16.08.2013 Views

Master's thesis Ørsted as a gravity mission satellite

Master's thesis Ørsted as a gravity mission satellite

Master's thesis Ørsted as a gravity mission satellite

SHOW MORE
SHOW LESS

Transform your PDFs into Flipbooks and boost your revenue!

Leverage SEO-optimized Flipbooks, powerful backlinks, and multimedia content to professionally showcase your products and significantly increase your reach.

FACULTY OF SCIENCE<br />

UNIVERSITY OF COPENHAGEN<br />

M<strong>as</strong>ter’s <strong>thesis</strong><br />

<strong>Ørsted</strong> <strong>as</strong> a <strong>gravity</strong> <strong>mission</strong><br />

<strong>satellite</strong><br />

Sofie Louise Sandberg Sørensen<br />

Niels Bohr Institute<br />

Academic advisor: Carl Christian Tscherning<br />

Submitted: 16/01/06


Contents<br />

1 Introduction 1<br />

1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2<br />

2 The <strong>Ørsted</strong> Satellite 3<br />

2.1 Instruments on board <strong>Ørsted</strong> . . . . . . . . . . . . . . . . . . . . . . . . . 3<br />

2.2 Satellite-to-<strong>satellite</strong> Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . 4<br />

3 <strong>Ørsted</strong> Data 6<br />

3.1 RINEX Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6<br />

3.2 JPL Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6<br />

3.3 ECI Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7<br />

4 B<strong>as</strong>ic Theory 8<br />

4.1 Spherical Harmonics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8<br />

4.2 The Gravitational Potential of the Earth . . . . . . . . . . . . . . . . . . . 10<br />

4.3 The Normal Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11<br />

4.4 Chauvenet’s Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12<br />

4.5 Rotation between CTS and CIS . . . . . . . . . . . . . . . . . . . . . . . . 13<br />

4.6 Simpson’s Rule of Integration . . . . . . . . . . . . . . . . . . . . . . . . . 15<br />

5 Selecting the Data 16<br />

5.1 Solar Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16<br />

5.2 Evaluation of the GPS Data . . . . . . . . . . . . . . . . . . . . . . . . . . 16<br />

5.3 Ground Track of the Satellite . . . . . . . . . . . . . . . . . . . . . . . . . 18<br />

6 Gravity Field Determination 20<br />

6.1 Dedicated Gravity Mission Satellites . . . . . . . . . . . . . . . . . . . . . 21<br />

6.1.1 CHAMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21<br />

6.1.2 GRACE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21<br />

6.1.3 GOCE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23<br />

6.2 <strong>Ørsted</strong> <strong>as</strong> a Gravity Mission Satellite. . . . . . . . . . . . . . . . . . . . . . 23<br />

7 Methods for Gravity Field Determination from State Vectors Only 25<br />

7.1 The Acceleration Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 25<br />

7.2 The Energy Conservation Theorem . . . . . . . . . . . . . . . . . . . . . . 26<br />

7.2.1 Tidal Gravitational Disturbances Caused by the Sun and Moon . . 26<br />

7.2.2 Rotational Potential . . . . . . . . . . . . . . . . . . . . . . . . . . 27<br />

7.2.3 Non-Gravitational Disturbing Potential . . . . . . . . . . . . . . . . 28


8 Deriving The Gravitational Potential 29<br />

8.1 Calculation of Vkin, Vrot, Vsun and Vmoon . . . . . . . . . . . . . . . . . . . . 29<br />

8.2 Determination of E0 and F . . . . . . . . . . . . . . . . . . . . . . . . . . 34<br />

8.3 Calculation of the Gravitational Potential Vearth . . . . . . . . . . . . . . . 39<br />

9 Theory of Le<strong>as</strong>t Squares Collocation 41<br />

9.1 The Covariance Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43<br />

10 Methods of Calculating the Spherical Harmonic Coefficients 45<br />

10.1 Method 1: Using Numerical Integration . . . . . . . . . . . . . . . . . . . . 45<br />

10.2 Method 2: Using Orthogonality Properties of the Spherical Harmonics . . . 46<br />

10.3 The Kepler Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . 47<br />

11 Collo.f90 50<br />

11.1 Program outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50<br />

11.2 Programming Problems and Solutions . . . . . . . . . . . . . . . . . . . . . 55<br />

11.2.1 The multi-grid function . . . . . . . . . . . . . . . . . . . . . . . . . 55<br />

11.2.2 Iterative Improvement . . . . . . . . . . . . . . . . . . . . . . . . . 56<br />

11.2.3 Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57<br />

11.2.4 Regularization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58<br />

12 Deriving the <strong>Ørsted</strong> Gravitational Potential Field Model ORSTED05 60<br />

12.1 Calculation of GM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60<br />

12.2 Comparing Results from Method 1 and Method 2 . . . . . . . . . . . . . . 60<br />

12.3 Collo.f90 error test on EGM96 test data . . . . . . . . . . . . . . . . . . . 62<br />

12.4 The Model ORSTED05 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64<br />

13 Final Considerations and Discussion 69<br />

14 Conclusion 74<br />

A Data Files 77<br />

A.1 ECI file example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77<br />

A.2 JPL file example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77<br />

A.3 RINEX file example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77<br />

B EGM96 coefficients 78<br />

C Fortran programs 79


Abstract<br />

The <strong>Ørsted</strong> <strong>satellite</strong> w<strong>as</strong> launched in 1999, with the main goal of modeling the Earth’s<br />

magnetic field.<br />

In this <strong>thesis</strong> <strong>Ørsted</strong> h<strong>as</strong> been evaluated <strong>as</strong> a <strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>, and the goal h<strong>as</strong><br />

been to derive a <strong>gravity</strong> field model, b<strong>as</strong>ed only on <strong>Ørsted</strong> data.<br />

By using <strong>satellite</strong>s like <strong>Ørsted</strong> <strong>as</strong> <strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>s, it will in theory be possible to<br />

gain new information on the variations in time of the <strong>gravity</strong> field, because these <strong>satellite</strong>s<br />

have been orbiting the Earth for a longer period of time that the dedicated <strong>gravity</strong> <strong>mission</strong><br />

<strong>satellite</strong>s. In this <strong>thesis</strong>, only the static <strong>gravity</strong> field h<strong>as</strong> been modeled.<br />

The <strong>Ørsted</strong> data, used in this project, are GPS me<strong>as</strong>urements giving the state vector<br />

of the <strong>satellite</strong>. Since the kinetic energy can be calculated directly from the state vector,<br />

the energy conservation method h<strong>as</strong> been used to derive the gravitational potential from<br />

the state vectors.<br />

When using the energy conservation method, information on the non-gravitational accelerations<br />

are needed. These are not me<strong>as</strong>ured by the <strong>Ørsted</strong> <strong>satellite</strong>, and therefore an<br />

approximations of these h<strong>as</strong> been found and applied.<br />

Le<strong>as</strong>t squares collocation h<strong>as</strong> been used to derive a global <strong>gravity</strong> field model from<br />

the calculated gravitational potential. Together with the le<strong>as</strong>t squares collocation, two<br />

different methods of calculating the spherical harmonic coefficients of the <strong>gravity</strong> model<br />

h<strong>as</strong> been used and compared; a method using numerical integration and a method using<br />

the orthogonality properties of the Legendre polynomials.<br />

It w<strong>as</strong> shown that the two methods were equally accurate, if the grid used in the numerical<br />

integration method, w<strong>as</strong> chosen sufficiently dense. The method using the orthogonality<br />

properties w<strong>as</strong> chosen for the full modeling.<br />

An <strong>Ørsted</strong> gravitational potential model, ORSTED05, w<strong>as</strong> found which contains spherical<br />

harmonic coefficients up to degree and order 6. The upper limit of obtainable coefficients<br />

w<strong>as</strong> determined by comparison with EGM96 coefficients and evaluation of the modeling<br />

errors from collo.f90, the program which h<strong>as</strong> been developed for the model calculations.<br />

The ORSTED05 potential field model, with the term ¯ C20 subtracted, is shown in the figure<br />

below.


Sammendrag<br />

Den danske satellit <strong>Ørsted</strong> blev sendt i kredsløb om Jorden i 1999. Hovedfom˚alet med<br />

<strong>Ørsted</strong> var at kortlægge Jordens magnetfelt.<br />

I dette speciale er mulighederne for, at benytte <strong>Ørsted</strong> satellitten som tyngde <strong>mission</strong>ssatellit<br />

blevet evalueret, og form˚alet med specialet har været, at bestemme en model af<br />

Jordens tyngdefelt kun b<strong>as</strong>eret p˚a data fra <strong>Ørsted</strong>.<br />

Ved at benytte satellitter som <strong>Ørsted</strong> i tyngdefelts-modelleringer er det i princippet<br />

muligt at forbedre billedet af de tidslige variationer af tyngdefeltet, da denne slags satellitter<br />

har været i kredsløb omkring Jorden i længere tid end de egentlige tyngde<strong>mission</strong>ssatellitter.<br />

I dette projekt er kun det statiske tyngdefelt blevet modelleret.<br />

De <strong>Ørsted</strong> data, som er blevet benyttet i dette speciale, er GPS-data der indeholder<br />

information om satellittens position og h<strong>as</strong>tighed (state vector). Da den kinetiske energi af<br />

satellitten kan beregnes direkte ud fra state vector’en, er energibevarelses metoden blevet<br />

valgt til bestemmelse af Jorden’s tyngdepotential. N˚ar man benytter energibevarelses metoden,<br />

er det nødvendigt at have kendskab til satellittens ikke-gravitationelle accelerationer,<br />

for at kunne isolere og bestemme de accelerationer, der skyldes Jordens tyngdepotential.<br />

De ikke-gravitationelle accelerationer m˚ales ikke af <strong>Ørsted</strong>, og derfor har det, i dette<br />

projekt, været nødvendigt at finde en m˚ade at approksimere disse p˚a.<br />

Mindste kvadraters kollokation er blevet benyttet til at bestemme en global model<br />

af Jorden’s tyngdepotential. I dette speciale er to forskellige metoder blevet benyttet<br />

sammen med kollokation, til at bestemme kuglefunktions-koefficienterne i denne model.<br />

Ved den ene metode benyttes numerisk integration, og ved den anden metode udnyttes<br />

ortogonalitetsegenskaberne af Legendre funktionerne til bestemmelsen af koefficienterne.<br />

Den globale model, ORSTED05, der er fundet i dette speciale, er en kuglefunktionsudvikling<br />

af Jorden’s tyngdepotential der indeholder led op til grad og orden 6. Den<br />

øvre grænse af led der kunne bestemmes ud fra <strong>Ørsted</strong> data, og som indg˚ar i ORSTED05,<br />

er blevet bestemt p˚a baggrund af sammenligning med EGM96 samt overvejelser af de<br />

modellerings-fejl der genereres af collo.f90, der er det program der er blevet udviklet<br />

til at udføre model beregningerne. ORSTED05 potentialfelts-modellen, med ¯ C20 leddet<br />

fratrukket, er vist i figuren herunder.


I would like to thank<br />

Acknowledgments<br />

- C. C. Tscherning for being a committed and helpful academic advisor.<br />

- Martin B. Sørensen for valuable help on working with the <strong>Ørsted</strong> data.<br />

- Anders Lynge Esbensen for great support,<br />

both personal and programming, throughout the project.<br />

- Henriette Hjorth for sharing office and opinions with me.<br />

- Eva Howe for always wanting to help.


1 Introduction<br />

1. Introduction<br />

The main purpose of this <strong>thesis</strong>, will be to explore the possibilities of deriving a model<br />

of the Earths <strong>gravity</strong> field, b<strong>as</strong>ed only on data from the danish micro <strong>satellite</strong> <strong>Ørsted</strong>. It<br />

h<strong>as</strong> to my knowledge not been done before, due to the fact that <strong>Ørsted</strong> is not designed for<br />

<strong>gravity</strong> field me<strong>as</strong>urement but for magnetic me<strong>as</strong>urements.<br />

The modeling will be b<strong>as</strong>ed on the energy conservation method and le<strong>as</strong>t squares collocation,<br />

and the aim is to determine which level of accuracy can be achieved for such an<br />

<strong>Ørsted</strong> <strong>gravity</strong> field model.<br />

When using the energy conservation method, information of the non-gravitational potential,<br />

such <strong>as</strong> the friction acting on the <strong>satellite</strong>, is needed. But since there is no accelerometer<br />

placed on board <strong>Ørsted</strong>, there are no direct me<strong>as</strong>urements of this. Therefore,<br />

one of the major challenges of this <strong>thesis</strong> will be to find an acceptable method to estimate<br />

this non-gravitational potential.<br />

Before getting started, one must recognize that it will not be possible to obtain better<br />

- or even <strong>as</strong> good - models derived from <strong>Ørsted</strong> data, <strong>as</strong> those derived from data<br />

from the dedicated <strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>s like for example CHAMP [Howe et al., 2003],<br />

[Gerlach et al., 2003a]. The re<strong>as</strong>ons being the lack of non-gravitational acceleration data,<br />

and the fact that <strong>Ørsted</strong> is orbiting the earth at a much higher altitude than the dedicated<br />

<strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>s.<br />

The <strong>Ørsted</strong> <strong>satellite</strong> and <strong>mission</strong> is explained in section 2, and the different available data<br />

files from <strong>Ørsted</strong> are described in section 3.<br />

In section 4 the b<strong>as</strong>ic theory, needed in this project, is explained. This section includes<br />

for instance an introduction to spherical harmonics, a description of the use of Chauvenet’s<br />

criterion and the theory of rotation between the Conventional Terrestrial System (CTS)<br />

and the Conventional Inertial System (CIS).<br />

The selection of usable data for this project is described in section 5.<br />

Gravity field determination in general is described in section 6, in which three different<br />

dedicated <strong>gravity</strong> <strong>mission</strong>s are also described. In this section <strong>Ørsted</strong> is evaluated <strong>as</strong> a<br />

<strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>, b<strong>as</strong>ed on considerations of the instruments on board the <strong>satellite</strong><br />

and the <strong>satellite</strong>s orbital parameters.<br />

The theory behind the energy conservation method is described in section 7 and the<br />

calculations in practice in section 8.<br />

In order to derive a model from the calculated gravitational potential, le<strong>as</strong>t squares<br />

collocation is used. The theory behind this method is explained in section 9. In section 10<br />

three different methods of calculating the spherical harmonic coefficients of the model are<br />

explained, and two of these methods are evaluated in section 12. The <strong>gravity</strong> modeling in<br />

practice is executed by the program collo.f90, which h<strong>as</strong> been developed during this project,<br />

and this program is described in section 11. In section 11.2 some of the programming<br />

problems and solutions related to collo.f90 are described.<br />

1


1. Introduction<br />

The result of this <strong>thesis</strong> is the gravitational potential field model ORSTED05, and in<br />

section 12 the ORSTED05 model is derived and described.<br />

In section 13 the main t<strong>as</strong>ks of the <strong>thesis</strong> are summarized and the choices, made throughout<br />

the project, are discussed.<br />

In section 14 the final conclusions on the <strong>thesis</strong> is made.<br />

1.1 Motivation<br />

I believe that it will be of great importance, if it can be shown to be possible, to derive<br />

useful information on the Earth’s <strong>gravity</strong> field from <strong>Ørsted</strong> data.<br />

This will mean that the number of <strong>satellite</strong>s, which can be used for the purpose of<br />

<strong>gravity</strong> field modeling, will incre<strong>as</strong>e. The re<strong>as</strong>on being that other <strong>satellite</strong>s similar to<br />

<strong>Ørsted</strong> also could be shown to be useful, in spite of the fact that they are not designed<br />

specifically for <strong>gravity</strong> field me<strong>as</strong>urements.<br />

Gravity field derivations from <strong>satellite</strong>s like <strong>Ørsted</strong> will, for example, give us new information<br />

on the variations of the <strong>gravity</strong> field in time. Hence these <strong>satellite</strong>s have been<br />

orbiting the earth for a longer period of time, than the specific <strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>s.<br />

Furthermore, these <strong>satellite</strong>s can contribute with additional information to the ongoing<br />

static <strong>gravity</strong> field modeling, since the coverage of the Earths surface will become better<br />

when the number of useful <strong>satellite</strong>s incre<strong>as</strong>es. The result will be a more accurate and<br />

detailed model of the earth’s <strong>gravity</strong> field.<br />

Gravity field models are of great importance in a wide variety of sciences. The solidearth<br />

geophysicists can derive information on the structure of the earth, and phenomena<br />

like sea level changes and land uplift can also be observed, when studying the <strong>gravity</strong> field<br />

variations. Furthermore glaciologists and oceanographers use <strong>gravity</strong> field variations to<br />

investigate the movement of ice sheets and ocean circulation.<br />

Satellites provide important data for the ongoing <strong>gravity</strong> field modeling, because their<br />

me<strong>as</strong>urements cover the Earth much better than it would ever be possible to do by me<strong>as</strong>urements<br />

on earth only. But since <strong>satellite</strong> data is not nearly <strong>as</strong> accurate <strong>as</strong> earth me<strong>as</strong>urements,<br />

a combination of both data is often used in modern <strong>gravity</strong> field modeling.<br />

2


2 The <strong>Ørsted</strong> Satellite<br />

2. The <strong>Ørsted</strong> Satellite<br />

The <strong>gravity</strong> field modeling which is the goal of this <strong>thesis</strong>, will be b<strong>as</strong>ed only on data from<br />

<strong>Ørsted</strong>. Therefore, the <strong>satellite</strong> and the <strong>Ørsted</strong> <strong>mission</strong> is shortly described in this section.<br />

The <strong>Ørsted</strong> <strong>satellite</strong> is the first and only danish <strong>satellite</strong>, and it is the result of a collaboration<br />

between institutions of research and private companies in Denmark. The <strong>satellite</strong><br />

is seen in figure 1.<br />

The <strong>satellite</strong> w<strong>as</strong> launched on the 23rd of February 1999, and it is still functional today.<br />

It transmits data to the Earth several times per day. Some facts about the <strong>Ørsted</strong> <strong>satellite</strong><br />

are listed in the box below [E. F. Christensen, 1994].<br />

Facts on the <strong>Ørsted</strong> <strong>satellite</strong>.<br />

-Inclination : 96.1 degrees.<br />

-Orbital period : ∼ 100 minutes.<br />

-Apogee altitude : ∼ 850 km.<br />

-Perigee altitude : ∼ 450 km<br />

-Weight : 60.7 kg.<br />

-Total cost: 126.1 mill dkr.<br />

-Eccentricity 0.015<br />

The <strong>Ørsted</strong> <strong>satellite</strong> is designed for me<strong>as</strong>urements of the Earths magnetic field and it’s<br />

variations in time. The first magnetic me<strong>as</strong>urements made by <strong>Ørsted</strong> were, at the time, the<br />

most precise ever made from space. It resulted in the <strong>Ørsted</strong> Initial Field Model (OIFM).<br />

Besides mapping the Earth’s magnetic field, other purposes of the <strong>Ørsted</strong> <strong>mission</strong> h<strong>as</strong><br />

been to investigate the electric current systems in the upper atmosphere and magnetosphere,<br />

and to incre<strong>as</strong>e our knowledge of the connection between the solar wind and the<br />

Earths magnetic field. A GPS receiver, placed on board the <strong>Ørsted</strong> <strong>satellite</strong>, h<strong>as</strong> made it<br />

possible also to derive information on some meteorological conditions in the atmosphere,<br />

such <strong>as</strong> temperature and air humidity, for further information see<br />

http://esa-spaceweather.net/sda/gpsvalidation/.<br />

2.1 Instruments on board <strong>Ørsted</strong><br />

An Overhaus magnetometer is placed on board <strong>Ørsted</strong>, which me<strong>as</strong>ures the strength of<br />

the magnetic field. It is located at the end of a long boom, in order to prevent it from<br />

me<strong>as</strong>uring the magnetic disturbances from the <strong>satellite</strong>’s electrical instruments. A vector<br />

magnetometer, also placed on the boom, me<strong>as</strong>ures the components of the field.<br />

3


2. The <strong>Ørsted</strong> Satellite<br />

Figure 1: The <strong>Ørsted</strong> <strong>satellite</strong>.<br />

There are two GPS receivers on board <strong>Ørsted</strong>; a TANS 2 receiver produced by Trimble<br />

and a TurboRogue receiver produced by Jet Propulsion laboratories (JPL). These are used<br />

for determination of the exact position of <strong>Ørsted</strong>.<br />

The majority of the time, the TANS 2 receiver is used for the positioning of the <strong>satellite</strong>,<br />

though it is not nearly <strong>as</strong> accurate at the TurboRogue receiver. This is due to the fact<br />

that the TurboRogue uses a lot of electric power compared to the TANS 2.<br />

The GPS me<strong>as</strong>urements do not give information on the attitude of the <strong>satellite</strong> in space.<br />

For this purpose there is a star camera on board the <strong>satellite</strong>. It continuously takes images<br />

of the visible stars and compare these with star catalogues. This gives a very precise<br />

determination of attitude.<br />

For the purpose of determining the density of charged particles (for example protons,<br />

electrons and helium nuclea) in space, there is also a particle detector on board.<br />

The power for all these instruments is provided by solar panels. They also charge<br />

batteries, which provide the power for the instruments, whenever the <strong>satellite</strong> is in the<br />

shadow of the Earth.<br />

The location of the individual instruments on the <strong>satellite</strong> is shown in figure 1<br />

