Investigation of geoid changes in

Northern Europe using latest data

from the dedicated satellite
gravity mission GRACE
Masterprogramm: Transport and Geoinformation Technology
KTH Stockholm

Marvin Mc Cutchan

Acknowledgement
I would like to thank my supervisor Prof. Huaan Fan for his support and constructive
comments on my thesis.
I would also like to thank Ilse, Jeffrey and Oma for their never ending support during
my studies. I could not have made it so far without you.

Abstract
Within this thesis, the temporal changes of the geoid are being analyzed in Sweden, Norway and Finland using geoid models derived from the satellite gravity mission
GRACE. Undulations are computed for the study area as well as for a selected set of
cities. These computations are then analyzed for temporal trends. Finally, the derived
signals are compared with the change of sea level, as well with the geoid computed from
the motion of the pole.
The thesis shows that the change of the geoid undulation does not underly local
variation, but exhibits a regional character. It is observed that the geoid undulation
exhibits a trend. This trend comes with certain periods, as well as an overall increase
over time can be detected.

Contents
1 Introduction

2 Mathematical fundamentals
2.1 Spherical Harmonics . . . . . . . . . . . . . . . . . . . . . . .
2.2 Series expansion of the gravitational potential . . . . . . . . .
2.3 Data from satellite gravity mission GRACE . . . . . . . . . .
2.4 Computation of the geoid models . . . . . . . . . . . . . . . .
2.4.1 GRS80 parameters and spherical harmonics coefficients
2.4.2 Compute disturbing potential . . . . . . . . . . . . . .
2.4.3 Compute normal gravity . . . . . . . . . . . . . . . . .
3 Numerical investigation of geoid changes
3.1 Workflow for computing the geoid undulations . . . . . . . . .
3.2 Analysis of temporal geoid undulation changes . . . . . . . . .
3.2.1 Workflow for processing the time series . . . . . . . . .
3.2.2 Time series of geoid undulation in Gothenburg . . . . .
3.2.3 Annual changes of the geoid undulation in Gothenburg
3.2.4 Amplitude spectra . . . . . . . . . . . . . . . . . . . .
3.3 Compare changes of geoid undulation with sea level changes .
3.4 Comparison with geoid changes derived from the polar motion
3.5 Comparison of all selected cities . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

6
7
8
11
12
12
12
13

.
.
.
.
.
.
.
.
.

14
18
22
22
23
27
29
32
36
39

4 Conclusions and Discussion

41

5 Future work

42

Introduction

The geoid is an equipotential surface which is needed to define heights, such as orthometric
heights, and to describe the shape of the earth. It is a physical model of the earth and
corresponds to the mean sea level, by definition [1]. It plays an essential role in modern
geodesy and therefore is required to exhibit an accuracy at a centimeter, or even millimeter
level. It is therefore expected that changes on the geoid have a significant impact on such
requirements. This thesis deals with computation and analysis of these temporal changes
which occur within three scandinavic countries, namely: Sweden, Norway and Finland.
The analysis includes: (1) computation of geoids for the whole study area (2)computation
of time series of the geoid undulation for selected places, (3) a frequency analysis of these
time series, (4) a comparison with sea level changes and the polar motion.
The geoid is created by two forces; the gravitational- and rotational potential of the
earth. The behavior of the rotational potential of the earth is well known whereas the
gravitational potential is more uncertain. Consequently, effort has been made to study
the earths gravitational potential field. The study of the gravity potential field not only
is important to gain knowledge about the geoid and to create more accurate geoid models, but also to determine satellites orbits more accurately or help to understand tectonic
movements. However, a variety of possibilities exist which allow to observe the gravitational potential, ranging from the use of an absolute or relative gravimeter to satellite
based gravimetry. Using a gravimeter only gives measurements for a specific and small
area, whereas satellite based gravimetry allows to measure on a global scale. Prominent satellite gravity missions are GRACE (Gravity Recovery and Climate Experiment),
CHAMP (CHAllenging Minisatellite Payload) and GOCE (Gravity Field and SteadyState Ocean Circulation Explorer) which all allow to measure velocity and position of
the corresponding satellite and consequently the gravitational potential field of the earth.
Another satellite based methodology which allows to compute the geoid undulation, i.e.
the height-difference between reference ellipsoid and geoid, is satellite altimetry. It measures the distance to the sea level and therefore enables to calculate the geoid undulation.
This thesis uses data from the GRACE to compute geoids.
The satellite gravity mission GRACE has been realized by a collaboration of NASA(National
Aeronautics and Space Administration) and the DLR (Deutschen Zentrum f
ur Luft- und
Raumfahrt). A distinct characteristic of GRACE is its set-up, which is based on two
satellites flying in the same orbit (see figure 1).

Figure 1: Set-up for the GRACE satellite mission: Two satellites collect data of the
gravitational field of the Earth. Source: NASA (eospso.nasa.gov)
The GRACE satellite twins constantly measure their distance to each other using
[2]. This way of measuring
microwaves (K-Band) with an accuracy better than 0.1 m
s
allows to detect anomalies in the earths gravitational field, and thus the modeling of the
geoid.
According to [1], following parameter for the orbit of the GRACE satellites are given:
eccentricity e < 0.005
inclination i = 89
initial altitude between 485km and 500km
both satellites are separated by a distance of approximately 220km.

The thesis is structured in following way: Chapter 2 introduces to the mathematical

foundations, which are paramount in order to understand and compute the gravitational
potential field. Chapter 3 outlines the investigation and computation of the geoid height
changes. Chapter 4 gives a conclusion.

