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 524 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 . . . . . . . . . 57
Õ ! 4. The normal gravity field . . . . . . . . . . . . . . . . . . . . . 87
Õ ! 5. Anomalous quantities of the gravity field . . . . . . . . . . . 111
Õ ! 6. Geophysical reductions . . . . . . . . . . . . . . . . . . . . . 129
Õ ! 7. Vertical reference systems . . . . . . . . . . . . . . . . . . . . 163
Õ ! 8. The Stokes equation and other integral equations . . . . . . 191
Õ ! 9. Spectral techniques, FFT . . . . . . . . . . . . . . . . . . . . . 231
Õ ! 10. Statistical methods . . . . . . . . . . . . . . . . . . . . . . . . 255
Õ ! 11. Gravimetric measurement devices . . . . . . . . . . . . . . . 299
Õ ! 12. The geoid, mean sea level, and sea-surface topography . . . 329
Õ ! 13. Satellite altimetry and satellite gravity missions . . . . . . . 351
Õ ! 14. Tides, the atmosphere, and Earth crustal movements . . . . 385
Õ ! 15. Earth gravity field research . . . . . . . . . . . . . . . . . . . 401

Õ ! A. Field theory and vector calculus — core knowledge . . . . . 405

Õ ! B. Function spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 423
Õ ! C. Why does FFT work? . . . . . . . . . . . . . . . . . . . . . . . 439
Õ ! D. Helmert condensation . . . . . . . . . . . . . . . . . . . . . . 443

– iii –
iv Contents

Õ ! E. The Laplace equation in spherical co-ordinates . . . . . . . . 453

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 . . . . . . . . . . . . . 3
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 The equations of Laplace and Poisson . . . . . . . . . . 16
1.9 Gauge invariance . . . . . . . . . . . . . . . . . . . . . . 17
1.10 Single mass-density layer . . . . . . . . . . . . . . . . . 18
1.11 Double mass-density layer . . . . . . . . . . . . . . . . . 20
1.12 The Gauss divergence theorem . . . . . . . . . . . . . . 22
1.13 Green’s theorems . . . . . . . . . . . . . . . . . . . . . . 28
1.14 The Chasles theorem . . . . . . . . . . . . . . . . . . . . 33
1.15 Boundary-value problems . . . . . . . . . . . . . . . . . 34
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Exercise 1–1: Core of the Earth . . . . . . . . . . . . . . . . . . 37
Exercise 1–2: Atmosphere . . . . . . . . . . . . . . . . . . . . . 38
Exercise 1–3: The Gauss divergence theorem . . . . . . . . . . 39

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 . . . . . 53
2.6 Dependence on height . . . . . . . . . . . . . . . . . . . 54
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3. Legendre functions and spherical harmonics 57

3.1 Legendre functions . . . . . . . . . . . . . . . . . . . . . 57
3.2 Symmetry properties of spherical harmonics . . . . . . 64
3.3 Orthogonality of Legendre functions . . . . . . . . . . . 67
3.4 Low-degree spherical harmonics . . . . . . . . . . . . . 69
3.5 Splitting a function into degree constituents . . . . . . 71
3.6 Spectral representations of various quantities . . . . . . 73
3.7 Often-used spherical-harmonic expansions . . . . . . . 76
3.8 Ellipsoidal harmonics . . . . . . . . . . . . . . . . . . . 77
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Exercise 3–1: Attenuation with height of a spherical-harmonic
expansion . . . . . . . . . . . . . . . . . . . . . . 83
Exercise 3–2: Symmetries of spherical harmonics . . . . . . . 84
Exercise 3–3: Algebraic-sign domains of spherical harmonics 85
Exercise 3–4: Escape velocity . . . . . . . . . . . . . . . . . . . 86

4. The normal gravity field 87

4.1 The basic idea of a normal field . . . . . . . . . . . . . . 87
4.2 The centrifugal force and its potential . . . . . . . . . . 89
4.3 Level surfaces and plumb lines . . . . . . . . . . . . . . 92
4.4 Natural co-ordinates . . . . . . . . . . . . . . . . . . . . 96
4.5 The normal potential in ellipsoidal co-ordinates . . . . 97
4.6 Normal gravity on the reference ellipsoid . . . . . . . . 99
4.7 Numerical values and calculation formulas . . . . . . . 100
4.8 The normal potential as a spherical-harmonic expansion 103
4.9 The disturbing potential . . . . . . . . . . . . . . . . . . 105

í ¤. û
vi Contents

Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 107

Exercise 4–1: The Somigliana–Pizzetti equation . . . . . . . . 108
Exercise 4–2: Centrifugal force . . . . . . . . . . . . . . . . . . 109

5. Anomalous quantities of the gravity field 111

5.1 Disturbing potential, geoid height, deflections of the
plumb line . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.2 Gravity disturbances . . . . . . . . . . . . . . . . . . . . 115
5.3 Gravity anomalies . . . . . . . . . . . . . . . . . . . . . 117
5.4 Units used for gravity anomalies . . . . . . . . . . . . . 120
5.5 The boundary-value problem of physical geodesy . . . 120
5.6 The telluroid mapping and the “quasi-geoid” . . . . . 122
5.7 Free-air anomalies . . . . . . . . . . . . . . . . . . . . . 124
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Exercise 5–1: The spectrum of gravity anomalies . . . . . . . 127
Exercise 5–2: Deflections of the plumb line and geoid tilt . . . 127
Exercise 5–3: Gravity anomaly, geoid height . . . . . . . . . . 127

6. Geophysical reductions 129

6.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.2 Bouguer anomalies . . . . . . . . . . . . . . . . . . . . . 130
6.3 Terrain effect and terrain correction . . . . . . . . . . . 134
6.4 Spherical Bouguer anomalies . . . . . . . . . . . . . . . 140
6.5 Helmert condensation . . . . . . . . . . . . . . . . . . . 142
6.6 Isostasy . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
6.7 Isostatic reductions . . . . . . . . . . . . . . . . . . . . . 152
6.8 The “isostatic geoid” . . . . . . . . . . . . . . . . . . . . 153
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 160
Exercise 6–1: Gravity anomaly . . . . . . . . . . . . . . . . . . 161
Exercise 6–2: Bouguer reduction . . . . . . . . . . . . . . . . . 161
Exercise 6–3: Terrain correction and Bouguer reduction . . . 161
Exercise 6–4: Isostasy . . . . . . . . . . . . . . . . . . . . . . . 162

í ¤. û
Contents
vii
7. Vertical reference systems 163
7.1 Levelling, orthometric heights and the geoid . . . . . . 163
7.2 Orthometric heights . . . . . . . . . . . . . . . . . . . . 165
7.3 Normal heights . . . . . . . . . . . . . . . . . . . . . . . 168
7.4 Difference between geoid height and height anomaly . 175
7.5 Difference between orthometric and normal heights . . 178
7.6 Calculating orthometric heights precisely . . . . . . . . 178
7.7 Calculating normal heights precisely . . . . . . . . . . . 181
7.8 Calculation example for heights . . . . . . . . . . . . . 181
7.9 Orthometric and normal corrections . . . . . . . . . . . 183
7.10 A vision for the future: relativistic levelling . . . . . . . 186
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 188
Exercise 7–1: Calculating orthometric heights . . . . . . . . . 188
Exercise 7–2: Calculating normal heights . . . . . . . . . . . . 188
Exercise 7–3: Difference between orthometric and normal
height . . . . . . . . . . . . . . . . . . . . . . . . 189

8. The Stokes equation and other integral equations 191

8.1 The Stokes equation and the Stokes integral kernel . . . 191
8.2 Example: The Stokes equation in polar co-ordinates . . 195
8.3 Plumb-line deflections and Vening Meinesz equations 201
8.4 The Poisson integral equation . . . . . . . . . . . . . . . 201
8.5 Gravity anomalies in the exterior space . . . . . . . . . 204
8.6 The vertical gradient of the gravity anomaly . . . . . . 207
8.7 Gravity reductions in geoid determination . . . . . . . 212
8.8 The remove–restore method . . . . . . . . . . . . . . . . 218
8.9 Kernel modification . . . . . . . . . . . . . . . . . . . . 219
8.10 Advanced kernel modifications . . . . . . . . . . . . . . 223
8.11 Block integration . . . . . . . . . . . . . . . . . . . . . . 225
8.12 Effect of the local zone . . . . . . . . . . . . . . . . . . . 227
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 228
Exercise 8–1: The Stokes equation in the near zone . . . . . . 229

í ¤. û
viii Contents

9. Spectral techniques, FFT 231

9.1 The Stokes equation as a convolution . . . . . . . . . . 231
9.2 Integration by FFT . . . . . . . . . . . . . . . . . . . . . 234
9.3 Solution in latitude and longitude . . . . . . . . . . . . 236
9.4 Bordering and tapering of the data area . . . . . . . . . 243
9.5 Computing a geoid model with FFT . . . . . . . . . . . 246
9.6 Use of FFT in other contexts . . . . . . . . . . . . . . . . 249
9.7 Computing terrain corrections with FFT . . . . . . . . . 249
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 253

10. Statistical methods 255

10.1 The role of uncertainty in geophysics . . . . . . . . . . 255
10.2 Linear functionals . . . . . . . . . . . . . . . . . . . . . 256
10.3 Statistics on the Earth’s surface . . . . . . . . . . . . . . 257
10.4 The covariance function of the gravity field . . . . . . . 260
10.5 Least-squares collocation . . . . . . . . . . . . . . . . . 263
10.6 Prediction of gravity anomalies . . . . . . . . . . . . . . 276
10.7 Covariance function and degree variances . . . . . . . 278
10.8 Propagation of covariances between various quantities 281
10.9 Global covariance functions . . . . . . . . . . . . . . . . 287
10.10 Collocation and the spectral viewpoint . . . . . . . . . 288
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 291
Exercise 10–1: Variance of prediction . . . . . . . . . . . . . . 292
Exercise 10–2: Hirvonen’s covariance equation and prediction 293
Exercise 10–3: Predicting gravity anomalies . . . . . . . . . . 293
Exercise 10–4: Predicting gravity anomalies (2) . . . . . . . . 294
Exercise 10–5: Propagation of covariances . . . . . . . . . . . 295
Exercise 10–6: Kaula’s rule for gravity gradients . . . . . . . . 295
Exercise 10–7: Underground mass points . . . . . . . . . . . . 297

11. Gravimetric measurement devices 299

11.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299

í ¤. û
Contents
ix
11.2 The relative or spring gravimeter . . . . . . . . . . . . . 301
11.3 The absolute or ballistic gravimeter . . . . . . . . . . . 308
11.4 Network hierarchy in gravimetry . . . . . . . . . . . . . 313
11.5 The superconducting gravimeter . . . . . . . . . . . . . 314
11.6 Gravity measurement and the atmosphere . . . . . . . 317
11.7 Airborne gravimetry and GNSS . . . . . . . . . . . . . 319
11.8 Measuring the gravity gradient . . . . . . . . . . . . . . 322
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 324
Exercise 11–1: Absolute gravimeter . . . . . . . . . . . . . . . 325
Exercise 11–2: Spring gravimeter . . . . . . . . . . . . . . . . . 326
Exercise 11–3: Air pressure and gravity . . . . . . . . . . . . . 326

12. The geoid, mean sea level, and sea-surface topography 329
12.1 Basic concepts . . . . . . . . . . . . . . . . . . . . . . . . 329
12.2 Geoid models and national height datums . . . . . . . 331
12.3 The geoid and post-glacial land uplift . . . . . . . . . . 333
12.4 Determining the sea-surface topography . . . . . . . . 336
12.5 Global sea-surface topography and heat transport . . . 337
12.6 The global behaviour of the sea level . . . . . . . . . . . 340
12.7 The sea-level equation . . . . . . . . . . . . . . . . . . . 342
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 347
Exercise 12–1: Coriolis force, ocean current . . . . . . . . . . . 348
Exercise 12–2: Land subsidence and the mechanism of land
uplift . . . . . . . . . . . . . . . . . . . . . . . . 349

13. Satellite altimetry and satellite gravity missions 351

13.1 Satellite altimetry . . . . . . . . . . . . . . . . . . . . . . 351
13.2 Crossover adjustment . . . . . . . . . . . . . . . . . . . 356
13.3 Choice of satellite orbit . . . . . . . . . . . . . . . . . . . 365
13.4 In-flight calibration . . . . . . . . . . . . . . . . . . . . . 372
13.5 Retracking . . . . . . . . . . . . . . . . . . . . . . . . . . 373
13.6 Oceanographic research using satellite altimetry . . . . 375
13.7 Satellite gravity missions . . . . . . . . . . . . . . . . . 376

í ¤. û
x Contents

Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 382

Exercise 13–1: Altimetry, crossover adjustment . . . . . . . . 383
Exercise 13–2: Satellite orbit . . . . . . . . . . . . . . . . . . . 384
Exercise 13–3: Kepler’s third law . . . . . . . . . . . . . . . . . 384

14. Tides, the atmosphere, and Earth crustal movements 385

14.1 The theoretical tide . . . . . . . . . . . . . . . . . . . . . 385
14.2 Deformation caused by the tidal potential . . . . . . . . 391
14.3 The permanent part of the tide . . . . . . . . . . . . . . 394
14.4 Tidal corrections between height systems . . . . . . . . 395
14.5 Loading of the Earth’s crust by sea and atmosphere . . 397
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 398
Exercise 14–1: The permanent tide . . . . . . . . . . . . . . . . 399

15. Earth gravity field research 401

15.1 Internationally . . . . . . . . . . . . . . . . . . . . . . . 401
15.2 Europe . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
15.3 The Nordic countries . . . . . . . . . . . . . . . . . . . . 403
15.4 Finland . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
15.5 Textbooks . . . . . . . . . . . . . . . . . . . . . . . . . . 404

A. Field theory and vector calculus — core knowledge 405

A.1 Vector calculus . . . . . . . . . . . . . . . . . . . . . . . 405
A.2 Scalar and vector fields . . . . . . . . . . . . . . . . . . . 409
A.3 Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
A.4 The continuity of matter . . . . . . . . . . . . . . . . . . 421

B. Function spaces 423

B.1 An abstract vector space . . . . . . . . . . . . . . . . . . 423
B.2 The Fourier function space . . . . . . . . . . . . . . . . 424
B.3 Sturm–Liouville differential equations . . . . . . . . . . 428
B.4 Legendre polynomials . . . . . . . . . . . . . . . . . . . 434
B.5 Spherical harmonics . . . . . . . . . . . . . . . . . . . . 434

í ¤. û
List of Tables
xi
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . 437
Exercise B–1: Orthonormality of the Fourier basis functions . 437

C. Why does FFT work? 439

D. Helmert condensation 443

D.1 The exterior potential of the topography . . . . . . . . . 444
D.2 The interior potential of the topography . . . . . . . . . 445
D.3 The exterior potential of the condensation layer . . . . 447
D.4 Total potential of Helmert condensation . . . . . . . . . 447
D.5 The dipole method . . . . . . . . . . . . . . . . . . . . . 450

E. The Laplace equation in spherical co-ordinates 453

E.1 Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . 453
E.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 456

Bibliography 459

Index 479

List of Tables
3.1 Legendre polynomials . . . . . . . . . . . . . . . . . . . . 58
3.2 Associated Legendre functions . . . . . . . . . . . . . . . 60
3.3 Semi-wavelengths for different degrees and orders . . . . 64
3.4 Plotting a surface spherical-harmonic map . . . . . . . . . 66
3.5 EGM96 coefficients and mean errors . . . . . . . . . . . . 78
3.6 Legendre functions of the second kind . . . . . . . . . . . 79

4.1 GRS80 normal potential spherical-harmonic coefficients . 104

5.1 Orders of magnitude of gravity variations . . . . . . . . . 120

8.1 Stokes equation in two dimensions, octave code . . . . . . 199

í ¤. û
xii List of Figures

8.2 Derivation of the kernel for the vertical gradient of the

gravity anomaly . . . . . . . . . . . . . . . . . . . . . . . . 209

13.1 Altimetric satellites through the ages . . . . . . . . . . . . 352

13.2 Calculating the height of a satellite from its period . . . . 372

14.1 The various periods in the theoretical tide . . . . . . . . . 390

List of Figures
1.1 Gravitation is universal . . . . . . . . . . . . . . . . . . . . 2
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 divergence 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 a field with height . . . . . . . . . . . . . . 47

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

3.1 Some Legendre polynomials . . . . . . . . . . . . . . . . . 58

3.2 Some associated Legendre functions . . . . . . . . . . . . 60
3.3 The algebraic signs of spherical harmonics . . . . . . . . . 62
3.4 Surface spherical harmonics as maps . . . . . . . . . . . . 63
3.5 Monopole, dipole, and quadrupole . . . . . . . . . . . . . 71
3.6 Radially shifting the field, commutative diagram . . . . . 74

í ¤. û
List of Figures
xiii
4.1 The normal gravity field of the Earth . . . . . . . . . . . . 88
4.2 Gravitation and centrifugal force . . . . . . . . . . . . . . 89
4.3 The curvature of level surfaces . . . . . . . . . . . . . . . 93
4.4 The curvature of the plumb line . . . . . . . . . . . . . . . 94
4.5 Natural co-ordinates . . . . . . . . . . . . . . . . . . . . . 96
4.6 Meridian ellipse and latitude types . . . . . . . . . . . . . 100
4.7 The normal field’s potential over the equator . . . . . . . 103

5.1 Geoid undulations and deflections of the plumb line . . . 112

5.2 A Finnish geoid model from 1984 . . . . . . . . . . . . . . 113
5.3 Equipotential surfaces of gravity and normal gravity fields 115
5.4 Various reference surfaces in gravity measurement . . . . 117
5.5 Free-air gravity anomalies for Southern Finland . . . . . 126

6.1 The attraction of a Bouguer plate . . . . . . . . . . . . . . 131

6.2 The Bouguer plate as an approximation to the topography 133
6.3 Behaviour of different anomaly types in the mountains . 135
6.4 Terrain-corrected Bouguer anomalies for Southern Finland 136
6.5 Calculating the terrain correction using the prism method 137
6.6 The steps in computing the Bouguer anomaly . . . . . . . 138
6.7 A special terrain shape . . . . . . . . . . . . . . . . . . . . 139
6.8 Helmert condensation and the gravity field . . . . . . . . 143
6.9 Friedrich Robert Helmert . . . . . . . . . . . . . . . . . . . 144
6.10 Isostasy and the deflection of plumb lines . . . . . . . . . 145
6.11 Pratt–Hayford isostatic hypothesis . . . . . . . . . . . . . 146
6.12 Airy–Heiskanen isostatic hypothesis . . . . . . . . . . . . 147
6.13 Quantities in isostatic compensation . . . . . . . . . . . . 147
6.14 The modern understanding of isostasy . . . . . . . . . . . 151
6.15 Isostatic gravity anomalies for Southern Finland . . . . . 154
6.16 Isostatic reduction as a pair of surface density layers . . . 157
6.17 Terrain shape . . . . . . . . . . . . . . . . . . . . . . . . . . 161

7.1 The principle of levelling . . . . . . . . . . . . . . . . . . . 164

í ¤. û
xiv List of Figures

7.2 Height reference benchmark in Kaivopuisto, Helsinki . . 166

7.3 Levelled heights and geopotential numbers . . . . . . . . 167
7.4 Lake Päijänne: water flows “uphill” . . . . . . . . . . . . 169
7.5 Mikhail Sergeevich Molodensky . . . . . . . . . . . . . . . 170
7.6 A graphic cartoon of the proof of Molodensky’s find . . . 173
7.7 Geoid, quasi-geoid, telluroid and topography . . . . . . . 174
7.8 An optical lattice clock . . . . . . . . . . . . . . . . . . . . 187

8.1 The principle of gravimetric geoid determination . . . . . 192

8.2 The geometry of integrating the Stokes equation . . . . . 193
8.3 The Stokes kernel function . . . . . . . . . . . . . . . . . . 194
8.4 The Stokes kernel function on the circle . . . . . . . . . . 200
8.5 Simulation of gravity anomalies and geoid undulations . 200
8.6 The geometry of the generating function of the Legendre
polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . 202
8.7 The Poisson kernel for gravity anomalies . . . . . . . . . . 207
8.8 Residual terrain modelling (RTM) . . . . . . . . . . . . . . 217
8.9 The remove–restore method as a commutative diagram . 219
8.10 Modified Stokes kernel functions . . . . . . . . . . . . . . 222
8.11 Simpson integration nodal weights in two dimensions . . 226

9.1 Map projection co-ordinates in the local tangent plane . . 232

9.2 Commutative diagram for FFT. . . . . . . . . . . . . . . . 237
9.3 “Tapering” 25 % . . . . . . . . . . . . . . . . . . . . . . . . 245
9.4 Example images for the FFT transform . . . . . . . . . . . 246
9.5 The Finnish FIN2000 geoid . . . . . . . . . . . . . . . . . . 248

10.1 Geocentric angular distance and azimuth . . . . . . . . . 261

10.2 Hirvonen’s covariance function in two dimensions . . . . 270
10.3 Least-squares collocation, example . . . . . . . . . . . . . 270
10.4 Least-squares collocation, calculation example . . . . . . 271
10.5 Global covariance functions as degree variances . . . . . 288
10.6 Circular geometry . . . . . . . . . . . . . . . . . . . . . . . 290

í ¤. û
List of Figures
xv
11.1 Jean Richer’s report . . . . . . . . . . . . . . . . . . . . . . 300
11.2 Autograv CG-5 spring gravimeter . . . . . . . . . . . . . . 301
11.3 Operating principle of a spring gravimeter . . . . . . . . 304
11.4 The idea of astatisation . . . . . . . . . . . . . . . . . . . . 307
11.5 Operating principle of a ballistic absolute gravimeter . . 309
11.6 Absolute gravimeter of type FG5 . . . . . . . . . . . . . . 310
11.7 The idea of operation of an atomic gravimeter . . . . . . . 314
11.8 International intercomparison of absolute gravimeters . . 315
11.9 Operating principle of a superconducting gravimeter . . 316

12.1 The mechanism of post-glacial land uplift . . . . . . . . . 335

12.2 The Fennoscandian gravity line on the 63rd parallel north 336
12.3 Connection between sea-surface topography and ocean
currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
12.4 Sea-surface topography map produced by GOCE . . . . . 341
12.5 The sea-level equation . . . . . . . . . . . . . . . . . . . . 343
12.6 Sea-level rise after the last ice age . . . . . . . . . . . . . . 345

13.1 Results from the TOPEX/Poseidon and Jason satellites . 354

13.2 Satellite altimetry as a measurement method . . . . . . . 355
13.3 A simple crossover geometry . . . . . . . . . . . . . . . . 358
13.4 Track geometry of satellite altimetry, example . . . . . . . 363
13.5 Kepler’s orbital elements . . . . . . . . . . . . . . . . . . . 366
13.6 The mechanism of a Sun-stationary orbit . . . . . . . . . . 368
13.7 The geometry of a “no-shadow” orbit . . . . . . . . . . . 369
13.8 A satellite orbit example . . . . . . . . . . . . . . . . . . . 370
13.9 Analysing the altimeter return pulse . . . . . . . . . . . . 373
13.10 Ice volume on the Arctic Ocean . . . . . . . . . . . . . . . 375
13.11 Gravity field determination from GPS orbital tracking . . 378
13.12 The principle of the GRACE satellites . . . . . . . . . . . . 379
13.13 GRACE mission results . . . . . . . . . . . . . . . . . . . . 380
13.14 Gravity field determination with GOCE . . . . . . . . . . 381
13.15 Satellite altimetric track geometry, example . . . . . . . . 383

í ¤. û
xvi List of Figures

14.1 Theoretical tide . . . . . . . . . . . . . . . . . . . . . . . . 386

14.2 The main components of the theoretical tide . . . . . . . . 391
14.3 The constituents of the permanent tide . . . . . . . . . . . 396

A.1 Exterior or vectorial product . . . . . . . . . . . . . . . . . 408

A.2 Kepler’s second law . . . . . . . . . . . . . . . . . . . . . . 409
A.3 The gradient . . . . . . . . . . . . . . . . . . . . . . . . . . 413
A.4 The divergence . . . . . . . . . . . . . . . . . . . . . . . . . 414
A.5 The curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
A.6 The Stokes curl theorem . . . . . . . . . . . . . . . . . . . 419
A.7 The Gauss divergence theorem . . . . . . . . . . . . . . . 421

B.1 Fourier analysis on a step function . . . . . . . . . . . . . 427

E.1 The Gauss divergence theorem applied to a co-ordinate

aligned volume element . . . . . . . . . . . . . . . . . . . 454

í ¤. û
 Acronyms

ABCDEFGHIJKLMNOPRSTUW
A
ACS Advanced Camera for Surveys, instrument on the Hubble Space Telescope
2 ^
AGU American Geophysical Union 402 ^

B
BGI Bureau Gravimétrique International, International Gravity Bureau 126, 136,
154, 401, 402 ^
BVP boundary-value problem 34 ^

C
CHAMP Challenging Minisatellite Payload for Geophysical Research and Ap-
plications 77, 249, 377 ^

D
DMA Defense Mapping Agency (USA) 76 ^
DORIS Doppler Orbitography and Radiopositioning Integrated by Satellite, a
French satellite positioning system 352, 353 ^
DTM digital terrain model 135 ^

E
EGM2008 Earth Gravity Model 2008 63, 77, 120, 126, 136 ^
EGM96 Earth Gravity Model 1996 76–78 ^

– xvii –
xviii ABCDEFGHIJKLMNOPRSTUW Acronyms

EGU European Geosciences Union 402 ^

ENSO El Niño Southern Oscillation 329, 354 ^
ESA European Space Agency 2, 341, 352, 353, 379 ^

F
FAS member of the French Academy of Sciences (Académie des sciences) 433,
434 ^
FFT Fast Fourier Transform 15, 153, 198, 214, 219, 234, 236–239, 241–247, 249,
250, 252, 253, 288, 291, 332, 439, 440 ^
FGI Finnish Geospatial Research Institute, formerly Finnish Geodetic Institute
404 ^
FIN2000 geoid model (Finland) 247, 248, 332 ^
FIN2005N00 geoid model (Finland) 247, 403 ^
FRAS Fellow of the Royal Astronomical Society 300 ^
FRS Fellow of the Royal Society (of London) 4, 17, 226, 299, 300, 390, 391, 433,
434 ^
FRSE Fellow of the Royal Society of Edinburgh 17, 390, 419, 434 ^

G
GDR geophysical data record 356, 373 ^
GFZ Geoforschungszentrum (Potsdam, Germany), German Research Centre for
Geosciences 377 ^
GIA glacial isostatic adjustment 333, 346 ^
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 116, 130, 187, 246, 247,
320, 321, 332, 333, 337, 352, 356, 372, 376, 380, 393, 394, 398, 403 ^
GOCE Geopotential and Steady-state Ocean Circulation Explorer 77, 249, 288,
322, 339, 341, 376, 379–382 ^
GPS Global Positioning System 70, 101, 173, 319, 325, 352, 377, 378 ^
GRACE Gravity Recovery And Climate Experiment 77, 249, 377–380 ^
GRAVSOFT Geopotential determination software, mainly developed in Den-
mark 247 ^
GRS80 Geodetic Reference System 1980 6, 101, 104, 109, 112, 119, 247, 318, 332,
333 ^

í ¤. û
Acronyms ABCDEFGHIJKLMNOPRSTUW
xix
GWR20 superconducting gravimeter built by GWR Instruments 315 ^

H
HUT Helsinki University of Technology, today part of Aalto University 404 ^

I
IAG International Association of Geodesy 247, 315, 401, 402 ^
IB inverted barometer 329 ^
ICET International Center for Earth Tides 402 ^
ICGEM International Center for Global Earth Models 402 ^
IDEMS International Digital Elevation Model Service 402 ^
IGeC International Geoid Commission (obsolete) 402 ^
IGeS International Geoid Service (obsolete) 402 ^
IGETS International Geodynamics and Earth Tide Service 315 ^
IGFS International Gravity Field Service 401, 402 ^
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 312 ^
IOC Initial Operating / Operational Capability 325 ^
ISG International Service for the Geoid 401, 402 ^
IUGG International Union of Geodesy and Geophysics 401 ^

J
J2 The dynamic flattening (“gravitational flattening”) of the Earth 77, 101, 104,
367, 368, 382 ^
Jason American-French-European radar altimetry satellite series, successors
of TOPEX/Poseidon 352–354, 369, 382 ^
JHU Johns Hopkins University 2 ^

K
KKJ National Grid Co-ordinate System (Finland) 124 ^

L
LAGEOS LAser GEOdynamic Satellite 366 ^
Lego™ “Leg Godt”, Engl. “Play Well”, Danish toy brand 26 ^
LLR Lunar laser ranging 308 ^

í ¤. û
xx ABCDEFGHIJKLMNOPRSTUW Acronyms

LSC least-squares collocation 153 ^

M
Mf Moon, fortnightly tide 390 ^

N
N2000 height system (Finland) 165, 168, 247, 331 ^
N60 height system (Finland) 164, 165, 247, 331, 332 ^
NAP Normaal Amsterdams Peil, height system (Western Europe) 165 ^
NASA National Aeronautics and Space Administration (USA) 2 ^
NAVD88 North American Vertical Datum 1988 165 ^
NC normal correction 185 ^
NGA National Geospatial-Intelligence Agency (USA, formerly NIMA) 76,
401 ^
NIMA National Imagery and Mapping Agency (USA, formerly DMA) 76 ^
NKG Nordiska Kommissionen för Geodesi, Nordic Geodetic Commission 403 ^
NKG2004 geoid model (Nordic area) 403 ^
NKG2015 geoid model (Nordic area) 403 ^
NN height system (Finland) 165 ^
NOAA National Oceanic and Atmospheric Administration (USA) 341 ^

O
OCorthometric correction 183, 184 ^
OSU Ohio State University, Columbus, Ohio, USA 76 ^

P
ppb parts per billion 120 ^
ppm parts per million 120 ^
PRARE Precise Range And Range-Rate Equipment, not operational 352 ^
PRS President of the Royal Society 3, 122, 146, 419 ^

R
RTM residual terrain modelling 215, 217, 218, 229 ^

S
SI Système international d’unités, International System of Units 11, 37, 101, 120,
126 ^

í ¤. û
Acronyms ABCDEFGHIJKLMNOPRSTUW
xxi
SK-42 Reference system of the Soviet Union, also known as the Krasovsky 1940
reference ellipsoid 102 ^
SLR satellite laser ranging 352, 366 ^
SRAL Synthetic Aperture Radar Altimeter 353 ^
Ssa Sun, semi-annual tide 390 ^
SST satellite-to-satellite tracking 379 ^
STScI Space Telescope Science Institute 2 ^
SWH significant wave height 355, 356 ^

T
TCterrain correction 134, 135, 137–140, 250–252 ^
TOPEX/Poseidon Topography Experiment / Poseidon, American-French radar
altimetry satellite xix, 331, 352–354, 369, 382 ^

U
UCO University of California Observatories 2 ^

W
WGS84 World Geodetic System 1984 70, 101 ^

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 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 and

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 1 Fundamentals of the theory of gravitation

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.

Helmert condensation, both of which we will study in later chapters.

We will learn about important integral theorems like the theorems

í Õ ! ¤. û
 Gravitation between two masses 1.2
3
of Gauss and Green, with the aid of which we may infer the whole
potential field in space from field values given only on a certain surface.
Other similar examples are the Chasles theorem and the solution to
Dirichlet’s problem.
In chapters 2 and 3, we apply these fundamentals of potential theory
to derive a spectral representation of the Earth’s gravitational field, a
spherical-harmonic expansion. pallofunktio-
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: 2

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 .
The value of G was determined for the first time by Henry Cavendish3 3

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
4 1 Fundamentals of the theory of gravitation

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
4 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

5 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
in which the three-dimensional vectors of place of both the attracting
6 and attracted masses are defined as follows in rectangular co-ordinates:6

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 AlbertEinstein (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, a theory which he however
never fully accepted.

í Õ ! ¤. û
 The potential of a point mass 1.3
5
r = xi + yj + zk, R = Xi + Yj + Zk,
{︁ }︁
where the triad of unit vectors i, j, k is an orthonormal basis7 in ortonormaali
Euclidean space and kanta
√︂ 7

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. Thanks to this, one may affix 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.
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
GM
V= , ℓ = ∥r − R∥ . (1.4)
ℓ
6 As vector notation, one uses either →−v (an arrow above the letter) or v (bolding the
letter). Here we use the bold notation where possible.
7 This means 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 and b defined in the space.

í Õ ! ¤. û
6 1 Fundamentals of the theory of gravitation

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

GM⊕ = 3.986 005 · 1014 m3/s2 .

The currently best value based on measurement is slightly more precise:

GM⊕ = 3.986 004 418 (8) · 1014 m3/s2 .

The number in parentheses (8) represents the uncertainty in units of

the last given decimal. The relative uncertainty is thus 2 : 109 .

^ 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.

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

í Õ ! ¤. û
 Potential of a spherical shell 1.4
7
b dθ

p b

P
r
O
θ
ℓ
b

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

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
2 sin θ dθ
VP = 2πGρpb √ .
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 = rb sin θ dθ,
√
and remembering that ℓ = r2 + b2 − 2rb cos θ we obtain
w ℓ2
dℓ
VP = 2πGρpb2 .
ℓ1 rb

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
2 ℓ 4πGρpb2
VP = 2πGρpb = r .
rb ℓ=r−b

í Õ ! ¤. û
8 1 Fundamentals of the theory of gravitation

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 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
8 spherical shell is8

r P − rO r −r
aP = ∇V|P = −4πGρpb2 3
= −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
ℓ
[︂
VP = 2πGρpb2 = 4πGρpb.
rb 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. The attraction vanishes
inside the shell.
In figure 1.3 we have drawn the curves of potential and attraction —
or more precisely acceleration, attractive force per unit of mass. If a
body consists of many concentric spherical shells (like rather precisely

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

í Õ ! ¤. û
 Computing the attraction from the potential 1.5
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.

the Earth and many celestial bodies), then inside the body only those
layers of mass “interior” to — meaning 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 1 Fundamentals of the theory of gravitation

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

∂ ∂ ∂
∇=i +j +k .
∂x ∂y ∂z
{︁ }︁
kanta As before, i, j, k is 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 potential V and
9 distance ℓ, 1.3 and 1.4:9

∂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 the gravitational acceleration or attraction
vector when the source of the field is one point mass M. So, in this
concrete case, vector equation 1.7 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,

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 1.6
11
as it is known to always be. However, the potential energy of 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 more briefly the gravi-
tation or attraction 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

the gradient theorem 1.7.
The potential of a solid body is obtained similarly by replacing the
sum by an integral, in the following way:10 10

í Õ ! ¤. û
12 1 Fundamentals of the theory of gravitation

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

The symbol ρ inside the integral stands for matter density at

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

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 potential and
volume, respectively.

í Õ ! ¤. û
Potential of a solid body 1.6
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 density integrated 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 11
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 1 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, and ∆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

(︂ √︁ )︂
ln ∆z + ∆x2 + ∆y2 + ∆z2

and substituting the integration bounds yields

√︂
H − z + ∆x2 + ∆y2 + (H − z)2
V = G ln √︁ .
−z + ∆x2 + ∆y2 + z2
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 ℓ
(X − x)2 + (Y − y)2 + (H − z)2
√︂
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
(︂ )︂
= = G · · =
∂H2 ∂H ℓ ∂ℓ ∂H
H−z
= G · −ℓ−2 · 21 ℓ−1 · 2 (H − z) = −G .
ℓ3

í Õ ! ¤. û
Example: The potential of a line of mass 1.7
15
The third derivative, obtained in the same way:

3 (H − z)2
(︃ )︃ (︃ )︃
∂3 V ∂ H−z 1
= −G =G − 3 =
∂H3 ∂H ℓ3 ℓ5 ℓ
3 (H − z)2 − ℓ2
=G ,
ℓ5
and so on. The Taylor expansion is

∂3H V |
∂2H V | ⏟ ⏞⏞H=0 ⏟
V|H=0 ∂H⏟⏞⏞⏟
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 deriva-
tives 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 terrain or 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).

í Õ ! ¤. û
16 1 Fundamentals of the theory of gravitation

We shall return to this subject more extensively in connection

with terrain-effect evaluation, in sections 6.3 and 9.7.

^ 1.8 The equations of Laplace and Poisson

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 = V + V + V, (1.12)
∂x2 ∂y2 ∂z2

in which
def ∂2 ∂2 ∂2
∆ = ⟨∇ · ∇⟩ = + +
∂x2 ∂y2 ∂z2
12 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.

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 1.9
17
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)

This equation is called the Poisson equation.13 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 14
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 an example of what 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

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 1 Fundamentals of the theory of gravitation

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
very well.
An often-chosen definition of potential is based on requiring that, if
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, or “sea level”.
In the limit r → ∞ the value of the potential 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 —

í Õ ! ¤. û
Single mass-density layer 1.10
19
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
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 coating 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 2
= 4πGκ r .
r r
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.

í Õ ! ¤. û
20 1 Fundamentals of the theory of gravitation

P n
ℓ
κ
−κ
δ

^ Figure 1.4. A double mass-density layer.

^ 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 normal.
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 ≈ κδ. (1.17)

í Õ ! ¤. û
Double mass-density layer 1.11
21
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
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

í Õ ! ¤. û
22 1 Fundamentals of the theory of gravitation

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.

^ 1.12 The Gauss divergence theorem

^ 1.12.1 Presentation
15 The Gauss divergence 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

16 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

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

genius. “Princeps mathematicorum”.

16 Assuming that for the vector field a a potential V exists, see section A.4.

í Õ ! ¤. û
 The Gauss divergence theorem 1.12
23
Field line, field a

Normal n

Body
Sources surface

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

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

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, nega-
tive charges17 correspond to “sinks” through which liquid disappears. 17
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.
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

17 But the ”charges” for gravitation, the masses, are always positive.

í Õ ! ¤. û
24 1 Fundamentals of the theory of gravitation

— just like a liquid flow — from the space inside the surface ∂V through
the surface to the outside.
The Gauss divergence 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” (↑ ↑ ↑ · · · ) over the whole boundary surface delimiting
this inner space.

^ 1.12.3 The potential version of the Gauss divergence theorem

Let us write the Gauss divergence 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
notation ∂V for the surface of the body V. The presentational forms 1.20
and 1.19 are connected by 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

í Õ ! ¤. û
 The Gauss divergence theorem 1.12
25
a

a+
3

+
a2
a+
∆z 1

a−
1
−
a2

a−
3
∆y
∆x

^ Figure 1.6. A little rectangular box.

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 a− 3 on its lower face. A box has of
course six faces, in each of three co-ordinate directions both “up- and
downstream”.
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

í Õ ! ¤. û
26 1 Fundamentals of the theory of gravitation
(︃ )︃
∂a1 ∂a2 ∂a3
= + + ∆x ∆y ∆z,
∂x ∂y ∂z

the same expression as 1.21. So, in this simple case, the Gauss divergence
theorem 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.

^ 1.12.5 Example 2: The Poisson equation for a sphere

The Poisson equation 1.14 states

∆V = −4πGρ. (1.14)

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
⃓
= r =− 2 ,
∂n ∂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 divergence theorem 1.20
requires.

í Õ ! ¤. û
 The Gauss divergence theorem 1.12
27
z

+1
z

0
GM(0, 0, 0)

x
−1 y

0 −1
0
y +1
x

^ Figure 1.7. Eight-unit cube.

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

See figure 1.7. Assume that there is 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 18

spike at the origin, being zero elsewhere, and producing a value of 1
upon volume integration.