[E. F. Christensen, 1994].<br />

2.2 Satellite-to-<strong>satellite</strong> Tracking<br />

The <strong>Ørsted</strong> <strong>satellite</strong> is part of a configuration called <strong>satellite</strong>-to-<strong>satellite</strong> tracking in highlow<br />

mode (SST-HL) [Seeber, 2003]. The low orbiting <strong>Ørsted</strong> <strong>satellite</strong>, with an average<br />

4


2. The <strong>Ørsted</strong> Satellite<br />

Figure 2: Illustration of <strong>satellite</strong>-to-<strong>satellite</strong> tracking in high-low mode.<br />

height of approximately 700km, is tracked by the high orbiting GPS <strong>satellite</strong>s, with a<br />

height of approximately 26.000km, and its position is thereby determined relative to an<br />

Earth-fixed coordinate system, CTS. The constellation of SST-HL can be seen in figure 2.<br />

The position of the low orbiting <strong>satellite</strong> is found by continuously me<strong>as</strong>uring the ranges<br />

and range changes between this and several GPS <strong>satellite</strong>s (at le<strong>as</strong>t 4), which positions are<br />

known with high accuracy. For the purpose of <strong>gravity</strong> field me<strong>as</strong>urements, the low orbiting<br />

<strong>satellite</strong> (in this c<strong>as</strong>e <strong>Ørsted</strong>) is used <strong>as</strong> a probe in the Earths <strong>gravity</strong> field. Information<br />

on the <strong>gravity</strong> field can be derived by studying the motion of the low orbiting <strong>satellite</strong>.<br />

5


3. <strong>Ørsted</strong> Data<br />

3 <strong>Ørsted</strong> Data<br />

The <strong>Ørsted</strong> data, which will be used in this <strong>thesis</strong>, consists purely of GPS-data. This<br />

GPS-data is available in three different types of files; RINEX files, JPL files and ECI files.<br />

Only GPS-data from the TurboRogue receiver will be used, since this is the most accurate<br />

of the two GPS receivers on board <strong>Ørsted</strong>, see section 2. The <strong>Ørsted</strong> GPS-data have been<br />

collected and processed by DMI.<br />

In the following, the three types of GPS data files are described, and in appendix A<br />

examples of the files are given.<br />

3.1 RINEX Files<br />

The RINEX files are the post processed GPS data files, and both the JPL and ECI files<br />

are derived from the RINEX files. RINEX is short for Receiver Independent Exchange<br />

Format, and it is a file format which h<strong>as</strong> been developed, in order to minimize the potential<br />

problems of international collaboration. It is an internationally accepted data format,<br />

by both users and manufacturers and being receiver independent, the RINEX data from<br />

different receiver types can e<strong>as</strong>ily be processed together. The RINEX files, used in this<br />

<strong>thesis</strong>, contain information on which GPS <strong>satellite</strong>s, the TurboRogue GPS receiver h<strong>as</strong><br />

been in contact with, for every single me<strong>as</strong>urement. Besides this information the time,<br />

range (pseudo-range) and ph<strong>as</strong>e of each signal are given in the RINEX files. The time<br />

given in the RINEX files, is the arrival time of the received signals, and it is given in GPS<br />

time.<br />

The pseudo-range P R is defined <strong>as</strong>:<br />

P R = d + c ∗ (dtreceiver − dt<strong>satellite</strong> + B), (1)<br />

where d is the distance between the receiver antenna and the <strong>satellite</strong> antenna, c is<br />

the velocity of light, dtreceiver is the receiver clock offset and dt<strong>satellite</strong> is the <strong>satellite</strong> clock<br />

offsets. B are the sum of other bi<strong>as</strong>es, <strong>as</strong> for example atmospheric delays. The pseudorange<br />

is given in meters.<br />

The ph<strong>as</strong>es given in the RINEX files are the carrier ph<strong>as</strong>es of both L1 and L2.<br />

For further information on the RINEX file format see for example<br />

http://igscb.jpl.n<strong>as</strong>a.gov/igscb/data/format/rinex2.txt.<br />

3.2 JPL Files<br />

The JPL (Jet Propulsion Laboratories) files are processed GPS data files. The JPL files<br />

contain information on the state vector ¯ VCT S of the <strong>satellite</strong>, calculated from the information<br />

in the RINEX files. The state vector is given every 10th second in Cartesian<br />

coordinates in an earth-fixed coordinate system CTS.<br />

¯VCT S = (¯x, ¯v)CT S = (x, y, z, vx, vy, vz)CT S, (2)<br />

6


3. <strong>Ørsted</strong> Data<br />

where ¯x is the position and ¯v is the velocity of <strong>Ørsted</strong>. The units of the state vector are<br />

[km] and [ km]<br />

of the position and the velocity respectively. The time is, <strong>as</strong> in the RINEX<br />

s<br />

files, given in GPS time.<br />

The ¯ VCT S state vector is calculated directly from the RINEX file, since the GPS data<br />

is given relative to a system of reference points on the Earth, meaning relative to the CTS<br />

system.<br />

3.3 ECI Files<br />

The ECI (Earth Centered Inertial) files contain information on the state vector, ¯ VCIS, of<br />

<strong>Ørsted</strong>’s motion relative to the Earth-centered inertial system CIS.<br />

¯VCIS = (¯x, ¯v)CIS = (x, y, z, vx, vy, vz)CIS, (3)<br />

The data in the ECI files are calculated from the data in the JPL files by rotating the<br />

vectors from the Earth-fixed system (CTS) to the inertial system (CIS):<br />

rCIS = P −1 N −1 S −1 rCT S<br />

where P is the precession matrix, N is the nutation matrix and S is the matrix, which<br />

rotates from the space-fixed system to the Earth-fixed system. The rotation from CTS to<br />

CIS is fully explained in section 4.5.<br />

The time in the ECI files is given in UTC time. Throughout the period of time relevant<br />

in this <strong>thesis</strong>, the difference between GPS time and UTC time h<strong>as</strong> been 13 seconds.<br />

GPS time − UTC1999 = GPS time − UTC2000 = +13seconds (5)<br />

The latter is an important information when comparison between the state vectors in<br />

the JPL files and the ECI files must be made.<br />

7<br />

(4)


4. B<strong>as</strong>ic Theory<br />

4 B<strong>as</strong>ic Theory<br />

In this section some of the b<strong>as</strong>ic theory needed in this project, is explained. For example,<br />

some potential field theory, a method for numerical integration and rotation between coordinate<br />

systems is described.<br />

Firstly the spherical harmonics are introduced, since these functions are of primary importance<br />

in the <strong>gravity</strong> field modeling.<br />

4.1 Spherical Harmonics.<br />

The functions, Y c<br />

nm and Y s<br />

nm, are called the Laplace surface spherical harmonics<br />

[Torge, 1999].<br />

Y c<br />

nm = Pnm(cosθ)cosmλ (6)<br />

Y s<br />

nm = Pnm(cosθ)sinmλ<br />

They are an orthogonal b<strong>as</strong>is for the Hilbert space on the unit sphere. Any continuous<br />

function on a sphere, <strong>as</strong> for example the Earth’s <strong>gravity</strong> field, can be represented by a<br />

linear combination of these functions. The functions Pnm(cosθ) are the <strong>as</strong>sociated Legendre<br />

functions of degree n and order m.<br />

The orthogonality of the b<strong>as</strong>is functions, are due to:<br />

<br />

Y i<br />

nmY k<br />

n ′ m ′dσ = 0, (7)<br />

form:<br />

σ<br />

for n = n ′ , m = m ′ or i = k. For n = n ′ , m = m ′ and i = k the latter equation h<strong>as</strong> the<br />

<br />

σ<br />

Y i<br />

nmY i<br />

nm =<br />

<br />

4π<br />

2n+1<br />

2π (n+m)!<br />

2n+1<br />

for m = 0<br />

for m = 0<br />

(n−m)!<br />

This means that the fully normalized spherical harmonics can be written <strong>as</strong> follows<br />

[Torge, 1999] :<br />

⎧ <br />

⎨ 2n+1<br />

¯Ynm<br />

4π<br />

=<br />

⎩<br />

Ynm for m = 0<br />

<br />

2n+1 (n−m)!<br />

2π (n+m)! Ynm<br />

(9)<br />

for m = 0<br />

The Laplace surface spherical harmonics of order zero (m = 0) are called the zonal<br />

harmonics. They have no dependence on the longitude λ since:<br />

Y c<br />

n0 = Pn0(cosθ)cos0λ = Pn0(cosθ) (10)<br />

Y s<br />

n0 = Pn0(cosθ)sin0λ = 0<br />

8<br />

(8)


(a) (b) (c)<br />

4. B<strong>as</strong>ic Theory<br />

Figure 3: Spherical harmonics. (a) Zonal harmonics with n = 3 and m = 0. (b) Tesseral<br />

harmonics with n = 5 and m = 2. (c) Sectorial harmonics with n = 3 and m = 3.<br />

The zonal harmonics have n nodes (zeros) on the interval (0 ≤ θ ≤ π), separating zones<br />

of alternating signs on the unit sphere.<br />

The spherical harmonics with m = 0 are longitude dependent and are called the tesseral<br />

harmonics. These contain 2m nodes in the interval 0 ≤ λ ≤ π and (n − m) nodes in the<br />

interval 0 ≤ θ ≤ π.<br />

The l<strong>as</strong>t kind of harmonics are called the sectorial harmonics, and are defined by same<br />

degree and order (n = m). These have no latitude dependency. Examples of the different<br />

kinds of spherical harmonics are shown in figure 3.<br />

9


4. B<strong>as</strong>ic Theory<br />

4.2 The Gravitational Potential of the Earth<br />

The gravitational field of the Earth is invariant to rotation, meaning that:<br />

curl ¯g = 0, (11)<br />

with ¯g being the gravitational acceleration vector [Torge, 1999], [Seeber, 2003]. This<br />

characteristic means that the following is valid:<br />

¯g = grad V, (12)<br />

where V is the Earth’s gravitational potential. The gravitational potential h<strong>as</strong> the<br />

form:<br />

<br />

1<br />

V (r) = G<br />

dm , lim V = 0, (13)<br />

earth r − r ′ r→∞<br />

where r − r ′ is the distance from the m<strong>as</strong>s volume dm(r ′ , θ ′ , λ ′ ) to the point P (r, θ, λ).<br />

Hence the gravitational potential in point P is the work needed, in order to move a unit<br />

m<strong>as</strong>s from r = ∞ to the point P .<br />

The Earth’s gravitational potential, V , is a continuous function with continuous derivatives<br />

of all orders, and it furthermore fulfills Laplace’s differential equation of second order<br />

in places with no m<strong>as</strong>s:<br />

∆V = 0 (14)<br />

Therefore, the Earth’s gravitational potential field is a harmonic function, and can be<br />

represented by a spherical harmonic expansion, <strong>as</strong> mentioned in section 4.1. The spherical<br />

harmonic expansion transforms the volume integral in eq. 13 into an infinite series by<br />

rewriting the reciprocal distance, 1<br />

r−r ′ , into a series ([Torge, 1999],eq. 3.79):<br />

1 1<br />

=<br />

r − r ′ r<br />

∞ ′ r<br />

n Pn(cos Ψ) (15)<br />

r<br />

n=0<br />

where Ψ is the angle distance between dm and P. Combining equation 13 and 15 the<br />

following expression can be derived, see also section 4.1:<br />

V = GM<br />

(1 +<br />

r<br />

∞<br />

n<br />

n=2 m=0<br />

<br />

a<br />

n Pnm(cos θ)(Cnm cos mλ + Snm sin mλ)) (16)<br />

r<br />

where a is the equatorial radius of earth, a = 6378136.6m. The term with n = 1 is zero<br />

since the center of the coordinate system is chosen to be at the Earth’s center of m<strong>as</strong>s.<br />

These equations are valid if all the m<strong>as</strong>s of the Earth is <strong>as</strong>sumed enclosed.<br />

The latter equation can equivalently be expressed in terms of the fully normalized<br />

Legendre functions ¯ Pnm and coefficients ¯ Cnm and ¯ Snm:<br />

10


V = GM<br />

(1 +<br />

r<br />

∞<br />

n<br />

n=2 m=0<br />

4. B<strong>as</strong>ic Theory<br />

<br />

a<br />

n ¯Pnm(cos θ)(<br />

r<br />

¯ Cnm cos mλ + ¯ Snm sin mλ)) (17)<br />

In geodesy, equation 16 is often written with another notation, namely:<br />

V = GM<br />

(1 −<br />

r<br />

∞<br />

n<br />

n=2 m=0<br />

<br />

a<br />

n Pnm(cos θ)(Jnm cos mλ + Knm sin mλ)) (18)<br />

r<br />

The spherical harmonic expansion is conventionally divided into two parts:<br />

V = U + T, (19)<br />

where T is called the disturbing potential and U is the normal potential defined subsequently.<br />

The spherical harmonic expansion of the Earth’s gravitational potential, given i eq. 16,<br />

simply represents a spectral decomposition of the gravitational field. The amplitude of<br />

each term in the expansion are determined by the harmonic coefficients Cnm and Snm (or<br />

Jnm and Knm), which are m<strong>as</strong>s integrals describing the m<strong>as</strong>s distribution in the Earth.<br />

The zero order term , GM also called the Keplerian term, represents the potential of a<br />

r<br />

homogeneous or radially layered Earth. The coefficient C20 = −J20 is called the dynamical<br />

form factor and it represents the flattening of the Earth.<br />

4.3 The Normal Potential<br />

The normal potential, U, is the potential of the normal <strong>gravity</strong>, γ:<br />

γ = grad U (20)<br />

The normal <strong>gravity</strong> field is the external <strong>gravity</strong> field of an equi-potential ellipsoid. The<br />

normal potential consists of a gravitational, Vellipsoid, and a centrifugal, Zellipsoid, term.<br />

U = Vellipsoid + Zellipsoid, (21)<br />

The spherical harmonic expansion of the normal potential h<strong>as</strong> the form:<br />

U = GM<br />

(1 −<br />

r<br />

∞<br />

n=1<br />

( a<br />

r )2n J2nP2n(cosθ)) + ω2<br />

2 r2 sin 2 (θ), (22)<br />

where a is the semi major axis of the ellipsoid, J2n are the even harmonic coefficients<br />

and P2n are the even Legendre functions.<br />

The odd terms of Jn are zero due to the symmetry of the ellipsoid with respect to the<br />

equator of the earth. Due to the symmetry of the ellipsoid with respect to the rotational<br />

axis of the earth, U contains no tesseral terms.<br />

In this <strong>thesis</strong>, the normal potential will be used <strong>as</strong> a reference field.<br />

11


4. B<strong>as</strong>ic Theory<br />

4.4 Chauvenet’s Criterion<br />

When evaluation of noisy data is needed, Chauvenet’s criterion can be used. In other<br />

words, Chauvenet’s criterion can be used to filter data sets.<br />

In this <strong>thesis</strong>, Chauvenet’s criterion will be used to evaluate the TurboRogue GPS<br />

me<strong>as</strong>urements, and in the following the criterion is described, [Taylor, 1997].<br />

If we suppose having a series of N me<strong>as</strong>urements, xi, of a quantity x, which probability<br />

distribution is governed by a normal distribution. The mean, ¯x, and standard deviation,<br />

σx, of the me<strong>as</strong>urements can be estimated using the following equations.<br />

¯x = 1<br />

N<br />

xi<br />

(23)<br />

N<br />

i=1<br />

<br />

<br />

<br />

σx = 1<br />

N<br />

(xi − ¯x)<br />

N − 1<br />

2 (24)<br />

Each me<strong>as</strong>urement can be evaluated by calculating the value tsus of the me<strong>as</strong>urement.<br />

tsus is the difference between the value of the me<strong>as</strong>urement and the mean value of the<br />

distribution, given in units of the standard deviations.<br />

tsus =<br />

i=1<br />

|x − ¯x|<br />

Then the probability, p, that a legitimate me<strong>as</strong>urement would differ from ¯x by more<br />

than tsus standard deviations is determined.<br />

σx<br />

p = P rob(outside(tsusσx))<br />

(25)<br />

= 1 − P rob(inside(tsusσx)) (26)<br />

= 1 − 1<br />

tsus<br />

√ e<br />

2π<br />

−z2 /2<br />

dz<br />

The latter equation must be calculated on a computer, since it is not possible to evaluate<br />

it analytically. p is converted into number of me<strong>as</strong>urements, n, by multiplying the<br />

probability p by the number of me<strong>as</strong>urements N. n is statistically the number of points<br />

which differ at le<strong>as</strong>t tsus standard deviations from ¯x, and therefore the size of n determines<br />

whether a me<strong>as</strong>urement must be rejected or accepted.<br />

−tsus<br />

n = pN (27)<br />

if (n ≥ 0.5) then accept data<br />

if (n < 0.5) then reject data<br />

If indeed a me<strong>as</strong>urement is rejected, and thereby filtered out, the mean and standard<br />

deviation of the remaining me<strong>as</strong>urements must be recalculated, in order to evaluate the<br />

next me<strong>as</strong>urement.<br />

12


4.5 Rotation between CTS and CIS<br />

4. B<strong>as</strong>ic Theory<br />

The <strong>Ørsted</strong> JPL state vectors are given in the CTS system, which is centered in the Earth’s<br />

center of m<strong>as</strong>s and is rotating with Earth. In this <strong>thesis</strong>, the energy conservation equation<br />

will be derived in the CIS system, which is also centered in the Earth’s center of m<strong>as</strong>s,<br />

but is fixed to some fundamental stars. Therefore a rotation between the two systems is<br />

needed when transforming the JPL state vectors into the ECI state vectors, described in<br />

section 3.<br />

The theory of this section is b<strong>as</strong>ed primarily on [Seeber, 2003]. A rotation of a vector<br />

r from the Earth-fixed system CTS to the space-fixed system CIS, must take into account<br />

both precession, nutation and Earth rotation:<br />

rCIS = P −1 N −1 S −1 rCT S, (28)<br />

where P is the precession matrix, N is the nutation matrix and S is the matrix which<br />

transforms the instantaneously space-fixed system to the Earth-fixed system.<br />

Because of the equatorial bulge on the Earth, the gravitational attraction from the sun<br />

and the moon induces a torque on the rotating Earth. Due to this torque, the Earth’s<br />

rotation axis moves in space with respect to an inertial system. It sweeps out a cone,<br />

centered on the ecliptic pole.<br />

This motion is not smooth, and it can be divided into parts of different periods. The<br />

motion of the longest period (∼26000 years) is called the precession, and it is shown<br />

schematically in figure 4.<br />

Figure 4: Figure showing precession and nutation.<br />

The precession matrix, P, is given by:<br />

13


4. B<strong>as</strong>ic Theory<br />

P = R3(−z)R2(θ)R3(ζ) (29)<br />

The matrices R1, R2 and R3 are the rotation matrices, which rotate around the x-axis,<br />

y-axis and z-axis respectively.<br />

The angles z, θ and ζ are functions of time, T since 01.01.2000 12:00 counted in Julian<br />

centuries:<br />

z = 0. ◦ 6406161T + 0. ◦ 0003041T 2 + 0. ◦ 0000051T 3<br />

θ = 0. ◦ 5567530T − 0. ◦ 0001185T 2 − 0. ◦ 0000116T 3<br />

ζ = 0. ◦ 6406161T + 0. ◦ 0000839T 2 + 0. ◦ 0000050T 3<br />

Besides the precission, there are additional periodic motions of the Earth’s rotation<br />

axis. These are called the nutation, and are also shown in figure 4. The main periods of<br />

the nutation are 13.66 days, half-year, one-year, 9.3 years and 18.6 years. The nutational<br />

motion is caused by the variations in the relative position of the moon, sun and Earth.<br />

The nutation matrix, N, is given by:<br />

with the angles ε, ∆ψ and ∆ε given by:<br />

(30)<br />

N = R1(−ε − ∆ε)R3(−∆ψ)R1(ε), (31)<br />

ε = 23 ◦ 26 ′ 21 ′′ .448 − 46 ′′ .815T − 0 ′′ .00059T 2 − 0 ′′ .001813T 3<br />

∆ψ = −17 ′′ .1996 sin Ω − 1 ′′ .3187 sin(2λM − 2D) − 0 ′′ .2274 sin(2λM) (32)<br />

∆ε = 9 ′′ .2025 cos Ω − 0 ′′ 5736 cos(2λM − 2D) + 0 ′′ .0977 cos(2λM),<br />

where Ω is the mean ecliptic longitude of the lunar <strong>as</strong>cending node, D is the mean<br />

elongation of the moon to the sun and λM is the mean ecliptic longitude of the moon.<br />

The transformation matrix account for the instantaneous polar coordinates xp, yp and<br />

the Greenwich apparent sidereal time GAST :<br />

S = R2(−xp)R1(−yp)R3(GAST ) (33)<br />

14


4.6 Simpson’s Rule of Integration<br />

4. B<strong>as</strong>ic Theory<br />

Simpson’s rule of integration is a method used for estimating integrals numerically<br />

[Kreyszig, 1999].<br />

The function f(x) will be piecewise quadratic approximated, in order to calculate the<br />

numerical approximation, ˜ J, of the integral.<br />

b<br />

a<br />

f(x)dx ≈ ˜ J (34)<br />

For the use of Simpson’s rule, the integration interval x ∈ [a, b] must be divided into<br />

an even number of subintervals, n, of equal length h = (b − a)/n. The points are thereby<br />

given by xr = a + r ∗ h.<br />

Using Simpson’s rule, two sub-intervals are used at the same time. This is why n must<br />

be an even number.<br />

A parabola is drawn through the three points (xr−1, f(xr−1)), (xr, f(xr)), (xr+1, f(xr+1))<br />

defining the two sub-intervals, see figure 5<br />

Figure 5: Simpson’s rule of integration<br />

The area, A, under the parabola, which is highlighted in figure 5, is given by:<br />

