Physical Geodesy

^ SpringerWienNewYork
Bernhard Hofmann-Wellenhof
Helmut Moritz
Physical Geodesy
Second,
corrected edition
SpringerWienNewYork
Dr. Bernhard Hofmann-Wellenhof
Dr. Helmut Moritz
Institut für Navigation und Satellitengeodäsie
Technische Universität Graz, Graz, Austria
This work is subject to copyright.

All rights are reserved, whether the whole or part of the material is concerned, specifically those of
translation, reprinting, re-use of illustrations, broadcasting, reproduction by photocopying machines
or similar means, and storage in data banks.
Product Liability: The publisher can give no guarantee for the information contained in this book.
This also refers to that on drug dosage and application thereof. In each individual case the respective
user must check the accuracy of the information given by consulting other pharmaceutical literature.
The use of registered names, trademarks, etc., in this publication does not imply, even in the absence
of a specific statement, that such names are exempt from the relevant protective laws and regulations
and therefore free for general use.
© 2006 and 2005 Springer-Verlag Wien
Printed in Austria
SpringerWienNewYork is a part of Springer Science + Business Media,
springeronline.com
Typesetting: Composition by authors

Cover illustration: Elmar Wasle, Graz
Printing: G. Grasl GmbH, 2540 Bad Vöslau, Austria
Printed on acid-free and chlorine-free bleached paper
SPIN 11737476
With 111 Figures
Library of Congress Control Number 2006931768
ISBN-10 3-211-33544-7 SpringerWienNewYork

ISBN-13 978-3-211-33544-4 SpringerWienNewYork
ISBN-10 3-211-23584-1 1st edn. SpringerWienNewYork
This book is dedicated to the memory of
Weikko Aleksanteri Heiskanen

(1895–1971)
Pioneer, initiator, and coauthor of “Physical Geodesy”,

whose dreams have come true in a way unexpected to
all of us.
Foreword
Almost the period of one generation has passed since 1967, the year of the
first release of Physical Geodesy by Weikko A. Heiskanen and Helmut Moritz.
Soon this book became a bestseller. Today, when studying publications deal-
ing with physical geodesy, not surprisingly the book is still frequently quoted.
Have the clocks been stopped since then? Not at all, time has flown as fast
as usual or maybe even faster – at least in someone’s imagination. However,
excellent quality is correlated with a long life expectation. This is the reason
why “the book” still plays an important role in geodetic science and beyond.
In the last decades, nevertheless, geodesy has certainly continually de-
veloped further – on the one hand by new computational methods and ideas
and on the other hand by modern measurement techniques. This is where
the story of this book starts.
Several years ago, I tried to convince Helmut Moritz on the necessity
of a new edition of Physical Geodesy. Even if I encountered some interest,
I did not manage to completely succeed. “Steter Tropfen höhlt den Stein”
(persistent drops hollow out the stone), I thought and started to repeat my
request regularly. The reason for my somehow obstinacy originated from
the past. In 1993, I got the chance to support Helmut Moritz in writing
the book entitled Geometry, Relativity, and Geodesy. For me, this was a
tremendously exciting time where we developed a great cooperation in any
respect. Immediately after this experience, I manifested my desire of another
chance for a cooperation. In these days, the idea of a new edition of Physical
Geodesy matured.
Finally, the persistent drops succeeded. I cannot tell you the Why and the
When; suddenly we had a contract with the Springer Publishing Company.
To me it seemed as if the wheel of time had been turned back – thank you,
Helmut!
Many persons deserve credit and thanks. Prof. Dr. Klaus-Peter Schwarz,
retired from the Department of Geomatics Engineering of the University
of Calgary, strongly influenced the balance between keeping, eliminating,
updating, and adding topics.
Prof. Dr. Herbert Lichtenegger, retired from the Institute of Navigation
and Satellite Geodesy of the Graz University of Technology, was a reviewer
of the book. He has critically read and corrected the full volume. His many
suggestions and improvements, critical remarks and proposals are gratefully
acknowledged.
Prof. Dr. Norbert Kühtreiber from the Institute of Navigation and Satel-
viii Foreword
lite Geodesy of the Graz University of Technology has helped with construc-
tive critique and valuable suggestions. Furthermore, he has strongly helped
to shape Chap. 11 by providing numerical examples and his valuable expe-
rience on the practical aspects of geoid computation.
In several fruitful discussions, Prof. Dr. Roland Pail from the Institute
of Navigation and Satellite Geodesy of the Graz University of Technology
has provided his rich experience on the space gravity missions. Parts of his
structured lecture notes are mirrored in the corresponding section. He also
deserves thanks for a careful proofreading of this section.
The cover illustration was designed and produced by Dipl.-Ing. Elmar
Wasle of TeleConsult Austria GmbH (www.teleconsult-austria.at). When
presenting this illustration to the Springer Publishing Company, the response
was extremely positive because of its eye-catcher quality.
The index of the book was produced using a computer program written
by Dr. Walter Klostius from the Institute of Geoinformation of the Graz
University of Technology. Also, his program helped in the detection of some
spelling errors.
The book is compiled with the text system LATEX2ε. One of the figures
included is also developed with LATEX2ε. The remaining figures are drawn
by using CorelDRAW 11. Primarily Dr. Klaus Legat from the Institute of
Navigation and Satellite Geodesy of the Graz University of Technology de-
serves the thanks for the figures. He was supported by Prof. Dr. Norbert
Kühtreiber. The highly academic level of the producers assures a seal of
quality. Many of these figures are redrawings of the originals in Heiskanen
and Moritz (1967).
I am also grateful to the Springer Publishing Company for their advice
and cooperation.
The inclusion by name of a commercial company or product does not
constitute an endorsement by the authors. In principle, such inclusions were
avoided whenever possible.
Finally, your ideas for a future edition of this book and your advice are
appreciated and encouraged.
The selection of topics is certainly different from the original book written
by Heiskanen and Moritz. However, basically we tried not only to keep the
overall structure wherever possible but also to leave the text unchanged.
The primary selection criteria of the topics were relevancy, tutorial content,
and the interest and expertise of the authors. A detailed description of the
contents is given in the Preface.
March 2005 B. Hofmann-Wellenhof

Preface
This book is a university-level introductory textbook. Physical geodesy is the

science of the figure of the earth and of its gravity field. Particular emphasis
is put on the interaction between geometry, especially GPS, and modern
gravitational techniques. The mathematical tool is potential theory. More
about the purpose and application of physical geodesy will be found in the
subsequent motivation. For better readability, some repetitions are purposely
used. The mathematical apparatus is kept as simple as possible.
The book is divided into 11 chapters, a section of references, and a de-
tailed index which should immediately help in finding certain topics of in-
terest.
The first chapter is an introduction to mathematical potential theory
to the extent needed in the present book. More precisely, it is “classical”
potential theory as represented, e.g., by the book of Kellogg (1929) (this is
our usual mode of quoting references, by name[s] and year). Mathematicians
will notice immediately that the presentation, as in most textbooks on the-
oretical physics, is informal: proofs are frequently omitted or replaced by
“heuristic” considerations.
The second chapter introduces the gravity field of the earth, e.g., the force
of gravity, level surfaces and plumb lines, the geoid, and coordinates naturally
related to them: astronomic latitude and longitude as well as heights above
the geoid. A powerful tool are developments in spherical harmonics. The nat-
ural reference surface is an ellipsoid of revolution equipped with a “normal”
gravity field. This gives us a “Geodetic Reference System” (GRS) or World
Geodetic System (WGS). Deviations of the real gravity field quantities from
the corresponding reference quantities are small and can be linearized. This
makes it possible to treat geodetic problems as relatively simple boundary-
value problems of potential theory. A well-known classical solution is Stokes’
integral formula.
The third chapter deals with gravity reductions, in particular reductions
using the theory of isostasy. A first link to geophysics is established in this
way.
The fourth chapter considers the problem of heights, which is more com-
plicated than one would think at the beginning. The first four chapters are
an update of the old book by Heiskanen and Moritz (1967), which serves as
the template of this book.
The fifth chapter is central in several respects. It is vastly expanded
as compared to the former book and has a completely different structure.
x Preface
The problem of interrelating geometric and physical aspects is met here

in all its complexity, from the global (Part I) to the classical local aspects
(Part III), the regional “three-dimensional geodesy” in the pre-satellite sense
(Part II) forming a transition. One could also say, Part I is integral and
Part III is differential. Part I, geocentric and global reference systems, has
been made possible only by highly precise geometric satellite methods. The
problem of the third dimension, one of the most difficult tasks of geodesy,
is formulated and solved here in the most direct and natural way. Part II
is an attempt to solve this problem classically in a nondifferential way, but
the weak link is the measurement of the zenith distances which are too
inaccurate because of atmospheric refraction. The classical way out of this
dilemma, still valid today, is the astrogeodetic integration of deflections of
the vertical as discussed in Part III.
The sixth chapter is relatively slight, treating the computation of the
gravity field up to about 10 km, with a view to application to airborne
gravimetry. It is a streamlined version of the old Chapter 6.
The seventh chapter corresponds to the old Chapter 9, but it is greatly
expanded to reflect the enormous progress of satellite methods for the de-
termination of the global gravitational field. The main problem has been
the gap between this global field at high elevations and the detailed but
ill-distributed terrestrial gravity measurements. The new dedicated satellite
missions, intended to close this gap, are described.
The eighth chapter, on Molodensky’s and related theories, is again con-
siderably expanded, owing to their great conceptual importance. Molodensky
was the first to base physical geodesy on boundary-value problems on the
physical earth’s surface rather than on the geoid. His conceptual frame in-
cludes also astrogeodetic methods. Although conceived by Molodensky as a
method to avoid the reduction of gravity to sea level, in mountain areas it
works best if combined with isostatic and similar gravity reductions which
are now familiar as “remove-restore” techniques. A great change with re-
spect to the earlier book is the fact that now we understand Molodensky’s
theory much better, so that we can treat it now completely with elementary
mathematics, without the use of integral equations.
The ninth chapter is a practically unchanged update of the old Chap-
ter 7. The statistical treatment of gravity has undergone an enormous and
unexpected increase in importance, theoretical as well as practical, so as to
warrant a special new chapter.
The tenth chapter deals with least-squares collocation. This is a great
synthesis of a generalization of least-squares prediction of gravity treated in
the ninth chapter, with the theory of Hilbert space with kernel functions, and
with ideas of gravity reduction in the theory of Molodensky. This synthesis
Preface xi
is due to a small publication by Krarup (1969). It has an extremely simple

mathematical structure and uses only matrix methods well-suited for elec-
tronic computation; still, it permits the combination of virtually all geodetic
data types. Thus it has become very popular for numerical computation.
The eleventh chapter illustrates these various methods by computations
in Austria, which combines a difficult topography with easily available data.
Internet citations within the text omit the part “http://” if the address con-
tains “www”; therefore, “www.esa.int” means “http://www.esa.int”. Usu-
ally, internet addresses given in the text are not repeated in the list of refer-
ences. Therefore, the list of references does not yield a complete picture of
the references of which we have been benefiting.
The use of the internet sources caused some troubles for the following
reason. When looking for a proper and concise explanation or definition,
quite often identical descriptions were found at different locations. So the
unsolvable problem arose to figure out the earlier and original source. In
these cases, sometimes the decision was made, to avoid a possible conflict of
interests, by omitting the citation of the source at all. This means that some
phrases or sentences may have been adapted from internet sources. On the
other side, as soon as this book is released, it may and will also serve as an
input source for several homepages.
For bibliographical references, the most readily accessible or most com-
prehensive publication of an author on a particular topic is given rather than
his first. The list of references does not aim at completeness; some important
publications may have been omitted but never on purpose.
The (American) spelling of a word is adopted from Webster’s Dictionary
of the English Language (third edition, unabridged). Apart from typical
differences like the American “leveling” in contrast to the British “level-
ling”, this may lead to other divergences when comparing dictionaries. Web-
ster’s Dictionary always combines the negation “non” and the following word
without hyphen unless a capital letter follows. Therefore “nongravitational”,
“nonpropulsed”, “nonsimultaneity” and “non-European” are corresponding
spellings.
Symbols representing a vector or a matrix are in boldface. The inner or
scalar product of two vectors is indicated by a dot “·”. The norm of a vector,
i.e., its length, is indicated by two double-bars “”. Vectors not related to
matrices are written either as column or as row vectors, whatever is more
convenient.
March 2005 B. Hofmann-Wellenhof H. Moritz

xii Preface
Preface to the second edition

Compared to the original version, this second edition only answers four times
the question marks resulting from a wrong label denotation and, for LATEX2ε
experts, some “overfull \hbox” warnings indicating too long lines without
finding appropriate linebreaks were eliminated. Furthermore, one misspelling
was corrected, some punctuation problems were solved differently, and a few
sentences were reformulated or updated. Therefore, it is considered a cor-
rected version and not a revised one. This is a significant difference because it
implies that possible advice for improvements from readers or reviewers were
not taken into consideration. The reason for this is not a haughty disregard
of other ideas but the tightness of the time schedule. To date only a very
small number of reviews has been released. This almost coincided with the
message of the Springer Publishing Company that the first edition will be
sold out shortly. Therefore, we are grateful to those who gave us advice and
we further encourage all readers accordingly because their support might
help to improve another edition.
April 2006 B. Hofmann-Wellenhof H. Moritz

Contents
Motivation 1
1 Fundamentals of potential theory 3

1.1 Attraction and potential . . . . . . . . . . . . . . . . . . . . . 3
1.2 Potential of a solid body . . . . . . . . . . . . . . . . . . . . . 6
1.3 Harmonic functions . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Laplace’s equation in spherical coordinates . . . . . . . . . . 9
1.5 Spherical harmonics . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Surface spherical harmonics . . . . . . . . . . . . . . . . . . . 12
1.7 Legendre’s functions . . . . . . . . . . . . . . . . . . . . . . . 14
1.8 Legendre’s functions of the second kind . . . . . . . . . . . . 19
1.9 Expansion theorem and orthogonality relations . . . . . . . . 21
1.10 Fully normalized spherical harmonics . . . . . . . . . . . . . . 23
1.11 Expansion of the reciprocal distance into zonal harmonics
and decomposition formula . . . . . . . . . . . . . . . . . . . 25
1.12 Solution of Dirichlet’s problem by means of spherical
harmonics and Poisson’s integral . . . . . . . . . . . . . . . . 27
1.13 Other boundary-value problems . . . . . . . . . . . . . . . . . 29
1.14 The radial derivative of a harmonic function . . . . . . . . . . 32
1.15 Laplace’s equation in ellipsoidal-harmonic coordinates . . . . 34
1.16 Ellipsoidal harmonics . . . . . . . . . . . . . . . . . . . . . . . 37
2 Gravity field of the earth 43

2.1 Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.2 Level surfaces and plumb lines . . . . . . . . . . . . . . . . . 46
2.3 Curvature of level surfaces and plumb lines . . . . . . . . . . 48
2.4 Natural coordinates . . . . . . . . . . . . . . . . . . . . . . . 54
2.5 The potential of the earth in terms of spherical harmonics . . 56
2.6 Harmonics of lower degree . . . . . . . . . . . . . . . . . . . . 61
2.7 The gravity field of the level ellipsoid . . . . . . . . . . . . . . 64
2.8 Normal gravity . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.9 Expansion of the normal potential in spherical harmonics . . 73
2.10 Series expansions for the normal gravity field . . . . . . . . . 76
2.11 Reference ellipsoid – numerical values . . . . . . . . . . . . . 83
2.12 Anomalous gravity field, geoidal undulations, and deflections
of the vertical . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
xiv Contents
2.13 Spherical approximation and expansion of the disturbing

potential in spherical harmonics . . . . . . . . . . . . . . . . . 96
2.14 Gravity anomalies outside the earth . . . . . . . . . . . . . . 99
2.15 Stokes’ formula . . . . . . . . . . . . . . . . . . . . . . . . . . 102
2.16 Explicit form of Stokes’ integral and Stokes’ function in
spherical harmonics . . . . . . . . . . . . . . . . . . . . . . . . 105
2.17 Generalization to an arbitrary reference ellipsoid . . . . . . . 109
2.18 Gravity disturbances and Koch’s formula . . . . . . . . . . . 115
2.19 Deflections of the vertical and formula of Vening Meinesz . . 116
2.20 The vertical gradient of gravity . . . . . . . . . . . . . . . . . 119
2.21 Practical evaluation of the integral formulas . . . . . . . . . . 122
3 Gravity reduction 129

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.2 Auxiliary formulas . . . . . . . . . . . . . . . . . . . . . . . . 130
3.3 Free-air reduction . . . . . . . . . . . . . . . . . . . . . . . . . 134
3.4 Bouguer reduction . . . . . . . . . . . . . . . . . . . . . . . . 135
3.5 Poincaré and Prey reduction . . . . . . . . . . . . . . . . . . 138
3.6 Isostatic reduction . . . . . . . . . . . . . . . . . . . . . . . . 141
3.6.1 Isostasy . . . . . . . . . . . . . . . . . . . . . . . . . . 141
3.6.2 Topographic-isostatic reductions . . . . . . . . . . . . 146
3.7 The indirect effect . . . . . . . . . . . . . . . . . . . . . . . . 149
3.8 The inversion reduction of Rudzki . . . . . . . . . . . . . . . 151
3.9 The condensation reduction of Helmert . . . . . . . . . . . . . 153
4 Heights 157
4.1 Spirit leveling . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.2 Geopotential numbers and dynamic heights . . . . . . . . . . 159
4.3 Orthometric heights . . . . . . . . . . . . . . . . . . . . . . . 161
4.4 Normal heights . . . . . . . . . . . . . . . . . . . . . . . . . . 166
4.5 Comparison of different height systems . . . . . . . . . . . . . 168
4.6 GPS leveling . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
5 The geometry of the earth 173

5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Part I: Global reference systems after GPS . . . . . . . . . . . . . 175
5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
5.3 The Global Positioning System . . . . . . . . . . . . . . . . . 176
5.3.1 Basic concept . . . . . . . . . . . . . . . . . . . . . . . 177
5.3.2 System architecture . . . . . . . . . . . . . . . . . . . 177
5.3.3 Satellite signal and observables . . . . . . . . . . . . . 181
Contents xv
5.3.4 System capabilities and accuracies . . . . . . . . . . . 183

5.3.5 GPS modernization concept . . . . . . . . . . . . . . . 186
5.4 From GPS to coordinates . . . . . . . . . . . . . . . . . . . . 187
5.4.1 Point positioning with code pseudoranges . . . . . . . 188
5.4.2 Relative positioning with phase pseudoranges . . . . . 189
5.5 Projection onto the ellipsoid . . . . . . . . . . . . . . . . . . . 192
5.6 Coordinate transformations . . . . . . . . . . . . . . . . . . . 193
5.6.1 Ellipsoidal and rectangular coordinates . . . . . . . . 193
5.6.2 Ellipsoidal, ellipsoidal-harmonic, and spherical
coordinates . . . . . . . . . . . . . . . . . . . . . . . . 197
5.7 Geodetic datum transformations . . . . . . . . . . . . . . . . 198
5.7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . 198
5.7.2 Three-dimensional transformation in general form . . 199
5.7.3 Three-dimensional transformation between WGS 84
and a local system . . . . . . . . . . . . . . . . . . . . 201
5.7.4 Differential formulas for other datum transformations 203
Part II: Three-dimensional geodesy: a transition . . . . . . . . . . 207
5.8 The three-dimensional geodesy of Bruns and Hotine . . . . . 207
5.9 Global coordinates and local level coordinates . . . . . . . . . 208
5.10 Combining terrestrial data and GPS . . . . . . . . . . . . . . 210
5.10.1 Common coordinate system . . . . . . . . . . . . . . . 210
5.10.2 Representation of measurement quantities . . . . . . . 211
Part III: Local geodetic datums . . . . . . . . . . . . . . . . . . . . 216
5.11 Formulation of the problem . . . . . . . . . . . . . . . . . . . 216
5.12 Reduction of the astronomical measurements to the ellipsoid 217
5.13 Reduction of horizontal and vertical angles and of distances . 221
5.14 The astrogeodetic determination of the geoid . . . . . . . . . 223
5.15 Reduction for the curvature of the plumb line . . . . . . . . . 228
5.16 Best-fitting ellipsoids and the mean earth ellipsoid . . . . . . 233
6 Gravity field outside the earth 239

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
6.2 Normal gravity vector . . . . . . . . . . . . . . . . . . . . . . 240
6.3 Gravity disturbance vector from gravity anomalies . . . . . . 242
6.4 Gravity disturbances by upward continuation . . . . . . . . . 247
6.5 Additional considerations . . . . . . . . . . . . . . . . . . . . 249
6.6 Gravity anomalies and disturbances compared . . . . . . . . . 252
7 Space methods 255

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
7.2 Satellite orbits . . . . . . . . . . . . . . . . . . . . . . . . . . 257
xvi Contents
7.3 Determination of zonal harmonics . . . . . . . . . . . . . . . . 261

7.4 Rectangular coordinates of the satellite and perturbations . . 267
7.5 Determination of tesseral harmonics and station positions . . 271
7.6 New satellite gravity missions . . . . . . . . . . . . . . . . . . 274
7.6.1 Motivation and introductory considerations . . . . . . 274
7.6.2 Measurement concepts . . . . . . . . . . . . . . . . . . 277
7.6.3 The CHAMP mission . . . . . . . . . . . . . . . . . . 280
7.6.4 The GRACE mission . . . . . . . . . . . . . . . . . . . 282
7.6.5 The GOCE mission . . . . . . . . . . . . . . . . . . . 284
8 Modern views on the determination of the figure

of the earth 289
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Part I: Gravimetric methods . . . . . . . . . . . . . . . . . . . . . 290
8.2 Gravity reductions and the geoid . . . . . . . . . . . . . . . . 290
8.3 Geodetic boundary-value problems . . . . . . . . . . . . . . . 294
8.4 Molodensky’s approach and linearization . . . . . . . . . . . . 296
8.5 The spherical case . . . . . . . . . . . . . . . . . . . . . . . . 302
8.6 Solution by analytical continuation . . . . . . . . . . . . . . . 303
8.6.1 The idea . . . . . . . . . . . . . . . . . . . . . . . . . . 303
8.6.2 First-order solution . . . . . . . . . . . . . . . . . . . . 304
8.6.3 Higher-order solution . . . . . . . . . . . . . . . . . . . 308
8.6.4 Problems of analytical continuation . . . . . . . . . . 308
8.6.5 Another perspective . . . . . . . . . . . . . . . . . . . 311
8.7 Deflections of the vertical . . . . . . . . . . . . . . . . . . . . 314
8.8 Gravity disturbances: the GPS case . . . . . . . . . . . . . . . 317
8.9 Gravity reduction in the modern theory . . . . . . . . . . . . 318
8.10 Determination of the geoid from ground-level anomalies . . . 324
8.11 A first balance . . . . . . . . . . . . . . . . . . . . . . . . . . 326
Part II: Astrogeodetic methods according to Molodensky . . . . . 329
8.12 Some background . . . . . . . . . . . . . . . . . . . . . . . . . 329
8.13 Astronomical leveling revisited . . . . . . . . . . . . . . . . . 336
8.14 Topographic-isostatic reduction of vertical deflections . . . . . 338
8.15 The meaning of the geoid . . . . . . . . . . . . . . . . . . . . 341
9 Statistical methods in physical geodesy 345

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
9.2 The covariance function . . . . . . . . . . . . . . . . . . . . . 346
9.3 Expansion of the covariance function in spherical harmonics . 350
9.4 Interpolation and extrapolation of gravity anomalies . . . . . 355
9.5 Accuracy of prediction methods . . . . . . . . . . . . . . . . . 358
Contents xvii
9.6 Least-squares prediction . . . . . . . . . . . . . . . . . . . . . 361

9.7 Correlation with height . . . . . . . . . . . . . . . . . . . . . 363
10 Least-squares collocation 369

10.1 Principles of least-squares collocation . . . . . . . . . . . . . . 369
10.2 Application of collocation to geoid determination . . . . . . . 375
11 Computational methods 379

11.1 The remove-restore principle . . . . . . . . . . . . . . . . . . 379
11.2 Geoid in Austria by collocation . . . . . . . . . . . . . . . . . 381
11.3 Molodensky corrections . . . . . . . . . . . . . . . . . . . . . 384
11.4 The geoid on the internet . . . . . . . . . . . . . . . . . . . . 388
References 389
Subject index 397

Motivation
As we have already indicated in the Preface, the subject of physical geodesy

is the study of the gravity field and the figure of the earth. In former times,
the scientifically relevant “figure of the earth” was the geoid, which is de-
fined as one of the equipotential surfaces of the earth’s gravity potential, of
which the (mean) surface of the oceans forms a part. So the gravity field
immediately enters into the very definition of “figure of the earth”. “Heights
above sea level” are heights above the geoid, and thus are both physically
and geometrically defined.
Gravity, essentialy caused by the earth’s gravitational attraction, has
always been determining the life of humankind, from walking on a hilly road
to crossing the oceans by ship or by airplane. It has also formed the shape
of our planet.
Scientific geodesy started when leading scientists such as Newton recog-
nized that the earth cannot be a sphere but must rather be flattened because
of the earth’s rotation. Not very much, but probably measurably. This was
one of the greatest scientific problems of that time.
Therefore, around 1740, the French Academy of Sciences undertook two
expeditions, one under Bouguer to Peru and one under Maupertuis to Lap-
land. Their purpose was to measure the length of a meridional arc of, say,
1 degree of latitude, one close to the equator and one close to the North
Pole. The difference between the two results is a measure of the flattening,
which is the deviation (with respect to the sphere) of the earth ellipsoid.
These measurements clearly indicated that the global figure of the earth is
an ellipsoid of revolution, at least approximately.
The next century was characterized by attempts to define the figure of
the earth more precisely. C.F. Gauss (1777–1855), the “princeps mathemati-
corum”, raised geodesy to the rank of a science. He did this by his theory of
surfaces – which finally led to General Relativity, cf. Moritz and Hofmann-
Wellenhof (1993) – and his adjustment by least squares, the first of all sta-
tistical estimation methods. He liked practical geodetic work and measured
a triangulation net. Gauss also introduced the geoid as the “mathematical
figure of the earth” defined as a level surface of the gravity field.
The geoid deviates from a well-fitting ellipsoid (e.g., the Geodetic Refer-
ence System 1980) by less than 100 meters. Geocentric positions nowadays
can be determined by GPS to an accuracy of better than 1 decimeter in
a purely geometric way. We may define these positions either in terms of
geocentric Cartesian coordinates or as ellipsoidal coordinates ϕ, λ, h.
2 Motivation
So the geometry of the earth can be determined largely independently of

the gravitational field, thanks to GPS and other satellite techniques. Still, the
gravitational field is needed, e.g., for determining the orbits of the satellites
themselves.
It is probable that, by the influence of GPS, gravity anomalies ∆g will
gradually be replaced by gravity disturbances δg. This is taken into account
in the present book.
Gravity has become one of the most sought-for and most interrelated
data in geophysics, and every increase of accuracy has immediately gener-
ated new needs. For instance, the ocean surface as determined by satellite
altimetry is not an exact equipotential surface because of small tilts due
to ocean currents. Thus, this “ocean topography”, measured by comparing
the results of satellite altimetry with precise gravimetric geoids determined
by combining various methods, provides important boundary conditions for
oceanography.
Already Clairaut related the density of the masses inside the earth with
internal gravity on the condition of hydrostatic equilibrium, and this question
with the classical title “The figure of the earth” can now be reconsidered in
the light of satellite data; cf. Moritz (1990).
Geological phenomena in the earth’s crust and upper mantle such as
isostasy and plate tectonics require an interaction between geodesy, geo-
physics, and geology.
Polar motion and anomalies in the earth’s rotation are largely caused by
the ceaseless circulation of the air masses defining weather. Earth rotation
is now monitored by laser and GPS, which provides an unexpected link
between geodesy and meteorology; cf. Moritz and Mueller (1987).
New measuring techniques related to inertial navigation systems (INS)
require an interaction between the geometry and the gravitational field. This
has considerable practical consequences, e.g., in tunnel surveying. GPS stops
short in front of a tunnel, and INS or conventional surveying methods must
take over. Either of them, however, does depend on the gravity field.
The terrestrial measurement of gravity is very time-consuming. Airborne
gravimetry has become operational only after the inertial and gravitational
forces have become separable by combination with GPS.
Not all of this can be treated in detail in the present introductory book. It
is, however, intended as a solid, mathematically oriented and not too difficult
treatise on graduate level leading to one’s own postgraduate research.
The book by Heiskanen and Moritz (1967) stood at the transition be-
tween classical and satellite geodesy. Similarly, the present book stands at
the beginning of an era characterized by “sensor integration”, data combi-
nation, and kinematic and navigational techniques.
1 Fundamentals of potential
theory
1.1 Attraction and potential

The purpose in this preparatory chapter is to present the fundamentals of
potential theory, including spherical and ellipsoidal harmonics, in sufficient
detail to assure a full understanding of the later chapters. Our intent is to
explain the meaning of the theorems and formulas, avoiding long derivations
that can be found in any textbook on classical (before 1950) potential theory;
we recommend Kellogg (1929). A simple rather than completely rigorous
presentation is offered in our book.
Nevertheless, the reader might consider this chapter perhaps more diffi-
cult and abstract than other parts of the book. Since later practical appli-
cations will give concrete meaning to the topics of the present chapter, the
reader may wish to read it only cursorily at first and return to it later when
necessary.
According to Newton’s law of gravitation, two points with masses m1 , m2 ,
separated by a distance l, attract each other with a force
m1 m2
F =G . (1–1)
l2
This force is directed along the line connecting the two points; G is Newton’s
gravitational constant. In SI units (Système International d’unités) based on
meter [m], kilogram [kg], and second [s], the gravitational constant has the
value
G = 6.6742 · 10−11 m3 kg−1 s−2 . (1–2)
The Newtonian gravitational constant G is somewhat of a scandal in
measuring physics. It is on the one hand one of the most important physi-
cal constants, and at the same time one of the least accurately determined
ones. The international authority in this field is the Committee on Data
for Science and Technology (CODATA), see under www.codata.org. In July
2002, CODATA recommended the value of G mentioned above, more pre-
cisely it assigned the value G = (6.6742 ± 0.0010) · 10−11 m3 kg−1 s−2 . The
symbol ± denotes the standard uncertainty, also called standard deviation
or standard error. This corresponds to a relative standard uncertainty of
1.5 · 10−4 or 150 ppm which is a deplorably high inaccuracy for such an
4 1 Fundamentals of potential theory
important constant, see http://physics.nist.gov/cuu/constants. (For other

constants we have a relative accuracy of 10−7 and better.) For comparison
of experimental results see the internet.
Although the masses m1 , m2 attract each other in a completely symmet-
rical way, it is convenient to call one of them the attracting mass and the
other the attracted mass. For simplicity we set the attracted mass equal to
unity and denote the attracting mass by m. The formula
m
F =G (1–3)
l2
expresses the force exerted by the mass m on a unit mass located at P at a

distance l from m.
We now introduce a rectangular coordinate system xyz and denote the
coordinates of the attracting mass m by ξ, η, ζ and the coordinates of the
attracted point P by x, y, z. The force may be represented by a vector F
with magnitude F (Fig. 1.1). The components of F are given by
l F
¯ Y || y
m y–´
z
P (x,y,z)
|| z
l
F
° z–³
m (»,´,³) ¯ Z
|| y
® X x–»
|| x y
x
Fig. 1.1. The components of the gravitational force; upper figure
shows y-component
1.1 Attraction and potential 5
Gm x − ξ x−ξ
X = −F cos α = − 2
= −G m 3 ,
l l l
Gm y − η y−η
Y = −F cos β = − = −G m 3 , (1–4)
l2 l l
Gm z − ζ z−ζ
Z = −F cos γ = − 2
= −G m 3 ,
l l l
where
l= (x − ξ)2 + (y − η)2 + (z − ζ)2 . (1–5)
We next introduce a scalar function
Gm
V = , (1–6)
l
called the potential of gravitation. The components X, Y, Z of the gravita-
tional force F are then given by
∂V ∂V ∂V
X= , Y = , Z= , (1–7)
∂x ∂y ∂z
as can be easily verified by differentiating (1–6), since

∂ 1 1 ∂l 1 x−ξ x−ξ
=− 2 =− 2 = − 3 ,... . (1–8)
∂x l l ∂x l l l
In vector notation, Eq. (1–7) is written
F = [X, Y, Z] = grad V ; (1–9)
that is, the force vector is the gradient vector of the scalar function V .
It is of basic importance that according to (1–7) the three components
of the vector F can be replaced by a single function V . Especially when we
consider the attraction of systems of point masses or of solid bodies, as we
do in geodesy, it is much easier to deal with the potential than with the three
components of the force. Even in these complicated cases the relations (1–7)
are applied; the function V is then simply the sum of the contributions of
the respective particles.
Thus, if we have a system of several point masses m1 , m2 , . . . , mn , the
potential of the system is the sum of the individual contributions (1–6):
mi n
G m1 G m2 G mn
V = + + ··· + =G . (1–10)
l1 l2 ln li
i=1
1.2 Potential of a solid body

Let us now assume that point masses are distributed continuously over a
volume v (Fig. 1.2) with density
dm
= , (1–11)
dv
where dv is an element of volume and dm is an element of mass. Then the
sum (1–10) becomes an integral (Newton’s integral),

dm
V =G =G dv , (1–12)
l l
v v
where l is the distance between the mass element dm = dv and the at-
tracted point P . Denoting the coordinates of the attracted point P by x, y, z
and of the mass element m by ξ, η, ζ, we see that l is again given by (1–5),
and we can write explicitly

(ξ, η, ζ)
V (x, y, z) = G dξ dη dζ , (1–13)
(x − ξ)2 + (y − η)2 + (z − ζ)2
v
since the element of volume is expressed by
dv = dξ dη dζ . (1–14)
This is the reason for the triple integrals in (1–12).
z
P (x,y,z)
l
v
d³
dm (»,´,³)d ´ d »
y
Fig. 1.2. Potential of a solid body

1.2 Potential of a solid body 7
The components of the force of attraction are given by (1–7). For in-
stance,
∂V ∂ (ξ, η, ζ)
X= =G dξ dη dζ
∂x ∂x l
v
(1–15)
∂ 1
=G (ξ, η, ζ) dξ dη dζ .
∂x l
v
Note that we have interchanged the order of differentiation and integration.
Substituting (1–8) into the above expression, we finally obtain

x−ξ
X = −G dv . (1–16)
l3
v
Analogous expressions result for Y and Z.

The potential V is continuous throughout the whole space and vanishes at
infinity like 1/l for l → ∞. This can be seen from the fact that for very large
distances l the body acts approximately like a point mass, with the result
that its attraction is then approximately given by (1–6). Consequently, in
celestial mechanics the planets are usually considered as point masses.
The first derivatives of V , that is, the force components, are also contin-
uous throughout space, but not so the second derivatives. At points where
the density changes discontinuously, some second derivatives have a discon-
tinuity. This is evident because the potential V may be shown to satisfy
Poisson’s equation
∆V = −4π G , (1–17)
where
∂2V ∂2V ∂2V
∆V = + + . (1–18)
∂x2 ∂y 2 ∂z 2
The symbol ∆, called the Laplacian operator, has the form
∂2 ∂2 ∂2
+ + . (1–19)
∂x2 ∂y 2 ∂z 2
From (1–17) and (1–18) we see that at least one of the second derivatives of
V must be discontinuous together with .
Outside the attracting bodies, in empty space, the density is zero and
(1–17) becomes
∆V = 0 . (1–20)
This is Laplace’s equation. Its solutions are called harmonic functions. Hence,
the potential of gravitation is a harmonic function outside the attracting
masses but not inside the masses: there it satisfies Poisson’s equation.
1.3 Harmonic functions

Earlier we have defined the harmonic functions as solutions of Laplace’s
equation
∆V = 0 . (1–21)
More precisely, a function is called harmonic in a region v of space if it
satisfies Laplace’s equation at every point of v. If the region is the exterior
of a certain closed surface S, then it must in addition vanish like 1/l for
l → ∞. It can be shown that every harmonic function is analytic (in the
region where it satisfies Laplace’s equation); that is, it is continuous and
has continuous derivatives of any order and can be developed into a Taylor
series.
The simplest harmonic function is the reciprocal distance
1 1
= (1–22)
l (x − ξ) + (y − η)2 + (z − ζ)2
2
between two points P (ξ, η, ζ) and P (x, y, z). It is the potential of a point
mass m = 1/G, situated at the point P (ξ, η, ζ); compare (1–5) and (1–6).
It is easy to show that 1/l is harmonic. We form the following partial
derivatives with respect to x, y, z in the fashion of (1–8):

∂ 1 x−ξ ∂ 1 y−η ∂ 1 z−ζ
=− 3 , =− 3 , =− 3 ;
∂x l l ∂y l l ∂z l l

∂2 1 −l2 + 3(x − ξ)2 ∂2 1 −l2 + 3(y − η)2
= , = , (1–23)
∂x2 l l3 ∂y 2 l l3

∂2 1 −l2 + 3(z − ζ)2
= .
∂z 2 l l3
Adding the last three equations and recalling the definition of ∆, we find

1
∆ = 0; (1–24)
l
that is, 1/l is harmonic.

The point P (ξ, η, ζ), where l is zero and 1/l is infinite, is the only point
to which we cannot apply the above derivation; 1/l is not harmonic at this
singular point.
As a matter of fact, the slightly more general potential (1–6) of an ar-
bitrary point mass m is also harmonic except at P (ξ, η, ζ), because (1–24)
remains unchanged if both sides are multiplied by G m.
1.4 Laplace’s equation in spherical coordinates 9
Not only the potential of a point mass but also any other gravitational
potential is harmonic outside the attracting masses. Consider the potential
(1–12) of an extended body. Interchanging the order of differentiation and
integration, we find from (1–12)

1
∆V = G ∆ dv = G ∆ dv = 0 ; (1–25)
l l
v v
that is, the potential of a solid body is also harmonic at any point P (x, y, z)
outside the attracting masses.
If P lies inside the attracting body, the above derivation breaks down,
since 1/l becomes infinite for that mass element dm(ξ, η, ζ) which coincides
with P (x, y, z), and (1–24) does not apply. This is the reason why the po-
tential of a solid body is not harmonic in its interior but instead satisfies
Poisson’s differential equation (1–17).
1.4 Laplace’s equation in spherical coordinates

The most important harmonic functions are the spherical harmonics. To find
them, we introduce spherical coordinates: r (radius vector; note that this is
a standard notation, although it does not represent a vector in the con-
temporary sense), ϑ (polar distance), λ (geocentric longitude), see Fig. 1.3.
Spherical coordinates are related to rectangular coordinates x, y, z by the
P
#
r
z
#
y
¸ r si
x n#
Fig. 1.3. Spherical and rectangular coordinates

equations
x = r sin ϑ cos λ ,
y = r sin ϑ sin λ , (1–26)
z = r cos ϑ ;
or inversely by

r= x2 + y 2 + z 2 ,

x2 + y 2
ϑ = tan−1 , (1–27)
z
y
λ = tan−1 .
x
To get Laplace’s equation in spherical coordinates, we first determine the
element of arc (element of distance) ds in these coordinates. For this purpose
we form
∂x ∂x ∂x
dx = dr + dϑ + dλ ,
∂r ∂ϑ ∂λ
∂y ∂y ∂y
dy = dr + dϑ + dλ , (1–28)
∂r ∂ϑ ∂λ
∂z ∂z ∂z
dz = dr + dϑ + dλ .
∂r ∂ϑ ∂λ
By differentiating (1–26) and substituting it into the elementary formula
ds2 = dx2 + dy 2 + dz 2 , (1–29)
we obtain
ds2 = dr 2 + r 2 dϑ2 + r 2 sin2 ϑ dλ2 . (1–30)
We might have found this well-known formula more simply by geometrical

considerations, but the approach used here is more general and can also be
applied to ellipsoidal (harmonic) coordinates.
In (1–30) there are no terms with dr dϑ, dr dλ, and dϑ dλ. This expresses
the evident fact that spherical coordinates are orthogonal: the spheres r =
constant, the cones ϑ = constant, and the planes λ = constant intersect each
other orthogonally.
The general form of the element of arc in arbitrary orthogonal coordinates
q1 , q2 , q3 is
ds2 = h21 dq12 + h22 dq22 + h23 dq32 . (1–31)
1.5 Spherical harmonics 11
It can be shown that Laplace’s operator in these coordinates is

1 ∂ h2 h3 ∂V ∂ h3 h1 ∂V ∂ h1 h2 ∂V
∆V = + + .
h1 h2 h3 ∂q1 h1 ∂q1 ∂q2 h2 ∂q2 ∂q3 h3 ∂q3
(1–32)
For spherical coordinates we have q1 = r, q2 = ϑ, q3 = λ. Comparison of
(1–30) and (1–31) shows that
h1 = 1 , h2 = r , h3 = r sin ϑ . (1–33)
Substituting these relations into (1–32) yields

1 ∂ 2 ∂V 1 ∂ ∂V 1 ∂2V
∆V = 2 r + 2 sin ϑ + 2 2 . (1–34)
r ∂r ∂r r sin ϑ ∂ϑ ∂ϑ r sin ϑ ∂λ2
Performing the differentiations we find
∂2V 2 ∂V 1 ∂2V cot ϑ ∂V 1 ∂2V

∆V ≡ + + + + = 0, (1–35)
∂r 2 r ∂r r 2 ∂ϑ2 r 2 ∂ϑ r 2 sin2 ϑ ∂λ2
which is Laplace’s equation in spherical coordinates. An alternative expres-
sion is obtained when multiplying both sides by r 2 :
∂2V ∂V ∂2V ∂V 1 ∂2V

r2 + 2r + + cot ϑ + = 0. (1–36)
∂r 2 ∂r ∂ϑ2 ∂ϑ sin2 ϑ ∂λ2
This form will be somewhat more convenient for our subsequent develop-
ment.
1.5 Spherical harmonics

We attempt to solve Laplace’s equation (1–35) or (1–36) by separating the
variables r, ϑ, λ using the trial substitution
V (r, ϑ, λ) = f (r) Y (ϑ, λ) , (1–37)
where f is a function of r only and Y is a function of ϑ and λ only. Performing

this substitution in (1–36) and dividing by f Y , we get

1 2 1 ∂2Y ∂Y 1 ∂2Y
(r f + 2r f ) = − + cot ϑ + , (1–38)
f Y ∂ϑ2 ∂ϑ sin2 ϑ ∂λ2
where the primes denote differentiation with respect to the argument (r, in
this case). Since the left-hand side depends only on r and the right-hand side
only on ϑ and λ, both sides must be constant. We can therefore separate the
equation into two equations:
r 2 f (r) + 2r f (r) − n(n + 1) f (r) = 0 (1–39)
and
∂2Y ∂Y 1 ∂2Y
+ cot ϑ + + n(n + 1) Y = 0 , (1–40)
∂ϑ2 ∂ϑ sin2 ϑ ∂λ2
where we have denoted the constant by n(n + 1).
Solutions of (1–39) are given by the functions
f (r) = r n and f (r) = r −(n+1) ; (1–41)
this should be verified by substitution. Denoting the still unknown solutions

of (1–40) by Yn (ϑ, λ), we see that Laplace’s equation (1–35) is solved by the
functions
Yn (ϑ, λ)
V = r n Yn (ϑ, λ) and V = . (1–42)
r n+1
These functions are called solid spherical harmonics, whereas the functions
Yn (ϑ, λ) are known as (Laplace’s) surface spherical harmonics. Both kinds
are called spherical harmonics; the kind referred to can usually be judged
from the context.
Note that n is not an arbitrary constant but must be an integer 0, 1, 2, . . .
as we will see later. If a differential equation is linear, and if we know several
solutions, then, as is well known, the sum of these solutions is also a solution
(this holds for all linear equation systems!). Hence, we conclude that
∞
∞
Yn (ϑ, λ)
V = r n Yn (ϑ, λ) and V = (1–43)
r n+1
n=0 n=0
are also solutions of Laplace’s equation ∆V = 0; that is, harmonic functions.

The important fact is that every harmonic function – with certain restrictions
– can be expressed in one of the forms (1–43).
1.6 Surface spherical harmonics

Now we have to determine Laplace’s surface spherical harmonics Yn (ϑ, λ).
We attempt to solve (1–40) by a new trial substitution
Yn (ϑ, λ) = g(ϑ) h(λ) , (1–44)

1.6 Surface spherical harmonics 13
where the functions g and h individually depend on one variable only. Per-
forming this substitution in (1–40) and multiplying by sin2 ϑ/g h, we find
sin ϑ h
sin ϑ g + cos ϑ g + n(n + 1) sin ϑ g = − , (1–45)
g h
where the primes denote differentiation with respect to the argument: ϑ in

g and λ in h. The left-hand side is a function of ϑ only, and the right-hand
side is a function of λ only. Therefore, both sides must again be constant;
let the constant be m2 . Thus, the partial differential equation (1–40) splits
into two ordinary differential equations for the functions g(ϑ) and h(λ):

m2
sin ϑ g (ϑ) + cos ϑ g (ϑ) + n(n + 1) sin ϑ − g(ϑ) = 0 ; (1–46)
sin ϑ
h (λ) + m2 h(λ) = 0 . (1–47)

Solutions of Eq. (1–47) are the functions
h(λ) = cos mλ and h(λ) = sin mλ , (1–48)
as may be verified by substitution. Equation (1–46), Legendre’s differential

equation, is more difficult. It can be shown that it has physically meaningful
solutions only if n and m are integers 0, 1, 2, . . . and if m is smaller than or
equal to n. A solution of (1–46) is the Legendre function Pnm (cos ϑ), which
will be considered in some detail in the next section. Therefore,
g(ϑ) = Pnm (cos ϑ) (1–49)
and the functions
Yn (ϑ, λ) = Pnm (cos ϑ) cos mλ and Yn (ϑ, λ) = Pnm (cos ϑ) sin mλ (1–50)
are solutions of the differential equation (1–40) for Laplace’s surface spherical
harmonics.
Since these solutions are linear, any linear combination of the solutions
(1–50) is also a solution. Such a linear combination has the general form
n

Yn (ϑ, λ) = [anm Pnm (cos ϑ) cos mλ + bnm Pnm (cos ϑ) sin mλ] , (1–51)
m=0
where anm and bnm are arbitrary constants. This is the general expression
for the surface spherical harmonics Yn (ϑ, λ).
Substituting this relation into equations (1–43), we see that

∞
n

n
Vi (r, ϑ, λ) = r [anm Pnm (cos ϑ) cos mλ + bnm Pnm (cos ϑ) sin mλ] ,
n=0 m=0
(1–52)
∞
n

1
Ve (r, ϑ, λ) = [anm Pnm (cos ϑ) cos mλ + bnm Pnm (cos ϑ) sin mλ]
r n+1
n=0 m=0
(1–53)
are solutions of Laplace’s equation ∆V = 0; that is, harmonic functions.
Furthermore, as we have mentioned, they are very general solutions indeed:
every function which is harmonic inside a certain sphere can be expanded
into a series (1–52), where the subscript i indicates the interior, and every
function which is harmonic outside a certain sphere (such as the earth’s
gravitational potential) can be expanded into a series (1–53), where the
subscript e indicates the exterior. Thus, we see how spherical harmonics can
be useful in geodesy.
1.7 Legendre’s functions

In the preceding section we have introduced Legendre’s function Pnm (cos ϑ)
as a solution of Legendre’s differential equation (1–46). The subscript n
denotes the degree and the subscript m the order of Pnm .
It is convenient to transform Legendre’s differential equation (1–46) by
the substitution
t = cos ϑ . (1–54)
In order to avoid confusion, we use an overbar to denote g as a function of
t. Therefore,
g(ϑ) = ḡ(t) ,
dg dg dt
g (ϑ) = = = −ḡ (t) sin ϑ , (1–55)
dϑ dt dϑ
g (ϑ) = ḡ (t) sin2 ϑ − ḡ (t) cos ϑ .
Inserting these relations into (1–46), dividing by sin ϑ, and then substituting
sin2 ϑ = 1 − t2 , we get

m2
(1 − t ) ḡ (t) − 2t ḡ (t) + n(n + 1) −
2
ḡ (t) = 0 . (1–56)
1 − t2
The Legendre function ḡ(t) = Pnm (t), which is defined by
1 n+m
2 m/2 d
Pnm (t) = (1 − t ) (t2 − 1)n , (1–57)
2n n! dtn+m
1.7 Legendre’s functions 15
satisfies (1–56). Apart from the factor (1 − t2 )m/2 = sinm ϑ and from a
constant, the function Pnm is the (n + m)th derivative of the polynomial
(t2 − 1)n . It can, thus, be evaluated. For instance,
(1 − t2 )1/2 d2 2 1
P11 (t) = (t − 1) = 1 − t 2·2= 1 − t2 = sin ϑ . (1–58)
2·1 dt2 2
The case m = 0 is of particular importance. The functions Pn0 (t) are often
simply denoted by Pn (t). Then (1–57) gives
1 dn 2
Pn (t) = Pn0 (t) = (t − 1)n . (1–59)
2n n! dtn
Because m = 0, there is no square root, that is, no sin ϑ. Therefore, the
Pn (t) are simply polynomials in t. They are called Legendre’s polynomials.
We give the Legendre polynomials for n = 0 through n = 5:
P0 (t) = 1 , P3 (t) = 52 t3 − 3
2 t,
P1 (t) = t , P4 (t) = 35 4
8 t − 15 2
4 t + 38 , (1–60)
P2 (t) = 32 t2 − 1
2 , P5 (t) = 63 5
8 t − 35 3
4 t + 15
8 t.
Remember that
t = cos ϑ . (1–61)
The polynomials may be obtained by means of (1–59) or more simply by the
recursion formula
n−1 2n − 1
Pn (t) = − Pn−2 (t) + t Pn−1 (t) , (1–62)
n n
by which P2 can be calculated from P0 and P1 , P3 from P1 and P2 , etc.
Graphs of the Legendre polynomials are shown in Fig. 1.4.
The powers of cos ϑ can be expressed in terms of the cosines of multiples
of ϑ, such as
cos2 ϑ = 1
2 cos 2ϑ + 1
2 , cos3 ϑ = 1
4 cos 3ϑ + 3
4 cos ϑ . (1–63)
Therefore, we may also express the Pn (cos ϑ) in this way, obtaining
P2 (cos ϑ) = 34 cos 2ϑ + 14 ,
P3 (cos ϑ) = 58 cos 3ϑ + 38 cos ϑ ,
P4 (cos ϑ) = 35
cos 4ϑ + 5
cos 2ϑ + 9
, (1–64)
64 16 64
63 35 15
P5 (cos ϑ) = 128 cos 5ϑ + 128 cos 3ϑ + 64 cos ϑ ,
··· = ··· .
P0
1.0
Pn(t)
0.5
P6 P6
0
–1.0 –0.5 0.5 1.0
P4 –0.5 P4
P2
–1.0
t = cos #
1.0
Pn(t) P1
P3 0.5 P5
P7
0
–1.0 –0.5 0.5 1.0
P7 P5 –0.5
P3
P1
–1.0
t = cos #
Fig. 1.4. Legendre’s polynomials as functions of t = cos ϑ: n even (top)

and n odd (bottom)
If the order m is not zero – that is, for m = 1, 2, . . . , n – Legendre’s functions

Pnm (cos ϑ) are called associated Legendre functions. They can be reduced to
the Legendre polynomials by means of the equation
dm Pn (t)
Pnm (t) = (1 − t2 )m/2 , (1–65)
dtm
which follows from (1–57) and (1–59). Thus, the associated Legendre func-
tions are expressed in terms of the Legendre
√ polynomials of the same degree
n. We give some Pnm , writing t = cos ϑ, 1 − t2 = sin ϑ:

2 cos ϑ − 2 ,
P31 (cos ϑ) = sin ϑ 15 2 3
P11 (cos ϑ) = sin ϑ ,
P21 (cos ϑ) = 3 sin ϑ cos ϑ , P32 (cos ϑ) = 15 sin2 ϑ cos ϑ , (1–66)
P22 (cos ϑ) = 3 sin2 ϑ , P33 (cos ϑ) = 15 sin3 ϑ .
We also mention an explicit formula for any Legendre function (polynomial

1.7 Legendre’s functions 17
or associated function):
r
(2n − 2k)!
Pnm (t) = 2−n (1 − t2 )m/2 (−1)k tn−m−2k ,
k! (n − k)! (n − m − 2k)!
k=0
(1–67)
where r is the greatest integer ≤ (n−m)/2; i.e., r is (n−m)/2 or (n−m−1)/2,
whichever is an integer. This formula is convenient for programming.
As this useful formula is seldom found in the literature, we show the
derivation, which is quite straightforward. The necessary information on fac-
torials may be obtained from any collection of mathematical formulas. The
binomial theorem gives
n n
n k n 2n−2k n!
(t − 1) =
2
(−1) t = (−1)k t2n−2k . (1–68)
k k! (n − k)!
k=0 k=0
Thus, (1–57) becomes

n
1 1 dn+m 2n−2k
Pnm (t) = n (1 − t2 )m/2 (−1)k (t ), (1–69)
2 k! (n − k)! dtn+m
k=0
the quantity n! having been cancelled out. The rth derivative of the power
ts is
dr s s!
r
(t ) = s(s − 1) · · · (s − r + 1) ts−r = ts−r . (1–70)
dt (s − r)!
Setting r = n + m and s = 2n − 2k, we have
dn+m 2n−2k (2n − 2k)! n−m−2k

(t )= t . (1–71)
dtn+m (n − m − 2k)!
Inserting this into the above expression for Pnm (t) and noting that the lowest
possible power of t is either t or t0 = 1, we obtain (1–67).
The surface spherical harmonics are Legendre’s functions multiplied by
cos mλ or sin mλ:
degree 0 P0 (cos ϑ) ;
degree 1 P1 (cos ϑ) ,
P11 (cos ϑ) cos λ , P11 (cos ϑ) sin λ ;
(1–72)
degree 2 P2 (cos ϑ) ,
P21 (cos ϑ) cos λ , P21 (cos ϑ) sin λ ,
P22 (cos ϑ) cos 2λ , P22 (cos ϑ) sin 2λ ;
and so on.
The geometrical representation of these spherical harmonics is useful.
The harmonics with m = 0 – that is, Legendre’s polynomials – are polyno-
mials of degree n in t, so that they have n zeros. These n zeros are all real
and situated in the interval −1 ≤ t ≤ +1, that is, 0 ≤ ϑ ≤ π (Fig. 1.4).
Therefore, the harmonics with m = 0 change their sign n times in this inter-
val; furthermore, they do not depend on λ. Their geometrical representation
is therefore similar to Fig. 1.5 a. Since they divide the sphere into zones, they
are also called zonal harmonics.
The associated Legendre functions change their sign n − m times in the
interval 0 ≤ ϑ ≤ π. The functions cos mλ and sin mλ have 2m zeros in the
interval 0 ≤ λ < 2π, so that the geometrical representation of the harmonics
for m = 0 is similar to that of Fig. 1.5 b. They divide the sphere into com-
partments in which they are alternately positive and negative, somewhat like
a chess board, and are called tesseral harmonics. “Tessera” means a square
or rectangle, or also a tile. In particular, for n = m, they degenerate into
functions that divide the sphere into positive and negative sectors, in which
case they are called sectorial harmonics, see Fig. 1.5 c.
P6(cos #)
P12,6(cos #)cos 6¸ P6,6(cos #)cos 6¸

(a)
(b) (c)
Fig. 1.5. The kinds of spherical harmonics: (a) zonal, (b) tesseral, (c) sectorial
1.8 Legendre’s functions of the second kind 19
1.8 Legendre’s functions of the second kind

The Legendre function Pnm (t) is not the only solution of Legendre’s differ-
ential equation (1–56). There is a completely different function which also
satisfies this equation. It is called Legendre’s function of the second kind, of
degree n and order m, and denoted by Qnm (t).
Although the Qnm (t) are functions of a completely different nature, they
satisfy relationships very similar to those satisfied by the Pnm (t).
The “zonal” functions
Qn (t) ≡ Qn0 (t) (1–73)
are defined by
n
1 1+t 1
Qn (t) = Pn (t) ln − Pk−1 (t) Pn−k (t) , (1–74)
2 1−t k
k=1
and the others by
dm Qn (t)
Qnm (t) = (1 − t2 )m/2 . (1–75)
dtm
Equation (1–75) is completely analogous to (1–65); furthermore, the func-

tions Qn (t) satisfy the same recursion formula (1–62) as the functions Pn (t).
If we evaluate the first few Qn , from (1–74) we find
1 1+t
Q0 (t) = ln = tanh−1 t ,
2 1−t
t 1+t
Q1 (t) = ln − 1 = t tanh−1 t − 1 , (1–76)
2 1−t

3 2 1 1+t 3 3 2 1 3
Q2 (t) = t − ln − t= t − tanh−1 t − t .
4 4 1−t 2 2 2 2
These formulas and Fig. 1.6 show that the functions Qnm are really
quite different from the functions Pnm . From the singularity ±∞ at t = ±1
(i.e., ϑ = 0 or π), we see that it is impossible to substitute Qnm (cos ϑ) for
Pnm (cos ϑ) if ϑ means the polar distance, because harmonic functions must
be regular.
However, we will encounter them in the theory of ellipsoidal harmon-
ics (Sect. 1.16), which is applied to the normal gravity field of the earth
(Sect. 2.7). For this purpose we need Legendre’s functions of the second
1.0
Q n ( t) Q0
Q2 0.5
0
–1.0 –0.5 0.5 1.0
–0.5
Q4
–1.0
t = cos #
1.0
Q n ( t)
Q3
Q5 0.5
0
–1.0 –0.5 0.5 1.0
–0.5
Q1
–1.0
t = cos #
Fig. 1.6. Legendre’s functions of the second kind: n even (top) and n odd (bottom)
kind as functions of a complex argument. If the argument z is complex, we

must replace the definition (1–74) by
n
1 z+1 1
Qn (z) = Pn (z) ln − Pk−1 (z) Pn−k (z) , (1–77)
2 z−1 k
k=1
where Legendre’s polynomials Pn (z) are defined by the same formulas as in

the case of a real argument t. Therefore, the only change as compared to
(1–74) is the replacement of
1 1+t
ln = tanh−1 t (1–78)
2 1−t
by
1 z+1
ln = coth−1 z . (1–79)
2 z−1
1.9 Expansion theorem and orthogonality relations 21
In particular, we have
1 z+1
Q0 (z) = ln = coth−1 z ,
2 z−1
z z+1
Q1 (z) = ln − 1 = z coth−1 z − 1 , (1–80)
2 z−1

3 2 1 z+1 3 3 2 1 3
Q2 (z) = z − ln − z= z − coth−1 z − z .
4 4 z−1 2 2 2 2
1.9 Expansion theorem and orthogonality relations

In (1–52) and (1–53), we have expanded harmonic functions in space into
series of solid spherical harmonics. In a similar way an arbitrary (at least in
a very general sense) function f (ϑ, λ) on the surface of the sphere can be
expanded into a series of surface spherical harmonics:
∞
∞
n
f (ϑ, λ) = Yn (ϑ, λ) = [anm Rnm (ϑ, λ) + bnm Snm (ϑ, λ)] , (1–81)
n=0 n=0 m=0
where we have introduced the abbreviations

Rnm (ϑ, λ) = Pnm (cos ϑ) cos mλ ,
(1–82)
Snm (ϑ, λ) = Pnm (cos ϑ) sin mλ .
The symbols anm and bnm are constant coefficients, which we will now
determine. Essential for this purpose are the orthogonality relations. These
remarkable relations mean that the integral over the unit sphere of the prod-
uct of any two different functions Rnm or Snm is zero:
⎫
Rnm (ϑ, λ) Rsr (ϑ, λ) dσ = 0 ⎪
⎬
σ
if s = n or r = m or both ;
Snm (ϑ, λ) Ssr (ϑ, λ) dσ = 0 ⎪
⎭ (1–83)
σ

Rnm (ϑ, λ) Ssr (ϑ, λ) dσ = 0 in any case .
σ
For the product of two equal functions Rnm or Snm , we have

4π
[Rn0 (ϑ, λ)]2 dσ = ;
2n + 1
σ

2π (n + m)!
2
[Rnm (ϑ, λ)] dσ = [Snm (ϑ, λ)]2 dσ = (m = 0) .
2n + 1 (n − m)!
σ σ
(1–84)
Note that there is no Sn0 , since sin 0λ = 0. In these formulas we have used
the abbreviation 2π π
= (1–85)
λ=0 ϑ=0
σ
for the integral over the unit sphere. The expression
dσ = sin ϑ dϑ dλ (1–86)
denotes the surface element of the unit sphere.
Now we turn to the determination of the coefficients anm and bnm in
(1–81). Multiplying both sides of the equation by a certain Rsr (ϑ, λ) and
integrating over the unit sphere gives

f (ϑ, λ) Rsr (ϑ, λ) dσ = asr [Rsr (ϑ, λ)]2 dσ , (1–87)
σ σ
since in the double integral on the right-hand side all terms except the one
with n = s, m = r will vanish according to the orthogonality relations (1–
83). The integral on the right-hand side has the value given in (1–84), so
that asr is determined. In a similar way we find bsr by multiplying (1–81)
by Ssr (ϑ, λ) and integrating over the unit sphere. The result is

2n + 1
an0 = f (ϑ, λ) Pn (cos ϑ) dσ ;
4π
σ
⎫
2n + 1 (n − m)! ⎪
anm = f (ϑ, λ) Rnm (ϑ, λ) dσ ⎪
⎪
⎪ (1–88)
2π (n + m)! ⎪
⎬
σ
(m = 0) .
2n + 1 (n − m)! ⎪
⎪
bnm = f (ϑ, λ) Snm (ϑ, λ) dσ ⎪
⎪
⎪
2π (n + m)! ⎭
σ
The coefficients anm and bnm can, thus, be determined by integration.
We note that the Laplace spherical harmonics Yn (ϑ, λ) in (1–81) may
also be found directly by the formula
π
2n + 1 2π
Yn (ϑ, λ) = f (ϑ , λ ) Pn (cos ψ) sin ϑ dϑ dλ , (1–89)
4π λ =0 ϑ =0
where ψ is the spherical distance between the points P , represented by ϑ, λ,
and P , represented by ϑ , λ (Fig. 1.7), so that
cos ψ = cos ϑ cos ϑ + sin ϑ sin ϑ cos(λ − λ) . (1–90)

Later, when being acquainted with Sect. 1.11, Eq. (1–89) may be verified by
straightforward computation, substituting Pn (cos ψ) from the decomposition
formula (1–105).
1.10 Fully normalized spherical harmonics 23
PN
¸' – ¸
#
#'
P
Ã P'
Fig. 1.7. Spherical distance ψ
1.10 Fully normalized spherical harmonics

The formulas of the preceding section for the expansion of a function into a
series of surface spherical harmonics are somewhat inconvenient to handle.
If we look at equations (1–84) and (1–88), we see that there are different
formulas for m = 0 and m = 0; furthermore, the expressions are rather
complicated and difficult to remember.
Therefore, it has been proposed that the “conventional” harmonics Rnm
and Snm , defined by (1–82) together with (1–57), be replaced by other func-
tions which differ by a constant factor and are easier to handle. We consider
here only the fully normalized harmonics, which seem to be the most conve-
nient and the most widely used.
The “fully normalized” harmonics are simply “normalized” in the sense
of the theory of real functions; we have to use this clumsy expression because
the term “normalized spherical harmonics” has already been used for other
functions, unfortunately often for some that are not “normalized” at all in
the mathematical sense.
We denote the fully normalized harmonics by R̄nm and S̄nm ; they are
defined by
√ √
R̄n0 (ϑ, λ) = 2n + 1 Rn0 (ϑ, λ) ≡ 2n + 1 Pn (cos ϑ) ;
⎫
(n − m)! ⎪
⎪
R̄nm (ϑ, λ) = Rnm (ϑ, λ) ⎪
⎪
2(2n + 1) ⎪
⎪ (1–91)
(n + m)! ⎬
(m = 0) .
⎪
⎪
(n − m)! ⎪
⎪
S̄nm (ϑ, λ) = 2(2n + 1) Snm (ϑ, λ) ⎪
⎪
⎭
(n + m)!
The orthogonality relations (1–83) also apply for these fully normalized har-
monics, whereas Eqs. (1–84) are thoroughly simplified: they become

1 1
R̄nm dσ =
2
S̄nm
2
dσ = 1 . (1–92)
4π 4π
σ σ
This means that the average square of any fully normalized harmonic is
unity, the average being taken over the sphere (the average corresponds to
the integral divided by the area 4π). This formula now applies for any m,
whether it is zero or not.
If we expand an arbitrary function f (ϑ, λ) into a series of fully normalized
harmonics, analogously to (1–81),
∞
n
f (ϑ, λ) = [ānm R̄nm (ϑ, λ) + b̄nm S̄nm (ϑ, λ)] , (1–93)
n=0 m=0
then the coefficients ānm , b̄nm are simply given by

1
ānm = f (ϑ, λ) R̄nm (ϑ, λ) dσ ,
4π
σ
(1–94)
1
b̄nm = f (ϑ, λ) S¯nm (ϑ, λ) dσ ;
4π
σ
that is, the coefficients are the average products of the function and the
corresponding harmonic R̄nm or S̄nm .
The simplicity of formulas (1–92) and (1–94) constitutes the main ad-
vantage of the fully normalized spherical harmonics and makes them useful
in many respects, even though the functions R̄nm and S̄nm in (1–91) are a
little more complicated than the conventional Rnm and Snm . We have
R̄nm (ϑ, λ) = P̄nm (cos ϑ) cos mλ ,
(1–95)
S̄nm (ϑ, λ) = P̄nm (cos ϑ) sin mλ ,
where
r

√ (2n − 2k)!
P̄n0 (t) = 2n + 1 2−n (−1)k tn−2k (1–96)
k! (n − k)! (n − 2k)!
k=0
for m = 0, and

(n − m)! −n
P̄nm (t) = 2(2n + 1) 2 (1 − t2 )m/2 ·
(n + m)!
(1–97)
r
(2n − 2k)!
(−1)k tn−m−2k
k! (n − k)! (n − m − 2k)!
k=0
1.11 Zonal harmonics and decomposition formula 25
for m = 0. This corresponds to (1–67); here, as in (1–67), r is the greatest

integer ≤ (n − m)/2.
There are relations between the coefficients ānm and b̄nm for fully normal-
ized harmonics and the coefficients anm and bnm for conventional harmonics
that are inverse to those in (1–91):
an0
ān0 = √ ;
2n + 1
⎫
1 (n + m)! ⎪
⎪
⎪
⎪
ānm =
2(2n + 1) (n − m)!
anm ⎪
⎪ (1–98)
⎬
(m = 0) .
⎪
⎪
1 (n + m)! ⎪
⎪
b̄nm = bnm ⎪
⎪
2(2n + 1) (n − m)! ⎭
1.11 Expansion of the reciprocal distance into zonal

harmonics and decomposition formula
The distance l between two points with spherical coordinates
P (r, ϑ, λ), P (r , ϑ , λ ) (1–99)
is given by
l2 = r 2 + r 2 − 2r r cos ψ , (1–100)
where ψ is the angle between the radius vectors r and r (Fig. 1.8), so that,
from (1–90),
cos ψ = cos ϑ cos ϑ + sin ϑ sin ϑ cos(λ − λ) (1–101)
l
r
P'
r'
Ã
O
Fig. 1.8. The spatial distance l

results. Assuming r < r, we may write
1 1 1
= = √ , (1–102)
l
r − 2r r cos ψ + r
2 2 r 1 − 2α u + α2
where we have put α = r /r and u = cos ψ. If r < r, this can be expanded

into a power series with respect to α. It is remarkable that the coefficients
of αn are the (conventional) zonal harmonics, or Legendre’s polynomials
Pn (u) = Pn (cos ψ):
∞
1
√ = αn Pn (u) = P0 (u)+α P1 (u)+α2 P2 (u)+ · · · . (1–103)
1 − 2α u + α2 n=0
Hence, we obtain
∞
1 r n
= Pn (cos ψ) , (1–104)
l r n+1
n=0
which is an important formula.

It would still be desirable in this equation to express Pn (cos ψ) in terms of
functions of the spherical coordinates ϑ, λ and ϑ , λ of which ψ is composed
according to (1–90). This is achieved by the decomposition formula
Pn (cos ψ) = Pn (cos ϑ) Pn (cos ϑ ) +

n
(n − m)!
2 [Rnm (ϑ, λ)Rnm (ϑ , λ ) + Snm (ϑ, λ)Snm (ϑ , λ )] .
(n + m)!
m=1
(1–105)
Substituting this into (1–104), we obtain
∞ n
(n − m)!
1 Pn (cos ϑ) n
= n+1
r Pn (cos ϑ ) + 2 ·
l r (n + m)!
n=0 m=1
(1–106)

Rnm (ϑ, λ) n Snm (ϑ, λ) n
r Rnm (ϑ , λ ) + r Snm (ϑ , λ ) .
r n+1 r n+1
The use of fully normalized harmonics simplifies these formulas. Replacing

the conventional harmonics in (1–105) and (1–106) by fully normalized har-
monics by means of (1–91), we find
n
1
Pn (cos ψ) = R̄nm (ϑ, λ)R̄nm (ϑ , λ ) + S̄nm (ϑ, λ)S¯nm (ϑ , λ ) ;
2n + 1
m=0
(1–107)
1.12 Solution of Dirichlet’s problem 27
∞
n
1 1
= ·
l n=0 m=0
2n + 1
(1–108)

R̄nm (ϑ, λ) n S̄nm (ϑ, λ) n
r R̄nm (ϑ , λ ) + r S̄nm (ϑ , λ ) .
r n+1 r n+1
The last formula will be fundamental for the expansion of the earth’s gravi-
tational field in spherical harmonics.
1.12 Solution of Dirichlet’s problem by means of

spherical harmonics and Poisson’s integral
We define Dirichlet’s problem, or the first boundary-value problem of potential
theory, as follows: Given is an arbitrary function on a surface S, to determine
is a function V which is harmonic either inside or outside S and which
assumes on S the values of the prescribed function.
If the surface S is a sphere, then Dirichlet’s problem can be solved by
means of spherical harmonics. Let us take first the unit sphere, r = 1, and
expand the prescribed function, given on the unit sphere and denoted by
V (1, ϑ, λ), into a series of surface spherical harmonics (1–81):
∞

V (1, ϑ, λ) = Yn (ϑ, λ) , (1–109)
n=0
the Yn (ϑ, λ) being determined by (1–89). (This series converges for very
general functions V .) The functions
∞

Vi (r, ϑ, λ) = r n Yn (ϑ, λ) (1–110)
n=0
and
∞
Yn (ϑ, λ)
Ve (r, ϑ, λ) = (1–111)
n=0
r n+1
assume the given values V (1, ϑ, λ) on the surface r = 1. The series (1–109)
converges, and we have for r < 1
r n Yn < Yn (1–112)
and for r > 1

Yn
< Yn . (1–113)
r n+1
Hence, the series (1–110) converges for r ≤ 1, and the series (1–111) con-
verges for r ≥ 1; furthermore, both series have been found to represent
harmonic functions. Therefore, we see that Dirichlet’s problem is solved by
Vi (r, ϑ, λ) for the interior of the sphere r = 1, and by Ve (r, ϑ, λ) for its exte-
rior.
For a sphere of arbitrary radius r = R, the solution is similar. We expand
the given function
∞

V (R, ϑ, λ) = Yn (ϑ, λ) . (1–114)
n=0
The surface spherical harmonics Yn are determined by
π
2n + 1 2π
Yn (ϑ, λ) = V (R, ϑ , λ ) Pn (cos ψ) sin ϑ dϑ dλ . (1–115)
4π
λ =0 ϑ =0
Then the series
∞
r n
Vi (r, ϑ, λ) = Yn (ϑ, λ) (1–116)
R
n=0
solves the first boundary-value problem for the interior, and the series
∞ n+1
R
Ve (r, ϑ, λ) = Yn (ϑ, λ) (1–117)
r
n=0
solves it for the exterior of the sphere r = R.

Thus, we see that Dirichlet’s problem can always be solved for the sphere.
It is evident that this is closely related to the possibility of expanding an
arbitrary function on the sphere into a series of surface spherical harmonics
and a harmonic function in space into a series of solid spherical harmonics.
Dirichlet’s boundary-value problem can be solved not only for the sphere
but also for any sufficiently smooth boundary surface. An example is given
in Sect. 1.16.
The solvability of Dirichlet’s problem is also essential to Molodensky’s
problem (Sect. 8.3). See also Kellogg (1929: Chap. XI).
Poisson’s integral
A more direct solution is obtained as follows. We consider only the exterior
problem, which is of greater interest in geodesy. Substituting Yn (ϑ, λ) from
(1–89) into (1–117), we obtain
Ve (r, ϑ, λ) =
∞ n+1 π
R 2n + 1 2π
V (R, ϑ , λ ) Pn (cos ψ) sin ϑ dϑ dλ .
r 4π
λ =0 ϑ =0
n=0
(1–118)
1.13 Other boundary-value problems 29
We can rearrange this as

2π π
1
Ve (r, ϑ, λ) = V (R, ϑ , λ ) ·
4π λ =0 ϑ =0
∞ n+1 (1–119)
R
(2n + 1) Pn (cos ψ) sin ϑ dϑ dλ .
n=0
r
The sum in the brackets can be evaluated. We denote the spatial distance
between the points P (r, ϑ, λ) and P (R, ϑ , λ ) by l. Then, using (1–104),
∞ n+1
1 1 1 R
= = Pn (cos ψ) (1–120)
l r 2 + R2 − 2R r cos ψ R
n=0
r
results. Differentiating with respect to r, we get

∞
r − R cos ψ 1 Rn+1
− = − (n + 1) Pn (cos ψ) . (1–121)
l3 R r n+2
n=0
Multiplying this equation by −2R r, multiplying the expression for 1/l by

−R, and then adding the two equations yields
∞ n+1
R(r 2 − R2 ) R
= (2n + 1) Pn (cos ψ) . (1–122)
l3 n=0
r
The right-hand side is the bracketed expression in (1–119). Substituting the

left-hand side, we finally obtain
π
R(r 2 − R2 ) 2π V (R, ϑ , λ )
Ve (r, ϑ, λ) = 3
sin ϑ dϑ dλ , (1–123)
4π
λ =0 ϑ =0 l
where
l= r 2 + R2 − 2R r cos ψ . (1–124)
This is Poisson’s integral. It is an explicit solution of Dirichlet’s problem for
the exterior of the sphere, which has many applications in physical geodesy.
1.13 Other boundary-value problems

There are other similar boundary-value problems. In Neumann’s problem, or
the second boundary-value problem of potential theory, the normal derivative
∂V /∂n is given on the surface S, instead of the function V itself. The normal
derivative is the derivative along the outward-directed surface normal n to
S. In the third boundary-value problem, a linear combination of V and of its

normal derivative
∂V
hV + k (1–125)
∂n
is given on S.
For the sphere, the solution of these boundary-value problems is also
easily expressed in terms of spherical harmonics. We consider the exterior
problems only, because these are of special interest to geodesy.
In Neumann’s problem, we expand the given values of ∂V /∂n on the
sphere r = R into a series of surface spherical harmonics:
∞

∂V
= Yn (ϑ, λ) . (1–126)
∂n r=R n=0
The harmonic function which solves Neumann’s problem for the exterior of
the sphere is then
∞ n+1
R Yn (ϑ, λ)
Ve (r, ϑ, λ) = −R . (1–127)
r n+1
n=0
To verify it, we differentiate with respect to r, getting
∞ n+2
∂Ve R
= Yn (ϑ, λ) . (1–128)
∂r n=0
r
Since for the sphere the normal coincides with the radius vector, we have

∂V ∂V
= , (1–129)
∂n r=R ∂r r=R
and we see that (1–126) is satisfied.

The third boundary-value problem is particularly relevant to physical
geodesy, because the determination of the undulations of the geoid from
gravity anomalies is just such a problem. To solve the general case, we again
expand the function defined by the given boundary values into surface spher-
ical harmonics:
∞
∂V
hV + k = Yn (ϑ, λ) . (1–130)
∂n
n=0
The harmonic function

∞ n+1
R Yn (ϑ, λ)
Ve (r, ϑ, λ) = (1–131)
r h − (k/R)(n + 1)
n=0
1.13 Other boundary-value problems 31
solves the third boundary-value problem for the exterior of the sphere r = R.
The straightforward verification is analogous to the case of (1–127).
In the determination of the geoidal undulations, the constants h, k have
the values
2
h = − , k = −1 , (1–132)
R
so that
∞ n+1
R Yn (ϑ, λ)
Ve (r, ϑ, λ) = R (1–133)
n=0
r n−1
solves the boundary-value problem of physical geodesy.

As we have seen in the preceding section, the first boundary-value prob-
lem can also be solved directly by Poisson’s integral. Similar integral formulas
also exist for the second and the third problem. The integral formula that
corresponds to (1–133) for the boundary-value problem of physical geodesy
is Stokes’ integral, which will be considered in detail in Chap. 2.
Remark on inverse problems

Boundary-value problems give the potential outside the earth, where there
are no masses and where the potential, satisfying Laplace’s equation, is har-
monic. The determination of the potential inside the earth is of a quite
different character since the earth is filled by masses, and the interior po-
tential satisfies Poisson’s rather than Laplace’s equation, as we have seen in
Sect. 1.2. Unfortunately, the density inside the earth is generally unknown.
To see the difficulties of the problem, let us consider Newton’s integral
(1–12). If the interior masses were known, we could easily use this formula
to compute the potential inside (and outside) the earth, in a direct and
straightforward way. The determination of the potential from the masses is
a “direct” problem. The “inverse” problem is to determine the masses from
the potential, finding a solution of Newton’s integral for the density , which
is essentially more difficult.
In fact, it is impossible to determine uniquely the generating masses
from the external potential. This inverse problem of potential theory has no
unique solution. Such inverse problems occur in geophysical prospecting by
gravity measurements: underground masses are inferred from disturbances
of the gravity field. To determine the problem more completely, additional
information is necessary, which is furnished, for example, by geology or by
seismic measurements.
Generally, nowadays we know that many problems in geophysics and
other sciences including medicine (e.g., seismic and medical tomography) are
inverse problems. We cannot pursue this interesting problem here and refer
only to the extensive literature, e.g., the book by Moritz (1995), the inter-
net page www.inas.tugraz.at/forschung/InverseProblems/AngerMoritz.html
or Anger and Moritz (2003).
1.14 The radial derivative of a harmonic function

For later application to problems related with the vertical gradient of gravity,
we will now derive an integral formula for the derivative along the radius
vector r of an arbitrary harmonic function which we denote by V . Such a
harmonic function satisfies Poisson’s integral (1–123):
π
R(r 2 − R2 ) 2π
V (R, ϑ , λ )
V (r, ϑ, λ) = sin ϑ dϑ dλ . (1–134)
4π λ =0 ϑ =0 l3
Forming the radial derivative ∂V /∂r, we note that V (R, ϑ , λ ) does not
depend on r. Thus, we need only to differentiate (r 2 − R2 )/l3 , obtaining
2π π
∂V (r, ϑ, λ) R
= M (r, ψ) V (R, ϑ , λ ) sin ϑ dϑ dλ , (1–135)
∂r 4π λ =0 ϑ =0
where

∂ r 2 − R2 1
M (r, ψ) ≡ = 5 (5R2 r−r 3 −R r 2 cos ψ −3R3 cos ψ) . (1–136)
∂r l3 l
Applying (1–135) to the special harmonic function V1 (r, ϑ, λ) = R/r, where
∂V1 R R
= − 2 and V1 (R, ϑ , λ ) = = 1, (1–137)
∂r r R
we obtain
2π π
R R
− 2 = M (r, ψ) sin ϑ dϑ dλ . (1–138)
r 4π λ =0 ϑ =0
Multiplying both sides of this equation by V (r, ϑ, λ) and subtracting it from
(1–135) gives
2π π
∂V R R
+ 2 VP = M (r, ψ) (V − VP ) sin ϑ dϑ dλ , (1–139)
∂r r 4π λ =0 ϑ =0
where
VP = V (r, ϑ, λ) , V = V (R, ϑ , λ ) . (1–140)
In order to find the radial derivative at the surface of the sphere of radius
R, we must set r = R. Then l becomes (Fig. 1.9)
1.14 The radial derivative of a harmonic function 33
P l0 P'
R sin(Ã/2)
R
Ã
Fig. 1.9. Spatial distance between two points on a sphere
ψ
l0 = 2R sin , (1–141)
2
and the function M takes the simple form
1 2R
M (R, ψ) = = . (1–142)
4R2 sin3 ψ2 l03
For ψ → 0 we have M (R, ψ) → ∞, and we cannot use the original formula

(1–135) at the surface of the sphere r = R. In the transformed equation
(1–139), however, we have V − VP → 0 for ψ → 0, and the singularity of
M for ψ → 0 will be neutralized (provided V is differentiable twice at P ).
Thus, we obtain the gradient formula
π
∂V 1 R2 2π V − VP
= − VP + sin ϑ dϑ dλ . (1–143)
∂r R 2π λ =0 ϑ =0 l03
This equation expresses ∂V /∂r on the sphere r = R in terms of V on this

sphere; thus, we now have
VP = V (R, ϑ, λ) , V = V (R, ϑ , λ ) . (1–144)
Solution in terms of spherical harmonics

We may express VP as
∞ n+1
R
VP = Yn (ϑ, λ) . (1–145)
n=0
r
Differentiation yields
∞
∂V Rn+1
=− (n + 1) n+2 Yn (ϑ, λ) . (1–146)
∂r r
n=0
For r = R, this becomes

∞
∂V 1
=− (n + 1) Yn (ϑ, λ) . (1–147)
∂r R n=0
This is the equivalent of (1–143) in terms of spherical harmonics. From this

equation, we get an interesting by-product. Writing (1–147) as
∞
∂V 1 1
= − VP − n Yn (ϑ, λ) (1–148)
∂r R R
n=0
and comparing this with (1–143), we see that

π ∞
R2 2π V − VP 1
sin ϑ dϑ dλ = − n Yn (ϑ, λ) . (1–149)
2π λ =0 ϑ =0 l03 R
n=0
This equation is formulated entirely in terms of quantities referred to the

spherical surface only. Furthermore, for any function prescribed on the sur-
face of a sphere, one can find a function in space that is harmonic outside
the sphere and assumes the values of the function prescribed on it. This is
done by solving Dirichlet’s exterior problem. From these facts, we conclude
that (1–149) holds for any (reasonably) arbitrary function V defined on the
surface of a sphere. These developments will be used in Sect. 2.20.
1.15 Laplace’s equation in ellipsoidal-harmonic

coordinates
Spherical harmonics are most frequently used in geodesy because they are
relatively simple and the earth is nearly spherical. Since the earth is more
nearly an ellipsoid of revolution, it might be expected that ellipsoidal har-
monics, which are defined in a way similar to that of the spherical harmonics,
would be even more suitable. The whole matter is a question of mathematical
convenience, since both spherical and ellipsoidal harmonics may be used for
any attracting body, regardless of its form. As ellipsoidal harmonics are more
complicated, however, they are used only in certain special cases which nev-
ertheless are important, namely, in problems involving rigorous computation
of normal gravity.
We introduce ellipsoidal-harmonic coordinates u, ϑ, λ (Fig. 1.10). In a
rectangular system, a point P has the coordinates x, y, z. Now we pass
through P the surface of an ellipsoid of revolution whose center is the origin
O, whose rotation axis coincides with the z-axis, and whose linear eccentric-
ity has the constant value E. The coordinate u is the semiminor axis of this
1.15 Laplace’s equation in ellipsoidal-harmonic coordinates 35
const
u=
E
2
# = const
u u2 +
P
u z
# ¯
F1 O E F2 xy - plane
x
z u 2 + E 2 sin # P
¸ ¸ = const
Gr y
ee
nw
i ch
Fig. 1.10. Ellipsoidal-harmonic coordinates: view from the front (top)

and view from above (bottom)
ellipsoid, ϑ is the complement of the “reduced latitude” β of P with respect

to this ellipsoid (the definition is seen in Fig. 1.10), i.e., ϑ = 90◦ − β, and λ
is the geocentric longitude in the usual sense.
It should be carefully noted that in spherical harmonics ϑ is the polar
distance, which is nothing but the complement of the geocentric latitude,
whereas in ellipsoidal-harmonic coordinates ϑ is the complement of the re-
duced latitude denoted by β.
The ellipsoidal-harmonic coordinates u, ϑ, λ are related to x, y, z by
√
x = u2 + E 2 sin ϑ cos λ ,
√
y = u2 + E 2 sin ϑ sin λ , (1–150)
z = u cos ϑ ,
√
which can be read from Fig. 1.10, considering that u2 + E 2 is the semi-
major axis of the ellipsoid whose surface passes through P . Because of
ϑ = 90◦ − β, we may equivalently write
√
x = u2 + E 2 cos β cos λ ,
√
y = u2 + E 2 cos β sin λ , (1–151)
z = u sin β .
Taking u = constant, we find
x2 + y 2 z2
+ = 1, (1–152)
u2 + E 2 u2
which represents an ellipsoid of revolution. For ϑ = constant, we obtain
x2 + y 2 z2
2 − = 1, (1–153)
E 2 sin ϑ E cos2 ϑ
2
which represents a hyperboloid of one sheet, and for λ = constant, we get

the meridian plane
y = x tan λ . (1–154)
The constant focal length E, i.e., the distance between the coordinate origin
O and one of the focal points F1 or F2 , which is the same for all ellipsoids
u = constant, characterizes the coordinate system. For E = 0 we have the
usual spherical coordinates u = r and ϑ, λ as a limiting case.
To find ds, the element of arc, in ellipsoidal-harmonic coordinates, we
proceed in the same way as in spherical coordinates, Eq. (1–30), and obtain
u2 + E 2 cos2 ϑ 2
ds2 = du +(u2 +E 2 cos2 ϑ) dϑ2 +(u2 +E 2 ) sin2 ϑ dλ2 . (1–155)
u2 + E 2
The coordinate system u, ϑ, λ is again orthogonal: the products du dϑ, etc.,
are missing in the equation above. Setting u = q1 , ϑ = q2 , λ = q3 , we have
in (1–31)
u2 + E 2 cos2 ϑ
h21 = , h22 = u2 + E 2 cos2 ϑ , h23 = (u2 + E 2 ) sin2 ϑ .
u2 + E 2
(1–156)
If we substitute these relations into (1–32), we obtain

1 ∂ 2 2 ∂V
∆V = 2 (u + E ) sin ϑ +
(u + E 2 cos2 ϑ) sin ϑ ∂u ∂u
(1–157)
∂ ∂V ∂ u2 + E 2 cos2 ϑ ∂V
sin ϑ + .
∂ϑ ∂ϑ ∂λ (u2 + E 2 ) sin ϑ ∂λ
1.16 Ellipsoidal harmonics 37
Performing the differentiations and cancelling sin ϑ, we get

2
1 2 ∂ V ∂V ∂2V
∆V ≡ 2 (u2
+ E ) + 2u + +
u + E 2 cos2 ϑ ∂u2 ∂u ∂ϑ2
(1–158)
∂V u2 + E 2 cos2 ϑ ∂ 2 V
cot ϑ + 2 = 0,
∂ϑ (u + E 2 ) sin2 ϑ ∂λ2
which is Laplace’s equation in ellipsoidal-harmonic coordinates. An alterna-
tive expression is obtained by omitting the factor (u2 + E 2 cos2 ϑ)−1 :
∂2V ∂V ∂2V ∂V u2 + E 2 cos2 ϑ ∂ 2 V
(u2 + E 2 ) + 2u + + cot ϑ + = 0.
∂u2 ∂u ∂ϑ2 ∂ϑ (u2 + E 2 ) sin2 ϑ ∂λ2
(1–159)
In the limiting case, E → 0, these equations reduce to the spherical expres-
sions (1–35) and (1–36).
1.16 Ellipsoidal harmonics

To solve (1–158) or (1–159), we proceed in a way which is analogous to
the method used to solve the corresponding equation (1–36) in spherical
coordinates. What we did there may be summarized as follows. By the trial
substitution
V (r, ϑ, λ) = f (r) g(ϑ) h(λ) , (1–160)
we separated the variables r, ϑ, λ, so that the original partial differential
equation (1–36) was decomposed into three ordinary differential equations
(1–39), (1–46), and (1–47).
In order to solve Laplace’s equation in ellipsoidal coordinates (1–159),
we correspondingly make the ansatz (trial substitution)
V (u, ϑ, λ) = f (u) g(ϑ) h(λ) . (1–161)
Substituting and dividing by f g h, we get

1 2 1 u2 + E 2 cos2 ϑ h
[(u + E 2 ) f + 2u f ]+ (g + g cot ϑ)+ 2 = 0 . (1–162)
f g (u + E 2 ) sin2 ϑ h
The variable λ occurs only through the quotient h /h, which consequently
must be constant. One sees this more clearly by writing the equation in the
form

(u2 + E 2 ) sin2 ϑ 1 2 1 h
− 2 [(u + E 2
) f + 2u f ] + (g + g cot ϑ) = .
u + E 2 cos2 ϑ f g h
(1–163)
The left-hand side depends only on u and ϑ, the right-hand side only on λ.
The two sides cannot be identically equal unless both are equal to the same
constant. Therefore,
h
= −m2 . (1–164)
h
The factor by which h /h is to be multiplied, i.e., the inverse of the main
factor on the left-hand side of (1–163), can be decomposed as follows:
u2 + E 2 cos2 ϑ 1 E2
= − . (1–165)
(u2 + E 2 ) sin2 ϑ sin2 ϑ u2 + E 2
Substituting (1–164) and (1–165) into (1–163) and combining functions of
the same variable, we obtain
1 2 E2 1 m2
[(u + E 2 ) f + 2u f ] + 2 m 2
= − (g + g
cot ϑ) + . (1–166)
f u + E2 g sin2 ϑ
The two sides are functions of different independent variables and must there-
fore be constant. Denoting this constant by n(n + 1), we finally get

E2
(u + E ) f (u) + 2u f (u) − n(n + 1) − 2
2 2 2
m f (u) = 0 ; (1–167)
u + E2

m2
sin ϑ g (ϑ) + cos ϑ g (ϑ) + n(n + 1) sin ϑ − g(ϑ) = 0 ; (1–168)
sin ϑ
h (λ) + m2 h(λ) = 0 . (1–169)
These are the three ordinary differential equations into which the partial
differential equation (1–159) is decomposed by the separation of variables
(1–161).
The second and third equations are the same as in the spherical case,
Eqs. (1–46) and (1–47); the first equation is different. The substitutions
u √
τ =i (where i = −1) and t = cos ϑ (1–170)
E
transform the first and second equations into

¯ ¯ m2
(1 − τ ) f (τ ) − 2τ f (τ ) + n(n + 1) −
2 f¯(τ ) = 0 ,
1 − τ2
(1–171)
m2
(1 − t ) ḡ (t) − 2t ḡ (t) + n(n + 1) −
2 ḡ(t) = 0 ,
1 − t2
where the overbar indicates that the functions f and g are expressed in terms
of the new arguments τ and t. From spherical harmonics we are already
familiar with the substitution t = cos ϑ and the corresponding equation for
ḡ(t).
Note that f¯(τ ) satisfies formally the same differential equation as ḡ(t),
namely, Legendre’s equation (1–56). As we have seen, this differential equa-
tion has two solutions: Legendre’s function Pnm and Legendre’s function of
the second kind Qnm . For ḡ(t), where t = cos ϑ, the Qnm (t) are ruled out for
obvious reasons, as we have seen in Sect. 1.8. For f¯(τ ), however, both sets of
functions Pnm (τ ) and Qnm (τ ) are possible solutions; they correspond to the
two different solutions f = r n and f = r −(n+1) in the spherical case. Finally,
(1–169) has as before the solutions cos mλ and sin mλ.
We summarize all individual solutions:
u u
f (u) = Pnm i or Qnm i ;
E E
g(ϑ) = P (cos ϑ) ; (1–172)
nm
h(λ) = cos mλ or sin mλ .

