Masterthesis Borander Pernille

Analysis of urban meteorological data
Comparison of local measurements and a numerical

mesoscale model under the aspect of local air quality
modeling
Pernille Borander
Thesis submitted for the degree of

Master of science in Meteorology and Oceanography
60 Credits
Department of Geosciences
Faculty of mathematics and natural sciences
UNIVERSITY OF OSLO
August 15, 2018

c 2018 Pernille Borander
Analysis of urban meteorological data

Comparison of local measurements and a numerical mesoscale model under the aspect of
local air quality modeling
This work is published digitally through DUO – Digitale Utgivelser ved UiO
http://www.duo.uio.no/
Printed: Reprosentralen, University of Oslo
All rights reserved. No part of this publication may be reproduced or transmitted, in any form
or by any means, without permission.
Abstract
This thesis explores the meteorology in an urban area during a winter period and how
well a numerical model is able to represent it. The air quality in the cities is of large
interest to the citizens and policymakers. However, effects of buildings and anthropogenic
activity on the meteorology in the city are complex. Lack of meteorological observations
makes it difficult to create good predictions of the dispersion of tracers and to evaluate the
models used as well. Meteorological modeling with focus on the planetary boundary layer
(PBL) over Bjørvika, Oslo is performed using the Weather Research and Forecast (WRF)
model combined with the single-layer urban canopy model (SLUCM). The simulated
meteorological parameters are evaluated by comparison with observational measurements
of the temperature at six levels, and the wind speed and the wind direction at two levels in
a tower crane mast in Bjørvika, from January 9 to March 1, 2018. The results show a large
difference of the meteorology between the model and the observations. The wind speeds
are overestimated and the temperatures are underestimated by WRF. Furthermore, the
atmosphere in the model is more stable compared to the observations when comparing
hourly values, indicating that the model parameters will not represent the local dispersion
correctly. However, the summed distribution of the atmospheric stability in WRF was
more similar to the observations, indicating that the parameters from WRF could work
better for dispersion estimations of longer timescales.
ii
Acknowledgements
Consuming all my time, energy and thoughts the last months, it is surreal to write the
final words of my master thesis. First, I would like to thank my supervisor, Erik Berge, for
your guidance, valuable discussions and that you have patiently answered my questions
(more than once). The work with this thesis brought me to new heights! 80 meters above
ground, to be exact. Many thanks also to my co-supervisor, Terje Koren Berntsen, for
early morning meetings and that your door always was open.
This thesis has been a technical challenge, and many have contributed to make this
study possible. Thank you to Veidekke AS and especially to their head of HSE, Geir
Erling Boman, for facilitating the use of your tower crane, loan of protective equipment
and the nice chats. A big thanks to John Hulth at the Department of Geosciences, for
all your time devoted to planning, instrumenting and maintaining the measuring sensors
in the Munch mast. At the MetOs section there are many incredible helpful people! A
huge thank you to Anne Claire Fouilloux. This lady is absolutely fantastic! Thank you
for all the hours you have assisted with WRF and python codes, and that you have found
solutions to frustrating errors, no matter the day of the week or the the hour of the day
they occurred. A big thanks to Kjetil Schanke Aas for always having time to answer my
WRF questions. Additionally, there is a lovely group of students at MetOs. Thanks to
you all for some great years with nice conversations and a good social environment. The
people of the Norwegian meteorological society are very helpful as well, and I am sending
a general thanks to the persons at Kjeller Vindteknikk, the Norwegian Meteorological
Institute, NILU, NIVA and Norconsult for answering my mails the last year
I want to thank my family. Pappa, Mamma, Marius, Jørgen, Erlend and the rest of
the clan, thank you for all your support and that you always have been a safe haven to
come home to. A special thanks to tante Anne-Katrine, Anne-Siri, Venke og Trond for
proofreading. A large thanks to my friends outside the University for coping with thesis
complains and for your support. Last, I need to thank "Faministene", my study friends
who I met during the first weeks of the bachelor. Without you, Ingrid, Mari, Elisabeth,
Vilde, Helene and Helle, it is difficult to tell if this girl who came straight out of a year
with musical at Follo FHS, would be sitting with her master thesis in meteorology in her
hands five years later. But the girl met you! Thank you for all our adventures and that
you gave me some fantastic years at the University of Oslo. Thank you!
Oslo - August 15, 2018

Pernille Borander
iv
Contents
Abstract ii
Acknowledgements iv
1 Introduction 2
1.1 Motivation and Background . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Outline of this Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Theory 6
2.1 The Planetary Boundary Layer . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.1 Heat Exchange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 Momentum Exchange . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Dispersion in PBL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.1 Advection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.2 Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Atmospheric Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.1 Static Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.2 Dynamic Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.3 The Richardson Number . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.4 The Obukhov Length . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.5 Pasquill’s Stability Classes . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.6 Effect of Stability on Wind Shear . . . . . . . . . . . . . . . . . . . 16
2.4 Special Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.1 Urban Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4.2 Ocean Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.3 Topographic Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3 The Observations 22
3.1 Description of Bjørvika . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.1 The Munch mast . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Instrumentation of the Munch mast . . . . . . . . . . . . . . . . . . . . . . 25
3.3 Calibration of Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3.1 The Temperature Observations . . . . . . . . . . . . . . . . . . . . 26
3.3.2 The Wind Observations . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 MET’s Weather Stations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4.1 Description of Hovin and Blindern . . . . . . . . . . . . . . . . . . 30
3.4.2 Instrumentation of the Hovin and Blindern masts . . . . . . . . . . 31
vi
1 CONTENTS
3.5 Comparison of Measurements . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.5.1 The Temperature Observations . . . . . . . . . . . . . . . . . . . . 32
3.5.2 The Wind Observations . . . . . . . . . . . . . . . . . . . . . . . . 34
4 The Model 38
4.1 The Weather Research and Forecast Model . . . . . . . . . . . . . . . . . . 38
4.1.1 WRF Modeling System . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.2 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.1 Domain Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2.2 Initialization and Boundary Conditions . . . . . . . . . . . . . . . . 42
4.2.3 Choices of Physic schemes . . . . . . . . . . . . . . . . . . . . . . . 43
4.3 Simulations and Processing . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.4 WRF Data Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.5 The SST-Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5 Results and Discussion 50

5.1 Model Comparison to Observations:
Selected Case (January 23 - January 26, 2018) . . . . . . . . . . . . . . . . 50
5.1.1 Meteorological Conditions . . . . . . . . . . . . . . . . . . . . . . . 50
5.1.2 The Munch mast . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.1.3 The MET Stations . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Whole Period (January 9 - March 1, 2018) . . . . . . . . . . . . . . . . . . 62
5.2.1 The Munch mast - Temperature . . . . . . . . . . . . . . . . . . . . 62
5.2.2 The Munch mast - Wind Speed . . . . . . . . . . . . . . . . . . . . 65
5.2.3 The Munch mast - Wind Direction . . . . . . . . . . . . . . . . . . 67
5.2.4 The Munch mast - Atmospheric Stability . . . . . . . . . . . . . . . 68
5.2.5 The MET Stations . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6 Summary and Conclusions 78
Appendices 82
A Additional Figures and Tables 82
B namelist.input 92
Bibliography 96
Chapter 1
Introduction
1.1 Motivation and Background

There is an increasing concern among inhabitants of Norway to what they inhale from the
air. Particulate matter and air quality are no longer topics only of interest for scientist,
but also for the average man on the street. This may not be surprising when you can
read on World Health Organization Europe’s (WHO) website that "Air pollution from
both outdoor and indoor sources represents the single largest environmental risk to health
globally" (WHO, 2017). WHO have estimated that six million premature deaths were
linked to exposure to air pollution in 2012 alone. Some may think that air quality is not a
big problem in a small nation as Norway. However, research has shown adverse health ef-
fects and serious ailments due to air pollutants at low concentrations, and thus something
that can not be ignored (Folkehelseinstituttet, 2013). As a matter of fact, Norway was
convicted in the European Free Trade Association (EFTA) Court on the 2th of October
2015 for violations of the EU Air Quality Directive for exceeding the local air quality limit
values for sulfur dioxide (SO2 ), particulate matter (P M10 ) and nitrogen dioxide (N O2 ).
In addition, the Norwegian municipalities’ local air quality investigations did not meet the
requirements of the Directive. The verdict included a number of municipalities including
Oslo (ESA, 2015).
Observations of tracers (e.g Gjerstad et al., 2012) and numerical dispersion models (e.g.
Borrego et al., 2003; Høiskar et al., 2016) are often used when investigating air quality. It
is crucial to have information on the meteorology when using a dispersion model. Mete-
orological circumstances control air motion. Numerical dispersion models need to be fed
with meteorological parameters like wind speed, wind direction and the atmospheric sta-
bility to be able to simulate dispersion and transport of the pollutants. The meteorological
data can come from observations, or they can come from numerical models developed to
simulate the atmosphere and its processes. Such atmospheric models are usually Eulerian.
Eulerian models divide the area of interest into grid boxes and solve complex equations
for each grid box. The Weather Research and Forecasting model (WRF) (Skamarock et.
al, 2008) is one example and is used in this study.
If we want to explore dispersion of pollutants close to the ground, in the planetary bound-
ary layer (PBL), the effect of turbulence has to be accounted for. Turbulence is what we
2
3 CHAPTER 1. INTRODUCTION
call the apparently random and chaotic nature of fluid flows (Arya, 1999). Turbulence is
very hard to measure and even harder to model. We need an infinite number of equa-
tions to describe turbulence in an exact mathematical manner (Wallace and Hobbs, 2006),
referred to as the "closure problem" of turbulence. So we turn to parameterization. An-
other challenge faced when investigating the meteorology close to the ground is the great
variation in topography that may exist. Whether the area studied have hills, mountains,
valleys or lakes, or have forests, crops or cities with tall buildings, there will be different
effects on the nearby meteorology (Oke, 1987). Turbulence and exchanges of momentum
and heat between the surface and the atmosphere are especially affected, which again
play a part in all other meteorological motions in the PBL. The local features are not
necessarily easy to resolve or parameterize in numerical models. There are quite a num-
ber of studies which have investigated the performance of the PBL parameterizations in
WRF (e.g. Shin and Hong, 2011; Krogsæter and Reuder, 2013; García-Díez et. al, 2013).
The complexity can not be ignored. There are naturally most interest in knowing the air
quality in areas where people live, which for many of are in the cities. However, there is
a lack of meteorological observations of the structure of the lower part of PBL in urban
areas for both air quality modeling and for evaluating the numerical models that are used
to simulate the meteorology. Trying to arrange field campaigns to acquire observations
are often expensive and the task of parameterize the PBL phenomenon is as García-Díez
et. al (2013) states ...one of the most challenging problems in numerical modeling of the
atmosphere.
A problem that arises when not having sufficient observations like the vertical profile
of wind and temperature, is during wintertime when there may be inversion situations.
Inversions events occur when air mases are very stable. Cold air lies closest to the surface
while warmer air lies on top, resulting in little mixing of the air. When there is little
mixing of the air, the concentration of pollutants can become very high and can damage
the air quality substantially. Under these events the wind flow and the temperature can
vary much in small vertical distances. If we do not have the means to forecast inversion
events, it is difficult to initiate restrictions to reduce the pollution, like a temporary ban
of diesel cars. In general, the stability of the PBL is important for the dispersion. A very
stable PBL challenges the already fragile theory of turbulence and surface fluxes. Many as-
sumptions are no longer valid with drastic increasing stability (Tastula and Vihma, 2011).
There is a wish, and need, to make good estimations and forecasts of the air quality.
The authorities demand it. They want reports in the areas where people live and breath
such as in cities. The atmospheric motion in these areas are influenced by the surround-
ing features of the surface. Hence, it is necessary to handle PBL meteorology which
can not be described completely by todays theory or numerical models. There is a lack
of crucial observations to make good estimations and good evaluations of the simulated
near-surface meteorology and air dispersion in the cities. To complicate things further,
seasonal phenomenons like winter inversions need to be considered.
1.2. OUTLINE OF THIS STUDY 4
1.2 Outline of this Study

The purpose of this master thesis was to investigate the local meteorology in an urban
area intended for use in an air quality dispersion model. This was attempted by collect-
ing observations at multiple vertical heights and apply an atmospheric numerical model,
which resolved local meteorological processes. This study wished to explore how good the
model available today represent local scale meteorological parameters of importance to
dispersion modeling.
Wind and temperature sensors were installed in the mast of an 80m tall tower crane
in Bjørvika, Oslo, currently involved in the construction of the new Munch museum. The
temperature was measured at 2m, 10m, 30m, 50m, 65m and 75m, and wind speed and
direction at 10m, 30m and 75m. As fare as the participants in this thesis know, such mea-
surements do not exist anywhere else in the Oslo area. The measuring period stretched
from 9th of January to 1th of March 2018 (i.e. a total of 52 days). The raw data from
the mast were compared to observations done by the Norwegian Meteorological Institute.
This thesis applied the Weather Research and Forecast Model (WRF), a popular and
widely used model due to its open source and the users possibility for modification. WRF
can resolve fine-scale meteorological processes and allow high spatial resolution (Ska-
marock et al., 2008). WRF is also commonly used as source of meteorological data in
local dispersion models (Solberg and Svendby, 2012). The meteorological parameters from
the WRF simulations were compared to the observations and differences were discussed.
In addition, a shorter "selected case" where a statically stable situation was examined
more closely.
In Chapter 2 we describe the atmospheric processes that have a crucial role in the mete-
orology in the PBL and the local dispersion meteorology. Chapter 3 present the observa-
tions, the instrumentation of the measuring mast and the validation of the observations,
while Chapter 4 gives a description of the numerical model, the model setup and sim-
ulations. The results are given and discussed in Chapter 5, before a summary and the
conclusions are presented in Chapter 6.
Chapter 2
Theory
In the following chapter the local meteorology theory, which is important for the trans-
port of tracers, will be discussed. Some surface characteristics that have special effects
on the dispersion, for example cities, ocean and topographic features will also be given.
The theory presented here is mainly based on Oke (1987), Stull (1988) and Arya (1999).
Atmospheric processes occur over a broad range of space and time scales. From a millionth
of a meter to ten thousands of kilometers on the space scale, and from under a second
to months on the timescale. Examples are shown in Fig. 2.1. Atmospheric motions
and related phenomena are divided into three broad categories: macroscale, mesoscale
and microscale. Micrometeorology is the study of the microscale or local-scale phenom-
ena and processes occurring on a horizontal length scale of a few kilometers or less, and a
timescale of an hour or less. A key issue with micrometeorology and boundary layer mete-
orology is the important exchanges of heat, mass and momentum. These exchanges occur
continuously between the atmosphere and the earth’s surface. The role of the vertical
distributions of meteorological variables as wind speed, wind direction and temperature
are important to the boundary layer meteorology as well. This is why observations of the
vertical wind and temperature profiles are of great interest.
2.1 The Planetary Boundary Layer

The planetary boundary layer (PBL), also commonly referred to as the atmospheric
boundary layer, is defined by Stull (1988) as "the part of the troposphere that is directly
influenced by the presence of the earth’s surface, and responds to the surface forcing with
a timescale of about an hour or less". The PBL is especially recognized by its turbulent
mixing driven by the frictional drag of the Earth’s surface (i.e. exchange of momentum)
and the vertical rising of air due to solar heating of the ground (i.e. exchange of heat).
The strength of the surface generated mixing in the PBL is not constant, thus the height
often called the mixing height/depth of the PBL varies both in time and in space. For
example if the Earth’s surface is well heated, as it will be by the sun during daytime in
the summer months, the heat will be transferred upwards into the cooler atmosphere.
This thermal mixing allows the boundary layer depth to expand up to 2km, limited by
a capping inversion separating the PBL from the above atmosphere (i.e. the free atmo-
sphere). On the contrary, if the surface cools more rapidly than the atmosphere as it will
6
7 CHAPTER 2. THEORY
Figure 2.1: Scale definitions and the characteristic time and horizontal length scales of
a variety of atmospheric phenomena, adapted from Markowski and Richardson (2010,
Fig. 1.1, p.4).
during the summer nighttime, the heat is transferred downwards from the atmosphere to
the ground. The effect now is that the turbulence is suppressed and the PBL depth may
shrink to just 100m. This picture of the varying PBL depth only dependent on the daily
cycle of the sun in the summer, is the ideal picture. During the winter when our study
took place, there is much less solar heating of the surface and thus low heating of the air
above. This results in little mixing of the air, hence the wintertime PBL do not have as
distinct diurnal cycle as the summertime PBL.
We often divide the PBL into different layers aligned approximately parallel to the sur-
face. The lowest layer is the roughness layer. The depth of this layer is dependent on the
dimensions of the surface roughness elements like buildings, trees etc, and extends usually
1 to 3 times their height. Because of the roughness elements the flow in the roughness
layer may be very complex and highly irregular. Above the roughness layer is the surface
layer. The surface layer is often assumed to be the bottom 10% of the PBL where the heat
and momentum fluxes varies less than 10% of their magnitude. Since the vertical fluxes
usually are said to be constant, the surface layer is often called the constant flux layer.
The flow in the surface layer is characterize by intense small-scale turbulence due to the
effect of the surface roughness and convection. Above the surface layer is the so called
2.1. THE PLANETARY BOUNDARY LAYER 8
Figure 2.2: Schematic of the vertical layers in the planetary boundary layer (PBL),
adopted from Arya (1999).
outer layer which extend up to the top of the PBL (i.e. 90% of its depth). The depth
and flow of this layer depends on the strength of the surface generated by turbulence as
mentioned before. The vertical layers of the PBL can be seen in Fig.2.2.
2.1.1 Heat Exchange

