Martin Vermeer

PHYSICAL
GEODESY
Aalto University publication series

SCIENCE + TECHNOLOGY 2/2020

Physical geodesy

Martin Vermeer

Aalto University
School of Engineering
Department of Built Environment
Aalto University publication series
SCIENCE + TECHNOLOGY 2/2020

© 2020 Martin Vermeer

ISBN (pdf) 978-952-60-8940-9

ISSN 1799-490X (pdf)
http://urn.fi/URN:ISBN:978-952-60-8940-9

Graphic design: Cover: Tarja Paalanen

Helsinki 2020

Finland
 Abstract
Aalto University, P.O. Box 11000, FI-00076 Aalto www.aalto.fi

Author
Martin Vermeer
Name of the publication
Physical geodesy
Publisher School of Engineering
Unit Department of Built Environment
Series Aalto University publication series SCIENCE + TECHNOLOGY 2/2020
Field of research Geodesy
Language English

Abstract

Physical geodesy studies the large-scale figure and gravity field of the Earth, which are
closely related. Our understanding of the gravity field is based on Newton’s theory of
gravitation. We present field theory, with partial differential equations describing the
behaviour of the field throughout space. Techniques for solving these equations using
boundary conditions on the Earth’s surface are explained. A central concept is the
geopotential.

The figure of the Earth is approximated by an ellipsoid of revolution, after which the
precise figure is described by small deviations from this ellipsoid. Vertical reference
systems are discussed in this context. Extending the approach to the Earth’s gravity
field yields small difference quantities, such as the disturbing potential and gravity
anomalies.

Approaches to modelling the gravity field explained are spectral development of the
field using spherical harmonics, the Stokes equation, numerical techniques based on
the Fast Fourier Transform, the remove-restore technique, and least-squares
collocation. Gravity measurement techniques are discussed, as are the multiple links
with geophysics, such as terrain effects, isostasy, mean sea level and the sea level
equation, and the tides.

Keywords figure of the Earth, gravity field, geopotential, reference ellipsoid, normal field,
disturbing potential, gravity anomaly, geoid, height system, spherical harmonics,
Stokes equation, remove-restore, least-squares collocation, gravimetry, isostasy,
mean sea level, tides
ISBN (pdf) 978-952-60-8940-9
ISSN (PDF) 1799-490X
Location of publisher Helsinki Location of printing Helsinki Year 2020
Pages 512 urn http://urn.fi/URN:ISBN:978-952-60-8940-9
 Preface

This book aims to present an overview of the current state of the study
of the Earth’s gravity field and those parts of geophysics closely related
to it, especially geodynamics, the study of the changing Earth. It has
grown out of over two decades of teaching at Helsinki’s two universities:
the Helsinki University of Technology — today absorbed into Aalto
University — and the University of Helsinki. As such, it presents
a somewhat Fennoscandian perspective on a very global subject. In
addition, the author’s own research on gravimetric geoid determination
helped shape the presentation. While there exist excellent textbooks on
all the different parts of what is presented here, I may still hope that
this text will find a niche to fill.

Helsinki, 9th December, 2020,

Martin Vermeer

–i–
ii Preface

Acknowledgements
Thanks are due to the many students and colleagues, both in academia
and at the Finnish Geodetic Institute, who have given useful comments
and corrections over the course of many years of lecturing both at the
University of Helsinki and at the Helsinki University of Technology,
today Aalto University.
Special thanks are due to the foreign students at Aalto University,
who forced me during recent years to provide an English version of this
text. The translation work also prompted a basic revision of the Finnish
text, which was long overdue as parts were written in the 1990s before
the author had had the benefit of pedagogical training. Thanks are thus
also due to Aalto University’s pedagogical training programme.
Olivier Francis is gratefully acknowledged for contributing figure
11.8.
Several map images were drawn using the Generic Mapping Tools
(Wessel et al., 2013).
The English language was competently checked by the Finnish Trans-
lation Agency Aakkosto Oy. Tarja Paalanen designed the cover. Laura
Mure and Matti Yrjölä helped with the practicalities of publishing.
This content is licenced under the Creative Commons Attribution 4.0
International (CC BY 4.0) licence, except as noted in the text or otherwise
apparent.

í ¤. û
 Contents

Chapters
Õ ! 1. Fundamentals of the theory of gravitation . . . . . . . . . . 1
Õ ! 2. The Laplace equation and its solutions . . . . . . . . . . . . 41
Õ ! 3. Legendre functions and spherical harmonics . . . . . . . . . 55
Õ ! 4. The normal gravity field . . . . . . . . . . . . . . . . . . . . . 85
Õ ! 5. Anomalous quantities of the gravity field . . . . . . . . . . . 109
Õ ! 6. Geophysical reductions . . . . . . . . . . . . . . . . . . . . . 127
Õ ! 7. Vertical reference systems . . . . . . . . . . . . . . . . . . . . 161
Õ ! 8. The Stokes equation and other integral equations . . . . . . 189
Õ ! 9. Spectral techniques, FFT . . . . . . . . . . . . . . . . . . . . . 227
Õ ! 10. Statistical methods . . . . . . . . . . . . . . . . . . . . . . . . 249
Õ ! 11. Gravimetric measurement devices . . . . . . . . . . . . . . . 293
Õ ! 12. The geoid, mean sea level, and sea-surface topography . . . 321
Õ ! 13. Satellite altimetry and satellite gravity missions . . . . . . . 341
Õ ! 14. Tides, the atmosphere, and Earth crustal movements . . . . 375
Õ ! 15. Earth gravity field research . . . . . . . . . . . . . . . . . . . 389

Õ ! A. Field theory and vector calculus — core knowledge . . . . . 393

Õ ! B. Function spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 411
Õ ! C. Why does FFT work? . . . . . . . . . . . . . . . . . . . . . . . 427
Õ ! D. Helmert condensation . . . . . . . . . . . . . . . . . . . . . . 431

– iii –
iv Contents

Õ ! E. The Laplace equation in spherical co-ordinates . . . . . . . . 441

Preface i

List of Tables xi

List of Figures xii

Acronyms xvii

1. Fundamentals of the theory of gravitation 1

1.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Gravitation between two masses . . . . . . . . . . . . . 2
1.3 The potential of a point mass . . . . . . . . . . . . . . . 5
1.4 Potential of a spherical shell . . . . . . . . . . . . . . . . 6
1.5 Computing the attraction from the potential . . . . . . 9
1.6 Potential of a solid body . . . . . . . . . . . . . . . . . . 11
1.7 Example: The potential of a line of mass . . . . . . . . . 14
1.8 Laplace and Poisson equations . . . . . . . . . . . . . . 16
1.9 Gauge invariance . . . . . . . . . . . . . . . . . . . . . . 17
1.10 Single mass-density layer . . . . . . . . . . . . . . . . . 18
1.11 Double mass-density layer . . . . . . . . . . . . . . . . . 19
1.12 The Gauss integral theorem . . . . . . . . . . . . . . . . 22
1.13 Green’s theorems . . . . . . . . . . . . . . . . . . . . . . 28
1.14 The Chasles theorem . . . . . . . . . . . . . . . . . . . . 32
1.15 Boundary-value problems . . . . . . . . . . . . . . . . . 34
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Exercise 1 – 1: Core of the Earth . . . . . . . . . . . . . . . . . . 37
Exercise 1 – 2: Atmosphere . . . . . . . . . . . . . . . . . . . . . 37
Exercise 1 – 3: The Gauss theorem . . . . . . . . . . . . . . . . 38

2. The Laplace equation and its solutions 41

2.1 The nature of the Laplace equation . . . . . . . . . . . . 41

í ¤. û
Contents
v
2.2 The Laplace equation in rectangular co-ordinates . . . 43
2.3 The Laplace equation in polar co-ordinates . . . . . . . 48
2.4 Spherical, geodetic, ellipsoidal co-ordinates . . . . . . . 50
2.5 The Laplace equation in spherical co-ordinates . . . . . 52
2.6 Dependence on height . . . . . . . . . . . . . . . . . . . 54

3. Legendre functions and spherical harmonics 55

3.1 Legendre functions . . . . . . . . . . . . . . . . . . . . . 55
3.2 Symmetry properties of spherical harmonics . . . . . . 62
3.3 Orthogonality of Legendre functions . . . . . . . . . . . 64
3.4 Low-degree spherical harmonics . . . . . . . . . . . . . 66
3.5 Splitting a function into degree constituents . . . . . . 68
3.6 Spectral representations of various quantities . . . . . . 70
3.7 Often-used spherical-harmonic expansions . . . . . . . 73
3.8 Ellipsoidal harmonics . . . . . . . . . . . . . . . . . . . 74
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Exercise 3 – 1: Attenuation with height of a spherical-harmonic
expansion . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Exercise 3 – 2: Symmetries of spherical harmonics . . . . . . . 81
Exercise 3 – 3: Algebraic-sign domains of spherical harmonics 82
Exercise 3 – 4: Escape velocity . . . . . . . . . . . . . . . . . . . 83

4. The normal gravity field 85

4.1 The basic idea of a normal field . . . . . . . . . . . . . . 85
4.2 The centrifugal force and its potential . . . . . . . . . . 87
4.3 Level surfaces and plumb lines . . . . . . . . . . . . . . 90
4.4 Natural co-ordinates . . . . . . . . . . . . . . . . . . . . 94
4.5 The normal potential in ellipsoidal co-ordinates . . . . 95
4.6 Normal gravity on the reference ellipsoid . . . . . . . . 97
4.7 Numerical values and calculation formulas . . . . . . . 99
4.8 The normal potential as a spherical-harmonic expansion 101
4.9 The disturbing potential . . . . . . . . . . . . . . . . . . 103
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 105

í ¤. û
vi Contents

Exercise 4 – 1: The Somigliana–Pizzetti equation . . . . . . . . 106

Exercise 4 – 2: Centrifugal force . . . . . . . . . . . . . . . . . . 107

5. Anomalous quantities of the gravity field 109

5.1 Disturbing potential, geoid height, deflections of the
plumb line . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.2 Gravity disturbances . . . . . . . . . . . . . . . . . . . . 114
5.3 Gravity anomalies . . . . . . . . . . . . . . . . . . . . . 115
5.4 Units used for gravity anomalies . . . . . . . . . . . . . 117
5.5 The boundary-value problem of physical geodesy . . . 118
5.6 The telluroid mapping and the “quasi-geoid” . . . . . 121
5.7 Free-air anomalies . . . . . . . . . . . . . . . . . . . . . 122
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Exercise 5 – 1: The spectrum of gravity anomalies . . . . . . . 125
Exercise 5 – 2: Deflections of the plumb line and geoid tilt . . . 125
Exercise 5 – 3: Gravity anomaly, geoid height . . . . . . . . . . 126

6. Geophysical reductions 127

6.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.2 Bouguer anomalies . . . . . . . . . . . . . . . . . . . . . 128
6.3 Terrain effect and terrain correction . . . . . . . . . . . 133
6.4 Spherical Bouguer anomalies . . . . . . . . . . . . . . . 138
6.5 Helmert condensation . . . . . . . . . . . . . . . . . . . 140
6.6 Isostasy . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.7 Isostatic reductions . . . . . . . . . . . . . . . . . . . . . 149
6.8 The “isostatic geoid” . . . . . . . . . . . . . . . . . . . . 152
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 157
Exercise 6 – 1: Gravity anomaly . . . . . . . . . . . . . . . . . . 157
Exercise 6 – 2: Bouguer reduction . . . . . . . . . . . . . . . . . 157
Exercise 6 – 3: Terrain correction and Bouguer reduction . . . 158
Exercise 6 – 4: Isostasy . . . . . . . . . . . . . . . . . . . . . . . 158

7. Vertical reference systems 161

í ¤. û
Contents
vii
7.1 Levelling, orthometric heights and the geoid . . . . . . 161
7.2 Orthometric heights . . . . . . . . . . . . . . . . . . . . 163
7.3 Normal heights . . . . . . . . . . . . . . . . . . . . . . . 166
7.4 Difference between geoid height and height anomaly . 173
7.5 Difference between orthometric and normal heights . . 176
7.6 Calculating orthometric heights precisely . . . . . . . . 176
7.7 Calculating normal heights precisely . . . . . . . . . . . 179
7.8 Calculation example for heights . . . . . . . . . . . . . 180
7.9 Orthometric and normal corrections . . . . . . . . . . . 181
7.10 A vision for the future: relativistic levelling . . . . . . . 184
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Exercise 7 – 1: Calculating orthometric heights . . . . . . . . . 186
Exercise 7 – 2: Calculating normal heights . . . . . . . . . . . . 187
Exercise 7 – 3: Difference between orthometric and normal height 187

8. The Stokes equation and other integral equations 189

8.1 The Stokes equation and the Stokes integral kernel . . . 189
8.2 Example: The Stokes equation in polar co-ordinates . . 193
8.3 Plumb-line deflections and Vening Meinesz equations 198
8.4 The Poisson integral equation . . . . . . . . . . . . . . . 199
8.5 Gravity anomalies in the exterior space . . . . . . . . . 202
8.6 The vertical gradient of the gravity anomaly . . . . . . 204
8.7 Gravity reductions in geoid determination . . . . . . . 209
8.8 The remove–restore method . . . . . . . . . . . . . . . . 215
8.9 Kernel modification . . . . . . . . . . . . . . . . . . . . 216
8.10 Advanced kernel modifications . . . . . . . . . . . . . . 220
8.11 Block integration . . . . . . . . . . . . . . . . . . . . . . 222
8.12 Effect of the local zone . . . . . . . . . . . . . . . . . . . 224
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Exercise 8 – 1: The Stokes equation . . . . . . . . . . . . . . . . 226

9. Spectral techniques, FFT 227

9.1 The Stokes theorem as a convolution . . . . . . . . . . . 227

í ¤. û
viii Contents

9.2 Integration by FFT . . . . . . . . . . . . . . . . . . . . . 230

9.3 Solution in latitude and longitude . . . . . . . . . . . . 232
9.4 Bordering and tapering of the data area . . . . . . . . . 239
9.5 Computing a geoid model with FFT . . . . . . . . . . . 241
9.6 Use of FFT in other contexts . . . . . . . . . . . . . . . . 244
9.7 Computing terrain corrections with FFT . . . . . . . . . 244
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 248

10. Statistical methods 249

10.1 The role of uncertainty in geophysics . . . . . . . . . . 249
10.2 Linear functionals . . . . . . . . . . . . . . . . . . . . . 250
10.3 Statistics on the Earth’s surface . . . . . . . . . . . . . . 251
10.4 The covariance function of the gravity field . . . . . . . 254
10.5 Least-squares collocation . . . . . . . . . . . . . . . . . 256
10.6 Prediction of gravity anomalies . . . . . . . . . . . . . . 268
10.7 Covariance function and degree variances . . . . . . . 270
10.8 Propagation of covariances . . . . . . . . . . . . . . . . 274
10.9 Global covariance functions . . . . . . . . . . . . . . . . 279
10.10 Collocation and the spectral viewpoint . . . . . . . . . 280
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 284
Exercise 10 – 1: Variance of prediction . . . . . . . . . . . . . . 286
Exercise 10 – 2: Hirvonen’s covariance equation and prediction 286
Exercise 10 – 3: Predicting gravity anomalies . . . . . . . . . . 287
Exercise 10 – 4: Predicting gravity anomalies (2) . . . . . . . . 288
Exercise 10 – 5: Propagation of covariances . . . . . . . . . . . 289
Exercise 10 – 6: Kaula’s rule for gravity gradients . . . . . . . . 289
Exercise 10 – 7: Underground mass points . . . . . . . . . . . . 291

11. Gravimetric measurement devices 293

11.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
11.2 The relative or spring gravimeter . . . . . . . . . . . . . 295
11.3 The absolute or ballistic gravimeter . . . . . . . . . . . 302
11.4 Network hierarchy in gravimetry . . . . . . . . . . . . . 307

í ¤. û
Contents
ix
11.5 The superconducting gravimeter . . . . . . . . . . . . . 308
11.6 Atmospheric influence on gravity measurement . . . . 310
11.7 Airborne gravimetry and GNSS . . . . . . . . . . . . . 312
11.8 Measuring the gravity gradient . . . . . . . . . . . . . . 315
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 317
Exercise 11 – 1: Absolute gravimeter . . . . . . . . . . . . . . . 318
Exercise 11 – 2: Spring gravimeter . . . . . . . . . . . . . . . . . 319
Exercise 11 – 3: Air pressure and gravity . . . . . . . . . . . . . 320

12. The geoid, mean sea level, and sea-surface topography 321
12.1 Basic concepts . . . . . . . . . . . . . . . . . . . . . . . . 321
12.2 Geoids and national height datums . . . . . . . . . . . 323
12.3 The geoid and post-glacial land uplift . . . . . . . . . . 325
12.4 Methods for determining the sea-surface topography . 329
12.5 Global sea-surface topography and heat transport . . . 329
12.6 The global behaviour of the sea level . . . . . . . . . . . 333
12.7 The sea-level equation . . . . . . . . . . . . . . . . . . . 334
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 339
Exercise 12 – 1: Coriolis force, ocean current . . . . . . . . . . . 340
Exercise 12 – 2: Land subsidence and the mechanism of land
uplift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340

13. Satellite altimetry and satellite gravity missions 341

13.1 Satellite altimetry . . . . . . . . . . . . . . . . . . . . . . 341
13.2 Crossover adjustment . . . . . . . . . . . . . . . . . . . 346
13.3 Choice of satellite orbit . . . . . . . . . . . . . . . . . . . 355
13.4 In-flight calibration . . . . . . . . . . . . . . . . . . . . . 362
13.5 Retracking . . . . . . . . . . . . . . . . . . . . . . . . . . 363
13.6 Oceanographic research using satellite altimetry . . . . 365
13.7 Satellite gravity missions . . . . . . . . . . . . . . . . . 366
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 372
Exercise 13 – 1: Altimetry, crossover adjustment . . . . . . . . 373
Exercise 13 – 2: Satellite orbit . . . . . . . . . . . . . . . . . . . 374

í ¤. û
x Contents

Exercise 13 – 3: Kepler’s third law . . . . . . . . . . . . . . . . . 374

14. Tides, the atmosphere, and Earth crustal movements 375

14.1 The theoretical tide . . . . . . . . . . . . . . . . . . . . . 375
14.2 Deformation caused by the tidal force . . . . . . . . . . 380
14.3 The permanent part of the tide . . . . . . . . . . . . . . 383
14.4 Tidal corrections between height systems . . . . . . . . 385
14.5 Loading of the Earth’s crust by sea and atmosphere . . 386
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 387
Exercise 14 – 1: Tide . . . . . . . . . . . . . . . . . . . . . . . . . 388

15. Earth gravity field research 389

15.1 Internationally . . . . . . . . . . . . . . . . . . . . . . . 389
15.2 Europe . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
15.3 The Nordic countries . . . . . . . . . . . . . . . . . . . . 391
15.4 Finland . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
15.5 Textbooks . . . . . . . . . . . . . . . . . . . . . . . . . . 392

A. Field theory and vector calculus — core knowledge 393

A.1 Vector calculus . . . . . . . . . . . . . . . . . . . . . . . 393
A.2 Scalar and vector fields . . . . . . . . . . . . . . . . . . . 397
A.3 Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
A.4 The continuity of matter . . . . . . . . . . . . . . . . . . 409

B. Function spaces 411

B.1 An abstract vector space . . . . . . . . . . . . . . . . . . 411
B.2 The Fourier function space . . . . . . . . . . . . . . . . 412
B.3 Sturm–Liouville differential equations . . . . . . . . . . 416
B.4 Legendre polynomials . . . . . . . . . . . . . . . . . . . 422
B.5 Spherical harmonics . . . . . . . . . . . . . . . . . . . . 422
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 425
Exercise B – 1: Orthonormality of the Fourier basis functions . 425

C. Why does FFT work? 427

í ¤. û
List of Tables
xi
D. Helmert condensation 431
D.1 The exterior potential of the topography . . . . . . . . . 432
D.2 The interior potential of the topography . . . . . . . . . 433
D.3 The exterior potential of the condensation layer . . . . 435
D.4 Total potential of Helmert condensation . . . . . . . . . 435
D.5 The dipole method . . . . . . . . . . . . . . . . . . . . . 438

E. The Laplace equation in spherical co-ordinates 441

E.1 Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . 441
E.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 444

Bibliography 447

Index 465

List of Tables
3.1 Legendre polynomials . . . . . . . . . . . . . . . . . . . . 56
3.2 Associated Legendre functions . . . . . . . . . . . . . . . 58
3.3 Semi-wavelengths for different degrees and orders . . . . 61
3.4 Plotting a surface spherical-harmonic map . . . . . . . . . 64
3.5 EGM96 coefficients and mean errors . . . . . . . . . . . . 75
3.6 Legendre functions of the second kind . . . . . . . . . . . 78

4.1 GRS80 normal potential spherical-harmonic coefficients . 102

5.1 Orders of magnitude of gravity variations . . . . . . . . . 119

8.1 Stokes equation in two dimensions, octave code . . . . . 197

8.2 Derivation of the kernel for the vertical gradient of gravity
anomaly . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

13.1 Altimetric satellites through the ages . . . . . . . . . . . . 342

í ¤. û
xii List of Figures

13.2 Calculating the height of a satellite from its period . . . . 362

14.1 The various periods in the theoretical tide . . . . . . . . . 379

List of Figures
1.1 Gravitation is universal . . . . . . . . . . . . . . . . . . . . 3
1.2 A thin spherical shell consists of rings . . . . . . . . . . . 7
1.3 Dependence of potential and attraction on distance. . . . . 9
1.4 A double mass-density layer . . . . . . . . . . . . . . . . . 20
1.5 A graphical explanation of the Gauss integral theorem . 23
1.6 A little rectangular box . . . . . . . . . . . . . . . . . . . . 25
1.7 Eight-unit cube . . . . . . . . . . . . . . . . . . . . . . . . 27
1.8 Green’s third theorem for an exterior point . . . . . . . . 30
1.9 Green’s third theorem for an interior point . . . . . . . . . 31
1.10 Green’s third theorem for the space external to a body . . 32
1.11 Iron ore body . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.1 Attenuation of the gravitational field with height . . . . . 46

2.2 Vertically shifting the potential field . . . . . . . . . . . . 47
2.3 Definition of spherical co-ordinates . . . . . . . . . . . . . 51
2.4 Definition of geodetic co-ordinates . . . . . . . . . . . . . 52

3.1 Some Legendre polynomials . . . . . . . . . . . . . . . . . 56

3.2 Associated Legendre functions . . . . . . . . . . . . . . . 58
3.3 The algebraic signs of spherical harmonics . . . . . . . . . 60
3.4 Surface spherical harmonics as maps . . . . . . . . . . . . 60
3.5 Monopole, dipole, and quadrupole . . . . . . . . . . . . . 68
3.6 Radially shifting the potential field . . . . . . . . . . . . . 71

4.1 The normal gravity field of the Earth . . . . . . . . . . . . 86

4.2 Gravitation and centrifugal force . . . . . . . . . . . . . . 87
4.3 The curvature of level surfaces . . . . . . . . . . . . . . . 91
4.4 The curvature of the plumb line . . . . . . . . . . . . . . . 93

í ¤. û
List of Figures
xiii
4.5 Natural co-ordinates . . . . . . . . . . . . . . . . . . . . . 94
4.6 Meridian ellipse and latitude types . . . . . . . . . . . . . 98
4.7 The normal field’s potential over the equator . . . . . . . 101

5.1 Geoid undulations and deflections of the plumb line . . . 110

5.2 A Finnish geoid model . . . . . . . . . . . . . . . . . . . . 112
5.3 Equipotential surfaces of gravity and normal gravity fields 113
5.4 Various reference surfaces . . . . . . . . . . . . . . . . . . 118
5.5 Free-air gravity anomalies for Southern Finland. . . . . . . 124

6.1 The attraction of a Bouguer plate . . . . . . . . . . . . . . 129

6.2 The Bouguer plate as an approximation to the topography 131
6.3 Behaviour of different anomaly types in the mountains . 132
6.4 Terrain-corrected Bouguer anomalies for Southern Finland 133
6.5 Calculating the terrain correction using the prism method 135
6.6 The steps in computing the Bouguer anomaly . . . . . . . 136
6.7 A special terrain shape . . . . . . . . . . . . . . . . . . . . 136
6.8 Helmert condensation and the gravity field . . . . . . . . 141
6.9 Friedrich Robert Helmert . . . . . . . . . . . . . . . . . . . 142
6.10 Isostasy and the bending of plumb lines . . . . . . . . . . 143
6.11 Pratt–Hayford isostatic hypothesis . . . . . . . . . . . . . 144
6.12 Airy–Heiskanen isostatic hypothesis . . . . . . . . . . . . 145
6.13 Quantities in isostatic compensation . . . . . . . . . . . . 145
6.14 The modern understanding of isostasy . . . . . . . . . . . 148
6.15 Isostatic gravity anomalies for Southern Finland. . . . . . 151
6.16 Isostatic reduction as a pair of surface density layers . . . 154
6.17 Terrain shape . . . . . . . . . . . . . . . . . . . . . . . . . . 158

7.1 The principle of levelling . . . . . . . . . . . . . . . . . . . 162

7.2 Height reference pillar in Kaivopuisto, Helsinki . . . . . . 164
7.3 Levelled heights and geopotential numbers . . . . . . . . 165
7.4 Lake Päijänne: water flows “uphill” . . . . . . . . . . . . 167
7.5 Mikhail Sergeevich Molodensky . . . . . . . . . . . . . . . 168

í ¤. û
xiv List of Figures

7.6 A graphic cartoon of the proof of Molodensky’s find . . . 172

7.7 Geoid, quasi-geoid, telluroid and topography . . . . . . . 172
7.8 An optical lattice clock . . . . . . . . . . . . . . . . . . . . 185

8.1 The principle of gravimetric geoid determination . . . . . 190

8.2 Integrating the Stokes equation geometrically . . . . . . . 191
8.3 The Stokes kernel function . . . . . . . . . . . . . . . . . . 192
8.4 The Stokes kernel function on the circle . . . . . . . . . . 196
8.5 Simulation of gravity anomalies and geoid undulations . 198
8.6 The geometry of the generating function of the Legendre
polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . 200
8.7 The Poisson kernel for gravity anomalies . . . . . . . . . . 205
8.8 Residual terrain modelling (RTM) . . . . . . . . . . . . . . 212
8.9 The remove–restore method as a commutative diagram . 216
8.10 Modified Stokes kernel functions . . . . . . . . . . . . . . 219
8.11 Simpson integration nodal weights in two dimensions . . 223

9.1 Map projection co-ordinates in the local tangent plane . . 228

9.2 Commutative diagram for FFT. . . . . . . . . . . . . . . . 232
9.3 “Tapering” 25 % . . . . . . . . . . . . . . . . . . . . . . . . 240
9.4 Example images for FFT transform . . . . . . . . . . . . . 241
9.5 The Finnish FIN2000 geoid . . . . . . . . . . . . . . . . . . 243

10.1 Geocentric angular distance and azimuth . . . . . . . . . 255

10.2 Hirvonen’s covariance function in two dimensions . . . . 262
10.3 An example of least-squares collocation . . . . . . . . . . 263
10.4 Collocation example . . . . . . . . . . . . . . . . . . . . . 265
10.5 Global covariance functions as degree variances . . . . . 281
10.6 Circular geometry . . . . . . . . . . . . . . . . . . . . . . . 283

11.1 Jean Richer’s report . . . . . . . . . . . . . . . . . . . . . . 294

11.2 Autograv CG-5 spring gravimeter . . . . . . . . . . . . . . 295
11.3 Operating principle of spring gravimeter . . . . . . . . . 298
11.4 The idea of astatisation . . . . . . . . . . . . . . . . . . . . 301

í ¤. û
List of Figures
xv
11.5 Operating principle of a ballistic absolute gravimeter . . 303
11.6 absolute gravimeter of type FG5 . . . . . . . . . . . . . . . 304
11.7 Principle of operation of an atomic gravimeter . . . . . . 307
11.8 International intercomparison of absolute gravimeters . . 308
11.9 Principle of operation of a superconducting gravimeter . 309

12.1 The mechanism of post-glacial land uplift . . . . . . . . . 327

12.2 The Fennoscandian gravity line on the 63rd parallel north 328
12.3 Connection between sea-surface topography and ocean
currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
12.4 Sea-surface topography map produced by GOCE . . . . . 332
12.5 The sea-level equation . . . . . . . . . . . . . . . . . . . . 335
12.6 Sea-level rise after the last ice age . . . . . . . . . . . . . . 337

13.1 Results from the TOPEX/Poseidon and Jason satellites . 344

13.2 Satellite altimetry as a measurement method . . . . . . . 345
13.3 Simple crossover geometry . . . . . . . . . . . . . . . . . . 348
13.4 Track geometry of satellite altimetry . . . . . . . . . . . . 353
13.5 Kepler’s orbital elements . . . . . . . . . . . . . . . . . . . 356
13.6 The mechanism of a Sun-stationary orbit . . . . . . . . . . 358
13.7 Geometry of a “no-shadow” orbit . . . . . . . . . . . . . . 359
13.8 A satellite orbit example . . . . . . . . . . . . . . . . . . . 360
13.9 Analysing the altimeter return pulse . . . . . . . . . . . . 363
13.10 Ice volume on the Arctic Ocean . . . . . . . . . . . . . . . 365
13.11 Gravity field determination from GPS orbital tracking . . 368
13.12 The principle of the GRACE satellites . . . . . . . . . . . . 369
13.13 GRACE mission results . . . . . . . . . . . . . . . . . . . . 370
13.14 Gravity field determination with GOCE . . . . . . . . . . 371
13.15 Satellite altimetric track geometry . . . . . . . . . . . . . . 373

14.1 Theoretical tide . . . . . . . . . . . . . . . . . . . . . . . . 376

14.2 The main components of the theoretical tide . . . . . . . . 380
14.3 Constituents of the permanent tide . . . . . . . . . . . . . 384

í ¤. û
xvi List of Figures

A.1 Exterior or vectorial product . . . . . . . . . . . . . . . . . 396

A.2 Kepler’s second law . . . . . . . . . . . . . . . . . . . . . . 397
A.3 The gradient . . . . . . . . . . . . . . . . . . . . . . . . . . 400
A.4 The divergence . . . . . . . . . . . . . . . . . . . . . . . . . 401
A.5 The curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
A.6 The Stokes curl theorem . . . . . . . . . . . . . . . . . . . 406
A.7 The Gauss integral theorem . . . . . . . . . . . . . . . . . 408

B.1 Fourier analysis on a step function . . . . . . . . . . . . . 415

E.1 Gauss integral theorem applied to a co-ordinate aligned

volume element . . . . . . . . . . . . . . . . . . . . . . . . 442

í ¤. û
 Acronyms

ABCDEFGHIJKLMNOPRSTUW
A
ACS Advanced Camera for Surveys, instrument on the Hubble Space Telescope
3 ^
AGU American Geophysical Union 390 ^

B
BGI Bureau Gravimétrique International, International Gravity Bureau 124, 133,
151, 389, 390 ^
BVP boundary-value problem 34 ^

C
CHAMP Challenging Minisatellite Payload for Geophysical Research and Ap-
plications 74, 244, 367, 368 ^

D
DMA Defense Mapping Agency (USA) 73 ^
DORIS Doppler Orbitography and Radiopositioning Integrated by Satellite, a
French satellite positioning system 342, 343 ^
DTM digital terrain model 134 ^

– xvii –
xviii ABCDEFGHIJKLMNOPRSTUW Acronyms

E
EGM2008 Earth Gravity Model 2008 61, 74, 117, 124, 133 ^
EGM96 Earth Gravity Model 1996 73–75 ^
EGU European Geosciences Union 390 ^
ENSO El Niño Southern Oscillation 321, 344 ^
ESA European Space Agency 3, 342, 343, 369 ^

F
FAS member of the French Academy of Sciences (Académie des sciences) 422 ^
FFT Fast Fourier Transform 15, 150, 197, 198, 214, 216, 230, 232, 234, 235, 237,
239–242, 244, 245, 247, 248, 281, 284, 324, 427, 428 ^
FGI Finnish Geospatial Research Institute, formerly Finnish Geodetic Institute
392 ^
FIN2000 geoid model (Finland) 242, 243, 324 ^
FIN2005N00 geoid model (Finland) 242, 391 ^
FRAS Fellow of the Royal Astronomical Society 294 ^
FRS Fellow of the Royal Society (of London) 4, 17, 223, 294, 379, 381, 422 ^
FRSE Fellow of the Royal Society of Edinburgh 17, 379, 422 ^

G
GDR geophysical data record 346, 363 ^
GFZ Geoforschungszentrum (Potsdam, Germany), German Research Centre for
Geosciences 367 ^
GIA glacial isostatic adjustment 325, 337, 338 ^
GNSS Global Navigation Satellite Systems, which comprise, besides American
GPS, the satellite positioning systems of other countries, such as the
Russian GLONASS and the European Galileo 115, 128, 186, 241, 242,
313, 314, 324, 325, 329, 342, 346, 362, 366, 370, 382, 383, 387, 391 ^
GOCE Geopotential and Steady-state Ocean Circulation Explorer 74, 244, 281,
315, 331, 332, 366, 369–372 ^
GPS Global Positioning System 68, 99, 171, 312, 318, 342, 367, 368 ^
GRACE Gravity Recovery And Climate Experiment 74, 244, 367–370 ^
GRAVSOFT Geopotential determination software, mainly developed in Denmark
242 ^
GRS80 Geodetic Reference System 1980 6, 99, 102, 106, 110, 117, 242, 312,

í ¤. û
Acronyms ABCDEFGHIJKLMNOPRSTUW
xix
324 ^
GWR20 superconducting gravimeter built by GWR Instruments 309 ^

H
HUT Helsinki University of Technology, today part of Aalto University 392 ^

I
IAG International Association of Geodesy 242, 309, 389, 390 ^
IB inverted barometer 321 ^
ICET International Center for Earth Tides 390 ^
ICGEM International Center for Global Earth Models 390 ^
IDEMS International Digital Elevation Model Service 390 ^
IGeC International Geoid Commission (obsolete) 390 ^
IGeS International Geoid Service (obsolete) 390 ^
IGETS International Geodynamics and Earth Tide Service 309 ^
IGFS International Gravity Field Service 389, 390 ^
IMGC-02 transportable absolute rise-and-fall gravimeter, built by the Istituto
di Metrologia «G. Colonnetti», formerly the Istituto Nazionale di Ricerca
Metrologica in Torino, Italy. 306 ^
ISG International Service for the Geoid 389, 390 ^

J
J2 second dynamic form factor, “gravitational flattening” 74, 99, 102, 357, 358,
372 ^
Jason American-French-European radar altimetry satellite series, successors of
TOPEX/Poseidon 342–344, 359, 372 ^
JHU Johns Hopkins University 3 ^

K
KCB Knight Commander of the Bath, British order of chivalry 379 ^
KKJ National Grid Co-ordinate System (Finland) 123 ^

í ¤. û
xx ABCDEFGHIJKLMNOPRSTUW Acronyms

L
LAGEOS LAser GEOdynamic Satellite 356 ^
Lego™ “Leg Godt”, Engl. “Play Well”, Danish toy brand 25 ^
LLR Lunar laser ranging 302 ^
LSC least-squares collocation 150 ^

M
Mf Moon, fortnightly tide 379 ^

N
N2000 height system (Finland) 163, 166, 242, 323 ^
N60 height system (Finland) 162, 163, 242, 323, 324 ^
NAP Normaal Amsterdams Peil, height system (Western Europe) 163 ^
NASA National Aeronautics and Space Administration (USA) 3 ^
NAVD88 North American Vertical Datum 1988 163 ^
NC normal correction 183 ^
NGA National Geospatial-Intelligence Agency (USA, formerly NIMA) 73,
389 ^
NIMA National Imagery and Mapping Agency (USA, formerly DMA) 73 ^
NKG Nordiska Kommissionen för Geodesi, Nordic Geodetic Commission 391 ^
NKG2004 geoid model (Nordic area) 391 ^
NKG2015 geoid model (Nordic area) 391 ^
NOAA National Oceanic and Atmospheric Administration (USA) 332 ^

O
OC orthometric correction 181 ^
OSU Ohio State University, Columbus, Ohio, USA 73 ^

P
ppb parts per billion 118 ^
ppm parts per million 118 ^
PRARE Precise Range And Range-Rate Equipment, not operational 342 ^
PRS President of the Royal Society 2, 120, 143 ^

R
RTM residual terrain modelling 210–212, 226 ^

í ¤. û
Acronyms ABCDEFGHIJKLMNOPRSTUW
xxi
S
SISystème international d’unités, International System of Units 11, 37, 99, 118,
119, 125 ^
SK-42 Reference system of the Soviet Union, also known as the Krasovsky 1940
reference ellipsoid 100 ^
SLR satellite laser ranging 342, 356 ^
SRAL Synthetic Aperture Radar Altimeter 343 ^
Ssa Sun, semi-annual tide 379 ^
SST satellite-to-satellite tracking 369 ^
STScI Space Telescope Science Institute 3 ^
SWH significant wave height 345, 346 ^

T
correction 132, 134, 135, 137, 138, 247 ^
TC terrain
TOPEX/Poseidon Topography Experiment / Poseidon, American-French radar
altimetry satellite xix, 323, 342–344, 359, 372 ^

U
UCO University of California Observatories 3 ^

W
WGS84 World Geodetic System 1984 68, 99 ^

x
x

í ¤. û
^ Fundamentals of the theory of
gravitation

^ 1.1 General
1
In this chapter we present the foundations of Newton’s theory of
gravitation. Intuitively, the theory of gravitation is easiest to understand
as “action at a distance,” Latin actio ad distans, where the force between
two masses is proportional to the masses themselves, and inversely
proportional to the square of the distance between them. This is the
form of Newton’s general law of gravitation familiar to all.
There exists an alternative but equivalent presentation, field theory,
which portrays gravitation as a phenomenon propagating through
space, a field. The propagation is expressed in field equations. The field
approach is not quite as intuitive, but is a powerful theoretical tool.1
In this chapter we acquaint ourselves with the central concept of field
theory, the gravitational potential. We also get to know the potential
fields of the theoretically interesting single and double mass-density
layers. Practical and theoretical applications of these include isostasy

1 There is also a philosophical side to this.

To many, for example to Leibniz, the idea of
a force that jumps from object to object was an abomination. Many tried to explain
gravitation — and also electromagnetism, etc. — by a “world aether”. It was not until
the advent of relativity theory that the understanding gained ground that a physical
theory does not have to satisfy our preconceptions about what constitutes a “sensible
explanation” — as long as it presents the physical phenomena correctly.

–1–
2 Fundamentals of the theory of gravitation

and Helmert condensation, the properties of which we will study in

later chapters.
We will learn about important integral theorems like the theorems
of Gauss and Green, with the aid of which we may infer the whole
potential field in space from field values given only for a certain surface.
Other similar examples are the Chasles theorem and the solution to
Dirichlet’s problem.
In chapter 3, we apply these fundamentals of potential theory to
derive a spectral representation of the Earth’s gravitational field, a
pallofunktio- spherical-harmonic expansion.
kehitelmä
Here, at the beginning of the text, we derive an extensive set of
mathematical equations, such as well-known integral equations. This is
unfortunately necessary preliminary work. These equations, however,
are not an end in themselves and are not worth committing to memory.
Try rather to understand their logic, and how these various results
have been arrived at historically. Try as well to acquire an intuition, a
fingertip feeling, about the nature of this theory.

^ 1.2 Gravitation between two masses

We start the investigation of the Earth’s gravity field suitably with Isaac
Newton’s2 general law of gravitation:
m1 m2
F=G . (1.1)
ℓ2
F is the attractional force between bodies 1 and 2, m1 and m2 are the
masses of the bodies, and ℓ is the distance between them. We assume the
masses to be points. The constant G, the universal gravitational constant,
has the value
G = 6.674 · 10−11 m3/kg s2 .

2 SirIsaac Newton PRS (1642–1727) was an English universal genius who derived the
mathematical underpinning of astronomy, and much of geophysics, in his major work,
Philosophiæ Naturalis Principia Mathematica (Mathematical Foundations of Physics).

í Õ ! ¤. û
 Gravitation between two masses
3

Figure 1.1. Gravitation is universal. A gravitational lens imaged by the Hubble

Space Telescope, the Abell 1689 cluster of galaxies at a distance of
2.2 billion light years. Benitez et al. (2003).
Credits: NASA, N. Benitez (JHU), T. Broadhurst (The Hebrew Uni-
versity), H. Ford (JHU), M. Clampin (STScI), G. Hartig (STScI), G.
Illingworth (UCO / Lick Observatory), the ACS Science Team and
^ the ESA.

í Õ ! ¤. û
4 Fundamentals of the theory of gravitation

The value of G was determined for the first time by Henry Cavendish3
torsiovaaka, using a sensitive torsion balance (Cavendish, 1798).
kiertoheiluri
Let us call the small body or test mass, for example a satellite, m, and
the large mass, the planet or the Sun, M. Then, m1 = M may be called
the attracting mass, and m2 = m the attracted mass, and we obtain

mM
F=G .
ℓ2
According to Newton’s law of motion

F = ma,

in which a is the gravitational acceleration of the body m. From this

follows
M
a=G 2.
ℓ
From this equation, the quantity m = m2 has vanished. This is the
famous observation by Galileo that all bodies fall equally fast,4 irrespective
of their mass. This is also known as Einstein’s principle of equivalence.5
Both the force F and the acceleration a have the same direction as the
line connecting the bodies. For this reason one often writes equation
1.1 as a vector equation, which is more expressive:

r−R
a = −GM , (1.2)
ℓ3

3 Henry Cavendish FRS (1731–1810) was a British natural scientist from a wealthy, noble

background. He did also pioneering work in chemistry. He was extremely shy, and
the renowned neurologist Oliver Sacks retrodiagnosed him as living with Asperger
syndrome (Sacks, 2001).
4 Atleast in a vacuum. The Apollo astronauts showed impressively how on the Moon
a feather and a hammer fall equally fast! YouTube, Hammer vs. Feather.
5 Albert
Einstein (1879–1955) was a theoretical physicist of Jewish German descent,
who created both the special and general theories of relativity, applying the latter to
cosmology, and did fundamental work in quantum theory.

í Õ ! ¤. û
 The potential of a point mass
5
in which the three-dimensional vectors of place of both the attracting
and attracted masses are defined as follows in rectangular co-ordinates:6

r = xi + yj + zk,
R = Xi + Yj + Zk,
{︁ }︁
where the triad of unit vectors i, j, k is an orthonormal basis7 in Eu- ortonormaali
clidean space and kanta
√︂
ℓ = ∥r − R∥ = (x − X)2 + (y − Y)2 + (z − Z)2 (1.3)

is the distance between the masses computed by the Pythagoras theorem.

Vector equation 1.2 contains a minus sign, which tells us that the
direction of the force is opposite to that of the vector r − R. This vector
is the location of the attracted mass m reckoned from the location of
the attracting mass M. In other words, this tells us that we are dealing
with an attraction and not a repulsion.

^ 1.3 The potential of a point mass

The gravitational field is a special field: if it is stationary, that is, not
time-dependent, it is conservative. This means that a body moving inside
the field along a closed path will, at the end of the journey, not have
lost or gained energy. Because of this, one may attach to each point in
the field a “label” onto which is written the amount of energy gained or
lost by a unit or test mass when travelling from an agreed-upon starting
point to the point under discussion. The value written on the label is
called the potential.

6 As vector notation, one uses either →

−
v (an arrow above the letter) or v (bolding the
letter). Here we use the bold notation, except for vectors designated by Greek letters,
which cannot be bolded.
7 Thismeans that ∥i∥ = ∥j∥ = ∥k∥ = 1 and ⟨i · j⟩ = ⟨i · k⟩ = ⟨j · k⟩ = 0, where the
def √︁
norm of vector a is defined as ∥a∥ = ⟨a · a⟩, and ⟨a · b⟩ is the inner or scalar product
of vectors a, b defined in the space.

í Õ ! ¤. û
6 Fundamentals of the theory of gravitation

Note that the choice of starting point is arbitrary. We shall return to

this significant matter.
The potential function defined in this way for a point-like body M is

V = GM ℓ ,
/︁
(1.4)

where ℓ is again, as above, the length of the vector r − R , ℓ = ∥r − R∥.

The constant GM has in the case of the Earth (according to the GRS80
system, conventionally) the value

GM⊕ = 3.986 005 · 1014 m3/s2 .

The currently best available physical value is slightly more precise:

GM⊕ = 3.986 004 4 · 1014 m3/s2 .

^ 1.4 Potential of a spherical shell

We may write, based on equation 1.4, the potential of an extended body
M into the following form:
w dm(R) w dm(R)
V (r) = G =G . (1.5)
M ℓ M ∥r − R∥

This is an integral over the mass elements dm of the body, where every
such mass element is located at its own place R. The potential V is
evaluated at location r, and the distance ℓ = ∥r − R∥.
We now derive the equation for the potential of a thin spherical shell,
see figure 1.2, where we have placed the centre of the sphere at the
origin O.
Because the circumference of a narrow ring, width b · dθ, is 2πb sin θ,
its surface area is
(2πb sin θ) (b · dθ) .
Let the thickness of the shell be p (small) and its density ρ. We obtain
for the total mass of the ring

2πpρb2 sin θ dθ.

í Õ ! ¤. û
 Potential of a spherical shell
7
b dθ

p b

P
r
O
θ
ℓ
b

^ Figure 1.2. A thin spherical shell consists of rings.

Because every point of the ring is at the same distance ℓ from point P,
we may write for the potential of the ring at point P:
2πGpρb2 sin θ dθ
VP = .
ℓ
With the cosine rule

ℓ2 = r2 + b2 − 2rb cos θ (1.6)

we obtain, using equation 1.5, for the potential of the whole shell
w
sin θ dθ
VP = 2πGρpb2 √ .
r + b2 − 2rb cos θ
2

In order to evaluate this integral, we must replace the integration

variable θ by ℓ. Differentiating equation 1.6 yields

ℓ dℓ = br sin θ dθ,
√
and remembering that ℓ = r2 + b2 − 2rb cos θ we obtain
w ℓ2
2 dℓ
VP = 2πGρpb .
ℓ1 br

í Õ ! ¤. û
8 Fundamentals of the theory of gravitation

In the case that point P is outside the shell, the integration bounds of ℓ
are ℓ1 = r − b and ℓ2 = r + b, and we obtain for the potential of point P
[︂ ]︂ℓ=r+b
ℓ 4πGρpb2
VP = 2πGρpb2 = r .
br ℓ=r−b
Because the mass of the whole shell is Mb = 4πb2 ρp, it follows that the
potential of the shell is the same as that of an equal-sized mass in its centre
O:
GM
VP = r b ,
where r is now the distance of computation point P from the centre of
the sphere O. We see that this is identical to equation 1.4.
In the same way, the attraction, or rather, acceleration, caused by the
spherical shell is8
r −r r −r
aP = ∇V|P = −4πGρpb2 P 3 O = −GMb P 3 O ,
r r
in which r = ∥rP − rO ∥ . This result is identical to the acceleration caused
by an equal-sized point mass located in point O, equation 1.2.
In the case that point P is inside the shell, ℓ1 = b − r and ℓ2 = b + r
and the above integral changes to the following:
[︂ ]︂ℓ=b+r
2 ℓ
VP = 2πGρpb = 4πGρpb.
br ℓ=b−r
As we see, this is a constant and not dependent upon the location of
point P. Therefore ∇VP = 0 and the attraction, being the gradient of
the potential, vanishes.
The end result is that the attraction of a spherical shell is, outside the
shell,
GM
a = ∥a∥ = 2 ,
r
where M is the total mass of the shell and r = ∥rP − rO ∥ the distance of
the observation point from the shell’s centre; and 0 inside the shell.
In figure 1.3 we have drawn the curves of potential and attraction
— or rather, acceleration, attractive force per unit of mass. If a body

8 Here, the ∇ (nabla) operator is used, to be explained in section 1.5.

í Õ ! ¤. û
 Computing the attraction from the potential
9

Acceleration
b2
4πGρp 2
r
4πGρpb

b
4πGρpb
r
Potential
0
0 b →r

Figure 1.3. Dependence of potential and attraction on distance r from the

^ centre of a spherical shell.

consists of many concentric spherical shells (like rather precisely the

Earth and many celestial bodies), then inside the body only those layers
of mass “internal” to — i. e., closer to the centre than — the observation
point participate in causing attraction, and this attraction is the same as
it would be if the mass of all these layers was concentrated in the centre
of the body. This case, where the distribution of mass density inside a
body only depends on the distance from its centre, is called an isotropic
density distribution.

^ 1.5 Computing the attraction from the potential

As we argued earlier, the potential V is a path integral. Conversely we
can compute, from the potential, the components of the gravitational
acceleration vector by differentiating V(x, y, z) with respect to place, by
applying the gradient operator, which is a vector operator:

∂V ∂V ∂V
a = ∇V = grad V = i+ j+ k. (1.7)
∂x ∂y ∂z

í Õ ! ¤. û
10 Fundamentals of the theory of gravitation

Here, the symbol ∇ (nabla), is a frequently used partial differential operator

∂ ∂ ∂
∇=i +j +k .
∂x ∂y ∂z

kanta Here, {i, j, k} is again a basis of mutually orthogonal unit vectors in

Euclidean space parallel to the (x, y, z) axes.
Let us try this differentiation in the case of the potential field of the
point mass M. Substitute the above equations for V 1.4 and ℓ 1.3:9

∂V ∂V ∂ℓ 1 x−X x−X
= = GM · − 2 · = −GM 3 .
∂x ∂ℓ ∂x ℓ ℓ ℓ
Similarly we compute the y and z components:

∂V y−Y ∂V z−Z
= −GM 3 , = −GM 3 .
∂y ℓ ∂z ℓ
These are the components of gravitational acceleration when the source
of the field is one point mass M. So, in this concrete case, the vector
equation given above applies:

a = grad V = ∇V.

Remark In physical geodesy — unlike in physics — the potential is

reckoned to be always positive if the attracting mass M is positive
(as it is known to always be). However, the potential energy of

9 From the equation

√︂ (︂ )︂ 1/2
ℓ= (x − X)2 + (y − Y)2 + (z − Z)2 = (x − X)2 + (y − Y)2 + (z − Z)2

it follows with the chain rule that

(︂ )︂ 1/2
∂ℓ ∂ (x − X)2 + (y − Y)2 + (z − Z)2 ∂ (x − X)2
= (︂ )︂ · =
∂x ∂ (x − X)2 + (y − Y)2 + (z − Z)2 ∂x
(︂ )︂− 1/2 x−X
= 12 (x − X)2 + (y − Y)2 + (z − Z)2 · 2 (x − X) = .
ℓ

í Õ ! ¤. û
 Potential of a solid body
11
body m inside the field V of mass M is negative! More precisely,
the potential energy of body m is

Epot = −Vm.

One calls the vector of gravitational acceleration in brief the “gravita-

tional vector”.

^ 1.6 Potential of a solid body

In the following we study a solid body, the mass of which is distributed
throughout space and thus not concentrated in a single point. For
example, the mass distribution of the Earth in space may be described
by a matter density function ρ:

dm(x, y, z)
ρ(x, y, z) = ,
dV(x, y, z)

in which dm is a mass element and dV is the corresponding element of

spatial volume. The dimension of ρ is density, its unit in the SI system,
kg/m3 .

Because the gravitational acceleration 1.7 is a linear expression in

the potential V, and both the force and acceleration vectors may be
summed linearly, it follows that the total potential of the body can also
be obtained by summing together the potentials of all its parts. For
example, the potential of a set of n mass points is

∑︂
n
m ∑︂
n
mi (Ri )
i
V(r) = G =G ,
ℓi ∥r − Ri ∥
i=1 i=1

from which we obtain the gravitational acceleration by simply using

gradient theorem 1.7.
The potential of a solid body is obtained similarly by replacing the

í Õ ! ¤. û
12 Fundamentals of the theory of gravitation

sum by an integral, in the following way:10

y y
dm ρ
V=G =G dV. (1.8)
body ℓ body ℓ

The symbol ρ inside the integral designates the matter’s den-

sity
√︂ at the location of dV. The quantity ℓ = ∥r − R∥ =
(x − X)2 + (y − Y)2 + (z − Z)2 is the distance between the point of
calculating the potential and the attracting mass element. More clearly:
y ρ(X, Y, Z)
V(x, y, z) = G √︂ dX dY dZ.
body 2 2 2
(x − X) + (y − Y) + (z − Z)

As we showed already above for mass points, the first derivative with
respect to place or gradient of the potential V of a solid body,

grad V = ∇V = a, (1.9)

is also the acceleration vector caused by the attraction of the body. This
applies generally.

^ 1.6.1 Behaviour at infinity

If a body is of finite extent — in other words, it lies completely within a
sphere of size ϵ around the origin — and its density is also bounded
everywhere, it follows that

∥r∥ → ∞ =⇒ V(r) → 0,

because, according to the triangle inequality,

ℓ = ∥r − R∥ ⩾ ∥r∥ − ∥R∥ ⩾ ∥r∥ − ϵ

and thus
∥r∥ → ∞ =⇒ 1 ℓ → 0.
/︁

10 Unfortunatelyalmost the same symbols V and V are used here for the potential and
volume, respectively.

í Õ ! ¤. û
Potential of a solid body
13
For the acceleration of gravitation, the same applies for all three compo-
nents, and thus also for the length of this vectorial quantity:

∥r∥ → ∞ =⇒ ∥∇V∥ → 0.

This result can still be sharpened: if ∥r∥ → ∞, then, again by the triangle
inequality,
ℓ = ∥r − R∥ ⩽ ∥r∥ + ∥R∥ ⩽ ∥r∥ + ϵ,
and thus
1 1 1 1 1/︁ 1 1 1/︁
⩽ ⩽ =⇒ ⩽ ⩽ .
∥r∥ + ϵ ℓ ∥r∥ − ϵ ∥r∥ 1 + ϵ ∥r∥ ℓ ∥r∥ 1 − ϵ ∥r∥

It is seen, again with the notation r = ∥r∥, that

r → ∞ =⇒ 1 ℓ → 1 r .
/︁ /︁

When we substitute this into integral 1.8, it follows that for large
distances r → ∞:
y y
ρ G GM
V=G dV ≈ r ρ dV = r ,
body ℓ body

in which M, the integral of density over the volume of the body, is

precisely its total mass. From this we see that, at great distance, the field
of a finite-sized body M is nearly identical with the field generated by
a point mass the mass of which is equal to the total mass of the body.
This important observation was already made by Newton. Thanks to
this phenomenon, we may treat, in the celestial mechanics of the solar
system, the Sun and planets11 as mass points, although we know that
they are not.

11 The
only important exception is formed by the forces between a planet and its
moons, due both to the flattening of the planet and tidal effects.

í Õ ! ¤. û
14 Fundamentals of the theory of gravitation

^ 1.7 Example: The potential of a line of mass

The potential of a vertical line of mass having a linear mass density of
unity is
wH
1
V(x, y, z) = G √︂ dZ, (1.10)
0 2 2 2
(X − x) + (Y − y) + (Z − z)

in which (X, Y) is the location in the plane of the mass line, (x, y, z) is
the location of the point at which the potential is being evaluated, and
the mass line extends from sea level Z = 0 to height Z = H.
Firstly we write ∆x = X − x, ∆y = Y − y, ∆z = Z − z, and the potential
becomes
w H−z
1
V(∆x, ∆y, ∆z) = G √︁ d(∆z).
−z ∆x + ∆y2 + ∆z2
2

The indefinite integral is

(︂ √︁ )︂
2 2
ln ∆z + ∆x + ∆y + ∆z 2

and substituting the integration bounds yields

√︂
H − z + ∆x2 + ∆y2 + (H − z)2
V = G ln √︁ .
−z + ∆x2 + ∆y2 + z2
Now we can expand this into a Taylor series in H around the point
H = 0: the first derivative of equation 1.10 is
∂V G G
= √︂ =
∂H 2 2 2 ℓ
(X − x) + (Y − y) + (H − z)
√︂
in which ℓ(H) = (X − x)2 + (Y − y)2 + (H − z)2 . The second deriva-
tive is, using the chain rule,

∂2 V ∂ G ∂ℓ−1 ∂ℓ 1
(︂ )︂
= = G · · = G · −ℓ−2 · ℓ−1 · 2 (H − z) =
∂H 2 ∂H ℓ ∂ℓ ∂H 2
H−z
= −G 3 .
ℓ

í Õ ! ¤. û
Example: The potential of a line of mass
15
The third derivative, obtained in the same way:
(︄ )︄
2
3 (H − z)2 − ℓ2
(︃ )︃
3 3 (H − z)
∂V ∂ H−z 1
= −G = G − = G ,
∂H3 ∂H ℓ3 ℓ5 ℓ3 ℓ5

and so on. The Taylor expansion is

⃓
⃓
∂H V|H=0 ∂2H V ⃓ ∂3H V ⃓H=0
V|H=0 ⏟⏞⏞⏟ ⏟⏞⏞⏟H=0 ⏟ ⏞⏞ ⏟
⏞ ⏟⏟ ⏞ 1 z 3z2 − ℓ20 3
V = 0 + G H + 21 G 3 H2 + 16 G H + ··· (1.11)
ℓ0 ℓ0 ℓ50
√︂
in which ℓ0 = (X − x)2 + (Y − y)2 + z2 , so the values of the derivatives
used in this expansion are obtained by substituting H = 0.
Question How can we exploit this result for computing the gravita-
tional potential of a complete, realistic topography?
Answer In this expansion, the coefficients 1 ℓ0 , 12 z ℓ30 , . . . , like ℓ0 ,
/︁ /︁

depend only on the differences ∆x = X − x and ∆y = Y − y

between the co-ordinates of the location of the mass line (X, Y)
and those of the evaluation location (x, y) — and of the height z
of the evaluation point. If the topography is given in the form of
a grid, we may evaluate the above expansion 1.11 term-wise for
the given z value and for all possible value pairs (∆x, ∆y).
Then, if the grid is, say, N × N in size, we need only N2 operations
to calculate every coefficient. The brute-force evaluation of the
Taylor expansion itself for the whole topography, for every point of
the terrain grid and every point of the evaluation grid, requires
after that N2 · N2 = N4 operations, but those are much simpler:
the coefficients themselves have been precomputed. And brute
force is not even the best approach: as we shall see, the above
convolution can be computed much faster using the Fast Fourier
Transform (FFT).
We shall return to this subject more extensively in connection
with terrain-effect evaluation, in sections 6.3 and 9.7.

í Õ ! ¤. û
16 Fundamentals of the theory of gravitation

^ 1.8 Laplace and Poisson equations

The second derivative with respect to the place of the geopotential, the
first derivative with respect to the place of the gravitational acceleration
vector called its divergence, is also of geophysical interest. We may write:

def
div a = ⟨∇ · a⟩ = ⟨∇ · (∇V)⟩ = ⟨∇ · ∇⟩ V =
∂2 ∂2 ∂2
= ∆V = 2
V + 2 V + 2 V, (1.12)
∂x ∂y ∂z
in which
def ∂2 ∂2 ∂2
∆ = ⟨∇ · ∇⟩ = 2
+ 2+ 2
∂x ∂y ∂z
is a well-known symbol called the Laplace operator.12
In equation 1.4 for the potential of a point mass we may show, by
performing all partial derivations 1.12, that

∆V = 0, (1.13)

the well-known Laplace equation. This equation applies outside a point

mass, and more generally everywhere in empty space: all masses can
in the limit be considered to consist of point-like mass elements. Or,
in equation 1.8 we may directly differentiate inside the triple integral
sign, exploiting the circumstance that it is permitted to interchange
integration and partial differentiation, if both exist.
A potential field for which Laplace equation 1.13 applies, is called a
harmonic field or function.
In the case where the mass density does not vanish everywhere, we
have a different equation, with ρ the mass density:

∆V = −4πGρ. (1.14)

12 Pierre-Simon marquis de Laplace (1749–1827) was a French universal genius who

contributed to mathematics and the natural sciences. He is one of the 72 French
scientists, engineers and mathematicians whose names were inscribed on the Eiffel
Tower, Eiffel Tower, 72 names.

í Õ ! ¤. û
 Gauge invariance
17
This equation is called the Poisson equation.13
The pair of equations

grad V = a, div a = −4πGρ

are known as the field equations of the gravitational field. They play
the same role as Maxwell’s14 field equations in electromagnetism.
Unlike in Maxwell’s equations, however, in the above there is no time
co-ordinate. Because of this, it is not possible to derive an equation
describing the propagation in space of gravitational waves, like the one
for electromagnetic waves in Maxwell’s theory.
We know today that these “Newton field equations” are only approx-
imate, and that Einstein’s general theory of relativity is a more precise
theory. Nevertheless, in physical geodesy Newton’s theory is generally
precise enough and we shall use it exclusively.

^ 1.9 Gauge invariance

An important property of the potential is that if we add a constant C to
it, nothing related to gravitation that can actually be measured, changes.
This is called gauge invariance. mitta-
invarianssi
Gravitation itself is obtained by differentiation of the potential, an
operation that eliminates the constant term. Therefore the definition
of potential is ambiguous: all potential fields V obtained by different
choices of C are equally valid.
Observations only give us potential differences, as spirit levellers know
all too well.

13 Siméon Denis Poisson (1781–1840) was a French mathematician, physicist and

geodesist, one of the 72 names inscribed on the Eiffel Tower, Eiffel Tower, 72 names.
14 James Clerk Maxwell FRS FRSE (1831–1879) was a Scottish physicist, the discoverer
of the field equations of electromagnetism. He found a wave-like solution to the
equations, and, based on propagation speed, identified light as such.

í Õ ! ¤. û
18 Fundamentals of the theory of gravitation

An often-chosen definition of potential is based on requiring that, if

def
r = ∥r∥ → ∞, then also V → 0, which makes physical sense and yields
simple equations. However, in work on the Earth’s surface, a more
practical alternative may be V = 0 at the mean sea surface — although
that is not free of problems, either.
For example, for the mass of the Earth M⊕ a physically sensible form
of the potential is, in spherical approximation,
GM⊕
V= r ,
which vanishes at infinity r → ∞, when again a geodetically sensible
definition would be
GM GM⊕
V= r⊕− ,
R
in which R = ∥R∥ is the radius of the Earth sphere. The latter potential
vanishes where r = R, on the surface of a spherical Earth. In the limit
r → ∞ its value is − GM⊕ R , not zero.
/︁

^ 1.10 Single mass-density layer

If we apply to the surface S of a body a mass “coating”, of mass-density
value
dm
κ= ,
dS
we obtain for the potential an integral equation otherwise similar to
equation 1.8, but a surface integral:
x x
dm κ
V=G =G dS. (1.15)
surface ℓ surface ℓ

Here again ℓ is the distance between the point at which the potential
is to be calculated, and the moving mass element in integration dm —
or surface element dS. The dimension of the mass surface density κ is
kg/m2 , different from the dimension of ordinary or volume mass density,

which is kg/m3 .
This case is theoretically interesting, although physically unrealistic.
The function V is everywhere continuous, also at the surface S, however

í Õ ! ¤. û
 Double mass-density layer
19
its first derivatives with place are already discontinuous. The discon-
tinuity appears in the direction perpendicular to the surface, in the
normal derivative.
Let us look at the simple case where a sphere of radius R has been
coated with a layer of constant surface density κ. By computing the
above integral 1.15 we may prove — in a complicated way, see section
1.4 — that the exterior potential is the same as it would be if all of the
mass of the body were concentrated in the sphere’s centre. Also in
section 1.4 we proved that the potential interior to the sphere is constant.
Thus, the exterior attraction (r > R) , with r the distance of the point
of computation from the centre of the sphere, is
(︂ )︂2
M κ · 4πR2 R
aext (r) = G 2 = G = 4πGκ r .
r r2
The interior attraction (r < R) is
aint (r) = 0.
This means that on the surface of the sphere, ℓ = R, the attraction is
discontinuous:
aext (R) − aint (R) = 4πGκ.
In this symmetric case we see that
∂V
a = ∥a∥ = , (1.16)
∂n
in which the differentiation variable n represents the normal direction,
the direction perpendicular to the surface S. If the surface S is an
equipotential surface of the potential V, equation 1.16 applies generally.
Then, the attraction vector — more precisely, the acceleration vector —
is perpendicular to the surface S, and its magnitude is equal to that of
the normal derivative of the potential.

^ 1.11 Double mass-density layer

A double mass-density layer may be interpreted as a dipole-density layer.
The dipoles are oriented in the direction of the surface’s normal.

í Õ ! ¤. û
20 Fundamentals of the theory of gravitation

P n
ℓ
κ
δ
−κ

^ Figure 1.4. A double mass-density layer.

If the dipole consists of two “charges” m and −m in locations r1

and r2 , in such a way that the vectorial separation between them is
∆r = r1 − r2 , then the dipole moment is d = m∆r, a vectorial quantity.
See figure 1.4.
Let the dipole surface density be
dD
µ= ,
dS
in which dD is a “dipole-layer element”. This layer may be seen as
being made up of two single layers. If we have a positive layer at density
κ and a negative layer at density −κ, and the distance between them is
δ, we get for small values of δ an approximate correspondence:

µ ≈ δκ. (1.17)

The combined potential of the two single mass-density layers computed

as explained in the previous section, equation 1.15, is
x
1 1
(︂ )︂
V=G κ − dS.
surface ℓ1 ℓ2

í Õ ! ¤. û
Double mass-density layer
21
The following relationship exists between the quantities ℓ1 , ℓ2 and δ
(Taylor expansion of the function 1 ℓ ):
/︁

1 1 ∂ 1
(︂ )︂
= +δ· + ··· ,
ℓ1 ℓ2 ∂n ℓ
∂
in which ∂n is again the derivative of the quantity in the normal direction
of the surface.
Substitution into the equation yields
x x
∂ 1 ∂ 1
(︂ )︂ (︂ )︂
V≈G κδ dS = G µ dS. (1.18)
surface ∂n ℓ surface ∂n ℓ

In the limit in which δ is arbitrarily small and κ correspondingly large,

this equation, like equation 1.17, holds exactly.
One can easily show that the above potential is not even continuous.
The discontinuity happens at the surface S. Let us look again, for the
sake of simplicity, at a sphere of radius R, coated with a double layer of
constant dipole density µ.
The exterior potential (r > R, r the distance from the centre of the
sphere) is
x
∂ 1
(︂ )︂
Vext = Gµ dS = 0,
surface ∂n ℓ
because the potential equals the difference between the potentials of
two concentric spherical shells of the same mass.
The interior potential (r < R) is
x
∂ 1 1
(︂ )︂ (︂ )︂
Vint = Gµ dS = Gµ · 4πR2 − 2 = −4πGµ,
surface ∂n ℓ R
by computing the surface integral using the sphere’s centre as the
evaluation point, and using the earlier established circumstance that
inside a sphere covered by a single constant-density mass-density layer
the potential is constant.
Now in the limit r → R the result is different for the exterior and
interior potentials. The difference is

Vext (R) − Vint (R) = 4πGµ.

í Õ ! ¤. û
22 Fundamentals of the theory of gravitation

^ 1.12 The Gauss integral theorem

^ 1.12.1 Presentation
The Gauss integral theorem,15 famous from physics, looks in vector
form like this: y x
div a dV = ⟨a · n⟩ dS, (1.19)
V ∂V
in which n is the exterior normal to surface S, now as a vector: the
length of the vector is assumed ∥n∥ = 1. ∂V is the total surface of body
V.
This theorem applies to all differentiable vector fields a and all
“well-behaved” bodies V on whose surface ∂V, a normal direction n
exists everywhere. In other words, this is not a special property of the
gravitational acceleration vector, though it also applies to this vector
field.

^ 1.12.2 Intuitive description

Let us note, that16
div a = ∆V = −4πGρ
is a source function. It stands for the amount, in the part of space
lähteet, nielut inside surface ∂V, of positive and negative “sources” and “sinks” of the
gravitational field.
The situation is analogous to the flow pattern of a liquid: positive
charges correspond to points where liquid is added to the flow, negative
charges17 correspond to “sinks” through which liquid disappears.
The vector a is in this metaphor the velocity of flow; in the absence
of “sources” and “sinks” it satisfies the condition div a = 0, which
expresses the conservation and incompressibility of matter.

15 Johann Carl Friedrich Gauss (1777–1855) was a German mathematician and universal

genius. Princeps mathematicorum.

16 Assuming that the potential V exists.
17 But the ”charges” for gravitation, the masses, are always positive.

í Õ ! ¤. û
 The Gauss integral theorem
23
Field line, field a

Normal n

Body
Sources surface

Figure 1.5. A graphical explanation of the Gauss integral theorem. The

concept of the field line was Michael Faraday’s invention. The flux
^ is the scalar product ⟨a · n⟩.

On the other hand, the function

∂V
⟨a · n⟩ =
∂n
is often called the flux, in other words, how much field stuff “flows out” vuo
— just like a liquid flow — from the space inside the surface ∂V through
the surface to the outside.
The Gauss integral theorem states the two amounts to be equal: it is
in a way a book-keeping statement demanding that everything which is
produced inside a surface — div a — also has to come out through the
surface — ⟨a · n⟩.
In figure 1.5 it is graphically explained that the sum of “sources” over
∑︁
the inner space of the body, (+ + + . . .), has to be the same as the sum
∑︁
of “flux” (↑↑↑ . . .) through the whole boundary surface delimiting

í Õ ! ¤. û
24 Fundamentals of the theory of gravitation

this inner space.

^ 1.12.3 The potential version of the Gauss theorem

Let us write the Gauss integral theorem a little differently, using the
potential V instead of the gravitational vector:
y x
∂V
∆V dV = dS, (1.20)
V ∂V ∂n

in which we have made the above substitutions. We also here see

the popular notation ∂V for the surface of the body V. The presenta-
tional forms 1.20 and 1.19 are connected by the equations 1.12 and 1.9,
connecting V and a.

^ 1.12.4 Example 1: A little box

Let us look at a little rectangular box with sides ∆x, ∆y and ∆z. The box
is so little that the field a(x, y, z) inside it is an almost linear function of
place. Let us write the vector a into components:

a = a1 i + a2 j + a3 k.

Now the volume integral

y (︃ )︃
∂a1 ∂a2 ∂a3
div a dV = + + ∆x ∆y ∆z (1.21)
V ∂x ∂y ∂z

while the surface integral

x
⟨a · n⟩ dS =
∂V

= (a+ − + − + −
1 − a1 ) ∆y ∆z + (a2 − a2 ) ∆x ∆z + (a3 − a3 ) ∆x ∆y.

Here, a+1 is the value of component a1 on the one face in the x direction,
and a1 its value on the other face, and so on. For example, a+
−
3 is the
−
value of a3 on the box’s upper and a3 on its lower face. A box has of
course six faces, in each of three co-ordinate directions both “up- and
downstream”.

í Õ ! ¤. û
 The Gauss integral theorem
25
a

a+
3

+
a2
a+
∆z 1

a−
1
−
a2

a−
3
∆y
∆x

^ Figure 1.6. A little rectangular box.

Then
∂a1 ∂a2 ∂a3
a+ −
1 − a1 ≈ ∆x, a+ −
2 − a2 ≈ ∆y, a+ −
3 − a3 ≈ ∆z,
∂x ∂y ∂z

and by substitution we see that

x
⟨a · n⟩ dS ≈
∂V
∂a1 ∂a ∂a
≈ ∆x · ∆y ∆z + 2 ∆y · ∆x ∆z + 3 ∆z · ∆x ∆y =
∂x ∂y ∂z
(︃ )︃
∂a1 ∂a2 ∂a3
= + + ∆x ∆y ∆z,
∂x ∂y ∂z

the same equation as 1.21. So, in this simple case, the Gauss equation
applies.
Obviously the equation also works, if we build out of these “Lego™
bricks” a larger body, because the faces of the bricks touching each
other are oppositely oriented and cancel from the surface integral of the
whole body. It is a little harder to prove that the equation also applies
to bodies having inclined surfaces.

í Õ ! ¤. û
26 Fundamentals of the theory of gravitation

^ 1.12.5 Example 2: The Poisson equation for a sphere

According to Poisson equation 1.14 we have

∆V = −4πGρ.

Assume a sphere of radius R, within which the mass density ρ is constant.

The volume integral over the sphere gives
y y
∆V dV = −4πGρ dV = −4πGρV = −4πGM, (1.22)
V V

in which M = ρV is the total mass of the sphere.

On the surface of the sphere, the normal derivative is
∂V ∂ GM ⃓ GM
⃓
= =− 2 ,
∂n ∂r r r=R R
⃓

a constant, and its integral over the surface of the sphere is

x
∂V GM GM
dS = − 2 · S = − 2 · 4πR2 = −4πGM. (1.23)
∂V ∂n R R
Results 1.23 and 1.22 are identical, as Gauss theorem 1.20 requires.

^ 1.12.6 Example 3: A point mass in an eight-unit cube

See figure 1.7.
Let us assume that we have a point mass of size GM in the centre of
a cube bounded by the co-ordinate planes x = ±1, y = ±1, and z = ±1.
In that case, the volume integral is
y y
∆V dV = −4πGM δ(r) dV = −4πGM,
V V

in which δ(r) is Dirac’s18 delta function in space, having an infinite

spike at the origin, being zero elsewhere, and producing a value of 1

18 PaulAdrien Maurice Dirac (1902–1984) was an English quantum physicist who

found the relativistic wave equation for the electron, and theoretically predicted the
existence of antimatter. He was a physics Nobel laureate 1933, shared with Erwin
Schrödinger. He is also believed to have been on the autism spectrum (Farmelo, 2011).

í Õ ! ¤. û
 The Gauss integral theorem
27
z

+1
z

0
GM(0, 0, 0)

x
−1 y

0 −1
0
y +1
x

^ Figure 1.7. Eight-unit cube.

upon volume integration. The surface integral is six times that over the
top face
w +1 w +1
(︄ )︄
1
−GM dx dy.
(x2 + y2 + 1) /2
−1 −1 3

Integrating with respect to x (expression in large parentheses) yields

]︄+1
w +1
[︄
1 x
dx = √︁ =
(x2 + y2 + 1) /2
3
−1 (y2 + 1) x2 + y2 + 1 −1
2
= √︁ .
(y + 1) y2 + 2
2

Integrating this with respect to y yields

]︄+1
w +1
[︄
2 y
√︁ dy = 2 arctan √︁ =
−1 (y2 + 1) y2 + 2 y2 + 2 −1
1 π
= 4 arctan √ = 4 · = 23 π.
3 6

í Õ ! ¤. û
28 Fundamentals of the theory of gravitation

Adding the six faces together yields

w +1 w +1
(︄ )︄
1
− 6 · GM 3/2
dx dy = −6 · GM · 23 π =
−1 −1
(x2 + y2 + 1)
= −4πGM,

agreeing with the volume result above.

^ 1.13 Green’s theorems

Apply the Gauss integral theorem to the vector field

F = U∇V.

Here, U and V are two different scalar fields. We obtain

y y
div F dV = ⟨∇ · (U∇V)⟩ dV =
V y V y
= U ⟨∇ · ∇⟩ V dV + ⟨∇U · ∇V⟩ dV =
V V
y y (︃ )︃
∂U ∂V ∂U ∂V ∂U ∂V
= U∆V dV + + + dV
V V ∂x ∂x ∂y ∂y ∂z ∂z

and
x x x
⟨F · n⟩ dS = ⟨U∇V · n⟩ dS = U ⟨∇V · n⟩ dS =
∂V ∂V ∂V
x
∂V
= U dS.
∂V ∂n
The result is Green’s19 first theorem:
y y (︃ )︃
∂U ∂V ∂U ∂V ∂U ∂V
U∆V dV + + + dV =
V V ∂x ∂x ∂y ∂y ∂z ∂z
x
∂V
= U dS.
∂V ∂n
19 George Green (1793–1841) was a British mathematical physicist, an autodidact,
working as a miller near Nottingham. He also invented the word “potential”. Green
(1828); O’Connor and Robertson (1998); Green’s Windmill.

í Õ ! ¤. û
Green’s theorems
29
This may be cleaned up, because the second term on the left is symmetric
for the interchange of the scalar fields U and V. Let us therefore
interchange U and V, and subtract the equations obtained from each
other. The result is Green’s second theorem:
y x (︂
∂V ∂U
)︂
(U∆V − V∆U) dV = U −V dS.
V ∂V ∂n ∂n
We assume in all operations that the functions U and V are “well-
behaved”: for example, all required derivatives exist everywhere in
body V.
A useful special case arises by choosing for the function U:

U= 1 ℓ,
/︁

in which ℓ is the distance from the given point of evaluation P. This

function U is well-behaved everywhere except precisely at point P, where
it is not defined.
In the case when point P is outside the surface ∂V, the result, Green’s
third theorem, is now obtained by substitution:
y x (︃ (︂ )︂)︃
1 1 ∂V ∂ 1
∆V dV = −V dS.
V ℓ ∂V ℓ ∂n ∂n ℓ
This case is depicted in figure 1.8.
In the case that point P is inside surface ∂V, the computation becomes
a little more complicated. Let us learn about a clever technique that —
in this case as in others — comes to the rescue.
We form a small sphere of radius ϵ called V2 around point P; now we
def
can formally define the body (containing a hole) V = V1 − V2 , and also
its surface ∂V which consists of two parts, ∂V = ∂V1 − ∂V2 .
Now we may write the volume integral into two parts:
y y y
1 1 1
∆V dV = ∆V dV − ∆V dV,
V ℓ V1 ℓ V2 ℓ

where the second term can be integrated in spherical co-ordinates:

y wϵ
1 1
∆V dV ≈ ∆VP 4πℓ2 dℓ = 2π∆VP ϵ2 ,
V2 ℓ 0 ℓ

í Õ ! ¤. û
30 Fundamentals of the theory of gravitation

Distance ℓ
Surface element dS

Surface
normal
n
Body V

Surface S = ∂V

Figure 1.8. Geometry for deriving Green’s third theorem if point P is outside
^ surface ∂V.

which will go to zero in the limit ϵ → 0.

For the first surface integral we obtain using Gauss integral theorem
1.20 :
x x y
1 ∂V 1 ∂V 1 1
dS = ϵ dS = ϵ ∆V dV ≈ ϵ ∆VP · 34 πϵ3 ,
∂V2 ℓ ∂n ∂V2 ∂n V2

which also goes to zero for ϵ → 0.

The second surface integral (the normal is pointing away from P):
x x
∂ 1 1 1
(︂ )︂
− V dS = − V · − 2 dS ≈ 4πϵ2 · 2 VP = 4πVP .
∂V2 ∂n ℓ ∂V2 ϵ ϵ
By combining all results with their correct algebraic signs we obtain for
the case where P is inside surface ∂V1 ∼ ∂V:
y x (︃ (︂ )︂)︃
1 1 ∂V ∂ 1
∆V dV = −4πVP + −V dS. (1.24)
V ℓ ∂V ℓ ∂n ∂n ℓ

After this it must be intuitively clear, and we present without formal

proof, that
y x (︃ (︂ )︂)︃
1 1 ∂V ∂ 1
∆V dV = −2πVP + −V dS
V ℓ ∂V ℓ ∂n ∂n ℓ

í Õ ! ¤. û
 Green’s theorems
31
Surface ∂V, part 1
Surface ∂V, part 2
Point P Volume V

Figure 1.9. Geometry for deriving Green’s third theorem if point P is inside
^ surface ∂V. .

if point P is precisely on the boundary surface ∂V of body V. This

however presupposes that the normal derivative, and especially the
normal direction, actually exist at precisely point P!
In geodesy, the typical situation is that where the body V over the
volume of which one wants to evaluate the volume integral, is the whole
space outside the Earth. In this case, we conveniently have ∆V = 0 and
the whole volume integral appearing above vanishes.
Result 1.24 may be generalised to this case, where V is the whole space
outside surface ∂V. This generalisation is made by now choosing as the
surface ∂V the three-part surface ∂V = ∂V1 + ∂V2 + ∂V3 , in which ∂V3 is
a sphere of large radius centred on P. Its radius is then allowed to grow
in the limit to infinity, and it can be shown that both integrals over the
surface ∂V3 vanish. The volume integral over the part of space outside
this surface also vanishes, being over empty space where ∆V = 0.
The end result is — note that n is now the exterior normal of the
Earth’s surface:
y x (︃ (︂ )︂)︃
1 1 ∂V ∂ 1
∆V dV = −4πVP − −V dS, (1.25)
V ℓ ∂V ℓ ∂n ∂n ℓ
Because in this case, in which V is the empty space exterior to the
Earth, the left-hand side volume integral vanishes, we may express
the potential VP at point P suitably as a two-term surface integral over
surface ∂V.

í Õ ! ¤. û
32 Fundamentals of the theory of gravitation

Point P
Integration space V

Boundary ∂V, part 2

Matter

Boundary ∂V, part 1

Boundary ∂V, part 3

(Limit)

^ Figure 1.10. Green’s third theorem for the space external to a body.

^ 1.14 The Chasles theorem

We study the above-described case where the “body” is the space
outside the surface ∂V — in practice: the space outside the Earth.
From Green’s theorem 1.25 derived above, we may derive for a
harmonic function V (so, ∆V = 0) in the exterior space:
x x
1 1 ∂ 1 ∂ 1
(︂ )︂
VP = − V dS + V dS. (1.26)
4π ∂V ℓ ∂n 4π ∂V ∂n ℓ
Interpretation The exterior, harmonic potential of an arbitrarily shaped
body can be represented as the sum of a single and a double
mass-density layer on the body’s surface.
Explanation We obtain the surface density of a single mass layer using
equation 1.15,
1 ∂
κ=− V, (1.27)
4πG ∂n

í Õ ! ¤. û
The Chasles theorem
33
and similarly the surface density of a double mass layer using
equation 1.18,
V
µ= .
4πG
If we plug these into equation 1.26, we obtain
x (︃ (︂ )︂)︃
κ ∂ 1
VP = G +µ dS.
∂V ℓ ∂n ℓ

In the case that the surface ∂V is an equipotential surface of the potential

V, so V = V0 , it follows that a single mass-density layer suffices, because
in that case
x (︃ )︃ x
∂ 1 ∂ 1
(︂ )︂
V dS = V0 dS = 0.
∂V ∂n ℓ ∂V ∂n ℓ

The right-hand side integral vanishes based on the Gauss integral

theorem. This is because the function 1 ℓ , with ℓ the distance from
/︁

point P, is harmonic inside the Earth’s body. The surface of the Earth is ∂V.
This is the Chasles20 theorem, also called the Green equivalent-layer
theorem.
The theorem is used in Molodensky’s21 theory. The representation of
the Earth’s gravity field by underground point-mass layers, for example
Vermeer (1984), could also be justified with this theorem.
The case where ∂V is an equipotential surface is realised if the body
is fluid and seeks by itself an external form equal to an equipotential
surface. For planet Earth, this applies for the ocean surface. In electro-
static theory, for a conductor in which the electrons can move freely, the
physical surface will also become an equipotential surface. And the

20 MichelChasles (1793–1880) was a French mathematician and geometrician, one of

the 72 whose names are inscribed on the Eiffel Tower, Eiffel Tower, 72 names.
21 Mikhail Sergeevich Molodensky (1909–1991) was an illustrious Russian physical
geodesist.

í Õ ! ¤. û
34 Fundamentals of the theory of gravitation

electric charges of a conductor are always on its outer surface.22

Equation 1.26, with substitution 1.27, simplifies in this case as follows:
x x
1 1 ∂ κ
VP = − V dS = G dS. (1.28)
4π ∂V ℓ ∂n ∂V ℓ

The equation tells us that we can compute the whole potential exterior
to the Earth, if only on the surface of the Earth — the shape of which
we also assume given in order to compute 1 ℓ ! — is given the gradient
/︁
∂
of the potential in the normal or vertical direction ∂n V. This gradient
is precisely the gravitational acceleration, a quantity obtainable from
gravimetric measurements. All of gravimetric geopotential determina-
tion (“geoid determination”), ever since G. G. Stokes, has been based on
this.

^ 1.15 Boundary-value problems

reuna-arvo- The boundary-value problem (BVP) is the problem of computing the
tehtävä potential V throughout space (or throughout a body’s exterior or interior
part of space) from given values relating to V on the boundary surface
of the body, for example on the surface of the Earth. The simplest
boundary-value problem is Dirichlet’s23 problem: the potential V itself
is given on the boundary surface. More complicated boundary-value
problems are based on linear functionals of the potential: on the boundary,
some linear expression in V is given, for example a derivative or linear
combination of derivatives, generally
{︁ }︁
L V ,
{︁ }︁
with L · being a linear operator.

22 This
is because the electrostatic potential inside a conductor must also be constant.
Even a single extra electron inside the body would make this impossible.
23 Peter
Gustav Lejeune Dirichlet (1805–1859) was a German mathematician also
known for his contributions to number theory.

í Õ ! ¤. û
Boundary-value problems
35
The Dirichlet boundary-value problem in the form popular in geodesy is:
determine the potential field V if its values are given on a closed surface
S, and furthermore it is given that V is harmonic (∆V = 0) outside
surface S. In the vacuum of space, the potential is always harmonic, as
already earlier noted: the potential of a point mass mP , V = GmP ℓ , is
/︁

a harmonic function everywhere except at point P — and an extended

body consists, in the limit, of many point masses or mass elements.
In the general case, this is a theoretically challenging problem. The
existence and uniqueness of the solution has been proven very generally,
see Heiskanen and Moritz (1967, page 18).
Based on the values of the potential function V on the surface S we
may thus compute the harmonic function V(x, y, z) throughout space
outside the surface. The boundary-value problem is a powerful general reuna-arvo-
method also applied in physical geodesy. One must however note tehtävä
that from potential values given on the surface it is not possible to
uniquely resolve the mass distribution inside the Earth, which generates
this potential.
This is clear already in the simple case of a constant potential on the
surface of a sphere. If additionally it is given that the mass distribution
is spherically symmetric, then nevertheless the density profile along
the radius remains indeterminate. All mass may be concentrated in
the centre, or it may form a thin layer just below the sphere’s surface,
or any alternative between these two extremes. Without additional
information — for example from seismic studies or geophysical density
models — we cannot resolve this issue.
The Chasles theorem mentioned above, equation 1.26, and its special
case, equation 1.28, are also examples of this: the theorem shows
how one may describe the exterior potential as generated by a mass
distribution on the surface of a body, although we know that the field’s
source is a mass distribution extending throughout the body!
This is a fundamental limitation of all methods that try to obtain

í Õ ! ¤. û
36 Fundamentals of the theory of gravitation

information on the situation inside the Earth based only on gravimetric

measurements on or outside the Earth.

^ Self-test questions
1. Which instrument was used to determine the constant G? Why is
it difficult to obtain a precise value for this constant?
2. Why do all objects, irrespective of their mass, undergo the same
acceleration of free fall, although the gravitational attraction on a
more massive body is obviously stronger?
3. What is a conservative force field?
(a) A force field for which the force can be written as the gradient
of a unique potential.
(b) A force field in which an object carried along a closed loop
will not gain and not lose energy.
(c) An attractive force field from which no object can escape.
(d) A force field the curl of which vanishes everywhere.
4. On the surface of a homogeneous, spherical asteroid the accel-
eration of free fall is 1 cm/s2 . What is the acceleration of free fall
on another asteroid that is otherwise similar, but has twice the
diameter?
(a) 0.25 cm/s2
(b) 1 cm/s2
(c) 2 cm/s2
(d) 4 cm/s2
5. What is a harmonic potential?
kertaluku 6. What is the order of the Laplace differential equation?
7. Is a linear potential, V(x, y, z) = a + bx + cy + dz (a, b, c, d con-
stants), harmonic?

í Õ ! ¤. û
 Exercise 1 – 1: Core of the Earth
37
8. If the potential in the previous question is a gravitational potential,
calculate its acceleration vector.
9. Under what condition is it possible to describe the external grav-
itational field emanating from a body as produced by a single
mass-density layer on the surface of that body?
10. The dipole-layer density µ is mentioned in section 1.11. What is
the SI unit of this quantity?

^ Exercise 1 – 1: Core of the Earth

1. Derive the equation giving the acceleration of gravity g on the
surface of a homogeneous-density sphere, if the density ρ and
radius Rcore are given.
2. The Earth’s iron-nickel core has a mean density of 11 g/cm3 and its
radius is 3500 km. Compute the acceleration of gravitation on its
surface gcore .
3. What is the attraction g at the centre of the core? What can you
say in general about the geopotential in this point (do not try to
calculate it)?
∂
4. Derive the equation for the radial gravitational gradient ∂r g on
the surface of a homogeneous-density sphere of density ρ.

^ Exercise 1 – 2: Atmosphere
1. The mean pressure of the atmosphere at sea level is 1013.25 hPa
(the unit for pressure, the Pascal, is defined as Pa = N/m2 .) On
the Earth’s surface gravity is 9.81 m/s2 . Calculate the mean surface
density as a thin layer κ in units of kg/m2 .
2. Calculate the total mass of the atmosphere using the spherical
shell approximation. You may take as its radius 6378 km.
3. Calculate the attraction generated by the atmosphere outside it,
both as acceleration and as a fraction of the total Earth attraction.

í Õ ! ¤. û
38 Fundamentals of the theory of gravitation

a a0

Σ1
d

∞ ∞

Σ2

^ Figure 1.11. Iron ore body.

4. How much is the attraction from the atmosphere inside the

atmosphere?

^ Exercise 1 – 3: The Gauss theorem

There is a deposit (body) of iron ore inside the Earth, which generates
(in the flat Earth approximation!) an attraction on the Earth’s surface,
which has been drawn as the a curve. See figure 1.11.
The true attraction curve is approximated by a simple function
⎧
⎨a if r ⩽ d
0
a=
⎩0 if r > d
(red dashed line), where r is the distance from the point on the Earth’s
surface straight above the ore deposit. So, the area where a ̸= 0 is a disk
of radius d on the surface of the Earth.
1. Compute, using the above approximation for a, the surface integral
x
a dS,
Σ1

í Õ ! ¤. û
Exercise 1 – 3: The Gauss theorem
39
where Σ1 is the surface of the Earth, see figure 1.11.
2. According to the Gauss integral theorem
x x y
⟨a1 · n1 ⟩ dS + ⟨a2 · n2 ⟩ dS = ∆V dV =
Σ1 Σ2 volume
y
= −4πGρiron dV = −4πGMbody ,
volume

in which Σ1 + Σ2 is the (two part) closed surface around the body.

The parts meet at infinity. a1 and a2 are the attraction vectors on
the surface of the Earth and on the surface Σ2 , and n1 and n2 are
the exterior normals to the surfaces.
Assuming that
x x x
⟨a1 · n1 ⟩ dS = ⟨a2 · n2 ⟩ dS = − a dS,
Σ1 Σ2 Σ1

calculate GMbody . Be careful with the algebraic signs!

3. Assuming that the deposit is a sphere at depth d, calculate GM
using Newton’s law of gravitation from the value a0 straight above
the deposit at the Earth’s surface.
4. Compare results 2 and 3 and draw conclusions. Is the function a
given above a good approximation?

í Õ ! ¤. û
^ The Laplace equation and its
solutions

^ 2.1 The nature of the Laplace equation

2
The Laplace equation is central to the study of the Earth’s gravitational
field: (︃ )︃
∂2 ∂2 ∂2
∆V = + + V = 0.
∂x2 ∂y2 ∂z2
We call the symbol ∆ the Laplace operator. Often, the alternative notation
∇2 is used.
If we study gravitation as a field, the Laplace equation is more natural
than Newton’s formalism. Newton’s equation is used when the mass
distribution is known: it yields directly the gravitational force caused
by the masses.
The Laplace equation, on the other hand, is a partial differential
equation. Its solution gives the potential V(x, y, z) of the gravitational
field throughout space or a part of space. From this potential one
may then calculate the effect of the field on a body moving in space
at the location where the body is. This is a two-phase process. It is
conceptually new here that a certain property, a field, is attributed to
empty space, and we no longer talk about action at a distance directly
between two bodies.
Solving the Laplace equation in the general case may be difficult. The
approach generally taken is that we choose some co-ordinate frame

– 41 –
42 The Laplace equation and its solutions

— a rectangular frame (as above), spherical co-ordinates, cylindrical

co-ordinates, toroidal co-ordinates, or whatever — which agrees best
with the geometry of the problem at hand. Then, we transform the
Laplace equation to those co-ordinates, we find special solutions of a
certain form, and finally we compose a general — or not-so-general —
solution as a linear combination of those special solutions, i. e., a series
expansion.
Fortunately, the theory of linear partial differential equations is well-
developed. Similar theoretical problems are encountered in the theories
of electromagnetic fields (Maxwell theory) and quantum mechanics
(Schrödinger1 equation), not to mention fluid and heat flow.
It is important to note that the Laplace equation is linear. This means
that, if two solutions are given

∆V1 = 0, ∆V2 = 0,

then their linear combinations

V = αV1 + βV2 , α, β ∈ R

are also good solutions: ∆V = 0. This property of linearity makes

it possible to seek general solutions as linear combinations or series
expansions of basic solutions.
A peculiarity that also distinguishes the Laplace equation from
Newton’s equation is that it is a local equation. It characterises the
behaviour of the potential field in a small neighbourhood of one point.
However, the solution is sought for a whole area. The solution approach
reuna-arvo- commonly used is the boundary-value problem. This means that the field
tehtävä
1 ErwinRudolf Josef Alexander Schrödinger (1887–1961) was a German physicist and
quantum theorist, the inventor of the wave equation of matter named after him, which
earned him the 1933 physics Nobel (shared with Paul Dirac), and of the eponymous
unobserved cat, which finds itself in a superposition state of being both alive and
dead.

í Õ ! ¤. û
 The Laplace equation in rectangular co-ordinates
43
values (“boundary values”) have to be given only on the boundary of a
certain part of space, for example on the Earth’s surface. From this, one
calculates the values of the field in outer space — the behaviour of the
field inside the Earth remains outside the scope of our interest. From
the perspective of the exterior gravitational field, one does not even
need to know the precise mass distribution inside the Earth — and one
cannot even determine it using only measurement values obtained on
and above the Earth’s surface!

^ 2.2 The Laplace equation in rectangular co-ordinates

It is a learning experience to write and solve the Laplace equation in
rectangular co-ordinates. The case is analogous to that of spherical
co-ordinates but the maths is much simpler.
Assume that the Earth’s surface, or sea level, is the level surface for z
co-ordinates z = 0. Then
(︃ )︃
∂2 ∂2 ∂2 (︁ )︁
∆V = 2
+ 2 + 2 V = ∆ X(x) · Y(y) · Z(z) ,
∂x ∂y ∂z
in which we have “experimentally” written

V(x, y, z) = X(x) · Y(y) · Z(z).