Here n and m < n are integers 0, 1, 2, . . ., as before. Hence, the functions
u
cos mλ
V (u, ϑ, λ) = Pnm i Pnm (cos ϑ) ,
E sin mλ
u (1–173)
cos mλ
V (u, ϑ, λ) = Qnm i Pnm (cos ϑ)
E sin mλ
are solutions of Laplace’s equation ∆V = 0, that is, harmonic functions.
From these functions we may form by linear combination the series
u
∞ n
nm iP
Vi (u, ϑ, λ) = E ·
b
n=0 m=0 Pnm i
E
[anm Pnm (cos ϑ) cos mλ + bnm Pnm (cos ϑ) sin mλ] ;
u (1–174)
∞ n Qnm i
Ve (u, ϑ, λ) = E ·
b
n=0 m=0 Qnm i
E
[anm Pnm (cos ϑ) cos mλ + bnm Pnm (cos ϑ) sin mλ] .
Here b is the semiminor axis of an arbitrary but fixed ellipsoid which may
be called the reference ellipsoid (Fig. 1.11). The division by Pnm (ib/E) or
Qnm (ib/E) is possible because they are constants; its purpose is to simplify
the expressions and to make the coefficients anm and bnm real.
P
u
b
F1 F2
E
a u2 + E 2
refere
nce ellipsoid
Fig. 1.11. Reference ellipsoid and ellipsoidal-harmonic coordinates
If the eccentricity E reduces to zero, the ellipsoidal-harmonic coordinates

u, ϑ, λ become spherical coordinates r, ϑ, λ; the ellipsoid u = b becomes the
sphere r = R because then the semiaxes a and b are equal to the radius R;
and we find
u u
Pnm i u n r n Qnm i n+1
E E R
lim = = , lim = ,
E→0 b b R E→0 b r
Pnm i Qnm i
E E
(1–175)
so that the first series in (1–174) becomes (1–116), and the second series in
(1–174) becomes (1–117). Thus, we see that the function Pnm (iu/E) corre-
sponds to r n and Qnm (iu/E) corresponds to r −(n+1) in spherical harmonics.
Hence, the first series in (1–174) is harmonic in the interior of the ellipsoid
u = b, and the second series in (1–174) is harmonic in its exterior; this case
is relevant to geodesy. For u = b, the two series are equal:
Vi (b, ϑ, λ) = Ve (b, ϑ, λ)
n
∞ (1–176)
= [anm Pnm (cos ϑ) cos mλ + bnm Pnm (cos ϑ) sin mλ] .
n=0 m=0
Thus, the solution of Dirichlet’s boundary-value problem for the ellipsoid

of revolution is easy. We expand the function V (b, ϑ, λ), given on the ellip-
soid u = b, into a series of surface spherical harmonics with the following
arguments: ϑ = complement of reduced latitude, λ = geocentric longitude.
Then the first series in (1–174) is the solution of the interior problem and
the second series in (1–174) is the solution of the exterior Dirichlet problem.
Formula (1–176) shows that not only functions that are defined on the
surface of a sphere can be expanded into a series of surface spherical har-
monics. Such an expansion is even possible for rather arbitrary functions
defined on a convex surface.
A remark on terminology
The ellipsoidal-harmonic coordinates u, ϑ (or β), λ are the generalization of
spherical coordinates for the sole use of getting closed solutions of Laplace’s
equation, in particular, for the gravity field of the reference ellipsoid in
Sect. 2.7. The brief name “ellipsoidal coordinates” frequently used for u, β, λ
might lead to a confusion with the ellipsoidal coordinates ϕ, λ, h. In the
present book, “ellipsoidal coordinates” will always denote “ellipsoidal ge-
ographic coordinates”, frequently also called “geodetic coordinates”, being
represented by ϕ, λ, h.
2 Gravity field of the earth
2.1 Gravity
The total force acting on a body at rest on the earth’s surface is the resultant
of gravitational force and the centrifugal force of the earth’s rotation and is
called gravity.
Take a rectangular coordinate system whose origin is at the earth’s center
of gravity and whose z-axis coincides with the earth’s mean axis of rotation
(Fig. 2.1). The x- and y-axes are so chosen as to obtain a right-handed
coordinate system; otherwise they are arbitrary. For convenience, we may
assume an x-axis which is associated with the mean Greenwich meridian (it
“points” towards the mean Greenwich meridian). Note that we are assuming
in this book that the earth is a solid body rotating with constant speed
around a fixed axis. This is a rather simplified assumption, see Moritz and
Mueller (1987). The centrifugal force f on a unit mass is given by
f = ω2p , (2–1)
where ω is the angular velocity of the earth’s rotation and

p= x2 + y 2 (2–2)
is the distance from the axis of rotation. The vector f of this force has the
z
!
P
p
f
x
y p y
x
Fig. 2.1. The centrifugal force

44 2 Gravity field of the earth
direction of the vector

p = [x, y, 0] (2–3)
and is, therefore, given by
f = ω 2 p = [ω 2 x, ω 2 y, 0] . (2–4)
The centrifugal force can also be derived from a potential
1 2 2
Φ= ω (x + y 2 ) , (2–5)
2
so that
∂Φ ∂Φ ∂Φ
f = grad Φ ≡ , , . (2–6)
∂x ∂y ∂z
Substituting (2–5) into (2–6) yields (2–4).

In the introductory remark above, we mentioned that gravity is the resul-
tant of gravitational force and centrifugal force. Accordingly, the potential of
gravity, W , is the sum of the potentials of gravitational force, V , cf. (1–12),
and centrifugal force, Φ:

1
W = W (x, y, z) = V + Φ = G dv + ω 2 (x2 + y 2 ) , (2–7)
l 2
v
where the integration is extended over the earth.

Differentiating (2–5), we find
∂2Φ ∂2Φ ∂2Φ

∆Φ ≡ + + = 2ω 2 . (2–8)
∂x2 ∂y 2 ∂z 2
If we combine this with Poisson’s equation (1–17) for V , we get the general-
ized Poisson equation for the gravity potential W :
∆W = −4π G + 2ω 2 . (2–9)
Since Φ is an analytic function, the discontinuities of W are those of V : some

second derivatives have jumps at discontinuities of density.
The gradient vector of W ,

∂W ∂W ∂W
g = grad W ≡ , , (2–10)
∂x ∂y ∂z
2.1 Gravity 45
with components

∂W x−ξ
gx = = −G dv + ω 2 x ,
∂x l3
v

∂W y−η
gy = = −G dv + ω 2 y , (2–11)
∂y l3
v

∂W z−ζ
gz = = −G dv ,
∂z l3
v
is called the gravity vector; it is the total force (gravitational force plus
centrifugal force) acting on a unit mass. As a vector, it has magnitude and
direction.
The magnitude g is called gravity in the narrower sense. It has the phys-
ical dimension of an acceleration and is measured in gal (1 gal = 1 cm s−2 ),
the unit being named in honor of Galileo Galilei. The numerical value of g
is about 978 gal at the equator, and 983 gal at the poles. In geodesy, another
unit is often convenient – the milligal, abbreviated mgal (1 mgal = 10−3 gal).
In SI units, we have
1 gal = 0.01 m s−2 ,

(2–12)
1 mgal = 10 µm s−2 .
The direction of the gravity vector is the direction of the plumb line, or the
vertical; its basic significance for geodetic and astronomical measurements
is well known.
In addition to the centrifugal force, another force called the Coriolis force
acts on a moving body. It is proportional to the velocity with respect to the
earth, so that it is zero for bodies resting on the earth. Since in classical
geodesy (i.e., not considering navigation) we usually deal with instruments
at rest relative to the earth, the Coriolis force plays no role here and need
not be considered.
Gravitational and inertial mass

The reader may have noticed that the mass m has been used in two con-
ceptually completely different senses: as inertial mass in Newton’s law of
inertia, force = mass × acceleration and as gravitational mass in Newton’s
law of gravitation (1–1). Thus, m in gravitation, which is a “true” force, is
the gravitational mass, but m in the centrifugal “force”, which is an accel-
eration, is the inertial mass. The Hungarian physicist Roland Eötvös had
shown experimentally already around 1890 that both kinds of masses are
equal within 10−11 , which is a formidable accuracy. He used the same type
of instrument by which experimental physicists have been able to determine
the numerical value of the gravitational constant G only to a poor accuracy
of about 10−4 , as we have seen at the beginning of this book. The coinci-
dence between the inertial and the gravitational mass was far too good to be
a physical accident, but, within classical mechanics, it was an inexplicable
miracle. It was not before 1915 that Einstein made it one of the pillars of
the general theory of relativity!
2.2 Level surfaces and plumb lines

The surfaces
W (x, y, z) = constant , (2–13)
on which the potential W is constant, are called equipotential surfaces or
level surfaces.
Differentiating the gravity potential W = W (x, y, z), we find
∂W ∂W ∂W
dW = dx + dy + dz . (2–14)
∂x ∂y ∂z
In vector notation, using the scalar product, this reads
dW = grad W · dx = g · dx , (2–15)
where
dx = [dx, dy, dz] . (2–16)
If the vector dx is taken along the equipotential surface W = constant, then
the potential remains constant and dW = 0, so that (2–15) becomes
g · dx = 0 . (2–17)
If the scalar product of two vectors is zero, then these vectors are orthogonal
to each other. This equation therefore expresses the well-known fact that the
gravity vector is orthogonal to the equipotential surface passing through the
same point.
The surface of the oceans, after some slight idealization, is part of a
certain level surface. This particular equipotential surface was proposed as
the “mathematical figure of the earth” by C.F. Gauss, the “Prince of Mathe-
maticians”, and was later termed the geoid. This definition has proved highly
suitable, and the geoid is still frequently considered by many to be the fun-
damental surface of physical geodesy. The geoid is thus defined by
W = W0 = constant . (2–18)
2.2 Level surfaces and plumb lines 47
P
H
g
geoid
W = W0
level surface
W = constant
Fig. 2.2. Level surfaces and plumb lines
If we look at equation (2–7) for the gravity potential W , we can see that
the equipotential surfaces, expressed by W (x, y, z) = constant, are rather
complicated mathematically. The level surfaces that lie completely outside
the earth are at least analytical surfaces, although they have no simple ana-
lytical expression, because the gravity potential W is analytical outside the
earth. This is not true of level surfaces that are partly or wholly inside the
earth, such as the geoid. They are continuous and “smooth” (i.e., without
edges), but they are no longer analytical surfaces; we will see in the next sec-
tion that the curvature of the interior level surfaces changes discontinuously
with the density.
The lines that intersect all equipotential surfaces orthogonally are not
exactly straight but slightly curved (Fig. 2.2). They are called lines of force,
or plumb lines. The gravity vector at any point is tangent to the plumb line at
that point, hence “direction of the gravity vector”, “vertical”, and “direction
of the plumb line” are synonymous. Sometimes this direction itself is briefly
denoted as “plumb line”.
As the level surfaces are, so to speak, horizontal everywhere, they share
the strong intuitive and physical significance of the horizontal; and they share
the geodetic importance of the plumb line because they are orthogonal to
it. Thus, we understand why so much attention is paid to the equipotential
surfaces.
The height H of a point above sea level (also called the orthometric
height) is measured along the curved plumb line, starting from the geoid
(Fig. 2.2). If we take the vector dx along the plumb line, in the direction of
increasing height H, then its length will be
dx = dH (2–19)
and its direction is opposite to the gravity vector g, which points downward,
so that the angle between dx and g is 180◦ . Using the definition of the scalar
product (i.e., for two vectors a and b it is defined as a · b = ab cos ω,
where ω is the angle between the two vectors), we get
g · dx = g dH cos 180◦ = −g dH (2–20)
accordingly, so that Eq. (2–15) becomes
dW = −g dH . (2–21)
This equation relates the height H to the potential W and will be basic for
the theory of height determination (Chap. 4). It shows clearly the insepara-
ble interrelation that characterizes geodesy – the interrelation between the
geometrical concepts (H) and the dynamic concepts (W ).
Another form of Eq. (2–21) is
∂W
g=− . (2–22)
∂H
It shows that gravity is the negative vertical gradient of the potential W , or
the negative vertical component of the gradient vector grad W .
Since geodetic measurements (theodolite measurements, leveling, but
also satellite techniques etc.) are almost exclusively referred to the system
of level surfaces and plumb lines, the geoid plays an essential part. Thus,
we see why the aim of physical geodesy has been formulated as the de-
termination of the level surfaces of the earth’s gravity field. In a still more
abstract but equivalent formulation, we may also say that physical geodesy
aims at the determination of the potential function W (x, y, z). At a first
glance, the reader is probably perplexed about this definition, which is due
to Bruns (1878), but its meaning is easily understood: If the potential W is
given as a function of the coordinates x, y, z, then we know all level surfaces
including the geoid; they are given by the equation
W (x, y, z) = constant. (2–23)
2.3 Curvature of level surfaces and plumb lines

The formula for the curvature of a curve y = f (x) is
1 y
κ= = , (2–24)
(1 + y 2 )3/2
2.3 Curvature of level surfaces and plumb lines 49
y
P
|| x
y = f (x )
Fig. 2.3. The curvature of a curve
where κ is the curvature, is the radius of curvature, and
dy d2 y
y = , y = 2 . (2–25)
dx dx
If we use a plane local coordinate system xy in which a parallel to the x-axis
is tangent at the point P under consideration (Fig. 2.3), then this implies
y = 0 and we get simply
1 d2 y
κ= = 2. (2–26)
dx
Level surfaces
Consider now a point P on a level surface S. Take a local coordinate system
xyz with origin at P whose z-axis is vertical, that is, orthogonal to the
surface S (Fig. 2.4). We intersect this level surface
W (x, y, z) = constant (2–27)
with the xz-plane by setting

y = 0. (2–28)
z
plumb
line
y
x
P
level surface
Fig. 2.4. The local coordinate system

Comparing Fig. 2.4 with Fig. 2.3, we see that z now takes the place of
y. Therefore, instead of (2–26) we have for the curvature of the intersection
of the level surface with the xz-plane:
d2 z
K1 = . (2–29)
dx2
If we differentiate W (x, y, z) = W0 with respect to x, considering that y
is zero and z is a function of x, we get
dz
Wx + Wz = 0,
dx
2 (2–30)
dz dz d2 z
Wxx + 2Wxz + Wzz + Wz 2 = 0 ,
dx dx dx
where the subscripts denote partial differentiation:
∂W ∂2W
Wx = , Wxz = , ... . (2–31)
∂x ∂x ∂z
Since the x-axis is tangent at P , we get dz/dx = 0 at P , so that
d2 z Wxx
=− . (2–32)
dx2 Wz
Since the z-axis is vertical, we have, using (2–22),
∂W ∂W
Wz = = = −g . (2–33)
∂z ∂H
Therefore, Eq. (2–29) becomes
Wxx
K1 = . (2–34)
g
The curvature of the intersection of the level surface with the yz-plane is
found by replacing x with y:
Wyy
K2 = . (2–35)
g
The mean curvature J of a surface at a point P is defined as the arith-
metic mean of the curvatures of the curves in which two mutually perpen-
dicular planes through the surface normal intersect the surface (Fig. 2.5).
Hence, we find
1 Wxx + Wyy
J = − (K1 + K2 ) = − . (2–36)
2 2g
surface normal
Fig. 2.5. Definition of mean curvature
Here the minus sign is only a convention. This is an expression for the mean
curvature of the level surface.
From the generalized Poisson equation
∆W ≡ Wxx + Wyy + Wzz = −4π G + 2ω 2 , (2–37)
we find
−2g J + Wzz = −4π G + 2ω 2 . (2–38)
Considering
∂g ∂g
Wz = −g , Wzz = − =− , (2–39)
∂z ∂H
we finally obtain
∂g
= −2g J + 4π G − 2ω 2 . (2–40)
∂H
This important equation, relating the vertical gradient of gravity ∂g/∂H to
the mean curvature of the level surface, is also due to Bruns (1878). It is
another beautiful example of the interrelation between the geometric and
dynamic concepts in geodesy.
Plumb lines
The curvature of the plumb line is needed for the reduction of astronomical
observations to the geoid. A plumb line may be defined as a curve whose
line element vector
dx = [dx, dy, dz] (2–41)
has the direction of the gravity vector
g = [Wx , Wy , Wz ] ; (2–42)
that is, dx and g differ only by a proportionality factor. This is best expressed
in the form
dx dy dz
= = . (2–43)
Wx Wy Wz
In the coordinate system of Fig. 2.4, the curvature of the projection of
the plumb line onto the xz-plane is given by
d2 x
κ1 = ; (2–44)
dz 2
this is equation (2–26) applied to the present case. Using (2–43), we have
dx Wx
= . (2–45)
dz Wz
We differentiate with respect to z, considering that y = 0:

d2 x 1 dx dx
= 2 Wz Wxz + Wxx − Wx Wzz + Wzx . (2–46)
dz 2 Wz dz dz
In our particular coordinate system, the gravity vector coincides with the
z-axis, so that its x- and y-components are zero:
Wx = Wy = 0 . (2–47)
Figure 2.4 shows that we also have

dx
= 0. (2–48)
dz
Therefore,
d2 x Wz Wxz Wxz Wzx
= = = . (2–49)
dz 2 Wz2 Wz Wz
Considering Wz = −g, we finally obtain
1 ∂g
κ1 = (2–50)
g ∂x
and, similarly,
1 ∂g
κ2 = . (2–51)
g ∂y
These are the curvatures of the projections of the plumb line onto the xz- and
yz-plane, the z-axis being vertical, that is, coinciding with the gravity vector.
The total curvature κ of the plumb line is given, according to differential
geometry (essentially Pythagoras’ theorem), by

1 2
κ = κ21 + κ22 = gx + gy2 . (2–52)
g
For reducing astronomical observations (Sect. 5.12), we need only the

projection curvatures (2–50) and (2–51).
We mention finally that the various formulas for the curvature of level
surfaces and plumb lines are equivalent to the single vector equation
grad g = (−2g J + 4π G − 2ω 2 ) n + g κ n1 , (2–53)
where n is the unit vector along the plumb line (its unit tangent vector) and
n1 is the unit vector along the principal normal to the plumb line. This may
be easily verified. Using the local xyz-system, we have
n = [0, 0, 1] ,
(2–54)
n1 = [cos α, sin α, 0] ,
where α is the angle between the principal normal and the x-axis (Fig. 2.6).
The z-component of (2–53) yields Bruns’ equation (2–40), and the horizontal
components yield
∂g ∂g
= g κ cos α , = g κ sin α . (2–55)
∂x ∂y
These are identical to (2–50) and (2–51), since κ1 = κ cos α and κ2 = κ sin α,
as differential geometry shows. Equation (2–53) is called the generalized
Bruns equation.
More about the curvature properties and the “inner geometry” of the
gravitational field will be found in books by, e.g., Hotine (1969: Chaps. 4–
20), Marussi (1985) and Moritz and Hofmann-Wellenhof (1993: Chap. 3).
z
plumb
n line
sin ® y
®
®
1
cos
n1
x
Fig. 2.6. Generalized Bruns equation

2.4 Natural coordinates

The system of level surfaces and plumb lines may be used as a three-
dimensional curvilinear coordinate system that is well suited to certain pur-
poses; these coordinates can be measured directly, as opposed to local rectan-
gular coordinates x, y, z. Note, however, that global rectangular coordinates
may be measured directly using satellites, see Sect. 5.3.
The direction of the earth’s axis of rotation and the position of the equa-
torial plane (normal to the axis) are well defined astronomically. The astro-
nomical latitude Φ of a point P is the angle between the vertical (direction
of the plumb line) at P and the equatorial plane, see Fig. 2.7. From this
figure, we also see that line P N is parallel to the rotation axis, plane GP F
normal to it, that is, parallel to the equatorial plane; n is the unit vector
along the plumb line; plane N P F is the meridian plane of P , and plane
N P G is parallel to the meridian plane of Greenwich.
Consider now a straight line through P parallel to the earth’s axis of
rotation. This parallel and the vertical at P together define the meridian
plane of P . The angle between this meridian plane and the meridian plane
of Greenwich (or some other fixed plane) is the astronomical longitude Λ of
P . (Exercise: define Φ and Λ without using the unit sphere. The solution
may be found in Sect. 5.9).
unit sphere
N
vertical
n
P F
G L
F
eridian
Greenwich m
equator
earth
Fig. 2.7. Definition of the astronomical coordinates Φ and Λ of P by

means of a unit sphere with center at P
2.4 Natural coordinates 55
The astronomical coordinates, latitude Φ and longitude Λ, form two of

the three spatial coordinates of P . As third coordinate we may take the
orthometric height H of P or its potential W . Equivalent to W is the geopo-
tential number C = W0 − W , where W0 is the potential of the geoid. The
orthometric height H was defined in Sect. 2.2; see also Fig. 2.2. The relations
between W , C, and H are given by the equations
H
W = W0 − g dH = W0 − C ,
0
H
C = W0 − W = g dH , (2–56)
0
W C
dW dC
H =− = ,
W0 g 0 g
which follow from integrating (2–21). The integral is taken along the plumb
line of point P , starting from the geoid, where H = 0 and W = W0 (see also
Fig. 2.8).
The quantities
Φ, Λ, W or Φ, Λ, H (2–57)
are called natural coordinates. They are the real-earth counterparts of the
ellipsoidal coordinates. They are related in the following way to the geocen-
tric rectangular coordinates x, y, z of Sect. 2.1. The x-axis is associated with
the mean Greenwich meridian; from Fig. 2.7 we read that the unit vector of
the vertical n has the xyz-components
n = [cos Φ cos Λ, cos Φ sin Λ, sin Φ] ; (2–58)
the gravity vector g is known to be
g = [Wx , Wy , Wz ] . (2–59)
earth's surface
P
level surface
W = constant
H
sea level geoid
0 W = W0
Fig. 2.8. The orthometric height H

On the other hand, since n is the unit vector corresponding to g but of

opposite direction, it is given by
g g
n=− =− , (2–60)
g g
so that
g = −g n . (2–61)
This equation, together with (2–58) and (2–59), gives
−Wx = g cos Φ cos Λ ,

−Wy = g cos Φ sin Λ , (2–62)
−Wz = g sin Φ .
Solving for Φ and Λ, we finally obtain
−Wz
Φ = tan−1 ,
Wx2 + Wy2
Wy (2–63)
Λ = tan−1 ,
Wx
W = W (x, y, z) .
These three equations relate the natural coordinates Φ, Λ, W to the rectan-

gular coordinates x, y, z, provided the function W = W (x, y, z) is known.
We see that Φ, Λ, H are related to x, y, z in a considerably more com-
plicated way than the spherical coordinates r, ϑ, λ of Sect. 1.4. Note also
the conceptual difference between the astronomical longitude Λ and the geo-
centric longitude λ.
2.5 The potential of the earth in terms of spherical

harmonics
Looking at the expression (2–7) for the gravity potential W , we see that the
part most difficult to handle is the gravitational potential V , the centrifugal
potential being a simple analytic function.
The gravitational potential V can be made more manageable for many
purposes if we keep in mind the fact that outside the attracting masses it is
a harmonic function and can therefore be expanded into a series of spherical
harmonics.
2.5 The potential of the earth in terms of spherical harmonics 57
r
l
Ã
O r'
dM
Fig. 2.9. Expansion into spherical harmonics
We now evaluate the coefficients of this series. The gravitational potential

V is given by the basic equation (1–12):

dM
V =G , (2–64)
l
earth
where we now denote the mass element by dM ; the integral is extended over
the entire earth. Into this integral we substitute the expression (1–104):
∞
1 r n
= Pn (cos ψ) , (2–65)
l r n+1
n=0
where the Pn are the conventional Legendre polynomials, r is the radius

vector of the fixed point P at which V is to be determined, r is the radius
vector of the variable mass element dM , and ψ is the angle between r and
r (Fig. 2.9).
Since r is a constant with respect to the integration over the earth, it
can be taken out of the integral. Thus, we get
∞
1
V = n+1
G r n Pn (cos ψ) dM . (2–66)
n=0
r
earth
Writing this in the usual form as a series of solid spherical harmonics,

∞
Yn (ϑ, λ)
V = , (2–67)
r n+1
n=0
we see by comparison that the Laplace surface spherical harmonic Yn (ϑ, λ)

is given by
Yn (ϑ, λ) = G r n Pn (cos ψ) dM , (2–68)
earth
the dependence on ϑ and λ arises from the angle ψ since
The spherical coordinates ϑ, λ have been defined in Sect. 1.4.

A more explicit form is obtained by using the decomposition formula
(1–108):
∞ n
1 1 R̄nm (ϑ, λ) n S̄nm (ϑ, λ) n
= r R̄nm (ϑ , λ ) + r S̄nm (ϑ , λ ) .
l n=0 m=0
2n + 1 r n+1 r n+1
(2–70)
Substituting this relation into the integral (2–64), we obtain
n
∞

R̄nm (ϑ, λ) S̄nm (ϑ, λ)
V = Ānm + B̄nm , (2–71)
r n+1 r n+1
n=0 m=0
where the constant coefficients Ānm and B̄nm are given by

(2n + 1) Ānm = G r n R̄nm (ϑ , λ ) dM ,
earth
(2–72)
(2n + 1) B̄nm = G r n S̄nm (ϑ , λ ) dM .
earth
These formulas are very symmetrical and easy to remember: the coefficient,
multiplied by 2n + 1, of the solid harmonic
R̄nm (ϑ, λ)
(2–73)
r n+1
is the integral of the solid harmonic
r n R̄nm (ϑ , λ ) . (2–74)
An analogous relation results for S̄nm .

Note the nice analogy: V is a sum and the coefficients are integrals!
Since the mass element is
dM = dx dy dz = r 2 sin ϑ dr dϑ dλ , (2–75)

2.5 The potential of the earth in terms of spherical harmonics 59
the actual evaluation of the integrals requires that the density be expressed
as a function of r , ϑ , λ . Although no such expression is available at present,
this fact does not diminish the theoretical and practical significance of spher-
ical harmonics, since the coefficients Anm , Bnm can be determined from the
boundary values of gravity at the earth’s surface. This is a boundary-value
problem (see Sect. 1.13) and will be elaborated later.
Recalling the relations (1–91) and (1–98) between conventional and fully
normalized spherical harmonics, we can also write equations (2–71) and (2–
72) in terms of conventional harmonics, readily obtaining
n
∞

Rnm (ϑ, λ) Snm (ϑ, λ)
V = Anm + Bnm , (2–76)
r n+1 r n+1
n=0 m=0
where

An0 =G r n Pn (cos ϑ ) dM ;
earth
⎫
(n − m)! ⎪
⎪
Anm = 2 G r n Rnm (ϑ , λ ) dM ⎪
⎪ (2–77)
(n + m)! ⎪
⎬
earth
(m = 0) .
(n − m)! ⎪
⎪
Bnm = 2 G r n Snm (ϑ , λ ) dM ⎪
⎪
(n + m)! ⎪
⎭
earth
In connection with satellite dynamics, the potential V is often written in

the form
∞ n
GM a n
V = 1+ Cnm Rnm (ϑ, λ) + Snm Snm (ϑ, λ) , (2–78)
r n=1 m=0
r
where a is the equatorial radius of the earth, so that

Anm = GM an Cnm
(n = 0) . (2–79)
Bnm = GM an Snm
Distinguish the coefficient Snm and the function Snm ! The coefficient Cn0
has formerly been denoted by −Jn . Note that C is related to cosine and S
is related to sine.
The corresponding fully normalized coefficients
1
C̄n0 = √ Cn0 ,
2n + 1
⎫
(n + m)! ⎪
⎪
⎪
⎪
C̄nm = Cnm ⎪
⎪ (2–80)
2(2n + 1)(n − m)! ⎬
(m = 0)
⎪
⎪
(n + m)! ⎪
⎪
S̄nm = Snm ⎪
⎪
2(2n + 1)(n − m)! ⎭
are also used.

It is obvious that the nonzonal terms (m = 0) would be missing in
all these expansions if the earth had complete rotational symmetry, since
the terms mentioned depend on the longitude λ. In rotationally symmetrical
bodies there is no dependence on λ because all longitudes are equivalent. The
tesseral and sectorial harmonics will be small, however, since the deviations
from rotational symmetry are slight.
Finally, we discuss the convergence of (2–71), or of the equivalent series
expansions, of the earth’s potential. This series is an expansion in powers
of 1/r. Therefore, the larger r is, the better the convergence. For smaller r
it is not necessarily convergent. For an arbitrary body, the expansion of V
into spherical harmonics can be shown to converge always outside the small-
est sphere r = r0 that completely encloses the body (Fig. 2.10). Inside this
sphere, the series is usually divergent. In certain cases it can converge partly
inside the sphere r = r0 . If the earth were a homogeneous ellipsoid of about
r = r0
r0
Fig. 2.10. Spherical-harmonic expansion of V

converges outside the sphere r = r0
2.6 Harmonics of lower degree 61
the same dimensions, then the series for V would indeed still converge at the
surface of the earth. Owing to the mass irregularities, however, the series of
the actual potential V of the earth can be divergent or also convergent at the
surface of the earth. Theoretically, this makes the use of a harmonic expan-
sion of V at the earth’s surface somewhat difficult; practically, it is always
safe to regard it as convergent. For a detailed discussion see Moritz (1980 a:
Sects. 6 and 7) and Sect. 8.6 herein.
It need hardly be pointed out that the spherical-harmonic expansion,
always expressing a harmonic function, can represent only the potential out-
side the attracting masses, never inside.
2.6 Harmonics of lower degree

It is instructive to evaluate the coefficients of the first few spherical harmonics
explicitly. For ready reference, we first state some conventional harmonic
functions Rnm and Snm , using (1–60), (1–66), and (1–82):
R00 = 1 , S00 = 0 ,
R10 = cos ϑ , S10 = 0 ,
R11 = sin ϑ cos λ , S11 = sin ϑ sin λ ,

(2–81)
R20 = 32 cos2 ϑ − 1
2 , S20 = 0 ,
R21 = 3 sin ϑ cos ϑ cos λ , S21 = 3 sin ϑ cos ϑ sin λ ,
R22 = 3 sin2 ϑ cos 2λ , S22 = 3 sin2 ϑ sin 2λ .

The corresponding solid harmonics r n Rnm and r n Snm are simply homoge-
neous polynomials in x, y, z. For instance,
r 2 S22 = 6r 2 sin2 ϑ sin λ cos λ = 6(r sin ϑ cos λ)(r sin ϑ sin λ) = 6xy . (2–82)
In this way, we find

R00 = 1 , S00 = 0 ,
r R10 = z , r S10 = 0 ,
r R11 = x , r S11 = y ,
(2–83)
r 2 R20 = − 12 x2 − 12 y 2 + z 2 , r 2 S20 = 0 ,
r 2 R21 = 3x z , r 2 S21 = 3y z ,
r 2 R22 = 3x2 − 3y 2 , r 2 S22 = 6x y .

Substituting these functions into the expression (2–77) for the coefficients
Anm and Bnm yields for the zero-degree term

A00 = G dM = GM ; (2–84)
earth
that is, the product of the mass of the earth times the gravitational constant.
For the first-degree coefficients, we get

A10 = G z dM , A11 = G x dM , B11 = G y dM ;
earth earth earth
(2–85)
and for the second-degree coefficients

1
A20 = G (−x2 − y 2 + 2z 2 ) dM ,
2
earth

A21 = G x z dM , B21 = G y z dM , (2–86)
earth earth

1 1
A22 = G (x2 − y 2 ) dM , B22 = G x y dM .
4 2
earth earth
It is known from mechanics that

1 1 1
xc = x dM , yc = y dM , zc = z dM (2–87)
M M M
are the rectangular coordinates of the center of gravity (center of mass,
geocenter). If the origin of the coordinate system coincides with the center
of gravity, then these coordinates and, hence, the integrals (2–85) are zero.
If the origin r = 0 is the center of gravity of the earth, then there will be
no first-degree terms in the spherical-harmonic expansion of the potential V .
Therefore, this is true for our geocentric coordinate system.
The integrals

x y dM , y z dM , z x dM (2–88)
are the products of inertia. They are zero if the coordinate axes coincide with
the principal axes of inertia. If the z-axis is identical with the mean rotational
axis of the earth, which coincides with the axis of maximum inertia, at least
the second and third of these products of inertia must vanish. Hence, A21 and
B21 will be zero, but not so B22 , which is proportional to the first product of
2.6 Harmonics of lower degree 63
inertia; B22 would vanish only if the earth had complete rotational symmetry
or if a principal axis of inertia happened to coincide with the Greenwich
meridian.
The five harmonics A10 R10 , A11 R11 , B11 S11 , A21 R21 , and B21 S21 –
all first-degree harmonics and those of degree 2 and order 1 – which must,
thus, vanish in any spherical-harmonic expansion of the earth’s potential,
are called forbidden or inadmissible harmonics.
Introducing the moments of inertia with respect to the x-, y-, z-axes by
the definitions
A= (y 2 + z 2 ) dM ,

B= (z 2 + x2 ) dM , (2–89)

C= (x2 + y 2 ) dM ,
and denoting the xy-product of inertia, which cannot be said to vanish, by

D= x y dM , (2–90)
we finally have
A00 = GM ,
A10 = A11 = B11 = 0 ,

A20 = G (A + B)/2 − C ,
(2–91)
A21 = B21 = 0 ,
A22 = 14 G (B − A) ,
B22 = 12 G D .
Now let the x- and y-axes actually coincide with the corresponding prin-
cipal axes of inertia of the earth. This is only theoretically possible, since
the principal axes of inertia of the earth are only inaccurately known. Then
B22 = 0; taking into account (2–78) and (2–79), we may write explicitly

GM G 1
V = + 3 C − (A + B)/2 (1 − 3 cos2 ϑ) +
r r 2
(2–92)
3
(B − A) sin ϑ cos 2λ + O(1/r 4 ) .
2
4
In rectangular coordinates this assumes the symmetrical form

GM G
V = + 5 (B + C − 2A) x2 + (C + A − 2B) y 2 +
r 2r
(2–93)

(A + B − 2C) z 2 + O(1/r 4 ) ,
which is obtained by taking into account the relations (1–26) between rect-
angular and spherical coordinates.
Terms of order higher than 1/r 3 may be neglected for larger distances
(say, for the distance to the moon), so that (2–92) or (2–93), omitting the
higher-order terms 0(1/r 4 ), are sufficient for many astronomical purposes,
cf. Moritz and Mueller (1987). Note that the notation 0(1/r 4 ) means terms
of the order of 1/r 4 . For planetary distances even the first term,
GM
V = , (2–94)
r
is generally sufficient; it represents the potential of a point mass. Thus, for
very large distances, every body acts like a point mass.
Using the form (2–78) of the spherical-harmonic expansion of V , then
the coefficients of lower degree are obtained from (2–79) and (2–91). We find
C10 = C11 = S11 = 0 ,
C − (A + B)/2
C20 = − ,
M a2
C21 = S21 = 0 , (2–95)
B−A
C22 = ,
4M a2
D
S22 = .
2M a2
The first of these formulas shows that the summation in (2–78) actually
begins with n = 2; the others relate the coefficients of second degree to the
mass and the moments and products of inertia of the earth.
2.7 The gravity field of the level ellipsoid

As a first approximation, the earth is a sphere; as a second approximation,
it may be considered an ellipsoid of revolution. Although the earth is not an
2.7 The gravity field of the level ellipsoid 65
exact ellipsoid, the gravity field of an ellipsoid is of fundamental practical

importance because it is easy to handle mathematically and the deviations
of the actual gravity field from the ellipsoidal “normal” field are so small that
they can be considered linear. This splitting of the earth’s gravity field into a
“normal” and a remaining small “disturbing” field considerably simplifies the
problem of its determination; the problem could hardly be solved otherwise.
Therefore, we assume that the normal figure of the earth is a level ellip-
soid, that is, an ellipsoid of revolution which is an equipotential surface of
a normal gravity field. This assumption is necessary because the ellipsoid is
to be the normal form of the geoid, which is an equipotential surface of the
actual gravity field. Denoting the potential of the normal gravity field by
U = U (x, y, z) , (2–96)
we see that the level ellipsoid, being a surface U = constant, exactly corre-
sponds to the geoid, which is defined as a surface W = constant.
The basic point here is that by postulating that the given ellipsoid be
an equipotential surface of the normal gravity field, and by prescribing the
total mass M , we completely and uniquely determine the normal potential
U . The detailed density distribution inside the ellipsoid, which produces the
potential U , is quite uninteresting and need not be known at all. In fact,
we do not know of any “reasonable” mass distribution for the level ellipsoid
(Moritz 1990: Chap. 5). Pizzetti (1894) unsuccessfully used a homogeneous
density distribution combined with a surface layer of negative density, which
is quite “unnatural”.
This determination is possible by Dirichlet’s principle (Sect. 1.12): The
gravitational potential outside a surface S is completely determined by know-
ing the geometric shape of S and the value of the potential on S. Originally
it was shown only for the gravitational potential V , but it can be applied to
the gravity potential
U = V + 12 ω 2 (x2 + y 2 ) (2–97)
as well if the angular velocity ω is given. The proof follows that in Sect. 1.12,
with obvious modifications. Hence, the normal potential function U (x, y, z)
is completely determined by
1. the shape of the ellipsoid of revolution, that is, its semiaxes a and b,
2. the total mass M , and
3. the angular velocity ω.
The calculation will now be carried out in detail. The given ellipsoid S0 ,
x2 + y 2 z 2
+ 2 = 1, (2–98)
a2 b
is by definition an equipotential surface
U (x, y, z) = U0 . (2–99)
It is now convenient to introduce the ellipsoidal-harmonic coordinates u, β, λ

of Sect. 1.15. The ellipsoid S0 is taken as the reference ellipsoid u = b.
Since V (u, β), the gravitational part of the normal potential U , will be
harmonic outside the ellipsoid S0 , we use the second equation of the series
(1–174). The field V has rotational symmetry and, hence, does not depend
on the longitude λ. Therefore, all nonzonal terms, which depend on λ, must
be zero, and there remains
u
∞
n i
Q
V (u, β) = E An Pn (sin β) , (2–100)
b
n=0 Qn i
E
where
E= a2 − b2 (2–101)
is the linear eccentricity. The centrifugal potential Φ(u, β) is given by
1
Φ(u, β) = 2 ω 2 (u2 + E 2 ) cos2 β . (2–102)
Therefore, the total normal gravity potential may be written

u
∞
n iQ
U (u, β) = E An Pn (sin β) + 12 ω 2 (u2 + E 2 ) cos2 β . (2–103)
b
n=0 Qn i
E
On the ellipsoid S0 we have u = b and U = U0 . Hence,
∞

An Pn (sin β) + 12 ω 2 (u2 + E 2 ) cos2 β = U0 . (2–104)
n=0
This equation applies for all points of S0 , that is, for all values of β. Since
b2 + E 2 = a2 (2–105)
and
cos2 β = 2
3 1 − P2 (sin β) , (2–106)
we have
∞

An Pn (sin β) + 1
3 ω 2 a2 − 13 ω 2 a2 P2 (sin β) − U0 = 0 (2–107)
n=0
2.7 The gravity field of the level ellipsoid 67
or

A0 + 13 ω 2 a2 − U0 P0 (sin β) + A1 P1 (sin β)
∞

+ A2 − 1
3 ω 2 a2 P2 (sin β) + An Pn (sin β) = 0 . (2–108)
n=3
This equation applies for all values of β only if the coefficient of every
Pn (sin β) is zero. Thus, we get
A0 = U0 − 13 ω 2 a2 , A1 = 0 ,
(2–109)
1
A2 = 3 ω 2 a2 , A3 = A4 = . . . = 0 .
Substituting these relations into (2–100) gives
u u

Q0 i Q2 i
V (u, β) = U0 − 13 ω 2 a2 E + 13 ω 2 a2 E P2 (sin β) . (2–110)
b b
Q0 i Q2 i
E E
This formula is basically the solution of Dirichlet’s problem for the level
ellipsoid, but we can give it more convenient forms. It is a closed formula!
First, we determine the Legendre functions of the second kind, Q0 and
Q2 . As
1 1
coth−1 (i x) = cot−1 x = −i tan−1 , (2–111)
i x
we find by (1–80) with z = i u/E:
u E
Q0 i = −i tan−1 ,
E u
(2–112)
u i u2 −1 E u
Q2 i = 1 + 3 2 tan −3 .
E 2 E u E
By introducing in (2–112) the abbreviations

1 u2 −1 E u
q = 1 + 3 2 tan −3 ,
2 E u E
(2–113)
1 b2 −1 E b
q0 = 1 + 3 2 tan −3
2 E b E
and substituting them in equation (2–110), we obtain
E

tan−1
u + q
V (u, β) = U0 − 13 ω 2 a2 1
3 ω 2 a2 q0 P2 (sin β) . (2–114)
E
tan−1
b
Now we can express U0 in terms of the mass M . For large values of u, we

have
E E
tan−1 = + O(1/u3 ) . (2–115)
u u
From the expressions (1–26) for spherical coordinates and from equations
(1–151) for ellipsoidal-harmonic coordinates, we find
x2 + y 2 + z 2 = r 2 = u2 + E 2 cos2 β , (2–116)
so that for large values of r we have

1 1
= + O(1/r 3 ) (2–117)
u r
and
E E
tan−1 = + O(1/r 3 ) , (2–118)
u r
where O(x) means “small of order x”, i.e., small of order 1/r 3 in our case.
For very large distances r, the first term in (2–114) is dominant, so that
asymptotically

E 1
V = U0 − 13 ω 2 a2 + O(1/r 3 ) . (2–119)
tan−1 (E/b) r
We know from Sect. 2.6 that
GM
V = + O(1/r 3 ) . (2–120)
r
Substituting this expression for V into the left-hand side of (2–119) yields
GM
E 1
= U0 − 1
ω 2 a2 + O(1/r 3 ) . (2–121)
r 3 tan−1 (E/b) r
Now multiply this equation by r and let then r → 0. The result is (rigor-
ously!)

E
GM = U0 − 13 ω 2 a2 −1 , (2–122)
tan (E/b)
which may be rearranged to
GM E
U0 = tan−1 + 13 ω 2 a2 . (2–123)
E b
This is the desired relation between mass M and potential U0 .
Substituting the result for U0 obtained in (2–123) into (2–114), simplifies
the expression for V to
GM E q
V = tan−1 + 13 ω 2 a2 P2 (sin β) . (2–124)
E u q0
2.8 Normal gravity 69
Expressing P2 as
P2 (sin β) = 3
2 sin2 β − 1
2 (2–125)
and, finally, adding the centrifugal potential Φ = ω 2 (u2 + E 2 ) cos2 β/2 from
(2–102), the normal gravity potential U results as
GM E q
2
U (u, β) = tan−1 + 12 ω 2 a2 sin β − 13 + 1
2 ω 2 (u2 + E 2 ) cos2 β .
E u q0
(2–126)
The only constants that occur in this formula are a, b, GM , and ω. This is
in complete agreement with Dirichlet’s theorem.
2.8 Normal gravity

Referring to the line element in ellipsoidal-harmonic coordinates according
to (1–155), replacing ϑ by its complement 90◦ − β, we get
ds2 = w2 du2 + w2 (u2 + E 2 ) dβ 2 + (u2 + E 2 ) cos2 β dλ2 , (2–127)
where
u2 + E 2 sin2 β
w= (2–128)
u2 + E 2
has been introduced. Thus, along the coordinate lines we have
u = variable, β = constant, λ = constant, dsu = w du ,

√
β = variable, u = constant, λ = constant, dsβ = w u2 + E 2 dβ ,
√
λ = variable, u = constant, β = constant, dsλ = u2 + E 2 cos β dλ .
(2–129)
The components of the normal gravity vector
γ = grad U (2–130)
along these coordinate lines are accordingly given by
∂U 1 ∂U
γu = = ,
∂su w ∂u
∂U 1 ∂U
γβ = = √ , (2–131)
∂sβ w u + E ∂β
2 2
∂U 1 ∂U
γλ = =√ = 0.
∂sλ u + E cos β ∂λ
2 2
The component γλ is zero because U does not contain λ. This is also evident
from the rotational symmetry.
Performing the partial differentiations, we find
GM ω 2 a2 E q
1
−w γu = 2 2
+ 2 2 2 sin2 β − 16 − ω 2 u cos2 β ,
u +E u + E q0
(2–132)
ω 2 a2 q
−w γβ = − √ 2 2 2
+ ω u + E sin β cos β ,
u2 + E 2 q0
where we have set

u2 + E 2 dq u2 u E
q = − =3 1+ 2 1− tan−1 − 1. (2–133)
E du E E u
Note that q does not mean dq/du; this notation has been borrowed from Hir-
vonen (1960), where q is the derivative with respect to another independent
variable which we are not using here.
For the level ellipsoid S0 itself, we have u = b and get
γβ,0 = 0 . (2–134)
(Note that we will often mark quantities referred to S0 by the subscript

0.) This is also evident because on S0 the gravity vector is normal to the
level surface S0 . Hence, in addition to the λ-component, the β-component
is also zero on the reference ellipsoid u = b. Note that the other coordinate
ellipsoids u = constant are not equipotential surfaces U = constant, so that
the β-component will not in general be zero.
Thus, the total gravity on the ellipsoid S0 , which we simply denote by
γ, is given by
GM
γ = |γu,0 | = ·
a a sin2 β + b2 cos2 β
2
(2–135)
ω 2 a2 E q0
1 ω 2 a2 b
· 1+ sin β − 6 −
2 1 2
cos β ,
GM q0 2 GM
since on S0 we get the relations
√ √
u2 + E 2 = b2 + E 2 = a ,
(2–136)
1 2 1
w0 = 2 2
b + E sin β = a2 sin2 β + b2 cos2 β .
a a
Now we introduce the abbreviation
ω 2 a2 b
m= (2–137)
GM
2.8 Normal gravity 71
and the second eccentricity

√
E a2 − b2
e = = . (2–138)
b b
The prime on e does not denote differentiation, but merely distinguishes the
second eccentricity from the first eccentricity which is defined as e = E/a.
Removing the constant terms by noting that
1 = cos2 β + sin2 β , (2–139)
we obtain
GM
γ= ·
a a2 sin2 β + b2 cos2 β
(2–140)
m e q0 m e q0
· 1+ sin β + 1 − m −
2 2
cos β .
3 q0 6 q0
At the equator (β = 0), we find

GM m e q0
γa = 1−m− ; (2–141)
ab 6 q0
at the poles (β = ±90◦ ), normal gravity is given by

GM m e q0
γb = 2 1+ . (2–142)
a 3 q0
Normal gravity at the equator, γa , and normal gravity at the pole, γb , satisfy
the relation
a − b γb − γa ω2b e q0
+ = 1+ , (2–143)
a γa γa 2q0
which should be verified by substitution. This is the rigorous form of an im-
portant approximate formula published by Clairaut in 1738. It is, therefore,
called Clairaut’s theorem. Its significance will become clear in Sect. 2.10.
By comparing expression (2–141) for γa and expression (2–142) for γb
with the quantities within parentheses in formula (2–140), we see that γ can
be written in the symmetrical form
a γb sin2 β + b γa cos2 β
γ= . (2–144)
a2 sin2 β + b2 cos2 β
We finally introduce the ellipsoidal latitude on the ellipsoid, ϕ, which is the
angle between the normal to the ellipsoid and the equatorial plane (Fig. 2.11).
Using the formula from ellipsoidal geometry,
b
¯
' '
a O
N
Fig. 2.11. Ellipsoidal latitude ϕ, geocentric latitude ϕ̄, reduced

(ellipsoidal-harmonic) latitude β for a point P on the ellipsoid
b
tan β = tan ϕ , (2–145)
a
we obtain
a γa cos2 ϕ + b γb sin2 ϕ
γ= . (2–146)
a2 cos2 ϕ + b2 sin2 ϕ
The computation is left as an exercise for the reader. This rigorous formula
for normal gravity on the ellipsoid is due to Somigliana from 1929.
We close this section with a short remark on the vertical gradient of
gravity at the reference ellipsoid, ∂γ/∂su = ∂γ/∂h. Bruns’ formula (2–40),
applied to the normal gravity field with the corresponding ellipsoidal height
h and with = 0, yields
∂γ
= −2γ J − 2ω 2 . (2–147)
∂h
The mean curvature of the ellipsoid is given by

1 1 1
J= + , (2–148)
2 M N
where M and N are the principal radii of curvature: M is the radius in the
direction of the meridian, and N is the normal radius of curvature, taken
in the direction of the prime vertical. From ellipsoidal geometry, we use the
formulas
c c
M= 2 2 3/2
, N= 2
, (2–149)
(1 + e cos ϕ) (1 + e cos2 ϕ)1/2
2.9 Expansion of the normal potential in spherical harmonics 73
where
a2
c= (2–150)
b
is the radius of curvature at the pole. The normal radius of curvature, N ,
admits a simple geometrical interpretation (Fig. 2.11). It is, therefore, also
known as the “normal terminated by the minor axis” (Bomford 1962: p. 497).
2.9 Expansion of the normal potential in spherical

harmonics
We have found the gravitational potential of the normal figure of the earth
in terms of ellipsoidal harmonics in (2–124) as
GM E 1 q
V = tan−1 + ω 2 a2 P2 (sin β) . (2–151)
E u 3 q0
Now we wish to express this equation in terms of spherical coordinates
r, ϑ, λ.
We first establish a relation between ellipsoidal-harmonic and spherical
coordinates. By comparing the rectangular coordinates in these two systems
according to Eqs. (1–26) and (1–151), we get
√
r sin ϑ cos λ = u2 + E 2 cos β cos λ ,
√
r sin ϑ sin λ = u2 + E 2 cos β sin λ , (2–152)
r cos ϑ = u sin β .
The longitude λ is the same in both systems. We easily find from these
equations
u
cot ϑ = √ tan β ,
u + E2
2
(2–153)

2 2 2
r = u + E cos β .
The direct transformation of (2–151) by expressing u and β in terms of
r and ϑ by means of equations (2–153) is extremely laborious. However, the
problem can be solved easily in an indirect way.
We expand tan−1 (E/u) into the well-known power series

E E 1 E 3 1 E 5
tan−1 = − + − ... . (2–154)
u u 3 u 5 u
The substitution of this series into the first equation of formula (2–113), i.e.,

1 u2 −1 E u
q= 1 + 3 2 tan −3 , (2–155)
2 E u E
leads, after simple manipulations, to

3 5 7
1 E 2 E 3 E
q=2 − + − ... . (2–156)
3·5 u 5·7 u 7·9 u
More concisely, we have

∞ 2n+1
E E 1 E
tan−1 = + (−1)n ,
u u 2n + 1 u
n=1
∞ 2n+1 (2–157)
2n E
q=− (−1)n .
(2n + 1)(2n + 3) u
n=1
By inserting these relations into (2–151) we obtain

∞ 2n+1
GM GM n 1 E
V = + (−1)
u E 2n + 1 u
n=1
∞ 2n+1
ω 2 a2 2n E
− (−1)n P2 (sin β) .
3q0 n=1 (2n + 1)(2n + 3) u
(2–158)
Introducing m, defined by (2–137), and the second eccentricity e = E/b, we
find 2n+1
∞
GM GM E
V = + (−1)n ·
u (2n + 1) E u
n=1

m e 2n
· 1− P2 (sin β) . (2–159)
3q0 2n + 3
We expand the potential V into a series of spherical harmonics. Because

of the rotational symmetry, there will be only zonal terms, and because of
the symmetry with respect to the equatorial plane, there will be only even
zonal harmonics. The zonal harmonics of odd degree change sign for negative
latitudes and must, therefore, be absent. Accordingly, the series has the form
GM P2 (cos ϑ) P4 (cos ϑ)
V = + A2 3
+ A4 + ··· . (2–160)
r r r5
We next have to determine the coefficients A2 , A4 , . . . . For this purpose, we
consider a point on the axis of rotation, outside the ellipsoid. For this point,
we have β = 90◦ , ϑ = 0◦ , and, by (2–153), u = r. Then (2–159) becomes
∞
GM n GM E
2n 2n m e 1
V = + (−1) 1− 2n+1
, (2–161)
r 2n + 1 2n + 3 3q0 r
n=1
2.9 Expansion of the normal potential in spherical harmonics 75
and (2–160) takes the form

∞
GM A2 A4 GM 1
V = + 3 + 5 + ··· = + A2n 2n+1 . (2–162)
r r r r n=1
r
Here we have used the fact that for all values of n
Pn (1) = 1 (2–163)
(see also Fig. 1.4). Comparing the coefficients in both expressions for V , we
2n
find
n GM E 2n m e
A2n = (−1) 1− . (2–164)
2n + 1 2n + 3 3q0
Equations (2–160) and (2–164) give the desired expression for the potential
of the level ellipsoid as a series of spherical harmonics.
The second-degree coefficient A2 is
A2 = G (A − C) . (2–165)
This follows from (2–91) by using A = B for reasons of symmetry. The C

is the moment of inertia with respect to the axis of rotation, and A is the
moment of inertia with respect to any axis in the equatorial plane. By letting
n = 1 in (2–164), we obtain

1 2 m e
A2 = − GM E 1 − 2
. (2–166)
3 15 q0
Comparing this with the preceding Eq. (2–165), we find

1 2 m e
G (C − A) = GM E 1 − 2
. (2–167)
3 15 q0
Thus, the difference between the principal moments of inertia is expressed
in terms of “Stokes’ constants” a, b, M , and ω.
It is possible to eliminate q0 from Eqs. (2–164) and (2–167), obtaining

n 3GM E 2n C −A
A2n = (−1) 1 − n + 5n . (2–168)
(2n + 1)(2n + 3) M E2
If we write the potential V in the form
a 2 a 4
GM
V = 1 + C2 P2 (cos ϑ) + C4 P4 (cos ϑ) + · · ·
r r r
(2–169)
GM ∞ a 2n
= 1+ C2n P2n (cos ϑ) ,
r r
n=1
then the C2n are given by

n 3e2n C −A
C2n = −J2n = (−1) 1 − n + 5n . (2–170)
(2n + 1)(2n + 3) M E2
Here we have introduced the first eccentricity e = E/a. For n = 1 this gives
the important formula
C −A
C20 = − (2–171)
M a2
or, equivalently,
C−A
J2 = , (2–172)
M a2
which is in agreement with the respective relation in (2–95) when taking into
account the rotational symmetry causing A = B.
Finally, we note that on eliminating q0 = (1/i) Q2 (i(b/E)) by using
Eq. (2–167), and U0 by using Eq. (2–122), we may write the expansion of V
in ellipsoidal harmonics, Eq. (2–110), in the form
i u
V (u, β) = GM Q0 i
E E
15i
u
+ G C − A − 1
3 M E 2 Q
2 i P2 (sin β) .
2E 3 E
(2–173)
This shows that the coefficients of the ellipsoidal harmonics of degrees zero
and two are functions of the mass and of the difference between the two
principal moments of inertia. The analogy to the corresponding spherical-
harmonic coefficients (2–91) is obvious. This is a closed formula, not a trun-
cated series!
2.10 Series expansions for the normal gravity field

Since the earth ellipsoid is very nearly a sphere, the quantities
√
E = a2 − b2 , linear eccentricity,
E
e = , first (numerical) eccentricity,
a
E (2–174)
e = , second (numerical) eccentricity,
b
a−b
f = , flattening,
a
2.10 Series expansions for the normal gravity field 77
and similar parameters that characterize the deviation from a sphere are
small. Therefore, series expansions in terms of these or similar parameters
will be convenient for numerical calculations.
Linear approximation
In order that the readers may find their way through the subsequent practical
formulas, we first consider an approximation that is linear in the flattening f .
Here we get particularly simple and symmetrical formulas which also exhibit
plainly the structure of the higher-order expansions.
It is well known that the radius vector r of an ellipsoid is approximately
given by
r = a (1 − f sin2 ϕ) . (2–175)
As we will see subsequently, normal gravity may, to the same approximation,
be written
γ = γa (1 + f ∗ sin2 ϕ) . (2–176)
For ϕ = ±90◦ , at the poles, we have r = b and γ = γb . Hence, we may write
b = a (1 − f ) , γb = γa (1 + f ∗ ) , (2–177)
and solving for f and f ∗ , we obtain
a−b
f = ,
a
(2–178)
γb − γa
f∗ = ,
γa
so that f is the flattening defined by (2–174), and f ∗ is an analogous quantity
which may be called gravity flattening.
To the same approximation, (2–143) becomes
f + f∗ = 5
2 m, (2–179)
where
. ω2a centrifugal force at equator
m= = . (2–180)
γa gravity at equator
This is Clairaut’s theorem in its original form. It is one of the most striking
formulas of physical geodesy: the (geometrical) flattening f in (2–178) can
be derived from f ∗ and m, which are purely dynamical quantities obtained
by gravity measurements; that is, the flattening of the earth can be obtained
from gravity measurements.
Clairaut’s formula is only a first approximation and must be improved,
first by the inclusion of higher-order ellipsoidal terms in f , and secondly by
taking into account the deviation of the earth’s gravity field from the normal
gravity field. But the principle remains the same.
Second-order expansion
We now expand the closed formulas of the two preceding sections into series
in terms of the second numerical eccentricity e and the flattening f , in
general up to and including e4 or f 2 . Terms of the order of e6 or f 3 and
higher will usually be neglected.
We start from the series

E E 1 E 3 1 E 5 1 E 7
tan−1 = − + − + ··· ,
u u 3 u 5 u 7 u
3 5 7
1 E 2 E 3 E
q=2 − + − ··· , (2–181)
3·5 u 5·7 u 7·9 u
3 5 7
1 E 1 E 1 E
q = 6 − + − ··· .
3·5 u 5·7 u 7·9 u
The first two series have already been used in the preceding section in (2–
154) and (2–156), respectively; the third is obtained by substituting the
tan−1 series into the closed formula (2–133) for q .
On the reference ellipsoid S0 , we have u = b and
E E
= = e , (2–182)
u b
so that
tan−1 e = e − 13 e3 + 1
5 e5 · · · ,

q0 = 2
15 e3 1 − 67 e2 · · · ,

(2–183)
q0 = 25 e2 1 − 37 e2 · · · ,
e q0

= 3 1 + 37 e2 · · · .
q0
We also need the series
a

b= √ = a 1 − 12 e2 + 38 e4 · · · . (2–184)
1+e2
Potential and gravity

By substituting these expressions into the closed formulas (2–123), (2–141),
(2–142), and (2–143), we obtain, up to and including the order e4 , the fol-
lowing relations.
Potential:
GM

U0 = 1 − 13 e2 + 15 e4 + 13 ω 2 a2 . (2–185)
b
Gravity at the equator and the pole:
GM

γa = 1− 3
2 m− 3
14 e2 m ,
ab
(2–186)
GM

γb = 2 1 + m + 3
7 e2 m .
a
Clairaut’s theorem:

5 ω2b
∗ 9 2
f +f = 1+ e . (2–187)
2 γa 35
The ratio ω 2 a/γa may be expressed as
ω2a 3
=m+ 2 m2 , (2–188)
γa
which is a more accurate version of (2–180).

From the first equation of (2–186), we find

3 2

GM = a b γa 1 + 32 m + 14 e m+ 9
4 m2 , (2–189)
which gives the mass in terms of equatorial gravity. Using this equation, we
can express GM in Eq. (2–185) in terms of γa , obtaining

U0 = a γa 1 − 13 e2 + 11 1 4 2 2
6 m+ 5e − 7e m+ 4 m .
11 2 (2–190)
Here we have eliminated ω 2 a by replacing it with GM m/b.

Now we can turn to Eq. (2–146) for normal gravity. A simple manipula-
tion yields
1 + b γba−a
γb
γa
sin2 ϕ
γ = γa . (2–191)
1 − a a−b
2 2 2
2 sin ϕ
The denominator is expanded into a binomial series:
1
√ = 1 + 12 x + 38 x2 + · · · . (2–192)
1−x
Then the abbreviated series

a2 − b2 e2
= = e2 − e4 ,
a2 1 + e2
(2–193)
b γb − a γa
= −e2 + 52 m2 + e4 − 13
7 e2 m + 15
4 m2
a γa
are introduced and we obtain, upon substitution,

γ = γa 1 + − 12 e2 + 52 m + 12 e 4 − 13 2
7 e m+
15
4 m2 sin2 ϕ
(2–194)

+ − 1
8 e4 + 5
4 e2 m 4
sin ϕ .
We may also express these quantities in terms of the flattening f by substi-

tuting the equation
1
e2 = − 1 = 2f + 3f 2 + · · · . (2–195)
(1 − f )2
The flattening f is most commonly used; it offers a slight advantage over

the second eccentricity e in that it is of the same order of magnitude as m:
it is not immediately apparent that m2 , e2 m, and e4 are quantities of the
same order of magnitude. We obtain

GM = a b γa 1 + 32 m + 37 f m + + 94 m2 , (2–196)

6 m− 5f − 7fm+ 4 m ,
U0 = a γa 1 − 23 f + 11 1 2 4 11 2 (2–197)

γ = γa 1 + − f + 52 m + 12 f 2 − 26 15
7 fm+ 4 m
2 sin2 ϕ
(2–198)

1 2 5
+ − 2 f + 2 f m sin ϕ . 4
The last formula is usually abbreviated as
γ = γa (1 + f2 sin2 ϕ + f4 sin4 ϕ) , (2–199)
so that we have
f2 = −f + 52 m + 12 f 2 − 26
7 fm+ 15
4 m2 ,
(2–200)
f4 = − 12 f2 + 5
2 f m.
By substituting
sin4 ϕ = sin2 ϕ − 1
4 sin2 2ϕ , (2–201)
we finally obtain
γ = γa (1 + f ∗ sin2 ϕ − 1
4 f4 sin2 2ϕ) , (2–202)
where
γb − γa
f∗ = = f2 + f4 (2–203)
γa
is the “gravity flattening”.
Coefficients of spherical harmonics

Equation (2–167) for the principal moments of inertia yields at once
C −A 1 2 m e
= − . (2–204)
M E2 3 45 q0
Expanding q0 by means of (2–183), we find
C −A 1

2
= 2 13 e2 − 13 m − 27 e2 m . (2–205)
ME e
Substituting this into (2–170) yields
C − A 1 2
−C20 = J2 = =3e − 1
3 m − 13 e4 + 1
21 e2 m
M E2 (2–206)
= f − m−
2
3
1
3
1
3 f2 + 2
21 f m,
−C40 = J4 = − 15 e4 + 2
7 e2 m = − 45 f 2 + 47 f m . (2–207)
The higher C or J, respectively, are already of an order of magnitude that
we have neglected.
Gravity above the ellipsoid

Denoting the height above the ellipsoid as ellipsoidal height h, then, in case
of a small height, the normal gravity γh at this height can be expanded into
a series in terms of h:
∂γ 1 ∂2γ 2
γh = γ + h+ h + ··· , (2–208)
∂h 2 ∂h2
where γ and its derivatives are referred to the ellipsoid, where h = 0.
The first derivative ∂γ/∂h may be obtained by applying Bruns’ formula
(2–147) together with (2–148) to the ellipsoidal height h (instead of H):

∂γ 1 1
= −γ + − 2ω 2 , (2–209)
∂h M N
where M, N are the principal radii of curvature of the ellipsoid, defined by

(2–149). Since
1 b
3/2 b

= 2 1 + e2 cos2 ϕ = 2 1 + 32 e2 cos2 ϕ · · · ,
M a a
(2–210)
1 b
1/2 b

= 2 1 + e2 cos2 ϕ = 2 1 + 12 e2 cos2 ϕ · · · ,
N a a
we have
1 1 b
2b
+ = 2 2 + 2e2 cos2 ϕ = 2 (1 + 2f cos2 ϕ) . (2–211)
M N a a
Here we have limited ourselves to terms linear in f , since the elevation h is
already a small quantity. Thus, we find from (2–209) after simple manipula-
tions:
∂γ 2γ
= − (1 + f + m − 2f sin2 ϕ) . (2–212)
∂h a
The second derivative ∂ 2 γ/∂h2 may be taken from the spherical approxima-
tion, obtained by neglecting e2 or f :
GM ∂γ ∂γ 2GM ∂2γ ∂2γ 6GM

γ= , = =− 3 , = = , (2–213)
a2 ∂h ∂a a ∂h2 ∂a2 a4
so that
∂2γ 6γ
2
= 2. (2–214)
∂h a
Thus we obtain

2 3 2
γh = γ 1 − (1 + f + m − 2f sin ϕ) h + 2 h .
2
(2–215)
a a
Using Eq. (2–198) for γ, we may also write the difference γh − γ in the form
2γa
3γa
γh − γ = − 1 + f + m + − 3f + 52 m sin2 ϕ) h + 2 h2 . (2–216)
a a
The symbol γh denotes the normal gravity for a point at latitude ϕ, situated
at height h above the ellipsoid; γ is the gravity at the ellipsoid itself, for the
same latitude ϕ, as given by (2–202) or equivalent formulas.
Second-order series developments for the inner gravity field are found in
Moritz (1990: Chap. 4); this is the main reason for such a development here,
because today one uses the closed formulas wherever possible.
2.11 Reference ellipsoid – numerical values 83
2.11 Reference ellipsoid – numerical values

Some history
The reference ellipsoid and its gravity field are completely determined by four
constants. Before the satellite era, one took the following four parameters:
a ... semimajor axis ,
f ... flattening ,
(2–217)
γa ... equatorial gravity ,
ω ... angular velocity .
The values best known and most widely used have been those of the Inter-
national Ellipsoid:
a = 6 378 388.000 m ,
f = 1/297.000 ,
(2–218)
γa = 978.049 000 gal ,
ω = 0.729 211 51 · 10−4 s−1 .
The geometric parameters a and f were determined by Hayford in 1909 from
isostatically reduced astrogeodetic data in the United States. They were
adopted for the International Ellipsoid by the assembly of the International
Association of Geodesy (IAG) at Madrid in 1924. The equatorial gravity
value γa was computed by Heiskanen (1924, 1928) from isostatically reduced
gravity data. The corresponding international gravity formula,
γ = 978.0490 (1 + 0.005 2884 sin2 ϕ − 0.000 0059 sin2 2ϕ) gal , (2–219)
was adopted by the assembly of IAG at Stockholm in 1930; whose coefficients

were computed from the assumed values for a, f, γa , ω by Cassinis (1930)
using Eqs. (2–200), (2–202), (2–203).
All parameters of the International Ellipsoid and its gravity field can
be computed from (2–218) to any desired degree of accuracy, which merely
expresses the inner consistency. In this way, we find (rounded values)
b = 6 356 912 m ,
E = 522 976 m ,
(2–220)
e2 = 0.006 7682 ,
m = 0.003 4499 .
For the constants in the spherical-harmonic expansion of the normal

gravity field, we find the values
C −A
−C20 = J2 = = 0.001 0920 ,
M a2 (2–221)
−C40 = J4 = −0.000 002 43 .
The change of normal gravity with elevation is given by the formula

(2–216), which for the International Ellipsoid becomes
γh = γ − (0.308 77 − 0.000 45 sin2 ϕ) h + 0.000 072 h2 , (2–222)
where γh and γ are measured in gal, and h is the elevation in kilometer.

Although the International Ellipsoid can no longer be considered the
closest approximation of the earth by an ellipsoid, it may still be used as
a reference ellipsoid for geodetic purposes. An official change of a reference
system must be very carefully considered because a large amount of data
may be referred to such a system.
The eastern countries have used the ellipsoid of Krassowsky:
a = 6 378 245 m ,
(2–223)
f = 1/298.3 .
Contemporary data
After the start of Sputnik, the first artificial satellite, in 1957, the Interna-
tional Astronomical Union, in 1964, adopted a new set of constants, among
them a = 6 378 160 m and f = 1/298.25. The value of a, which is consid-
erably smaller than that for the International Ellipsoid, incorporates astro-
geodetic determinations; the change in the value of J2 , and consequently of
f , is due to the results from artificial satellites.
In 1967, these values were taken by the International Union of Geodesy
and Geophysics (IUGG) as the Geodetic Reference System 1967.
This decision was soon seen to be wrong; especially the value of a was
recognized to be too large: now we believe to be on the order of 6 378 137 m,
the value of the Geodetic Reference System 1980 (GRS 1980) and, based on
it, the World Geodetic System 1984 (WGS 84). More details of these two
systems are given below.
Geodetic Reference System 1980 (GRS 1980)

The GRS 1980 has been adopted at the XVII General Assembly of the IUGG
in Canberra, December 1979, by Resolution No. 7. Inherently, this resolu-
tion recognizing that the Geodetic Reference System 1967 adopted at the
XIV General Assembly of IUGG, Lucerne, 1967, no longer represents the
size, shape, and gravity field of the earth to an accuracy adequate for many
geodetic, geophysical, astronomical, and hydrographic applications and con-
sidering that more appropriate values are now available, recommends that
the Geodetic Reference System 1967 be replaced by the new Geodetic Refer-
ence System 1980 which is also based on the theory of the geocentric equipo-
tential ellipsoid. The four defining parameters of the GRS 1980 are given in
Table 2.1. Defining parameters of the GRS 1980
Parameter and value Description

a = 6 378 137 m semimajor axis of the ellipsoid
GM = 3 986 005 · 108 m3 s−2 geocentric gravitational constant of the
earth (including the atmosphere)
J2 = 108 263 · 10−8 dynamical form factor of the earth (ex-
cluding the permanent tidal deforma-
tion)
ω = 7 292 115 · 10−11 rad s−1 angular velocity of the earth
Table 2.1. Note that these parameters, as given in the table, are defined
as exact! Note also that GM , the “geocentric gravitational constant” of the
earth, may also more figuratively be denoted as “product of the (Newtonian)
gravitational constant and the earth’s mass”.
On the basis of these defining parameters and by the computational
formulas given in Moritz (1980 b), the geometrical and physical constants of
Table 2.2 may be derived.
The GRS 1980 is still (2005) valid as the official reference system of the
IUGG and it forms the fundamental basis of the WGS 84.
World Geodetic System 1984 (WGS 84)

As just mentioned, the WGS 84 may be regarded as a descendant of the
GRS 1980. Due to its still increasing importance, we consider it appropriate
to describe the WGS 84 in some more detail.
Following the National Imagery and Mapping Agency (2000) of the USA,
the definition of the WGS 84 may be described in the following way. The
WGS 84 is a Conventional Terrestrial Reference System (CTRS). The def-
inition of this coordinate system follows the criteria as outlined by the In-
ternational Earth Rotation Service (IERS). The criteria for this system are
the following:
• it is geocentric, the center of mass being defined for the whole earth
including oceans and atmosphere;
• its scale is that of the local earth frame, in the meaning of a relativistic
theory of gravitation;
• its orientation was initially given by the Bureau International de l’Heure
(BIH) orientation of 1984.0;
• its time evolution in orientation will create no residual global rotation
with regards to the crust.
Table 2.2. GRS 1980 derived constants

Geometrical constants
b = 6 356 752.3141 m semiminor axis of the ellipsoid
E = 521 854.0097 m linear eccentricity
c = 6 399 593.6259 m polar radius of curvature
e2 = 0.006 694 380 022 90 first eccentricity squared
e2 = 0.006 739 496 775 48 second eccentricity squared
f = 0.003 352 810 681 18 flattening
1/f = 298.257 222 101 reciprocal flattening
Physical constants
U0 = 62 636 860.850 m2 s−2 normal potential at the ellipsoid
J4 = −0.000 002 370 912 22 spherical-harmonic coefficient
J6 = 0.000 000 006 083 47 spherical-harmonic coefficient
J8 = −0.000 000 000 014 27 spherical-harmonic coefficient
m = 0.003 449 786 003 08 m = ω 2 a2 b/(GM )
γa = 9.780 326 7715 m s−2 normal gravity at the equator
γb = 9.832 186 3685 m s−2 normal gravity at the pole
The WGS 84 is a right-handed, earth-fixed orthogonal coordinate system.

The origin and axes are defined in the following way:
• Origin: earth’s center of mass.
• Z-axis: the direction of the IERS Reference Pole (IRP); this direction
corresponds to the direction of the BIH Conventional Terrestrial Pole
(CTP) (epoch 1984.0). In other terms, the Z-axis is, by convention,
identical to the mean position of the earth’s rotational axis.
• X-axis: intersection of the IERS Reference Meridian (IRM) and the
plane passing through the origin and normal to the Z-axis; the IRM is
coincident with the BIH Zero Meridian (epoch 1984.0); in other terms,
the X-axis is associated with the mean Greenwich meridian.
• Y -axis: this axis completes a right-handed, earth-centered-earth-fixed
(ECEF) orthogonal coordinate system.
The WGS 84 origin also serves as the geometric center of the WGS 84 ellip-
soid and the Z-axis serves as the rotational axis of this ellipsoid of revolution.
This completes the definition of the WGS 84 as given in National Imagery
and Mapping Agency (2000). Note that the definition of the WGS 84 CTRS
has not changed in any fundamental way.
Reference frames: WGS 84 and ITRF

Now we need the distinction between definition and realization. When using
the term “coordinate system” or “reference system”, then this implies the
definition only; however, when using the term “coordinate frame”, then a
realization is implied (Mueller 1985). So far, we have only given a definition
of the WGS 84; therefore, we ought to denote this as WGS 84 CTRS. Now
we consider a realization and, therefore, use the term “coordinate frame”.
Following closely National Imagery and Mapping Agency (2000) and
Hofmann-Wellenhof et al. (2001: Sect. 3.2.1), an example of a terrestrial
reference frame is – on the basis of the previous definition – the WGS 84
reference frame (often simply denoted as WGS 84 – as we will also do).
Associated to this frame is a geocentric ellipsoid of revolution, originally de-
fined by the four parameters (1) semimajor axis a, (2) normalized second
degree zonal gravitational coefficient C̄20 , (3) truncated angular velocity of
the earth ω, and (4) earth’s gravitational constant G. This frame has been
used for GPS since 1987.
Another example for a terrestrial reference frame is the one produced
by the IERS and is called International Terrestrial Reference Frame (ITRF)
(McCarthy 1996). The definition of the axes is analogous to the WGS 84, i.e.,
the Z-axis is defined by the IERS Reference Pole (IRP) and the X-axis lies
in the IERS Reference Meridian (IRM); however, the realization differs! The
ITRF is realized by a number of terrestrial sites where temporal effects (plate
tectonics, tidal effects) are also taken into account. Thus, ITRF is regularly
updated (almost every year) and the acronym is supplemented by the last
two digits of the last year whose data were used in the formation of the
frame, e.g., ITRF89, ITRF90, ITRF91, ITRF92, ITRF93, ITRF94, ITRF95,
ITRF96, ITRF97, or the full designation of the year, e.g., ITRF2000.
The comparison of the original WGS 84 and ITRF revealed remarkable
differences (Malys and Slater 1994):
1. The WGS 84 was established through Doppler observations from the

TRANSIT satellite system, while ITRF is based on Satellite Laser
Ranging (SLR) and Very Long Baseline Interferometry (VLBI) obser-
vations. The accuracy of the TRANSIT reference stations was esti-
mated to be in the range of 1 to 2 meters, while the accuracy of the
ITRF reference stations is at the centimeter level.
2. The numerical values for the original defining parameters differ from
those in the ITRF. The only significant difference, however, was in
the earth’s gravitational constant GWGS − GITRF = 0.582 · 108 m3 s−2 ,
which resulted in measurable differences in the satellite orbits.
On the basis of this information, the former U.S. Defense Mapping Agency
(DMA) has proposed to replace the value of G in the WGS 84 by the standard
IERS value and to refine the coordinates of the GPS tracking stations. The
revised WGS 84, valid since January 2, 1994, has been given the designation
WGS 84 (G 730), where the ‘G’ indicates that the respective coordinates
used were obtained through GPS and the following number 730 indicates
the GPS week number when DMA has implemented the refined system.
In 1996, the U.S. National Imagery and Mapping Agency (NIMA) – the
successor of DMA – has implemented a revised version of the frame denoted
as WGS 84 (G 873) and being valid since September 29, 1996. The frame is
realized by monitor stations with refined coordinates. The associated ellip-
soid and its gravity field are now defined by the four parameters a, f, GM, ω,
which are slightly different compared to the respective ITRF values, e.g., the
current WGS 84 (G 873) frame and the ITRF97 show insignificant systematic
differences of less than 2 cm. Hence, they are virtually identical.
Note that the refinements applied to the WGS 84 reference frame have
reduced the uncertainties in the coordinates of the frame, the uncertainty
of the gravitational model, and the uncertainty of the geoid undulations;
however, they have not changed the WGS 84 coordinate system in the sense
of definition !
More general, the relationship between the WGS 84 and the ITRF is
characterized by two statements: (1) WGS 84 and ITRF are consistent; (2)
the differences between WGS 84 and ITRF are in the centimeter range world-
wide (National Imagery and Mapping Agency 2000).
However, if a transformation between reference frames is required, this
is accomplished by a datum transformation (see Sect. 5.7).
Numerical values for the WGS 84 (reference frame)

As mentioned at the very beginning of Sect. 2.11, the reference ellipsoid and
its gravity field are completely determined by four constants. The current
defining parameters for WGS 84 are listed in Table 2.3.
Table 2.3. Defining parameters of the WGS 84

a = 6 378 137 m semimajor axis of the ellipsoid
f = 1/298.257 223 563 flattening of the ellipsoid
GM = 3 986 004.418 · 108 m3 s−2 geocentric gravitational constant of
the earth (including the atmosphere)
ω = 7 292 115 · 10−11 rad s−1 angular velocity of the earth
Table 2.4. WGS 84 reference ellipsoid derived constants

Geometrical constants
C̄20 = −0.484 166 774 985 · 10−3 normalized second-degree harmonic
b = 6 356 752.3142 m semiminor axis of the ellipsoid
e = 8.181 919 084 2622 · 10−2 first eccentricity
e2 = 6.694 379 990 14 · 10−3 first eccentricity squared
e = 8.209 443 794 9696 · 10−2 second eccentricity
e2 = 6.739 496 742 28 · 10−3 second eccentricity squared
E = 5.218 540 084 2339 · 105 linear eccentricity
c = 6 399 593.6258 m polar radius of curvature
b/a = 0.996 647 189 335 axis ratio
Physical constants
U0 = 62 636 851.7146 m2 s−2 normal potential at the ellipsoid
γa = 9.780 325 3359 m s−2 normal gravity at the equator
γb = 9.832 184 9378 m s−2 normal gravity at the pole
γ̄ = 9.797 643 2222 m s−2 mean value of normal gravity
M = 5.973 3328 · 1024 kg mass of the earth (includes atmosphere)
m = 0.003 449 786 506 84 m = ω 2 a2 b/(GM )
Some history (even if only some years old) is important here because
the parameters selected to originally define the WGS 84 reference ellipsoid
were the semimajor axis a, the product of the earth’s mass and the grav-
itational constant GM (also denoted as “geocentric gravitational constant
of the earth”), the normalized second-degree zonal gravitational coefficient
C̄20 , and the earth’s angular velocity ω. Due to significant refinements of
these original defining parameters, the DMA recommended, e.g., a refined
value for the GM parameter.
Anyway, a decision was made to retain the original WGS 84 reference
ellipsoid values for the semimajor axis a = 6 378 137 m and for the flattening
f = 1/298.257 223 563. For this reason, the four defining parameters were
chosen to be a, f, GM, ω.
Readers who like some confusion may continue right here; otherwise skip
this short paragraph. Due to this new choice of the defining parameters,
there are in addition two distinct values for the C̄20 term, one is dynami-
cally derived and the other geometrically by the defining parameters. The
geometric derivation based on the four defining parameters a, f, GM, ω yields

C̄20 = −0.484 166 774 985 · 10−3 which differs from the original value by
7.5015 · 10−11 . For many more details refer to National Imagery and Map-
ping Agency (2000).
We conclude these considerations by a useful table. Using the four defin-
ing parameters, it is possible to derive the more commonly used geometric
constants and physical constants (Table 2.4) associated with the WGS 84
reference ellipsoid.
Numerical comparison of GRS 1980 and WGS 84

As mentioned previously, the GRS 1980 is the basis of the WGS 84. However,
due to different defining parameters on the one hand and, e.g., a refined value
for GM for the WGS 84 on the other hand, numerical differences between
the GRS 1980 and the WGS 84 arise. Some of these differences are given in
Table 2.5.
Table 2.5. Numerical comparison between GRS 1980 and WGS 84
Parameter GRS 1980 WGS 84

GM 3 986 005 · 108 m3 s−2 3 986 004.418 · 108 m3 s−2
1/f 298.257 222 101 298.257 223 563
b 6 356 752.3141 m 6 356 752.3142 m
e2 0.006 694 380 022 90 0.006 694 379 990 14
e2 0.006 739 496 775 48 0.006 739 496 742 28
E 521 854.0097 m 521 854.0084 m
c 6 399 593.6259 m 6 399 593.6258 m
U0 62 636 860.850 m2 s−2 62 636 851.7146 m2 s−2
γa 9.780 326 7715 m s−2 9.780 325 3359 m s−2
γb 9.832 186 3685 m s−2 9.832 184 9378 m s−2
m 0.003 449 786 003 08 0.003 449 786 506 84
2.12 Anomalous gravity field, geoidal undulations,

and deflections of the vertical
The small difference between the actual gravity potential W and the normal
gravity potential U is denoted by T , so that
W (x, y, z) = U (x, y, z) + T (x, y, z) ; (2–224)

2.12 Anomalous gravity field 91
n n'
P geoid
W = W0
N gP
Q reference
ellipsoid
°Q U = W0
Fig. 2.12. Geoid and reference ellipsoid
T is called the anomalous potential, or disturbing potential. We compare the

geoid
W (x, y, z) = W0 (2–225)
with a reference ellipsoid
U (x, y, z) = W0 (2–226)
of the same potential U0 = W0 . A point P of the geoid is projected onto
the point Q of the ellipsoid by means of the ellipsoidal normal (Fig. 2.12).
The distance P Q between geoid and ellipsoid is called the geoidal height, or
geoidal undulation, and is denoted by N . Unfortunately, there is a conflict of
notation here. Denoting both the normal radius of curvature of the ellipsoid
and the geoidal height by N is well established in geodetic literature. We
continue this practice, as there is little chance of confusion.
Consider now the gravity vector g at P and the normal gravity vector γ
at Q. The gravity anomaly vector ∆g is defined as their difference:
∆g = gP − γ Q . (2–227)
A vector is characterized by magnitude and direction. The difference in mag-

nitude is the gravity anomaly
∆g = gP − γQ ; (2–228)
the difference in direction is the deflection of the vertical.

The deflection of the vertical has two components, a north-south com-
ponent ξ and an east-west component η (Fig. 2.13). As the direction of the
vertical is directly defined by the astronomical coordinates latitude Φ and
longitude Λ, the components ξ and η can be expressed by them in a sim-
ple way. The actual astronomical coordinates of the geoidal point P , which
al
line
orm
||z
ln
mb
n
da
n'
plu
so i
ip
L–¸
ell
ian
L–¸ »
ri d
co
me
s'
´ F
wich
Green
P
'
¸
L
Fig. 2.13. The deflection of the vertical as illustrated by means of a

unit sphere with center at P
define the direction of the plumb line n or of the gravity vector g, can be
determined by astronomical measurements. The ellipsoidal coordinates (or
geodetic coordinates in the sense of geographical coordinates on the ellip-
soid) given by the direction of the ellipsoidal normal n have been denoted
by ϕ and λ – these coordinates should not be confused with the ellipsoidal-
harmonic coordinates of Sect. 1.15 ! It is evident that this λ is identical with
the geocentric longitude (and also with the ellipsoidal-harmonic longitude).
Thus,
geoidal normal n, astronomical coordinates Φ, Λ ;

(2–229)
ellipsoidal normal n , ellipsoidal coordinates ϕ, λ .
From Fig. 2.13, we read
ξ = Φ − ϕ,
(2–230)
η = (Λ − λ) cos ϕ .
It is also possible to compare the vectors g and γ at the same point P . Then
we get the gravity disturbance vector
δg = gP − γ P . (2–231)
Accordingly, the difference in magnitude is the gravity disturbance
δg = gP − γP . (2–232)
The difference in direction – i.e., the deflection of the vertical – is the same
as before, since the directions of γP and γQ coincide virtually.
The gravity disturbance is conceptually even simpler than the gravity
anomaly, but it has not been that important in terrestrial geodesy. The
significance of the gravity anomaly is that it is given directly: the gravity
g is measured on the geoid (or reduced to it), see Chap. 3, and the normal
gravity γ is computed for the ellipsoid.
A very important remark

So far, for historical reasons, much more gravity anomalies ∆g are available
and are being processed than gravity disturbances δg. By GPS, however, the
point P is determined rather than Q. Therefore, in future, we may expect
that δg will become more important than ∆g.
However, mirroring the present state of practice of physical geodesy, we
continue mainly to work with ∆g. Most statements about ∆g will also apply
for δg, with obvious modifications, such as with Molodensky’s corrections
(see Chap. 8), and Stokes’ formula will be replaced by Koch’s formula (see
below in this chapter).
Relations
There are several basic mathematical relations between the quantities just
defined. Since
∂U
UP = UQ + N = UQ − γ N , (2–233)
∂n Q
we have
WP = UP + TP = UQ − γ N + TP . (2–234)
Because
WP = UQ = W0 , (2–235)
we find
T =γN (2–236)
(where we have omitted the subscript P on the left-hand side) or
T
N= . (2–237)
γ
This is the famous Bruns formula, which relates the geoidal undulation to
the disturbing potential.
Next we consider the gravity disturbance. Since
g = grad W ,
(2–238)
γ = grad U ,
the gravity disturbance vector (2–231) becomes

∂T ∂T ∂T
δg = grad (W − U ) = grad T ≡ , , . (2–239)
∂x ∂y ∂z
Then
∂W ∂U . ∂U
g=− , γ=− =− , (2–240)
∂n ∂n ∂n
because the directions of the normals n and n almost coincide. Therefore,
the gravity disturbance is given by

∂W ∂U . ∂W ∂U
δg = gP − γP = − − =− − (2–241)
∂n ∂n ∂n ∂n
or
∂T
δg = − . (2–242)
∂n
Since the elevation h is reckoned along the normal, we may also write
∂T
δg = − . (2–243)
∂h
Comparing (2–242) with (2–239), we see that the gravity disturbance δg, be-
sides being the difference in magnitude of the actual and the normal gravity
vector, is also the normal component of the gravity disturbance vector δg.
We now turn to the gravity anomaly ∆g. Since
∂γ
γP = γQ + N, (2–244)
∂h
we have
∂T ∂γ
− = δg = gP − γP = gP − γQ − N. (2–245)
∂h ∂h
Remembering the definition (2–228) of the gravity anomaly and taking into
account Bruns’ formula (2–237), we find the following equivalent relations:
∂T ∂γ
− = ∆g − N, (2–246)
∂h ∂h
∂T ∂γ
∆g = − + N, (2–247)
∂h ∂h
∂T 1 ∂γ
∆g = − + T, (2–248)
∂h γ ∂h
∂γ
δg = ∆g − N, (2–249)
∂h
1 ∂γ
δg = ∆g − T, (2–250)
γ ∂h
relating different quantities of the anomalous gravity field.
Another equivalent form is
∂T 1 ∂γ
− T + ∆g = 0 . (2–251)
∂h γ ∂h
This expression has been called the fundamental equation of physical geodesy,
because it relates the measured quantity ∆g to the unknown anomalous
potential T . In future, the relation
∂T
+ δg = 0 (2–252)
∂h
may replace it.
It has the form of a partial differential equation. If ∆g were known
throughout space, then (2–251) could be discussed and solved as a real par-
tial differential equation. However, since ∆g is known only along a surface
(the geoid), the fundamental equation (2–251) can be used only as a bound-
ary condition, which alone is not sufficient for computing T . Therefore, the
name “differential equation of physical geodesy”, which is sometimes used
for (2–251), is rather misleading.
One usually assumes that there are no masses outside the geoid. This is
not really true. But neither do we make observations directly on the geoid;
we make them on the physical surface of the earth. In reducing the measured
gravity to the geoid, the effect of the masses outside the geoid is removed by
computation, so that we can indeed assume that all masses are enclosed by
the geoid (see Chaps. 3 and 8).
In this case, since the density is zero everywhere outside the geoid, the
anomalous potential T is harmonic there and satisfies Laplace’s equation
∂2T ∂2T ∂2T
∆T ≡ + + = 0. (2–253)
∂x2 ∂y 2 ∂z 2
This is a true partial differential equation and suffices, if supplemented by
the boundary condition (2–251), for determining T at every point outside
the geoid. If we write the boundary condition in the form
∂T 1 ∂γ
− + T = ∆g , (2–254)
∂n γ ∂n
where ∆g is assumed to be known at every point of the geoid, then we see that
a linear combination of T and ∂T /∂n is given upon that surface. According
to Sect. 1.13, the determination of T is, therefore, a third boundary-value
problem of potential theory. If it is solved for T , then the geoidal height,
which is the most important geometric quantity in physical geodesy, can be
computed by Bruns’ formula (2–237).
Therefore, we may say that the basic problem of physical geodesy, the
determination of the geoid from gravity measurements, is essentially a third
boundary-value problem of potential theory.
2.13 Spherical approximation and expansion of the

disturbing potential in spherical harmonics
The reference ellipsoid deviates from a sphere only by quantities of the order
.
of the flattening, f = 3 · 10−3 . Therefore, if we treat the reference ellipsoid
as a sphere in equations relating quantities of the anomalous field, this may
cause a relative error of the same order. This error is usually permissible in
N, T, ∆g, δg, etc. For instance, the absolute effect of this relative error on
the geoidal height is of the order of 3· 10−3 N ; since N hardly exceeds 100 m,
this error can usually be expected to be less than 1 m.
As a spherical approximation, we have
GM ∂γ ∂γ GM 1 ∂γ 2
γ= , = = −2 3 , =− . (2–255)
r2 ∂h ∂r r γ ∂h r
We introduce a mean radius R of the earth. It is often defined as the radius of
a sphere that has the same volume as the earth ellipsoid; from the condition
4 4
3 π R3 = 3 π a2 b , (2–256)
we get √
3
R= a2 b . (2–257)
In a similar way, we may define a mean value of gravity, γ0 , as normal gravity
at latitude ϕ = 45◦ (Moritz 1980b: p. 403). Numerical values of about
R = 6371 km, γ0 = 980.6 gal (2–258)
are usual. Then

1 ∂γ 2
=− ,
γ ∂h R
(2–259)
∂γ 2γ0
=− .
∂h R
2.13 Spherical approximation 97
Since the normal to the sphere is the direction of the radius vector r, we
have to the same approximation
∂ ∂ ∂
= = . (2–260)
∂n ∂h ∂r
In Bruns’ theorem (2–237) we may replace γ by γ0 , and Eqs. (2–246)
through (2–250) and (2–251) become
∂T 2γ0
− = ∆g + N, (2–261)
∂h R
∂T 2γ0
∆g = − − N, (2–262)
∂r R
∂T 2
∆g = − − T, (2–263)
∂r R
2γ0
δg = ∆g + N, (2–264)
R
2
δg = ∆g + T , (2–265)
R
∂T 2
+ T + ∆g = 0 . (2–266)
∂r R
The last equation is the spherical approximation of the fundamental bound-
ary condition.
Remark
The meaning of this spherical approximation should be carefully kept in
mind. It is used only in equations relating the small quantities T, N, ∆g, δg,
etc. The reference surface is never a sphere in any geometrical sense, but
always an ellipsoid. As the flattening f is very small, the ellipsoidal formulas
can be expanded into power series in terms of f , and then all terms containing
f, f 2 , etc., are neglected. In this way one obtains formulas that are rigorously
valid for the sphere, but approximately valid for the actual reference ellipsoid
as well. However, normal gravity γ in the gravity anomaly ∆g = g − γ
must be computed for the ellipsoid to a high degree of accuracy. To speak
of a “reference sphere” in space, in any geometric sense, may be highly
misleading.
Since the anomalous potential T = W − U is a harmonic function, it can be

expanded into a series of spherical harmonics:
∞ n+1
R
T (r, ϑ, λ) = Tn (ϑ, λ) . (2–267)
r
n=0
Tn (ϑ, λ) is Laplace’s surface harmonic of degree n. On the geoid, which as a

spherical approximation corresponds to the sphere r = R, we have formally
∞

T = T (R, ϑ, λ) = Tn (ϑ, λ) . (2–268)
n=0
We need not be concerned with questions of convergence here. Differentiating

the series (2–267) with respect to r, we find
∞ n+1
∂T 1 R
δg = − = (n + 1) Tn (ϑ, λ) . (2–269)
∂r r r
n=0
On the geoid, where r = R, this becomes
∞
∂T 1
δg = − = (n + 1) Tn (ϑ, λ) . (2–270)
∂r R
n=0
These series express the gravity disturbance in terms of spherical harmonics.