The absolute greatest source of energy to atmospheric motion is the sun. The radiation
from the sun is absorbed directly by the atmosphere through aerosols, clouds and gas
molecules such as ozone, O3 , but mostly absorbed by the surface of the earth or reflected
back depending on the surface albedo. The exchange of energy between the earths surface
and the atmosphere when looking at a flat, homogeneous surface can be expressed by the
surface energy balance equation
Rn = Qh + Qe + Qg (2.1)
where Rn is the net radiation received by the surface, Qh is the sensible heat flux to the
atmosphere, Qe is the latent heat flux to the atmosphere and Qg is the heat flux to the
ground. The net radiation is the difference between the incoming shortwave and longwave
radiation and the outgoing longwave radiation. The net radiation is defined positive when
incoming exceeds the outgoing. The sensible heat flux depend on the temperature differ-
ence between the surface and the air above. This is normally transferred down the mean
temperature gradient (i.e. from a warm surface to colder air or vice versa). The latent
heat flux is due to evaporation or condensation from or to the surface, and is transferred
9 CHAPTER 2. THEORY
Figure 2.3: Schematic of the surface energy budget during daytime and nighttime,
adopted from Markowski and Richardson (2010).
similarly to the sensible heat flux down the mean gradient of specific humidity or mixing
ratio. The ground heat flux depends on the soil type and moisture content and is usually
just 10% of the net radiation. The non-radiative fluxes are defined to be positive when
they are directed away from the surface and negative when directed to the surface, as
shown in Fig. 2.3. The transfer of heat is mainly done by turbulence, but in the first
millimeters of air in the interface between the PBL and the surface, molecular conduction
of heat initiates the process. After that, the turbulent motion takes over, often referred
to as thermal generated turbulence or free convection.
Sometimes it can be more appropriate when discussing the surface energy balance, to
picture the surface as a volume or layer that can have heat content itself and not as a
massless plane. The Eq. 2.1 needs then an additional term, the changes of energy storage,
∆Qs
Rn = Qh + Qe + Qg + ∆Qs (2.2)
The change of energy storage describes the energy absorption or release from the volume.
∆Qs 6= 0, indicates that the input and output to one or more of the other fluxes in the sur-
face energy balance do not balance and there is either a net energy storage gain (+∆Qs ),
hence warming of the volume, or a net energy storage loss (−∆Qs ), hence cooling of the
volume. ∆Qs = 0, indicate no net change in energy status or the temperature of the
volume, and the fluxes are in balance
During winter the heating from the sun gets weaker and the exchange of heat between
the surface and the PBL decrease, and it gets colder. The likelihood of snow and ice on
the surface is high. The greatest impacts of snow and ice on the surface energy balance is
that ice and snow allow transmission of short-wave radiation, and that their high albedo
greatly reflects the incoming radiation. Even a thin layer of snow cover can have a great
effect on the surface albedo (Oke, 1987). If the temperature is below 0◦ C, Qe in Eq. 2.2
2.1. THE PLANETARY BOUNDARY LAYER 10
Figure 2.4: Illustration of the effect of roughness on the wind speed profile near the
ground, in a strong wind and weak thermal environment. The height zg is the top of
the PBL. Adopted from Oke (1987).
is likely negligible because of lack of liquid water evaporation and little vapor which can
condensate. Qg and ∆Qs are also negligible because of the lack of solar heating and the
low conductivity of snow. The resulting energy balance is between Rn and Qh with a net
radiative sink and a convective sensible heat source.
2.1.2 Momentum Exchange

Because the PBL is in contact with the surface of the earth, the rigid surface exerts
a frictional drag on the atmosphere. A consequence of this is exchange of atmospheric
momentum from the PBL to the surface. In the same way as with heat exchange, the
momentum transfer is dominated by molecular processes in the first millimeters of the
air before turbulent transport takes over. The surface drag controls the wind field and
since air is a viscous fluid, it must come to rest right at the surface. The effect of this is
that the mean horizontal wind speed decrease sharply towards the surface. The depth of
the influence of the frictional drag depend on the roughness of the surface in absence of
strong thermal effects. Increasing surface roughness increase the depth of influence, thus
the vertical gradient of mean wind speed is greatest over smooth surfaces and decrease
with increasing roughness. This is illustrated in Fig. 2.4 where zg is the PBL height.
The surface drag is negligible above this height. Another consequence of the momen-
tum exchange and wind shear is the generation and maintenance of turbulence. This
is called mechanical generated turbulence of forced convection, which in turn influences
other turbulent exchanges as seen in dispersion of pollutants.
11 CHAPTER 2. THEORY
2.2 Dispersion in PBL

The surface of the Earth is as we have already established a large source of different
species. When there are emissions to the atmosphere the rate of change of the concentra-
tion c(x, t) of the specie with time can be expressed by the mass balance equation for a
chemical specie (Seinfeld and Pandis, 2016)
∂c
= −∇ · (uc) + R + E − S (2.3)
∂t
where the first term on the right hand side of the equation represents the atmospheric
transport of the specie, where u is the velocity vector, the second term, R, represent the
chemical generation for the specie, E is the emission and the last term, S, is the removal
fluxes of the specie. Since we are most interested in the atmospheric transport which can
be separated into transport by advection and transport by turbulence in this study. Thus
we can write the mass balance equation as (Seinfeld and Pandis, 2016)
∂c
= −u · ∇c + K · ∇2 c + R + E − S (2.4)
∂t
where the first term on the right hand side is the transport due to advection or the mean
wind, u = u, and the second term is the transport due to turbulence also called turbulent
diffusion where K is the eddy diffusivity coefficient.
2.2.1 Advection
The advection may be responsible for rapid horizontal transport since the horizontal winds
in the PBL often has a strength of 2-10ms−1 . The vertical winds are usually on the order
of millimeters to centimeters. They are much smaller compared to horizontal winds. This
is why we are most interested in the horizontal advection when we look at dispersion of
species. The distribution of the concentration of a specie due to horizontal advection can
be expressed as
∂c(x, y) ∂c ∂c
= −u · −v· (2.5)
∂t ∂x ∂y
where u = u and v = v is the horizontal mean wind components.
2.2.2 Turbulence
Turbulence is defined as highly irregular fluctuating motions both temporal and spatial,
and is often visualized as consisting of irregular swirls of many different sizes called eddies.
When we look at dispersion in the PBL turbulence must be considered. When the flow of
interest is turbulent, not laminar, it is common practice to represent the flows variables as
sums of their mean (x) and fluctuating (x0 ) part called Reynolds’ decomposition, where
2.2. DISPERSION IN PBL 12
the mean is smooth and slowly varying and the fluctuations is extremely space-time vary-
ing. Examples on Reynolds’ decomposition is
u = u + u0 , v = v + v0, θ = θ + θ0 , etc. (2.6)
Considering the advection equation mentioned previously Eq. (2.5). If we solve the
equations with u = u and v = v, and compare it to observations, we would discover that
the specie is spreading out more than expected. This additional spreading is a consequence
of the fluctuating part of the wind, u0 and v 0 . This imply that we need to consider
the precise velocity field when looking at dispersion in a turbulent environments. The
horizontal turbulent transport is usually negligible compared to the horizontal advection
transport. The vertical turbulent diffusion of a specie can be expressed as

∂c ∂ ∂c
= Kz (2.7)
∂t ∂z ∂z
where Kz is the eddy diffusivity coefficient. Due to the closure problem as mentioned in
the introduction, there are uncertainties related to calculations of turbulence.
Turbulence is dissipative and the turbulent kinetic energy (TKE) is not conserved. Un-
less energy is continuously supplied trough either mechanical or thermal generation, the
turbulence will die out by loss of energy to inertial generation of smaller eddies on the
edge of the larger eddies. This loss or cascade of energy continues until reaching total
dissipation by molecular viscosity, which is illustrated by the energy spectrum of TKE,
Fig. 2.5. The energy spectrum show how each eddy scale is linked to the total TKE. The
larger the value of TKE, the greater intensity of the microscale turbulence.
Figure 2.5: The energy spectrum of the turbulent kinetic energy (TKE), indicating the
link between the total TKE and the associated eddy scale. Adopted from Wallace and
Hobbs (2006).
2.3 Atmospheric Stability

2.3.1 Static Stability
The static stability of the atmosphere refers to how able the atmosphere is to resist or
enhance a vertical displacement of a parcel of air. Air parcels in the atmosphere are
continually displaced upward and downward owing to turbulence and overall atmospheric
motion. If the parcel is displaced for example upward, it can continue to rise, remain
at its new height, or return back to its initial position. If the parcel continues to rise, it
means that the density of the parcel is lower than the surrounding air and the atmosphere
is said to be unstable. If the parcel remains at its new position the density of the parcel
and surrounding air is equal and the atmosphere is neutral. And if the parcel returns
to its original position it means that the parcel density is higher than its surroundings
and the atmosphere is called stable. A general static stability parameter is the vertical
gradient of virtual potential temperature, ∂θ v
∂z
, and is summarized as

>0, Stable atmosphere
∂θv 
=0, Neutral atmosphere (2.8)
∂z 
<0, Unstable atmosphere

The virtual potential temperature is the temperature dry air must have to equal the den-
sity of moist air at the same pressure and is defined as
θv = θ · (1 + 0.61 · r) (2.9)
where r is the mixing ratio. The potential temperature, θ, is defined as

(R/cp )
P0
θ=T (2.10)
P
where T is the temperature at a pressure level P , P0 is a reference pressure (often 1013

hPa) and R/cp = 0.286. The difference between θv and θ is small, but not negligible.
During winter the air is usually sufficiently dry, so we can disregard this difference. Our
study takes place in the winter, so we will from now on assume a dry atmosphere and use θ.
Sometimes, if the timing is right, we can get inversion events. During night, especially if
there are no clouds aloft and the winds are light, the radiative cooling of the ground can
lead to surface air which is colder than the air above it. This leads to a situation where
the atmospheric temperatures increases with altitude and we have high static stability.
This is called an inversion. Turbulence is reduced by high stability, in addition to low ad-
vection, we can see from the mass balance equation (Eq.2.4) that the transport of species
will be small. Pollutants emitted into an inversion layer get trapped and can reach high
concentrations, since there is no mixing and transport of the air. We often experience
inversions at higher latitudes in the wintertime, usually when a high pressure system with
cloudless and dry air dominates.
2.3. ATMOSPHERIC STABILITY 14
2.3.2 Dynamic Stability

Even if the PBL is statically stable, the wind shear can generate turbulence which can
lead to dynamically unstable flows. If a critical value of shear is reached across a density
interface in the air, the flow becomes dynamically unstable and waves begin to form on the
interface. The waves grow and eventually breaks. This is called Kelvin-Helmholtz waves
(KHW). The KHWs will mix air of different density resulting in small local generations
of static instability, creating turbulence, which will spread throughout the layer causing a
mixing of the air layers of different density. The mixing will ultimately reduce the shear
due to momentum exchange and reduce the dynamic instability and the turbulence.
2.3.3 The Richardson Number

The stability of the PBL can, as discussed previously, be a good indicator of how turbu-
lent it is. The dimensionless Richardson number is a parameter which can be used as a
measure on the presence of turbulence, and is defined as
g ∂θ
θ ∂z
Ri = (2.11)
∂u 2 ∂v 2

∂z
+ ∂z
where g is the acceleration due to gravity, θ is the mean potential temperature and u, v
are the components of the mean wind vector. The Richardson number relates of buoyancy
∂θ
(numerator) to shear forces (denominator) in a flow. When Ri < 0, ∂z is < 0, and the at-
mosphere is buoyantly unstable. If Ri is small and negative, wind shear is large compared
to buoyancy, and mechanical generated turbulence dominates. If Ri is large and negative,
∂θ
thermal generated turbulence dominates. When Ri > 0, ∂z is > 0, and the atmosphere is
buoyantly stable, and thermal turbulence does not occur. If Ri is small and positive, wind
shear is large compared with buoyant stability, and mechanical turbulence occurs. If Ri is
large and positive, wind shear is low compared to buoyant stability, and neither mechani-
cal turbulence nor thermal turbulence occur, instead the flow is said to be laminar. Table
2.1 summarizes the flow regimes mentioned. The so called "Critical Richardson number",
Ric = 0.25, is when a laminar flow becomes turbulent. An already turbulent flow will con-
Richardson Type of flow Level of turbulence Level of turbulence

number, Ri due to buoyancy due to shear
Large, negative Turbulent Large Small

Small, negative Turbulent Small Large
Small, positive Turbulent None (weakly stable) Large
Large, positive Laminar None (strongly stable) Small
Table 2.1: Characteristics of vertical flow of air for different values of the Richardson
number, adopted from Jacobsen (2005).
tinue to be turbulent as long as Ri < 1.0. The flow is dynamically unstable if Ri < Ric .
The Richardson number is an often used measure of turbulence in the surface layer, where
vertical gradients of θ and u, v are most significant. The Richardson number is also an ap-
propriate stability parameter above the surface layer. When assuming the approximations
∂θ ∆θ ∂u ∆u ∂v ∆v
≈ , ≈ , ≈ (2.12)
∂z ∆z ∂z ∆z ∂z ∆z
where ∆ represents a vertical distance difference, we can define the bulk Richardson num-
ber, RiB
g∆θ∆z
RiB = (2.13)
θ ((∆u)2 + (∆v)2 )
RiB is a form of the Richardson number which is most commonly used in meteorology,
since observations and numerical models often can give values at discrete points in space.
When using the RiB , the critical value for when a laminar flow becomes turbulent, do not
apply. Ric is only appropriate for local gradients and not finite differences across layers.
So even though RiB often is easier to calculate, we bring higher uncertainties when trying
to predict if a flow is turbulent or not. However, the thinner the layer we look at, the
closer the critical value will be to 0.25 when using RiB .
2.3.4 The Obukhov Length

Another much used stability parameter in the surface layer is the Obukhov length. Since
the fluxes in the surface layer can be assumed to be constant due to the low variation in
magnitude as mentioned earlier, we can use the surface values of heat and momentum flux
to define turbulence. This is why the Obukhov length is not relevant above the surface
layer. We can define the Obukhov length as,
(τ0 /ρ)3/2 u2∗

L=− = (2.14)
k(g/T0 )(H0 /ρcp ) k(g/T0 )θ∗
where τ0 is the surface shear stress, ρ is the mass density of air, k is the von Karman
constant, H0 is the sensible heat flux at the surface and cp is the specific heat at con-
stant pressure. And so τ0 /ρ is the kinematic surface stress, H0 /ρcp is the kinematic heat
flux, and g/T0 is by P.S Arya (1999) called the buoyancy variable. u∗ = (τ0 /ρ)1/2 rep-
resent the friction velocity (i.e. a measure of the wind shear close to the ground and
so the mechanical turbulence) and θ∗ = −H0 /ρcp u∗ the friction temperature (i.e. the
characteristic temperature scale). The Obukhov length is interpreted by Stull (1988) as
"proportional to the height above the surface at which buoyancy factors first dominate
over mechanical (shear) production of turbulence". The Obukhov length is a part of the
widely accepted Monin-Obukhov similarity theory of turbulence. A similarity theory uses
dimensional analysis and observational data to derive relationships that can be used to
define turbulence, since our knowledge of the governing physics is not complete.
2.3. ATMOSPHERIC STABILITY 16
Bulk Richardson numb. limits Stability classes
RiB < −5.34 Very unstable, A

−5.34 ≤ RiB < −2.26 Unstable, B
−2.26 ≤ RiB < −0.569 Weakly unstable, C
−0.569 ≤ RiB < 0.083 Neutral, D
0.083 ≤ RiB < 0.196 Weakly stable, E
0.196 ≤ RiB < 0.49 Stable, F
RiB ≥ 0.49 Very Stable, G
Table 2.2: The Pasquills’s stability classes and the associated limits for the bulk Richard-
son number, RiB , adapted from Krogsæter and Reuder (2013).
2.3.5 Pasquill’s Stability Classes

The stability classification proposed by Pasquill (1961) and later modified by Turner
(1970), is one of the simplest and much used methods when classifying stability calcu-
lations. The original advantage with this classification system is that it only requires
standard observations from meteorological stations, such as near surface wind speed, so-
lar radiation and cloudiness. With surface information, the atmospheric stability can be
divided into seven classes from ’A - very stable’ to ’G - very unstable’. The Pasquill
classification can also be related to the bulk Richardson number, RiB , and the Obukhov
length, L (Mohan and Siddiqui, 1998). The limits of RiB for the different Pasquill stabil-
ity classes are calculated from the nomogram of Golder (1972). The relations can be seen
in Table 2.2. The Pasquill stability categories are commonly used in Gaussian dispersion
models (Seinfeld and Pandis, 2016). It provides a basis for correlation of readily available
meteorological data with the more difficult dispersion parameters σy and σz . These are
the standard deviations of the wind velocity fluctuations in the y and z direction. The
most widely used correlation is called the Pasquill-Gifford curves (Seinfeld and Pandis,
2016).
2.3.6 Effect of Stability on Wind Shear

The stability of the PBL influence the vertical structure and therefore also the variation
of wind with height and the turbulent eddies. Fig. 2.6 illustrate the different effects of
different stability environment. Under neutral conditions the form of the wind shear has
been found to be accurately described by a logarithmic decay curve, expressed as
u z
∗
uz = ln (2.15)
k z0
where uz is the mean wind speed at height z, z0 is the roughness length, u∗ is the frictional
velocity and k is the von Karman constant. The buoyancy is unimportant in a neutral
Figure 2.6: Schematic of the wind speed profile and the eddy structure close to the
ground due to the effect of a) a neutral PBL, b) an unstable PBL and c) a stable PBL.
In d) the profiles of a),b) and c) are re-plotted with a neutral logarithmic height scale,
adopted from Oke (1987).
PBL, while mechanical generated eddies are dominant and so the simplest interpretation
of these eddies are being circular and increasing in diameter with height. Mechanical forces
are still dominant at low heights under unstable conditions, but the thermal effect becomes
increasingly important with increasing height. This results in a vertical "stretching" of
the eddies and a reduction of the wind gradient. A stable PBL on the other hand dampens
the vertical movement, compresses the eddies and strengthens the wind gradient, thus a
steeper wind profile. The different effects on the wind shear is illustrated in Fig. 2.3.6.
2.4 Special Effects

Turbulence, stability and the mixing height have all an important role in the dispersion
of species, as we have seen. Unfortunately, they are more or less complex to calculate
and measure. In addition, the surface is often assumed to be flat for its simplicity, which
obviously rarely is the case. When we include surface details such as large cities, fjords or
mountains to the estimations, we need to consider the special boundary layer phenomena
that arises and the effect that they have.
2.4. SPECIAL EFFECTS 18
Figure 2.7: Schematic representation of the urban atmosphere illustrating a two-layer

classification of urban modification: the urban boundary layer (UBL) and the urban
canopy layer (UCL), adopted from Oke (1987).
2.4.1 Urban Effect