A = h<br />

3 [f(xr−1) + 4f(xr) + f(xr+1)]<br />

By adding all of the are<strong>as</strong>, Simpson’s rule of integration of the whole function, h<strong>as</strong> the<br />

form:<br />

˜J = h<br />

3 (f0 + 4f1 + 2f2 + 4f3 + · · · + 2fn−2 + 4fn−1 + fn), (35)<br />

where fr = f(xr).<br />

15


5. Selecting the Data<br />

5 Selecting the Data<br />

The first t<strong>as</strong>k in this project h<strong>as</strong> been to select the <strong>Ørsted</strong> GPS-data, which subsequently<br />

will be used for the modeling of the Earth’s <strong>gravity</strong> field. Since the TurboRogue GPS<br />

receiver is the most accurate of the two GPS receivers on board <strong>Ørsted</strong>, only data from<br />

this receiver will be used in the modeling. The disadvantage of this choice is, that this<br />

receiver only h<strong>as</strong> been turned on for limited periods of time.<br />

5.1 Solar Activity<br />

There are some things that have to be taken into consideration, when selecting the data.<br />

Firstly, one h<strong>as</strong> an interest in eliminating some of the disturbing effects which can influence<br />

the later calculations, <strong>as</strong> for example the solar activity. Variations in the solar<br />

activity induce changes in the atmospheric density, which again result in changes in the<br />

deacceleration of the <strong>satellite</strong> due to the incre<strong>as</strong>ed friction.<br />

Hence, by choosing to work with data from a period of time, in which the solar activity<br />

does not vary significantly in magnitude, the estimate of the non-gravitational accelerations<br />

acting on <strong>Ørsted</strong>, will be simplified.<br />

For the purpose of selecting a period of data, where the solar activity h<strong>as</strong> been somewhat<br />

stable, the radio flux h<strong>as</strong> been plotted <strong>as</strong> a function of time since <strong>Ørsted</strong> w<strong>as</strong> launched in<br />

1999, see figure 6.<br />

Solar flares generate electron beams, which again generate electromagnetic waves which<br />

can be observed on Earth. These electromagnetic waves are called the radio flux, and it<br />

is a good indicator of the magnitude of the solar activity [Kivelson and Russell, 1997]. In<br />

figure 6, the red line shows the variations of the radio flux, and the green and blue marks<br />

indicate which days, the TurboRogue GPS-receiver on board <strong>Ørsted</strong>, h<strong>as</strong> been turned on.<br />

The period of data, which runs from 1999-09-02 to 2000-02-22, h<strong>as</strong> been chosen for the<br />

further calculations, because it is a period of time in which the solar activity w<strong>as</strong> rather<br />

stable, and at the same time it contains many TurboRogue GPS-me<strong>as</strong>urements not too far<br />

spaced in time. In figure 6 the period of data, that h<strong>as</strong> been selected, is marked with blue.<br />

5.2 Evaluation of the GPS Data<br />

Originally the chosen period of time contained more than 400.000 TurboRogue GPSme<strong>as</strong>urements.<br />

It can be <strong>as</strong>sumed that they are not all equally accurate. Unfortunately<br />

there are no direct uncertainty estimates, and another parameter therefore must be found<br />

to evaluate the data.<br />

For this purpose, it would be preferable to know the PDOP values of every single GPS<br />

positioning, since these could be used for weighting the individual me<strong>as</strong>urements in the<br />

modeling. The PDOP value is an indicator of the geometry of the GPS-<strong>satellite</strong> constellation<br />

used to determine the position, and it is given by:<br />

16


Figure 6: Solar activity indicated by the radio flux.<br />

17<br />

5. Selecting the Data


5. Selecting the Data<br />

a b<br />

Figure 7: (a) The position of the <strong>satellite</strong> in the X-Y plane before the filtering. (b) After<br />

the filtering<br />

P DOP = Vmax<br />

, (36)<br />

V<br />

where V is the volume formed by the <strong>satellite</strong> constellation and Vmax is the largest<br />

volume possible expanded by the <strong>satellite</strong> constellation ([Seeber, 2003], p.301).<br />

Unfortunately the PDOP values for each me<strong>as</strong>urement are not given, and they would<br />

be both difficult and time consuming to calculate. Therefore another parameter must be<br />

used in the determination of whether a me<strong>as</strong>urement is good (accurate enough) or not.<br />

The accuracy of a GPS-me<strong>as</strong>urement depends strongly on the number of <strong>satellite</strong>s used<br />

in the positioning, and therefore this parameter will be used in the selection of data. The<br />

choice is made, that all me<strong>as</strong>urements b<strong>as</strong>ed on signals from 6 or more GPS <strong>satellite</strong>s will<br />

be considered <strong>as</strong> good me<strong>as</strong>urements.<br />

The program rinex p<strong>as</strong>s2.f90, see appendix C, h<strong>as</strong> been developed for the purpose<br />

of filtering the GPS-data. The output of this program is a file containing only data<br />

from the GPS-me<strong>as</strong>urements b<strong>as</strong>ed on 6 or more <strong>satellite</strong>s. Figure 7 shows the effect<br />

of rinex p<strong>as</strong>s2.f90, by showing the data points of the <strong>satellite</strong> for one day in the X-Y plane<br />

both before and after the filtering. It is seen that the filtering leaves ”holes” in the orbit<br />

of the <strong>satellite</strong>.<br />

The chosen period of time contains 254463 good GPS-me<strong>as</strong>urements. The distribution<br />

of good me<strong>as</strong>urements in the chosen period, can be seen in figure 8.<br />

5.3 Ground Track of the Satellite<br />

It must be ensured that the ground track of the <strong>satellite</strong>, in the chosen period of time, covers<br />

the whole surface of the Earth. This must be fulfilled, in order to derive a satisfactory<br />

18


Figure 8: Distribution of good data in the chosen period.<br />

5. Selecting the Data<br />

model, which describes the complete <strong>gravity</strong> potential field of the Earth, without any<br />

”holes”. The ground track of the <strong>Ørsted</strong> <strong>satellite</strong>, in the selected period, is seen in figure 9.<br />

It is seen that the ground track covers the Earth satisfactory except for the polar regions.<br />

The polar gaps are a result of the inclination of the <strong>satellite</strong>.<br />

Now the data, which will be used in the modeling of the Earth’s <strong>gravity</strong> field, h<strong>as</strong> been<br />

selected, b<strong>as</strong>ed on considerations of the solar activity, accuracy of the positioning and the<br />

coverage of the ground track of the <strong>satellite</strong>.<br />

Figure 9: The ground track of <strong>Ørsted</strong> in the chosen period of time.<br />

19


6. Gravity Field Determination<br />

6 Gravity Field Determination<br />

There are many different ways of determining the Earth’s <strong>gravity</strong> field by <strong>satellite</strong>. In this<br />

section three different <strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>s, using different me<strong>as</strong>uring techniques will<br />

be described. This is done to clarify the limitations and possibilities of <strong>Ørsted</strong> <strong>as</strong> a <strong>gravity</strong><br />

<strong>mission</strong> <strong>satellite</strong>.<br />

The result of a <strong>gravity</strong> field determination can be represented in many different ways.<br />

The aim is to produce a result which also can be used <strong>as</strong> a tool in other projects. Therefore<br />

the goal of this <strong>thesis</strong> is not only to calculate the values of the <strong>gravity</strong> potential from<br />

the <strong>Ørsted</strong> data, but rather to be able to determine a global model representing the results.<br />

The global <strong>gravity</strong> model determined in this project, will be expressed by a spherical<br />

harmonic expansion of the gravitational potential field. This type of global model h<strong>as</strong><br />

been chosen because of the advantages of spherical harmonics, when describing a function<br />

on a sphere, see also section 4.1. Furthermore the models widely used today, such <strong>as</strong><br />

EGM96 [Lemoine et al., 1998], are spherical harmonic series representing the gravitational<br />

potential field. By choosing this kind of representation, the results found in this <strong>thesis</strong> can<br />

and will be directly compared to EGM96.<br />

Figure shows the EGM96 model to degree and order 360, represented by the geoid<br />

heights.<br />

Figure 10: The geoid model EGM96.<br />

Many different parameters can be used to represent the Earth’s <strong>gravity</strong> field. For example<br />

both the gravitational potential, the geoid height and the gravitational acceleration<br />

represents the field, and these parameters can be derived from one another. The choice of<br />

representation depends on the purpose of the modelling. In this <strong>thesis</strong>, the <strong>gravity</strong> field<br />

will be represented by the gravitational potential field.<br />

Contrary to <strong>Ørsted</strong>, the <strong>satellite</strong>s CHAMP, GRACE and GOCE are dedicated <strong>gravity</strong><br />

<strong>mission</strong> <strong>satellite</strong>s. They have different characteristics, but they are designed to complement<br />

each other in the goal of refining the knowledge of the Earths <strong>gravity</strong> field. This is<br />

20


6. Gravity Field Determination<br />

seen in figure 11, which shows the expected geoid errors of the harmonic coefficients of the<br />

<strong>gravity</strong> model.<br />

Figure 11: Geoid error estimates on the spherical harmonic coefficients. It is seen that<br />

the three <strong>mission</strong>s complement each other. ([Seeber, 2003] fig. 10.3)<br />

6.1 Dedicated Gravity Mission Satellites<br />

6.1.1 CHAMP<br />

The CHAMP (Challenging Mini-Satellite Payload) <strong>satellite</strong> is, like the <strong>Ørsted</strong> <strong>satellite</strong>,<br />

part of a SST-HL constellation, see section 2.2. [Seeber, 2003]<br />

CHAMP w<strong>as</strong> launched in July 2000 into a near polar orbit of height 450 km. The<br />

<strong>satellite</strong> will drop to a height of approximately 300 km during a period of about 5 years<br />

due to external forces, such <strong>as</strong> friction. This decre<strong>as</strong>e in height makes the <strong>satellite</strong> able to<br />

me<strong>as</strong>ure signal of varying wavelength throughout it’s lifetime.<br />

CHAMP acts like a probe in the Earth’s gravitational field, which orbit perturbations<br />

are registered by the GPS <strong>satellite</strong>s. On board the <strong>satellite</strong> are accelerometers me<strong>as</strong>uring<br />

the non-gravitational accelerations of the <strong>satellite</strong>, in order to extract these disturbances<br />

from the signal. The <strong>satellite</strong> is seen in figure 12. For further information on the CHAMP<br />

<strong>satellite</strong>, see http://www.gfz-potsdam.de/pb1/op/champ/index CHAMP.html<br />

6.1.2 GRACE<br />

The two GRACE (Gravity Recovery and Climate Experiment) <strong>satellite</strong>s are identical and<br />

are orbiting the Earth at an altitude of about 500 km in a near polar orbit [Seeber, 2003].<br />

The <strong>satellite</strong>s are seen in figure 13.<br />

The distance between the two <strong>satellite</strong>s is approximately 220 km. The GRACE <strong>mission</strong><br />

is using the concept of <strong>satellite</strong>-to-<strong>satellite</strong> tracking in low-low mode (SST-LL). Each<br />

21


6. Gravity Field Determination<br />

Figure 12: The CHAMP <strong>satellite</strong> - part of a SST-HL mode constellation.<br />

<strong>satellite</strong> transmits signals, which are received by the other <strong>satellite</strong>, in order to obtain a<br />

very precise determination of the relative motion of the <strong>satellite</strong>s.<br />

The goal of SST in general is to setup equations connecting the coefficients of the<br />

terrestrial <strong>gravity</strong> field, Cnm and Snm, with the observed radial velocity range rate, ˙ρ, and<br />

the range rate change, ¨ρ. The parameters ˙ρ and ¨ρ are not actually me<strong>as</strong>ured, but are the<br />

result of the data processing.<br />

The relative motion is caused by the different accelerations of the two <strong>satellite</strong>s, caused<br />

by the different gravitational attraction from the Earth, and from non-gravitational accelerations.<br />

As with the CHAMP <strong>satellite</strong>, the non-gravitational accelerations of the GRACE<br />

<strong>satellite</strong>s are me<strong>as</strong>ured by 3D accelerometers placed in the centre of m<strong>as</strong>s of both <strong>satellite</strong>s.<br />

These me<strong>as</strong>urements are done in order to separate the gravitational and non-gravitational<br />

disturbances and thereby determine the <strong>gravity</strong> field of the earth. For more information<br />

on the GRACE <strong>mission</strong> <strong>satellite</strong>s, see for example: http://www.csr.utex<strong>as</strong>.edu/grace/.<br />

Figure 13: The GRACE <strong>satellite</strong> - part of a SST-LL mode constellation.<br />

22


6.1.3 GOCE<br />

6. Gravity Field Determination<br />

The GOCE (Gravity Field and Steady-State Ocean Circulation Explorer) <strong>satellite</strong> is planned<br />

for launch in 2006. The <strong>satellite</strong> is seen in figure 14. The principles of this <strong>gravity</strong> <strong>mission</strong><br />

is GPS SST-HL and <strong>satellite</strong> <strong>gravity</strong> gradiometry (SGG). The <strong>satellite</strong> will be send into<br />

a near polar orbit of a height of approximately 250 km. A <strong>gravity</strong> gradiometer on board<br />

GOCE provides two me<strong>as</strong>urements of the accelerations in every direction with a distance<br />

of about 50 cm, from which the difference in acceleration can be derived.<br />

The advantage of the SGG method compared to SST-HL and SST-LL is the fact that<br />

the non-gravitational accelerations are the same for all the accelerometers since they are<br />

placed in the same spacecraft. Therefore they cancel out when finding the acceleration<br />

differences. With SGG, the second derivatives of the gravitational potential, or <strong>gravity</strong><br />

gradients, are determined.<br />

The <strong>satellite</strong> is designed to be drag free, meaning that the cross section area in the<br />

direction of maximum drag is minimised. There is an attitude and drag controller on board<br />

the <strong>satellite</strong>, and the <strong>satellite</strong>s attitude will be corrected continuously using thrusters. The<br />

correction demands fuel, and therefore the lifetime of the <strong>satellite</strong> will be relatively short,<br />

namely about 2 years. As seen in figure 11, GOCE is expected to refine the higher order<br />

coefficients of the gravitational potential field.<br />

The GOCE <strong>mission</strong> is described in detail at:<br />

http://www.esa.int/esaLP/ESA1MK1VMOC LPgoce 0.html.<br />

Figure 14: The GOCE <strong>satellite</strong>.<br />

6.2 <strong>Ørsted</strong> <strong>as</strong> a Gravity Mission Satellite.<br />

When comparing <strong>Ørsted</strong> with GRACE, CHAMP and GOCE, it is clear that it is most<br />

comparable with CHAMP, mainly because they are both part of a SST-HL constellation.<br />

There are several differences between CHAMP and <strong>Ørsted</strong> though, which make the <strong>gravity</strong><br />

derivations b<strong>as</strong>ed on <strong>Ørsted</strong> data more complicated than the CHAMP derivations. There<br />

are three important criteria which a <strong>satellite</strong> must fulfil, in order to qualify <strong>as</strong> a good<br />

23


6. Gravity Field Determination<br />

<strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>, and these are stated below. ([Seeber, 2003] p. 471)<br />

-The <strong>satellite</strong> must have a relatively low orbital height (200-500 km).<br />

-The GPS-tracking of the <strong>satellite</strong> must be uninterrupted.<br />

-There must be instruments on board, which make it possible to separate the gravitational<br />

and non-gravitational accelerations.<br />

<strong>Ørsted</strong> does not fully satisfy the first criteria, since the apogee altitude of <strong>Ørsted</strong> is<br />

about 850 km and the perigee altitude is approximately 450 km. The criteria stems from<br />

the fact that the gravitational acceleration of a <strong>satellite</strong> caused by a m<strong>as</strong>s anomaly in the<br />

Earth, decre<strong>as</strong>es with the square of the incre<strong>as</strong>ing orbital height, r. Furthermore the term<br />

A = ( ae<br />

r )n , describes the incre<strong>as</strong>ing attenuation of the gravitational field with incre<strong>as</strong>ing<br />

orbital height. As a result of this, high orbiting <strong>satellite</strong>s will only be able to me<strong>as</strong>ure<br />

the components of the field, which have long wavelengths. Because of this, the <strong>Ørsted</strong><br />

data is expected only to reveal the low degree harmonic coefficients, if any at all, of the<br />

gravitational field.<br />

The TurboRogue GPS-tracking cannot fully fulfil the second criteria, since it is only<br />

turned on for limited periods of time <strong>as</strong> it is seen in section 5.2. Though this is compensated<br />

for by the fact that data from a long period of time will be used in further calculations. In<br />

this way a satisfactory coverage can be obtained, in spite of the interrupted GPS-tracking.<br />

The l<strong>as</strong>t criteria is not fulfilled by <strong>Ørsted</strong> either, since there is no accelerometer on<br />

board, to me<strong>as</strong>ure the non-gravitational accelerations. Therefore a method, in which these<br />

accelerations can be modelled and separated from the gravitational accelerations, must<br />

be used in this <strong>thesis</strong>, in order to be able to convert the state vectors into gravitational<br />

potential.<br />

One advantage of the <strong>Ørsted</strong> <strong>satellite</strong>, when evaluating it <strong>as</strong> a <strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>,<br />

is that it is orbiting the Earth in a near polar orbit, see section 2. This results in a very<br />

good ground track coverage of the Earth, which is also clearly seen in figure 9 where it is<br />

also seen that the polar gaps are of a limited size.<br />

All in all, it must be concluded that <strong>Ørsted</strong> h<strong>as</strong> many limitations <strong>as</strong> a <strong>gravity</strong> <strong>mission</strong><br />

<strong>satellite</strong>, and it will be interesting to see the level of accuracy which can be achieved.<br />

24


7. Methods for Gravity Field Determination from State Vectors Only<br />

7 Methods for Gravity Field Determination from State<br />

Vectors Only<br />

The only available data from <strong>Ørsted</strong> are the state vectors, and this is the re<strong>as</strong>on why<br />

the energy conservation method h<strong>as</strong> been chosen for the calculations of the gravitational<br />

potential in this project. Hence the kinetic energy can be directly calculated from the<br />

state vector. In the following, the theory of the energy conservation theorem will be fully<br />

explained.<br />

Another way of calculating the gravitational potential field only from state vectors is<br />

the acceleration method. This method is also described shortly subsequently, although it<br />

h<strong>as</strong> not been used in this <strong>thesis</strong>.<br />

7.1 The Acceleration Method<br />

When using the acceleration method, the accelerations, Ā, of the <strong>satellite</strong> are calculated<br />

from the state vectors, by deriving the first order derivatives of the velocity, ¯v, or the<br />

second order derivatives of the position, ¯x.<br />

Ā = d¯v<br />

dt = d2¯x dt2 (37)<br />

In order to calculate the first and second order derivatives from the state vectors, these<br />

must be changed from a point-wise function into a continuous function. This can for<br />

example be done by the use of Newton interpolation [Abt, 2004], [Reubelt et al., 2003],<br />

[Ditmar and van Eck van der Sluijs, 2004].<br />

When the interpolation is executed, the accelerations of the <strong>satellite</strong> can be calculated<br />

directly from the continuous state vector function. The acceleration found is the total<br />

acceleration of the <strong>satellite</strong>, and from this the tidal accelerations and non gravitational<br />

acceleration must be subtracted in order to isolate the accelerations only caused by the<br />

Earth’s gravitational pull.<br />

Agravitational = A − Anon−gravitational − Atidal<br />

The acceleration method can for example be used in combination with Le<strong>as</strong>t squares<br />

collocation, to derive a global model of the <strong>gravity</strong> field.<br />

The main re<strong>as</strong>on why the acceleration method h<strong>as</strong> not been used in this <strong>thesis</strong>, is that<br />

the method w<strong>as</strong> not fully developed or tested, when this project w<strong>as</strong> initiated.<br />

Furthermore the acceleration method is more sensitive to noise in the state vector, than<br />

the energy conservation method is [Abt, 2004].<br />

25<br />

(38)


7. Methods for Gravity Field Determination from State Vectors Only<br />

7.2 The Energy Conservation Theorem<br />

The energy conservation method h<strong>as</strong> been chosen for the calculation of the Earths gravitational<br />

potential field in this <strong>thesis</strong>.<br />

In this section, the b<strong>as</strong>ic ide<strong>as</strong> and considerations behind the energy conservation<br />

method will be explained. As mentioned earlier, the energy conservation method h<strong>as</strong><br />

been chosen for this project, because the kinetic energy can be directly calculated from the<br />

state vectors [Visser et al., 2003], [Gerlach et al., 2003b].<br />

One of the most b<strong>as</strong>ic theorems of physics is the energy conservation theorem<br />

[Knudsen and Hjorth, 1996].<br />

The mechanical energy of a particle is<br />

conserved for motion in a conservative<br />

force field.<br />

This theorem does also apply on the <strong>Ørsted</strong> <strong>satellite</strong> in the Earth centered inertial coordinate<br />

system CIS. If it is <strong>as</strong>sumed that the only force acting on <strong>Ørsted</strong> is the gravitational<br />

force from the Earth, the energy conservation theorem will have the simple form:<br />

Ekin + Epot = constant, (39)<br />

where Ekin is the kinetic energy of <strong>Ørsted</strong> and Epot is the potential energy. Since the<br />

potential energy is caused only by the Earth’s gravitational pull, it will subsequently be<br />

referred to <strong>as</strong> Eearth.<br />

Equivalently, the latter equation may be expressed in terms of potentials instead of<br />

energy, namely by dividing each term with the m<strong>as</strong>s, m, of the <strong>satellite</strong>.<br />

