Eth 41591 02

Diss. ETH No.
MANTLE ELECTRICAL CONDUCTIVITY ESTIMATES

FROM GEOMAGNETIC JERK OBSERVATIONS
A dissertation submitted to
ETH Z URICH
(Swiss Federal Institute of Technology)
for the degree of
D OCTOR OF S CIENCES
presented by
K ATIA JASBINSCHEK DOS R EIS P INHEIRO
M.Sc. Geophysics,
National Observatory, Brazil
born 21st February, 1979
citizen of Brazil
accepted on the recommendation of
Prof. Dr. A. Jackson, examiner
Prof. Dr. M. Mandea, co-examiner
P.D. Dr. A. Kuvshinov, co-examiner
2009
"What is really good is to fight with determination,
embrace life and live it with passion!
Loose your battles with class and dare to win
because the world belongs to those who dare and life,
Life is worth too much to be insignificant..."
Charles Chaplin
ii
Abstract
The electrical conductivity of the Earths mantle has been a subject of much debate in the last
few years. Induction studies agree mainly in the first 1000 km of the mantle, however in the
lower mantle the conductivity is still very uncertain. Experimental studies of mineral physics
simulating the conditions of the deep mantle have been performed and results disagree by 3 orders
of magnitude depending, for example, on the considered geotherm and composition of the lower
mantle. As a complement to mineralogical and induction studies, time variations of the magnetic
field in the core can also contribute towards a better understanding of mantle conductivity.
Geomagnetic jerks involve abrupt temporal changes in the secular variation of Earths magnetic
field and are believed to be due to motions in the fluid core. In this thesis we use geomagnetic jerk
observations to constrain information about the mantle electrical conductivity. We modelled the
secular variation, around the time of a jerk, by two straight-line segments; their intersection defines
the jerk occurrence time and the difference between the two slopes defines the jerk amplitude. The
jerks morphology was obtained by data analysis of observatory annual means from which we built
a spherical harmonic model. The innovative aspect of the data analysis was the evaluation of error
bars in both jerk occurrence time and amplitude. We detected global jerks occurring at 1969, 1978,
and 1991 that show different time arrivals at the surface of the order of 3 years.
There are two possible hypotheses to explain the different occurrence jerk times at the Earths
surface: the first is to consider these differential time delays generated by dynamical processes in
the core which do not occur simultaneously; the second is to consider jerks generated instanta-
neously in the core and the time delays caused by a conducting mantle. In this thesis we analyze
the second hypothesis so that the geomagnetic field observed at the surface will correspond to a
filtered version of the original field generated in the core.
We developed the forward approach to this problem using Backus [1983] mantle filter theory,
in which a 1D mantle conductivity model acts as linear, causal and time-invariant filter. The
jerk is simulated as an impulse in time at the CMB and its morphology in the core obtained by
a global spherical harmonic model. The key point is that the mantle filter is different for each
harmonic degree. Therefore, as the mixing of harmonics varies with location at the Earths surface,
distinct time delays will exist in different location at the Earths surface and for different jerk events
and components of the magnetic field. We have demonstrated, by simple examples, exactly how
different emergence times can occur at the Earths surface.
However, in order to constrain some information about the electrical conductivity of the mantle
one needs to solve the inverse problem. A very interesting aspect is that the same mantle electrical
conductivity would generate different delay times in different global jerks (as the 1969, 1978 and
1991 jerks). We used Velimsky and Martinecs [2005] approach to solve exactly the diffusion
equation for 1D mantle models to calculate the Impulse Response Functions (IRFs).
The 1-D mantle conductivity model of Kuvshinov & Olsen (2006) was adopted up to 700 km
depth and below that four simulations were performed with electrical conductivities varying from
iii
1000 S/m to 0.4 S/m: the first calculation considers a uniform one-layer model, the second a two-
layer model with the lower mantle divided into two layers of 1250 km (bottom) and 950 km (top);
a third simulation considering a uniform lower mantle 1900 km thick and a D 300 km thick.
The three-layer model assumes the lower mantle divided into two layers of 950 km thick, and a
bottom layer 300 km thich, simulating the D . The output model secular variation was treated
in the same way as the data and the misfit between data and synthetics was evaluated in order to
find possible electrical conductivity models for the lower mantle that explain the observed jerk
differential delays.
We used both spherical harmonic models of the 1969 jerk (Y component) calculated in this
work and by Le Huy et al. (1998). The results showed to be consistent with similar patterns of
low misfit values for a lower mantle about 1 S/m and allowing a broad variation for the D . Our
results of variations of electrical conductivity in D showed that our data is not sensitive to those
variations in deep and thin layers. The amplitude of most low misfit models indicate differential
delays on the order of 0.1 yr and by our data analysis the differential delays are on the order of 1.5
yr. However, the patterns early/late for differential delays of the minimum misfit models are in a
good agreement with the results found in the data analysis of the Y component of the 1969 jerk.
iv
Zusammenfassung
Die elektrische Leitfhigkeit des Erdmantels wird seit einigen Jahren kontrovers diskutiert. Auch
wenn frhere Studien in den ersten 1000 km des Mantels bereinstimmungen aufweisen, ist die
elektrische Leitfhigkeit des oberen Erdmantels noch nicht gnzlich geklrt. Experimentelle Un-
tersuchungen der Mineralphysik, welche die Gegebenheiten des unteren Mantels simulieren, we-
ichen um die Grssenordnung von 3 von den simulierten Ergebnissen ab, wenn die geothermische
Zusammensetzung des unteren Mantels bercksichtigt wird. Ergnzend zu den mineralogischen
und einfhrenden Studien, knnen zeitliche Variationen des magnetischen Feldes im Erdkern eben-
falls zu einem besseren Verstndnis der Mantelleitfhigkeit beitragen.
Geomagnetische Jerks, die abrupte nderungen der Sekularvariation des Erdmagnetfeldes mit
sich bringen, sind auf Bewegungen im flssigen Teil des Erdkerns zurckzufhren. In dieser
Dissertation werden geomagnetische Beobachtungen genutzt, um Informationen ber die elek-
trische Leitfhigkeit des Mantels zu extrahieren. Die Sekularvariationen, die zeitlich geomagnetis-
che Jerks einschliessen, wurde mit zwei linearen Abschnitten modelliert, deren Schnittpunkt das
Ereignis zeitlich definiert. Die Differenz der zwei Anstiege beschreibt die Amplitude der geo-
magnetischen Jerks. Die Form der geomagnetischen Jerks wurde anhand von Datenanalysen der
jhrlichen Mittel eines Observatoriums erzielt, um daraus ein sphrisch harmonisierendes Mod-
ell zu erstellen. Der neuartige Aspekt der Datenanalyse war die Entwicklung der Abweichungen
des zeitlichen Auftretens und der Amplitude der geomagnetischen Jerks. Es konnten globale geo-
magnetische Jerks 1969, 1978 und 1991 nachgewiesen werden, deren zeitlich Eintrittszeit auf der
Erdoberflche um 3 Jahre variiert. Es existieren zwei Hypothesen, diese zeitlichen Unterschiede
zu erklren: Zum einem sind die verschiedenen Zeitverzgerungen zu bercksichtigen, die durch
dynamische Prozesse im Kern entstehen, welche aber nicht gleichzeitig auftreten, zum anderen
ist zu bercksichtigt, dass Jerks sofort im Kern generiert werden und die Zeitverzgerung durch
den leitfhigen Mantel erzeugt wird. In dieser Dissertation wird die zweite Hypothese analysiert,
so dass das geomagnetische Feld, welches an der Oberflche beobachtet wurde mit der gefilterten
Version des originalen Feldes des Kerns selbst korrespondiert.
Fr dieses Problem wurde die Mantelfilter Theorie nach Backus [1983] weiter entwickelt,
in dem ein 1D Mantelleitfhigkeitsmodell linear, kausal und zeitlich unverzgert fungiert. Jerks
sind simuliert als ein Impuls der Zeit zur CMB und seine Morphologie im Kern wird durch
ein globales sphrisch harmonisches Modell erreicht. Besondere Aufmerksamkeit sollte auf den
Mantelfilter gelegt werden, der fr jedes harmonische Grad unterschiedlich ist. Deutlich aus-
geprgte Zeitverzgerungen an unterschiedlichen Orten der Erdoberflche, fr unterschiedliche
Jerk Ereignisse und Komponenten des Magnetfeldes existieren als Folge einer Vermischung von
harmonischen Abweichungen. Weiterhin haben wir anhand von einfachen Beispielen aufgezeigt,
wie sich die unterschiedlichen Entstehungszeiten an der Erdoberflche ereignen. Um jedoch einige
Informationen ber die elektrische Leitfhigkeit des Mantels einzuschrnken, muss das inverse
Problem gelst werden. Sehr interessant ist, dass dieselben elektrischen Leitfhigkeiten des Man-
v
tels fr verschiedene Zeitverzgerungen und verschiedene globale Jerks (1969,1978 und 1991)
generiert wurden. Wir nutzten die Herangehensweise von Velmnsk und Martinec [2005], um die
Durchdringungsformel fr das 1D Mantelmodell exakt zu lsen und die Impulsantwortfunktion
(IRFs) zu berechnen.
Das 1D Mantelleitfhigkeitsmodell von Kuvshinov & Olsen [2006] wurde fr eine Tiefe bis
700 km bernommen. Der darunterliegende Teil wurde mit Hilfe von den folgenden 4 Simula-
tionen , denen unterschiedliche elektrische Leitfhigkeiten zu Grunde liegen (1000 S/m bis 0.4
S/m) durchgefhrt. Die erste Berechnung basiert auf einem einheitlichen Ein-Lagen Modell. In
einer zweiten Berechnung wird ein Zwei-Lagen Modell angenommen, dass den unteren Mantel
in eine untere, 1250km dicke und eine obere, 950 km dicke Schicht unterteilt. Bei der dritten
Simulation handelt es sich um einen 1900 km dicken, einheitlichen unteren Mantel und einem
D mit einer Dicke von 300 km. In dem drei-Lagen Modell wird der untere Mantel in zwei
Schichten von je 950 km Dicke und einen obere Mantel hat eine Dicke von 300 km, welches D
simuliert. Das Ergebnismodell der Sekularvariationen wurde so bearbeitet wie die Daten selbst,
wobei die beobachteten und synthetisierten Daten evaluiert wurden. Ein denkbares elektrisches
Leitfhigkeitsmodell fr den unteren Mantel kann daraus abgeleitet werden, dass die beobachteten
differentiellen Verzgerungen der Jerks erklren kann.
In dieser Arbeit wurde das sphrisch harmonische Modell, basierend auf Le Huy et al. (1998),
genutzt, um das Jerk von 1969 (Y-Komponente) zu berechnen. Die Ergebnisse waren konsistent,
denn mit elektrischen Leitfhigkeiten von ungefhr 1 S/m fr den unteren Mantel wurden auch
nur geringe Abweichungen simuliert und der Verlauf der Kurven war sehr hnlich. Dies erlaubt
eine weite Variation fr D . Variationen der elektrischen Leitfhigkeit in D zeigen, dass diese
Daten nicht sensitiv gegenber varierender Tiefe und Schicktdicke sind. Wenn die Amplituder
der Abweichung von den beobachteten und den synthetischen Daten klein ist, deutet dies auf dif-
ferenzielle Verzgerungen von etwa 0.1 Jahren hin. Die differenziellen Verzgerungen der Jerks
unserer Datenanalyse ergab etwa 1.5 Jahre. Dennoch, das Muster des Modells mit der geringsten
Abweichung fr den zeitlichen Verlauf (frher/spter) der Jerks, stimmt gut mit den Ergebnissen
berein, die uns die Danenanalyse ber die Y-Komponente des Jerks von 1969 lieferte.
vi
Acknowledgements
I would like first to acknowledge Prof. Andrew Jackson for supervising my PhD research over the
past four years. During this time, he has brought many innovative, interesting and challenging ideas
that have provided constant incentives for the development of this research. His broad knowledge
in geophysics and enthusiasm about science have guided me on my learning process.
In Leeds, I would like to thank Ashley Willis and Chris Davies for discussions on geomag-
netism and for the enjoyable days; Mathieu Dumberry and Nicolas Gillet from whom I have
received important suggestions for my PhD project; my friends Christine Souque and Cristiano
Lana for his friendship and support; and to those who were my family in England: Jayne Harnett,
Regina, Claire and Mike King. In Australia, I am grateful to Malcolm and Julie Sambridge for
their support and all the happy moments; to Catriona Macfarlane and Ian Dias for their friendship
and help in Canberra.
During the three years I spent in Zurich, I have had the opportunity to be part of the Earth
and Planetary Magnetism Group from which I have benefited from the interesting discussions and
seminars. I would like to thank my working colleagues from the EPM group for these enjoy-
able and profitable years: Bjarne Almqvist, Claudio DAddario, Hakon Fischer, Andreas Gehring,
Prof. Ann Hirt, Amir Khan, Jessica Kind, Kuan Li, Prof. William Lowrie, Philippe Marti, Sanja
Panovska, Alexey Semenov and Andrey Sheyko. In particular, I shall express my gratitude to Chris
Finlay who has made an important contribution to my thesis through many discussions, advices
and excellent ideas. I am also very grateful to Alexei Kuvshinov for his always constructive crit-
icism, encouragement and for kindly providing his electrical conductivity model results which I
have used in this thesis. In the last 2 years of my PhD, Jakub Velmnsk has given me great sug-
gestions and help. In addition, he has provided me his forward solver code which was fundamental
for the development of the inverse problem. I thank my dear colleagues, who worked in the EPM
group up until last year, Erik Spence and Martha Evonuk, for the important advice and for the
pleasant days in Zurich.
During conferences and workshops, I had the opportunity to discuss my PhD project, particu-
larly, with Nils Olsen, Cathy Constable, Susan Macmillan and Mioara Mandea, who have provided
constructive criticism as well as valuable insights into my project. I am also grateful to Nils Olsen
for kindly providing the 2003 geomagnetic jerk morphology model. I would also like to mention
that the two most impressive courses I have taken during my PhD were mantle dynamics given by
Geoffrey Davies in ANU, Australia and Inverse Theory by Albert Tarantola, in Neuchatel. I am
grateful to both Davies and Tarantola for inspiration in science and proving that a scientist can also
be concerned about social issues.
For their williness in helping and solving administrative issues I thank Elisabeth Laederach,
Andr Blanchard and Sabine Raess. For financial support I thank the Dorothy Hodgkin Scholarship
Award, the Australian National University (Canberra) and ETH-Zurich. I am also grateful for an
ETH student award that contributed to the air ticket which allowed me to participate in a conference
vii
in Brazil (2007) and workshop in Japan (2008).

I would like to thank very much Beatriz Quintal for her friendship, dedication and help, espe-
cially in the difficult and decisive moments of my PhD. My dear friend Manuele Faccenda with
who I shared most of my journey in Zurich, I thank for the constant encouragement and friend-
ship. I thank also Manuele for the assistance in Matlab, which enable me to build some beautiful
figures for the thesis. I am glad to have met Negar Haghipour and I am grateful for her friendship
and for many happy moments we had together. To my friend Claudio Villatoro I thank for the
important conversations, which provided me creativity, courage and important tools to go forward.
Other special friends I want to mention are Paul Hilary, Silke Nissen, Sanja Panovska, Weronika
Gorczyk and Claudio Delle Piane. I thank them for their friendship and enjoyable days in Zurich.
I thank Chris Streule for the instigating conversation about other interesting matters besides
science, for the happy moments and for his care and dedication during the last months of my PhD.
I am very grateful to who I consider my dear Brazilian family in Zurich: Crisogono Vasconcelos
and Gisele Ferola, Beatriz Quintal, Sibylle Leuzinger, Cintia de Oliveira, Fabiola Dellatti and
Luciene Penz, Ana Velloso and Fernando, for all the support, help, care and great company during
these 3 years in Zurich.
I wish to make a special acknowledgement to my dearest friend Patricia Galvo who, even
from Brazil, was always present during these 4 years of my PhD. Her friendship, care and advices
were very valuable and always a source of strength and encouragement to me. I also thank Jandyr
Travassos for believing in me since the beginning of my journey and for constantly challenging me
to confront problems with courage and determination.
I want to express my deep gratitude to my dear parents, Euclides and Marcia Pinheiro, to whom
I dedicate my work and all my victories during my PhD. I thank them for their constant love and
dedication, and for being the source of inspiration for my work.
viii
Contents
1 Introduction 1
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Earth Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Geomagnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.1 Geomagnetic Field Sources . . . . . . . . . . . . . . . . . . . . . . . . . 4
Laplaces Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3.2 Geomagnetic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Geomagnetic Jerks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Mantle Conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5.1 Mineral Physics Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5.2 Induction Studies: external and internal magnetic fields . . . . . . . . . . 11
1.6 Thesis objectives and structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2 Theory of mantle screening 17

2.1 Mathematical description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Backus mantle filter theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.1 Backus low frequency approximation . . . . . . . . . . . . . . . . . . . . 21
2.2.2 Backus [1983] linear kernels . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 Limitations of Backus mantle filter theory . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3 Data Analysis of Geomagnetic Jerks 35

3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 Methods to Detect Geomagnetic Jerks . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 Limitations on the Analysis of Geomagnetic Jerks . . . . . . . . . . . . . . . . . . 37
3.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.5 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.5.1 Jerk occurrence time and amplitude detection . . . . . . . . . . . . . . . . 43
3.5.2 Spherical harmonic model for jerk amplitudes . . . . . . . . . . . . . . . . 44
3.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.6.1 Jerk time occurrence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.6.2 Jerk morphology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
ix
4 The Forward Problem 69

4.1 Synthetic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.1.1 A simple model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.1.2 A more complex model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2 Geophysical application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5 The Inverse Problem 81

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.3.1 One-layer model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.3.2 Two-layer models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Sensitivity to D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.3.3 Three-layer models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6 Conclusions 100
6.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
A Data analysis tables 103
B Tables with misfit values obtained in the inversion 118

x
List of Figures
1.1 Schematic cross section of the Earth indicating seismological regions and pres-
sures as a function of depth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Profiles of seismic velocities and density with depth in the earth. . . . . . . . . . 2
1.3 Schematic cross section of the D zone. . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Magnetic field components. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Distribution of magnetic observatories currently operating on the Earths surface. . 5
1.6 Spherical coordinate system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.7 Spatial spectrum of the internal geomagnetic field at 2000. . . . . . . . . . . . . . 6
1.8 Eastward component registered in five European magnetic observatories. . . . . . 8
1.9 Secular variation of East component of Chambon la Fort observatory in France
and Niemegk observatory in Germany . . . . . . . . . . . . . . . . . . . . . . . 9
1.10 Simple schematic of secular variation for a hypothetical component. . . . . . . . 9
1.11 Secular variation of the East component (Y) of Eskdalemuir and Gnangara obser-
vatories. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.12 Electrical conductivity as a function of depth obtained from mineral physics studies. 12
1.13 Results of 1-D profiles of mantle electrical conductivity obtained from induction
studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1 Radial models of mantle electrical conductivity. . . . . . . . . . . . . . . . . . . 23

2.2 Impulse Response Functions (A, B) and Transfer Functions (C, D) for the two ra-
dial electrical conductivity models, shown in equation 2.27 and Figure 2.1. Plots
A and C correspond to the higher conducting model (c = 3000 and = 11) while
plots B and D correspond to the weaker conducting model (c = 100 and = 8). . 24
2.3 Example of the delay and smoothing caused by a 1D mantle conductivity given
by equation 2.27, with c = 1000 S/m and = 3 for the 1978 jerk spherical
harmonic model given by Le Huy et al. (1998). The black line would be the jerk
for an insulating mantle and the red line for a conducting mantle. . . . . . . . . . 26
2.4 Kernels (equation 2.69, (r)) for the harmonic degrees: = 1, = 2, = 5,
= 6, = 11 and = 12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.5 Examples of CIRFs calculated for the hypothetical locations in Table 2.1. . . . . . 29
2.6 Examples of CIRFs (solid line) calculated for the 1978 spherical harmonic model
of Le Huy et al. (1998) of ten magnetic observatories (Table 2.2). A radial electri-
cal conductivity model (equation 2.27 for c = 1000 S/m and = 3) was adopted. 30
2.7 Integrands of the smoothing time (F(t)(t 1 )2 ) given in equation 2.47 of ten
observatories (Table 2.2). A radial electrical conductivity model (equation 2.27
for c = 1000 S/m and = 3) was adopted. . . . . . . . . . . . . . . . . . . . . 30
xi
2.8 Convolution of a jerk with A = 1 (black line) with the CIRF (red line) and with
the Gaussian approximation (blue line) of the ten observatories used as example
(Table 2.2). The same radial model as in Figures 2.6 and 2.7 (equation 2.27 for
c = 1000 S/m and = 3) was adopted. . . . . . . . . . . . . . . . . . . . . . . 31
2.9 Fit of two straight-lines segments, by least-squares, to the convolved jerk with
CIRF of observatory 6 (a) and 9 (c) and for the Gaussian for observatory 6 (b)
and 9 (d). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.10 Example of the calculation of composite kernels. . . . . . . . . . . . . . . . . . . 33
2.11 Composite and differential kernels. . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1 Secular variation estimates for the X component of the magnetic field using dif-
ferent datasets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Secular variation estimates for the Y component of the magnetic field using dif-
ferent datasets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3 Secular variation estimates for the Z component of the magnetic field using dif-
ferent datasets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4 Observatory annual differences in geographical (red triangles) and geomagnetic
dipole (black triangles) coordinates. . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.5 Examples of normalized Probability Distribution Functions (PDFs). . . . . . . . . 45
3.6 Example of L2 fitting of two straight line segments to the secular variation of Y
component of Fuerstenfeldbruck observatory. . . . . . . . . . . . . . . . . . . . . 46
3.7 Results for the 1969 jerk occurrence time at the Earths surface obtained by the
L2 method for the X , Y and Z components of the magnetic field. . . . . . . . . . . 49
3.8 Error bars for the 1969 jerk occurrence time for a selection of magnetic observa-
tories in the X , Y and Z components of the magnetic field. . . . . . . . . . . . . . 50
3.9 Results for the 1978 jerk occurrence time at the Earths surface by the least-
squares method for the X , Y and Z components of the magnetic field. . . . . . . . 51
3.10 Error bars for the 1978 jerk occurrence time for a selection of observatories in the
X , Y and Z components of the magnetic field. . . . . . . . . . . . . . . . . . . . 52
3.11 Results of the 1991 jerk occurrence time at the Earths surface by the L2 method
for the X , Y and Z components of the magnetic field. . . . . . . . . . . . . . . . . 53
3.12 Error bars for the 1991 jerk occurrence time for a selection of observatories in
the X , Y and Z components of the magnetic field using first differences of annual
means and the L2 method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.13 Results for the 1999 jerk occurrence time at the Earths surface by the L2 method
to the X , Y and Z components of the magnetic field by considering first differences
of annual means. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.14 Examples of the first differences of annual means (red triangles) at four magnetic
observatories used to analyse the possible 1999 jerk. . . . . . . . . . . . . . . . . 56
3.15 Jerk amplitudes (nT/yr2 ) for the 1969 jerk detected by the L2 method, by using
first differences of observatory annual mean data. . . . . . . . . . . . . . . . . . 57
xii
3.18 Spherical harmonic models of the 1969 geomagnetic jerk using simultaneous L2
analysis of the first differences of observatory annual means for the X , Y and Z
components and different damping parameters. . . . . . . . . . . . . . . . . . . . 60
3.19 Lowes spectra for the 1969 jerk spherical harmonic models shown in Figure 3.18
for each damping parameter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.20 Trade-off curves considering the spherical harmonic models, using simultaneous
analysis of the slopes obtained by the L2 method applied to first differences of
annual means for the X , Y and Z components of the 1969, 1978 and 1991 jerks. . 62
3.21 Spherical harmonic models of the X , Y and Z components of the 1969 geomag-
netic jerk using different damping parameters. . . . . . . . . . . . . . . . . . . . 62
3.24 Lowes spectra of the 1969, 1978 and 1991 spherical harmonic models using si-
multaneous analysis of the jerk amplitudes in X , Y and Z components derived
from L2 analysis of first differences of observatory annual means for the damping
parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.1 Delay times calculated for different harmonic degrees and considering a radial
mantle conductivity model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 Secular variation of the X (A), Y (B) and Z (C) components of Niemegk observa-
tory, in Germany. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.3 Hypothetical spherical harmonic models of X , Y and Z used as a simple example
to calculate the CIRF and output jerk. . . . . . . . . . . . . . . . . . . . . . . . . 71
4.4 Examples of CIRFs (A, B) and output jerks (C,D) in two locations for X (green
line), Y (purple line) and Z (red line) components. . . . . . . . . . . . . . . . . . 72
4.5 Power Spectrum of the two synthetic models of jerk amplitudes. . . . . . . . . . . 73
4.6 Synthetic spherical harmonic models evaluated from the synthetic models at the
core-mantle boundary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.7 Results of the forward modelling calculation of time delays for the two and three-
peaked models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.8 Synthetic examples the first, second and third time derivatives for an insulating
mantle and for a radial mantle conductivity model. . . . . . . . . . . . . . . . . . 75
4.9 Spherical harmonic model obtained from Le Huy et al. (1998) of the East com-
ponent of the 1969 geomagnetic jerk at the Earths surface (in A). Spherical har-
monic models of each harmonic degree at the Earths surface for: = 1 (in B),
= 2 (in C), = 3 (in D) and = 4 (in E). . . . . . . . . . . . . . . . . . . . . . 76
4.10 Morphology of the 2003 geomagnetic jerk for the East component obtained from
Olsen & Mandea (2007). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.11 Impulse Response Function multiplied by the amplitude value for each harmonic
degree () at the location corresponding to LAquila magnetic observatory. The
sum of these curves result in the Composite Impulse Response Function. . . . . . 77
4.12 Results of delay times of the 1969 and 2003 jerks for the Y component. . . . . . . 78
xiii
4.13 Synthetic jerks in the case of an insulating mantle (black V shape) and of a
conducting mantle, which is the result of the convolution between the CIRF of
each observatory and the input jerk at the CMB. . . . . . . . . . . . . . . . . . . 79
5.1 Comparison between analytical and numerical solutions of IRFs. . . . . . . . . . 83

5.2 Spherical harmonic models of the 1969 jerk amplitudes calculated in this work
(Chapter 3) and by Le Huy et al. (1998) that are used to calculate the CIRFs. . . . 84
5.3 Scheme of time delays (1 ) and differential delays (1 ) simulating four different
locations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.4 Histogram of the differential delays measured on the annual means data corre-
sponding to the 1969 jerk (Y component). . . . . . . . . . . . . . . . . . . . . . 85
5.5 Results for the 1969 jerk occurrence time at the Earths surface obtained by the
L2 method for the Y component of the magnetic field. . . . . . . . . . . . . . . . 86
5.6 IRFs of a one-layer model simulating the lower mantle with varying electrical
conductivities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.7 Examples of CIRF s and CIRFs for a given electrical conductivity model with
three layers and for considering the spherical harmonic model of the 1969 jerk
(Chapter 3). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.8 Examples of annual means of the output jerk for a one-layer model, and the fitting
of two straight line segments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.9 Misfit for the one-layer model considering the 1969 jerk spherical harmonic model
in Chapter 3 and taken from Le Huy et al. (1998). . . . . . . . . . . . . . . . . . 89
5.10 Examples of differential delays for a one-layer electrical conductivity model for
the 1969 jerk and a typical distribution of observatories. . . . . . . . . . . . . . . 90
5.11 Misfit between the differential delays of the data and model considering the 1969
jerk spherical harmonic model in Chapter 3 and Le Huy et al. (1998) model for
the two-layer model with the mantle divided into two layers of 1250 km and 950
km. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.12 IRFs of a two-layered model simulating the lower mantle and the D with varying
electrical conductivities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
jerk spherical harmonic model of Chapter 3 for the two-layer model with the
mantle divided into two layers of 300 km and 1900 km. . . . . . . . . . . . . . . 92
5.14 Examples of differential delays for two electrical conductivity models for the
1969 jerk and a typical distribution of observatories. . . . . . . . . . . . . . . . . 95
5.15 Misfit for the two-layer model considering the 1969 jerk spherical harmonic
model in Chapter 3 and taken from Le Huy et al. (1998), for a higher D electrical
conductivity and finer variations. . . . . . . . . . . . . . . . . . . . . . . . . . . 96
jerk spherical harmonic model in Chapter 3 and the three-layer model. . . . . . . 97
5.17 Misfit between the differential delays of the data and model considering Le Huys
et al. (1998) spherical harmonic model of the 1969 jerk and the three-layer model. 98
5.18 Examples of differential delays for three electrical conductivity models for the
1969 jerk and a typical distribution of observatories. . . . . . . . . . . . . . . . . 99
xiv
List of Tables
2.1 Delay times and amplitudes of two chosen locations. . . . . . . . . . . . . . . . . 29

2.2 Number, code and location of 10 observatories used as example for the calculation
of the time delay using Backus low frequency approximation and the CIRF directly. 31
3.1 Summary of seven previous works about detection of geomagnetic jerks, speci-
fying the data, the methods used and which jerks were detected. . . . . . . . . . . 67
5.1 Setup of Velmsk and Martinecs (2005) code to calculate the IRFs. . . . . . . . 83
A.1 Occurrence date () of the late 1960s geomagnetic jerk in the annual means of
the X , Y and Z components of each magnetic observatory, detected by fitting two
straight-line segments to data in the least-squares sense. . . . . . . . . . . . . . . 104
A.3 Occurrence date () of the early 1990s geomagnetic jerk in the annual means of
A.5 Number of observatories included, excluded and not detected in the data analysis
of each geomagnetic jerk for the annual mean datasets. . . . . . . . . . . . . . . 117
B.1 Misfit values for the electrical conductivity models used for a one-layer model
simulation considering the 1969 jerk morphology model in Chapter 3 and from
Le Huy et al. (1998). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
B.2 Misfit values for the electrical conductivity models used for a two-layer model
simulation considering the 1969 jerk morphology model in Chapter 3. . . . . . . 120
simulation considering the 1969 jerk morphology model in Chapter 3 using a
thinner grid for the electrical conductivity. . . . . . . . . . . . . . . . . . . . . . 120
simulation considering the 1969 jerk morphology model of Le Huy et al. (1998)
using a thinner grid for the electrical conductivity. . . . . . . . . . . . . . . . . . 121
B.5 Misfit values for the three-layer model and using two different spherical harmonic
models of the 1969 jerk amplitude. . . . . . . . . . . . . . . . . . . . . . . . . . 122
xv
List of Symbols
a radius of the Earths surface
error bar
(r, 0 , 0 ) differential delays
(n )n nth central moment
22 smoothing time
data error bars in t0
c radius of the core
2m minimum value of 2
(t) Dirac delta
harmonic degree
0 dielectric constant
magnetic diffusivity
power exponent in Lahiri-Price model
(r) kernels
damping parameter
0 permeability of free space (4 107 H/m)
order of Bessels equation
longitude
total electric charge density
electrical conductivity
c electrical conductivity at the CMB
nn nth ordinary moment
1 delay time
1 (r, , ) differential delay times
colatitude
angular frequency
i ith basis function
Am (r, , ) jerk amplitude
B magnetic field
Bt toroidal magnetic field
Bp poloidal magnetic field
C arbitrary geomagnetic component
D Declination
d data vector
E electrical field
E misfit
Ere f reference misfit
E% variance reduction
xvi
e error bar (lower bound)

e+ error bar (upper bound)
em () complex coefficients (primary external potential)
||e|| norm
F () Transfer Function
F (t) Impulse Response Function