A city affects the atmosphere a lot. As air flows from the countryside and rural area
toward the city, different boundary conditions need to be accounted for. We can even
define an urban boundary layer (UBL) and the main features can be seen in Fig. 2.7.
The UBL develops downwind from the leading-edge of the city and is usually defined as
a phenomenon of mesoscale (Fig. 2.1), but it depends on the size of the city. The char-
acteristics of this layer are defined by the nature of the general urban surface. The urban
canopy layer (UCL) is produced by microscale motion and processes occurring underneath
the roof-level of the buildings.
Tall and large buildings exert a stronger drag on the wind, since the surface roughness is
larger in the UBL compared to the PBL over rural surfaces, as illustrated in Fig. 2.4. On
the other hand within the UCL the buildings can restrict the air motion and reduce the
average wind speed. The wind speeds in the UCL are usually lower compared to rural
winds at the same height. But there are exceptions. An example is if the wind flow is
directed through streets oriented in the same direction as the wind. In these situations,
the wind can result into a local ’jet’ because of a channeling effect by the buildings. When
large roughness elements like buildings are placed close to each other, the height of the
average roof-top can act like a displaced surface. The wind profile will now increase loga-
rithmically with height above the displaced surface as shown in Fig. 2.8. We can define a
displacement distance, d, and a roughness length, z0 . The mean wind speed in statically
neutral conditions which we defined in Sec. 2.3.6 can now be expressed as:
u (z − d)
∗
uz = ln (2.16)
k z0
where uz is the mean wind speed at height z, u∗ is the frictional velocity, k is the von
Karman constant, and we set uz = 0 at z = d + z0 .
Figure 2.8: Schematic of the wind flow above the displaced surface created by close
standing buildings, where the wind speed, M, is a function of the height, z. d is the
displacement height above the true surface and z0 is the roughness length, inspired by
Oke (1987, Fig. 9.7).
Cities are also large sources of anthropogenic heat and pollution. In contrast to other
natural surfaces, the surface energy balance in a city area is not only dependent on the
solar cycle. It is directly dependent on human activity and the surface heat equation
needs an additional term that represent the source of heat due to human activity, the
anthropogenic heat flux, Qf . The urban surface energy balance of a building-air volume
is
Rn + Qf = Qh + Qe + ∆Qs + ∆Qa (2.17)
where ∆Qa is advection of horizontal sensible and latent heat fluxes through the sides
of the building-air volume. Qg from the original balance equation, Eq. 2.2 is negligible
since the urban surface energy balance applies to volumes of sufficient depths and so the
conduction of heat is very small. Urban areas are usually covered by asphalt and concrete,
which are dry, water-proof surfaces with albedo and heat capacities that convert and
store incoming solar radiation as sensible heat better than the surrounding rural areas.
In addition the sky view factor in an urban area is often blocked by buildings which
reduce the loss of energy by longwave radiation. And so, the air temperature in cities is
generally warmer than that of their rural surroundings, which is often referred to as the
urban heat island. The greatest temperature difference between urban and rural areas
are usually observed during the night under clear and calm weather in the winter season.
The additional anthropogenic heat can enhance the thermal generated turbulence over
the city and so the mixing of the air.
2.4.2 Ocean Effect

Oceans have larger heat capacities than land. Due to the great heat it can store, the
climate by oceans generally vary less during the different seasons compared to the inland.
The main reasons for this difference in addition to great heat capacity is that water bodies
allows transmission of shortwave radiation to great depths. They are able of heat transfer
2.4. SPECIAL EFFECTS 20
by convection and mixing and they convert much of the excess energy into latent heat
instead of sensible heat. The water surface energy balance, where there is assumed no
vertical heat transfer (Qg ), can be expressed as
Rn = Qh + Qe + ∆Qs + ∆Qa (2.18)
where ∆Qa represents the net horizontal heat transfer due to water currents. The water
and land temperature differences can be clearly seen during wintertime in for example
Oslo. The temperatures can be as low as −15◦ C, but the Oslo fjord is still not frozen.
Another effect of the sea that can be experienced at the coast in the winter is when the
winds are coming from the sea. The wind will bring warmer air from the ocean to the
shore. This warm advection can often lead to warmer air aloft, while we have colder air
by the ground, and so an inversion situation may occur.
2.4.3 Topographic Effect

The variation in topography in an area have several consequences to the airflow. Val-
leys can produce local wind systems as a result of thermal differences. In cloudless and
fair-weather conditions, differential warming or cooling of the landscape can result in hor-
izontal temperature and pressure gradients. These gradients generates local winds often
referred to as anabatic and katabatic winds. Anabatic winds are a result of heating of
the air above the slope and floor of the valley which give rise to unstable upslope flows.
Katabatic winds are a result of radiative cooling of the lower air layers in the valley which
slide downslope under the influence of gravity. A more obvious effect of topography is the
modification of the wind flow, where it forces the flow towards other directions.
Chapter 3
The Observations
In this chapter the observations will be presented. Bjørvika, the instrumentation of the
Munch mast, and the calibration of the observations are described, before a comparison
of the measured variables to those made by the Norwegian Meteorological Institute at
Blindern and Valle Hovin in the same period are given.
Weather stations usually only measure temperature at 2m and wind at 10m. This study
wanted to make observations at additional heights, since wind speed, wind direction and
temperature often vary considerably in short vertical distances during for example an
inversion. Observational data like this have not been recorded in Oslo, to our knowl-
edge. The first ideas for doing the measurements were either to use a drone or a tethered
weather balloon. When exploring these options, it was discovered that there are very
strict rules for disturbing the airspace above Oslo. As a contribution to the scientific
research, Veidekke AS gave permission to install wind and temperature sensors in their
tower crane currently involved with the building of the new Munch museum in Bjørvika,
Oslo.
3.1 Description of Bjørvika

Bjørvika is an area east of Oslo city center and is situated in the inner Oslo fjord. Since
the start of the construction of the Opera House in 2003, Bjørvika has been a district of
growing urbanization. This is one of the reasons we found Bjørvika especially interesting
for this study. As shown in the overview of Bjørvika in Fig. 3.1, 1350 houses and 3
kindergartens were already built by the end of 2017 (marked in red), 850 new houses and
one new kindergarten are under construction (in blue), and an elementary school and a
high school in this area by 2022. A dispersion study has been completed, marked by the
yellow line to the right in Fig. 3.1 by Elin Aas, a fellow master student at the University
of Oslo. Aas has used passive samplers to measure concentrations of N O2 at seven points
along the marked road, and uses both the observed and modeled meteorological data from
this study in the dispersion simulations.
The climate in Oslo and Bjørvika is affected to a great extent by the Oslo fjord. The
fjord is an important source of heat and can have a large impact on the energy balance in
this area. As discussed in Sec. 2.4.2, the sea will contribute to a generally mild climate
22
23 CHAPTER 3. THE OBSERVATIONS
Figure 3.1: Illustration of Bjørvika, with the already finished construction work marked
in red and the ongoing projects marked in blue. The yellow circle is where the Munch
mast is situated and the yellow line to the east of the Munch mast is where a dispersion
study is taking place, which will use the meteorological data from this study, adapted
from https://www.bjorvikautvikling.no/portfolio-item/bolig-og-naering/.
during the winter. However, the fjord effect will also vary with how much of the sea that is
frozen and the wind direction. As Bjørvika is a part of a larger city and has become more
urbanized, the urban heat island effect is also expected to have an impact. Oslo has an
increasing terrain height in the northern direction, called Nordmarka with highest points
between 600-700m above mean sea level. Nordmarka includes Sørkedalen and Maridalen
(Thorsnæs, 2018). The upland to the east is called Ekeberg, which is about 140m above
mean sea level (Tvedt, 2016). Due to the topography, wind from the north has a tendency
to veer to north-east due to Groruddalen, while the southerly winds is affected by the
Oslo fjord, the Bunne fjord and Ekeberg.
3.1.1 The Munch mast

The Munch mast was located at the end of Paulsenkaia southeast from the Oslo Opera
House where the Aker River flows out to the Oslo fjord. The mast is marked as a yellow
circle in Fig. 3.1. The area where the mast was placed, was a construction area. The pier
with the mast was counted as the reference surface, and the heights of the sensors were
measured as the distance above the pier. The mast was situated about one meter from
the south west corner of the Munch museum building. As the measurements was taken so
close to a building, disturbances of the wind speed and wind direction were expected. The
Munch museum consist of a 13 floor high tower encircled by a three floor high podium
1
. A picture of the building in the vertical can be seen to the right in Fig. 3.2. The
framework and exterior of the building was finished before the measuring period started.
1
https://www.kulturbyggene.no/om-prosjektet/category539.html
3.1. DESCRIPTION OF BJØRVIKA 24
Figure 3.2: To the left: map showing an overview of the area where the Munch mast is
located, the mast marked as a red and white circle. The red rectangle indicate the area
called Sukkerbiten. To the right: The Munch mast from a ground point of view, marked
with the same red and white circle.
It was assumed that the disturbance due to the building was the same the whole time.
There were also placed barracks for the workers at Sukkerbiten. This is marked with
a red square to the left in Fig. 3.2. These were about 10m high and was situated at
Sukkerbiten the whole measuring period. Thus it was assumed "constant" effect from
these as well. An overview of the direction zones which was expect to be disturbed by the
surrounding buildings, at respectively 10m and 30m, can be seen in Fig. 3.3. The 10m
wind sensor would probably be disturbed largely by both the museum (56◦ −147◦ ) and the
barracks (214◦ − 282◦ ). It was expected that the sensor at 30m was mainly disturbed by
the museum. In addition there would also be some disturbances of other wind directions
a both heights as well, although not so obvious. A CDF (Computational Fluid Dynamics)
wind analysis could quantify these effects, but this was out of the scope of our study.
(a) (b)
Figure 3.3: The approximate disturbed wind direction zones due to surrounding build-
ings at a) 10m and b) 30m.
3.2 Instrumentation of the Munch mast

John Hulth, the head engineer at the Section of Physical geography and Hydrology at the
Department of Geoscience, was leading the instrumentation of the Munch mast and was
assisted by the author of this thesis. He planned the technical and practical aspects of
selecting and placing the sensors. Also the maintenance and removal of the equipment.
Before the instrumentation could start, it was necessary to complete the safety training.
This is mandatory for all who are involved at Veidekke AS’s construction sites. The
Munch mast was installed with six temperature sensors of the type DS18B20 digital
thermometer, placed at 2, 10, 30, 50, 65 and 75 meters. They were mounted in radiation
shields or “huts” to avoid direct sunlight and placed on the North side of the mast. The
thermometers measured the temperature twice every minute. The horizontal wind speed
and the wind direction were measured by DS-2 Sonic anemometers, which were placed
at 10, 30 and 75 meters. The anemometers had no moving parts and measured the wind
components by transmitting acoustic signals. The wind sensors were placed on metal rod
reaching 50 cm out from the mast on the west side. The logging frequency of the wind
sensors were twice every minute as well. To be able to supply the sensors with electricity,
it was decided to bring a 100m long wire up in the mast instead of using batteries, which
would have needed more maintenance work. We got up into the mast by climbing ladders
fastened inside the mast, and there was a deck every fifth meter which made it easer to
attach the sensors. The instrumentation of the Munch mast can be seen in Fig.3.4.
(a) (b)
Figure 3.4: Illustrations showing a) the Munch mast and the heights where the tem-
perature sensor (T) and wind sensors (W) was installed, and b) how the sensors were
fastened, adapted from original images made by John Hulth.
3.3. CALIBRATION OF MEASUREMENTS 26
3.3 Calibration of Measurements

3.3.1 The Temperature Observations
The temperature sensors were tested for their sensitivity in a climate chamber before
installed on the Munch mast. A climate chamber is a test chamber where you can main-
tain a certain climate, for example temperature as in this thesis, to do a calibration of
the instruments. The observations were taken during the winter, so the temperature in-
struments were tested from −21◦ C to +10◦ C. Since Oslo is situated by the sea, it is
unlikely that it would get colder than −20◦ C. The results from the climate chamber can
be seen in the Appendix (Fig. A.1). The documentation from the manufacturer of the
instrument claimed an accuracy of ±0.5◦ C which was confirmed during the chamber test.
By linear interpolation the error functions were calculated for their respective sensors
and the measurements were corrected. The correction functions can be seen in Table
3.1. The manufacturer claimed a drift of less than 0.2◦ C. There were unfortunately no
chamber tests of the temperature sensor at 2m. Due to measuring inconsistencies by the
first installed 2m sensor, it was replaced before the period started. The new sensor was
not tested in the climate chamber due to the time schedule. It was decided to keep the
observations at 2m since the measuring accuracy of the other five sensors showed good
agreement with the documentation from the manufacturer.
Sensor Correction func.
10m y = −6 · 10−6 T + 0.0623

30m y = −0.0072x + 0.1849
50m y = 0.0074x − 0.2795
65m y = 0.0037x − 0.0429
75m y = 0.0005x + 0.0609
Table 3.1: The correction functions for each temperature sensor, calculated by linear
interpolation of the result from the climate chamber test (Fig. A.1).
The shape of the distribution of the corrected temperatures for the whole measuring
period were similar for all heights with equally spaced bars of 0.5◦ C, as seen in Fig. 3.5.
The peak of the distributions were between 1.5◦ C and 2.5◦ C. The correlation between
the sensors and the sensor at 10m can be seen in Fig. 3.6. The correlation were close to
the on-to-one line for all sensors, with slightly colder measurements by the sensors above
10m and slightly warmer measurements by the sensor at 2m. This indicates that there
were more static unstable than stable situations in which the temperature decrease with
increasing height. These results also confirmed that it was reasonable to include the 2m
sensor, since it did not deviate much from the other sensors.
Figure 3.5: The distribution of the corrected temperature observations for the whole
period, January 9 - March 1, and with equally spaced bars of 0.5◦ C. There is two
observations every minute.
Figure 3.6: Corrected observed temperature (◦ C) at 10m along the x-axis versus the
corrected observed temperatures (◦ C) at 2m, 30m, 50m, 65m and 75m along the y-axis
for the whole measuring period in the Munch mast, January 9 - March 1. There is two
observations for every minute.
3.3. CALIBRATION OF MEASUREMENTS 28
3.3.2 The Wind Observations

The wind sensors were not tested for their measuring accuracy, since there were not
access to such instruments. The manufacturer claimed an accuracy of ±0.30 ms−1 , or
±3% (whichever is greater) for the wind speed and an accuracy of ±3 degrees for the
wind direction.
Wind Speed
The distribution of the observed wind speed was a bit varying. The sensors at 10m and
30m in the Munch mast had mainly measured speeds less than 10ms−1 , while the sensor
at 75m had measured wind speeds up to 50ms−1 due to erroneous measurements. The
distribution can be seen in Fig. 3.7. The erroneous measurements of the sensor at 75m
were obvious in the scatter plot, Fig. 3.8b. There was unfortunately not possible to replace
the sensor at 75m. The comparison of the sensor at 10m and 30m, Fig. 3.8a, shows a
more expected result, with slightly larger wind speeds at 30m. There was measured some
speeds exceeding 20 ms−1 at 30m, but these where limited to one day, so it was reasonable
to see this as erroneous measurements. The upper wind speed limit was set to 15 ms−1
and measurements exceeding this limit were not included. The restrain was based on
the climatology of the wind speed measured at Blindern, Fig. A.2, which seldom exceed
10ms−1 . With these restraints the clear deviations at some points between the sensor at
10m and 30m (Fig.3.8a) were avoided.
Figure 3.7: The distribution of the wind speed observations in the Munch mast at (from
left to right) 10m, 30m and 75m for the whole period, January 9 - March 1, using 10
minute averages and with equally spaced bars of 0.5ms−1
(a) (b)
Figure 3.8: Observed wind speed (ms−1 ) at 10m versus a) the observed wind speed
(ms−1 ) at 30m and b) the observed wind speed (ms−1 ) at 75m along the y-axis for the
whole measuring period, January 9 - March 1, using 10 minute averages.
Wind Direction
When analyzing the wind direction, it was applied a calm wind limit at 0.5 ms−1 . If the
wind speed was less than 0.5ms−1 , it was defined as calm wind and the wind direction was
set to "no data". The reason for this limit is that when the wind speed is low the wind
direction is alternating, almost turbulent, and it is not possible to measure a direction.
The wind directions observed at both 10m and 30m were mainly wind from the northeast
and south, as seen from the wind roses in Fig. 3.9. There was no wind recorded from east
for both sensors. An obvious explanation for this, was the location of the museum being
on the east side of the mast as shown in the picture in Fig. 3.2. The Munch Museum was
working as a wind shield. This can be seen from the wind roses in Fig. 3.9. There was
a distinct increase in westerly wind from 10m to 30m. This was most likely due to the
disturbances to the west at 10m, as seen in Fig. 3.3, which the sensor at 30m did not have.
There would also be an effect on the wind speed and direction from the mast itself,
as mentioned in Sec. 3.1.1. In a study done by Stickland et al. (2013) they reported that
the wind anemometers need to be placed on a rod 3.5m away from the meteorological
mast to achieve a 99% free stream. The study was based on results from wind tunnel
tests and CFD simulations of a triangular meteorological mast. The mast used in this
study was squared. Thus it can not be expected to have the exact same terms for free
stream, but the required boom length would probably need to be around 3.5m. However,
it was restrictions to use the mast only so that the tower crane could be operative for
construction work during the whole measuring period. It was therefore not possible to
attach the anemometers further out than 50cm (Fig. 3.4b).
3.4. MET’S WEATHER STATIONS 30
(a) (b)
Figure 3.9: The distribution of the observed wind direction and wind intensity at a)
10m and b) 30m in the Munch mast for the whole period, January 9 - March 1, using
10 minute averages. The calm wind limit is 0.5ms−1
3.4 MET’s Weather Stations

The Norwegian Meteorological Institute (MET) has several operational weather stations
in Oslo. This thesis used MET’s observations from selected stations and compared them
with the Munch mast observations, as a validation of the measurements done by the
instruments. Observations from the station at Valle Hovin (from now only termed Hovin)
and Blindern, were used. These were the closest stations measuring temperature, wind
speed and wind direction.
3.4.1 Description of Hovin and Blindern

Hovin is an area about 3km northeast from the Munch mast. It lies about 100m above
sea level with no steep topography in immediate vicinity but is generally affected by the
same upland as the rest of Oslo (Sec. 3.1). On a local scale, the weather mast at Hovin
has flat terrain with football courts to the south and east. However, there are groups of
trees within 20m to the north and east. There are industrial buildings and houses about
100m in every direction. As Hovin lies a few kilometers away from the fjord and the effect
of the fjord is highly relevant further into the inland, we would expect temperatures at
Hovin to be a bit colder compared to the Munch mast in winter.
Blindern is the area where the campus of the University of Oslo is situated and one
of the offices to MET. Blindern lies about 4.5km northwest from the Munch mast, and
about 100m above sea level. Locally, there are some trees within 20m to the south and
west of the mast at Blindern. While the MET building is situated within 30m to the
east. Within 20m to the north of the mast is the metro. The foot of the topography
connected to Nordmarka, starts about 2 kilometer northeast of Blindern. Blindern is as
Hovin situated a few kilometers away from the fjord and we would also expect lower here
temperatures than at the Munch mast.
3.4.2 Instrumentation of the Hovin and Blindern masts