Mathematical fundamentals

In this section the mathematical foundations for describing the geopotential field of the
earth are delineated. It will be shown that (1)finding a closed equation for the gravitational field at a point requires to solve Laplaces equation,(2) that the solution to
Laplaces equation is a series of spherical harmonics, (3) how the geopotential field can be
expanded in a series of spherical harmonics and (4) how the geoid undulation is calculated.
Derivations in this chapter are based on [1] and [3].
Consider two masses, e.g. the earth and a satellite, one of them (PE ) significantly bigger than the other one (PS ). Newtons law describes the attraction of the smaller mass as:
F =G

M
l2

(2.0.1)

Where l is the euclidean distance between the two masses, M the weight of the heavier mass and G Newtons gravitational constant. Knowing the coordinates of both masses
PE (x, y, z) and PS (x0 , y 0 , z 0 ), the gravitational force can be split up into three components.

x0 x
GM

F~ = 3 y 0 y
l
z0 z

(2.0.2)

In order to avoid a vector representation one can search for a scalar function V which
satisfies following condition:
F~ = grad(V )
(2.0.3)
lim V = 0
l

Following potential function meets this requirements:

V =

GM
l

(2.0.4)

However, this potential function describes (PE ) as a point mass. In order to describe the
attraction caused by a volume v having a density , 2.0.4 yields to:
Z Z Z
V =
x

Gp

Z Z Z

(x, y, z)
(x x0 )2 + (y y 0 )2 + (z z 0 )2

dxdydz = G
v

dv
l

(2.0.5)

The first derivatives of V from equation 2.0.5 are the components of the force in X-,Yand Z direction. These components are continuous. In case of the second derivatives, it
is known that a discontinuity is present which can appear at a change of density . The
potential function V satisfies the Poisson equation for the interior and exterior space of
6

the attracting mass according to:

V = 4G . . . interior space

(2.0.6)

V = 0 . . . exterior space
is the Laplace operator, which is defined as:
(V ) =

2V
2V
2V
+
+
X 2 Y 2 Z 2

(2.0.7)

V = 0, known as Laplaces equation, is a differential equation. Based on this equation

and the requirement of a function describing the potential field, a boundary value problem
is present. This boundary value problem can be solved using spherical harmonics.

2.1

Spherical Harmonics

Laplaces equation is required to be solved. Firstly Laplaces equation shall be trans using following
formed from cartesian coordinates(x, y, z) to spherical coordinates(r, , )
relationship:
p
x2 + y 2 + z 2
p
x2 + y 2
1
= tan (
)
z
y
= tan1 ( )
x

r=

(2.1.1)

Which yields Laplaces equation in spherical coordinates:

2
2
1
2V
= r2 V + 2r V + V + cot()
V +
V (r, , )

2 = 0
r2
r
2
sin2 ()

(2.1.2)

This differential equation can be solved by separating its variables. Consequently, three
different functions shall describe Laplaces equation, in which every function depends on
one variable:
= f ()g()h(r)

V (r, , )

(2.1.3)

Every function is derived from a separate homogeneous differential equation of second

order. Their solutions can be seen bellow:

h(r) =

rn+1
f () = cos(m) and sin(m)
= Pnm (cos())

g()

(2.1.4)

Where Pnm is called a Legendre function. Legendre functions form an orthogonal set
are
of functions. Their computation will be described in Chapter 2.2. f () and g()
Since Laplaces
summarized in one function named surface harmonic function Y (, ).
equation is a linear differential equation, any linear combination of its solutions also forms
a solution.
=
V (r, , )

X
n=0

h(r)

n
X

Y (, )

m=0

X
1 X

anm Pnm (cos())cos(m)

+ bnm Pnm (cos())sin(m)
V (r, , ) =
n+1
r
m=0
n=0

(2.1.5)

2.1.5 states a general solution to Laplaces equation. The coefficients of this linear combination allow to create a series expansion, for example of the gravitational potential field
of the earth.

2.2

Series expansion of the gravitational potential

The general solution to Laplaces equation is given by 2.1.5. Therefore it is possible to

express equation 2.0.5 in form of a series of harmonic functions. Changing 2.0.5 from
cartesian to spherical coordinates one gets:
) = G
V (r, ,

Z Z Z
v

(r0 , 0 , 0 )
dv(r0 , 0 , 0 )
l

(2.2.1)

) is attracted by a mass, such as the earth (see figure 2) in which

A point PS (r, ,
dv(r0 , 0 , 0 ) is a volume element.
Changing from cartesian to spherical coordinates and using the cosine rule one gets
for l:
1
1
=p
l
r2 + r02 2rr0 cos()

(2.2.2)

Figure 2: A point PS it attracted by a body(e.g. earth). dv is a volume element.

Where can be computed using the cosine rule from spherical trigonometry:

cos() = cos()cos(
0 ) + sin()sin(
0 )cos( 0 )

(2.2.3)

Using the Leibniz rule (see equation 2.2.4), one can expand 2.2.2 in a series of Legendre
polynomials. In order to perform a series expansion of 2.2.2 two substitutions have to be
made.

X
1

=
Pn (x)tn
(2.2.4)
r 1 2xt + t n=0
r0
r
x = cos()
t=

(2.2.5)

1 X r0 n
q
=
( ) Pn (cos())
r
r
r0
r0
r 1 2cos() r + ( r )
n=0
1

The Legendre polynomial in 2.2.6 can be substituted with normalized spherical harmonics
Ynm (see equation 2.2.7. Normalized spherical harmonics form a orthonormal system of
functions:

1
< Yn0 m0 , Ynm >=
4

1, if n = n0 and m = m0 ,

Ynm Ynm d =
0, if n =

6 n0 and/or m 6= m0 .