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
28 1 Fundamentals of the theory of gravitation

The surface integral is six times that over the top face
x w +1 w +1
(︄ )︄
⟨︁ ⟩︁ 1
a · n dS = −GM dx dy.
(x2 + y2 + 1) /2
top face −1 −1 3

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

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

2
= √︁ .
(y2 + 1) y2 + 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
Adding the six faces together yields
w +1 w +1
(︄ )︄
1
− 6 · GM dx dy = −6 · GM · 23 π =
−1 −1
(x2 + y2 + 1)
3/2

= −4πGM,
agreeing with the volume result above.

^ 1.13 Green’s theorems

Apply the Gauss divergence 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

í Õ ! ¤. û
Green’s theorems 1.13
29
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: 19

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
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:
1
U= ,
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 (remember that ∆U = 0
inside ∂V):
y x (︃ (︂ )︂)︃
1 1 ∂V ∂ 1
∆V dV = −V dS.
V ℓ ∂V ℓ ∂n ∂n ℓ

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
30 1 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.

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
which will go to zero in the limit ϵ → 0.
For the first surface integral we obtain using Gauss divergence theo-
rem 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

í Õ ! ¤. û
 Green’s theorems 1.13
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.

which also goes to zero for ϵ → 0.

The second surface integral (note that the normal to ∂V2 is pointing
inward to 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

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 in which 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.

í Õ ! ¤. û
32 1 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.

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 containing both the material body and
point 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 end result is — note that n is now the exterior normal to 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 limit, in which V is the entire empty space exterior to the
Earth where ∆V = 0, 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
 The Chasles theorem 1.14
33
^ 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
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 divergence

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.

í Õ ! ¤. û
34 1 Fundamentals of the theory of gravitation

20 This is the Chasles20 theorem, also called the Green equivalent-layer

theorem.
21 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
22 electric charges of a conductor are always on its outer surface.22

Equation 1.26, with substitution 1.27, simplifies in this case to

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 must be
known 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 determination (“geoid
determination”), ever since G. G. Stokes, has been based on this idea.

^ 1.15 Boundary-value problems

reuna-arvo- The boundary-value problem (BVP) is the problem of computing the
tehtävä
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.
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.

í Õ ! ¤. û
Boundary-value problems 1.15
35
potential V throughout the part of space exterior or interior to a given
boundary surface from given values relating to V on that boundary
surface, for example on the surface of the Earth. The simplest boundary-
value problem is Dirichlet’s23 problem: the potential V itself is given on 23
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 functional, see section 10.2.
The Dirichlet boundary-value problem in the form used in geodesy is:
determine the potential field V if its values are given on a closed surface
S, and V is harmonic (∆V = 0) outside surface S. In the vacuum of space,
the potential is always harmonic, as already noted earlier: 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
method also applied in physical geodesy. One must however note
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

23 Peter
Gustav Lejeune Dirichlet (1805–1859) was a German mathematician also
known for his contributions to number theory.

í Õ ! ¤. û
36 1 Fundamentals of the theory of gravitation

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
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-

í Õ ! ¤. û
 Exercise 1 – 1: Core of the Earth
37
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?
6. What is the order of the Laplace differential equation? kertaluku
7. Is a linear potential, V(x, y, z) = a + bx + cy + dz (a, b, c, d
constants), harmonic?
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)?

í Õ ! ¤. û
38 1 Fundamentals of the theory of gravitation

a a0

Σ1
d

∞ ∞

Σ2

^ Figure 1.11. Iron ore body.

∂
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 6371 km.
3. Calculate the attraction generated by the atmosphere outside it,
both as acceleration and as a fraction of the total Earth attraction.
4. How much is the attraction from the atmosphere inside the
atmosphere?

í Õ ! ¤. û
 Exercise 1 – 3: The Gauss divergence theorem
39
^ Exercise 1 – 3: The Gauss divergence theorem
There is a deposit (body) of iron ore inside the Earth, figure 1.11. The
deposit generates an attraction on the Earth’s surface, which has been
drawn as the a curve. We use the flat Earth approximation.
The true attraction curve is approximated by a simple function
⎧
⎨a if s ⩽ d
0
a=
⎩0 if s > d
(red dashed line), where s 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

where Σ1 is the surface of the Earth, see figure 1.11.

2. According to the Gauss divergence 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.

í Õ ! ¤. û
40 1 Fundamentals of the theory of gravitation

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
^ 2.1 The nature of the Laplace equation
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. Sometimes the alternative
notation ∇2 is used.
If we study gravitation as a field, the Laplace equation is more natural
than Newton’s approach. 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 attached 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 2 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: 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
1 (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 valid 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 2.2
43
values (“boundary values”) have to be given only on the boundary of
the part of space of interest.
For example, field values are given on the Earth’s surface. From this,
one calculates the values of the field in outer space, where the Laplace
equation applies — 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, sea level, is the level surface for z
co-ordinates z = 0. Write
(︃ )︃
(︁ )︁ ∂2 ∂2 ∂2 (︁ )︁
∆V = ∆ V(x, y, z) = 2
+ 2+ 2 X(x) · Y(y) · Z(z) ,
∂x ∂y ∂z

containing a trial solution

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

í Õ ! ¤. û
44 2 The Laplace equation and its solutions

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

Because this has to be true throughout the space, that is for all combi-
nations of values x, y, and 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 ∂2
X(x) = −ω2x X(x), Y(y) = −ω2y Y(y),
∂x2 ∂y2

(the reason for choosing a negative constant will become apparent), and

∂2 (︁ 2 2
)︁
Z(z) = ω x + ω y Z(z).
∂z2
Now, the solution is readily found at least to the first two equations:
2 they are harmonic oscillators, and their basis solutions2 are
harmoninen (︁ )︁ (︁ )︁
värähtelijä X(x) = exp ±iωx x , Y(y) = exp ±iωy y .

The solution of the Z equation, on the other hand, is exponential:

(︃ √︂ )︃
2
Z(z) = exp ±z ωx + ωy . 2

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.

í Õ ! ¤. û
The Laplace equation in rectangular co-ordinates 2.2
45
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 3
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: 4

2 Alternativebasis solutions are X(x) = sin ωx x, X(x) = cos ωx x etc. They

are equivalent to those presented because exp(iωx x) = cos ωx x + i sin ωx x and
exp(−iωx x) = cos ωx x − i sin ωx x.
3. . . although real-world shoeboxes are rarely square.

í Õ ! ¤. û
46 2 The Laplace equation and its solutions

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
kantafunktio are two-dimensional basis functions on the Earth’s surface, 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

V(x, y, z) = Vjk (x,y)

⏟ (︂ )︂⏞⏞ (︂ )︂⏟
∑︂
∞ ∑︂
∞
jx ky
(︂ √︁
z
)︂
vjk sin π sin π exp ±π j2 + k2 . (2.2)
L L L
j=1 k=1

Inside the z expression there may be 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
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.
kommutoiva We summarise equations 2.1 and 2.2 in commutative diagram 2.2.
kaavio
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
 The Laplace equation in rectangular co-ordinates 2.2
47
L

13πx 5πx
sin sin
Sea level L L

Figure 2.1. The exponential attenuation of the Fourier waviness of a harmonic

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

The takeaway from this is that the operation of vertically shifting the
field V from zero level to the level z, which is not straightforward in the
space domain, becomes simple — as in a straightforward multiplication

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 harmonic field V in the space and frequency
(wave-number) domains, commutative diagram. Rectangular
^ geometry.

í Õ ! ¤. û
48 2 The Laplace equation and its solutions

5 — in the frequency domain.5 The same applies in spherical co-ordinates,

where the frequency domain means spherical-harmonic coefficients, as
we shall see.

^ 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,

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

í Õ ! ¤. û
The Laplace equation in polar co-ordinates 2.3
49
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).

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
(︁ )︁

=⇒ q 2 − k2 = 0
=⇒ q 2 = 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
∑︂
∞
int
V (α, r) = a0 + rk (ak cos kα + bk sin kα) ,
k=1

but if we require existence — or, at least, good behaviour — at infinity6 6

í Õ ! ¤. û
50 2 The Laplace equation and its solutions

r → ∞, we need the second solutions,

∑︂
∞
V ext (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, spherical

co-ordinates, case.

^ 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.
pyörähdys- As the Earth is not precisely a sphere but rather an oblate ellipsoid of
ellipsoidi 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

6 In fact, limr→∞ V ext → ∞ but limr→∞ ∂ ext

∂r V = 0.

í Õ ! ¤. û
 Spherical, geodetic, ellipsoidal co-ordinates 2.4
51
Pole Z
r cos ϕ P
r
Equator
r sin ϕ

ϕ Y

λ
X

Greenwich meridian

^ Figure 2.3. Definition of 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
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, b is the polar vertaus-
radius, ellipsoidi
2 2
def a − b
e2 = 2
(2.7)
a
is the square of the first eccentricity,7 and in equations 2.5, h is the height 7

í Õ ! ¤. û
52 2 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.

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◦ .
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, (2.8)
Z = u sin β.

isoakselin If the semimajor axis of the Earth ellipsoid is a and its semiminor axis
puolikas
pikkuakselin 7 The parameter is connected to the Earth’s flattening f through the equation e2 =
puolikas 2f − f2 .

í Õ ! ¤. û
 The Laplace equation in spherical co-ordinates 2.5
53
b, it follows that E2 = a2 − b2 . Equation 2.7 tells us that E2 = a2 e2 .
The first eccentricity e is the eccentricity of the meridian ellipse, and
E = ae is the distance from the geocentre of the two focal points of this
ellipse, which lie on both sides of the geocentre in the equatorial plane.
Equations 2.8 tell us that all ellipses u = constant in the meridian plane
for different values of u have the same two focal points: they are confocal.
See figure 4.6.
We note still — see Heiskanen and Moritz (1967) figure 1-14 — that the
curves β = constant describe hyperbolas with these same focal points.

^ 2.5 The Laplace equation in spherical co-ordinates

The Laplace equation transformed to spherical co-ordinates reads (see
appendix E for a geometric proof):

∂2 V 2 ∂V 1 ∂2 V tan ϕ ∂V 1 ∂2 V
∆V = 2
+r + 2 − 2 + 2 = 0, (2.9)
∂r ∂r r ∂ϕ 2 r ∂ϕ r cos ϕ ∂λ2
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.
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
Yn (ϕ, λ)
Vnint (ϕ, λ, r) = rn Yn (ϕ, λ), Vnext (ϕ, λ, r) = , n = 0, 1, . . . ,
rn+1
(2.10)

of which the first is 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 Vn (ϕ, λ, r) are solid spherical pallofunktio
harmonics. The latter are harmonic functions everywhere in space except avaruus-
pallofunktio
í Õ ! ¤. û
54 2 The Laplace equation and its solutions

at the origin (2.10, rightmost equation) or at infinity (leftmost equation).

The functions Yn , called Laplace’s spherical harmonics, are
∑︂
n
Yn (ϕ, λ) = Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) . (2.11)
m=0

The functions Pnm are Legendre functions, on which more later on. With
the help of expression 2.11, we obtain, by using the second, physically
pallofunktio- realistic alternative from equations 2.10, the following solution or series
kehitelmä expansion for the potential V in outer space:

∑︂
∞
1 ∑︂
n
V(ϕ, λ, r) = Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) .
rn+1
n=0 m=0
(2.12)
The coefficients anm and bnm are called the coefficients of the spherical-
harmonic expansion, in short, spectral coefficients. Together they re-
present the function V, in somewhat the same way that the Fourier
coefficients vjk do in rectangular co-ordinates in equation 2.2. The
asteluku, subscripts n and m are called degree and order.
järjestysluku
Often we will be using a somewhat freer notation for the scaled
functions Yn Rn+1 . For example, if we expand the disturbing potential
/︁
häiriö-
potentiaali T into spherical harmonics, we shall use the notation Tn (ϕ, λ) for its
pallofunktio 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 2.10 one sees that for different values of
the degree n the function Vn (ϕ, λ, r) has a different dependence on

í Õ ! ¤. û
 Self-test questions
55
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 or sea level.
The dependence is
Y (ϕ, λ)
Vn (ϕ, λ, r) = n n+1 .
r
At sea level
Y (ϕ, λ) def
Vn (ϕ, λ, R) = n n+1 = Vn (ϕ, λ).
R
Therefore we may write
(︂ )︂n+1
R R + H −(n+1)
(︂ )︂
Vn (ϕ, λ, r) = 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 exponential,
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.

^ Self-test questions
1. In what essential way does the approach of the Laplace equation
differ from Newton’s approach?
2. How does the linearity of the Laplace equation help in finding
solutions?
3. How does separation of variables work?
4. Why does solving the Laplace equation require boundary condi-
tions?
5. Show, using equations 2.8, that for the curves u = constant in the
meridian plane Y = 0, the sum of the distances from the curve point
[︂ √ ]︂T [︂ ]︂T
2 2
u + E cos β 0 sin β to the focal points ±E 0 0
is constant (and that these curves thus are confocal ellipses), and
that for the curves β = constant the difference of these distances is

í Õ ! ¤. û
56 2 The Laplace equation and its solutions

constant (and that these curves thus are confocal hyperbolas). See
figure 4.6.

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

3
^ 3.1 Legendre functions
In equations 2.11 and 2.12, the functions P are Legendre1 functions that 1
pop up whenever we solve a Laplace-like equation in spherical co-
ordinates. There exist various effective, so-called recursive algorithms
for their calculation, 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 as 64-bit integers — not to mention 360!, the
factorial of 360, which overflows even the standard 64-bit floating-point
format. Heiskanen and Moritz’s (1967) equation 1-62, contrary to what
is stated there, is not directly suitable for computer use!

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.

– 57 –
58 3 Legendre functions and spherical harmonics

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

As a 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) = 32 t2 − 1
2 P2 (sin ϕ) = − 34 cos 2ϕ + 1
4
5 3 3 5 3
P3 (t) = 2t − 2t P3 (sin ϕ) = − 8 sin 3ϕ + 8 sin ϕ
1 4 2 35 5 9
(︁ )︁
P4 (t) = 8 35t − 30t + 3 P4 (sin ϕ) = 64 cos 4ϕ − 16 cos 2ϕ + 64
1
(︁ 5 3 + 15t
)︁
P5 (t) = 8 63t − 70t
1 6 4 2
(︁ )︁
P6 (t) = 16 231t − 315t + 105t −5

The first Legendre polynomials are listed in table 3.1. Higher polyno-
mials than this are rarely needed in manual computation.
kantafunktio For comparison, the Fourier basis functions (like, in a more complicated

−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 3.1
59
way, sines and cosines together as well!)
x
(︂ )︂
Fj (x) = exp 2πij ,
L
in which i2 = −1, can also be calculated 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, the equatorial plane: Pn (−t) = Pn (t), or equiv-
(︁ )︁
alently Pn sin(−ϕ) = Pn (sin ϕ). This means that their values
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
open 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

í Õ ! ¤. û
60 3 Legendre functions and spherical harmonics

15
P32
P31
10 P21
P25,25
P33 5 · 1030 P33

5 P32
P22
P11
0
P21 P31

P32
−5
−1 −0.5 0 −→ t 0.5 1

Figure 3.2. Some associated Legendre functions. Note the extremely different
^ scale used for the function P25,25 , see equation 3.8.

◦ 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-
asteluku, ing on the values of degree number n and order number m.
järjestysluku [︁ ]︁
◦ Figure 3.2 suggests that the polynomials Pnm (t) go on t ∈ −1, 1 ,

^ 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 3.1
61
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”.

^ 3.1.3 Surface spherical harmonics

Starting from equation 2.11, 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,
def nm
Ynm (ϕ, λ) = (3.3)
⎩Pn|m| (sin ϕ) sin |m| λ if m < 0,
⎧
⎨a if m ⩾ 0,
def nm
vnm = (3.4)
⎩bn|m| if m < 0.

These are the surface spherical harmonics of degree n and order m. pinta-
pallofunktio
Surface spherical harmonics, like also solid spherical harmonics,
come in three kinds:
Zonal harmonics m = 0. These functions depend only on latitude. vyöhyke-
funktiot
Sectorial harmonics m = n. the algebraic signs of these functions
sektorifunktiot
depend only on longitude and not on latitude. The functions
themselves however do depend on both latitude and longitude!
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

í Õ ! ¤. û
62 3 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
^ fashion similar to that of sine or cosine functions.

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 proper wave functions (sines or cosines): the
connection to those is complicated at least. It nevertheless makes sense
to speak of their wavelength.
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 2.11, the expressions cos mλ and sin mλ go around a full
circle, the equator, 0◦ ⩽ λ < 360◦ , or 0 ⩽ λ < 2π, precisely 2m times
puoli- through zero. The “semi-wavelength” is thus
aallonpituus
2πR R
= πm,
2m

í Õ ! ¤. û
 Legendre functions 3.1
63

Figure 3.4. Surface spherical harmonics as maps. Horizontal axis λ ∈

0, 360◦ = 0, 2π , vertical axis ϕ ∈ −90◦ , 90◦ = − π 2 , π 2 .
[︁ )︁ [︁ )︁ [︁ ]︁ [︁ /︁ /︁ ]︁

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

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, a representative 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

í Õ ! ¤. û
64 3 Legendre functions and spherical harmonics

^ 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

that only details the size of continents — order of magnitude 1000 km —

pallofunktio- will be visible. On the other hand, experimental spherical-harmonic
kehitelmä expansions of the topography go even up to degree 10 800 (Balmino et al.,
2012).

^ 3.2 Symmetry properties of the spherical-harmonic

expansion
We recapitulate the spherical-harmonic expansion:

∑︂
∞
1 ∑︂
n
V(ϕ, λ, r) = Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) .
rn+1
n=0 m=0
(2.12)

^ 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
or antisymmetric in the argument ϕ. This means that either (symmetric
case)
(︁ )︁
Pnm (sin ϕ) = Pnm sin(−ϕ)

í Õ ! ¤. û
Symmetry properties of spherical harmonics 3.2
65
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
(︁
. (3.2)
dtm
We need to answer two questions:
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, equation 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 2.12 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
visually judge its symmetry properties. Do not believe, test.

í Õ ! ¤. û
66 3 Legendre functions and spherical harmonics

^ Tableau 3.4. Plotting a surface spherical-harmonic map. Note that the octave
code for associated Legendre functions used here contains an additional factor
(−1)m , so compared to equation 3.2 all odd orders have the opposite sign.

% 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’)

^ 3.2.2 Dependence on longitude λ

This dependence works though the “Fourier basis functions” cos mλ
and sin mλ. The interesting property here is rotational symmetry: does
the spherical-harmonic expansion 2.12 change when we change the
longitude λ?
We see immediately that there will be dependence on λ if any coef-
ficient anm , bnm , m ̸= 0 is non-zero. So, in order to obtain rotational
symmetry, all coefficients anm and 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

í Õ ! ¤. û
 Orthogonality of Legendre functions 3.3
67
identically, so the coefficients b00 , b10 , b20 , . . . simply do not matter.
They may be any value, including zero. The coefficients a00 , a10 , a20 ,
. . . 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) = an Pn (sin ϕ),
rn+1
n=0

def def
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
⟩︁ def +1 ⎨ 2 if n = n ′ ,
2n + 1
⟨︁
Pn · Pn ′ t = Pn (t) Pn ′ (t) dt = (3.5)
−1 ⎩0 if n ̸= n ′ .
This orthogonality is just one example of a more general way to look
at functions and integrals over functions. There exists here 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 2
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 3

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.

í Õ ! ¤. û
68 3 Legendre functions and spherical harmonics

co-ordinates (ψ, α). So, we have

⎧
⟩︁ def x ⎨ 4π if n = n ′ ,
Pn (cos ψ) Pn ′ (cos ψ) dσ = 2n + 1
⟨︁
Pn · Pn ′ σ =
σ ⎩0 if n ̸= n ′ ,
(3.6)
in which ψ is the angular distance from the integration point to the
origin ψ = 0 of the parametrisation (ψ, α). Equation 3.6 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). (3.7)
Now the modified scalar product — the mean product over the unit
sphere — is
⎧
⟨︁ ⟩︁ def 1 x ⎨1 if n = n ′ ,
Pn · Pn ′ σ = P (cos ψ) Pn ′ (cos ψ) dσ =
4π σ n ⎩0 if n ̸= n ′ ,

4 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), m ̸= 0. (3.8)
(n + m)! nm

In this case, the orthonormal functions are those of equation 3.3, but

3 Carl
Gustav Jacob Jacobi (1804–1851) was a German mathematician known for his
work on elliptic functions.
4 And also ⎧
⟨︁ ⟩︁ def 1 w +1 ⎨1 if n = n ′ ,
Pn · Pn ′ t = 2 Pn (t) Pn ′ (t) dt =
−1 ⎩0 if n ̸= n ′ ,

again, the mean product over the domain of integration.

í Õ ! ¤. û
 Low-degree spherical harmonics 3.4
69
normalized:
⎧
⎨P
nm (cos ψ) cos mα if m ⩾ 0,
Y nm (ψ, α) =
⎩Pn|m| (cos ψ) sin |m| α if m < 0.

The scalar product that applies is

⟨︁ ⟩︁ def 1 x
Y nm · Y n ′ m ′ σ = Y (ψ, α) Y n ′ m ′ (ψ, α) dσ =
4π σ nm ⎧
⎨1 if n = n ′ and m = m ′ ,
=
⎩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 2.12 for degree n = 0
is
1 1 a
V0 (ϕ, λ, r) = r a00 P00 (sin ϕ) = r a00 P0 (sin ϕ) = r00 ,
from which
a00 = GM.
So, a00 represents the force field of a point mass or spherically symmetric
mass distribution centred at the origin. The higher spherical-harmonic pallofunktio-
coefficients are “perturbations” on top of this. kerroin

The expansion for the degree-one coefficients looks as follows:

1
V1 (ϕ, λ, r) = (a cos ϕ cos λ + b11 cos ϕ sin λ + a10 sin ϕ) .
r2 11
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
kanta
í Õ ! ¤. û
70 3 Legendre functions and spherical harmonics

yielding
1 ⟨︁ ⟩︁
V1 (r) = (a 11 i + b 11 j + a 10 k) · r .
r3
The potential field of a dipole is

G ⟨︁ ⟩︁
V(r) = 3
d·r ,
r
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

Earth’s gravitational field’s dipole moment.
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:
t
y y y ρr dV
d⊕ = r dm = ρr dV = ρ dV · t⊕ = M⊕ · rcom ,
⊕ ρ dV
⊕ ⊕ ⊕

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 , and a10 vanish. If the equations of motion
of 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.
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.

í Õ ! ¤. û
 Splitting a function into degree constituents 3.5
71
d

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

^ 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) = (3.9)
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
this degree constituent equation 3.9 there is a certain similarity with the asteosuus-
projection or coefficient computation equation B.11. Nevertheless, here yhtälö
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) = f(ϕ, λ, R) = fn (ϕ, λ)
n=0

on the surface of the sphere r = R.

To prove the degree constituent equation, we choose without loss of
generality as the “north pole” of the co-ordinate system the evaluation

í Õ ! ¤. û
72 3 Legendre functions and spherical harmonics

point (ϕ, λ), so ϕ = 90◦ . Then, ϕ ′ = 90◦ − ψ. By writing (see equation

2.12):

(︁ ′ ′ )︁ ∑︂
∞ ∑︂
n
Pnm cos ψ (anm cos mλ ′ + bnm sin mλ ′ ) ,
(︁ )︁
f ϕ ,λ =
n=0 m=0

substituting this into the degree constituent equation 3.9, and exploiting
the orthogonality of the Legendre functions, we obtain on the right-hand
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 3.6:

2n + 1 4π
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 2.11 and

Pn0 (1) = 1,
Pnm (1) = 0, m ̸= 0.

As this applies for every point (ϕ, λ), it follows that the degree con-
stituent equation 3.9 is universally true. Note that the values an depend
on the point choice!

í Õ ! ¤. û
 Spectral representations of various quantities 3.6
73
^ 3.6 Spectral representations of various quantities

^ 3.6.1 Potential
Starting from equation 2.10, we write the spectral expansion of the
geopotential V in space:
∑︂
∞ (︂ )︂n+1
R
V(ϕ, λ, r) = r Vn (ϕ, λ), (3.10)
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
1 ∑︂
m=0 n
= n+1 vnm Ynm (ϕ, λ).
R
m=−n

Here, the basis functions Ynm are given by equation 3.3:

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

and the coefficients, equation 3.4:

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

On the Earth’s surface (r = R) this yields

∑︂
∞ ∑︂
∞
1 ∑︂
n
V(ϕ, λ, R) = Vn (ϕ, λ) = vnm Ynm (ϕ, λ). (3.11)
Rn+1
n=0 n=0 m=−n

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

Again, as in section 2.2 for rectangular geometry, it is seen that the kaavio
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
74 3 Legendre functions and spherical harmonics

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. Vertically or radially shifting the harmonic field V, in the space
^ and frequency (degree number) domains. Spherical geometry.

^ 3.6.2 Gravitation
5In the Neumann5 boundary-value problem we solve a function V of
reuna-arvo- which the normal derivative, ∂ V, is given on a closed surface in space,
∂n
tehtävä for example the surface of a body.
∂ ∂
If the body is a sphere, one may assume ∂n V = ∂r V and work with
spherical-harmonic expansions. By differentiating equation 3.10 we
obtain

∂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 CarlGottfried Neumann (1832–1925) was a German mathematician who studied the

Dirichlet boundary-value problem.

í Õ ! ¤. û
Spectral representations of various quantities 3.6
75
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 . (3.12)
n+1
n=0 n=0

For the gravitation one may write, analogously to expression 3.11 for
the potential:

∑︂
∞
def
∑︂
∞
1 ∑︂
n
g(ϕ, λ, R) = gn (ϕ, λ) = gnm Ynm (ϕ, λ), (3.13)
Rn+1
n=0 n=0 m=−n

and comparison yields

n+1
gnm = − v . (3.14)
R nm
This is an interesting result worth thinking about:
1. Firstly, note how simple the connection 3.14 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
may derive from these the degree constituent functions gn (ϕ, λ) asteosuus-
using the method presented earlier. In this way we can then funktio
obtain the solution by means of equation 3.12 for the whole
exterior geopotential field! This is the basic idea of geopotential —
or geoid — determination, from the spectral perspective.

í Õ ! ¤. û
76 3 Legendre functions and spherical harmonics

^ 3.7 Often-used spherical-harmonic expansions

Of the existing global spherical-harmonic expansions we must mention
the already outdated EGM96. It was developed by researchers from
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
6 360. Its standard presentation6 is

GM
(︃ ∑︂
360 (︂ )︂ ∑︂
a n
n )︃
V= r⊕
(︁ )︁
1+ r Pnm (sin ϕ) Cnm cos mλ + Snm sin mλ . (3.15)
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
ratahäiriöt derived primarily from analysis of satellite orbit perturbations. Because
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)

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

í Õ ! ¤. û
 Ellipsoidal harmonics 3.8
77
and satellite radar altimetric data (over the ocean). After the launches
of the gravimetric satellite missions CHAMP, GRACE, and GOCE, and as a
result of their measurements, nowadays at least the harmonic degree
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
model, the newest and best spherical-harmonics model from the time pallofunktio-
just before the satellite gravity missions. The values tabulated are n, m, malli
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) . (3.16)
n=2 m=0

def
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 3.7 and 3.8,
the relationship with the parameters C, S is
{︄ }︄ {︄ }︄
Jn0 √ Cn0
= − 2n + 1 ,
Kn0 Sn0
{︄ }︄ √︄ {︄ }︄ (3.17)
Jnm (n − m)! Cnm
= − 2 (2n + 1) , m ̸= 0.
Knm (n + m)! Snm

^ 3.8 Ellipsoidal harmonics

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

í Õ ! ¤. û
78 3 Legendre functions and spherical harmonics

^ 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

because the maths needed is more complicated. Moreover, ellipsoidal

co-ordinates are mostly only theoretically interesting and not in any
broad use within geodesy.
The form of presentation is

V(β, λ, u) =

í Õ ! ¤. û
 Ellipsoidal harmonics 3.8
79
^ 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
(︁ 3 2 1 )︁ z + 1 3
Q2 (z) = 4 z − 4 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 +3
z−1 2

∑︂
∞ ∑︂n
Qnm i u
(︁ )︁
= (︁ Eb )︁ Pnm (sin β) (Aenm cos mλ + Benm sin mλ) , (3.18)
n=0 m=0
Qnm 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 3.18
gives a real result for real-valued coefficients Aenm , Benm .
Those interested in the derivation of the above equation 3.18 can find
it in Heiskanen and Moritz (1967) section 1-20 or other textbooks on
potential theory.

^ 3.8.1 The scaling to standard form of the expansion

Assume the origin to be in the centre of mass of the Earth, so Ae10 ≈ 0,
Ae11 ≈ 0, Be11 ≈ 0.
We can also show that in expansion 3.18 the first coefficient must be

GM⊕ E
Ae00 = Ae0 = arctan
E b
and the expansion specialised for a rotationally symmetric field becomes

∑︂
∞ ∑︂
∞
Qn i u
(︁ )︁
∗
V (β, u) = Vn∗ (β, u) = (︁ Eb )︁ Ae∗
n0 Pn (sin β). (3.19)
n=0 n=0
Qn i E
Also
Q0 i u
(︁ )︁
GM⊕ E
V0 (u) = V0∗ (u) = (︁ Eb )︁ arctan ,
Q0 i E E b

í Õ ! ¤. û
80 3 Legendre functions and spherical harmonics

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 (3.20)
E E b
we obtain
GM⊕ E
V0 (u) = V0∗ (u) = arctan u . (3.21)
E
This corresponds to the “central field” of a spherical-harmonic expansion
GM⊕ r . Using this, we may scale equation 3.18 by substituting the
/︁

above identities 3.20. The coefficients need to be divided by the constant

expression
GM⊕ E
arctan ,
E b
as the central field, expression 3.21, 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
in which we have also introduced fully normalized coefficients Cnm ,
e
Snm and Legendre functions Pnm (sin β).
This is an ellipsoidal-harmonic expansion that agrees with the
standard-form spherical-harmonic expansion 3.15, with the total mass
of the Earth outside the parentheses and the coefficients dimension-
less. 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 3.15 or
3.16 and ellipsoidal expansion 3.18, if the flattening of the Earth → 0,

í Õ ! ¤. û
 Ellipsoidal harmonics 3.8
81
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 3.18 yields

V(u, β, λ) = V(r, ϕ, λ) =
∑︂∞ ∑︂n (︂ )︂
a n+1
= r Pnm (sin ϕ) (Aenm cos mλ + Benm sin mλ) , (3.22)
n=0 m=0

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

/︁

and relations 3.17, 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) , m ̸= 0.
(n + m)! Snm

Substituting these into equation 3.22 affirms its equivalence with equa-
tions 3.15 and 3.16 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.
◦ The convergence behaviour on a flattened Earth will be better. suppeneminen

í Õ ! ¤. û
82 3 Legendre functions and spherical harmonics

This is because, due to the Earth’s flattening, the equator is some

21 km further from the Earth’s centre than the poles. Therefore,
high-degree spherical harmonics in particular will have difficulty
converging efficiently both at the poles and in the equatorial
region. This problem is worst for very high-degree expansions
(for example Wenzel, 1998). Already for a degree number of 360,
puoli- the semi-wavelength of a spherical harmonic will be only 55 km!
aallonpituus

^ 3.8.4 Disadvantage of using ellipsoidal harmonics

Evaluation of an ellipsoidal-harmonic expansion is clearly more labori-
ous and expensive than that of a spherical-harmonic one, in terms of
computer resources.

^ Self-test questions
1. What are the harmonic degree and harmonic order in a spherical-
harmonic expansion? How do they relate to the resolution of the
expansion on the Earth’s surface?
2. What types of spherical harmonics are there? Explain their
dependence on latitude and longitude.
3. 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?
4. 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.
5. How does the attenuation of spherical harmonics with height
behave? Why does a gravimetric satellite that is trying to map the
gravity field of the Earth at a high resolution fly in as low an orbit
as possible?

í Õ ! ¤. û
 Exercise 3 – 1: Attenuation with height of a spherical-harmonic expansion
83
6. What does the degree constituent equation express?
7. Which spherical-harmonic coefficients are associated with the
dipole moment of the Earth’s mass distribution? Why are they
missing from tableau 3.5?

^ Exercise 3 – 1: Attenuation with height of a

spherical-harmonic expansion
If
∑︂
∞ ∑︂
∞ (︂ )︂n+1
R
V(ϕ, λ, r) = Vn (ϕ, λ, r) = r Vn (ϕ, λ),
n=0 n=0
we may call
Vn (ϕ, λ, r) (︂ R )︂n+1
= r
Vn (ϕ, λ)
the attenuation factor of the potential with height.
Differentiation with respect to r yields
∂Vn (ϕ, λ, r) n + 1 R n+2
(︂ )︂
=− r Vn (ϕ, λ), (3.23)
∂r R
or, because, at sea level, similarly
⃓
∂Vn (ϕ, λ, r) ⃓⃓ n+1
=− Vn (ϕ, λ), (3.24)
∂r ⃓
r=R
R
it follows that the attenuation factor for the attraction is the ratio of
expressions 3.23 and 3.24:
(︂ )︂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.
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?

í Õ ! ¤. û
84 3 Legendre functions and spherical harmonics

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 2.12. In it, Pnm (sin ϕ) = Pnm (t) 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.
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) = t −1 .
2n n! dtn
We can observe the following properties:
◦ Differentiating a symmetric function of t will produce an antisym-
metric function, and vice versa.
(︁ )︁
◦ The function t2 − 1 and its powers are symmetric.
◦ 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 hemispheres.
(︁ )︁
◦ Similarly, for even n, Pn (sin ϕ) = Pn sin(−ϕ) , and for odd n,
(︁ )︁
Pn (sin ϕ) = −Pn sin(−ϕ) .

Questions
1. What is the corresponding rule for the functions Pnm , in
other words, for which values n and m 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:

í Õ ! ¤. û
 Exercise 3 – 3: Algebraic-sign domains of spherical harmonics
85
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, so V(ϕ, λ, r) = V(−ϕ, λ, r), which of
the spherical-harmonic coefficients anm and bnm drop out
of 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
We have seen in section 3.1 that the associated Legendre functions Pnm (t)
have precisely n − m + 1 algebraic-sign intervals on their interval of
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
[︁ )︁

circle. How many algebraic-sign domains — grey or white areas, visible or

occluded — are there in figure 3.3 for each surface spherical harmonic
⎧
⎨P (sin ϕ) cos mλ if m ⩾ 0,
nm
Ynm (ϕ, λ) = ?
⎩Pn|m| (sin ϕ) sin |m| λ if m < 0

í Õ ! ¤. û
86 3 Legendre functions and spherical harmonics

^ 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 pyörähdys-
revolution, the gravity field of the Earth can also be approximated by a ellipsoidi
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 model 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 häiriö-
potential and the gravity anomaly, which then again will be intercom- potentiaali
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 an luotiviiva

– 87 –
88 4 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.

amount called the plumb-line deflection. This deflection of the plumb

luotiviivan line is also a very small angle.
poikkeama
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 by V ∗ — a quantity rarely used in geodesy
— then the normal gravity potential or normal potential U is

U = V ∗ + Φ,

í Õ ! ¤. û
 The centrifugal force and its potential 4.2
89
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: V ∗ , like V,

is defined in a non-rotating or 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 or attraction.

^ 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
90 4 The normal gravity field

It follows that
√︂⟨︁ ⟩︁ √︁
p = ∥p∥ = p · p = X2 + Y 2 .

Now the centrifugal force — or rather, acceleration — is

fω = ω2⊕ p = ω2⊕ (Xi + Yj) , (4.1)

with ω⊕ the rotation rate of the Earth in radians per unit of time. If X
and Y are in metres and ω⊕ is in radians per second, then fω is obtained
in m/s2 .
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
1 Coriolis1 force. Fluids — water, air — on the Earth’s surface, if they

are at rest, only sense gravity, which includes the centrifugal force.
Currents in addition also sense the Coriolis force, which deflects them
sideways and causes the well-known vortex phenomena in the oceans
and atmosphere, such as hurricanes.
We may describe centrifugal force as the gradient of a potential. If
we write for this centrifugal potential

Φ = 12 ω2⊕ X2 + Y 2 ,
(︁ )︁

we may directly calculate the gradient

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

which corresponds to the above centrifugal-force equation 4.1.

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 4.2
91
If we add to the gravitational potential V the centrifugal potential Φ,
we obtain the gravity potential or geopotential W:

W = V + Φ.

We may also derive from the centrifugal potential Φ the following

equation by differentiating it twice:
∂ 2 ∂ 2 ∂
∆Φ = ∇2 Φ = ⟨∇ · fω ⟩ = 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 geopotential or gravity potential.

The difference between gravitation and gravity is essential. The
force, or acceleration, of gravitation g∗ = ∇V is just an attractive force,
whereas 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 Loránd
Eötvös.2 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 geopotential W. 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
92 4 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.
{︁ }︁
Let i, j, k be an orthonormal basis along the (x, y, z) axes. Then, 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 · ∇W⟩ ,
∂e ∂x ∂y ∂z
which vanishes if and only if the vectors e and ∇W are perpendicular to
each other. 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

The plane that at P has the same direction as the level surface is called
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
From this follows the difference in W values between the surfaces
(g = ∥g∥ = ∥∇W∥):
g
δW ≈ −ϵg ≈ − (x − x0 )2 .
2ρx

í Õ ! ¤. û
 Level surfaces and plumb lines 4.3
93
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.

Differentiating (note that W here is now the geopotential on the tangent

or horizontal plane), yields3 3

∂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

we obtain the mean or Germain4 curvature, in most locations a positive 4

number:
Kx + Ky ∂xx W + ∂yy W
J= =− ,
2 2g

3 We use here compact Euler notation for partial derivatives, ∂xx , ∂yy , ∂zz , which is
often convenient.

í Õ ! ¤. û
94 4 The normal gravity field

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.

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).

^ 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

4 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 4.3
95
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 .
/︁

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
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, or normal plumb lines, in the same way. The difference is, however,
that we can find a simple mathematical expression for gravity on the
surface of the reference ellipsoid, for example equation 4.8. A good
approximation is
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
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
This also means that the direction of the normal plumb line at height h
is φ(h) = φ(0) + κ∗x h, in which numerically κ∗x = 0.171 ′′ km−1 · sin 2φ
(Heiskanen and Moritz, 1967, equation 5-34).

í Õ ! ¤. û
96 4 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.

^ 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
same time the height h from the reference ellipsoid, a purely geometric
quantity.
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
plumb line n was measured astronomically: astronomical latitude
Φ 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

í Õ ! ¤. û
 The normal potential in ellipsoidal co-ordinates 4.5
97
natural co-ordinates.
Instead of the potential, orthometric height H may be 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
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 3.18, the expansion of the geopo-
tential into ellipsoidal harmonics. The normal potential U is required to ellipsoidi-
be constant on the reference ellipsoid u = b. We expand the centrifugal funktio
potential Φ into ellipsoidal harmonics, obtaining

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) =

(︁ )︁ (︁ )︁

= − 31 ω2⊕ u2 + E2 P2 (sin β) − P0 (sin β) .