In other words, we write experimentally V as the product of three factor

functions, with each factor function depending only on one co-ordinate
— the “separation of variables”. A realistic potential function V will of muuttujien
course usually not be in this form. We may however hope to write it as erottaminen
a linear combination of terms that are in the above form, thanks to the
linearity of the Laplace equation.
By taking all the partial derivatives, we obtain
∂2 ∂2 ∂2
YZ X + XZ Y + XY Z = 0.
∂x2 ∂y2 ∂z2
Dividing by the expression XYZ yields
1 ∂2 1 ∂2 1 ∂2
X(x) + Y(y) + Z(z) = 0.
X(x) ∂x2 Y(y) ∂y2 Z(z) ∂z2

í Õ ! ¤. û
44 The Laplace equation and its solutions

Because this has to be true in the whole space, i. e., for all combinations
of values x, y, z, it follows that each term must be a constant. If we take for
the first and second constants −ω2x and −ω2y , we get in conclusion for
the third constant ω2x + ω2y . By writing this definition and result out
and moving the denominator to the other side, we obtain
∂2
X(x) = −ω2x X(x),
∂x2
(why the minus sign? We shall presently see. . . )
∂2
2
Y(y) = −ω2y Y(y),
∂y
and
∂2
Z(z) = ω2x + ω2y Z(z).
(︁ )︁
∂z 2

Now, the solution is readily found at least to the first two equations:
harmoninen they are harmonic oscillators, and their basis solutions2 are
värähtelijä (︁ )︁
X(x) = exp ±iωx x ,
(︁ )︁
Y(y) = exp ±iωy y .
The solution of the Z equation, on the other hand, is exponential:
(︂ √︂ )︂
Z(z) = exp ±z ω2x + ω2y .
We can now form basis solutions in space:
(︂ √︂ )︂
2 2
Vωx ωy (x, y, z) = exp i (±ωx x ± ωy y) ± z ωx + ωy .
The general solution is obtained by summing the terms Vωx ωy for
different values of ωx and ωy with varying coefficients.
We cannot choose the value pair (ωx , ωy ) entirely freely. The values
which are allowed will depend on the boundary conditions given.
Let us assume that in both the x and y directions the size of our
world is L (“shoebox world”3 ). Let us make things a little simpler by

2 Alternative basis solutions are X(x) = sin ωx x, X(x) = cos ωx x etc. They are equiv-
alent to those presented because exp(iωx x) = cos ωx x + i sin ωx x, exp(−iωx x) =
cos ωx x − i sin ωx x.
3 . . . although real-world shoeboxes are rarely square.

í Õ ! ¤. û
The Laplace equation in rectangular co-ordinates
45
assuming that on the boundary surfaces of our shoebox world we have
the boundary conditions

V(0, y, z) = V(L, y, z) = V(x, 0, z) = V(x, L, z) = 0.

It then follows that the only pairs (ωx , ωy ) that yield a solution that fits
the box are
πj πk
ωx = , ωy = , j, k ∈ Z,
L L
and the only suitable functions are sine functions. Thus we obtain as a
solution:
x y z
(︂ )︂ (︂ )︂ (︂ √︁ )︂
Vjk (x, y, z) = sin πj sin πk exp ±π (j2 + k2 ) .
L L L
This particular solution may now be generalised by multiplying it
by suitable coefficients, and summing it over different index values
j = 0, ±1, ±2, . . . and k = 0, ±1, ±2, . . . We may however remark that the
terms for which j = 0 or k = 0 will always vanish, and the terms that
contain j = +n and j = −n, or k = +n and k = −n, n ∈ N, are (apart
from their algebraic signs) identical. Therefore in practice we sum over
the values j = 1, 2, . . . and k = 1, 2, . . .
Different boundary conditions will give slightly different general
solutions. Their general form is, however, always similar.
The zero-level z = 0 expansion resulting from the general solution is
the familiar Fourier4 sine expansion:
Vjk (x,y)
⏟ (︂ )︂⏞⏞ (︂ )︂⏟
∑︂
∞ ∑︂
∞
jx ky
V(x, y, 0) = vjk sin π sin π , (2.1)
L L
j=1 k=1

in which vjk are Fourier coefficients, and the expressions

jx ky
(︂ )︂ (︂ )︂
def
Vjk (x, y) = sin π sin π
L L
are two-dimensional basis functions on the Earth’s surface, more pre- kantafunktio

4 JosephFourier (1768–1830) was a French mathematician and physicist — and some

would say, climatologist — one of the Eiffel Tower’s 72 names, Eiffel Tower, 72 names.

í Õ ! ¤. û
46 The Laplace equation and its solutions

13πx 5πx
Sea level sin sin
L L

Figure 2.1. The exponential attenuation of gravitational field Fourier waviness

with height. Rectangular geometry, one dimension. Long waves
(small wave numbers, red) attenuate more slowly with height than
^ short waves (green), meaning that height acts as a low-pass filter.

cisely, on level z = 0.
We refer to section B.2.2 in appendix B for a description with illustra-
tion of how a Fourier analysis and synthesis on a simple function is done,
and how the Fourier expansion approximates the original function as
terms are added.
A complete three-dimensional expansion again is
Vjk (x,y)
⏟ (︂ )︂⏞⏞ (︂ )︂⏟
∑︂
∞ ∑︂
∞
jx ky
(︂ √︁
z
)︂
V(x, y, z) = vjk sin π sin π 2
exp ±π j + k 2 . (2.2)
L L L
j=1 k=1

The z expression may have a positive as well as a negative algebraic

sign! Of course the solution with a positive sign goes to → ∞ when
z → ∞, which is not physically realistic in the exterior space.
Note also that V(x, y, 0) and vjk represent the same gravitational field

í Õ ! ¤. û
 The Laplace equation in rectangular co-ordinates
47
Space domain Frequency domain
Fourier F
V(x, y, 0) −−−−−−−−−−−−−−−−→ vjk
⏐ ⏐
× (easy)↓
⏐(hard) ⏐
↓
Inverse Fourier F −1
(︂ √︁ )︂
V(x, y, z) ←−−−−−−−−−−−−−−−− vjk exp −π j2 + k2 Lz

Figure 2.2. Vertically shifting the potential field V in the space and frequency
^ domains. Rectangular geometry.

in two essentially different ways: in the space domain, and in the —

spatial — frequency, or wave-number, domain. The information content
in the two is the same. They can be transformed into each other by the
forward and inverse Fourier transforms F and F−1 .
In fact, the information content in V(x, y, 0) is in principle the same
as that in V(x, y, z) for any level z: knowing the potential on one
surface means — with the Laplace equation — knowing the potential
throughout space.
We summarise equations 2.1 and 2.2 in commutative diagram 2.2. kommutoiva
kaavio
The takeaway from this is that the operation of vertically shifting the
potential field V from zero level to the level z, which is not straightfor-
ward in the space domain, becomes simple — as in a straightforward
multiplication — in the frequency domain.5 The same applies in
spherical co-ordinates, where the frequency domain means spherical-
harmonic coefficients, as we shall see.

5 Thereason for this, as we shall later discuss more generally, is that the vertical shift
operation is a convolution.

í Õ ! ¤. û
48 The Laplace equation and its solutions

^ 2.3 The Laplace equation in polar co-ordinates

In polar co-ordinates, two-dimensionally, the Laplace equation is

∂2 V 1 ∂V 1 ∂2 V
∆V = + r ∂r + = 0.
∂r2 r2 ∂α2
We carry out on this the same kind of separation of variables as was
done in section 2.2. Write first

V(α, r) = A(α)R(r)

and then split the above equation into two equations, one for the right-
hand side function R(r) and one for the function A(α). Substitution
yields

∂2 R(r) A(α) ∂R(r) R(r) ∂2 A(α)

A(α) + r + 2 = 0.
∂r2 ∂r r ∂α2
2/︁
Multiply by the expression r A(α)R(r) :

r2 ∂2 R(r) 1 ∂2 A(α)
(︃ )︃
r ∂R(r)
+ + = 0.
R(r) ∂r2 R(r) ∂r A(α) ∂α2
Both terms must be constant:
(︃ 2 )︃
∂ R(r) ∂R(r)
r r + − k2 R(r) = 0,
∂r2 ∂r
∂2 A(α)
+ k2 A(α) = 0.
∂α2
Here, the algebraic sign of the constant k2 has been chosen so that A(α)
gets a periodic solution. Such a general solution would be

Ak (α) = ak cos kα + bk sin kα,

in which, because angle α has a period of 2π, k has to be a non-negative

integer: k = 0, 1, 2, . . . . Negative k does not give different solutions,
because

ak cos kα = ak cos(−kα), bk sin kα = −bk sin(−kα).

í Õ ! ¤. û
The Laplace equation in polar co-ordinates
49
The other equation, in the function R(r), is harder to solve. A test
solution is a power law:
R(r) = rq .
Substitution yields

r rq (q − 1) rq−2 + qrq−1 − k2 rq = 0
(︁ )︁

=⇒ q2 − k2 = 0
=⇒ q2 = k2 .

This works for positive q = 2, 3, . . . and negative q = −1, −2, . . . For

q = 1 we find
r − k2 r = 0 =⇒ k2 = 1 = q2 .
For k = 0, besides the trivial constant solution, the non-trivial solution
R(r) = ln r is found:
1 1
(︂ )︂
r r · − 2 + r − k2 ln r = 0 =⇒ k = 0.
r
Thus we obtain the general solution
⎧
⎨1 or ln r if k = 0,
Rk (r) =
⎩rk or r−k if k = 1, 2, . . .

We see that, if we require the solution to exist at the origin r = 0, we

need the first solutions, obtaining
∑︂
∞
V1 (α, r) = a0 + rk (ak cos kα + bk sin kα) ,
k=1

but if we require existence — or, at least, good behaviour — at infinity6

r → ∞, we need the second solutions,
∑︂
∞
V2 (α, r) = a0 + b0 ln r + r−k (ak cos kα + bk sin kα) . (2.3)
k=1

There is a clear similarity here to the three-dimensional, i. e., spherical

co-ordinates, case.
6 In fact, limr→∞ V2 → ∞ but limr→∞ ∂
∂r V2 = 0.

í Õ ! ¤. û
50 The Laplace equation and its solutions

^ 2.4 Spherical, geodetic, ellipsoidal co-ordinates

In physical geodesy we use geometrical and physical concepts side by
side. For example, the co-ordinates of a point can be given in the form
(X, Y, Z) , which are in principle geometric — except for the physical
assumption that the origin of the co-ordinate system is in the centre of
mass of the Earth.
As the Earth is not precisely a sphere but rather an oblate ellipsoid of
revolution, one cannot use geographical co-ordinates as if they were
spherical co-ordinates. Because the flattening of the Earth — some
0.3 % — cannot be ignored, this difference is significant. The connection
between spherical co-ordinates (ϕ, λ, r) and rectangular ones (X, Y, Z)
is the following:

X = r cos ϕ cos λ,
Y = r cos ϕ sin λ, (2.4)
Z = r sin ϕ.

Here ϕ is the geocentric latitude and λ is the (ordinary — geocentric,

geodetic or geographical, all three are the same) longitude, while r is
the distance from the Earth’s centre. The X axis points in the direction
of the Greenwich meridian. See figure 2.3.
On the Earth’s surface, these spherical co-ordinates are not very useful
because of the Earth’s flattening, but in space, spherical co-ordinates
are much-used. On the Earth’s surface most often geodetic — also called
geographical — co-ordinates (φ, λ, h) are used:

X = (N + h) cos φ cos λ,
Y = (N + h) cos φ sin λ, (2.5)
Z = N + h − e2 N sin φ,
(︁ )︁

in which
a a2
N(φ) = √ = √ . (2.6)
1 − e2 sin2 φ a2 cos2 φ + b2 sin2 φ

í Õ ! ¤. û
 Spherical, geodetic, ellipsoidal co-ordinates
51
Pole Z
r cos ϕ P

Equator
r r sin ϕ

ϕ Y

λ
X

Greenwich meridian

^ Figure 2.3. Definition of spherical co-ordinates.

The quantity N defined in equation 2.6 is the west-east direction, or

transversal, radius of curvature of the reference ellipsoid. In the equation, poikittainen
a is the equatorial radius of the reference ellipsoid used, vertaus-
ellipsoidi
a2 − b 2
2 def
e =
a2
is the square of the first eccentricity,7 and in equations 2.5, h is the height
of the point above the reference ellipsoid, see figure 2.4.
Converting rectangular co-ordinates into geodetic ones is easiest to
do iteratively, although the literature also offers closed formulas.
Spherical co-ordinates and geodetic, also called geographical, co-
ordinates are considerably different. In latitude, the difference is up to
11 minutes of arc, or almost 20 kilometres. This maximum is attained
for latitudes ±45◦ .

7 The parameter is connected to the Earth’s flattening f through the equation e2 = 2f−f2 .

í Õ ! ¤. û
52 The Laplace equation and its solutions

Z Ellipsoidal
P normal
x

(︂(︁ )︂ h
1 − e2 N + h sin φ
)︁

φ X, Y
O (N + h) cos φ
Reference
ellipsoid

^ Figure 2.4. Definition of geodetic co-ordinates.

In theoretical work one also uses ellipsoidal co-ordinates (β, λ, u). The
redukoitu co-ordinate β is called the reduced latitude. The relationship with
leveysaste rectangular co-ordinates is
√︁
X = u2 + E2 cos β cos λ,
√︁
Y = u2 + E2 cos β sin λ, (2.7)
Z = u sin β.

If the semimajor axis of the Earth ellipsoid is a and its semiminor axis
b, it follows that E2 = a2 − b2 .

^ 2.5 The Laplace equation in spherical co-ordinates

The Laplace equation transformed to spherical co-ordinates reads (see
appendix E for a simple geometric proof):
∂2 V 2 ∂V 1 ∂2 V tan ϕ ∂V 1 ∂2 V
∆V = 2
+r + 2 2
− + 2 = 0, (2.8)
∂r ∂r r ∂ϕ r ∂ϕ r sin ϕ ∂λ2
2 2

in which ϕ is the (geocentric) latitude, λ is the longitude, and r is the

distance from the origin or centre of the Earth.

í Õ ! ¤. û
The Laplace equation in spherical co-ordinates
53
Here we shall not derive the solution of this equation by separation
of variables, as it is pretty complicated. It can be found in section E.2
and in the literature (Heiskanen and Moritz, 1967, section 1–9). What is
significant is that the solution looks somewhat similar to the solution
in rectangular co-ordinates presented earlier, section 2.2. The basis
solutions of the Laplace equation are

˜︁n,1 (ϕ, λ, r) = rn Yn (ϕ, λ),

V ˜︁n,2 (ϕ, λ, r) = Yn (ϕ, λ) ,
V n = 0, 1, . . .
rn+1
(2.9)

of which the first is again nonphysical in outer space, because, unlike the
true geopotential, these expressions grow to infinity for r → ∞.
In the above equations, the functions Yn (ϕ, λ) are called surface pinta-
spherical harmonics, whereas the functions V ˜︁n (ϕ, λ, r) are solid spherical pallofunktio
harmonics. The latter are harmonic functions everywhere in space except avaruus-
at the origin (2.9, rightmost equation) or at infinity (leftmost, physically pallofunktio
unrealistic equation).
The functions Yn are
∑︂
n
Yn (ϕ, λ) = Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) . (2.10)
m=0

The functions Pnm are Legendre functions, on which more later on. With
the help of expression 2.10, we obtain, by using the second, physically
realistic alternative from equations 2.9, the following solution or series pallofunktio-
expansion for the potential V in space: kehitelmä

∑︂
∞
1 ∑︂
n
V(ϕ, λ, r) = Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) .
rn+1
n=0 m=0
(2.11)
The coefficients anm and bnm are called the coefficients of the spherical-
harmonic expansion, in short, spectral coefficients. Together they rep-
resent the function V, in somewhat the same way that the Fourier
coefficients vjk do in rectangular co-ordinates in equation 2.2. The
subscripts n and m are called degree and order. asteluku,
järjestysluku
í Õ ! ¤. û
54 The Laplace equation and its solutions

Often we will be using a somewhat freer notation for the scaled

functions Yn Rn+1 For example, if we expand the disturbing potential
/︁

pallofunktio T into spherical harmonics, we shall use the notation Tn (ϕ, λ) for its
surface harmonics. Similarly, ∆gn (ϕ, λ) is the surface harmonic of the
gravity anomaly ∆g for degree n, and so on. Then, it holds on the
asteosuus- Earth’s surface r = R (degree constituent expansion) that
hajotelma
∑︂
∞ ∑︂
∞
T (ϕ, λ, R) = Tn (ϕ, λ), ∆g(ϕ, λ, R) = ∆gn (ϕ, λ),
n=0 n=0

and so on.

^ 2.6 Dependence on height

From the above equations 2.9 one sees that for different values of
the degree n the function V˜︁n (ϕ, λ, r) has a different dependence on
the distance r from the Earth’s centre, or equivalently, on the height
H = r − R, if by R we denote the radius of the Earth sphere, i. e., sea
level. The dependence is

˜︁n (ϕ, λ, r) = Yn (ϕ, λ) .

V
rn+1
At sea level
V˜︁n (ϕ, λ, R) = Yn (ϕ, λ) def
= Vn (ϕ, λ).
Rn+1
Therefore, we may write
(︂ )︂n+1
˜︁n (ϕ, λ, r) = R R + H −(n+1)
(︂ )︂
V r Vn (ϕ, λ) = Vn (ϕ, λ) =
R
H −(n+1) H
(︂ )︂ (︂ )︂
= 1+ Vn (ϕ, λ) ≈ exp − (n + 1) Vn (ϕ, λ).
R R
We see that the attenuation of the potential with height is again expo-
nential, and the harmonic degree number n appears in the exponent,
as also did the wave number in rectangular geometry, see equation 2.2
and figure 2.1. The analogy works.

í Õ ! ¤. û
^ Legendre functions and spherical
harmonics

^ 3.1 Legendre functions

3
In equations 2.10 and 2.11, the functions P are Legendre1 functions that pop
up whenever we solve a Laplace-like equation in spherical co-ordinates.
There exist various effective, so-called recursive algorithms, for example
the following algorithm only for ordinary Legendre polynomials Pn =
Pn0 :
nPn (t) = − (n − 1) Pn−2 (t) + (2n − 1) tPn−1 (t). (3.1)
Similar equations also exist for the functions Pnm , m > 0. There are even
alternatives to choose from, although most equations are complicated.
One should be careful that, in their computation, the factorials do not kertoma
go overboard. Already 30! (factorial of 30) is a larger number than
computers can handle even as 64-bit integers — not to mention 360!, the
factorial of 360. Heiskanen and Moritz’s (1967) equation 1-62, contrary
to what is stated there, is not directly suitable for computer use!
The first Legendre polynomials are listed in table 3.1. Higher polyno-
mials than this are rarely needed in manual computation.

1 Adrien-Marie Legendre (1752–1833) was a French mathematician known for his work

on number theory, statistics — he invented the method of least-squares independently

from Gauss — and elliptical functions. His name is inscribed on the Eiffel Tower,
Eiffel Tower, 72 names.

– 55 –
56 Legendre functions and spherical harmonics

^ Table 3.1. Legendre polynomials. t = sin ϕ.

Function of t Expressed in sines and cosines

P0 (t) = 1 P0 (sin ϕ) = 1
P1 (t) = t P1 (sin ϕ) = sin ϕ
P2 (t) = 23 t2 − 12 P2 (sin ϕ) = − 43 cos 2ϕ + 14
P3 (t) = 25 t3 − 32 t P3 (sin ϕ) = − 85 sin 3ϕ + 38 sin ϕ
P4 (t) = 18 35t4 − 30t2 + 3
(︁ )︁

P5 (t) = 18 63t5 − 70t3 + 15t

(︁ )︁
1
231t6 − 315t4 + 105t2 − 5
(︁ )︁
P6 (t) = 16

kantafunktio For comparison, the Fourier basis functions (like, in a more complicated
way, sines and cosines as well!)
x
(︂ )︂
Fj (x) = exp 2πij ,
L

−90◦ −30◦ 0◦ −→ ϕ 30◦ 90◦

1 P0
P3 P6
P4 P10 P1
0.5 P5 P25
P5

P6
−0.5 P4 P2 P3
P1

−1
−1 −0.5 0 −→ t 0.5 1

Figure 3.1. A number of Legendre polynomials P0 (t), . . . , P25 (t) as functions

^ of the argument t = sin ϕ.

í Õ ! ¤. û
 Legendre functions
57
in which i2 = −1, can also be computed recursively:

Fj+1 (x) = Fj (x) · F1 (x).

^ 3.1.1 Properties of Legendre polynomials

◦ The even polynomials — those polynomials of which the harmonic
degree number n is even — are mirror symmetric through the
origin ϕ = t = 0, i. e., the equatorial plane: Pn (−t) = Pn (t), or
(︁ )︁
equivalently Pn sin(−ϕ) = Pn (sin ϕ). This means that their val-
ues at the same latitude north and south of the equator are identical.
The odd polynomials are again antisymmetric: Pn (−t) = −Pn (t)
(︁ )︁
or Pn sin(−ϕ) = −Pn (sin ϕ), meaning that their values at the
same latitude north and south of the equator are of opposite signs.
◦ From figure 3.1, we see that the polynomials Pn (t) go, on the
whole interval t ∈ [−1, 1], or ϕ ∈ [−90◦ , 90◦ ], precisely n times
through zero.
◦ As the values in the end points t = ±1, ϕ = ±90◦ are ±1, it follows
that there are precisely n + 1 “algebraic-sign intervals”, meaning
intervals of t or ϕ on which the polynomial assumes only positive
or only negative values.

^ 3.1.2 Properties of associated Legendre functions

We give several of the associated Legendre functions Pnm , m ̸= 0 in Legendren
table 3.2 for illustration. liitännäis-
funktio
One defining equation for these is
)︁ m/2 dm Pn (t)
Pnm (t) = 1 − t2
(︁
. (3.2)
dtm
◦ The associated Legendre functions are also either mirror sym-
metric through the origin or equatorial plane, Pnm (−t) = Pnm (t),
(︁ )︁
or equivalently, Pnm sin(−ϕ) = Pnm (sin ϕ), or antisymmetric,
(︁ )︁
Pnm (−t) = −Pnm (t) or Pnm sin(−ϕ) = −Pnm (sin ϕ), depend-
ing on the values of order number m and degree number n. järjestysluku,
asteluku
í Õ ! ¤. û
58 Legendre functions and spherical harmonics

5
P32
P22
P21
P31
0
P11
P31 P21

−5
P32 P33
P25,25
P33
5 · 1030
−10 P21
P31
P32
−15
−1 −0.5 0 −→ t 0.5 1

Figure 3.2. Associated Legendre functions. Note the extremely different scale
^ used for the function P25,25 , see equation 3.7.

◦ Figure 3.2 suggests that the polynomials Pnm (t) go on t ∈ [−1, 1],
or ϕ ∈ [−90◦ , 90◦ ], precisely n − m times through zero. This is
indeed the case.
◦ As the values in the end points t = ±1, ϕ = ±90◦ are also zero, it
follows that there are precisely n−m+1 “algebraic-sign intervals”.

^ Table 3.2. Associated Legendre functions.

Function of t Trigonometric function

√
P11 (t) = 1 − t2 P11 (sin ϕ) = cos ϕ
√
P21 (t) = 3t 1 − t2 P21 (sin ϕ) = 3 sin ϕ cos ϕ
P22 (t) = 3 1 − t2 P22 (sin ϕ) = 3 cos2 ϕ
(︁ )︁
)︁ √
P31 (t) = 32 5t2 − 1 1 − t2 P31 (sin ϕ) = 32 5 sin2 ϕ − 1 cos ϕ
(︁ (︁ )︁

P32 (t) = 15t 1 − t2 P32 (sin ϕ) = 15 sin ϕ cos2 ϕ

(︁ )︁
)︁ 3/
P33 (t) = 15 1 − t2 2 P33 (sin ϕ) = 15 cos3 ϕ
(︁

í Õ ! ¤. û
 Legendre functions
59
^ 3.1.3 Surface spherical harmonics
Starting from equation 2.10, we may write

Yn (ϕ, λ) =
∑︂
n
(︁ )︁
= anm Pnm (sin ϕ) cos mλ + bnm Pnm (sin ϕ) sin mλ =
m=0 ∑︂
n
= vnm Ynm (ϕ, λ),
m=−n

in which m now runs from −n to +n. Here

⎧
⎨P (sin ϕ) cos mλ if m ⩾ 0,
nm
Ynm (ϕ, λ) = (3.3)
⎩Pn|m| (sin ϕ) sin |m| λ if m < 0.

These are the surface spherical harmonics of degree n and order m. pinta-
pallofunktio
Surface spherical harmonics come in three kinds:
asteluku,
Zonal harmonics m = 0. These functions depend only on latitude. järjestysluku
Sectorial harmonics m = n. the algebraic signs of these functions vyöhyke-
depend only on longitude and not on latitude. The functions funktiot
themselves however do depend on both latitude and longitude! sektorifunktiot
Tesseral harmonics 0 < m < n. These functions, the algebraic sign ruutufunktiot
of which changes with both latitude and longitude, form a
checkerboard pattern on the surface of the sphere, if the positive
values are painted white and the negative ones grey (Latin tessera
= a tile, as used in a mosaic).

Every function will, on the interval sin ϕ ∈ [−1, +1], go precisely n − m

times through zero. Every function is either symmetric or antisymmetric
through the origin as a function of ϕ or of t = sin ϕ.
Spherical harmonics thus represent a wave phenomenon of sorts.
They are however not wave functions (sines or cosines): the connection
to those is complicated at least. It nevertheless makes sense to speak of
their wavelength.

í Õ ! ¤. û
60 Legendre functions and spherical harmonics

(a) (b) (c)

Zonals: P50 (sin ϕ) Sectorials: Tesserals:
P66 (sin ϕ) cos 6λ P11,6 (sin ϕ) cos 6λ

Figure 3.3. The algebraic signs of spherical harmonics on the Earth’s surface.
White means positive, grey negative. The functions “wave” in a
^ sine or cosine function-like fashion.

pallofunktio Figure 3.3 depicts how the algebraic signs of the different spherical
harmonics behave on the Earth’s surface — and above. This is a
perspective sketch and not all white and grey areas are visible!
In equation 2.10, the expressions cos mλ and sin mλ go around a full

Figure 3.4. Surface spherical harmonics as maps. Horizontal axis λ ∈ [0, 360◦ ),
vertical axis ϕ ∈ [−90◦ , 90◦ ]. Functions depicted are

P50 (sin ϕ) P66 (sin ϕ) cos 6λ P11,6 (sin ϕ) cos 6λ

.
^ P40 (sin ϕ) P65 (sin ϕ) cos 5λ P10,6 (sin ϕ) cos 6λ

í Õ ! ¤. û
 Legendre functions
61
^ Table 3.3. Semi-wavelengths for different degrees and orders of spherical
harmonics.

m or n − m Semi-wavelength (km) In degrees

10 2000 18
40 500 4◦. 5
180 111 1
360 55 30 = 0◦. 5
′

1800 11 6 ′ = 0◦. 1
10 800 1.85 1 ′ = 0◦. 017

circle, the equator, 0◦ ⩽ λ < 360◦ , or 0 ⩽ λ < 2π, precisely 2m times

through zero. The “semi-wavelength” is thus puoli-
aallonpituus
2πR R
= πm,
2m
in which R is again the radius of the Earth.
A similar formula also applies for the functions Pnm (sin ϕ): as the
function passes through zero n − m times on the interval — from pole
to pole — −90◦ < ϕ < 90◦ or − π 2 < ϕ < π 2 , it follows that also in
/︁ /︁

this case, the semi-wavelength is

πR
n − m.
If we plug various values for m and n − m into this, we obtain table 3.3.
This table also gives the resolution that can be achieved with a spherical-
harmonic expansion, or in how detailed a fashion the expansion can
describe the Earth’s gravity field. The expansions available today, like
the model EGM2008, go to harmonic degree n = 2159; the “sharpness”
of a geopotential image based on them is thus 9 km. Models based on
satellite orbit perturbations often extend only to degree 20, meaning
that only details the size of continents — order of magnitude 1000 km —
will be visible. On the other hand, experimental spherical-harmonic pallofunktio-
kehitelmä

í Õ ! ¤. û
62 Legendre functions and spherical harmonics

expansions of the topography go even up to degree 10 800 (Balmino

et al., 2012).

^ 3.2 Symmetry properties of the spherical-harmonic

expansion
pallofunktio- We recapitulate the spherical-harmonic expansion given at the begin-
kehitelmä ning, equation 2.11:

∑︂
∞
1 ∑︂
n
V(ϕ, λ, r) = Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) .
rn+1
n=0 m=0

^ 3.2.1 Dependence on latitude ϕ

It is seen that the dependence on ϕ only works through the Legendre
function Pnm (sin ϕ). This function can, in terms of mirror symmetry
between the northern and southern hemispheres, be either symmetric in
ϕ or antisymmetric in ϕ. This means that either (symmetric case)
(︁ )︁
Pnm (sin ϕ) = Pnm sin(−ϕ)

or (antisymmetric case)
(︁ )︁
Pnm (sin ϕ) = −Pnm sin(−ϕ) .

Equivalently, it means, with t = sin ϕ, that either (symmetric case)

Pnm (t) = Pnm (−t)

or (antisymmetric case)

Pnm (t) = −Pnm (−t).

Which case applies depends on the values of both n and m. To work it

out, one can look at, say, equation 3.2:
)︁ m/2 dm Pn (t)
Pnm (t) = 1 − t2
(︁
.
dtm
We need to answer two questions:

í Õ ! ¤. û
 Symmetry properties of spherical harmonics
63
1. For which values n is the polynomial Pn (t) symmetric, for which
is it antisymmetric in t? For this, you may look at the recursive
algorithm for computation of the polynomials, eq. 3.1. We
already know that P0 (t) = 1 is symmetric, and that P1 (t) = t is
antisymmetric. The rule for other n values follows recursively (or
you could cheat by looking at table 3.1).
d
2. What does differentiation dt do to the symmetry or antisymmetry
of the function?
√
Multiplication by 1 − t2 = cos ϕ changes nothing, as this factor is
symmetric in t or ϕ.
So, in order to make expansion 2.11 mirror symmetric between northern
and southern hemispheres, one has to set the coefficients anm , bnm
for which the corresponding Pnm is antisymmetric, to zero. Then,
those terms vanish from the expansion. The coefficients, and terms,
remaining are those for which the corresponding Pnm is symmetric.
In tableau 3.4 we give a code fragment in the octave rapid-prototyping
language to plot an arbitrary surface spherical harmonic, in order to pinta-
visually judge its symmetry properties. Do not believe, test. pallofunktio

^ 3.2.2 Dependence on longitude λ

This dependence works though the “Fourier basis functions” cos mλ kantafunktio
and sin mλ. The interesting property here is rotational symmetry: does
the spherical-harmonic expansion 2.11 change when we change the pallofunktio-
longitude λ? kehitelmä

We see immediately that, for m ̸= 0, there will be dependence

on λ if any coefficient anm , bnm is non-zero. So, in order to obtain
rotational symmetry, all coefficients anm , bnm for values m > 0 must
be suppressed: a11 = b11 = a21 = b21 = a22 = b22 = · · · = 0.
Of the remaining coefficients, we can say that for m = 0, sin mλ = 0
identically, so the coefficients b00 , b10 , b20 , . . . simply do not matter. They
may be any value, including zero. The coefficients a00 , a10 , a20 , . . .

í Õ ! ¤. û
64 Legendre functions and spherical harmonics

^ Tableau 3.4. Plotting a surface spherical-harmonic map.

% Plotting surface spherical harmonics

phi=linspace(-90,90,72);
lab=linspace(0,360,144);
[f,l]=meshgrid(phi,lab);
n=5; m=-3;
leg=legendre(n,sin(phi.*pi./180));
if m >= 0
cs=cos(m.*lab.*pi./180);
else
cs=sin(abs(m).*lab.*pi./180);
end
v=leg(abs(m)+1,:)’*cs;
contourf(l,f,v’)
xlabel(’Longitude’, ’FontSize’, 16)
ylabel(’Latitude’, ’FontSize’, 16)
str=sprintf(’Surface spherical harmonic n=%d, m=%d’, n, m)%
title(str, ’FontSize’, 20)
axis ([0 360 -90 90])
colorbar()
print(’legendre2D.jpg’,’-djpg’)

however do matter, as for m = 0, cos mλ = 1 identically. So we obtain

as the rotationally symmetric expansion
∑︂
∞
1
V(ϕ, λ, r) = V(ϕ, r) = a P (sin ϕ),
rn+1 n n
n=0

in which Pn = Pn0 are the familiar Legendre polynomials, and an = an0 .

^ 3.3 Orthogonality of Legendre functions

Legendre’s polynomials are orthogonal: the integral — formally, a scalar
product of vectors — is
⎧
w +1 ⎨ 2 if n = n ′ ,
Pn (t)Pn ′ (t) dt = 2n + 1 (3.4)
−1 ⎩0 if n ̸= n ′ .

í Õ ! ¤. û
Orthogonality of Legendre functions
65
This orthogonality is just one example of a more general way to look at
functions and integrals over functions. There exists a useful analogy
with vector spaces, see appendix B.
Alternatively we may write, on the surface of a unit sphere σ, using
a parametrisation2 (ψ, α) by angular distance and azimuth, see figure
10.1:
x
Pn (cos ψ)Pn ′ (cos ψ) dσ =
σ
w 2π w π
= Pn (cos ψ)Pn ′ (cos ψ) sin ψ dψ dα =
0 0
w −1 w +1
= −2π Pn (t)Pn ′ (t) dt = 2π Pn (t)Pn ′ (t) dt,
+1 −1
in which t = cos ψ and the surface element of the unit sphere dσ =
sin ψ dψ dα, in which again sin ψ is the determinant of Jacobi3 of the
co-ordinates (ψ, α). So, we have
⎧
x ⎨ 4π if n = n ′ ,
Pn (cos ψ)Pn ′ (cos ψ) dσ = 2n + 1 , (3.5)
σ ⎩0 if n ̸= n ′ ,
in which ψ is the angular distance from some point on the surface of the
sphere. Equation 3.5 tells us that Legendre polynomials are mutually
orthogonal if the vectorial product is defined as an integral over the
surface of the unit sphere σ.
Alternatively, we may also define fully normalised Legendre polyno-
mials
def √
Pn (cos ψ) = 2n + 1Pn (cos ψ). (3.6)
Now the modified scalar product — the mean product over the unit
sphere — is
⎧
x ⎨1 if n = n ′ ,
1
P (cos ψ)Pn ′ (cos ψ) dσ =
4π σ n ⎩0 if n ̸= n ′ ,

2 This parametrisation may be regarded as a latitude, longitude co-ordinate system:

the latitude is 90◦ − ψ = 12 π − ψ, the longitude is α.
3 Carl Gustav Jacob Jacobi (1804–1851) was a German mathematician.

í Õ ! ¤. û
66 Legendre functions and spherical harmonics

showing the polynomials now to be orthonormal.4 Similarly fully nor-

Legendren malised associated Legendre functions also exist, see Heiskanen and
liitännäis- Moritz 1967, page 31:
funktio √︄
def (n − m)!
Pnm (cos ψ) = 2 (2n + 1) P (cos ψ). (3.7)
(n + m)! nm

In this case, the orthonormal functions are those of equation 3.3, but
normalized:
⎧
⎨P (cos ψ) cos mα if m ⩾ 0,
nm
Y nm (ψ, α) =
⎩Pn|m| (cos ψ) sin |m| α if m < 0.

The scalar product that applies is

⎧
x ⎨1 if n = n ′ and m = m ′ ,
1
Y nm (ψ, α)Y n ′ m ′ (ψ, α) dσ =
4π σ ⎩0 otherwise.

^ 3.4 Low-degree spherical harmonics

The potential field of a point mass is (equation 1.4):
GM
V= r .
The corresponding term in the potential expansion 2.11 for degree n = 0
is
˜︁0 (ϕ, λ, r) = 1 a00 P0 (sin ϕ) = a00 ,
V r r
from which
a00 = GM.
So, a00 represents the force field of a point mass or spherically symmetric
pallofunktio- mass distribution centred at the origin. The higher spherical-harmonic
kerroin
4 And also ⎧
1 w +1 ⎨1 if n = n ′ ,
Pn (t)Pn ′ (t) dt =
2 −1 ⎩0 if n ̸= n ′ ,

again, the mean product over the domain of integration.

í Õ ! ¤. û
Low-degree spherical harmonics
67
coefficients are “perturbations” on top of this.
The expansion for the degree-one coefficients looks as follows:

˜︁1 (ϕ, λ, r) = 1 (a11 cos ϕ cos λ + b11 cos ϕ sin λ + a10 sin ϕ) .
V
r2
Write this in vector form using the expression for the location vector

r = (r cos ϕ cos λ) i + (r cos ϕ sin λ) j + (r sin ϕ) k

{︁ }︁
— in which i, j, k is an orthonormal basis of the Euclidean space — ortonormaali
yielding kanta
˜︁1 (r) = 1 (a11 i + b11 j + a10 k) · r .
⟨︁ ⟩︁
V 3
r
The potential field of a dipole is
G
V(r) = ⟨d · r⟩ ,
r3
in which d is the dipole moment. Comparison yields

a11 i + b11 j + a10 k = Gd,

so the first-degree n = 1 spherical-harmonic coefficients represent the pallofunktio-

Earth’s gravitational field’s dipole moment. kerroin

Every mass element dm of our Earth may be taken to consist of

◦ a monopole at the origin of the co-ordinate system, magnitude dm
◦ a dipole, magnitude r · dm, in which r is the location vector of the
mass element.
In that case we may compute the dipole moment of the whole Earth by
integration:
y y
d⊕ = r dm = ρr dV = M⊕ · rcom ,
Earth Earth

in which, by definition, rcom is the location of the centre of mass of the

Earth. From this follows that, if we choose our co-ordinate system
so that the origin is in the centre of mass of the Earth, the spherical-
harmonic coefficients a11 , b11 , a10 vanish. If the equations of motion of

í Õ ! ¤. û
68 Legendre functions and spherical harmonics

Figure 3.5. Monopole, dipole, and quadrupole at the Earth’s centre and their
^ effects on the geoid.

satellites are formulated in a certain co-ordinate system, like in the

case of GPS satellites the WGS84 system, then the origin of the system
is automatically in the centre of mass of the Earth, and the degree-one
spherical-harmonic coefficients are really zero.
pallofunktio The same logic applies to higher degrees of spherical harmonics. The
degree-two coefficients represent the quadrupole moment of the Earth —
corresponding to her inertial tensor — and so on.

^ 3.5 Splitting a function into degree constituents

There exists a useful integral equation for surface spherical harmonics,
if the function itself f on the surface of the sphere has been given.
The equation is Heiskanen and Moritz (1967) equation 1-71, using our
notation Yn → fn :
x
2n + 1
f ϕ ′ , λ ′ Pn (cos ψ) dσ ′ ,
(︁ )︁
fn (ϕ, λ) = (3.8)
4π σ

in which ψ is the geocentric angular distance between evaluation point

(ϕ, λ) and moving data or integration point (ϕ ′ , λ ′ ), see figure 8.2. In
asteosuus- this degree constituent equation 3.8 there is a certain similarity with the
yhtälö projection or coefficient computation formula B.11. Nevertheless, here

í Õ ! ¤. û
Splitting a function into degree constituents
69
we do not have a computation of spectral coefficients, but of “spectral
constituent functions” fn .
We bring to mind the core property of the functions fn ,
∑︂
∞
f(ϕ, λ) = fn (ϕ, λ)
n=0

on the surface of the sphere.

For the proof, we choose without loss of generality as the “north pole”
of the co-ordinate system the evaluation point (ϕ, λ), so ϕ = 90◦ . Then,
ϕ ′ = 90◦ − ψ. By writing (see equation 2.11):
(︁ ′ ′ )︁ ∑︂
∞ ∑︂
n
Pnm cos ψ (anm cos mλ ′ + bnm sin mλ ′ ) ,
(︁ )︁
f ϕ ,λ =
n=0 m=0

substituting this into the degree constituent equation 3.8, and exploiting asteosuus-
the orthogonality of the Legendre functions, we obtain on the right-hand yhtälö
side:
x (︁
2n + 1
f ϕ ′ , λ ′ Pn (cos ψ) dσ ′ =
)︁
IR =
4π σ
x
2n + 1
= an0 2
Pn (cos ψ) dσ ′
4π σ
Then, using equation 3.5:
2n + 1 4π def
IR = an0 = an0 = an .
4π 2n + 1
On the left-hand side of the degree constituent equation we obtain,
because on the assumed north pole ϕ = 90◦ and thus sin ϕ = 1:

IL = fn (ϕ, λ) = fn (90◦ , λ) =
∑︂
n
= Pnm (1) (anm cos mλ + bnm sin mλ) = Pn0 (1)an0 = an ,
m=0

by using equation 2.10 and

Pn0 (1) = 1, Pnm (1) = 0 if m ̸= 0.
As this applies for every point (ϕ, λ), it follows that the degree con-
stituent equation 3.8 is generally true. Note that the values an depend
on the point choice!

í Õ ! ¤. û
70 Legendre functions and spherical harmonics

^ 3.6 Spectral representations of various quantities

^ 3.6.1 Potential
Starting from equation 2.9, we write the spectral expansion of the geopo-
tential V in space:
∑︂
∞ (︂ )︂n+1
R
V(ϕ, λ, r) = r Vn (ϕ, λ), (3.9)
n=0

in which the degree constituents Vn are

Yn (ϕ, λ)
Vn (ϕ, λ) = =
Rn+1
1 ∑︂
n
= n+1 Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) =
R
m=0
1 ∑︂
n
= n+1 vnm Ynm (ϕ, λ).
R
m=−n

kantafunktio Here, the basis functions Ynm are given by equation 3.3:
⎧
⎨P
nm (sin ϕ) cos mλ if m ⩾ 0,
Ynm (ϕ, λ) =
⎩Pn|m| (sin ϕ) sin |m| λ if m < 0,

and the coefficients are

⎧
⎨a if m ⩾ 0,
nm
vnm =
⎩bn|m| if m < 0.

On the Earth’s surface (r = R) we obtain

∑︂
∞ ∑︂
∞
1 ∑︂
n
V(ϕ, λ, R) = Vn (ϕ, λ) = vnm Ynm (ϕ, λ). (3.10)
Rn+1
n=0 n=0 m=−n

kommutoiva We may summarise the relationships found in commutative diagram 3.6.

kaavio Again, as in section 2.2 for rectangular geometry, it is seen that the
shift of the potential function V from the spherical level R to the level
r = R + H is essentially easier in the frequency domain — the degree
constituents Vn (ϕ, λ) — than it is in the space domain.

í Õ ! ¤. û
 Spectral representations of various quantities
71
Space domain Frequency domain
2n + 1 x ∞
Vn (ϕ, λ) =
4π σ
V(ϕ ′ , λ ′ , R)Pn (cos ψ)dσ ′ ∑︂
V(ϕ, λ, R) −−−−−−−−−−−−−−−−−−−−−−−−−−→ = Vn (ϕ, λ)
n=0
⏐ ⏐
× (easy)↓
⏐(hard) ⏐
↓
∑︂ ∞
∑︂ (︁ R )︁n+1
V(ϕ, λ, r) ←−−−−−−−−−−−−−−−−−−−−−−−−−− = r Vn (ϕ, λ)
n=0

Figure 3.6. Radially shifting the potential field V, in the space and frequency
^ domains. Spherical geometry.

^ 3.6.2 Gravitation
In the Neumann5 boundary-value problem we solve a function V of reuna-arvo-
∂
which the normal derivative, ∂n V, is given on the surface of a body or tehtävä
a closed surface in space.
∂ ∂
In the case of a spherical body, we may assume ∂n V = ∂r V and work
with spherical-harmonic expansions. By differentiating equation 3.9 pallofunktio-
we obtain kehitelmä

∂V ∑︂
∞
n + 1 R n+1
(︂ )︂ ∑︂
∞
n + 1 R n+2
(︂ )︂
=− r r Vn (ϕ, λ) = − r Vn (ϕ, λ).
∂r R
n=0 n=0

At sea level this means

∂V ⃓
⃓ ∑︂
∞
n+1
=− Vn (ϕ, λ).
∂r r=R R
⃓
n=0

If we also write at sea level for the gravitation

∂V ⃓
⃓
def def
∑︂
∞
g(ϕ, λ, R) = = gn (ϕ, λ),
∂r r=R
⃓
n=0

5 Carl Gottfried Neumann (1832–1925) was a German mathematician.

í Õ ! ¤. û
72 Legendre functions and spherical harmonics

it follows by analogy that

n+1
gn (ϕ, λ) = − Vn (ϕ, λ),
R
and conversely that

R
Vn (ϕ, λ) = − g (ϕ, λ).
n+1 n
As a result of this, we obtain the spectral representation of the solution to a
certain Neumann problem:

∑︂
∞ (︂ )︂n+1
R ∑︂
∞ (︂ )︂n+1
R gn (ϕ, λ)
V(ϕ, λ, r) = r Vn (ϕ, λ) = −R r . (3.11)
n+1
n=0 n=0

We may write equivalently to expression 3.10 for the potential:

∑︂
∞
def
∑︂
∞
1 ∑︂
n
g(ϕ, λ, R) = gn (ϕ, λ) = gnm Ynm (ϕ, λ),
Rn+1
n=0 n=0 m=−n

and comparison yields

n+1
gnm = − v . (3.12)
R nm
This is an interesting result worth thinking about:
1. Firstly, note how simple the connection 3.12 between potential
vnm and gravitation gnm is in the frequency domain!
2. Secondly, if measurement values of gravitational acceleration
g(ϕ, λ) are available over the whole surface area of the Earth, we
asteosuus- may derive from these the degree constituent functions gn (ϕ, λ)
funktio using the method explained earlier. In this way we can then
obtain the solution by means of equation 3.11 for the whole
exterior geopotential field! This is the basic idea of geopotential —
or geoid — determination, from the spectral perspective.

í Õ ! ¤. û
 Often-used spherical-harmonic expansions
73
^ 3.7 Often-used spherical-harmonic expansions
Of the existing global spherical-harmonic expansions we must mention pallofunktio-
the already outdated EGM96. It was developed by researchers from kehitelmä
Ohio State University using very extensive, mostly gravimetric, data
collected by the American NIMA (National Imagery and Mapping Agency,
the former Defense Mapping Agency DMA, the current National Geospatial-
Intelligence Agency NGA). This expansion goes up to harmonic degree
360. Its standard presentation6 is

GM
(︃ 360 (︂ )︂ ∑︂
∑︂ a n
n )︃
V= r⊕
(︁ )︁
1+ r Pnm (sin ϕ) Cnm cos mλ + Snm sin mλ . (3.13)
n=2 m=0

This form of presentation — the algebraic sign in front of the expansion,

which starts from degree number n = 2, the number one inside the
parentheses which represents the point mass in the origin equal in
magnitude to the total mass of the Earth, and the dimensionless and
“fully normalised” coefficients C and S — is an industry standard in the
global research community in the field of computing spherical-harmonic
expansions as models of the Earth’s gravitational field. Professor
Richard H. Rapp at Ohio State University has been a pioneer, which is
why the models are often called OSU models.
Generally in these models the lower terms — 2 ⩽ n ⩽ 20 — are
derived primarily from analysis of satellite orbit perturbations. Because ratahäiriöt
of this, the models are in a co-ordinate system with the origin in the
Earth’s centre of mass. This explains the absence of the degree-one
coefficients, as explained earlier.
The higher coefficients again — 20 < n ⩽ 360 — were before the year
2000 mostly the result of the analysis of gravimetric data (over land)
and satellite radar altimetric data (over the ocean). After the launches

6 Here a = a⊕ is used to signify the equatorial radius of the Earth’s reference ellipsoid,
not R, and ϕ, signifying geocentric latitude. The co-ordinates (ϕ, λ, r) form a spherical
co-ordinate system.

í Õ ! ¤. û
74 Legendre functions and spherical harmonics

of the gravimetric satellite missions CHAMP, GRACE, and GOCE, and as

a result of their measurements, nowadays at least the degree number
interval 20 < n ⩽ 200 is the product of space geodesy. Only the still
higher-degree coefficients continue to come from terrestrial data. The
newer model EGM2008 (Pavlis et al., 2012) goes up to degree 2159.
In tableau 3.5 we give the first and last coefficients of the EGM96
pallofunktio- model, the newest and best spherical-harmonics model from the time
malli just before the satellite gravity missions. The values tabulated are n, m,
Cnm , Snm and the mean errors (standard deviations) of both coefficients
from their computation. Note that all Sn0 vanish!
Sometimes non-normalised coefficients are also used, and we write

GM⊕
(︃ ∑︂
∞ (︂ )︂ ∑︂
a n
n )︃
V= r 1− r Pnm (sin ϕ) (Jnm cos mλ + Knm sin mλ) . (3.14)
n=2 m=0

Then we use the notation Jn = Jn0 . The coefficient J2 is the most

important spherical-harmonic coefficient of the Earth’s gravity field,
expressing the flattening of the Earth. Based on equations 3.6 and 3.7,
the relationship with the parameters C, S is
{︄ }︄ {︄ }︄
Jn0 √ Cn0
= − 2n + 1 ,
Kn0 Sn0
{︄ }︄ √︄ {︄ }︄ (3.15)
Jnm (n − m)! Cnm
= − 2 (2n + 1) if m ̸= 0.
Knm (n + m)! Snm

^ 3.8 Ellipsoidal harmonics

Laplace differential equation 1.13 may be written and solved in el-
lipsoidal co-ordinates instead of spherical co-ordinates. The result
ellipsoidi- is known as an ellipsoidal-harmonic expansion.7 They are little-used,
funktio- because the maths needed is more complicated. Moreover, ellipsoidal
kehitelmä
7 Thisexpansion for the ellipsoid of revolution differs from the expansion into Lamé
functions found for the triaxial ellipsoid.

í Õ ! ¤. û
 Ellipsoidal harmonics
75
^ Tableau 3.5. Coefficients and mean errors of the EGM96 spherical-harmonic
expansion.

n m Cnm Snm Cnm mean error Snm mean error

2 0 −0.484165371736E−03 0.000000000000E+00 0.35610635E−10 0.00000000E+00
2 1 −0.186987635955E−09 0.119528012031E−08 0.10000000E−29 0.10000000E−29
2 2 0.243914352398E−05 −0.140016683654E−05 0.53739154E−10 0.54353269E−10
3 0 0.957254173792E−06 0.000000000000E+00 0.18094237E−10 0.00000000E+00
3 1 0.904627768605E−06 0.248513158716E−06 0.13965165E−09 0.13645882E−09
3 2 0.904627768605E−06 −0.619025944205E−06 0.10962329E−09 0.11182866E−09
3 3 0.721072657057E−06 0.141435626958E−05 0.95156281E−10 0.93285090E−10
4 0 0.539873863789E−06 0.000000000000E+00 0.10423678E−09 0.00000000E+00
4 1 −0.536321616971E−06 −0.473440265853E−06 0.85674404E−10 0.82408489E−10
4 2 0.350694105785E−06 0.662671572540E−06 0.16000186E−09 0.16390576E−09
4 3 0.990771803829E−06 −0.200928369177E−06 0.84657802E−10 0.82662506E−10
4 4 −0.188560802735E−06 0.308853169333E−06 0.87315359E−10 0.87852819E−10
5 0 0.685323475630E−07 0.000000000000E+00 0.54383090E−10 0.00000000E+00
5 1 −0.621012128528E−07 −0.944226127525E−07 0.27996887E−09 0.28082882E−09
5 2 0.652438297612E−06 −0.323349612668E−06 0.23747375E−09 0.24356998E−09
5 3 −0.451955406071E−06 −0.214847190624E−06 0.17111636E−09 0.16810647E−09
5 4 −0.295301647654E−06 0.496658876769E−07 0.11981266E−09 0.11849793E−09
5 5 0.174971983203E−06 −0.669384278219E−06 0.11642563E−09 0.11590031E−09
6 0 −0.149957994714E−06 0.000000000000E+00 0.14497863E−09 0.00000000E+00
6 1 −0.760879384947E−07 0.262890545501E−07 0.22415138E−09 0.21957296E−09
6 2 0.481732442832E−07 −0.373728201347E−06 0.27697363E−09 0.28105811E−09
6 3 0.571730990516E−07 0.902694517163E−08 0.19432407E−09 0.18682712E−09
6 4 −0.862142660109E−07 −0.471408154267E−06 0.15229150E−09 0.15328004E−09
6 5 −0.267133325490E−06 −0.536488432483E−06 0.89838470E−10 0.87820905E−10
6 6 0.967616121092E−08 −0.237192006935E−06 0.11332010E−09 0.11518036E−09
.. ..
. .
360 358 0.709604781531E−10 0.691761006753E−10 0.50033977E−10 0.50033977E−10
360 359 0.183971631467E−10 −0.310123632209E−10 0.50033977E−10 0.50033977E−10
360 360 −0.447516389678E−24 −0.830224945525E−10 0.50033977E−10 0.50033977E−10

co-ordinates are mostly only theoretically interesting and not in any

broad use within geodesy.

í Õ ! ¤. û
76 Legendre functions and spherical harmonics

The form of presentation is

V(β, λ, u) =
∑︂
∞ ∑︂ n
Qnm i u
(︁ )︁
= (︁ Eb )︁ Pnm (sin β) (Aenm cos mλ + Benm sin mλ) , (3.16)
n=0 m=0
Q nm i E

in which Qnm (z) are the Legendre functions of the second kind, sampled in
table 3.6. Although the general argument z is complex, equation 3.16
gives a real result for real-valued coefficients Aenm , Benm .
Those interested in the derivation of the above equation can find it in
Heiskanen and Moritz (1967) or other textbooks on potential theory.

^ 3.8.1 The scaling to standard form of the expansion

Assume Ae10 = Ae11 = Be11 = 0, meaning the vanishing of the dipole
moment.
We can also show that in expansion 3.16 the first coefficient is
GM⊕ E
Ae00 = Ae0 = arctan
E b
and the expansion specialised for a rotationally symmetric field becomes

∑︂
∞ ∑︂
∞ (︁ u )︁
Q n iE
Ψ(β, u) = Ψ
˜︁ n (β, u) = (︁ b )︁ Aen0 Pn (sin β). (3.17)
n=0 n=0
Q n iE

Also (︁ u )︁
Q 0 i E GM⊕ E
V
˜︁0 (u) = Ψ
˜︁ 0 (u) = (︁ b )︁ arctan ,
Q0 i E E b
the gravitational potential of the field constituent of ellipsoidal degree
zero.
With the substitutions (Heiskanen and Moritz, 1967, page 66)
u E b E
(︂ )︂ (︂ )︂
Q0 i = −i arctan u , Q0 i = −i arctan (3.18)
E E b
we obtain
V ˜︁ 0 (u) = GM⊕ arctan E .
˜︁0 (u) = Ψ (3.19)
E u

í Õ ! ¤. û
 Ellipsoidal harmonics
77
This corresponds to the “central field” of a spherical-harmonic expansion pallofunktio-
GM⊕ r . Using this, we may “scale” equation 3.16 by substituting the kehitelmä
/︁

above identities 3.18. The coefficients need to be divided by the constant

expression
GM⊕ E
arctan ,
E b
as the central field, expression 3.19, is moved outside the expansion.
The result is
GM⊕ E
V(β, λ, u) = arctan u ·
E
∑︂
∞ ∑︂n
arctan Eb Qnm i u
(︃ (︁ )︁ (︂ )︂)︃
E e e
· 1+ E
(︁ b )︁ Pnm (sin β) Cnm cos mλ + Snm sin mλ ,
n=2 m=0
arctan u
Q nm iE
e e
in which we have also introduced fully normalized coefficients Cnm , Snm
and Legendre functions Pnm (sin β).
This is an ellipsoidal-harmonic expansion that agrees with the ellipsoidi-
spherical-harmonic expansion 3.13, with the total mass of the Earth out- funktio-
kehitelmä
side the parentheses and the coefficients dimensionless. This equation
has apparently not been used for any geopotential determination.

^ 3.8.2 Equivalence of the Rapp and ellipsoidal expansions

We can demonstrate the equivalence of spherical expansions 3.13 or
3.14 and ellipsoidal expansion 3.16, if the flattening of the Earth → 0,
and thus also b → a, β → ϕ, and u → r. We assume that Heiskanen
and Moritz (1967) equation 1-112,

Qnm i u
(︁ )︁ (︂ )︂
a n+1
lim (︁ Eb )︁ = r
E→0 Qnm i
E

is valid. Substitution into equation 3.16 yields

V(u, β, λ) = V(r, ϕ, λ) =
∑︂∞ ∑︂ n (︂ )︂
a n+1
= r Pnm (sin ϕ) (Aenm cos mλ + Benm sin mλ) , (3.20)
n=0 m=0

í Õ ! ¤. û
78 Legendre functions and spherical harmonics

^ Table 3.6. Legendre functions of the second kind.

1 z+1
Q0 (z) = 2 ln
z−1 (n + 1) Qn+1 (z) − (2n + 1) zQn (z) + nQn−1 (z) = 0
1 z+1
Q1 (z) = 2 z ln −1
z−1
)︁ z + 1 3
Q2 (z) = 34 z2 − 14 ln
(︁
− z )︁ m/2 dm
z−1 2 Qnm (z) = 1 − z2
(︁
Qn (z)
(︁ 5 3 3 )︁ z + 1 5 2 2 dzm
Q3 (z) = 4 z − 4 z ln − z +
z−1 2 3

which, with the identifications Ae00 = GM⊕ a , Ae10 = Ae11 = Be11 = 0

/︁

and relations 3.15, suggests

{︄ }︄ {︄ }︄ {︄ }︄
Aen0 GM Jn0 GM √︁ Cn0
=− a⊕ = a ⊕ (2n + 1) ,
Ben0 Kn0 Sn0
{︄ }︄ {︄ }︄
Aenm GM Jnm
=− a⊕ =
Benm Knm
√︄ {︄ }︄
GM (n − m)! Cnm
= a⊕ 2 (2n + 1) if m ̸= 0.
(n + m)! Snm

Substituting these into equation 3.20 affirms its equivalence with equa-
pallofunktio tions 3.13 and 3.14 for spherical harmonics.

^ 3.8.3 Advantages of using ellipsoidal harmonics

◦ The expression for the normal gravitational potential is simple
in this form of presentation, see Heiskanen and Moritz (1967)
equation 2-56. A spherical-harmonic expansion of the same
field would instead require theoretically an infinite number of
coefficients. In practice, this number is only 3 to 4, so an expansion
up to J6 or J8 will suffice.
suppeneminen ◦ The convergence will be more rapid, as less terms are needed.
This is because, due to the Earth’s flattening, the equator is some
23 km further from the Earth’s centre than the poles. Therefore,

í Õ ! ¤. û
 Self-test questions
79
high-degree spherical harmonics in particular will have difficulty
converging efficiently both at the poles and in the equatorial suppeneminen
region. This problem is worst for very high-degree expansions
(for example Wenzel, 1998). Already for a degree number of 360,
the semi-wavelength of a spherical harmonic will be only 55 km! puoli-
aallonpituus
^ 3.8.4 Disadvantage of using ellipsoidal harmonics pallofunktio

Evaluation of an ellipsoidal-harmonic expansion is clearly more labori- ellipsoidi-

ous and expensive than that of a spherical-harmonic one, in terms of funktio-
kehitelmä
computer resources.

^ Self-test questions
1. How does separation of variables work?
2. Why does solving the Laplace equation require boundary condi-
tions?
3. What are the harmonic degree and harmonic order in a spherical- asteluku,
harmonic expansion? How do they relate to the resolution of the järjestysluku
expansion on the Earth’s surface?
4. What types of spherical harmonics are there? Explain their
dependence on latitude and longitude.
5. How many times does a surface spherical harmonic Ynm (ϕ, λ)
change its algebraic sign travelling along a meridian from the
south pole to the north pole? How many times when travelling
around the Earth along the equator?
6. What does it mean if it is said that two functions are mutually
orthogonal? Give a possible definition of the scalar product of two
functions.
7. How does the attenuation of spherical harmonics with height
behave? Why does a gravimetric satellite that is trying to map the

í Õ ! ¤. û
80 Legendre functions and spherical harmonics

gravity field of the Earth at a high resolution fly in as low an orbit

as possible?
asteosuus- 8. What does the degree constituent equation express?
yhtälö
9. Which spherical-harmonic coefficients are associated with the
pallofunktio-
dipole moment of the Earth’s mass distribution? Why are they
kerroin
missing from tableau 3.5?

^ Exercise 3 – 1: Attenuation with height of a

spherical-harmonic expansion
If
∑︂
∞ ∑︂
∞ (︂ )︂n+1
R
V(ϕ, λ, r) = V
˜︁n (ϕ, λ, r) =
r Vn (ϕ, λ),
n=0 n=0

we may call
˜︁n (ϕ, λ, r) (︂ R )︂n+1
V
= r
Vn (ϕ, λ)
the attenuation factor of the potential with height.
Differentiation with respect to r yields

∂V
˜︁n (ϕ, λ, r) n + 1 R n+2
(︂ )︂
=− r Vn (ϕ, λ), (3.21)
∂r R
or, because, at sea level, similarly
⃓
∂V˜︁n (ϕ, λ, r) ⃓⃓ n+1
=− Vn (ϕ, λ), (3.22)
∂r R
⃓
⃓
r=R

it follows that the attenuation factor for the attraction is the ratio of
expressions 3.21 and 3.22:
(︂ )︂n+2
R
r .
1. Draw a log-linear graph of the attenuation factors of both the
potential and the attraction for values n = 0, 1, 2, , . . . , 100, by
hand or by machine. Choose R = 6378 km, r = 7378 km — a
height 1000 km above the Earth’s surface.

í Õ ! ¤. û
 Exercise 3 – 2: Symmetries of spherical harmonics
81
2. Based on this, if the satellite is 1000 km above the Earth’s surface,
∂ ˜︁
for what degree number n will the accelerations ∂r Vn (ϕ, λ, r)
caused by the attraction at the satellite’s level be less than 1 % of
what they are on the Earth’s surface?
3. For what degree number n will they be less than 10−4 × of what
they are on the Earth’s surface?

^ Exercise 3 – 2: Symmetries of spherical harmonics

See equation 2.11. In it, Pnm (sin ϕ) = Pnm (t), t = sin ϕ is only a function
of latitude ϕ. When ϕ runs from the south pole through the equator
to the north pole, −90◦ ⩽ ϕ ⩽ +90◦ , parameter t will run through the
values −1 ⩽ t ⩽ +1 on the interval [−1, 1].
For the Legendre functions exists the closed expression 3.2:
)︁ m/2 dm
Pnm (t) = 1 − t2
(︁
P (t),
dtm n
in which the Pn (t) are ordinary Legendre polynomials:

1 dn (︁ 2
(︂ )︁n )︂
Pn (t) = n t −1 .
2 n! dtn
We can observe the following properties:
1. Differentiating a symmetric function of t will produce an antisym-
metric function, and vice versa.
(︁ )︁
2. The function t2 − 1 and its powers are symmetric.
3. Thus: for even n values, Pn (t) = +Pn (−t): Pn is symmetric
between the northern and southern hemispheres, and for odd
n values Pn (t) = −Pn (−t): Pn is antisymmetric between hemi-
spheres.
(︁ )︁
4. Similarly, for even n, Pn (sin ϕ) = +Pn sin(−ϕ) , and for odd n,
(︁ )︁
Pn (sin ϕ) = −Pn sin(−ϕ) .

í Õ ! ¤. û
82 Legendre functions and spherical harmonics

Questions
1. What is the corresponding rule for the functions Pnm , in
other words, for which (n, m) values is it symmetric and for
which values antisymmetric?
2. Fill in the diagram (n = 0, . . . , 5, m = 0, . . . , n) with either
‘S’ (symmetric) or ‘A’ (antisymmetric) in each framed cell:
n= 0 1 2 3 4 5
m=0
1
2
3
4
5 ×
3. What is the logic of symmetry?
4. If the field is mirror symmetric between the northern and
southern hemispheres, i. e., V(ϕ, λ, r) = V(−ϕ, λ, r), which
pallofunktio- of the spherical-harmonic coefficients anm , bnm drop out of
kerroin the series expansion? Why?
(Hint: see the example formulas and graphs for Pnm (sin ϕ)
in this chapter and try to guess a general rule. Then, verify.)
5. The same question if the potential is rotationally symmetric
about the Earth’s rotation axis: V(ϕ, λ, r) = V(ϕ, r).

^ Exercise 3 – 3: Algebraic-sign domains of spherical

harmonics
Legendren We have seen in section 3.1 that the associated Legendre functions Pnm (t)
liitännäis- have precisely n − m + 1 algebraic-sign intervals on their interval of
funktio ◦ ◦
definition ϕ ∈ [−90 , 90 ] . We can show that the functions cos mλ and
sin mλ each have 2m zero crossings and 2m algebraic-sign intervals
on their domain of definition λ ∈ [0, 360◦ ), assumed to form a closed

í Õ ! ¤. û
 Exercise 3 – 4: Escape velocity
83
circle. How many algebraic-sign domains — grey or white areas, visible or
occluded — are there in figure 3.3 for each surface spherical harmonic pinta-
⎧ pallofunktio
⎨P (sin ϕ) cos mλ if m ⩾ 0,
nm
Ynm (ϕ, λ) =
⎩Pn|m| (sin ϕ) sin |m| λ if m < 0

^ Exercise 3 – 4: Escape velocity

1. Given a spherically symmetric planet, mass GM, radius R, from
the surface of which a cannon shoots projectiles at flight velocity
v. What is the minimum value for v — the escape velocity — if it is
desired that the projectile can travel to arbitrarily large distances
from the planet and never fall back? The kinetic energy of the
projectile is Ekin = 12 mv2 , in which m is the projectile’s mass.
2. Given, in two-dimensional geometry, a circularly symmetric planet,
mass GM, radius R, with the gravitational field of the planet
represented by potential V as given in section 2.3, what does V
look like in terms of those parameters? Make an educated guess.
3. There is again a cannon on the edge of the circle planet. What can
you now say about the escape velocity v (do not try to compute
it!)?

í Õ ! ¤. û
^ The normal gravity field

4
^ 4.1 The basic idea of a normal field
Just as the figure of the Earth can be approximated by an ellipsoid of
revolution, the gravity field of the Earth can also be approximated by a
field of which one equipotential surface, or level surface, is precisely this
ellipsoid of revolution, the reference ellipsoid. vertaus-
ellipsoidi
This brings a logical idea to mind: why not define intercompatibly
a reference ellipsoid and a geopotential or normal potential, one of the
equipotential surfaces of which is the reference ellipsoid? After that,
a gravity formula is obtained by taking the gradient of this normal
potential.
After this we may define anomalous quantities, such as the disturbing
potential and the gravity anomaly, which then again will be intercom-
patible, while being numerically much smaller.
Let the normal potential be U(x, y, z). Then, normal gravity will be

∂U
γ (x, y, z) = ⃦→
⃦− ⃦
γ ⃦ = ∥∇U∥ = − →
⟨︁− ⟩︁
γ ·n =− ,
∂n
∂
in which ∂n denotes differentiation in the direction of the exterior
surface normal n to a level surface of the normal field, itself an ellipsoid
as well, see figure 4.1. This direction will differ from the direction of
the normal to the level surfaces of the gravity field, or plumb line, by luotiviiva

– 85 –
86 The normal gravity field

X Field lines
of the normal
n gravity field
Normal
gravity
n
→
−
γ
→
−
γ

→
−
γ Equipotential surfaces of
the normal gravity field
X
Reference ellipsoid
(flattening exaggerated)

^ Figure 4.1. The normal gravity field of the Earth.

luotiviivan precisely the plumb-line deflection. The deflection of the plumb line is
poikkeama also typically a very small angle.

We shall see in the next section that the pseudo-force generated by

the Earth’s rotation may, in a system rotating along with the Earth, be
described by a rotational potential Φ — also called centrifugal potential.
The normal potential U is also defined in such a way that the rotational
vertaus- potential Φ is included in it: the normal potential is the reference
potentiaali potential of the gravity field, not the gravitational field. If we denote
the normal gravitational potential using Ψ — a quantity rarely used in
geodesy — then the normal gravity potential or normal potential U is

U = Ψ + Φ,

í Õ ! ¤. û
 The centrifugal force and its potential
87
Z

p Centrifugal
force
k X
Gravitation
Gravity
i j
Y
X

^ Figure 4.2. Gravitation and centrifugal force.

in which Φ is the centrifugal potential. In other words: Ψ, like V, is

defined in a non-rotating (inertial) system, whereas U, like W, is defined
in a system that co-rotates with the Earth and is non-inertial. The word
gravity refers to a force acting in a co-rotating system, whereas in an
inertial system we use the word gravitation.

^ 4.2 The centrifugal force and its potential

The rotation of the Earth affects the gravity field. In an inertial reference vertaus-
system one speaks of gravitation and gravitational potential V. On the järjestelmä
Earth’s surface, however, in a non-inertial or co-rotating system, we talk
of gravity and gravity potential W. They are different things, and the
rotational motion and its centrifugal force are the cause of the difference.
See figure 4.2.
To derive the equation for centrifugal force, write first

p = Xi + Yj.
{︁ }︁
The vectors i, j, k form an orthonormal basis along the (X, Y, Z) axes. ortonormaali
kanta

í Õ ! ¤. û
88 The normal gravity field

It follows that √︂ √︁
p = ∥p∥ = ⟨p · p⟩ = X2 + Y 2 .
Now the centrifugal force — or rather, acceleration — is, in metres per
second squared,
f⊕ = ω2⊕ p = ω2⊕ (Xi + Yj) , (4.1)
with ω⊕ the rotation rate of the Earth in radians per second.
Here on Earth, gravity measurements are generally made using an
instrument that is at rest with respect to the Earth’s surface: it follows
the rotation of the Earth. If the instrument moves, one must, in addition
to the centrifugal force, take into account another pseudo-force: the
Coriolis1 force. Fluids — water, air — on the Earth’s surface, if they
are at rest, only sense the centrifugal force. Currents in addition also
sense the Coriolis force, which deflects them sideways and causes the
well-known eddy phenomena in the oceans and atmosphere.
We may describe centrifugal force as the gradient of a potential. If we
write for this centrifugal potential

Φ = 21 ω2⊕ X2 + Y 2 ,
(︁ )︁

we may directly calculate the gradient

∂Φ ∂Φ ∂Φ
f⊕ = ∇Φ = i+ j+ k=
∂X ∂Y ∂Z
= 12 ω2⊕ · 2X · i + 21 ω2⊕ · 2Y · j + 0 = ω2⊕ (Xi + Yj) ,

which corresponds to the above centrifugal-force equation.

If we add to the gravitational potential V the centrifugal potential Φ,
we obtain the gravity potential W:

W = V + Φ.

1 Gaspard-Gustave Coriolis (1792–1843) was a French mathematician, physicist and

mechanical engineer. His name is inscribed on the Eiffel Tower, Eiffel Tower, 72 names.

í Õ ! ¤. û
The centrifugal force and its potential
89
We may also derive from the centrifugal potential Φ the following
equation by differentiating it twice:

∂ 2 ∂ 2 ∂
∆Φ = ∇2 Φ = ⟨∇ · f⊕ ⟩ = ω⊕ X + ω⊕ Y + 0 = 2ω2⊕ , (4.2)
∂X ∂Y ∂Z
from which follows, with Poisson equation 1.14,

∆W = −4πGρ + 2ω2⊕ , (4.3)

the Poisson equation for the gravity potential.

The difference between gravitation and gravity is essential. The force,
g = ∇V is just an attractive force, whereas
or acceleration, of gravitation ˜︁
the acceleration of gravity g = ∇W is the vector sum of gravitation
and centrifugal force. Attraction and centrifugal force act in the same
fashion: the force is proportional to the mass of the test object. In other
words, the acceleration is always the same independently of the mass of
the test object. This is the famous equivalence principle (Galileo, Einstein),
which has been proven to hold to very great precision. We may mention
in particular the clever tests by the Hungarian baron Loránd Eötvös.2
Water masses on the Earth’s surface, as also the atmosphere — and on
a vastly longer time-scale also the “solid” Earth rock forming mountain
ranges and ocean depths — react to gravity without distinguishing
between attraction and centrifugal force. For this reason, the sea surface
coincides within a metre or so with an equipotential or level surface of the
W function or geopotential. Moreover, on dry land, we measure heights
from this surface, the geoid (according to Gauss, the “mathematical
figure of the Earth”).

2 Loránd baron Eötvös de Vásárosnamény (1848–1919) was a Hungarian physicist and

student of gravitation.

í Õ ! ¤. û
90 The normal gravity field

^ 4.3 Level surfaces and plumb lines

Surfaces of constant gravity potential or geopotential, equipotential
surfaces or level surfaces, are the following surfaces:

W(x, y, z) = constant.
{︁ }︁
ortonormaali Let i, j, k again be an orthonormal basis along the (x, y, z) axes. Then,
kanta in the direction of the unit vector

e = e1 i + e2 j + e3 k

the potential changes as follows:

∂W ∂W ∂W ∂W
= e1 + e2 + e3 ,
∂e ∂x ∂y ∂z

which vanishes if and only if

⟨e · ∇W⟩ = 0,

in other words, the potential is stationary only in directions that are

perpendicular to the Earth’s gravity vector

∇W = g.

Level surfaces and gravity vectors, or plumb lines, are always perpen-
dicular to each other.

^ 4.3.1 Curvature of level surfaces

Let there be given a plane that at P has the same direction as the level
surface: its tangent plane, figure 4.3. If the local curvature of the level
surface in the x direction is ρx and the x co-ordinate of point P is x0 , we
may develop the distance between the surfaces in a Taylor series:

1
ϵ≈ (x − x0 )2 .
2ρx

í Õ ! ¤. û
 Level surfaces and plumb lines
91
Tangent plane
P W = WP − δW

ϵ Equipotential
surface W = WP
x0 X
x
x axis

Radius of
curvature ρx

^ Figure 4.3. The curvature of level surfaces.

From this we obtain the difference in W values between the surfaces

(g = ∥g∥ = ∥∇W∥):
g
δW ≈ −ϵg ≈ − (x − x0 )2 .
2ρx

By differentiating (note that W here is now the geopotential on the

tangent or horizontal plane), we obtain

∂2 ∂2 g
2
δW = 2 W = ∂xx W = − ρ ,
∂x ∂x x

from which
g
ρx = − .
∂xx W
By determining the curvature in the x and y directions,

def 1 ∂ W def 1 ∂yy W

Kx = ρ = − xxg , Ky = ρ = − g , (4.4)
x y

í Õ ! ¤. û
92 The normal gravity field

we obtain the mean or Germain3 curvature, in most locations a positive

number:
Kx + Ky ∂xx W + ∂yy W
J= =− ,
2 2g
and by using Poisson equation 4.3,

∆W = ∂xx W + ∂yy W + ∂zz W = −4πGρ + 2ω2⊕ ,

we obtain
−2gJ + ∂zz W = −4πGρ + 2ω2⊕ .
By using
∂g ∂g
∂zz W = − =− ,
∂z ∂H
in which H is the height co-ordinate, we obtain for the vertical gradient
of gravity (Heiskanen and Moritz, 1967, equation 2-20):
∂g
= −2gJ + 4πGρ − 2ω2⊕ ,
∂H
an equation found by Ernst Heinrich Bruns (Bruns, 1878, page 13).
We also showed the compact Euler notation for partial derivatives,
∂xx , ∂yy , ∂zz , which is often convenient.

^ 4.3.2 Curvature of plumb lines

luotiviiva Plumb lines are curved because gravity is not constant in the horizontal
direction. If gravity increases in a horizontal direction, then the equi-
potential surfaces will come closer together too, and they will not be
parallel. This means that the plumb lines, being perpendicular to all
equipotential surfaces, must be curved in that direction.
Consider two equipotential surfaces, one for potential WP and one for
potential WP + ∆W. The distance separating them will be ∆H = ∆W g .
/︁

3 Marie-Sophie Germain (1776–1831) was a brilliant French mathematician, number

theorist and student of elasticity. She corresponded with Gauss on number theory
(Friedelmeyer, 2014) and did foundational work towards a proof of Fermat’s last
theorem. Her name is missing from the Eiffel Tower.

í Õ ! ¤. û
 Level surfaces and plumb lines
93
Plumb line
Radius of curvature ρx
P
WP
W=
g
∆W g
/︁
g

∆W x −→
+
W = WP
^ Figure 4.4. The curvature of the plumb line.

In the direction of co-ordinate x the relative tilt between the two surfaces
will be (︃ )︃
∂ ∂ ∆W ∆W ∂g
∆H(x) = =− 2 .
∂x ∂x g(x) g ∂x
If the starting distance between the surfaces is ∆H, it will take a distance
of
)︂/︄(︃
∆W
/︂ (︂ )︃ /︂
ρx = − ∆H ∂ ∆W ∂g g ∂g
∆H = − g − 2 =
∂x g ∂x ∂x
to bring the tangents together, see figure 4.4. The curvature of the
plumb line is the inverse of this, in both the x and the y co-ordinate luotiviiva
directions:
1 1 ∂g 1 1 ∂g
κx = ρ = g , κy = ρ = g .
x ∂x y ∂y
We can derive the curvature of the field lines of the normal gravity field
in the same way. The difference is, however, that we can find a simple
mathematical expression for gravity on the surface of the reference vertaus-
ellipsoid, for example equation 4.6. A good approximation is ellipsoidi

γ(φ) ≈ γa cos2 φ + γb sin2 φ.

With the chain rule

∂γ ∂γ ∂φ 1 ∂γ 1
= = = (−2γa cos φ sin φ + 2γb sin φ cos φ) =
∂x ∂φ ∂x R ∂φ R
γ − γa
= b sin 2φ.
R

í Õ ! ¤. û
94 The normal gravity field

Astronomical
co-ordinates Φ, Λ

Plumb line
n

Greenwich

n
O Φ

Figure 4.5. Natural co-ordinates Φ and Λ. In addition, a natural height

^ co-ordinate, for example the geopotential W, is needed.

This means in the x or south-north and y or west-east direction:

1 ∂γ 1 γb − γa 1 ∂γ
κ∗x = γ ≈ sin 2φ, κ∗y = γ = 0.
∂x R γa ∂y

^ 4.4 Natural co-ordinates

Before the satellite era it was impossible to directly measure the geocen-
tric co-ordinates X, Y and Z. Today this is possible, and we obtain at the
vertaus- same time the height from the reference ellipsoid h, a purely geometric
ellipsoidi quantity.

luotiviiva In earlier times, one could measure only the direction of the plumb
line as shown in figure 4.5, as well as the potential difference between
an observation point and sea level by levelling. The direction of the

í Õ ! ¤. û
 The normal potential in ellipsoidal co-ordinates
95
plumb line n was measured astronomically: astronomical latitude Φ
(not to be confused with the centrifugal potential) and astronomical
longitude Λ. The third co-ordinate, the gravity potential difference
W(x, y, z) − W0 from the potential W0 of sea level, was determined by
levelling. Co-ordinates Φ, Λ and W are called natural co-ordinates.
Often, instead of the potential, orthometric height H is used. Its
definition is easy to understand if one writes
w WP
∂W 1 1
= −g =⇒ dH = − g dW =⇒ HP = − dW, (4.5)
∂H W0 g (W)
∂
in which the integral is taken along the plumb line of point P. ∂H is luotiviiva
the derivative in the direction of the plumb line, the local normal to
the level surfaces. g is the acceleration of gravity along the plumb
line as a function of place — or of geopotential level. In this case
of orthometric heights, g is the true gravity inside the rock, which
is a non-linear function of place and will also depend on the rock
density. This trickiness of their determination is a problem specific to
orthometric heights. We will return to this later (Heiskanen and Moritz,
1967, chapter 4).
The co-ordinates Φ, Λ and H also form a natural co-ordinate system.

^ 4.5 The normal potential in ellipsoidal co-ordinates

We have already presented equation 3.16 for the expansion of the
geopotential into ellipsoidal harmonics. It is required of the normal ellipsoidi-
potential U that it is constant on the reference ellipsoid u = b. We funktio
expand the centrifugal potential Φ into ellipsoidal harmonics, obtaining vertaus-
ellipsoidi

Φ(β, u) = 12 ω2⊕ x2 + y2 = 12 ω2⊕ u2 + E2 cos2 β =

(︁ )︁ (︁ )︁

= 21 ω2⊕ u2 + E2 1 − sin2 β =
(︁ )︁ (︁ )︁

= 21 ω2⊕ u2 + E2 − 32 P2 (sin β) + 23 P0 (sin β) =

(︁ )︁ (︁ )︁

= − 13 ω2⊕ u2 + E2 P2 (sin β) − P0 (sin β) .

(︁ )︁ (︁ )︁

í Õ ! ¤. û
96 The normal gravity field

In addition, based on equation 3.17 we have for the rotationally sym-

metric normal gravitational potential Ψ:
∑︂
∞ ∑︂
∞ (︁ u )︁
Q n iE
Ψ(β, u) = Ψ
˜︁ n (β, u) = (︁ b )︁ Aen Pn (sin β).
n=0 n=0
Q n iE

Now
U(β, u) = Ψ(β, u) + Φ(β, u).
vertaus- On the reference ellipsoid u = b we have as a requirement U(β, b) = U0 ,
ellipsoidi which is possible only if

U0 = Ae0 + 13 ω2⊕ b2 + E2 = Ae0 + 13 ω2⊕ a2 ,

(︁ )︁

0 = Ae1 ,
0 = Ae2 − 31 ω2⊕ b2 + E2 = Ae2 − 13 ω2⊕ a2 ,
(︁ )︁

0 = Aen , n = 3, 4, 5, . . . .

The quantity U0 can be computed uniquely, if the Earth’s mass GM⊕

and the measures of the reference ellipsoid a, b are known. The result,
given in Heiskanen and Moritz (1967) as equation 2-61, is
GM⊕ E
U0 = arctan + 13 ω2⊕ a2 .
E b
From this follows
GM⊕ E
Ae0 = U0 − 13 ω2⊕ a2 = arctan .
E b
The normal gravity potential U is obtained as follows (remember the
identities 3.18):
Ψ
˜︁ (u)
⏟ ⏞⏞
0
⏟
GM⊕ E
U(β, u) = Ψ(β, u) + Φ(β, u) = arctan u +
E
Ae P (sin β) Φ(β,u)
⏟ ⏞⏞2 ⏟ Q (︁i u )︁ ⏟(︁ 2 ⏞⏞ ⏟ ⏟ ⏞⏞ )︁ ⏟
1 2 2 2 E 3 2 1
)︁ 1 2
(︁ 2 2 2
+ 3 ω⊕ a (︁ )︁ sin β − 2 + 2 ω⊕ u + E cos β =
Q2 i Eb 2
= C0 (u) + C1 (u) sin2 β + C2 (u) cos2 β,

í Õ ! ¤. û
 Normal gravity on the reference ellipsoid
97
in which C0 , C1 , and C2 are suitable functions of u. The function Ψ
˜︁ 0 is
the term for n = 0 in expansion 3.17.
On the surface of the reference ellipsoid (u = b), using a2 = b2 + E2 :

U(β, b) =
Ψ
˜︁ (b)
⏟ ⏞⏞
0
⏟ ⏟ Ae2 P2 (sin β)
⏞⏞ ⏟ ⏟
Φ(β,b)
⏞⏞ ⏟
GM⊕ E 1 2 2 2 1 2 2 1 2 2 2
= arctan + 2 ω⊕ a sin β − 6 ω⊕ a + 2 ω⊕ a cos β =
E b
GM⊕ E
= arctan + 13 ω2⊕ a2 ,
E b
the constant U0 , as it had better be!

^ 4.6 Normal gravity on the reference ellipsoid

∂
Without proof, we mention that for normal gravity (the quantity γ = ∂h
U)
the following equation applies on the reference ellipsoid: vertaus-
ellipsoidi
aγ sin2 β + bγa cos2 β
γ(β) = √︁b .
a2 sin2 β + b2 cos2 β

By substitution we find immediately that γa is normal gravity on the

equator (β = 0) and γb normal gravity on the poles (β = ±90◦ ).
Equations 2.5 and 2.7 yield

Z b
/︁
sin β a Z a
tan β = = √ 2 /︁ = b √ 2 = tan ϕ
cos β X +Y a2 X +Y 2 b

and
/︂(︁
Z 1 − e2 N
)︁
sin φ 1 Z a2
tan φ = cos φ = √ 2 = 2
√ = tan ϕ,
X + Y2 N
/︁ 1 − e X2 + Y 2 b2

in which ϕ is the geocentric latitude, see equations 2.4. From this follows
directly
b
tan β = a tan φ,

í Õ ! ¤. û
98 The normal gravity field

Z
Q

b P

β
a ϕ X/Y
O

^ Figure 4.6. Geometry of the meridian ellipse and various types of latitude.

in which the latitude angle φ is the geodetic or geographic latitude. β

redukoitu is still the reduced latitude. Now it can be shown (exercise!) that
leveysaste
aγ cos2 φ + bγb sin2 φ
γ(φ) = √ a . (4.6)
a2 cos2 φ + b2 sin2 φ
This is the famous Somigliana–Pizzetti4 equation. These geodesists
demonstrated for the first time that an “ellipsoidal” normal gravity
vertaus- field which has the reference ellipsoid as one of its equipotential or
ellipsoidi level surfaces exists exactly, and that the gravity formula is also a closed
expression in geographic latitude.

4 Carlo Somigliana (1860–1955) was an Italian mathematician and physicist. Paolo

Pizzetti (1860–1918) was an Italian geodesist.

í Õ ! ¤. û
 Numerical values and calculation formulas
99
^ 4.7 Numerical values and calculation formulas
When the reference ellipsoid has been chosen, we may calculate the nor-
mal potential and normal gravity corresponding to it. The fundamental
quantities are
a the equatorial radius of the ellipsoid of revolution, its semimajor
axis
f the flattening
defa−b
f= a ,
in which b is the polar radius or semiminor axis
ω⊕ the rotation rate of the Earth
GM⊕ the total mass of the Earth, including the atmosphere.

Nowadays the most commonly used reference ellipsoid cum normal vertaus-
potential is the GRS80, the Geodetic Reference System 1980: ellipsoidi

a = 6 378 137 m, ω⊕ = 7 292 115 · 10−11 s−1 ,

1
= 298.257 222 101, GM⊕ = 3 986 005 · 108 m3/s2 .
f
In reality, f is not a defining constant of GRS80, but the constant J2 is
used instead, which is a defining quantity for the gravitational field, see
equation 3.14.
The WGS84 (World Geodetic System 1984) used by the GPS system has
a reference ellipsoid that is almost identical to that of the GRS80.

The normal potential is (Heikkinen, 1981), in SI units:

U = 62 636 860.8500 +
(︄ )︄
− 9.780 326 77 − 0.051 630 75 sin2 φ −
+ h+
− 0.000 227 61 sin4 φ − 0.000 001 23 sin6 φ
(︄ )︄
+ 0.015 438 99 · 10−4 − 0.000 021 95 · 10−4 sin2 φ −
+ −4 4
h2 +
− 0.000 000 10 · 10 sin φ
+ − 0.000 024 22 · 10−8 + 0.000 000 07 · 10−8 sin2 φ h3 ,
(︁ )︁

í Õ ! ¤. û
100 The normal gravity field

and normal gravity (note the minus sign, U is positive and diminishes
going upwards):

∂U
γ=− = + 9.780 326 77 + 0.051 630 75 sin2 φ +
∂h
+ 0.000 227 61 sin4 φ + 0.000 001 23 sin6 φ +
(︄ )︄
+ 0.030 877 98 · 10−4 − 0.000 043 90 · 10−4 sin2 φ −
− h+
− 0.000 000 20 · 10−4 sin4 φ
− − 0.000 072 65 · 10−8 + 0.000 000 21 · 10−8 sin2 φ h2 . (4.7)
(︁ )︁

Here, the unit of potential is m2/s2 , and the unit of gravity, m/s2 . φ is
vertaus- geodetic latitude; h (in metres) is the height above the reference ellipsoid.
ellipsoidi More precise equations can be found from Heikkinen (1981). In these
equations, the coefficient 9.780 32 . . . m/s2 is equatorial gravity, and the
value −0.030 87 . . . s−2 is the vertical gradient of gravity on the equator.
Other gravity formulas and reference ellipsoids still in legacy use (and
slowly vanishing) are Helmert’s 1906 ellipsoid, the Krasovsky ellipsoid
or SK-42 in Eastern Europe, the International or Hayford ellipsoid (1924)
and its gravity formula, and the Geodetic Reference System 1967.

^ 4.7.1 Numerical example

According to the above equation, the normal potential over the equator
is

U = 62 636 860.8500 − 9.780 326 77 h + 0.015 438 99 · 10−4 h2 −

− 0.000 024 22 · 10−8 h3 .

◦ Draw this function for values of h in the range 0–7000 km.

◦ Draw for comparison the quadratic version, from which the last
term is left off.
Questions
1. What is the minimum of the quadratic function?

í Õ ! ¤. û
 The normal potential as a spherical-harmonic expansion
101
X
Cubic
80 000 000 Quadratic
Linear
60 000 000 Realistic

40 000 000

20 000 000

0 1000 2000 3000 4000 5000 6000 7000

Figure 4.7. The normal field’s potential over the equator. Heights in kilometres,
^ potential in m2/s2 .

2. How physically realistic is this?

Answers
1. See figure 4.7. The minimum of the quadratic function
is at height 3000 km. The cubic function does not have a
minimum.
2. Not very physical: the stationary point for potential U (the
normal potential in a co-rotating system) should be located at
approximately 36 000 km height, at the geostationary orbit.
This tells us that polynomial approximation cannot be extrapo-
lated very far. In this case, the interval of extrapolation is of the
same order as the radius of the Earth, and that will no longer
work.

^ 4.8 The normal potential as a spherical-harmonic

expansion
The spherical-harmonic expansion of an ellipsoidal gravitational field pallofunktio-
kehitelmä

í Õ ! ¤. û
102 The normal gravity field

^ Table 4.1. GRS80 normal potential spherical-harmonic coefficients (Heikkinen,

1981; Heiskanen and Moritz, 1967).

Non-normalised Fully normalised

J2 = J2,0 = 1082.63 · 10−6 J2 = −C2,0 = 484.166 854 896 · 10−6

J4 = J4,0 = −2.370 912 22 · 10−6 J4 = −C4,0 = −0.790 304 073 · 10−6
J6 = J6,0 = +0.006 083 47 · 10−6 J6 = −C6,0 = +0.001 687 251 · 10−6
J8 = J8,0 = −0.000 014 27 · 10−6 J8 = −C8,0 = −0.000 003 461 · 10−6

also contains, besides the second-degree harmonic, higher-degree

harmonics. If we write, as is customary, the potential outside the Earth
in the following form (Heiskanen and Moritz, 1967 equation 2-39, also
equation 3.14):

GM⊕
(︃ ∑︂
∞ (︂ )︂ ∑︂
a n
n )︃
V(ϕ, λ, r) = r 1− r Pnm (sin ϕ) (Jnm cos mλ + Knm sin mλ) ,
n=2 m=0

then we may also write the normal gravitational potential, Ψ, into the
form
GM⊕
(︃ ∑︂
∞ (︂ )︂n
a
)︃
Ψ(ϕ, r) = r 1− Jn r Pn (sin ϕ) ,
n=2
even

which contains only even coefficients Jn = Jn0 , because the normal field
is symmetric about the equatorial plane.
The coefficients for the GRS80 normal gravitational potential are
found5 in table 4.1. Higher terms are usually not needed. The rela-
tionship between fully normalised and non-normalised coefficients is
√
Jn = Jn 2n + 1.

5 They can also be calculated from equation (2-92) given in Heiskanen and Moritz
(1967): (︁ )︁n
3 e2
(︃ )︃
J2
J2n = (−1)n+1 1 − n + 5n 2 ,
(2n + 1) (2n + 3) e
starting from the values J2 and e2 . The results are the same as in the table’s left
column.

í Õ ! ¤. û
 The disturbing potential
103
For comparison: in section 4.5 it was shown that in the expansion
of the same field into ellipsoidal harmonics, only the degree-zero and ellipsoidi-
degree-two coefficients are non-zero! This is one reason why these funktio
functions are used at all.
Instead of using an ellipsoidal model, we may also use as a normal
gravity potential formula the first two or three terms of the spherical-
harmonic expansion of the real geopotential. Then we obtain, taking
the centrifugal potential along:

Y Y (ϕ, λ) 1 2 (︁ 2
U = r0 + 2 3 + ω⊕ X + Y 2 ,
)︁
r 2
with the corresponding equipotential surface U = U0 being the “Bruns
spheroid”, or

Y Y (ϕ, λ) Y4 (ϕ, λ) 1 2 (︁ 2
U = r0 + 2 3 + ω⊕ X + Y 2 ,
)︁
+
r r5 2
def
the “Helmert spheroid”. Here, Y0 = GM⊕ while Y2 (ϕ, λ) and Y4 (ϕ, λ)
are taken from the true geopotential.
These equations are easy to compute, but their equipotential or level
surfaces are not ellipsoids of revolution, and in fact not even rotationally
symmetric. They are quite complicated surfaces (Heiskanen and Moritz,
1967, section 2-12)!
However, in geometric geodesy we always use a reference ellipsoid, vertaus-
so this is also a wise thing to do in physical geodesy. ellipsoidi

^ 4.9 The disturbing potential

Write the gravity potential

W = V + Φ,

in which Φ is the centrifugal potential (see above), and the normal

potential
U = Ψ + Φ.

í Õ ! ¤. û
104 The normal gravity field

The difference between them is the disturbing potential.

def
T = W − U = V − Ψ.

pallofunktio Both V and Ψ can be expanded into spherical harmonics. If we write

the gravity potential

GM
W =V +Φ=Φ+ r ⊕ ·
(︃ ∑︂
∞ (︂ )︂ ∑︂
a n
n )︃
· 1− r Pnm (sin ϕ) (Jnm cos mλ + Knm sin mλ) ,
n=2 m=0

and the normal potential

GM
(︃ ∑︂
∞ (︂ )︂
a n ∗
)︃
U=Φ+ r ⊕ 1− r Jn Pn (sin ϕ) ,
n=2
even

we obtain by subtraction for the disturbing potential

GM
T =W−U=− r ⊕ ·
∞ (︂ )︂ ∑︂
(︃∑︂ n )︃
a n
· r Pnm (sin ϕ) (δJnm cos mλ + Knm sin mλ) ,
n=2 m=0

in which
⎧
⎨δJ = Jn0 − J∗n if n even,
n0
⎩δJnm = Jnm otherwise.

The above equation for the disturbing potential T is shortened as follows

(Heiskanen and Moritz, 1967, equation 2-152):
∑︂
∞ (︂ )︂
a n+1
T (ϕ, λ, r) = r Tn (ϕ, λ), (4.8)
n=2

asteosuus where, in every term, the degree constituent Tn has the same dimension
as T , and

GM⊕ ∑︂
n
Tn (ϕ, λ) = − a Pnm (sin ϕ) (δJnm cos mλ + Knm sin mλ) .
m=0

í Õ ! ¤. û
 Self-test questions
105
On the surface of the reference sphere of radius a:6 vertauspallo
∑︂
∞
T (ϕ, λ) = Tn (ϕ, λ),
n=2

from which we see that on the reference level, the terms Tn (ϕ, λ) are
really the degree constituents of the disturbing potential T for a certain asteosuus
degree number n.
The above expansions are all missing the terms n = 0, 1. Of these,
T0 (ϕ, λ) = T0 is a constant — the global average of the disturbing
potential — and T1 (ϕ, λ) has the form of a dipole field. Its value is
proportional to the cosine of the angle between the geocentric location
vector of the point of calculation and that of the dipole vector. Both
vanish because it is assumed that
◦ the total mass of the Earth GM⊕ assumed by the normal field is
realistic
◦ the origin of the co-ordinate reference system is assumed to be at
the centre of mass of the Earth.
See section 3.4 for more.

^ Self-test questions
1. What is the basic idea behind using a normal gravity field?
2. What is the difference between gravity and gravitation?
3. Given the centrifugal potential

Φ = 21 ω2 X2 + Y 2 ,
(︁ )︁

derive the centrifugal acceleration as a vector. X, Y, Z are rectan-

gular co-ordinates of a frame rotating at angular rate ω around
the Z axis.
6 Earlier
on we also used for this reference radius (in spherical approximation) the
symbol R.