Z Z

(2.2.6)

The substitution can be made according to the decomposition formula:

n
X
1
)Ynm (0 , 0 ))
Pn (cos()) =
(Ynm (,
2n + 1 m=n

(2.2.7)

Inserting 2.2.7 into 2.2.6 yields:

X
1 X

(
anm cos(m) + bnm sin(m))Pnm (sin())
n+1
r
Z Zm=0
Z
0 0

(r0 )n cos(m0 )Pnm (cos())p(r

, , 0 )(r0 , 0 , 0 )dv
a
nm =
Z Z Zv
0 0

bnm =
, , 0 )(r0 , 0 , 0 )dv
(r0 )n cos(m0 )Pnm (sin())p(r

) =
V (r, ,

(2.2.8)

Both coefficients bnm and a

nm can grow large due to the fact that they have a dimension.
In practice, this is being avoided by making them unitless:
Anm =
nm =
B

1
an GM

a
nm

1
bnm
n
a GM

(2.2.9)

Pnm (t)are called normalized Legendre Functions. They are computed based on [4]:
Pnm (t) = bnm Pn2,m (t) + bnm tPn1,m (t) n 3; 1 m n 2

Pn,n1 (t) = 2n + 1tPn1,n1 (t) n 1; m = n 1

r
2n + 1
Pnn (t) =
1 t2 Pn1,n1 (t) n 2; m = n
2n

2n
+
1
n

1
2n + 1

Pn0 (t) =
Pn2,0 +
2n 1tPn1,0 n 2; m = 0
n
n
2n 3
(2.2.10)
the coefficients anm and bnm are calculated
s

(2n 1)(2n + 1)
(n m)(n + m)

(2n + 1)(n + m 1)(n m 1)

(n m)(n + m)(2n 3)

anm =
bnm =

(2.2.11)

10

This iterative approach is numerical stable compared with other approaches for computing Legendre functions [4]. However, using this approach requires to have the first 3
Legendre functions computed a priori:

P0,0 = 1 P1,0 =

3t P1,1 =

3 1 t2

(2.2.12)

in equation 2.2.10.
It has to be noted that t is substituted with sin()
Adding the first degree component separately to 2.2.8 and using the coefficients defined in 2.2.9 leads to:
n

GM X a n+1 X

nm sin(m))Pnm (sin())
(Anm cos(m) + B
V (r, , ) =
+
( )
r
r
m=0
n=2

(2.2.13)

Equation 2.2.13 is the expansion of the gravitational potential field using spherical harmonics. It is a closed equation which can be used to calculate the geopotential for a point
) (a is the equatorial radius of the earth). In practice, the spherical harmonics
PS (r, ,
nm are unknown and have to be determined. This can be done by
coefficients Anm and B
using satellite gravity missions, such as GRACE. Having the spherical coefficients at hand,
it is possible to calculate the geoid undulation N . The computation of N is described in
chapter 2.4.3. Chapter 2.3 describes the data type provided by the GRACE mission.

2.3

Data from satellite gravity mission GRACE

There are several data types which are provided from the GRACE project which allow
the modelling of the geoid [2]. The data type chosen for this thesis contains following
parameters which are needed to describe the geoid using spherical harmonics:
GM , i.e. the gravitational constant multiplied with the mass of the earth
r, i.e. the Radius of the earth
Anm and Bnm which are the coefficients of the spherical harmonics
Based on these provided parameters, 2.2.13 can be computed.
It is important to note that a temporal resolution of 1 month is provided. Each provided
file (a txt file) describes the geoid for a specific month. According to [5], original GRACE
data is subject to a correlation among the spherical harmonics. This correlation appears
in form of stripes (from north to south) when performing a temporal analysis of the data.
In order to minimize this influence the data has been smoothed a priori, applying [6].
135 txt files are downloaded, describing the geoid for 135 months. This files contain geoid
11

models starting from April 2002 till September 2014, in an interval of one month. The data
for 15 months is not provided by GRACE. In order to compensate for this missing data, a
linear interpolation is performed, this is described in chapter 3 Numerical investigation
of geoid changes.

2.4

Computation of the geoid models

In this section the workflow of the computation for the geoid undulation for a location/area
is described. Therefore following steps are performed(1-6): (1) Load GRS80 parameters
and spherical harmonics coefficients, (2) Compute spherical coordinates, (3) Compute
normalized Legendre functions, (4) Compute disturbing potential T , (5) Compute normal
gravity and (6) compute the geoid undulation.

2.4.1

GRS80 parameters and spherical harmonics coefficients

The data derived from GRACE is based on the GRS80 (Geodetic Reference System 80)
which defines parameter for a global reference ellipsoid [7] (see table 1).
Name
Geocentric gravitational constant GM
Semi-major axis a
Dynamic form factor J2
Angular velocity of rotation

Value
3
3.98605x1014 [ ms2 ]
6378137[m]
1.08263x103 []
7.292115x105 rad
s

Table 1: GRS80 parameter

Besides the GRS80 parameter, the spherical harmonics coefficients from the GRACE
data are loaded. This coefficients are called Anm and Bnm . The coefficients are organized
within a hashmap in Matlab [8] for each month. Each month is represented by an object [9]
which allows to access the data in an object-based manner.

2.4.2

Compute disturbing potential

The disturbing potential T expresses the difference between the spherical harmonics for
the gravitational potential V 0 of the the ellipsoid and the spherical harmonics for the
gravitational potential V of the geoid [1]:

) = V V 0
T (r, ,

12

(2.4.1)

Where V 0 is:
) =
V 0 (r, ,

X
GM
a

(1 +
( )2n J2n P2n (sin()))
r
r
n=1

J2n is given with :

(2.4.2)

2n
4n + 1
5n
n 3e

(1 n + 2 J2 )
J2n = (1)
(2n + 1)(2n + 3)
e
Subtracting both potentials from each other leads to a subtraction of the corresponding
coefficients:
n
GM X

(Cnm cos(m) + Snm sin(m))Pnm (sin)

T (r, , ) =
a m=0

Cnm = Anm Jnm

nm
Snm = B

(2.4.3)

nm are obtained from the GRACE data.

The coefficients Anm and B

2.4.3

Compute normal gravity

The normal gravity is computed by:

1 + sin2 ()
= e p
1 e2 sin2 ()
bp ae
=
aa

(2.4.4)

In this step, the geoid undulation is obtained by dividing the disturbing potential by
the normal gravity :

) = T (r, , )
(2.4.5)
N (r, ,

The geoid undulation is the final outcome of this procedure and describes the height
difference between the reference ellipsoid and the geoid for any location specified by a
geodetic position.