The Hovin mast measured temperatures in a MI-hut at 2m and 10m and wind speed and
direction at 25m. The Blindern station also measured temperatures in a MI-hut at 2m
and 10m, and wind speed and direction at 26.5m. The locations of the MET stations and
the Munch mast can be seen in Fig. 3.10.
The data from the MET’s masts were downloaded from eKlima. eKlima is a web portal by
the Norwegian Meteorological Institute which gives free access to their climate database.
The database contain data from all weather stations of the institute, present and past, as
well as data from other institutions.
Figure 3.10: Map (from Google Maps) with the locations of the MET weather stations
at Blindern, Valle Hovin and the Munch mast in Bjørvika.
3.5. COMPARISON OF MEASUREMENTS 32
3.5 Comparison of Measurements

3.5.1 The Temperature Observations
The distribution of the observed temperatures at 2m and 10m at the Munch mast were
similar compared to the same heights at Blindern and Hovin. The peak for the 2m distri-
bution was between −1◦ C and −1.5◦ C at the Munch mast, while it was between −2◦ C and
−2.5◦ C at Blinder and Hovin. The peak for the 10m distribution was between −1.5◦ C and
−2◦ C at the Munch mast, while between −2◦ C and −2.5◦ C at the Blindern and Hovin
masts, Fig. 3.11. The coldest temperatures in the Munch mast were about −12.5◦ C at
2m, and −13◦ C at 10m, while both measuring heights at Blindern and Hovin observed
temperatures down to −16◦ C. Blindern observed more temperatures below −15◦ C than
the Munch mast and the Hovin mast. The Munch mast observed the warmest tempera-
tures up to 7.5◦ C at 2m and 7◦ C at 10m, while the warmest observations at Hovin were
up to 6.5◦ C and Blindern up to 6◦ C. The average temperature for the whole period in the
Munch mast was −2.0◦ C at 2m and −2.3◦ C at 10m. The average temperature at Hovin
was −3.2◦ C at 2m and −3.3◦ C at 10m, which was similar to the averages at Blindern
being −3.2◦ C at both 2m and 10m. So the Munch mast was warmer in general, compared
to the MET masts. This was as expected when looking at the locations of the masts in
Fig. 3.10. The MET masts are further away from the fjord, so the warmth from the
ocean has a greater impact on the Munch mast than the MET masts, as discussed in Sec.
3.4.1. The impact from the fjord and the sea surface temperature would especially be
dominating at the Munch mast when the wind direction was southerly. If there is weak
wind from the city towards the Munch mast, the urban heat effect will be dominating.
(a) (b)
(c) (d)
(e) (f )
Figure 3.11: The distribution of the observed temperature at 2m at a) the Munch mast
c) Hovin and e) Blindern, and at 10m at b) the Munch mast, d) Hovin and f ) Blindern
for the whole measuring period, January 9 - March 1, using hourly averages and equally
spaced bars of 0.5◦ C.
3.5.2 The Wind Observations

Wind Speed
The distribution of the wind speed was a bit different at the Munch mast compared to
those for Blindern and Hovin. The main observed wind speeds were below 2ms−1 for all
locations. The peak of the distribution at the Munch mast was 0.5-0.75ms−1 at both 10m
and 30m, 1-1.25ms−1 at Hovin and 1.5-1.75ms−1 at Blindern, Fig. 3.12. The frequency
of the low wind speeds were much higher at the Munch mast compared to the masts
ar Blindern and Hovin. It was expected more strong wind speed when comparing the
10m at the Munch mast to the measuring heights at Hovin (25m) and Blindern (26.5m)
because of the frictional drag, as discussed in Sec. 2.1.2. But the observations at 30m in
the Munch mast were also much lower compared to Hovin and Blindern. On explanation
may be the differences in disturbed wind fields around the Munch mast due to the Munch
museum being very close, as discussed in Sec. 3.3.2. This was something to keep in mind
when comparing the observations to the model runs as well. The average wind speed for
the whole measuring period was in the Munch mast 1.3ms−1 at 10m and 1.6ms−1 at 30m.
At Hovin the average wind speed was 3.1ms−1 and at Blindern it was 2.7ms−1 .
Wind Direction
The wind direction distribution between the MET masts and the Munch mast were sim-
ilar. Dominant wind directions were winds from the northeast and the south as can be
seen in Fig. 3.13. The distribution at 30m in the Munch mast was most like the obser-
vations done by the MET masts. This was as expected since they were approximately of
the same height. There were few observations of wind from the east or the west at both
Blindern and Hovin, indicating that not much information was lost due to the sheltering
by the Munch museum and the barracks. One difference when comparing the intensities
of the wind directions, was that at the Munch mast the strongest winds came from the
south, while at Blindern and Hovin the intensity of the winds was generally stronger in
all directions.
(a)
(b)
Figure 3.12: The distribution of the observed wind speed at a) the Munch mast at 10m
to the left and 30m to the right, and at b) Valle Hovin (25m) to the left and Blindern
(26.5m) to the right for the whole measuring period, January 9 - March 1, using 10
minute averaged and equally spaced bars of 0.25ms−1 .
(a) (b)
(c) (d)
Figure 3.13: The distribution of the observed wind direction and wind intensity at a)
10m in the Munch mast, b) 30m in the Munch mast, c) 25m at Hovin and d) 26.5m
at Blindern for the whole period, January 9 - March 1, using 10 minute averages. The
calm wind limit is 0.5ms−1 .
Chapter 4
The Model
The model used in this study was the Weather Research and Forecasting model (WRF),
which is described in the first part of this chapter. A presentation of the setup and the
simulations from the study follows. The model description is mainly based on Wang et
al. (2017) and Skamarock et al. (2008).
4.1 The Weather Research and Forecast Model

WRF is a state-of-the-art mesoscale numerical weather prediction system. It is used for
both operational weather forecasting and atmospheric research with possible application
across a broad range of spatial scales. The development of WRF began in the late 1990’s
and was a grand collaboration between the National Center of Atmospheric Research
(NCAR), the National Oceanic and Atmospheric Administration (NOAA, represented by
the National Centers for Environmental Prediction (NCEP) and the Forecast Systems
Laboratory (FSL)), the Air Force Weather Agency (AFWA), the Naval Research Labora-
tory (NRL), the University of Oklahoma and the Federal Aviation Administration (FAA).
WRF is today maintained by NCAR’s Mesoscale and Microscale Meteorology Laboratory
(MMM).
4.1.1 WRF Modeling System

The WRF system consist of four major programs. The WRF Preprocessing System
(WPS), the WRF-Data assimilation (WRF-DA), the dynamical solvers and the Post-
processing and Visualization tools. It is also possible to couple WRF with chemistry,
the so called WRF-Chem. WRF-Chem makes it possible to simulate the chemistry of
the atmosphere. WRF-Chem was not used in this study. There are two choices for
the dynamical solver; the Nonhydrostatic mesoscale Model(NMM) and the Advanced
Research WRF (ARW, which is also non-hydrostatic). The ARW core was the default
solver version 3.8.1.1 was used in this thesis. The total flowchart of WRF with the various
components can be seen in Fig. 4.1, which illustrate a full overview of the potential with
WRF though not all components were used in this thesis. WRF is an Eulerian, fully
compressible and non-hydrostatic model, and the solver use the time-split Runge-Kutta
scheme for integration of the governing equations.
38
39 CHAPTER 4. THE MODEL
Figure 4.1: Schematic showing the flowchart of the WRF modeling system, adopted
from Wang et al. (2017).
The governing equations are cast in flux form and is formulated using the terrain-following
vertical coordinate η, defined by a normalized hydrostatic pressure as:
(ph − pht )
η= , where µ = phs − pht (4.1)
µ
ph is the hydrostatic component of the pressure, phs refer to values along the surface and
pht refer to the values at the top boundaries. η is the traditional sigma coordinate, σ,
used in many hydrostatic atmospheric models, and varies from 1 at the model surface
to 0 at the model top of the domain, shown in Fig. 4.2. The number of vertical levels
and the value for each eta-level can be specified by the user, and was done in this study.
The vertical and horizontal grid staggering is the Arakawa C-grid, seen in Fig. 4.3. That
is, normal velocities (u, v, w) are staggered one-half grid length from the thermodynamic
variables. This means that for example the vertical velocity, w, is calculated for the full
levels, the η-levels (levels of equal value) and the thermodynamic variables such as poten-
tial temperature, θ, are calculated between the η-levels for half levels. These half levels
are also called mass levels.
4.1. THE WEATHER RESEARCH AND FORECAST MODEL 40
Figure 4.2: Schematic of the terrain-following vertical η coordinate in WRF, adopted

from Skamarock et al. (2008).
Figure 4.3: Schematic of the Arakawa C staggering in a) the horizontal and b) the
vertical, adopted from Skamarock et al. (2008).
4.2 Model Setup

This thesis used WRF for simulation of the period January 9, 2018 to March 9, 2018. The
focus was on Oslo, capital of Norway, with Bjørvika in center, as this was the location of
our weather station which is discussed later. One of the main tasks for this study was to
manage to run WRF with high resolution in the inner domain and to be able to study
small scale processes of the model. WRF is, as mentioned, a mesoscale model. Too high
spatial resolution will contradict its mesoscale physics. It was desided that a resolution
of 300m × 300m was satisfactory for the inner domain. The domain setup was chosen
according to recommendations from Emilie Iversen at Kjeller Vindteknikk. The choices
of the physic parameterizations were chosen according to recommendations from Kjetil
Schanke Aas at the Department of Geosciences, University of Oslo, and the supervisors.
There was not anyone with experience in using the urban canopy scheme at either the
university or at Kjeller Vindteknikk, but it was included since the scheme takes into
account effects which are important in this study. An example of the namelist.input used
in this study can be seen in the Appendix B.
Figure 4.4: The four one-way nested domains that are used in the WRF simulations.
4.2. MODEL SETUP 42
Domain Horizontal res. Numb. of grid cells ∆t
d01 13.5×13.5 km 103×100 67 s

d02 2.7×2.7 km 106×106 13.4 s
d03 0.9×0.9 km 103×100 4.5 s
d04 0.3×0.3 km 103×100 1.5 s
Table 4.1: Table with an overview of the horizontal resolution and number of grid cells
in the four model domains.
4.2.1 Domain Setup

WRF supports horizontal nesting. This means that it is possible to introduce several
grids to the simulation to focus the desired resolution to the area of interest. There are
two nesting possibilities: two-way and one-way nesting which define how the domains
interact with each other. With one-way nesting the integrations of the coarse outer or
parent domain is used as boundary condition for the finer child domain. With two-way
nesting, there is a feedback of the solutions from the integrations in the child domain
back to the parent grid points lying inside the fine grid. The model in this thesis was
set up as a one-way nested model with four one-way nested domains. It was sufficient
with a one-way nested model since this study was looking at local scale motions in the
innermost domain. Having the additional information by a two-way nesting on the larger
processes from the other domains would not change the processes of interest notably. Fig.
4.4 shows the four domains. The largest domain covers most of Mid-Scandinavia and had
a horizontal grid resolution of 13,5 km. Then zooming in on east of Norway and Oslo
in three nesting levels, of 2.7km, 0.9km and 0.3km. The horizontal dimensions of each
domain, number of grid cells and time step are shown in Table 4.1.
There were 51 levels with the model top at 50 hPa in the vertical for all four domains.
WRF does not offer vertical nesting. The η-values for the lowest levels were specified for
this study, to be close to some of the measuring points in the Munch mast where there
were perform observations. There were six levels in the lowest 120m, where the η-value
was decreasing monotonically with height. An overview of the lowest six levels can be
seen in Table A.1 in the Appendix.
4.2.2 Initialization and Boundary Conditions

The initial conditions and boundary conditions of the parent domain was set by the oper-
ational data from the European Center for Medium-Range Weather Forecast (ECMWF).
The ECMWF data also initialized the other three domains, while the boundary condi-
tions were calculated from the parent of each domain. The horizontal resolution of the
ECMWF data was 0.0625◦ × 0.0625◦ and were updated with 6 hour intervals. The sea
surface temperature (SST) was also set by the ECMWF data. It was not updated, since
(a) (b)
Figure 4.5: Figure (a) shows the land use of the fourth domain. WRF have captured the
main feature of the fjord and city from the satellite image from google map in Figure
(b) (The land use plot with a complete bar of color descriptions can be seen in the
Appendix, Fig. A.3).
we ran over short periods. Thus the SST’s were not changed over the course of the runs.
It was important for this study that the model had a realistic definition of the land
use, since different types of land use comes with different physical properties that change
the radiation and energy balance. In this thesis, the fjord and the city in the inner domain
were the most important features to be resolved. Several land use databases was tested,
and the U.S. Geological Survey (USGS) had the best land resolution. The land use in the
inner domain compared to a satellite picture of Oslo can be seen in Fig. 4.5.
4.2.3 Choices of Physic schemes

The selection of physical schemes in WRF are many and fall into different categories:
microphysics, cumulus parameterization, surface layer, planetary boundary layer (PBL),
land-surface model and atmospheric radiation. Each category containing several choices.
WRF offers numerous ways of dealing with the physical aspects. This thesis chose par-
ticularly to focus on the choices of the planetary boundary layer scheme, the land-surface
scheme, the surface layer scheme and the urban surface scheme or urban canopy model
(an additional specification to the land-surface scheme). These were the definitions of
most interest for this study. Additionally, the physical parameterization used for the mi-
crophysics was the Eta-Ferrier scheme (Rogers et al., 2001). The shortwave and longwave
radiation were parameterized with the Rapid Radiative Transfer Model for GCM appli-
cations (RRTMG; Iacono et al, 2008).
4.2. MODEL SETUP 44
Planetary Boundary Layer Scheme

The PBL scheme computes the flux profiles within the PBL, and provides the atmospheric
tendencies of temperature, moisture, horizontal momentum and vertical diffusion (turbu-
lence) due to eddy transport. This study used the Mellor-Yamada-Janjic (MYJ) or Eta
scheme. It is a local TKE closure scheme based on Monin-Obukhov theory. MYJ uses the
Mellor-Yamada Level 2.5 turbulence closure model (Mellor and Yamada (1982)), which
solves a prognostic equation for the TKE. The PBL top is dependent on TKE, buoyancy
and shear of the flow, where the scheme applies an upper limit to the length-scale charac-
terizing the effect of turbulence. The implementation of MYJ is described in Janic (1990,
1996, 2002).
Land-Surface Scheme
The land-surface model (LSM) provides the information about the heat and moisture
fluxes over land points and sea-ice points. This study used the Noah LSM scheme de-
veloped by NCEP, NCAR and AFWA which calculates soil temperature and moisture in
four layers, and predicts fractional snow cover and soil ice. It provides the sensible and
latent heat fluxes to the PBL scheme. This scheme has almost identical code as used in
the NCEP North American Mesoscale Model.
Surface Layer Scheme

The PBL scheme is used together with a surface layer scheme. The surface layer scheme
will calculate the friction velocities and exchange coefficient that make it possible to cal-
culate the surface heat and moisture fluxes by the land-surface scheme and surface stress
by the PBL scheme. This study used the Eta surface layer scheme, or the so called MYJ
surface scheme, since it must be run with the MYJ PBL scheme. It is also based on
Monin-Obukhov similarity theory with Zilitinkevich thermal roughness length (Zilitinke-
vich, 1995).
Urban Surface Scheme

The Urban surface scheme is coupled to WRF through the Noah LSM. It takes into ac-
count the effect roofs, walls and streets has on the surface temperature and heat fluxes,
and the momentum exchange between the urban surface and the atmosphere. This study
used the single-layer urban canopy model (SLUCM) which considers the shadowing, re-
flections, and trapping of radiation in a street canyon. An exponential wind profile is
prescribed. The prognostic variables are the surface skin temperatures (TSKs) at the
roofs, walls and roads, and temperature profiles within roof, wall, and road layers (Chen
et al., 2011). The SLUCM is embedded within the first model layer, as shown in Fig.
4.6. It is developed by Kusaka et al. (2001) and Kusaka and Kimura (2004), and later
modified by Chen et al. (2006). See the articles for more details.
Figure 4.6: A schematic of the single-layer urban canopy model, adopted from Chen et
al. (2011).
4.3 Simulations and Processing

This thesis simulated the period January 9 - March 1, 2018. Due to time-limitations it
was decided to split the period into 7 parallel runs with four hours for spin up.
• Run 1: January 9 - January 18
• Run 2: January 19 - January 25
• Run 3: January 26 - February 1
• Run 4: February 2 - February 8
• Run 7: February 23 - March 1
All runs had the same setup described previously. The WRF output files was given every
hour. It was also performed a shorter simulation, with WRF output files every ten minutes
as a selected case.
• Selected case: January 23 - January 26 (ten minute outputs)

4.3. SIMULATIONS AND PROCESSING 46
The input data described in previous sections were prepared by the WRF Pre-processing
System (WPS). The three main programs in the WPS is geogrid, ungrib and metgrid.
Geogrid defines the coarse domain and the other nested domains. Ungrib extracts the
meteorological fields from GRIB data sets for the simulation period. Metgrid interpolates
horizontally the meteorological fields on to the domains. The data flow is illustrated
in Fig. 4.7. In addition, the calc_cmwf_p was used to create fields of pressure and
geopotential heights from the ECMWF model-level data sets to the same level as the other
atmospheric fields. This was required by the real program which initialize WRF. For the
post-processing the NCAR Command Language (NCL) was used. NCL has many built in
functions to easily extract WRF model output and do basic diagnostics calculations 1 like
rotate u and v components of the wind to earth coordinates and do vertical interpolations
which was needed in this study to be able to compare the model output to the different
measuring heights in the Munch mast. NetCDF Operators (NCO) was applied for file
manipulations. All analysis and visualizations of results ere done with Python version
3.6.1. The python modules Pandas and wind rose were also used.
Figure 4.7: The flow between the three main programs in the WRF Pre-processing
System; geogrid, ungrib and metgrid, adapted from Weng et al. (2017).
1
https://www.ncl.ucar.edu/Document/Functions/wrf.shtml
4.4 WRF Data Selection

