Arabian Journal for Science and Engineering

https://doi.org/10.1007/s13369-024-09734-8

RESEARCH ARTICLE - SPECIAL ISSUE - CHALLENGES AND RECENT

ADVANCEMENTS IN NUCLEAR ENERGY SYSTEMS

Numerical Study of Near-Field Radionuclides Dispersion Around

Barakah Power Plant During Postulated Accidental Release Scenarios
Fatema Ali Almazrouei1,2 · Yacine Addad1,2 · Peter Rodgers1

Received: 21 April 2024 / Accepted: 28 October 2024

© The Author(s) 2024

Abstract
This study explores the assessment of hazards arising from nuclear power plant incidents, informed by the Fukushima
catastrophe. It evaluates the environmental impact of noble gases, such as iodine-131 releases, recognizing the limitations
of current local computational tools, particularly in predicting near-field dispersion accurately. Utilizing computational fluid
dynamics (CFD), this study validates this approach’s effectiveness in predicting pollutant dispersion around buildings. Among
the five turbulence models tested, the Lag Elliptic Blending (EB) k-ε model emerges as the most suitable for simulating
radioactive pollutant dispersion due to its superior performance in capturing flow dynamics. The findings underscore the
inadequacy of traditional Gaussian plume models in accounting for the effects of buildings on dispersion patterns. Notably,
simulations around the Barakah nuclear site located in the United Arab Emirates reveal the significant influence of buildings
on the trajectory of radioactive pollutants from hypothetical cracks. Consequently, it advocates caution in relying solely on
classical Gaussian plume models for evacuation plans, as they may overlook crucial flow patterns due to building presence,
potentially leading to distorted assessments of gas distribution and deposition rates.

Keywords Atmospheric boundary layer · RANS turbulence models validation · Air pollution

List of symbols l Turbulence length scale [m]

v2 Wall-normal stress component
Bq Becquerel [disintegration/s] f Elliptic relaxation parameter
b Width [m] q Released gas emission amount [m3 /s]
Cs Caesium y+ Dimensionless wall distance [-]
C Gas concentration [ppm] U Inflow Velocity [m/s]
Cgas Release gas concentration [ppm] Y Vertical distance [m]
C0 Reference gas concentration [ppm] y0 Aerodynamic roughness [m]
H Height [m] XR /b Normalized reattachment length on building top
k Turbulence kinetic energy [m2 /s2 ] [-]
L Length [m] XF /b Normalized reattachment length behind the
building [-]
B Yacine Addad z0 Aerodynamic roughness [m]
[email protected]
Fatema Ali Almazrouei
[email protected] Greek letters
Peter Rodgers
[email protected] ε Turbulence dissipation rate [m2 /s3]
1 Department of Mechanical and Nuclear Engineering, Khalifa ω Specific turbulence dissipation rate [s−1 ]
University of Science and Technology, Abu Dhabi, UAE ϕ v 2 /k [-]
2 Emirates Nuclear Technology Center (ENTC), Khalifa ρ Density [kg/m3 ]
University of Science and Technology, Abu Dhabi, UAE κ Von Karman constant

123
 Arabian Journal for Science and Engineering

Subscripts/Superscripts Navier Stokes (RANS), a task that will be thoroughly exam-

ined, evaluated, and authenticated in this study using the
H Variable value at the building height level commercial CFD software, Star-CCM + .
max Maximum For the validation and verification process of the RANS
turbulence models, two test cases (Cases A and B) are
selected. Case A considers flow around a building without
Abbreviation pollutants emission. This case was selected to assess the
RANS models’ performance in predicting the wind speed
APR Advanced power reactor distribution, recirculation zones size, and turbulence level at
AIJ Architectural institute of Japan different locations. Details about this case are provided in
BNPP Barakah nuclear power plant Sect. 3. Case B involves a flow around a similar building as
CMAQ Community multi-scale air quality model in Case A, but with different ground boundary condition and
CWE Computational wind engineering pollutant release. Given the outcomes of Cases A and B, the
EB Elliptic blending optimal RANS turbulence model has been discerned and sub-
EPA Environmental protection agency sequently employed to forecast the dispersion of accidental
GIS Geographical information system radionuclides around a nuclear power plant. This endeavor
NWP Numerical weather prediction entails the investigation of two distinct source release pos-
RLZ Realizable tulated scenarios emanating from the nuclear power plant’s
containment dome.
In the above context the originality of the present study lies
in several key aspects. Firstly, the Lag Elliptic Blending (EB)
k-ε turbulence model is used which has not been previously
1 Introduction tested for the type of external flows examined in this study.
This model is particularly well-suited for capturing chal-
In the Arabian Peninsula, diverse types of nuclear power lenging flow characteristics, such as non-equilibrium effects,
plants, encompassing both operational facilities and planned flow separation, and reattachment zones, which are com-
ventures, are paving the way for a dynamic energy landscape. monly encountered around nuclear plant structures. Unlike
During normal operation of a nuclear power plant (NPP), traditional eddy-viscosity models, the Lag EB k-ε model
a small fraction of the produced radionuclides is released includes blending functions that allow it to more accurately
to the atmosphere. Such release is estimated to be of the handle streamline curvature and anisotropy effects. These
order of 1014 Bq of activity per year for Kr-85 for instance in characteristics make it ideal for simulating pollutant disper-
the typical APR1400 design [1]. However, as demonstrated sion in environments with large geometric obstructions like
during the Fukushima accident, much higher values (mostly buildings, providing more precise predictions of flow behav-
noble gases, I-131 and Cs-137) can be released in the event ior and pollutant transport. The present research evaluates
of plant malfunction [2]. Accordingly, the critical elements this model’s performance in predicting radioactive pollutant
of decision-support systems for emergency preparedness and dispersion around nuclear power plant buildings and com-
response to hazardous nuclear events involve assessing the pares its findings with those from simulations using classical
level of threat to the population based on the modeling of the turbulence models. Secondly, the study employs computa-
radionuclides release to the environment. Hence, a rigorous tional fluid dynamics (CFD)-based modelling to account for
evaluation of the release impact on the environment is vital the effects of buildings on dispersion patterns, providing a
and must be assertively estimated. Especially near the release more accurate representation of pollutant transport paths,
source, and around the plant buildings due to direct impact concentration distribution, and deposition rates compared to
on the workers while accessing the plant during such urgent traditional Gaussian plume models. The research includes a
situations. rigorous validation process against benchmark experimen-
For this reason, the aim of this study is to concentrate on tal data for two specific test cases, ensuring the reliability
predicting how radioactive pollutants disperse in the vicinity of the CFD models used. Thirdly, the simulations conducted
of nuclear power plant buildings, employing computational around the Barakah nuclear site in the UAE aim to high-
fluid dynamics (CFD). The reason being that the buildings light the significance of buildings’ influence on the trajectory
influence the wind direction, which directly affects how the of radioactive pollutants from hypothetical release points,
plume behaves, including the path of pollutant transport, a finding crucial for developing more effective evacuation
distribution of concentration, rate of deposition, and their plans and emergency responses. By addressing these inno-
whereabouts. To accomplish this, it is crucial to select the vative aspects, the study aims to contribute to the field of
appropriate turbulence model based on Reynolds Averaged environmental impact assessment of nuclear power plants