Ekin<br />

m<br />

− Eearth<br />

m = Vkin − Vearth = E0, (40)<br />

where E0 is a constant, Vkin is the kinetic potential and Vearth is the Earth’s gravitational<br />

potential.<br />

The <strong>satellite</strong> and the Earth can not be considered an isolated system though, since<br />

there are other forces than the gravitational force from the Earth acting on <strong>Ørsted</strong>, and<br />

these must also be present in the energy conservation equation.<br />

7.2.1 Tidal Gravitational Disturbances Caused by the Sun and Moon<br />

The sun and the moon induce tidal gravitational disturbances, Vsun and Vmoon, on the<br />

Earth. These disturbances generate a movement of m<strong>as</strong>ses, which again influence the<br />

<strong>satellite</strong>, and this effect must also be present in the energy equation.<br />

In principle the tidal effects caused by the other planets in the solar system, must also<br />

be considered. Though the gravitational tidal influences generated by these planets are of<br />

such a small size, that they are <strong>as</strong>sumed negligible. When including the disturbing solid<br />

26


7. Methods for Gravity Field Determination from State Vectors Only<br />

Figure 15: Figure showing angles and distances used in equation 42 and 43.<br />

Earth tides in the energy conservation equation, and expressing it in terms of potentials,<br />

it h<strong>as</strong> the form:<br />

Vkin + Vearth + Vsun + Vmoon = E0<br />

The size of the solid Earth tidal potential caused by the sun and the moon are given<br />

by equation 19 and equation 20 respectively [Seeber, 2003]:<br />

Vsun = − k2<br />

2<br />

Vmoon = − k2<br />

2<br />

Gmsun<br />

r 3 sun<br />

Gmmoon<br />

r 3 moon<br />

(41)<br />

a 5 e<br />

r 3 ((3 − 15 cos(θsun) 2 ) + 6 cos(θsun)) (42)<br />

a 5 e<br />

r 3 ((3 − 15 cos(θmoon) 2 ) + 6 cos(θmoon)), (43)<br />

where k2 is the Love number which describes the el<strong>as</strong>ticity of the Earth, ae is the<br />

equatorial radius of the Earth, msun is the m<strong>as</strong>s of the sun and mmoon is the m<strong>as</strong>s of the<br />

moon. The distances and angles, used in the equations, are shown in figure 15.<br />

7.2.2 Rotational Potential<br />

In this <strong>thesis</strong>, the calculations considering the energy conservation will be executed in<br />

the conventional inertial system, CIS. In this system, the gravitational force field is not a<br />

conservative force field due to the rotation of the Earth.<br />

Since the energy conservation theorem only applies in a conservative force field, a rotational<br />

potential, Vrot, must be included in the energy conservation equation, to compensate<br />

for this.<br />

Vkin + Vearth + Vsun + Vmoon + Vrot = E0<br />

27<br />

(44)


7. Methods for Gravity Field Determination from State Vectors Only<br />

The rotational potential is derived in [Jekeli, 1999], and h<strong>as</strong> the form:<br />

Vrot = −ω(xvy − yvx), (45)<br />

−5 rad<br />

where ω = 7.292115 · 10 is the rotational velocity of the Earth.<br />

s<br />

7.2.3 Non-Gravitational Disturbing Potential<br />

There are also non-gravitational forces, acting on the <strong>satellite</strong> in its orbit. One of these<br />

forces is the friction force, caused by particles in the atmosphere de-accelerating the <strong>satellite</strong>.<br />

The friction force is dependent on many different parameters such <strong>as</strong> solar activity,<br />

the composition of the atmosphere, the velocity, altitude, geometry and attitude of the<br />

<strong>satellite</strong>.<br />

Including the non-gravitational disturbing potential, F , in the energy conservation<br />

theorem, it now h<strong>as</strong> the form:<br />

Vkin + Vearth + Vsun + Vmoon + Vrot + F = E0<br />

Theoretically, the non-gravitational potential can be calculated by [Howe et al., 2003]:<br />

<br />

F = (v · a)dt, (47)<br />

where v is the velocity vector and a is the non-gravitational acceleration vector of the<br />

<strong>satellite</strong>. But since the non-gravitational acceleration of <strong>Ørsted</strong> is not known, another<br />

approach to the determination of F must be found. This approach will be explained in<br />

section 8.2.<br />

If the latter equation is <strong>as</strong>sumed to be the complete energy conservation equation, it<br />

is possible to isolate and calculate the Earths gravitational potential, Vearth, if the other<br />

terms in the equation can be calculated.<br />

−Vearth = Vkin + Vsun + Vmoon + Vrot + F − E0<br />

This equation shows the b<strong>as</strong>ic idea behind calculating the Earths gravitational potential,<br />

using the energy conservation theorem.<br />

28<br />

(46)<br />

(48)


8 Deriving The Gravitational Potential<br />

8. Deriving The Gravitational Potential<br />

One of the most important t<strong>as</strong>ks in this project is, to derive the gravitational potential<br />

from the state vectors, in every single GPS me<strong>as</strong>uring point. It h<strong>as</strong> proved to be a process,<br />

which demands many different considerations. Given are the state vectors in both CIS and<br />

CTS coordinates, and from these the Earth’s <strong>gravity</strong> potential must be derived.<br />

For this purpose the program stat2pot2 h<strong>as</strong> been developed. It is b<strong>as</strong>ed on the program<br />

stat2pot programmed by C.C.Tscherning (personal communication) for the work<br />

with different dedicated <strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>s. The most significant difference between<br />

stat2pot and stat2pot2 is the fact that the non-gravitational accelerations of <strong>Ørsted</strong> are<br />

not known, and therefore must be calculated in stat2pot2, where<strong>as</strong> they are used <strong>as</strong> input<br />

in stat2pot.<br />

The gravitational potential, Vearth, is calculated using the conservation of energy theorem,<br />

which is explained in section 7.2. The energy conservation method h<strong>as</strong> also been used<br />

by others for <strong>gravity</strong> field modeling [Visser et al., 2003], [Gerlach et al., 2003b], [Gerlach et al., 2003a],<br />

[Howe, 2005]. In this section, the computation of each term in the theorem is explained<br />

and all of the considerations, related to these computations, are likewise described. All<br />

of the calculations, described in this section, are executed by stat2pot2. A flow chart of<br />

stat2pot2 is shown in figure 16 and the program can be seen in appendix C.<br />

As mentioned in section 7.2, the coordinate system chosen for the calculations is the<br />

CIS, and in this inertial system the conservation of energy theorem h<strong>as</strong> the form:<br />

Vearth = Vkin − Vrot − Vsun − Vmoon − E0 + F (49)<br />

In the latter equation, the signs of every term h<strong>as</strong> been incorporated. Hence the difference<br />

between this equation and equation 48. This h<strong>as</strong> been done to be able to work with<br />

Vearth <strong>as</strong> a positive term.<br />

The first thing that is done in stat2pot2, is to open and read the ECI and JPL files.<br />

Therefore this is described <strong>as</strong> activity 1 in the flowchart of stat2pot2.f90 in figure 16.<br />

8.1 Calculation of Vkin, Vrot, Vsun and Vmoon<br />

The kinetic energy per m<strong>as</strong>s, Vkin, of the <strong>satellite</strong> in the CIS coordinate system is directly<br />

calculated from the CIS state vector by:<br />

Vkin = 1<br />

2 v2 CIS = 1<br />

2 (v2 x + v 2 y + v 2 z)CIS<br />

This calculation is executed <strong>as</strong> activity 2 in the flowchart of stat2pot2.f90. The kinetic<br />

potential is plotted <strong>as</strong> a function of time in figure 17. Figure 17(a) shows the whole period<br />

of time, and figure 17(b) shows a time period corresponding approximately to one orbit<br />

(∼100 minutes).<br />

29<br />

(50)


8. Deriving The Gravitational Potential<br />

[irec=irec+1]<br />

Start program stat2pot2.f90<br />

[Start counting : irec=1]<br />

1:Open and read line "irec" of the inputfiles<br />

2: Calculate kinetic energy per m<strong>as</strong>s<br />

3: Calculate rotational potential<br />

4: Define referencefield: normalpotential<br />

5: Calculate tidal potential from Sun and Moon<br />

6: Calculate E0-F<br />

7: Write position and the terms calculated in 3-7 to file<br />

[if not end of file]<br />

[if end of file]<br />

8: Finds linear regression on (E0-F)(dt) and subtract the line found<br />

9: Use Chauvenet's criterion to filter the data from 8<br />

10: The line found in 8 is added to (E0-F)(dt) again<br />

11: Linear regression is done on the filtered (E0-F)(dt) to determine E0 and F<br />

12: Calculate gravitational potential from every data point<br />

Diagram: stat2pot2 Page 1<br />

13: Write position and gravitational potential to file<br />

End program<br />

Input files are the<br />

JPL and ECI files<br />

Filename :<br />

Vjord.out<br />

Filename:<br />

collodata.inp<br />

Figure 16: Flowchart of the program stat2pot2.f90<br />

30


8. Deriving The Gravitational Potential<br />

(a) (b)<br />

Figure 17: The kinetic energy per m<strong>as</strong>s <strong>as</strong> a function of time, Vkin(t). (a) The entire<br />

period of time. (b) Approximately one orbit.<br />

The rotational potential, Vrot, is also calculated from each GPS-me<strong>as</strong>urement directly<br />

from the CIS state vector by [Jekeli, 1999]:<br />

Vrot = ω(xvy − yvx)CIS<br />

This is done <strong>as</strong> activity number 3 in stat2pot2.f90, In figure 18, the rotational potential<br />

<strong>as</strong> a function of time is shown.<br />

The disturbing potential of the solid Earth tidal effects caused by the Sun, Vsun, and<br />

the Moon, Vmoon, are functions of both distance, latitude and longitude of the <strong>satellite</strong> in<br />

each me<strong>as</strong>uring point [Seeber, 2003]. This potential is calculated in activity 5, figure 16.<br />

Vsun/moon = k2<br />

2<br />

Gmsun/moon<br />

r 3 sun/moon<br />

(51)<br />

a 5 e<br />

r 3 ((3 − 15 cos(θsun/moon) 2 ) + 6 cos(θsun/moon)), (52)<br />

Vsun and Vmoon are plotted <strong>as</strong> functions of time in figure 19 and 20 respectively.<br />

The minimum, maximum and mean values of the four terms ,Vkin, Vrot, Vsun and Vmoon<br />

are shown in table 1. The values in table 1 give an impression of the order of magnitude<br />

of the individual terms.<br />

It is seen that the kinetic energy per m<strong>as</strong>s and the rotational potential are the most<br />

dominant by far. The small size of the tidal disturbances caused by the sun and the moon,<br />

indicate that it in fact w<strong>as</strong> legitimate, to discard the tidal potential disturbances caused<br />

by other planets, <strong>as</strong> stated in section 7.2.1, since these terms would be even smaller and<br />

insignificant.<br />

31


8. Deriving The Gravitational Potential<br />

(a) (b)<br />

Figure 18: The rotational potential <strong>as</strong> a function of time, Vrot(t) for (a) the whole period<br />

of time and (b) approximately one orbit.<br />

(a) (b)<br />

Figure 19: The tidal potential caused by the Sun, <strong>as</strong> a function of time, Vsun(t). (a) For<br />

the whole period of time and (b) for approximately one orbit.<br />

32


8. Deriving The Gravitational Potential<br />

(a) (b)<br />

Figure 20: The tidal potential caused by the Moon, <strong>as</strong> a function of time, Vmoon(t). (a)<br />

The entire time period and (b) approximately one orbit.<br />

Term min. value max. value mean value<br />

Vkin 26989069.76 28837625.66 27899061.48<br />

Vrot -441914.88 -434234.46 -438169.90<br />

Vsun -3.88 7.74 -0.57<br />

Vmoon -7.02 14.48 -1.64<br />

Table 1: Max.-, min.- and mean values of Vkin, Vrot, Vsun and Vmoon. Units are [ m2<br />

s s ].<br />

33


8. Deriving The Gravitational Potential<br />

Left is now to determine the constant E0 and the non-gravitational potential, F , in<br />

order to finally be able to calculate the Earth’s <strong>gravity</strong> potential Vearth from every GPS<br />

data point. The equation of energy conservation, eq. 49, is rewritten with the known and<br />

the unknown parameters separated:<br />

8.2 Determination of E0 and F<br />

E0 − F + Vearth = Vkin − Vrot − Vsun − Vmoon<br />

As stated in the preface, one of the major challenges of this <strong>thesis</strong> is to find a valid way to<br />

calculate the non-gravitational potential, F , since the non-nn-gravitational accelerations of<br />

the <strong>satellite</strong> are not known. The fact that the friction force acting on the <strong>satellite</strong> depends<br />

on many different parameters, <strong>as</strong> stated in section 7.2.3, makes it difficult to set up a model<br />

describing the force.<br />

There are several different methods of estimating the non-gravitational potential without<br />

directly modeling it. The way that E0 and F are determined in this <strong>thesis</strong>, will be<br />

explained in this section.<br />

In order to estimate and examine the non-gravitational potential of the <strong>satellite</strong> <strong>as</strong><br />

a function of time, the <strong>gravity</strong> potential of the earth, Vearth, must be known according<br />

to equation 53. Since this is obviously not the c<strong>as</strong>e, it is approximated with an already<br />

known reference potential field. The simpler the reference field is, the better, since one of<br />

the goals of this <strong>thesis</strong> is to use <strong>as</strong> little a priori knowledge <strong>as</strong> possible, in the computation<br />

of the Earths <strong>gravity</strong> field. But since it is not possible to estimate F and E0 without any<br />

knowledge of Vearth, some a priori <strong>gravity</strong> potential field, Vearth,ref., is necessary.<br />

E0 − F = Vkin − Vrot − Vsun − Vmoon − Vearth,ref.<br />

With the aim of keeping the a priori information at a minimum, the <strong>gravity</strong> potential<br />

field is, <strong>as</strong> a start, approximated by:<br />

Vearth,ref1 = GM<br />

r =<br />

GM<br />

(x 2 + y 2 + z 2 )<br />

When this reference field is calculated, and subtracted, in every me<strong>as</strong>uring point, E0−F<br />

can be examined, see figure 21.<br />

Here E0 − F is shown <strong>as</strong> a function of longitude and latitude, and the large bulge near<br />

equator indicates that the first approximation, Vearth,ref1., of the <strong>gravity</strong> potential w<strong>as</strong> too<br />

simple.<br />

Intuitively the values of E0 − F should be almost constant, except from a small decre<strong>as</strong>e in<br />

time, due to the non-gravitational forces acting on the <strong>satellite</strong>. The bulge is a consequence<br />

of the fact that the ellipsoidal form of the earth w<strong>as</strong> not taken into account in the chosen<br />

reference potential field Vearth,ref1.. B<strong>as</strong>ed on this observation it is clear that a new reference<br />

field, containing more information about the earth’s <strong>gravity</strong> field must be used. The first<br />

34<br />

(53)<br />

(54)<br />

(55)


8. Deriving The Gravitational Potential<br />

Figure 21: (E0 − F ) calculated, using Vearth,ref1. = GM<br />

r<br />

4 terms of the normal potential, without the centrifugal term, is used <strong>as</strong> the reference<br />

potential instead. See section 4.3, [Torge, 1999].<br />

Vearth,ref2. = GM<br />

(1 +<br />

r<br />

3<br />

n=1<br />

( ae<br />

r )2n J2nP2n(cosθ)), (56)<br />

where ae = 6378136.4m is the Earth’s mean equatorial radius. This reference field is<br />

defined in activity 4 in the stat2pot2 flowchart.<br />

When the latter reference field is calculated and subtracted in every me<strong>as</strong>uring point,<br />

E0 − F can be examined again, see figure 22.<br />

In figure 22 it is seen that E0 − F calculated from Vearth,ref2. is almost constant all over<br />

the earth, which indicates that the chosen reference field contains sufficient information.<br />

B<strong>as</strong>ed on these considerations, Vearth,ref2. will be used <strong>as</strong> the reference field in the further<br />

calculations. E0 − F is calculated <strong>as</strong> activity 6 in stat2pot2, and it is plotted <strong>as</strong> a function<br />

of time in figure 23.<br />

When looking at figure 23, the overall trend is clear. Even though there are of course<br />

daily variations in the friction potential due to changes in the atmosphere’s density etc, the<br />

non-gravitational potential, F , is decre<strong>as</strong>ing at an almost constant rate in time when looking<br />

at the dominating trend during the chosen period of time. B<strong>as</strong>ed on this observation,<br />

the friction is <strong>as</strong>sumed linear in time, meaning that it can be written <strong>as</strong>:<br />

F = A · t, (57)<br />

35


8. Deriving The Gravitational Potential<br />

Figure 22: (E0 − F ) calculated, using the normal potential <strong>as</strong> reference field, Vearth,ref2..<br />

Figure 23: (E0 − F )(dt)<br />

36


8. Deriving The Gravitational Potential<br />

where A is a constant and t is the time. This is clearly an approximation, and the<br />

approach is chosen mainly because of its apparent simplicity. The later result must tell<br />

whether or not it is a valid one. If it is shown not to be valid, an other approach will be<br />

used.<br />

When (E0 − F )(dt) is found, linear regression is used to find the best fit of the function<br />

to a straight line, and according to the following statement, E0 is found <strong>as</strong> the crossing of<br />

the straight line with the y-axis:<br />

dt = t − t0<br />

(E0 − F )(t0) = E0(t0) − F (t0) = E0(t0) − 0 = E0, (58)<br />

due to the fact that F is <strong>as</strong>sumed linear in time and zero at t0<br />

Before fitting (E0 − F )(dt) by linear regression, some considerations must be made.<br />

It is clearly seen in figure 21 - 23 that some of the (E0 − F ) values are so extreme, that<br />

they most likely are a result of an error in the GPS positioning. Therefore, a criterion<br />

to exclude extreme error-b<strong>as</strong>ed data must be stated, since the deviant data points will<br />

wrongly influence the linear regression and thereby also the determination of E0 and F .<br />

Chauvenets criterion is used to filter out very deviant or error-b<strong>as</strong>ed data. Chauvenets<br />

criterion is described in section 4.4. In order to use Chauvenets criterion on a data set, the<br />

data set must have mean 0. In order to fulfil this requirement, linear regression is used to<br />

fit the data shown i figure 23 by a straight line, a ∗ t + b, and the value of the found line is<br />

then subtracted in every data point.<br />

C(t) = (E0 − F )(t) − (a · t + b) (59)<br />

Another requirement is that the data must be governed by a normal distribution. To<br />

check if this is actually the c<strong>as</strong>e a histogram, showing the distribution of the data, is seen<br />

in figure 24. Only the central part of the distribution is shown in figure 24, since the tails of<br />

the distribution are very long due to the very extreme me<strong>as</strong>urements. Though it is clearly<br />

seen that the data are in fact governed by a normal distribution. Therefore Chauvenets<br />

criterion, will be used to filter this data set, C(t), with mean zero.<br />

Finally, the line (a ∗ t + b) is added to the filtered data again, and the result is the data<br />

set, which is seen marked with red, in figure 25. These steps are executed in activity 8-10<br />

in stat2pot2.<br />

The number of data points before the filtering w<strong>as</strong> 203443 and the number of points<br />

after the filtering is 202125, which corresponds to a 0.647% decre<strong>as</strong>e in data points due to<br />

the filtering. As stated in eq. 58, E0 can now be derived by finding the crossing between<br />

a line found by linear regression and the y-axis, and the rate of the decre<strong>as</strong>e in F is found<br />

<strong>as</strong> the slope of the line. The straight line, found by linear regression <strong>as</strong> activity 11, figure<br />

16 is described by the function:<br />

−F + E0 = At + B = −3.699402494578402 · 10 −4 · t − 27501928.8889737 (60)<br />

37


8. Deriving The Gravitational Potential<br />

Figure 24: Histogram of the data set C with mean 0.<br />

Figure 25: The filtered data and straight line found by linear regression.<br />

38


The line is shown in figure 25 marked with green.<br />

The result is:<br />

E0 = −27501928.89<br />

F = 3.699402494578402 ∗ 10 −4 · t<br />

8. Deriving The Gravitational Potential<br />

8.3 Calculation of the Gravitational Potential Vearth<br />

The gravitational potential in all the me<strong>as</strong>uring points can now be calculated, hence all of<br />

the necessary terms have been calculated, see eq. 49.<br />

In figure 26 the calculated <strong>gravity</strong> potential in <strong>satellite</strong> altitude, can be seen <strong>as</strong> a<br />

function of longitude and latitude. The gravitational potential is calculated <strong>as</strong> activity 12<br />

in stat2pot2.f90.<br />

Figure 26: Earths <strong>gravity</strong> potential [ m2<br />

s 2 ] in every me<strong>as</strong>uring point.<br />

An important part of the <strong>thesis</strong> h<strong>as</strong> now been completed, since the <strong>gravity</strong> potential of<br />

the Earth h<strong>as</strong> been derived from the state vectors in every me<strong>as</strong>uring point.<br />

The minimum-, maximum- and mean value of the calculated gravitational potential are<br />

listed in table 2.<br />

The result shown in figure 26 is the full gravitational potential derived from every data<br />

point. In order to show this result in a more intuitive way, the first two terms of the spherical<br />

harmonic series of the gravitational potential GM GM R and ( r r r )2C20P20(cos(θ)) cos(mλ)<br />

have been subtracted. This h<strong>as</strong> been done, since these terms are completely dominating,<br />

hence by subtracting them in every point, the variations of the field can be seen. This<br />