When extracting the WRF output data, it
was decided to get the data from the grid
box closest to the location of the Munch
mast (59◦ 540 20.800 N, 10◦ 450 17.500 E) and the
data from an additional box in each di-
rection (i.e. north, east, south and west).
The reason for this was to see if any of the
other grid boxes showed a better fit with
the observational data, and to see the dif-
ference between sea grid boxes and land
grid boxes. To be able to get land grid
boxes, it was needed to use the second
grid boxes in northern and eastern direc-
tion. A schematic of the extracted grid Figure 4.8: Schematic of the selected grid boxes
boxes can be seen in Fig.4.8. Tempera- for extraction of data from the WRF output files.
ture, wind speed, wind direction, heat flux, Letters representing: M-the closest grid to the
PBL height and the friction velocity were loaction of the Munch mast, N - closest grid to
collected from the chosen grids. The wa- the North of M being land, E - closest grid to
the East of M being land, S - closest grid to the
ter grids (i.e. grid M, S, W from Fig.
South of M being sea and W - closest grid to the
4.8) had the best fit with the observations.
West of M being sea.
The differences in root-mean-square-error
(RMSE) and the bias among the best grids
was very small and so it was decided to use the grid box closest to the Munch mast when
presenting the results in Chapter 5. The same were the case for Hovin and Blindern,
where the simulated data was checked in the same way as for the Munch mast to get the
best dataset.
4.5 The SST-Problem

When starting the analysis of the WRF simulation results by comparing them to the
observations from the Munch mast, it was discovered that the temperatures from WRF
had a large negative bias. WRF was colder than the observations at all levels. The near
surface temperatures were closely connected to the SST and the TSK. An investigate was
done to see if the negative bias could have something to do with the definition of the fjord
and the SST from the ECMWF data, since the sea has a thermal effect as discussed in Sec.
2.4. It was discovered that the mask of the ECMWF data did not match the landmask
used in the inner domain (i.e. USGS (Sec. 4.2.2)). It had excluded the water. Because
of this, the SST value was not used, as there was no water detected. It was decided
after discussing the problem with Kjetil Schanke Aas and Anne Claire Fouilloux at the
Department of Geosciences, University of Oslo and the WRF user support, to modify the
initialization by changing the TSK. The landmask from the USGS resolved the Oslo fjord
well in the fourth domain, Fig. 4.5, so the TSK was modified for all USGS water bodies
which were not lakes. The ECMWF data was too coarse in the third domain and had
4.5. THE SST-PROBLEM 48
some very low SST’s in the fjord in the second domain. It was not observed that much ice
on the fjord during January and February. It was therefore decided to modify the TSK in
the third and second domain as well. The TSK in the three inner domains was set to 276K
(i.e. 3◦ C), as this was the SST from the few defined water grids in the fjord in the third
domain set by the ECMWF data. 276K was also close to the average observed SST in
the Oslo fjord in January, which was 276.6K. The average was based on SST observations
at 1m depth at the Norwegian Institute for Water Research (NIVA) research facility at
Solbergstrand, Drøbak, provided by Oddbjørn Pettersen. The modification of the TSK
gave model results more similar to the observations, as can be seen in Fig. 4.9 where the
temperature at 2m of the observed and simulated data, before and after the modification,
are compared. As be seen, the temperature in WRF has generally become more similar to
the observations for the whole period. Especially from January 9 to January 23 and from
February 19 to March 1, which showed the largest deviations in the original run. The
comparison of the temperature at the additional five heights before and after modification
of the TSK can be seen in the Appendix (Fig. A.4 - Fig. A.8).
(a)
(b)
Figure 4.9: Comparison of the time series of the observed temperature (in Celsius) at
2m and the model temperature (in Celsius) at 2m from a) the original run and b) the
modified (TSK=276K) run, during the whole period, January 9 - March 1. The closest
10 minute average to every hour is used for the observations, and hourly values are
used for WRF.
Chapter 5
Results and Discussion
In this chapter the results from the Munch mast observations are discussed and compared.
The aim is to compare and discuss the observed meteorological parameters at the Munch
mast and the derived parameters, which are important for local dispersion calculations.
First the selected case, where the meteorological parameters during a static stable situa-
tion are studied in detail. Then a more statistical comparison of the WRF model with the
observations for the whole data period is discussed. The comparison of the MET masts
to WRF is also included, to look closer on the urban effect on the meteorology.

Selected Case (January 23 - January 26, 2018)
The theory of turbulence and, thus the calculation of the surface fluxes, are challenged in
a statically stable atmosphere. At the same time, stable environments are important for
the dispersion of tracers. This thesis wanted to see how good WRF would perform under
such circumstances. The longest period that was statically stable (following the definition
in Eq. 2.8), according to the Munch mast observations, was from January 23 18UTC to
January 26 18UTC. The time series of the full vertical profile of the potential temperature
from the selected case can be seen in the appendix (Fig. A.9 - Fig. A.12). 10 minutes
averages from the Munch mast were compared to 10 minutes data from WRF, which will
be discussed after a short evaluation of the meteorological conditions on a larger scale
during this period.
5.1.1 Meteorological Conditions

Fig. 5.1 shows the analysis charts of the geopotential height and surface pressure during
the selected case. On January 23 multiple low pressure systems were located west of
Norway. Since air is rotating anticlockwise around the low pressure systems, the air was
brought up from the warmer south/south-west of Europe, indicated by the isobars. The
low pressure systems moved toward north and east during January 24 and January 25,
and the wind continued from the south. On the afternoon of January 26, the weather
situation had changed. The low pressure situated over Great Britain had moved eastward
out of the Norwegian sea and a high pressure ridge had moved into the Norwegian sea and
50
51 CHAPTER 5. RESULTS AND DISCUSSION
down south of Norway. High pressure systems are associated with clockwise flow which
in this case resulted in winds from a more north/northeasterly direction bringing cold air
from higher latitudes. An inversion situations is expected when a high pressure system
dominates, with its dry air and few clouds. However, from the observations in the Munch
mast from this period (Fig. A.9 - Fig. A.12) it can be seen that in the beginning of the
period there were cold temperatures by the surface, and there were already a statically sta-
ble situation. The warm advection which started around January 23 09UTC, increased
the temperatures. The characteristics of the air in the warm advection enhanced and
maintained the inversion. The warm advection continued until the afternoon of January
26 when colder air from the north was transported by the high pressure system. The
temperature decreased and the inversion was subdued.
(a) (b)
(c) (d)
Figure 5.1: Analysis charts of the 500hPa geopotential height (gpdm) and the sur-
face pressure (hPa) over Europe at a) 18UTC January 23, 2018 b) 18UTC January
24, 2018 c) 18UTC January 25, 2018 and d) 18UTC January 26, 2018 (Source:
www.wetterzentrale.de).
5.1. MODEL COMPARISON TO OBSERVATIONS:
SELECTED CASE (JANUARY 23 - JANUARY 26, 2018) 52
5.1.2 The Munch mast

In Fig. 5.2 the results from the WRF simulations are compared with the observed wind
direction at 10m and 30m (Fig.5.2a), wind speed at 10m and 30m (Fig. 5.2b), the wind
speed difference between 10m and 30m (∆u = u30 − u10 ) (Fig. 5.2c), the potential tem-
perature at 10m and 30m (Fig. 5.2d), the potential temperature difference between 10m
and 30m, ∆θ = θ30 − θ10 (Fig. 5.2e), the calculated bulk Richardson number (Fig. 5.2f)
and the Pasquill’s stability classes defined from Table 2.2 (Fig. 5.2g). The maximum and
minimum values of the bulk Richardson number have been set to -20 and 20 respectively,
to easier see the trends. The model result of the the PBL height (Fig. 5.2h), the fric-
tional velocity (Fig. 5.2i) and the heat flux (Fig. 5.2j) were also included. It was not
observations of these variables for comparison. They were include to see them in context
of the previous mentioned plots, as they are important to the dispersion.
(a)
(b)
(c)
(d)
(e)
(f )
(g)
(h)
(i)
(j)
Figure 5.2: Comparison of the time series of the observed and simulated a) wind di-
rection at 10m and 30m, b) wind speed at 10m and 30m c) wind shear between 10m
and 30m (∆u = u30 − u10 ), d) temperature at 10m and 30m e) difference in potential
temperature (∆θ = θ30 − θ10 ), f ) calculated Bulk Richardson number (Eq. 2.13) where
the minimum and maximum values are set to -20 and 20, respectively for better view
of the details and g) Pasquill’s stability classes, at the Munch mast during the selected
case using 10 minute averages of the observations and 10 minute values for WRF.
Additionally are h) the PBL height, i) the frictional velocity (u∗ ) and j) the heat flux
extracted from WRF for the stable case using 10 minute values.
The model was consistent with the observations of the wind direction in the first part
of the period. The wind direction was mainly from the south during the first 48 hours,
at both 10m and 30m, which also is consistent with the large scale weather conditions
shown in Fig. 5.1. There was a clear deviation around 18UTC on January 25, where
WRF had changed to a northeasterly wind direction while the observations measured
southerly winds, though maybe a bit more varying around 180◦ . This indicates that
WRF had switched from air coming from the sea to air transported over land, and there
was a change of the air masses earlier than expected when comparing with the observed
weather conditions in Fig. 5.1.
The wind speed was generally higher in WRF compared to the observations both at
10m and 30m. While the observations were between 0-12.5 ms−1 , WRF was between
0-19ms−1 . This indicates a stronger near-surface winds in WRF than in the observations.
At some points the deviation between the observations and WRF was almost 10ms−1 , as
seen for example at January 24 00UTC in Fig. 5.2b. At this point the wind speed obser-
vation at 10m was close to 0ms−1 , while WRF is almost 7.5ms−1 . This would give high
concentrations of tracers in a dispersion model if using the observations, but the WRF
data would give larger mixing. The same tendency of WRF overestimating the surface
wind speed was also reported by Tastula and Vihma (2011) and Carvalho et. al (2014b),
using the MYJ, Eta and Noah scheme during winter. As with this study, Carvalho et. al
(2014b) analyzed data close to the coast. They suggested the overestimation to be due to
coastal winds’ strong dependence on the surrounding topography and to high-resolution
coastal interaction mechanisms. This could be for example abrupt roughness changes.
WRF is limited to resolve such features as being a mesoscale model. In this case, the
defined roughness length in WRF at the Munch mast was a clear contributor to the bias.
The land use at the Munch mast grid was defined by WRF as water, as discussed in Sec.
4.4, with a roughness length of 0.0001m. This is smoother than in reality considering for
example the buildings at Bjørvika. When the complexity of the terrain is smoothened by
the model, this will result in a lower drag between the surface and the atmosphere and
so WRF produces stronger wind flow near the surface, as discussed in Sec. 2.1.2. On the
other side, there are uncertainties with the observations as well. The observations were
taken close to the buildings and the wind flow was probably disrupted by the mast itself
and by the close buildings, as discussed in Sec. 3.3.2. However, WRF seemed to follow
the same underlying trend as the observations with a decrease in the beginning to around
January 24 06UTC, with an increase from January 24 12UTC to January 25 06UTC and
from this point on to decrease. This is positive considering the WRF values to be spatial
averages inside one grid box. WRF was closer to the observations from around January
26 00UTC and the rest of the period than in the first part. When seeing this in context
with the wind direction in Fig. 5.2a, it can be seen that the southerly wind speeds were
higher than the northerly in WRF, which is in agreement with what we saw for the ob-
servations saw for the whole measuring period in Fig. 3.13. Further indications of this
can be seen at January 24 between 00UTC and 12UTC, where WRF had some low wind
speeds which fit with some cases of northerly winds at the same times in Fig. 5.2a. On
the other side, in the presence of low wind speed the variation of the wind direction is
higher and determination of the direction accurately is affected. This could explain why
WRF simulated northerly winds in the last part of period, seen in Fig. 5.2a. Carvalho et.
al (2014b) did also report WRF to have the largest errors for the wind direction during
presence of low wind speeds. Deviations in the local pressure distribution in WRF could
be an additional explanation for the difference in wind direction between the observations
and WRF. This was not investigated further or in detail in this thesis.
Even though the wind speed was generally overestimated in WRF, the difference be-
tween the wind speed at 10m and 30m were similar to the observations, as seen in Fig.
5.2c. There were even periods where ∆u is greater in the observations compared to WRF,
especially between January 24 12UTC and the early hours of January 26. The largest ∆u
in the observations was greater than 5ms−1 , while the largest ∆u in WRF were little less
than 3ms−1 . This indicates a generally larger wind shear and probably more mechanical
generated turbulence (Sec. 2.2.2) in reality than WRF simulated. This deviation could
be due to the disturbances in the Munch mast, where the 10m sensor had more buildings
at its level while the 30m sensor was above some of these, as discussed in Sec. 3.3.2.
WRF had also a period from the afternoon of January 25 where the wind was almost
2ms−1 larger at 10m than at 30m. This was not reflected in the observations. We have
not investigated this further.
The temperature in WRF showed the same tendency as the observed values with temper-
atures slowly increasing to January 24 12UTC, which was followed by an almost constant
period until the morning of January 25. The last period showed decreasing temperatures,
Fig. 5.2d. The observations was generally warm the whole period, varying between 0◦ C
and 8◦ C, while WRF varied between -4◦ C and 5◦ C. WRF is consistent with the ob-
servations during the first hours, being almost equal. From around January 24 12UTC,
WRF underestimated with the largest deviations of around 4◦ C. WRF was close to the
observations again at January 25 12UTC, but from around January 25 18UTC it underes-
timated quite a lot. The deviations seen in the temperature in WRF, between January 24
12UTC and January 25 12UTC, could be explained by a weak effect from the sea surface
temperature in WRF during the period of southerly winds. This was also a period of
high wind speeds. The clear deviation in WRF from January 25 18 UTC could be linked
to the change in wind direction from southerly to northerly. Northerly wind direction
brings cold air masses from the interior around Oslo. We do not see this change in the
observations.
The difference in potential temperature between 10m and 30m was not large in neither
WRF nor the Munch mast, Fig. 5.2e. The observed ∆θ was slightly negative during the
first hours, increasing quickly around January 24 00UTC to a difference of 0.5K. It then
lies around a difference of 0.25K with a small increase at around January 26 00UTC, be-
fore decreasing to below zero again. The trend in WRF followed the observations, apart
from around the afternoon of January 25 where ∆θ increased sharply to almost 1.5K.
From the definition of static stability in Eq. 2.8, both WRF and the observations were
weakly stable almost the whole period, but WRF was generally more stable at the end.
Again, the most clear deviation happened around the time when the wind direction in
WRF changed. But unlike the wind speed, this seemed to affect the air closest to the
ground the most with colder temperatures than aloft. This created a more static stable
atmosphere. It is hard to tell if this is due to local effects, the parametrization of the
energy balance, the vertical transport of heat or if the change in wind direction results
in advection of inversions developed in the inland. WRF had a couple of large negative
values between January 26 00UTC-12UTC indicating some fluctuating atmospheric sta-
bility. A more statically stable model compared to the observations could explain the
negative bias seen for the surface temperatures at the end of the period.
The calculated bulk Richardson number varied in both WRF and the Munch mast, Fig.
5.2f. They started around zero, then the observed RiB drastically increased in the obser-
vations around January 24 00UTC with multiple large values until January 24 12UTC.
WRF had only one large positive value in the early morning of January 24. Then both
the simulated and observed RiB decreased to around zero. WRF had large positive values
again at around January 25 06UTC. Large positive values continued in WRF from this
point on, whereas the observed RiB only had a few large positive values around January
26 00UTC. WRF had also two large negative values between January 26 00UTC and
12UTC. The bulk Richardson number would naturally vary a lot with ∆u and ∆θ, which
we have seen vary a bit differently for WRF compared to the observations. However, there
were clearly more positive values than negative. WRF having more, most of them in the
end of the period, and the Munch mast having a bit fewer, but in both the beginning
and the end of the period. Looking back at Table 2.1, the table shows that positive RiB
indicates weak to strong stable atmosphere. This fits with what Fig. 5.2e shows, and that
the occurrence of turbulence was more due to shear forces than thermal forces. With the
Pasquill’s stability classes and their relationship to different RiB limits shown in Table.
2.2, it is easier to get a better picture of the different stability classes involved. As seen
in Fig. 5.2g, there were most occurrence of the classes from D-neutral to G-very stable.
Unlike what we saw in the RiB figure, Fig. 5.2f, there seemed to be a larger part of the
period that was neutral in both WRF and the observations. WRF had more occurrences
of G-very stable compared to the observations.
The PBL height is important as it determines the upper boundary of the PBL as discussed
in Sec. 2.1. As we see from Fig. 5.2h, the PBL height in WRF varied from a couple of
tens of meters (18m) to above 1750m. In the beginning of the period it decreased to the
minimum height 18m, before it increased sharply around January 24 06UTC. The PBL
height continued to vary a lot, but was high to around January 25 06UTC. It decreased
to the minimum height at the end of January 25. The PBL height was less than 250m
the rest of the period. The varying PBL height determines the vertical extension of tur-
bulent mixing of momentum, heat and tracers, and is (with the PBL scheme chosen in
this study) defined by the TKE, Sec. 4.2.3. By comparing the trend of the PBL height
in Fig. 5.2h to that of the frictional velocity in Fig. 5.2i and the heat flux in Fig. 5.2j,
we see that u∗ follows the same trend, while the heat flux is some what opposite. u∗ is
a measure of mechanical generated turbulence (Sec. 2.2.2). This was highest around the
time the PBL height was highest. The heat flux on the other hand, was most negative
when the PBL was at its highest. From Sec. 2.1.1 we know that a negative flux means
it is directed downwards to the surface, thus heat is transfered from the warmer air to
the colder surface. A downward heat flux is typically associated with low PBL height,
because of low thermal generated turbulence (Sec. 2.1), which we do not see in our case.
5.1.3 The MET Stations

