0.91 0.98 0.94 0.97 0.97 0.95 0.93 0.96 0.90 0.90 0.94 0.99 0.94
SC
steady
NMSE
M AN U
Sim. type
AC C
Table 6: Comparison of modeled and experimental results (ro/p = Cobs /Cpred ) of ground-
level maximum concentrations (Case C1).
x/H y/H steady (ro/p ) unsteady (ro/p )
5.5 5.0 0.77 0.73
11.0 5.6 0.76 0.73
11.0 5.1 0.58 0.63
19.5 6.0 0.65 0.67
44
30.0 6.3 0.73 0.76
30.0 7.0 0.91 0.91
37.5 7.0 1.01 1.01
45.0 6.8 0.88 0.84
45.0 7.0 1.02 0.97
56.3 7.0 1.02 0.98
67.5 6.8 0.93 0.91
67.5 7.0 1.08 1.04
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
Figure 1: Layout and the dimension of the computational domain considered for the EMU
AC C
EP
TE
D
Case A1.
45
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
Figure 2: The site features and dimension of the computational domain considered for the EMU project Case C1. The black lines are terrain height contours close to the site.
46
0
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
C2H4 (kg/m3) 0.002
0.001
0.003
0.004
D
Figure 3: Ground level concentration contours from steady RANS solution for the EMU
AC C
EP
TE
Case A1.
Figure 4: Zoom on velocity vectors computed from steady RANS solution in vertical plane (x − z) behind the building in EMU Case A1.
47
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
Figure 5: Scatter (first row) and Q-Q (second row) plots between the normalized predicted and observed concentrations at all vertical levels of each sampling line for the EMU Case A1. First and second columns are respectively for steady and unsteady RANS simulated concentrations. The middle solid line is one-to-one line between observed and simulated concentrations whereas the dotted lines correspond to factor of two.
48
1.2
0.2
Steady Unsteady
(a)
SC
FB
0.0 0.6
M AN U
-0.1 0.3
Steady Unsteady
(b)
0.1
0.9
-0.2 -0.3
0.0 0.5
1.0
1.00
2.0 x/H
5.0
TE
COR
0.92
0.80
1.0
2.0 x/H
5.0
5.0
10.0 Steady Unsteady
(d)
70 68 66 64 62
10.0
AC C
0.5
EP
0.84
2.0 x/H
72
D
0.96
1.0
74
Steady Unsteady
(c)
0.88
0.5
10.0
FAC2
NMSE
RI PT
ACCEPTED MANUSCRIPT
0.5
1.0
2.0 x/H
5.0
10.0
Figure 6: The statistical performance measures (a) NMSE, (b) FB, (c) COR, and (d) FAC2 with increasing downwind distances from the source (at each horizontal lines) in EMU Case A1
49
ACCEPTED MANUSCRIPT
Unsteady
Steady z/H = 0.16 z/H = 0.16 z/H = 1.02 z/H = 1.02
obs pred obs pred
2.0
1.5
1.0
1.5
1.0
SC
-3.0
-2.5
-2.0
-1.5
-1.0
-0.5
0.0
0.5
-3.0
1.0
(a2) EMU case A1: x/H = 2.0
z/H = 0.34 z/H = 0.34 z/H = 1.51 z/H = 1.51
1.4
obs pred obs pred
-2.0
-1.5
1.2
1.0 0.8 0.6
-1.0
-0.5
0.0
0.5
1.0
0.0
0.5
1.0
(b2) EMU case A1: x/H = 2.0
1.4
Normalized concentration
1.2
-2.5
1.6
z/H = 0.34 z/H = 0.34 z/H = 1.51 z/H = 1.51
obs pred obs pred
-2.0
-1.5
M AN U
1.6
1.0 0.8 0.6 0.4
0.4
0.2
0.2
0.0
0.0 -3.0
-2.5
-2.0
-1.5
-1.0
-0.5
0.0
0.5
-3.0
1.0
0.7
z/H = 0.34 z/H = 0.34 z/H = 1.49 z/H = 1.49
obs pred obs pred
TE
0.6
D
(a3) EMU case A1: x/H = 5.0
0.5 0.4
EP
0.3 0.2 0.1 0.0 -3
-2
-1
0.3
0.7
z/H = 0.34 z/H = 0.34 z/H = 1.49 z/H = 1.49
0.6
-0.5
obs pred obs pred
0.5 0.4 0.3 0.2
-4
2
obs pred obs pred
-2
z/H = 0.16 z/H = 0.16 z/H = 1.83 z/H = 1.83
0.3
0.2
-3
-1
0
1
2
(b4) EMU case A1: x/H = 10.0
0.4
Normalized concentration
z/H = 0.16 z/H = 0.16 z/H = 1.83 z/H = 1.83
-1.0
(b3) EMU case A1: x/H = 5.0
0.0 1
(a4) EMU case A1: x/H = 10.0
0.4
-2.5
0.1
0
AC C
-4
Normalized concentration
Normalized concentration
2.0
0.0
0.0
Normalized concentration
obs pred obs pred
0.5
0.5
Normalized concentration
z/H = 0.16 z/H = 0.16 z/H = 1.02 z/H = 1.02
2.5
Normalized concentration
Normalized concentration
2.5
(b1) EMU case A1: x/H = 1.0
3.0
RI PT
(a1) EMU case A1: x/H = 1.0
3.0
obs pred obs pred
0.2
0.1
0.1
0.0
0.0 -6
-4
-2
0 y/H
2
-6
4
50
-4
-2
0
2
y/H
Figure 7: Comparison of the measured and simulated normalized concentrations for crossstream profiles at two vertical levels for x/H = 1.0, 2.0, 5.0, 10.0 in Case A1. First and second columns are respectively for steady and unsteady RANS simulated concentrations.