The equivalent of (2–263) outside the earth is
∂T 2
∆g = − − T. (2–271)
∂r r
Its exact meaning will be discussed at the end of the following section. The
substitution of (2–269) and (2–267) into this equation yields
∞ n+1
1 R
∆g = (n − 1) Tn (ϑ, λ) . (2–272)
r n=0 r
On the geoid, this becomes
∞
1
∆g = (n − 1) Tn (ϑ, λ) . (2–273)
R n=0
This is the spherical-harmonic expansion of the gravity anomaly.

Note that even if the anomalous potential T contains a first-degree spher-
ical term T1 (ϑ, λ), it will in the expression for ∆g be multiplied by the factor
1 − 1 = 0, so that ∆g can never have a first-degree spherical harmonic – even
if T has one.
2.14 Gravity anomalies outside the earth 99
2.14 Gravity anomalies outside the earth

If a harmonic function H is given at the surface of the earth, then, as a
spherical approximation, the values of H outside the earth can be computed
by Poisson’s integral formula (1–123)
2
R r − R2
HP = H dσ . (2–274)
4π l3
σ

The symbol σ is the usual abbreviation for an integral extended over the
whole unit sphere. The meaning of the other notations is read from Fig. 2.14.
The value of the harmonic function at the variable surface element R2 dσ is
denoted simply by H, whereas HP refers to the fixed point P . Then we get

l = r 2 + R2 − 2R r cos ψ . (2–275)
The harmonic function H can be expanded into a series of spherical har-

monics: 2 ∞ n+1

R R R
H= H0 + H1 + Hn . (2–276)
r r r
n=2
By omitting the terms of degrees one and zero, we get a new function
2 ∞ n+1

R R R
H = H − H0 − H1 = Hn . (2–277)
r r r
n=2
r l
Ã R 2d ¾
Ã
d¾ R
unit sphere
r=1
terrestrial sphere
r=R
Fig. 2.14. Notations for Poisson’s integral and derived formulas

The surface harmonics are given by

1 3
H0 = H dσ , H1 = H cos ψ dσ (2–278)
4π 4π
σ σ
according to equation (1–89). Hence, we find from (2–277), on expressing H

by Poisson’s integral (2–274) and substituting the integrals (2–278) for H0
and H1 , the basic formula
2
1 r − R2 1 3R
HP = − − cos ψ H dσ . (2–279)
4π l3 r r2
σ
The reason for this modification of Poisson’s integral is that the formulas
of physical geodesy are simpler if the functions involved do not contain har-
monics of degrees zero and one. It is therefore convenient to split off these
terms. This is done automatically by the modified Poisson integral (2–279).
We now apply these formulas to the gravity anomalies outside the earth.
Equation (2–272) yields at once
∞ n+1
R
r ∆g = (n − 1) Tn (ϑ, λ) . (2–280)
r
n=0
Just as Tn (ϑ, λ) is a Laplace surface harmonic, so is (n−1) Tn . Consequently,

r ∆g, considered as a function in space, can be expanded into a series of
spherical harmonics and is, therefore, a harmonic function. Hence, we can
apply Poisson’s formula to r ∆g, getting
2
R r − R2 1 3R
r ∆gP = − − 2 cos ψ R ∆g dσ (2–281)
4π l3 r r
σ
or
R2 r 2 − R2 1 3R
∆gP = − − 2 cos ψ ∆g dσ . (2–282)
4π r l3 r r
σ
This is the formula for the computation of gravity anomalies outside the
earth from surface gravity anomalies, or for the upward continuation of grav-
ity anomalies.
Finally, we discuss the exact meaning of the gravity anomaly δgP outside
the earth. We start with a convenient definition. The level surfaces of the
actual gravity potential, the surfaces
W = constant , (2–283)
2.14 Gravity anomalies outside the earth 101
geopotential
P W = WP = const.
surface
NP
Q spheropotential
U = WP = const.
surface
P0
geoid W = W0
N
Q0 ellipsoid U = W0
Fig. 2.15. Geopotential and spheropotential surfaces
are often called geopotential surfaces; the level surfaces of the normal gravity
field, the surfaces
U = constant , (2–284)
are called spheropotential surfaces.
We consider now the point P outside the earth (Fig. 2.15) and denote
the geopotential surface passing through it by
W = WP . (2–285)
There is also a spheropotential surface
U = WP (2–286)
of the same constant WP . The normal plumb line through P intersects this
spheropotential surface at the point Q, which is said to correspond to P .
We see that the level surfaces W = WP and U = WP are related to each
other in exactly the same way as are the geoid W = W0 and the reference
ellipsoid U = W0 . If, therefore, the gravity anomaly is defined by
∆gP = gP − γQ , (2–287)
as in Sect. 2.12, then all derivations and formulas of that section also apply
for the present situation, the geopotential surface W = WP replacing the
geoid W = W0 , and the spheropotential surface U = WP replacing the
ellipsoid U = W0 . This is also the reason why (2–271) applies at P as well
as at the geoid.
Note that P in Sect. 2.12 is a point at the geoid, which is denoted by P0
in Fig. 2.15.
This situation will be taken up again in Chap. 8, in the context of Molo-
densky’s problem.
2.15 Stokes’ formula

The basic Eq. (2–271),
∂T 2
∆g = −− T, (2–288)
∂r r
can be regarded as a boundary condition only, as long as the gravity anoma-
lies ∆g are known only at the surface of the earth. However, by the up-
ward continuation integral (2–282), we are now able to compute the gravity
anomalies outside the earth. Thus, our basic equation changes its meaning
radically, becoming a real differential equation that can be integrated with
respect to r. Note that this is made possible only because T , in addition to
the boundary condition, satisfies Laplace’s equation ∆T = 0.
Multiplying (2–288) by −r 2 , we get
∂T ∂ 2
−r 2 ∆g = r 2 + 2r T = (r T ) . (2–289)
∂r ∂r
Integrating the formula
∂ 2
(r T ) = −r 2 ∆g(r) (2–290)
∂r
between the limits ∞ and r, we find
r r
2
r T = − r 2 ∆g(r) dr , (2–291)
∞ ∞
where ∆g(r) indicates that ∆g is now a function of r, computed from sur-

face gravity anomalies by means of the formula (2–282). Since this formula
automatically removes the spherical harmonics of degrees one and zero from
∆g(r), the anomalous potential T , as computed from ∆g(r), cannot contain
such terms. Thus, we have
∞ n+1
R R3 R4
T = Tn = 3 T2 + 4 T3 + · · · . (2–292)
n=2
r r r
Therefore,

R3 R4
2
lim (r T ) = lim T2 + 2 T3 + · · · = 0, (2–293)
r→∞ r→∞ r r
so that r

r T = r 2 T − lim (r 2 T ) = r 2 T
2
(2–294)
∞ r→∞
and r
r T =−
2
r 2 ∆g(r) dr . (2–295)
∞
2.15 Stokes’ formula 103
The fact that r is used both as an integration variable and as an upper

limit should not cause any difficulty. Substituting the upward continuation
integral (2–282), we get
⎡ ⎤
r 3
R 2
⎣ r −R r2 3R
r2 T = − +1+ cos ψ ∆g dσ ⎦ dr . (2–296)
4π ∞ l3 r
σ
Interchanging the order of the integrations gives

r 3
R2 r − R2 r 3R
r2 T = − + 1 + cos ψ dr ∆g dσ . (2–297)
4π ∞ l3 r
σ
The integral in brackets can be evaluated by standard methods. The indefi-

nite integral is
3
r − R2 r 3R
− +1+ cos ψ dr
l3 r
2r 2
= − 3l − 3R cos ψ ln(r − R cos ψ + l) + r + 3R cos ψ ln r .
l
(2–298)
The reader is advised to perform this integration, taking into account (2–
275), or at least to check the result by differentiating the right-hand side
with respect to r.
For large values of r, we have

R
l =r 1− cos ψ · · · = r − R cos ψ · · · (2–299)
r
and, hence, we find that as r → ∞, the right-hand side of the above indefinite
integral approaches
5R cos ψ − 3R cos ψ ln 2 . (2–300)
If we subtract this from the indefinite integral, we get the definite integral,
since infinity is its lower limit of integration. Thus,
r
r 3 − R2 r 3R
− +1+ cos ψ dr
l3 r
∞ (2–301)

2r 2 r − R cos ψ + l
= + r − 3l − R cos ψ 5 + 3 ln .
l 2r
Hence, we obtain Pizzetti’s formula

R
T (r, ϑ, λ) = S(r, ψ) ∆g dσ , (2–302)
4π
σ
where

2R R R l R2 r − R cos ψ + l
S(r, ψ) = + − 3 2 − 2 cos ψ 5 + 3 ln . (2–303)
l r r r 2r
On the geoid itself, we have r = R, and denoting T (R, ϑ, λ) simply by T , we

find
R
T = ∆g S(ψ) dσ , (2–304)
4π
σ
where

1 ψ ψ ψ
S(ψ) = −6 sin +1−5 cos ψ−3 cos ψ ln sin + sin2 (2–305)
sin(ψ/2) 2 2 2
is obtained from S(r, ψ) by setting
ψ
r = R and l = 2R sin . (2–306)
2
By Bruns’ theorem, N = T /γ0 , we finally get

R
N= ∆g S(ψ) dσ . (2–307)
4π γ0
σ
This formula was published by G.G. Stokes in 1849; it is, therefore, called
Stokes’ formula, or Stokes’ integral. It is by far the most important formula
of physical geodesy because it performs to determine the geoid from gravity
data. Equation (2–304) is also called Stokes’ formula, and S(ψ) is known as
Stokes’ function.
Using formula (2–302), which was derived by Pizzetti (1911) and later on
by Vening Meinesz (1928), we can compute the anomalous potential T at any
point outside the earth. Dividing T by the normal gravity at the given point
P (Bruns’ theorem), we obtain the separation NP between the geopotential
surface W = WP and the corresponding spheropotential surface U = WP ,
which, outside the earth, takes the place of the geoidal undulation N (see
Fig. 2.15 and the explanations at the end of the preceding section).
We mention again that these formulas are based on a spherical approx-
imation; quantities of the order of 3 · 10−3 N are neglected. This results in
an error of probably less than 1 m in N , which can be neglected for many
practical purposes. Sagrebin (1956), Molodenskii et al. (1962: p. 53), Bjer-
hammar, and Lelgemann have developed higher approximations, which take
into account the flattening f of the reference ellipsoid; see Moritz (1980 a:
Sect. 39).
2.16 Stokes’ integral and Stokes’ function 105
We next see from the derivation of Stokes’ formula by means of the

upward continuation integral (2–282) that it automatically suppresses the
harmonic terms of degrees one and zero in T and N . The implications of this
will be discussed later. We will see that Stokes’ formula in its original form
(2–304) and (2–307) only applies for a reference ellipsoid that (1) has the
same potential U0 = W0 as the geoid, (2) encloses a mass that is numerically
equal to the earth’s mass, and (3) has its center at the center of gravity of
the earth. Since the first two conditions are not accurately satisfied by the
reference ellipsoids that are in current practical use, and can hardly ever be
rigorously fulfilled, Stokes’ formula will later be modified for the case of an
arbitrary reference ellipsoid.
Finally, T is assumed to be harmonic outside the geoid. This means that
the effect of the masses above the geoid must be removed by suitable gravity
reductions. This will be discussed in Chaps. 3 and 8.
A bonus application to satellite geodesy

As a somewhat unexpected application, not related to Stokes’ formula, we
note that Eq. (2–280) can be used to compute gravity anomalies ∆g from
a satellite-determined spherical-harmonic series of the external gravitational
potential V !
2.16 Explicit form of Stokes’ integral and Stokes’

function in spherical harmonics
We now write Stokes’ formula (2–307) more explicitly by introducing suitable
coordinate systems on the sphere.
The use of spherical polar coordinates with origin at P offers the ad-
vantage that the angle ψ, which is the argument of Stokes’ function, is one
coordinate, the spherical distance. The other coordinate is the azimuth α,
reckoned from north. Their definitions are seen in Fig. 2.16. Denoting by P
both a fixed point on the sphere r = R (or in space) and its projection on
the unit sphere is common practice and will not cause any trouble.
If P coincides with the north pole, then ψ and α are identical with ϑ and
λ. According to Sect. 1.9, the surface element dσ is then given by
dσ = sin ψ dψ dα . (2–308)
Since all points of the sphere are equivalent, this relation applies for an
arbitrary origin P . In the same way, we have
2π π
= . (2–309)
α=0 ψ=0
σ
PN
north pole
®
P
Ã
d¾
dÃ
sinÃd®
Fig. 2.16. Polar coordinates on the unit sphere
Hence, we find
2π π
R
N= ∆g(ψ, α) S(ψ) sin ψ dψ dα (2–310)
4π γ0 α=0 ψ=0
as an explicit form of (2–307). Performing the integration with respect to α

first, we obtain
π 2π
R 1
N= ∆g(ψ, α) dα S(ψ) sin ψ dψ . (2–311)
2γ0 ψ=0 2π α=0
The expression in brackets is the average of ∆g along a parallel of spherical

radius ψ. We denote this average by ∆g(ψ), so that
2π
1
∆g(ψ) = ∆g(ψ, α) dα . (2–312)
2π α=0
Thus, Stokes’ formula may be written
π
R
N= ∆g(ψ) F (ψ) dψ , (2–313)
γ0 ψ=0
where we have introduced

1
2 S(ψ) sin ψ = F (ψ) . (2–314)
The functions S(ψ) and F (ψ) are shown in Fig. 2.17. Alternatively, we may
2.16 Stokes’ integral and Stokes’ function 107
0
30° 60° 90° 120° 150° 180°
F(Ã)
–1
–2
S(Ã)
Fig. 2.17. Stokes’ functions S(ψ) and F (ψ)
use ellipsoidal coordinates ϕ, λ. As a spherical approximation, ϑ is the com-

plement of ellipsoidal latitude:
ϑ = 90◦ − ϕ . (2–315)
Hence, we have
2π π/2
dσ = cos ϕ dϕ dλ , (2–316)
λ=0 ϕ=−π/2
σ
so that Stokes’ formula now becomes

2π π/2
R
N (ϕ, λ) = ∆g(ϕ , λ ) S(ψ) cos ϕ dϕ dλ , (2–317)
4π γ0 λ =0 ϕ =−π/2
where ϕ, λ are the ellipsoidal coordinates of the computation point and ϕ , λ

are the coordinates of the variable surface element dσ. The spherical distance
ψ is expressed as a function of these coordinates by
cos ψ = sin ϕ sin ϕ + cos ϕ cos ϕ cos(λ − λ) . (2–318)
Stokes’ function in terms of spherical harmonics

In Sect. 2.13, Eq. (2–273), we have found
∞
1
∆g(ϑ, λ) = (n − 1) Tn (ϑ, λ) . (2–319)
R
n=0
We may also directly express ∆g(ϑ, λ) as a series of Laplace surface spherical

harmonics:
∞

∆g(ϑ, λ) = ∆gn (ϑ, λ) . (2–320)
n=0
Comparing these two series yields
n−1 R
∆gn (ϑ, λ) = Tn (ϑ, λ) or Tn = ∆gn , (2–321)
R n−1
so that
∞
∞
∆gn
T = Tn = R . (2–322)
n−1
n=0 n=0
This equation shows again that there must not be a first-degree term in the
spherical-harmonic expansion of ∆g; otherwise the term ∆gn /(n − 1) would
be infinite for n = 1. As usual, we now assume that the harmonics of degrees
zero and one are missing. Therefore, we start the summation with n = 2.
By Eq. (1–89), we may write

2n + 1
∆gn = ∆g Pn (cos ψ) dσ , (2–323)
4π
σ
so that the preceding formula becomes

∞
R 2n + 1
T = ∆g Pn (cos ψ) dσ . (2–324)
4π n−1
n=2 σ
By interchanging the order of summation and integration, we get

∞

R 2n + 1
T = Pn (cos ψ) ∆g dσ . (2–325)
4π n=2
n−1
σ
Comparing this with Stokes’ formula (2–304), we find the expression for
Stokes’ function in terms of Legendre polynomials (zonal harmonics):
∞
2n + 1
S(ψ) = Pn (cos ψ) . (2–326)
n=2
n−1
In fact, the analytic expression (2–305) of Stokes’ function could have

been derived somewhat more simply by direct summation of this series,
but we believe that the derivation given in the preceding section is more
instructive because it also throws sidelights on important related problems
such as the “bonus equation” (2–280).
2.17 Generalization to an arbitrary reference ellipsoid 109
2.17 Generalization to an arbitrary reference

ellipsoid
As we have seen, Stokes’ formula, in its original form, suppresses the spherical
harmonics of degrees zero and one in the anomalous potential T and is,
therefore, strictly valid only if these terms are missing. This fact and the
condition U0 = W0 impose restrictions on the reference ellipsoid and on its
normal gravity field that are difficult to fulfil in practice.
Therefore, we generalize Stokes’ formula so that it will apply to an arbi-
trary ellipsoid of reference, which must satisfy only the condition that it is
so close to the geoid that the deviations of the geoid from the ellipsoid can
be treated as linear.
Consider the anomalous potential T at the surface of the earth. Its ex-
pression in surface spherical harmonics is given by
∞

T (ϑ, λ) = Tn (ϑ, λ) . (2–327)
n=0
By separating the terms of degrees zero and one, we may write
T (ϑ, λ) = T0 + T1 (ϑ, λ) + T (ϑ, λ) , (2–328)
where
∞

T (ϑ, λ) = Tn (ϑ, λ) . (2–329)
n=2
In the general case this function T ,

rather than T itself, is the quantity given
by Stokes’ formula. It is equal to T only if T0 and T1 are missing. Otherwise,
we have to add T0 and T1 in order to get the complete function T .
The zero-degree term in the spherical-harmonic expansion of the poten-
tial is equal to
GM
, (2–330)
r
where M is the mass. Hence, the zero-degree term of the anomalous potential
T = W − U at the surface of the earth, where r = R, is given by
G δM
T0 = , (2–331)
R
where
δM = M − M (2–332)
is the difference between the mass M of the earth and the mass M of the
ellipsoid. It would be zero if both masses were equal – but since we do not
know the exact mass of the earth, how can we make M rigorously equal to
M?
Subsequently, we will see that the first-degree harmonic can always be
assumed to be zero. Under this assumption, we can substitute (2–331) into
(2–328) and express T by the conventional Stokes formula (2–304). Thus we
obtain
G δM R
T = + ∆g S(ψ) dσ . (2–333)
R 4π
σ
This is the generalization of Stokes’ formula for T . It holds for an arbitrary
reference ellipsoid whose center coincides with the center of the earth.
First-degree terms
The coefficients of the first-degree harmonic in the potential W are, according
to (2–85) and (2–87), given by
GM xc , GM yc , GM zc , (2–334)
where xc , yc , zc are the rectangular coordinates of the earth’s center of grav-

ity. For the normal potential U , we have the analogous quantities
GM xc , GM yc , GM zc . (2–335)
As xc , yc , zc are very small in any case, these are practically equal to
GM xc , GM yc , GM zc . (2–336)
The coefficients of the first-degree harmonic in the anomalous potential T =

W − U are, therefore, equal to
GM (xc − xc ) , GM (yc − yc ) , GM (zc − zc ) . (2–337)
They are zero, and there is no first-degree harmonic T1 (ϑ, λ) if and only if
the center of the reference ellipsoid coincides with the center of gravity of
the earth. This is usually assumed.
In the general case, we find from the first-degree term of (2–76), on
putting r = R and using the coefficients (2–85) together with (2–87),
GM
T1 (ϑ, λ) = 2 (zc − zc ) P10 (cos ϑ) + (xc − xc ) P11 (cos ϑ) cos λ
R
(2–338)
+ (yc − yc ) P11 (cos ϑ) sin λ .
If the origin of the coordinate system is taken to be the center of the reference
ellipsoid, then xc = yc = zc = 0. With P10 (cos ϑ) = cos ϑ, P11 (cos ϑ) = sin ϑ,
and GM/R2 = γ0 we then obtain the following expression for the first-degree
harmonic of T :
T1 (ϑ, λ) = γ0 (xc sin ϑ cos λ + yc sin ϑ sin λ + zc cos ϑ) . (2–339)
Dividing by γ0 , we find the first-degree harmonic of the geoidal height:
N1 (ϑ, λ) = xc sin ϑ cos λ + yc sin ϑ sin λ + zc cos ϑ . (2–340)
Introducing the vector

xc = [xc , yc , zc ] (2–341)
and the unit vector of the direction (ϑ, λ),
e = [sin ϑ cos λ, sin ϑ sin λ, cos ϑ] , (2–342)
(2–340) may be written as
N1 (ϑ, λ) = xc · e , (2–343)
which is interpreted as the projection of the vector xc onto the direction

(ϑ, λ).
Hence, if the two centers of gravity do not coincide, then we need only
add the first-degree terms (2–339) and (2–340) to the generalized Stokes
formula (2–333) and to its analogue for N , respectively, in order to get the
most general solution for Stokes’ problem, the computation of T and N from
∆g. Equation (2–273) shows that any value of T1 (ϑ, λ) is compatible with a
given ∆g field because, for n = 1, the quantity (n − 1) T1 is zero and so T1 ,
whatever be its value, does not at all enter into ∆g.
Hence, the most general solution for T and N contains three arbitrary
constants xc , yc , zc , which can, thus, be regarded as the constants of integra-
tion for Stokes’ problem. In actual practice, one always sets xc = yc = zc = 0,
thus placing the center of the reference ellipsoid at the center of the earth.
This constitutes an essential advantage of the gravimetric determination of
the geoid over the astrogeodetic method, where the position of the reference
ellipsoid with respect to the center of the earth remains unknown.
Zero-degree terms in N and ∆g

Let us first extend Bruns’ formula (2–237) to an arbitrary reference ellipsoid.
Suppose
W (x, y, z) = W0 ,
(2–344)
U (x, y, z) = U0
are the equations of the geoid and the ellipsoid, where in general the con-
stants W0 and U0 are different. As in Sect. 2.12, we have, using Fig. 2.12,
WP = UQ − γ N + T , but now UQ = U0 = W0 = WP , so that
γ N = T − (W0 − U0 ) . (2–345)
Denoting the difference between the potentials by
δW = W0 − U0 , (2–346)
we obtain the following simple generalization of Bruns’ formula:
T − δW
N= . (2–347)
γ
We also need the extension of Eqs. (2–246) through (2–250). Those for-
mulas which contain N instead of T are easily seen to hold for an arbitrary
reference ellipsoid as well, but the transition from N to T is now effected by
means of (2–347). Hence, Eq. (2–247), i.e.,
∂T ∂γ
∆g = − + N, (2–348)
∂h ∂h
remains unchanged, but (2–248) becomes
∂T 1 ∂γ 1 ∂γ
∆g = − + T− δW . (2–349)
∂h γ ∂h γ ∂h
Therefore, the fundamental boundary condition is now
∂T 1 ∂γ 1 ∂γ
− + T = ∆g + δW . (2–350)
∂h γ ∂h γ ∂h
The spherical approximations of these equations are
T − δW
N= (2–351)
γ0
and
∂T 2 2
∆g = − − T + δW (2–352)
∂r R R
and
∂T 2 2
− − T = ∆g − δW . (2–353)
∂r R R
Relations between T , N , and ∆g

By (2–347), we have
T = γ0 N + δW . (2–354)
Substituting this into (2–333) and dividing by γ0 , we obtain

G δM δW R
N= − + ∆g S(ψ) dσ . (2–355)
R γ0 γ0 4π γ0
σ
This is the generalization of Stokes’ formula for N . It applies for an arbitrary

reference ellipsoid whose center coincides with the center of the earth.
While formula (2–333) for T contains only the effect of a mass difference
δM , the formula (2–355) for N contains, in addition, the potential differ-
ence δW . These formulas also show clearly that the simple Stokes integrals
(2–304) and (2–307) hold only if δM = δW = 0, that is, if the reference el-
lipsoid has the same potential as the geoid and the same mass as the earth.
Otherwise, they give N and T only up to additive constants: putting
G δM δW
N0 = − (2–356)
R γ0 γ0
and taking into account (2–331), we have

R
T = T0 + ∆g S(ψ) dσ , (2–357)
4π
σ

R
N = N0 + ∆g S(ψ) dσ . (2–358)
4π γ0
σ
Alternative forms of (2–355), which are sometimes useful, are obtained in

the following way. Substituting the series (2–268) and (2–270) into (2–352),
we get
∞
1 2
∆g(ϑ, λ) = (n − 1) Tn (ϑ, λ) + δW (2–359)
R R
n=0
as the generalization of (2–273). Expanding the function ∆g(ϑ, λ) into the

usual series of Laplace surface spherical harmonics,
∞

∆g(ϑ, λ) = ∆gn (ϑ, λ) , (2–360)
n=0
and comparing the constant terms (n = 0) of these two equations, we get

1 2
− T0 + δW = ∆g0 , (2–361)
R R
where, by (1–89),
1
∆g0 = ∆g dσ . (2–362)
4π
σ
Expressing T0 by (2–331) in terms of δM , we obtain

1 2
∆g0 = − 2
G δM + δW . (2–363)
R R
The two equations (2–356) for N0 and (2–363) for ∆g0 can now be solved
for δM and δW :
G δM = R (R ∆g0 + 2γ0 N0 ) ,
(2–364)
δW = R ∆g0 + γ0 N0 .
The constant N0 may be expressed by either of these equations:
R G δM
N0 = − ∆g0 + ,
2γ0 2γ0 R
(2–365)
R δW
N0 = − ∆g0 + .
γ0 γ0
A final note
A direct consequence of Eq. (2–356) is that N0 has an immediate geometrical
meaning: if a is the equatorial radius (semimajor axis) of the given reference
ellipsoid, then
aE = a + N0 (2–366)
is the equatorial radius of an ellipsoid whose normal potential U0 is equal to
the actual potential W0 of the geoid, and which encloses the same mass as
that of the earth, the flattening f remaining the same. The reason is that
for such a new ellipsoid E the new N0 = 0 by (2–356) with δM = 0 and
δW = 0.
A small additive constant N0 is equivalent to a change of scale for a nearly
spherical earth. To see this, imagine a nearly spherical orange. Increasing the
thickness of the peel of an orange everywhere by 1 mm (say) is equivalent
to a similarity transformation (uniform increase of the size) of the orange’s
surface.
So, the usual Stokes formula, without N0 , gives a global geoid that is
determined only up to the scale which implicitly is contained in N0 . It is,
however, geocentric, at least in theory, because it contains no spherical har-
monic of first degree, T1 (ϑ, λ). It would be exactly geocentric if the earth
were covered uniformly by gravity measurements. The scale was formerly de-
termined astrogeodetically, historically by grade measurements dating back
2.18 Gravity disturbances and Koch’s formula 115
to the 18th century (Clairaut, Maupertuis; see Todhunter [1873]). Today,

the scale is furnished by satellites (laser, GPS).
2.18 Gravity disturbances and Koch’s formula

It is easy to find Koch’s formula, which is the alternative of Stokes’ formula
for gravity disturbances δg. We just indicate the road in its general outlines,
leaving the reader to generate a four-lane highway.
Compare equations (2–269) and (2–270) with (2–272) and (2–273). We
see that the main difference between gravity disturbances δg and gravity
anomalies ∆g is the spherical harmonic factor n + 1 and n − 1, respectively.
The other – very small – difference is that we omit in ∆g the terms n = 0
and 1 (see comment after (2–273)), which is not necessary in δg.
Using almost literally the development of Sect. 2.14, we get an equa-
tion for δg which is the exact equivalent of (2–282) for ∆g. Following the
integration in Sect. 2.15, we get a formula of form (2–302)

R
T (r, ϑ, λ) = K(r, ψ) δg dσ , (2–367)
4π
σ
and on the sphere r = R we get a formula of form (2–304) which we call

Koch’s formula:
R
T = K(ψ) δg dσ , (2–368)
4π
σ
where K(ψ) is the Hotine–Koch function

1 1
K(ψ) = − ln 1 + , (2–369)
sin(ψ/2) sin(ψ/2)
which is very similar to the Stokes function (2–305). By Bruns’ theorem, we
finally get
R
N= K(ψ) δg dσ . (2–370)
4πγ0
σ
In absolute analogy with (2–326), we have, simply by replacing n − 1 by

n + 1 and leaving n = 0 as the lower limit of the sum,
∞
2n + 1
K(ψ) = Pn (cos ψ) (2–371)
n=0
n+1
as the expression of the Koch–Hotine function in terms of Legendre polyno-

mials (zonal harmonics). It is really that simple!
A historical remark
This remark is due to Mrs. M. I. Yurkina, Moscow. Mathematically, the
above is the solution of Neumann’s problem (the second boundary-value
problem of potential theory) for the sphere, cf. Sect. 1.13. It is a classical
problem of potential theory, with a history of at least 150 years, similarly
to Stokes’ formula. “Neumann’s problem” is named after the mathematician
Carl Neumann, who edited his father’s (Franz Neumann) lectures from the
1850s (Neumann 1887: see especially p. 275). The external spherical Neu-
mann problem also occurs in Kellogg (1929: p. 247). It is again found in
Hotine (1969: pp. 311, 318).
Their basic significance for modern physical geodesy with a known earth
surface was recognized and elaborated by Koch (1971). So the present inte-
gral formula should perhaps be called F. Neumann – C. Neumann – Kellogg –
Hotine – Koch formula. For brevity, we refer to it as Koch’s formula.
2.19 Deflections of the vertical and formula of

Vening Meinesz
Stokes’ formula permits the calculation of the geoidal undulations from grav-
ity anomalies. A similar formula for the computation of the deflections of the
vertical from gravity anomalies has been given by Vening Meinesz (1928).
Figure 2.18 shows the intersection of geoid and reference ellipsoid with
a vertical plane of arbitrary azimuth. If ε is the component of the deflection
of the vertical in this plane, then
dN = −ε ds (2–372)
or
dN
ε=− ; (2–373)
ds
the minus sign is a convention, its meaning will be explained later.
plumb line ellipsoidal normal

"
" geoid
dN
ds
ds
s ellipsoid
Fig. 2.18. The relation between the geoidal undulation and the
deflection of the vertical
2.19 Deflections of the vertical and formula of Vening Meinesz 117
In a north-south direction, we have

ε = ξ and ds = dsϕ = R dϕ ; (2–374)
in an east-west direction,
ε = η and ds = dsλ = R cos ϕ dλ . (2–375)
In the formulas for dsϕ and dsλ , we have again used the spherical approxi-
mation; according to (1–30), the element of arc on the sphere r = R is given
by
ds2 = R2 dϕ2 + R2 cos2 ϕ dλ2 . (2–376)
By specializing (2–373), we find
dN 1 ∂N
ξ=− =− ,
dsϕ R ∂ϕ
(2–377)
dN 1 ∂N
η=− =− ,
dsλ R cos ϕ ∂λ
which gives the connection between the geoidal undulation N and the com-
ponents ξ and η of the deflection of the vertical.
As N is given by Stokes’ integral, our problem is to differentiate this
formula with respect to ϕ and λ. For this purpose, we use the form (2–317),
2π π/2
R
N (ϕ, λ) = ∆g(ϕ , λ ) S(ψ) cos ϕ dϕ dλ , (2–378)
4π γ0 λ =0 ϕ =−π/2
where ψ is defined in (2–318) as a function of ϕ, λ and ϕ , λ .

The integral on the right-hand side of this formula depends on ϕ and
λ only through ψ in S(ψ). Therefore, by differentiating under the integral
sign,
2π π/2
∂N R ∂S(ψ)
= ∆g(ϕ , λ ) cos ϕ dϕ dλ (2–379)
∂ϕ 4π γ0 λ =0 ϕ =−π/2 ∂ϕ
is obtained and a similar formula for ∂N/∂λ. Here we have
∂S(ψ) dS(ψ) ∂ψ ∂S(ψ) dS(ψ) ∂ψ
= , = . (2–380)
∂ϕ dψ ∂ϕ ∂λ dψ ∂λ
Differentiating (2–318) with respect to ϕ and λ, we obtain
∂ψ
− sin ψ = cos ϕ sin ϕ − sin ϕ cos ϕ cos(λ − λ) ,
∂ϕ
(2–381)
∂ψ
− sin ψ = cos ϕ cos ϕ sin(λ − λ) .
∂λ
north pole
¸'– ¸
90°– ' 90°– ''
®
P
Ã
d¾
Fig. 2.19. Relation between geographical and polar coordinates on the sphere
We now introduce the azimuth α, as shown in Fig. 2.16. From the spherical
triangle of Fig. 2.19 we get, using well-known formulas of spherical trigonom-
etry,
sin ψ cos α = cos ϕ sin ϕ − sin ϕ cos ϕ cos(λ − λ) ,
(2–382)
sin ψ sin α = cos ϕ sin(λ − λ) .
Substituting these relations into the preceding equations, we find the simple
expressions
∂ψ ∂ψ
= − cos α , = − cos ϕ sin α , (2–383)
∂ϕ ∂λ
so that
∂S(ψ) dS(ψ) ∂S(ψ) dS(ψ)
=− cos α , =− cos ϕ sin α . (2–384)
∂ϕ dψ ∂λ dψ
These are substituted into (2–379) and the corresponding formula for ∂N/∂λ
and from equations (2–377) we finally obtain
2π π/2
1 dS(ψ)
ξ(ϕ, λ) = ∆g(ϕ , λ ) cos α cos ϕ dϕ dλ ,
4π γ0 λ =0 ϕ =−π/2 dψ
2π π/2
1 dS(ψ)
η(ϕ, λ) = ∆g(ϕ , λ ) sin α cos ϕ dϕ dλ
4π γ0 λ =0 ϕ =−π/2 dψ
(2–385)
or, written in the usual abbreviated form,

1 dS(ψ)
ξ= ∆g cos α dσ ,
4π γ0 dψ
σ
(2–386)
1 dS(ψ)
η= ∆g sin α dσ .
4π γ0 dψ
σ
2.20 The vertical gradient of gravity 119
These are the formulas of Vening Meinesz. Differentiating Stokes’ function

S(ψ), Eq. (2–305), with respect to ψ, we obtain Vening Meinesz’ function
dS(ψ) cos(ψ/2) 1 − sin(ψ/2)

=− + 8 sin ψ − 6 cos(ψ/2) − 3
dψ 2 sin2(ψ/2) sin ψ (2–387)

+ 3 sin ψ ln sin(ψ/2) + sin2(ψ/2) .
This can be readily verified by using the elementary trigonometric identities.

The azimuth α is given by the formula
cos ϕ sin(λ − λ)
tan α = , (2–388)
cos ϕ sin ϕ − sin ϕ cos ϕ cos(λ − λ)
which is an immediate consequence of (2–382).

The form (2–385) is an expression of (2–386) in terms of ellipsoidal co-
ordinates ϕ and λ. As with Stokes’ formula (Sect. 2.15), we may also use an
expression in terms of spherical polar coordinates ψ and α:
2π π
1 dS(ψ)
ξ= ∆g(ψ, α) cos α sin ψ dψ dα ,
4π γ0 α=0 ψ=0 dψ
(2–389)
2π π
1 dS(ψ)
η= ∆g(ψ, α) sin α sin ψ dψ dα .
4π γ0 α=0 ψ=0 dψ
The reader can easily verify that these equations give the deflection compo-
nents ξ and η with the correct sign corresponding to the definition (2–230);
see also Fig. 2.13. This is the reason why we introduced the minus sign in
(2–373).
We note that the formula of Vening Meinesz is valid as it stands for an
arbitrary reference ellipsoid, whereas Stokes’ formula had to be modified by
adding a constant N0 . If we differentiate the modified Stokes formula with
respect to ϕ and λ, to get Vening Meinesz’ formula, then this constant N0
drops out and we get Eqs. (2–386).
For the practical application of Stokes’ and Vening Meinesz’ formulas
and problems, the reader is referred to Sect. 2.21 and to Chap. 3.
2.20 The vertical gradient of gravity

Bruns’ formula (2–40), with = 0,
∂g
= −2g J − 2ω 2 , (2–390)
∂H
cannot be directly applied to determine the gradient ∂g/∂H because the

mean curvature J of the level surfaces is unknown. Therefore, we proceed in
the usual way by splitting ∂g/∂H into a normal and an anomalous part:
∂g ∂γ ∂∆g
= + . (2–391)
∂H ∂H ∂H
The normal gradient ∂γ/∂H is given by (2–147) and (2–148). The anomalous
.
part, ∂∆g/∂H = ∂∆g/∂r, will be considered now.
Expression in terms of ∆g
Equation (2–272) may be written as (note that r ∆g is harmonic and the
factor must be 1 for r = R)
∞ n+2
R
∆g(r, ϑ, λ) = ∆gn (ϑ, λ) . (2–392)
r
n=0
By differentiating with respect to r and setting r = R, we obtain at sea level:

∞ ∞
∂∆g 1 1 2
=− (n + 2) ∆gn = − n ∆gn − ∆g . (2–393)
∂r R R R
n=0 n=0
Now we can apply (1–149), setting V = ∆g and Yn = ∆gn . The result is

∂∆g R2 ∆g − ∆gP 2
= dσ − ∆gP . (2–394)
∂r 2π l03 R
σ
In this equation, ∆gP is referred to the fixed point P at which ∂∆g/∂r is

to be computed; l0 is the spatial distance between the fixed point P and the
variable surface element R2 dσ, expressed in terms of the angular distance ψ
by
ψ
l0 = 2R sin . (2–395)
2
Compare Fig. 1.9 of Sect. 1.14; the element R2 dσ is at the point P .
The important integral formula (2–394) expresses the vertical gradient
of the gravity anomaly in terms of the gravity anomaly itself. Since the
integrand decreases very rapidly with increasing distance l0 , it is sufficient in
this formula to extend the integration only over the immediate neighborhood
of the point P , as opposed to Stokes’ and Vening Meinesz’ formulas, where
the integration must include the whole earth, if a sufficient accuracy is to be
obtained.
2.20 The vertical gradient of gravity 121
Expression in terms of N
By differentiating equation (2–271),
∂T 2
∆g = − − T, (2–396)
∂r r
with respect to r, we get
∂∆g ∂2T 2 ∂T 2
=− 2 − + 2 T. (2–397)
∂r ∂r r ∂r r
To this formula we add Laplace’s equation ∆T = 0, which in spherical
coordinates has the form
∂2T 2 ∂T tan ϕ ∂T 1 ∂2T 1 ∂2T
+ − + + = 0; (2–398)
∂r 2 r ∂r r 2 ∂ϕ r 2 ∂ϕ2 r 2 cos2 ϕ ∂λ2
see Eq. (1–35), modify by replacing V by T and substitute ϑ = 90 − ϕ. The

result, on setting r = R, is
∂∆g 2 tan ϕ ∂T 1 ∂2T 1 ∂2T

= 2T− 2
+ 2 2
+ 2 . (2–399)
∂r R R ∂ϕ R ∂ϕ R cos ϕ ∂λ2
2
Since T = γ0 N , we may also write
∂∆g 2γ0 γ0 ∂N γ0 ∂ 2 N γ0 ∂2N

= 2 N − 2 tan ϕ + 2 + , (2–400)
∂r R R ∂ϕ R ∂ϕ2 R2 cos2 ϕ ∂λ2
where γ0 is a global mean value as usual. This equation expresses the vertical
gradient of the gravity anomaly in terms of the geoidal undulation N and
its first and second horizontal derivatives. It can be evaluated by numerical
differentiation, using a map of the function N . However, it is less suited for
practical application than (2–394) because it requires an extremely accurate
and detailed local geoidal map, which is hardly ever available; inaccuracies
of N are greatly amplified by forming the second derivatives.
Expression in terms of ξ and η

From equations (2–377), we find
∂N ∂N
= −R ξ , = −R η cos ϕ , (2–401)
∂ϕ ∂λ
so that
∂2N ∂ξ ∂2N ∂η
2
= −R , 2
= −R cos ϕ . (2–402)
∂ϕ ∂ϕ ∂λ ∂λ
Substituting these relations into (2–400) yields

∂∆g 2γ0 γ0 γ0 ∂ξ γ0 ∂η
= 2 N+ ξ tan ϕ − − . (2–403)
∂r R R R ∂ϕ R cos ϕ ∂λ
Introducing local rectangular coordinates x, y in the tangent plane, we have
R dϕ = dsϕ = dx ,
(2–404)
R cos ϕ dλ = dsλ = dy ,
so that (2–403) becomes

∂∆g 2γ0 γ0 ∂ξ ∂η
= 2 N+ ξ tan ϕ − γ0 + . (2–405)
∂r R R ∂x ∂y
The first two terms on the right-hand side can be shown to be very small in
comparison to the third term; hence, to a sufficient accuracy

∂∆g ∂ξ ∂η
= −γ0 + (2–406)
∂r ∂x ∂y
may be used. These beautiful formulas express the vertical gradient of the
gravity anomaly in terms of the horizontal derivatives of the deflection of the
vertical. They can again be evaluated by means of numerical differentiation
if a map of ξ and η is available. They are somewhat better suited for practical
application than (2–400) because only first derivatives are required.
2.21 Practical evaluation of the integral formulas

Integral formulas such as Stokes’ and Vening Meinesz’ integrals must be
evaluated approximately by summations. The surface elements dσ are re-
placed by small but finite compartments q, which are obtained by suitably
subdividing the surface of the earth. Two different methods of subdivision
are used:
1. Templates (Fig. 2.20). The subdivision is achieved by concentric circles
and their radii. The template is placed on a gravity map of the same
scale so that the center of the template coincides with the computation
point P on the map. The natural coordinates for this purpose are polar
coordinates ψ, α with origin at P .
2. Grid lines (Fig. 2.21). The subdivision is achieved by the grid lines of
some fixed coordinate system, in particular of ellipsoidal coordinates
ϕ, λ. They form rectangular blocks – for example, of 10 ×10 or 1◦ ×1◦ .
These blocks are also called squares, although they are usually not
squares as defined in plane geometry.
2.21 Practical evaluation of the integral formulas 123
®1
=
®
P q
® = ®2
Ã = Ã2
Ã = Ã1
Fig. 2.20. A template
¸= 36°20' 30' 40' 36°50'
45°30'
q
20'
'= 45°10'
Fig. 2.21. Blocks formed by a grid of ellipsoidal coordinates
The template method is wonderfully easy to understand and to use for

theoretical considerations, but completely old-fashioned. Only the gridline
method has survived in the computer world.
As a simple and instructive example illustrating the principles of numer-
ical integration consider Stokes’ formula

R
N= ∆g S(ψ) dσ (2–407)
4π γ0
σ
with its explicit forms (2–310) for the template method and (2–317) for the
method that uses fixed blocks.
For each compartment qk , the gravity anomalies are replaced by their

average value ∆g k in this compartment. Hence, the above equation becomes

R R
N= ∆g k S(ψ) dσ = ∆gk S(ψ) dσ (2–408)
4π γ0 4π γ0
k qk k qk
or
N= ck ∆g k , (2–409)
k
where the coefficients

R
ck = S(ψ) dσ (2–410)
4π γ0
qk
are obtained by integration over the compartment qk ; they do not depend

on ∆g.
If the integrand – in our case, Stokes’ function S(ψ) – is reasonably
constant over the compartment qk , it may be replaced by its value S(ψk ) at
the center of qk . Then we have

R S(ψk )
ck = S(ψ) dσ = R2 dσ . (2–411)
4π γ0 4π γ0 R
qk qk
The final integral is simply the area Ak of the compartment and we obtain
Ak S(ψk )
ck = . (2–412)
4π γ0 R
The advantage of the template method is its great flexibility. The influ-
ence of the compartments near the computation point P is greater than that
of the distant ones, and the integrand changes faster in the neighborhood
of P . Therefore, a finer subdivision is necessary around P . This can easily
be provided by templates. Yet, the method is completely old-fashioned and
thus obsolete.
The advantage of the fixed system of blocks formed by a grid of ellipsoidal
coordinates lies in the fact that their mean gravity anomalies are needed for
many different purposes. These mean anomalies of standard-sized blocks,
once they have been determined, can be easily stored and processed by a
computer. Also, the same subdivision is used for all computation points,
whereas the compartments defined by a template change when the template
is moved to the next computation point. The flexibility of the method of
standard blocks is limited; however, one may use smaller blocks (5 × 5 , for
example) in the neighborhood of P and larger ones (1◦ × 1◦ , for example)
farther away. With current electronic computation, this method is the only
one used in practice. The theoretical usefulness of polar coordinates will be
shown now.
Effect of the neighborhood

This issue is interesting and instructive. In the innermost zone, even the
template method may pose difficulties if the integrand becomes infinite as
ψ → 0. This happens with Stokes’ formula, since
. 2
S(ψ) = (2–413)
ψ
for small ψ. This can be seen from the definition (2–305), because the first
term is predominant and is, for small ψ, given by
1 . 1 2
= = . (2–414)
sin(ψ/2) (ψ/2) ψ
Vening Meinesz’ function becomes infinite as well, since to the same approx-
imation,
dS(ψ) . 2
=− 2. (2–415)
dψ ψ
In the gradient formula (2–394), the integrand
1 . 1
= 3 3 (2–416)
l03 R ψ
behaves in a similar way.
Therefore, it may be convenient to split off the effect of this innermost
zone, which will be assumed to be a circle of radius ψ0 around the compu-
tation point. For instance, Stokes’ integral becomes in this way
N = Ni + Ne , (2–417)
where
2π ψ0
R
Ni = ∆g S(ψ) dσ ,
4π γ0 α=0 ψ=0
(2–418)
2π π
R
Ne = ∆g S(ψ) dσ .
4π γ0 α=0 ψ=ψ0
The radius ψ0 of the inner zone corresponds to a linear distance of a few
kilometers. Within this distance, we may treat the sphere as a plane, using
polar coordinates s, α, where
. . . ψ
s = R ψ = R sin ψ = 2R sin , (2–419)
2
so that the element of area becomes
R2 dσ = s ds dα . (2–420)
It is consistent with this approximation to use (2–413) through (2–416),

putting
. 2R dS . 2R2 1 . 1
S(ψ) = , =− 2 , = 3. (2–421)
s dψ s l03 s
In Stokes’ and Vening Meinesz’ functions as well, the relative error of these
approximations is about 1% for s = 10 km, and about 3% for s = 30 km.
In 1/l03 it is even less. Hence, the effect of this inner zone on our integral
formulas becomes
2π s0
1 ∆g
Ni = s ds dα , (2–422)
2π γ0 α=0 s=0 s
2π s0
1 ∆g
ξi = − cos α s ds dα ,
2π γ0 α=0 s=0 s2
(2–423)
2π s0
1 ∆g
ηi = − sin α s ds dα ,
2π γ0 α=0 s=0 s2
s0
∂∆g 1 2π
∆g − ∆gP
= s ds dα . (2–424)
∂H i 2π α=0 s=0 s3
In order to evaluate these integrals, we expand ∆g into a Taylor series at
the computation point P :
1
2
∆g = ∆gP + x gx + y gy + x gxx + 2x y gxy + y 2 gyy + · · · . (2–425)
2!
The rectangular coordinates x, y are defined by
x = s cos α , y = s sin α , (2–426)
so that the x-axis points north. We further have

∂∆g ∂ 2 ∆g
gx = , gxx = , etc. (2–427)
∂x P ∂x2 P
This Taylor series may also be written as
∆g = ∆gP + s (gx cos α + gy sin α)

(2–428)
s2

+ gxx cos2 α + 2gxy cos α sin α + gyy sin2 α + · · · .
2
Inserting this into the above integrals, we can easily evaluate them. Perform-
ing the integration with respect to α first and noting that
2π
0 dα = 2π ,
2π 2π 2π
0 sin α dα = 0 cos α dα = 0 sin α cos α dα = 0 , (2–429)
2π 2π
0 sin2 α dα = 0 cos2 α dα = π ,
we find
s0
1 s2
Ni = ∆gP + (gxx + gyy ) + · · · ds , (2–430)
γ0 0 4
s0
1
ξi = − (gx + · · ·) ds ,
2γ0 0
s0 (2–431)
1
ηi = − (gy + · · ·) ds ,
2γ0 0

∂∆g 1 s0
= (gxx + gyy + · · ·) ds . (2–432)
∂H i 4 s=0
We now perform the integration over s, retaining only the lowest nonvanish-
ing terms. The result is
s0
Ni = ∆gP , (2–433)
γ0
s0 s0
ξi = − gx , ηi = − gy , (2–434)
2γ0 2γ0

∂∆g s0
= (gxx + gyy ) . (2–435)
∂H i 4
We see that the effect of the innermost circular zone on Stokes’ formula
depends, to a first approximation, on the value of ∆g at P ; the effect on
Vening Meinesz’ formula depends on the first horizontal derivatives of ∆g;
and the effect on the vertical gradient depends on the second horizontal
derivatives.
Note that the contribution of the innermost zone to the total deflection
of the vertical has the same direction as the line of steepest inclination of
the “gravity anomaly surface”, because the plane vector
ϑ = [ξi , ηi ] (2–436)
is proportional to the horizontal gradient of ∆g,
grad ∆g = [gx , gy ] . (2–437)

line of steepest descent

grad Dg
Dg = 40 mgal
Dg = 30 mgal
Dg = 20 mgal
Fig. 2.22. Lines of constant ∆g and lines of steepest descent
The direction of grad ∆g defines the line of steepest descent (Fig. 2.22).
The values of gx and gy can be obtained from a gravity map. They are the
inclinations of north-south and east-west profiles through P . Values for gxx
and gyy may be found by fitting a polynomial in x and y of second degree
to the gravity anomaly function in the neighborhood of P .
A remark on accuracy
Deflections of the vertical ξ, η, if combined with astronomical observations
of astronomical latitude Φ and astronomical longitude Λ, furnish positions
on the reference ellipsoid, expressed by ellipsoidal coordinates
ϕ=Φ−ξ,
(2–438)
λ = Λ − η sec ϕ ,
just as vertical position is obtained by
h=H +N. (2–439)
Unfortunately, to get the same precision for horizontal as for vertical po-
sition, is much more difficult, keeping in mind the relation 1 ∼ = 30 m on
the earth’s surface. So to get an accuracy of 1 m, which is not too difficult
with Stokes’ formula, means an accuracy better than 0.03 in both Φ and ξ
(analogously to Λ and η), which is almost impossible to achieve practically.
3 Gravity reduction
3.1 Introduction
Gravity g measured on the physical surface of the earth must be distin-
guished from normal gravity γ referring to the surface of the ellipsoid. To
refer g to sea level, a reduction is necessary. Since there are masses above
sea level, the reduction methods differ depending on the way how to deal
with these topographic masses. Gravity reduction is essentially the same for
gravity anomalies ∆g and gravity disturbances δg.
Gravity reduction serves as a tool for three main purposes:
• determination of the geoid,

• interpolation and extrapolation of gravity,
• investigation of the earth’s crust.
Only the first two purposes are of a direct geodetic nature. The third is of
interest to theoretical geophysicists and geologists, who study the general
structure of the crust, and to exploration geophysicists.
The use of Stokes’ formula for the determination of the geoid requires
that the gravity anomalies ∆g represent boundary values at the geoid. This
implies two conditions: first, gravity g must refer to the geoid; second, there
must be no masses outside the geoid (Sect. 2.12). Hence, figuratively speak-
ing, gravity reduction consists of the following steps:
1. the topographic masses outside the geoid are completely removed or

shifted below sea level;
2. then the gravity station is lowered from the earth’s surface (point P )
to the geoid (point P0 , see Fig. 3.1).
H earth's surface
geoid
P0
Fig. 3.1. Gravity reduction
130 3 Gravity reduction
The first step requires knowledge of the density of the topographic masses,
which is somewhat problematic.
By such a reduction procedure certain irregularities in gravity due to
differences in height of the stations are removed so that interpolation and
even extrapolation to unobserved areas become easier (Sect. 9.7).
3.2 Auxiliary formulas

Let us compute the potential U and the vertical attraction A of a homoge-
neous circular cylinder of radius a and height b at a point P situated on its
axis at a height c above its base (Fig. 3.2).
P outside cylinder
Assume first that P is above the cylinder, c > b. Then the potential is given
by the general formula (1–12),

U =G dv . (3–1)
l
Introducing polar coordinates s, α in the xy-plane by
x = s cos α , y = s sin α , (3–2)
we have
l= s2 + (c − z)2 (3–3)
and
dv = dx dy dz = s ds dα dz . (3–4)
z z
P P
l
a c
s a
b dm z b
xy
Fig. 3.2. Potential and attraction of a circular cylinder on an external point

3.2 Auxiliary formulas 131
Hence, we find, with the density = constant,

2π a b
s ds dz dα
U = G
α=0 s=0 z=0 s2 + (c − z)2
(3–5)
a b
s ds dz
= 2π G .
s=0 z=0 s2 + (c − z)2
The integration with respect to s yields

a
a
s ds
= s2 + (c − z)2
s=0 s2 + (c − z)2 0 (3–6)

= a2 + (c − z)2 −c+z,
so that we have
b
U = 2π G −c+z+ a2 + (c − z)2 dz . (3–7)
0
The indefinite integral is 2π G times

1
2 (c−z)2 − 12 (c−z) a2 + (c − z)2 − 12 a2 ln c−z + a2 + (c − z)2 , (3–8)
as may be verified by differentiation. Hence, U finally becomes

! √
Ue = π G (c − b)2 − c2 − (c − b) a2 + (c − b)2 + c a2 + c2
√ " (3–9)
− a2 ln c − b + a2 + (c − b)2 + a2 ln c + a2 + c2 ,
where the subscript e denotes that P is external to the cylinder.

The vertical attraction A is the negative derivative of U with respect to
the height c [see Eq. (2–22)]:
∂U
A=− . (3–10)
∂c
Differentiating (3–9), we obtain

Ae = 2π G b + a2 + (c − b)2 − a2 + c2 . (3–11)
P on cylinder
In this case we have c = b, and Eqs. (3–9) and (3–11) become
√
b + a2 + b2
U0 = π G −b + b a2 + b2 + a ln
2 2
, (3–12)
a

A0 = 2π G a + b − a2 + b2 . (3–13)
P inside cylinder
We assume that P is now inside the cylinder, c < b. By the plane z = c we
separate the cylinder into two parts, 1 and 2 (Fig. 3.3), and compute U as
the sum of the contributions of these two parts:
Ui = U1 + U2 , (3–14)
where the subscript i denotes that P is now inside the cylinder. The term
U1 is given by (3–12) with b replaced by c, and U2 by the same formula with
b replaced by b − c. Their sum is

√
Ui = π G − c2 − (b − c)2 + c a2 + c2 + (b − c) a2 + (b − c)2
√ (3–15)
c+ a2 + c2 b−c+ a2 + (b − c)2
+ a2 ln + a2 ln .
a a
It is easily seen that the attraction is the difference A1 − A2 :

Ai = 2π G 2c − b − a2 + c2 + a2 + (b − c)2 ; (3–16)
this formula may also be obtained by differentiating (3–15) according to

(3–10).
2 P b -c
b
1 c
Fig. 3.3. Potential and attraction on an internal point

3.2 Auxiliary formulas 133
Circular disk
Let the thickness b of the cylinder go to zero such that the product
κ = b (3–17)
remains finite. The quantity κ may then be considered as the surface density
with which matter is concentrated on the surface of a circle of radius a. We
need potential and attraction for an exterior point. By setting
κ
= (3–18)
b
in (3–9) and (3–11) and then letting b → 0, we get by well-known methods

of the calculus √
Ue0 = 2π G κ a2 + c2 − c ,
(3–19)
c
A0e = 2π G κ 1 − √ .
a + c2
2
Sectors and compartments

For a sector of radius a and angle
2π
α= , (3–20)
n
we must divide the above formulas by n. For a compartment subtending the

same angle and bounded by the radii a1 and a2 (Fig. 3.4), we get, in an
obvious notation,
1
∆U = U (a2 ) − U (a1 ) ,
n
(3–21)
1
∆A = A(a2 ) − A(a1 ) .
n
P ®
a = a1
a = a2
Fig. 3.4. Template compartment

Since Ae and Ai differ only by a constant, this constant drops out in the
second equation of (3–21), and we obtain from (3–11) and (3–16)
2π
∆Ae = ∆Ai = G a22 + (c − b)2 − a21 + (c − b)2
n
(3–22)
− a22 + c2 + a21 + c2 .
On the other hand, ∆Ue = ∆Ui .

Note that we have for didactic reasons purposely used the compartments
corresponding to polar coordinates (Fig. 3.4) because they are so simple
and instructive, but also still useful for many purposes. For practical com-
putation, rectangular blocks (see Fig. 2.21) are almost exclusively used. For
conceptual purposes, however, the polar coordinate template remains invalu-
able; cf. Sect. 2.21.
3.3 Free-air reduction

For a theoretically correct reduction of gravity to the geoid, we need ∂g/∂H,
the vertical gradient of gravity. If g is the observed value at the surface of the
earth, then the value g0 at the geoid may be obtained as a Taylor expansion:
∂g
g0 = g − H ··· , (3–23)
∂H
where H is the height between P , the gravity station above the geoid, and
P0 , the corresponding point on the geoid (Fig. 3.1). Suppose there are no
masses above the geoid and neglecting all terms but the linear one, we have
g0 = g + F , (3–24)
where
∂g
F =− H (3–25)
∂H
is the free-air reduction to the geoid. Note that the assumption of no masses
above the geoid may be interpreted in the sense that such masses have been
mathematically removed beforehand, so that this reduction is indeed carried
out “in free air”.
For many practical purposes it is sufficient to use instead of ∂g/∂H the
normal gradient of gravity (associated with the ellipsoidal height h) ∂γ/∂h,
obtaining
. ∂γ .
F =− H = +0.3086 H [mgal] (3–26)
∂h
for H in meters.
3.4 Bouguer reduction 135
3.4 Bouguer reduction

The objective of the Bouguer reduction of gravity is the complete removal
of the topographic masses, that is, the masses outside the geoid.
The Bouguer plate

Assume the area around the gravity station P to be completely flat and
horizontal (Fig. 3.5), and let the masses between the geoid and the earth’s
surface have a constant density . Then the attraction A of this so-called
Bouguer plate is obtained by letting a → ∞ in (3–13), since the plate,
considered plane, may be regarded as a circular cylinder of thickness b = H
and infinite radius. By well-known rules of the calculus, we obtain
AB = 2π G H (3–27)
as the attraction of an infinite Bouguer plate. With standard density =

2.67 g cm−3 this becomes
AB = 0.1119 H [mgal] (3–28)
for H in meters.
Removing the plate is equivalent to subtracting its attraction (3–27) from
the observed gravity. This is called incomplete Bouguer reduction. Note that
this is the usual “plane” Bouguer plate; for a truly “spherical” Bouguer plate
we would have 4π instead of 2π (Moritz 1990: p. 235).
To continue and complete our gravity reduction, we must now apply the
free-air reduction F as given in (3–26). This combined process of removing
the topographic masses and applying the free-air reduction is called complete
Bouguer reduction. Its result is Bouguer gravity at the geoid:
gB = g − AB + F . (3–29)
P0
Fig. 3.5. Bouguer plate

With the assumed numerical values, we have
gravity measured at P g
minus Bouguer plate − 0.1119 H
plus free-air reduction + 0.3086 H (3–30)
Bouguer gravity at P0 gB = g + 0.1967 H .
Since gB now refers to the geoid, we obtain genuine gravity anomalies in the
sense of Sect. 2.12 by subtracting normal gravity γ referred to the ellipsoid:
∆gB = gB − γ . (3–31)
They are called Bouguer anomalies.
Terrain correction
This simple procedure is refined by taking into account the deviation of the
actual topography from the Bouguer plate of P (Fig. 3.6). This is called
terrain correction or topographic correction. At A the mass surplus ∆m+ ,
which attracts upward, is removed, causing g at P to increase. At B the
mass deficiency ∆m− is made up, causing g at P to increase again. The
terrain correction is always positive.
The practical determination of the terrain correction At is carried out
by means of a template, similar to that shown in Fig. 2.20, using (3–22) and
adding the effects of the individual compartments:

At = ∆A . (3–32)
A earth's surface
Dm +
H - HP
B P
Bouguer plate
Dm –
H
HP
geoid
P0
Fig. 3.6. Terrain correction

3.4 Bouguer reduction 137
Again, we can use a template in polar coordinates (Fig. 2.20) for theoretical
considerations or a rectangular grid (Fig. 2.21) for numerical computations.
For a surplus mass ∆m+ , H > Hp , we have
b = H − HP , c = 0; (3–33)
and for a mass deficiency ∆m− , H < HP ,
b = c = HP − H . (3–34)
By adding the terrain correction At to (3–29), the refined Bouguer gravity
gB = g − AB + At + F (3–35)
is obtained. The Bouguer reduction and the corresponding Bouguer anoma-

lies ∆gB are called refined or simple, depending on whether the terrain cor-
rection has been applied or not.
In practice it is convenient to separate the Bouguer reduction into the
effect of a Bouguer plate and the terrain correction, because the amount of
the latter is usually much less. Even for mountains 3000 m in height, the
terrain correction is only of the order of 50 mgal (Heiskanen and Vening
Meinesz 1958: p. 154).
Unified procedure
It is also possible to compute the total effect of the topographic masses,
AT = AB − At , (3–36)
in one step by using columns with base at sea level (Fig. 3.7), again sub-
HP
P0
Fig. 3.7. Bouguer reduction

dividing the terrain by means of a template. Note the difference between

AT , the attraction of the topographic masses, and the terrain correction At !
Then
AT = ∆A , (3–37)
where we now have b = H, c = HP . Use (3–13) with b = HP for the

innermost circle.
Instead of (3–35), we now have
gB = g − AT + F . (3–38)
The Bouguer reduction may be still further refined by the consider-

ation of density anomalies, anomalies in the free-air gradient of gravity
(Sect. 2.20), and spherical effects. More computational formulas may be
found in Jung (1961: Sect. 6.4).
3.5 Poincaré and Prey reduction

Suppose we need the gravity g inside the earth. Since g cannot be measured,
it must be computed from the surface gravity. This is done by reducing the
measured values of gravity according to the method of Poincaré and Prey.
We denote the point at which g is to be computed by Q, so that g = gQ .
Let P be the corresponding surface point so that P and Q are situated on
the same plumb line (Fig. 3.8). Gravity at P , denoted by gP , is measured.
The direct way of computing gQ would be to use the formula
P
∂g
gQ = gP − dH , (3–39)
Q ∂H
plumb line
Q
H = HP
earth's surface
z = HQ
geoid
P0 W = W0
Fig. 3.8. Prey reduction

3.5 Poincaré and Prey reduction 139
provided that the actual gravity gradient ∂g/∂H inside the earth were known.
It can be obtained by Bruns’ formula (2–40),
∂g
= −2g J + 4π G − 2ω 2 , (3–40)
∂H
if the mean curvature J of the geopotential surfaces and the density are
known between P and Q.
The normal free-air gradient is given by (2–147):
∂γ
= −2γ J0 − 2ω 2 , (3–41)
∂h
where J0 is the mean curvature of the spheropotential surfaces. If the ap-
proximation
.
g J = γ J0 (3–42)
is sufficient, then we get from (3–40) and (3–41)
∂g ∂γ
= + 4π G . (3–43)
∂H ∂h
Numerically, neglecting the variation of ∂γ/∂h with latitude, we find for the
density = 2.67 g cm−3 and (truncated) G = 6.67 · 10−11 m3 kg−1 s−2
∂g
= −0.3086 + 0.2238 = −0.0848 gal km−1 , (3–44)
∂H
so that (3–39) becomes
gQ = gP + 0.0848 (HP − HQ ) (3–45)
with g in gal and H in km. This simple formula, although being rather crude,
is often applied in practice.
The accurate way to compute gQ would be to use (3–39) and (3–40) with
the actual mean curvature J of the geopotential surfaces, but this would
require knowledge of the detailed shape of these surfaces far beyond what is
attainable today.
Another way of computing gQ , which is more practicable at present, is
the following. It is similar to the usual reduction of gravity to sea level (see
Sect. 3.4) and consists of three steps:
1. Remove all masses above the geopotential surface W = WQ , which
contains Q, and subtract their attraction from g at P .
2. Since the gravity station P is now “in free air”, apply the free-air
reduction, thus moving the gravity station from P to Q.
3. Restore the removed masses to their former position, and add alge-
braically their attraction to g at Q.
The purpose of this slightly complicated but logically clear procedure

is that in step 2 the free-air gradient can be used. If we here replace the
actual free-air gradient by the normal gradient ∂γ/∂h, then the error will
presumably be smaller than in using (3–43).
Note that the free-air gradient can also be accurately computed alterna-
tively by (2–394); the gravity anomalies ∆g in this formula are the gravity
anomalies obtained after performing step 2, that is, Bouguer anomalies re-
ferred to the lower point Q.
The effect of the masses above Q (steps 1 and 3) may be computed, e.g.,
by means of some kind of template or computer procedure for numerical
three-dimensional integration. If the terrain correction is neglected and only
the infinite Bouguer plate between P and Q of the normal density =
2.67 g cm−3 is taken into account, then we obtain with the steps numbered
as above:
gravity measured at P gP
1. remove Bouguer plate − 0.1119 (HP − HQ )
2. free-air reduction from P to Q + 0.3086 (HP − HQ )
3. restore Bouguer plate − 0.1119 (HP − HQ )
together: gravity at Q gQ = gP + 0.0848 (HP − HQ ) .
(3–46)
This is the same as (3–45), which is, thus, confirmed independently. We see
now that the use of (3–43) or (3–45) amounts to replacing the terrain with
a Bouguer plate.
Finally, we note that the reduction of Poincaré and Prey, abbreviated
as Prey reduction, yields the actual gravity which would be measured inside
the earth if this were possible. Its purpose is, thus, completely different from
the purpose of the other gravity reductions which give boundary values at
the geoid.
It cannot be directly used for the determination of the geoid but is needed
to obtain orthometric heights as will be discussed in Sect. 4.3. Actual gravity
g0 at a geoidal point P0 is related to Bouguer gravity gB , Eq. (3–38), by
g0 = gB − AT, P0 . (3–47)
It is obtained by subtracting from gB the attraction AT, P0 of the topographic

masses on P0 , which corresponds to restoring the topography after the free-
air reduction of Bouguer gravity from P to P0 .
3.6 Isostatic reduction 141
3.6 Isostatic reduction

3.6.1 Isostasy
One might be inclined to assume that the topographic masses are simply
superposed on an essentially homogeneous crust. If this were the case, the
Bouguer reduction would remove the main irregularities of the gravity field
so that the Bouguer anomalies would be very small and would fluctuate
randomly around zero. However, just the opposite is true. Bouguer anoma-
lies in mountainous areas are systematically negative and may attain large
values, increasing in magnitude on the average by 100 mgal per 1000 m of
elevation. The only explanation possible is that there is some kind of mass
deficiency under the mountains. This means that the topographic masses are
compensated in some way.
There is a similar effect for the deflections of the vertical. The actual
deflections are smaller than the visible topographic masses would suggest.
In the middle of the nineteenth century, J.H. Pratt observed such an effect
in the Himalayas. At one station in this area he computed a value of 28
for the deflection of the vertical from the attraction of the visible masses
of the mountains. The value obtained through astrogeodetic measurements
was only 5 . Again, some kind of compensation is needed to account for this
discrepancy.
Two different theories for such a compensation were developed at almost
exactly the same time, by J.H. Pratt in 1854 and 1859 and by G.B. Airy in
1855. According to Pratt, the mountains have risen from the underground
somewhat like a fermenting dough. According to Airy, the mountains are
floating on a fluid lava of higher density (somewhat like an iceberg floating
on water), so that the higher the mountain, the deeper it sinks.
Pratt–Hayford system
This system of compensation was outlined by Pratt and put into a mathe-
matical form by J.F. Hayford, who used it systematically for geodetic pur-
poses.
The principle is illustrated in Fig. 3.9. Underneath the level of compen-
sation there is uniform density. Above, the mass of each column of the same
cross section is equal. Let D be the depth of the level of compensation, reck-
oned from sea level, and let 0 be the density of a column of height D. Then
the density of a column of height D + H (H representing the height of the
topography) satisfies the equation
(D + H) = D 0 , (3–48)
4 km 6 km
2 km 3 km
5 km
H
H'
.
D = 100 km 2.67 2.62 2.57 2.59 2.52 2.67 2.76
level of compensation
Fig. 3.9. Pratt–Hayford isostasy model
which expresses the condition of equal mass. It may be assumed that

0 = 2.67 g cm−3 . (3–49)
According to (3–48), the actual density is slightly smaller than this normal
value 0 . Consequently, there is a mass deficiency which, according to (3–48),
is given by
h
∆ = 0 − = 0 . (3–50)
D+H
In the oceans, the condition of equal mass is expressed as
(D − H ) + H w = D 0 , (3–51)
where
w = 1.027 g cm−3 (3–52)
is the density and H
the depth of the ocean. Hence, there is a mass surplus
of a suboceanic column given by
H
− 0 = (0 − w ) . (3–53)
D − H
As a matter of fact, this model of compensation is idealized and schematic.
It can be only approximately fulfilled in nature. Values of the depth of com-
pensation around
D = 100 km (3–54)
are assumed.
For a spheroidal earth, the columns will converge slightly towards its
center, and other refinements may be introduced. We may postulate either
equality of mass or equality of pressure; each postulate leads to somewhat
different spherical refinements. It may be mentioned that for computational
reasons Hayford used still another, slightly different model; for instance, he
reckoned the depth of compensation D from the earth’s surface instead of
from sea level.
Airy–Heiskanen system
Airy proposed this model, and Heiskanen gave it a precise formulation for
geodetic purposes and applied it extensively. Figure 3.10 illustrates the prin-
ciple. The mountains of constant density
0 = 2.67 g cm−3 (3–55)
float on a denser underlayer of constant density
1 = 3.27 g cm−3 . (3–56)
The higher they are, the deeper they sink. Thus, root formations exist under
mountains, and “antiroots” under the oceans.
We denote the density difference 1 − 0 by ∆. On the basis of assumed
numerical values, we have
∆ = 1 − 0 = 0.6 g cm−3 . (3–57)
Denoting the height of the topography by H and the thickness of the cor-
responding root by t (Fig. 3.10), then the condition of floating equilibrium
is
t ∆ = H 0 , (3–58)
so that
0
t= H = 4.45 H (3–59)
∆
results. For the oceans, the corresponding condition is
t ∆ = H (0 − w ) , (3–60)
where H and w are defined as above and t is the thickness of the antiroot
(Fig. 3.10), so that we get
0 − w
t = H = 2.73 H (3–61)
1 − 0
2 km
H
H'
T = 30 km density 2.67
t'
8.9 km
density 3.27
Fig. 3.10. Airy–Heiskanen isostasy model
for the numerical values assumed.

Again, spherical corrections must be applied to these formulas for higher
accuracy, and the formulations in terms of equal mass and equal pressure
lead to slightly different results.
The normal thickness of the earth’s crust is denoted by T (Fig. 3.10);
values of around
T = 30 km (3–62)
are assumed. The crustal thickness under mountains is then
T +H +t (3–63)
and under the oceans it is

T − H − t . (3–64)
Vening Meinesz regional system

Both systems just discussed are highly idealized in that they assume the
compensation to be strictly local; that is, they assume that compensation
takes place along vertical columns. This presupposes free mobility of the
masses to a degree that is obviously unrealistic in this strict form.
For this reason, Vening Meinesz modified the Airy floating theory in 1931,
introducing regional instead of local compensation. The principal difference
between these two kinds of compensation is illustrated by Fig. 3.11. In Vening
regional compensation
local compensation
Fig. 3.11. Local and regional compensation
Meinesz’ theory, the topography is considered as a load on an unbroken but

yielding elastic crust.
In a very sloppy way which is only good for memorizing, we may say that,
standing on thin ice, Airy will break through, but under Vening Meinesz the
ice is stronger and will bend but not break.
Although Vening Meinesz’ refinement of Airy’s theory is more realistic, it
is more complicated and is, therefore, seldom used by geodesists because, as
we will see, any isostatic system, if consistently applied, serves for geodetic
purposes as well.
Geophysical and geodetic evidence shows that the earth is about 90% iso-
statically compensated, but it is difficult to decide, at least from gravimetric
evidence alone, which model best accounts for this compensation. Although
seismic results indicate an Airy type of compensation, in some places the
compensation seems to follow the Pratt model. Nature will never conform
to any of these models to the degree of precision which we have assumed
above. However, a well-defined and consistent mathematical formulation is
certainly a necessary prerequisite for the application of isostasy for geodetic
purposes.
For an extensive presentation of several types of isostasy, see Moritz (1990:
Chap. 8). The Vening Meinesz model has been treated in detail by Abd-
Elmotaal (1995); much information is also available in the internet. A clas-
sic on isostasy and its geophysical applications is Heiskanen and Vening
Meinesz (1958: Chaps. 5 and 7).
3.6.2 Topographic-isostatic reductions

The objective of the topographic-isostatic reduction of gravity is the regular-
ization of the earth’s crust according to some model of isostasy. Regulariza-
tion here means that we are trying to make the earth’s crust as homogeneous
as possible. The topographic masses are not completely removed as in the
Bouguer reduction but are shifted into the interior of the geoid in order
to make up the mass deficiencies that exist under the continents. In the
topographic-isostatic model of Pratt and Hayford, the topographic masses
are distributed between the level of compensation and sea level, in order to
bring the crustal density from its original value to the constant standard
value 0 . In the Airy–Heiskanen model, the topographic masses are used to
fill the roots of the continents, bringing the density from 0 = 2.67 g/cm3
to 1 = 3.27 g/cm3 .
In other terms, the topography is removed together with its compensa-
tion, and the final result is ideally a homogeneous crust of density 0 and
constant thickness D (Pratt–Hayford) or T (Airy–Heiskanen).
Thus we have three steps:
1. removal of topography,
2. removal of compensation,
3. free-air reduction to the geoid.
Steps 1 and 3 are known from Bouguer reduction, so that the techniques of
Sect. 3.4 can be applied to them. Step 2 is new and will be discussed now
for the two main topographic-isostatic systems.
Pratt–Hayford system
The method is the same as for the terrain correction, Sect. 3.4, Eq. (3–32).
The attraction of the (negative) compensation is again computed by

AC = ∆A , (3–65)
where the attraction of a vertical column representing a compartment is

given by (3–22) with
b = D , c = D + HP (3–66)
and replaced by the density defect ∆. If the preceding Bouguer reduction
were done with the original density of the column expressed by
D
= 0 (3–67)
D+H
according to (3–48), then ∆ would be given by (3–50).
Usually the Bouguer reduction is performed using a constant density 0 ;

the density defect ∆ must then be computed by
H
∆ = 0 , (3–68)
D
which differs slightly from (3–50), in order to restore equality of mass ac-
cording to
(0 − ∆) D + 0 H = 0 D . (3–69)
The first term on the left-hand side represents the mass of the layer between
the level of compensation and sea level; the second term represents the mass
of the topography, now assumed to have a density 0 .
Airy–Heiskanen system
Again we use
AC = ∆A , (3–70)
where b and c in (3–22) are, according to Fig. 3.12, given by
b = t, c = HP + T + t , (3–71)
and is replaced by ∆ = 1 − 0 = 0.6 g/cm3 .
P
HP H
P0
T
HP + T + t
T+t
root t
Fig. 3.12. Topography and compensation – Airy–Heiskanen model
Total reduction
In analogy with (3–38), the topographic-isostatically reduced gravity on the
geoid becomes
gTI = g − AT + AC + F , (3–72)
where −AC is the attraction of the compensation which is actually negative,

so that its removal is equivalent to the term +AC . The quantity AT is the
attraction of topography, to be computed as the effect of a Bouguer plate
combined with terrain correction, Eq. (3–36), or in one step, as described in
Sect. 3.4; F is the free-air reduction approximated by (3–26).
Oceanic stations
Here the terms AT and F of (3–72) are zero, since the station is situated on
the geoid, but the term AC is more complicated.
In the Pratt–Hayford model, the procedure is as follows. The mass sur-
plus (3–53) of a suboceanic column of height D − H (Fig. 3.9) is removed
and used to fill the corresponding oceanic column of height H to the proper
density 0 . In mathematical terms, this is
AC = −A1 + A2 , (3–73)
where both A1 and A2 are of the form (3–32), ∆A is given by (3–22). For
A1 we have
b = D − H , c = D, (3–74)
and density − 0 ; for A2 we have
b = c = H (3–75)
and density 0 − w .
In the Airy–Heiskanen model, the mass surplus of the antiroot, 1 − 0 ,
is used to fill the oceans to the proper density 0 . The corresponding value
is again given by (3–73), where for A1 we now have
b = t , c=T, (3–76)
and density 1 − 0 ; and for A2 we have, as before,
b = c = H (3–77)
and density 0 − w .
In both models, Eq. (3–72) reduces for oceanic stations to
gTI, ocean = g + AC . (3–78)

3.7 The indirect effect 149
Topographic-isostatic anomalies
The topographic-isostatic gravity anomalies are – in analogy to the Bouguer
anomalies – defined by
∆gTI = gTI − γ . (3–79)
If any of the topographic-isostatic systems were rigorously true, then the
topographic-isostatic reduction would fulfil perfectly its goal of complete reg-
ularization of the earth’s crust, which would become level and homogeneous.
Then, with a properly chosen reference model for γ, the topographic-isostatic
gravity anomalies (3–79) would be zero.
The actual topographic-isostatic compensation occurring in nature can-
not completely conform to such abstract models. As a consequence, nonzero
topo-graphic-isostatic gravity anomalies will be left, but they will be small,
smooth, and more or less randomly positive and negative. On account of
this smoothness and independence of elevation, they are better suited for in-
terpolation or extrapolation than any other type of anomalies; see Chap. 9,
particularly Sect. 9.7.
It may be stressed again that for geodetic purposes the topographic-
isostatic model used must be mathematically precise and self-consistent, and
the same model must be used throughout. Refinements include the consid-
eration of irregularities of density of the topographic masses and the consid-
eration of the anomalous gradient of gravity.
3.7 The indirect effect

The removal or shifting of masses underlying the gravity reductions change
the gravity potential and, hence, the geoid. This change of the geoid is an
indirect effect of the gravity reductions.
Thus, the surface computed by Stokes’ formula from topographic-isostatic
gravity anomalies, is not the geoid itself but a slightly different surface, the
cogeoid. To every gravity reduction there corresponds a different cogeoid.
Let the undulation of the cogeoid be N c . Then the undulation N of the
actual geoid is obtained from
N = N c + δN (3–80)
by taking into account the indirect effect on N , which is given by

δW
δN = , (3–81)
γ
where δW is the change of potential at the geoid. Equation (3–81) is an
application of Bruns’ theorem (2–237).
The change of potential, δW , is for the Bouguer reduction expressed by
δWB = UT (3–82)
and for the topographic-isostatic reduction by
δWTI = UT − UC , (3–83)
the subscripts of the potential U corresponding to those of the attraction A

used in the preceding sections.
For the practical determination of UT and UC , the template technique,
as expressed in (3–32), may again be used (at least, conceptually):

U= ∆U , (3–84)
where the relevant formulas are the first equation of (3–21), (3–9), (3–12),
and (3–15). The point U refers to is always the point P0 at sea level (Fig. 3.1).
For UT we use U0 , see (3–12), with b = H and density 0 (see Fig. 3.12). For
UC in the continental case, we use Ue , see (3–9), with the following values:
Pratt–Hayford,
H
b = c = d , density 0 ; (3–85)
D
Airy–Heiskanen,
b = t, c = t+T , density 1 − 0 . (3–86)
The corresponding considerations for the oceanic case are left as an exercise
for the reader.
The indirect effect with Bouguer anomalies is very large, of the or-
der of ten times the geoidal undulation itself. See the map at the end of
Helmert (1884: Tafel I), where the maximum value is 440 m! The reason is
that the earth is in general topographic-isostatically compensated. There-
fore, the Bouguer anomalies cannot be used for the determination of the
geoid.
With topographic-isostatic gravity anomalies, as might be expected, the
indirect effect is smaller than N , of the order of 10 m. It is necessary, how-
ever, to compute the indirect effect δNI carefully, using exactly the same
topographic-isostatic model as for the gravity reductions.
Furthermore, before applying Stokes’ formula, the topographic-isostatic
gravity anomalies must be reduced from the geoid to the cogeoid. This is
done by a simple free-air reduction, using (3–26), by adding to ∆gI the
correction
δ = +0.3086 δN [mgal] , (3–87)
3.8 The inversion reduction of Rudzki 151
δN in meters. This correction δ is the indirect effect on gravity; it is of the

order of 3 mgal.
Now the topographic-isostatic gravity anomalies refer strictly to the co-
geoid. The application of Stokes’ formula gives N c , which according to (3–80)
is to be corrected by the indirect effect δN to give the undulation N of the
actual geoid.
Deflections of the vertical

The indirect effect on the deflections of the vertical is, in agreement with
Eqs. (2–377), given by
1 ∂ δN
δξ = − ,
R ∂ϕ
(3–88)
1 ∂ δN
δη = − .
R cos ϕ ∂λ
The indirect effect is essentially identical with the so-called topographic-

isostatic deflection of the vertical (Heiskanen and Vening Meinesz 1958:
pp. 252–255).
The topographic-isostatic reduction as such is very much alive, however.
It is practically the only gravity reduction used for geoid determination at
the present time (with the possible exception of free-air reduction, which is
a case by itself).
The last purely gravimetric geoid, before the advent of satellites, was the
Columbus Geoid (Heiskanen 1957).
3.8 The inversion reduction of Rudzki

It is possible to find a gravity reduction where the indirect effect is zero.
This is done by shifting the topographic masses into the interior of the geoid
in such a way that
UC = UT . (3–89)
Then
δW = UT − UC = 0 . (3–90)
This procedure was given by M. P. Rudzki in 1905. For the present purpose,
we may consider the geoid to be a sphere of radius R (Fig. 3.13). Let the
mass element dm at Q be replaced by a mass element dm at a certain point
Q inside the geoid situated on the same radius vector. The potential due to
P0
l
R l'
r
Ã Q
Q' (dm)
r' (dm')
Fig. 3.13. Rudzki reduction as an inversion in a sphere
these mass elements at the geoidal point P0 is

dm G dm
dUT = G = ,
l r + R2 − 2R r cos ψ
2
(3–91)
dm G dm
dUC = G = .
l r 2 + R2 − 2R r cos ψ
We should have
dUC = dUT (3–92)
if
R
dm = dm (3–93)
r
and
R2
r = . (3–94)
r
This is readily verified by substitution into the second equation of (3–91).
The condition (3–94) means that Q and Q are related by inversion in the
sphere of radius R (Kellogg 1929: p. 231). Therefore, this reduction method
is called inversion reduction or Rudzki reduction.
The condition (3–93) expresses the fact that the compensating mass dm
is not exactly equal to dm but is slightly smaller. Since this relative decrease
of mass is of the order of 10−8 , it may be safely neglected by setting
dm = dm . (3–95)
Usually it is even sufficient to replace the sphere by a plane. Then Q is the

ordinary mirror image of Q (Fig. 3.14).
3.9 The condensation reduction of Helmert 153
P Q
HP H
P0
Q0
Q0Q = Q0Q' H
Q'
Fig. 3.14. Rudzki reduction as a plane approximation
Rudzki gravity at the geoid becomes, in analogy to (3–72),
gR = g − AT + AC + F , (3–96)
#
where AC = ∆A with b = H, c = H + HP , the density being equal to
that of topography.
Since the indirect effect is zero, the cogeoid of Rudzki coincides with the
actual geoid, but the gravity field outside the earth is changed, which to-
day is in the center of attention. In addition, the Rudzki reduction does not
correspond to a geophysically meaningful model. Nevertheless, it is impor-
tant conceptually. Regard it an interesting historic curiosity, but never even
consider to use it!
3.9 The condensation reduction of Helmert

Here the topography is condensed so as to form a surface layer (somewhat
like a glass sphere made of very thin but very heavy and robust glass) on the
geoid so that the total mass remains unchanged. Again, the mass is shifted
along the local vertical (Fig. 3.15).
We may consider Helmert’s condensation as a limiting case of an isostatic
H
%
geoid
· = %H
Fig. 3.15. Helmert’s method of condensation

reduction according to the Pratt–Hayford system as the depth of compen-

sation D goes to zero. This is sometimes useful.
Again we have
gH = g − AT + AC + F , (3–97)
#
where AC = ∆A is now to be computed using the second equation of
(3–19) with c = HP and κ = H; HP is the height of the station P and H
the height of the compartment.
The indirect effect is
δW = UT − UC . (3–98)
#
The potential UC = ∆U is to be computed using the first equation of
(3–19) with κ = H as before, but c = 0 since it refers to the geoidal point
P0 . The corresponding δN is very small, amounting to about 1 m per 3 km of
average elevation. It may, therefore, usually be neglected so that the cogeoid
of the condensation reduction practically coincides with the actual geoid.
Even the “direct effect”, −AT + AC , can usually be neglected, as the
attraction of the Helmert layer nearly compensates that of the topography.
There remains
gH = g + F , (3–99)
that is, the simple free-air reduction. In this sense, the simple free-air reduc-
tion may be considered as giving approximate boundary values at the geoid,
to be used in Stokes’ formula. To the same degree of approximation, the
“free-air cogeoid” coincides with the actual geoid.
Hence, the free-air anomalies
∆gF = g + F − γ (3–100)
may be considered as approximations of “condensation anomalies”
∆gH = gH − γ . (3–101)
The many facets of free-air reduction

This is one of the most basic, most difficult, and most fascinating topics of
physical geodesy. In fact, the free-air anomaly means several conceptually
different but related concepts.
1. The term F above has been seen to be part of every gravity reduction
rather than a full-fledged gravity reduction itself.
2. Approximately, free-air anomalies may be identified with Helmert’s
condensation anomalies as we have seen above.
3.9 The condensation reduction of Helmert 155
3. Rigorously, free-air anomalies can even be considered as resulting from

a mass-transporting gravity reduction, in a similar sense as the iso-
static anomaly. Just imagine that you transport the masses above the
geoid into its interior in such a way that the external potential remains
unchanged ! This reminds us of Rudzki’s reduction (geoid potential re-
mains constant) but is rather different. The most important advantage
is that the free-air anomaly in the present sense leaves the external po-
tential unchanged which nowadays is much more important than the
geoid. The greatest disavantage is that it cannot be computed: we do
not know how to shift the masses so that the external masses remain
unchanged. In logical terms, the Rudzki reduction is constructive –
we are told how to do it –, whereas the present reduction is non-
constructive – we do not know how to do it directly. More about this
in Sects. 8.2, 8.6, 8.9, and 8.15. We shall, thus, attempt to cut the
difficult cake into easier pieces.
These are the main methods that have been proposed for the reduction of
gravity. A simple overview is given by Fig. 3.16.
Pratt Airy Rudzki Helmert
H H H H
T H
D t
Fig. 3.16. Topography and compensation for different gravity reductions

4 Heights
4.1 Spirit leveling

The principle of spirit leveling is well known. To measure the height difference
δHAB between two points A and B, vertical rods are set up at each of
these two points and a level (leveling instrument) somewhere between them
(Fig. 4.1). Since the line Ā B̄ is horizontal, the difference in the rod readings
l1 = AĀ and l2 = B B̄ is the height difference:
δHAB = l1 − l2 . (4–1)
If we measure a circuit, that is, a closed leveling line where we finally return
to the initial point, then the algebraic sum of all measured differences in
height will not in general be rigorously zero, as one would expect, even if we
had been able to observe with perfect precision. This misclosure indicates
that leveling is more complicated than it appears at first sight.
Let us look into the matter more closely. Figure 4.2 shows the relevant
geometrical principles. Let the points A and B be so far apart that the pro-
cedure of Fig. 4.1 must be applied repeatedly. Then the sum of the leveled
height differences between A and B will not be equal to the difference in the
orthometric heights HA and HB . The reason is that the leveling increment
δn, as we henceforth denote it, is different from the corresponding increment
δHB of HB (Fig. 4.2), due to the nonparallelism of the level surfaces. De-
noting the corresponding increment of the potential W by δW , we have by
(2–21)
−δW = g δn = g δHB , (4–2)
_ _
A B
l2
l1
B
±HAB
A
Fig. 4.1. Spirit leveling

158 4 Heights
±n ±HB
A HB
HA
geoid
Fig. 4.2. Leveling and orthometric height
where g is the gravity at the leveling station and g is the gravity on the
plumb line of B at δHB . Hence,
g
δHB = δn = δn . (4–3)
g
There is, thus, no direct geometrical relation between the result of lev-
eling and the orthometric height, since (4–3) expresses a physical relation.
What, then, if not height, is directly obtained by leveling? If gravity g is also
measured, then
δW = −g δn (4–4)
is determined, so that we obtain
B

WB − WA = − g δn . (4–5)
A
Thus, leveling combined with gravity measurements furnishes potential dif-

ferences, that is, physical quantities.
It is somewhat more rigorous theoretically to replace the sum in (4–5)
by an integral, obtaining
B
WB − WA = − g dn . (4–6)
A
Note that this integral is independent of the path of integration; that is,
different leveling lines connecting the points A and B (Fig. 4.3) should give
the same result. This is evident because W is a function of position only;
therefore, to every point there corresponds a unique value W . If the leveling
line returns to A, then the total integral must be zero:
$
g dn = −WA + WA = 0 . (4–7)
4.2 Geopotential numbers and dynamic heights 159
A
Fig. 4.3. Two different leveling lines connecting A and B (taken to-
gether, they form a circuit)
%
The symbol denotes an integral over a circuit.
On the other hand, the measured height difference, that is, the sum of
the leveling increments
B
B
∆nAB = δn = dn , (4–8)
A A
depends on the path of integration and is, thus, not in general zero for a
circuit: $
dn = misclosure = 0 . (4–9)
In mathematical terms, dn is not a perfect differential (the differential of a

function of position), whereas dW = −g dn is perfect, so that dn becomes a
perfect differential when it is multiplied by the integrating factor (−g).
Thus, potential differences are the result of leveling combined with grav-
ity measurements. They are basic to the whole theory of heights; even ortho-
metric heights must be considered as quantities derived from potential differ-
ences. Leveling without gravity measurements, although applied in practice,
is meaningless from a rigorous point of view, for the use of leveled heights
(4–8) as such leads to contradictions (misclosures); it will not be considered
here.
4.2 Geopotential numbers and dynamic heights

Let O be a point at sea level, that is, simplifying speaking, on the geoid;
usually a suitable point on the seashore is taken. Let A be another point,
connected to O by a leveling line. Then, by formula (4–6), the potential
difference between A and O can be determined. The integral
A
g dn = W0 − WA = C , (4–10)
0
160 4 Heights
which is the difference between the potential at the geoid and the potential
at the point A, has been introduced as the geopotential number of A in
Sect. 2.4. It is defined so as to be always positive.
As a potential difference, the geopotential number C is independent of
the particular leveling line used for relating the point to sea level. It is the
same for all points of a level surface; it can, thus, be considered as a natural
measure of height, even if it does not have the dimension of a length.
The geopotential number C is measured in geopotential units (g.p.u.),
where
1 g.p.u. = 1 kgal m = 1000 gal m. (4–11)
.
Using g = 0.98 kgal in (4–10), we get
. .
C = g H = 0.98 H , (4–12)
so that the geopotential numbers in g.p.u. are almost equal to the height
above sea level in meters.
The geopotential numbers were adopted at a meeting of a Subcommission
of the IAG at Florence in 1955. Formerly, the dynamic heights were used,
defined by
C
H dyn = , (4–13)
γ0
where γ0 is normal gravity for an arbitrary standard latitude, usually 45◦ :
γ45◦ = 9.806 199 203 m s−2 = 980.6 199 203 gal (4–14)
for the GRS 1980. Just note and keep in mind that 1 gal = 10−2 m s−2 and,
accordingly, 1 mgal = 10−5 m s−2 .
The dynamic height differs from the geopotential number only in the
scale or the unit: The division by the constant γ0 in (4–13) merely con-
verts a geopotential number into a length. However, the dynamic height has
no geometrical meaning whatsoever, so that the division by an arbitrary γ0
somehow obscures the true physical meaning of a potential difference. Hence,
the geopotential numbers are, for reasons of theory and for practically es-
tablishing a national or continental height system, preferable to the dynamic
heights.
Dynamic correction
It is sometimes convenient to convert the measured height difference ∆nAB
of (4–8) into a difference of dynamic height by adding a small correction.
Using Eqs. (4–13) and (4–10) gives
B
1 1
∆HAB = HB − HA =
dyn dyn dyn
(CB − CA ) = g dn , (4–15)
γ0 γ0 A
4.3 Orthometric heights 161
which may be rewritten as

B B B
1 g − γ0
∆HABdyn
= (g − γ0 + γ0 ) dn = dn + dn , (4–16)
γ0 A A A γ0
so that
dyn
∆HAB = ∆nAB + DCAB , (4–17)
where
B
. g − γ0
B
g − γ0
DCAB = dn = δn (4–18)
A γ0 γ0
A
is the dynamic correction.