(︁ )︁ (︁ )︁

In addition, based on equation 3.19 we have for the rotationally sym-

metric normal gravitational potential V ∗ :
∑︂
∞ ∑︂
∞ (︁ u )︁
Q n i
V ∗ (β, u) = Vn∗ (β, u) = (︁ Eb )︁ Ae∗
n0 Pn (sin β). (3.19)
n=0 n=0
Qn i E

í Õ ! ¤. û
98 4 The normal gravity field

Now
U(β, u) = V ∗ (β, u) + Φ(β, u).
On the reference ellipsoid u = b we have as a requirement U(β, b) = U0 ,
def
which is possible only if (Ae∗ e∗
n = An0 ):

1 2 1 2 2
U0 = Ae∗ = Ae∗
(︁ 2 2
)︁
0 + ω
3 ⊕
b + E 0 + 3 ω⊕ a ,

0 = Ae∗
1 ,
1 2 1 2 2
0 = Ae∗ = Ae∗
(︁ 2 2
)︁
2 − 3 ω⊕ b + E 2 − 3 ω⊕ a ,

0 = Ae∗
n, 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 and b are known. The
result, Heiskanen and Moritz (1967) equation 2-61, is

GM⊕ E
U0 = arctan + 13 ω2⊕ a2 . (4.6)
E b
From this follows

1 2 2 GM⊕ E
Ae∗
0 = U0 − 3 ω⊕ a = arctan .
E b
The normal gravity potential U is obtained as follows:
V ∗ (u)
⏟ ⏞⏞
0
⏟
∗ GM⊕ E
U(β, u) = V (β, u) + Φ(β, u) = arctan u +
E
A e∗
P2 (sin β) Φ(β,u)
⏟ ⏞⏞2 ⏟ Q (︁i u )︁ (︁⏟ ⏞⏞ ⏟ ⏟ ⏞⏞ )︁ ⏟
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,

in which C0 , C1 , and C2 are suitable functions of u. The function V0∗ is

the term for n = 0 in expansion 3.19, equation 3.21.
On the surface of the reference ellipsoid (u = b), using a2 = b2 + E2 :

U(β, b) =

í Õ ! ¤. û
 Normal gravity on the reference ellipsoid 4.6
99
V ∗ (b)
⏟ ⏞⏞
0
⏟ ⏟ Ae∗
2 P2 (sin β)
⏞⏞ ⏟ ⏟
Φ(β,b)
⏞⏞ ⏟
GM⊕ E
= arctan + 12 ω2⊕ a2 sin2 β − 61 ω2⊕ a2 + 12 ω2⊕ a2 cos2 β =
E b
GM⊕ E
= arctan + 13 ω2⊕ a2 ,
E b
the constant U0 (equation 4.6), 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:

aγ sin2 β + bγa cos2 β

γ(β) = √︁ b . (4.7)
a2 sin2 β + b2 cos2 β
It is seen that γa is normal gravity on the equator (β = 0) and γb normal
gravity on the poles (β = ±90◦ ).
Equations 2.5 and 2.8 yield

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

and /︂(︁
Z 1 − e2 N
)︁
sin φ 1 Z a2
tan φ = cos φ = √ 2 = √ = tan ϕ,
X + Y2 N
/︁ 1 − e2 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,
in which the latitude angle φ is the geodetic or geographic latitude. β
is still the reduced latitude. Now it can be shown (exercise!) that redukoitu
leveysaste
aγ cos2 φ + bγb sin2 φ
γ(φ) = √︁ a . (4.8)
a2 cos2 φ + b2 sin2 φ
This is the famous Somigliana–Pizzetti5 equation. These geodesists de- 5

í Õ ! ¤. û
100 4 The normal gravity field

Z
Q

a
P

β
F2 E ϕ φ F1 X/Y
a O

Hyperbola β = constant

Figure 4.6. Geometry of the meridian ellipse and various types of latitude, as
^ well as the focal points F1 and F2 .

monstrated for the first time that an “ellipsoidal” normal gravity field
which has the reference ellipsoid as one of its equipotential or level
surfaces exists exactly, and that the gravity formula is also a closed
expression in geographic latitude.

^ 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

5 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 4.7
101
a the equatorial radius of the ellipsoid of revolution, its semimajor isoakselin
axis puolikas

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

potential is the GRS80, the Geodetic Reference System 1980:

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 the GRS80, but the constant J2
is used instead, which is a defining quantity for the gravitational field,
see equation 3.16.
The WGS84 (World Geodetic System 1984) used by the GPS 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 , (4.9)
(︁ )︁

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

í Õ ! ¤. û
102 4 The normal gravity field

+ 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 . (4.10)
(︁ )︁

Here, the unit of potential is m2/s2 , and the unit of gravity, m/s2 . φ
is geodetic latitude; h (in metres) is the height above the reference
ellipsoid. 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 . . . · 10−4 s−2 is the vertical gradient of gravity
on the equator.
Other gravity formulas and reference ellipsoids still in legacy use
(and slowly fading away) 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 equation 4.9, 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?
2. How physically realistic is this?

í Õ ! ¤. û
 The normal potential as a spherical-harmonic expansion 4.8
103
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 .

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 realistic: 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-
also contains, besides the second-degree harmonic, higher-degree kehitelmä
harmonics. If we write, as is customary, the potential outside the Earth

í Õ ! ¤. û
104 4 The normal gravity field

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

1981; Heiskanen and Moritz, 1967).

Non-normalised Fully normalised

∗ ∗
J∗2 = J∗2,0 = 1082.63 · 10−6 J2 = −C2,0 = 484.166 854 896 · 10−6
∗ ∗
J∗4 = J∗4,0 = −2.370 912 22 · 10−6 J4 = −C4,0 = −0.790 304 073 · 10−6
∗ ∗
J∗6 = J∗6,0 = +0.006 083 47 · 10−6 J6 = −C6,0 = +0.001 687 251 · 10−6
∗ ∗
J∗8 = J∗8,0 = −0.000 014 27 · 10−6 J8 = −C8,0 = −0.000 003 461 · 10−6

in the following form (Heiskanen and Moritz, 1967 equation 2-39, also
equation 3.16):

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, V ∗ , into the
form
GM⊕
(︃ ∑︂
∞ (︂ )︂n
a
)︃
∗ ∗
V (ϕ, r) = r 1− Jn r Pn (sin ϕ) ,
n=2
even
def
which contains only even coefficients J∗n = J∗n0 , because the normal
field is symmetric about the equatorial plane.
The coefficients for the GRS80 normal gravitational potential are
6 found6 in table 4.1. Higher terms are usually not needed. The rela-

tionship between fully normalised and non-normalised coefficients is

∗√
J∗n = Jn 2n + 1.
For comparison: in section 4.5 it was shown that in the expansion
asteluku of the same field into ellipsoidal harmonics, only the degree-zero and

6 They can also be calculated from equation 2-92 given in Heiskanen and Moritz (1967):

(︁ )︁n
3 e2
(︃ )︃
n+1 J2
J∗2n = (−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 4.9
105
degree-two coefficients are non-zero! This is one reason why these
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 + 2 ω⊕ X + Y 2 ,
)︁
r
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 + 2 ω⊕ X + Y 2 ,
)︁
+ 5
r r
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,
so this is also a wise thing to do in physical geodesy.

^ 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 = V ∗ + Φ.
The difference between them is the disturbing potential häiriö-
potentiaali

í Õ ! ¤. û
106 4 The normal gravity field

def
T = W − U = V − V ∗.

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

the gravity potential

W =V +Φ=Φ+
GM⊕
(︃ ∞ (︂ )︂ ∑︂
∑︂ a n
n )︃
+ r 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⊕ ∑︂ (︂ a )︂n ∑︂
(︃ ∞ n )︃
T =W−U=− r r Pnm (sin ϕ) (δJnm cos mλ + Knm sin mλ) ,
n=2 m=0
(4.11)
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), (4.12)
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

vertauspallo On the surface of the reference sphere of radius a:7

7

í Õ ! ¤. û
 Self-test questions
107
∑︂
∞
T (ϕ, λ) = T (ϕ, λ, a) = 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
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
values 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.

7 Earlier
on we also used for this reference radius (in spherical approximation) the
symbol R.

í Õ ! ¤. û
108 4 The normal gravity field

4. In figure 4.1 are drawn the equipotential surfaces of the normal

gravity field. It is seen that they are farther apart over the equator
than over the poles, because normal gravity over the equator is
less than over the poles.
How would the situation be for the normal gravitational field, that
is, without the centrifugal force? Please explain your reasoning.
5. Explain the idea of natural co-ordinates.
6. What was the relationhip between M. Le Blanc and C. F. Gauss?
Use Google.
7. Derive the Somigliana–Pizzetti equation 4.8 from equation 4.7.
What makes the equation valuable?
8. What are the defining parameters of the Geodetic Reference
System 1980?
9. Why does the spherical-harmonic expansion of the normal poten-
tial contain only a small number of terms and coefficients?
10. Why does the spherical-harmonic expansion of the normal poten-
järjestysluku tial not contain any terms of order m ̸= 0?
11. Why does the spherical-harmonic expansion of the normal poten-
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 latitude
types (geodetic φ, geocentric ϕ, and reduced β) at most, in minutes
of arc? Assume that the maximum happens at latitudes ±45◦ .

í Õ ! ¤. û
 Exercise 4 – 2: Centrifugal force
109
4. Compute for both a geodetic and a reduced latitude of 45◦ numer-
ical values of gravity for the case of the GRS80 reference ellipsoid.
By how much do they differ?

^ 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,
in other words, 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 and minutes. 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: häiriö-
potentiaali
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 ξ luotiviivan
and 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.

Here, (Φ, Λ) are astronomical latitude and longitude, that together make
up the direction of the local plumb line, and (φ, λ) are the geodetic luotiviiva
latitude and longitude that similarly make up the direction of the normal
gravity vector or “normal plumb line”.1 See figure 5.1. 1

– 111 –
112 5 Anomalous quantities of the gravity field

Plumb-line
deflections (ξ, η)
Topography

Geoid Geoid height N

Reference ellipsoid

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

The geoid height or geoid undulation is

def
N = h − H,

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

and h the height above the reference ellipsoid.
Deflections of the plumb line in Finland are a few seconds of arc ( ′′ ) in
magnitude. Geoid undulations range from 15 to 32 m (for comparison:
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, one may calculate for each gravity field
quantity the corresponding normal quantity. By subtracting the normal
quantity from the original one, the corresponding anomalous quantity is
obtained.
1 Thisassumes that Φ and Λ have been reduced to sea level, because of the curvature
of the plumb line, section 4.3.2, and that φ and λ have been computed on the reference
ellipsoid.

í Õ ! ¤. û
 Disturbing potential, geoid height, deflections of the plumb line 5.1
113

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

For heights above the reference ellipsoid there exists an expression

analogous to expression 4.5 for orthometric heights. Let U be the normal

í Õ ! ¤. û
114 5 Anomalous quantities of the gravity field

2 potential and γ normal gravity:2

w UP
1
dU = −γ dh =⇒ 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 g(z) − γ(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 variable U → W and changing it to a

length: dW = −g dz.
3 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 difference NP does not. Therefore, instead of point P, we use its
projection P ′ to mean sea level — the zero point of the height system
used. Then, the first term also vanishes: HP ′ = 0. So
w UP ′ )︁ T ′
1 1 (︁
NP ′ = − dU ≈ γ ′ WP ′ − UP ′ = γP ′ ,
WP ′ γ(U) P P

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

quantities are now at sea level. More compactly:

T
N = γ. (5.2)

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

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
 Gravity disturbances 5.2
115

−γ = − grad U
−g = − grad W
P
WP Geoid

UP
N Ellipsoid
UQ (= WP )
Q

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

tion 2-144).
Figure 5.3 depicts the situation still better. In this figure, the normal
∂
gravity vector γ = grad U has a length of γ = ∥γ∥ = − ∂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.

^ 5.2 Gravity disturbances

The difference between the true and normal gravity accelerations is
called the gravity disturbance, painovoima-
häiriö
∂W ∂U
(︂ )︂
def
δg = g − γ = ∥g∥ − ∥γ∥ ≈ − − ,
∂H ∂h
4 Ernst
Heinrich Bruns (1848–1919) was a German mathematician and mathematical
geodesist.

í Õ ! ¤. û
116 5 Anomalous quantities of the gravity field

in which the differentiation is done along the plumb line for W and —
slightly imprecisely — along the normal on the reference ellipsoid for
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 4.12, 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
(︂ )︂ ∑︂
∞
n + 1 R n+2
(︂ )︂
= r r T n (ϕ, λ) = r Tn (ϕ, λ), (5.4)
R
n=2 n=2

and at sea level (r = R):

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

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

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.
We can determine gravity disturbances from observations 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 in the following
section.

í Õ ! ¤. û
 Gravity anomalies 5.3
117
Topography Plumb line
Telluroid Q P (measurement point)
ζ
Mean sea
level (geoid)
HP
hQ
Ellipsoid
N

Figure 5.4. Reference ellipsoid, mean sea level (geoid), telluroid, and gravity
^ measurement.

^ 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
level (the geoid), determined, for example, through a national levelling
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

í Õ ! ¤. û
118 5 Anomalous quantities of the gravity field

According to this, we derive an expression for the gravity anomaly 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
⃓
∂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
∆g is given everywhere on the Earth’s surface.
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 = − − r T, (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 spectral representations 3.10

í Õ ! ¤. û
Gravity anomalies 5.3
119
(but for T ) and 5.4 for δg:

∑︂
∞ (︂
n+1 2
)︂ (︂ )︂n+1
R
∆g(ϕ, λ, r) = r − r r Tn (ϕ, λ) =
n=2
∑︂
∞
n − 1 R n+1
(︂ )︂ ∑︂
∞
n − 1 R n+2
(︂ )︂
= r r Tn (ϕ, λ) = r Tn (ϕ, λ) =
R
n=2 n=2
∑︂
∞ (︂ )︂n+2
R
= r ∆gn (ϕ, λ), (5.7)
n=2

with the notation

def n−1
∆gn (ϕ, λ) = T (ϕ, λ). (5.8)
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
of mass of the Earth, but if the origin is not located there, at least gravity
anomalies do not change.
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 5
the total mass and volume of the reference ellipsoid. The assumption is
largely justified because GM⊕ can be, and has been, determined very
precisely by satellites, and modern models for the normal potential, like
GRS80, are based on these determinations.6 6

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

í Õ ! ¤. û
120 5 Anomalous quantities of the gravity field

^ 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

^ 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
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 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

like EGM2008 give a smaller value of 6 378 136.3 m as the mean location of global mean
sea level. This is something to be well aware of when using the model in production
work. Uncertainty continues to be of decimetre order.

í Õ ! ¤. û
The boundary-value problem of physical geodesy 5.5
121
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
point’s location in space. This location depends additionally on the
location of sea level, the geoid, in space; more precisely its height above
or below the reference ellipsoid.
This is how we arrive at the third boundary-value problem.7 The 7
boundary-value problem of physical geodesy is to determine the potential V
outside a body if given on its surface is the linear combination

∂V
c1 V + c2 ,
∂n
with c1 and c2 suitable coefficients. The variable n represents here
differentiation 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 = − +γ T. (5.5)
∂H ∂H
∂
We see that c1 = −1 and c2 = γ−1 ∂H γ. This equation, the definition
5.5 of gravity anomalies, is known as the fundamental equation of physical
geodesy.
Again in spherical approximation, inverting equation 5.8 yields

R
Tn (ϕ, λ) = ∆gn (ϕ, λ).
n−1
Remember that the functions ∆gn (ϕ, λ) are computable using the degree asteosuus-
constituent equation 3.9 when ∆g(ϕ, λ) is known all over the Earth. yhtälö

This is how we also obtain the solution of this boundary-value

problem in spectral representation — which is thus valid in the whole

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
122 5 Anomalous quantities of the gravity field

exterior space:

∑︂
∞ (︂ )︂n+1
R ∑︂
∞ (︂ )︂n+1
R R
T (ϕ, λ, r) = r Tn (ϕ, λ) = r ∆gn (ϕ, λ) =
n−1
n=2 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
8 5.9 is known as the Stokes8 equation:
x
R
S(ψ, r, 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

In section 8.1 we will give closed form 8.3 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
mapping adds to every point P a corresponding point Q, the geodetic
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

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” 5.6
123
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.
This 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-
sion to “reference-surface thinking”, which only really works within the vertauspinta
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).

í Õ ! ¤. û
124 5 Anomalous quantities of the gravity field

^ 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 formal
definition, but 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
9 gravity is constant.9

Thus, free-air anomalies can be calculated in a simpler way. The

gravity formula of the normal field 4.10 gives for latitude 60◦ :

γ = 981 917.838 − 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

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

9 For the greatest precision, one should consider that the latitude Φ may also not
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
 Self-test questions
125
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 4.10 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.
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
1. How do deflections of the plumb line and geoid heights relate to
each other?
2. What is the fundamental equation of physical geodesy in spherical
approximation?
3. In what way is a gravity disturbance different from a gravity
anomaly?
4. What units are used for measuring gravity anomalies and gravity

í Õ ! ¤. û
126 5 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.

gradients? How are they related to the SI system?

5. How do the geoid height and the disturbing potential relate to
each other?
6. Explain telluroid mapping and height anomalies.

í Õ ! ¤. û
 Exercise 5 – 1: The spectrum of gravity anomalies
127
^ Exercise 5 – 1: The spectrum of gravity anomalies
Use equation 5.8. If we assume that the quadratic mean of the degree
constituents ∆gn of gravity anomalies,
√︃ x
def 1
∥∆gn ∥σ = ∆g2 (ϕ, λ) dσ,
4π σ n

does not depend on the chosen degree number n, how then does the
similarly defined ∥Tn ∥σ 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,
there is a systematic error of one arc second, what error does this cause
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.

^ Exercise 5 – 3: Gravity anomaly, geoid height

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 .
1. Using the fundamental equation of physical geodesy 5.6:

∂T 2
∆g = − − r T,
∂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

í Õ ! ¤. û
128 5 Anomalous quantities of the gravity field

2. Assume that the point is close to sea level. Using the Bruns
equation
T
N = 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 — or a linear combination — observed on the boundary surface
only. A requirement is, that there are no masses outside the boundary
surface.
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

– 129 –
130 6 Geophysical reductions

into account when the boundary-value problem is formulated

and solved.
◦ 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. Only then
will the Laplace equation 1.13 apply in the exterior space of the
Earth, as required by the boundary-value problem of physical
geodesy, see section 5.5.
Then, the geopotential or geoid solution obtained is not any more
that of the original mass distribution, 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
epäsuora is determined and reversed. This influence is called the “indirect
vaikutus effect”.
poistamis- In the literature, this method is also referred to as a remove–restore
entistämis- method.
menetelmä
◦ 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
reference ellipsoid. Nowadays this is possible using GNSS, for
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

í Õ ! ¤. û
 Bouguer anomalies 6.2
131
z
P
dz
β ds ⏟ ⏞⏞ ⏟
ℓ
dβ
cos β
H ℓ

d
s dV x

^ Figure 6.1. The attraction of a Bouguer plate.

way of removing the effect of the topography is the so-called Bouguer1 1

reduction.

^ 6.2.1 Calculation of the Bouguer reduction

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
H of point P above the lower surface of the plate. 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). Transform this to the co-ordinates

(β, z, α). Forget about α and study the surface element (figure 6.1, top

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

í Õ ! ¤. û
132 6 Geophysical reductions

right)
ℓ
ds dz = dβ dz,
cos β
in which the determinant of Jacobi needed, ℓ cos β , is seen.
/︁

Carry out the integration:

y w 2π w d w ∞
def cos β cos β
a = ∥a∥ = G 2
ρ dV = Gρ · ds dz · s dα =
ℓ 0 0 0 ℓ2
w 2π w d w π/2
cos β ℓ
= Gρ · dβ dz · s dα =
0 0 0 ℓ2 cos β
w d w 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

cos(β0 (z))
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

í Õ ! ¤. û
 Bouguer anomalies 6.2
133
Evaluation point P

II

I Bouguer plate
I
Topography d=H

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

We obtain for the whole integral

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

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, 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-
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

í Õ ! ¤. û
134 6 Geophysical reductions

equation 5.11 into this, we obtain

(︁ )︁
∆gB = gP − γ0 (φ) + 0.3084 − 0.1119 H = gP − γ0 (φ) + 0.1965 H.
(6.3)

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 T C (volumes I and II
in figure 6.2). We shall return to its calculation later.

^ 6.2.2 Properties
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 gravi-
metric material is geographically sparse. This requires that topographic
heights are known to a better spatial density.

^ 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
nothing there.
◦ The attraction of volumes II, where there actually is something, is
ignored.

í Õ ! ¤. û
 Terrain effect and terrain correction 6.3
135
Topography

Free-air anomaly

Bouguer anomaly

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

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 — acts 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 T C — 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. 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, and in the mountains one needs even 50 m. The systematic
nature of the terrain correction makes a too-sparse terrain model cause,

í Õ ! ¤. û
136 6 Geophysical reductions

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 126, there is a strong negative bias of Bouguer
anomalies — although part of this is due to post-glacial isostatic
imbalance 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.

í Õ ! ¤. û
 Terrain effect and terrain correction 6.3
137
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.

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
(︂ (︁
′ ′
= 12 Gρ dx ′ dy ′ ,
)︁
T C(x, y) H x , y − H(x, y)
−D −D ℓ3
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
equation does not consider.

í Õ ! ¤. û
138 6 Geophysical reductions

Given: gravity g
on terrain Bouguer-
Terrain
plate
correction
correction

Free-air
reduction Subtraction of
to sea normal gravity
level at sea 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.

The values of the terrain correction T C vary from fractions of a milligal

(Southern Finland) to hundreds of milligals (high mountain ranges).
käsivarsi 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-
ilmareduktio tion and free-air reduction.

^ 6.3.1 Example: applying the terrain correction in a special

case
Given is the special terrain shape rendered in quasi-3-D in figure 6.7.
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

í Õ ! ¤. û
 Terrain effect and terrain correction 6.3
139
P

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.

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,
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
TC = 2
· 2πGρ · H =
1
= 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,

TC = 5.595 mGal,

í Õ ! ¤. û
140 6 Geophysical reductions

and, because a semi-plate is removed that is above the level of

point Q, the algebraic sign of T C 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

4. Bouguer to free air:

∆gB (Q) 22.000 mGal
Bouguer plate addition, 200 m +22.380 mGal
T C “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

í Õ ! ¤. û
Spherical Bouguer anomalies 6.4
141
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 2
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,
also in low-lying areas, where the planar Bouguer reduction is
typically small.
4. As the terrain correction is now calculated over the whole globe in
spherical geometry — though only for the topography. Therefore
the rule that all its contributions are positive no longer applies:
Abrehdary et al. (2016) report that in places near mountain ranges
below the local horizon, the spherical terrain correction can be as
negative as −200 mGal.
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

2 One can also do so, and often does, in connection with the Bouguer-plate correction.

í Õ ! ¤. û
142 6 Geophysical reductions

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
puskutraktori rock. If we try this bulldozing experiment, we will end up 1381 m
short or 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 spherical and planar Bouguer reductions, on
average over the globe, will be half of this, ≈ 155 mGal. As the
global mean free-air anomaly is zero, the global mean spherical
Bouguer anomaly will be 309 mGal, the bulk of the positivity
being found over the deep ocean.

^ 6.5 Helmert condensation

3 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

3 Friedrich
Robert Helmert (1843–1917) was a German geodesist known for his work
on mathematical and statistical geodesy.

í Õ ! ¤. û
 Helmert condensation 6.5
143
Equipotential surface

Topography

g′ g

Condensation layer

Figure 6.8. Helmert condensation and the changes it causes in the gravity
^ field.

For a spherical Earth, the corresponding integral is

w R+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 exterior and the interior potential as functions
of the “degree constituents” of the various powers of the topography
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

í Õ ! ¤. û
144 6 Geophysical reductions

^ Figure 6.9. Friedrich Robert Helmert. Humboldt University Berlin (2017).

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
luotiviivan mountains, specifically the deflections of the plumb line, could only be
poikkeama explained by assuming that under every mountain range there was also
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
 Isostasy 6.6
145
Plumb-line deflections

Geoid
Mountain

Earth’s crust

“Root”

Earth’s mantle

^ Figure 6.10. Isostasy and the deflection of plumb lines towards the mountain.

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 classical isostatic hypothesis is the Pratt–Hayford hypothesis.
This was proposed by J. H. Pratt4 in the middle of the 19th century 4
(Pratt, 1855, 1859, 1864), and J. F. Hayford5 developed the mathematical 5
tools needed for computation. According to this hypothesis, the matter
density of the “root” under a mountain would vary with the height
of the mountain, so that under the highest mountains would be the

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

and the figure of the Earth.

í Õ ! ¤. û
146 6 Geophysical reductions

Mountains

Sea
Compen-
sation Earth’s crust
depth

Compen-
sation
level
Mantle

^ Figure 6.11. Pratt–Hayford isostatic hypothesis.

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.
6 Another classical isostatic hypothesis is due to G. B. Airy.6 Because V.
7 A. Heiskanen7 used it extensively and developed its mathematical 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 GeorgeBiddell Airy PRS (1801–1892) was an English mathematician and astronomer,

“Astronomer Royal” 1835–1881.
7 Veikko Aleksanteri Heiskanen (1895–1971), “the great Heiskanen” (Hermans, 2007)
was a 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
 Isostasy 6.6
147
Mountains
Sea ρw
Earth’s crust

Compensation depth t0

ρc
Anti-root Compen-
sation
level
Mountain root

ρm

Mantle

^ Figure 6.12. Airy–Heiskanen isostatic hypothesis.

^ 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

t r
t0 t
r
Anti-root

Root

^ Figure 6.13. Quantities in isostatic compensation.

í Õ ! ¤. û
148 6 Geophysical reductions

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

8 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”.
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, at least from the viewpoint of isostatic
equilibrium, 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”, for short compensation depth, 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

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
pressure of the column of displaced mantle material.

í Õ ! ¤. û
 Isostasy 6.6
149
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.

^ 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?

í Õ ! ¤. û
 150 6 Geophysical reductions

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.

^ 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 variation of the
oceanic water masses. 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 the Earth’s crust.
More in chapter 12.

í Õ ! ¤. û
 Isostasy 6.6
151
Mid-oceanic ridge
Plate motion
Deep-sea trench Conrad
discontinuity
Sea X
Earth’s crust
Mohorovičić
Lithosphere discontinuity

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.

^ 6.6.5 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.
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 greatest, 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

í Õ ! ¤. û
152 6 Geophysical reductions

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:
◦ 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 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.
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

í Õ ! ¤. û
 The “isostatic geoid” 6.8
153
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 similar
to those in other reductions. We will discuss them later: numerical
integration in the space domain — grid integration, spherical-cap
integration, least-squares collocation (LSC), finite elements, etc. — or in pienimmän
the spectral domain, for example FFT and “Fast Collocation”. neliösumman
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
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.

^ 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
order to formulate a boundary-value problem on the geoid.
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
geoid. In other words, the indirect effect (the “restore” step) is of this
order. Under the oceans, similarly the isostatic co-geoid is somewhat

9 FelixAndries Vening Meinesz (1887–1966) was a Dutch geophysicist, geodesist and

gravimetrist. He wrote with V. A. Heiskanen the textbook The Earth and its Gravity Field
(1958).

í Õ ! ¤. û
154 6 Geophysical reductions

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 126.

above the geoid.

As one of the requirements for geoid determination methods is a

í Õ ! ¤. û
The “isostatic geoid” 6.8
155
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 10
(1967) call on their page 152 the indirect effect ”moderate”.
However, isostatic methods are very suitable for elucidating the
interior structure of the Earth, because both the topography and the
imprint 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
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

10 Of
course Bouguer reduction is even worse! The indirect effect can be hundreds of
metres.

í Õ ! ¤. û
156 6 Geophysical reductions

layer at level H = 0, and the second, density

)︂2
R
(︂
−κ ,
R−D
at compensation depth D. This choice preserves the total mass of the
Earth. The situation is depicted — in flat-Earth approximation — in
figure 6.16.
In the following we use the “generating function” equation 8.7,

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 σ 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):

Vcomp =
x (︃ (︂ )︂2 )︃ ∑︂
∞ (︂ )︂n+1
R R−D
= G (R − D) −κ Pn (cos ψ) dσ =
σ R−D R
n=0
x ∑︂
∞ (︂
R−D n
)︂
= −GR κ Pn (cos ψ) dσ,
σ R
n=0

from which the combined effect of the reduction is (n = 0 drops out):

(︁ )︁
δViso = − Vtop + Vcomp =
x ∑︂ ∞ (︃ )︂ )︃
R−D n
(︂
= −GR κ 1− Pn (cos ψ) σ. (6.7)
σ R
n=1

í Õ ! ¤. û
 The “isostatic geoid” 6.8
157
Sea level

Compensation depth

^ Figure 6.16. Isostatic reduction as a pair of surface density layers.

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 11

use again the degree constituent equation, Heiskanen and Moritz (1967) asteosuus-
equation 1-71, or our equation 3.9, in the following form: yhtälö
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
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

11 This works on dry land and on the ocean. Lakes, glaciers and areas like the Dead Sea
are more complicated.

í Õ ! ¤. û
158 6 Geophysical reductions

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 ′ = δViso .
σ R
n=1

It follows that

∑︂
∞
4πGR
(︃ (︂
R−D n
)︂ )︃
δViso =− 1− κn (ϕ, λ) =
2n + 1 R
n=1
∑︂
∞
2
(︃ (︂
R−D n
)︂ )︃
=− 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 .
12 def /︁
Let us first look at the contribution from12 1 < n ⩽ N = R D . Then,
as
R−D n nD
(︂ )︂
≈1− ,
R R
the following approximation holds:

∑︂
N
2nD ∑︂
N
δViso ≈− (A ) ≈ − D (AB )n ≈ −DA
˜︁ B ,
2n + 1 B n
n=1 n=1

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
The “isostatic geoid” 6.8
159
and
δViso DA˜︁ B DAB
δNiso = γ ≈ − γ ≈− γ . (6.8)
This is the indirect effect of isostatic reduction.
Let us substitute realistic values. Let the depth of the Mohorovičić13 13
discontinuity be on average ∼ 20 km.14 14

On land H ≈ 0.8 km, the Earth’s mean topographic height, and we

obtain δNiso, land ≈ −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, sea ≈
+5.1 m.

In other words, this effect can be sizeable.

Note, however, that the above calculation used the equivalent Bouguer-
plate attraction
∑︂
N
AB = 2πGκn ,
n=0
whereas equation 6.8 contains
∑︂
N
2n ∑︂
N
A
˜︁ B = 2πGκn ≈ 2πGκn ,
2n + 1
n=1 n=1

which lacks the zeroth-degree constituent κ0 . in other words, the global

average of A
˜︁ B , and thus of δNiso , over continents and oceans would have
to vanish, because of the assumption made that isostatic reduction does
not change the total mass of the Earth. The calculated value is, however,

δNiso = 0.29 · δNiso, land + 0.71 · δNiso, sea = 3.1 m.

13 AndrijaMohorovičić (1857–1936) was a Croatian meteorologist and a pioneer of

modern seismology.
14 Underthe continents, the depth is 35 km, under the oceans 7 km below the sea floor
(Encyclopaedia Britannica, Moho).

í Õ ! ¤. û
160 6 Geophysical reductions

With this value as a correction, we obtain

δNiso, land ≈ −1.8 m − 3.1 m = −4.9 m,

δNiso, sea ≈ +5.1 m − 3.1 m = +2.0 m.

These are values for typical, extended continental or ocean areas, and
only indicative. Precise calculation needs to be numerical.
Equation 6.8 is, through the Bouguer-plate attraction AB , linear in the
height H. This means that every added kilometre of topography causes
about −2.2 m in the quantity δNiso, land , and every added kilometre of
bathymetry similarly +1.4 m in the quantity δNiso, sea . 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.

^ 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.

í Õ ! ¤. û
 Exercise 6 – 1: Gravity anomaly
161
P

Q
600 m
300 m

Q′ Sea level

^ Figure 6.17. Terrain shape.

^ 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.
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 .

^ Exercise 6 – 3: Terrain correction and Bouguer

reduction
Given the terrain shape, figure 6.17.

í Õ ! ¤. û
162 6 Geophysical reductions

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 equation for the attraction of a Bouguer plate. Here
is a half Bouguer plate, with only half the attraction of a full one.
2. Compute the terrain correction at point Q. What is the algebraic
sign?
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 that at point Q 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
vertaustaso 700 kg/m3 . Let the reference level for the interface corresponding to zero
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. The volcano Mauna Kea, Hawaii, 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.

– 163 –
164 7 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 7.2
165
system is called N60. However, the precise realisation is a special pillar
in the garden of the Helsinki astronomical observatory in Kaivopuisto.1 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
N2000 heights are, at the decimetric precision level, heights over the
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 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 The district 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. 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
166 7 Vertical reference systems

Figure 7.2. Height reference benchmark 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 benchmark of the precise levelling of Finland,
^ 30.4652 m above zero).

í Õ ! ¤. û
 Orthometric heights 7.2
167
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
∑︁3 ′
correct height above the geoid: i=1 ∆Hi computed along the
plumb line.
The equipotential or level surfaces of the geopotential are not 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
168 7 Vertical reference systems

This is a very physical definition, however not a very operational

one, because one (generally) do not get to measure along a plumb line
inside the Earth, and the geoid is not even 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 equation
C
H = P,
g
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 equation 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 C
H∗ = ,
γ0H

í Õ ! ¤. û
 Normal heights 7.3
169
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 T
ζ= , (7.1)
γHh
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
170 7 Vertical reference systems

Figure 7.5. Mikhail Sergeevich Molodensky (1909–1991), source obscure.

More photographs and background information in Brovar et al.
^ (2000).

point and h its height from the reference ellipsoid. The parameter z is
the distance from the reference ellipsoid reckoned along the ellipsoidal
häiriö- normal. T is the disturbing potential at the point.
potentiaali
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
plumb line is very close to a linear function of place, one could define a

í Õ ! ¤. û
Normal heights 7.3
171
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.
The geometric height h from the reference ellipsoid may be connected
to the potential U of the normal gravity field through 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 3

By defining
wh
def 1
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
as the corresponding new geoid height type. It has however the aesthetic
flaw that we divide here by the average normal gravity computed

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
172 7 Vertical reference systems

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 − N ≈ γ0h 1 + , (7.3)
2 dz R
(︃ )︃
1 + dγ H+
γHh ≈ γ0h + 2 H ≈ γ0h 1 − . (7.4)
dz R
d d
R is the Earth’s radius in spherical approximation: dz γ ≈ dr γ ≈
− 2γ R .
/︁
+/︁ +/︁
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+ ,
(︃ )︃
def T 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)
The quantity γ0H , and thus also normal height H∗ , can be, unlike γ0h ,
computed using only information obtained by (spirit or trigonometric)

í Õ ! ¤. û
 Normal heights 7.3
173
+ +
− 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).

levelling, without having to know the height h above the reference

ellipsoid, which would again require knowledge of the local geoid.
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.
Figure 7.6 attempts to visualise the derivation.

í Õ ! ¤. û
174 7 Vertical reference systems

Topography
ζ
Telluroid
H∗ H h

H∗ N ζ
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 an area
where N > 0. The separation between geoid and quasi-geoid is
^ exaggerated.

^ 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
H∗ above the reference ellipsoid (and thus a distance ζ below or
−ζ above the topography), is called the telluroid. It is a mapping

í Õ ! ¤. û
 Difference between geoid height and height anomaly 7.4
175
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
corresponding point P on the topography. 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
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.
Orthometric heights — for example Helmert heights — on the other
hand are always used together with geoid heights N. For computing

í Õ ! ¤. û
176 7 Vertical reference systems

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 dT (z)
FA
TFA (H) − TFA (0) = dz ≈ −∆gFA H, (7.7)
0 dz
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

/︁
4 (”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 the 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 exterior
field, meaning that the Bouguer-plate attraction acting at P
must also be continued downwards, in other words, taken
fully into account.

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 7.4
177
Because the surface mass density of the plate is Hρ, its assumed
attraction is everywhere on the plumb line of point P:

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
178 7 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,
in which h is the height above the reference ellipsoid, and H is the
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 — 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
in which the mean gravity along the plumb line is
w
1 H
g= g(z) dz.
H 0

í Õ ! ¤. û
Calculating orthometric heights precisely 7.6
179
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
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 painovoiman
0.2238 mGal/m greater (twice the standard-density 2670 kg/m3 Bouguer- ilmagradientti
plate effect): the end result is the total inside-rock gravity gradient,
−0.0846 mGal/m .
This is called the Prey5 reduction. As the end result we obtain the 5
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
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 AdalbertPrey (1873–1949) was an Austrian astronomer and geodesist and an author

of textbooks.

í Õ ! ¤. û
180 7 Vertical reference systems

◦ 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 should use a sufficient number of support points
along the plumb line for computing this effect.
◦ The assumption that the free-air vertical gravity gradient is every-
where the same, −0.3084 mGal/m . The real gradient can easily vary
by ±10 % around this value.
◦ The assumption that the rock density 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
6 by using Niethammer’s6 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 determination. 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
free-air gravity gradient instead of a standard value. To compute the
gradient, the integral equation presented in section 8.6 may be used.
The precise calculation of orthometric heights is thus laborious: just
as laborious as the precise determination of the geoid, and for the
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).

6 Theodor Niethammer (1876–1947) was a Swiss astronomer and geodesist who created

the gravimetric base network of Switzerland.

í Õ ! ¤. û
 Calculating normal heights precisely 7.7
181
^ 7.7 Calculating normal heights precisely
For this we use equation 7.6:
C W − W0
H∗ = =− , (7.6)
γ γ
where the average value of normal gravity along the plumb line is
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 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, in other words the
reference ellipsoid.

^ 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 .

í Õ ! ¤. û
182 7 Vertical reference systems

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):

(1) 5000 m2/s2

H = =
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.

3. The Bouguer anomaly is (equation 6.2):

∆gB = ∆gFA − 0.1119 mGal/m · H = −49.951 mGal.

4. The first attempt is again

C
H∗(0) = γ = 509.087 m.
0

í Õ ! ¤. û
 Orthometric and normal corrections 7.9
183
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−1
def
∑︂
B
∆Hi,i+1 = ∆H,
i=A A
levelling line

after which the non-exactness of this method is accounted for by

applying the “orthometric correction” (OC):

∑︂
B
HB = HA + ∆H + OCAB .
A

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, it holds that CB − CA −

í Õ ! ¤. û
184 7 Vertical reference systems

∑︁B
∆C = 0 because of the conservative nature of the geopotential.
A
Dividing by a constant γ0 yields

CB CA ∑︂ ∆C
B

γ0 − γ0 − γ0 = 0.
A

On the other hand

∑︂
B
CB CA ∑︂ ∆C
B
OCAB = HB − HA − ∆H = − − g ,
gB gA
A A

with gA and gB average gravity values along the plumb lines of A and
B and g gravity along the levelling line. This expression compares
∑︁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
(︃
CB CB
)︃ (︃
CA CA
)︃ ∑︂
B (︂
∆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 = γ0 ∆H + γ0 HA − γ0 HB , (7.15)
A

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.

í Õ ! ¤. û
Orthometric and normal corrections 7.9
185
Similarly we may also calculate the normal correction (NC) in calculating
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 =
γ ∆H + γ HA − γ H∗B . (7.17)
0 0 0
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.
Both the orthometric correction 7.15 and the normal correction 7.17
are calculated one benchmark interval at a time: one must know, in
addition to the levelled height difference ∆H, local gravity g along
the levelling line. Furthermore one must know g(H) or γ(0) at both
end points in 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 also
if one wants 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.

í Õ ! ¤. û
186 7 Vertical reference systems

^ 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
7 the Schwarzschild7 metric for a spherically symmetric field:

metriikka
c2 dτ2 =
2GM 2 2 2GM −1 2
(︂ )︂ (︂ )︂
dr − r2 dϕ2 + cos2 ϕ dλ2 =
(︁ )︁
= 1− 2 c dt − 1 − 2
c r c r
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
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: extreme cooling produces a so-called
Bose–Einstein condensate of atoms, which is trapped inside an optical
valohila lattice formed by six laser beams, an electromagnetic pattern of standing
waves. The clock oscillation is on a different frequency. A Bose–Einstein

7 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 7.10
187
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.

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.
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.

í Õ ! ¤. û
188 7 Vertical reference systems

^ 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.

^ 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.