123
Arabian Journal for Science and Engineering

and enhance the accuracy of predictive models for emergency modelling does have its advantages and is regarded as one of
preparedness and response. the preferred methods for evaluating potential releases and
dispersals at the vicinity of the plant, it is important to exer-
cise caution to ensure the reliability of its results. CFD allows
2 Literature Review simulating various release scenarios, accounting for changes
in the release intensity and environmental conditions, thereby
Modeling Scale Urban air pollution problems are character- providing detailed information about flow characteristics at
ized by varying source points, which are often located several every location simultaneously [8].
hundred meters from one another. The complex urban geom- Prior gas dispersion modelling studies Atmospheric dis-
etry is yet another dynamic affecting the dispersion problem persion modeling is used to estimate and quantify the
[3]. According to Blocken et al. [4], outdoor pollutant dis- deposition and concentration of radionuclides following
persions are often observed within macro-scale, mesoscale, unintentional releases. The goal is often to quantify the expo-
and micro-scale. Each of these scales is differentiated by sure dosage to humans and model countermeasure strategies.
the horizontal dispersion length (L). From macro, meso to According to Leelőssy et al. [10] a combination of numerical
micro scales, L > 6500 km, 10 km < L < 6500 km, and L weather predictions (NWP) systems developed by respective
< 10 km, respectively. Blocken [5] illustrated spatial mod- environmental protection agencies per country and several
elling scales in the built environment of pollutants dispersion, numerical models are required to predict, estimate, and
where CFD is utilized for microscale modeling at less than precisely quantify atmospheric radionuclides release, depo-
2 km. Near-field dispersion model is of interest since wind sition, and dispersion. For optimal results, the input of such
and buildings can affect plume behavior. Micro-scale pollu- data as meteorological conditions, including the quantity of
tant dispersion in the built environment can be assessed by atmospheric turbulence as well as the magnitude and direc-
field measurements, wind-tunnel testing and by numerical tion of the wind, emission parameters such as the height
simulations using the CFD approach [4]. and location of release source, terrain elevations, and the
Why the CFD approach For years, the Gaussian plume widths, heights, and precise positions of possible obsta-
and Puff gas dispersion models have been utilized in the pre- cles is paramount. Based on differing physical assumptions,
diction of atmospheric dispersion. For example, Ren et al. implementations, and numerical techniques, various model-
[6] devised an integrated solution for assessing radionuclide ing software can simulate atmospheric radionuclide transport
release accidents and formulating emergency evacuation and dispersion. According to Leelőssy et al. [10], spatial
schemes employing the Gaussian model. Similarly, Liu et al. scale, physical and chemical changes as well as climatic
[7] introduced a dispersion modelling scheme tailored to complexity and land surface must all be considered when
accommodate the building layout of a Chinese nuclear power developing a viable modeling tool. The success thereof is
plant site, employing the Puff dispersion model. However, determined by both the computational resources and avail-
it is evident that these models tend to yield errors when it able data. Leelőssy et al. [10] provide a list of atmospheric
comes to dispersion at the vicinity of numerous buildings and modeling systems widely used across the globe with limited
intricate sources of emissions. These models can simulate coverage. These models are fundamentally associated with
the spread of pollution considering various environmental their country of origin. For instance, in the United States,
and atmospheric factors, including wind direction and local the Environmental Protection Agency (EPA) uses models
topography. For instance, the Gaussian model’s effective- such as CMAQ and CALPUFF for regulatory modeling [10].
ness and precision are limited to cases involving flat terrains. Each of these models has varying operations capacities. The
Additionally, the original form of the Gaussian model can- dispersion modeling in France [10] considers one puff, one
not be applied when there are barriers obstructing the path, plume, and two Eulerian chemistry-transport models in the
between the source of emission and the receiver [8]. Gaus- French Polyphemus to compute radioactive decay, atmo-
sian models are known to provide poor results in situations spheric transport, and associated chemistry. SPEEDI systems
with low wind speeds, where the three-dimensional diffusion are used by the Japan Atomic Energy Agency for turbulence
is significant [3]. modeling [10].
Urban air pollution poses a challenge as both the sources Various studies have covered different aspects of atmo-
and the receptors are often situated in proximity within a spheric dispersion modeling. For instance, Blocken [9]
few hundred meters. The intricate geometry of areas makes illustrates the perspective of computational wind engineering
it difficult for Numerical Weather Prediction (NWP) models (CWE) analysis based on different spatial scales. According
to accurately depict the wind patterns within regions. Con- to Blocken et al. [11], atmospheric dispersions in the built
sequently, conventional dispersion models, whether Eulerian environment are often influenced by obstacles, which make
or Lagrangian fall short in their applicability [9]. In such a certain models, such as the original Gaussian model, inappli-
scenario utilizing CFD emerges as the choice. While CFD cable. Consequently, high-accuracy experimental data is thus

123
 Arabian Journal for Science and Engineering

crucial in validating turbulent flow CFD simulations, partic- escalate computational costs. Consequently, despite its avail-
ularly when they are based on RANS equations. Bogaersa ability in the commercial code Star-CCM+ , it was not used
and Rensburg [8] compared different CFD and Gaussian in the current study.
dispersion models and conclude that the Gaussian disper- Previous studies have predominantly relied on traditional
sion model is not suitable for predicting the dispersal of gas turbulence models, such as the Realizable k-ε and Gaus-
in proximity when obstacles or barriers are present. CFD sian plume models, to simulate pollutant dispersion around
predictions of possible radioactive releases were studied by nuclear power plants [4, 15]. However, these models exhibit
Addad and Al Noamani [12] based on United Arab Emi- significant limitations when applied to environments where
rates (UAE) environmental conditions. They observed that complex flow interactions with buildings are present. For
the different structures within the nuclear power plant have a instance, the Realizable k-ε model is known to underpre-
notable impact on how pollution spreads, disperses and set- dict reattachment lengths and recirculation zones, leading to
tles in the surrounding area. Additionally, the way hazardous inaccuracies in predicting pollutant dispersion in built envi-
releases are transported is significantly influenced by vari- ronments [5, 15]. Similarly, Gaussian plume models, while
ations in wind direction and temperature. Tang et al. [13] suitable for open and flat terrains, fail to account for the
used CFD analysis, in conjunction with geographical infor- influence of buildings on pollutant transport and deposition,
mation system (GIS) data, to examine how wind patterns, resulting in distorted predictions in urban or industrial set-
variations, in building induced pressure and the movement tings [7].
of pollutants is interacting within the atmosphere. This study Considering the limitations identified in the literature, the
indicates how topography impacts the wind patterns conse- primary objective of this study is to evaluate the dispersion
quently the spread of radioactive particles providing insights of radioactive pollutants around nuclear power plants using
into hazards and safety measures, in nuclear facility settings advanced Computational Fluid Dynamics (CFD) models.
[13]. This research will focus on validating several RANS tur-
Blocken et al. [14] evaluated numerically pollutant dis- bulence models, including the newly developed Lag Elliptic
persion within the built environment, comparing the perfor- Blending (EB) k-ε model, which has not been extensively
mance of various models and experiments. The study findings tested for complex flows around nuclear plant buildings.
were that prediction of the plume parallel to wind direction Through benchmark test cases and simulations at the Barakah
from stacks disperse within a range of about three times the Nuclear Power Plant (BNPP), the goal is to determine the
wind direction. However, the prediction underestimated lat- most accurate model for predicting pollutant dispersion and
eral dispersion [14]. Lateb et al. [15] compared different k–ε flow behavior in such environments. This study also aims
models for pollutant emissions using a configuration of two to highlight the shortcomings of traditional Gaussian plume
buildings. The study noted that the Realizable (RLZ) k–ε models when applied to complex geometries.
model was the one that accurately predicted the concentra- Accordingly, this study follows similar approach to prior
tion distribution in the region between the two buildings. CFD studies, where different RANS turbulence models are
Moreover, the standard k–ε model was proven inadequate in carefully assessed, and the most suitable one is selected.
reproducing the vertical concentration distribution. Leelossy Studies [9, 12, 15, 16] primarily concentrated on conven-
et al. [3], reiterate the importance of developing appropriate tional RANS turbulence models such as Realizable k–ε [17],
model strategies to handle atmospheric air pollutants. They k-ω SST [18] and the v2 -f [19] models. Nevertheless, this
also described several modelling tools and strategies, such study in addition to those models, also incorporates recently
as Gaussian, Eulerian, Lagrangian, and CFD models, noting developed models, namely elliptic blending (EB) k–ε [20]
their drawbacks and advantages. and Lag EB k–ε [21] models, into the validation process.
Near nuclear power plants, convection–diffusion equa- Notably, these models had not been evaluated in previous
tions or Gaussian plume models are mostly used to predict the studies.
associated air flows. According to de Sampaio et al. [16] such The rationale behind choosing these models lies in their
models are beneficial at the atmospheric mesoscale, espe- ability to accurately predict flow behavior in various condi-
cially, between 2 and 200 km. In the order of 0–2 km, the tions, such as separated flow, rotation, or strong streamline
local scale, where the material stream is released, the models curvature. Unlike the Realizable k–ε model, all other selected
cannot account for the ensuing turbulence, which is mostly models have demonstrated good predictive capabilities in
caused by the interaction of objects and wind. CFD based these scenarios. For example, the EB k–ε model, accord-
modelling, on the other hand, has the capability to handle ing to its authors [20], is one of the most robust models that
such critical details. Moreover, as indicated by Sampaio et al. offers improved accuracy compared to the Realizable k–ε
[16], the CFD code can be developed to incorporate mesh model, particularly near walls. It also outperforms the k-ω
adaptive techniques, thereby improving local flow resolution. SST model in terms of numerical stability and reliability.
Nonetheless, this approach is acknowledged to significantly