As a matter of fact, the dynamic correction may also be used for com-
puting differences of geopotential numbers. We at once obtain
CB − CA = γ0 ∆nAB + γ0 DCAB . (4–19)
4.3 Orthometric heights

We denote the intersection of the geoid and the plumb line through point P
by P0 (Fig. 4.4). Let C be the geopotential number of P , that is,
C = W0 − W , (4–20)
and H its orthometric height, that is, the length of the plumb-line segment
between P0 and P . Perform the integration in (4–10) along the plumb line
P0 P . This is permitted because the result is independent of the path. We
then get
H
C= g dH . (4–21)
0
H earth's surface
geoid
P0
Fig. 4.4. Gravity reduction

162 4 Heights
This equation contains H in an implicit way. It is also possible to get H

explicitly. From
dW dC
dC = −dW = g dH , dH = − = , (4–22)
g g
we obtain
W C
dW dC
H=− = . (4–23)
W0 g 0 g
As before, the integration is extended over the plumb line.
The explicit formula (4–23), however, is of little practical use. It is better
to transform (4–21) in a way that at first looks entirely trivial:
H
1 H
C= g dH = H · g dH , (4–24)
0 H 0
so that
C = ḡ H , (4–25)
where
1 H
ḡ = g dH (4–26)
H 0
is the mean value of the gravity along the plumb line between the geoid,
point P0 , and the surface point P . From (4–25) it follows that
C
H= , (4–27)
ḡ
which permits H to be computed if the mean gravity ḡ is known. Since ḡ
does not strongly depend on H, Eq. (4–27) is a practically useful formula
and not merely a tautology. For determining the mean gravity ḡ, Eq. (4–26)
may be written

1 H
ḡ = g(z) dz , (4–28)
H 0
where g(z) is the actual gravity at the variable point Q which has the height z
(Fig. 3.8). The simplest approximation is to use the simplified Prey reduction
of (3–45):
g(z) = g + 0.0848 (H − z) , (4–29)
where g is the gravity measured at the surface point P . The integration
(4–28) can now be performed immediately, giving

1 H
ḡ = g + 0.0848 (H − z) dz
H 0
(4–30)
0.0848 z 2 H
=g+ Hz−
H 2 0
or
ḡ = g + 0.0424 H (g in gal, H in km) . (4–31)
The factor 0.0424 refers to the normal density = 2.67 g/cm3 . The corre-
sponding formula for arbitrary constant density is, by (3–43),

1 ∂γ
ḡ = g − + 2π G H . (4–32)
2 ∂h
If we use ḡ according to (4–31) or (4–32) in the basic formula (4–27), we

obtain the so-called Helmert height:
C
H= (4–33)
g + 0.0424 H
with C in g.p.u., g in gal and H in km.

As we have seen in Sect. 3.5, this approximation replaces the terrain
with an infinite Bouguer plate of constant density and of height H. This is
often sufficient. Sometimes, in high mountains and for highest precision, it
is necessary to apply to g a more rigorous Prey reduction, such as the three
steps described in Sect. 3.5. A practical and very accurate method for this
purpose has been given by Niethammer in 1932. It takes the topography into
account, assuming only that the free-air gradient is normal and the density
is constant down to the geoid.
It is also sufficient to calculate ḡ as the arithmetic mean of gravity g mea-
sured at the surface point P and of gravity g0 computed at the corresponding
geoidal point P0 by the Prey reduction:
1
ḡ = 2 (g + g0 ) . (4–34)
This presupposes that gravity g varies linearly along the plumb line. This
can usually be assumed with sufficient accuracy, even in extreme cases, as
shown by Mader (1954) and by Ledersteger (1955).
Orthometric correction
The orthometric correction is added to the measured height difference, in
order to convert it into a difference in orthometric height.
We let the leveling line connect two points A and B (Fig. 4.5) and apply
a simple trick first:
∆HAB = HB − HA = HB − HA − HB
dyn dyn
+ HA dyn
+ (HB − HA
dyn
)
(4–35)
dyn
= ∆HAB + (HB − HB
dyn
) − (HA − HA
dyn
).
164 4 Heights
earth's
B surface
geoid
A0 B0
Fig. 4.5. Orthometric and dynamic correction
From (4–17), we have

dyn
∆HAB = ∆nAB + DCAB . (4–36)
Consider now the differences between the orthometric and dynamic heights,
HA − HA dyn
and HB − HB dyn
. Imagine a fictitious leveling line leading from
point A0 at the geoid to the surface point A along the plumb line. Then the
measured height difference would be HA itself: ∆nA0A = HA , so that
dyn
DCA0 A = ∆HA 0A
− ∆nA0 A = HA
dyn
− HA (4–37)
and
HA − HA
dyn
= −DCA0 A ,
(4–38)
HB − HB
dyn
= −DCB0 B .
Substituting (4–36) and (4–38) into (4–35), we finally have
∆HAB = ∆nAB + DCAB + DCA0 A − DCB0 B (4–39)
or
∆HAB = ∆nAB + OCAB , (4–40)
where
OCAB = DCAB + DCA0 A − DCB0 B (4–41)
is the orthometric correction. This is a remarkable relation between the or-
thometric and dynamic corrections (Ledersteger 1955). We may write this
OCAB = DCAB + DCA0 A + DCBB0 , (4–42)
where we have reversed the sequence of the subscripts of the last term and,
consequently, the sign. With DCB0 A0 = 0 (why?), we may write
OCAB = DCAB + DCBB0 + DCB0 A0 + DCA0 A . (4–43)

Accordingly, this may be written
OCAB = DCABB0 A0 A . (4–44)
Thus, the orthometric correction from A to B equals the dynamic correc-

tion over the loop ABB0 A0 A, a curious, but practically completely useless
relation equivalent to (4–41). (Question: Why is this independent of γ0 ?)
From (4–18), we find
B B
g − γ0
g − γ0
DCAB = dn = δn ,
A γ0 γ0
A
A
g − γ0 ḡA − γ0
DCA0 A = dH = HA , (4–45)
A0 γ0 γ0
B
g − γ0 ḡB − γ0
DCB0 B = dH = HB ,
B0 γ0 γ0
where ḡA and ḡB are the mean values of gravity along the plumb lines of A
and B. Thus, the orthometric correction (4–41) becomes
B
g − γ0 ḡA − γ0 ḡB − γ0
OCAB = δn + HA − HB . (4–46)
γ0 γ0 γ0
A
Here again we need the mean values of gravity along the plumb lines, ḡA and
ḡB ; γ0 is an arbitrary constant for which we always take normal gravity at
45◦ latitude.
Accuracy
Let us first evaluate the effect on H of an error in the mean gravity ḡ. From
H = C/ḡ, we obtain by differentiation
C H
δH = − 2
δḡ = − δḡ . (4–47)
ḡ ḡ
Since ḡ is about 1000 gal, we have, neglecting the minus sign, the simple
formula
.
δH[mm] = δḡ[mgal] H[km] , (4–48)
the subscripts denoting the units; δH is the error in H, caused by an error
δḡ in ḡ.
For H = 1 km,
.
δH[mm] = δḡ[mgal] , (4–49)
166 4 Heights
which shows that an error δḡ in the order of 100 mgal falsifies an elevation
of 1000 m by only 10 cm.
Let us now estimate the effect of an error of the density on ḡ. Differ-
entiating (4–32) and omitting the minus sign we find
δḡ = 2π G H d . (4–50)
If δ = 0.1 g cm−3 and H = 1 km, then
δḡ = 4.2 mgal , (4–51)
which causes an error of 4 mm in H. A density error of 0.6 g/cm3 , which

corresponds to the maximum variation of rock density occurring in practice,
falsifies H = 1000 m by only 25 mm.
Mader (1954) has estimated the difference between the simple computa-
tion of mean gravity according to Helmert, Eq. (4–32), and more accurate
methods that take the terrain correction into account. He found for Hochtor,
in the Alps, H = 2504 m:
Helmert ḡ = 980.263 (Bouguer plate only),

Niethammer 286
(also terrain correction) . (4–52)
1
ḡ = 2 (g + g0 ) 285
Mean gravity ḡ according to (4–34) differs from Niethammer’s value by only

1 mgal, which shows the linearity of g along the plumb line even in an extreme
case. This corresponds to a difference in H of 3 mm. The simple Helmert
height differs by about 6 cm from these more elaborately computed heights.
Therefore, the differences are very small even in this rather extreme case;
we see that orthometric heights can be obtained with very high accuracy.
This is of great importance for a discussion of the recent theory of Moloden-
sky from a practical point of view. See Chap. 8, particularly Sect. 8.11.
4.4 Normal heights

Assume for the moment the gravity field of the earth to be normal, that is,
W = U , g = γ, T = 0. On this assumption compute “orthometric heights”;
they will be called normal heights and denoted by H ∗ . Thus, Eqs. (4–21)
through (4–26) become
H∗
W0 − W = C = γ dH ∗ , (4–53)
0
4.4 Normal heights 167
C
∗ dC
H = , (4–54)
0 γ
C = γ̄ H ∗ , (4–55)
where H∗
1
γ̄ = ∗ γ dH ∗ (4–56)
H 0
is the mean normal gravity along the plumb line.
As the normal potential U is a simple analytic function, these formulas
can be evaluated straightforwards; but since the potential of the earth is
evidently not normal, what does all this mean? Consider a point P on the
physical surface of the earth. It has a certain potential WP and also a certain
normal potential UP , but in general WP = UP . However, there is a certain
point Q on the plumb line of P , such that UQ = WP ; that is, the normal
potential U at Q is equal to the actual potential W at P . The normal height
H ∗ of P is nothing but the ellipsoidal height of Q above the ellipsoid, just
as the orthometric height of P is the height of P above the geoid.
For more details the reader is referred to Sect. 8.3; Fig. 8.2 illustrates
the geometric relations.
We now give some practical formulas for the computation of normal
heights from geopotential numbers. Writing (4–56) in the form
H∗
1
γ̄ = ∗ γ(z) dz (4–57)
H 0
corresponding to (4–28), then we can express γ(z) by (2–215) as

2 3 2
γ(z) = γ 1 − 1 + f + m − 2f sin ϕ z + 2 z ,
2
(4–58)
a a
where γ is the gravity at the ellipsoid, depending on the latitude ϕ but not
on z. Thus, straightforward integration with respect to z yields

1 2 z2 3 z 3 H ∗
γ̄ = ∗ γ z − 1 + f + m − 2f sin ϕ
2
+ 2
H a 2 a 3 0
(4–59)
1 ∗ 1 ∗2 1 ∗3
= ∗γ H − 1 + f + m − 2f sin ϕ H + 2 H
2
H a a
or H ∗2 H ∗2
γ̄ = γ 1 − 1 + f + m − 2f sin ϕ 2
+ 2 . (4–60)
a a
This formula may be used for computing H ∗ by the formula
C
H∗ = . (4–61)
γ̄
168 4 Heights
The mean theoretical gravity itself depends on H ∗ , by (4–60), but not

strongly, so that an iterative solution is very simple.
It is also possible to give a direct expression of H ∗ in terms of the geopo-
tential number C by substituting (4–60) into (4–61) and expanding into a
series of powers of H ∗ :

∗ C 1
∗ ∗2
H = 1+ 1 + f + m − 2f sin ϕ H + O(H ) .
2
(4–62)
γ a
Solving this equation for H ∗ and expanding H ∗ in powers of C/γ, we obtain

2
∗ C
C C
H = 1 + 1 + f + m − 2f sin ϕ
2
+ , (4–63)
γ aγ aγ
where γ is normal gravity at the ellipsoid, for the same latitude ϕ. The
accuracy of this formula will be sufficient for almost all practical purposes;
still more accurate expressions are given in Hirvonen (1960).
Corresponding to the dynamic and orthometric corrections, there is a
normal correction NC of the measured height differences. Equation (4–46)
immediately yields, on replacing ḡ by γ̄ and H by H ∗ :
B
g − γ0 γ̄A − γ0 ∗ γ¯B − γ0 ∗
NCAB = δn + HA − HB , (4–64)
γ0 γ0 γ0
A
so that
∗ ∗ ∗
∆HAB = HB − HA = ∆nAB + NCAB . (4–65)
The normal heights were introduced by Molodensky in connection with his
method of determining the physical surface of the earth; see Chap. 8.
4.5 Comparison of different height systems

By means of the geopotential number
point
C = W0 − W = g dn , (4–66)
geoid
we can write the different kinds of height in a common form which is very
instructive:
C
height = , (4–67)
G0
4.5 Comparison of different height systems 169
where the height systems differ according to how the gravity value G0 in the
denominator is chosen. We have:
dynamic height: G0 = γ0 = constant ,
orthometric height: G0 = ḡ , (4–68)
normal height: G0 = γ̄ .
It is seen that one can devise an unlimited number of other height systems
by selecting G0 in a different way.
The geopotential number C is, in a way, the most direct result of leveling
and is of great scientific importance. However, it is not a height in a geomet-
rical or practical sense. While the dynamic height has at least the dimension
of a height, it has no geometrical meaning. One advantage is that points of
the same level surface have the same dynamic height; this corresponds to the
intuitive feeling that if we move horizontally, we remain at the same height.
Note that the orthometric height differs for points of the same level surface
because the level surfaces are not parallel. This gives rise to the well-known
paradoxes of “water flowing uphill”, etc.
The dynamic correction can be very large, because gravity varies from
equator to pole by about 5000 mgal. Take, for instance, a leveling line of
.
1000 m difference of height at the equator, where g = 978.0 gal, computed
.
with γ0 = γ45◦ = 980.6 gal. Then (4–18) gives a dynamic correction of
approximately
978.0 − 980.6
DC = · 1000 m = −2.7 m . (4–69)
980.6
Because of these large corrections, dynamic heights are not suitable as prac-
tical heights, and the geopotential numbers are preferable for scientific pur-
poses.
Orthometric heights are the natural “heights above sea level”, that is,
heights above the geoid. Therefore, they have an unequalled geometrical
and physical significance. Their computation is relatively laborious, unless
Helmert’s simple formula (4–33) is used, which is sufficient in most cases.
The orthometric correction is rather small. In the Alpine leveling line of
Mader (1954), leading from an elevation of 754 m to 2505 m, the orthometric
correction is about 15 cm per 1 km of measured height difference. See also
Sect. 8.15.
The physical and geometrical meaning of the normal heights is less ob-
vious; they depend on the reference ellipsoid used. Although they are basic
in the new theories of physical geodesy, they have a somewhat artificial
character as compared to the orthometric heights. They are, however, easy
170 4 Heights
to compute rigorously; the order of magnitude of the normal corrections is

about the same as that of the orthometric corrections. In some countries
they have replaced the orthometric heights in practice.
For estimates of the difference between orthometric height H and normal
height H ∗ , we refer the reader to Sect. 8.13.
All these height systems based on C are functions of position only. There
are, thus, no misclosures, as there are with measured heights. From a purely
practical point of view, the desired requirements of a height system are that
1. misclosures be eliminated,
2. corrections to the measured heights be as small as possible.
Empirical height systems have been devised to give smaller corrections than
either the orthometric or the normal heights. They have no clear physical
significance, however, and are beyond the scope of this book.
Accuracy
Leveling is one of the most accurate geodetic measurements. A standard
error of ±0.1 mm per km distance is possible; it increases with the square
root of the distance.
If the error of measurement and interpolation, etc., of gravity is negligible,
then the differences in the geopotential number C can be determined with
an accuracy of ±0.1 gal m per km distance; this corresponds to ±0.1 mm
in measured height. Referring to gravity measurements, it is sufficient to
measure at distances of some kilometers.
Dynamic heights and normal heights are clearly as accurate as the geopo-
tential numbers, because normal gravity γ is errorless. Orthometric heights,
however, are also affected by imperfect knowledge of density, etc., but only
slightly; see the end of Sect. 4.3.
Triangulated heights
Historically and for the sake of completeness, the determination of heights
by triangulation, that is, by means of zenith angles, should be mentioned.
The main problem is the atmospheric refraction affecting the zenith an-
gles. Thus, the accuracy of triangulated heights is much less than that of
leveling. Consequently, triangulated heights are not considered any longer
here.
For small distances (e.g., < 1 km), trigonometric height measurements,
referred to the local plumb line, have the character of a leveled height dif-
ference δn This fact may be used (with care!) to fill small gaps in a leveling
network.
4.6 GPS leveling 171
Remark on misclosures
All misclosures in any acceptable system of heights denoted for the moment
by h (not to be confused with ellipsoidal heights) must be zero:
$
dh = 0 (4–70)
for any closed path. Height networks consisting of triangles, if computed by

least-squares adjustment, thus must satisfy the condition that the sum of
height differences must be zero for each triangle. Mathematically, this can
be shown to be equivalent to the commutativity of second derivatives:
∂2h ∂2h
= . (4–71)
∂x ∂y ∂y ∂x
4.6 GPS leveling

Spirit leveling (Fig. 4.6) is a very time-consuming operation. GPS has intro-
duced a revolution also here. The basic equation is
H =h−N. (4–72)
This equation relates the orthometric height H (above the geoid), the el-
lipsoidal height h (above the ellipsoid), and the geoidal undulation N . If
any two of these quantities are measured, then the third quantity can be
computed.
If h is measured by GPS, and if there exists a reliable digital geoid map
of N , then the orthometric height H can be obtained immediately.
Equation (4–72) can also be used for geoid determination: if h is measured
by GPS, and H is available from leveling, then the geoid N can be determined
as N = h − H. The same principle can be applied even on the oceans as
satellite altimetry, as we will see later in Chap. 7, e.g., Eq. (7–47).
GPS leveling implies replacing to some extent the classical leveling by
GPS. Referring to Fig. 4.6 and applying (4–72) to A and B leads to
HA = hA − NA ,
(4–73)
HB = hB − NB ,
and the height difference
HB − HA = hB − hA − NB + NA . (4–74)
Introducing the notations δHAB = HB − HA , δhAB = hB − hA , and δNAB =
NB − NA , the relation reduces to
δHAB = δhAB − δNAB . (4–75)
172 4 Heights
_ _
A B
l2
l1 B
earth's surface
HB – HA
A
HB
HA hB
hA geoid
NA NB
ellipsoid
Fig. 4.6. GPS leveling
With GPS leveling, δhAB is obtained, so that with a known geoid, i.e., known
δNAB , the orthometric height difference δHAB may be computed accord-
ing to (4–75). This is a tremendous advantage since otherwise the classical
leveling together with gravity measurements is required to determine the
orthometric height difference, see Eqs. (4–40) and (4–46).
Note that only the difference of the geoidal undulations impacts the
result.
5 The geometry of the earth
5.1 Overview
This chapter consists of three parts.
Part I: Global reference systems after GPS

A fundamental task is to define a global reference system based on a reference
ellipsoid which is a good global representation of the earth and which should
ideally be characterized by the following properties: Its center coincides with
the geocenter (the earth’s center of mass), its z-axis represents a suitably
defined mean rotation axis of the earth, and the xz-plane is parallel to a
mean plane close to the Greenwich meridian. The reference ellipsoid itself is
defined to be an ellipsoid of revolution that globally approximates the geoid
best in some global sense.
Actually, such a geometric or physical definition cannot be absolutely ac-
curately and unambiguously realized; the final definition will always contain
an arbitrary conventional element.
To make things even more complicated, the earth is not a completely
rigid body. It can (again approximately!) be regarded as an elastic body
with a liquid core. It undergoes small more or less periodic changes. So it
must be referred to a mean ellipsoid that does not change with time.
All this will be taken for granted in the present introductory treatment.
We shall assume a well-defined geocentric reference ellipsoid with rigid di-
mensions, a fixed origin, and a time-invariant orientation – close to reality
but, in principle, conventionally adopted. For temporal changes in the earth’s
body and rotation, the reader may be referred to Moritz and Mueller (1987).
Before the advent of satellite geodesy, a geocentric reference system could
not be realized. Thus, we had to work with a local geodetic system displaced
with respect to the geocenter by an unknown amount on the order of up to
a few hundred meters. Therefore, we must take into account a translation
(parallel shift) of the local reference ellipsoid with respect to a geocentric
system. This implies three translation parameters.
Note that “local” here is used in the sense of “regional”, i.e., for a country,
territory, or region, in contrast to “global”.
Usually, the orientation of a local reference system is accurately known
since the direction of the xyz-axes was accessible by astronomical measure-
ments quite accurately at least for the last two centuries. Thus, the orienta-
tion of a local geodetic datum is known to the order of 0.1 (arc seconds).
174 5 The geometry of the earth
Today, we can readily determine the deviation of a local system or local

datum from a global reference system. We have the deviation of
• size and shape of the reference ellipsoid (a, f ),
• translation (x0 , y0 , z0 ), and
• orientation (three very small Euler angles ε1 , ε2 , ε3 ).
Since GPS is very well established (cf. Hofmann-Wellenhof et al. 2001),
we assume a general knowledge for granted and recapitulate in this book
only some basic facts.
Part II: Three-dimensional geodesy: a transition

This part considers how the concepts of geodesy in the modern sense of Molo-
densky, Marussi, and Hotine would look shortly before the advent of satel-
lites, but already including electronically measured spatial distances (trilat-
eration). We work with local Cartesian coordinates rotated in a known way
by the astronomically measurable quantities Φ, Λ, A (astronomical latitude,
longitude, azimuth), considered as Eulerian angles of rotation of the local
with respect to the global axes. However, we have no means to determine
the geocenter. So the situation is somewhat more complicated but still geo-
metrically well defined and transparent. “Local” here means “strictly local”,
varying from point to point together with their plumb lines defined by (Φ, Λ).
The main problem with this approach is the impossibility of measuring
precise zenith angles because of atmospheric refraction. We may say that
the vertical dimension is much worse defined than the horizontal dimension.
Finally, we shall consider how terrestrial and GPS data can be combined.
Part III: Local geodetic datum

The way out of the dilemma of the worse vertical dimension is a complete
separation of horizontal and vertical and determining the latter by the dif-
ferential method of astrogeodetic geoid determination. This was a “2+1-
dimensional” rather than a three-dimensional approach, logically more com-
plicated but practically more accurate. In fact, the former (and present)
astrogeodetic methods can be understood much better by deriving them
from the global situation. Thus, today with GPS we are in a much better
position practically as well as theoretically: the classical local datums can be
understood best by their relation with the global geometry. “Local geodetic
system” or “local geodetic datum” is again meant in the sense of “regional”,
e.g., the North-American Datum or the European Datum.
GPS permits to separate the geometry from the gravity field, which con-
tinues to be a challenge for physical geodesy to be solved by a combination
of terrestrial and satellite data.
5.2 Introduction 175
Part I: Global reference systems after GPS
5.2 Introduction
Geodesy, as the theory of size and shape of the earth, is not a purely geo-
metrical science since the earth’s gravity field, a physical entity, is involved
in many geodetic measurements, especially terrestrial ones.
The gravimetric methods are usually considered to constitute physical
geodesy in the narrower sense. The measurements of triangulation, leveling,
and geodetic astronomy, all make essential use of the plumb line, which,
being the direction of the gravity vector, is no less physically defined by
nature than its magnitude, that is, the gravity g. All determinations of the
geoid by various methods and its use as well as the use of deflections of the
vertical belong to physical geodesy, quite as well as the gravimetric methods.
Even in the age of GPS, we have many previous geodetic data which
continue to be useful and have to be understood in order to be optimally
combined with the new satellite data. In precise operations of engineering
geodesy such as tunnel surveying, the plumb line and deflections of the ver-
tical must be taken into account.
For an optimal understanding and use of local (or rather regional) geo-
detic datums, we must know their relation to a global geodetic system as
used in GPS. Therefore, it is appropriate to start with global geometry in a
rather elementary way.
A few introductory ideas may help in comprehending this subject. To
fix the position of a point in space, we need three coordinates. We can use,
and have used, a rectangular Cartesian coordinate system. This is the basic
geometric coordinate system. It may be easily converted computationally to
ellipsoidal coordinates ϕ, λ, h referred to any given reference ellipsoid.
For many special purposes, however, it is preferable to take what we have
called the natural coordinates: Φ (astronomical latitude), Λ (astronomical
longitude), and H (orthometric height), which directly refer to the gravity
field of the earth (Sect. 2.4). The height H may be obtained by geometric
leveling, combined with gravity measurements, and Φ and Λ are determined
by astronomical measurements.
As long as the geoid can be identified with an ellipsoid, the use of these
coordinates for computations is very simple. Since this identification is suf-
ficient only for results of rather low accuracy, the deviations of the geoid
from an ellipsoid must be taken into account. As we have seen, the geoid has
rather disagreeable mathematical properties. It is a complicated surface with
discontinuities of curvature. Thus, it is not suitable as a surface on which to
perform mathematical computations directly, as on the ellipsoid.
To repeat, the ellipsoidal coordinates ϕ, λ, h are defined such as to refer

to the ellipsoid exactly as the natural coordinates refer to the geoid, hence
their names.
Since the deviations of the geoid from the ellipsoid are small and com-
putable, it is convenient to add small reductions to the original coordinates
Φ, Λ, H, so as to get values which refer to an ellipsoid. In this way we shall
find in Sect. 5.12:
ϕ=Φ−ξ,
λ = Λ − η sec ϕ , (5–1)
h =H +N;
ϕ and λ are the ellipsoidal coordinates on the ellipsoid, sometimes also called
geodetic latitude and geodetic longitude to distinguish them from the astro-
nomical latitude Φ and the astronomical longitude Λ. Astronomical and el-
lipsoidal coordinates differ by the deflection of the vertical (components ξ
and η). The quantity h is the geometric height above the ellipsoid; it differs
from the orthometric height H above the geoid by the geoidal undulation N .
Geodetic measurements (angles, distances) are treated similarly. The
principle of triangulation is well known: historically, distances were obtained
indirectly by measuring the angles in a suitable network of triangles; only
one baseline was necessary in principle to furnish the scale of the network.
Triangulation was indispensable in former times, because angles could be
measured much more easily than long distances.
Nowadays, however, long distances can be measured directly just as eas-
ily as angles by means of electronic instruments, so that triangulation, using
angular measurements, is often replaced or supplemented by trilateration,
using distance measurements. The computation of triangulations and trilat-
erations on the ellipsoid is easy. It is, therefore, convenient to reduce the
measured angles, baselines, and long distances to the ellipsoid, in much the
same way as the astronomical coordinates are treated. Then the ellipsoidal
coordinates ϕ, λ obtained (1) by reducing the astronomical coordinates and
(2) by computing triangulations or trilaterations on the ellipsoid can be
compared; they should be identical for the same point.
Today, of course, GPS is the best method for determining ϕ, λ, and h
directly.
5.3 The Global Positioning System

The following sections on the Global Positioning System (GPS) are extracted
from Hofmann-Wellenhof et al. (2003: Sect. 9.3) which in return is based on
5.3 The Global Positioning System 177
Hofmann-Wellenhof et al. (2001: Chap. 2). For details supplementing the

compact description here, the reader is referred to these books.
5.3.1 Basic concept

GPS is the responsibility of the Joint Program Office (JPO), a component
of the Space and Missile Center at El Segundo, California. In 1973, the JPO
was directed by the U.S. Department of Defense (DOD) to establish, develop,
test, acquire, and deploy a spaceborne positioning system. The present nav-
igation system with timing and ranging is the result of this initial directive.
GPS was conceived as a ranging system from known positions of satellites
in space to unknown positions on land, at sea, in air, and in space. The
original objectives of GPS were the instantaneous determination of position
and velocity on a continuous basis, and the precise coordination of time (i.e.,
time transfer).
Based on code or carrier phase measurements, GPS uses pseudoranges
derived from the broadcast satellite signal.
Using the code measurements, the pseudorange is derived from measuring
the travel time of the coded signal and multiplying it by its velocity. Since
the clocks of the receiver and the satellite are never perfectly synchronized, a
clock error must be taken into account. Consequently, each equation of this
type comprises four unknowns: the three point coordinates contained in the
true range and the clock error. Thus, four satellites are necessary to solve
for the four unknowns. Indeed, the GPS concept assumes that – without
obstruction – four or more satellites are in view at any location on or near
the earth 24 hours a day.
Using carrier phase measurements, ambiguities must be taken into ac-
count as additional unknowns. For more details see Hofmann-Wellenhof et
al. (2001: Sect. 6.1.2).
5.3.2 System architecture

Space segment
Constellation
The GPS satellites have nearly circular orbits with an altitude of about
20200 km above the earth, i.e., they are mean earth orbit (MEO) satellites,
yielding a period of nominally 12 sidereal hours. The nominal constellation
consists of 24 operational satellites deployed in six evenly spaced planes (A
to F) with an inclination of 55◦ against the equator and with four satel-
lites per plane. Furthermore, active spare satellites for replenishment may
be operational. See http://tycho.usno.navy.mil/gpscurr.html for the current

status.
With the nominal constellation, the space segment provides global cover-
age with four to eight simultaneously observable satellites above 15◦ elevation
angle at any time of day. If the elevation mask is reduced to 10◦ , occasionally
up to 10 satellites will be visible; and if the elevation mask is further reduced
to 5◦ , occasionally 12 satellites will be visible.
Satellites categories
Essentially, the GPS satellites provide a platform for radio transceivers,
atomic clocks, computers, and various ancillary equipment. The electronic
equipment of each satellite allows the user to measure a pseudorange to the
satellite, and each satellite broadcasts a message which allows the user to
determine the spatial position of the satellite for arbitrary instants. The aux-
iliary equipment of each satellite, among others, consists of solar panels for
power supply and a propulsion system for orbit and stability control.
There are several classes or types of GPS satellites. These are the Block I,
Block II, Block IIA, Block IIR, Block IIR-M, and the future Block IIF and
Block III satellites. An up-to-date description is difficult because new nota-
tions are introduced in a rather arbitrary way; an example is the recently
introduced notation Block IIR-M.
Eleven Block I satellites were launched in the period between 1978 to
1985. Today, none of them is in operation anymore.
The essential difference between Block I and Block II satellites is related
to U.S. national security. Block I satellite signals were fully available to
civilian users. Starting with Block II, satellite signals may be restricted for
civilian use. The Block II satellites are equipped with mutual communication
capability. Some of them carry retroreflectors and can be tracked by laser
ranging.
The Block IIR satellites (“R” denotes replenishment or replacement)
have a design life of 10 years. They are equipped with improved facilities for
communication and intersatellite tracking. Block IIR-M satellites incorpo-
rate two new military signals and a second civil signal. The first Block IIR-M
was launched on September 25, 2005.
Currently (April 2006), the first launch of a Block IIF satellite (“F”
denotes follow on) is scheduled for 2008 (instead of the previously projected
dates mid of 2006 and 2007). These satellites will broadcast a third civil
signal on L5 (see Sect 5.3.5).
Presently, the DOD undertakes studies for the next generation of GPS
satellites, called Block III satellites. Preliminary dates (likely to change) are
2011/12 for first launches and on-orbit tests (Civil GPS Service Interface
Committee 2002). These satellites will be characterized by an assured and

improved level of integrity without the need of augmentation.
Satellite signal
The key to the accuracy of the system is the fact that all signal compo-
nents are precisely controlled by atomic clocks. These highly accurate fre-
quency standards of GPS satellites produce the fundamental frequency of
10.23 MHz. Coherently derived from this frequency are (presently) two sig-
nals in the L-band, the L1 and the L2 carrier waves generated by multiplying
the fundamental frequency by 154 and 120, respectively, yielding
L1 = 1575.42 MHz ,
L2 = 1227.60 MHz .
These dual frequencies are essential for eliminating the major source of error,
i.e., the ionospheric refraction.
The pseudoranges that are derived from measured travel times of the
signal from each satellite to the receiver use two pseudorandom noise (PRN)
codes that are modulated onto the two carriers.
The C/A-code (coarse/acquisition-code) is available for civilian use. Each
C/A-code is a unique sequence of 1023 bits, called chips, which is repeated
each millisecond. The duration of each C/A-code chip is about 1 µs. Equiv-
alently, the chip length – denoted also as wavelength or chip width (Misra
and Enge 2001: Sect. 2.3.1) – is about 300 m. The C/A-code is presently
modulated upon L1 only and is purposely omitted from L2. This omission
allows the JPO to control the information broadcast by the satellite and,
thus, denies full system accuracy to nonmilitary users.
The P-code (precision-code) has been reserved for U.S. military and other
authorized users. This is achieved by using the W-code to encrypt the P-
code to the Y-code (anti-spoofing). The P-code has an effective chip length
of about 30 m. The P-code is modulated on both carriers L1 and L2.
In addition to the PRN codes, a data message is modulated onto the
carriers consisting of status information, satellite clock bias, and satellite
ephemerides. The orbit data are given as Kepler-like elements and are de-
noted as broadcast ephemerides. The full set of elements is given in, e.g.,
Montenbruck and Gill (2001: Sect. A.2.2). It is worth noting that the present
signal structure will be improved in the near future (see Sect. 5.3.5).
Control segment
The operational control system (OCS) consists of a master control station,
monitor stations, and ground control stations. The main tasks of the OCS
are tracking of the satellites for the orbit and clock determination and predic-
tion, time synchronization of the satellites, and upload of the data message
to the satellites.
Master control station

The master control station is located at the Consolidated Space Opera-
tions Center (CSOC) at Shriver Air Force Base, Colorado Springs, Colorado.
CSOC collects the tracking data from the monitor stations and calculates
the satellite orbit and clock parameters by a Kalman estimator. These re-
sults are then passed to one of the three ground control stations for eventual
upload to the satellites. The satellite control and system operation is also
the responsibility of the master control station.
Monitor stations
There are five monitor stations located at Hawaii, Colorado Springs, As-
cension Island in the South Atlantic Ocean, Diego Garcia in the Indian
Ocean, and Kwajalein in the North Pacific Ocean. Each of these stations
is equipped with a precise atomic time standard and receivers which con-
tinuously measure pseudoranges to all satellites in view. Pseudoranges are
measured every 1.5 seconds and, using ionospheric and meteorological data,
they are smoothed to produce 15-minute interval data which are transmitted
to the master control station.
Ground control stations

These stations collocated with the monitor stations at Ascension, Diego Gar-
cia, and Kwajalein are the communication links to the satellites and mainly
consist of the ground antennas. The satellite ephemerides and clock informa-
tion, calculated at the master control station and received via communication
links, are uploaded to each GPS satellite via S-band radio links.
User segment
The diversity of the military and civilian users is matched by the type of
receivers available today.
On the basis of the type of observables (i.e., code pseudoranges or phase
pseudoranges) and of the availability of codes (i.e., C/A-code, P-code, or
Y-code), GPS receivers can be classified. For the majority of navigation ap-
plications, C/A-code pseudorange receivers will suffice. With this type of
receiver, only code pseudoranges using the C/A-code on L1 are measured.
Typical devices output the three-dimensional position either in latitude, lon-
gitude, and height or in some map projection systems, e.g., universal trans-
verse Mercator (UTM) coordinates and height.
5.3.3 Satellite signal and observables

Components of the signal
The official description of the GPS signal is given in the GPS Interface
Control Document ICD-GPS-200, available at www.navcen.uscg.gov. Details
may also be found in Spilker (1996).
The (current) components of the signal are summarized in Table 5.1.
Note that the nominal fundamental frequency f0 is intentionally reduced by
about 0.005 Hz to compensate for relativistic effects.
The navigation message essentially contains information about the satel-
lite health status, the satellite clock, the orbit, and various correction data.
The parameters in the block of orbit information are the reference epoch,
six parameters to describe a Kepler ellipse at the reference epoch, three
secular correction terms and six periodic correction terms.
Observables
In concept, the GPS observables are ranges which are deduced from mea-
sured time or phase differences based on a comparison between received
signals and receiver-generated signals. As mentioned earlier, the ranges are
biased by satellite and receiver clock errors and, consequently, they are de-
noted as pseudoranges. Essentially, pseudoranges differ from distances by an
unknown additive constant.
Apart from the satellite and the receiver clock bias, further error sources
can be classified into three groups, i.e., satellite-related errors (e.g., orbital
errors), signal propagation medium-related errors (e.g., ionospheric and tro-
pospheric refraction), and receiver-related errors (e.g., antenna phase center
variation, multipath), but are omitted in the subsequent simplified models.
Extended models are given in Hofmann-Wellenhof et al. (2003: Sect. 10.2.2).
Table 5.1. Components of the satellite signal
Component Frequency or code Wavelength

chipping rate [MHz]
Fundamental frequency f0 = 10.23
Carrier L1 154 f0 = 1575.42 19.0 cm
Carrier L2 120 f0 = 1227.60 24.4 cm
P-code f0 = 10.23
C/A-code f0 /10 = 1.023
Navigation message f0 /204 600 = 50 · 10−6
Code pseudoranges
The measured time difference ∆t is affected by the satellite clock error δS
and the receiver clock error δ. The error δS of the satellite clock can be
modeled by a polynomial with the coefficients being transmitted in the nav-
igation message. Assuming the δS correction is applied, the time interval ∆t
multiplied by the speed of light c yields the code pseudorange R and, hence,
R = c ∆t . (5–2)
Assuming a common time reference for satellite and receiver, e.g., GPS time,
the term ∆t may be decomposed into the run time ∆t(GPS) and the receiver
clock errors δ leading to
R = c ∆t(GPS) + c δ = + c δ , (5–3)
where is the geometric range between the satellite and the receiver. The
receiver module responsible for code pseudorange measurements is denoted
as delay lock loop (DLL). Details on the DLL functionality are given in Misra
and Enge (2001: Sect. 9.5).
Phase pseudoranges
Assuming again that the satellite clock error correction is applied, the phase
pseudorange Φ is modeled by
λΦ = + cδ + λN , (5–4)
where the carrier wavelength λ has been introduced. The range represents
the distance between the satellite at emission epoch t and the receiver at
reception epoch t + ∆t. Phase measurements are ambiguous, since the initial
integer number N of cycles between satellite and receiver is unknown. As
long as the tracking of a satellite is not interrupted, the ambiguity remains
constant within the tracking loop of the receiver. The responsible receiver
hardware is denoted as phase lock loop (PLL). Compared to (5–3), the phase
pseudorange differs from the code pseudorange only by the phase ambiguity
term λ N . Dividing the above equation by λ scales the phase to cycles.
As mentioned previously, the majority of navigation applications does not
need carrier phase measurements. Only for increased accuracy requirements
(e.g., relative positioning; see below), phase measurements become relevant.
Doppler data
Some of the first solution models proposed for GPS were to use the Doppler
observable. Considering Eq. (5–4), the equation for the observed Doppler
shift scaled to range rate is given by
D = λ Φ̇ = ˙ + c δ̇ , (5–5)
where the derivatives with respect to time are indicated by a dot. The raw
Doppler shift is less accurate than integrated Doppler.
The Doppler shift is measured in the carrier tracking loop of a GPS re-
ceiver (Misra and Enge 2001: Sect. 9.6). Assuming a known satellite velocity,
the Doppler shift can be used to estimate the velocity of the user.
5.3.4 System capabilities and accuracies

Two operational capabilities are distinguished: firstly, the initial operational
capability (IOC) and, secondly, the full operational capability (FOC).
IOC was attained in July 1993, when 24 (Block I/II/IIA) GPS satellites
were operating and were available for navigation. Officially, IOC was declared
by the DOD on December 8, 1993.
FOC was achieved when 24 Block II/IIA satellites were operational in
their assigned orbits and the constellation was tested for operational military
performance. Even though 24 Block II and Block IIA satellites were available
since March 1994, FOC was not declared before July 17, 1995 which indicates
an extensive testing phase.
The selection of the GPS observation technique depends upon the par-
ticular requirements of the project; especially the desired accuracy plays a
dominant role.
Point positioning
When using a single receiver, usually point positioning with code pseudo-
ranges is performed. The concept of point positioning is simple (Fig. 5.2).
Without clock errors, trilateration in space (i.e., using three ranges) solves
the task to determine the point coordinates. Using pseudoranges, four ob-
servations are necessary to account for the three coordinate components and
the receiver clock error. For point positioning, GPS provides two levels of
service: the standard positioning service (SPS) with access for civilian users
and the precise positioning service (PPS) with access for authorized users.
SPS performance standards are based on signal-in-space performance.
Contributions of ionosphere, troposphere, receiver, multipath, topography, or
interference are not included. Furthermore, SPS is provided on the L1 signal
only; the L2 signal is not part of the SPS (Department of Defense 2001).
The global average positioning domain accuracy amounts to 13 m horizontal
error (95% probability level) and 22 m vertical error (95% probability level).
The PPS has access to both codes and provides accuracies down to the
meter level.
Differential GPS
Selective availability (SA), the deliberate degradation of the point position-
ing accuracy by “dithering” (i.e., distorting on purpose) the satellite clock
(called δ-process) and manipulating the ephemerides (called ε-process), has
led to the development of differential GPS (DGPS). Only the basic idea is
explained here.
DGPS is based on the use of two (or more) receivers, where one (station-
ary) reference or base receiver is located at a known point and the position
of the (mostly moving) remote receiver is to be determined. Using code pseu-
doranges, at least four common satellites must be tracked simultaneously at
both sites. The known position of the reference receiver is used to calculate
corrections to the observed pseudoranges. These corrections are then trans-
mitted via telemetry (i.e., controlled radio link) to the roving receiver and
allow the computation of the rover position with far more accuracy than for
the single-point positioning mode.
Using DGPS based on C/A-code pseudoranges, real-time accuracies at
the 1–5 m level can be routinely achieved. Phase-smoothed code ranges yield
the submeter level (Lachapelle et al. 1992). Even higher accuracies can be
reached by the use of carrier phases (precise DGPS). For ranges up to some
20 km, accuracies at the subdecimeter level can be obtained in real time (De-
Loach and Remondi 1991). To achieve this accuracy, the ambiguities must
be resolved “on the fly” and, therefore, (generally) dual-frequency receivers
are required. Furthermore, five satellites per epoch are required.
After the deactivation of SA in May 2000, DGPS must be seen from a
different viewpoint. The increased point positioning accuracy achieved with
a single receiver may suffice for some kinds of applications.
Relative positioning
At present, highest accuracies are achieved in the relative-positioning mode
with observed carrier phases. Relative positioning is associated with base-
lines, i.e., the three-dimensional vector between a known reference station
and the location to be determined. Processing a baseline requires that the
phases are simultaneously observed at both baseline endpoints (Fig. 5.1).
Originally, relative positioning was only possible by postprocessing data.
Today, (near) real-time data transfer over short baselines is routinely possi-
ble, which enables real-time computation of baseline vectors and has led to
the real-time kinematic (RTK) technique.
satellite
vector
baseline
unknown station
reference station
Fig. 5.1. Concept of relative positioning
Static relative positioning

The reference station and the unknown station are static, i.e., no motion
occurs between the two points of the baseline. When highest accuracy is an
issue, then this is the preferred method. Fully depending on the application
and on the length of the baseline, the observation time may amount from
several tens of minutes to many hours. Referring to navigation, where usually
motion is involved, static relative positioning is of minor importance. The
reader is referred to Hofmann-Wellenhof et al. (2001: Sect. 7.1.2) for details.
Kinematic relative positioning

The kinematic method is very productive because the greatest number of
points can be determined in the least time.
The drawback is that after initialization a continuous lock on at least
four satellites must be maintained.
The semikinematic or stop-and-go technique is characterized by alterna-
tively stopping and moving one receiver to determine the positions of fixed
points along the trajectory. The most important feature of this method is the
increase in accuracy when several measurement epochs at the stop locations
are accumulated and averaged. This technique is often referred to simply as
kinematic method. Relative positional accuracies at the centimeter level can
be achieved for baselines up to some 20 km.
The kinematic technique requires the resolution of the phase ambiguities
by initialization which can be performed by static or kinematic techniques.
Currently available commercial software (for dual-frequency receivers) only
requires 1–2 minutes of observation for baselines up to 20 km to resolve the
ambiguities kinematically (“on the fly”).
5.3.5 GPS modernization concept

In January 1999, the USA announced the GPS modernization concept, a
$400 million initiative. The key feature is the implementation of a new signal
structure in future satellites.
Future GPS satellites

The Block IIR satellites increase their presence in the GPS constellation. A
new effort will bring modernized functionality to IIR satellites. These mod-
ernized satellites, denoted as IIR-M (replenishment-modernization), will pro-
vide new services to military and civilian users. New signals and increased
L-band power will significantly improve the navigation performance (Mar-
quis 2001).
The Block IIF and the Block III satellites are the next generations.
These next generations of satellites will have many improvements over
the present satellites. It is planned to include the capability to transmit data
between satellites to make the system more independent. The autonomous
navigation (auto-nav) capability via intersatellite cross-link ranging will al-
low the satellites to essentially position themselves without extensive ground
tracking. In summary, the future satellites will have the following mainly mil-
itary advantages:
• Navigation accuracy will be maintained for six months without ground

support and control.
• Uplink jamming concerns will be minimized.
• One upload per spacecraft per month instead of one or even more per
day will be performed.
• Need for overseas stations to support navigation uploads will be re-
duced.
• Improved navigation accuracy will be achieved.
New signal structure

Referring to codes, presently civil users have unlimited access only to the
C/A-code on the carrier L1. The modernization will provide new signals:
implementing military codes (M-codes) on L1 and on L2 and a civilian code
on L2 (abbreviated as L2c). The M-code will provide the authorized users
with more signal security, improved acquisition options, and more jamming
resistance. The new civilian L2c signal will provide nonauthorized users dual-
frequency operation to perform ionospheric error correction. In addition to
these codes, a new L5 frequency will be provided for civilian users to en-
hance aviation applications. The notation L5 is chosen because, actually,
5.4 From GPS to coordinates 187
the satellites transmit additional signals at frequencies referred to as L3 and

L4. These signals are classified and for military purposes only (Misra and
Enge 2001: Sect. 2.3).
According to the modernization initiative released in 1999, the Inter-
agency GPS Executive Board concept will be realized with the following
specifications. Future GPS signals will be transmitted by three carriers where
L1 and L2 remain unchanged, and the new carrier L5 is specified as
L5 = 115f0 = 1176.45 MHz ,
where f0 = 10.23 MHz denotes the basic GPS frequency. The carrier L5,
placed in a protected aeronautical radio navigation service band, was re-
cently allocated by the World Radio Conference organized regularly by the
International Telecommunication Union (Vorhies 2000).
Note that both new civil GPS signals will have two codes. L5 will not
share with military signals and use two equal-length codes in phase quadra-
ture, each clocked at 10.23 MHz. L2 is shared between civil and military
signals. The new L2c signal provides two codes by time multiplexing. The
two codes are of different length (Fontana et al. 2001). The existing military
Y-code will be replaced by new (split) M-codes.
The linear carrier phase combination of L2 with L5 results in a signal
with a wavelength of about 5.9 m. Long wavelengths facilitate ambiguity
resolution. By contrast, the linear combination of L1 with L5 will be used
as ionosphere-free combination because large frequency differences are ad-
vantageous for calculating ionospheric corrections. The common processing
of phase data from all three carriers will be performed in the three-carrier
ambiguity resolution approach (Vollath et al. 1999).
A perspective for the implementation is given in the 2001 Federal Radio-
navigation Plan: IOC (18 satellites in orbit with the new L2c signal and
M-code capability) is planned for 2008 and FOC (24 satellites in orbit) is
planned for 2010. At least one satellite is planned to be operational with the
new L5 capability no later than 2005, with IOC planned for 2012 and FOC
planned for 2014.
5.4 From GPS to coordinates

So far, we have got an introductory GPS overview. Now we are interested in
applying elementary GPS approaches to demonstrate how coordinates are
obtained. Two examples, as simple as possible, are selected: point positioning
and relative positioning.
5.4.1 Point positioning with code pseudoranges

The situation is shown in Fig. 5.2. The coordinates of A are to be determined
by using GPS. As we know from Sect. 5.3.4, four pseudoranges to different
satellites are necessary to determine the three coordinate components of A
and the receiver clock error. Generalizing (5–3), we obtain
j
RA (t) = jA (t) + c δA (t) . (5–6)
j
This is the code pseudorange at an epoch t, where RA (t) is the measured
code pseudorange between the observing site A (as indicated in Fig. 5.2) and
the satellite j, and jA (t) is the geometric distance between the satellite and
the observing point, and c is the speed of light. The last item is the receiver
clock error δA (t). Note that we assume the simplest possible model, thus,
we do not consider ionospheric and tropospheric influences, other biases and
errors.
Examining Eq. (5–6), the desired point coordinates to be determined are
implicitly comprised in the distance jA (t), which can explicitly be written
as

j
A (t) = (X j (t) − XA )2 + (Y j (t) − YA )2 + (Z j (t) − ZA )2 , (5–7)
where the WGS 84 (World Geodetic System 1984, see Sect. 2.11) coordinates
X j (t), Y j (t), Z j (t) are the components of the geocentric position vector of
the satellite at epoch t, and XA , YA , ZA are the three unknown WGS 84 co-
ordinates of the observing site, which might be denoted (XA , YA , ZA )WGS 84
or, which means the same, (XA , YA , ZA )GPS .
How many unknowns are involved? Note that the satellite coordinates
X j (t), Y j (t), Z j (t) may always be assumed known (more precisely, are cal-
k
j l
%Ak (t) m
Z WGS-84 %Al (t)
j
%A (t) %A
m
(t)
A (XA ,YA ,ZA)
Y WGS-84
ZA
XA
X WGS-84
YA
Fig. 5.2. Point positioning

culable) from the information broadcast by the satellite. Therefore, there re-
main the three unknown station coordinates XA , YA , ZA and the unknown
receiver clock error δA (t). In other terms, at least four satellites are required
to set up four equations of type (5–6). Denoting the satellites by j, k, l, m,
the corresponding system of equations
j
RA (t) = jA (t) + c δA (t) ,
k (t) = k (t) + c δ (t) ,
RA A A
(5–8)
l (t) = l (t) + c δ (t) ,
RA A A
m (t) = m (t) + c δ (t)

RA A A
is obtained or, by substituting (5–7) accordingly,

j

RA (t) = (X j (t) − XA )2 + (Y j (t) − YA )2 + (Z j (t) − ZA )2 + c δA (t) ,

k (t) =
RA (X k (t) − XA )2 + (Y k (t) − YA )2 + (Z k (t) − ZA )2 + c δA (t) ,

l (t) =
RA (X l (t) − XA )2 + (Y l (t) − YA )2 + (Z l (t) − ZA )2 + c δA (t) ,
m (t) =

RA (X m (t) − XA )2 + (Y m (t) − YA )2 + (Z m (t) − ZA )2 + c δA (t)
(5–9)
results. This system of equations comprises only the previously mentioned
four unknowns XA , YA , ZA and the unknown receiver clock error δA (t) and
may, thus, be solved. We do not consider linearization, possible redundant
measurements, etc. We just intended to demonstrate the principle. The clock
error is a by-product, but the desired result obtained from (5–9) are the GPS
coordinates XA , YA , ZA ; this means, the resulting coordinates are obatined
in the WGS 84.
As described in Sect. 5.3.4, the accuracy of the point positioning method
based on code ranges may be expected to amount some 10 m (nominally). A
much higher accuracy is achieved by relative positioning treated in the next
section.
5.4.2 Relative positioning with phase pseudoranges

The objective of relative positioning is to determine the coordinates of an
unknown point with respect to a known point. In other words, relative po-
sitioning aims at the determination of the vector between the two points
which is often called the baseline vector or simply baseline (Fig. 5.3). Let
now A denote the known reference point, B the unknown point, and bAB
the baseline vector. Introducing the corresponding position vectors XA , XB ,
k
l
j
%Ak (t) %Al (t)
j %A
m
(t) m
Z WGS-84 %A (t)
j
%Bk (t) %Bl (t)
%B (t)
%Bm(t)
B
A bAB
ZA ZB
Y WGS-84
YB XB
X WGS-84 XA
YA
Fig. 5.3. Relative positioning
the relation
XB = XA + bAB (5–10)
may be formulated, and the components of the baseline vector bAB are
⎡ ⎤ ⎡ ⎤
XB − XA ∆XAB
bAB = ⎣ YB − YA ⎦ = ⎣ ∆YAB ⎦ . (5–11)
ZB − ZA ∆ZAB
The coordinates of the reference point must be given in the WGS 84 and are
usually approximated by a code pseudorange solution. Relative positioning
can be performed with code pseudoranges (cf. Eq. (5–3)) or with phase
pseudoranges (cf. Eq. (5–4)). Subsequently, only phase pseudoranges are
explicitly considered. We repeat (5–4),
λΦ = + cδ + λN , (5–12)
where we have already explained the wavelength λ, the phase Φ, the distance
(which is the same as for the code pseudorange model), the speed of light
c, the receiver clock error δ, and the ambiguity N in Sect. 5.3.3.
Introducing f , the frequency of the corresponding satellite signal, and
taking into account the relation f = c/λ, we may divide (5–12) by λ obtain-
ing
1
Φ= +fδ+N. (5–13)
λ
This may be generalized to
1 j
Φji (t) = (t) + f δi (t) + Nij , (5–14)
λ i
where Φji (t) is the measured carrier phase expressed in cycles referred to
station i and satellite j at epoch t. The time-independent phase ambiguity
Nij is an integer number and, therefore, often called integer ambiguity or
integer unknown or simply ambiguity.
Relative positioning requires simultaneous observations at both the ref-
erence and the unknown point. This means that the observation time tags
for the two points must be the same. Assuming such observations (5–14) at
the two points A and B to satellite j and another satellite k simultaneously
at epoch t, the following measurement equations may be set up:
1 j
ΦjA (t) = (t) + f δA (t) + NAj ,
λ A
1
ΦkA (t) = kA (t) + f δA (t) + NAk ,
λ (5–15)
1
ΦB (t) = jB (t) + f δB (t) + NBj ,
j
λ
1
ΦkB (t) = kB (t) + f δB (t) + NBk .
λ
Introducing the short-hand notations
Φjk k j k j
AB (t) = ΦB (t) − ΦB (t) − ΦA (t) + ΦA (t) ,
jk k j k j
AB (t) = B (t) − B (t) − A (t) + A (t) ,
(5–16)
jk
NAB = NBk − NBj − NAk + NAj ,
we form the double-difference model which is defined as

1 jk
Φjk
AB (t) =
jk
(t) + NAB . (5–17)
λ AB
Note that the receiver clock biases have canceled; this is the reason why
double-differences are preferably used. This cancellation resulted from the
assumptions of simultaneous observations and equal frequencies of the satel-
lite signals (which is justified for GPS).
Assuming A as reference station with known coordinates, the remain-
ing unknowns of the double-difference model are the desired coordinates
XB , YB , ZB – which are comprised in jB (t) and kB (t) – and the ambiguities.
To solve for these unknowns, we need more satellites (to set up additional
double-differences) and also more epochs.
We do not consider linearization, possible redundant measurements, etc.
We just intended to demonstrate the principle. The desired result obtained
from (5–17) is the baseline vector bAB with the components ∆XAB , ∆YAB ,
∆ZAB or, finally, the GPS coordinates XB , YB , ZB derived from (5–10) via
the known station A to achieve the high accuracy. Note that the resulting
coordinates are obtained in the WGS 84.
This concludes the short introduction how the user of GPS gets WGS 84
coordinates, i.e., geocentric rectangular coordinates X, Y, Z or, computed
from them, ellipsoidal coordinates ϕ, λ, h; see Sect. 5.6.1.
5.5 Projection onto the ellipsoid

Let us establish the position of a point P by means of the natural coor-
dinates Φ, Λ, H. Then we may project it onto the geoid along the (slightly
curved) plumb line. The orthometric height is the distance between P and its
projection P0 onto the geoid, measured along the plumb line (Fig. 5.4). Al-
though this mode of projection is entirely natural, the geoid is not suited for
performing computations on it directly; the point P0 is, therefore, projected
onto the reference ellipsoid by means of the straight ellipsoidal normal, thus
getting a point Q0 on the ellipsoid. In this way, the earth’s surface point P
and the corresponding point Q0 on the ellipsoid are connected by a double
projection, that is, by two projections which are performed one after the
other and which are quite analogous, the orthometric height H = P P0 cor-
responding to the geoidal undulation N = P0 Q0 . This double projection is
called Pizzetti’s projection.
It is much simpler to project the point P from the physical surface of the
earth directly onto the ellipsoid through the straight ellipsoidal normal, thus
obtaining a point Q. The distance P Q = h is the ellipsoidal height, i.e., the
height above the ellipsoid. The earth’s surface point P is then determined
by the ellipsoidal height h and the ellipsoidal coordinates ϕ, λ of Q on the
ellipsoid so that the ellipsoidal coordinates ϕ, λ, h take the place of the natural
coordinates Φ, Λ, H. This is called Helmert’s projection.
" earth's
surface
±
hH
P0 geoid
N
Q Q0 ellipsoid
Fig. 5.4. The projection of Helmert and of Pizzetti

5.6 Coordinate transformations 193
The practical difference between Pizzetti’s and Helmert’s projection is

small. The ellipsoidal height h is equal to H + N within a fraction of a
millimeter. The ellipsoidal coordinates ϕ and λ, with respect to the two
projections, are related by the equations
H
ϕHelmert = ϕPizzetti + ξ,
R
(5–18)
H
λHelmert = λPizzetti + η sec ϕ ,
R
.
which can be read from Fig. 5.4, since QQ0 = H ε; R = 6371 km is the
mean radius of the earth. Even if ε = 1 arc minute and H = 1000 m, the
distance QQ0 is only about 30 cm and the ellipsoidal coordinates differ by
less than 0.01 , which is below the accuracy of astronomical observations.
For most purposes, we may, therefore, neglect the difference between the two
projections.
Pizzetti’s projection is better adapted to the geoid, because there is an
exact correspondence between a geoidal point P0 and an ellipsoidal point
Q0 . Helmert’s projection has overwhelming practical advantages, notably
the straightforward conversion of the ellipsoidal coordinates ϕ, λ, h into rect-
angular coordinates x, y, z; it is also simpler in other respects. The decisive
advantage of Helmert’s projection is its direct relation to GPS. It is, there-
fore, exclusively used now in practice.
5.6 Coordinate transformations

5.6.1 Ellipsoidal and rectangular coordinates
We now derive the relation between the ellipsoidal coordinates ϕ, λ, h and
the corresponding rectangular coordinates x, y, z.
The equation of the reference ellipsoid in rectangular coordinates is
x2 + y 2 z 2
+ 2 = 1. (5–19)
a2 b
The representation of this ellipsoid in terms of ellipsoidal coordinates is given
by
x = N cos ϕ cos λ ,
y = N cos ϕ sin λ , (5–20)

b2
z= N sin ϕ ,
a2
z
P
p
h
nQ
X Q
x
'
x,y plane
N
Fig. 5.5. Ellipsoidal and rectangular coordinates
where N is the normal radius of curvature (2–149):
a2
N= . (5–21)
a2 cos2 ϕ + b2 sin2 ϕ
These equations are known from ellipsoidal geometry; it may also be verified
by direct substitution that a point with xyz-coordinates (5–20) satisfies the
equation of the ellipsoid (5–19) and so lies on the ellipsoid. The components
of the unit normal vector n are

n = cos ϕ cos λ , cos ϕ sin λ , sin ϕ , (5–22)
because ϕ is the angle between the ellipsoidal normal and the xy-plane,
which is the equatorial plane (Fig. 5.5). Now let the coordinates of a point
P outside the ellipsoid form the vector
X = [X , Y , Z ] ; (5–23)
similarly we have, for the coordinates of the point Q on the ellipsoid,
x = [x , y , z ] . (5–24)
From Fig. 5.5, we read

X = x + hn, (5–25)
that is
X = x + h cos ϕ cos λ ,
Y = y + h cos ϕ sin λ , (5–26)
Z = z + h sin ϕ .
By (5–20), this becomes
X = (N + h) cos ϕ cos λ ,
Y = (N + h) cos ϕ sin λ ,
(5–27)
2
b
Z = N + h sin ϕ .
a2
These equations are the basic transformation formulas between the ellip-
soidal coordinates ϕ, λ, h and the rectangular coordinates X, Y, Z of a point
outside the ellipsoid. The origin of the rectangular coordinate system is the
center of the ellipsoid, and the z-axis is its axis of rotation; the x-axis has
the Greenwich longitude 0◦ and the y-axis has the longitude 90◦ east of
Greenwich (i.e., λ = +90◦ ).
A possible source of confusion is that the normal radius of curvature of
the ellipsoid and the geoidal undulation are both denoted by the symbol N ;
in (5–27), N is, of course, the normal radius of curvature. Generally, let the
context decide between quantities of such different magnitude (6000 km and
60 m).
Equations (5–27) permit the computation of rectangular coordinates
X, Y, Z from the ellipsoidal coordinates ϕ, λ, h.
The inverse procedure, the computation of ϕ, λ, h from given X, Y, Z, is
frequently performed iteratively, although a solution in closed form exists.
A possible iterative
√ procedure is as follows.
Denoting X + Y 2 by p, we get from the first two equations of (5–27)
2
or from Fig. 5.5

p = X 2 + Y 2 = (N + h) cos ϕ , (5–28)
so that
p
h= −N. (5–29)
cos ϕ
The third equation of (5–27) may be transformed into

a2 − b2
Z= N− N + h sin ϕ = (N + h − e2 N ) sin ϕ , (5–30)
a2
where e2 = (a2 − b2 )/a2 . Dividing this equation by the above expression for
p, we find
Z 2 N
= 1−e tan ϕ , (5–31)
p N +h
so that −1
Z N
tan ϕ = 1 − e2 . (5–32)
p N +h
Given X, Y, Z, and hence p, Eqs. (5–29) and (5–32) may be solved iteratively
for h and ϕ. As a first approximation, we set h = 0 in (5–32), obtaining
Z
tan ϕ(1) = (1 − e2 )−1 . (5–33)
p
Using ϕ(1) , we compute an approximate value N(1) by means of (5–21). Then

(5–29) gives h(1) . Now, as a second approximation, we set h = h(1) in (5–32),
obtaining
−1
Z N(1)
tan ϕ(2) = 1−e 2
. (5–34)
p N(1) + h(1)
Using ϕ(2) , improved values for N and h are found, etc. This procedure is
repeated until ϕ and h remain practically constant.
The result for λ is immediately obtained from the first two equations of
(5–27):
Y
λ = arctan . (5–35)
X
Many other computation methods have been devised. One example for
the transformation of X, Y, Z into ϕ, λ, h without iteration but with an
inherent approximation is
Z + e2 b sin3 θ
ϕ = arctan ,
p − e2 a cos3 θ
Y (5–36)
λ = arctan ,
X
p
h= −N,
cos ϕ
where
Za
θ = arctan (5–37)
pb
is an auxiliary quantity and
e2 = (a2 − b2 )/a2 , e2 = (a2 − b2 )/b2 (5–38)

are
√ first and second numerical eccentricity. As introduced in (5–28), p =
X 2 + Y 2 . Actually, there is no reason why these formulas are less popular
than the iterative procedure since there is no significant difference between
the two methods. Computation methods with neither iteration nor approx-
imation are, e.g., given by Sünkel (1977) and Zhu (1993).
5.6.2 Ellipsoidal, ellipsoidal-harmonic, and spherical

coordinates
Even if we have several times pointed out the different definitions, it is
very important to stress once more the need not to confuse the following
coordinate triples (see Fig. 5.6):
• ellipsoidal coordinates: ϕ, λ, h;
• ellipsoidal-harmonic coordinates: β, λ, u,
alternatively: ϑellipsoidal-harmonic , λ, u;
• spherical coordinates: ϕ̄, λ, r, alternatively: ϑspherical , λ, r.
The longitude λ is the same in all triples. The ellipsoidal coordinates latitude
ϕ and longitude λ are sometimes also denoted geodetic latitude and geodetic
z
2 2
radius u + E
ere of
sp h
hP
throug P
oid
l lips
le
a soid
oc ellip h
nf c e #ell-har.
ren Q
co
efe
#sph. r
r
¯
F1 ' ' F2
x,y plane
E = a2 – b 2 u2 + E 2
Fig. 5.6. Ellipsoidal, ellipsoidal-harmonic, and spherical coordinates

longitude. The ellipsoidal-harmonic coordinate β is the reduced latitude, and

the spherical coordinate ϕ̄ is the geocentric latitude.
The latitude ϕ refers to the reference ellipsoid. The reduced latitude β
refers to the coordinate ellipsoid u = constant (confocal ellipsoid through P
in Fig. 5.6).
So far so clear. Real attention is necessary when using the coordinate ϑ,
which has been introduced as complement of the spherical coordinate ϕ̄ and
as the complement of the ellipsoidal harmonic β as well.
Therefore, a correct but clumsy notation would be
ϑellipsoidal-harmonic = 90◦ − β ,
(5–39)
ϑspherical = 90◦ − ϕ̄ .
Note, however, that we did not use these indications to distinguish be-
tween the spherical and the ellipsoidal-harmonic ϑ! Thus, the reader is chal-
lenged to attentively distinguish between these quantities. Wherever possi-
ble, we tried to avoid conflicts.
Some examples: we used the spherical coordinates r, ϑ, λ in Sects. 1.4,
1.11, 1.12, 1.14, 2.5, 2.6, 2.13, 2.18, etc. We used the ellipsoidal-harmonic
coordinates u, ϑ, λ in Sects. 1.15, 1.16; we used the ellipsoidal-harmonic co-
ordinates u, β, λ in Sects. 2.7, 2.8, and we used the spherical coordinates
r, ϑ, λ as well as the ellipsoidal-harmonic coordinates u, β, λ in Sect. 2.9.
The following equations express the rectangular coordinates in these
three systems:
√
X = (N + h) cos ϕ cos λ = u2 + E 2 cos β cos λ = r cos ϕ̄ cos λ ,
√
Y = (N + h) cos ϕ sin λ = u2 + E 2 cos β sin λ = r cos ϕ̄ sin λ ,
(5–40)
2
b
Z = N + h sin ϕ = u sin β = r sin ϕ̄ .
a2
These relations, which follow from combining Eqs. (1–26), (1–151), and (5–
27), can be used if we wish to compute u and β from h and ϕ or from r and
ϕ̄, etc.
5.7 Geodetic datum transformations

5.7.1 Introduction
First we define a geodetic datum or a geodetic reference system. It is de-
fined by (1) the dimensions of the reference ellipsoid (semimajor axis a and
5.7 Geodetic datum transformations 199
flattening f ) and (2) its position with respect to the earth or the geoid.
This relative position is most simply defined by the coordinates x0 , y0 , z0
of the center of the reference ellipsoid with respect to the geocenter. Since
the geocenter was not accessible to classical geodetic measurements before
the satellite era, a fundamental or initial point P1 on the earth surface was
chosen, such as Meades Ranch for North America and Potsdam for Central
Europe. It turns out that a convenient but conventional choice of the el-
lipsoidal coordinates ϕ1 , λ1 , h1 of the fundamental point P1 is equivalent to
x0 , y0 , z0 of the geocenter.
Thus, we have 5 defining parameters:
• 2 parameters a (semimajor axis) and f (flattening) as form parameters,
and
• 3 parameters x0 , y0 , z0 or ϕ1 , λ1 , h1 as position parameters.
Later on we shall also admit a scale factor and small rotations around the
three coordinate axes.
A (geodetic) datum transformation defines the relationship between a
global (geocentric) and a local (in general nongeocentric) three-dimensional
Cartesian coordinate system; therefore, a datum transformation transforms
one coordinate system of a certain type to another coordinate system of the
same type. This is one of the primary tasks when combining GPS data with
terrestrial data, i.e., the transformation of geocentric WGS 84 coordinates
to local terrestrial coordinates. The terrestrial system is usually based on a
locally best-fitting ellipsoid, e.g., the Clarke ellipsoid or the GRS-80 ellipsoid
in the U.S. and the Bessel ellipsoid in many parts of Europe. The local
ellipsoid is linked to a nongeocentric Cartesian coordinate system, where the
origin coincides with the center of the ellipsoid.
5.7.2 Three-dimensional transformation in general form

Consider two arbitrary sets of three-dimensional Cartesian coordinates form-
ing the vectors X and XT (Fig. 5.7). The 7-parameter transformation, also
denoted as Helmert transformation or similarity transformation in space,
between the two sets can be formulated by the relation
XT = x0 + µ R X , (5–41)
where x0 is the translation (or shift) vector, µ is a scale factor, and R is a
rotation matrix.
The components of the shift vector
⎡ ⎤
x0
x0 = ⎣ y0 ⎦ (5–42)
z0
ZT
Z
"3
XT X Y
"2
x0
YT
"1
X
XT
Fig. 5.7. Three-dimensional transformation
account for the coordinates of the origin of the X system in the XT system.
Note that a single scale factor is considered. More generally (but with GPS
not necessary), three scale factors, one for each axis, could be used. The
rotation matrix is an orthogonal matrix which is composed of three successive
rotations
R = R 3 {ε3 } R2 {ε2 } R1 {ε1 } . (5–43)
Explicitly,
⎡ ⎤
cos ε2 cos ε3 cos ε1 sin ε3 sin ε1 sin ε3
⎢ + sin ε1 sin ε2 cos ε3 − cos ε1 sin ε2 cos ε3 ⎥
⎢ ⎥
⎢ ⎥
R=⎢
⎢− cos ε2 sin ε3 cos ε1 cos ε3 sin ε1 cos ε3 ⎥
⎥
⎢ − sin ε1 sin ε2 sin ε3 + cos ε1 sin ε2 sin ε3 ⎥
⎣ ⎦
sin ε2 − sin ε1 cos ε2 cos ε1 cos ε2
(5–44)
is obtained.
In the case of known transformation parameters x0 , µ, R, a point from
the X system can be transformed into the XT system by (5–41).
If the transformation parameters are unknown, they can be determined
with the aid of common (identical) points, also denoted as control points.
This means that the coordinates of the same point are given in both systems.
Since each common point (given by XT and X) yields three equations, two
common points and one additional common component (e.g., height) are
sufficient to solve for the seven unknown parameters. In practice, redun-
dant common point information is used and the unknown parameters are
calculated by least-squares adjustment.
Since the parameters are mixed nonlinearly in Eq. (5–41), a linearization
must be performed, where approximate values x 0 approx , µapprox , R approx are
required.
5.7.3 Three-dimensional transformation between WGS 84

and a local system
In the case of a datum transformation between WGS 84 and a local sys-
tem, some simplifications will arise. Referring to the necessary approximate
values, the approximation µapprox = 1 is appropriate and the relation
µ = µapprox + δµ = 1 + δµ (5–45)
is obtained. Furthermore, the rotation angles εi in (5–44) are small and may
be treated as differential quantities. Introducing these quantities into (5–44),
setting cos εi = 1 and sin εi = εi , and considering only first-order terms gives
⎡ ⎤
1 ε3 −ε2
R = ⎣ −ε3 1 ε1 ⎦ = I + δR , (5–46)
ε2 −ε1 1
where I is the unit matrix and δR is a (skewsymmetric) differential rotation

matrix. Thus, the approximation R approx = I is appropriate. Finally, the
shift vector is split up in the form
x0 = x 0 approx + δx0 , (5–47)
where the approximate shift vector
x 0 approx = X T − X (5–48)
follows by substituting the approximations for the scale factor and the rota-
tion matrix into Eq. (5–41).
Introducing Eqs. (5–45), (5–46), (5–47) into (5–41) and skipping de-
tails which can be found, for example, in Hofmann-Wellenhof et al. (1994:
Sect. 3.3) gives the linearized model for a single point i. This model can be
written in the form
X Ti − X i − x 0 approx = A i δp , (5–49)
where the left side of the equation is known and may formally be considered
as an observation. The design matrix A i and the vector δp, containing the
unknown parameters, are given by
⎡ ⎤
1 0 0 Xi 0 −Zi Yi
Ai = ⎣ 0 1 0 Yi Zi 0 −Xi ⎦ ,
0 0 1 Zi −Yi Xi 0 (5–50)
δp = [ δx0 δy0 δz0 δµ ε1 ε2 ε3 ] .

Recall that Eq. (5–49) is now a system of linear equations for point i. For n
common points, the design matrix A is
⎡ ⎤
A1
⎢ A2 ⎥
⎢ ⎥
A=⎢ . ⎥. (5–51)
⎣ .. ⎦
An
In detail, for three common points the design matrix is

⎡ ⎤
1 0 0 X1 0 −Z1 Y1
⎢ 0 1 0 Y1 Z 0 −X ⎥
⎢ 1 1 ⎥
⎢ 0 0 1 Z1 −Y1 X 0 ⎥
⎢ 1 ⎥
⎢ ⎥
⎢ 1 0 0 X2 0 −Z2 Y2 ⎥
⎢ ⎥
⎢
A = ⎢ 0 1 0 Y2 Z2 0 −X2 ⎥, (5–52)
⎥
⎢ 0 0 1 Z2 −Y2 X 0 ⎥
⎢ 2 ⎥
⎢ ⎥
⎢ 1 0 0 X3 0 −Z3 Y3 ⎥
⎢ ⎥
⎣ 0 1 0 Y3 Z3 0 −X3 ⎦
0 0 1 Z3 −Y3 X3 0
which leads to a slightly redundant system. Least-squares adjustment yields

the parameter vector δp and the adjusted values by (5–45), (5–46), (5–47).
Once the seven parameters of the similarity transformation are determined,
formula (5–41) can be used to transform other than the common points.
For a specific example, consider the task of transforming GPS coordi-
nates of a network, i.e., global geocentric WGS 84 coordinates, to (three-
dimensional) coordinates of a (nongeocentric) local system indicated by the
subscript LS. The GPS coordinates are denoted by (X, Y, Z)GPS and the lo-
cal system coordinates are the plane coordinates (y, x)LS and the ellipsoidal
height hLS . To obtain the transformation parameters, it is assumed that the
coordinates of the common points in both systems are available. The solution
of the task is obtained by the following algorithm:
1. Transform the plane coordinates (y, x)LS of the common points into
the ellipsoidal surface coordinates (ϕ, λ)LS by using the appropriate
mapping formulas.
2. Transform the ellipsoidal coordinates (ϕ, λ, h)LS of the common points

into the Cartesian coordinates (X, Y, Z)LS by (5–27).
3. Determine the seven parameters of a Helmert transformation by using

the coordinates (X, Y, Z)GPS and (X, Y, Z)LS of the common points.
4. For network points other than the common points, transform the co-
ordinates (X, Y, Z)GPS into (X, Y, Z)LS via Eq. (5–41) using the trans-
formation parameters determined in the previous step.
5. Transform the Cartesian coordinates (X, Y, Z)LS computed in the pre-

vious step into ellipsoidal coordinates (ϕ, λ, h)LS , e.g., by the iterative
procedure given in (5–28) through (5–34).
6. Map the ellipsoidal surface coordinates (ϕ, λ)LS computed in the pre-
vious step into plane coordinates (y, x)LS by the appropriate mapping
formulas.
The advantage of the three-dimensional approach is that no a priori infor-

mation is required for the seven parameters of the similarity transformation.
The disadvantage of the method is that for the common points ellipsoidal
heights (and, thus, geoidal heights) are required. However, as reported by
Schmitt et al. (1991), incorrect heights of the common points often have
a negligible effect on the plane coordinates (y, x). For example, incorrect
heights may cause a tilt of a 20 km × 20 km network by an amount of 5 m
in space; however, the effect on the plane coordinates is only approximately
1 mm.
For large areas, the height problem can be solved by adopting approx-
imate ellipsoidal heights for the common points and performing a three-
dimensional affine transformation instead of the similarity transformation.
5.7.4 Differential formulas for other datum transformations

Now we consider simplified cases. Suppose that the geocenter does not co-
incide with the center of the reference ellipsoid, but that the geocentric axes
and the ellipsoidal axes are parallel. Such a parallel shift is also called a
translation (Fig. 5.8). Assume a rectangular coordinate system XY Z whose
origin is the geocenter, the axes being directed as usual. Let the coordinates
of the center of the ellipsoid with respect to this system be x0 , y0 , z0 , as
stated previously. Then Eqs. (5–27) must obviously be modified so that they
become
X = x0 + (N + h) cos ϕ cos λ ,
Y = y0 + (N + h) cos ϕ sin λ ,
(5–53)
2
b
Z = z0 + N + h sin ϕ .
a2
These equations form the starting point for various important differential
formulas of coordinate transformation.
b
geocenter
Y
x0 x0
z0 a
y0 ellipsoid center
X
Fig. 5.8. Translation problem
First we ask how the rectangular coordinates X, Y, Z change if we vary

the ellipsoidal coordinates ϕ, λ, h by small amounts δϕ, δλ, δh and if we also
alter the geodetic datum, namely, the reference ellipsoid a, f and its position
x0 , y0 , z0 , by δa, δf and δx0 , δy0 , δz0 . Note that δx0 , δy0 , δz0 correspond to a
small translation (parallel displacement) of the ellipsoid, its axis remaining
parallel to the axis of the earth.
The solution of this problem is found by differentiating (5–53):
∂X ∂X ∂X ∂X ∂X
δX = δx0 + δa + δf + δϕ + δλ + δh ,
∂a ∂f ∂ϕ ∂λ ∂h
∂Y ∂Y ∂Y ∂Y ∂Y
δY = δy0 + δa + δf + δϕ + δλ + δh , (5–54)
∂a ∂f ∂ϕ ∂λ ∂h
∂Z ∂Z ∂Z ∂Z ∂Z
δZ = δz0 + δa + δf + δϕ + δλ + δh ,
∂a ∂f ∂ϕ ∂λ ∂h
since, according to Taylor’s theorem, small changes can be treated as differ-

entials.
In these differential formulas we shall be satisfied with an approximation.
Since the flattening f is small, we may expand (2–149) as

a2 a2

N= (1 + e2 cos2 ϕ)−1/2 = 1 − 12 e2 cos2 ϕ · · ·
b b (5–55)
= a (1 + f · · ·)(1 − f cos2 ϕ · · ·) = a(1 + f − f cos2 ϕ · · ·)
yielding
.
N = a (1 + f sin2 ϕ) (5–56)
and
b2 .
N = (1 − 2f · · ·) a (1 + f sin2 ϕ · · ·) = a (1 − 2f + f sin2 ϕ) (5–57)
a2
and
b = a (1 − f ) , e2 = 2f · · · . (5–58)
Thus, Eqs. (5–53) are approximated by
X = x0 + (a + af sin2 ϕ + h) cos ϕ cos λ ,
Y = y0 + (a + af sin2 ϕ + h) cos ϕ sin λ , (5–59)
Z = z0 + (a − 2af + af sin2 ϕ + h) sin ϕ .
Now we can form the partial derivatives in (5–54), for instance,

∂X .
= (1 + f sin2 ϕ) cos ϕ cos λ = cos ϕ cos λ , (5–60)
∂a
since we may neglect the flattening in these coefficients. This amounts to
using for the coefficients, and only for them, a spherical approximation anal-
ogous to that of Sect. 2.13. Similarly, all coefficients are easily obtained as
partial derivatives, and Eqs. (5–54) become
δX = δx0 − a sin ϕ cos λ δϕ − a cos ϕ sin λ δλ

+ cos ϕ cos λ (δh + δa + a sin2 ϕ δf ) ,
δY = δy0 − a sin ϕ sin λ δϕ + a cos ϕ cos λ δλ

(5–61)
+ cos ϕ sin λ (δh + δa + a sin2 ϕ δf ) ,
δZ = δz0 + a cos ϕ δϕ + sin ϕ (δh + δa + a sin2 ϕ δf )

− 2a sin ϕ δf .
These formulas give the changes in the rectangular coordinates X, Y, Z in
terms of the variation in the position (x0 , y0 , z0 ) and the dimensions (a, f )
of the ellipsoid and in the ellipsoidal coordinates ϕ, λ, h referred to it.
Transformation of the ellipsoidal coordinates

Several important formulas for the transformation of coordinates may be
derived from Eqs. (5–61). First, let the position of P in space remain un-
changed; that is, let
δX = δY = δZ = 0 . (5–62)
Determine the change of the ellipsoidal coordinates ϕ, λ, h if the dimensions
of the reference ellipsoid and its position are varied. Geometrically, this is
illustrated by Fig. 5.9. The problem is, thus, to solve equations (5–61) for
δϕ, δλ, δh, the left-hand sides being set equal to zero. To get δϕ, multiply the
first equation of (5–61) by − sin ϕ cos λ, the second equation of (5–61) by
− sin ϕ sin λ, and the third equation of (5–61) by cos ϕ and add all equations
obtained in this way. For δλ, the factors are − sin λ, cos λ, and 0; for δh, they
are cos ϕ cos λ, cos ϕ sin λ, and sin ϕ. The result is
a δϕ = sin ϕ cos λ δx0 + sin ϕ sin λ δy0 − cos ϕ δz0 + 2a sin ϕ cos ϕ δf ,
a cos ϕ δλ = sin λ δx0 − cos λ δy0 ,
δh = − cos ϕ cos λ δx0 − cos ϕ sin λ δy0 − sin ϕ δz0 − δa + a sin2 ϕ δf .

(5–63)
These formulas express the variations δϕ, δλ, δh at an arbitrary point in
terms of the variations δx0 , δy0 , δz0 at a given point and the changes δa and
δf of the parameters of the reference ellipsoid. Thus, they relate two different
systems of ellipsoidal coordinates, provided these systems are so close to each
other that their differences may be considered as linear. Mathematically,
Eqs. (5–63) are infinitesimal coordinate transformations (essentially but not
exclusively orthogonal transformations); to the geodesist, they give the effect
P
E2
Z
h + ±h
E1 h
' + ±'
±x0 a + ±a
'
a X
Y
Fig. 5.9. A small change of the reference ellipsoid together with a small
parallel shift
5.8 The three-dimensional geodesy of Bruns and Hotine 207
of a change in the geodetic datum.