í Õ ! ¤. û
 Exercise 7 – 3: Difference between orthometric and normal height
189
^ 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
Assume 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 häiriö-

field T = T (ϕ, λ), and ∆gn = ∆gn (ϕ, λ) those of the gravity anomaly potentiaali
field ∆g = ∆g(ϕ, λ). 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 3.9, one obtains the asteosuus-
integral equation yhtälö

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

– 191 –
192 8 The Stokes equation and other integral equations

Mass excess

Mass
deficit
−N

^ Figure 8.1. The principle of gravimetric geoid determination.

x ∑︂
∞
(︄ )︄
R 2n + 1
= P (cos ψ) ∆g(ϕ ′ , λ ′ ) dσ ′ =
4π σ n−1 n
n=2
x
R
= S(ψ) ∆g(ϕ ′ , λ ′ ) dσ ′ , (8.1)
4π σ
in which
∑︂
∞
2n + 1
S(ψ) = P (cos ψ),
n−1 n
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,

í Õ ! ¤. û
 The Stokes equation and the Stokes integral kernel 8.1
193

N(ϕ, λ)
Evaluation S(ψ)
point

∆g ϕ ′ , λ ′ dσ ′
(︁ )︁

Moving data or
ψ integration point
Earth’s centre

^ Figure 8.2. The geometry of integrating the Stokes equation.

of the disturbing potential T , and from that, the geoid height N using
Bruns equation 5.2, N = T γ . The result is
/︁

T (ϕ, λ) x
R
S(ψ) ∆g ϕ ′ , λ ′ dσ ′ ,
(︁ )︁
N(ϕ, λ) = γ = (8.2)
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 8.2 is the classical Stokes integral 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 8.7, Heiskanen and
/︁

Moritz’s (1967) equation 1-81. Of course 1 ℓ is also the kernel function

/︁

of an integral equation, equation 1.28, 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

í Õ ! ¤. û
194 8 The Stokes equation and other integral equations

25
S(ψ)
20 1
sin 12 ψ

15 − 6 sin 12 ψ + 1 − 5 cos ψ

− 3 cos ψ ln sin 12 ψ + sin2 21 ψ

(︁ )︁
10
S(ψ)−→

0
30◦ 60◦ 90◦ 180◦
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 8.3
^ with their different asymptotic behaviours.

equation 5.10:
∑︂
∞ (︂ )︂n+1
R 2n + 1
S(ψ, r, R) = P (cos ψ). (5.10)
r n−1 n
n=2

The Stokes kernel function on the Earth’s surface is depicted in figure

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
S(ψ) = 1
− 6 sin 21 ψ + 1 − 5 cos ψ −
sin 2 ψ (︂ )︂
1 2 1
− 3 cos ψ ln sin 2 ψ + sin 2 ψ . (8.3)

This closed expression helps us to understand better how the function

behaves close to the origin ψ = 0: the first term, 1 sin 21 ψ , goes to
/︁

í Õ ! ¤. û
 Example: The Stokes equation in polar co-ordinates 8.2
195
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 value 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.

^ 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”
computation 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) = V ext (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 1 ∂γ
∆g(α, r) = − + T=
∂r γ ∂r

í Õ ! ¤. û
196 8 The Stokes equation and other integral equations

∑︂
∞
k −k
= r r (ak cos kα + bk sin kα) −
k=1

1 ∑︂ −k
∞
−r r (ak cos kα + bk sin kα) =
k=1
∑︂
∞
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). (8.4)
R k
kantafunktio According to Fourier theory, the basis functions cos kα and sin kα are
orthonormal 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 we may expand

∑︂
∞
∆g(α, R) = ∆gk (α)
k=2

í Õ ! ¤. û
Example: The Stokes equation in polar co-ordinates 8.2
197
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.

This yields the following Fourier coefficients:

{︄ }︄ {︄ }︄
Ak a k
= (k − 1) R−(k+1) , 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

í Õ ! ¤. û
198 8 The Stokes equation and other integral equations

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

Define the Stokes kernel for this two-dimensional situation:

T (α, R) w
R 2π (︁ ′ )︁ (︁ ′
dα ′ ,
)︁
N(α) = γ = πγ 0 ∆g α , R S α − α
∑︂
∞
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 8.4 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, into a function of the spatial wave number
(number of waves around the circle) k.
This formulation has the merit of being able to use any standard
FFT software library offering compatible versions of both the forward
{︁ }︁ {︁ }︁
Fourier transform F · and the inverse transform F−1 · .
For more about FFT, see appendix C.
Figure 8.5 shows a simulation result in which a randomly generated
set of gravity anomalies on the circle r = R has been used to calculate

í Õ ! ¤. û
 Example: The Stokes equation in polar co-ordinates 8.2
199
^ Tableau 8.1. Stokes equation in two dimensions, octave code.

% Stokes equation simulator in two dimensions

R = 6378137;
g = 9.8;
ak(1:180) = 0.0;
bk(1:180) = 0.0;
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
dg(1:360) = 0.0;
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);
% Without degree one
dg(i) = dg(i) + ak(k)*cos(k*i*pi/180) + bk(k)*sin(k*i*pi/180);
end
end
hold on
plot(1:360, dg, ’b’) plot(1:360, 0.00001*T/g, ’m’)
print -dpdf stokes2D-out.pdf

í Õ ! ¤. û
200 8 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.

geoid undulations on the same circle. Both curves display fairly realistic
statistical behaviour. The code used is given in tableau 8.1.

200
−→ ∆g (mGal)
−→ N (m)

100

−100

−200 −→ 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.

í Õ ! ¤. û
 Plumb-line deflections and Vening Meinesz equations 8.3
201
^ 8.3 Plumb-line deflections and Vening Meinesz
equations
By differentiating the Stokes equation with respect to place, we obtain
integral equations for the components of the deflection of the plumb luotiviivan
line (Heiskanen and Moritz, 1967, equation 2-210’): poikkeama

{︄ }︄ {︄ }︄
ξ(ϕ, λ) x dS(ψ) cos α
1
= ∆g(ϕ ′ , λ ′ ) dσ ′ =
η(ϕ, λ) 4πγ σ dψ sin α
{︄ }︄
x dS(ψ) cos α
1
= ∆g(ϕ ′ , λ ′ ) sin ψ dα dψ, (8.5)
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
dσ ′ = sin ψ dα dψ, in which sin ψ is Jacobi’s determinant of the (ψ, α)
co-ordinates.
These equations were derived for the first time by the Dutch geo-
physicist F. A. 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 transforma-
tions 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, in other words the angular distance as seen from
the origin, is ψ. The distance between points P and Q is ℓ.

í Õ ! ¤. û
202 8 The Stokes equation and other integral equations

z P

ℓ r

Q ψ y
R
O

Figure 8.6. The geometry of the generating function of the Legendre polyno-
^ mials.

def def
With the definitions R = ∥R∥ and r = ∥r∥, we may write (cosine
rule):
ℓ2 = r2 + R2 − 2rR cos ψ. (8.6)
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), (8.7)
ℓ r2 + R2 − 2Rr cos ψ R
n=0

in which r and R are the distances of points P and Q from the origin O,
usually the centre of the Earth. The function 1 ℓ is called the generating
/︁

function of the Legendre polynomials.

Differentiating equation 8.7 with respect to r yields
1 ∑︂ n + 1 R n+1
∞
r − R cos ψ
(︂ )︂
− =− r r Pn (cos ψ).
ℓ3 R
n=0

Multiply this by 2r:

1 ∑︂
∞ (︂ )︂n+1
2r2 − 2rR cos ψ R
− 3
= − (2n + 2) r Pn (cos ψ).
ℓ R
n=0

í Õ ! ¤. û
The Poisson integral equation 8.4
203
Now add together this equation and equation 8.7:

1 ∑︂
∞ (︂ )︂n+1
−2r2 + 2rR cos ψ + ℓ2 R
3
= − (2n + 1) r Pn (cos ψ).
ℓ R
n=0

Substitute ℓ2 from equation 8.6:

−2r2 + 2rR cos ψ + ℓ2 R2 − r2

= ,
ℓ3 ℓ3
and the result is, after multiplying with −R,

∑︂
∞
(︁ )︁
R r2 − R2 (︂ )︂n+1
R
3
= (2n + 1) r Pn (cos ψ). (8.8)
ℓ
n=0

Applying degree constituent equation 3.9 to the harmonic potential

field V on the spherical Earth’s surface, radius R:
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 3.10:

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

we obtain

V(ϕ, λ, r) =
1 ∑︂ R n+1
∞ (︂ )︂ x (︁
V ϕ ′ , λ ′ , R Pn (cos ψ) dσ ′ =
)︁
= r (2n + 1)
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
4π σ ℓ3
by replacing the expression in square brackets by equation 8.8.

í Õ ! ¤. û
204 8 The Stokes equation and other integral equations

Thus we have obtained the Poisson integral for computing a harmonic

field V from values given on the Earth’s surface:
x R (︁r2 − R2 )︁
1
VP = VQ dσQ , (8.9)
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).
Still a third way to write the same equation, useful when the harmonic
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. . . .
reuna-arvo- Equation 8.9 solves for this special case the so-called Dirichlet boundary-
tehtävä 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 8.9, applies for an arbitrary
harmonic field V, meaning any field for which ∆V = 0. The equation ap-
plies conveniently to the expression r∆g, the gravity anomaly multiplied
by the geocentric radius, which is also a harmonic field. This is how

í Õ ! ¤. û
Gravity anomalies in the exterior space 8.5
205
we can express the gravity anomaly in the external space ∆g(ϕ, λ, r) as
a function of gravity anomalies ∆g(ϕ ′ , λ ′ , R) on a reference sphere of vertauspallo
radius R. The function r∆g is harmonic, because according to equation
5.7
1 ∑︂
∞ (︂ )︂n+1
R
∆g = r (n − 1) r Tn ,
n=2
so
∑︂
∞ (︂ )︂n+1
R ∑︂
∞ (︂ )︂n+1
R
r∆g = r (n − 1) Tn = r Tn′ ,
n=2 n=2

in which Tn′ (ϕ, λ) = (n − 1) Tn (ϕ, λ) is a perfectly legal surface spherical pinta-

harmonic just like Tn (ϕ, λ) itself: the dependence on the radius r, the pallofunktio
(︁ /︁ )︁n+1
factor R r , is the same as for the (harmonic) potential. So, Poisson’s
integral equation 8.9 applies to function r∆g:
x R (︁r2 − R2 )︁ [︂
[︁ ]︁ 1 (︁ ′ ′ )︁]︂ ′
r ∆g(ϕ, λ, r) = R∆g ϕ , λ , R dσ
4π σ ℓ3
or
x (︁ 2 2
)︁
1 RR r −R
∆g ϕ ′ , λ ′ , R dσ ′ .
(︁ )︁
∆g(ϕ, λ, r) = r (8.10)
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 8.10 we may lift the closed form of the kernel, which
is dimensionless: (︁ 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

í Õ ! ¤. û
206 8 The Stokes equation and other integral equations

Using the approximation r + R ≈ 2r still yields

x
1 r − R (︁ ′ ′ )︁ ′
∆g(ϕ, λ, r) ≈ R2 ∆g ϕ , λ , R dσ .
2π σ ℓ3

Alternatively, we derive the spectral form:

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

Degree constituent equation 3.9 gives the functions ∆gn :

x
2n + 1
∆gn (ϕ, λ) = ∆g(ϕ ′ , λ ′ , R) Pn (cos ψ) dσ ′ ,
4π σ

with the aid of which

∆g(ϕ, λ, r) =
1 ∑︂ R n+2
∞ (︂ )︂ x
= 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
∑︂
∞ (︂ )︂n+2
def 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

í Õ ! ¤. û
 The vertical gradient of the gravity anomaly 8.6
207
2
Poisson kernel K
1 1 km K 1 r2 − R2
2
=
2 km R r ℓ3
0

−1 Gravity-anomaly gradient kernel K ′

1 km (︄ (︁ 2 2 2
)︁ )︄
−2 K′ 1 r − R
1.25 km = 3 2 − 32
R3 ℓ r2 ℓ2
1.5 km
−3 2 km

−4
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 for various
height differences r − R. These kernels are used when evaluating
^ the surface integral in map co-ordinates (x, y) in kilometres.

radius 1◦ . See figure 8.7. The main use of Poisson’s kernel is the harmonic
continuation, upwards or downwards, of gravity anomalies measured
and computed at various levels, shifting them to the same reference vertaustaso
level.
In the limit r → R (sea level becomes the level of evaluation), this
kernel function goes asymptotically to the two-dimensional Dirac δ
function. This is inevitable for a kernel that computes gravity anomalies
from gravity anomalies.

^ 8.6 The vertical gradient of the gravity anomaly

Differentiate an equation obtained from equations 5.8 and 5.7:
∑︂
∞ (︂ )︂n+2
R ∂∆g 1 ∑︂ R n+3
∞ (︂ )︂
∆g = r ∆gn =⇒ =− r (n + 2) ∆gn .
∂r R
n=2 n=2

í Õ ! ¤. û
208 8 The Stokes equation and other integral equations

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.
∆gn is expressed, using degree constituent equation 3.9, as an integral
over the anomaly field at sea level:
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 ′ , (8.11)
(︁ )︁
=
4πR σ
in which the (dimensionless) kernel function is now
∑︂
∞ (︂ )︂n+3
R
′
K (ψ, r, R) = − r (2n + 1) (n + 2) Pn (cos ψ).
n=2

Alternatively, derive a closed expression. Start from Poisson equation

8.10 for gravity anomalies, and differentiate with respect to r. See
tableau 8.2.
In the result, the last term is small, 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 similar in magnitude; the
second term goes however rapidly to zero for ℓ ≫ r − R. The factor 1 ℓ3
/︁

does so however even more rapidly.

Write
x ∆g(︁ϕ ′ , λ ′ , R)︁
∂∆g(ϕ, λ, r) R2 5
= κ dσ ′ − ∆g(ϕ, λ, r), (8.13)
∂r 4π σ ℓ 3 2r
with the definition (︁ 2 )︁2
def r − R2 3
κ = 2− . (8.14)
r2 ℓ2
2

í Õ ! ¤. û
 The vertical gradient of the gravity anomaly 8.6
209
^ Tableau 8.2. Derivation of the kernel K ′ for the vertical gradient of gravity
anomaly. Used is the definition of ℓ, equation 8.6, as well as Poisson’s integral
equation 8.10.

x (︃ R
(︄ )︃ )︄
∂∆g(ϕ, λ, r) 1 ∂ (︁ 2 2
)︁ −3 (︁ ′ ′ )︁ ′
= ·R r −R ·ℓ ∆g ϕ , λ , R dσ =
∂r 4π ∂r σ r
R2 x ∂ 1 (︁ 2
(︃ )︃
2
∆g ϕ ′ , λ ′ , R dσ ′ =
)︁ −3 (︁ )︁
= · r −R ·ℓ
4π σ ∂r r
(︁ 2 )︁− 3/2
R2 x
(︄(︃ )︄
r2 − R2 2r 2
)︃
1 1 (︁ 2 d ℓ ∂ℓ
· 3 + · r − R2 · ∆g ϕ ′ , λ ′ , R dσ ′ =
)︁ (︁ )︁
= − 2
+ 2
·
4π σ r r ℓ r d (ℓ ) ∂r

R2 x
(︄ (︃ )︄
r2 − R2 r2 − R2 (︁ 3 −5 )︁
)︃
1
· (2r − 2R cos ψ) ∆g ϕ ′ , λ ′ , R dσ ′ =
(︁ )︁
= 3
− 2
+2 + · −2ℓ
4π σ ℓ r r
R2 x 1 2 2 2 2 2
(︃ )︃
3r −R 1 ℓ +r −R
∆g ϕ ′ , λ ′ , R dσ ′ −
(︁ )︁
= 3
2− 2 2
4π σ ℓ r ℓ r
1 1 x R R r2 − R2
(︁ )︁
− · 3
∆g(ϕ ′ , λ ′ , R) dσ ′ =
r 4π σ r ℓ
R2 x 1
(︄ (︁ 2 )︁ )︄
2
)︁ (︁ 2
2
3r −R
2
3 r −R r − R2 (︁ ′ ′ )︁ ′ 1
= 2 − 2 − 2 ∆g ϕ , λ , R dσ − ∆g(ϕ, λ, r) =
4π σ ℓ3 r2 r2 ℓ2 r

R2 x 1 2 2
(︄ (︁ 2 )︁ )︄ (︃ )︃
3 r −R 1 3
(︁ ′ ′ )︁ ′
= 2− 2 ∆g ϕ , λ , R dσ − + ∆g(ϕ, λ, r) =
4π σ ℓ3 r2 ℓ2 r 2r

R2 x 1 2 2
[︄ (︁ 2 )︁ ]︄
3 r −R 5
∆g ϕ ′ , λ ′ , R dσ ′ − ∆g(ϕ, λ, r). (8.12)
(︁ )︁
= 3
2− 2 2 2
4π σ ℓ r ℓ 2r

Looking at figure 8.7, it is seen that the Poisson kernel K gets narrower
in proportion to r − R and taller in proportion to (r − R)−2 . As the
integral over the Poisson kernel is two-dimensional and scales with the
square of the width, it remains constant when r → R, and in fact the
kernel converges to the two-dimensional Dirac δ function.
For the kernel K ′ of the vertical gradient of the gravity anomaly, the
behaviour is more unpleasant: it narrows in the same way, but, as figure

í Õ ! ¤. û
210 8 The Stokes equation and other integral equations

8.7 shows, gets taller in proportion to (r − R)−3 . This makes its integral
over the sphere diverge in proportion to (r − R)−1 .
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), (8.15)
∂r
but also, like equation 8.13:
x ∆g
˜︂ (ϕ ′ , λ ′ , R)
∂ ∆g
˜︂ 0 (ϕ, λ, r) R2 5 ˜︂
= κ 0 3 dσ ′ − ∆g (ϕ, λ, r). (8.16)
∂r 4π σ ℓ 2r 0
Subtract equation 8.16 from equation 8.13 and substitute equation 8.15,
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
= κ dσ ′ −
4π 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
= κ dσ ′ −
4π σ 3
ℓ (︃ )︃
(︂ )︂2
5 R 2 R 2
(︂ )︂
− ∆g(ϕ, λ, r) − r ∆g0 − r r ∆g0 .
2r

def
Choose the constant ∆g0 = ∆g(ϕ, λ, R), the anomaly at sea level of the
evaluation point:

x ∆g(︁ϕ ′ , λ ′ , R)︁ − ∆g(ϕ, λ, R)

∂∆g(ϕ, λ, r) R2
= κ dσ ′ −
∂r(︃ 4π σ 3
ℓ )︃
(︂ )︂2
5 R 2 R 2
(︂ )︂
− ∆g(ϕ, λ, r) − r ∆g(ϕ, λ, R) − r r ∆g(ϕ, λ, R) ≈
2r

í Õ ! ¤. û
The vertical gradient of the gravity anomaly 8.6
211
x ∆g ϕ ′ , λ ′ , R − ∆g(ϕ, λ, R) ′ 2 (︂ R )︂2
(︁ )︁
R2
≈ κ dσ − r r ∆g(ϕ, λ, R).
4π σ ℓ3
(8.17)

For κ = 2, this would correspond to Heiskanen and Moritz (1967)

equation 2-217, however for an evaluation point at level r ̸= R. For a
well-behaved gravity-anomaly field,
nicely
∆g ϕ ′ , λ ′ , R − ∆g(ϕ, λ, R) → 0 for (ϕ ′ , λ ′ ) → (ϕ, λ) ,
(︁ )︁

and the integral 8.17 will converge for r → R. We posit without proof
that for r → R, convergence will be to the same limit as the Heiskanen
and Moritz equation, in other words, the second term in expression
8.14 fades away and, effectively, κ → 2.
If we are integrating over the surface of a spherical Earth of radius
R rather than the unit sphere σ of radius 1 — or, equivalently, in local
metric co-ordinates (x, y) — we can make the substitution dS = R2 dσ,
with dS a surface element on a sphere of radius R. This removes the
factor R2 from integral equations such as 8.10, 8.12, and 8.17.
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
sea level (on the reference sphere). In our equations 8.17 and 8.11
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 8.17, we may perform a real reduction to sea level, in linear

í Õ ! ¤. û
212 8 The Stokes equation and other integral equations

approximation:

(0)
(︁ )︁ ⃓⃓
∂∆g ϕ, λ, z
∆g(1) ϕ, λ, R ≈ ∆g ϕ, λ, r −
(︁ )︁ (︁ )︁
H.
⃓
∂z
⃓
⃓
z=r

This may be iterated.

^ 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 measured gravity down from the topographic
epäsuora surface to sea level is easy. The indirect effect (the effect of the
vaikutus mass shifts on the geoid, the “restore” step) is small.
◦ 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.

í Õ ! ¤. û
 Gravity reductions in geoid determination 8.7
213
^ 8.7.2 Downwards continuation in linear approximation
The approach described above can, following Molodensky, be linearised:

⎛ (︁ )︁ ⎞
∆g∗ ϕ ′ ,λ ′
⎜⏟ ⏞⏞ ⃓ ⏟⎟
x ⎜
R ∂∆g ⃓⃓ ∂T ⃓
⎟ ⃓
′⎟ ′
T= ⎜∆g − H ⎟ S(ψ) dσ + H. (8.18)
⎜
4π ∂z ⃓z=H ′ ⎟ ∂z z=H
⃓
σ⎜
⎝ ⎠

⏞ ⏟⏟ ⏞
T ∗ (ϕ,λ)

So, first we reduce the ∆g measured and calculated at the topographic

surface to sea level using the vertical gradient of the anomalies and the
terrain height H ′ of the measurement point, with the result
(︁ ′ ′ )︁ ⃓⃓
∂∆g ϕ ,λ ,z ⃓
∆g∗ ϕ ′ , λ ′ = ∆g ϕ ′ , λ ′ , H ′ − 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 ϕ, λ, z
T (ϕ, λ, H) = T ∗ (ϕ, λ) + H.
⃓
∂z
⃓
⃓
z=H
∂
In these equations T , its vertical derivative T,
∆g, and its vertical
∂H
∂
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 = − − r T, (5.6)
∂r
in which r = R + H. Here, we need firstly the vertical derivative of the
disturbing potential. This is easy:
∂T ∂T 2
= = −∆g − r T,
∂H ∂r

í Õ ! ¤. û
214 8 The Stokes equation and other integral equations

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, that is the
anomalous 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.

^ 8.7.3 The evaluation point as the reference level

In the above equation 8.18 we used as the reference level the sea surface.
This is arbitrary: we may use whatever reference level, for example H0 ,
in which case
x (︃ ⃓ )︃
R + H0 ∂∆g ⃓⃓
T= ∆g − (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

x (︃ ⃓ )︃
R+H ∂∆g ⃓⃓
T= ∆g − (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
1 that the linearisation error will remain smaller.1 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.

1 The linearisation error could be even further tuned down by choosing as the evaluation

level for the vertical gradient z = 1

2
(H ′ + H).

í Õ ! ¤. û
 Gravity reductions in geoid determination 8.7
215
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)
equation 7.1. Here, γ is normal gravity calculated for point latitude2 ϕ 2
and topographic height 12 (H + h) ≈ H + 12 ζ.

^ 8.7.4 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
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 may exist
but contains very large positive and negative masses close to each other,
which is physically unrealistic.
One expresses this by saying that the problem is “ill-posed”. In such huonosti
cases, one uses regularisation: one changes the exterior field a little asetettu
— 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. jäännösmaasto-
mallinnus
2 Inan actual calculation one would calculate γHh using the true geodetic latitude φ
and equation 4.10. 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 8 The Stokes equation and other integral equations

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
HRTM for the height of the smoothed, or low-pass filtered, terrain
at the horizontal 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
8.17 for the vertical gradient of the free-air gravity anomaly may
be used for this.
If this anomalous vertical gradient is ignored as it often is, then
the vertical gravity gradient of the terrain-reduced external field is
assumed to be the vertical gradient of normal gravity — according

í Õ ! ¤. û
 Gravity reductions in geoid determination 8.7
217

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, the deviations 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 dashed 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 ′ .

to section 5.4, −0.3 mGal/m — and this operation will leave the
gravity anomaly unchanged.
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
also 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

í Õ ! ¤. û
218 8 The Stokes equation and other integral equations

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
the mass shifts that has to be applied in reverse to arrive at the final
geopotential or geoid solution — is so small as to often be negligible.
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).

^ 8.8 The remove–restore method

All current geoid determination methods are in one way or another
poistamis- “remove–restore” methods, even in several different ways.
entistämis-
menetelmä 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
pallofunktio- of a spherical-harmonic expansion. Thus, a residual gravity field is
kehitelmä obtained
◦ 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
of the order of a few kilometres in size — caused by the
topography, are gone. After this, prediction of gravity values

í Õ ! ¤. û
 Kernel modification 8.9
219
“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.

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
reduction and isostatic reduction. Also Helmert condensation may be
mentioned, although its prediction properties are poorer.
We may illustrate the remove–restore method by means of commu- kommutoiva
tative diagram 8.9. In this diagram, the black arrows with text denote kaavio
calculations that are recommended, because they are easy and accurate.
The grey arrow with text refers to direct computation, which again is
computing intensive and numerically troublesome.

^ 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

í Õ ! ¤. û
220 8 The Stokes equation and other integral equations

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 ′ , (8.19)
4πγ σ0
in which σ0 is a cap on the unit sphere the radius of which is, say, ψ0 .
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 both parts of the integrand in equation 8.19 into spectral form:

∑︂
∞
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
subtracted from the data — and that the model is accurate up to that
degree number.
Now, because ∆gn is a certain linear combination of the surface
spherical harmonics
⎧
⎨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, compare equation 3.13:

1 ∑︂
n
∆gn (ψ, α) = n+1 ∆gnm Ynm (ψ, α),
R
m=−n

and also

Pn (cos ψ) = Pn0 (cos ψ) cos(0 · α) = Yn0 (ψ, α),

í Õ ! ¤. û
Kernel modification 8.9
221
it follows from the orthogonality of the Y functions that
x
Pn (cos ψ) ∆gn ′ (ϕ ′ , λ ′ ) dσ ′ = 0 if n ̸= n ′ .
σ

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–Gore3 modification (Wong and Gore, 1969). A desirable property 3
of the new kernel function SL is that it would be — at least compared
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 8.19) 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.

3 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
222 8 The Stokes equation and other integral equations

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 using a cosine taper.

The reason for the oscillation is that in the frequency or degree-number

domain the modified kernel’s cut-off is quite sharp. Transforming
such a sharp edge between the space and frequency domains will
4 invariably produce an oscillation, which is related to the so-called Gibbs4

Gibbsin ilmiö phenomenon.

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.

4 Josiah
Willard Gibbs (1839–1903) was an American physicist, chemist, thermody-
namicist, mathematician and engineer.

í Õ ! ¤. û
 Advanced kernel modifications 8.10
223
^ 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), (8.20)
n−1 n
n=2

in which the modification coefficients sn , n = 2, . . . , L can be chosen.5 5

They are chosen so as to minimise the values of the kernel SL in the area
outside the cap, σ − σ0 . In this way one may eliminate the truncation
error of equation 8.19 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 8.20 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,
integrate over the area σ − σ0 outside the local cap, and require the
result to vanish:\{
w
S(ψ) Pn (cos ψ) dσ −
σ−σ0
∑︂
L
2n ′ + 1
w
− sn ′ ′ Pn ′ (cos ψ) Pn (cos ψ) dσ = 0,
n −1 σ−σ0
n ′ =2

n = 2, . . . , L,

5 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
224 8 The Stokes equation and other integral equations

a system of L − 1 equations in the L − 1 unknowns sn ′ :

∑︂
L
2n ′ + 1
e ′ s ′ = Qn ,
′
n ′ − 1 nn n
n =2

in which
w wπ
1
Qn = S(ψ) Pn (cos ψ) dσ = S(ψ) Pn (cos ψ) sin ψ dψ
2π σ−σ0 ψ0

and
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
⟨︁ L w
SL (ψ) Pn (cos ψ) dσ,
⟩︁
S · Pn σ−σ0 = (8.21)
σ−σ0

for all values n from 2 to L.

Expressions 8.21 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 inspection, is “smooth”. Depending of course on the values of

í Õ ! ¤. û
 Block integration 8.11
225
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: an
element of the vector space spanned by the polynomials Pn , n = 2, . . . ,
L. But if this is so, and the scalar products 8.21 with each of the basis kantavektori
vectors vanish, then SL must be the zero function on σ − σ0 .
See also Featherstone (2003).
Appendix A section 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 , (8.22)
4πγ
i

in which ∆gi is the mean of block i:

x x
def 1 1
∆gi = ω ∆g(ϕ, λ) dσ = ω ∆g(ϕ, λ) cos ϕ dϕ dλ,
i σi i σi

and the block integral of the Stokes kernel similarly

x
def
S ψ(ϕ, λ; ϕ ′ , λ ′ ) cos ϕ ′ dϕ ′ dλ ′ ,
(︁ )︁
Si (ϕ, λ) =
σi

in which σi is the area of block i and its size on the unit sphere is
x x
def
ωi = dσ = cos ϕ dϕ dλ.
σi σi

í Õ ! ¤. û
226 8 The Stokes equation and other integral equations

4 j=
1 36 1
36 36
1

16
4 36 4
36 36 0

1 1 -1
36 4 36
36
k = -1 0 1

^ Figure 8.11. Simpson integration nodal weights in two dimensions.

Numerical evaluation of such an integral, or quadrature, is done conve-

6 niently using Simpson’s rule:6

w λi + ∆λ/2 w ϕi + ∆ϕ/2 (︁
′ ′
cos ϕ ′ dϕ ′ dλ ′ ≈
)︁
Si (ϕ, λ) = S ψ(ϕ, λ; ϕ , λ )
λi − ∆λ/2 ϕi − ∆ϕ/2

∑︂ 1 ∑︂
1
≈ ∆ϕ ∆λ wj wk Sjk
i ,
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.
(︂ (︁ )︁)︂
Sjk 1 1
cos ϕi + 21 j ∆ϕ ,
(︁ )︁
i (ϕ, λ) = S ψ ϕ, λ; ϕ i + 2
j ∆ϕ, λ i + 2
k ∆λ
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.

6 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
 Effect of the local zone 8.12
227
^ 8.12 Effect of the local zone
One can show that the effect of the local (inner) zone on the geoid height
at the evaluation point (ϕ, λ) is proportional to the gravity anomaly
in the point itself, ∆g(ϕ, λ). Starting from Stokes equation 8.2 with
S(ψ) ≈ 1 sin 12 ψ ≈ 2 ψ , we find, for a circular inner zone of radius ψ0 :
/︁ /︁

w 2π w ψ0
R 2  dψ dα ≈
δN0 = sin
∆g(ψ, α)  ψ
4πγ 0 0 ψ


w ψ0 (︃ w 2π )︃
R 1 R s
≈γ ∆g(ψ, α) dα dψ ≈ γ · ψ0 · ∆g0 = γ0 ∆g0 .
0 2π 0

Here s0 = Rψ0 is the radius of the local block or cap in units of length.
The quantity
w ψ0 (︃ w 2π )︃
def 1 1
∆g0 = ∆g(s, α) dα dψ =
ψ0 0 2π 0
w (︃ w )︃
1 s0 1 2π
=s ∆g(s, α) 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.
The local contributions to the deviations of the plumb line are again
proportional to the horizontal gradients of gravity anomalies. We start
from Vening Meinesz equations 8.5, with the above approximation for
a local cap, specifically

2 d 2
S(ψ) ≈ =⇒ S(ψ) = − 2 :
ψ dψ ψ
{︄ }︄ {︄ }︄
δξ0 w ψ0 w 2π cos α
1 2
(︂ )︂
≈ − 2 ∆g(ϕ ′ , λ ′ ) 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

í Õ ! ¤. û
228 8 The Stokes equation and other integral equations
(︃ )︃
∂∆g ∂∆g
≈ ∆g0 + R ψ cos α + sin α ,
∂x ∂y

and substitute:
{︄ }︄
δξ0 1
≈ ·
δη0 4πγ
{︄ }︄
w ψ0 w 2π (︃ )︂ )︃ cos α
2 ∂∆g ∂∆g
(︂
· − 2 ∆g0 + Rψ 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 =r0. So do the mixed
r 2π
terms in sin α cos α. The only nonzero
2π 2
terms contain 0 sin α dα = 0 cos2 α dα = π :
w ψ0 w 2π
1 2 ∂∆g
δξ0 ≈ − Rψ cos α cos α sin
ψ dα dψ ≈
4πγ 0 0
ψ 2  ∂x
w ψ0 w 2π
R ∂∆g Rψ ∂∆g
≈− cos2 α dα dψ ≈ − 0 ,
2πγ 0 0 ∂x 2γ ∂x
w ψ0 w 2π
1 2 ∂∆g
δη0 ≈ − R sin α
ψ sin α 
sin
ψ dα dψ ≈
4πγ 0 0 ψ 2 ∂y
w ψ0 w 2π
R ∂∆g Rψ ∂∆g
≈− sin2 α dα dψ ≈ − 0 .
2πγ 0 0 ∂y 2γ ∂y
Evaluating these integrals assumes 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
8.22, behaves numerically poorly in the immediate surroundings of the
evaluation point if the kernel function is singular at the origin ψ = 0.
Both the Stokes 8.2 and Vening Meinesz 8.5 kernels are of this kind.

^ Self-test questions
1. What do the Stokes equation and its spectral form look like?

í Õ ! ¤. û
 Exercise 8 – 1: The Stokes equation in the near zone
229
(︁ )︁
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?
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 approach.
10. Why, in geoid determination, is the Stokes kernel function often
modified? What does such a modification look like?
11. What is the Gibbs phenomenon?

^ Exercise 8 – 1: The Stokes equation in the near zone

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 2π w ∞
· · · 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 data point for the
gravity anomaly ∆g.

í Õ ! ¤. û
230 8 The Stokes equation and other integral equations

Hint: you need to consider Jacobi’s determinant for the polar co-
ordinates (s, α).
3. Compute N (as an equation) 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 equation as a convolution
We start from the Stokes equation 8.1,
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,
on the surface of a spherical Earth. So, the locations of both points are
given in spherical co-ordinates. 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 is 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 cos α, y = ψR sin 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

– 231 –
232 9 Spectral techniques, FFT

x
Data point
α
y
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, 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 cos α, y = 2 tan R sin 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 equation as a convolution 9.1
233
Furthermore, we must account for Jacobi’s determinant R2 of the
projection:
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 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- pallofunktio-
harmonic expansions, because these represent the long-wavelength kehitelmä
part of the 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 poistamisvaihe
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 over greater
distances.
Of course, once the integral has been computed and the local disturb- häiriö-
potentiaali
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, a small part of the whole
Earth’s surface. This is the case for the modified kernels discussed in section 8.9.

í Õ ! ¤. û
 234 9 Spectral techniques, FFT

ing potential Tloc , and the corresponding geoid undulation Nloc , have
been obtained, we must remember to add to them again the effect of
the global spherical-harmonic expansion on the disturbing potential T
entistämisvaihe and 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 multiple 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 build a discrete grid representation of
the function ∆g(x, y), a rectangular table of ∆g values on an equispaced
(xi , yj ) grid of points. The values may be, say, the function values
2 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 Alternatively,
one could for example calculate for every grid point the average over a
square cell surrounding the point.

í Õ ! ¤. û
Integration by FFT 9.2
235
so we write
(︁ )︁
Sij = S ∆xi , ∆yj ,
where again

∆xi = i δx, ∆yj = j δy, i, j = 0, 1, . . . , N − 1.

Now the central peak at the origin of the kernel 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 , in one corner, and the grid
contains only one quadrant of the peak. This is not acceptable.
The periodicity inherent in the discrete Fourier transform means that
the values i = 12 N, . . . , N − 1 may be replaced by the negative values
def
i ′ = i − N = − 21 N, . . . , −1 without formally changing anything: see
footnote 1 in appendix C. In this interpretation

∆xi ′ = i ′ δx, ∆yj ′ = j ′ δy, i ′ , j ′ = − 12 N, . . . , 12 N − 1,

landing the origin in the centre of the grid. This is the correct way to
compute the values of the true, non-periodic kernel, with both positive
and negative values ∆x and ∆y from an area symmetrically surrounding
the origin.
Next:
1. The grid representations ∆gij and Sij thus obtained of the func-
tions ∆g and S are transformed to the frequency domain — they
{︁ }︁ {︁ }︁
become functions Suv = F Sij and Guv = F ∆gij of the two
“frequencies”, the wave indices u and v in the x and y direc-
tions. The spatial frequencies or wave numbers3 ν ˜︁ and spatial 3
u L, ν v L , in which
/︁ /︁
wavelengths λ are ν˜︁x = λ−1
x = ˜︁y = λ−1
y =
L = Nδx = Nδy is the size of the area, assumed to be square.

3 This is the so-called linear wave number, whole waves per unit of length. The angular
wave number is k = 2π˜︁
ν, radians of phase angle per unit of length.

í Õ ! ¤. û
236 9 Spectral techniques, FFT

2. They are multiplied with each other “one frequency pair at a

time”: we calculate
1
Tuv = S ·G , u, v = 0, 1, . . . , 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
/︁

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 8.17. The only requirement is that the equation can be
expressed as a convolution.
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 that the input data, the
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.
The FFT method may be depicted as a commutative diagram, figure 9.2.
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

í Õ ! ¤. û
 Solution in latitude and longitude 9.3
237
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.

grid by interpolating from the given (φ, λ) one through a map projection
calculation is avoided. However, working in geographical co-ordinates
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
λ − λ′
cos λ − λ ′ = 1 − 2 sin2
(︁ )︁
,
2
ψ
cos ψ = 1 − 2 sin2 ,
2
ϕ − ϕ′
cos ϕ − ϕ ′ = 1 − 2 sin2
(︁ )︁
,
2

í Õ ! ¤. û
238 9 Spectral techniques, FFT

and obtain the half-angle cosine rule:

λ − λ′
cos ψ = cos ϕ − ϕ ′ − 2 cos ϕ cos ϕ ′ sin2
(︁ )︁
2
′
ψ ϕ − ϕ λ − λ′
=⇒ sin2 = sin2 + cos ϕ cos ϕ ′ sin2 .
2 2 2
Here we may use the following approximation:

cos ϕ ′ , cos ϕ ≈ cos ϕ0 ,

vertaus- in which ϕ0 is a reference latitude in the middle of the calculation area.

leveysaste Now the above equation becomes

ψ ϕ − ϕ′ λ − λ′
sin2 ≈ sin2 + cos2 ϕ0 sin2 , (9.4)
2 2 2
def def
which depends only on the differences ∆ϕ = ϕ − ϕ’ and ∆λ = λ − λ ′ .
After this, the FFT method may be applied by using co-ordinates
4 (ϕ, λ)4 and the Stokes kernel written as

(︄ √︃ )︄
∆ϕ ∆λ
S(ψ) = S(∆ϕ, ∆λ) = S 2 arcsin sin2 + cos2 ϕ0 sin2 ,
2 2

which is now a function of only the differences ∆ϕ and ∆λ, as the con-
volution theorem requires. This clever way of using FFT in geographical
5 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.
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 9.3
239
Write the Stokes equation as follows:
x (︁
R )︁[︂ (︁ ]︂
S ∆ϕ, ∆λ; ϕ ∆g ϕ ′ , λ ′ cos ϕ ′ dϕ ′ dλ ′ ,
)︁
N(ϕ, λ) = (9.5)
4πγ
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 sufficiently
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:

(︃ )︃
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 =

Equation 9.6 is the linear combination of two convolutions. Both are

evaluated by FFT. The equation forms the weighted mean from the
solutions obtained.
In this method we use, instead of approximative equation 9.4, an
exact equation in which ϕ ′ is expressed into ϕ and ∆ϕ:

ψ ϕ − ϕ′ λ − λ′
sin2 = sin2 + cos ϕ cos ϕ ′ sin2 =
2 2 2
∆ϕ ∆λ
= sin2 + cos ϕ cos (ϕ − ∆ϕ) sin2 .
2 2
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

í Õ ! ¤. û
240 9 Spectral techniques, FFT

interpolate N(ϕ, λ) according to equation 9.6 when ϕi ⩽ ϕ < ϕi+1 .