F(0) total jerk amplitude
F(r, , ,t) Composite Impulse Response Function
G geometrical attenuation
G design matrix or data kernel
gm and h
m Gauss coefficients for internal sources
H Horizontal component of the magnetic field
I Inclination
im () complex coefficients (secondary internal potential)
J electric current density
J Bessel function
K(r, 0 , 0 ) composite kernel
K(r, 0 , 0 ) differential kernels
L truncation degree
m harmonic order
m model vector
N number of independent data
N norm
Pm (cos Schmidt quasi-normalized associated Legendre functions
P(r, , ,t) poloidal scalar
pm (r,t) poloidal functions in the spherical harmonic decomposition
qm and s
m Gauss coefficients for external sources
R spatial spectrum of the magnetic field
R spatial spectrum of the jerk morphology
R resolution matrix
r radius
r radius vector from the geocentre
T(r, , ,t) toroidal scalar
Tr(R) trace of the resolution matrix
t time
t0 jerk occurrence time
V potential
V jerk amplitude in the potential function
v velocity
X Northward component of the magnetic field
X (a, ,) jerk amplitude in X
Y Eastward component of the magnetic field
Y (a, , ) jerk amplitude in Y
Ym (, ) Schmidt quasi-normalized spherical harmonics
Z vertical component of the magnetic field
Z(a,
,) jerk amplitude in Z
Dedicated to my parents Euclides and Marcia
1
Chapter 1
Introduction
1.1 Overview
The Earths mantle electrical conductivity reflects chemical and physical properties of the planets
interior, places constraints on core-mantle coupling and controls the transmission of geomagnetic
signals from the core to the surface. Despite many mineral physics studies, induction studies
and considerations of the transmission of the core magnetic field to the surface, the lower mantle
electrical conductivity is still mostly unknown and remains an open question. The aim of this
thesis is to obtain constraints on mantle electrical conductivity by using subtle signals that are often
globally seen in components of the Earths magnetic field. These signals take the form of rapid
variations in the rate-of-change of the field with time, and are termed geomagnetic jerks. Since
jerks are time variations of the magnetic field generated in the core, they will pass through the
electrical conducting mantle, before arriving at the surface. Consequently, the geomagnetic field
observed at the surface will correspond to a filtered version, delayed and smoothed, of the original
field generated in the core. This delayed and smoothed version of the input contains information
on the filter though which it has just passed, and the challenge is to decode this information in
terms of Earth properties.
1.2 Earth Structure

The main seismological and compositional divisions of the Earth are illustrated in Figure 1.1: the
crustal thickness varies from 20 km to 70 km in the continental crust, and 5 km to 8.5 km thick in
the ocean crust; beneath the crust lies the mantle up to 2890 km depth separated from the liquid
metallic outer core by a major change of seismic properties.
Solids transmit compressional waves (P-waves) and shear waves (S-waves), while liquids have
no shear strength and propagate only P-waves. The presence of only P-waves and absence of S-
waves in the outer core indicates its liquid nature, while the presence of P-waves and S-waves in
the inner core indicates that it is solid (from 5150 km depth). The zone approximately 300 km
thick between the lower mantle and top of the core is called D (Davies (1999)).
In this thesis we focus on the Earths mantle, especially on its electrical conductivity. In the
mantle, there are two main divisions in terms of seismic discontinuities (at 410 km and 660 km
depth), as shown in Figure 1.2. The region between the bottom of the crust and the first seismic
discontinuity at 410 km is called the upper mantle; and between the two seismic discontinuities is
2
Figure 1.1: Schematic cross section of the Earth considering seismological regions and pressures in function
of depth. Taken from Jeanloz (1990).
Figure 1.2: Profiles of seismic velocities and density with depth in the earth. Compressional velocity
(solid line), shear velocity (dashed line) and density (grey line). The right panel shows a zoom of the
mantle and the arrows indicate the discontinuities in seismic velocities in the P and S waves. From the model
ak135 (Kennett et al. (1995)), taken from Davies (1999).
referred to as the transition zone, and below 660 km it is called lower mantle. The first discontinu-
ity marks the phase transition of olivine ((Mg,Fe)2 SiO4 ) into a denser spinel polymorphs1 called
wadsleyite and at 520 km depth another transition to ringwoodite ((Mg,Fe)2 SiO4 ). The second dis-
continuity results from a breakdown of higher-pressure polymorph ringwoodite ((Mg,Fe)2 SiO4 ) to
perovskite ((Mg,Fe)SiO3 ) and magnesiowustite ((Mg,Fe)O). Silicate perovskite is the most abun-
dant mineral in the Earth (Yagi (2007)), since the volume of the lower mantle exceeds half of the
Earths volume.
Until 2004 it was believed that perovskite was stable in the lower mantle down to the CMB.
However, recent experimental studies simulating the conditions near the D layer (above 125 GPa
1 Spinel polymorph: any class of minerals of general formulation XY O which can crystallize in different forms
2 4
(such as the cubic crystal system) and happen due to changes of temperature and/or pressure.
3
and 2500 K) led to the discovery of a new phase change from MgSiO3 perovskite to post-perovskite
(ppv, Murakami et al. (2004)). The experimental evidence of post-perovskite was later confirmed
by ab initio calculations performed by computational mineral physics. This discovery is motivat-
ing studies in diverse fields of Earth sciences such as: seismology, geodynamics, geomagnetism,
mineral physics and geochemistry. The reason is that specific properties of post-perovskite are
related to chemical, compositional and thermal processes in the D layer (Hirose et al. (2007)).
Some open questions concerning the observed special properties of D , such as the strong
elastic anisotropy and the intermittent presence of ultra low-velocity zones, could be explained
with the discovery of the post-perovskite transition (Yagi (2007)). However, there are many other
open questions regarding D such as whether iron from the core enters into the post-perovskite
structure, causing an increase on mantle electrical conductivity. Mao et al. (2004) demonstrated
that the incorporation of Fe reduces the pressure for the phase transition and the major component
of D would be post-perovskite. Figure 1.3 schematically shows that hotter regions at the base of
D generate stronger interactions between D and the outer core. As a consequence, larger amounts
of Fe would be added into the ferromagnesian post-perovskite silicate which would increase the
electrical conductivity in those regions.
Figure 1.3: Schematic cross section of the D zone.

The shading in the squares (perovskite, pv) and striped
rectangles (post-perovskite, ppv) reflects the changing
Fe/Mg ratio, with an increasing red color indicating
higher Fe content in that phase (Mao et al. (2004)).
1.3 Geomagnetic Field Description

The Earths magnetic field is subdivided in internal field, generated in the liquid outer core and
crust; and external fields generated in the ionosphere and magnetosphere2 that is the region where
the solar wind interacts with the magnetic field of the Earth. The geomagnetic field varies spatially
and temporally from periods of milliseconds to million of years. The temporal fluctuations of the
external field ranges from milliseconds to a few decades including the slowest periods related to
the variations of the solar magnetic field. The longest time variations of the magnetic field, from
a few years to millions of years, are due to the changes in the internal field that are called secu-
lar variation, or paleomagnetic secular variation when the time variations are longer than several
hundred of years (Wardinsky (2007)).
The magnetic field is measured in observatories, at repeat stations, by satellites, and by oceano-
graphic, land and aircraft surveys. In the 15th century, the declination and inclination (Figure 1.4)
2 The ionosphere is the region from altitudes between 50 km and 1500 km and the magnetosphere from about 10-12
Earth radii.
4
were first measured by land observations, while in the 16th century the first observations of the
magnetic field in ocean areas were performed. After Gausss method for measuring absolute inten-
sities was invented in 1832, the horizontal and total intensities started to be observed. Nowadays,
the northward (X ), eastward (Y ) and downward (Z) components are most commonly measured
(Jackson et al. (2000)).
Figure 1.4: Magnetic field components, where X is the northward, Y the eastward and Z the vertical com-
ponent. Some simple geometric relations
between the magnetic field components are: X = H cos(D);
Y = H sin(D), tan(I) = HZ and F = X 2 +Y 2 + Z 2 , where D is the declination, I the inclination, H the
horizontal component and F the total field. Taken from Campbell (2003)
Magnetic observatories (Figure 1.5) provide the most continuous and accurate observations
of the magnetic field with measurements at least every minute starting in the early 19th century
(Macmillan (2007)). However, the uneven spatial distribution of observatories (Figure 1.5) is a
clear limitation for the global analysis of the magnetic field. Satellites measurements provide good
spatial data coverage and may contribute in a global analysis of the magnetic field. However, the
operation time of satellite missions was not regular during the last 43 years, when the first satellite
mission (POGO) carrying a scalar magnetometer was launched (from October 1965 to April 1971).
The first vector satellite mission (Magsat) only flew in November 1979 to May 1980 followed by a
gap of 20 years, when the rsted satellite was launched (February, 1999). CHAMP, rsted 2/SAC-
C satellites were launched in 2000 and, with rsted, are currently mapping the geomagnetic field.
1.3.1 Geomagnetic Field Sources

By spherical harmonic analysis one can separate the different sources of the magnetic field into
internal fields originated in the core and crust, and external fields generated in the ionosphere and
magnetosphere.
5
Figure 1.5: Distribution of magnetic observatories currently operating on the Earths surface (Macmillan
(2007)).
Laplaces Equation
Above the Earths surface and in regions where there is no current flow, the magnetic field is the
gradient of a potential:
B = V, (1.1)
that satisfies Laplaces equation:
2V = 0. (1.2)
In spherical coordinates (r is the radius of the sphere, is the colatitude and the longitude, see
Figure 1.6) Laplaces equation can be written as:

1 2 V 1 V 1 2V
r + sin + 2 = 0. (1.3)
r2 r r sin sin 2
Figure 1.6: Spherical coordinate system. Taken from Kreyzig (2001).

6
The solution of Laplaces equation in spherical coordinates (equation 1.3) is found by the
method of separation of variables and it is in the form:
h a +1 r i
V (r, , ) = a [gm m
cos m + h sin m] + [qm m
Pm (cos )
cos m + s sin m]
=1 m=0 r a
(1.4)
where P (cos ) are the Schmidt quasi-normalized associated Legendre functions, and (g , hm
m m
)
and (qm and s m ) are the Gauss coefficients for internal and external sources respectively. In equa-

tion 1.4 terms in (a/r) go to zero as r goes to infinity and describe fields from sources inside the
Earth, while terms in (r/a) go to zero as r tends to zero and describe sources outside the Earths
surface (r > a, where a is the Earths surface radius taken to be 6371.2km). Since we are interested
only in the internal sources (Langel (1987)):
a +1
V (r, ,) = a cos m + h sin m) P (cos ).
(gm m m
(1.5)
=1 m=0 r
The separation into core and crustal fields is usually based on the assumption that deep sources
will give much longer wavelength signals than the shallow crustal sources. The Lowes spatial
spectrum, defined as:
R = ( + 1) (gm 2 m 2
) + (h ) , (1.6)
m
for each spherical harmonic degree, expresses the contribution of the internal field to |B|2 . Figure
1.7 presents the spatial spectrum of the internal part of the magnetic field, using coefficients from
the model CM4 at epoch 2000. A significant change of slope of the spatial spectrum (Figure 1.7)
happens around harmonic degree 14, which would separate core and crustal fields. However, there
are uncertainties on this division, since the spectrum of the core field will continue after = 14 and
the crustal field will also continue below this value (Lowes (2007)).
Figure 1.7: Spatial spectrum of the internal geomag-

netic field at 2000. The circles are values of R (scale
on the right), while the squares are corresponding val-
ues for the secular variation (scale on the left) over the
Earths surface. Taken from Lowes (2007) using CM4
coefficients (Sabaka et al. (2004)).
From equation 1.5 one can calculate the magnetic field components (X , Y and Z) in spherical
coordinates, expressed in terms of gradients of the potential:
+2 Pm (cos )
1 V a
X= = (gm cos m + h m
sin m) (1.7)
r =1 m=0 r

7
1 V 1 a +2
Y =
r sin
= r
sin =1 sin m h cos m)P (cos )
m(gm m m
(1.8)
m=0
V a +2
Z= = ( + 1) cos m + h sin m)P (cos )
(gm m m
(1.9)
r =1 m=0 r
1.3.2 Geomagnetic Models

Time dependent models of the magnetic field, using different methodologies and datasets, have
been developed in the last few decades. Some of the field models were used to detect geomag-
netic jerks and are going to be discussed in Chapter 3. Most field models use spherical harmonic
expansion in terms of Gauss coefficients (gm m
and h ):
(t) = g i (t),
gm i m
(1.10)
i
for the internal field, and some recent models also do so for the external field. In equation 1.10
the i are a set of basis functions and i gm
a set of unknown coefficients. The choice of the basis
functions (i ) is what mostly differs between different field models (Jackson (2007)). The linear
inverse problem (d = Gm), using X , Y and Z components, can be described as finding the Gauss
coefficients (m) given the data (such as observations of X , Y and Z components, represented by
the vector d) and the design matrix (G), usually by the least-squares method. The most recent time
dependent models of the geomagnetic field, which are going to be mentioned in chapters 3 and 4,
are summarized below:
CM4 from Sabaka et al. (2004): from 1960 to 2002.5;

In addition to core secular variations, comprehensive models take into account effects of
crustal (from harmonic degree 14 to 65 in CM4) and external fields. The data set is a com-
pilation of satellite data (POGO, Magsat, rsted and CHAMP) and observatory data. They
also represent the secular variation in terms of B-splines with knots every 2.5 years.
Currents responsible for the ionospheric field are considered to flow in a thin spherical shell
at 110 km altitude and vary, with 24, 12, 8 and 6 hour periods, also annually and semiannu-
ally. The magnetospheric fields are also modelled considering daily periodicity and based on
the Dst index. The induced ionospheric and magnetospheric fields are calculated assuming
a four layer conductivity model obtained by Olsen (1998).
CHAOS model from Olsen et al. (2006): from March 1999 to December 2005;
Derived using data from three satellites: rsted, CHAMP and SAC-C. The temporal vari-
ations of the Gauss coefficients are also described by cubic splines with a 1 year knot sep-
aration. CHAOS model provides a representation of the internal magnetic field (core and
crustal) up to spherical harmonic degree L = 40, while for the secular variation L = 15.
Olsen et al. (2006) mention that the most important advantage of this model compared to the
previous ones, is its ability to map small scale structure of the secular variation.
8
1.4 Geomagnetic Jerks

In this work we are interested to analyse the main field generated in the core, consequently we used
monthly and annual mean data from magnetic observatories that average out some of the external
signal in the data. Figure 1.8 shows examples of 5 European observatories where it is possible to
observe some smooth changes in the trend of the magnetic field (in this case, the Y component).
However, in order to analyse these time variations we evaluate the secular variation (here defined
as first time differences) of the magnetic field that enhances the signal, but also the noise contained
in the data. The secular variation trend suffers abrupt changes called geomagnetic jerks. Some
smoothing can be applied in the secular variation in order to better identify geomagnetic jerks, as
shown in Figure 1.9.
Figure 1.8: Eastward component registered in five European magnetic observatories: Niemegk (NGK),
Furstenfeldbruck (FUR), both in Germany, Chambon la Foret (CLF) in France, Belsk (BEL) in Poland and
LAquila (AQU) in Italy.
Jerks have a typical V shape and may be approximated as straight-line segments, dividing
intervals of linear secular variation. In this simplistic model, jerks can be seen in the second
time differences as a step function and in the third differences as an impulse (Figure 1.10). Their
amplitudes are evaluated by the difference between the slope of consecutive straight line segments
of the secular variation. In this work we assume that jerks happen simultaneously at the CMB as
an impulse in the third time differences of the magnetic field.
The internal origin of jerks was demonstrated by Malin & Hodder (1982) using spherical har-
monic analysis, in contradiction with other works (Alldredge (1977), Alldredge (1984)) that be-
lieved jerks were caused by the external field. Other recent works have confirmed the internal
origin of jerks, for example Le Huy et al. (1998) who performed spherical harmonic models of the
1969, 1978 and 1991 events (X , Y and Z components). Jerks are usually detected in the secular
variation of the Y magnetic field component because of reduced influence of external signals. For
example, the predictions of external and internal fields by the CM4 model show that the intensity
of the primary and induced magnetospheric field are much larger in the X and Z components.
9
Figure 1.9: Secular variation of East component of Chambon la Fort observatory in France (yellow circles)
and Niemegk observatory in Germany (green circles). The data shown are first differences of observatory
monthly means that have been smoothed by a 12-month running average.
Figure 1.10: Simple schematic of secular variation for a hypothetical component of the magnetic field that
are used as input in the forward and inverse problems. The first graph (A) shows the typical V-shape of
jerks in the secular variation and B and C show sketched the secular acceleration and the third differences,
respectively. In this case, we would have an impulse in the third differences at the time when the jerk
occurred.
Since jerks are generated in the core, they will pass through the electrical conducting mantle,
before arriving at the surface. Consequently, the geomagnetic field observed at the surface will
correspond to a filtered version, delayed and smoothed, of the original field generated in the core.
The most relevant characteristic of geomagnetic jerks for constraining mantle electrical conductiv-
ity is their non-simultaneous behaviour at the Earths surface. This means that it would be possible
for the same event originating at the CMB to arrive at different times at the Earths surface. We
10
call differential time delays the difference between each jerk arrival time at each observatory and
a reference arrival time of the jerk is defined in this work as the mean time arrival considering
all observatories analysed. For instance, if the mean jerk occurrence time was at 1968, then the
differential delay of Eskdalemuir would be 1 year and of Gnangara 4 years (Figure 1.11).
Figure 1.11: Secular variation (annual means) of the East component (Y) of Eskdalemuir observatory in the
UK (left), wherea jerk happens around 1969, and Gnangara observatory in Australia (right) where a jerk is
clearly seen at around 1972.
The reason why jerks are not simultaneous at the Earths surface and why there are distinct
delays patterns for different jerks and magnetic components is still not well understood. There
are at least two simple hypotheses that could be envisaged as an explanation for this: the first is to
consider these differential time delays as being generated by dynamical processes in the core which
do not occur simultaneously; the second, which is the one we analyse in this work, is to consider
jerks generated instantaneously in the core and the time delays caused entirely by a conducting
mantle.
1.5 Mantle Electrical Conductivity

The study of physical properties in the deep Earth is fundamental for the understanding of geo-
dynamic and geochemical processes in the mantle. Studies involving electromagnetic methods,
mineral physics experiments and ab initio calculations have been used to estimate the mantle elec-
trical conductivity. In the last 5 years the efforts to obtain information about the lower mantle
layers have increased, especially with the recent discovery of post-perovskite in the D layer.
1.5.1 Mineral Physics Studies

Estimates of electrical conductivity using experiments at the lower mantle conditions were per-
formed by Peyronneau & Poirier (1989), where they showed that the conductivity is directly pro-
portional to the iron content, temperature and pressure. They found an electrical conductivity of the
order of 1 S/m at 1100 km depth and extrapolation of these experimental results led to a conductiv-
ity of 50-100 S/m at the CMB. Shankland et al. (1993) generated the conditions of the lower mantle
at pressures of 1.2 to 40 GPa and temperatures from 20 to 400o C and extrapolated the results to
lower mantle conditions adopting a functional form for the electrical conductivity that includes the
11
effect of pressure and temperature. They found 3-10 S/m as the maximum value of the electrical
conductivity in the lower mantle (Figure 1.12A). They highlighted that across the lower mantle the
conductivity increases by a factor of five and that lateral variations in the temperature (of about
200 K) probably do not produce resolvable departures from the radial conductivity, unless there is
also compositional heterogeneity.
Wood & Nell (1991) measured the electrical conductivity of (Mg,Fe)O at temperatures from
900 to 1500o C. They tested changes in oxygen partial pressure (pO2 ) and demonstrated that it
has a large influence on the electrical conductivity, about 10 times larger than the effect of an
increase of pressure by 10 GPa. Xu et al. (2000) considered also effects of oxygen fugacity and
the presence of other mineral components as pyroxene and magnesiowustite in their experiments.
They argue that the composition they use for magnesiowustite and the oxygen activity of their
measurements are close to the estimated lower mantle conditions. However, they ignore some
factors such as grain boundaries (which make a small contribution to the bulk conductivity at high
pressures) and possible effects of water and melt. The justification is that these are complications
which are not expected below a few hundred kilometers depth for a spherically symmetric Earth
(Figure 1.12B). Dobson & Brodholt (2000) highlighted that the presence of 4 % of Al2 O3 increases
the electrical conductivity by a factor of three. In order to represent possible limits of the lower
mantle conductivity, they used candidate lower mantle geotherms considering both aluminuous
and alumina-free compositions. They found that the lower mantle must be cool if lower mantle
alumina is incorporated into perovskite and it must be hot if alumina is not incorporated.
After the discovery of the post-perovskite phase transition in 2004, Mao et al. (2004) analysed
three ferromagnesian silicates, submitted to a pressure of the order of 130-165 GPa and a temper-
ature of 2500 K, and observed the transformation to the post-perovskite phase in all three samples.
They concluded that the incorporation of iron reduces the pressure needed for the transition of per-
ovskite to post-perovskite and that hotter regions have a stronger interaction with the outer core,
adding a larger amount of Fe in the post-perovskite and producing ultra-low-velocity zones. Mao
et al. (2007) posed the important question of whether iron enters in the structure of post-perovskite
and how it would affect its physical and chemical properties.
Ono et al. (2006) estimated the electrical conductivity of the lower mantle based on the analogy
of post-perovskite with Al2 O3 (Figure 1.12C). They used two models of the lower mantle geotherm
and found that when the temperature of the top of the outer core is lower than 4000 K, D is
composed of post-perovskite and when it is higher than 4000 K a thin layer of low electrical
conductivity may appear. Their estimates of electrical conductivity for the bottom of D reach
about 3000 S/m for the cool core and about 50 S/m for the hot core (Figure 1.12C). More recently,
Ohta et al. (2008) reported direct measurements of the electrical conductivity of (Mg, Fe)SiO3
post-perovskite at D conditions by using a laser-heated diamond-anvil cell. Their measurements
of electrical conductivity showed a minimal dependence on temperature and the conductance of
D was found to be 4 107 S, corresponding to a uniform conductivity of about 140 S/m (Figure
1.12D). They also stressed that the addition of Al2 O3 in the post-perovskite phase would increase
the the electrical conductivity of the D layer.
1.5.2 Induction Studies: external and internal magnetic fields

The electrical conductivity of the mantle can be studied in two ways by using the geomagnetic
field: from below by the time variations originating in the outer core and from the top using time
12
Figure 1.12: In A, the results taken from Shankland et al. (1993) that include E & B (Egbert & Booker
(1992), dotted line), T & W (Tarits and Wahr [1992], solid line), and the upper-bound curve ALMC
(Achache et al. (1981), dashed line). The hollow symbols represent extrapolated laboratory models. In B
the results presented in Xu et al.s [2003] paper: laboratory-based conductivity-depth profile compared with
geophysical models. Shaded areas illustrate the effect on the model of a 100o C temperature variation.
Geophysical models shown are AGLHS99 (Mandea Alexandrescu et al. (1999)), SKCJ93 (Schultz et al.
(1993)), BD (Bahr et al. (1993)), and Olsen99 (Olsen (1999)). In C the temperature profile in the lowermost
mantle shown in the plot on the left and on the electrical conductivity profiles derived from the perovskite
(right plot), CaIrO3 -type (post-perovskite) phase that is assumed to be two orders of magnitude higher than
that of perovskite, and magnesiowustite, taken from Ono et al. (2006). In D the results of Ohta et al. (2008)
electrical conductivity measurements of perovskite (open symbols) and post-perovskite (close symbols) as a
function of temperature. Previous data on perovskite by Katsura et al. (1998) are also presented by crosses.
changes of external origin. The potential is expanded in terms of spherical harmonics:

L r a +1
V =a m
() + m
() Pm (cos ) exp(im), (1.11)
=1 m= a r
where m
() and () are the complex expansion coefficients for the primary external and sec-
m
ondary internal parts of the potential, respectively. The Q-response is defined as the ratio between
13
the internal and external magnetic fields:
m
()
Q () = , (1.12)
m
()
and if several frequencies are known, it is possible to derive electrical conductivity models for
the mantle. Another alternative is to calculate the C-response that is the ratio between the vertical
magnetic field and its radial gradient (Banks (2007)):
Br
C= . (1.13)
Br /r
C-responses can also be defined in relation to Q-responses. By defining:
m
() = () + (),
m m
(1.14)
and
zm m m
() = () ( + 1) (), (1.15)
By equations 1.14 and 1.15, the magnetic field components can be represented at the Earths sur-
face by:
dPm (cos )
X (, , ) = ()
m
exp(m), (1.16)
,m d
m m
Y (, , ) = m
() P (cos ) exp(m), (1.17)
,m sin
and
Z(, , ) = zm
()P (cos ) exp(m).
m
(1.18)
,m
The C-response is usually the basis for estimating 1-D mantle electrical conductivity profiles
by induction studies and it is possible to be applied in global and local studies.
Information about electrical conductivities from 400 km to 1200 km requires time variations
with periods from 10 hours to about 42 days. In this period of time there are two main sources for
induction studies (Olsen (1998)):
(1) Electric currents in the ionosphere: the day side ionosphere is more conductive than the night
side and in each hemisphere there are currents circulating. Because of the Earths rotation, daily
variation in the magnetic field are created (Constable (2007)).
(2) Magnetic storms in the magnetosphere: the solar wind adds charged particles into the ring
currents, therefore increases in the solar wind generate what are called magnetic storms. The ring
current creates fields at the surface mainly in the P01 geometry and its strength is characterized by
the Dst index (Constable (2007)).
One of the earliest applications of induction to obtain information about the electrical con-
ductivity started with Lahiri and Prices [1939] work, who considered the general theory for any
radially-varying conductivity in a spherical geometry and obtained a formal solution. They applied
this theory to solar diurnal and storm-time variations to estimate the conductivities up to 1000 km
depth (Figure 1.13). In addition, geomagnetic time variations generated in the core have been used
in attempts to infer some information about the mantles electrical conductivity. Runcorn (1955)
14
discussed the theory of the diffusion of the magnetic field by considering the mantle as an infinite
sheet of uniform thickness and conductivity. He applied a step-like impulse at the bottom of the
slab and evaluated what the magnetic signal observed at the top would be.
McDonald (1957) used longer periods which characterize the secular variation and estimated
the lower mantle conductivity. He joined his estimates with the mantle model of Lahiri & Price
(1939), creating what is referred as the Price-McDonald model (see Figure 1.13). Banks (1969)
used fluctuations in the strength of the ring current in the magnetosphere, with a dominant P10
spherical harmonic geometry and calculated a 1D conductivity profile up to about 2000 km (Figure
1.13). Roberts (1986) reviewed methods for analysing long period data (few hours), the electro-
magnetic response functions and the effects of lateral heterogeneity.
Figure 1.13: Results of 1-D profiles of mantle electrical conductivity: Lahiri & Price (1939) developed
the general theory for any non-uniform conducting mantle and used solar diurnal variations and storm-time
variations; McDonald (1957) used longer periods that characterize the geomagnetic secular variation; Banks
(1969) used geomagnetic deep sounding techniques; Olsen (1998) estimated C-responses using variations of
the external field in Europe, from periods of 3 hours and 30 days and extended this study by analysing longer
periods from 1 month to 1 year (Olsen (1999)); Kuvshinov et al. (2005) derived 3-D forward solutions and
calculated the misfit between the observed and the modelled responses; all of the mentioned works used data
registered in magnetic observatories while Kuvshinov & Olsen (2006) analysed five years of satellite data
(from rsted, CHAMP and SAC-C) from 2001 to 2005 and estimated C-responses in the period between 14
hours and 4 months, and Velimsky (pers. comm.) also used satellite data.
Olsen (1998) used two different external field sources to evaluate C-responses in Europe: geo-
magnetic daily variations generated in the ionosphere and variations caused by the decay of mag-
netic storms in the magnetosphere (periods between 3 hours and 30 days). He concluded that there
is a monotonic increase of the electrical conductivity from 0.01 S/m at 200 km depth to about 1.4
S/m below 1000 km depth (Figure 1.13). Olsen (1999) used longer periods (from 30 days to 1
year) to estimate the electrical conductivity in deeper mantle layers. C-responses were calculated
15
using daily mean values at 42 magnetic observatories and these were used to infer an increase of
the electrical conductivity from 800 km depth ( = 2 S/m) to 2000 km ( = 3-10 S/m) shown
in Figure 1.13. Below this depth it was not possible to interpret the results in terms of mantle
conductivity due to lack of resolution (longer periods are needed).
The time variations of the magnetic field near ocean areas is strongly affected by the electrical
conductivity of the sea water. Olsen & Kuvshinov (2004) showed the importance of considering the
ocean effect by using 1D and 3D electrical conductivity models of the Earth and a temporal-spatial
structure for the storms. Olsen et al. (2006) developed a field model (CHAOS) based on satellite
data (CHAMP, rsted and SAC-C). Kuvshinov & Olsen (2006) used CHAOS model predictions
to subtract the core and crustal fields from 5 years of satellite data. They calculated C-responses in
the range of 14 hours to 4 months and 1D estimates of mantle electrical conductivity (from 400 km
depth) were found to be similar to studies using observatories (Figure 1.13). They compared their
results with others (Constable & Constable (2004), Olsen (1999), Kuvshinov et al. (2005)) and
found that all models monotonically increase from 0.03-0.08 S/m (at 400 km depth) to 1-2 S/m (at
900 km depth). However, their results for depths shallower than 400 km are more conducting than
in other works, this difference diminishing when they applied a correction for the oceans. Most
induction studies agree that electrical conductivity rises from 0.03-0.09 S/m at 400 km depth to
1-2.5 S/m at 900 km depth (Kuvshinov & Olsen (2006)), but in the lowermost mantle induction
study estimates are still very uncertain (see Figure 1.13) due to lack of resolution.
Studies involving geomagnetic jerks to analyse the mantle electrical conductivity were also
developed. Achache et al. (1980) performed the same calculation as Runcorn (1955), but they
use instead a ramp-function at the bottom of the slab (2000 km). They calculated responses for
different values of electrical conductivity and, based on the 1969 jerk, they estimated the average
conductivity of the slab as being 60 S/m which corresponds to a conductance of 1.2 108S. More
recently, Mandea Alexandrescu et al. (1999) assumed the 1969 jerk as a pure singularity at the
CMB and evaluated the effect caused by a uniformly conducting mantle. The estimated electrical
conductivity is less than 10 S/m on average assuming a uniform conducting lower mantle 2000
km thick and an insulating upper mantle. Nagao et al. (2003) also solved the diffusion equation in
the mantle for an abrupt change occurring simultaneously at the CMB. They believe that the later
arrival of jerks in South Africa and South Pacific Ocean may be explained by higher conductivities
beneath these regions. The electrical conductivity is assumed to be constant apart from anomalies
around South Africa and South Pacific. There is a contradiction between the estimated of conduc-
tivity in the region of South Africa from the delay times (> 200 S/m) and from the duration time of
jerks (< 40 S/m). However, the duration of the jerk (defined as the time for the change of the linear
trend of the secular variation) is very difficult to measure in the data due to the presence of noise.
In addition, if there are higher conductivities under South Africa and the South Pacific ocean, all
jerks would be late in these regions, a feature not seen in 1991.
1.6 Thesis objectives and structure

The aim of this work is to obtain constraints on the mantle electrical conductivity based on obser-
vations of geomagnetic jerks. Chapter 2 describes the theory of the screening effect of the mantle
on the magnetic field generated in the core. The data analysis of geomagnetic jerks is performed
in Chapter 3 using annual and monthly mean observatory data. The method applied to detect jerks
was of fitting two straight-line segments to the secular variation (X , Y and Z components), thus
16
allowing the determination of appropriate error bars. By using this method it is possible to identify
jerk occurrence times and their amplitudes used subsequently for building jerk spherical harmonic
models. The main questions we want to answer in Chapter 3 are:
(1) Which geomagnetic jerks can we detect in observatory data and what are their morphologies
and error bars?
(2) Are geomagnetic jerks distributed worldwide?
(3) Are jerk differential delay times the same in all events?
The forward problem is solved in Chapter 4, where we applied Backus mantle filter theory
which considers the mantle as a linear, causal and time-invariant filter. The input is assumed to be
an impulse in the third time derivative of the magnetic field that occurs simultaneously at the CMB;
the filter is represented by the electrical conducting mantle and the output the convolution between
the input and the filter, which would be the jerk we observe at the Earths surface. Some radial
models from Lahiri & Price (1939) are used as examples to illustrate the answers to the following
questions:
(4) What is the effect of the electrical conducting mantle on the input jerk signal assumed simulta-
neous at the CMB?
(5) Can a 1-D mantle electrical conductivity model cause jerk differential time delays at the Earths
surface?
(6) Is it possible that each jerk event (1969, 1978 and 1991 jerks) generates distinct time delays?
Chapter 5 involves solving the inverse problem to find what constraints jerk differential delay times
impose on the mantle electrical conductivity. The questions to be answered in this chapter are:
(7) How sensitive are the results in relation to the morphology of spherical harmonic model of the
jerk?
(8) What is the sensitivity of our model to the conductivity of the D ? How do the differential
delays vary when the D conductivity is changed?
(9) By considering each jerk separately, can we find possible 1-D mantle models of electrical con-
ductivity that explain their differential delays for X , Y and Z components?
(10) Is it possible to find candidate models for the lower mantle electrical conductivity that can
explain the differential jerk time delays of all jerks analysed together?
Finally, Chapter 6 contains the main conclusions obtained in this work and presents some ideas for
future work.
17
Chapter 2
Theory of mantle screening
The main magnetic field generated in the core has to pass through the conducting mantle before it
is observed at the Earths surface. The Earth may be modelled as a linear filter system where the
input would be a jerk at the CMB and the output the jerk observed at the surface. In this chapter, the
mathematical description of this problem, namely Backus mantle filter theory, and its limitations
are discussed.
2.1 Mathematical description