123
Arabian Journal for Science and Engineering

Fig. 1 Sketch of the CFD domain with 6H height from the ground for test case A

The Lag EB k–ε model [21] goes further by better predict- 3.1 CFD Model Validation without Dispersion
ing flow physics in areas with non-equilibrium effects, where
traditional linear eddy viscosity models tend to overestimate Test case A was selected for the validation of CFD model
certain terms. This model addresses this issue by consider- without dispersion to evaluate the ability of different RANS
ing the angle between relevant components and integrating turbulence models to reproduce the flow physics. The valida-
additional terms in the transport equation of the normalized tion reference data used for this test case was obtained from
(reduced) wall-normal stress component (ϕ) to account for the wind tunnel experiments series conducted at the Institute
anisotropy, curvature, and rotational effects. Concerning the of Japan (AIJ) [24]. Test case A consists of a flow around
Realizable k–ε model, although it has been previously tested a building with a height H = 0.16 m and width b = H/2.
and was found inadequate for accurately simulating flow A sketch of the test case A building, the domain size and
around buildings, in this study it underwent further scrutiny the boundary conditions are illustrated in Fig. 1. It is worth
to confirm its limitations and ensure consistency with pre- mentioning that the domain size and boundary conditions are
vious findings, as well as to conduct cross-validation with identical to the ones reported in Liu et al. [25].
other CFD codes.
In the following sections, the methodology used to vali-
3.1.1 CFD Domain Setting and Boundary Conditions
date the selected turbulence models and apply them to the
Barakah site is presented.
Mesh sensitivity analysis was carried out to aid in selecting
a mesh resolution satisfying grid independent solution. Con-
sidering both the accuracy and the physical time needed to
3 Methodology complete the simulations, a mesh size of 2,601,918 control
volumes was selected for this test case. The boundary layer
The present study employs a specific strategy to validate the has 10 prism layers with prism layer stretching factor of 1.1
CFD modelling methodology utilized in predicting scenarios and y + ~ 2.4. Convergence criteria were set below 10–5 for
for the dispersion of radioactive pollutants from the Barakah continuity, momentum, turbulence dissipation rate and turbu-
Nuclear Power Plant (BNPP). This methodology is subjected lent kinetic energy k. Five low-Reynolds RANS turbulence
to validation against benchmark data provided by the Archi- models (Realizable k–ε, elliptic blending (EB) k–ε, Lag EB
tectural Institute of Japan (AIJ) [22], encompassing Case k–ε, k-ω SST and v2 -f) were selected to assess their ability in
Studies A and B. Test case A is designed to validate the flow predicting the airflow conditions of the test case under con-
field physics, while test case B aims to validate dispersion sideration. The simulations were conducted in steady-state
phenomena. Upon successful validation of these aspects, the mode.
BNPP Facility is modelled at a 1:1 scale to evaluate: Inlet wind velocity U(z) and turbulent kinetic energy k(z)
profiles are defined using wind tunnel experiment data as
• The impact of release points of radionuclides. shown in Fig. 2, whereas the following equation is used to
• The impact of various BNPP buildings on the flow struc- define the profile of turbulence dissipation rate:
ture and potential deposition of radionuclides at local
levels.
ε = Cμ3/4 k 3/2 l −1 (1)
It is noteworthy that for all numerical simulations con-
ducted in this study, the chosen commercial CFD software is where cμ = 0.09, l = κ*z, κ = 0.41, z is the vertical distance
Star CCM + whose computational method is given in [23]. to the wall.

123
 Arabian Journal for Science and Engineering

Table 2 Test case A summary normalized reattachment length

Flow Condition Turbulence Model Reattachment

Length

XR /b XF /b

Steady state RANS Realizable k-ε – 3.25

EB k-ε 0.337 3.5
Lag EB k-ε 0.558 2.88
v2 -f >1 4.75
k-ω SST >1 4.125
Exp. data [24] – 0.52 1.42

XR /b and XF /b are normalized reattachment lengths on building top and

behind, respectively
Fig. 2 Test case A experimental data inflow profiles [22]

Table 1 Computational parameters and boundary conditions for test

case A vortex shedding and recirculation in the wake. The produc-
tion and dissipation terms in the k and ε equations are overly
Parameter CFD model setting smoothed in the wake, extending the recirculation zone.
While the model excels in near-wall regions, more advanced
Mesh type Unstructured polyhedral
turbulence models like LES or RSM are needed to accurately
Solver Steady, segregated flow,
resolve the vortex dynamics in the wake. It is worth mention-
pressure–velocity coupling (SIMPLE),
second-order discretization ing here that these elliptic blending models have been more
Wall treatment Lag EB k-ε: All y + recently developed, in comparison to the remaining RANS
Realizable k-ε: Two layers all y + wall models, and were not tested for this test case before, which
Boundary conditions Symmetry for the sides and top, non-slip makes this study an even more necessary one.
condition for the ground Figure 3 presents streamwise velocity (U) profiles for
Velocity inlet for the inflow boundary position locations -a-, -b-, and -c-. For the positions -a- and
Pressure outlet for the outlet boundary
-c-, overall, the velocity (U) profile predictions are in good
agreement with measurement with all k–ε based models but
not with k-ω SST model, which is observed to significantly
The numerical parameters and boundary conditions for deviate from the experimental data. However, at position -b-
the numerical simulation are given in Table 1. (top of the building), significant discrepancies exist between
the models’ predictions and the measurements. The excep-
3.1.2 Test Case A Results and Discussion tion to this observation is the velocity profile predicted with
the Lag EB k–ε model which is seen to agree better with
Table 2 summarizes the reattachment lengths obtained from the experimental measurements. This is in alignment with
the different models’ prediction in comparison with the wind the recirculation zone size predictions reported in Table 2.
tunnel measurements. This table shows that the Realizable Figure 4 shows turbulent kinetic energy (k) profiles at the
k–ε model is not able to predict the recirculation zone above locations -a-, -b-, and -c-. It is interesting to notice that, at the
the building, while this recirculation zone is over predicted location -a-, significant discrepancies exist between Realiz-
by the v2 -f and k-ω SST models. On the other hand, the able k–ε, Lag EB k–ε models’ predictions and measurement.
elliptic blending models, in particular the Lag EB k–ε, are Whereas k-ω SST predictions are much closer to the refer-
returning values much closer to the experiments. The Lag ence data at this location. This discrepancy at position -a-
EB k-ε model, for instance, closely predicts the reattachment might stem from known issues highlighted in studies such as
length on top of the building (XR /b) due to the elliptic blend- Blocken et al. [26], namely, inlet profiles significantly influ-
ing function, which enhances the prediction of the turbulent ence CFD accuracy for atmospheric boundary layer flows
kinetic energy (k) and dissipation rate (ε) near walls, cap- over flat terrain, with minor profile modifications causing
turing anisotropic turbulence and non-equilibrium effects. notable flow field changes. To address this, one suggested
However, it overpredicts the reattachment length behind the remedy is to reduce the upstream computational domain
building (Xf /b) because the model’s isotropic eddy-viscosity length. Another option is to artificially decrease turbulent
assumption struggles to capture the impact of the unsteady kinetic energy at the inlet to reduce momentum transfer and

