S I M U L AT I N G AT M O S P H E R I C D I S P E R S I O N W I T H I N T H E C I T Y

EINDHOVEN

A thesis submitted to the Delft University of Technology in partial fulfillment

of the requirements for the degree of

Master of Science in Environmental Engineering of Civil Engineering and Geosciences

by

Hannah Herbschleb

September 2021

Supervisors: Dr. Stephan de Roode (TU Delft)

Dr. Ir. Ruud Janssen (TNO)
Prof. Dr. Pier Siebesma (TU Delft)
Prof. Dr. Ir. Herman Russchenberg (TU Delft)

The work in this thesis was made in association with:

Climate, Air and Sustainability

TNO

Faculty of Civil Engineering and Geosciences

Delft University of Technology
Hannah Herbschleb: Simulating atmospheric dispersion within the city Eindhoven (2021)
Student number: 4312589
Abstract
The Dutch Atmospheric Large Eddy Simulation (DALES) model is used to investi-
gate the addition of an urban area to DALES and its impact on the dispersion of a
pollutant in non-neutral boundary conditions. Two different states of the boundary
layer are investigated, a stable boundary layer (SBL) and a convective boundary
layer (CBL), which are primarily achieved by a prescribed time-dependent surface
flux, negative for a SBL and positive for a CBL.

A module is added to DALES to provide the addition of heterogeneous emis-

sion sources. Several emission sources are implemented, one line source and three
point sources which locations depend on their location in the city. All sources emit
from the surface level. For the applied urban area, a part of Eindhoven is selected,
containing several relatively high buildings, a high way and a measurement cite.
Unfortunately, the results from these simulations could not be validated with the
measurements due to several model constraints.

The dispersion of a pollutant within a stable and a convective boundary layer

is first compared without buildings to the Gaussian plume model, to analyze and
attempt to verify the dispersion results obtained from DALES. The concentrations
of the pollutant in a SBL as calculated by DALES and the Gaussian plume model
are remarkably similar. The largest differences in concentration are at the source
and very close to the source, which is difficult to calculate for both models. The
differences between the CBL of DALES and the Gaussian plume model are more
significant, as the concentrations calculated by DALES are a factor 10 smaller than
obtained from the Gaussian plume model. The difference in concentration levels
between DALES’ SBL and CBL is almost a factor 20, demonstrating the importance
of turbulence on the dispersion.

The presence of an urban are influences the concentration of a pollutant severely,

regardless the stability of the atmosphere. The effect is larger in stable night-time
conditions, than convective day-time conditions. Due to the presence of buildings,
the concentrations of the pollutant is more concentrated at and around the source,
while at larger distances the concentration decrease with respect to simulations
without buildings. In a stable boundary layer, the surface concentrations up to a
downwind distance of 200m from the source are larger with the presence of build-
ings than without. In a convective boundary layer, the downwind distance from
which the surface concentration increases due to the presence of buildings is con-
siderably smaller: 50m, due to the larger turbulences present in the CBL. These
results could be highly dependent on the wind velocity, placing of buildings and
the source, vertical domain size, the lack of moisture, and turbulence. These fac-
tors require further investigation to advance the study of the implementation of an
urban area in DALES.

iii
Contents
1 introduction 3
1.1 Modeling the atmosphere . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Objectives and outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 dales 7
2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Prognostic variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Turbulent Kinetic Energy . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4.1 Subfilter-scale TKE . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4.2 TKE Budget Equation . . . . . . . . . . . . . . . . . . . . . . . . 10
2.5 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5.1 Surface flux model . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5.2 Immersed Boundary Method . . . . . . . . . . . . . . . . . . . . 12
2.5.3 Boundary conditions: lateral domain . . . . . . . . . . . . . . . 14
2.6 Advection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.7 Addition of heterogeneous emission sources . . . . . . . . . . . . . . . 15
3 gaussian plume model 17
3.1 Point source model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Line source model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2.1 Dispersion equations . . . . . . . . . . . . . . . . . . . . . . . . . 18
4 set-up of the simulations 21
4.1 Simulation time and domain size . . . . . . . . . . . . . . . . . . . . . . 21
4.2 Subsidence and inversion jump . . . . . . . . . . . . . . . . . . . . . . . 22
4.3 Surface temperature flux . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.4 Initial geostrophic wind velocity and advection scheme . . . . . . . . . 24
4.5 Subgrid Turbulent Kinetic Energy . . . . . . . . . . . . . . . . . . . . . 25
4.6 Surface boundary condition: implemented elevation map . . . . . . . 25
4.7 Source locations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.8 Differences between the GABLS1 and this study’s SBL . . . . . . . . . 27
5 analysis of atmospheric conditions 29
5.1 Establishing the simulations . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.1.1 Stable Boundary Layer (SBL) . . . . . . . . . . . . . . . . . . . . 29
5.1.2 Convective Boundary Layer (CBL) . . . . . . . . . . . . . . . . . 30
5.2 Effect of the addition of obstacles . . . . . . . . . . . . . . . . . . . . . . 32
5.2.1 Expectations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.2.2 SBL results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.2.3 CBL results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6 dispersion analysis 35
6.1 Comparison with the Gaussian plume model . . . . . . . . . . . . . . . 35
6.1.1 Surface concentrations . . . . . . . . . . . . . . . . . . . . . . . . 37
6.1.2 Vertical profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.1.3 Development over time . . . . . . . . . . . . . . . . . . . . . . . 40
6.1.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.2 Comparison with the urban area . . . . . . . . . . . . . . . . . . . . . . 41
6.2.1 Surface concentrations . . . . . . . . . . . . . . . . . . . . . . . . 43
6.2.2 Vertical profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.2.3 Development over time . . . . . . . . . . . . . . . . . . . . . . . 45
6.2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.3 Influence of a different advection scheme . . . . . . . . . . . . . . . . . 46

v
 6.3.1 Development over time . . . . . . . . . . . . . . . . . . . . . . . 48
6.3.2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
7 conclusions and recommendations 51
7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.1.1 Effect of the addition of obstacles . . . . . . . . . . . . . . . . . 51
7.1.2 Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
7.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
a inputfiles for the dales simulations 59
a.1 Settings - namoptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
a.2 Vertical profiles - prof.inp . . . . . . . . . . . . . . . . . . . . . . . . . . 61
a.3 Large scale forcings - lscale.inp . . . . . . . . . . . . . . . . . . . . . . . 61
a.4 Large scale fluxes - ls flux.inp . . . . . . . . . . . . . . . . . . . . . . . . 61
a.5 Additional inputfiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
b transition of gabls1 to sbl 63
c results of the point sources 67
List of Figures
Figure 1.1 Measured surface concentrations of NOx and PM10 . . . . . . 5
Figure 2.1 Immersed Boundary Method: schematic overview . . . . . . . 13
Figure 2.2 Example of the input file for the immersed boundary method 13
Figure 2.3 Boundary conditions: default DALES and new . . . . . . . . . 14
Figure 2.4 Schematic overview of nudging boundaries . . . . . . . . . . . 15
Figure 2.5 Example of the input file for source locations . . . . . . . . . . 16
Figure 2.6 Example of the emission rates for source locations . . . . . . . 16
Figure 4.1 Course of the modeled potential temperature . . . . . . . . . . 23
Figure 4.2 Vertical profile of the potential liquid temperature of the SBL
and CBL simulations . . . . . . . . . . . . . . . . . . . . . . . . 24
Figure 4.3 Initial vertical profile of the initial Subgrid-TKE values for
SBL and CBL simulations . . . . . . . . . . . . . . . . . . . . . 25
Figure 4.4 Locations of the measuring sites in Eindhoven and the simu-
lated area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Figure 4.5 Elevation map of Eindhoven: original and cleaned . . . . . . . 26
Figure 4.6 The locations of the scalars in the IBM domain. . . . . . . . . 27
Figure 5.1 Comparison of the GABLS1 simulations versus this study’s
SBL simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Figure 5.2 Vertical profiles of the CBL simulation of the buoyancy flux,
TKE, vertical velocity variance and the TKE budget. . . . . . . 31
Figure 5.3 Surface concentrations of the SBL simulation with obstacles . 32
Figure 5.4 Impact of the obstacles on the stable boundary layer, ex-
pressed with vertical profiles of several variables . . . . . . . . 33
Figure 5.5 Impact of the obstacles on the convective boundary layer, ex-
pressed with vertical profiles of several variables . . . . . . . . 34
Figure 6.1 Visualisation of the pollutant concentration of the DALES
and Gaussian plume SBL simulations. . . . . . . . . . . . . . . 36
Figure 6.2 Visualisation of the pollutant concentration of the DALES
and Gaussian plume CBL simulations. . . . . . . . . . . . . . . 37
Figure 6.3 Concentrations of the line sources in the downwind direction 38
Figure 6.4 Differences in simulated concentrations between SBL and CBL 38
Figure 6.5 Informative figure of the sources and analyzed downwind
distances form the source . . . . . . . . . . . . . . . . . . . . . 39
Figure 6.6 Vertical profiles of the concentration from the DALES and
Gaussian plume model at several downwind distances from
the line source . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Figure 6.7 Development of the concentration over time of DALES with-
out IBM and the Gaussian plume model . . . . . . . . . . . . . 41
Figure 6.8 Vertical and horizontal profiles of time-averaged concentra-
tions of DALES with obstacles and without . . . . . . . . . . . 42
Figure 6.9 Comparison with and without obstacles of concentrations of
the line sources in the downwind direction . . . . . . . . . . . 43
Figure 6.10 Vertical profiles of the concentrations within the SBL and
CBL, with and without IBM at several downwind distances
from the source . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Figure 6.11 Vertical distribution in percentages of concentrations of SBL
and CBL, with and without IBM . . . . . . . . . . . . . . . . . 45
Figure 6.12 Development of the surface concentration over time of DALES
of both the stable and convective boundary layer . . . . . . . . 45

vii
Figure 6.13 Vertical and horizontal profiles of time-averaged concentra-
tions of the Kappa advection scheme, in comparison to 2nd
order and Gaussian plume . . . . . . . . . . . . . . . . . . . . . 47
Figure 6.14 Surface concentrations of different advection schemes . . . . . 48
Figure 6.15 Vertical profile distribution of concentrations of the two dif-
ferent advection schemes . . . . . . . . . . . . . . . . . . . . . . 48
Figure 6.16 Development of surface concentrations over time of the sim-
ulation with the Kappa advection scheme . . . . . . . . . . . . 49
Figure B.1 Vertical profiles of the total buoyancy flux of the GABLS sim-
ulation in comparison to small alterations to the desired sim-
ulation state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Figure B.2 Vertical profiles of the total TKE and the SFS-TKE of the
GABLS simulation in comparison to small alterations to the
desired simulation state . . . . . . . . . . . . . . . . . . . . . . 65
Figure B.3 Vertical profiles of the ratio between the total TKE and the
SFS-TKE of the GABLS simulation in comparison to small
alterations to the desired simulation state . . . . . . . . . . . . 65
Figure C.1 Vertical and horizontal profiles of time-averaged concentra-
tions of the SBL as calculated by the Gaussian plume model
and DALES with obstacles and without . . . . . . . . . . . . . 68
Figure C.2 Vertical and horizontal profiles of time-averaged concentra-
tions of the CBL as calculated by the Gaussian plume model
and DALES with obstacles and without . . . . . . . . . . . . . 69
Figure C.3 Concentrations of the point sources in the downwind direction 70
Figure C.4 Vertical profiles of the concentration from the DALES and
Gaussian plume model at several downwind distances from
the point source . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

1
1
Introduction
The presence of industries providing our way of life has severely impacted the
earth in the past 100 years, resulting in the current climate crisis. One of the most
important aspects of the climate crisis is the state of the atmosphere, specifically
air pollution and the dispersion hereof. The increasing air pollution in urban areas,
which are inhabited by the majority of the world population, is a growing problem
[WHO, 2016]. To gain insight into the current situation and estimate the impact of
possible changes to the urban layout or emission patterns, atmospheric models are
required to accurately simulate the lower part of the atmosphere (the Atmospheric
Boundary Layer) on a scale smaller than 10 meters between two grid points.

The atmospheric boundary layer is the part of the atmosphere influenced by the
surface, and follows a diurnal cycle. During the day, the atmospheric boundary
layer depth increases and typically grows to several kilometers. The sun warms
the Earth’s surface causing convection. The resulting turbulence increases the mix-
ing layer height, which is the level at which the atmospheric boundary layer meets
the free atmosphere. This turbulence impacts the dispersion strength of particles,
especially the vertical dispersion. During the night, the cooling of the ground sur-
faces causes a strong reduction of turbulence, resulting in a lower, stable boundary
layer which typically reaches around the 200m. The remaining turbulence is mainly
caused by the wind shear in the horizontal direction. In a boundary layer without
obstacles like buildings, the direction and the strength of the dispersion is mostly
dictated by the wind direction and turbulence. The addition of an urban area will
influence this dispersion strength.

1.1 Modeling the atmosphere

Many studies simulating the atmosphere are based upon computational fluid dy-
namics (CFD). There are several different implementations of CFD. In 2019, Yang
et al. proposed a model to investigate the influences between urban form, urban
air pollution, and the wind environment using a Reynolds-averaged Navier-Stokes
model (RANS). This method is based on statistical averaging of turbulence, and as-
sumes the turbulent diffusion to be proportional to the fields’ mean concentration.
This leads to relatively fast and steady, two-dimensional equations.

3
 Another widely used type of CFD is Large-Eddy Simulation (LES), which is based
on resolving turbulence using local filtering. It is able to resolve turbulent scales
with a filter width1 between 1 and 50 meters [Heus et al., 2010]. LES can resolve
up to 90% of all turbulence. This method can calculate the turbulence in 3D, and
therefore, it is more time-expensive than RANS.

Fortunately, multiple comparisons between RANS with LES with respect to the
dispersion of pollutants in urban areas have been performed [Jia and Kikumoto,
2021; Yang et al., 2019; Tominaga et al., 2008]. All concluded that LES is more accu-
rate in terms of concentration’s strength and location. Therefore, this study will not
take a RANS model into account. The LES model this study uses is the fourth ver-
sion of the Dutch Atmospheric Large-Eddy Simulation model (DALES4). A more
in-depth description of DALES is given in chapter 2.