After this, the solution is not 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 series with respect to latitude
6 about a reference latitude located in the middle of the computation area.6

Each term in the expansion depends only on the difference in latitude.

The integral to be calculated similarly expands into terms, of which
each contains a pure convolution.
Let us write the general problem as follows:
w 2π w + π/2 )︁[︂ (︁ ]︂
C ϕ, ϕ ′ , ∆λ m ϕ ′ , λ ′ cos ϕ ′ dϕ ′ dλ ′ ,
(︁ )︁
ℓ(ϕ, λ) =
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 and 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 and ϕ ′ into a dependence
upon ϕ and ∆ϕ:

C = C ϕ, ϕ ′ , ∆λ = C(∆ϕ, ∆λ; ϕ).

(︁ )︁

6 Inthe literature the method has been generalised by also expanding the kernel with
respect to height.

í Õ ! ¤. û
Solution in latitude and longitude 9.3
241
Linearise:

C = C0 (∆ϕ, ∆λ) + (ϕ − ϕ0 ) Cϕ (∆ϕ, ∆λ) + · · ·

where we define for a suitable reference latitude ϕ0 :

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 be accurate 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)

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.

í Õ ! ¤. û
242 9 Spectral techniques, FFT

Let us choose, say, a 256 × 256 size grid (so N = 256) and fill in
the missing values by extrapolation.
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
7
[︁ ]︁
Cϕ ⊗ m cos ϕ by means of FFT — like7
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 “stan-
8 dard operations”,8 multiplication with the coefficients (ϕ − ϕ0 )
and adding together, again 65 536 standard operations.
The grid matrices corresponding to kernel functions C0 and Cϕ
are obtained as follows: for three reference latitudes ϕ−1 , ϕ0 , ϕ+1
we 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
9 follows:9

7 Fourier transforms are multiplied by multiplying the corresponding elements, see

section 9.2 equation 9.3.
8A standard operation is a multiplication plus either an addition or a subtraction.

í Õ ! ¤. û
 Bordering and tapering of the data area 9.4
{︁ }︁ }︃
243
{︃
{︁ }︁ {︁ }︁ {︁ }︁ ]︁(0) F ℓ
F C0 · F m cos ϕ = F ℓ = F−1 {︁ }︁ .
[︁
=⇒ m cos ϕ
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

{︃ {︁ }︁
(0) }︃
]︁(1) ]︁(0) F ℓ − ℓ
+ F−1 {︁ }︁
[︁ [︁
m cos ϕ = m cos ϕ ,
F C0
and so on, iteratively. Two, three iterations are enough. This method has
been used to compute underground mass points from gravity anomalies
to represent the exterior gravity field of the Earth.10 More is explained 10
in Forsberg and Vermeer (1992); Vermeer (1992).

^ 9.3.4 “1-D 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

9A Fourier transform is divided by another one by dividing the corresponding

elements, see section 9.2.
10 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
244 9 Spectral techniques, FFT

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
11 edges.11 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.
The calculation area for the kernel function is also made similarly
four times larger. In this case the whole grid including the border
area is filled with real (computed) values.
The grid of the kernel function must be filled in such a way,
that index values i, j > N 2 are interpreted as negative values
/︁

i − N and j − N, representing also negative ∆xi and ∆yj . Then,

the peak of the function will be in the centre of the grid. If the
function is symmetric, the four quadrants of the grid will look
like mirror images of each other. Then, the grid 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.

11 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
 Bordering and tapering of the data area 9.4
245
25 % 50 % 25 %
Data area
1

^ Figure 9.3. “Tapering” 25 %.

If the values at the edges are not zero, they may be forced to
zero by multiplying the whole calculation 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 transform. These artefacts are related to the Gibbs Gibbsin ilmiö
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, The group
of Klaus-Peter Schwarz and Michael Sideris in Calgary, Canada, the
Delft group (Strang van Hees, Haagmans, De Min, Van Gelderen),
the Milanese group (Sansò, Barzaghi, Brovelli), Heiner Denker at the
Hannover “Institut für Erdmessung”, and many others.

í Õ ! ¤. û
246 9 Spectral techniques, FFT

Figure 9.4. Example images for the FFT transform without (above) and with
(below) tapering. The online FFT service from Watts (2004) was
used. The images are greytone amplitude spectra |Fuv | plotted
^ with the origin u = v = 0 in the centre, see appendix C.

^ 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
 Computing a geoid model with FFT 9.5
247
^ 9.5.1 GRAVSOFT software
The GRAVSOFT geoid determination software has been mainly produced
in Denmark. Authors include Carl Christian Tscherning,12 René Fors- 12
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 provides, in addition
to variants of FFT geoid determination, for example least-squares col- pienimmän
location, as well as routines for evaluating various terrain effects. Its neliösumman
kollokaatio
popularity can be partly explained by it being free for scientific use, and
being distributed as source code. It is also well-documented. There-
fore 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
is a reference surface for the N60 height system: using it together with vertauspinta
GNSS positioning allows determination of the N60 heights of points.
The model gives geoid heights above the GRS80 reference ellipsoid. The vertaus-
second model is similarly a reference surface for the new N2000 height ellipsoidi
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.

12 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
248 9 Spectral techniques, FFT

20˚ ◦
20 24◦
24˚ 28◦
28˚ 32˚ ◦
32

20
◦
0 ◦˚
770 7700˚
19

8 ◦˚
668 6688˚◦

24 25

22
21
23
31 28
29
30

20
26
27

6 ◦˚
666 19 6666˚◦

17

18
18

664
4 ◦˚ 6644˚◦

18
5
23 24 2

20 212
2

19

2 ◦˚
662 6622˚◦
18

17
19 16

660
0 ◦˚ 6600˚◦
15

16

20˚ ◦
20
28◦ 32˚◦
32
24◦
24˚ 28˚

2010 Oct 20 13:27:28

^ Figure 9.5. The Finnish FIN2000 geoid. Data © Finnish Geodetic Institute.

í Õ ! ¤. û
 Use of FFT in other contexts 9.6
249
^ 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 inverting satellite
altimetry derived “geoid heights” to gravity anomalies (Andersen et al.,
2010). A pioneer of this method has been David Sandwell from the
Scripps Institute of Oceanography in California, for example 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- ilma-
ping of large areas. In the pioneering days, the gravity field over gravimetria
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 strength of airborne gravimetry
is that one rapidly measures large areas in a homogeneous way.

^ 9.7 Computing terrain corrections with FFT

The terrain correction is a very localised phenomenon, the calculation
of which requires high-resolution terrain data from a relatively small

í Õ ! ¤. û
250 9 Spectral techniques, FFT

area surrounding the computation point. Thus, calculating the terrain

correction is ideally suited for the FFT method.
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 yields
the vertical component of the attraction of rock columns (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 3
dx ′ dy ′ . (9.8)
−∞ ℓ
/︂ 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.
vinoetäisyys 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 follows 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
Computing terrain corrections with FFT 9.7
251
From equation 9.8 we obtain by expansion into terms:
x +∞ x +∞ ′
1 ′ ′ H
T C(x, y) = 12 GρH2 dx dy − GρH dx ′ dy ′ +
−∞ 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 an
unpleasant situation. There is a solution for this which we present 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
√︂ √ /︂
2 2 u 2 + v2
in which q = ν ˜︁x + ν
˜︁y = L , u and v are wave indices, and
˜︁x = u L and ν ˜︁y = v L are (linear) “spatial frequencies” or wave
/︁ /︁
ν
numbers 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 ≈ 21 GρH2 F 1 · (1 − 2πδq) −
δ
{︁ ′ }︁ (︂ 2π )︂
− GρH F H · (1 − 2πδq) +
δ
{︁ 2 }︁
(︂
2π
)︂
+ 12 Gρ F (H ′ ) · (1 − 2πδq) ,
δ

í Õ ! ¤. û
252 9 Spectral techniques, FFT

where we left off all terms in higher powers of δ.

Re-order the terms:
{︁ }︁ 2π (︂ {︁ }︁ {︁ }︁ {︁ 2 }︁
)︂
F TC = Gρ 12 H2 F 1 − H F H ′ + 21 F (H ′ ) −
δ (︂ {︁ }︁ {︁ ′ }︁ 1 {︁ ′ 2 }︁)︂
− 2πGρ · 2πq · 2 H F 1 − H F H + 2 F (H )
1 2
.
{︁ }︁
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):
{︁ }︁ 2π (︂ {︁
′ 2
}︁)︂
F TC = Gρ 2 F H − HH + 2 (H )
1 2 ′ 1
+
δ (︂ {︁ }︁ {︁ 2 }︁
)︂
+ 2πGρ · 2πq · H F H ′ − 21 F (H ′ )
and the reverse Fourier transform yields
π
(︂ )︂
2 ′ ′ 2
T C = Gρ H − 2H H + (H ) +
δ {︃ }︃
(︂ {︁ ′ }︁ 1 {︁ ′ 2 }︁)︂
+ 2πGρ F −1
2πq · H F H − 2 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
{︃ (︂ }︃
{︁ ′ }︁ 1 {︁ ′ 2 }︁)︂
T C = 4π Gρ F
2 −1
q · H F H − 2 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): the evaluation happens at the Earth’s surface.
The Fourier transforms above are evaluated by the FFT method.
For calculating the terrain correction T C 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 T C as a sum of convolutions, a Taylor
series expansion. An early paper on this is Parker (1972).

í Õ ! ¤. û
 Self-test questions
253
^ 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 calculation
area 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 always 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

– 255 –
256 10 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 :

d
⃓
f ↦→ f(x)⃓ .
⃓
dx x=x0

A trivial functional is also the evaluation functional, the function value

itself (one could say 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

d⃓
⃓ {︁ }︁ d
⃓
L= , meaning L f = f(x)⃓ .
⃓
dx x=x0 dx
⃓
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
häiriö- function T (ϕ, λ, R) = T (ϕ, λ, r)|r=R , that is, of the disturbing potential
potentiaali at the surface of a spherical Earth. The theory thus uses the spherical
1 approximation,1 and the surface of the sphere of radius R corresponds

í Õ ! ¤. û
 Statistics on the Earth’s surface 10.3
257
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) ↦→ ξ(ϕ, λ, r),
T (·, ·, R) ↦→ ∆g(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

ϕ ∈ − π 2 ,+ π 2 ,
[︁ /︁ /︁ ]︁ [︁ )︁
T (ϕ, λ, R), 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 value set or codomain of which is a function space. arvojoukko
In other words, it is a random variable, the realisation values of which
are functions. A stochastic process may be a quantity developing over
time, the precise behaviour of which is uncertain, for example a satellite

í Õ ! ¤. û
258 10 Statistical methods

orbit. In the same way as for a (real-valued) stochastic quantity x we

{︁ }︁
odotusarvo may define an expected value or expectancy E x and a variance
{︃(︂ }︃
def {︁ }︁ {︁ }︁)︂2
Σxx = Var x = E x − E x ,

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:
def {︁ }︁
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:
(︁ )︁ (︁ )︁ def {︁ }︁
Ax t1 , t2 = Cxx 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:
(︁ )︁ def {︁ }︁
Cxy t1 , t2 = Cov x(t1 ), y(t2 ) =
{︃(︂ }︃
{︁ }︁)︂(︂ {︁ }︁)︂
= E x(t1 ) − E x(t1 ) y(t2 ) − E y(t2 ) .

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: 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, a special problem arises.
A stochastic quantity is generally defined as a quantity x from which

1 This is not mandatory, but the error of approximation is usually small.

í Õ ! ¤. û
Statistics on the Earth’s surface 10.3
259
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
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 — for example some geophysical quantity
— on the surface of the Earth, x(ϕ, λ), we define a quantity similar to
{︁ }︁
the statistical expectancy E · , the geographic mean odotusarvo

{︁ }︁ def 1 x w w
1 2π + π/2
M x = x(ϕ, λ) dσ = x(ϕ, λ) cos ϕ dϕ dλ.
4π σ 4π 0 − π/2
(10.1)
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 field x.
Similarly to the statistical variance based on expectancy, we may
define the geographic variance:
{︃(︂ }︃
{︁ }︁ 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 2

í Õ ! ¤. û
260 10 Statistical methods

their definition:
{︁ }︁
M ∆g = 0.
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
{︁ }︁
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 :
(︁ )︁ (︁ )︁
∆gP = ∆g ϕP , λP , ∆gQ = ∆g ϕQ , λQ ,

can move independently over the whole Earth’s surface.

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

2 Thisis 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.

í Õ ! ¤. û
 The covariance function of the gravity field 10.4
261
α
P
Q

Earth’s centre of mass

^ Figure 10.1. Definition of geocentric angular distance and azimuth.

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 .

ϕQ and λQ can be computed3 if we know ϕP and λP as well as both the 3

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 = M ∆gP ϕP , λP ∆gQ ϕP , λP , ψPQ , αPQ =
x
1 (︁ )︁ (︁ )︁
= ∆gP ϕP , λP ∆gQ ϕP , λP , ψPQ , αPQ dσP . (10.3)
4π σ
3 This is called the geodetic forward problem on the sphere.

í Õ ! ¤. û
262 10 Statistical methods

Also here, M is a geographic-mean operator. First, fix point Q in relation

4 to point P, while both azimuth αPQ and distance ψPQ are held fixed.4
Move point P, and with it, point Q, over the whole of the Earth’s surface.
Compute the corresponding integral over the unit sphere σ, and divide
by 4π:
{︁ }︁ x
(︁ )︁ 1
C∆g∆g ψPQ , αPQ = M ∆gP ∆gQ(P) = ∆g ∆gQ(P) dσ =
4π σ P
w w
1 2π + π/2
= ∆gP ∆gQ(P) cos ϕ dϕ dλ,
4π 0 − π/2
in which dσ = cos ϕ dϕ dλ is used, and cos ϕ is Jacobi’s determinant for
the co-ordinates (ϕ, λ) = (ϕP , λP ) on the unit sphere.
In addition to the assumption of homogeneity, we make still the
assumption of isotropy: the covariance function — 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 averaging also over all azimuth angles αPQ ∈ 0, 2π :
)︁ def {︁ }︁
C∆g∆g ψPQ = M ′ ∆gP ∆gQ(P) =
(︁
w
1 2π {︁ }︁
= M ∆gP ∆gQ(P) dαPQ =
2π 0
w 2π w 2π w + π/2
1
= 2 ∆gP ∆gQ(P) cos ϕ dϕ dλ dαPQ . (10.4)
8π 0 0 − π/2
4 The critical reader may interject that, while the angular distance ψAB exists inde-
pendently from the definition of geographical co-ordinates, that is not the case for
the azimuth angle αAB : it depends on the local direction of the meridian. If in
equation 10.3 the angle αAB is the conventional geodetic azimuth, this definition
of the covariance function considers only one specific possible pattern of azimuth
dependence. This suggests a generalisation to azimuth angles defined with respect to
general curvilinear co-ordinates on the Earth’s surface.
Also isotropy should then be understood as the absence of azimuth dependence not
only in geographical co-ordinates, but in all possible curvilinear co-ordinates.

í Õ ! ¤. û
 Least-squares collocation 10.5
263
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

Collocation is a statistical estimation technique used to estimate the pienimmän
values of a stochastic process and calculate the uncertainties (like mean neliösumman
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, 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 travelled.
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 t1 , t1 C t1 , t2 · · · C t1 , tN
{︁ }︁ ⎢ C t2 , t1
⎢ (︁ )︁ (︁ )︁ (︁ )︁ ⎥
C t2 , t2 · · · C t2 , tN ⎥
Var si = ⎢ .. .. .. .. ⎥.
⎢
⎣ . . . .
⎥
⎦
(︁ )︁ (︁ )︁ (︁ )︁
C tN , t1 C tN , t 2 · · · 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(t, t ′ ), or C(∆t), is known, then the whole

í Õ ! ¤. û
264 10 Statistical methods

matrix and all of its elements can be calculated, provided all argument
values (observation times) ti are also known.
Let the shape of the problem now be that one should estimate, or
predict, the value of process s at the moment in time T , that is 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
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 , t 1 C T1 , t 2 · · · C T1 , t N
{︂ (︁ )︁ }︂ ⎢
(︁ )︁ (︁ )︁ (︁ )︁ ⎥
⎢ C T2 , t 1 C T2 , t 2 · · · C T2 , t N ⎥
Cov s Tp , s(ti ) = ⎢ .. .. .. ⎥.
. . .
⎢ ⎥
⎣ ⎦
(︁ )︁ (︁ )︁ (︁ )︁
C TM , t 1 C TM , t 2 · · · C TM , t N
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 . (10.5)

í Õ ! ¤. û
 Least-squares collocation 10.5
265
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 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 and nj of two different observations ℓi and
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

as a linear combination of the observations ℓi at our disposal. 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-
{︂ (︁ )︁ (︁ )︁}︂ varianssi
def
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 ,

and5 5

í Õ ! ¤. û
266 10 Statistical methods
{︃(︂ }︃
(︁ )︁ (︁ )︁)︂ (︂ (︁ )︁ (︁ )︁ )︂
Σpq = Cov ˆ︁ s Tp − s Tp , ˆ︁ s Tq − 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 . (10.6)

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. Choose
def
Λpj = Cpi (Cij + Dij )−1 .

Then, from equation 10.6 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 . (10.7)

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 ,

in which

5 The matrix Ciq is the transpose of Cpj , the matrix Λjq the transpose of Λpi .

í Õ ! ¤. û
 Least-squares collocation 10.5
267
I = Λpi (Cij + Dij ) Λjp =
(︂ )︂ (︂ )︂
−1 −1
= Cpi (Cij + Dij ) + δΛpj (Cij + Dij ) (Cjk + Djk ) Ckp + δΛkp =
hhhh ˂
˂˂˂˂
−1
=˂
Cpi
˂˂ (C˂ijh
˂+h
˂Dij ) hh
˂h
h Ch
jp
h
+ Cpi δΛ
hhhh
hip
h˂+˂δΛ
˂˂pi Cip + δΛpi (Cij + Dij ) δΛjp ,
˂˂

(︂ )︂
II = − Λpj Cjp = − Cpi (Cij + Dij )−1 + δΛpj Cjp =
hhhh ˂
˂˂˂˂
−1
= −˂C˂ ˂(C ij )hhC
h+ Dh −˂δΛ pi Cip
h h˂
˂ ˂˂
˂˂
˂ pi
˂ij
hjp
˂ h
h˂

and
(︂ )︂
III = − Cpi Λip = − Cpi (Cij + Dij )−1 Cjp + δΛip =
= − Cpi (Cij + Dij )−1 Cjph
−hCh δΛ
pih h,
hip

with the final result

III
⏟ ⏞⏞ ⏟⏟ I
⏞⏞ ⏟
′ −1
Σpp = Cpp − Cpi (Cij + Dij ) Cjp + δΛpi (Cij + Dij ) δΛjp .

Here, the last term — the only difference from result 10.7 — 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 s Tp = Cpi (Cij + Dij )−1 ℓj ,

(︁ )︁
=⇒ ˆ︁

is optimal in the sense of least-squares — more precisely, in the sense of

minimising the variance of prediction Σpp .

^ 10.5.5 The covariance function of gravity anomalies

Least-squares collocation is used extensively to optimally estimate
gravity values and other functionals of the gravity field on the Earth’s
surface.

í Õ ! ¤. û
268 10 Statistical methods

If we have two points, P and Q, with measured gravity anoma-

(︁ )︁ (︁ )︁
lies ∆gP = ∆g ϕP , λP and ∆gQ = ∆g ϕQ , λQ , one would like to
determine 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). The covariance will be
{︁ }︁ {︁ }︁
Cov ∆gP , ∆gQ = M ′ ∆gP ∆gQ(P) = C ψPQ .
(︁ )︁

6 A popular covariance function for gravity anomalies is Hirvonen’s6

covariance function:
C0
C(ψ) = (︁ /︁ )︁2 , (10.8)
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
7 correlation between the gravity anomalies in two points is still 50 %.7

In local applications, instead of the angular distance ψ one uses the

linear distance
s = R ψ,
where R is the radius of the Earth. Then
C0
C(s) = (︁ /︁ )︁2 .
1+ s d

6 ReinoAntero Hirvonen (1908–1989) was a Finnish physical and mathematical

geodesist.

í Õ ! ¤. û
 Least-squares collocation 10.5
269
This equation was derived from gravimetric data for Ohio state, USA,
but it has broader validity. C(0) = C0 , the signal variance. The variable
d = R ψ0 is also called the correlation length. It is the distance d for
which C(d) = 12 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,
and a quadratic variant:
(︃ (︂ )︂ )︃
ψ ψ 2
(︂ )︂
C(ψ) = C0 exp − , C(ψ) = C0 exp − .
ψ0 ψ0

^ 10.5.6 Least-squares collocation for gravity anomalies

If N points Pi , i = 1, . . . , N are given, where were measured gravity
(︁ )︁
values and calculated anomalies — ∆gi = ∆g ϕi , λi , we may, as above,
construct a signal variance matrix
{︂ }︂
def
Cij = Var ∆gi =
⎡ (︁ )︁ (︁ )︁ ⎤ ⎡ ⎤
C0 C ψ12 · · · C ψ1N C0 C12 · · · C1N
⎢ (︁ )︁ (︁ )︁ ⎥ ⎢
⎢ C ψ21 C0 · · · C ψ2N ⎥ ⎢ C21 C0 · · · C2N ⎥
⎥
=⎢ .. .. .. ⎥=⎢ .
⎥ ⎢ . .. ⎥,
.. ⎥
. . . . . . ⎦
⎢
⎣ ⎦ ⎣
(︁ )︁ (︁ )︁
C ψN1 C ψN2 · · · C0 CN1 CN2 · · · C0

in which all elements C(ψij ) are calculated using covariance function

10.8 given above.

7 The correlation is
{︂ }︂ C0
{︂ }︂ Cov ∆gP , ∆gQ
/︁ )︁2
1 + ψ ψ0
(︁
1
Corr ∆gP , ∆gQ = √︃ {︂ }︂ {︂ }︂ = √ = (︁ /︁ )︁2 ,
C0 C0 1+ ψ ψ0
Var ∆gP Var ∆gQ

which is 0.5 for ψ = ψ0 .

í Õ ! ¤. û
270 10 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.

If we also compute for the point P at which gravity is unknown:

{︂ }︂ [︂ (︁ )︁ (︁ )︁ (︁ )︁ ]︂ def
Cov ∆gP , ∆gi = C ψP1 C ψP2 · · · C ψPN = CPi ,

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 ∆g1 and ∆g2 (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.

í Õ ! ¤. û
 Least-squares collocation 10.5
271
x
1 (15 mGal)
30
P′
2 (20 mGal)
20

10 P

y
10 20 30

^ Figure 10.4. Least-squares collocation, calculation example.

we obtain, in the same way as before, for the least-squares collocation

solution
ˆ︂ P = CPi (Cij + Dij )−1 ℓj ≈ CPi C−1 ℓj ,
∆g ij

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,
again describes the random observation error, observation uncertainty,
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
variance of prediction, equation 10.11:

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 .

í Õ ! ¤. û
272 10 Statistical methods

^ 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
determination of the gravity points! — are errorless. So, Dij = 0, i,
j = 1, 2.
Kysymys Calculate an estimate of the gravity anomaly ∆g
ˆ︂ P at point P
√
and its mean error σ∆gP = ΣPP .
Vastaus Calculate first the distances s and the corresponding covari-
ances 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 .
500
/︁
1+ 400

From this follows

Cij + Dij ≈ [︄ ]︄ [︄ ]︄
C11 C12 1000 666.66
≈ Cij = = mGal2 ,
C21 C22 666.66 1000

í Õ ! ¤. û
Least-squares collocation 10.5
273
and its inverse matrix
[︄ ]︄
0.0018 −0.0012
(Cij + Dij )−1 = 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

∆g
ˆ︂ P = [︄ ]︄ [︄ ]︄
[︂ ]︂ 0.0018 −0.0012 15
= 444.44 444.44 mGal =
−0.0012 0.0018 20
= 9.333 mGal.

The precision, the variance of prediction, equation 10.11:

ΣPP = CPP − CPi (Cij + Dij )−1 CjP =

[︄ ]︄ [︄ ]︄
[︂ ]︂ 0.0018 −0.0012 444.44
= C0 − 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

í Õ ! ¤. û
274 10 Statistical methods

zero. In fact, not using the observational data at all would leave
us with the a priori estimate
√
ˆ︂ P = 0 ±
∆g 1000 mGal = 0 ± 31.623 mGal,

almost as good.
If, instead, we had used point P ′ in between points 1 and 2, at
location (25 km, 25 km), then
/︂
2 (︁
CP ′ 1 = CP ′ 2 = 1000 mGal 1 + 50 400 = 888.89 mGal
2
/︁ )︁

ˆ︂ P ′ = 18.667 ± 7.201 mGal, which is clearly better than the

and ∆g
a priori 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 = 7.664 ± 29.272 mGal

ˆ︂ P ′ = 16.460 ± 18.426 mGal
for the original point location, and ∆g
for the shifted point location.

^ 10.5.8 Theory of least-squares collocation

pienimmän Above we presented one popular application of least-squares collocation.
neliösumman Here we look at the method more generally. The basic equation is
kollokaatio
ˆ︁f = Cfg (Cgg + Dgg )−1 g + n .
(︁ )︁
(10.9)

The vector g contains observed quantities gi , the vector n contains the

observational errors: uncertainty or noise, and ˆ︁f is a vector of quantities
fˆ︁p to be predicted.
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

í Õ ! ¤. û
Least-squares collocation 10.5
275
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
can be expressed as follows:8 8

{︁ }︁ {︁ }︁ {︁ }︁
= M ′ fp gi , = M ′ g i gj ,
[︁ ]︁ [︁ ]︁ [︁ ]︁
Cfg pi
Cgg ij
Dgg ij
= E ni nj ,

in which ni , an element of vector n, represents the uncertainty of the

observation process appearing in observation equation 10.5:

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.
The great strength of least-squares collocation is its flexibility. Different
observation types may be handled with a single unified theory 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.

8 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
276 10 Statistical methods

^ 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, one speaks of homogeneous prediction. For example,
the prediction equation for gravity anomalies already presented in
subsection 10.5.6 is obtained from equation 10.9 by substitution:
ˆ︂ P = CPi (Cij + Dij )−1 ℓj .
∆g (10.10)

Here are several points j where gravity has been measured: let us say,
N observations ℓj = ∆gj + nj , j = 1, . . . , N. The number of points to be
predicted may be one: 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, CPi is a size 1 × N row matrix.
9 The prediction error is now the difference quantity9 ∆gˆ︂ P − ∆g , and
P
its variance (“variance of prediction”) is
{︂ }︂
def
ΣPP = Var ∆g ˆ︂ P − ∆g =
P
{︁ }︁ {︂ }︂ {︂ }︂ {︂ }︂
= Var ∆g ˆ︂ P + Var ∆g − Cov ∆g
P
ˆ︂ P , ∆g − Cov ∆g , ∆g
P P
ˆ︂ P .

Here (propagation of variances applied to equation 10.10):

{︁ }︁
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 =
(︃ {︂ }︂ )︃
−1
= CPi (Cij + Dij ) Cov ∆gj , ∆gP + 0 =

= CPi (Cij + Dij )−1 CjP ,

9 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
Prediction of gravity anomalies 10.6
277
and also
{︂ }︂
ˆ︂ P = CPi (Cij + Dij )−1 CjP
Cov ∆gP , ∆g
{︂ }︂
and finally, the signal variance Var ∆gP = CPP .
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 . (10.11)

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 10.10 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 .

í Õ ! ¤. û
278 10 Statistical methods

^ 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 ) = M TP TQ(P) ,

or alternatively using equation 10.4:

)︁ def {︁ }︁
K(P, Q) = K ψPQ = M ′ TP TQ(P) =
(︁
w 2π w 2π w + π/2
1
= 2 T T cos ϕ dϕ dλ dαPQ . (10.12)
8π 0 0 − π/2 P Q(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) = (10.13)
⎩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 10.13 can only be
independent of α if m = 0. So

∑︂
∞ ∑︂
∞
K(ψ) = kn0 Yn0 (ψ) = kn Pn (cos ψ). (10.14)
n=2 n=2

í Õ ! ¤. û
 Covariance function and degree variances 10.7
279
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.

^ 10.7.2 Degree variances and spherical-harmonic coefficients

Multiply equation 10.14 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 ∑︂
∞ w +1
2
= kn Pn (t) Pn ′ (t) dt = kn ′ ,
−1 2n + 1
n=2

using orthogonality condition 3.5. It follows that

w
2n + 1 π
kn = K(ψ) Pn (cos ψ) sin ψ dψ, (10.15)
2 0

meaning that, if K(ψ) is given, we can calculate all kn .

Substituting K(ψPQ ) from equation 10.12 yields, with abbreviations
ψ = ψPQ , α = αPQ :
K(ψ)
⏟ ⏞⏞ ⏟
wπ w 2π w 2π w + π/2
2n + 1 1
kn = TP TQ(P) cos ϕ dϕ dλ dα Pn (cos ψ) sin ψ dψ =
2 0 8π2 0 0 − π/2
I
⏟ ⏞⏞ ⏟
w 2π w + π/2 w 2π w π
2n + 1
= TP TQ(P) Pn (cos ψ) sin ψ dψ dα cos ϕ dϕ dλ.
16π2 0 − π/2 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 2π w 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

í Õ ! ¤. û
280 10 Statistical methods

in which Tn,P = Tn (ϕP , λP ) = Tn (ϕ, λ). Tn is the constituent of the

disturbing potential T for the harmonic degree number n, compare the
asteosuus- degree constituent equation 3.9. Substitution yields
yhtälö
w 2π w + π/2
1
kn = TP Tn,P cos ϕ dϕ dλ =
4π 0 − π/2
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.

Write, following equation 4.11, but using the definitions of equation

3.15:

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
)︂
Tn2
⟨︁ ⟩︁
kn = dσ = Tn · Tn σ = δCnm + Snm .
4π σ R
m=0

Here, we have exploited the orthonormality of the fully normalised

kantafunktio basis functions Pnm (sin ϕ) cos mλ and Pnm (sin ϕ) sin mλ on the surface
of unit sphere σ. So

í Õ ! ¤. û
 Propagation of covariances between various quantities 10.8
281
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 between various

quantities
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

í Õ ! ¤. û
282 10 Statistical methods

in which the fn are defined according to degree constituent equation

3.9, 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
n R
L = 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
LnP = rP .
(︁ )︁
Concretely, for the disturbing potential T ϕP , λP , rP in point P, this
means
(︁ )︁ {︁ }︁ ∑︂
∞
n
∑︂
∞ (︂
R n+1 (︁
)︂ )︁
T ϕP , λP , rP = LP T (·, ·, R) = LP Tn,P = r T n ϕ P , λ P .
P
n=2 n=2

The covariance function in space of the disturbing potential T is

{︂ (︁ )︁}︂
K rP , rQ , ψPQ = M ′ T ϕP , λP , rP T ϕQ(P) , λQ(P) , rQ =
(︁ )︁ )︁ (︁
{︂ {︁ }︁ {︁ }︁}︂
′
= M LP T (·, ·, R) LQ(P) T (·, ·, R) =
{︃ ∑︂
∞ ∑︂
∞ }︃
′ n
(︁ n ′ )︁
=M (LP Tn,P ) LQ Tn ′ ,Q(P) =
n=2 n ′ =2
∑︂ ∞ ∑︂
∞
{︁ }︁
n′ ′
= Ln
P LQ M Tn,P Tn ′ ,Q(P) .
n=2 n ′ =2

í Õ ! ¤. û
Propagation of covariances between various quantities 10.8

{︁ }︁
283
Based on orthogonality,10 M ′ Tn,P Tn ′ ,Q(P) = 0 if n ̸= n ′ . So 10

)︁ ∑︂
∞
′
{︁ }︁
Ln n
(︁
K rP , rQ , ψPQ = P LQ M Tn,P Tn,Q(P) . (10.16)
n=2

Now at sea level, all Ln n

P = LQ = 1, so

)︁ ∑︂
∞
{︁ }︁
M ′ Tn,P Tn,Q(P) .
(︁
K ψPQ =
n=2

Comparing with equation 10.14,

(︁ )︁ ∑︂
∞
K ψPQ = kn Pn (cos ψ), (10.14)
n=2
we see that
{︁ }︁
M ′ Tn,P Tn,Q(P) = kn Pn (cos ψPQ ).

10 Likein the proof of the harmonicity of r ∆g in section 8.5, one must take along the
third dimension. Write
{︁ }︁ 1 w 2π {︁ }︁
M ′ Tn,P Tn ′ ,Q(P) = M Tn,P Tn ′ ,Q(P) dαPQ =
2π 0 {︃ }︃
1 w 2π {︂ }︂
= M Tn,P · Tn ′ ,Q(P) dαPQ = M Tn,P Tn⃝′ ,P ,
2π 0

with the definition

def 1 w 2π
Tn⃝′ ,P = Tn ′ ,Q(P) dαPQ .
2π 0
With the radial dimension r:

1 w 2π 1 w 2π (︂ r )︂n ′ +1
Tn⃝′ ,P (r) = Tn ′ ,Q(P) (r) dαPQ = Tn ′ ,Q(P) (R) dαPQ =
2π 0 2π 0 R
(︂ r )︂n ′ +1 1 w 2π (︂ r )︂n ′ +1
= · Tn ′ ,Q(P) (R) dαPQ = Tn⃝′ ,P (R).
R 2π 0 R

This shows that Tn⃝′ ,P (r) is a perfectly legal solid spherical harmonic of degree
n ′ , Tn⃝′ ,P = Tn⃝′ ,P (R) a legal surface harmonic, and the orthogonality of spherical
harmonics applies:
{︁ }︁ {︂ }︂
M ′ Tn,P Tn ′ ,Q(P) = M Tn,P Tn⃝′ ,P = 0 if n ̸= n ′ .

í Õ ! ¤. û
284 10 Statistical methods

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 K(ψ) in equation 10.14, and
for the same reason:

Kψ
(︁ )︁
n ψPQ = kn Pn (cos ψPQ ).

(︁ )︁
At sea level Krn R, R = 1 and the coefficients kn are those given by
equation 10.15.
11 Equation 10.16 now becomes11

)︁ ∑︂
∞
Ln n
(︁ (︁ )︁
K rP , rQ , ψPQ = L
P Q k n Pn cos ψPQ =
n=2
∑︂ (︂
∞
R
)︂n+1 (︂
R n+1
)︂ (︁ )︁
= rP rQ k n Pn cos ψ PQ =
n=2
∑︂
∞ (︃
R2
)︃n+1
(︁ )︁
= rP rQ k n Pn cos ψPQ . (10.17)
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
kn of the corresponding sea-level disturbing potential T (ϕ, λ, R), by
kovarianssien applying propagation of covariances on expansion 10.14 of function K.
kasautuminen Thus we have obtained the three-dimensional covariance function for the
disturbing potential, needed for example in mountainous countries and
in air and space applications.

11 Thisonly works this cleanly because in this case, the operator Ln is of multiplier
(︁ /︁ )︁n+1
type, R r .

í Õ ! ¤. û
 Propagation of covariances between various quantities 10.8
285
^ 10.8.2 Example: the covariance function of gravity anomalies
We know from equation 5.7 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 L∆g :

{︁ }︁ ∑︂
∞
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
(︂ )︂
= Ln T
∆g,P n,P = rP rP Tn,P .
n=2 n=2

We can show in the same way as above that

{︁ }︁ {︁ }︁
Cov ∆gP , ∆gQ = M ′ ∆gP ∆gQ(P) =
∑︂
∞
{︁ }︁
′
= Ln Ln
∆g,P ∆g,Q M T n,P T n,Q(P) =
n=2
∑︂
∞
n − 1 R n+1 n − 1 R n+1
(︂ )︂ (︂ )︂ (︁ )︁
= rP rP rQ rQ k n Pn cos ψ PQ =
n=2
∑︂
∞ (︃
R2
)︃n+2 (︂
n−1 2
)︂ (︁ )︁
= rP rQ k n Pn cos ψPQ .
R
n=2

Often, we write
)︁ def {︁ }︁ {︁ }︁
C rP , rQ , ψPQ = Cov ∆gP , ∆gQ = M ′ ∆gP ∆gQ(P) =
(︁

∑︂∞ (︃
R2
)︃n+2
(︁ )︁
= rP rQ cn P n cos ψPQ ,
n=2

í Õ ! ¤. û
286 10 Statistical methods

in which the degree variances of gravity anomalies are

)︂2
n−1
(︂
cn = kn .
R
Similarly we also calculate the “mixed covariances” between disturbing
potential and gravity anomaly:
{︁ }︁
Cov TP , ∆gQ =
{︁′
}︁ ∑︂
∞
′
{︁ }︁
= M TP ∆gQ(P) = Ln n
P L∆g,Q M Tn,P Tn,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

series expansion:
{︂ {︁ }︁ {︁ }︁}︂ ∑︂ {︁ }︁
′
Cov L1 T P , L2 T Q = Ln n
1,P L2,Q M Tn,P Tn,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 T Q = 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 coefficients found from the expansion. These coefficients
are indicated above by red and blue colouring.

í Õ ! ¤. û
 Global covariance functions 10.9
287
^ 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.
They are commonly given in the form of a degree variance formula. The
best known is the rule observed by William Kaula (Rapp, 1989):12 12

2n + 1
kn = α .
n4
By writing
n−1 2
(︂)︂
cn = kn ,
R
in which cn are the degree variances of gravity anomalies, we obtain
2n + 1 n − 1 2 2α
(︂ )︂
cn = α ≈ .
n4 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 Bjerhammar13 sphere inside the Earth (R is the Earth’s 13
mean radius). The form of the equation is chosen so the covariance
functions of various quantities will be closed expressions.

12 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.
13 Arne Bjerhammar (1917–2011) was a Swedish geodesist.

í Õ ! ¤. û
 288 10 Statistical methods

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.

^ 10.10 Collocation and the spectral viewpoint

The calculations in least-squares collocation can also be executed effi-
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 when collo-
cation equations depend only on differences in longitude ∆λ between
points, not on absolute longitudes λ.
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 the
def
edge of a circle, in points ψi = 2π i N , i = 0, 1, 2, . . . , N − 1. Assume
/︁

í Õ ! ¤. û
Collocation and the spectral viewpoint 10.10
289
that also the results of the calculation, estimates fˆ︁i of the result function
f(ψ), are desired at the same points. Then we have equation 10.9:

ˆ︁f = Cfg (Cgg + Dgg )−1 g + n ,

(︁ )︁
(10.9)

with
[︁ ]︁ (︁ )︁ (︁ )︁
Cfg = Cfg f(ψi ), g(ψj ) = Cfg ψi , ψj ,
ij
[︁ ]︁ (︁ )︁ (︁ )︁
Cgg ij = Cgg g(ψi ), g(ψj ) = Cgg ψi , ψj ,
[︁ ]︁ (︁ )︁ (︁ )︁
Dgg ij = Dgg g(ψi ), g(ψj ) = Dgg ψi , ψj .

