ed

Smart placement of depolluting panels in urban areas as

iew
regards of the air flow using adjoint framework and
district digital twin
T. Hamadaa , F. Chabia , R. Chakira , D. Lejrib , F. Dugayc , J. Waeytensa,∗

ev
a
Univ Gustave Eiffel, COSYS, , F-77420, Champs-sur-Marne, France
b
Univ Lyon, Univ Gustave Eiffel, ENTPE, LICIT-Eco7, , F-69675, Lyon, France
c
Airparif, F-75004, Paris, France

r
Abstract

er
Air pollution is a major worldwide health issue. Depolluting panels can be
deployed in urban areas to reduce pollutant human exposure from traffic.
pe
Using a district digital twin, we propose a two-step numerical strategy for
the smart placement of these panels. In the diagnosis stage, CFD is used to
get a detailed pollutant concentration map and to determine critical highly
ot

polluted areas. In the remediation stage, we aim to mitigate the air pol-
lution in these crucial areas by selecting relevant placement of depolluting
panels. In this way, we propose a spatial sensitivity indicator based on an
tn

adjoint framework. The strategy is applied to the full-scale laboratory dis-

trict “Sense-City” under controlled conditions and a district area in Paris
rin

using realistic NOx traffic emission and wind conditions. We note that NOx
concentration is high on some sidewalks, street intersections, and first-floor
building windows leading to outdoor/indoor pollutant transfer. Results show
ep

that a non-smart massive deployment of depolluting panels in urban areas

∗
corresponding author email: [email protected]
Pr

Preprint submitted to Building and Environment September 8, 2023

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
does not improve the air quality in the areas of interest. In the article, we give

iew
practical recommendations concerning depolluting panel placement. Lastly,
this strategy can provide a help-decision tool for city managers to design
depolluting panels-based mitigation actions.

ev
Keywords: outdoor air quality, traffic pollution, depolluting panels, CFD,
adjoint problem, digital twin, real-case study

r
1. Introduction

er
Air pollution is a major public health problem. According to the World
Health Organization, about 4.2 million deaths per year are related to poor
pe
air quality (World Health Organization, 2016). A wide variety of sectors,
e.g traffic, industry, agriculture, and housing, contribute to air pollution.
Urban cities produce around 78% of CO2 and atmospheric pollutants that
affect 50% of the population living in these areas (Bereitschaft and Debbage,
ot

2013). The main pollutants in outdoor air are NOx (NO2 +NO), VOC, SO2 ,
CO, O3 , and PM. Traffic is an important contributor to NOx emission in
tn

cities notably in Paris region. It has been shown that short or long-term
human exposure to nitrogen dioxide (NO2 ) causes respiratory and cardiovas-
cular diseases (Stafoggia et al., 2022; Meng et al., 2021). Through apartment
rin

openings, it can be noted that outdoor pollutants can also affect indoor air
quality (Leung, 2015). It is therefore important to (i) understand how these
pollutants disperse in the urban environment by making detailed cartogra-
ep

phies, (ii) identify the critical areas where pollution is high, and (iii) propose
an appropriate strategy to improve the air quality.
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
Concerning the monitoring of pollution, air quality stations are mainly
used to get accurate measurements but at a significant financial cost. The
number of air quality stations and their distribution on the territory comply
with European rules. These rules are essential to ensure quality and compa-

ev
rable measurement across regions and countries. In general, few air quality
stations are deployed in cities and regions. To counterbalance this sparse spa-
tial coverage of the instrumentation, physical models and numerical simula-

r
tions can be employed. Hence, data assimilation techniques combining sensor

er
outputs and physical models are often considered to get accurate air pollu-
tant cartography (Tilloy et al., 2013; Elbern and Schmidt, 2001; Le Dimet
and Talagrand, 1986). In practice, to make air pollution cartography at the
pe
region scale (1 − 3 km resolution) in Ile de France, Airparif agency uses the
chemistry-transport model CHIMERE (Bessagnet et al., 2008) where the dif-
ferent pollutant sources are taken into account notably for the traffic-related
ot

emissions through the model HEAVEN. At the district scale, to get a knowl-
edge of the pollution close to the roads, many cities like Paris, Roma, London,
tn

Johannesburg, and Beijing use the model ADMS-Urban (Carruthers et al.,

1994) based on Gaussian dispersion model. The Airparif agency combines
the ADMS-Urban model and the CHIMERE model to produce maps on the
rin

Ile de France region at a resolution ranging from 12.5m to 50m. These mod-
els enable numerical predictions at a reasonable computational cost however
they are not appropriate for detailed air pollutant cartography at a 1-meter
ep

resolution.
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
To precisely localize highly polluted areas and to design efficient urban

iew
planning using a virtual testing strategy with the view of locally reducing
air pollution, most sophisticated models such as “Computational Fluid Dy-
namics” can be considered. From geographic data and City Information
Modeling (CIM), the numerical mock-up of the district domain can be con-

ev
structed, providing access to the digital twin of the district. The computation
of the airflow pattern and scalar fields like pollutant concentration and tem-
perature can be valuable for a wide range of urban applications. CFD has

r
been used to get a better understanding of the urban airflow and the pol-

er
lutant dispersion (Ramponi et al., 2015; Tominaga and Stathopoulos, 2013;
Di Sabatino et al., 2013; Gousseau et al., 2011; Hang et al., 2011) where prac-
tical recommendations can be found in (Tominaga and Stathopoulos, 2016).
pe
A special attention has been given to street canyons (Koutsourakis et al.,
2012; Tominaga and Stathopoulos, 2011; Di Sabatino et al., 2008) where tur-
bulences induced by traffic (Zheng and Yang, 2021; Zhao et al., 2021; Solazzo
ot

et al., 2008) , the roof shapes (Yassin, 2011) and the vegetation (Salim et al.,
2011; Balczo et al., 2009; Gromke et al., 2008) were studied using CFD. An
tn

overview of CFD studies in street canyon applications is proposed in the

review article (Li et al., 2006). CFD was also previously used to evaluate dif-
ferent urban planning strategies through virtual testing. In (Gromke et al.,
rin

2015), CFD simulations were made at a district scale to quantify the cool-
ing effect, during a heat wave, of several revegetation scenarios such as tree
planting and wall and or roof vegetation. Moreover, the detailed description
ep

of the district airflow pattern using CFD allows to study the pedestrian wind
discomfort in urban areas (Kang et al., 2020; Blocken and Persoon, 2009) and
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
also the wind-driven natural ventilation efficiency for the cooling of buildings

iew
by considering the influence of balconies (Montazeri and Blocken, 2013), the
position and shape of openings (Peren et al., 2015; Kosutova et al., 2019) and
other building specificities. Furthermore, the CFD models are also involved
in inverse methods to identify the pollutant source characteristics (position

ev
and intensity) and wind condition (Mons et al., 2017, 2019), and in adjoint
sensitivity analysis method to determine the optimal number and placement
of gas sensors (Waeytens and Sadr, 2018). To predict the turbulent flows

r
encountered in urban areas, two categories of CFD turbulence models are

er
commonly employed: Steady or unsteady Reynolds Averaged Navier-Stokes
(RANS or URANS, see (Rodriguez, 2019)) and Large Eddy Simulation (LES,
see (Sagaut, 2006)). Despite their lack of accuracy compared to LES model
pe
(Gousseau et al., 2011), RANS models remain largely used in operational
studies due to their lower computational cost (Blocken, 2015). In the present
article, RANS models are to be considered to get a detailed NOx pollution
ot

map at the street and the district scales.

tn

To reduce human exposure to pollutants in critical areas, actions have to

be carried out. The different types of mitigation strategies can be categorized
based on the partial differential equation (1) representing transport-diffusion-
rin

reaction phenomena of pollutants. Herein, we mainly focus on traffic pollu-

tion.
∂C
+ ⃗v · ∇C − div(D∇C) + R(C) = |{z}
S (1)
∂t
ep

| {z } | {z }
II III I

where C is the pollutant concentration, ⃗v and D are the flow velocity and
the diffusion parameter, R(C) represents chemical reaction of species and S
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
is associated to source emissions.

iew
As shown in Eq (1), obviously the pollutant concentration C can be decreased
by reducing the source emission S. This strategy “I” is often employed by mu-
nicipalities and regional authorities via traffic restrictions like Low-Emission
Zone (LEZ) and the replacement of a portion of the oldest vehicles. A good

ev
knowledge of the local fleet and the virtual testing of different scenarios via a
simulation platform are recommended to efficiently design the Low-Emission
Zones (Andre et al., 2020; André et al., 2018; Duque et al., 2016).

r
The strategy “II” consists in modifying the pollutant dispersion by acting

er
on the urban airflow. A first way is to change the urban morphology. In-
deed, in the literature numerous researches have shown the effects of the
urban morphology at the building, the street and the district scales on the
pe
air quality (da Silva et al., 2022; Yang et al., 2020; An et al., 2019; Shen
et al., 2017). At the district scale, the mitigation of the air pollution through
action planning on urban morphology can be performed in new districts or
ot

in local urban renewal, e.g demolition of old buildings in a district to cre-

ate wind corridors leading to the improvement of air quality and thermal
tn

comfort (Peng et al., 2015). These actions can be long-term and expensive.
At the street scale, barriers can be installed just next to traffic roads in or-
der to reduce the air pollution on the sidewalk especially in street canyon
rin

(Issakhov and Omarova, 2021; Gromke et al., 2016; Vos et al., 2013; Ha-
gler et al., 2011). The studies show that barriers must be continuous, high
(more than 2 meters height) and impermeable, e.g solid walls or green bar-
ep

riers with very low porosity/permeability. As for nature-based solutions,

vegetation can contribute to decreasing heat island phenomena, noise reduc-
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
tion, preservation of biodiversity and the contribution of human well-being.

iew
Nevertheless, contrary to popular belief, the diminution of air pollution by
addition of vegetation in urban areas is not so evident (Vos et al., 2013). On
the one hand, the pollutant concentrations can be reduced via the deposition
on leaves. On the other hand, the presence of trees can modify the pollu-

ev
tant dispersion by obstructing the urban air flow. Studies point out that
trees can increase traffic concentrations on the sidewalk in street canyons
but show a beneficial decrease of concentrations in open terrain configura-

r
tions, which are mainly found outside city center (Jeanjean et al., 2015). In

er
summary, trees in streets and more generally in urban parks have a positive
global impact on city air pollution but they can be locally deteriorated (Xing
and Brimblecombe, 2019; Selmi et al., 2016). At the street scale, the urban
pe
airflow can also be modified using passive or active ventilation strategy. In
(Mirzaei and Haghighat, 2010), the authors proposed a pedestrian ventilation
system in order to enhance the air quality and thermal comfort in the pedes-
ot

trian walking area of street canyons. The controlled air movement is ensured
by a ventilation mechanism guiding polluted air from the near-surface level
tn

through a designed vertical duct system to the surrounding street level. The
updraft movement of air can also be achieved using a passive solar strategy.
Hence, in (Daghistani, 2021) the author developed a solar chimney street-
rin

lighting pole where the inner part of the lamp structure is employed for the
air pollutant removal.
Lastly, the air quality improvement strategy “III” is based on the degra-
ep

dation of pollutant species via a chemical reaction mechanism. In recent

decades, the depollution structures made of the integration of photocatalyst
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
semiconductor oxides such as titanium dioxide (TiO2 ) and zinc oxide (ZnO)

iew
in construction materials or as direct use on a surface layer, have been used
to remove various pollutants (O3, COx, NOx, VOCs) (Le Pivert et al., 2021;
Darvish et al., 2020; Le Pivert et al., 2020; Binas et al., 2017). The func-
tioning of these panels relies on the photocatalysis mechanism. Thus, with

ev
sufficient light and under the presence of photocatalyst (ZnO or TiO2 ), a
catalytic oxidation effect is triggered, degrading certain polluting particles
(He et al., 2017; Garrido et al., 2019). Also, despite that TiO2 photocat-

r
alyst in the form of nanoparticles is the most used for the production of

er
depolluting surfaces, recent studies are moving towards the use of other pho-
tocatalysts such as ZnO given the harmful effect of TiO2 on health and the
environment (Grande and Tucci, 2016; Wu and Ren, 2020). For these reasons
pe
(Le Pivert et al., 2021) developed depolluting panels, which consist of pho-
tocatalyst nanostructures ZnO grown on construction materials (tiling, rock
aggregates), to test their efficiency to reduce air pollution from road traffic.
ot

These first designed ZnO panels have a surface area of about 1 m2 . The
authors have shown that at the district scale, in the presence of ZnO depol-
tn

luting panels on the road surfaces, the concentrations of pollutants emitted

by road traffic diminish. To conclude this paragraph, let us note that a simi-
lar classification of local mitigation strategies can also be found in the review
rin

article (Li et al., 2021).