Several studies have investigated the processes and physics in LES models of
urban areas. Walton and Cheng modeled an urban area in steady-state, neutral
conditions, where the implemented LES model lacked the complexity provided by
DALES in terms of the governing turbulence equations. Others have investigated
flow and dispersion in urban areas with the LES model PALM (Parallelized Large-
Eddy Simulation Model), taking a stable atmosphere and radiation into account.
However, these versions of the PALM model have a 2D urban model and energy
balance solvers [Letzel et al., 2008; Xie et al., 2008, 2013; Resler et al., 2017], whereas
Grylls et al. is currently working or a more complex version of simulating an urban
area in DALES4, uDALES. This version was not available for this study, and a pre-
vious version of this model is applied which is an model based on the findings of ?
implemented by [Koene, 2020].

Verzijlbergh et al. [2009] has investigated the dispersion of pollutants within

DALES, comparing cloudy and non-cloudy conditions. However, the dispersion
in urban areas in non-neutral conditions is not yet investigated. The objective of
this study is to simulate the dispersion of a pollutant in an convective and sta-
ble boundary layer, with and without structures, to investigate the capabilities of
DALES to simulate an urban area. The main research question is:

What is the influence of obstacles on the dispersion of a pollutant within Eindhoven, in

stable and convective boundary layers?

This study will not take any form of moisture into account, for example clouds
or precipitation. The only varying aspects of the simulations are the addition of the
obstacles and the stability of the atmosphere; all other variables are constant when
differentiating between simulations.

1.2 Objectives and outline

The selected area is the Dutch city Eindhoven, which has had problems with air
pollutions due to its highway through the city, the nearby airport, and the city’s
agriculture [Ascenso et al., 2021]. The Dutch National Institute for Public Health
and the Environment (RIVM) has been measuring the particulate matter (PM) and
the NOx concentrations across the country for several years, including certain loca-
tions within Eindhoven. The topology implemented in the DALES simulations will
be based on one of these locations. Figure 1.1 gives the hourly surface concentra-
tions of days with clear-sky conditions, measured at one of these sites in Eindhoven.

1 The filter width is a function of the grid size

Ideally, the results obtained from the DALES simulations are compared against the
measurements, to statistically determine the accuracy of the simulations. This study
is a small step in that direction.

(a) (b)

Figure 1.1: Measured surface concentrations of NOx and PM10 of all days meeting the re-
quirements, and the hourly average.

The approach of this study is to first validate the simulated stable and convective
boundary layer with the help of a widely used dispersion model, the Gaussian
plume model, and investigate the effect of the added urban area by comparing the
results from the DALES simulations with and without buildings.
First, chapter 2 elaborates on the current version of DALES and the additions
made for this study, followed by an explanation of the Gaussian plume model in
chapter 3. Chapter 4 gives insight into the set-up of all simulations by elaborat-
ing on the different prescribed variables and settings. The results are divided in
two chapters, first the atmospheric conditions are analyzed to establish a proper
simulation of both the stable and convective boundary layer in chapter 5. The dis-
persion of all simulations is analyzed in chapter 6, starting with the stable boundary
layer, with and without obstacles, followed by the results of the convective bound-
ary layer, and finishing with a comparison between the two. The conclusions and
recommendations are given in chapter 7.

5
2
DALES
First, this chapter will give insight into the background of DALES, the prognostic
variables, and the governing equations it uses, in respectively section 2.1, 2.2 and
2.3. This is followed by the turbulence kinetic energy (TKE) calculations in sec-
tion 2.4, and an explanation of the boundary conditions of DALES, the immersed
boundary layer and the lateral domain in section 2.5. Finally, section 2.6 gives the
implemented advection scheme and section 2.7 describes the additional model for
the implementation of multiple sources at different locations.

2.1 Background
Results of DALES simulations were first published in 1986 and is based on the pio-
neering LES modeling principles of Lilly [1967], Deardorff [1974], Sommeria [1976],
and Nieuwstadt and Brost [1986]. Especially the latter forms the base of the current
LES-codes of DALES, written in the language Fortran95. Since, DALES has had
many contributors, which resulted in the current version, DALES4, which is main-
tained and further developed by researchers of the Technical University of Delft,
the Royal Netherlands Meteorology Institute (KNMI), the University of Wagenin-
gen, and the Max Planck Institute for Meteorology [Heus et al., 2010].
DALES consists of a multitude of modules that perform specific parts of the
calculations. All the modules are called upon when needed from the main module,
which can be run on command. This clear set-up enables the user to easily add
or modify a module. The simulations require several input files which contain the
initial and boundary conditions of the simulated environment and the required
settings for the simulation. While all the input files and most of the output is
ASCII files, the larger datasets are stored in Network Common Data Form (NetCDF)
version 3 or higher.

7
2.2 Prognostic variables
DALES takes several time-dependent prognostic variables as input which determine
the course of the simulations. These variables are the three velocity components ui
in x, y, and z direction, the total water specific humidity qt , rain water specific
humidity qr , the liquid potential temperature θl , rain droplet number concentration
Nr and a maximum of 100 passive or active scalar φ. The wind in x-direction is
specified from west to east and in the y-direction from south to north [Heus et al.,
2010]. The total water specific humidity is defined as:

qt = qv + qc , (2.1)

where the total water specific humidity is equal to the sum of the water vapor
specific humidity qv and the cloud liquid water specific humidity qc . Given that
there will not be any moisture present in any of the simulations, the values of qv
and qc are:

qv = 0, (2.2)
qc = 0 (2.3)

making the value of the total water specific humidity equal to zero (qt = 0).
The liquid water potential temperature and the liquid water virtual potential
temperature are calculated with the following equations:

L
θl ≈ θ − qc (2.4)
c pd Π
!   
L Rv Rv
θv ≈ θ− qc 1− 1− qt − qc (2.5)
c pd Π Rd Rd

where θ, which is a function of the temperature T = θΠ, is the potential tempera-

ture, L and c pd are constants of the latent and specific heat respectively and Π is
a function of the pressure (the Exner function). Rd and Rv are respectively the gas
constants for dry air and water vapor, with Rd = 287.0 Jkg−1 K−1 and Rv = 461.5
Jkg−1 K−1 . Due to the lack of moisture (with qt = qv = qc = 0) equation 2.5 results
in θv ≈ θl ≈ θ.

2.3 Governing Equations

DALES core equations are based on the Navier-Stokes equations with the Boussi-
nesq approximation with a reference state (ρ0 , p0 , θ0 ). The equations of motion are
given by the continuity equation (2.6), the Boussinesq approximation of the Navier-
Stokes equation (2.7), and the filtered transport equation for scalars (2.8).

∂uei
= 0, (2.6)
∂xi
∂uei ∂uei uej ∂π g ∂τij
=− − + θev δi3 + Fi − , (2.7)
∂t ∂x j ∂xi θ0 ∂x j
∂ϕ ∂ue ϕ
e ∂Ru j ,ϕ
=− i −
e
+ Sϕ (2.8)
∂t ∂x j ∂x j

The tildes represent the filtered mean variables, Fi the large scale forcings, and with
pe
the modified pressure π = φ0 + 32 e with pe the filtered pressure, ρ0 the reference state
density and e the subfilter-scale turbulence kinetic energy (SFS-TKE). τij represents
 ei uej − 23 e, which is part of the subgrid momentum flux, which is parame-
i , uj − u
u]
∂2 π ∂
terized. π is computed with a Poisson equation of ∂xi2
, where the divergence ∂xi
is taken of equation 2.7. Combined with the continuity equation 2.6, the left-hand
side of equation 2.7, ∂∂tuei is equal to zero. This results in the following equation:
!
∂2 π ∂ ∂uei uej g ∂τij
= − + θev δi3 + Fi − . (2.9)
∂xi2 ∂xi ∂x j θ0 ∂x j

With π and τij , DALES can calculate the resolved and subgrid fluxes due to eddies,
which is elaborated in section 2.4.
Similarly, in the transport equation for scalars ϕ, the Ru j ϕ represents the SFS scalar
fluxes by ugj, ϕ − u e. S ϕ denotes the source terms of the scalar.
ej ϕ

2.4 Turbulent Kinetic Energy

The turbulent kinetic energy (TKE) is a useful parameter to analyze the turbulence
in a boundary layer. The total TKE consists of the resolved TKE and the subfilter-
scale turbulent kinetic energy (SFS-TKE). All simulations must be designed in terms
of required mesh size and resolution, such that the total TKE is not dominated by
the SFS-TKE, and most is resolved.
The resolved TKE is best analyzed with the help of the TKE budget equation.
Section 2.4.2 gives an explanation of the equations and implementations. The cal-
culations of the subfilter-scale TKE are key closure equations of DALES, where the
turbulent kinetic energy a of a scale smaller than the filter width are parameterized,
which is elaborated below.

2.4.1 Subfilter-scale TKE

The SFS fluxes in DALES are parameterized with the help of a downgradient eddy
diffusivity approach:

∂ϕ
Ru j ,ϕ = −Kh
e
, (2.10)
∂x j
!
∂uei ∂uej
τij = −Km + (2.11)
∂x j ∂xi

where Kh and Km are the eddy viscosity and diffusivity coefficients. These can
be modeled in DALES by two methods: as a function of SFS-TKE e proposed by
Deardorff [1974] or the Smagorinsky closure [Smagorinsky, 1963]. This study will
only focus on the SFS-TKE model of Deardorff, where ∂e
∂t is a function of Ru j ,ϕ and
τij :

∂e ∂uej e ∂ue g ∂Ru j ,e 1 ∂Ru j ,π

=− − τij i + Rw,θv − − −ε (2.12)
∂t ∂x j ∂x j θ0 ∂x j ρ0 ∂x j

where ε represents the SFS-TKE dissipitation rate. The first and the second terms
on the right-hand-side are resp. solved and calculated with equation 2.11. The third
term represents the buoyancy SFS-TKE production, which is calculated with

g g
R = ( ARw,θl + BRw,qt ) (2.13)
θ0 w,θv θ0

9
where A and B are dependent on a dry or moist local thermodynamic state. As
stated before, only a dry atmosphere will be simulated in this study, resulting in
the following equations for A and B:
Rv
A = 1+ R qet (2.14)
d
 
B= R R − 1 θ0
v
(2.15)
d

where θ0 is the reference potential temperature. The remaining terms of equation

2.12 are together parameterized with:
  !
∂ 1 ∂ ∂e
− Ru j ,e + Ru j ,π = 2Km (2.16)
∂x j ρ0 ∂x j ∂x j

Kh and Km are parameterized after Deardorff [1980] as shown in the equations

below. These equations take several constants, which are given in table 2.1.
Km = cm λe1/2 (2.17)
1/2
Kh = ch λe (2.18)
with
  3
cf 3 −2
cm = α (2.19)
2π 2
 
λ
ch = 1 + 2 cm (2.20)
∆
and λ depending on the grid size ∆ = (∆x · ∆y · ∆z)1/3 , and the stability of the flow
g θev 2
 
N = θ0 ∂∂z :
!
e1/2
λ = min ∆, c N . (2.21)
N

α cm cf cN
1.5 0.12 2.5 0.76

Table 2.1: Constants in equations 2.17 - 2.21

Kh
In the case of a very stable boundary layer, λ << ∆, resulting in Km ≈ 1 and in
Kh
the case of neutral or convective conditions, Km = 3.

2.4.2 TKE Budget Equation

The turbulent kinetic energy budget E gives insight into the contribution of which
turbulence term to the total kinetic energy. This can be calculated by applying
Reynolds decomposition to the Boussinesq approximation of the Navier-Stokes
equation (2.7). The total tendency of the TKE is then given by:
   h  i
∂E ∂ 1 2 2 2
≡ ue” + ve” + w”
∂t 2
e
∂t
∂huei ∂hvei
 
g g
= − hue”w”e i + hve”w”
e i + hwθv i
∂z ∂z θe0
| {z } | {z } (2.22)
shear production buoyancy production

∂hw”E
f i ∂hw”π”
^i
− − − hε τ i
| ∂z
{z } | ∂z
{z } |{z}
viscous dissipation
turbulent transport pressure correlation
Here, the double prime ” denotes the deviation from the slab-average indicated by
hi, and the dissipation term is given by:

" #!
∂ ∂uei ” ∂uej ”
ε τ = uej ” Km + (2.23)
∂x j ∂x j ∂xi

where Km is the eddy diffusivity. All these different terms are evaluated at different
positions due to DALES’ staggered grid, requiring interpolations. Therefore the
TKE budget is not entirely closed, leaving a residual. Fortunately, the implemented
budget equations and interpolation techniques are well-defined, and therefore the
residual is insignificant.

2.5 Boundary Conditions

2.5.1 Surface flux model

To enable the exchange between scalars at the surface and the atmosphere, and
to parameterize the turbulent drag, DALES requires a surface model. The surface
f −w
fluxes here are denoted by Fs,φ = wφ eφe of arbitrary variable φ. This differs from
the previously defined value R, since the fluctuations in the vertical velocity at the
surface are zero. As given in Heus et al. [2010], this model is based on the following
equations:

Fs,u = −C M hU1 iue1 (2.24)

Fs,v = −C M hU1 ive1 (2.25)
Fs,ϕ = −C ϕ hU1 i( f
ϕ1 − ϕ0 ) (2.26)

where equations 2.24 and 2.24 calculate the momentum fluxes along the horizontal
wind vectors, and equation 2.26 the scalar flux. C M and C ϕ are the drag coefficients,
which are calculated with

u2∗0
CM = (2.27)
hU1 i2
u2∗0 ϕ∗0
Cϕ = . (2.28)
hU1 ih f
ϕ1 − ϕ0 i

All variables withp hφi i represent the horizontally averaged gradient at the model
level i, hU1 i = hue1 i2 + hve1 i2 , u∗0 is the averaged friction velocity and ϕ∗0 the
scalar scales. u∗0 is unknown beforehand and therefore the Monin-Obukhov simi-
larity theory is applied. This relates the bulk Richardson number (Ri B ), which can
be calculated with equation 2.30, to the Obukhov length (L) of the lowest layer of
the atmosphere between the surface to z1. The values of u∗0 and ϕ∗0 can be calcu-
lated with the help of these parameterizations. The applied equations are:

11
 h
  i
z1 z1
z0h

z1 ln z0h − ψ H L + ψ H L
Ri B = i2 (2.29)
L
h  
ln zz0m − ψM zL1 + ψM z0m

1


L
 
g 1z h θ
f v1 i − h θ
f v0 i
Ri B = (2.30)
θ0 hU1 i2
u3∗0
L=− g (2.31)
κ hθ h Fs,θv i
v0 i

h Fs,φ i
ϕ ∗0 = − (2.32)
u ∗0

where z0 m and z0 h denote the roughness length from momentum and heat, ψ M
and ψH are integrated stability functions. hθf
v0 i and h θv1 i are the spatially averaged
f
filtered virtual potential temperature of the surface level and the first level of the
model, respectively and κ is the Von Karman coefficient.