Remark. The differential formulas could also be replaced by a successive
application of the original finite formulas. Try!
Part II: Three-dimensional geodesy: a transition
5.8 The three-dimensional geodesy of Bruns and

Hotine
The idea of a computation of a triangulation network in space dates back to
Bruns (1878). On the basis of his ideas, Hotine (1969), and earlier in 1959, de-
veloped extensively the concept of a classical (pre-satellite) geodetic network
in a rigorous three-dimensional way. For a comparison, see Levallois (1963).
Consider the polyhedron formed by triangulation benchmarks on the sur-
face of the earth and the straight lines of sight connecting them (Fig. 5.10).
Another set of straight lines – one through each corner – represents the
plumb line at the stations.
In order to determine this figure, we need five parameters for each station
– three coordinates and two parameters defining the direction of the plumb
line. The main terrestrial observational data for this purpose are
1. horizontal angles and zenith angles, obtained by theodolite observa-

tions;
2. straight spatial distances, obtained by electronic distance measure-

ments; and
3. astronomical observations of latitude and longitude to fix the direction

of the plumb line, and of azimuth to determine the orientation of the
polyhedron.
Fig. 5.10. Bruns’ polyhedron

We may use a rectangular coordinate system; then the three coordinates

to be determined will be X, Y, Z. The parameters defining the direction of
the plumb line are conveniently taken to be Φ and Λ, astronomical latitude
and longitude. We can express the astronomical azimuth A, the measured
zenith angle z , and the spatial distance s in terms of these five parameters.
This will be the scope of Sect. 5.9.
This information is purely “geometric”. We need the terrestrial measure-
ments (especially Φ, Λ, A) in order to link this geometry to the gravity field
as represented by the plumb lines. The Bruns polyhedron is the best way to
show this geometrically.
Today, GPS is the best way to determine global rectangular coordinates
X, Y, Z or ellipsoidal coordinates ϕ, λ, h directly.
5.9 Global coordinates and local level coordinates

We shall use a Cartesian coordinate system XY Z introduced in Sect. 5.6.1,
global but not necessarily geocentric. The coordinates X, Y, Z form a vector
X. Thus, the vectors X i and X j represent two terrestrial points Pi and
Pj . We define the vector between these two points in the global coordinate
system by X ij = X j − X i .
In addition, we introduce a “local level system” referred to the tangential
plane to the level surface at a point Pi and to the local vertical, which is
the tangent at Pi to the natural plumb line defined by the astronomical
coordinates Φ and Λ, see Sect. 2.4. The axes n i , e i , u i of this local (tangent
plane) coordinate system at Pi corresponding to the north, east, and up
ui
ni
. ei
Pi
Xi
Fi Y
Li
Fig. 5.11. Global and local level coordinates

5.9 Global coordinates and local level coordinates 209
ui (up, zenith)
Xij Pj
z'ij sij uij

Pi
ni
Aij
eij (north)
nij
ei (east)
Fig. 5.12. Measurement quantities in the local level system
direction, are thus represented in the global system by

⎡ ⎤ ⎡ ⎤ ⎡ ⎤
− sin Φi cos Λi − sin Λi cos Φi cos Λi
n i = ⎣ − sin Φi sin Λi ⎦ , e i = ⎣ cos Λi ⎦ , u i = ⎣ cos Φi sin Λi ⎦ ,
cos Φi 0 sin Φi
(5–64)
where the vectors n i and e i span the tangent plane at Pi (Fig. 5.11). The
third coordinate axis of the local level system, i.e., the vector u i , is orthog-
onal to the tangent plane and has the direction of the plumb line.
Now the components nij , eij , uij of the vector x ij in the local level system
are introduced. These coordinates are sometimes denoted as ENU (east,
north, up) coordinates. Considering Fig. 5.12, these components are obtained
by a projection of vector X ij onto the local level axes n i , e i , u i . Analytically,
this is achieved by scalar products. Therefore,
⎡ ⎤ ⎡ ⎤
nij n i · X ij
x ij = ⎣ eij ⎦ = ⎣ e i · X ij ⎦ (5–65)
uij u i · X ij
is obtained. Assembling the vectors n i , e i , u i of the local level system as

columns in an orthogonal matrix D i , i.e.,
⎡ ⎤
− sin Φi cos Λi − sin Λi cos Φi cos Λi
D i = ⎣ − sin Φi sin Λi cos Λi cos Φi sin Λi ⎦ , (5–66)
cos Φi 0 sin Φi
relation (5–65) may be written concisely as
x ij = DTi X ij . (5–67)
The components of x ij may also be expressed by the spatial distance sij ,

the azimuth Aij , and the zenith angle zij , which is assumed to be corrected
for refraction. The appropriate relation is

⎡ ⎤
⎡ ⎤ sij sin zij cos A
ij
nij ⎢ ⎥
x ij = eij ⎦ = ⎢
⎣
⎣ s ij sin z sin A
ij ij
⎥,
⎦ (5–68)
uij s cos z
ij ij
refer to P , i.e.,
where the terrestrial measurement quantities sij , Aij , zij i
the measurements were taken at Pi . Inverting (5–68) gives the measurement
quantities explicitly:

sij = n2ij + e2ij + u2ij ,
eij
tan Aij = ,
nij (5–69)
= uij
cos zij .
n2ij + e2ij + u2ij
Substituting (5–65) for nij , eij and uij , the measurement quantities may be
expressed by the components of the vector X ij in the global system.
A note on azimuth and zenith distance

Since the local level coordinates refer to the local plumb line defined by the
astronomical coordinates Φ, Λ (Sect. 2.4), A and z are called astronomical
azimuth and astronomical zenith distance (zenith angle). They will also play
a basic role in Part III.
A final word on the zenith distance. The measured (“astronomical”) az-
imuth is denoted by A, and the corresponding ellipsoidal azimuth is denoted
by α. Since the ellipsoidal zenith distance is conventionally denoted by z, it
would be consistent to indicate the measured (“astronomical”) zenith dis-
tance by Z. This symbol, however, is firmly reserved for the third axis of the
XY Z system, so we exceptionally, but consistently with the rest of the book,
use the symbol z . (Both A and z will return in the following sections.)
5.10 Combining terrestrial data and GPS

5.10.1 Common coordinate system
So far, GPS and terrestrial networks have been considered separately with
respect to the adjustment. The combination, for example, by a datum trans-
formation, was supposed to be performed after individual adjustments. Now
5.10 Combining terrestrial data and GPS 211
the common adjustment of GPS observations and terrestrial data is inves-

tigated. The problem encountered here is that GPS data refer to the three-
dimensional geocentric Cartesian system WGS 84, whereas terrestrial data
refer to the individual local level (tangent plane) systems at each measure-
ment point referenced to plumb lines. Furthermore, terrestrial data are tra-
ditionally separated into position and height, where the position refers to an
ellipsoid and the (orthometric) height to the geoid.
For a joint adjustment, a common coordinate system is required to which
all observations are transformed. In principle, any arbitrary system may be
introduced as common reference. One possibility is to use two-dimensional
(plane) coordinates in the local system as proposed by Daxinger and Stir-
ling (1995). Here, a three-dimensional coordinate system is chosen. The ori-
gin of the coordinate system is the center of the ellipsoid adopted for the
local system, the Z-axis coincides with the semiminor axis of the ellipsoid,
the X-axis is obtained by the intersection of the ellipsoidal Greenwich merid-
ian plane and the ellipsoidal equatorial plane, and the Y -axis completes the
right-handed system. Position vectors referred to this system are denoted by
X LS , where LS indicates the reference to the local system.
After the decision on the common coordinate system, the terrestrial mea-
surements referring to the individual local level systems at the observing
sites must be represented in this common coordinate system. Similarly, GPS
baseline vectors regarded as measurement quantities are to be transformed
to this system.
5.10.2 Representation of measurement quantities

Distances
The spatial distance sij as function of the local level coordinates is given in
(5–69). If nij , eij , uij , the components of x ij , are substituted by (5–65), the
relation
sij = n2ij + e2ij + u2ij
(5–70)
= (Xj − Xi ) + (Yj − Yi ) + (Zj − Zi )
2 2 2
is obtained, where (5–64) has also been taken into account, namely, the
fact that n i , e i , u i are unit vectors. Obviously, the second expression arises
immediately from the Pythagorean theorem. Differentiation of (5–70) yields
Xij Yij Zij

dsij = (dXj − dXi ) + (dYj − dYi ) + (dZj − dZi ) ,
sij sij sij
(5–71)
where
Xij = Xj − Xi ,
Yij = Yj − Yi , (5–72)
Zij = Zj − Zi
have been introduced accordingly. The relation (5–71) may also be expressed
as
Xij Yij Zij
δsij = (δXj − δXi ) + (δYj − δYi ) + (δZj − δZi )
sij sij sij
(5–73)
if the differentials are replaced by differences.
Azimuths
Again the same principle applies: the measured azimuth Aij as a function of
the local level coordinates is given in (5–69). If nij , eij , uij , the components
of x ij , are substituted by (5–65), the relation
tan Aij = eij /nij

−Xij sin Λi + Yij cos Λi (5–74)
=
−Xij sin Φi cos Λi − Yij sin Φi sin Λi + Zij cos Φi
is obtained. After a lengthy derivation, the relation
sin ϕi cos λi sin αij − sin λi cos αij
δAij = (δXj − δXi )
sij sin zij
sin ϕi sin λi sin αij + cos λi cos αij
+ (δYj − δYi )
sij sin zij
cos ϕi sin αij (5–75)
− (δZj − δZi )
sij sin zij
+ cot zij sin αij δΦi
+ (sin ϕi − cos αij cos ϕi cot zij ) δΛi

is obtained. Approximate values are sufficient in the coefficients, denoted by
ϕ, λ, α, z instead of Φ, Λ, A, z .
Directions
Measured directions Rij are related to azimuths Aij by the orientation un-
known oi . The relation reads
Rij = Aij − oi , (5–76)

and the expression

δRij = δAij − δoi (5–77)
is immediately obtained.
Zenith angles
The zenith angle zij as function of the local level coordinates is given in
(5–69). If nij , eij , uij , the components of x ij , are substituted by (5–65), the
relation
= u /s
cos zij ij ij
Xij cos Φi cos Λi + Yij cos Φi sin Λi + Zij sin Φi (5–78)

=
2 + Y 2 + Z2
Xij ij ij
is obtained, where (5–70) and (5–72) have been used. After a lengthy deriva-
tion, the relation
= Xij cos zij − sij cos ϕi cos λi

δzij (δXj − δXi )
s2ij sin zij
Yij cos zij − sij cos ϕi sin λi
+ (δYj − δYi )
s2ij sin zij (5–79)
Zij cos zij − sij sin ϕi
+ (δZj − δZi )
s2ij sin zij
− cos αij δΦi − cos ϕi sin αij δΛi
is obtained.
It is presupposed that the zenith angles are reduced to the chord of the
light path. This reduction may be modeled by
sij
zij = zij meas
+ k, (5–80)
2R
where zijmeas is the measured zenith angle, R is the mean radius of the
earth, and k is the coefficient of refraction. For k either a standard value
may be substituted or the coefficient of refraction is estimated as additional
unknown. In the case of estimation, there are several choices, e.g., one value
for k for all zenith angles or one value for a group of zenith angles or one
value per day. (It is known that measured zenith angles are “weaker” than
other observations, which can be taken into account by giving them lower
weights.)
Ellipsoidal height differences

The “measured” ellipsoidal height difference is represented by
hij = hj − hi . (5–81)
The heights involved are obtained by transforming the Cartesian coordinates
into ellipsoidal coordinates according to (5–36) or by using the iterative pro-
cedure given in Sect. 5.6.1. The height difference is approximately (neglecting
the curvature of the earth) given by the third component of x ij in the local
level system. Hence,
hij = u i · X ij (5–82)
or, by substituting u i according to (5–64), the relation
hij = cos Φi cos Λi Xij + cos Φi sin Λi Yij + sin Φi Zij (5–83)
is obtained. This equation may be differentiated with respect to the Carte-
sian coordinates. If the differentials are replaced by the corresponding dif-
ferences,
δhij = cos Φj cos Λj δXj + cos Φj sin Λj δYj + sin Φj δZj
(5–84)
− cos Φi cos Λi δXi − cos Φi sin Λi δYi − sin Φi δZi
is obtained, where the coordinate differences were decomposed into their
individual coordinates.
Baselines
From relative GPS measurements, baselines X ij(GPS) = X j(GPS) − X i(GPS) in
the WGS 84 are obtained. The position vectors X i(GPS) and X j(GPS) may be
transformed by a three-dimensional (7-parameter) similarity transformation
to a local system indicated by LS. According to Eq. (5–41), the transforma-
tion formula reads
X LS = x0 + µ R X GPS , (5–85)
where the meaning of the individual quantities is the following:
X LS ... position vector in the local system ,
X GPS ... position vector in the WGS 84 ,
x0 ... shift vector ,
R ... rotation matrix ,
µ ... scale factor .
Forming the difference of two position vectors, i.e., the baseline X ij , the
shift vector x0 is eliminated. Using (5–85), there results
X ij(LS) = µ R X ij(GPS) (5–86)
for the baseline. Similar to (5–49), the linearized form is
X ij(LS) = X ij(GPS) + A ij δp , (5–87)
where now the vector δp and the design matrix A ij are given by
δp = [δµ ε1 ε2 ε3 ]T ,
⎡ ⎤
Xij 0 −Zij Yij
(5–88)
⎢ ⎥
A ij = ⎣ Yij Zij 0 −Xij ⎦ .
Zij −Yij Xij 0 (GPS)
Note that the rotations εi refer to the axes of the system used in GPS. If
they should refer to the local system, then the signs of the rotations must be
changed, i.e., the signs of the elements of the last three columns of matrix
A ij must be reversed.
The vector X ij(LS) on the left side of (5–87) contains the points X i(LS)
and X j(LS) in the local system. If these points are unknown, then they are
replaced by known approximate values and unknown increments
X i(LS) = X i0(LS) + δX i(LS) ,

(5–89)
X j(LS) = X j0(LS) + δX j(LS) ,
where the coefficients of these unknown increments (+1 or −1) together with
matrix A ij form the design matrix.
The vector X ij(GPS) in (5–87) is regarded as measurement quantity. Thus,
finally,
X ij(GPS) = δX j(LS) − δX i(LS) − A ij δp + X j0(LS) − X i0(LS) (5–90)
is the linearized observation equation.

In principle, any type of geodetic measurement can be employed if the
integrated geodesy adjustment model is used. The basic concept is that any
geodetic measurement can be expressed as a function of one or more posi-
tion vectors X and of the gravity field W of the earth. The usually non-
linear function must be linearized where the gravity field W is split into
the normal potential U of an ellipsoid and the disturbing potential T , thus,
W = U + T . Applying a minimum principle leads to the collocation formulas
(Moritz 1980 a: Chap. 11).
Many examples integrating GPS and other data can be found in technical
publications. For example, there are attempts to detect earth deformations
from GPS and terrestrial data.
Part III: Local geodetic datums
5.11 Formulation of the problem

As we have remarked several times, the weak point of the Bruns–Hotine
method is the insufficient accuracy of the zenith angle measurement preclud-
ing the practical use of this method for larger triangulations. The trigono-
metric heights obtained in this way are significantly less accurate than the
horizontal positions.
A practical solution of this problem was to separate positions and heights.
The horizontal position was calculated on the reference ellipsoid in the way
we shall see later. Accurate heights were obtained by leveling referred to the
“actual” level surfaces, in particular to the geoid.
Thus, this theoretically and practically unsatisfactory procedure used
two different reference surfaces: the ellipsoid for horizontal position and the
geoid for heights. The mutual position of these two surfaces was not even
known because of lack of knowledge of the geoidal height N . It has been
rightfully ridiculed as “2+1-dimensional geodesy”.
There is a way out of this dilemma even for local (or rather regional)
geodetic systems. The trigonometric height h is not determined by zenith-
angle measurements but by using the simple formula
h=H +N (5–91)
from leveled orthometric heights H by adding the geoid height N !

But how do we get the geoid? Even before the satellite era, there existed
two methods:
1. the astrogeodetic method, determining N from deflections of the vertical

ξ and η;
2. the gravimetric method, using for this purpose gravity anomalies ∆g.
The theories of both methods were known as early as 1850, but what was
lacking were data, especially gravimetric ones. Serious practical applications
started not much before 1950, a hundred years later, just before the advent
of satellites. This will be discussed in detail later in this book.
A reasonable measuring accuracy was achievable, but another difficulty
appeared. Both methods require the evaluation of integrals of the data (ξ and
η, or ∆g) as continuous functions. The data, however, are always measured
at discrete points only. Interpolation is necessary and introduces additional
errors. If the data are distributed uniformly and densely, resulting errors
may be kept small. The fundamental problem exists, however.
Summarizing, we may say: (1) The method of zenith angles is theoreti-

cally rigorous but not in general sufficiently accurate; (2) the astrogeodetic
method using integration of vertical deflections is not theoretically rigorous
in this sense but still may be accurate enough.
Method 1 has been treated in Part II of this chapter, so method 2 war-
rants detailed considerations in the present Part III.
5.12 Reduction of the astronomical measurements

to the ellipsoid
Now we establish the relation between the natural coordinates Φ, Λ, H and
the ellipsoidal coordinates ϕ, λ, h referring to an ellipsoid according to Hel-
mert’s projection.
The ellipsoidal height h and the orthometric height H have been consid-
ered, e.g., in Sect. 4.6 (see also Fig. 5.4 and Eq. (5–91)). They are related
by h = H + N .
Thus, there remains the reduction of the astronomical coordinates Φ and
Λ to the ellipsoid and, if we also include the astronomical observation of the
azimuth, the astronomical azimuth A to the ellipsoid in order to obtain the
ellipsoidal coordinates ϕ and λ and the ellipsoidal azimuth α.
We introduce the auxiliary quantities
∆ϕ = Φ − ϕ ,
∆λ = Λ − λ , (5–92)
∆α = A − α .
The reduction of Φ and Λ to the corresponding ellipsoidal coordinates ϕ and
λ is implicitly contained in Eq. (2–230):
ξ = Φ − ϕ = ∆ϕ ,
(5–93)
η = (Λ − λ) cos ϕ = ∆λ cos ϕ ,
where we have substituted the respective auxiliary quantities. Thus, the con-
version formulas from natural coordinates Φ, Λ, H to ellipsoidal coordinates
ϕ, λ, h are
ϕ=Φ−ξ,
λ = Λ − η/ cos ϕ , (5–94)
h =H +N.
Now we turn to the reduction of the azimuth. Thus, the question is which
∆α arises from ∆ϕ and ∆λ. The answer is found in Eq. (5–75), where we
only consider the last two terms on the right-hand side (i.e., we do not take
into account changes of the point coordinates). Omitting all subscripts and
introducing the auxiliary quantities of (5–92), we immediately get
∆α = cot z sin α ∆ϕ + (sin ϕ − cos α cos ϕ cot z) ∆λ (5–95)
or, using ∆ϕ = ξ and ∆λ cos ϕ = η, yields
∆α = ξ sin α cot z + sin ϕ ∆λ − η cos α cot z . (5–96)
This equation may be rearranged to
∆α = sin ϕ ∆λ + (ξ sin α − η cos α) cot z . (5–97)
Alternatively, by using ∆λ = η/ cos ϕ, we get
∆α = η tan ϕ + (ξ sin α − η cos α) cot z . (5–98)
In first-order triangulation, the lines of sight are usually almost horizontal
. .
so that z = 90◦ , cot z = 0. Therefore, the corresponding term can in general
be neglected and we get
∆α = η tan ϕ = ∆λ sin ϕ . (5–99)
This is Laplace’s equation in its usual simplified form. It is remarkable that
the differences ∆α = A − α and ∆λ = Λ − λ should be related in such a
simple way. Laplace’s equation is fundamental for the classical astrogeodetic
computation of triangulations (Sect. 5.14).
For later reference we note that the total deflection of the vertical – that
is, the angle ϑ between the actual plumb line and the ellipsoidal normal – is
given by
ϑ = ξ 2 + η2 (5–100)
and that the deflection component ε in the direction of the azimuth α is
ε = ξ cos α + η sin α . (5–101)
It is clear that ϑ in (5–100) has nothing to do with the two different ϑ
used for spherical and ellipsoidal-harmonic coordinates (polar distances).
Returning to the reduction of astronomical to the corresponding ellip-
soidal quantities, we have (5–94) for the reduction of Φ, Λ, H to ϕ, λ, h and,
finally, the formula
α = A − η tan ϕ (5–102)
reduces the astronomical azimuth A to the ellipsoidal azimuth α.
For the application of these formulas, we need the geoidal undulation N
and the deflection components ξ and η with respect to the reference ellipsoid
used. Two points should be noted:
1. The vertical axis of the reference ellipsoid is parallel to the earth’s

axis of rotation, but it need not be in an absolute position, its center
coinciding with the earth’s center of gravity. To repeat the reason:
the earth axis is accessible to (astronomical) observation, whereas the
geocenter is physically defined and inaccessible to direct geometrical
observation.
2. The geocenter is accessible in two physically defined ways: (1) gravi-

metrically through Stokes’ formula and (2) physically by the first Kep-
ler law applied to satellite motion and responsible for the geocentricity
of GPS orbits.
Note that unless otherwise stated, we always assume that our observa-
tions are made at sea level. This is not so unnatural for an inhabitant of a
large plain region but causes headache to a geodesist working in the Alps
or in the Rocky Mountains. We have already been confronted with this sit-
uation before, in gravity reduction, and will meet it repeatedly later, most
prominently under the heading of Molodensky’s problem.
It should also be mentioned that the ellipsoidal azimuth α in (5–102)
refers to the actual target, which does not in general lie on the ellipsoid.
For the conventional method of computation on the ellipsoid, one wishes the
azimuth to refer to a target on the ellipsoid, which is the point at the foot of
the normal through the actual target. Furthermore, α refers to what is called
a normal section of the ellipsoid, rather than to a geodesic line, which is used
in computation. In either case very small azimuth reductions are necessary;
since these reductions are purely problems in ellipsoidal geometry, the reader
is referred to any appropriate textbook.
Effect of polar motion

The direction of the earth’s axis of rotation is not rigorously fixed, neither
in space nor with respect to the earth, but undergoes very small, more or
less periodic variations. Astronomers know it by the name of nutation (with
respect to inertial space), geodesists know it by the name of polar motion
(with respect to the earth’s body). This phenomenon arises from a minute
difference between the axes of rotation and of maximum inertia, the angle
between these axes being about 0.3 , and is somewhat similar to the preces-
sion of a spinning top. This motion of the pole has a main period of about
430 days, the Chandler period, but is rather irregular, presumably because
of the movement of masses, atmospheric variations, etc. (Fig. 5.13).
The International Earth Rotation Service (IERS), initially International
Latitude Service and then Polar Motion Service, which is maintained by
the International Astronomical Union and by the International Union of
1996 –0.2
1995.0 1900.0
1998.5
+0.6 +0.4 +0.2 1997
y ['']
1997.0
1998.0
+0.2
x ['']
Fig. 5.13. Polar motion: mean pole displacement 1900–1997 (solid line),
detailed polar motion 1995–1998 (dotted line)
Geodesy and Geophysics, continuously observes the variation of a number of

parameters at a considerable number of stations distributed over the whole
earth. Thus, it monitors variations of the earth’s axis (polar motion) and of
its angular speed of rotation.
The results are published as the rectangular coordinates of the instanta-
neous pole PN with respect to a mean pole PN0 . The astronomically observed
values of Φ, Λ, and A naturally refer to the instantaneous pole PN and must,
therefore, be reduced to the mean pole, using the published values of x and
y.
This is accomplished by means of the equations
Φ = Φobs − x cos λ + y sin λ ,
Λ = Λobs − (x sin λ + y cos λ) tan ϕ + y tan ϕGr , (5–103)
A = Aobs − (x sin λ + y cos λ) sec ϕ .
Now Φ, Λ, A are referred to the mean pole; these values are used in geodesy
because they do not vary with time. Longitude, throughout this book, is
reckoned positive to the east, as is usual in geodesy; it should be mentioned
that in the past literature these formulas are often written for west longi-
tude, according to the former practice of astronomers. Since the correction
terms containing x and y are extremely small (of the order of 0.1 ), we may
use either the ellipsoidal values ϕ and λ or the astronomical values Φ and Λ
in these terms. The term containing ϕGr (the latitude of Greenwich) in the
formula for Λ is usually omitted, so that the mean meridian of Greenwich re-
mains fixed as the conventional zero meridian, rather than the astronomical
5.13 Reduction of horizontal and vertical angles and of distances 221
longitude of Greenwich itself.

These formulas (5–103) are Eqs. (7-13), (7-14), and (7-15) of Moritz and
Mueller (1987: pp. 419–420). It is interesting to note the close similarity
between the azimuth reduction (5–98) because of the “zenith variation” –
that is, the deflection of the vertical – and the longitude reduction of (5–103)
because of the polar variation. Actually, the geometry for both cases is the
same. The quantities ξ, η, 90◦ − z, ϕ correspond to x, y, ϕ, ϕGr ; the difference
in sign of sin α and sin λ is due to the fact that, when viewed from the
zenith, azimuth is reckoned clockwise and, when viewed from the pole, east
longitude is reckoned counterclockwise.
5.13 Reduction of horizontal and vertical angles

and of distances
Horizontal angles
To reduce an observed horizontal angle ω to the ellipsoid, we note that every
angle may be considered as the difference between two azimuths:
ω = α2 − α1 . (5–104)
Hence, we can apply formula (5–98). In the difference α2 − α1 , the main
term η tan ϕ drops out, so that for nearly horizontal lines of sight the whole
reduction may be neglected.
Vertical angles
The relation between the measured zenith angle z and the corresponding
ellipsoidal zenith angle z may be given as
z = z + ε = z + ξ cos α + η sin α , (5–105)
where α is the azimuth of the target.
Spatial distances
Electronic measurement of distance yields straight spatial distances l be-
tween two points A and B (Fig. 5.14). These distances may either be used
directly for computations in the ellipsoidal coordinate system ϕ, λ, h, as in
“three-dimensional geodesy” (see Sect. 5.9), or they may be reduced to the
surface of the ellipsoid to obtain chord distances l0 or geodesic distances s0 .
We again approximate the ellipsoidal arc A0 B0 by a circular arc of radius
R that is the mean ellipsoidal radius of curvature along A0 B0 . By applying
the law of cosines to the triangle OAB, we find
l2 = (R + h1 )2 + (R + h2 )2 − 2(R + h1 )(R + h2 ) cos ψ . (5–106)
B
A l
h1 s0 h2
A0 l0 B0
R R
Ã
0
Fig. 5.14. Reduction of spatial distances
With
ψ
cos ψ = 1 − 2 sin2 , (5–107)
2
this is transformed into

h1 h2 ψ
l2 = (h2 − h1 )2 + 4R2 1 + 1+ sin2 ; (5–108)
R R 2
and with
ψ
l0 = 2R sin (5–109)
2
and the abbreviation ∆h = h2 − h1 , we obtain

h1 h2 2
l2 = ∆h2 + 1 + 1+ l . (5–110)
R R 0
Hence, the chord l0 and the arc s0 are expressed by

(
) l2 − ∆h2
)
l0 = * ; (5–111)
1 + hR1 1 + hR2
5.14 The astrogeodetic determination of the geoid 223
l0
s0 = R ψ = 2R sin−1 . (5–112)
2R
Ellipsoidal refinements of these formulas may be found in Rinner (1956).
As a matter of fact, spatial distances are independent of the vertical.
Therefore, the reduction formula (5–111) does not contain the deflection of
the vertical ε.
5.14 The astrogeodetic determination of the geoid

Helmert’s formula
The shape of the geoid can be determined if the deflections of the vertical
are given. Helmert’s formula
dN = −ε ds (5–113)
as given in (2–372) is the basic equation (Fig. 5.15). Integrating this relation,
we get B
NB = NA − ε ds , (5–114)
A
where
ε = ξ cos α + η sin α (5–115)
is the component of the deflection of the vertical along the profile AB, whose
azimuth is α (see Eq. (5–101)).
Formula (5–114) expresses the geoidal undulation as an integral of the
vertical deflections along a profile. Since N is a function of position, this
integral is independent of the form of the line that connects the points A and
B. This line need not necessarily be a geodesic on the ellipsoid, and α may in
the general case be variable. In practice, north-south profiles (ε = ξ) or east-
west profiles (ε = η) are often used. The integral (5–114) is to be evaluated
plumb line ellipsoid normal

"
" geoid
dN
ds
ds
s ellipsoid
Fig. 5.15. Relation between geoidal undulation and deflection of

the vertical
by a numerical or graphical integration. The deflection component ε must

be given at enough stations along the profile such that the interpolation
between these stations can be done reliably. Sometimes a map of ξ and η
is available for a certain area. Such a map is constructed by interpolation
between well-distributed stations at which ξ and η have been determined
(Grafarend and Offermanns 1975). Then the profiles of integration may be
suitably selected; loops may be formed to obtain redundancies which must
be adjusted.
If the deflection components ξ and η are obtained directly from the equa-
tions
ξ = Φ − ϕ , η = (Λ − λ) cos ϕ , (5–116)
that is, by comparing the astronomical and ellipsoidal (or geodetic) coordi-
nates of the same point, then this method is called the astrogeodetic deter-
mination of the geoid.
The astronomical coordinates are directly observed; the ellipsoidal coor-
dinates are obtained in the following way.
Determination of a local astrogeodetic datum

This is of historic interest only, but indispensible for an understanding of
the present classical triangulation system. In agreement with Part I, but in
contrast to Part II, “local” again means “regional”, referring to a country
(e.g., France) or even a continent (e.g., European Datum or North-American
Datum). In a larger triangulation system, a certain “initial point” P1 is
chosen for which the undulation N1 and the components ξ1 and η1 of the
deflection of the vertical are prescribed. Here ξ1 , η1 , and N1 may be assumed
arbitrarily in principle; the position of the reference ellipsoid with respect
to the earth is thereby fixed. For the sake of definiteness let us consider
the case that has been of greatest practical importance, that is, the case
in which ξ1 = η1 = N1 = 0. In this case, because ξ1 = η1 = 0, the geoid
and the ellipsoid have the same surface normal so that, because N1 = 0, the
ellipsoid is tangent to the geoid below P1 (Fig. 5.16). The condition that
the axis of the reference ellipsoid be parallel to the earth’s axis of rotation
finally determines the orientation of the triangulation net because Laplace’s
equation (5–99) then gives ∆α1 = η1 tan ϕ1 = 0, so that α1 = A1 ; that is, at
the initial point the ellipsoidal azimuth is equal to the astronomical azimuth.
Now we can reduce the measured distances and angles to the ellipsoid
and compute on it the position of the points of the triangulation net (their
ellipsoidal coordinates ϕ and λ) in the usual way. After measuring the coor-
dinates Φ and Λ astronomically at the same points, we can then compute the
deflection components ξ and η by (5–116). Starting from the assumed value
P1
rotation axis
ellipsoid
geoid
Fig. 5.16. The reference ellipsoid is tangent to the geoid at P1
N1 at the initial point P1 (in our case, N1 = 0), we can finally compute the
geoidal heights N of any point of the triangulation net by repeated applica-
tion of (5–114). These geoidal heights refer to the ellipsoid that was fixed by
prescribing ξ1 , η1 , N1 , and, of course, its semimajor axis a and its flattening
f . To employ a frequently used term, they refer to the given astrogeodetic
datum (a, f ; ξ1 , η1 , N1 ).
By means of N and the orthometric height H, the height h above the
ellipsoid is obtained via h = H + N , so that the rectangular spatial coordi-
nates X, Y, Z can be computed by (5–27). But unless ξ and η are absolute
(geocentric) deflections, the origin of the coordinate system will not be at
the center of the earth (see Sect. 5.7).
A flaw in the procedure described above apparently is that N, ξ, η are
already needed for the reduction of the measured angles and distances to
the ellipsoid. However, for this purpose approximate values of N, ξ, η are
sufficient. These are obtained by performing the process just explained with
unreduced angles and distances. We can also get suitable values for N, ξ, η
in other ways, for instance, by Stokes’ formula.
Use and misuse of Laplace’s equation

It should be mentioned that in practice the component η has been often
obtained from azimuth measurements using (5–102) in rearranged form, that
is,
η = (A − α) cot ϕ , (5–117)
because astronomical measurements of azimuth are simpler than those of

longitude. This is a misuse which may lead to a systematic distortion of the
net. Longitude and azimuth are often measured at the same point. Then
Laplace’s condition
∆α = ∆λ sin ϕ (5–118)
furnishes an important check on the correct orientation of the net and forces
the axis of the ellipsoid to be parallel with the earth’s axis of rotation. Thus
it may be used for adjustment purposes. Astronomical stations with longi-
tude and azimuth observations are, therefore, called Laplace stations. For
these purposes, the measuring accuracy of astronomical field observations is
sufficient, in contrast to the use for directly determining horizontal positions
by ϕ = Φ − ξ, etc. in Sect. 2.21.
The astrogeodetic determination of the geoid, also called astronomical
leveling, was known to Helmert (1880) and even before.
Comparison with the Stokes method

It is quite instructive to compare Helmert’s formula
B
N = NA − ε ds (5–119)
A
for the astrogeodetic method with Stokes’ formula

R
N= ∆g S(ψ) dσ (5–120)
4π γ0
σ
for the gravimetric method. Both methods use the gravity vector g. It is
compared with a normal gravity vector γ. The components ξ = ∆ϕ and
η = ∆λ cos ϕ of the deflection of the vertical represent the differences in
direction, and the gravity anomaly ∆g represents the difference in magnitude
of the two vectors. Helmert’s formula determines the geoidal undulation N
from ξ and η, that is, by means of the direction of g, and Stokes’ formula
determines N from ∆g, that is, by means of the magnitude of g. Both
formulas are somewhat similar: they are integrals which contain ε, or ξ and
η, and ∆g in linear form.
Otherwise, the two formulas show marked differences which are charac-
teristic for the respective method. In Helmert’s formula, the integration is
extended over part of a profile; thus, it is sufficient to know the deflection
of the vertical in a limited area. The position of the reference ellipsoid with
respect to the earth’s center of gravity is unknown, however, and can be
determined only by means of the gravimetric method or, more practically,
the analysis of satellite orbits (Sect. 7.2). Furthermore, the astrogeodetic
method can be used only on land, because the necessary measurements are
impossible at sea.
In Stokes’ formula, however, the integration should be extended over

the whole earth. The gravity anomaly ∆g must be known all over the earth;
however, accurate gravity measurements at sea are possible. The gravimetric
method yields, for the whole earth, absolute geoidal undulations: the center
of the reference ellipsoid coincides with the center of the earth. Nowadays,
this is only a theoretical possibility because the required complete cover-
age of the whole earth is not available; again, GPS helps. Nevertheless, the
gravimetric method is still basic: it furnishes, not the geocenter, but details
of the geoid, together with the astrogeodetic method!
The astrogeodetic method has often been applied to the determination
of geoidal sections. We mention, because of its pioneering character and its
romantic title, “Das Geoid im Harz” by Galle (1914). In the years following
1970 it is becoming rare to use Helmert’s integral formula in its original
form, and deflections of the vertical are more and more combined with other
data (gravity, GPS, and other satellite data) for a uniform determination of
geoid and gravity field (see Chaps. 10 and 11).
Adjustment of nets of astrogeodetic geoidal heights

With a sufficiently dense net of astrogeodetic stations (preferably Laplace
points) with an average station distance of 10–20 km, the Helmert integral
(5–119) can be approximated by
B B
εA + εB
∆NAB ≡ NB − NA = − ε ds = − ds (5–121)
A 2 A
or
εA + εB
∆NAB = − sAB . (5–122)
2
Thus, the undulation difference can be computed for the line AB, and simi-
larly for other lines BC and CA in the triangle ABC (Fig. 5.17). The closure
condition
∆NAB + ∆NBC + ∆NCA = 0 (5–123)
must be satisfied and imposed as a condition in the least-squares adjustment
A sAB
B
Fig. 5.17. Triangular net for an astrogeodetic geoid

of the net. Accordingly, the other triangles can be computed as in any other
height network (e.g., leveling net).
It is curious that it may be shown that such closures are mathematically
equivalent to the well-known relation
∂2N ∂2N
= . (5–124)
∂x ∂y ∂y ∂x
See also Sect. 4.5.
5.15 Reduction for the curvature of the plumb line

Motivation
The astronomical coordinates Φ and Λ, as observed on the surface of the
earth, are not rigorously equal to their corresponding values at the geoid
because the plumb line, the line of force, is not straight, or in other words,
because the level surfaces are not parallel. Thus, if we wish our astronomical
coordinates to refer to the geoid, we must reduce our observations accord-
ingly.
Examples of such cases are the following:
1. The gravimetric deflections have usually been computed by Vening

Meinesz’ formula for the geoid, so that either the gravimetric deflec-
tions must be reduced upward to the ground point or the astronomical
observations must be reduced downward to the geoid, in order to make
the two quantities comparable.
2. If astronomical observations are used for the determination of the

geoid, the same reduction, in principle, must be applied.
Important remark
The principle of reduction of the plumb line is of fundamental theoretical
importance for understanding the geometry of the earth’s gravity field. In
practice, it is usually disregarded if the topography is sufficiently flat, or
replaced by more sophisticated methods in mountainous areas, as we shall
see later (Sects. 8.12 and 8.13). The present section may be skimmed at first
reading, except for the normal curvature of the plumb line at its very end.
Principles
Consider the projection of the plumb line onto the meridian plane. According
to the well-known definition of the curvature of a plane curve, the angle
5.15 Reduction for the curvature of the plumb line 229
between two neighboring tangents of this projection of the plumb line is
dϕ = −κ1 dh , (5–125)
where the minus sign is conventional and the curvature κ1 is given by (2–50):
1 ∂g
κ1 = . (5–126)
g ∂x
The x-axis is horizontal and points northward. Hence, the total change of
latitude along the plumb line between a point on the ground, P , and its
projection onto the geoid, P0 , is given by
P P
δϕ = dϕ = − κ1 dh (5–127)
P0 P0
or P
1 ∂g
δϕ = − dh . (5–128)
P0 g ∂x
Using κ2 of (2–51), we similarly find for the change of longitude
P
1 ∂g
δλ cos ϕ = − dh , (5–129)
P0 g ∂y
where the y-axis is horizontal and points eastward.
Alternative formulas
There is a close relationship between the curvature reduction of astronomical
coordinates and the orthometric reduction of leveling, considered in Sect. 4.3.
The orthometric correction d(OC) has been defined as the quantity that
must be added to the leveling increment dn in order to convert it into the
orthometric height difference dH:
d(OC) = dH − dn . (5–130)
From Fig. 5.18, we see that, for a north-south profile, the curvature reduction
and the orthometric correction are related by the simple formula
∂(OC)
δϕ = . (5–131)
∂x
Similarly, we find
∂(OC)
δλ cos ϕ = . (5–132)
∂y
earth's surface
dn
dH local horizon
P ±' d (OC)
parallel to geoid
plumb line
±'
Fgeoid Fground
x (north)
P0 geoid
parallel to
equatorial plane
Fig. 5.18. Plumb-line curvature and orthometric correction
According to Sect. 4.3, we have

C
dC = g dn = −dW , H= . (5–133)
ḡ
Hence, (5–130) becomes
1 1
d(OC) = dH − dC = dH + dW , (5–134)
g g
so that
∂H 1 ∂W
δϕ = + ,
∂x g ∂x
(5–135)
∂H 1 ∂W
δλ cos ϕ = + .
∂y g ∂y
These equations relate the reduction for the curvature of the plumb line
to the orthometric height H and the potential W . In view of the irregular
shape of the plumb lines, it is remarkable that such simple general relations
as (5–131), (5–132), and (5–135) exist.
5.15 Reduction for the curvature of the plumb line 231
These relations may be used to find computational formulas for the cur-
vature reductions δϕ and δλ. We have

dC C dC
d(OC) = dH − =d −
g ḡ g
(5–136)
dC C dC C g − ḡ dC
= − 2 dḡ − = − 2 dḡ +
ḡ ḡ g ḡ ḡ g
or
H g − ḡ
d(OC) = − dḡ + dn . (5–137)
ḡ ḡ
By substituting this into (5–131) and (5–132), we obtain
H ∂ḡ g − ḡ
δϕ = − + tan β1 ,
ḡ ∂x ḡ
(5–138)
H ∂ḡ g − ḡ
δλ cos ϕ = − + tan β2 ,
ḡ ∂y ḡ
where we have set

∂n ∂n
tan β1 = , tan β2 = , (5–139)
∂x ∂y
so that β1 and β2 are the angles of inclination of the north-south and east-
west profiles with respect to the local horizon; ḡ is the mean value of gravity
between the geoid and the ground. In these formulas, we need only this
mean value ḡ, together with its horizontal derivatives, and the ground value
g, whereas in (5–128) and (5–129), we must know the horizontal derivatives
of gravity all along the plumb line. The detailed shape of the plumb lines
does not directly enter into (5–138) as it does into (5–128) and (5–129).
The mean value ḡ is found by a Prey reduction of the measured gravity
g. In order that the numerical differentiations ∂g/∂x and ∂g/∂y give reliable
results, a dense gravity net around the station is necessary, and the Prey
reduction must be performed carefully. The inclination angles β1 and β2 are
taken from a topographical map.
The sign of these corrections may be found in the following way. If g
decreases in the x-direction, then formulas (5–128) and (5–138) give δϕ > 0
and Fig. 5.18 shows that Φ at P0 is then greater than at P . The same holds
for Λ, so that we have
Φgeoid = Φground + δϕ ,
(5–140)
Λgeoid = Λground + δλ .
Integrated form
In formula (5–114), the deflection components ξ and η refer to the geoid.
This means that the astronomical observations of Φ and Λ must be reduced
to the geoid.
It is also possible and often more convenient to apply this correction
for plumb-line curvature not to the astronomical coordinates Φ and Λ but
to the geoidal height differences computed from the unreduced deflection
components.
These N values, denoted by N , are obtained by using in (5–116) the
directly observed Φ and Λ, which define the direction of the plumb line at
the station P (Fig. 5.19). The notation N will be reserved for the correct
geoidal heights. Then we read from Fig. 5.19:
dh = dN + dH = dN + dn , (5–141)
where h is the geometric height above the ellipsoid. Thus, we see that the
difference between the unreduced and the correct element of geoidal height,
dN − dN = dH − dn = d(OC) , (5–142)
is equal to the difference between the element dH of orthometric height and
earth's surface
dn
plumb ellipsoidal dH
line normal local horizon
W = WP
P dN dN '
h
geoid
normal
geoid W = W0
dN
P0
ds ellipsoid
Fig. 5.19. Reduction of astronomical leveling

5.16 Best-fitting ellipsoids and the mean earth ellipsoid 233
the leveling increment dn, which is the orthometric reduction d(OC). Thus,
NB − NA = NB − NA − OCAB , (5–143)
so that we can immediately apply Eq. (4–46):

B B
g − γ0 ḡB − γ0 ḡA − γ0
NB −NA = − ε ds− dn+ HB − HA , (5–144)
A A γ0 γ0 γ0
where γ0 is our usual constant γ45◦ ; the deflection components ε are com-
puted from the observed ground values Φ and Λ by (5–116) and (5–115).
These ideas go back to Helmert, but they are hardly used anymore.
Curvature of the normal plumb line

If, instead of the actual gravity g, the normal gravity γ is applied for the
computation of the plumb-line curvature, we find, using

2
γ = γa 1 + f ∗ sin2 ϕ − h · · · , (5–145)
a
that
∂γ . 1 ∂γ . 2γa ∗ . 2γ ∗
= = f sin ϕ cos ϕ = f sin ϕ cos ϕ ,
∂x R ∂ϕ R R
(5–146)
∂γ . 1 ∂γ
= = 0.
∂y R cos ϕ ∂λ
Hence, the integrand (1/γ)(∂γ/∂x) in (5–128) does not depend on h, so that
the integration can be performed immediately. We find
f∗
δϕnormal = − h sin 2ϕ = −0.17 h [km] sin 2ϕ ,
R (5–147)
δλnormal = 0.
The curvature of the normal plumb line in the east-west direction is zero,
owing to the rotational symmetry of the ellipsoid of revolution. The normal
reduction (5–147) is very simple and practically important, see especially
Sect. 8.13.
5.16 Best-fitting ellipsoids and the mean

earth ellipsoid
We define the mean earth ellipsoid physically as that ellipsoid of revolution
which shares with the earth the mass M , the potential W0 , the difference
between the principal moments of inertia G(C − Ā), where Ā = (A + B)/2,

and the angular velocity ω.
It is also possible to define the mean earth ellipsoid geometrically as
that ellipsoid which approximates the geoid most closely. This definition is
perhaps more appealing to the geodesist; it may, for instance, be formulated
by the condition that the sum of the squares of the deviations N of the geoid
from the ellipsoid be a minimum:

N 2 dσ = minimum (5–148)
σ
(this integral is to be considered the limit of a sum). The condition of clos-

est approximation may also be expressed in terms of the deflections of the
vertical:
(ξ 2 + η 2 ) dσ = minimum , (5–149)
σ
minimizing the sum of the squares of the total deflection of the vertical

ϑ = ξ 2 + η2 . (5–150)
Many other similar definitions of closest approximation are possible.

The first definition, based on the condition (5–148), is the most plausible
and the most appropriate intuitively, as has been already noted by Helmert;
in principle, however, all definitions are more or less conventional and are
equivalent theoretically as we shall see below.
The second definition, based on the condition (5–149), uses deflections
of the vertical and is, thus, particularly well adapted to the astrogeodetic
method. However, since this method can be applied only over limited areas,
at most spanning the continents, the integral (5–149) must be replaced by a
sum covering the astronomical stations of a restricted region:

(ξ 2 + η 2 ) dσ = minimum . (5–151)
In this way, we can get only the best-fitting ellipsoid for the region consid-
ered, rather than a general earth ellipsoid. As Fig. 5.20 indicates, a locally
best-fitting ellipsoid may be quite different from the mean earth ellipsoid,
which can be considered a best-fitting ellipsoid for the whole earth.
If a reasonably good approximation of the earth ellipsoid by a local best-
fitting ellipsoid is desired, it is advisable to subtract the effect of the topog-
raphy and of its isostatic compensation from the astrogeodetic deflections of
the vertical before the minimum condition (5–151) is applied. The purpose
of this procedure is to smooth the irregularities of the geoid. In this way,
best-fitting ellipsoid
for region AB A
geoid
mean earth ellipsoid
Fig. 5.20. A locally best-fitting ellipsoid and the mean earth ellipsoid
Hayford computed the international ellipsoid as ellipsoid that best fits the
isostatically reduced vertical deflections in the United States. Rapp (1963)
made an interesting recomputation.
Please note: Don’t use formula (5–151) in spite of its historical impor-
tance: the determination of local best-fitting ellipsoids is hopelessly obsolete
now!
The previously described method is impaired by unknown density anoma-
lies and by the lack of complete isostatic compensation. Therefore, it is better
to go still one step further and subtract the gravimetrically computed values
ξ g , η g from the astrogeodetic deflections ξ a , η a . Then the minimum condition

(ξ a − ξ g )2 + (η a − η g )2 = minimum (5–152)
results. Thus, we may say that Hayford’s method is equivalent to the use of
(5–152), the gravimetric values ξ g , η g being approximated by deflections that
represent the effect of topography and of its isostatic compensation only. If
the isostatic compensation were complete, and if we had perfect knowledge
of the density above the geoid, both methods would give exactly the same
result if applied properly.
Equivalence of different definitions of the earth ellipsoid

It is quite remarkable that the minimum definitions (5–148) or (5–149) and
a similar definition due to Rudzki, using the condition

(∆g)2 dσ = minimum , (5–153)
σ
yield results which, to the usual spherical approximation, are identical with
each other and with the physical definition in terms of M , W0 , C − Ā, and
ω. This can be seen as follows. We write the spherical-harmonic expansion

of the disturbing potential in the form
∞ n
G δM
T = + anm Rnm (ϑ, λ) + bnm Snm (ϑ, λ) . (5–154)
R
n=1 m=0
Then, according to Sect. 2.17, Eqs. (2–351) and (2–359) or (2–363), we have
∞ n
G δM δW 1
N= − + anm Rnm (ϑ, λ) + bnm Snm (ϑ, λ) (5–155)
R γ0 γ0 γ0 n=1 m=0
and
G δM 2δW
∆g = − +
R2 R
∞ n
1
+ (n − 1) anm Rnm (ϑ, λ) + (n − 1) bnm Snm (ϑ, λ) ;
R n=1 m=0
(5–156)
remember that γ0 denotes a global mean value of gravity. The condition of
equal masses, δM = 0, is very natural and will be assumed. If we square
the formulas for N and ∆g and integrate over the whole earth, then all
the integrals of products of different harmonics Rnm and Snm will be zero,
according to the orthogonality property (1–83), and the remaining integrals
will be given by (1–84). Thus, we find

4π
N 2 dσ = 2 δW 2
γ0
σ
∞
n

4π 1 (n + m)!

+ 2 a2n0 + a2 + b2nm ,
γ0 n=1 2n + 1 2(n − m)! nm
m=1
(5–157)

16π
(∆g)2 dσ = δW 2
R2
σ
∞
n

4π (n − 1)2 (n + m)!

+ 2 a2n0 + a2 + b2nm .
R 2n + 1 2(n − m)! nm
n=1 m=1
(5–158)
By a more complicated derivation, which we omit here but which can be
found in Molodenskii et al. (1962: p. 87), one gets the similar formula
∞
n

2 2 4π n (n + 1) 2
(n + m)!
2 2

(ξ + η ) dσ = 2 2 an0 + a + bnm .
R γ0 2n + 1 2(n − m)! nm
σ n=1 m=1
(5–159)
Varying the size and shape of the reference ellipsoid and its position with
respect to the earth changes only the coefficients δW , a10 , a11 , b11 , and a20 ,
leaving the other coefficients practically invariant. Thus, the minimum of any
of the integrals (5–157), (5–158), (5–159) is obtained if all these coefficients
are equal to zero. Now, δW = 0 means equal potential U0 = W0 ; a10 = a11 =
b11 = 0 means absolute position (coincident centers of gravity); and a20 = 0
means equality of J2 or of C − (A + B)/2.
Therefore, the equivalence of the physical definition by means of M, W0 ,
C − Ā, ω and of the condition of closest approximation in any of the forms
(5–148), (5–149), or (5–153) has been established. (It may be noted that
(5–158) contains no first-degree term, because of the factor (n − 1)2 , and
that (5–159) contains no term of degree zero, so that these equations do not
determine the missing terms.)
Best-fitting ellipsoid and World Geodetic System

It should be remembered, however, that the mean earth ellipsoid, defined
in this manner, is not necessarily the best reference surface for practical
geodetic purposes. It is essentially defined empirically by means of empiri-
cal determinations of GM, W0 , etc. Its parameters will change with every
improvement in the quality or the number of the relevant measurements
(gravity, distances, etc.). Since an enormous amount of numerical data is
based on an assumed reference ellipsoid, it would be highly impractical to
change it very often, for this would involve repeated transformations of all
the data. It is much better to use a fixed reference ellipsoid with rigidly
assumed parameters, which can be more or less arbitrary if only they give
a reasonably good approximation. In this respect, the Geodetic Reference
System 1980 is still (2005) perfectly acceptable.
A certain amount of conflict exists between the interests of geodesists and
astronomers regarding the earth ellipsoid. The geodesist needs a permanent
reference surface, whereas the astronomer wants the best approximation of
the earth by an ellipsoid. A good compromise is to use a fixed geodetic
reference ellipsoid, but from time to time to compute the “best” corrections
to the assumed parameters for astronomical and other purposes. This has
been the practice of the IAG since 1974.
6 Gravity field outside
the earth
6.1 Introduction
The gravity field outside the earth is particularly important at satellite al-
titude; this will be treated mainly in Chap. 7. The considerations of the
present chapter are applicable to gravitational forces also at satellites (see
Sect. 7.2), but their main practical purpose is to compute test values for the
gravity vector, gravity disturbances, and gravity anomalies at flight eleva-
tions for comparison with airborne gravimetry for reference and calibration
purposes. Airborne gravimetry is much faster than both terrestrial and ship-
borne gravimetry, so it is of interest also for geophysical prospecting.
For computational reasons, it is again convenient to split the gravity
potential W and the gravity vector
g = grad W (6–1)
into a normal potential U and a normal gravity vector
γ = grad U , (6–2)
and the disturbing potential T = W − U and the gravity disturbance vector
δg = grad T = g − γ . (6–3)
The normal gravity field is usually taken to be the gravity field of a suit-
able equipotential ellipsoid. This permits closed formulas and offers other
advantages of mathematical simplicity (see Sect. 2.12).
Thus, U and γ are computed first, and W and g are then obtained by
W =U +T ,
(6–4)
g = γ + δg .
For some purposes, we need the vector of gravitation, grad V (pure at-
traction without centrifugal force), rather than the vector of gravity. The
gravitational vector is computed from the gravity vector by subtracting the
vector of centrifugal force:
⎡ 2 ⎤
ω x
grad V = g − grad Φ = g − ⎣ω 2 y ⎦ , (6–5)
0
240 6 Gravity field outside the earth
where the notations of Sect. 2.1 are used. The rectangular coordinate system
x, y, z will be applied in this chapter in the usual sense: it is geocentric, the
x- and y-axes lying in the equatorial plane with Greenwich longitudes 0◦ and
90◦ East, respectively, and the z-axis being the rotation axis of the earth.
The sign of the components of g, γ, δg, etc., will always be chosen so
that they are positive in the direction of increasing coordinates.
6.2 Normal gravity vector

The gravity field of an equipotential ellipsoid is best expressed in terms of
ellipsoidal-harmonic coordinates u, β, λ, introduced in Sects. 1.15 and 2.7.
They are related to rectangular coordinates x, y, z by
√
x= u2 + E 2 cos β cos λ ,
√
y= u2 + E 2 cos β sin λ , (6–6)
z = u sin β .
If x, y, z are given, then u, β, λ can be computed by closed formulas. First

we find
x2 + y 2 = (u2 + E 2 ) cos2 β , z 2 = u2 sin2 β . (6–7)
Eliminating β between these two equations, we obtain a quadratic equation

for u2 , whose solution is

1 1 4E 2 z 2
u2 = (x2 + y 2 + z 2 − E 2 ) + 1+ 2 . (6–8)
2 2 (x + y 2 + z 2 − E 2 )2
Then β is given by
√
u2 + E 2
z
tan β = , (6–9)
u x2 + y 2
and for λ we simply have
y
tan λ = . (6–10)
x
With known ellipsoidal-harmonic coordinates, the normal potential U is
given by (2–126):
GM E q
2
U (u, β) = tan−1 + 1
2 ω 2 a2 sin β − 13 + 12 ω 2 (u2 + E 2 ) cos2 β .
E u q0
(6–11)
6.2 Normal gravity vector 241
The components of γ along the coordinate lines are, by (2–131) and (2–132),

1 ∂U 1 GM ω 2 a2 E q
1
γu = =− + sin β − 6 − ω u cos β ,
2 1 2 2
w ∂u w u2 + E 2 u2 + E 2 q0 2

1 ∂U 1 ω 2 a2 q
γβ = √ =− −√ 2 2 2
+ ω u + E sin β cos β ,
w u2 + E 2 ∂β w u2 + E 2 q0
1 ∂U
γλ = √ = 0.
u2 + E 2 cos β ∂λ
(6–12)
To get the components of γ in the xyz-system, we compute
∂U ∂U ∂x ∂U ∂y ∂U ∂z
= + + , etc. (6–13)
∂u ∂x ∂u ∂y ∂u ∂z ∂u
The partial derivatives of x, y, z with respect to u, β, λ are obtained by

differentiating equations (6–6); we find
∂U u ∂U u ∂U ∂U
=√ cos β cos λ +√ cos β sin λ + sin β ,
∂u u2 + E 2 ∂x u2 + E 2 ∂y ∂z
∂U √ ∂U 2 ∂U ∂U
= − u2 + E 2 sin β cos λ − u + E 2 sin β sin λ + u cos β ,
∂β ∂x ∂y ∂z
∂U √ ∂U 2 ∂U
= − u2 + E 2 cos β sin λ + u + E 2 cos β cos λ .
∂λ ∂x ∂y
(6–14)
Introducing the components
∂U 1 ∂U
γx = , ··· ; γu = , ··· , (6–15)
∂x w ∂u
we obtain
u u 1
γu = √ cos β cos λ γx + √ cos β sin λ γy + sin β γz ,
2
w u +E 2 2
w u +E 2 w
1 1 u
γβ = − sin β cos λ γx − sin β sin λ γy + √ cos β γz ,
w w w u + E2
2
γλ = − sin λ γx + cos λ γy .
(6–16)
These are the formulas of an orthogonal rectangular coordinate transforma-
tion. The inverse transformation is obtained by interchanging the rows and
columns in the matrix of this equation system. Thus, we obtain
u 1
γx = √ cos β cos λ γu − sin β cos λ γβ − sin λ γλ ,
w u2 + E2 w
u 1
γy = √ cos β sin λ γu − sin β sin λ γβ + cos λ γλ , (6–17)
w u2 + E 2 w
1 u
γz = sin β γu + √ cos β γβ .
w w u2 + E 2
This follows from the definition of these coefficients as direction cosines.

Equations (6–17) may also be found by solving the linear Eqs. (6–16) with
respect to γx , γy , γz in some other way.
The formulas of the present section are completely rigorous. They can
easily be programmed. Here it would not be appropriate to use the spherical
approximation because they are relatively large quantities of the normal
ellipsoidal field.
6.3 Gravity disturbance vector from gravity

anomalies
In Sect. 1.4, we have introduced spherical coordinates: r (radius vector), ϑ
(polar distance), λ (geocentric longitude) (see Fig. 1.3). Now we use these
coordinates again but replace the polar distance ϑ by its complement, the
geocentric latitude ϕ̄ (Fig. 6.3). In analogy to (1–26), these spherical coor-
r
z
# ' y
¸ r co
x s'
Fig. 6.1. Spherical coordinates r, ϕ̄ (or ϑ, respectively), λ and

rectangular coordinates x, y, z
6.3 Gravity disturbance vector from gravity anomalies 243
dinates are related to rectangular coordinates x, y, z by the equations
x = r cos ϕ̄ cos λ ,
y = r cos ϕ̄ sin λ , (6–18)
z = r sin ϕ̄
or inversely by

r = x2 + y 2 + z 2 ,
z
ϕ̄ = tan−1 , (6–19)
x2 + y2
y
λ = tan−1 .
x
Now it is convenient to start with the components δgr , δgϕ̄ , δgλ of the
gravity disturbance vector δg, Eq. (6–3), in the spherical coordinates r, ϕ̄, λ.
In analogy to (2–377), we have
∂T 1 ∂T 1 ∂T
δgr = , δgϕ̄ = , δgλ = . (6–20)
∂r r ∂ ϕ̄ r cos ϕ̄ ∂λ
Since we are dealing with the relatively small quantities of the disturbing
field, a spherical approximation may be sufficient (Sect. 2.13), as it was in
the case of Stokes’ formula.
The disturbing potential T may be expressed in terms of the free-air
anomalies at the earth’s surface by the formula of Pizzetti, Eqs. (2–302) and
(2–303),

R
TP = T (r, ϕ̄, λ) = ∆g S(r, ψ) dσ , (6–21)
4π
σ
where S(r, ψ) is the extended Stokes function,

2R R R l R2 r − R cos ψ + l
S(r, ψ) = + − 3 2 − 2 cos ψ 5 + 3 ln , (6–22)
l r r r 2r
and

l= r 2 + R2 − 2R r cos ψ . (6–23)
According to (6–20), we must differentiate (6–21) with respect to r, ϕ̄, and

λ. Here we note that the integral on the right-hand side of (6–21) depends
on r, ϕ̄, λ only through the function S(r, ψ). Thus, ∆g being constant with
respect to the differentiation, we have

R ∂S(r, ψ)
δgr = ∆g dσ ,
4π ∂r
σ

R ∂S(r, ψ)
δgϕ̄ = ∆g dσ , (6–24)
4π r ∂ ϕ̄
σ

R ∂S(r, ψ)
δgλ = ∆g dσ .
4π r cos ϕ̄ ∂λ
σ
The point P at which δg is to be computed has the coordinates ϕ̄, λ; let
the corresponding coordinates of the variable point P , to which ∆g and dσ
refer, be denoted by ϕ̄ , λ . Then dσ will be expressed by
dσ = cos ϕ̄ dϕ̄ dλ (6–25)
and ψ, the angular distance between P and P , is represented via
cos ψ = sin ϕ̄ sin ϕ̄ + cos ϕ̄ cos ϕ̄ cos(λ − λ) . (6–26)
We have
∂S(r, ψ) ∂S(r, ψ) ∂ψ ∂S(r, ψ) ∂S(r, ψ) ∂ψ
= , = . (6–27)
∂ ϕ̄ ∂ψ ∂ ϕ̄ ∂λ ∂ψ ∂λ
Now we recall the corresponding derivations in Sect. 2.19, leading to
Vening Meinesz’ formula. As a spherical approximation which is sufficient
for T , δg, etc., we may identify the geocentric latitude ϕ̄ with the ellipsoidal
latitude ϕ. Thus, Eqs. (6–27) and (2–380) are completely analogous, and
(2–383) may be borrowed from Sect. 2.19:
∂ψ ∂ψ
= − cos α , = − cos ϕ̄ sin α . (6–28)
∂ ϕ̄ ∂λ
The azimuth α is given by formula (2–388):
cos ϕ̄ sin(λ − λ)
tan α = . (6–29)
cos ϕ̄ sin ϕ̄ − sin ϕ̄ cos ϕ̄ cos(λ − λ)
By means of (6–27) and (6–28), Eqs. (6–24) become

R ∂S(r, ψ)
δgr = ∆g dσ ,
4π ∂r
σ

R ∂S(r, ψ)
δgϕ̄ = − ∆g cos α dσ , (6–30)
4π r ∂ψ
σ

R ∂S(r, ψ)
δgλ = − ∆g sin α dσ .
4π r ∂ψ
σ
6.3 Gravity disturbance vector from gravity anomalies 245
Now we form the derivatives of the extended Stokes function (6–22) with
respect to r and ψ. By differentiating (6–23), we get
∂l r − R cos ψ ∂l Rr
= , = sin ψ . (6–31)
∂r l ∂ψ l
By means of these auxiliary relations, we find
∂S R (r 2 − R2 ) 4R R 6R l
=− 3
− − 2+ 3
∂r rl rl r r

R2 r − R cos ψ + l
+ 3 cos ψ 13 + 6 ln ,
r 2r
(6–32)
∂S 2R2 r 6R2 8R2
= sin ψ − 3 − + 2
∂ψ l rl r

3R2 r − R cos ψ − l r − R cos ψ + l
+ 2 + ln .
r l sin2 ψ 2r
Somewhat more convenient expressions are obtained by substituting
R
t= , (6–33)
r
l
D= = 1 − 2t cos ψ + t2 . (6–34)
r
Then the extended Stokes function (6–22) and its derivatives (6–32) become

2 1 − t cos ψ + D
S(r, ψ) = t + 1 − 3D − t cos ψ 5 + 3 ln , (6–35)
D 2

∂S(r, ψ) t2 1 − t2 4
=− + + 1 − 6D
∂r R D3 D

1 − t cos ψ + D
− t cos ψ 13 + 6 ln ,
2
(6–36)
∂S(r, ψ) 2 6
= −t sin ψ
2 + −8
∂ψ D3 D

1 − t cos ψ − D 1 − t cos ψ + D
−3 − 3 ln .
D sin2 ψ 2
These expressions are used in (6–21) and (6–30) to compute T and δg.
The separation NP of the geopotential surface through P, W = WP , and
the corresponding spheropotential surface U = WP is according to Bruns’
theorem given by
TP
NP = ; (6–37)
γQ
see also Sect. 2.14 and Fig. 2.15.

The deflection of the vertical, which is the deviation of the actual plumb
line from the normal plumb line at P , is represented by its north-south and
east-west components,
1 ∂NP 1 ∂NP
ξP = − , ηP = − ; (6–38)
r ∂ ϕ̄ r cos ϕ̄ ∂λ
these equations correspond to (2–377). Since γ varies very little with latitude
and is independent of longitude, we have

∂NP ∂ TP 1 ∂TP TP ∂γQ . 1 ∂TP
= = − 2 = (6–39)
∂ ϕ̄ ∂ ϕ̄ γQ γQ ∂ ϕ̄ γQ ∂ ϕ̄ γQ ∂ ϕ̄
and
∂NP 1 ∂TP
= . (6–40)
∂λ γQ ∂λ
Substituting the results of (6–39) and (6–40) into (6–38) and comparing then
with (6–20) shows that
1 1
ξP = − δgϕ̄ , ηP = − δgλ . (6–41)
γQ γQ
We see that NP , ξP , ηP are given by Eqs. (6–21) and (6–30), apart from
the factor ±1/γQ . Hence, these equations are the extensions of Stokes’ and
Vening Meinesz’ formulas for points outside the earth and reduce to these
formulas for r = R, t = 1.
Writing Eqs. (6–41) in the form
δgϕ̄ = −γ ξ , δgλ = −γ η , (6–42)
we see that the horizontal components of δg are directly related to the de-
flection of the vertical, which is the difference in direction of the vectors
g and γ. The radial component δgr , however, represents the difference in
magnitude of these vectors, since as a spherical approximation
−δgr = δg = gP − γP , (6–43)
which is the scalar gravity disturbance (see Sect. 2.12).

Note that here the gravity disturbance δg is the basic quantity to be
computed, rather than the gravity anomaly ∆g, because both g and γ refer
to the computation point P .
6.4 Gravity disturbances by upward continuation 247
H l
s dx dy
xy
F R2d¾
R R
Fig. 6.2. Plane approximation
6.4 Gravity disturbances by upward continuation

We apply Poisson’s integral formula (1–123) to the harmonic function T :

R (r 2 − R2 ) T
TP = dσ . (6–44)
4π l3
σ
In the neighborhood of P (Fig. 6.2), the sphere practically coincides with

its tangent plane at F . Since the value of the integrand is very small at
greater distances from P , we may extend the integration over the tangent
plane instead of over the sphere. Then, according to Fig. 6.2,

l= s2 + H 2 . (6–45)
We introduce a rectangular coordinate system x, y, z, the x-axis pointing

north and the y-axis pointing east in the tangent plane. Then we may also
write
l = x2 + y 2 + H 2 , (6–46)
the surface element becomes

.
R2 dσ = dx dy , (6–47)
and we further have

r =R+H,
. (6–48)
r 2 − R2 = (r + R)(r − R) = 2R H .
Thus, (6–44) becomes the plane formula

H ∞ ∞ T H ∞ ∞ T
TP = 3
dx dy = dx dy .
2π −∞ −∞ l 2π −∞ −∞ (x + y + H 2 )3/2
2 2
(6–49)
This important formula is called the “upward continuation integral”. It per-
forms the computation of the value of the harmonic function T at a point
above the xy-plane from the values of T given on the plane, that is, the up-
ward continuation of a harmonic function. Both T and its partial derivatives,
∂T /∂x, ∂T /∂y, ∂T /∂z, are harmonic, because if
∂2T ∂2T ∂2T
+ + = 0, (6–50)
∂x2 ∂y 2 ∂z 2
then we also have

∂ 2 ∂T ∂ 2 ∂T ∂ 2 ∂T ∂ ∂2T ∂2T ∂2T
+ 2 + 2 = + + = 0.
∂x2 ∂x ∂y ∂x ∂z ∂x ∂x ∂x2 ∂y 2 ∂z 2
(6–51)
Thus, the upward continuation integral (6–49), which applies for any har-
monic function, may also be applied to ∂T /∂x, ∂T /∂y, and ∂T /∂z.
As T is the disturbing potential, its partial derivatives are the compo-
nents of the gravity disturbance:
∂T ∂T ∂T
= δgϕ̄ , = δgλ , = δgr . (6–52)
∂x ∂y ∂z
We are not writing δgx , δgy , δgz because we wish to reserve this notation
for the components in the geocentric global coordinate system, which should
not be confused with the local system introduced in this section. As usual,
r, ϕ̄, λ denote geocentric spherical coordinates (see Sect. 6.3) corresponding
to the spherical approximation.
Thus, we have in addition to (6–49)

∞
H δgr
δgr = dx dy , (6–53)
2π l3
−∞
6.5 Additional considerations 249

∞
H δgϕ̄
δgϕ̄ = dx dy ,
2π l3
−∞
(6–54)
∞

H δgλ
δgλ = dx dy .
2π l3
−∞
On the left-hand side of these equations, the components of δg refer to the

elevated point P ; in the integral on the right-hand side, they are taken at
sea level and are to be computed from the expressions

2γ0
δgr = −δg = − ∆g + N , (6–55)
R
δgϕ̄ = −γ0 ξ ,
(6–56)
δgλ = −γ0 η ,
which follow from (2–264) together with (6–42) and (6–43) applied to sea
level. The symbols R and γ0 denote, as usual, a mean earth radius and a
mean value of gravity on the earth’s surface.
Hence, we may compute T and δg by means of the upward continuation
integral if the geoidal undulations N and the deflection components ξ and η
at the earth’s surface are given.
The plane approximation is sufficient except for very high altitudes (e.g.,
> 250 km). Otherwise, we must use the spherical formula (6–44) for T . For
the radial component δgr , formula (6–44) may also be applied with T re-
placed by r δg, since r δg and r ∆g are harmonic as we know from Sect. 2.14.
The corresponding spherical formulas for the upward continuation of the hor-
izontal components δgϕ̄ and δgλ are not known. The reason why the same
formula, the upward continuation integral, applies for T and the components
of δg in the planar case only is that the derivatives of T are harmonic only
when referred to a Cartesian coordinate system.
6.5 Additional considerations

Reference surface
The preceding formulas for the disturbing potential T and the gravity dis-
turbance vector δg are rigorously valid if the reference surface is a sphere.
In practice, the gravity anomalies are referred to an ellipsoid. The above
formulas for T and δg are also valid for an ellipsoidal reference surface if a
.
relative error of the order of the flattening f = 0.3% is neglected, that is, as
a spherical approximation. The reader is reminded that this does not mean
that the ellipsoid is replaced by a sphere in any geometrical sense; rather

it means that in the originally elliptical formulas the first and higher pow-
ers of the flattening are neglected, whereby they formally become spherical
formulas.
Since the gravity anomalies, etc., are referred to an ellipsoid, we must be
very careful in computing t, which enters into the formulas of Sect. 6.3. If
an exact sphere of radius R were used as a reference surface, then we should
have r = R + H, where H is the elevation of the computation point above
the sphere. Actually, we use a reference ellipsoid; then we again have
R
r =R+H, t= , (6–57)
R+H
but H is now the elevation above the ellipsoid (or, to a sufficient accuracy,
above sea level), the constant R = 6371 km being the earth’s mean radius.
Thus,r as computed by (6–57) differs from the geocentric radius vector
r = x2 + y 2 + z 2 . We have already mentioned that we may replace the
geocentric latitude ϕ̄ by the ellipsoidal latitude ϕ, as far as T and δg are
concerned – for instance, by putting ϕ̄ = ϕ in (6–26) or (6–29).
Data
For all computations dealing with the external gravity field of the earth,
free-air gravity anomalies must be used for ∆g, since all other types of grav-
ity anomalies correspond to some removal or transport of masses whereby
the external field is changed. If, in addition to ∆g, deflections of the vertical
ξ, η (in the upward continuation) are used, then these quantities should be
computed from free-air anomalies. If, as usually done, the normal free-air
.
gradient ∂y/∂h = 0.3086 mgal/m is used for the free-air reduction, then the
free-air anomalies refer, strictly speaking, to the earth’s physical surface (to
ground level) rather than to the geoid (to sea level). The N values com-
puted from them by Stokes’ formula are height anomalies ζ, referring to the
ground, rather than heights of the actual geoid. However, this distinction is
insignificant and can be ignored in most cases, so that we may consider ∆g
as sea-level anomalies (see Sect. 8.6).
If we cannot neglect this distinction, aiming at highest accuracy in high
and steep mountains for low altitudes H, then we may proceed as follows. We
reduce the free-air anomaly ∆g from the ground point A to the corresponding
point A0 at sea level (Fig. 6.3):
∂∆g
∆gharmonic = ∆g − h, (6–58)
∂h
and use the sea level anomaly ∆gharmonic so obtained. The vertical gradient
∂∆g/∂h may be computed by applying formula (2–394) using the ground-
6.5 Additional considerations 251
H1
A
H
h1 level of F
F h W = W1
sea level
P0 A0 W = W0
Fig. 6.3. Reduction to sea level and to the level of F
level anomalies ∆g. Or we may reduce to any other level surface W = W1 ,

for instance, to that passing through F (Fig. 6.3), using h1 instead of h in
(6–58). Then we should also use H1 , rather than H, in (6–57). For large-
scale purposes, reduction to sea level appears to be preferable. Probably such
a reduction will attain a considerable amount only in exceptional cases so
that it can usually be neglected and H in the formulas of Sects. 6.3 and 6.4
may be taken as the height of P above sea level or above ground. See also
Sect. 8.6.
Computation of the gravity vector

After computing the components δgr , δgϕ̄ , δgλ by numerical integration, we
may transform them into Cartesian coordinates δgx , δgy , δgz with respect
to the global coordinate system.
We may go via ellipsoidal-harmonic coordinates according to Sect. 6.2.
For the small quantities δgu , δgβ , δgλ , we may apply the spherical approxi-
mation, neglecting a relative error of the order of the flattening. If the flat-
tening is neglected, then the ellipsoidal-harmonic coordinates u, β, λ reduce
to the spherical coordinates r, ϕ̄, λ so that as a spherical approximation
δgu = δgr , δgβ = δgϕ̄ , (6–59)
δgλ being rigorously the same in both systems. Thus, δgr , δgϕ̄ , δgλ may also
be considered as the components of δg in ellipsoidal-harmonic coordinates.
Then we have
gu = γu + δgr , gβ = γβ + δgϕ̄ , gλ = δgλ ; (6–60)
and gx , gy , gz are obtained by (6–17), the components of g replacing the cor-

responding components of γ. It is evident that the spherical approximation
can only be used for δg so that γu and γβ must be computed by the rigorous
formulas (6–12).
The gravity potential W may be computed by the first equation of (6–
4); the gravitational potential V is obtained by subtracting the centrifugal
potential ω 2 (x2 + y 2 )/2; and the vector of gravitation is given by (6–5).
6.6 Gravity anomalies and disturbances compared

Suppose gravity g is to be computed at some point P outside the earth
(Fig. 6.4); we consider here only the magnitude of the gravity vector. This
is conveniently done by adding a correction to the normal gravity γ. From
Sect. 2.12 and later, we recall the two different kinds of such a correction,
g − γ:
1. the gravity disturbance δg, in which g and γ both refer to the same
point P ;
2. the gravity anomaly ∆g, in which g refers to P , but γ refers to the
corresponding point Q, which is situated on the plumb line of P and
whose normal potential U is the same as the actual potential W of P ,
that is, UQ = WP .
These two quantities are connected by
2γ0
∆g = δg − NP ; (6–61)
R
this simple relation is sufficient for moderate altitudes.
The gravity disturbance is used when the spatial position of P is given,
that is, its geocentric rectangular coordinates x, y, z are measured. With
GPS measurements of the position of the aircraft, the use of gravity distur-
bances is natural.
The use of gravity anomalies ∆g had been traditional. This is the case, for
instance, in airborne gravity measurements, where the height of the aircraft
P W = WP
NP
U = WP
Q
H1
F earth's
surface
Fig. 6.4. Gravity anomalies and disturbancies

6.6 Gravity anomalies and disturbances compared 253
above ground is measured. This case seems rather to belong to the past. If
the case should arise, gravity anomalies ∆g can be upward continued just as
δg as described in Sect. 6.4.
Again, free-air anomalies referred to ground level or, more accurately,
to some level surface, are to be used. If the ground is elevated above sea
level but reasonably flat, it is somewhat better to regard H as elevation
above ground rather than above sea level, because the ground may then be
considered locally part of a level surface.
The inverse problem, the downward continuation of gravity anomalies or
rather gravity disturbances, occurs in the reduction of gravity measured on
board an aircraft. There is, of course, a relation to harmonic downward con-
tinuation in the solution of Molodensky’s problem as described in Sect. 8.6.
Upward and downward continuation are also tools of geophysical explo-
ration, but here the objective is quite different. Several methods have been
developed in this connection, some of which are also applicable for geodetic
purposes; see, e.g., Dobrin and Savit (1988) or Telfort et al. (1990).
Upward and downward continuation are related as direct and inverse
problems in the theory of inverse problems, see Anger et al. (1993) and also
www.inas.tugraz.at under forschung/InverseProblems/AngerMoritz.html,
where additional references can be found.
7 Space methods
7.1 Introduction
The subject of this chapter is the use of satellite observations for determining
features of the gravity field and the figure of the earth. Only the barest
essentials can be presented within the scope of a chapter. The reader will
find more information in special textbooks such as Hofmann-Wellenhof et
al. (2001), Montenbruck and Gill (2001), and Seeber (2003).
Historical remarks
Immediately after the first launch of artifical satellites (Sputnik 1957, Ex-
plorer 1958), their use for geodetic purposes was initiated, and by now the
Global Positioning System (GPS) has become the most important method
for a fast and precise determination of geodetic positions (see Sect. 5.3).
Historically, the first observational methods were intended to determine the
spatial direction and the distance to the satellite. Most of these methods are
now obsolete, but some principles may be still useful.
Directions
They may be measured by photographing the satellite against the back-
ground of stars, or by means of radio waves transmitted from the satellite,
using the principle of interference. Photography can only achieve an accuracy
of about 0.2 arc seconds and is not used any more in its original sense. The
principle of the photographic method was as follows. On the photographic
plate, the image of the satellite is surrounded by images of stars. The direc-
tions to the surrounding stars are defined by their right ascensions α and
declinations δ, which are known from astronomy. Therefore, by interpolation
we find the right ascension and declination of the satellite representing the
desired direction. This technique is now obsolete.
Ranges
They are measured by radar or by laser. Radar is used for measuring ranges
to space probes orbiting in the solar system, which is important to space
sciences rather than to geodesy. Lunar Laser Ranging (LLR) and Satellite
Laser Ranging (SLR) are useful for determining the earth rotation param-
eters because of their high (subcentimeter) accuracy; however, their use is
restricted to a limited number of fundamental stations.
256 7 Space methods
Range rates
This measurement quantity is found by observing the Doppler effect with

radio waves transmitted from a satellite. It is still used within GPS and in
satellite-to-satellite tracking (SST).
Satellite altimetry
Here a short-wave electronic ray is sent, from a satellite flying over the
oceans, vertically down to the ocean surface, reflected there and received by
the satellite again. The measured travel time immediately gives the height
H of the satellite above the ocean surface. Knowing the orbital position of
the satellite with respect to the global reference system, we can compute the
satellite height h above the ellipsoid. Then the difference h− H is the geoidal
height N . This is the case if the sea surface is assumed to coincide with the
geoid. In reality, because of ocean currents, etc., both surfaces are separated
by the “sea surface topography”, which may reach the order of 1 m and is
interesting to oceanography. It can be determined if an accurate ocean geoid
is known from the gravitational field.
The principles of these methods are illustrated in Fig. 7.1, where e indi-
cates the direction observation, s between tracking station and satellite refers
to the range measurement, and, accordingly, ds/dt corresponds to Doppler
observation, whereas ds/dt between the two satellites is obtained by SST;
finally, H is measured by satellite altimetry.
satellite
satellite
s
H
tracking
station e
geoid
land W = W0
ocean
Fig. 7.1. Principles of satellite techniques

7.2 Satellite orbits 257
7.2 Satellite orbits

The first spectacular result from satellite observations, well advertised by
NASA around 1960, was the “discovery” that “the earth is not an ellipsoid
but rather shaped like a pear”. This pear shape is caused by the spherical
harmonic J3 . Its effect, at the North and South Poles, is on the order of
30 m, by three orders of magnitude less than the ellipticity coming from J2 ,
whose linear effect a − b is about 20 km (!).
The first real result, also found around 1960, was a dramatic improvement
in the accuracy of the flattening f itself, which lead to a change from 1/297.3,
generally believed before, to 1/298.25, corresponding to a linear improvement
of the earth size of about 70 m!
The earth’s flattening causes the largest but not the only deviation of the
earth gravitational field from that of a homogeneous sphere. Generally, the
gravitational potential can be expanded into a series of spherical harmonics
according to Sect. 2.5, Eq. (2–78):
∞
GM a n
V = 1− Jn Pn (cos ϑ)
r n=2
r
∞ n
a n
+ [Cnm cos mλ + Snm sin mλ] Pnm (cos ϑ) .
r
n=2 m=1
(7–1)
Here the terms containing Jn are the zonal harmonics, and those containing
Snm and Cnm are the tesseral harmonics.
The former notations Jnm = −Cnm and Knm = −Snm are not used any
more for the tesseral harmonic coefficients; for the zonal harmonics, the use
of Jn has prevailed so far, but also Cn0 = −Jn is being used.
Considering the moon, the only term of appreciable influence is J2 , which
represents the flattening. Artificial satellites are, compared to the moon,
much closer to the earth; typical heights above ground of a geodetically
used satellite range from some 300 km up to 20 000 km. Hence, they are
also influenced by harmonics other than J2 and can, therefore, be used to
determine harmonics of low degree. For this purpose, we must study the
effect of gravitational disturbances on the orbits of close satellites.
Before we can do this, we must briefly review the theory of an undisturbed
orbit, which means that the gravitational potential has the form
GM
V = , (7–2)
r
all C s and S s being zero. This represents the gravitational field of a point
mass or a homogeneous sphere. Then the motion of a satellite is described
258 7 Space methods
Z0
K'
line of nodes S
À
P
A ! t
bi
or i or
W equa
t
vernal equinox
K
X0
Fig. 7.2. Satellite orbit as projected onto a unit sphere
by Kepler’s three laws for planetary motion. Satellites with parabolic or

hyperbolic orbits are of no interest in this context.
According to Kepler’s first law, the orbit is an ellipse of which the center
of the earth occupies one focus. The position of the orbit in space is defined
by the six orbital elements:
a semimajor axis,
e eccentricity,
i inclination,
(7–3)
Ω right ascension of the node,
ω argument of perigee,
T time of perigee passage .
If a and b are the semiaxes of the orbital ellipse (there is no danger of

confusion with those of the terrestrial ellipsoid!), then the eccentricity is
defined by √
a2 − b2
e= . (7–4)
a
Figure 7.2 shows the projection of the orbit onto a geocentric unit sphere,
where P is the perigee, A the apogee, K is the ascending node, K the
descending node, S is the instantaneous position of satellite. The line of
nodes is the intersection of the orbital plane with the plane of the equator;
it connects the ascending node K and the descending node K . The right
ascension of the node, Ω, is the angle between the line of nodes and the
direction to the vernal equinox. The symbol Ω is also called longitude of
7.2 Satellite orbits 259
the node, but in conformity with astronomical terminology it is the right

ascension of the (ascending) node. The major axis of the orbit intersects the
orbital ellipse at the perigee P , the position where the satellite is closest to
the earth, and at the apogee A, where the satellite is farthest away. The
angle ω between the nodes and the major axis is the argument of perigee.
The angular distance of the satellite S from perigee is called true anomaly
and denoted by v; it is a function of time. Note that this strange name comes
from the history of astronomy; there is nothing anomalous with it!
The equation of the orbital ellipse may be written
p
r= , (7–5)
1 + e cos v
where r is the distance of the satellite from the earth’s center of mass and
b2
p= = a (1 − e2 ) (7–6)
a
is the length of the radius vector r for v = 90◦ . The radius vector r and
the true anomaly v form a pair of polar coordinates in the orbital plane,
and (7–5) is the well-known polar equation of an ellipse. See Fig. 7.3 for an
illustration of these quantities, where F , the focal point, is the earth’s center
of mass.
According to Kepler’s second law, the area of the elliptical sector swept by
the radius vector r between any two positions of the satellite is proportional
to the time it takes the satellite to pass from one position to the other. In
other words, the time rate of change of the area swept by the radius vector
is constant. Since the element of area of a sector in polar coordinates r and
e2
S
b p
r
A ae À P
a F r0 e1
Fig. 7.3. Orbital ellipse

260 7 Space methods
v is 12 r 2 dv, this law may be formulated mathematically as

dv
r2 = GM a (1 − e2 ) , (7–7)
dt
where the constant has already been given its proper value.
Kepler’s third law reads
n2 a3 = GM , (7–8)
where the satellite mass has been neglected and where
2π
n= (7–9)
P
is the “mean motion” (mean angular velocity) of the satellite, P being its
period.
So far we have assumed that all Jn , Cnm , and Snm in (7–1) are zero. This
is not true because of the irregularities of the earth’s gravitational field, even
though these coefficients are small. Therefore, the satellite is subject to small
perturbing forces. We may still consider the satellite orbit as an ellipse, but
then the parameters of this ellipse, the orbital elements, will no longer be
constant but will change slowly. At each instant, this osculating ellipse will
be slightly different. It is defined as follows. Imagine that at the instant
under consideration all perturbing forces suddenly vanish. Then the satellite
will continue its motion along an exact ellipse; this is the osculating ellipse.
If we resolve the total perturbing force into rectangular components S,
T , and W , where S is directed along the radius vector, W is normal to the
orbital plane, and T is normal to S and W – note that this notation follows
astronomical usage; there is no relation to the geodetic use of T and W for
potentials! –, then the time rate of change of the orbital parameters can be
expressed in terms of these components:
+
2a2 a p
ȧ = e S sin v + T ,
b GM r
+
b a r+p er
ė = S sin v + cos v + T ,
a GM r p
+
r a
ı̇ = W cos(ω + v) ,
b GM
+
r a sin(ω + v)
Ω̇ = W ,
b GM sin i
+
b a 1 r+p r
ω̇ = − S cos v + T sin v − W sin(ω + v) cot i .
a GM e ep p
(7–10)
7.3 Determination of zonal harmonics 261
As usual, ȧ denotes da/dt, etc. The derivation of these equations may be

found in any textbook on celestial mechanics, e.g., Plummer (1918: p. 151),
Brouwer and Clemence (1961: p. 301), and Seeber (2003: Sect. 3.2.1.3), who
uses the symbols K1 , K2 , K3 instead of W, S, R.
7.3 Determination of zonal harmonics

The effect of the zonal harmonics on satellite orbits is much greater than that
of the tesseral harmonics. Only zonal harmonics (J2 , J3 , J4 , . . .) will give
observable variations of the orbital elements themselves. The tesseral har-
monics cause oscillatory disturbances that rapidly change their sign, whereas
the effect of the zonal harmonics is cumulative. For this reason, we consider
first the effect of zonal harmonics, that is, the effect of those independent of
longitude λ. Hence we set
GM
V = + R, (7–11)
r
where the perturbing potential
G M ae n+1
∞
R=− Jn Pn (cos ϑ) (7–12)
ae n=2 r
is a function of r and ϑ only. Note that the main difference between the
perturbing potential R of celestial mechanics and the disturbing potential
T of physical geodesy is that R, but not T , also incorporates the effect
of the flattening through J2 . There are also other perturbing forces acting
on a satellite, such as the resistance of the atmosphere (atmospheric drag),
radiation pressure exerted by the sunlight, etc. These nongravitational per-
turbances must be taken into account separately and will not be considered
here.
Note that the equatorial radius of the earth (the semimajor axis of the
terrestrial ellipsoid) has been denoted by ae , in order to distinguish it from
a, which now denotes the semimajor axis of the orbital ellipse. This notation
will be used in what follows.
Since S is the component of the perturbing force along the radius vector,
we have
∂R
S= . (7–13)
∂r
The components of the perturbing force along the meridian and the prime
vertical are
1 ∂R 1 ∂R
− and . (7–14)
r ∂ϑ r sin ϑ ∂λ
262 7 Space methods
pole
–1 @R
r @#
W
® T
satellite 1 @R
r sin # @¸
! +À ®
90° – #
i
node
Fig. 7.4. Components of the perturbing force
The components T and W are obtained from them by a plane rotation

(Fig. 7.4):
1 ∂R 1 ∂R
T =− cos α + sin α ,
(7–15)
1 ∂R 1 ∂R
W =− sin α − cos α .
From the rectangular spherical triangle in Fig. 7.4 it follows that
cos(ω + v) sin i cos i

cos α = , sin α = , (7–16)
sin ϑ sin ϑ
so that finally
cos(ω + v) sin i ∂R cos i ∂R

T =− + ,
r sin ϑ ∂ϑ r sin2 ϑ ∂λ
(7–17)
cos i ∂R cos(ω + v) sin i ∂R
W =− − .
r sin ϑ ∂ϑ r sin2 ϑ ∂λ
We have included ∂R/∂λ because of the presence of longitude-dependent
tesseral harmonics in the general case (see Sect. 7.5). In our present case,
where R is given by (7–12), ∂R/∂λ is zero.
Now we must differentiate (7–12) with respect to r and ϑ, compute the
components S, T , W from Eqs. (7–13) and (7–17), and substitute them
into the system (7–10). In this way, we can express the rates of change
ȧ, ė, . . . of the orbital elements in terms of the coefficients J2 , J3 , J4 , . . . . We
cannot, however, observe these rates of change directly. Rather, we observe

the changes of the orbital elements after several revolutions. The changes
after one revolution, with period P , are
t0 +P t0 +P t0 +P
∆a = ȧ dt , ∆e = ė dt , ∆i = ı̇ dt , etc. (7–18)
t0 t0 t0
The t0 is an arbitrary “epoch” (instant of time). In order to perform these

integrations, we must express ȧ, ė, . . . in terms of one independent variable.
For this independent variable, we may take the time t or the true anomaly
v. The second possibility will be adopted here.
The polar distance ϑ is expressed as a function of v through the relation
cos ϑ = sin(ω + v) sin i , (7–19)
which follows from the rectangular spherical triangle in Fig. 7.4. The radius
vector r is also a function of v according to (7–5). Finally, Kepler’s second
law (7–7) furnishes the relation between v and the time t:
dt r2
= . (7–20)
dv G M a (1 − e2 )
Hence, we may change the integration variable from t to v, obtaining, for

instance,
t0 +P 2π
da
∆a = ȧ dt = dv , (7–21)
t0 v=0 dv
where
da da dt r2
= = ȧ . (7–22)
dv dt dv G M a (1 − e2 )
Analogous formulas result for the other orbital elements.