4
Steady x/H = 0.50 x/H = 1.00 x/H = 2.00 x/H = 5.00 x/H = 10.0
10
(a)
Unsteady
x/H = 0.50 x/H = 1.00 x/H = 2.00 x/H = 5.00 x/H = 10.0
(b)
M AN U
1
1
0.1
0.1
TE
D
0.1
1
0.01 0.01
EP
Normalized observed concentration (Cobs/Cn)
10
0.1
1
Normalized observed concentration (Cobs/Cn)
Figure 8: Scatter plot between the normalized predicted and observed mean arc-maximum concentrations at all vertical levels of each sampling line for the EMU Case A1.
AC C
0.01 0.01
Normalized predicted concentration (Cpred/Cn)
Normalized predicted concentration (Cpred/Cn)
10
SC
RI PT
ACCEPTED MANUSCRIPT
51
10
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
0
0.005
C2H4 (kg/m3) 0.01
0.015
0.02
Figure 9: Ground level concentration contours computed from steady RANS solution for the EMU Case C1.
52
ACCEPTED MANUSCRIPT
Steady
Unsteady
0.1
Normalized observed concentration (Cobs/Cn)
0.001 0.001
Normalized predicted concentration (Cpred/Cn)
0.1
AC C 0.01
0.1
Normalized observed concentration (Cobs/Cn)
0.01
0.1
1
0.1
1
Normalized observed concentration (Cobs/Cn)
x/H = 11.0 x/H = 30.0 x/H = 45.0 x/H = 67.5
(b2)
0.1
0.01
EP
0.01
0.001 0.001
1
1
(a2)
x/H = 11.0 x/H = 30.0 x/H = 45.0 x/H = 67.5
(b1)
M AN U
0.1
TE
Normalized predicted concentration (Cpred/Cn)
1
0.01
D
0.001 0.001
0.1
0.01
0.01
x/H = 11.0 x/H = 30.0 x/H = 45.0 x/H = 67.5
SC
(a1)
x/H = 11.0 x/H = 30.0 x/H = 45.0 x/H = 67.5
RI PT
1
Normalized predicted concentration (Cpred/Cn)
Normalized predicted concentration (Cpred/Cn)
1
1
0.001 0.001
0.01
Normalized observed concentration (Cobs/Cn)
Figure 10: Scatter (first row) and Q-Q (second row) plots between the normalized predicted and observed concentrations at all vertical levels of each sampling line for the EMU Case C1. First and second columns are respectively for steady and unsteady CFD simulated concentrations.