There are four different options for this surface model in DALES, differentiating
in complexity and required input, where the surface scalar fluxes, scalar values,
and u∗0 are prescribed or parameterized as described above, or with a Land Sur-
face Model as input. The possibility to add obstacles to the model (the Immersed
Boundary Method) is currently only implemented for a surface model where the
surface scalar and momentum fluxes are parameterized and the scalar fluxes at the
surface can be prescribed (in DALES option 2).

2.5.2 Immersed Boundary Method

The Immersed Boundary Method is a technique that models the surface boundary
conditions in fluid dynamics simulations with obstacles. This method was first
implemented in DALES by Tomas et al. [2016] and is further developed by a TU
Delft Masters Graduate Koene [2020].

Methodology

Every time step, the Immersed Boundary Method first calculates the flow without
any obstacles, subsequently changing the flow properties at the wall. First, the
modeled velocities perpendicular to the wall of the object, in the adjacent grid cell
ui,j,k , are changed to zero, by imposing a counterforce that forces the momentum
tendencies to zero, which is called direct forcing. On flows parallel to the wall, the
presence of a wall has a shear effect.
Consider a grid point with its cell center at location (i, j, k), adjacent to an obstacle
in the direction of the flow (here u), as shown in figure 2.1.

Figure 2.1: A schematic overview of two adjacent grid cells, with an obstacle present at
location (i − 1, j, k), and flow in u-direction. Modeled after a figure from Koene
[2020]

The velocities parallel to the wall, v and w, experience shear stress. Here, the tur-
bulent diffusions between the two cells, which would have been modeled without
an obstacle, are replaced with the wall shear stress. It is applied to both vi,j,k and
vi,j+1,k at the edges of the grid cell, and similarly to wi,j,k and wi,j,k+1 . The shear
stress is calculated with

" # 2
 1+ B B 1+ B
1 − B 1+ B
 
ν 1+B ν
|τ | = ρ A 1− B + |utan | (2.33)
2 ∆xi A ∆xi

with
ν 2
|utan | = A 1− B (2.34)
2∆xi

where A = 8.3, B = 17 and the kinematic viscosity ν is denoted with that of air,
which equals 1.41 × 10−5 m2 s−1 .

Implementation

To implement the height of the obstacles, DALES requires an input file containing
a 2D matrix with the grid size of the surface (itot ×jtot ). This matrix contains the
elevation from the surface of every grid point. It is programmed such that DALES
sees these elevations as the top of the building, and vertically models every under-
lying grid point as an obstacle. This elevation is scaled to fit the vertical resolution.
Figure 2.2 gives an example of an input-file for an immersed boundary layer with
two square objects. With an ∆z = 5m, the upper left object translates to a building
with a height of 5m, and the lower right 10m. The IBM currently implemented in
DALES models stationary, impermeable obstacles.

Figure 2.2: An example of the input file for the implemented objects with the immersed
boundary method.

13
2.5.3 Boundary conditions: lateral domain

The boundary conditions of the sides and top of the domain in DALES are im-
plemented as periodic boundary conditions; what leaves one side of the domain,
reappears on the opposite side. This is illustrated in 2D in figure 2.3a. Periodic
boundary layers are undesirable in this study, due to the fact that a concentration
could then be present or even accumulating in behind the source, considering the
direction of the flow. Therefore, a small piece of code is added to DALES following
a method of Dorp [2016], where an additional nudging variable is implemented
which smoothly nudges the tendencies of the scalars to zero, from a variable dis-
tance from the boundaries, defined as the nudging depth dnudge . Figure 2.3b gives
a schematic overview of a nudged domain.

(a) (b)

Figure 2.3: The boundary conditions as implemented in DALES with in (a) the original pe-
riodic boundary conditions and (b) the nudged boundary conditions. Modeled
after a figure of Dorp [2016].

The nudging variable decreases the tendencies of the concentration of source φ:

∂φ original
the original tendency before the nudging ∂t is multiplied with the nudging
value f nudge to obtain the nudged tendency. f nudge increases gradually from 0 to 1
when a grid cell nears a boundary. Equations 2.35 and 2.36 give the implemented
equations in DALES.
nudged original
∂φ ∂φ
= (1 − f nudge ) ∗ (2.35)
∂t ∂t

where f nudge is a matrix of size (itot × jtot ), where all values with a distance to a
boundary larger than dnudge are zero. The entries inside the nudging depth distance
from the boundary are calculated with:
!
1 1 π
f nudge = 2 + 2 cos i . (2.36)
dnudge

where i denotes the absolute distance in grid points from the closest boundary.
When taken dnudge = 5, figure 2.4 gives the calculated values of the domain. A
cosine function is implemented to create a nudging gradient from 1 to 0 rather than
a steep line, to prevent computational errors.

2.6 Advection
DALES gives the option of five different advection schemes. All advection schemes
operate on with a global equation for advection in the x-direction is given as:

δuei φi Fi+ 1 − Fi− 1

2 2
= , (2.37)
δx ∆x
Figure 2.4: The calculated values of f nudge at the edges of a domain with a nudge depth
dnudge = 5.

where Fi− 1 represents the convective flux of φ through the plane i − 12 perpendicular
2
to the wind velocity uei . The values of the convective flux Fi− 1 are calculated by the
2
one of the available advection schemes. In these equations φ can represent the
e e1/2 or ϕ
variables ue, ve, w, e. The IBM module is currently only implemented for a
second order advection scheme, for the reason that it is a relatively simple advection
scheme:

φi + φi−1
Fi2nd
−1
= uei− 1 . (2.38)
2 2 2
The central difference in this equation can lead to negative concentrations, which
lead to the addition of the Kappa advection scheme to DALES to prevent this prob-
lem. This scheme is modeled after the work of Vreugdenhil and Koren [1993], and
is given as:

 
1
Fiκ− 1 = uei− 1 φi−1 + κi− 1 (φi−1 − φi−2 ) . (2.39)
2 2 2 2

In this equation, κi− 1 is defined as a switch that takes the magnitude of the up-
2
wind gradient of φ into account: a third-order upwind scheme when small, and
first-order when stronger [Heus et al., 2010]. However, the IBM is not yet imple-
mented for this advection scheme and therefore will only be used for a comparisons
without IBM.

2.7 Addition of heterogeneous emission sources

The current version of DALES only allows the application of pollutants to be ho-
mogeneous, indicating that the predefined source flux will be emitted from every
surface point at an equal rate. To investigate the effect of objects on the disper-
sion in a stable boundary layer and convective boundary layer, a non-homogeneous
source is preferred. A point or line source allows the results to show the vertical
and horizontal reach of the pollutant.

Therefore, an additional module is written into DALES. The main purpose of this
module is to read an input file similar to to input file needed for the IBM, where
every source can be prescribed to grid point at the surface level. Consequently,
the module prescribing the source fluxes to grid points is altered. Previously, the
emission of the source was prescribed to every location at every timestep. To only
prescribe the emission to the specified location, a conditional function is added. To
explain the details of this function, first the outline of the input file needs to be
described.

15
The required input file consist of a matrix with size (itot × jtot ) representing the
surface area expressed in grid points, where every location ai,j contains an integer
in between zero and the total number of different sources. This can be defined in the
settings for the simulations. This method allows the user to define the desired area
of emissions, for all points with the same number represent the same scalar. Figure
2.5 gives an example of a part of an input file, where three sources are defined: two
point sources (1 and 2) and one line source (3).

Figure 2.5: An example of the input file for the source locations.

The source fluxes are prescribed per source, per surface grid point in multiple
loops. As explained before, with heterogeneous sources, every source is prescribed
to every grid point. The input file provides a matrix where every source has a
specific integer, and grid points without source are equal to zero, as seen in figure
2.5. The added conditional function only prescribes the source flux to grid point
matching the index of the source flux. For example, the are 3 source fluxes at the
locations given in figure 2.5, with flux(1) = 3 g g−1 ms−1 , flux(2) = 2 g g−1 ms−1 , and
flux(3) = 7 g g−1 ms−1 . The first flux will only be prescribed to the grid points which
have a value of 1 in the input file, etc. and the emission rates (in g g−1 ms−1 ) per
grid point will be as given in figure 2.6.

Figure 2.6: An example of the emission rates in [g g−1 ms−1 ] for the source locations match-
ing the example in figure 2.5.
3
Gaussian Plume Model
The goal of this study is to analyse the dispersion within an urban area of DALES
simulations in non-neutral conditions. Unfortunately, verification with measure-
ments is not yet possible and therefore a different approach in an attempt of val-
idation is applied. The dispersion results of the simulations in DALES without
obstacles are compared to the dispersion results of a widely used model: the Gaus-
sian plume model. The influence of the objects is measured in comparison with
the DALES simulations without obstacles. This chapter gives an explanation of the
Gaussian plume model and an overview of the used equations and parameters.

A study by Pirhalla et al. [2021] compared the dispersion results of a Gaussian

plume model with LES results and measurements, and concluded that close to the
source (5 - 10m) the Gaussian plume model is not as accurate as the LES model,
however, beyond that distance, the results of both are highly accurate, even though
LES slightly underestimated the peak concentration.
The Gaussian plume model is a highly idealized model for homogeneous terrain
conditions without obstacles, to estimate the downwind concentrations of an emis-
sion source. The investigated sources are one line source and three point sources.

3.1 Point source model

The Gaussian plume model for a point source is described with the following equa-
tion:
!
y2 ( z − H )2 ( z + H )2
   
Q
χ( x, y, z) = exp − 2 exp − + exp − . (3.1)
2πuσy σz 2σy 2σz2 2σz2

Here, χ( x, y, z) is the concentration at distance x downwind of the source, and

cross-wind distance y and height z. Q is the intensity of the source in mass per
unit time, u the wind velocity, H the height of the release. The release height H in
all simulations will be zero, as will the height z in all calculations, resulting in the
following equation:
!
y2 z2
 
Q
χ( x, 0, 0) = exp − 2 2exp − 2 . (3.2)
2πuσy σz 2σy 2σz

The most important variables here are σy and σz , which are resp. the lateral and
vertical dispersion coefficient. In the literature, there are multiple approaches to

17
these variables, several studies have compared these methods on the basis of their
performance in rural and urban areas. The best performing method according to
the literature is a combination of different approaches [Mao et al., 2020; Wu and
Liu, 2018; Essa et al., 2011; Carrascal et al., 1993]. A power-law, distance dependent
approach is used for the vertical dispersion, and the lateral dispersion is based on
an angular half-width of the plume. Section 3.2.1 elaborates on the implemented
equations and parameters for these variables.

3.2 Line source model

The line source Gaussian plume Model is approached in the same manner as the
implementation of the line source in DALES; as several point sources in formation.
This enables the possibility to match the LES-modelled line source exactly.

3.2.1 Dispersion equations

The used equations for σy and σz in the implemented Gaussian plume model are as
follows:

σz = cx d (3.3)
σy = 465.116x · tan(θ ) (3.4)

where θ is defined with the following equation:

θ = 0.01745( a − b · ln( x )). (3.5)

All parameters are dependable on the Pasquill stability, class ranging from very
unstable or convective (A) to very stable (F), and parameters c and d are addition-
ally dependent on the downwind distance x in km of the source. The values of
these parameters are given table 3.1 and 3.2 below. The implemented values are
corresponding with the values of stability class A for the convective boundary layer
simulations and E for the stable boundary layer. Here, the stability class E is chosen
and not F, because class F contains very little to no turbulence and is mostly reliant
on small scale eddies. The Large-Eddy Simulation method is therefore not able to
properly calculate these small eddies.

Parameters
Class Stability a b
A Very Unstable 24.1670 2.5334
B Unstable 18.3330 1.8096
C Neutral/Slightly unstable 12.5000 1.0857
D Neutral/Slightly stable 8.3330 0.7238
E Stable 6.2500 0.5429
F Very stable 4.1667 0.3619

Table 3.1: Coefficients for equations 3.4 and 3.3, based on Singer and Smit (1968).
 Class x (km) c d
0.10-0.15 158.08 1.0542
0.16-0.20 170.22 1.0932
0.21-0.25 179.52 1.1262
A 0.26-0.30 217.41 1.2644
0.31-0.40 358.89 1.4094
0.41-0.50 346.75 1.7283
0.50-3.11 453.85 2.1166
> 3.11 σz = 5000

0.10-0.20 90.673 0.93198

B 0.21-0.40 94.483 0.98332
> 0.40 109.300 1.09710

C > 0.10 61.141 0.91465

0.10-0.30 34.459 0.86974

0.31-1.00 32.093 0.81066
1.01-3.00 32.093 0.64403
D 3.01-10.00 33.504 0.60486
10.01-30.00 36.650 0.56589
> 30 44.053 0.51179

0.10-0.30 23.331 0.81956

0.31-1.00 21.628 0.75660
1.01-2.00 21.628 0.63077
E 2.01-4.00 22.534 0.57154
4.01-10.00 24.703 0.50527
10.01-20.00 26.970 0.46713
20.01-40.00 35.420 0.37615

0.10-0.20 15.209 0.81558

0.21-0.70 14.457 0.78407
0.71-1.00 13.953 0.68465
F 1.01-2.00 13.953 0.63227
2.01-3.00 14.823 0.54503
3.01-7.00 16.178 0.46490
7.01-15.00 17.836 0.41507
15.01-30.0 22.551 0.32681

Table 3.2: Coefficients for equations 3.4 and 3.3, based on Cramer (1979).