After performing all these operations, which are lengthy but not too
264 7 Space methods
difficult, we find
∆a = 0 ,
1 − e2
∆e = − tan i ∆i ,
e
3
ae 5
∆i = 3π e 1 − sin2 i cos i cos ω J3
p 4
4
45 ae 7
+ πe 1 − sin i sin 2i sin 2ω e J4 · · · ,
2
16 p 6
2
ae
∆Ω = −3π cos i J2
p
3
ae 15
+ 3π 1− 2
sin i cot i sin ω e J3
p 4 (7–23)
4
15 ae 7
+ π 1 − sin2 i cos i J4 · · · ,
2 p 4
2
ae 5
∆ω = 6π 1 − sin2 i J2
p 4
3
ae 5
+ 3π 1 − sin i sin i sin ω e J3
2
p 4
4
ae 31 49
− 15π 1− 2
sin i + 4
sin i
p 8 19

3 7
+ − sin2 i sin2 i cos 2ω J4 · · · .
8 16
Terms of the order of e2 J3 and e2 J4 , which are very small, have been ne-
glected in these equations. The proportionality of ∆e and ∆i is more or less
accidental: it applies only with respect to long-periodic disturbances; ė and
di/dt themselves are not proportional. The quantity p is defined by (7–6);
it is hardly necessary to repeat that a, p, e, etc., refer to the orbital ellipse
and not to the terrestrial ellipsoid, of which ae is the equatorial radius.
By integrating over one revolution, we have removed the short-periodic
terms of periods P, 2P, 3P, . . . , such as cos v, cos 2v, etc. What remains are
secular terms, which are constant for one revolution and increase steadily
with the number of revolutions, and the long-periodic terms, which change
very slowly with time in a periodic manner. The argument of perigee ω
increases slowly but steadily, so that the perigee of a satellite orbit also
rotates around the earth, but much slower than the satellite itself; a typical
period of ω is two months. Therefore, terms containing cos ω, sin ω, or sin 2ω
are called long-periodic.
The first equation of (7–23) shows that the semimajor axis of the orbit
does not change secularly or long-periodically. The eccentricity and the in-
clination undergo long-period, but not secular, variations, whereas Ω and ω
change both secularly and long-periodically.
Equations (7–23) are linear in J2 , J3 , J4 , . . . . For practical applications,
nonlinear terms containing J22 , J2 J3 , J2 J4 , etc., must also be taken into ac-
count, since J22 is of the order of J4 . The derivation of these nonlinear terms
is much more difficult, and their expressions are different in the various or-
bital theories that have been proposed. For these reasons, such expressions
will not be given here.
Equations (7–23), supplemented by certain nonlinear terms, can be used
to determine coefficients J2 , J3 , J4 , etc. Since the secular or long-periodic
variations ∆Ω, ∆ω, ∆e, ∆i are known from observation for a sufficient num-
ber of satellites, we obtain equations of the form
a2 J2 + a3 J3 + a4 J4 + · · · + a22 J22 + a23 J2 J3 + · · · = A ,
b2 J2 + b3 J3 + b4 J4 + · · · + b22 J22 + b23 J2 J3 + · · · = B, (7–24)

.. .. .. .. .. ..
. . . . . .
which can be solved for J2 , J3 , J4 , . . . . Since there can be only a finite

number of these equations, we must neglect all Jn with n greater than a
certain number n0 , which depends on the number of equations available, on
their degree of mutual independence, etc. This used to be a difficulty with
this method, but it has been overcome long ago by least-squares collocation
(Moritz 1980 a: Sect. 21). For details see Schwarz (1976).
From (7–23) it is seen that the coefficients of the Jn depend essentially
on the inclination i. It is, therefore, important to use satellites with a wide
variety of inclinations, in order to get equations with a high mutual inde-
pendence.
Now the question arises which orbital elements are to be used for deter-
mining the coefficients Jn . The semimajor axis a clearly cannot be used at
all. As for the other elements, we must distinguish between coefficients of
even and of odd degree n. The even coefficients J2 , J4 , . . . can be determined
well from the regression of the node, ∆Ω, and the rotation of perigee, ∆ω.
To see this, inspect (7–23). The even harmonics cause secular disturbances
of Ω and ω, which are much larger than the long-periodic effects of the odd
coefficients, since J3 , J5 , . . . are multiplied by the small eccentricity e.
266 7 Space methods
On the other hand, in ∆e and ∆i the odd coefficients J3 , J5 , . . . have

a much larger effect than the even coefficients, which here appear with the
small factor e. Therefore, the odd coefficients are determined from ∆e or ∆i,
or from the change of perigee distance r0 = F P (Fig. 7.3). Since r0 is the
radius vector for v = 0, we have from (7–5) and (7–6)
p
r0 = = a (1 − e) , (7–25)
1+e
so that
∆r0 = −a ∆e (7–26)
because ∆a = 0. Thus, the variation of perigee distance is proportional to
the variation of eccentricity and may be used instead of ∆e.
Numerical values
Helmert (1884: p. 472) used the regression of the node of the moon’s orbit
to determine J2 , which is the only coefficient to have an appreciable effect
. .
on it. Note that for e = 0 and p = a
ae , the equation for ∆Ω in (7–23)
becomes a 2
e
∆Ω = −3π J2 cos i . (7–27)
a
Helmert found
J2 = 1086.5 · 10−6 (7–28)
by averaging two widely different values. This corresponds to a flattening of
1/f = 297.8 ± 2.2 . (7–29)
This value is quite close to the recent results but has a much larger uncer-
tainty.
Reliable values by this method can only be obtained from close artificial
satellites. Currently accepted values are, for example,
J2 = 1082.6359 · 10−6 ,
J3 = −2.5324 · 10−6 , (7–30)
J4 = −1.6198 · 10−6 ,
whose standard errors are assumed to be better than ±0.01 · 10−6 . The value
for J2 has been taken from the report of the IAG by Groten (2004), accessible
from www.gfy.ku.dk/∼iag/HB2004/part5/51-groten.pdf. J3 and J4 are from
the recent mission GRACE (see Sect. 7.5).
7.4 Rectangular coordinates of the satellite and perturbations 267
The most significant geodetic result is the reliable determination of J2

and, therefore, of the flattening f , around 1/298.25. Already in 1964, the In-
ternational Astronomical Union (IAU) adopted the value 298.25 correspond-
ing to J2 = 1082.7 · 10−6 (see Sect. 2.11), followed by the IAG International
Geodetic Reference Systems 1967 and then 1980, which in the slightly dif-
ferent form of the World Geodetic System 1984 (WGS 84) is standard even
today (2005).
7.4 Rectangular coordinates of the satellite and

perturbations
We now describe how the rectangular coordinates of the satellite are com-
puted from the orbital elements. Then we will outline how they are affected
by the irregularities of the gravity field. These considerations are necessary
for the determination of tesseral harmonics from satellite observations.
We introduce an equatorial coordinate system X 0 Y 0 Z 0 that is at rest
with respect to the stars. The origin is at the earth’s center of mass. The
Z 0 -axis coincides with its axis of rotation; the X 0 Y 0 -plane is the equatorial
plane. The X 0 -axis is the line of intersection of the equatorial plane and
the ecliptic (the plane of the earth’s orbit around the sun); according to
astronomical terminology, it points to the vernal equinox. This coordinate
system X 0 Y 0 Z 0 is fundamental in spherical astronomy. Note that the di-
rections of the coordinate axes so defined are not completely constant in
time. This fact requires certain refinements for which the reader is referred
to Moritz and Mueller (1987: Chap. 7). In the present context, we consider
the X 0 Y 0 Z 0 -system as constant in time.
The relation between the rectangular coordinates of a satellite and the
elements of its osculating ellipse (Sect. 7.2) at a certain time is found as fol-
lows. Consider Fig. 7.3 and the coordinate system e1 , e2 defining the orbital
plane. Assuming e3 orthogonal to this plane,
⎡ ⎤
cos v
⎢ ⎥
r⎢ ⎥
⎣ sin v ⎦ (7–31)
0
is the representation of the satellite in this system. This result may be trans-
formed into the equatorial system X 0 Y 0 Z 0 by a rotation matrix R and re-
sults in a vector denoted as X0 = [X 0 , Y 0 , Z 0 ]. The transformation is
268 7 Space methods
obtained by ⎡ ⎤ ⎡ ⎤
X0 cos v
⎢ 0 ⎥ ⎢ ⎥
⎢ Y ⎥ = R r ⎢ sin v ⎥ , (7–32)
⎣ ⎦ ⎣ ⎦
Z 0 0
where the matrix R is composed of three successive rotation matrices (see
Figs. 7.2 and 7.3) and is given by
R = R 3 {−Ω} R1 {−i} R 3 {−ω}

⎡ ⎤
cos Ω cos ω − cos Ω sin ω sin Ω sin i
⎢ − sin Ω sin ω cos i − sin Ω cos ω cos i ⎥
⎢ ⎥
⎢ ⎥
= ⎢
⎢ sin Ω cos ω − sin Ω sin ω − cos Ω sin i ⎥ , (7–33)
⎥
⎢ + cos Ω sin ω cos i + cos Ω cos ω cos i ⎥
⎣ ⎦
sin ω sin i cos ω sin i cos i
see Hofmann-Wellenhof et al. (2001: p. 43). The column vectors of the or-
thonormal matrix R are the axes of the orbital coordinate system represented
in the equatorial system X0i .
Substituting (7–33) into (7–32) and carrying out the multiplication (Mon-
tenbruck and Gill 2001: Eq. (2.51)) yields
X 0 = r [cos Ω cos(ω + v) − sin Ω sin(ω + v) cos i] ,
Y 0 = r [sin Ω cos(ω + v) + cos Ω sin(ω + v) cos i] , (7–34)
Z 0 = r sin(ω + v) sin i ,
where, according to (7–5),
a (1 − e2 )
r= . (7–35)
1 + e cos v
This expresses the rectangular coordinates of the satellite in terms of the
elements of its osculating orbit, the true anomaly v fixing its position as a
function of time.
Since the osculating ellipse does not remain constant, it is convenient
to use a fixed reference orbit – for instance, the osculating ellipse E0 at
a certain instant t0 , having the elements a0 , e0 , i0 , Ω0 , ω0 , T0 . At a later
instant t, the orbital elements will have changed to a0 + ∆t a, e0 + ∆t e, i0 +
∆t i, Ω0 + ∆t Ω, ω0 + ∆t ω, T0 + ∆t T , which corresponds to an osculating
ellipse Et .
7.4 Rectangular coordinates of the satellite and perturbations 269
The orbital elements in (7–34) refer to this instantaneous osculating el-

lipse, so that a = a0 + ∆t a, etc. Therefore, the coordinates X 0 , Y 0 , Z 0
depend on the time in two ways: explicitly, through the true anomaly v, and
implicitly, through the variable elements of the osculating orbit. We elimi-
nate the implicit dependence in the following way. We evaluate (7–34) using
the elements a0 , etc., of the fixed reference ellipse. Then the coordinates so
obtained depend on the time only explicitly and correspond to a Keplerian
motion in space along a fixed ellipse. To convert them into true coordinates
X 0 , Y 0 , Z 0 , they must be corrected by ∆t X 0 , ∆t Y 0 , ∆t Z 0 , for which the
linear terms of a Taylor expansion of (7–34) give
∆t X 0 =
∂X 0 ∂X 0 ∂X 0 ∂X 0 ∂X 0 ∂X 0
∆t a ∆t e + ∆t i + ∆t Ω + ∆t ω + ∆t v ,
∂a ∂e ∂i ∂Ω ∂ω ∂v
∆t Y 0 =
∂Y 0 ∂Y 0 ∂Y 0 ∂Y 0 ∂Y 0 ∂Y 0
∆t a ∆t e + ∆t i + ∆t Ω + ∆t ω + ∆t v ,
∂a ∂e ∂i ∂Ω ∂ω ∂v
∆t Z 0 =
∂Z 0 ∂Z 0 ∂Z 0 ∂Z 0 ∂Z 0 ∂Z 0
∆t a ∆t e + ∆t i + ∆t Ω + ∆t ω + ∆t v .
∂a ∂e ∂i ∂Ω ∂ω ∂v
(7–36)
The partial derivatives are readily obtained by differentiating (7–34); note
that r is a function of a, e, and v.
In these equations, we have used the perturbation of the true anomaly,
∆t v, instead of the perturbation of perigee epoch, ∆t T .
Perturbations expressed in terms of Cnm and Snm

The perturbations of the orbital elements are found by integrating (7–10):
t t
∆t a = ȧ dt , ∆t e = ė dt , . . . . (7–37)
t0 t0
A similar expression can be written for ∆t v. The components S, T , W of the
perturbing force are expressed in terms of Jn , Cnm , and Snm using equations
(7–12), (7–13), and (7–17), where the perturbing potential

G M ae n+1
∞
R=− Jn Pn (cos ϑ)
ae r
n=2
n (7–38)
− (Cnm cos mλ + Snm sin mλ) Pnm (cos ϑ)
m=1
270 7 Space methods
now also contains the tesseral harmonics.

By performing the integrations in (7–37), we obtain equations of the form

∆t a = Anm Cnm + Ānm Snm ,
n,m

∆t e = Bnm Cnm + B̄nm Snm , (7–39)
n,m
.. ..
. .
where the coefficients Anm , etc., are functions of the time t and are, as a
rule, periodic. Zonal and tesseral harmonics have been combined in (7–39)
by setting Jn = −Cn0 and admitting the value m = 0; this practice will be
continued in what follows.
The substitution of (7–39) into (7–36) gives the perturbation of the rect-
angular coordinates X 0 , Y 0 , Z 0 as functions of the harmonic coefficients
Cn0 = −Jn , Cnm , and Snm in the form

∆t X 0 = Lnm Cnm + L̄nm Snm ,
n,m

∆t Y 0 = Mnm Cnm + M̄nm Snm , (7–40)
n,m

∆t Z 0 = Nnm Cnm + N̄nm Snm ,
n,m
where again Lnm , L̄nm , Mnm , etc., are functions of the time t.
These perturbations are added to the coordinates computed from (7–34)
using the orbital elements of the reference ellipse E0 . In this way, we obtain
the rectangular coordinates of the satellite in the form
X 0 = X 0 (t; a0 , e0 , i0 , Ω0 , ω0 , T0 ; Cnm , Snm ) ,
Y 0 = Y 0 (t; a0 , e0 , i0 , Ω0 , ω0 , T0 ; Cnm , Snm ) , (7–41)
Z 0 = Z 0 (t; a0 , e0 , i0 , Ω0 , ω0 , T0 ; Cnm , Snm )
as explicit functions of the time t, containing as constant parameters the
orbital element of the reference ellipse E0 and the gravitational coefficients
Cnm and Snm . This is the advantage of (7–41) over the system (7–34), which
formally is much simpler but depends on the variable orbital parameters of
the osculating ellipse.
The actual expressions for (7–41) are very complicated. Therefore, we
have been satisfied with outlining the procedure, referring the reader for
details to the pioneering book by Kaula (1966 a) and to his papers given
there.
7.5 Determination of tesseral harmonics and station positions 271
7.5 Determination of tesseral harmonics and

station positions
Zonal harmonics give rise to secular and long-periodic perturbations of the

orbital elements a, e, etc. Therefore, their influence can be detected in
changes of orbital parameters obtained by integrating over many revolutions
of the satellite.
The perturbations due to tesseral harmonics have a much shorter period.
The longest period of a harmonic of the order m = 1 is one day, for m = 2 it
is only half a day, etc. Therefore, we must look for another method, which is
sensitive enough to detect even short-periodic effects and extracts as much
information as possible from the observations.
The observed elements are essentially spatial polar coordinates of the
satellite with respect to the observing station: the distance s and the direc-
tion as determined by two angles. Corresponding to our coordinate system
X 0 , Y 0 , Z 0 introduced in the preceding section, these two angles are the
right ascension α and the declination δ, whose definition may be seen in
Fig. 7.5. The angles α and δ are polar coordinates in three-dimensional space
and were obtained by photographing the satellite against the background of
stars, as outlined in Sect. 7.1. They are outdated nowadays but retained for
geometrical intuition and symmetry. Most important are distances s mea-
sured by GPS, radar, or laser. Note that the measurement of the range rate
ds/dt of the satellite by means of the Doppler effect is also important for the
determination of tesseral harmonics and station positions.
Denoting in the equatorial system X 0 Y 0 Z 0 the rectangular coordinates of
the terrestrial station P by XP0 , YP0 , ZP0 and of the satellite S by XS0 , YS0 , ZS0 ,
||Z 0 Z0 unit sphere
S S
Z0
P ± ZS0 –ZP0
P ± ±
® ||Y 0 Y0
0 XS0 –XP0 ®
||X 0
Y –Y
S P
0
X0 ®
0
Y
X0
earth
Fig. 7.5. Direction to the satellite defined by right ascension α and

declination δ
272 7 Space methods
we find by inspecting Fig. 7.5

XS0 − XP0 = s cos δ cos α ,
YS0 − YP0 = s cos δ sin α , (7–42)
ZS0 − ZP0 = s sin δ ,

so that
YS0 − YP0
α = tan−1 ,
XS0 − XP0
ZS0 − ZP0
δ = tan−1 , (7–43)
(XS0 − XP0 )2 + (YS0 − YP0 )2

s = (XS0 − XP0 )2 + (YS0 − YP0 )2 + (ZS0 − ZP0 )2 .
We now compute the rectangular coordinates XP0 , YP0 , ZP0 of the observ-
ing station P . The system X 0 Y 0 Z 0 , being fixed with respect to the stars,
rotates with respect to the earth. The coordinates of P in this system are,
therefore, functions of time. Let XP , YP , ZP be the coordinates of P in the
usual geocentric coordinate system fixed with respect to the earth. In this
system, the Z-axis, coinciding with the Z 0 -axis, is the earth’s axis of rota-
tion; the X-axis lies in the mean meridian plane of Greenwich, corresponding
to the longitude λ = 0◦ ; and the Y -axis points to λ = 90◦ east. Figure 7.6
shows that
XP0 = XP cos θ0 − YP sin θ0 ,
YP0 = XP sin θ0 + YP cos θ0 , (7–44)
ZP0 = ZP .
The angle θ0 is called Greenwich sidereal time; its value is
θ0 = ω t , (7–45)
where ω is the angular velocity of the earth’s rotation. It is proportional
to the time t and, in appropriate units, measures it. Thus, absolute Green-
wich time is needed to convert the terrestrial coordinates XP , YP , ZP to the
celestial coordinates XP0 , YP0 , ZP0 that are required in (7–42) and (7–43).
As a final step, we substitute the station coordinates, as given by (7–44),
and the satellite coordinates, as symbolized by (7–41), into (7–43), obtaining
expressions of the form
α = α(XP , YP , ZP ; t; a0 , e0 , i0 , Ω0 , ω0 , T0 ; Cnm , Snm ) ,
δ = δ(XP , YP , ZP ; t; a0 , e0 , i0 , Ω0 , ω0 , T0 ; Cnm , Snm ) , (7–46)
s = s(XP , YP , ZP ; t; a0 , e0 , i0 , Ω0 , ω0 , T0 ; Cnm , Snm ) .
7.5 Determination of tesseral harmonics and station positions 273
Z 0= Z
Y
Z 0= Z X
Y0
Y 0 X0
P
Y q0 Y
q0
Y0 X
X0 X0
X
Fig. 7.6. Geocentric coordinate systems X 0 Y 0 Z 0 (celestial) and

XY Z (terrestrial)
Besides depending on the station coordinates and the time, they also contain
the orbital and gravitational parameters.
Every observation furnishes an equation of type (7–46). Provided we have
a sufficient number of such observation equations, we can solve them for the
station coordinates XP , YP , ZP , for the orbital parameters a0 , e0 , etc., of the
reference ellipse, and for a certain number of gravitational parameters Cnm
and Snm . This is the principle of the orbital method. In practice, differential
formulas will be applied to determine corrections to assume approximate val-
ues by means of a least-squares adjustment. Therefore, the actual analytical
developments are from the outset directed toward obtaining differential for-
mulas corresponding to (7–46). The substitutions indicated above are, thus,
consistently performed in terms of the corresponding differential expressions.
In this way we are able to operate with linear equations and to employ that
efficient tool of linear analysis, matrix calculus. Simple though the princi-
ple of this procedure is, the details when written out are nevertheless so
complicated that the reader must again be referred to the literature, e.g.,
Kaula (1966 a), Montenbruck and Gill (2001). Computer formula manipula-
tion is also used.
Besides these analytical problems, which have been satisfactorily solved,
the geodetic application of (7–46) raises difficulties similar in principle to
those involved in the determination of zonal harmonics by means of (7–24),
but even more serious in practice. Strictly speaking, an infinite number of
unknowns, Cnm , Snm , etc., are to be determined from a finite number of
observations. In order to get a definite solution, it must be assumed that the
effect of higher-degree terms is negligibly small. But even then there are very
many unknowns: coordinates of the observing stations, parameters of the
274 7 Space methods
reference orbit, and gravitational parameters; in addition, other unknowns

must be included to take into account nongravitational forces acting on the
satellite, such as air drag. An appropriate computational tool is least-squares
collocation with parameters (Moritz 1980 a: Sect. 16).
To get a strong solution, observations should be evenly distributed both
in space (with respect to the inclination of the satellites used) and in time.
Present results
At present (2005), several determinations of tesseral harmonics up to the
degree 360 are available from a combination of satellite and terrestrial data.
Soon the degree 1800 will be achieved. These coefficients represent the large-
scale features of the disturbing potential T and, hence, of the geoid, since
the geoidal height is given by N = T /γ. There is a general agreement be-
tween the essential aspects of these determinations as expressed in geoidal
maps, although the details of these maps, and even more so the individual
coefficients, are rather different.
As an example we take the first nonzonal coefficients, C22 and S22 , which,
according to Sect. 2.6, Eq. (2–95), express the inequality of the earth’s prin-
cipal equatorial moments of inertia or, somewhat loosely speaking, its triax-
iality. According to Groten (2004), we have C22 = (1574.5 ± 0.7) · 10−9 and
S22 = (−903.9 ± 0.7) · 10−9 .
Concerning the order of magnitude, J2 is on the order of magnitude of
10 , where all the other coefficients are of order 10−6 . This is why the earth
−3
can be approximated by an ellipsoid so well.
7.6 New satellite gravity missions

7.6.1 Motivation and introductory considerations
Accuracy requirements in geodesy, geophysics, and oceanography for detailed
gravity field information amount to 1 mgal for gravity anomalies. The related
accuracy for the geoid ranges from 1 to 2 cm. In the presatellite era, the
earth’s gravity field was known with high accuracy only in a few regions
of the world. Primarily, the available accurate gravity field information was
based on terrestrial and airborne measurements. This implied that in large
parts of the world there were virtually no gravity data available.
Why do we need the earth’s gravity field at all? Following Pail (2003),
first, the gravity field reflects the mass inhomogeneities in the earth’s interior
and on the earth’s surface. Second, it is fundamental for the determination
of the geoid (see Chap. 11) which, in its turn, may be regarded as a physical
7.6 New satellite gravity missions 275
reference surface for a number of geodynamic processes (subject to conti-

nents, oceans, ice masses, atmosphere, etc.) and their interaction. The mass
inhomogeneities are a necessary prerequisite to understand convection mo-
tions in the earth’s mantle which are responsible for plate tectonics. Some
large and many small lithospheric plates with a thickness of some 100–200 km
move with a relative velocity of some centimeters per year. At the edges of
the plates, seismic zones and volcanoes are situated.
Many time-dependent earth-related processes can be regarded as changes
of the mass distribution and, thus, influence the gravity field, e.g., ocean cir-
culation, ice mass variations, sea level change, tides, volcanism, post-glacial
rebound. These variations may be categorized according to their periodicity.
Some of these effects are extremely long-periodic or secular, e.g., plate tec-
tonics with about 100 million years. In contrast, changes of the ice masses
may amount to some 10 years only; even immediate events like earthquakes
may occur.
These variations are referred to a global physical reference surface, the
geoid. Therefore, the more accurately we know the geoid, the better we
accurately understand the previously mentioned effects. Referring to various
disciplines, the earth’s gravity field is important for, e.g., geodesy, geophysics,
oceanography, and climatology.
Geodesy
As mentioned in Sect. 5.3, GPS has revolutionized geodesy in many respects.
Despite the tremendous importance of GPS, in Sect. 5.4 it was shown that
the user of GPS gets only geometric quantities: WGS 84 coordinates, i.e.,
geocentric rectangular coordinates X, Y, Z or, computed from them, ellip-
soidal coordinates ϕ, λ, h (see Sect. 5.6.1). Therefore, the height obtained by
GPS, i.e., the ellipsoidal height h, is purely geometric. To transform these
heights into orthometric heights H by H = h − N , the geoidal undulation N
is required. Using satellites to determine the earth’s gravity field, a globally
uniform height system will result.
Additionally, an accurate knowledge of the earth’s gravity field improves
the orbit determination of satellites.
Oceanography
The sea surface topography (SST), i.e., the difference between the geoid and
the mean sea surface, can be determined when combining satellite altimetry
data and the earth’s gravity field data. From Fig. 7.7 we obtain the relation
h = N + SST + ∆H + a , (7–47)
276 7 Space methods
satellite orbit
a
h
DH
mean sea surface
SST instantaneous
sea surface
geoid
N
ellipsoid
Fig. 7.7. Satellite altimetry
where h is the ellipsoidal height of the altimeter satellite (based on orbit

computations), N is the geoidal height, SST is the sea surface topography
to be derived, ∆H is caused by the instantaneous tidal effect, and a is
the altimeter measurement. Note that (7–47) is a simplified representation,
since, e.g., usually SST is split into a dynamic and a constant part. Refer to
Seeber (2003: Sect. 9.3.1) for more details.
Knowing the sea surface topography, ocean currents and circulations may
be explained, which is highly interesting for our understanding of the global
energy transport. Ocean currents together with their time variations are an
important indicator for climatic changes.
This method suffers from different accuracy influences in the results:
when referring the mean sea surface to the ellipsoid, centimeter accuracy
could be achieved. Involving the gravity model and referring the sea surface
topography to the geoid as in Fig. 7.7, an improved geoid is required for a
consistent accuracy level.
Geophysics
As mentioned earlier, the earth’s gravity field reflects the mass inhomo-
geneities in the interior of the earth. Knowing gravity values on the earth’s
surface and, in addition, complementary data (e.g., magnetic and seismic
data), improved models for the structure and processes in the earth’s inte-
rior may be obtained. These processes may cause the movement of tectonic
plates which are responsible for earthquakes. Thus, we see that the gravity
field is the fundamental link in a chain of interactive processes. Using more
descriptive terms, an improved knowledge on the gravity field may yield
more accurate methods to predict earthquakes. This justifies any effort on
the determination of the earth’s gravity field.
7.6.2 Measurement concepts

From the introduction above, the need for an accurate determination of the
earth’s gravity field becomes evident. Three different measurement concepts
evolved, leading to three different gravity field satellite missions:
• satellite-to-satellite (SST) tracking in high-low mode being realized by

the “Challenging Minisatellite Payload” (CHAMP) mission,
• satellite-to-satellite tracking in low-low mode being realized by the
“Gravity Recovery and Climate Experiment” (GRACE), and
• satellite gravity gradiometry, the objective of the “Gravity Field and
Steady State Ocean Circulation Explorer” (GOCE) mission.
Before giving some details on the objectives and payloads of the missions,
the different concepts are briefly described.
Satellite-to-satellite tracking in high-low mode

The principle is shown in Fig. 7.8. The orbit of the low earth orbit (LEO)
satellite is continuously determined by satellites of global systems such as
GPS, GLONASS or, in the future, Galileo. Note that the term “high-low
mode” is not really appropriate because the satellites of GPS, GLONASS,
and Galileo belong to the mean earth orbit (MEO) satellites and not to the
high earth orbit (HEO) satellites. However, we keep the notation as used in
Seeber (2003: Sect. 10.1). Apart from satellite-to-satellite tracking, the LEO
satellite uses an accelerometer. In principle, three-dimensional perturbing
accelerations caused by the earth’s gravity field are measured. These ac-
celerations correspond to first derivatives of the gravitational potential V .
GPS satellites
LEO satellite
3D accelerometer
earth's surface
Fig. 7.8. Satellite-to-satellite tracking in high-low mode

278 7 Space methods
The gravity field is derived by inverting (in the sense of inverse problems,
cf. the remark on inverse problems at the end of Sect. 1.13) the information
obtained from the satellite orbit
Satellite-to-satellite tracking in low-low mode

The principle is shown in Fig. 7.9. Two LEO satellites are placed in the same
orbit but separated by some hundreds of kilometers (about 220 km in the
case of GRACE). Ranges and range rates between the satellites are measured
to utmost accuracy. Individually, the orbit of each LEO satellite is affected
by perturbing accelerations which correspond to the first derivatives of the
gravitational potential. In combination, differences of accelerations result.
In addition, the position of the LEOs is determined by GPS satellites. This
means that inherently satellite-to-satellite tracking in high-low mode is also
implied. The effect of nongravitational forces on the satellite, e.g., due to air
drag, must either be compensated or measured by an accelerometer.
GPS satellites
LEO satellites
3D accelerometers
earth's surface
Fig. 7.9. Satellite-to-satellite tracking in low-low mode
Satellite gravity gradiometry

Compared to the just decribed low-low mode of satellite-to-satellite tracking
with a long baseline between the two LEOs, the baseline between the ac-
celerometer units tends to zero in case of satellite gravity gradiometry. This
is achieved by placing both units into a single satellite (Fig. 7.10). Therefore,
satellite gradiometry is the measurement of acceleration differences in three
spatial orthogonal directions between the test masses of the six accelerometer
units (two on each of the three axes) inside the satellite. In other words, the
measured signal is the difference in gravitational acceleration at the satellite,
GPS satellites
LEO satellite
gradiometer
earth's surface
Fig. 7.10. Satellite gravity gradiometry with a three-axis gradiometer
where the gravitational signal arises from the attracting masses of the earth.
Thus, the measured signal corresponds to the gradients of the component
of the gravity acceleration, i.e., the second derivatives of the gravitational
potential. For instance, in obvious notation we read from Fig. 7.11
Vx2 − Vx1 ∆Vx . ∂Vx
= = = Vxz . (7–48)
∆z ∆z ∂z
Summarizing the briefly described three methods, satellite-to-satellite track-

ing in high-low mode, satellite-to-satellite tracking in low-low mode, and
satellite gravity gradiometry, we may say that the basic observable is grav-
itational acceleration. Following Rummel et al. (2002), the case of satellite-
to-satellite tracking in high-low mode corresponds to a three-dimensional
position, velocity or acceleration determination of a LEO satellite. The three-
dimensional accelerometry corresponds to gravity acceleration. Mathemati-
cally, this is expressed by the first derivatives of the gravitational potential.
Considering the low-low mode, the principle corresponds to the line-of-
Vx1
Dz
x
Vx2
Fig. 7.11. Measuring the second derivative Vxz

280 7 Space methods
sight measurement of the range, range rate or acceleration difference between

the two low-orbiting satellites. The intersatellite link corresponds to accel-
eration differences between the two LEO satellites. Mathematically, this is
expressed by the difference of first derivatives of the gravitational potential
over a long baseline (i.e., the distance between the two LEO satellites).
In the case of satellite gradiometry, three-dimensional acceleration dif-
ferences referring to the very short baseline realized by the gradiometer are
measured. The gradient of gravity components corresponds to the accelera-
tion gradient. Mathematically, this is expressed by the second derivatives of
the gravitational potential.
Another feature inherent to satellite gravity missions should be kept in
mind: the amplification of errors by the factor (r/R)n+1 when transferring
the measurement comprising the signal and noise from satellite altitude to
the earth’s surface. The factor (r/R)n+1 describes the field attenuation with
altitude. This error amplification effect is minimized by using an orbit as low
as possible and by not measuring the potential V itself or its gradient but
rather its second-order derivatives as in gravity gradiometry.
7.6.3 The CHAMP mission

The information on the challenging minisatellite payload (CHAMP) mission
has been extracted primarily from http://op.gfz-potsdam.de/champ.
The Geoforschungszentrum Potsdam initiated the CHAMP idea and has
the main responsibility. The primary CHAMP objectives are the following:
• mapping of the global gravity field, or, more specifically, to accurately

determine the long-wavelength features of the static earth gravity field
and its temporal variations (caused, e.g., by atmospheric mass redis-
tributions, ocean circulation, sea level changes resulting from polar ice
melting);
• mapping of the global magnetic field, or, more specifically, to accurately
determine the main and crustal magnetic field of the earth and its
space-time variations;
• profiling of the ionosphere and the troposphere, or, more specifically, to
derive from GPS signal refraction data information on the temperature,
water vapor, and electronic content of the atmosphere.
The CHAMP mission was launched on July 15, 2000 from the Russian Ple-
setsk cosmodrome. The main mission parameters of the respective satellite
are the following:
• almost circular (eccentricity e < 0.004) and near-polar (i = 87◦ ) orbit,

• initial altitude of 454 km,

• designed lifetime of five years for the mission (but the life expectation
is much higher!),
• weight of 522 kg, length of 8.3 m (including a “boom” of 4 m length),
width of 1.6 m, height of 0.75 m.
This initial altitude may be regarded as a compromise between gravity

and magnetic field measurements. Considering the gravity field, a lower al-
titude would be desirable. Primarily due to atmospheric drag, the altitude
will decrease to about 300 km and even less which is important because of an
increasing sensitivity with respect to gravity field coefficient determination.
The reason for the curious 4 m boom is that the magnetometry assembly
must be separated from the main body of the satellite (“magnetic cleanliness
reasons”, see http://op.gfz-potsdam.de/champ).
To achieve the mission goals, the following payload is on board of the
satellite:
• dual-frequency GPS receiver connected to a multiple antenna system

to determine the orbit of the CHAMP satellite using code and phase
pseudoranges;
• three-axis accelerometer to measure the nongravitational accelerations
acting on the spacecraft (air drag, solar radiation pressure, albedo,
etc.);
• laser retroreflector for backup tracking to measure two-way ranges be-
tween ground stations and the satellites with 1–2 cm accuracy; these
measurements support the precise orbit determination;
• fluxgate magnetometer to measure the vector components of the mag-
netic field of the earth (this instrument is supported by a scalar mag-
netometer to provide a calibration capability of the fluxgate magne-
tometer);
• equipment to determine the ion density and temperature, the drift
velocity, and the electric field;
• two advanced star trackers to provide high-precision attitude informa-
tion as required for the three-axis accelerometer, the digital ion drift
meter, but also for the attitude control of the satellite.
Typical other equipment required for a proper operation of the satellite

but with no specific relation to the scientific objectives of the mission is not
detailed here, such as the cold gas propulsion system, the thermal control
system, the power generation, the data handling, the telemetry, tracking and
command system. Furthermore, we do not list items of the control segment
282 7 Space methods
of the CHAMP mission, but refer the reader to the previously mentioned
homepage.
As explained before, the measuring principle for CHAMP is satellite-to-
satellite tracking in high-low mode. The gravity field of the earth perturbes
the CHAMP satellite orbit. These perturbing accelerations correspond to
first derivatives of the gravitational potential V . This implies that the gravity
field of the earth may be derived from observed gravitational satellite orbit
perturbations applying numerical orbit integration (Montenbruck and Gill
2001) or using the energy balance principle (Ilk 1999, Jekeli 1999, Sneeuw
et al. 2002).
For further reading see Reigber et al. (2003), Seeber (2003: Sect. 10.2.2).
7.6.4 The GRACE mission

The information on the gravity recovery and climate experiment (GRACE)
mission has been extracted primarily from http://op.gfz-potsdam.de/grace.
The GRACE mission is a joint project between the U.S. National Aero-
nautics and Space Administration (NASA) and the Deutsches Zentrum für
Luft- und Raumfahrt (DLR). The primary objectives of the mission are the
following:
• determination of the global high-resolution gravity field of the earth,

• temporal gravity variations.
In addition, another task is the determination of the total electron content

by GPS measurements to get knowledge on the refractivity in the ionosphere
and troposphere. The two satellites of this mission were launched simultane-
ously on March 17, 2002 from the Russian Plesetsk cosmodrome. The main
mission parameters of the two satellites are the following:
• almost circular (eccentricity e < 0.005) and near-polar (i = 89◦ ) orbit,

• initial altitude between 485 km and 500 km,
• the two satellites are some 220 km apart (this requires orbit maneuvers
every one or two months to maintain the separation between the two
spacecraft),
• design lifetime of the mission is five years (but extended operation is
envisaged),
• the weight of each satellite is about 480 kg and the length about 3 m.
As with CHAMP, also the altitude of the GRACE satellites will decrease
in the course of their lifetime primarily because of atmospheric drag. The
amount of this decrease depends on the solar activity cycle and may accumu-
late in the mission lifetime to some 50 km on low activity, and up to 200 km
on high activity, see http://op.gfz-potsdam.de/grace.
The range between the two satellites must be determined extremely ac-
curately. Its range rate must be known to better than 1 µm s−1 , which is
achieved by intersatellite microwave measurements. The basic idea is that
variations in the gravity field cause variations in the range between the
two satellites; areas of stronger gravity will affect the lead satellite first and,
therefore, accelerate it away from the following satellite (Seeber 2003: p. 479).
GRACE will not only provide a static global gravity field but also its
temporal variations.
To achieve the mission goals, the following payload is on board of the
two satellites:
• The K-band ranging system is the key instrument of GRACE to mea-
sure the range changes between both satellites using dual-band mi-
crowave signals (i.e., two one-way ranges) with a precision of about
1 µm s−1 . The ranges are obtained at a sampling rate of 10 Hz.
• The GPS receiver serves for the precise orbit determination of the
GRACE spacecraft and provides data for atmospheric and ionospheric
profiling. To achieve this, satellite-to-satellite tracking between the
GRACE satellites and the GPS satellites is realized. A navigation solu-
tion comprising position, velocity, and a time mark is derived on board.
The navigation solution is required for the attitude control system. The
precise orbit based on code and carrier pseudoranges is determined on
ground.
• The attitude and orbit control system comprises a cold gas propulsion
system, three magnetic torque rods, star trackers, a three-axis inertial
reference unit to measure angular rates, and a three-axis magnetome-
ter.
• The accelerometer measures all nongravitational accelerations on the
GRACE spacecraft, e.g., due to air drag or solar radiation pressure.
• The laser retroreflector is a passive payload instrument used to reflect
short laser pulses transmitted by ground stations. The distance be-
tween a ground station and a GRACE satellite can be measured with
an accuracy of 1–2 cm. The laser retroreflector data are primarily used
together with the GPS receiver data for the precise orbit determina-
tion.
In 2004, the GRACE science team released to the public a first version
of a new earth gravity field model complete to degree and order 150. The
resulting improved geoid together with satellite altimetry will advance the
284 7 Space methods
knowledge on oceanographic, geodetic, and solid earth issues such as oceanic

heat flux, change of sea level, ocean currents, precise positioning, orbit de-
termination, and leveling.
The GRACE concept can be regarded as a one-dimensional gradiometer
with a very long baseline of 220 km (Seeber 2003: p. 480). In contrast to this
concept, GOCE uses very short baselines (50 cm) in three directions.
7.6.5 The GOCE mission

The main sources of this section are www.esa.int/export/esaLP/goce.html,
ESA (1999), Müller (2001), Drinkwater et al. (2003), and Pail (2003).
The gravity field and steady-state ocean circulation explorer (GOCE)
mission is a Core Mission of the ESA Living Planet Programme. The primary
objectives of the GOCE mission are to measure the earth’s stationary gravity
field and to model the geoid with extremely high accuracy. More specifically:
• to determine the gravity anomalies with an accuracy of 1 mgal,
• to determine the geoid with an accuracy of 1–2 cm,
• to achieve these results at a spatial resolution better than 100 km.
According to the above mission requirements, GOCE is intended for a
representation of the gravity potential by spherical harmonics complete at
least to degree and order 200 (corresponding to the spatial resolution of
100 km), but 250 is envisaged.
From the geodetic point of view, a global geoid of 1–2 cm accuracy and
a gravity field model accurate to 1 mgal at about 100 km spatial resolution
may be used – among many other important applications – for the following
purposes:
• Control (or replacement) of traditional leveling by leveling with GPS.
In Sect. 4.6 we have learned the basic equation (4–72), H = h − N ,
relating the orthometric height H (above the geoid), the ellipsoidal
height h (above the ellipsoid), and the geoidal undulation N . With N
accurately known from GOCE and h measured by GPS (Sects. 5.5,
5.6.1), the orthometric height H is readily obtained.
• Worldwide unification of height systems so as to refer to one height
datum which allows for comparison of different sea levels (e.g., in the
North Sea and in the Mediterranean) and sea-level changes (which
may be caused by melting continental ice sheets). Remember that the
geoid is defined as an equipotential surface which follows a hypothetical
ocean surface at rest (in the absence of tides and currents and other
smaller influences). Consequently, a precise geoid is crucial in deriving
accurate measurements of ocean currents and sea-level changes.
• Providing a significant improvement in satellite orbit determination

and prediction. This especially applies to low-orbiting satellites. The
highly accurate gravity field will enable a better separation of the per-
turbations caused by the static gravity field and other perturbing forces
(not only the nongravitational forces caused by air drag and solar ra-
diation pressure but also perturbations caused by the solid earth and
ocean tides).
The duration of the mission is scheduled with nominally 20 months,

including a 3-month commissioning and calibration phase and two measure-
ment phases, each lasting six months and separated by a long eclipse period.
The other main mission parameters are the following:
• due for launch in 2007 from Plesetsk in Russia,

• sun-synchronous orbit, inclination 96.5◦ ,
• measurement altitudes: approximately 250 km,
• single ground station in Kiruna, Sweden, to exchange data and com-
mands; the European Space Operations Center (ESOC) at Darmstadt
will be used for mission and satellite control.
The main payload components are the following:
• three-axis gravity gradiometer based on three pairs of electrostatic

servo-controlled accelerometers to measure gravity gradients in three
spatial orthogonal directions: the desired signal is the difference in
gravitational acceleration (between a pair of accelerometers separated
by 0.5 m) at the test mass location inside the spacecraft caused by
gravity anomalies from attracting masses of the earth;
• geodetic dual-frequency (to compensate for ionospheric delays) multi-
channel GPS receiver with codeless tracking capability to (1) determine
the orbit of the GOCE satellite and (2) derive gravity information from
this orbit (the first task is performed by satellite-to-satellite tracking
in the high-low mode: this provides knowledge of the precise position
of the [low] spacecraft relative to [high] reference satellites such as the
GPS satellites; the second task is solved by orbit perturbation analysis
yielding gravity information);
• laser retroreflector to enable tracking by ground-based laser stations;
• attitude control accomplished by actuators comprising an ion thruster
assembly, star trackers, a three-axis magnetic torquer, and some other
sensors;
• length of the satellite about 5 m, cross section of 1 m2 , weight about
1000 kg.
286 7 Space methods
Referring to the results, the main output of this mission will be the following:
• spherical-harmonic coefficients for the gravitational potential, see, e.g.,
(2–80),
• corresponding variance-covariance matrix.
Derived products from this main output are geoidal heights, gravity anoma-
lies, and also oceanographic data.
It is important to mention that the GPS orbit analysis of GOCE will
rather yield long-wavelength information of the gravity field, while the satel-
lite gravity gradiometry will yield the short-wavelength information.
GOCE is the first “drag-free” mission, which implies that the satellite
moves in free fall around the earth. Therefore, a drag compensation and
attitude control system is required to compensate for drag forces and torques.
This and more information may be found in Rebhahn et al. (2000),
Drinkwater et al. (2003), Pail (2003), www.esa.int/livingplanet/goce.
Measurements
The basic principle of gradiometry in GOCE is the measurement of accel-
eration differences for a very short baseline. Considering two accelerometers
separated by 50 cm on one axis, Müller (2001) and Pail (2003) write the two
observation equations as

a1 = M + Ω̇ + ΩΩ ∆x + f ng ,
(7–49)
a2 = − M + Ω̇ + ΩΩ ∆x + f ng ,
where a1 and a2 are the measured accelerations of the two accelerometers

on the axis, and M is the Marussi tensor,
⎡ 2 ⎤
∂ V ∂2V ∂2V
⎢ ∂x2 ∂x ∂y ∂x ∂z ⎥
⎢ ⎥
⎢ 2 ⎥
⎢ ∂ V ∂2V ∂2V ⎥
M=⎢ ⎢ ∂x ∂y
⎥,
⎥ (7–50)
⎢ ∂y 2 ∂y ∂z ⎥
⎢ ⎥
⎣ ∂2V ∂2V ∂2V ⎦
∂x ∂z ∂y ∂z ∂z 2
which comprises the second derivatives of the gravitational potential (our
target quantity!). Furthermore, the skewsymmetric matrix
⎡ ⎤
0 ω3 −ω2
Ω = ⎣ −ω3 0 ω1 ⎦ (7–51)
ω2 −ω1 0
comprises the components of the angular velocity and is used to describe the
orientation of the gradiometer. Since Ω is skewsymmetric, the tensor ΩΩ
is symmetric. Finally, ∆x in (7–49) is the vector from the intersection of
the three coordinate axes to the respective accelerometer (where the same
length is assumed), and f ng comprises all nongravitational effects (air drag,
solar radiation pressure, etc.).
Now we once add (“common mode”) and once subtract (“differential
mode”) the two accelerations in (7–49) and obtain
(a1 + a2 )/2 = f ng ,
(7–52)
(a1 − a2 )/2 = M + Ω̇ + ΩΩ ∆x ,
where we can extract the nongravitational effects f ng in the common mode.

Introducing the quantity
Γ = M + Ω̇ + ΩΩ (7–53)
and assuming a known geometry of the gradiometer, i.e., ∆x may safely

assumed to be known, then the remaining task is to extract the gravity
gradient tensor M from Γ. This can be achieved by the two relations
(Γ − ΓT )/2 = Ω̇ ,
(7–54)
(Γ + ΓT )/2 = M + ΩΩ ,
where the superscript T denotes transposition. To verify these relations, a

little matrix calculus is needed. If, generally, K is a symmetric matrix, then
we have K = KT . If K is a skewsymmetric matrix, then we have K = −KT .
Referring now to (7–53), we know that M is symmetric, Ω̇ is skewsym-
metric, and ΩΩ is symmetric. Therefore, transposing (7–53) yields
ΓT = M − Ω̇ + ΩΩ . (7–55)
Using now (7–53) and (7–55), we get immediately
Γ − ΓT = 2Ω̇ (7–56)
and, finally,
(Γ − ΓT )/2 = Ω̇ , (7–57)
which completes our proof for the first relation of (7–54). To prove the second
relation of (7–54), we add Eqs. (7–53) and (7–55):
Γ + ΓT = 2M + 2ΩΩ (7–58)
288 7 Space methods
or
(Γ + ΓT )/2 = M + ΩΩ , (7–59)
which concludes our proof.
Since we have determined Ω̇ in (7–57), we can get Ω by integration:
t
Ω(t) = Ω(t0 ) + Ω̇ dt , (7–60)
t0
where the initial orientation Ω(t0 ) is obtained from the star trackers. Squar-
ing the result for Ω(t) yields ΩΩ, which is needed in (7–59) so that we
find
M = (Γ + ΓT )/2 − ΩΩ (7–61)
as final result for the desired Marussi tensor M. Many more details may be
found in Rummel (1986).
8 Modern views on the
determination of the figure of
the earth
8.1 Introduction
In the preceding chapters we have usually followed what might be called the
conservative approach to the problems of physical geodesy using classical
observations. The geodetic measurements – astronomical coordinates and
azimuths, horizontal angles, gravity observations, etc. – are reduced to the
geoid, and the “geodetic boundary-value problem” is solved for the geoid by
means of Stokes’ integral and similar formulas. The geoid then serves as a
basis for establishing the position of points of the earth’s surface.
The advantage of this approach is that the geoid is a level surface, capable
of a simple definition in terms of the physically meaningful and geodetically
important potential W . The geoid represents the most obvious mathematical
formulation of a horizontal surface at sea level. This is why the use of the
geoid simplifies geodetic problems and makes them accessible to geometrical
intuition.
The disadvantage is that the potential W inside the earth, and hence
the geoid W = constant, depends on the density because of Poisson’s
Eq. (2–9),
∆W = −4π G + 2ω 2 . (8–1)
Therefore, in order to determine or to use the geoid, the density of the
masses at every point between the geoid and the ground must be known, at
least theoretically. This is clearly impossible, and therefore some assumptions
concerning the density must be made, which is unsatisfactory theoretically,
even though the practical influence of these assumptions is usually rather
small.
For this reason it is of basic importance that M.S. Molodensky in 1945 was
able to show that the physical surface of the earth can be determined from
geodetic measurements alone, without using the density of the earth’s crust.
This requires that the concept of the geoid be abandoned. The mathematical
formulation becomes more abstract and more difficult. Both the gravimetric
method and the astrogeodetic method can be modified for this purpose. The
gravity anomalies and the deflections of the vertical now refer to the ground,
290 8 Determination of the figure of the earth
and no longer to sea level; the “height anomalies” at ground level take the
place of the geoidal undulations.
These developments have considerably broadened our insight into the
principles of physical geodesy and have also introduced powerful new meth-
ods for tackling classical problems. Hence their basic theoretical significance
is by no means lessened by the fact that many scientists prefer to retain the
geoid because of its conceptual and practical advantages.
In this chapter, we first give a concise survey of the conventional determi-
nation of the geoid by means of gravity reductions, in order to understand
better the modern ideas. After an exposition of Molodensky’s theory, we
show how the new methods may be applied to classical problems such as
gravity reduction or the determination of the geoid by gravimetric and as-
trogeodetic methods. It should be mentioned that the terms “modern” and
“conventional” merely serve as convenient labels; they do not imply any con-
notation of value or preferability.
Part I: Gravimetric methods

8.2 Gravity reductions and the geoid
The integrals of Stokes and of Vening Meinesz and similar formulas presup-
pose that the disturbing potential T is harmonic on the geoid, which implies
that there are no masses outside the geoid. This assumption – no masses
outside the bounding surface – is necessary if we wish to treat any problem
of physical geodesy as a boundary-value problem in the sense of potential
theory. The reason is that the boundary-value problems of potential theory
always involve harmonic functions, that is, solutions of Laplace’s equation
∆T = 0 . (8–2)
This is equivalent to ∆V = 0. Proof: T = W −U (U is the normal potential),
∆W = 2ω 2 outside the earth (density zero, only rotation, ∆U = 2ω 2 for
the same reason, hence ∆T = ∆W − ∆U = 2ω 2 − 2ω 2 = 0). Since then
∆W = 2ω 2 rather than zero by Eq. (2–9), it is not quite correct to call the
external gravity potential W harmonic as well, but we may nevertheless do
so for simplicity. No misunderstanding is possible.
We know, for instance, that the determination of T or N from gravity
anomalies ∆g may be considered as a third boundary-value problem (see
Sect. 1.13).
Since there are masses outside the geoid, they must be moved inside the
geoid or completely removed before we can apply Stokes’ integral or related
8.2 Gravity reductions and the geoid 291
earth's
surface
P0 geoid
±N P c
cogeoid
N c N
ellipsoid
Q0
Fig. 8.1. Geoid and cogeoid
formulas. This is the purpose of the various gravity reductions. They were
considered extensively in Chap. 3; we therefore can limit ourselves to pointing
out those theoretical features that are relevant to our present problem.
If the external masses, the masses outside the geoid, are removed or
moved inside the geoid, then gravity changes. Furthermore, gravity is ob-
served at ground level but is needed at sea level. Thus, the reduction of
gravity involves the consideration of these two effects, in order to obtain
boundary values on the geoid.
This regularization of the geoid by removing the external masses unfortu-
nately also changes the level surfaces and hence, in general, the geoid. This is
the indirect effect; the changed geoid is called the cogeoid or the regularized
geoid.
The principle of this method may be described as follows (Jung 1956:
p. 578); see Fig. 8.1.
1. The masses outside the geoid are, by computation, either removed

entirely or else moved inside the geoid. The effect of this procedure on
the value of gravity g at the station P is considered.
2. The gravity station is moved from P down to the geoid, to the point
P0 . Again, the corresponding effect on the gravity is considered.
3. The indirect effect, the distance δN = P0 P c , is obtained by dividing

the change in potential at the geoid, δW , by normal gravity (Bruns’
theorem):
δW
δN = . (8–3)
γ
4. The gravity station is now moved from the geoidal point P0 to the
cogeoid, to the point P c (hence the notation with upper index c). This
gives the boundary value of gravity at the cogeoid, gc .
5. The shape of the cogeoid is computed from the reduced gravity anoma-
lies
∆gc = gc − γ (8–4)
by Stokes’ formula, which gives N c = QP c .
6. Finally, the geoid is determined by considering the indirect effect. The

geoidal undulation N is thus obtained as
N = N c + δN . (8–5)
Remark. At first sight it may seem that the masses between the geoid
and the cogeoid should be removed if the cogeoid happens to be below the
geoid, because Stokes’ formula is applied to the cogeoid. However, this is
not necessary, and therefore we need not be concerned with a “secondary
indirect effect”. The argument is a little too technical to be presented here;
see Moritz (1965: p. 26).
In principle, every gravity reduction that gives boundary values at the
geoid is equally suited for the determination of the geoid, provided the in-
direct effect is properly taken into account. Thus, the selection of a good
reduction method should be made from other points of view, such as the
geophysical meaning of the reduced gravity anomalies, the simplicity of com-
putation, the feasibility of interpolation between the gravity stations, the
smallness or even absence of the indirect effect, etc. (see Sect. 3.7).
The Bouguer reduction corresponds to a complete removal of the ex-
ternal masses. In the isostatic reduction, these masses are shifted vertically
downward according to some theory of isostasy. In Helmert’s condensation
reduction, the external masses are compressed to form a surface layer on
the geoid. The Bouguer reduction and especially the isostatic reduction (in
modern terminology topographic-isostatic reduction) are used as auxiliary
quantities for computational purposes, especially to facilitate interpolation.
The free-air anomaly is nowadays used in three senses:
1. at ground level (on the physical surface of the earth) it is simply the
gravity anomaly in the sense of Molodensky (Sect. 8.4);
2. at sea level it may be identified with the analytical continuation of
the Molodensky anomaly from ground down to sea level. This will
be considered in detail in Sect. 8.6. A final review will be found in
Sect. 8.15.
8.2 Gravity reductions and the geoid 293
3. The free-air anomaly can be theoretically interpreted as an approxi-

mation of the classical condensation anomaly in the sense of Helmert
(Sect. 3.9). This is one of the interpretations of the frequent practice to
simply apply Stokes’ formula to the classical free-air anomaly, where
only the standard normal free-air reduction is applied to measured
gravity g, see Eq. (8–6) below.
This is pretty rigorously the gravity anomaly in the sense of Molo-
densky (item 1 above), so there is another interpretation of this fre-
quent practice: it is a (conscious or unconscious) use of Molodensky’s
method in the zero approximation (i.e., only Stokes’ formula without
Molodensky correction g1 , see Sect. 8.6). Of course, this works only in
a reasonably flat terrain.
Important remark. Curiously enough, it helps if the terrain correction
(Sect. 3.4) is applied; this is explained in Moritz (1980 a: Sect. 48) as
some kind of Molodensky correction g1 and in Moritz (1990: p. 244)
by isostatic reduction.
Also the Poincaré–Prey reduction is quite different (Sect. 3.5). It gives

the actual gravity inside the earth. It does not give boundary values but is
used for orthometric heights (Chap. 4).
In all reduction methods it is necessary to know the density of the masses
above the geoid. In practice, this involves some kind of an assumption – for
instance, putting = 2.67 g cm−3 . A second assumption is usually made in
the free-air reduction, which is part of the reduction of gravity to the geoid:
the actual free-air gravity gradient is assumed to be equal to the normal
gradient
∂γ .
= −0.3086 mgal m−1 . (8–6)
∂h
These two assumptions falsify our results, at least theoretically.
The second assumption can be avoided by using the actual free-air gradi-
ent as computed by the methods of Sect. 2.20. The anomalies ∆g to be used
in formula (2–394) must be gravity anomalies reduced to the geoid: gravity
g after steps l and 2 of the above description, minus normal gravity γ on the
ellipsoid. This presupposes that in step 2 a preliminary free-air reduction
using the normal gradient has been applied first.
Deflections of the vertical

The indirect effect affects the deflection of the vertical as well as the geoidal
height. We have found
N = N c + δN , (8–7)
where N c is the undulation of the cogeoid, the immediate result of Stokes’

formula, and δN is the indirect effect. By differentiating N in a horizontal
direction, we get the deflection component along this direction:
∂N ∂N c ∂(δN )
ε=− =− − . (8–8)
∂s ∂s ∂s
This means that we must add to the immediate result of Vening Meinesz’
formula, −∂N c /∂s, a term representing the horizontal derivative of δN (see
also Sect. 3.7).
To repeat, the main purpose is to obtain a simple boundary surface.
The geoid approximated by an ellipsoid or even a sphere is a much easier
boundary surface than the physical surface of the earth, to which we turn
now.
8.3 Geodetic boundary-value problems

It is, however, quite easy to understand the general principles. In space we
have the well-known fact that the gravity vector g and the gravity potential
(geopotential) W are related by

∂W ∂W ∂W
g = grad W ≡ , , , (8–9)
∂x ∂y ∂z
which shows that the force g is the gradient vector of the potential.
Let S be the earth’s topographic surface and let W and g be the geopo-
tential and the gravity vector on this surface. Then there exists a relation
g = f (S, W ) , (8–10)
the gravity vector g on S is a function of the surface S and the geopotential

W on it. This can be seen in the following way. Let the surface S and the
geopotential W on S be given. The gravitational potential V is obtained
by subtracting the potential of the centrifugal force Φ, which is simple and
perfectly known (Sect. 2.1):
V = W − Φ. (8–11)
The potential V outside the earth is a solution of Laplace’s equation ∆V = 0

and consequently harmonic (Sect. 1.3). Thus, knowing V on S, we can ob-
tain V outside S by solving Dirichlet’s boundary-value problem, the first
boundary-value problem of potential theory, which is practically always
uniquely solvable (Sect. 1.12) at least if V is sufficiently smooth on S. After
8.3 Geodetic boundary-value problems 295
having found V as a function in space outside S, we obtain the gravitational

force grad V . Adding the well-known and simple vector of the centrifugal
force, we obtain the gravity vector g outside and, by continuity, on S.
This is precisely what (8–10) means. The modern general concept of a
function can be explained as a rule of computation, indicating that given S
and W on S, we can uniquely calculate g on S. Note that f is not a function
in the elementary sense but rather a “nonlinear operator”, but we disregard
this for the moment. Therefore we may formulate:
(1) Molodensky’s boundary-value problem is the task to determine S, the

earth’s surface, if g and W on it are given. Formally, we have to solve
(8–10) for S:
S = F1 (g, W ) , (8–12)
that is, we get geometry from gravity.
(2) GPS boundary-value problem. Since we have GPS at our disposal, we

can consider S as known, or at least determinable by GPS. In this case,
the geometry S is known, and we can solve (8–10) for W :
W = F2 (S, g) , (8–13)
that is, we get potential from gravity. As we shall see, this is far from
being trivial: we have now a method to replace leveling, a tedious
and time-consuming old-fashioned method, by GPS leveling, a fast and
modern technique (Sect. 4.6).
In spite of all similarities, we should bear in mind a fundamental differ-

ence: (8–13) solves a fixed-boundary problem (boundary S given), whereas
(8–12) solves a free-boundary problem: the boundary S is a priori unknown
(“free”). Fixed-boundary problems are usually simpler than free ones.
This is only the principle of both solutions. The formulation is quite
easy to understand. The direct implementation of these formulas is difficult,
however, because that would imply the solution of “hard inverse function
theorems” of nonlinear functional analysis. For numerical computations, we
know series solutions, in the form of “Molodensky series”, which are suffi-
cient for all present purposes and which can, furthermore, be derived in an
elementary fashion, without needing integral equations (Molodenski 1958;
Molodenskii et al. 1962; Moritz 1980 a: Sect. 45). Here we shall outline the
known elementary solution for Molodensky’s problem and immediately ex-
tend it to the GPS problem. Both problems will be solved by very similar
Molodensky series.
The simplest possible example

Let the boundary surface S be a sphere of radius R. The earth is represented
by this sphere which is considered homogeneous and nonrotating. The po-
tential W is identical to the gravitational potential V , so that on the surface
S we have constant values
GM
W= ,
R
(8–14)
GM
g = 2 .
R
Knowing W and g, we have
W
R= , (8–15)
g
the radius of the sphere S. Thus, we have solved Molodensky’s problem in
this trivial but instructive example. We have indeed got geometry (i.e., R)
from physics (i.e., g and W )!
8.4 Molodensky’s approach and linearization

We have just seen that the reduction of gravity to sea level necessarily in-
volves assumptions concerning the density of the masses above the geoid.
This is equally true of other geodetic computations when performed in the
conventional way.
To see this, consider the problem of computing the ellipsoidal coordi-
nates ϕ, λ, h from the natural coordinates Φ, Λ, H, as described in Chap. 5.
The geometric ellipsoidal height h above the ellipsoid is obtained from the
orthometric height H above the geoid and the geoidal undulation N by
h=H +N. (8–16)
The determination of N was considered in Chap. 2 and elsewhere in this

book. To compute H from the results of leveling, we need the mean gravity
ḡ along the plumb line between the geoid and the ground (Sect. 4.3). Since
gravity g cannot be measured inside the earth, we compute it by Prey’s
reduction, for which we must know the density of the masses above the
geoid.
The ellipsoidal coordinates ϕ and λ are obtained from the astronomical
coordinates Φ and Λ and the deflection components ξ and η by
ϕ = Φ−ξ, λ = Λ − η sec ϕ . (8–17)
The coordinates Φ and Λ are measured on the ground; ξ and η can be

computed for the geoid by Vening Meinesz’ formula, the indirect effect being
8.4 Molodensky’s approach and linearization 297
taken into account according to Sect. 8.2. To apply the above formulas, either
Φ and Λ must be reduced down to the geoid or ξ and η must be reduced up
to the ground. In both cases this involves the reduction for the curvature of
the plumb line (Sect. 5.15), which also depends on the mean value ḡ through
its horizontal derivatives. Hence Prey’s reduction enters here too.
Thus we see that in the conventional approach to the problems of physical
geodesy we must know the density of the outer masses or make assumptions
concerning it. To avoid this, Molodensky proposed a different approach in
1945.
Figure 8.2 shows the geometrical principles of this method, which is es-
sentially a linearization of Eq. (8–10). The ground point P (i.e., point on the
earth’s surface S) is again projected onto the ellipsoid according to Helmert.
However, the ellipsoidal height h is now determined by
h = H∗ + ζ , (8–18)
the normal height H ∗ replacing the orthometric height H, and the height
anomaly ζ replacing the geoidal undulation N .
This will be clear if one considers the surface whose normal potential U at
every point Q is equal to the actual potential W at the corresponding point
P , so that UQ = WP , corresponding points P and Q being situated on the
same ellipsoidal normal. This surface is called the telluroid (Hirvonen 1960,
1961). The vertical distance from the ellipsoid to the telluroid is the normal
height H ∗ (Sect. 4.4), whereas the ellipsoidal height h is the vertical distance
from the ellipsoid to the earth’s surface. Thus, the difference between these
two heights is the height anomaly
ζ = h − H∗ , (8–19)
closely corresponding to the geoidal undulation N = h − H, which is the

difference between the ellipsoidal and the orthometric height.
P
³ earth's
surface S
Q
telluroid S
H* h
ellipsoid E
Q0
Fig. 8.2. Telluroid, normal height H ∗ , and height anomaly ζ

The normal height H ∗ , and hence the telluroid Σ, can be determined by

leveling combined with gravity measurements, according to Sect. 4.4. First
the geopotential number of P , C = W0 − WP , is computed by
P
C= g dn , (8–20)
0
where g is the measured gravity and dn is the leveling increment. The normal
height H ∗ is then related to C by an analytical expression such as (4–63),
2
∗ C C C
H = 1 + (1 + f + m − 2f sin ϕ)
2
+ , (8–21)
γQ0 a γQ0 a γQ0
where γQ0 is the normal gravity at the ellipsoidal point Q0 . Note that H ∗ is
independent of the density.
The normal height H ∗ of a ground point P is identical with the ellipsoidal
height h, the height above the ellipsoid, of the corresponding telluroid point
Q. If the geopotential function W were equal to the normal potential function
U at every point, then Q would coincide with P , the telluroid would coincide
with the physical surface of the earth, and the normal height of every point
would be equal to its ellipsoidal height. Actually, however, WP = UP ; hence
the difference
ζP = hP − HP∗ = hP − hQ (8–22)
is not zero. This explains the term “height anomaly” for ζ.
The gravity anomaly is now defined as
∆g = gP − γQ ; (8–23)
it is the difference between the actual gravity as measured on the ground
and the normal gravity on the telluroid. The normal gravity on the telluroid,
which we shall briefly denote by γ, is computed from the normal gravity at
the ellipsoid, γQ0 , by the normal free-air reduction, but now applied upward:
∂γ ∗ 1 ∂ 2 γ ∗2
γ ≡ γQ = γQ0 + H + H + ··· . (8–24)
∂h 2! ∂h2
For this reason, the new gravity anomalies (8–23) are called free-air anoma-
lies. They are referred to ground level, whereas the conventional gravity
anomalies have been referred to sea level. Therefore, the new free-air anoma-
lies have nothing in common with a free-air reduction of actual gravity to
sea level, except the name. This distinction should be carefully kept in mind.
A direct formula for computing γ at Q is (2–215),
∗ 2
H ∗ H
γ = γQ0 1 − 2(1 + f + m − 2f sin2 ϕ) +3 , (8–25)
a a
where γQ0 is the corresponding value on the ellipsoid.

The height anomaly ζ may be considered as the distance between the
geopotential surface W = WP = constant and the corresponding spheropo-
tential surface U = WP = constant at the point P . In Sect. 2.14 (Fig. 2.15),
we have denoted this distance by NP and have found that Bruns’ formula
(2–237) also applies to this quantity. Hence, for ζ = NP we have
T
ζ= , (8–26)
γ
where T = WP − UP is the disturbing potential at ground level, and γ the

normal gravity at the telluroid.
It may be expected that ζ is connected with the ground-level anomalies
∆g by an expression analogous to Stokes’ formula for the geoidal height N .
This is indeed true. However, the telluroid is not a level surface, and to every
point P on the earth’s surface corresponds in general a different geopotential
surface W = WP . Therefore, the relation between ∆g and ζ in the new theory
is considerably more complicated than for the geoid. In Molodensky’s original
formulation, the problem involves an integral equation, which may be solved
by an iteration, the first term of which is given by Stokes’ formula. We shall
use an equivalent but much simpler approach without integral equation.
Finally, we remark that we may also plot the height anomalies ζ above
the ellipsoid. In this way we get a surface that is identical with the geoid over
the oceans, because there ζ = N , and is very close to the geoid anywhere
else. This surface has been called the quasigeoid by Molodensky. However,
the quasigeoid is not a level surface and has no physical meaning whatever.
It must be considered as a concession to conventional conceptions that call
for a geoidlike surface. From this point of view, the normal height of a point
is its elevation above the quasigeoid, just as the orthometric height is its
elevation above the geoid.
Gravity disturbance
As usual, the gravity disturbance is defined by
δg = gP − γP . (8–27)
It is a typical new feature introduced into the practice of physial geodesy

by GPS, because GPS determines the ellipsoidal coordinates ϕ, λ, h directly
at the surface point P , so that now δg can be considered observational data
instead of ∆g.
Linearization
The linearization applies equally well for the Molodensky problem and the
GPS problem. The geometry is familiar (Fig. 8.2).
We recall the surface Σ, the telluroid, which is defined by the condition
U (Q) = W (P ) . (8–28)
We note that (8–28) is the surface equivalent to the classical relation for sea
level (Fig. 8.3)
U (Q0 ) = W (P0 ) . (8–29)
Equation (8–28) would apply with
W (P0 ) = W0 = constant (8–30)
if S were an equipotential surface, the geoid, which is the case only over the
oceans with the usual simplifying assumption that the surface of the ocean
is an equipotential surface not changing with time (Fig. 8.3).
Molodensky’s theory does not use the geoid directly but the physical
earth’s surface. We repeat once more that this is Molodensky’s epochal idea
which radically changed the course of physical geodesy since 1945.
We shall, however, use the fictitious case of S being an equipotential sur-
face, but only as a first (or zero-order) assumption in a perturbation approach
for the real earth’s surface (Molodensky series). This first approximation is
the spherical case to be considered in the next section.
Now we consider the linearization in more detail. The ellipsoidal height
h is directly determined by GPS. It may be decomposed into
h = H∗ + ζ . (8–31)
Here, H ∗ is the normal height and ζ is the height anomaly, whose definitions
are seen from Fig. 8.2. In the GPS case we do know the earth’s surface S
directly, but the telluroid Σ and the height anomalies ζ are still required for
formulating the boundary condition, just as the knowledge of the geoid does
not make superfluous the reference ellipsoid.
P0 geoid
N
ellipsoid
Q0
Fig. 8.3. Geoid and ellipsoid

The definition of the gravity anomaly ∆g and the gravity disturbance δg

has, on the earth’s surface, the same form as in the classical case of geoid
and sea level:
∂T 1 ∂γ
∆g = gP − γQ = − + T, (8–32)
∂h γ ∂h
∂T
δg = gP − γP = − . (8–33)
∂h
The gravity disturbance δg has become practically important only through
GPS, since h, the ellipsoidal height of P , can be measured using GPS and
hence γP , the normal gravity γ at P , can be determined.
As usual, Bruns’ formula applies at P0 (classical geoid height N ) and P
(Molodensky height anomaly ζ) as well:
T (P0 )
N= , (8–34)
γ
T (P )
ζ= , (8–35)
γ
with some approximate value for γ such as γ45◦ . Equation (8–32) can be
reformulated as the boundary conditions for the Molodensky problem
∂T 1 ∂γ
− T + ∆g = 0 , (8–36)
∂h γ ∂h
cf. (2–251), and for the GPS problem, cf. (2–252),
∂T
+ δg = 0 . (8–37)
∂h
These two boundary conditions apply at the surface S (Molodensky) and at
sea level as well.
Finally we introduce the spherical approximation, disregarding the flat-
tening f in the equations (which are linear relations between small quanti-
ties).
Note: The spherical approximation is a formal operation (disregarding f
in small ellipsoidal quantities) and does not mean using a “reference sphere”
instead of a reference ellipsoid in any geometrical sense (Moritz 1980 a: p. 15).
This would imply geoidal heights on the order of 20 km!
Then (8–36) and (8–37) reduce to
∂T 2
+ T + ∆g = 0 , (8–38)
∂r r
∂T
+ δg = 0 . (8–39)
∂r
These equations, for the Molodensky and the GPS problem, are valid both
at sea level (classical) and at S (Molodensky).
8.5 The spherical case

As we have agreed, we work formally with a sphere (the reference ellipsoid
stays at its geometric place!). This means putting r = R = constant. Fur-
thermore, we assume (fictitiously!) that S is a level surface.
Expanding T and ∆g into a series of Laplace spherical harmonics, see
(2–322) and (2–320), we find
∞

T (ϑ, λ) = Tn (ϑ, λ) , (8–40)
2
∞

∆g(ϑ, λ) = ∆gn (ϑ, λ) (8–41)
2
on the surface of the sphere, whence by (8–38) and (2–321) with r = R,
∞
∆gn
T =R . (8–42)
n=2
n−1
The summation starts conventionally with n = 2, rather than n = 0, for sev-

eral reasons, one of them being that n = 1 would lead to a zero denominator
in (8–42).
Using (2–325) and (2–326) leads to the well-known Stokes’ formula

R
T = S(ψ) ∆g dσ , (8–43)
4π
σ
where
∞
2n + 1
S(ψ) = Pn (cos ψ) , (8–44)
n−1
n=2
where P (cos ψ) are Legendre polynomials. Here ψ denotes the spherical dis-
tance from the point at which T is to be computed.
In exactly the same way, we obtain for the gravity disturbance with the
boundary condition (8–39), summarizing the derivation in Sect. 2.18,
∞

δg(ϑ, λ) = δgn (ϑ, λ) , (8–45)
0
∞
δgn
T (ϑ, λ) = R , (8–46)
n+1
n=0
8.6 Solution by analytical continuation 303
and the formula of Neumann–Koch

R
T = K(ψ) δg dσ , (8–47)
4π
σ
where
∞
2n + 1
K(ψ) = Pn (cos ψ) (8–48)
n+1
n=0
and, by summation of this series,

1 1
K(ψ) = − ln 1 + (8–49)
sin(ψ/2) sin(ψ/2)
being the Neumann–Koch function.

So in the GPS boundary problem on the sphere, the solution (8–47)
is completely analogous to the formula of Stokes (8–43) for the classical
problem.
The fact that the GPS problem is conceptually simpler (fixed-boundary
surface) than Molodensky’s problem (free-boundary surface) is expressed by
the fact that Stokes’ function must start with n = 2, since n = 1 gives a zero
denominator, whereas Neumann–Koch’s function (8–48) is regular for all n.
In both cases, the height anomaly ζ (here the geoidal height) is given by
Bruns’ formula
T . T
ζ= = . (8–50)
γ γ0
In the spherical approximation, γ may be, in formulas of Bruns’ and Stokes’
type, replaced by our usual mean value γ0 = γ45◦ .
We will see that these spherical solutions form the base for an elemen-
tary solution of Molodensky’s problem and the GPS problem for the earth’s
surface. We only mention the well-known fact that, for the earth’s surface S,
these two problems are oblique-derivative problems, since the direction of the
plumb line does not coincide with the normal to the earth’s surface, at least
on land. Thus the GPS boundary problem for S is not a spherical Neumann
problem, which always involves the normal derivative!
8.6 Solution by analytical continuation

8.6.1 The idea
The idea is very simple (Fig. 8.4). Our observations ∆g or δg, given on the
earth’s surface S, are “reduced”, or rather “analytically continued” (upward
or downward, see below and Fig. 8.5), to a level surface (or normal level
Dg earth's surface S
z
P Dg' h
point level U = UP
hP
ellipsoid U = U0
Fig. 8.4. Analytical continuation from the earth’s surface to point level
surface U = UP , which for our purpose is the same). In the spherical ap-
proximation, both surfaces U = UP and U = U0 are concentric spheres, but
only in the precise sense of the spherical approximation as explained above.
We also use the term “harmonic continuation” because the analytically
continued function satisfies Laplace’s equation. This will be explained in
detail later.
An expansion into a Taylor series gives immediately
∂∆g∗ 1 ∂ 2∆g∗ 1 3 ∂ 3∆g∗
∆g = ∆g∗ + z + z2 + z + ···
∂z 2! ∂z 2 3! ∂z 3
∞ (8–51)
1 n ∂ n∆g∗
= ∆g∗ + z ,
n! ∂z n
n=1
where
z = h − hP (8–52)
is the elevation difference with respect to the computation point P . For
the present, we assume the series (8–51) to be convergent. Here ∆g∗ is the
gravity anomaly at point level (Fig. 8.4). The use of a Taylor series is typical
for analytical continuation. For instance, Taylor series are a standard tool
for analytical continuation of functions of a complex variable.
8.6.2 First-order solution

It is particularly easy to give a solution as a first approximation. With γ0
from (8–50) we have

R ∂∆g ∂ζ
ζ= ∆g − h S(ψ) dσ + h. (8–53)
4π γ0 ∂h ∂h
σ
This follows from the geometrical interpretation of this equation which is

evident from Fig. 8.5 a. We see that the free-air anomalies ∆g at ground
level are “reduced” downward to sea level to become
∂∆g
∆gharmonic = ∆g − h (8–54)
∂h
(a)
Dg
P h
hP
Dg harmonic sea level
(b)
Dg
P h – h0
hP – h0 reference level
h0
(c)
Dg h – hP
P point level Dg *
hP
Dg harmonic
Fig. 8.5. Harmonic continuation to sea level (a), to an

arbitrary level (b), and to the level of point P (c)
(the superscript “harmonic” denotes harmonic continuation to sea level; see

Fig. 8.5 and the paragraph “A note on terminology” below); then Stokes’
integral gives height anomalies at sea level which are reduced upward to
∂ζ
ground level by adding the term ∂h h.
Harmonic continuation to point level

The elevation h in (8–53) is taken above sea level (see Fig. 8.5 a). If we
examine the arguments leading to this equation, we will find that the sea
level is not distinguished from any other level. If we reckon the elevation
above some other reference level, which has the elevation h0 above sea level,
we must replace h by h − h0 (see Fig. 8.5 b). Thus (8–53) is equivalent to

R ∂∆g ∂ζ
ζ= ∆g − (h − h0 ) S(ψ) dσ + (h − h0 ) . (8–55)
4π γ0 ∂h ∂h
σ
In particular we may take as reference level the level of the point P itself,
so that
h0 = hP , (8–56)
where P is the point at which the height anomaly ζ is computed. If this
choice is made, the last term in the above expression will be zero, because
outside the integral h always means hP , so that h − h0 = hP − hP = 0. Thus
we have
R ∂∆g
ζ= ∆g − (h − hP ) S(ψ) dσ . (8–57)
4π γ0 ∂h
σ
This formula is particularly simple; geometrically it means that the free-
air anomalies are “reduced” (in the sense of “analytically or harmonically
continued”) from the ground to the level of the computation point P (see
Fig. 8.5 b). Thus, the reference level is different for different computation
points.
As we have already indicated at the beginning of Sect. 8.6.1, Fig. 8.5 c
shows that harmonic continuation by Eq. (8–57) is upward for surface points
below the level of P and downward for surface points above the level of P .
Important remark
Equation (8–57) is really a genuine spherical Stokes formula applied to a
“reference sphere”, namely, to the spherical “point level”! An immediate
consequence: this formula can be simply differentiated horizontally to give a
genuine Vening Meinesz formula in the sense of Sect. 2.19 for the deflections
of the vertical. This remark is relevant for Sect. 8.7.
Vertical derivative
The vertical derivative ∂/∂r can be expressed in terms of surface values by
the well-known spherical formula (Sect. 1.14)

∂f 1 R2 f − fQ
=− f+ dσ . (8–58)
∂r R 2π l03
σ
Q is the surface point where ∂f /∂r is computed and to which f in the first
term on the right-hand side refers, σ denotes the unit sphere, and
ψ
l0 = 2R sin . (8–59)
2
This gives ∂∆g/∂r if we put f = ∆g in (8–58). We may also introduce the
linear gradient operator L by

R2 f − fQ
L(f ) = dσ . (8–60)
2π l03
σ
(The first term on the right-hand side of (8–58) is much smaller and can be
omitted.)
The term ∂ζ/∂r no longer occurs in (8–57) as it did in (8–53) and (8–55),
and will not be needed.
Computational formulas; the Molodensky correction

Our computational formula is (8–57). We split it up as follows: The free-air
anomaly ∆g is continued (downward or upward) from ground level to the
level of point P , obtaining
∆g∗ = ∆g + g1 , (8–61)
where the Molodensky correction is

∂∆g ∂∆g
g1 = − (h − hP ) = − (h − hP ) (8–62)
∂h ∂r
(in spherical approximation) with

∂∆g R2 ∆g − ∆gQ
= dσ . (8–63)
∂r 2π l03
σ
Then we finally get

ζ = ζ0 + ζ1 , (8–64)
where
R
ζ0 = ∆g S(ψ) dσ (8–65)
4π γ0
σ
is the simple Stokes formula applied to ground-level free-air anomalies ∆g,

and the Molodenski correction for ζ is

R
ζ1 = g1 S(ψ) dσ . (8–66)
4π γ0
σ
This is the first-order solution, or linear solution.
Important remark
Please note carefully that we are using “linear”, or “first-order”, in two very
different senses:
• general linearization, linear in quantities of the anomalous potential,

such as N or ζ, as introduced in Sect. 2.12 and Sect. 8.4 and implied
everywhere throughout the book, and
• linear approximation in h used very generally in first-order “Moloden-

sky corrections” such as (8–62) or (8–66) but not in (8–67) or (8–68).
In fact, to a higher approximation
∆g∗ = ∆g + g1 + g2 + g3 + · · · (8–67)
and
ζ = ζ0 + ζ1 + ζ2 + ζ3 + · · · . (8–68)
Generalizing (8–66), we have

R
ζi = gi S(ψ) dσ , (8–69)
4π γ0
σ
where i = 1, 2, 3, . . . . For the deflection of the vertical we have similar

expressions, see Sect. 8.7; compare also (8–75) and (8–76).
8.6.3 Higher-order solution

The following recursion formulas are somewhat advanced and may be omit-
ted. From Moritz (1980 a: Sect. 45) we may take the recursion formula for
the correction terms gn , which are evaluated recursively by
n

gn = − z r Lr (gn−r ) , (8–70)
r=1
starting from
g0 = ∆g . (8–71)
Here the operator Ln is also defined recursively:
Ln (∆g) = n−1 L1 [Ln−1 (∆g)] (8–72)
starting with
L1 = L (8–73)
with the gradient operator L defined above, (8–60), and z given by (8–52).
8.6.4 Problems of analytical continuation

Analytical continuation comes from the theory of complex variables and
means extending the domain, on which the function is defined, by the use of
Taylor series. Complex functions always satisfy Laplace’s equations in two
dimensions and are therefore harmonic.
Also in three dimensions, functions satisfying Laplace’s equation are

called harmonic, as we know well. Analytical continuation is again best de-
fined by Taylor series, and analytical continuation is frequently called har-
monic continuation (Kellogg 1929: Chap. X).
Above we have been misusing the all-round word “reduced” in the sense
of “analytical” or “harmonic” continuation and will continue to do so for
brevity. As we have seen in Sect. 8.2 and will see in Sect. 8.9, it is not a
gravity reduction in the standard sense of explicit mass removal. The Taylor
series whose first term is (8–54) is an analytical operation performed on the
external potential directly at ground level, preserving the Laplace equation
∆W = 0. (In fact, ∆W = 2ω 2 , but let us, as we did in (8–2), for a while for-
get earth rotation, which implies ω = 0 and W = V .) Thus, it is a harmonic
function and our “reduction” is really analytical continuation as a harmonic
function or briefly harmonic continuation. Harmonic continuation is the
key notion in modern physical geodesy, from Molodensky’s prob-
lem to least-squares collocation. Its full meaning will gradually emerge
in what follows, as a notion which is surprisingly simple and general. Sym-
bols like ∆gharmonic will relate to harmonic continuation. In what follows, we
shall sometimes continue to use “reduce downward” or “continue downward”
instead of “harmonically continue downward” and use “reduce upward” in
a similar sense. We also use “continue upward”. Only in doubt, the clumsy
expression “harmonically continue upward” should be employed. Also “ana-
lytical continuation” is used. It all means the same. In the present context,
confusion is hardly possible.
Hence we see why gravity anomalies ∆g at ground level may be used
for f in (8–58), whereas the equivalent expression (2–394) was originally
derived for gravity anomalies at sea level. Since ∆gharmonic and ∆g differ only
by terms of the order of h, the difference between using ∆gharmonic or ∆g
in (8–62) causes only an error of the order of h2 , which is negligible in the
linear approximation.
Analytical continuation: historical remarks

The use of analytical continuation has an interesting history. It was first
considered as a possibility by Molodensky himself, already before 1945, but
he soon rejected this method! Molodensky was a profound mathematician,
with a high regard for mathematical rigor. He would not be satisfied with
intuitive heuristic approaches so common in mathematical physics, also in
the present book.
In fact, the analytical continuation of the external gravitational potential
into the interior of the earth’s masses is very likely to become singular at
some points. As a serious mathematician, Molodensky rejected the use of
singular functions for regular purposes.