If the physics of the whole situation, including the physics of the

measurement process, is rotationally symmetric, we must have

{︂ )︁}︂ 1 ∑︂
N−1
[︁ ]︁ (︁ (︁ )︁ def [︁ ]︁
Cfg = M⃝ f(ψi ) g ψj(i) = f(ψi ) g ψj(i) = Cfg k ,
i,j(i) N
i=0

with j(i) = (i + k) mod N. Here, the operator M⃝ is the “circle mean”

of a function,
∑︂
{︁ }︁ def 1 N−1
M⃝ h = h(ψi ),
N
i=0

which, like the geographic average in section 10.4, replaces the statistical
average.
In the same way we obtain
[︁ ]︁ {︂ (︁ )︁}︂
Cgg i,j(i)
= M⃝ g(ψi ) g ψj(i) =

1 ∑︂
N−1
(︁ )︁ def [︁ ]︁
= g(ψi ) g ψj(i) = Cgg k .
N
i=0

Now Cfg , Cgg are functions of only k, and we may write them
[︁ ]︁ (︁ )︁ [︁ ]︁
Cfg
ij
= C fg ψ i , ψj = C fg (∆ψk ) = Cfg k ,
[︁ ]︁ (︁ )︁ [︁ ]︁
Cgg ij = Cgg ψi , ψj = Cgg (∆ψk ) = Cgg k ,

í Õ ! ¤. û
290 10 Statistical methods

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 ij
= Dgg ψi , ψj = Dgg (∆ψk ) = Dgg k = E ni nj(i) ,

the traditional statistical variance of the observation noise. Also because

14 generally the observations do not correlate with each other, we have14

Dgg = σ2 IN ,

σ2 (the variance of observations, assumed equal for all) times the N × N

sized unit matrix.
Toeplitz- Matrices of this form are called Toeplitz circulant matrices.15 Thanks
sirkulantti to this property, equation 10.9 is a string of convolutions.
matriisi
15
14 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.
15 OttoToeplitz (1881–1940) was a German Jewish mathematician who contributed to
functional analysis.

í Õ ! ¤. û
 Self-test questions
291
Without proof we present that the spectral version of equation 10.9
looks like this:
{︁ }︁ {︂ }︂ {︁ }︁ {︂ }︂
{︁ }︁ F C fg F Cfg
f = {︁ }︁
F ˆ︁ {︁ }︁ F g + n = {︁ }︁ F g + n .
F Cgg + F Dgg F Cgg + σ N
2/︁

This is an easy and fast way to calculate the solution using FFT. If for a
{︁ }︁
suitable operator L we have f = L g , the equation becomes
{︁ }︁ {︁ }︁ {︂ }︂
{︁ }︁ 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 =F L ·F g
F ˆ︁ ⇐⇒ ˆ︁f = L g .

For example, if g are gravity anomalies and f are values of the disturbing
potential, then16 16
{︁ }︁ 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?
2. What is a functional?
16 In
real computation though, the equation must be changed to use, instead of the
harmonic degree number n, which refers to global spherical geometry, the Fourier
wave number expressed on the computation grid used.

í Õ ! ¤. û
292 10 Statistical methods

3. What is a linear functional?

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
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.
8. Explain degree variances. What is the difference between degree
variances kn and cn ?
9. What does Kaula’s rule express?
10. What is a Toeplitz circulant matrix?

^ Exercise 10 – 1: Variance of prediction

The equation for the variance of prediction at a point P is

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 .

í Õ ! ¤. û
 Exercise 10 –2: Hirvonen’s covariance equation and prediction
293
^ Exercise 10 – 2: Hirvonen’s covariance equation and
prediction
Hirvonen’s covariance equation is
C0
C(s) = (︁ /︁ )︁2 , (10.18)
1+ s d
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
{︂ }︂
{︂ }︂ Cov ∆gP
, ∆gQ
def
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
which is at a distance sPQ = 10 km from the (precisely!) given
anomaly in point P? Apply equation 10.11 as follows:

ΣQQ = 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

í Õ ! ¤. û
294 10 Statistical methods

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

where ℓj = ∆gj + nj is the (abstract) vector of gravity anomaly observa-

tions, ⎡ {︂ }︂ {︂ }︂ ⎤
Var ∆g1 Cov ∆g1 , ∆g2
Cij = ⎣ {︂ }︂ {︂ }︂ ⎦
Cov ∆g1 , ∆g2 Var ∆g2
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 n1 Cov n1 , n2
Dij = {︁ }︁ {︁ }︁ .
Cov n1 , n2 Var n2

1. Compute the matrix Cij , assuming again Hirvonen’s covariance

equation 10.18 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.
4. Compute the variance of prediction (note CjP = CTPi ) using

ΣPP = 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

í Õ ! ¤. û
 Exercise 10 –5: Propagation of covariances
295
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.
1. Compute Cij , CPi , ∆g
ˆ︂ P and ΣPP .
2. Compare the result with the previous one. Conclusion?

^ Exercise 10 – 5: Propagation of covariances

Given the covariance function 10.17 of the disturbing potential

{︁ }︁ ∑︂
∞ (︃ )︃n+1
R2 (︁ )︁
Cov T P , T Q = r r k n Pn cos ψPQ .
P Q
n=2

1. Derive 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. Derive the covariance function of the gravity-gradient disturbance

∂2 ∂
2
T = − δg,
∂r ∂r
in other words, 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)
n=2

í Õ ! ¤. û
296 10 Statistical methods

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

2n + 1
kn = α .
n4
From these one can derive, using propagation of variances, the degree
variances of gravity anomalies

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

as follows:
n−1 2 2α
)︁2 (︂ )︂
cn = Ln
(︁
∆g (R) k n = kn ≈ .
R nR2
Analogously, differentiate expansion 5.7 for the gravity anomaly

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

yielding
∂ ∆g ∑︂ (n − 1) (n + 2) (︂ R )︂n+1
∞
=− r Tn ,
∂r r2
n=2

the connection between the disturbing potential and the anomalous

gravity gradient in the spectral domain.
On the Earth’s surface r = R:
⃓
∂ ∆g ⃓⃓ ∑︂∞
(n − 1) (n + 2) def ∑︂ n
∞
=− Tn = L∆g ′ (R) Tn (ϕ, λ)
∂r ⃓r=R R2
n=2 n=2

with
(n − 1) (n + 2)
Ln
∆g ′ (R) = − .
R2

í Õ ! ¤. û
 Exercise 10 –7: Underground mass points
297
1. Derive an approximate equation for the degree variances for the
anomalous gravity gradient. Designate them with the symbol cn′ ,
in an analogue fashion as above for the gravity anomaly degree
variances cn :
cn′ = x(n) · kn ≈ y · nz .
Find the expression x(n) and the constants y and z for the case of
the Earth.
2. Conclusion?

^ 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
change in ambient gravity g it causes on the Earth’s surface, for
which C(s) = 12 C0 ?
2. Thus, if we wish to build a model made of mass points, in which
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 1
built in 1657 the first useable pendulum clock based on this principle
(Wikipedia, Pendulum clock).
When the young French researcher Jean Richer2 visited French 2
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 FRS (1629–1695) was a leading Dutch natural scientist and
mathematician. 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.

– 299 –
300 11 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
3
more precise special devices were built, for example Kater’s3 reversion
4 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 a British physicist who made contributions to
scientific instruments and metrology.
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 11.2
301

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 pivot, 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 measurement values obtained with
pendulum apparatus in the so-called Potsdam system are systematically
14 mGal too high.

^ 11.2 The relative or spring gravimeter

A spring gravimeter is at its simplest the same as a spring balance.

í Õ ! ¤. û
302 11 Gravimetric measurement devices

In a linear spring balance the equation of motion of the test mass is

(︃ )︃
d2 ℓ
m − g = −k (ℓ − ℓ0 ) , (11.1)
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 , (11.2)
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 equilib-
värähtely- rium position. The oscillation equation, obtained by summing equations
yhtälö 11.1 and 11.2, is
d2 (︁ )︁ k (︁ )︁
ℓ − ℓ = − m ℓ − ℓ .
dt2
The period is
√︃ √︃
m ℓ − ℓ0 δℓ
√︂
P = 2π = 2π g = 2π g , (11.3)
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
11.2 in the form
(︁ )︁
mg = k ℓ − ℓ0 = k δℓ
with the result
dℓ d (δℓ) m P2
= = = 2. (11.4)
dg dg k 4π
Substitution of, for example, δℓ = 5 cm and g = 10 m/s2 into equation
11.3 yields P = 0.44 s. One milligal of change in gravity g produces,

í Õ ! ¤. û
 The relative or spring gravimeter 11.2
303
according to equation 11.4, a lengthening of only 5 · 10−8 m = 50 nm
(check!), one-twelfth the wavelength of a helium-neon laser. 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 length without loading, the rest length.
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. (11.5)

í Õ ! ¤. û
304 11 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 a spring gravimeter. On the right, how to

^ build a “zero-length spring”.

By differentiation

mpℓ dg + mgp dℓ − kbc dℓ = 0,

from which is obtained, by substituting equation 11.5, a sensitivity

equation:

dℓ mpℓ mpℓ ℓ ℓ − ℓ0
=− =− )︁ = g ℓ0 .
dg mgp − kbc
/︂(︁
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 11.2
305
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 5
ratio is precisely ℓ − ℓ0 ℓ0 .
/︁

This is the operating principle of an astatised gravimeter, like the

LaCoste-Romberg.6 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 11.5:
(︁ )︁
mgpℓ − kbc ℓ − ℓ0 = 0, (11.6)

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.

Use equation 11.6 to eliminate kbc, yielding

(︁ )︁ ℓ (︁ )︁
mgp ℓ − ℓ + mapℓ − mgp ℓ − ℓ = 0.
ℓ − ℓ0

5 For
comparability we should still multiply by p b sin β , if the position of the test
/︁

mass is measured.
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
306 11 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 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
7 no material that is perfectly elastic, but it is always plastic7 as well, the

í Õ ! ¤. û
 The relative or spring gravimeter 11.2
307
δ(ε)

ε)
α F(ε) cos(α + δ + ε)

F(
−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 almost 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

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
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
308 11 Gravimetric measurement devices

the values obtained from the measurement in proportion to their

times of measurement.
◦ 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.
ilmagravimetri 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
contains 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
8 of Colorado at Boulder by Jim Faller’s group,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 main

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 11.3
309
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.

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 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 LaCoste-Romberg relative gravimeters are capable of.

í Õ ! ¤. û
310 11 Gravimetric measurement devices

Figure 11.6. Absolute gravimeter of type FG5. Photograph United States

^ National Oceanic and Atmospheric Administration (NOAA).

The instrument is however large and, although transportable, cannot

be called a field instrument. Of late, development has gone in the
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, z = z0 + v0 t + 12 gt2 ,
dt
from which we obtain the observation equations of the measurement

í Õ ! ¤. û
The absolute or ballistic gravimeter 11.3
311
process ⎡ ⎤
[︂ ]︂ z 0
1 2
zi = 1 ti t · ⎣ v0 ⎦ + ni .
⎢ ⎥
2 i
g
Here, the unknowns9 to be estimated are z0 , v0 and g. The quantities 9
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 a moment close to that of release of the prism is of course essential.
The number of measurement values obtained from each individual
drop is substantial.
We write the observation equations in matric form:

ℓ = A x + n,

in which
⎡ ⎤ ⎡ ⎤ ⎡ 1 2
⎤
z1 n1 1 t1 t
2 1
⎡ ⎤
1 2 z0
⎢ z2 n2 1 t2 t
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
2 2
⎥ ⎢ ⎥ ⎢ ⎥
ℓ=⎢ ⎥, n = ⎢ ⎥, A = ⎢ ⎥, x = ⎢
⎣ v0 ⎦ .
⎥
⎢ .. .. .. .. ..
⎣ . . . . .
⎥ ⎢ ⎥ ⎢ ⎥
⎦ ⎣ ⎦ ⎣
1 2
⎦ g
zn nn 1 tn t
2 n

The solution follows from this according to the method of least-squares pienimmän
adjustment, from the normal equations neliösumman
tasoitus
T T
x = A ℓ,
A A ˆ︁

giving the solution, or 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 ,

9 It
would be easy (exercise!) to add an unknown representing the vertical gradient of
gravity to this. Whether a useable value for this unknown can be obtained from the
measurements is a good question.

í Õ ! ¤. û
312 11 Gravimetric measurement devices

in which σ is the uncertainty (mean error) of a single observation zi ,

painoyksikön also known as the mean error of unit weight.
keskivirhe
An alternative type of absolute gravimeter throws the prism up (inside
the tube), after which it moves along a symmetric path. An 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 by extreme cooling. The condensate
consists of perhaps a million atoms in identical quantum states, with
their phase angles in lockstep like marching soldiers.
The interaction between laser light and the atoms is based on the Ra-
man effect, an inelastic scattering process, in which the atoms exchange
both energy and momentum with photons in the laser beam, while
switching from one quantum state to another. The scattering involves
two photons: the transition is “forbidden”, with a very precisely defined
energy and momentum change.
The condensate is dropped, and the first laser pulse splits it into
10 two. Half of the atoms10 fall first slowly, then faster. The other half

fall fast at first and then slower. To bring this about, a second laser
pulse pair is used that acts like a mirror, or perhaps a tennis racket. The
third and last laser pulse recombines the beams. Then, constructive or
destructive interference is observed using a fluorescence detector. The

10 Thisis a quantum theoretically erroneous statement. The matter wave of each

individual atom splits into two! Wikipedia, Double-slit experiment.

í Õ ! ¤. û
 Network hierarchy in gravimetry 11.4
313
phase difference between the two arms of the interferometer is inferred
from these observations.
As the atoms travel through space-time along two different paths on
which the gravity potentials are different,11 a phase difference is formed 11
between these, which in principle can be measured. See figure 11.7, in
which the horizontal axis is time. Without gravity (dashed lines) this
phase difference would be zero.
Like in all (non-kinematic) interferometric methods, the ambiguity
problem — the circumstance that a measured phase is always in the
[︁ )︁
interval 0, 2π , although a phase change or difference may contain
any number of whole cycles — poses its own challenge. Ambiguity
resolution is possible by measuring at several different inter-pulse time
intervals T , figure 11.7.

^ 11.4 Network hierarchy in gravimetry

Network hierarchy is just as important in gravimetry as in measurements
of location or height. The procedure has typically been that the highest
measurement order consisted of points measured by absolute gravime-
ters — in the old days, this meant pendulum measurements. Stepwise
densification of this highest-order network, base network measurement,
was subsequenty done with relative or spring gravimeters, like also
the lowest-order measurements or gravity mapping surveys. In base
network measurement, fast transportation was used, such as aircraft,
and national or regional reference points were often located at airports. vertauspiste
Because pendulum instruments are not sufficiently precisely 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

11 In
fact, the spinning of the phase angle of the atom’s wave function acts like a clock,
and the speed at which time elapses depends on the local geopotential (Vermeer,
1983a).

í Õ ! ¤. û
314 11 Gravimetric measurement devices

t
T T

Recombination
beam
z
Splitting
beam

Mirroring
beam

g Without gravity
In gravity field

Figure 11.7. The idea of operation of an atomic or quantum gravimeter. The

^ horizontal axis is time.

by no means non-existent, in the order of microgals. As there are

no better, meaning more absolute, instruments than these, the issue
cannot ultimately be resolved. For this reason, 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 levitat-
ing on a magnetic field. The precise place of the sphere is measured elec-
tronically. Because a superconducting material is impenetrable by a mag-
netic field, the sphere will remain forever in the same spot inside the field:

í Õ ! ¤. û
 The superconducting gravimeter 11.5
315

Figure 11.8. International intercomparison of absolute gravimeters in Walfer-

^ dange, Luxembourg. Image courtesy © Olivier Francis.

the Meissner effect. Of course the field itself must be constant. It is gen-
erated by superconducting solenoids inside a vessel made of mu-metal käämi
(Wikipedia, Mu-metal), which keeps out the Earth’s magnetic field.
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.

í Õ ! ¤. û
316 11 Gravimetric measurement devices

Upper capacitor plate

Middle capac-
itor plate

Levitation coil
Feedback
coil

Lower capacitor plate

Levitation coil

Figure 11.9. Operating principle of a superconducting gravimeter. Reading

^ out the sphere position is done capacitively.

The most important property of a superconducting gravimeter is,

12 in addition to its superior precision,12 its stability, the absence of any
drift. For this reason, it is extremely suited to monitoring long-period
phenomena, like the free oscillations of the solid Earth after large
13 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 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

12 Virtanen(2006) reports how the instrument at Metsähovi detected the change in

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
 Gravity measurement and the atmosphere 11.6
317
remotely controllable superconducting gravimeters, for example the
GWR iGrav® , weighing 30 kg and not expending 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 Gravity measurement and the atmosphere

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
at least the pressure effect go away. In practice it is easier to
calibrate the instrument in the laboratory and calculate a correction
to the field measurements based on the calibration.
◦ 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 using a Bouguer-
plate approximation: if the air pressure is p, then the surface 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 14
of the Bouguer plate is

2πGκ = 0.43 mGal (11.7)

14 Soyes, the force acting on a standard 14-inch laptop screen (aspect ratio 16 : 9) is
547 kg. . . but it doesn’t matter as it is not an old-fashioned vacuum cathode-ray tube.

í Õ ! ¤. û
318 11 Gravimetric measurement devices

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, local 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.
Remember that the reference or normal gravity field of GRS80 is
defined in such a way that the parameter GM⊕ refers to the entire
mass of the Earth, including the atmosphere, the Earth’s attraction
as satellites are sensing 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 mea-
surement 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
2
= γ ,
R

í Õ ! ¤. û
 Airborne gravimetry and GNSS 11.7
319
twice the Bouguer-plate atmospheric reduction 11.7 calculated
above. This is the value that should be added to the measured
gravity values.
One may also 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 planar 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, changed airborne gravimetry ilma-
from a difficult technology to something completely operational. To gravimetria
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
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 sensed in a reference frame connected vertauskehys
to the solid Earth and rotating with it
2. the pseudo-forces caused by the inevitable accelerations of the
aircraft even in cruise flight.

í Õ ! ¤. û
320 11 Gravimetric measurement devices

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∗i+1 + x∗i−1 − 2x∗i
a∗i ≈ . (11.8)
∆t2
When the acceleration measured by the gravimeter is g˜︁ and the direction
luotiviiva of the local plumb line (upwards) n, local gravity g follows:
⟨︂(︁ )︁ ⟩︂
∗
˜︁ − a∗ · n − ω2⊕ N(φ) cos φ,
⟨︁ ⟩︁
g=g ˜︁ − a + fω · n = g

in which fω = ω2⊕ (X i + Y j) is the centrifugal acceleration of Earth

rotation, equation 4.1. N(φ) is the transversal radius of curvature of
the Earth ellipsoid, equation 2.6.
If working in a co-rotating (Earth centred, Earth fixed) frame, the
centrifugal acceleration term should be omitted. However, then a
Coriolis acceleration due to the interaction of aircraft velocity v with
Earth rotation needs to be introduced instead. This acceleration term is
′
⟨︁ ⟩︁
fω = −2 ω⊕ × v = 2ω⊕ (vY i − vX j) ,

and we obtain
⟨︂(︁ )︁ ⟩︂
⊕ ′
˜︁ − a⊕ · n + 2ω⊕ veast cos φ .
⟨︁ ⟩︁
˜︁ − a − fω · n = g
g=g

The choice of the time constant ∆t is critical in this method. It is best

to choose it as long as possible, as then, the precision of the 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.

í Õ ! ¤. û
Airborne gravimetry and GNSS 11.7
321
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 11.8 the uncertainty of the
vertical acceleration is √
σz 6
σa = .
∆t2
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!
The flight height H is chosen 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.
Since then, other large uninhabited areas in the Arctic and Antarctic
regions have also been 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 example, for sea areas near the coast or inside archipelagos,
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).

í Õ ! ¤. û
322 11 Gravimetric measurement devices

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 is similar, see section
13.7. An essential difference is, however, that the instrumentation on
the satellite is in a state of weightlessness: g
˜︁ = 0 (in a high orbit, or
ilmanvastus when using an air drag compensation mechanism), or g ˜︁ is small and
kiihtyvyys- is measured using a sensitive accelerometer (in a low orbit, where air
mittari drag is significant).

The greatest challenge in planning a satellite gravity mission is

choosing the flight height. The lowest possible height is some 200 km.
ajoaine At that height, a tankload of propellant is already needed, or the flight
will not 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.
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 ∂y2 ∂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

í Õ ! ¤. û
Measuring the gravity gradient 11.8
323
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

the vertical free-air gravity gradient: Newton’s law gives for a spherical painovoiman
Earth ilmagradientti
GM
gz = − .
(R + z)2
The minus sign is because g points downwards while the z co-ordinate
increases going up. Differentiation gives

∂ GM ∂ (R + z) 2gz
g =2 · =− ≈
∂z z (R + z) 3 ∂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 = = −ρ , ∂yy W = = −ρ ,
∂x2 x ∂y2 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 .

/︁

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.

í Õ ! ¤. û
324 11 Gravimetric measurement devices

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 11.4 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?
3. By which method choices do we, in practical measurements, take
the drift of a relative gravimeter into account?
4. Why were, before the advent of absolute gravimeters, the reference
points of international fundamental gravimetric networks often
on airports?
5. In an absolute or ballistic gravimeter, what is the role of:

í Õ ! ¤. û
 Exercise 11 –1: Absolute gravimeter
325
(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 used GPS widely.
◦ By December 1993, GPS achieved initial operational capabil-
ity (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 .

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 accurately should the
laser interferometer measure the falling distance of the prism?
Feel free to choose the method of analysis to be used: analytical,
numerical, . . . . Consider that you are in a purchasing situation
when building an absolute gravimeter. A ballpark estimate is
good enough!
3. The same question for the measurement accuracy of the falling
time.

í Õ ! ¤. û
326 11 Gravimetric measurement devices

^ 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 spring gravimeter always has a drift: the connection
between measured value and true value changes slowly over time. In
a non-factory-fresh gravimeter, this drift is however very regular and
almost linear.
Question How is the behaviour of a spring gravimeter, especially 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?

^ Exercise 11 – 3: Air pressure and gravity

1. How much does a low-pressure zone of 100 hPa — meaning
that the air pressure is 100 hPa lower than average air pressure
1013.25 hPa — affect gravity measured on the Earth’s surface?
Assume the low-pressure zone to be of great areal extent, as
low-pressure zones are.

í Õ ! ¤. û
Exercise 11 –3: Air pressure and gravity
327
2. How much does sea water rise due to the “inverted barometer ylösalainen
effect” under a low-pressure zone? ilmapuntari

3. To how much does the effect from point 2 amount in local gravity
measured on a ship? Assume that you are on the open sea, that
the free-air vertical gravity gradient is −0.3 mGal/m, and that the
density of sea water is 1030 kg/m3 . Analyse the situation carefully.15 15

15 Meaning 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 makea vesi
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 place to place
◦ 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

– 329 –
 330 12 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.
suolaisuus ◦ 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, the permanent separation between the mean sea
surface and the geoid. See figure 12.4.
The classical definition of the geoid is

“The level surface of the Earth’s gravity field that agrees on average best
with 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 are the time series that are
mareografi available, as tide gauges have been widely operating already for about

í Õ ! ¤. û
 Geoid models and national height datums 12.2
331
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.

^ 12.2 Geoid models 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
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

í Õ ! ¤. û
332 12 The geoid, mean sea level, and sea-surface topography

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 “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. Calculate a gravimetric geoid model by using the Stokes inte-
poistamis- gral equation and remove–restore, for example using the FFT
entistämis- calculation technique.
menetelmä
2. 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 a, b,

c, . . . from the differences between the two heights in these known
pienimmän comparison points by using the least-squares method.
neliösumman
menetelmä
í Õ ! ¤. û
 The geoid and post-glacial land uplift 12.3
333
^ 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. In particular the ongoing isostatic response of
the solid Earth to the ending of the last ice age, 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 — or equivalently, relative to the
surface of a geocentric reference ellipsoid such as GRS80. This
land uplift is measured when using satellites the orbits of which
are determined in a co-ordinate reference system tied to the vertaus-
Earth’s centre of mass, for example, positioning of tide gauges by järjestelmä
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
internal to the Earth from one place to another, it is clear that

í Õ ! ¤. û
334 12 The geoid, mean sea level, and sea-surface topography

the geoid must also change. The geoid rise is, however, small
compared to the land uplift, only a few percent of it.

Equation (the dot above a quantity denotes the derivative with respect
1 to time1 ):
̇ = ḣ − N
H ̇ =H ̇r+H ̇e+H ̇ t,

in which
Ḣ relative land uplift from the geoid
ḣ absolute land uplift from the reference ellipsoid
̇r
H relative land uplift from the local mean sea level
̇e
H eustatic (global mean sea level) rise
̇t
H change over time of the sea-surface topography (likely small)
̇
N geoid rise from the reference ellipsoid.

The rise of the geoid as a result of land uplift can be simply calculated
with the Stokes integral 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, filling
it. This matter may be roughly modelled as a Bouguer plate.

1 This dot notation, fluxion, was introduced by Newton.

í Õ ! ¤. û
 The geoid and post-glacial land uplift 12.3
335

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.

◦ 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
standard deviation), which would seem to settle the issue. It looks like
the Bouguer hypothesis is closer to physical reality. The flow of mass is

í Õ ! ¤. û
336 12 The geoid, mean sea level, and sea-surface topography

5◦ ◦
65 ◦ ◦
35 ◦
10 ◦ ◦ 30 65
15◦ 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.

assumed to happen 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.

^ 12.4 Determining the sea-surface topography

In principle three geodetic methods exist:
◦ satellite radar altimetry and gravimetric geoid determination

í Õ ! ¤. û
 Global sea-surface topography and heat transport 12.5
337
◦ positioning of tide gauges along the coast using GNSS, together
with gravimetric geoid determination
◦ precise levelling along the coast connecting tide gauges.
In addition to this, we still have the oceanographic method: 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 if 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 caused by storms resembling
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
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 Sun’s
energy from the equator to higher latitudes. There are many other
things that a better knowledge of sea currents would help to explore:
for example carbon dioxide dissolved into the water, chlorophyll (phy-
toplankton), and salinity.

í Õ ! ¤. û
338 12 The geoid, mean sea level, and sea-surface topography

The Coriolis force, or acceleration, caused by the Earth’s rotation is

′
⟨︁ ⟩︁
fω = −2 ω⊕ × v , (12.1)

in which v is the velocity vector of a freely moving particle 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 amounts to ω⊕ · n = ω⊕ sin φ, or as a vector,

def ⟨︁ ⟩︁
ω⊕ = ω⊕ · n n = (ω⊕ sin φ) n.

Now, the Coriolis vector projected onto the horizontal plane is

′ def
⟨︁ ⟩︁ ⟨︁ ⟩︁
fω = −2 ω⊕ × v = −2ω⊕ sin φ n × v ,

the length of which is the scalar

′
fω ′
= ∥fω ∥ = 2v ω⊕ sin |φ| .

Here, v = ∥v∥ and 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, and 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
′
fω ω⊕
γ = 2v γ sin |φ| .

Here, γ is local gravity. This equilibrium between Coriolis force and

gravity 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.

í Õ ! ¤. û
Global sea-surface topography and heat transport 12.5
339
For example, in the case of the Gulf Stream, the height changes
caused by this effect are 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
may write for the sea-surface topography H the geostrophic equations
∂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
the flow velocity vector field2 2

[︂ ]︂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 world ocean. The GOCE satellite fits this need like a
glove, see subsection 13.7.3. 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

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
340 12 The geoid, mean sea level, and sea-surface topography

Figure 12.3. Connection between sea-surface topography and ocean currents.

^ Arrows depict ocean currents, curves, sea-surface topography.

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
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
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 mean

í Õ ! ¤. û
 The global behaviour of the sea level 12.6
341

Figure 12.4. Sea-surface topography map produced by GOCE. Base map ©

European Space Agency (ESA). Unit: cm. Ocean currents super-
^ imposed: NOAA / Rick Lumpkin (NOAA, Ocean currents).

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 3
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,

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
342 12 The geoid, mean sea level, and sea-surface topography

vajoaminen a subsidence of the land is taking place at an annual rate of 0.5–1.7

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.
In order to complicate the picture, the sea-level rise caused by the
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).
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.
merenpohjan The subsidence of the sea floor has not been globally uniform: at
vajoaminen 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 remain, 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,
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,

í Õ ! ¤. û
 The sea-level equation 12.7
343
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.

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
relative to the solid Earth’s surface: they are changes in sea depth.
S is also what tide gauges measure.
◦ 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 A o
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.

í Õ ! ¤. û
344 12 The geoid, mean sea level, and sea-surface topography

◦ 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
{︁ }︁
Gs ⊗o S (ω, t) =
wt x
Gs ψ(ω, ω ′ ), (t − t ′ ) S(ω ′ , t ′ ) dω ′ dt ′ , (12.4)
(︁ )︁
=
−∞ 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), (12.5)

def
in which ∆t = t − t ′ ⩾ 0.
This equation expresses simply that the sea depth S is the distance
between sea surface and sea floor, and that a change in it is the
difference between the vertical displacements of those two: that
of the sea surface caused by a change in potential V, and that of
the sea floor, meaning a change in local radius r.
Here the Green’s function of the geopotential is

GV (ψ, ∆t) = GrV (ψ, ∆t) + GeV (ψ, ∆t) + GvV (ψ, ∆t),
/︂
r
The function GV (ψ, ∆t) = δ(∆t) 2 sin 12 ψ is the rigid partial
(︁ )︁

Green’s function, representing the change in potential caused by

a mass of water or ice before any deformation takes place.

í Õ ! ¤. û
 The sea-level equation 12.7
345

^ Figure 12.6. Sea-level rise after the last ice age (Rohde, 2005).

The functions GeV and GvV are the elastic and viscous deformation
partial Green’s functions of the geopotential. These thus charac-
terise the rheological behaviour of the Earth, and their theoretical
calculation requires the internal density and viscosity distribu-
tions 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.
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 vertical difference

í Õ ! ¤. û
346 12 The geoid, mean sea level, and sea-surface topography

in location between 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. And also 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,
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 similar elastic Love numbers as
those that appear in the theory of tidal deformation, for long (though
geologically short) periods P. 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 also appearing in equations 14.4. δ(∆t) is Dirac’s delta

function.
GIA,
however, is viscous deformation on a range of geological time-
scales. Equation 12.5 becomes
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

í Õ ! ¤. û
 Self-test questions
347
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 = 1, . . . , I counts the

viscous relaxation modes for every degree number n. The number of
different modes I is in practice a handful, each related to a different
discontinuity surface in the Earth’s density and viscosity model used.
k /︁ h /︁
The ratios rni sni and rni sni are called “modal strengths”, and the
def /︁
τni = 1 sni are relaxation times in which the mode in question will
decay over time.
Generally, uplift patterns that are of large spatial extent — low degree
numbers n — decay slower, when again the local patterns — high
degree numbers — tend to decay faster. The local uplift patterns of
the last deglaciation have today already vanished: the Fennoscandian
land uplift is already geographically 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 uplift patterns 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?
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?

í Õ ! ¤. û
348 12 The geoid, mean sea level, and sea-surface topography

6. Which two main models are on offer for the mechanism of land
uplift?
7. Which 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?
14. What does the sea-level equation describe?
15. How would sea-level equation 12.3 change if the convolution inte-
grals like equation 12.4 were over the unit sphere dσ = cos ϕ dϕ dλ
instead of over dω = R2 cos ϕ dϕ dλ?
16. Why does the mean sea level in the Baltic Sea not rise when the
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?
Which edge is higher? Assume that the current is at latitude 45◦
north.
2. If the same current were 200 km broad and the velocity of flow
0.05 m/s (so, if the same depth is assumed, the amount of water
transported would also be the same), compute for that case the

í Õ ! ¤. û
 Exercise 12 –2: Land subsidence and the mechanism of land uplift
349
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
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 1
was the first large satellite altimetry data set used for determining
the global 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 ocean, more
precisely the deflections of the plumb line, which are needed luotiviivan
to impart the correct departure direction to ballistic missiles poikkeama

1 But read this: Wikipedia, Seasat conspiracy theory.

– 351 –
 352 13 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, 17.07 0.08
Geosat 1985 108.05 786 3, 17.07 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 17.07 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
Haiyang-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

launched from submarines. The 17-day repeat data from the

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 instrument 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-
tioning device, which allowed the determination of the location

í Õ ! ¤. û
Satellite altimetry 13.1
353
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
information 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 2
TOPEX/Poseidon in 2002 as “the most successful ocean experiment
of all time” (Munk, 2002).
◦ Haiyang-2A (2011-043A) is a Chinese satellite also launched by
China.
◦ SARAL (2013-009A) is a satellite launched by India. The altimeter
AltiKa 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. synteettinen
aukko
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.
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 physical oceanographer.

í Õ ! ¤. û
354 13 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. Data © Colorado University at Boulder’s Sea Level
Research Group; Nerem et al. (2010). Comparison with ENSO (“El
Niño”), SOI = Southern Oscillation Index, Climate Research Unit,
^ Climate Research Unit; Ropelewski and Jones (1987).

— 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

í Õ ! ¤. û
 Satellite altimetry 13.1
355
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.

pulses on board the satellite, averaging them down to a measurement

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 return 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 merkitsevä
phase, one should make a careful correction so no bias is created: if aallonkorkeus

í Õ ! ¤. û
356 13 Satellite altimetry and satellite gravity missions

the SWH is large, the altimeter footprint — the area on the sea from
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
1. the height of sea waves (SWH)
2. solid-Earth tides
3. ocean tides
4. the “wet” tropospheric propagation delay, best derived from
measurements with a downlooking water vapour radiometer on
the satellite, otherwise from meteorological modelling
5. the “dry” tropospheric propagation delay
6. the ionospheric delay, only for the part of the ionosphere below
the satellite, depending on flight height
7. the altimeter’s own calibration correction — nowadays “in-flight”
calibration is always strived for, using an ensemble of GNSS-
mareografi positioned tide gauges, see section 13.4.
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 13.2
357
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
(including the known corrections 1–7 in the previous section), 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: the
permanent deviation of the sea surface from an equipotential surface,
∆h is the orbit-error correction, ϵ is the residual variation of the 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 risteyskohta-
the complication that sea-surface residual variation and orbit corrections tasoitus
appear 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 and 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 figure 13.3 we have three tracks and two crossing points. The
observation equations, which describe the discrepancies in the known

í Õ ! ¤. û
358 13 Satellite altimetry and satellite gravity missions

∆h2 ∆h3

∆h2 − ∆h3
∆h1
∆h1 − ∆h3

Crossover 1

Crossover 2

2
1

^ Figure 13.3. A simple crossover geometry.

crossover points as functions of the orbit corrections, are

ℓ1 = ∆h2 − ∆h3 + n1 ,
ℓ2 = ∆h1 − ∆h3 + n2 ,

3 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 13.2
359
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 T
A A A ℓ = A−1 AT A ℓ = A−1 AT A 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
360 13 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 1 c ⎦ ,
/︁ ⎥
= 3 ⎣ 2 −1 1 ⎦ ⎣ 1 ⎦= 3⎣ 2
⎥⎢
1 c −1 1 c
/︁ /︁
−1 −1 1 −1
and the solution is
⎡ ⎤
∆h
ˆ︂ 1
⎣ ∆h2 ⎦ = A−1 ℓ =
⎢ ˆ︂ ⎥

∆h
⎡ /︁ ⎤ ⎡ ⎤ ⎡ ⎤
−1 2 1 c ℓ1 −1 2
ˆ︂ 3 [︄ ]︄
ℓ1
= 13 ⎣ 2 −1 1 c ⎦ ⎣ ℓ2 ⎦ = 1 ⎢
/︁ ⎥ ⎢
3 ⎣ 2 −1 ⎦ ,
⎢ ⎥ ⎥
ℓ2
−1 −1 1 c
/︁
0 −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 13.2
361
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 rakennematriisi
±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
362 13 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
4 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.
korkeus- The situation is somewhat similar to when one is defining a height
järjestelmä or vertical reference system for a country. One needs to fix one height,
for example that of Helsinki harbour. If alternatively one fixes another
height, for example that of 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 and bi to be derived, for example,
from the known uncertainties of the orbit determination available. The
pienimmän least-squares adjustment equation then becomes
neliösumman )︁−1 T
x = AT A + σ2 Σ−1
(︁
tasoitus ˆ︁ A ℓ,
5 in which Σ is the diagonal matrix containing the a priori variances5 σ2a,i

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 13.2
363

^ Figure 13.4. Example of track geometry of satellite altimetry.

and σ2b,i of the parameters of each track i. This approach is referred to

as Tikhonov6 regularisation. 6

^ 13.2.3 Another example

In figure 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
are there?
5 σ is the mean error of unit weight, in this case the mean error, assumed constant, of a

crossover observation.
6 Andrey Nikolayevich Tikhonov (1906–1993) was a Russian mathematician and
geophysicist.

í Õ ! ¤. û
364 13 Satellite altimetry and satellite gravity missions

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 a trend,
a term changing linearly with time, 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 − (16 − 4) = 4 redundancies.
6. Any such choice would be arbitrary. Rather use the method
described above, Tikhonov regularisation.

í Õ ! ¤. û
 Choice of satellite orbit 13.3
365
^ 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 reckoned 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 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 isoakselin
distance from the geocentre — while h is called the satellite’s mean puolikas
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
Earth multiplied by Newton’s universal gravitational constant.8 The 8

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
366 13 Satellite altimetry and satellite gravity missions

Z
Earth’s
rotation E
Perigee Nodal line
axis
Y
Nodal line
ν ω
ν Peri-
Satellite Apogee
i a gee
Ω ̇ b
Ω ω
X
θ
line
dal

Earth ea
X
si

rotation
Ap

X X ′, (Greenwich)
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.

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
laseretäisyys- from satellite laser ranging (SLR) observations. For this purpose the
mittaus well-known LAGEOS (Laser Geodynamic Satellite) satellites (1976-039A
and 1992-070B), which orbit the Earth at a height of 5900 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
seen from table 13.1 at the start of the chapter. The height is fine-tuned

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 13.3
367
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 affect kerroin
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 or nodal precession. This typically 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
√︃
dΩ GM⊕ (︂ a⊕ )︂2
= − 32 a J2 cos i,
dt a3
in which a⊕ is the equatorial radius of the Earth reference ellipsoid, M⊕
the mass of the Earth, and i the inclination of the orbital plane relative
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
368 13 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 /
9 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 eastwards
direction of the Earth’s rotation. The orbital inclination i, or for a
retrograde orbit, its supplement 180◦ − i, is also the greatest northern
or southern geocentric latitude a satellite can fly over. This means that,

9 If
the height of the orbit is less than 1400 km, it cannot be completely no-shadow. In
midwinter or in midsummer the satellite will then fly through the Earth’s shadow.

í Õ ! ¤. û
 Choice of satellite orbit 13.3
369
Spring

Summer Winter
Autumn

Autumn

Winter
Summer
Morning hemisphere
Evening hemisphere

Spring

Figure 13.7. The geometry of a “no-shadow” orbit. In this figure, the satellite
flies north over places where it is morning and south over places
^ where it is evening.

unless the inclination is precisely 90◦ , there will be areas around both
poles that the satellite will never overfly: the “polar holes”.
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 (“resonance”), and thus they cannot be observed with a
satellite in this type of orbit. 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
370 13 Satellite altimetry and satellite gravity missions

θ̇

i
3 2 1

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⊕ =

í Õ ! ¤. û
Choice of satellite orbit 13.3
371
6 378 137 m.
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, a day being 1440
minutes: P = 1440 min 14 = 102.857 min.
/︁

2. The satellite completes 43 orbits in three days or 3 × 1440

minutes: P = 3 × 1440 min 43 = 100.465 min.
/︁

3. The satellite completes 502 orbits in 35 days or 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.
/︁

í Õ ! ¤. û
372 13 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;
fac=4*pi*pi;
a=(GM*P*P/fac)^0.33333333;
h = a - ae;
printf(’\n\nOrbital height: %8.3f km.\n’, h/1000);

8.
(a) The 35-day orbit would be excellent for detailed map-
ping. The three-day orbit would be able to see, for
example, tides or weather-related phenomena, albeit at
poorer 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
10 thrusters.10 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
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).

10 Disclosure: the total velocity change needed is ∆v = 1.6 m/s , a brisk walking pace.

í Õ ! ¤. û
 Retracking 13.5
373
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”.