123
Arabian Journal for Science and Engineering

[24]
a b c

Inlet U

X=-0.06 X=0 X=0.06

-a- X = -0.06

-b- X = 0.00 -c- X = 0.06

Fig. 3 Test case A normalized mean streamwise velocities at positions a, b, and c

subsequent wind speed acceleration near the surface. How- 3.2 CFD Model Validation with Dispersion
ever, to facilitate fair comparison, this study did not include
these adjustments, where the domain size followed Liu et al.’s Test case B was selected for the validation of RANS turbu-
study [25] and the inlet profiles were obtained from the exper- lence models with dispersion in isothermal flow to further
imental data. At location -b-, however, the k profiles predicted investigate their prediction ability of the flow around a build-
with Realizable k–ε, and the Lag EB k–ε models are in bet- ing with pollutants release. The reference data utilized for
ter agreement with reference data in comparison to the ones test case B were acquired from the experimental study con-
obtained with k-ω SST model. Similarly, to the velocity pro- ducted by Tanaka et al. [27] in an isothermal wind tunnel.
files, these results also agree with the recirculation zone size A building model (2:1:1) installed in the wind tunnel has
predictions reported in Table 2. dimensions of 0.2 m in height and 0.1 m in width. The cor-
The results obtained in the present study reveal that pre- responding Reynolds number, based on the building’s height
dictions obtained with the recently developed turbulence (H), is 56,000. The gas is emitted from a 4 mm diameter
models, specifically the Lag EB k–ε model, are much supe- orifice in the ground, positioned 0.05 m behind the model
rior to the remaining RANS models. Accordingly, the Lag EB building. The emission point comprises tracer gas, namely,
k–ε model will be further assessed for flow around building ethylene (C2H4). The cross-section of test case B and the
predictions in test case B with pollutants release. model geometry of the wind tunnel experiment are depicted
in Figs. 5 and 6, respectively. Table 3 summarizes the gas
release concentration, amount, and the reference gas con-
centration as reported in the Architectural Institute of Japan

123
 Arabian Journal for Science and Engineering

Fig. 4 Test case A turbulent

kinetic energy profiles at
positions a, b, and c
[24]
a b c

Inlet U

X=-0.06 X=0 X=0.06

-a- X = -0.06

-b- X = 0.00 -c- X = 0.06

wind wind
50mm

Gas flow rate = 350 cc/min

Dia = 4 mm
Gas exit
200mm
100mm dia=4mm

100mm

100mm 52mm

Fig. 5 Experimental flow field measurement for test case B [24]

(AIJ), website [24]. Figure 6 schematic diagram of the com- experiment and a CFD simulation conducted, in this study,
putational domain is similar to Keshavarzian et al. [28]. It is using the commercial code STAR CCM + . The inlet wind
worth noting that the domain size entered here adheres to the velocity U(z) and turbulent kinetic energy k(z) profiles are
guidelines established by the AIJ [24]. defined using wind tunnel experiment data as shown in Fig. 7.
Hence, the fitted profile of inlet velocity is prescribed as:
3.2.1 Domain Setting and Boundary Conditions
U (z) = (u ∗ /κ) × ln((z + z 0 )/z 0 ) (2)
Test case B involves an airflow around a simple cuboid (1:1:2)
building model with dispersion, aimed at elucidating the flow
physics in the presence of pollutants. Several CFD studies where κ = 0.42, z vertical distance to the wall, z0 aerody-
[28–30] have endeavored to numerically predict this test case namic roughness = 0.013 m, and the friction velocity, u ∗ , is
to evaluate and validate the outcomes of different turbulence equal to 0.484533.
models. In this investigation, the results obtained from these Turbulent kinetic energy profile is fitted as:
studies, particularly those reported by Keshavarzian et al.
[28], are juxtaposed with the outcomes of a wind tunnel k(z) = (1.03) × (0.26z)exp(−2.3z) (3)

123
Arabian Journal for Science and Engineering

Finally, the profile of turbulence dissipation rate is esti-

mated as:

ε(z) = u ∗ 3 /(κ(z + z 0 )) (4)

For these numerical simulations, the steady Lag EB k-ε

turbulence model was selected. The novelty of this work is
a result of the fact that this turbulence model has not yet
been tested in this type of external flows. Simulations using
the Realizable k-ε turbulence model were also carried out
to compare the findings to those of the Keshavarzian et al.
[28] using that model. The computational parameters and
boundary conditions for the present simulation are outlined
in Table 4. As for the convergence criteria, these were set
below 10–5 for continuity, momentum, turbulence dissipation
Fig. 6 Schematic diagram of CFD calculated computational domain for
test case B [28] rate and turbulent kinetic energy.

Table 3 Test case B experimental conditions data 3.2.2 Mesh Sensitivity

Variable Symbol (unit) Isothermal A mesh sensitivity analysis was conducted to assist in the
State
selection of a mesh with optimal computational efficiency
Released tracer gas Cgas (ppm) 1.0 × 106 and to ensure the independence of results from the grid. Five
concentration meshes were generated for the grid sensitivity analysis. The
Released gas emission q (m3 /s) 5.83 × 10–6 mesh underwent gradual refinement, resulting in 297,912
amount control volumes for Mesh-0, 489,846 control volumes for
Building height H (m) 0.2 Mesh-1, 713,405 control volumes for Mesh-2, 2,288,246
Inflow velocity at < uH > (m/s) 3.226 control volumes for Mesh-3, and 3,165,727 control volumes
computational domain for Mesh-4. The steady Lag EB k–ε turbulence model was
height H selected for mesh sensitivity analysis. As illustrative exam-
Reference gas concentration C0 (ppm) 45.21 ples, Fig. 8 shows the grid resolution for the coarse mesh,
[cgas q / < uH > Hˆ2]
Mesh-0, and fine one Mesh-4. At the wall vicinity, boundary
layer with 5 prism layers and prism layer stretching factor of

[28]

-a- Inflow turbulence dissipation -b- Inflow velocity profile (U). -c- Inflow turbulent kinetic
rate (ε). energy profile (k).

Fig. 7 CFD inflow profiles for test case B

123
 Arabian Journal for Science and Engineering

Table 4 Computational parameters and boundary conditions for test Table 5 Summary of normalized reattachment length for test case B
case B
Turbulence model Normalized reattachment length
Parameter CFD model setting
XR /H XF /H
Mesh type Unstructured polyhedral
Lag EB k-ε 0.197 1.28
Solver Steady, segregated flow,
pressure–velocity coupling RLZ k-ε No recirculation 1.522
(SIMPLE), second order Experiment [28] 0.266 ± 0.021 0.686 ± 0.064
discretization
Wall treatment Lag EB k-ε: All y + , XR /H and XF /H are normalized reattachment lengths on building top
Realizable k-ε: High Y + and behind, respectively
Computational domain boundary Symmetry for the sides and
conditions top, non-slip condition for
the ground an evaluation of the capacity of various RANS models to
Velocity inlet for the inflow mimic and replicate diverse flow phenomena in the vicinity
boundary of the building.
Pressure outlet for the outlet
Table 5 provides a summary of the reattachment lengths
boundary
extracted from the predictions of the steady-state RANS
Source inlet Wall passive scalar flux =
7806.4928 ppm kg/m2 s simulations, specifically using the Realizable k–ε and Lag
Schmidt Number Molecular diffusivity =
EB k–ε models. The reattachment length atop is named the
1.35059 m2 /s XR and while the one behind the building is XF . These
Turbulence = 0.9 predictions are compared with wind tunnel measurements
documented by Tanaka et al. [27]. The normalization of the
recirculation length was achieved by utilizing the height (H)
1.1 was employed. The resulting dimensionless distance, y+ , of the building.
to the wall is about 15.7. As illustrated in Table 5, in contradiction to the experi-
Figures 9, 10 and 11 illustrate a comparison of the normal- mental data, the predictions derived from the Realizable k–ε
ized mean velocity magnitude, the turbulent kinetic energy, model predictions did not indicate a reattachment on the roof
k, and the mean concentration (C/C0) profiles, respectively, of the building. However, behind the building, the recircula-
using the five meshes. The simulation outcomes for the pro- tion zone (XF /H) obtained with this model was significantly
files behind the building at x/H = 0.25, 0.5 and 1.0 do not greater than observed in the experiment. The simulation con-
differ noticeably between the five meshes. As there was no ducted with the Lag EB k–ε model exhibited a recirculation
discernible mesh sensitivity, Mesh-1 resolution was deemed zone on the top of the building, approximately 30% shorter
adequate and used for this study’s validation scenario. than the measured value, and a recirculation zone nearly dou-
ble the size of the experimentally determined value behind
3.2.3 Test Case B Results and Discussion the building. Although both values still deviate consider-
ably from an excellent agreement with the reference data,
The proficiency of RANS turbulence models in accurately the model is evidently more adept at capturing realistic flow
predicting the reattachment lengths atop and behind the physics when compared to its eddy-viscosity-based counter-
building is initially evaluated. These flow characteristics bear part models as demonstrated in the previous test case (Case
paramount significance, given their direct influence on the A).
transport and distribution of pollutant emissions. Specifi- Figure 12 depicts the vertical profiles of the mean stream-
cally, the pollutant behavior in proximity to the recirculation wise velocity (U/UH ) at various sections, aiming to evaluate
zone’s periphery is directly influenced by the local flow the efficacy of diverse turbulence models. In reference to the
dynamics. Furthermore, the present sensitivity study enables measurement data, Willemsen et al. [31] observed that the