The diffusion equation will be the starting point to develop the whole theory of the effect of the
screening of the magnetic field by the mantle. In order to deduce the diffusion equation, one needs
to use the pre-Maxwells equations, in which the effects of displacement current is neglected; this
approximation is justified for the Earth (see Backus et al. (1996), chapter 2). The relevant equations
are
1
E = , (2.1)
0
which is Gausss law in differential form, and E is the electrical field, is the total electric charge
density and 0 is the dielectric constant. The divergence of the magnetic field (B) is zero
B = 0, (2.2)
which means there are no magnetic monopoles. Based on a series of experiments, Faraday reached
the conclusion that a changing magnetic field induces an electric field, given by
B
E = , (2.3)
t
which is Faradays law in differential form. Amperes law is given by
B = 0 J, (2.4)
where 0 is the permeability of free space (4 107 H/m) and J the electric current density.
Ohms law, for a material moving with velocity v in the presence of E and B is:
J = (E + v B), (2.5)
18
where is the electrical conductivity. Combining Amperes law (equation 2.4) and Ohms law
(equation 2.5) gives
B = 0 (E + v B).
If we define the magnetic diffusivity by
1
= , (2.6)
0
then we can write
B = E + v B. (2.7)
Equation 2.7 can be considered in three different situations: The first is when both the velocity
v and the conductivity are not zero, which will lead to the induction equation. In the second
case, we consider the magnetic field passing through a conductor ( 6= 0) which is static (v = 0).
That is the case of the mantle and it will be used to derive the diffusion equation. The last case,
considering an insulating medium ( = 0) and also v = 0 leads to Laplaces equation, the solutions
of which were presented in Chapter 1 (Section 1.3.1).
If we apply the curl to equation 2.7, we find
B = E + (v B), (2.8)
and when is a constant we can re-write equation 2.8 as
B = E + (v B). (2.9)
Now by Faradays law (equation 2.3) we have

B
( B) 2 B = + (v B) (2.10)
t
and since B = 0, we arrive at
B
= (v B) + 2 B, (2.11)
t
which is the induction equation. The diffusion equation is a special case of the induction equation
where v = 0:
B
= 2 B. (2.12)
t
However, in this thesis, the electrical conductivity is only assumed to vary as function of radius
= (r). For a non-uniform conductor, the diffusion equation is written as (see equation 2.8):
B
= ( B) , (2.13)
t
The appropriate way to solve the diffusion equation when the conductor has a spherical sym-
metry with a radial conductivity dependence is by using a toroidal-poloidal decomposition of the
magnetic field Gubbins & Roberts (1987):
B = Bt + B p , (2.14)
19
the vectors Bt and B p are called the toroidal and poloidal magnetic fields, respectively and can be
written in terms of poloidal and toroidal scalars (P(r, , ,t) and T(r, , ,t)):
B = (T(r, , ,t)r) + (P(r, , ,t)r), (2.15)

where r is the radius vector from the geocentre, is the colatitude and the longitude. Only the
relationship between the poloidal magnetic field at the CMB (r = c = 3485 km) and at the Earths
surface (r = a = 6371 km) is considered in this thesis. Toroidal magnetic fields are absent in the
region r > a because of the insulating atmosphere and in a 1-D mantle model, T is separated from
P. In short hand notation P = P(r, , ,t) and the magnetic field B in equation 2.15 can be written:

B = (Pr) = (rP) r2 P, (2.16)
r
and the curl of the magnetic field will be:
B = [(2 P)r]. (2.17)
Substituting equations 2.16 and 2.17 into 2.13 we obtain the diffusion equation for the poloidal
scalar:
P 1
= 2 P, (2.18)
t 0 (r)
which can be rewritten in terms of spherical harmonics, since the poloidal scalar P can be expanded
as:

P= (r,t)Y (, ),
pm m
(2.19)
=1 m=
where Y (, ) are the spherical harmonics, pm
m
(r,t) are functions in the spherical harmonic decom-
position, is the degree, m the order and Ym the Schmidt quasi-normalized spherical harmonics. If
the Laplacian is represented by:
1 2 L2 (.)
2 (.) = (r(.)) , (2.20)
r r2 r2
and the operator L2 applied to spherical harmonics is:
L2Ym (, ) = ( + 1)Ym (, ), (2.21)
we can rewrite equation 2.18 in the form:
2
pm (r,t) 1 1 m ( + 1) m
= (rp (r,t)) p (r,t) (2.22)
t 0 (r) r r2 r2
pm (a,t)
subject to [pm m
(c,t)] = 0 at the CMB and [p (a,t)] = [ r ] = 0 at the Earths surface, where [

] means jump. Equation 2.22 illustrates the important point that in the case of a 1-D conductivity
profile (r), the diffusion equation separates in spherical harmonic degree , which is a great
advantage used in this work.
In order to solve the diffusion equation in the frequency domain, one must define the Fourier
transform of the poloidal scalar:
Z
,, ) =
P(r, P(r, , ,t)et dt, (2.23)

20
and its inverse Fourier transform as:

1
Z
P(r, ,,t) = P(r, , , )et d. (2.24)
2
The Fourier equivalent of the poloidal scalar diffusion equation (2.18) is:
1
=
P 2 P,
(2.25)
0 (r)
or in terms of spherical harmonics (Fourier transform of equation 2.22):

1 1 2 ( + 1) m
pe (r, ) =
m
(r pe (r, ))
m
pe (r, ) . (2.26)
0 (r) r r2 r2
In order to illustrate the effect of an electrical conductor we adopt a 1D mantle model that increases
with depth as used by Lahiri and Price (1939):
c 2+2
(r) = c , (2.27)
r
where c is the electrical conductivity at the CMB, a positive constant and r = c is the core radius.
By defining:
1
= ( + 0.5) (2.28)

and c
1
zr = (0 c c2 )1/2 , (2.29)
r
and assuming the radial mantle conductivity model of equation 2.27, Gubbins & Roberts (1987)
1
show how equation 2.26 reduces to Bessels equation of order for f (r) = r 2 pem (r, ):

2 f (r) 1 f (r) 2
+ + 1 f (r) = 0. (2.30)
r2 r r r2
We suppose that is so large that the electrical conductivity falls rapidly to zero with increasing
r, being zero before r = a is reached. Since r = corresponds to z = 0 (equation 2.29), the Bessel
function should vanish at z = 0, consequently f (r) = J (zr ) , where J is the Bessel function of
order .
21 J (r) be the solution of equation 2.26, for any fixed harmonic degree
(r, ) = r
If we let pem
(), pem
(c, ) can be regarded as the input to the mantle which produces output pem (a, ) at the
Earths surface. In the frequency domain any input and output functions can be related by:
(a, ) = G F () p
pem em (c, ),
(2.31)
where G is the geometrical attenuation:
c +1
G = , (2.32)
a
and F () is the transfer function given by:

zc J (za )
F () = , (2.33)
za J (zc )
Note that the mantle filter can be described as a combination of geometrical and electromagnetic
filters, that depend on the electrical conductivity and on the spherical harmonic degree , but not
on the order m.
21
2.2 Backus mantle filter theory

Backus (1983) developed a theory that considers the mantle as a linear, causal and time invariant
filter. In what follows we adopt the separation by spherical harmonic degree as in equation 2.22.
Any output at the Earths surface (r = a, pm (a,t)) can be calculated by the convolution between the
input at the CMB (r = c, (pm (c,t))) and the electromagnetic (F (t)) and geometrical (G ) filters:
pm m
(a,t) = G F (t) p (c,t)
Z
= G F (t t )pm
(c,t )dt . (2.34)

The input jerk is assumed as a simultaneous second order impulse at the CMB, represented by
pm
(c,t) = (t), where (t) is a Dirac delta. The output can be written as
Z
pm
(a,t) = G F (t t )(t )dt

= G F (t), (2.35)
where F (t) is called the Impulse Response Function (IRF) that characterizes the mantle filter. The
Fourier transform of the IRF, Z

F () = F (t)et dt, (2.36)

is called the Transfer Function (TF) and for zero frequency = 0,
Z
F (0) = F (t)dt, (2.37)

which is the area under the

R
IRF curve for each harmonic degree . We use the same convention as
Backus (1983) in which F (t)dt = 1.
2.2.1 Backus low frequency approximation

Backus (1983) assumes a low frequency approximation, where higher powers than frequency 2
are neglected in the series expansion of the Transfer Function. This assumption simplifies Backus
[1983] mantle filter theory, since the IRF can be represented by a Gaussian, and it leads to a
linear relationship between the mantle electrical conductivity and the time an impulse takes to pass
through the conducting mantle.
We start from Backus[1983] equation 2.2:

1

F()
= F(0) n! (n)n , (2.38)
n=0
where nn are called the nth ordinary moments, given by

Z
1
(n )n = F(t)t n dt. (2.39)

F(0)
and (n )n the nth central moments:

Z
1
(n ) =n
F(t)(t 1 )n dt. (2.40)

F(0)
22
Since the TF can be expressed by the Fourier transform of the IRF (shown in equation 2.36),
we have:
Z

F() F(t)et dt

Z
1
= F(t) (t)n dt
n=0 n!
Z Z
2
Z
3
Z
= F(t)dt + F(t)tdt F(t)t 2 dt F(t)t 3 dt + ...
2 6
22 F(0)
2 3

= F(0)
+ F(0) 1 F(0) 33 + ..., (2.41)
2 6
and if we truncate this expansion at 2 , the TF is rewritten by
2

F()
= F(0) 1 + 1 22 . (2.42)
2
A simple expansion of equation 2.40 shows that the central moments can be expressed in terms of
ordinary moments; we find
22 = 22 21 , (2.43)
and substituting 22 from equation 2.43, in equation 2.42, we have

2

F()
F(0) 1 + 1 + 2
2 2
2 1
2
1 22 2
F(0)e . (2.44)
In order to calculate the IRF, we take the inverse Fourier transform of equation 2.44:
1
Z
F(t) = F()et d
2
Z 22 2
F(0)
= e(1 t) 2 d
2
s ( t)2

F(0) 2 122
= e 2
2 22
(1 t)2

F(0)
222
= e , (2.45)
2 2
F(t)
where F(0)
has the form of a Gaussian with mean 1 , given by the first ordinary moment (n = 1 in
equation 2.39) Z
1
1 = F(t)tdt (2.46)

F(0)
and called the delay time, which is the time an impulse at the CMB takes to pass through the
conducting mantle. In equation 2.45, 22 is the variance, called the smoothing time and given by
the second central moment (n = 2 in equation 2.40)
Z
1
22 = F(t)(t 1 )2 dt. (2.47)

F(0)
23
If one aims to calculate the delay and smoothing time constants for each harmonic degree,
equations 2.46 and 2.47 become: Z
1 () = F (t)tdt. (2.48)

and Z
(2 ()) =2
F (t)(t 1 ())2 dt, (2.49)

since F (0) = 1 (equation 2.37). In order to illustrate the effect of the electrical conducting mantle,
we used two radial conductivity models, given by equation 2.27, to calculate the IRFs and TFs
(Figure 2.1).
Figure 2.1: Radial models of mantle electrical conductivity (equation 2.27): the blue curve represents a
weakly conducting model (c = 100 and = 8) and the red curve a highly conducting model (c = 3000
and = 11).
The delay and smoothing time constants were evaluated, by equations 2.46 - 2.49, for the
two radial models in Figure 2.1. These examples, illustrated in Figure 2.2, show that higher con-
ductivity models and lower spherical harmonic degrees generate more delayed and usually more
smoothed outputs. It is possible to compare, in Figure 2.2, the delay time of the different models
and spherical harmonic degrees: for the most conducting model 1 (1) = 2.57 yr and for = 5 the
delay time is smaller (1 (5) = 1.99 yr), while for the less conducting model 1 (1) = 0.15 yr and
1 (5) = 0.11 yr.
We note that every spherical harmonic component of the poloidal scalar travels independently
through the mantle. We introduce our simple model for the jerk at the CMB, namely
pm
(c,t) = G p (a)(t).
1 m
(2.50)
and at the Earths surface the output is
pm
(a,t) = p (a)F (t).
m
(2.51)
24
Figure 2.2: Impulse Response Functions (A, B) and Transfer Functions (C, D) for the two radial electrical
conductivity models, shown in equation 2.27 and Figure 2.1. Plots A and C correspond to the higher con-
ducting model (c = 3000 and = 11) while plots B and D correspond to the weaker conducting model
(c = 100 and = 8).
The matching conditions between an insulator and a conductor give two representations at the
Earths surface:
B = V (2.52)
with
L
V =a Bm Ym (2.53)
=1 m=0
and for the conductor
B = Pr, (2.54)
where the relationship between the two is
pm

Bm
= . (2.55)
a
25
We now illustrate Backus phenomenon of mode mixing in different locations correspondent

to magnetic observatories, and in different components (X , Y and Z). Backus (1983) expresses the
magnetic components in terms of the poloidal scalar by a spherical harmonic expansion:

1 Ym (, ) m Y m (, )
X (r, ,,t) = p (r,t) =
(r, , ),
Bm (2.56)
a =1 m= =1 m=

1 Ym (, ) m Y m (, ) m
Y (r, , ,t) = csc p (r,t) = csc B (r, , ),
a =1 m= =1 m=
(2.57)
and
1
Z(r, , ,t) =
a =1 m=
[( + 1)Ym (, )] pm
(r,t) = ( + 1)Ym (, )Bm
(r, , ),
=1 m=
(2.58)
where Ym (, ) are the spherical harmonics. Alternatively to equation 2.50 one can imagine that
the impulse is in the radial component, given by
1
B r (c, , ,t) = ( + 1) pm (c)(t)Ym(, ).
c =1
(2.59)
m=
If the input in the poloidal field is a second order impulse and a given jerk morphology is
considered at a given location (0 is the colatitude and 0 the longitude) at the Earths surface
(r = a), we find for Br
L Z
Br (a, 0, 0 ,t) = A (r, 0, 0 )F(t t )(t )dt
=1
L
= A(r, 0, 0)F(t)
=1
= F(r, 0 , 0 ,t) (2.60)
where F(r, 0 , 0 ,t) is called the Composite Impulse Response Function (CIRF). We have for
Br (a, 0, 0 )
(a, 0 , 0 ) = ( + 1)Y B
Am m m
(2.61)
that simply represent the spherical harmonic components of the amplitude (or morphology) of the
jerk at the Earths surface. If one wants to calculate the delay time in different locations at the
surface, one may use equation 2.46 but for the CIRF instead of the IRF:
L Z
1
1 (r, 0, 0 ) =

F(0)
(r, 0 , 0 )
Am

F (t)tdt
=1 m=0
L
1
= Am
F(0) =1 m=0

(r, 0 , 0 ). (2.62)
26

Since F(0) is now the area under the CIRF curve, we can say that:
Z

F(0) = F(r, 0 , 0 ,t)dt

L Z
= (r, 0 , 0 )
Am

F (t)dt
=1 m=0
L
= Am (r, 0, 0) = A(r, 0, 0 ), (2.63)
=1 m=0
which is called the total amplitude. If we let:
(r, 0 , 0 )
Am
(r, 0 , 0 ) = , (2.64)
m=0 A(r, 0 , 0 )
then equation 2.62 becomes

L
1 (r, , ) = (r, 0, 0)1(). (2.65)
=1
If we use the same radial mantle conductivity model given in equation 2.27, but changing the
parameters to c = 1000 S/m and = 3, and taking the 1978 jerk spherical harmonic model of
Le Huy et al. (1998), it is possible to calculate the CIRF and the delay time (1 (a, 0 , 0 )) at a
specific location. For example, by convolving the CIRF at the location corresponding to Hermanus
observatory, with an assumed impulsive jerk happening at time t = 1969 at the CMB, the output
jerk that happens at around 1974, will result in a delayed and smoothed version of the original
signal (Figure 2.3).
Figure 2.3: Example of the delay and smoothing caused by a 1D mantle conductivity given by equation
2.27, with c = 1000 S/m and = 3 for the 1978 jerk spherical harmonic model given by Le Huy et al.
(1998). The black line would be the jerk for an insulating mantle and the red line for a conducting mantle.
However, it is not possible to measure the delay time in the secular variation data because we do
not know when the jerk was generated at the CMB. But one can measure what we call differential
27
delays in which we choose a reference location and subtract all the other jerk occurrence times
from the reference. Consequently, in equation 2.64 becomes
Am (r, , )
(r, re f , re f )
Am
(r, 0, 0 ) = 0 0
, (2.66)
m=0 A(r, 0 , 0 ) A(r, re f , re f )
and the differential delay times are calculated by:

L
1 (r, , ) = (r, 0, 0)1(). (2.67)
=1
2.2.2 Backus [1983] linear kernels

The most important advantage of Backus [1983] low frequency approximation is the resulting lin-
ear relationship between mantle electrical conductivity and delay time. For each harmonic degree,
Z a
1 () = (r)(r)dr. (2.68)
c
where (r) are called kernels of equation 2.68, given by

c 2+1
0 0
(r) = r 1 = (r) (2.69)
2 + 1 r 2 + 1
where c 2+1
(r) = 1 , (2.70)
r
leading to Backus equation 3.17a:
Z a c 2+1
0
1 () = r 1 (r)dr. (2.71)
2 + 1 c r
Some examples of kernels ( (r)) for different harmonic degrees () are shown in Figure 2.4.
Higher harmonic degrees are more sensitive to lower mantle layers, which shows it might be
possible to obtain some constraints on the electrical conductivity of the deep mantle by using 1 ().
The difference between kernels is bigger for small harmonic degrees: notice that for high degrees
such as = 11 and = 12 the kernels are approximately the same.
If one wants to find the kernels for different locations, one has to combine equations 2.65 and
2.68:
L Z a Z a
1 (r, 0 , 0 ) = (r, 0, 0) c
(r)(r)dr =
c
K(r, 0, 0 )(r)dr, (2.72)
=1
where we call,
L
K(r, 0 , 0 ) = (r, 0, 0 )(r), (2.73)
=1
the composite kernel calculated in a specific location (r, 0 , 0 ). Because in the data we measure
the differential delay times, we also need to calculate differential kernels, given by
L
K(r, 0 , 0 ) = (r, 0, 0)(r), (2.74)
=1
28
Figure 2.4: Kernels (equation 2.69, (r)) for the harmonic degrees: = 1, = 2, = 5, = 6, = 11 and
= 12.
and the differential delays are calculated by

Z a L
(r, 0, 0 ) = K(r, 0, 0)(r)dr,
c =1
(2.75)
2.3 Limitations of Backus mantle filter theory

If we follow the same expansion of the Transfer Function (equation 2.41) considering the trunca-
tion up to 2 , but calculating at a specific location, we find for F(r,
0 , 0 , )
Z
1
0 , 0 , ) =
F(r, F(r, 0, 0 ,t) (t)n dt
n=0 n!
L Z
1
= A(r, 0, 0)
F (t)
n!
(t)n dt
=1 n=0
L Z Z Z
1
= A (r, 0 , 0 ) F (t)dt + F (t)(t)dt 2
F (t)(t) dt
=1 2

L
2
= A (r, 0 , 0 ) 1 + 1 () 2 ()2
=1 2
L 2 ()2 2
= A(r, 0, 0)e1() 2 , (2.76)
=1
which is a sum of Gaussians scaled by the amplitudes of each harmonic degree (A (r, 0 , 0 )). The
problem is that the sum of Gaussians may not be another Gaussian, and in these cases the Transfer
0, 0 , )) would not be well represented. In order to illustrate that,
Function in one location (F(r,
we present an example considering hypothetical delay times with specific amplitudes (Table 2.1).
29
1 () in years A (a, 1 , 1 ) A (a, 2 , 2 ) (a, 1 , 1 ) (a, 2 , 2 )

1 3 4 4 0.1538 -0.6666
2 2 6 6 0.2308 -1.0000
3 1 16 -16 0.6154 2.6666
Table 2.1: Delay times and amplitudes simulated for two chosen locations, considering harmonic de-
gree truncation L = 3. The total amplitude (A(a, 1 , 1 )) of location 1 is 26 nT/yr2 and of location 2,
A(a, 2 , 2 ) = -6 nT/yr2 . This example is created in order to illustrate the limitations of Backus low
frequency approximation, but it might be possible that this combination of delays and amplitudes do not
correspond to a given (r) distribution.
The IRFs for each harmonic degree are represented by Gaussians with mean 1 () and vari-
ance 22 = 1 (Figure 2.5 A and C). We calculated the delay times by using equation 2.65, and we
conclude that a Gaussian is a good representation for location 1, where 1 (a, 1 , 1 ) = 1.54 yr and
22 (a, 1 , 1 ) = 1.56 yr2 .
Figure 2.5: Examples of CIRFs calculated for the hypothetical locations in Table 2.1. A and B are the CIRF
for each spherical harmonic degree of location 1 and 2, respectively and C and D are the CIRF resulting
from the sum of CIRF of location 1 and 2, respectively. The dotted line in C is the Gaussian representation
of the CIRF (solid line).
However, in location 2 it is not possible to represent this fictitious oscillatory CIRF by a Gaus-
sian (see Figure 2.5 B and D) since it would mean that the filter is acausal (1 (a, 2 , 2 ) = 1.33
yr) and the Gaussian representation does not exist for negative variance (22 (a, 2, 2 ) = 8.11
yr2 ).
In order to give a more realistic example of amplitudes we consider the 1978 jerk morphology
of the Y component, given by Le Huy et al. (1998) with truncation degree L = 4 and the mantle
radial conducting model given by equation 2.27 for c = 1000 S/m and = 3. We choose ten loca-
30
tions corresponding to magnetic observatories, described in Table 2.2. For each of these locations,
we calculated the CIRFs and their approximation by Gaussians, as shown in Figure 2.6.
Figure 2.6: Examples of CIRFs (solid line) calculated for the 1978 spherical harmonic model of Le Huy
et al. (1998) of ten magnetic observatories (Table 2.2). A radial electrical conductivity model (equation 2.27
for c = 1000 S/m and = 3) was adopted.
Figure 2.7: Integrands of the smoothing time (F(t)(t 1 )2 ) given in equation 2.47 of ten observatories
(Table 2.2). A radial electrical conductivity model (equation 2.27 for c = 1000 S/m and = 3) was adopted.
31
OBS no OBS code latitude longitude CIRF 1 (yr) Gaussian 1 (yr)

1 MEA 54.616 246.653 1980.63
2 NGK 52.07 12.68 1975.32 1976.29
3 VIC 48.52 236.58 1972.77
4 NEW 48.27 242.88 1971.72
5 MMB 43.91 144.189 1975.29 1976.37
6 SJG 18.11 293.85 1974.45 1975.45
7 GNA -31.8 116 1975.43 1976.35
8 HER -34.425 19.225 1974.30 1974.73
9 CZT -46.431 51.86 1978.59 1980.81
10 MCQ -54.5 158.95 1974.31 1974.73
Table 2.2: Number, code and location of 10 observatories used as example for the calculation of the time
delay using Backus low frequency approximation and the CIRF directly. We used the 1978 jerk spherical
harmonic model of Le Huy et al. (1998). We calculated the convolution between the input jerk assumed
to happen at time t = 1969, with the CIRFs and their approximations by Gaussians. The delay time was
detected by using a least-squares fitting of two straight-line segments of the annually sampled convolved
signal (output).
Figure 2.8: Convolution of a jerk with A = 1 (black line) with the CIRF (red line) and with the Gaussian
approximation (blue line) of the ten observatories used as example (Table 2.2). The same radial model as in
Figures 2.6 and 2.7 (equation 2.27 for c = 1000 S/m and = 3) was adopted.
In some locations, such as Victoria (VIC, Canada, observatory no 3) and Newport (NEW, US,
Observatory no 4), the delay and smoothing times are negative and the approximation of a CIRF
by a Gaussian will not be valid. Even when the CIRF is slightly oscillatory, as in NEW observa-
tory, it will affect strongly the resulting delay (equation 2.46) and smoothing (equation 2.47) time
32
Figure 2.9: Fit of two straight-lines segments, by least-squares, to the convolved jerk with CIRF of observa-
tory 6 (a) and 9 (c) and for the Gaussian for observatory 6 (b) and 9 (d).
constants because their integrands, shown in Figure 2.7, tend to enhance the oscillatory behaviour,
especially for larger times (t).
Even if the CIRF is oscillatory it is possible to calculate the output by convolution. In order to
evaluate how Backus low frequency approximation would affect our measurements of the delay
time, we calculated the convolution of the input jerk with the PDF and with the CIRF (Figure 2.8).
For the same ten locations corresponding to magnetic observatories (in Table 2.2) the resulting
outputs look similar, with exception of observatory 9 (CZT). But if one evaluates annual means
for both results and by fitting two straight-line segments, in the same way as the data analysis
of Chapter 3, the resulting jerk occurrence times are significantly different (from 0.42 to 2.22
year, see Table 2.2 and Figure 2.9). This important result shows that in some cases the Gaussian
approximation will not be appropriate.
The composite and differential kernels (equations 2.73 and 2.74, respectively) were calculated
for the two examples in Tables 2.1 and 2.2. In the first hypothetical example, we show in Figure
2.10 how the composite kernels are calculated through the sum of composite kernels for each
harmonic degree. In addition, the second location with an oscillatory CIRF has more sensitivity to
the deep layers of the mantle than in the well behaved CIRF of location 1 (Figure 2.10).
This shows that some oscillatory CIRFs, which can not be represented by Gaussians, may
give more useful information than constant sign CIRFs, therefore should not be ignored. In the
second example (Figure 2.11), we chose as the reference kernel the one in which the jerk arrived
earliest (observatory 1, MEA in Table 2.2). In order to calculate differential kernels (equation
2.74) we subtract all other composite kernels from the reference (observatory 1, MEA). Since
the reference kernel is greater than all the others, the differential kernels are all totally negative.
That shows another limitation of Backuss filter theory to solve the inverse problem: since the
electrical conductivity and differential jerk delays (in this example) are positive, at least part of the
33
differential kernels should be positive.
Figure 2.10: Example of calculation of composite kernels: in A the kernels for each harmonic degree and
for locations 1 and 2 (Table 2.1) and in B the resulting composite kernels for both locations.
Figure 2.11: Composite (A) and differential (B) kernels evaluated for the ten observatories used as example
(Table 2.2) and considering a radial mantle conductivity model as shown in equation 2.27.
2.4 Summary
In this chapter we have illustrated Backus [1983] mantle filter theory by using hypothetical and
realistic examples. We showed that higher electrical conductivity models and lower harmonic
degrees generate more delayed jerks. Backus developed a low frequency approximation of the
Transfer Function that makes it possible to represent the Impulse Response Functions by Gaus-
sians. The most advantageous consequence of Backus low frequency approximation is the linear
relationship between the delay time (1 ()) and the radial electrical conductivity ((r)), in the
spectral domain. However, there are some limitations on this approximation when one wants to
represent an IRF at a given location (called CIRF). The delay and smoothing time constants may
be negative due to a mixing of harmonics and if that is the case, the Gaussian approximation to the
CIRF and the linear relationship between 1 () and (r) are no longer valid.
34
Because the time when the jerk was generated at the CMB is unknown, we measure differential
delays (1 ) between observatories instead of delay times (1 ). That is the reason why we calcu-
lated composite and differential kernels, and for the examples presented in Tables 2.1 and 2.2 we
concluded that: it is possible that oscillatory CIRFs give useful information about lower mantle
layers, therefore should not be ignored; we found totally negative differential kernels which should
not happen since in those examples all differential delays and electrical conductivity are positive.
Because of these limitations on Backus mantle filter theory we use the results of the data analysis
of geomagnetic jerks (Chapter 3) to solve the inverse problem (Chapter 5), but abandoning Backus
linear approximation and instead using numerical calculations of exact IRF using Velmsk and
Martinecs [2005] code.
35
Chapter 3
Data Analysis of Geomagnetic Jerks
3.1 Overview
Geomagnetic jerks are abrupt variations in the rate of change of the magnetic field, termed the
secular variation (SV), that have a typical V shape. They can be approximated as straight-line
segments, dividing intervals of linear secular variation. The most debated aspects of geomagnetic
jerks are their morphologies (the spatial distribution of amplitude), the allied question of local or
global visibility, and whether the same jerk is observed simultaneously or not at the Earths surface.
The non-simultaneity of jerks has been used in attempts to obtain some information about mantle
electrical conductivity (e.g. Mandea Alexandrescu et al. (1999), Nagao et al. (2003) and Pinheiro
& Jackson (2008)). These studies depend critically on accurate measurements of the time when
jerks occur at magnetic observatories.
In this chapter, we assume that the SV around the time of geomagnetic jerks can be modelled
by a set of straight line segments and that the point where the lines intersect defines the occurrence
time of the geomagnetic jerk. Our aim is to find the best fit of the linear segments to the data by
L2 (least-squares) and L1 (1-norm) methods and to determine their slopes. The difference between
the slope of consecutive straight lines defines the jerk amplitude at each location; these are used
to build spherical harmonic models of each jerk in each field component. In sections 3.2 and
3.3 we describe the methods and their limitations to detect geomagnetic jerks, respectively. The
datasets and pre-processing of the data are presented in section 3.4 and the methodology used for
the modelling, calculating the error bars and for the spherical harmonic analysis of jerk amplitudes
are reported in section 3.5. The results and discussion of the best known global jerks that have
occurred in 1969, 1978, 1991 and possibly in 1999 are described in sections 3.6 and 3.7.
3.2 Methods to Detect Geomagnetic Jerks