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
offset may slowly change or drift over time, and may be temperature- käynti
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 an
approach called retracking (Altimetry, Retracking).
The method of analysis uses the point on the leading edge of the
return pulse which is at half the height of 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

í Õ ! ¤. û
 374 13 Satellite altimetry and satellite gravity missions

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,
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
consider freeboard in the processing: how high the ice sticks out of
the water.
mannerjäätikkö ◦ Over continental ice sheets. Here, the shape of the return pulse
will be very different from that over open water. Furthermore, the
travel time of the return pulse varies rapidly as the satellite flies
11 on, and the reception window cannot keep track.11
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
12 about 1030 kg m3 , the ice thickness is about 8× freeboard.12 If there is
/
additional remote-sensing data providing the area of ice cover, one can
calculate the total volume and mass of the sea ice.
The Arctic ice cover has diminished radically over recent decades.
The most radical reduction has been that of ice volume, see figure

11 Thenewest satellites such as Sentinel-3 use a digital terrain model for steering the
tracking window when not over the open ocean.
12 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 13.6
375
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).
^

13.10. In addition to surface area, thickness is also decreasing: of the

multi-year, thicker ice, a large part has already vanished.

^ 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, 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:

í Õ ! ¤. û
376 13 Satellite altimetry and satellite gravity missions

– 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
something (though not everything) about the variability of
the sea surface.
– Crossovers may also provide information on sea-surface vari-
ability. When the sea surface varies, the results from the
crossover adjustment will get poorer: the root-mean-square
of a posteriori (after calculation) crossover differences will
grow. Using this method to actually study sea-surface vari-
ability is more challenging. It can however be used to estimate
the magnitude of variability.
– Nowadays altimetric satellites always carry a GNSS positioning
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
direct measurement, assuming that both temporal and spatial
measurement densities are sufficient.
◦ The deviations of the 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 measure-
ments are available for an area, these may be used to estimate the
geoid, and after that one may calculate the sea-surface topography.
Collecting sufficiently precise and dense gravimetric data is possi-
ble with sea or airborne gravimetry. Measurement with a special
satellite (gravitational gradiometry, GOCE satellite) has also long
been planned and has finally been realised, see subsection 13.7.3.

^ 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

í Õ ! ¤. û
 Satellite gravity missions 13.7
377
or geopotential; in other words, to determine a global high-resolution
model of the geoid.

^ 13.7.1 CHAMP
CHAMP (Challenging Minisatellite Payload for Geophysical Research
and Applications, 2000-039B) was a German satellite project under
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 some 300 km due to atmospheric drag. The orbital inclination was
87◦ . On 19th 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) over 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 11.8.
The satellite also contained an accelerometer, which eliminated the kiihtyvyys-
satellite’s accelerations caused by the atmosphere’s aerodynamic forces, mittari
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.

^ 13.7.2 GRACE
GRACE (Gravity Recovery And Climate Experiment Mission, 2002-
012 A and B) measured temporal changes in the Earth’s gravity field

í Õ ! ¤. û
378 13 Satellite altimetry and satellite gravity missions

Magnetometer GPS-2 GPS-3

beam GPS-4
GPS-1
GPS antenna
Acceleration vector
Nom
inal
orbi
“CHAMP” True t
orbit

Solar
Earth internal mass
cells
density variations
(example)

Figure 13.11. Determining the Earth’s gravity field from GPS orbital tracking
^ of a low flying satellite.

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 total mass, variable in time, 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 ±1 —m/s . Both satellites also carried
sensitive accelerometers for measuring and eliminating the effect of
atmospheric drag.

í Õ ! ¤. û
 Satellite gravity missions 13.7
379
Difference between line-of-sight accelerations

1 Satellite 2
Satellite Distance 220 km

Precise
ranging,
wavelength
45 0 km 1.5 cm
e ight
H

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 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.
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.

^ 13.7.3 GOCE
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

í Õ ! ¤. û
380 13 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).

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
ajoaine a stock of propellant in order to maintain the orbit against atmospheric
13 drag. The orbital inclination was 96◦. 7, so the orbit was Sun-stationary.13

GOCEcarried a very sensitive gravitational gradiometer, a device for

measuring 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 extremely sensitive,
three-axes accelerometers mounted pairwise on a frame. The mission
ended in 2013 and the satellite was seen to burn up in the atmosphere
on 11th 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 some 100 km in diameter, and their

13 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 13.7
381
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.

precision is as good as ±2 cm.

With a global geoid model this precise, we may calculate the de-
viations of the sea surface from the geoid, an equipotential or level
surface, at similar 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

í Õ ! ¤. û
382 13 Satellite altimetry and satellite gravity missions

12.5 and figure 12.4. This is 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?
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. What is, in satellite altimetry crossover adjustment, a datum defect,
and how can it be corrected?
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 satellites with
non-integer repeat periods seem to 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
383

^ 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 crossovers, 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 unknowns a and b are needed in total?

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
not?
4. How many unknowns need to be fixed for an unique solution?
Which coefficients would you fix?

í Õ ! ¤. û
384 13 Satellite altimetry and satellite gravity missions

5. Are there any redundant observations? Was it wise to choose an

orbit-error correction model with two unknowns per track?

^ Exercise 13 – 2: Satellite orbit

A satellite moves in a Sun-synchronous orbit. After 419 orbits and 30
days, the satellite 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. As a Sun-synchronous orbit is not polar, the highest northern
latitude that the satellite can fly over is less than 90◦ . In what
compass direction is the satellite flying at that point?

^ Exercise 13 – 3: Kepler’s third law

1. What is the height h of a satellite of period 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.
2. What is the orbital inclination i of the satellite, if it is given that
the orbit is circular and Sun-synchronous? See section 13.3.

í Õ ! ¤. û
^ Tides, the atmosphere, and Earth
crustal movements

^ 14.1 The theoretical tide

14
The tide is the result of the attraction of external celestial bodies, in
the Earth’s case Moon and Sun, and of the free-fall motion of the Earth
towards these attracting bodies. We may write the potential field of the
attraction as follows:
GM
V′ = ,
ℓ
where ℓ is the distance of the attracting body from the point of evaluation
of the potential, see figure 14.1. GM is the mass of the Sun or Moon
multiplied by Newton’s gravitational constant. The attraction can be
expressed as an acceleration or “force” field, of magnitude
GM
a′ = .
ℓ2
The Earth, floating freely in space, responds to this by freely falling
towards the attracting body, with an acceleration (see section 1.4):
GM
a ′′ = ,
d2
with d the distance of the attracting body from the Earth’s centre.
This acceleration a ′′ is a constant. Considered as a field in space, we
can associate a potential with it:
GM GM R 2 z GM R 2
(︂ )︂ (︂ )︂
′′
V = 2 z= = cos ζ ,
d R d R R d

– 385 –
386 14 Tides, the atmosphere, and Earth crustal movements
Lo
Hi w
gh tid
t id e
e

R
X ζ ζ′
z
Hi
gh
Lo tid
w e ℓ
tid
e
d

Parallax

Figure 14.1. Theoretical tide. ζ ′ is the local zenith angle of the Moon (or Sun),
^ ζ the corresponding geocentric angle.

where z is the co-ordinate defined along the line connecting the Earth’s
centre and the attracting body, see figure 14.1. R is the radius of the
Earth, assumed spherical.
The net tidal potential as felt by mass elements on the Earth’s surface
is now

GM GM R 2
(︂ )︂
′ ′′
V =V −V = − cos ζ =
ℓ R d
GM ∑︂ R n+1
∞ (︂ )︂
GM R 2
(︂ )︂
= Pn (cos ζ) − cos ζ
R d R d
n=0

using expansion 8.7. Here, the term for n = 0 is constant and thus
arbitrary for a potential, and we delete it. The term n = 1 produces a

í Õ ! ¤. û
The theoretical tide 14.1
387
precise cancellation. Remains
GM ∑︂ R n+1
∞ (︂ )︂
V= Pn (cos ζ),
R d
n=2

in which the term for degree 2 is dominant. asteluku

The tidal potential V can now be written as follows:
GMR2 GMR2 (︁
3 cos2 ζ − 1 + · · · ,
)︁
V= 3
P2 (cos ζ) + · · · = 3
d 2d
ζ is the local geocentric zenith angle of the Sun or Moon: the local zenith
angle ζ ′ corrected for parallax, see figure 14.1. P2 (cos ζ) is the Legendre
polynomial of degree two. In the case of the Sun and Moon, the extra
terms (· · · ) for higher degree numbers 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 1
the hour angle2 of the Moon. 2

The spherical-harmonic addition theorem (Wolfram MathWorld, Spheri- pallo-

cal Harmonic Addition Theorem) yields funktioiden
summauslause
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,

P2 (cos ζ) = P2 (sin ϕ) P2 (sin δ) +

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
the geocentre.
2 Thehour 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, at its greatest elevation in the local meridian.

í Õ ! ¤. û
388 14 Tides, the atmosphere, and Earth crustal movements

+ 31 P21 (sin ϕ) P21 (sin δ) cos h + 1

P (sin ϕ) P22 (sin δ) cos 2h.
12 22

According to table 3.2,

P21 (sin ϕ) = 3 sin ϕ cos ϕ, P21 (sin δ) = 3 sin δ cos δ,

P22 (sin ϕ) = 3 cos2 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 +
(︁ )︁ (︁ )︁

3
+ 4
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= 3
4d ⎝ ⎠
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 1
sin2 δ = sin2 ϵ sin2 ℓ = sin2 ϵ
)︁
=⇒ 2
− 2
cos 2ℓ , (14.1)

in which ℓ is the longitude of the Moon in its orbit, reckoned from

the equator crossing, and ϵ is the inclination of the Moon’s orbital

í Õ ! ¤. û
The theoretical tide 14.1
389
plane with respect to the equator, on average 23◦. 5 but varying
between 18◦. 3 and 28◦. 6. Thus we obtain
GMR2 (︁ )︁ (︂ (︁ 1 1 )︂
2 2
)︁
V1 = 3 sin ϕ − 1 3 sin ϵ 2 − 2 cos 2ℓ − 1 ,
4d3
where we have used result 14.1. We split V1 = V1a + V1b into two
parts, a constant3 and a periodic, semi-monthly (“fortnightly”) 3
part:
GMR2 (︁
3 sin2 ϕ − 1 23 sin2 ϵ − 1 ,
)︁ (︁ )︁
V1a = 3
(14.2)
4d
GMR2 (︁
3 sin2 ϕ − 1 23 sin2 ϵ cos 2ℓ .
)︁ (︁ )︁
V1b =− 3
4d

◦ In addition, there are a couple of terms in which the Moon’s 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
= 1 − sin2 ϵ
)︁
2
− cos
2
2ℓ ,
1
(︁ )︁
cos 2ℓ cos 2h = 2 cos(2ℓ + 2h) + cos(2ℓ − 2h) ,
√︂ (︁ )︁
sin 2δ = 2 sin δ cos δ = 2 sin2 δ 1 − sin2 δ =
√︃
(︁ 1 1 )︁ (︂ 2
(︁ 1 1 )︁)︂
= 2 sin ϵ 2
− 2
cos 2ℓ 1 − sin ϵ 2
− 2
cos 2ℓ ,

leading to a trigonometric expansion in lunar longitude ℓ, and so

on. See for example Melchior’s4 famous book (1978). 4

3 In case of the Moon, not precisely, because ϵ$ is (slowly) time-dependent.

í Õ ! ¤. û
390 14 Tides, the atmosphere, and Earth crustal movements

^ Table 14.1. The various periods in the theoretical tide. The widely used
symbols were standardised by George Darwin.

Changing Period Darwinsymbol

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

The coefficient (︂ )︂2

3GMR2
def 3 GM R
D= = , (14.3)
4d3 4 d d
5 “Doodson’s5 constant, ” is taken separately from the above equations.

The value for the Moon equals D$ = 26.8 cm × γ and for the Sun
D⊙ = 12.3 cm × γ, with γ ≈ 9.81 m/s2 . See figure 14.2.
6 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
orbit, like the Earth’s orbit, is significantly eccentric.

4 PaulJacques Léon Camille baron Melchior (1925–2004) was a 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 FRS FRSE (1845–1912) was an English astronomer and
mathematician, son of Charles Darwin of Origin of Species fame.

í Õ ! ¤. û
 Deformation caused by the tidal potential 14.2
391
V1a , permanent V1b , fortnightly, ϵ = 23◦
80 0.2 80 0.1

40 0 40 0.05
Latitude ( ◦ )

−0.2 0
0 0
−0.05
−40 −0.4 −40
−0.1
−80 −0.6 −80
−0.15
Obliquity ϵ, 0◦ − 90◦ Lunar longitude, 0◦ − 360◦
X
V2 , diurnal, δ = 23◦ V3 , semidiurnal, δ = 23◦
80 80 0.6
0.4
0.4
40 0.2 40
Latitude ( ◦ )

0.2
0 0
0 0
−0.2 −0.2
−40 −0.4 −40 −0.4
−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.

^ 14.2 Deformation caused by the tidal potential

The tidal potential, 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 properties are often characterised by elastic Love7 numbers (Love, 7

7 Augustus Edward Hough Love FRS (1863–1940) was a British mathematician and
student of Earth elasticity.

í Õ ! ¤. û
392 14 Tides, the atmosphere, and Earth crustal movements

1909; Melchior, 1978).

Let us first write the tidal potential V = V(ϕ, λ, r) in the following
way:
∑︂
∞ (︂ )︂
r n ∑︂
∞
V(ϕ, λ, r) = Vn (ϕ, λ) = Vnint (ϕ, λ, r),
R
n=2 n=2

in which the index n denotes the spherical-harmonic degree number.

asteosuus Vn (ϕ, λ) is the degree constituent, and
(︂ )︂n
def r
Vnint (ϕ, λ, r) = Vn (ϕ, λ)
R
avaruus- the interior solid spherical harmonic of potential V for degree number
pallofunktio n.
8 Call the linear8 displacement of an element of matter of the solid
Earth in the radial direction, ur , in the north direction, uϕ , and in the
east direction, uλ . The following equations apply:

1 ∑︂ ∑︂
∞ ∞
int
ur (ϕ, λ, r) = γ Hn (r) Vn (ϕ, λ, r) = Hn (r) ζn (ϕ, λ, r),
n=2 n=2

1 ∑︂
∞
∂Vnint (ϕ, λ, r) ∑︂
∞
uϕ (ϕ, λ, r) = γ Ln (r) =r Ln (r) ξn (ϕ, λ, r),
∂ϕ
n=2 n=2

1 ∑︂
∞
∂Vnint (ϕ, λ, r) ∑︂
∞
uλ (ϕ, λ, r) = γ Ln (r) =r Ln (r) ηn (ϕ, λ, r).
cos ϕ ∂λ
n=2 n=2

Here, 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
luotiviiva of the plumb line.
epäsuora The deformation of the Earth also causes a change, the “indirect effect”
vaikutus

8 So their unit is metre, not degree, also for uϕ and uλ !

í Õ ! ¤. û
Deformation caused by the tidal potential 14.2
393
in addition to the Moon’s original tidal potential V, in the geopotential.
We write
∑︂
∞
δV(ϕ, λ, r) = Kn (r) Vnint (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 , k n = Kn R . (14.4)

Because of the large distances to Sun and Moon, the only non-negligible
part of the tidal potential V is the part for the degree number n = 2, the
“rugby-ball part” V2int .
The Love numbers will still depend on the frequency, or on the tidal
period P:
(︁ )︁ (︁ )︁ (︁ )︁
hn = hn P , ℓn = ℓn P , k n = 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 different
amplitudes and phase angles.9 In this way we may determine at 9
least those Love numbers that correspond to periods occurring in the
theoretical tide.
The numbers h and ℓ are nowadays obtained for example by GNSS
positioning. The GNSS processing software contains a built-in reduction
for this phenomenon. From gravity measurements one obtains infor-
mation on a certain linear combination of h and k, δ = 1 + h − 32 k: the
lunar tidal force 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 tilt meter is also a useful research instrument,
like the instrument of the Finnish Geodetic Institute that has long been

9 The phase angles may be represented by making the Love numbers complex.

í Õ ! ¤. û
394 14 Tides, the atmosphere, and Earth crustal movements

in use in the Tytyri limestone mine (Tytyri Mine Experience) in Lohja

(Kääriäinen and Ruotsalainen, 1989). A modern, improved version of
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
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
10 et al. (2010). The Love number ℓ10 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 elastic properties 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
pallofunktio- in defining spherical-harmonic expansions of the Earth’s gravity field,
kehitelmä the tidal reduction does include this term which it is theoretically and

10 Alsocalled the Shida number. Toshi Shida (1876–1936) was a Japanese Earth tide
researcher.

í Õ ! ¤. û
 Tidal corrections between height systems 14.4
395
practically impossible to know. See Poutanen et al. (1996).
More generally, the reduction of a geodetic quantity, for example the
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,
directly suitable for use in oceanography.
◦ The direct effect of the tidal field of Sun and Moon 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 effect of the tidal field 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.
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 1
)︁
≈ −0.7615 · sin ϕ − .
4d3 3

With the combined Doodson’s constant 14.3 for Sun and Moon equal to

3GM⊙ R2 3GM$ R2
D= + =
4d3⊙ 4d3$

í Õ ! ¤. û
396 14 Tides, the atmosphere, and Earth crustal movements

Earth permanent tidal deformation Tide-free Earth crust

Direct permanent effect on geoid of Mean (zero) Earth crust
Moon and Sun Tide-free geoid
Zero geoid
Effect of Earth permanent tidal
deformation (mass displacement) on Mean geoid
geoid Reference ellipsoid

Figure 14.3. Conceptual diagram showing the constituents of the permanent

^ tide.

= (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
 Loading of the Earth’s crust by sea and atmosphere 14.5
397
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 the Earth’s crust enter into the picture,
we need values for the Love numbers h and k for the permanent
tidal deformation, expressing this deformation and its potential as
fractional parts of the original external tidal potential.11 As we have 11
seen, these numbers cannot be empirically determined. Values often
used are h ≈ 0.6, k ≈ 0.3. With this, the above equations apply
with the coefficient 29.77 cm multiplied by the linear combination
γ = 1 − h + k ≈ 0.7. This yields
def (︁ 1
∆mean 2
)︁
tidefree H = Hmean − Htidefree = −20.84 cm × 3
− sin ϕ ,

∆mean mean
(︁ 2 2
)︁
tidefree H(ϕ2 ) − ∆tidefree H(ϕ1 ) = 20.84 cm × sin ϕ2 − sin ϕ1 .

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.
11 Both are needed: the tidal deformation displaces both the Earth’s crust and the
geoid.

í Õ ! ¤. û
398 14 Tides, the atmosphere, and Earth crustal movements

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
12 is the package Eterna written by the German Hans-Georg Wenzel,12

which also has found use 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.
For measuring the deformation, a registering gravimeter is generally
used. The Earth’s crust moves up and down elastically, which to
painovoiman 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?

12 Hans-Georg Wenzel (1945–1999) was a German physical geodesist and geophysicist.

í Õ ! ¤. û
 Exercise 14 –1: The permanent tide
399
4. What is Doodson’s constant?
5. What do Love numbers express?
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: The permanent tide

The equation for the permanent part of the tide is

GMR2 (︁
3 sin2 ϕ − 1 32 sin2 ϵ − 1 ,
)︁ (︁ )︁
V1a = 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

– 401 –
402 15 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, Italy (http://www.isgeoid.polimi.it/). 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. The
Italian state has provided significant funding for these activities.
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
1 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, while American, is a very cosmopolitan player.

í Õ ! ¤. û
 The Nordic countries 15.3
403
^ 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 models
are actually based on the Nordic NKG2004 gravimetric geoid, and are
fitted to a Finnish set of GNSS levelling control points, serving as a
transformation surface for heights.
In 2015, the Finnish Geodetic Institute was merged into the National

í Õ ! ¤. û
404 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 ¤. 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. The defining property of a
vector is that under co-ordinate transformations it behaves identically
to 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. The effect of the transformation on the components is the
same as if the vector were a location difference between two points.
This is fundamentally why it is possible to draw vectors as arrows.
About notation In printed text, vectors are often written in bold: v. In
handwritten text one may use an arrow above the symbol: → −v.

^ A.1.1 The 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, we speak of a scalar field: each scalar value is

– 405 –
406 A Field theory and vector calculus — core knowledge

tied to a location, but, even if a co-ordinate transformation changes

the co-ordinate values of the location, the scalar value itself remains
unchanged: it is an invariant.
An example of a scalar product: work ∆E is
⟨︁ ⟩︁
∆E = F · ∆r ,
⟨︁ ⟩︁
the scalar product of force F and path ∆r. Often, the angle brackets ·
are left 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 w B ⟨︁ ⟩︁
∆EAB = dE = F · dr .
A 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 , (homogeneity)
⟨︁ ⟩︁ ⟨︁ ⟩︁ ⟨︁ ⟩︁
a · (b + c) = a · b + a · c , (distributivity)
⟨︁ ⟩︁ ⟨︁ ⟩︁
a·b = b·a , (commutativity)
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 A.1
407
^ A.1.3 The 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 L:
⟨︁ ⟩︁
L= 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
⟨︁ ⟩︁
L = m r × ṙ . (A.2)

^ A.1.4 The vectorial product, formally

Let
def ⟨︁ ⟩︁
x = a×b
be the vectorial product of the two vectors a and b. Then (µ ∈ R):
⟨︁ ⟩︁ ⟨︁ ⟩︁ ⟨︁ ⟩︁
µa × b = a × µb = µ a × b , (homogeneity)
⟨︁ ⟩︁ ⟨︁ ⟩︁ ⟨︁ ⟩︁
a × (b + c) = a × b + a × c , (distributivity)
⟨︁ ⟩︁ ⟨︁ ⟩︁
a×b =− b×a , (anticommutativity)
⟨︁ ⟩︁
and thus a × a = 0.

The resulting vector x is always perpendicular to the vectors a and

b. The length of vector x corresponds to the surface area of the
parallelogram spanned by vectors a and b: suunnikas

∥x∥ = ∥a∥ ∥b∥ sin α, (A.3)

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 = µ b for some suitable value of µ.

í Õ ! ¤. û
408 A Field theory and vector calculus — core knowledge

⟨︁ ⟩︁
x= a×b

α
∥x∥

b
a

^ Figure A.1. Exterior or vectorial product.

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 x = a × b .

^ 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.4)
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.4:
⟨︃ ⟩︃
d ⟨︁ ⟩︁ ⟨︂ dr dr ⟩︂ d2 r ⟨︁ ⟩︁ ⟨︁ ⟩︁
r × ṙ = × + r × 2 = ṙ × ṙ + r × r̈ . (A.5)
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 the
associated acceleration,
d2 r
r̈ = 2 ,
dt
keskeisvoima are central:
GMm
F = m r̈ = − r.
∥r∥3

í Õ ! ¤. û
 Scalar and vector fields A.2
409
Angular momentum
⟨︁ ⟩︁
r × ṙ

Planet

Sun
ṙ
1⃦
⃦⟨︁ ⟩︁⃦
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.

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.5:
d ⟨︁ ⟩︁ GM ⟨︁ ⟩︁
r × ṙ = 0 − r × r = 0.
dt ∥r∥3

So: the quantity on the left-hand side, angular momentum L per unit
of mass m, equation A.2, is conserved:
⟨︁ ⟩︁ L
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.

í Õ ! ¤. û
410 A Field theory and vector calculus — core knowledge

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).

^ 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, and
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.6)
Orthonormality means in addition that

∥i∥ = ∥j∥ = ∥k∥ = 1. (A.7)

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 b1 + a2 b2 + a3 b3 = ai bi ,
i=1

using the identities stated above for the basis vectors A.6 and A.7.

í Õ ! ¤. û
 Scalar and vector fields A.2
411
For the vectorial product, the calculation is more involved. For
orthogonal vectors, the angle α in equation A.3 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 .

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 b 2 − a2 b 1 .

These expressions are determinants as well:

⎡ ⎤
c1 [︄ [︄ ]︄ [︄ ]︄ [︄ ]︄ ]︄T
a 2 a 3 a 3 a 1 a 1 a 2
⎣ 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

í Õ ! ¤. û
412 A Field theory and vector calculus — core knowledge

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

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. Imagine 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.

^ A.2.5 The divergence

Given is a vector field a(x, y, z) = a1 i + a2 j + a3 k. Form the scalar
product s of this and the nabla operator:
⟨︁ ⟩︁ ∂a1 ∂a2 ∂a3
s = div a = ∇ · a = + + .
∂x ∂y ∂z

í Õ ! ¤. û
 Scalar and vector fields A.2
413

^ Figure A.3. The gradient. The level curves of the scalar field are dashed blue.

Interpretation The divergence describes the sources of a vector field,

both the positive and negative ones. Think of 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 is again a vector field a(x, y, z). Form the vectorial product c of
this and the nabla operator, again producing a vector field:
⎡ ⎤
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

í Õ ! ¤. û
414 A Field theory and vector calculus — core knowledge

Figure A.4. The divergence. Positive divergences (“sources”) and negative

^ ones (“sinks”). Field lines red dashed.

using the evaluation rules for determinants.

Interpretation The curl describes the eddiness or vorticity 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 rotation, whereas 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

í Õ ! ¤. û
 Scalar and vector fields A.2
415

^ Figure A.5. The curl. Positive (anticlockwise) and negative (clockwise) eddies.

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.
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,

í Õ ! ¤. û
416 A Field theory and vector calculus — core knowledge

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 conser-
vative, 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),

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

í Õ ! ¤. û
 Integrals A.3
417
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 .
The differential form of this is
⟨︁ ⟩︁
dE = F · dr ,

from which one obtains the integral form, the work integral
w B ⟨︁ ⟩︁
∆EAB = F · dr .
A

Here, the amount of work done by a body moving 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
w B ⟨︁ ⟩︁
∆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

í Õ ! ¤. û
418 A Field theory and vector calculus — core knowledge

vector field, the projection of a onto the normal vector of the surface,
the vector perpendicular to the surface in the surface element dS.
Let the normal vector on 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,
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 ⟨︁ ⟩︁ w 2π w + π/2 ⟨︁ ⟩︁ 2
a · dS = a · n dS = a · n R cos ϕ dϕ dλ.
S S 0 − π/2

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

í Õ ! ¤. û
 Integrals A.3
419
Tangent vector t

curl a

curl a
z Integral Closed
s ⟨︁ Integral ⟩︁
⟨︁ ⟩︁
a · t ds path ∂S
∂S
S curl a · n dS

curl a

^ Figure A.6. The Stokes curl theorem.

necessary integrations and differentiations to be possible. Then (also

called the Kelvin1 –Stokes theorem): 1
x ⟨︁ ⟩︁ z ⟨︁ ⟩︁
curl a · dS = a · dr ,
S ∂S

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

1 William Thomson, Lord Kelvin PRS FRSE (1824–1907) was a British physicist, mathe-
matician, engineer and inventor, and of course best known for the absolute temperature
scale proposed by him in 1848.

í Õ ! ¤. û
420 A Field theory and vector calculus — core knowledge

so also w B ⟨︁ ⟩︁ w B ⟨︁ ⟩︁
a · dr = a · dr .
A A
path 1 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 field g(x, y, z) is the gradient of the Earth’s
gravity potential or geopotential 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.

^ A.3.4 The Gauss divergence 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 A.4
421
n

∂V
div a V

Figure A.7. The Gauss divergence theorem. n is the normal vector to the
exterior surface. The Gauss divergence 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.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

í Õ ! ¤. û
422 A Field theory and vector calculus — core knowledge

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

When talking about liquid or gas flow, it is good to be aware that

the vorticity 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. In fact, turbulent flow is very common, and even for
laminar flow, usually curl v ̸= 0.

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 in the space can be expressed 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 .

– 423 –
424 B 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 = ⟨︁ ⟩︁ = 2
. (B.1)
i=1
e i · e i ∥e i ∥
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 , r i = r · ei . (B.2)
i=1

The quantity √︂⟨︁

def ⟩︁
∥ei ∥ = e i · 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
1 the scalar product of two functions f and g as the following integral1
⟩︁ ⟨︂→− → w 2π
− def 1
⟨︁ ⟩︂
f·g = 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.
kanta One basis in this vector space (a function space) is formed by the Fourier

1 The arrows over the function designators f and g here are just psychology: the
functions are really “vectors”. These arrows are not normally used.

í Õ ! ¤. û
 The Fourier function space B.2
425
basis functions,

1√
e0 = 2,
2
ek = cos kx, k = 1, 2, 3, . . . , (B.4)
e−k = sin kx, k = 1, 2, 3, . . . .

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 0 f(x) cos kx dx, k = 1, 2, 3, . . . ,
w 2π
⟨︁ ⟩︁ 1
bk = f · e−k = π 0 f(x) sin kx dx, k = 1, 2, 3, . . . .

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)︁.

í Õ ! ¤. û
426 B 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 2π w 2π
1 1
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 2π w 2π
1 1
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 =
kπ ⎩− 2 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 a0 + bk sin kx = −π sin kx.
2 2 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 √ ∑︂
K
2 ∑︂ 1
K
f(K) (x) = 12 a0 2 + bk sin kx = 1
−π sin kx, (B.5)
2 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 B.2
427
K=1
K = 25 K=3 K=5
1.0 f(x)

0.8 a0

0.6
X
0.4 a b

0.2 b5
b3
0 f(x)
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 π f(K) (x) − f(x) dx = 0.
K→∞ 0

This does not mean that for an arbitrarily small value ε it holds that
⃓f (x) − f(x)⃓ < ε for every x ∈ 0, 2π , when K → ∞. Looking at
⃓ (K) ⃓ [︁ )︁

figure B.1, there will always remain a small neighbourhood of x = π

within which there will be points x ′ ̸= π where the absolute difference
⃓ (K) ′ ⃓
⃓f (x ) − f(x ′ )⃓ > 0.1 (or any other positive bound < 0.5), even for
arbitrarily large values of K. We say that the Fourier expansion is
convergent, but not uniformly convergent. tasainen
suppeneminen
The Fourier expansion converges pointwise “almost everywhere” in
[︁ )︁
x ∈ 0, 2π : at all points except for the two special points x = 0 and
def
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

í Õ ! ¤. û
428 B Function spaces

shoulder will get narrower for higher K, but not any lower, remaining
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 λ for which

one or more solutions or eigenvectors x 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 ∑︂
n (︃∑︂
n )︃
Lx = xi · aij ej = aij xi ej . (B.6)
i=1 j=1 j=1 i=1

í Õ ! ¤. û
 Sturm–Liouville differential equations B.3
429
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
{︁ }︁
ei . Matrix A is again the numerical representation of operator L on
the same basis.2 2

^ B.3.2 A self-adjoint operator

Let L be a linear operator in a vector space where there exists a scalar
⟨︁ ⟩︁
product: a bilinear form x · y which is symmetric or commutative.
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
430 B Function spaces

or
∑︂
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, or 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 different, then the eigenvectors xp , p = 1, . . . , n will
täydellinen constitute a complete orthogonal basis3 in the vector space Rn .
ortogonaali
The proof is not hard. We start from the equation for the eigenvalue
kanta
3 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 =⇒ λ p xq · xp = λ q 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 B.3
431
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, and 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 ΣPP − λI = det = 0,
σxy σ2y − λ
or
(︁ 2
σx − λ σ2y − λ − σ2xy = 0.
)︁ (︁ )︁

This yields

1
(︁ 2 2
)︁ 1 √︂(︁ )︁2 (︁ )︁
λ1,2 = 2
σx + σy ± 2 σ2x + σ2y − 4 σ2x σ2y − σ2xy =
1
(︁ 2 2
)︁ 1 √︂(︁ )︁2
= 2 σx + σy ± 2 σ2x − σ2y + 4σ2xy .

The visual presentation of the variance matrix is the 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
432 B 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 B.3
433
it holds that (integration by parts):

⟩︁ w T
]︃T w
d2 y(t)
[︃
⟨︁→
− →
− dy(t) T dx(t) dy(t)
x ·Ly = x(t) dt = x(t) − dt,
0 dt2 dt 0 0 dt dt
⟩︁ w T d2 x(t)
[︃ ]︃T w
⟨︁ →
− →
− dx(t) T dx(t) dy(t)
Lx · y = 2
y(t)dt = y(t) − dt.
0 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 4
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.
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 5
6
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.
5 JacquesCharles François Sturm FRS FAS (1803–1855) was a French mathematician,
one of the 72 names engraved on the Eiffel Tower. Eiffel Tower, 72 names.

í Õ ! ¤. û
434 B Function spaces

type problems”. It includes the oscillation equation, Legendre’s equa-

tion, Bessel’s equation, and many more. Every one of them generates, in
a natural way, its own set of mutually orthogonal functions that serve as
kantafunktio 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
−
⟩︂
def
f · g = f(t) g(t) dt.
−1

They do not however constitute an orthonormal basis, but only an

ortogonaali orthogonal one:
kanta
⟩︁ w +1 2 2
∥Pn ∥2 = 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
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.

6 Joseph Liouville FRS FRSE FAS (1809–1882) was a French mathematician.

í Õ ! ¤. û
Spherical harmonics B.5
435
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
f ·−
⟩︂
g = f(ϕ, λ) g(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 this
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 Yn ′ m ′ , are orthogonal with respect to each other:
x
⟨︁ ⟩︁ 1
Ynm · Yn ′ m ′ = Y (ϕ, λ) Yn ′ m ′ (ϕ, λ) dσ = 0
4π σ nm
if n ̸= n ′ or m ̸= m ′ .
{︁ }︁
The basis Ynm , n = 0, 1, 2, . . . , m = −n, . . . , n is orthogonal but not
orthonormal: the lengths of the vectors differ from unity.

∥Ynm ∥2 = Ynm · Ynm =

⟨︁ ⟩︁
⎧
x ⎪
⎨ 1
if m = 0,
1 2 2n + 1
= Ynm (ϕ, λ) dσ = (n + |m|)!
4π σ ⎪
⎩
1
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-
x 2 pallofunktio
⃦Y nm ⃦2 = 1
⃦ ⃦
Y (ϕ, λ) dσ = 1.
4π σ nm

í Õ ! ¤. û
436 B Function spaces

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
This is, in the geometric analogy, a straightforward projection onto the
unit vectors of the basis.
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 2.12 is

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

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

í Õ ! ¤. û
 Self-test questions
437
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

⟩︁ Rn+1 x
vnm = Rn+1 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
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 data 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

˜︁j = j L ,
/︁
The frequency argument, spatial frequency or[︂wave number, /︂ ]︂ ν
j = 0, . . . , N − 1 is defined on the interval1 0, (N − 1) L . i is the 1

– 439 –
440 C Why does FFT work?

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

FFT is just a brutally efficient way of computing both these equations

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∑︂ ∑︂
N−1
1 jk jk ⎠
(︂ )︂ (︂ )︂
F(˜︁
νj ) = ⎝ fk exp −2πi + fk exp −2πi =
N N N
k=0 1
k= N
⎛ 1 N−1 2 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∑︂ j + if j even
1 ⎣fk ± f 1 ⎦ exp −2πi jk ,
(︂ )︂
= (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/subtractions, not counting pre-calculations.
(︁ )︁j
Here we used Euler’s identity exp(−πi) = −1, so e−πij = e−πi =
2 (−1) j , either +1 or −1.2 The expression in square brackets, for each k

[︂ (︁ 1 )︁/︂ ]︂
1 Alternatively,the interval of definition can be chosen as − 12 N L , 2 N − 1 L .
/︁

vj − N L , or j → j−N, for j > 12 N−1. This has the merit

/︁
vj → ˜︁
This is done by mapping ˜︁
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
 441
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 origi-
nal 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 , 16
1 1
N2 , 32 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 first 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 ′ ) (︃ H(ϕ ′ , λ ′ ) H2 (ϕ ′ , λ ′ )
)︃
=R 2
1+ + dσ ′ , (D.1)
σ ℓ(ψ, r, R) R 3R 2

with H the height of the topography.

– 443 –
444 D 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 8.7):

1 ∑︂ 1 r ′ ∑︂
∞ (︃ )︃n+1 ∞ (︃ )︃n
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
1
that may be especially close to a jagged topographic surface. For the
philosophically inclined, read 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
(︂ )︂
= Gρ (R + H)n+3 − Rn+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 r ⃓
n=0

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 D.2
445
(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 H2 H3
· r + 1
(n + 2) 2 + 1
(n + 2) (n + 1) 3 + · · · Pn (cos ψ) dσ ′ .
σ R 2 R 6 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 and does not
diverge. In mountainous topography, this may be a problem.

^ D.2 The interior potential of the topography

In the same way we may derive the equation for the interior potential of
the topography, the masses between the sea level and terrain surface. For
the spatial distance ℓ between those points we use the interior expansion,
equation 8.7, 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
446 D 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 ′ )
⎢ ∑︂
∞
(︄ )︄
(r ′ )−(n−2)
+ r2 ln r ′ ⎥
⎥
=⎣
⎢ rn − =
n−2 ⎦
n=0
n̸=2
r ′ =R
∑︂
∞
rn
(︂ )︂
R+H
= R−(n−2) − (R + H)−(n−2) + r2 ln ,
n−2 R
n=0
n̸=2

yielding

int
Vtop (ϕ, λ, r) =
⎛ ⎞
x ⎜ ∑︂
∞
rn R + H⎟
(︂ )︂
−(n−2) −(n−2) 2 ⎟ P (cos ψ) dσ ′ .
= Gρ R − (R + H) + r ln
R ⎠ n
⎜
σ⎝ n−2
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 R3 2·3·4 R4
is cleanly included into the following expression obtained by substitu-
tion:
int
Vtop (ϕ, λ, r) =
x ∑︂ ∞ (︃ )︃
rn H H2 H3
= Gρ n−2
− 1
(n − 1) 2 + 1
(n − 1) n 3 − · · · Pn (cos ψ) dσ ′ . (D.4)
σ R R 2 R 6 R
n=0

í Õ ! ¤. û
 The exterior potential of the condensation layer D.3
447
^ 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 . (6.4)
R R
Substituting this into the previous yields (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 ′ .
σ R R R
n=0
(D.5)

^ 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 2

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 H3
= −Gρ r
1 2 1
nH + 6 n (n + 3) + · · · Pn (cos ψ) dσ ′ .
σ 2 R
n=0

2 Theoretically speaking, the exterior space is the space outside a sphere that encloses
all of the Earth’s topography, a so-called Brillouin sphere. Practice is less restrictive.

í Õ ! ¤. û
448 D Helmert condensation

Often, we define the degree constituents of powers of height H (compare

asteosuus- the degree constituent equation 3.9), as follows:
yhtälö x
def 2n + 1
Hνn (ϕ, λ) = Hν (ϕ ′ , λ ′ ) Pn (cos ψ) dσ ′ , (D.6)
4π σ
with which it holds that
∑︂
∞
ν
H (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 D.4
449
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, 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 ∑︂ ∞ (︃ )︃
H3
= Gρ 1 2 1
(n + 1) H − 6 (n − 2) (n + 1) + · · · Pn (cos ψ) dσ ′ .
σ 2 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 1 H3n
)︃
int
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
∞
H3
(︃ )︃
1
= γ H − (n − 2) n + · · · . (D.8)
2n + 1 2 n 6 R
n=0

í Õ ! ¤. û
450 D Helmert condensation

The term n = 0 yields the indirect effect of a constant terrain H = H =

H0 : 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.
3 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 in spherical geometry:

x (︃ )︃
2 ∂ 1
VP = GR µQ dσQ .
σ ∂rQ ℓPQ

We use the expansion into Legendre polynomials, equation 8.7:

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 D.5
451
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 3.9: kehitelmä
x
2n + 1
(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: regionally though not globally.
Thus we obtain again an estimate for the indirect effect of Helmert
condensation. In geoid computation by means of this method this
represents the shift in geoid surface caused by the condensation, which
must be undone, that is, accounted for with the opposite algebraic
sign. In other words, when looked upon as a remove–restore method, it poistamis-
constitutes its “restore” step: entistämis-
menetelmä
V AB H πGρH2
δNHelmert = γ ≈ 12 γ = γ .
For comparison, the more precise expansion D.8 yields in approxima-
tion for larger n values
4πGρ ∑︂
∞
n+1 2 πGρ ∑︂ 2
∞
πGρH2
1
δNHelmert ≈ γ · Hn ≈ γ Hn = γ ,
2 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 of
def
∆ϕ, ∆λ, and ∆r. Look at the difference in flux of vector field a = ∇V vuo
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
the divergence 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 ortonormaali
of type “north-east-up”. The vector e1 points to the local north, the kanta
vector e2 to the east, and the vector e3 “up”, in the radial direction. 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− ).