19
4
Set-up of the Simulations
The dispersion of emissions within DALES is simulated in two different boundary
layer conditions: a convective boundary layer (CBL) representing the day, and a sta-
ble boundary layer (SBL) representing the night. Both simulations are performed
with and without the addition of an urban area, to investigate the effect hereof on
the boundary layer, and examine the possible differences in their impact on a SBL
and CBL.

This chapter elaborates on the setup of the different simulations and the imple-
mented initial boundary conditions. First, the simulations time and chosen domain
size are described in section 4.1, followed by the inversion jump (section 4.2) and
the virtual potential surface temperature fluxes, and the initial vertical profiles of
the potential temperature (section 4.3). The initial geostrophic wind velocity and ap-
plied advection scheme are discussed in section 4.4, and the initial vertical profiles
of the subgrid-TKE are discussed for both the stable and the convective boundary
layer in section 4.5. Section 4.6 and 4.7 show the implemented elevation map and
the source locations, respectively. This chapter concludes with an elaboration of the
case whereafter the SBL simulation is modeled, and the differences between the two.

The stable boundary layer is modeled after the GABLS1 study, which stands for
Global Energy and Water Cycle Experiment Atmospheric Boundary Layer Study,
and simulates a verified stable boundary layer [Beare et al., 2006]. The GABLS1
study investigated different models simulating a SBL at different resolutions, and
compared the outcomes. All input files and settings are available in DALES, and
therefore this study has been chosen to use as a starting point for simulating a
reliable SBL. Furthermore, the GABLS1 case is simulated in DALES and used as a
validation model for the SBL simulation without obstacles. Therefore, their exact
variables and the differences between this study’s SBL simulation and the GABLS1
simulation will be described in section 4.8. The convective boundary layer is mainly
achieved through the addition of a positive surface temperature flux.

4.1 Simulation time and domain size

The initial simulation time and horizontal domain size and resolutions are equal for
all simulations, whereas the vertical domain size and resolution differs per stability.
Firstly, all simulations should produce a sufficient amount of output to properly

21
study all the data. As most simulation models, DALES requires some initialization
time before it reaches a steady state, which is generally two hours. A total simula-
tion time of eight hours will provide sufficient data.

The horizontal domain size of a convective boundary layer often requires a rela-
tively large number of grid points, and should be able to provide sufficient surface
area for numerous obstacles. Therefore the number of grid points in both x-, and
y-direction is chosen to be 160. The implemented elevations of obstacles in this
study are based on an elevation map with a horizontal resolution of ∆x = ∆y = 5m.
This results in a horizontal domain size of 800 × 800m.

The vertical resolution depends on the size of the simulated eddies. A stable
boundary layer contains relatively smaller scale eddies than a convective boundary
layer due to the lack of turbulence, and therefore requires a smaller vertical resolu-
tion. To resolve most of the small scale eddies, the ideal vertical resolution should
be approximately 3m [Beare et al., 2006]. To model an adequate portion of an urban
area, and to resolve sufficient amount of the turbulence, the vertical resolution of
the SBL will be set to 3m. This is very similar to the vertical resolution of the DALES
- GABLS1 case of 3.125m. The simulations with a convective boundary layer could
be properly modeled with a coarser grid, and will therefore have a vertical resolu-
tion of ∆z = 5 meters.

The boundary layer height and the increase hereof determines the required verti-
cal domain size. Due to the higher intensity of turbulence in a CBL the boundary
layer increases over time, while the boundary layer height of the SBL remains ap-
procimately constant through time. Therefore the CBL requires a larger vertical
domain than the SBL. A rule of thumb is that the boundary layer height is maxi-
mally 23 of the modeled height of the domain. The initial boundary layer depth in
both SBL and CBL simulations is 100m, as is the GABLS1 case. As the boundary
layer height of the SBL is expected to remain approximately constant throughout
the simulation, a vertical domain size of 500m should be more than sufficient, which
results in 166 grid points.

However, the vertical domain size of the convective boundary layer is limited by
the relatively small horizontal domain size. To properly implement an urban area in
the lowest part of the atmosphere, the vertical resolution should be relatively small,
and is therefore set to 5m. The total amount of grid point should be in the same
order of greatness as the amount horizontal grid points of 160. A vertical domain
size of 900m is chosen, which equals 180 grid points.

4.2 Subsidence and inversion jump

Typically, a convective boundary layer can grow several kilometers during the day.
However, due to the limited vertical domain and the requirement of a boundary
layer depth, hbl , of maximally two-thirds of the total vertical domain size, the bound-
ary layer growth, ∂h ∂t , is to be controlled. In this study, the growth of the boundary
layer is restrained by an inversion jump at the top of the boundary layer and the
subsidence; the slowly downward motion of air. The subsidence generally is a func-
tion of the height and the divergence of the large-scale horizontal wind. Here a
more pragmatic approach is applied to ensure the boundary layer height remains
within the simulated domain, where the subsidence is prescribed for every height.
A typical value for the divergence is 10−5 s−1 , which is applied in the CBL simula-
tions.
 To determine the minimum inversion jump ∆θv , the boundary layer growth can
be calculated with:

∂h w0 θ 0 sr f
=A + wh (4.1)
∂t ∆θ
wh = − Div ∗ z (4.2)
hbl = h0 + ∂h
∂t ∗t (4.3)

with A = 0.2, w0 θ 0 sr f the potential surface temperature flux, ∆θ the inversion jump
and wh the subsidence, which is a function of the divergence Div and the height
z in meters (equation 4.2). The boundary layer height after a simulation of eight
hours can be calculated with equation 4.3, where h0 is the initial boundary layer
height. With these equations, the boundary layer height after eight hours can be
determined, if the value of w0 θ 0 sr f is known. This value is obtained by averaging
the potential surface temperature flux of several test simulations with an attempted
CBL. From this is concluded that a ∆θ = 6K would ensure a boundary layer depth
after a simulated time of 8 hours to stay below 600m (two thirds of the total vertical
domain of 900m).

4.3 Surface temperature flux

The stable and the convective boundary layer are both achieved in DALES by pre-
scribing a time-dependent large-scale forcings of a negative and positive surface
heat flux respectively, in the form of an hourly prescribed potential surface temper-
ature, which decreases or increases every hour. Both heat fluxes are derived from
observations in Eindhoven of the Royal Netherlands Meteorology Institute (KNMI)
of two clear summer days, August 16th and 17th of 2016. The course of the poten-
tial liquid surface temperature θl,sr f is calculated from the measured temperature
p R
T with θl,sr f = T · p0 cp , and is given in figure 4.1. Here, the implemented surface
values for the stable and convective boundary layer are given in light blue and dark
blue respectively.

Figure 4.1: Course of the potential temperature θl,sr f After which the CBL simulations are
modeled.

DALES requires the initial vertical profile of the liquid potential temperature in
one of its input files. This is constant throughout the stable boundary layer until the
inversion height zi = 100m for both the SBL and CBL simulations. At the inversion
height, in the CBL simulation θl will experience an inversion jump of ∆θ = 6K,

23
hereafter the potential temperature is a function of the height and the adiabatic
lapse rate of 6 Kkm−1 :

z < zi −→ θl = θ0 (4.4)
z ≥ zi −→ θl = θ0 + ∆θ + 6
1000 z. (4.5)

Here θ0 denotes the potential liquid surface temperature at t = 0s, which is 290.16K
in the SBL simulation and 285.89K in the CBL simulation derived from figure 4.1,
and z is expressed in meters. The vertical profiles of both simulations are given in
figure 4.2.

Figure 4.2: The vertical profile of the potential liquid temperature of the SBL and CBL simu-
lations.

4.4 Initial geostrophic wind velocity and advection

scheme
One of the most important prognostic variables is the initial geostrophic wind
(geowind) velocity and direction. Both the SBL and CBL simulations are tested
with several initial geowind velocities ranging from 2 ms−1 to 6 ms−1 . These tests
showed that with an initial geowind velocity smaller than 5 ms−1 , the percentages of
resolved turbulences with respect to the total turbulences become too small to speak
of a properly resolved turbulent flow. Furthermore, wind velocities larger than 5
ms−1 reduced the desired differences between the stable and convective boundary
layer, in terms of vertical component of the turbulent fluxes. Therefore, the initial
geowind velocity for all simulations is 5 ms−1 .
To determine the wind direction, the locations of the sources and obstacles in the
simulated urban area are examined, which are elaborated in section 4.7. From this
is concluded that an Eastern wind is most desirable.

As described in section 2.6, there are several advection schemes available in

DALES, with each its own advantages and disadvantaged. Due to the fact that
the IBM is only implemented for a second-order advection scheme, the four sim-
ulations (SBL with and without IBM, and CBL with and without IBM) will all be
simulated with this advection scheme. The disadvantage of this advection scheme
is the tendency to model negative concentrations. Therefore, an additional sim-
ulation without obstacles is performed of a stable boundary layer with a Kappa
advection scheme to investigate the effect of a more advanced advection scheme on
the dispersion.
 4.5 Subgrid Turbulent Kinetic Energy
DALES requires a vertical profile of the initial subgrid turbulent kinetic energy
(Subgrid-TKE). The Subgrid-TKE in both the stable and convective boundary layer
are prescribed following the GABLS1 case, where the Subgrid-TKE is a function of
the height z, while taking the inversion height zi into account [Beare et al., 2006].
Above the inversion height, Subgrid-TKE = 0. Figure 4.3 shows the vertical profile of
the implemented initial Subgrid-TKE values of both the SBL and CBL simulations.
 3
z
z ≤ zi −→ Subgrid-TKE = 0.4 1 − (4.6)
zi
z > zi −→ Subgrid-TKE = 0 (4.7)

Figure 4.3: The initial vertical profile of the subgrid turbulent kinetic energy of the SBL and
CBL simulations.

4.6 Surface boundary condition: implemented

elevation map
To analyze the impact of objects on the dispersion, an elevation map of the city of
Eindhoven is implemented. Several high-traffic streets are present near high build-
ings and residential areas, making this an interesting area to model the dispersion.
Furthermore, there are two air quality measuring sites managed by the RIVM in
Eindhoven, which locations are shown in figure 4.4a. Site A and B are respectively
located at the Genovalaan and the Noord Brabantlaan. Location A is near a hospital
and a shopping mall, making it an interesting location in terms of building heights.
However, there is no large emitting source nearby, which is the case in location B.
This measurement site is located near one of the main highways, which is known to
be a large polluter. Therefore, all experiments will use an elevation map including
this location, as figure 4.4b shows.
Multiple sections of the government have collaborated to a free to the public,
elevation map, mapped from helicopter and airplane laser data: the AHN (Dutch:
Actueel Hoogtebestand Nederland). These elevation maps are downloadable here1
at the resolution of 0.5 or 5m. In figure 4.5a the original elevation map is shown,
with vegetation, cars, and other irregularities. The dimensions used for this research
1 Link: https://downloads.pdok.nl/ahn3-downloadpage/

25
 (a) (b)

Figure 4.4: a) Locations of the two measuring sites in Eindhoven, and b) the simulated loca-
tion.

are five meters in the horizontal directions (x, y). Developers of DALES are currently
working on incorporating vegetation in the model [Grylls and van Reeuwijk, 2021].
However, the version implemented here is incapable of modelling other structures
than rectangle shaped object. Figure 4.5b shows the implemented elevation map of
the stable boundary layer, corrected for obstacles other than buildings and layered
per 5m.
The large circle-shaped object is the Evoluon in Eindhoven, and in figure 4.5a
this is accompanied by a pond below street level. This pond and the part of the
highway below street level (h ≈ 6.4)m in the mid-left section of the figure, are
shown at h = 0m. In the DALES simulations, h = 0m is located at the street, which
explains the difference in height between the to figures.
The structures in this domain within five nodes from the boundary are removed,
due to the nudging of the variables as explained in section 2.5.3.

(a) (b)

Figure 4.5: The elevation map of Eindhoven with location B at its center, (a) the original and
(b) corrected for vegetation and formatted for the IBM input file.

4.7 Source locations

Two large highways are present in the simulated domain, and on the location of the
Eastern highway, a line source is implemented, due to the Eastern wind direction.
Around this line source are three point sources added, to investigate the dispersion
of point sources, as well as line sources. Figure 4.6 shows their location with respect
to the IBM map. One point source (location 2) is modeled in a relatively wide street
in approximately the wind direction, the location of the next source (location 3)
is chosen at the same y-level as the measurement site, with respect to the wind
direction. The last source (location 4) is located directly in front of the Evoluon to
investigate the dispersion of a point source with respect to a large, nearby building.
The prescribed surface fluxes of the passive tracer are for all simulations and sources
identical: 0.001 kg kg−1 ms−1 , which is defined in the simulation settings.

Figure 4.6: The locations of the scalars in the IBM domain.

4.8 Differences between the GABLS1 and this study’s

SBL
The available GABLS1 has a dimension of 400 × 400 × 400 m3 and a resolution of
3.125 m [Beare et al., 2006]. The vertical profile of the virtual potential temperature
has a constant value of 265 K up to 100m, and thereafter increases with 0.01 Km−1 .
The initial geostrophic wind velocity in the East-West direction (u g geo) is equal to 8
ms−1 , and 0 ms−1 in the North-South direction (v geo ). A prescribed surface cooling
of 0.25 Kh−1 is applied for 9 hours. The initial subgrid turbulent kinetic energy
is a function of height: 0.4(1 − z/250)3 ms−2 for the first 100m, after which it is
equal to 0 ms−2 . The latter is equal to the initial subgrid-TKE values implemented
in the SBL simulation for this study, whereas all of the other named variables differ.
Table 4.1 gives an overview of the differences in the case of GABLS1 and the SBL
simulation.

Variable
Case Domain ∆x = ∆y ∆z θv,sr f u geo ∂θv,sr f /∂t
(x × y × z)[m3 ] [m] [m] [K] [ms−1 ] [Kh−1 ]
GABLS 400 × 400 × 400 3.125 3.125 265 8 -0.25
SBL 800 × 800 × 500 5 3 290.16 5 -0.75

Table 4.1: Overview of the differences between the GABLS1 case and the implemented SBL
for this study.