Many methods to detect jerk time occurrence have been explored in the past 30 years, the most
common being the fitting of two straight line segments to the SV by the least-squares method. Jerk
amplitudes are evaluated by the difference between the slope of consecutive straight line segments.
Le Moul et al. (1982) analyzed more than 130 observatories by fitting straight lines to the SV
before and after 1970. They found that the 1970 jerk is worldwide and its amplitude varies with lo-
cation but in some observatories, such as Honolulu and Coimbra, this jerk cannot be seen because
its amplitude is close to zero. The main disadvantage of the method applied by Le Moul et al.
36
(1982) is the assumption that the jerk occurs at the same time at all observatories. Gubbins & Tom-
linson (1986) demonstrated that the 1970 jerk was not simultaneous at Earths surface; they found
a time lag of about 2 years between Apia (Samoa) and Amberley (New Zealand) observatories.
Whaler (1987) explored an alternative approach in which all possible straight line segments are
fitted to the SV to detect non-simultaneous worldwide jerks. However, she assumed that the jerk
in the X , Y and Z components happened at the same time at the Earths surface. Stewart & Whaler
(1995) developed an algorithm involving optimal piecewise regression analysis applied to the main
field data directly. They concluded that geomagnetic annual means can be well represented by
piecewise quadratic models, but their method permits overlaps and discontinuities in the model.
Our modelling approach follows these studies in assuming that the jerk can be represented as a V
shape in the secular variation, a step function in the secular acceleration and a spike in the third
derivative.
Wavelet analysis of geomagnetic jerks was first globally applied by Alexandrescu et al. (1996);
this method involves only the assumption that sudden events of an unknown nature at undefined
dates may have occurred in the geomagnetic field. The sensitivity of wavelets to local character-
istics of a signal is an important advantage of this method (Alexandrescu et al. (1995)). Jerks are
assumed to be singularities introduced at the core-mantle boundary (CMB), the singularity being
defined as a discontinuity in an derivative of the signal, where is its regularity. Alexandrescu
et al. (1996) considered a linear combination of X and Y components of observatory monthly
means and detected seven different events: the 1901, 1913 and 1925 jerks being possibly global
in extent, the 1932 and 1949 jerks were observed only in the Pacific and American areas, and the
1969 and 1978 jerks found to be worldwide.
De Michelis & Tozzi (2005) also applied wavelet analysis to detect jerks, by using monthly
means from 44 observatories (only 6 being in the Southern hemisphere) and detected global jerks
in 1978, 1991 and 1999 and a local jerk in 1986. They used an alternative approach called local
intermittency measure (LIM), which has the advantage that it only selects discontinuities that are
not related to noise or external contributions, while in the classical wavelet method, an analysis of
ridge functions for each discontinuity is needed, even if the discontinuity is caused by a source un-
related to the core field. A limitation of wavelet analysis is that the boundary effects are important,
preventing detection of events close to the beginning and end of the time series. If the jerk is in one
of the data limits it is necessary to apply more classical methods, as was carried out by Mandea
et al. (2000) for the 1999 event.
The internal origin of geomagnetic jerks was demonstrated by Malin & Hodder (1982) us-
ing spherical harmonic analysis, in contradiction with other works (Alldredge (1977), Alldredge
(1984)) that believed jerks were caused by the external field. Other recent works have confirmed
the internal origin of jerks, for example Le Huy et al. (1998) who performed spherical harmonic
models of the 1969, 1978 and 1991 events (X, Y and Z components). In a different approach,
Nagao et al. (2002) applied a statistical time series model to geomagnetic monthly means data, ad-
justing seasonal and short time scale time variations. They tested the possibility of the ring current
as the origin of jerks and concluded that jerks cannot be explained by changes in latitudinally-
dependent external current intensities, thus ruling out an external origin of jerks. Bloxham et al.
(2002) attempt to demonstrate that geomagnetic jerks may be represented as torsional oscillations
modelled as a sum of waves of varying period and damping/growth timescale.
Le Huy et al. (1998) used smoothed annual means from 160 observatories and performed a
least-squares fit of two straight-line segments to the secular variation and a subsequent spherical
harmonic analysis of the 1969, 1978 and 1991 jerk amplitudes (L = 4, L being the maximum
37
harmonic degree). The 1991 jerk was also analyzed by De Michelis et al. (1998) by fitting two
straight-line segments to the secular variation of the Y component from 74 magnetic observato-
ries, with effects of the external field removed. They proved that the 1991 jerk is worldwide and
produced spherical harmonic models of its amplitude. Later, De Michelis et al. (2000) performed
the same analysis but using the X , Y and Z components of the 1991 jerk calculating the mean jerk
occurrence time as 1990.1 0.6. They employed spherical harmonic analysis to study the trend
of the energy density spatial spectrum for each harmonic degree.
3.3 Limitations on the Analysis of Geomagnetic Jerks

The problem of the poor geographical distribution of observatories can be improved by the use of
satellite data; such data has recently been used in addition to observatory data to build global field
models in which geomagnetic jerks are detected. Sabaka et al. (2004) developed a comprehensive
approach model (CM4) in the time interval from 1960 to mid-2002 that includes POGO (1967
to 1971), Magsat (1979 to 1980), Oersted, which is in operation from 1999 to present time, and
CHAMP that was launched in 2000 and working in present time. They were able to detect the
well known jerks and discussed the possibility of an additional event around 1997. Chambodut
& Mandea (2005) used the CM4 comprehensive model to separate internal from external sources
and performed a global search for geomagnetic jerks. They fitted two straight-line segments to
synthetic values of the secular variation obtained from CM4 and confirmed the non-simultaneous
behavior of the 1969, 1978 and 1991 jerks.
Wardinski & Holme (2006) developed another time-dependent model (C3 FM) of the core mag-
netic field from 1980 to 2000 by inverting a selection of observatory monthly means, annual means
and repeat station measurements. They found that the 1991 jerk was best seen in the Y component,
but they did not find the proposed 1999 jerk probably because it was too close to the end of the
interval covered by their model. Olsen & Mandea (2007) used satellite monthly means obtained
from virtual observatories at 400 km altitude from the CHAMP satellite and detected a new jerk
in 2003 observed simultaneously in a restricted area near 90o E with maximum jerk amplitudes
of about 25 nT/yr2 (in the Z component) at 30o latitude. Olsen & Mandea (2008) inferred the
existence of another localised jerk at 2005 centred on Southern Africa, using 9 years of satellite
data and annual differences of magnetic observatory monthly means.
The main limitation of previous studies concerned with the detection and characterization of
jerks is that they do not include jerk time and amplitude error analysis. This is essential because of
the difficulty of dealing with non core field signals when using geomagnetic data. Ducruix et al.
(1980) presented details of external field contamination in the 1969 jerk; they fitted two parabolas
in the annual mean data (from 1947 to 1977) and the residuals were found to be associated with
the solar cycle especially in the X and Z components. Gubbins (1984) also found evidence for
external signals in the residuals of a double parabola fitting (centred at 1969.5) to 106 observatory
annual mean data. Verbanac et al. (2006) recently analyzed the influence of the external field
on the X component and confirmed that it affects the secular variation estimates derived from
observatory annual means. They evaluated differences of 46 European observatory data and the
core contribution predicted by the CM4 model (Sabaka et al. (2004)) and compared the residuals
with the external fields predicted by the POMME model (Maus et al. (2005)). As the POMME
model uses the Dst index as a controlling parameter, Verbanac et al. (2006) believe that the good
coherency they found between the residuals and POMME predictions means that the dominant part
38
of the external signal in observatory annual means is caused by the ring current. Because of the
geometry of the ring current, the external disturbing field affects mainly the X ; that is the reason
why jerks are most commonly analyzed in the Y component (see, for example McLeod (1994)).
Besides the external field contamination, there are also errors due to instrument drift, malfunc-
tion, calibration, man-made magnetic signals, and possible errors in the data processing (McLeod
(1994)). Pais & Miranda (1995) give an example of data errors identified in records from Coim-
bra observatory; diverse problems, such as modification of the base line because of change in the
instruments, introduction of new buildings and new tram lines near the observatory are reported.
As we are interested in the data of all possible observatories during their operating time, it is clear
that problems such as these will be common in our analysis. De Michelis et al. (2000) performed
a visual inspection in 119 observatories annual mean datasets, and discarded five observatories
(ABG, FUQ, ALE, SYO, CWE) due to base line control problems plus observatories in which
only absolute values were reported for the 1991 jerk (QUE, MLT and MIR). In this work we take
into consideration other studies pre-selection of data to analyse magnetic jerks, since we do not
perform a detailed analysis in this matter. Concerning the monthly means we used data reported
by Chulliat & Telali (2007) in which a pre-selection was carefully performed.
3.4 Data
We analyzed three different datasets in order to study geomagnetic jerks in the X , Y and Z compo-
nents of the geomagnetic field: observatory annual means; 12-month running averages of observa-
tory monthly means rotated to geomagnetic dipole coordinates and annual means of the core field
contribution calculated from the CM4 field model. Jerks were first studied using annual means
but rapid events such as the 2003 and 2005 jerks have recently been studied using the higher tem-
poral resolution provided by the secular variation estimates calculated from monthly means. The
advantage of using a comprehensive model (CM4) is that it is able to separate different magnetic
field sources: core, crustal, ionospheric, magnetospheric, induced and toroidal field produced by
electrical currents at satellite altitudes (Sabaka et al. (2002)).
In this paper, the secular variation is evaluated as the annual differences of main field data
(e.g. dY 1 1
dt |t0 = Y (t0 + 2 ) Y (t0 2 )). Examples of observatory annual mean secular variation to-
gether with the CM4 prediction of the core field and monthly mean secular variation rotated to
geomagnetic dipole coordinates are shown for the X , Y and Z components in Figures 3.1, 3.2 and
3.3, respectively. Notice that using observatory 12-month running average of first differences of
monthly means gives the advantage of a higher temporal resolution but these data also present a
considerable amount of scatter that makes the identification of jerks more difficult than in the first
difference of annual means. Furthermore, comparing the X , Y and Z components (Figures 3.1,
3.2 and 3.3) it is clear that jerks are not simultaneous in the three components and that they are
often more visible in the Y component, probably because it is less contaminated by external field
variations.
Annual mean data are provided by the World Data Centre (WDC) for Geomagnetism at the
British Geological Survey (BGS, Edinburgh), while monthly means for all INTERMAGNET ob-
servatories were obtained by the World Monthly Means Database Projects reported by Chulliat
& Telali (2007). The later were derived from WDC hourly means with three consistency tests
performed: visual inspection, comparison of monthly means with annual means, and compar-
ison with results of the CM4 model. We excluded data when the monthly mean was calcu-
39
lated with less than 15 days. Some of the error reported in monthly mean data, such as large
spikes not related with strong activity and jumps are described in htt p : //www app3.g f z
potsdam.de/COM/comparison_obs.html .
Figure 3.1: Secular variation estimates for the X component of the magnetic field using different datasets.
Shown in A to D are first differences of annual means (red triangles) and the core contribution secular vari-
ation evaluated by the CM4 model (black line) and in E to H the annual difference of 12-month running
average of monthly means in geographical coordinates (red triangles) at the following magnetic observa-
tories: Gnangara in Australia (A,E), Kakioka in Japan (B,F), Meanook in Canada (C,G) and Niemenk in
Germany (D,H).
In order to decrease the influence of magnetospheric fields in the X and Y components, monthly
means data were rotated to geomagnetic dipole coordinates. This involves aligning the Northward
and Eastward components with the geomagnetic dipole axis (e.g. Olsen & Mandea (2007)). The
40
Figure 3.2: Secular variation estimates for the Y component of the magnetic field using different datasets.
Shown in A to D are first differences of the annual means (red triangles) and the core contribution secular
variation from the CM4 model (black line) and in E to H the annual difference of 12-month running aver-
age of monthly means in geographical coordinates (red triangles) at the following magnetic observatories:
Gnangara in Australia (A,E), Kakioka in Japan (B,F), Meanook in Canada (C,G) and Niemenk in Germany
(D,H).
dipole field direction for the 1969 jerk is taken to be the International Geomagnetic Reference Field
(IGRF) model at 1970 and the position of the dipole axis is given by a colatitude dip = 11.41o and
longitude dip = 289.82o E. In some observatories such as Niemengk (Germany) this rotation of
coordinates substantially decreases the amount of scatter present in the data, while at others such
as Gnangara the result does not differ greatly from the original data (Figure 3.4, E-H). We do not
apply this rotation to annual means since in most cases it does not greatly reduce the noise (Figure
41
Figure 3.3: Secular variation estimates for the Z component of the magnetic field using different datasets.
Shown in A to D are first differences of the annual means (red triangles) and the core contribution secular
variation evaluated by the CM4 model (black line) and in E to H the annual difference of 12-month running
average of monthly means in geographical coordinates (red triangles) at the following magnetic observa-
tories: Gnangara in Australia (A,E), Kakioka in Japan (B,F), Meanook in Canada (C,G) and Niemenk in
Germany (D,H).
3.4, A-D).
The 1969 jerk was chosen to test the application of different analysis methods (L2 and L1 ) to
the three datasets. The 1978, 1991 and 1999 jerks are only analysed by the least-squares method
applied to the observatory annual mean secular variation (X , Y and Z components). The reasons
for this choice are discussed in section 3.6.
The analysis of each geomagnetic jerk is done separately: a complete set of data is considered
to be 15 years of data, with 7 years before and 7 years after the supposed jerk. Most European
42
Figure 3.4: Observatory annual differences in geographical (red triangles) and geomagnetic dipole (black
triangles) coordinates. First differences of annual mean values are shown in A-D and annual differences of
12-month running average of monthly means, from E-H. Two observatories are used as examples: Gnangara
for the X component shown in A and E and for the Y component in B and F; Niemengk observatory being
the X component in C and G and the Y component in D and H.
observatories have complete annual mean dataset available for the 1969-1970 jerk, that is data
from 1962 to 1976. However, we also include observatories with incomplete datasets, by which
we mean a minimum of 11 years of data of which 5 years is centred at the date of the jerk (from
1968 to 1972 for the 1969-1970 jerk).
43
3.5 Methodology
3.5.1 Jerk occurrence time and amplitude detection
Geomagnetic jerks are modelled here as two straight-line segments fitted to the secular variation
estimates. Denoting with C an arbitrary geomagnetic component, our model is defined as:
= a1 (t t0 ) + b,
C(t) (3.1)
for t t0 and
= a2 (t t0 ) + b,
C(t) (3.2)
for t t0. In this model the intersection of the two straight-lines defines the jerk occurrence time
and the jerk amplitude is given by
A = a2 a1 . (3.3)
The advantage of using this simple model is that we were able to calculate the required error bars
in the time occurrence and amplitude of geomagnetic jerks.
In order to find the model that best fits the data, we explore different norms indicated by ||e||,
the norm of the error (or residual) vector e, where:
e = d Gm, (3.4)
where d and m are the data and model vectors respectively, and G the data kernel. We calculated
the model parameters (m = {a1 , a2 , b}) for all t0 at intervals of 0.001 yr.
We used L1 and L2 norms, ||e||1 = i |ei | and ||e||2 = i |ei |2 , respectively. The choice of
a norm implies an assumption that data errors obey a particular type of statistics: the L2 norm
(or least-squares method) assumes a Gaussian distribution of errors while the L1 norm assumes a
Laplacian distribution. Long-tailed distributions, such as the Laplacian, imply many large residu-
als while short-tailed distributions, such as the Gaussian, tolerate less the presence of large outliers
(e.g. Menke (1989)). In geomagnetism the least-squares method is usually used and data errors
are usually assumed to be Gaussian. However, situations in which the distribution of the errors
is known a priori are rare in geophysical problems and the assumption of Gaussian errors in ge-
omagnetism is often motivated on grounds of simplicity. Walker & Jackson (2000) showed that
in certain geomagnetic contexts a Laplacian error distribution may be more appropriate than the
Gaussian. For this reason, both L1 and L2 measures of datafit are considered in this work.
The preferred model for t0 , for each geomagnetic jerk at each observatory and for each field
component, is chosen according to the minimum of the misfit curve and the error bars determined
by intervals of confidence on the associated probability distribution function (PDF) curve. When
the PDF is broad, there are many possible models in the neighbourhood of the minimum misfit
value that can fit the data to a satisfactory level; in this case the error bars are large. Conversely,
when the PDF curve is sharp, the error bars become smaller since the minimum value of the misfit
is well defined.
In the general case of a Gaussian distribution of errors, the probability of the data (d) given
their predicted value (Gm) and their error bars () is:

1 (d Gm)2
prob(d|m, , L2 ) = exp , (3.5)
2 22
44
but the problem is that the data error bars () are unknown. An estimate of , called ,
can be
inferred by using the data, as shown in chapter 3 of Sivia & Skilling (2006):
s
2m
= , (3.6)
N Tr(R)
where 2m is the minimum value of 2 , N is the number of independent data whose noise is assumed
uncorrelated, Tr(R) is the trace of the resolution matrix (R = G[GT G]1 GT ) and considering the
3 parameters in our model, Tr(R) = 3. 2 is defined as:
!2
2 = d i Gi j m j = e2i = ||e||2 (3.7)
i j i
We assume that the noise in the secular variation estimates obtained from annual means is
uncorrelated. This is reasonable because averaging over a year removes much of the correlated
noise due to external field variations. However, this assumption is not found to be valid for monthly
mean data where a strong influence of temporally correlated external field variations remains. In
order to correctly calculate error bars in t0 for the monthly means, a more complex noise model that
accounts for the time variations in the external field, should be adopted. Because of this limitation
and due to the more scatter present in monthly mean secular variation estimates, in sections 3.6
and 3.7 only error bars derived from annual means are reported.
Substituting the estimate of the data error bars (equation 3.6) in the PDF (equations 3.5):
2
(N 3)
prob(d|L2 ) exp , (3.8)
22m
from which we evaluated the error bars in t0 using 67 % confidence interval. The jerks are classified
as not detected when the minimum of the misfit curve (or maximum in the PDF curve) is in one
of the extremes of the time interval studied, and as excluded when it is not possible to obtain
error bars. That happens when the 67 % of the PDF area can not be completely calculated because
much of the PDF is outside the time interval considered (see examples in Figure 3.5).
3.5.2 Spherical harmonic model for jerk amplitudes

Jerk amplitudes are measured at magnetic observatories and used to build global spherical har-
monic models, where each component of the magnetic field presents a different jerk morphology
(X (a, ,), Y (a, , ) and Z(a,
, ) for jerk amplitudes) because they represent components
of the gradient of the potential function (V ) associated with the jerk morphology which can be
expanded in spherical harmonics:
L a +1

V (a, ,) = a gm
cos m + h m sin m Pm (cos )

=1 m=0 r
where gm and h are the Gauss coefficients (in nT/yr ), r is the radius (r = a Earths radius),
m 2
is the colatitude and the longitude. For each spherical harmonic model we calculate its Lowes
spatial spectrum at the Earths surface:
h i
R = ( + 1) (gm
) 2
+ (h m )2 .
(3.9)
m=0
45
Figure 3.5: Examples of normalized Probability Distribution Functions (PDFs) in (A) Gnangara observa-
tory (GNA, Y component), (B) Port aux Francais observatory (PAF, X component) and (C) misfit curve of
Vernadsky observatory (AIA, Z component), where the 1969 jerk was classified as: detected (A), excluded
(B) and not detected (C).
We solve the linear inverse problem of finding the Gauss coefficients given measurements
of jerk amplitudes in different observatories. We apply the traditional regularization method of
stabilizing the inversion (see, for example Menke (1989), Parker (1994) or Gubbins (2004)) that
involves minimizing a combination of solution norm (N) and error or misfit (E):
T () = E + N, (3.10)
where is called the trade-off or damping parameter which determines the relative importance
of E and N (Gubbins (2004)). We used the norm suggested by Gubbins & Bloxham (1985), to
measure the spatial complexity of an internal potential field model, which in our notation is:
( + 1)2 3 a 2
N = 4 (gm 2 m 2
) + (h ) , (3.11)
,m (2 + 1) c
where c and a are the Earths core and surface radius, respectively. We choose one of the candidate
models at the knee of the trade-off curve (N versus E for each damping parameter ) that represents
a compromise between model complexity and misfit.
46
3.6 Results
3.6.1 Jerk time occurrence
We applied our method of fitting two straight-line segments to secular variation estimates derived
from annual means and monthly means observatory data and annual means synthetic data obtained
from CM4 field model, using both L2 and L1 measures of misfit. Examples of the L2 fit to ob-
servatory data at Fuerstenfeldbruck observatory (Germany) and the corresponding misfit curve is
shown in Figure 3.6. We find that the detected jerk occurrence times are similar for both the L1
and L2 methods and for the different datasets used. Results for occurrence time and error bars, de-
rived using the L2 method, for the X , Y and Z components from the annual mean secular variation
estimates, are shown in Appendix 1.
Figure 3.6: Example of L2 fitting of two straight line segments to the secular variation of Y component of
Fuerstenfeldbruck observatory (FUR, Germany) obtained from first differences of observatory annual means
(A) and its associated misfit curve (B). The solid line (in A) represents the minimum misfit chosen model
and the dashed lines the limits of the error bars.
We tested the L1 method and we found that there are no substantial discrepancies in the results
of t0 detection compared to the L2 method: the patterns of late/early jerks are similar for both. The
L1 method would be advantageous if there were only many outliers in the data. For the 1969 jerk,
occurrence times were determined by the L2 method using the first differences of annual mean
data, first differences of 12-month running average and the CM4 synthetic data. The results for t0
are similar in all three datasets: the patterns of early/late jerks are consistent. One difference of
the results obtained using CM4 synthetic dataset and observatory annual means is that we were
able to detect the 1969 jerk with more confidence (smaller error bars) with the CM4 dataset. The
probable reason is that the synthetic data considers only the core field contribution, which makes
the data less noisy, this jerks appear slightly more clearly. However, since the observatory annual
means constitute the raw data that is geographically better distributed than the monthly mean data,
and largely constraining the CM4 model, we prefer to analyse the first differences of annual means
data directly to determine the robustness of jerk occurrence times.
47
The following features were found to be robust for the 1969 jerk: in the X component, an early
jerk occurrence time was found around China, Korea and Japan, while later jerks were seen in
Europe and Africa (4 observatories, see Figure 3.7 top). In the Y component (Figure 3.7, middle)
jerks appeared earlier in Europe and later in South America (2 observatories) and Africa (3 ob-
servatories). The time occurrence pattern of the Z component is more complex; the most striking
feature is the early arrival in South East Asia region (Figure 3.7 bottom).
The innovative aspect of this work is that we derive error bars in the analysis of jerks non-
simultaneous behaviour. The error bars for European observatories in all three components mostly
overlap (see Figure 3.8 A, B and C) demonstrating the consistency of our analysis. This overlap-
ping is not expected in error bars of observatories in the South hemisphere because they are not
close to each other.
In general, the error bars of all the jerk occurrence times turned out to be asymmetrical (e 6=
e+ ) because of the non-symmetric shape of the calculated PDF curves. The mean value of the
jerk occurrence time is calculated using all observatories, and the locations where the jerk is later
(earlier) than the mean are classified as a late (early) jerk only if the error bars do not cross the
mean value. For example, in the X component of Huancayo observatory (Figure 3.8 A), the error
bar crosses the mean value (1969.94). This also happened for most observatories in the North and
South America, demonstrating that we do not have enough confidence to describe the early/late
patterns in these regions, for the X component.
The mean error bar of the Y component of the 1969 jerk is shown to be smaller than in the X
and Z components (Tables in Appendix 1) probably due to less contamination by the external field.
A generic pattern can be extracted from the occurrence times and error bars in Figure 3.8 B: the
1969 jerk arrives earlier in European observatories compared to the ones in the South hemisphere.
However, in detail our results show a more complex configuration with more geographical variation
in the occurrence times; we found a pattern more similar to P20 than to P10 (see Figure 3.7 middle).
Some examples of the occurrence times for the 1978 jerk and their associated error bars, using
the L2 method and first differences of observatory annual means, are presented in Figures 3.9 and
3.10. The most prominent features found for the 1978 jerk were: early arrival of X component
in the South East Asia region; late arrival of the Y component mostly in Europe, South Africa (3
observatories) and South America (2 observatories). On average the error bars for the 1978 jerk are
larger than in the 1969 jerk, but also showed smaller error bars in the Y component compared to
the X and Z components (see Table in Appendix 1). Examples of large error bars, found at Boulder
(in X ), Gnangara (in Y ) and Kakioka (in Z) are shown in Figure 3.10.
The occurrence time patterns for the 1991 jerk are even more complex than for the previous
jerks (see Figure 3.11): it is not possible to observe this jerk in the X component in most of Europe
and it occurred earlier in most of the US and in the Indian Ocean (3 observatories), while in the
region around India (8 observatories) it arrived later. Conversely, in the Y component a late jerk
is detected in North America and early in Europe. The error bars of European observatories (see
Figure 3.12) mostly overlap, but there are large error bars, for example in the Y component of
Teoloyucan (Mexico).
The possible 1999 jerk (Mandea et al. (2000); De Michelis & Tozzi (2005)) was not identified
in our analysis (Figure 3.13): the most favorable result was in the X component, where it was
possible to detect this jerk in 21 % of the observatories available, most of them being in Europe.
Even in these few observatories where this possible jerk was seen, the typical V shape of a jerk
did not appear clearly, see for example in the case of Chambon-la-Fret and Manhay observatories
(see Figure 3.14).
48
3.6.2 Jerk morphology