Given the very encouraging results of the depolluting panels in reducing ur-
ban pollution, it can be promising to deploy them in urban areas. They can
ep

be placed as paving demountable blocks on road surfaces, building walls, mo-

torway sound barriers, etc. To concretely evaluate the potential of pollutant
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
removal by photocatalytic coatings in a street canyon, the authors in (Pul-

iew
virenti et al., 2020) conducted CFD simulations and experimental validation
in Bologna, Italy. To go one step further, we propose herein to optimize
the placement of depolluting panels through a CFD-based strategy in order
to locally improve the air quality in critical urban areas with a restricted

ev
surface of panels. The article is organized as follows. After this introduction
section, the computer-aided strategy for the smart placement of depollut-
ing panels is presented in Section 2. Firstly, in the diagnosis step, CFD is

r
used at the district scale to identify critical highly polluted areas. Then, in a

er
remediation step, a virtual testing strategy based on CFD and adjoint frame-
work is developed to determine a relevant placement of depolluting panels for
goal-oriented air pollutant mitigation. In Section 3, the numerical method
pe
is illustrated with a controlled laboratory district named “Sense-City”. The
smart placement strategy is lastly applied to a Paris district under in-situ
meteorological and traffic conditions in Section 4.
ot
tn

2. Numerical method for the optimal placement of depolluting sys-

tem as regards of urban air flow

2.1. Outline
rin

In this section we give an overview of the virtual testing strategy to de-

termine a smart placement of depolluting panels at the district scale. Our
approach relies on the formulation of an optimization problem and a sensitiv-
ep

ity analysis through the adjoint framework. The main steps of the proposed
method are summarized below.
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
Diagnosis stage:
From detailed pollutant cartographies at the district scale, the goal is to
identify high polluted areas.

• determine the air flow at the district scale using computational fluid

ev
dynamics;

• solve the direct advection-diffusion-reaction model to reconstruct pre-

r
cisely the pollution map at the district scale;

er
• identify from the pollution map the critical areas where people can be
exposed to high level of pollutant. The pollutant concentration in these
pe
critical areas are defined as quantities of interest.

Remediation stage:
The objective is to improve air quality in the determined critical areas by
ot

decreasing the quantity of interest. In the proposed method, local sensitivity

analysis with respect to the selected quantity of interest is performed con-
tn

sidering both direct and adjoint problems in the actual urban configuration
(without depolluting panels).

• solve the adjoint advection-diffusion-reaction problem associated to the

rin

chosen quantities of interest;

• evaluate the spatial sensitivity indicator using the numerical concen-

ep

tration of the direct problem determined in the diagnosis stage and the
numerical solution of the adjoint problem;
Pr

10

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
• determine a relevant position of depolluting panels given by significant

iew
value of the sensitivity indicator.

Each of these steps are detailed in the following sections.

2.2. Technical description

ev
2.2.1. Diagnosis: district pollutant map using direct model
We denote by Ω the computational domain of the district. After simu-

r
lating the airflow ⃗v over Ω, the detailed cartography of the pollutant con-
centration C (x, t) is obtained by solving the convection-diffusion-reaction
equation:
er
pe

∂C

 + ⃗v · ∇C − div(D∇C) + R(C) = S in Ω × [0, T ]
∂t





C = Ci on Γi × [0, T ]

(2)
∇C · ⃗n = 0 on Γo ∪ Γw × [0, T ]




ot





C (t = 0) = C in Ω
0

where Ci and C0 denotes respectively the background concentration enter-

tn

ing the domain Ω through the inlets Γi and the initial concentration in the
domain. The boundaries Γo and Γw are associated to outlet and wall surfaces.
rin

In the urban air quality application, the airflow is considered newtonian,

incompressible and turbulent. In practice, the airflow velocity ⃗v and the tur-
bulent viscosity µt (kg/(m.s)) are obtained using Reynolds Averaged Navier-
ep

Stokes simulations (RANS). The diffusion term D involved in Eq. (2) is the
sum of the molecular diffusion and the turbulent diffusion Dt . The turbulent
Pr

11

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
diffusion Dt (m2 /s) is given by the formula

iew
µt
Dt = , (3)
ρ Sct

where ρ (kg/m3 ) is the density of air and Sct is the turbulent Schmidt
number. Lastly, R(C) and S in Eq.(2) are the reaction and the source terms.

ev
The source term S in the domain Ω is used to define the position and the
amplitude of source emissions.

r
In the applications to be presented in Sections 3 and 4, the pollutant

er
cartography of the pollutant is obtained from the deterministic solution of
the direct problem. To better take into account the uncertain parameters
pe
of the physical model like boundary air flow conditions and term source
characteristics, uncertainty quantification’s (UQ) and data assimilation can
be considered (see,(Hammond et al., 2019; Mons et al., 2017; Tilloy et al.,
2013)).
ot

From the detailed pollutant concentration map C(x, t), the critical areas
tn

where people are exposed to high pollution levels are determined. Hence, the
concentrations in these areas are defined as quantities of interest J .
Z TZ
rin

J = fq (x)ξ(t)C(x, t)dΩdt (4)

0 Ω

where fq (x) is the space function to extract the pollution in a critical area
Ωq and ξ(t) is the time extractor function to define the period of interest.
ep

In our case, we choose the pollutant concentration at the location where we

want to improve the air quality as a quantity of interest. Thereby, herein Ωq
Pr

12

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
is a subdomain of Ω representing the location where we want to improve the

iew
air quality.

2.3. Remediation: Smart placement of depolluting panels solving minimiza-

tion problem

ev
To model the effects of depolluting panels on the district air quality, we
consider in the optimization problem the advection-diffusion-reaction equa-
tion: 

r
∂C
+ ⃗v · ∇C − div(D∇C) = S in Ω × [0, T ]





 ∂t










C = Ci on Γi × [0, T ]
er
∇C · ⃗n = 0 on Γo ∪ Γn × [0, T ] (5)
pe

−D∇C · ⃗n = kC on Γp × [0, T ]







C (t = 0) = C0 in Ω


To remain simple, the degradation of the pollutant by the photocatalytic

depolluting system is described using a first-order-reaction where k (m/s)
ot

denotes the reaction rate function. The reaction rate is an approximation of

Langmuir-Hinshelwood kinetic model and was used in previous works (Pul-
tn

virenti et al., 2020; Yusuf et al., 2020; Yusuf and Palmisano, 2021). The
degradation reaction occuring at the surfaces of the depolluting panels, it is
considered as a boundary condition (Yusuf et al., 2020; Yusuf and Palmisano,
rin

2021). For operational reasons, the depolluting panels can not be placed on
any surfaces of the district. Consequently, we limit the potential placement
ep

of depolluting panels on the boundary Γp which is a restricted part of the wall

boundary Γw . The boundaries Γp and Γn are defined such that Γw = Γp ∪ Γn .
Lastly, let us underline that k(x) is a scalar field defined on the boundary Γp .
Pr

13

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
The constrained minimization problem to be solved is:

min J (C, k) (6)

k,C∈C

where C ∈ C, imply that the concentration C has to satisfies the direct prob-

ev
lem equation (5).

The constrained minimization problem (6) is rewritten using the La-

r
grangian L

L(C, k; λ) = J (C, k) −
0
Z TZ
Ω
λ1
∂C
∂t
Z T Z 
er 
+ ⃗v.∇C − div(D∇C) − S dΩ dt
Z TZ
pe
− λ2 (C − Ci )dS dt − λ3 ∇C · ⃗n dS dt
0 Γi 0 Γo ∪Γn
Z T Z Z
+ λ4 (D∇C · ⃗n + kC)dS dt − λ5 (C(t = 0) − C0 )dΩ
0 Γp Ω
(7)
ot

where λi , i ∈ {1, .., 5} are Lagrange multipliers.

tn

The derivative of the Lagrangian L according to:

• the Lagrange multipliers λi , i.e. ∂L/∂λi = 0, gives as expected the

“direct problem” where the concentration C has to satisfy Eq. (5);
rin

• the concentration C, i.e. ∂L/∂C = 0, leads to the formulation of the

ep
Pr

14

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
“adjoint problem” whose the adjoint concentration noted C̃ verifies

iew

 ∂ C̃

 − − ⃗v · ∇C̃ − div(D∇C̃) = fq ξ in Ω × [0, T ]
∂t





C̃ = 0 on Γi × [0, T ]






∇C̃ · ⃗n = 0 on Γn × [0, T ]


ev
(8)
−D∇C̃ · ⃗n = k C̃ on Γp × [0, T ]











 D∇C̃ · ⃗n + (⃗v · ⃗n) C̃ = 0 on Γo × [0, T ]

r



C̃(t = T ) = 0 in Ω.


er
The main steps of the calculations to obtained the adjoint problem and the
relation between the Lagrange multipliers λi and the adjoint field C̃ are given
pe
in Appendix A. We notably show that the adjoint field C̃ corresponds to
the Lagrange multiplier λ1 , i.e. C̃ = λ1 . The adjoint problem corresponds
to a backward advection-diffusion equation which involves a fictitious source
emission located in the area of interest Ωq . Even in stationary conditions,
ot

the adjoint problem is different from the direct problem (5).

The adjoint concentration field C̃ corresponds to a sensitivity function as-
tn

sociated with the considered quantity of interest J . Hence, if the adjoint

concentration is null in a part of the domain, it physically means that what
happens in this part of the domain will not affect the chosen quantity of
rin

interest.
This type of adjoint model was previously used in by the authors to study
the optimal placement of air quality sensors (Waeytens and Sadr, 2018).
ep

Lastly, the sensitivity of the quantity of interest (QoI) to the reaction

Pr

15

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
rate scalar field k to the depolluting panels can be obtained by

iew
Z T Z T
∂J ∂L
= = λ4 Cdt = − C̃Cdt on Γp (9)
∂k ∂k 0 0

As shown in Appendix A, one has λ4 = −C̃ on Γp × [0, T ].