-a- Mesh-0 -b- Mesh-4

Fig. 8 X–Z plane taken at y = 0 to illustrate the coarse and fine grids resolution for test case B

123
Arabian Journal for Science and Engineering

c d f

Inlet U

Gas Exit

X/H= X/H= X/H=

0.25 0.50 1

-c- X/H = 0.25

-d- X/H = 0.50 -f- X/H = 1.00

Fig. 9 Mean velocity profiles behind the building obtained as a function of mesh for test case B

c d f

Inlet U

Gas Exit

X/H= X/H= X/H=

0.25 0.50 1

-c- X/H = 0.25

-d- X/H = 0.50 -f- X/H = 1.00

Fig. 10 Turbulent kinetic energy profiles behind the building obtained as a function of mesh for test case B

123
 Arabian Journal for Science and Engineering

c d f

Inlet U

Gas Exit

X/H= X/H= X/H=

0.25 0.50 1

-c- X/H = 0.25

-f- X/H = 1.00

-d- X/H = 0.50
Fig. 11 Mean concentration (C/C0) behind the building obtained with the four meshes for test case B

standard measurement error for these wind tunnel experi- to numerical parameter selection issues or other test-case
ments can be reasonably estimated at 20%. This underscores setup parameters such as wrongly implemented boundary
the importance of acknowledging potential errors in wind conditions. For instance, a higher inlet velocity may cause
tunnel data. Consequently, Fig. 12 includes error bars, illus- such a behavior but is not immediately noticeable, especially
trating a deviation from the measurement data of no more when all variables presented in the figure are normalized.
than 20%. Nevertheless, all models effectively reproduce the mea-
For sections -a- and -b-, situated upstream of the build- surement data in the region above the building (z/H ≥ 1.0)
ing, no considerable discrepancy with the experimental data at all locations from -a- to -f -. This is understandable by
is observed in the velocity profile. This initially instills con- recognizing that the velocity gradients are less sharp in that
fidence that the prescribed inlet velocity has been correctly elevation, thereby diminishing the impact of viscous effects.
implemented in the commercial code. Secondly, an agree- Figure 13 illustrates the vertical profiles of turbulence
ment with previous numerical studies for this specific region kinetic energy (k/UH 2 ) at various sections. A notable discrep-
of the domain is also evident. Moving further downstream, ancy is apparent in the turbulent kinetic energy obtained from
however, at locations -c-, -d-, -e-, and -f -, the differences both RANS models across all sections, spanning from -b- to
between different models’ predictions and the reference data -f -. This encompasses the region upstream of the building to
are more pronounced. In line with the aforementioned results, the entirety of the wake zone. This implies that although the
the Lag EB k–ε model indicates a slower flow recovery, trans- different turbulence models can reproduce the velocity field,
lating to a longer reattachment length as reported in Table 5. the same cannot be said for the turbulence levels. For exam-
Although not with the same amplitude as the experiment, ple, the RANS models underestimate the turbulent kinetic
this trend aligns with the findings that this model predic- energy in the circulation zone at the building wake, sections
tions are superior to the remaining RANS models tested in -c- to -f -. This underestimation of k values is consequently
this study. Predictions from the Realizable k–ε model reveal leading to the overestimation of reattachment length which
a significantly larger recirculation zone, consistent with the is associated with the vortex shedding and flow separation.
reattachment length reported in Table 5. These flow phenomena often occur at the lee side and top
Surprisingly, the velocity profiles reported by parts of the building [25, 28]. Also, for the turbulent kinetic
Keshavarzian et al. [28] for these locations are some- energy profiles the predictions obtained with the Lag EB k-
what peculiar, suggesting that the RANS model may be ε turbulence model are superior to the ones obtained with
also struggling to capture the recirculation zone behind Realizable k–ε model, especially at the sections from -c- to
the building. Given that the authors claimed to have used -f -.
the same RANS model (Realizable k–ε model), it remains Figure 14 illustrates the vertical profiles of gas concentra-
unclear whether these significant differences are attributed tion (C/Co) at different sections behind the building, aiming

123
Arabian Journal for Science and Engineering

Fig. 12 Profiles of the a b c d e f

normalized mean streamwise
velocity at the plane (y = 0)
[28]
[27]

Gas Exit

X/H X/H= X/H= X/H= X/H= X/H=

=-1 -0.75 0.25 0.50 0.75 1

-a- X/H = -1.00 -b- X/H = -0.75

-c- X/H = 0.25 -d- X/H = 0.50

-e- X/H = 0.75 -f- X/H = 1.00

to examine the impact of various turbulence models on the particularly at elevations where z/H ≤ 0.5 and z/H ≥ 1. Within
dispersion of pollutants. Additionally, an exploration into the the range of 0.5 ≤ z/H ≤ 1, the models effectively capture the
influence of turbulent Schmidt number on pollutant concen- pollutant profile trend, albeit with an overestimated ampli-
tration was conducted employing the Lag EB k-ε turbulence tude. However, it is worth noting that the predictions reported
model. Due to the considerable variation in concentration by Keshanvrzian et al. [28] do not appear to signify a com-
values near the release point, the concentration profiles are parable level of agreement with the experimental data. This
presented in a logarithmic scale. discrepancy raises concerns, suggesting that these numerical
In section -c-, the pollutant concentration predicted by the predictions should be approached with caution. Therefore, it
RANS models selected in this study exhibits minor discrep- is important to exercise prudence before passing judgment on
ancies among the various models. Nonetheless, all models the model’s capacity to accurately replicate the flow physics
demonstrate notable agreement with the experimental data,

123
 Arabian Journal for Science and Engineering

a b c d e f

[28]
[27]
Gas Exit

X/H X/H= X/H= X/H= X/H= X/H=

=-1 -0.75 0.25 0.50 0.75 1

-a- X/H = -1.00 -b- X/H = -0.75

-c- X/H = 0.25 -d- X/H = 0.50

-e- X/H = 0.75 -f- X/H = 1.00

Fig. 13 Profiles of the normalized turbulent kinetic energy at the vertical center plane (y = 0)

123
Arabian Journal for Science and Engineering

c d e f

[28]
[27]
Gas Exit

X/H= X/H= X/H= X/H=

0.25 0.50 0.75 1

-c- X/H = 0.25 -d- X/H = 0.50

-e- X/H = 0.75 -f- X/H = 1

Fig. 14 Profiles of the normalized mean concentration at the vertical center plane (y = 0)

in this test case, particularly when relying solely on the pre- the impact of underpredicted turbulent kinetic energy pro-
dictions reported in [28]. files on the magnitude and distribution of gas concentration,
In Sections -d- to -f -, the normalized pollutant concentra- the turbulence Schmidt number was reduced from the stan-
tion profiles exhibit comparable trends to the experimental dard value of 0.9 to 0.7, effectively increasing the turbulent
data. Nevertheless, the values are overpredicted at eleva- diffusion term in the scalar transport equation. The findings
tions where z/H ≥ 0.5. The predictions of the mean pollutant indicate that decreasing the turbulence Schmidt number from
concentration near the ground, however, are underestimated. 0.9 to 0.7 results in no discernible difference in gas concen-
The marked disparities between the RANS simulation and tration.
experimental measurements arise from the overestimation of Hence, it is conceivable to deduce that, overall, the out-
the reattachment length, XF /H, which was observed to be comes derived from the Lag EB k-ε and Realizable k-ε RANS
twice the experimental measured value. Finally, to evaluate