Wind observations at multiple heights and temperature observations above 10m was not
available from the weather mast at Hovin or Blindern, thus the different stability param-
eters was not calculated as was done for the Munch mast.
Hovin
For the weather mast at Hovin, the wind direction at 25m, Fig. 5.3a, the wind speed at
25m, Fig 5.3b, the temperature at 10m, Fig. 5.3c were compared.
(a)
(b)
(c)
Figure 5.3: Comparison of the time series of the observed and simulated a) wind di-
rection at 25m, b) wind speed at 25m and c) temperature at 10m, at Hovin during the
stable case using 10 minute averages for the observed wind data, hourly averages for
the observed temperature, and 10 minute values for all WRF data.
The change in wind direction with time was very similar to what was seen at the Munch
mast. The wind was mainly southerly until around January 25 12UTC for both the obser-
vations and WRF, Fig. 5.3a, but after this point the wind seemed to be more fluctuating
in both the observations and WRF compared to the Munch mast. The same tendency
of a change towards northerly winds was seen in WRF, while the observations were more
southerly. Around January 26 06UTC the observations changed gradually towards west-
erly winds and at the end of the period the observations were also close to northerly
winds. This change was not seen in the observations at the Munch mast. The reason
that the simulated and observed wind direction were more similar at Hovin, may be due
to the local disturbances of the wind by the buildings and the mast itself at the Much
mast. The tendency of the wind speed was similar at Hovin as it was at the Munch mast.
The wind speed decreased in the beginning, increased from around January 24 00UTC
with a max point at January 25 00UTC, before it decreased and stayed low for the rest of
the period, Fig. 5.3b. WRF overestimated the wind speed throughout almost the whole
period, but was closest to the observations at the end of the period. This occurred around
January 25 12UTC. The most striking difference at Hovin compared to the Munch mast,
was that the observed wind speed varied between 0-10ms−1 and WRF varied between
0-14ms−1 . This was 2.5ms−1 lower for the observations and 5ms−1 lower for WRF than
was seen at the Munch mast. This could indicate that there was some weakening effect
by buildings. Even though Hovin was not a greatly urban area, the roughness length was
in WRF 0.14m. So WRF recognized Hovin as an urban area with a larger surface rough-
ness than that at the Munch mast. This resulted in smaller near surface wind speeds at
Hovin compared to the Munch mast. The urban canopy model assumes the buildings to
be below the first model level (Fig. 4.6) and has no direct effects of buildings included,
which could explain why WRF still overestimated the wind speed. Overestimation of the
wind speed, when using the single-layer urban canopy model (SLUCM), was reported by
Kim et al. (2013). They also suggested the cause to be a smoothening of the surface
roughness in WRF by not sufficiently parameterize urban areas. The lowest speeds occur
at the same time as the wind direction varied most, which could again be the reason for
the fluctuating direction in the first place.
The same tendency was seen with the temperature in Fig. 5.3c as for the Munch mast.
WRF underestimated the temperature from around January 24 12UTC at this site also.
The greatest difference compared to the Munch mast, was that both the observations and
the WRF results at Hovin were about 1◦ C colder throughout the period. This was prob-
ably due to greater distance from the warmth of the fjord, as expected. This could also
indicate that there was not any compensating heat from the urban heat island. This can
be due to the parameterization of the surface temperature and heat flux from buildings,
roads and walls by the UCM. The simplest version was used in this study, the single-layer
UCM (Sec. 4.2.3) and the default urban morphology. A more complex UCM could prob-
ably perform model results closer to the observations. Kim et al. (2013) also reported an
underestimation of the temperature.
Blindern
For the weather mast at Blindern the wind direction at 26.5m, Fig. 5.4a, the wind speed
at 26.5m, Fig 5.4b, the temperature at 10m, Fig. 5.4c were compared.
(a)
(b)
(c)
Figure 5.4: Comparison of the time series of the observed and simulated a) wind direc-
tion at 26.5m, b) wind speed at 26.5m and c) temperature at 10m, at Blindern during
the stable case using 10 minute averages for the observed wind data, hourly averages
for the observed temperature, and 10 minute values for all WRF data.
WHOLE PERIOD (JANUARY 9 - MARCH 1, 2018) 62
The observed and modeled values at Blindern were not surprisingly very similar to what
we saw at Hovin. The wind speed was still generally overestimated by WRF, but WRF
was closer to the observations at Blindern than what we saw for both the Hovin mast
and the Munch mast. This could be due to the surface roughness being represented bet-
ter at Blindern in WRF, than that simulated for Hovin and the Munch mast. Another
explanation can be that the observations at Blindern were less disturbed, compared to
the measurements at Hovin and Blindern. The wind direction was also like we found at
Hovin, but seemed to have more fluctuating periods where the wind direction varied a lot,
Fig. 5.4a. These periods fits with the periods of winds below 2ms−1 in both the observed
and simulated values for Blindern. This indicates the strength of the wind flow to be the
reason for fluctuations in the wind direction.
The tendency of the temperature at 10m at Blindern compared to Hovin was similar.
However, Hovin seemed to be half a degree warmer at the warmest and a degree colder at
the coldest compared to Blindern. Where Hovin varied from -4◦ C to a little above 6◦ C,
Blindern varied from -3◦ C to a little under 6◦ C, Fig. 5.4c. Blindern had the smallest tem-
perature interval of the three stations. Even if WRF was still underestimating compared
to the observations, this could indicate that also the parameterization of the heating effect
of the urban heat island on the surface temperature and heat flux were closer to reality
at Blindern.

Whole Period (January 9 - March 1, 2018)
In this section the hourly WRF results are compared against the observed values for
the whole measuring period, from January 9 to March 1. The result of the temperature
and wind speed are presented as scatter plot, the wind direction as wind roses, and the
different stability parameters as bar plots. The statistical metrics used for the scatter
plots to evaluate the simulations were the bias, the Root-Mean-Square-Error (RMSE)
and the correlation coefficient (R). The 1:1 line indicating a perfect agreement between
the observations and WRF is also included in the scatter plots.
5.2.1 The Munch mast - Temperature

Fig. 5.5 shows the observed temperatures in the Munch mast along the x-axis compared
with the corresponding WRF values along the y-axis, for each height in the Munch mast.
(a) (b)
(c) (d)
(e) (f )
Figure 5.5: Observed v.s. modeled temperature (◦ C) at the Munch mast, January 9 to
March 1, using the closest 10 min. averaged value to every hour for the observations
and hourly values for WRF. The root mean square error (RMSE), the bias, the number
of valid values (n), and the correlation coefficient (R) are also shown.
WRF reproduced the observed temperatures with a general underestimation at all heights,
thus negative bias as seen in Fig. 5.5. What was seen in the selected case, seemed to
be the case for the whole period as well. The 2m comparison had the smallest bias and
RMSE, which also can seen by its symmetry in the scatter plot Fig. 5.5a. This suggests
that the chosen value for the TSK for the sea, as discussed in Sec. 4.5, was satisfying
and similar to the sea surface temperature in reality. The largest bias occurred for the
10m comparison, Fig. 5.5b. The bias decreased for each increasing level. The 75m com-
parison had the second smallest bias after the 2m comparison, Fig. 5.5f. The RMSE
varied not much, but decreased a little with height as well. The correlation coefficient
(R) was similar for all heights. It was somewhat surprising that the largest bias was
at 10m, when the best estimation was at 2m. Since the effect of the TSK seemed to
decrease considerably already at 10m, it could indicate that WRF was not transporting
enough heat in the vertical. This could have something to do with how the surface layer
scheme connects the surface to the air above. There was unfortunately not enough time
to investigate the cause of the weak heat transport further. The improvement in the bias
with height, indicates that WRF managed to make simulations closer to the observations
as the physics near the ground got less influential on the atmospheric motion. This was
as expected since the physics close to the ground is difficult to handle, as established in
the theory. Since the Munch mast was located in an area where both land-sea interaction
and urban features affected the meteorology, the surface parameterization is a complex
matter alone. The general negative bias could be explained by the PBL parameterization
as well. Previous studies have shown that local PBL schemes generally tend to produce
unrealistically shallow and moist PBL’s (Alapaty et al., 1997; Bright and Mullen, 2002;
Stensrud and Weiss, 2002) as pointed out by García-Díez et. al (2013). Tastula and
Vihma reported also negative surface temperature biases in WRF when comparing the
result of different PBL scheme options to each other for the Antarctic atmosphere during
winter. They suggested the nonlocal Yonsei University PBL scheme (Hong et al., 2006)
together with the fifth-generation Pennsylvania State University-National Center for At-
mospheric Research Mesoscale Model (MM5) surface layer scheme (Zhang and Anthes,
1982) to be the best choice for the PBL parameterization.
The differences in the snow cover between the observations and WRF could contribute
to the temperature bias in WRF. Fig. 5.6 show the snow cover in the fourth domain
at four different days during our measuring period. The whole area which was defined
as urban built up land in Fig. 4.5a has a snow cover throughout the period. The Oslo
area starts out with a snow depth of 0.5m and has a snow depth above 1m at the end in
WRF. This would probably not be accurate for the city center. In reality the snow gets
transported away and the roads are cleared in the city center. So the surface would not
be fully covered of snow even in a snow rich winter. The Munch mast was situated in a
construction area which most likely was cleared of snow as well. This has an effect on
the surface energy balance in the model, as discussed in Sec. 2.1.1. When we have some
centimeters of snow instead of bare ground there is a greater cooling of the air above the
surface, because snow is isolating the heat from the ground. This thesis did not investi-
gate the snow cover, or a possible snow fraction in the city center, further.
(a) (b)
(c) (d)
Figure 5.6: The snow depth (m) in the fourth domain in WRF at a) January 15 -
2018, 12UTC, b) January 30 - 2018, 12UTC, c) February 15 - 2018, 12UTC and d)
February 28 - 2018, 12UTC.
5.2.2 The Munch mast - Wind Speed

Fig. 5.7 shows the observed wind speed in the Munch mast along the x-axis compared
with the corresponding WRF values along the y-axis, at 10m and 30m.
(a) (b)
Figure 5.7: As Fig. 5.5 but for wind speed (ms−1 ).
As was seen in the selected case, WRF continued to overestimate the wind speed through
the whole period. While it was decided to set a limit of the observations of 15ms−1 (Sec.
3.3.2), WRF had some wind speed values of almost 17ms−1 . As seen from the clusters of
points for both 10m and 30m in Fig. 5.7, the main part of the observations were lower
than 4ms−1 . They were overestimated by almost 10ms−1 at some time steps by WRF.
This gave the impression that WRF have problems with reproducing the weak wind flows,
as also was seen in the selected case. The overestimation of the wind speed was larger at
30m. The wind speed had also a larger spreading of the points in the scatter plot than
the wind speed at the 10m level due to some larger observed wind speeds at 30m. There
could be different reasons for this. Carvalho et. al (2014b) reported as mentioned in
the selected case the same findings of overestimation of the wind speeds, especially wind
speeds below 8ms−1 . Another study by Carvalho et al. (2014a) detected similar tenden-
cies under different conditions for the initialization and boundaries. They suggested this
to be due to the chosen PBL scheme, where the lowest errors for low wind speeds were
performed by an non-local closure scheme and not a local closure scheme as used in this
study. They argued that the non-local closure scheme was more suitable for unstable at-
mospheric conditions. Since weak winds often can be related to an unstable atmosphere,
when convective turbulence dominates over mechanical turbulence. They expected the
non-local scheme to perform better. And as the urban scheme connects the surface to
the surface layer scheme and in turn the PBL scheme, an inaccurate representation of the
surface roughness would contribute to the simulation of momentum exchange and so the
overestimation of the wind speed. This was also mentioned to be a cause in the selected
case. As was pointed out previously, WRF do not directly represent the buildings and
the measuring mast. Hence, it can not represent the local wind flows generated around
the buildings and mast either, which probably was contributing to the bias in WRF. The
RMSE and correlation did not vary a lot between the to heights, although RMSE was
largest at 30m and the correlation was largest at 10m.
5.2.3 The Munch mast - Wind Direction

Fig. 5.8 shows the distribution of the observed wind direction in the Munch mast and
the WRF values at 10m and 30m. These are presented as wind roses.
(a) (b)
(c) (d)
Figure 5.8: The observed wind direction and wind intensity at a) 10m and b) 30m,
and the model values at c) 10m and d) 30m, for the Munch mast, from January 9 to
March 1, using the closest 10 minute averaged value to every hour for the observations
and hourly values for the model data. The direction is given as relative to a northward
heading (0◦ ) and the limit of calm wind is 0.5ms−1 .
The wind roses in Fig. 5.8 show that the dominating wind direction in WRF was more
easterly at both 10m, Fig. 5.8b, and 30m, Fig. 5.8d, whereas the observed wind direction
was more northerly at both 10, Fig. 5.8a, and 30m, Fig. 5.8c. The dominating wind
direction sectors in WRF were not the same as for the observations, which was not
surprising as the observations were highly affected by the close standing buildings and
the measuring mast, as pointed out before. It was also noticed that WRF had only winds
from 0◦ − 180◦ . Winds from west were absent and the southerly winds were close to be
missing as well. This seems strange, but we do not have a clear explanation to the cause
of this. One suggestion could be, as seen from the observed wind roses, that the winds
from west had low intensity. As was suggested in the previous section, WRF seems to fail
to depict the calmest specter of the wind speed. This could explain why the model does
not depict the whole direction range as well. This, however, does not explain why WRF
had little southerly winds, as these were the strongest in intensity in both the observations
and the model. This was not investigated further in this thesis.
5.2.4 The Munch mast - Atmospheric Stability

Fig. 5.9 shows the comparison of the difference in potential temperature between 10m and
30m in Fig. 5.9a, the wind shear between 10m and 30m in Fig. 5.9c, the bulk Richardson
number in Fig. 5.9e, the stability distribution based on Pasguill’s stability classes in Fig.
5.9f, the stability distribution based on the potential temperature gradient in Fig. 5.9b,
and the distribution of the wind shear in Fig. 5.9d. Since there was wind observations
at 10m and 30m in the Munch mast, this layer was used to consider the stability. To
better reveal the variation in the distribution of ∆θ and ∆u, it was decided to separate
the result into five groups, which is indicated on the x-axis.
(b)
(a)
(d)
(c)
(e) (f )
Figure 5.9: Point comparison/the statistical distribution of the observed and the modeled
a)/b) potential temperature gradient c)/d) wind speed gradient and e) bulk Richardson
number/f ) Pasquill’s stability classes at the Munch mast, between 10m and 30m for the
whole period from January 9 to March 1 using the closest 10 minute average to every
hour for the observations and hourly values for WRF.
The comparison of ∆θ in Fig. 5.9a shows little correlation between the observations and
WRF. The largest deviation occurred when ∆θ was close to zero in the observations. ∆θ
was largely overestimated in the model, up to a deviation of two degrees at some points.
This overestimation indicate a more statically stable model, which can be seen from the
distribution in Fig. 5.9b as well. The distribution shows that the largest part of ∆θ from
both WRF and the observations falls into the category −0.25K ≤ ∆θ < 0.25K. WRF
had larger frequency of ∆θ>0.25K compared to the observations. Large negative values
of ∆θ were not frequent in neither the model nor the observations, but WRF had the
largest frequency of values in the range -0.75K≤ ∆θ<-0.25K. As mentioned previously,
the overestimation of statically stable events could be due to this study choice of the local
PBL scheme. A nonlocal PBL scheme could improve the overestimation.
The difference in wind speed, ∆u had little correlation between WRF and observations
as well. WRF was generally overestimating the wind speed difference. However, there
were some larger values in the observations, especially when the wind speeds in WRF
were between 0ms−1 and 3ms−1 . The distribution in Fig. 5.9d shows that WRF had
most occurrences of high wind speed differences, ∆u>1ms−1 . This was consistent with
what we saw of the distribution of ∆θ. Statically stable atmosphere strengthens the wind
gradient, as discussed in Sec. 2.3.6.
The calculated bulk Richardson number shows no correlation between the observations
and WRF in Fig. 5.9e. This was not that surprising since the bulk Richardson number
is very sensitive in changes of ∆u and partly sensitive to ∆θ, as discussed in the selected
case as well. Nevertheless, the distribution of Pasquill’s stability classes based on the bulk
Richardson number in Fig. 5.9f, shows that the model and the observations match quite
well. It is also seen in Fig. 5.9f that even if WRF seemed to be more statically stable and
had larger wind shear, when they where combined in the bulk Richardson number there
were actually less "very stable" cases (Pasquill’s stability class G) in WRF compared to
the observations. The rest of the WRF distribution in the other classes was also close
to the observations. There could be some "compensating effects" in term of ∆θ and ∆u
which makes the statistical distribution as good as it is. The similar distribution of the
stability classes between WRF and the observations also show that the point comparison
in Fig. 5.9e of the bulk Richardson number could be misguiding. The bulk Richardson
number would give completely different atmospheric conditions with the model result
compared to the observed if it was used for predictions of for example the air quality
hour by hour. But the distribution in Fig. 5.9f show that the WRF data could be more
meaningful to use in for example calculations for larger mean values, like yearly means or
other longer periods.
As it would be interesting to see the distribution of ∆θ at the higher levels in the Munch
mast, the difference in potential temperature between 30m-50m and 50m-75m were com-
pared. The result is shown in Fig. 5.10a and Fig. 5.10b. The difference between 50m and
75m, is shown in Fig. 5.10c and Fig. 5.10d.
(b)
(a)
(d)
(c)
Figure 5.10: Point comparison/the statistical distribution of the observed and the mod-
eled potential temperature gradient between a)/b) 30m and 50m and c)/d) 50m and 75m
at the Munch mast, between 10m and 30m for the whole period January 9 to March
1 using the closest 10 minute average to every hour for the observations and hourly
values for WRF.
The comparison of ∆θ between 30m and 50m in Fig. 5.10a, and ∆θ between 50m and
75m in Fig. 5.10c shows the same as we saw for the difference of ∆θ between 10m and
30m in Fig. 5.9a. There was no correlation between WRF and the observations. The
most striking feature was that the observations for ∆θ30−10 lies between -1K and 1K, while
∆θ50−30 lies mainly between 0K and 0.5K, and ∆θ75−50 lies between -1K and 0K. WRF
overestimated here as well, but managed to have similar distribution for ∆θ50−30 with no
negative values less than -0.25K, as seen in Fig. 5.10b. The distribution of ∆θ75−50 on the
other hand, was totally opposite. Where the observations had almost no values on the
positive side larger than 0.25K and a large group lower than -0.25K, WRF had no values
lower than -0.25K and a large group at both 0.25K≤ ∆θ<0.75K and ∆θ ≥0.75K as seen
in Fig. 5.10d. There is not a good explanation for the deviation of the distribution in the
50m-75m layer, which sticks out compared to the other layers where the distribution was
more similar to the observed. It could be due to the impact of the buildings or the mast
on the observations. There could be some small errors in the measurements at 75m or in
the calibration of the observations at 75m, which could have an impact on the distribution
of the groups, but this is uncertain.
5.2.5 The MET Stations