To evaluate the effectiveness of depolluting panels placed at a position x, we

ev
define the positive sensitivity indicator:
Z T
∂J
I(x) = = C̃(x, t)C(x, t)dt (10)
∂k

r
0

Let us note that the indicator I(x) is significant when both direct concentra-

er
tion C and the adjoint concentration C̃ have high values. The depolluting
panels must therefore be placed in high-polluted zones (C(x, t) high) which
pe
can impact the chosen quantity of interest (C̃ high) due to the transport and
the dispersion of the pollutant.
For an operational purpose and an objective of limiting computational cost,
we only perform a local sensitivity analysis by solving the direct problem
ot

(5) and the adjoint problem (8) in the pre-existing configuration, i.e. no
depolluting panels (k = 0). Then, the direct and adjoint numerical solutions
tn

are used to evaluate the sensitivity indicator (10) on all the surfaces of Γp .
Finally, the depolluting panels are to be placed only on surfaces where the
indicator I(x) is significant.
rin

In the following, the spatial sensitivity indicator I(x) will be used to select
smart placement of depolluting panels for two urban area applications. The
ep

sensitivity indicator is computed on all the wall surfaces of the domain, i.e.
Γp = Γw and Γn = ∅.
Pr

16

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
3. Application in controlled conditions: the Sense-City district

iew
To illustrate the numerical strategy for the smart placement of depolluting
panels, we first consider a controlled scenario in a full-scale small district em-
bedded in the climatic chamber of the equipment “Sense-City”. In this part,
the airflow within the climatic chamber is considered steady and turbulent.

ev
The numerical study concerns the transient dispersion of a non-reactive gas
pollutant emitted at a given position on the road of the Sense-City district

r
within a period of 180 seconds1 . In what follow, we present the detailed de-
scription of the controlled scenario in Sense-City district and the application
of depolluting panels strategic position.
er
3.1. Presentation of Sense-City equipment and the studied district under con-
pe
trolled conditions

Sense-City is an instrumented district build in 2018 and located on the

campus of Université Gustave Eiffel at Champs-sur-Marne, France (Derkx
ot

et al., 2012). This district is designed for the validation of numerical physical
models and innovative technologies used in the field of urban environment. It
tn

is composed of two small district of 400 m2 . These urban areas can be studied
in natural conditions or in controlled climatic conditions using a large mobile
climatic chamber of dimension (20 m × 20 m × 10 m). In addition, more than
rin

150 sensors (temperature sensors, gas sensors, etc..) are installed at different
locations in the district.
ep

1
Due to the closed-cycle air circulation within the climatic chamber (Streichenberger
et al., 2021), we had to limit the simulation period to 180 seconds.
Pr

17

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
In the present article, the small district #1 covered by climatic chamber

iew
is used for our numerical study (see Figure 1). It consist in two small houses,
a two story precast concrete building and a street layout. Figure 1 also
shows the digital twin of this district. On the south and north walls of the
climatic chamber are installed six fans, two horizontal air-forced ducts and

ev
two vertical rectangular columns for air extraction called “extractor”. The
boundary surfaces of the computational domain are: inlet fans (surfaces with
red and blue colors), outlet fans (surfaces with green color), horizontal air-

r
forced ducts (surfaces with black and purple colors), extractors (surfaces with

er
yellow) and walls (buildings and climatic chamber walls). For more details
about the airflow characteristics, turbulent air flow simulation using RANS
model and the its validation using 3D anemometer measurement, the reader
pe
can refer to (Streichenberger et al., 2021).

zoom

zoom
ot

N
Z W
X E
tn

Y S

FANS COLOR LEGEND SUCTION AND FORCED AIR DUCTS COLOR LEGEND
Inflow surfaces (blowing sides of fans with reduced airflow) Outflow surfaces (extraction surfaces of suction ducts)
Inflow surfaces (blowing sides of regular fans) Inflow surfaces (perforation on upper forced-air ducts )
Outflow surfaces (extraction sides of all fans) Inflow surfaces (perforation on lower forced-air ducts)
rin

Figure 1: Sense-City district #1 covered by the climatic chamber (at left) and its associated
digital twin (at right).

.
ep
Pr

18

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
3.2. Time evolution cartography of the pollutant dispersion in a controlled

iew
scenario without depolluting system
In this first application, the time-evolution of the concentration in Sense-
City district in controlled conditions on the time interval [0, 180s] is studied.
Concerning the Sense-City numerical mock-up, let us note that the center

ev
of the Cartesian coordinate system, represented in Fig. 2, is placed on the
ground surface at the center of the district such that the district domain is
described by {x ∈ [−11.22m, 11.22m], y ∈ [−11.1m, 11.1m], z ∈ [0m, 10m]}.

r
Herein, we consider the scenario of a continuous release of a pollutant tracer,

er
i.e. non-reactive, from a parallelepiped volumic source (0.5 × 0.5 × 0.5 =
0.125 m3 ). The center of the source is located in the middle of the road at
pe
the coordinates (x = 0 m, y = −2.5 m, z = 1 m) between the concrete building
and the two houses (see Fig. 2) at height z = 1 m. The parallelepiped volume
Ωs of the source is expressed by the space function fs :

1 for x ∈ Ω
ot

s
fs (x) = . (11)
0 elsewhere

The source amplitude is taken to 2000 µg/m3 /s. The volume of the source
tn

being of 0.125 m3 , the considered emission is 250 µg/s. Let us note that it
corresponds to the order of magnitude of NOx mean emission at 20km/h for
rin

EURO 6 passengers cars with small petrol engine, i.e 0.039g/km emission
factor that gives a value of 217 µg/s (Ntziachristos et al., 2009).
ep

In the absence of depolluting system, to determine the time-evolution

cartography of the pollutant, we solve the direct advection-diffusion problem
defined in Eq. (2). Since the pollutant concentration C(x, t) is considered
Pr

19

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Mast 10 Mast 5

iew
Mast 6
Mast 1 Building window

ev
Y
Mast 3
Z X

r
Mast 9 Source
Mast 2

er Mast 4
pe
Mast 8

Mast 7

Figure 2: Position of the pollutant source and the mast sensors in Sense-City district.
ot

.
tn

herein as a passive scalar, i.e. non-reactive pollutant such as in the articles

(Tee et al., 2020; Tominaga and Stathopoulos, 2018), the reaction term in
Eq. (2) is taken to 0.
rin

The time-averaged turbulent velocity field ⃗v and the turbulent viscosity µt

are obtained from a previous work (Streichenberger et al., 2021), where un-
ep

steady Reynolds Averaged Navier-Stokes (URANS) turbulent model within

software Code Saturne was used to simulate airflow in Sense-City district.
The turbulent diffusion is Dt = µt /(ρSc ) and the Schmidt number Sc is set
Pr

20

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
to 0.7 (Tominaga and Stathopoulos, 2007; Wang and McNamara, 2006). We

iew
consider null initial concentration, homogeneous Dirichlet conditions (i.e.
C0 = Ci = 0) at the inlets (inlet fans, horizontal air-forced ducts) and ho-
mogeneous Neumann condition (i.e. ∇C · ⃗n = 0) on the walls and outlets.

ev
In this application, we use the Finite Element Method with SUPG sta-
bilization in the software FreeFem++ (Hecht, 2012) to solve the direct gas
dispersion model Eq. (2). The variational formulation of the problem reads:

r
Z Z Z
∂C ∗
C dΩ + (⃗v · ∇C)C dΩ + D(∇C · ∇C ∗ ) dΩ
∗
Ω ∂t
−
ZΩ

∂Ω
∗
D(∇C · ⃗n)C dΩ + αer
Ω
Z

Ω
∂C
∂t
(⃗v · ∇C ∗ ) dΩ
pe
Z Z
+ α (⃗v · ∇C)(⃗v · ∇C ) dΩ − α D∆C(⃗v · ∇C ∗ ) dΩ
∗

Z Ω Z Ω
∗ ∗
= f C dΩ + α f (⃗v · ∇C ) dΩ (12)
Ω Ω

where α represents the stabilization term defined by SUPG method (Hughes

ot

et al., 1986; Franca et al., 1992). Euler implicit scheme is used for the time
discretization. The time step is taken to ∆t = 0.1 s.
tn

An unstructured mesh with a local volume refinement around the source

emission was build using the open-source software SALOME with the auto-
matic mesh generator NETGEN. The mesh has a total cell number of about
rin

3,947,186.
ep

In Fig. 3, the spatial maps of pollution in the Sense-City district are

plotted at pedestrian level (z = 1.5 m) for different times (t = 5 s ; t =
30 s; t = 60 s). The vertical profile of pollutant concentration along the
Pr

21

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
south-oriented building wall which faces the road is also represented. Since

iew
the concentration is almost unchanged after 60 seconds, only the pollutant
maps from t = 0 s to t = 60 s are given. From Fig. 3, it can be noted
that the pollutant goes from the source to smash on the concrete building
and then bypasses it on one side and goes towards the extractor of Sense-

ev
City ventilation system, which is located in the lower right corner in Fig. 3.
Moreover, a non-negligible part of the pollution, i.e. about 20 µg/m3 as
shown in Fig. 4 at Mast 6, also reaches the window placed at the first floor

r
of the building depicted by a white rectangular in Fig. 3.

er
To emphasize the non-homogeneous nature of the concentration level in the
district, we have also extracted from the CFD numerical simulations the
local pollutant concentration in several mast positions defined in Fig. 2. The
pe
time evolution of the concentration at the masts is given in Fig. 4. As the
concentration was negligible at Masts 7, 8, 9 and 10, there are not represented
in Fig. 4. Due to urban airflow, the pollutant concentration can be very
ot

different from one side of the road to the other. Indeed, Fig. 4 shows that
concentration at Mast 2 (on the building sidewalk) is 20 times higher than
tn

the one at Mast 1 (near the two houses). Even on the building sidewalk, the
pollutant concentration varies significantly, i.e. from 10 µg/m3 to 60 µg/m3
according to the spatial position (see Masts 2, 3 and 4).
rin

To sum up, high concentrations are observed obviously at the source po-
sition, at the vicinity of the building on the sidewalk and at the first floor
building window.
ep
Pr

22

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
ev
t = 5s t = 5s

r
er
pe

t = 30s t = 30s
ot
tn
rin
ep

t = 60s t = 60s

Figure 3: Spatial pollution maps (µg/m3 ) for Sc = 0.7 as a function of time, in the x-y
plane at z = 1.5m (left) and in the x-z plane at y = −5.4m (right). The white rectangular
Pr

contour the represent the building and the window of the first floor.
23

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
60

iew
Mast 1
50 Mast 2
Mast 3
Concentration ( g/m 3)

Mast 4
40
Mast 5
Mast 6

ev
30

20

r
10

0
0 20 40 60 er
80
Time (s)
100 120 140 160 180
pe
Figure 4: Pollutant concentration from simulation at the masts of Sense-City district as a
function of time - no depolluting panels.

.
ot

3.3. Determination of high-polluted areas and definition of quantities of in-

terest
tn

From the pollution maps in Sense-City (Fig. 3), two critical pollution
areas are selected :
rin

• The first area is located on the sidewalk near the concrete building and
is defined by the subdomain
ep

Ωq1 = {x ∈ [−5m, 5m], y ∈ [−5.5m, −4m], z ∈ [1m, 2m]}

• The second area corresponds to the vicinity of the first floor building
Pr

24

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
window and is defined by the subdomain

iew
Ωq2 = {x ∈ [1.5m, 2.5m], y ∈ [−5.5m, −5.3m], z ∈ [3.7m, 5m]}

For each high-polluted area, we define a quantity of interest Ji , i ∈ {1, 2}

ev
(see Eq. (4)) associated to the spatial-averaged concentration over the whole
time interval [0, 180s]. Hence, the space function fqi (x), i ∈ {1, 2} involved
in Eq. (4) to extract the averaged concentration in the critical areas Ωqi , i ∈

r
{1, 2} is given by 
1/|Ω | for x ∈ Ω
qi qi
fqi (x) =
er
0 elsewhere
.

and the time function for both quantities of interest is ξ(t) = 1/T on the
(13)
pe
whole time interval [0, T = 180s].