123
 Arabian Journal for Science and Engineering

models manifest a tendency closely aligned with the experi- 2019. Furthermore, during these months, wind speeds mea-
mental data. Nevertheless, it is noteworthy that the accuracy sured at a height of 10 m above the ground range from 3 to
of the Lag EB k-ε model surpasses that of the Realizable k-ε 6 m/s. Consequently, an average freestream value of 5 m/s
model to a considerable extent. was selected for this study, as elaborated in Sect. 3.3.1 below.
It is also crucial to observe that, for further enhanced In relation to the emission of radionuclides, two hypothet-
performance in CFD simulations, one must transition from ical cracks in containment have been randomly generated as
RANS models to a more advanced approach, such as Large circular holes with a diameter of 50 cm each. The release
Eddy Simulation (LES), as exemplified in reference [28] from each crack is simulated separately to evaluate the release
for this test case and in [32], and [33] for flow around location impact on the transport path of the radioactive pollu-
bluff bodies. However, this shift entails significantly higher tant (specifically, radionuclide I-131 gas) and the location of
computational resource requirements. Additionally, it is peak concentration near the power plant units. As illustrated
noteworthy that without the utilization of a well-resolved in Fig. 21a and b, Source 1 is positioned at the front of the
mesh, LES predictions may potentially be inferior to RANS dome, while Source 2 is situated at the rear. A total amount of
predictions, as evidenced by Addad et al. [34]. This presents I-131 released into the air during these hypothetical accident
yet another challenge in the execution of LES simulations scenarios is approximately 2.11 × 1017 Bq (refer to Table 6
for urban air pollution problems. for the prescribed volumetric rate value), which corresponds
Henceforth, drawing upon the collective findings of Cases to the total amount released during the Fukushima nuclear
A and B, the following conclusions emerge: The Realizable accident as documented in the work of An et al. [30].
k-ε model falls short in predicting reattachment on the roof of
the building, whereas the Lag EB k-ε model demonstrates a 3.3.1 Domain Setting and Boundary Conditions
closer agreement with experimental data. The mean velocity
fields on the windward side, as predicted by RANS mod- The study under consideration accounts for various site char-
els, exhibit strong concordance with experimental results, acteristics. These include the flatness of the terrain, the
albeit discrepancies arise in the near wake region. Notably, inclusion of all significant structures within the power plant,
significant disparities in turbulence kinetic energy profiles the use of scaled buildings within the computational domain,
are discernible when juxtaposed with measurement data. and alignment of velocity direction and magnitude with
Moreover, RANS models’ predictions regarding pollutant previously documented measurements. In Fig. 16, the inlet
concentration display variances across different models and boundary profiles pertaining to wind velocity U(y), turbulent
experimental measurements yet maintain satisfactory confor- kinetic energy k(y), and turbulence dissipation rate ε(y) are
mity with reference data. Ultimately, the predictions derived established in accordance with the Richard and Hoxey Law
from the Lag EB k-ε model emerge as the closest to the [36], as delineated as follows:
experimental results for both Cases A and B. Consequently,
the Lag EB k-ε model is deemed the optimal choice for sim- U (y) = (u ∗ /κ) × ln((y + y0 )/y0 )) (5)
ulating the flow around a Nuclear Power Plant with dispersed
radioactive pollutants.
u∗2
k(y) = √ ; cmu = 0.22 (6)
cmu
3.3 CFD Model for Barakah Nuclear Site
ε(y) = u ∗ 3 /(κ(y + y0 )) (7)
Figure 15 depicts the study area of the Barakah Nuclear
Power Plant (BNPP) site illustrating the computational where κ = 0.42, y is the normal distance to the wall, y0 is the
domain’s extension. This domain stretches from 5 × Hmax aerodynamic roughness = 0.005 m [37], the friction velocity,
upstream (to the north of) the plant’s units, to 15 × Hmax in u ∗ , is equal to 0.2477, estimated using following equation:
the streamwise direction (towards the south). Additionally, it
extends by 5 × Hmax in the lateral directions (east and west). u∗ = κU H /ln((h + y0 )/y0 ) (8)
Here, Hmax corresponds to the height of the units’ domes,
which is 75 m [1]. All the buildings are 1:1 scale with the where UH = 5 m/s, h = 24 m.
BNPP site. The prevailing wind direction is from the north, In conducting the numerical simulation for these simula-
perpendicular to the power plant buildings. This information tions, the steady Lag EB k-ε turbulence model was selected.
aligns with findings from Abida et al. [35], which indicated A comprehensive summary of all computational parameters
that winds around the BNPP predominantly blow from the and boundary conditions pertinent to the current simulation
northeast, north, and north-northeast, as per wind roses cal- can be found in Table 6. As for the criteria ensuring conver-
culated from weather station data for June 2019 and January gence, these were established to be below 10–5 for continuity,

123
Arabian Journal for Science and Engineering

Fig. 15 Nuclear power plant location and size of the computational domain

Table 6 The computational parameters and boundary conditions for Barakah NPP case

Parameter CFD model setting

Mesh type Unstructured polyhedral

Solver Steady, segregated flow, pressure–velocity coupling (SIMPLE), second order discretization
Wall treatment Lag EB k-ε: All y +
Boundary conditions Symmetry for the sides and top, non-slip condition for the ground
Velocity inlet for the inflow boundary
Pressure outlet for the outlet boundary
Source inlet Q I −131gasemission = 5.27 ∗ 10−12 m3 /s based on the average release rate (Bq/h) from the Fukushima accident in
which the release lasted for 475 h and the total amount of I-131released into the air was around 2.11 ∗ 1017 Bq
(Terada et al. [38])
Schmidt Number Molecular diffusivity = 51.8342 m2 /s
Turbulence = 0.9

123
 Arabian Journal for Science and Engineering

Fig. 16 Inflow profiles prescribed for the Barakah NPP site simulations

momentum, turbulence dissipation rate, and turbulent kinetic other, contrasting with the somewhat underpredicted profiles
energy. obtained with the coarse grid (Mesh-1). Consequently, based
on these observations, Mesh-2 resolution was deemed ade-
quate and was consequently employed for the simulations
3.3.2 Mesh Sensitivity Assessment concerning the two postulated accident scenarios.

In a manner akin to the preceding validation tests outlined

above, a mesh sensitivity analysis has been also undertaken 3.3.3 Results and Discussion for the BNPP Case
in this instance to ascertain the attainment of a solution inde-
pendent of grid size. Accordingly, three grids, employing an 3.3.3.1 Recirculation Zones Figure 21 illustrates the distri-
unstructured mesh type (as depicted in Fig. 17a and b), were bution of field on a 2D planes, showcasing the streamwise
generated. These are Mesh-1, comprising 3,588,292 control velocity (U) as predicted by the EB k–ε turbulence model.
volumes; Mesh-2, consisting of 11,759,988 control volumes; Within this depiction, both plots delineate the Z-X plane sit-
and Mesh-3, with at total control volumes of 16,710,751. uated at a height of 10 m above ground level (refer to Fig. 21b
Adjacent to the walls, a boundary layer featuring 30 prism for the plane location) and the X–Y plane at Z = − 1 m (refer
layers and a prism layer stretching factor of 1.12 was used. to Fig. 21a for the plane location). These representations
Consequently, for all three grids, the initial cell maintains unveil the presence of diverse recirculation zones, observ-
a consistent distance from the wall, distributed between 30 able upstream, above, alongside, and behind the buildings.
and 100, as delineated in Fig. 17c. Notably, exceptions to this Such observations vividly underscore the intricacies inherent
uniformity arise in impingement zones where the maximum in the flow surrounding the plant units, which are anticipated
value unavoidably exceeds this range, and in recirculation to directly influence the dispersion of radioactive pollutants,
zones where the normalized distance unavoidably falls below particularly in the vicinity of the facility. Furthermore, it sug-
the threshold value of 30 as the wall shear stress drops to small gests that the trajectory of pollutant transport and diffusion is
values in the recirculation zones. susceptible to the influence of local flow dynamics, even in
Figures 18, 19 and 20 depict a comparison of pre- scenarios where the incoming flow remains at zero incidence.
dictions for the normalized streamwise velocity, turbulent This observation gains further significance upon examination
kinetic energy and mean concentration profiles, respectively, of gas distribution in the context of two accident scenarios
acquired from three different grids. These profiles are sit- considered in this study, a discussion which will follow in
uated downstream of the plant buildings at x = 120 m, the following two subsections.
200 m, and 250 m, as shown in the figures. The plots reveal
a notable agreement between the velocity and concentration
profiles. However, for the turbulent kinetic energy, Mesh- 3.3.3.2 Gas trajectories at the Plant’s Vicinity Figure 22a
2 and Mesh-3 predictions exhibit good agreement with each and b depict the trajectory of gas upon its release from the