13

Numerical investigation of geoid changes

The analysis of the change of the geoid undulation is carried out within two domains:
(1) a spatial domain (2) a frequency domain. (1) expresses changes of the geoid undulation for any location in the area of interest in centimeters. (2) expresses the change of
the geoid undulation within the frequency domain. After that, the change of the geoid
undulation is compared with the change of the sea level. Then, the change of the polar
motion is compared to the change of the geoid undulation. In the end of this chapter,
an investigation of the change of the geoid undulation on a regional scale is carried out,
searching for spatial patterns in the study area.
Based on the theory provided in the previous chapter, geoid undulations are computed
using equation 2.4.5 for:
A GRID covering the entire study area
A selected set of cities within the study area

Figure 3: Study area for computing a GRID.

In figure 3, the study area can be examined. It is covered by a bounding box defined
by the coordinates(, ): (55 ,5 ) and (72 ,31 ). Each grid-cell has a size of 34x34[arcsec].
Therefore the geoid undulation is computed for each of this cells in an iterative manner
using equation 2.4.5 .Thus, the product of this process illustrates the different geoid
14

undulations for the whole study area, for each grid-cell. It has to be noted that this
procedure is based on the data from GRACE for just one month. Geoid undulations are
calculated for July 2003 and January 2008, respectively (see figure 4 and 5).

Figure 4: Geoid for northern Europe in July 2003.

Inspecting the geoid for the entire study area (see figures 4 and 5) visually, no significant changes can be detected. A transition, going from east to west, of a lower to a
higher geoid undulation can be observed. The geoid undulation changes more sudden on
land as on sea, where the transition seems to be more smooth.

15

Figure 5: Geoid for northern Europe in January 2008.

Using a set of selected cities withing the study area, the task is to identify the geoid
undulation for each given month per city. This will result in a time series of geoid
undulations for each selected city. The cities are selected in such a manner that all
capital cities are included, and, that they are well distributed over the region of the study
area in order to be able to detected potential local variations. This time series are than
used for further analysis. Following cites are selected:
Byrvik, Norway
Karasjok, Norway
Oslo, Norway
Stockholm, Sweden
Gothenburg, Sweden
Skellefte
a, Sweden
Helsinki, Finland
16

 Kuusamo, Finland
In figure 6 the location of the selected cities can be seen.

Figure 6: Selected cities for computing a time series. For each city the geoid undulation
for each given month is computed.
The data of 15 months is not provided by the GRACE dataset (see table 2).
June, 2002
July, 2002
June, 2003
January, 2004
January, 2011
June, 2011
November, 2011
April, 2012
May, 2012
October, 2012
March, 2013
August, 2013
September, 2013
February, 2014
July, 2014
Table 2: Missing months in GRACE data.

17

These months have been interpolated linearly. Thus, if k months (k 1) are missing
between any two months M onth1 and M onth2 , the geoid undulation N for the ith
month is interpolated linearly according to:

Ni = N1 +

i
X
N2 N1
n=1

k+1

N1 . . . geoid undulation of M onth1

(3.0.6)

N2 . . . geoid undulation of M onth2

{i Z, k Z, 1 i k}
The accuracy of the computed geoid undulation depends on different factors. The
inclination of the satellites influences the accuracy of zonal, tesseral and sectorial harmonics in different ways [10]. The accuracy of the geoid undulation varies depending on
the geographical latitude [11], but however, lays at millimeter level [10]. This accuracy is
sufficient for the analysis carried out within this thesis.

3.1

Workflow for computing the geoid undulations

The workflow for computing geoid undulations is realized by programing a software in

Matlab. The program is structured procedural. Its outline can be seen in figure 7. First,
the program is being passed either a specific location or a defined area for which the geoid
undulations are supposed to be computed. The study area is defined by the coordinates of
the borders of the bounding box and a cell size resolution. Afterwards, a data structure
for storing all geoid undulations is created. The data structure is a three dimensional
array. Its aim is to save the computed geoid undulations for each GRID cell for a specific
month. Thus, each element of the array corresponds to a specific cell in the GRID of the
study area for a specific time. Then, the folder which contains all provided text files of
the GRACE mission is being accessed. The program than proceeds by computing the
geoid undulations based on every test files and the desired study area. The results are
then saved in the three dimensional array. This procedure iterates until no more text files
can be found.
For computing the geoid undulations, a separate function is invoked. Its outline can
be seen in figure 8. The function starts by creating an object.
This object is an instance of a class which represents the geoid for a specific month.
It consist of five attributes which store the file name of the GRACE txt file, the geoid
18

Figure 7: The overall workflow for reading in GRACE data and computing the geoid
undulations.

Figure 8: The workflow of the function for computing the geoid undulation.
undulations for this month, the name of the month, the year and the spherical harmonics
(see figure 9). The object also has two methods: A constructor which is used to initialize

19

the object, and, a method which enables to access the geoid undulations once they are
computed. First, the object reads the text file line by line extracting information out of
its header and then the main part which includes the degree, order and corresponding
value of the spherical harmonics.

Figure 9: The Geoid month class. It stores the information of a GRACE textfiles in a
structured way.

In order to increase the performance of the program, the spherical harmonics are
stored in a Hash Map. A Hash Map is a specific data structure which enables to access its
elements in a quick manner. For this purpose a so called hash function has to be defined
which computes the index of the desired element (see figure 10). The hash function takes
the degree and order of the spherical harmonics and computes the index in the Hash Map.
The definition of the hash function Hash f unction used in this program can be seen in
equation 3.1.1.

Hash f unction(N, M ) =

N
X

n+M +1

(3.1.1)

n=0

Figure 10: The Geoid month class. It stores the information of a GRACE text files in a
structured way.

20

Then the function calculates the required Legendre functions and initializes a GRID,
represented by a two dimensional array. An iteration goes through every location the geoid
undulation shall be computed for, transforms it from ellipsoidal coordinates to spherical
coordinates, computes the geoid undulation and adds it to the corresponding GRID cell
in the GRID. Once all locations are processed, the function returns the GRID, which
contains all geoid undulations for a specific month.