The differences in terms of domain size and resolution are due to the used eleva-
tion map and required horizontal domain for the convective boundary layer. The
vertical resolution is deliberately similar with ∆z = 3m. It is however taken slightly
smaller to more accurately accommodate the buildings. Furthermore, the surface
temperature differs significantly, because the simulation for this study is modeled

27
after a summers day in The Netherlands, setting the initial virtual potential surface
temperature (θv,sr f ) to 290.16K. Another result from the implementation of the mea-
surements is the larger decrease in potential surface temperature over time, where
the potential surface temperature of the GABLS1 simulation decreases with −0.25K
per hour, the SBL simulation performed for this study decreases on average with
−0.75K per hour.

Furthermore, the inversion height is identical, as is the adiabatic lapse rate from
of the potential temperature above this inversion height om 100m and the vertical
profile of the TKE. The wind velocity differs significantly, because a wind velocity
of 8ms−1 would heavily influence the vertical dispersion in both the SBL an CBL
simulation, decreasing the differences between the two.
The differences between these two cases will be extensively investigated in chap-
ter 5.
5
Analysis of atmospheric
conditions
To investigate DALES’ capabilities to simulate an urban area faithfully, several steps
are demanded. First, this chapter will analyse the SBL and CBL simulations with-
out obstacles in section 5.1, to establish realistic simulated atmospheric conditions.
As stated before, the stable boundary layer is constructed from the GABLS1 case
[Beare et al., 2006], and several variables from the SBL simulation of this study will
be compared against the results of the GABLS1 case. Even though both cases pro-
duce an SBL, the SBL simulated for this study will further be referred to as the
SBL simulation. The convective cases will be verified by a theoretical approach, by
analysing the vertical profiles of the buoyancy flux, TKE, velocity variances and the
TKE budget equation.
When the reliability of the simulations is established, mainly in terms of the
resolved turbulence and the velocity variances, section 5.2 will investigate the effect
of the implemented urban area on the meteorological variables, like the vertical
profiles of the wind velocity, and the buoyancy flux.

5.1 Establishing the simulations

The reliability of the simulated environment is analyzed by the results of several
variables, the buoyancy flux, the vertical profiles of the total TKE and the SFS-TKE
and the ratio between the two. The total buoyancy flux indicates the vertical kine-
matic flux of the virtual potential temperature, positive in the upwards direction.
The vertical profiles of the TKE give insight into the magnitude of the simulated
turbulence and the SFS-TKE ratio to the total TKE into the subgrid contribution to
the total TKE, where a high value indicates a large dependency on subfilter turbu-
lence. This variable is an important addition because a reliable simulation should
not rely on subfilter turbulence.

5.1.1 Stable Boundary Layer (SBL)

As stated in chapter 4.8, the simulated SBL is based on the GABLS1 case, and the
convective boundary layer is on surface measurements. To ensure a reliable SBL,
the GABLS1 case is gradually modified to build the desired case. An elaboration

29
of the results of these simulations with on change per simulation are given in Ap-
pendix B. The largest differences between the GABLS1 case and the SBL simulation
are the geostrophic wind velocity (u geo,gabls = 8ms−1 and u geo,sbl = 5ms−1 ), the ini-
tial surface temperature with a difference of 25K and the hourly decrease in surface
temperature, where the surface temperature of the SBL simulation decreases three
times as fast a the GABLS1 simulation.

Figure 5.1 gives the time-averaged vertical profiles of the buoyancy flux, the SFS
and total TKE, and the ratio between the two, of both the GABLS1 and the SBL
simulation. The buoyancy flux of both simulations are remarkably similar, follow-
ing almost the same path. The only difference is that the total buoyancy flux of
the SBL simulation at the surface is larger than the GABLS1 simulation, which can
be explained by the larger hourly surface temperature decrease. The magnitude
of the total TKE of both the simulations match as well. The vertical profile of the
GABLS1 simulation however, contains more turbulence above 50m, where the SBL
simulation decreases steadily with height, after the first levels. This difference could
be due to the lower geowind, which in a stable boundary layer is one of the main
causes of the turbulence, in the form of shear turbulence.

The subfilter-scale TKE of the SBL simulation is significantly smaller than the
GABLS1 simulation, which in figure 5.1.b is behind the solid red line for the lowest
heights. This translates to the ratios seen in figure 5.1.c, where the subgrid con-
tribution to the total TKE is given. The SFS-TKE/Total TKE ratio of the GABLS1
case is relatively high, which indicates that most of the turbulence is not resolved,
but parameterized. The ratio in the SBL simulation is smaller than 0.4, which is
relatively small meaning that most turbulences are resolved.

Figure 5.1: The vertical profiles of the total buoyancy flux, the total and SFS TKE and the
ratio of SFS to Total TKE. Blue indicates the results of the GABLS1 simulation
and red the SBL simulation.

From these results could be concluded that the simulated stable boundary layer
is a reliable simulation, because of the favorable ratio of the subfilter surface TKE to
the total turbulent kinetic energy, and the similar vertical profiles of the total TKE
and buoyancy flux to the GABLS1 simulation.

5.1.2 Convective Boundary Layer (CBL)

The convective boundary layer is achieved with an increasing prescribed surface

temperature to ensure the rise of the mixing layer. Due to the unaltered horizontal
domain with respect to the SBL simulation (800 × 800m), the vertical domain size
is limited. The domain should be able to contain multiple eddies, and since their
horizontal size is approximately equal to their vertical size, the boundary layer
depth is limited in its size and growth. The exact equations and values of the
implemented subsidence to control the boundary layer growth are described in
chapter 4.2.
Similarly to the SBL, the reliability of the CBL simulation is investigated through
the total buoyancy flux, the resolved TKE and sub-filter scale TKE and their ratio,
the resolved horizontal and vertical velocity variance and the TKE budget eqaution.
Figure 5.2 shows the values of these variables over time, as the boundary layer
grows with time.

Figure 5.2: The vertical profiles of the total buoyancy flux, the resolved TKE, the ratio of
SFS/Total TKE, the resolved horizontal and vertical velocity variance and the
TKE budget equation, of the CBL simulation. The results range from t = 2 hours
to t = 6h, with blue the first timestep and red the last. If present, the solid black
line gives the time-averaged value.

These figures show the growth of the boundary layer very clearly, starting with a
height of 100m, developing slowly in these six hours to a boundary layer height of
600m. The last two simulated hours are discarded because the boundary layer grew
beyond the 600m mark, creating unreliable results. First, the vertical profile of the
buoyancy flux is following a vertical profile known to occur during the day: from a
positive value on the surface, decreasing with height due to the excess in updrafts,
to become negative in the mixing layer and thus having a decelerating effect on the
rising of air parcels, and finally above the boundary layer, reaching a value of zero.

Furthermore, significantly more turbulence is resolved than parameterized as SFS-

TKE, with an average SFS/Total TKE ratio in the boundary layer of 0.1. This indi-
cates a well-performing simulation and a reliable convective boundary layer. The re-
solved horizontal and vertical variance all have comparable magnitudes, indicating
sufficient turbulence in all directions, and follow the vertical profile as is expected
of these variables and given in literature [Efstathiou et al., 2018]. The TKE-budget
gives an indication of the main different components of the TKE. The buoyancy,

31
shear and transport factors are significantly enough to balance the dissipation from
the resolved turbulence to small-scale eddies.

5.2 Effect of the addition of obstacles

In this section, the implementation of an urban area to a DALES simulation is ana-
lyzed through several variables, focusing on the impact on boundary layer height,
buoyancy, the horizontal and vertical velocity variances and the TKE budget terms.
The goal of this chapter is to examine whether DALES simulates a reliable urban
area and give insights into topics requiring further investigation. This section starts
with a first impression of the surface level concentrations with the addition of ob-
stacles. This is followed by the expectations, and two separate analyses of several
meteorological variables of the SBL simulation and the CBL simulation, to inves-
tigate the impact of obstacles of the simulated boundary layer. The two different
boundary layers will be compared with each other in chapter 6.

The time-averaged surface concentrations of the line source are given in figure
5.3, of the SBL simulation with obstacles. Here can be seen that the concentrations
within the buildings are zero and the fluxes flow properly around the obstacles.

Figure 5.3: The time-averaged surface concentrations of the line source of the SBL simulation
with obstacles.

5.2.1 Expectations

The expected influence of an urban area on a boundary layer regardless of the

stability, is that the presence of buildings increases the boundary layer height. The
horizontal turbulent fluxes at the building heights experience a change in direction
when encountering a building by going around and above.
The expected effect on a stable boundary layer would be larger than the effect on
a convective boundary layer. Considering that a convective boundary layer already
has a strong vertical turbulence caused by convection, whereas the addition of a
vertical flux to the stable boundary layer can significantly increase the boundary
layer height. A convective boundary layer typically grows to several kilometers
during the day, where the addition of several buildings with a maximum height of
30m, probably does not impact the boundary layer depth significantly. However,
the simulated CBL in this study has a maximum boundary layer depth under 600m,
due to domain restrictions. Therefore the addition of an elevation map will probably
add significantly to the boundary layer depth of a CBL, though considerably less
than to the SBL.

5.2.2 SBL results

In figure 5.4 is the impact of the use of the IBM (dotted lines) shown for the SBL
simulation, from the first level without obstacles up to a height of 180m. As can be
seen clearly in 5.4.a) the vertical profile of the wind velocity, the stable boundary
layer increases exactly with a factor of 1.5 due to the addition of an urban area.
This explains the difference in the virtual temperature profile, because it increases
linearly to the top of the boundary layer. The surface temperature and the tem-
perature above the boundary layer are equal, only the boundary layer differs and
therefore the slope of θv in the boundary layer. The difference in boundary layer
height also results in the shown difference in the buoyancy flux.
The horizontal velocity variance in the wind direction is a factor 10 larger than in
the other two direction, which is expected in a stable boundary layer where shear
turbulence mostly dominates. Remarkably, the addition of an urban area to the sim-
ulation does not significantly change the horizontal velocity variance in the wind
direction near the surface. The largest differences are the v0 v0 res , and w0 w0 res near
the surface, which is caused by the presence of buildings. This results in a larger
dispersion rate with the addition of obstacles.

Figure 5.4: The impact of the obstacles on the stable boundary layer over time, expressed
with the vertical profiles up to a height of 180m, of 4 different times in the simu-
lation, and the average values in black. The dotted lines represent the simulations
with obstacles and the solid line without. The following variables are shown: a)
the wind velocity, b) the virtual potential temperature, c) the total buoyancy flux,
d) and e) the resolved horizontal velocity variance in u and v direction respec-
tively, and f) the resolved vertical velocity variance.

33
5.2.3 CBL results

The same calculations and figures as performed for figure 5.4 are repeated for the
convective boundary layer. The results are shown in figure 5.5. Due to the constant
growth of the boundary layer during this simulation, the time averaged values are
not calculated for this analyses.
As expected, the boundary layer growth due to an urban area is not as significant
for a convective boundary layer as a stable one. The convective boundary layer has
grown with a factor of 1.12. Furthermore, figure 5.5.a) shows that with obstacles,
the wind velocity in the boundary layer is significantly reduced, with the difference
increasing over time. The presence of buildings halts the horizontal wind velocity
in comparison with an area without buildings. However, it is remarkable to see the
influence of buildings with a maximum height of 30m, have such a large impact on
the wind velocity throughout the whole boundary layer. The impact of an urban
area on a boundary layer of several kilometers is therefore an interesting topic to
further investigate.

The slight difference between the two virtual potential temperature profiles is
due to a larger upwards motion in the simulation with an urban area, therefore,
the surface temperature is slightly better mixed with the boundary layer. With the
addition of the obstacles, the buoyancy flux especially has a larger negative value in-
dicating a larger mixing layer between the boundary layer and the free atmosphere
above. The horizontal velocity variances in the wind direction (u) does not change
near the surface, however the variances in the other two direction increase visibly
in the lowest part of the atmosphere.

Figure 5.5: The impact of the obstacles on the convective boundary layer over time, of 4
different times in the simulation ranging from t = 2h to t = 6h. The dotted lines
represent the simulations with obstacles and the solid line without. The following
variables are shown: a) the wind velocity, b) the virtual potential temperature, c)
the total buoyancy flux, d) and e) the resolved horizontal velocity variance in u
and v direction respectively, and f) the resolved vertical velocity variance.
6
Dispersion analysis
The diffusion of a pollutant in the atmosphere is directly related to the stability and
thus the amount of turbulence. For this study the prescription of heterogeneous
sources at the surface level is added to DALES, which allows the implementation of
multiple sources, in this case one line source and three point sources, located at the
surface. The in-depth description of this extra module and the locations of these
sources is given in chapter 2.7 and 4.7 respectively.
First, the time-averaged concentrations of the pollutant of the simulations without
obstacles are compared to the Gaussian plume model. They will be compared in
the vertical as well as the horizontal plane, for both the stable and the convective
boundary layer. Thereafter, the influence of the obstacles on the dispersion of the
pollutant is inspected to assess the usability of DALES to properly model an urban
area. This chapter gives the results for the line source, and the results for the point
sources are given and elaborated in Appendix C, due to the fact that the results
from the point sources are considerably similar to the results from the line source.
Therefore, these analyses did not lead to different conclusions.
Finally, in section 6.3 two additional analyses are performed to investigate whether
a different advection scheme has a significant influence on the dispersion. All time-
averaged values do not take the first two hours into consideration due to the spin
up of the simulation.

6.1 Comparison with the Gaussian plume model

The time-averaged concentrations of the SBL simulations without obstacles, of both
DALES and the Gaussian plume model, are displayed in figure 6.1. The results are
all scaled equally, and to ensure visibility at larger distances limited to a value of
20µg/m3 , which is lower that the maximum concentration. The left figures show
the concentration in an x-y-plane at the lowest obtained level, z = 1.5m, and the
figures on the right give a height slice in the downwind direction, of the lowest
100m.

35
 (a) (b)

(c) (d)

Figure 6.1: A visualisation of the pollutant concentration of the DALES and Gaussian plume
SBL simulations, horizontally at z = 1.5m and vertically downwind.