123
Arabian Journal for Science and Engineering

Fig. 17 The unstructured mesh used in the simulation and near wall resolution

Fig. 18 Normalized mean velocity profiles behind the plant’s buildings

123
 Arabian Journal for Science and Engineering

Fig. 19 Normalized turbulent kinetic energy profiles behind the plant’s buildings

Fig. 20 Normalized I-131 Gas concentration (C/Cmax ) behind the plant’s buildings

123
Arabian Journal for Science and Engineering

Fig. 21 Mean streamwise velocity (U) field distribution

Fig. 22 Transport path and field distribution for radioactive I-131 gas

hypothetical first and second cracks, respectively. It is evi- the roofs is notably more pronounced, with gas even reaching
dent from the illustrations that in the former scenario, the the second unit, as observed in Fig. 22a and b. This contrast is
gas veers towards one side of the building, resulting in only further apparent in Fig. 22c and d, wherein the concentration
a relatively small portion being deposited on the roofs. Con- maximum values are seen to vary between the two scenarios.
versely, in the latter scenario, the spread of pollutants onto Consequently, these findings furnish compelling evidence of

123
 Arabian Journal for Science and Engineering

the significant influence exerted by buildings on pollutant tra- research study by proposing an evacuation plan using the
jectory and diffusion in the nearby vicinity of the nuclear site. CFD approach and comparing outputs with the one based
Hence, it is advisable that evacuation plans for the workers, on the Gaussian model, hence, providing insights into errors
traditionally reliant on classical Gaussian plume models such magnitudes when such plans rely on simpler dispersion mod-
as those proposed in [6], be approached with caution. Such els.
classical models inherently fail to explicitly incorporate the
Acknowledgements This study was conducted in the frame of the
flow patterns in the vicinity of buildings, potentially leading Project “Modelling of Radionuclides Dispersion in the UAE Envi-
to distorted assessments of gas distribution and deposition ronment (MORAD)” funded by the Federal Authority for Nuclear
rates. It should be clarified that the focus of the compari- Regulation (FANR), UAE. The opinions presented herein are exclu-
sively those of the authors and do not necessarily reflect the viewpoints
son illustrated in Fig. 22 was primarily on the trajectory of
of the MORAD project collaborators or their team members.
the radioactive gas plume, for which the present RANS tur-
bulence model has demonstrated reliable predictions when Funding This work was supported by Federal Authority for Nuclear
Regulation (UAE), (Grant No. 8434000306), Yacine Addad.
compared to experimental data in the second validation test
case (Case B). Predictions for the deposition rate have been
excluded from the manuscript to avoid misleading readers, as Open Access This article is licensed under a Creative Commons
discrepancies were observed between the model and experi- Attribution 4.0 International License, which permits use, sharing, adap-
mental data in the sublayer region for the validation test case. tation, distribution and reproduction in any medium or format, as
long as you give appropriate credit to the original author(s) and the
source, provide a link to the Creative Commons licence, and indi-
cate if changes were made. The images or other third party material
4 Conclusion in this article are included in the article’s Creative Commons licence,
unless indicated otherwise in a credit line to the material. If material
is not included in the article’s Creative Commons licence and your
The present study aims to predict the dispersion of radioac- intended use is not permitted by statutory regulation or exceeds the
tive pollutants in the vicinity of a nuclear site using the permitted use, you will need to obtain permission directly from the copy-
Computational Fluid Dynamics (CFD) approach. Five dif- right holder. To view a copy of this licence, visit http://creativecomm
ferent turbulence models were tested to assess their accuracy ons.org/licenses/by/4.0/.
and capability in predicting flow physics around buildings.
These models include the Realizable k–ε model, the k-ω SST
model, the v2 -f model, the elliptic blending (EB) k–ε model,
and the Lag EB k–ε model. Two validation test cases were References
utilized for evaluating the prediction capabilities of these
1. Korea Electric Power Corporation and Korea Hydro & Nuclear
models. The first case involved flow around a building with-
Power Co., APR1400 design control document tier 2, Chapter 11
out gas emissions, while the second case simulated flow with Radioactive Waste management. APR1400-K-X-FS-14002-NP,
a similar building geometry but with gas emissions. The pre- Revision 3, (2018).
dictions from these models indicated that the Realizable k-ε 2. Ain Sulaiman, S.N.; Mohamed, F.; Ab Rahim, A.N.: Radioactive
release during nuclear accidents in Chernobyl and Fukushima. IOP
model performed the worst, particularly in predicting reat-
Conf. Ser. Mater. Sci. Eng. 298, 012011 (2018). https://doi.org/10.
tachment on the roof of the building. Conversely, the Lag 1088/1757-899X/298/1/012011.
EB k-ε model demonstrated much superior agreement with 3. Leelőssy, Á.; Molnár, F.; Izsák, F.; Havasi, Á.; Lagzi, I.; Mészáros,
experimental data. Consequently, predictions derived from R.: Dispersion modeling of air pollutants in the atmosphere: a
review. Open Geosci. 6(3), (2014). https://doi.org/10.2478/s13533-
the Lag EB k-ε model were the closest to the experimental
012-0188-6.
results for both validation test cases. Therefore, the Lag EB 4. Blocken, B.; Tominaga, Y.; Stathopoulos, T.: CFD simulation of
k-ε model is considered the optimal choice for simulating micro-scale pollutant dispersion in the built environment. Build.
flow around a Nuclear Power Plant with dispersed radioac- Environ. 64, 225–230 (2013). https://doi.org/10.1016/j.buildenv.
2013.01.001
tive pollutants. 5. Blocken, B.: Computational fluid dynamics for urban physics:
Simulations conducted for flow around the Barakah importance, scales, possibilities, limitations and ten tips and tricks
nuclear site revealed that the trajectory of radioactive pol- towards accurate and reliable simulations. Build. Environ. 91,
lutants from hypothetical cracks is significantly influenced 219–245 (2015). https://doi.org/10.1016/j.buildenv.2015.02.015
6. Ren, Y.; Zhang, G.; Zheng, J.; Miao, H.: An integrated solution
by the presence of the plant’s buildings. This suggests that for nuclear power plant on-site optimal evacuation path planning
evacuation plans for workers, traditionally based on classical based on atmospheric dispersion and dose model. Sustainability
Gaussian plume models, should be approached with caution. 16(6), 2458 (2024). https://doi.org/10.3390/su16062458
These models inherently fail to explicitly incorporate flow 7. Liu, Y.; Li, H.; Sun, S.; Fang, S.: Enhanced air dispersion modelling
at a typical Chinese nuclear power plant site: coupling RIMPUFF
patterns due to the presence of buildings, potentially leading with two advanced diagnostic wind models. J. Environ. Radioact.
to distorted assessments of gas distribution and deposition 175–176, 94–104 (2017). https://doi.org/10.1016/j.jenvrad.2017.
rates. Future work in this study will aim to extend the current 04.016