In the configuration without depolluting panels, the numerical estimation

of the two quantities of interest are:
ot


J1 = 33.0 µg/m3

(14)
tn

J2 = 17.9 µg/m3



3.4. Smart placement of depolluting panels to improve air quality in high-

polluted areas
rin

After defining the quantities of interest, the adjoint problem (see Eq. (8))
needs to be solved. Fig. 5 shows the adjoint concentration on the surfaces of
ep

the domain. As illustrated in Fig. 5 for the quantity of interest J2 (i.e the
mean value of the pollutant concentration at the building window), the ad-
joint problem corresponds to a backward in time advection-diffusion problem
Pr

25

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
where a virtual source is emitted in the quantity of interest’s area. Contrary

iew
to the direct problem defined in Eq. (5), the adjoint problem has a final
condition in time. Let us note that the adjoint solution corresponds to a
sensitivity function as regards of the chosen quantity of interest. Thus, if the
adjoint solution is almost null in a part of the domain, it means that actions

ev
in this part of the domain, like emissions of pollutant or placement of de-
polluting panels, will have no influence on the considered quantity of interest.

r
To smartly place depolluting panels for improving the chosen quantities

er
of interest, the spatial sensitivity indicator defined in Eq. (10) is computed
on all the wall surfaces of the domain (Γp = Γw ). It is obtained from the
product of the pollutant concentration field (direct concentration) and the
pe
adjoint concentration. Hence, the depolluting panels are to be positioned in
areas where both direct and adjoint concentrations are high. In Fig. 6, the
sensitivity indicator maps are shown for the two selected quantities of inter-
ot

est: mean pollutant concentration values on the building sidewalk (J1 ) and
at the building window (J2 ). The maximum value of the sensitivity indicator
tn

I1 is about 25 whereas the value of I2 can exceed 50. It means that placing
a given surface in m2 of depolluting panels in the most sensitive areas can
have more impact for reducing the pollutant concentration at the building
rin

window (J2 ) than the one at the building sidewalk (J1 ). From Fig. 6, we
remark that depolluting panels have to be positioned on the bottom of the
building surface, and on a restricted part of the sidewalk and on the road to
ep

reduce the quantity of interest J1 (sidewalk pollutant concentration). Con-

cerning the pollutant concentration at the building window J2 , air pollution
Pr

26

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
r ev
er
pe
ot

Figure 5: Adjoint solution represented at the surfaces of Sense-City district - Quantity of

tn

interest J2 associated to the pollutant concentration at the vicinity of the window building.

.
rin

exposures can be decreased by placing panels below the window on the build-
ing facade, and on a portion of the sidewalk and on the road. We recall that
ep

the placement of depolluting panels is determined thanks to the knowledge

of the pollutant dispersion via the airflow.
Pr

27

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
ev
Figure 6: Sensitivity indicator maps I1 associated to the pollutant concentration at the
building sidewalk J1 (at left) and I2 associated to the pollutant concentration at the

r
building window J2 .

er
In conclusion, we propose to smartly placed depolluting panels to de-
pe
crease the pollutant concentrations for both quantities of interest J1 and J2
using the spatial sensitivity indicators I1 and I2 . In Fig. 7, the two sensitiv-
ity indicators are displayed on the most influential surfaces, i.e. the building
walls, the building sidewalk, and the road. Three depolluting panel areas,
ot

noted DP1, DP2, and DP3, are determined by taking the sensitivity thresh-
old value to 5. The lowest is the sensitivity threshold value, highest is the
tn

deployment surface of the panels. In practice, the threshold value can be

chosen to get a deployment surface of panels strictly less than the maximum
deployment surface allowed by the urban planner. In the Sense-City appli-
rin

cation, the total surface of depolluting panels corresponds to 29.7m2 . Let us

give more details about the three optimal depolluting areas represented in
Fig. 7:
ep

• “DP1” is located on the building wall above the window. It is defined by

{x ∈ [−0.3m, 2.7m], y = −5.5m, z ∈ [2m, 3.7m]}. As shown in Fig. 6,
Pr

28

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
panels in “DP1” allow mainly to improve the air quality at the building

iew
window and thus to reduce outdoor/indoor transfer of pollutant;

• “DP2” is located on the bottom building wall. It is given by {x ∈

[−3.4m, 4m], y = −5.5m, z ∈ [0m, 2m]}. Panels in this area contribute
to improve both quantities of interest;

ev
• “DP3” is placed on a part of the sidewalk and the road, and it is
defined by {x ∈ [−1.9m, 1.3m], y ∈ [−2.5m, −5.5m], z = 0m}. As for

r
“DP2”, panels in “DP3” can reduce the pollutant concentrations on
the building sidewalk and window.
er
For operational urban purposes, we preferred rectangular areas to more com-
pe
plex shapes.
ot

DP3

DP1
tn

DP2
rin

Figure 7: Smart placement of depolluting panels, indicated by black rectangles, in Sense-

City district on the road and the sidewalk (at left) and on the building facade (at right)
to improve the quantities of interest J1 and J2 - White rectangular denotes the window
ep

position.

.
Pr

29

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
3.5. Numerical validation of the optimized placement of depolluting panels

iew
In Section 3.4, we determined smart placement of depolluting panels to
improve both quantities of interest J1 and J2 using the spatial sensitivity in-
dicator defined in Eq. (10). To validate numerically the relevance of the pre-
dicted position of panels, the pollutant concentration in Sense-City district

ev
and the quantities of interest are evaluated by solving the advection-diffusion
direct problem (see Eq. (5)) in three different configurations:

r
• “no depollution case” - it corresponds to the initial district urban plan-
ning, i.e. in the absence of depolluting device. It was detailed in
Sections 3.2 and 3.3;
er
• “optimal depollution placement” - the depolluting panels DP1, DP2
pe
and DP3, determined from the smart placement strategy in Section
3.4, are integrated in the Sense-City district;

• “bad depollution placement” - depolluting panels are massively de-

ot

ployed on all the district roads (see Fig. 2) except in the sensitive area
DP3.
tn

For these three urban planning configurations, we compute the quanti-

ties of interest J1 (mean pollutant concentration at the building sidewalk)
rin

and J2 (mean concentration at the building window). The numerical re-

sults are summarized in Table 1. In the case of “optimal depolluting panels
placement”, different values of the reaction rate k are investigated for all the
ep

panels DP1, DP2, and DP3. If the reaction rate k of the depolluting panels
is less or equal to 0.01m/s, the reduction of the pollutant concentration in
the areas of interest is not significant, i.e. less than 2 µg/m3 . As predicted
Pr

30

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
from the sensitivity indicator in Fig. 6, the smart placement of depollut-

iew
ing panels may have more impact on the improvement of the air quality at
the building window (J2 ) than the one on the building sidewalk (J1 ). Ta-
ble 1 confirms this expectation. Indeed, when considering depolluting panels
(DP1,DP2,DP3) with k = 1m/s, the quantity of interest J2 decreases by

ev
about 10 µg/m3 (−59% variation) whereas the decrease is less than 5 µg/m3
(−14% variation) for the quantity of interest J1 . According to the study of
the airflow in the Sense-City district (Streichenberger et al., 2021), the pol-

r
lutant may pass closer to the district surfaces (road, sidewalk, and building

er
surfaces) for the trajectory between the source and the building window than
the one between the source and the sidewalk. In summary, depolluting panel
strategy will not be efficient if the pollutant does not pass close to district
pe
surfaces.

In the last configuration noted as “bad depollution placement”, the pan-

ot

els are largely deployed on all the road surfaces of the Sense-City district
except the area DP3. Hence, it represents 172m2 of depolluting panels over
tn

a total ground surface of 328m2 . Even for a high value of depolluting panel
reaction rate k, i.e. k = 1m/s, Table 1 shows that a massive deployment of
panels can result in no improvement of the air quality in the areas of interest
rin

when the panels are badly positioned. In conclusion, the use of district air
flow in numerical strategies can allow the design of efficient urban planning
to improve the air quality in critical areas with a reduced surface area of
ep

depolluting panels.
Pr

31

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Configuration
J1 (µg/m3 ) J2 (µg/m3 )

iew
Panel placement k(m/s)

“No depolluting panel” - 33.0 17.9

“Optimal depolluting 0.01 32.5 16.6
panel placement” 0.1 30.4 11.4

ev
1 28.3 7.3
“Bad depolluting placement” 1 33.0 17.9

r
Table 1: Numerical values of the quantities of interest in different urban planning config-
urations for Sense-City district, k denotes the reaction rate of the depolluting panels.

4. Real case application - Paris district

er
pe
In Section 3, the numerical strategy for smart placement of depolluting
panels was applied in a simple urban geometry and under controlled weather
conditions in Sense-City to demonstrate its effectiveness. However, actual
ot

urban flows depend on meteorological conditions that vary with time. In this
section, therefore, we examine the applicability of the numerical strategy to
tn

a real urban district of Paris under different wind conditions.

In this study, realistic weather conditions and pollution situations are
considered using the two most dominant wind directions encountered in Paris
rin

and actual measured NOx pollutant concentrations on the roads. In Paris

region, generally, the two dominant airflows are North-East (NE) and South-
West (SW). Thanks to the cooperation with Airparif agency, two critical days
ep

and specific hours are determined, corresponding to high pollutant levels and
NE or SW wind conditions. Accordingly, we selected the 4th of December
Pr

32

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
2021 at 7 pm with a wind direction of 230 degrees and the 16th of December

iew
2021 at 11 am with a 70 degrees wind direction. These two studied winter
days are outside of the intensive photochemical period (April-September).

4.1. Computational setting

ev
The direct simulation for flow and pollutant dispersion is performed with
the finite-volume CFD software “Code Saturne”(Frederic et al., 2004). In or-
der to compromise between accuracy and computational cost, we consider the

r
pseudo-steady-state incompressible Reynolds-Averaged Navier-Stokes (RANS)
with K −ω SST turbulence model. Second-order upwind schemes are chosen
er
to compute velocity, turbulence kinetic energy, and specific dissipation. As
with Sense-City case, the turbulent Schmidt number Sct is fixed at 0.7.
pe
OpenFOAM (Greenshields et al., 2015), an open source software, is used
to solve the pseudo-steady-state adjoint advection-diffusion equation. It cor-
responds to Eq. (8) without the time derivative term. A first-order upwind
ot

scheme is chosen to discretize the advection term. The velocity field ⃗v and
turbulent viscosity field of the adjoint equation are imported from the direct
simulation made with Code Saturne.
tn

4.2. From real urban geometry to mesh generation

The study area is a part of the 8th arrondissement of Paris (France),

rin

which is located on the west side of the city (see Fig. 8 (a)). This area
includes housing, business, political activities, and many tourist spots, such
as the Avenue des Champs-Élysées. The white zone in Fig. 8 (b) indicates
ep

the main domain of focus in this study. To imitate an actual incoming flow
affected by windward buildings, we put the extra region represented with
Pr

33

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
yellow lines in Fig. 8 (b). The domain size is about 2 km in the NS direction

iew
and 1.55 km in the EW direction. An air quality measurement station of the
Airparif agency is present in the studied domain at the Avenue des Champs-
Élysées (see red circle in Fig. 8 (b)). Many streets in this area are known
to be highly polluted because of the traffic, e.g. Av des Champs-Élysées, Bd

ev
Malesherbes, and Rue de Miromesnil.

r
er
pe
ot
tn

Figure 8: (a) Area of simulation: overall view of Paris (at left) and (b) detail view of the
rin

study area (at right).

ep

The numerical mock-up of the urban area is generated via a software

chain. From IGNMap, 2D shapes with height information of each building are
exported. Then, 3D geometry is obtained by the automatic extrusion process
Pr

34

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
of each building in CityEngine software. Let us note that precise heights of

iew
the buildings and courtyards are taken into account in the main domain of
interest (white zone in Fig. 8) whereas geometrical simplifications are made
in the extra region (yellow line in Fig. 8) such as constant height in a building
block and removal of courtyard in building blocks. Lastly, an unstructured

ev
mesh of the computational domain is generated in SALOME software (Ribes
and Caremoli, 2007). The computational domain is adapted according to
the wind direction. Fig. 9 shows the computational domain and mesh when

r
considering wind from the SW direction. For NE case, only the building

er
geometries are rotated to make a new computational domain. Each distance
from the outer buildings to the boundary of the domain is determined by
reference to Guidebook (AIJ, 2020), suggesting maintaining enough length
pe
to avoid the interaction between flow and the domain boundary except for
the inlet face where the distance between them should be small enough to
let the inflow keep its flow profile until the main domain. Vertical length is
ot

more than 10Hmax . Herein, Hmax is equal to 42.5 m which corresponds to the
maximum building height. As mentioned, the wind direction, the days, and
tn

the time chosen for the simulations are 230 degrees on 4th December 2021
at 7 pm (SW) and 70 degrees on 16th December 2021 at 11 am (NE). The
number of cells is about 25 million for SW and NW cases. In the studied
rin

domain, 1m mesh size is used where traffic pollutant sources are put, and 3m
mesh size elsewhere.
ep