21

3.2

Analysis of temporal geoid undulation changes

Based on the time series created for each selected city, an analysis of the local geoid
undulation change is being performed. It includes an analysis of the time series, the
annual behaviour of the time series and a trend analysis. The analysis in this subchapter
is based on the time series for a single city, namely Gothenburg (Sweden). Gothenburg
was chosen because of several reasons. It is decided that a city in Sweden should be
chosen. Since there was no data available for the sea level change in Stockholm, it is
decided to choose Gothenburg, because it is the second biggest city of Sweden, in terms
of population. However, subchapter 3.5 contains a comparison of the data which was
computed for all selected cities.

3.2.1

Workflow for processing the time series

For processing the geoid undulation changes, a program in Matlab is realized. Its workflow
can be seen in figure 11. It takes the computed geoid undulations for one specific location
as an input argument. Using the information which geoid undulation corresponds to
which date, the program knows when data for a month is missing. Therefore a separate
interpolation function is defined. Based on the newly interpolated values, the monthly
variation is being calculated, i.e., the geoid undulation for a month minus the average
value of all geoid undulations for this specific location. Labels are required for plotting
the data. These labels are set up by strings which consist of the month and year for the
corresponding geoid undulation. For interpolated geoid undulations, a separate function
is invoked which creates the missing string. After that, another function performs a least
square adjustment using the geoid undulations and the number of the month (starting
at 0) as input arguments. The input is used for creating a design matrix which than is
used to compute the coefficients of an adjusted line as well as the standard error of the
coefficients. In the end, the program plots the data, including the adjusted lines.

It has to be noted, that the program is invoked for every city separately. Therefore
it reads in the location of the city, in ellipsoidal coordinates, and saves the final plots as
images in a folder system on the computer.

22

Figure 11: The workflow for processing and plotting the monthly time series of the geoid
undulations which is realized in Matlab.
3.2.2

Time series of geoid undulation in Gothenburg

The geoid undulation in Gothenburg for every month is visualized in figure 12. Additional
lines are added to the plot which indicate a global trend in the increase or decrease
of the geoid undulation. These lines are adjusted to the time series in a least square
manner. Their functional model is expressed by f (x) = ax + b. The errors of the adjusted
parameters a and b are computed as well.

Looking at figure 12, it can be seen that from year 2002 to 2006 the geoid undulation
raises stronger than from 2006 to 2014. In the time period around 2012 the highest values,
but also the highest fluctuations of the geoid undulation are exhibited. The errors of the
adjusted coefficients can be seen in the boxes in figure 12. The upper box corresponds to
the left line and the lower box corresponds to the right line.
In order to reduce the time series to its variation, the overall mean is computed and
subtracted from the geoid undulation of every month. Thus, every month represents
the difference between the corresponding geoid undulation and the mean. The geoid
undulation variation can be seen in figure 13. The variation is similar to the normal time
23

series. Months of extrema and time periods of high fluctuation are the same. However, one
special characteristic can be identified: There seems to be an almost harmonic behavior:
An overall decrease from year 2002 to 2004, than an increase to 2006, after that an
decrease till 2010, from 2010 to 2012 an increase and finally, from 2012 to 2014, the geoid
undulation decreases. It seems a local maximum or minimum every 3-4 months is reached.
The highest geoid undulations can be observed in the interval from 2011 to 2013.

24

25

Figure 12: Geoid undulation over time in Gothenburg. The red lines indicate the temporal trends.

26

Figure 13: Variation of the geoid undulation over time in Gothenburg. The variation is computed by substracting the average from the
signal.

3.2.3

Annual changes of the geoid undulation in Gothenburg

Annual changes of the geoid undulation are computed. Therefore the geoid undulation of
the first month of a year(N1 ) and the last month (N12 ) of a year is taken, subtracted and
mm
]:
divided by 12, this gives a overall change per year in terms of a velocity [ month

Change per year =

N12 N1 mm
[
]
12
month

(3.2.1)

For processing the annual change of the geoid undulation, a Matlab program is realized.
Its outline can be seen in figure 14. It takes the geoid undulations for a specific location
as an input and interpolates missing months. It than computes the velocity. Therefore a
separate function uses the interpolated geoid undulation values. Afterwards the labels for
the missing months are calculated and the data is plotted and saved in a folder system.

Figure 14: Workflow for processing the annual change of the geoid undulation.

27

The plot of the annual change for Gothenburg can be seen in figure 15.

Figure 15: Changes of the geoid undulation within 12 months in Gothenburg.

Considering the annual changes of the geoid undulation no local difference can be
detected. Peaks are observed in 2002 (local minimum), 2004 (local maximum), 2008
(local minimum), 2011 (local maximum) and 2012 (local minimum). The magnitude of
change seems to gain over the years.

28

3.2.4

Amplitude spectra

Computing an amplitude spectrum allows to detect periodicities in a signal. In the context

of this thesis, the computation of an amplitude spectrum can be utilized for detecting
potential periods in the change of the geoid undulation.
For computing and visualizing the spectrum of the amplitude An , a Fourier transformation is utilized according to [3]:
N
1 X
a0 =
f (i)
N i=1
N
2n
N
2 X
f (i)cos(
)i (n = 1, 2, . . . , 1)
an =
N i=1
N
2
N
2 X
N
2n
bn =
)i (n = 1, 2, . . . , 1)
f (i)sin(
N i=1
N
2

aN =
2

(3.2.2)

N
2 X
f (i)cos(i)
N i=1

bN = 0
2
p
An = a2n + b2n

Additionally a rearranged amplitude spectrum according to [3] is computed:

Tn = N/n
Tn . . . period(month)
(3.2.3)

N . . . number of samples (150)

n . . . index of fourier coeffiecient n = 2, . . . ,

N
2

Tn in equation 3.2.3 is a new index: An is used for the y-axis on position Tn on the
x-axis in the plot. Using Tn gives an indication which time period, in terms of months,
is represented by which amplitude. Accordingly, a time period having a high amplitude
corresponds to a time period which comes along with a high change in the geoid undulation.
The workflow for this frequency analysis is realized in Matlab (see figure 16). Missing
months of the original data of the geoid undulation are first interpolated. The interpolated
data is then subject to a Fourier transformation. The Fourier transformation is realized
29

using a separate function. The function computes the Fourier coefficients in an iterative
manner. These coefficients are then rearranged, resulting in the rearranged amplitude
spectrum (see equation 3.2.3). The spectra are than plotted and saved in a folder system
on the computer.