123
Arabian Journal for Science and Engineering

8. Bogaersa, A.E.J.; Rensburg, G.J.J.: A comparison of different CFD 25. Liu, Z.; Li, W.; Shen, L.; Han, Y.; Zhu, Z.; Hua, X.: Numerical
and Gaussian dispersion models. In: Eleventh South African Con- study of stable stratification effects on the wind over simplified tall
ference on Computational and Applied Mechanics, Vanderbijlpark, building models using large-eddy simulations. Build. Environ. 193,
South Africa, (2018). 107625 (2021). https://doi.org/10.1016/j.buildenv.2021.107625
9. Blocken, B.: 50 years of computational wind engineering: past, 26. Blocken, B.; Stathopoulos, T.; Carmeliet, J.: CFD simulation of the
present and future. J. Wind Eng. Ind. Aerodyn. 129, 69–102 (2014). atmospheric boundary layer: wall function problems. Atmos. Env-
https://doi.org/10.1016/j.jweia.2014.03.008 iron. 41(2), 238–252 (2007). https://doi.org/10.1016/j.atmosenv.
10. Leelőssy, Á.; Lagzi, I.; Kovács, A.; Mészáros, R.: A review of 2006.08.019
numerical models to predict the atmospheric dispersion of radionu- 27. Tanaka, H.; Yoshie, R.; Hu, C.H.: Uncertainty in measurements
clides. J. Environ. Radioact. 182, 20–33 (2018). https://doi.org/10. of velocity and concentration around a building. 4th International
1016/j.jenvrad.2017.11.009 Symposium on Computational Wind Engineering, pp. 549–552,
11. Blocken, B.; Stathopoulos, T.: Evaluation of CFD for simulating (2006)
air pollutant dispersion around buildings. ASHRAE Trans. 116(2), 28. Keshavarzian, E.; Jin, R.; Dong, K.; Kwok, K.C.S.: Effect of
597 (2010) building cross-section shape on air pollutant dispersion around
12. Alnoamani, F.; Addad, Y.: Predictions of radioactive material buildings. Build. Environ. 197, 107861 (2021). https://doi.org/10.
releases in gaseous effluents under the UAE environmental con- 1016/j.buildenv.2021.107861
ditions. Porto Alegre, Brazil, (2015). 29. An, K.; Fung, J.C.H.: An improved SST k-ω model for pollu-
13. Tang, B.; Wang, H.; Xu, J.; Lin, J.; Hu, J.; Chen, R.: High-resolution tant dispersion simulations within an isothermal boundary layer.
simulation of the near-field pollutant dispersion in a nuclear power J. Wind Eng. Ind. Aerodyn. 179, 369–384 (2018). https://doi.org/
plant community with high-performance computing. J. Nonlinear 10.1016/j.jweia.2018.06.010
Math. Phys. 31(1), 6 (2024). https://doi.org/10.1007/s44198-024- 30. An, K.; Wong, S.M.; Fung, J.C.H.: Exploration of sustainable
00171-7 building morphologies for effective passive pollutant dispersion
14. Blocken, B.; Stathopoulos, T.; Saathoff, P.; Wang, X.: Numeri- within compact urban environments. Build. Environ. 148, 508–523
cal evaluation of pollutant dispersion in the built environment: (2019). https://doi.org/10.1016/j.buildenv.2018.11.030
comparisons between models and experiments. J. Wind Eng. Ind. 31. Willemsen, E.; Wisse, J.A.: Accuracy of assessment of wind
Aerodyn. 96(10–11), 1817–1831 (2008). https://doi.org/10.1016/ speed in the built environment. J. Wind Eng. Ind. Aero-
j.jweia.2008.02.049 dyn. 90(10), 1183–1190 (2002). https://doi.org/10.1016/S0167-
15. Lateb, M.; Masson, C.; Stathopoulos, T.; Bédard, C.: Comparison 6105(02)00231-3
of various types of k–ε models for pollutant emissions around a 32. Addad, Y.; Laurence, D.; Talotte, C.; Jacob, M.C.: Large-eddy
two-building configuration. J. Wind Eng. Indus. Aerodyn. Vol. 115, simulation of a forward-backward facing step for acoustic source
pp. 9–21, (2013). https://doi.org/10.1016/j.jweia.2013.01.001. identification. Int. J. Heat Fluid Flow 24(4), 562–571 (2003).
16. de Sampaio, P.A.B.; Junior, M.A.G.; Lapa, C.M.F.: A CFD https://doi.org/10.1016/S0142-727X(03)00050-X
approach to the atmospheric dispersion of radionuclides in the 33. Addad, Y.; Prosser, R.; Laurence, D.; Moreau, S.; Mendonca, F.:
vicinity of NPPs. Nucl. Eng. Des. 238(1), 250–273 (2008). https:// On the use of embedded meshes in the LES for external flows. J.
doi.org/10.1016/j.nucengdes.2007.05.009 Flow Turbulence Combust. 80, 393–408 (2007). https://doi.org/10.
17. Shih, T.H.; Liou, W.W.; Shabbir, A.; Yang, Z.; Zhu, J.: A new k- 1007/s10494-007-9131-1
eddy viscosity model for high Reynolds number turbulent flows. 34. Addad, Y.; Gaitonde, U.; Laurence, D.: Optimal unstructured mesh-
Comput. Fluids 24(3), 227–238 (1995). https://doi.org/10.1016/ ing for Large Eddy Simulations In: Meyers, J., Geurts, B.J., Sagaut,
0045-7930(94)00032-T P. (eds) Quality and Reliability of Large-Eddy Simulations. Ercof-
18. Menter, F.R.: Two-equation eddy-viscosity turbulence models tac Series, Vol. 12. Springer, Dordrecht, 2008. https://doi.org/10.
for engineering applications. AIAA J. 32(8), 1598–1605 (1994). 1007/978-1-4020-8578-9_8.
https://doi.org/10.2514/3.12149 35. Abida, R.; Addad, Y.; Francis, D.; Temimi, M.; Nelli, N.; Fonseca,
19. Lien, F.S.; Kalitzin, G.; Durbin, P.A.: RANS modeling for R.; Nesterov, O.; Bosc, E.: Evaluation of the performance of the
compressible and transitional flows. Center for Turbulence WRF model in a hyper-arid environment: a sensitivity study. Atmo-
Research—Proceedings of the Summer Program (1998). sphere 13(6), 985 (2022). https://doi.org/10.3390/atmos13060985
20. Billard, F.; Laurence, D.: A robust k−ε-ν 2 / k elliptic blending tur- 36. Richards, P.J.; Hoxey, R.P.: Appropriate boundary conditions for
bulence model applied to near-wall, separated and buoyant flows. computational wind engineering models using the k- turbulence
Int. J. Heat Fluid Flow 33(1), 45–58 (2012). https://doi.org/10. model. J. Wind Eng. Ind. Aerodyn. 46–47, 145–153 (1993). https://
1016/j.ijheatfluidflow.2011.11.003 doi.org/10.1016/0167-6105(93)90124-7
21. Lardeau, S.; Billard, F.: Development of an elliptic-blending lag 37. Wieringa, J.: Updating the Davenport roughness classification. J.
model for industrial applications. In: 54th AIAA Aerospace Sci- Wind Eng. Ind. Aerodyn. 41(1–3), 357–368 (1992). https://doi.org/
ences Meeting, Reston, Virginia: American Institute of Aeronautics 10.1016/0167-6105(92)90434-C
and Astronautics, (2016). https://doi.org/10.2514/6.2016-1600. 38. Terada, H.; Nagai, H.; Tsuduki, K.; Furuno, A.; Kadowaki, M.;
22. Architectural Institute of Japan (AIJ), Guidebook for prac- Kakefuda, T.: Refinement of source term and atmospheric disper-
tical applications of CFD to pedestrian wind environment sion simulations of radionuclides during the Fukushima Daiichi
around buildings. Architectural Institute of Japan. Accessed: Apr. nuclear power station accident. J. Environ. Radioact. 213, 106104
29, 2023. [Online]. Available: https://www.aij.or.jp/jpn/publish/ (2020). https://doi.org/10.1016/j.jenvrad.2019.106104
cfdguide/index_e.htm.
23. Simcenter STAR-CCM+ software, Accessed: October 8,
2023. [Online]. Available: https://plm.sw.siemens.com/en-
US/simcenter/fluids-thermal-smulation/star-ccm/.
24. Meng, Y.; Hibi, K.: Turbulent measurements of the flow field
around a high-rise building. Wind Eng. JAWE 1998(76), 55–64
(1998). https://doi.org/10.5359/jawe.1998.76_55

123