4.3. Boundary conditions and traffic pollutant source

At the inlet faces, as shown in Eqs. (15-17), we impose a logarithmic pro-

file for velocity, a constant value for kinematic energy and a height-dependent
Pr

35

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
r ev
Figure 9: Computational area and mesh of simulation for Paris SW case.

er
profile for specific dissipation assuming fully developed inflow conditions
(P.J. Richards, 2011).
pe
 
u∗ z + z0
U= ln (15)
κ z0

u∗ 2
ot

k= p (16)
Cµ
tn

u∗
ω=p ′ (17)
β κk−ω z
where u∗ , z0 , and κ are the friction velocity, the roughness length, and Von
Karman constant. The roughness length is taken to 1 m. To define u∗ for
rin

the inlet velocity profile, we use velocity measurement data at 10 m height at

Montsouris station in Paris. It corresponds to 2 m/s for SW and 2.5 m/s for
′
ep

NE case. The details concerning the calculation of the coefficients Cµ , β and

κk−ω can be found in (P.J. Richards, 2011). On the outlet face, zero-gradient
conditions are imposed. Wall function is used on the bottom face and all the
Pr

36

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
building surfaces. For other boundaries, symmetry conditions are given on

iew
laterals and upper faces.
Concerning NOx emissions from traffic, they are estimated on one-hour
intervals by the Airparif agency using emission factors from COPERT (Ntzi-
achristos et al., 2009) and data assimilation, which combines observed data

ev
and a traffic model. In the studied domain, we consider as pollutant sources
44 main roads (see red lines in Fig. 10), which contribute the most to traffic
pollution. In the CFD calculations, the pollutant sources on the main roads

r
are modeled as volumetric source terms in the advection-diffusion equation.

er
The pollutant volumetric sources are defined on the width of the roads from
0 m to 1 m height to mimic traffic emissions with a one-hour average pollu-
tant concentration. NOx emissions on each street are given in µg/m3 /s. For
pe
the details of the NOx values, see Appendix B.
ot
tn
rin
ep

Figure 10: Position of main traffic pollutant sources (44 roads in the considered domain).

.
Pr

37

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
4.4. Simplifications and limitations of Paris study case

iew
The main assumptions made to simplify the Paris study case in terms of
geometry, airflow, pollutant dispersion, and chemical modeling are listed and
summarized in this section.

ev
- Wind and traffic pollution scenarios: the airflow and the traffic pollu-
tant dispersion in the Paris case are studied on hourly-averaged time in-
tervals using pseudo-steady simulations. Only two characteristic study

r
cases, i.e. highly-polluted periods from traffic (morning and evening)
with Paris region dominant winds (South-West and North-East), are
er
analyzed to propose first recommendations on the placement of depol-
luting panels in the urban area;
pe
- Estimated traffic pollution emission: as mentioned in Section 4.3, the
pollutant sources are estimated from a traffic model and pollutant emis-
sion factors. The fleet and the velocity of the vehicles not being well-
ot

known, these input data are subject to uncertainties;

- No height variation in the geometry: in the present Paris domain, the

tn

elevation is about 20 m lower from north to south. Although the to-

pography can influence the airflow, herein the slope being slight, we do
rin

not consider the height variation of the terrain;

- Airflow disturbance induced by urban equipment, trees, traffic, and tem-

perature: trees and vehicles can increase turbulence and air mixing in
ep

urban areas. Thereby, they may facilitate the dispersion of pollutants

and reducing pollutant concentrations. However, for simplicity, they
Pr

38

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
are not considered in modeling the case. In the same way, some ge-

iew
ometrical details, such as urban equipment, are not included in the
digital twin. Furthermore, we neglected the buoyancy effects caused
by temperature-induced density changes and the atmospheric stability
caused by temperature differences between the ground and the atmo-

ev
sphere.

- Chemical reaction of traffic pollutants: from NOx traffic emissions pro-

r
vided by AirParif agency, NOx background concentration, and the de-
tailed airflow numerical fields, we compute the cartography of NOx

er
concentration in the Paris domain using advection-diffusion PDE. We
neglected the possible reaction with other chemical species. We recall
pe
that the depolluting action of the panel surface is simply modeled with
a first-order reaction. Its reaction rate function k is assumed to be
independent of UV radiation levels and air temperature.
ot

4.5. Cartography of the pollutant concentration and critical pollution areas

in Paris
tn

The first step of the proposed numerical strategy is to make a cartog-

raphy of pollutant concentration. Fig. 11 shows the mean velocity and
NOx concentration at 1.5 m height for the SW and NE cases. Concerning
rin

the numerical estimation of NOx concentrations, measurement values of the

2
air quality subway station Chatelet-Les Halles (in the center of Paris) are
used as background concentrations: 45.5 µg/m3 and 164.5 µg/m3 for SW
ep

2
Since this station is far from any large street, the measured concentration is considered
as a background concentration by AirParif.
Pr

39

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
(µg/m3 )

iew
SW NE
Background concentration at Chatelet-Les Halles station 45.5 164.5
CFD simulation (traffic-derived emission) 52.3 5.4
Total predicted concentration (background + CFD) 97.8 169.9

ev
Measurement station at Av des Champs-Elysées 123.0 154.1

Table 2: Comparison of NOx concentration between CFD and measurement station.

r
er
and NE cases, respectively. CFD is used to compute the cartography of
NOx concentration resulting from traffic emissions. Hence, the total NOx
concentration is obtained by adding the background concentration to traffic-
pe
derived concentration predicted by CFD. Table 2 compares the numerical
NOx concentration with the measured concentration at the air quality sta-
tion of AirParif agency located at Av des Champs-Élysées (see red circle in
ot

Fig. 8). The traffic-derived pollutant concentration at the air quality station
is lower for NE wind direction case than for SW wind. In fact, few traffic
tn

emissions are conveyed to the measurement station as (i) the wind direction
is NE, (ii) the air quality station is placed on the NE sidewalk of Champs
Elysées and (iii) an important green park is present at the vicinity of the air
rin

quality station in the NE direction (no big traffic emission). At the contrary,
in the SW case, the measurement station observes the high traffic pollutant
emission from Av des Champs Elysées. Overall, in both simulated cases, an
ep

acceptable gap, i.e. less than 25%, is obtained between the simulated and
measured NOx concentrations at Av des Champs-Élysées.
Pr

40

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
r ev
er
pe
ot
tn
rin

Figure 11: Mean velocity and not Mean NOx concentration on 1 hour-interval at 1.5 m
height: SW case (top) and NE case (bottom).

.
ep
Pr

41

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
CFD results represented in Fig. 11 highlight that the pollutant concen-

iew
tration is inhomogeneous at the district and street levels. As expected, high
pollutant concentrations are predicted on the 44 main streets where volu-
mic sources are placed (e.g., Av des Champs-Élysées and Bd Malesherbes),
while lower concentrations are obtained on the other streets due to only the

ev
contribution of the advection and diffusion of pollutants from the 44 streets.
Low airflow velocities in some areas result in an pollution accumulation such
as in Rue de Miromesnil. Moreover, it is observed on many streets that the

r
pollutant concentrations are high only on one side of the road. This is due

er
to the typical pollutant distribution of the street canyon where highest con-
centrations are noticed on the leeward side of roads, as the street primary
vortex (Dabberdt et al., 1973) conveys pollutants leeward. For example,
pe
Fig. 12 shows pollutant concentration and wind direction on a street’s cross
section at 58 Bd Malesherbes for the SW case. On the left side (leeward),
the NOx concentration is higher than the right side and above 500 µg/m3 at
ot

the pedestrian level, with decreasing concentration as height increases. Let

us note that due to the simplification assumptions considered and detailed
tn

in Section 4.4, the predicted NOx concentration may give over-estimation.

The next step is to select critically polluted areas (areas of interest) where
rin

the air quality should be improved. As regards the two wind direction cases,
four areas of interst are selected from the NOx pollutant cartographies:
ep

• Ωq1 : southwest sidewalk from 69 to 85 Bd Malesherbes represented in

Fig. 13 (a) in red, and in Fig. 14 (a) for SW case. Dimension of the
area is 5 m width, 140 m length, and height belongs to [0.5 m, 2 m].
Pr

42

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
r ev
er
pe
Figure 12: Mean NOx concentration and wind direction on vertical direction in SW case
at 58 Bd Malesherbes.

.
ot

The associated mean pollutant concentration J1 is 627 µg/m3 .

tn

• Ωq2 : building facade of the lower floors on the southwest side from 69
to 81 Bd Malesherbes represented in Fig. 13 (a) in blue and in Fig.
rin

14 (a) for SW case. The dimension of the area is 0.5 m width, 110 m
length, and height belongs to [4 m, 9.2 m]. The associated mean NOx
concentration J2 is 427 µg/m3 .
ep

• Ωq3 : sidewalks at the intersection of Rue de Miromesnil and Rue la

Boétie represented in Fig. 13 (b) and in Fig. 14 (b)) for SW case. The
Pr

43

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
dimension of the area is 1 ∼ 2.5 m width, 20 ∼ 50 m length, and height

iew
belongs to [0.5 m, 2 m]. The associated mean NOx concentration J3 is
252 µg/m3 .

• Ωq4 : north-east sidewalk from 48 to 60 Bd Malesherbes represented in

Fig. 13 (c) and in Fig. 14 (a) for NE case. Dimension of the area

ev
is 4 m width, 140 m length, and height belongs to [0.5 m, 2 m]. The
associated mean NOx concentration J4 is 326 µg/m3 .

r
The quantities of interest J1 , J3 and J4 have been selected in order to

er
improve the air quality on sidewalks frequented by pedestrians. J3 is also of
particular interest as it is located at a road intersection having the crowded
metro station entrance “Miromesnil” and many city shops. Lastly, reducing
pe
air pollution on the building facade, e.g. J2 at Bd Malesherbes, is important
in order to decrease the transfer of traffic pollutants from the outdoors to
the indoors.
ot

4.6. Adjoint solution and smart placement of depolluting panels in Paris

To smartly place the depolluting panels on the surfaces of the urban

tn

domain, the last step is to calculate the adjoint concentration and the sensi-
tivity indicator for each selected quantity of interest defined in the previous
rin

section. Herein, we consider that the depolluting panels can be potentially

placed on all the surfaces of the domain, i.e. Γp = Γw . Hence, the sensitivity
indicator is not evaluated only in the areas of interest. As explained before,
ep

the sensitivity indicator becomes significant when both direct and adjoint
concentrations are high. Fig. 15 shows the cartographies of the NOx concen-
tration (solution of the direct problem), the adjoint concentration, and the
Pr

44

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
r ev
er
Figure 13: Selected areas of interest in the Paris district where air quality should be
pe
improved: (a) the south-west sidewalk in red and the lower floors of the building facade in
blue from 69 to 81 Bd Malesherbes, (b) sidewalks at the intersection of Rue de Miromesnil
and Rue de Boétie, (c) the north-east sidewalk from 48 to 60 Bd Malesherbes.
ot

resulting sensitivity indicator on the surfaces of the domain for each quan-
tity of interest. The bold yellow lines in the sensitivity indicator map denote
tn

the contour lines of the sensitivity indicator at 0.1 for both the south-west
sidewalk and the lower floor building facade at Bd Malesherbes, at 0.2 at the
intersection of Rue de Miromesnil and at 0.003 for the north-east sidewalk
rin

at Bd Malesherbes. These values are associated to the chosen sensitivity

indicator thresholds. Herein, the thresholds have been selected such that the
surface of depolluting panels does not exceed 300 m2 .
ep

On the southwest sidewalk at Bd Malesherbes, we define the north and

south parts of the sidewalk (see Fig. 16). Firstly, concerning the quantity of
Pr

45

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
r ev
er
pe
Figure 14: Pictures of the areas of interest in the Paris district: (a) 58 Bd Malesherbes
and (b) 43 Rue de Miromesnil.