í Õ ! ¤. û
106 The normal gravity field

4. Explain the idea of natural co-ordinates.

5. What was the relationhip between M. Le Blanc and C. F. Gauss?
Use Google.
6. What makes Somigliana–Pizzetti equation 4.6 valuable?
7. What are the defining parameters of the Geodetic Reference
System 1980?
pallofunktio- 8. Why does the spherical-harmonic expansion of the normal poten-
kehitelmä tial contain only a small number of terms and coefficients?
9. Why does the spherical-harmonic expansion of the normal poten-
järjestysluku tial not contain any terms of order m ̸= 0?
10. Why does the spherical-harmonic expansion of the normal poten-
asteluku tial contain only terms with even degree numbers n?

^ Exercise 4 – 1: The Somigliana–Pizzetti equation

1. Given gravity on the equator γa and on the poles γb , what is the
gravity on geodetic latitude φ = 45◦ ? Derive an expression that
may also contain a and b.
2. And what is the gravity on the reduced latitude β = 45◦ ? Compare
with the previous.
3. Given are the semimajor axis a and semiminor axis b. What are
the differences, for the same point, between the different latitudes
(geodetic φ, geocentric ϕ, and reduced β) at most, in minutes of
arc? Assume that the maximum happens at latitudes ±45◦ .
redukoitu 4. Compute for both a geodetic and a reduced latitude of 45◦ numer-
leveysaste ical values of gravity for the case of the GRS80 reference ellipsoid.
vertaus- By how much do they differ?
ellipsoidi

í Õ ! ¤. û
 Exercise 4 – 2: Centrifugal force
107
^ Exercise 4 – 2: Centrifugal force
Given is the rotation rate of the Earth in radians per second: ω⊕ =
7292 115 · 10−11 s−1 .
1. Compute (roughly) the centrifugal force caused by the Earth’s
rotation at Southern Finland (φ = 60◦ , R = 6378 km, spherical
Earth). In what direction does the force point (sketch!)?
2. How much does the centrifugal force contribute to local gravity,
i. e., by how much does it change gravity, both as an acceleration
and as a percentage?
3. Compute from the ω⊕ value given above, the rotation time of the
Earth in hours. Why is it not precisely 24h ?

í Õ ! ¤. û
^ Anomalous quantities of the
gravity field

^ 5.1 Disturbing potential, geoid height, deflections of

5
the plumb line
The first anomalous quantity, which we already discussed above, is the
difference between the true gravity potential W and the normal gravity
potential U, the disturbing potential:
def
T = W − U.

All other anomalous quantities are various functions of the disturbing

potential, such as the geoid height N and the plumb-line deflections luotiviivan
ξ, η. They are generally obtained by subtracting from each other poikkeama

◦ a natural quantity related to the Earth’s real gravity field, and

◦ a corresponding quantity related to the normal gravity field of
the reference ellipsoid of the Earth. vertaus-
ellipsoidi
For example, deflections of the plumb line
def def
ξ = Φ − φ, η = (Λ − λ) cos φ,

in which (Φ, Λ) are astronomical latitude and longitude, that together

make up the direction of the local plumb line, and the geodetic latitude
and longitude (φ, λ), that similarly make up the direction of the normal

– 109 –
110 Anomalous quantities of the gravity field

Plumb-line
deflections ξ, η
Topography

Geoid Geoid height N

Reference ellipsoid

^ Figure 5.1. Geoid undulations and deflections of the plumb line.

gravity vector or “normal plumb line”.1 See figure 5.1.

The geoid height or geoid undulation is
def
N = H − h,

in which H is the orthometric height — reckoned from mean sea level —

vertaus- and h the height above the reference ellipsoid.
ellipsoidi
Deflections of the plumb line in Finland are a few seconds of arc ( ′′ ) in
luotiviivan magnitude. Geoid undulations range from 15 to 32 m (for comparison,
poikkeama
globally the range of variation is −107 m to +85 m), relative to the GRS80
ellipsoid as is today customary. At sea level, the plumb-line deflections
— expressed in radians! — equal the horizontal gradients of the geoid
undulation. See figures 5.1 and 5.2.
For any reference ellipsoid, for example the GRS80 ellipsoid, there
exists its own mathematically exact standard or normal gravity field,
of which one equipotential or level surface is precisely that reference
ellipsoid. Using this field, we may calculate for each gravity field

1 Thisholds exactly only on the reference ellipsoid. Elsewhere one should add to φ a
correction for the curvature of the field lines of the normal gravity field or “normal
plumb lines”, see figure 4.1. The correction is δφn = 0.171 ′′ km−1 · h sin 2φ, with h
the height in kilometres from the reference ellipsoid (Heiskanen and Moritz, 1967,
equation 5-34). See also section 4.3.2.

í Õ ! ¤. û
Disturbing potential, geoid height, deflections of the plumb line
111
quantity the corresponding normal quantity. By subtracting the normal
quantity from the original one, we obtain the corresponding anomalous
quantity.
For heights above the reference ellipsoid there exists an expression vertaus-
analogous to expression 4.5 for orthometric heights. In the expression, ellipsoidi
U is the normal potential and γ normal gravity:2
w UP
1
hP = − dU.
U0 γ (U)

The geoid height in point P is now

w WP w UP
1 1
NP = hP − HP = dW − dU =
W0 g(W) U0 γ(U)
w WP w WP w UP w U0
1 1 1 1
= dW − dU − dU + dU
W0 g(W) W0 γ(U) WP γ(U) W0 γ(U)
w WP γ(W ′ ) − g(W ′ ) w UP w U0
′ 1 1
= dW − dU + dU
W0 g(W ′ )γ(W ′ ) WP γ(U) W0 γ(U)
w HP γ(z) − g(z) w UP w U0
1 1
= dz − dU + dU, (5.1)
0 γ(z) WP γ(U) W0 γ(U)

by re-naming the integration variables W, U → W ′ and changing it to a

length: dW ′ = −g dz.
In equation 5.1 the last term vanishes if we assume3 U0 = W0 . The
first and second terms both depend on the height of point P, but their
sum NP does not. Therefore we place point P at mean sea level — the
zero point of the height system used. It follows that the first term also
vanishes: HP = 0. So
w UP
1 1 (︁ )︁ T
NP = − dU ≈ γ WP − UP = γP ,
WP γ(U) P P

2 This
is not exactly true, due to the “normal plumb line” not being the same as the
normal on the reference ellipsoid. The error made is tiny.
3 This is not self-evident!
In a local vertical datum the potential of the zero point could
well differ by as much as the equivalent of a metre from the normal potential of a
global reference ellipsoid.

í Õ ! ¤. û
112 Anomalous quantities of the gravity field

Figure 5.2. A geoid model for Finland from 1984. Deflections of the plumb
^ line computed from observations in red (Vermeer, 1984).

where we have substituted T = W − U, the disturbing potential. All

quantities are assumed to be at sea level. More compactly:

í Õ ! ¤. û
 Disturbing potential, geoid height, deflections of the plumb line
113

−g = − grad W −→
−
γ = − grad U

P
WP Geoid

UP
N Ellipsoid
UQ (= WP )
Q

Figure 5.3. Equipotential surfaces of the gravity field (W) and the normal
^ gravity field (U).

N= T γ.
/︁
(5.2)

This is the famous Bruns4 equation (Heiskanen and Moritz, 1967,

equation 2-144).
Figure 5.3 depicts the situation still better. In this figure, the normal
gravity vector →
−
γ = grad U has a length of ∂h ∂
U, from which it follows,
with equation T = W − U, that the separation between “matching”
surfaces W = WP and U = UQ , when WP = UQ , is

UQ − UP W −U T
N≈ γ = P γ P = γ.

4 Ernst
Heinrich Bruns (1848–1919) was an eminent German mathematician and
mathematical geodesist.

í Õ ! ¤. û
114 Anomalous quantities of the gravity field

^ 5.2 Gravity disturbances

The difference between the true and normal gravity accelerations is
painovoima- called the gravity disturbance,
häiriö
∂W ∂U
δg = g − γ = ∥g∥ − ⃦→
def ⃦− ⃦ (︂ )︂
γ⃦ ≈ − − ,
∂H ∂h
luotiviiva in which the differentiation is done along the plumb line for W and —
vertaus- slightly imprecisely — along the normal on the reference ellipsoid for
ellipsoidi U. The directions of the plumb line and surface normal on the ellipsoid
are actually very close to each other. Therefore, to good approximation

∂W ∂U ∂T
(︂ )︂
δg ≈ − − =− .
∂H ∂H ∂H
In spherical approximation we have

∂T
δg ≈ − . (5.3)
∂r
asteosuus We already expanded the disturbing potential T into constituents for
different spherical-harmonic degree numbers, equation 4.8, and now
we obtain by differentiating with respect to r:

∂ ∑︂ R n+1
(︃ ∞ (︂ )︂ )︃
∂T (ϕ, λ, r)
δg(ϕ, λ, r) = − =− r Tn (ϕ, λ) =
∂r ∂r
n=2
∑︂
∞
n + 1 R n+1
(︂ )︂
= r r Tn (ϕ, λ), (5.4)
n=2

and at sea level (r = R):

∑︂
∞
n+1
δg(ϕ, λ, R) = Tn (ϕ, λ).
R
n=2

This is the spectral representation of the gravity disturbance at sea level,

i. e., on an Earth sphere of radius R. As a suitable value for the reference
radius R one may take the equatorial radius a⊕ of a reference ellipsoid
for the Earth.

í Õ ! ¤. û
 Gravity anomalies
115
We can observe gravity disturbances only if, in addition to measuring
∂
(︁ ⃓ )︁
the acceleration of gravity gP = ∂H W ⃓P at a point P, we have a way
to measure P’s location in space, relative to the geocentre, so one may
∂
⃓
calculate normal gravity γP = ∂h U⃓P at the same point. Nowadays this
is even easy using GNSS, but traditionally it has been impossible. For
this reason, gravity disturbances are little-used. One rather uses gravity
anomalies, about which more below.

^ 5.3 Gravity anomalies

Normal gravity is calculated as a function of location expressed in
geodetic co-ordinates (φ, λ, h). However, in traditional gravimetric field
work, before the satellite positioning era, one only had access to the
geodetic co-ordinates φ and λ, but not to any accurate height h above
the reference ellipsoid. One only had access to the height H above sea vertaus-
level (the geoid), determined, for example, through a national levelling ellipsoidi
network — or, in the worst case, barometrically.
This means that, although the true gravity g is measured at point P,
the height of which above sea level is HP , the normal gravity γ must of
necessity be calculated at another point Q, the height of which above the
reference ellipsoid is hQ = HP . See figure 5.4.
In other words, the measured height of point P above mean sea level is
substituted, brute-force style, into the normal gravity formula, which,
however, expects a height above the reference ellipsoid! This special trait of
the definition of gravity anomalies may be called a “free boundary-value reuna-arvo-
problem”. tehtävä

According to this, we calculate gravity anomalies as follows:

∆gP = gP − γQ = (gP − γP ) + (γP − γQ ) =

∂W ⃓ ∂U ⃓
(︂ ⃓ ⃓ )︂
=− ⃓ − ⃓ + (γP − γQ )
∂H P ⃓ ∂h P ⃓
∂ (W − U) ⃓⃓ ∂γ ⃓⃓
≈− ⃓ + (hP − hQ ) ∂H ⃓ =
∂H P P

í Õ ! ¤. û
116 Anomalous quantities of the gravity field
⃓
∂T ⃓
⃓ (︁ )︁ ∂γ ⃓
=− ⃓ + hP − HP ⃓ =
∂H P ∂H ⃓P
⃓ (︃ )︃⃓
∂T ⃓ ∂γ ⃓⃓ ∂T T ∂γ ⃓⃓
⃓
=− ⃓ + NP = − + ,
∂H P ∂H ⃓P ∂H γ ∂H ⃓P

using almost all the equations above. This equation,

∂T 1 ∂γ
∆g = − +γ T, (5.5)
∂H ∂H

fysikaalisen is known as the fundamental equation of physical geodesy. It is the boundary

geodesian condition of the third boundary-value problem (Heiskanen and Moritz,
perusyhtälö
1967, section 1-17). It enables the solution of T in the exterior space, if
reuna-arvo-
∆g is given everywhere on the Earth’s surface.
tehtävä
If we assume that the Earth is a sphere of radius R and that the normal
gravity field is spherically symmetric, we may approximate:
∂T 2
∆g = − − rT, (5.6)
∂r
in which r = R + H is the distance from the Earth’s centre.
By substituting into this equation 5.3 for δg we obtain
2
∆g = δg − r T .

From this we obtain by substituting the above spectral representations

4.8 and 5.4 for T and δg:
∑︂
∞ (︂
n+1 2
)︂ ∑︂
∞
n−1
∆g(ϕ, λ, r) = r − r Tn (ϕ, λ) = r Tn (ϕ, λ).
n=2 n=2

Choose the following notation:

def n−1
∆gn (ϕ, λ) = T (ϕ, λ). (5.7)
R n
The presence of the factor n − 1 shows that gravity anomalies cannot
contain constituents of degree n = 1, even if T would contain them. It
is always wise to place the origin of the co-ordinate system in the centre

í Õ ! ¤. û
 Units used for gravity anomalies
117
of mass of the Earth, but if the origin is not located there, at least gravity
anomalies do not change.
We obtain by substitution

∑︂
∞
n − 1 R n+1
(︂ )︂
∆g(ϕ, λ, r) = r r Tn (ϕ, λ) =
n=2
∑︂
∞
n − 1 R n+2
(︂ )︂ ∑︂
∞ (︂ )︂n+2
R
= r Tn (ϕ, λ) = r ∆gn (ϕ, λ). (5.8)
R
n=2 n=2

At sea level r = R we find

∑︂
∞
∆g(ϕ, λ, R) = ∆gn (ϕ, λ),
n=2

showing the ∆gn to be the degree constituents of gravity anomaly ∆g.

Observe that the term n = 1 is missing: ∆g1 = 0. It is also assumed
that ∆g0 = − T0 R = 0, meaning that the true exterior potential is on
/︁

global average the same as the normal potential. Also the total mass
GM⊕ and geoid volume5 of the Earth are assumed to be the same as
the total mass and volume of the reference ellipsoid. The assumption is vertaus-
largely justified because GM⊕ can be, and has been, determined very ellipsoidi
precisely by satellites, and modern models for the normal potential, like
GRS80, are based on these determinations.6

^ 5.4 Units used for gravity anomalies

A common unit of measurement for gravity variations is the milligal.
The connection with the SI system is 1 mGal = 10−5 m/s2 . The unit µGal
or 10−8 m/s2 is also used. The units µm/s2 and nm/s2 , which formally belong

5 In fact, the atmosphere complicates this matter.

6 However, GRS80 has an equatorial radius of 6 378 137.0 m, while the newer models
like EGM2008 give a smaller value of 6 378 136.3 m as the mean location of global mean
sea level. Uncertainty continues to be of decimetre order.

í Õ ! ¤. û
118 Anomalous quantities of the gravity field

Topography
Telluroid Q P (measurement point)
ζ
Mean sea
level (geoid)
HP
hQ
Ellipsoid
N

Figure 5.4. Reference ellipsoid, mean sea level (geoid), and gravity measure-
^ ment.

to the SI system, are also used in modern books. Nevertheless, milligals

and microgals are more familiar still, and correspond to 1 ppm (part
per million) and 1 ppb (part per billion) of ambient gravity close to the
Earth’s surface.
In table 5.1 we give a few values in order to get an idea of the orders
of magnitude of phenomena.
A popular unit for measuring gravity gradients is the eötvös, symbol
E. In SI units it is 10−9 s−2 , corresponding to 10−4 mGal/m. On the Earth’s
∂
surface the vertical gradient ∂H g is on average some −0.3 mGal/m =
−3000 E.

^ 5.5 The boundary-value problem of physical geodesy

As we explained in the above section, gravimetric measurement is more
∂ ∂
complicated than just measuring the quantity ∂H W ≈ ∂r W. When we
measure the vertical derivative of the geopotential, we do it in a place we
do not precisely know. Even if we know the height of the measurement
location above sea level, that still does not give us the measurement

í Õ ! ¤. û
 The boundary-value problem of physical geodesy
119
^ Table 5.1. Orders of magnitude of gravity variations.

Phenomenon Fraction of gravity SI units mGal

Ambient gravity 1 9.81 981 000

Variation with location ±10−4 ±10−3 ±100
Difference equator – poles 0.5 % 0.05 5000
Difference sea surface – 10 km high 0.3 % 0.03 3000
Gravimeter accuracy ± 10 –10−7
−8 ± 10 –10−6
−7 ± 0.01–0.1

point’s location in space. This location depends additionally on the

location of sea level, i. e., the geoid, in space; more precisely its height
above or below the reference ellipsoid. vertaus-
ellipsoidi
This is how we arrive at the third boundary-value problem.7 The
boundary-value problem of physical geodesy is to determine the potential V reuna-arvo-
tehtävä
outside a body if given on its surface is the linear combination

∂V
c1 V + c2 ,
∂n
with c1 , c2 suitable coefficients. The variable n represents here differ-
entiation in the direction of the normal to the boundary surface, in
practice the same as H or r.
In physical geodesy, the following linear combination is given as a
boundary condition:

∂T 1 ∂γ
∆g = −+ T.
∂H γ ∂H
∂ /︁
We see that c1 = −1 and c2 = ∂H γ γ . This equation, the definition 5.5
of gravity anomalies, is known as the fundamental equation of physical fysikaalisen
geodesy. geodesian
perusyhtälö

7 The third or mixed boundary-value problem is associated with Victor Gustave Robin
(1855–1897), a French mathematician. Then, the Dirichlet problem could be called the
first and the Neumann problem the second boundary-value problem.

í Õ ! ¤. û
120 Anomalous quantities of the gravity field

In spherical approximation we have equations 5.6 and 5.8:

∂T 2
∆g = − − rT,
∂r
∑︂
∞
n − 1 R n+1
(︂ )︂ ∑︂
∞ (︂ )︂n+2
R
∆g(ϕ, λ, r) = r r Tn (ϕ, λ) = r ∆gn (ϕ, λ),
n=2 n=2

or at sea level (r = R):

∑︂
∞
n−1 ∑︂
∞
∆g = Tn (ϕ, λ) = ∆gn (ϕ, λ),
R
n=2 n=2

in which the degree constituents ∆gn (ϕ, λ) are defined in equation 5.7.
Remember that these functions are computable with the help of the
asteosuus- degree constituent equation 3.8 when ∆g(ϕ, λ) is known all over the
yhtälö Earth.

reuna-arvo- Thus we also obtain the solution of this boundary-value problem in

tehtävä spectral representation (which is thus valid in the whole exterior space)
by using degree constituent equation 3.8:
∑︂
∞ (︂ )︂n+1
R ∆gn (ϕ, λ)
T (ϕ, λ, r) = R r =
n−1
n=2

R ∑︂ 2n + 1 R n+1
∞ (︂ )︂ x (︁ ′ ′ )︁
= ∆g ϕ , λ , R Pn (cos ψ)dσ ′ . (5.9)
4π n−1 r σ
n=2

This is precisely the boundary-value problem that is created if surface

gravity anomalies are given everywhere on the Earth, land and sea.
The integral equation corresponding to the above spectral equation
5.9 is known as the Stokes8 equation:
x
R
S(ψ, r)∆g ϕ ′ , λ ′ , R dσ ′ ,
(︁ )︁
T (ϕ, λ, r) =
4π σ
in which the Stokes kernel is
∑︂∞
2n + 1 R n+1
(︂ )︂
S(ψ, r, R) = Pn (cos ψ). (5.10)
n−1 r
n=2

8 Sir
George Gabriel Stokes PRS (1819–1903) was an Irish-born, gifted mathematician
and physicist who made his career in Cambridge.

í Õ ! ¤. û
 The telluroid mapping and the “quasi-geoid”
121
In section 8.1 we will give closed expression 8.2 for this function, for
the case r = R, and a graph.

^ 5.6 The telluroid mapping and the “quasi-geoid”

If we measure the astronomical latitude and longitude (Φ, Λ) and
interpret them as geodetic (geographical) co-ordinates (φ, λ), and also
interpret the potential difference − (W − W0 ) as a measure for the height
above the reference ellipsoid h, we perform, as it were, a mapping. This vertaus-
mapping adds to every point P a corresponding point Q, the geodetic ellipsoidi
co-ordinates of which are the same as the natural co-ordinates of point
P.
This approach is called telluroid mapping. The telluroid is the surface
that follows the shapes of the Earth’s topography, but is everywhere
below the topography by an amount ζ if positive, or above it by an
amount −ζ otherwise. The quantity ζ is called a height anomaly.
Telluroid mapping is an important tool in Molodensky’s gravity field
theory. It is, however, a pretty abstract concept. One may say that the
telluroid is a model of the Earth’s surface, obtained by assuming that
◦ The true potential field of the Earth is the normal potential.
◦ The mathematical mean sea surface or geoid, the reference surface
for height measurement, coincides with the reference ellipsoid.
In other words, the telluroid is a model for the Earth’s topographic
surface that is obtained by taking levelled heights — more precisely,
geopotential numbers obtained from levelling — as if they represented
differences of the normal potential from that of the reference ellipsoid.
In practice, a map of values ζ is often called a “quasi-geoid” model.
The quasi-geoid is usually close to the geoid, except in the mountains,
where the differences can exceed a metre.
One should however remember that the height anomaly ζ is defined
on the topographic surface; a surface that is quite rough in many places.

í Õ ! ¤. û
122 Anomalous quantities of the gravity field

This also means that all variations in topographic height will also be
reflected as variations in this quasi-geoid, in such a way that the quasi-
geoid correlates strongly with the small details in the topography. One
can thus not say that the shape of the quasi-geoid only expresses the
figure of the Earth’s potential field. In it, variations in geopotential and
in topographic height are hopelessly mixed up.
This is why the quasi-geoid is an unfortunate compromise, a conces-
vertauspinta sion to “reference-surface thinking”, which only really works within the
classical geoid concept. Better stick — within Molodensky’s theory —
to the concept of height anomaly, which is a three-dimensional function
or field
ζ(X, Y, Z) = ζ(φ, λ, h).

^ 5.7 Free-air anomalies

If we measure gravity g at point P, the height of which over sea level is
H and the latitude Φ, we may calculate the gravity anomaly ∆g at the
point as follows:
def
∆g = g − γ(Φ, H),
in which γ(Φ, H) is normal gravity calculated according to its definition
at height H and latitude Φ.
ilma-anomalia This is how we define free-air anomalies.
We linearise this as follows:
(︃ )︃
∂γ ∂γ
∆g = g − γ(Φ, H) ≈ g − γ(φ, h) + (Φ − φ) + (H − h) ≈
∂φ ∂h
(︃ )︃
∂γ ∂γ ∂γ
≈ g − γ(φ, 0) + h + (H − h) = g − γ(φ, 0) − H ,
∂h ∂h ∂h
∂
making the approximation that the vertical gradient ∂h
γ of normal
gravity is constant.9

9 For the greatest precision, one should consider that the latitude Φ may also not

í Õ ! ¤. û
Free-air anomalies
123
Thus, free-air anomalies can be calculated in a simpler way. The
gravity formula of the normal field 4.7 gives for latitude 60◦ :

γ = 981 919.178 − 0.308 449 4 H + · · · mGal.

So, in linear approximation (close to the Earth’s surface) gravity atten-

uates some 0.3 mGal for every metre in height. This value is worth
committing to memory.
An approximate equation for calculating free-air anomalies then is ilma-anomalia

∆gP = gP − γ0 (φ) + 0.3084mGal/m H, (5.11)

def
in which γ0 (φ) = γ(φ, 0), normal gravity at sea level, is only a function
of latitude. In a country like Finland, equation 5.11 is often sufficiently
precise, although the evaluation of the original equation 4.7 is also easy.
Free-air anomalies are widely used. Generally, when one discusses
gravity anomalies, one means just this, free-air anomalies. They express
the Earth’s exterior gravity field, including mountains, valleys and
everything.
Questions
1. If gravity at sea level is 9.81 m/s2 , at what height will gravity
disappear, as computed according to the above-mentioned
vertical gravity gradient −0.3 mGal/m?
2. How physically realistic is this?
Answers
(︂ /︂ )︂
5
1. At −0.3 mGal/m, it takes 9.81 · 10 0.3 m = 3270 km to go
to zero.

be a latitude on a geocentric reference ellipsoid. It could be astronomical latitude,

or latitude in some old national co-ordinate system computed on a non-geocentric
ellipsoid, like in Finland KKJ, the National Map Grid Co-ordinate System, which was
computed on the Hayford ellipsoid. The error caused by this is however two, three
orders of magnitude smaller than the effect caused by H − h.

í Õ ! ¤. û
124 Anomalous quantities of the gravity field

18 20 22 24 26 28 30 32

64 64

62 62

60 60

18 20 22 24 26 28 30 32

Figure 5.5. Free-air gravity anomalies for Southern Finland, computed from
the EGM2008 spherical-harmonic expansion. Data © Bureau
Gravimétrique International (BGI) / International Association of
^ Geodesy. Web service BGI, EGM2008.

2. Not very. The gravity gradient itself drops quickly from the
value of −0.3 mGal/m going up, so this linear extrapolation is
simply wrong.

^ Self-test questions
luotiviiva 1. How do deflections of the plumb line and geoid heights relate to

í Õ ! ¤. û
 Exercise 5 – 1: The spectrum of gravity anomalies
125
each other?
2. What is the fundamental equation of physical geodesy in spherical fysikaalisen
approximation? geodesian
perusyhtälö
3. In what way is a gravity disturbance different from a gravity
anomaly?
4. What units are used for measuring gravity anomalies and gravity
gradients? How are they related to the SI system?
5. How does the geoid height and the disturbing potential relate to
each other?
6. Explain telluroid mapping and height anomalies.

^ Exercise 5 – 1: The spectrum of gravity anomalies

Use equation 5.7. If we assume that the quadratic mean of the degree
constituents ∆gn of gravity anomalies,
√︃ x
def 1
∆gn = ∆g2 (ϕ, λ) dσ,
4π σ n
does not depend on the chosen degree number n, how then does the
similarly defined T n depend on n?
In other words: which degree numbers of the gravity field are
relatively strongest in the disturbing potential, and which in the gravity
anomalies?

^ Exercise 5 – 2: Deflections of the plumb line and geoid

tilt
If, in the south-north components of plumb-line deflections in a country, luotiviivan
there is a systematic error of one arc second, what error does this cause poikkeama
in the difference N2 − N1 between the geoid heights in points 1 and
2 of which the inter-point distance is approximately 1000 km in the
south-north direction? See figures 5.1 and 5.2.

í Õ ! ¤. û
126 Anomalous quantities of the gravity field

^ Exercise 5 – 3: Gravity anomaly, geoid height

ilma-anomalia In Finland there is a place where the (free-air) gravity anomaly is
∆g = 100 mGal = 10−3 m/s2 . In the same place the disturbing potential
T is 200 m2/s2 .
fysikaalisen 1. Using the fundamental equation of physical geodesy 5.6:
geodesian
perusyhtälö ∂T 2
∆g = − − rT,
∂r
∂
T and compare it with the quantity 2T r . Assume
/︁
calculate ∂r
∂
T or 2T r , dominates?
/︁
r ≈ R. Which of the two, ∂r
2. Assume that the point is close to sea level. Using the Bruns
equation
N= T γ,
/︁

where γ is average gravity 9.81 m/s2 , compute the geoid height N

of the point.

í Õ ! ¤. û
^ Geophysical reductions

^ 6.1 General
6
We see that integral equations, like Green’s third theorem 1.25, offer a
possibility to calculate the whole exterior potential of the Earth — as
well as all the quantities that may be calculated from the potential, such
as, for example, the gravitational acceleration — using values of V or
∂
∂n
V observed on the boundary surface only. Green’s third theorem
is but one example of many: every integral theorem is the solution of
some boundary-value problem. reuna-arvo-
tehtävä
There are three alternatives for choosing a boundary surface:
1. Choose the topographic surface of the Earth.
2. Choose mean sea level, more precisely, an equipotential surface
close to mean sea level called the geoid.
3. Choose the reference ellipsoid. vertaus-
ellipsoidi
◦ Alternative 1 has been developed mostly by the Molodensky
school (Molodensky et al., 1962) in the former Soviet Union. The
advantage of the method is that we need no gravity reduction, as all
masses are already inside the boundary surface. Its disadvantage
is that the, often complex, shape of the topography must be taken
into account when the boundary-value problem is formulated
and solved.

– 127 –
128 Geophysical reductions

◦ Alternative 2 is classical geoid or geopotential determination. In

this case geophysical reductions are needed to the input gravity data:
some masses are outside the computation boundary and need to
be computationally removed or moved to the inside.
A further complication of the method then is that the geopotential
or geoid solution obtained is not that of the original mass distri-
bution, but of the reduced one. This surface is called the co-geoid.
We need a “restoration step” where this influence of the reduction
step on the geopotential and geoid is determined and reversed.
epäsuora This influence is called the “indirect effect”.
vaikutus
In the literature, this method is also referred to as the remove–restore
method.
◦ Alternative 3 has been used rarely, because it has not been tra-
ditionally possible to make gravity measurements in a location
known in the absolute sense, relative to the geocentre or the
vertaus- reference ellipsoid. Nowadays this is possible using GNSS, for
ellipsoidi example in Antarctica and Greenland’s interior, where there is no
sea-level-bound height system.
We may expect this approach to gain more traction as heights of
gravimetric stations are more and more determined directly using
GNSS. See for example Märdla (2017).

^ 6.2 Bouguer anomalies

ilma-anomalia Free-air anomalies depend on the topography, because gravity itself
contains the attractive effect of topographic masses. A map of free-air
anomalies shows the same small details as seen in the topography. One
way of removing the effect of the topography is the so-called Bouguer1

1 Pierre Bouguer (1698–1758) was a French professor in hydrography, who participated

in the public discussion on the figure of the Earth, and in 1735–1743 led an expedition
of the French Academy of Sciences doing a grade measurement in Peru, South America,
at the same time as De Maupertuis was carrying out a similar grade measurement in

í Õ ! ¤. û
 Bouguer anomalies
129
z
P
dz
β ds ⏟ ⏞⏞ ⏟
ℓ
dβ
cos β
H ℓ

d
s dV x

^ Figure 6.1. The attraction of a Bouguer plate.

reduction.

^ 6.2.1 Calculation
We calculate the effect of a homogeneous plate on gravity. Assume that
the plate is infinite in size; thickness d, matter density ρ, and height
of point P above the lower surface of the plate H. See figure 6.1. The
attraction at point P (which is directed straight downwards for reasons
of symmetry) is obtained by integrating. The volume integral to be
computed has a volume element

dV = ds · dz · s dα

in the cylindrical co-ordinates (s, z, α). We transform this to the co-

ordinates (β, z, α). We forget about α and study the surface element
(figure 6.1, top right)
ℓ
ds dz = dβ dz,
cos β

in which the determinant of Jacobi needed, ℓ cos β , is seen.

/︁

Lapland. In addition to geodesy, he was also active in astronomy.

í Õ ! ¤. û
130 Geophysical reductions

We carry out the integration:

y w 2πw dw ∞
def cos β cos β
a = ∥a∥ = G 2
ρ dV = Gρ · ds dz · s dα =
ℓ 0 0 0 ℓ2
w 2πw dw π/2
cos β ℓ
= Gρ · dβ dz · s dα =
0 0 0 ℓ2 cos β
w dw π/2 w d (︃w π/2 )︃
s
= 2πGρ dβ dz = 2πGρ sin β dβ dz.
0 0 ℓ 0 0

Here, the integral

w π/2 [︁ ]︁ π/2
sin β dβ = − cos β 0 = 1,
0

and the end result is

a = 2πGρd. (6.1)
This is the formula for the attraction of a Bouguer plate. As a side result,
we obtain the attraction of a circular disk of radius r:
w β0 (z) [︁ ]︁β (z) (︁ )︁
sin β dβ = − cos β 0 0 = 1 − cos β0 (z) ,
0

and the whole integral

wd
(︄ )︄
H−z
a = 2πGρ 1 − √︂ dz.
0
(H − z)2 + r2

The indefinite integral is

w √︂
H−z
√︂ dz = − (H − z)2 + r2 .
(H − z)2 + r2

Substituting the bounds yields

wd
(︄ )︄ √︂
H−z √︁
1 − √︂ dz = d + (H − d)2 + r2 − H2 + r2 .
0
(H − z)2 + r2

We obtain for the whole integral

(︃ √︂ √︁ )︃
2 2 2 2
a = 2πGρ d + (H − d) + r − H + r .

í Õ ! ¤. û
 Bouguer anomalies
131
Evaluation point P

II

Bouguer plate
I I
d=H
Topography

^ Figure 6.2. The Bouguer plate as an approximation to the topography.

In the limit r → ∞, and thus

√︂ √︁
(H − d)2 + r2 − H2 + r2 → 0,

this is identical to equation 6.1.

Bouguer anomalies are computed in order to remove the attraction
of masses of the Earth’s crust above sea level, i. e., above the geoid. The
true topography is approximated by a Bouguer plate, see figure 6.2.
There is no standard way to treat sea-covered areas:
◦ Some maps show Bouguer anomalies over land and free-air anoma- ilma-anomalia
lies over the sea. This is an option if no good quality depth data is
available.
◦ A more correct way is to replace sea water by a rocky Bouguer plate,
the thickness of which equals the local sea depth or bathymetry.
The calculation goes as follows:

∆gB = ∆gFA − 2πGρH = ∆gFA − 0.1119 H, (6.2)

where we assume for the density ρ of the plate an often-used value for
the average density of the Earth’s crust, ρ = 2670 kg/m3 . By substituting
into this equation 5.11, we obtain
(︁ )︁
∆gB = gP − γ0 (φ) + 0.3084 − 0.1119 H = gP − γ0 (φ) + 0.1965 H.
(6.3)

í Õ ! ¤. û
132 Geophysical reductions

Topography

Geoid

Free-air anomaly

Bouguer anomaly

^ Figure 6.3. Behaviour of different anomaly types in mountainous terrain.

The quantity ∆gB is called a (simple) Bouguer anomaly.

The difference between the attraction of a Bouguer plate and that of
the true topography is called the terrain correction TC (volumes I and II
in figure 6.2). We shall return to its computation later.

^ 6.2.2 Properties
ilma-anomalia Unlike free-air anomalies which vary on both sides of zero, Bouguer
anomalies are strongly negative, especially in the mountains. For example,
if the mean elevation of a mountain range is H = 1000 m, the Bouguer
systematiikka anomalies will, as a consequence of this, contain a bias of 1000 ×
(−0.1119 mGal) = −112 mGal, about −100 mGal for every kilometre of
elevation.
The advantage of Bouguer anomalies is their smaller variation with
place. For this reason they are suited especially for the interpolation
and prediction of gravity anomalies, in situations where the available
gravimetric material is geographically sparse. However, one then has
to have access to topographic heights of a better spatial density.

í Õ ! ¤. û
 Terrain effect and terrain correction
133
18 20 22 24 26 28 30 32

64 64

62 62

60 60

18 20 22 24 26 28 30 32

Figure 6.4. Terrain-corrected Bouguer anomalies for Southern Finland, com-

puted from the spherical-harmonic expansion EGM2008. Data
© Bureau Gravimétrique International (BGI) / International Asso-
ciation of Geodesy. Web service BGI, EGM2008. In comparison to
figure 5.5 on page 124, there is a strong negative bias of Bouguer
anomalies — although part of this is due to post-glacial isostatic
unbalance and also visible in the free-air map. Bouguer anomalies
are also smoother, but that is harder to see here, as Southern
^ Finland is already a smooth area.

^ 6.3 Terrain effect and terrain correction

Using the simple Bouguer reduction does not remove precisely from
gravity anomalies the attractive effect of the whole topography. Figure
6.2 shows that we make two types of error:
◦ The attraction of volumes I is taken along, although there is

í Õ ! ¤. û
134 Geophysical reductions

nothing there.
◦ The attraction of volumes II, where there actually is something, is
ignored.
Both errors work in the same direction! Because volumes I are below
the point of evaluation, their attraction — which the simple Bouguer
reduction considers present, and removes — would act downwards.
And because volumes II are above the point of evaluation, their attraction
— which in the simple Bouguer reduction is not corrected for — acts
upwards. The error made is in the same direction as in the previous
case.

The terrain correction is always positive.

We write
∆gB′ = ∆gB + T C,
where TC — the “terrain correction” — is positive. ∆gB′ is called the
terrain-corrected Bouguer anomaly.
The terrain correction is calculated by numerical integration. Figure
6.5 shows the prism integration method, and how both prisms, I and II,
lead to a positive correction, because prism I is computationally added
and prism II removed when applying the terrain correction. One needs
a digital terrain model, DTM, which must be, especially around the
evaluation point, extremely dense: according to experience, 500 m is
the maximum inter-point separation in a country like Finland, in the
mountains one needs even 50 m. The systematic nature of the terrain
correction makes a too-sparse terrain model cause, possibly serious,
biases in the insufficiently corrected gravity anomalies.
To compute the terrain correction with the prism method, we use the
following equation, assuming a constant crustal density ρ and a flat
Earth, in rectangular map co-ordinates x, y:
w +Dw +D (︂ (︁ )︂2
1 1
H x , y − H(x, y) dx ′ dy ′ ,
′ ′
)︁
T C(x, y) = 2 Gρ 3
−D −D ℓ

í Õ ! ¤. û
 Terrain effect and terrain correction
135
H (x ′ , y ′ )
Topography
θ II
H (x, y) P

I
H (x ′ , y ′ )
Geoid
x ′, y ′ x, y x ′, y ′

^ Figure 6.5. Calculating the classical terrain correction using the prism method.

in which
√︃ (︂ (︁ )︁)︂2
ℓ= (x − x ′ )2 + (y − y ′ )2 + 1
2
H(x ′ , y ′ ) − H(x, y)

is the distance between the evaluation point

[︂ ]︂T
x y H(x, y)

and the centre point on the central axis of the prism

[︂ (︂ )︁)︂ ]︂T
x ′ y ′ 12 H(x ′ , y ′ ) + H x, y
(︁
.

Of course, this is only an approximation, but it works well enough in

terrain where slopes generally do not exceed 45◦ . In the integral above,
the limit D is typically tens or hundreds of kilometres. In the latter case,
the curvature of the Earth already starts having an effect, which the
formula does not consider.
The values of the terrain correction TC vary from fractions of a milligal
(Southern Finland) to hundreds of milligals (high mountain ranges).
In the “arm” of Finland — the north-western, somewhat mountainous
border area with Sweden and Norway — the terrain correction may be
tens of milligals.
Figure 6.6 shows the stages of calculating Bouguer anomalies from
gravity observations through terrain correction, Bouguer-plate correc-
tion and free-air reduction. ilmareduktio

í Õ ! ¤. û
136 Geophysical reductions

Given: gravity g
on terrain Bouguer
Terrain
plate
correction
correction

Free-air
reduction Subtract normal gravity
to sea at sea level
level −γ0 (φ)

Figure 6.6. The steps in calculating the Bouguer anomaly. The reduction
to sea level uses the standard free-air vertical gravity gradient,
^ −0.3084 mGal/m, the vertical gradient of normal gravity.

^ 6.3.1 Example: applying the terrain correction in a special case

The special terrain shape rendered in quasi-3D in figure 6.7 is given.
Here, the height differences are PQ ′ = 300 m and QQ ′ = 200 m. The
rock density is the standard crustal density, 2670 kg/m3 .
Questions
1. Calculate the terrain correction at point P (hint: use the

Q
300 m
200 m

Q′ Sea level

Figure 6.7. A special terrain shape. The vertical rock wall at PQ is also straight
^ on a map and extends to infinity in both directions.

í Õ ! ¤. û
Terrain effect and terrain correction
137
attraction formula for the Bouguer plate). Algebraic sign?
2. Calculate the terrain correction at point Q. Algebraic sign?
3. If at point P it is given that the free-air anomaly is 50 mGal, ilma-anomalia
how much is the Bouguer anomaly at the point?
4. If at point Q it is given that the Bouguer anomaly is 22 mGal,
how much is the free-air anomaly at the point?
Answers
1. The terrain correction at point P is the change in gravity
if the terrain is filled up on the left side up to a level of
300 m. This means adding half a Bouguer plate, thickness
100 m, below the level of P. The effect (projected onto the
vertical direction) is
1 1
TC = 2
· 2πGρ · H = 2
· 0.1119 mGal/m · 100 m =
= 5.595 mGal.

2. The terrain correction at point Q is the change in gravity if

we remove the half Bouguer plate to the right of and above the
point, which is 100 m thick. Its vertical gravity effect is, as
calculated above,
T C = 5.595 mGal,
and, because a semi-plate is removed that is above the level of
point Q, the algebraic sign of TC is again positive.
3. Free air to Bouguer:
∆gFA (P) 50.000 mGal
TC +5.595 mGal
Bouguer plate removal, 300 m −33.570 mGal
∆gB (P) 22.025 mGal

∆gFA +T C −Plate ∆gB

í Õ ! ¤. û
138 Geophysical reductions

4. Bouguer to free air:

∆gB (Q) 22.000 mGal
Bouguer plate addition, 200 m +22.380 mGal
TC “uncorrection” −5.595 mGal
∆gFA (Q) 38.785 mGal

∆gB +Plate −T C ∆gFA

^ 6.4 Spherical Bouguer anomalies

More recently, spherical Bouguer anomalies have also been calculated,
for example Balmino et al. (2012); Kuhn et al. (2009); Hirt and Kuhn
(2014). In this calculation, the topography and bathymetry of the whole
Earth is taken into account, in spherical geometry (the error caused
by neglecting the Earth’s flattening is in this calculation negligible).
Spherical Bouguer anomalies differ from Bouguer-plate anomalies in
four ways:
1. the attraction of a Bouguer shell of thickness H is 4πGρH, twice
as much as the corresponding Bouguer-plate attraction. The
remote part of the shell contributes as much attraction as the
neighbourhood of the evaluation point!
2. The bathymetry of the oceans is accounted for2 by replacing the
water with standard-density crustal rock. This contribution to the
anomalies is positive.
3. The topography and bathymetry of remote parts of the globe
are also taken into account realistically. As most of the Earth is
covered by deep ocean, this causes a strong positive general bias,
which in moderately elevated areas like Southern Finland more

2 One can also do so, and often does, in connection with the Bouguer-plate correction.

í Õ ! ¤. û
Spherical Bouguer anomalies
139
than cancels the negative one caused by the local topography!
4. As the terrain correction is now calculated over the whole globe,
in spherical geometry, it is no longer a small number and may be
strongly negative as well as positive (Abrehdary et al., 2016).
There is a large systematic difference between the planar and spherical
Bouguer anomalies, which however is very long-wavelength in nature,
and even in an area the size of Australia almost a constant, −18.6 mGal
within a variation interval of a few milligals. The details in the Bouguer
maps look the same (Kuhn et al., 2009).
Just for fun, we compute the net mass effect of doing the complete
spherical Bouguer reduction globally. The mean height of the land
topography is 800 m, land occupying 29 % of the globe. The mean ocean
depth is 3700 m, corresponding to an equivalent rock depth to be “filled
in” of
2670 − 1030
3700 × m = 2272 m,
2670
assuming a density for crustal rock of 2670 kg/m3 , a sea-water density of
1030 kg/m3 , and ocean occupying 71 % of the globe. The sum weighted
by area is thus

(0.29 × 800 − 0.71 × 2272) m = −1381 m.

Interpretation There is not enough topography to fill all of the oceans,

even if we are allowed to compress sea water into standard crustal
rock. If we try this bulldozing experiment, we will end up 1381 m
short, i. e., below the current sea level.
If, instead, we add standard crustal rock to end up at current sea
level — the definition of spherical Bouguer reduction! — we will
add to the Earth’s attraction as sensed from space an amount of
4πGρ × 1381 m = 309 mGal.
The global mean planar Bouguer reduction, as well as the dif-
ference between planar and spherical Bouguer reductions, on
average over the globe, will be half of this, ≈ +155 mGal. Coastal

í Õ ! ¤. û
140 Geophysical reductions

plains and seas, both close to sea level, will have similarly positive
spherical Bouguer anomalies.

^ 6.5 Helmert condensation

An often-used method, proposed by Friedrich Robert Helmert,3 for
removing the effect of the masses exterior to the geoid is condensation.
In this method, we shift mathematically all the continental masses
vertically downwards to mean sea level into a simple mass-density layer

κ = ρH,

where H is the height of the topography above sea level and ρ its mean
matter density. This mass surface density can be interpreted as a column
mass integral:
w R+H
κ=ρ dz.
R

For a spherical Earth, the corresponding integral is

w H (︂ )︂2 (︃ )︃
r 1 [︁ 1 3 ]︁R+H H 1 H2
κ=ρ dr = ρ 2 3 r R = ρH 1 + + 3 2 , (6.4)
R R R R R

where it is understood that mass is moved from a column cross-section

2/︁
of r R2 to sea level, where the cross-section is 1.
The advantage of Helmert condensation over Bouguer reduction
is that no mass is being removed. The Bouguer reduction amounts to
the computational removal of topographic masses on a large scale.
Therefore, unlike with Bouguer reduction, in Helmert condensation
gravity anomalies will not change systematically.
In appendix D we derive series expansions in spherical geometry
which express both the external and the internal potential as functions
of the “degree constituents” of the various powers of the topography

3 Friedrich
Robert Helmert (1843–1917) was an eminent German geodesist known for
his work on mathematical and statistical geodesy.

í Õ ! ¤. û
 Isostasy
141
Equipotential surface

Topography

g
g′

Condensation layer

Figure 6.8. Helmert condensation and the changes it causes in the gravity
^ field.

H(ϕ, λ). The extensively presented derivation in the appendix is much-

used in gravity field theory to model the gravity effect of the topography.
In this theory, issues of convergence are difficult, although we gloss suppeneminen
over those here.

^ 6.6 Isostasy
^ 6.6.1 Classical hypotheses
As early as in the 18th and 19th centuries, also thanks to Bouguer’s work
in South America as well as that of British geodesists in the Indian
Himalayas, it was understood that mountain ranges are not just piles
of rock on top of the Earth’s crust. The gravity field surrounding the
mountains, specifically the deflections of the plumb line, could only be luotiviivan
explained by assuming that under every mountain range there was also poikkeama
a “root” made from lighter rock species. The origin of this root was
speculated to be the almost hydrostatic behaviour of the Earth’s crust

í Õ ! ¤. û
142 Geophysical reductions

^ Figure 6.9. Friedrich Robert Helmert. Humboldt University Berlin (2017).

over geological time-scales. This assumption of hydrostatic equilibrium

was called the hypothesis of isostasy, also isostatic compensation.
Back then, unlike now, it was not yet possible to get a precise or
even correct picture using physical methods (seismology) of how these
mountain roots are really shaped. That is why simplified working
hypotheses were formulated.
One classic isostatic hypothesis is the Pratt–Hayford hypothesis. This
was proposed by J. H. Pratt4 in the middle of the 19th century (Pratt,
1855, 1859, 1864), and J. F. Hayford5 developed the mathematical tools

4 John
Henry Pratt (1809–1871) was a British clergyman and mathematician who
worked as the archdeacon of Kolkata, India. Wikipedia, John Pratt.
5 John Fillmore Hayford (1868–1925) was a United States geodesist who studied isostasy

í Õ ! ¤. û
 Isostasy
143
Plumb-line deflections

Geoid
Mountain

Earth’s crust

“Root”

Earth’s mantle

^ Figure 6.10. Isostasy and the bending of plumb lines towards the mountain.

needed for computation. According to this hypothesis, the density

of the “root” under a mountain would vary with the height of the
mountain, so that under the highest mountains would be the lightest
material, and the boundary between this light root material and the
denser material of the Earth’s mantle would be at a fixed depth. This
model, which nowadays finds little acceptance, is illustrated in figure
6.11.
Another classical isostatic hypothesis is due to G. B. Airy.6 Because
V. A. Heiskanen7 used it extensively and developed its mathematical

and the figure of the Earth.

6 GeorgeBiddell Airy PRS (1801–1892) was an English mathematician and astronomer,
“Astronomer Royal” 1835–1881.
7 VeikkoAleksanteri Heiskanen (1895–1971), “the great Heiskanen” (Hermans, 2007)
was an eminent Finnish geodesist who also worked in Ohio, USA. He is known for his
work on isostasy and global geoid modelling, the “Columbus geoid”. See Kakkuri
(2008).

í Õ ! ¤. û
144 Geophysical reductions

Mountains

Sea
Compen-
sation Earth’s crust
depth

Compen-
sation
level
Mantle

^ Figure 6.11. Pratt–Hayford isostatic hypothesis.

form, it is called the Airy–Heiskanen model. In this model, it is

assumed that the mass density of the “root” is fixed, and that the
isostatic compensation is realised by varying the depth to which the
root extends down into the Earth’s mantle. In our current understanding,
this corresponds better to what is really happening inside the Earth.
This hypothesis is illustrated in figure 6.12.

^ 6.6.2 Calculation formulas

Airy’s isostatic hypothesis assumes that in every place the total mass of
a column of matter is the same. So, let the density of the Earth’s crust be
ρc , the density of the mantle ρm , the density of sea water ρw , sea depth
d, crustal thickness t, and topographic height H. We obtain

d (ρm − ρw ) + c
tρc + dρw − (t + d) ρm = c =⇒ t = − ρm − ρc
on the sea, and
Hρ − c
tρc − (t − H) ρm = c =⇒ t = ρ m− ρ
m c

í Õ ! ¤. û
 Isostasy
145
Mountains
Sea ρw

Earth’s crust
t0

ρc
Anti-root

Mountain root

ρm

Mantle

^ Figure 6.12. Airy–Heiskanen isostatic hypothesis.

on land. c is a suitable constant.8 Here we have conveniently forgotten

about the curvature of the Earth: we use the “flat Earth model”.

t r
t0 t
r
Anti-root

Root

^ Figure 6.13. Quantities in isostatic compensation.

8 Its
dimension, after multiplication with ambient gravity g, is pressure: according
to Archimedes’ law, the pressure of the crustal (plus sea-water) column minus the

í Õ ! ¤. û
146 Geophysical reductions

Under land, the depth of a mountain root is

Hρ − c Hρ − hρ Hρ − c
r = t − H = ρ m− ρ − ρ m − ρ c = ρ c− ρ .
m c m c m c

Similarly under the sea

d (ρm − ρw ) + c dρm − dρc d (ρc − ρw ) + c
r=t+d=− ρm − ρc + ρ −ρ =− ρm − ρc .
m c

In the equations, the constant c is arbitrary and expresses the fact that
the level from which one computes the depth of the root — less precisely,
the “average thickness of the crust” — can be chosen arbitrarily.
Another approach: instead of c, use the “zero topography compen-
sation level” t0 , to be computed from the above equations by setting
H = d = 0:
t0 (ρc − ρm ) = c.
This yields under the land the root depth
Hρc − t0 (ρc − ρm ) ρ
r= ρm − ρc = t0 + H ρ −c ρ , (6.5)
m c

and under the sea

d (ρc − ρw ) + t0 (ρc − ρm ) ρ −ρ
r=− ρm − ρc = t0 − d ρc − ρw , (6.6)
m c

somewhat simpler equations that are also more intuitive.

Still a third form:
Hρc + (−r) (ρm − ρc ) = c,
(−d) (ρc − ρw ) + (−r) (ρm − ρc ) = c.
In other words,
∑︂
(deviation × density contrast) = constant.
interfaces

The effect of the different isostatic hypotheses on gravity is pretty much

the same: the hypotheses cannot be distinguished based on gravity
measurements alone. The effect of the choice of hypothesis on the geoid
is stronger.

pressure of the column of displaced mantle material.

í Õ ! ¤. û
 Isostasy
147
^ 6.6.3 Example: Norway
The southern Norwegian Hardanger plateau (Hardangervidda) is a high-
land at, on average, 1100 m above sea level. It is the largest peneplain in puolitasanko
Europe, a national park, and a popular tourist attraction, being traversed
by the Bergensbanen, the highest regular railway in Northern Europe.
The Norwegian Sea is the part of the Atlantic Ocean adjoining Norway,
and does not belong to the continental shelf. It is on average 2 km deep. mannerjalusta
Questions
1. What is the depth of the root under the Hardanger plateau,
relative to the compensation depth t0 ?
2. What is the negative depth of the anti-root under the Nor-
wegian Sea, relative to the same compensation depth?
3. What is the relative depth of the root of the Hardanger
plateau, compared to the nearby Norwegian Sea?
Answers
1. We use equation 6.5, finding
ρ
r − t0 = H ρ −c ρ =
m c
2670 kg/m3
= 1100 m × = 4196 m.
(3370 − 2670) kg/m3
Here we have used standard densities for crustal and mantle
rock, respectively.
2. We use equation 6.6, finding
ρ −ρ
r − t0 = −d ρc − ρw =
m c
(2670 − 1030) kg/m3
= −2000 m × = −4686 m,
(3370 − 2670) kg/m3
using the standard density value for sea water.
3. The depth contrast between root and anti-root is 4196 −
(−4686) m = 8882 m. For perspective, Mount Everest is
8848 m above sea level.

í Õ ! ¤. û
148 Geophysical reductions

Mid-Atlantic ridge
Plate motion
Deep-sea trench Conrad
Earth’s crust discontinuity
Sea X
Mohorovičić
discontinuity
Lithosphere

Convection Subduction Bottom of

lithosphere
Asthenosphere

Mantle Benioff zone

660 km
discontinuity

Figure 6.14. The modern understanding of isostasy and plate tectonics. Deep-
^ sea trenches are known to be in isostatic disequilibrium.

^ 6.6.4 The modern understanding of isostasy

Nowadays we have a much better understanding of the internal situation
in the Earth. However, isostasy continues to be a valid concept. A more
realistic understanding of the internal structure of the Earth is given in
figure 6.14.
An important subject for current research is the effect on vertical
motion of the Earth’s crust of the growing and melting of the ice masses
mannerjäätikkö of the Earth, like the continental ice sheets. This includes both the direct
effect of the varying ice masses and the effect of the changes caused
in the water masses of the ocean. Paleo-research concentrates on the
changes over the glacial cycle, while modern retreats of glaciers, as in
Alaska and on Spitsbergen, cause their own, observable local uplift of

í Õ ! ¤. û
 Isostatic reductions
149
the Earth’s crust. More in chapter 12.

^ 6.6.5 Example: Fennoscandian land uplift

During the last glacial maximum, some 20 000 years ago, Fennoscandia
was covered by a continental ice sheet of thickness up to 3 km. mannerjäätikkö

Questions
1. How much was the Earth’s surface depressed by this load,
assuming isostatic equilibrium?
2. Currently the land is rising in central Fennoscandia, where
the ice thickness was maximal, at a rate of 10 mm/a. How
long would it take at this rate for the depression to vanish?
Answers
1. We assume for the ice density a value of 920 kg/m3 . With an
upper mantle density of 3370 kg/m3 — note that it is Earth’s
mantle material that is being displaced by the ice, the Earth’s
crust just transmits the load! See figure 12.1a — we find for
the depression

920 kg/m3
∆H = 3000 m × = 819 m.
3370 kg/m3

2. At the rate of 10 mm/a it will take 819 m 0.01 m/a = 81 900

/︁

years total for the depression to vanish. Part of this uplift

has already taken place since the last deglaciation.
In reality, of course, the rate has decreased substantially, and
will continue to decrease, over time.

^ 6.7 Isostatic reductions

The computational removal of both the topography and its isostatic
compensation from the measured quantities of the gravity field is called
isostatic reduction. It serves two purposes:

í Õ ! ¤. û
150 Geophysical reductions

◦ By removing as many as possible “superficial” effects from the

gravity field, we are left with a field where only the effect of the
Earth’s deep layers remains. This is useful for geophysical studies.
◦ These “superficial” effects are also generally very local: in spectral
language, very short-wavelength. By removing those, we are left
with a residual field that is much smoother, and that can be
interpolated or predicted better. This is important especially in
areas where there is a paucity of real measurement data, like the
oceans, deserts, polar areas, etc.
ilma-anomalia Isostatic anomalies, free-air anomalies to which isostatic reduction has
been applied, are very smooth (like Bouguer anomalies), and their
predictive properties are good. However, unlike Bouguer anomalies,
isostatic anomalies are on average zero. They lack the large bias that
makes Bouguer anomalies strongly negative especially in mountainous
areas, section 6.2. This of course is because isostatic reduction is only
the shifting of masses from one place to another — from mountains to
roots beneath the same mountains, the mass deficit of which is pretty
precisely the same as the mass of the mountains themselves sticking out
above sea level — rather than removal of masses, which is what Bouguer
reduction does.
The reduction methods used in isostatic calculations are the same as
in other reductions. We will discuss them later: numerical integration
in the space domain — grid integration, spherical-cap integration,
pienimmän least-squares collocation (LSC), finite elements, etc. — or in the spectral
neliösumman domain, for example FFT and “Fast Collocation”.
kollokaatio
The question of the hypothesis assumed to apply is a more interest-
ing one. Traditionally, the Pratt or Airy hypotheses have been used,
developed into quantitative methodologies by Hayford or Heiskanen or
Vening Meinesz.9 A newer approach has been to use real measurement

9 Felix
Andries Vening Meinesz (1887–1966) looseness=1was a Dutch geophysicist,
geodesist and gravimetrist. He wrote with V. A. Heiskanen the textbook The Earth and

í Õ ! ¤. û
 Isostatic reductions
151
18 20 22 24 26 28 30 32

64 64

62 62

60 60

18 20 22 24 26 28 30 32

Figure 6.15. Isostatic gravity anomalies for Southern Finland. Airy–Heis-

kanen hypothesis, compensation depth 30 km. Data © Bureau
Gravimétrique International (BGI) / International Association of
Geodesy, World Gravity Map project. Web service BGI, WGM2012.
Here, on the thick, rigid Fennoscandian Shield, the local features
of the topography are not isostatically compensated and the map
^ looks rather similar to the free-air anomaly map 5.5 on page 124.

data from seismic tomography in order to model the interior structure

of the Earth. With real measurement data, if reliable, one should get
better results.

its Gravity Field (1958).

í Õ ! ¤. û
152 Geophysical reductions

^ 6.8 The “isostatic geoid”

Let us look at how the “isostatic geoid”, more precisely the co-geoid
of isostatic reduction, is computed. Isostatic reduction is one possible
method for computationally removing the masses outside the geoid, in
reuna-arvo- order to formulate a boundary-value problem on the geoid.
tehtävä
We can show (Heiskanen and Moritz, 1967, page 142), that under the
continents, the isostatic co-geoid is as much as several metres below the
epäsuora geoid. In other words, the indirect effect (the “restore” step) is of this
vaikutus order. Under the oceans, similarly the isostatic co-geoid is somewhat
above the geoid.
As one of the requirements for geoid determination methods is a small
indirect effect, it follows that isostatic methods are not perhaps the best
possible if the intent is to calculate a model of the geoid or quasi-geoid
representing the exterior potential.10 Heiskanen and Moritz (1967, page
152) call the indirect effect ”moderate”.
However, isostatic methods are very suitable for elucidating the
interior structure of the Earth, because both the topography and the
“impression” it makes on the Earth’s mantle, the isostatic compensation,
are computationally removed.
Research has shown that the great topographic features of the Earth
are some 85–90 % isostatically compensated (Heiskanen, 1960). This is
valuable information if no other knowledge is available.
This is the second reason why the isostatic geoid is of interest: the
gravity field of an Earth from which the effect of mountains has been
removed completely — mountain roots and all — can uncover physical
unbalances existing in deeper layers, and processes causing these. Such
processes include especially convection currents in the Earth’s mantle
as well as the possible effect of the liquid outer core of the Earth
on these currents. Interesting correlations have been found between

10 Of
course Bouguer reduction is even worse! The indirect effect can be hundreds of
metres.

í Õ ! ¤. û
The “isostatic geoid”
153
mantle convection patterns, the global map of the geoid, and the electric
current patterns in the core causing the Earth’s magnetic field (Wen and
Anderson, 1997; Prutkin, 2008; Kogan et al., 1985).
Isostatic reduction consists of two parts:
◦ computational removal of the topography
◦ computational removal of the isostatic compensation of the topog-
raphy.
It is possible to calculate both parts exactly using prism integration, see
section 6.3. Here however we shall gain understanding by a qualitative
approach. We approximate both parts with a single mass-density layer,
with density for example κ = ρH for the topography. We place the first
layer at level H = 0, and the second, density -κ, at compensation depth
H = −D. The situation is depicted in figure 6.16.
In the following we use the “generating function” equation 8.6,

1 ∑︂ R n+1
∞ (︂ )︂
1
= r Pn (cos ψ),
ℓ R
n=0

together with the single mass-density layer equation 1.15:

x x
κ κ
V=G dS = GR2 dσ.
surface ℓ surface ℓ

We obtain for the potential field of the mass-density layer at sea level,
when the evaluation point is also placed at sea level, H = 0 =⇒ r = R:
x ∑︂
∞
Vtop = GR κ Pn (cos ψ) dσ
σ
n=0

and with the density layer at compensation depth (source level R − D,

evaluation level R):

GR2
x ∑︂
∞ (︂
R − D n+1
)︂
Vcomp = (−κ) Pn (cos ψ) dσ =
R−D σ R
n=0
x ∑︂
∞ (︂
R−D n
)︂
= −GR κ Pn (cos ψ) dσ,
σ R
n=0

í Õ ! ¤. û
154 Geophysical reductions

Sea level

Compensation depth

^ Figure 6.16. Isostatic reduction as a pair of surface density layers.

from which the combined effect (n = 0 drops out):

(︁ )︁
δViso = − Vtop + Vcomp =
x ∑︂ ∞ (︃ )︂ )︃
R−D n
(︂
= −GR κ 1− Pn (cos ψ) dσ. (6.7)
σ R
n=1

Here, the mass density per unit of surface area κ is

⎧
⎨ρ H if H ⩾ 0,
c
κ=
⎩(ρc − ρw ) H if H < 0,

so we replace ocean depths with equivalent “dry” depths.11 Now we

asteosuus- use again the degree constituent equation, Heiskanen and Moritz (1967)
yhtälö equation 1-71, or our equation 3.8, in the following form:
x
def 2n + 1
κn (ϕ, λ) = κ(ϕ ′ , λ ′ )Pn (cos ψ) dσ ′ .
4π σ

Multiplying both sides by the factor

(︃ )︂ )︃
4πGR R−D n
(︂
− 1−
2n + 1 R

11 This works on dry land and on the ocean. Lakes, glaciers and areas like the Dead Sea
are more complicated.

í Õ ! ¤. û
The “isostatic geoid”
155
and moving it inside the integral, we obtain
(︃ )︂ )︃
4πGR R−D n
(︂
− 1− κn (ϕ, λ) =
2n + 1 R
x (︁ (︃ )︂ )︃
R−D n
(︂
′ ′
Pn (cos ψ) dσ ′ .
)︁
= − GR κ ϕ ,λ 1−
σ R

Summation yields expression 6.7 above:

∑︂
∞
4πGR
(︃ (︂
R−D n
)︂ )︃
− 1− κn (ϕ, λ) =
2n + 1 R
n=1
x ∑︂
∞ (︃ (︂ )︂ )︃
R−D n
′ ′
= − GR κ(ϕ , λ ) 1− Pn (cos ψ) dσ ′ .
σ R
n=1

It follows using equation 6.7 that

∑︂
∞
2
(︃ (︂
R−D n
)︂ )︃
δViso =− R 1− 2πGκn =
2n + 1 R
n=1
∑︂
∞
2
(︃ (︂ )︂ )︃
R−D n
=− R 1− (AB )n .
2n + 1 R
n=1

Here we have used the notation AB = 2πGκ. This represents the

equivalent Bouguer-plate attraction of a mass-density layer κ, and its
degree constituents are (AB )n = 2πGκn .
Let us first look at the contribution from12 0 < n ⩽ N = R D . Then,
/︁

as
R−D n nD
(︂ )︂
≈1− ,
R R

12 The contribution from degree numbers n > R D is

/︁

∞
∑︂ 2R
δViso ≈ − (AB )n ,
2n + 1
n=N+1

where the terms are small and rapidly falling to zero. In this degree range, the mass-
density layer approximation for the topography and its compensation breaks down,
but it hardly matters as these short wavelengths aren’t even isostatically compensated.

í Õ ! ¤. û
156 Geophysical reductions

the following approximation holds:

∑︂
N
2nD ∑︂
N
δViso ≈− (A ) ≈− D (AB )n ≈ −DAB ,
2n + 1 B n
n=1 n=1

and
δTiso DAB
δNiso = γ ≈− γ . (6.8)
epäsuora This is the indirect effect of isostatic reduction.
vaikutus
Let us substitute realistic values. Let the depth of the Mohorovičić13
discontinuity be on average ∼ 20 km.14
On land H ≈ 0.8 km, the Earth’s mean topographic height, and we
obtain δNiso ≈ −1.8 m.
On the ocean H ≈ −3.7 km on average. We must still multiply by the
ratio
ρc − ρw 2670 − 1030
ρc = ,
2670
in order to take the water into account. We obtain δNiso ≈ +5.2 m.

In other words, this effect can be sizeable.

Equation 6.8 is linear in the height H. This means that, under the
continents, the isostatic co-geoid will run on the order of a few metres
below the classical geoid, when on the oceans again it must be a few
metres above the geoid (mean sea level). We may also conclude that
in the isostatic reduction’s effect on the geoid – at least for longer
wavelengths 2πR n , longer than the compensation depth D — all
/︁

wavelengths are represented in the spectrum in approximately in the

same proportions as in the topography itself, and the effect is in fact
proportional to the topography.

13 AndrijaMohorovičić (1857–1936) was a Croatian meteorologist and a pioneer of

modern seismology.
14 Under the continents, the depth is 35 km, under the oceans 7 km below the sea floor,
according to Encyclopædia Britannica (Encyclopaedia Britannica, Moho). Using these
values, we find δNiso = −3.2 m on land, +2.8 m on the ocean.

í Õ ! ¤. û
 Self-test questions
157
^ Self-test questions
1. Which effects are computationally removed in
(a) the simple Bouguer reduction?
(b) the terrain-corrected Bouguer reduction?
(c) the isostatic reduction?
2. Why is the terrain correction always positive?
3. Why do Bouguer anomalies have good interpolation properties,
and on what condition — in other words, which additional infor-
mation must be available at the interpolation stage?
4. How was it discovered that mountains have roots?
5. Explain the isostatic hypotheses of Pratt–Hayford and Airy–
Heiskanen.

^ Exercise 6 – 1: Gravity anomaly

Given point P, height above sea level H = 500 m, local gravity gP =
9.82 m/s2 . Normal gravity at sea level for local latitude φ is γ0 (φ) =
9.820 192 m/s2 .
1. Compute point P’s free-air anomaly ∆g. ilma-anomalia
2. Compute point P’s Bouguer anomaly (without terrain correction)
∆gB .

^ Exercise 6 – 2: Bouguer reduction

1. Point P is 500 m above sea level. Its free-air anomaly is ∆gFA =
25 mGal. Calculate the Bouguer anomaly ∆gB of the point. Forget
about the terrain correction.
2. See section 6.2: Bouguer anomalies. Derive equations 6.2 and
6.3 anew, assuming that the mean density of the Earth’s crust is
ρ = 3370 kg/m3 .

í Õ ! ¤. û
158 Geophysical reductions

Q
600 m
300 m

Q′ Sea level

^ Figure 6.17. Terrain shape.

^ Exercise 6 – 3: Terrain correction and Bouguer

reduction
Given the terrain shape, figure 6.17.
The vertical rock wall PQ is also straight on a map and extends in
both directions (“into” and “out of” the paper) to infinity.
Height differences: PQ ′ = 600 m, QQ ′ = 300 m.
1. Compute the terrain correction at point P (hint: use the formula
for the attraction of a Bouguer plate. We have here a half Bouguer
plate, with only half the attraction of a full one.)
2. Compute the terrain correction at point Q. Algebraic sign?
ilma-anomalia 3. If at point P it is given that the free-air anomaly is 60 mGal, how
much is the Bouguer anomaly at the point? (Use the complete
Bouguer reduction.)
4. If it is given at point Q that the Bouguer anomaly is 10 mGal, how
much is the point’s free-air anomaly?

^ Exercise 6 – 4: Isostasy
Assume Airy–Heiskanen isostatic compensation (figure 6.12). The
density of the Earth’s crust ρc = 2670 kg/m3 , density of the mantle
ρm = 3370 kg/m3 , so the density contrast at the crust-mantle interface is

í Õ ! ¤. û
Exercise 6 – 4: Isostasy
159
700 kg/m3 . Let the reference level for the interface corresponding to zero vertaustaso
topography be −25 km, so t0 = 25 km.
1. Calculate the depth of the “root” of an 8 km high mountain
below the reference level −25 km, assuming it is isostatically
compensated.
2. Mauna Kea is 4 km above sea level, however the surrounding
sea is 5 km deep. How deep is the root of Mauna Kea below the
reference level?
3. How much is the “anti-root” of the surrounding sea above the
reference level? Let the density of sea water be 1030 kg/m3 .
4. So, how deep is the root of Mauna Kea relative to its surroundings?

í Õ ! ¤. û
^ Vertical reference systems

7
^ 7.1 Levelling, orthometric heights and the geoid
Heights have traditionally been determined by levelling. Levelling is
a technique for determining height differences using a level (levelling
instrument) and two rods or staffs. The level comprises a telescope and
a spirit level, and in the measurement situation the telescope’s optical
axis, the sight axis, is pointing along the local horizon. Levelling staffs
are placed on two measurement points, and through the measuring
telescope, measurement values are read off them. The difference
between the two values gives the height difference between the two
points in metres.
The distance between level and staffs is 40–70 m, as longer distances
would cause too large errors due to atmospheric refraction. Longer dis-
tances are measured by repeat measurements using several instrument
stations and intermediate points.
The height differences ∆H thus obtained are not, however, directly
useable. The “height difference” between two points P and Q, obtained
by directly summing the height differences ∆H, depends namely also
on the path chosen when levelling from P to Q. Also the sum of height
∑︁
differences ⃝ ∆H around a closed path is (generally) not zero.

Geometric height is not a conservative field.

– 161 –
162 Vertical reference systems

Levelling staffs Horizontal

line of sight

00

10
Level
back fore
20

View
∆H = back − fore

^ Figure 7.1. The principle of levelling.

This is why, in precise levelling, the height differences are always

converted to potential differences: ∆W = −∆H · g, in which g is the local
gravity, which is either measured or — like in Finland — interpolated
from an existing gravity survey data base. The sum of potential
∑︁
differences around a closed loop is always zero: ⃝ ∆W = 0.
For the potential at an arbitrary terrain point P we find

∑︂
P
WP = W0 − (∆H · g) ,
sea level

the summation being done directly from sea level (potential W0 ) up to

point P. The quantity

∑︂
P
CP = − (WP − W0 ) = (∆H · g) ,
sea level

which is positive above sea level, is called the geopotential number of

point P.
korkeus- W0 is the potential of the national height reference level. In Finland,
vertaustaso the reference level of the old N60 system is in principle the mean sea
level in Helsinki harbour at the beginning of 1960, which is why the

í Õ ! ¤. û
 Orthometric heights
163
system is called N60. However, the precise realisation is a special pillar
in the garden of the Helsinki astronomical observatory in Kaivopuisto.1
The new Finnish height system is called N2000, and the realisation of its
reference level is a pillar at the Metsähovi research station. In practice korkeus-
N2000 heights are, at the decimetric precision level, heights over the vertaustaso
Amsterdam NAP datum.
Other countries have their own, similar height reference or datum
points: Russia has Kronstadt, Western Europe the widely used Ams-
terdam datum NAP, southern Europe has the old Austro-Hungarian
harbour city of Trieste, North America the North American Vertical Da-
tum 1988 (NAVD88) with the datum point Father Point (Pointe-au-Père)2
in Rimouski, Quebec, Canada, etc.

^ 7.2 Orthometric heights

To create a vertical reference, it would be simplest to use the original
geopotential differences from sea level, the geopotential numbers de-
fined above, C = − (W − W0 ), directly as height values. However, this
is psychologically and practically difficult: people want their heights to
be in metres.
Geopotential numbers have clear advantages: they represent the
amount of energy that is needed (for a unit test mass) to move to the
point from the reference level. Fluids — sea water, but also air, or, on
geological time-scales, even bedrock! — always flow downwards and
seek the state of minimum energy.

1 However, the value engraved on the pillar is the reference height of the still older
NN system, not N60. The correct reference value for N60 for this pillar, 30.513 76 m, is
given in the publication Kääriäinen (1966, page 49).
2 Thedistrict Pointe-au-Père of the city of Rimouski was named after the Jesuit priest
Father Henri Nouvel (1621?–1701?), who served forty years with the native population
of New France, today’s Quebec. Pointe-au-Père is also notorious as the location of
the RMS Empress of Ireland shipwreck in 1914, in which over a thousand passengers
perished.

í Õ ! ¤. û
164 Vertical reference systems

Figure 7.2. Height reference pillar in the garden of the Helsinki astronomical
observatory in Kaivopuisto, Kääriäinen (1966). Text:

Suomen Utgångspunkt för

tarkka- precisionsnivellementet
vaakituksen i Finland
pääkiintopiste 30,4652 m öfver noll
30,4652 m yli nollan

(Reference bench mark of the precise levelling of Finland, 30.4652 m

^ above zero).

í Õ ! ¤. û
 Orthometric heights
165
P
WP
∆H3 g
∆H3′

g ∆H2 H ∆H2′

∆H1 ∆H1′
O W0
Geoid

Figure 7.3. Levelled heights and geopotential numbers. The height obtained
∑︁
by summing the levelled height differences, 3i=1 ∆Hi , is not the
∑︁
“correct” height above the geoid, i. e., 3i=1 ∆Hi′ computed along
the plumb line.
The equipotential or level surfaces of the geopotential are not at
all parallel: because of this, a journey along the Earth’s surface
may well go “upwards”, to increasing heights above the geoid,
although the geopotential number decreases. Thus, water may
flow “upwards”.
The gravity vector g is everywhere perpendicular to the level
surfaces, and its length is inversely proportional to the distance
^ separating the surfaces.

In Finland, as in many other countries, orthometric heights have been

long in use. They are physically defined heights above “mean sea level”
or the geoid. See figure 7.3.
The classical geoid is defined as

“The level surface of the Earth’s gravity field that fits on average best to
the mean sea level.”

The orthometric height H of point P is defined as the height obtained

by measuring the distance of P from the geoid along the plumb line. luotiviiva

í Õ ! ¤. û
166 Vertical reference systems

This is a very physical definition, however not a very operational one,

because we (generally) do not get to measure along a plumb line inside
the Earth, and the geoid is not visible there. This is why orthometric
heights are calculated from geopotential numbers: if the geopotential
number of point P is CP , we calculate the orthometric height using the
formula
H = CP g ,
/︁

luotiviiva where g, the average gravity along the plumb line, is

wH
1
g= g(z) dz,
H 0

and z is the measured distance from the geoid along the plumb line.
Because the formula for g already itself contains H, we obtain the
solution iteratively, starting from a crude initial estimate for H. The
suppeneminen iteration converges fast.
We shall see that determining precise orthometric heights is challeng-
ing, especially in the mountains.

^ 7.3 Normal heights

In Finland, currently, with the N2000 height system, normal heights are
used. They are, like orthometric heights, heights above mean sea level.
The mathematical representation of mean sea level in this case is the
quasi-geoid. In sea areas, the quasi-geoid is identical to the geoid. Over
land, it differs a little from the geoid, and in mountainous areas the
difference may be substantial.

^ 7.3.1 Molodensky’s theory

The renowned theorist M. S. Molodensky (figure 7.5) developed a theory
in which the height of a point from “mean sea level” is defined by the
following equation:
def
H∗ = C γ0H ,
/︁

í Õ ! ¤. û
 Normal heights
167
South
North
Lake Päijänne: C = − (W − W0 ) = 76.9 GPU gS

gN HS
Lake Päijänne
HN
Geoid: W = W0

Figure 7.4. In terms of orthometric heights, water may sometimes flow “up-
wards”. Although the north and south ends of Lake Päijänne are
on the same geopotential level — 76.9 geopotential units below
that of mean sea level — the orthometric height of the south end
HS is greater than that of the north end HN , because local gravity
g is stronger in the north than in the south. The height difference
in the case of Lake Päijänne is 8 mm (Jaakko Mäkinen, personal
communication). Calculation using the normal gravity field yields
6 mm. The balance of 2 mm comes from the difference between
^ the gravity anomalies at the northern and southern ends.

in which γ0H is the average normal gravity computed between the zero
level (reference ellipsoid) and H∗ along the ellipsoidal normal. So, the vertaus-
method of computing is the same as in the case of orthometric heights, ellipsoidi
but using the normal gravity field instead of the true gravity field.
Heights “above sea level” are for practical reasons given in metres.
For large, continental networks we want to give heights above a compu-
tational reference ellipsoid in metres, and thus heights above “sea level”
also have to be in metres.
Molodensky also proposed that instead of the geoid, height anomalies
would be used, the definition of which is
def /︁
ζ = T γHh , (7.1)

in which now γHh is the average normal gravity at terrain level; more
precisely: the average of normal gravity along the ellipsoidal normal
over the interval z ∈ H∗ , h , in which H∗ is the normal height of the
[︁ ]︁

í Õ ! ¤. û
168 Vertical reference systems

Figure 7.5. Mikhail Sergeevich Molodensky (1909–1991), source obscure.

More photographs and background information in Brovar et al.
^ (2000).

vertaus- point and h its height from the reference ellipsoid. The parameter z is
ellipsoidi the distance from the reference ellipsoid reckoned along the ellipsoidal
normal. T is the disturbing potential at the point.
Based on these assumptions, Molodensky showed that

H∗ + ζ = h.

This equation is very similar to the corresponding one for orthometric

heights and geoid heights

H + N = h.

Also otherwise ζ, the height anomaly, also called “quasi-geoid height”, is

very close to N, and correspondingly H∗ is close to H.

^ 7.3.2 Molodensky’s realisation

The Molodensky school realised that, because normal gravity along the
luotiviiva plumb line is very close to a linear function of place, one could define a

í Õ ! ¤. û
Normal heights
169
height type that can be computed directly from geopotential numbers,
and that also would be compatible with similarly defined, so-called
height anomalies, and with geometric heights h reckoned from the
reference ellipsoid. vertaus-
ellipsoidi
The geometric height h from the reference ellipsoid may be connected
to the potential U of the normal gravity field indirectly, though the
following integral equation:
wh
U = U0 − γ(z) dz.
0

Here, U is the normal potential and γ normal gravity. One level surface
of U, U = U0 , is also the reference ellipsoid. The variable z is the
distance from the ellipsoid along its local normal.3
By defining
wh
1
def
γ0h = γ(z) dz (7.2)
h 0

we obtain
U − U0
h=− .
γ0h
By using W = U + T and dividing by γ0h we obtain

W − W0 T
= −h
γ0h γ0h

assuming W0 = U0 , the normal potential on the reference ellipsoid.

Next, one could define
? W − W0
H+ = −
γ0h

as a new height type, and

? T
N+ = h − H+ =
γ0h

3 Here we ignore that the normal gravity vector →

−γ (z) is for z ̸= 0 not precisely parallel
with the ellipsoidal normal: the curvature of the field lines of the normal gravity field
or normal plumb lines, section 4.3.2.

í Õ ! ¤. û
170 Vertical reference systems

as the corresponding new geoid height type. It has however the aesthetic
flaw that we divide here by the average normal gravity computed
between the levels 0 and h. This quantity is not operational without a
means of determining the ellipsoidal height h.
This suggests the following improvement based on the circumstance
that γ(z) is a nearly linear function. This means that the vertical
d
derivative dz γ is nearly constant in the height interval considered.
We define in addition to equation 7.2:
w H+ wh
def 1 def 1
γ0H = γ(z) dz, γHh = γ(z) dz.
H+ 0 N+ H+

Now
(︃ )︃
1 + dγ N+
γ0H ≈ γ0h − 2
N ≈ γ0h 1 − , (7.3)
dz R
(︃ )︃
1 + dγ H+
γHh ≈ γ0h + 2 H ≈ γ0h 1 + . (7.4)
dz R
d d
γ ≈ 2γ R .
/︁
R is the Earth’s radius in spherical approximation: dz γ ≈ dr
+/︁ +/︁
Next, we also exploit that both N R and H R are ≪ 1, so
(︃ )︃−1 (︃ )︃ (︃ )︃−1 (︃ )︃
N+ N+ H+ H+
1− ≈ 1+ , 1+ ≈ 1− ,
R R R R

and with equations 7.3, 7.4, and the definitions above of H+ and N+ ,
(︃ )︃
defT T γ0h + H+ N+ H+
ζ= = · ≈N 1− = N+ − ,
γHh γ0h γHh R R
(︃ )︃
∗ def W − W0 W − W0 γ0h + N+
H =− =− · ≈H 1+ =
γ0H γ0h γ0H R
N+ H+
= H+ + .
R
+ +/︁
Because the, already small, correction terms N H R cancel, we finally
obtain
H∗ + ζ = H+ + N+ = h. (7.5)

í Õ ! ¤. û
 Normal heights
171
The quantity γ0H , and thus also normal height H∗ , can be, unlike γ0h ,
computed using only information obtained by (spirit or trigonometric)
levelling, without having to know the height h above the reference vertaus-
ellipsoid, which would again require knowledge of the local geoid. ellipsoidi

This was Molodensky’s realisation (Molodensky et al., 1962) as early

as in 1945, long before the Global Positioning System GPS, or a global,
geocentric reference ellipsoid, existed. Back then, continental triangula-
tion networks, like the one of the Soviet Union, were computed on their
own, regionally defined reference ellipsoids.
+ +/︁
The size of the correction term N H R is, for heights of the global
geoid up to 110 m, 17 mm for each kilometre of terrain height. The
errors remaining after applying this term are microscopically small,
because normal gravity is, unlike true gravity, extremely linear along the
plumb line — as equations 7.3 and 7.4 already assumed. luotiviiva
Figure 7.6 attempts to visualise the derivation.

^ 7.3.3 Normal height and height anomaly

Normal height

C W − W0
H∗ = =− , (7.6)
γ γ
in which (recursive definition!)
w H∗
1
γ = γ0H = γ(z) dz.
H∗ 0

Height anomaly
W−U T
ζ= = ,
γHh γHh
in which w
1 h
γHh = γ(z) dz.
ζ H∗
The height anomaly ζ, which otherwise is a quantity similar to the
geoid height N, is however located at the level of the topography,
not at sea level. The surface formed by points which are a distance

í Õ ! ¤. û
172 Vertical reference systems

+ +
− N RH
N+
ζ

h
H∗
H+
N + H+
R

γ(z)

Figure 7.6. A graphic cartoon of the proof of Molodensky’s realisation. The

blue and red areas, which are equal, represent the correction terms
which convert N+ to ζ and H+ to H∗ , respectively. The red and
blue arrows stand for the conversion process. The balls represent
^ midpoints of averaging intervals for the function γ (z).

Topography
ζ
Telluroid
H∗ H h

H∗ Nζ
Geoid Reference ellipsoid Quasi-geoid

Figure 7.7. Geoid, quasi-geoid, telluroid and topography. Note the correlation
between the quasi-geoid and topography. Depicted is the situation
^ where N > 0.

í Õ ! ¤. û
 Difference between geoid height and height anomaly
173
H∗ above the reference ellipsoid (and thus a distance ζ below or vertaus-
−ζ above the topography), is called the telluroid. It is a mapping ellipsoidi
of sorts of the topographic surface: the set of points Q whose
normal potential UQ is the same as the true potential WP of the
true topography’s corresponding point P. See figure 5.4.
Often, as a concession to old habits, we construct a surface that
is at a distance ζ above or a distance −ζ below the reference
ellipsoid. This surface is called the quasi-geoid. It lacks physical
meaning: it is not a level surface, although out at sea it coincides
with the geoid. Its short-wave features, unlike those of the geoid,
correlate with the short-wavelength features of the topography.
Height above the ellipsoid (assumed U0 = W0 )

U − U0
h= ,
γ0h

where wh
1
γ0h = γ(z) dz.
h 0

The relationship between the three quantities is

h = H∗ + ζ.

In all three cases, the quantity is defined by dividing the potential

difference by some sort of “average normal gravity”, suitably computed
along a segment of the local plumb line. In the case of the height luotiviiva
anomaly ζ, a piece of plumb line is used high up, close to the topographic
surface, between level H∗ (telluroid) and level h (topography).

^ 7.4 Difference between geoid height and height

anomaly
Normal heights are very operational. They are always used together
with “quasi-geoid” heights — more correctly: height anomalies — ζ.

í Õ ! ¤. û
174 Vertical reference systems

Orthometric heights — for example Helmert heights — on the other

hand are always used together with geoid heights N. For computing
both, H and N, one needs the topographic mass density ρ, for which
a standard constant value is often assumed (2670 kg/m3 ), and the local
vertical gradient of gravity, for which generally the vertical normal
gravity gradient (−0.3084 mGal/m) is assumed.
The difference between height anomaly and geoid height is calculated
as follows.
1. First, calculate the separation between the quasi-geoid and the
“free-air geoid”. The free-air geoid is an equipotential surface of
the harmonically downwards continued exterior potential. If TFA is
the disturbing potential of the exterior, harmonically downwards
continued field, then its difference between topography level and
sea level is:
w H ∂T (z)
FA
TFA (H) − TFA (0) = dz ≈ −∆gFA H, (7.7)
0 ∂z

and by using the Bruns equation twice, ζ = T (H) γ = TFA (H) γ

/︁ /︁

(height anomaly or quasi-geoid height) and NFA = TFA (0) γ

/︁

(”free-air geoid height”, FA = Free Air), we obtain4

∆gFA H
ζ − NFA ≈ − γ . (7.8)

2. Thus we have obtained the difference between the height anoma-

lies and heights of the “free-air geoid”. It remains to determine
the separation between the “free-air geoid” and geoid.
Let us approximate the topography by a Bouguer plate. Then
◦ In the case of the “free-air geoid” NFA the thickness of the
plate is the height H of point P. This is because the free-air
geoid is based on the harmonically downwards continued

4 Here we made the approximation that γ is the same on the topography level as at
sea level.

í Õ ! ¤. û
Difference between geoid height and height anomaly
175
exterior field, meaning that the Bouguer-plate attraction
acting at P must also be continued downwards, i. e., taken
fully into account.
Because the surface mass density of the plate is Hρ, its assumed
attraction is everywhere on the plumb line of point P: luotiviiva

2πGρ H. (7.9)

◦ Now in the case of the geoid height N = T (0) γ , we have to be

/︁

physically realistic: in an arbitrary location z on the plumb

line of point P, the Bouguer plate is partly below the location,
and partly above it. The attraction is then only
2πGρ z − 2πGρ (H − z) = 2πGρ (2z − H) . (7.10)

By integrating the difference between equations 7.9 and 7.10, like

we did for equation 7.7, we obtain
w H (︁ )︁
T (0) − TFA (0) = 2πGρ (2z − H) − H dz =
0
]︁z=H
= 2πGρ z2 − 2Hz z=0 = −2πGρH2 ≈ −AB H,
[︁

in which AB is the attraction of a Bouguer plate of thickness H.

We obtain again by dividing the equation by normal gravity:
AB H
N − NFA = − γ .

By subtracting this latest result from equation 7.8, we find

(−∆gFA + AB ) H ∆g H
ζ−N= γ = − γB . (7.11)

See also Heiskanen and Moritz (1967, pages 327–328). As the Bouguer
anomaly ∆gB is strongly negative in the mountains, it follows that the
quasi-geoid is there always above the geoid: approximately, using
equation 6.2:
0.1119 mGal/m 2
ζ−N≈ H ≈ 10−7 m−1 · H2 .
9.81 m/s2
Or, if H is in units of km and ζ − N in units of m:
ζ − N ≈ 0.1 m/km2 · H2 .

í Õ ! ¤. û
176 Vertical reference systems

^ 7.5 Difference between orthometric and normal

heights
The geoid is the level from which orthometric heights are measured.
Therefore, we may write
h = H + N,
vertaus- in which h is the height above the reference ellipsoid, and H is the
ellipsoidi orthometric height.

We may also bring back to memory equation 7.5:

h = H∗ + ζ ,

in which ζ is the height anomaly, and H∗ is the normal height.

We obtain simply
∆gB H
H − H∗ = ζ − N = − γ , (7.12)

using equation 7.11.

^ 7.6 Calculating orthometric heights precisely

Orthometric heights are a traditional way of expressing height “above
sea level”. Orthometric heights are heights above a real geoid, i. e., a
level surface inside the Earth that is, in the mean, located at the same
level as the mean sea level.
We may write
wH
W = W0 − g(z) dz,
0

in which g is the true gravity inside the topographic masses. From this
we obtain
C − (W − W0 )
H= = ,
g g
luotiviiva in which the mean gravity along the plumb line is
wH
1
g= g(z) dz.
H 0

í Õ ! ¤. û
Calculating orthometric heights precisely
177
The method is recursive: H appears on both the left and right sides. This
is not a problem: both H and g are obtained iteratively. Convergence is suppeneminen
fast.
In practice one calculates orthometric height using an approximate
formula. In Finland, Helmert orthometric heights have long been used, for
(︁ )︁
which gravity measured on the Earth’s surface, g H , is extrapolated
downwards by using the estimated vertical gravity gradient interior
to the rock. It is assumed that its standard value outside the rock, the
value −0.3084 mGal/m (the free-air gradient), changes to a value that is ilmagradientti
0.2238 mGal/m greater (twice the standard-density 2670 kg/m3 Bouguer-
plate effect5 ): the end result is the total inside-rock gravity gradient,
−0.0846 mGal/m.
This is called the Prey6 reduction. As the end result we obtain the
following equations (the coefficient is half the gravity gradient, so the
mean gravity along the plumb line is the same as gravity at the midpoint luotiviiva
of the plumb line):
g = g(H) − 0.0846 mGal/m − 12 H = g(H) + 0.0423 mGal/m · H,
(︁ )︁

thus
C C
H= = , (7.13)
g g(H) + 0.0423 mGal/m · H
in which C is the geopotential number (potential difference with mean
sea level) and g(H) is gravity at the Earth’s surface. See also Heiskanen
and Moritz (1967, pages 163–167). The term 0.0423 mGal/m · H is typically
much smaller than g(H), which is about 9.81 m/s2 = 981 000 mGal! So, an
iteration in which the denominator is first calculated using a crude H
value, converges very fast.
The use of Helmert heights as an approximation to orthometric
heights is imprecise for the following reasons:

5 In Finland, however, density values read from a geological map were used.
6 AdalbertPrey (1873–1949) was an Austrian astronomer and geodesist and an author
of textbooks.

í Õ ! ¤. û
178 Vertical reference systems

luotiviiva ◦ The assumption that gravity changes linearly along the plumb
line. This is not the case, especially not because of the effect of the
surrounding terrain. In the precise computation of orthometric
heights, one ought to compute the terrain correction separately
for every point on the plumb line.
◦ The assumption that the free-air vertical gravity gradient is a
constant, −0.3084 mGal/m. This is not the case, the gradient can
easily vary by ±10 %.
◦ The assumption that the rock density is ρ = 2670 kg/m3 . The true
density value may easily vary by ±10 % or more around this
assumed value.
The first approximation, neglecting the terrain effect, can be corrected
by using Niethammer’s7 method, see Heiskanen and Moritz (1967,
page 167). It requires that, in geoid computation, too, the terrain is
correspondingly taken into account.
The third approximation, the density, can be removed as a problem
by conventionally agreeing to also use a standard density ρ = 2670 kg/m3
in the corresponding geoid computation. The surface thus obtained
is not any more a true geoid then, but a “fake geoid”, for which no
suitable name comes to mind.
The second approximation could be eliminated by using the true
painovoiman free-air gravity gradient instead of a standard value. However, the true
ilmagradientti free-air gradient depends sensitively on local crustal density variations.
Moreover, the value of the free-air gradient on the Earth’s surface is not
precisely representative for the downwards continued free-air gradient
along the whole plumb line. To compute the gradient, one can use the
Poisson equation, on which more later.
The precise calculation of orthometric heights is thus laborious: just
as laborious as the precise determination of the geoid, and for the

7 Theodor Niethammer (1876–1947) was a Swiss astronomer and geodesist who was
the first to map the gravity field of the Swiss Alps.

í Õ ! ¤. û
 Calculating normal heights precisely
179
same reasons. Fortunately in non-mountainous countries, Helmert
heights are good enough. In Finland they were even computed using
for the ρ values “true” crustal densities according to a geological map
(Kääriäinen, 1966, page 32).

^ 7.7 Calculating normal heights precisely

For this we use equation 7.6:
C W − W0
H∗ = =− ,
γ γ
where the average value of normal gravity along the plumb line is luotiviiva
w ∗
1 H
γ = γ0H = ∗ γ(z) dz.
H 0
Because normal gravity is in good approximation a linear function of z,
we may write
∂γ
γ = γ0 + 12 H∗ ,
∂z
∂ def
in which ∂z γ = −0.3084 mGal/m and γ0 (φ) = γ(φ, 0) is normal gravity
computed at height zero. We obtain

γ = γ0 − 0.1542 mGal/m · H∗ .

The solution is again obtained iteratively:

C C
H∗ = = (7.14)
γ γ0 − 0.1542 mGal/m · H∗
in which γ0 (φ) can be calculated exactly when local latitude φ is
known. H∗ appears on both sides of the equation, but the iterative
solution converges fast due to the first term of the denominator γ0 , some suppeneminen
9.81 m/s2 = 981 000 mGal, being a lot larger than 0.1542 mGal/m · H∗ .
Calculation of normal heights, unlike calculation of orthometric
heights, is not sensitive to Earth crustal density hypotheses. It depends,
however, on the choice of normal gravity field, i. e., the reference vertaus-
ellipsoid. ellipsoidi

í Õ ! ¤. û
180 Vertical reference systems

^ 7.8 Calculation example for heights

At point P the potential difference with the sea level is C = 5000 m2/s2 .
Local gravity is g = 9.820 000 m/s2 .
Normal gravity calculated at level zero under point P equals γ0 =
9.821 500 m/s2 .
Questions
1. Calculate the orthometric height of point P.
ilma-anomalia 2. Calculate the free-air gravity anomaly ∆gFA of point P.
3. Calculate the Bouguer anomaly (without terrain correction)
∆gB of point P.
4. Calculate the normal height of point P.
5. If the geoid height at point P is N = 25.000 m, how much is
the height anomaly (“quasi-geoid height”) ζ?
Answers
1. First attempt:

C 5000
H(0) = g = m = 519.165 m.
9.82
Second attempt (equation 7.13):

5000 m2/s2
H(1) = =
9.820 000 m/s2 + 0.0423 · 10−5 s−2 · 519.165 m
= 509.154 m.

After that, the millimetres no longer change.

2. The free-air anomaly is

∆gFA = 9.820 000 m/s2 −

− 9.821 500 − 0.3084 · 10−5 · 509.154 m/s2 =
(︁ )︁

= 7.023 mGal.

í Õ ! ¤. û
 Orthometric and normal corrections
181
3. The Bouguer anomaly is (equation 6.2):

∆gB = ∆gFA − 0.1119mGal/m · H = −49.951 mGal.

4. The first attempt is again

C
H∗(0) = γ = 509.087 m.
0

The second, equation 7.14:

5000 m2/s2
H∗(1) = =
9.821 500 m/s2 − 0.1542 · 10−5 s−2 · 509.087 m
= 509.128 m,

also final on the millimetre level.

5. The difference equation 7.12 yields
∆gB H
ζ−N=− γ = 0.026 m.
Also (check) H − H∗ = 0.026 m. So

ζ = N + 0.026 m = 25.026 m.