Jerk amplitudes were measured using the L2 method applied to first differences of observatory an-
nual mean data, for each magnetic component and for each jerk event. The patterns of positive and
negative amplitudes are well defined geographically, especially for the Y component. For instance,
in Europe the 1969 jerk (Y component) intensities are positive while in Southern Hemisphere,
North America, and in the South-East Asian region the amplitudes are negative (Figure 3.15).
In the subsequent 1978 jerk, the opposite pattern appeared: mostly positive amplitudes in
Northern Hemisphere, South America and South East Asian region and negative in Europe and
Africa (Figure 3.16).
The jerk morphology in 1991 showed similar signs to the 1969 jerk, but with some differences
for example at observatories in North America in which the amplitudes maintained positive values
in the X component (Figure 3.17A). The amplitude range in the 1969, 1978 and 1991 jerks was
about 15 nT/yr2 . The jerk amplitudes were found to be a more robust quantity than the jerk
occurrence times, as illustrated by their small error bars (mean in the Y component of the 1969
jerk: 0.41 nT/yr2).
Using the amplitude measurements from magnetic observatories, we calculated global spheri-
cal harmonic models for the jerk amplitudes in the X , Y and Z components for each geomagnetic
jerk, as described in section 3.5. The spherical harmonic expansion is truncated at degree L = 14
and we explored the influence of varying the damping parameter on jerk morphology. Our pre-
ferred models thus avoid small scale structures in areas with few measurements. Results of spheri-
cal harmonic models at the Earths surface, with different damping parameters (from = 109 up
to = 102 ) are shown in Figure 3.18, for the Y component of the 1969 jerk.
It is clear that in models with very small damping parameters such as = 109 to = 107 ,
the values of amplitude are much higher than those observed in the data. In addition, these models
generate small scale features in locations where there are few data points such as South America
and oceanic areas. Conversely, when one uses a high value of the damping parameter such as
= 101 to = 102 the models are far too simple to represent the observed patterns in the data
and the misfit values increase considerably.
We evaluated the Lowes spectra (Figure 3.19) for each of the 1969 spherical harmonic models
using different damping parameters (Figure 3.18). When the damping parameter is small (such as
= 109 ) the power is spread out to high harmonic degree, generating the models small scale
structure. On the contrary, when one applies a high value for the damping parameter (such as
= 1) the power is confined to low frequencies, mainly at = 1 and = 2. By plotting the trade-
off curve (norm versus misfit for each damping parameter) it is possible to choose a model that is
a compromise between being the simplest model possible and fitting the data well. The trade-off
curve for the 1969 jerk (Figure 3.20) suggested that damping parameters between 105 and 104
are probably the most appropriate since they are located in the knee of the curve. Note that for
the 1978 and 1991 jerks the chosen damping parameters are also = 105 and = 104 (Figure
3.20), though higher misfit values are present in the 1978 jerk. The resulting spherical harmonic
models for the 1969, 1978 and 1991 jerks for all three components are shown in Figures 3.21, 3.22
and 3.23, respectively. It is possible to check that the chosen models for each jerk are in reasonable
agreement with amplitude measurements at each point at the Earths surface (Figures 3.15 to 3.17).
49
Figure 3.7: Results for the 1969 jerk occurrence time at the Earths surface obtained by the L2 method for
the X (top), Y (middle) and Z (bottom) components of the magnetic field by considering first differences of
observatory annual means. The mean occurrence time (e.g. 1969.94 for the X component) is shown close to
the vertical bar (from 1 to 6 years) which gives the height of the blue and red bars. The red bars represent
locations where the jerk appeared later than the mean occurrence time and the blue bars where it appeared
earlier. The occurrence time when the bar is red, is given by the sum of the mean occurrence time and the
height of the bar in a specific location; while the occurrence time of blue bars in a given location is given
by subtracting the mean occurrence time by the height of the bar. Dark red (blue) bars represent locations
where the limits of the error bars are later (earlier) than the mean occurrence time and light red (blue) bars
where the limits of the error bar are earlier (later). The green squares represent the locations where the jerk
was not detected and the black squares where data was excluded.
50
Figure 3.8: Error bars for the 1969 jerk occurrence time for a selection of magnetic observatories in the X
(top), Y (middle) and Z (bottom) components of the magnetic field derived using first differences of annual
means and the L2 method. The observatories used as examples are: Eskdalemuir (ESK, UK), Wingst (WNG,
Germany), Niemegk (NGK, Germany), Chambon-la-Fret (CLF, France), LAquila (AQU, Italy), Huancayo
(HUA, Peru), La Quiaca (LQA, Argentina), Hermanus (HER, South Africa), Gnangara (GNA, Australia)
and Trelew (TRD, Argentina).
51
Figure 3.9: Results for the 1978 jerk occurrence time at the Earths surface by the least-squares method for
the X (top), Y (middle) and Z (bottom) components of the magnetic field by considering first differences of
observatory annual means. The labelling scheme is the same as in Figure 3.7.
52
Figure 3.10: Error bars for the 1978 jerk occurrence time for a selection of observatories in the X (top), Y
(middle) and Z (bottom) components of the magnetic field using first differences of annual means and the L2
method. The observatories used as examples are: Boulder (BOU, US), Fredericksburg (FRD, US), Ottawa
(OTT, Canada), Victoria (VIC, Canada), Leirvogur (LRV, Iceland), Kakioka (KAK, Japan), Port Moresby
(PMG, Papua New Guinea), Apia (API, Samoa), Gnangara (GNA, Australia), Huancayo (HUA, Peru) and
Hermanus (HER, South Africa).
53
Figure 3.11: Results of the 1991 jerk occurrence time at the Earths surface by the L2 method for the X (top),
Y (middle) and Z (bottom) components of the magnetic field by considering first differences of observatory
annual means. The labelling scheme is the same as in Figure 3.7.
54
Figure 3.12: Error bars for the 1991 jerk occurrence time for a selection of observatories in the X (top), Y
(middle) and Z (bottom) components of the magnetic field using first differences of annual means and the L2
method. The observatories used as examples are: Boulder (BOU, US), Fredericksburg (FRD, US), Ottawa
(OTT, Canada), Asahikawa (ASH, Japan), Qeqertarsuaq (GDH, Greeland), Teoloyucan (TEO, Mexico),
Alibag (ABG, India), Port Alfred (CZT, Antarctica), Port aux Franais (PAF, Antarctica), Hermanus (HER,
South Africa), Patamai (PPT, French Polynesia) and Gnangara (GNA, Australia).
55
Figure 3.13: Results for the 1999 jerk occurrence time at the Earths surface by the L2 method to the X
(top), Y (middle) and Z (bottom) components of the magnetic field by considering first differences of annual
means. The labelling scheme is the same as in Figure 3.7.
56
Figure 3.14: Examples of the first differences of annual means (red triangles) at four magnetic observatories
used to analyse the possible 1999 jerk. The solid line is the best fit by using the L2 method and the two
dashed lines represent the 67 % probability error bounds. A and C show the Y component of Stennis Space
Center (BSL, US) and Valentia (VAL, Ireland) observatories; while B and D show the X component of
Chambon-la-Fret (CLF, France) and Manhay (MAB, Belgium) observatories, respectively.
57
Figure 3.15: Jerk amplitudes (nT/yr2 ) for the 1969 jerk detected by the L2 method, by using first differences
of observatory annual mean data. The three maps displayed are the amplitudes of X (top), Y (middle) and
Z (bottom) components. The red and blue bars represent negative and positive values of jerk amplitude,
respectively.
58
of observatory annual mean data. The three maps displayed are the amplitudes of X (top), Y (middle) and Z
(bottom) components. The labelling scheme is the same as in Figure 3.15.
59
of observatory annual mean data. The three maps displayed are the amplitudes of X (top), Y (middle) and Z
(bottom) components. The labelling scheme is the same as in Figure 3.15.
60
Figure 3.18: Spherical harmonic models of the 1969 geomagnetic jerk using simultaneous L2 analysis of
the first differences of observatory annual means for the X , Y and Z components and different damping
parameters A ( = 102 ), B ( = 101 ), C ( = 100 ), D ( = 101 ), E ( = 102 ), F ( = 103 ), G ( = 104 ),
H ( = 105 ), I ( = 106 ), J ( = 107 ), K ( = 108 ) and L ( = 109 ).
61
Figure 3.19: Lowes spectra for the 1969 jerk spherical harmonic models shown in Figure 3.18 for each
damping parameter: A ( = 102 ), B ( = 101 ), C ( = 100 ), D ( = 101 ), E ( = 102 ), F ( = 103 ), G
( = 104 ), H ( = 105 ), I ( = 106 ), J ( = 107 ), K ( = 108 ) and L ( = 109 ).
62
Figure 3.20: Trade-off curves considering the spherical harmonic models, using simultaneous analysis of the
slopes obtained by the L2 method applied to first differences of annual means for the X , Y and Z components
of the 1969 (A), 1978 (B) and 1991 (C) jerks.
Figure 3.21: Spherical harmonic models of the X (A, D), Y (B, E) and Z (C, F) components of the 1969
geomagnetic jerk using different damping parameters shown in the trade-off curve (Figure 3.20 A): = 105
(left column) and = 104 (right column).
63
geomagnetic jerk. The labelling scheme is the same as in Figure 3.21.
geomagnetic jerk. The labelling scheme is the same as in Figure 3.15.
64
3.7 Discussion
The reasons why we choose annual mean values for the detection of geomagnetic jerks at magnetic
observatories are: (i) the annual mean data are less contaminated by the external field compared
to the monthly means, which are in general more scattered (see Figures 3.1 to 3.4); (ii) the annual
mean data are more globally distributed than the available monthly mean dataset (Chulliat & Telali
(2007)); (iii) we presented in the previous section that our results for the jerk detection time from
the monthly means and annual means are similar.
In addition, our results mostly corroborate the ones of Nagao et al. (2003) and Chambodut
& Mandea (2005) which used monthly means and more sophisticated methods to detect jerks.
However, we believe that monthly means may be more appropriate when analysing rapid events
such as the 2003 jerk, because more temporal resolution would be needed. Because the geographic
distribution of magnetic observatories is uneven it is difficult to describe the global behavior of
geomagnetic jerks; we therefore believe that when generalizing jerk global patterns, one should
always mention the number of observatories in which any proposed jerk was detected. In addition,
it is possible to analyze synthetic data evaluated by global field models (such as CM4) but one must
also recognize that they are largely constrained by observatory data with non-uniform distribution.
It is probable that our chosen model of two straight lines is too simple to represent jerks, but
previous works using wavelet analysis and statistical models are equally not based on a physical
argument. All those previous works and the present one aim to find discontinuities in the secular
variation, using different methodologies. The great advantage of our analysis is that we are able
to estimate error bars, which are essential for: (i) characterizing the patterns early/late occurrence
times of geomagnetic jerks with a level of confidence and (ii) further studies relating jerks with
mantle electrical conductivity.
The results of jerk occurrence times in nearby observatories can be inconsistent by a few years
in some regions such as in Europe for the Y component of the 1969 and 1978 jerks (Figures 3.7
and 3.9) and in North America for the X component for the 1978 jerk (Figure 3.9). Although
the majority of the error bars overlap in nearby locations, it is possible that some large departures
occur due to problems in the data, such as outliers. However, we should not exclude the possibility
of early/late jerks in locations close to each other, because if one considers the hypothesis of
jerk differential occurrence times generated totally by a conducting mantle it would be possible
to generate these kinds of patterns depending on the mode mixing (Pinheiro & Jackson (2008)).
The jerk amplitudes (in Figures 3.15 to 3.17) are more robust measurements than the occurrence
times and there are only a few exceptions where there is a change in sign in observatories close to
each other (e.g. Europe for the X component on the 1969 jerk) probably due to errors in the data.
However, there is not much influence of these outliers when we built spherical harmonic models
of jerk amplitudes (Figures 3.21 to 3.23).
A summary of the results of all mentioned papers is shown in Table 3.1. Nagao et al. (2003)
applied a statistical model to find the 1969, 1978 and 1991 jerks by using the monthly means of the
Y component of 47, 57 and 45 observatories, respectively, of which a maximum of 12 observatories
were in the Southern Hemisphere for the 1991 jerk. This illustrates how the monthly mean data
are more unevenly distributed than the annual mean data. Chambodut & Mandea (2005) used the
secular variation of the Y component including monthly means data, where a wavelet analysis was
performed, and synthetic data from CM4, in which they applied the conventional method of fitting
straight lines. One has to be careful when comparing maps of the global occurrence of geomagnetic
jerks because late and early jerks are relative to a given value; in Chambodut & Mandea (2005)
65
the analysis of the 1969 jerk is centered at 1971.0 while in Nagao et al. (2003) it is discriminated
as before or after 1971.5. We believe that the most appropriate way to define early and late jerks
is relative to the mean occurrence time considering all observatories where the jerk was detected
(e.g. 1969.90 for the Y component of the 1969 jerk, see Figure 3.7).
Concerning the 1969 jerk (Y component) our results mostly agree with Nagao et al. (2003)
and Chambodut & Mandea (2005): in Europe the jerk appeared earlier and in the few observa-
tories available in Africa and Australia, it appeared later with a time lag of about 3 years. In the
South East Asia region this jerk also appeared approximately at 1971, which is considered to be
slightly late in our analysis and slightly early in Nagao et al. (2003) and Chambodut & Mandea
(2005). The South American results also agree with Chambodut & Mandea (2005) and could not
be compared with Nagao et al. (2003) due to an insufficient number of observatories. We also
found some agreement of zones where this jerk was not identified as in parts of Africa and India;
this corresponds to blind zones in Chambodut & Mandea (2005) analysis and to green squares in
ours (Figure 3.7). Obviously some differences in the results will appear due to the use of different
datasets and methodologies.
The zero time delay chosen for the 1978 jerk in Nagao et al. (2003) and Chambodut & Mandea
(2005) was 1980.0, while in our analysis the mean jerk occurrence time was found to be 1978.08
for the Y component. Therefore in Europe, for example, the small red bars (Figure 3.9) agree with
early jerks found in both previous works. However, even in Europe there are some late jerk arrivals
found geographically close to early ones, being most of them consistent considering the limits
of the error bars. The results in South Africa, Australia are considered late in all three analysis
and in Western Africa we also found two observatories with an earlier arrival that corresponds to
the blue spot shown in Figure 7 of Chambodut & Mandea (2005). De Michelis & Tozzi (2005)
also analyzed the 1978 jerk applying a new approach involving wavelet analysis, and chose the
reference date to be 1979.0. They found a more simple pattern of early/late jerks that agrees mostly
with Nagao et al. (2003) but disagree specially in South America with Chambodut & Mandea
(2005) and our present results.
For the Y component of the 1991 jerk, Nagao et al. (2003) chose 1993.58 to be the reference
date, while in our work and Chambodut & Mandea (2005) the reference date is around 1991.
De Michelis et al. (2000) also analysed this jerk and found the average value to be 1990.1 0.6.
The morphology of this jerk turns out to be more complicated than for previous jerks in all the
studies. However, there is also more disagreement between different works, for example in Africa
we were not able to detect the 1991 jerk while Chambodut & Mandea (2005) and De Michelis
et al. (2000) found it to be earlier than the time inferred by Nagao et al. (2003). De Michelis &
Tozzi (2005) found 1992 as the average value for this jerk with an early jerk arrival in Europe and
Asia and a late arrival in most of Africa and North and South America, in contrast with the other
studies that identified more complex patterns. We are not able to compare directly our results to
Alexandrescu et al. (1996) since they used instead a linear combination of X and Y components.
Mandea et al. (2000) were the first to mention the possible presence of a jerk at the end of the
20th century by analysing the first differences of 12-month running average of the Y component
and detected visually the 1999 jerk in 9 observatories: CLF, NGK, BFE, DOU, ESK, FUR, HAD,
LER, NUR. De Michelis & Tozzi (2005) could not apply their wavelet analysis to perform a global
search for the 1999 possible jerk, due to boundary effects in the wavelet analysis. However, they
applied the traditional straight-line fit to the monthly mean data for this jerk, including 44 geomag-
netic observatories (only 5 in the Southern Hemisphere) and claimed that the 1999 jerk could be
identified in observatories in Europe, North America, Asia, Africa and Pacific Ocean (identified in
66
total in 29 observatories). We believe that De Michelis and Tozzis [2005] argument is not enough
to classify the 1999 jerk as global; for example, the 1999 jerk was detected only in 4 observatories
in the Southern Hemisphere, and the observatories where it was detected are mostly concentrated
in Europe and North America. In addition, our results suggest that there is no global jerk in 1999
and in most of the few observatories where these two straight lines could be fitted to the data, we
found that the typical jerk V shape can not usually be seen in the data. It is also possible that
the lack of temporal resolution by using annual mean data make it more difficult to detect this jerk
than by using monthly mean data. However, in De Michelis & Tozzi (2005) it was not specified
which correction they performed in the monthly means used to analyse the 1999 jerk; therefore it
is possible that their results are influenced by the external fields.
In relation to the jerk amplitudes our results can be compared to Courtillot & Le-Mouel (1984)
for the 1969 jerk, Le Huy et al. (1998) for all three jerks (1969, 1978 and 1991) and De Michelis
et al. (2000) for the 1991 jerk. In general, our jerk morphology models are similar to the previous
results although they vary according to the damping parameter we applied. It is clear that there are
small differences between Courtillot & Le-Mouel (1984), Le Huy et al. (1998), De Michelis et al.
(2000) and the present works spherical harmonic models, since different datasets and methods
are used. We found similar Lowes spectra for the 1969 and 1978 jerks with a peak at = 3 and
for the 1991 jerk two peaks at = 1 and = 3 (Figure 3.24). The harmonic content of the jerk
signal, which differs from the present work to the previous papers, will be crucial in our subsequent
analysis of jerk observations to infer mantle electrical conductivity. Backus (1983) and Pinheiro
& Jackson (2008) have shown that the phenomena of mode mixing, where different spherical
harmonic degrees diffuse at different rates through the mantle, is fundamental for explaining jerk
differential occurrence times at the Earths surface.
Figure 3.24: Lowes spectra of the 1969 (A), 1978 (B) and 1991 (C) spherical harmonic models using
simultaneous analysis of the jerk amplitudes in X , Y and Z components derived from L2 analysis of first
differences of observatory annual means for the damping parameters: = 105 (blue triangles) and = 104
(green triangles).
3.8 Summary
The data analysis of geomagnetic jerks that are reported to have occurred in 1969, 1978, 1991
and 1999 was performed using observatory annual mean data, observatory monthly mean data and
the core contribution calculated from the CM4 model. Two methods of fitting straight lines to the
Data Method Jerks
Le Moul et al. (1982) annual means (X , Y , Z) Least-squares fit two straight-lines 1969
130 observatories
Le Huy et al. (1998) smoothed annual means Least-squares fit two straight-lines 1969, 1978, 1991 global
160 observatories
Alexandrescu et al. (1996) monthly means: linear Wavelet analysis 1901, 1913, 1925, 1932,
comb. of X and Y 74 observatories 1949, 1969, 1978
De Michelis et al. (2000) annual means (X , Y and Z) Least-squares fit two straight-lines 1991 global
109 observatories
Mandea et al. (2000) monthly means (Y ) Visual 1999 jerk in 9 observatories
12 month running average
67
Nagao et al. (2003) monthly means (X , Y and Z) Statistical model 1969, 1978, 1991 global
50 observatories Least-squares fit two straight-lines
Chambodut & Mandea (2005) monthly means (Y ) and Wavelet analysis / 1969, 1978, 1991 global(?)
synthetic data from CM4 Least-squares fit two straight-lines
De Michelis & Tozzi (2005) monthly means Wavelet analysis 1978, 1991, 1999 global
44 observatories Local Intermittency Measure 1986 local
Olsen & Mandea (2007) satellite monthly means Spherical Harmonic Expansion / 2003
(virtual observatories at 400 km) Least-squares fit two straight-lines
present work annual and monthly means and Least-squares fit two straight-lines 1969,1978,1991 global
synthetic data from CM4 1999 not detected
Table 3.1: Summary of seven previous works about detection of geomagnetic jerks, specifying the data, the methods used and which jerks were detected.
68
secular variation were applied: L1 and L2 . It was expected that the L1 approach would allow a
better fit to the observatory data since it is not as sensitive to noise and outliers as the L2 method.
However, analysis of the 1969 jerk showed that there are only small differences between the two
methods and that there is not much advantage in applying the L1 instead of L2 method. Since jerk
time delays obtained from the annual means data, monthly means data and the CM4 model showed
similar patterns, we presented in detail only results of the L2 method applied to the first differences
of observatory annual means.
The analysis of error bars in the occurrence time of geomagnetic jerks proved to be essential
in order to reliably distinguish between early and late jerks. In general, the error bars in t0 were
smaller in the Y component than in the X and Z components. The 1969, 1978 and 1991 jerks
were detected worldwide in the secular variation estimates derived from annual means, while the
proposed event in 1999 was detected only locally. The non-simultaneous behaviour of geomagnetic
jerks presented a different pattern for each magnetic component and for each jerk event. Our
analysis showed that the error bars for the Y component are in general smaller than in the X and Z
components, probably due to a smaller influence of external field variations, therefore we can more
confidently describe the non-simultaneous behaviour in the Y component. In general, the Southern
hemisphere observatories presented larger error bars.
Instead of a general description of late in the Southern Hemisphere and early in the Northern
Hemisphere, our analysis showed a more complex pattern with generally early arrivals in Europe
and later arrivals in South America and Africa (both have few observatories available) in the Y
component of the 1969 jerk. Conversely, in the X and Z components a later arrival of the 1969 jerk
appeared in Europe and Africa and mostly early were found in South America. In the Y component
of the 1978 jerk we found a late time occurrence in most of Europe and South Africa and for the
same component of the 1991 jerk we observed an early arrival in Europe and a late arrival in North
America and South East Asia.
The patterns of positive/negative jerk amplitudes were better defined in the Y component and
at most locations presented opposite signs in consecutive jerks. The amplitude measurements were
used to build spherical harmonic models. By using different damping parameters we evaluated the
trade-off curve for each geomagnetic jerk, in order to choose the most appropriate models for the
jerk morphology. The chosen models in all three jerks have damping parameter values between
= 105 and = 104 and power spectra that vanish at harmonic degrees bigger than five (L = 5)
for the 1969 and 1978 jerks and bigger than six (L = 6) for the 1991 jerk. The strongest peaks in
the power spectra for the 1969 and 1978 jerks are at = 2 while for the 1991 jerk the strongest
peak is found at = 3.
The results obtained in this jerk data analysis may contribute to more information about the
Earths interior, such as the core processes generating geomagnetic jerks and the estimates of the
electrical conductivity of the mantle. One possible way to explain the geomagnetic jerk time delay
patterns is by considering a 1D electrically conducting mantle and the morphologies that vary for
each magnetic component and each magnetic jerk. Pinheiro & Jackson (2008) demonstrated that
even by assuming a simple model of a radial conducting mantle and the effects of some harmonic
mixing (varying for each jerk and component) it is possible to generate differential jerk occurrence
times (Chapter 4). Since the deep mantle electrical conductivity is poorly known, we solve the
inverse problem of determining mantle electrical conductivity from jerk differential occurrence
times (Chapter 5), where their associated error bars that were estimated in this Chapter.
69
Chapter 4
The Forward Problem
There are two simple hypotheses to explain the differential delays pattern of geomagnetic jerks: the
first is to consider these differential time delays as being generated by dynamical processes in the
core which do not occur simultaneously; the second is to consider jerks generated instantaneously
in the core and the time delays as being caused entirely by a conducting mantle. In this chapter,
we analyse the second hypothesis and we assume a jerk to be an impulse in time at the CMB with
a morphology given by two synthetic models. A summary of Backus [1983] theory was given
in chapter 2 and applied to an arbitrary 1D mantle electrical conductivity model to illustrate the
physical effect of the mantle filter. We calculate the resulting time delays for the same radial model
(equation 2.27) but changing c and (Figure 4.1).
Figure 4.1: Delay times calculated for a radial man-

tle conductivity model (in equation 2.27: (r) =
2+2
c cr ) for different parameters c and . The re-
sults are shown for different harmonic degrees: = 1
represented by solid lines and for = 2 by dashed
lines.
Figure 4.1 summarizes the effect of the mantle filter by using a specific radial model: the delay
times increase with a decrease of and with an increase of c ; lower harmonic degrees are more
delayed than higher harmonic degrees. But in this chapter we are interested in demonstrating that
a 1D mantle conductivity model is able to generate differential jerk time delays varying spatially
and, which are different in the X , Y and Z components at the Earths surface. We show that the
mixing of harmonics is the reason for such differential delays and that it is not necessary to invoke
laterally varying conductivity models. Since the jerk morphology varies for different magnetic
70
components and jerk events (such as the 1969, the 1978 and the 1991 jerks), the differential time
delays at the Earths surface also do so.
We apply this theory to a real morphology of a jerk taken from Le Huy et al. [1998] spherical
harmonic analysis of the 1969 jerk. The resulting differential time delays that would be seen in
a realistic distribution of observatories is shown. This also leads to the conclusion that since jerk
morphologies are different in distinct jerk events, then the pattern of differential time delays will
also change as a function of the particular event.
4.1 Synthetic Models

4.1.1 A simple model
One of the characteristics of geomagnetic jerks that is not completely understood is that the jerk
arrival times are not the same for each component of the magnetic field. To illustrate this, Figure
4.2 shows the secular variation of the North, East and Vertical components of Niemegk observatory
where the 1969 jerk occurs at slightly different times.
Figure 4.2: Secular variation of the X (A), Y (B) and Z (C) components of Niemegk observatory, in Germany.
The green lines are the secular variation using annual means and the gray lines are the secular variation
corrected for external and induced effects by averaging of hourly predictions from model CM4 (Sabaka
et al. (2004)).
71
It is also important to notice that jerks will be usually better seen in the East component because
it suffers less influence of the magnetospheric field. In addition, the jerk time detection will depend
on the method applied, as for example, if it is detected by the fitting of two straight lines or by more
sophisticated methods, such as wavelet analysis.
Jerk amplitudes are measured at the Earths surface (A(a, , )), at magnetic observatories, and
used to build global spherical harmonic models. One important issue is that each component of
the magnetic field at the Earths surface (X (a,, ), Y (a, , ) and Z(a, , )) presents a different
jerk morphology (notation X(a, , ), Y (a, , ) and Z(a,
, ) for jerk amplitudes) since they
represent components of the gradient of the potential (V ):
1 V (a, ,) 1 V (a, , ) ,) = V (a, , ) ,

X (a, ,) = , Y (a, , ) = , Z(a,
r r sin r
where the potential can be expanded in spherical harmonics:
L =1 a +1
V (a, ,) = a r g cos m + h sin m Pm (cos )
m m
=1 m=0
and the coefficients gm

l and hl are in nT/yr . In order to illustrate that it is possible that different
m 2
components have not the same delay time, we present a simple example in Figure 4.3.
Figure 4.3: Hypothetical spherical harmonic models of X (A), Y (B) and Z (C) used as a simple example to
calculate the CIRF and output jerk in Figure 4.4.
72
We consider the more conducting radial model given in Figure 2.1 (c = 3000 and = 11) and
assume only two non-zero Gauss coefficients: g10 = 1 nT/yr2 and g12 = 1 nT/yr2 . For this model,
the jerk amplitudes for each magnetic component can be calculated by:

X = sin 3 cos2 cos2 sin2 , (4.1)

Y = 3 sin cos cos , (4.2)

Z = 2 cos 3 3 cos2 cos sin , (4.3)
which shows that when = 90o , Y and Z are zero and when = 90o then Y = 0. Note that
for Y there is only contribution of = 2 and if = 45o then X will be zero for any longitude .
We choose arbitrary locations: 1 = 45, 1 = 45 for location 1 and 2 = 120, 2 = 0 for location 2
in order to calculated the CIRF and the output jerks for the X , Y and Z components (Figure 4.4).
Results of location 2 illustrate that it is possible that the same jerk event is not observed in the three
components for the same location. In this case, the jerk would be seen in the X and Z component
but not in the Y (see equations 4.1 - 4.3). The delay time for the three components is measured
by the first ordinary moment (in Backus theory) and by the fitting of two straight-line segments
in which the difference of the delay time of X and Z reach up to 0.38 yr and up to 0.07 yr if one
compares the two locations.
Figure 4.4: Examples of CIRFs (A,B) and output jerks (C,D) in two locations for X (green line), Y (purple
line) and Z (orange line) components. We chose location 1 (A,C) and 2 (B,D) at 1 = 45, 1 = 45 and
2 = 120, 2 = 0, respectively. The Gauss coefficients for the jerk morphology were simulated to be:
g10 = 1 nT/yr2 and g12 = 1 nT/yr2 . These examples were calculated for a highly conducting radial model
given in Figure 2.1 (c = 3000 and = 11).
73
4.1.2 A more complex model

We consider two synthetic models defined in terms of their Lowes spatial power spectrum (equation
3.9):
h i
R = ( + 1) (gm ) 2
+ (h m )2 .

m=0
The first model has two peaks at = 1 and = 2 with a truncation degree L = 2, while the
second model presents three peaks at = 2, = 5 and = 8 and L = 8 (Figure 4.5). Both models
were built with a power spectrum with peak values of R = 1, where each coefficient (gm m
and h )
of each harmonic degree has the same value. For example, in the case of the two peaked model,
the three non-zero coefficients
of = 1 are all equal to 1/ 6 and the five non-zero coefficients of
= 2 are equal to 1/ 15 which gives as a result two peaks R 1 = 1 and R 2 = 1.
Figure 4.5: Power Spectrum of the two synthetic models of jerk amplitudes.
The synthetic spherical harmonic models are shown in Figure 4.6 for the North (X ), East (Y )
and vertical (Z) components of the magnetic field. Two radial models are created, following equa-
tion 2.27: the first is a weakly conducting model with c = 100 S/m and = 8 and the second
a very strong conductor with c = 3000 and = 11, both shown in Figure 2.1. In the weakly
conducting model, the electrical conductivity at the surface is 0.0019 S/m and 0.0416 S/m at 1000
km depth, while the highly conducting model goes from 0.0015 S/m at the surface up to 0.0930
S/m at 1000 km. It is clear that both are not realistic models for the Earth, but instead they are
used to illustrate two issues: how the delay times depend on the conductivity and how a 1D mantle
conductivity model can cause differential time delays at the Earths surface.
We used as an example the highly conducting model in Figure 2.1 (c = 3000 and = 11) and
calculated the time delays of the two synthetic morphology models (Figure 4.7). The patterns of
delay times of X , Y and Z components are different depending on their morphologies. This would
explain why jerks are not seen at the same time in different components. If there is more mixing
of harmonics, the jerk morphology will be more complex as also will be the pattern of time delays
(see comparison in Figure 4.7).
74
Figure 4.6: Synthetic spherical harmonic models evaluated from the synthetic models (Figure 4.5) at the
core-mantle boundary. The synthetic model with two peaks in the power spectrum is shown in A for the
North component of the magnetic field (X ), in B for the East component (Y ) and in C for the vertical
component (Z). The three peaked synthetic model is shown in D for X , in E for Y and in F for Z. The color
scale in nT/yr2 .
Figure 4.7: Results of the forward modelling calculation of time delays for the two-peaked model (A,B,C)
and for the three-peaked model (D,E,F) by using the highly conducting model. In A and D the time delays
are for the North component of the magnetic field (X ), in B and E for the East component (Y ) and in C and F
for the vertical component (Z). The color scale is in years and the red would mean late jerks, while the blue
earlier jerks. The areas with gray color correspond to locations where the time delay could not be detected
because of zero jerk amplitude (see Figure 4.6).
75
4.2 Geophysical application

Let the input be written in a general form as pm
(c,t) = (t)A (c, , ), where (t) is the Dirac
m
delta function, and equation 2.34 becomes:

Z
pm
(a,t) = (c, , )
G Am F (t t )(t )dt

(c, , )F (t),
= G Am (4.4)
Spherical harmonic models are constructed by using measurements of jerk amplitudes at the
Earths surface. However, to obtain the input model at the CMB (Am (c, , )), the mantle is as-
sumed initially to be an insulator and a downward continuation, from the surface to the CMB, is
applied:
a (+1)
A (c, , ) = A (a, , )
m m
. (4.5)
c
R
The mantle filter does not modify jerk amplitudes since F (t)dt = 1, it only delays and smooths
the input jerk. Some examples of the first, second and third differences of the output jerk are
shown in Figure 4.8 for two radial mantle electrical conductivity models (in equation 2.27 for
c = 1000 S/m and = 3). This example illustrates that the asymptotes are not affected by the
mantle conductivity.
Figure 4.8: Synthetic examples of the third derivative of the magnetic field (in A), the second derivative (in
B) and the first derivative corresponding to the secular variation (in C). The black lines represent the results
for an insulating mantle, the purple line for the location correspondent to LAquila observatory (AQU, Italy),
the blue line in Huancayo observatory (HUA, Peru) and the green line in Addis Ababa observatory (AAE,
Ethiopia). These results were calculated considering Le Huy et al.s [1998] model of the 1969 jerk and for a
radial mantle conductivity model (in equation 2.27 for c = 1000 S/m and = 3).
In addition, by using equations 4.4 and 4.5:

c +1
p (a,t) =
m
Am (c, , )F (t)
a
c +1 a (+1)
= Am (a, , ) F (t)
a c
= Am (a, ,)F (t), (4.6)
76
we show that to solve the forward problem we can use the spherical harmonic model of jerk am-
(a, ,)) and ignore the geometrical attenuation, instead of using the
plitudes at the surface (Am
model at the CMB. In this section, Backus(1983) mantle filter theory is applied to two real geo-
magnetic jerk morphologies (East component) of the 1969 jerk taken from Le Huy et al. (1998)
who analysed 123 observatories and of the 2003 jerk taken from Olsen & Mandea (2007) who used
CHAMP satellite data (Figures 4.9 and 4.10).
Figure 4.9: Spherical harmonic model obtained from Le Huy et al. (1998) of the East component of the
1969 geomagnetic jerk at the Earths surface (in A). Spherical harmonic models of each harmonic degree at
the Earths surface for: = 1 (in B), = 2 (in C), = 3 (in D) and = 4 (in E).
The truncation degree used by Le Huy et al. (1998) was taken as L = 4, and for each spherical
harmonic degree we evaluated the associated jerk amplitude model (Figure 4.9), while for Olsen &
Mandea (2007) the truncation degree was taken as L = 14. Conversely, we assume an unrealistic
highly conducting model for the mantle, as shown in Figure 2.1 (c = 3000 and = 11). The aim
of this section is to demonstrate that the same arbitrary 1D electrical model is able to generate
differential time delays at the surface at a realistic distribution of observatories.
By having both jerk amplitudes and IRF for each we can introduce the Composite Impulse
Response Function (CIRF, F(r, , ,t), in short notation F(t) ) that is the sum of the IRF (F (t))
multiplied by jerk amplitudes (Am (r, ,)), for each harmonic degree (equation 2.60). It is obvious
that since jerk amplitudes vary spatially, the CIRFs will also do so (equation 2.60). If we look at
one observatory location for Le Huy et al.s model, for example LAquila Observatory (Italy), we
77
Figure 4.10: Morphology of the 2003 geomagnetic jerk for the East component obtained from Olsen &
Mandea (2007). Their model is truncated to spherical harmonic degree L = 14.
can notice how a CIRF is generated by the linear combination of the IRFs multiplied by the jerk
amplitude at that location, for each (Figure 4.11).
Figure 4.11: In A the Impulse Response Function multiplied by the amplitude value for each harmonic
degree (), of the location corresponding to LAquila magnetic observatory (Italy). By comparison with
Figure 4.9 it is possible to check that in Italy, for = 1 and = 4 the amplitudes are negative, while for = 2
and = 3 the amplitudes are positive. The sum of these curves result in the Composite Impulse Response
Function (in B).
By evaluating the CIRF of each location we can measure its delay time by equation 2.65. As the
CIRFs vary spatially it is a straightforward conclusion that delay times (1 ) will also be different at
each location. We can then evaluate 1 , using Le Huy et al.s (1998) spherical model and the same
1D highly conducting mantle model (Figures 4.9), in different locations. The result is presented
78
in Figure 4.12 for a typical distribution of observatories, where the blue bars would correspond to
early jerks and red bars to late jerks. For instance, in Europe the bars are small because the delay
time values are close to the mean delay time. In this example, the differential delays are of the
order of 2 years (see vertical scale in Figure 4.12).
Figure 4.12: Results of delay times of the 1969 (A) and 2003 (B) jerks for the Y component. In A, the
Le Huy et al. (1998) spherical harmonic model of the 1969 jerk (Figure 4.9) is used (mean delay time of
2.29 years), while in B, Olsen and Mandeas [2007] model for the 2003 jerk (Figure 4.10) with a mean delay
time of 2.25 years. In both models a highly electrically conducting mantle model (Figure 2.1) was used.
The yellow areas correspond to extremely high delay time values, blue bars represent early jerks and red
bars are late jerks. The stars (in A) are at the location of observatories used as example in Figure 4.13.
However, it is important to stress that these results should not be compared with real differential
79
delays measured in observatories because we are assuming an arbitrary 1D mantle electrical model.
In addition, small changes in the jerk morphology can cause substantial differences in the pattern
of delay times. We also applied the highly conducting mantle model to the 2003 jerk morphology
(Y component) and we found differential delay times of the order of 3 years. The 2003 jerk was
reported to be concentrated near 90o E, with maximum jerk strength at about 30o latitude and
with small differential delay times of the order of a month (Olsen & Mandea (2007)). By the
application of this forward approach we found also very small differential delays in this area.
Some examples of CIRF and the resulting jerk at the surface are shown in Figure 4.13 for the
location corresponding to the three observatories shown in Figure 4.12. The extremely high values
of delay time (yellow area in Figure 4.12) are caused by an oscillatory CIRF, as shown in Figure
4.13 for the location corresponding to Addis Ababa (AAE) observatory.
Figure 4.13: Synthetic jerks in the case of an insulating mantle (black V shape) and of a conducting
mantle (red line), which is the result of the convolution between the CIRF of each observatory (in the right
column) and the input jerk at the CMB. These results consider a high mantle electrical conductivity model
(in Figure 2.1 for c = 3000 and = 11) and the jerk morphology of Le Huy et al. (1998) (in Figure 4.9).
Different observatory locations are considered to illustrate the delay and smoothing caused by a hypothetical
electrical conducting mantle: in A Niemegk with a positive CIRF, in B Vassouras with a negative CIRF and
in C Addis Ababa (AAE) with an oscillatory CIRF. The amplitudes of these hypothetic jerks are roughly the
same as the real ones. Note that in C the amplitude of the jerk is close to zero, therefore it is not visible.
80
The estimated delay times in oscillatory CIRFs have extremely high values because when the