.
ot

interest J1 , the sensitivity indicator reaches highest values on the ground of

tn

the north part. It corresponds to a portion of the sidewalk, of the road and
of the lower floor’s building facade (Fig. 15 (a)). Similarly, for the quantity
of interest J2 , significant sensitivity indicator values are mainly located on
rin

the north part of Bd Malesherbes on the lower floor building facade (Fig. 15
(b)). In the other regions, the indicator is relatively low. For J1 and J2 , we
observe important variation of the sensitivity indicator even on the same side-
ep

walk and building facade which are attributed to the magnitude of adjoint
concentration. Looking at the airflow characteristics in Bd Malesherbes (Fig.
Pr

46

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
16), a typical street canyon flow occurs on the north part. Hence, in the north

iew
part of Bd Malesherbes, the pollutant source of the adjoint problem defined
on Ωq1 and Ωq2 are backwardly transported along the ground and building
facade by the primary vortex. It leads to high adjoint concentration on the
sidewalk, road and building facade in the north part of Bd Malesherbes. On

ev
the other hand, in the center and south part (except for the south corner),
the adjoint concentration is low on the ground and building facade. In these
areas, the vertical velocity is tiny compared to the north part, and the wind

r
direction near the building surfaces is oriented parallel to the road (Fig. 16).

er
This airflow does not transport the adjoint pollutant concentration near the
ground and building facade, resulting in low indicator values in these regions.
To sum up, urban airflow plays a major role in the sensitivity indicator level
pe
and in determining the smart placement of depolluting panels. The amount
of adjoint concentration passing close to the ground or the building surface
is a major contributer for the selection of panel location and size.
ot

For the quantity of interest J3 dealing with the NOx concentration at the
street intersection of Rue de Miromesnil and Rue la Boétie, Fig. 15 (c) shows
tn

that high sensitivity indicator values are obtained on the sidewalk surfaces,
especially in Rue de Miromesnil, a well-known road having heavy traffic,
and also on lower floor’s building facades. Moreover, the sensitivity indica-
rin

tor is significant at the entrance to the subway station. In this intersection,

complex airflow is created as winds come from many directions and merge.
Hence, this air mixing favors the direct and adjoint concentrations to reach
ep

the ground surfaces and building facades as illustrated.

Concerning the quantity of interest J4 on the north-east sidewalk of Bd
Pr

47

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Malesherbes, Fig. 15 (d) indicates very local high sensitivity indicator at the

iew
building corner on the south part mainly due to high adjoint concentrations.
In addition, a high indicator is obtained around the center of the sidewalk
near the vertical setback. (Ng and Chau, 2014) pointed out that vertical
setbacks increase airflow and pollutant dispersion in the vertical direction.

ev
Its effect contributes to the high adjoint concentration in this region, thereby
the high indicator. This finding implies that combining measures to facili-
tate airflow mixing with depolluting panels can be a more efficient way to

r
improve air quality. However, in this northeast sidewalk case, note that the

er
magnitude of the adjoint concentration and the sensitivity indicator is much
smaller than in the other cases. Therefore, putting depolluting panels in the
northeast sidewalk can be less efficient.
pe
To conclude, for the improvement of the air quality at Bd Malesherbes
(69-85 and 48-60) and at the intersection of Rue de Miromesnil and of Rue
ot

la Boétie under the considered wind conditions, the suggested placement of

depolluting panels based on the chosen sensitivity contour lines in Fig. 15
tn

are the following:

• Bd Malesherbes - panels should be placed on the north part of the

south-west sidewalks (4 m wide × 30 m long) and of the road adjacent
rin

to the sidewalk (3.5 m × 30 m), on the building facade in the north part
(height∈ [2.5m, 9.5m] × 35 m) and in the south part (height∈ [2m, 9m]
ep

× 10 m). Although the placement of panels can be less effective on the

north-east sidewalk due to the low sensitivity indicator in comparison
with the south-west sidewalk, candidate positions are on the sidewalk
Pr

48

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
r ev
er
pe
ot
tn

Figure 15: Direct concentration, adjoint concentration, and sensitivity indicator associated
with each quantity of interest: (a) the south-west sidewalk from 69 to 85 Bd Malesherbes,
rin

(b) the building facade at the lower floor on south-west side from 69 to 81 Bd Malesherbes,
(c) the subway station and shops at the intersection of Rue de Miromesnil and Rue la
Boétie, (d) the north-east sidewalk from 48 to 60 Bd Malesherbes.
ep

at the south corner (6 m × 35 m) and in the center of the sidewalk (4

m × 30 m).
Pr

49

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
iew
r ev
Figure 16: Vertical velocity and wind direction at Bd Malesherbes at 1m height.

.
er
• Intersection of Rue de Miromesnil and of Rue la Boétie - panels should
pe
be placed on a part of the sidewalk and road on the north side of Rue de
Miromesnil (1.5 m wide × 50 m long) and (1.5 m × 50 m) respectively,
at the entrance of the metro station (6 m × 15 m) and at the part
ot

of the sidewalk on Rue la Boetie (6 m × 5 m). The other candidate

position is a part of the south sidewalk on Rue de Miromesnil (1.5 m
tn

× 12 m).

The total areas enclosed by each contour line Ii (x) for each quantity of in-
terest are summarized in Table 3. Let us recall that the contour lines of the
rin

sensitivity indicators are used as the thresholds to identify the smart place-
ments. Herein, the sensitivity indicator thresholds are chosen empirically for
a limited and reasonable deployment of depolluting panels. In practice, the
ep

threshold can be chosen depending on how much local authorities want to

improve air quality and the cost that can be afforded.
Pr

50

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
(m2 )

iew
South-West sidewalk and road at Bd Malesherbes (I1 ) 255.7
SW South-West building facade at Bd Malesherbes (I2 ) 236.2
Sidewalk and road at the intersection of Rue de 252.3
Miromesnil (I3 )

ev
NE North-East sidewalk at Bd Malesherbes (I4 ) 239.6

Table 3: Area of depolluting panel enclosed by I1 (x) = 0.1, I2 (x) = 0.1, I3 (x) = 0.2 and

r
I4 (x) = 0.003.

5. Conclusions and perspectives er

pe
To improve the air quality in critical highly-polluted urban areas, we pro-
posed a numerical strategy for the smart placement of depolluting panels.
These panels can adsorb and degrade a part of traffic pollutants provided
that the pollutants pass near these depolluting surfaces. Hence, we focused
ot

on an efficient placement of panels as regards to urban airflows using CFD

simulations. Our approach can be decomposed into two steps. Firstly, in the
tn

diagnosis stage, detailed airflow and pollutant cartographies are computed

at the district scale to identify critical highly-polluted areas which are des-
ignated as “quantities of interest” in the article. Then, in the remediation
rin

stage, local sensitivity analysis is performed through an adjoint framework

to determine a relevant and limited placement of depolluting panels with the
aim of reducing the pollutant concentration in localized critical areas. In-
ep

deed, the adjoint concentration field provides the location of the areas of the
domain having an impact on the chosen quantities of interest. As shown in
Pr

51

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
the studied cases, the adjoint advection-diffusion problem can be solved using

iew
finite volume or finite element CFD softwares. A spatial sensitivity indicator
was introduced and computed on all the surfaces of the domain to determine
the best placement of depolluting panels. The indicator involves the prod-
uct of the pollutant concentration field and the adjoint concentration field.

ev
To summarize, the depolluting panels should be deployed on surfaces having
high pollutant concentration and significant impact on the chosen quantities
of interest.

r
The numerical strategy for the smart placement of depolluting accord-

er
ing to urban airflow was illustrated in two real cases. The first application
deals with a small real-scale district named “Sense-City” under controlled
airflow and environmental conditions provided by a huge climatic chamber.
pe
In this case, we considered a scenario with a localized source of pollutants
on the road. Then, in the second application, a real district of Paris was
studied. Two representative wind conditions of the Region Ile de France and
ot

realistic NOx source emissions given by the Airparif agency were considered.
Iin both cases, inhomogeneous, highly polluted areas are observed on the
tn

sidewalks, road and building facades. In order to improve the air quality
in these critical areas, smart placement of depolluting panels was proposed
using the numerical strategy. It was shown that depolluting panels should
rin

be placed on a part of the sidewalks, of the building facades and of the

roads adjacent to the sidewalks. By comparing improvement effectiveness
in two panel deployment pattern: smart and non-smart placements, it was
ep

revealed that a massive aleatory deployment of depolluting panels may not

be efficient. In fact, we proved that the urban airflow plays a major role
Pr

52

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
regarding the pollutant concentration distribution, the adjoint concentration

iew
and ultimately the sensitivity indicator for the smart placement of depol-
luting panels. In particular, the sensitivity indicator tends to be significant
when the airflow path is on a line connecting the quantity of interest down-
wind and the buildings and ground surface windward. By fixing a threshold

ev
value on the sensitivity indicator, we can determine the surface of depollut-
ing panels to be deployed. In practice, the local authorities can select the
threshold value by considering the balance between the objective of air qual-

r
ity improvement and expenditure on depolluting panels. Although it was

er
not examined in this study, combinations with other air pollutant mitigation
actions, especially those that enhance airflow mixing, may enable the panels
to be placed more efficiently.
pe
The goal of this article was to propose a simple and preliminary opera-
tional numerical strategy for a better deployment of depolluting panels in ur-
ban areas. In the presented results, the first recommendations on depolluting
ot

panels placement are essentially given regarding the transport of pollutants

via the urban air flows. Herein, many simplifications have been considered
tn

for easier practical use. Even though the pollution scenarios in Paris were
studied in the winter period on small time intervals of 1 hour, we didn’t take
into account buoyancy effects, traffic-induced turbulence terms, and tree veg-
rin

etation which can modify the airflow and the pollutant dispersion. Moreover,
to limit the complexity and the time computation, the pollutant was modeled
as passive (multiple pollutant reactions not considered). Lastly, the degra-
ep

dation of the pollutant by the depolluting system is described in a simple

way using a Robin boundary condition with a first-order reaction. Despite
Pr

53

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
these simplifying assumptions, it can be underlined that the actual proposed

iew
method can distinguish “useless panel placement areas” corresponding to
surfaces having no impact on the improvement of the quantities of interest
and “promising panel placement areas”. In future works, a more sophisti-
cated strategy can be developed to take into account the above limitations

ev
and to get a more precise quantitative evaluation of the “promising panel
placement areas”. For that, we can move toward multi-pollutant reactions
and multi-physics simulations. The pollutant degradation mechanism of the

r
panels is based on photocatalysis, the panels’ efficiency depends on the solar

er
irradiance, the temperature, and many other physical parameters. Research
works have to be conducted on both experimental and numerical aspects. An
extensive experimental characterization of the panels at the material scale in
pe
controlled conditions in the laboratory is required. From these experiments,
a constitutive law can be obtained and implemented in multi-physics simu-
lation at the district scale. The coupled-problem solutions providing notably
ot

airflow, solar irradiance, temperature, and pollutant cartographies will allow

an improved prediction of panel pollutant degradation and thus an enhanced
tn

evaluation of the panel placement. At Université Gustave Eiffel, forthcom-

ing controlled air pollution experiments are to be conducted in the district
of the Sense-City equipment to validate the proposed numerical strategy for
rin

the smart placement of ZnO depolluting panels.