39


8. Deriving The Gravitational Potential<br />

min. value max. value mean value<br />

54932086.69 56781344.53 55843136.9841374<br />

Table 2: Maximum-, minimum- and mean value of the gravitational potential in <strong>satellite</strong><br />

altitude.<br />

anomalous gravitational potential for a selected period of the data can be seen in figure<br />

27.<br />

Figure 27: Earths anomalous gravitational potential [ m2<br />

s 2 ] in every data point.<br />

The next major t<strong>as</strong>k will be to derive a model which describes the gravitational potential<br />

all over the Earth, from the gravitational potential just calculated. The model will be<br />

expressed in terms of spherical harmonics.<br />

40


9 Theory of Le<strong>as</strong>t Squares Collocation<br />

9. Theory of Le<strong>as</strong>t Squares Collocation<br />

The theory behind the le<strong>as</strong>t squares collocation method is described in this section, and it<br />

is b<strong>as</strong>ed on<br />

[Moritz, 1980], [Heiskanen and Moritz, 1967], [Moritz, 1978], [Goovaerts, 1997].<br />

Le<strong>as</strong>t squares collocation is a mathematical statistical method, which can be used to<br />

determine the anomalous gravitational potential field, T , through a combination of various<br />

geodetic observations, ¯ l. The me<strong>as</strong>urements and the gravitational potential are <strong>as</strong>sumed<br />

statistically correlated.<br />

Le<strong>as</strong>t squares collocation is widely used <strong>as</strong> a tool for interpolation and for transforming<br />

between different types, distributions and qualities of me<strong>as</strong>urements. One of the great<br />

advantages is that the correlation of one field, such <strong>as</strong> the the gravitational potential, can<br />

be transformed into the correlation of another potential field, such <strong>as</strong> the <strong>gravity</strong> anomaly<br />

field. This is possible because these <strong>gravity</strong> field parameters are connected through linear<br />

functionals.<br />

In the following, T represents the anomalous gravitational potential:<br />

T = V − Vr, (61)<br />

with V being the Earth’s external gravitational potential, and Vr a reference potential.<br />

Two examples of the linear functionals relating T and other <strong>gravity</strong> parameters by ¯ l = ¯ L ¯ T ,<br />

are listed below ([Torge, 1999] eq. 6.101b, 6.97b):<br />

N = T<br />

<br />

1<br />

= T (62)<br />

γ0 γ0<br />

∆g = − ∂T 2T<br />

−<br />

∂r r =<br />

<br />

− ∂<br />

<br />

2<br />

− T,<br />

∂r r<br />

where N is the geoid height and ∆g is the <strong>gravity</strong> anomaly. γ0 is the normal potential<br />

at the ellipsoid, see section 4.3.<br />

The b<strong>as</strong>is functions:<br />

<br />

GM<br />

a n Pnm(cos θ) cos mλ (63)<br />

r r<br />

<br />

GM<br />

a n Pnm(cos θ) sin mλ<br />

r r<br />

expand the Hilbert space, which is a space containing the functions defined in a spherical<br />

coordinate system. In le<strong>as</strong>t squares collocation, these b<strong>as</strong>is functions are weighted such<br />

that the overall estimation error, mP , is minimized [Moritz, 1980]<br />

mP 2 = M(ɛP 2 ) (64)<br />

ɛP = Testimated(P ) − Tme<strong>as</strong>ured(P ),<br />

41


9. Theory of Le<strong>as</strong>t Squares Collocation<br />

where M is the averaging or mean function and ɛP is the local estimation error in point<br />

P.<br />

Using this definition, le<strong>as</strong>t squares collocation finds the ”best” estimate of the potential<br />

field, b<strong>as</strong>ed on the me<strong>as</strong>ured data.<br />

The estimated value T ′ (A) of the field in point A is given by the following equation,<br />

<strong>as</strong>suming that the data is noise-free:<br />

T ′ (A) = ¯Cst ¯C −1¯l (65)<br />

T ′ ⎡<br />

⎤<br />

K1,1 K1,2 . . . K1,i<br />

⎢ K2,1 K2,2 . . . K2,i<br />

⎥<br />

(A) = [ KA,1 KA,2 . . . KA,i ] ⎢<br />

⎥<br />

⎣ . . . ⎦<br />

Ki,1 Ki,2 . . . Ki,i<br />

−1 ⎡<br />

where ¯Cst is a vector containing the covariances, K, between all of the i data points<br />

and the point A, in which the estimation is wanted. The estimation (or interpolation)<br />

points can for example be defined by a grid, covering the Earth.<br />

The matrix ¯C, is the auto covariance matrix, containing the covariances between all of<br />

the data points. It is a symmetric matrix, since K1,2 = K2,1. The covariance function K<br />

is explained subsequently.<br />

The vector ¯ l is the observation vector, and it contains the me<strong>as</strong>urements in all the i<br />

data points.<br />

The inverse of the matrix ¯C exists because it h<strong>as</strong> full rank, due to the fact that covariance<br />

matrices are positive definite, and that no observations are used more than once,<br />

meaning that they are all linear independent.<br />

As mentioned, equation 65 is only valid if the observations are noise free which is never<br />

the c<strong>as</strong>e for real me<strong>as</strong>urements. Therefore the noise must be accounted for in equation 65,<br />

and this is done by adding a matrix ¯N, containing the noise, to the auto covariance matrix<br />

¯C. In this way, the noise may also be minimized.<br />

T ′ (A) = ¯Cst[¯C + ¯N] −1¯l (66)<br />

If it can be <strong>as</strong>sumed that all of the observations have the same variance, and that the<br />

noise on the data are uncorrelated, the noise matrix is simplified and becomes a diagonal<br />

matrix containing the variance, n 2 .<br />

¯N =<br />

⎡<br />

⎢<br />

⎣<br />

n 2 0<br />

...<br />

0 n 2<br />

42<br />

⎤<br />

⎢<br />

⎣<br />

l1<br />

l2<br />

.<br />

li<br />

⎤<br />

⎥<br />

⎦<br />

⎥<br />

⎦ (67)


9.1 The Covariance Function<br />

9. Theory of Le<strong>as</strong>t Squares Collocation<br />

A covariance function exists which describes the correlation between all the me<strong>as</strong>urements,<br />

hence it describes the form of the field, and this information is needed when doing an estimation<br />

or interpolation. [Moritz, 1980], [Moritz, 1978], [Torge, 1999], [Heiskanen and Moritz, 1967]<br />

The spacial covariance function, K(P, Q), between two points P = (θ, λ) and Q =<br />

(θ ′ , λ ′ ) on the sphere r=R, separated by the angle ψ is given by ([Moritz, 1980] eq. 10-9):<br />

K(P, Q) = M(T (P ), T (Q)) (68)<br />

= 1<br />

8π2 2π π 2π<br />

T (θ, λ) · T (θ ′ , λ ′ ) sin θdθdλdα.<br />

λ=0<br />

θ=0<br />

α=0<br />

M is an averaging or mean operator and α is the azimuth [Moritz, 1978].<br />

When the disturbing potential is written <strong>as</strong> a spherical harmonic series ([Torge, 1999]<br />

eq.6.4):<br />

T (r, θ, λ) = GM<br />

r<br />

∞<br />

n=2<br />

<br />

a<br />

n n<br />

r<br />

m=0<br />

or equivalently with fully normalized coefficients:<br />

T (r, θ, λ) = GM<br />

r<br />

n=2<br />

m=0<br />

[Cnm cos(mλ) + Snm sin(mλ)]Pnm(cos(θ)), (69)<br />

∞ <br />

a<br />

n n<br />

[<br />

r<br />

¯ Cnm cos(mλ) + ¯ Snm sin(mλ)] ¯ Pnm(cos(θ)), (70)<br />

and the above mentioned covariance function is extended into space outside r=R, it<br />

can also be written <strong>as</strong> a spherical harmonic expansion ([Torge, 1999] eq.6.191a):<br />

K(P, Q) =<br />

∞<br />

n=2<br />

σn 2<br />

2 n+1<br />

R<br />

Pn(cos ψ). (71)<br />

(rP rQ)<br />

The angle ψ is calculated from the relationship ([Moritz, 1980] eq. 10-3):<br />

cos ψ = cos θ cos θ ′ + sin θ sin θ ′ cos(λ ′ − λ) (72)<br />

The coefficients σn, called the degree variances, are determined by the relationship<br />

([Torge, 1999] eq.6.192 and 6.141 ):<br />

σn 2 =<br />

GM<br />

R<br />

2 n <br />

m=0<br />

( ¯ C 2 nm + ¯ S 2 nm) (73)<br />

Two parameters characterize the covariance function, namely the variance, K0, the<br />

correlation length, ξ ([Moritz, 1978] eq.35):<br />

43


9. Theory of Le<strong>as</strong>t Squares Collocation<br />

Figure 28: Characteristic parameters of the covariance function.<br />

K0 = K(ψ = 0) = var(T )<br />

K(ξ) = 1<br />

2 K0<br />

These parameters are also shown in figure 28.<br />

If the covariance between two points is zero, the points in question are statistically<br />

uncorrelated. As the distance between the two points incre<strong>as</strong>e, the covariance decre<strong>as</strong>es<br />

since the points become more and more uncorrelated. The covariance does not converge at<br />

zero, but tends to oscillates between small values around zero. This oscillation is caused by<br />

the fact that the gravitational potential in a point is not only affected by the local density<br />

variations, but also from large scale density variations. Positive covariances indicates that<br />

the two points have potentials of same size and sign, and negative covariances indicates<br />

potentials of same size but opposite signs.<br />

44<br />

(74)<br />

(75)


10. Methods of Calculating the Spherical Harmonic Coefficients<br />

10 Methods of Calculating the Spherical Harmonic<br />

Coefficients<br />

If the anomalous potential field T is known everywhere on a sphere with surface area σ, the<br />

fully normalized spherical harmonic coefficients, ¯ Cnm and ¯ Snm, can be calculated directly<br />

using the following equation ([Torge, 1999], eq. 6.138a):<br />

¯Cnm =<br />

¯Snm =<br />

<br />

1<br />

4πGM<br />

<br />

1<br />

4πGM<br />

σ<br />

σ<br />

<br />

r<br />

n r T (θ, λ)<br />

a<br />

¯ Pnm(cos(θ)) cos(mλ)dσ (76)<br />

<br />

r<br />

n r T (θ, λ)<br />

a<br />

¯ Pnm(cos(θ)) sin(mλ)dσ,<br />

but since this is clearly not the c<strong>as</strong>e, other methods must be used in order to calculate<br />

the coefficients. The two methods used in this project, are explained in the following<br />

sections. The first method is the numerical integration method, which subsequently will be<br />

called method 1, and the other method using the orthogonality properties of the spherical<br />

harmonics, will be called method 2. These two methods are used together with le<strong>as</strong>t<br />

squares collocation to derive an <strong>Ørsted</strong> <strong>gravity</strong> field model.<br />

The Kepler element method is also described in this section, though this method h<strong>as</strong><br />

not been used here.<br />

10.1 Method 1: Using Numerical Integration<br />

Since the anomalous potential T is not known everywhere, but only in the data points,<br />

an interpolation of the field can be found using le<strong>as</strong>t squares collocation, such that it is<br />

possible to execute a numerical calculation of the integrations in eq. 76, and determine the<br />

spherical harmonic coefficients.<br />

Equation 66 is used to determine the anomalous potential T ′ in a number of points<br />

defined by a grid covering the Earth, by interpolating between the data points. When this<br />

interpolation of the field is executed, numerical integration can be used to calculate the<br />

coefficients by [Moritz, 1980]:<br />

¯Cnm =<br />

¯Snm =<br />

<br />

1<br />

4πGM<br />

<br />

1<br />

4πGM<br />

σ<br />

σ<br />

<br />

r<br />

n r T<br />

a<br />

′ (θ, λ) ¯ Pnm(cos(θ)) cos(mλ)dσ<br />

<br />

r<br />

n r T<br />

a<br />

′ (θ, λ) ¯ Pnm(cos(θ)) sin(mλ)dσ, (77)<br />

Here Simpson’s rule of numerical integration h<strong>as</strong> been chosen for the calculations, and<br />

this method is explained in section 4.6. When using Simpson’s numerical integration,<br />

the grid defining the interpolation points must be equi-angular and it must have an even<br />

number of subintervals in both latitude and longitude.<br />

An example of such a grid is shown in figure 29, and here the number of grid points is<br />

9 × 9 = 81.<br />

45


10. Methods of Calculating the Spherical Harmonic Coefficients<br />

Figure 29: Example of grid.<br />

10.2 Method 2: Using Orthogonality Properties of the Spherical<br />

Harmonics<br />

The method, which is b<strong>as</strong>ed on the orthogonality properties of the spherical harmonics,<br />

is another method of calculating the spherical harmonic coefficients ¯ Cnm and ¯ Snm,<br />

[Tscherning, 2001]. This method will be described in this section.<br />

The estimated anomalous potential in the point P , T ′ (P ), can be calculated by equation<br />

66, using le<strong>as</strong>t squares collocation. This equation can be rewritten in the following<br />

way:<br />

T ′ (P ) = ¯C T st(P )[¯C + ¯N] −1¯ l<br />

= ¯C T st(P ) ¯ k (78)<br />

N<br />

= K(P, Qi)ki<br />

i=1<br />

Where N is the number of observations, K is the covariance and the elements ki can<br />

be calculated directly by ki = ([ ¯C + ¯N] −1¯l)i.<br />

The fully normalized coefficients are given by equation 76, and by substituting the<br />

expression for the anomalous potential just derived and equation 71 into this equation of<br />

¯Cnm, the following is obtained:<br />

¯Cnm =<br />

<br />

1<br />

N<br />

rP<br />

n<br />

rP<br />

4πGM σ R<br />

i=1<br />

¯Pnm(cos θ) cos(mλ)dσ.<br />

46<br />

∞<br />

j=2<br />

σj 2<br />

<br />

2 j+1<br />

R<br />

Pj(cos ψ)ki · (79)<br />

rP rQi


10. Methods of Calculating the Spherical Harmonic Coefficients<br />

Here rP is the altitude of point P from equation 78 and φ is the angle between the point<br />

P and the me<strong>as</strong>urepoint Qi.<br />

In the following, only the ¯ Cnm coefficient will be derived. Equivalent derivations of ¯ Snm<br />

are e<strong>as</strong>ily found.<br />

Rearranging the latter equation,<br />

<br />

and normalizing the Legendre polynomial Pj(cos ψ)<br />

2j+1<br />

by the normalization factor , the following expression is obtained:<br />

2<br />

¯Cnm =<br />

<br />

σ<br />

1 r<br />

4πGM<br />

n+1<br />

P<br />

Rn N ∞<br />

σj<br />

i=1 j=2<br />

2<br />

2 j+1<br />

R<br />

ki · (80)<br />

rP rQi<br />

<br />

2j + 1<br />

¯Pj(cos ψ)ki<br />

2<br />

¯ Pnm(cos θ) cos(mλ)dσ<br />

Here it is seen that the point P is cancled out from the equation. By writing the<br />

equation with spherical harmonics instead of Legendre polynomials, using the relation:<br />

¯Pj(cos ψ) =<br />

j<br />

k=−j<br />

¯Yjk(θP , λP ) · ¯ Yjk(θQi , λQi ),<br />

and also using the orthogonality properties of the spherical harmonics:<br />

<br />

equation 79 h<strong>as</strong> the form:<br />

σ<br />

¯Cnm = R<br />

4πGM<br />

¯Yjk · ¯ Ynmdσ = 0, if j = n or k = m<br />

N<br />

i=1<br />

σn 2<br />

<br />

2 n+1<br />

R 2n + 1<br />

ki<br />

¯Ynm(Qi) (81)<br />

rP rQi<br />

2<br />

From the latter equation, the harmonic coefficients can be directly calculated, if the<br />

σn’s are known.<br />

10.3 The Kepler Element Method<br />

As mentioned, the Kepler element method is another way of finding the coefficients of the<br />

spherical harmonic expansion of the gravitational potential of the earth, see section 4.2.<br />

This method is an alternative to the two methods, which are used in this project and are<br />

described in section 10.1 and 10.2.<br />

When using the Kepler element method [Kaula, 2000][Seeber, 2003], the motion of the<br />

<strong>satellite</strong> in question must be expressed in terms of the Kepler elements (or Keplerian orbital<br />

parameters):<br />

47


10. Methods of Calculating the Spherical Harmonic Coefficients<br />

a semi-major axis of the <strong>satellite</strong> orbit<br />

e numerical eccentricity<br />

i orbit inclination<br />

Ω right <strong>as</strong>cension of <strong>as</strong>cending node<br />

ω argument of perigee<br />

ν true anomaly<br />

The Kepler elements are also shown in figure 30, in which Pe is the pericenter and S is<br />

the position of the <strong>satellite</strong>.<br />

Figure 30: Figure showing the Keplerian orbital parameters ([Seeber, 2003], fig.3.4).<br />

Equations <strong>as</strong> the folowing can be derived. This equation gives the variation of the<br />

Kepler element, Ω in time is given by ([Kaula, 2000], p.39):<br />

dΩ<br />

dt = 3nC20ae 2 cos i<br />

2(1 − e 2 ) 2 , (82)<br />

with ae being the equatorial radius of the earth and n = √ GMa −3/2<br />

From this equation, the coefficient C20 can be directly calculated.<br />

From similar expressions, the rest of the coefficients can be derived ([Kaula, 2000],p.40).<br />

Equation 82 is valid when <strong>as</strong>suming that:<br />

da de<br />

= 0 and = 0. (83)<br />

dt dt<br />

And this is clearly not the c<strong>as</strong>e due to the non-gravitational forces which are indeed<br />

acting on the <strong>satellite</strong> and disturbing orbit. If the Kepler element method should be used,<br />

48


10. Methods of Calculating the Spherical Harmonic Coefficients<br />

one would have to estimate an average of the semi-major axis of the <strong>satellite</strong> orbit, which<br />

would induce an error, and this is the re<strong>as</strong>on why this method h<strong>as</strong> not been used in this<br />

<strong>thesis</strong>.<br />

49


11. Collo.f90<br />

11 Collo.f90<br />

The program collo.f90 h<strong>as</strong> been developed during this project, with the purpose of calculating<br />

the spherical harmonic coefficients of the <strong>Ørsted</strong> gravitational potential model. The<br />

program can be seen in appendix C. The calculations are b<strong>as</strong>ed on le<strong>as</strong>t-squares collocation,<br />

which is described in section 9 and the two methods of calculating the coefficients<br />

described in section 10.1 and 10.2.<br />

11.1 Program outline<br />

In this section, the program outline of the main program collo.f90 and of subroutine ’collocation’<br />

will be described.<br />

Main Program ’collo.f90’<br />

A flowchart of the main program collo.f90 is seen in figures 31.<br />

Figure 31: Flowchart of collo.f90<br />

50


11. Collo.f90<br />

The first activity in collo.f90 is to read the arguments from the command line which<br />

are the input file name and the parameter nte. The input file for collo.f90 is the file<br />

collodata.inp which is generated by stat2pot2.f90, see section 8 figure 16.<br />

The parameter nte indirectly defines the number of input data which will be used in<br />

the following calculations. For example by choosing nte = 2, every second input data will<br />

be used and by choosing nte = 3, every third input data will be used etc. This is what is<br />

shown in activity 2 and 3 in the flowchart of collo.f90.<br />

The theory of le<strong>as</strong>t squares collocation which is explained in section 9 is valid when wanting<br />

to determine the anomalous potential, T . Often, the anomalous potential is defined<br />

<strong>as</strong> T = V − U, see equation 19. This is not a good definition here since the purpose of<br />

collo.f90 is also to determine the coefficients of the normal potential U, see eq. 22, and it<br />

therefore should not be subtracted from the data. Therefore the anomalous potential will<br />

be defined <strong>as</strong> T = V − GM in the following.<br />

r<br />

Hence, in order to determine the anomalous potential, the first term of the spherical<br />

harmonic expansion of the gravitational potential, namely GM must be calculated and<br />

r<br />

subtracted from the data. The latter is executed in activity 4. The calculation of GM<br />

r will<br />

be described in section 12.1.<br />

As described in activity 5 in figure 31, the subroutine ’collocation’ is called to execute<br />

the calculations on the selected data. The subroutine ’collocation’ contains the essential<br />

parts of the collo.f90 program, and it is described in the following.<br />

Subroutine ’collocation’<br />

A flowchart of ’collocation’ is shown in figure 32. One of the b<strong>as</strong>ic ide<strong>as</strong> behind this<br />

program is to estimate the individual coefficients in an iterative process, <strong>as</strong> it is outlined<br />

the flowchart.<br />

The first activity shown in the ’collocation’ flowchart is the definition of a grid. This<br />

grid will be used when performing the numerical integration used in method 1, see section<br />

10.1. An example of such an equi-angular grid is shown in figure 29.<br />

In the program, the grid points are defined by two vectors containing the intervals in<br />

longitude λ and latitude θ:<br />

(λ1, λ2, λ3, . . . , λj) = [0; 2π<br />

; 2π] (84)<br />

j<br />

(θ1, θ2, θ3, . . . , θj) = [0; π<br />

; π],<br />

j<br />

such that j × j is the number of grid points. The chosen grid height is mean <strong>satellite</strong><br />

height.<br />

In activity 2, an initial guess of the coefficient ¯ C20 is used <strong>as</strong> input (it is known that<br />

¯S20 = 0), and b<strong>as</strong>ed on this guess the degree variance σ2, see eq. 73, is calculated <strong>as</strong><br />