Figure 16: The workflow of the computation of the amplitude spectra.

The amplitude spectrum of the geoid undulation of Gothenburg as well as the rearranged amplitude spectrum can be seen in figures 17 and 18.

30

Figure 17: The amplitude spectrum of the change of the geoid undulation in Gothenburg.

Figure 18: The rearranged amplitude spectrum of the change of the geoid undulation in
Gothenburg
The amplitude spectrum in figure 17 shows higher coefficients for lower frequencies,
indicating that the geoid undulation changes higher over a longer timer rather than within
short periods. This is confirmed by the rearranged amplitude spectrum in figure 18, it
can be observed that the greatest variability comes with an interval of half a year, a
31

year, 49 months, and the biggest amplitude for a period of 74 months. It has to be
noted that the plot showing the annual changes (see figure 15) and figure 18 confirm
the similiar information, i.e., that the geoid undulation has significant changes within 49
months (approximately 4 years) which is likewise to the change from 2008 to 2012 (see
figure 15), and, a significant change of a period of 74 months (approximately 6 years)
which is likewise to the change from 2002 to 2008.

3.3

Compare changes of geoid undulation with sea level changes

The geoid is defined by the equipotential surface, which corresponds to the mean ocean [1].
In order to investigate similarities between the sea level change and the change of the geoid
undulation per month, data from a sea level measure station, located in Gothenburg, is
gathered [12]. Changes of the geoid undulations for Gothenburg are multiplied by 35 to
enable a visual inspection.
Two averages are computed for the sea level change, a monthly and an annual average.
The averages are ploted together with the corresponding change of the geoid undulation
and their correlation C(n) according to a cross correlation:

C(n) =

N (m) S(m + n)
(3.3.1)

m=

{C R, 1 C 1}
In equation 3.3.1, N expresses the geoid undulation change and S to the sea level
change. n is the lag which indicates the relative shift of both signals to each other for a
certain correlation.

32

Figure 19: The workflow of the computation of the amplitude spectra.

A Matlab program realizes the workflow (see figure 19) for comparing the sea level
change and the change of the geoid undulation. First, the program initializes two data
structures . The program then reads in a txt file which contains the sea level recordings.
Since the temporal resolution of the sea level recording is one hour, an average of the
sea level for a month is computed. Therefore, the recordings for one month are stored
in one data structure. As soon as the end of one month is reached and the recordings
of a new month are reached, the average of the previous months sea level recordings is
calculated and stored in the second data structure. Afterwards the first data structure is
being emptied and will store the sea level recordings of the next month. This procedure
continues until the end of the sea level recording file is reached. The data structure which
contains the average values of the sea level is then subject to a correlation calculation
with the geoid undulations. For the computation of the correlation, a function which is
built-in in Matlab is used. The computed data is than plotted and stored in a folder
system of the computer. This procedure is run twice. One time computing a monthly
average of the sea level and one time an annual average.
Figure 20 illustrates the sea level change and the change of the geoid undulation in
Gothenburg as well as their correlation for every month. It should be noted that the geoid
undulations are mutliplied by 35 in order enable a visual inspection.
33

Figure 20: Sea level changes in contrast to the change of the geoid unduation and their
correlations.
Figure 21 illustrates the annual sea level change and the change of the geoid undulation
in Gothenburg as well as their correlation for every year. The sea level change is averaged
for every year.

Figure 21: Sea level changes in contrast to the change of the geoid unduation and their
correlations.

34

The monthly sea level change are autocorrelated with the geoid undulation change,
but never reaches a significant level of correlation (see figure 20). The annual changes of
the sea level and the geoid undulation change (see figure 21) are maximal correlated with
a value of -0.6 with a delay of 1 year (lag of 13 years). Visually inspecting the plots, no
similarities among the sea level change and the change of the geoid undulation can be
detected.

35

3.4

Comparison with geoid changes derived from the polar motion

In this section the comparison of the previously computed change of the geoid undulation,
which is based on the gravitational potential, with the geoid undulation change computed
from the polar motion is delineated. This allows to investigate if both models confirm the
same information.
The rotation axis of the earth is not fixed. It moves in an approximately circular
behavior around a reference pole,the CIO (Conventional International Origin), which is
defined as the average pole over the years 1900-1905 [3]. The IERS (International Earth
Rotation Service) determines the offset to the CIO in spherical distances x and y. The
change of the relative position of the pole can be seen in figure 22.

Figure
22:
The
pole
offset
from
http://www.iers.org/IERS/EN/DataProducts)

2000

to

2014

(Source:

The shape of the geoid depends on the centrifugal potential , which changes its
magnitude depending on the latitude [3]:
1

= 2 r2 cos2 ()
2
r. . . radius of the earth

(3.4.1)

. . . angular velocity of earth rotation

The change of the pole over time causes a change in the centrifugal potential:
1

= 2 r2 cos2 (2)
2
36

(3.4.2)

 is the longitude of a location where shall be computed. xp and yp are the offset of
the pole at a certain time relative to the CIO. Based on equation 3.4.2, a change of the
geoid undulation can be described as:
N =

1 2 2 2
r cos (2)
2g

(3.4.3)

= (xp cos() yp sin())

Earth rotation data has been downloaded from the IERS and is used for computing N .
The offset of the pole is given in arcsec[] which can be directly used for equation 3.4.3.
Computing N for every month in which GRACE also provides data, allows to compare
both changes of the geoid undulation.
A Matlab program realizes the workflow (see figure 23) for computing the correlation
between the polar motion and the change of the geoid undulation. It creates two data
structures. One for storing the x- and y-offset values of the rotation axis for a month
and one for storing the geoid undulation change computed from the polar motion for the
corresponding months. The program than reads in the txt file which contains the data of
the x- and y- offsets of the rotation axis of the earth. The offset is recorded in an interval
of one day and thus must be averaged for each month. The program then calculates the
change of the geoid undulation based on the average x- and y- offset of the rotation axis
for every month and stores it in the designated data structure. Afterwards, the program
performs a calculation of the correlation which uses a built-in Matlab function. Finally
the data get plotted and stored in a folder system of the computer.