Acknowledgments
ep

The authors acknowledge the supports from the innovative research project
“E3S” funded by Université Gustave Eiffel and Eiffage group and from the
Pr

54

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
project “Sense-City” funded by ANR (France) within the Investment for the

iew
Future Program under reference number ANR-10-EQPX-48.

Appendix A. Calculation of Lagrange multipliers and adjoint prob-

lem

ev
In Section 2, the constrained minimization problem (6) was rewritten
using the Lagrangian L defined in Eq. (7). After some calculations to find
the saddle point of the Lagrangian, we obtain the following relations between

r
the Lagrange multipliers

er
λ2 = −D∇λ1 · ⃗n − (⃗v · ⃗n)λ1 on Γo × [0, T ] (A.1)
pe
λ3 = Dλ1 on Γo ∪ Γn × [0, T ] (A.2)

λ4 = −λ1 on Γp × [0, T ] (A.3)

λ5 = λ1 (t = 0) on Ω (A.4)
ot

and we can show that the Lagrange multiplier λ1 satisfies


∂λ1
− − ⃗v · ∇λ1 − div(D∇λ1 ) = fq ξ in Ω × [0, T ]



tn




 ∂t




 λ1 = 0 on Γi × [0, T ]



∇λ1 · ⃗n = 0 on Γn × [0, T ]

(A.5)
rin

−D∇λ1 · ⃗n = kλ1 on Γp × [0, T ]












 D∇λ1 · ⃗n + (⃗v · ⃗n) λ1 = 0 on Γo × [0, T ]



λ1 (t = T ) = 0 in Ω.
ep

Thus, Eq. (A.5) corresponds to the adjoint problem. By convenience, we

note C̃(= λ1 ) the solution of this adjoint problem.
Pr

55

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Appendix B. NOx road emissions in Paris district

iew
The NOx emission magnitude (µg/m3 /s) of the 44 roads used in the
direct simulation are summarized in Fig. B.17. These data were provided
by the Airparif agency. Emission values represent only traffic-derived NOx.
These values are given as source terms and are emitted as a constant value

ev
of NOx throughout the simulations.

r
er
pe
ot
tn

Figure B.17: Data of NOx emission on 44 roads: position of roads (left) and emission
magnitude (right). Note that No.2 and No.3 is summed up and treated as one road.

.
rin

References
ep

AIJ, 2020. Guidebook for CFD Predictions of Urban Wind Environment.

Architectural Institute of Japan.
Pr

56

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
An, K., Wong, S.M., Chi-Hung Fung, J., 2019. Exploration of sustainable

iew
building morphologies for effective passive pollutant dispersion within com-
pact urban environments. Building and Environment 148, 508–523.

André, M., Pasquier, A., Carteret, M., 2018. Experimental determination of

the geographical variations in vehicle fleet composition and consequences

ev
for assessing low-emission zones. Transportation Research Part D: Trans-
port and Environment 65, 750–760.

r
Andre, M., Sartelet, K., Moukhtar, S., Andre, J., Redaelli, M., 2020. Diesel,

er
petrol or electric vehicles: What choices to improve urban air quality in the
ile-de-france region? a simulation platform and case study. Atmospheric
Environment 241, 117752.
pe
Balczo, M., Gromke, C., Ruck, B., 2009. Numerical modeling of flow and
pollutant dispersion in street canyons with tree planting. Meteorologische
Zeitschrift 18, 197–206.
ot

Bereitschaft, B., Debbage, K., 2013. Urban form, air pollution, and co2
tn

emissions in large us metropolitan areas. The Professional Geographer 65,

612–635.

Bessagnet, B., Menut, L., Curci, G., Hodzic, A., Guillaume, B., Liousse, C.,
rin

Moukhtar, S., Pun, B., Seigneur, C., Schulz, M., 2008. Regional modeling
of carbonaceous aerosols over europe-focus on secondary organic aerosols.
Journal of Atmospheric Chemistry 61, 175–202.
ep

Binas, V., Venieri, D., Kotzias, D., Kiriakidis, G., 2017. Modified tio2 based
Pr

57

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
photocatalysts for improved air and health quality. Journal of Materiomics

iew
3, 3–16.

Blocken, B., 2015. Computational Fluid Dynamics for urban physics: Im-
portance, scales, possibilities, limitations and ten tips and tricks towards
accurate and reliable simulations. Building and Environment 91, 219–245.

ev
Blocken, B., Persoon, J., 2009. Pedestrian wind comfort around a large
football stadium in an urban environment: CFD simulation, validation

r
and application of the new Dutch wind nuisance standard. Journal of

er
wind engineering and industrial aerodynamics 97, 255–270.

Carruthers, D., Holroyd, R., Hunt, J., Weng, W., Robins, A., Apsley, D.,
pe
Thompson, D., Smith, F., 1994. Uk-adms: A new approach to modelling
dispersion in the earth’s atmospheric boundary layer. Journal of Wind
Engineering and Industrial Aerodynamics 52, 139–153.
ot

da Silva, F., Costa Reis, N., Santos, J., Goulart, E., de Alvarez, C., 2022.
Influence of urban form on air quality: The combined effect of block ty-
tn

pology and urban planning indices on city breathability. Science of The

Total Environment 814, 152670.

Dabberdt, W.F., Ludwig, F., Johnson Jr, W.B., 1973. Validation and appli-
rin

cations of an urban diffusion model for vehicular pollutants. Atmospheric

Environment (1967) 7, 603–618.
ep

Daghistani, F., 2021. Solar chimney street-lighting pole for ventilating pol-
luted urban areas. Sustainable Cities and Society 72, 103057.
Pr

58

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Darvish, S.M., Ali, A.M., Sani, S.R., 2020. Designed air purifier reactor for

iew
photocatalytic degradation of co2 and no2 gases using mwcnt/tio2 thin
films under visible light irradiation. Materials Chemistry and Physics 248,
122872.

Derkx, F., Lebental, B., Bourouina, T., Bourquin, F., Cojocaru, C.S.,

ev
Robine, E., Van Damme, H., 2012. The Sense-City project, in: XVII-
Ith Symposium on Vibrations, Shocks and Noise, p. 9p.

r
Di Sabatino, S., Buccolieri, R., Pulvirenti, B., Britter, R.E., 2008. Flow and

er
pollutant dispersion in street canyons using FLUENT and ADMS-Urban.
Environmental Modeling & Assessment 13, 369–381.
pe
Di Sabatino, S., Buccolieri, R., Salizzoni, P., 2013. Recent advancements in
numerical modelling of flow and dispersion in urban areas: a short review.
International Journal of Environment and Pollution 7 52, 172–191.
ot

Duque, L., Relvas, H., Silveira, C., Ferreira, J., A., M., Gama, C., Rafael,
S., Freitas, S., Borrego, C., Miranda, A., 2016. Evaluating strategies to
tn

reduce urban air pollution. Atmospheric Environment 127, 196–204.

Elbern, H., Schmidt, H., 2001. Ozone episode analysis by four-dimensional

variational chemistry data assimilation. Journal of Geophysical Research:
rin

Atmospheres 106, 3569–3590.

Franca, L.P., Frey, S.L., Hughes, T.J., 1992. Stabilized finite element meth-
ep

ods: I. application to the advective-diffusive model. Computer Methods

in Applied Mechanics and Engineering 95, 253–276.
Pr

59

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Frederic, A., Namane, M., Marc, S., 2004. Code saturne: A finite volume

iew
code for the computation of turbulent incompressible flows-industrial ap-
plications. International Journal on Finite Volumes 1.

Garrido, I., Pastor-Belda, M., Campillo, N., Viñas, P., Yañez, M.J., Vela, N.,
Navarro, S., Fenoll, J., 2019. Photooxidation of insecticide residues by zno

ev
and tio2 coated magnetic nanoparticles under natural sunlight. Journal of
Photochemistry and Photobiology A: Chemistry 372, 245–253.

r
Gousseau, P., Blocken, B., Stathopoulos, T., van Heijst, G.J.F., 2011. CFD

er
simulation of near-field pollutant dispersion on a high-resolution grid: A
case study by LES and RANS for a building group in downtown Montreal.
Atmospheric Environment 45, 428–438.
pe
Grande, F., Tucci, P., 2016. Titanium dioxide nanoparticles: a risk for human
health? Mini reviews in medicinal chemistry 16, 762–769.
ot

Greenshields, C.J., et al., 2015. Openfoam user guide. OpenFOAM Founda-

tion Ltd, version 3, 47.
tn

Gromke, C., Blocken, B., Janssen, W., Merema, B., van Hooff, T., Timmer-
mans, H., 2015. CFD analysis of transpirational cooling by vegetation:
Case study for specific meteorological conditions during a heat wave in
rin

Arnhem, Netherlands. Building and Environment 83, 11–26.

Gromke, C., Buccolieri, R., Di Sabatino, S., Ruck, B., 2008. Dispersion
ep

study in a street canyon with tree planting by means of wind tunnel and
numerical investigations - Evaluation of CFD data with experimental data.
Atmopheric Environment 42, 8640–8650.
Pr

60

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Gromke, C., Jamarkattel, N., Ruck, B., 2016. Influence of roadside hedgerows

iew
on air quality in urban street canyons. Atmospheric Environment 139, 75–
86.

Hagler, G., Tang, W., Freeman, M., Heist, D., Perry, S., Vette, A., 2011.
Model evaluation of roadside barrier impact on near-road air pollution.

ev
Atmospheric Environment 45, 2522–2530.

Hammond, J.K.H., Chakir, R., Bourquin, F., Maday, Y., 2019. Pbdw: A non-

r
intrusive reduced basis data assimilation method and its application to an

er
urban dispersion modeling framework. Applied Mathematical Modelling
76, 1–25.
pe
Hang, J., Li, Y., Sandberg, M., 2011. Experimental and numerical studies of
flows through and within high-rise building arrays and their link to venti-
lation strategy. Journal of wind engineering and industrial aerodynamics
99, 1036–1055.
ot

He, D., Li, Y., Wu, J., Yang, Y., An, Q., et al., 2017. Carbon wrapped
tn

and doped tio2 mesoporous nanostructure with efficient visible-light pho-

tocatalysis for no removal. Applied Surface Science 391, 318–325.

Hecht, F., 2012. New development in freefem++. Journal of Numerical

rin

Mathematics 20, 251–265. URL: https://freefem.org/.

Hughes, T.J., Mallet, M., Akira, M., 1986. A new finite element formulation
ep

for computational fluid dynamics: Ii. beyond supg. Computer methods in

applied mechanics and engineering 54, 341–355.
Pr

61

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Issakhov, A., Omarova, P., 2021. Modeling and analysis of the effects of

iew
barrier height on automobiles emission dispersion. Journal of Cleaner Pro-
duction 296, 126450.

Jeanjean, A., Hinchliffe, G., McMullan, W., Monks, P., Leigh, R., 2015. A
cfd study on the effectiveness of trees to disperse road traffic emissions at

ev
a city scale. Atmospheric Environment 120, 1–14.

Kang, G., Kim, J.J., Choi, W., 2020. Computational fluid dynamics simula-

r
tion of tree effects on pedestrian wind comfort in an urban area. Sustain-
able Cities and Society 56, 102086.
er
Kosutova, K., van Hooff, T., Vanderwel, C., Blocken, B., Hensen, J., 2019.
pe
Cross-ventilation in a generic isolated building equipped with louvers:
Wind-tunnel experiments and cfd simulations. Building and Environment
154, 263 – 280.
ot

Koutsourakis, N., Bartzis, J.G., Markatos, N.C., 2012. Evaluation of

Reynolds stress, k-epsilon and RNG k-epsilon turbulence models in street
tn

canyon flows using various experimental datasets. Environmental Fluid

Mechanics 12, 379–403.