^ 7.9 Orthometric and normal corrections

In practical orthometric height calculations, one often starts by adding
together the height differences ∆H measured by levelling (“staff-reading
differences”) between points A and B as a tentative or crude height
difference
∑︂
B
∆H,
A

after which the non-exactness of this method is accounted for by

applying the “orthometric correction” (OC):

∑︂
B
HB = HA + ∆H + OCAB .
A

í Õ ! ¤. û
182 Vertical reference systems

The fact that the difference in orthometric heights between two points
A and B is not equal to the sum of the levelled height differences is due
to gravity not being the same everywhere.
With CA , CB and ∆C the geopotential numbers at A and B, and the
geopotential differences along the levelling line, we have CB − CA −
∑︁B
A ∆C = 0 because of the conservative nature of the geopotential.
Dividing by a constant γ0 yields

CB CA ∑︂ ∆C
B

γ0 − γ0 − γ0 = 0.
A

On the other hand, we have

∑︂
B
C C ∑︂ ∆C B
OCAB = HB − HA − ∆H = B − A − g ,
gB gA
A A

luotiviiva with gA , gB average gravity values along the plumb lines of A and B
and g gravity along the levelling line. In this expression, we compare
∑︁B
A ∆H, the naively calculated sum of levelled height differences, with
the difference between the orthometric heights of the end points A and
B, calculated according to the definition.
Subtraction yields
(︃ )︃ (︃ )︃ ∑︂
B (︂
CB CB CA CA ∆C ∆C
)︂
OCAB − 0 = − γ − − γ − g − γ0 ,
gB 0 gA 0
A

in which
(︃ )︃ (︃ )︃
CB CB γ0 − gB CB γ0 − gB
− γ = γ0 = γ0 HB ,
gB 0 gB
(︃ )︃
CA CA γ0 − gA
− γ = γ0 HA ,
gA 0

∆C ∆C γ0 − g
(︂ )︂
g − γ0 = γ0 ∆H,

yielding the orthometric correction

∑︂
B (︂
g − γ0
)︂ (︃
gA − γ0
)︃ (︃
gB − γ0
)︃
OCAB = γ ∆H + γ HA − γ HB , (7.15)
0 0 0
A

í Õ ! ¤. û
Orthometric and normal corrections
183
which is identical to Heiskanen and Moritz’s (1967) equation 4-33.
The choice of the constant γ0 is arbitrary; it is wise to choose it close
to the average gravity in the area of the levelling line AB, so as to keep
the numerics small.
Similarly we may also calculate the normal correction (NC) in calculat-
ing normal heights. Start from the equation

∑︂
B
CB CA ∑︂ ∆C
B
NCAB = H∗B − H∗A − ∆H = − − g , (7.16)
γB γA
A A

from which, as above, follows by subtraction:

∑︂
B (︂
g − γ0
)︂ (︃
γA − γ0
)︃ (︃
γB − γ0
)︃
∗
NCAB = γ0 ∆H + γ0 HA − γ0 H∗B . (7.17)
A

The identical first term in both equation 7.15 and equation 7.17 can be
traced back to the term
∑︂
B
∆C ∑︂
B

g = ∆H,
A A

the naive summation of height differences ∆H in the case of both

orthometric and normal correction, which is the generic basis of the
concept of both corrections.
Equation 7.16 yields

∑︂
B
H∗B = H∗A + ∆H + NCAB .
A

What changes between the orthometric and normal corrections is the

definition of heights: H∗ instead of H, requiring division by the average
of normal gravity along the plumb line γ, not by that of true gravity g. luotiviiva
Both the orthometric correction 7.15 and the normal correction 7.17
can be calculated one staff interval at a time: one must know, in addition
to the levelled height difference ∆H, local gravity g along the levelling

í Õ ! ¤. û
184 Vertical reference systems

line. Furthermore one must know g(H) or γ(0) at both end points in
luotiviiva order to calculate mean gravity g or γ along the plumb lines of those end
points. All this goes well with the equations given above. Remember
that gravity g along the levelling line is needed in any case in order to
reduce the individual levelled height differences ∆H to geopotential
number differences ∆C. This reduction is part of the computation of
both the orthometric and the normal correction.

^ 7.10 A vision for the future: relativistic levelling

According to general relativity, the deeper a clock is inside the potential
well of masses, the slower it ticks. This is most easily seen by looking at
metriikka the Schwarzschild8 metric for a spherically symmetric field:

c2 dτ2 =
2GM 2 2 2GM −1 2
(︂ )︂ (︂ )︂
dr − r2 dϕ2 + cos2 ϕ dλ2 =
(︁ )︁
= 1− 2 c dt − 1 − 2
cr cr
2W 2 2 2W −1 2
(︂ )︂ (︂ )︂
dr − r2 dϕ2 + cos2 ϕ dλ2 ,
(︁ )︁
= 1 − 2 c dt − 1 − 2
c c
in spherical co-ordinates plus time (ϕ, λ, r, t) . Here we see how the rate
ominaisaika of proper time τ is slowed down compared to stationary co-ordinate
time t (time at infinity r → ∞), when the geopotential W increases
closer to the mass. The slowing-down ratio is
√︃
∂τ 2W W
= 1− 2 ≈1− 2.
∂t c c

Now c2 , the speed of light squared, is, in the units of daily life, a huge
number: 1017 m2/s2 . This means that measuring a potential difference
of 1 m2/s2 — corresponding to a height difference of 10 cm — using this

8 Karl Schwarzschild (1873–1916) was a German physicist who was the first to derive, in

1915 while serving on the Russian front, a closed spherically symmetric, non-rotating
solution to the field equation of Albert Einstein’s general theory of relativity, the
Schwarzschild metric.

í Õ ! ¤. û
 A vision for the future: relativistic levelling
185
Braunschweig

100 km

Garching

Figure 7.8. An optical lattice clock: the ultra-precise atomic clock of the future
operates at optical wavelengths. To the right, the trajectory of the
^ Predehl et al. (2012) experiment.

method, requires a precision of 1 : 1017 . More traditional, microwave-

based atomic clocks can do precisions of 10−12 –10−14 (Vermeer, 1983a).
With the new optical clocks, the objective should be achievable and
relativistic levelling may become a reality.
The clock works in this way: an extremely cold, so-called Bose–
Einstein condensate of atoms is trapped inside an optical lattice formed valohila
by six laser beams, an electromagnetic pattern of standing waves. The
readout beam of the clock oscillation uses a different frequency. A
Bose–Einstein condensate has the property that all atoms are in precisely
the same quantum state — like the photons in an operating laser: their
matter waves are coherent. In a way, all the atoms together act as one
virtual atom.
The condensate may consist of millions of atoms, and can actually be
seen through the window of the vacuum chamber as a small plasma blob.

í Õ ! ¤. û
186 Vertical reference systems

Unfortunately it is not enough that just one laboratory measures time

to extreme precision. One has also to be able to compare the ticking
rates of different clocks over geographical distances. For this, a solution
has also been found: existing fibreoptic cables already in global use for
Internet and telephony are useable for this with small modifications.
The modifications concern the amplifiers in the cables at distances of
some 100 km, which must be replaced by modified ones (Predehl et al.,
2012). In this way, both the traditional precise levelling networks and
the height systems based on GNSS technology and geoid determination
may be replaced by this hi-tech (and hi-science!) solution.

^ Self-test questions
1. Why are heights calculated directly from levelled height differ-
ences not good enough as a height system?
2. What is a geopotential number?
3. What are orthometric heights?
4. What are normal heights?
5. What is the classical definition of the geoid?
6. What is a height anomaly?
7. What is the quasi-geoid?
8. Why might water sometimes flow in the “wrong” direction, to a
greater height?
9. What is the telluroid?
10. What are the orthometric correction and the normal correction?

^ Exercise 7 – 1: Calculating orthometric heights

The potential difference with sea level at point P, − (W − W0 ), equals
1000 m2/s2 . Gravity at the point is gP = 9.820 000 m/s2 . Calculate the
orthometric height of the point. Aim for millimetre precision.

í Õ ! ¤. û
 Exercise 7 – 2: Calculating normal heights
187
^ Exercise 7 – 2: Calculating normal heights
At point P, the potential difference with sea level is

− (W − W0 ) = 5000 m2/s2 .

Below the point at sea level, normal gravity is γ0 = 9.821 500 m/s2 .
Calculate the normal height of the point.

^ Exercise 7 – 3: Difference between orthometric and

normal height
At point P, the Bouguer anomaly is ∆gB = −120 mGal. The orthometric
height of the point is 1150 m.
1. Calculate the normal height of point P.
2. If the geoid height in point P is N = 21.75 m, calculate the height
anomaly ζ of the point.

í Õ ! ¤. û
^ The Stokes equation and other
integral equations

^ 8.1 The Stokes equation and the Stokes integral

kernel
8
The following assumes a spherical Earth. By suitably combining the
equations in section 5.3, one obtains at sea level
∑︂
∞ ∑︂
∞
∆gn
T= Tn = R ,
n−1
n=2 n=2

with Tn = Tn (ϕ, λ) the degree constituents of the disturbing potential

field T = T (ϕ, λ), and ∆gn = ∆gn (ϕ, λ) those of the gravity anomaly
field. The summation starts from n = 2; for the degree numbers n = 0, 1,
the ∆gn are assumed to vanish, as ∆g0 ̸= 0 would mean a different total
mass for the normal field than for the Earth, and ∆g1 ̸= 0 an offset of
the co-ordinate origin from the Earth’s centre of mass, see section 3.4.
This is now the Stokes equation’s spectral form.
Substituting into this degree constituent equation 3.8, one obtains the asteosuus-
integral equation yhtälö

R ∑︂ 2n + 1
∞ x
T= ∆gPn (cos ψ) dσ =
4π n−1 σ
n=2
∑︂
(︄ ∞
x x
)︄
R 2n + 1 R
= Pn (cos ψ) ∆g dσ = S(ψ)∆g dσ,
4π σ n−1 4π σ
n=2

– 189 –
190 The Stokes equation and other integral equations

Mass excess

Mass
deficit
−N

^ Figure 8.1. The principle of gravimetric geoid determination.

in which
∑︂
∞
2n + 1
S(ψ) = Pn (cos ψ),
n−1
n=2

the Stokes kernel function. The angle ψ is the geocentric angular

distance between the evaluation point and moving observation point,
see figure 8.2. The equation above allows the calculation, from global
gravimetric data and for every point on the surface of the Earth sphere,
of the disturbing potential T , and from that, the geoid height N using
Bruns equation 5.2, N = T γ . The result is
/︁

í Õ ! ¤. û
 The Stokes equation and the Stokes integral kernel
191

N(ϕ, λ) S(ψ)
Evaluation
point
∆g ϕ ′ , λ ′ dϕ dλ
(︁ )︁

Moving data or
ψ integration point
Earth’s centre

^ Figure 8.2. Integrating the Stokes equation geometrically.

T (ϕ, λ) x
R (︁ ′ ′ )︁ ′
N(ϕ, λ) = γ = S(ψ) ∆g ϕ , λ dσ , (8.1)
4πγ σ

in which (ϕ, λ) and (ϕ ′ , λ ′ ) are the evaluation point and the moving
point (“data point”), respectively, and the angular distance between
them is ψ. Equation 8.1 is the classical Stokes equation of gravimetric
geoid determination.
The above illustrates the correspondence between integral equations
and spectral expansions. There are other examples of this, like the
spectral representation of the function 1 ℓ , equation 8.6, Heiskanen
/︁

and Moritz’s (1967) equation 1-81. Of course 1 ℓ is also the kernel

/︁

function of an integral equation, the one yielding the potential V if the

single-layer mass density κ is given.
A version of the Stokes equation for the exterior space also exists. We
gave it earlier, equation 5.9. The spectral form of its kernel function is
equation 5.10:
∑︂
∞ (︂ )︂n+1
R 2n + 1
S(ψ, r, R) = P (cos ψ).
r n−1 n
n=2

The Stokes kernel function on the Earth’s surface is depicted in figure

í Õ ! ¤. û
192 The Stokes equation and other integral equations

25
S(ψ)
20 1
sin 21 ψ
−6 sin 12 ψ + 1 − 5 cos ψ
15
−3 cos ψ ln sin 21 ψ + sin2 12 ψ
(︁ )︁

10
S(ψ)−→

ψ −→
−5
0 0.5 1 1.5 2 2.5 3 3.5

Figure 8.3. The Stokes kernel function S(ψ). The argument ψ is in radians
[0, π). Also plotted are the three parts of analytical expression 8.2
^ with their different asymptotic behaviours.

8.3, in which the angle ψ is in radians (1 rad ≈ 57◦. 29578). The curve
was calculated using the following closed expression (Heiskanen and
Moritz, 1967, section 2-16, equation 2-164):
1
− 6 sin 12 ψ + 1 − 5 cos ψ − 3 cos ψ ln sin 21 ψ + sin2 12 ψ .
(︁ )︁
S(ψ) = 1
sin 2 ψ
(8.2)
This closed expression helps us to understand better how the function
behaves close to the origin ψ = 0: the first term, 1 sin 12 ψ , goes to
/︁

infinity when ψ → 0. The next three terms, −6 sin 12 ψ + 1 − 5 cos ψ, are

all bounded on the whole interval [0, π) and the limit for ψ → 0 is −4.
The last, complicated term −3 cos ψ ln sin 12 ψ + sin2 21 ψ also goes to
(︁ )︁

infinity — positive infinity! — for ψ → 0, but much more slowly because

of the logarithm.

í Õ ! ¤. û
 Example: The Stokes equation in polar co-ordinates
193
^ 8.2 Example: The Stokes equation in polar
co-ordinates
In section 2.3 we derived a general solution to the Laplace equation
in two dimensions in polar co-ordinates. Below we develop a “toy”
computational framework for gravimetric geoid determination in two
dimensions, which allows us to do simple numerical simulations in
order to get a feel for the behaviour of these things.
Firstly we derive the disturbing potential, gravity anomaly, and
Stokes integral kernel for this solution, equation 2.3, assuming a normal
potential U(r) = a0 + b0 ln r.
◦ Disturbing potential:

T (α, r) = V2 (α, r) − (a0 + b0 ln r) =

∑︂
∞
= r−k (ak cos kα + bk sin kα) .
k=1

◦ Normal gravity:
∂U b
γ(r) = − = − r0 .
∂r
◦ Normal gravity gradient:

∂γ ∂2 U b
= − 2 = 20 .
∂r ∂r r
◦ Gravity anomaly, equation 5.5:

∂T T ∂γ
∆g(α, r) = − + =
∂r γ ∂r
∑︂
∞
k −k
= r r (ak cos kα + bk sin kα) +
k=1

1 ∂γ ∑︂ −k
∞
+γ r (ak cos kα + bk sin kα) =
∂r
k=1
∑︂
∞ (︃ )︃
k 1 ∂γ −k
= r + γ ∂r r (ak cos kα + bk sin kα) =
k=1

í Õ ! ¤. û
194 The Stokes equation and other integral equations

∑︂
∞
k − 1 −k
= r r (ak cos kα + bk sin kα) .
k=2

We see that, if we write

∑︂
∞ (︂ )︂k
R def
T (α, r) = r Tk (α), Tk (α) = R−k (ak cos kα + bk sin kα) ,
k=1

it follows that
∑︂
∞ (︂ )︂k+1
R
∆g(α, r) = r ∆gk (α),
k=2
def
∆gk (α) = (k − 1) R−(k+1) (ak cos kα + bk sin kα) ,

and, like in the case of spherical co-ordinates,

k−1
∆gk (α) = T (α). (8.3)
R k
kantafunktio According to Fourier theory, the basis functions cos kα, sin kα are ortho-
normal on the circle r = R when choosing the following integral as the
scalar product:
⎧
w 2π w 2π ⎨0 if k ̸= m,
1 1
cos kα cos mα dα = sin kα sin mα dα =
π 0 π 0 ⎩1 if k = m,
w
1 2π
π cos kα sin mα dα = 0 always.
0

This means that, with

∑︂
∞
∆g(α, R) = ∆gk (α),
k=2

we may decompose ∆g(α) into its Fourier terms as follows:

def
∆gk (α) = (k − 1) R−(k+1) (ak cos kα + bk sin kα) =
A B
⏟ ⏞⏞k ⏟ ⏟ ⏞⏞k ⏟
−(k+1) −(k+1)
= (k − 1) R ak cos kα + (k − 1) R bk sin kα.

í Õ ! ¤. û
Example: The Stokes equation in polar co-ordinates
195
This yields the following Fourier coefficients:
{︄ }︄ {︄ }︄
Ak −(k+1) ak
= (k − 1) R , k = 2, 3, · · ·
Bk bk

and on the circle r = R the expansion is

∑︂
∞ ∑︂
∞
∆g(α, R) = ∆gk (α) = (Ak cos kα + Bk sin kα) .
k=2 k=2

The substitutions
{︄ }︄ {︄ }︄
ak Rk+1 Ak
=
bk k−1 Bk

yield

∑︂
∞ ∑︂
∞
T (α, R) = Tk (α) = R−k (ak cos kα + bk sin kα) =
k=2 k=2
∑︂
∞ (︃
Rk+1 Rk+1
)︃
−k
= R A cos kα + B sin kα =
k−1 k k−1 k
k=2
∑︂
∞
R
= (A cos kα + Bk sin kα) .
k−1 k
k=2

Using the equations for the Fourier coefficients,

{︄ }︄ {︄ }︄
Ak w cos kα
1 2π
=π ∆g(α, R) dα,
Bk 0 sin kα

and the cosine difference equation (Wolfram Demonstrations, Difference

formula for cosine), we obtain

1
T (α, R) = π ·
∑︂
∞
R
(︃ w 2π w 2π )︃
′ ′ ′ ′ ′ ′
· cos kα ∆g(α , R) cos kα dα + sin kα ∆g(α , R) sin kα dα =
k−1 0 0
k=2

1 ∑︂ R
∞ w 2π
∆g(α ′ , R) · cos k (α − α ′ ) dα ′ .
(︁ )︁
=π
k−1 0
k=2

í Õ ! ¤. û
196 The Stokes equation and other integral equations

−2π −π π 2π
−1

Figure 8.4. The Stokes kernel function on the circle r = R in two-dimensional

geometry. Note the symmetry and periodicity. Compare with the
^ spherical Stokes kernel, figure 8.3.

Define the Stokes kernel for this two-dimensional situation:

T (α, R) w
R 2π (︁ ′ )︁ (︁ ′
)︁ ′
N(α) = γ = πγ 0 ∆g α , R S α − α dα ,
∑︂
∞
cos k (α − α ′ )
(︁ )︁
′ def
(︁ )︁
with S α − α = .
k−1
k=2

For small values of α − α ′ we may approximate (Wolfram Functions,

∑︁∞ cos kx
k=1 k
):

∑︂
∞
cos (k ′ + 1) (α − α ′ ) ∑︂
∞
cos k ′ (α − α ′ )
(︁ )︁ (︁ )︁
′
S(α − α ) = ≈ =
′
k′ ′
k′
k =1 k =1
(︃ )︃
1 1 )︁ ≈ − ln(α − α ′ ).
= ln
2
(︁
2 1 − cos(α − α ) ′

More abstractly, we may also write relationship 8.3 in terms of the

discrete Fourier transform and its inverse, as
{︂ }︂
k−1 R
F{∆g} = F{T } =⇒ T = F−1 F{∆g} .
R k−1
Here, F{f} represents the Fourier transform of a function f(α) of spatial
co-ordinate α on the circle, as a function of the spatial wave number
(number of waves around the circle) k.

í Õ ! ¤. û
 Example: The Stokes equation in polar co-ordinates
197
^ Tableau 8.1. Stokes equation in two dimensions, octave code.

% Stokes equation simulator in two dimensions

R = 6378137;
g = 9.8;
ak(1:180) = 0;
bk(1:180) = 0;
dg(1:360) = 0.0;
T(1:360) = 0.0;
for i=1:359
% Gauss-Markov
dg(i+1) = 0.8*dg(i) + 50*(rand()-0.5);
end
dgsum = 0.0;
for i=1:360
% Enforce circularity
dg(i) = dg(i) - (dg(360) - dg(1)) * (i/359);
dgsum = dgsum + dg(i);
end
for i = 1:360
% Enforce zero expectancy
dg(i) = dg(i) - dgsum/360;
for k = 2:180
ak(k) = ak(k) + dg(i) * cos(k*i*pi/180)/180;
bk(k) = bk(k) + dg(i) * sin(k*i*pi/180)/180;
end
end
for i=1:360
for k = 2:180
T(i) = T(i) +(ak(k)*cos(k*i*pi/180)+bk(k)*sin(k*i*pi/180))*R/(k-1);
end
end
hold on
plot(1:360, dg, ’b’) plot(1:360, 0.00001*T/g, ’m’)
print -dpdf stokes2D-out.pdf

This formulation has the merit of being able to use any standard
FFT software library offering compatible versions of both the forward

í Õ ! ¤. û
198 The Stokes equation and other integral equations

100

−→ ∆g (mGal)
−→ N (m)
50

−50

−100 −→ α (◦ )
0 90 180 270 360

Figure 8.5. Simulation of gravity anomalies (Gauss–Markov process) and

geoid undulations (blue) in two-dimensional geometry on the
^ circle. Note the spectral behaviour of both.

Fourier transform F{·} and the inverse transform F−1 {·}.

For more about FFT, see appendix C.
In figure 8.5 we show a simulation result in which a randomly
generated set of gravity anomalies on the circle r = R has been used
to calculate geoid undulations on the same circle. Both curves display
fairly realistic statistical behaviour. The code used is given in tableau
8.1.

^ 8.3 Plumb-line deflections and Vening Meinesz

equations
By differentiating the Stokes equation with respect to place, we obtain
luotiviivan integral equations for the components of the deflection of the plumb
poikkeama line (Heiskanen and Moritz, 1967, equation 2-210’):
{︄ }︄ {︄ }︄
ξ x dS(ψ) cos α
1
= ∆g dσ =
η 4πγ σ dψ sin α

í Õ ! ¤. û
 The Poisson integral equation
199
{︄ }︄
x dS(ψ) cos α
1
= ∆g · sin ψ dα dψ, (8.4)
4πγ σ dψ sin α

in which ξ and η are the south-north and west-east direction de-

flections of the plumb line, and the unit-sphere surface element is luotiviivan
dσ = sin ψ dα dψ, in which sin ψ is Jacobi’s determinant of the (ψ, α) poikkeama
co-ordinates.
These equations were derived for the first time by the Dutch geo-
physicist Vening Meinesz. The angle α is the azimuth or direction
angle between the calculation or evaluation point (ϕ, λ) and the moving
integration or observation point (ϕ ′ , λ ′ ). These equations are much
harder to write in spectral form, as the kernel functions are now also
functions of the azimuth direction α; in other words, they are anisotropic.
The disturbing potential, the gravity disturbance, and the gravity
anomaly, are all so-called isotropic quantities: they do not depend on
the azimuth and therefore, in the spectral representation the transfor-
mations between them are functions of harmonic degree n only.

^ 8.4 The Poisson integral equation

Look at figure 8.6. The point Q of the body is located at R, and the
observation point P at r. The geocentric angular distance between the
two location vectors, i. e., the angular distance as seen from the origin,
is ψ. The distance between points P and Q is ℓ.
def def
With the definitions R = ∥R∥ and r = ∥r∥, we may write (cosine
rule):
√︁
ℓ = r2 + R2 − 2rR cos ψ. (8.5)
We may also write the function 1 ℓ as the following expansion (Heiska-
/︁

nen and Moritz, 1967, equation 1-81):

1 ∑︂ R n+1
∞ (︂ )︂
1 1
= √︁ = r Pn (cos ψ), (8.6)
ℓ r2 + R2 − 2Rr cos ψ R
n=0

í Õ ! ¤. û
200 The Stokes equation and other integral equations

z
P

ℓ r

Q ψ
R
O

Figure 8.6. The geometry of the generating function of the Legendre polyno-
^ mials.

in which r and R are the distances of points P and Q from the origin
O, usually the centre of the Earth. Function 8.6 is called the generating
function of the Legendre polynomials.
Differentiating equation 8.6 with respect to r yields

1 ∑︂ n + 1 R n+1
∞
r − R cos ψ
(︂ )︂
− = − r r Pn (cos ψ).
ℓ3 R
n=0

This we multiply by 2r:

1 ∑︂
∞
2r2 − 2rR cos ψ
(︂ )︂n+1
R
− 3
= − (2n + 2) r Pn (cos ψ).
ℓ R
n=0

Now we add together this equation and equation 8.6:

1 ∑︂
∞
−2r2 + 2rR cos ψ + ℓ2
(︂ )︂n+1
R
3
=− (2n + 1) r Pn (cos ψ).
ℓ R
n=0

Substitute ℓ2 from equation 8.5:

−2r2 + 2rR cos ψ + ℓ2 R2 − r2
= ,
ℓ3 ℓ3

í Õ ! ¤. û
The Poisson integral equation
201
and the result is, after multiplying with −R,
∑︂
∞
(︁ )︁
R r2 − R2 (︂ )︂n+1
R
= (2n + 1) r Pn (cos ψ). (8.7)
ℓ3
n=0

Applying degree constituent equation 3.8 to the harmonic potential asteosuus-

field V on the spherical Earth’s surface, radius R: yhtälö
x (︁
2n + 1
V ϕ ′ , λ ′ , R Pn (cos ψ) dσ ′ ,
)︁
Vn (ϕ, λ) =
4π σ

as well as the spectral expansion of the field in space 3.9:

∑︂
∞ (︂ )︂n+1
R
V(ϕ, λ, r) = r Vn (ϕ, λ),
n=0

we obtain

V(ϕ, λ, r) =
1 ∑︂ R n+1
∞ (︂ )︂ x (︁
′ ′
Pn (cos ψ) dσ ′ =
)︁
= r (2n + 1) V ϕ , λ , R
4π σ
n=0
∑︂
[︄ ∞
x (︁
]︄
(︂ )︂n+1
1 R
V ϕ ′, λ ′, R Pn (cos ψ) dσ ′ =
)︁
= (2n + 1) r
4π σ
n=0
x R (︁r2 − R2 )︁ (︁
1
V ϕ ′ , λ ′ , R dσ ′
)︁
= 3
4π σ ℓ
by replacing the expression in square brackets by equation 8.7.
Thus we have obtained the Poisson equation for computing a harmonic
field V from values given on the Earth’s surface:
x R (︁r2 − R2 )︁
1
VP = VQ dσQ , (8.8)
4π σ ℓ3
in which ℓ is again the straight-line distance between evaluation point P
(where VP is being computed) and moving data point Q (on the surface
of the sphere, VQ under the integral sign). In this equation we have
given the points symbolic names: the co-ordinates of evaluation point
P are (ϕ, λ, r), the co-ordinates of data point Q are (ϕ ′ , λ ′ , R).

í Õ ! ¤. û
202 The Stokes equation and other integral equations

Still a third way to write the same equation, useful when the function
or field V is not actually defined between the topographic Earth’s surface
and sea level, is
x R (︁r2 − R2 )︁
1
V= V ∗ dσ.
4π σ ℓ3
Here, V ∗ denotes the value of a harmonically downwards continued function
V — downwards continued into the topography, all the way down to
sea level, or, in spherical approximation, to the surface of the sphere
r = R. This is a function that above the topography is identical to
V, is harmonic, and exists also between the topography and sea level.
The question of the existence of such a function has been a classical
theoretical nut to crack. . .
Equation 8.8 solves for this special case the so-called Dirichlet boundary-
value problem, finding a harmonic function in an area of space when the
value of the function on the boundary of the area has been given.

^ 8.5 Gravity anomalies in the exterior space

The equation derived in section 8.4, equation 8.8, applies for an arbitrary
harmonic field V, i. e., any field for which ∆V = 0. The equation applies
conveniently to the expression r∆g, the gravity anomaly multiplied
by the geocentric radius, which is also a harmonic field. This is how
we can express the gravity anomaly in the external space ∆g(ϕ, λ, r) as
vertauspallo a function of gravity anomalies ∆g(ϕ ′ , λ ′ , R) on a reference sphere of
radius R. The function r∆g is harmonic, because according to equation
5.8
1 ∑︂
∞ (︂ )︂n+1
R
∆g = r (n − 1) r Tn ,
n=2
so
∑︂
∞ (︂ )︂n+1
R ∑︂
∞ (︂ )︂n+1
R
r∆g = r (n − 1) T n = r Tn′ ,
n=2 n=2

pinta- in which Tn′ (ϕ, λ) = (n − 1) Tn (ϕ, λ) is a perfectly legal surface spherical

pallofunktio

í Õ ! ¤. û
Gravity anomalies in the exterior space
203
harmonic just like Tn (ϕ, λ) itself. Also, the dependence on the radius
(︁ /︁ )︁n+1
r, the factor R r , is the same as for the (harmonic) potential. So,
Poisson’s integral equation 8.8 applies to function r∆g:
x R (︁r2 − R2 )︁ [︁
1
R∆g ϕ ′ , λ ′ , R dσ ′
(︁ )︁]︁
[r∆g(ϕ, λ, r)] = 3
4π σ ℓ
or
x (︁ 2 2
)︁
1 RR r −R (︁ ′ ′ )︁ ′
∆g(ϕ, λ, r) = r ∆g ϕ , λ , R dσ . (8.9)
4π σ ℓ3
An alternative notation is
x (︁ 2 2
)︁
1 RR r −R
∆g = r ∆g∗ dσ,
4π σ ℓ3
in which ∆g∗ denotes the gravity anomaly at sea level, again calculated
by harmonic downwards continuation of the exterior field, in this case the
expression r∆g.
From equation 8.9 we may lift the closed form of the kernel:
(︁ 2 2
)︁
RR r −R
K(ℓ, r, R) = r ,
ℓ3
with which
x
1
K(r, ψ, R) ∆g ϕ ′ , λ ′ , R dσ ′ .
(︁ )︁
∆g(ϕ, λ, r) =
4π σ

Using the approximation r + R ≈ 2r still yields

x
1 r − R (︁
R2 3 ∆g ϕ ′ , λ ′ , R dσ ′ .
)︁
∆g(ϕ, λ, r) ≈
2π σ ℓ
Alternatively, we derive the spectral form:

1 ∑︂ R n+1 ∑︂
∞ (︂ )︂ ∞ (︂ )︂n+2
R
∆g = r r (n − 1) Tn (ϕ, λ) = r ∆gn (ϕ, λ).
n=2 n=2

Degree constituent equation 3.8 gives the functions ∆gn : asteosuus-

x yhtälö
2n + 1
∆gn (ϕ, λ) = ∆g(ϕ ′ , λ ′ , R) Pn (cos ψ) dσ ′ ,
4π σ

í Õ ! ¤. û
204 The Stokes equation and other integral equations

with the aid of which

1 ∑︂ R n+2
∞ (︂ )︂ x
∆g = r (2n + 1) ∆g(ϕ ′ , λ ′ , R) Pn (cos ψ) dσ ′ =
4π σ
n=2
x (︃∑︂∞ (︂ )︂n+2 )︃
1 R
= r (2n + 1) Pn (cos ψ) ∆g(ϕ ′ , λ ′ , R) dσ ′ =
4π σ
n=2
x
1
= K ∆g(ϕ ′ , λ ′ , R) dσ ′ ,
4π σ mod
in which
def
∑︂
∞ (︂ )︂n+2
R
Kmod (ψ, r, R) = r (2n + 1) Pn (cos ψ)
n=2

is the modified spectral version of the Poisson kernel for gravity anoma-
lies. From this kernel, the constituents of degree number 0 and 1 have
been removed, see Heiskanen and Moritz (1967, equation 2-159).
Compared to the Stokes kernel, the Poisson kernel drops off fast to
zero for growing values of ℓ. In other words, the evaluation of the
kalotti integral equation may be restricted to a very local area, like a cap of
radius 1◦ . See figure 8.7. The main use of Poisson’s kernel is the harmonic
continuation, upwards or downwards, of gravity anomalies measured
vertaustaso and computed at various levels, shifting them to the same reference
level.
In the limit r → R (sea level becomes the level of evaluation), this
kernel function goes asymptotically to the Dirac δ function.

^ 8.6 The vertical gradient of the gravity anomaly

Differentiate an equation obtained from equations 5.7 and 5.8:
∑︂
∞ (︂ )︂n+2
R ∂∆g 1 ∑︂ R n+3
∞ (︂ )︂
∆g = r ∆gn =⇒ = − r (n + 2) ∆gn .
∂r R
n=2 n=2

This equation is exact in spherical approximation. Its kernel function

is well localised, in other words, it drops off to zero very fast. For
calculation, a small “cap” also suffices here.

í Õ ! ¤. û
 The vertical gradient of the gravity anomaly
205
0.2
Poisson kernel, dimensionless
0.1
1 km
2 km
0

Gravity-gradient kernel, unit km−1

−0.1
1 km
1.25 km
−0.2 1.5 km
2 km
−0.3
Distance (km) −→

0 1 2 3 4 5 6 7

Figure 8.7. The Poisson kernel function for gravity anomalies as well as the
kernel for the anomalous vertical gravity gradient, both at various
height levels. These kernels are used when evaluating the surface
^ integral in map co-ordinates x, y in kilometres.

∆gn is expressed, using degree constituent equation 3.8, as an integral asteosuus-

over the anomaly field at sea level: yhtälö

x
2n + 1
∆g ϕ ′ , λ ′ , R Pn (cos ψ) dσ ′ ,
(︁ )︁
∆gn (ϕ, λ) =
4π σ

so

1 ∑︂ R n+3
∞ (︂ )︂ x
∂∆g(ϕ, λ, r) (︁ ′ ′ )︁
=− r (2n + 1) (n + 2) ∆g ϕ , λ , R Pn (cos ψ) dσ ′ =
∂r 4πR σ
n=2
x
1
K ′ (ψ, r, R) ∆g ϕ ′ , λ ′ , R dσ ′ , (8.10)
(︁ )︁
=
4πR σ
in which the kernel function is now
∑︂
∞ (︂ )︂n+3
R
′
K (ψ, r, R) = − r (2n + 1) (n + 2) Pn (cos ψ).
n=2

í Õ ! ¤. û
206 The Stokes equation and other integral equations

^ Tableau 8.2. Derivation of the kernel for the vertical gradient of gravity
anomaly.

R2 ∂ x 1
(︄ )︄
∂∆g(ϕ, λ, r) r2 − R2 (︁ ′ ′ )︁ ′
= 3
∆g ϕ , λ , R dσ =
∂r 4π ∂r (r + R2 − 2rR cos ψ) /2
σ r 2

R2 x 1
(︄ )︁ )︄
r2 − R2 3 (2r − 2R cos ψ) r2 − R2
(︁
∆g ϕ ′ , λ ′ , R dσ ′ =
(︁ )︁
3
2− 2
− 2
4π σ ℓ r 2rℓ

R2 x 1
(︄ )︄
3 ℓ2 + r2 − R2 r2 − R2
(︁ )︁ (︁ )︁
∆g ϕ ′ , λ ′ , R dσ ′ −
(︁ )︁
= 3
2 − 2 2
4π σ ℓ 2r ℓ
1 1 x R R r2 − R2
(︁ )︁
− · 3
∆g(ϕ ′ , λ ′ , R) dσ ′ =
r 4π σ r ℓ
R2 x 1
(︄ )︁ )︄
3 r2 − R2 3 r2 − R2 r2 − R2
(︁ )︁ (︁ )︁ (︁
1
∆g ϕ ′ , λ ′ , R dσ ′ − ∆g(ϕ, λ, r) =
(︁ )︁
= 3
2− 2
− 2 2
4π σ ℓ 2r 2r ℓ r
2
R2 x 1
(︄ )︄
3 r2 − R2
(︁ )︁ (︃ )︃
(︁ ′ ′ )︁ ′ 1 3
= 2− ∆g ϕ , λ , R dσ − + ∆g(ϕ, λ, r),
4π σ ℓ3 2r2 ℓ2 r 2r
)︁2 ]︄
R2 x 1
[︄
3 r2 − R2
(︁
5
∆g ϕ ′ , λ ′ , R dσ ′ − ∆g(ϕ, λ, r). (8.11)
(︁ )︁
= 3
2− 2 2
4π σ ℓ 2r ℓ 2r

Alternatively, we derive a closed expression. We start from Poisson

equation 8.9 for gravity anomalies, and differentiate1 with respect to r.
See tableau 8.2.
In the result, the last term is small, of an order of less than one part
in a thousand, compared to the preceding term.
The terms inside the square brackets require their own consideration.
In the local zone ℓ ≈ r − R the terms are of the same order of magnitude;
the second term goes however rapidly to zero for ℓ ≫ r − R. This means
that the contribution of the second term to the integral will be only of
2
/︂
order (r − R) R times, or less than a millionth part of, that of the
2

1 Hint: use symbolic algebra software.

í Õ ! ¤. û
The vertical gradient of the gravity anomaly
207
first term. We may thus write
x ∆g(︁ϕ ′ , λ ′ , R)︁
∂∆g(ϕ, λ, r) R2 5
≈ dσ ′ − ∆g(ϕ, λ, r). (8.12)
∂r 2π σ ℓ 3 2r
This equation will only behave well for r > R: for r → R the kernel
function 1 ℓ3 will go to infinity for ψ → 0. Regularisation can be done
/︁

by observing that a globally constant gravity anomaly field

(︂ )︂2
∆g ˜︂ 0 (r) = R ∆g0
˜︂ 0 (ϕ, λ, r) = ∆g
r

has a gradient of

∂∆g
˜︂ 0 (ϕ, λ, r) 2 ˜︂
= − r ∆g 0 (ϕ, λ, r), (8.13)
∂r
but also, like equation 8.12:
x ∆g
∂∆g
˜︂ 0 (ϕ, λ, r) R2 ˜︂ 0 (ϕ, λ, r) 5 ˜︂
≈ dσ ′ − ∆g 0 (ϕ, λ, r). (8.14)
∂r 2π σ ℓ3 2r
Subtract equation 8.14 from equation 8.12 and substitute equation 8.13,
yielding
(︁ )︁
∂∆g(ϕ, λ, r) ∂ ∆g(ϕ, λ, r) − ∆g ˜︂ 0 (ϕ, λ, r) ∂∆g˜︂ 0 (ϕ, λ, r)
= + =
∂r ∂r ∂r
x ∆g(︁ϕ ′ , λ ′ , R)︁ − ∆g˜︂ 0 (ϕ ′ , λ ′ , R)
R2
= dσ ′ −
2π σ ℓ3
5 ˜︂ 0 (ϕ, λ, r) − 2 ∆g
(︂ )︂
− ∆g(ϕ, λ, r) − ∆g r 0 (ϕ, λ, r) =
˜︂
2r
x ∆g(︁ϕ ′ , λ ′ , R)︁ − ∆g
R2 0
= dσ ′ −
2π σ ℓ3
(︂ )︂2
2 R 2
(︃ )︃
5 R
(︂ )︂
− ∆g(ϕ, λ, r) − r ∆g0 − r r ∆g0 .
2r

def
Choose the constant ∆g0 = ∆g(ϕ, λ, r), the anomaly in the evaluation

í Õ ! ¤. û
208 The Stokes equation and other integral equations

point:
x ∆g(︁ϕ ′ , λ ′ , R)︁ − ∆g(ϕ, λ, r)
∂∆g(ϕ, λ, r) R2
= dσ ′ −
∂r 2π σ ℓ 3
(︂ )︂2 )︃
2 R 2
(︃
5 R
(︂ )︂
− 1− r ∆g(ϕ, λ, r) − r r ∆g(ϕ, λ, r) ≈
2r
x ∆g(︁ϕ ′ , λ ′ , R)︁ − ∆g(ϕ, λ, r)
R2 2 R 2
(︂ )︂
′
≈ dσ − r r ∆g(ϕ, λ, r). (8.15)
2π σ ℓ3
This corresponds to Heiskanen and Moritz (1967, equation 2-217). For
well-behaved gravity anomalies, the integrand will now be well-behaved
also when r → R.
If we are integrating over the surface of a spherical Earth of radius R
rather than the unit sphere σ of radius 1, the factor R2 drops out from
equations 8.9, 8.11, 8.12 and 8.15.
In Molodensky’s method this or similar equations can be rapidly
evaluated from very local gravimetric data.
The closed expression given in Heiskanen and Moritz (1967, expres-
sion 2-217), is the anomalous vertical gravity gradient evaluated at
vertauspallo sea level (on the reference sphere). In our equations 8.15 and 8.10
we also need gravity anomalies at sea level. However, anomalies at
the topographic surface level are available. In practice, we may proceed
iteratively, by initially assuming that the anomaly values observed at
topography level are at sea level:

∆g(0) ϕ ′ , λ ′ , R ≈ ∆g ϕ ′ , λ ′ , r = ∆g ϕ ′ , λ ′ , R + H ,
(︁ )︁ (︁ )︁ (︁ )︁

in which H = H(ϕ ′ , λ ′ ) is the topographic height at point (ϕ ′ , λ ′ ). When

a crude anomalous gradient has been calculated, for example using
equation 8.15, we may perform a real reduction to sea level, in linear
approximation:
⃓
(1)
(︁ ′ ′ )︁ (︁ ′ ′ )︁ ∂∆g(0) ⃓
∆g ϕ , λ , R ≈ ∆g ϕ , λ , r − ⃓ H.
∂z ⃓z=r

This may be iterated.

í Õ ! ¤. û
 Gravity reductions in geoid determination
209
^ 8.7 Gravity reductions in geoid determination
^ 8.7.1 Classical methods
Use of the Stokes equation for gravimetric geoid determination presup-
poses that all masses are inside the geoid — and that the exterior field
is thus harmonic. For this reason we move the topographic masses
computationally to inside the geoid, in a way that needs to be specified.
The classical methods for this are
◦ Helmert’s (second) condensation method, section 6.5: the masses
are shifted vertically down to the geoid into a surface density layer.
After this, shifting gravity down from the topographic surface to
sea level is easy. The indirect effect (the effect of the mass shifts epäsuora
on the geoid, the “restore” step) is small. vaikutus

◦ Isostatic reduction, in which the effects of both the topography

and its compensation, the “roots” of mountains below sea level,
are computationally removed. The indirect effect of this method
is larger. See section 6.7 and equation 6.8.
◦ Bouguer reduction, section 6.2: the effect of the topographic
masses is brutally removed from the observed gravity data, and,
after geoid calculation, it is equally brutally restored to the result.
Bouguer anomalies contain large negative biases in the mountains
and therefore, the indirect effect of Bouguer reduction is excessive
and extends over a large area. This is why Bouguer reduction is
used more rarely.

^ 8.7.2 The residual terrain modelling method

Imagine that, conceptually, the topographic masses are shifted to below
the geoid in a way that does not change the exterior field. This is materially
the same as determining the geoid associated with the harmonically
downwards continued exterior field.
The problem here is that such a mass distribution below sea level

í Õ ! ¤. û
210 The Stokes equation and other integral equations

which produces the harmonically downwards continued external po-

tential in the space between topographic surface and geoid does not
always precisely exist. Or that a suitable mass distribution exists but
contains extremely large positive and negative masses close to each
other, which is physically unrealistic.
huonosti One expresses this by saying that the problem is “ill-posed”. In such
asetettu cases, one uses regularisation: one changes the exterior field a little
— as little as possible, so that it becomes a sensible field that can be
harmonically continued below the topographic surface. Then, some
sensible mass distribution interior to the geoid producing this field will
also exist.
One can start, for example, by filtering out the short-wave parts
caused by the topography using a high-resolution digital terrain model.
This is called the RTM (residual terrain modelling) method.
In this method, we do not actually move all topographic masses to
puskutraktori below the geoid. Instead, we use a bulldozer technique, figure 8.8: only
masses close to the topographic surface are either removed or filled in,
in a way that creates a smooth replacement topography that is long-
wavelength only. The exterior field of this smoothed topography, unlike
that of the original topography, lacks the shortest wavelengths. It may
thus be downwards continued to the geoid with sufficient precision.
First, we computationally remove from the topography only the short
wavelengths (under 30 km) by moving the masses of the peaks into the
ilma-anomalia valleys: a low-pass filtering. The effect of this on the free-air gravity
anomalies ∆g calculated from measurements is evaluated and taken
into account: the “remove” step.
In detail:
1. At each point P we apply the terrain correction to the gravity
anomalies as described in section 6.3.
2. Next, we remove the attraction of a Bouguer plate of thickness
H − HRTM , in which H stands for the terrain height of point P, and

í Õ ! ¤. û
Gravity reductions in geoid determination
211
HRTM for the height of the smoothed, or low-pass filtered, terrain
at the location of point P. This effect is, according to equation 6.1,
equal to
2πGρ (H − HRTM ) ,
in which ρ is the rock density assumed in the calculation.
3. After this, the location of the gravity anomaly is moved down (or
up!) — “downwards continuation” — from the original terrain
level H to the surface of the new, smoothed terrain, HRTM . Equation
8.15 for the vertical gradient of the free-air gravity anomaly may ilma-anomalia
be used for this.
If this anomalous vertical gradient is small, meaning that the
vertical gravity gradient of the terrain-reduced external field
equals the vertical gradient of normal gravity — according to
section 5.4, −0.3 mGal/m — this operation will leave the anomaly
unchanged. Typically, there will be a change: one may show
— exercise 1 item 4 — that on the surface of a buried sphere of
anomalous density ∆ρ, there will be a radial anomalous gravity
gradient of 83 πG∆ρ. For ∆ρ ≪ ρ, this will be negligible compared
to the Bouguer-plate coefficient in item 2.
4. Rigorously speaking, an inverse terrain correction for the shapes
of the smoothed terrain should be applied, to arrive at gravity
anomalies realistic for this new replacement topography. Often
this step is left out as the effect is small.
5. After that, harmonic downwards continuation of the exterior field
succeeds: almost only long wavelengths are left in the exterior
field.
Because the mass shifts in the RTM method are so small, take place
over such small distances, and are of such a short wavelength in nature,
the indirect effect or “restore” step — the change in geopotential due to epäsuora
the mass shifts that has to be applied in reverse to arrive at the final vaikutus
geopotential or geoid solution — is so small as to often be negligible.

í Õ ! ¤. û
212 The Stokes equation and other integral equations

P −
− − P′ +
+ +
−
Bouguer plate, down- Inverse
Terrain correction wards continuation terrain correction

Figure 8.8. Residual terrain modelling (RTM). One removes the short wave-
lengths, i. e., the differences from the red dashed line, from the
terrain computationally: the masses rising above it are removed,
the valleys below it are filled. After reduction, the red line,
smoother than the original terrain, is the new terrain surface. The
exterior potential of the new mass distribution will differ only
little from the original one, but may be harmonically downwards
continued to sea level.
Left, terrain correction for point P, middle, Bouguer-plate and
gradient reduction to the level of smoothed terrain point P ′ , and
^ right, the inverse terrain correction for point P ′ .

For the same reason, the effect of unknown topographic density will
also remain small.
Finally we note that, because the RTM method removes the effect
of the short-wavelength topography, it is also a suitable method for
interpolating gravity anomalies. See Märdla (2017).

^ 8.7.3 Downwards continuation in linear approximation

The approach described above can, following Molodensky, be linearised:
x (︃ (︁ ⃓
)︁ ∂∆g ⃓
)︃
R ′ ′ ′
T (ϕ, λ, H) = ∆g ϕ , λ , H − ⃓ H S(ψ) dσ ′ +
′
4π σ ∂z ⃓z=H ′

í Õ ! ¤. û
Gravity reductions in geoid determination
213
∂T ⃓
⃓
+ H. (8.16)
∂z z=H
⃓

So, first we reduce the ∆g measured and calculated at the topographic

surface to sea level using the gradient of the anomalies and the terrain
height H ′ of the measurement point, with the result
⃓
∗ ∂∆g ⃓⃓
∆g = ∆g − H ′.
∂z ⃓z=H ′

After this, we apply, at sea level, the Stokes equation, and obtain the
disturbing potential at sea level T ∗ . After this, the disturbing potential
is “unreduced” back to terrain level, to the evaluation point, with the
equation
∂T ⃓
⃓
T = T∗ + H.
∂z z=H
⃓
∂
In these equations T , its vertical derivative ∂H T , ∆g, and its vertical
∂
derivative ∂H ∆g always belong to the exterior harmonic gravity field.
The connection between them is the fundamental equation of physical fysikaalisen
geodesy, equation 5.5, in spherical geometry geodesian
perusyhtälö
∂T 2
∆g = − − rT,
∂r
in which r = R + H. Here, we need firstly the vertical derivative of the
disturbing potential. This is easy: we have

∂T ∂T 2
= = −∆g − r T ,
∂H ∂r
where the first term on the right is directly measured, and the second
term’s T is obtained iteratively from the main product of the solution
process.
Calculating the vertical gradient of gravity anomalies, i. e., the anoma-
lous vertical gradient of gravity, is harder. For this task, section 8.6 offers
calculation options. Luckily for practical calculations, the kernels of
the integral equations are very localised and one does not need gravity
anomalies from a very large area.

í Õ ! ¤. û
214 The Stokes equation and other integral equations

^ 8.7.4 The evaluation point as the reference level

vertaustaso In the above equation 8.16 we used as the reference level the sea surface.
This is arbitrary: we may use whatever reference level, for example H0 ,
in which case

T (ϕ, λ, H) =
x (︃ (︁ ⃓
)︁ ∂∆g ⃓
)︃
R + H0 ′ ′ ′
= ∆g ϕ , λ , H − ⃓ (H − H0 ) S(ψ) dσ ′ +
′
4π σ ∂z ⃓z=H ′
∂T ⃓
⃓
+ (H − H0 ) .
∂z z=H
⃓

If we now choose H0 = H, the last term drops off, and we obtain

T (ϕ, λ, H) =
x (︃ (︁ ⃓ )︃
R+H ∂∆g
∆g ϕ ′ , λ ′ , H ′ − (H − H) S(ψ) dσ ′ .
′
)︁ ⃓
= ⃓
4π σ ∂z ⃓ z=H ′

In this case, the reduction takes place from the height of the ∆g mea-
surement point to the height of the T evaluation point. This is likely to
be a shorter distance than from sea level to evaluation height, especially
in the immediate surroundings of the evaluation point. This means
that the linearisation error will remain smaller.2 What is bad, on the other
hand, is that the expression in parentheses is now different for each
evaluation point. This complicates the use of FFT-based computation
techniques, on which more later.
Here, we were all the time discussing the determination of the
disturbing potential T (ϕ, λ, H); this is in practice the same as determining
the height anomaly

T (ϕ, λ, H) T (ϕ, λ, H)
ζ(ϕ, λ, H) = ≈ (︁ 1
γHh
)︁,
γ ϕ, 2 (H + h)

2 Thelinearisation error could be even further tuned down by choosing as the

evaluation level for the vertical gradient z = 12 (H ′ + H).

í Õ ! ¤. û
 The remove–restore method
215
equation 7.1. Here, γ is normal gravity calculated for point latitude3 ϕ
and topographic height 12 (H + h) = H + 12 ζ = h − 12 ζ.

^ 8.8 The remove–restore method

All current geoid determination methods are in one way or another
“remove–restore” methods, even in several different ways.
1. From the observed gravity values, first the effect of a global gravity
field model is removed. This model is generally given in the form
of a spherical-harmonic expansion. Thus, a residual gravity field is pallofunktio-
obtained kehitelmä

◦ that has numerically smaller values which are easier to work

with
◦ that is more local: the long “wavelengths”, the patterns extend-
ing over large areas, have been removed from the residual
field, only the local details remain.
2. From the observed gravity, the effects are removed of all masses
that are outside the geoid — in practice, the topography. The
purpose of this is to obtain a residual gravity field
◦ to which the Stokes equation may be applied, because no
masses are left outside the boundary surface
◦ from which especially the very small “wavelengths” — details
the size of which is of the order of a few kilometres — caused
by the topography, are gone. After this, prediction of gravity
values from sparse measurement values will work better.
Some gravity reduction methods — methods which computationally
remove the gravity effect of the exterior masses — with good predic-
tion properties were already presented in subsection 8.7.1: Bouguer

3 Inan actual calculation one would calculate γHh using the true geodetic latitude φ
and equation 4.7. The height 12 (H + h) has to be correct within a few metres in order
to attain millimetre precision in ζ.

í Õ ! ¤. û
216 The Stokes equation and other integral equations

“Remove” “Restore”
Brute force
∆g −−−−−−−−−−−−−−−−−−→ N
⏐ ↑
⏐ Global gravity ⏐ Global gravity
−↓ +⏐
field model field model
∆gloc Nloc
⏐ ↑
⏐ Exterior masses ⏐ Exterior masses
−↓ (topography) +⏐ (topography)
Stokes
∆gred −−−−−−−−−−−−−−−−−−→ Nred

^ Figure 8.9. The remove–restore method as a commutative diagram.

reduction and isostatic reduction. Also Helmert condensation may be

mentioned, although its prediction properties are poor.
kommutoiva We may illustrate the remove–restore method by means of commu-
kaavio tative diagram 8.9. In this diagram, the black arrows with text denote
calculations that are recommended, because they are easy and accurate.
The grey arrow with text refers to direct computation, which again is
troublesome and computing intensive.

^ 8.9 Kernel modification

In the remove–restore method described above, the handling of reduced
gravity anomalies ∆gred and geoid heights Nred happens ordinarily
within a small area. For example, when using the FFT method, the area
of computation is often a rectangular area in the map projection plane,
drawn generously around the country or area for which a geoid model
is being computed.
Furthermore, if we compute a geoid model directly by integrating the
Stokes equation, we will evaluate this integral, after removing the effect
of the global model from the given gravity data, only over a limited
kalotti area or cap: evaluate the equation
x
R
Nred = S(ψ)∆gred (ϕ ′ , λ ′ ) dσ ′ , (8.17)
4πγ σ0

í Õ ! ¤. û
Kernel modification
217
in which σ0 is a cap on the unit sphere the radius of which is, say, ψ0 . kalotti
The (possibly dangerous) assumption behind this is that, outside the
cap, ∆gred is both small and rapidly varying, because the longer wave-
lengths have been removed from it with the global-model reduction.
Write, in the above equation 8.17,

∑︂
∞
2n + 1
S(ψ) = Pn (cos ψ)
n−1
n=2

and
∑︂
∞
′ ′
∆gred (ϕ , λ ) = ∆gn (ϕ ′ , λ ′ ),
n=L+1

assuming that L is the largest degree number that is still along in

the global spherical-harmonic expansion, or gravity model, that was pallofunktio-
subtracted from the data — and that the model is accurate up to that kehitelmä
degree number.
Now, because ∆gn is a certain linear combination of the surface pinta-
spherical harmonics pallofunktio
⎧
⎨P (cos ψ) cos mα if m = 0, . . . , n,
nm
Ynm (ψ, α) =
⎩Pn|m| (cos ψ) sin |m| α if m = −n, . . . , −1,

for example like this:

1 ∑︂
n
∆gn (ψ, α) = ∆gnm Ynm (ψ, α),
Rn+1
m=−n

and also

Pn (cos ψ) = Pn0 (cos ψ) cos(0 · α) = Yn0 (ψ, α),

it follows from the orthogonality of the Y functions that

x
Pn (cos ψ) ∆gn ′ (ϕ ′ , λ ′ ) dσ ′ = 0 if n ̸= n ′ .
σ

í Õ ! ¤. û
218 The Stokes equation and other integral equations

Now we may write — the terms n ⩽ L drop away:

x
S(ψ) ∆gred (ϕ ′ , λ ′ ) dσ ′ =
σ
∑︂ 2n + 1 ∑︂
(︄ ∞ )︄ (︄ ∞
x
)︄
= P (cos ψ) ∆gn (ϕ ′ , λ ′ ) dσ ′ =
σ n−1 n
n=2 n=L+1
∑︂ 2n + 1 ∑︂
(︄ ∞ )︄ (︄ ∞
x
)︄
= P (cos ψ) ∆gn (ϕ ′ , λ ′ ) dσ ′ =
σ n−1 n
n=L+1 n=L+1
x
= SL (ψ) ∆gred (ϕ ′ , λ ′ ) dσ ′ ,
σ

in which
∑︂
∞
2n + 1
L
S (ψ) = P (cos ψ)
n−1 n
n=L+1

is a so-called modified Stokes kernel function. The harmonic degree number

L is called the modification degree. The size of the evaluation area σ0 is
chosen to be compatible with this.
The modification method described here, restricting the Legendre
expansion of the S function to higher degree numbers, is called the
Wong–Gore4 modification (Wong and Gore, 1969). A desirable property
of the new kernel function SL is that it would be — at least compared
kalotti to the original function S — small outside the cap area σ0 . In that
case, restricting the integral to the cap instead of the whole unit sphere
(equation 8.17) does not do much damage. It is clear that SL is much
narrower than S, as only the higher harmonic degrees are represented
in it. This can be verified by plotting a graph of both curves (figure
8.10). The graph does not however go totally to zero outside the cap
but oscillates somewhat.
The reason for the oscillation is that in the frequency, i. e., degree-
number, domain the modified kernel’s cut-off is quite sharp. Trans-
forming such a sharp edge between the space and frequency domains

4 L.Wong and R. C. Gore worked at the Aerospace Corporation, a space technology

research institution in California. Wikipedia, The Aerospace Corporation.

í Õ ! ¤. û
 Kernel modification
219
25
S(ψ)
20 S2 (ψ)
S3 (ψ)
15 S4 (ψ)
S5 (ψ)
10 S6 (ψ)
S2−5 (ψ)
5
S(ψ)
S4 (ψ) S2 (ψ) S(ψ)
0
S(ψ)
S6 (ψ) Angular distance ψ (rad) −→
−5
0 0.5 1 1.5 2 2.5 3 3.5

Figure 8.10. Modified Stokes kernel functions. Note how the kernel values
for higher modification degree numbers L approach zero outside
the local area. The red curve has been ”soft modified” over a
^ modification degree range of 2–5.

will invariably produce an oscillation, which is related to the so-called

Gibbs5 phenomenon. Gibbsin ilmiö
In figure 8.10 we have drawn in red a Stokes kernel that was modified
or tapered ”softly”, by instead of removing the lower-degree terms
altogether, forcing them gradually to zero going from degree number 5
down to 2. The curve is seen to go even better to zero than the ”sharply”
modified kernels.

5 Josiah
Willard Gibbs (1839–1903) was an American physicist, chemist, thermody-
namicist, mathematician and engineer.

í Õ ! ¤. û
220 The Stokes equation and other integral equations

^ 8.10 Advanced kernel modifications

Other kernel modification methods are found in the literature. Their
general form is

∑︂
∞
2n + 1 ∑︂
L
2n + 1
L
S (ψ) = Pn (cos ψ) + (1 − sn ) P (cos ψ) =
n−1 n−1 n
n=L+1 n=2

∑︂
L
2n + 1
= S(ψ) − sn P (cos ψ), (8.18)
n−1 n
n=2

in which the modification coefficients sn , n = 2, . . . , L can be chosen.6

They are chosen so as to minimise the values of the kernel SL in the area
kalotti outside the cap, σ − σ0 . In this way one may eliminate the truncation
error of equation 8.17 and the oscillation of the Wong–Gore modification
almost entirely. Molodensky et al. (1962) had already developed such a
method earlier. See also Bucha et al. (2019).
In the above equation 8.18 we want to minimise the function
∑︂
L
2n + 1
L
S (ψ) = S(ψ) − sn P (cos ψ)
n−1 n
n=2

over the area outside a local cap, σ − σ0 . Let us multiply this expression
with each of the Legendre polynomials Pn (cos ψ), n = 2, . . . , L in turn,
and integrate over the area σ − σ0 outside the local cap:
w
S(ψ)Pn (cos ψ) dσ −
σ−σ0
∑︂
L
2n ′ + 1
w
− s
n′ Pn ′ (cos ψ)Pn (cos ψ) dσ = 0, n = 2, . . . , L,
n′ − 1 σ−σ0
n ′ =2

a system of L − 1 equations in the L − 1 unknowns sn ′ :

∑︂
L
2n ′ + 1
e ′ s ′ = Qn ,
′
n ′ − 1 nn n
n =2

6 The
choice sn = 1 again gives the simply (Wong–Gore) modified Stokes kernel from
which the low degrees have been completely removed.

í Õ ! ¤. û
Advanced kernel modifications
221
with
w wπ
1
Qn = S(ψ)Pn (cos ψ) dσ = S(ψ)Pn (cos ψ) sin ψ dψ,
2π σ−σ0 ψ0

and similarly
w
1
enn ′ = Pn (cos ψ)Pn ′ (cos ψ) dσ =
2π σ−σ0
wπ
= Pn (cos ψ)Pn ′ (cos ψ) sin ψ dψ.
ψ0

The coefficients Qn are known as Molodensky’s truncation coefficients,

enn ′ as Paul’s (1973) coefficients.
From this, we can solve the sn for every degree number n from 2 to L.
This solution also sets to zero the expressions
w
SL (ψ)Pn (cos ψ) dσ, (8.19)
σ−σ0

for all values n from 2 to L.

Expressions 8.19 can be understood as inner or scalar products, between
functions SL and Pn . Similarly, the elements of enn ′ contain the scalar
products between functions Pn and Pn ′ . These scalar products do not
vanish: when integrating over σ − σ0 , unlike over the whole sphere σ,
the Legendre polynomials are not mutually orthogonal. Therefore, e is
a full matrix, not a diagonal matrix like when integrating over the full
unit sphere σ.
The Legendre polynomials are, however, independent of each other
on the domain σ − σ0 , and together span an L − 1 -dimensional linear
vector space.
Now, outside the cap σ0 of radius ψ0 , the Stokes kernel S(ψ), by visual kalotti
inspection, is “smooth”. Depending of course on the values of cap radius
ψ0 and modification degree L, it may be so smooth that it does not
contain any significant contribution from degree numbers higher than
L. If this applies for S, it will also apply for SL . This means that SL will
be a linear combination of the Legendre polynomials, i. e., an element

í Õ ! ¤. û
222 The Stokes equation and other integral equations

of the vector space spanned by the polynomials Pn , n = 2, . . . , L. But

kantavektori if this is so, and the scalar products 8.19 with each of the basis vectors
vanish, then SL must be the zero function on σ − σ0 .
See also Featherstone (2003).
Appendix A.1 explains more about linear vector spaces and the scalar
product of vectors.

^ 8.11 Block integration

In numerical gravimetric geoid determination one uses averages of
anomalies computed over standard-sized cells or blocks, generally 5 ′ ×5 ′ ,
10 ′ × 10 ′ , 30 ′ × 30 ′ etcetera. At European latitudes, often sizes like
3 ′ × 5 ′ , 5 ′ × 10 ′ , 6 ′ × 10 ′ are used, which are approximately square.
The following equation applies when evaluating an integral using
block averages:

R ∑︂
N(ϕ, λ) = Si (ϕ, λ)∆gi , (8.20)
4πγ
i

in which ∆gi is the mean of block i

x x
1 1
∆gi = ∆g(ϕ, λ) dσ = ∆g(ϕ, λ) cos ϕ dϕ dλ,
ω(σi ) σi ω(σi ) σi

and the block value of the Stokes kernel similarly

x
1
S ψ(ϕ, λ; ϕ ′ , λ ′ ) cos ϕ ′ dϕ ′ dλ ′ ,
(︁ )︁
Si (ϕ, λ) =
ω(σi ) σi

in which σi is the area of block i and its size on the unit sphere is
x x
ω(σi ) = dσ = cos ϕ dϕ dλ.
σi σi

Numerical evaluation of such an integral, or quadrature, is done conve-

í Õ ! ¤. û
 Block integration
223
4 k=
1 36 1
36 36
1

16
4 36 4
36 36 0

1 1 -1
36 4 36
36
j = -1 0 1

^ Figure 8.11. Simpson integration nodal weights in two dimensions.

niently using Simpson’s rule:7

Si (ϕ, λ) =
w λi + ∆λ/2 w ϕi + ∆ϕ/2 (︁
1 ′ ′
cos ϕ ′ dϕ ′ dλ ′ ≈
)︁
= S ψ(ϕ, λ; ϕ , λ )
ω(σi ) λi − ∆λ/2 ϕi − ∆ϕ/2

∆ϕ ∆λ ∑︂ ∑︂
1 1
≈ wj wk Si,jk ,
ω(σi )
j=−1 k=−1

in which ∆λ and ∆φ are the block sizes in the latitude and longitude
directions, and w−1 = w1 = 16 , w0 = 46 are the weights.
(︂ (︁ )︁)︂
Si,jk (ϕ, λ) = S ψ ϕ, λ; ϕi + 21 j∆ϕ, λi + 12 k∆λ cos ϕi + 12 j∆ϕ ,
(︁ )︁

j, k = −1, 0, 1

are the values of expression S ψ(ϕ, λ; ϕ ′ , λ ′ ) cos ϕ ′ at the nodal points

(︁ )︁

used in the evaluation, 3×3 of them. See figure 8.11. More complicated
formulas (repeated Simpson or Romberg) can also be employed.

7 Thomas Simpson FRS (1710–1761) was an English mathematician and textbook writer.

Actually Simpson’s rule was already being used a century earlier by Johannes Kepler.

í Õ ! ¤. û
224 The Stokes equation and other integral equations

^ 8.12 Effect of the local zone

One can show that the effect of the local (inner) zone on the geoid
at the evaluation point (ϕ, λ) is proportional to the gravity anomaly
in the point itself, ∆g(ϕ, λ). Starting from Stokes equation 8.1 with
S(ψ) ≈ 1 sin ψ2 ≈ 2 ψ , we find, for a circular inner zone of radius ψ0 :
/︁ /︁

w 2πw ψ0
R 2
δN0 = ∆g(ψ, α) sin ψ dψ dα ≈
4πγ 0 0 ψ
R s
≈ · 2π · ∆g0 · 2ψ0 = γ0 ∆g0 .
4πγ

kalotti Here s0 = Rψ0 is the radius of the local block or cap in units of length.
The quantity
w s0 (︃ w 2π )︃
1
def 1
∆g0 = s ∆g dα ds
0 0 2π 0

is a special average of the gravity anomaly, the average of “ring averages”

for radii s between zero and s0 . If s0 is small, one may take for this the
anomaly value ∆g(ϕ, λ) at the centre without incurring much error.
luotiviivan The local contributions to the deflections of the plumb line are again
poikkeama proportional to the horizontal gradients of gravity anomalies. We start
from Vening Meinesz equations 8.4, with the above approximations for
a local cap:
{︄ }︄ {︄ }︄
δξ0 w ψw
0 2π
(︃ )︃
cos α
1 2
≈ − 2 ∆g · sin ψ dα dψ.
δη0 4πγ 0 0 ψ sin α

We expand ∆g into local linear rectangular co-ordinates x, y:

∂∆g ∂∆g
∆g ≈ ∆g0 + x +y ≈
∂x ∂y
(︃ )︃
∂∆g ∂∆g
≈ ∆g0 + R sin ψ cos α + sin α ,
∂x ∂y

and substitute:

í Õ ! ¤. û
 Self-test questions
225
{︄ }︄
δξ0 1
≈ ·
δη0 4πγ
w ψw
(︄ (︃ )︃ )︄ {︄ }︄
0 2π 2 ∂∆g ∂∆g cos α
· − 2 ∆g0 + R sin ψ cos α + sin α sin ψ dα dψ.
0 0 ψ ∂x ∂y sin α
r 2π
Here, the terms in ∆g0 drop out in α integration (because 0 sin α dα =
r 2π
0 cos α dα = 0). So do the mixed terms in sin α cos α. What remains
is
w ψw0 2π 2
1 ∂∆g
δξ0 ≈ − R sin ψ cos α cos α · sin ψ dα dψ ≈
4πγ 0 0 ψ2 ∂x
w ψw
0 2π ∂∆g
R Rψ ∂∆g
≈− cos2 α · dα dψ ≈ − 0 ,
2πγ 0 0 ∂x 2γ ∂x
w ψw0 2π 2
1 ∂∆g
δη0 ≈ − 2
R sin ψ sin α sin α · sin ψ dα dψ ≈
4πγ 0 0 ψ ∂y
w ψw0 2π ∂∆g
R Rψ ∂∆g
≈− sin2 α · dα dψ ≈ − 0 .
2πγ 0 0 ∂y 2γ ∂y
Evaluating these integrals assumed the partial derivatives to be constant
within the cap. Using Rψ0 = s0 yields now
s0 ∂∆g s0 ∂∆g
δξ0 ≈ − , δη0 ≈ − .
2γ ∂x 2γ ∂y
These equations might be useful as standard block integration, equation
8.20, behaves numerically poorly in the immediate surroundings of the
evaluation point if the kernel function is singular in the origin ψ = 0.
Both the Stokes 8.1 and Vening Meinesz 8.4 kernels are of this kind.

^ Self-test questions
1. What do the Stokes equation and its spectral form look like?
(︁ )︁
2. What does the Stokes kernel function S ψ look like when ex-
panded in Legendre polynomials?
3. What is a suitable approximation of the Stokes kernel when ψ is
small?

í Õ ! ¤. û
226 The Stokes equation and other integral equations

4. What is an isotropic, what an anisotropic quantity on the Earth’s

surface? Give an example of the latter.
5. What does the Poisson integral equation describe?
6. Why are gravity reductions necessary when using the Stokes
equation for computing a geoid model?
7. Which gravity reduction methods are available?
8. Explain the residual terrain modelling (RTM) method.
9. Explain the remove–restore method.
10. Why, in geoid determination, is the Stokes kernel function often
modified? What does such a modification look like?
Gibbsin ilmiö 11. What is the Gibbs phenomenon?

^ Exercise 8 – 1: The Stokes equation

1. Derive a simpler form of the Stokes function S(ψ) which is valid
when the angular distance ψ is small. This simpler form really
consists of only one term!
2. Using this form, write the integral equation
x
R
N= S(ψ) ∆g dσ
4πγ σ
into polar co-ordinates, as an integral of the form
w 2πw ∞
· · · ds dα,
0 0

in which s = ψR is the linear distance from the evaluation point,

and α the azimuth angle (direction angle) from the evaluation
point for the geoid height N to the moving integration point for
the gravity anomaly ∆g.
Hint: you need to consider Jacobi’s determinant for the polar co-
ordinates (s, α).
3. Compute N (as a formula) if ∆g = ∆g0 only within a circular area
s ⩽ s0 , and outside it ∆g = 0. Assume that s0 is small.

í Õ ! ¤. û
^ Spectral techniques, FFT

9
^ 9.1 The Stokes theorem as a convolution
We start from the Stokes equation
x
R
S(ψ)∆g ϕ ′ , λ ′ dσ ′ ,
(︁ )︁
T (ϕ, λ) =
4π σ
in which (ϕ ′ , λ ′ ) is the location of the moving integration or observation
point, and (ϕ, λ) is the location of the evaluation point, both at sea
level, that is on the surface of a spherical Earth. So, the locations of
both points are given in spherical co-ordinates (ϕ, λ). The integration
is carried out over the surface of the unit sphere σ: a surface element is
dσ = cos ϕ dϕ dλ, in which cos ϕ represents the determinant of Jacobi,
for the spherical co-ordinates (ϕ, λ).
However locally, in a sufficiently small area, one may also write
the point co-ordinates in rectangular form and express the integral
in rectangular co-ordinates. Suitable rectangular co-ordinates are, for
example, map projection co-ordinates, see figure 9.1.
A simple example of rectangular co-ordinates in the tangent plane
would be

x = ψR sin α, y = ψR cos α, (9.1)

in which α is the azimuth of the line connecting evaluation point and

moving data point. The centre of this projection is the point where the

– 227 –
228 Spectral techniques, FFT

y
Data point
α
x
Evaluation point

R ψ

Earth’s centre

^ Figure 9.1. Map projection co-ordinates x, y in the local tangent plane.

tangent plane touches the sphere. The locations of other points are
measured by the angle ψ at the Earth’s centre, i. e., the geocentric angular
distance, and by the direction angle in the tangent plane or azimuth α.
A more realistic example uses a popular conformal map projection
called the stereographic projection:

ψ ψ
(︂ )︂ (︂ )︂
x = 2 tan R sin α, y = 2 tan R cos α.
2 2
In the limit for small values of ψ this agrees with equations 9.1.
Taking the squares of equations 9.1, summing them, and dividing
the result by R2 yields
x2 + y2
ψ2 ≈ .
R2
More generally ψ is the angular distance between the points (x, y)
(evaluation point) and (x ′ , y ′ ) (data, integration or moving point) seen
from the Earth’s centre, approximately
(︃ )︃2 (︃ )︃2
2 x − x′ y − y′
ψ ≈ + .
R R

í Õ ! ¤. û
The Stokes theorem as a convolution
229
Furthermore, we must account for Jacobi’s determinant R2 of the projec-
tion:
dσ = R−2 dx dy ⇐⇒ dx dy = R2 dσ,
and the Stokes equation now becomes
x∞
1
S x − x ′ , y − y ′ ∆g x ′ , y ′ dx ′ dy ′ ,
(︁ )︁ (︁ )︁
T (x, y) ≈ (9.2)
4πR −∞

a two-dimensional convolution.1
Convolutions have nice properties in Fourier theory. If we designate
the Fourier transform with the symbol F, and convolution with the
symbol ⊛, we may abbreviate the above equation as follows:

1
T= S ⊛ ∆g,
4πR
and according to the convolution theorem (“Fourier transforms a convolu-
tion into a multiplication”):

1
F{T } = F{S} · F{∆g}.
4πR
This approximation in the (x, y) plane works only if integration can be
restricted to a local area, where the curvature of the Earth’s surface may be
neglected. This is possible thanks to the use of global spherical-harmonic pallofunktio-
expansions, because these represent the long-wavelength part of the kehitelmä
spatial variability of the Earth’s gravity field. After we have removed
the effect of the global spherical-harmonic model from the observed
gravity anomalies ∆g (the “remove” step) we may safely forget the effect
of areas far removed from the evaluation point: after this removal, the
anomaly field ∆gloc will contain only the remaining short-wavelength
parts, the effect of which cancels out at greater distances.

1 Theintegration extends from minus to plus infinity in both co-ordinates x and y.

This can only be realistic on a curved Earth if it is assumed that the kernel S is of
bounded support: it differs from zero only in a bounded area. This is the case for the
modified kernels discussed in section 8.9.

í Õ ! ¤. û
230 Spectral techniques, FFT

Of course, once the integral has been computed and the local disturb-
ing potential Tloc , and the corresponding geoid undulation Nloc , have
been obtained, we must remember to add to it again the effect of the
pallofunktio- global spherical-harmonic expansion on the disturbing potential T and
kehitelmä geoid undulation N to be calculated separately. This is the “restore”
kommutoiva step of the computation; see the commutative diagram 8.9.
kaavio

^ 9.2 Integration by FFT

The Fourier transform needed for applying the convolution theorem
is calculated as a discrete Fourier transform. The highly efficient Fast
Fourier Transform, FFT, exists for this purpose, for example Vermeer
(1993). There are several slightly differing formulations of the discrete
Fourier transform to be found in the literature. It does not really matter
which is chosen, as long as it is a compatible pair of a forward Fourier
transform F and a reverse Fourier transform F−1 .
hilaesitys In preparation for this, we first compute a discrete grid representation of
the function ∆g(x, y), a rectangular table of ∆g values on an equidistant
(x, y) grid of points. The values may be, say, the function values
themselves at the grid points:2
(︁ )︁
∆gij = ∆g xi , yj ,

in which the co-ordinates of the grid points are

xi = i δx, yj = j δy, i, j = 0, 1, . . . , N − 1,

for suitably chosen grid spacings (δx, δy) . The integer N is the grid size,
assumed for simplicity to be the same in both directions.
Next, we do the same for the kernel function

S(ψ) = S x − x ′ , y − y ′ = S(∆x, ∆y),

(︁ )︁

2 Thereexist alternatives to this. For example, one could calculate for every grid point
the average over a square cell surrounding the point.

í Õ ! ¤. û
Integration by FFT
231
so we write
(︁ )︁
Sij = S ∆xi , ∆yj ,
where again

∆xi = i δx, ∆yj = j δy, i, j = 0, 1, . . . , N − 1.

Note that the central peak at the origin of the symmetric function S
— S(∆x, ∆y) → ∞ when (∆x, ∆y) → (0, 0) — is placed at the origin
i = j = 0 of the grid of function values Sij . The peak of the function is
thus split into four “quadrants”, one in each corner of the grid.
Next:
1. The grid representations Sij and ∆gij thus obtained of the func-
tions S and ∆g are transformed to the frequency domain — they
become functions Suv and Guv of the two “frequencies”, the
wave indices u and v in the x and y directions. The spatial
frequencies or wave numbers3 ν ˜︁ and spatial wavelengths λ are
u L, ν v L , in which L = Nδx = Nδy is
/︁ /︁
˜︁x = λ−1
ν x = ˜︁y = λ−1
y =
the size of the area, assumed to be square.
2. They are multiplied with each other “one frequency pair at a
time”: we calculate
1
Tuv = S ·G , u, v = 0, . . . , N − 1. (9.3)
4πR uv uv
{︁ }︁
3. We transform the result, Tuv = F Tij , back to the space domain:
{︁ }︁
Tij = F−1 Tuv , a point grid Tij = T xi , yj of the disturbing
(︁ )︁

potential T . The disturbing potential of an arbitrary point can be

obtained from this grid by interpolation. The co-ordinates xi , yj
run as functions of i, j in the same way as described above for ∆g.
The method described is good for computing the disturbing potential T
— and similarly the geoid height N = T γ — from gravity anomalies
/︁

3 This is the so-called linear frequency, whole waves per unit of length. The angular
def
frequency is ω = 2π˜︁
ν, radians of phase angle per unit of length.

í Õ ! ¤. û
232 Spectral techniques, FFT

Observation
Interpolation Regular
points in their −−−−−−−−−−−−−−→
point grid
own places
⏐ ⏐
⏐ ⏐
↓Direct solution ↓FFT
Free solution Interpolation Regular
←−−−−−−−−−−−−−−
point selection point grid

^ Figure 9.2. Commutative diagram for FFT.

using the Stokes equation. It is just as good for evaluating other

quantities, like for example the vertical gradient of the gravity anomaly
using equation 8.12. The only requirement is that the equation can be
expressed as a convolution.
The inversion calculation is also easy, as we shall see: in the frequency
domain it is just a simple division.
Using the discrete Fourier transform requires the input data in a field
to be integrated — in the example, gravity anomalies — is given on
a regular grid covering the area of computation, or can be converted
into one. The result — in the example, the disturbing potential — is
obtained on a regular grid in the same geometry. Values can then be
interpolated to chosen locations.
kommutoiva The FFT method may be depicted as a commutative diagram, figure 9.2.
kaavio
Appendix C offers a short explanation of why FFT works and what
makes it as efficient as it is.

^ 9.3 Solution in latitude and longitude

In the above equation 9.2, the grid co-ordinates x and y are rectangular.
For practical reasons, we would rather use latitude and longitude (φ, λ)
as grid co-ordinates. In that way, the need to generate a new (x, y) point
grid by interpolating from the given (φ, λ) one through a map projection
calculation is avoided. However, working in geographical co-ordinates

í Õ ! ¤. û
 Solution in latitude and longitude
233
causes errors due to meridian convergence — as a latitude and longitude
co-ordinate system is not actually rectangular. The co-ordinate pair
(φ, λ cos φ) would be slightly more suitable.
The problem has also been addressed on a more conceptual level.

^ 9.3.1 The Strang van Hees method

The Stokes kernel function S(ψ) depends only on the geocentric angular
distance ψ between evaluation point (ϕ, λ) and observation point
(ϕ ′ , λ ′ ). The angular distance may be written as follows (cosine rule on
the sphere):

cos ψ = sin ϕ sin ϕ ′ + cos ϕ cos ϕ ′ cos λ − λ ′ .

(︁ )︁

Substitute

cos λ − λ ′ = 1 − 2 sin2 12 (λ − λ ′ ) ,
(︁ )︁

cos ψ = 1 − 2 sin2 12 ψ,
cos ϕ − ϕ ′ = 1 − 2 sin2 12 (ϕ − ϕ ′ ) ,
(︁ )︁

and obtain the half-angle cosine rule:

cos ψ = cos ϕ − ϕ ′ − 2 cos ϕ cos ϕ ′ sin2 21 (λ − λ ′ )

(︁ )︁

=⇒ sin2 21 ψ = sin2 21 (ϕ − ϕ ′ ) + cos ϕ cos ϕ ′ sin2 21 (λ − λ ′ ) .

Here we may use the following approximation:

def
cos ϕ ′ ≈ cos ϕ = cos ϕ0 ,

in which ϕ0 is a reference latitude in the middle of the calculation area. vertaus-

Now the above equation becomes leveysaste

sin2 21 ψ ≈ sin2 21 (ϕ − ϕ ′ ) + cos2 ϕ0 sin2 12 (λ − λ ′ ) , (9.4)

def def
which depends only on the differences ∆ϕ = ϕ − ϕ’ and ∆λ = λ − λ ′ ,
a requirement for convolution.

í Õ ! ¤. û
234 Spectral techniques, FFT

After this, the FFT method may be applied by using co-ordinates

(ϕ, λ)4 and the re-written Stokes kernel
(︄ √︂ )︄
def
S(∆ϕ, ∆λ) = S 2 arcsin sin2 12 ∆ϕ + cos2 ϕ0 sin2 21 ∆λ .

This clever way of using FFT in geographical co-ordinates was invented

by the Dutchman G. Strang van Hees5 in 1990.

^ 9.3.2 “Spherical FFT”, multi-band model

We divide the area into several narrow latitude bands. In each band
we apply the Strang van Hees method using its own optimal central
latitude.
Write the Stokes equation as follows:
x (︁
R
S ϕ − ϕ ′ , λ − λ ′ ; ϕ ∆g ϕ ′ , λ ′ cos ϕ ′ dϕ ′ dλ ′ ,
)︁ [︁ (︁ )︁ ]︁
N(ϕ, λ) =
4πγ
(9.5)
where we have expressed S(·) as a function of latitude difference,
longitude difference and evaluation latitude. Now, choose two support
latitudes, ϕi and ϕi+1 . Assume furthermore that between these S is a
linear function of ϕ. In that case we may write

(ϕ − ϕi ) Si+1 (∆ϕ, ∆λ) + (ϕi+1 − ϕ) Si (∆ϕ, ∆λ)

S(∆ϕ, ∆λ; ϕ) = ,
ϕi+1 − ϕi

where ∆ϕ = ϕ − ϕ ′ , ∆λ = λ − λ ′ , and

Si (∆ϕ, ∆λ) = S ϕ − ϕ ′ , λ − λ ′ ; ϕi ,
(︁ )︁

Si+1 (∆ϕ, ∆λ) = S ϕ − ϕ ′ , λ − λ ′ ; ϕi+1 .

(︁ )︁

We obtain by substitution into integral equation 9.5:

4 Inpractice one uses the geodetic or geographical latitude φ instead of ϕ without

significant error.
5 Govert L. Strang van Hees (1932–2012) was a Dutch gravimetric geodesist.

í Õ ! ¤. û
 Solution in latitude and longitude
235
(︃ )︃
R ϕi+1 − ϕ ϕ − ϕi
N(ϕ, λ) = I + I , (9.6)
4πγ ϕi+1 − ϕi i ϕi+1 − ϕi i+1
with
x
Si (∆ϕ, ∆λ) ∆g ϕ ′ , λ ′ cos ϕ ′ dϕ ′ dλ ′ ,
[︁ (︁ )︁ ]︁
Ii =
x
Si+1 (∆ϕ, ∆λ) ∆g ϕ ′ , λ ′ cos ϕ ′ dϕ ′ dλ ′ .
[︁ (︁ )︁ ]︁
Ii+1 =

This equation is the linear combination of two convolutions. Both are

evaluated by FFT. One forms the weighted mean from the solutions
obtained according to equation 9.6.
In this method we use, instead of approximative equation 9.4, an
exact equation in which ϕ ′ is expressed into ϕ and ∆ϕ:

sin2 12 ψ = sin2 12 (ϕ − ϕ ′ ) + cos ϕ cos ϕ ′ sin2 12 (λ − λ ′ ) =

= sin2 12 ∆ϕ + cos ϕ cos (ϕ − ∆ϕ) sin2 12 ∆λ.

Here again, we calculate Si and Si+1 for the support latitude values
ϕi and ϕi+1 , we evaluate the integrals with the aid of the convolution
theorem, and interpolate N(ϕ, λ) according to equation 9.6 when ϕi ⩽
ϕ < ϕi+1 . After this, the solution is not entirely exact, because inside
every band we still use linear interpolation. However by making the
bands narrower, we can keep the error arbitrarily small.

^ 9.3.3 “Spherical FFT”, Taylor expansion model

This somewhat more complicated but also more versatile approach
expands the Stokes kernel into a Taylor expansion with respect to latitude
about a reference latitude located in the middle of the computation area.6 vertaus-
Each term in the expansion depends only on the difference in latitude. leveysaste
The integral to be calculated similarly expands into terms, of which
each contains a pure convolution.

6 Inthe literature the method has been generalised by also expanding the kernel with
respect to height.

í Õ ! ¤. û
236 Spectral techniques, FFT

Let us write the general problem as follows:

w 2πw + π/2 (︁
′
)︁ [︁ (︁ ′ ′ )︁ ′
dϕ ′ dλ ′ ,
]︁
ℓ(ϕ, λ) = C ϕ, ϕ , ∆λ m ϕ , λ cos ϕ
0 − π/2

in which ℓ contains values to be computed, m values given, and C is the

coefficient or kernel function. Here only rotational symmetry around the
Earth’s axis is assumed for the geometry: the kernel function depends
only on the difference between longitudes ∆λ rather than the absolute
longitudes λ, λ ′ .
In a concrete case, m contains for example gravity anomaly values ∆g
in various points (ϕ ′ , λ ′ ), ℓ contains geoid heights N in various points
(ϕ, λ), and C contains coefficients calculated using the Stokes kernel
function.
We first change the dependence upon ϕ, ϕ ′ into a dependence upon
ϕ, ∆ϕ:
C = C ϕ, ϕ ′ , ∆λ = C(∆ϕ, ∆λ; ϕ).
(︁ )︁

Linearise:
C = C0 (∆ϕ, ∆λ) + (ϕ − ϕ0 ) Cϕ (∆ϕ, ∆λ) + . . .
vertaus- where we define for a suitable reference latitude ϕ0 :
leveysaste def (︁ )︁
C0 (∆ϕ, ∆λ) = C ∆ϕ, ∆λ; ϕ0 ,
⃓
∂def ⃓
Cϕ (∆ϕ, ∆λ) = C(∆ϕ, ∆λ; ϕ)⃓⃓ .
∂ϕ ϕ=ϕ 0

This expansion into two terms will work only for a limited range in ∆φ,
and the kernel function C is assumed to be of bounded support. In this
case, the integrals may be calculated within a limited area instead of
over the whole Earth.
Substitution yields
x
C(∆ϕ, ∆λ; ϕ) · m ϕ ′ , λ ′ cos ϕ ′ dϕ ′ dϕ ′ =
(︁ )︁
ℓ(ϕ, λ) =
x (︂ )︂
= C0 + (ϕ − ϕ0 ) Cϕ · m cos ϕ ′ dϕ ′ dϕ ′ =
x x
= C0 · m cos ϕ ′ dϕ ′ dλ ′ + (ϕ − ϕ0 ) Cϕ · m cos ϕ ′ dϕ ′ dλ ′ . (9.7)

í Õ ! ¤. û
Solution in latitude and longitude
237
It is important here now that the integrals in the first and second terms,
x
C0 (∆ϕ, ∆λ) m ϕ ′ , λ ′ cos ϕ ′ dϕ ′ dλ ′ = C0 ⊛ [m cos ϕ] ,
[︁ (︁ )︁ ]︁
x
Cϕ (∆ϕ, ∆λ) m ϕ ′ , λ ′ cos ϕ ′ dϕ ′ dλ ′ = Cϕ ⊛ [m cos ϕ] ,
[︁ (︁ )︁ ]︁

are both convolutions: both C functions depend only on ∆ϕ and ∆λ.

Both integrals can be calculated if both the data grid m cos ϕ and the
coefficient grids C0 and Cϕ are calculated first in preparation. After
this — in principle expensive, but, thanks to FFT and the convolution
theorem, a lot cheaper — integration, computing compound 9.7 is fast:
one multiplication and one addition for each evaluation point (ϕ, λ).
Example Let the evaluation area at latitude 60◦ be 10◦ × 20◦ in size. If
the grid mesh size is 5 ′ × 10 ′ , the number of cells is 120 × 120. Let
us choose, say, a 256 × 256 grid (so N = 256) and fill the missing
values with extrapolated values.
The values of the kernel functions C0 and Cϕ are calculated on
a 256 × 256 grid (∆ϕ, ∆λ) as well. The number of these is thus
also 65 536. Calculating the convolutions C0 ⊛ [m cos ϕ] and
Cϕ ⊛ [m cos ϕ] by means of FFT — i. e.,7
x
C0 (∆ϕ, ∆λ) m ϕ ′ , λ ′ cos ϕ ′ dϕ ′ dλ ′ =
(︁ )︁
{︁ }︁
= C0 ⊛ [m cos ϕ] = F−1 F{C0 } · F{m cos ϕ} ,
x
Cϕ (∆ϕ, ∆λ) m ϕ ′ , λ ′ cos ϕ ′ dϕ ′ dλ ′ =
(︁ )︁
{︁ }︁
= Cϕ ⊛ [m cos ϕ] = F−1 F{Cϕ } · F{m cos ϕ} ,
(︁ )︁
requires N2 × 2 log N2 = 65 536 × 16 = more than a million
operations, multiplication with (ϕ − ϕ0 ) and adding together,
each again 65 536 operations.
The grid matrices corresponding to functions C0 and Cϕ are
obtained as follows: for three reference latitudes ϕ−1 , ϕ0 , ϕ+1 we vertaus-
leveysaste
7 Fourier transforms are multiplied by multiplying the corresponding elements, see
section 9.2 equation 9.3.

í Õ ! ¤. û
238 Spectral techniques, FFT

compute numerically the grids

(︁ )︁
C−1 = C ∆ϕ, ∆λ; ϕ−1 ,
(︁ )︁
C0 = C ∆ϕ, ∆λ; ϕ0 ,
(︁ )︁
C+1 = C ∆ϕ, ∆λ; ϕ+1 ,

in which C0 is directly available, and

C+1 − C−1
Cϕ ≈ .
ϕ+1 − ϕ−1

Inversion calculation is thus also directly feasible. Let ℓ be given in

suitable point grid form. We compute the first approximation to m as
follows:8
{︃ }︃
(0) F{ℓ}
F{C0 } · F{m cos ϕ} = F{ℓ} =⇒ [m cos ϕ] = F −1
.
F{C0 }

The second approximation is obtained by first calculating

(0) (0)
ℓ(0) = C0 ⊛ [m cos ϕ] + (ϕ − ϕ0 ) · Cϕ ⊛ [m cos ϕ] ,

after which we make the improvement:

{︃ {︁ }︁ }︃
(1) (0) F ℓ − ℓ(0)
[m cos ϕ] = [m cos ϕ] +F −1
,
F{C0 }

and so on, iteratively. Two, three iterations are usually enough. This
method has been used to compute underground mass points from
gravity anomalies to represent the exterior gravity field of the Earth.9
More is explained in Forsberg and Vermeer (1992); Vermeer (1992).

8A Fourier transform is divided by another one by dividing the corresponding

elements, see section 9.2.
9 Because the relationship between the mass points and the observed gravity anomalies

on the Earth’s surface can be described exactly in geodetic co-ordinates, the method
may be used with geodetic latitude φ instead of geocentric latitude ϕ. Thus, errors
caused by ignoring the Earth’s flattening are avoided.

í Õ ! ¤. û
 Bordering and tapering of the data area
239
^ 9.3.4 “1D-FFT”
This is a limiting case of the previous ones, in which FFT is used only in
the longitude direction. In other words, this is a zones method in which
the zones have a width of only a single grid row. This method is exact
if all longitudes 0◦ ⩽ λ < 360◦ are along in the calculation. It requires
somewhat more computing time compared to the previous methods.
In fact, it is identical to a Fourier transform in variable λ, longitude.
Details are found in Haagmans et al. (1993).

^ 9.4 Bordering and tapering of the data area

The discrete Fourier transform presupposes the data to be periodically
continuous. In other words, it is assumed that when connecting the
eastern edge of the data area to the western edge, and the northern
edge to the southern edge, the data has to be continuous across these
edges.10 In practice, this is not the case. We are faced with two different
requirements:
◦ The data on the other side of an edge must be so far away as to
have no noticeable influence across the edge on the result of the
calculation.
◦ The data must be continuous across the edges.
Therefore, always when using FFT with the convolution theorem, two
measures need to be taken.
1. We continue the data by adding a border area to the data area,
so-called bordering. Often, the width of the border area is 25 % of
the size of the data area, making the surface area of the whole
calculation area four times that of the data area itself. The border
is filled with measured values where those exist, otherwise with
predicted (inter- or extrapolated) values.

10 Topologically the area with the edges thus connected is equivalent to a torus, and
the data is presupposed to be continuous on the surface of the torus.

í Õ ! ¤. û
240 Spectral techniques, FFT

25 % 50 % 25 %
Data area
1

^ Figure 9.3. “Tapering” 25 %.

The calculation area for the kernel function is also made similarly
four times larger. In this case the border area is filled with
real (computed) values. If the function is symmetric, it will
automatically be periodically continuous.
2. Because the discrete Fourier transform assumes periodicity, one
must make sure that the data is continuous across the edges. If
the values at the edges are not zero, they may be forced to zero by
multiplying the whole data area by a so-called tapering function,
which goes smoothly to zero towards the edges. Such a function
can easily be built: examples are a cubic spline polynomial or a
Tukey or cosine taper. See figure 9.3, showing a 25 % tapering
function, as well as example images 9.4, which show how non-
continuity — sharply differing left and right, and upper and lower,
edges — causes horizontal and vertical artefacts in the Fourier
Gibbsin ilmiö transform. These artefacts are related to the Gibbs phenomenon,
already mentioned in section 8.9: a sharp cut-off or edge in the
space domain will generate signal on all frequencies, up to the
highest ones.
Many journal articles have appeared on these technicalities. Groups that
were already involved in early development of FFT geoid determination
in the 1980s include Forsberg’s group in Copenhagen, Schwarz and
Sideris’ group in Calgary, Canada, the Delft group (Strang van Hees,
Haagmans, De Min, Van Gelderen), the Milanese group (Sansò, Barza-

í Õ ! ¤. û
 Computing a geoid model with FFT
241

Figure 9.4. Example images for FFT transform without (above) and with
(below) tapering. The online FFT service from Watts (2004) was
used. The images are amplitude spectra |Fuv | plotted with the
^ origin u = v = 0 in the centre, see appendix C.

ghi, Brovelli), Heiner Denker at the Hannover “Institut für Erdmessung”,

and many others.

^ 9.5 Computing a geoid model with FFT

Nowadays computing a geoid or quasi-geoid model is easy thanks to
increased computing power, especially using FFT. On the other hand, the
spread of precise geodetic satellite positioning has made the availability
of precise geoid models an important issue, so that one can use GNSS
technology for rapid and affordable height determination.

í Õ ! ¤. û
242 Spectral techniques, FFT

^ 9.5.1 GRAVSOFT software

The GRAVSOFTgeoid determination software has been mainly produced
in Denmark. Authors include Carl Christian Tscherning,11 René Fors-
berg, Per Knudsen, the Norwegian Dag Solheim, and the Greek Dimitris
Arabelos. The manual for the software is Forsberg and Tscherning
(2008).
This package is in widespread use and also offers, in addition to vari-
pienimmän ants of FFT geoid determination, for example least-squares collocation,
neliösumman as well as routines for evaluating various terrain effects. Its popularity
kollokaatio
can be partly explained by it being free for scientific use, and being
distributed as source code. It is also well-documented. Therefore it has
also found commercial use, for example in the petroleum extraction
industry.
GRAVSOFT has also been used a great deal for teaching, for example at
many research schools organised by the IAG (International Association
of Geodesy) in various countries. ISG, Geoid Schools.