The concentrations of both the Gaussian plume model and DALES have a similar
magnitude, however the concentration within the Gaussian plume model reaches
higher into the boundary layer than simulated by DALES. Furthermore, the hori-
zontal distribution downwind of the concentration differs between the two models.
The results obtained by DALES have higher concentration at the source, while the
Gaussian plume results show higher concentrations at larger distances.

Figure 6.1d shows an oscillating effect of the concentration upwind and above
the plume, resulting in negative concentrations. This is due to the implemented
advection scheme. Chapter 6.3 gives a more in-depth analysis of this aspect.

Figure 6.2 repeats the procedure as 6.1 for the CBL simulations. Immediately
noticeable is the difference in magnitude of the concentrations. The concentrations
obtained from the Gaussian plume model are significantly larger than obtained
from DALES. This will be further investigated in this chapter. The downwind and
upwards distribution of the concentrations however, seems to correspond quite well.
 (a) (b)

(c) (d)

Figure 6.2: A visualisation of the pollutant concentration of the DALES and Gaussian plume
CBL simulations, horizontally at z = 2.5 and vertically downwind.

6.1.1 Surface concentrations

In figure 6.3, the spatial distribution of the downwind concentrations are displayed,
expressed with the actual concentrations in µg m−3 and in percentages. The lat-
ter is calculated as the part of the sum of all directly downwind concentrations,
calculated in percentages. E.g.: in figure 6.3.b., approximately 8.7% of all emitted
downwind concentrations obtained by the Gaussian plume model is present at the
source, whereas in the DALES results, this is ≈ 2.1%. The results of all simulations
are shown, grouped per stability, at the surface level (z = 0m), time-averaged and
averaged over for all x-values along the black lines in figure 6.5.

This figure gives insight into the concentrations at the lowest modeled elevation.
As seen in figure 6.1 and 6.2, the concentrations of the SBL simulations are quite sim-
ilar, apart from the concentrations at and close to the source. The Gaussian plume
model calculates a higher concentration directly at the source, however, from 5m
downwind of the source DALES calculates a higher surface concentration. These
differences could be due to the fact that a Gaussian plume model is essentially
build for sources with a certain stack height from the ground, where in this study
the stack height equals 0m. The model first determines the locations downwind and
the distances in x and y direction from the source, and then calculates the concen-
trations with a fixed set of parameters. The calculation of the concentration follows
this parameterization and includes a double height term following the reflection
assumption (see chapter 3 for further elaborations and the applied equations). This
could cause such a high concentration at the source in the Gaussian plume model.

37
Figure 6.3: The time-averaged line source concentrations χ in the downwind direction and
the percentages of the total concentration per downwind distance. The blue lines
represent the results of the Gaussian plume model, red the DALES simulation
without an urban area. a) and b) give the concentration and the percentages of
the SBL simulations, and c) and d) of the CBL simulations.

The differences between the two CBL simulations are however more significant.
The maximum concentration obtained from the convective Gaussian plume model
is almost 10 times larger than the maximum concentration simulated by DALES.
The spatial distribution of the CBL simulations however is significantly similar.

Figure 6.4 gives an equally scaled comparison of the SBL and CBL concentrations
of both models, and shows that the concentrations of the SBL and CBL calculated
by the Gaussian plume model differ with a factor of approximately 2, while the
difference in DALES between the concentrations of the SBL and CBL simulations
are a factor 10 apart. A reason for these differences in surface concentrations could
be that most of the emission modeled by DALES disperses upwards instead of re-
maining at the surface. This will be further investigated in this section.

Figure 6.4: A comparison between the concentrations at downwind distances from the line
source between the SBL and CBL simulations of a) the Gaussian plume model
and b) DALES.
In conclusion, there are several significant differences between the surface concen-
trations of the Gaussian plume model and the DALES simulations of a line source
pollutant. The distribution of the surface concentrations in a SBL is more evenly
distributed in the downwind direction when simulated by DALES. The Gaussian
plume model has a relatively high concentration at the source, but lower concentra-
tions downwind than DALES calculates. For the CBL, the distribution of the down-
wind concentration of both simulations are more alike. However, the magnitude of
the calculated concentrations differ considerably for this simulated boundary layer:
the concentrations calculated by the Gaussian plume model are a factor of 10 larger
than the surface concentrations obtained by DALES. The immense difference be-
tween the surface concentrations of the CBL simulation and the distribution of the
SBL could be further investigated. The vertical distribution of the pollutant could
be an explanation, and will be investigated in the next section.

6.1.2 Vertical profiles

The vertical profiles of the concentrations are analyzed at specific downwind dis-
tances from the line source. For practical reasons, this paragraph starts with an
overview of these locations with respect to the line source. Figure 6.5 displays the
location of the line source at 0m, and the concentrations are analyzed along the solid
black lines, from ∆x = −5m to ∆x = 500m. Here the time-averaged surface con-
centration levels of the passive scalar are shown of the stable boundary layer, with
and without the buildings. Here the purpose of this figure is merely to reiterate the
exact line source location and to provide a visual aid to the analyses.

Figure 6.5: An informative figure for better understanding of the performed analyses in this
chapter. Shown are the time-averaged surface concentrations (z = 1.5m), of the
DALES - SBL simulations, with and without obstacles.

All analyses containing specific downwind distances from the source ∆x are av-
eraged over the black lines in figure 6.5. For example, if the vertical profile 200m
from the source is given, is this the average of all concentrations along the black line
denoted with 200m in figure 6.5, excluding the buildings if present. Furthermore,
these lines specify which distances are analyzed.

In figure 6.6 the vertical profiles of the time- and spatial-averaged concentration
are shown of the lowest part of the atmosphere. On the left are the DALES sim-
ulations without obstacles, and the results from the Gaussian plume model are
portrayed on the right. The specific maximum height shown is dependent on the
stability, 23m for the SBL simulations and 45m for the CBL simulations. As dis-
cussed in the previous paragraph, the surface concentrations in the Gaussian plume
model are considerably higher for a CBL than calculated by DALES. Seen in this
figure, is that this is also true for the vertical profiles of the concentration.

39
Figure 6.6: The time-averaged, spatial averaged, vertical profiles of the concentration as ob-
tained from the DALES and Gaussian plume model at several downwind dis-
tances from the line source, as shown in figure 6.5. a) and b) give the concen-
tration and the percentages of the SBL simulations, and c) and d) of the CBL
simulations.

The vertical profile of the SBL simulation differs considerably between the two
models. The concentration in the Gaussian plume model reaches higher elevation
than the concentrations obtained from DALES. In both the stable and convective
boundary layer, most of the concentration remains at the surface, and the surface
concentrations of the SBL are significantly higher than the CBL.

6.1.3 Development over time

Figure 6.7.a. and b. show the development over time of both the DALES simulations
without the immersed boundary method compared to the Gaussian plume model,
at several predefined distances from the source. The concentrations calculated by
the Gaussian plume model are constant due to the constant input variables of the
emission rate Q and the wind velocity. All concentration are shown at z = 1.5m for
the SBL and z = 2.5m for the CBL. DALES takes approximately two hours to initiate
the simulation, this is denoted with a dotted grey line in both figures. Interesting
is the increase at t = 2h, followed by a decrease, and a steady increase again in
concentration over time in the stable boundary layer at all distances from the source,
whereas in the convective boundary layer, the concentration oscillates around a
stable amplitude. From this could be concluded that in a stable boundary layer, a
percentage of the emission will remain close to the source instead of dispersing out
of the domain. In a convective boundary layer, this does not seem the case.

6.1.4 Conclusion

There are considerable differences between the calculated diffusion of the two mod-
els, in terms of magnitude, the vertical and horizontal distribution, especially in the
CBL simulation. The magnitude of the concentrations in the CBL differs consider-
ably while the horizontal and vertical distribution are quite similar, they only differ
Figure 6.7: The Development of the concentration over time of the Gaussian plume model
and a) the simulated stable boundary layer and b) the simulated convective
boundary layer.

with a factor of 10 in magnitude. The dispersion in the SBL simulations are more
similar, however, DALES calculates a higher value at the surface, but lower down-
wind and reaches a lower height than in the Gaussian plume model. Remarkable
is the large difference in concentrations between the stable and convective bound-
ary layer, which are significantly larger between the two DALES simulations than
the Gaussian plume calculations. All these factors contribute to the conclusion that
more research is necessary to investigate the diffusion profiles of a line or point
source in DALES in comparison with the Gaussian plume model.

6.2 Comparison with the urban area

The effect of the added obstacles on the diffusion of an pollutant is investigated
with the same analyses as performed in the previous section. Figure 6.8 gives a
global insight into the concentration distribution of SBL and CBL simulations per-
formed by DALES. Similarly to 6.1, the left figures are the horizontal distributions
of the concentrations at the z = 1.5m for the SBL and z = 2.5m for the CBL, and the
vertical distributions are shown at the right.

These figures immediately show a considerable impact of the obstacles on the

surface concentrations. Moreover, these figures display that the implementation of
an urban area in DALES seems to function properly, when looking at dispersion.
The concentrations of the passive tracer move around the obstacles and the maxi-
mum value of the concentration is not only centered at the source, but more evenly
distributed around the source.

Another observation is that in both the SBL and CBL the emission seems to stay in
the lowest part of the atmosphere and concentrates around the source, downwind
and upwind. It does reach a higher altitude in a city, due to the upwards motion
of the fluxes corresponding with the presence of buildings. The effect is larger in a
stable boundary layer, however the IBM seems to still have a significant impact on
the surface concentrations. Further investigation into these values is performed in
following sections.

41
 (a) (b)

(c) (d)

(e) (f)

(g) (h)

Figure 6.8: Vertical and horizontal profiles of time-averaged concentrations of the SBL and
CBL DALES simulation, with obstacles and without.
6.2.1 Surface concentrations

Figure 6.9 displays the horizontal distribution of the surface level concentrations of
the SBL and CBL simulations, comparing the results with and without obstacles.
The locations where a building is present are excluded.

Figure 6.9: A comparison of the time-averaged line source concentrations χ between the sim-
ulation with (blue) and without obstacles (red). These values are the averaged in
the downwind direction and the percentages of the total concentration per down-
wind distance. a) and b) give the concentration and the percentages respectively
of the SBL simulations, and c) and d) of the CBL simulations.

As expected, this figure shows that the surface concentrations levels are much
more concentrated when surrounded by buildings. The concentration at the source
of the stable boundary layer has increased with a factor of 1.8 and the concentra-
tion in the convective layer with 2.3. In the SBL simulation, the concentrations at a
short distance from the source are increased by the presence of the obstacles, while
in the CBL simulation, this is not the case. This is probably due to the increase
of turbulence present in the convective boundary layer, resulting in an increased
concentration at higher elevations. This effects also applies to the horizontal distri-
bution of the concentration.

6.2.2 Vertical profiles

As introduced before, figure 6.10 given the vertical profiles of the concentrations,
spatially averaged along the line source, at several distances from the source lo-
cation. Here can be seen that not only have the obstacles a large impact on the
horizontal diffusion, it also influences the vertical distribution of a tracer concen-
tration. Moreover, without obstacles the concentrations upwind are negative, while
figure b) and d) show that the concentration upwind at ∆x = −5ms−1 is nearly
as high as downwind. The concentration in the simulations without obstacles is
more concentrated around the source, regardless of the stability. However, at fur-
ther distances from the source, the concentration levels decreases when an urban
area is added. In the SBL simulation, this effect is seen after 200m, while in the CBL

43
simulation, the concentrations obtained by DALES with obstacles are lower from a
downwind distance of 50m.

Figure 6.10: Vertical profiles of the concentrations within the stable and convective boundary
layer, with and without IBM at several downwind distances from the line source.

Figure 6.11 gives the vertical distributions expressed as the part of the sum of the
vertical concentration per height, of both the SBL and the CBL simulation. E.g.: at
0m from the source location (figure 6.11.a), the simulation without buildings (blue)
calculated that approximately 90% of the all the vertically distributed concentrations
are present at he surface level (z = 1.5m). All blue lines denote the simulation
without buildings, and red with buildings. This metric is introduced to compare
the vertical distributions if the absolute concentration values differ greatly in their
magnitude, as per the CBL simulation.
Here the effect of the obstacles can clearly be seen, as that even at the source
location (∆x = 0) the percentage of the total vertical concentration at the surface
is 54%, whereas in both the stable and convective boundary layer without an IBM,
this percentage is nearly 90%. Interesting is the development of dispersion in the
stable boundary layer with distance, where at ∆x = 500m, the most concentration
is not at the surface, but rather at z = 25m. This effect is also seen in the convective
boundary layer, at ∆x = 100m, however less significant.
Figure 6.11: Vertical distribution of the concentrations within the stable and convective
boundary layer of both the simulations with and without an IBM, at several
distances from the line source.

6.2.3 Development over time

Figure 6.12 gives an overview of the development of the surface concentrations over
time. The SBL simulation without obstacles has, from t = 5h an increase with time,
as has the SBL simulation with obstacles from distances larger than 10m from the
source. All simulations have a considerable large oscillation effect. On a stable
boundary layer, the addition of the urban area, have on both the SBL as the CBL an
enlarging effect on the surface concentration of the passive tracer.

Figure 6.12: The Development of the surface concentration over time of the stable (a. and b.)
and the convective (c. and d.) boundary layer. The different colors denote the
different downwind distances from the source.

6.2.4 Conclusion

The surface concentration of a pollutant is severely influenced by the presence of

buildings, regardless the stability of the atmosphere. The effect is smaller on the
time-averaged concentrations in stable night-time conditions, than convective day-

45
time conditions. The concentration can grow with a factor of 1.5 − 5 times the
concentration without an urban area, depending on the distance and height from
the source. The dispersion is restrained by the presence of buildings within 200m in
a stable boundary layer and 50m in a convective boundary layer. These results are
presumably highly dependent on the wind velocity, placing of buildings and source,
vertical domain size, and turbulence. These factors require further investigation to
advance the study of the implementation of an urban area in DALES.
Furthermore, the addition of the urban area has a significant impact on the devel-
opment of the downwind surface concentrations over time in the stable boundary
layer. Whereas the downwind surface concentrations increase over time in the sta-
ble boundary layer without buildings, the presence of buildings has a large impact
on these concentrations.