The result from the simulations to the observations at Hovin and Blindern were compared,
as with the selected case, to examine the urban effect and the parameterization done by
the urban canopy model closer.
Hovin
Fig. 5.11 shows the comparison of the 2m temperature, Fig. 5.11a, the 10m temperature,
Fig. 5.11b, the wind speed at 25m, Fig 5.11c, the wind rose for WRF, Fig. 5.11d and the
wind rose for the observed wind directions, Fig. 5.11e.
(a) (b)
(c)
(d) (e)
Figure 5.11: a) and b) as for Fig. 5.5, c) as for Fig. 5.7 and d) and e) as for Fig. 5.8
but for Hovin at its respective measuring heights.
The 2m temperature comparison in Fig. 5.11a shows weak correlation between the ob-
servations and WRF. There were a larger RMSE and a larger negative bias compared to
those at the Munch mast, as seen in Table 5.1. One explanation to this difference is that
the Munch mast grid was a water grid while Hovin was a land grid. Hence, Hovin does
not get as direct effect of the fjord as the Munch mast. However, Hovin was an urban
grid, and the negative bias can be explained by a too weak anthropogenic heat flux. So
there was not a large contribution of human activity in the urban surface energy balance,
Eq. 2.17, by the SLUCM. This could in turn explain the similar results in the selected
case. The RMSE and bias decreased a little at 10m shown in Fig. 5.11b, but WRF was
still pretty low compared to the observations. The SLUCM default category for Hovin
was suburban. This seemed to be a good description of Hovin as it was not as urban as for
example the city center of Oslo (Sec. 3.4.1). The temperature simulations would probably
have been improved if the geometric and thermal parameters (e.g. building height, diur-
nal maximum of anthropogenic heat, etc.) were specified for this study, instead of using
default settings. Urban features affecting the meteorology varies a lot from city to city.
It is dependent on many factors such as the size, the urban density and the population
density. The author of this thesis was unsuccessful in the search of other studies done
in Norwegian cities including an UCM. However, studies which used an urban model has
been done in other cities. The evaluation study of the performance of WRF in Paris
done by Kim et al. (2013) found a small improvement in the underestimation of the tem-
perature when including the UCM. They suggested that a more complex UCM like the
the Building Environment Parametrization (BEP) (Martilli et al., 2002) or the Building
Energy Model (BEM) (Salamanca et al., 2010) could improve the result. In addition they
suggested to introduce multiple urban land-use categories (e.g. high intensity, medium
intensity, and low intensity). Kim et al. (2013) compared their results to five other studies
done in other cities (Flagg and Taylor, 2011; Grossman-Clarke et al., 2010; Lee et al.,
2011; Salamanca et al., 2011, Miao et al., 2009). Kim et al. (2013) temperature bias
ranged between the maximum and the minimum of the previous studies, indicating that
the complexity of the urban effect was a problem in the range of studies. However, there
could be a problem with WRF having a general cold bias. Areas, which are not directly
connected to urban areas, could be colder compared to reality as well, but this was not
investigated further in this thesis. Hovin had snow cover in WRF similar to the Munch
mast, as seen in Fig. 5.6.
The wind speed was also at Hovin overestimated, as seen in Fig. 5.11c. This was the same
pattern as was seen for the selected case. This indicated that the suggested reason of too
smooth roughness in WRF, was generally a problem for the whole period. The rough-
ness length defined at Hovin was 0.14m, whereas a typical suburban roughness length
according to Oke (1987) lies between 0.4m and 1.8m depending on the density of the
roughness elements. As discussed previously, the smoothening of the surface roughness
will result in less drag and turbulence. This results in a narrower zone of frictional drag
and not as reduced wind speeds as expected. Overestimations of the wind speed was also
reported by Kim et al. (2013). They suggested that a more complex UCM could improve
the modeled wind speed as well. Compared to the other five studies mentioned earlier,
Kim et al. (2013) wind speed bias was lower. When comparing the distribution of the
wind direction in the model in Fig. 5.11d, and the observations in Fig. 5.11e, the wind
direction seems to be more similar to what was found for the Munch mast. This indicated
that the observations was not as locally disturbed as the Munch mast, and so the model
was closer to the observations.
Blindern
Fig. 5.12 shows the comparison of the 2m temperatures in Fig. 5.12a, the 10m tempera-
tures in Fig. 5.12b, the wind speeds at 26.5m in Fig 5.12c, the wind rose for the simulated
wind directions in Fig. 5.12d and the wind rose for the observed wind directions in Fig.
5.12e.
(a) (b)
(c)
(d) (e)
Figure 5.12: a) and b) as for Fig. 5.5, c) as for Fig. 5.7 and d) and e) as for Fig. 5.8
but for Blindern at its respective measuring heights.
The results for Blindern was not surprisingly very similar to those at Hovin. The temper-
ature at both 2m and 10m at Blindern had a similar correlation as those at Hovin and the
Munch mast, and the bias and RMSE were similar as well. WRF had also underestimated
the temperature at Blindern, as can be seen with a negative bias in Fig. 5.12a and Fig.
5.12b. This was probably due to a weak urban heat island effect as discussed earlier for
Hovin. Blindern had snow cover similar to that at Hovin an the Munch Mast as well,
as seen in Fig. 5.6. At Blindern the snow depth was measured, and it can be seen from
Fig. 5.13 that it was deeper in WRF. The wind speed results in Fig. 5.12c shows an
overestimation of the wind speed. This was probably again due to a smoothing of the
surface roughness at Blindern by WRF and that WRF do not include the buildings ex-
plicitly. The wind direction however, have the best correlation between the observed and
simulated values at Blindern. The greatest frequency was recorded for the same direction
in the observations and the simulations. This fit of the dominating wind direction has
not been seen at any of the other mast locations.
The results from the Munch mast, Blindern and Hovin for the temperatures and wind
speeds are shown in Table 5.1. There were a few results which stood out. First, the
bias at 2m at the Munch mast, which was a lot smaller compared to the other stations.
Second is the bias for the wind speed at Blindern, which was also much smaller compared
to the other wind speed biases. The third result we want to point out was the correlation
coefficient for the wind speed at Blindern as well, which was quite larger than the others.
Apart from this, the values of the bias, RMSE and correlation were relatively similar
between the three stations.
Figure 5.13: The observed and simulated snow depth (m) at Blindern, from January 9
to March 1, using hourly values.
Station Parameters BIAS RMSE R
Munch mast T2m (◦ C) -1.522 1.781 0.766

Hovin T2m (◦ C) -5.055 2.927 0.747
Blindern T2m (◦ C) -4.430 2.668 0.737
Munch mast T10m (◦ C) -3.423 2.665 0.760

Hovin T10m (◦ C) -4.102 2.752 0.770
Blindern T10m (◦ C) -3.178 2.252 0.798
Munch mast T30m (◦ C) -3.307 2.677 0.766

T50m (◦ C) -3.199 2.602 0.773
T65m (◦ C) -2.916 2.566 0.775
T75m (◦ C) -2.679 2.528 0.780
Munch mast W10m (ms−1 ) 2.762 1.658 0.526

Munch mast W30m (ms−1 ) 3.849 2.265 0.432
Hovin W25m (ms−1 ) 2.735 1.724 0.599
Blindern W26.5m (ms−1 ) 0.913 2.033 0.705
Table 5.1: Overview of the biases, root mean square errors (RMSE) and correlation
coefficients (R) from the comparisons of the simulated temperatures (T) and wind speeds
(W) in WRF with the observations at the Munch mast, Hovin and Blindern.
Chapter 6
Summary and Conclusions
In this thesis, meteorological observations were collected in Bjørvika, Oslo, compared to

observations done by the Norwegian Meteorological Institute at Hovin and Blindern, be-
fore compared to simulations carried out with the Weather Research and Forecast model
(WRF). WRF had a horizontal resolution of 0.3km in the inner domain and was combined
with the single-layer urban canopy model. This have been done in order to investigate
how well the model represent the meteorology in an urban area during winter conditions
and how well it resolves important meteorological parameters used in local air dispersion
modeling. The period January 9 to March 1, 2018, was simulated. Additionally, a shorter
selected period was simulated, January 23 to January 26, 2018, with more frequent out-
puts from WRF to look closer on an event with stable stratification.
In the selected case the wind direction and wind speed at 10m and 30m, the wind speed
difference between 10m-30m, the temperature at 10m and 30m, the temperature difference
between 10m-30m, the Bulk Richardson number and the Pasquill stability classes in the
10m-30m layer were compared to WRF and discussed. The results from the selected case
show that WRF mainly underestimated the temperatures and overestimated the wind
speeds compared to the observations. Both the observations and WRF were mostly stati-
cally stable during the selected case. When looking at the combined dynamical and static
stability represented in the Pasquill classes they were both in the neutral to very stable
categories. However, WRF seemed to be more stable than the observations. The largest
deviations were connected to the clear change from southerly to northerly wind flow in
the last part of the period in WRF. This change was not reflected in the observations.
The results from Hovin and Blindern continued the pattern that was seen for the Munch
mast, with underestimation of the temperature and overestimation of the wind speed. For
these locations, only the temperature at 10m, and the wind direction and wind speed at
one level (Hovin-25m, Blindern-26.5m) were compared with WRF.
In the comparison of the whole period the temperatures at 2m, 10m, 30m, 50m, 65m
and 75m, the wind speed and wind direction at 10m and 30m, the difference in potential
temperature between 10m-30m, 30m-50m and 50m-75m, the difference in wind speed be-
tween 10m-30m, and the bulk Richardson number and Pasquills stability classes in the
10m-30m layer at the Munch mast were compared to WRF and discussed. The results
for the whole measuring period was consistent with what we saw for the selected case.
78
79 CHAPTER 6. SUMMARY AND CONCLUSIONS
The temperatures were underestimated, the wind speeds were overestimated and it was
a more stable atmosphere in WRF in terms of both the potential temperature gradient
and the Pasquill stability classification. The same parameters as for the selected case
were compared for the whole measuring period at Hovin and Blindern as well, which
showed the same results as for the Munch mast. There were a few results that stood out:
the 2m temperature bias at the Munch mast, BIAS = −1.522, which was the smallest
compared to the other locations. The wind speed bias, BIAS = 0.913, at Blindern,
which was the smallest compared to the other locations. And the wind speed correlation
coefficient, R = 0.705, at Blindern, which was the largest compared to the other locations.
It can be conclude from this thesis results that there were large differences between the
observed and simulated meteorology in Bjørvika, Oslo during January 9 to March 1, 2018.
From the comparison of WRF to the observations, it was seen that they differed in atmo-
spheric structure. The observations were more unstable to neutral in terms of the static
stability, while WRF was more neutral to stable. This could be due to WRF not reflecting
the effect of the fjord or the urban heat island well. This could also be due to our choice
in PBL scheme. Local PBL schemes have shown to overestimate the occurrence of stable
events in previous studies as well. Using a nonlocal scheme could improve this. Especially
the simulated wind speed in the urban area was different from what was observed. This
was not surprising, since WRF has no direct effects of buildings or the measuring mast
included. It was also seen that the sea surface temperature in the fjord in WRF was
of great importance. The simulations turned out to be closer to the observations in the
Munch mast after modifying the too coarse sea surface temperature originally defined by
the ECMWF data.
Using WRF in local dispersion studies will be demanding, according to this study. There
can not be expected to get correct results of the dispersion in an urban area or an area
with buildings, since the model results deviated significantly from the observations. This
was especially seen when comparing the hourly values of the wind speed, wind direc-
tion and the stability in WRF with hourly averages of the observations. However, when
summing up the observed and simulated potential temperature gradient and Pasquill sta-
bility classes it was seen that the general distributions in WRF was more similar to the
observed. This indicates that the WRF data could be more useful when evaluating tracer
concentrations on a larger temporal scale. Although the distribution results gave some
differences as well. Especially the large occurrence of static stability and high wind speeds
in WRF, as already mentioned, which will affect concentration averages.
In terms of future work we would suggest using observations of surface fluxes when evalu-
ating WRF. Surfaces fluxes would probably be easier to compare to the simulated values
instead of observations at different levels, which requires interpolation of the simulated
data. A sensitivity test to compare the different PBL scheme choices combined with the
single-layer urban canopy model during winter at higher latitudes, could also give better
understanding in which combination that gives the best representation of the meteorolog-
ical parameters in an urban area in the wintertime. We would also suggest to investigate
the use of more complex urban canopy models and introducing different urban intensities
for the land use categories (e.g. high intensity, medium intensity, low intensity), which
80
may correct some of the deviations seen in this study.
We wish to point out that this project should be regarded as a kind of pilot observa-
tional study. Our observations in the Munch mast were measured close to a building,
introducing uncertainties to the quality of the measurements. The opportunity to get
observations at a less disrupted area, where not both the effect of the sea and the city
were important, would have been preferable. Because of the circumstances this were not
possible to achieve. However, we would suggest the possibility to do a correction of the
wind data from this study. Simulations of the wind field around the mast and the Munch
museum with CDF (Computational Fluid Dynamics), could improve the wind data and
make it more comparable to the WRF simulations.
Appendix A
Additional Figures and Tables
Result of the climate chamber test
Figure A.1: The result of the sensitivity test of the temperature sensors from the climate
chamber.
82
83 APPENDIX A. ADDITIONAL FIGURES AND TABLES
Wind climatology at Blindern
Figure A.2: The climatology of the wind measured at Blindern during winter from 1995
to 2014.
Overview of the six lowest η-levels in WRF
η -levels Approx. Height (m)

1.0 0
0.9976 18
0.9937 48
0.9905 72
0.9874 96
0.9847 120
Table A.1: Table with an overview of the lowest six η-levels and corresponding approx-
imate height in meter.
84
Land use with full color description
Figure A.3: The land use of the fourth domain with the complete bar of color descrip-
tions.
Comparison of the temperature results from WRF before and after the TSK
modification at 10m, 30m, 50m, 65m and 75m
(a)
(b)
Figure A.4: Comparison of the time series of the observed temperature (in Celsius) at
10m and the model temperature (in Celsius) at 10m from a) the original run and b)
the modified (TSK=276K) run, during the whole period, January 9 - March 1. The
closest 10 minute average to every hour is used for the observations, and hourly values
are used for WRF.
86
(a)
(b)
are used for WRF.
(a)
(b)
are used for WRF.
88
(a)
(b)
are used for WRF.
(a)
(b)
are used for WRF.
90
The vertical profile of the potential temperature for the Selected Case
Figure A.9: The time series of the observed potential temperature (K) at all heights in
the Munch mast on January 23, using hourly means.
Figure A.10: The time series of the observed potential temperature (K) at all heights
in the Munch mast on January 24, using hourly means.
Appendix B
namelist.input
92
&time_control
run_days = 0,
run_hours = 0,
run_minutes = 0,
run_seconds = 0,
start_year = 2018, 2018, 2018, 2018,
start_month = 01, 01, 01, 01,
start_day = 09, 09, 09, 09,
start_hour = 00, 00, 00, 00,
start_minute = 00, 00, 00, 00,
start_second = 00, 00, 00, 00,
end_year = 2018, 2018, 2018, 2018,
end_month = 01, 01, 01, 01,
end_day = 18, 18, 18, 18,
end_hour = 00 ,00, 00, 00,
end_minute = 00, 00, 00, 00,
end_second = 00, 00, 00, 00,
interval_seconds = 21600,
input_from_file = .true.,.true.,.true.,.true.,
history_interval = 60, 60, 60, 60,
frames_per_outfile = 1, 1, 1, 1,
restart = .false.,
restart_interval = 180,
io_form_history = 2,
io_form_restart = 2,
io_form_input = 2,
io_form_boundary = 2,
debug_level = 0,
/
&domains
eta_levels = 1.000, 0.9976, 0.9937, 0.9905, 0.9874,
0.9842, 0.9811, 0.978, 0.9749, 0.9717,
0.9684, 0.9646, 0.9606, 0.956, 0.9509,
0.9451, 0.9386, 0.9311, 0.9225, 0.9127,
0.9014, 0.8885, 0.8737, 0.8569, 0.8378,
0.8165, 0.7927, 0.7664, 0.7378, 0.7068,
0.6738, 0.6389, 0.6025, 0.565, 0.5266,
0.4878, 0.4489, 0.4101, 0.3719, 0.3343,
0.2977, 0.2621, 0.2277, 0.1945, 0.1627,
0.1322, 0.103, 0.0752, 0.0488, 0.0237,
0.000,
time_step = 60,
time_step_fract_num = 0,
time_step_fract_den = 1,
max_dom = 4,
e_we = 103, 106, 103, 103,
e_sn = 100, 106, 100, 100,
e_vert = 51, 51, 51, 51,
p_top_requested = 5000,
num_metgrid_levels = 138,
num_metgrid_soil_levels = 4,
dx = 13500, 2700, 900, 300,
dy = 13500, 2700, 900, 300,
grid_id = 1, 2, 3, 4,
parent_id = 1, 1, 2, 3,
i_parent_start = 1, 48, 38, 40,
j_parent_start = 1, 40, 34, 32,
parent_grid_ratio = 1, 5, 3, 3,
parent_time_step_ratio = 1, 5, 3, 3,
feedback = 0,
smooth_option = 0,
/
&physics
mp_physics = 5,5,5,5,
ra_lw_physics = 4,4,4,4,
ra_sw_physics = 4,4,4,4,
radt = 3,3,3,3,
sf_sfclay_physics = 2,2,2,2,
sf_surface_physics = 2,2,2,2,
bl_pbl_physics = 2,2,2,2,
bldt = 0,0,0,0,
cu_physics = 0,0,0,0,
cudt = 5,5,5,5,
isfflx = 1,
icloud = 1,
ifsnow = 0,
surface_input_source = 1,
num_soil_layers = 4,
num_land_cat = 28,
fractional_seaice = 0,
sf_urban_physics = 1,1,1,1,
sst_update = 0,
/
&dynamics
w_damping = 1,
diff_opt = 1,
km_opt = 4,
diff_6th_opt = 0, 0, 0, 0,
diff_6th_factor = 0.12, 0.12, 0.12, 0.12,
base_temp = 290.,
damp_opt = 0,
zdamp = 5000.,5000.,5000.,5000.,
dampcoef = 0.2, 0.2, 0.2, 0.2,
khdif = 0, 0, 0, 0,
kvdif = 0, 0, 0, 0,
non_hydrostatic = .true.,.true.,.true.,.true.,
moist_adv_opt = 1, 1, 1, 1,
scalar_adv_opt = 1, 1, 1, 1,
epssm = 0.7, 0.7, 0.7, 0.7,
/
&bdy_control
spec_bdy_width = 5,
spec_zone = 1,
relax_zone = 4,
specified = .true.,.false.,.false.,.false.,
nested = .false.,.true.,.true.,.true.,
/
&grib2
/
&namelist_quilt
nio_tasks_per_group = 0,
nio_groups = 1,
/
Bibliography
Alapaty, Kiran et al. (1997). “Simulation of Atmospheric Boundary Layer Processes Us-
ing Local- and Nonlocal-Closure Schemes”. In: Journal of Applied Meteorology 36.3,
pp. 214–233. issn: 0894-8763. doi: 10.1175/1520- 0450(1997)036<0214:SOABLP>
2 . 0 . CO ; 2. url: http : / / journals . ametsoc . org / doi / abs / 10 . 1175 / 1520 -
0450%281997%29036%3C0214%3ASOABLP%3E2.0.CO%3B2.
Arya, S. Pal (1999). Air Pollution Meteorology and Dispersion. New York: Oxford Uni-
versity Press, Inc. isbn: 0-19-507398-3.
Borrego, C. et al. (2003). “Emission and dispersion modelling of Lisbon air quality at
local scale”. In: Atmospheric Environment 37.37, pp. 5197–5205. issn: 1352-2310. doi:
10 . 1016 / J . ATMOSENV . 2003 . 09 . 004. url: https : / / www . sciencedirect . com /
science/article/pii/S1352231003007404.
Bright, David R. and Steven L. Mullen (2002). “The Sensitivity of the Numerical Sim-
ulation of the Southwest Monsoon Boundary Layer to the Choice of PBL Turbu-
lence Parameterization in MM5”. In: Weather and Forecasting 17.1, pp. 99–114. doi:
10.1175/1520-0434(2002)017<0099:TSOTNS>2.0.CO;2. url: http://journals.
ametsoc.org/doi/abs/10.1175/1520- 0434%282002%29017%3C0099%3ATSOTNS%
3E2.0.CO%3B2.
Carvalho, D. et al. (2014a). “Sensitivity of the WRF model wind simulation and wind en-
ergy production estimates to planetary boundary layer parameterizations for onshore
and offshore areas in the Iberian Peninsula”. In: Applied Energy 135, pp. 234–246.
issn: 0306-2619. doi: 10 . 1016 / J . APENERGY . 2014 . 08 . 082. url: https : / / www .
sciencedirect.com/science/article/pii/S0306261914008939.
— (2014b). “WRF wind simulation and wind energy production estimates forced by dif-
ferent reanalyses: Comparison with observed data for Portugal”. In: Applied Energy
117, pp. 116–126. issn: 0306-2619. doi: 10.1016/J.APENERGY.2013.12.001. url:
https://www.sciencedirect.com/science/article/pii/S0306261913009847.
Chen, Fei et al. (2011). “The integrated WRF/urban modelling system: Development,
evaluation, and applications to urban environmental problems”. In: International Jour-
nal of Climatology. issn: 08998418. doi: 10.1002/joc.2158.
96
97 BIBLIOGRAPHY
Chen, F. et al. (2006). “Current status of urban modeling in the community Weather
Research and Forecast (WRF) model”. In: Joint with Sixth Symposium on the Ur-
ban Environment and AMS Forum: Managing our Physical and Natural Resources:
Successes and Challenges. Atlanta, GA, USA: Amer. Meteor. Soc.
ESA (The Europeian Free Trade Assosiation (EFTA) Surveillance Authority) (2015).
Judgment of the court (2. October 2015). Luxembourg. url: http://www.eftacourt.
int/uploads/tx_nvcases/7_15_Judgment.pdf.
Flagg, D D and P A Taylor (2011). “Atmospheric Chemistry and Physics Sensitivity of