area under the CIRF curve is close to zero (F(0) = 0) then 1 values will be also extremely high
(see equation 2.46). These anomalous values of delay times (yellow areas in Figure 4.12) coincide
with locations of zero amplitude shown by the Le Huy et al. (1998) and Olsen & Mandea (2007)
spherical harmonic models (Figures 4.9 and 4.10). Locations of jerk zero amplitude would mean
that the jerk could not be observed and therefore delay times could not be measured.
4.3 Summary
We applied Backus [1983] mantle filter theory and verified that the mantle filter acts differently
on each harmonic degree. The mixing of harmonics varies with location, depending on the jerk
morphology at the Earths surface. As a consequence, the arrival times of geomagnetic jerks that
are measured by the first moment of the composite impulse response curve, will be different at
each location. In order to illustrate that, we used an arbitrary 1D radial conducting model of the
mantle, first applied to two synthetic models of jerk morphology and then to the 1969 and 2003
jerk morphologies obtained from Le Huy et al. (1998) and Olsen & Mandea (2007).
The present work demonstrates that: (1) assuming a simultaneous jerk at the CMB, a 1D man-
tle conductivity model is able to generate jerk differential delay times at the Earths surface; (2)
synthetic examples of jerk morphology of the X , Y and Z components demonstrated that the pat-
tern of delay times would be different for each component. This would explain why jerks are not
seen in observatory data at the same time in different components of the magnetic field; (3) the
application of this theory to the 1969 and 2003 jerk morphologies showed that it would also be a
natural consequence to have different delay times for each jerk. It is clear that the pattern of delay
time at the Earths surface is very sensitive to the input model of jerk morphology.
We used two unrealistic mantle conductivity profiles for the Earth, but they illustrated that
the observed differential delay times of the order of 2 years seems to require a highly electrically
conducting lower mantle. In Chapter 5 we solve the inverse problem in order to obtain some
constraints on mantle electrical conductivity models based on the observation of differential delay
times at the Earths surface.
81
Chapter 5
The Inverse Problem
The aim of this chapter is to obtain information about mantle electrical conductivity by using ob-
servations of differential delay times of the 1969 global geomagnetic jerk (Y component). We
consider Kuvshinov and Olsens (2006) model of conductivity of the upper mantle and models of
one to three layers below 700 km. In the first section we present a brief introduction about inverse
problems, while in section 5.2 we present the methodology, comparisons between analytical and
numerical (code of Velmsk & Martinec (2005)) solutions for the IRF using radial mantle con-
ductivity models, and the setup of the inverse problem. The results and discussion of simulations
are presented in section 5.3.
5.1 Introduction
Menke (1989) defines inverse theory as a set of mathematical techniques for reducing data to obtain
useful information about the physical world on the basis of inferences drawn from observations. In
this work, our observations are differential delay times for the 1969, 1978 and 1991 geomagnetic
jerks, measured in the X , Y and Z components in magnetic observatories.
The main difference between discrete and continuous inverse theory is whether the model is
treated as continuous functions or discrete parameters. The data will be always discrete and a con-
tinuous problem can be converted in a discrete one by approximating the integral as a summation
using, for example, the trapezoidal rule. When one sets up an inverse problem it is necessary to
choose variables to represent data and model parameters. In some cases there might be no strong
reasons to select a specific parametrization.
There are three very important topics associated with the solutions of inverse theory: existence,
uniqueness and stability. If the mathematical model is inconsistent with the observations, no model
will exist. For example, if our assumption that jerks are simultaneous in the core is not valid, it
is possible that a solution does not exist. Because the measurements are not exact, there has to be
a criteria to decide when the predictions of the mathematical model agree well with observations
(such as by using the misfit). The residuals are defined as the difference between the data and the
model:
r = d G(m)
where G(m) is the forward functional generating synthetic data from the model m. The solution(s)
is (are) based in the minimization of some measure of misfit.
If the condition of existence of a solution is satisfied, one needs to know how many solutions
82
may exist. While the forward problem has a unique solution, the inverse problem does not. That is
the reason one needs to make explicit any available a priori information on the model parameters
and a careful representation of the data uncertainties should be performed (Tarantola (2005)). Our
calculations of jerk occurrence time and amplitudes included error bars (see Chapter 3) that are
used to solve the inverse problem.
An unstable problem happens when a small variation (or error) in the data generates a large
departure from the original solution; it can be tested by changing slightly the data and comparing
the solutions. We have not analyzed this issue in this study. For linear inverse problems this issue
is well understood: it is associated with a large condition number in the matrix linking the model
to the data, the so-called equations of condition matrix. The condition number measures the ratio
of the smallest to the largest eigenvalue. It is the presence of small eigenvalues in the matrix that
can lead to a catastrophic amplification of any errors present in the data. Since our problem is
non-linear, the tools of linear inverse theory are not available to us. Instead we have analyzed an
allied problem of stability, namely the sensitivity to the forward model. We tested the stability of
the present inverse problem by using the same parametrization, but varying slightly the model for
the jerk morphology used as input, which tests the sensitivity to errors in the forward functional;
the results are discussed in section 5.3.
5.2 Methodology
In view of the fact that the inverse problem is non-linear, we solve it by exhaustive search, where
the forward problem is solved repeatedly by changing the electrical conductivity of mantle layers
and by calculating the jerk differential delays at the surface. The forward problem (discussed in
Chapters 2 and 4) can be summarized as: the output (jerk at the surface, pm (a,t)) is evaluated by
the convolution between the input that simulates an impulsive jerk generated simultaneously (at
t = 0) at the CMB and the Composite Impulse Response Function (CIRF, equation 2.60) for each
jerk.
We followed seven main steps to solve the inverse problem:
1. Calculation of the Impulse Response Function (IRF) is performed by a modified version of
Velmsk and Martinecs [2005] code. The method is based on direct time-integration of the EM
induction equation, and uses spatial discretization by spherical harmonics and piecewise-linear fi-
nite elements in the lateral and radial direction, respectively. The electrical conductivity is given as
a piecewise constant function over layers of arbitrary thickness. Lateral variations of conductivity
are neglected in this case. The boundary condition at the CMB requires prescription of the time
series spherical harmonic coefficients of the vertical field. When these time series take the form
of a delta function, which is numerically approximated by a narrow, zero centred and triangular
function normalized to unit integral, the predicted vertical field at the Earths surface corresponds
to the product of the IRF and the geometrical attenuation Gl . The details of the setup in Velmsk
and Martinecs [2005] code is given in Table 5.1. The Impulse Response Function calculated for
each model was truncated where it reached 5% of the peak value considering = 1 since the IRFs
for higher harmonic degrees are shorter in time.
We compared Velmsk and Martinecs [2005] numerical calculations of IRFs, using piece-
wise constant conductivity models, with analytic solutions using a power law (equation 2.27). We
considered two radial 1D mantle electrical conductivity models: the first model setting c = 100
S/m and = 11 and in the second model, c = 1000 S/m and = 11. The difference between the
83
time step 1 hour

half width of delta function 2 time steps
parametrization layers 5 km
variation of conductivity from 1000 S/m to 0.4 S/m in steps of 3
truncation IRF 5% of the peak value
Table 5.1: Setup of Velmsk and Martinecs (2005) code to calculate the IRFs.
results clearly increases with the harmonic degree number (see Figure 5.1). When one increases
the number of layers used to calculate the radial model the agreement between the two solutions
improves considerably: with 10000 layers the difference between the solutions is imperceptible.
Such a high resolution is required to describe the rapidly growing power-law conductivity model
by piecewise constant functions with sufficient accuracy.
2. Evaluation of the Composite Impulse Response Function (CIRFs, see Pinheiro & Jackson
(2008) and equation 2.60) by using the spherical harmonic models of jerk amplitude. In this chapter
we only use the Y component of the 1969 jerks and jerk amplitude models of Chapter 3, and Le
Huy et al.s (1998) models (see Figure 5.2);
3. Convolution of those CIRFs with the input jerk (Figure 1.10), simulated as a second order
impulse in the poloidal field, simultaneous at the CMB and with unit amplitude;
Figure 5.1: Comparison between analytical and numerical solutions of IRFs for the two mantle radial con-
ductivity models (given in equation 2.27) for = 1 (red), = 3 (blue) and = 5 (black): c = 100 S/m and
= 11 (in A and C) and in the second model c = 1000 S/m and = 11 (in B and D). In Velmsk and
Martinecs [2005] code the number of layers is set as 1000 (A, B) and as 10000 (C, D).
84
Figure 5.2: Spherical harmonic models of the 1969 jerk amplitudes calculated in this work (A) with trunca-
tion degree L = 5 and by Le Huy et al. (1998) (B) with truncation degree L = 4, both used in this Chapter to
calculate the Composite IRFs.
4. Annual mean evaluation of the output jerk: since we aim to treat the model in the same way
as the data (see Chapter 3);
5. Fitting of two straight-line segments to the output annual means by the least-squares method:
the intersect is defined as the delay time (1 ) and the error bars in 1 are evaluated as described in
Chapter 3;
6. Calculation of jerk differential delay times for all locations corresponding to the analysed
observatories. The differential delay times (1 ) are calculated in relation to the mean delay value
(m ): negative values are considered as early jerks and positive as late jerks (see scheme in Figure
5.3). The histogram of the differential delays measured in the data of the 1969 jerk (Y component)
is shown in Figure 5.4 where the zero corresponds to the mean delay time. The data differential
delays can vary up to 3 years, but most of the observatories vary in the range of 1.5 year.
7. We compare the differential delays of data (Figure 5.5) with model predictions, by calculat-
ing the misfit value that is the measure of how well the differential delays of the model fit the data
differential delays: v
u N
u1 (obsi modi )2
E= t

N i=1 2obsi
, (5.1)
85
Figure 5.3: Scheme of time delays (1 ) and differential delays (1 ) simulating four different locations. The
differential delays are calculated in relation to the mean time delay (m ) shown by the dashed line, where
the locations 1 and 3 would correspond to later jerks and locations 2 and 4, earlier jerks.
where N is the number of observatories, obsi are the observed differential delays, modi the
model differential delays and obsi the data error bars in obsi .
Figure 5.4: Histogram of the differential delays measured on the annual means data corresponding to the
1969 jerk (Y component).
The Earths mantle was modelled considering Kuvshinov and Olsens (2006) 1-D profile for
the first 700 km and below that depth four model setups were built:
1. One-layer model with 2200 km thickness;
2. Two-layer model with a bottom layer 1250 km thick and top layer with 950 km;
3. Two-layer model with a bottom layer simulating the D (300 km thick) and a top layer (1900
km) simulating the lower mantle;
4. Three-layer model with the lower mantle divided into two layers 950 km thick and D with
thickness of 300 km.
The variation of electrical conductivity in each of these models is described in the next section and
86
Figure 5.5: Results for the 1969 jerk occurrence time at the Earths surface obtained by the L2 method for
Y component of the magnetic field by considering first differences of observatory annual means. The mean
occurrence time is shown close to the vertical bar (from 1 to 6 years) which gives the height of the blue and
red bars. The red bars represent locations where the jerk appeared later than the mean occurrence time and
the blue bars where it appeared earlier. The occurrence time when the bar is red, is given by the sum of the
mean occurrence time and the height of the bar in a specific location; while the occurrence time of blue bars
in a given location is given by subtracting the mean occurrence time by the height of the bar. Dark red (blue)
bars represent locations where the limits of the error bars are later (earlier) than the mean occurrence time
and light red (blue) bars where the limits of the error bar are earlier (later). The green squares represent the
locations where the jerk was not detected and the black squares where data was excluded.
by Tables in Appendix B.
5.3 Results and Discussion

We calculated the misfit values for the one, two and three-layer models, where for all of them we
adopt the 1-D electrical conductivity model of Kuvshinov & Olsen (2006) for the first 700 km. For
each of the models, some examples of IRFs, differential delays for the locations corresponding to
observatories and the misfit, are shown. The differential delay times are calculated in relation to
the mean delay, as was performed in the data analysis. In order to have a basis for comparison
between different values of misfit, we define the reference misfit as:
v
u N
u1 (obsi )2
Ere f = t
2
N i=1
(5.2)
obsi
that is the misfit for an insulating mantle model, where the differential delays are equal to zero
(mod = 0). For the observatories used in the data analysis of the 1969 jerk, the reference misfit
is Ere f = 1.542 considering our spherical harmonic model (Figure 5.2 A) and Ere f = 1.456 using
the model of Le Huy et al. (1998), shown in Figure 5.2 B. This difference is because we exclude
observatory locations with amplitudes in the range 1 nT/yr2 . Since the amplitude isolines are
87
different in our and Le Huy et al.s (1998) model, the excluded observatories will not be the same:
18 locations in Le Huy et al.s model and 8 locations excluded in our model. In order to quan-
tify how better the minimum misfit model is compared to the reference (insulating mantle), we
calculate: " #
Emin 2
E% = 1 100, (5.3)
Ere f
which we call variance reduction in an analogous way to that used in seismology.
5.3.1 One-layer model

We calculated the IRFs for a one-layer model varying the electrical conductivity from 1000 S/m to
0.4 S/m. Some examples of the IRFs are shown in Figure 5.6. If one measures the delay times of
these IRFs curves by using Backus mantle filter theory (see Chapter 2), as a first approximation,
for = 1 we obtain: 1 (1) 10.79 years for the most conducting model (100 S/m), 1 (1) 0.64
years (for the model with 10 S/m) and 1 (1) 0.07 years (for the model with 1 S/m).
But in order to calculate the differential delays, we need to evaluate first the Composite Impulse
Response Functions (CIRFs) for each electrical conductivity model. Some examples of CIRF and
CIRF are shown in Figure 5.7, for a uniform electrical conductivity of 10 S/m and 100 S/m, for
locations corresponding to Hermanus and Vassouras magnetic observatories. Each of the CIRFs
were convolved with an arbitrary jerk simulated at time t = 0 and the result of this convolution is a
delayed and smoothed output jerk that we would observe at the Earths surface. For each electrical
conductivity and each jerk morphology model the output jerk will be different (as shown in the
forward problem, Chapter 4).
Since we want to treat the model in the same way as the data, annual means were calculated
for the output jerk. We selected the same number of data points (7 data points before and after
the peak, N = 15) for the fitting of two straight line segments, by least-squares. The resulting
fitting is shown in Figure 5.8 for Hermanus and Gnangara observatories and assuming an uniform
conducting mantle of 10 S/m and 100 S/m.
Figure 5.6: Impulse Response Functions (IRFs) of a one-layer model simulating the lower mantle for = 1
(red), = 3 (black) and = 5 (blue) and with electrical conductivities of: 1.0 S/m (A); 10.0 S/m (B); and
100.0 S/m (C). These IRFs have been resampled to 1 day intervals.
88
Figure 5.7: Examples of CIRF and CIRF (black curves) calculated for the 1969 jerk considering the spher-
ical harmonic model of the Y component in Chapter 3. We used in this example the electrical conducting
model of Kuvshinov & Olsen (2006) for the first 700 km depth and below that a uniform one-layered model
with 10 S/m (A and B) and with 100 S/m (C). The CIRFs for locations corresponding to Hermanus (A) and
Vassouras (B and C) magnetic observatories are shown.
We calculated the misfit values for each of the electrical conductivity models that are shown in
Table B.1 and Figure 5.9 for the two jerk spherical harmonic models of the 1969 jerk amplitude.
In both models, the misfit values increase rapidly with conductivities larger than 20 S/m and below
that it oscillates in a range smaller than 0.1 of the misfit value (about 1.45 for Le Huy et al.s (1998)
model and about 1.55 by using our model). There are only two acceptable conductivity models
that are in common when using the two spherical harmonic models: = 3 S/m and = 5 S/m. By
Figure 5.8: Examples of annual means and the fitting of two straight line segments to the output data (red
symbols) at Gnangara observatory (GNA) in Australia (A) and Hermanus observatory (HER) in South Africa
(B and C). We used in this example the electrical conducting model of Kuvshinov & Olsen (2006) for the
first 700 km depth and below that a layer of 10 S/m (A and B) and 100 S/m (C).
89
using the model from Le Huy et al. (1998) two more models are also smaller than the reference:
= 1 S/m and = 10 S/m. Consequently, the range of acceptable models in this analysis vary
from 1 S/m 10 S/m.
Figure 5.9: Misfit between the differential

delays of the data and model for the one-
layer mantle electrical conductivity model.
We consider the spherical harmonic model
of the 1969 jerk in Chapter 3 (red symbols)
and that taken from Le Huy et al. (1998)
(blue symbols). The plot in the top left is a
zoom of the yellow area in the bigger plot.
Another way to analyse the results is by looking at the differential delays we would observe
at the surface for each conducting model. We show three examples of differential delays for the
minimum misfit models of Le Huy et al. (1998) (Figure 5.10 A) and our spherical harmonic model
(Figure 5.10 B and C) that are both reasonably similar to the patterns early/late seen in the data
(Figure 5.5), however the amplitude of the model differential delays (about 1 month) are more
than one order of magnitude smaller than in the data (about 1.5 year). The delays observed by
using our amplitude models are in general larger than those using Le Huy et als (1998) model. We
also looked at the patterns of a highly conducting mantle with 200 S/m (Figure 5.10 C) which does
not fit well the patterns early/late observed in the data, but the amplitudes of the differential delays
are in better accordance with the data.
90
Figure 5.10: Examples of differential delays for one-layer electrical conductivity model and a typical distri-
bution of observatories. We used the spherical harmonic model of Le Huy et al. (1998) (A) and the one in
Chapter 3 (B and C) for mantle electrical conductivities of 1 S/m (A), 5 S/m (B) and 200 S/m (C).
91
5.3.2 Two-layer models

In the first two-layer model we divided the lower mantle into two thick layers of 1250 km (bottom)
and 950 km (top) in the low conducting range of 0.457 S/m to 12.346 S/m. The misfit values are
shown in Figure 5.11, considering the two models of jerk amplitudes (see Figure 5.2). Both results
show low misfit values in the range of 3 S/m for the top and bottom of the lower mantle. They
also present a minima in the left top corner, which correspond to low values of conductivity in the
top lower mantle (up to 1 S/m) and high values in the bottom mantle (about 12 S/m that is the
maximum conductivity in this simulation). We increase the electrical conductivity values in the
second simulation of a two-layer model, but considering a thin mantle layer in the bottom (300 km)
that simulates the D . We want to answer the question whether it is possible to have sensitivity to
variations of electrical conductivity in such a thin layer in the bottom of the mantle.
Figure 5.11: Misfit between the differential delays of the data and model considering the 1969 jerk spherical
harmonic models (see Figure 5.2) in Chapter 3 and from Le Huy et al. (1998) for the two-layer model with
the mantle divided into two layers of 1250 km and 950 km.
Sensitivity to D
In the second two-layer model the mantle is divided into a top layer with 1900 km thickness and a
bottom layer 300 km thick simulating the D . The first issue to be analysed is whether there is a
large difference in the IRF if one varies the electrical conductivity of the simulated D . In Figure
5.12 we present the IRFs of three profiles with the following electrical conductivities of the D and
lower mantle: 111.111 S/m and 12.346 S/m (A); 111.111 S/m and 4.115 S/m (B); and 12.346 S/m
and 4.115 S/m (C).
Since this is a non-linear problem, the relationship between the electrical conductivity distri-
bution and delay times does not vary linearly, in contrast to Backus theory (equation 2.71). By the
examples of IRFs, in Figure 5.12, the delay times vary considerably between the different models
indicating that in this kind of regime we would be able to detect conductivity variations in both
the lower mantle and D . The limitation is that those variations are of the order of 0.1 year and
could not be detected from measuring noisy occurrence times in the data. However, the differential
92
Figure 5.12: Impulse Response Functions (IRFs) of a two-layered model with the top layer simulating
the lower mantle (1900 km thick) and the bottom layer simulating D (300 km thick) for = 1 (red),
= 3 (black) and = 5 (blue) and with varying electrical conductivities of: 111.111 S/m and 12.346 S/m
(A); 111.111 S/m and 4.115 S/m (B); and 12.346 S/m and 4.115 S/m (C) for the lower mantle and D ,
respectively.
delay times are also going to depend on the spherical harmonic model for the jerk and the mixing
of harmonics is able to generate larger differential delays.
Figure 5.13: Misfit between the differential delays of the data and model for different mantle electrical
conductivities. We consider the spherical harmonic model of Chapter 3 for the 1969 jerk for the two-layer
model with the mantle divided into two layers of 300 km and 1900 km.
93
We calculated differential delays for models ranging from 111.111 S/m to 0.457 S/m (see Table
B.2) using our spherical harmonic model for the 1969 jerk amplitudes. This result shows that the
misfits (Figure 5.13) increase rapidly for electrical conductivities of the lower mantle larger than
10 S/m and that there are two minima in a conductivity of the lower mantle corresponding to 1
S/m and for the D about 1 S/m and from about 37 S/m to 100 S/m (in the limits of this simulation
range). The most acceptable models are in the range of 1.372 S/m in the lower mantle (see Table
B.2) but there is a high misfit value at a lower mantle with 1.372 S/m and a D with 12.346 S/m. In
order to investigate what caused such a jump in the misfit values in this range, we plotted in Figure
5.14 the differential delays for a lower mantle with 1.372 S/m and D with 37.037 S/m (A), 12.346
S/m (B) and 4.115 S/m (C). The patterns for a D with 37.037 S/m and 4.115 S/m are similar to
the early/late patterns in the data (see Figure 5.5), while when the D has a conductivity of 12.346
S/m, the pattern is basically the opposite, causing the high misfit value (see Table B.2).
In the second calculation of the two-layer model, we increased the range of the D to 1000 S/m
and calculated finer variations of electrical conductivity (from steps of three to two, see Tables B.2
and B.3), as shown in Figure 5.15. This result also suggests a low electrical conductivity for the
lower mantle of about 1 S/m and a broad range of acceptable conductivities for the D . Considering
this misfit calculation, it seems the data is sensitive to variation in the thick lower mantle, but not
sensitive to variations of the D electrical conductivity.
In order to analyse the stability of this inverse problem in terms of jerk amplitudes, we calcu-
lated misfit values considering the two spherical harmonic models in Figure 5.2 for the 1969 jerk
(Tables B.3 and B.4). Considering only one example it is not possible to prove if the inverse prob-
lem is stable or not, but instead we want to address how slightly different results of morphology,
for the same jerk event, can influence the results of differential delay times and consequently misfit
values for the two-layer model. By comparing those results (Figure 5.15 A and B) we conclude
that the patterns of low/high misfit are similar by using any of the models. By using our ampli-
tude model for the same jerk (Figure 5.2 A), 24 out of 72 misfit values are smaller or equal to the
reference while in Le Huy et al.s model only 18 models are acceptable. Both results favoured an
electrical conductivity of the lower mantle of 0.977 S/m and a broad range of variation for the D .
However, considering 0.977 S/m as the minimum misfit value for the lower mantle, both models
have in common small misfit values for a D with 1000 S/m or 125 S/m.
5.3.3 Three-layer models

We also developed models with three layers in order to test whether variations of conductivity in
the lower mantle would cause a significant effect in the results of differential jerk delay times, and
consequently in the misfit values. We simulated a lower mantle divided into two layers of same
thickness (950 km) and D with 300 km of thickness. We calculated also this three-layer model for
our and Le Huy et al.s (1998) models for the 1969 jerk amplitudes (Figures 5.16 and 5.17). Each
of the plots represent a fixed value of the electrical conductivity in D and the misfit is shown as
a function of the conductivities of the two lower mantle layers. The patterns of misfit models are
similar in both cases and there appears a common minimum misfit for high electrical conductivities
of the D (Figure 5.16 B and Figure 5.17 A).
The minimum misfits, considering our amplitude model, were found to be with a D with
111.111 S/m, bottom lower mantle with 0.457 S/m and top lower mantle with 1.372 S/m and
another one with the D and bottom lower mantle with 12.346 S/m and top lower mantle with
0.457 S/m. By using Le Huy et al.s (1998) model, the minimum misfit values are considering a
94
D with 333.333 S/m, bottom lower mantle with 1.372 S/m and top lower mantle with 0.457 S/m
and the second with a D with 0.457 S/m, bottom lower mantle with 12.346 S/m and top lower
mantle with 0.457 S/m. Both minimum misfit results favour a highly conducting D , however there
are also similar misfit values considering some low conducting D which shows that it is difficult
to constrain the D conductivity. If we analyse the intersection of small misfit values by using ours
and Le Huy et als models, we find a consistent range of 12.346 S/m for the D , 0.457 S/m for the
bottom mantle and a variation of 0.457 S/m to 4.115 S/m in the top lower mantle (See Table B.5).
It is interesting to look at the delay patterns (early/late) from those models which provide
smaller misfit values and compare with the data results. Figure 5.18 shows results of the three layer
model with variation of conductivity in the lower mantle for the 1969 jerk. Notice that although
the patterns (late/early jerks) are reasonably in accordance with data results, the amplitudes of
differential delays in the model are in general more than one order of magnitude smaller than the
ones in the data.
5.4 Summary
The inverse problem was solved, by exhaustive search, considering the same methodology devel-
oped in the forward problem: evaluation of IRFs, CIRFs, convolution of the input jerk with the
CIRF, generation of annual means of the output signal and fitting of two-straight line segments to
this synthetic data. Following this methodology, we calculated the differential delay times in the
data and models, from which we evaluated the misfits. The misfit values for all simulations, favour
a low conductivity of the lower mantle of about 1 S/m allowing a broad range of conductivities
for the D . We believe that there is not much sensitivity of our data to detect changes in the D
electrical conductivity.
The patterns early/late jerks in the minimum misfit models were found to be similar to the data
analysis, considering the 1969 jerk. However the amplitude of such differential delays is on the
order of 0.1 year, while in the data it is about 1.5 year. The minimum misfit models favour the two-
layer models with a lower mantle 1900 km thick and a D 300 km thick, which gives a variance
reduction of about 3.3 % when using Le Huy et al.s (1998) model and of 2.3 % when using our
spherical harmonic model.
95
Figure 5.14: Examples of differential delays for two-layer electrical conductivity model and a typical distri-
bution of observatories. All models are simulating two layers below 700 km depth: the lower mantle with
1900 km thick and the D is 300 km thick. We used the spherical harmonic model in Chapter 3 of 1969 jerk.
The three plots correspond to values of misfit all for a lower mantle of 1.372 and a D : A with 37.037 S/m,
B with 12.346 S/m, and C 4.115 S/m.
96
Figure 5.15: Misfit for the two-layer model considering the 1969 jerk spherical harmonic model in Chapter
3 (A) and taken from Le Huy et al. (1998) (B), for a higher D electrical conductivity and finer variations.
97
conductivity models, considering the spherical harmonic model of the 1969 jerk, shown in Chapter 3. This
misfit calculation is relative to the three-layer model with a lower mantle divided into two layers of 950 km
thick and a D of thickness 300 km (see Table B.5). The electrical conductivity of D is fixed for each misfit
plot: in A for 333.33 S/m, in B for 111.11 S/m, in C for 37.04 S/m, in D for 12.35 S/m, in E for 4.11 S/m,
in F for 1.37 S/m and in G for 0.46 S/m.
98
conductivity models, considering Le Huy et al.s (1998) spherical harmonic model of the 1969 jerk. This
misfit calculation is relative to the three-layer model with a lower mantle divided into two layers of 950 km
thick and a D of thickness 300 km (see Table B.5). The electrical conductivity of D is fixed for each misfit
plot: in A for 333.33 S/m, in B for 111.11 S/m, in C for 37.04 S/m, in D for 12.35 S/m, in E for 4.11 S/m,
in F for 1.37 S/m and in G for 0.46 S/m.
99
Figure 5.18: Examples of differential delays for three-layer electrical conductivity model and a typical
distribution of observatories. All models are simulating three layers below 700 km depth: the lower mantle
is divided into two layers of 950 km thick and the D is 300 km thick. We used the spherical harmonic
model of Le Huy et al. (1998) (A) and the model in Chapter 3 (B,C), both for the 1969 jerk. The three plots
correspond to low values of misfit: A and B for a D with 0.457 S/m, bottom lower mantle with 12.346
S/m and top lower mantle with 0.457 S/m and C presents results for a D with 333.333 S/m, bottom lower
mantle with 4.115 S/m and top lower mantle with 12 S/m.
100
Chapter 6
Conclusions
The objective of this thesis is to investigate constraints on the mantles electrical conductivity by
using observations of geomagnetic jerks. The results are divided in three parts: data analysis of ge-
omagnetic jerks in observatory data; the forward problem using a radial mantle conductivity model
and calculating the differential jerk time delays at the surface; and solving the inverse problem in
order to constrain some information about the mantle electrical conductivity.
We used different datasets in order to detect geomagnetic jerks: observatory annual mean data,
observatory monthly mean data and the core contribution calculated from the CM4 model. We
adopted a simple method of fitting two straight line segments to the secular variation by 1-norm
(L1 ) and 2-norm or least-squares (L2 ). We used all datasets and L1 and L2 in order to detect the
1969 jerk. This test showed that the patterns of early/late jerks did not change significantly when
using L1 and L2 and when applied to different datasets. Consequently, we concluded that there is
not much advantage in applying the L1 instead of L2 method and we chose to analyse all the other
jerks by using annual means data since monthly means have more influence of the external field and
the geographical distribution of the dataset is more restricted than for the annual means. We also
preferred to analyse annual means than synthetic data extracted from CM4, which is nevertheless
strongly constrained by magnetic observatory data.
The main advantage of using this simple model of fitting two straight line segments is that we
are able to calculate error bars in the jerk occurrence time and amplitude. The analysis of error bars
in the occurrence time of geomagnetic jerks proved to be essential in order to reliably distinguish
between early and late jerks. In general, the error bars in t0 were smaller in the Y component than
in the X and Z components, probably due to a smaller influence of external field variations; they
are also usually larger for the Southern hemisphere observatories.
The 1969, 1978 and 1991 jerks were detected worldwide in the secular variation estimates
derived from annual means, while the proposed event in 1999 was detected only locally. The
results of the Y component of the 1969 jerk occurrence time showed a more complex pattern with
generally early arrivals in Europe and later arrivals in South America and Africa (both have few
observatories available). Conversely, in the X and Z components a later arrival of the 1969 jerk
appeared in Europe and Africa, while and mostly early arrivals were found in South America. In
the Y component of the 1978 jerk we found a late time occurrence in most of Europe and South
Africa and for the same component of the 1991 jerk we observed an early arrival in Europe and a
late arrival in North America and South East Asia.
Concerning the results of the jerk amplitude, we concluded that the patterns of positive/negative
jerk amplitudes were better defined in the Y component and at most locations presented opposite
101
signs in consecutive jerks. We evaluated the trade-off curve for each geomagnetic jerk and the most
appropriate models for the jerk morphology have damping parameter values between = 105 and
= 104 and power spectra that vanish at harmonic degrees bigger than five (L = 5) for the 1969
and 1978 jerks and bigger than six (L = 6) for the 1991 jerk.
In the forward problem we applied Backus (1983) mantle filter theory and verified that the
mantle filter acts differently on each harmonic degree. The mixing of harmonics varies with loca-
tion, depending on the jerk morphology at the Earths surface and consequently the arrival times
of geomagnetic jerks vary for each location, jerk and component. We used as example a 1D ra-
dial conducting model of the mantle applied to two synthetic models of jerk morphology and to
realistic models of the 1969 and 2003 jerks (Le Huy et al. (1998) and Olsen & Mandea (2007)).
We used two radial mantle conductivity profiles and illustrated that the observed differential time
delays of the order of 2 years seem to suggest a highly electrically conducting mantle.
We concluded, by solving the forward problem that: (1) assuming a simultaneous jerk at the
CMB, a 1D mantle conductivity model is able to generate jerk differential time delays at the Earths
surface; (2) synthetic examples of jerk morphology of the X , Y and Z components demonstrated
that the pattern of time delays would be different. This would explain why jerks are not seen in
observatory data at the same time in different components of the magnetic field; (3) the application
of this theory to the 1969 and 2003 jerk morphologies showed that it would also be a natural
consequence to have different time delays for each jerk.
In the inverse problem the aim is to obtain some constraints on mantle electrical conductivity
models based on the observation of differential time delays at the Earths surface, obtained in the
data analysis. We simulated mantle electrical conductivity models of two layers (upper mantle
1900 km thick and with a D of 300 km) and different configurations for the three-layer model:
a lower mantle divided in two layers of 950 km thick and a D 300 km thick; or a unique layer
1900 km thick for the lower mantle and two layers in the D : a bottom layer of 100 km thick and
top layer 200 km thick. We varied the electrical conductivity in steps of 3 from 333.33 S/m to
0.05 S/m in the two-layer model and from 111.11 S/m to 0.05 S/m in the three-layer model. The
results showed that the IRFs and differential time delays are sensitive to variations of electrical
conductivity of the lower mantle and D in a non-linear relationship.
In order to test how slightly different spherical harmonic models for the same jerk would af-
fect the misfit results, we calculated the 2-layer simulation for the 1969 jerk, by using the model
evaluated in Chapter 3 and Le Huys et al. [1998] model. We concluded that the limits where the
misfits are smaller agree in the two cases. We calculated the misfit value by considering an insu-
lating mantle which is called reference misfit, which is Ere f = 1.542 when using our model of jerk
amplitudes (Chapter 3) and Ere f = 1.456 when using Le Huy et al.s (1998) model. We compared
the misfit values for the models of electrical conducting mantle.
The results of the one-layer simulation point to a range of conductivities from 1 S/m to 10
S/m for the lower mantle in agreement with Olsen (1999) and Mandea Alexandrescu et al. (1999).
In the two-layer model considering a bottom layer 1250 km thick and top layer with 950 km the
preferred values are in the range of 0.5 S/m to 3 S/m for both layers. We tested whether our data
would be sensitive to variations in the conductivity of the D , but for both two and three-layer
models it allowed a broad range. However, in the three-layer model the results favour mainly an
electrical conductivity from 12.346 S/m to 333.333 S/m (limit in this simulation).
The minimum misfit models agree relatively well with the jerk pattern of early/late differential
time delays but not with the amplitude of those delays. For example, the minimum misfit model
was found to have delays on the order of 0.5-1 year while in the data these are of the order of
102
1.5 years. In addition, the variance reduction was found to be about 3 %, which may not be a
significant improvement from the insulating model. Therefore, one may consider whether there
are other physical factors not captured by our model that would influence the differential time
arrivals of geomagnetic jerks at the Earths surface.
6.1 Future Work

The 2003 and possibly the 1999 geomagnetic jerks would require more temporal resolution to be
characterized since they have a shorter duration compared to the previous ones. For this reason, I
aim to deduce error bars in the monthly means data making it possible to analyse recent satellite
data, which provide global coverage. For the inverse problem there are some topics I wish to
develop as future work: the first is to calculate misfit values for other configurations of mantle
electrical conductivity models, varying the layers thicknesses. The second topic is calculate a
joint misfit considering all the jerks and components of the magnetic field (1978, 1991 and 2003
jerks and X and Z components). The last subject I wish to explore is the effect of a 3D mantle
electrical conductivity models on the jerk differential delay times. At the moment, there are no
3D candidate models for the mantle electrical conductivity. However, with increasing efforts from
many research areas, such as seismology and mineral physics, I believe that in the near future there
will be some suggestions of 3D models of mantle electrical conductivity, especially in the D .
Therefore, it will be interesting to analyse 3D candidate models and evaluate their effects on jerk
differential delay times.
103
Appendix A
Data analysis tables