53
0.9
0.2
Steady Unsteady
(a)
RI PT
ACCEPTED MANUSCRIPT
M AN U
FB
NMSE
0.3
0.0
SC
0.1 0.6
Steady Unsteady
(b)
-0.1
0.0 11.0
30.0
45.0 x/H
1.00
30.0
100
Steady Unsteady
(c)
0.96
45.0
67.5
x/H Steady Unsteady
(d)
TE
COR
0.88
EP
0.84
30.0
45.0
FAC2
D
80
0.92
0.80 11.0
-0.2 11.0
67.5
60
40
67.5
20 11.0
30.0
45.0
67.5
x/H
AC C
x/H
Figure 11: The statistical performance measures (a) NMSE, (b) FB, (c) COR, and (d) FAC2 with increasing downwind distances from the source (at each horizontal lines) in EMU Case C1
54
ACCEPTED MANUSCRIPT
Steady 0.16
obs pred obs pred
0.08
0.04
-5
0 5 10 15 20 (a2) EMU case C1: (z-zg)/H = 0.0
0.040
0.04
0.032
0.024
0.016
0.008
-10
obs pred obs pred
-5
obs pred obs pred
0.032
0 5 10 15 20 (b2) EMU case C1: (z-zg)/H = 0.0
M AN U
x/H = 45.0 x/H = 45.0 x/H = 67.5 x/H = 67.5
25
x/H = 45.0 x/H = 45.0 x/H = 67.5 x/H = 67.5
0.024
obs pred obs pred
0.016
0.008
0.000
0.000 -10
-5
0
5
10
15
(a3) EMU case C1: (z-zg)/H = 3.6
0.025
obs pred obs pred
EP
0.010
0.005
0.000 -4
-2
0
2
4
6
8
10
5
10
15
20
x/H = 11.0 x/H = 11.0 x/H = 30.0 x/H = 30.0
0.020
obs pred obs pred
0.015
0.010
0.000 12
14
AC C
0.012
0
(b3) EMU case C1: (z-zg)/H = 3.6
0.005
(a4) EMU case C1: (z-zg)/H = 3.6 x/H = 45.0 x/H = 45.0 x/H = 67.5 x/H = 67.5
0.016
Normalized concentration
TE
0.015
-5
0.025
D
x/H = 11.0 x/H = 11.0 x/H = 30.0 x/H = 30.0
0.020
-10
20
0.008
0.004
-4
16 obs pred obs pred
-2
0
2
4
6
8
10
12
14
(b4) EMU case C1: (z-zg)/H = 3.6 x/H = 45.0 x/H = 45.0 x/H = 67.5 x/H = 67.5
0.016
Normalized concentration
Normalized concentration
0.08
25
Normalized concentration
-10
Normalized concentration
0.12
0.00
0.00
Normalized concentration
x/H = 11.0 x/H = 11.0 x/H = 30.0 x/H = 30.0
RI PT
0.12
(b1) EMU case C1: (z-zg)/H = 0.0
SC
x/H = 11.0 x/H = 11.0 x/H = 30.0 x/H = 30.0
Normalized concentration
0.16
Normalized concentration
Unsteady
(a1) EMU case C1: (z-zg)/H = 0.0
0.012
16 obs pred obs pred
0.008
0.004
0.000
0.000 -2
0
2
4
6
y/H
8
10
12
14
16
-2
0
2
4
6
y/H
8
10
12
14
55
Figure 12: Comparison of measured and simulated normalized concentrations for crossstream profiles at two vertical levels ((z − zg )/H = 0.0, 3.6) for four downwind distances (x/H = 11.0, 30.0, 45.0, 67.5) from the releases in Case C1. First and second columns are respectively for steady and unsteady RANS simulated concentrations.
16
RI PT
ACCEPTED MANUSCRIPT
Steady
0.02
0.04
0.06
0.08
normalized concentration
10
EP
4
2
0.005
0.010
0.015
0.020
0.02
0.04
0.06
0.08
0.10
0.12
normalized concentration
D
TE
6
0 0.000
0 0.00
0.10
(a2) EMU case C1: vertical concentration (far-field) x/H = 45.0 obs x/H = 45.0 pred x/H = 67.5 obs x/H = 67.5 pred
AC C
(z-zg)/H
8
4
2
2
10
6
(a1) EMU case C1: vertical concentration (near-field) x/H = 11.0 obs x/H = 11.0 pred x/H = 30.0 obs x/H = 30.0 pred
SC
4
0 0.00
8
(z-zg)/H
6
10
M AN U
(z-zg)/H
8
(a1) EMU case C1: vertical concentration (near-field) x/H = 11.0 obs x/H = 11.0 pred x/H = 30.0 obs x/H = 30.0 pred
(z-zg)/H
10
Unsteady
8
(a2) EMU case C1: vertical concentration (far-field) x/H = 45.0 obs x/H = 45.0 pred x/H = 67.5 obs x/H = 67.5 pred
6
4
2
0.025
0.030
0 0.000
0.005
normalized concentration
0.010
0.015
0.020
0.025
normalized concentration
Figure 13: Comparison of measured and simulated concentrations for vertical profiles in the near-field (left) and in the far-field (right) (Case C1).
56
0.030
ACCEPTED MANUSCRIPT
! #
$
!
" !
!
'
!!
TE D
M AN U
SC
! & &
EP
% !
!
AC C
"
!
RI PT
"