^ 9.5.2 The Finnish FIN2000 geoid

Currently two geoid models are in use in Finland: FIN2000 (figure 9.5)
and FIN2005N00 (Bilker-Koivula and Ollikainen, 2009). The first model
vertauspinta is a reference surface for the N60 height system: using it together with
GNSS positioning allows determination of the N60 heights of points.
vertaus- The model gives geoid heights above the GRS80 reference ellipsoid. The
ellipsoidi second model is similarly a reference surface for the new N2000 height
system. It, too, gives heights from the GRS80 reference ellipsoid.
The precisions (mean errors) of FIN2000 and FIN2005N00 are on the
level of ± 2–3 cm.

11 Carl Christian Tscherning (1942–2014) was a Danish physical geodesist well-known

for his research into the gravity field of the Earth. He did ground-breaking work on
statistical computation methods for modelling the Earth’s gravity field from many
different measurement types.

í Õ ! ¤. û
 Computing a geoid model with FFT
243
20˚ 24˚ 28˚ 32˚

20
70˚ 70˚
19

68˚ 68˚
24 25

22
21
23

31
28
29
30

20
26
27

66˚ 19 66˚

17
18

18

64˚ 64˚

18
5
23 24 2

20 212
2

19

62˚ 62˚
18

17
19 16

60˚ 60˚
15

16

20˚ 32˚
24˚ 28˚

2010 Oct 20 13:27:28

^ Figure 9.5. The Finnish FIN2000 geoid. Data © Finnish Geodetic Institute.

í Õ ! ¤. û
244 Spectral techniques, FFT

^ 9.6 Use of FFT computation in other contexts

^ 9.6.1 Satellite altimetry
The Danish researchers Per Knudsen and Ole Balthasar Andersen
have computed a gravity map of the world ocean by starting from
satellite altimetry derived “geoid heights” and inverting them to gravity
anomalies (Andersen et al., 2010). A pioneer of this method has been
David Sandwell from the Scripps Institute of Oceanography in California
(e.g., Garcia et al., 2014). The short-wavelength features in the map can
tell us about the sea-floor topography.

^ 9.6.2 Satellite gravity missions and airborne gravimetry

The data from satellite gravity missions (like CHAMP, GRACE and GOCE)
can also be regionally processed using the FFT method: in the case
of GOCE, the inversion of gradiometric measurements yields geoid
heights on the Earth’s surface from measurements made at satellite
level. Airborne gravity measurements are also processed in this way
using FFT. The problem is called “harmonic downwards continuation”
and is in principle unstable.
Airborne gravimetry is a practical method for the gravimetric map-
ping of large areas. In the pioneering days, the gravity field over
Greenland was mapped, as well as many areas around the Arctic and
Antarctic. Later, areas were measured like the Brazilian Amazonas,
Mongolia, and Ethiopia (Bedada, 2010), for which no full-coverage
terrestrial gravimetric data existed. The advantage of this method is
that one measures rapidly large areas in a homogeneous way.

^ 9.7 Computing terrain corrections with FFT

Terrain correction is a very localised phenomenon, the calculation of
which requires high-resolution terrain data from a relatively small
area surrounding the computation point. Thus, calculating the terrain
correction is ideally suited for the FFT method.

í Õ ! ¤. û
Computing terrain corrections with FFT
245
We show how, with FFT, we can simply and efficiently evaluate the
terrain correction. We make the following simplifying assumptions:
◦ Terrain slopes are relatively gentle.
◦ The density ρ of the Earth’s crust is constant.
◦ The Earth is flat — the “shoebox world”.
These assumptions are not mandatory. The general case, however, leads
us into a jungle of equations without aiding the conceptual picture.
The terrain correction, the removal of the joint effect of all the topo-
graphic masses, or lacking topographic masses, above and below the
height level H of the evaluation point, can be calculated under these
assumptions using the following rectangular equation, which gives the
attraction of rock columns projected onto the vertical (figure 6.5):

x +∞ Gρ (︁H ′ (x ′ , y ′ ) − H(x, y) )︁
T C(x, y) = cos θ dx ′ dy ′ =
−∞ ℓ2
x +∞ Gρ (H ′ − H) ′
1H −H
= · dx ′ dy ′ =
−∞ ℓ2 2 ℓ
x +∞ (H ′ − H)2
1
= 2 Gρ dx ′ dy ′ . (9.8)
−∞ ℓ3
/︂ 1
/︂
′ ′
Here, Gρ (H − H) ℓ is the attraction of the column and 2
2 (H − H) ℓ
is the cosine of the angle θ between the force vector — assumed coming
from the midpoint of the rock column — and the vertical direction. This
is the so-called prism method.
We will make a linear approximation, wherein ℓ, the slant distance
between the evaluation point (x, y) and the moving data point (x ′ , y ′ ),
is the horizontal distance as well:
2 2
ℓ2 ≈ (x − x ′ ) + (y − y ′ ) .

Equation 9.8 is easy to check straight from Newton’s law of gravitation.

When it is assumed that the terrain is relatively free of steep slopes,
then ℓ is large compared to H ′ − H.

í Õ ! ¤. û
246 Spectral techniques, FFT

From equation 9.8 we obtain by expansion into terms:

x +∞ x +∞
1 1 H′
T C(x, y) = GρH2 dx ′ dy ′ − GρH dx ′ dy ′ +
2 −∞ ℓ3 −∞ ℓ3
x +∞ (H ′ )2
+ 12 Gρ dx ′ dy ′ , (9.9)
−∞ ℓ3
in which every integral is a convolution with kernel ℓ−3 , and the functions
to be integrated are 1, H ′ , and (H ′ )2 .
Unfortunately the function ℓ−3 as implicitly defined above has no
Fourier transform. Therefore, we change the above definition a tiny bit
by adding a small term:

ℓ2 = (x − x ′ ) 2 + (y − y ′ ) 2 + δ2 . (9.10)

The terms in the above equation 9.9 are large numbers that almost cancel
each other, giving a nearly correct result. Numerically this is however
an unpleasant situation. There is a solution for this which we describe
next.
If ℓ is defined according to equation 9.10, then the Fourier transform
of kernel ℓ−3 is (Harrison and Dickinson, 1989; Forsberg, 1984):

{︁ −3 }︁ 2π
(︃ )︃
2π 4π2 δ2 q2
F ℓ = exp(−2πδq) = 1 − 2πδq + − ··· ,
δ δ 1·2

def
√︂ √ /︂
in which q = ν ˜︁2y = u2 + v2 L , u and v are wave indices, and
˜︁2x + ν
˜︁x = u L and ν ˜︁y = v L (linear) “spatial frequencies” or wave numbers
/︁ /︁
ν
in the x and y directions in the (x, y) plane. If we substitute this into
equation 9.9, we notice that the terms containing 1 δ sum to zero, and
/︁

of course the terms containing positive powers of δ vanish as well when

δ → 0. We obtain (Harrison and Dickinson, 1989):

2π
(︂ )︂
F{T C} ≈ 12 GρH2 F{1} · (1 − 2πδq) −
δ
{︂ }︂ (︂
2π 2π
(︂ )︂ )︂
2
− GρH F{H ′ } · (1 − 2πδq) + 12 Gρ F (H ′ ) · (1 − 2πδq)
δ δ

í Õ ! ¤. û
Computing terrain corrections with FFT
247
where we left off all terms in higher powers of δ.
Re-order the terms:
{︁ }︁ π (︂ {︁ ′ 2 }︁)︂
F T C = Gρ H F{1} − 2HF{H } + F (H )
2 ′
+
δ (︂ {︁ 2 }︁
)︂
+ 2πGρ · 2πq · − 12 H2 F{1} + HF{H ′ } − 21 F (H ′ ) .

Because F{1} = 0 if q ̸= 0, the first term inside the second term will
always vanish. We obtain (remember that H is a constant, the height of
the evaluation point):
(︂ {︁
π 2 }︁
)︂
F{T C} = Gρ F H2 − 2HH ′ + (H ′ ) +
δ (︂ {︁ ′ 2 }︁)︂
+ 2πGρ · 2πq · HF{H } − 2 F (H )
′ 1

and the reverse Fourier transform yields

π
(︂ )︂
2
T C = Gρ H2 − 2H ′ H + (H ′ ) +
δ {︂ (︂ {︁ 2 }︁
)︂}︂
+ 2πGρ F−1 2πq · HF{H ′ } − 12 F (H ′ ) .

In the first term

2 2
H2 − 2H ′ H + (H ′ ) = (H − H ′ ) = 0

in point (x, y) in which H ′ = H, and we obtain

{︂ {︁ 2 }︁
(︂ )︂}︂
T C = 2πGρ F−1 2πq · H F{H ′ } − 12 F (H ′ ) ,

from which the troublesome 1 δ has now vanished.

/︁

A condition for this “regularisation” or “renormalisation” is that

′
H = H at point (x, y), i. e., the evaluation happens at the Earth’s surface.
The convolutions above are evaluated by the FFT method.
For calculating the terrain correction TC in the exterior space — exam-
ples are airborne gravimetry, the effect of the sea floor at the sea surface,
and the effect of the Mohorovičić discontinuity at the Earth’s surface —
there are techniques that express TC as a sum of convolutions, a Taylor
series expansion. An early paper on this is Parker (1972).

í Õ ! ¤. û
248 Spectral techniques, FFT

^ Self-test questions
1. What is the definition of a convolution?
2. Explain the convolution theorem.
3. Check that the dimensions of the quantities on both sides of
equation 9.2 match.
4. What is spatial frequency? What is the difference between linear
and angular spatial frequency?
5. Explain the basic idea of the Strang van Hees method.
6. What other approaches are there to applying the FFT method on a
curved (spherical or ellipsoidal) surface?
7. Why are bordering of the data area and tapering of the data
necessary?
8. In addition to geoid determination, where in physical geodesy is
the FFT method also used?
9. When computing the terrain correction on the Earth’s surface,
explain the “δ trick” used in the derivation. Why is it necessary,
and how does one make the δ vanish again?

í Õ ! ¤. û
^ Statistical methods

10
^ 10.1 The role of uncertainty in geophysics
In geophysics, we often obtain results based on uncertain, incomplete,
or otherwise deficient observational data. This also applies in the
study of the Earth’s gravity field: the density of gravity observations on
the Earth’s surface, for example, varies greatly, and large areas of the
oceans and polar regions are covered only by a very sparse network of
measurements. We speak of spatial undersampling.
Measurement technologies that work from space, on the other hand,
usually provide coverage of the whole globe, oceans, poles and all.
They, however, do not measure at a very high resolution. Either the
resolution of the method is limited — this holds for example for the
gravity-field parameters calculated from satellite orbit perturbations ratahäiriöt
— or the instruments measure only directly underneath the satellite’s
path, like satellite altimetry.
Another often relevant uncertainty factor is that one can do precise
measurements on the Earth’s surface, but inside the Earth the uncer-
tainty is much larger and the data is obtained much more indirectly.
In previous chapters we described techniques by which we could
calculate desired values or parameters for the Earth’s gravity field,
assuming that, for example, gravity anomalies are available everywhere
on the Earth’s surface, and with arbitrarily high resolution. In this

– 249 –
250 Statistical methods

chapter we look at mathematical means to handle real-world situations

where this is not the case.

^ 10.2 Linear functionals

In mathematics, a mapping that associates with every function in a
given function space a certain numerical value is called a functional.
One such is, for example, a (partial) derivative at a certain point x0 :
∂
⃓
f ↦→ f(x)⃓ .
⃓
∂x x=x0

A trivial functional is also the evaluation functional, the function value

itself (i. e., the “zeroth derivative”) for a certain argument value,
(︁ )︁
f ↦→ f x0 .

Other functionals are for example the integral over a given area σ:
w
f ↦→ f(x) dx,
σ

and so on.
We may write symbolically
∂⃓ ∂
⃓ ⃓
L= , meaning L{f} = f(x)⃓ .
⃓
∂x x=x0 ∂x
⃓
x=x0

A functional or operator is linear if

L{αf + βg} = αL{f} + βL{g}, α, β ∈ R.

Remember that all partial derivatives, as also the Laplace operator ∆, are
linear.
In physical geodesy, all interesting functionals are functionals of the
function T (ϕ, λ, R) = T (ϕ, λ, r)|r=R , i. e., of the disturbing potential at
the surface of a spherical Earth. The theory thus uses the spherical
approximation,1 and the surface of the sphere of radius R corresponds

1 This is not mandatory, but the error of approximation is usually small.

í Õ ! ¤. û
 Statistics on the Earth’s surface
251
def
to mean sea level. For example, the disturbing potential TP = T (ϕ, λ, R)
at a point P at sea-level location (ϕ, λ) is such a functional:

T (·, ·, R) ↦→ T (ϕ, λ, R).

If point P is not at sea level, a suitable functional also exists:

T (·, ·, R) ↦→ T (ϕ, λ, r).

If the quantity is not the disturbing potential, but, say, the gravity
anomaly or the deflection of the plumb line: luotiviivan
poikkeama
T (·, ·, R) ↦→ ∆g(ϕ, λ, r),
T (·, ·, R) ↦→ ξ(ϕ, λ, r),
T (·, ·, R) ↦→ η(ϕ, λ, r).

All these are also linear functionals. In fact, if we write

∑︂
∞
1 ∑︂
n
T (ϕ, λ, r) = Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) ,
rn+1
n=2 m=0

even the spherical-harmonic coefficients anm , bnm are all linear func- pallofunktio-
tionals of the disturbing potential T : kerroin

T (·, ·, R) ↦→ anm ,
T (·, ·, R) ↦→ bnm .

Here, T (·, ·, R) is shorthand for the whole function

T (ϕ, λ, R), ϕ ∈ − π 2 , + π 2 , λ ∈ [0, 2π) .

[︁ /︁ /︁ ]︁

^ 10.3 Statistics on the Earth’s surface

In statistics, we define a stochastic process as a stochastic quantity, or
random variable, the domain of which is a function space. In other words, määrittely-
it is a random variable, the realisation values of which are functions. A joukko

í Õ ! ¤. û
252 Statistical methods

stochastic process may be a quantity developing over time, the precise

behaviour of which is uncertain, for example a satellite orbit. In the
same way as for a (real-valued) stochastic quantity x we may define an
odotusarvo expected value or expectancy E{x} and a variance
{︁ }︁
Cxx = Var{x} = E (x − E{x})2 ,

we may also do so for a stochastic process. The only difference is that

by doing so we obtain functions.
Let, for example, the stochastic process x(t) be a function of time.
Then we may define its variance function as follows:
{︁ }︁
Cxx (t) = Var x(t) .

However, much more may be defined for a stochastic process, for

example the covariance of values of the same process taken at different
points in time, the autocovariance:
(︁ )︁ {︁ }︁
Axx t1 , t2 = Cov x(t1 ), x(t2 ) =
{︂(︁ {︁ }︁)︁ (︁ {︁ }︁)︁}︂
= E x(t1 ) − E x(t1 ) x(t2 ) − E x(t2 ) .

Similarly if we have two different processes, we may define the cross

covariance between them.
The argument of a stochastic process is commonly time t. However in
geophysics we study stochastic processes the arguments of which are
locations on the Earth’s surface, i. e., we talk of processes of the form
x(ϕ, λ). The definitions of auto- and cross-covariances work otherwise
in the same way, but in the case of the Earth we have a special prob-
lem. A stochastic quantity is generally defined as a quantity x from
which realisations x1 , x2 , x3 , . . . are obtained, which together have certain
statistical properties.
The classical example is the dice throw. A die can be thrown again
and again, and one can practise the art of statistics on the results of the
throws. Another classic example is measurement. Measurement of the

í Õ ! ¤. û
Statistics on the Earth’s surface
253
same quantity can be repeated, and is repeated, in order to improve
precision.
For a stochastic process defined on the Earth’s surface, the situation
is different.

We have only one Earth.

For this reason, statistics must be done in a somewhat different fashion.

Given a stochastic process on the surface of the Earth, x(ϕ, λ), we
define a quantity similar to the statistical expectancy E{·}, the geographic odotusarvo
mean
x w w
def 1 1 2π + π/2
M{x} = x(ϕ, λ)dσ = x(ϕ, λ) cos ϕ dϕ dλ. (10.1)
4π σ 4π 0 − π/2

Here x(ϕ, λ) is the one and only realisation of process x that is available
on this Earth.
Clearly this definition makes sense only in the case where the statis-
tical behaviour of the process x(ϕ, λ) is the same everywhere on Earth,
independently of location (ϕ, λ). This is called the assumption of homo-
geneity. It is in fact the assumption that the spherical symmetry of the
Earth extends to the statistical behaviour of her gravity field.
Similarly to the statistical variance based on expectancy, we may
define the geographic variance:
def {︁ }︁ def {︂(︁ )︁2 }︂
Cxx (ϕ, λ) = Var x(ϕ, λ) = M x − M{x} . (10.2)

The global average of gravity anomalies ∆g(ϕ, λ) vanishes2 based on

their definition:
M{∆g} = 0.

2 This is not exactly valid if, for example, the normal gravity field used in calculating
the anomalies contains the mass of the atmosphere, but gravity values measured close
to sea level do not contain the attraction of the atmosphere. There are other small
effects due to the non-realism of the normal field.

í Õ ! ¤. û
254 Statistical methods

In that case, equation 10.2 is simplified as follows:

{︁ }︁ {︁ }︁ x (︁
1 )︁2
C∆g∆g (ϕ, λ) = Var ∆g(ϕ, λ) = M ∆g2 = ∆g(ϕ, λ) dσ.
4π σ
The definition given here of the geographic mean M{·} is based on
integration of the one and only realisation over the surface of the Earth.
As has been seen, in statistics the mean is defined slightly differently, as
{︁ }︁
odotusarvo the expectancy of a stochastic process. For gravity anomalies it is E ∆g ,
in which ∆g is the anomaly considered as a stochastic process, the series
of values of ∆g that results if we look at an endless series of randomly
formed Earths. Not very practical!
If the expectancy of a stochastic process is the same as the mean of one
realisation computed by integration — and other statistical properties
are similarly the same — we speak of an ergodic process. Establishing
empirically in geophysics that a process is ergodic is typically difficult
to impossible.

^ 10.4 The covariance function of the gravity field

Defining a covariance function between points P and Q is more compli-
cated. Something like equations 10.1 and 10.2 cannot be used directly,
because both ∆gP and ∆gQ can move independently over the whole
Earth’s surface. We have
(︁ )︁ (︁ )︁
∆gP = ∆g ϕP , λP , ∆gQ = ∆g ϕQ , λQ .

In the following we assume that the covariance to be calculated will only

depend on the relative location of points P and Q. In a homogeneous
gravity field, the covariance function will not depend on the absolute
location of the points, but only on the difference in location between
points P and Q.
Write
(︁ )︁ (︁ )︁
ϕQ = ϕQ ϕP , λP , ψPQ , αPQ , λQ = λQ ϕP , λP , ψPQ , αPQ .

í Õ ! ¤. û
 The covariance function of the gravity field
255
α
P
Q

Earth’s centre of mass

^ Figure 10.1. Definition of geocentric angular distance and azimuth.

ϕQ and λQ can be computed3 if we know ϕP and λP as well as both the

geocentric angular distance ψPQ and the azimuth angle αPQ . See figure
10.1.
Now we may write
(︂ (︁ )︁ (︁ )︁)︂
∆gQ = ∆gQ ϕQ ϕP , λP , ψPQ , αPQ , λQ ϕP , λP , ψPQ , αPQ =
(︁ )︁
= ∆gQ ϕP , λP , ψPQ , αPQ ,

and we may define as the covariance function

(︁ )︁ def {︁ (︁ )︁ (︁ )︁}︁
C∆g∆g ψPQ , αPQ = MP ∆gP ϕP , λP ∆gQ ϕP , λP , ψPQ , αPQ =
x
1 (︁ )︁ (︁ )︁
= ∆gP ϕP , λP ∆gQ ϕP , λP , ψPQ , αPQ dσP .
4π σ
Also here, MP is a geographic-mean operator. First we fix point Q in
relation to point P: both azimuth αPQ and distance ψPQ are held fixed.
Point P, and with it, point Q, is moved over the whole of the Earth’s
surface. We compute the corresponding integral over the unit sphere

3 This is called the geodetic forward problem on the sphere.

í Õ ! ¤. û
256 Statistical methods

σP , and divide by 4π:

{︁ }︁ x
(︁ )︁ 1
C∆g∆g ψPQ , αPQ = MP ∆gP ∆gQ(P) = ∆g ∆g dσ =
4π σ P Q(P) P
w w
1 + π/2 2π
= ∆gP ∆gQ(P) dλP cos ϕP dϕP ,
4π − π/2 0
in which dσ = cos ϕ dλ dϕ was used, cos ϕ being Jacobi’s determinant
for the co-ordinates (ϕ, λ) on the unit sphere.
In addition to the assumption of homogeneity, we may make still
the assumption of isotropy: the covariance function — or more generally,
the statistical behaviour of the gravity field — does not depend on the
relative direction or azimuth αPQ of point pair (P, Q), but only on the
angular distance ψPQ between them. (This, too, is, like homogeneity, one
of the forms in which the Earth’s spherical symmetry is expressed.) In
this case we may compute the geographic mean in a slightly different
way, by also averaging over all azimuth angles αPQ ∈ [0, 2π):
(︁ )︁ def
C∆g∆g ψPQ =
{︁ }︁ w
def 1 2π {︁ }︁
= MP′ ∆gP ∆gQ(P) = MP ∆gP ∆gQ(P) dαPQ =
2π 0
w 2πw + π/2 w 2π
1
= 2 ∆gP ∆gQ(P) dλP cos ϕP dϕP dαPQ . (10.3)
8π 0 − π/2 0
Remark The true gravity field of the Earth is not terribly homogeneous
or isotropic, but in spite of this, both hypotheses are widely used.

^ 10.5 Least-squares collocation

^ 10.5.1 Stochastic processes in one dimension
pienimmän Collocation is a statistical estimation technique used to estimate the
neliösumman values of a stochastic process and calculate the uncertainties (like mean
kollokaatio
errors) of the estimates.
Let s(t) be a stochastic process, the autocovariance function of which
(︁ )︁
is C ti , tj . Let the process furthermore be stationary, in other words,

í Õ ! ¤. û
Least-squares collocation
257
(︁ )︁ (︁ )︁
for any two moments in time ti , tj it holds that C ti , tj = C tj − ti =
C(∆t). The argument t is generally time, but could be any parameter,
for example the distance of a journey.
Of this process, we have observations made at times t1 , t2 , . . . , tN ,
when the corresponding process values for those times are
s(t1 ), s(t2 ), . . . , s(tN ). Let us assume, for the moment, that these values
are error-free observations. Then the observations are function values of
process s, stochastic quantities, the variance matrix of which we may
write as follows:
⎡ (︁ )︁ (︁ )︁ (︁ )︁ ⎤
C t 1 , t1 C t2 , t1 · · · C t 1 , tN
..
{︁ }︁ ⎢ C t1 , t2
⎢ (︁ )︁ (︁ )︁ ⎥
C t 2 , t 2 · · · . ⎥
Var si = ⎢ .. .. ..
⎥.
. . .
⎢ ⎥
⎣ ⎦
(︁ )︁ (︁ )︁ (︁ )︁
C t1 , t N C t2 , t N · · · C tN , tN
We also call this autocovariance matrix the signal variance matrix of s.
(︁ )︁
We use the symbol Cij for this, both for one[︂ element Cij = C ti , tj of ]︂
(︁ )︁
the matrix and for the whole matrix: Cij = C ti , tj , i, j = 1, . . . , N .
[︂ ]︂
The symbol si again denotes a vector s(ti ), i = 1, . . . , N consisting
of process values — or one of its elements s(ti ).
(︁ )︁
Note that, if the function C ti , tj , or C(∆t), is known, then the whole
matrix and all of its elements can be calculated, provided all argument
values (observation times) ti are known.
Let the shape of the problem now be that one should estimate, i. e.,
predict, the value of process s at the moment in time T , i. e., s(T ), based
on our knowledge of the above-described observations s(ti ), i = 1, . . . , N.
In the same way as we calculated above the covariances between s(ti )
and s(tj ) (elements of the signal variance matrix Cij ), we also compute
the covariances between s(T ) and all s(ti ), i = 1, . . . , N. We obtain
{︁ }︁ [︂ (︁ )︁ (︁ )︁ (︁ )︁ ]︂
Cov s(T ), s(ti ) = C T , t1 C T , t2 · · · C T , tN .

For this we may use the notation CT j . It is assumed here that there is
only one point in time T for which estimation is done. Generalisation to

í Õ ! ¤. û
258 Statistical methods

the case where there are several Tp , p = 1, . . . , M, is straightforward. In

that case, the signal covariance matrix will be of size M × N:
⎡ ⎤
C(T1 , t1 ) C(T1 , t2 ) · · · C(T1 , tN )
{︁ (︁ )︁ }︁ ⎢ C(T2 , t1 ) C(T2 , t2 ) · · · C(T2 , tN ) ⎥
⎢ ⎥
Cov s Tp , s(ti ) = ⎢ ⎢ .. .. .. ⎥.
. . .
⎥
⎣ ⎦
C(TM , t1 ) C(TM , t2 ) · · · C(TM , tN )

For this we may use the more general notation Cpj .

^ 10.5.2 Signal and noise

The process s(t) is called the signal. It is a physical phenomenon that we
are interested in. There also exist physical phenomena that are otherwise
similar, but that we are not interested in: on the contrary, we wish to
remove their influence. Such stochastic processes are called noise.
When we make an observation, the purpose of which is to obtain
a value for the quantity s(ti ), we obtain in reality a value that is not
absolutely precise. The real observation thus is

ℓi = s(ti ) + ni . (10.4)

Here, ni is a stochastic quantity: observational error or noise. Let its

variance — or more precisely, the joint noise variance matrix of multiple
observations — be Dij . This is a very similar matrix to the above Cij ,
and also symmetric and positive definite. The only difference is that
Dij designates noise, which we are not interested in. Often it may be
assumed that the errors ni , nj of two different observations ℓi , ℓj do not
correlate, in which case Dij is a diagonal matrix.

^ 10.5.3 Estimator and variance of prediction

Now we construct an estimator
(︁ )︁ def ∑︂
s Tp =
ˆ︁ Λpi ℓi ,
i

í Õ ! ¤. û
 Least-squares collocation
259
as a linear combination of the observations at our disposal ℓi . The
purpose in life of this estimator is to get as close as possible to s(Tp ). So,
the quantity to be minimised is the difference
(︁ )︁ (︁ )︁ (︁ )︁ (︁ )︁ (︁ )︁
s Tp − s Tp = Λpi ℓi − s Tp = Λpi s(ti ) + ni − s Tp .
ˆ︁

Here, for the sake of writing convenience, we left the summation sign
∑︁
off (Einstein summation convention): We always sum over adjacent,
identical indices, in this case i.
Study the variance of this difference, the so-called variance of prediction: ennustus-
def {︁ (︁ )︁ (︁ )︁}︁ varianssi
Σpp = Var ˆ︁ s Tp − s Tp .

We exploit propagation of variances, the notations introduced above, and varianssien

our knowledge that surely there is no physical relationship, or correlation, kasautuminen
between observation process noise n and signal s:
{︂(︁ )︁ (︁ )︁}︂
Cov s(ti ) + ni , s(tj ) + nj =
{︁ }︁ {︁ }︁
= Cov s(ti ), s(tj ) + Cov ni , nj = Cij + Dij ,

and4
{︂(︂ (︁ )︁ (︁ )︁ )︂ (︂ (︁ )︁ (︁ )︁ )︂}︂
Σpq = Cov ˆ︁ s Tp − s Tp , ˆ︁ s T q − s Tq =
{︂(︁ )︁ (︁ )︁}︂ {︂ (︁ )︁ (︁ )︁}︂
= Λpi Cov s(ti ) + ni , s(tj ) + nj Λjq + Cov s Tp , s Tq −
{︁ (︁ )︁}︁ {︁ (︁ )︁ }︁
− Λpi Cov s(ti ), s Tq − Cov s Tp , s(tj ) Λjq =
= Λpi (Cij + Dij ) Λjq + Cpq − Λpi Ciq − Cpj Λjq . (10.5)

The variances, or diagonal elements, Σpp of the matrix are now obtained
by setting q = p.

^ 10.5.4 Showing optimality

Here we show that the optimal estimator is indeed the one producing
the minimum possible variances.

4 The matrix Ciq is the transpose of Cpj , the matrix Λjq the transpose of Λpi .

í Õ ! ¤. û
260 Statistical methods

Choose
def
Λpj = Cpi (Cij + Dij )−1 .
Then, from equation 10.5 and exploiting the symmetry of the C and D
matrices, we obtain

Σpp = Cpi (Cij + Dij )−1 Cjp + Cpp −

− Cpi (Cij + Dij )−1 Cjp − Cpi (Cij + Dij )−1 Cjp =
= Cpp − Cpi (Cij + Dij )−1 Cjp . (10.6)

Let us study next the alternative choice

Λpj = Cpi (Cij + Dij )−1 + δΛpj .

In this case we obtain by substitution

I
⏟ ⏞⏞ ⏟ II
⏟ ⏞⏞ ⏟ ⏟ ⏞⏞ ⏟
III
′
Σpp = Λpi (Cij + Dij ) Λjp + Cpp − Λpj Cjp − Cpi Λip =
I
⏟[︂ ]︂ ⏞⏞ [︂ ]︂⏟
−1 −1
= Cpi (Cij + Dij ) + δΛpj (Cij + Dij ) (Cjk + Djk ) Ckp + δΛkp +
II III
⏟ [︂ ⏞⏞ ]︂ ⏟ ⏟ [︂ ⏞⏞ ]︂⏟
−1 −1
+ Cpp − Cpi (Cij + Dij ) + δΛpj Cjp − Cpi (Cij + Dij ) Cjp + δΛip =
I
⏟
hhhh ˂
⏞⏞ ⏟
˂−1 ˂˂˂
=˂
Cpi
˂ (C
˂ ˂
h
ij
˂ +˂D
h ˂h
hij ) h C
hhjp
h
+ Cpi δΛ
hhhh
hip
h˂+˂δΛ
˂˂pi Cip + δΛpi (Cij
˂˂ + Dij ) δΛjp +
II
⏟
hhhh
⏞⏞ ⏟ ⏟ ⏞⏞
III
⏟
˂ ˂˂˂˂
−1 −1
+ Cpp −˂C˂ (C
˂˂+ hDhij )hhC ˂ pi Cip − Cpi (Cij
−˂δΛ + Dij ) Cjp − Cpi δΛ h=
hh ˂
h ˂ ˂ ˂ ˂ hhhh
˂ pi˂ ij hjp hip
h˂
= Cpp − Cpi (Cij + Dij )−1 Cjp + δΛpi (Cij + Dij ) δΛjp .
Here, the last term — the only difference from result 10.6 — is positive,
′
because the matrices Cij and Dij are positive definite: Σpp > Σpp , except
when δΛpi = 0. In other words, the solution given above,

Λpj = Cpi (Cij + Dij )−1 =⇒ ˆ︁ s Tp = Cpi (Cij + Dij )−1 ℓj ,

(︁ )︁

is optimal in the sense of least-squares — more precisely, in the sense of

ennustus- minimising the variance of prediction Σpp .
varianssi
í Õ ! ¤. û
 Least-squares collocation
261
^ 10.5.5 The covariance function of gravity anomalies
Least-squares collocation is used extensively to optimally estimate pienimmän
gravity values and other functionals of the gravity field on the Earth’s neliösumman
kollokaatio
surface.
If we have two points, P and Q, with measured gravity anomalies
(︁ )︁ (︁ )︁
∆gP = ∆g ϕP , λP and ∆gQ = ∆g ϕQ , λQ , we would like to have the
covariance between these two anomalies,
{︁ }︁
Cov ∆gP , ∆gQ .

As argued in section 10.4, we can only empirically derive such a

covariance by looking at all point pairs (P, Q) that are in the same
relative position around the globe, and averaging over them using the
M or M ′ operator.
Normally the covariance is assumed to depend only on the geocentric
angular distance ψ between points P and Q. Then, we speak of an
isotropic process ∆g(ϕ, λ). Then, the covariance will also be
{︁ }︁ {︁ }︁
Cov ∆gP , ∆gQ = MP′ ∆gP ∆gQ(P) = C ψPQ .
(︁ )︁

A popular covariance function for gravity anomalies is Hirvonen’s5

equation:
C0
C(ψ) = (︁ /︁ )︁2 , (10.7)
1 + ψ ψ0
in which C0 = C(0) and ψ0 are parameters describing the behaviour of
{︁ }︁ {︁ }︁
the gravity field. C0 = Var ∆g(ϕ, λ) = M ∆g2 is called the signal
variance, ψ0 the correlation length. ψ0 gives the distance at which the
correlation between the gravity anomalies in two points is still 50 %.6
In local applications, instead of the angular distance ψ one uses the
linear distance
s = ψR,

5 ReinoAntero Hirvonen (1908–1989) was a Finnish physical and mathematical

geodesist.

í Õ ! ¤. û
262 Statistical methods

0.8

0.6
C(ψ)
0.4

0.2
4
2
−4 −2 0
0 2 −2 y
x 4 −4

Figure 10.2. Hirvonen’s covariance function in two dimensions. C0 = ψ0 = 1

^ is assumed.

where R is the radius of the Earth. Then

C0
C(s) = (︁ /︁ )︁2 .
1+ s d
This equation was derived from gravimetric data for Ohio state, USA,
but it has broader validity. C(0) = C0 , the signal variance when s = 0.
The variable d = Rψ0 is also called the correlation length. It is the
distance d for which C(d) = 21 C0 , as seen from the equation.
The quantity C0 varies considerably between areas, from hundreds to
thousands of mGal2 , and tends to be largest in mountainous areas. The
quantity d is generally of the order of magnitude of tens of kilometres.
Alternative functions that are also often used in local applications
include the covariance function of the stationary Gauss–Markov process,

6 The correlation is
C0
{︁ }︁ (︁ /︁ )︁2
{︁ }︁ Cov ∆gP
, ∆g Q 1 + ψ ψ0 1
Corr ∆gP , ∆gQ = √︂ {︁ }︁ {︁ }︁ = √ = (︁ /︁ )︁2 ,
Var ∆gP Var ∆gQ C0 C0 1 + ψ ψ0

which is 0.5 if ψ = ψ0 .

í Õ ! ¤. û
 Least-squares collocation
263
20

16

12
∆g
8

30 40
y 20 30
20
10 10 x

Figure 10.3. An example of least-squares collocation. Here are given two data
points (stars); the surface plotted gives the estimated value ∆g
ˆ︂ P
for each point P in the area. We use least-squares collocation for
^ inter- and extrapolating gravimetric data.

and a quadratic variant:

(︂ (︁ /︁ )︁2 )︂
C(ψ) = C0 exp − ψ ψ0 , C(ψ) = C0 exp − ψ ψ0
(︁ /︁ )︁
.

^ 10.5.6 Least-squares collocation for gravity anomalies

If N points Pi , i = 1, . . . , N are given, where were measured gravity
(︁ )︁
values — more precisely, anomalies — ∆gi = ∆g ϕi , λi , we may, as
above, construct a signal variance matrix
def {︁ }︁
Cij = Var ∆gi =
⎡ (︁ )︁ (︁ )︁ ⎤ ⎡ ⎤
C0 C ψ21 · · · C ψN1 C0 C21 · · · CN1
⎢ (︁ )︁ (︁ )︁ ⎥ ⎢
⎢ C ψ12 C0 · · · C ψN2 ⎥ ⎢ C12 C0 · · · CN2 ⎥
⎥
=⎢ .. .. .. ⎥ = ⎢ .. .. ⎥,
.. ⎥
⎢ ⎥ ⎢
⎣ . . . ⎦ ⎣ . . . ⎦
(︁ )︁ (︁ )︁
C ψ1N C ψ2N · · · C0 C1N C2N · · · C0
(︁ )︁
in which all elements C ψij are calculated using covariance function
10.7 given above.

í Õ ! ¤. û
264 Statistical methods

If we also compute for the point P at which gravity is unknown:

{︁ }︁ [︂ (︁ )︁ (︁ )︁ (︁ )︁ ]︂ def
Cov ∆gP , ∆gi = C ψP1 C ψP2 · · · C ψPN = CPi ,

pienimmän we obtain, in the same way as before, for the least-squares collocation
neliösumman solution
kollokaatio −1 −1
ˆ︂ ∆gP = CPi (Cij + Dij ) ℓj ≈ CPi Cij ℓj ,
in which the ℓj = ∆gj + nj are gravity anomaly observations made in
points j = 1, . . . , N. The matrix Dij , which we leave out of consideration
here, again describes the random observation error, observation uncer-
tainty, or noise ni associated with making those observations. Often
Dij is a diagonal matrix, meaning that the observations are statistically
independent and do not correlate with each other.
We may also compute a precision assessment of this solution, the
ennustus- variance of prediction, equation 10.10:
varianssi
ΣPP = C0 − CPi (Cij + Dij )−1 CjP ≈ C0 − CPi C−1
ij CjP

in the case of one unknown prediction point P. Its square root

√︁
σ∆gP = ΣPP

is the mean error of estimator ∆g

ˆ︂ P .

^ 10.5.7 Calculation example

See figure 10.4. Two points are given where gravity has been measured
and gravity anomalies calculated: ∆g1 = 15 mGal, ∆g2 = 20 mGal. The
co-ordinates in the x and y directions are in kilometres. It is assumed
that between the gravity anomalies of different points, Hirvonen’s
covariance function,
C0
C(s) = (︁ /︁ )︁2 ,
1+ s d
applies, in which d = 20 km and C0 = ±1000 mGal2 . In addition, it
is assumed that the gravity measurements done — including height

í Õ ! ¤. û
 Least-squares collocation
265
y
1 (15 mGal)
30
P′
2 (20 mGal)
20

10 P

x
10 20 30

^ Figure 10.4. Collocation example.

determination of the gravity points! — were errorless. So, Dij = 0, i, j =

1, 2.
Calculate an estimate of the gravity anomaly ∆g
ˆ︂ P at point P and its
√
mean error σPP = ΣPP .
Calculate first the distances s and the corresponding covariances C.
(︂ )︂
s212 = (30 − 20)2 + (20 − 30)2 km2 = 200 km2 ,
2
1000 mGal
C12 = C21 = = 666.66 . . . mGal2 ,
200
/︁
1+ 400
(︂ )︂
s21P = (30 − 10)2 + (20 − 10)2 km2 = 500 km2 ,
2
1000 mGal
C1P = = 444.44 . . . mGal2 ,
1 + 500 400
/︁
(︂ )︂
2 2
s22P = (20 − 10) + (30 − 10) km2 = 500 km2 ,
2
1000 mGal
C2P = = 444.44 . . . mGal2 .
1 + 500 400
/︁

From this follows

[︄ ]︄ [︄ ]︄
C11 C12 1000 666.66
Cij + Dij = Cij = = mGal2 ,
C21 C22 666.66 1000

í Õ ! ¤. û
266 Statistical methods

and its inverse matrix

[︄ ]︄
−1 0.0018 −0.0012
(Cij + Dij ) = mGal−2 .
−0.0012 0.0018
Furthermore
[︂ ]︂ [︂ ]︂
CPi = CP1 CP2 = 444.44 444.44 mGal2 .

As the vector of observations is

[︄ ]︄ [︄ ]︄
∆g1 15
∆gj = = mGal,
∆g2 20

we obtain the result

[︄ ]︄ [︄ ]︄
[︂ ]︂ 0.0018 −0.0012 15
∆g
ˆ︂ P = 444.44 444.44 mGal =
−0.0012 0.0018 20
= 9.333 mGal.

ennustus- The precision, the variance of prediction, equation 10.10:

varianssi
ΣPP = CPP − CPi (Cij + Dij )−1 CjP = C0 −
[︄ ]︄ [︄ ]︄
[︂ ]︂ 0.0018 −0.0012 444.44
− 444.44 444.44 mGal2 =
−0.0012 0.0018 444.44
= 762.96 mGal2 ,

so
√︁
σ∆gP = ΣPP = ±27.622 mGal.
Summarising the result:

ˆ︂ P = 9.333 ± 27.622 mGal.

∆g

Observe that the gravity anomaly estimate found is much smaller than
its own uncertainty, and thus does not differ significantly from zero. In fact,
not using the observational data at all would leave us with the a priori
estimate
√
ˆ︂ P = 0 ± 1000 mGal = 0 ± 31.623 mGal,
∆g

í Õ ! ¤. û
 Least-squares collocation
267
almost as good.
If, instead, we had used point P ′ in between points 1 and 2, at location
(25 km, 25 km) , then
/︂
2 (︁
1000 mGal 1 + 50 400 = 888.89 mGal
2
/︁ )︁
CP 1 = CP 2 =
′ ′

ˆ︂ P ′ = 18.667 ± 7.201 mGal, which is clearly better than the a priori

and ∆g
estimate of zero.
And if we had chosen instead the Gauss–Markov covariance function

C = C0 exp − s d
(︁ /︁ )︁

we would have obtained the results ∆g ˆ︂ P = 7.663 ± 29.272 mGal for the
ˆ︂ P ′ = 16.460 ± 18.426 mGal for the shifted
original point location, and ∆g
point location.

^ 10.5.8 Theory of least-squares collocation

Above we presented one popular application of least-squares collocation. pienimmän
Here we look at the method more generally. The basic equation is neliösumman
kollokaatio
−1 (︁ )︁
ˆ︁f = Cfg (Cgg + Dgg ) g+n . (10.8)

The vector g contains observed quantities gi , the vector n contains the

observational noise, and ˆ︁f is a vector of quantities fˆ︁p to be predicted. The
hat is a commonly used symbol for an estimator.
Both vectors g and ˆ︁f can, for example, be gravity anomalies, in which
case we have homogeneous prediction, a type of inter- or extrapolation.
More generally ˆ︁f and g are of different types: for example ˆ︁f consists of
geoid heights Np and g of gravity anomalies ∆gi . In the latter case, the
Stokes equation is “covertly” along in the structure of the C matrices.
These matrices are built from covariance functions. Their elements

í Õ ! ¤. û
268 Statistical methods

can be expressed as follows:7

[︂ ]︂ [︂ ]︂ [︂ ]︂ {︁ }︁
Cfg = M{fp gi }, Cgg = M{gi gj }, Dgg = E ni nj ,
pi ij ij

in which ni , an element of vector n, represents the uncertainty of the

observation process appearing in observation equation 10.4:

ℓi = gi + ni , or equivalently ℓ =g + n.

ℓ is the vector of the observation values themselves, including observa-

tion uncertainty n.
The D matrix is the variance matrix of observational uncertainty, the
noise variance matrix describing a property of the observational process,
not of the gravity field. While the values of M{∆gi ∆gj } can be as large as
{︁ }︁
1200 mGal2 , the values of E ni nj can be much smaller, depending on
the measurement technique used, for example as small as 0.01 mGal2 .
This does not apply however in the case of block averages — for
example averages over blocks of size 1◦ × 1◦ , computed from scattered
measurements — which are often very imprecise.
pienimmän The great advantage of least-squares collocation is its flexibility. Dif-
neliösumman ferent observation types may be handled with a single unified theory
kollokaatio
and method, the locations of observation points are free, and the result
is obtained directly as freely choosable quantities in locations where
one wants them.

^ 10.6 Prediction of gravity anomalies

If the quantity to be calculated or estimated, ˆ︁f, is of the same type as the
observed quantity, g, we often speak of homogeneous prediction. For ex-
ample, the prediction equation for gravity anomalies already presented

7 Here,we use the geographic mean M{·} for evaluating the signal covariances. In
doing so, f and g are no longer considered stochastic. It is assumed that their global
geographic mean vanishes: M{f} = M{g} = 0.

í Õ ! ¤. û
Prediction of gravity anomalies
269
in subsection 10.5.6 is obtained from equation 10.8 by substitution:

ˆ︂ P = CPi (Cij + Dij )−1 ℓj .

∆g (10.9)

Here are several points j where gravity is given: let us say, N observations
ℓj = ∆gj + nj , j = 1, . . . , N. The number of points to be predicted may
be one, i. e., point P, or many. The matrices Cij and Dij are square, and
the inverse of their sum exists. CPi is a rectangular matrix. If there is
only one point P, it is a size 1 × N row matrix.
The prediction error is now the difference quantity8 ∆g ˆ︂ P − ∆g , and
P
its variance (“variance of prediction”) is ennustus-
varianssi
def {︁ }︁
ΣPP = Var ∆g ˆ︂ P − ∆g =
{︁ }︁ {︁P }︁ {︁ }︁ {︁ }︁
= Var ∆gP + Var ∆gP − Cov ∆g
ˆ︂ ˆ︂ P , ∆g − Cov ∆g , ∆g
P P
ˆ︂ P .

Here (propagation of variances applied to equation 10.9): varianssien

kasautuminen
{︁ }︁
Var ∆gˆ︂ P = CPi (Cij + Dij )−1 (Cjk + Djk ) (Ckℓ + Dkℓ )−1 CℓP =
= CPi (Cij + Dij )−1 CjP

and
{︁ }︁ {︂ (︂ )︂ }︂
−1
Cov ∆gP , ∆gP = Cov CPi (Cij + Dij )
ˆ︂ ∆gj + nj , ∆gP =
(︂ {︁ }︁ )︂
= CPi (Cij + Dij )−1 Cov ∆gj , ∆gP + 0 =
= CPi (Cij + Dij )−1 CjP ,

and also
{︁ }︁
ˆ︂ P = CPi (Cij + Dij )−1 CjP
Cov ∆gP , ∆g
{︁ }︁
and finally, the signal variance Var ∆gP = CPP .

8 Beaware that here, ∆gP is the true value of the gravity anomaly at point P, which we
do not know empirically. The measured value would be ℓP = ∆gP + nP , in which nP is
the random error or “noise” of the gravimetric observation.

í Õ ! ¤. û
270 Statistical methods

Here, CiP (also called CjP , or even CℓP ) is the transpose of CPi . The
matrix (Cij + Dij )−1 is symmetric and its own transpose.
The end result is

ΣPP = CPi (Cij + Dij )−1 CjP + CPP −

− CPi (Cij + Dij )−1 CjP − CPi (Cij + Dij )−1 CjP =
= CPP − CPi (Cij + Dij )−1 CjP .

In case Dij ≪ Cij , we obtain a simpler, often-used result:

ΣPP ≈ CPP − CPi C−1

ij CjP . (10.10)

Borderline cases
◦ Point P is far from all points i. Then CPi ≈ 0 and ΣPP ≈ CPP ,
so prediction is impossible in practice, and the prediction
equation 10.9 will yield the value zero. The mean error of
√ √
prediction σ∆gP = ΣPP is the same as the variability CPP
of the gravity anomaly signal, the square root of the signal
variance.
◦ Point P is identical with one of the points i. Then, if we use
only that point i, we obtain

ΣPP = CPP − CPP C−1

PP CPP = 0,

no prediction error whatsoever — as the value at the predic-

tion point was already known!.
However, if DPP ̸= 0 (but small), the result is ΣPP ≈ DPP .

^ 10.7 Covariance function and degree variances

^ 10.7.1 The covariance function of the disturbing potential
In theoretical work we use, instead of gravity anomalies, the covariance
function of the disturbing potential T on the Earth’s surface:
def {︁ }︁
K(P, Q) = K(ψPQ , αPQ ) = MP TP TQ(P) ,

í Õ ! ¤. û
Covariance function and degree variances
271
or alternatively using equation 10.3:
)︁ def {︁ }︁
K(P, Q) = K ψPQ = MP′ TP TQ(P) =
(︁
w 2πw + π/2 w 2π
1
= 2 T T dλ cos ϕP dϕP dαPQ . (10.11)
8π 0 − π/2 0 P Q(P) P

Here it is assumed that the disturbing potential is isotropic: K does not

depend on α but only on ψ.
We choose on the unit sphere a co-ordinate system where point P is a
“pole”. In this system, the parameters αPQ and ψPQ are the spherical
co-ordinates of point Q. The covariance function is expanded into the
following sum:

∑︂
∞ ∑︂
n
K(ψ) = knm Ynm (ψ, α)
n=2 m=−n

with Ynm defined as in equation 3.3:

⎧
⎨P (cos ψ) cos mα if m ⩾ 0,
nm
Ynm (ψ, α) = (10.12)
⎩Pn|m| (cos ψ) sin |m| α if m < 0.

Based on isotropy, all coefficients for which the order m ̸= 0, vanish:

the expressions on the right-hand side of equation 10.12 can only be
independent of α if m = 0. So

∑︂
∞ ∑︂
∞
K(ψ) = kn0 Yn0 (ψ) = kn Pn (cos ψ). (10.13)
n=2 n=2

The coefficients kn are called the degree variances (of the disturbing astevarianssi
potential). For isotropic covariance functions K(ψ), the information
content of the degree variances kn , n = 2, 3, . . . is the same as that of
the function itself, and is in fact its spectral representation.

í Õ ! ¤. û
272 Statistical methods

^ 10.7.2 Degree variances and spherical-harmonic coefficients

Multiply equation 10.13 with Pn ′ (cos ψ) sin ψ and integrate:
wπ
K(ψ)Pn ′ (cos ψ) sin ψ dψ =
0
∑︂
∞ wπ
= kn Pn (cos ψ)Pn ′ (cos ψ) sin ψ dψ =
0
n=2
∑︂
∞ w1
2
= kn Pn (t)Pn ′ (t) dt = kn ′
−1 2n + 1
n=2

using orthogonality condition 3.4. It follows that

w
2n + 1 π
kn =K(ψ)Pn (cos ψ) sin ψ dψ, (10.14)
2 0

meaning that, if K(ψ) is given, we can calculate all kn .

Substituting K(ψPQ ) from equation 10.11 yields, with abbreviations
ϕ = ϕP , λ = λP , ψ = ψPQ , α = αPQ :

K(ψ)
⏟ ⏞⏞ ⏟
wπ w 2πw + π/2 w 2π
2n + 1 1
kn = TP TQ(P) dλ cos ϕ dϕ dα Pn (cos ψ) sin ψ dψ =
2 0 8π2 0 − π/2 0
I
w + π/2 w 2π ⏟w w ⏞⏞ ⏟
2n + 1 2π π
= TP TQ(P) Pn (cos ψ) sin ψ dψ dαdλ cos ϕ dϕ.
16π2 − π/2 0 0 0

Here we have interchanged the order of the integrals, as is allowed,

and moved TP to another place.
The expression I is a surface integral over the unit sphere:
w 2πw π (︁ )︁
I= TQ(P) Pn cos ψPQ sin ψPQ dψPQ dαPQ =
0 0
x (︁ )︁ 4π
= TQ(P) Pn cos ψPQ dσQ = T ,
σ 2n + 1 n,P
in which Tn,P = Tn (ϕP , λP ). Tn is the constituent of the disturbing
asteosuus- potential T for the harmonic degree number n, compare the degree
yhtälö

í Õ ! ¤. û
Covariance function and degree variances
273
constituent equation 3.8. Substitution yields

w + π/2 w 2π
1
kn = TP Tn,P dλP cos ϕP dϕP =
4π − π/2 0
x x {︁ }︁
1 1
= T Tn dσ = M{T Tn } = Tn2 dσ = M Tn2 ,
4π σ 4π σ
according to the definition of operator M and considering the mutual
orthogonality of the functions Tn .

The degree variances are the geographic variances of the degree con-
stituents of the disturbing potential.

We write based on equation 3.13 :

T (ϕ, λ, r) =
GM⊕ ∑︂ (︂ R )︂n+1 ∑︂
∞ n
(︁ )︁
= r P nm (sin ϕ) δC nm cos mλ + Snm sin mλ ,
R
n=2 m=0

∗
in which the normal field, coefficients Cn , has been subtracted out:
⎧
⎨δC = C − C∗ if n even,
n0 n0 n
⎩δCnm = Cnm otherwise.

We see that

GM⊕ ∑︂
n
(︁ )︁
Tn (ϕ, λ) = Pnm (sin ϕ) δCnm cos mλ + Snm sin mλ .
R
m=0

We obtain
x (︃ )︃2 ∑︂
n (︂
1 GM⊕ 2 2
)︂
kn = Tn2 dσ = δCnm + Snm .
4π σ R
m=0

Here, we have exploited the orthonormality of the fully normalised basis kantafunktio
functions Pnm (sin ϕ) cos mλ and Pnm (sin ϕ) sin mλ on the surface of
unit sphere σ. So

í Õ ! ¤. û
274 Statistical methods

The degree variances kn of the disturbing potential can be calculated

directly from the spherical-harmonic coefficients.

The literature offers many alternative notations, such as

kn = σ2n = σTi T .

^ 10.8 Propagation of covariances

The covariance function K of the disturbing potential derived above
can also be used to derive the covariance functions of other quantities.
This works in principle for quantities that can be expressed as linear
functionals of the disturbing potential T (·, ·, R) on the surface of the
spherical Earth, as explained in section 10.2.

^ 10.8.1 Example: upwards continuation of the potential

Let us write the disturbing potential in space T (ϕ, λ, r) as a functional
of the surface disturbing potential T (ϕ, λ, R) = T (·, ·, R). With the
definition of the degree constituents Tn ,

def
∑︂
∞
T (ϕ, λ, R) = Tn (ϕ, λ),
n=2

it holds that
∑︂
∞ (︂ )︂n+1
R
T (ϕ, λ, r) = r Tn (ϕ, λ).
n=2

Symbolically
{︁ }︁
T (ϕ, λ, r) = L T (·, ·, R) .
Here, L is the linear functional
∑︂
∞ (︂ )︂n+1
R
L{f} = r fn ,
n=2

asteosuus- in which the fn are defined according to degree constituent equation

yhtälö

í Õ ! ¤. û
Propagation of covariances
275
3.8, so that on the sea level of a spherical Earth

∑︂
∞
f= fn .
n=2

Symbolically

∑︂
∞
L{f} = Ln fn ,
n=2

in which (︂ )︂n+1
R
Ln = r
is the spectral representation of the functional L.
We may write at a certain point P, location (ϕP , λP , rP ) in space:

∑︂
∞
LP {f} = Ln
P fn,P ,
n=2

in which )︂n+1
R
(︂
Ln
P = r .
P
(︁ )︁
Concretely, for the disturbing potential T ϕP , λP , rP in point P, this
means
{︁ }︁ ∑︂
∞ ∑︂
∞ (︂
R n+1 (︁
)︂
Ln
(︁ )︁ )︁
T ϕP , λP , rP = LP T (·, ·, R) = T
P n,P = r T n ϕ P , λ P .
P
n=2 n=2

Now, the covariance function in space of T is obtained:

{︂ (︁ )︁}︂
K rP , rQ , ψPQ = MP′ T ϕP , λP , rP T ϕQ(P) , λQ(P) , rQ =
(︁ )︁ )︁ (︁
{︂ {︁ }︁ {︁ }︁}︂
′
= MP LP T (·, ·, R) LQ(P) T (·, ·, R) =
{︃∑︂∞ ∑︂∞ (︂ )︂}︃
′ n n′
= MP (LP Tn,P ) LQ Tn ′ ,Q(P) =
n=2 n ′ =2
∑︂∞ ∑︂
∞
{︁ }︁
n′ ′
= Ln
P LQ MP Tn,P Tn ′ ,Q(P) .
n=2 n ′ =2

í Õ ! ¤. û
276 Statistical methods

According to equation 10.11:

{︁ }︁
MP′ Tn,P , Tn ′ Q(P) =
w 2πw + π/2 w 2π
1
= 2 T T ′ dλ cos ϕP dϕP dαPQ .
8π 0 − π/2 0 n,P n ,Q(P) P

A trial expansion is
{︁ }︁
MP′ Tn,P , Tn ′ ,Q(P) = ˜︁
kn Pn (sin ψPQ ),

the coefficients of which are according to equation 10.14:

wπ {︁ }︁
2n + 1
kn =
˜︁ MP′ Tn,P , Tn ′ ,Q(P) Pn (cosψPQ ) sin ψPQ dψPQ .
2 0

Substitution yields, with the abbrevations ϕ = ϕP , λ = λP , ψ =

ψPQ , α = αPQ :
{︁ }︁
MP′ Tn,P ,Tn ′ ,Q(P)
⏟ ⏞⏞ ⏟
w πw 2πw + π/2 w 2π
2n + 1 1
kn = T T ′ dλ cos ϕ dϕ dα Pn (cosψ) sin ψ dψ =
2 8π2 0 0 − π/2 0 n,P n ,Q(P)
˜︁

I
w + π/2 w 2π ⏟ w
w ⏞⏞ ⏟
2n + 1 2π π
= Tn,P Tn ′ ,Q(P) Pn (cosψ) sin ψ dψ dα dλ cos ϕ dϕ.
16π2 − π/2 0 0 0

asteosuus- The integral, using degree constituent equation 3.8:

yhtälö
w 2πw π
I= Tn ′ ,Q(P) Pn (cosψPQ ) sin ψPQ dψPQ dαPQ =
0 0
w
4π
= Tn ′ ,Q(P) Pn (cosψPQ )dσQ = T ′ ,
σ 2n ′ + 1 n ,P
so
w + π/2 w 2π
2n + 1 4π
kn =
˜︁ Tn,P Tn ′ ,P dλP cos ϕP dϕP =
16π2 2n ′ + 1 − π/2 0
w
2n + 1 1 2n + 1
= · T T ′ dσ = M {Tn Tn ′ } .
2n ′ + 1 4π σ n n 2n ′ + 1

í Õ ! ¤. û
Propagation of covariances
277
Using orthogonality yields
⎧ {︁ }︁
{︁ }︁ ⎨M Tn2 = kn Pn cos ψPQ if n = n ′ ,
(︁ )︁
′
MP Tn,P Tn ′ ,Q(P) =
⎩0 if n ̸= n ′ ,
the harmonic components of the surface covariance function, equation
10.13.
This should not surprise us: if the spatial covariance function is
isotropic, it must have the general form
)︁ ∑︂
∞
Krn rP , rQ Kψ
(︁ (︁ )︁ (︁ )︁
K rP , rQ , ψPQ = n ψPQ ,
n=2
(︁ )︁
and Kψn ψPQ must have the same form as in equation 10.13:

Kψ
(︁ )︁
n ψPQ = kn Pn (cos ψPQ ).
(︁ )︁
When Krn rP , rQ = 1, i. e., at sea level, the coefficients kn are those
given by equation 10.14.
Thus we obtain9
)︁ ∑︂
∞
Ln n
(︁ (︁ )︁
K rP , rQ , ψPQ = P LQ kn Pn cos ψPQ =
n=2
∑︂
∞ (︂
R n+1 R n+1
)︂ (︂ )︂ (︁ )︁
= rP rQ k n Pn cos ψPQ =
n=2
∑︂∞ (︃
R2
)︃n+1
(︁ )︁
= r r kn Pn cos ψPQ . (10.15)
P Q
n=2

Here we have expressed the covariance function of the disturbing

potential in space T (ϕ, λ, r) into an expansion into the degree variances astevarianssi
kn of the corresponding sea-level disturbing potential T (ϕ, λ, R), by
applying propagation of covariances on expansion 10.13 of function K. kovarianssien
Thus we have obtained the three-dimensional covariance function for the kasautuminen
disturbing potential, needed for example in mountainous countries and
in air and space applications.

9 Thisonly works this cleanly because in this case, the operator Ln is of multiplier
(︁ /︁ )︁n+1
type, R r .

í Õ ! ¤. û
278 Statistical methods

^ 10.8.2 Example: the covariance function of gravity anomalies

We know from equation 5.8 that there exists the following relationship
between gravity anomalies and the disturbing potential:
∑︂
∞
n − 1 R n+1
(︂ )︂
∆g = r r Tn ,
n=2

symbolically: ∆g = L∆g {T } for a suitable operator Lg :

∑︂
∞
n − 1 R n+1
(︂ )︂
L∆g {f} = Ln
∆g fn , Ln
∆g = r r .
n=2

Again, at a concrete point P,

{︁ }︁
∆g(ϕP , λP , rP ) = L∆g,P T (·, ·, R) =
∑︂∞ ∑︂
∞
n − 1 R n+1
(︂ )︂
n
= L∆g,P Tn,P = r r Tn,P .
P P
n=2 n=2

We can show in the same way as above that

{︁ }︁ {︁ }︁
Cov ∆gP , ∆gQ = MP′ ∆gP ∆gQ(P) =
∑︂
∞
{︁ }︁
′
= Ln n
∆g,P L∆g,Q MP Tn,P Tn,Q(P) =
n=2
∑︂
∞
n − 1 R n+1 n − 1 R n+1
(︂ )︂ (︂ )︂ (︁ )︁
= rP rP rQ rQ k n Pn cos ψPQ =
n=2
∑︂
∞ (︃
R2
)︃n+2 (︂
n−1 2
)︂ (︁ )︁
= rP rQ kn Pn cos ψPQ .
R
n=2

Often, we write
)︁ def ′ {︁ }︁ ∑︂
∞ (︃ )︃n+2
(︁ R2 (︁ )︁
C ψPQ , rP , rQ = MP ∆gP ∆gQ(P) = rP rQ cn Pn cos ψPQ ,
n=2

astevarianssi in which the degree variances of gravity anomalies are

)︂2
n−1
(︂
cn = kn .
R

í Õ ! ¤. û
 Global covariance functions
279
Similarly we also calculate the “mixed covariances” between disturbing
potential and gravity anomaly:
{︁ }︁
Cov T P , ∆gQ =
{︁ }︁ ∑︂
∞
{︁ }︁
= MP′ TP ∆gQ(P) = Ln Ln
P ∆g,Q MP
′
T n,P T n,Q(P) =
n=2
∑︂
∞ (︂
R
)︂n+1
n − 1 R n+1
(︂ )︂ (︁ )︁
= rP rQ rQ k n Pn cos ψPQ =
n=2
∑︂
∞
n−1
(︃
R2
)︃n+1
(︁ )︁
= r r r k n Pn cos ψPQ .
Q P Q
n=2

All these are examples of propagation of covariances, when applied to a kovarianssien

series expansion: kasautuminen

{︂ {︁ }︁ {︁ }︁}︂ ∑︂ {︁ }︁
′
Cov L1 T P , L2 T Q = Ln Ln
1,P 2,Q MP T n,P T n,Q(P) =
n
∑︂
Ln n
(︁ )︁
= 1,P L2,Q kn Pn cos ψPQ ,
n

for arbitrary linear functionals

{︁ }︁ ∑︂ n {︁ }︁ ∑︂ n
L1 TP = L1,P Tn,P , L2 TQ = L2,Q Tn,Q ,
n n
(︁ )︁ (︁ )︁
in which Tn,P = Tn ϕP , λP and Tn,Q = Tn ϕQ , λQ are the degree
constituents of the disturbing potential on the Earth’s surface. The
problem, in each case, is identifying the spectral form of this linear
functional. This is done by expanding the quantity concerned into Tn ,
and lifting the coefficient found from the equation. These coefficients
are indicated above by red and blue colouring.

^ 10.9 Global covariance functions

Empirical covariance functions have been calculated often. There have
been only a few empirical covariance functions for the whole Earth.

í Õ ! ¤. û
280 Statistical methods

They are commonly given in the form of a degree variance formula. The
best known is the rule observed by William Kaula (Rapp, 1989):10
2n + 1
kn = α .
n4
By writing
n−1 2
(︂
)︂
cn = kn ,
R
astevarianssi in which cn are the degree variances of gravity anomalies, we obtain

2n + 1 n − 1 2 2α
(︂ )︂
cn = α ≈ .
n 4 R nR2
Here, α is a planet specific constant, according to Kaula’s estimate
/︁ )︁2
α = 10−10 GM⊕ a⊕ .
(︁

The Kaula rule does not hold very precisely. It applies, by the way,
fairly well for the gravity field of Mars, of course with a different
constant (Yuan et al., 2001).
Another well-known rule is the Tscherning–Rapp equation (Tschern-
ing and Rapp, 1974):
A (n − 1) n−1 2
(︂ )︂
cn = = kn .
(n − 2) (n + B) R
The constants are, according to the authors, A = 425.28 mGal2 and B =
24 (exactly). As a technical detail, one usually chooses R = RB = 0.999R,
the radius of a Bjerhammar11 sphere inside the Earth (R is the Earth’s
mean radius). The form of the equation is chosen so the covariance
functions of various quantities will be closed expressions.

^ 10.10 Collocation and the spectral viewpoint

pienimmän The calculations in least-squares collocation can also be executed effi-
neliösumman
kollokaatio 10 William M. Kaula (1926–2000) was an American geophysicist and space geodesist
who studied the determination of the Earth’s gravity field by means of satellite
geodesy.
11 Arne Bjerhammar (1917–2011) was an eminent Swedish geodesist.

í Õ ! ¤. û
 Collocation and the spectral viewpoint
281
Kaula EGM96
108 EGM2008
Tscherning–Rapp
GOCE, Gatti et al. (2014)
106 EGM96 error variances
EGM2008 error variances
104
)︁

GOCE error variances

m4/s4
(︁

102
Degree variance kn

100

10−2

10−4

10−6

0 50 100 150 200 250 300 350 400

Harmonic degree n

Figure 10.5. Global covariance functions as degree variances. The GOCE model
^ cuts off at degree 280.

ciently by way of FFT. For this one should study the symmetries present
in the geometry, especially the rotational symmetry, which exists, for
example, in the direction of longitude on the whole Earth: nothing
changes when we turn the whole Earth by a certain angle θ around her
axis of rotation: for all longitudes, what happens is λ ↦→ λ + θ.
In the following we discuss a simplified example in one dimension.
Let observations ℓi = gi + ni of a field g(ψ), ψ ∈ [0, 2π) be given on
def
the edge of a circle, in points ψi = 2π i N , i = 0, 1, 2, . . . , N − 1. Let
/︁

us assume that also the results of the calculation, estimates fˆ︁i of the
result function f(ψ), are desired at the same points. Then we have

í Õ ! ¤. û
282 Statistical methods

equation 10.8:
ˆ︁f = Cfg [Cgg + Dgg ]−1 g + n ,
(︁ )︁

with
[︂ ]︂ (︁ )︁
Cfg = Cfg f(ψi ), g(ψj ) = Cfg (ψi , ψj ),
[︂ ]︂ij (︁ )︁
Cgg = Cgg g(ψi ), g(ψj ) = Cgg (ψi , ψj ),
[︂ ]︂ij (︁ )︁
Dgg = Dgg g(ψi ), g(ψj ) = Dgg (ψi , ψj ).
ij

If the physics of the whole situation, including the physics of the

measurement process, is rotationally symmetric, we must have
[︂ ]︂ {︁ (︁ )︁}︁
Cfg = M⃝ f(ψn )g ψj(n) =
i,j(i)

1 ∑︂
N−1
(︁ )︁ def [︂ ]︂
= f(ψn )g ψj(n) = Cfg ,
N k
n=0

with j (i) = (i + k) mod N. Here, the operator M⃝ is the “circle average”

of a function,
def 1
∑︂
N−1
M⃝ {h} = h(ψn ),
N
n=0

which, like the geographic average in section 10.4, replaces the statistical
average.
In the same way we obtain
[︂ ]︂ {︁ (︁ )︁}︁
Cgg = M⃝ g(ψn )g ψj(n) =
i,j(i)

1 ∑︂
N−1
(︁ )︁ def [︂ ]︂
= g(ψn )g ψj(n) = Cgg .
N k
n=0

Now Cfg , Cgg are functions of only k, and we may write them
[︂ ]︂ [︂ ]︂
Cfg = Cfg (ψi , ψj ) = Cfg (∆ψk ) = Cfg ,
[︂ ]︂ij [︂ k
]︂
Cgg = Cgg (ψi , ψj ) = Cgg (∆ψk ) = Cgg ,
ij k

í Õ ! ¤. û
 Collocation and the spectral viewpoint
283
2 1
i
0
N−1
N−2
j
∆ψk
ψj

ψi

^ Figure 10.6. Circular geometry.

def
in which ∆ψk = (ψj − ψi ) mod 2π and k = (j − i) mod N.
Furthermore
[︂ ]︂ [︂ ]︂ {︁ }︁
Dgg = Dgg (ψi , ψj ) = Dgg (∆ψk ) = Dgg = E ni nj(i) ,
ij k

the traditional statistical variance of the observation noise. Also because

generally the observations do not correlate with each other, we have12

Dgg = σ2 IN ,

σ2 (the variance of observations, assumed equal for all) times the N × N

sized unit matrix.
Matrices of this form are called Toeplitz circulant matrices.13 Thanks Toeplitz-
to this property, equation 10.8 is a string of convolutions. sirkulantti
matriisi

12 Infact, the unit or identity matrix is also known as the Kronecker delta, and as a
Toeplitz matrix may be interpreted as a discrete version of the Dirac delta function.
Its discrete Fourier transform is “white”:
1
F{IN } = ,
N
with the same power for all frequencies.
13 OttoToeplitz (1881–1940) was a German Jewish mathematician who contributed to
functional analysis.

í Õ ! ¤. û
284 Statistical methods

Without proof we present that the spectral version of equation 10.8

looks like this:
{︁ }︁ F{Cfg } {︁ }︁ F{Cfg } {︁ }︁
F ˆ︁f = ·F g+n = · F g + n .
F{Cgg } + F{Dgg } 2
F{Cgg } + σ N
/︁

This is an easy and rapid way to calculate the solution using FFT. If for a
{︁ }︁
suitable operator L we have f = L g , the equation simplifies as follows:

{︁ }︁ F{L} · F{Cgg } {︁ }︁
F ˆ︁f = 2/︁
·F g+n .
F{Cgg } + σ N

In the limit in which the observations are exact, σ2 = 0 and thus n = 0,

it holds that
{︁ }︁ {︁ }︁ {︁ }︁
F ˆ︁f = F{L} · F g ⇐⇒ ˆ︁f = L g .
For example, if g are gravity anomalies and f are values of the disturbing
potential, then14
R
F{L} = .
n−1
The approach is called Fast Collocation, for example Bottoni and Barzaghi
(1993). Of course it is used in two dimensions on the Earth’s surface,
although our example is one-dimensional. As always, it requires that
the observations are given on a grid, and in this case also that the
precision of the material is homogeneous — the same everywhere — over
the area. This requirement is hardly ever precisely fulfilled.

^ Self-test questions
1. What is the difference between signal and noise?
(a) Signal is not a random stochastic process, whereas noise is.

14 Inreal computation it is not so simple. . . the harmonic degree number n, which

refers to global spherical geometry, must first be converted to the Fourier wave number
expressed on the computational grid used.

í Õ ! ¤. û
Self-test questions
285
(b) Signal is a stochastic process that we are interested in and
wish to estimate, while noise is a stochastic process that we
are not interested in and that we would like to filter out.
(c) Signal is a stochastic process with a greater variance and is
therefore more easily detectable than noise.
(d) Signal is a property of a real-world system, while noise is a
property of an observation instrument or method.
2. What is a functional?
(a) A mapping from a function space to a set of numbers, for
example the real numbers.
(b) A random-valued function.
(c) A functional associates with every (well-behaved) function
defined on some domain, a number.
(d) A function of a vectorial argument.
3. What is a linear functional?
(a) A linear functional associates a number L{f} with any linear
function f(x) = a + bx defined on some domain
(b) If, for any functions f and g it holds for a functional L that

L{af + bg} = aL{f} + bL{g}

for any real values a, b, then L is a linear functional.

(c) A linear functional associates with any (well-behaved) func-
tion defined on some domain, a linear expression L{f} =
a + bx.
4. The statistical behaviour of a stochastic process defined on the
Earth’s surface is the same independently of where on Earth you
are. This property is called isotropy | ergodicity | homogeneity
| stationarity.
5. The statistical behaviour of a stochastic process of time is the same

í Õ ! ¤. û
286 Statistical methods

independently of where on the time axis you are. This property is

called isotropy | ergodicity | homogeneity | stationarity.
6. Why, in the study of the Earth’s gravity field, does one use as the
average of quantities the geographical average rather than the
statistical average?
7. Which two different kinds of covariance functions are used for
gravity anomalies on the Earth’s surface? Give the formulas and
name the free parameters.
astevarianssi 8. Explain degree variances. What is the difference between degree
variances kn and cn ?
Toeplitz- 9. What does Kaula’s rule express?
sirkulantti
matriisi 10. What is a Toeplitz circulant matrix?

^ Exercise 10 – 1: Variance of prediction

ennustus- The equation for the variance of prediction at a point P is
varianssi
ΣPP = CPP − CPi (Cij + Dij )−1 CjP ,

in which the observation points are i = 1, . . . , N. Assume there is only

one observation point, point P. Then

ΣPP = CPP − CPP (CPP + DPP )−1 CPP .

Show that, if Dij ̸= 0 but however Dij ≪ Cij ,

ΣPP ≈ DPP .

^ Exercise 10 – 2: Hirvonen’s covariance equation and

prediction
Hirvonen’s covariance equation is
C0
C(s) = (︁ /︁ )︁2 ,
1+ s d

í Õ ! ¤. û
 Exercise 10 – 3: Predicting gravity anomalies
287
with the Ohio parameters C0 = 337 mGal2 and d = 40 km (Heiskanen
and Moritz, 1967, equation 7-9). The equation gives the covariance
between the gravity anomalies at two points P and Q
(︁ )︁ {︁ }︁
C sPQ = Cov ∆gP , ∆gQ .

sPQ is the linear distance between the points.

{︁ }︁ {︁ }︁
1. Calculate Var ∆gP and Var ∆gQ . Remember that, according
{︁ }︁ {︁ }︁
to the definition, Var x = Cov x, x !
{︁ }︁
2. Calculate Cov ∆gP , ∆gQ if sPQ = 20 km.
3. Calculate the correlation
{︁ }︁
{︁ }︁ def Cov ∆gP , ∆gQ
Corr ∆gP , ∆gQ = √︂ {︁ }︁ {︁ }︁ .
Var ∆gP Var ∆gQ

4. Assume now that we only have a measurement in point P. What is

the “variance of prediction” of the gravity anomaly in point Q ennustus-
which is at a distance sPQ = 10 km from the (precisely!) given varianssi
anomaly in point P? Apply equation 10.10 as follows:

σ2QQ = CQQ − CQP C−1

PP CPQ .

5. And item 4 if the distance is sPQ = 80 km?

^ Exercise 10 – 3: Predicting gravity anomalies

Let the measured gravity anomalies ℓ1 = ∆g1 + n1 and ℓ2 = ∆g2 + n2 be
given at two points 1 and 2. The distance between the points is 80 km
and between them, at the same distance of 40 km from both, is located
point P. Compute the gravity anomaly of point P, ∆gP by means of
prediction. The prediction equation is

ˆ︂ P = CPi (Cij + Dij )−1 ℓj ,

∆g

í Õ ! ¤. û
288 Statistical methods

where ℓj = ∆gj + nj is the (abstract) vector of gravity anomaly observa-

tions, [︄ {︁ }︁ {︁ }︁ ]︄
Var ∆gi Cov ∆gi , ∆gj
Cij = {︁ }︁ {︁ }︁
Cov ∆gi , ∆gj Var ∆gj
is the signal variance matrix of the vector ∆gi , and
[︂ {︁ }︁ {︁ }︁ ]︂
CPi = Cov ∆gP , ∆g1 Cov ∆gP , ∆g2

is the signal covariance matrix between ∆gP and ∆gi . Dij is the variance
matrix of the observation random uncertainty or noise ni , i = 1, 2:
[︄ {︁ }︁ {︁ }︁ ]︄
Var ni Cov ni , nj
Dij = {︁ }︁ {︁ }︁ .
Cov ni , nj Var nj

1. Compute the matrix Cij , assuming again Hirvonen’s covariance

formula (previous exercise) and a parameter value of d = 40 km.
2. Compute CPi .
3. Compute ∆gˆ︂ P expressed in the observed values ℓ1 and ℓ2 . Assume
Dij = 0 (and thus ni = 0). Inverting the Cij matrix is possible by
hand, but just use Matlab or octave.
ennustus- 4. Compute the variance of prediction (note CjP = CTPi ) using
varianssi
σ2PP = CPP − CPi C−1
ij CjP .

^ Exercise 10 – 4: Predicting gravity anomalies (2)

Let us again have points 1 and 2 with measured gravity anomalies
ℓ1 = ∆g1 and ℓ2 = ∆g2 . Now however the points 1, 2 and P are in a
triangular configuration, with a right angle at point P, and the distances
from P to points 1 and 2 still 40 km. The distance between points 1 and
√
2 is now only 40 2 km.
ˆ︂ P and σ2 .
1. Compute Cij , CPi , ∆g PP

2. Compare the result with the previous one. Conclusion?

í Õ ! ¤. û
 Exercise 10 – 5: Propagation of covariances
289
^ Exercise 10 – 5: Propagation of covariances
Given the covariance function 10.15 of the disturbing potential

{︁ }︁ ∑︂
∞ (︃ )︃n+1
R2 (︁ )︁
Cov T P , T Q = r r kn Pn cos ψPQ .
P Q
n=2

1. Calculate the covariance function of the gravity disturbance δg

(equation 5.4). Hint: write first an expansion of the form
∑︂
∞
δg = Ln
δg Tn
n=2

in order to find the expression for the coefficient Ln

δg . After this

{︁ }︁ ∑︂
∞
Ln n
(︁ )︁
Cov δgP , δgQ = δg,P Lδg,Q kn Pn cos ψPQ .
n=2

2. Compute the covariance function of the gravity-gradient disturbance

∂2 ∂
T = − δg,
∂r2 ∂r
i. e., the vertical gradient of the gravity disturbance.

^ Exercise 10 – 6: Kaula’s rule for gravity gradients

For the disturbing potential
∑︂
∞ (︂ )︂n+1
R
T (ϕ, λ, r) = r Tn (ϕ, λ) (10.16)
n=2

or on the Earth’s surface (r = R)

∑︂
∞
T (ϕ, λ, R) = Tn (ϕ, λ)
n=2

Kaula’s rule applies, with the degree variances astevarianssi

2n + 1
kn = α .
n4

í Õ ! ¤. û
290 Statistical methods

varianssien From these one can derive, using propagation of variances, the degree
kasautuminen variances of gravity anomalies

∑︂
∞ ∑︂
∞ (︂
n−1
)︂
∆g = Ln
g Tn = Tn
R
n=2 n=2

as follows:
n−1 2 2α
(︁ n )︁2 (︂ )︂
cn = L∆g kn = kn ≈ .
R nR2
By differentiating the above expansion 10.16 for the disturbing potential

∑︂
∞ (︂ )︂n+1
R
T (r, φ, λ) = r Tn (φ, λ)
n=2

we obtain the second derivative

∂2 T ∑︂
∞
(n + 1) (n + 2) (︂ R )︂n+1
= r Tn ,
∂r2 r2
n=2

the connection between the disturbing potential and the gravity gradient
in the spectral domain.
On the Earth’s surface r = R, or

∑︂
∞
(n + 1) (n + 2) def ∑︂ n
∞
⃓
∂2 T ⃓⃓
= Tn = Lgg Tn
∂r2 ⃓r=R R2
n=2 n=2

with
(n + 1) (n + 2)
Ln
gg = .
R2
astevarianssi 1. Derive an (approximate) equation for the degree variances for
the gravity gradient. Designate them with the symbol ggn , in
an analogue fashion as above for the gravity anomaly degree
variances cn :
ggn = ? · kn ≈ ? · n ? .

2. Conclusion?

í Õ ! ¤. û
 Exercise 10 – 7: Underground mass points
291
^ Exercise 10 – 7: Underground mass points
1. If a mass point is placed inside the Earth at a depth D beneath
an observation point P, what then is the correlation length s of
the gravitational field it causes on the Earth’s surface, for which
C(s) = 12 C0 ?
2. Thus, if we wish to construct a model made of mass points, where
under each observation point ∆gP there is one mass point, how
deep should we place them if the correlation length d is given?

í Õ ! ¤. û
^ Gravimetric measurement devices

11
^ 11.1 History
The first measurement device ever built based on a pendulum was a
clock. The pendulum equation,
√︃
ℓ
P = 2π g ,

tells that the swinging time or period P of a pendulum of a given length

is a constant that depends only on the length ℓ and local gravity g, on
condition that the swings are small. The Dutch Christiaan Huygens1
built in 1657 the first useable pendulum clock based on this (Wikipedia,
Pendulum clock).
When the young French researcher Jean Richer2 visited French
Guyana in 1671 with a pendulum clock, he noticed that the clock
ran clearly slower. The matter was corrected simply by shortening the
pendulum. The cause of the effect could not be the climatic conditions
in the tropics, like the thermal expansion of the pendulum. The right

1 Christiaan Huygens (1629–1695) was a leading Dutch natural scientist and mathe-
matician. Besides inventing the pendulum clock, he also was the first to realise (in
1655) that the planet Saturn has a ring.
2 JeanRicher (1630–1696) was a French astronomer. He is really only remembered for
his pendulum finding.

– 293 –
294 Gravimetric measurement devices

^ Figure 11.1. Jean Richer’s report.

explanation was that in the tropics, gravity g is weaker than in Europe.

After his return to France in 1673, Richer had to make his pendulum
longer again. The observation is described in just one paragraph in his
report Observations astronomiques et physiques faites en l’isle de Caïenne,
(Richer, 1731, pages 87–88).
This is how the pendulum gravimeter was invented. Later, much
more precise special devices were built, for example Kater’s3 reversion
pendulum, and the four-pendulum Von Sterneck4 device, which was
also used in Finland in the 1920s and 1930s (Pesonen, 1930; Hirvonen,
1937). We must also mention the submarine measurements, including
in the Java Sea by the Dutch Vening Meinesz, in which it was observed
syvänmeren that above the trenches in the ocean floor there is a notable shortage of
hauta gravity, and that they thus are in a state of strong isostatic disequilibrium
(Vening Meinesz, 1928).
Pendulum gravimeters are however too hard to operate and too slow
for high-productivity gravimetric observations. For that purpose, the
spring gravimeter was developed, see section 11.2.

3 Henry Kater FRS FRAS (1777–1835) was an English physicist.

4 RobertFreiherr (baron) Daublebsky von Sterneck (1839–1910) was a major general in
the Austro-Hungarian army and a geophysicist, astronomer and geodesist.

í Õ ! ¤. û
 The relative or spring gravimeter
295

Figure 11.2. Autograv™ CG-5 spring gravimeter from Scintrex. Image Mon-
^ niaux (2011).

A pendulum gravimeter is in principle an absolute measurement

device, as gravity is obtained directly as an acceleration. There are,
however, systematic effects associated with the suspension, or fulcrum,
of the pendulum, because of which one cannot trust in the absoluteness
of the measurement after all. One attempted solution is the very long wire
pendulum, for example Hytönen (1972). However, nowadays absolute
measurements are made with ballistic gravimeters, see section 11.3. It
has been observed that the older measurements made with pendulum
apparatus in the so-called Potsdam system are systematically 14 mGal
too large. . .

^ 11.2 The relative or spring gravimeter

A spring gravimeter is at its simplest the same as a spring balance.

í Õ ! ¤. û
296 Gravimetric measurement devices

In a linear spring balance the equation of motion of the test mass is

(︃ )︃
d2 ℓ
m − g = −k (ℓ − ℓ0 ) ,
dt2

where m is the test mass, g the local (to be measured) gravity, and k the
spring constant. The quantity ℓ0 is the “rest length” of the spring; the
length it would have if no external forces were acting on it. ℓ is the true,
instantaneous length of the string.
The equilibrium between the spring force and gravity is

d2 ℓ (︁ )︁
= 0 =⇒ mg = k (ℓ − ℓ0 ) = k ℓ − ℓ0 , (11.1)
dt2
in which ℓ is the mean length of the spring during the oscillation, and
also the equilibrium length in the absence of oscillations.
When the test mass is disturbed, it starts oscillating about its equi-
värähtely- librium position. The oscillation equation, obtained by summing the
yhtälö above two equations, is

d2 (︁ )︁ k (︁ )︁
ℓ − ℓ = − m ℓ − ℓ .
dt2
The period is
√︃ √︃
m ℓ − ℓ0 δℓ
√︂
P = 2π = 2π g = 2π g, (11.2)
k

in which δℓ = ℓ − ℓ0 denotes the difference between the equilibrium

length and the length in the state of rest: the lengthening of the spring by
gravity.
The sensitivity of the instrument is obtained by differentiating equation
11.1 in the form
(︁ )︁
mg = k ℓ − ℓ0 = k δℓ
with the result
dℓ d (δℓ) m P2
= = = 2. (11.3)
dg dg k 4π

í Õ ! ¤. û
 The relative or spring gravimeter
297
Substitution of, for example, δℓ = 5 cm and g = 10 m/s2 into equation
11.2 yields P = 0.44 s. One milligal of change in gravity g produces,
according to equation 11.3, a lengthening of only 5 · 10−8 m (check).
Clearly then, the sensor observing or compensating this displacement
must be extremely sensitive!

^ 11.2.1 Astatisation
An astatised gravimeter uses a different measurement geometry. The
LaCoste-Romberg gravimeter, which long enjoyed great popularity,
serves as an example. Inside it, the test mass is at the end of a lever
beam, see figure 11.3. Two torques operate on the beam, which are in
equilibrium. The torque caused by the spring is
(︁ )︁
τs = k ℓ − ℓ0 b sin β,

in which ℓ is the spring’s true, stretched equilibrium length, and ℓ0 the

theoretical or state-of-rest length without loading.
According to the sine rule

ℓ sin β = c sin(90◦ + ϵ) = c cos ϵ,

substitution of which in the previous equation yields

(︁ )︁ bc
τs = k ℓ − ℓ0 cos ϵ.
ℓ

The force of gravity pulling at the mass is mg, and the corresponding
torque
τg = mgp cos ϵ.
Between these there has to be equilibrium:
(︁ )︁ bc
τg − τs = mgp cos ϵ − k ℓ − ℓ0 cos ϵ = 0
ℓ
or
(︁ )︁
mgpℓ − kbc ℓ − ℓ0 = 0. (11.4)

í Õ ! ¤. û
298 Gravimetric measurement devices

Reality

Working range

Force
Spring, length ℓ Hooke’s law

c
(︁ )︁
k ℓ − ℓ0 sin β
Length
β
ϵ
b Test mass beam
p

mg

Figure 11.3. Operating principle of spring gravimeter. On the right, how to

^ build a “zero-length spring”.

By differentiation

mpℓ dg + mgp dℓ − kbc dℓ = 0,

from which we obtain, by substituting equation 11.4, a sensitivity

equation:

dℓ mpℓ mpℓ ℓ ℓ − ℓ0
dg
=−
mgp − kbc
=− /︂(︁ )︁ = g ℓ0 .
mgp − mgp ℓ ℓ − ℓ0

From this we see that the sensitivity can be driven up arbitrarily by

choosing ℓ0 as short as possible, almost zero — a so-called zero-length
spring solution (Wikipedia, Zero-length springs).
tasaus Of course, levelling the instrument, using its bull’s-eye level and three
rasiatasain footscrews, is critical.

í Õ ! ¤. û
 The relative or spring gravimeter
299
For example, assuming ℓ = 5 cm, ℓ0 = 0.1 cm, g = 10 m/s2 gives

dℓ
= 2.5 · 10−6 m/mGal,
dg
a 50 times5 better(︁ result)︁/︂
than earlier! The improvement or astatisation
ratio is precisely ℓ − ℓ0 ℓ0 .
This is the operating principle of an astatised gravimeter, like the
LaCoste-Romberg.6

^ 11.2.2 Period of oscillation

There is another way to look at this: if the instrument is not in equilib-
rium, the lever beam will slowly oscillate about the equilibrium position.
We start from equation 11.4:
(︁ )︁
mgpℓ − kbc ℓ − ℓ0 = 0, (11.5)

but for a state of disequilibrium. Then, the test mass will be undergoing
an acceleration a, positive downwards, and we have

m (g − a) pℓ − kbc (ℓ − ℓ0 ) = 0,

where, instead of the equilibrium spring length ℓ, the instantaneous

length ℓ appears. Subtracting the above two equations yields
(︁ )︁ (︁ )︁
mgp ℓ − ℓ + mapℓ − kbc ℓ − ℓ = 0.

We use equation 11.5 again to eliminate kbc, yielding

(︁ )︁ ℓ (︁ )︁
mgp ℓ − ℓ + mapℓ − mgp ℓ − ℓ = 0.
ℓ − ℓ0

5 Forcomparability we should still multiply by p b sin β , if we measure the position

/︁

of the test mass.

6 LucienLaCoste (1908–1995) was an American physicist and metrologist, who, as
an undergraduate, together with his physics professor Arnold Romberg (1882–1974)
discovered the principle of the astatised gravimeter and zero-length spring.

í Õ ! ¤. û
300 Gravimetric measurement devices

Rearranging terms gives

ℓ0 (︁ )︁
mapℓ = mgp ℓ−ℓ
ℓ − ℓ0
or
g ℓ0 (︁ )︁
a=− ℓ−ℓ .
ℓ ℓ − ℓ0
(︁ )︁/︁
Here we see again the “astatisation ratio” ℓ − ℓ0 ℓ0 appear, which for
a zero-length spring (ℓ0 ≈ 0) is very large.
Now the string length disequilibrium ℓ − ℓ is connected with the
vertical displacement z (reckoned downwards) of the test mass, as
follows:
(︁ )︁ p
z= ℓ−ℓ .
b sin β
With this, we obtain
d2 g ℓ0 b sin β
a= z=− z.
dt 2 ℓ ℓ − ℓ0 p
This is again an oscillation equation in z. The oscillation period is
√︄
ℓ p ℓ − ℓ0
P = 2π g .
b sin β ℓ0

For the same values as above, ℓ0 = 0.1 cm, ℓ = 5 cm ≈ ℓ, g = 10 m/s2 , and

p b sin β = 2, we find
/︁

P = 4.4 s.
What this long oscillation period also means is that the instrument is
less sensitive to high-frequency vibrations, for example from passing
traffic or microseismicity. This is a significant operational advantage.

^ 11.2.3 Practicalities of measurement

An ordinary spring gravimeter is based on elasticity. Because there is
no material that is perfectly elastic, but it is always plastic7 as well, the

7 Plastic deformation in a metal crystal is mediated by crystal-lattice defects called

dislocations. As dislocations travel through the crystal lattice under load, the properties

í Õ ! ¤. û
 The relative or spring gravimeter
301
δ (ε)

)
F (ε
α F (ε) cos (α + δ + ε)

−ε
mg

mg

Figure 11.4. The idea of astatisation. The elastic force of an ordinary spring
grows steeply with extension (left), whereas the weight of the
test mass is constant. The lever beam and diagonal arrangement
(right) causes the part of the force of the spring in the direction
of motion of the lever (red) to diminish with extension, while
the spring force itself grows similarly with extension. This near-
cancellation boosts sensitivity. The spring used is a zero-length
^ spring.

gravimeter itself changes during the measurement process. This change

is called drift. The drift is managed in practical measurements by the käynti
following measures:
◦ We measure along lines starting from a known point and ending
on a known point, producing a closing error. A line is traversed as
rapidly as possible. The closing error is eliminated by adjusting
the values obtained from the measurement in proportion to their
times of measurement.

of the metal change, which may eventually result in metal fatigue, a known problem
for example in aviation. Wikipedia, Dislocation. The art of making metals stronger
by inhibiting the motion of dislocations, for example by adding carbon to iron to
form steel, forms a large part of practical metallurgy. Wikipedia, Strengthening
mechanisms of materials.

í Õ ! ¤. û
302 Gravimetric measurement devices

◦ The gravimeter is transported carefully without bumping it.

arretointi ◦ We remember to always arrest (clamp down the lever beam) during
transport!
◦ Because the elastic properties of the spring and the instrument
geometry both depend on temperature, precision gravimeters are
always thermostated.
A sea gravimeter differs from an ordinary (land) gravimeter in hav-
vaimennus ing powerful damping. This applies also for an airborne gravimeter.
Both types are mounted on a stabilised platform, keeping the axis of
measurement along the local vertical in spite of vehicle motion.

^ 11.3 The absolute or ballistic gravimeter

The ballistic or absolute gravimeter is a return to roots, the definition of
gravity: it measures directly the acceleration of free fall. The instrument is
made up of a vacuum tube, inside of which an object, a prism reflecting
light, falls freely. See figure 11.5.
Here we describe briefly the JILA gravimeter, built at the University
of Colorado at Boulder by Jim Faller,8 of which the Finnish Geodetic
Institute has acquired two. Figure 11.6 shows the newer model, FG5,
built by the same group. In Finland this instrument, serial number 221,
has served as the national standard for the acceleration of free fall. It
was upgraded to model FG5X in 2012.
During the fall of the prism, a “cage” with a window in the bottom
moves along with the prism inside it without touching it. The purpose
of the cage is to prevent the last remaining traces of air from affecting
the motion of the prism. Approaching the bottom, the cage, which

8 James E. Faller (born 1934) is an American physicist, metrologist, geodesist, and

student of gravitation. He proposed the installation of laser retroreflectors on the
lunar surface in the context of the Apollo project, in order to measure the distance to
the Moon — LLR, lunar laser ranging.

í Õ ! ¤. û
 The absolute or ballistic gravimeter
303
Vacuum pump system
Cage transporter

Prism protective cage

Falling prism

“Superspring”

g
Reference prism

Semi-transparent mirror

Laser
Mirror

Interference observation device

^ Figure 11.5. Operating principle of a ballistic absolute gravimeter.

moves along a rail under computer control, decelerates, and the prism
lands relatively softly on its base. After that, the cage moves back to the
top of the tube and a new measurement cycle starts.
A laser interferometer measures the locations of the prism during its
fall. The measurements are repeated thousands of times to get good
precision through averaging. Another prism, the reference prism, is sus- vertausprisma
pended in another tube from a very soft spring (actually an electronically
simulated “superspring”) to protect it from microseismicity.
The instrument is designed to achieve the greatest precision possible;
for example, the vibration caused by the drop is controlled by a well-
designed mount. Precisions are of the order of several microgals, similar
to what ordinary LaCoste-Romberg relative gravimeters are capable of.
The instrument is however large and, although transportable, cannot
be called a field instrument. Of late, development has gone in the

í Õ ! ¤. û
304 Gravimetric measurement devices

Figure 11.6. Absolute gravimeter of type FG5. Figure United States National
^ Oceanic and Atmospheric Administration.

direction of smaller devices, which are essentially better portable.

The motion of a freely falling mass is given by the equation
d2
z = g(z),
dt2
where it is assumed — realistically — that gravity g depends on the
location z within the drop tube, reckoned downwards. If we nevertheless
take g to be constant, we obtain by integration
d
z = v0 + gt,
dt
z = z0 + v0 t + 21 gt2 ,
from which we obtain the observation equations of the measurement
process ⎡ ⎤
[︂ ]︂ z 0
zi = 1 ti 12 t2i · ⎣ v0 ⎦ + ni .
⎢ ⎥

í Õ ! ¤. û
The absolute or ballistic gravimeter
305
Here, the unknowns9 to be estimated are z0 , v0 and g. The quantities
zi are the interferometrically measured vertical locations of the falling
prism, and ni are the measurement errors or “noise”. Determining
precisely the corresponding measurement time or epoch ti reckoned from
the moment of release of the prism is of course essential. The volume
of measurements obtained from each individual drop is substantial.
We write the observation equations in matric form:

ℓ = Ax + n,

in which
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
z1 n1 1 t1 t21 ⎡ ⎤
z0
z2 n2 1 t2 t22 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢
⎥, ⎥,
ℓ=⎢ n=⎢ A=⎢ ⎥, x = ⎣ v0 ⎦ .
⎢ ⎥
.. .. .. .. .. ⎥
. . . . . ⎦
⎢ ⎥ ⎢ ⎥ ⎢
g
⎣ ⎦ ⎣ ⎦ ⎣
zn nn 1 tn t2n

The solution follows from this according to the method of least-squares pienimmän
adjustment, from the normal equations neliösumman
tasoitus
AT Aˆ︁
x = AT ℓ,

giving the solution (estimate)

)︁−1 T
x = AT A
(︁
ˆ︁ A ℓ.