6.3 Influence of a different advection scheme

As explained in chapter 2.6, DALES offers the option to implement different advec-
tion schemes. The Immersed Boundary Method which provides the addition of an
urban area, is only implemented for a second order advection scheme. However, as
seen in the results earlier this chapter, this advection scheme can cause an oscillating
effect in some fluxes.
An additional simulation is performed to investigate the influence of a different
advection scheme on the dispersion, the kappa advection scheme. This simulation
is in all aspects identical to the stable boundary layer simulation without the ad-
dition of obstacles, with one difference: the advection scheme. This simulation
will be referred to in this chapter as the Kappa simulation. Figure 6.13 gives the
vertical and horizontal distribution as obtained from the three models; the DALES
simulation with a Kappa advection scheme, the DALES simulation with a second
order advection scheme and the Gaussian plume model. Only the SBL simulations
is investigated. The left figures give the horizontal distribution at the surface level
(z = 1.5m) and the right figures give the vertical distribution.
The distribution of the pollutant as calculated by the kappa advection scheme
seems to resemble the Gaussian plume model to a greater degree, with a higher
concentration at the source and a lower concentration downwind. The implementa-
tion of this advection scheme has resolved the oscillation present in the simulation
displayed in 6.13.c and d.
 (a) (b)

(c) (d)

(e) (f)

Figure 6.13: Vertical and horizontal profiles of time-averaged concentrations of the Kappa
advection scheme, in comparison to the second order advection scheme and the
Gaussian plume model.

Interesting is to investigate the seemingly higher concentrations at the source in

the DALES-Kappa simulation. Figure 6.14 gives these values, spatially- and time-
averaged, with x = 0 the location of the source. Remarkable is the magnitude of
the concentration at the source. The maximum value of the Kappa simulation is
over 10 times the size of the second order-simulation and 6 times of the Gaussian
plume. The distribution of the concentration over the downwind distance is almost
identical to the Gaussian plume model, possibly indicating the correctness of both.

An explanation for the low concentrations calculated with the second order ad-
vection scheme could be that the oscillating effect possibly impacts the magnitude
of the highest concentration. The concentrations fields within DALES are conserved,
meaning that the presence of negative values impacts the size of the positive values.
Therefore, with the absence of the negative values, the concentrations calculated in
the Kappa-simulation are locally higher than calculated by the second order advec-

47
tion scheme.

Figure 6.14: The surface concentrations of the SBL Gaussian plume model, the DALES sim-
ulation with a 2nd order advection scheme and the DALES simulation with a
Kappa advection scheme.

Figure 6.15 displays the vertical distribution of the concentration at different

downwind distances. From this graph can be concluded that the Kappa advec-
tion scheme has a minimal effect on the vertical diffusion. The surface values differ
slightly, however, the overall distribution is remarkable similar. While the horizon-
tal distribution resembles the Gaussian plume model, the vertical are significantly
different.
These results show the difficulty in comparing the dispersion of a pollutant close
to the source. LES may not be the best tool to investigate the concentrations at this
distance. However, further downwind, it benefits in terms of spatial distributions,
time variations and taking into account presence of obstacles.

Figure 6.15: The vertical profile distribution of the Gaussian plume model, the Kappa ad-
vection scheme and the second order advection scheme, displayed at several
downwind distances.

6.3.1 Development over time

In figure 6.16 the development of the surface concentrations over time are displays
of the Gaussian plume model, the SBL DALES simulation without the implemen-
tation of the IBM, and the Kappa simulation at different distances from the source.
As seen before, the concentration at the source is significantly higher in the Kappa
simulation, and increases heavily over time. Therefore, the same results are shown
three times, on three different vertical scales. The development of the concentration
over time at the Kappa simulation is experiencing almost no oscillation in both the
absolute value and time, but increases steadily.

Figure 6.16: Development of surface concentrations over time of the simulation with the
Kappa advection scheme, compared with the SBL simulation without obstacles
and the Gaussian plume model

6.3.2 Conclusion

The kappa advection scheme is rather similar to the Gaussian plume model in terms
of the horizontal distribution. The concentration at the source is considerably higher
than the second order advection scheme, questioning the reliability of these values.
Perhaps the oscillating effect observed with the second order advection scheme,
meaning the erroneous negative values and positive values near the source, had an
impact on the magnitude of the concentration. Therefore a favourable improvement
to the DALES is the implementation of the kappa advection scheme in combination
with the addition of obstacles. Additionally, the lack of oscillation and negative
values in the concentration over time when calculated with the Kappa advection
scheme given this advection scheme preference over the second order advection
scheme.

49
7
Conclusions and
Recommendations
This chapter concludes all findings in the analyses performed in chapter 5 and
6, followed by the recommendations for further studies. First the conclusion will
inspect the effect of the addition of an urban area on different variables of the stable
and convective boundary layer. This is followed by the conclusions of the dispersion
analyses: the comparison with the Gaussian plume model and the influence of
obstacles on the dispersion. The impact of a different advection scheme of the
dispersion is discussed last. This chapter concludes with the recommendations for
further studies.

7.1 Conclusions

7.1.1 Effect of the addition of obstacles

The impact of the obstacles is measured in terms of boundary layer height and the
vertical profiles of the wind velocity, buoyancy flux and turbulence. The expected
influence of an urban area on a boundary layer regardless of the stability, is that the
presence of buildings increases the boundary layer height. The horizontal turbulent
fluxes of the TKE at the building heights experience a change in direction when
encountering a building by going around it. A part of the horizontal turbulence
translates thus to the vertical turbulent transport. The simulation results show that
the stable boundary layer height grows with 50%, while the convective boundary
layer height increase with 12%.

In regard of the vertical wind velocity profile, the presence of buildings halts the
horizontal wind velocity of both the stable and convective boundary layer, through-
out the whole boundary layer almost with 50%. As the maximum height of the
buildings in the simulated part of Eindhoven for this study is merely 30m, it is
interesting that these relatively small buildings have such a large impact on the ver-
tical wind profile.

51
7.1.2 Dispersion

First the DALES simulations without an urban area are compared to a Gaussian
plume model, in terms of horizontal and vertical distribution and the development
of the concentrations over time. Then the effect of the obstacles on the DALES sim-
ulations is assessed on the same aspects.

Gaussian plume comparison

There are considerable differences between the calculated diffusion of the Gaussian
plume model and DALES, in terms magnitude for the CBL simulation, and the hori-
zontal distribution for the SBL simulation. The concentrations calculated by DALES
are approximately ten times smaller than calculated by the Gaussian plume model,
while the concentrations between the SBL simulations of both models differ with a
factor smaller than two. In the SBL simulation results of DALES, the surface concen-
tration is larger at the source than the Gaussian plume model estimates, while fur-
ther downwind, the surface concentration obtained by the Gaussian plume model is
higher. Moreover, the concentration calculated by the Gaussian plume model reach
a higher elevation in the stable boundary layer than simulated by DALES.

Effect of an urban area

The concentration of a pollutant is severely influenced by the presence of buildings,
regardless the stability of the atmosphere. The effect is larger on the time-averaged
concentrations in stable night-time conditions, than convective day-time conditions.
The results of the simulations showed that the surface concentrations of the pol-
lutant in the SBL where higher up to a downwind distance of 200m due to the
presence of buildings, but after this distance the surface concentration decreased
with respect to the SBL simulation without buildings. In a convective boundary
layer, the downwind distance from which the surface concentration decreases due
to the presence of buildings is much closer: 50m due to the larger turbulences
present in the CBL. These results could be highly dependent on the wind velocity,
placing of buildings and the source, vertical domain size, and turbulence. These
factors require further investigation to advance the study of the implementation of
an urban area in DALES.

Furthermore, the surface concentration calculated by DALES increases with time

in a stable boundary layer, which needs further investigation. The development
over time of the surface concentration in the CBL simulations do nu decrease nor
increase over time.

Kappa advection scheme

The results as obtained by the simulation with the kappa advection scheme are
rather similar to the Gaussian plume model in terms of the horizontal distribution.
The concentration at the source is considerably higher than the second order advec-
tion scheme, questioning the reliability of these values. Perhaps the oscillating effect
of the surface concentration observed with the second order advection scheme has
an impact on the magnitude of the concentration. Therefore a favourable improve-
ment to the IBM in DALES is the implementation of the kappa advection scheme in
combination with an IBM. Additionally, the lack of oscillation in the concentration
over time when calculated with the Kappa advection scheme given this advection
scheme preference over the second order advection scheme. Similar to the second
order advection scheme, the surface concentrations increase over time.
7.2 Recommendations
From these analyses, several aspects come to light which require further research.
First, the impact of an urban area on a deep convective boundary layer is necessary
to investigate whether the results observed in this study also apply there, such as
the large impact on the boundary layer height and the vertical profile of the wind
velocity.

The comparison between the Gaussian plume model, DALES and the results ob-
tained from the two advection schemes illustrated the extremely large differences
between magnitude of the simulated concentrations of the CBL and the SBL, and
the horizontal and the vertical distribution of the concentrations. Additional inves-
tigation into the relevance of the stack height of the source is recommended. The
addition of the line and point sources build for this study only allow the source to
be located at the surface.

Finally, several additions and adjustments to DALES are needed to accurately

simulate the dispersion of pollutions in a city: the possibility to simulate vegetation,
and the possibility to add the building height as a 3D matrix instead of a birds-
eye-view. Currently the top of the buildings are given in a 2D matrix and DALES
assumes all grid-points below the top of the building as such. For the most part of
an urban area a 3D matrix would not make a enormous difference as most of the
buildings straight, and the data for a 3D matrix of a city is not to my knowledge
available. However, for example the modeled Evoluon in this study, is spherical,
and therefore the flow would be more realistic if a 3D matrix could be implemented.

Several aspects were out of scope for this research which should still be investi-
gated. This research excluded any form of moisture in the atmosphere, which are
very interesting topics: cloudiness and precipitation. The wind speed and direction
could also be differentiated, to investigate their effect on the dispersion.

53
Bibliography
Ascenso, A., Augusto, B., Silveira, C., Rafael, S., Coelho, S., Monteiro, A., Ferreira, J.,
Menezes, I., Roebeling, P., and Miranda, A. I. (2021). Impacts of nature-based
solutions on the urban atmospheric environment: a case study for eindhoven,
the netherlands. Urban Forestry and Urban Greening, 57. Cited By :1.

Beare, R. J., Macvean, M. K., Holtslag, A. A., Cuxart, J., Esau, I., Golaz, J.-C., Jimenez,
M. A., Khairoutdinov, M., Kosovic, B., Lewellen, D., et al. (2006). An intercom-
parison of large-eddy simulations of the stable boundary layer. Boundary-Layer
Meteorology, 118(2):247–272.

Carrascal, M. D., Puigcerver, M., and Puig, P. (1993). Sensitivity of gaussian plume
model to dispersion specifications. Theoretical and Applied Climatology, 48(2-
3):147–157. Cited By :22.

Deardorff, J. W. (1974). Three-dimensional numerical study of the height and mean

structure of a heated planetary boundary layer. Boundary-Layer Meteorology,
7(1):81–106.

Deardorff, J. W. (1980). Stratocumulus-capped mixed layers derived from a three-

dimensional model. Boundary-Layer Meteorology, 18(4):495–527.

Dorp, P. v. (2016). Large-eddy simulation of wind farms in clear and cloud-topped

boundary layers.

Efstathiou, G. A., Plant, R. S., and Bopape, M.-J. M. (2018). Simulation of an evolv-
ing convective boundary layer using a scale-dependent dynamic smagorinsky
model at near-gray-zone resolutions. Journal of Applied Meteorology and Climatol-
ogy, 57(9):2197–2214.

Essa, K. S., Ghobrial, R. A., Mina, A., and Higazy, M. (2011). Calculation of cross-
wind integrated concentration by using different dispersion schemes. Mausam,
62(1):51–60.

Grylls, T., Suter, I., and van Reeuwijk, M. (2020). Steady-state large-eddy simula-
tions of convective and stable urban boundary layers. Boundary-Layer Meteorol-
ogy, 175(3):309–341. Cited By :3.

Grylls, T. and van Reeuwijk, M. (2021). Tree model with drag, transpiration, shading
and deposition: Identification of cooling regimes and large-eddy simulation.
Agricultural and Forest Meteorology, 298:108288.

Heus, T., Van Heerwaarden, C. C., Jonker, H. J. J., Pier Siebesma, A., Axelsen, S.,
Van Den Dries, K., Geoffroy, O., Moene, A. F., Pino, D., De Roode, S. R., and
De Arellano, J. V. . (2010). Formulation of the dutch atmospheric large-eddy
simulation (dales) and overview of its applications. Geoscientific Model Develop-
ment, 3(2):415–444. Cited By :133.

Jia, H. and Kikumoto, H. (2021). Source term estimation in complex urban environ-
ments based on bayesian inference and unsteady adjoint equations simulated
via large eddy simulation. Building and Environment, 193.

Koene, M. (2020). Implementation of ibm in dales.

55
Letzel, M. O., Krane, M., and Raasch, S. (2008). High resolution urban large-eddy
simulation studies from street canyon to neighbourhood scale. Atmospheric
Environment, 42(38):8770–8784.

Lilly, D. K. (1967). The representation of small-scale tubulence in numerical simula-

tion experiments. Proceedings of IBM Scientific Computing Symposium on Environ-
mental Sciences, Edited by: Goldstine, H. H.:195–210.

Mao, S., Lang, J., Chen, T., Cheng, S., Cui, J., Shen, Z., and Hu, F. (2020). Compari-
son of the impacts of empirical power-law dispersion schemes on simulations
of pollutant dispersion during different atmospheric conditions. Atmospheric
Environment, 224. Cited By :2.

Nieuwstadt, F. and Brost, R. (1986). The decay of convective turbulence. Journal of

Atmospheric Sciences, 43(6):532–546.

Pirhalla, M., Heist, D., Perry, S., Tang, W., and Brouwer, L. (2021). Simulations of
dispersion through an irregular urban building array. Atmospheric Environment,
258.

Resler, J., Krč, P., Belda, M., Juruš, P., Benešová, N., Lopata, J., Vlček, O., Damašková,
D., Eben, K., Derbek, P., et al. (2017). Palm-usm v1. 0: A new urban sur-
face model integrated into the palm large-eddy simulation model. Geoscientific
Model Development, 10(10):3635–3659.