Figure 24 show change of the geoid undulation at Gothenburg, derived from GRACE
and the motion of the pole, as well as the correlation of both signals. The correlation is
calculated based on 3.3.1
The geoid undulation, computed from GRACE and the polar motion (see figure 24)
exhibit very little correlation. Looking arround month 50 in figure 24), the reader can
observe a fluctuation in the signal derived from the polar motion. This fluctuation is
caused by the Chandler Wobble [13].The Chandler Wobble introduces a change of the
polar motion in a period of approximately 7 years. This periodicity cannot be detected
in the change of the geoid undulation derived from GRACE data.

37

Figure 23: The workflow of the computation of the amplitude spectra.

Figure 24: Change of the geoid undulation, derived from the motion of the pole and
GRACE. The plot on the bottom shows their correlation.

38

3.5

Comparison of all selected cities

This section delineates the findings among the plots which are created for all selected
cities. The plots can be found in the appendix and are organized according to the city.
Table 3 summarizes the standard errors of the adjusted lines which can be found
in the plots describing the time series of the geoid undulation for every month (see Appendix). Looking at this table, several things can be observed about the regional behavior
of the geoid undulations. The strongest variation appears to be in the south in the region around Gothenburg and Oslo. The geoid undulation in Oslo exhibits the strongest
variation, Gothenburg the second strongest biggest variation. However, the magnitude of
the standard errors seems to be positively correlated with the magnitude of the absolute
value of the geoid undulation. Thus, locations with a higher geoid undulation tend to have
a higher variation in their geoid undulation. It can be also observed that the standard
errors for the slope increase, whereas the standard errors for the constant decrease.
City name
Gothenburg
Oslo
Stockholm
Helsinki
Boyrvik
Skelleftea
Kuusamo
Karasjok

a1 [m]
0.12
0.13
0.08
0.06
0.12
0.08
0.06
0.09

b1 [m]
12.48
13.64
8.12
6.40
12.22
7.86
6.10
8.84

a2 [m]
0.38
0.42
0.25
0.20
0.38
0.24
0.19
0.27

a2 [m]
10.80
11.81
7.03
5.54
10.58
6.80
5.28
7.65

Table 3: Standard errors of the adjusted lines for all selected cities.The errors with index
1 correspond to the first (left) adjusted line in the plots, errors with index 2 to the second
adjusted line (right) in the plots.

City name
Gothenburg
Oslo
Stockholm
Helsinki
Boyrvik
Skelleftea
Kuusamo
Karasjok

2002(Min)
-1.4
-1.6
-1.9
-2.3
-1.6
-1.9
-2.1
-1.4

4.4
5
5.2
5.6
5.7
5.9
6.7
5.3

2004(Max)
3
3.4
3.3
3.3
4.1
4
3.6
3.9

5.5
6.1
5.8
5.8
7.1
7.1
6.7
7.4

2008(Min)
-2.5
-2.7
-2.5
-2.5
-3.3
-3.1
-3.1
-3.5

8.5
9.6
8.9
9
11.7
11.2
11
12.1

2011(Max)
6
6.9
6.4
6.5
8.4
8.1
7.9
8.6

9.8
11.4
10.2
10.1
13.5
12.6
12
13.4

2012(Min)
-3.8
-4.5
-3.8
-3.6
-5.1
-4.5
-4.1
-4.8

Table 4: Local minima and maxima of the annual velocity of the geoid undulation. The
mm
denotes their differences. The values are in [ month
]
39

Table 4 summarizes the annual velocity of the geoid undulations. The columns describes either a local maximum or a local minimum as well as their differences. It can be
seen that the differences between the extremes seem to get larger by time. The biggest
difference can be observed between the maxima in 2011 and the minima in 2012, for all
selected cities. It can be also observed that the magnitude of the difference depends on
the latitude. Cities higher up in the north exhibit the strongest differences, between all
minima and maxima. Taking a look at the rearranged amplitude spectra, it can be observed that cities which are higher up in the north are stronger affected by variations in
periods of approximately 6 months and 49 months, since the corresponding coefficients
are higher.

40

Conclusions and Discussion

Within this thesis, the geoid undulations are computed for northern Europe, based on data
provided by the satellite gravity mission GRACE. Using the gathered data, an analysis is
carried out. The analysis is based on a visual inspection of plots of the geoid undulation
for every month, the change of the geoid undulation per year, an amplitude spectrum
as well as a rearranged amplitude spectrum. Afterwards the geoid undulation change is
compared with the change of the sea level and the change of the polar motion. In the
end, a comparison of the geoid undulation change for various cities in performed.
The computations for all selected cities (see figures 25-59 in the Appendix) show similar
results. However, it is possible to detect variations of the geoid undulation which seem
to be spatially dependent. The area around Oslo and Gothenburg exhibit the greatest
variation. The velocity of the annual change of the geoid undulation seems to be positively
correlated with the latitude.
The geoid undulation exhibits a stronger change in an annual temporal resolution than
a monthly temporal resolution. The annual change of the geoid undulation is confirmed by
the amplitude spectra and the rearranged amplitude spectra, as well as the computation
of the annual change of the geoid undulation. It is also detected that there is an overall
increase of the geoid undulation.
The change of the sea level and the geoid undulation show no significant similarities.
The sea level change fluctuates approximately 35 times stronger than the geoid undulation.
The cause of this could be the influence of atmospheric conditions such as rain and wind.
Thus, one might question the efficiency of methods which use the sea level to determine
the geoid undulation, such as satellite altimetry.
The comparison of the change of the geoid undulation derived from the polar motion
and the one derived from GRACE data leads to the conclusion that there is almost no
similarity present. Neither the magnitude nor the periodicity of the signals had any
observable similarities. The Chandler Wobble which is detectable in the geoid undulation
change derived from the polar motion is not visible in the geoid undulation change derived
from the GRACE data.