The uncertainty of the estimates is given by the variance matrix

{︁ }︁ )︁−1
x = σ2 AT A
(︁
Var ˆ︁ ,

in which σ is the uncertainty (mean error) of a single observation zi ,

also known as the mean error of unit weight. painoyksikön
keskivirhe
An alternative type of absolute gravimeter throws the prism up (inside
the tube), after which it moves along a symmetric parabolic path. An

9 It
would be easy (exercise!) to add an unknown representing the vertical gradient of
gravity to this.

í Õ ! ¤. û
306 Gravimetric measurement devices

example of such a “rise-and-fall” instrument is the Italian IMGC-02

(d’Agostino et al., 2008). Theoretically, this method would give more
precise results; however, the technical challenges are larger than in the
case of the dropping method. Intercomparisons between instruments
of these two types have helped to identify error sources.
Recently so-called atomic or quantum gravimeters have also been built,
in which the falling of individual atoms is measured interferometrically
(de Angelis et al., 2009).
The idea of the device is that it measures the effect of gravity on the
phase angle of the matter wave of falling atoms. Firstly a so-called
Bose–Einstein condensate is prepared, which is extremely cold and
consists of perhaps a million atoms in identical quantum states, with the
same phase angle like marching soldiers. The condensate is dropped,
and the first laser pulse splits it into two. Half of the atoms10 fall first
slowly, then faster. The other half fall fast at first and then slower. In
order to achieve this, a second laser pulse pair is used that acts like
a mirror, or perhaps a tennis racket. The third and last laser pulse is
for reading out interferometrically the phase difference between the
two merging atomic beams. The interaction between light and atoms is
based on the Raman effect.
As the atoms travel through space-time along two different paths
on which the gravity potentials are different,11 a phase difference is
formed between these which can be measured. Without gravity (dashed
lines) this phase difference would be zero, see figure 11.7, in which the
horizontal axis is time.

10 Thisis a quantum theoretically erroneous statement. The matter wave of each

individual atom splits into two! Wikipedia, Double-slit experiment.
11 In
fact, the spinning of the atom’s phase angle acts like a clock, and the speed at
which time elapses depends on the local geopotential (Vermeer, 1983a).

í Õ ! ¤. û
 Network hierarchy in gravimetry
307
t

Splitting Readout
beam beam

Mirroring
beam

g Without gravity
In gravity field

Figure 11.7. Principle of operation of an atomic gravimeter. The horizontal

^ axis is time.

^ 11.4 Network hierarchy in gravimetry

Network hierarchy is just as important in gravimetry as in measure-
ments of location or height. The procedure has typically been that the
highest measurement order consisted of points measured by absolute
gravimeters — in the old days, this meant pendulum measurements.
Stepwise densification of this network, i. e., measurement of the base
network, was then done with relative or spring gravimeters, like the
lowest-order measurements, gravity mapping surveys. In base net-
work measurement, fast transportation was used, such as aircraft, and
national or regional reference points were often located at airports. vertauspiste
Because pendulum instruments were not genuinely absolute, the old
Potsdam system collected a 14 mGal systematic error: all values were
that much too high. Nowadays we use ballistic free-fall gravimeters
instead, the possible systematics of which are much smaller — but by

í Õ ! ¤. û
308 Gravimetric measurement devices

Figure 11.8. International intercomparison of absolute gravimeters in Walfer-

^ dange, Luxembourg. Image courtesy © Olivier Francis.

no means non-existent, in the order of magnitude of microgals. As

there are no better, i. e., more absolute, instruments than these, the
issue cannot really be resolved. Nevertheless, international instrument
intercomparisons, like the International Intercomparison of Absolute
Gravimeters, are regularly organised and are valuable.
In Finland, regular absolute gravimetric measurements have been
made in Metsähovi, and also in Vaasa (two points), Joensuu (two points),
Kuusamo, Sodankylä, Kevo, and Eurajoki.

^ 11.5 The superconducting gravimeter

This gravimeter type is based on a superconducting metal sphere levi-
tating on a magnetic field. The precise place of the sphere is measured
electronically. Because a superconducting material is impenetrable by a
magnetic field, the sphere will remain forever in the same spot inside
the field: the Meissner effect. Of course the field itself must be constant.
käämi It is generated by superconducting solenoids inside a vessel made of
mu-metal (Wikipedia, Mu-metal), which keeps out the Earth’s magnetic
field.

í Õ ! ¤. û
 The superconducting gravimeter
309
Upper capacitor plate

Middle capac-
itor plate

Levitation coil
Feedback
coil

Lower capacitor plate

Levitation coil

Figure 11.9. Principle of operation of a superconducting gravimeter. Reading

^ out the sphere position is done capacitively.

Superconduction applied for this still demands working at the tem-

perature of liquid helium (He). For this reason the device is not only
expensive, but also requires an expensive laboratory in an environment
where the societal infrastructure allows.
There are more than thirty superconducting gravimeters in the world.
The work is co-ordinated by the IAG service IGETS, the International
Geodynamics and Earth Tide Service. One GWR20-type instrument has
been working since 1994 in Kirkkonummi at the Metsähovi research
station of the then Finnish Geodetic Institute, now the National Land
Survey, Virtanen and Kääriäinen (1995), Virtanen (1998). The instrument
was upgraded in 2014.
The most important property of a superconducting gravimeter is, in
addition to its superior precision,12 its stability, i. e., its absent drift. For käynti

12 Virtanen (2006) reports how the instrument at Metsähovi detected the change in

í Õ ! ¤. û
310 Gravimetric measurement devices

this reason, it is extremely suited to monitoring long-period phenomena,

like the free oscillations of the solid Earth after large earthquakes,13 in
which the whole Earth tolls like a church bell. Thus it is suitable for
measurements that are unsuitable for an ordinary gravimeter because
käynti of its larger drift and poorer sensitivity, and measurements for which a
seismometer is unsuited because the frequencies are too low.
A recent trend is the development of lightweight, “portable”, and
remotely controllable superconducting gravimeters, for example the
GWR iGrav, weighing 30 kg and not consuming any liquid helium at
all. On the other hand it needs over a kilowatt in grid power for its
refrigeration system (GWR Instruments, Inc., iGRAV® Gravity Sensors).
Perhaps this will lead to improvement over the current situation where
the bulk of the instruments are located in Europe and North America.

^ 11.6 Atmospheric influence on gravity measurement

The atmosphere has the following effects on gravity:
◦ Instrumental effects. These are due to the way the gravimeter
is constructed. Putting the instrument in a pressure chamber
makes these effects go away. In practice it is easier to calibrate
the instrument (in the laboratory) and calculate a correction term
according to the calibration certificate to be applied to the field
measurements.
◦ Attraction of the atmosphere. This is real gravitation. It contains
irregular variations with place and time that we need to remove
from the observed gravity values.
The effect of the atmosphere can be evaluated with the aid of the
Bouguer-plate approximation: if air pressure is p, then the surface

gravity as workers cleared snow from its laboratory roof, including a tea break!
“Weighing” visitors to the lab by their gravitational attraction is also standard fare.
13 Their periods are in the range of about 300–30 000 seconds — frequencies 0.03–3 mHz

— and they are of considerable geophysical interest, Wikipedia, Earth normal modes.

í Õ ! ¤. û
Atmospheric influence on gravity measurement
311
mass density of the atmosphere is

κ= p γ,
/︁

where γ is a representative gravity value inside the atmosphere.

We do not make a large error by using the sea-level value
γ ≈ 9.81 m/s2 . The standard value of air pressure at sea level
is 1013.25 hPa, giving us on sea level14 κ ≈ 10 329 kg/m2 . The effect
of the Bouguer plate is

2πGκ = 0.43 mGal (11.6)

in the upwards direction.

It would however be wrong to apply this value as a correction! The
standard atmosphere is in reality a spherical shell inside which
the measurements are made, and inside the shell its attraction
vanishes, see section 1.4.
Instead, variations in air pressure have a proportional effect. If the
air pressure disturbance is ∆p = p − p0 , in which p0 is mean air
pressure, the correction to be made to a gravity measurement will
be
∆p
δgA = 0.43 p mGal.
0

During the passage of a storm or weather front, this beautiful

theory collapses, and simple equations give misleading results.
Then it is best to just not do any gravity measurements!
◦ Including the atmosphere in the mass of the Earth. This is not a
correction to be applied to gravity measurements. It is a reduction
which is applied in the calculation of gravity anomalies, if one
wants anomalies in which the effect of the atmosphere does not
produce a bias.

14 So yes, the force acting on a standard 14-inch laptop screen is 540 kg. . . but it doesn’t
matter as it is not an old-fashioned vacuum cathode-ray tube.

í Õ ! ¤. û
312 Gravimetric measurement devices

Remember that the reference or normal gravity field of GRS80

is defined in such a way that the parameter GM⊕ contains the
whole mass of the Earth, including the atmosphere, the Earth’s
attraction as satellites are feeling it (Heikkinen, 1981).
Therefore, if one wishes to calculate gravity anomalies that have
a global mean of zero, one should also reduce measured gravity
by computationally moving the whole atmosphere above the point of
measurement to below the measurement point, for example to sea
level.
The total mass of the atmosphere is
p
MA = 4πκR2 = 4π γ R2 .

According to Newton its attraction is

GMA 4πGp
= γ ,
R2
twice the Bouguer-plate atmospheric reduction 11.6 calculated
above. This is the value that should be added to the measured
gravity values.
One may think of this value as the change in gravity if the local
atmospheric Bouguer plate were condensed, Helmert condensation
style, to below the measurement location, producing a double
Bouguer correction.
At sea level, the correction is 0.87 mGal. At height, the correction
is
p(H)
0.87 p mGal,
0
in which p(H) and p0 are the air pressures at height H and at sea
level, respectively.

^ 11.7 Airborne gravimetry and GNSS

In the early years of the 1990s GPS, the Global Positioning System, and
more generally, satellite positioning, has changed airborne gravimetry

í Õ ! ¤. û
Airborne gravimetry and GNSS
313
from a difficult technology to something completely operational. To
understand this, one must know the principle of operation of airborne
gravimetry.
An aircraft carries an airborne gravimeter, an instrument that, in the
same way as a sea gravimeter, is strongly damped. The measurement vaimennus
is done automatically, generally using electrostatic compensation. The
instrument is mounted on a stabilised platform that follows the local
vertical.
During flight, the gravimeter measures total gravity on board the
aircraft, consisting of two parts:
1. gravity proper — gravity as felt in a reference frame connected to vertauskehys
the solid Earth and rotating with it
2. the pseudo-forces caused by the inevitable accelerations of the
aircraft even in cruise flight.
Attached to the aircraft are a number of GNSS antennas. With these and
a geodetic GNSS instrument, the motions of the aircraft can be monitored
with centimetre accuracy. From these motions, we may then calculate
the pseudo-forces mentioned above under item 2.
If we measure the position of the plane (or instrument) xi at moments
ti , ∆t = ti+1 − ti , we obtain estimated acceleration values as follows (in
an inertial frame):
x + xi−1 − 2xi
ai ≈ i+1 . (11.7)
(∆t)2
When the acceleration measured by the gravimeter is Γi and the direction
of the local plumb line (upwards) ni , local gravity gi follows: luotiviiva

gi = Γi − ⟨(ai − f⊕,i ) · ni ⟩ ,

in which f⊕,i is the centrifugal acceleration of Earth rotation, equation

4.1.
The choice of the time constant ∆t is critical in the whole method.
It is best to choose it as long as possible, as then, the precision of the

í Õ ! ¤. û
314 Gravimetric measurement devices

calculated GNSS accelerations ai is as good as possible. The damping of

the gravimeter is also chosen in accordance with ∆t, and the observations
are filtered digitally: all frequencies above the bound ∆t−1 are removed,
because they are almost entirely caused by the motions of the aircraft.
Often the high-frequency part removed from the signal is 10 000 times
stronger than the gravity signal we are after! See for example Lu et al.
(2017) figure 2.
If the uncertainty (mean error) of one GNSS vertical position co-ordinate
measurement is σz (and the different co-ordinates do not correlate with
each other!), then according to equation 11.7 the uncertainty of the
vertical acceleration is √
σz 6
σa = .
(∆t)2
Making the time interval ∆t as long as possible without resolution
suffering requires a low flight speed. Generally a propeller aircraft or
even a helicopter is used. Of course the price of the measurement grows
with the duration of the flight — a helicopter rotor hour is expensive!
For the flight height H we choose in accordance with resolution ∆x:

H ∼ ∆x = v ∆t,

where v is the flight speed. The separation between adjacent flight lines
is chosen similarly.
The first major airborne gravimetry project was probably the Greenland
Aerogeophysics Project (Brozena, 1992). In this ambitious American-
Danish project in the summers of 1991 and 1992, over 200 000 km was
flown, all the time measuring gravity and the magnetic field, and the
height of the ice surface using altimetry.
After that, other large uninhabited areas in the Arctic and Antarctic
regions were also mapped, see Brozena et al. (1996), Brozena and Peters
(1994). Already in subsection 9.6.2 we made mention of other large
surveys. Activity continues, see Coakley et al. (2013), Kenyon et al. (2012).
The method is very suitable for large, uninhabited areas, but also, for

í Õ ! ¤. û
 Measuring the gravity gradient
315
example, for sea areas close to the coast where ship gravimeters would
have difficulty navigating long straight tracks. In 1999, an airborne
gravimetry campaign was undertaken over the Baltic Sea, including the
Gulf of Finland (Jussi Kääriäinen, personal communication).
In addition to the economic viewpoint, an important advantage of
airborne gravimetry is that homogeneous coverage by gravimetric data
is obtained from a large area. The homogeneity of surface gravimetric
data collected over many decades is difficult to guarantee in the same
way. Moreover, the effect of the very local terrain, which for surface
measurements is a hard-to-remove systematic error source, especially
in mountainous terrain (see section 6.3), does not come into play in the
same way for airborne gravimetry.
The operating principle of satellite gravimetry, for example GOCE
(Geopotential and Steady-state Ocean Circulation Explorer), is similar.
An essential difference is, however, that the instrumentation on the
satellite is in a state of weightlessness: Γ = 0 (in a high orbit, or
when using an air drag compensation mechanism), or Γ is small and ilmanvastus
is measured using a sensitive accelerometer (in a low orbit, where air kiihtyvyys-
drag is significant). mittari
The greatest challenge in planning a satellite mission is choosing
the flight height. The lowest possible height is some 200 km. At that
height, a tankload of propellant is already needed, or the flight will not ajoaine
last long. However, the resolution of the measurements on the Earth’s
surface is limited: for example, the smallest details in the Earth’s gravity
field “seen” by the GOCE satellite are 50–100 km in diameter.

^ 11.8 Measuring the gravity gradient

The acceleration of gravity g is the gradient of the geopotential W. The
acceleration of gravity varies itself with place, especially close to masses.

í Õ ! ¤. û
316 Gravimetric measurement devices

We speak of the gravity-gradient tensor or Eötvös tensor:

∂2 ∂2 ∂2
⎡ ⎤
⎢ ∂x2 ∂x∂y ∂x∂z ⎥
⎡ ⎤
∂xx ∂xy ∂xz
∂2 ∂2 ∂2 ⎥
⎢ ⎥
def ⎢
M=⎢ ⎥ W = ⎣ ∂yx ∂yy ∂yz ⎦ W.
⎢ ⎥
⎢ ∂y∂x ∂y 2 ∂y∂z ⎥
⎣
∂2 ∂2 ∂2
⎦ ∂zx ∂zy ∂zz
∂z∂x ∂z∂y ∂z2
We know that gravity increases going down, at least in free air. Going
up, gravity diminishes, about 0.3 mGal for every metre of height.
In topocentric co-ordinates (x, y, z), where z points to the zenith, this
matrix is approximately
⎡ ⎤
−0.15 0 0
M≈⎣ 0 −0.15 0 ⎦ mGal/m,
⎢ ⎥

0 0 0.3
in which ∂zz W = ∂z gz = −∂z g ≈ 0.3 mGal/m is the standard value for
ilmagradientti the vertical free-air gravity gradient: Newton’s law gives for a spherical
Earth
GM
gz = − .
(R + z)2
The minus sign is because g points downwards while the z co-ordinate
increases going up. Derivation gives

∂ GM ∂ (R + z) 2gz
gz = 2 3
· =− ≈
∂z (R + z) ∂z (R + z)
≈ 3 · 10−6 m/s2 m = 0.3 mGal/m.
/︁

The quantities ∂xx W and ∂yy W again represent the curvatures of the
equipotential or level surfaces in the x and y directions, equations 4.4:
∂2 W g ∂2 W g
∂xx W = 2
= −ρ , ∂yy W = 2
= −ρ ,
∂x x ∂y y

in which ρx and ρy are the radii of curvature in the x and y directions.

The substitution ρx = ρy = R yields

∂xx W = ∂yy W ≈ −1.5 · 10−6 m/s2 m = −0.15 mGal/m.

/︁

í Õ ! ¤. û
 Self-test questions
317
The Hungarian researcher Loránd Eötvös did a number of clever exper-
iments (Eötvös, 1998) in order to measure components of the gravity-
gradient tensor with torsion balances built by him. The method continues torsiovaaka,
to be in use in geophysical research, because the gravity gradient as a kiertoheiluri
measured quantity is very sensitive to local variations in matter density
in the Earth’s crust.
In honour of Eötvös, we use as the unit of gravity gradient the eötvös,
symbol E:
1 E = 10−9 m/s2 m = 10−4 mGal/m.
/︁

The above tensor is now

⎡ ⎤
−1500 0 0
M≈⎣ 0 −1500 0 ⎦ E.
⎢ ⎥

0 0 3000

Note that
∂2 W ∂2 W ∂2 W
+ + = ∂xx W + ∂yy W + ∂zz W ≈ 0,
∂x2 ∂y2 ∂z2

the familiar Laplace differential equation. However, the equation is not

exact here: in a co-ordinate system co-rotating with the Earth, the term
for the centrifugal force divergence, 2ω2⊕ , must be added, equation 4.2.
The gravity-gradient field of Sun and Moon is known on the Earth’s
surface as the tidal field, see section 14.1.

^ Self-test questions
1. For the spring gravimeter described in section 11.2, one milligal
of change in gravity g produces according to equation 11.3 a
lengthening of 5 · 10−8 m. Do a calculational check.
2. Why is a pendulum gravimeter, although theoretically absolute,
not very accurate as an absolute gravimeter?

í Õ ! ¤. û
318 Gravimetric measurement devices

3. By which method choices do we, in practical measurements, take

käynti the drift of a relative gravimeter into account?
vertauspiste 4. Why were the reference points of international fundamental gravi-
metric networks often on airports before the advent of absolute
gravimeters?
5. In an absolute or ballistic gravimeter, what is the role of:
(a) the “cage” surrounding the falling prism
(b) the “superspring”?
6. According to Google
◦ The Gulf War from 1990 to 1991 was the first conflict in which
the military widely used GPS.
◦ By December 1993, GPS achieved initial operational capability
(IOC), indicating a full constellation (24 satellites) being
available.
◦ The Greenland Aerogeophysics Project, the first ever large-
scale airborne gravimetric mission, mapped the gravity field
of Greenland during the summers of 1991 and 1992.
Why are these three dates so close together?

^ Exercise 11 – 1: Absolute gravimeter

The observation process of absolute gravimetry is described by the
equation
z = z0 + v0 t + 21 gt2 .

Let us assume that the distance of falling is 30 cm.

1. How long is the time of falling?
2. If we aim at an accuracy of ±10 µGal, how precisely should the
laser interferometer then measure the falling distance of the prism?
A crude order-of-magnitude guesstimate is enough!

í Õ ! ¤. û
 Exercise 11 – 2: Spring gravimeter
319
3. The same question for the time registration of the falling time.

^ Exercise 11 – 2: Spring gravimeter

When we use a spring gravimeter in the field, we place the device at
every measurement station on a solid base, for example bedrock, for
measurement, and level it.
Furthermore we always take care that
◦ The device is arrested during transport: the beam is clamped to be
motionless.
◦ The internal temperature of the device is kept constant by a
thermostat system.
The reason for this is that the functioning of a spring gravimeter depends
on the properties of the spring material, which may change as a result
of careless handling or temperature variations.
Furthermore a gravimeter always has a drift: the relationship between käynti
measured value and true value changes slowly over time. In a non-
factory-fresh gravimeter, this drift is however very regular and almost
linear.
As a result of the drift, a spring gravimeter cannot be used for absolute
gravity measurement and is therefore called a relative gravimeter.
Question How is the relative nature of a spring gravimeter and its drift
taken into account
1. in planning the topology of the measurement network?
2. in planning the time order of the different measurements in
a network?
3. in the choice of vehicles and point locations?

í Õ ! ¤. û
320 Gravimetric measurement devices

^ Exercise 11 – 3: Air pressure and gravity

1. How much does a low-pressure zone of 100 hPa (meaning that the
pressure is 100 hPa lower than average air pressure 1013.25 hPa)
affect gravity measured on the Earth’s surface? You may assume
the low-pressure zone to be very extended in area.
ylösalainen 2. How much does sea water rise due to the “inverted barometer
ilmapuntari effect” under a low-pressure zone?
3. How much does the effect from point 2 amount to in local gravity
measured on a ship? Assume that you are on the open sea, that
the free-air gravity gradient is −0.3 mGal/m, and that the density of
sea water is 1030 kg/m3 . Analyse the situation carefully.15

15 And I mean really carefully.

í Õ ! ¤. û
^ The geoid, mean sea level, and
sea-surface topography

^ 12.1 Basic concepts

12
On the ocean, the geoid is on average at the same level as the mean sea
level, the surface obtained by removing all periodic and quasi-periodic
variations from the instantaneous sea surface. These variations include
◦ tidal phenomena, caused by Sun and Moon, of an order of magni-
tude of ±1 m, locally even more
◦ variations caused by air pressure variations (“inverted barometer ylösalainen
effect”, IB). Typically of an order of decimetres, under tropical ilmapuntari
cyclones up to metres
◦ “wind pile-up”, water being pushed by winds
◦ in littoral seas, variation in the volume of fresh water flowing out
from rivers into the sea
◦ eddies that are formed in the oceans in connection with, for exam-
ple, the Gulf Stream and the Agulhas Stream (“mesoscale eddies”)
that may live for months, and inside of which the sea surface may
be even decimetres above or below that of the surroundings
◦ the continual shifting of ocean currents from one place to another
◦ the ENSO, El Niño Southern Oscillation, is a very long time-scale,
quasi-periodic weather phenomenon happening in the waters
of the Pacific Ocean and the air above it, but affecting weather

– 321 –
322 The geoid, mean sea level, and sea-surface topography

phenomena worldwide. The time-scale of variability ranges from

two to seven years. See figure 13.1.
If we remove all these periodic and quasi-periodic variations, we are
left with the mean sea level. If the water of the seas were in a state of
equilibrium, this mean sea surface would be an equipotential or level
surface of the Earth’s gravity field, the geoid.
This is, however, not how things really are. Mean sea level differs
from a level surface due to for example the following phenomena:
◦ Permanent ocean currents cause, through the Coriolis force, per-
manent differences in mean water level.
◦ Permanent differences in temperature and salinity also cause
permanent differences in the mean water level, the latter for
example in front of the mouths of rivers.
These physical phenomena, among other things, cause the so-called
meritopografia sea-surface topography, a permanent separation between the mean sea
surface and the geoid.
The classical definition of the geoid is

“The level surface of the Earth’s gravity field that fits on average best to
the mean sea level.”

The practical problem with this definition is that determining the correct
level of the geoid requires knowledge of the mean sea level everywhere
on the world ocean. This is why many “geoid” models in practice do
not coincide with global mean sea level, but with some locally defined
mean sea level — and often only approximately.
Mean sea level in its turn is also a problematic concept. It is sea level
from which all periodic effects have been computationally removed
— but who can know if a so-called secular effect in reality is perhaps
long-period? The measure of permanency is the time series that are
mareografi available, as tide gauges have been widely operating already for about

í Õ ! ¤. û
 Geoids and national height datums
323
a century, when again modern satellite time series — TOPEX/Poseidon
and its successors — are just about a quarter of a century long.
A sensible compromise may be the average sea level over 18 years, an
important periodicity, saros (Wikipedia, Saros), in the orbital motion of
the Moon.
The sea-surface topography is defined as the difference between the meritopografia
mean sea level and the geoid. See figure 12.4.

^ 12.2 Geoids and national height datums

A locally determined geoid model is generally relative. Locally, at the
current state of the art, we have no access to global mean sea level
at an acceptable precision. This is likely to change with technology
development.
In general, a local geoid model is tied to a national height system,
and the difference from the classical definition is thus the same as the
difference of the national height system from the global mean sea level.
In Finland, heights were determined for a long time in the N60 height
system, which is tied to the mean sea level in Helsinki harbour at the
start of 1960. The difference between it and the global mean sea level
is about 30 cm, due to the sea-surface topography in the Baltic Sea, see
figure 12.4. The reference benchmark of the system is located in nearby pääkiintopiste
Kaivopuisto, figure 7.2. Precise levelling disseminated heights from
here all over Finland.
The modern Finnish height system is N2000, which is in principle
tied to the mean sea level in Amsterdam, which is close to global mean
sea level. The reference benchmark in Finland is similarly located at the
Metsähovi research station in the Kirkkonummi municipality, west of
Helsinki.
At the beginning of 1960, the reference surface of the Finnish height
system N60 was an equipotential or level surface of the Earth’s gravity
field. However, due to uplift, that is no longer the case: the post-glacial

í Õ ! ¤. û
324 The geoid, mean sea level, and sea-surface topography

land uplift varies from some four millimetres per year in the Helsinki
area to ten millimetres per year in the area of maximum land uplift near
Pohjanmaa Ostrobothnia. This is the main reason why in Fennoscandia, height
systems have a “best before” date and must be modernised a couple of
times per century.
Generally, geoid maps for practical use, like FIN2000, the Finnish
geoid model (figure 9.5), are constructed so that they transform heights
in the national height system, for example N60 heights (Helmert heights)
vertaus- above “mean sea level” to heights above the reference ellipsoid of the
ellipsoidi GRS80 system.

As land uplift is an ongoing process, it must be tied to a certain epoch,

a point in time at which the GNSS measurements were done to which
the original gravimetric geoid solution has been fitted. In the case of
FIN2000 this was 1997.0 (Matti Ollikainen, personal communication;
Bilker-Koivula and Ollikainen, 2009; Häkli et al., 2009).
Strictly speaking then, FIN2000 is not a model of the geoid. A better
name might be a “transformation surface”. This holds true, in fact, for
all national or regional geoid models that are built primarily for the
purpose of enabling the use of GNSS in height determination (“GNSS
levelling”). These “geoid-like surfaces” are generally constructed in the
following way:
1. We calculate a gravimetric geoid model by using the Stokes equa-
tion and remove–restore, for example using the FFT calculation
technique.
2. We fit this geoid surface solution to a number of comparison
points, in which both the height from levelling — “above sea level”
— and from the GNSS method — above the reference ellipsoid
— are known. The fit takes place for example by modelling the
differences as a polynomial function:

δN = a + b (λ − λ0 ) + c (φ − φ0 ) + ...

or something more complicated, and solving the coefficients

í Õ ! ¤. û
 The geoid and post-glacial land uplift
325
a, b, c, . . . from the differences between the two heights in these
known comparison points by using the least-squares method. pienimmän
neliösumman
menetelmä
^ 12.3 The geoid and post-glacial land uplift
Global mean sea level is not constant. It rises slowly by an amount that,
over the past century, has slowly grown. Over the whole 20th century,
the rate has been 1.5 – 2.0 mm/a, for example 1.6 mm/a (Wöppelmann et al.,
2009). Over the last several decades, the rate has accelerated and is now
over 3 mm/a, see figure 13.1.
This value is called the eustatic rise of mean sea level. It is caused partly
by the melting of glaciers, ice caps, and continental ice sheets, partly by mannerjäätikkö
the thermal expansion of sea water. A precise value for the eustatic rise
is hard to determine: almost all tide gauges used for monitoring sea mareografi
level have their own vertical motions, and distinguishing these from
the rise in sea level requires a representative geographic distribution
of measurement locations. The ongoing response of the solid Earth
to the ending of the last ice age in particular, the latest deglaciation
or termination, the so-called GIA (glacial isostatic adjustment), is a
global phenomenon that it has only been possible to observe by satellite
positioning in the most recent decades.
Because of eustatic sea-level rise, a distinction must be made between
absolute and relative land uplift:
Absolute land uplift is the motion of the Earth’s crust relative to the
centre of mass of the Earth. This land uplift is measured when
using satellites the orbits of which are determined in a co-ordinate vertaus-
reference system tied to the Earth’s centre of mass, for example, järjestelmä
positioning of tide gauges by means of GNSS.
Relative land uplift is the motion of the Earth’s crust relative to the
mean sea level. This motion is measured by tide gauges, also
called mareographs.
Geoid rise As the post-glacial land uplift is the shifting of masses

í Õ ! ¤. û
326 The geoid, mean sea level, and sea-surface topography

internal to the Earth from one place to another, it is clear that

the geoid must also change. The geoid rise is, however, small
compared to the land uplift, only a few percent of it.

Equation (the point above a quantity denotes the time derivative with
respect to time1 ):

̇ = ḣ − N
H ̇ =H
̇r+H
̇e+H
̇ t,

in which
Ḣ absolute land uplift from the geoid
vertaus- ḣ absolute land uplift from the reference ellipsoid
ellipsoidi ̇
Hr relative land uplift from the local mean sea level
̇e
H eustatic (global mean sea level) rise
̇t
meritopografia H change over time of the sea-surface topography (likely small)
̇
N geoid rise.

The rise of the geoid as a result of land uplift can be simply calculated
with the Stokes equation:
x
dN R d
(︂ )︂
= S(ψ) ∆g dσ.
dt 4πγ σ dt
d
Here, dt ∆g is the change of gravity anomalies over time due to land
uplift. Unfortunately we do not precisely know the mechanism by which
mass flows in the Earth’s mantle to underneath the land-uplift area. We
may posit
d dH ̇,
∆g = c = cH
dt dt
in which the constant c may range from −0.16 to −0.31 mGal/m.
◦ The value −0.16 mGal/m is called the “Bouguer hypothesis”: it
corresponds to the situation in which upper mantle matter flows
into the space freed up underneath the rising Earth’s crust, in

1 This dot notation, fluxion, was introduced by Newton.

í Õ ! ¤. û
 The geoid and post-glacial land uplift
327

Earth’s crust

Asthenosphere

(a)
Bouguer hypothesis. . .

Earth’s crust

Upper mantle

(b)
. . . and free-air hypothesis.

Figure 12.1. The two different hypotheses on the mechanism of post-glacial

^ land uplift.

order to fill it. This matter may be roughly modelled as a Bouguer

plate.
◦ The value −0.31 mGal/m is the opposite extreme, the “free-air hy-
pothesis”. By this hypothesis, the ice load during the last ice
age has only compressed the Earth’s mantle, and now it is slowly
expanding again to its former volume (the “rising dough model”). pullataikina-
malli
Up until fairly recently, the most likely value was about −0.2 mGal/m,
with substantial uncertainty. The latest results (Mäkinen et al., 2010;
Olsson et al., 2019) may be summarized as −0.16 ± 0.02 mGal/m (one

í Õ ! ¤. û
328 The geoid, mean sea level, and sea-surface topography

5 35
65 10 65
15 30
20 25

Föllinge
Meldal
Kopperå Stugun Vaasa Joensuu
Vågstranda Kramfors
Äänekoski

60 60
5 35
10 30
15 20 25

^ Figure 12.2. The Fennoscandian gravity line on the 63rd parallel north.

standard deviation), which would seem to settle the issue. It looks like
the Bouguer hypothesis is closer to physical reality. The flow of mass
probably happens within the asthenosphere.
This problem has been studied much in the Nordic countries. The
method used has been gravimetric measurement along the 63rd parallel
north (the “Blue Road Geotraverse” project). The measurement stations
extend from the Norwegian coast to the Russian border, and have been
chosen such that the gravity along them varies within a narrow range.
In this way, the effect of the scale error of the gravimeters is avoided.
Clearly, absolute gravity is of no interest here, only the change in gravity
differences over time between the stations.
These measurements have been made over many years using high-
precision spring or relative gravimeters. In recent years, there has been a
shift to using absolute gravimeters, obviating the need for measurement
lines.

í Õ ! ¤. û
 Methods for determining the sea-surface topography
329
^ 12.4 Methods for determining the sea-surface
topography
In principle three geodetic methods exist:
◦ satellite radar altimetry and gravimetric geoid determination
◦ positioning of tide gauges along the coast using GNSS, together mareografi
with gravimetric geoid determination
◦ precise levelling along the coast connecting tide gauges.
In addition to this, we still have the oceanographic method, i. e., physical
modelling. The method is termed steric levelling if temperature and
salinity measurements along vertical profiles are used on the open
ocean, and geostrophic levelling of ocean current measurements are used
to determine the Coriolis effect, generally close to the coast.
All methods should give the same results. The Baltic Sea is a textbook
example where all three geodetic methods have been used. It has been
found that the whole Baltic Sea surface is tilted: relative to a level
surface, the sea surface goes up from the Danish straits to the bottoms pohjukka
of the Gulf of Finland and the Bothnian Bay by 25–30 cm.
Oceanographic model calculations show that this tilt is mainly due
to a salinity gradient: in the Atlantic Ocean, the salinity is 30–35 o/oo,
when in the Baltic it drops to 5–10 o/oo, due to the massive production
of fresh water by the rivers (Ekman, 1992). Of course on top of this
come temporal variations, like oscillations such as those in a bathtub,
the amplitude of which can be over a metre.
In Ekman (1992) more is said about the sea-surface topography of the meritopografia
Baltic and its determination.

^ 12.5 Global sea-surface topography and heat transport

One important reason why researchers are interested in the global
sea-surface topography is that it offers an opportunity to study more
precisely the currents in the oceans and thus the transport of the

í Õ ! ¤. û
330 The geoid, mean sea level, and sea-surface topography

Sun’s energy from the equator to higher latitudes. There are many
matters whose study would be helped by more precise knowledge of
ocean currents: carbon dioxide dissolved into the water, chlorophyll
(phytoplankton), salinity, among many others.
The Coriolis force, or acceleration, caused by the Earth’s rotation is
a=2 v×→ −
⟨︁ ⟩︁
ω⊕ , (12.1)
in which v is the velocity vector in a system attached to the rotating
Earth, and →
−
ω ⊕ is the rotation vector of the Earth. This is an axial vector,
pointing in the direction of the Earth’s axis of rotation.
If a fluid flows on the Earth’s surface, then, in the above equation
12.1, only the part of →
−
ω ⊕ in the normal direction n to the ocean surface
will have an effect: this part has a length of →
⟨︁− ⟩︁
ω ⊕ · n = ω⊕ sin φ, and
the vector equation 12.1 may be replaced by the simpler scalar equation
a = 2vω⊕ sin φ,
def
in which a = ∥a − ⟨a · n⟩ n∥, the length of the projection of a onto the
local horizon plane, and v = ∥v∥, ω⊕ = ⃦→
⃦− ⃦
ω ⊕ ⃦ in the familiar way. The
direction of the Coriolis acceleration is always perpendicular to the flow
velocity: when watching along the flow direction, to the right in the
northern hemisphere, to the left in the southern hemisphere.
As a result of the Coriolis force, the sea surface in the area of an ocean
current is tilted sideways with respect to the current, at an angle
a ω⊕
γ = 2v γ sin φ.
Here, γ is local gravity. This equilibrium between Coriolis force and
the horizontal gradient of pressure is called the geostrophic equilibrium.
On the equator, it can be seen from the equation that the tilt is zero, but
everywhere else, ocean currents are tilted.
For example, in the case of the Gulf Stream, the height variation
caused by this effect is several decimetres. If we define a local (x, y) co-
ordinate system in which x(φ, λ) is pointing north and y(φ, λ) east, we
meritopografia may write for the sea-surface topography H the geostrophic equations

í Õ ! ¤. û
 Global sea-surface topography and heat transport
331

Figure 12.3. Connection between sea-surface topography and ocean currents.

^ Red arrows depict ocean currents, curves, sea-surface topography.

∂H ω ∂H ω
= −2vy γ⊕ sin φ, = +2vx γ⊕ sin φ. (12.2)
∂x ∂y
As we will see in chapter 13, we can measure the location in space of
the sea surface at a precision of a few centimetres using satellite radar
altimetry. If we furthermore have a precise geoid map, we may calculate
the sea-surface topography, and with the aid of equations 12.2 solve for meritopografia
the flow velocity vector field2
[︂ ]︂T [︂ ]︂T
vx (x, y) vy (x, y) = vx (φ, λ) vy (φ, λ) .

An elegant property of these equations is that we do not even have

to know the absolute level of the field H(x, y) = H(φ, λ), because that
vanishes in differentiation.
The method described, figure 12.3, requires a sufficiently precise
geoid map of the oceans of the world. The GOCE satellite fits this need

2A popular, though unofficial, unit for ocean current is the sverdrup (Wikipedia,
Sverdrup), a million cubic metres per second. All the rivers of the world together
make about one sverdrup, while the Gulf Stream is 30–150 Sv. “There is a river in the
ocean” – Matthew Fontaine Maury (1806–1873), American polymath and pioneer of
oceanography.

í Õ ! ¤. û
332 The geoid, mean sea level, and sea-surface topography

Figure 12.4. Sea-surface topography map produced by GOCE. Base map ©

European Space Agency. Unit: cm. Ocean currents superimposed:
^ NOAA / Rick Lumpkin (NOAA, Ocean currents).

like a glove, see section 13.7. One objective of the mission was, as the
name indicates, to get a full picture of ocean currents and especially
their capacity for heat transport. This knowledge helps understand
how the Earth’s climate functions and how it is changing, also as a
result of human activity. This is an important issue for Europe and
Fennoscandia, and also Finland, as the heat energy brought by the Gulf
Stream helps to keep these areas habitable (Caesar et al., 2018).
Even without a geoid model, we can study the variations of ocean
currents using satellite altimetry. It has long been known that in the
North Atlantic Ocean, mesoscale eddies have been moving alongside the
Gulf Stream; eddies of size 10–100 km which show up in altimetric
imagery. It is interesting that the eddies also show up in maps of the

í Õ ! ¤. û
 The global behaviour of the sea level
333
ocean surface temperature, and biologists have observed that life inside
the eddies differs from that outside them (Godø et al., 2012). The life
span of the eddies can be weeks, even months.
A good, though somewhat dated, introduction to “geodetic oceanog-
raphy” is given by Rummel and Sansó (1992).

^ 12.6 The global behaviour of the sea level

Water exists on the Earth in three phases: liquid, ice, and vapour. During
geological history, the ratio between liquid water and ice in particular
has varied substantially. Also today, a large amount of ice is tied up
in continental ice sheets, specifically in Antarctica and Greenland. Of mannerjäätikkö
these, the Eastern Antarctic ice sheet is overwhelmingly the largest.
When the amount of water tied up in continental ice sheets varies, so
does the sea level. The end of the last ice age has raised the sea level
by as much as 120 m, a process that ran to completion some 7000 years
ago3 (Wikipedia, Sea level rise). Not until the last century or two has the
sea level again started rising, and the rise accelerating, as a consequence
of global warming.
We still live in the aftermath of the last glaciation. There were
large continental ice sheets which have since melted away, like in
Fennoscandia and in Canada (the Laurentide ice sheet): the land is still
rising at an even pace, up to 10 and 14 millimetres per year, respectively.
Around the land-uplift areas, in central Europe and the United States,
a subsidence of the land is taking place at an annual rate of 0.5–1.7 vajoaminen
millimetres, for example DeJong et al. (2015). Directly underneath the
hard crust of the Earth or lithosphere, in the upper mantle layer called
asthenosphere, material is flowing slowly inwards under the rising Earth’s
crust.

3 7000years “before present”, 7 ka BP. BP conventionally means: before 1950. Nowa-

days b2k, before the year 2000, is also used.

í Õ ! ¤. û
334 The geoid, mean sea level, and sea-surface topography

In order to complicate the picture, the sea-level rise caused by the

mannerjäätikkö melting of continental ice sheets also presses the ocean floor down —
by as much as 0.3 mm per year; the so-called Peltier effect (Peltier, 2009).
mareografi Therefore, the measured sea-level rise — whether on the coast by tide
gauges, or from space using satellite altimetry — does not represent the
whole change in total ocean water volume. If the latter is what interests us,
as it always does in climate research, this Peltier correction must still be
added to the observation values.
vajoaminen The subsidence of the sea floor has not even been globally uniform: at
the edges of the continents a “lever motion” happens when the sea floor
subsides but the dry land does not. And in the tropics in the Indian and
Pacific Oceans, the sea level reached its maximum level, the mid-Holocene
highstand, relative to the Earth’s crust approximately 7000 years ago.
After this, the local sea level subsided and the coral formations from
that age remained, dead, some 2–3 m above the modern sea level. This
is how, for example, Tuvalu and the Maldives were formed, which are
now being threatened by the modern sea-level rise again.

^ 12.7 The sea-level equation

Scientifically the variations in sea level are studied using the sea-level
equation. A pioneer in this field has been the Canadian Richard Peltier
(Peltier), who has constructed physics-based models of how both the
solid Earth and sea level respond when the total mass of the continental
ice sheets changes.
The sea-level equation is (Farrell and Clark, 1976; Spada and Melini,
2015):
G
(︂ (︁ )︁ (︁ )︁ )︂
S = SE + ρi Gs ⊛i I − Gs ⊛i I + ρw Gs ⊛o S − Gs ⊛o S , (12.3)
R
in which
◦ S = S(ω, t) = S(ϕ, λ, t) means the variations of sea level as a
function of place ω = (ϕ, λ) and time t. These variations are

í Õ ! ¤. û
 The sea-level equation
335
Sea level
drops Sea level Sea level
rises drops

Greenland
Antarctica

Figure 12.5. The sea-level equation. Sea level reacts in a complicated way
^ when continental ice sheets melt.

relative to the solid Earth’s surface, i. e., changes in sea depth. S is

also what tide gauges measure. mareografi
◦ I = I(ω, t) is similarly a function of place and time describing the
variations in thickness of ice sheets and glaciers.
◦ SE is the eustatic term, the variation in ice volume converted into
“equivalent global sea-level variation”, in an equation
mi (t)
SE (t) = − ,
ρw Ao
in which mi (t) is the variation in total ice mass as a function of
time, ρw the density of sea water, and Ao the total surface area of
the oceans.
◦ R is the mean radius of the Earth, G Newton’s universal gravita-
tional constant, section 1.2.
◦ ρ is the density of matter: ρi that of ice, and ρw that of sea water.
◦ ⊛ is the symbol of a convolution over the surface of the Earth and
the time axis, ⊛i over land ice, ⊛o over the oceans: Green’s function
is multiplied with the ice and sea functions and integrated over
the domain in question. These integrals are, by the way, very
similar to the ones discussed in section 8.1. For example

í Õ ! ¤. û
336 The geoid, mean sea level, and sea-surface topography

{︁ }︁
Gs ⊛o S (ω, t) =
wt x
Gs ψ(ω, ω ′ ), (t − t ′ ) S(ω ′ , t ′ ) dω ′ dt ′ ,
(︁ )︁
=
−∞ ocean

in which ψ(ω, ω ′ ) is the geocentric angular distance between

evaluation point ω = (ϕ, λ) and data point ω ′ = (ϕ ′ , λ ′ ). The
measure of the surface integral is dω = R2 dσ = R2 cos ϕ dϕ dλ.
As can be seen, we have here a convolution applied both over the
Earth’s surface ω and over the time axis t.
◦ The overbar designates averaging over the whole relevant surface
area.
◦ Gs is the Green’s function of sea level

1
Gs (ψ, ∆t) = γ GV (ψ, ∆t) − Gr (ψ, ∆t). (12.4)

Here the Green’s function of the geopotential is

GV (ψ, ∆t) = GrV (ψ, ∆t) + GeV (ψ, ∆t) + GvV (ψ, ∆t),

where ψ is the angular distance of the evaluation point from

′ r
the data
/︂ point and ∆t = t − t ⩾ 0. The function GV (ψ, ∆t) =
δ(∆t) 2 sin 1 ψ is the rigid partial Green’s function, representing
2
the change in potential caused by a mass of water or ice before
any deformation takes place.
The functions GeV and GvV are the elastic and viscous partial Green’s
functions of deformation. These thus characterise the rheological
behaviour of the Earth, and their theoretical calculation requires
the internal density and viscosity distributions ρ(r) and η(r) of
the Earth, assuming they are isotropic, only dependent upon r.

Gr (ψ, ∆t) = Ger (ψ, ∆t) + Gvr (ψ, ∆t)

is similarly Green’s function of vertical or radial displacement of the sea

floor, in the same way split into elastic and viscous parts. There
trivially is no “rigid” part.

í Õ ! ¤. û
 The sea-level equation
337

^ Figure 12.6. Sea-level rise after the last ice age (Rohde, 2005).

The behaviour of the sea level can now be computed in this way where
one first tries to construct an “ice-load history” I(ω, t). Then, from this
one tries to calculate iteratively, using sea-level equation 12.3, S(ω, t).
S signifies relative sea-level variation, changes in the relative positions
of sea level and the Earth’s solid body or Earth’s crust. It is a function
of place: one may not assume that it would be the same everywhere.
Mitrovica et al. (2001) show how, for example, the meltwater from
Greenland flees to the southern hemisphere, when the meltwater from
Antarctica again comes similarly to the north. This is a consequence of
the change in the Earth’s gravity field and the geoid, when large volumes
of ice melt. Another factor is also the fact that the physical shape of
the Earth changes when the ice load changes: glacial isostatic adjustment,
GIA.

This also complicates the monitoring of the global mean sea level
from local measurements: the problem is familiar in Fennoscandia,

í Õ ! ¤. û
338 The geoid, mean sea level, and sea-surface topography

where the Earth’s crust, for now, is rising faster than the global sea
level. . .
Green’s functions in the sea-level equation are functions of both ψ
and time difference ∆t. This tells us that GIA is a function of both place
and time. On a spherically symmetric Earth, the functions may be
written as expansions. See Wieczerkowski et al. (1999).
The elastic response of the Earth to loading is instantaneous on the
geological time-scale. It is described by the same elastic Love numbers
that appear in the theory of tidal deformation, see section 14.2. Like
this:
GeV (ψ,∆t) Ge (ψ,∆t)
⏟ ⏟
⏞⏞ ⏟ r
⏞⏞ ⏟
1 ∑︂
∞
1 ∑︂
∞
Ges (ψ, ∆t) = γ · δ(∆t) kn Pn (cos ψ) − γ · δ(∆t) hn Pn (cos ψ),
n=0 n=0

with kn and hn as defined in equations 14.4. δ(∆t) is Dirac’s delta

function.
GIA, however, is viscous deformation on a range of geological time-
scales. Equation 12.4 yields
Gv (ψ,∆t) Gv (ψ,∆t)
⏟ V
⏞⏞ ⏟ ⏟ r
⏞⏞ ⏟
1 ∑︂
∞
1 ∑︂
∞
Gvs (ψ, ∆t) = γ · kvn (∆t)Pn (cos ψ) − γ hvn (∆t)Pn (cos ψ),
n=0 n=0

with the viscous Love numbers for potential and vertical displacement:

∑︂
I ∑︂
I
kvn (∆t) = rkni exp(−sni ∆t), hvn (∆t) = rh
ni exp(−sni ∆t).
i=1 i=1

Here, n is the degree number, and the index i counts the viscous relaxation
k /︁ h /︁
modes for every degree number n. The ratios rni sni and rni sni are
called “modal strengths”, and the τni = 1 sni are relaxation times in
/︁

which the mode in question will decay over time.

Generally, the modes that are of large spatial extent — low degree
numbers n — decay slower, when again the local modes — high

í Õ ! ¤. û
 Self-test questions
339
degree numbers — tend to decay faster. The local modes of the last
deglaciation have today already vanished: the geographic pattern of
the Fennoscandian land uplift is already very smooth and the seismicity
accompanying the deglaciation is pretty much over. Back then, during
the retreat of the ice sheet at its edge, there were strong earthquakes,
traces of which are visible in the landscape (Kuivamäki et al., 1998).
The now-dominant viscous modes are many hundreds of kilometres in
geographic extent and correspondingly of time-scales of thousands of
years.

^ Self-test questions
1. List all the causes of sea-level variations that you are aware of.
2. What is the sea-surface topography? meritopografia
3. What is eustatic sea-level rise?
4. What is the origin of the name “El Niño”?
5. What is absolute, and what is relative land uplift? What does the
difference between the two consist of?
6. Which two main models are on offer for the mechanism of land
uplift?
7. What three geodetic techniques are available for determining the
sea-surface topography?
8. What is the shape of the sea-surface topography of the Baltic Sea,
and what is its cause?
9. What is the Coriolis force, and how does it affect ocean currents?
10. What is the geostrophic balance?
11. In whose honour is the unit sverdrup named?
12. How can one invert a map of the sea-surface topography into a
map of ocean currents? Where on Earth does this not work?
13. What is the Peltier effect? What is the mid-Holocene highstand?

í Õ ! ¤. û
340 The geoid, mean sea level, and sea-surface topography

14. What does the sea-level equation describe?

15. Why does the mean sea level in the Baltic Sea not rise when the
mannerjäätikkö Greenland continental ice sheet melts? What will happen in the
Baltic Sea when the West Antarctic ice sheet melts?

^ Exercise 12 – 1: Coriolis force, ocean current

It is given that the velocity of flow of an ocean current is 0.1 m/s and its
width 100 km.
1. How much is the height difference between its left and right edges?
Assume that the current is at latitude 45◦ north.
2. If the same current was 200 km broad and the velocity of flow
0.05 m/s (so, assuming the same depth, the transport of water would
also be the same), compute for that case the height difference
between the left and the right edges.
3. [For fun] if the depth of the current is 1 km, what is the water
transport in sverdrup?

^ Exercise 12 – 2: Land subsidence and the mechanism of

land uplift
vajoaminen How does the post-glacial land subsidence observed in the United States
and central Europe support a Bouguer type of land-uplift mechanism
(figure 12.1a), rather than a free-air mechanism?

í Õ ! ¤. û
^ Satellite altimetry and satellite
gravity missions

^ 13.1 Satellite altimetry

13
Satellite altimetry is a measurement method in which the distance from a
satellite straight down to the sea surface is measured using a microwave
radar. Historically there have been many satellites carrying an altimetry
radar: see table 13.1, which may not be complete.
◦ The GEOS-3 (1975-027A) and Seasat satellites were American
testing satellites aimed at developing the altimetric technique.
The measurement precision of GEOS-3 was still modest. Before
that, satellite altimetry was also tested with a device, accuracy
±1 m, on board the orbital laboratory Skylab (1973-027A).
◦ Seasat (1978-064A) broke down only three months after launch,
probably due to a short-circuit.1 However, the data from Seasat
was the first large satellite altimetry data set used for determining
the mean sea surface, and that of the Baltic Sea (Vermeer, 1983b).
◦ Geosat (1985-021A) was a satellite launched by the US Navy,
intended to map the gravity field on the world’s oceans; more
precisely the deflections of the plumb line, which are needed luotiviivan
to impart the correct departure direction to ballistic missiles poikkeama
launched from submarines. The 17-day repeat data from the

1 But read this: Wikipedia, Seasat conspiracy theory.

– 341 –
342 Satellite altimetry and satellite gravity missions

^ Table 13.1. Altimetric satellites through the ages.

Satellite Launch Orbital Orbital Repeat Measure- Positioning

year inclina- height periods ment pre- technique
tion (◦ ) (km) (days) cision (m)

GEOS-3 1975 115.0 843 ∼ 38 0.20

Seasat 1978 108.0 780 3, 17 0.08
Geosat 1985 108.0 780 3, 17 0.04
ERS-1 1991 98.5 780 3, 35, 168 0.03
TOPEX/Poseidon 1992 66.0 1337 9.9156 0.033 GPS, DORIS
ERS-2 1995 98.5 780 3, 35 0.03 PRARE
Geosat follow-on 1998 108.0 800 17 0.035
Envisat 2001 98.5 784 35 0.045 GPS, DORIS
Jason-1 2001 66.1 1336 9.9156 0.025 GPS, DORIS
Jason-2 2008 66.04 1336 9.9156 0.025 GPS, DORIS
Cryosat-2 2010 92.0 725 369 DORIS
HY-2A 2011 99.3 970 14, 168 0.085 DORIS, GPS
SARAL/AltiKa 2013 98.5 781 35 DORIS
Jason-3 2016 66.04 1338 9.9927 0.025 GPS, DORIS
Sentinel-3A 2016 98.62 804 27 0.03 DORIS, SLR, GNSS

geodetic mission was initially classified. Later however, the data

from the southern hemisphere was published for scientists to use,
and still later, the whole data set was made public.
◦ The satellites ERS-1/2 (1991-050A, 1995-021A) and Envisat (2002-
009A) were launched by the ESA, the European Space Agency.
The altimeter was just one among many packages. A German
positioning device called PRARE was on the ERS satellites, but it
only functioned after launch on ERS-2.
◦ TOPEX/Poseidon (1992-052A) was an American-French collabo-
meritopografia ration, one goal of which was to precisely map the sea-surface
topography. A special feature was the on-board precise GPS posi-

í Õ ! ¤. û
Satellite altimetry
343
tioning device, which allowed the determination of the location of
the sea surface geocentrically. Together with its successors Jason-1,
2 and 3 (2001-055A, 2008-032A, 2016-002A), this satellite mission
has also produced, and continues to produce, valuable informa-
tion on the global rise of the sea level over the last 25 years, of
about 3 mm per year. See figure 13.1.
The famous oceanographer Walter Munk2 characterised
TOPEX/Poseidon in 2002 as “the most successful ocean experiment
of all time” (Munk, 2002).
◦ HY-2A (2011-043A) is a Chinese satellite also launched by China.
◦ SARAL/AltiKa (2013-009A) is a satellite launched by India. The
altimeter and DORIS are French contributions.
◦ Cryosat-2 (2010-013A) is a satellite launched by the ESA to study
polar sea ice. Of special interest is the freeboard, the amount by varalaita
which the ice sticks out of the water. From this, the thickness, and,
with the surface area, the total volume may be calculated. In-orbit
positioning is done using the French DORIS system.
The launch of Cryosat-1 failed.
◦ Sentinel-3A (2016-011A) is a versatile ESA remote-sensing satellite,
the first of a planned constellation. It carries several instruments,
among them the SRAL, or Synthetic Aperture Radar Altimeter.
The measurement method of satellite radar altimetry is presented
in figure 13.2. The figure shows all the quantities playing a role in
altimetry: the measured range s is the height h of the satellite above
the reference ellipsoid corrected for the geoid height N, the sea-surface vertaus-
topography H, and variations of the sea surface, like tides, eddies, ellipsoidi
annual variation, and so on. meritopografia

Furthermore, if the satellite does not contain a precise positioning

device, the true orbit of the satellite will differ from the calculated orbit

2 Walter Heinrich Munk (1917–2019) was a famous American fhysical oceanographer.

í Õ ! ¤. û
344 Satellite altimetry and satellite gravity missions

60
TOPEX/Poseidon, Jason satellites, global mean sea level, mm

Trend 3.1 ± 0.4 mm/a

40
SOI, shifted, scaled and inverted

20
Jason-3
TOPEX/Poseidon

Jason-2
0

Jason-1

−20
1998, 2016 Super El Niño

−40
5

−60
1990 1995 2000 2005 2010 2015 2020

Figure 13.1. Results from the TOPEX/Poseidon and Jason satellites. Annual
cycle removed. Comparison with ENSO (“El Niño”), SOI = South-
ern Oscillation Index. Data © Colorado University at Boulder’s
^ Sea Level Research Group; Nerem et al. (2010).

— even from the orbit calculated afterwards. Therefore,

h = h0 + ∆h,

in which h0 is the calculated orbit, and ∆h the orbit-error correction.

The measurements are performed by sending thousands of pulses
down each second, measuring the travel times of the reflected return
pulses on board the satellite, averaging them down to a measurement

í Õ ! ¤. û
 Satellite altimetry
345
True orbit

Calculated orbit

h s

Sea-surface topography H Footprint

Geoid

Reference ellipsoid

Geoid height N
Sea surface Mean sea surface

^ Figure 13.2. Satellite altimetry as a measurement method, concepts.

rate of 10–20 values per second, and transmitting these to Earth. Of

these values, the largest and smallest are thrown away as possibly
erroneous, and from the remainder, a mean value is calculated for the
central epoch of the pulse train using linear regression. The value thus
obtained from the regression line is the actual “measurement”: one
every second, making the effective measurement frequency 1 Hz.
The details will vary from satellite to satellite. The pulse shape is
never quite crisp, the place of the reflection on the ocean surface, or
footprint, has a diameter of several kilometres. Especially if the ocean
has wave motion (significant wave height, SWH), then, in the processing
phase, one should make a careful correction so no bias is created: if
the SWH is large, the altimeter footprint — the area on the sea from

í Õ ! ¤. û
346 Satellite altimetry and satellite gravity missions

which radio energy returns to the receiver — will also be larger, and the
distance travelled by the radio waves will on average be a little longer.
The newest satellites use an interferometric technique that differs
somewhat from the description above.
Of all the corrections related to instrumentation, atmosphere, ocean,
and solid Earth, we mention
◦ the height of sea waves (SWH)
◦ solid-Earth tides
◦ ocean tides
◦ the “wet” tropospheric propagation delay, best derived from
measurements with a water vapour radiometer on the satellite,
otherwise from an atmospheric model
◦ the “dry” tropospheric propagation delay
◦ the ionospheric delay, only for the part of the ionosphere below
the satellite, depending on flight height
◦ the altimeter’s own calibration correction — nowadays “in-flight”
calibration is always strived for, using an ensemble of GNSS-
mareografi positioned tide gauges.
The measurements and all corrections to be made to them are collected
into a “geophysical data record” (GDR), one per observation epoch. The
files built this way are distributed to researchers. This allows all kind of
experimentation; for example, the replacement of a correction by one
calculated from improved models.

^ 13.2 Crossover adjustment

When a satellite orbits the Earth over months or years, thousands of
points are formed where the tracks cross each other. If we assume that
the sea level is the same for both satellite overflights, then this forms a
condition that can be used to adjust away orbit errors.

í Õ ! ¤. û
 Crossover adjustment
347
The observation equation is

s = h − N − H − ϵ + n = h0 + ∆h − N − H − ϵ + n,

in which s is the altimetric measurement of the height of the sea surface,

h the actual, and h0 the calculated height of the satellite above the
reference ellipsoid. N is the geoid height, H is the sea-surface topography, vertaus-
i. e., the permanent deviation of the sea surface from an equipotential ellipsoidi
surface, ∆h is the orbit-error correction, ϵ is the residual variation of the meritopografia
sea surface, the variation remaining after correcting for the tides and
other effects that can be modelled, and n is the random uncertainty, or
noise, in the radar altimetry observations.
From this we obtain in the crossing point of tracks i and j:
def (︁ i i
)︁ (︂ j j
)︂ (︁ )︁
ℓk = s − h0 − s − h0 = (∆hi − ∆hj ) − (ϵi − ϵj ) + ni − nj .

This is the observation equation of crossover adjustment. Here we see

the complication that in both, sea-surface residual variation and orbit
corrections come along in the equation in the same way. They cannot
be separately determined by crossover adjustment.
If we forget for now the sea-surface residual variation — or assume
that it behaves randomly, in other words it is part of the noise n — we
may write more simply
def (︁ )︁
ℓk = ∆hi − ∆hj + nk , in which nk = ni − nj − (ϵi − ϵj ) .

The index k counts crossover points, the indices i, j count tracks.

Next, we choose a suitable model for the satellite orbit error. The
simplest choice, sufficient for a small area, is the assumption that the
orbit correction is a constant for each track. See a simple example,
figure 13.3.

^ 13.2.1 A simple example

In the figure we have three tracks and two crossing points. The
observation equations, which describe the discrepancies in the known

í Õ ! ¤. û
348 Satellite altimetry and satellite gravity missions

∆h2 ∆h3

∆h2 − ∆h3
∆h1
∆h1 − ∆h3

Crossover 1

Crossover 2

2
1

^ Figure 13.3. Simple crossover geometry.

crossover points as functions of the orbit corrections, are

ℓ1 = ∆h2 − ∆h3 + n1 ,
ℓ2 = ∆h1 − ∆h3 + n2 ,

or in matric form3
x
ℓ A ⏟⎡ ⏞⏞ ⎤⏟ n
⏟ ⏞⏞ ]︄⏟ ⏟[︄
[︄ ⏞⏞ ]︄⏟ ∆h1 ⏟[︄ ⏞⏞ ]︄⏟
ℓ1 0 1 −1 n1
= ⎣ ∆h2 ⎦ + , (13.1)
⎢ ⎥
ℓ2 1 0 −1 n2
∆h3

í Õ ! ¤. û
Crossover adjustment
349
symbolically
ℓ = Ax + n.
If one now tries to calculate the solution using ordinary least-squares,
)︁−1 T
x = AT A
(︁
ˆ︁ A ℓ,

this will not work. The normal matrix AT A is singular (check!). This
makes sense, as one can move the whole track network up or down
without the observations ℓk changing. No unique solution can be found
for such a system.
Finding a solution requires that something must be fixed: for example,
one track — or, more democratically, the mean level of all tracks. This
fixing is achieved by adding the following “observation equation”:
[︂ ]︂
def
ℓ3 = 0 = c c c · x, (13.2)

in which c is some suitable constant. Then, matrix A becomes

⎡ ⎤
0 1 −1
A = ⎣ 1 0 −1 ⎦ ,
⎢ ⎥

c c c

and the least-squares solution

⎡ ⎤ ⎡ ⎤
∆h
ˆ︂ 1 ℓ1
⎢ ˆ︂ ⎥ (︁ T )︁−1 T )︁−1 T ⎢
A ℓ = AT A
(︁
x = ⎣ ∆h 2 ⎦ = A A A ⎣ ℓ2 ⎦ ,
ˆ︁ ⎥

∆h
ˆ︂ 3 0

where the matrix inversion is now possible. In this particular case,

x = A−1 ℓ will give the same solution, as A is square and invertible:
ˆ︁
(︁ T )︁−1 T (︁ )︁−1 T (︂(︁ )︁−1 )︂
A A A ℓ = A−1 AT A ℓ = A−1 AT AT ℓ = A−1 ℓ.

3 Notethe similarity with the observation equations for levelling! Instead of bench-
marks, we have tracks, instead of levelling lines, crossover points.

í Õ ! ¤. û
350 Satellite altimetry and satellite gravity missions

Now the symbolic algebra system maxima (SourceForge, Maxima) — or

brute-force calculation — gives the readily verified inverse
⎡ ⎤−1 ⎛⎡ ⎤⎡ ⎤⎞−1
0 1 −1 1 0 1 −1
A−1 = ⎣ 1 0 −1 ⎦ = ⎝⎣ 1 ⎦ ⎣ 1 0 −1 ⎦⎠ =
⎢ ⎥ ⎜⎢ ⎥⎢ ⎥⎟

c c c c 1 1 1
⎡ ⎤−1 ⎡ ⎤−1
0 1 −1 1
= ⎣ 1 0 −1 ⎦ ⎣ 1 ⎦ =
⎢ ⎥ ⎢ ⎥

1 1 1 c
⎡ ⎤⎡ ⎤ ⎡ /︁ ⎤
−1 2 1 1 −1 2 1 c
1⎢ ⎥ 1⎢ 1 c ⎥
/︁
= ⎣ 2 −1 1 ⎦ ⎣ 1 ⎦ = 3 ⎣ 2 −1 ⎦,
⎥⎢
3
1 c 1 c
/︁ /︁
−1 −1 1 −1 −1
and the solution is
⎡ ⎤ ⎡ /︁ ⎤ ⎡ ⎤
∆h
ˆ︂ 1 −1 2 1 c ℓ1
1⎢
⎣ ∆h2 ⎦ = A−1 ℓ = 3 ⎣ 2 −1 1 c ⎦ ⎣ ℓ2 ⎦ =
⎢ ˆ︂ ⎥ /︁ ⎥ ⎢ ⎥

−1 −1 1 c
/︁
∆h
ˆ︂ 3 0 ⎡ ⎤
−1 2 [︄ ]︄
1⎢ ℓ1
= ⎣ 2 −1 ⎦ ,
⎥
3 ℓ2
−1 −1
from which c has vanished.
Another way to look at this is to first write the observation equations
13.1 and 13.2 together as
ℓ A x n
⏟ ⏞⏞ ⎤⏟ ⏟⎡
⎡ ⏞⏞ ⎤⏟⏟⎡ ⏞⏞ ⎤⏟ ⏟⎡ ⏞⏞ ⎤⏟
ℓ1 0 1 −1 ∆h1 n1
⎣ ℓ2 ⎦ = ⎣ 1 0 −1 ⎦⎣ ∆h2 ⎦ + ⎣ n2 ⎦,
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥

0 c c c ∆h3 0
and then multiply the left-hand side and both terms on the right with
the diagonal matrix ⎡ ⎤
1 0 0
def ⎢
D = ⎣ 0 1 0 ⎦.
⎥

0 0 1 c
/︁

í Õ ! ¤. û
 Crossover adjustment
351
The result is
Dℓ DA Dn
⏟⎡ ⏞⏞ ⎤⏟ ⎡⏟ ⏞⏞ ⎤⏟ ⎡ ⎤ ⏟⎡ ⏞⏞ ⎤⏟
ℓ1 0 1 −1 ∆h1 n1
⎣ ℓ2 ⎦ = ⎣ 1 0 −1 ⎦ ⎣ ∆h2 ⎦ + ⎣ n2 ⎦,
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥

0 1 1 1 ∆h3 0

from which c has also vanished.

The principle applies generally:

Minimal constraints added to observation equations with a datum

defect do not essentially change the solution.

^ 13.2.2 A more advanced orbit correction model

A more advanced representation of orbit corrections more suitable for
use in a larger area, is a linear function:

∆h = a + bτ,

where the parameter τ is the location along the track reckoned from its
starting point. The dimension of this location can be time (seconds) or
distance (degrees or kilometres). Now, the set of observation equations
for the situation described above is
x
⏟ ⏞⏞ ⏟
⎡ ⎤
a1
ℓ A n
⏟[︄ ⏞⏞ ]︄⏟ ⏟[︄ ⏞⏞ ]︄⏟⎢ ⏟[︄ ⏞⏞ ]︄⏟
⎢ b ⎥
1 ⎥
1 τ21 −1 −τ31 ⎢
⎢ ⎥
ℓ1 0 0 ⎢ a2 ⎥ + n1 .
⎥
=
ℓ2 1 τ12 3
0 0 −1 −τ2 ⎢ b2 ⎥⎢
⎥ n2
⎢ ⎥
⎣ a3 ⎦
b3

The design matrix A contains, besides the values 1 and −1, also values
±τik , in which i is the number of the track and k that of the crossover
point. These values are computable when the geometry of the tracks is
known.

í Õ ! ¤. û
352 Satellite altimetry and satellite gravity missions

Now there are two unknowns for every track, a and b, a constant and
a trend. Of course also this system will prove to be singular. Removing
the singularity can be done by fixing all three parameters b and one
parameter a.4
The phenomenon that no solution can be found unless something
is fixed is called a datum defect. Fixing something suitable will define
a certain datum. Between two different datums exists a transformation
formula: in the case of one orbit correction parameter per track, this
transformation is a simple parallel shift or translation of all tracks up or
down.
The situation is somewhat similar to when one is defining a height
korkeus- or vertical reference system for a country. One needs to fix one point,
järjestelmä for example Helsinki harbour. If alternatively one fixes another point,
for example Turku harbour, the result is another datum, in which all
height values differ from the corresponding ones in the first datum by a
certain fixed amount.
The argument continues to hold if there is a large number of tracks:
say, ten north-going and ten south-going tracks, crossing at 10×10
crossover points. Here, for two parameters per track, we would have 40
unknowns and no less than 100 observations. Still, we must constrain the
absolute level and the various trends and possible other deformations
of the whole network of tracks. A simple approach is to attach a priori
uncertainties to the unknowns ai , bi to be derived, for example, from the
pienimmän known uncertainties of the orbit prediction available. The least-squares
neliösumman adjustment equation then becomes
tasoitus
)︁−1 T
x = AT A + σ2 Σ−1
(︁
ˆ︁ A ℓ,

4 Inorder to understand this, build, say, a three-track “wire-frame model” from pieces
of iron wire, tied together by pieces of string at the crossover points. The crossover
conditions do not in any way fix the values of the trends b, and the whole absolute
level of the frame continues to be unconstrained.

í Õ ! ¤. û
 Crossover adjustment
353

^ Figure 13.4. Example of track geometry of satellite altimetry.

in which Σ is the diagonal matrix containing the a priori variances5

σ2a,i and σ2b,i of the parameters of each track i. This is referred to as
Tikhonov6 regularisation.

^ 13.2.3 Another example

In diagram 13.4 describing a satellite altimetry geometry, there are 16
crossover points. We attempt a crossover adjustment.
Questions
1. If the orbit correction ∆h of each satellite track is described
by a model with a single bias term, how many unknowns

5σ is the mean error of unit weight, in this case the mean error of a crossover
observation.
6 Andrey Nikolayevich Tikhonov (1906–1993) was a Russian mathematician and
geophysicist.

í Õ ! ¤. û
354 Satellite altimetry and satellite gravity missions

are there?
2. If we have available 16 “observations” or crossover differ-
ences, how many of them are redundant?
3. Is it geometrically possible to calculate this network?
4. If we fix one track in advance (a priori information), how
many redundant observations are there? Can this network
be calculated?
5. If every track has two unknowns, a bias as well as an error
growing linearly with time or trend, what then needs to be
fixed in order to make the network calculable? How many
redundancies are there then?
6. If, in case 3, we fix one track, which one would you choose?
Propose alternatively a solution where you do not have to
make a choice.
Answers
1. As many as there are tracks: 8.
2. 16 − 8 = 8.
3. No, because the absolute level of the whole network is
indeterminate.
4. 16 − (8 − 1) = 9. Now the network can be calculated.
5. If we assume that the tracks are “straight” in (x, y) co-
ordinates, then the set of allowable transformations on the
whole network is

∆h = a00 + a10 x + a01 y + a11 xy,

having four degrees of freedom. So, one needs to fix for

example one bias and three trends, not all north- or all south-
going. Then there are 16 − (8 − 4) = 12 redundancies.
6. Any such choice would be arbitrary. Rather use the method
described above instead, Tikhonov regularisation.

í Õ ! ¤. û
 Choice of satellite orbit
355
^ 13.2.4 Global crossover adjustment
In a global crossover adjustment, often a still more sophisticated model
is used,
∆h = a + b sin τ + c cos τ, (13.3)
in which now τ is an angular measure, for example the place along the
track measured from the last south-north equator crossing or ascending
node. See Schrama (1989), where this problem is treated more extensively.
In this model, a represents the size of the orbit, while b and c denote
the offset of the centre of the orbit from the geocentre. This model
is three-dimensional: the orbital arcs with their crossovers form a
spherical network surrounding the Earth. The degrees of freedom left
by the crossover conditions are now the size of this sphere and the offset
of its centre from the geocentre:

∆h = a0 + a1 cos ϕ cos λ + a2 cos ϕ sin λ + a3 sin ϕ (13.4)

with four degrees of freedom.7

^ 13.3 Choice of satellite orbit

In choosing a satellite orbit, Kepler’s orbital laws are central. Kepler’s
third law says
GM⊕ P2 = 4π2 a3 , (13.5)
in which a = a⊕ + h is the satellite orbit’s semimajor axis — the mean
distance from the geocentre — while h is called the satellite’s mean
height. P is the orbital period; a⊕ is the equatorial radius of the Earth
ellipsoid.
From equation 13.5 one can already infer that using satellite observa-
tions one can precisely determine the quantity GM⊕ , the mass of the

7 One could argue that, in equation 13.3, the parameter a should be zero, as Kepler’s
third law allows a very precise determination of the orbital size, see section 13.3. Then,
also a0 = 0 in equation 13.4.

í Õ ! ¤. û
356 Satellite altimetry and satellite gravity missions

z
Earth’s
rotation E
Perigee Nodal line
axis
y
Nodal line
ν ω
ν
Satellite Apogee Perigee
i a
Ω ̇ b X
Ω ω
θ
ine

Earth
al l

ea
rotation
sid

x
Ap

X (Greenwich)
x ′,
vernal equinox
Apogee Ascending node

Figure 13.5. Kepler’s orbital elements: a — semimajor axis, e — eccentricity,

i — inclination, Ω — right ascension (celestial longitude) of
the ascending node, ω — argument of perigee, and ν — true
^ anomaly.

Earth multiplied by Newton’s universal gravitational constant.8 The

period P can be precisely determined from long observation series, and
the size of the orbit a can also be obtained very precisely, for example
from satellite laser ranging (SLR) observations. For this purpose the
well-known LAGEOS (Laser Geodynamic Satellite) satellites (1976-039A
and 1992-070B), which orbit the Earth at a height of 6000 km, have
been used. Ranges are nowadays obtained with better than centimetre
precision.
The orbits of altimetric satellites are chosen to be much lower, as is

8 Thisis why it is said that Henry Cavendish was the first to “weigh the Earth”. . .
Determining GM⊕ was already straightforward back then using the orbital motion of
the Moon, or even gravity on the Earth’s surface. The challenge was separating G and
the mass of the Earth M⊕ from each other, obtaining the latter in ordinary units of
mass.

í Õ ! ¤. û
Choice of satellite orbit
357
seen from table 13.1 at the start of the chapter. The height is fine-tuned
using on-board thrusters, so that the satellite passes over the same rakettimoottori
place, for example once a day, after 14 orbital periods. Alternatively one
chooses an orbit that flies over the same place every third, seventeenth,
168th day. . . This is called the repeat period.
The choice of the repeat period depends on the mission objective:
◦ If one wishes to study the precise shape of the mean sea surface,
one chooses a long repeat period, in order to get the tracks as close
together as possible on the Earth’s surface.
◦ If one wishes to study the variability of the sea surface, one chooses
an orbit that returns to the same location after a short time interval.
Then, the grid of tracks on the Earth’s surface will be sparser.
Parameters describing the figure of the Earth also affect satellite motion,
for example the quantity J2 , the dynamic flattening, having a value of
J2 = 1082.63 · 10−6 . It is the largest of the many spherical-harmonic pallofunktio-
coefficients that together represent the figure of the Earth and that kerroin
affect satellite orbits. In the case of J2 , the effect is that the plane of the
satellite orbit rotates at a certain angular rate around the Earth’s rotation
axis: orbital precession. This makes the satellite, if it flies over the same
location the next day, do so several minutes earlier. For a circular orbit
of radius a, the equation is
√︃
2
dΩ 3 GM⊕ a⊕
=− J cos i,
dt 2 a3 a2 2
in which a⊕ is the equatorial radius of the Earth reference ellipsoid, M⊕ vertaus-
the mass of the Earth, and i the inclination of the orbital plane relative ellipsoidi
to the equator.
Substituting numerical values into this yields
dΩ cos i
= −1.318 95 · 1018 m3.5 s−1 · ,
dt (a⊕ + h)3.5
in which h is the mean height of the satellite orbit, conventionally above
a sphere of size equatorial radius a⊕ . If we substitute into this, say, the

í Õ ! ¤. û
358 Satellite altimetry and satellite gravity missions

Attraction Solar
caused by Satellite
(apparent)
Earth’s orbital
daily motion
flattening motion

X
X

Daily motion of satellite

orbit ascending node

^ Figure 13.6. The mechanism of a Sun-stationary orbit.

satellite height h = 800 km (and use a⊕ = 6 378 137 m), we obtain

dΩ
(︂ /︂ )︂
◦
= − 1.331 03 · 10−6 rad/s · cos i = − 6 . 589 day · cos i.
dt
For practical reasons — solar panels! — we often choose the satellite
orbit such that the orbital plane turns along with the annual apparent
◦/︁ ◦
motion of the Sun, 360 365.25 days = 0 . 9856 day . See figure 13.6.
/︁

If the inclination i is chosen in the range 96◦ –102◦ , depending on

the orbital height, then the Earth’s dynamic flattening J2 will cause
just the suitable rotational motion of the orbital plane (“no-shadow /
Sun-synchronous / Sun-stationary orbit”),9 see figure 13.7.
An orbit with an inclination i > 90◦ is called a retrograde orbit: the
satellite is moving westwards in longitude, opposite to the direction
of the Earth’s rotation, which is eastwards. The orbital inclination i,
or for a retrograde orbit, its supplement 180◦ − i, is also the greatest
northern or southern latitude a satellite can fly over. This means that,

9 If
the height of the satellite orbit is less than 1400 km, it is not completely no-shadow.
Either in midwinter or in midsummer, the satellite will fly through the Earth’s shadow.

í Õ ! ¤. û
 Choice of satellite orbit
359
Spring

Summer Winter

Autumn

^ Figure 13.7. Geometry of a “no-shadow” orbit. Season names are boreal.

unless the inclination is precisely 90◦ , there will be areas around both
poles that the satellite will never overfly: the “polar holes”, or “poles of
ignorance”.
A drawback of a Sun-stationary orbit is that the altimetric observations
are always made at the same local time of day. For example, the diurnal
and semidiurnal tides caused by the Sun will always have the same
phase angle, and thus they cannot be observed with a satellite in this
type of orbit (“resonance”). Therefore, the oceanographic satellite
TOPEX/Poseidon, and the follow-up Jason satellites, were placed in
non-Sun-stationary orbits.

^ 13.3.1 Example
A satellite moves in a Sun-stationary orbit, in other words, always, day
after day, flies over the same latitude at the same local (mean) solar
time.

í Õ ! ¤. û
360 Satellite altimetry and satellite gravity missions

i
3 2 1 θ̇

Figure 13.8. A satellite in a retrograde orbit around the rotating Earth, crossing
the equator south to north three successive times. The angle
between the orbit and the equator, the inclination i, or for a
retrograde orbit, its supplement 180◦ − i, is also the highest
northern or southern latitude that the satellite can fly over. The
^ unreachable “polar holes” are indicated by blue dashed lines.

Questions
1. What is the period of the satellite if it always flies again over
the same spot after 14 revolutions?
2. The same question if the satellite always flies over the same
spot after 43 revolutions (3 days)?
3. And after 502 revolutions (35 days)?
4. What is the height of the satellite in a “three-day orbit”? Use
Kepler’s third law, equation 13.5. GM⊕ = 3 986 005·108 m3/s2 ,
and the height of the satellite is h = a − a⊕ , with a⊕ =
6 378 137 m.

í Õ ! ¤. û
Choice of satellite orbit
361
5. What is the satellite height in a “35-day orbit”? And the
height difference from the previous question?
6. What is, for the three-day orbit, the mean separation between
north-going orbital tracks (so, at what level of detail is the
altimeter able to image the sea surface!)?
7. The same question for a 35-day orbit.
8. Questions for reflection:
(a) For what purpose would you use a 35-day orbit, for
what purpose a three-day orbit?
(b) Would it be possible, or easy, to fly both orbits with the
same satellite (see question 5)?
Answers
1. The satellite completes 14 orbits per day, i. e., per 1440
minutes: P = 1440 min 14 = 102.857 min.
/︁

2. The satellite completes 43 orbits in three days, i. e., per

3 × 1440 minutes: P = 3 × 1440 min 43 = 100.465 min.
/︁

3. The satellite completes 502 orbits in 35 days, i. e., per 35×1440

minutes: P = 35 × 1440 min 502 = 100.398 min.
/︁

4. Execute the octave code in tableau 13.2. The result is

780.604 km.
5. The same code, with P=100.398*60, yields 777.421 km. The
difference from the previous is 3.183 km.
6. There are 43 orbits with different ground tracks. That
◦/︁
means a separation of 360 43 = 8◦. 372. At the equator this
is 40 000 km 43 = 930 km. The distance is shorter at higher
/︁

latitudes.
◦/︁
7. 360 502 = 0◦. 717, or 40 000 km 502 = 80 km.
/︁

8.
(a) The 35-day orbit would be excellent for detailed map-

í Õ ! ¤. û
362 Satellite altimetry and satellite gravity missions

^ Tableau 13.2. Calculating the height of a satellite from its period.

format long
GM=3986005e8;
ae=6378137;
P=100.465*60; % seconds
fac=4*pi*pi; % four pi square
a=(GM*P*P/fac)^0.33333333;
h = a - ae;
printf(’\n\nOrbital height: %8.3f km.\n’, h/1000);

ping. The three-day orbit would be able to see, for

example, tides or weather-related phenomena, albeit at
poor spatial resolution.
(b) The difference in height being only 3 km and in period
4 s, the change in orbit between the two repeat periods
should be easily within reach of even small on-board
rakettimoottori thrusters. So, yes.

^ 13.4 In-flight calibration

The highly precise, GNSS-positioned satellite radar altimeters in use
today require proper calibration. The technique of choice for this is
in-flight calibration, using an ocean area — or sometimes a lake area —
the geocentric location of the water surface of which is known thanks
mareografi to surrounding GNSS-positioned tide gauges combined with a precise
geoid model of the area. An example of such measurements is Vu et al.
(2018).
One reason for in-flight calibration is the circumstance that radar
altimeters not only have an unknown zero offset — due to the not
precisely known signal paths through the electronic circuitry — but this
käynti offset may slowly change or drift over time, and may be temperature-

í Õ ! ¤. û
 Retracking
363
Travel time

Half-
height
rule

Sent pulse Received pulse

Figure 13.9. Analysing the altimeter return pulse. The classical return pulse
^ time measurement uses the “half-height point”.

dependent.

^ 13.5 Retracking
The results of a satellite altimetry mission are published already during
flight in the form of a geophysical data record (GDR) file, containing
everything related to the measurement, such as atmospheric correction
terms, tidal corrections, and sea-state parameters.
It is common practice today to re-process older altimetry measure-
ments, applying improved methodologies in order to extract additional
useful information. The complete return pulse is analysed again in a
method called retracking (Altimetry, Retracking).
The method of analysis uses the point on the leading edge of the
return pulse which is at half-height from the maximum value of the
pulse. This is according to experience a good way to get the travel
time associated with the point at the centre of the footprint, directly
underneath the satellite. In the back part of the pulse are reflections
from the further-away peripheral areas of the footprint.
There are three situations where the automatic analysis technique
applied during flight does not work properly, and a more careful a
posteriori analysis of the pulse is worthwhile:
◦ Archipelagos like Indonesia or Åland. Here it may happen, for
example, that the centre point of the footprint is on land. Then,

í Õ ! ¤. û
364 Satellite altimetry and satellite gravity missions

the first strong bounces will come under an angle from the nearest
coast. A precise coastline mask is then essential for processing.
But already over open water close to coastlines the return pulse
will be distorted.
◦ Sea ice areas in the Arctic and Antarctic Oceans. Bounces may
come from the surface of the sea ice, in which case one should
varalaita consider freeboard in the processing, i. e., how much the ice sticks
out of the water.
mannerjäätikkö ◦ Over continental ice sheets. Here, the shape of the return puls
will be very different from that over open water. Furthermore, the
travel time of the return pulse varies rapidly as the satellite flies
on, and the reception window cannot keep track.10
In these cases the traditional real-time processing on-board produces
erroneous measurements, or no measurements at all. With retracking,
such measurements have been saved, and the area covered by altimet-
ric measurements has been extended, especially into the Arctic and
Antarctic areas.
Freeboard is an important quantity in determining the thickness of
the ice. As the density of ice is about 920 kg/m3 and the density of sea
water about 1030 kg/m3 , the ice thickness is about 8× freeboard.11 If
there is additional remote-sensing data on the area of ice cover, one can
calculate the total volume and mass of sea ice.
The Arctic ice cover has diminished radically over recent decades.
The most radical reduction has been that of ice volume, see figure
13.10. In addition to surface area, thickness is also decreasing: of the
multi-year, thicker ice, a large part has already vanished.

10 Thenewest satellites such as Sentinel-3 use a digital terrain model when not over
the open ocean.
11 Assuming that there is no snow on the ice. Also, ice density varies, and differs
between one-year and multi-year ice.

í Õ ! ¤. û
 Oceanographic research using satellite altimetry
365
40

30
Ice volume (1000 km3 )

20

10

Year
0
1970 1980 1990 2000 2010 2020

Figure 13.10. Ice volume on the Arctic Ocean. PIOMAS; Schweiger et al.
(2011).
^

^ 13.6 Oceanographic research using satellite altimetry

The first geodetic application of satellite altimetry was geoid determi-
nation. Altimetric geoid determination works only if we assume that
the sea surface
◦ is constant in time
◦ coincides with a level surface, i. e., is the same as the geoid.
In practice, however, the ocean surface is variable in time and is also
not a level surface. Therefore, other approaches have been developed.
◦ Sea-surface variability can be studied by satellite altimetry using
three methods:
– Repeat tracks from the same satellite. The tracks can be
stacked and adjusted together using a simple orbit-error
correction model, and the remaining per-track residuals tell

í Õ ! ¤. û
366 Satellite altimetry and satellite gravity missions

something (but not everything!) about the variability of the

sea surface.
– The crossovers may also provide information on sea-surface
variability. When the sea surface varies, the results from the
crossover adjustment will get poorer: the root-mean-square
a posteriori (after calculation) crossover difference will grow.
Using this method to actually study sea-surface variability is
more difficult: it is mostly just able to establish that it exists,
and estimate its magnitude.
– Nowadays altimetric satellites always carry a GNSS position-
ing instrument, providing the absolute geocentric location of
the microwave radar device at the moment of measurement.
With it, the variations of sea level can be monitored by di-
rect measurement, assuming that both temporal and spatial
measurement densities are sufficient.
◦ The deviations of sea level from a level surface — the geoid — can
be studied only if we have access to independent information on the
true geoid surface. If dense, high-quality gravity measurements
are available for an area, these may be used to estimate the geoid,
meritopografia and after that one may calculate the sea-surface topography.
Collecting sufficiently precise and dense gravimetric data is pos-
sible with a sea gravimeter or airborne gravimetry. Measurement
with a special satellite (gravitational gradiometry, GOCE satellite)
has also long been planned and has finally been realised, see
section 13.7.

^ 13.7 Satellite gravity missions

During the early years of the 21st century, three satellite missions were
launched to investigate the fine structure of the Earth’s gravity field
or geopotential; in other words, to determine a global high-resolution
model of the geoid.

í Õ ! ¤. û
Satellite gravity missions
367
CHAMP (Challenging Minisatellite Payload for Geophysical Research
and Applications, 2000-039B) was a German satellite project un-
der the auspices of the German Research Centre for Geosciences
GFZ. The satellite was launched into orbit from Plesetsk, Russia,
in 2000. The orbit height was initially 454 km, coming down
over the time of the mission to ∼ 300 km due to atmospheric
drag. The orbital inclination was 87◦ . On 19 September 2010,
the satellite returned into the atmosphere. Project description:
CHAMP Mission.
CHAMP contained a GPS receiver in order to determine the satellite
location in space x(t) for any moment in time t. From successive
satellite locations one may calculate the geometric acceleration
a(t) by differentiation:
d2
a(t) = x(t).
dt2
The differentiation is done numerically as presented in the part
on airborne gravimetry, equation 11.7.
The satellite also contained an accelerometer, which eliminated kiihtyvyys-
the satellite’s accelerations caused by the atmosphere’s aero- mittari
dynamic forces, the deviations from free-fall motion. Then, only
the accelerations caused by the Earth’s gravitational field remain,
from which a precise global geopotential or geoid model may be
calculated using the techniques described earlier.
A number of global geopotential models based on CHAMP data
have been calculated and published.
GRACE (Gravity Recovery And Climate Experiment Mission, 2002-012
A and B) measured temporal changes in the Earth’s gravity field
extremely precisely, but at a rather crude geographic resolution.
These temporal changes are caused by motions in the Earth’s
“blue film”: her atmosphere and hydrosphere. The quantity
measured is also called the “sea-floor pressure”, a somewhat
surprising expression, until one sees that it really represents the

í Õ ! ¤. û
368 Satellite altimetry and satellite gravity missions

Magnetometer GPS-2 GPS-3

GPS-4
beam GPS-1

GPS antenna
Acceleration vector

Nom
inal
orbi
True t X
“CHAMP” orbit

Solar
cells Earth internal mass
density variations
(example)
Figure 13.11. Determining the Earth’s gravity field from GPS orbital tracking
^ of a low flying satellite.

total mass of a column of air and water.

The effective time resolution was one complete mapping of the
globe every month. Project description: GRACE Mission. The
project was a collaborative American-German undertaking under
the leadership of the Center for Space Research, University of
Texas at Austin.
GRACE was a satellite pair (“Tom and Jerry”): the satellites flew in
the same orbit in a tandem configuration at initially about 500 km
height, at an inter-satellite separation of 220 km. The orbital
inclination was 89◦ , so the orbit was almost polar, providing
complete global coverage. The changes in distance between the
satellites were measured by a microwave link at a precision of
kiihtyvyys- ±1 µm/s. Both satellites also carried sensitive accelerometers for
mittari measuring and eliminating the effect of atmospheric drag.
The measurement system was so sensitive that even the movement
of a water layer of one millimetre thickness could be noticed, as
long as the layer extended over an area the size of a continent
(some 1000 km).

í Õ ! ¤. û
 Satellite gravity missions
369
Difference between line-of-sight accelerations

Satel
ite 1 lite 2
Satell Distance 220 km
Precise ranging, wavelength 1.5 cm

m
450 k

Figure 13.12. The principle of the GRACE satellites: measuring the minute
variations in time of the gravity field using SST (satellite-to-
satellite tracking). The changes are due to mass shifts in the
“blue film” — the atmosphere and hydrosphere — and expressed
^ as variations in “total sea-floor pressure” (↓).

The published results show impressively, for example, the wet

and dry monsoons, seasonal variations in opposite phases in
the northern and southern hemispheres, in the great tropical
river basins: Amazonas, Congo, the Mekong, India, Indonesia. . .
GRACE Mission, hydrology.
The mission ended in 2017 after 15 years, three times the planned
mission duration. A GRACE follow-on mission was launched in
2018, GRACE Follow-On Mission.
GOCE (2009-013A, Geopotential and Steady-state Ocean Circulation
Explorer) was the most ambitious of all the satellites. Built
by the European Space Agency ESA, the satellite was launched
successfully from Plesetsk in March 2009. The orbital height was
only 270–235 km during the mission and the satellite contained an
ionic rocket engine with a stock of propellant in order to maintain ajoaine

í Õ ! ¤. û
370 Satellite altimetry and satellite gravity missions

Figure 13.13. GRACE mission results: surface mass layer in centimetres of

^ water equivalent. Click for animation (e-book).

the orbit against atmospheric drag. The orbital inclination was

96◦. 7, so the orbit was Sun-stationary.12
GOCE carried a very sensitive gravitational gradiometer, a device
that measured precisely components of the gradient of the Earth’s
attraction, the dependence of components of the attraction vector
on the co-ordinates of place. The gradiometer consisted of six
kiihtyvyys- extremely sensitive, three-axes accelerometers mounted pairwise
mittari on a frame. The mission ended in 2013 and the satellite was seen
to burn up in the atmosphere on 11 November over the Falkland
Islands (Scuka, 2013).
Theoretical analysis has shown that a gravitational gradiometer
is the best way to measure the very local features of the Earth’s
gravity field, better than orbital tracking by GNSS. The smallest
details in the geoid map seen by GOCE are only 100 km in diameter,
and their precision is as good as ±2 cm.

12 Because of this inclination angle, there was a cap of radius 6◦. 7 at each pole within
which no measurements were obtained. Over recent years these “poles of ignorance”
have been gradually filled in by airborne gravimetry campaigns, for example Forsberg
et al. (2017).

í Õ ! ¤. û
 Satellite gravity missions
371
Measuring
acceleration
GOCE satellite
differences
1
5
4 X
3
6

2
Gradiometer
Accelerometer

Unknown
density variations

Figure 13.14. Determining the Earth’s gravity field with the gravitational
^ gradiometer on the GOCE satellite.

With a global geoid model this precise, we may calculate the

deviations of the sea surface from the geoid, an equipotential
or level surface, at the same precision. We saw that the true
location in space of the sea surface is obtained from satellite radar
altimetry, also at a few centimetres’ precision. This separation
between sea surface and equipotential surface can again be
inverted to ocean currents, see section 12.5 and figure 12.4. This is

í Õ ! ¤. û
372 Satellite altimetry and satellite gravity missions

the background for the name of the GOCE satellite.

^ Self-test questions
1. What is the footprint of a radar altimeter? How does it depend on
wave height?
varalaita 2. What is the freeboard of ocean ice? How can it be used to determine
the volume of the ice?
3. What three alternative models for the satellite orbit-error correc-
tion exist?
4. In satellite altimetry, what is a datum defect, and how can it be
fixed?
5. How can Kepler’s third law be used to determine the mean height
of a satellite orbit if the satellite’s period is given?
6. What is the repeat period of a satellite orbit?
7. What is J2 , and how does it affect the motion of a satellite?
8. What is a Sun-synchronous orbit, and why is it useful?
9. What is a retrograde orbit?
10. Why are the orbits of the TOPEX/Poseidon and Jason satellites not
Sun-synchronous?
11. In table 13.1 some satellites have a repeat period that is an integer
number of days, some satellites do not. What do the satellites
with integer repeat periods have in common?
12. With which three satellite altimetric methods can one study sea-
surface variability?
13. Three satellite missions have been launched so far to study the fine
structure of the Earth’s gravity field and its temporal variability.
Present them and the methods used by them.

í Õ ! ¤. û
 Exercise 13 – 1: Altimetry, crossover adjustment
373

^ Figure 13.15. Example of a satellite altimetric track geometry.

^ Exercise 13 – 1: Altimetry, crossover adjustment

it is given that there are two north-going satellite tracks and three
south-going ones. There are six crossover sites, see figure 13.15.
1. If the orbit-error corrections for every track are described as a
linear function of place:

∆h = a + bτ,

how many coefficients a and b are needed?

2. Write out the observation equations. The observations are the
crossover differences, the unknowns are the coefficients a and b
for the different tracks.
3. Can these observation equations give a unique solution? Why /
why not?

í Õ ! ¤. û
374 Satellite altimetry and satellite gravity missions

^ Exercise 13 – 2: Satellite orbit

A satellite moves in a Sun-synchronous orbit, where after 419 orbits
and 30 days, it again moves over exactly the same spot.
1. What is the period of the satellite?
2. How long is the distance (west to east), in kilometres, between the
north-going tracks at the equator?
3. What is the highest northern latitude that the satellite can fly over?
4. In what compass direction is the satellite flying at that point?

^ Exercise 13 – 3: Kepler’s third law

What is the satellite height h if the period is 98 minutes? Use Kepler’s
third law 13.5,
GM⊕ P2 = 4π2 a3 ,
GM⊕ = 3 986 005 · 108 m3/s2 , and the height of the satellite is h = a − a⊕ ,
in which a⊕ = 6 378 137 m.