Smagorinsky, J. (1963). General circulation experiments with the primitive equa-

tions: I. the basic experiment. Monthly weather review, 91(3):99–164.

Sommeria, G. (1976). Three-dimensional simulation of turbulent processes in

an undisturbed trade wind boundary layer. Journal of Atmospheric Sciences,
33(2):216–241.

Tomas, J., Pourquie, M., and Jonker, H. (2016). Stable stratification effects on flow
and pollutant dispersion in boundary layers entering a generic urban environ-
ment. Boundary-Layer Meteorology, 159(2):221–239.

Tominaga, Y., Mochida, A., Murakami, S., and Sawaki, S. (2008). Comparison of
various revised k– models and les applied to flow around a high-rise building
model with 1:1:2 shape placed within the surface boundary layer. Journal of
Wind Engineering and Industrial Aerodynamics, 96(4):389–411.

Verzijlbergh, R., Jonker, H., Heus, T., and Vilà-Guerau de Arellano, J. (2009). Tur-
bulent dispersion in cloud-topped boundary layers. Atmospheric Chemistry and
Physics, 9(4):1289–1302.

Vreugdenhil, C. B. and Koren, B. (1993). Numerical Methods for Advection-Diffusion

Problems, volume 45. Vieweg.

Walton, A. and Cheng, A. (2002). Large-eddy simulation of pollution dispersion

in an urban street canyon—part ii: idealised canyon simulation. Atmospheric
Environment, 36(22):3615–3627.

WHO, W. H. O. (2016). World health statistics 2016: monitoring health for the SDGs
sustainable development goals. World Health Organization.

Wu, Z. and Liu, C. . (2018). Source depletion analogy for reactive plume dispersion
over schematic urban areas. Atmospheric Environment, 190:226–231. Cited By :2.

Xie, Z.-T., Coceal, O., and Castro, I. P. (2008). Large-eddy simulation of flows over
random urban-like obstacles. Boundary-layer meteorology, 129(1):1–23.
Xie, Z.-T., Hayden, P., and Wood, C. R. (2013). Large-eddy simulation of
approaching-flow stratification on dispersion over arrays of buildings. Atmo-
spheric Environment, 71:64–74.

Yang, G., Li, X., Ding, L., Zhu, F., Wang, Z., Wang, S., Xu, Z., Xu, J., Qiu, P., and
Guo, Z. (2019). Cfd simulation of pollutant emission in a natural draft dry cool-
ing tower with flue gas injection: Comparison between les and rans. Energies,
12(19).

57
A
Inputfiles for the DALES
simulations

DALES requires several inputfiles in order to perform a simulation properly, with

the file-type ASCII. The settings are given in file called namoptions, and several
initial vertical profiles are given in ’prof.inp’ and ’lscale.inp’, where the later con-
tains the large scale forcings. The large scale fluxes causing the convective of stable
boundary layer are prescribed to dales in the file ’ls flux’. The different settings
between the SBL and the CBL are noted as such.

A.1 Settings - namoptions

&RUN
iexpnr = 001
lwarmstart = .false.
runtime = 28800 (s)
irandom = 43
randthl = 0.1
randqt = 0
nsv = 4 number of passive scalars: 1 line source and 3 point sources
ladaptive = .true.

&DOMAIN all specifications containing the domain size

itot = 160 amount of horizontal grid points in x-direction
jtot = 160 amount of horizontal grid points in y-direction
kmax = 167 amount of vertical grid points
xsize = 800. domain size in x-direction (m)
ysize = 800. domain size in y-direction (m)

59
&PHYSIC
ltimedep = .true. setting to include timedependent fluxes
lmoist = .false. no moisture
iradiation = 0 no radiation
lcoriol = .false. no coriolis force
wsvsurf = 0.001, 0.001, 0.001, 0.001
prescribed surface scalar flux (kg kg−1 ms−1 )

&NAMSURFACE
z0 = 0.1
ps = 101900.00 (Pa)
isurf = 2
thls = SBL: 290.16, CBL: 285.98 (K)
wqsurf = 0 (kg kg−1 ms−1 )

&DYNAMICS
llsadv = .false.
lqlnr = .false.
cu = 0.
cv = 0.
iadv mom = 2 , Kappa:7 advection scheme, 2 = second order, 7 = Kappa
iadv tke = 2
iadv thl = 2
iadv qt = 2
iadv sv = 2

&NAMSCALAR additional settings for heterogeneous scalars

lscalar = .true. switch to implement the scalar module
lboundary = .true. switch to nudge turbulent fluxes at domain edges
nudgedepthgr = 5 number of nodes from domain edges from which
the nudging is applied

The addition of obstacles:

&NAMRURALBOUNDARY
lruralboundary = .true. switch to implement the IBM module
ldefrural = .true. input file with the obstacle height is prescribed
lnoslip = .false.
lwallfunc = .true.
lfluxform = .true.
lpoislast = .true.
lthlwalltimedep = .true. switch to depend the wall temperature on the
prescribed timedependent temperatures
thlwall = SBL: 290.16, CBL: 285.98 (K), initial wall temperature of
buildings
ct = 0.
A.2 Vertical profiles - prof.inp
This inputfile contains the initial vertical profiles at the height of the half-nodes,
e.g. the SBL simulation has a vertical resolution of 3m, thus the vertical profiles
are specified starting from 1.5m, to 4.5m, etc., and For the CBL simulation with a
∆z = 5m, the heights specified are 2.5m, 7.5m, etc.
The prescribed variables are the initial profiles of the liquid potential temperature θl
(K), the total water specific humidity qt (kg k−1 ), the wind velocity in North-South
direction (u) and East-West direction (v) in ms−1 , and the turbulent kinetic energy
TKE in m2 s2 . The vertical profiles of θl differs for the SBL and CBL simulations
which are specified in chapter 4.3, as is the vertical profile of the TKE. This study
does not take any moisture into account, resulting the qt = 0 kg k−1 for all heights.
The desired wind direction is East, making u = 0 ms−1 for all heights. The imple-
mented initial wind velocity is 5 ms−1 , in the Eastern direction makes the value of
v for all heights −5 ms−1 .

A.3 Large scale forcings - lscale.inp

The large scale forcings, such as the geostrophic wind velocity u geo and v geo and the
subsidence are specified in this inputfile. The values of u geo and v geo are identical to
the values of u and v as given above, and the prescribed subsidence is a function of
height and the divergence. Chapter 4.2 gives insight into the equations, values and
vertical profile of the subsidence.

A.4 Large scale fluxes - ls flux.inp

The ls flux inputfile consist of two parts. The first part contains the fluxes of several
variables which can be specified for several timesteps, and the second is identical
to the lscale.inp input file for every specified timestep in the first part.
The only time dependent variable is the liquid surface temperature (K), which is
here specified for every hour, following the course of the nocturnal surface temper-
ature of 16 August 2016 for the SBL simulation and the diurnal θl,sur f of 17 August
2016 for the CBL simulation. Figure 4.1 in chapter 4.3 given a schematic overview
of the temperature courses of both simulations.

A.5 Additional inputfiles

Beside these four inputfiles, the performed simulations also require three other
input files. If passive scalars are prescribed DALES, also requires an initial vertical
profile of all the specified scalars, which are all 0 at the start of the simulation. The
other two inputfiles are the locations of the passive scalars and the locations of
the buildings. The specifics of these input files are given is chapter 2.7 and 2.5.2,
respectively.

61
B
Transition of GABLS1 to SBL
The stable boundary layer implemented for this study is based on the SBL simulated
by the GABLS1 case. However, as stated in chapter 4.8, there are several important
differences between the two simulations. These variables are changed one by one
to study the impact of the different modifications on the stability of the boundary
layer of the simulation.

Of these alterations, the decrease in wind speed and the hourly decline of the
virtual temperature are expected to have the largest impact. In a stable boundary
layer at night time, when the surface is no longer heated by the radiation of the sun,
the largest remaining factor of turbulence in the shear turbulence, driven by the
wind velocity. A smaller wind velocity automatically translates to a smaller shear
turbulence. A larger decrease of the surface temperature per hour will probably
mean that the (negative) buoyancy will increase. Over all, the impact of all changes
are expected to have a dampening effect on all considered variables. First, figure
B.1 gives insights into the buoyancy flux of the GABLS1 simulations in comparison
with all incremental alteration.

The first three changes ( a)domain, b)wind velocity and c)surface temperature)
don’t seem to impact the buoyancy flux significantly. However, the larger decrease
of the hourly surface temperature flux, which as table 4.1 presents, is tripled. This
almost directly translates to the buoyancy flux, which also triples. The buoyancy
is the upward acceleration of air, mainly caused by temperature differences. There-
fore it holds that an increase in the (negative) temperature flux brings an increase
in the buoyancy flux. Interesting is the decrease in the buoyancy flux due to a lower
wind velocity, whereas this increase can not be directly seen when this is the only
changed variable (figure B.1.b.). A smaller wind velocity has a direct impact on the
turbulence, where the shear turbulence is mostly based on the size of the wind ve-
locity. This decrease in shear TKE due to a smaller wind velocity, could impact the
ratio at which the air at the surface is mixed with the air above. In the simulation
with all differences combined, the surface temperature decrease over time is larger,
however, its effect on the buoyancy flux is probably restrained by the decrease of
shear TKE.

63
Figure B.1: The vertical profiles of of the total buoyancy flux of the GABLS simulation in com-
parison to small alterations to the desired simulation state, where a) changes the
domain size, b) the wind velocity, c) the virtual surface temperature, d) the hourly
virtual surface temperature flux and e) gives a comparison of all the changes to
the original case study. The alterations in variables are given in the legend of
the graph, all unnamed variables are the unchanged variables of the GABLS
simulation as displayed in table 4.1.

A similar effect is seen in figure B.2, where the total and the sub-surface TKE re-
main relatively constant, with a significant increase of the total TKE when ∂θvs ur f /∂t
increases and a decrease in turbulence when the wind velocity of 8 ms−1 is replaced
by 5 ms−1 . Advantageous is the decreases vertical profile of the SFS-TKE in the de-
sired simulation (in red in figure B.2.e). This means that most of the turbulence is
resolved by the model, whereas the other simulations portray a SFS-TKE/Total TKE
ratio proximating 0.9. This ratio in the simulated SBL implemented in this study is
smaller than 0.4. These are shown in figure B.3.
Figure B.2: The vertical profiles of of the total turbulent kinetic energy (TKE) and the sub-
filter surface TKE of the GABLS simulation in comparison to small alterations to
the desired simulation state, where a) changes the domain size, b) the wind veloc-
ity, c) the virtual surface temperature, d) the hourly virtual surface temperature
flux and e) gives a comparison of all the changes to the original case study.

Figure B.3: The vertical profiles of of the ratio of the SFS-TKE to the total turbulent kinetic
energy (TKE) of the GABLS simulation in comparison to small alterations to the
desired simulation state, where a) changes the domain size, b) the wind velocity,
c) the virtual surface temperature, d) the hourly virtual surface temperature flux
and e) gives a comparison of all the changes to the original case study.

65
C
Results of the Point Sources
The analyses performed for the line source are repeated for the the point sources.
Figure C.1 gives the horizontal and vertical distribution of the concentrations of the
SBL simulations of first the Gaussian plume model, than the DALES simulation
without obstacles, followed by the DALES simulation with obstacles. If an horizon-
tal or vertical distribution is given of one point source, the results of the middle
point source are displayed, at y = 430m.

The plumes of the concentrations obtained from the Gaussian plume model are
larger than the plumes calculated by DALES, and are larger downwind form the
sources. Moreover, they reach higher elevations than the concentrations in the
DALES simulations. The reason for these differences in results should be further in-
vestigated, as to which model is a more accurate portrayal of a surface point source
in a stable boundary layer without obstacles.

As already seen with the result of the line source, the concentrations upwind from
the source and around the edges of the plume have negative values in the DALES
simulations, due to the applied second order advection scheme. Furthermore, the
concentration of the pollutant is significantly more concentrated near the source
when buildings are present, then when they are not. However, the plume direction
with the presence of buildings is still heavily influenced by the wind direction.

Figure C.2 display the same figures as figure C.1 for the convective boundary layer.
As concluded by the line source analyses, the concentrations as obtained from the
Gaussian plume model of the CBL are considerably higher than the concentrations
as calculated by DALES.
The presence of buildings has a similar effect on the concentrations in the CBL as
in the SBL, however less significant: the plume of the concentration of the pollutant
is more concentrated at the source. The plume direction is however highly affected
by the obstacles in a CBL.

67
 (a) (b)

(c) (d)

(e) (f)

Figure C.1: Vertical and horizontal profiles of the three point sources of time-averaged con-
centrations of the SBL as obtained from the Gaussian plume model and the
DALES simulation, with obstacles and without.
 (a) (b)

(c) (d)

(e) (f)

Figure C.2: Vertical and horizontal profiles of the three point sources of time-averaged con-
centrations of the CBL as calculated by the Gaussian plume model and the
DALES simulation, with obstacles and without.

69
 Figure C.3 gives the surface concentrations of both the SBL and CBL simulations
as obtained from DALES, comparing the concentrations with and without obstacles.
As stated before, the concentrations of the simulations with buildings are highly
concentrated at the source, regardless of the stability of the atmosphere.

Figure C.3: The time-averaged point source concentrations χ in the downwind direction and
the percentages of the total concentration. The red lines represent the results of
the simulation with buildings and the red without. a) and b) give the concen-
tration and the percentages of the SBL simulations, and c) and d) of the CBL
simulations.

Figure C.4 gives the vertical profiles of the pollutants concentrations several dis-
tances from the source of the SBL and CBL simulations with and without buildings.
The presence of buildings increases concentrations at higher levels in the atmo-
sphere, and the concentrations is more concentrated around the source.

Figure C.4: The time-averaged, spatial averaged, vertical profiles of the concentration at sev-
eral downwind distances from one of the point sources. a) and b) give the con-
centration and the percentages of the SBL simulations, and c) and d) of the SBL
simulations.
Colophon
This document was typeset using LATEX. The document layout was generated using
the arsclassica package by Lorenzo Pantieri, which is an adaption of the original
classicthesis package from André Miede.