104
Table A.1: Occurrence date () of the late 1960s geomagnetic jerk in the annual means of
the X , Y and Z components of each magnetic observatory, detected by fitting two straight-
line segments to data in the least-squares sense. The symbols in this table are: observa-
tories where the jerk was not detected, observatories excluded when the minimum of the
misfit curve is in one of the extremities, and when there was not enough data to perform
the analysis. The left and right limits of 67 % confidence are given by e and e+ .
OBS X Y Z
e e+ e e+ e e+
AAA 1971.00 -0.37 0.21
ABG 1966.67 -0.51 0.88 1972.59 -1.34 0.81
ABK 1970.78 -0.83 0.66
AIA 1966.43 -1.28 4.70 1970.00 -0.72 2.19
ALE * * 1973.00 -6.35 0.24
ALM 1972.00 -4.08 1.96 1968.72 -0.72 1.55
AML *
ANN 1970.38 -1.95 1.07
API 1967.00 -1.38 5.30 1970.21 -1.95 1.39 1973.00 -0.31 0.55
AQU 1971.28 -2.00 2.50 1969.38 -0.60 0.61
BEL * *
BJI 1966.71 -0.66 0.57 1970.28 -0.35 0.46 1969.29 -0.45 0.46
BJN 1972.35 -5.55 0.80 1970.00 -0.37 0.42 1973.66 -1.37 1.30
BLC 1970.33 -1.64 3.12 1970.63 -1.47 1.68 1969.65 -0.84 0.83
BNG 1973.65 -1.98 0.98 1972.00 -0.98 2.85 1972.00 -1.36 1.92
BOU 1970.24 -0.66 0.65 1970.69 -0.67 0.45 *
BRW 1970.00 -1.12 1.07 *
CCS 1969.61 -0.56 0.55
CLF 1971.00 -1.67 3.53 1969.24 -0.90 0.77 1972.59 -2.01 1.37
CNH 1966.68 -1.03 1.00 1969.59 -0.70 0.67
COI 1970.00 -1.08 0.91 1969.79 -0.41 0.56 1968.88 -2.42 1.34
CPA 1968.65 -2.01 3.48 1970.76 -1.39 2.47
CWE 1970.75 -3.26 3.41 1969.00 -0.61 5.56 1968.00 -0.71 3.83
DAL *
DIK 1970.00 -1.43 1.20 1969.18 -1.79 1.89
DOB 1968.54 -0.49 0.60 1972.48 -0.76 1.76
DOU 1971.00 -2.05 2.57 1969.55 -0.41 0.42 1973.63 -1.40 0.88
DRV 1970.00 -1.33 1.96
EBR 1971.00 -2.46 2.92 1969.32 -0.45 0.35 1973.00 -4.13 0.71
ESK 1971.00 -1.01 3.59 1969.43 -0.28 0.28 1972.39 -1.45 1.40
FCC 1968.53 -0.39 1.30 *
FRD 1970.16 -0.36 0.54 1967.27 -0.71 0.73 1967.23 -0.72 0.53
FUQ 1969.00 -1.20 1.28
FUR 1971.00 -1.31 1.76 1969.74 -0.26 0.27 1974.00 -0.84 0.72
GCK 1971.00 -1.79 2.44 1969.52 -0.27 0.28
GNA 1974.00 -1.99 0.97 1971.56 -0.46 0.42 1970.97 -0.42 0.28
GUA 1972.31 -1.05 1.44 1966.21 -0.98 0.84 1971.06 -0.79 2.30
105
GZH 1969.00 -1.92 1.40 1970.58 -2.35 3.40 1972.00 -1.73 0.97
HAD 1971.00 -1.58 2.23 1969.62 -0.29 0.30 1970.18 -0.64 2.29
HER 1972.17 -1.66 1.31 1972.44 -0.58 0.63 1966.00 -0.56 0.99
HIS 1970.00 -0.81 1.58
HLP 1971.00 -2.09 1.84 1970.39 -0.96 0.48
HRB 1966.48 -0.96 1.03 1969.30 -1.12 1.17
HUA 1967.82 -1.39 4.17 1973.00 -0.53 0.77 1969.44 -0.41 0.42
IRK 1971.71 -1.06 0.47 1970.27 -0.63 0.50
ISK 1969.60 -0.72 0.75 1971.00 -2.42 1.31
KAK 1967.00 -0.79 0.66 1970.49 -0.72 0.63 1969.00 -0.55 0.44
KGD 1971.00 -1.15 1.13
KIV 1971.00 -0.62 0.79 1969.48 -0.33 0.35 *
KNY 1966.82 -0.54 0.55 1971.00 -0.68 0.62 1969.10 -0.70 0.98
KNZ 1966.74 -0.60 0.63 1970.00 -1.03 1.32 1968.86 -0.52 0.39
KOD 1970.46 -1.19 1.51 1972.20 -0.59 1.10
LAS 1969.00 -1.19 5.27
LER 1969.32 -0.40 0.42 1972.21 -2.40 0.76
LGR 1971.00 -1.73 0.79 *
LMM 1972.13 -1.09 1.67 1972.18 -1.26 1.43 1967.46 -2.24 3.20
LNN 1971.50 -0.98 2.47 1969.80 -0.35 0.27
LOV 1971.13 -0.86 2.19 1969.78 -0.29 0.31 1974.00 -0.74 0.78
LQA 1970.00 -2.29 2.91 1972.64 -2.89 1.50 *
LRV 1967.00 -1.54 4.09 1969.48 -0.40 0.40 1970.19 -0.76 1.35
LSA *
LUA 1973.17 -3.19 1.23
LVV 1970.21 -0.56 0.33
LZH 1966.00 -0.37 0.44 1971.79 -0.44 0.58 1970.13 -0.79 0.72
MAW 1971.00 -0.80 1.92 1968.54 -0.95 0.91 1971.65 -2.64 3.19
MBC 1970.00 -0.46 0.47
MBO 1966.34 -0.42 1.55
MCQ 1965.87 -0.58 0.68 1967.00 -1.42 5.19
MEA 1969.69 -0.47 0.49 1970.00 -0.86 0.78
MIR 1971.00 -5.01 0.97 1971.00 -1.52 3.73
MLT 1971.63 -4.38 1.71
MMB 1966.82 -0.96 0.65 1969.55 -0.63 0.59 1969.54 -0.35 0.36
MMK 1970.00 -0.90 1.49 1970.00 -1.04 0.78 1970.09 -2.76 2.49
MNK 1971.00 -1.39 1.98 1969.46 -0.71 0.97
MOL
MOS 1971.50 -0.84 1.86 1970.22 -0.29 0.38
MUT 1966.25 -0.72 1.32 1966.00 -0.36 0.70
NCK 1966.11 -0.65 0.49
NGK 1971.00 -1.29 2.65 1969.73 -0.28 0.44 1974.00 -1.11 0.95
NUR 1971.03 -0.92 2.26 1969.58 -0.20 0.20
NVL 1971.00 -3.86 2.77
ODE 1971.55 -1.45 3.16 1970.08 -0.55 0.41 1966.90 -0.95 0.96
PAB *
106
PAF 1966.54 -0.48 0.71 1971.73 -0.25 0.24

PAG 1970.23 -0.62 0.46 1969.00 -1.94 1.16
PIL 1972.88 -2.15 1.61
PMG 1967.00 -0.87 4.62 1970.11 -1.60 0.94 1970.29 -2.76 2.36
QUE *
RES 1969.54 -2.38 1.50 1970.49 -0.50 0.48 1973.50 -0.57 0.53
RSV 1971.20 -0.81 1.81 1969.64 -0.39 0.48
SAB *
SBA 1969.31 -1.13 1.25 *
SFS 1967.13 -1.52 1.94
SJG 1969.00 -0.75 0.30 1971.40 -1.01 0.58 1971.00 -1.97 2.00
SMG 1970.90 -1.91 0.47 *
SOD 1969.59 -0.29 0.30
SSH 1966.49 -0.34 0.40 1970.99 -1.29 0.61 1969.82 -3.16 1.52
SSO 1967.00 -0.70 0.95 1971.00 -0.96 1.02 1970.00 -2.14 0.86
SUA 1969.00 -0.77 1.43
SVD 1971.00 -1.31 0.78 1969.86 -0.37 0.23 *
TAM 1971.00 -2.72 3.60 1970.00 -3.62 2.97 1972.74 -2.23 1.58
TAN *
TEO
TFS 1972.61 -0.65 1.17 1971.00 -1.80 0.82
THL 1969.81 -0.35 0.64 1966.00 -0.97 5.68 1973.56 -1.32 0.67
THY 1972.60 -1.78 0.89
TKT 1970.32 -0.70 0.42 *
TNG *
TOL 1971.00 -2.42 3.43 1969.22 -0.63 0.76 1974.00 -3.98 0.55
TOO 1970.00 -0.48 0.88 1970.00 -3.80 1.83
TRD 1972.20 -0.65 0.77
TRO 1971.37 -1.19 1.47 1969.37 -0.45 0.53
TRW 1973.70 -4.96 1.02
TSU 1972.00 -0.52 0.48
TTB * *
VAL 1971.00 -1.40 2.65 1969.63 -0.23 0.23 1970.24 -0.83 1.26
VIC 1971.00 -2.47 2.23 1971.35 -0.70 0.35 1970.01 -0.92 2.17
VLA 1966.00 -0.35 0.59 1968.59 -0.76 0.69 1968.51 -1.73 1.54
VOS *
VSS 1971.00 -2.66 0.63
WHN 1969.00 -2.29 1.59
WIK 1971.00 -2.44 1.89 1969.49 -0.29 0.30
WIT 1971.00 -1.66 2.28 1969.77 -0.23 0.23 1974.00 -1.53 0.86
WNG 1971.00 -0.92 2.55 1969.72 -0.23 0.23 1973.75 -1.12 0.74
YAK 1967.29 -1.19 1.46 1969.11 -1.04 1.16
YSS 1972.66 -1.96 1.21 1967.71 -1.17 2.14
107
OBS X Y Z
e e+ e e+ e e+
AAA * 1974.00 -0.83 5.46 1974.32 -0.91 0.53
ABG 1979.00 -4.79 1.56 1976.25 -0.47 2.39
ABK 1978.00 -0.24 0.40 1974.26 -0.80 2.41
AIA 1978.00 -0.72 1.92 1977.82 -1.37 0.85 *
ALE * *
ALM * 1977.56 -1.18 1.02
ANN
API 1977.00 -2.88 3.37 1976.95 -0.57 0.51 1978.29 -0.60 0.59
AQU 1982.00 -3.66 0.84
ARS 1978.00 -1.25 0.92
ASH 1978.08 -0.55 0.72
BDV 1978.00 -0.25 0.59
BEL 1980.59 -1.52 0.71
BJI 1977.51 -1.10 0.60
BJN 1978.39 -0.48 0.54
BLC 1975.48 -1.32 3.84 1978.00 -0.58 0.74 1977.00 -0.86 2.59
BNG 1975.08 -1.31 4.63 1976.48 -1.71 0.86 1980.80 -5.96 0.55
BOU 1978.00 -0.88 3.09 1977.34 -0.52 0.39 1976.16 -0.99 0.88
BRW 1975.00 -1.98 4.40
CBB 1976.53 -0.45 0.79
CCS 1974.19 -1.02 2.66
CLF 1979.65 -0.37 0.35 1980.00 -4.60 1.00
COI 1977.51 -0.91 0.72 1979.00 -4.10 2.53
CWE 1980.00 -2.56 0.67
DIK 1974.00 -0.63 0.74
DOB 1978.00 -0.41 0.31 1974.50 -1.45 1.51
DOU 1978.29 -0.48 0.48
DRV
ESK 1978.00 -0.44 0.32
FCC * 1977.57 -2.34 0.74 1977.00 -1.29 2.24
FRD * 1978.61 -1.48 0.76 1975.79 -0.46 0.65
FUQ * 1976.54 -1.42 2.49
FUR 1978.27 -0.42 0.51
GCK 1978.40 -0.54 0.68
GNA * 1981.53 -2.82 1.45 1976.43 -0.58 0.66
GUA * 1979.32 -2.37 1.09 1978.13 -0.41 0.23
GWC 1977.00 -0.71 1.88 1977.00 -0.96 1.67 1979.00 -1.37 0.93
108
GZH *
HAD 1977.94 -0.36 0.28
HBK *
HER 1981.98 -0.36 0.47 1982.00 -0.94 0.84
HIS * *
HLP 1979.94 -1.67 0.95
HRB 1978.10 -0.35 0.56
HUA * 1975.69 -0.81 1.44
HYB 1974.00 -0.87 5.69 1979.55 -1.80 0.58
IRK 1974.61 -1.53 2.98 1974.13 -0.83 1.57
ISK * 1978.31 -1.13 1.59
KAK * 1973.83 -0.75 2.55 1979.00 -3.18 1.56
KGD
KIR 1981.00 -3.24 1.65 1977.50 -0.65 0.76 1977.00 -2.11 0.94
KIV * 1978.00 -0.43 0.87
KNY * 1978.00 -0.54 0.47 1979.05 -1.31 1.34
KNZ * 1979.36 -2.93 1.33
KOD 1975.70 -1.23 0.83
LAS 1976.00 -2.81 3.57
LER 1978.00 -0.38 0.43
LMM 1981.72 -0.46 0.48 1976.16 -2.92 1.43
LNN 1978.27 -0.38 0.48 1974.41 -1.29 2.40
LNP * 1981.00 -2.05 0.45
LOV 1978.00 -0.16 0.35
LQA 1979.00 -0.69 1.13 *
LRV * 1977.89 -0.47 0.25 1980.00 -2.80 1.98
LUA 1976.00 -2.92 2.54
LVV 1976.33 -0.73 0.64 1974.61 -0.81 0.59
LZH 1979.00 -0.50 0.45 1975.40 -1.04 1.25
MAW 1980.91 -4.16 1.64
MBC * 1974.13 -0.92 3.67
MBO 1979.28 -1.37 1.34
MCQ 1981.00 -5.27 1.13 1977.39 -2.92 2.50
MEA 1974.39 -1.04 1.10
MGD 1978.00 -0.87 0.29 1975.00 -0.73 0.61
MIR
MIZ * 1975.00 -1.39 4.60
MLT *
MMB *
MMK 1980.30 -1.26 1.03 1978.00 -3.86 2.12
MNK 1975.22 -0.90 1.14 1976.44 -2.09 1.80
MOL
MOS * 1978.00 -1.78 1.51 1977.92 -3.40 0.58
MUT * 1978.60 -2.24 1.74 1979.00 -3.19 2.56
NAL 1978.15 -2.20 1.56 1974.00 -0.87 3.29
NCK * 1978.37 -1.65 0.70
109
NEW *
NGK 1978.16 -0.26 0.31
NUR 1977.92 -0.69 0.65 1974.24 -1.14 1.74
NVL 1979.38 -4.99 1.46
NVS 1979.53 -3.59 1.58 1976.45 -1.75 1.29
ODE 1978.19 -0.76 0.70
OTT * 1979.82 -0.88 0.41 1976.28 -0.40 0.40
PAF 1976.00 -1.10 0.88 1977.94 -3.48 0.64
PAG 1979.51 -0.80 1.44
PIL 1979.00 -0.68 0.44
PMG 1982.00 -4.64 0.47 1978.27 -0.82 0.81 1976.00 -0.83 2.35
PPT * 1977.40 -0.50 0.51
RES 1979.00 -0.67 0.64
SAB 1980.82 -2.04 1.07 1975.84 -1.61 0.60
SBA 1974.97 -1.59 4.16
SJG * 1979.49 -1.43 1.01 1981.20 -4.21 1.07
SOD 1978.00 -0.54 0.36 1974.57 -1.08 1.49
SSH 1977.95 -2.19 1.60 1978.79 -3.47 1.57
STJ 1976.00 -0.93 1.97 1978.65 -1.61 1.05
SUA 1981.44 -3.14 0.96 1978.03 -3.59 1.93
TAM
TEN 1978.00 -0.83 1.53 1975.34 -0.68 4.68
TEO
TFS * 1978.23 -0.70 2.02
THL * 1980.25 -0.43 0.95 1980.00 -1.60 1.57
THY 1977.00 -1.72 4.21 1977.84 -0.49 0.50
TIK 1978.00 -3.00 1.06 *
TKT * 1975.72 -0.48 0.91
TNG 1979.00 -0.96 3.65
TRD * 1978.00 -2.59 2.77 1979.00 -0.95 1.04
TRO 1978.00 -0.67 0.87
TRW
TSU 1981.77 -0.50 0.63 1981.00 -1.85 1.82
TTB * *
VAL 1977.92 -0.23 0.30 1979.62 -2.93 1.21
VIC * 1977.68 -0.33 0.51 1982.00 -1.28 0.73
VLA 1980.00 -5.29 0.91 1977.00 -2.42 3.48
VOS *
VSS 1978.60 -3.06 1.01
WHN * 1975.15 -0.73 0.78
WIK 1978.17 -0.37 0.48
WIT 1978.27 -0.31 0.31
WNG 1978.00 -0.16 0.28
YAK 1975.00 -1.01 0.45 1975.97 -1.24 1.12
YSS
110
Table A.3: Occurrence date () of the early 1990s geomagnetic jerk in the annual means of
OBS X Y Z
e e+ e e+ e e+
AAA 1992.00 -0.07 0.24 1990.06 -0.16 0.42
ABG 1992.21 -0.24 0.45 1994.98 -0.88 0.17 1990.76 -0.12 0.12
AIA 1991.68 -1.26 0.86
ALE * *
AMS 1988.74 -0.25 0.26 1988.47 -0.69 0.66 1994.85 -0.30 0.24
API 1989.00 -0.19 0.15
AQU 1990.07 -0.54 0.88
ARC 1990.53 -1.03 0.64
ARS 1992.40 -0.47 0.58
ASH 1995.00 -0.03 0.11 1989.30 -0.95 0.37 1988.70 -0.71 0.75
BDV 1990.00 -0.71 0.91
BEL 1990.35 -0.76 0.89
BFE 1990.80 -0.68 0.88
BJI 1991.40 -0.47 0.52 1993.46 -0.30 0.27
BJN 1988.00 -1.07 0.15 1991.00 -0.95 1.65
BLC 1992.10 -0.43 0.47
BNG 1991.00 -0.37 0.07
BOU 1988.88 -0.49 0.23 1993.00 -0.42 0.56 1991.00 -0.37 0.37
BOX 1990.00 -1.74 0.59 1993.00 -0.64 0.61
BRW 1991.00 -0.73 0.86 1992.00 -2.32 2.71
CBB
CCS 1993.00 -0.37 0.89 1990.00 -0.33 0.66
CDP 1995.59 -0.26 0.22 1991.29 -0.42 0.29 1988.00 -1.09 0.67
CLF 1991.95 -0.92 0.50
CNB 1988.00 -0.31 0.64 1988.26 -0.51 0.77
COI 1989.00 -0.31 0.18 1990.00 -0.14 0.16
CSY
CTA
CTS 1990.93 -1.30 0.22
CWE 1990.00 -0.22 0.08
CZT 1988.00 -0.15 0.44 1994.00 -2.82 0.96 1995.30 -0.53 0.26
DIK 1992.18 -0.26 0.18
DLR 1988.00 -0.09 0.80 1993.20 -0.82 1.77 1991.62 -0.32 0.31
DOB 1988.00 -1.50 0.99 1991.04 -0.70 0.70
DOU 1991.19 -0.60 0.77
DRV 1991.56 -0.88 1.13 1987.00 -0.62 3.97
DVS 1988.01 -0.10 0.61
111
ESK 1988.00 -0.38 0.94 1991.85 -0.86 0.62

ETT 1992.00 -2.81 2.91 1989.27 -0.88 1.22 1990.71 -0.11 0.11
EYR
FCC 1992.00 -0.35 0.42
FRD 1988.37 -0.38 0.43 1992.48 -0.95 0.96 1991.19 -0.27 0.29
FRN 1988.00 -0.14 0.64 1994.39 -0.87 0.55 1989.00 -0.28 0.67
FUQ 1992.00 -0.08 0.10 1992.13 -0.31 0.64
FUR 1990.54 -0.68 0.86
GCK 1990.04 -0.68 0.84
GDH 1993.00 -0.30 0.65 1992.00 -1.06 3.95 1989.00 -0.64 1.87
GLN 1990.81 -1.05 0.45
GNA * 1991.41 -0.42 0.42
GUA 1987.43 -0.48 0.58
GZH 1992.20 -0.48 0.20 1991.00 -0.60 0.50
HAD 1988.00 -0.57 0.95 1991.90 -0.90 0.51
HBK 1986.39 -0.38 0.54
HER 1989.00 -0.57 0.14
HLP 1990.92 -0.78 0.21
HRB 1990.83 -0.71 0.56
HRN 1990.09 -0.74 0.71
HTY 1994.08 -0.26 0.40
HYB 1992.00 -0.12 0.39 1991.44 -0.11 0.11
IRK 1989.00 -0.54 1.80 1989.66 -0.34 0.82
ISK 1990.06 -0.51 0.52
KAK 1994.00 -0.29 0.28
KIR 1992.00 -0.73 0.90 1993.00 -0.40 0.50
KIV 1990.00 -0.92 0.19
KNY 1994.00 -2.37 1.38 1994.02 -0.20 0.37
KNZ 1994.00 -0.21 0.38
KOD 1990.00 -0.17 0.16
KRC
LAS
LER 1991.65 -1.09 0.77
LNN 1990.00 -1.46 0.65
LNP 1992.00 -0.48 1.13 1992.00 -0.21 0.40
LOV 1990.51 -0.63 0.98
LRV 1989.31 -0.77 1.27 1991.79 -1.07 0.85 1988.07 -0.22 0.78
LVV 1987.68 -1.27 2.02 1993.00 -0.66 2.11
LZH 1990.56 -0.36 0.37
MAB 1990.17 -0.54 1.59
MAW 1993.00 -0.30 0.21
MBC
MBO 1986.43 -0.40 0.89
MCQ 1988.00 -0.74 0.80 1990.00 -0.89 3.53 1992.00 -0.50 0.74
MEA 1988.71 -0.43 0.41 1995.41 -0.57 0.42
MGD 1994.00 -0.88 1.64 1991.00 -0.28 0.11
112
MIR
MIZ 1994.00 -0.32 0.28
MMB 1989.00 -0.91 0.50 1993.73 -0.46 0.36
MNK 1990.25 -0.45 1.34
MOL 1989.91 -0.14 0.09 1990.00 -0.56 0.23
MOS 1989.25 -0.41 0.73 1992.00 -0.38 0.19
MZL 1991.37 -0.75 0.56
NAL
NAQ 1991.52 -0.73 1.07 1989.00 -0.72 0.12
NCK 1991.00 -0.12 0.29
NEW 1989.00 -0.55 0.29 1994.71 -0.94 0.91 1989.00 -0.82 0.45
NGK 1990.49 -0.64 1.02
NUR 1990.02 -0.51 0.97
NVS 1994.00 -0.53 0.40 1991.00 -0.85 0.90
ODE 1989.57 -1.70 0.54
OTT 1988.75 -0.38 0.31 1993.00 -1.26 0.76 1992.55 -0.55 0.56
PAF 1988.15 -0.33 0.51 1988.00 -0.47 1.18 1993.67 -0.37 0.64
PAG 1990.00 -0.99 0.86
PBQ 1991.00 -0.16 0.31 1992.00 -0.73 0.21
PIL
PPT 1989.02 -0.41 0.91
QGZ 1988.58 -0.33 0.38 1992.51 -0.27 0.26 1990.19 -0.65 1.03
QIX 1991.61 -0.36 0.35
QUE 1992.00 -0.08 0.13 1993.00 -0.03 0.06
QZH 1993.85 -0.33 0.20 1991.89 -0.51 0.49
RES 1992.53 -0.89 1.05
SAB 1993.26 -0.50 0.83 1988.00 -0.73 1.84
SBA 1988.00 -1.88 0.48 1990.70 -0.94 1.25
SHL
SJG 1995.66 -0.28 0.23 1992.44 -0.87 0.74 1988.00 -0.36 0.41
SOD 1990.27 -0.73 1.06 1993.26 -0.54 1.68
SPT 1988.00 -0.35 0.61 1991.58 -1.51 0.68
SSH 1993.35 -0.68 0.46 1992.35 -0.31 0.29
STJ 1989.00 -0.43 0.41 1989.67 -0.43 0.37
SUA 1995.60 -0.49 0.35
TEO 1991.00 -0.59 0.59 1994.39 -0.72 0.38 1994.00 -0.03 0.07
TFS 1995.68 -0.33 0.26 1990.00 -0.40 0.68 1992.30 -0.63 0.52
THJ 1990.60 -0.25 0.26
THL 1991.63 -1.06 1.34
THY 1991.00 -0.81 0.74 1995.00 -0.18 0.19
TKT 1995.00 -1.64 0.50 1989.59 -0.90 1.36
TNB 1990.50 -1.19 0.11
TNG 1989.58 -0.29 0.31 1995.00 -0.16 0.01 1991.00 -0.05 0.01
TRD 1988.71 -0.29 0.29 1991.00 -0.03 0.10
TRO 1989.66 -0.70 0.90
TRW 1988.00 -0.23 0.20 1988.43 -0.54 0.72
113
TTB
UJJ 1992.95 -0.40 0.16 1994.40 -0.44 0.51 1988.96 -0.13 0.17
VAL 1992.00 -0.75 1.08 1988.00 -0.46 0.68
VIC 1988.55 -0.42 0.46 1993.65 -0.85 1.07
VLA 1992.37 -0.48 0.34
VSS 1991.00 -0.26 0.18
WHN 1992.46 -0.39 0.37 1994.40 -1.18 0.37
WIK 1990.52 -0.85 0.82
WMQ 1992.65 -0.40 0.38 1991.75 -0.47 0.50 *
WNG 1990.83 -0.66 0.87
YAK 1991.80 -1.03 0.73
YKC 1989.46 -0.51 0.83
114
OBS X Y Z
e e+ e e+ e e+
ABG
ABK
AIA 2000.00 -2.00 0.44
ALE
AMS 1999.26 -0.76 0.81
API 1999.00 -0.43 1.24
AQU
ARS
ASH
ASP
BDV
BEL 1999.69 -1.87 0.47
BFE 1998.00 -0.28 2.00
BJN
BLC
BOU
BRW
BSL 2000.00 -1.75 0.59
CBB
CBI
CLF 1998.00 -0.26 1.55 2000.00 -1.30 0.72
CNB
COI 2000.00 -0.29 0.61 2000.00 -0.87 0.29
CSY
CTA
CZT 1998.91 -0.77 0.80
DLR
DOB 1999.61 -0.87 0.53
DOU 1998.21 -0.35 1.60 2000.00 -1.58 0.52
DRV
ESA 1998.00 -0.23 0.77
ESK 1998.50 -0.57 1.48 2000.00 -0.52 0.68
EYR 2000.00 -1.63 0.66
FCC 1998.00 -0.63 0.45
FRD
FRN
FUR
115
GDH
GNA 1999.00 -0.57 0.54
GUA 2000.00 -2.03 0.43 1998.34 -0.26 0.23
GUI
HAD 1998.45 -0.58 1.38
HBK 1998.00 -0.21 0.35
HER
HLP 1999.00 -1.22 1.14
HRB 1999.00 -1.24 1.35 1999.00 -1.86 0.20
HRN
HTY 1998.26 -0.35 0.51
IRK
KAK 2000.00 -0.24 0.32
KIR 2000.39 -0.36 0.31
KNY
KNZ
LER 1998.00 -0.85 1.36 2000.00 -1.72 0.48
LOV
LRM 2000.40 -1.05 0.32
LRV 1998.00 -0.62 1.27 2000.00 -0.60 0.64
LZH
MAB 1998.00 -0.37 1.40 1999.91 -1.42 0.69
MAW 2000.57 -0.66 0.38
MCQ 1998.00 -0.64 0.62
MEA 1998.00 -0.82 0.65
MIR
MIZ 1998.00 -0.15 1.15
MMB 1998.00 -0.39 0.24
MOS
NAQ
NCK
NEW
NGK
NUR 1998.00 -0.97 0.51 1998.12 -0.47 1.37
NVS
ODE
OTT
OUL 1998.59 -0.77 1.25
PAF
PAG 1998.00 -0.38 0.82 1998.00 -0.77 0.89
PBQ
PPT
QUE
RES
SBA
SJG 1999.00 -1.46 0.99
116
SOD 1999.92 -0.63 0.63

SPT
STJ
SUA
TAM
TAN
TFS
THL
THY
TKT * *
TRO
TSU 1998.00 -0.61 0.14
TTB 2000.00 -2.11 0.45
VAL 1998.38 -0.49 1.58 1998.00 -0.46 0.86
VIC 1998.00 -0.58 1.48
VNA
WNG
YKC
Annual means
Jerk no of OBS included not detected excluded error on date (yr) error on amplitude (nT /yr2 )
X Y Z X Y Z X Y Z X Y Z X Y Z
1970 78 90 69 31 18 19 14 15 19 -1.46 -0.89 -1.50 -0.83 -0.42 -0.97
1.86 0.95 1.29 0.83 0.41 0.87
1978 43 93 66 15 12 24 67 20 35 -1.66 -1.17 -1.75 -1.13 -0.52 -1.24
1.73 1.09 1.57 1.34 0.44 1.05
117
1991 43 90 59 73 32 58 23 17 23 -1.45 -1.07 -1.15 -1.45 -0.43 -0.94
1.89 1.05 1.35 0.91 0.42 0.87
1999 21 9 14 70 73 77 9 36 10 -0.99 -1.22 -1.30 -0.42 -0.58 -0.52
1.40 0.76 1.00 0.40 0.55 0.62
Table A.5: Number of observatories included, excluded and not detected in the data analysis of each geomagnetic jerk for the annual mean datasets. The
mean error bars upper limit (first line for each jerk) and lower limit (second line) of the date occurrence and amplitude for each geomagnetic jerk are also
given.
118
Appendix B
Tables with misfit values obtained in the

inversion
119
Table B.1: Misfit values for the electrical conductivity models used for a one-layer model
simulation considering the 1969 jerk morphology model of Chapter 3 (misfit 1) and from
Le Huy et al. (1998) (misfit 2). The misfit values in red are all smaller or equal to the
reference misfit (1.542 for our model and 1.456 for Le Huy et al.s model), while the other
misfit values (in black) are larger than the reference.
mantle misfit 1 misfit 2

0.400 1.547 1.460
0.500 1.547 1.461
0.600 1.547 1.462
0.700 1.548 1.462
0.800 1.548 1.463
0.900 1.544 1.463
1.000 1.543 1.446
2.000 1.552 1.466
3.000 1.541 1.452
4.000 1.556 1.472
5.000 1.536 1.450
6.000 1.556 1.465
7.000 1.546 1.467
8.000 1.554 1.469
9.000 1.561 1.465
10.000 1.547 1.448
20.000 1.554 1.476
30.000 1.588 1.521
40.000 1.595 1.520
50.000 1.673 1.530
60.000 1.724 1.564
70.000 1.731 1.629
80.000 1.762 1.575
90.000 1.774 1.626
100.000 1.772 1.606
200.000 3.096 2.346
300.000 2.919 5.229
400.000 4.279 14.938
500.000 6.521 16.659
600.000 8.015 15.859
700.000 9.882 19.205
800.000 10.690 22.865
900.000 11.068 26.190
1000.000 14.105 28.595
120
111.111 37.037 12.346 4.115 1.372 0.457

111.111 1.751 1.619 1.541 1.548 1.535 1.549
37.037 1.769 1.594 1.561 1.546 1.537 1.547
12.345 1.794 1.593 1.555 1.545 1.550 1.542
4.115 1.792 1.588 1.553 1.554 1.539 1.547
1.372 1.793 1.589 1.552 1.555 1.538 1.547
0.457 1.781 1.588 1.554 1.552 1.538 1.547
Table B.2: Misfit values for the electrical conductivity models used for a two-layer model simulation con-
sidering the 1969 jerk morphology model of Chapter 3: the first line (in blue) shows the variation of con-
ductivity for the lower mantle (1900 km thick), the first column (in green) the values for the D (300 km
thick). The misfit values in red are all smaller than the reference misfit (1.542), while the other misfit values
(in black) are larger than the reference.
15.625 7.812 3.906 1.953 0.977 0.488