mesoscale model urban boundary layer meteorology to the scale of urban representa-
tion”. In: Atmos. Chem. Phys 11, pp. 2951–2972. doi: 10.5194/acp-11-2951-2011.
url: www.atmos-chem-phys.net/11/2951/2011/.
Folkehelseinstituttet (2013). Luftkvalitetskriterier - Virkninger av luftforurensning på helse

(2013:9). Tech. rep. Oslo: Nasjonalt folkehelseinstitutt, Miljødirektoratet.
García-Díez, M et al. (2013). “Seasonal dependence of WRF model biases and sensitivity to
PBL schemes over Europe”. In: Quarterly Journal of the Royal Meteorological Society
Q. J. R. Meteorol. Soc. J. R. Meteorol. Soc 139.139, pp. 501–514. doi: 10.1002/qj.
1976. url: https://rmets.onlinelibrary.wiley.com/doi/pdf/10.1002/qj.1976.
Gjerstad, Karl Idar, Ingrid Sundvor, and Dag Tønnesen (2012). Vurdering av luftkvalitet,
Måledataanalyse og litteraturstudie (Oppdragsrapport for Klima- og forurensningsdi-
rektoratet (Klif )). Tech. rep. Norwegian Institute for Air Research, p. 102. url: http:
//luftkvalitet.info/Libraries/Rapporter/43-2012-kig-klif_1.sflb.ashx.
Golder, D (1972). “Relations among stability parameters in the surface layer”. In: Boundary-
Layer Meteorology 3.1, pp. 47–58. doi: 10.1007/BF0076910.
Grossman-Clarke, Susanne et al. (2010). “Contribution of Land Use Changes to Near-

Surface Air Temperatures during Recent Summer Extreme Heat Events in the Phoenix
Metropolitan Area”. In: American Meteorological Society 49, pp. 1649–1664. doi: 10.
1175/2010JAMC2362.1. url: https://journals.ametsoc.org/doi/pdf/10.1175/
2010JAMC2362.1.
Høiskar, Britt Ann Kåstad, Ingrid Sundvor, and Mathias Vogt (2016). Effekt av lavut-
slippssoner på luftkvaliteten i Oslo Utslipps-og spredningsberegninger. Tech. rep. Oslo:
Norsk institutt for luftforskning(NILU), p. 23. url: https : / / brage . bibsys . no /
xmlui/bitstream/handle/11250/2419841/22-2016+Effekt+av+lavutslippssoner+
p%C3%A5+luftkvaliteten+i+Oslo.pdf?sequence=3.
Hong, Song-You, Yign Noh, and Jimy Dudhia (2006). “A New Vertical Diffusion Package
with an Explicit Treatment of Entrainment Processes”. In: Monthly Weather Review
134.9, pp. 2318–2341. doi: 10.1175/MWR3199.1. url: http://journals.ametsoc.
org/doi/abs/10.1175/MWR3199.1.
BIBLIOGRAPHY 98
Iacono, Michael J. et al. (2008). “Radiative forcing by long-lived greenhouse gases: Cal-
culations with the AER radiative transfer models”. In: Journal of Geophysical Re-
search 113.D13, p. D13103. issn: 0148-0227. doi: 10.1029/2008JD009944. url: http:
//doi.wiley.com/10.1029/2008JD009944.
Jacobsen, Mark Z. (2005). Fundamentals of Atmospheric Modeling. 2nd ed. Cambridge

University Press, pp. 228–271. url: http://www.dca.ufcg.edu.br/mna/jacobson.
pdf.
Janjic, Z. I (1990). “The step-mountain coordinate: physical package”. In: Mon. Wea. Rev.
118, pp. 1429–1443.
Janjic, Z. I. (1996). “The surface layer in the NCEP Eta Model”. In: Eleventh Conference
on Numerical Weather Prediction, Norfolk, VA, 19–23 August. Boston, MA: Amer.
Meteor. Soc., pp. 354–355.
— (2002). “Nonsingular Implementation of the Mellor–Yamada Level 2.5 Scheme in the

NCEP Meso model”. In: NCEP Office Note 437, p. 61.
Kim, Youngseob et al. (2013). “Evaluation of the Weather Research and Forecast/Urban
Model Over Greater Paris”. In: Boundary-Layer Meteorology 149.1, pp. 105–132. issn:
0006-8314. doi: 10.1007/s10546-013-9838-6. url: http://link.springer.com/
10.1007/s10546-013-9838-6.
Krogsæter, O and J Reuder (2013). “Validation of boundary layer parameterization schemes

in the Weather Research and Forecasting model (WRF) under the aspect of offshore
wind energy applications -Part II: Boundary layer height and atmospheric stability
WRF Validation for Offshore Wind Energy Applications -Part II”. In: Wind Energy
00, pp. 1–14. doi: 10.1002/we. url: http://bora.uib.no/bitstream/handle/
1956/17049/WRF_wind_ii_rev.pdf?sequence=3&isAllowed=y.
Kusaka, H. and F. Kimura (2004). “Coupling a single-layer urban canopy model with a
simple atmospheric model: Impact on urban heat island simulation for an idealized
case”. In: J. Meteor. Soc. Japan 82, pp. 67–80.
Kusaka, H., H. Kondo, et al. (2001). “A simple single-layer urban canopy model for at-
mospheric models: Comparison with multi-layer and slab models”. In: Bound.-Layer
Meteor. 101, pp. 329–358.
Lee, S.-H et al. (2011). “Evaluation of urban surface parameterizations in the WRF model
using measurements during the Texas Air Quality Study 2006 field campaign”. In:
Atmos. Chem. Phys 11, pp. 2127–2143. doi: 10 . 5194 / acp - 11 - 2127 - 2011. url:
www.atmos-chem-phys.net/11/2127/2011/.
Markowski, P. and Y. Richardson (2010). Mesoscale Meteorology in Midlatitudes. 1st ed.

John Wiley & Sons, Ltd. isbn: 978-0-470-74213-6.
99 BIBLIOGRAPHY
Martilli, Alberto, Alain Clappier, and Mathias W. Rotach (2002). “An Urban Surface
Exchange Parameterisation for Mesoscale Models”. In: Boundary-Layer Meteorology
104.2, pp. 261–304. issn: 0006-8314. doi: 10.1023/A:1016099921195. url: http:
//link.springer.com/10.1023/A:1016099921195.
Mellor, G. L. and T. Yamada (1982). “Development of a Turbulence Closure Model for

Geophysical Fluid Problems”. In: Rev. Geophys. Space Phys. 20, pp. 851–875. url:
https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/RG020i004p00851.
Miao, Shiguang et al. (2009). “An Observational and Modeling Study of Characteristics
of Urban Heat Island and Boundary Layer Structures in Beijing”. In: American Me-
teorological Society 48, pp. 484–501. doi: 10.1175/2008JAMC1909.1. url: https:
//journals.ametsoc.org/doi/pdf/10.1175/2008JAMC1909.1.
Mohan M and Siddiqui T. (1998). “Analysis of various schemes for the estimation of
atmospheric stability classification.” In: Atmospheric Environment 32.21, pp. 3775–
3781. doi: 10.1016/S1352-2310(98)00109-5.
Oke, Timothy R. (1987). Boundary Layer Climates. 2nd ed. Methuen & Co. Ltd, p. 435.
isbn: 0-415-04319-0.
Pasquill, F (1961). “Estimation of the dispersion of windborne material”. In: Quarterly

journal of the Royal Meteorological Society 90.1063, pp. 33–49.
Rogers, E et al. (2001). Changes to the NCEP Meso Eta Analysis and Forecast System:
Increase in resolution, new cloud microphysics, modified precipitation assimilation,
modified 3DVAR analysis. url: [Available % 20online % 20at % 20http : %20 / /www .
emc . ncep . noaa . gov / mmb / mmbpll / eta12tpb ; %20also % 20available % 20from %
20National % 20Weather % 20Service , %20Office % 20of % 20Meteorology , %201325 %
20East-West%20Highway,%20Silver%20Spring,%20MD%2020910.
Salamanca, Francisco, Andrea Krpo, et al. (2010). “A new building energy model coupled
with an urban canopy parameterization for urban climate simulations—part I. formu-
lation, verification, and sensitivity analysis of the model”. In: Theoretical and Applied
Climatology 99.3-4, pp. 331–344. issn: 0177-798X. doi: 10.1007/s00704-009-0142-
9. url: http://link.springer.com/10.1007/s00704-009-0142-9.
Salamanca, Francisco, Alberto Martilli, et al. (2011). “A Study of the Urban Boundary
Layer Using Different Urban Parameterizations and High-Resolution Urban Canopy
Parameters with WRF”. In: American Meteorological Society 50, pp. 1107–1128. doi:
10.1175/2010JAMC2538.1. url: https://journals.ametsoc.org/doi/pdf/10.
1175/2010JAMC2538.1.
Seinfeld, John P. and Spyros N. Pandis (2016). Atmospheric Chemistry and Physics -
From Air Pollution to Climate Change. 3rd ed. Hoboken, New Jersey: John Wiley &
Sons, Inc. isbn: 9781118947401.
BIBLIOGRAPHY 100
Shin, Hyeyum Hailey and Song-You Hong (2011). “Intercomparison of Planetary Boundary-
Layer Parametrizations in the WRF Model for a Single Day from CASES-99”. In:
Boundary-Layer Meteorol 139, pp. 261–281. doi: 10 . 1007 / s10546 - 010 - 9583 - z.
url: http : / / home . chpc . utah . edu / ~u0818471 / Documents / 2014 / Continuity /
Relevant%20Papers/Intercomparison%20of%20PBLs%20in%20WRF%20from%20CASES-
99%20(Shin%20_%20Hong%202011).pdf.
Skamarock, William C. et al. (2008). A Description of the Advanced Research WRF Ver-
sion 3 (Technical repor). Tech. rep. Boulder, Colorado, USA: Mesoscale and Microscale
Meteorology Division, National Center for Atmospheric Research.
Solberg, S. and T Svendby (2012). Development of a nested WRF/EMEP modelling system

at NILU (EMEP status report 1/2012). Tech. rep. Oslo: Norwegian Meteorological
Institute - MSC-W., pp. 81–89. url: http : / / emep . int / publ / reports / 2012 /
status_report_1_2012.pdf.
Stensrud, David J. and Steven J. Weiss (2002). “Mesoscale Model Ensemble Forecasts
of the 3 May 1999 Tornado Outbreak”. In: Weather and Forecasting 17.3, pp. 526–
543. doi: 10 . 1175 / 1520 - 0434(2002 ) 017<0526 : MMEFOT > 2 . 0 . CO ; 2. url: http :
//journals.ametsoc.org/doi/abs/10.1175/1520-0434%282002%29017%3C0526%
3AMMEFOT%3E2.0.CO%3B2.
Stickland, Matthew et al. (2013). “Measurement and simulation of the flow field around a
triangular lattice meteorological mast”. In: Journal of Energy and Power Engineering
13. url: https://strathprints.strath.ac.uk/43674/.
Stull, Roland B. (1988). An Introduction to Boundary Layer Meteorology. Dordrecht, The

Netherlands: Kluwer Academic Publisher. isbn: 90-277-2769-4.
Tastula, Esa-Matti and Timo Vihma (2011). “WRF Model Experiments on the Antarc-
tic Atmosphere in Winter”. In: American Meteorological Society. doi: 10 . 1175 /
2010MWR3478 . 1. url: https : / / journals . ametsoc . org / doi / pdf / 10 . 1175 /
2010MWR3478.1.
Thorsnæs, Geir. (2018). Oslo. url: https://snl.no/Oslo.
Turner, DB (1970). Workbook of Atmospheric Dispersion Estimates. Ohio, USA: U.S.

Department of Health, Education and Welfare, p. 84.
Tvedt, Knut Are (2016). Ekeberg: fjell i Oslo. url: https://snl.no/Ekeberg_-_fjell_

i_Oslo.
Wallace, John M and Peter V Hobbs (2006). Atmospheric Science - An Introductory

Survey. 2nd ed. Elsevier Inc. isbn: 13:978-0-12-732951-2.
101 BIBLIOGRAPHY
Wang, Wei et al. (2017). WRF(ARW) Version 3 Modeling System User’s Guide. Boulder,
Colorado, USA: Mesoscale and Microscale Meteorology Division, National Center for
Atmospheric Research.
WHO (Europe) (2017). WHO’s commitment to air quality: from the 1950s to today. url:
http://www.euro.who.int/en/health-topics/environment-and-health/air-
quality/news/news/2017/02/whos- commitment- to- air- quality- from- the-
1950s-to-today.
Zhang, Dalin and Richard A. Anthes (1982). “A High-Resolution Model of the Plane-
tary Boundary Layer—Sensitivity Tests and Comparisons with SESAME-79 Data”.
In: Journal of Applied Meteorology 21.11, pp. 1594–1609. doi: 10 . 1175 / 1520 -
0450(1982)021<1594:AHRMOT> 2.0.CO;2. url: http://journals.ametsoc.org/
doi/abs/10.1175/1520-0450%281982%29021%3C1594%3AAHRMOT%3E2.0.CO%3B2.
Zilitinkevich, S. S. (1995). “Non-local turbulent transport: pollution dispersion aspects of

coherent structure of convective flows. Air Pollution III — Volume I. Air Pollution The-
ory and Simulation”. In: Transactions on Ecology and the Environment. Southampton
Boston: Computational Mechanics Publications, pp. 53–60.

Masterthesis Borander Pernille

Uploaded by

Copyright:

Available Formats

Masterthesis Borander Pernille

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Masterthesis Borander Pernille

Uploaded by

Copyright:

Available Formats

Analysis of urban meteorological data

Comparison of local measurements and a numerical

Thesis submitted for the degree of

August 15, 2018

Analysis of urban meteorological data

Printed: Reprosentralen, University of Oslo

Oslo - August 15, 2018

3.5 Comparison of Measurements . . . . . . . . . . . . . . . . . . . . . . . . . 32

5 Results and Discussion 50

6 Summary and Conclusions 78

A Additional Figures and Tables 82

1.1 Motivation and Background

1.2 Outline of this Study

2.1 The Planetary Boundary Layer

2.1.1 Heat Exchange

2.1.2 Momentum Exchange

2.2 Dispersion in PBL

where u = u and v = v is the horizontal mean wind components.

u = u + u0 , v = v + v0, θ = θ + θ0 , etc. (2.6)

2.3 Atmospheric Stability

where r is the mixing ratio. The potential temperature, θ, is defined as

where T is the temperature at a pressure level P , P0 is a reference pressure (often 1013

2.3.2 Dynamic Stability

2.3.3 The Richardson Number

Richardson Type of flow Level of turbulence Level of turbulence

Large, negative Turbulent Large Small

2.3.4 The Obukhov Length

(τ0 /ρ)3/2 u2∗

Bulk Richardson numb. limits Stability classes

RiB < −5.34 Very unstable, A

2.3.5 Pasquill’s Stability Classes

2.3.6 Effect of Stability on Wind Shear

2.4 Special Effects

Figure 2.7: Schematic representation of the urban atmosphere illustrating a two-layer

2.4.1 Urban Effect

2.4.2 Ocean Effect

Rn = Qh + Qe + ∆Qs + ∆Qa (2.18)

2.4.3 Topographic Effect

3.1 Description of Bjørvika

3.1.1 The Munch mast

3.2 Instrumentation of the Munch mast

3.3 Calibration of Measurements

Sensor Correction func.

10m y = −6 · 10−6 T + 0.0623

3.3.2 The Wind Observations

3.4 MET’s Weather Stations

3.4.1 Description of Hovin and Blindern

3.4.2 Instrumentation of the Hovin and Blindern masts

3.5 Comparison of Measurements

3.5.2 The Wind Observations

4.1 The Weather Research and Forecast Model

4.1.1 WRF Modeling System

Figure 4.2: Schematic of the terrain-following vertical η coordinate in WRF, adopted

4.2 Model Setup

Domain Horizontal res. Numb. of grid cells ∆t

d01 13.5×13.5 km 103×100 67 s

4.2.1 Domain Setup

4.2.2 Initialization and Boundary Conditions

4.2.3 Choices of Physic schemes