51


11. Collo.f90<br />

Figure 32: Flowchart of subroutine collocation<br />

activity number 3. The final result of the coefficient estimation is independent on this<br />

initial guess, since an iterative process is used. The number of iterations needed to obtain<br />

a satisfactory result is dependent on the initial guess, though.<br />

The next t<strong>as</strong>k in the program is to calculate the elements of the auto covariance matrix<br />

¯ C, b<strong>as</strong>ed on the degree variances found in activity 3, and this is done in activity 4.<br />

Hence in the iterative process, the covariance is calculated from the current guess of the<br />

coefficients. This auto covariance matrix is used in activity 5 to solve the matrix equation<br />

¯C ¯ k = ¯ lobs for ¯ k. ¯ lobs is the vector containing all of the chosen input data.<br />

In the program, an ordering of the different matrices used must be determined, such that<br />

the multiplication of the matrices is done correctly. The auto covariance matrix is ordered<br />

<strong>as</strong> follows:<br />

52


⎛<br />

K(l1, l1)<br />

⎜<br />

¯C<br />

⎜ K(l2, l1)<br />

= ⎜<br />

⎝ .<br />

K(l1, l2)<br />

K(l2, l2)<br />

.<br />

. . .<br />

. . .<br />

...<br />

⎞<br />

K(l1, li)<br />

K(l2, li) ⎟<br />

. ⎠<br />

K(li, l1) K(li, l2) . . . K(li, li)<br />

11. Collo.f90<br />

Since the auto covariance matrix is symmetric, such that K(l1, l2) = K(l2, l1), it is<br />

reduced to:<br />

⎛<br />

K(l1, l1)<br />

⎜<br />

¯C<br />

⎜ 0<br />

= ⎜<br />

⎝ .<br />

K(l1, l2)<br />

K(l2, l2)<br />

.<br />

. . .<br />

. . .<br />

.. .<br />

⎞<br />

K(l1, li)<br />

K(l2, li) ⎟<br />

. ⎠<br />

0 0 . . . K(li, li)<br />

At this point, two different methods are used for the calculation of d(nm), which is the<br />

estimate of the (nm) coefficient. In activity 9, method 2 described in section 10.2 is used<br />

to estimate the intermediate coefficient d(nm) which can be either ¯ Cnm or ¯ Snm. This is<br />

done using only the vector ¯ k.<br />

In activity 6-8 method 1, which is described in section 10.1, is executed. In activity 6<br />

the cross covariance matrix ¯Cst, see eq. 65, is calculated such that the vector ¯t = ¯Cst ¯ k can<br />

be calculated in activity 7. ¯Cst is written <strong>as</strong> a matrix containing the covariances between<br />

all of the grid points and all the data points instead of being a vector. This is done in<br />

order to execute all of the calculations at once, in stead of calculating the interpolation<br />

in the grid points one at the time. This choice is time saving, which is a very important<br />

element when dealing with many data. The ordering of the cross covariance is <strong>as</strong> follows:<br />

⎛<br />

K((θ1, λ1), l1)<br />

⎜<br />

K((θ2, λ1), l1)<br />

⎜ .<br />

¯Cst<br />

⎜<br />

= ⎜ K((θj, λ1), l1)<br />

⎜ K((θ1, λ2), l1)<br />

⎜<br />

⎝ .<br />

K((θ1, λ1), l2)<br />

K((θ2, λ1), l2)<br />

.<br />

K((θj, λ1), l2)<br />

K((θ1, λ2), l2)<br />

.<br />

. . .<br />

. . .<br />

...<br />

. . .<br />

. . .<br />

...<br />

⎞<br />

K((θ1, λ1), li)<br />

K((θ2, λ1), li) ⎟<br />

. ⎟<br />

K((θj, λ1), li) ⎟<br />

K((θ1, λ1), li) ⎟<br />

. ⎠<br />

K((θj, λj), l1) K((θj, λj), l2) . . . K((θj, λj), li)<br />

The ¯t vector contains all of the interpolation values, T ′ , in all of the grid points, and it<br />

h<strong>as</strong> the form:<br />

53<br />

(85)<br />

(86)


11. Collo.f90<br />

⎛<br />

⎜<br />

¯t<br />

⎜<br />

= ⎜<br />

⎝<br />

T ′ (θ1, λ1)<br />

T ′ (θ2, λ1)<br />

.<br />

T ′ (θ1, λ2)<br />

T ′ (θ2, λ2)<br />

.<br />

T ′ (θ?, λ?)<br />

Finally the estimate of the coefficient d(nm) i calculated using numerical integration in<br />

activity 8.<br />

Independently of the method used, the d(nm) coefficient is added to the sum of calculated<br />

d(nm) coefficients (nm) ′ in activity 10. In the first iteration (nm) ′ = 0 since the<br />

calculated d(nm) is the first contribution to the coefficient.<br />

In activity 11, the d(nm) potential is subtracted from the data, in order to be able to<br />

perform further calculations on the residual potential, such that the spherical harmonic<br />

coefficient estimation is done in an iterative way.<br />

The d(nm) estimate is now being evaluated. If the found d(nm) is less than 1·10 −16 , it is<br />

<strong>as</strong>sumed that the full coefficient h<strong>as</strong> been found and no further iteration on this coefficient<br />

will be executed. As a result, the order or degree is incre<strong>as</strong>ed, such that the next coefficient<br />

can be calculated. This is what is shown in activity 12b. This can be done because the<br />

¯Cnm and ¯ Snm coefficients are all orthogonal.<br />

Assuming that the full coefficient h<strong>as</strong> been found and is equal to (nm) ′ this is printed<br />

to the screen, and (nm) ′ is set to zero for the calcuation of the next coefficient, see activity<br />

13. When the next coefficient must be found, another first guess must be made, and this<br />

is determined b<strong>as</strong>ed on the EGM96 coefficients <strong>as</strong> it is stated in activity 14. The guess is<br />

b<strong>as</strong>ed on the EGM96 model coefficients to decre<strong>as</strong>e the number of iterations, and thereby<br />

also the computational time needed to obtain a satisfactory result.<br />

If, on the other hand, the coefficient d(nm) is larger that 1 · 10 −16 , it is <strong>as</strong>sumed that<br />

the full coefficient is still not found, and the guess for the new iteration is set to d(nm)/2<br />

and hereafter another iteration of the program is initiated. The covariance function of this<br />

new iteration h<strong>as</strong> the same form <strong>as</strong> in the first iteration, see eq. 71, only with the degree<br />

variances calculated from the new guess d(nm)/2.<br />

54<br />

⎞<br />

⎟<br />

⎠<br />

(87)


11.2 Programming Problems and Solutions<br />

11. Collo.f90<br />

During the development of the program collo.f90 and especially the subroutine collocation.f90,<br />

many problems arose and solutions had to be found. In this section, these problems<br />

and solutions are described and discussed. For the process of programming problem<br />

solving, [Press et al., 1992] h<strong>as</strong> been a great help. In order to be able to find and eliminate<br />

the programming problems, the program h<strong>as</strong> continuously been tested on test data.<br />

The testdata consists of potentials directly calculated from the coefficients of the EGM96<br />

model, see [Lemoine et al., 1998] and appendix B.<br />

The advantage of testing the program on the EGM96 test data, is that it is e<strong>as</strong>y to<br />

conclude whether the result is satisfactory or not. This is not e<strong>as</strong>y with results found<br />

from the <strong>Ørsted</strong> data, since it is not yet known how much <strong>gravity</strong> information these data<br />

actually contain.<br />

11.2.1 The multi-grid function<br />

Any arithmetic operation using floating numbers, are introducing roundoff error. These<br />

roundoff errors accumulate, when the number of calculations incre<strong>as</strong>e, and may consequently<br />

swamp the true solution to the problem. For linear equation systems, this happens<br />

if the number of unknowns (N) is too large [Press et al., 1992].<br />

In this project, several thousand data points must be evaluated, and hence a system of<br />

equations with dimension equal to the number of data equations, must be solved. To test<br />

if roundoff errors are a problem, the program w<strong>as</strong> tested on the EGM96 test data, and it<br />

w<strong>as</strong> found that the solution, found by collo.f90 ,w<strong>as</strong> indeed swamped in roundoff errors.<br />

In order to minimize the roundoff errors, a multi-grid function w<strong>as</strong> implemented in collocation.f90.<br />

This multi-grid function divides the Surface of the Earth into boxes containing<br />

only a fraction of the data points. The calculations given in activity 5 in the ’collocation’<br />

flowchart (figure 32), are executed within each box, meaning that the size of the equation<br />

systems is decre<strong>as</strong>ed dramatically. The decre<strong>as</strong>e in number of equations, should decre<strong>as</strong>e<br />

the roundoff errors and thereby improve the solution.<br />

It is important though, that the number of points in each box is neither too large such<br />

that roundoff errors dominate the solutions of some boxes, nor that it is too small such<br />

that possible me<strong>as</strong>uring errors dominate the results instead of being averaged out. Also, it<br />

is important that the box size is sufficiently large, such that the boxes in the polar regions<br />

are not empty due to the polar gaps of the <strong>Ørsted</strong> <strong>satellite</strong> coverage. To continuously check<br />

this condition, the boxes and the number of data are plotted in plots like the one shown<br />

in figure 33.<br />

Another advantage of dividing the problems into smaller linear systems is a great improvement<br />

in the computational speed, because the calculations of the boxes can be executed<br />

paralleled.<br />

One disadvantage of using the multi-grid method, is that the points in one box are treated<br />

<strong>as</strong> if they are completely uncorrelated with points from all other boxes. This is not a<br />

55


11. Collo.f90<br />

180˚<br />

180˚<br />

225˚<br />

225˚<br />

270˚<br />

270˚<br />

315˚<br />

315˚<br />

0˚<br />

80˚ 80˚<br />

60˚ 60˚<br />

40˚ 40˚<br />

20˚ 20˚<br />

0˚ 0˚<br />

-20˚ -20˚<br />

-40˚ -40˚<br />

-60˚ -60˚<br />

-80˚ -80˚<br />

0˚<br />

Figure 33: Boxes used in the multi-grid function and the number of data, N, in every<br />

box.<br />

fair <strong>as</strong>sumption, since points lying close to a box edge will indeed be correlated with the<br />

points in the neighboring boxes. Therefore the concept of overlap w<strong>as</strong> also implemented in<br />

collo.f90 such that the data of all of the neighboring boxes are also used when solving for<br />

each box. In this way solutions of each box is found 9 times, and the final result is then<br />

given by the average value of these 9 solutions.<br />

The choice of using the multi-grid function did reduce the roundoff errors and the results<br />

of the computations were improved considerably. But unfortunately the errors were not<br />

eliminated, and therefore additional programming considerations were necessary. These<br />

will be described subsequently.<br />

11.2.2 Iterative Improvement<br />

A problem that can occur, when working with large sets of equations, is that it is not<br />

always possible to obtain a level of precision, which is comparable with the limit of the<br />

computer. Sometimes even one or two figures can be lost, for matrices which were far<br />

from singular. An effective way of restoring the full computer precision is to use ”iterative<br />

improvement” of the solution [Press et al., 1992]. This method is shown schematically in<br />

figure 34.<br />

In order to obtain the wanted level of precision, it w<strong>as</strong> chosen to implement the method<br />

of iterative improvement in the subroutine collocation.f90. In the following, the activities<br />

mentioned refer to the flowchart of collocation.f90, see figure 32.<br />

The iterative improvement w<strong>as</strong> implemented by calculating a first estimate of the coefficient,<br />

d(nm) in activity 9 and adding this to the the (nm) ′ coefficient in activity 10.<br />

Before the first iteration (nm) ′ = 0 and after the l<strong>as</strong>t iteration (nm) ′ will be the the sum<br />

of all of the calculated d(nm) coefficients and the result is the full estimated coefficient<br />

either ¯ Cnm or ¯ Snm. In activity 11 the d(nm) coefficient is subtracted from the data and in<br />

45˚<br />

45˚<br />

56<br />

90˚<br />

90˚<br />

135˚<br />

135˚<br />

180˚<br />

180˚<br />

N<br />

95<br />

90<br />

85<br />

80<br />

75<br />

70<br />

65<br />

60<br />

55


11. Collo.f90<br />

Figure 34: The principle of iterative improvement of the solution to Ā · ¯x = ¯ b. The<br />

first guess ¯x + δ¯x is multiplied with Ā to give ¯ b + δ ¯ b. ¯ b is subtracted to give δ ¯ b which is<br />

inverted to give δ¯x. This is subtracted from the first guess to givean improved solution ¯x.<br />

([Press et al., 1992] figure 2.5.1)<br />

activity 9, a new coefficient d(nm) is calculated on the residual data set.<br />

11.2.3 Decomposition<br />

When wanting to optimize a program which deals with large matrices, matrix decomposition<br />

can be introduced with great computational benefit.<br />

Assuming that the matrix Ā can be written <strong>as</strong> a product of two matrices, such that<br />

Ā = ¯ L · Ū. Here Ū is an upper triangular matrix containing elements in and above<br />

the diagonal while ¯ L is a lower triangular matrix containing elements in and below the<br />

diagonal.<br />

Such a LU decomposition can be used to solve the following linear equation; Ā · ¯x =<br />

( ¯ L· Ū)· ¯x = ¯ L·( Ū· ¯x) = ¯ b. This is done by first solving for the vector ¯y such that ¯ L· ¯y = ¯ b<br />

and then Ū · ¯x = ¯y [Press et al., 1992].<br />

The great computational advantage of this kind of decomposition, is that the solution<br />

of a triangular set of equations is quite trivial.<br />

In the c<strong>as</strong>e of Ā being symmetric and positive definite, it h<strong>as</strong> a more efficient triangular<br />

decomposition called Cholesky decomposition. A positive definite matrix Ā is defined<br />

by<br />

¯v · Ā · ¯v > 0 for all ¯v<br />

Cholesky decomposition is extremely stable numerically without any pivoting at all,<br />

and it is used by the LAPACK subroutines used in collocation.f90 for the symmetric and<br />

positive definite auto covariance matrix C in activity 5 in the flowchart, figure 32.<br />

57


11. Collo.f90<br />

LAPACK is a collection of standard subroutine packages, which h<strong>as</strong> been developed at<br />

Argonne National Laboratories and is published, well documented and available for free<br />

use, see http://www.netlib.org/lapack/.<br />

Several of these subroutines h<strong>as</strong> been used in collocation.f90 since these are developed<br />

to solve problems involving large systems of equations, and it w<strong>as</strong> indeed found that the<br />

use of the LAPACK subroutines did give better results.<br />

11.2.4 Regularization<br />

A problem is said to be ill-posed if the following criteria are not fulfilled (see e.g.<br />

http://en.wikipedia.org/wiki/Ill-posed problem):<br />

1. A solution exists<br />

2. The solution is unique<br />

3. The solution depends continuously on the initial data.<br />

If a problem is ill-posed, it must be reformulated in order to be solved numerically.<br />

This reformulation is done by regularization, where new <strong>as</strong>sumptions on the solution are<br />

added, <strong>as</strong> for example smoothness. Therefore regularization is often a trade off between<br />

smoothness and data fitting.<br />

The condition number, κ, is a parameter which defines whether a problem of the form<br />

x = ¯ b in numerical analysis is well-posed or not.<br />

κ( Ā) = Ā−1 · Ā (88)<br />

A well-conditioned (well-posed) problem h<strong>as</strong> a low condition number, while a problem<br />

with a high condition number is said to be ill-conditioned (ill-posed).<br />

In collocation.f90 the regularization is performed by adding a number α to the diagonal<br />

of the auto covariance matrix, C ′ = C + αI, in activity 4, figure 32. The number α h<strong>as</strong><br />

been chosen by trial and error, in the search for a good trade off between smoothness and<br />

data fitting.<br />

As stated in [Metzler and Pail, 2005] the presence of polar gaps in the distribution of<br />

data points is partly responsible for the ill-posed problem of calculating the gravitational<br />

potential from GOCE data. The problem becomes ill-posed because the b<strong>as</strong>is functions (the<br />

spherical harmonics) are globally defined, and therefore dependent on globally distributed<br />

data. Therefore a Spherical Cap Regularization Approach (SCRA) h<strong>as</strong> been developed for<br />

GOCE data. Since the inclination of <strong>Ørsted</strong> also creates polar gaps, similar considerations<br />

h<strong>as</strong> been made in this project, and regularization h<strong>as</strong> been used to stabilize the set of linear<br />

equations.<br />

To determine whether the problem is ill-posed because of to the polar gaps, synthetic<br />

data generated from the EGM96 model h<strong>as</strong> been used to cover the polar gaps. This<br />

synthetic data h<strong>as</strong> been added to the real <strong>Ørsted</strong> data and collo.f90 h<strong>as</strong> been used to<br />

find a solution. It w<strong>as</strong> shown that this <strong>Ørsted</strong>-synthetic solution did only vary little (a<br />

few percent) from the <strong>Ørsted</strong>-only solution, and it therefore must be concluded that the<br />

58


11. Collo.f90<br />

presence of polar gaps is not a significant problem in this project. This may be due to<br />

the use of the multi-grid function which divides the Earth into boxes which also cover the<br />

polar regions and do all contain a certain amount of data, see 11.2.1<br />

59


12. Deriving the <strong>Ørsted</strong> Gravitational Potential Field Model ORSTED05<br />

12 Deriving the <strong>Ørsted</strong> Gravitational Potential Field<br />

Model ORSTED05<br />

In this section the <strong>Ørsted</strong> gravitational potential model will be derived, expressed <strong>as</strong> a<br />

spherical harmonic series. The program collo.f90 h<strong>as</strong> been developed during the project<br />

for this purpose, and it w<strong>as</strong> described in section 11.<br />

12.1 Calculation of GM<br />

Since the term GM is by far the most dominating term, it is <strong>as</strong>sumed that it can be<br />

r<br />

determined by a simple method described in the following. Since GM is the dominant term<br />

r<br />

of the gravitational potential, we disregard all other terms.<br />

Hence all of the calculated values of the potential multiplied with the distance Vearth · r<br />

are added up, and divided by the number of me<strong>as</strong>urements, N. This means that GM is<br />

simply approximated by the mean of the Vearth · r. This can only be done due to the great<br />

and the following terms in the gravitational potential.<br />

difference in size of GM<br />

r<br />

GM ≈<br />

N i=1 (Vearth(i) · r(i))<br />

N<br />

= (GM)approx<br />

Using the latter approximation, the following result is achieved:<br />

(GM)approx = 398514652297117m 3 /s 2<br />

The difference between the approximated and the ’true’ value is evaluated in percent:<br />

(GM)true − (GM)approx<br />

(GM)true<br />

= (3.986009 ∗ 1014 m 3 /s 2 − 398514652297117m 3 /s 2 )<br />

3.986009 ∗ 10 14 m 3 /s 2<br />

= −2.1638 ∗ 10 −4 ≈ −0.0216%<br />

The latter equation shows that it is an acceptable approximative method to find GM,<br />

since the difference is of such a small size.<br />

Hence, the anomalous potential, T , can now be used in the further calculations. This<br />

is done in activity number 4 in the flowchart of collo.f90, which is shown in figure 31.<br />

12.2 Comparing Results from Method 1 and Method 2<br />

Both method 1 and 2 have been tested on the EGM96 test data to determine the spherical<br />

harmonic coefficients of degree 2 to 4. This h<strong>as</strong> been done in order to find which of the<br />

two methods is the f<strong>as</strong>test and most accurate, when used for the full model determination.<br />

The results of the two methods are compared to each other and to the equivalent<br />

EGM96 coefficients. In table 3, the coefficients found by the two methods are listed.<br />

In this test, 2520 test data points generated from EGM96 were used, and the grid used<br />

in method 1 h<strong>as</strong> 20 × 20 points. The overlap multi-grid method h<strong>as</strong> been used, and the<br />

60<br />

(89)


12. Deriving the <strong>Ørsted</strong> Gravitational Potential Field Model ORSTED05<br />

largest system of equations to be solved is of size 64 while the smallest is of size 42. The<br />

box size is approximately 12 ◦ × 12 ◦ and the number of multi-grid boxes is 15 × 30.<br />

n m ¯ Cnm<br />

Method 2 Method 1<br />

¯Snm<br />

¯Cnm<br />

¯Snm<br />

2 0 −0.48445497 · 10−3 0 −0.48447268 · 10−3 0<br />

2 1 −0.99037859 · 10−10 −0.82713757 · 10−10 −0.10038595 · 10−9 −0.93629297 · 10 10<br />

2 2 0.20923471 · 10−5 -0.10190048 · 10−5 0.20809231 · 10−5 −0.10426476 · 10−5 3 0 0.92019799 · 10−6 0 0.96282684 · 10−6 0<br />

3 1 0.16576926 · 10−5 0.20233258 · 10−6 0.16588057 · 10−5 0.24617314 · 10−6 3 2 0.62848981 · 10−6 −0.42572507 · 10−6 0.62489276 · 10−6 −0.44617471 · 10−6 3 3 0.35437913 · 10−6 0.72325547 · 10−6 0.41794515 · 10−6 0.80057489 · 10−6 4 0 0.12895032 · 10−5 0 0.13100918 · 10−5 0<br />

4 1 −0.29858347 · 10−6 −0.25282049 · 10−6 −0.28729297 · 10−6 −0.28710544 · 10−6 4 2 0.23121831 · 10−6 0.26341738 · 10−6 0.24554158 · 10−6 0.29924076 · 10−6 4 3 0.42740805 · 10−6 −0.76283535 · 10−7 0.42692479 · 10−6 −0.59335934 · 10−7 4 4 −0.56833794 · 10−7 0.86072184 · 10−7 −0.50901670 · 10−7 0.87327298 · 10−7 Table 3: Spherical harmonic coefficients found by two different methods.<br />