Still, analytical continuation continued to exert an irresistable fascina-
tion because its use is so easy. It was rediscovered around 1960 by A. Bjer-
hammar. At the General Assembly of the International Union of Geodesy
and Geophysics in Berkeley, California, in 1963, one of the authors (H.M.)
talked to Bjerhammer about these difficulties, but Bjerhammar refused to
take them seriously. After a long discussion he convinced H.M. that analyt-
ical continuation was rigorously possible for discrete boundary data (all our
terrestrial gravity measurements are discrete) and approximately possible
for continuous boundary data.
This admittedly intuitive thinking was made rigorous by the idea of
Krarup (1969) that Runge’s theorem, well known for approximation of ana-
lytical functions of a complex variable, should be applied to the problem of
analytical continuation of harmonic functions in space. Runge’s theorem, in
the form of Krarup, loosely speaking says that, even if the external geopoten-
tial cannot be regularly continued from the earth surface S into its interior,
it can be made continuable by an arbitrarily small change of the geopotential
at S. Another historical remark: the Krarup–Runge theorem for harmonic
functions in space goes back at least to Szegö and to Walsh (both around
1929), cf. Frank and Mises (1930: pp. 760–762). It is always dangerous to
talk about priorities! A detailed discussion will be found in Moritz (1980 a:
Sects. 6 to 8).
More on the validity of this method

Let us summarize. The presupposition of this method is that the earth’s
external gravitational potential can be continued, as a regular harmonic
function, analytically down to sea level. This is the case if and only if it
is possible to shift the masses outside the ellipsoid into its interior in such
a way that the potential outside the earth remains unchanged or, in other
words, if the analytical continuation of the disturbing potential T is a regular
function everywhere between the earth’s surface and the ellipsoid. Thus, the
question arises whether the external potential can be analytically continued
down to sea level.
Rigorously, as we have just remarked, the answer must be in the negative,
in view of the irregularities of topography (Molodenski et al. 1962: p. 120;
Moritz 1965: Sect. 6.4). This fact is also related to the divergence at the
earth’s surface of the spherical-harmonic expansion for the external potential
(Sect. 2.5).
However, by Krarup–Runge’s theorem, the analytical continuation of the
external potential down to sea level is possible with sufficient accuracy for
all practical purposes. Actually it is possible with any accuracy you wish; if
you are not satisfied with 1 mgal, prescribe 10−3 mgal or 10−1000 mgal!
Bjerhammar has pointed out that the assumption of a complete contin-
uous gravity coverage at every point of the earth’s surface, from which the
above negative answer follows, is unrealistic because we can measure gravity
only at discrete points. If the purpose of physical geodesy is understood as
the determination of a gravity field that is compatible with the given discrete
observations, then it is always possible to find a potential that can be an-
alytically continued down to the ellipsoid. This is the theoretical basis for
least-squares collocation.
Here we need only one result: Do not worry about analytical contin-
uation! It is always possible with an arbitrarily small error being
not equal to 0 (though not for one being 0).
So, in the same year 1969, Marych and Moritz independently found an el-
ementary solution by analytical continuation in the form of an infinite series
denoted as “Molodensky series”. Details can be found in Moritz (1980 a):
The original form of Molodensky’s series obtained by solving an integral
equation is found in Sect. 45. Pellinen’s equivalence proof that the simple
“analytical continuation solution” and Molodensky’s integral equation solu-
tion are equivalent (that means, the series are termwise equal!) is found in
Sect. 46.
We remark that analytical continuation is a purely mathematical concept
independent of the density of the topographic masses. Thus, it is not an
“introduction of gravity reduction through the backdoor”, which would be
contrary to the spirit of Molodensky’s theory.
8.6.5 Another perspective

Consider Fig. 8.6. Let us assume that the analytical downward continuation
of ∆g to the sea level surface has been performed, obtaining ∆gharmonic . The
sea-level anomalies ∆gharmonic then generate, on the physical surface of the
earth, a field of gravity anomalies ∆g that is identical with the actual gravity
anomalies on the earth’s surface as obtained from observation. Therefore, the
gravity anomalies that they generate outside the earth must also be identical
P
Dg
hP ground
h
sea level
Dg harmonic
Fig. 8.6. Free-air anomalies at ground level, ∆g, and at sea level, ∆g harmonic
with the actual gravity anomalies outside the earth, since the function r ∆g
is harmonic according to Sect. 2.14.
(Remark: we are consistently using the notation ∆g for ground level,
∆gharmonic for sea level, and ∆g∗ for point level; see Fig. 8.5.)
It follows that the harmonic function T that is produced by ∆gharmonic
according to Pizzetti’s generalization (2–302) of Stokes’ formula

R
T (r, ϑ, λ) = ∆gharmonic S(r, ψ) dσ (8–74)
4π
σ
is identical with the actual disturbing potential of the earth outside and on
its surface.
Applications
Assume that we got in some way (e.g., by the Taylor series mentioned above
or by collocation to be treated in Chap. 10 or by a high-resolution gravita-
tional field from satellite observations) the downward continuation ∆gharmonic
to sea level. Then we can compute the external gravity field, its spherical
harmonics, etc., rigorously by means of the conventional formulas of Chaps. 2
and 6, provided we use ∆gharmonic rather than ∆g in the relevant formulas.
For instance, the coefficients of the spherical harmonics of the gravitational
potential may be obtained by expanding the function ∆gharmonic according to
Sect. 1.9 together with Sect. 1.6. If we wish to compute the height anomaly
ζ at a point P at ground level, we must remember that P lies above the el-
lipsoid, so that the formulas for the external gravity field are to be applied.
By Bruns’ formula ζ = T /γ0 (8–50), we get

R
ζ= ∆gharmonic S(r, ψ) dσ , (8–75)
4π γ0
σ
where r = R + h and h is the topographic height of P in some sense of

Chap. 4. (We do not need it very accurately, but it means that h is formally
added to the constant radius R of the mean terrestrial sphere, which has no
real-world geometric interpretation!) Cf. Eq. (6–57). The function S(r, ψ) is
expressed by (2–303), (6–22) or (6–35). Similarly, ξ and η, being deflections
of the vertical above sea level, must be computed by (6–41) and the second
and third equation of (6–30). This gives

t ∂S(r, ψ)
ξ= ∆gharmonic cos α dσ ,
4π γ0 ∂ψ
σ
(8–76)
t harmonic ∂S(r, ψ)
η= ∆g sin α dσ ,
4π γ0 ∂ψ
σ
where ∂S(r, ψ)/∂ψ is expressed by the second equation of (6–32) or the

second equation of (6–36). The linear approximation of (8–74) is evidently
equivalent to (8–53).
This indirect procedure, downward continuation to sea level and again
upward continuation to ground level or above, has the advantage that only
the conventional spherical formulas are needed; yet at the same time the
irregularities of the earth’s topography are fully taken into account. The
downward continuation of ∆g need be performed only once; the resulting
anomalies ∆gharmonic may be stored and used for all further computations.
Just as ∆g is related to ∆gharmonic by analytical continuation, so are ζ and
harmonic
N , the height of a “harmonic geoid”. A final and hopefully instructive
and not too difficult review will be found in Sect. 8.15.
An elementary explanation from daily life

Generally, “analytical continuation” means continuation by the same math-
ematical formula: Taylor series, Laplace equation, or even an elementary
explicit equation.
Let us illustrate the meaning of analytical continuation by means of an
almost trivial example from everyday life (Fig. 8.7). A person is driving a
car along a road which at first is completely straight; at point B, however,
it suddenly turns into a circular curve. Thus, our person first drives along
the straight segment of the road. Unfortunately, he is tired and sleepy just
when the straight road suddenly turns into a circular curve. Thus, our sleepy
driver fails to turn the steering wheel and goes straight ahead, the car leaving
the road. Fortunately, the slope is mild, the driver immediately takes control
again and manages to bring the car to a stop at C without major damages.
The driver (one of the authors of this book) has even found the experience
an excellent example to illustrate analytical continuation in his courses!
The gravitational potential corresponds to the road ABC, which, after
some idealization, can be considered “piecewise analytic”, consisting of the
straight line AB and the circular arc BC. The transition from the straight
line to the circle is continuous and differentiable at B, but the curvature
C'
B
A
Fig. 8.7. An illustration of analytical continuation

changes discontinuously from 0 to 1/R, where R is the radius of the circular

arc. Therefore, the function “road” is continuous and continuously differ-
entiable but has a discontinuous second derivative at point B, just as the
function “gravitational potential” is everywhere continuous and continuously
differentiable but has discontinuous second derivatives at the earth surface
as we have seen in Sect. 1.2. The straight line has the equation y = 0
(which is the “one-dimensional Laplace equation”!), corresponding to the
external potential satisfying ∆V = 0. Thus, neither the “function road” nor
the “function gravitational potential” may be considered an everywhere ana-
lytical function, but each may be said to consist of a “linear” or “harmonic”
piece (y = 0 or ∆V = 0: Laplace, respectively) and a “nonlinear” piece
(y = 0 or ∆V = −4π G : Poisson). For the road, the analytical continu-
ation is the straight line for which y = 0 even beyond point B, the path
followed by the car without action of the sleepy driver towards C , and for
the potential it is a function satisfying ∆Vanalytical continuation = 0 even in the
interior of the earth.
8.7 Deflections of the vertical

The effect of Molodensky-type corrections is even much more important
on the deflections of the vertical ξ, η than on the height anomalies ζ. This
is shown by their order of magnitude in high mountains: the Molodensky
correction for the height anomaly might be of the order of 0.3 m, whereas
for vertical deflections they may be on the order of 0.3 arc seconds, which
corresponds to 10 m (1 arc second corresponds to 30 m in position). The
difference is more than one order of magnitude!
The consideration of a Molodensky type of correction to the deflec-
tions of the vertical is easiest by using analytical continuation to point level
(Sect. 8.6). Differentiating (8–57) in north-south and east-west direction, we
get the corresponding Vening Meinesz equations

1 ∂∆g dS
ξ= ∆g − (h − hP ) cos α dσ ,
4π γ0 ∂h dψ
σ
(8–77)
1 ∂∆g dS
η= ∆g − (h − hP ) sin α dσ .
4π γ0 ∂h dψ
σ
Its geometrical interpretation is analogous to that of (8–57). The gravity

anomalies ∆g are “reduced” to the level of point P so that we obtain
∂∆g
∆gharmonic = ∆g − (h − hP ) . (8–78)
∂h
8.7 Deflections of the vertical 315
Since these anomalies refer to a level surface, Vening Meinesz’ formula can
now be applied directly and gives (8–77).
Relation with the ellipsoidal geodetic coordinates

The deflection components ξ and η as given by the above expressions rep-
resent the deviation of the actual plumb line from the normal plumb line at
the ground point P . Therefore, they are defined by
ξ = Φ − ϕ∗ ,
(8–79)
η = (Λ − λ∗ ) cos ϕ .
The symbols Φ and Λ represent the astronomical coordinates of P referred to

the ground. The symbols ϕ∗ and λ∗ represent the “normal coordinates” of P ,
defining the direction of the normal plumb line at P ; they are not identical
with the ellipsoidal coordinates ϕ and λ of P , which are the coordinates of
the foot point Q0 of the straight perpendicular to the ellipsoid (Fig. 8.8).
ellipsoidal normal
normal
plumb line P
earth's
surface
parallel to
'* equator
' ''
Q0 Q00 ellipsoid
Fig. 8.8. Normal latitude ϕ∗ and ellipsoidal latitude ϕ
The normal coordinates of P , ϕ∗ and λ∗ , differ from the normal coordi-

nates of Q00 , ϕ and λ , by the correction for the normal curvature of the
plumb line (see Sect. 5.15). Formula (5–147) gives
h
ϕ∗ = ϕ + f ∗ sin 2ϕ ,
R (8–80)
λ∗ = λ .
Because of the rotational symmetry, we have rigorously
λ = λ , (8–81)
since Q0 and Q00 lie on the same ellipsoidal meridian. Furthermore, even
in extreme cases the distance between Q0 and Q00 can never exceed a few
centimeters. For this reason, we may also set
ϕ = ϕ (8–82)
without introducing a perceptible error. Hence, we can identify ϕ and λ with

ϕ and λ, which are the ellipsoidal coordinates of P according to Helmert’s
projection (Sect. 5.5). Therefore, we may replace the above equations for ϕ∗
and λ∗ by
h
ϕ∗ = ϕ + f ∗ sin 2ϕ ,
R (8–83)
λ∗ = λ .
Introducing the deflection components according to Helmert’s projection,
defined as
ξHelmert = Φ − ϕ ,
(8–84)
ηHelmert = (Λ − λ) cos ϕ ,
we see that they are related to ξ and η by the equations
h
ξHelmert = ξ + f ∗ sin 2ϕ ,
R (8–85)
ηHelmert = η .
Therefore, ξ and ξHelmert differ by the normal reduction for the curvature of
the plumb line,
h
−δϕnormal = f ∗ sin 2ϕ . (8–86)
R
The deflection components ξHelmert and ηHelmert are used in astrogeodetic com-
putations; ξ and η are those obtained gravimetrically from formulas such as
(8–77) and (8–88) below.
These relations are mathematically quite analogous to the corresponding
equations (5–138) for the conventional method using the geoid, but now,
with the use of the normal curvature, the once formidable obstacle of the
correction for plumb-line curvature practically belongs to the past.
Remark on accuracy
With Molodensky’s theory, the accuracy problem mentioned at the end of
Sect. 2.21 even aggravates, because in a mountainous terrain it is almost
impossible to compute the Molodensky corrections with an accuracy of 0.03
(say), so that these observations cannot be directly used for precise horizontal
positions.
8.8 Gravity disturbances: the GPS case 317
Astronomical field observations for latitude, longitude, and azimuth have

an accuracy around 0.3 , which is sufficient for classical trigonometric net
computation and astrogeodetic observation of the geoid (Sect. 5.14).
8.8 Gravity disturbances: the GPS case

The basic fact is that for gravity disturbances the derivation of “Molodensky
corrections” gn is identical to the ∆g case. The reason is that the gravity dis-
turbance δg has exactly the same analytical behavior as the gravity anomaly
∆g since r δg, as a function in space, is harmonic together with r ∆g. Thus,
the arguments are literally the same, only ∆g has to be replaced by δg, and
Stokes’ formula must be replaced by the Neumann–Koch formula (8–47) and
similarly for Vening Meinesz’ formula.
Therefore, we obtain
∞

R R
ζ= δg K(ψ) dσ + gn K(ψ) dσ , (8–87)
4π γ0 4π γ0
σ n=1 σ
1∞
1 dK dK
ξ= δg cos α dσ + gn cos α dσ ,
4π γ0 dψ 4π γ0 dψ
σ n=1 σ
∞ (8–88)
1 dK 1 dK
η= δg sin α dσ + gn sin α dσ .
4π γ0 dψ 4π γ0 dψ
σ n=1 σ
For the “Vening Meinesz GPS formula” (8–88), we find by differentiation of

(8–49):
dK 1 cos(ψ/2) 1
=− 2 . (8–89)
dψ 2 sin (ψ/2) 1 + sin(ψ/2)
The correction terms gn are evaluated recursively by
n

gn = − z r Lr (gn−r ) , (8–90)
r=1
but now we start from

g0 = δg . (8–91)
We only have to replace ∆g by δg and S(ψ) by K(ψ). The operators L
remain the same.
Let us summarize again our trick for solving the modern boundary-value
problems (Molodensky and Koch). It is difficult to directly work with the
complicated earth’s surface S. Therefore, by analytical continuation of ∆g or
δg, respectively, we reduce these complicated problems to the corresponding

spherical problems, for which the solution is simple and well known.
The similarity of the Molodensky series for the Molodensky problem, on
the one hand, and for the GPS boundary problem, on the other hand, is very
clear because ∆g and δg have the same analytical and geometric structure.
At the same time, this similarity is very surprising since the two underly-
ing boundary problems are mathematically quite different, as we have seen
in Sect. 8.3 (compare Eqs. (8–12) and (8–13)). Nonetheless, (8–87) does give
the potential as (8–13) requires: by Bruns’ theorem, which is the omnipresent
link between geometry and physics, we have
T =γζ. (8–92)
Then
W =U +T (8–93)
is the geopotential required by (8–13), and
C = W0 − W (8–94)
is the geopotential number, the physical measure of height above sea level,
conventionally obtained by the cumbersome method of leveling, but now
computed in a direct way from gravity data. This is the physical, more
general, equivalent of the geometric determination of the normal height by
H ∗ = h − ζ, according to Eq. (8–31).
It can be shown that, in the linear approximation, the Molodensky cor-
rection for the gravity disturbance has the same form as for the gravity
anomaly and can for each quantity be computed using either ∆g or δg.
The formulas for the Molodensky corrections and their numerical values
are the same to the linear approximation.
All this shows the power of Molodensky’s approach even in problems he
never treated himself.
8.9 Gravity reduction in the modern theory

In Sect. 8.2, we have considered gravity reductions from the point of view
of the determination of the geoid. It is quite remarkable that these reduc-
tions, such as the Bouguer or the isostatic reduction, can also be incorpo-
rated into the new method of direct determination of the earth’s physical
surface, although with essentially changed meaning (Pellinen 1962; Moritz
1965: Sect. 5.2).
8.9 Gravity reduction in the modern theory 319
Let the masses outside the geoid be removed or moved inside the geoid, as
described in Sect. 8.2, and consider the effect of this procedure on quantities
referred to the ground.
We denote the changes in potential and in gravity by δW and δg; then
the new values at ground will be
W c = W − δW ,
(8–95)
gc = g − δg .
(It is clear that δg here is not the gravity disturbance!) The disturbing
potential T = W − U becomes
T c = T − δW . (8–96)
The physical surface S as such will remain unchanged, but the telluroid Σ will
change, because its points Q are defined by UQ = WP , and the potential W at
any surface point P will be affected by the mass displacements according to
(8–95). The distance Q Qc between the original telluroid Σ and the changed
telluroid Σc (Fig. 8.9) is given by
δW
Q Qc = (8–97)
γ
according to Bruns’ theorem. This is identical with the variation of the height
anomaly ζ, so that
δW
δζ = ζ − ζ c = . (8–98)
γ
Normal gravity γ on the telluroid Σ becomes on the changed telluroid Σc
∂γ 1 ∂γ
γc = γ + δζ = γ + δW , (8–99)
∂h γ ∂h
³c
³ earth's surface S
Qc
±³
Q changed telluroid Sc
telluroid S
ellipsoid E
Fig. 8.9. Telluroid before and after gravity reduction, Σ and Σc

so that the new gravity anomaly will be

1 ∂γ
∆g = g − γ = (g − δg) − γ +
c c c
δW (8–100)
γ ∂h
or
1 ∂γ
∆gc = ∆g − δg − δW . (8–101)
γ ∂h
The reduced gravity anomaly ∆gc consists of the free-air anomaly (in
the Molodensky sense) ∆g and two reductions:
1. the direct effect, −δg, of the shift of the outer masses on g; and
2. the “indirect effect”,

1 ∂γ
− δW , (8–102)
γ ∂h
of this shift on γ, because of the change of the telluroid to which γ
refers.
Let us repeat once more that all these anomalies ∆gc refer to the physical
surface of the earth, to “ground level”!
If the masses outside the geoid are completely removed, then ∆gc is
a Bouguer anomaly; if the outer masses are shifted vertically downward
according to some isostatic hypothesis, then ∆gc is an isostatic anomaly, etc.
In this way we may get a “ground equivalent” for each conventional gravity
reduction. The two are always related by analytical continuation. See below
for the isostatic anomalies; for analytical continuation see Sect. 8.6.
Now we may describe the determination of the height anomalies ζ in a
way that is similar to the corresponding procedure for the geoidal undula-
tions N of Sect. 8.2:
1. The masses outside the geoid are, by computation, removed entirely or

else moved inside the geoid; W and g change to W c and gc according
to (8–95).
2. The point at which normal gravity is computed is moved from the
ellipsoid upward to the telluroid point Q.
3. The indirect effect, the distance Q Qc = δζ, is computed by (8–98).
4. The point to which normal gravity refers is now moved from the point
Q of the telluroid Σ to the point Qc of the changed telluroid Σc , ac-
cording to (8–99).
5. The changed height anomalies ζ c are computed from the “reduced”
gravity anomalies ∆gc (8–101) by any solution of Molodensky’s prob-
lem, such as Eq. (8–57) or (8–68).
6. Finally, the original height anomalies ζ are obtained by considering the

indirect effect according to
ζ = ζ c + δζ . (8–103)
The purpose of this somewhat complicated procedure is to make use of

the well-known advantages of Bouguer and isostatic anomalies. The Bouguer
anomalies, and even more so the isostatic anomalies, are smoother and more
representative than the free-air anomalies and can, therefore, be interpolated
more easily and more accurately.
The isostatic gravity anomalies ∆gc in the new sense are thus quite anal-
ogous to the conventional isostatic anomalies; accordingly for any other type
of gravity reduction. The difference is that now the isostatic anomalies, etc.,
refer to the physical surface of the earth as well as the free-air anomalies. If
the isostatic anomalies in this new sense are analytically continued from the
earth’s surface down to the geoid, then isostatic anomalies in the conven-
tional sense are obtained. Nowadays, in view of the “remove-restore princi-
ple”, one speaks usually of topographic-isostatic reduction while continuing
to speak of isostatic anomalies.
Hence, the isostatic anomalies according to the conventional definition
(at sea level) and those according to the new definition (at ground level) are
related through analytical continuation. This fact leads to two conclusions.
First, the difference between the isostatic anomalies according to these two
definitions will be small, because the distance along which this analytical
continuation is made is only the height above sea level and because the
isostatic reduction achieves a strong smoothing of the anomalous gravity
field. This difference is considerably smaller than the corresponding differ-
ence between free-air anomalies at ground level and at sea level. This fact
clearly provides a computational advantage if isostatic anomalies are used
in a formula such as (8–74).
Second, we obtain a relation between the conventional and the modern
use of gravity reduction if the method of downward continuation, as discussed
in the preceding section, is applied for obtaining the height anomalies. As we
have just seen, the gravity anomalies ∆gc∗ at sea level, obtained by downward
continuation of the isostatic ground-level anomalies ∆gc , are identical with
the isostatic anomalies in the conventional sense. Hence, we obtain on the
one hand the height anomalies by

R c∗ δW
ζ= ∆g S(R + h, ψ) dσ + (8–104)
4π γ γ ground
σ
according to (8–75) and (8–103), and on the other hand the geoidal undula-
tions by
R c∗ δW
N= ∆g S(ψ) dσ + (8–105)
4π γ0 γ geoid
σ
according to the ordinary Stokes formula applied to ∆gc∗ and (8–5). Since
the height anomalies refer to the elevation h, the function S(R+h, ψ) replaces
in (8–104) the original function of Stokes S(ψ) ≡ S(R, ψ), which occurs in
(8–105) because the geoidal undulation refers to zero elevation. We could
use γ0 in (8–104) as well. Summarizing, we have the following steps:
1. Computation of the free-air anomaly at ground level, ∆g, according to
(8–23).
2. Computation of the isostatic anomaly at ground level, ∆gc , according
to (8–101).
3. Downward continuation of ∆gc by (8–54), where ∆g and ∆gharmonic are
replaced by ∆gc and ∆gc∗ . The resulting isostatic anomalies at sea
level, ∆gc∗ , may now be used for two purposes: either for
4a. the determination of the physical surface of the earth according to
(8–104), or for
4b. the determination of the geoid according to (8–105).
An error in the assumed density of the masses below the earth’s surface
affects the geoidal undulations as determined from (8–105) but does not
influence the height anomalies resulting from (8–104). This is clear because
a wrong guess of the density means only that the masses above sea level are
not completely removed, which is no worse than not removing them at all
when using free-air anomalies.
This method is of particular interest for practical computations, as we
will see later. It has become popular by the name “remove-restore method”,
invented by K. Colic and others, see Sect. 11.1.
An almost final remark on free-air reduction

The apparently so simple topic of free-air reduction in reality is formidably
complex and complicated. Therefore, it is not possible to treat it in one
block. The problem is rather like a mountain which can only be investigated
by accessing it from various sides. An initial glance has been given as early
as in Chap. 3, and the reader is asked to return to the paragraph “The
many facets of free-air reduction” in Sect. 3.9. Now it is much easier to un-
derstand the remarks made there. What we now understand as harmonic
continuation offers a possibility to interpret free-air reduction as a mass-
transporting gravity reduction: the topographic masses are transported into
the interior of the earth in such a way that the exterior potential remains
unchanged. This is not unlike the Rudzki reduction, where the geoid remains
unchanged. Whereas the Rudzki reduction is, however, “constructive” in the
sense that a way of performing it can be described, our present interpreta-
tion of free-air reduction as harmonic continuation is nonconstructive, it is
an “improperly posed” inverse problem; cf. Anger and Moritz (2003) and
www.inas.tugraz.at/forschung/InverseProblems/AngerMoritz.html, as well
as Fig. 8.10.
S topographic masses
(a)
geoid
W W
(DW = 0) (DW = –4¼G%)
W =W0
S S
(b) cogeoid (c) harmonic
geoid
W harmonic
Wc (DW harmonic = 0)
Wc (DW = 0)
c W
(DW = 0)
(DW c = 0)
W c =W0 W harmonic =W0
Fig. 8.10. (a) Geoid and topographic masses, (b) mass displacement in gravity
reduction, (c) “ill-defined” mass displacement in free-air reduction as harmonic
continuation
Important remark
The isostatic gravity anomalies and the topographic-isostatically reduced
deflections of the vertical (Sect. 8.14) are fundamental for least-squares col-
location in mountain areas (Sect. 11.2). Thus, the spatial approach due
to Molodensky is basic even for least-squares collocation!
Exercise
Collecting all these remarks into a separately readable paper on the various
aspects of free-air reduction would be a nice task for a seminar work. The
present authors offer a prize of Euro 500, the “Molodensky Prize”, to the
first excellent review paper on this topic.
8.10 Determination of the geoid from ground-level

anomalies
We have seen that it is possible to determine the physical surface of the earth
by means of the height anomalies ζ, and the direction of the plumb line on
it by means of the deflection components ξ and η, from free-air anomalies
referred to the ground. If we plot the orthometric height H downward along
the plumb line, starting from the physical surface, then the locus of the
points so obtained will be the geoid (Fig. 8.11).
This geometrical idea may be formulated analytically in the following
way. Conventionally, the height h above the ellipsoid is given by
h=H +N; (8–106)
according to the modern theory, by
h = H∗ + ζ . (8–107)
From these two equations we get
N − ζ = H∗ − H . (8–108)
P
³
Q earth's
surface
H
telluroid
H*
P0 geoid
N
ellipsoid
Q0
Fig. 8.11. Geoid at a depth H below the earth’s surface

8.10 Determination of the geoid from ground-level anomalies 325
This means that the difference between the geoidal undulation N and the
height anomaly ζ is equal to the difference between the normal height H ∗ and
the orthometric height H. Since ζ is also the undulation of the quasigeoid,
this difference is also the distance between geoid and quasigeoid.
According to Sect. 4.5, the two heights are defined by
C C
H= , H∗ = , (8–109)
ḡ γ̄
where C is the geopotential number, ḡ is the mean gravity along the plumb
line between geoid and ground, and γ̄ is the mean normal gravity along the
normal plumb line between ellipsoid and telluroid. By eliminating C between
these two equations, we readily find
ḡ − γ̄
H∗ − H = H, (8–110)
γ̄
which is also the distance between the geoid and the quasigeoid, see (8–108);
hence
ḡ − γ̄
N =ζ+ H. (8–111)
γ̄
The height anomaly ζ may be expressed, for instance, by Molodensky’s
formula (8–57). Then we obtain

R R ḡ − γ̄
N= ∆g S(ψ) dσ + g1 S(ψ) dσ + H , (8–112)
4π γ0 4π γ0 γ̄
σ σ
where g1 is the term (8–62). Thus N is given by Stokes’ integral, applied to

free-air anomalies at ground level, and two small corrections, where
1. the term containing g1 represents the effect of topography;
2. the term containing ḡ − γ̄ represents the distance between the geoid

and the quasigeoid.
If we neglect these two corrections, then the geoidal undulations N are

given by Stokes’ integral using free-air anomalies. This was first noted by
Stokes in 1849. A new approach by Jeffreys (1931) by means of Green’s iden-
tities started several developments which culminated in the work of Molo-
densky and others.
The advantage of this method for the determination of N is that the
density of the masses above sea level enters only indirectly, as an effect
on the orthometric height H through the mean gravity ḡ, which must be
computed by a Prey reduction (Sect. 3.5). Hence, as far as errors in the
density are concerned, the geoidal undulation N as obtained by this method

is as accurate as the orthometric height.
As a matter of fact, the gravity anomaly ∆g in this method refers to
ground level; it is the difference between gravity at ground and normal grav-
ity at the telluroid. Instead of using directly this free-air anomaly, we may
also use other gravity anomalies – for instance, the isostatic anomaly in the
sense of Sect. 8.9.
To repeat a simple but fundamental principle: ∆g, δg, ξ, η, ζ as obtained
by Molodensky’s theory primarily always refer to the physical earth’s sur-
face and not to sea level!
8.11 A first balance

The new methods described in this chapter are primarily intended for the
determination of the physical surface of the earth, but they are also well
suited for the determination of the geoid (Sect. 8.10). Their essential fea-
ture is that the gravity anomalies now refer to the ground, whether we deal
with free-air anomalies or with isostatic or other similarly reduced gravity
anomalies (Sect. 8.9).
The immediate result is the height anomaly ζ, the separation between
the geopotential and the corresponding spheropotential surface at ground
level. By plotting the height anomalies above the ellipsoid, we get the quasi-
geoid. This geoid-like surface has no physical significance, but it furnishes a
convenient visualization of the height anomalies. By plotting the orthometric
height from the earth’s surface vertically downward, we obtain the geoid.
It is instructive to compare the geoid and the quasigeoid. The geoidal
undulation N and ζ, the undulation of the quasigeoid, are related by (8–
111), or
ḡ − γ̄
N −ζ = H = H∗ − H . (8–113)
γ̄
The term ḡ − γ̄ is approximately equal to the Bouguer anomaly; this may
be seen by using (4–32) for γ together with
. 1 ∂γ
γ̄ = γ − H. (8–114)
2 ∂h
The quantity γ̄ in the denominator can be replaced by our usual constant
γ0 . Since the Bouguer anomaly is rather insensitive to local topographic
irregularities, the coefficient is locally constant so that there is approximately
a linear relation between ζ and the local irregularities of the height H. In
other words, the quasigeoid mirrors the topography (Fig. 8.12).
8.11 A first balance 327
earth's
surface
quasigeoid
³
ellipsoid
Fig. 8.12. Quasigeoid
To get a quantitative estimate of the difference N − ζ, we again use the

fact that
ḡ − γ̄ . ∆gB .
= = 10−3 ∆gB , (8–115)
γ̄ 981 gal
where ∆gB is the Bouguer anomaly in gal, so that
.
(ζ − N )[m] = −∆gB [gal] · H[km] . (8–116)
Since ∆gB is usually negative on the continents, the differences ζ − N are

usually positive there. In other words, the height anomaly ζ is in general
greater than the corresponding geoidal undulation N on land. We have ζ =
N on the oceans. If ∆gB = −100 mgal = −0.1 gal and H = 1 km, then
ζ − N = 0.1 m . (8–117)
Furthermore, the Bouguer anomaly depends on the mean elevation of the

terrain, decreasing approximately by 0.1 gal per 1 km average elevation. As-
suming as a rough estimate, which may be verified by inspecting maps of
Bouguer anomalies,
∆gB [gal] = −0.1 H[av
km]
, (8–118)
we obtain
.
(ζ − N )[m] = +0.1 H[av
km]
H[km] , (8–119)
where H is the height of the station and H av is an average height of the
area considered. We see that the difference ζ − N increases faster than the
elevation, almost as the square of the elevation. As a matter of fact, this
formula is suited only to give an idea of the order of magnitude (see also
Sect. 11.3).
Since ζ − N = H − H ∗ , the approximate formulas given above may also
be used to estimate the differences between the orthometric height H and
the normal height H ∗ .
A theoretically important point is that the quasigeoid can be determined
without hypothetical assumptions concerning the density, but not so the
geoid. The avoidance of such assumptions has been the guiding idea of Molo-
densky’s research. However, orthometric heights are but little affected by er-
rors in density. The error in H due to the imperfect knowledge of the density
hardly ever exceeds 1–2 decimeters even in extreme cases (Sect. 4.3). It is
presumably smaller than the inaccuracy of the corresponding ζ even with
very good gravity coverage, because of inevitable errors of interpolation, etc.
If, therefore, the method of Sect. 8.10 is used, the geoid can be determined
with virtually the same accuracy as the quasigeoid. Note that it is theoret-
ically even possible to eliminate completely the errors arising from the use
of the geoid (Moritz 1962, 1964). Thus, we may well retain the geoid with
its physical significance and its other advantages.
How much do Molodensky’s formulas differ from the corresponding equa-
tions of Stokes and Vening Meinesz? The deviation of ζ from the result of
the original Stokes formula is given by the equivalent expressions

R R ∂∆g
ζ1 = g1 S(ψ) dσ or ζ1 = − (h − hP ) S(ψ) dσ
4π γ0 4π γ0 ∂h
σ σ
(8–120)
according to Eqs. (8–62) and (8–66). This correction may even be smaller
than the difference ζ − N (see Sect. 11.3).
It is appropriate again to point out that the deflection of the vertical
is relatively more affected by the Molodensky correction than is the height
anomaly. In extreme cases, this correction may attain values of a few seconds,
as studies of models by Molodensky (Molodenski et al. 1962: pp. 217–225)
indicate. This is considerable, since 1 in the deflection corresponds to 30 m
in position. Numerical estimates will be found in Chap. 11.
We may summarize the result of applying Stokes’ and Vening Meinesz’
formulas to free-air anomalies directly, without any corrections. Stokes’ for-
mula yields height anomalies ζ with high accuracy; for many practical pur-
poses, we may, in addition, identify these height anomalies with the corre-
sponding geoidal undulations N . Vening Meinesz’ formula gives deflections
of the vertical at ground level that are relatively less accurate but often
acceptable.
An advantage of the modern theory is its direct relation to the external
gravity field of the earth, which is particularly important nowadays for the
computation of the effect of gravitational disturbances on spacecraft trajec-
tories and satellite orbits. It is immediately clear that ground-level quan-
tities, such as free-air gravity anomalies, are better suited for this purpose
than the corresponding quantities referred to the geoid, which is separated
from the external field by the outer masses. For the computation of the ex-
ternal field and of spherical harmonics, the method described in Sect. 8.6.5
8.12 Some background 329
is particularly appropriate (see also Sect. 6.5).

Practically it is usually adequate to consider only the linear approxima-
tion by using (8–57). In many cases it is even possible to neglect the correc-
tion −(∂∆g/∂h) h, identifying the sea-level free-air anomalies ∆gharmonic with
the corresponding ground-level anomalies ∆g. In agreement with Sect. 3.9,
these free-air anomalies ∆gharmonic = ∆g may also be considered approxima-
tions to condensation anomalies in the sense of Helmert. This approximation
is particularly sufficient for the external gravity field, spherical harmonics,
and geoidal undulations or height anomalies. For deflections of the vertical,
it is often necessary to use a more careful approach, such as the consideration
of the indirect effect with mass-transporting gravity reductions (Sect. 8.2)
or the modern methods of Sect. 8.9.
In high and steep mountains, the approach of Molodensky and others
through free-air anomalies encounters practical difficulties, such as unrelia-
bility of interpolation, large corrections, and other computational problems.
To avoid this, isostatic reduction in the modern sense shoud be used. Thus
the clash between “conventional” (geoid) and “modern” (Molodensky-type)
ideas gives way to an important synthesis. For another synthesis, see least-
squares collocation in Sects. 10.2 and 11.2.
For further study, especially of the historic aspects, the reader is referred
to the book by Molodenski et al. (1962) and the M.S. Molodensky Anniver-
sary Volume edited by Moritz and Yurkina (2000).
Part II: Astrogeodetic methods according to

Molodensky
8.12 Some background

The computation of a detailed geoid, or of a detailed gravity potential field,
in limited areas, especially in mountainous regions, has not been very much
in the focus of attention recently. There may be various reasons for this.
For decades now, global geoid determinations, either from satellite data
or from a combination of satellite and gravimetric data have been in the
center of interest (Lerch et al. 1979, Reigber et al. 1983, Rapp 1981). Even
(almost) purely gravimetric global and local geoids have been successfully
computed (March and Chang 1979), between the classic Heiskanen (1957)
and the modern local geoid (Kühtreiber 2002 b). An excellent recent reference
volume is that by Tsiavos (2002).
Over the oceans, the geoid is now known to an accuracy of perhaps a
few centimeters, due to satellite altimetry. Unfortunately, satellite altimetry
does not work over land areas. The classical method for a detailed geoid
determination on the continents has been the gravimetric method, in spite
of the fact that it is severely handicapped by lack of an adequate gravity
coverage (or lack of information on such a coverage). Thus, we have the
paradoxical situation that on the oceans, long a stepchild of geodesy, the
geoid is now in general known much better than on the continents.
Still, the gravimetric method has continued to fascinate theoreticians
because it gives rise to very interesting and deep mathematical problems,
related to the geodetic boundary-value problem discussed above in this chap-
ter.
These enormous practical and theoretical developments concerning global
satellite and gravimetric gravity field determination have somewhat over-
shadowed the determination of detailed geoids in smaller areas, in partic-
ular, astrogeodetic geoids. Especially in mountainous regions, local geoid
determinations are difficult. The gravimetric method does not work very
well in high mountains. The astrogeodetic method, using astronomical ob-
servations of latitude and longitude, does work well there but is considered
time-consuming and somewhat old-fashioned, perhaps also because work-
ing during the night is not very popular nowadays. An appropriate use
of gravity and astrogeodetic data in high mountains must involve some
topographic-isostatic reduction. Furthermore, the theory behind the astro-
geodetic method is not nearly as attractively difficult as the theory of Molo-
densky’s problem. Last but not least, high-mountain areas are exceptional
and, apart from such countries as Switzerland and Austria, are frequently
regions of little economic interest. For these and similar reasons, the main-
stream of geodetic practice and theory has flown with grand indifference
around high mountains, ignoring such trivial obstacles.
Still, a country such as Switzerland has made a virtue out of necessity
and has traditionally been very active in local astrogeodetic geoid determi-
nation (Elmiger 1969, Gurtner 1978, Gurtner and Elmiger 1983). Austria
has followed up (Österreichische Kommission für die Internationale Erdmes-
sung 1983). It has been found that, even besides the problem of getting the
required observations, the underlying theory is not so trivial as one might
think and shows quite interesting features.
Concerning measurements, astronomical observations have again proved
very feasible in mountains; see the articles by Erker, Bretterbauer and Gerst-
bach, Lichtenegger and Chesi in Chap. 2 of Österreichische Kommission für
die Internationale Erdmessung (1983), followed by Sünkel et al. (1987). The
main advantages of astrogeodetic versus gravimetric data for local geoid
determination in mountainous regions may be summarized as follows:
1. It is sufficient to have astrogeodetic deflections of the vertical in the
region of geoid determination; no data are needed outside that region

as they would be in the gravimetric method.
2. Errors in the topographic height have significantly less influence on
deflections than on gravity data. Thus, a relatively crude terrain model
will be sufficient for the use of astrogeodetic data.
As a matter of fact, the two types of data are not mutually exclusive;
an optimal geoid determination will combine astrogeodetic deflections of
the vertical, gravity anomalies, and possibly data of other type. A suitable
technique for this purpose is least-squares collocation to be discussed in
Chap. 10.
From the observational point of view it is interesting to note that inertial
surveying techniques will be able to furnish deflections of the vertical and
gravity anomalies rapidly and with sufficient accuracy for many purposes.
Let us finally try to give a list of various methods of geoid determination:
• conventional satellite techniques (Doppler, laser, etc.),
• satellite-to-satellite tracking,
• satellite gradiometry,
• satellite altimetry,
• gravimetry,
• astrogeodesy, and,
• most directly, GPS leveling (Sect. 4.6).
As a general rule, these methods are listed in such a way as to start with
the most global and end up with the most local method, that is, according
to decreasing globality or increasing locality. In general, going down the list
also corresponds to increasing resolution and accuracy.
Again it should be stressed that these methods complement each other
and should be combined for best results.
New satellite gravity missions have been discussed in Sect. 7.6.
Astrogeodetic method according to Molodensky

The remaining part of this chapter deals primarily with the lower end of the
list, providing a detailed theory of astrogeodetic local geoid determination
in areas with difficult topography. The role (and necessity) of topographic-
isostatic reduction is investigated in detail. The computations for Austria
give concrete numerical results for questions which have been much dis-
cussed theoretically, such as the difference between geoidal heights and height
anomalies according to Molodensky (quasigeoidal heights), or the numeri-
cal effect of analytical continuation from the earth’s surface to sea level
(Österreichische Kommission für die Internationale Erdmessung 1983).
earth's surface
telluroid
P
actual level surface W = WP
³
Q normal level surface U = WP
"
–± "
normal plumb line
(curved)
actual plumb line
ellipsoidal normal
(straight)
P0 P0' P0'' geoid W = W0
N "0
ellipsoid U = W0
Q0 Q0' Q0''
Fig. 8.13. The basic geometry
As a warm-up, let us return to basics and remember some main principles

of Molodensky’s geometry. Figure 8.13 illustrates the basic quantities. In
the classical theory, the geoid is defined by its deviation N from a reference
ellipsoid; N is the geoidal height. The geoid is a level surface W = W0 =
constant of the gravity potential W ; the ellipsoid is defined to be the level
surface U = U0 = constant of a normal gravity potential U ; the constants
W0 and U0 are usually assumed to be equal (Sect. 2.12).
For the modern theory according to Molodensky (Sect. 8.4), to each point
P of the earth’s surface we associate a point Q in such a way that Q lies on
the straight ellipsoidal normal through P and that
U (Q) = W (P ) . (8–121)
That is, Q is defined such that its normal potential U equals the actual
potential W of P .
This corresponds to the classical relation
U0 = U (Q0 ) = W (P0 ) = W0 (8–122)
mentioned above, by which U0 is taken to be equal to W0 (Fig. 8.13). By the

same correspondence, the height anomaly according to Molodensky,
ζ = QP , (8–123)
is the modern equivalent of the classical geoidal height,
N = Q0 P0 . (8–124)
Using the anomalous potential
T =W −U, (8–125)
we have according to Bruns’ theorem

T T
ζ= , N= , (8–126)
γ Q γ Q0
where γ denotes the ellipsoidal normal gravity.

The points P0 form the geoid, and the points Q0 constitute the ellipsoid,
both being level surfaces (of W and U , respectively). On the other hand, the
points P form the earth’s surface, and the set of points Q defines an auxiliary
surface, denoted as telluroid according to R.A. Hirvonen. As a matter of fact,
neither the earth’s surface nor the telluroid are level surfaces, which makes
matters more complicated than in the classical situation, where we deal with
level surfaces.
Following a suggestion of Molodensky, one could plot the height anoma-
lies ζ as vertical distances from the reference ellipsoid. Thus one obtains a
geoid-like surface, the quasigeoid, and ζ could be considered as quasigeoidal
heights. In contrast to the geoid, however, the quasigeoid is not a level surface
and does not admit of a natural physical interpretation. Therefore, working
with height anomalies ζ, it is best to consistently consider them quantities
referred to the earth’s surface (vertical distances between earth surface and
telluroid), rather than using the quasigeoidal concept. A summary will be
given in Sect. 8.15.
The classical gravity anomaly ∆g0 at sea level is defined as
∆g0 = g(P0 ) − γ(Q0 ) , (8–127)
where g denotes gravity and γ normal gravity. So far, g(P0 ) denotes the ac-
tual gravity on the geoid; we are not yet here considering mass-transporting
gravity reductions.
Analogously we have according to Molodensky:
∆g = g(P ) − γ(Q) . (8–128)
Generally we will, as far as feasible, use the subscript “0” to designate quan-
tities referred to sea level, to distinguish them from quantities referred to
the earth’s surface, which do not carry such a subscript. For instance, ∆g0
refers to sea level and ∆g to the earth’s surface. With GPS we have gravity
disturbances
∆g = g(P ) − γ(P ) . (8–129)
Regarding plumb line definition, we must distinguish three lines (Fig. 8.13):
1. the straight ellipsoidal normal Q0 P ,
2. the actual plumb line P0 P ,
3. the normal plumb line P0 P .
Geometrically, the ellipsoidal normal is defined as the straight line through
P perpendicular to the ellipsoid. The (actual) plumb line is defined by the
condition that, at each point of the line, the tangent coincides with the
gravity vector g at that point; the plumb line is very slightly curved, but its
curvature is irregular, being determined by the irregularities of topographic
masses. The normal plumb line, at each of its points, is tangent to the normal
gravity vector γ; it possesses a curvature that is even smaller and completely
regular.
The points P0 , P0 , and P0 coincide within a few decimeters, and we will
not distinguish them in what follows. The reason is that the distance, in arc
seconds, between P0 and P0 is much smaller than the effect of plumb line
curvature (Sect. 5.15). The same applies for Q0 , Q0 , and Q0 .
The direction of the gravity vector g is the direction of (the tangent to)
the plumb line. It is determined by two angles, the astronomical latitude Φ
and the astronomical longitude Λ. Let Φ, Λ be referred to the earth’s surface
(to point P ) and Φ0 , Λ0 to the geoid (strictly speaking, to point P0 ). The
differences
δϕ = Φ0 − Φ , δλ = Λ0 − Λ (8–130)
express the effect of plumb line curvature (Fig. 8.14). You may also wish to
refer back to Fig. 5.18. Hence, we have
Φ0 = Φ + δϕ , Λ0 = Λ + δλ . (8–131)
Knowing the plumb line curvature δΦ, δΛ, we could use these simple formulas
to compute the sea-level values Φ0 , Λ0 from the observed surface values Φ, Λ.
In the same way as Φ, Λ are related to the actual plumb line, the ellip-
soidal latitude ϕ and the ellipsoidal longitude λ refer to the straight ellip-
soidal normal. The quantities
ξ = Φ − ϕ, η = (Λ − λ) cos ϕ (8–132)
are the components of the deflection of the vertical in a north-south and an
east-west direction. For an arbitrary azimuth α, the vertical deflection ε is
given by
ε = ξ cos α + η sin α . (8–133)
±'
parallel to F0
equator F
P0''
Fig. 8.14. Curvature of the plumb line along a north-south profile
These quantities ξ, η, ε refer to the earth’s surface. Figure 8.13 shows ε.

Similarly, we have for the geoid
ξ0 = Φ 0 − ϕ , η0 = (Λ0 − λ) cos ϕ , (8–134)
ε0 = ξ0 cos α + η0 sin α . (8–135)

See again Fig. 8.13 for ε0 , noting that we do not distinguish the normals in
Q0 and Q0 as we have mentioned above.
In addition, we need the normal direction of the plumb line at the surface
point P ; it is defined as the tangent to the normal plumb line at P ; the
corresponding latitude and longitude will be denoted by ϕ̄, λ̄. In this “local”
notation, there is no danger of confusion with the spherical coordinate ϕ̄
used in earlier chapters. Hence, we have
ϕ = ϕ̄ + δϕnormal , λ = λ̄ + δλnormal , (8–136)
where δϕ, δλ express the normal plumb line curvature. These equations are
the “normal equivalent” to (8–131): the “normal surface values” ϕ̄, λ̄ cor-
respond to the “actual surface values” Φ, Λ and the ellipsoidal values ϕ, λ
correspond to the geoidal values Φ0 , Λ0 . To make the analogy complete, we
should replace ϕ = ϕ(P0 ) by ϕ(P0 ), but we have consistently neglected such
differences.
In contrast to the actual plumb line curvature, it is very easy to compute
the normal curvature of the plumb line: from (5–147) we have
δϕnormal = −0.17 h [km] sin 2ϕ , δλnormal = 0 , (8–137)
where h [km] denotes elevation in kilometers.

Since the ellipsoidal normal and hence ϕ, λ are geometrically defined, we
may call the quantities (8–132) “geometric deflections of the vertical” at the
earth’s surface. On the other hand, the normal plumb line is physically (or
dynamically) defined by means of the external gravity field of an equipo-
tential ellipsoid. Hence also ϕ̄, λ̄ are dynamically defined. The quantities
obtained by replacing ϕ, λ by ϕ̄, λ̄ so that
ξ̄ = Φ − ϕ̄ , η̄ = (Λ − λ̄) cos ϕ , (8–138)
are called “dynamical deflections of the vertical” at the earth’s surface. By

(8–136) and (8–137) we have
ξ̄ = ξ + δϕnormal , η̄ = η , (8–139)
since δλnormal = 0. For an azimuth α we accordingly have
ε̄ = ξ¯ cos α + η̄ sin α . (8–140)
Compare ε and ε̄ in Fig. 8.13 and note that in this figure δ denotes the
curvature of the normal plumb line for the azimuth α given by the analogous
formula
δ = δϕnormal cos α + (δλnormal cos ϕ) sin α = δϕnormal cos α . (8–141)
8.13 Astronomical leveling revisited

From Fig. 8.15 we take the well-known differential relation
dN = −ε0 ds , (8–142)
where ε0 denotes the deflection of the vertical at the geoid. Integration be-
tween two points A and B yields the difference between their geoidal heights:
B
NB − NA = − ε0 ds , (8–143)
A
plumb line ellipsoidal normal

"0 geoid
"0 dN
N
ds ellipsoid
Fig. 8.15. Astronomical leveling according to Helmert

8.13 Astronomical leveling revisited 337
or, approximately,
ε0A + ε0B
NB − NA = − sAB , (8–144)
2
where sAB denotes the horizontal distance between A and B. The minus
sign is conventional. Cf. Sect. 5.14.
A corresponding relation to height anomalies according to Molodensky
is found as follows (Molodensky et al. 1962: p. 125):
∂ζ ∂ζ
dζ = ds + dh , (8–145)
∂s ∂h
notations following Fig. 8.16. Since the earth’s surface is not a level surface,
we also have a vertical part (∂ζ/∂h) h in addition to the usual horizontal
part (∂ζ/∂s) ds. The vertical part arises from change in height and is usually
smaller than the horizontal part.
In analogy to (8–142), the horizontal part is given by
∂ζ
= −ε̄ , (8–146)
∂s
where ε̄ denotes the dynamical deflection of the vertical at the earth’s surface;
cf. (8–140) and Fig. 8.13. For the vertical part we have from (8–126):

∂ζ ∂ T 1 ∂T 1 ∂γ
= = − T (8–147)
∂h ∂h γ γ ∂h γ ∂h
or
∂ζ ∆g g−γ
=− =− (8–148)
∂h γ γ
according to the fundamental equation of physical geodesy (8–36).
Hence (8–145) becomes
g−γ
dζ = −ε̄ ds − dh . (8–149)
γ
earth's surface
dh
W = WP
P @³
" @s
ds
ds U = WP
Fig. 8.16. Astronomical leveling according to Molodensky

Integrating this relation yields the difference of the height anomaly

B B
∆g
ζB − ζA = − ε̄ ds − dh ; (8–150)
A A γ
the gravity anomaly ∆g refers to the earth’s surface according to (8–128).

The first term on the right-hand side represents the Helmert integral (8–143)
of the surface deflection ε̄, and the second term is Molodensky’s correction to
the Helmert integral, necessary to obtain height anomalies. This correction
depends on the gravity g at the earth’s surface.
8.14 Topographic-isostatic reduction of vertical

deflections
For the reasons mentioned at the end of the preceding section, it is nat-
ural to try and find a way which makes use of the clear advantages of the
topographic-isostatic reduction but avoids the problems inherent in a free-air
reduction from the surface point P to the geoidal point P0 .
In Sect. 8.9, we have treated the reduction of gravity from the modern
point of view. The second formula of (8–95) is
gc = g − δg . (8–151)
Everything is referred to the ground point P , and δg = δgTI is the effect

of gravity reduction on g, also at P . In the topographic-isostatic reduction
which we use here exclusively, it is the gravitational attraction of the to-
pography minus the gravitational attraction of the compensating isostatic
masses, topography minus isostasy.
To get the topographic-isostatic gravity anomaly, we subtract normal
gravity γ, also referred to ground level, more precisely, to the corresponding
telluroid point Q. Thus,
∆gc = ∆g − δgTI . (8–152)
The explanation is trivial: you are standing at point P and watch how the
topography is removed to fill the isostatic mass deficits, but by a miracle
you are still hovering at P , now in “free air”.
Application to deflections of the vertical

The gravity anomaly is only one component of the anomalous gravity vector,
the other two being the vertical components ξ and η, both, of course, mul-
tiplied by γ to get the dimensions right. Thus, ξ and η can be isostatically
reduced in exactly the same way.
8.14 Topographic-isostatic reduction of vertical deflections 339
For ξ and η, (8–152) becomes

ξ c = ξ − ξTI + δϕnormal , η c = η − ηTI . (8–153)
By means of (8–139) this may be written
ξ c = ξ̄ − ξTI , η c = η̄ − ηTI . (8–154)
The interpretation of (8–154), however, is clear, simple, and rigorous: from
the dynamic deflections of the vertical at P , which are the very quantities
ξ̄ and η̄, we subtract the effect of the topographic-isostatic masses, ξTI and
ηTI likewise at P . The vertical deflections so obtained, ξ c and η c , thus do
not really refer to the (co-)geoid; in reality, they refer to the earth’s surface!
But what, then, means the normal plumb line curvature δϕnormal in (8–
153)? Does it not mean a reduction from the earth’s surface to sea level? No,
in Eqs. (8–139) it only denotes the transformation between the geometrical
and the dynamical deflection of the vertical, both referred to the point P
of the earth’s surface. This is also clear from Fig. 8.13, which illustrates the
formula
ε̄ = ε + δ , (8–155)
extending (8–139) to an arbitrary azimuth, δ being defined by (8–141).
This interpretation of (8–153) or (8–154) as isostatically reduced deflec-
tions of the vertical at the earth’s surface is exact, whereas the interpretation
of (8–8) as deflections at the cogeoid was only approximate. This is the de-
sired rigorous interpretation of our isostatically reduced vertical deflections.
This interpretation exactly corresponds to the modern view of gravity
reduction according to the theory of Molodensky. According to this view,
the isostatically (or in some other way) reduced gravity anomalies continue
to refer to the earth’s surface. The classical gravity reduction (Sect. 8.2) had
comprised two procedures: mass transport and shift P → P0 ; the new view
of gravity reduction only considers the mass transport; the problematic shift
P → P0 is avoided.
Formally, a “normal free-air reduction”
∂γ
F =− h (8–156)
∂h
may be said to occur also in Molodensky’s theory: normal gravity γ in the
new definition (8–128) of the gravity anomaly, where it refers to the telluroid
point Q, is computed by
∂γ
γ = γQ0 + h, (8–157)
∂h
with h = Q0 Q denoting the normal height of P . But instead of reducing
actual gravity g downward, from P to P0 , now normal gravity is reduced
upward from Q0 to Q. Whereas for the first process the use of the normal
gradient ∂γ/∂h is problematic, it is fully justified for the second process.
In a similar way, we might interpret δϕnormal as a reduction of ϕ for
normal curvature of the plumb line upwards, say, from P0 to P . This is
possible because in (8–136) ϕ could be said to refer to P0 (because P0 and
P0 practically coincide), and because ϕ̄ denotes the latitude of the tangent
to the normal plumb line at P . This interpretation is instructive because of
the analogy with gravity reduction, though regarding ϕ and ϕ̄ as ellipsoidal
and dynamic latitude of the same point P appears more natural. Refer again
to our key figure (Fig. 8.13).
As pointed out above, the present interpretation of ξ c and η c as isostati-
cally reduced deflections of the vertical at the earth’s surface is conceptually
rigorous and therefore also practically more accurate, but this decisive ad-
vantage implies a computational drawback if integration along a profile is
used: Since this integration must now be performed along the earth’s surface
and not along a level surface such as the geoid, computation will be more
complicated. Instead of the simple Helmert formula (8–143), we now must
use the Molodensky formula (8–150):
B B
gc − γ
c
ζB − c
ζA =− ε ds −
c
dh (8–158)
A A γ
with
εc = ξ c cos α + η c sin α , (8–159)
and ∆gc = gc −γ, where gc is the isostatically reduced surface value of gravity
(measured value g minus attraction of the topographic-isostatic masses).
From the isostatic height anomalies ξ c obtained in this way, we then get
the actual height anomalies ζ by applying the indirect effect:
ζ = ζ c + δζ (8–160)
with
TTI
δζ = . (8–161)
γ
This is completely analogous to (8–5) and (8–3), but now TTI is the potential
of the topographic-isostatic masses at the surface point P . As a matter of
fact, normal gravity in (8–3) refers to the ellipsoid, and in (8–161) to the
telluroid, but the difference is generally small.
For higher mountains, the isostatic reduction procedure described in the
present section is preferable in practice to a direct application of Moloden-
sky’s formula (8–150) because the isostatically reduced vertical deflections
8.15 The meaning of the geoid 341
are much smoother and easier to interpolate. It is, however, extremely la-
borious from a computational point of view since the integration must be
performed along the earth’s surface (or, what is practically the same, along
the telluroid).
We remark that the computational drawback of the present method,
the Molodensky integration along the earth’s surface, can be completely
avoided if we perform our computations in space: instead of integrating
along a surface, we perform collocation in space. This modern procedure,
to be described in the next chapter, permits a simple and computationally
convenient use of surface deflections and also their combination with gravi-
metric and other data. Still, the present developments are necessary for a
full understanding of the collocation approach.
Final remarks
In these last sections we tried to apply the same principle for topographic-
isostatic reduction (the “remove-restore method”) at point level to all terres-
trial data related to the gravity vector: gravity anomalies and disturbances
(Sect. 8.9) and deflections of the vertical (Sect. 8.14). This unified view of
isostatically reduced data thus makes them directly suitable for combined
solutions by least-squares collocation to be treated in Chap. 10.
8.15 The meaning of the geoid

We now review the geoid and some surfaces that might be able to replace it.
We will again confirm the unique role of the geoid as a standard surface of
physical geodesy.
The meaning of the geoid is very simple. It is defined in Sect. 2.2 as
one of the equipotential surfaces (level surfaces, surfaces of constant gravity
potential)
W (x, y, z) = constant. (8–162)
The constant is chosen so that, on the oceans, the geoid coincides with mean
sea level:
W (x, y, z) = W0 . (8–163)
This is the usual classical equation of the geoid. So what is the problem?
Well, theory and practice are different, in geodesy as well as in daily life.
First, we must disregard small tidal effects (on the order of 50 cm). This is
done by applying a suitable tidal model and is not too problematic. In fact,
we have numerous geoids determined from satellite observations. Second,
they are usually expressed in terms of a series of spherical harmonics. If
taken at sea level, such a series may diverge (this is related to the difficulties
of downward continuation, cf. Sect. 8.6). Such a possible divergence may

concern mathematicians, but it should not concern geodesists, for several
reasons:
1. Our spherical-harmonic expansions are not infinite series but finite
polynomials, by their very determination and computations. So diver-
gence problems do not exist; the question is only good approximation.
2. Such approximating polynomials of spherical harmonics always exist
for arbitrary accuracy requirements (Frank and Mises 1930: p. 760).
In geodesy we usually speak of Runge’s theorem. The whole subject is
thorougly discussed in Moritz (1980 a: Sects. 6 to 8).
3. If you use spatial collocation, the behavior (harmonic or not, conver-
gent or divergent, ...) of the solution is completely determined by the
covariance function used. One always uses “good” covariance functions,
which are harmonic and analytic down to a sphere completely inside
the earth.
So forget all about the convergence problem. It is practically solved. Further

discussions beyond the results obtained so far would have to be made at a
very high mathematical level. The question can be made as complicated as
desired; if looked at it from the right angle, it is simple.
Geoid and downward continuation

Therefore, and by the reasoning at the end of Sect. 10.1, the geoid computed
by (harmonic!) spherical-harmonic expressions and by collocation is not a
level surface of the actual geopotential W but a level surface of a harmonic
downward continuation of W , for the simple reason that the base functions
both of spherical harmonics and of collocation satisfy Laplace’s equation
(8–2). We may speak of a “harmonic geoid”. This again emphasizes the
importance of analytical continuation (Sect. 8.6). We have deliberately used
the indefinite article “a” in the italicized expression above, because harmonic
downward continuation is an inverse problem and thus has no unique solution
(see below).
The application of collocation to ξ, η, ∆g without gravity reduction gives
height anomalies ζ and undulations of the harmonic geoid, N harmonic , by
simply varying the elevation parameter (h and zero, respectively) in the col-
location program. A completely analogous fact was remarked at the end of
the last section for the case of height anomalies ζ c and cogeoidal heights
N c . In the case of Molodensky’s problem (without or with gravity reduc-
tion), we have seen a completely similar behavior with the application of the
generalized Stokes and Vening Meinesz formulas, (8–75) and (8–76).
8.15 The meaning of the geoid 343
Geoid, harmonic geoids, and quasigeoid

The geoid in the usual sense of Eqs. (2–18) or (8–163) is defined purely by
nature and is independent of geodetic observations (except for the tidal cor-
rections). Its disadvantage is that it depends on the “topographic masses”
above the geoid whose density is unknown, at least in principle. This draw-
back seems to be theoretical rather than practical.
The harmonic geoids are equipotential surfaces of an analytical down-
ward continuation. We shall be careful to denote the harmonically continued
potential by W harmonic so that
W harmonic = W0 = constant (8–164)
denote harmonic geoid(s).

To repeat, analytical downward continuation based on discrete data at
the earth’s surface is an inverse problem (Sect. 1.13; for more informa-
tion see www.inas.tugraz.at/forschung/InverseProblems/AngerMoritz.html)
which has infinitely many possible solutions. For collocation, e.g., each solu-
tion corresponds to the choice of a different covariance function.
Thus, the “harmonic geoid” is not uniquely defined. It is a product not
only of nature but also of the computational method used. It cannot, there-
fore, replace the real geoid as a standard surface.
The “cogeoids” of the various gravity reductions (Sect. 8.2) are inter-
mediate computational concepts and should never be used in place of the
geoid. The topographic-isostatic height anomalies at point level, ζ c , and the
heights of the topographic-isostatic cogeoid, N c , are related to each other by
analytical continuation. The same collocation formula applies if the height
anomaly f (P ) is computed at sea level with elevation parameter 0 to give
N c , or, if f (P ) is computed at point level with elevation parameter h, to
give ζ c . (The elevation parameter h is a height above sea level in any of the
definitions of Chap. 4.) See item 5 at the end of Sect. 10.2.
For the limiting case of Fig. 8.5 c, take the question: “How is the undu-
lation N harmonic of a ‘harmonic geoid’ related to the height anomaly ζ above
it on the ground and on the same vertical?” Answer: “By analytic continu-
ation!”
Another special question to which the answer is also easy: “Which gravity
reduction leaves the geoid unchanged?” Answer: “The Rudzki reduction”
(Sect. 3.8). So why not use it? It changes the external potential, which today
is of paramount importance.
“What is the difference between the Rudzki reduction and the harmonic
downward continuation?” Answer: “The Rudzki reduction leaves the geoid
unchanged but changes the external geopotential: there is W c = W = W0
only on the geoid, but W c = W outside the earth, which is inadmissible.
The harmonic continuation leaves the external geopotential unchanged but

changes the geoid: W harmonic = W outside the earth and on the earth’s sur-
face, but W harmonic = W at sea level.
Height anomalies and quasigeoid

The height anomalies ζ refer to the physical earth’s surface. They find their
natural physical interpretation in Hirvonen’s telluroid. Molodensky proposed
to plot ζ above the reference ellipsoid and get the “quasigeoid”. Thus, ζ gives
the quasigeoid in exactly the same way as the geoidal height N gives the
geoid. However, this analogy is purely formal. There is no way to interpret
the quasigeoid as a surface of constant potential or find any other physical
interpretation for it. Again, it cannot replace the real geoid as a standard
surface.
Thus, in spite of all modern developments, the geoid retains its role as
a standard reference for physical geodesy. However, the reader must have
a clear view of all the concepts reviewed in this section, see Forsberg and
Tscherning (1997).
A final remark on the many facets of free-air reduction

Now, dear reader, having struggled through almost the whole book, you will
be able to understand the disjecta membra on free-air reduction strewn all
over it, such as Sects. 3.3, 3.9, 8.2, 8.6, 8.9, and the present section. Have a
couple of nice mountaineering tours!
9 Statistical methods in
physical geodesy
9.1 Introduction
Some of the most important problems of gravimetric geodesy are formulated
and solved in terms of integrals extended over the whole earth. An example
is Stokes’ formula. Thus, in principle, we need the gravity g at every point
of the earth’s surface. As a matter of fact, even in the densest gravity net
we measure g only at relatively few points so that we must estimate g at
other points by interpolation. In large parts of the oceans we have made no
observations at all; these gaps must be filled by some kind of extrapolation.
Mathematically, there is no difference between interpolation and extrap-
olation; therefore they are denoted by the same term, prediction.
Prediction (i.e., interpolation or extrapolation) cannot give exact values;
hence, the problem is to estimate the errors that are to be expected in the
gravity g or in the gravity anomaly ∆g. As usual, gravity disturbances δg
are appropriately comprised whenever we speak of gravity anomalies.
Since ∆g is further used to compute other quantities, such as the geoidal
undulation N or the deflection components ξ and η, we must also investigate
the influence of the prediction errors of ∆g on N, ξ, η, etc. This is called
error propagation, which will play a basic role.
It is also important to know which prediction method gives highest ac-
curacy, either in ∆g or in derived quantities N, ξ, η, etc. To be able to find
these “best” prediction methods, it is necessary to have solved the previous
problem, to know the prediction error of ∆g and its influence on the derived
quantities.
Summarizing, we have the following problems:
1. estimation of interpolation and extrapolation errors of ∆g (or δg);

2. estimation of the effect of these errors on derived quantities (N, ξ, η,
etc.);
3. determination of the best prediction method.
Since we are interested in the average rather than the individual errors,
we are led to a statistical treatment. This will be the topic of the present
chapter.
346 9 Statistical methods in physical geodesy
9.2 The covariance function

It is quite remarkable that all the problems mentioned above can be solved
by means of only one function of one variable, without any other information.
This is the covariance function of the gravity anomalies.
First we need a measure of the average size of the gravity anomalies ∆g.
If we form the average of ∆g over the whole earth, we get the value zero:

1
M {∆g} ≡ ∆g dσ = 0 . (9–1)
4π
σ
The symbol M stands for the average over the whole earth (over the unit
sphere); this average is equal to the integral over the unit sphere divided
by its area 4π. The integral is zero if there is no term of degree zero in the
expansion of the gravity anomalies ∆g into spherical harmonics, that is, if a
reference ellipsoid of the same mass as the earth and of the same potential
as the geoid is used. This will be assumed throughout this chapter.
Note that if this is not the case, that is, if M {∆g} = m = 0, then we
may form new gravity anomalies ∆g∗ = ∆g − m by subtracting the average
value m. Then M {∆g∗ } = 0 and all the following developments apply to the
“centered” anomalies ∆g∗ .
Clearly, the quantity M {∆g}, which is zero, cannot be used to charac-
terize the average size of the gravity anomalies. Consider then the average
square of ∆g,

1
var{∆g} ≡ M {∆g2 } = ∆g2 dσ . (9–2)
4π
σ
It is called the variance of the gravity anomalies. Its square root is the root
mean square (rms) anomaly:

rms{∆g} ≡ var{∆g} = M {∆g2 } . (9–3)
The rms anomaly is a very useful measure of the average size of the gravity
anomalies; it is usually given in the form
rms{∆g} = ± 35 mgal ; (9–4)
the sign ± expresses the ambiguity of the sign of the square root and sym-
bolizes that ∆g may be either positive or negative. The rms anomaly is very
intuitive; but the variance of ∆g is more convenient to handle mathemati-
cally and admits an important generalization.
9.2 The covariance function 347
Instead of the average square of ∆g, consider the average product of

the gravity anomalies ∆g ∆g at each pair of points P and P that are at a
constant distance s apart. This average product is called the covariance of
the gravity anomalies for the distance s and is defined by
covs {∆g} ≡ M {∆g ∆g } . (9–5)
The average is to be extended over all pairs of points P and P for which
P P = s = constant.
The covariances characterize the statistical correlation of the gravity
anomalies ∆g and ∆g , which is their tendency to have about the same
size and sign. If the covariance is zero, then the anomalies ∆g and ∆g are
uncorrelated or independent of one another (note that in the precise lan-
guage of mathematical statistics, zero correlation and independence are not
quite the same, but we may neglect the difference here!); in other words,
the size or sign of ∆g has no influence on the size or sign of ∆g . Gravity
anomalies at points that are far apart may be considered uncorrelated or
independent because the local disturbances that cause ∆g have almost no
influence on ∆g and vice versa.
If we consider the covariance as a function of distance s = P P , then we
get the covariance function C(s) mentioned at the beginning:
C(s) ≡ covs {∆g} = M {∆g ∆g } (P P = s) . (9–6)
For s = 0, we have
C(0) = M {∆g2 } = var{∆g} (9–7)
according to (9–2). The covariance for s = 0 is the variance.
A typical form of the function C(s) is shown in Fig. 9.1. For small dis-
tances s (1 km, say), ∆g is almost equal to ∆g, so that the covariance is
almost equal to the variance; in other words, there is a very strong corre-
lation. The covariance C(s) decreases with increasing s because then the
C (s)
Fig. 9.1. The covariance function

anomalies ∆g and ∆g become more and more independent. For very large
distances, the covariance will be very small but not in general exactly zero
because the gravity anomalies are affected not only by local mass distur-
bances but also by regional factors. Therefore, we may expect an oscillation
of the covariance between small positive and negative values.
Note that positive covariances mean that ∆g and ∆g tend to have the
same size and the same sign; negative covariances mean that ∆g and ∆g
tend to have the same size and opposite sign. The stronger this tendency,
the larger is C(s); the absolute value of C(s) can, however, never exceed the
variance C(0).
The practical determination of the covariance function C(s) is somewhat
problematic. If we were to determine it exactly, we should have to know grav-
ity at every point of the earth’s surface. This we obviously do not know; and
if we knew it, then the covariance function would have lost most of its signif-
icance because then we could solve our problems rigorously without needing
statistics. As a matter of fact, we can only estimate the covariance function
from samples distributed over the whole earth. But even this is not quite
possible at present because of the imperfect or completely missing gravity
data over the oceans. For a discussion of sampling and related problems see
Kaula (1963, 1966 b).
The first comprehensive estimate of the covariance function was made by
Kaula (1959). Some of his values are given in Table 9.1 for historical interest.
They refer to free-air anomalies. The argument is the spherical distance
s
ψ= (9–8)
R
corresponding to a linear distance s measured on the earth’s surface; R is a
mean radius of the earth. The rms free-air anomaly is
√
rms{∆g} = 1201 = ± 35 mgal . (9–9)
We see that C(s) decreases with increasing s and that, for s/R > 30◦ , very
small values oscillate between plus and minus.
For some purposes we need a local covariance function rather than a
global one; then the average M is extended over a limited area only, instead
of over the whole earth as above. Such a local covariance function is useful
for more detailed studies in a limited area – for instance, for interpolation
problems. As an example we mention that Hirvonen (1962), investigating the
local covariance function of the free-air anomalies in Ohio, found numerical
values that are well represented by an analytical expression of the form
C0
C(s) = , (9–10)
1 + (s/d)2
9.2 The covariance function 349
Table 9.1. Estimated values of the covariance

function for free-air anomalies [unit mgal2 ]
ψ C(ψ) ψ C(ψ) ψ C(ψ)

0.0◦ +1201 8◦ +124 27◦ +18
0.5◦ 751 9◦ 104 29◦ +6
1.0◦ 468 10◦ 82 31◦ +8
1.5◦ 356 11◦ 76 33◦ +5
2.0◦ 332 13◦ 54 35◦ −8
2.5◦ 306 15◦ 47 40◦ −12
3.0◦ 296 17◦ 45 50◦ −20
4.0◦ 272 19◦ 34 60◦ −30
5.0◦ 246 21◦ 35 90◦ −4
6.0◦ 214 23◦ 10 120◦ +12
7.0◦ 174 25◦ 20 150◦ −21
where
C0 = 337 mgal2 , d = 40 km . (9–11)
This function is valid for s < 100 km.

In the meantime it has been recognized that a proper determination of
global and local covariance functions is a central practical problem in this
context.
The Tscherning–Rapp covariance model and the COVAXN sub-

routine
The fundamental covariance model by Tscherning and Rapp (1974) and the
subroutine COVAXN (Tscherning 1976) are still very much up to date, as
the following quotation from Kühtreiber (2002 b) shows:
“The global covariance function of the gravity anomalies Cg (P, Q) given
by Tscherning and Rapp (1974, p. 29) is written as
∞
n−1
Cg (P, Q) = A sn+2 Pn (cos ψ) , (9–12)
(n − 2)(n + B)
n=3
where Pn (cos ψ) denotes the Legendre polynomial of degree n; ψ is the spher-

ical distance between P and Q; and A, B and s are the model parameters.
A closed expression for (9–12) is available in (ibid., p. 45).
The local covariance function of gravity anomalies C(P, Q) given by

Tscherning–Rapp can be defined as
∞
n−1
C(P, Q) = A sn+2 Pn (cos ψ) . (9–13)
(n − 2)(n + B)
n=N +1
Modeling the covariance function means in practice fitting the empirically

determined covariance function (through its three essential parameters: the
variance C0 , the correlation length ξ and the variance of the horizontal
gradient G0 ) to the covariance function model. Hence the four parameters
A, B, N and s are to be determined through this fitting procedure. A simple
fitting of the empirical covariance function was done using the COVAXN-
subroutine (Tscherning 1976).
The essential parameters of the empirical covariance parameters for 2489
gravity stations in Austria are 740.47 mgal2 for the variance C0 and 43.5 km
for the correlation length ψ1 . The value of the variance for the horizontal
gradient G0 was roughly estimated as 100 E2 (note that E indicates the
Eötvös unit, where 1 E = 10−9 s−2 ).
With a fixed value B = 24, the following Tscherning–Rapp covariance
function model parameters were fitted: s = 0.997 065, A = 746.002 mgal 2
and N = 76. The parameters were used for the astrogeodetic, the gravimetric
as well as the combined geoid solution.” (End of quotation.)
The Tscherning–Rapp model can be summed to get closed expressions.
Its popularity is due to its comprehensiveness: there are expressions for co-
variances of various quantities derived by covariance propagation (Sect. 10.1),
and to its flexibility since it contains several parameters which can be given
various numerical values.
Remark. The spherical-harmonic expression of the covariance function
is considered in Sect. 9.3. The theory of global and local covariance func-
tions is described in great detail in Moritz (1980 a: Sects. 22 and 23). The
three essential parameters of a local covariance function (variance C0 , corre-
lation length ξ, and curvature parameter G0 ) are also defined there. Funda-
mental numerical studies on local covariance functions have been made by
Kraiger (1987, 1988).
9.3 Expansion of the covariance function

in spherical harmonics
The more or less complicated integral formulas of physical geodesy frequently
take on a much simpler form if they are rewritten in terms of spherical
harmonics. A good example is Stokes’ formula (see Sect. 2.15).
9.3 Expansion of the covariance function in spherical harmonics 351
Unfortunately, this theoretical advantage is in most cases balanced by

the practical disadvantage that the relevant series converge very slowly. In
certain cases, however, the convergence is good. Then the use of spherical
harmonics is very convenient practically; we consider such a case in the next
section.
The spherical-harmonic expansion of the gravity anomalies ∆g may be
written in different ways, such as
∞

∆g(ϑ, λ) = ∆gn (ϑ, λ) , (9–14)
n=2
where ∆gn (ϑ, λ) is the Laplace surface harmonic of degree n; or, more ex-
plicitly,
∞
n

∆g(ϑ, λ) = anm Rnm (ϑ, λ) + bnm Snm (ϑ, λ) , (9–15)
n=2 m=0
where
Rnm (ϑ, λ) = Pnm (cos ϑ) cos mλ ,
(9–16)
Snm (ϑ, λ) = Pnm (cos ϑ) sin mλ
are the conventional spherical harmonics; or in terms of fully normalized
harmonics (see Sect. 1.10):
∞
n

∆g(ϑ, λ) = ānm R̄nm (ϑ, λ) + b̄nm S̄nm (ϑ, λ) . (9–17)
n=2 m=0
Here ϑ is the polar distance (complement of geocentric latitude) and λ is the

longitude.
Let us now find the average products of two Laplace harmonics
n

∆gn (ϑ, λ) = ānm R̄nm (ϑ, λ) + b̄nm S̄nm (ϑ, λ) . (9–18)
m=0
These average products are

2π π
1
M {∆gn ∆gn } = ∆gn (ϑ, λ) ∆gn (ϑ, λ) sin ϑ dϑ dλ , (9–19)
4π λ=0 ϑ=0
since the averaging is extended over the whole earth, that is, over the whole
unit sphere. Take first n = n, which gives the average square of the Laplace
harmonic of degree n:
2π π
1 2
M {∆gn } =
2
∆gn (ϑ, λ) sin ϑ dϑ dλ . (9–20)
4π λ=0 ϑ=0
Substituting (9–18) and taking into account the orthogonality relations (1–
83) and the normalization (1–91), we easily find
n

M {∆gn2 } = (ā2nm + b̄2nm ) . (9–21)
m=0
Consider now the average product (9–19) of two Laplace harmonics of differ-
ent degree, n = n. Owing to the orthogonality of the spherical harmonics,
the integral in (9–19) is zero:
M {∆gn ∆gn } = 0 if n = n . (9–22)
In statistical terms this means that two Laplace harmonics of different de-
grees are uncorrelated or, broadly speaking, statistically independent.
In a way similar to that used for the gravity anomalies, we may also
expand the covariance function C(s) into a series of spherical harmonics. Let
us take an arbitrary, but fixed, point P as the pole of this expansion. Thus
spherical polar coordinates ψ (angular distance from P ) and α (azimuth)
are introduced (Fig. 9.2). The angular distance ψ corresponds to the linear
distance s according to (9–8). If we expand the covariance function, with
argument ψ, into a series of spherical harmonics with respect to the pole P
and coordinates ψ and α, we have
∞
n

C(ψ) = cnm Rnm (ψ, α) + dnm Snm (ψ, α) , (9–23)
n=2 m=0
north pole
Ã = const. ®
P Ã
P'
equ
ator
Fig. 9.2. Spherical coordinates ψ, α

9.3 Expansion of the covariance function in spherical harmonics 353
which is of the same type as (9–15). But since C depends only on the distance
ψ and not on the azimuth α, the spherical harmonics cannot contain any
terms that explicitly depend on α. The only harmonics independent of α are
the zonal functions
Rn0 (ψ, α) ≡ Pn (cos ψ) , (9–24)
so that we are left with
∞

C(ψ) = cn Pn (cos ψ) . (9–25)
n=2
The cn ≡ cn0 are the only coefficients that are not equal to zero. We also
use the equivalent expression in terms of fully normalized harmonics:
∞

C(ψ) = c̄n P̄n (cos ψ) . (9–26)
n=2
The coefficients in these series, according to Sects. 1.9 and 1.10, are given
by
2n + 1 2π π
cn = C(ψ) Pn (cos ψ) sin ψ dψ dα
4π α=0 ψ=0
(9–27)
2n + 1 π
= C(ψ) Pn (cos ψ) sin ψ dψ
2 ψ=0
and
cn
c̄n = √ . (9–28)
2n + 1
We now determine the relation between the coefficients cn of C(ψ) in
(9–25) and the coefficients ānm and b̄nm of ∆g in (9–18). For this purpose
we need an expression for C(ψ) in terms of ∆g, which is easily obtained by
writing (9–27) more explicitly. Take the two points P (ϑ, λ) and P (ϑ , λ ) of
Fig. 9.2. Their spherical distance ψ is given by
Here ψ and the azimuth α are the polar coordinates of P (ϑ , λ ) with respect
to the pole P (ϑ, λ).
The symbol M in (9–6) denotes the average over the unit sphere. Two
steps are required to find it. First, we average over the spherical circle of
radius ψ (denoted in Fig. 9.2 by a broken line), keeping the pole P fixed and
letting P move along the circle so that the distance P P remains constant.
This gives 2π
∗ 1
C = ∆g(ϑ, λ) ∆g(ϑ , λ ) dα , (9–30)
2π α=0
where C ∗ still depends on the point P chosen as the pole ψ = 0. Second, we

average C ∗ over the unit sphere:
2π π
1
C ∗ sin ϑ dϑ dλ
4π λ=0 ϑ=0
2π π 2π (9–31)
1
= 2 ∆g(ϑ, λ) ∆g(ϑ , λ ) sin ϑ dϑ dλ dα .
8π λ=0 ϑ=0 α=0
This is equal to the covariance function C(ψ), the symbol M in (9–6) now
being written explicitly:
2π π 2π
1
C(ψ) = 2 ∆g(ϑ, λ) ∆g(ϑ , λ ) sin ϑ dϑ dλ dα . (9–32)
8π λ=0 ϑ=0 α=0
The coordinates ϑ , λ in this formula are understood to be related to ϑ, λ
by (9–29) with ψ = constant, but to be arbitrary otherwise; this expresses
the fact that in (9–6) the average is extended over all pairs of points P and
P for which P P = ψ = constant.
To compute the coefficients cn , substitute (9–32) into (9–27), obtaining

2n + 1 π
cn = C(ψ) Pn (cos ψ) sin ψ dψ
2 ψ=0
2π π
1 2n + 1 2π π (9–33)
= ∆g(ϑ, λ) ∆g(ϑ , λ ) ·
4π 4π λ=0 ϑ=0 α=0 ψ=0
· Pn (cos ψ) sin ψ dψ dα · sin ϑ dϑ dλ .
Consider first the integration with respect to α and ψ. According to (1–89),

we have

2n + 1 2π π
∆g(ϑ , λ ) Pn (cos ψ) sin ψ dψ dα
4π α=0 ψ=0
π
2n + 1 2π
= ∆g(ϑ , λ ) Pn (cos ψ) sin ϑ dϑ dλ = ∆gn (ϑ, λ) ,
4π
λ =0 ϑ =0
(9–34)
the change of integration variables being evident. Hence (9–33) becomes
2π π
1
cn = ∆g(ϑ, λ) ∆gn (ϑ, λ) sin ϑ dϑ dλ . (9–35)
4π λ=0 ϑ=0
This may also be written
cn = M {∆g ∆gn } . (9–36)

9.4 Interpolation and extrapolation of gravity anomalies 355
Into this we now insert (9–14), which we write

∞

∆g(ϑ, λ) = ∆gn (ϑ, λ) , (9–37)
n =2
denoting the summation index by n instead of n. We get

, ∞ ∞

cn = M ∆gn ∆gn = M {∆gn ∆gn } . (9–38)
n =2 n =2
According to (9–22), only the term with n = n is different from zero so that
from (9–21) we finally obtain
n

cn = M {∆gn2 } = (ā2nm + b̄2nm ) . (9–39)
m=0
Hence, cn is the average square of the Laplace harmonic ∆gn (ϑ, λ) of degree
n, or its variance. For these reasons the cn are also called degree variances.
The “degree covariances” are zero because of (9–22).
Equation (9–39) relates the coefficients ānm and b̄nm of ∆g and cn of C(s)
in the simplest possible way. Note that ānm and b̄nm are coefficients of fully
normalized harmonics, whereas cn are coefficients of conventional harmonics.
As a matter of fact, we may also use the anm and bnm (conventional) or
the c̄n (fully normalized); but then (9–39) will obviously become slightly
more complicated. It should be mentioned that the mathematics behind the
statistical description of the gravity anomalies is the theory of stochastic
processes. The gravity anomaly field is treated as a stationary stochastic
process on a sphere; the spherical-harmonic expansions of this section are
nothing but the spectral analysis of that process. A comprehensive treatment
of this topic is found in Moritz (1980 a).
9.4 Interpolation and extrapolation of gravity

anomalies
As pointed out in Sect. 9.1, the purpose of prediction (interpolation and
extrapolation) is to supplement the gravity observations, which can be made
at only relatively few points, by estimating the values of gravity or of gravity
anomalies at all the other points P of the earth’s surface.
If P is surrounded by gravity stations, we must interpolate; if the gravity
stations are far away from P , we extrapolate. Evidently, there is no sharp
distinction between these two kinds of prediction and the mathematical for-
mulation is the same in both cases.
In order to predict a gravity anomaly at P , we must have information
about the gravity anomaly function. The values observed at certain points
are the most important information. In addition, we need some information
on the form of the anomaly function. If the gravity measurements are very
dense, then the continuity or “smoothness” of the function is sufficient – for
instance, for linear interpolation. Otherwise we may try to use statistical
information on the general structure of the gravity anomalies. Here we must
consider two kinds of statistical correlation: the autocorrelation – the corre-
lation between each other – of gravity anomalies and the correlation of the
gravity anomalies with height.
Correlation with height will for the moment be disregarded; Sect. 9.7
will be devoted to this topic. The autocorrelation is characterized by the
covariance function considered in Sect. 9.2.
Mathematically, the purpose of prediction is to find a function of the
observed gravity anomalies ∆g1 , ∆g2 , . . . , ∆gn in such a way that the un-
known anomaly ∆gP at P is approximated by the function
.
∆gP = F (∆g1 , ∆g2 , . . . , ∆gn ) . (9–40)
Here ∆gi denotes the value of ∆g at a point i, not a spherical harmonic! In

practice, only linear functions of the ∆gi are used. If we denote the predicted
value of ∆gP by ∆g - P , such a linear prediction has the form
n

- P = αP 1 ∆g1 + αP 2 ∆g2 + . . . + αP n ∆gn ≡
∆g αP i ∆gi . (9–41)
i=1
The coefficients αP i depend only on the relative position of P and the grav-
ity stations 1, 2, . . . , n; they are independent of the ∆gi . Depending on the
way we choose these coefficients, we obtain different interpolation or extrap-
olation methods. Here are some examples.
Geometric interpolation
The “gravity anomaly surface”, as represented by a gravity anomaly map,
may be approximated by a polyhedron by dividing the area into triangles
whose corners are formed by the gravity stations and passing a plane through
the three corners of each triangle (Fig. 9.3). This is approximately what is
done in constructing the contour lines of a gravity anomaly map by means
of graphical interpolation.
Analytically, this interpolation may be formulated as follows. Let point
P be situated inside a triangle with corners 1, 2, 3 (Fig. 9.3). To each point
9.4 Interpolation and extrapolation of gravity anomalies 357
2
P
1
3
Fig. 9.3. Geometric interpolation
we assign its value ∆g as its z-coordinate, so that the points 1, 2, and 3

have “spatial” coordinates (x1 , y1 , z1 ), (x2 , y2 , z2 ), and (x3 , y3 , z3 ); x and y
are ordinary plane coordinates. The plane through 1, 2, 3 has the equation
(x2 − x)(y3 − y2 ) − (y2 − y)(x3 − x2 )

z= z1
(x2 − x1 )(y3 − y2 ) − (y2 − y1 )(x3 − x2 )
(x3 − x)(y1 − y3 ) − (y3 − y)(x1 − x3 )
+ z2 (9–42)
(x3 − x2 )(y1 − y3 ) − (y3 − y2 )(x1 − x3 )
(x1 − x)(y2 − y1 ) − (y1 − y)(x2 − x1 )
+ z3 .
(x1 − x3 )(y2 − y1 ) − (y1 − y3 )(x2 − x1 )
If we replace z1 , z2 , z3 by ∆g1 , ∆g2 , ∆g3 , then z is the interpolated value

-
∆gP at point P , which has the plane coordinates x, y. Thus,
- P = αP 1 ∆g1 + αP 2 ∆g2 + αP 3 ∆g3 ,

∆g (9–43)
where the αP i are the coefficients of zi in the preceding equation.
Representation
Often the measured anomaly of a gravity station 1 is made to represent the
whole neighborhood so that
- P ≡ ∆g1
∆g (9–44)
as long as P lies within a certain neighborhood of point 1. Then
αP 1 = 1 , αP 2 = αP 3 = . . . = αP n = 0 . (9–45)
This method is rather crude but simple and accurate enough for many pur-
poses.
Zero anomaly
If there are no gravity measurements in a large area – for instance, on the
oceans –, then the estimate
-P ≡ 0
∆g (9–46)
is used in this area. In this trivial case all αP i are zero.
If all known gravity stations are far away, and if we know of nothing
better, then this primitive extrapolation method is applied, although the
accuracy is poor. At best, this method may work with isostatic anomalies.
None of these three methods gives optimum accuracy. In the next section we
investigate the accuracy of the general prediction formula (9–41) and find
those coefficients αP i that yield the most accurate results.
9.5 Accuracy of prediction methods

In order to compare the various possible methods of prediction, to determine
their range of applicability, and to find the most accurate method, we must
evaluate their accuracy.
Consider the general case of Eq. (9–41). The correct gravity anomaly at
P is ∆gP , the predicted value is
n

-P =
∆g αP i ∆gi . (9–47)
i=1
The difference is the error εP of prediction,

- P = ∆gP −
εP = ∆gP − ∆g αP i ∆gi . (9–48)
i
By squaring we find

ε2P = ∆gP − αP i ∆gi ∆gP − αP k ∆gk
i k
(9–49)
= ∆gP2 − 2 αP i ∆gP ∆gi + αP i αP k ∆gi ∆gk .
i i k
Let us now form the average M of this formula over the area considered
(either a limited region or the whole earth). Then we have from (9–6),
M {∆gi ∆gk } = C(i k) ≡ Cik ,
M {∆gP ∆gi } = C(P i) ≡ CP i , (9–50)
M {∆gP2 } = C(0) ≡ C0 .
9.5 Accuracy of prediction methods 359
These are particular values of the covariance function C(s), for s = i k,

s = P i, and s = 0; for instance, i k is the distance between the gravity
stations i and k. The abbreviated notations Cik and CP i are self-explanatory.
We further set
M {ε2P } = m2P . (9–51)
Thus mP is the root mean square error of a predicted gravity anomaly at P ,
or briefly, the standard error of prediction (interpolation or extrapolation).
Taking all these relations into account, we find the average M of (9–49)
to be
n n
n
m2P = C0 − 2 αP i CP i + αP i αP k Cik . (9–52)
i=1 i=1 k=1
This is the fundamental formula for the standard error of the general predic-
tion formula (9–41). For the special cases described in the preceding section,
the particular values of αP i are to be inserted.
Einstein’s summation convention

At least at this point the reader will be grateful to Albert Einstein for having
invented not only the theory of relativity – well, even the general theory of
relativity has been used in geodesy (Moritz and Hofmann-Wellenhof 1993),
but the reader of the present book will be saved from it – but also the very
practical summation convention which has eradicated myriads of unneces-
sary summation signs from the mathematical literature. This convention
simply says that, if an index occurs twice in a product, summation is auto-
matically implied. Using this convention, the preceding equation is simply
written
m2P = C0 − 2 αP i CP i + αP i αP k Cik . (9–53)
In the future we shall take this equation for granted unless stated otherwise.
Such formulas are also handsome for programming (a loop).
Now back to reality in the form of examples.

As an example consider the case of representation, Eq. (9–44); all α are
zero except one. Here (9–53) yields
m2P = C0 − 2CP 1 + C0 = 2C0 − 2CP 1 . (9–54)
For the case of zero anomaly, there is m2p = C0 , as should be expected.

Often we need not only the standard error mP of prediction but also the
correlation of the prediction errors εP and εQ at two different points P and
Q, expressed by the “error covariance” σP Q , which is defined by
σP Q = M {εP εQ } . (9–55)
If the errors εP and εQ are uncorrelated, then the error covariance σP Q = 0.

From (9–48) we have generally
.

/
σP Q = M ∆gP − αP i ∆gi ∆gQ − αQk ∆gk
. /
= M ∆gP ∆gQ − αP i ∆gQ ∆gi − αQk ∆gP ∆gk + αP i αP k ∆gi ∆gk
(9–56)
and finally
σP Q = CP Q − αP i CQi − αQi CP i + αP i αQk Cik . (9–57)
The notations are self-explanatory; for instance, CP Q = C(P Q).
The error covariance function

The values of the error covariance σP Q , for different positions of the points
P and Q, form a continuous function of the coordinates of P and Q. This
function is called the error covariance function, or briefly, the error function,
and is denoted by σ(xP , yP , xQ , yQ ). If P and Q are different, then we simply
have
σ(xP , yP , xQ , yQ ) = σP Q ; (9–58)
if P and Q coincide, then (9–57) reduces to (9–53) so that
σ(xP , yP , xP , yP ) = m2P (9–59)
is the square of the standard prediction error at P .
Thus the error covariances σP Q may be considered as special values of
the error covariance function, just as the covariances CP Q of the gravity
anomalies may be considered as special values of the covariance function
C(s). To repeat, the error function is the covariance function of the prediction
errors, defined as
M {εP εQ } , (9–60)
whereas C(s) is the covariance function of the gravity anomalies, defined as
M {∆gP ∆gQ } . (9–61)
The term “covariance function” in the narrower sense will be reserved for
C(s) – in contrast to least-squares adjustment, where “covariances” auto-
matically mean error covariances. Covariances are “isotropic”, which means
independent of directions; the error covariances are nonisotropic.
From (9–53) and (9–57) the error function can be expressed in terms of
the covariance function; we may write more explicitly
σ(xP , yP , xQ , yQ ) = C(P Q) − αP i C(Qi) − αQi C(P i) + αP i αQk C(i k) .
(9–62)
9.6 Least-squares prediction 361
Thus we recognize the basic role of the covariance function in accuracy stud-
ies. The error function, on the other hand, is fundamental for problems of
error propagation.
9.6 Least-squares prediction

The values of αP i for the most accurate prediction method are obtained by
minimizing the standard prediction error expressed by (9–53) as a function
of the α. The familiar necessary conditions for a minimum are
∂m2P
≡ −2CP i + 2αP k Cik = 0 (i = 1, 2, . . . , n) (9–63)
∂αP i
or
Cik αP k = CP i . (9–64)
This is a system of n linear equations in the n unknowns αP k ; the solution
is
(−1)
αP k = Cik CP i , (9–65)
(−1)
where Cik denote the elements of the inverse of the symmetric matrix
[Cik ].
Substituting (9–65) into (9–41) gives
- P = αP k ∆gk = C (−1) CP i ∆gk .