í Õ ! ¤. û
^ Tides, the atmosphere, and Earth
crustal movements

^ 14.1 The theoretical tide

14
The tidal potential V can be written as follows:

GMR2 GMR2 (︁ 2
)︁
V= P2 (cos ζ) + · · · = 3 cos ζ − 1 + ··· ,
d3 2d3
in which d is the distance to either the Moon or the Sun, R is the radius
of the Earth, and ζ is the local geocentric zenith angle of the Sun or
Moon, i. e., the local zenith angle ζ ′ corrected for parallax, see figure
14.1. P2 (cos ζ) is the Legendre polynomial of degree two. GM is the
mass of the Sun or Moon multiplied by Newton’s gravitational constant.
In the case of the Sun and Moon, the extra terms (. . .) can be neglected,
because these are such remote bodies: d ≫ R.
The cosine rule on the sphere tells us that

cos ζ = sin ϕ sin δ + cos ϕ cos δ cos h,

in which ϕ is the latitude, δ is the declination1 of the Moon, and h is

the hour angle2 of the Moon.
According to the spherical-harmonic addition theorem (Wolfram Math- pallo-
funktioiden
summauslause
1 Thedeclination of a celestial body is its latitude on the celestial sphere, its angular
distance from the celestial equator (Wikipedia, Declination), in this case as seen from

– 375 –
376 Tides, the atmosphere, and Earth crustal movements
Lo
w
Hi tid
gh e
tid
e

X ζ ζ′

Hi
gh
Lo tid
w e
tid
e

Parallax

Figure 14.1. Theoretical tide. ζ ′ is the local zenith angle of the Moon (or Sun),
^ ζ the corresponding geocentric angle.

World, Spherical Harmonic Addition Theorem) we have now

Pn (cos ζ) = Pn (sin ϕ)Pn (sin δ) +

∑︂n
(n − m)!
+2 P (sin ϕ)Pnm (sin δ) cos mh,
(n + m)! nm
m=1

or for n = 2,

the geocentre.
2 The hour angle is the angle, or difference in longitude, between the meridian of the
Moon and the local meridian, measured along the celestial equator Wikipedia, Hour
angle, in this case as seen from the geocentre. It vanishes when the Moon is in upper
culmination, in the local meridian due south when seen from northern non-tropical
latitudes.

í Õ ! ¤. û
The theoretical tide
377
P2 (cos ζ) = P2 (sin ϕ)P2 (sin δ) +
+ 31 P21 (sin ϕ)P21 (sin δ) cos h + 1
P (sin ϕ)P22 (sin δ) cos 2h.
12 22

In this, according to table 3.2,

P21 (sin ϕ) = 3 sin ϕ cos ϕ, P21 (sin δ) = 3 sin δ cos δ,

2
P22 (sin ϕ) = 3 cos ϕ, P22 (sin δ) = 3 cos2 δ,

and we obtain

P2 (cos ζ) = P2 (sin ϕ)P2 (sin δ) +

+ 3 sin ϕ cos ϕ sin δ cos δ cos h + 43 cos2 ϕ cos2 δ cos 2h =
= 21 3 sin2 ϕ − 1 12 3 sin2 δ − 1 + 43 sin 2ϕ sin 2δ cos h +
(︁ )︁ (︁ )︁

+ 34 cos2 ϕ cos2 δ cos 2h.

From this
⎛(︁ )︁ ⎞
3 sin2 ϕ − 1 3 sin2 δ − 1 +
)︁ (︁
GMR2 ⎜
⎜ + 3 sin 2ϕ sin 2δ cos h + ⎟ .
⎟
V=
4d3 ⎝ ⎠
2 2
+ 3 cos ϕ cos δ cos 2h

This is the Laplace tidal decomposition equation.

It has three parts:
◦ A slowly varying part,

GMR2 (︁
3 sin2 ϕ − 1 3 sin2 δ − 1 ,
)︁ (︁ )︁
V1 = 3
4d
that still depends on the lunar declination δ and is therefore
periodic with a 14-day (half-month) period. Using spherical
trigonometry:

sin δ = sin ϵ sin ℓ

(︁ 1
=⇒ sin2 δ = sin2 ϵ sin2 ℓ = sin2 ϵ − 21 cos 2ℓ , (14.1)
)︁
2

í Õ ! ¤. û
378 Tides, the atmosphere, and Earth crustal movements

in which ℓ is the longitude of the Moon in its orbit, reckoned from

the ascending node (equator crossing), and ϵ is the inclination of
the Moon’s orbital plane with respect to the equator, on average
23◦. 5 but rather variable, between 18◦. 3 and 28◦. 6. Thus we obtain
GMR2 (︁ )︁ (︂ (︁ 1 1 )︂
2 2
)︁
V1 = 3 sin ϕ − 1 3 sin ϵ − cos 2ℓ − 1 ,
4d3 2 2

where we have used result 14.1. We split V1 = V1a + V1b into two
parts, a constant3 and a periodic or semi-monthly (“fortnightly”)
part:

GMR2 (︁ 2
)︁ (︁ 3 2
)︁
V1a = 3 sin ϕ − 1 sin ϵ − 1 , (14.2)
4d3 2

GMR2 (︁
3 sin2 ϕ − 1 23 sin2 ϵ cos 2ℓ .
)︁ (︁ )︁
V1b =− 3
4d

◦ In addition, we have a couple of terms in which the hour angle h

appears, periods roughly a day and roughly half a day:

GMR2
V2 = · 3 sin 2ϕ sin 2δ cos h,
4d3
GMR2
V3 = · 3 cos2 ϕ cos2 δ cos 2h.
4d3
In both, we have in addition to h, still δ as a “slow” variable. These
equations could be written out as the sums of various functions
of the longitude of the Moon ℓ.
Use basic trigonometry again, with equation 14.1:

cos2 δ = 1 − sin2 δ = 1 − sin2 ϵ sin2 ℓ =

(︁ 1
= 1 − sin2 ϵ− 12 cos 2ℓ ,
)︁
2

cos 2ℓ cos 2h = 21 cos(2ℓ + 2h) + cos(2ℓ − 2h) ,

(︁ )︁
√︂
sin 2δ = 2 sin δ cos δ = 2 sin2 δ 1 − sin2 δ =
(︁ )︁
√︂(︁
1
− 12 cos 2ℓ 1 − sin2 ϵ 21 − 12 cos 2ℓ ,
)︁ (︁ (︁ )︁)︁
= 2 sin ϵ 2

3 Not precisely, because ϵ is (slowly) time-dependent.

í Õ ! ¤. û
 The theoretical tide
379
^ Table 14.1. The various periods in the theoretical tide. The widely used
symbols were standardised by George Darwin.

Changing Period Darwin symbol

Name
function
Moon Sun Moon Sun

V1a - - - M0 S0 Permanent tide

V1b cos 2ℓ 14d 182d Mf a Ssab Declination tide
V2 cos h 24h 50m 24h K1 , O1 S1 , P1 Diurnal
V3 cos 2h 12h 25m 12h M2 S2 Semidiurnal

a Lunar fortnightly
b Solar semi-annual

leading to a trigonometric expansion in lunar longitude ℓ, and so

on. See for example Melchior’s4 famous book (1978).
The coefficient
3GMR2
def
D= , (14.3)
4d3
“Doodson’s5 constant” is often taken separately from the above equations.
The value for the Moon equals D = 26.8 cm × γ and for the Sun
12.3 cm × γ, with γ ≈ 9.81 m/s2 . See figure 14.2.
The periods are listed in table 14.1 with their Darwin6 symbols.
In practice, the diurnal and semidiurnal tides can be divided further
into many “spectral lines” close to each other, also because the lunar

4 PaulJacques Léon Camille baron Melchior (1925–2004) was an eminent Belgian

geophysicist and Earth tides researcher and founder of the Walferdange underground
laboratory for geodynamics in Luxembourg.
5 Arthur Thomas Doodson FRS (1890–1968) was a British oceanographer, a pioneer of
tidal theory, also involved in designing machines for computing the tides. He was
completely deaf.
6 Sir George Howard Darwin KCB FRS FRSE (1845–1912) was an English astronomer and

mathematician, son of Charles Darwin of Origin of Species fame.

í Õ ! ¤. û
380 Tides, the atmosphere, and Earth crustal movements

V1a , permanent V1b , fortnightly, ϵ = 23◦

0.2 0.15
80 80
60 0.1 60 0.1
40 0 40
Latitude ( ◦ )

20 −0.1 20 0.05
0 −0.2 0 0
−20 −0.3 −20
−40 −0.4 −40 −0.05
−60 −0.5 −60
−0.6 −0.1
−80 −80
Obliquity ϵ, 0◦ − 90◦ Lunar longitude, 0◦ − 360◦
X
V2 , diurnal, δ = 23◦ V3 , semidiurnal, δ = 23◦
80 80 0.6
60 0.4 60 0.4
40 0.2 40
Latitude ( ◦ )

0.2
20 20
0 0
0 0
−20 −0.2 −20 −0.2
−40 −0.4 −40 −0.4
−60 −60 −0.6
−80 −0.6
−80 −0.8
0 5 10 15 20 0 5 10 15 20
Hour angle Hour angle

Figure 14.2. The main components of the theoretical tide. These values must
^ still be multiplied by Doodson’s constant D.

orbit (like the Earth’s orbit) is an ellipse, not a circle.

^ 14.2 Deformation caused by the tidal force

The tidal force, or theoretical tide, of which we spoke above, is not the
same as the deformation it causes in the solid Earth. This deformation
will depend upon the elastic properties inside the Earth. These elastic

í Õ ! ¤. û
Deformation caused by the tidal force
381
properties are often characterised by elastic Love7 numbers (Love, 1909;
Melchior, 1978).
Let us first write the external or tidal potential V = V(ϕ, λ, r) in the
following way:

∑︂
∞ (︂ )︂
r n ∑︂
∞
V(ϕ, λ, r) = Vn (ϕ, λ) = V
˜︁n (ϕ, λ, r),
R
n=2 n=2

in which the index n denotes the spherical-harmonic degree number. Vn

˜︁n = r R n Vn the solid spherical harmonic asteosuus
(︁ /︁ )︁
is the degree constituent, V
of potential V for degree number n. avaruus-
Call the linear displacement of an element of matter of the solid Earth pallofunktio
in the radial direction, ur , in the north direction, uϕ , and in the east asteluku
direction, uλ . The following equations apply:

1 ∑︂ ∑︂
∞ ∞
ur (ϕ, λ, r) = γ Hn (r) V
˜︁n (ϕ, λ, r) = Hn (r) ζn (ϕ, λ, r),
n=2 n=2

1 ∑︂
∞
∂ ˜︁ ∑︂
∞
uϕ (ϕ, λ, r) = γ Ln (r) V n (ϕ, λ, r) = r Ln (r) ξn (ϕ, λ, r),
∂ϕ
n=2 n=2

1 ∑︂
∞
∂ ∑︂∞
uλ (ϕ, λ, r) = γ Ln (r) V
˜︁n (ϕ, λ, r) = r Ln (r) ηn (ϕ, λ, r).
cos ϕ ∂λ
n=2 n=2

All displacements are in length units. r is the distance from the geocentre.
It is assumed here that the Love numbers Hn and Ln depend only on
r, so that the elastic properties of the Earth are spherically symmetric.
The symbols ζn , ξn and ηn represent the effect of the tidal potential of
harmonic degree n on the level of an equipotential surface and on the
components of the direction of the plumb line. luotiviiva
The deformation of the Earth also causes a change, the “indirect effect” epäsuora
in addition to the Moon’s original tidal potential V, in the geopotential. vaikutus

7 Augustus Edward Hough Love FRS (1863–1940) was a British mathematician and
student of Earth elasticity.

í Õ ! ¤. û
382 Tides, the atmosphere, and Earth crustal movements

We write
∑︂
∞
δV(ϕ, λ, r) = Kn (r)V
˜︁n (ϕ, λ, r),
n=2

in which we already use a third type of Love numbers.

On the Earth’s surface r = R we make the following specialisation:
def (︁ )︁ def (︁ )︁ def (︁ )︁
hn = Hn R , ℓn = Ln R , kn = K n R . (14.4)

In practice, because of the large distances to Sun and Moon, the only
important part of the tidal potential V is the part for the degree number
n = 2, the “rugby-ball part” V˜︁2 .
The Love numbers will still depend on the frequency, i. e., on the tidal
period P:
(︁ )︁ (︁ )︁ (︁ )︁
hn = hn P , ℓn = ℓn P , kn = kn P .

The tides offer an excellent means of determining all these Love numbers
(︁ )︁ (︁ )︁ (︁ )︁
h2 P , ℓ2 P , and k2 P empirically, because, being periodic variations,
they cause Earth deformations at the same periods, but with differ-
ent amplitudes and phase angles. In this way we may determine at
least those Love numbers that correspond to periods occurring in the
theoretical tide.
The h and ℓ numbers are nowadays obtained for example by GNSS
positioning. The GNSS processing software contains a built-in reduc-
tion for this phenomenon. From gravity measurements one obtains
information on a certain linear combination of h and k, δ = 1 + h − 23 k:
lunar attraction changes gravity directly, vertical displacement changes
gravity though its gradient, and deformation of the Earth, the shifting
of masses, also changes gravity directly.
The long water-tube clinometer is also a useful research instrument,
like the tube of the Finnish Geodetic Institute that has long been in
use in the Tytyri limestone mine (Tytyri Mine Experience) in Lohja
(Kääriäinen and Ruotsalainen, 1989). A modern, improved version of

í Õ ! ¤. û
 The permanent part of the tide
383
this instrument is presented in Ruotsalainen (2017). The same applies
for sensitive clinometers in general, like the Verbaandert–Melchior
pendulum. A clinometer measures the changes in orientation between
the Earth’s crust and the local plumb line. This can again provide luotiviiva
information on a different linear combination of h and k, γ = 1 − h + k.
Measuring the absolute direction of the plumb line, for example with
a zenith tube, can again provide information on the linear combination
Λ = 1 − ℓ + k, but only after various reductions (Earth orientation
parameters like polar motion and variations in rotation rate), Vondrák
et al. (2010). The Love number ℓ8 comes in through the horizontal
displacement of the zenith tube, to a location where the plumb-line
direction is different.

^ 14.3 The permanent part of the tide

As shown above, the theoretical tide equation contains a constant part
that does not even vary in a long-period way. Of course the Earth
also responds to this part of the tidal force. However, because the
deformation is not periodic, it is not possible to measure it. And the
mechanical theory of the solid Earth, and our knowledge of the state
of matter inside the Earth, are just not good enough for a theoretical
calculation of the response.
For this reason the understanding is generally accepted that the effect
of the permanent part of the tide on the Earth’s state of deformation
should not be included in any tidal reduction (Ekman, 1992). Many
times, however, for example in the processing of GNSS observations or
in defining spherical-harmonic expansions of the Earth’s gravity field, pallofunktio-
the tidal reduction does include this term which it is theoretically and kehitelmä
practically impossible to know. See Poutanen et al. (1996).
More generally, the reduction of a geodetic quantity, for example the

8 Alsocalled the Shida number. Toshi Shida (1876–1936) was an eminent Japanese
Earth tide researcher.

í Õ ! ¤. û
384 Tides, the atmosphere, and Earth crustal movements

Earth permanent tidal deformation Tide-free Earth crust

Direct effect on geoid of Moon and Sun Mean (zero) Earth crust
Tide-free geoid
Effect of Earth permanent tidal
deformation (mass displacement) on Zero geoid
geoid Mean geoid
Reference ellipsoid

Figure 14.3. Conceptual diagram showing the constituents of the permanent

^ tide.

height of the geoid, for the permanent part of the tide can be carried
out in three different ways:
◦ No reduction whatever is made for the permanent part. The
quantity thus obtained is called the “mean geoid”. The surface
obtained is in the hydrodynamic sense an equilibrium surface,
suitable for use in oceanography.
◦ The direct effect of the gravitational field emanating from celestial
objects is removed in its entirety from the quantity, but the effect
of the Earth’s deformation caused by it is left uncorrected. The
quantity thus obtained is called the “zero geoid”.
◦ Both the gravitational effect of a celestial body, and the effect
of the deformation it causes, can be calculated according to a
certain deformation model (Love numbers), and removed. The
result obtained is called the “tide-free geoid”. Its problem is, as
explained, the empirical indeterminacy of the elasticity model
used.

í Õ ! ¤. û
 Tidal corrections between height systems
385
See figure 14.3. It is good to be critical and precisely analyse the way
in which the data reduction has been done!

^ 14.4 Tidal corrections between height systems

We see from equation 14.2 that, with ϵ = 23◦. 5, the permanent part of
the tidal potential is equal to

GMR2 (︁ 2
)︁ (︁ 3 2
)︁
Vperm = 3 sin ϕ − 1 sin ϵ − 1 ≈
4d3 2

3GMR2 (︁ 2
sin ϕ − 13 .
)︁
≈ −0.7615 · 3
4d
With the combined Doodson’s constant 14.3 for Sun and Moon equal to

3GM⊙ R2 3GM$ R2
D= + =
4d3⊙ 4d3$
= (12.3 cm + 26.8 cm) × γ = 39.1 cm × γ

we obtain
(︁ 1
− sin2 ϕ × γ.
)︁
Vperm = 29.77 cm × 3

We can express this, with Bruns equation 5.2, as a permanent tidal geoid
effect:
Nperm = 29.77 cm × 13 − sin2 ϕ .
(︁ )︁

From this, Nperm (0◦ ) = 9.92 cm on the equator, and Nperm (±90◦ ) =
−19.85 cm on the poles.
This, the geoid effect of the permanent part of the external potential
of the Sun and Moon is also equal to the difference between the mean
geoid and the zero geoid as defined above:
def (︁ 1
∆mean − sin2 ϕ .
)︁
zero N = Nmean − Nzero = 29.77 cm × 3

For heights H above sea level, with H = h − N, we have

def (︁ 1
∆mean − sin2 ϕ ,
)︁
zero H = Hmean − Hzero = −29.77 cm × 3

í Õ ! ¤. û
386 Tides, the atmosphere, and Earth crustal movements

and for two different latitudes ϕ1 and ϕ2 we have for the effect on the
height difference

∆mean mean
(︁ 2 2
)︁
zero H(ϕ2 ) − ∆zero H(ϕ1 ) = 29.77 cm × sin ϕ2 − sin ϕ1 .

This is the value to be added when going from a zero-geoid to a mean-

geoid height system, and subtracted when going from a mean-geoid to a
zero-geoid height system.
When the tide-free geoid and Earth crust enter into the picture, we
need values for the Love numbers h2 and k2 for the permanent tidal
deformation, expressing this deformation and its potential as fractional
parts of the original external tidal potential. As we have seen, these
numbers cannot be empirically determined. Values often used are
h2 ≈ 0.6, k2 ≈ 0.3. With this, the above equations apply with the
coefficient 29.77 cm multiplied by 1 − h2 + k2 . This yields
def (︁ 1
∆mean − sin2 ϕ ,
)︁
tidefree H = Hmean − Htidefree = −20.84 cm × 3

∆mean mean
(︁ 2 2
)︁
tidefree H(ϕ 2 ) − ∆tidefree H(ϕ 1 ) = 20.84 cm × sin ϕ 2 − sin ϕ 1 .
Any other correction equation can be obtained from these, like

∆zero zero
(︁ 2 2
)︁
tidefree H(ϕ2 ) − ∆tidefree H(ϕ1 ) = −8.93 cm × sin ϕ2 − sin ϕ1 .

^ 14.5 Loading of the Earth’s crust by sea and

atmosphere
In addition to the deformation caused by the tidal force, the Earth’s
crust also deforms due to the loading by sea and atmosphere. Especially
close to the coast, the tidal motion of the sea causes a multi-period
deformation that moves the Earth’s crust up and down by as much as
centimetres.
This phenomenon can be computationally modelled if the elastic
properties of the solid Earth, the tidal motion of the sea, and the precise
shape of the coastline are known. One known program for this purpose

í Õ ! ¤. û
 Self-test questions
387
is the package Eterna written by the German Hans-Georg Wenzel,9
which also has been used in Finland.
On the other hand, when such tools exist, tidal loading offers also
an excellent opportunity for studying precisely the very local elastic
properties of the Earth’s crust.
A registering gravimeter is generally used for measuring the defor-
mation. The Earth’s crust moves up and down elastically, which to
first order changes gravity in proportion to the free-air gradient value ilmagradientti
−0.3 mGal/m. For a description of the method, see Torge (1992) section
4.2.
The use of GNSS for measuring the ocean tidal loading has not yet
become common.
Like the ocean, the atmosphere also causes, through changes in air
pressure, varying deformations of the Earth’s crust. The phenomenon
is very small, at most a couple of centimetres. Gravity measurement is
not a very good way to study this phenomenon, because many more
local, often poorly known, factors affect local gravity. Measurement by
GNSS is promising but also challenging.

^ Self-test questions
1. Present in words the three components of the theoretical tide
produced by the Laplace decomposition method.
2. How may the slowly varying part of the theoretical tide be further
decomposed into two parts? Present the parts in words.
3. What are the declination and hour angle of a celestial body, for
example the Moon?
4. What is Doodson’s constant?
5. What do Love numbers express?

9 Hans-Georg Wenzel (1945–1999) was a German physical geodesist and geophysicist.

í Õ ! ¤. û
388 Tides, the atmosphere, and Earth crustal movements

6. Why is it not possible to empirically determine the deformation

caused by the permanent part of the tide?
7. Present the three different ways to take the permanent part of the
tide into account when defining the geoid.

^ Exercise 14 – 1: Tide
The equation for the permanent part of the tide is

GMR2 (︁
3 sin2 φ − 1 23 sin2 ϵ − 1 ,
)︁ (︁ )︁
W1a = 3
4d
in which φ is latitude and ϵ is the obliquity of the Earth’s axis of rotation,
currently about 23◦. 5.
1. For what value φ does the permanent part of the tide vanish?
What is your interpretation?
2. For what value ϵ does the permanent part of the tide vanish?
What is your interpretation?

í Õ ! ¤. û
^ Earth gravity field research

15
^ 15.1 Internationally
In the framework of the IAG, the International Association of Geodesy,
research into the Earth’s gravity field is currently the responsibility of
the International Gravity Field Service (IGFS). The IGFS was created in
2003 at the IUGG General Assembly in Sapporo, Japan, and it operates
under the IAG’s new Commission 2 “Gravity Field”. The United States
National Geospatial-Intelligence Agency (NGA) serves as its technical
centre.
An important and well-reputed IAG service is the International Gravity
Bureau, the BGI, Bureau Gravimétrique International located in Toulouse,
France (http://bgi.obs-mip.fr/). The bureau works as an international
broker to which countries can submit their gravimetric materials. If a
researcher needs gravimetric material from another country, for example
in order to do a geoid computation, they can request it from the BGI, who
will provide it with the permission of the country of origin, provided the
country of the researcher has in its turn submitted its own gravimetric
materials for BGI use.
The French state has invested significant funds into this vital interna-
tional activity.
Another important IAG service in this field is the ISG, the International
Service for the Geoid. It has in fact been operating since as early as

– 389 –
390 Earth gravity field research

1992 under the name International Geoid Service (IGeS), the executive
arm of the International Geoid Commission (IGeC). The ISG office is
located in Milan (http://www.isgeoid.polimi.it/), also with substantial
support by the Italian state. The task of this service is to support
geoid determination in different countries. Existing geoid solutions are
collected into a common database, and international research schools
are organised to develop awareness about and skills in the art of geoid
computation, especially in developing countries.
Both services, BGI and ISG, are under the auspices of the IGFS, as two
of the many official services of the IAG. Other IGFS services include the
International Center for Earth Tides (ICET), the International Center for
Global Earth Models (ICGEM), and the International Digital Elevation
Model Service (IDEMS).

^ 15.2 Europe
The EGU, the European Geosciences Union, operates in Europe, co-
ordinating many publication and meeting activities relating to the
gravity field and geoid. The EGU organises annual symposia, where
sessions are always also included on subjects related to the gravity
field and geoid. American scientists also participate. Conversely the
American Geophysical Union’s (AGU) fall and spring meetings1 are also
favoured by European researchers.
The Geodetic Institute (“Institut für Erdmessung”) of Leibniz University
in Hannover, Germany has acted since 1990 as the computing centre
of the International Geoid Commission’s (IGeC) Subcommission for
Europe, and produced high-quality European geoid models (Denker,
1998; European geoid calculations). The work continues since 2011
within the framework of the IAG Subcommission 2.4a Gravity and Geoid
in Europe.

1 Fall
(autumn) meetings are in San Francisco, spring meetings somewhere in the
world. The AGU, although American, is a very cosmopolitan player.

í Õ ! ¤. û
 The Nordic countries
391
^ 15.3 The Nordic countries
In the Nordic countries, important work is being co-ordinated by the
NKG, the Nordiska Kommissionen för Geodesi, and its Working Group for
Geoid and Height Systems. Its activities include geoid determination,
studying the preconditions for still more precise geoid models, new
levelling technologies, and the study of post-glacial land uplift.
The group has for a long time computed high-quality geoid models at
its computing centre in Copenhagen, the next to last one being NKG2004
(Forsberg and Kaminskis, 1996; Forsberg and Strykowski, 2010). The
newest model, NKG2015, is the result of calculations by the computing
centres of several countries, including Sweden and Estonia. It was
published in October 2016.

^ 15.4 Finland
In Finland the study of the Earth’s gravity field has mainly been in
the hands of the Finnish Geodetic Institute, founded in 1918, one year
after Finnish independence. The institute has been responsible for
the national fundamental levelling and gravimetric networks and their
international connections. In 2001 the Finnish Geodetic Institute’s
gravity and geodesy departments were joined into a new department
of geodesy and geodynamics, to which gravity research also belongs.
Among topics studied are solid-Earth tides, the free oscillations of
the solid Earth, post-glacial land uplift, and vertical reference or height korkeus-
systems. järjestelmä

Geoid models have been computed all the time, starting with Hir-
vonen’s global model (Hirvonen, 1934) and ending, for now, with the
Finnish model FIN2005N00 (Bilker-Koivula, 2010). These geoid mod-
els are actually based on the Nordic NKG2004 gravimetric geoid, and
are fitted to a Finnish set of GNSS levelling control points through a
transformation surface.
In 2015, the Finnish Geodetic Institute was merged into the National

í Õ ! ¤. û
392 Earth gravity field research

Land Survey as its geospatial data centre and research facility. The
English-language acronym continues as FGI, the Finnish Geospatial
Research Institute (https://www.maanmittauslaitos.fi/en/research).
Helsinki University of Technology (today part of Aalto University)
has also been active in research on the Earth’s gravity field. Heiskanen,
a professor at HUT in 1928–1949, acted in 1936–1949 as the director
of the International Isostatic Institute. After moving to Ohio State
University, he worked with many other, including Finnish and Finnish-
born, geodesists on calculating the first major global geoid model, the
“Columbus geoid” (Kakkuri, 2008).

^ 15.5 Textbooks
There are many good textbooks on the study of the Earth’s gravity field.
In addition to the already mentioned classic, Heiskanen and Moritz
(1967), which is in large part obsolete, we may mention Wolfgang Torge’s
book (1989). Moritz (1980) is difficult but good. Similarly difficult is
Molodensky et al. (1962). Worth reading also from the perspective of
physical geodesy is Vaníček and Krakiwsky (1987). A newer book in
the field is Hofmann-Wellenhof and Moritz (2006).

í Õ ! ¤. û
^ Field theory and vector calculus
— core knowledge

^ A.1 Vector calculus

A
In physics, many quantities are vector quantities; for example, force,
velocity, electrostatic field, and many more. A vector behaves in the
same way as the location difference between two neighbouring points.
Let the location difference be ∆r = r2 − r1 , in which r1 and r2 are the
location vectors of points 1 and 2. In a co-ordinate transformation, the
vector considered as an object does not change, but the numerical values
of its components, subsection A.2.2, are co-ordinate-system dependent
and will change.
About notation In printed text, vectors are often written in bold. In
handwritten text one may use an arrow above the symbol: → −v.

^ A.1.1 Scalar product

Between two vectors, a scalar product or dot product can be defined,
which is itself a scalar value. A scalar is in physics a single numerical
value; say, pressure or temperature. In the case of a scalar product of
two vector fields, the value is tied to a location, but, even if a co-ordinate
transformation changes the co-ordinate values of the location, the scalar
itself remains unchanged.

– 393 –
394 Field theory and vector calculus — core knowledge

An example of a scalar product: work ∆E is

∆E = ⟨F · ∆r⟩ ,

the scalar product of force F and path ∆r. Often, we leave the angle
brackets ⟨·⟩ off.
Later we shall see that if the points 1 and 2, ∆r = r2 − r1 , are very
close to each other, we may write

dE = ⟨F · dr⟩ ,

in which dr and dE are infinitesimal elements of path and energy. If

now there is a curved path between points A and B, we may get from
this an integral equation, the work integral:
wB
∆EAB = ⟨F · dr⟩ .
A

^ A.1.2 The scalar product, formally

Let
def
s = ⟨a · b⟩
be the scalar product of the vectors a and b. It holds (µ ∈ R) that

⟨µa · b⟩ = ⟨a · µb⟩ = µ ⟨a · b⟩ ,
⟨a · b⟩ = ⟨b · a⟩ ,

and we call
def
√︁
∥a∥ = ⟨a · a⟩
the norm or length of vector a.
The following also applies:

s = ∥a∥ ∥b∥ cos α,

where α is the angle between the directions of the vectors a and b.

í Õ ! ¤. û
 Vector calculus
395
^ A.1.3 Exterior or vectorial product
The exterior product, or cross product, of two vectors is itself a vector
called the vectorial product, at least in three-dimensional Euclidean space.
For example, the angular momentum q:
q = ⟨r × p⟩ ,
where p = mṙ is linear momentum, r the location vector of the body
relative to some origin, m the mass of the body, and
dr
ṙ = (A.1)
dt
is the time derivative of the location, or velocity. We write
q = m ⟨r × ṙ ⟩ .

^ A.1.4 The vectorial product, formally

Let
def
c = ⟨a × b⟩
be the vectorial product of the two vectors a and b. Then (µ ∈ R):
⟨µa × b⟩ = ⟨a × µb⟩ = µ ⟨a × b⟩ ,
⟨a × b⟩ = − ⟨b × a⟩ ,
and thus ⟨a × a⟩ = 0.
The resulting vector c is always orthogonal to the vectors a and b. The
length of vector c corresponds to the surface area of the parallelogram suunnikas
spanned by vectors a and b:
∥c∥ = ∥a∥ ∥b∥ sin α, (A.2)
in which again α is the angle between the directions of vectors a and
b. If the angle is zero, then the vectorial product is also zero (because
then, a = µb for some suitable value of µ).
If the angle is not zero we need in addition a corkscrew rule saying
that, if a corkscrew is turned from vector a to vector b, it will move
forward in the direction of the product vector c = ⟨a × b⟩.

í Õ ! ¤. û
396 Field theory and vector calculus — core knowledge

c = ⟨b × a⟩

α
∥c∥

b
a

^ Figure A.1. Exterior or vectorial product.

^ A.1.5 Kepler’s second law

Let r be the location vector of the body (planet) relative to the centre
of motion (the Sun), and ṙ (equation A.1) its velocity vector. Then, the
vectorial product
dr
⟨︂ ⟩︂
⟨r × ṙ ⟩ = r × (A.3)
dt
is precisely twice the surface area of the triangle or “area” swept over
in a unit of time.
Let us take the time derivative of this product, the expression A.3:
⟩︂ ⟨︃ ⟩︃
d dr dr d2 r
⟨︂
⟨r × ṙ ⟩ = × + r × 2 = ⟨ṙ × ṙ ⟩ + ⟨r × r̈⟩ . (A.4)
dt dt dt dt
Here, the first term vanishes, because for an arbitrary vector ⟨a × a⟩ = 0.
In the second term, we can use our knowledge that the attractive force F
emanating from the Sun that causes planetary orbital motion, and also
the acceleration it causes,
d2 r
r̈ = 2 ,
dt
keskeisvoima are central:
GMm
F = mr̈ = − r.
∥r∥3
G is the universal gravitational constant, M is the mass of the Sun, and
m is the mass of the planet.
Substitute this into equation A.4:
d GM
⟨r × ṙ ⟩ = 0 − ⟨r × r⟩ = 0.
dt ∥r∥3

í Õ ! ¤. û
 Scalar and vector fields
397
Angular momentum
⟨r × ṙ ⟩

Planet

Sun
ṙ
1
∥⟨r × ṙ ⟩∥
2 Velocity
Radius vector vector
r

Figure A.2. Kepler’s second law. In the same amount of time, the radius vector
of a planet will “sweep over” a same-sized area — conservation
^ of angular momentum.

So: the quantity on the left-hand side, angular momentum q per unit
of mass m, is conserved:
q
⟨r × ṙ ⟩ = m .
Like, for example, the total amount of energy, electric charge and many
other quantities, the amount of angular momentum in a closed system
is also constant.

^ A.2 Scalar and vector fields

^ A.2.1 Definitions
In the Euclidean space we may define functions or fields.
A scalar field is a scalar-valued function, which is defined throughout
the space (or a part of it), for example temperature T (r). So, for every
value of the location vector r there is a temperature value T (r).
A vector field is a vector-valued function that again is defined through-
out space, for example the electrostatic field E(r).

í Õ ! ¤. û
398 Field theory and vector calculus — core knowledge

^ A.2.2 A basis in space

In the space we may choose a basis made up of three vectors which span
the space in question. Generally we choose three basis vectors i, j, k,
that are orthogonal to each other, and the norms, or lengths, of which
ortonormaali are equal to 1, an orthonormal basis. Orthogonality of two vectors means
kanta that their scalar product vanishes; so

i ⊥ j, i ⊥ k, j⊥k

means that
⟨i · j⟩ = ⟨i · k⟩ = ⟨j · k⟩ = 0. (A.5)
Orthonormality means in addition that

∥i∥ = ∥j∥ = ∥k∥ = 1. (A.6)

Now we may expand vectors in the space into their components:

a = a1 i + a2 j + a3 k,

and scalar and vectorial products can now also be calculated with the
aid of their components:
⟨︂ ⟩︂
s = ⟨a · b⟩ = (a1 i + a2 j + a3 k) · (b1 i + b2 j + b3 k) =
∑︂
3
= a1 b 1 + a2 b 2 + a3 b 3 = ai bi ,
i=1

using the identities stated above for the basis vectors A.5 and A.6.
For the vectorial product, the calculation is more involved. For
orthogonal vectors, the angle α in equation A.2 is 90◦ , so

∥⟨i × j⟩∥ = ∥⟨i × k⟩∥ = ∥⟨j × k⟩∥ = 1.

The corkscrew rule now tells us that

k = ⟨i × j⟩ = − ⟨j × i⟩ ,
i = ⟨j × k⟩ = − ⟨k × j⟩ ,
j = ⟨k × i⟩ = − ⟨i × k⟩ .

í Õ ! ¤. û
 Scalar and vector fields
399
We get as the final outcome the determinant
⎡ ⎤
i j k
c = ⟨a × b⟩ = det ⎣ a1 a2 a3 ⎦ =
⎢ ⎥

b1 b2 b3
= (a2 b3 − a3 b2 ) i + (a3 b1 − a1 b3 ) j + (a1 b2 − a2 b1 ) k.

So

c1 = a2 b3 − a3 b2 , c2 = a3 b1 − a1 b3 , c3 = a1 b2 − a2 b1 .

These expressions are determinants as well:

⎡ ⎤
c1 [︄ [︄ ]︄ [︄ ]︄ [︄ ]︄ ]︄T
a2 a3 a3 a1 a1 a2
⎣ c2 ⎦ = det det det .
⎢ ⎥
b2 b3 b3 b1 b1 b2
c3

^ A.2.3 The nabla operator

{︁ }︁
The location vector r can be written on the i, j, k basis as follows: kanta

r = xi + yj + zk,

which defines (x, y, z) co-ordinates in space.

Let us define a vector operator called nabla (∇) as follows:
def ∂ ∂ ∂
∇=i +j +k .
∂x ∂y ∂z
The operator is on its own without meaning. It acquires meaning only
when it operates on something, in which case the three partial derivatives
on the right-hand side can be calculated.

^ A.2.4 The gradient

Let V(r) = V(x, y, z) be a scalar field in space. The nabla operator will
give its gradient g, a vector field in the same space:
∂V ∂V ∂V
g = grad V = ∇V = i +j +k .
∂x ∂y ∂z

í Õ ! ¤. û
400 Field theory and vector calculus — core knowledge

^ Figure A.3. The gradient. The level curves of the scalar field in blue.

So, the field g(r) = g(x, y, z) is the gradient field of V. In physics g is

often a force field and V its potential.
Interpretation The gradient describes the slope of the scalar field. The
direction of the vector is the direction in which the value of the
scalar field changes fastest, and its length describes the rate of
change with location. Imaging a hilly landscape: the height of
the ground above sea level is the scalar field, and its gradient is
pointing uphill everywhere, away from the valleys towards the
hilltops. The longer the g arrows, the steeper the slope of the
ground surface.
The gradient operator (like also the divergence and the curl, see
later) is linear:

grad (U + V) = grad U + grad V.

í Õ ! ¤. û
 Scalar and vector fields
401

Figure A.4. The divergence. Positive divergences (“sources”) and negative

^ ones (“sinks”). Field lines dashed.

^ A.2.5 The divergence

Given a vector field a(x, y, z) = a1 i+a2 j+a3 k, we form the scalar product
s of this and the nabla operator:

∂a1 ∂a2 ∂a3

s = div a = ⟨∇ · a⟩ = + + .
∂x ∂y ∂z

Interpretation The divergence describes the “sources” of a vector field,

both the positive and negative ones. Imagine the velocity of the
flow of water as a vector field. At the locations of the “sources”
the divergence is positive, at the locations of the “sewer holes”
or sinks, negative; everywhere else zero (because liquid cannot lähteet, nielut
appear out of nothing or disappear into nothing).

^ A.2.6 The curl

Given again a vector field a(x, y, z), we form the vectorial product c of
this and the nabla operator, again producing a vector field:

í Õ ! ¤. û
402 Field theory and vector calculus — core knowledge
⎡ ⎤
i j k
⎢ ∂ ∂ ∂ ⎥
c = curl a = ∇ × a = det ⎢
⎣ ∂x ∂y ∂z ⎦ =
⎥

a1 a2 a3
⎡ ⎤ ⎡ ⎤
∂ ∂ [︄
∂ ∂
]︄ ∂ ∂
= det ⎣ ∂y ∂z ⎦ i − det ∂x ∂z j + det ⎣ ∂x ∂y ⎦ k =
a2 a3 a1 a3 a1 a2
(︃ )︃ (︃ )︃
∂a3 ∂a2 ∂a1 ∂a3 ∂a2 ∂a1
(︂ )︂
= − i+ − j+ − k,
∂y ∂z ∂z ∂x ∂x ∂y

using the evaluation rules for determinants.

Interpretation The curl describes the eddiness or turbulence present in a
vector field.

Imagine a weather map, where low- and high-pressure zones are

drawn. Our vector field is the wind field. The wind circulates (in the
northern hemisphere) clockwise around the high-pressure zones, and
anticlockwise around the low-pressure zones. We may say that the curl
of the wind field is positive at the high pressures and negative at the
low pressures.
(This is a poor metaphor, as it is two-dimensional. In R2 , the curl is a
scalar, not a vector, just like we need only one angle to characterise a
rotational motion, when in R3 we need the three Euler angles.)

^ A.2.7 Conservative fields

What happens if a vector field a is the gradient of a scalar field V, and
we try to calculate its curl b, which is a vector as well? Write
⎡ ⎤
i j k
⎢ ∂ ∂ ∂ ⎥
b = curl a = curl grad V = det ⎢ ∂x ∂y ∂z ⎥ V
⎢ ⎥
∂ ∂ ∂
⎣ ⎦
∂x ∂y ∂z
and let
b = b1 i + b2 j + b3 k.

í Õ ! ¤. û
 Scalar and vector fields
403

^ Figure A.5. The curl. Positive (clockwise) and negative (anticlockwise) eddies.

Then, expanding the determinant yields

∂ ∂ ∂ ∂
b1 = V− V = 0,
∂y ∂z ∂z ∂y
∂ ∂ ∂ ∂
b2 = V− V = 0,
∂z ∂x ∂x ∂z
∂ ∂ ∂ ∂
b3 = V− V = 0,
∂x ∂y ∂y ∂x
thus
b = curl a = 0!
In other words, if the vector field a(x, y, z) is the gradient of the scalar
field V(x, y, z), its curl will vanish:

curl grad V = ⟨∇ × ∇V⟩ = ⟨∇ × ∇⟩ V = 0,

so the vectorial product of ∇ with itself vanishes just as if it were an

ordinary vector!
Definition A vector field a of which the curl vanishes is called conserva-
tive, and the corresponding scalar field V, a = grad V, is called
the potential of field a.

We note immediately that, if

a(x, y, z) = grad V(x, y, z),

í Õ ! ¤. û
404 Field theory and vector calculus — core knowledge

then also
a(x, y, z) = grad (V(x, y, z) + V0 ) ,
with V0 an arbitrary constant, because
∂V0 ∂V ∂V
grad V0 = i + j 0 + k 0 = 0.
∂x ∂y ∂z
So the potential is not uniquely defined.

^ A.2.8 The Laplace operator

Assume a conservative field a, so curl a = 0. Then we may write

a = grad V = ∇V,

in which V is the potential.

Let us now express the divergence of field a into the potential:
∂ ∂ ∂ ∂ ∂ ∂
div a = ⟨∇ · a⟩ = ⟨∇ · ∇V⟩ = V+ V+ V=
∂x ∂x ∂y ∂y ∂z ∂z
(︃ )︃
∂2 ∂2 ∂2 def
= 2
+ 2 + 2 V = ∆V,
∂x ∂y ∂z
where we have introduced a new differential operator, the Delta operator
invented by the French Pierre-Simon Laplace,
∂2 ∂2 ∂2
∆= 2
+ 2 + 2 = ⟨∇ · ∇⟩ = ∇2 .
∂x ∂y ∂z
When operating on the potential of a “source free” field — for example
the gravitational potential in a vacuum or the electrostatic potential in
an area of space free of electric charges — the result of this Delta, or
Laplace, operator vanishes.

^ A.3 Integrals
^ A.3.1 The curve integral
We saw earlier that work ∆E can be written as the scalar product of
force F and path ∆r:
∆E = ⟨F · ∆r⟩ .

í Õ ! ¤. û
 Integrals
405
The differential form of this is

dE = ⟨F · dr⟩ ,

from which one obtains the integral form, the work integral
wB
∆EAB = ⟨F · dr⟩ .
A

Here, the amount of work needed to move a body from point A to point
B is computed by integrating ⟨F · dr⟩ along the path AB.
If we parametrise the path according to arc length s, and the tangent
vector to the path is called
def dx dy dz
t= i+ j + k,
ds ds ds
we may also write
wB
∆EAB = ⟨F · t⟩ ds,
A
the parametrised version of the integral.

^ A.3.2 The surface integral

Assume we are given again a vector field a and a surface in space S.
Often, one seeks to integrate over surface S the normal component of a
vector field, the projection of a onto the normal vector of the surface.
Let the normal vector of the surface be n. Then we must integrate
x
⟨a · n⟩ dS,
S

symbolically written x
⟨a · dS⟩ ,
S
in which the notation dS is called an oriented surface element. It is a vector
pointing in the same direction as the normal vector n.
Like a curve, a surface can also be parametrised. For example, the
Earth’s surface (assumed a sphere) can be parametrised by latitude ϕ
and longitude λ: r = r(ϕ, λ). In this case we write as the surface element

dS = R2 cos ϕ dϕ dλ,

í Õ ! ¤. û
406 Field theory and vector calculus — core knowledge

Tangent vector t

curl a

curl a

Closed Integral
z
path ∂S s Integral ∂S
⟨a · t⟩ ds
S ⟨curl a · n⟩ dS

curl a

^ Figure A.6. The Stokes curl theorem.

in which R2 cos ϕ is Jacobi’s determinant of the parameter pair (ϕ, λ). In

this parametrisation, the integral is calculated as follows:
x x x
⟨a · dS⟩ = ⟨a · n⟩ dS = ⟨a · n⟩ R2 cos ϕ dϕ dλ.
S S S

Other surfaces and parametrisations have other Jacobi’s determinants.

The determinant always represents the true area of a “parameter surface
element” dϕ dλ “in nature”. For example, on the Earth’s surface, a
degree times degree patch is largest near the equator. In polar co-
ordinates (ρ, θ) in the plane (x = ρ cos θ, y = ρ sin θ), the determinant
of Jacobi is ρ. In the ordinary (x, y) parametrisation in the plane, Jacobi’s
determinant is 1 and thus can be left out altogether.

^ A.3.3 The Stokes curl theorem

Let S be a surface in space (not necessarily flat) and ∂S its edge curve.
Assume that the surface and its edge are well-behaved enough for all
necessary integrations and differentiations to be possible. Then (Stokes):
x z
⟨curl a · dS⟩ = ⟨a · dr⟩ ,
S ∂S

í Õ ! ¤. û
Integrals
407
with r the location vector of the edge curve. The parametrised form of
the theorem is
x z
⟨curl a · n⟩ dS = ⟨a · t⟩ ds,
S ∂S

with n is the normal to surface S and t the tangent vector of edge curve
∂S.
In words The surface integral of the curl of a vector field over a surface
is the same as the closed path integral of the field around the
edge of the surface.
Special case For a conservative vector field a it holds that curl a = 0
everywhere. Then z
⟨a · dr⟩ = 0,
∂S

so also wB wB
⟨a · dr⟩ = ⟨a · dr⟩ .
A, path 1 A, path 2

Let a be the force vector of a field, like the acceleration, or force

per unit mass, caused by the gravity field. Then this has the
following interpretation:

The work integral from point A to point B does not depend on the path
chosen. And the work done by a body transported around a closed path
is zero.

This perhaps explains better the essence of a conservative force

field. A conservative field can be represented as the gradient of a
potential: a = grad V, in which V is the potential of the field. The
Earth’s gravity vector field g(x, y, z) is the gradient of the Earth’s
gravity potential W(x, y, z). At mean sea level — more precisely,
at the geoid — the gravity potential is constant; the gravity vector
g is everywhere perpendicular to the geoid.

í Õ ! ¤. û
408 Field theory and vector calculus — core knowledge

a
∂V

div a
V

Figure A.7. The Gauss integral theorem. n is the normal vector to the exterior
surface. The Gauss integral theorem can also be presented with
the aid of (Michael Faraday’s) field lines: a field line starts or
terminates on an electric charge (a place where div a ̸= 0) or runs
^ to infinity (through the surface ∂V).

^ A.3.4 The Gauss integral theorem

Let V be a part of space, and ∂V its closed boundary, a union of surfaces.
Assume again that both are mathematically well-behaved. Then the
following theorem applies (Gauss):
y x x
div a dV = ⟨a · dS⟩ = ⟨a · n⟩ dS.
V ∂V ∂V

In words What is created inside a body (“sources”, divergence) must

come out through its surfaces.
Usually, the orientation of surface ∂V is taken as positive on the
outside: the normal vector n of the surface points outwards.

í Õ ! ¤. û
 The continuity of matter
409
^ A.4 The continuity of matter
An often-used equation in hydro- and aerodynamics is the continuity
equation. This expresses that matter cannot just disappear or increase in
amount. In the general case, the equation looks like this:

d
div(ρv) + ρ = 0.
dt
Here, the expression ρv stands for mass currents, ρ is the matter density,
v is the velocity of flow. The term div(ρv) expresses how much more
matter, in a unit of time, exits the volume element than enters it, per unit
of volume. The second term again, the time derivative of the density ρ,
stands for the change in the amount of matter inside the volume element
over time. The two terms must balance for the “matter accounting” to
close.
If the moving fluid is incompressible, then ρ is constant:

d
ρ = 0 =⇒ div(ρv) = ρ div v = 0 =⇒ div v = 0.
dt
Remember, however, that curl v does not necessarily vanish — so, the flow
is not necessarily eddy-free — so a potential V for which v = grad V
does not necessarily exist.

í Õ ! ¤. û
^ Function spaces

^ B.1 An abstract vector space

B
In an abstract vector space we may create a basis, with the help of which kanta
each vector can be written as a linear combination of the basis vectors:
for example, if the basis, in a concrete three-dimensional space, is
{e1 , e2 , e3 }, we may write an arbitrary vector r in the form

∑︂
3
r = r1 e1 + r2 e2 + r3 e3 = ri ei .
i=1

Precisely because three basis vectors (not in the same plane) are always
enough, we call the ordinary (Euclidean) space three-dimensional.
In a vector space one can define a scalar product, which is a linear
mapping from two vectors to one number (“bilinear form”):

⟨r · s⟩ .

Linearity means that

⟨(αr1 + βr2 ) · s⟩ = α ⟨r1 · s⟩ + β ⟨r2 · s⟩ , α, β ∈ R

and commutativity that vaihdan-

naisuus
⟨r · s⟩ = ⟨s · r⟩ .

– 411 –
412 Function spaces

kantavektori If the basis vectors are orthogonal to each other, in other words, ⟨ei · ej ⟩ =
0 if i ̸= j, we may calculate the coefficients ri in a simple way:

∑︂
3
⟨r · ei ⟩ ⟨r · ei ⟩
r= ri ei , ri = = . (B.1)
i=1
⟨ei · ei ⟩ ∥ei ∥2

If, in addition,

⟨ei · ei ⟩ = ∥ei ∥2 = 1, i ∈ {1, 2, 3},

in other words, the basis vectors are orthonormal, equation B.1 becomes
simpler still:

∑︂
3
r= ri ei , ri = ⟨r · ei ⟩ . (B.2)
i=1

The quantity
√︁
∥ei ∥ = ⟨ei · ei ⟩
is called the norm of the vector ei .
Unlike ordinary space, which is three-dimensional, a function space
is an infinite-dimensional, abstract vector space, that nevertheless helps
us to make certain abstract, but very useful fundamentals of function
theory more concrete!

^ B.2 The Fourier function space

^ B.2.1 Description
Functions can also be considered elements in a vector space. If we define
the scalar product of two functions f and g as the following integral1
⟨︂→
− → w 2π
def 1
f ·−
⟩︂
g = π f(x)g(x) dx, (B.3)
0

it is easily verified that the above requirements for a scalar product are
met.

í Õ ! ¤. û
 The Fourier function space
413
One basis in this vector space (a function space) is formed by the Fourier kanta
basis functions,
→
− 1√
e0 = 2,
2
→
−ek = cos kx, k = 1, 2, 3, . . . (B.4)
→
−
e = sin kx, k = 1, 2, 3, . . .
−k

This basis is orthonormal (proof: exercise). It is also a complete basis,

which we shall not prove. As the number of basis vectors is countably
infinite, we say that this function space is infinitely dimensional.
Now every function f(x) meeting certain conditions can be expanded
in the way of equation (B.2), as follows:

1 √ ∑︂ ∞
f(x) = a0 2 + (ak cos kx + bk sin kx) ,
2
k=1

— the familiar Fourier-series expansion — in which the coefficients are

w
1 √ 2π √
⟨︂→
− − ⟩︂
a0 = f · → e0 = 2 f(x) dx = 2 · f(x),
2π 0
⟨︂→
− → w
− 1 2π
⟩︂
ak = f · e k =π f(x) cos kx dx, k = 1, 2, 3, . . .
0
⟨︂→
− − w
1 2π
bk = f · →
⟩︂
e−k = π f(x) sin kx dx, k = 1, 2, 3, . . .
0

This is the familiar way in which the coefficients of a Fourier series are
calculated.

^ B.2.2 Example
As an example of Fourier analysis, we may take a step function on the
interval [0, 2π): ⎧
⎨0 x ∈ [0, π)
f(x) = .
⎩1 x ∈ [π, 2π)

1 Thearrows over the function designators try to psychologically instil the notion that
they are “vectors”.

í Õ ! ¤. û
414 Function spaces

We can calculate the Fourier coefficients of this function as follows:

w 2π
1 √ 1 √ 1√
a0 = 2· f(x) dx = 2·π= 2,
2π 0 2π 2
w w
1 2π 1 2π
ak = π f(x) cos kx dx = π cos kx dx =
0 π
]︂2π
1 1 1
[︂
=π sin kx = (sin 2kπ − sin kπ) = 0,
k π kπ
w w
1 2π 1 2π
bk = π f(x) sin kx dx = π sin kx dx =
0 π
]︂2π
1 1 1
[︂
= π − cos kx = (cos kπ − cos 2kπ) =
k π kπ ⎧
)︂ ⎨0 if k even,
1
(︂
= (−1)k − 1 = 2
kπ ⎩− if k odd.
kπ
√
In numbers: a0 = 12 2 = 0.707 10 . . . , b1 = − 2 π = −0.636 62 . . . , b3 =
/︁

− 2 3π = −0.212 20 . . . , b5 = −0.127 32 . . ., and so forth. The expansion

/︁

now becomes
√ ∑︂
∞
2 ∑︂ 1
∞
1 1
f(x) = 2 2a0 + bk sin kx = 2 − π sin kx.
k
k=1 k=1
odd

We see that it only contains sines, no cosines. This is a consequence of

the function’s symmetry properties.
In figure B.1 we show truncated expansions for this function:

def 1 √ ∑︂
K
2 ∑︂ 1
K
(K) 1
f (x) = 2 a0 2 + bk sin kx = 2
−π sin kx, (B.5)
k
k=1 k=1
odd

with K the truncation parameter.

^ B.2.3 Convergence
suppeneminen The Fourier expansion converges in the square integral sense: if we
define the truncated expansion

def 1 √ ∑︂
K
(K)
f (x) = 2 a0 2 + (ak cos kx + bk sin kx) ,
k=1

í Õ ! ¤. û
 The Fourier function space
415
K=1
K = 25 K=3 K=5
1.0

0.8 a0
f(x)
0.6
X
0.4 a b

0.2 b5
b3
0
b1
−0.2
0 1 2 3 4 5 6

Figure B.1. Fourier analysis on a step function. Plotted are the truncated
Fourier expansions f(K) (x), equation B.5, for values of K of 1, 3, 5,
^ and 25. The inset gives the spectrum of the function.

then w 2π (︁
1 )︁2
lim f(K) (x) − f(x) dx = 0.
K→∞ π 0
⃓ ⃓
This does not mean that there exists a ε for which ⃓f(K) (x) − f(x)⃓ < ε
for every x ∈ [0, 2π), when K → ∞. Looking at figure B.1, there will
always remain a small neighbourhood of x = π where the absolute
⃓ ⃓
difference ⃓f(K) (x) − f(x)⃓ will reach 0.5, even for arbitrarily large values
of K. We say that the Fourier expansion is convergent, but not uniformly tasainen
convergent. suppeneminen

The Fourier expansion converges pointwise “almost everywhere” in

x ∈ [0, 2π): at all points except for the two special points x = 0 and
x = π. By defining f(0) = f(π) = 0.5, the expansion is made pointwise pisteittäinen
convergent everywhere. suppeneminen

Also, note the “shoulder” of the expansion, even for K = 25. This
shoulder will get narrower for higher K, but not any lower, remaining

í Õ ! ¤. û
416 Function spaces

Gibbsin ilmiö at approximately 0.09. This is known as the Gibbs phenomenon.

^ B.3 Sturm–Liouville differential equations

^ B.3.1 The eigenvalue problem
In an abstract vector space we may formulate an eigenvalue problem:
given a linear operator (mapping) L, we may write

Lx − λx = 0,

where the problem consists of determining the eigenvalues λp for which

one or more solutions or eigenvectors xp exist.
ortonormaali In a concrete n-dimensional vector space in which there is an or-
{︁ }︁
kanta thonormal basis e , i = 1, . . . , n we may write the vector
i

∑︂
n
x= xi ei ,
i=1

and, thanks to linearity,

(︃∑︂
n )︃ ∑︂
n
Lx = L xi ei = xi · Lei .
i=1 i=1

On the other hand, we may write n different vectors Lei on the basis
{︁ }︁
ej in the following way:

∑︂
n
Lei = aij ej , i = 1, . . . , n.
j=1

This defines the coefficients aij , which may be collected into a size n × n
matrix A.
Now substitution yields

∑︂
n (︃∑︂
n )︃
Lx = aij xi ej . (B.6)
j=1 i=1

í Õ ! ¤. û
 Sturm–Liouville differential equations
417
Also
∑︂
n ∑︂
n
λx = λ xi ei = (λxj ) ej . (B.7)
i=1 j=1

By combining equations B.6 and B.7, of which all coefficients must be

identical, we obtain
∑︂
n
aij xi − λxj = 0, j = 1, . . . , n,
i=1

or, as a matric equation,

Ax − λx = 0, (B.8)

in which A is a matrix consisting of the coefficients aij , and x a column

[︂ ]︂T
vector consisting of the coefficients xi : x = x1 x2 · · · xn .
Of course equation B.8 also represents an eigenvalue problem, but
now in the linear vector space Rn consisting of all coefficient vectors x.
Every x is the numerical representation of a vector x on the chosen basis kanta
{︁ }︁
ei . Matrix A is again the numerical representation of operator L on
the same basis.2

^ B.3.2 A self-adjoint operator

Let L be a linear operator in a vector space where there exists a scalar
product, i. e., a bilinear form ⟨x · y⟩ which is symmetric or commutative. vaihdan-
naisuus
Then L is self-adjoint, if for each pair of vectors x, y it holds that

⟨x · Ly⟩ = ⟨Lx · y⟩ .

If the corresponding matrix A is self-adjoint, that means that

⟨x · Ay⟩ = ⟨Ax · y⟩ ,

2 An advantage of the numerical representations is of course that one can actually

calculate with them.

í Õ ! ¤. û
418 Function spaces

i. e.,
∑︂
n (︃∑︂
n )︃ ∑︂
n (︃∑︂
n )︃
xi aij yj = aij xj yi ,
i=1 j=1 i=1 j=1

which is trivially true if

aij = aji , i, j ∈ 1, . . . , n ⇐⇒ A = AT .

In other words,

A symmetric matrix is a self-adjoint operator.

From linear algebra it is undoubtedly familiar that the eigenvectors xp , xq

belonging to different eigenvalues λp ̸= λq of a symmetric, size n × n matrix
are mutually orthogonal: xp ⊥ xq . If all eigenvalues λp , p = 1, . . . , n are
täydellinen different, then the eigenvectors xp , p = 1, . . . , n will constitute a complete
ortogonaali orthogonal basis3 in the vector space Rn .
kanta
The proof is not hard. We start from the equation for the eigenvalue
problem for eigenvectors and -values xp , λp :

Lxp = λp xp ,

and multiply from the left by vector xq :

⟨xq · Lxp ⟩ = λp ⟨xq · xp ⟩ .

Similarly for eigenvectors and -values xq , λq multiplied from the left by

vector xp :
⟨xp · Lxq ⟩ = λq ⟨xp · xq ⟩ .
If L is self-adjoint, then

⟨xq · Lxp ⟩ = ⟨Lxq · xp ⟩ = ⟨xp · Lxq ⟩ =⇒ λp ⟨xq · xp ⟩ = λq ⟨xp · xq ⟩ .

3 Actually the eigenvectors may be arbitrarily re-scaled: if x is an eigenvector, then also

def
e = x ∥x∥ is. Thus we obtain an orthonormal basis.
/︁

í Õ ! ¤. û
Sturm–Liouville differential equations
419
It follows that
(λp − λq ) ⟨xp · xq ⟩ = 0.
Remember that the scalar product is symmetric. If λp ̸= λq , we thus
must have ⟨xp · xq ⟩ = 0, or xp ⊥ xq , what was to be proven.
Example The variance matrix of location in the plane. The variance matrix
of the co-ordinates of point P in the plane is
{︄[︄
{︁ }︁
]︄}︄ [︄ ]︄
xP σ2x σxy
Var xP = Var = ΣPP = ,
yP σxy σ2y

a symmetric matrix. Here, σ2x and σ2y are the variances, or squares
of the mean errors, of the x and y co-ordinates, whereas σxy is
the covariance between the co-ordinates.
The eigenvalues of this matrix ΣPP are the solutions of the charac-
teristic equation
[︄ ]︄
σ2x − λ σxy
det = 0,
σxy σ2y − λ
or
(︁ 2
σx − λ σ2y − λ − σ2xy = 0.
)︁ (︁ )︁

This yields

1
(︁ 2 2
)︁ 1 √︂[︁ ]︁2 [︁ ]︁
λ1,2 = σx + σy ± 2
2
σ2x + σ2y − 4 σ2x σ2y − σ2xy =
1
(︁ 2 2
)︁ 1 √︂[︁ ]︁2
= 2 σx + σy ± 2 σ2x − σ2y + 4σ2xy .

The variance matrix has a variance or error ellipse. The semi-lengths

√ √
of its principal axes are λ1 and λ2 , and the directions of the
principal axes are the eigenvectors of ΣPP : x1 and x2 , mutually
orthogonal. If the co-ordinate axes are turned into the directions
of x1,2 , the matrix ΣPP will assume the form
[︄ ]︄ [︄ ]︄
2
′ σx ′ 0 λ 1 0
ΣPP = = .
0 σ2y ′ 0 λ2

í Õ ! ¤. û
420 Function spaces

The sum of the eigenvalues (and the trace of the matrix), λ1 + λ2 =

σ2x + σ2y , is an invariant called the point variance.

^ B.3.3 Self-adjoint differential equations

A function space also features self-adjoint or “symmetric” differential
equations. In fact, the most famous equations of physics are of this type.
Take a good look at, for example, the oscillation equation, in which
x(t) is the position as a function of time:

d2
x(t) + ω2 x(t) = 0. (B.9)
dt2
The solution has the general form (α amplitude, ϕ phase constant)

x(t) = α sin(ωt − ϕ).

On the interval t ∈ [0, T ] we require periodicity:

d ⃓ d ⃓
(︁ )︁ ⃓ ⃓
x(0) = x T , x⃓ = x⃓ .
dt x=0 dt x=T
These boundary conditions are an essential part of being self-adjoint.
Then, a solution is found only for certain values of ω — quantisation.
Equation B.9 is an eigenvalue problem, form-wise:

Lx + ω2 x = 0,

in which the operator is

d2
L= .
dt2
We first show that this operator is on the interval [0, T ] self-adjoint. If the
scalar product is defined as follows:
⟨︁→
− →
− ⟩︁ def w T
x · y = x(t)y(t) dt,
0

í Õ ! ¤. û
Sturm–Liouville differential equations
421
it holds that (integration by parts):

⟨︁→
− ⟩︁ w T d2 y(t)
x · L→
−
y = x(t) dt =
0 dt2
[︃ ]︃T w
dy(t) T d d
= x(t) − x(t) y(t) dt,
dt 0 0 dt dt
⟨︁ → w 2
T d x(t)
L−x ·→− ⟩︁
y = y(t) dt =
0 dt2
[︃ ]︃T w
dx(t) T d d
= y(t) − x(t) y(t) dt.
dt 0
0 dt dt

As, on the right-hand side, the first terms vanish and the second terms
are identical, it follows that
⟨︁→
−
x · L→
−
y = L→
⟩︁ ⟨︁ − →
x ·−
⟩︁
y ,

which was to be proven.

Self-adjoint operators have eigenvalues and eigenvectors, in this case
functions, that are mutually orthogonal for different values of ω.4 For
the oscillation equation with the above periodicity conditions they are
just the solution functions

2πk
(︂ )︂
sin(ωk t − ϕ) = sin t−ϕ , (B.10)
T
in which the frequency
2πk
ωk =
T
is quantised by a “quantum number” k ∈ N.
If we let T → ∞, the frequencies ωk get closer and closer to each
other, and in the end morph into a continuum.

4 In fact, for the same value ω there exist two mutually orthogonal periodic solutions,
k

2πkt 2πkt
sin ωk t = sin , cos ωk t = cos .
T T
Any linear combination of these is a valid solution as well, and is of the general form
B.10.

í Õ ! ¤. û
422 Function spaces

In physics there is a broad class of differential equations that are

self-adjoint in some function space. The class is known as “Sturm5 –
Liouville6 type problems”. It includes the oscillation equation, Legen-
dre’s equation, Bessel’s equation, and many more. Every one of them
generates, in a natural way, its own set of mutually orthogonal functions
kantafunktio that serve as the basis functions for the general solution of many partial
differential equations.

^ B.4 Legendre polynomials

The ordinary Legendre polynomials Pn (t) also constitute a basis in a
function space, with the scalar product definition
⟨︂→
− → w +1
f ·−
⟩︂
def
g = f(t)g(t) dt.
−1

They do not however constitute an orthonormal basis, but only an

ortogonaali orthogonal one:
kanta ⃦→
⃦− ⃦2
⃦ ⟨︂→
− → − ⟩︂ w +1 2 2
⃦ Pn ⃦ = Pn · Pn = Pn (t) dt = .
−1 2n + 1

^ B.5 Spherical harmonics

On the surface of a sphere, all functions can also be considered elements
of a function space. Every function meeting certain well-behavedness
requirements — like integrability — is an element. The functions

Rnm (ϕ, λ) = Pnm (sin ϕ) cos mλ, n = 0, 1, 2, . . . ,m = 0, . . . , n,

Snm (ϕ, λ) = Pnm (sin ϕ) sin mλ, n = 0, 1, 2, . . . ,m = 1, . . . , n,

together form a complete basis for this vector space in such a way that
every function can be written as an — if necessary infinite — linear

5 JacquesCharles François Sturm FRS FAS (1803–1855) was an eminent French mathe-
matician, one of the 72 names engraved on the Eiffel Tower. Eiffel Tower, 72 names.
6 Joseph Liouville FRS FRSE FAS (1809–1882) was an eminent French mathematician.

í Õ ! ¤. û
Spherical harmonics
423
combination of these basis functions. The situation is analogous to three-
dimensional space, where a complete basis consists of three vectors not
in the same plane.
An alternative, more compact way of writing this is
⎧
⎨P (sin ϕ) cos mλ if m ⩾ 0,
nm
Ynm (ϕ, λ) =
⎩Pn|m| (sin ϕ) sin |m| λ if m < 0,

for values n = 0, 1, 2, . . . , m = −n, . . . , n.

In this function space, a scalar product is defined:
⟨︂→
− −→⟩︂ x
1
V ·W = V(ϕ, λ)W(ϕ, λ) dσ,
4π σ
in which σ is the surface of the unit sphere (“directional sphere”, or
even “celestial sphere”), dσ = cos ϕ dϕ dλ is a surface element of the
sphere, and cos ϕ is the determinant of Jacobi of the co-ordinates (ϕ, λ).
According to this definition, we can show that two different functions,
Ynm and Ysr , are orthogonal with respect to each other:
⟨︂→
− →
− ⟩︂ x
1
Ynm · Ysr = Y (ϕ, λ) Ysr (ϕ, λ) dσ = 0
4π σ nm
if n ̸= s or m ̸= r.
{︂→
− }︂
The basis Ynm , n = 0, 1, 2, . . . , m = −n, . . . , n , is orthogonal but not kanta
orthonormal: the “length” of the vectors differs from unity.
⃦→
⃦− ⃦2
⃦ ⟨︂→
− →
− ⟩︂
⃦ Ynm ⃦ = Ynm · Ynm =
⎧
x ⎪
⎨ 1
if m = 0,
1 2 2n + 1
= Ynm (ϕ, λ)dσ = 1 (n + |m|)!
4π σ ⎪
⎩ if m ̸= 0,
2(2n + 1) (n − |m|)!
see Heiskanen and Moritz (1967, equation 1-69). Proving this orthogo-
nality is not straightforward.
If we now divide the functions Ynm (or, equivalently, Rnm , Snm ) by
the square roots of the above factors, we obtain the fully normalised
surface spherical harmonics Y nm , for which it holds that pinta-
pallofunktio
í Õ ! ¤. û
424 Function spaces

⃦→
⃦
− ⃦2
⃦ x 2
⃦ Y nm ⃦ = 1 Y (ϕ, λ) dσ = 1.
⃦ ⃦ 4π σ nm

With those it is again easy to calculate the coefficients fnm of a given

general function on the sphere f(ϕ, λ) (the overline means that these
are fully normalised coefficients):
⟨︃
− →
→ −
⟩︃ x
1
fnm = f · Y nm = f(ϕ, λ) Y nm (ϕ, λ) dσ. (B.11)
4π σ

kanta This is a straightforward projection on the unit vectors of the basis

(geometric analogue).
In the above integral, f(ϕ, λ) is the function f on the Earth’s surface: if
the radius of the spherical Earth is R, then f(ϕ, λ) = f(ϕ, λ, R).
The fully normalised equation corresponding to expansion 2.11 is

∑︂
∞
1 ∑︂
n
(︁ )︁
V(ϕ, λ, r) = P nm (sin ϕ) a nm cos mλ + b nm sin mλ .
rn+1
n=0 m=0

We may also write

⎧
⎨P
nm (sin ϕ) cos mλ if m ⩾ 0,
Y nm (ϕ, λ) =
⎩Pn|m| (sin ϕ) sin |m| λ if m < 0,

which corresponds to the definition of the fully normalised Legendre

functions:
√
Pn0 (sin ϕ) = 2n + 1 Pn0 (sin ϕ) ,
√︄
(n − m)!
Pnm (sin ϕ) = 2(2n + 1) P (sin ϕ), m > 0.
(n + m)! nm

Now, the above equation for the potential becomes

∑︂
∞
1 ∑︂
n
V(ϕ, λ, r) = vnm Y nm (ϕ, λ),
rn+1
n=0 m=−n

í Õ ! ¤. û
 Self-test questions
425
in which ⎧
⎨a if m ⩾ 0,
nm
vnm =
⎩bn|m| if m < 0.

On the sphere r = R this becomes

∑︂
∞
1 ∑︂
n
V(ϕ, λ, R) = vnm Y nm (ϕ, λ),
Rn+1
n=0 m=−n

from which by orthogonal projection (equation B.11) follows

⟨︃
− → −
⟩︃ x
n+1 → Rn+1
vnm = R V · Y nm = V(ϕ, λ, R) Y nm (ϕ, λ)dσ
4π σ

or
x
Rn+1
anm = V(ϕ, λ, R) Pnm (ϕ, λ) cos mλ dσ,
4π σ
x
Rn+1
bnm = V(ϕ, λ, R) Pnm (ϕ, λ) sin mλ dσ.
4π σ

^ Self-test questions
1. The identity ⟨r · s⟩ = ⟨s · r⟩, for two elements r and s of a vector
space, expresses the property of linearity | commutativity |
associativity.

^ Exercise B – 1: Orthonormality of the Fourier basis

functions
Show the orthonormality of the Fourier basis functions, equation B.4 by kantafunktio
deriving their scalar products by equation B.3.

í Õ ! ¤. û
^ Why does FFT work?

FFT
C
is a factorisation method for computing the discrete Fourier trans-
form that spectacularly reduces the number of calculations needed and
speeds up the calculation. It requires the number of grid points to be a
factorisable number.
There are alternatives in choosing precisely which FFT method to use.
The fastest FFT requires a grid the number of points of which is a power
of 2. The size of the grid is then 2n × 2m . Alternative, “mixed-radix”
methods may also be considered and perform well if the grid size is
something like 360 × 480, for example N = 360 = 2 × 2 × 2 × 3 × 3 × 5.
If the grid size is a prime number, FFT is no better than the ordinary
discrete Fourier transform.
If the function f(x) is given on the interval x ∈ [0, L), on an equi-
spaced grid, xk = kL N , as values fk = f(xk ), k = 0, . . . , N − 1, the
/︁

discrete Fourier transform in one dimension is

F{f(x)} = F(˜︁
ν),

in which

1 ∑︂
N−1
jk
(︂ )︂
F(˜︁
νj ) = f(xk ) exp −2πi , j = 0, . . . , N − 1. (C.1)
N N
k=0

The frequency argument, spatial frequency or wave number, ν

˜︁j =

– 427 –
428 Why does FFT work?
[︂ /︂ ]︂
j L , j = 0, . . . , N − 1 is defined on the interval1 0, (N − 1) L . i is
/︁

the imaginary unit: i2 = −1. We use exp(x) to denote ex .

Correspondingly, the inverse discrete Fourier transform,

F−1 {F(˜︁
ν)} = f(x),

is
∑︂
N−1 (︂
jk
)︂
f(xk ) = νj ) exp 2πi
F(˜︁ , k = 0, . . . , N − 1. (C.2)
N
j=0

is just a brutally efficient way of computing both these equations

FFT
C.1 and C.2. A brute-force calculation of these formulas requires an
order of N2 “standard operations”, each of them a single multiplication
plus a single addition or subtraction. If N is even, we may write
⎛ ⎞
1
N−1
⎜ 2∑︂
1 ⎜
(︂
jk
)︂ ∑︂
N−1 (︂
jk ⎟
)︂
F(˜︁
νj ) = fk exp −2πi + fk exp −2πi ⎟=
N⎝ N N ⎠
k=0 1
k= N
2
⎛ 1 N−1 1
N−1
⎞
2∑︂ (︃ 1 )︃ 2∑︂ ′
1 jk N jk ⎠
(︂ )︂ (︂ )︂
= ⎝ fk exp −2πi + exp −2πij 2 f ′ 1 exp −2πi =
N N N ′
k+ N
2
N
k=0 k =0
⎛ 1 N−1 1
N−1
⎞
2∑︂ 2∑︂
1 jk jk ⎠
(︂ )︂ (︂ )︂
= ⎝ fk exp −2πi + exp(−πij) f ′ 1 exp −2πi =
N N k+ N
2
N
k=0 k=0
1
N−1 [︄ {︄ }︄
2∑︂
]︄
1 j (︂
jk
)︂ + if j even
= fk ± f 1 exp −2πi , (C.3)
N k+ N
2
N − if j odd
k=0

the computation of which sum requires only N · 12 N multiplications and

additions, not counting pre-calculations.
[︂ (︁ 1 )︁/︂ ]︂
1 Alternatively,the interval of definition could be chosen as − 12 N L , 2 N − 1 L .
/︁

vj − N L , or j → j−N, for j > 21 N−1. This has the merit

/︁
This is done by mapping ˜︁vj → ˜︁
of placing the frequency zero in the middle. It does not materially change anything,
νj ) with unity: exp −2πi Nk N = exp(−2πik) = 1, the
(︁ /︁ )︁
as it simply multiplies F(˜︁
periodicity property of the discrete Fourier transform.

í Õ ! ¤. û
 429
(︁ )︁j
Here we used Euler’s identity exp(−πi) = −1, so e−πij = e−πi =
(−1) j , either +1 or −1.2 The expression in square brackets, for each k
value k = 0, 1, . . . , 12 N − 1, is either a summation, for even values of j, or
a subtraction, for odd values of j. In total, 12 N sums and 12 N differences
are pre-calculated. Also the exp expressions are pre-calculated into a
lookup table.
Altogether some 12 N2 standard operations are needed, half the original
number.
Equation C.3 is itself recognised as a Fourier series, but the number
of support points is only 12 N instead of N. If 12 N is also even, we may
repeat the above trick, resulting in an expression requiring only an order
of 14 N2 operations. Lather, rinse, repeat, and the number of operations
becomes 18 N2 , 161 N2 , 32
1
N2 , etc. . . A more precise analysis shows that if N
is a power of 2, the whole discrete Fourier transform may be computed
in order N × 2 log N operations!
In the literature, smart algorithms are found implementing the
method described, for example fftw (“Fastest Fourier Transform in
the West”, FFTW Home Page; Frigo and Johnson, 2005).

2 These values are called the “twiddle factors”.

í Õ ! ¤. û
^ Helmert condensation

D
In order to derive the equation for Helmert condensation, we derive the
equation for the potential of the topography:
y ρ(ϕ ′ , λ ′ , r ′ ) y
′ 1
Vtop (ϕ, λ, r) = G dV ≈ Gρ dV ′ ,
top ℓ(ψ, r, r ′ ) top ℓ(ψ, r, r ′ )

in which ψ is the geocentric angular distance between the evaluation

point (ϕ, λ, r) and the data point (ϕ ′ , λ ′ , r ′ ). We assume a standard
density ρ.
We similarly derive the equation for the potential of the condensation
layer:
y
1
Vcond (ϕ, λ, r) = Gρ dV ′ .
cond ℓ(ψ, r, R)

We integrate in spherical co-ordinates:

y w w R+H(ϕ ′ ,λ ′ )
1 ′ 1 2
′
dV = ′
(r ′ ) dr ′ dσ ′ ,
top ℓ(ψ, r, r ) σ R ℓ(ψ, r, r )
y w w R+H(ϕ ′ ,λ ′ )
1 1 2
dV ′ = (r ′ ) dr ′ dσ ′ =
cond ℓ(ψ, r, R) σ ℓ(ψ, r, R) R
w (︃
H(ϕ ′ , λ ′ ) H2 (ϕ ′ , λ ′ )
)︃
1 ′ ′
=R 2
H(ϕ , λ ) 1 + + 2
dσ ′ , (D.1)
σ ℓ(ψ, r, R) R 3R

with H the height of the topography.

– 431 –
432 Helmert condensation

^ D.1 The exterior potential of the topography

In order to derive the exterior potential of the topography, we use the
expansion of the inverse distance (equation 8.6):

1 ∑︂ 1 ∑︂
∞ (︃ )︃n+1 ∞ (︃ )︃n
r′ 1 r′
= Pn (cos ψ) = Pn (cos ψ).
ℓ r′ r r r
n=0 n=0

tasainen This expansion converges uniformly1 with respect to ψ if r > r ′ . In

suppeneminen the following, we shall assume convergence throughout, dangerous as
that may be especially close to a jagged topographic surface. For the
philosophically inclined, see Moritz (1980).
Substitution yields

∑︂ ∞ y
(︃ )︃n
1 r′
ext
Vtop (ϕ, λ, r)
= Gρ r r Pn (cos ψ) dV ′ =
top
n=0
x w R+H(ϕ ′ ,λ ′ ) ∑︂
∞
(︄ (︃ )︃n )︄
1 r′ 2
= Gρ r r (r ′ ) dr ′ Pn (cos ψ) dσ ′ =
σ R
n=0
]︄R+H
x ∑︂
∞
[︄
1 1 n+3
= Gρ (r ′ ) Pn (cos ψ) dσ ′ =
σ rn+1 n + 3
n=0 r ′ =R
x ∑︂
∞
1 1
(︂ )︂
n+3
= Gρ (R + H) − R n+3
Pn (cos ψ) dσ ′ .
σ rn+1 n + 3
n=0

We now use the following Taylor expansion:

1 Uniform convergence means that, given r and r ′ , for every ϵ > 0 there is an Nmin for
which
⃓ 1 1 ∑︂
⃓ N (︃ ′ )︃n
⃓
r ⃓
⃓ − Pn (cos ψ)⃓ < ϵ
⃓ ⃓
⃓ℓ r n=0 r ⃓

for all N > Nmin , and for all values of ψ. This is a stronger property than mere
convergence.

í Õ ! ¤. û
 The interior potential of the topography
433
(R + H)n+3 =
(︃ )︃
n+3 H (n + 3) (n + 2) H2 (n + 3) (n + 2) (n + 1) H3
=R 1 + (n + 3) + + + ··· .
R 2 R2 2·3 R3
(D.2)
Substitution yields

ext
Vtop (ϕ, λ, r) = GρR2 ·
x ∑︂ ∞ (︂ )︂n+1 (︃ )︃
R H 1 H2 1 H3
· r + (n + 2) 2 + 6 (n + 2) (n + 1) 3 + · · · Pn (cos ψ) dσ ′ .
σ R 2 R R
n=0
(D.3)
This is thus the exterior potential of the topography, or, inside the
topographic masses, the harmonic downwards continuation of the exterior
potential, assuming that this is mathematically possible (in the case of
mountainous topography, generally not) and does not diverge.

^ D.2 The interior potential of the topography

In the same way we may derive the equation for the interior potential
of the topography, i. e., the masses between the sea level and terrain
surface. For the spatial distance ℓ between those points we use the
interior expansion, equation 8.6, valid for r < r ′ :

1 ∑︂ r n+1
∞ (︂ )︂
1
= r Pn (cos ψ).
ℓ r′
n=0

Substitute:

1 ∑︂ r n+1
y ∞ (︂ )︂
int
Vtop (ϕ, λ, r) = Gρ Pn (cos ψ) dV ′ =
top r r′
n=0
I
⏟ ⏞⏞ ⏟
x w R+H(ϕ ′ ,λ ′ )
1 ∑︂
∞ (︂ )︂
r n+1 ′ 2 ′
= Gρ (r ) dr Pn (cos ψ) dσ ′ .
σ R r r′
n=0

í Õ ! ¤. û
434 Helmert condensation

Here, the height integral I is

1 ∑︂ r n+1 ′ 2 ′
w R+H(ϕ ′ ,λ ′ ) ∞ (︂ )︂
I= (r ) dr =
R r r′
n=0
⎡ ⎤R+H(ϕ ′ ,λ ′ )
∑︂
∞
(︄ )︄
′ −(n−2)
(r )
=⎣ rn − + r2 ln r ′ ⎦ =
⎢ ⎥
n−2
n=0
n̸=2 r ′ =R
∑︂
∞
rn
(︂
−(n−2)
)︂
R+H
= R−(n−2)
− (R + H) + r2 ln ,
n−2 R
n=0
n̸=2

yielding

int
Vtop (ϕ, λ, r) =
⎛ ⎞
⎜∑︂ rn
x ∞
R + H⎟
(︂ )︂
= Gρ R−(n−2) − (R + H)−(n−2) + r2 ln P (cos ψ) dσ ′ .
σ
⎝ n−2 R ⎠ n
n=0
n̸=2

For this we use the Taylor expansion

(R + H)−(n−2) =
(︃ )︃
−(n−2) H (n − 2) (n − 1) H2 (n − 2) (n − 1) n H3
=R 1 − (n − 2) + − + ··· .
R 2 R2 2·3 R3
Also, the special case n = 2,
(︃ )︃
2 R+H 2 H 1 H2 1 H3 1 H4
r ln =r − + − + ... =
R R 2 R2 3 R3 4 R4
(︃ )︃
rn H n − 1 H2 (n − 1) n H3 (n − 1)n(n + 1) H4
= n−2 − + − + ··· ,
R R 2 R2 2 · 3 R4 2·3·4 R4
is cleanly included into the following expression obtained by substitu-
tion:
int
Vtop (ϕ, λ, r) =
x ∑︂ ∞ (︃ )︃
rn H 1 H2 1 H3
= Gρ − (n − 1) 2 + 6 (n − 1) n 3 − · · · Pn (cos ψ) dσ ′ . (D.4)
σ Rn−2 R 2 R R
n=0

í Õ ! ¤. û
 The exterior potential of the condensation layer
435
^ D.3 The exterior potential of the condensation layer
This is derived by specialising equation D.3 to the case H → 0, but
nevertheless ρ → ∞, so that κ = ρH remains finite. in this limit, all
terms containing H2 , H3 and higher powers go to zero. The result is
then
x ∑︂
∞ (︂ )︂n+1
R H
ext
Vcond (ϕ, λ, r) = GρR2
P (cos ψ) dσ ′ =
σ r R n
n=0
x ∑︂
∞ (︂ )︂n+1
R
= GR r κPn (cos ψ) dσ ′ .
σ
n=0

Earlier on we had a more precise formula 6.4 for κ on the surface of a

spherical Earth: (︃ )︃
H 1 H2
κ = ρH 1 + + 3 2 .
R R
By substituting this into the previous, we obtain (see also equation D.1):

x ∑︂
∞ (︂ )︂n+1 (︃ )︃
R H H2 1 H3
ext
Vcond = GρR 2
r + 2 + 3 3 Pn (cos ψ) dσ ′ . (D.5)
σ R R R
n=0

^ D.4 Total potential of Helmert condensation

This is obtained by subtracting equations D.5 and D.3 from each other.
The result — which applies in the exterior space2 — is

ext ext ext

δVHelmert (ϕ, λ, r) = Vcond (ϕ, λ, r) − Vtop (ϕ, λ, r) = −GρR2 ·
x ∑︂ ∞ (︂ )︂n+1 (︃
)︁ H2 (︁ 1 )︁ 3
)︃
R (︁ 1 1 H
· r (n + 2) − 1 2 + 6 (n + 2) (n + 1) − 3 + ··· ·
σ 2 R R3
n=0

· Pn (cos ψ) dσ ′ =
x ∑︂
∞ (︂ )︂n+1 (︃ )︃
R 1 1 H3
= −Gρ r 2
2
nH + 6 n (n + 3) + · · · Pn (cos ψ) dσ ′ .
σ R
n=0

2 Theoretically speaking, the exterior space is the space outside a geocentric sphere
that encloses all of the Earth’s topography. Practice is less restrictive.

í Õ ! ¤. û
436 Helmert condensation

Often, we define the degree constituents of powers of height H (compare

asteosuus- the degree constituent equation 3.8), as follows:
yhtälö x
2n + 1
ν
Hn (ϕ, λ) = Hν (ϕ ′ , λ ′ )Pn (cos ψ) dσ ′ , (D.6)
4π σ

with which it holds that

∑︂
∞
Hν (ϕ, λ) = Hνn (ϕ, λ).
n=0

Then
ext
δVHelmert =
∑︂
∞ (︂ )︂n+1
R 1
(︃
H3n
)︃
1 2 1
= −4πGρ r nHn + 6 n (n + 3) + ··· .
2n + 1 2 R
n=0

If the topography is constant, all terms vanish for which n ̸= 0. In the

above expansion, in that case the first and second terms also vanish. In
this case n = 0, the following terms do not even exist: the expansion
D.2 is the binomial expansion
(R + H)3 = R3 + 3R2 H + 3RH2 + H3 .
So
ext
δVHelmert =0
as was to be expected according to section 1.4: condensing a spherical
shell will not change the exterior field.

^ D.4.1 The gravity effect of Helmert condensation

Let us calculate the effect of the Helmert condensation potential on
gravity anomalies:

∂ 2
∆gext
Helmert = − δV ext − δV ext ≈
∂r Helmert r Helmert
∑︂ 1 (︂ − (n + 1) 2 )︂ (︂ R )︂n+1 (︃
∞
H3n
)︃
1 2 1
≈ 4πGρ r +r r nHn + 6 n (n + 3) + ··· =
2n + 1 2 R
n=0

1 ∑︂ n − 1 R n+1 1
∞
H3n
(︂ )︂ (︃ )︃
2 1
= −4πGρ · r nHn + 6 n (n + 3) + · · · . (D.7)
2n + 1 r 2 R
n=0

í Õ ! ¤. û
 Total potential of Helmert condensation
437
Now, n = 1 also gives a zero result, expected as gravity anomalies do
not contain any constituents of degree number 1.
Result D.7 is approximate and not to be used on or close to the
topography. Note the strong dependence upon n: the gravity effect of
Helmert condensation is dominated by short wavelengths, i. e., the local
features of the topography.

^ D.4.2 The interior potential of Helmert condensation

This quantity is evaluated on the level of the geoid. It represents the indirect epäsuora
effect of Helmert condensation, the shift of the geoid surface in space vaikutus
caused by the mass shifts. Subtract equations D.5 and D.4 from each
other:

int ext int

δVHelmert (ϕ, λ, R) = Vcond (ϕ, λ, R) − Vtop (ϕ, λ, R) =
x ∑︂ ∞ (︃ )︃
H H2 1 H3
= GρR2 + 2 + 3 3 Pn (cos ψ) dσ ′ −
σ R R R
n=0
x ∑︂ ∞ (︃ )︃
H 1 H2 1 H3
− GρR 2
− (n − 1) 2 + 6 (n − 1) n 3 − · · · Pn (cos ψ) dσ ′ =
σ R 2 R R
n=0
x ∑︂ ∞ (︃ )︃
1 1 H3
= Gρ 2
2
(n + 1) H − 6 (n − 2) (n + 1) + · · · Pn (cos ψ) dσ ′ .
σ R
n=0

Using again the definition of the degree constituents of the powers of

height H, equation D.6, we obtain
∑︂
∞
n+1 1 2
(︃
H3n
)︃
int 1
δVHelmert = 4πGρ H − (n − 2) + ··· ,
2n + 1 2 n 6 R
n=0

from which one obtains with Bruns equation 5.2 the indirect effect of
Helmert condensation:
int
δVHelmert
δNHelmert = γ =

4πGρ ∑︂ n + 1 1 2
∞
H3n
(︃ )︃
1
= γ H − (n − 2) + · · · . (D.8)
2n + 1 2 n 6 R
n=0

í Õ ! ¤. û
438 Helmert condensation

epäsuora The term n = 0 yields the indirect effect of a constant terrain H = H = H0 :

vaikutus using only the first term inside the parentheses yields

2πGρ 2
δNHelmert, const = γ H,

which cannot be neglected.

^ D.5 The dipole method

As a sanity test, we may describe the effect of Helmert condensation
in first approximation as a dipole-density layer field µ. The topographic
mass, surface density κ = ρH, moves downwards by on average 12 H.
The effect would be the same if the mean sea level3 were covered by a
double mass-density layer

µ = 21 ρH2 . (D.9)

The potential of this layer is, in spherical approximation (equation 1.18):

x x
∂ 1 ∂ 1
(︂ )︂ (︂ )︂
V=G µ dS ≈ GR2 µ dσ.
S ∂n ℓ σ ∂n ℓ

Written more explicitly

x (︃ )︃
2 ∂ 1
VP = GR µQ dσQ .
σ ∂rQ ℓPQ

We use the expansion into Legendre polynomials, equation 8.6:

1 ∑︂ rQ n+1
∞ (︂
1
)︂
=r rP Pn (cos ψPQ ),
ℓPQ Q
n=0

differentiate with respect to rQ , and substitute:

x
1 ∑︂
∞ (︂ r )︂n+1
Q
2
VP = GR 2
µQ n r Pn (cos ψPQ ) dσQ .
σ rQ n=0
P

3 In fact, a better place for this replacement layer would be the 41 H level.

í Õ ! ¤. û
The dipole method
439
By substituting into this equation D.9 for the double mass-density layer
µQ we obtain, by taking the limit rP , rQ ↓ R:

1 ∑︂
∞ x
V= n (2πGρH) HPn (cos ψ) dσ ′ =
4π σ
n=0
1 ∑︂
∞ x
= n AB HPn (cos ψ) dσ ′ .
4π σ
n=0

Here, we have left off the designations P and Q again as they are no
longer needed for clarity.
The symbol AB denotes the attraction of a Bouguer plate of thickness
H and matter density ρ.
Let us develop the quantity (AB H) into a spherical-harmonic expan- pallofunktio-
sion. According to degree constituent equation 3.8: kehitelmä
asteosuus-
x
2n + 1 yhtälö
(AB H)n = (AB H) Pn (cos ψ) dσ ′ ,
4π σ

yielding
∑︂
∞
n
V= (A H) ≈ 1 (A H) ,
2n + 1 B n 2 B
n=0

at least for the higher n values, i. e., regionally though not globally. The
term n = 0 vanishes, which is not realistic.
Thus we obtain again an estimate for the indirect effect of Helmert epäsuora
condensation. In geoid computation by means of this method this vaikutus
represents the shift in geoid surface caused by the condensation, which
must be undone, i. e., accounted for with the opposite algebraic sign. In
other words, when looked upon as a remove–restore method, it constitutes
its “restore” step:

V AB H πGρH2
δNHelmert = γ ≈ 21 γ = γ .

í Õ ! ¤. û
440 Helmert condensation

For comparison, the more precise expansion D.8 yields in approxima-

tion for larger n values

4πGρ 1 ∑︂ n + 1 2 πGρ ∑︂ 2
∞ ∞
πGρH2
δNHelmert ≈ γ ·2 Hn ≈ γ Hn ≈ γ ,
2n + 1
n=0 n=0

essentially the same result.

í Õ ! ¤. û
^ The Laplace equation in spherical
co-ordinates

^ E.1 Derivation
E
Consider a small volume element with sizes in co-ordinate directions
def
of ∆ϕ, ∆λ, and ∆r. Look at the difference in flux of vector field a = ∇V
between what comes in and what goes out through opposite faces.
We do the analogue of what was shown in subsection 1.12.4, using a
body or volume element with surfaces aligned along co-ordinate lines,
allowing the size of the element to go to zero in the limit, and exploiting
integral theorem 1.19 of Gauss. The quantity div a = ∆V is a source
density in space, and its average value multiplied by the volume of an
element must equal the total flux through the surfaces of the element.
{︁ }︁
Define at the location of the body an orthonormal basis e1 , e2 , e3 of ortonormaali
type “north east up”. The vector e1 points to the local north, the vector kanta
e2 to the east, and the vector e3 “up”, in the radial direction. Now we
may write
a = a1 e1 + a2 e2 + a3 e3 .
Part of the difference in flux f between opposing faces is due to a
change in the normal component of a between the faces, part is due to a
difference in face surface area ω:
I II
⏟ ⏞⏞ ⏟ ⏟ ⏞⏞ ⏟
+ − + − + −
f − f ≈ ω (a − a ) + a (ω − ω ).

– 441 –
442 The Laplace equation in spherical co-ordinates

e1 e3
a
e2
r∆ϕ

r cos ϕ ∆r

r r∆λ cos ϕ

∆ϕ

∆λ
λ

Equatorial plane

Figure E.1. Gauss integral theorem applied to a co-ordinate aligned volume

^ element.

See figure E.1.

◦ Latitudinal direction, ϕ, “south–north”:

ω−
ϕ = r cos ϕ∆r∆λ, ω+
ϕ = r cos(ϕ + ∆ϕ)∆r∆λ,

difference
ω+ −
ϕ − ωϕ ≈ −r sin ϕ∆ϕ · ∆r∆λ.

Multiply by
∂V 1 ∂V
a1 = = r
∂ (rϕ) ∂ϕ

í Õ ! ¤. û
Derivation
443
and divide by element volume r2 cos ϕ∆r∆ϕ∆λ, yielding
tan ϕ ∂V
∆IIϕ V = − .
r2 ∂ϕ
This of course in addition to the first contribution
a+
1 − a1
−
∆Iϕ V = ⟨∇a1 · e1 ⟩ = ,
r · ∆ϕ
with [︃ ]︃+ [︃ ]︃+
∂V 1 ∂V
a+
1 − a−
1 = = r ,
∂ (rϕ) −
∂ϕ −
yielding
[︂ ]︂+
∂
V
1 1 ∂ϕ − 1 ∂2 V
∆Iϕ V = r · r · ≈ .
∆ϕ r2 ∂ϕ2
◦ Longitudinal direction, λ, “west–east”: no change in surface area
ωλ = r∆r∆ϕ because of rotational symmetry:

∆IIλ V = 0.

We only have

a+
2 − a2
−
∆Iλ V = ⟨∇a2 · e2 ⟩ = ,
r cos ϕ · ∆λ
with
[︃ ]︃+ [︂ ]︂+
∂V 1 ∂V
a+
2 − a−
2 = = .
∂ (λr cos ϕ) −
r cos ϕ ∂λ −

Substitution yields
[︁ ∂
]︁+
1 1 ∂λ
V − 1 ∂2 V
∆Iλ V = · · ≈ .
r cos ϕ r cos ϕ ∆λ r2 cos ϕ ∂λ2
2

◦ In the radial direction, the surface areas of opposing faces —

“inner–outer” — are
2 2
ω−
r = r cos ϕ∆ϕ∆λ, ω+
r = (r + ∆r) cos ϕ∆ϕ∆λ,

í Õ ! ¤. û
444 The Laplace equation in spherical co-ordinates

the difference being

ω+ −
r − ωr ≈ 2r∆r · cos ϕ∆ϕ∆λ.