From table 3 it is seen that the two methods give similar although not identical results.<br />

For the purpose of comparison, the lowest order coefficients of the EGM96 model are listed<br />

in appendix 10. When comparing the results given in table 3 to the EGM96 coefficients it<br />

is seen that that the results are of the same order of magnitude <strong>as</strong> the EGM96 coefficients,<br />

but that they are not identical.<br />

This difference must be due to errors in the calculations done by collo.f90, and will be<br />

discussed later.<br />

To determine which of the two methods should be used for the further calculations, a<br />

more detailed examination of the determination of ¯ C20 h<strong>as</strong> been executed.<br />

For this examination, the number of data points h<strong>as</strong> been held fixed, while the number<br />

of integration grid points used in method 1, h<strong>as</strong> been incre<strong>as</strong>ed.<br />

In table 4, the coefficient ¯ C20 found by method 2 and the computation time tcomp are<br />

seen, and in table 5 the values of the ¯ C20 coefficient found by method 1 are listed together<br />

with the chosen number of grid points, Ngrid, and the computation time, tcomp. The number<br />

of boxes used in the multi-grid function is the same <strong>as</strong> in the first test leading to the results<br />

in table 3.<br />

When comparing the results given in table 5 and 4, it is seen that they are almost<br />

identical. This means that both methods are usable, but there are major differences when<br />

looking at the computation time.<br />

It is intuitively clear that the numerical integration depends on the chosen number of<br />

grid points, and it is seen in the tables that at constant number of data points, an incre<strong>as</strong>e<br />

61


12. Deriving the <strong>Ørsted</strong> Gravitational Potential Field Model ORSTED05<br />

Ndata<br />

Method 2<br />

¯C20(orth) tcomp[sec]<br />

2520 −0.48445497 · 10−3 4.020252<br />

Table 4: Results of the determination of ¯ C20 using the orthogonality of Legendre<br />

polynomials-method. Ndata is the number of data points used <strong>as</strong> input and tcomp is the<br />

computation time.<br />

Method 1<br />

Ndata 2520 2520 2520 2520<br />

Ngrid 9 × 9 15 × 15 25 × 25 45 × 45<br />

¯C20(int)[10 −3 ] -0.48453368 -0.48457688 -0.48446463 -0.48445504<br />

tcomp[sec] 7.156447 8.528533 10.64067 15.74098<br />

diff. 0.01624 0.02516 0.0019939 1.444 · 10 −5<br />

Table 5: Results of the determination of ¯ C20 using the numerical integration method.<br />

Ndata is the number of data points used <strong>as</strong> input, Ngrid is the number of grid points, diff is<br />

is the relative deviation between the result found by method 1 and 2 in percent and tcomp is<br />

the computation time.<br />

in grid points causes the result from the method 1, to become more and more similar to<br />

the result from method 2.<br />

From this fact it must be concluded that in order to obtain identical results by the two<br />

methods, the grid must be very dense, and therefore method 2 is preferable, due to the<br />

low computation time. Method 2 will therefore be used in the following.<br />

12.3 Collo.f90 error test on EGM96 test data<br />

As described in the latter section, method 2 h<strong>as</strong> been chosen for the full spherical harmonic<br />

coefficient determination, leading to the <strong>Ørsted</strong> potential model ORSTED05. Many<br />

considerations on eliminating programming errors were made while developing collo.f90,<br />

and these were described in section 11.2. To examine the collo.f90 program accuracy, the<br />

program h<strong>as</strong> been tested on EGM96 generated data. Since this data is noise-free it can be<br />

<strong>as</strong>sumed that errors determined from this test run entirely is a result of roundoff errors,<br />

instability and other problems caused by collo.f90. These errors are the so-called model<br />

errors.<br />

The spherical harmonic coefficients up to degree and order 6, which are the result of<br />

this test run on EGM96 generated data, are shown in table 6.<br />

It is important to know how well the program collo.f90 can determine the coefficients of<br />

62


12. Deriving the <strong>Ørsted</strong> Gravitational Potential Field Model ORSTED05<br />

Coefficients found from EGM96 test data<br />

n m Cnm<br />

¯ ¯Snm Erel(Cnm) Erel(Snm)<br />

2 0 −0.48453712 · 10−3 0 0.08<br />

2 1 −0.18703043 · 10−9 −0.15624559 · 10−9 0.02 -113.07<br />

2 2 0.24100089 · 10−5 −0.12058656 · 10−5 -1.19 -13.88<br />

3 0 0.10425540 · 10−5 0 8.91 0.00<br />

3 1 0.19925720 · 10−5 0.22108229 · 10−6 -1.84 -11.04<br />

3 2 0.78616977 · 10−6 −0.56062048 · 10−6 -13.09 -9.44<br />

3 3 0.49183355 · 10−6 0.10355649 · 10−5 -31.79 -26.78<br />

4 0 0.17563786 · 10−5 0 225.33 0.00<br />

4 1 −0.38399811 · 10−6 −0.32358149 · 10−6 -28.40 -31.65<br />

4 2 0.83840095 · 10−7 0.47463012 · 10−6 -76.09 -28.38<br />

4 3 0.64279324 · 10−6 −0.11453986 · 10−6 -35.12 -42.99<br />

4 4 −0.71160763 · 10−7 0.17085108 · 10−6 -62.26 -44.68<br />

5 0 0.30338033 · 10−7 0 -55.73 0.00<br />

5 1 0.71377222 · 10−7 −0.18314154 · 10−7 -214.94 -80.60<br />

5 2 0.42677613 · 10−6 −0.17286081 · 10−6 -34.59 -46.54<br />

5 3 −0.22963628 · 10−6 −0.16159178 · 10−6 -49.19 -24.79<br />

5 4 −0.13503873 · 10−6 0.12584736 · 10−7 -54.27 -74.66<br />

5 5 0.68520757 · 10−7 −0.31734344 · 10−6 -60.84 -52.59<br />

6 0 −0.53044678 · 10−6 0 253.73 0<br />

6 1 −0.37407780 · 10−7 −0.13960414 · 10−7 -50.84 -153.10<br />

6 2 0.16893802 · 10−7 −0.13834618 · 10−6 -64.93 -62.98<br />

6 3 0.30180637 · 10−7 −0.74285183 · 10−8 -47.21 -182.29<br />

6 4 −0.40652870 · 10−7 −0.19502480 · 10−6 -52.85 -58.63<br />

6 5 −0.12069107 · 10−6 −0.22178500 · 10−6 -54.82 -58.66<br />

6 6 −0.67023069 · 10−10 −0.71904223 · 10−7 -100.69 -69.69<br />

Table 6: Spherical harmonic coefficients and their relative errors, determined from<br />

EGM96 test data.<br />

a known noise-free field, when evaluating the coefficients found from <strong>Ørsted</strong> data. In figure<br />

35 it is shown how much the results in table 6 deviates from the true EGM96 coefficients<br />

<strong>as</strong> a function of degree and order, given by the relative error. This relative error Erel is<br />

given by:<br />

<br />

(nm)estimated − (nm)EGM96<br />

Erel =<br />

· 100%, (90)<br />

(nm)EGM96<br />

, where (nm) is the coefficient in question.<br />

It is seen that the relative error from coefficient to coefficient varies a lot, the smallest<br />

being 0.02% on coefficient ¯ C21 and the largest being 253.73% on coefficient ¯ C60 . It is<br />

difficult to determine from where exactly these errors stem. It h<strong>as</strong> not been possible,<br />

63


12. Deriving the <strong>Ørsted</strong> Gravitational Potential Field Model ORSTED05<br />

¯Cnm<br />

Figure 35: Relative error on the coefficients from table 6 given in %.<br />

during this project, to incre<strong>as</strong>e the accuracy of collo.f90 further than these results show.<br />

Neither h<strong>as</strong> it been possible to explain the great variation in the relative error on the<br />

coefficients.<br />

When examining the relative error estimates in figure 35, it is seen that for the majority<br />

of the coefficients, the relative error is negative. This means that the coefficients<br />

estimated by collo.f90 are often too small. Unfortunately it h<strong>as</strong> not been possible to find<br />

an explanation of this either.<br />

In order to convert the relative errors shown in table 35 into errors given in potentials,<br />

the potential field of the coefficients in table 6 h<strong>as</strong> been calculated and subtracted from<br />

the potential field calculated from the true EGM96 coefficients up to order and degree 6.<br />

The potential fields are calculated in ellipsoid height 400km. The result is seen in figure<br />

36.<br />

When examining the result in figure 36, a pattern in the relative errors is seen. The<br />

general pattern shows that the relative errors changes sign in three ”belts” across the earth.<br />

This is a result of the coefficient C40 having a large relative error. It also can be noted<br />

that the error in general follows the amplitude of the field. This is result of the fact that<br />

collo.f90 in general underestimates the coefficients.<br />

12.4 The Model ORSTED05<br />

In table 7 the spherical harmonic coefficients up to degree and order 6, calculated from the<br />

<strong>Ørsted</strong> potential data by collo.f90, are listed. These calculations are b<strong>as</strong>ed on 30319 <strong>Ørsted</strong><br />

data, and the multi-grid box size is 4.5 ◦ × 4.5 ◦ . The box with the le<strong>as</strong>t data contains 0 and<br />

the one with the most contains 26 data. Hence the maximum number of equations solved<br />

64<br />

¯Snm


12. Deriving the <strong>Ørsted</strong> Gravitational Potential Field Model ORSTED05<br />

Figure 36: Difference between the collo.f90 potential field calculated from EGM96 test<br />

data, and the true EGM96 potential field [ m2<br />

s 2 ].<br />

simultaneously by collo.f90 is 9 × 26 = 234.<br />

The relative error on the ORSTED05 coefficients are shown in figure 37. The relative<br />

errors are found by comparison with the true EGM96 coefficients.<br />

¯Cnm<br />

Figure 37: Relative error on the coefficients from table 7 given in %.<br />

It is seen that the relative errors are generally higher that the ones shown in figure 35,<br />

which w<strong>as</strong> off course expected, since the <strong>Ørsted</strong> data contain noise.<br />

It w<strong>as</strong> experienced that the result of the <strong>Ørsted</strong> modeling depended to some extend on<br />

the input parameter which defines the number of data points, and on the box size chosen<br />

for the multi-grid function. Therefore some effort w<strong>as</strong> made to find the best box size for<br />

the calculation of ORSTED05.<br />

65<br />

¯Snm


12. Deriving the <strong>Ørsted</strong> Gravitational Potential Field Model ORSTED05<br />

Coefficients found from <strong>Ørsted</strong> data<br />

n m Cnm<br />

¯ ¯Snm Erel(Cnm) Erel(Snm)<br />

2 0 −0.48430152 · 10−3 0 0.03 0<br />

2 1 0.20918906 · 10−10 0.11526763 · 10−9 -111.19 -90.36<br />

2 2 0.54956953 · 10−7 0.37681631 · 10−7 -97.75 -102.69<br />

3 0 0.49093894 · 10−6 0 -48.71 0<br />

3 1 0.65031951 · 10−7 0.15956609 · 10−6 -96.80 -35.79<br />

3 2 −0.25690223 · 10−8 0.25657993 · 10−7 -100.28 -104.14<br />

3 3 −0.20055761 · 10−7 −0.45693563 · 10−7 -102.78 -103.23<br />

4 0 0.31344998 · 10−6 0 -41.94 0<br />

4 1 0.26589597 · 10−8 0.13587627 · 10−7 -100.50 -102.87<br />

4 2 0.10974940 · 10−6 −0.10235117 · 10−6 -68.71 -115.45<br />

4 3 0.54036757 · 10−7 0.26473021 · 10−8 -94.55 -101.32<br />

4 4 −0.68176511 · 10−7 0.29326264 · 10−7 -63.84 -90.50<br />

5 0 −0.82332096 · 10−8 0 -112.01 0<br />

5 1 0.17745786 · 10−8 0.19571046 · 10−7 -102.86 -120.73<br />

5 2 −0.29136486 · 10−7 0.38796906 · 10−7 -104.47 -112.00<br />

5 3 0.21662527 · 10−7 −0.18331542 · 10−7 -104.79 -91.47<br />

5 4 0.10677581 · 10−7 0.14816158 · 10−7 -103.62 -70.17<br />

5 5 0.26782299 · 10−7 0.35216889 · 10−7 -84.69 -105.26<br />

6 0 0.27876531 · 10−7 0 -118.59 0<br />

6 1 0.22034075 · 10−8 0.18914277 · 10−7 -102.90 -28.05<br />

6 2 0.69187286 · 10−7 −0.79483553 · 10−7 43.62 -78.73<br />

6 3 0.62885721 · 10−8 0.55044245 · 10−8 -89.00 -39.02<br />

6 4 0.19604171 · 10−7 0.10455823 · 10−7 -122.74 -102.22<br />

6 5 0.16016148 · 10−7 0.17809609 · 10−7 -106.00 -103.32<br />

6 6 0.26971528 · 10−7 0.32539687 · 10−7 178.74 -113.72<br />

Table 7: Spherical harmonic coefficients and the relative errors, determined from <strong>Ørsted</strong><br />

data.<br />

66


12. Deriving the <strong>Ørsted</strong> Gravitational Potential Field Model ORSTED05<br />

The re<strong>as</strong>on why only the coefficients up to degree and order 6 h<strong>as</strong> been listed in this<br />

section and the latter, is that it w<strong>as</strong> found that the relative error on the coefficients, found<br />

from the <strong>Ørsted</strong> data, incre<strong>as</strong>ed significantly at the higher degree coefficients. For example<br />

the relative error on coefficient ¯ C70 is Erel = −725.32%. This relative error could not be<br />

explained by model errors, since the relative error on the ¯ C70 coefficient found from EGM96<br />

generated data w<strong>as</strong> only Erel = −52.14%.<br />

B<strong>as</strong>ed on this, it w<strong>as</strong> determined that the potential, derived from <strong>Ørsted</strong> data does at<br />

le<strong>as</strong>t contain information on the Earth’s <strong>gravity</strong> field, which can be described by the first<br />

coefficients up to degree and order 6. Therefore the model ORSTED05 will only consists<br />

of these 25 coefficients and the GM<br />

r found in section 12.1.<br />

For the purpose of evaluating the ORSTED05 model, the potential of the EGM96<br />

coefficients from degree 2 to 6 is shown in figure 38. As expected it is seen that the ¯ C20 is<br />

dominating the image, and in order to get a more detailed potential image, this coefficient<br />

h<strong>as</strong> been subtracted, and the resulting field is shown in figure 39.<br />

Figure 38: The potential field from the true EGM96 coefficients up to degree and order<br />

6 [ m2<br />

s 2 ].<br />

In figure 40, the ORSTED05 potential field calculated from the coefficients in table 7,<br />

is shown and in figure 41 this field is shown with the ¯ C20 term subtracted.<br />

Comparing figure 41 and 39 it is seen that the model calculated from Orsted data<br />

indeed looks similar to EGM96. Obviously there are some derivations, but main features<br />

such <strong>as</strong> the large minimum over India and the maximums over Europe and Indonesia are<br />

in fact observed.<br />

From this it is seen that it h<strong>as</strong> been possible to derive an all-<strong>Ørsted</strong> model, which indeed<br />

contains some information on the Earth’s <strong>gravity</strong> field.<br />

67


12. Deriving the <strong>Ørsted</strong> Gravitational Potential Field Model ORSTED05<br />

Figure 39: The potential field from the true EGM96 coefficients up to degree and order<br />

6 with ¯ C20 subtracted [ m2<br />

s 2 ].<br />

Figure 40: The potential field from the true ORSTED05 coefficients [ m2<br />

s 2 ].<br />

Figure 41: The potential field from the ORSTED05 coefficients with ¯ C20 subtracted [ m2<br />

s 2 ].<br />

68


13 Final Considerations and Discussion<br />

13. Final Considerations and Discussion<br />

In this section, I will summarize and comment on the choices made and results found in<br />

this <strong>thesis</strong>. This is done in acknowledgment of the fact that many of the choices made, h<strong>as</strong><br />

merely been one of many possible.<br />

In section 6.2 the <strong>Ørsted</strong> <strong>satellite</strong> w<strong>as</strong> evaluated <strong>as</strong> a <strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>. Three<br />

criteria were stated which must be fulfilled by a good <strong>gravity</strong> <strong>mission</strong> <strong>satellite</strong>, namely a<br />

low orbital height, uninterrupted GPS tracking and finally it must have instruments on<br />

board which me<strong>as</strong>ure the non-gravitational accelerations.<br />

It w<strong>as</strong> concluded though, that <strong>Ørsted</strong> does not fulfill any of these criteria completely,<br />

and therefore the expectations for the result of the <strong>gravity</strong> field modeling were not high.<br />

It w<strong>as</strong> concluded that it would probably only be possible to derive the coefficients of low<br />

degree of the spherical harmonic expansion of the gravitational potential.<br />

A <strong>gravity</strong> potential field model, ORSTED05, w<strong>as</strong> derived from the <strong>Ørsted</strong> data. This<br />

model contains coefficients up to degree and order 6, which to some extent are comparable<br />

to the known EGM96 coefficients.<br />

Subsequently, some comments on the different parts of this <strong>thesis</strong> are made. Firstly the<br />

data selection is discussed.<br />

Data Selection.<br />

The t<strong>as</strong>k of selecting the <strong>Ørsted</strong> GPS data, which should be used in the <strong>gravity</strong> field<br />

modeling, is described in section 5. The period of data running from 1999-09-02 to 2000-<br />

02-22 w<strong>as</strong> chosen.<br />

This period w<strong>as</strong> chosen b<strong>as</strong>ed on the considerations that the period should be stable in<br />

respect to the solar activity, and the period should also contain a certain amount of useful<br />

data.<br />

It would have been preferable to work with data from a later period. The re<strong>as</strong>on being<br />

that the height of the <strong>satellite</strong> decre<strong>as</strong>es with time, and thereby the number of spherical<br />

harmonic coefficients which can be calculated will incre<strong>as</strong>e. This is due to the attenuation<br />

of the gravitational field with height, which w<strong>as</strong> mentioned in section 6.2.<br />

During the data selection, a choice w<strong>as</strong> made of accepting all me<strong>as</strong>urements b<strong>as</strong>ed on<br />

tracking by 6 or more GPS <strong>satellite</strong>s. It w<strong>as</strong> estimated that 6 <strong>satellite</strong>s were a minimum,<br />

since no DOP values of the me<strong>as</strong>urements were available, but that it is very likely to have<br />

a low DOP value when having at le<strong>as</strong>t 6 <strong>satellite</strong>s.<br />

If a higher number of <strong>satellite</strong>s needed for accepting a me<strong>as</strong>urement had been chosen,<br />

the uncertainties of the me<strong>as</strong>urements would decre<strong>as</strong>e, since the uncertainty of a GPS<br />

69


13. Final Considerations and Discussion<br />

a b<br />

Figure 42: The position of the <strong>satellite</strong> in the X-Y plane (a) after filtering with 7 <strong>satellite</strong>s<br />

(b) after filtering with 8 <strong>satellite</strong>s.<br />

me<strong>as</strong>urement depends on the number of <strong>satellite</strong>s available for the me<strong>as</strong>urement. A decre<strong>as</strong>e<br />

in me<strong>as</strong>urement uncertainty would be an advantage in the modeling, but the choice<br />

of a higher number of <strong>satellite</strong>s would be in conflict with the criteria for uninterrupted<br />

me<strong>as</strong>urements, which is stated in section 6.2.<br />

The direct result of changing the number of <strong>satellite</strong>s from 6 to 7 or 8 is shown in figure<br />

42, which shows that when the number of <strong>satellite</strong>s needed for a good me<strong>as</strong>urement w<strong>as</strong><br />

incre<strong>as</strong>ed to 7 or 8, the GPS me<strong>as</strong>urement can indeed not be considered uninterrupted,<br />

and this w<strong>as</strong> why the choice of 6 <strong>satellite</strong>s w<strong>as</strong> used when selecting the data.<br />

Figure 42 should be compared with figure 7, and it is seen that the incre<strong>as</strong>e in <strong>satellite</strong>s<br />

does create many more ”holes” in the me<strong>as</strong>urements, and it therefore is concluded that<br />

the choice of 6 <strong>satellite</strong>s w<strong>as</strong> preferable.<br />

The Energy Conservation Method.<br />

The energy conservation method w<strong>as</strong> chosen for gravitational potential derivations in this<br />

project. This choice w<strong>as</strong> made primarily due to the fact that the kinetic energy can be<br />

calculated directly from the state vectors. Furthermore the method had earlier been tested<br />

by e.g Eva Howe on CHAMP data with success [Howe et al., 2003, Howe, 2005].<br />

An alternative to the energy conservation method is the acceleration method described<br />

in section 7.1, and the main re<strong>as</strong>on this method w<strong>as</strong> not chose is the fact that it had not<br />

been tested when this project w<strong>as</strong> initialized. Though, it could be interesting to test if<br />

this method could in fact be used with success on <strong>Ørsted</strong> data, since the method h<strong>as</strong> been<br />

tested on CHAMP data with success [Abt, 2004].<br />

70


13. Final Considerations and Discussion<br />

Determination of the Non-Gravitational Disturbing Potential.<br />

In this project, it w<strong>as</strong> chosen to model the non-gravitational potential, F , <strong>as</strong> linear in<br />

time.<br />

F = A · t + E0<br />

This choice w<strong>as</strong> b<strong>as</strong>ed on figure 25, which shows an overall trend which corresponds<br />

with the <strong>as</strong>sumption that the non-gravitational potential decre<strong>as</strong>es at a constant rate in<br />

time.<br />

It is clear though, that this is just an approximation since a complete model of F<br />

should take e.g. <strong>satellite</strong> height, particle height, solar activity and cross section area of the<br />