Le Dimet, F.X., Talagrand, O., 1986. Variational algorithms for analysis and
rin

assimilation of meteorological observations: theoretical aspects. Tellus A

38A, 97–110.
ep

Le Pivert, M., Kerivel, O., Zerelli, B., Leprince-Wang, Y., 2021. Zno nanos-
tructures based innovative photocatalytic road for air purification. Journal
of Cleaner Production 318, 128447.
Pr

62

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Le Pivert, M., Zerelli, B., Martin, N., Capochichi-Gnambodoe, M., Leprince-

iew
Wang, Y., 2020. Smart zno decorated optimized engineering materials
for water purification under natural sunlight. Construction and Building
Materials 257, 119592.

Leung, D.Y., 2015. Outdoor-indoor air pollution in urban environment: chal-

ev
lenges and opportunity. Frontiers in Environmental Science 2, 69.

Li, X.X., Liu, C.H., Leung, D.Y.C., Lam, K.M., 2006. Recent progress in

r
CFD modelling of wind field and pollutant transport in street canyons.
Atmospheric Environment 40, 5640–5658.
er
Li, Z., Ming, T., Shi, T., Zhang, H., Wen, C.Y., Lu, X., Dong, X., Wu, Y.,
pe
de Richter, R., Li, W., Peng, C., 2021. Review on pollutant dispersion in
urban areas-part b: Local mitigation strategies, optimization framework,
and evaluation theory. Building and Environment 198, 107890.
ot

Meng, X., Liu, C., Chen, R., Sera, F., Vicedo-Cabrera, A., Milojevic, A.,
Guo, Y., Tong, S., Coelho, M., Saldiva, P., Lavigne, E., Correa, P., Or-
tega, N., Osorio, S., Kyselý, J., Urban, A., Orru, H., Maasikmets, M.,
tn

Jaakkola, J., Ryti, N., Huber, V., Schneider, A., Katsouyanni, K., Analitis,
A., Hashizume, M., Honda, Y., Ng, C., Nunes, B., Teixeira, J., Holobaca,
rin

I., Fratianni, S., Kim, H., Tobias, A., Íñiguez, C., Forsberg, B., Åström,
C., Ragettli, M., Guo, Y.L., Pan, S.C., Li, S., Bell, M., Zanobetti, A.,
Schwartz, J., Wu, T., Gasparrini, A., Kan, H., 2021. Short term asso-
ep

ciations of ambient nitrogen dioxide with daily total, cardiovascular, and

respiratory mortality: multilocation analysis in 398 cities. BMJ-British
Medical Journal 372.
Pr

63

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Mirzaei, P., Haghighat, F., 2010. A novel approach to enhance outdoor

iew
air quality: Pedestrian ventilation system. Building and Environment 45,
1582–1593.

Mons, V., Margheri, L., Chassaing, J.C., Sagaut, P., 2017. Data assimilation-
based reconstruction of urban pollutant release characteristics. Journal of

ev
Wind Engineering and Industrial Aerodynamics 169, 232–250.

Mons, V., Wang, Q., Zaki, T.A., 2019. Kriging-enhanced ensemble varia-

r
tional data assimilation for scalar-source identification in turbulent envi-

er
ronments. Journal of Computational Physics 398, 108856.

Montazeri, H., Blocken, B., 2013. CFD simulation of wind-induced pres-

pe
sure coefficients on buildings with and without balconies: Validation and
sensitivity analysis. Building and Environment 60, 137–149.

Ng, W.Y., Chau, C.K., 2014. A modeling investigation of the impact of street
ot

and building configurations on personal air pollutant exposure in isolated

deep urban canyons. Science of The Total Environment 468-469, 429–448.
tn

Ntziachristos, L., Gkatzoflias, D., Kouridis, C., Samaras, Z., 2009. Cop-
ert: A european road transport emission inventory model, in: Information
Technologies in Environmental Engineering, Springer Berlin Heidelberg.
rin

pp. 491–504.

Peng, C., Li, C., Zou, Z., Shen, S., Sun, D., 2015. Improvement of air
ep

quality and thermal environment in an old city district by constructing

wind passages. Sustainability 7, 12672–12692.
Pr

64

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Peren, J.I., van Hooff, T., Leite, B.C.C., Blocken, B., 2015. CFD analysis

iew
of cross-ventilation of a generic isolated building with asymmetric open-
ing positions: Impact of roof angle and opening location. Building and
Environment 85, 263–276.

P.J. Richards, S.N., 2011. Appropriate boundary conditions for computa-

ev
tional wind engineering models revisited. Journal of Wind Engineering
and Industrial Aerodynamics 99, 257–266.

r
Pulvirenti, B., Baldazzi, S., Barbano, F., Brattich, E., Di Sabatino, S., 2020.

er
Numerical simulation of air pollution mitigation by means of photocat-
alytic coatings in real-world street canyons. Building and Environment
186, 107348.
pe
Ramponi, R., Blocken, B., De Coo, L.B., Janssen, W.D., 2015. CFD simu-
lation of outdoor ventilation of generic urban configurations with different
urban densities and equal and unequal street widths. Building and Envi-
ot

ronment 92, 152–166.

tn

Ribes, A., Caremoli, C., 2007. Salomé platform component model for nu-
merical simulation. Proceedings - International Computer Software and
Applications Conference 2, 553 – 564.
rin

Rodriguez, S., 2019. Applied Computational Fluid Dynamics and Turbulence

Modeling. Springer Nature Switzerland.
ep

Sagaut, P., 2006. Large Eddy Simulation for Incompressible Flows. Springer-
Verlag Berlin Heidelberg.
Pr

65

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Salim, S.M., Cheah, S.C., Chan, A., 2011. Numerical simulation of disper-

iew
sion in urban street canyons with avenue-like tree plantings: Comparison
between RANS and LES. Building and Environment 46, 1735–1746.

Selmi, W., Weber, C., Rivière, E., Blond, N., Mehdi, L., Nowak, D., 2016.
Air pollution removal by trees in public green spaces in strasbourg city,

ev
france. Urban Forestry & Urban Greening 17, 192–201.

Shen, J., Gao, Z., Ding, W., Yu, Y., 2017. An investigation on the effect

r
of street morphology to ambient air quality using six real-world cases.
Atmospheric Environment 164, 85–101.
er
Solazzo, E., Cai, X., Vardoulakis, S., 2008. Modelling wind flow and vehicle-
pe
induced turbulence in urban streets. Atmospheric Environment 42, 4918–
4931.

Stafoggia, M., Oftedal, B., Chen, J., Rodopoulou, S., Renzi, M., Atkinson,
ot

R., Bauwelinck, M., Klompmaker, J., Mehta, A., Vienneau, D., Ander-
sen, Z., Bellander, T., Brandt, J., Cesaroni, G., de Hoogh, K., Fecht,
tn

D., Gulliver, J., Hertel, O., Hoffmann, B., Hvidtfeldt, U., Joeckel, K.H.,
Jorgensen, J., Katsouyanni, K., Ketzel, M., Kristoffersen, D., Lager, A.,
Leander, K., Liu, S., Ljungman, P., Nagel, G., Pershagen, G., Peters, A.,
rin

Raaschou-Nielsen, O., Rizzuto, D., Schramm, S., Schwarze, P., Severi, G.,
Sigsgaard, T., Strak, M., van der Schouw, Y., Verschuren, M., Weinmayr,
G., Wolf, K., Zitt, E., Samoli, E., Forastiere, F., Brunekreef, B., Hoek,
ep

G., Janssen, N., 2022. Long-term exposure to low ambient air pollution
concentrations and mortality among 28 million people: results from seven
Pr

66

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
large european cohorts within the elapse project. Lancet Planetary Health

iew
6, E9–E18.

Streichenberger, B., Chakir, R., Jouy, B., Waeytens, J., 2021. Simulation and
validation of cfd turbulent airflow at pedestrian level using 3d ultrasonic
anemometer in the controlled urban area “sense-city”. Journal of Wind

ev
Engineering and Industrial Aerodynamics 219, 104801.

Tee, C., Ng, E., Xu, G., 2020. Analysis of transport methodologies for

r
pollutant dispersion modelling in urban environments. Journal of Envi-

er
ronmental Chemical Engineering 8, 103937.

Tilloy, A., Mallet, V., Poulet, D., Pesin, C., Brocheton, F., 2013. Blue-based
pe
no 2 data assimilation at urban scale. Journal of Geophysical Research:
Atmospheres 118, 2031–2040.

Tominaga, Y., Stathopoulos, T., 2007. Turbulent schmidt numbers for cfd
ot

analysis with various types of flowfield. Atmospheric Environment 41,

8091–8099.
tn

Tominaga, Y., Stathopoulos, T., 2011. Cfd modeling of pollution dispersion

in a street canyon: Comparison between les and rans. Journal of Wind
Engineering and Industrial Aerodynamics 99, 340–348.
rin

Tominaga, Y., Stathopoulos, T., 2013. Cfd simulation of near-field pollu-

tant dispersion in the urban environment: A review of current modeling
ep

techniques. Atmospheric Environment 79, 716–730.

Tominaga, Y., Stathopoulos, T., 2016. Ten questions concerning modeling

Pr

67

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
of near-field pollutant dispersion in the built environment. Building and

iew
Environment 105, 390–402.

Tominaga, Y., Stathopoulos, T., 2018. Cfd simulations of near-field pollutant

dispersion with different plume buoyancies. Building and Environment 131,
128–139.

ev
Vos, P., Maiheu, B., Vankerkom, J., Janssen, S., 2013. Improving local air
quality in cities: To tree or not to tree? Environmental Pollution 183,

r
113–122.

er
Waeytens, J., Sadr, S., 2018. Computer-aided placement of air quality sen-
sors using adjoint framework and sensor features to localize indoor source
pe
emission. Building and Environment 144, 184–193.

Wang, X., McNamara, K., 2006. Evaluation of cfd simulation using rans tur-
bulence models for building effects on pollutant dispersion. Environmental
ot

Fluid Mechanics 6, 181–202.

World Health Organization, 2016. Ambient air pollution: A global assess-

tn

ment of exposure and burden of disease.

Wu, A., Ren, W., 2020. TiO2 Nanoparticles: Applications in Nanobiotech-

rin

nology and Nanomedicine. John Wiley & Sons.

Xing, Y., Brimblecombe, P., 2019. Role of vegetation in deposition and

dispersion of air pollution in urban parks. Atmospheric Environment 201,
ep

73–83.
Pr

68

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615
 ed
Yang, J., Shi, B., Shi, Y., Marvin, S., Zheng, Y., Xia, G., 2020. Air pollution

iew
dispersal in high density urban areas: Research on the triadic relation of
wind, air pollution, and urban form. Sustainable Cities and Society 54,
101941.

Yassin, M.F., 2011. Impact of height and shape of building roof on air quality

ev
in urban street canyons. Atmospheric Environment 45, 5220–5229.

Yusuf, A., Oladipo, H., Yildiz Ozer, L., Garlisi, C., Loddo, V., Abu-Zahra,

r
M., Palmisano, G., 2020. Modelling of a recirculating photocatalytic

er
microreactor implementing mesoporous n-tio2 modified with graphene.
Chemical Engineering Science 391, 123574.
pe
Yusuf, A., Palmisano, G., 2021. Three-dimensional cfd modelling of a pho-
tocatalytic parallel-channel microreactor. Chemical Engineering Science
229, 116051.
ot

Zhao, Y., Jiang, C., Song, X., 2021. Numerical evaluation of turbulence
induced by wind and traffic, and its impact on pollutant dispersion in
tn

street canyons. Sustainable Cities and Society 74, 103142.

Zheng, X., Yang, J., 2021. Cfd simulations of wind flow and pollutant dis-
persion in a street canyon with traffic flow: Comparison between rans and
rin

les. Sustainable Cities and Society 75, 103307.

ep
Pr

69

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4572615