1000.000 1.559 1.545 1.540 1.527 1.524 1.524
500.000 1.573 1.553 1.555 1.538 1.539 1.567
250.000 1.543 1.549 1.546 1.537 1.546 1.535
125.000 1.546 1.554 1.548 1.549 1.533 1.550
62.500 1.567 1.547 1.550 1.546 1.540 1.548
31.250 1.561 1.553 1.535 1.542 1.550 1.543
15.625 1.543 1.551 1.546 1.544 1.540 1.5328
7.812 1.539 1.554 1.553 1.553 1.537 1.547
3.906 1.535 1.555 1.554 1.553 1.536 1.547
1.953 1.535 1.546 1.550 1.553 1.540 1.547
0.977 1.536 1.548 1.551 1.552 1.541 1.547
0.488 1.539 1.549 1.550 1.551 1.542 1.547
sidering the 1969 jerk morphology model of Chapter 3: the first line (in blue) shows the variation of con-
ductivity for the lower mantle (1900 km thick), the first column (in green) the values for the D (300 km
thick). The misfit values in red are all smaller than the reference misfit (1.542), while the other misfit values
(in black) are larger than the reference. This table is a zoom of Table B.2 and with a thinner grid of electrical
conductivities.
121
15.625 7.812 3.906 1.953 0.977 0.488

1000.000 1.461 1.465 1.459 1.441 1.432 1.435
500.000 1.494 1.457 1.471 1.457 1.455 1.491
250.000 1.442 1.474 1.457 1.453 1.462 1.449
125.000 1.447 1.466 1.464 1.466 1.447 1.464
62.500 1.479 1.475 1.455 1.463 1.455 1.464
31.250 1.477 1.479 1.456 1.459 1.465 1.456
15.625 1.460 1.472 1.457 1.459 1.456 1.447
7.812 1.460 1.468 1.473 1.467 1.450 1.462
3.906 1.465 1.465 1.470 1.467 1.450 1.462
1.953 1.461 1.464 1.466 1.467 1.447 1.462
0.977 1.460 1.463 1.466 1.466 1.464 1.461
0.488 1.460 1.464 1.466 1.467 1.467 1.462
sidering the 1969 jerk morphology model of Le Huy et al. (1998): the first line (in blue) shows the variation
of conductivity for the lower mantle (1900 km thick), the first column (in green) the values for the D (300
km thick). The misfit values in red are all smaller than the reference misfit (1.542), while the other misfit
values (in black) are larger than the reference.
122
Table B.5: Misfit values for the three-layer model, where the lower mantle is divided into
two layers of 950 km and the D with 300 km thick. For the upper mantle we consider
Kuvshinov & Olsen (2006) electrical conductivity model. The red numbers in this table
correspond to the values smaller or equal to the reference misfits for considering our spher-
ical harmonic model (1.542) called misfit 1 and by Le Huy et al.s model (1.456) called
misfit 2.
of D layer of bottom of top misfit 1 misfit 2

lower mantle lower mantle
0.457 0.457 0.457 1.547 1.461
0.457 0.457 1.372 1.545 1.461
0.457 0.457 4.115 1.553 1.469
0.457 0.457 12.346 1.549 1.457
0.457 1.372 0.457 1.549 1.463
0.457 1.372 1.372 1.538 1.451
0.457 1.372 4.115 1.549 1.451
0.457 1.372 12.346 1.546 1.465
0.457 4.115 0.457 1.548 1.463
0.457 4.115 1.372 1.551 1.468
0.457 4.115 4.115 1.552 1.471
0.457 4.115 12.346 1.558 1.482
0.457 12.346 0.457 1.537 1.437
0.457 12.346 1.372 1.538 1.462
0.457 12.346 4.115 1.547 1.469
0.457 12.346 12.346 1.554 1.462
1.372 0.457 0.457 1.547 1.462
1.372 0.457 1.372 1.543 1.447
1.372 0.457 4.115 1.555 1.470
1.372 0.457 12.346 1.550 1.455
1.372 1.372 0.457 1.545 1.462
1.372 1.372 1.372 1.538 1.452
1.372 1.372 4.115 1.546 1.451
1.372 1.372 12.346 1.550 1.465
1.372 4.115 0.457 1.551 1.463
1.372 4.115 1.372 1.550 1.467
1.372 4.115 4.115 1.555 1.472
1.372 4.115 12.346 1.554 1.481
1.372 12.346 0.457 1.537 1.441
1.372 12.346 1.372 1.538 1.463
1.372 12.346 4.115 1.547 1.468
1.372 12.346 12.346 1.552 1.469
4.115 0.457 0.457 1.547 1.461
4.115 0.457 1.372 1.536 1.448
4.115 0.457 4.115 1.550 1.470
4.115 0.457 12.346 1.549 1.459
4.115 1.372 0.457 1.543 1.446
123
4.115 1.372 1.372 1.539 1.455

4.115 1.372 4.115 1.543 1.452
4.115 1.372 12.346 1.548 1.466
4.115 4.115 0.457 1.551 1.465
4.115 4.115 1.372 1.547 1.466
4.115 4.115 4.115 1.554 1.476
4.115 4.115 12.346 1.558 1.481
4.115 12.346 0.457 1.537 1.442
4.115 12.346 1.372 1.545 1.463
4.115 12.346 4.115 1.547 1.457
4.115 12.346 12.346 1.553 1.460
12.346 0.457 0.457 1.542 1.445
12.346 0.457 1.372 1.538 1.455
12.346 0.457 4.115 1.541 1.451
12.346 0.457 12.346 1.548 1.465
12.346 1.372 0.457 1.538 1.450
12.346 1.372 1.372 1.550 1.466
12.346 1.372 4.115 1.546 1.462
12.346 1.372 12.346 1.554 1.469
12.346 4.115 0.457 1.551 1.465
12.346 4.115 1.372 1.539 1.450
12.346 4.115 4.115 1.545 1.458
12.346 4.115 12.346 1.561 1.466
12.346 12.346 0.457 1.532 1.450
12.346 12.346 1.372 1.551 1.470
12.346 12.346 4.115 1.549 1.462
12.346 12.346 12.346 1.555 1.463
37.037 0.457 0.457 1.547 1.460
37.037 0.457 1.372 1.551 1.466
37.037 0.457 4.115 1.552 1.471
37.037 0.457 12.346 1.558 1.476
37.037 1.372 0.457 1.549 1.464
37.037 1.372 1.372 1.537 1.449
37.037 1.372 4.115 1.548 1.456
37.037 1.372 12.346 1.560 1.470
37.037 4.115 0.457 1.541 1.456
37.037 4.115 1.372 1.554 1.468
37.037 4.115 4.115 1.546 1.466
37.037 4.115 12.346 1.548 1.443
37.037 12.346 0.457 1.552 1.467
37.037 12.346 1.372 1.545 1.459
37.037 12.346 4.115 1.551 1.472
37.037 12.346 12.346 1.561 1.476
111.111 0.457 0.457 1.549 1.464
111.111 0.457 1.372 1.532 1.452
111.111 0.457 4.115 1.543 1.459
124
111.111 0.457 12.346 1.555 1.464

111.111 1.372 0.457 1.550 1.464
111.111 1.372 1.372 1.535 1.451
111.111 1.372 4.115 1.546 1.463
111.111 1.372 12.346 1.555 1.466
111.111 4.115 0.457 1.542 1.460
111.111 4.115 1.372 1.550 1.449
111.111 4.115 4.115 1.548 1.465
111.111 4.115 12.346 1.549 1.470
111.111 12.346 0.457 1.556 1.470
111.111 12.346 1.372 1.551 1.455
111.111 12.346 4.115 1.536 1.463
111.111 12.346 12.346 1.541 1.466
333.333 0.457 0.457 1.541 1.456
333.333 0.457 1.372 1.540 1.461
333.333 0.457 4.115 1.543 1.457
333.333 0.457 12.346 1.561 1.487
333.333 1.372 0.457 1.545 1.434
333.333 1.372 1.372 1.546 1.463
333.333 1.372 4.115 1.543 1.458
333.333 1.372 12.346 1.565 1.472
333.333 4.115 0.457 1.542 1.445
333.333 4.115 1.372 1.541 1.461
333.333 4.115 4.115 1.551 1.455
333.333 4.115 12.346 1.533 1.457
333.333 12.346 0.457 1.549 1.463
333.333 12.346 1.372 1.544 1.464
333.333 12.346 4.115 1.554 1.461
333.333 12.346 12.346 1.570 1.478
125
Bibliography
Achache, J., Courtillot, J., Ducruix, J., & Le Moul, J.-L., 1980. The late 1960s secular variation
impulse: further constraints on deep mantle conductivity, Phys. Earth Planet. Int., 23, 7275.
Achache, J., Le Moul, J.-L., & Courtillot, J., 1981. Geophys. J. R. Astr. Soc., 65, 579?601.
Alexandrescu, M., Gilbert, D., Hulot, G., Le Moul, J.-L., & Saracco, G., 1995. Detection of
geomagnetic jerks using wavelet analysis, J. Geophys. Res., 100, 1255712572.
Alexandrescu, M., Gilbert, D., Hulot, G., Le Moul, J.-L., & Saracco, G., 1996. Worldwide
wavelet analysis of geomagnetic jerks, J. Geophys. Res., 101(B10), 2197521994.
Alldredge, L. R., 1977. Deep mantle conductivity, J. Geophys. Res., 82(33), 54275431.
Alldredge, L. R., 1984. A discussion of impulses and jerks in the geomagnetic field, J. Geophys.
Res., 89, 44034412.
Backus, G., Parker, R., & Constable, S., 1996. Foundations of Geomagnetism, Cambridge Univer-
sity Press.
Backus, G. E., 1983. Application of mantle filter theory to the magnetic jerk of 1969, Geophys. J.
R. Astr. Soc., 74, 713746.
Bahr, K., Olsen, N., & Shankland, T., 1993. Geophys. Res. Lett., 20(24), 29372940.
Banks, R., 2007. Geomagnetic deep sounding, in Encyclopedia of Geomagnetism, pp. 307309,
Kluwer Press.
Banks, R. J., 1969. Geomagnetic variations and the electrical conductivity of the upper mantle,
Geophys. J. R. Astr. Soc., 17, 457487.
Bloxham, J., Zatman, S., & Dumberry, M., 2002. The origin of geomagnetic jerks, Nature,
420(6911), 6568.
Campbell, W. H., 2003. Introduction to Geomagnetic Fields, Cambridge University Press, 2nd
edn.
Chambodut, A. & Mandea, M., 2005. Evidence for geomagnetic jerks in comprehensive models,
Earth Planets Space, 57, 139149.
Chulliat, A. & Telali, K., 2007. World monthly means database project, Publs. Inst. Geophys. Pol.
Acad. Sc., 398.
126
Constable, S., 2007. Geomagnetism in Treatise on Geophysics, vol. 5, pp. 237276, Elsevier.
Constable, S. & Constable, C., 2004. Observing geomagnetic induction in magnetic satellite
measurements and associated implications for mantle conductivity, Geochemistry, Geophysics,
Geosystems, 5, doi:10.1029/2003GC000,634.
Courtillot, V. & Le-Mouel, J.-L., 1984. Geomagnetic secular variation impulses., Nature, Review
article, 311, 709716.
Davies, G. F., 1999. Dynamic Earth- Plates, Plumes and Mantle Convection, Cambridge Univer-
sity Press.
De Michelis, P. & Tozzi, R., 2005. A local intermittency measure (LIM) approach to the detection
of geomagnetic jerks, Geophys. Res. Lett., 235, 261272.
De Michelis, P., Cafarella, L., & Meloni, A., 1998. Worldwide character of the 1991 geomagnetic
jerk, Earth Planet. Sci. Lett., 25(3), 377380.
De Michelis, P., Cafarella, L., & Meloni, A., 2000. A global analysis of the 1991 geomagnetic
jerk, Geophys. J. Int., 143, 545556.
Dobson, D. P. & Brodholt, J. P., 2000. The electrical conductivity of the lower mantle phase
magnesiowstite at high temperatures and pressures, J. Geophys. Res., 105(B1), 531538.
Ducruix, J., Courtillot, V., & Le Mou/"el, J.-L., 1980. The late 1960s secular variation impulse,
the eleven year magnetic variation and the electrical conductivity of the deep mantle, Geophys.
J. R. Astr. Soc., 61, 708711.
Egbert, G. D. & Booker, J. R., 1992. Very long period magneto-tellurics at tucson observatory:
Implications for mantle conductivity., J. Geophys. Res., 97, 15,09915,112.
Gubbins, D., 1984. Geomagnetic field analysis - II. secular variation consistent with a perfectly
conducting core, Geophys. J. R. Astr. Soc., 77, 753766.
Gubbins, D., 2004. Time series analysis and inverse theory for geophysicists, Cambridge Univer-
sity Press.
Gubbins, D. & Bloxham, J., 1985. Geomagnetic field analysis - III. magnetic fields on the core -
mantle boundary, Geophys. J. R. Astr. Soc., 80, 695713.
Gubbins, D. & Roberts, P. H., 1987. Chapter 1: Magnetohydrodynamics of the Earths core in
Geomagnetism, vol. 2, Academic Press, editor J. A. Jacobs.
Gubbins, D. & Tomlinson, L., 1986. Secular variation from monthly means from Apia and Am-
berley magnetic observatories, Geophys. J. R. Astr. Soc., 86, 603616.
Hirose, K., J., B., T., L., & Yuen, D., 2007. An introduction to post-perovskite: the last mantle
phase transition in Post-perovskite: the last mantle phase transition, American Geophysical
Union.
Jackson, A., 2007. Time-dependent models of the geomagnetic field, in Encyclopedia of Geomag-
netism, pp. 948953, Kluwer Press.
127
Jackson, A., Jonkers, A. R. T., & Walker, M. R., 2000. Four centuries of geomagnetic secular
variation from historical records, Phil. Trans. R. Soc. Lond. A, 358, 957990.
Jeanloz, R., 1990. The nature of the Earths core, Annu. Rev. Earth Planet. Sci., 18, 35786.
Katsura, T., Sato, K., & Ito, E., 1998. Electrical conductivity of silicate perovskite at lower-mantle
conditions, Nature, 395, 493495.
Kennett, B. L. N., Engdahl, E. R., & Buland, R., 1995. Constraints on seismic velocities in the
earth from travel times, Geophys. J. Int., 122, 108124.
Kreyzig, E., 2001. Advanced Engineering Mathematics, John Wiley, NY.
Kuvshinov, A. & Olsen, N., 2006. A global model of mantle conductivity derived from 5 years of
CHAMP, rsted, and SAC-C magnetic data, Geophys. Res. Lett., 33.
Kuvshinov, A., Utada, D., Avdeev, D., & Koyama, T., 2005. 3-D modelling and analysis of dst
c-responses in the north pacific ocean region, revisited, Geophys. J. Int., 160, 505526.
Lahiri, B. N. & Price, A. T., 1939. Electromagnetic induction in non-uniform conductors, and the
determination of the conductivity of the earth from terrestrial magnetic variations, Phil. Trans.
R. Soc., A237, 509540.
Langel, R. A., 1987. Chapter 4: The main field. In Geomagnetism, vol. 1, Academic Press, editor
J. A. Jacobs.
Le Huy, M., Alexandrescu, M., Hulot, G., & Le Moul, J.-L., 1998. On the characteristics of
successive geomagnetic jerks, Earth Planets Space, 50, 723732.
Le Moul, J.-L., Ducruix, J., & Duyen, C. H., 1982. The worldwide character of the 1969-1970
impulse of the secular acceleration rate, Phys. Earth Planet. Int., 28, 337350.
Lowes, F., 2007. Geomagnetic spectrum, spatial, in Encyclopedia of Geomagnetism, pp. 350353,
Kluwer Press.
Macmillan, S., 2007. Observatories, overview, in Encyclopedia of Geomagnetism, pp. 3746,

Kluwer Press.
Malin, S. R. C. & Hodder, B. M., 1982. Was the 1970 geomagnetic jerk of internal or external
origin?, Nature, 296, 726728.
Mandea, M., Bellanger, E., & J-L., L. M., 2000. A geomagnetic jerk for the end of the 20th
century?, Earth Planet. Sci. Lett., 183, 369373.
Mandea Alexandrescu, M., Gilbert, D., J-L., L. M., Hulot, G., & Saracco, G., 1999. An estimate of
average lower mantle conductivity by wavelet analysis of geomagnetic jerks, J. Geophys. Res.,
104(B8), 17,73517,745.
Mao, W. L., Shen, G., Prakapenka, V. B., Meng, Y., Campbell, A. J., Heinz, D. L., Shu, J., Hemley,
R. J., & Mao, H.-K., 2004. Ferromagnesian postperovskite silicates in the D" layer of the earth,
PNAS, 101(45), 1586715869.
128
Mao, W. L., Campbell, A. J., Prakapenka, V. B., Hemley, R. J., & Mao, H.-k., 2007. Effect of Iron
on the properties of post-perovskite silicate in Post-perovskite, the last phase transition, chap. 1,
pp. 3746, AGU.
Maus, S., Lhr, H., Balasis, G., Rother, M., & Mandea, M., 2005. Introducing POMME, Potsdam
Magnetic Model of the Earth in Earth Observation with CHAMP: results from Three Years in
Orbit, Springer, Berlin.
McDonald, K., L., 1957. Penetration of the geomagnetic secular field through a mantle with
variable conductivity, J. Geophys. Res., 62(1), 117140.
McLeod, M. G., 1994. Magnetospheric and ionospheric signals in magnetic observatory monthly
means: Electrical conductivity of the deep mantle, J. Geophys. Res., 99(B7), 1357713590.
Menke, W., 1989. Geophysical data analysis: discrete inverse theory, vol. 45 of International
Geophysics Series, Academic Press, INC.
Murakami, M., Hirose, K., Kawamura, K., Sata, N., & Ohishi, Y., 2004. Post-perovskite phase
transition in MgSiO3 , Science, 304, 855858.
Nagao, H., Iyemori, T., Higuchi, T., Nakano, S., & Araki, T., 2002. Local time features of geo-
magnetic jerks, Earth Planets Space, 54, 119131.
Nagao, H., Iyemori, T., Higuchi, T., & Araki, T., 2003. Lower mantle conductivity anomalies esti-
mated from geomagnetic jerks, J. Geophys. Res., 108(B5), 2254, doi:10.1029/2002JB0011786.
Ohta, K., Onoda, S., Hirose, K., Sinmyo, R., Sata, N., Ohishi, Y., & Yasuhara, A., 2008. The
electrical conductivity of post-perovskite in the Earths D layer, Science, 320, 8991.
Olsen, N., 1998. Estimation of C-responses (3h to 720h) and the electrical conductivity of the
mantle beneath Europe, Geophys. J. Int., 133, 298308.
Olsen, N., 1999. Long-period (30 days 1 year) electromagnetic sounding and the electrical
conductivity of the lower mantle beneath Europe, Geophys. J. Int., 138, 179187.
Olsen, N. & Kuvshinov, A., 2004. Modelling the ocean effect of geomagnetic storms, Earth
Planets Space, 56, 252530.
Olsen, N. & Mandea, M., 2007. Investigation of a secular variation impulse using satellite data:
The 2003 geomagnetic jerk, Earth Planets Space, 255, 94105.
Olsen, N. & Mandea, M., 2008. Rapidly changing flows in the Earths core, Nature Geoscience.
Olsen, N., Luhr, H., Sabaka, T., Mandea, M., Rother, M., Tofner-Clausen, L., & Choi, S., 2006.
CHAOS-a model of the earths magnetic field derived from CHAMP, rsted, and SAC-C mag-
netic satellite data, Geophys. J. Int., 166, 6775.
Ono, S., Oganov, A. R., Koyama, T., & Shimizu, H., 2006. Stability and compressibility of the
high-pressure phases of Al2 O3 up to 200 GPa: Implications for the electrical conductivity of the
base of the lower mantle, Earth Planet. Sci. Lett., 246, 326335.
129
Pais, M. A. & Miranda, J. M. A., 1995. Secular variation in Coimbra (Portugal) since 1866, J.
Geomagn. Geoelectr., 47, 267282.
Parker, R., 1994. Geophysical Inverse Theory, Princeton, NJ Princeton University Press.
Peyronneau, J. & Poirier, J. P., 1989. Electrical conductivity of the Earths lower mantle, Nature,
342, 537539.
Pinheiro, K. & Jackson, A., 2008. Can a 1D mantle electrical conductivity model generate mag-
netic jerk differential time delays?, Geophys. J. Int., 173(3), 781792.
Roberts, R. G., 1986. Global electromagnetic induction, Surveys in Geophysics, (8), 339374.
Runcorn, S. K., 1955. The electrical conductivity of the earths mantle, Transactions American
Geophysical Union, 36(2), 191198.
Sabaka, T. J., Olsen, N., & Langel, R. A., 2002. A comprehensive model of the quiet-time, near-
earth magnetic field: phase 3, Geophys. J. Int., 151, 3268.
Sabaka, T. J., Olsen, N., & Purucker, M. E., 2004. Extending comprehensive models of the earths
magnetic field with rsted and CHAMP data, Geophys. J. Int., pp. 127.
Schultz, A., Kurtz, R. D., Chave, A., & Jones, A. G., 1993. Conductivity discontinuities in the
upper mantle beneath a stable craton, Geophys. Res. Lett., 20, 29412944.
Shankland, T. J., Peyronneau, J., & Poirier, J.-P., 1993. Electrical conductivity of the Earths lower
mantle., Nature, 366, 453455.
Sivia, D. S. & Skilling, J., 2006. Data Analysis, A Bayesian Tutorial, Oxford University Press,
2nd edn.
Stewart, D. N. & Whaler, K. A., 1995. Optimal piecewise regression analysis and its application
to geomagnetic time series, Geophys. J. Int., 121, 710724.
Tarantola, A., 2005. Inverse Problem Theory and Methods for Model Parameter Estimation, Soci-
ety for Industrial and Applied Mathematics.
Velmsk, J. & Martinec, Z., 2005. Time-domain, spherical harmonic-finite element approach to
transient three-dimensional geomagnetic induction in a spherical heterogeneous earth, Geophys.
J. Int., 161, 81101.
Verbanac, G., Luhr, H., & Rother, M., 2006. Evidence of the ring current effect in geomagnetic
observatories annual means, Geofizika, 23, 1320.
Walker, M. & Jackson, A., 2000. Robust modelling of the earths magnetic field, Geophys. J. Int.,
143(3), 799808.
Wardinski, I. & Holme, R., 2006. A time-dependent model of the Earths magnetic field and its
secular variation for the period 1980-2000, J. Geophys. Res., 111.
Wardinsky, I., 2007. Geomagnetic secular variation, in Encyclopedia of Geomagnetism, pp. 346
350, Kluwer Press.
130
Whaler, K. A., 1987. A new method for analysing geomagnetic impulses, Phys. Earth Planet. Int.,
48, 221240.
Wood, B. J. & Nell, J., 1991. High-temperature electrical conductivity of the lower-mantle phase
(Mg,Fe)O, Nature, 351, 309311.
Xu, Y., Shankland, T. J., & Poe, B. T., 2000. Laboratory-based electrical conductivity in the earth
mantle, J. Geophys. Res., 105(B12), 27,865 27,875.
Yagi, T., 2007. Review of Experimental Studies on Mantle Phase Transitions in Post-perovskite,
the last phase transition, chap. 1, pp. 918, AGU.
131
Katia Jasbinschek Pinheiro

Institute of Geophysics, Earth and Planetary Magnetism Group
ETH Zurich, NO H27, 8087 Switzerland
1. Education
PhD in Geophysics (started 12/2004 expected by 02/2009) ETH Zurich, Institute of Geo-
physics, Earth and Planetary Magnetism Group
Project: Mantle electrical conductivity estimates from geomagnetic jerk observations

PhD at the University of Leeds (School of Earth & Environment, Earth Sciences), Eng-
land, U.K. from 12/2004 to 04/2007:
Visiting student at the Australian National University
(RSES- Research School of Earth Sciences) from 01/2005 to 04/2005
Scholarship: Dorothy Hodgkin Post-Graduate Award - NERC/BP
PhD at ETH (Swiss Federal Institute of Technology) Zurich, Institute of Geophysics,
Earth and Planetary Magnetism Group, from 12/2004 to 04/2007
MSc degree in Geophysics (03/2001 - 06/2003) Observatrio Nacional (National Observa-

tory, Dept. Geophysics)
Rio de Janeiro, Brazil
MSc dissertation: Deteco de impulsos da variao secular geomagntica no Obser-

vatrio de Vassouras (Detection of geomagnetic jerks at Vassouras Magnetic Observa-
tory, in Portuguese)
Scholarship: CAPES
BSc degree in Oceanography (03/1997 - 02/2001)

UERJ (State University of Rio de Janeiro, Inst. Geosciences, Dept. Oceanography)
Undergraduate Project: Caracterizao geofsica da sedimentao e estrutura do em-

basamento da Baia de Ilha Grande, Rio de Janeiro (Geophysical characterization of
the sedimentation and basement structure of the Ilha Grande Bay, Rio de Janeiro, in
Portuguese)
Scholarships: PIBIC/UERJ (06/1997 - 07/1998); SR3/UERJ (08/1998 - 05/1999); FAPERJ
(02/2000 - 01/2001)
2. Work Experience
Geophysicist at PETROBRAS (Brazilian State Oil Company)-

(11/2002 - 10/2004)
132
SMS (Safety, Environment and Health Division), Rio de Janeiro, Brazil

Duties: Survey planning and supervising of data collection, processing and interpreta-
tion of geophysical data (GPR- Ground Penetrating Radar and VES- Vertical Electrical
Sounding) collected by contractors on impacted areas.
3. Computational Experience
LINUX, WINDOWS and MAC systems
Programming in MATLAB, MATHEMATICA and FORTRAN
4. Publications
Pinheiro, K. and Jackson, A.; Can a 1D mantle electrical conductivity model generate
magnetic jerk differential time delays? (2008)
Paper published and awarded in Geophysical Journal International, v. 173, p. 781-792,

2008
Student Author Award (http://www.wiley.com/bw/journal.asp?ref=0956-540x).
Pinheiro, K. and Jackson, A.; Data analysis of the 1969, 1978, 1991 and 1999 geomag-
netic jerks (2008)
Paper submitted to Geophysical Journal International in August 2008
5. Conference Proceedings
Pinheiro, K. and Jackson, A.; Mantle electrical conductivity estimates from geomag-
netic jerk observations (October, 2008)
Talk presented at the workshop Transport Properties of the Lower Mantle .
Pinheiro, K. and Jackson, A.; Geomagnetic jerks and mantle electrical conductivity:
the forward and inverse approaches (2008)
Poster presented at ETH PhD Assembly, 2008, Zurich, Switzerland.
Pinheiro, K. and Jackson, A.; Geomagnetic jerks and mantle electrical conductivity
(November, 2007)
Talk presented at So Paulo University (USP), So Paulo, Brazil.
the forward approach (November, 2007)
133
Talk presented at Luiz Muniz Barreto -VI Latin American School of Geomagnetism
-ELAG, Rio de Janeiro, Brazil.
the inverse approach (November, 2007)
Talk presented at Luiz Muniz Barreto -VI Latin American School of Geomagnetism
-ELAG, Rio de Janeiro, Brazil.
Pinheiro, K. and Jackson, A.; Geomagnetic jerk time delays (July, 2007)
Talk presented at IUGG XXIV, Perugia. Italy.
Pinheiro, K. and Jackson, A.; Estimates of geomagnetic jerks time delays from 1D
mantle conductivity models: a forward approach (NOvember, 2006)
Talk presented at 4th Swiss Geoscience Meeting, 2006, Bern, Switzerland.
Pinheiro, K. and Jackson, A.; Mantle conductivity models and geomagnetic jerks (July,
2006)
Poster presented at SEDI, 2006, Prague, Czech Republic.
Pinheiro, K. and Jackson, A.; Geomagnetic jerks and mantle conductivity (May, 2006)
Poster presented at First Swarm International Science Meeting, 2006, Nantes, France.
Pinheiro, K. and Jackson, A.; Mantle conductivity models and geomagnetic jerks (Novem-
ber, 2005)
Poster presented at William Smith Meeting - The Deep Earth: The Structure and Evo-
lution of the Interior of our Planet. The Geological Society, Burlington House, London.
Pinheiro, K. and Jackson, A.; Mantle conductivity models and geomagnetic jerks (Septem-
ber, 2005)
Poster presented at Annual British Geophysical Association (BGA) New Horizons
Post-Graduate Research Conference, Galway, Ireland (commended poster)
Pinheiro, K. and Travassos, J. M.; Impulses in the geomagnetic secular variation at
Vassouras Observatory (Brazil) (April, 2004)
Poster presented at EGU 1st General Assembly, Nice, France
Travel subsidised by Keith Runcorn Travel Award for Non-Europeans (KRTA)
Pinheiro, K.J.R. and Travassos, J.M.; Detection of geomagnetic jerks at Vassouras Ob-
servatory (November, 2003)
Talk presented at the 8th International Congress of the Brazilian Geophysical Society,
Pinheiro, K.J.R. and Benyosef, L.C.C.; Secular variation in Brazil from 1900 to 2000:
a comparison with ELEMAG model (October, 2001)
Poster presented at the 7th International Congress of the Brazilian Geophysical Society,
Salvador, Brazil

Eth 41591 02

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Eth 41591 02

Hochgeladen von

Copyright:

Verfügbare Formate

Diss. ETH No.

MANTLE ELECTRICAL CONDUCTIVITY ESTIMATES

for the degree of

K ATIA JASBINSCHEK DOS R EIS P INHEIRO

National Observatory, Brazil

born 21st February, 1979

accepted on the recommendation of

Prof. Dr. A. Jackson, examiner

Prof. Dr. M. Mandea, co-examiner

P.D. Dr. A. Kuvshinov, co-examiner

in Brazil (2007) and workshop in Japan (2008).

2 Theory of mantle screening 17

3 Data Analysis of Geomagnetic Jerks 35

4 The Forward Problem 69

5 The Inverse Problem 81

A Data analysis tables 103

B Tables with misfit values obtained in the inversion 118

2.1 Radial models of mantle electrical conductivity. . . . . . . . . . . . . . . . . . . 23

5.1 Comparison between analytical and numerical solutions of IRFs. . . . . . . . . . 83

2.1 Delay times and amplitudes of two chosen locations. . . . . . . . . . . . . . . . . 29

e error bar (lower bound)

1.2 Earth Structure

Figure 1.3: Schematic cross section of the D zone.

1.3 Geomagnetic Field Description

1.3.1 Geomagnetic Field Sources

Figure 1.6: Spherical coordinate system. Taken from Kreyzig (2001).

Figure 1.7: Spatial spectrum of the internal geomag-

1.3.2 Geomagnetic Models

CM4 from Sabaka et al. (2004): from 1960 to 2002.5;

1.4 Geomagnetic Jerks

1.5 Mantle Electrical Conductivity

1.5.1 Mineral Physics Studies

1.5.2 Induction Studies: external and internal magnetic fields

changes of external origin. The potential is expanded in terms of spherical harmonics:

the internal and external magnetic fields:

C-responses can also be defined in relation to Q-responses. By defining:

1.6 Thesis objectives and structure

Theory of mantle screening

2.1 Mathematical description

and when is a constant we can re-write equation 2.8 as

Now by Faradays law (equation 2.3) we have

B = (T(r, , ,t)r) + (P(r, , ,t)r), (2.15)

and its inverse Fourier transform as:

2.2 Backus mantle filter theory

which is the area under the

2.2.1 Backus low frequency approximation

where nn are called the nth ordinary moments, given by

and (n )n the nth central moments:

and at the Earths surface the output is

We now illustrate Backus phenomenon of mode mixing in different locations correspondent

which is called the total amplitude. If we let:

then equation 2.62 becomes

and the differential delay times are calculated by:

2.2.2 Backus [1983] linear kernels

where (r) are called kernels of equation 2.68, given by

and the differential delays are calculated by

2.3 Limitations of Backus mantle filter theory

1 () in years A (a, 1 , 1 ) A (a, 2 , 2 ) (a, 1 , 1 ) (a, 2 , 2 )

OBS no OBS code latitude longitude CIRF 1 (yr) Gaussian 1 (yr)

differential kernels should be positive.

Data Analysis of Geomagnetic Jerks

3.2 Methods to Detect Geomagnetic Jerks

3.3 Limitations on the Analysis of Geomagnetic Jerks