<strong>satellite</strong> into account. But it w<strong>as</strong> not realistic to derive such a complicated model in this<br />

project.<br />

Two other models of the non-gravitational potential were tested in this <strong>thesis</strong>, namely<br />

1. F = A · t + B · t 2 + E0<br />

2. F (h)<br />

In the first <strong>as</strong>sumption the non-gravitational potential is a second order polynomial in<br />

time, and in the second <strong>as</strong>sumption, F , is a function of <strong>satellite</strong> height instead of time.<br />

The latter is tested since the friction on the <strong>satellite</strong> depends on the particle density in the<br />

atmosphere which again is a function of ellipsoid height, h. It therefore could be expected<br />

that F and h are correlated.<br />

When testing the first <strong>as</strong>sumption, it w<strong>as</strong> found that when fitting the data to a second<br />

order polynomial, the second order term w<strong>as</strong> very small such that:<br />

F = A · t + B · t 2 + E0 ≈ A · t + E0<br />

which brings us back to the <strong>as</strong>sumption of F being linear in time.<br />

When testing the second <strong>as</strong>sumption, non-gravitational potential F w<strong>as</strong> plotted <strong>as</strong> a<br />

function of ellipsoid height instead of time, and the result is seen in figure 43.<br />

It is seen that there is no clear correlation between F and h, and from this observation<br />

it w<strong>as</strong> concluded that this <strong>as</strong>sumption would not be used in this <strong>thesis</strong>.<br />

71


13. Final Considerations and Discussion<br />

Figure 43: The non-gravtational potential F <strong>as</strong> a function of ellipsoid height h.<br />

Calculation of the Spherical Harmonic Coefficients.<br />

Two methods of calculating the spherical harmonic coefficients have been tested in this<br />

project, namely method 1 using numerical integration described in section 10.1 and method<br />

2 using the orthogonality properties of the Legendre polynomials, which w<strong>as</strong> described in<br />

section 10.2.<br />

It w<strong>as</strong> shown that equal results could be achieved using the two methods, when choosing<br />

a number of grid point for the numerical integration method, which w<strong>as</strong> sufficiently<br />

large. But since the computation time of method 1 w<strong>as</strong> found to be longer than method<br />

2, the latter method w<strong>as</strong> chosen for the further modeling.<br />

The kind of grid which w<strong>as</strong> defined for the interpolation using method 1 w<strong>as</strong> shown in<br />

section 10.1, figure 29. This kind of grid is simple and it fulfills the criteria for the Simpson’s<br />

rule of integration stated in section 4.6.<br />

The disadvantage of using this kind of grid is the fact that the grid point density<br />

is largest at the poles, where there are no me<strong>as</strong>urements, see figure 9. Method 1 could<br />

probably be improved by using a more appropriate grid. But since the orthogonality<br />

method w<strong>as</strong> chosen for the modeling, this w<strong>as</strong> not tested.<br />

72


Collo.f90<br />

13. Final Considerations and Discussion<br />

The program collo.f90 w<strong>as</strong> developed during this project for the purpose of determining<br />

the gravitational potential model ORSTED05. The program outline w<strong>as</strong> described in<br />

section 11.1.<br />

During the development of the program collo.f90, many problems arose and solutions<br />

had to be found. These were described in section 11.2.<br />

The multi-grid function w<strong>as</strong> introduced in order to minimize the roundoff errors in the<br />

program, by dividing the Earth into boxes containing a limited amount of data points<br />

and executing the calculations in each box, see figure 33. The use of overlap w<strong>as</strong> also<br />

implemented in the multi-grid function, in order to avoid errors on the edges of the boxes<br />

which stems from the fact, that data in neighboring boxes are <strong>as</strong>sumed uncorrelated when<br />

they are in fact correlated. If this problem should be eliminated, the box-size should be<br />

chosen to be equal to the correlation length found from the covariance function of the<br />

spherical harmonic coefficient in question. In this way, all correlated data points are used<br />

when finding the solutions of each box. But a look at the covariance function of for example<br />

¯C20 shows that the correlation length is so large that the purpose of using the multi-grid<br />

function vanishes if choosing the box-size from this correlation length. Therefore this idea<br />

w<strong>as</strong> not implemented in collo.f90.<br />

Figure 44: The covariance function of ¯ C20.<br />

73


14. Conclusion<br />

14 Conclusion<br />

As described in section 12, an <strong>Ørsted</strong> <strong>gravity</strong> potential model, ORSTED05, h<strong>as</strong> been<br />

calculated. A test run of collo.f90 on EGM96 generated data, showed the errors which<br />

were a result of the numerical calculations themselves.<br />

The first term in the spherical harmonic series, GM, w<strong>as</strong> successfully determined from<br />

<strong>Ørsted</strong> data, in section 12.1. The found value w<strong>as</strong> within 0.0216% of the known value.<br />

The program collo.f90 w<strong>as</strong> used to determine the following spherical coefficients from<br />

the <strong>Ørsted</strong> data. The second term ¯ C20 w<strong>as</strong> found with an accuracy of 0.03%. Higher<br />

degree terms w<strong>as</strong> found but with larger errors compared to EGM96. By comparing the<br />

results with the EGM96 coefficients and the numerical errors found from the test run, it<br />

w<strong>as</strong> concluded that the ORSTED05 model should contain coefficients of degree up to 6.<br />

This choice w<strong>as</strong> b<strong>as</strong>ed on the fact that the relative errors of the higher order coefficients<br />

incre<strong>as</strong>ed significantly.<br />

All in all, it must be concluded that the result w<strong>as</strong> satisfactory. Some <strong>gravity</strong> information<br />

could indeed be derived from the <strong>Ørsted</strong> data, <strong>as</strong> it is clearly seen when comparing<br />

figure 39 and 41.<br />

If the program collo.f90 could be developed further, it is very likely that even more<br />

<strong>gravity</strong> information could be derived from the <strong>Ørsted</strong> data.<br />

74


REFERENCES REFERENCES<br />

References<br />

[Abt, 2004] Abt, T. L. (2004). Determination of a <strong>gravity</strong> field model from one month of<br />

champ <strong>satellite</strong> data using accelerations. Study Thesis.<br />

[Ditmar and van Eck van der Sluijs, 2004] Ditmar, P. and van Eck van der Sluijs, A. A.<br />

(2004). A technique for modeling the earth’s <strong>gravity</strong> field on the b<strong>as</strong>is of <strong>satellite</strong> accelerations.<br />

Journal of Geodesy, pages 12–33.<br />

[E. F. Christensen, 1994] E. F. Christensen, D. (1994). <strong>Ørsted</strong>-satellitten - dansk geomagnetisk<br />

forskningsprojekt. GeologiskNyt, pages 6–8.<br />

[Gerlach et al., 2003a] Gerlach, C., Foldvary, L., Svehla, D., Gruber, T., Wermuth, M.,<br />

Sneeuw, N., Frommknecht, B., Oberndorfer, H., Peters, T., Rothacher, M., Rummel,<br />

R., and Steigenberger, P. (2003a). A champ <strong>gravity</strong> field model from kinematic orbits<br />

using the energy integral. Geophysical research letters, pages 1–4.<br />

[Gerlach et al., 2003b] Gerlach, C., Sneeuw, N., Visser, P., and Svehla, D. (2003b). Champ<br />

<strong>gravity</strong> field recovery using the energy balance approach. Advances in Geosciences, pages<br />

73–80.<br />

[Goovaerts, 1997] Goovaerts, P. (1997). Geostatistics for Natural Resources Evaluation.<br />

Oxford.<br />

[Heiskanen and Moritz, 1967] Heiskanen, W. A. and Moritz, H. (1967). Physical Geodesy,<br />

chapter 7. W. H. Freeman and company, 1st edition.<br />

[Howe, 2005] Howe, E. (2005). Champ <strong>satellite</strong> <strong>gravity</strong> field determination by collocation.<br />

PhD Thesis.<br />

[Howe et al., 2003] Howe, E., Stenseng, L., and Tscherning, C. C. (2003). Analysis of one<br />

month of champ state vector and accelerometer data for the recovery of the <strong>gravity</strong><br />

potential. Advances in Geosciences, pages 1–4.<br />

[Jekeli, 1999] Jekeli, C. (1999). The determination of gravitational potential differences<br />

from <strong>satellite</strong>-to-<strong>satellite</strong> tracking. Celestial Mechanics and Dynamical Astronomy, pages<br />

85–101.<br />

[Kaula, 2000] Kaula, W. M. (2000). Theory of Satellite Geodesy - Applications of <strong>satellite</strong>s<br />

to Geodesy. Dover Publications, inc.<br />

[Kivelson and Russell, 1997] Kivelson, M. G. and Russell, C. T. (1997). Introduction to<br />

space physics. Cambridge University Press.<br />

[Knudsen and Hjorth, 1996] Knudsen, J. M. and Hjorth, P. (1996). Elements of Newtonian<br />

Mechanics. Springer, 2nd edition.<br />

75


REFERENCES REFERENCES<br />

[Kreyszig, 1999] Kreyszig, E. (1999). Advanced Mathematical engineering. John Wiley and<br />

sons, inc., 8th edition.<br />

[Lemoine et al., 1998] Lemoine, F. G., Kenyon, S. C., Factor, J. K., Trimmer, R.,<br />

Pavlis, N. K., Chinn, D. S., Cox, C. M., Klosko, S. M., Luthcke, S. B., Torrence,<br />

M. H., Wang, Y. M., Williamson, R. G., Pavlis, E. C., Rapp, R. H., and<br />

Olson, T. R. (1998). Egm96 - the n<strong>as</strong>a gsfc and nima joint geopotential model.<br />

http://cddis.gsfc.n<strong>as</strong>a.gov/926/egm96/egm96.html.<br />

[Metzler and Pail, 2005] Metzler, B. and Pail, R. (2005). Goce data processing: The spherical<br />

cap regularization approach. Stud. Geophys. Geod., pages 441–462.<br />

[Moritz, 1978] Moritz, H. (1978). Le<strong>as</strong>t-squares collocation. Reviews of Geophysics and<br />

Space Physics, pages 27–35.<br />

[Moritz, 1980] Moritz, H. (1980). Advanced Physical Geodesy. Herbert Wichmann Verlag.<br />

[Press et al., 1992] Press, W. H., Vetterling, W. T., Teukolsky, S. A., and Flannery, B. P.<br />

(1992). Numerical Recipes in Fortran 77 - The Art of Scientific Computing. Cambridge<br />

University Press, 2nd edition.<br />

[Reubelt et al., 2003] Reubelt, T., Austen, G., and Grafarend, E. W. (2003). Harmonic<br />

analysis of the earth’s gravitational field by means of semi-continuous ephemerides of a<br />

low earth orbiting gps-tracked <strong>satellite</strong>. c<strong>as</strong>e study:champ. Journal of Geodesy, pages<br />

257–278.<br />

[Seeber, 2003] Seeber, G. (2003). Satellite Geodesy. de Gruyter, 2nd edition.<br />

[Taylor, 1997] Taylor, J. R. (1997). An introduction to Error Analysis. University Science<br />

Books, 2nd edition.<br />

[Torge, 1999] Torge, W. (1999). Geodesy. de Gruyter, 3rd edition.<br />

[Tscherning, 2001] Tscherning, C. C. (2001). Computation of spherical harmonic coefficients<br />

and their error estimates using le<strong>as</strong>t squares collocation. Journal of Geodesy,<br />

pages 14–18.<br />

[Visser et al., 2003] Visser, P. N. A. M., Sneeuw, N., and Gerlach, C. (2003). Energy<br />

integral method for <strong>gravity</strong> field determination from <strong>satellite</strong> orbit coordinates. Journal<br />

of Geodesy, pages 207–216.<br />

76


A Data Files<br />

A.1 ECI file example<br />

A. Data Files<br />

The eci file contains 13 coulombs. Time is in UTC, positions and velocities are relative to an inertial reference system centered<br />

in the Earth.<br />

#1: <strong>satellite</strong> ID, #2: year, #3: month, #4: day, #5: hour, #6: minute, #7: second #8: x[km], #9: y[km], #10: z[km],<br />

#11: Vx[km/s], #12: Vy[km/s], #13: Vz[km/s]<br />

#1 #2 #3 #4 #5 #6 #7 #8 #9 #10 #11 #12 #13<br />

92 1999 10 10 15 19 7.00 −3983.28270 441.90984 6020.27834 6.11989181 0.83137380 4.01909742<br />

92 1999 10 10 15 26 34.00 −936.32142 754.30908 7130.21862 7.29508572 0.54584941 0.87326629<br />

92 1999 10 10 15 26 35.00 −929.02711 754.85445 7131.08799 7.29606278 0.54506542 0.86576632<br />

92 1999 10 10 15 29 37.00 407.22290 840.35887 7163.81869 7.34574325 0.39209955 −0.50729543<br />

92 1999 10 10 15 29 47.00 480.69588 844.29333 7158.42934 7.34101518 0.38333734 −0.58268589<br />

92 1999 10 10 15 29 57.00 554.08061 848.08429 7152.22797 7.33557524 0.37444586 −0.65802639<br />

92 1999 10 10 15 30 7.00 627.40914 851.81300 7145.28661 7.32918350 0.36535724 −0.73368695<br />

92 1999 10 10 15 30 17.00 700.65498 855.39499 7137.57624 7.32216730 0.35648691 −0.80897753<br />

92 1999 10 10 15 30 27.00 773.84133 858.92222 7129.10386 7.31443313 0.34748437 −0.88403814<br />

92 1999 10 10 15 30 37.00 846.95207 862.37157 7119.88547 7.30593231 0.33832164 −0.95901880<br />

92 1999 10 10 15 30 47.00 919.95377 865.69513 7109.90308 7.29665938 0.32915586 −1.03405949<br />

92 1999 10 10 15 30 57.00 992.87952 868.95237 7099.19768 7.28659732 0.32001301 −1.10895022<br />

92 1999 10 10 15 31 7.00 1065.69333 872.10356 7087.73228 7.27578059 0.31085708 −1.18379099<br />

92 1999 10 10 15 31 17.00 1138.38322 875.16309 7075.51586 7.26420487 0.30180251 −1.25846179<br />

92 1999 10 10 15 31 27.00 1210.93610 878.06984 7062.49644 7.25190662 0.29246722 −1.33319264<br />

92 1999 10 10 15 31 37.00 1283.44160 881.04045 7048.87401 7.23864238 0.28308958 −1.40790353<br />

A.2 JPL file example<br />

The eci file contains 13 coulombs. Time is in GPS-time, positions and velocities are relative to an Earth-fixed reference<br />

system centered in the Earth. #1: <strong>satellite</strong> ID, #2: x[km], #3: y[km], #4: z[km], #5: Vx[km/s], #6: Vy[km/s] , #7:<br />

Vz[km/s], #8: year, #9: month, #10: day, #11: hour, #12: minute, #13: second<br />

#1 #2 #3 #4 #5 #6 #7 #8 #9 #10 #11 #12 #13<br />

92 1043.59400 −3869.18100 6020.45900 −3.29200000 5.31719000 4.01878000 1999 10 10 15 19 20.00<br />

92 −397.59030 −1134.56600 7130.24400 −3.03950000 6.72036000 0.87290000 1999 10 10 15 26 47.00<br />

92 −400.62860 −1127.84600 7131.11300 −3.03811000 6.72198000 0.86540000 1999 10 10 15 26 48.00<br />

92 −927.03960 115.01480 7163.77700 −2.72954000 6.89527000 −0.50766000 1999 10 10 15 29 50.00<br />

92 −954.30270 183.99780 7158.38400 −2.70971000 6.89759000 −0.58305000 1999 10 10 15 30 0.00<br />

92 −981.30140 252.98300 7152.17900 −2.68953000 6.89924000 −0.65839000 1999 10 10 15 30 10.00<br />

92 −1008.12300 321.97430 7145.23400 −2.66886000 6.90002000 −0.73405000 1999 10 10 15 30 20.00<br />

92 −1034.67900 390.97330 7137.52000 −2.64820000 6.90010000 −0.80934000 1999 10 10 15 30 30.00<br />

92 −1061.06400 459.97220 7129.04400 −2.62719000 6.89951000 −0.88440000 1999 10 10 15 30 40.00<br />

92 −1087.25100 528.96280 7119.82200 −2.60579000 6.89821000 −0.95938000 1999 10 10 15 30 50.00<br />

92 −1113.18400 597.92830 7109.83600 −2.58415000 6.89614000 −1.03442000 1999 10 10 15 31 0.00<br />

92 −1138.93000 666.88080 7099.12700 −2.56229000 6.89328000 −1.10931000 1999 10 10 15 31 10.00<br />

92 −1164.44000 735.79840 7087.65800 −2.54019000 6.88967000 −1.18415000 1999 10 10 15 31 20.00<br />

92 −1189.72400 804.66480 7075.43800 −2.51796000 6.88527000 −1.25882000 1999 10 10 15 31 30.00<br />

92 −1214.72000 873.48640 7062.41500 −2.49525000 6.88023000 −1.33355000 1999 10 10 15 31 40.00<br />

92 −1239.66200 942.27980 7048.78900 −2.47221000 6.87426000 −1.40826000 1999 10 10 15 31 50.00<br />

A.3 RINEX file example<br />

year month day hour minute second ok # <strong>satellite</strong>s ID ID ID<br />

02 2 20 8 20 50.0000000 0 2 1 21<br />

sat 1 data −0.05148 −0.04040 28912088.24948 28912088.24940 139.00048<br />

0.00040<br />

sat 21 data −0.85148 −0.66340 28324541.45848 28324550.65340 256.00048<br />

0.00040<br />

02 2 20 8 21 0.0000000 0 2 1 21<br />

sat 1 data 303265.83247 236311.03840 28969796.76447 28969796.76440 89.00047<br />

0.00040<br />

sat 21 data −20438.44348 −15926.05940 28320654.28048 28320646.97640 258.00048<br />

0.00040<br />

02 2 20 8 21 10.0000000 0 3 1 21 11<br />

sat 1 data 606423.40747 472537.72040 29027486.58147 29027480.28440 55.00047<br />

0.00040<br />

sat 21 data −42479.17848 −33100.65840 28316462.55548 28316451.22740 262.00048<br />

0.00040<br />

sat 11 data −0.03748 −0.02940 28570978.87048 28570978.87040 308.00048<br />

0.00040<br />

02 2 20 8 21 20.0000000 0 2 21 11<br />

sat 21 data −66133.20148 −51532.36540 28311963.94948 28311975.08940 260.00048<br />

0.00040<br />

sat 11 data 399865.47348 311583.48640 28647062.12348 28647062.12340 306.00048<br />

0.00040<br />

77


B. EGM96 coefficients<br />

B EGM96 coefficients<br />

EGM96 spherical harmonic coefficients of<br />

degree n and order m up to (n,m)=(7,7)<br />

n m Cnm<br />

¯ ¯Snm<br />

2 0 -0.484165371736E-03 0.000000000000E+00<br />

2 1 -0.186987635955E-09 0.119528012031E-08<br />

2 2 0.243914352398E-05 -0.140016683654E-05<br />

3 0 0.957254173792E-06 0.000000000000E+00<br />

3 1 0.202998882184E-05 0.248513158716E-06<br />

3 2 0.904627768605E-06 -0.619025944205E-06<br />

3 3 0.721072657057E-06 0.141435626958E-05<br />

4 0 0.539873863789E-06 0.000000000000E+00<br />

4 1 -0.536321616971E-06 -0.473440265853E-06<br />

4 2 0.350694105785E-06 0.662671572540E-06<br />

4 3 0.990771803829E-06 -0.200928369177E-06<br />

4 4 -0.188560802735E-06 0.308853169333E-06<br />

5 0 0.685323475630E-07 0.000000000000E+00<br />

5 1 -0.621012128528E-07 -0.944226127525E-07<br />

5 2 0.652438297612E-06 -0.323349612668E-06<br />

5 3 -0.451955406071E-06 -0.214847190624E-06<br />

5 4 -0.295301647654E-06 0.496658876769E-07<br />

5 5 0.174971983203E-06 -0.669384278219E-06<br />

6 0 -0.149957994714E-06 0.000000000000E+00<br />

6 1 -0.760879384947E-07 0.262890545501E-07<br />

6 2 0.481732442832E-07 -0.373728201347E-06<br />

6 3 0.571730990516E-07 0.902694517163E-08<br />

6 4 -0.862142660109E-07 -0.471408154267E-06<br />

6 5 -0.267133325490E-06 -0.536488432483E-06<br />

6 6 0.909789371450E-07 0.000000000000E+00<br />

7 1 0.279872910488E-06 0.954336911867E-07<br />

7 2 0.329743816488E-06 0.930667596042E-07<br />

7 3 0.250398657706E-06 -0.217198608738E-06<br />

7 4 -0.275114355257E-06 -0.123800392323E-06<br />

7 5 0.193765507243E-08 0.177377719872E-07<br />

7 6 -0.358856860645E-06 0.151789817739E-06<br />

7 7 0.109185148045E-08 0.244415707993E-07<br />

78


C Fortran programs<br />

C. Fortran programs<br />

This cd contains the three fortran programs rinexp<strong>as</strong>s2.f90, stat2pot2.f90<br />

and collo.f90, which have been developed for this project.<br />

79

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!