Multiply by
∂V
a3 =
∂r
and divide by the volume of the element r2 cos ϕ∆r∆ϕ∆λ, yielding
for the second contribution to the Laplace operator
2 ∂V
∆IIr V = r .
∂r
This in addition to the first contribution
[︁ ∂ ]︁+
a+ − a− ∂r
V − ∂2 V
∆Ir V = ⟨∇a3 · e3 ⟩ = 3 3
= ≈ ,
∆r ∆r ∂r2
in which ]︂+
∂V
[︂
a+ −
3 − a3 = .
∂r −

All of this gives us the end result

∆V = ∆Ir V + ∆Iλ V + ∆Iϕ V + ∆IIr V + ∆IIϕ V =

∂2 V 1 ∂2 V 1 ∂2 V 2 ∂V tan ϕ ∂V
= + + + − ,
∂r2 r2 cos2 ϕ ∂λ2 r2 ∂ϕ2 r ∂r r2 ∂ϕ
equivalent to equation 2.8.

^ E.2 Solution
^ E.2.1 Separating the radial dependency
Let us attempt separation of variables as follows:

V(ϕ, λ, r) = R(r)Y(ϕ, λ).

2/︁
Substitution into equation 2.8 and multiplication by r RY yields
(︃ )︃ (︃ )︃
1 2 ∂2 R ∂R 1 1 ∂2 Y ∂2 Y ∂Y
r + 2r =− + − tan ϕ .
R ∂r2 ∂r Y cos2 ϕ ∂λ2 ∂ϕ2 ∂ϕ

í Õ ! ¤. û
 Solution
445
This must again apply for all values r, ϕ, and λ and thus can only be a
constant, p. This yields two equations:
(︃ )︃
2
2∂ R ∂R
r + 2r − pR = 0,
∂r2 ∂r
(︃ )︃
1 ∂2 Y ∂2 Y ∂Y
+ − tan ϕ + pY = 0.
cos2 ϕ ∂λ2 ∂ϕ2 ∂ϕ

For the first equation we try a power law,

R(r) = rq ,

yielding

q (q − 1) rq + 2qrq − prq = 0 =⇒ (q (q + 1) − p) rq = 0

with the solution

p = q (q + 1) .
Solving the second equation for Y(ϕ, λ),
(︃ )︃
1 ∂2 Y ∂2 Y ∂Y
+ − tan ϕ + q (q + 1) Y = 0, (E.1)
cos2 ϕ ∂λ2 ∂ϕ2 ∂ϕ

is trickier. It turns out that q must be an integer. One finds, for n ∈ N0 ,

that there are non-negative solutions q = n and negative solutions
q = − (n + 1), with n = 0, 1, 2, . . . With this, the full set of special
solutions is

˜︁n,1 = rn Yn (ϕ, λ),

V ˜︁n,2 = Yn (ϕ, λ) ,
V n ∈ N0 ,
rn+1
equations 2.9.

^ E.2.2 Solving for surface harmonics

Both solutions q, the non-negative and the negative one, yield on
substitution into equation E.1 the same equation for n:
(︃ )︃
1 ∂2 Y ∂2 Y ∂Y
+ − tan ϕ + n (n + 1) Y = 0.
cos2 ϕ ∂λ2 ∂ϕ2 ∂ϕ

í Õ ! ¤. û
446 The Laplace equation in spherical co-ordinates

We attempt separation of variables:

Y(ϕ, λ) = F(ϕ)L(λ).
/︂
2
Substitution and multiplication by cos ϕ FL yields
(︃ )︃
cos2 ϕ ∂2 F ∂F 1 ∂2 L
− tan ϕ + n (n + 1) F = − .
F ∂ϕ2 ∂ϕ L ∂λ2
Both sides must be again equal to the same constant, which we shall
assume positive and call m2 :
(︃ )︃
∂2 F ∂F m2
− tan ϕ + n (n + 1) − F = 0,
∂ϕ2 ∂ϕ cos2 ϕ
∂2 L
+ m2 L = 0.
∂λ2
The first equation is known as Legendre’s equation. Its solutions are the
Legendre functions Pnm (sin ϕ), with the integer m = 0, 1, . . . , n. The
harmoninen second is the classical harmonic oscillator, with solutions1
värähtelijä
Lm,1 (λ) = cos mλ, Lm,2 (λ) = sin mλ.
pinta- With this, we find for the surface spherical harmonics the linear combi-
pallofunktio nations
∑︂
n
Yn (ϕ, λ) = Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) .
m=0

The general solution is now formed as follows:

∑︂
∞ ∑︂
n
V1 (ϕ, λ, r) = rn Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) ,
n=0 m=0
∑︂
∞
1 ∑︂
n
V2 (ϕ, λ, r) = Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) .
rn+1
n=0 m=0

pallofunktio- Here, anm and bnm are the spherical-harmonic coefficients specifying
kerroin the linear combination of special solutions. Only the second solution is
physically realistic for representing the Earth’s gravitational field, going
to zero at infinity r → ∞.
1 This also explains why m must be an integer: the longitude λ is circular with a period
of 2π.

í Õ ! ¤. û
 Bibliography

ABCDEFGHIKLMNOPRSTVWY
A
M. Abrehdary, L. E. Sjöberg, and M. Bagherbandi. The spherical terrain
correction and its effect on the gravimetric-isostatic Moho determination.
Geophysical Journal International, 204(1):262–273, 2016. URL
https://doi.org/10.1093/gji/ggv450. 139
G. d’Agostino, S. Desogus, A. Germak, C. Origlia, D. Quagliotti, G. Berrino,
G. Corrado, V. d’Errico, and G. Ricciardi. The new IMGC-02 transportable
absolute gravimeter: Measurement apparatus and applications in
geophysics and volcanology. Annals of Geophysics, 51(1):39–49, 2008. URL
https://doi.org/10.4401/ag-3038. 306
Altimetry, Retracking. Radar altimetry tutorial & toolbox. ESA and CNES.
URL http://www.altimetry.info/radar-altimetry-tutorial/data-flow/data-
processing/retracking/. Accessed 1st March, 2022. 363
O. B. Andersen, P. Knudsen, and P. Berry. The DNSC08GRA global marine
gravity field from double retracked satellite altimetry. Journal of Geodesy, 84:
191–199, 2010. URL https://doi.org/10.1007/s00190-009-0355-9. 244
M. de Angelis, A. Bertoldi, L. Cacciapuoti, A. Giorgini, G. Lamporesi,
M. Prevedelli, G. Saccorotti, F. Sorrentino, and G. M. Tino. Precision
gravimetry with atomic sensors. Measurement Science and Technology, 20(2):
022001, 2009. URL http://dx.doi.org/10.1088/0957-0233/20/2/022001.
306

– 447 –
448 ABCDEFGHIKLMNOPRSTVWY Bibliography

B
G. Balmino, N. Vales, S. Bonvalot, and A. Briais. Spherical harmonic modeling
to ultra-high degree of Bouguer and isostatic anomalies. Journal of Geodesy,
86:499–520, 2012. URL https://doi.org/10.1007/s00190-011-0533-4. 62, 138
T. B. Bedada. Absolute geopotential height system for Ethiopia. PhD thesis,
University of Edinburgh, 2010. URL https:
//era.ed.ac.uk/bitstream/handle/1842/4726/Bedada2010_small.pdf.
Accessed 22nd September, 2021. 244
N. Benitez, T. J. Broadhurst, H. C. Ford, M. Clampin, G. Hartig, G. D.
Illingworth, et al. Hubble Looks Through Cosmic Zoom Lens, 2003. URL
https://esahubble.org/images/opo0301a/. © 2003 ESA/Hubble (CC BY
4.0). Accessed 22nd September, 2021. 3
BGI, EGM2008. EGM2008 anomaly maps visualization. URL http://bgi.obs-
mip.fr/data-products/outils/egm2008-anomaly-maps-visualization/.
Accessed 17th September, 2021. 124, 133
BGI, WGM2012. WGM2012 maps visualization/extraction. URL
http://bgi.obs-mip.fr/data-products/outils/wgm2012-maps-
visualizationextraction/. Accessed 17th September, 2021. 151
M. Bilker-Koivula. Development of the Finnish Height Conversion Surface
FIN2005N00. Nordic Journal of Surveying and Real Estate Research, 7(1):76–88,
2010. URL https://journal.fi/njs/article/download/3663/3432. Accessed
22nd September, 2021. 391
M. Bilker-Koivula and M. Ollikainen. Suomen geoidimallit ja niiden
käyttäminen korkeuden muunnoksissa. Research note (in Finnish) 29,
Finnish Geodetic Institute, 2009. URL https://www.maanmittauslaitos.fi/
sites/maanmittauslaitos.fi/files/fgi/GLtiedote29.pdf. Accessed 11th May,
2019. 242, 324
G. P. Bottoni and R. Barzaghi. Fast collocation. Bulletin Géodésique, 67(2):
119–126, 1993. URL https://doi.org/10.1007/BF01371375. 284
V. V. Brovar, M. I. Yurkina, M. Heifets, M. S. Molodensky, and H. Moritz. M. S.
Molodensky In Memoriam. Online PDF, Mitteilungen der geodätischen
Institute der Technischen Universität Graz Folge 88, 2000. URL
http://www.helmut-moritz.at/SciencePage/Molodensky.pdf. Helmut
Moritz and Maria I. Yurkina, editors. Accessed 1st March, 2020. 168

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
449
J. M. Brozena. The Greenland Aerogeophysics Project: Airborne gravity,
topographic and magnetic mapping of an entire continent. International
Association of Geodesy Symposia 110, pages 203–214, Vienna, Austria, 20th
August, 1992. Springer, New York, NY. URL
https://doi.org/10.1007/978-1-4613-9255-2_19. 314
J. M. Brozena and M. F. Peters. State-of-the-art airborne gravimetry.
International Association of Geodesy Symposia 113, pages 187–197, Graz,
Austria, 1994. Springer-Verlag. URL
https://doi.org/10.1007/978-3-642-79721-7_20. 314
J. M. Brozena, M. F. Peters, and R. Salman. Arctic airborne gravity
measurements program. In Segawa et al. (1996), pages 131–146. URL
https://doi.org/10.1007/978-3-662-03482-8_20. 314
H. Bruns. Die Figur der Erde: Ein Beitrag zur europäischen Gradmessung.
Stankievicz, Berlin, 1878. URL
https://play.google.com/books/reader?id=DP0-
AAAAYAAJ&hl=en&pg=GBS.PP5. Accessed 31st January, 2020. 92
B. Bucha, C. Hirt, and M. Kuhn. Cap integration in spectral gravity forward
modelling: near- and far-zone gravity effects via Molodensky’s truncation
coefficients. Journal of Geodesy, 93:65–83, 2019. URL
https://doi.org/10.1007/s00190-018-1139-x. 220

C
L. Caesar, S. Rahmstorf, A. Robinson, G. Feulner, and V. S. Saba. Observed
fingerprint of a weakening Atlantic Ocean overturning circulation. Nature,
556:191–196, 2018. URL https://doi.org/10.1038/s41586-018-0006-5. 332
H. Cavendish. Experiments to determine the density of the earth. Philosophical
Transactions of the Royal Society, 88, 1798. URL
https://doi.org/10.1098/rstl.1798.0022. 4
CHAMP Mission. CHAMP – CHAllenging Minisatellite Payload. Deutsches
Geoforschungszentrum, Helmholtz-Zentrum Potsdam. URL
https://www.gfz-potsdam.de/champ/. Accessed 1st March, 2020. 367
B. J. Coakley, S. C. Kenyon, and R. Forsberg. Updating the Arctic Gravity
Project grid with new airborne and Extended Continental Shelf data. AGU
Fall Meeting Abstracts, page C3, December 2013. URL
https://ui.adsabs.harvard.edu/abs/2013AGUFM.G13C..03C/abstract.
Accessed 17th January, 2020. 314

í ¤. û
450 ABCDEFGHIKLMNOPRSTVWY Bibliography

J. J. O’Connor and E. F. Robertson. George Green (1793–1841). MacTutor

History of Mathematics archive, School of Mathematics and Statistics,
University of St Andrews, Scotland, 1998. URL
https://mathshistory.st-andrews.ac.uk/Biographies/Green/. Accessed
22nd September, 2021. 28

D
B. D. DeJong, P. R. Bierman, W. L. Newell, Tammy, M. Rittenour, S. A. Mahan,
G. Balco, and D. H. Rood. Pleistocene relative sea levels in the Chesapeake
Bay region and their implications for the next century. GSA Today, 25(8):
4–10, 2015. URL https://doi.org/10.1130/GSATG223A.1. 333
H. Denker. Evaluation and improvement of the EGG97 quasigeoid model for
Europe by GPS and leveling data. In Vermeer and Ádám (1998), pages
53–61. 390

E
Eiffel Tower, 72 names. List of the 72 names on the Eiffel Tower. URL https:
//en.wikipedia.org/wiki/List_of_the_72_names_on_the_Eiffel_Tower.
Accessed 7th April, 2019. 16, 17, 33, 45, 55, 88, 422
M. Ekman. Postglacial rebound and sea level phenomena with special
reference to Fennoscandia and the Baltic Sea. In Kakkuri (1993), pages 7–70.
329, 383
Encyclopaedia Britannica, Moho. URL
https://www.britannica.com/science/Moho. Accessed 22nd January, 2020.
156
L. Eötvös. Three Fundamental Papers of Loránd Eötvös. Loránd Eötvös
Geophysical Institute of Hungary, 1998. ISBN 963-7135-02-2. Editor Zoltán
Szabó. 317
European geoid calculations. Leibniz Universität Hannover, Institute of
Geodesy. URL
https://www.ife.uni-hannover.de/en/research/main-research-focus/re
gional-gravity-field-and-geoid-modelling/european-geoid-calculations/.
Accessed 11th May, 2019. 390

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
451
F
G. Farmelo. The Strangest Man. Basic Books, reprint edition, 2011. ISBN
978-0-4650-2210-6. 26
W. E. Farrell and J. A. Clark. On postglacial sea level. Geophysical Journal of the
Royal Astronomical Society, 46(3):647–667, 1976. URL
https://doi.org/10.1111/j.1365-246X.1976.tb01252.x. 334
W. E. Featherstone. Software for computing five existing types of
deterministically modified integration kernel for gravimetric geoid
determination. Computers and Geosciences, 29:183–193, 2003. URL
https://doi.org/10.1016/S0098-3004(02)00074-2. 222
FFTW Home Page. URL https://www.fftw.org. Accessed 15th May, 2019. 429
R. Forsberg. A study of terrain reductions, density anomalies and geophysical
inversion methods in gravity field modelling. Report 355, Ohio State
University, Department of Geodetic Science and Surveying, 1984. URL
https://earthsciences.osu.edu/sites/earthsciences.osu.edu/files/report-
355.pdf. Accessed 17th February, 2020. 246
R. Forsberg and J. Kaminskis. Geoid of the Nordic and Baltic region from
gravimetry and satellite altimetry. In Segawa et al. (1996), pages 540–547.
URL https://doi.org/10.1007/978-3-662-03482-8_72. 391
R. Forsberg and G. Strykowski. NKG Gravity Data Base and NKG Geoid.
Technical University of Denmark, DTU Space, National Space Institute,
2010. URL
https://www.nordicgeodeticcommission.com/wp-content/uploads/20
14/10/8-WG_geoid_2010_March_presentation_ForsbergStrykowski.pdf.
Accessed 11th May, 2019. 391
R. Forsberg, A. V. Olesen, F. Ferraccioli, T. Jordan, H. Corr, and K. Matsuoka.
PolarGap 2015/16 - Filling the GOCE polar gap in Antarctica and ASIRAS
flight around South Pole. Final report, European Space Agency, 2017. URL
https://earth.esa.int/documents/10174/134665/PolarGap-2015-2016-
final-report. Accessed 8th January, 2020. 370
R. Forsberg and C. C. Tscherning. An overview manual for the GRAVSOFT
Geodetic Gravity Field Modelling Programs, 2008. URL
https://www.academia.edu/9206363/An_overview_manual_for_the_G
RAVSOFT_Geodetic_Gravity_Field_Modelling_Programs. Accessed 11th
May, 2019. 242

í ¤. û
452 ABCDEFGHIKLMNOPRSTVWY Bibliography

R. Forsberg and M. Vermeer. A generalized Strang van Hees approach to fast

geopotential inversion. Manuscripta geodaetica, 17:302–314, 1992. 238
J.-P. Friedelmeyer. Du côté des lettres (2) : une lettre de Sophie Germain à Carl
Friedrich Gauss (20 février 1807), et la réponse de celui-ci (30 avril 1807),
2014. URL https://images.math.cnrs.fr/Du-cote-des-lettres-une-lettre-de-
Sophie-Germain-a-Carl-Friedrich-Gauss-20. Accessed 22nd September,
2021. 92
M. Frigo and S. G. Johnson. The design and implementation of FFTW3.
Proceedings IEEE, 93(2):216–231, 2005. URL
http://www.fftw.org/fftw-paper-ieee.pdf. Accessed 14th February, 2020.
429

G
E. S. Garcia, D. T. Sandwell, and W. H. Smith. Retracking CryoSat-2, Envisat
and Jason-1 radar altimetry waveforms for improved gravity field recovery.
Geophysical Journal International, 2014. URL
https://doi.org/10.1093/gji/ggt469. 244
A. Gatti, M. Reguzzoni, F. Sansò, and F. Migliaccio. Space-wise grids of
gravity gradients from GOCE data at nominal satellite altitude. Presented
at the 5th International GOCE User Workshop, UNESCO, Paris, France,
25th – 28th November, 2014. URL
https://www.researchgate.net/publication/275029640_SPACE-
WISE_GRIDS_OF_GRAVITY_GRADIENTS_FROM_GOCE_DATA_AT_N
OMINAL_SATELLITE_ALTITUDE. Accessed 11th May, 2019. 281
O. R. Godø, A. Samuelsen, G. J. Macaulay, R. Patel, S. S. Hjøllo, J. Horne,
S. Kaartvedt, and J. A. Johannessen. Mesoscale eddies are oases for higher
trophic marine life. PLoS One, 7(1):e30161, 2012. URL
https://doi.org/10.1371/journal.pone.0030161. 333
GRACE Follow-On Mission. Jet Propulsion Laboratory. URL
https://gracefo.jpl.nasa.gov/mission/overview/. Accessed 1st March,
2020. 369
GRACE Mission. Measuring Earth’s Surface Mass and Water Changes. Jet
Propulsion Laboratory. URL https://grace.jpl.nasa.gov/. Accessed 1st
March, 2020. 368

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
453
GRACE Mission, hydrology. NASA. URL
https://commons.wikimedia.org/wiki/File:
Global_Gravity_Anomaly_Animation_over_LAND.gif. Accessed 1st March,
2020. 369
G. Green. An Essay on the Application of Mathematical Analysis to the Theories of
Electricity and Magnetism. 1828. URL
https://play.google.com/books/reader?id=GwYXAAAAYAAJ. Accessed
23rd April, 2019. 28
Green’s Windmill. Green’s windmill and science centre. URL
https://www.greensmill.org.uk/. Accessed 10th May, 2019. 28
GWR Instruments, Inc., iGRAV® Gravity Sensors. URL
https://www.gwrinstruments.com/igrav-gravity-sensors.html. Accessed
22nd September, 2022. 310

H
R. Haagmans, E. de Min, and M. van Gelderen. Fast evaluation of convolution
integrals on the sphere using 1D FFT, and a comparison with existing
methods for Stokes’ integral. Manuscripta geodaetica, 18:227–241, 1993. 239
P. Häkli, J. Puupponen, H. Koivula, and M. Poutanen. Suomen geoidimallit ja
niiden käyttäminen korkeuden muunnoksissa. Research note (in
Finnish) 30, Finnish Geodetic Institute, 2009. URL https://www.maanmitt
auslaitos.fi/sites/maanmittauslaitos.fi/files/fgi/GLtiedote30.pdf.
Accessed 26th January, 2020. 324
J. C. Harrison and M. Dickinson. Fourier transform methods in local gravity
modelling. Bulletin Géodésique, 63:149–166, 1989. URL
https://doi.org/10.1007/BF02519148. 246
M. Heikkinen. Solving the shape of the Earth by using digital density models.
Report 81:2, Finnish Geodetic Institute, Helsinki, 1981. 99, 100, 102, 312
W. A. Heiskanen. The latest achievements of physical geodesy. Journal of
Geophysical Research, 65(9):2827–2836, 1960. URL
https://doi.org/10.1029/JZ065i009p02827. 152
W. A. Heiskanen and H. Moritz. Physical Geodesy. W. H. Freeman and
Company, San Francisco, London, 1967. 35, 53, 55, 66, 68, 76, 77, 78, 92, 95,
96, 102, 103, 104, 110, 113, 116, 152, 154, 175, 177, 178, 183, 191, 192, 198, 199,
204, 208, 287, 392, 423

í ¤. û
454 ABCDEFGHIKLMNOPRSTVWY Bibliography

W. F. Hermans. Beyond Sleep. Harry N. Abrams, reprint edition, 2007. ISBN

978-1-5856-7583-8. 143
C. Hirt and M. Kuhn. Band-limited topographic mass distribution generates
full-spectrum gravity field: Gravity forward modeling in the spectral and
spatial domains revisited. Journal of Geophysical Research: Solid Earth, 119,
2014. URL https://doi.org/10.1002/2013JB010900. 138
R. A. Hirvonen. The continental undulations of the geoid. PhD thesis, Helsinki
University of Technology, 1934. Finnish Geodetic Institute publication 19.
391
R. A. Hirvonen. Relative Bestimmungen der Schwerkraft in Finnland in den Jahren
1931, 1933 und 1935. Publication 23, Finnish Geodetic Institute, Helsinki,
1937. 294
B. Hofmann-Wellenhof and H. Moritz. Physical Geodesy. Springer-Verlag Wien
GmbH, 2006. Second, revised edition. 392
Humboldt University Berlin. Friedrich Robert Helmert with a relative
pendulum, 2017. URL https://www.researchgate.net/publication/3189949
32_Friedrich_Robert_Helmert_founder_of_modern_geodesy_on_the_occ
asion_of_the_centenary_of_his_death. © 2017 Humboldt-Universität zu
Berlin, Universitätsbibliothek (CC BY 3.0). Accessed 19th May, 2019. 142
E. Hytönen. Absolute gravity measurement with long wire pendulum. PhD thesis,
University of Helsinki, 1972. Finnish Geodetic Institute publication 75. 295

I
International Intercomparison of Absolute Gravimeters. University of
Luxembourg, European Center for Geodynamics and Seismology. URL
http://www.ecgs.lu/international-intercomparison-of-absolute-
gravimeters/. Accessed 6th February, 2020. 308
ISG, Geoid Schools. International Service for the Geoid. URL
http://www.isgeoid.polimi.it/Schools/schools.html. Accessed 1st March,
2020. 242

K
E. Kääriäinen. The Second Levelling of Finland in 1935–1955. Publication 61,
Finnish Geodetic Institute, Helsinki, 1966. 163, 164, 179

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
455
J. Kääriäinen and H. Ruotsalainen. Tilt Measurements in the Underground
Laboratory Lohja 2, Finland, in 1977–1987. Publication 110, Finnish Geodetic
Institute, Helsinki, 1989. 382
J. Kakkuri, editor. Geodesy and Geophysics, lecture notes, NKG Autumn School
1992, Finnish Geodetic Institute publication 115, 1993. 450, 460, 461
J. Kakkuri. Surveyor of the Globe – Story of the Life of V. A. Heiskanen. Finnish
National Land Survey, 2008. URL
https://readymag.com/u95015526/508134/. Accessed 13th May, 2019. 143,
392
S. C. Kenyon, R. Forsberg, A. V. Olesen, and S. A. Holmes. NGA’s use of
aerogravity to advance the next generation of Earth Gravitational Models.
AGU Fall Meeting Abstracts, page A4, December 2012. URL
https://ui.adsabs.harvard.edu/abs/2012AGUFM.G12A..04K/abstract.
Accessed 11th May, 2019. 314
M. G. Kogan, M. Diament, A. Bulot, and G. Balmino. Thermal isostasy in the
South Atlantic Ocean from geoid anomalies. Earth and Planetary Science
Letters, 74:280–290, 1985. URL
https://doi.org/10.1016/0012-821X(85)90028-7. 153
M. Kuhn, W. Featherstone, and J. Kirby. Complete spherical Bouguer gravity
anomalies over Australia. Australian Journal of Earth Sciences, 56(2):213–223,
2009. URL https://espace.curtin.edu.au/handle/20.500.11937/34751.
Accessed 1st March, 2020. 138, 139
A. Kuivamäki, P. Vuorela, and M. Paananen. Indications of postglacial and
recent bedrock movements in Finland and Russian Karelia. Report YST-99,
Geological Survey of Finland, 1998. URL
http://tupa.gtk.fi/julkaisu/ydinjate/yst_099.pdf. Accessed 17th January,
2020. 339

L
A. E. H. Love. The Yielding of the Earth to Disturbing Forces. Proceedings of the
Royal Society of London A, 82(551):73–88, 1909. URL
https://doi.org/10.1098/rspa.1909.0008. 381
B. Lu, F. Barthelmes, S. Petrovic, C. Förste, F. Flechtner, Z. Luo, K. He, and
M. Li. Airborne gravimetry of GEOHALO mission: Data processing and

í ¤. û
456 ABCDEFGHIKLMNOPRSTVWY Bibliography

gravity field modeling. Journal of Geophysical Research: Solid Earth, 122:

10 586–10 604, 2017. URL https://doi.org/10.1002/2017JB014425. 314

M
J. Mäkinen, A. Engfeldt, L. Engman, B. G. Harsson, T. Oja, S. Rekkedal,
K. Røthing, P. Rouhiainen, H. Ruotsalainen, H. Skatt, G. Strykowski,
H. Virtanen, K. Wieczerkowski, and D. Wolf. The Fennoscandian Land
Uplift Gravity Lines: comparison of observed gravity change with observed
vertical motion and with GIA models. Report, Nordiska Kommissionen för
Geodesi, 2010. URL https://www.nordicgeodeticcommission.com/wp-cont
ent/uploads/2014/10/1-Makinen_et_al_land_uplift_gravity_lines.pdf.
Accessed 22nd September, 2021. 327
S. Märdla. Regional Geoid Modelling by the Least Squares Modified Hotine Formula
Using Gridded Gravity Disturbances. PhD thesis, Tallinn University of
Technology, 2017. URL https://digikogu.taltech.ee/en/Item/baba5f3f-
22ce-43b4-8f81-7b1fe015ac19. Accessed 17th September, 2021. 128, 212
P. J. Melchior. The Tides of the Planet Earth. Pergamon Press, Oxford, 1978.
ISBN 978-0-0802-6248-2. 379, 381
J. X. Mitrovica, M. E. Tamisiea, J. L. Davis, and G. A. Milne. Recent mass
balance of polar ice sheets inferred from patterns of global sea level change.
Nature, 409:1026–1029, February 2001. URL
https://doi.org/10.1038/35059054. 337
M. S. Molodensky, V. F. Eremeev, and M. I. Yurkina. Methods for the Study of the
External Gravitational Field and Figure of the Earth. Israel Program of Scientific
Translations, Jerusalem, 1962. (Transl. from Russian). 127, 171, 220, 392
D. Monniaux. Autograv CG5, 2011. URL https:
//commons.wikimedia.org/wiki/File:Autograv_CG5_P1150838.JPG. ©
2011 David Monniaux (GFDL). Accessed 13th May, 2019. 295
H. Moritz. Advanced Physical Geodesy. H. Wichmann Verlag, Karlsruhe, 1980.
ISBN 978-3-87-907106-7. 392, 432
W. H. Munk. The U.S. Commission on Ocean Policy, Testimony, 18th April
2002. URL https://govinfo.library.unt.edu/oceancommission/meetings/a
pr18_19_02/munk_statement.pdf. Accessed 14th May, 2019. 343

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
457
N
R. S. Nerem, D. P. Chambers, C. Choe, and G. T. Mitchum. Estimating mean
sea level change from the TOPEX and Jason altimeter missions. Marine
Geodesy, 33(1):435, 2010. URL
https://doi.org/10.1080/01490419.2010.491031. 344
NOAA, Ocean currents. How does the ocean affect climate and weather on
land? NOAA Ocean Exploration and Research. URL
https://oceanexplorer.noaa.gov/facts/climate.html. Accessed 1st March,
2020. 332

O
P.-A. Olsson, K. Breili, V. Ophaug, H. Steffen, M. Bilker-Koivula, E. Nielsen,
T. Oja, and L. Timmen. Postglacial gravity change in Fennoscandia – three
decades of repeated absolute gravity observations. Geophysical Journal
International, 217:1141–1156, 2019. URL
https://doi.org/10.1093/gji/ggz054. 327

P
R. L. Parker. The rapid calculation of potential anomalies. Geophysical Journal
of the Royal Astronomical Society, 31:447–455, 1972. URL
https://doi.org/10.1111/j.1365-246X.1973.tb06513.x. 247
M. K. Paul. A method of evaluating the truncation error coefficients for
geoidal height. Bulletin Géodésique, 110:413–425, 1973. URL
https://doi.org/10.1007/BF02521951. 221
N. K. Pavlis, S. A. Holmes, S. C. Kenyon, and J. K. Factor. The development
and evaluation of the Earth Gravitational Model 2008 (EGM2008). Journal of
Geophysical Research, 117(B4), 2012. URL
https://doi.org/10.1029/2011JB008916. 74
W. R. Peltier. Home page, W. R. Peltier, FRSC. URL
https://www.atmosp.physics.utoronto.ca/~peltier/data.php. Accessed
22nd September, 2021. 334
W. R. Peltier. Closure of the budget of global sea level rise over the GRACE
era: the importance and magnitudes of the required corrections for global
glacial isostatic adjustment. Quaternary Science Reviews, 28(17–18):
1658–1674, 2009. URL https://doi.org/10.1016/j.quascirev.2009.04.004.

í ¤. û
458 ABCDEFGHIKLMNOPRSTVWY Bibliography

Special issue: Quaternary Ice Sheet-Ocean Interactions and Landscape

Responses. 334
U. Pesonen. Relative Bestimmungen der Schwerkraft in Finnland in den Jahren
1926–1929. Publication 13, Finnish Geodetic Institute, Helsinki, 1930. 294
PIOMAS. Polar Science Center, PIOMAS Arctic Sea Ice Volume Reanalysis.
URL http://psc.apl.washington.edu/research/projects/arctic-sea-ice-
volume-anomaly/. Accessed 11th May, 2019. 365
M. Poutanen, M. Vermeer, and J. Mäkinen. The Permanent Tide in GPS
Positioning. Journal of Geodesy, 70:499–504, 1996. URL
https://doi.org/10.1007/BF00863622. 383
J. H. Pratt. II. On the attraction of the Himalaya mountains, and of the
elevated regions beyond them, upon the plumb-line in India. Philosophical
Transactions of the Royal Society of London, 145:53–100, 1855. URL
https://doi.org/10.1098/rstl.1855.0002. 142
J. H. Pratt. II. On the deflection of the plumb-line in India caused by the
attraction of the Himalaya mountains and the elevated regions beyond, and
its modification by the compensating effect of a deficiency of matter below
the mountain mass. Proceedings of the Royal Society of London, 9:493–496,
1859. URL https://doi.org/10.1098/rspl.1857.0096. 142
J. H. Pratt. On the degree of uncertainty which local attraction, if not allowed
for, occasions in the map of a country, and in the mean figure of the earth as
determined by geodesy ; a method of obtaining the mean figure free from
ambiguity by a comparison of the Anglo-Gallic, Russian, and Indian Arcs ;
and speculations on the constitution of the earth’s crust. Proceedings of the
Royal Society of London, 13:253–276, 1864. URL
https://doi.org/10.1098/rspl.1863.0061. 142
K. Predehl, G. Grosche, S. M. Raupach, S. Droste, O. Terra, J. Alnis, Th. Legero,
T. W. Hänsch, Th. Udem, R. Holzwarth, and H. Schnatz. A 920 km Optical
Fiber Link for Frequency Metrology at the 19th Decimal Place. Science, 27th
April 2012. URL https://doi.org/10.1126/science.1218442. 185, 186
I. Prutkin. Gravitational and magnetic models of the core-mantle boundary
and their correlation. Journal of Geodynamics, 45:146–153, 2008. URL https:
//www.researchgate.net/publication/257097255_Gravitational_and_mag
netic_models_of_the_core-mantle_boundary_and_their_correlation.
Accessed 1st March, 2020. 153

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
459
R
R. H. Rapp. The decay of the spectrum of the gravitational potential and the
topography for the Earth. Geophysical Journal International, 99(3):449–455,
1989. URL https://doi.org/10.1111/j.1365-246X.1989.tb02031.x. 280
J. Richer. Observations astronomiques et physiques faites en l’isle de Caïenne.
In Ouvrages de Mathematique de M. Picard. P. Gosse and J. Neaulme, 1731.
URL http://www.e-rara.ch/zut/content/pageview/815403. Accessed 11th
May, 2019. 294
R. A. Rohde. Post-glacial sea level rise, 2005. URL
https://commons.wikimedia.org/wiki/File:Post-Glacial_Sea_Level.png. ©
2005 Robert A. Rohde (GFDL). Accessed 21st July, 2019. 337
R. Rummel and F. Sansó, editors. Satellite Altimetry in Geodesy and
Oceanography. Proceedings, International Summer School of Theoretical Geodesy,
Lecture Notes in Earth Sciences, 50, Trieste, Italy, 25th May – 6th June 1992.
Springer-Verlag. URL https://doi.org/10.1007/BFb0117924. 333
H. Ruotsalainen. Interferometric water level tilt meter development in Finland
and comparison with combined earth tide and ocean loading models. Pure
and Applied Geophysics, 2017. URL
https://doi.org/10.1007/s00024-017-1562-6. 383

S
O. Sacks. Henry Cavendish: An early case of Asperger’s syndrome? Neurology,
57(7):1347–1347, 2001. URL https://doi.org/10.1212/WNL.57.7.1347. 4
E. J. O. Schrama. The Role of Orbit Errors in Processing of Satellite Altimeter Data.
PhD thesis, Delft University of Technology, 1989. URL
https://www.ncgeo.nl/downloads/33Schrama.pdf. Accessed 11th May,
2019. 355
A. R. Schweiger, R. Lindsay, L. Zhang, M. Steele, H. Stern, and R. Kwok.
Uncertainty in modeled Arctic sea ice volume. Journal of Geophysical Research,
116(C00D06), 2011. URL http://dx.doi.org/10.1029/2011JC007084. 365
D. Scuka. GOCE burning: Last orbital view. ESA blog, 2013. URL
https://blogs.esa.int/rocketscience/2013/11/11/goce-burning-last-
orbital-view/. Accessed 22nd September, 2021. 370

í ¤. û
460 ABCDEFGHIKLMNOPRSTVWY Bibliography

Sea Level Research Group. University of Colorado. URL

https://sealevel.colorado.edu/. Accessed 22nd September, 2021. 344
J. Segawa, H. Fujimoto, and S. Okubo, editors. Proceedings, IAG International
Symposium on Gravity, Geoid and Marine Geodesy (GraGeoMar96),
International Association of Geodesy Symposia 117, Tokyo, Japan, 30th
September – 5th October 1996. Springer-Verlag. 449, 451
SourceForge, Maxima. Maxima, a computer algebra system. URL
https://maxima.sourceforge.io/. Accessed 22nd September, 2021. 350
G. Spada and D. Melini. SELEN: a program for solving the “Sea Level
Equation”. User manual for version 2.9, Computational Infrastructure for
Geodynamics (CIG), 2015. URL
http://geodynamics.org/cig/software/selen/selen-manual.pdf. Accessed
1st March, 2020. 334

T
W. Torge. Gravimetry. de Gruyter, Berlin, New York, 1989. ISBN
978-3-11-010702-9. 392
W. Torge. Gravity and tectonics. In Kakkuri (1993), pages 131–172. 387
C. C. Tscherning and R. H. Rapp. Closed covariance expressions for gravity
anomalies, geoid undulations, and deflections of the vertical implied by
anomaly degree variances. Report 208, Dept. of Geodetic Science and
Surveying, The Ohio State University, Columbus, OH, USA, 1974. URL
https://earthsciences.osu.edu/sites/earthsciences.osu.edu/files/report-
208.pdf. Accessed 17th January, 2020. 280
Tytyri Mine Experience. URL
https://www.tytyrielamyskaivos.fi/en/mine-experience. Accessed 17th
September, 2021. 382

V
P. Vaníček and E. Krakiwsky. Geodesy – The Concepts. Elsevier Science
Publishers, second edition, 1987. ISBN 978-0-4448-7777-2. 392
F. A. Vening Meinesz. Gravity survey by submarine via Panama to Java. The
Geographical Journal, 71(2):144–156, 1928. URL
https://doi.org/10.2307/1782700. 294

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
461
M. Vermeer. Chronometric levelling. Report 83:2, Finnish Geodetic Institute,
1983a. 185, 306
M. Vermeer. A new Seasat altimetric geoid for the Baltic. Report 83:4, Finnish
Geodetic Institute, 1983b. 341
M. Vermeer. Geoid studies on Finland and the Baltic. PhD thesis, University of
Helsinki, 1984. Report 84:3, Finnish Geodetic Institute. 33, 112
M. Vermeer. FGI studies on satellite gravity gradiometry. 3. Regional high
resolution geopotential recovery in geographical coordinates using a Taylor
expansion FFT technique. Report 92:1, Finnish Geodetic Institute, 1992. 238
M. Vermeer. Geoid determination using frequency domain techniques. In
Kakkuri (1993), pages 183–200. 230
M. Vermeer and J. Ádám, editors. Proceedings, Second Continental Workshop on
the Geoid in Europe, Report 98:4, Finnish Geodetic Institute, Masala,
10th – 14th March, 1998. 450, 462
H. Virtanen. On superconducting gravimeter observations above 8 mHz at
the Metsähovi station. Report 98:5, Finnish Geodetic Institute, Masala, 1998.
309
H. Virtanen. Studies of Earth dynamics with superconducting gravimeter. PhD
thesis, University of Helsinki, 2006. URL
http://urn.fi/URN:ISBN:952-10-3057-7. Publication 133, Finnish Geodetic
Institute. Accessed 11th May, 2019. 309
H. Virtanen and J. Kääriäinen. The installation and first results from the
superconducting gravimeter GWR20 at the Metsähovi station, Finland.
Report 95:1, Finnish Geodetic Institute, Helsinki, 1995. 309
J. Vondrák, C. Ron, and V. Štefka. Earth orientation parameters based on
EOC-4 astrometric catalog. Acta Geodynamica et Geomaterialia, 7(3):245–251,
2010. URL https:
//www.irsm.cas.cz/materialy/acta_content/2010_03/2_Vondrak.pdf.
Accessed 20th February, 2020. 383
P. Vu, F. Frappart, J. Darrozes, V. Marieu, F. Blarel, G. Ramillien, P. Bonnefond,
and F. Birol. Multi-satellite altimeter validation along the French Atlantic
Coast in the Southern Bay of Biscay from ERS-2 to SARAL. Remote Sensing,
93(10), 2018. URL http://dx.doi.org/10.3390/rs10010093. 362

í ¤. û
462 ABCDEFGHIKLMNOPRSTVWY Bibliography

W
D. Watts. Fourifier, 2004. URL
https://ejectamenta.com/imaging-experiments/fourifier/. Accessed 17th
September, 2021. 241
L. Wen and D. L. Anderson. Layered mantle convection: A model for geoid
and topography. Earth and Planetary Science Letters, 146(3–4):367–377, 1997.
ISSN 0012-821X. URL
http://222.195.83.195/wen/Reprints/WenAnderson97EPSL.pdf. Accessed
4th March, 2020. 153
H.-G. Wenzel. Ultra high degree geopotential model GPM3E97A to degree
and order 1800 tailored to Europe. In Vermeer and Ádám (1998), pages
71–80. 79
P. Wessel, W. Smith, R. Scharroo, J. Luis, and F. Wobbe. Generic Mapping
Tools: Improved Version Released. EOS Trans. AGU, 94(45):409–410, 2013.
URL http://dx.doi.org/10.1002/2013EO450001. ii
K. Wieczerkowski, J. X. Mitrovica, and D. Wolf. A revised relaxation-time
spectrum for Fennoscandia. Geophysical Journal International, 139:69–86,
1999. URL https://doi.org/10.1046/j.1365-246X.1999.00924.x. 338
Wikipedia, The Aerospace Corporation. URL
https://en.wikipedia.org/wiki/The_Aerospace_Corporation. Accessed
23rd April, 2019. 218
Wikipedia, Declination. URL https://en.wikipedia.org/wiki/Declination.
Accessed 29th February, 2020. 375
Wikipedia, Dislocation. URL https://en.wikipedia.org/wiki/Dislocation.
Accessed 23rd April, 2019. 301
Wikipedia, Double-slit experiment. URL
https://en.wikipedia.org/wiki/Double-slit_experiment. Accessed 18th
February, 2020. 306
Wikipedia, Earth normal modes. URL
https://en.wikipedia.org/wiki/Seismic_wave#Normal_modes. Accessed
16th March, 2020. 310
Wikipedia, Hour angle. URL https://en.wikipedia.org/wiki/Hour_angle.
Accessed 29th February, 2020. 376

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
463
Wikipedia, John Pratt. URL
https://en.wikipedia.org/wiki/John_Pratt_(Archdeacon_of_Calcutta).
Accessed 23rd April, 2019. 142
Wikipedia, Mu-metal. URL https://en.wikipedia.org/wiki/Mu-metal.
Accessed 23rd April, 2019. 308
Wikipedia, Pendulum clock. URL
https://en.wikipedia.org/wiki/Pendulum_clock. Accessed 23rd April,
2019. 293
Wikipedia, Saros. URL https://en.wikipedia.org/wiki/Saros_(astronomy).
Accessed 23rd April, 2019. 323
Wikipedia, Sea level rise. URL https://en.wikipedia.org/wiki/Sea_level_rise.
Accessed 23rd April, 2019. 333
Wikipedia, Seasat conspiracy theory. URL
https://en.wikipedia.org/wiki/Seasat#Conspiracy_theory. Accessed 23rd
April, 2019. 341
Wikipedia, Strengthening mechanisms of materials. URL https:
//en.wikipedia.org/wiki/Strengthening_mechanisms_of_materials.
Accessed 10th May, 2019. 301
Wikipedia, Sverdrup. URL https://en.wikipedia.org/wiki/Sverdrup.
Accessed 23rd April, 2019. 331
Wikipedia, Zero-length springs. URL
https://en.wikipedia.org/wiki/Spring_(device)#Zero-length_springs.
Accessed 23rd April, 2019. 298
Wolfram Demonstrations, Difference formula for cosine. URL
https://demonstrations.wolfram.com/DifferenceFormulaForCosine/.
Accessed 7th April, 2019. 195
∑︁
Wolfram Functions, ∞ k=1
cos kx
k . URL https:
//functions.wolfram.com/ElementaryFunctions/Cos/23/02/0001/.
Accessed 25th February, 2020. 196
Wolfram MathWorld, Spherical Harmonic Addition Theorem. URL https:
//mathworld.wolfram.com/SphericalHarmonicAdditionTheorem.html.
Accessed 11th May, 2019. 375

í ¤. û
464 ABCDEFGHIKLMNOPRSTVWY Bibliography

L. Wong and R. Gore. Accuracy of geoid heights from modified Stokes kernels.
Geophysical Journal of the Royal Astronomical Society, 18(1):81–91, 1969. URL
https://doi.org/10.1111/j.1365-246X.1969.tb00264.x. 218
G. Wöppelmann, C. Letetrel, A. Santamaría-Gómez, M.-N. Bouin,
X. Collilieux, Z. Altamimi, S. D. Williams, and B. Martín-Míguez. Rates of
sea-level change over the past century in a geocentric reference frame.
Geophysical Research Letters, 36(12), 2009. URL
https://doi.org/10.1029/2009GL038720. 325

Y
YouTube, Hammer vs. Feather, 2010. URL https://www.youtube.com/watc
h?feature=player_embedded&v=KDp1tiUsZw8#! Accessed 7th April, 2019.
4
D.-N. Yuan, W. L. Sjogren, A. S. Konopliv, and A. B. Kucinskas. Gravity field
of Mars: A 75th degree and order model. Journal of Geophysical Research, 106
(E10):23 377–23 401, 2001. URL https://doi.org/10.1029/2000JE001302. 280

í ¤. û
 Index

ABCDEFGHIJKLMNOPQRSTUVWZ

A Amazonas (Brazil), 244, 369

a priori information, 354 American Geophysical Union (AGU), 390
a priori variance, 353 amplifier, in fibreoptic cable, 186
Aalto University, 392 Amsterdam (The Netherlands), 163, 323
Abell 1689, 3 analysis, a posteriori, 363
acceleration angular distance, geocentric
geometric, 367 definition, 255
measured by GNSS, 313, 314 covariance function, 256, 261
measured by gravimeter, 313 degree constituent equation, 68
of aircraft, 313 forward geodetic problem, 255
of free fall, 302 generating function, 199
satellite, 367 Helmert condensation, 431
accelerometer, 315, 367, 370 Stokes kernel, 190, 233
on GRACE, 368 tangent plane, 228
action at a distance, 1 anisotropy, 199
adjustment, least-squares, 352 anomalous quantity, 85, 109, 111
Agulhas Stream, 321 Antarctic Ocean, 364
air pressure, variations, 311 Antarctica, 128, 244, 314, 333, 337
airborne gravimetry, 244, 366 antimatter, 26
description, 312 anti-root, under sea, 147
flight height, 314 Apollo project, 4, 302
homogeneity, 315 Arabelos, Dimitris, 242
Airy, George Biddell, 143 Archimedes’ law, 145
Airy–Heiskanen hypothesis, 145, 151 Arctic, 244, 314
Airy–Heiskanen model, 144 Arctic Ocean, 364
Åland, 363 ice cover, 364
Alaska (USA), 148 ice volume, 364, 365
algebraic-sign domain, 83 argument of perigee, 356
algebraic-sign interval, 57, 58 arrest (gravimeter), 302
altimetric satellite, 342, 356 ascending node, 355, 356

– 465 –
466 ABCDEFGHIJKLMNOPQRSTUVWZ Index

of the Moon, 378 in a function space, 422

Asperger syndrome, 4 in a vector space, 411, 417
associated Legendre function, 57 basis vector, 411–413
fully normalised, 66 bathymetry, 131, 138
symmetry, 62, 82 Bergensbanen (Norway), 147
associated Legendre functions Bessel’s equation, 422
figure, 58 BGI, 124, 133, 151, 389
table, 58 bias, of measurement, 345
astatisation, 301 bilinear form, 411, 417
astatisation ratio, 299, 300 Bjerhammar sphere, 280
asthenosphere, 328, 333 Bjerhammar, Arne, 280
Atlantic Ocean blue film, of the Earth, 367
North, 332 Blue Road Geotraverse project, 328
salinity, 329 bordering, 239
atmosphere Bose–Einstein condensate, 185, 306
attraction, 310 Bothnian Bay (Finland, Sweden), 329
surface mass density, 311 Bouguer anomaly, 131, 157
total mass, 312 bias, 132, 133
atmospheric drag, 367, 368, 370 calculation, 180
compensation, 315 calculation steps, 135, 136
atmospheric loading, 386, 387 example, 137
atmospheric refraction, 161 interpolation, 132
atomic clock, 185 prediction, 132
attenuation factor, 80 properties, 132
attenuation with height, 46, 54 simple, 132
attraction, 5 Southern Finland, 133
exterior, 19 spherical, 138
interior, 19 bias, 138
of a spherical shell, 8, 9 terrain corrected, 134
autocovariance, 252 Bouguer hypothesis, of land uplift,
average density, of the Earth’s crust, 131 326–328
averaging over ocean surface, 336 Bouguer plate
axial vector, 330 as approximation, 131, 174
azimuth, 199, 227, 228, 255, 256 attraction, 129, 130, 132, 439
definition, 255 double, 177
half, 137
B of air, 310
Baltic Sea Bouguer reduction, 129, 209, 216
airborne gravimetry, 315 simple, 133
mean sea surface, 341 Bouguer shell, attraction, 138
salinity, 329 Bouguer, Pierre, 128, 141
sea-surface topography, 323, 329 Boulder, University of Colorado at, USA,
barometric heighting, 115 302, 344
Barzaghi, Riccardo, 241 boundary condition, 44, 45, 119
base network measurement, 307 periodicity, 420
basis boundary surface, choice, 127
complete, 413, 422, 423

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
467
boundary-value problem, 34, 35, 42, 127 clock, 184
definition, 34 pendulum, 293
free, 115 closing error, 301
of Dirichlet, 34, 35, 119, 202 coastline, 386
of Neumann, 71, 72, 119 coastline mask, 364
of physical geodesy, 119 coefficient vector, 417
spectral solution, 120 co-geoid, 128
third, 116, 119 of isostatic reduction, 152, 156
bounded support, 229 coherence, of matter waves, 185
box, rectangular, 24, 25 collocation, least-squares (LSC), 150, 242,
Brovelli, Maria, 241 261, 263, 267
Bruns equation, 113, 126, 174, 190, 385 description, 256, 264
Bruns vertical-gradient equation, 92 FFT, 280
Bruns, Ernst Heinrich, 92, 113 flexibility, 268
bulldozer, 210 Columbus geoid (model), 143, 392
Bureau Gravimétrique International, see BGI commutative diagram, 47, 70, 216, 232
comparison point, for geoid
C determination, 324
cage, in absolute gravimeter, 302 compensation depth, 151, 153
Calgary (Canada), 240 component, of a vector, 398
calibration Congo (Africa), 369
gravimeter, 310 conservative field, 161, 182
in-flight, 346, 362 definition, 5, 403
radar altimeter, 346, 362 curl, 407
calibration certificate, 310 potential, 403, 407
cannon, 83 continental ice sheet, 148, 149, 325, 333
carbon dioxide, 330 continental ice sheets, total mass, 334
Cavendish, Henry, 4, 356 continental shelf, 147
celestial mechanics, 13 continuity equation, 409
celestial sphere, see unit sphere convection, in the Earth’s mantle, 153
central force field, 396 convolution, 15, 47, 237, 283
centrifugal acceleration, 88 calculation, 237
centrifugal force, 87, 107 linear combination, 235
divergence, 317 notation, 229
centrifugal potential, 86, 87 sea-level equation, 335
expression, 88 terrain correction, 246
CHAMP (satellite), 74, 244, 367 two-dimensional, 229
characteristic equation, 419 convolution theorem, 229, 230, 239
Chasles theorem, 2, 33, 35 co-ordinate conversion, 51
Chasles, Michel, 33 co-ordinate reference system
checkerboard, 59 co-rotating, 87, 317
chlorophyll, 330 inertial, 87
circular disk, attraction, 130 co-ordinate time, 184
climate research, 334 co-ordinate transformation, 393
climate, of Earth, 332 co-ordinates
clinometer, 383 cylindrical, 42
clinometer, long water-tube, 382

í ¤. û
468 ABCDEFGHIJKLMNOPQRSTUVWZ Index

ellipsoidal, 52 crossover point, 348, 352, 353

geodetic, 50 crustal density, 144, 245
definition, 52 Cryosat-2 (satellite), 343
geographical, 50, 234 curl (operator), 402, 403
natural, 94, 95 interpretation, 402
polar, 406 linearity, 400
rectangular, 5, 50 of gradient, 402
spherical, 29, 42, 50, 55 of wind field, 402
definition, 51 curvature
topocentric, 316 of a level surface, 90, 91, 316
toroidal, 42 of field line, 93, 110, 169
Copenhagen (Denmark), 240, 391 of plumb line, 92, 93
coral, 334 of the Earth, 135, 145, 229
Coriolis acceleration, 330 cyclone, tropical, 321
direction, 330
Coriolis effect, 329 D
Coriolis force, 88, 322, 330 damping, of gravimeter, 302, 313, 314
Coriolis, Gaspard-Gustave, 88 Danish straits, 329
corkscrew rule, 395, 398 Darwin, Sir George, 379
correlation, 259 datum, 352
correlation length, 261, 262, 291 datum defect, 352
correlation, quasi-geoid & topography, datum point, 163
172, 173 datum transformation, 352, 354
correspondence, integral & spectral Dead Sea (Levant), 154
equations, 191 declination, of the Moon, 375
cosine rule, 233, 375 Defense Mapping Agency, US (DMA), 73
half-angle, 233 deformation
cosine taper, 240 of the Earth, 380
covariance function, 254, 267 viscous, 338
definition, 255 deglaciation, last, 149, 325, 339
empirical, 279 degree constituent
Gauss–Markov, 267 of powers of height, 436, 437
global, 281 of the disturbing potential, 105
isotropic, 271 of the gravity anomaly, 117
of gravity anomalies, 278 degree constituent equation, 68, 69, 120,
of Hirvonen, 261–264 189
of the disturbing potential, 270, 289 data point, 68
in space, 275, 277 evaluation point, 68
on the Earth’s surface, 277 harmonic field, 201
spectral representation, 271 degree number, of tidal force, 382
cross covariance, 252 degree of freedom, 354, 355
cross product, see vectorial product degree variance
crossover adjustment, 353, 366, 373 notation, 274
global, 355 of gravity anomalies, 278, 280
crossover condition, 355 of the disturbing potential, 271, 274
crossover difference, 354, 366, 373 on the Earth’s surface, 277
degree variance formula, 280

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
469
degree, harmonic, 54, 57, 61, 218 spherical-harmonic expansion, 104
definition, 53 surface harmonics, 54
Delft (The Netherlands), 240 divergence (operator), 16, 22, 401, 404, 408
delta function, Dirac’s, 26, 204, 283, 338 interpretation, 401
Denker, Heiner, 241 linearity, 400
density Doodson, Arthur Thomas, 379
ice, 149, 335, 364 Doodson’s constant, 379, 385
mantle, 144 DORIS (positioning instrument), 343
rock, 95, 178 dot product, see scalar product
sea water, 144, 147, 335, 364 downwards continuation, harmonic, 174,
standard crustal, 178 209, 211, 212, 244, 433
topography, 174, 212 existence, 202, 210
upper mantle, 149 of r∆g, 203
density model, 35 drift (gravimeter), 301, 309, 310
density profile, 35
density, SI unit, 11 E
developing country, education, 390 Earth
dice throw, 252 centre of mass, 50, 67, 73, 105, 117
difference, geoid – free-air geoid, 175 flattening, 50, 51, 74, 99
difference, height anomaly – free-air gravitational field, 41, 367
geoid, 174 spectral representation, 2
difference, height anomaly – geoid height, gravity field, 74
174 radius, 18
difference, orthometric height – normal rotation rate, 88, 99, 383
height, 176 earthquake, 310, 339
difference, quasi-geoid – geoid, 175, 180 eccentricity, orbital, 356
differential operator, 404 eddiness, in a vector field, 402
digital terrain model (DTM), 134, 210, 364 eddy, 343
dipole, 20, 67, 68 eddy phenomenon, 88
dipole field, 105 eddy-free flow, 409
dipole moment, 67 EGM96 (geopotential model), 73
definition, 20 coefficients, mean errors, 75
of the Earth, 67 EGM2008 (geopotential model), 61, 74,
vanishing, 76 124, 133
dipole surface density, 20 Eiffel Tower, 92
dipole-density layer, 19, 438 72 names, 16, 17, 33, 45, 55, 88, 422
dipole-layer element, 20 eigenvalue problem, 416–418
Dirac, Paul, 26 eigenvectors of a symmetric matrix, 418
directional sphere, see unit sphere eight-unit cube, 27
Dirichlet, Peter Gustav Lejeune, 34 Einstein summation convention, 259
Dirichlet’s problem, 2 Einstein, Albert, 4
dislocation (crystal), 300 El Niño Southern Oscillation (ENSO), 321
disturbing potential, 109, 125 elasticity, 92, 300
definition, 104 of the Earth, 380, 381, 386
at terrain level, 213 of the Earth’s crust, 387
local, 230 elasticity model, 384
electric charge, conservation, 397

í ¤. û
470 ABCDEFGHIJKLMNOPQRSTUVWZ Index

electric currents, in the Earth’s core, 153 European climate, 332

electrostatic compensation, 313 European Geosciences Union (EGU), 390
electrostatic potential, 404 European Space Agency (ESA), 342, 343,
ellipsoid of revolution, 50 369
ellipsoidal harmonic, 95 eustatic rise, of mean sea level, 325, 326,
ellipsoidal-harmonic expansion, 77 335
definition, 74 evaluation functional, 250
centrifugal potential, 95 evaluation latitude, 234
computation, 79 Everest, Mount, 147
convergence, 78 expectancy
normal potential, 78 of a stochastic process, 254
RMS Empress of Ireland, 163 statistical, 253
energy exterior product, see vectorial product
conservation, 397
of place, 163 F
Envisat (satellite), 342 factorial, 55
eötvös (unit), 118, 317 Falkland Islands, 370
Eötvös tensor, 316 Faller, James E., 302
Eötvös, Loránd, 89 Faraday, Michael, 23, 408
epoch Fast Collocation, 150, 284
of land uplift, 324 Fast Fourier Transform (FFT), 15, 150, 232
of measurement, 305 algorithms, 429
equations of motion, of satellites, 67 and convolution, 230
equatorial radius, see semimajor axis, and tapering, 241
Earth ellipsoid collocation, 281, 284
equilibrium length, of spring, 296, 297 mixed-radix, 427
equipotential surface, 19 radix 2, 427
as boundary, 33 terrain correction, 244, 247
figure, 113 Fastest Fourier Transform in the West
equivalence principle, 4, 89 (fftw, software), 429
ergodicity, 254 Father Point / Pointe-au-Père (Rimouski,
error ellipse, 419 Quebec, Canada), 163
ERS-1 (satellite), 342 Fennoscandia, 149, 324, 337
ERS-2 (satellite), 342 Fennoscandian Shield, 151
escape velocity, 83 fibreoptic cable, 186
estimation, 257 field equations
estimator, 258 of electromagnetism, 17
mean error, 264 of gravitation, 1, 17
optimal, 259 field line, 23, 401, 408
Eterna (software), 387 field theory
Ethiopia, 244 of electromagnetism, 42
Euclidean space, 5, 10, 411 of gravitation, 1
Euler angle, 402 field, the concept, 41
Euler notation, 92 figure of the Earth, 85, 89, 128, 143, 357
Euler’s identity, 429 FIN2000 (geoid model), 242, 243, 324
Eurajoki (Finland), 308 precision, 242
FIN2005N00 (geoid model), 242, 391

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
471
precision, 242 freeboard, 343, 364, 372
finite element method (FEM), 150 French Academy of Sciences, 128
Finland, 162, 177, 242, 302 frequency domain, 47, 231, 232
Finnish climate, 332 fresh water, 321, 329
Finnish Geodetic Institute (FGI), 302, 309, fulcrum, of a pendulum, 295
382, 391 function space, 250, 251, 412, 413, 420, 422
Finnish Geospatial Research Institute on the sphere, 422
(FGI), 392 function theory, 412
first eccentricity, 51 functional, 274
flat Earth model, 145 definition, 250
flattening, of a planet, 13 linear, 34, 251, 274, 279
flow velocity, 22, 401, 409 definition, 250
vector field, 331 of the disturbing potential, 251
fluid motion in gravity field, 163 of the disturbing potential, 250
flux, 23 fundamental equation of physical
fluxion, 326 geodesy, 116, 119, 213
footprint, radar altimeter, 345, 363
footscrews, 298 G
Forsberg, René, 240, 242 Galilei, Galileo, 4
Fourier basis function, 45, 56, 63, 413 gauge invariance, 17
Fourier coefficient, 45, 53 Gauss integral theorem, 2, 28, 30, 33, 408,
Fourier series, 413, 429 441, 442
Fourier sine expansion, 45 figure, 23, 408
Fourier theory, 194 presentation, 22
Fourier transform, 47 book-keeping, 23
artefacts, 240 in terms of potential, 24
discrete, 230, 232, 427, 429 Gauss, Carl Friedrich, 22
periodicity, 239, 240 Gelderen, Martin van, 240
reverse, 428 general relativity, 17, 184
forward, 230 generating function
of ℓ−3 , 246 geometry, 200
notation, 229 of the Legendre polynomials, 153,
reverse, 230, 247 200
Fourier, Joseph, 45 geodetic forward problem on the sphere,
France (funding), 389 255
free oscillations, of the solid Earth, 310, Geodetic Reference System 1967 (GRS67),
391 100
periods, 310 geographic mean, 254–256, 268
free-air anomaly definition, 253
definition, 122 geographic variance, 253
calculation, 123, 180 geoid, 118, 153, 166, 321, 322
linearisation, 122 definition, 89, 165
Southern Finland, 124 classical, 322
use, 123 fake, 178
free-air hypothesis, of land uplift, 327 free-air, 174
free-air reduction, 135 true, 366

í ¤. û
472 ABCDEFGHIJKLMNOPQRSTUVWZ Index

geoid determination, 34, 72, 216, 365, 389, geostrophic equilibrium, 330
390 Germain curvature, 92
1D-FFT, 239 Germain, Marie-Sophie, 92
classical, 128 German Research Centre for Geosciences
FFT, 240, 242 (GFZ), 367
gravimetric, 191, 222, 329 Gibbs phenomenon, 219, 240, 416
principle, 190 Gibbs, Josiah Willard, 219
computational framework, 193 glacial isostatic adjustment (GIA), 325,
NKG, 391 337
precise, 178 glacial maximum, last, 149
software, 242 glacier, 154, 325, 335
spherical FFT retreat, 148
multi-band, 234 global warming, 333
Taylor expansion, 235 GM⊕ , best value, 6
geoid height, 109, 119, 174, 180, 343, 347 GNSS
definition, 110 height of gravimetric stations, 128
from satellite altimetry, 244 in airborne gravimetry, 313
reduced, 216 in aircraft, 313
geoid map, 370 in altimetric satellite, 366
geoid model, 152 in height determination, 241
computation, 241 measuring atmospheric, ocean tidal
global loading, 387
high resolution, 366 orbital tracking, 370
precise, 367, 371 positioning of tide gauges, 325, 329
gravimetric, 324 GNSS levelling, 324
local, 323 GOCE, 74, 244, 315, 331, 332, 371
of Finland, 112 description, 369
geoid rise, 325, 326 name, 372
geoid undulation, 110 GPS
global, 110 on satellite, 342, 367
in Finland, 110 orbital tracking, 368
geological map, 179 reference system, 99
geometric geodesy, 103 GRACE (satellite pair), 74, 244
geophysical data record (GDR), 346, 363 description, 367, 369
geophysical reduction, 128 GRACE follow-on mission, 369
geopotential, 95 grade measurement, 128
gradient, 315 gradient
level surface, 89 of Earth attraction, 370
on the tangent plane, 91 of gravity disturbance, 289
spectral expansion, 70 of potential, normal direction, 34
geopotential image, sharpness, 61 gradient (operator), 9, 12, 400, 402
geopotential model, global, 367 interpretation, 400
geopotential number, 163, 165 linearity, 400
definition, 162 of scalar field, 399
GEOS-3 (satellite), 341 gravimeter
Geosat (satellite), 341 absolute, 302, 328

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
473
principle of operation, 303 measurement, 118
airborne, 302, 313 on the rotating Earth, 88
astatised, 297, 299 prediction, 215
atomic, 306 total, 313
principle of operation, 307 gravity acceleration, 89
ballistic, 295, 302 from pendulum, 295
FG5, 302, 304 on the plumb line, 95
IMGC-02, 306 gravity anomaly, 115, 270
JILA, 302 a priori estimate, 266
LaCoste-Romberg, 297, 299, 303 definition, 119
pendulum, 294 as a boundary condition, 119
quantum, see gravimeter, atomic as a functional, 251
registering, 387 at sea level, 203, 208
relative, 328 at topography level, 208
sea, 302, 366 atmospheric reduction, 311
spring, 294, 295, 298, 300, 307 availability, 249
superconducting, 308–310 block average, 222, 224, 268
gravimetry, 73 calculation, 115, 122
airborne, see airborne gravimetry change, 326
satellite, 315 estimate, 265, 266
study of Earth interior, 36 from satellite altimetry, 244
gravitation, 3 global average, 253
gravitational acceleration, 4, 11, 13, 34, 89 horizontal gradient, 224
measurements, 72 in the external space, 202
gravitational acceleration vector, 9–11, 16 mean error, 265
gravitational constant, universal, 2, 335, observations, 264
356, 375, 396 reduced, 216
gravitational field, 5, 22 surface harmonics, 54
of celestial objects, 384 gravity disturbance, 289
gravitational force, 41 definition, 114
gravitational gradiometer (GOCE), 366 observing, 115
description, 370, 371 spectral representation, 114
gravitational lens, 3 gravity field, 2, 85, 165
gravitational potential, 1 change, 337
in a vacuum, 404 determination, 280, 368, 371
rotationally symmetric, 96 exterior, 123, 213
gravitational vector, 11 fine structure, 366
gravitational wave, 17 GOCE resolution, 315
gravity, 129, 295, 313, 328 observation density, 249
definition, 302 oceanic, 341
absolute measurement, Finland, 308 of mountains, 141
along levelling line, 184 research in Europe, 390
equatorial, 100 research in Finland, 391
in airborne gravimetry, 313 research in HUT, 392
in the tropics, 294 research internationally, 389
local, 167, 293, 313 residual, 215

í ¤. û
474 ABCDEFGHIJKLMNOPQRSTUVWZ Index

spatial variability, 229 grid representation, 230

statistical behaviour, 261 GRS80
temporal change, 367 definition, 99
textbooks, 392 GM⊕ , 6
very local features, 370 spherical-harmonic coefficients, 102
gravity formula, 85, 98, 100, 123 Gulf of Finland, 329
gravity gradient, 118, 317, 382 airborne gravimetry, 315
free air, 387 Gulf Stream, 321, 330–332
gravity mapping survey, 307 Guyana, French, 293
gravity potential GWR iGrav (gravimeter), 310
as sum of gravitational and GWR20 (gravimeter), 309
centrifugal potentials, 88
gradient, 407 H
in spherical harmonics, 104 Haagmans, Roger, 240
of falling atoms, 306 Hannover (Germany), 241
gravity vector, 165 Hardanger plateau (Norway), 147
gravity vector field, 407 harmonic continuation
gravity versus gravitation, 87 of gravity anomalies, 204
gravity-gradient field of Sun and Moon, of the potential, 274
see tidal field harmonic field
gravity-gradient tensor, 316, 317 definition, 16
GRAVSOFT (software), 242 r∆g, 202
Green equivalent-layer theorem, 33 harmonic oscillator, 44, 446
Green, George, 28 hat notation, of estimator, 267
Greenland, 128, 244, 333, 337 Hayford ellipsoid, 100, 123
Greenland Aerogeophysics Project (GAP), Hayford, John Fillmore, 142
314 heat transport, 329, 332
Green’s first theorem, 28 height
Green’s function, 335, 338 above mean sea level, 165, 166
of sea level, 336 above the geoid, 165
of the geopotential, 336 above the reference ellipsoid, 51, 111,
of vertical displacement, 336 168, 343
partial, of deformation, 336 height anomaly, 121, 122, 168, 171, 173
Green’s second theorem, 29 definition, 167
Green’s theorems, 2 height system, 391
Green’s third theorem, 29, 127 national, 323, 324, 352
exterior point, 30 Heiskanen, Veikko Aleksanteri, 143, 392
external space, 32 helicopter, 314
interior point, 31 helium, liquid, 309, 310
Greenwich meridian, 50 Helmert condensation, 2, 140, 141, 209,
grid 216, 431, 437, 438
of gravity anomalies, 230 gravity effect, 437
of the disturbing potential, 231 Helmert ellipsoid, 100
of the Stokes kernel, 231 Helmert height, 174, 324
grid integration, 150 definition, 177
grid matrix calculation, 237 as approximation, 177
Helmert, Friedrich Robert, 140

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
475
Helsinki (Finland), 324 inertial tensor, 68
Helsinki astronomical observatory, 163, instantaneous length, of spring, 299
164 Institut für Erdmessung (Hannover,
Helsinki harbour, 162, 323, 352 Germany), 241, 390
Helsinki University of Technology (HUT, integrability, 422
TKK), 392 integral equation, 2
Himalayas, 141 integration by parts, 421
Hirvonen, Reino Antero, 261 intercomparison, of absolute gravimeters,
Hirvonen’s geoid model, 391 306, 308
Hofmann-Wellenhof, Bernhard, 392 International Association of Geodesy
homogeneity (IAG), 242, 389
of data precision, 284 International Center for Earth Tides
of gravimetric data, 315 (ICET), 390
homogeneity assumption, 253, 254 International Center for Global Earth
horizon plane, 330 Models (ICGEM), 390
hour angle, of the Moon, 375 International Digital Elevation Model
Hubble Space Telescope, 3 Service (IDEMS), 390
Huygens, Christiaan, 293 International Geodynamics and Earth
HY-2A (satellite), 343 Tide Service (IGETS), 309
hydrodynamics, 384 International Geoid Commission (IGeC),
hydrosphere, 367 390
International Geoid Service (IGeS), 390
I International Gravimetric Bureau, see BGI
ice age, last, 327, 333, 337 International Gravity Field Service
ice cap, 325 (IGFS), 389, 390
ice load, 327, 337 International Isostatic Institute, 392
history, 337 International Service for the Geoid (ISG),
ice sheet, 335, 339 389
Laurentide, 333 International Union of Geodesy and
ice, multi-year, 364 Geophysics (IUGG), 389
identity matrix, 283 interpolation from grid, 231, 232
ill-posed problem, 210 invariant, 420
inclination, orbital, 356–358, 360 inversion calculation, 232, 238
of the Moon, 378 inverted barometer, 321
incompressibility, 22, 409 isostasy, 1, 142, 143
independence, statistical, 264 modern understanding, 148
India, 343 isostasy hypothesis, 142
indirect effect, 128 isostatic anomaly
of Bouguer reduction, 152, 209 definition, 150
of Helmert condensation, 209, 437, interpolation, 150
439 prediction, 150
constant terrain, 438 Southern Finland, 151
of isostatic reduction, 152, 156, 209 isostatic compensation, 144, 145
of residual terrain modelling (RTM), definition, 142
211 percentage, 152
of the tidal potential, 381, 384 isostatic equilibrium, 294
Indonesia, 363, 369

í ¤. û
476 ABCDEFGHIJKLMNOPQRSTUVWZ Index

isostatic geoid, 152 coefficients, 220

isostatic hypothesis, 142, 144, 150 degree, 218
isostatic reduction, 149, 150, 216 sharp cut-off, 218
description, 153 soft, 219
and density layers, 154 Wong–Gore, 218, 220
residual field, 150 Kevo (Finland), 308
isotropic density distribution, 9 Kirkkonummi (Finland), 309, 323
isotropic process, 261 KKJ, 123
isotropy, 199 Knudsen, Per, 242
and spectral representation, 199 Kolkata (India), 142
of the disturbing potential, 271 Krasovsky ellipsoid, 100
of the viscosity, 336 Kronecker delta, 283
isotropy assumption, 256 Kronstadt (Russia), 163
Italy (funding), 390 Kuusamo (Finland), 308
iteration
calculation of normal height, 179 L
calculation of orthometric height, LaCoste, Lucien, 299
166, 177 LAGEOS (satellite), 356
land uplift
J absolute, 325, 326
J2 (dynamic flattening), 74, 99, 357, 358 Fennoscandian, 339
Jacobi, Carl Gustav Jacob, 65 post-glacial, 323, 325, 391
Jacobi’s determinant, 129 mechanism, 327
definition, 406 relative, 325, 326
map projection co-ordinates, 229 Laplace equation, 16, 41, 317
polar co-ordinates, 226, 406 definition, 41
spherical co-ordinates, 65, 199, 227, basis solutions, 53
256, 406, 423 in ellipsoidal co-ordinates, 74
Jason (satellites), 343, 359 in polar co-ordinates, 48
Java Sea (Dutch Indies, Indonesia), 294 in rectangular co-ordinates, 43
Jerry (GRACE satellite), 368 in spherical co-ordinates, 52, 441
Joensuu (Finland), 308 linearity, 42, 43
solution, 41
K transformation, 42
Kääriäinen, Jussi, 315 Laplace operator (∆), 404
Kaivopuisto (Helsinki, Finland), 163, 164, definition, 16, 41
323 linearity, 250
Kater, Henry, 294 Laplace, Pierre-Simon, 16, 404
Kater’s reversion pendulum, 294 Lapland (Northern Europe), 129
Kaula, William, 280 laser interferometer, 303
Kepler, Johannes, 223, 355 latitude
Kepler’s laws, 355 astronomical, 95, 109, 123
Kepler’s orbital elements, 356 geocentric, 50, 52, 97
Kepler’s second law, 396, 397 geodetic, 98, 106, 109, 234
Kepler’s third law, 355, 360, 374 geographic, 98
kernel modification reduced, 52, 98, 106
advanced, 220 types, 98

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
477
least-squares method, 305, 325 geodetic, 109
ordinary, 349 of the Moon, 378
Legendre function, 53, 55, 446 lookup table, 429
fully normalised, 424 Love number, 381, 382, 384, 386
of the second kind, 76, 78 determination, 382
Legendre polynomial, 55 by GNSS, 382
fully normalised, 65 elastic, 338, 381
Legendre polynomials viscous, 338
as a basis, 422 Love, Augustus, 381
figure, 56 LSC, see collocation, least-squares
orthogonality on the interval lunar laser ranging (LLR), 302
[−1, +1], 64
orthogonality on the unit sphere, 65 M
orthonormality on the unit sphere, magnetic field, 308
66 of the Earth, 153, 308
table, 56 Mäkinen, Jaakko, 167
Legendre, Adrien-Marie, 55 Maldives (Indian Ocean), 334
Legendre’s equation, 422, 446 map projection co-ordinates, 227, 228
Lego™ brick, 25 map projection plane, 216
Leibniz University (Hannover, Germany), mareograph, see tide gauge
390 Mars (planet), gravity field, 280
Leibniz, Gottfried Wilhelm, 1 mass density, 16
level surface, 90, 165, 322, 371 mass distribution inside the Earth, 35
definition, 90 mass point, underground, 238, 291
level, bull’s-eye, 298 mass surface density, 18, 140
levelling, 95, 165 mass-density layer, 1, 209
principle, 161, 162 double, 19, 20, 32, 439
geostrophic, 329 single, 18, 32, 140, 153, 191
new technologies, 391 matter density, 11, 12, 409
relativistic, 185 matter, conservation, 22
steric, 329 Mauna Kea, 159
levelling instrument, 161 Maupertuis, Pierre de, 128
levelling line, 184 Maxwell, James Clerk, 17
levelling staff, 161 mean error of unit weight, 305, 353
levelling, of gravimeter, 298 mean geoid, 384
lever beam, 297, 302 mean sea level, 118, 165, 322
lever motion, 334 definition, 321, 322
linear partial differential equations, global, 323, 325
theory of, 42 location, 117
linear regression, 345 mean sea surface, 357
Liouville, Joseph, 422 measurement axis, 302
localised kernel, 204, 213 measuring telescope, 161
location vector of a mass element, 67 mechanics, of the solid Earth, 383
Lohja (Finland), 382 Meissner effect, 308
longitude Mekong (river), 369
astronomical, 95, 109 Melchior, Paul, 379
meridian convergence, 233

í ¤. û
478 ABCDEFGHIJKLMNOPQRSTUVWZ Index

meridian ellipse, 98 nabla (∇, operator), 10, 399

mesoscale eddy, 321, 332 National Geospatial-Intelligence Agency,
metal fatigue, 301 US (NGA), 73, 389
metallurgy, 301 National Imagery and Mapping Agency,
Metsähovi research station (Finland), 163, US (NIMA), 73
308, 309, 323 National Land Survey of Finland, 309, 392
microgal (µGal), 117 National Map Grid Co-ordinate System,
microseismicity, effect on gravimeter, 300, see KKJ
303 NAVD88, North American Vertical
microwave link (GRACE), 368 Datum 1988, 163
mid-Holocene highstand, 334 Navy, United States, 341
Milan (Italy), 240, 390 network densification, stepwise, 307
milligal (mGal), 117 network hierarchy, 307
Min, Erik de, 240 Neumann, Carl Gottfried, 71
minimum energy state, 163 Newton, Isaac, 2, 13
mirror antisymmetry, 57 Newton’s law of gravitation, 1, 245
mirror symmetry, 57, 63 definition, 2
missile, submarine-launched ballistic, 341 for a spherical Earth, 316
mixed covariance, 279 Newton’s law of motion, 4
modal strength, 338 Newton’s theory of gravitation, 1, 17
Mohorovičić, Andrija, 156 Niethammer, Theodor, 178
Mohorovičić discontinuity (“Moho”), 156, Niethammer’s method, 178
247 NKG Working Group for Geoid and
Molodensky theory, 33, 121 Height Systems, 391
Molodensky, Mikhail Sergeevich, 33, 121, NKG2004 (geoid model), 391
127, 166–168, 392 NKG2015 (geoid model), 391
Molodensky’s method, 208 NN (height system), 163
height anomaly, 214 noise
linearisation, 212 definition, 258
linearisation error, 214 in altimetry observations, 347
Molodensky’s realisation, 168, 171, 172 observational, 258
momentum noise variance matrix, 258, 268
angular, 395 Nordiska Kommissionen för Geodesi (NKG),
conservation, 397 391
per unit of mass, 397 norm, of a vector, 5, 394, 412
linear, 395 Normaal Amsterdams Peil (NAP), 163
Mongolia, 244 normal component, of a vector field, 405
monopole, 67, 68 normal correction (NC), 183
monsoon, 369 equation, 183
Moritz, Helmut, 392 for staff interval, 183
mosaic, 59 normal derivative, 19
mu-metal, 308 existence, 31
Munk, Walter, 343 of potential, 19
on surface of sphere, 26
N normal direction, 19
N60 (height system), 162, 242, 323 existence, 31
N2000 (height system), 163, 242, 323

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
479
normal equation, 305 GOCE mission, 332
normal gravitational potential, 86 inversion problem, 371
normal gravity measurements, 329
definition, 85 permanent, 322
at sea level, 123 tilt, 330
calculation, 115 unit, 331
GRS80, 100 variation, 332
linearity along the plumb line, 168, ocean loading, 386
171, 179 oceanography, 384
on the equator, 97 octave (programming language), 63, 361
on the poles, 97 Ohio (USA), 143, 262
on the reference ellipsoid, 97, 99 Ohio State University, 73, 392
normal gravity field, 86 one-Earth problem, 253
and reference ellipsoid, 110 operator
choice, 179 linear, 34, 416, 417
ellipsoidal, 98 self-adjoint, 417, 418, 421
GRS80, 312 spectral representation, 275
normal gravity vector, 110, 113, 169 optical lattice clock, 185
normal height, 166, 171, 173 optimality, least-squares, 260
calculation, 179, 180, 183 orbit
precise calculation, 179 35-day, 361
normal matrix, 349 no-shadow, 358, 359
normal plumb line, 110, 111, 169 retrograde, 358, 360
normal potential, 85, 86, 96 Sun-stationary, 358, 359, 370
definition, 85 Sun-synchronous, 358, 374
global average, 117 three-day, 362
gradient, 85 orbit correction, 344, 347, 351
GRS80, 99 bias, 353, 354
in a co-rotating system, 101 trend, 352, 354
in spherical harmonics, 104 orbit error, 346
on the reference ellipsoid, 97, 99, 169 correction, 365
over the equator, 100, 101 orbit prediction, 352
normal vector, of a surface, 405, 408 order, harmonic, 53
Norwegian Sea, 147 orientation, of a surface, 408
Nottingham (Great Britain), 28 orthogonal basis, 412, 423
Nouvel, Henri SJ, 163 complete, 418
Legendre polynomials, 422
O orthogonality
obliquity, of the Earth’s rotation axis, 388 of degree constituents, 273
observation equation, 268 of functions, 422
adjustment constraint, 349 orthometric correction (OC), 181, 183
of ballistic gravimetry, 304, 305 equation, 182
of crossover adjustment, 347, 351, for staff interval, 183
373 orthometric height, 110, 165, 174
of levelling, 349 definition, 95, 165
of satellite altimetry, 347 calculation, 166, 180, 181
ocean current, 331, 332

í ¤. û
480 ABCDEFGHIJKLMNOPQRSTUVWZ Index

precise calculation, 176, 178 definition, 85

orthonormal basis, 10, 67, 87, 90, 412, 413, bending, 143
418 plumb-line deflection, 109, 110, 125, 141,
definition, 5, 398 198
orthonormality, 194 definition, 86, 109
oscillation equation, 296, 300, 420, 422 and the geoid, 110
as an eigenvalue problem, 420 as a functional, 251
Ostrobothnia (Finland), 324 at sea, 341
OSU model, 73 in Finland, 110
local contribution, 224
P observed, 112
paleo-research, 148 plumb-line direction, 109, 114, 313
parallax, 375 absolute, 383
parallelogram, 395 measurement, 94
path integral, 9, 407 Päijänne, Lake (Finland), 167
Peltier effect, 334 point mass, underground, 33
Peltier, W. Richard, 334 point variance, 420
pendulum clock, 293 Poisson equation, 26, 178
pendulum equation, 293 definition, 17
peneplain, 147 data point, 201
period evaluation point, 201
fortnightly tide, 377 for computing a harmonic field from
orbital, 355, 356 surface values, 201
oscillation, 296, 300 for r∆g, 203
pendulum, 293 gravity potential, 89
theoretical tide, 379, 382 spectral form, 203
Peru (South America), 128 Poisson kernel for gravity anomalies,
petroleum extraction industry, 242 203–205
phase angle Poisson, Siméon Denis, 17
of an atom, 306 polar hole, 359, 360
of matter waves, 306 polar motion, Earth, 383
phase angle, tidal, see tide polar radius, see semimajor axis, Earth
phase, of water, 333 ellipsoid
physical geodesy poles of ignorance, 370
geometry and physics, 50 polynomial fit, of geoid surface, 324
potential convention, 10 potential
textbook, 392 definition, 5
physical modelling, 329 at a terrain point, 162
physical theory, nature of, 1 exterior, 21, 35, 43, 117, 127, 152
phytoplankton, 330 of the topography, 432
Pizzetti, Paolo, 98 interior, 21
planet as a mass point, 13 of the topography, 433
plasma, 185 of a mass line, 14
plasticity, 300 of a point mass, 10, 16, 35, 66
plate tectonics, 148 of a pointlike body, 6
Plesetsk Cosmodrome (Russia), 367, 369 of a set of mass points, 11
plumb line, 90

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
481
of a solid body, 11 quantisation, 420
of a spherical shell, 6, 7, 9 quantum mechanics, 42
of an extended body, 6 quantum number, 421
of the topography, 15 quantum state, 185, 306
origin of word, 28 quantum theory, 306
uniqueness, 404 quasi-geoid, 121, 122, 166, 172, 173
potential difference, 17, 95, 162 quasi-geoid height, 168, 173
potential energy, 10 quasi-geoid model, 121, 152
potential field, 2, 41 computation, 241
local behaviour, 42
of a dipole, 67 R
of a mass-density layer, 1 radar, microwave, 341, 366
vertical shift, 47, 71 radio energy, 346
potential theory, 2 radius of curvature
Potsdam system, 295, 307 transversal, 51
PRARE (positioning instrument), 342 Raman effect, 306
Pratt, John Henry, 142 Rapp, Richard H., 73, 280
Pratt–Hayford hypothesis, 142, 144 recursion
precession, orbital, 357 calculation of normal height, 171
precise levelling, 162, 323, 329 calculation of orthometric height,
prediction, 239, 257 177
homogeneous, 267, 268 computation of Fourier basis
Prey reduction, 177 functions, 57
Prey, Adalbert, 177 computation of Legendre
principal axes, of error ellipse, 419 polynomials, 63
Principia (book), 2 recursive algorithm, 55
prism method of terrain correction, 135 reduction to sea level, 208
propagation delay redundancy, 354
ionospheric, 346 reference benchmark, 164
tropospheric N60, 323
dry, 346 N2000, 323
wet, 346 reference ellipsoid, 51, 118
propagation of covariances, 277, 279 as a level surface, 85, 98
propagation of variances, 259, 269 reference latitude, 233, 235, 236
propellant (GOCE), 315, 369 reference radius, 105
propeller aircraft, 314 reference-surface thinking, 122
proper time, 184 regularisation, 210, 247
pseudo-force relativity theory, 1
aircraft motions, 313 relaxation time, 338
Earth’s rotation, 86 remote sensing of sea ice, 364
moving on a rotating Earth, 88 “remove” step, 210, 229
Pythagoras theorem, 5 remove–restore method, 128, 215, 216,
439
Q renormalisation, 247
quadrature, 222 repeat track, 365
quadrupole, 68 research school, international, 242, 390
quadrupole moment, of the Earth, 68

í ¤. û
482 ABCDEFGHIJKLMNOPQRSTUVWZ Index

residual terrain modelling (RTM), 210, satellite geodesy, 280

212 satellite laser ranging (SLR), 356
resolution, of measurement, 249 satellite orbit, 355
resonance, 359 mean height, 355, 357
rest length, of spring, 296, 297 perturbation, 73, 249
“restore” step, 209, 211, 230, 439 repeat period, 357
rheology, 336 Sun-stationary, 358
Richer, Jean, 293 satellite-to-satellite tracking (SST), 369
rising dough model, 327 Saturn (planet), ring, 293
Robin, Victor Gustave, 119 scalar field, definition, 397
rocket engine (GOCE), 369 scalar product, 221
Romberg, Arnold, 299 definition, 393
Romberg, Werner, 223 commutativity, 411
root, of mountain, 141, 142, 150 in a function space, 420
density, 143, 144 in a vector space, 5, 411, 417
depth, 146 linearity, 411
rotational potential, 86 of force and path, 394
rotational symmetry, 82, 236, 281, 282, 443 of Legendre polynomials, 64, 422
Royal Society of Edinburgh, 17 of operator and vector, 401
Royal Society of London, 2, 4, 17, 120 of two functions, 412
of vector fields, 393
S of vectors, 394
Sacks, Oliver, 4 on the sphere, 423
salinity (sea water), 322, 330 Schrödinger equation, 42
salinity gradient (Baltic Sea), 329 Schrödinger, Erwin, 42
sampling, spatial, 249 Schrödinger’s cat, 42
San Francisco (USA), 390 Schwarz, Klaus Peter, 240
Sandwell, David, 244 Schwarzschild metric, 184
Sansò, Fernando, 240 Schwarzschild, Karl, 184
SARAL/AltiKa (satellite), 343 sea ice, 364
saros (lunar motion periodicity), 323 sea-floor pressure, 367
satellite altimetry, 73, 329, 341 sea-level equation, 334, 335
concepts, 345 angular distance, geocentric, 336
geoid, 365 convolution, 335, 336
in archipelagos, 363 data point, 336
location of sea surface, 371 evaluation point, 336
measured range, 343 sea-level oscillation, 329
measurement method, 343 sea-level rise, 325
orbital geometry, 353 global, 343
results, 363 Holocene, 337
retracking, 363, 364 Seasat (satellite), 341
return pulse sea-surface residual variation, 347
analysis, 363 sea-surface topography, 322, 330, 331, 343,
half-height point, 363 347, 366
time measurement, 363 definition, 323
travel time, 364 change, 326
sea-surface variability, 365

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
483
determination, 329 space geodesy, 74
global, 329 spatial frequency, 231, 246
map, 332 spatial wavelength, 231
mapping, 342 spectral coefficients, 53
sea-surface variability, 357, 365 spectral constituent function, 69
secular effect, in sea level, 322 spherical Bouguer reduction, 139
seismicity, 339 mass effect, 139
seismology, 35, 142 spherical cap, 216
seismometer, 310 spherical harmonic
self-adjoint differential equation, 420, 422 algebraic sign, 60
semimajor axis, Earth ellipsoid, 52, 99, sectorial, 59, 60
106 semi-wavelength, 61, 79
semimajor axis, orbital, 355, 356 tesseral, 59, 60
semiminor axis, Earth ellipsoid, 52, 99, wavelength, 59
106 zonal, 59, 60
sensitivity, instrument, 296, 298 spherical shell, 7, 9
Sentinel-3A (satellite), 343 spherical symmetry, 338, 381
separation of variables, 43, 48, 53, 444, 446 of mass distribution, 35
shoebox world, 44, 45, 245 of the Earth, 253, 256
Sideris, Michael, 240 spherical-cap integration, 150
sight axis, of a level, 161 spherical-harmonic coefficient, 274, 357
signal covariance matrix, 258 as a linear functional, 251
signal variance, 261 fully normalised, 73, 424
signal variance matrix, 257 spherical-harmonic expansion, 2, 62, 229,
gravity anomalies, 263 230
signal, definition, 258 coefficients, 53
significant wave height (SWH), 345, 346 degree-one, 67
Simpson integration, 223 first terms, 103
Simpson, Thomas, 223 global, 73
Simpson’s rule, 223 high-degree, 79
singularity, of normal matrix, 349, 352 model, 73, 217
sink (vector field), 22, 401 of the normal gravitational potential,
Skylab (space station), 341 102
slowing-down ratio, of time, 184 of the topography, 62
snow clearing, 310 resolution, 61
Sodankylä (Finland), 308 rotational symmetry, 63, 64
solar panel, 358 spheroid
solar time, 359 Bruns, 103
Solheim, Dag, 242 Helmert, 103
solid body, 11 spirit level, 161
solid spherical harmonic, 53 Spitsbergen (island), 148
Somigliana, Carlo, 98 spline, 240
Somigliana–Pizzetti equation, 98 spring balance, linear, 296
source (vector field), 10, 22, 401 spring constant, 296
source function, 22 spring lengthening, 296
space domain, 47, 231 stabilised platform, 302, 313

í ¤. û
484 ABCDEFGHIJKLMNOPQRSTUVWZ Index

stability (gravimeter), 309 surface element, oriented, 405

staff-reading difference, 181 surface normal
stationarity, of a stochastic process, 256 Earth, 119
stationary field, 5 ellipsoidal, 114
steel manufacture, 301 exterior, 85
stereographic map projection, 228 exterior, 22
Sterneck, Robert von, 294 surface spherical harmonic, 53, 59, 446
stochastic process as a map, 60
definition, 251 fully normalised, 423
ergodic, 254 plotting, 63, 64
of location, 252 surface spherical harmonics
on the Earth’s surface, 253 linear combination of, 217
stochastic quantity, 251 orthogonality, 217
realisation, 252 orthonormality, 273
Stokes curl theorem, 406 sverdrup (unit), 331
Stokes equation, 120, 191, 216, 227, 326 swinging time (pendulum), 293
at sea level, 213 symbolic algebra software, 206
data point, 191 Synthetic Aperture Radar Altimeter
differentiation, 198 (SRAL), 343
evaluation point, 191, 227
exterior space, 191 T
geoid height, 190 tangent plane, 227, 228
in collocation, 267 to a level surface, 90
in plane co-ordinates, 229 tapering, 240
inner zone, 224 tapering function, 240
integration, 191 Taylor-series expansion, 14, 15, 21
spectral form, 189 Helmert condensation, 432, 434
Stokes equation and harmonicity, 209 tea break, 310
Stokes kernel, 120, 190, 192, 235 telluroid, 121, 172
closed expression, 192 definition, 121
modified, 218, 219 telluroid mapping, 121, 173
on the Earth’s surface, 191 temperature (sea water), 322
re-written, 234 terrain correction (T C), 132, 134, 137, 158,
smoothness, 221 178, 244, 245
spectral form, 191 bias, 134
Taylor-series expansion, 235 calculation, 134, 135
two-dimensional, 196 data point, 245
Stokes, George Gabriel, 34, 120 equation, 134
Strang van Hees, Govert, 234, 240 evaluation point, 245, 247
Sturm, Jacques, 422 example, 136
Sturm-Liouville problem, 422 in spherical geometry, 139
submarine measurements, 294 in the exterior space, 247
subsidence, land, 333 prism method, 134, 245
superconduction, 308, 309 values, 135
superspring, 303 terrain effect, local, 315
support latitude, 234 thermal expansion
of a pendulum, 293

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
485
of sea water, 325 of a body, 13
thermostat (gravimeter), 302 of a column of air and water, 368
thruster, on-board, 357, 362 of a column of matter, 144
tidal decomposition equation of Laplace, of the Earth, 73, 99, 105, 117
377 Toulouse (France), 389
tidal field, 317 trace, of a matrix, 420
tidal force, 380, 383 transformation surface, geoidal, 324, 391
tidal loading, ocean, 387 trench, ocean, 294
tidal potential, 375, 382 triangle inequality, 12, 13
tidal reduction, permanent deformation, Trieste (Italy), 163
383 true anomaly, 356
tide, 13, 321, 343, 347 Tscherning, Carl Christian, 242, 280
amplitude, 382 Tscherning-Rapp formula, 280
diurnal, 359, 378, 379 Tukey taper, 240
fortnightly, 378 turbulence, in a vector field, 402
frequency, 382 Turku (Finland), harbour, 352
ocean, 346, 386 Tuvalu (Pacific Ocean), 334
period, 382 twiddle factor (FFT), 429
permanent part, 383–385, 388 Tytyri limestone mine (Lohja, Finland),
deformation, 383 382
effect on height difference, 386
effect on the geoid, 385 U
phase angle, 382 uncertainty
semidiurnal, 359, 378, 379 a priori, 352
solid-Earth, 346, 391 inside the Earth, 249
theoretical, 376, 380, 383 of co-ordinate measurement, 314
tide gauge, 322, 325, 335 of estimate, 256, 266
tide-free Earth crust, 386 of observation, 249, 264, 268
tide-free geoid, 384, 386 of vertical acceleration, 314
Tikhonov regularisation, 353, 354 uniform convergence, 415, 432
Tikhonov, Andrey Nikolayevich, 353 unit matrix, 283
Toeplitz circulant matrix, 283 unit sphere, 65, 199, 208, 272, 423
Toeplitz, Otto, 283 unknown (adjustment parameter), 305,
Tom (GRACE satellite), 368 354
tomography, seismic, 151 upper culmination, of the Moon, 376
TOPEX/Poseidon (satellite), 323, 342, 344,
359 V
topographic-potential integral Vaasa (Finland), 308
data point, 431 variance, 252
evaluation point, 431 variance function, of a stochastic process,
topography, 131, 138 252
topography shift to inside geoid, 209 variance matrix of location, 419
Torge, Wolfgang, 392 variance of prediction, 264, 266, 269, 287
torque, 297 definition, 259
torsion balance, 4, 317 minimisation, 260
total mass vector, 393
vector field

í ¤. û
486 ABCDEFGHIJKLMNOPQRSTUVWZ Index

definition, 397 Z
differentiable, 22 zenith angle, of the Moon, 375, 376
vector space, 413 zenith tube, 383
abstract, 411, 412 zero geoid, 384
n-dimensional, 416 zero potential
vector sum, 89 at infinity, 18
vectorial product, 395, 396 at mean sea surface, 18
of operator and vector, 401 zero topography compensation level, 146
Vening Meinesz equations, 199, 224 zero-length spring, 298–301
Vening Meinesz, Felix Andries, 150, 199 how to build, 298
submarine measurements, 294
Verbaandert–Melchior pendulum, 383
vertical displacement, of test mass, 300
vertical gravity gradient, 100, 123, 305
anomalous, 208, 213, 232
at sea level, 208
kernel, 205
free-air, 177, 178, 316
inside-rock, 177
on the Earth’s surface, 118
vertical normal gravity gradient, 122, 136,
174
vertical reference system, see height
system
viscosity, 336
viscous relaxation mode, 338
Von Sterneck device, 294

W
water flowing upwards, 167
water vapour radiometer, 346
wave equation
of matter, 42
relativistic, for the electron, 26
weighing visitors, 310
weightlessness, 315
Wenzel, Hans-Georg, 387
wind field, 402
wind pile-up, 321
wire pendulum, very long, 295
wire-frame model, 352
work integral, 394, 405, 407
world aether, 1
World Geodetic System 1984 (WGS84), 68,
99

í ¤. û