41

Future work

A deeper understanding and exploration of the temporal changes of the geoid, can be
gained by an extended observation of the geoid over more years. In order to be able to
proof the existence of temporal changes of the geoid, which come in periods of many years,
longer time series are required. This not only would increase our knowledge about the
geoid, but as a consequence, would also help in answering related topics such as global
warming.

42

References
[1] H. M. Bernhard Hofmann-Wellenhof, Pysical Geodesy. Springer, 2005.
[2] (2015, June) about grace. [Online]. Available: http://www.gfz-potsdam.de/grace/
[3] H. Fan, Pysical Geodesy, script for physical geodesy, KTH.
[4] W. E. F. S. A. Holmes, A unified approach to the clenshaw summation and the
recursive computation of very high degree and order normalised associated legendre
functions, Journal of Geodesy, 2002.
[5] S. Swenson and J. Wahr, Post-processing removal of correlated errors in grace data,
GEOPHYSICAL RESEARCH LETTERS, vol. 33, 2006.
[6] J. Kusche, R. Schmidt, S. Petrovic, and R. Rietbroek, Decorrelated grace
time-variable gravity solutions by gfz, and their validation using a hydrological
model, Journal of Geodesy, vol. 83, no. 10, pp. 903913, 2009. [Online]. Available:
http://dx.doi.org/10.1007/s00190-009-0308-3
[7] J. Adam, International Association of Geodesy (IAG): Geodesists Handbook 2008.
Springer, 2008.
[8] (2015,
June)
about
hashmaps
in
matlab.
[Online].
http://se.mathworks.com/help/matlab/ref/containers.map-class.html

Available:

[9] (2015,
June)
about
objects
in
matlab.
[Online].
Available:
http://se.mathworks.com/help/matlab/matlabp rog/matlab objects.html
[10] W. Zheng, C. Shao, J. Luo, and H. Xu, Improving the accuracy of {GRACE}
earths gravitational field using the combination of different inclinations, Progress
in Natural Science, vol. 18, no. 5, pp. 555 561, 2008. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S1002007108000592
[11] J. Wahr, S. Swenson, and I. Velicogna, Accuracy of grace mass estimates, Geophysical
Research Letters, vol. 33, no. 6, pp. n/an/a, 2006, l06401. [Online]. Available:
http://dx.doi.org/10.1029/2005GL025305
[12] (2015, June) Sea level change gothenburg. [Online]. Available:
catalog.smhi.se/explore/

http://opendata-

[13] L. Kurt, The Earths Variable Rotation : Geophysical Causes and Consequences. Cambridge University Press, 1980.

43

Appendix

44

45

Figure 25: geoid undulation over time in Stockholm. The red lines indicate the temporal trends.

Timeseries for selected cities Stockholm

46

Figure 26: Variation of the geoid undulation over time in Stockholm. The variation is computed by substracting the average from the
signal.

47

Figure 27: Changes of the geoid undulation within 12 months in Stockholm.

Figure 28: The amplitude spectrum of the change of the geoid undulation in Stockholm.

Figure 29: The rearranged amplitude spectrum of the change of the geoid undulation in
Stockholm

48

49

Figure 30: geoid undulation over time in Helsinki. The red lines indicate the temporal trends.

Helsinki

50

Figure 31: Variation of the geoid undulation over time in Helsinki. The variation is computed by substracting the average from the signal.

51

Figure 32: Changes of the geoid undulation within 12 months in Helsinki.

Figure 33: The amplitude spectrum of the change of the geoid undulation in Helsinki.

Figure 34: The rearranged amplitude spectrum of the change of the geoid undulation in
Helsinki

52

53

Figure 35: geoid undulation over time in Karasjok. The red lines indicate the temporal trends.

Karasjok

54

Figure 36: Variation of the geoid undulation over time in Karasjok. The variation is computed by substracting the average from the
signal.

55

Figure 37: Changes of the geoid undulation within 12 months in Karasjok.

Figure 38: The amplitude spectrum of the change of the geoid undulation in Karasjok.

Figure 39: The rearranged amplitude spectrum of the change of the geoid undulation in
Karasjok

56

57

Figure 40: geoid undulation over time in Kuusamo. The red lines indicate the temporal trends.

Kuusamo

58

Figure 41: Variation of the geoid undulation over time in Kuusamo. The variation is computed by substracting the average from the
signal.

59

Figure 42: Changes of the geoid undulation within 12 months in Kuusamo.

Figure 43: The amplitude spectrum of the change of the geoid undulation in Kuusamo.

Figure 44: The rearranged amplitude spectrum of the change of the geoid undulation in
Kuusamo

60

61

Figure 45: geoid undulation over time in Oslo. The red lines indicate the temporal trends.

Oslo

62

Figure 46: Variation of the geoid undulation over time in Oslo. The variation is computed by substracting the average from the signal.

63

Figure 47: Changes of the geoid undulation within 12 months in Oslo.

Figure 48: The amplitude spectrum of the change of the geoid undulation in Oslo.

Figure 49: The rearranged amplitude spectrum of the change of the geoid undulation in
Oslo

64

65

Figure 50: geoid undulation over time in Skelleftea. The red lines indicate the temporal trends.

Skelleftea

66

Figure 51: Variation of the geoid undulation over time in Skelleftea. The variation is computed by substracting the average from the
signal.

67

Figure 52: Changes of the geoid undulation within 12 months in Skelleftea.

Figure 53: The amplitude spectrum of the change of the geoid undulation in Skelleftea.

Figure 54: The rearranged amplitude spectrum of the change of the geoid undulation in
Skelleftea

68

69

Figure 55: geoid undulation over time in Boyrvik. The red lines indicate the temporal trends.

Boyrvik

70

Figure 56: Variation of the geoid undulation over time in Boyrvik. The variation is computed by substracting the average from the signal.

71

Figure 57: Changes of the geoid undulation within 12 months in Boyrvik.

Figure 58: The amplitude spectrum of the change of the geoid undulation in Boyrvik.

Figure 59: The rearranged amplitude spectrum of the change of the geoid undulation in
Boyrvik

72