∆g (9–66)
ik
In matrix notation this is written

⎡ ⎤−1 ⎡ ⎤
C11 C12 . . . C1n ∆g1
⎢
C21 C22 . . . C2n ⎥ ⎢ ⎥
- P = CP 1 , CP 2 , . . . , CP n ⎢ ⎥ ⎢ ∆g2 ⎥
∆g ⎢ .. .. .. ⎥ ⎢ .. ⎥ . (9–67)
⎣ . . . ⎦ ⎣ . ⎦
Cn1 Cn2 . . . Cnn ∆gn
We see that for optimal prediction we must know the statistical behavior of
the gravity anomalies through the covariance function C(s).
There is a close connection between this optimal prediction method
and the method of least-squares adjustment. Although they refer to some-
what different problems, both are designed to give most accurate results.
The linear equations (9–64) correspond to the “normal equations” of ad-
justment computations. Prediction by means of formula (9–67) is therefore
called “least-squares prediction”. A generalization to heterogeneous data is
“least-squares collocation” to be treated in Chap. 10. In its most general
form, least-squares collocation also includes parameter estimation by least-

squares adjustment. This is an advanced subject treated in great detail in
Moritz (1980 a).
It is easy to determine the accuracy of least-squares prediction. Insert
the α of Eq. (9–65) into (9–53), after appropriate changes in the indices of
summation. This gives
m2P = C0 − 2αP k CP k + αP k αP l Ckl
(9–68)
(−1) (−1) (−1)
= C0 − 2Cik CP i CP k + Cik CP i Cjl CP j Ckl .
For the reader to appreciate the Einstein summation convention, we give
this equation in its original form:

m2P = C0 − 2 αP k CP k + αP k αP l Ckl
k k l
(−1)
(−1) (−1)
= C0 − 2 Cik CP i CP k + Cik CP i Cjl CP j Ckl .
i k i j k l
(9–69)
But now back to normal! We have

(−1) 1 if j = k
Cjl Ckl = δjk = (9–70)
0 k.
if j =
The matrix [δkl ] is the unit matrix. This formula states that the product of
a matrix and its inverse is the unit matrix. Thus, we further have
(−1) (−1) (−1) (−1)
Cik Cjl Ckl = Cik δjk = Cij (9–71)
because a matrix remains unchanged on multiplication by the unit matrix.
Hence, we get
(−1) (−1)
m2P = C0 − 2Cik CP i CP k + Cij CP i CP j
(−1) (−1)
= C0 − 2Cik CP i CP k + Cik CP i CP k (9–72)
(−1)
= C0 − Cik CP i CP k .
Thus, the standard error of least-squares prediction is given by
(−1)
m2P = C0 − Cik CP i CP k
⎡ ⎤−1 ⎡ ⎤
C11 C12 . . . C1n CP 1
⎢
⎢ C21 C22 . . . C2n ⎥ ⎢ ⎥
⎥ ⎢ CP 2 ⎥
= C0 − CP 1 , CP 2 , . . . , CP n ⎢ . .. .. ⎥ ⎢ .. ⎥ .
⎣ .. . . ⎦ ⎣ . ⎦
Cn1 Cn2 . . . Cnn CP n
(9–73)
9.7 Correlation with height 363
In the same way we find the error covariance in the points P and Q:
(−1)
σP Q = CP Q − Cik CP i CQk
⎡ ⎤−1 ⎡ ⎤
C11 C12 . . . C1n CQ1
⎢ C
⎢ 21 C22 . . . C2n ⎥ ⎢
⎥ ⎢ CQ2 ⎥
⎥
= CP Q − CP 1 , CP 2 , . . . , CP n ⎢ . .. .. ⎥ ⎢ .. ⎥ .
⎣ .. . . ⎦ ⎣ . ⎦
Cn1 Cn2 . . . Cnn CQn
(9–74)
These two formulas give the error covariance function for least-squares pre-
diction. Both formulas have a form similar to that of (9–67) and are equally
well suited for computations so that ∆g - and its accuracy can be calculated
at the same time.
It is clear that, after appropriate slight changes, this theory applies au-
tomatically to gravity disturbances δg.
Practical considerations
Geometric interpolation (Sect. 9.4) is suited for the interpolation of point
anomalies in a dense gravity net, with station distances of 10 km or less. If
mean anomalies for blocks of 5 × 5 or larger are needed rather than point
anomalies, then some kind of representation, such as that considered in the
previous section, may be simpler and hardly less accurate.
Least-squares prediction is, by its very definition, more accurate than
either geometric interpolation or representation, but the improvement in ac-
curacy is not striking. The main advantage of least-squares prediction is
that it permits a systematic, purely numerical, digital processing of gravity
data; gravity anomalies are stored in data bases, and gravity anomaly maps,
if necessary, are generated automatically. The same formula applies to both
interpolation and extrapolation so that gaps in the gravity data make no dif-
ference in the method of computation, which becomes completely schematic
(Moritz 1963). For practical and computational details see Rapp (1964) and
many other papers published since.
For larger station distances, of 50 km or more, prediction of individual
point values becomes meaningless. In this case we must work with mean
anomalies of, say, 1◦ × 1◦ blocks.
9.7 Correlation with height

So far we have taken into account only the mutual correlation of the gravity
anomalies, their autocorrelation, disregarding the correlation with height,
h [m] h [m]
' = 46° N ' = 45° N

2000 2000
¸ = 09° E ¸ = 03° E
1500 1500
1000 1000
500 500
®
Dg [mgal] Dg [mgal]
0 0
–150 –100 –50 0 +50 –50 0 +50 +100 +150
Fig. 9.4. Correlation of the free-air anomalies with height
which is important in many cases. Therefore our formulas were valid only
for gravity anomalies uncorrelated with height, such as isostatic or, to a
certain extent, Bouguer anomalies; or for free-air anomalies in moderately
flat areas. Free-air anomalies in mountains must be treated differently.
Figure 9.4 due to U.A. Uotila shows the correlation of free-air anomalies
with height. The gravity anomalies ∆g are plotted against the height h. If
there were an exact functional dependence between ∆g and h, then all points
would lie on a straight line (or, more generally, on a curve). In reality, there
is only an approximate functional relation, a general trend or tendency of
the free-air anomalies to increase linearly with height; exceptions, even large
ones, are possible. This shows very well the meaning of correlation.
We have characterized the mutual correlation of the gravity anomalies
by the “autocovariance function” (9–6),
C(s) = M {∆g ∆g } , (9–75)
where s = P P . Similarly, we may form the “cross-covariance function”
B(s) = M {∆g ∆h } = M {∆g ∆h} , (9–76)
expressing the correlation between gravity and height, and
A(s) = M {∆h ∆h } , (9–77)
which is the autocovariance function of the height differences
∆h = h − M {h} , (9–78)
where the symbol M {h} denotes the mean height of the whole area consid-
ered.
If ∆g and ∆h are not correlated, then the function B(s) is identically
zero. If this is not the case, then we should also take the height into account
in our interpolation.
It is easy to extend the prediction formula (9–41) for this purpose, but
this has turned out to be of little practical importance.
Application to Bouguer anomalies

Of great practical importance, however, is the question whether it is possible
to render the free-air anomalies independent of height by adding a term that
is proportional to the height. In other words, when is the quantity
z = ∆g − b ∆h , (9–79)
with a certain coefficient b, uncorrelated with height? In statistical termi-

nology, correlation with height is a trend, which may be capable of being
removed.
The trend z has the form of a Bouguer anomaly; for a real Bouguer
anomaly we have, according to Sect. 3.4,
b = 2π G . (9–80)
For the density = 2.67 g/cm3 we get
b = +0.112 mgal/m . (9–81)
Let us form the covariance function Z(s) between the “Bouguer anomaly” z
of (9–79) and height difference ∆h
Z(s) ≡ M {z ∆h } = M {∆g ∆h − b ∆h ∆h } = B(s) − b A(s) . (9–82)
If z is to be uncorrelated with h, then Z(s) must be identically zero. The

condition is
B(s) − b A(s) ≡ 0 , (9–83)
which must be satisfied for all s and a certain constant b at least approxi-
mately.
We see that the “Bouguer anomaly” z is uncorrelated with height if
the functions A(s) and B(s) are proportional for the area considered; the
constant b is then represented by
B(s)
b= . (9–84)
A(s)
elevation ®
Bouguer anomaly ®
% = 2.2 g/cm3
% = 2.4 g/cm3
% = 2.6 g/cm3
distance along profile ®
Fig. 9.5. Bouguer anomalies corresponding to different densities : the

best density is = 2.4 g/cm3 (no correlation); for other densities the
Bouguer anomalies are correlated with height (positive correlation for
3 3
= 2.2 g/cm , negative correlation for = 2.6 g/cm )
It may be shown that this is equivalent to the condition that the points of
Fig. 9.4 lie approximately on a straight line. The coefficient b is then given
by
b = tan α (9–85)
as the inclination of the line towards the h-axis.
In practice these conditions are very often fulfilled to a good approx-
imation. Furthermore, by computing b from Eq. (9–84) or determining it
graphically by means of (9–85), we often get a value that is close to the
normal Bouguer gradient (9–81).
If we assume that b depends only on the rock density , then we obtain a
means for determining the average density, which is often difficult to measure
directly. This is the “Nettleton method”, used in geophysical prospecting: the
coefficient b is found statistically by means of Eqs. (9–84) or (9–85), and
the rock density is then computed from (9–80). Figure 9.5 illustrates the
principle of this method; see also Jung (1956: p. 600).
If the condition (9–83) is fulfilled, then we may consider the “Bouguer
anomaly” z as a gravity anomaly that is completely uncorrelated with height;
we can directly apply to it the whole theory of the preceding sections. But
even when this condition is not quite satisfied, Bouguer anomalies will in
general be far less correlated with height than free-air anomalies. The fact
that in (9–79) gravity is reduced to a mean height and not to sea level,
is quite irrelevant in this connection because this is only a question of an
additive constant. More recent developments are discussed by Moritz (1990:

p. 244).
It is thus possible to consider the Bouguer reduction as a means of ob-
taining gravity anomalies that are less dependent on height and hence more
representative than free-air anomalies. More precisely, the Bouguer anoma-
lies take care of the dependence on the local irregularities of height. The
isostatic anomalies are, in addition, also largely independent of the regional
features of topography. See also Chaps. 3 and 8.
10 Least-squares collocation
10.1 Principles of least-squares collocation

The principle of collocation is very simple. The anomalous potential T out-
side the earth is a harmonic function, that is, it satisfies Laplace’s differential
equation
∂2T ∂2T ∂2T
∆T = + + = 0. (10–1)
∂x2 ∂y 2 ∂z 2
An approximate analytical representation of the external potential T is ob-
tained by
q
.
T (P ) = f (P ) = bk ϕk (P ) , (10–2)
k=1
a linear combination f of suitable base functions ϕ1 , ϕ2 , . . . , ϕq with appro-
priate coefficients bk . All these are functions of the space point P under
consideration.
As T is harmonic outside the earth’s surface, it is natural to choose base
functions ϕk which are likewise harmonic, so that
∆ϕk = 0 , (10–3)
in correspondence to (10–1).
There are many simple systems of functions satisfying the harmonicity
condition (10–3), and thus we have many possibilities for a suitable choice
of base functions ϕk . We might, for instance, choose spherical harmonics or
potentials of suitably distributed point masses, depending on whether we
emphasize global or local applications.
The coefficients bk may be chosen such that the given observational values
are reproduced exactly – for instance, all deflections of the vertical in a given
area. This means that the assumed approximating function f in (10–2) gives
the same deflections of the vertical at the observation stations as the actual
potential and hence may well be considered a suitable approximation for T .
Let us now try to put these ideas into a mathematical form.
Interpolation
Let errorless values of T be given at q spatial points P1 , P2 , . . . , Pq ; these
points may lie on the earth’s surface or in space above the earth’s surface.
We put
T (Pi ) = fi , i = 1, 2, . . . , q (10–4)
370 10 Least-squares collocation
and postulate that in approximating T (P ) by f (P ), the observations (10–4)

will be reproduced exactly. The condition for this is
q

bk ϕk (Pi ) = T (Pi ) = fi , (10–5)
k=1
which may be written as a system of linear equations

q

Aik bk = fi with Aik = ϕk (Pi ) (10–6)
k=1
or in matrix notation
Ab = f . (10–7)
If the square matrix A is regular, then the coefficients bk are uniquely de-
termined by
b = A−1 f . (10–8)
This model is suitable, for instance, for a determination of the geoid by
satellite altimetry, since this method, rather directly, yields geoidal heights
Ni and hence, by Bruns’ theorem (2–236), T (Pi ) = γi Ni . For the astro-
geodetic geoid determination, we must generalize this model, which leads us
to collocation.
Collocation
Here we wish to reproduce, by means of the approximation (10–2), q mea-
sured values which again are assumed to be errorless (this assumption is
not essential and will be dropped later). These measured values are assumed
to be linear functionals L1 T, L2 T, . . . , Lq T of the anomalous potential T .
“Linear functional” means nothing else than a quantity LT that depends
linearly on T but need not be an ordinary function but may, say, also con-
tain a differentiation or an integral; essentially, it is the same as a “linear
operator”.
In fact, deflections of the vertical,
1 ∂T 1 ∂T
ξ=− , η=− , (10–9)
γ ∂x γ ∂y
but also gravity anomalies,
∂T 2
∆g = − − T, (10–10)
∂z R
and gravity disturbances
∂T
δg = − (10–11)
∂z
10.1 Principles of least-squares collocation 371
are such linear functionals of T ; here, x, y, z denotes a local coordinate sys-

tem in which the z-axis is vertical upwards and the x- and y-axes are directed
towards north and east, and R = 6371 km is a mean radius of the earth.
Equation (10–9) is a consequence of equations such as (2–377), with ∂s = ∂x
or ∂y; normal gravity γ may be considered constant with respect to horizon-
tal derivation. Equation (10–10) is the well-known fundamental equation of
physical geodesy in spherical approximation (2–263). Equations (10–9) and
(10–10) refer to the earth’s surface.
To repeat, by saying that deflections of the vertical and gravity distur-
bances and anomalies are linear functionals of T , we simply indicate the fact
that ξ, η, δg, ∆g depend on T by the expressions (10–9) and (10–10), which
clearly are linear; they are the linear terms of a Taylor expansion, neglect-
ing quadratic and higher terms. In the above notation Li T , the symbol Li
denotes, for instance, the operation
1 ∂
Li = (10–12)
γ ∂x
applied to T at some point.

Putting
Li f = Li T = i (10–13)
and substituting (10–2), we get
q

Bik bk = i with Bik = Li ϕk , (10–14)
i=1
where Li ϕk denotes the number obtained by applying the operation Li to the

base function ϕk ; the coefficient Bik obtained in this way does not depend
on the measured values. Equation (10–14) is a linear system of q equations
for q unknowns, which is quite similar to (10–6). This method of fitting an
analytical approximating function to a number of given linear functionals is
called collocation and is frequently used in numerical mathematics.
It is clear that interpolation is a simple special case of collocation in
which
Li f = f (Pi ) (10–15)
is the “evaluation functional”, giving the value of f at a point Pi . Thus

we see that in both interpolation and collocation the coefficients bk require
the solution of a linear system of equations (which in general will not be
symmetric).
Least-squares interpolation
Let us consider a function
K = K(P, Q) , (10–16)
in which two points P and Q are the independent variables. Let this function
K be
• symmetric with respect to P and Q,
• harmonic with respect to both points, everywhere outside a certain
sphere, and
• positive-definite (the positive definitiveness of a function is defined
similarly as in the case of a matrix).
Then the function K(P, Q) is called a (harmonic) kernel function (Moritz
1980 a: p. 205). A kernel function K(P, Q) may serve as “building material”
from which we can construct base functions. Taking for the base functions
the form
ϕk (P ) = K(P, Pk ) , (10–17)
where P denotes the variable point and Pk is a fixed point in space, we obtain
least-squares interpolation already treated by a quite different approach in
Chap. 9.
This name originates from the statistical interpretation of the kernel
function as a covariance function (Sect. 9.2); then least-squares interpola-
tion has some minimum properties (least-error variance, similarly as in least-
squares adjustment). This interpretation is not essential, however; one may
also work with arbitrary analytical kernel functions, considering the proce-
dure as a purely analytical mathematical approximation technique. Normally
one tries to combine both aspects in a reasonable way.
Substituting (10–17) into (10–6), we get
Aik = K(Pi , Pk ) = Cik ; (10–18)
this square matrix now is symmetric (in the general case, Aik is not sym-
metric!) and positive definite because of the corresponding properties of the
function K(P, Q). Then the coefficients bk follow from (10–8) and may be
substituted into (10–2). With the notation
ϕk (P ) = K(P, Pk ) = CP k , (10–19)
the result may be written in the form
⎡ ⎤−1 ⎡ ⎤
C11 C12 . . . C1q f1
⎢ C
⎢ 21 C 22 . . . C ⎥
2q ⎥
⎢ f
⎢ 2⎥
⎥
f (P ) = CP 1 CP 2 . . . CP q ⎢ .
.. ⎦ ⎣ .. ⎥
⎥ ⎢
. . . , (10–20)
⎣ .. .. ⎦
Cq1 Cq2 . . . Cqq fq
10.1 Principles of least-squares collocation 373
formally identical with Eq. (9–67) obtained in a completely different way.
Least-squares collocation
Here we again derive the base functions from a kernel function K(P, Q), but
in a way slightly different from (10–17): we put
ϕk (P ) = LQ
k K(P, Q) , (10–21)
where LQ k means that the functional Lk is applied to the variable Q; the

result no longer depends on Q (since the application of a functional results
in a definite number). Thus, in (10–14) we must put
Bik = LPi LQ
k K(P, Q) = Cik , (10–22)
which gives a matrix which again is symmetric. Solving (10–14) for bk and
substituting into (10–2) gives with
ϕk (P ) = LQ
k K(P, Q) = CP k (10–23)
the formula
⎡ ⎤−1 ⎡ ⎤
C11 C12 . . . C1q 1
⎢
⎢ C21 C 22 . . . C 2q
⎥ ⎢ 2 ⎥
⎥ ⎢ ⎥
f (P ) = CP 1 CP 2 . . . CP q ⎢ . .. .. ⎥ ⎢ .. ⎥ . (10–24)
⎣ .. . . ⎦ ⎣ . ⎦
Cq1 Cq2 . . . Cqq q
This is formally the same expression as (10–20), but with fi replaced by

i and with “covariances” Cik and CP i defined by “covariance propagation”
(10–22) and (10–23). The concept of covariance propagation is a straight-
forward generalization of the formal structure of error propagation known
from adjustment computations. However, this structure as such is purely
mathematical rather than statistical. We know that a “linear functional” is
the continuous analogue (in infinite-dimensional Hilbert space) to the usual
concept of a linear function in n-dimensional vector space. We try not to bur-
den the reader with too much mathematical formalism, but this is treated in
great detail in Moritz (1980 a) and in Moritz and Hofmann-Wellenhof (1993:
Chap. 10). We cannot, however, resist the temptation to compare the struc-
ture
bi = Lji aj (10–25)
leading to
cov(bi , bj ) = Lki Llj cov(ak , al ) (10–26)
for finite-dimensional vectors a and matrix L using the usual summation

over two equal indices, and Ni = LPi ∆gP leading to
cov(NP , NQ ) = LPi LQ
j cov(∆gP , ∆gQ ) , (10–27)
where Ni denotes the geoidal height at point i and ∆g is the gravity anomaly
at point P , and L denotes the Stokes formula. Explicit expressions are found
in Moritz (1980 a: Sect. 15).
In this statistical interpretation, we take the kernel function K(P, Q)
as the covariance function C(P, Q). Then f (P ) is an optimal estimate (in
the sense of least variance) for the anomalous potential T and hence for
the height anomaly ζ = T /γ, on the basis of arbitrary measurement data.
For geoid determination in mountainous areas, relevant terrestrial measure-
ment data primarily are ξ, η, and ∆g. The covariances Cik and CP i are
given by known analytical expressions, see Tscherning and Rapp (1974) or
Moritz (1980 a: Sect. 15). A general computer program for collocation is
described in Sünkel (1980).
Least-squares collocation may easily be generalized to observational data
affected by random errors; systematic effects may also be taken into consid-
eration. In addition to the estimated quantities (f in our present case) we
may also compute their standard error by a formula similar to (10–24). A
comprehensive presentation of a least-squares collocation may be found in
Moritz (1980 a). You cannot learn collocation from this slight chapter only!
Harmonicity of the covariance functions.

In three-dimensional space, the covariance functions, being kernel functions
and their linear functional transformations L, are always harmonic. If we
have (9–25),
∞
C(ψ) = cn Pn (cos ψ) (10–28)
n=2
on the sphere, then in space there will be

∞
n+2
R2
C(r, r , ψ) = cn Pn (cos ψ) (10–29)
r r
n=2
(Moritz 1980 a: Sect. 23, Eq. (32-1)). The point P (r, θ, λ) is the computation
point, and Q(r , θ , λ ) is a current data point; ψ is the spherical distance
between (θ, λ) and (θ , λ ), and R is the mean radius of the earth. The de-
pendence on r is given by the factor
r −(n+2) (10–30)
10.2 Application of collocation to geoid determination 375
because r∆g is harmonic, and similarly for r . The factor

2
R
(10–31)
r r
is chosen to become equal to 1 if both points P and Q lie on sea level; in

this case, Eq. (10–29) reduces to (10–28).
So, each of the terms of (10–29) is harmonic, that is, it satisfies Laplace’s
equation. Thus, the whole series (10–29) is harmonic (if it converges), being
a linear combination of harmonic terms. This is a well-known consequence
of the linearity of Laplace’s equation: the linear combination of solutions of
any linear equation is itself a solution of this equation.
Thus, also the spherical harmonics series of T = r∆g is harmonic down
to the reference sphere r = R, with respect both to r and r . Harmonic
functions, by their very definition, are regular analytic functions down to
r = R, so T and all its linear combinations are regular and thus admit
downward continuation down to the reference sphere (cf. Sect. 8.6).
10.2 Application of collocation to geoid

determination
It is well known that the direct interpolation of free-air gravity anomalies,
which essentially are surface gravity anomalies (8–128) in high mountains,
e.g., by least-squares interpolation, leads to relatively poor results because
of the correlation of the free-air anomalies with elevation (Sect. 9.7). This
correlation with elevation constitutes a considerable trend which must be
removed before the interpolation. Bouguer anomalies take care of the de-
pendence on the local irregularities of elevation; isostatic anomalies are, in
addition, also largely independent on the regional features of topography; in
Sect. 11.1 we shall consider, in addition, also the removal of global trends
by spherical-harmonic earth gravity models (e.g., EGM 96, see www.iges
.polimi.it/index/geoid repo/global models.htm) obtainable from the inter-
net.
In exactly the same way we must remove the main trend of the vertical
deflections ξ, η and the gravity anomalies ∆g by an isostatic reduction be-
fore applying collocation. Thus, isostatic reduction, pragmatically regarded
as trend removal, is essential for the practical application of least-squares
collocation in mountainous regions (Forsberg and Tscherning 1981).
Physically speaking, we transport the topographic masses to the interior
of the geoid in such a way that the isostatic mass deficiencies are filled. The
observation point P remains in its position on the earth’s surface. In this way,
not only the harmonic character of the anomalous potential T outside the
earth’s surface is preserved, but in addition, the computational removal of
the topographic masses above sea level makes the function T harmonic down
to sea level. Hence, the collocation formula (10–24) can be applied also at
sea level, giving cogeoid heights N c . By applying the inverse reduction (the
indirect effect) to the computed height anomalies ζ c and cogeoid heights N c ,
we get actual ζ and N . It can be expected that errors in the isostatic model
used (e.g., an Airy–Heiskanen model) will largely cancel in this combined
procedure of reduction and “anti-reduction” (remove-restore technique; see
Sect. 11.1).
The procedure is theoretically optimal and practically well suited for
computer use. The integrability conditions, which in Helmert integration
are represented by the closures of the individual triangles (see Sect. 5.14),
are automatically taken into account. The fact that the deflections of the
vertical are given only in a certain region has the effect that the geoid can
only be computed in that region. Since, even by collocation, differences in
geoidal heights between two neighboring stations A and B depend essentially
only on the deflections in those two stations, the lack of data outside the
region under consideration will hardly cause a noticeable distortion. Note,
however, that the addition of a constant to all geoidal heights N will not
affect the deflections of the vertical; hence, astrogeodetic data determine
the geoidal heights only up to an additive constant. This constant may be
chosen such that the average value of the computed N is zero, and the result
of collocation comes near to this case.
To get immediately almost geocentric geoidal heights, it is appropriate to
take into consideration a global trend which mainly affects ζ and N itself, by
subtracting the effect of a suitable global gravity field, e.g., the gravity earth
model given as a spherical-harmonic expansion up to degree 180◦ × 180◦ of
Rapp (1981), say, following Sünkel (1983). This will be described in the next
section; in the present section we limit ourselves to the isostatic reduction.
Computational procedure
The computational procedure consists of the following steps:
1. Transformation of the astrogeodetic surface deflections ξ, η from the
local datum used for the geocentric Geodetic Reference System 1980
by the well-known differential formulas of Vening Meinesz (see Heiska-
nen and Moritz 1967: Eq. (5-59)). This is necessary since collocation
requires a reference system which is as realistic as possible.
2. Application of the normal plumb line curvature (8–137) to the “geo-

metric” surface deflections ξ, η gives the “dynamic” surface deflections
10.2 Application of collocation to geoid determination 377
ξ̄, η̄ by (8–136).
3. Computation of the gravity anomalies ∆g, also referred to the earth’s

surface according to (8–128).
¯ η̄, ∆g by (8–154) and (8–101)
4. The topographic-isostatic reduction of ξ,
c c c
gives values ξ , η , ∆g which continue to refer to the surface point P .
5. The application of collocation to ξ c , η c , ∆g gives height anomalies ζ c

and cogeoid heights N c , by simply varying the elevation parameter (h
and zero, respectively) in the collocation program (see Sünkel 1983).
6. By applying the indirect effect (10–2) and (8–153), we get actual height
anomalies ζ and geoidal heights N .
11 Computational methods
11.1 The remove-restore principle

Let us start with gravity reduction according to the modern view of mea-
suring and calculating the gravity field in principle always at the earth’s
surface, or briefly, on the ground, or equivalently, at point level. This is used
in the sense of Sects. 8.9 and 8.14. More precisely, it is topographic-isostatic
reduction at ground level.
The most practical way to realize this idea is least-squares collocation,
because it automatically works in three-dimensional space, by simply putting
the desired topographic height h as parameters for input (measurements:
gravity anomalies, deflections of the vertical, etc.) and output (potential T
or its functionals to be computed). Symbolically, this means
T = L() (11–1)
or
output = L(input) , (11–2)
where L denotes the linear operation of least-squares collocation (not to be
confused with a linear functional L as used, e.g., in Eq. (10–13)).
In Sect. 8.9 we have introduced gravity reduction from the point of view
of the modern theory. To repeat, immediately specializing to topographic-
isostatic reduction, we have
• measured gravity anomalies ∆g at ground level,

• reduced topographic-isostatic anomalies ∆gc obtained by removing the
attraction of the topographic-isostatic masses δgTI ,
• “co-potential” T c = L(∆gc ) computed by collocation, and
• “real potential” T by restoring the “indirect effect” of the topographic-
isostatic masses δTTI .
Mathematically this may be written
T = L(∆g − δgTI ) + δTTI . (11–3)
This is a reinterpretation of the gravity reduction of Sect. 8.9. It must

be correct since if
δTTI = L(δgTI ) (11–4)
380 11 Computational methods
then Eq. (11–3) gives

δT = L(∆g), (11–5)
as it should be.
The same principle works also with deflections of the vertical ξ, η at the
earth’s surface, both as input data and as output results (Sects. 8.14 and
10.2).
The underlying isostatic model is in principle arbitrary. For practical
purposes it should provide a good approximation (small residuals δT ) and
be computationally convenient.
We see, however, a change of perspective. Collocation is no longer applied
to the “real” anomalous gravity field as in (11–5) but to the residual field, re-
moving the field generated by the assumed topographic-isostatic model. The
model is arbitrary, but the derived quantities must be computed in a rigorous
consistent fashion. (Consistency for the quantities computed by collocation
as guaranteed by a correct covariance propagation; see Sect. 10.2.)
This change of perspective may not seem important because it is just
a change of nomenclature: what formerly was importantly called “isostatic
anomaly” is now degraded to a miserable “residual”. However, the remove-
restore principle permits also the use of other approximate fields to remove
trends; especially one of the numerous existing “earth (gravity) models” (EM
or EGM) consisting of spherical-harmonic expansions of the potential T up
to degree 180 or higher.
Therefore, we “remove” from the observations – gravity anomalies,
gravity disturbances, deflections of the vertical, etc. – the effect EM com-
puted from the earth model used, and after collocation “restore” the effect of
the EM on the result. The mathematics is the same as in (11–4) and (11–5):
δTEM = L(δEM ) (11–6)
and
δT = L(). (11–7)
We have only slightly generalized from ∆g to .
Now we proceed an important step further. The remove-restore principle
has only two requirements:
1. the removed auxiliary potentials must be harmonic, precomputable,

and used in a mathematically consistent way: what is removed in the
input must be restored in the output;
2. in the usual case of linearity, two or more different auxiliary potentials

may be used (removed-restored) simultaneously.
11.2 Geoid in Austria by collocation 381
Thus, we use simultaneously the earth model EM for the longer wave-
lengths and the topographic-isostatic geological model TI for the shorter
wavelengths. Since the spherical-harmonic expansions are generalizations,
for the sphere, of Fourier series for the circle, we can speak of wavelengths.
Denoting the maximum degree of the spherical-harmonic expansion with N ,
this can be associated with a shortest resolvable wavelength λ according to
2π 360◦
λ= = . (11–8)
N N
For an expansion to degree N = 180 (say), we have λ = 360◦ /180 = 2◦ ,
which roughly corresponds to 200 km on a meridian or on the equator. In
many cases, the half wavelength λ/2 is considered (see Seeber 2003: p. 469).
Since EM (approximately) takes care of the long waves up to a certain
maximum degree N , it is resonable to represent the remaining short waves
from N to infinity. This sequence N + 1, N + 2, . . . , ∞ will be denoted by
CN , where CN is the abbreviation of the “complement” of the sequence
from 2 to N .
Thus, we may write for the residuals
N − T CN ,
δT = T − TEM TI
(11–9)
δ = − N CN
EM − TI .
The collocation procedure will be applied to these residuals.
Remark
As we have noted at the beginning of Sect. 10.2, the remove-restore process
aims at removing all known major trends:
• the local topography produces Bouguer anomalies,
• the regional features (i.e., their isostatic compensation), in addition to
the Bouguer effect, produce topographic-isostatic anomalies,
• the global irregularities are expressed by an earth model and lead to
what is modestly called the “residual anomalies”.
It is clear that what is “removed” before the computation, must be fully
“restored” after the computation.
11.2 Geoid in Austria by collocation

Austria is a nice country, and in spite of being small, it has all types of
topography: flat, hilly, and alpine, up to an elevation of 3800 m. Thus, beyond
being a pleasant place to live, it is an interesting geodetic test area.
The pioneering work has been done by Sünkel (1983). Later work, espe-
cially by Sünkel et al. (1987), Kühtreiber (1998, 2002 a, 2002 b), and Erker
et al. (2003) has refined, extended and perfected the gravity field in Austria,
but the 1983 work is good for an introduction.
Sünkel (1983) used least-squares collocation to calculate the geoid for the
main part of Austria from a very good material of deflections of the vertical.
Gravity anomalies of a comparable quality were not yet available in 1983. In
addition to an isostatic reduction (Sect. 8.14) according to Airy–Heiskanen
(T = 30 km), he also removed a global trend by means of an earth gravity
model, represented by a spherical-harmonic expansion up to a certain degree
N . In particular, he used the model of Rapp (1981) with N = 180.
After removing the topographic-isostatic trend TTI and this global trend
N (remember, EM denotes earth model), there remains a residual anoma-
TEM
lous potential δT , given by
N N
δT = T − TTI − TEM + TTI . (11–10)
Since the earth model potential TEMN is represented by a spherical-harmonic
expansion up to degree N , it may be appropriate to consider, for the isostatic

reduction, only the effect for degrees N > 180 (or, say, N > 360), replacing
TTI by
CN N
TTI = (TTI )N > 180 = TTI − TTI , (11–11)
N represents a spherical-harmonic expansion for T
where TTI TI truncated at
degree N = 180. This explains Eq. (11–11).
The observations i = [ξ, η, ∆g], which represent linear functionals Li T ,
are reduced in the same way, obtaining
N N
i − Li TTI − Li TEM + Li TTI = Li δT . (11–12)
Adding the earth model reduction to the computational procedure outlined

at the end of the preceding section, we thus have the flow diagram of Ta-
ble 11.1.
Data
The topography in Austria is rather varied, with elevations up to 3800 m.
The density of astrogeodetic stations was 10 to 20 km; the total number of
deflections data used was 521. No gravity anomalies were used in this first
computation.
The topographic-isostatic reduction of the deflections of the vertical was
made using a rather crude digital terrain model consisting of mean elevations
for 20 × 20 rectangles. It has been obtained by digitizing a map 1 : 500 000.
11.2 Geoid in Austria by collocation 383
Table 11.1. From observations to the geoid

observations referred to
(Li T )
Geodetic Reference System 1980
↓
reduction N − TN)
−Li (TTI + TEM TI
TI, EM
↓
(Li δT )
↓
collocation
↓
δT
↓
inverse reduction N − TN)
+ (TTI + TEM TI
TI, EM
↓
T
↓
N = (T /γ)0 , ζ = (T /γ)h
The standard error of this model is on the order of 100 m. Investigations have
shown that, in spite of its poor accuracy, the model is reasonably adequate for
reduction of deflections of the vertical; it is, however, totally inadequate for
gravity! In fact, the reduction error for ξ, η is approximately proportional
to the terrain inclination; it is thus very small if the deflection station is
situated in an area of inclination zero. This is the case not only if the station
lies in a horizontal plane but also if it lies on the top of a mountain, as most
deflection stations do.
Results
It turned out that almost all of the signal (T, N, ζ) comes from the topo-
graphic-isostatic model and the N = 180 gravity model used. This part,
TI + EM, lies between 41.5 m and 49.5 m. The contribution of collocation
(γ −1 T ) lies between −0.5 m and 1.5 m, after removal of a pronounced trend
on the order of 3 m.
The efficiency of topographic-isostatic reduction can also be seen from
the fact that it has reduced the variance of the deflections of the vertical in
Austria (the square of the average size of ξ and η) from 30 (arc second)2 to
5 (arc second)2 .
So we may say that we can determine the Austrian geoid to 1–2 m without
measurements (deflections of the vertical) and without collocation, knowing
only a topographic map! This is even more surprising since Austria is not
particularly well isostatically compensated.
Of considerable interest is the effect of analytical continuation on the
isostatically (plus earth model) reduced anomalous potential TTI . It is ex-
pressed by the difference γ −1 T at the earth’s surface minus γ −1 T at sea
level. This difference reaches a maximum of 13 cm in the Central Alps and is
otherwise positive and negative. In the terminology of the present book, this
is the separation between the real geoid and the harmonic geoid (Sect. 8.15).
Of the same interest is the difference between the height anomalies ζ
(= γ −1 T at the earth’s surface) and the geoidal heights N (= γ −1 T at sea
level). The maximum of 35 cm for ζ − N is reached at the Grossglockner
mountain (the highest peak in Austria, H = 3797 m). The results are in
excellent agreement with the approximate formula
ζ − N = −(981 gal)−1 ∆gB H , (11–13)
where ∆gB is the Bouguer anomaly in gal and H is the elevation in the
same units as ζ and N . The agreement may easily be verified, since the
Bouguer anomalies in the investigated area range from 10 mgal to −170 mgal,
corresponding to topographic heights from 200 m to 3000 m (Sünkel 1983:
p. 140). In Sünkel et al. (1987: p. 69), the differences ζ − N for the whole of
Austria range between −2 cm and +56 cm.
All this has been computed only from the measured deflections of the
vertical. Gravity observations have been included by Kühtreiber (2002 a,
2002 b) and Erker et al. (2003), leading to what might be a “few-centimeter
geoid”.
Important: the astrogeodetic geoid and the gravimetric geoid are com-
pared and finally combined after systematic trends have been eliminated by
Kühtreiber (2002 b) and Erker et al. (2003).
11.3 Molodensky corrections

In Sect. 8.6 we have given a solutions of Molodensky’s problem by means of
a series obtained on the basis of analytical continuation. It can be written
in the form of Eqs. (8–68), (8–69), (8–67),
ζ = ζ0 + ζ1 + ζ2 + ζ3 + · · · , (11–14)
11.3 Molodensky corrections 385

R
ζi = gi S(ψ) dσ , (11–15)
4π γ
σ
∗
∆g = ∆g + g1 + g2 + g3 + · · · . (11–16)
The correction terms gn are evaluated recursively by
n

gn = − z r Lr (gn−r ) , (11–17)
r=1
starting from
g0 = ∆g . (11–18)
Here the operator Ln is also defined recursively:
Ln (∆g) = n−1 L1 [Ln−1 (∆g)] (11–19)
starting with
L1 = L (11–20)
with the gradient operator L defined by the integral (8–60), that is,

R2 f − fQ
L(f ) = dσ . (11–21)
2π l03
σ
This means: take g0 = ∆g , where ∆g is the free-air anomaly at ground level

in the sense of Molodensky, then compute g1 by (11–17) with n = 1, then
compute g2 by (11–17) with n = 2 and L2 by (11–19), then g3 by (11–17)
with n = 3 and L3 by (11–19), etc.
The operator L behaves like differentiation (L(f ) = ∂∆g ∂r ) and, there-
fore, “roughens” the function f ; this means that each successive L becomes
rougher and rougher. This is not conducive to the convergence of Molodens-
ky’s series unless the original ∆g is very smooth, which cannot be assumed
in mountainous areas.
In such cases, some smoothing of ∆g is inevitable. Numerical analysis
is constantly confronted with problems of smoothing, so many techniques
of smoothing have been developed such as the sliding average. For evalu-
ating the integral L, fast Fourier methods are available. The problem is to
find an appropriate degree of smoothing which makes consecutive correc-
tions g1 , g2 , g3 , . . . decrease in order to achieve practical convergence with-
out “oversmoothing”. At any rate, smoothing must ensure that g5 , g6 , . . .
are practically negligible since they cannot be meaningfully computed be-
cause of the inevitable accumulation of round-off errors, which finally tends
to producing pure noise.
Table 11.2. Characteristic values in arc seconds for Molodensky cor-

rections ξi and ηi for deflections of the vertical until i = 4, computed
from free-air gravity anomalies
ξ1 η1 ξ2 η2 ξ3 η3 ξ4 η4
min −2.44 −1.94 −0.92 −0.84 −0.35 −0.24 −0.08 −0.12
max 2.36 3.654 0.88 0.86 0.21 0.20 0.05 0.09
mean 0.19 0.32 −0.02 −0.02 −0.01 −0.01 0.00 0.00
rms 0.90 0.96 0.29 0.27 0.06 0.06 0.02 0.02
Table 11.3. Characteristic values in arc seconds for Molodensky cor-

rections ξi and ηi for deflections of the vertical until i = 4, computed
from isostatic gravity anomalies
ξ1 η1 ξ2 η2 ξ3 η3 ξ4 η4
min −0.57 −0.36 −0.06 −0.07 −0.01 −0.02 0.00 0.00
max 0.33 0.46 0.09 0.05 0.01 0.01 0.00 0.00
mean −0.04 0.01 0.00 −0.01 0.00 0.00 0.00 0.00
rms 0.11 0.09 0.02 0.02 0.00 0.00 0.00 0.00
As Kühtreiber (1990) showed in his thorough work, there is no rough-

and-ready prescription for finding an optimal smoothing. Trial and error may
be the best approach.
Isostatic reduction might be considered a smoothing method on a geo-
physical basis, cf. Tables 11.2 and 11.3.
Just to give an idea of the order of magnitudes, we take some typical
sizes of the Molodensky corrections in high mountains.
We take two tables from Kühtreiber (1990): the following Tables 11.2 and
11.3 are Kühtreiber’s Tables (8-3) and (8-6). The gravity data are assumed to
be given in a rectangular grid of size 11.25 × 18.75 . A suitable smoothing
is presupposed. Much better is, of course, the use of isostatic reduction,
which should provide a physically meaningful and efficient smoothing. This
is shown by Table 11.3.
To provide some contrast and to include also Molodensky corrections for
the height anomaly ζ, we quote also Table 11.4 of a somewhat earlier work
by Kraiger et al. (1987) (denoted as Table 6.1 there). The values are not
directly comparable because test areas and selected methods of integration,
smoothing, data density, etc., are different. Still, they lead to interesting
11.3 Molodensky corrections 387
Table 11.4. Comparison of direct numerical integration and fast

Fourier transform (FFT): maximum and (arithmetic) mean val-
ues of Molodensky corrections ζi , ξi , ηi for i = 1, 2; test area:
46.788◦ ≤ ϕ ≤ 46.512◦ , 13.438◦ ≤ λ ≤ 14.646◦ , 600 m ≤ topographic
height ≤ 2400 m
ζ1 [cm] ξ1 [ ] η1 [ ] ζ2 [cm] ξ2 [ ] η2 [ ]
maximum 40.8 2.0 2.0 0.8 0.2 0.2 direct int.
values 47.6 1.5 1.4 0.7 0.1 0.1 FFT
mean 31.3 0.4 0.4 0.2 0.03 0.03 direct int.
values 36.7 0.3 0.4 0.5 0.02 0.02 FFT
conclusions:
1. The method of Molodensky corrections depends very much on the de-
tails of numerical integration (data density, smoothing, etc.).
2. The corrections decrease for increasing i = 1, 2, 3, . . . . This is what

they have to do. Higher corrections may be expected finally to consist
of “pure noise” because of general roughening and increasing round-off
errors, so that the question of convergence becomes practically as well
as theoretically meaningless: higher terms must simply be put equal to
zero by higher force.
3. The Molodensky correction ζ1 may reach a few decimeters, ζ2 and

higher-order terms might frequently be negligible.
4. At the end of Sects. 2.21 and 8.8, we have remarked a curious phe-
nomenon. Using the same data, gravimetric methods seem to furnish
the vertical position (expressed by ζ or N ) roughly by one order of
magnitude better than the horizontal position (as expressed by ξ, η).
If we take the old astronomer’s rule that 1 ∼
= 30 m in position, then

1 m corresponds to 0.03 . Assume that we get 1 m in vertical position
and wish to get the same accuracy for horizontal position. This would
mean that we have to get the astronomical measurements Φ, Λ and
the deflections of the vertical ξ, η with better than 0.03 . This also
seems to apply with the order of magnitude of the Molodensky correc-
tions, where a Molodensky correction ζ1 = 0.41 m comes along with a
ξ1 = η1 = 2 , which corresponds to 60 m.
In this sense, gravimetry is weaker by one order of magnitude in deter-
mining the horizontal than the vertical position. This is an admittedly
one-sided perspective, but it was used against scientists who claimed,

still around 1960, that the gravimetric method was able to do every-
thing that satellite geodesy could. With GPS now we know better, and
without ideological scruples we combine satellite data with terrestrial
gravity.
(A second perspective of astronomical observations is the astrogeode-
tic geoid determination. Here the accuracy of astronomy is sufficient;
cf. Sec. 5.14.)
Final remark
The computation of Molodensky reductions is heavy work. So in mountain-
ous areas, least-squares collocation is definitely preferable to integration,
except for certain test computations (Sideris 1987, 1990).
Collocation also permits comparison and combination of astrogeodetic
and gravimetric data; a key paper is Kühtreiber (2002 b).
All this, however, builds on the fundamental ideas of M.S. Molodensky.
In his landmark publication (Krarup 1969) one clearly sees the transition
from Molodensky’s problem to least-squares collocation.
11.4 The geoid on the internet

The International Association of Geodesy (IAG) has a very active Interna-
tional Geoid Service (IGeS–IAG). Before you try to compute your own geoid,
look at www.iges.polimi.it to see what is available there. You can find global
and regional geoids, data, software, references, plans for future work, etc.
We particularly mention the geoid repositories:
• www.iges.polimi.it/index/geoid repo/global models.htm ,
• www.iges.polimi.it/index/geoid repo/regional models.htm .
In the latter file you can find:
• USA gravimetric Geoid 1996 (Dru Smith),
• European Geoid/Quasigeoid EGG97 (H. Denker),
• Austrian Geoid 1996 (H. Sünkel).
Other important internet addresses:
• International Gravity Bureau (Toulouse):
http://bgi.cnes.fr8110/bgi a.html ,
• International Association of Geodesy:
www.iag aig.org .
References
Abd-Elmotaal H (1995): Attraction of the topographic masses. Bulletin Géodésique,

69: 304–307.
Anger G, Moritz H (2003): Inverse problems and uncertainties in science and medi-
cine. Sitzungsberichte der Leibniz-Sozietät, 61(5): 171–212.
Anger G, Gorenflo R, Jochmann H, Moritz H, Webers W (1993): Inverse problems:
principles and applications in geophysics, technology, and medicine. Mathe-
matical Research 74: 37–44.
Bomford G (1962): Geodesy, 2nd edition. Oxford University Press.
Brouwer D, Clemence GM (1961): Methods of celestial mechanics. Academic Press,
New York.
Bruns H (1878): Die Figur der Erde. Publikation des Preussischen Geodätischen
Instituts, Berlin.
Cassinis G (1930): Sur l’adoption d’une formule internationale pour la pesanteur
normale. Bulletin Géodésique, 26: 40–49.
Civil GPS Service Interface Committee (2002): Summary report of the 40th meeting
of the Civil GPS Service Interface Committee (CGSIC), Oregon, March 22–
24. Available at www.navcen.uscg.gov/cgsic/default.htm.
Daxinger W, Stirling R (1995): Kombinierte Ausgleichung von terrestrischen und
GPS-Messungen. Österreichische Zeitschrift für Vermessung und Geoinfor-
mation, 83(1+2): 48–55.
DeLoach SR, Remondi B (1991): Decimeter positioning for dredging and hydro-
graphic surveying. In: Proceedings of the First International Symposium on
Real Time Differential Applications of the Global Positioning System, vol 1.
TÜV Rheinland, Köln: 258–263.
Department of Defense (2001): Global Positioning System standard positioning ser-
vice performance standard. Available from the U.S. Assistant for GPS, Posi-
tioning and Navigation, Defense Pentagon, Washington (DC).
Dobrin MB, Savit CH (1988): Introduction to geophysical prospecting, 4th edition.
McGraw Hill, New York.
Drinkwater MR, Floberghagen R, Haagmans R, Muzi D, Popescu A (2003): GOCE:
ESA’s first earth explorer core mission. In: Beutler GB, Drinkwater MR,
Rummel R, Steiger R von (eds): Earth gravity from space – from sensors to
earth sciences. Space Sciences Series of ISSI, 18: 419–432. Kluwer, Dordrecht.
Elmiger A (1969): Studien über Berechnung von Lotabweichungen aus Massen, In-
terpolation von Lotabweichungen und Geoidbestimmung in der Schweiz. Doc-
toral Dissertation, ETH Zürich.
Erker E, Höggerl N, Imrek E, Hofmann-Wellenhof B, Kühtreiber N (2003): The
Austrian geoid – recent steps to a new solution. Österreichische Zeitschrift
für Vermessung & Geoinformation, 91(1): 4–13.
ESA (1999): Gravity field and steady-state ocean circulation mission. Reports for
mission selection. The four candidate earth explorer core missions, SP-1233(1).
Available at http://esamultimedia.esa.int/docs/goce sp1233 1.pdf.
390 References
Fontana RD, Cheung W, Stansell T (2001): The modernized L2 civil signal. GPS
World, 12(9): 28–34.
Forsberg R, Tscherning CC (1981): The use of height data in gravity field approxi-
mation by collocation. Journal of Geophysical Research, 86 (B9): 7843–7854.
Forsberg R, Tscherning CC (1997): Topographic effects in gravity field modelling
for BVP. Available at www.gfy.ku.dk/∼cct/comored.htm.
Frank P, Mises R von (eds) (1930): Die Differential- und Integralgleichungen der
Mechanik und Physik, 2nd edition, part 1: Mathematischer Teil. Vieweg,
Braunschweig (reprint 1961 by Dover, New York and Vieweg, Braunschweig).
Galle A (1914): Das Geoid im Harz. Veröffentlichung des Geodätischen Instituts
Potsdam, vol 61.
Grafarend E, Offermanns G (1975): Eine Lotabweichungskarte Westdeutschlands
nach einem geodätisch konsistenten Kolmogorov-Wiener-Modell. Deutsche
Geodätische Kommission bei der Bayerischen Akademie der Wissenschaften,
Reihe A: Theoretische Geodäsie, vol 82.
Groten E (2004): Fundamental parameters and current (2004) best estimates of the
parameters of common relevance to astronomy, geodesy, and geodynamics.
Available at www.gfy.ku.dk/∼iag/HB2004/part5/51-groten.pdf.
Gurtner W (1978): Das Geoid in der Schweiz. Institut für Geodäsie und Photogram-
metrie der ETH Zürich, vol 20.
Gurtner W, Elmiger H (1983): Computation of geoidal heights and vertical deflec-
tions in Switzerland. Presented at the XVIII General Assembly of the IUGG
at Hamburg, August 15–27.
Heiskanen W (1924): Untersuchungen über Schwerkraft und Isostasie. Finnish Geo-
detic Institute, Helsinki, vol 4.
Heiskanen W (1928): Ist die Erde ein dreiachsiges Ellipsoid? Gerlands Beiträge zur
Geophysik, 19: 356–377.
Heiskanen WA (1957): The Columbus geoid. EOS, Transactions, American Geo-
physical Union, 38: 841–848.
Heiskanen WA, Moritz H (1967): Physical geodesy. Freeman, San Francisco London.
Heiskanen WA, Vening Meinesz FA (1958): The earth and its gravity field. McGraw-
Hill, New York.
Helmert FR (1880): Die mathematischen und physikalischen Theorien der höheren
Geodäsie, part 1. Teubner, Leipzig (reprint 1962).
Helmert FR (1884): Die mathematischen und physikalischen Theorien der Höheren
Geodäsie, part 2. Teubner, Leipzig (reprint 1962).
Hirvonen RA (1960): New theory of the gravimetric geodesy. Publications of the
Isostatic Institute of the International Association of Geodesy, Helsinki, vol
32.
Hirvonen RA (1961): The reformation of geodesy. Journal of Geophysical Research,
66: 1471–1478.
Hirvonen RA (1962): On the statistical analysis of gravity anomalies. Publications
of the Isostatic Institute of the International Association of Geodesy, Helsinki,
vol 37.
Hofmann-Wellenhof B, Kienast G, Lichtenegger H (1994): GPS in der Praxis. Sprin-
ger, Wien New York.
References 391
Hofmann-Wellenhof B, Lichtenegger H, Collins J (2001): GPS – theory and practice,

5th edition. Springer, Wien New York.
Hofmann-Wellenhof B, Legat K, Wieser M (2003): Navigation – principles of posi-
tioning and guidance. Springer, Wien New York.
Hotine M (1969): Mathematical geodesy. ESSA Monograph 2, U.S. Department of
Commerce, Washington (reprint 1992 by Springer).
Ilk KH (1999): Energiebetrachtungen für die Bewegung zweier Satelliten im Gravi-
tationsfeld der Erde. In: Krumm F, Schwarze VS (eds): Quo vadis geodesia?
Part 1. Schriftenreihe des Instituts des Studienganges Geodäsie und Geoin-
formatik der Universität Stuttgart, Report 1999/6: 191–205.
Jeffreys H (1931): An application of the free-air reduction of gravity. Gerlands
Beiträge zur Geophysik, 31: 378–386.
Jekeli C (1999): The determination of gravitational potential differences from satel-
lite-to-satellite tracking. Celestial Mechanics and Dynamical Astronomy, 75:
85–101.
Jung K (1956): Figur der Erde. In: Flügge S (ed): Handbuch der Physik, vol 47,
geophysics 1. Springer, Berlin, 534–639.
Jung K (1961): Schwerkraftverfahren in der Angewandten Geophysik. Akademische
Verlagsgesellschaft Geest & Portig, Leipzig.
Kaula WM (1959): Statistical and harmonic analysis of gravity. Journal of Geo-
physical Research, 64: 2401–2421.
Kaula WM (1963): Determination of the earth’s gravitational field. Reviews of Geo-
physics, 1: 507–551.
Kaula WM (1966 a): Theory of satellite geodesy. Blaisdell, Waltham (Massachusetts)
Toronto London.
Kaula WM (1966 b): Global harmonic and statistical analysis of gravimetry. In: Or-
lin H (ed): Proceedings of the Symposium on Extension of Gravity Anomalies
to Unsurveyed Areas. American Geophysical Union, Geophysical Monograph,
9: 58–67.
Kellogg OD (1929): Foundations of potential theory. Springer, Berlin (reprint 1954
by Dover, New York, and 1967 by Springer, Berlin Heidelberg New York).
Koch KR (1971): Die geodätische Randwertaufgabe bei bekannter Erdoberfläche.
Zeitschrift für Vermessungswesen, 96: 218–224.
Kraiger G (1987): Untersuchungen zur Prädiktion nach kleinsten Quadraten mittels
empirischer Kovarianzfunktion unter besonderer Beachtung des Krümmungs-
parameters. Mitteilungen der geodätischen Institute der Technischen Univer-
sität Graz, vol 53.
Kraiger G (1988): Influence of the curvature parameter on least-squares prediction.
Manuscripta geodaetica, 13: 164–171.
Kraiger G, Kühtreiber N, Wang YM (1987): The correction terms of the solution
of Molodensky’s problem by analytical continuation in the Central Alps of
Austria. In: Austrian Geodetic Commission (ed): The gravity field in Austria.
Geodätische Arbeiten Österreichs für die Internationale Erdmessung, Neue
Folge, vol IV: 95–109.
Krarup T (1969): A contribution to the mathematical foundation of physical geodesy.
Danish Geodetic Institute, Copenhagen, vol 44.
392 References
Kühtreiber (1990): Untersuchungen zur gravimetrischen Bestimmung von Lotab-

weichungen im Hochgebirge nach Molodensky mittels Fast-Fourier-Transfor-
mation. PhD thesis, Department of Physical Geodesy, Graz University of
Technology, Graz, Austria.
Kühtreiber (1998): Improved gravimetric geoid AGG97 of Austria. In: Forsberg R,
Feissel M, Dietrich R (eds): Geodesy on the move: gravity, geoid, geodynamics
and Antarctica. IAG Scientific Assembly, Rio de Janeiro, Brazil, September
3–9, 1997. Springer, Berlin Heidelberg, 306–311 [Schwarz K-P (ed): Interna-
tional Association of Geodesy Symposia, vol 119].
Kühtreiber (2002 a): High precision geoid determination for Austria. Habilitation,
Department of Positioning and Navigation, Graz University of Technology,
Graz, Austria.
Kühtreiber N (2002 b): High precision geoid determination of Austria using het-
erogeneous data. In: Tziavos IN (ed): Gravity and geoid 2002. Proceedings
of the Third Meeting of the International Gravity and Geoid Commission,
Thessaloniki, Greece, August 26–30, 144–149.
Lachapelle G, Cannon ME, Erickson C, Falkenberg W (1992): High precision C/A
code technology for rapid static DGPS surveys. In: Proceedings of the Sixth
International Geodetic Symposium on Satellite Positioning, Columbus, Ohio,
March 17–20, vol 1: 165–173.
Ledersteger K (1955): Der Schwereverlauf in den Lotlinien und die Berechnung der
wahren Geoidschwere. Publication dedicated to Weikko A. Heiskanen. Finnish
Geodetic Institute, Helsinki, vol 46: 109–124.
Lerch FJ, Klosko SM, Lauhscher RE, Wagner CA (1979): Gravity model improve-
ment using Geos 3 (GEM 9 and 10). Journal of Geophysical Research, 84(B8):
3897–3916.
Levallois JJ (1963): La réhabilitation de la géodésique classique et la géodésie tridi-
mensionelle. Bulletin Géodésique, 68: 193–199.
Mader K (1954): Die orthometrische Schwerekorrektion des Präzisions-Nivellements
in den Hohen Tauern. Österreichische Zeitschrift für Vermessungswesen, Son-
derheft, vol 15.
Malys S, Slater J (1994): Maintenance and enhancement of the World Geodetic
System 1984. In: Proceedings of ION GPS-94, 7th International Technical
Meeting of the Satellite Division of the Institute of Navigation, Salt Lake
City, Utah, September 20–23, part 1: 17–24.
March JG, Chang ES (1979): Global detailed gravimetric geoid. Marine Geodesy,
2(2): 145–159.
Marquis W (2001): M is for modernization. GPS World, 12(9): 36–42.
Marussi A (1985): Intrinsic geodesy. Springer, Berlin Heidelberg New York Tokyo.
McCarthy DD (ed) (1996): IERS Conventions. Observatoire de Paris, IERS Tech-
nical Note 21.
Misra P, Enge P (2001): Global Positioning System – signals, measurements, and
performance. Ganga-Jamuna, Lincoln (Mass.).
Molodenski MS (1958): Grundbegriffe der geodätischen Gravimetrie. VEB Verlag
Technik, Berlin (Russian originals 1945 and 1948).
References 393
Molodenskii MS, Eremeev VF, Yurkina MI (1962): Methods for study of the external
gravity field and figure of the earth. Israel Program of Scientific Translations,
Jerusalem (Russian original 1960).
Montenbruck O, Gill E (2001): Satellite orbits – models, methods, and applications,
corrected 2nd printing. Springer, Berlin.
Moritz H (1962): Studies on the accuracy of the computation of gravity in high ele-
vations. Publications of the Isostatic Institute of the International Association
of Geodesy, Helsinki, vol 38.
Moritz H (1963): Interpolation and prediction of point gravity anomalies. Publica-
tions of the Isostatic Institute of the International Association of Geodesy,
Helsinki, vol 40.
Moritz H (1964): Zur Bestimmung des Geoides und seiner Verwendung als Reduk-
tionsfläche. Zeitschrift für Vermessungswesen, 89: 200–202.
Moritz H (1965): Schwerevorhersage und Ausgleichungsrechnung. Zeitschrift für
Vermessungswesen, 90: 181–184.
Moritz H (1980 a): Advanced physical geodesy. Wichmann, Karlsruhe (reprint 2001
by Civil and Environmental Engineering and Geodetic Science, Ohio State
University, Columbus, Ohio).
Moritz H (1980 b): Geodetic Reference System 1980. Bulletin Géodésique, 54: 395–
405.
Moritz H (1990): The figure of the earth – theoretical geodesy and the earth’s
interior. Wichmann, Karlsruhe.
Moritz H (1995): Science, mind and the universe – an introduction to natural phi-
losophy. Wichmann, Heidlberg.
Moritz H, Hofmann-Wellenhof B (1993): Geometry, relativity, geodesy. Wichmann,
Karlsruhe.
Moritz H, Mueller II (1987): Earth rotation – theory and observation. Ungar, New
York.
Moritz H, Yurkina MI (eds) (2000): M.S. Molodensky – in memoriam. Mitteilungen
der geodätischen Institute der Technischen Universität Graz, vol 88.
Mueller II (1985): Reference coordinate systems and frames: concepts and realiza-
tion. Bulletin Géodésique, 59: 181–188.
Müller J (2001): Die Satellitengradiometriemission GOCE – Theorie, technische Re-
alisierung und wissenschaftliche Nutzung. Deutsche Geodätische Kommission
bei der Bayerischen Akademie der Wissenschaften, Reihe C, vol 541.
National Imagery and Mapping Agency (2000): Department of Defense World Geo-
detic System 1984 – its definition and relationships with local geodetic sys-
tems, 3rd edition, amendment 1. NIMA Technical Report TR 8350.2, Bethes-
da, Maryland. Available as PDF file at www.nima.mil.
Neumann F (1887): Vorlesungen über die Theorie des Potentials und der Kugel-
funktionen (edited by C. Neumann). Teubner, Leipzig.
Österreichische Kommission für die Internationale Erdmessung (ed) (1983): Das
Geoid in Österreich. Geodätische Arbeiten Österreichs für die Internationale
Erdmessung. Neue Folge, vol III.
Pail R (2003): Satellitengeodäsie. Lecture manuscript available at the Institute of
Navigation and Satellite Geodesy of the Graz University of Technology.
394 References
Pellinen LP (1962): Accounting for topography in the calculation of quasigeoidal

heights and plumb-line deflections from gravity anomalies. Bulletin Géodé-
sique, 63: 57–65.
Pizzetti P (1894): Sulla espressione della gravita alla superficie del geoide, supposto
ellissoidico. Atti della Reale Accademia dei Lincei, Rome, V(3): 166.
Pizzetti P (1911): Sopra il calcolo teorico delle deviazioni del geoide dall’ ellissoide.
Atti della Reale Accademia della Scienze di Torino, 46: 331.
Plummer HC (1918): An introductory treatise on dynamical astronomy. Cambridge
University Press (reprint 1960 by Dover, New York).
Rapp RH (1963): A consideration of Hayford’s best fitting ellipsoid data using the
differential change equations of Vening Meinesz. Geofisica pura e applicata,
54: 1–5.
Rapp R (1964): The prediction of point and mean gravity anomalies through the
use of a digital computer. Institute of Geodesy, Photogrammetry and Car-
tography, Ohio State University, vol 43.
Rapp RH (1981): The earth’s gravity field to degree and order 180 using SEASAT
altimeter data, terrestrial gravity data, and other data. Ohio State University,
Department of Geodetic Sciences, Columbus, Ohio, vol 322.
Rebhan H, Aguirre M, Johannessen J (2000): The gravity field and steady-state
ocean circulation explorer mission – GOCE. ESA Earth Observation Quar-
terly, 66: 6–11.
Reigber C, Muller H, Rizos C, Bosch W (1983): An improved earth gravity model
(GRIM 3B). Presented at the XVIII General Assembly of the IUGG at Ham-
burg, August 15–27.
Reigber C, Lühr H, Schwintzer P (eds) (2003): First CHAMP mission results for
gravity, magnetic and atmospheric studies. Springer, Berlin Heidelberg New
York.
Rinner K (1956): Über die Reduktion grosser elektronisch gemessener Entfernungen.
Zeitschrift für Vermessungswesen, 81: 47–55.
Rummel R (1986): Satellite gradiometry. In: Sünkel H (ed): Mathematical and nu-
merical techniques in physical geodesy. Springer, Berlin Heidelberg New York
London Paris Tokyo, 317–363 [Bhattacharji S, Friedman GM, Neugebauer HJ,
Seilacher A (eds): Lecture Notes in Earth Sciences, vol 7].
Rummel R, Balmino G, Johannessen J, Visser P, Woodworth P (2002): Dedicated
gravity field missions – principles and aims. Journal of Geodynamics, 33:
3–20.
Sagrebin DW (1956): Die Theorie des regularisierten Geoids. Veröffentlichung des
Geodätischen Instituts Potsdam, vol 9.
Schmitt G, Illner M, Jäger R (1991): Transformationsprobleme. Deutscher Verein für
Vermessungswesen, special issue: GPS und Integration von GPS in bestehende
geodätische Netze, vol 38: 125–142.
Schwarz KP (1976): Least-squares collocation for large systems. Bollettino di Geode-
sia e Scienze Affini, 35(3): 309–324.
Seeber G (2003): Satellite geodesy, 2nd edition. Walter de Gruyter, Berlin New
York.
Sideris M (1987): Spectral methods for the numerical solution of Molodensky’s prob-
lem. University of Calgary, Department of Surveying Engineering, vol 20024.
References 395
Sideris M (1990): Rigorous gravimetric terrain modelling using Molodensky’s oper-

ator. Manuscripta geodaetica, 15: 97–106.
Sneeuw N, Gerlach C, Svehla D, Gruber C (2002): A first attempt at time-variable
gravity recovery from CHAMP using the energy balance approach. In: Tziavos
IN (ed): Gravity and geoid 2002. Proceedings of the Third Meeting of the
International Gravity and Geoid Commission, Thessaloniki, Greece, August
26–30, 237–242.
Spilker JJ (1996): GPS signal structure and theoretical performance. In: Parkinson
BW, Spilker JJ (eds): Global Positioning System: theory and applications,
vol 1. American Institute of Aeronautics and Astronautics, Washington DC,
57–119 (Progress in Astronautics and Aeronautics, vol 163).
Sünkel H (1977): Ein nichtiteratives Verfahren zur Transformation geodätischer Ko-
ordinaten. Österreichische Zeitschrift für Vermessungswesen und Photogram-
metrie, 64(1): 29–33.
Sünkel H (1980): A general surface representation module designed for geodesy.
Ohio State University, Department of Geodetic Sciences, Columbus, Ohio,
vol 292.
Sünkel H (1983): Geoidbestimmung: Berechnungen an der TU Graz, 2. Teil. In:
Österreichische Kommission für die Internationale Erdmessung (ed): Das Geo-
id in Österreich. Geodätische Arbeiten Österreichs für die Internationale
Erdmessung, Neue Folge, vol III: 125–143.
Sünkel H, Bartelme N, Fuchs H, Hanafy M, Schuh W-D, Wieser M (1987): The
gravity field in Austria. In: Austrian Geodetic Commission (ed): The grav-
ity field in Austria. Geodätische Arbeiten Österreichs für die Internationale
Erdmessung, Neue Folge, vol IV: 47–75.
Telford WM, Geldart LP, Sheriff RE (1990): Applied geophysics, 2nd edition. Cam-
bridge University Press, Cambridge New York Port Chester Melbourne Syd-
ney.
Todhunter I (1873): A history of the mathematical theories of attraction and of the
figure of the earth from the time of Newton to that of Laplace. Macmillan,
London (reprint 1962 by Dover Publications, New York).
Tscherning CC (1976): Implementation of Algol-procedures for covariance compu-
tation on the RC 4000-computer. Danish Geodetic Institute, Copenhagen, vol
12.
Tscherning CC, Rapp RH (1974): Closed covariance expressions for gravity anoma-
lies, geoid undulations, and deflections of the vertical implied by anomaly
degree variance models. Ohio State University, Department of Geodetic Sci-
ences, Columbus, Ohio, vol 208.
Tziavos IN (ed): Gravity and geoid 2002. Proceedings of the Third Meeting of the
International Gravity and Geoid Commission, Thessaloniki, Greece, August
26–30.
Uotila UA (1960): Investigations on the gravity field and shape of the earth. Publi-
cations of the Isostatic Institute of the International Association of Geodesy,
Helsinki, vol 33.
Vening Meinesz FA (1928): A formula expressing the deflection of the plumb-line in
the gravity anomalies and some formulae for the gravity field and the grav-
ity potential outside the geoid. Proceedings of the Koninklijke Nederlandse
Akademie van Wetenschappen, 31(3): 315–331.
396 References
Vollath U, Birnbach S, Landau H, Fraile-Ordoñez JM, Martı́n-Neira M (1999):

Analysis of three-carrier ambiguity resolution technique for precise relative
positioning in GNSS-2. Navigation, 46(1): 13–23.
Vorhies J (2000): WRC 2000 results – GPS. Available at www.igeb.gov.
Zhu J (1993): Exact conversion of earth-centered, earth-fixed coordinates to geodetic
coordinates. Journal of Guidance, Control, and Dynamics, 16: 389–391.
Subject index
Accuracy of prediction 358 Baseline 87, 176, 184, 185, 189–191,

Airy–Heiskanen isostasy 146–148, 150, 211, 214, 215, 278, 280, 284, 286
376, 382 best-fitting ellipsoid 199, 233–235
analytical continuation 292, 303, 304, BIH conventional terrestrial pole 86
308–311, 313, 314, 317, 320, 321, Bouguer plate 135–137, 140, 148, 163
331, 342, 343, 384 Bouguer reduction 135, 137, 138, 141,
analytic function 44, 56, 167 146, 147, 150, 292, 367
angular velocity 43, 65, 85, 87–89, 234, boundary-value problem 29, 294, see
260, 272, 287 first –, fixed –, free –, geodetic –,
anomalous gravity field 90, 95, 321, 380 GPS –, Molodensky’s –,
anomalous potential 91, 95, 97, 98, 102, oblique-derivative –, second –, third –
104, 109, 110, 307, 333, 369, 370, 374, boundary-value problem of physical
376, 382, 384 geodesy 31
approximation, see linear –, spherical – Bruns’ equation 53
associated Legendre functions 16 Bruns’ formula 119
astrogeodetic geoid 174, 227, 330, 370, Bruns’ polyhedron 207
384, 388
astrogeodetic geoid determination 174, C/A-code 179–181
223, 330, 370, 388 center of gravity 43, 62, 105, 110, 219,
astrogeodetic method 111, 216, 217, 226
226, 227, 234, 289, 330 centrifugal force 43–45, 239, 294, 295
astrogeodetic method according to centrifugal potential 56, 66, 69, 252
Molodensky 329–331 CHAMP mission 277, 280, 282
astronomical coordinates 54, 55, 91, 92, Chandler period 219
176, 208, 210, 217, 224, 228, 229, 232, Clairaut’s theorem 71, 77, 79
289, 296, 315 cogeoid 149–151, 153, 154, 291, 292,
astronomical latitude 54, 128, 174–176, 294, 339, 376, 377
208, 334 collocation 215, 265, 274, 309, 311, 312,
astronomical leveling 226, 232, 336 323, 329, 331, 341–343, 362, 369–371,
astronomical longitude 54, 56, 128, 175, 374–377, 379, 380, 381–384, 388
176, 220, 334 comparison of different height systems
astronomical measurements 45, 92, 168
173, 175, 217, 225, 387 condensation reduction 153, 154, 292,
attraction 1, 3, 5, 7, 130–133, 135, see Helmert reduction
138–141, 146, 148, 150, 154, 239, 338, conventional spherical harmonics 23
340, 379 coordinates, see astronomical –,
autocorrelation 356, 363 ellipsoidal –, ellipsoidal harmonic –,
autocovariance function 364 geodetic –, local level –, natural –,
azimuth 105, 116, 119, 174, 207, 208, rectangular –,
210, 212, 217–219, 221, 223–226, 244, coordinate transformations 193
317, 334, 336, 339, 352, 353 Coriolis force 45
398 Subject index
correlation 346, 347, 350, 356, 359, ellipsoid, see best-fitting –, gravity field
363–366, 375 of –, mean earth –
correlation length 350 ellipsoidal coordinates 1, 37, 41, 55, 92,
correlation with elevation 346, 375 107, 119, 122–124, 128, 175, 176, 192,
correlation with height 363 193, 195, 197, 199, 202–206, 208,
covariance 342, 343, 346–350, 352, 354, 214, 217, 224, 275, 296, 299, 315, 316
356, 359–361, 363, 365, 372–374, 380 ellipsoidal harmonic coordinates 10,
covariance function 342, 346–350, 352, 34, 197
354, 356, 359–361, 363, 365, 372, 374 ellipsoidal harmonics 3, 19, 34, 37, 73,
covariance propagation 350, 373, 380 76
curvature of level surfaces 48, 53 ellipsoidal height 72, 81, 134, 167, 171,
curvature of normal plumb line 233 192, 193, 202, 203, 214, 217, 275, 276,
curvature of plumb line 48, 228 284, 296–298, 300, 301
curvature parameter 350 ellipsoidal normal 91, 92, 192, 194, 218,
297, 332–335
Eötvös unit 350
Datum, see geodetic –, local –, local
equipotential ellipsoid 84, 239, 240, 336
astrogeodetic –, three-dimensional
equipotential surface 1, 2, 46, 47, 65,
transformation of –
66, 70, 284, 300, 341, 343
datum transformations 203
error covariance function 360, 363
decomposition formula 22, 25, 26, 58 error propagation 345, 373
deflection of the vertical 91–93, 116, Euler angles 174
117, 122, 127, 141, 151, 176, 218, 221, expansion theorem 21
223, 224, 226, 234, 246, 293, 308, extrapolation of gravity anomalies 355
314, 328, 334, 336–338, 339
density 6, 7, 31
First boundary-value problem 27, 28,
DGPS, see differential GPS
31, 294
differential formulas for datum fixed boundary-value problem 295
transformations 203 flattening 1, 77, 78, 80, 86, 88, 89, 96,
differential GPS 184 97, 104, 114, 199, 205, 225, 249–251,
Dirichlet’s problem 27–29, 67 257, 261, 266, 267, 301
discontinuity 7 forbidden spherical harmonics 63
disturbing potential 91, 93, 96, 215, free-air anomaly 154, 155, 250, 292,
236, 239, 243, 248, 249, 261, 274, 290, 293, 307, 320, 322, 326, 348, 385
299, 310, 312, 319 free-air reduction 134–136, 139, 140,
Doppler 87, 182, 183, 256, 271, 331 146, 148, 150, 151, 154, 250, 293, 298,
downward continuation 253, 311–313, 322–324, 338, 344
321, 322, 342, 343, 375 free boundary-value problem 295
dynamic correction 161, 164, 165, 169 fully normalized spherical harmonics
dynamic height 159, 160, 169, 170 23, 24, 59
function, see analytic –, harmonic –
Earth’s crust 2, 129, 144, 146, 149, 289 functional 373
earth’s rotation 1, 2, 43, 272 fundamental equation of physical
eccentricity 34, 40, 66, 71, 74, 76, 78, geodesy 95, 337, 371
80, 86, 89, 197, 258, 265, 266, 280,
282 Gal 45, 84, 139, 160, 163, 169, 227,
Einstein summation convention 362 327, 384
Subject index 399
generalization of Stokes’ formula 110, GPS combination with terrestrial data

113 210
geocenter 62, 173, 174, 199, 203, 219, GPS control stations 179, 180
227 GPS leveling 171, 172, 295, 331
geocentric latitude 35, 72, 198, 242, GPS modernization concept 186
244, 250, 351 GPS satellite categories 178
geocentric longitude 9, 35, 40, 56, 92, GPS satellite signal 178, 181, 191
242 GRACE mission 282
geodetic boundary-value problem 289, gradient 5, 32, 33, 44, 48, 51, 72,
294, 330 119–122, 125, 127, 134, 138–140, 149,
geodetic coordinates 92, 224 163, 250, 280, 287, 293, 294, 306, 308,
geodetic datum 173, 174, 198, 199, 340, 350, 366, 385
204, 207 gradient formula 33, 125
Geodetic Reference System 1980 84–86
gradient vector 5, 44, 48, 294
geoid 290, 324, 341, 375, 381, 388,
gradiometry 277–280, 286, 331
see astrogeodetic –
gravimetric method 216, 226, 227, 289,
geoidal height 91, 96, 111, 216, 232,
290, 330, 331, 388
256, 274, 276, 293, 299, 303, 332, 333,
344, 374 gravitation 3, 5, 7, 43–45, 85, 239, 252
geoidal undulation 91, 93, 104, 116, gravitational constant 3
117, 121, 150, 171, 176, 192, 195, 218, gravitational mass 45, 46
223, 226, 275, 284, 292, 296, 297, 322, gravity anomaly 91, 93–95, 97, 98,
325–327, 345 99–101, 120–122, 128, 226, 227, 242,
geoid and downward continuation 342 246, 252, 292, 293, 298, 301, 304,
geoid and internet 388 317, 318, 320, 326, 333, 338, 339, 345,
geoid and quasigeoid 325 355, 356, 358, 359, 363, 366, 374
geoid by collocation 381, 382 gravity disturbance 92–94, 98, 115,
geoid determination 375 239, 243, 246, 247–249, 252, 299,
geometry of the earth 173 301, 302, 317–319
geopotential number 55, 159–161, gravity disturbance vector 92, 94, 239,
168–170, 298, 318, 325 242, 243, 249
geopotential surface 101, 104, 139, 245, gravity field 1, 2, 19, 31, 41, 48, 64, 65,
299 72, 77, 82–84, 88, 95, 101, 109, 141,
global coordinates 208 153, 166, 174, 175, 208, 215, 227, 228,
global covariance function 349 239, 240, 250, 255, 267, 274–278,
Global Positioning System 1, 2, 87, 88, 280–286, 311, 312, 321, 328–330, 336,
93, 115, 171–175, 176–184, 186, 376, 379, 380, 382
187–193, 199, 200, 202, 208, 210, gravity field of ellipsoid 64, 83, 88
211, 214, 215, 219, 227, 252, 255, 256, gravity field of the earth 43
271, 275, 277, 278, 280–286, 295, gravity field outside the earth 153, 239
299–301, 303, 317, 318, 331, 334, 388 gravity flattening 77
global reference systems 173, 175 gravity reduction 129, 135, 149–151,
GOCE mission 277, 284 154, 155, 219, 290, 292, 309, 311,
GPS, see Global Positioning System 318–323, 338–340, 342, 343, 379
GPS architecture 177 Greenwich 43, 54, 55, 63, 86, 173, 195,
GPS boundary-value problem 295 211, 220, 221, 240, 272
400 Subject index
GRS 1980, see Geodetic Reference isostatic reduction 141, 154, 292, 293,
System 1980 318, 321, 329, 340, 375, 376, 382, 386
ITRF, see International Terrestrial
Harmonic continuation 305, 306, 309, Reference Frame
322, 323, 344
harmonic function 7, 8, 12, 28, 30, 32, Kepler’s laws 258
56, 61, 97, 99, 100, 247, 248, 290, kernel function 372–374
309, 310, 312, 369 Koch’s formula 93, 115, 116
harmonics 61, see ellipsoidal –, Krarup–Runge theorem 310
spherical –, tesseral –, zonal – Krassowsky’s ellipsoid 84
height 157, see comparison of different
– systems, dynamic –, ellipsoidal –, Laplace’s equation 7–9, 10–12, 14, 31,
Helmert –, normal –, orthometric – 37, 39, 41, 95, 102, 121, 218, 224,
height above sea level 160, 318, 321, 290, 294, 304, 309, 342, 375
343 Laplace’s equation in
Helmert height 163, 166 ellipsoidal-harmonic coordinates 34,
Helmert reduction 153 37
Helmert’s formula 226 Laplace’s equation in spherical
Helmert’s projection 192 coordinates 9
Laplace spherical harmonics 22, 302
IERS 85–88, 219 latitude, see astronomical –, geocentric
IERS reference pole 86, 87 –, reduced –
least-squares collocation 265, 274, 309,
indirect effect 149–151, 153, 154,
311, 323, 329, 331, 341, 362, 369,
291–294, 296, 320, 321, 329, 340, 376,
374, 375, 379, 382, 388
377
least-squares interpolation 372, 375
inertial mass 45
least-squares prediction 361
initial point 157, 199, 224, 225
Legendre, see associated – functions
inner zone effects 125
Legendre’s differential equation 13, 14,
International Association of Geodesy 19
83, 388 Legendre’s functions 13, 14
International Ellipsoid 83, 84, 235 Legendre’s functions of the second kind
International Geoid Service 388 19
International Gravity Bureau 388 Legendre’s polynomials 15, 16, 20, 26
International Terrestrial Reference level ellipsoid 65, 67, 70, 75
Frame 87, 88 leveling 48, 157–160, 163, 164, 169–172,
International Union of Geodesy and 175, 216, 226, 228, 229, 232, 233, 284,
Geophysics 84, 310 295, 296, 298, 318, 331, 336, 337
interpolation of gravity anomalies 355 leveling, see astronomical –, GPS –
inverse problem 31, 253, 323, 342, 343 level surface 46–48, 53, 54, 100, 101,
inversion reduction of Rudzki, see 120, 157, 169, 216, 228, 291, 333, 341
Rudzki reduction linear approximation 308, 309, 313,
isostasy 141, see Airy–Heiskanen –, 318, 329
Vening Meinesz regional system, linear eccentricity 34, 66, 86, 89
Pratt–Hayford model linearization 189, 191, 200, 296, 297,
isostatic anomalies 321 300, 307
isostatic compensation 234, 235, 381 local astrogeodetic datum 224
Subject index 401
local coordinates 49 normal potential 73

local covariance function 348, 350 nutation 219
local datum 174, 216, 376
local level coordinates 208, 210–213 Oblique-derivative boundary-value
longitude, see astronomical –, problem 303
geocentric – oceanography 2, 256, 274, 275
orbital elements 258, 260–263, 265,
Marussi tensor 286 267–271
mass, see gravitational –, inertial – orthogonality 21–23, 236, 352
mass of the earth 62, 89, 110 orthogonality relations 21
mean curvature 50, 51, 72, 120, 139 orthometric correction 163–165, 169,
mean earth ellipsoid 233–237 229, 230
mean gravity 124, 162, 165, 166, 296, orthometric height 47, 55, 157, 158,
325 161, 163, 167, 169–172, 175, 176,
mean radius of the earth 193, 213, 348, 192, 211, 217, 225, 229, 230, 232, 284,
371, 374 296, 297, 299, 324–327
milligal 45 osculating ellipse 260, 267–270
misclosure 157
misclosure and second derivatives 171 P-code 179–181
Molodensky correction 293, 307, 314, Pellinen’s equivalence proof 311
318, 328, 384, 387 perturbing force 260–262, 269
Molodensky’s boundary-value problem perturbing potential 261, 269
295 physical surface of the earth 95, 129,
Molodensky series 311 167, 168, 192, 289, 292, 294, 298, 311,
Molodensky’s theory 290, 300, 311, 320–322, 324, 326
316, 326, 339 Pizzetti’s formula 103
moment of inertia 75 Pizzetti’s projection 192
plumb line 45, 46–48, 51–55, 92, 101,
Natural coordinates 54–56, 122, 175, 138, 158, 161–164, 166, 167, 170, 175,
176, 192, 217, 296 192, 207–211, 218, 228–233, 246, 252,
Neumann–Koch formula 317 296, 297, 303, 315, 324, 325, 334–336,
Neumann–Koch function 303 339, 340, 376
Neumann’s problem 29, 30, 116 Poincaré and Prey reduction 138
Newtonian gravitational constant 3 point level 304, 312, 314, 341, 343, 379
Newton integral 6, 31 point positioning with code
normal gravity 19, 34, 65, 66, 69, 71, pseudoranges 188
72, 76, 77, 79, 81–84, 86, 89–91, 93, Poisson’s equation 7, 44
94, 96, 97, 101, 104, 109, 129, 136, Poisson’s integral 27–29, 31, 32, 99,
160, 165, 167, 168, 170, 226, 233, 239, 100, 247
252, 291, 293, 298, 299, 301, 319, 320, polar motion 2, 219
325, 326, 332–334, 338–340, 371 potential 3, 5, 6–9, 14, 27, 29, 31, 44,
normal gravity field 76 46–48, 55–57, 59-66, 69, 73–75, 79,
normal gravity vector 240 86, 89–91, 93, 95–98, 100, 102, 104,
normal height 166, 167, 170, 297–300, 105, 109, 110, 113, 114, 116, 130, 133,
318, 325, 327, 339 149, 150, 152, 154, 155, 157, 158–160,
normal plumb line 101, 233, 246, 315, 167, 215, 230, 233, 236, 237, 239, 240,
325, 334–336, 339, 340, 376 243, 248, 249, 252, 257, 261, 269, 274,
402 Subject index
277–280, 282, 284, 286, 289–291, remove-restore 376, 379–381

294–299, 307, 309–314, 318, 319, 323, rotation matrix 199, 201, 214
329, 332, 333, 340, 343, 344, 346, 369, Rudzki reduction 151–153, 155, 323,
370, 374, 376, 379, 380, 382, 384 343
potential of the earth 56
potential theory 3 Satellite altimetry 2, 171, 256, 275,
Pratt–Hayford model 146, 148, 150, 154 283, 329, 331, 370
prediction 180, 285, 345, 355, 356, satellite gradiometry 278, 280, 331
358–363, 365 satellite gravity missions 274, 331
Prey reduction 140, 162, 163, 231, 325 satellite orbits 87, 226, 257, 261, 328
principal radii of curvature 72, 82
satellite-to-satellite tracking 256,
pseudorange 177, 178, 180, 182, 188,
277–279, 282, 283, 285, 331
190
scale factor 199–201
second boundary-value problem 29, 116
Quasigeoid 299, 325–328, 333, 344 sectorial spherical harmonics 18
separation of variables 38
Radius vector 9, 32, 57, 77, 97, 152, shift vector 199, 201, 214
242, 250, 259–261, 263, 266 solid spherical harmonics 12, 21, 28, 57
rectangular coordinates 9, 54–56, 62, spherical approximation 82, 96–99,
64, 73, 110, 122, 126, 192, 193–195, 104, 107, 117, 205, 235, 242–244, 246,
198, 204, 205, 208, 220, 240, 242, 243, 248, 249, 251, 301, 303, 304, 307, 371
252, 267, 268, 270–272, 275 spherical coordinates 197
recursion formula for Legendre’s spherical harmonics 11, 56, 73, 350,
polynomials 15, 19 see conventional –, forbidden –, fully
recursion formula for Molodensky normalized –, Laplace –, sectorial –,
corrections 308 solid –, surface –, tesseral –, zonal –
reduced latitude 35, 40, 198
spheropotential surface 101, 104, 139,
reduction for the curvature of the
245, 299, 326
plumb line 228
spirit leveling 157
reduction of astronomical
Stokes’ formula 93, 102, 104–110, 113,
measurements 217
115, 116, 119, 123, 125, 127–129,
reduction of azimuth 217, 218
149–151, 154, 219, 225–227, 243, 250,
reduction of gravity 134, 135, 139, 146,
292–294, 299, 302, 312, 317, 328, 345,
155, 253, 291, 293, 296, 338
reduction of horizontal angles 221 350
reduction of spatial distances 221, 222 Stokes’ function 104, 105, 108, 119,
reduction of vertical angles 221 124, 303
reference ellipsoid 39, 41, 66, 70, 72, 78, Stokes’ integral 31, 104, 105, 117, 125,
83, 84, 88–91, 96, 97, 101, 104, 105, 289, 290, 305, 325
109–114, 116, 119, 128, 169, surface, see equipotential –,
173–175, 192, 193, 198, 199, 203, 204, geopotential –, level –,
206, 216, 218, 219, 224–227, 237, 250, spheropotential –
300–302, 332, 333, 344, 346 surface spherical harmonics 12, 13, 17,
reference systems 173, 175, 267 23, 27, 28, 30, 40, 41, 108, 109, 113
regional isostasy 144
relative positioning 182, 184, 185, 187, Telluroid 297–300, 319, 320, 325, 326,
189–191 333, 338–341, 344
Subject index 403
template 122–125, 134, 136–138, 140,

150
terrain correction 136–138, 140, 146,
148, 166, 293
tesseral harmonics 18, 257, 261, 262,
267, 270, 271, 274
third boundary-value problem 29, 116
three-dimensional geodesy 174, 207
three-dimensional transformation of
datum 199–201, 203
topographic-isostatic reduction 146,
149–151, 292, 321, 330, 331, 338, 341,
377, 379, 382, 383
topography 136, 140, 141, 143,
145–148, 153, 154, 163, 183, 228, 234,
235, 256, 275, 276, 310, 313, 325, 326,
331, 338, 367, 375, 381, 382
triangulation 1, 170, 175, 176, 207, 216,
218, 224, 225
Undulation 91, 93, 104, 116, 117, 121,

149–151, 171, 176, 192, 195, 218, 223,
224, 226, 227, 275, 284, 290, 292, 294,
296, 297, 322, 325–328, 343, 345
upward continuation 100, 102, 103,
105, 247–250, 313
Variance 346–348, 355, 372, 374, 383

Vening Meinesz’ formula 116, 119, 127,
228, 244, 246, 294, 296, 315, 317, 328
Vening Meinesz’ function 119, 125
Vening Meinesz GPS formula 317
Vening Meinesz regional system 144
vertical derivative 306
vertical gradient of gravity 119
WGS 84, see World Geodetic System

1984
World Geodetic System 1984 84,
188–190, 192, 199, 201, 202, 211,
214, 267
Zenith angle 208, 210, 213, 216, 221

zenith distance, see zenith angle
zonal harmonics 18, 25, 26, 74, 108,
115, 257, 261, 271, 273

Physical Geodesy

Uploaded by

Copyright:

Available Formats

Physical Geodesy

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Physical Geodesy

Uploaded by

Copyright:

Available Formats

^ SpringerWienNewYork

This work is subject to copyright.

Typesetting: Composition by authors

With 111 Figures

Library of Congress Control Number 2006931768

ISBN-10 3-211-33544-7 SpringerWienNewYork

Weikko Aleksanteri Heiskanen

Pioneer, initiator, and coauthor of “Physical Geodesy”,

March 2005 B. Hofmann-Wellenhof

This book is a university-level introductory textbook. Physical geodesy is the

The problem of interrelating geometric and physical aspects is met here

is due to a small publication by Krarup (1969). It has an extremely simple

March 2005 B. Hofmann-Wellenhof H. Moritz

Preface to the second edition

April 2006 B. Hofmann-Wellenhof H. Moritz

1 Fundamentals of potential theory 3

2 Gravity ﬁeld of the earth 43

2.13 Spherical approximation and expansion of the disturbing

3 Gravity reduction 129

5 The geometry of the earth 173

5.3.4 System capabilities and accuracies . . . . . . . . . . . 183

6 Gravity ﬁeld outside the earth 239

7 Space methods 255

7.3 Determination of zonal harmonics . . . . . . . . . . . . . . . . 261

8 Modern views on the determination of the ﬁgure

9 Statistical methods in physical geodesy 345

9.6 Least-squares prediction . . . . . . . . . . . . . . . . . . . . . 361

10 Least-squares collocation 369

11 Computational methods 379

Subject index 397

As we have already indicated in the Preface, the subject of physical geodesy

So the geometry of the earth can be determined largely independently of

1.1 Attraction and potential

important constant, see http://physics.nist.gov/cuu/constants. (For other

expresses the force exerted by the mass m on a unit mass located at P at a

as can be easily veriﬁed by diﬀerentiating (1–6), since

In vector notation, Eq. (1–7) is written

F = [X, Y, Z] = grad V ; (1–9)

1.2 Potential of a solid body

since the element of volume is expressed by

This is the reason for the triple integrals in (1–12).

Fig. 1.2. Potential of a solid body

Analogous expressions result for Y and Z.

1.3 Harmonic functions

that is, 1/l is harmonic.

1.4 Laplace’s equation in spherical coordinates

Fig. 1.3. Spherical and rectangular coordinates

ds2 = dx2 + dy 2 + dz 2 , (1–29)

We might have found this well-known formula more simply by geometrical

It can be shown that Laplace’s operator in these coordinates is

Substituting these relations into (1–32) yields

Performing the diﬀerentiations we ﬁnd

∂2V 2 ∂V 1 ∂2V cot ϑ ∂V 1 ∂2V

∂2V ∂V ∂2V ∂V 1 ∂2V

1.5 Spherical harmonics

V (r, ϑ, λ) = f (r) Y (ϑ, λ) , (1–37)

where f is a function of r only and Y is a function of ϑ and λ only. Performing

r 2 f (r) + 2r f (r) − n(n + 1) f (r) = 0 (1–39)

f (r) = r n and f (r) = r −(n+1) ; (1–41)

this should be veriﬁed by substitution. Denoting the still unknown solutions

for m = 0. This corresponds to (1–67); here, as in (1–67), r is the greatest