– 453 –
454 E 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. The Gauss divergence 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 E.1
455
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
ω− 2
r = r cos ϕ ∆ϕ ∆λ, ω+
r = (r + ∆r) cos ϕ ∆ϕ ∆λ,

í Õ ! ¤. û
456 E 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, yield-
ing for the second contribution to the Laplace operator

2 ∂V
∆IIr V = r .
∂r
This in addition to the first contribution
[︁ ∂ ]︁+
a+ −
3 − a3 ∂r
V − ∂2 V
∆Ir V = ⟨∇a3 · e3 ⟩ = = ≈ .
∆r ∆r ∂r2
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
= + + + − 2 , (E.1)
∂r2 r2 cos2 ϕ ∂λ2 r2 ∂ϕ2 r ∂r r ∂ϕ

equivalent to equation 2.9.

^ 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 E.1 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 E.2
457
This must again apply for all values r and ϕ and thus both expressions
can only be equal to 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.2)
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
Yn (ϕ, λ)
Vnint (ϕ, λ, r) = rn Yn (ϕ, λ), Vnext (ϕ, λ, r) = , n ∈ N0 ,
rn+1
equations 2.10.

^ E.2.2 Solving for surface harmonics

Both solutions q, the non-negative and the negative one, yield on
substitution into equation E.2 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
458 E 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 to be positive and call m2 :
(︃ )︃
∂2 F ∂F m2 ∂2 L
2
− tan ϕ + n (n + 1) − F = 0, + m2 L = 0.
∂ϕ ∂ϕ 2
cos ϕ ∂λ 2

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.
harmoninen The second is the classical harmonic oscillator, with solutions1
värähtelijä
1 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
V int (ϕ, λ, r) = rn Pnm (sin ϕ) (anm cos mλ + bnm sin mλ) ,
n=0 m=0
∑︂
∞
1 ∑︂
n
ext
V (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 exterior 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. 141

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. 312

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. 373

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. 249

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.
312

– 459 –
460 ABCDEFGHIKLMNOPRSTVWY Bibliography

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. 64, 140

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. 249

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. 2

BGI, EGM2008. EGM2008 anomaly maps visualization. URL http://bgi.obs-

mip.fr/data-products/outils/egm2008-anomaly-maps-visualization/.
Accessed 17th September, 2021. 126, 136

BGI, WGM2012. WGM2012 maps visualization/extraction. URL

http://bgi.obs-mip.fr/data-products/outils/wgm2012-maps-
visualizationextraction/. Accessed 17th September, 2021. 154

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. 403

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. 247, 332

G. P. Bottoni and R. Barzaghi. Fast collocation. Bulletin Géodésique, 67(2):

119–126, 1993. URL https://doi.org/10.1007/BF01371375. 291

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

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
461
http://www.helmut-moritz.at/SciencePage/Molodensky.pdf. Helmut
Moritz and Maria I. Yurkina, editors. Accessed 1st March, 2020. 170

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. 321

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. 321

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. 321

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. 94

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. 223

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. 339

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. 377

í ¤. û
462 ABCDEFGHIKLMNOPRSTVWY Bibliography

Climate Research Unit. University of East Anglia. URL

https://crudata.uea.ac.uk/cru/data/soi/. Accessed 13th January, 2022. 354

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. 321

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. 29

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. 342

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. 402

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, 34, 46, 57, 90, 433

M. Ekman. Postglacial rebound and sea level phenomena with special

reference to Fennoscandia and the Baltic Sea. In Kakkuri (1993), pages 7–70.
337, 394

Encyclopaedia Britannica, Moho. URL

https://www.britannica.com/science/Moho. Accessed 22nd January, 2020.
159

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
463
L. Eötvös. Three Fundamental Papers of Loránd Eötvös. Loránd Eötvös
Geophysical Institute of Hungary, 1998. ISBN 963-7135-02-2. Editor Zoltán
Szabó. 323

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. 402

G. Farmelo. The Strangest Man. Basic Books, reprint edition, 2011. ISBN
978-0-4650-2210-6. 27

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. 342

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. 225

FFTW Home Page. URL https://www.fftw.org. Accessed 15th May, 2019. 441

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. 251

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. 403

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

í ¤. û
464 ABCDEFGHIKLMNOPRSTVWY Bibliography

14/10/8-WG_geoid_2010_March_presentation_ForsbergStrykowski.pdf.
Accessed 11th May, 2019. 403

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. 380

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. 247

R. Forsberg and M. Vermeer. A generalized Strang van Hees approach to fast

geopotential inversion. Manuscripta geodaetica, 17:302–314, 1992. 243

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. 94

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.
441

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. 249

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

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
465
https://www.researchgate.net/publication/275029640_SPACE-
WISE_GRIDS_OF_GRAVITY_GRADIENTS_FROM_GOCE_DATA_AT_N
OMINAL_SATELLITE_ALTITUDE. Accessed 11th May, 2019. 288

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. 340

GRACE Follow-On Mission. Jet Propulsion Laboratory. URL

https://gracefo.jpl.nasa.gov/mission/overview/. Accessed 1st March,
2020. 379

GRACE Mission. Measuring Earth’s Surface Mass and Water Changes. Jet
Propulsion Laboratory. URL https://grace.jpl.nasa.gov/. Accessed 1st
March, 2020. 378

GRACE Mission, hydrology. NASA. URL

https://commons.wikimedia.org/wiki/File:
Global_Gravity_Anomaly_Animation_over_LAND.gif. Accessed 1st March,
2020. 379

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. 29

Green’s Windmill. Green’s windmill and science centre. URL

https://www.greensmill.org.uk/. Accessed 10th May, 2019. 29

GWR Instruments, Inc., iGRAV® Gravity Sensors. URL

https://www.gwrinstruments.com/igrav-gravity-sensors.html. Accessed
22nd September, 2022. 317

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. 243

í ¤. û
466 ABCDEFGHIKLMNOPRSTVWY Bibliography

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. 332

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. 251

M. Heikkinen. Solving the shape of the Earth by using digital density models.
Report 81:2, Finnish Geodetic Institute, Helsinki, 1981. 101, 102, 104, 318

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. 155

W. A. Heiskanen and H. Moritz. Physical Geodesy. W. H. Freeman and

Company, San Francisco, London, 1967. 35, 53, 57, 68, 71, 79, 80, 81, 94, 95,
97, 98, 104, 105, 106, 114, 118, 153, 155, 157, 177, 179, 180, 184, 193, 194, 201,
202, 206, 211, 293, 404, 435

W. F. Hermans. Beyond Sleep. Harry N. Abrams, reprint edition, 2007. ISBN

978-1-5856-7583-8. 146

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. 140

R. A. Hirvonen. The continental undulations of the geoid. PhD thesis, Helsinki

University of Technology, 1934. Finnish Geodetic Institute publication 19.
403

R. A. Hirvonen. Relative Bestimmungen der Schwerkraft in Finnland in den Jahren

1931, 1933 und 1935. Publication 23, Finnish Geodetic Institute, Helsinki,
1937. 300

B. Hofmann-Wellenhof and H. Moritz. Physical Geodesy. Springer-Verlag Wien

GmbH, 2006. Second, revised edition. 404

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
467
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. 144

E. Hytönen. Absolute gravity measurement with long wire pendulum. PhD thesis,
University of Helsinki, 1972. Finnish Geodetic Institute publication 75. 301

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. 314

ISG, Geoid Schools. International Service for the Geoid. URL

http://www.isgeoid.polimi.it/Schools/schools.html. Accessed 1st March,
2020. 247

E. Kääriäinen. The Second Levelling of Finland in 1935–1955. Publication 61,

Finnish Geodetic Institute, Helsinki, 1966. 165, 166, 180

J. Kääriäinen and H. Ruotsalainen. Tilt Measurements in the Underground

Laboratory Lohja 2, Finland, in 1977–1987. Publication 110, Finnish Geodetic
Institute, Helsinki, 1989. 394

J. Kakkuri, editor. Geodesy and Geophysics, lecture notes, NKG Autumn School
1992, Finnish Geodetic Institute publication 115, 1993. 462, 473, 474

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. 146,
404

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.

í ¤. û
468 ABCDEFGHIKLMNOPRSTVWY Bibliography

AGU Fall Meeting Abstracts, page A4, December 2012. URL

https://ui.adsabs.harvard.edu/abs/2012AGUFM.G12A..04K/abstract.
Accessed 11th May, 2019. 321

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. 155

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. 140, 141

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. 347

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. 391

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
gravity field modeling. Journal of Geophysical Research: Solid Earth, 122:
10 586–10 604, 2017. URL https://doi.org/10.1002/2017JB014425. 321

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

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
469
ent/uploads/2014/10/1-Makinen_et_al_land_uplift_gravity_lines.pdf.
Accessed 22nd September, 2021. 335

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. 130, 218

P. J. Melchior. The Tides of the Planet Earth. Pergamon Press, Oxford, 1978.
ISBN 978-0-0802-6248-2. 389, 392

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. 346

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). 129, 173, 223, 404

D. Monniaux. Autograv CG5, 2011. URL https:

//commons.wikimedia.org/wiki/File:Autograv_CG5_P1150838.JPG. ©
2011 David Monniaux (GFDL). Accessed 13th May, 2019. 301

H. Moritz. Advanced Physical Geodesy. H. Wichmann Verlag, Karlsruhe, 1980.

ISBN 978-3-87-907106-7. 404, 444

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. 353

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. 354

NOAA, Ocean currents. How does the ocean affect climate and weather on
land? NOAA Ocean Exploration and Research. URL

í ¤. û
470 ABCDEFGHIKLMNOPRSTVWY Bibliography

https://oceanexplorer.noaa.gov/facts/climate.html. Accessed 1st March,

2020. 341

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. 335

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. 252

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. 224

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. 77

W. R. Peltier. Home page, W. R. Peltier, FRSC. URL

https://www.atmosp.physics.utoronto.ca/~peltier/data.php. Accessed
22nd September, 2021. 342

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.
Special issue: Quaternary Ice Sheet-Ocean Interactions and Landscape
Responses. 342

U. Pesonen. Relative Bestimmungen der Schwerkraft in Finnland in den Jahren

1926–1929. Publication 13, Finnish Geodetic Institute, Helsinki, 1930. 300

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
471
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. 375

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. 395

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. 145

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. 145

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. 145

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. 187

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. 155

í ¤. û
472 ABCDEFGHIKLMNOPRSTVWY Bibliography

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. 287

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. 300

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. 345

C. F. Ropelewski and P. D. Jones. An extension of the Tahiti-Darwin Southern

Oscillation Index. Monthly Weather Review, 115:2161–2165, 1987. URL
https://doi.org/10.1175/1520-0493(1987)115<2161:AEOTTS>2.0.CO;2.
Accessed 13th January, 2022. 354

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. 340

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. 394

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. 365

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
473
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. 375

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. 380

Sea Level Research Group. University of Colorado. URL

https://sealevel.colorado.edu/. Accessed 22nd September, 2021. 354

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. 461, 463

SourceForge, Maxima. Maxima, a computer algebra system. URL

https://maxima.sourceforge.io/. Accessed 22nd September, 2021. 360

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. 342

G. Strang van Hees. Stokes’ Formula Using Fast Fourier Techniques.

Manuscripta geodaetica, 15:235–239, 1990. 238

W. Torge. Gravimetry. de Gruyter, Berlin, New York, 1989. ISBN

978-3-11-010702-9. 404

W. Torge. Gravity and tectonics. In Kakkuri (1993), pages 131–172. 398

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. 287

í ¤. û
474 ABCDEFGHIKLMNOPRSTVWY Bibliography

Tytyri Mine Experience. URL

https://www.tytyrielamyskaivos.fi/en/mine-experience. Accessed 17th
September, 2021. 394

P. Vaníček and E. Krakiwsky. Geodesy – The Concepts. Elsevier Science

Publishers, second edition, 1987. ISBN 978-0-4448-7777-2. 404

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. 300

M. Vermeer. Chronometric levelling. Report 83:2, Finnish Geodetic Institute,

1983a. 186, 313

M. Vermeer. A new Seasat altimetric geoid for the Baltic. Report 83:4, Finnish
Geodetic Institute, 1983b. 351

M. Vermeer. Geoid studies on Finland and the Baltic. PhD thesis, University of
Helsinki, 1984. Report 84:3, Finnish Geodetic Institute. 34, 113

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. 243

M. Vermeer. Geoid determination using frequency domain techniques. In

Kakkuri (1993), pages 183–200. 234

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. 462, 475

H. Virtanen. On superconducting gravimeter observations above 8 mHz at

the Metsähovi station. Report 98:5, Finnish Geodetic Institute, Masala, 1998.
315

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. 316

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
475
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. 315

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. 394

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. 372

D. Watts. Fourifier, 2004. URL

https://ejectamenta.com/imaging-experiments/fourifier/. Accessed 17th
September, 2021. 246

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. 155

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. 82

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. 346

í ¤. û
476 ABCDEFGHIKLMNOPRSTVWY Bibliography

Wikipedia, The Aerospace Corporation. URL

https://en.wikipedia.org/wiki/The_Aerospace_Corporation. Accessed
23rd April, 2019. 221

Wikipedia, Declination. URL https://en.wikipedia.org/wiki/Declination.

Accessed 29th February, 2020. 387

Wikipedia, Dislocation. URL https://en.wikipedia.org/wiki/Dislocation.

Accessed 23rd April, 2019. 307

Wikipedia, Double-slit experiment. URL

https://en.wikipedia.org/wiki/Double-slit_experiment. Accessed 18th
February, 2020. 312

Wikipedia, Earth normal modes. URL

https://en.wikipedia.org/wiki/Seismic_wave#Normal_modes. Accessed
16th March, 2020. 316

Wikipedia, Hour angle. URL https://en.wikipedia.org/wiki/Hour_angle.

Accessed 29th February, 2020. 387

Wikipedia, John Pratt. URL

https://en.wikipedia.org/wiki/John_Pratt_(Archdeacon_of_Calcutta).
Accessed 23rd April, 2019. 145

Wikipedia, Mu-metal. URL https://en.wikipedia.org/wiki/Mu-metal.

Accessed 23rd April, 2019. 315

Wikipedia, Pendulum clock. URL

https://en.wikipedia.org/wiki/Pendulum_clock. Accessed 23rd April,
2019. 299

Wikipedia, Saros. URL https://en.wikipedia.org/wiki/Saros_(astronomy).

Accessed 23rd April, 2019. 331

Wikipedia, Sea level rise. URL https://en.wikipedia.org/wiki/Sea_level_rise.

Accessed 23rd April, 2019. 341

Wikipedia, Seasat conspiracy theory. URL

https://en.wikipedia.org/wiki/Seasat#Conspiracy_theory. Accessed 23rd
April, 2019. 351

í ¤. û
Bibliography ABCDEFGHIKLMNOPRSTVWY
477
Wikipedia, Strengthening mechanisms of materials. URL https:
//en.wikipedia.org/wiki/Strengthening_mechanisms_of_materials.
Accessed 10th May, 2019. 307

Wikipedia, Sverdrup. URL https://en.wikipedia.org/wiki/Sverdrup.

Accessed 23rd April, 2019. 339

Wikipedia, Zero-length springs. URL

https://en.wikipedia.org/wiki/Spring_(device)#Zero-length_springs.
Accessed 23rd April, 2019. 304

Wolfram Demonstrations, Difference formula for cosine. URL

https://demonstrations.wolfram.com/DifferenceFormulaForCosine/.
Accessed 7th April, 2019. 197
∑︁
Wolfram Functions, ∞ k=1
cos kx
k . URL https:
//functions.wolfram.com/ElementaryFunctions/Cos/23/02/0001/.
Accessed 25th February, 2020. 198

Wolfram MathWorld, Spherical Harmonic Addition Theorem. URL https:

//mathworld.wolfram.com/SphericalHarmonicAdditionTheorem.html.
Accessed 11th May, 2019. 387

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. 221

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. 333

YouTube, Hammer vs. Feather, 2010. URL https://www.youtube.com/watc

h?feature=player_embedded&v=KDp1tiUsZw8#! Accessed 7th April, 2019.
4

í ¤. û
478 ABCDEFGHIKLMNOPRSTVWY Bibliography

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. 287

í ¤. û
 Index

ABCDEFGHIJKLMNOPQRSTUVWZ

A American Geophysical Union (AGU), 402

Aalto University, 404 amplifier, in fibreoptic cable, 187
Abell 1689, 2 Amsterdam (The Netherlands) datum,
acceleration 165, 331
geometric, 377 Andersen, Ole Balthasar, 249
of free fall, 308 angular distance, geocentric, 237
acceleration, measured by GNSS, 320 figure, 261
accelerometer on a satellite, 322 angular momentum
action at a distance, 1 as a vectorial product, 407
airborne gravimeter, 308 conservation, 409
airborne gravimetry figure, 409
description, 319 anomalous quantity, 87, 111, 112
aircraft motions, 319 Antarctica
Coriolis acceleration, 320 continental ice sheet, 340
flight height, 321 meltwater, 346
GNSS, 320 antimatter, 27
gravimetric mapping, 249 anti-root, under sea, 149
gravity, 320 Apollo project, 4
homogeneity, 322 Arabelos, Dimitris, 247
on-board measured gravity, 319 Archimedes’ law, 148
uncertainty of vertical acceleration, Arctic Ocean
321 ice cover, 374
Airy, George Biddell, 146 ice volume, 374, 375
Airy–Heiskanen hypothesis, 147, 154 argument of perigee, 366
Airy–Heiskanen model, 146 arrest (gravimeter), 308
Airy’s isostatic hypothesis, calculation ascending node, orbital, 366
formulas, 147 Asperger syndrome, 4
AltiKa (altimeter), 353 astatisation ratio, 305, 306
altimetric missions (table), 352 astatisation, the concept (figure), 307
altimetric satellite, orbit choice, 366 asthenosphere, 342

– 479 –
480 ABCDEFGHIJKLMNOPQRSTUVWZ Index

atmosphere Bouguer hypothesis, of land uplift, 334,

surface mass density, 317 335
total mass, 318 Bouguer plate
atmospheric drag compensation, 322 as approximation, 133
atmospheric loading, 398 figure, 133
atomic clock, 186 attraction, 132
autocovariance, 258 figure, 131
azimuth (figure), 261 double, 179
half, 139
B of air, 317
ballistic gravimetry Bouguer reduction, 131, 212
normal equations, 311 indirect effect, 155, 212
observation equations, 310, 311 mass effect of spherical reduction,
unknowns, 311 141
Baltic Sea simple, 134
airborne gravimetry, 321 Bouguer shell, attraction, 140
salinity gradient, 337 Bouguer, Pierre, 131
Seasat data, 351 Boulder, U. of Colorado at, USA, 308, 354
sea-surface topography, 337 boundary condition
base network measurement in gravity anomaly, 121
gravimetry, 313 periodicity, 432
“before present” (BP), 341 shoebox, 45
Bergensbanen (Norway), 149 boundary-value problem
BGI, 126, 136, 154, 401 definition, 34, 42
Bjerhammar sphere, 287 boundary surface, choice, 129
Bjerhammar, Arne, 287 free, 117
Blue Road Geotraverse project, 336 of Dirichlet, 35, 121, 204
body (extended), potential, 6 of Neumann, 74, 75, 121
body (pointlike), potential, 5 of physical geodesy, 121
body, exterior potential, 36 spectral solution, 121
bordering, of data and kernel (Fourier), third, 118, 121
244 bounded support, 233
Bose–Einstein condensate, 186, 312 Bruns equation, 114, 128
Bouguer anomaly Bruns vertical-gradient equation, 94
bathymetry, 133 Bruns, Ernst Heinrich, 94, 114
bias, 134, 136 bulldozer (metaphor), 142, 216
calculation steps, 138 Bureau Gravimétrique International, 126,
example, 139 136, 154, 401
interpolation, 134
prediction, 134 C
properties, 134 cap, spherical (geoid determination), 220
simple, 134 Cavendish, Henry, 3, 366
Southern Finland, 136 celestial sphere, 435
spherical, 140 Center for Space Research, U. of Texas,
bias, 141 USA, 378
terrain corrected, 135 centrifugal force, 89, 109
why, 133 expression, 90

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
481
figure, 89 calculation by FFT, 242
centrifugal potential, 88 notation, 233
expression, 90 convolution theorem, 233
CHAMP (satellite), 377 co-ordinate time (relativity), 186
accelerometer, 377 co-ordinates
figure, 378 ellipsoidal, 52
GPS receiver, 377 geodetic, 51
characteristic equation, 431 definition, 52
Chasles theorem, 34 natural, 97
equipotential surface as boundary, figure, 96
33, 34 polar, 418
non-uniqueness of mass distribution, rectangular, 4, 50
36 Copenhagen (Denmark),
Chasles, Michel, 34 geoid determination, 403
circle mean, M⃝ , 289 Coriolis acceleration
circular disk, attraction, 132 direction, 338
closing error (gravimetry), 307 in airborne gravimetry, 320
co-geoid, 130 of an ocean current, 338
collocation, least-squares (LSC), 247 Coriolis force, 90
description, 263 Coriolis, Gaspard-Gustave, 90
FFT, 288 corkscrew rule, of the vectorial product,
figure, 270 408, 411
flexibility, 275 correlation length, 268, 269, 297
solution, 271 correlation, quasi-geoid & topography,
theory, 274 174, 175
Columbus geoid (model), 146, 404 correspondence, integral & spectral
commutative diagram equations, 193
FFT, 237 cosine rule on the sphere, 237, 387
radial shift, 74 half-angle, 238
remove--restore, 219 covariance function
vertical shift, 47 definition, 261
compensation depth, 148, 154 empirical, 287
components, of a vector, 410 Gauss–Markov, 269
condensation layer, exterior potential, 447 global, 288
confocality, 53 isotropy, 279
conservation law, 23 of gravity anomalies, 285
conservative field of Hirvonen, 268, 272
definition, 5, 416 figure, 270
as the gradient of a potential, 420 of the disturbing potential, 278, 295
divergence, 416 in space, 282, 284
potential, 416 spectral representation, 279
continental ice sheets and sea level, 340 cross product, see vectorial product
continuity equation, 421 cross-covariance, 258
convection, in the Earth’s mantle, 155 crossover adjustment
convergence, uniform, 427, 444 a priori uncertainties, 362
convolution allowed datum transformations, 364,

í ¤. û
482 ABCDEFGHIJKLMNOPQRSTUVWZ Index

365 potential field, 70

constant and trend, 361 dipole moment, 20
constant orbit correction, 357 dipole-density layer, 20
constraint, 359, 362 Dirac, Paul, 27
datum defect, 362 directional sphere, 435
example, 358, 363 Dirichlet, Peter Gustav Lejeune, 35
global, 365 dislocation (crystal), 307
least-squares solution, 359 disturbing potential, 111
observation equation, 357 definition, 105
sea-surface variability, 376 at terrain level, 213
wire-frame model, 362 degree constituents, 107
CryoSat-2 (satellite), 353 degrees 0 and 1, 107, 119
curl (operator), 414 isotropy, 278
figure, 415 spherical-harmonic expansion, 106
interpretation, 414 divergence (operator)
of a gradient, 415 conservative field, 416
of a vector field, 413 figure, 414
of the wind field, 414 interpretation, 413
is a source function, 22
D vector field, 412
Darwin, Sir George, 390 Doodson, Arthur Thomas, 390
declination, of the Moon, 387 Doodson’s constant, 390
Defense Mapping Agency (DMA), US, 76 numerical value, 395
deformation, viscous (GIA), 346 dot product, see scalar product
degree constituent equation, 71 double-slit experiment (quantum theory),
proof, 71 312
degree variance drift (gravimeter), 307
notation, 281
of gravity anomalies, 286, 287 E
of the disturbing potential, 279, 281, Earth
284 blue film, 378
degree variance formula, 287 dipole moment, 70
degree, harmonic, 54 exterior potential, 155
delta function, Dirac’s, 27 flattening, 101
limit of Poisson kernel, 207 gravitational field, 41
delta, Kronecker’s and Dirac’s, 290 inertial tensor, 70
density profile, 36 internal mass distribution, 35
difference magnetic field, 155
geoid – free-air geoid, 177 quadrupole moment, 70
height anomaly – free-air geoid radius, 18
height, 176 radius of curvature, transversal, 51
height anomaly – geoid height, 176 rheology, 345
orthometric height – normal height, rotation rate, 101
178 total mass, 76, 101
quasi-geoid – geoid, 177 Earth centre of mass as
dipole co-ordinate origin, 70, 119
at Earth’s centre, 71 Earth ellipsoid

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
483
equatorial radius, 51, 101 Euler’s identity, 440
first eccentricity, 51 European Geosciences Union (EGU), 402
polar radius, 101 European Space Agency (ESA)
semimajor axis, 101 CryoSat-2, 353
semiminor axis, 101 ERS-1/2, 352
earthquake, 347 GOCE, 379
eccentricity, orbital, 366 sea-surface topography map, 341
EGM96 (geopotential model), 76 Sentinel-3A, 353
coefficients, mean errors, 78 eustatic rise, of mean sea level, 333, 343
EGM2008 (geopotential model), 63, 77, evaluation functional, 256
126, 136 Everest, Mount, 150
Eiffel Tower, 94 exterior product, see vectorial product
72 names, 16, 17, 34, 46, 57, 90, 433
eigenvalue problem F
linear operator, 428 factorial, 57
matrix, 429 Falkland Islands, 380
self-adjoint operator, 430 Faller, James E., 308
variance matrix of location, 431 falling atoms, gravity potential, 313
Einstein summation convention, 265 Faraday, Michael, 23, 421
Einstein, Albert, 4 Fast Collocation, 291
El Niño Southern Oscillation (ENSO), 329 Fast Fourier Transform (FFT)
elastic properties algorithms, 441
of the Earth, 391 and convolution, 234
of the Earth’s crust, 398 collocation, 288, 291
electric currents in the Earth’s core, 155 commutative diagram, 236
ellipsoidal-harmonic expansion mixed-radix, 439
definition, 77 radix 2, 439
centrifugal potential, 97 terrain correction, 250, 252
computation, 82 fast Fourier transform (FFT)
convergence, 81 geoid determination, 236
normal potential, 81 Fastest Fourier Transform in the West
standard form, 80 (fftw, software), 441
RMS Empress of Ireland, 165 Father Point / Pointe-au-Père (Rimouski,
Envisat (satellite), 352 Quebec, Canada), 165
eötvös (unit), 120, 324 Fennoscandia
Eötvös tensor, 322 continental ice sheet, 151, 341
Eötvös, Loránd, 91 gravity line, 336
equivalence principle, 4, 91 longevity of heights, 332
ERS-1/2 (satellites), 352 Fennoscandian Shield,
escape velocity, 86 isostatic anomalies, 154
estimator, 265 Fermat’s last theorem, 94
mean error, 271 field equations
optimal, 266 of electromagnetism, 17
Eterna (software), 398 of gravitation, 1, 17
Euclidean space, 10 field line, 23, 421
Euler notation, 93 field theory of gravitation, 1
field, the concept, 41

í ¤. û
484 ABCDEFGHIJKLMNOPQRSTUVWZ Index

figure of the Earth, 91 free-air hypothesis, of land uplift, 335

FIN2000 (geoid model), 247, 332 freeboard (sea ice), 353, 374
construction, 332 French Academy of Sciences, 131
figure, 248 frequency domain (Fourier), 46, 235
land-uplift epoch, 332 fresh river water, variation, 329
precision, 247 function space, 424
FIN2005N00 (geoid model), 247, 403 on the circle, 424
precision, 247 on the sphere, 434
Finland, 247, 308 scalar product, 432
Finnish Geodetic Institute (FGI), function theory, 424
gravity-field research, 403 functional
Finnish Geospatial Research Institute definition, 256
(FGI), 404 linear
flow velocity (vector field), 23, 413 definition, 256
flux, 23 of the potential, 35
fluxion, 334 of the disturbing potential, 256
footprint, radar altimeter, 355, 373 spectral representation, 282
Forsberg, René, 247 fundamental equation of
Fourier basis functions physical geodesy, 118, 121
as a basis, 425
recursive calculation, 58 G
two-dimensional, 46 Galilei, Galileo, 4
Fourier coefficient, 46 gauge invariance, 17
Fourier series, 425 Gauss divergence theorem, 420
Fourier sine expansion, 45 presentation, 22
Fourier transform book-keeping, 24
and tapering, 246 box, 24, 25
artefacts, 245 co-ordinate blob, 454
discrete, 234, 236, 439 eight-unit cube, 27
periodicity, 235, 243, 244 figure, 23, 421
reverse, 440 in terms of potential, 24
forward and reverse, 234 sphere, 26
of ℓ−3 , 251 Gauss, Carl Friedrich, 22
notation, 233 general relativity, 17, 186
step function, 427 Geodetic Reference System 1980,
Fourier, Joseph, 45 see GRS80
France (BGI, funding), 401 geographic mean
Francis, Olivier, 315 definition, 259
free oscillations, solid Earth, 316, 403 M, 260
periods, 316 M ′ , 262
free-air anomaly geographic variance, 259
definition, 124 geoid, 155
calculation, 124 definition, 91, 167
Southern Finland, 126 classical, 330
use, 125 fake, 180
free-air geoid, 176 true, 376

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
485
geoid computation education in geostrophic equations, 339
developing countries, 402 geostrophic equilibrium, 338
geoid computation research school, Germain curvature, 93
international, 402 Germain, Marie-Sophie, 93
geoid determination German Research Centre for Geosciences
1-D FFT, 243 (GFZ), 377
classical, 130 Gibbs phenomenon
comparison point, 332 Fourier transform, 245
FFT, 247 kernel modification, 222
research groups, 245 step function, 428
gravimetric, 34, 193 Gibbs, Josiah Willard, 222
principle, 192 glacial isostatic adjustment (GIA), 333,
2-D computation framework, 195 346
NKG, 403 glacier retreat, 150
satellite altimetry, 375 GM⊕ , best value, 6
software, 247 GNSS
spectral viewpoint, 75 height of gravimetric stations, 130
spherical cap, 220 in airborne gravimetry, 320
spherical FFT in height determination, 246
multi-band, 238 measuring ocean tidal loading, 398
Taylor expansion, 240 on an altimetric satellite, 376
standard crustal density, 180 positioning of tide gauges, 333, 337
geoid height or undulation GNSS levelling, 332
definition, 112 GOCE (satellite)
from satellite altimetry, 249 description, 379
global, 112 figure, 381
in Finland, 112 name, 382
geoid model ocean currents and heat transport,
Columbus, 146 339
computation, 246 precision, 381
global high resolution, 377 resolution, 322, 380
of Finland, 113 sea-surface topography, 341
geoid rise, 333, 334 GPS
geological map, density values, 180 on CHAMP satellite, 377
geophysical data record (GDR), 356, 373 reference system, 101
geophysical reduction, 130 GRACE (satellite pair)
geopotential description, 377, 379
image sharpness, 63 accelerometer, 378
level surface, 91 microwave link, 378
spectral expansion, 73 results, video, 380
geopotential number GRACE follow-on mission, 379
definition, 164 gradient
and levelled height, 167 of Earth attraction, 380
as energy level, 165 of gravity disturbance, 295
GEOS-3 (satellite), 351 of potential, normal direction, 34
Geosat (satellite), 351 gradient (operator), 9

í ¤. û
486 ABCDEFGHIJKLMNOPQRSTUVWZ Index

figure, 413 gravitation

interpretation, 412 field theory of, 1
linearity, 412 is an attraction, 5
of a scalar field, 412 law of, 3
gravimeter theory of, 1
absolute or ballistic, 308 gravitational constant, universal, 3, 385
principle of operation, 309 gravitational field, 23
cage, 308 conservativeness, 5
laser interferometer, 309 stationarity, 5
superspring, 309 gravitational gradiometer (GOCE)
air pressure variations, 318 accelerometer, 380
astatised, 303, 305 description, 380
invention, 305 figure, 381
atmospheric attraction, 317 theory, 380
atomic or quantum gravitational lens, 2
principle of operation, 312, 314 gravitational wave, 17
ambiguity problem, 313 gravity, 131, 336
figure, 314 definition, 308
calibration, 317 absolute measurement, Finland, 314
damping, 308, 319, 320 along levelling line, 185
FG5, 308 equatorial, 102
photo, 310 in the tropics, 300
IMGC-02, 312 local, 169
JILA, 308 variations, size, 120
LaCoste-Romberg, 303, 305, 309 gravity anomaly, 278
lever beam, 303 definition, 121
pendulum, 300 as a boundary condition, 121
submarine measurement, 300 as a functional, 257
registering, 398 atmospheric reduction, 318
sensitivity, 302, 304 block average, 225
spring precision, 275
equilibrium length, 302, 303 calculation, 124
instantaneous length, 305 degree constituents, 119
lengthening, 302 expression, 118
rest length, 302, 303 from satellite altimetry, 249
spring or relative, 301, 304 global average, 259
arrest, 308 harmonic continuation, 207
drift, 307 observations, 271
material properties, 306 special average, 227
thermostat, 308 gravity disturbance, 295
superconducting, 314, 315 definition, 115
principle of operation, 316 observing, 116
stability, 316 spectral representation, 116
trend, 317 gravity field
gravimetry, airborne, see airborne and geopotential, 420
gravimetry determination

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
487
CHAMP, 378 grid (Fourier)
GOCE, 381 formation, 234
fine structure, 376 interpolation, 236
observation density, 255 of gravity anomalies, 234
of mountains, 144 of the disturbing potential, 236
research of the Stokes kernel, 235
in Europe, 402 GRS80
in Finland, 403 definition, 101
in HUT, 404 atmospheric mass, 318
internationally, 401 GM⊕ , 6
residual, 218 spherical-harmonic coefficients, 104
statistical behaviour, 268 Gulf of Finland, airborne gravimetry, 321
temporal change, 377 Gulf Stream, 340
textbooks, 404 Guyana, French, 299
gravity formula, 87, 100, 124 GWR iGrav (gravimeter), 317
legacy, 102 GWR20 (gravimeter), 315
gravity gradient, 324
eötvös unit, 120 H
of Sun and Moon, 324 Haiyang-2A (satellite), 353
gravity mapping survey, 313 Hardanger plateau (Norway), 149
gravity measurement, reference surfaces harmonic downwards continuation
(figure), 117 existence, 204, 215
gravity potential, 91 exterior field, 217
gravity versus gravitation, 89, 91 Helmert condensation, 445
gravity-gradient tensor, 322 of gravity anomaly, 216
measurement, 323 of r ∆g, 205
GRAVSOFT (software), 247 of the exterior field, 217
Green equivalent-layer theorem, 34 harmonic field
Green, George, 29 definition, 16
Greenland attenuation with height, 55
continental ice sheet, 340 figure, 47
meltwater, 346 radial shift, 74
Greenland Aerogeophysics Project (GAP), r ∆g, 204
321 vertical shift, 47
Green’s first theorem, 29 harmonic oscillator, 44, 458
Green’s function Hayford, John Fillmore, 145
of sea level, 344 height
of the geopotential, 344 above mean sea level, 167
of vertical displacement, 345 above the reference ellipsoid, 52, 113
Green’s second theorem, 29 and geopotential number, 167
Green’s third theorem, 29 height anomaly
boundary point, 31 definition, 169
exterior point, 30 telluroid mapping, 123
exterior space, 32 three-dimensional, 123
interior point, 30, 31 height system, national, 331
Greenwich meridian, 50 height transformation surface, 332, 403
Heiskanen, Veikko Aleksanteri, 146, 404

í ¤. û
488 ABCDEFGHIJKLMNOPQRSTUVWZ Index

helicopter (airborne gravimetry), 321 photo, 315

Helmert condensation, 212 International Association of Geodesy
description, 142 (IAG), 247, 401
as a dipole-density layer, 450 International Geodynamics and Earth
condensation-layer potential, 443 Tide Service (IGETS), 315
figure, 143 International Geoid Commission (IGeC),
gravity effect, 449 402
indirect effect, 449 Subcommission for Europe, 402
constant terrain, 450 International Geoid Service (IGeS), 402
dipole method, 451 International Gravimetric Bureau (BGI),
mass conservation, 143 126, 136, 154, 401
topographic potential, 443 International Gravity Field Service
total potential, 447 (IGFS), 401, 402
Helmert height International Isostatic Institute, 404
definition, 179 International Service for the Geoid (ISG),
as approximation, 179 401
Helmert, Friedrich Robert, 142 International Union of Geodesy and
Helsinki astronomical observatory, Geophysics (IUGG), 401
reference benchmark, 165, 166 invariant, 406, 432
Helsinki harbour (N60), 331 inversion calculation (FFT), 242
Helsinki University of Technology inverted barometer, variation, 329
(HUT, TKK), 404 isostasy
Hirvonen, Reino Antero, 268 continental ice sheets, 150
Hirvonen’s geoid model, 403 figure, 145
Hofmann-Wellenhof, Bernhard, 404 modern understanding, 150, 151
mid-Holocene highstand, 342 paleo-research, 150
homogeneity assumption, 259 isostasy hypothesis, 145
and the covariance function, 260 isostatic anomaly
homogeneity of gravimetric data, 322 definition, 152
homogeneous prediction, 274, 276 Southern Finland, 154
hour angle, of the Moon, 387 isostatic compensation, 147
Hubble Space Telescope, 2 definition, 145
Huygens, Christiaan, 299 percentage, 155
isostatic geoid, 153
I why of interest, 155
IAG, 247, 401 isostatic hypothesis, 153
ice, multi-year, 375 isostatic reduction
ice-load history, 345 description, 155
ill-posed problem, 215 co-geoid, 153
inclination, orbital, 366, 370 indirect effect, 153, 159, 212
of the Moon, 388 mass conservation, 153
incompressibility, 23, 422 mass-density layer method, 155
indirect effect, 130 figure, 157
Institut für Erdmessung (Hannover, purposes, 152
Germany), 245, 402 residual field, 152
intercomparison, of absolute gravimeters, isotropic density distribution, 9
312, 314

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
489
isotropic process, 268 Wong–Gore, 221
isotropy and spectral representation, 201 KKJ, 124
isotropy assumption, 262 Knudsen, Per, 247, 249
Italy (ISG, funding), 402 Kolkata (India), 145
iteration Kronstadt (Russia) datum, 165
calculation of normal height, 181
calculation of orthometric height, L
179 LaCoste, Lucien, 305
LAGEOS (satellite), 366
J land uplift
J2 (dynamic flattening), 77, 101, 367, 368 absolute, 333
Jacobi, Carl Gustav Jacob, 68 effect on height system, 331
Jacobi’s determinant post-glacial
definition, 418 mechanism, 335
Bouguer-plate transformation, 132 relative, 333
map projection co-ordinates, 233 Laplace equation, 16
polar co-ordinates, 418 definition, 41
spherical co-ordinates (ϕ, λ), 231 basis solutions, 53
spherical co-ordinates (ψ, α), 68 co-ordinate transformation, 42
Jason (satellites) in ellipsoidal co-ordinates, 77
description, 353 in polar co-ordinates, 48
orbit choice, 369 in rectangular co-ordinates, 43
Java Sea (Dutch Indies, Indonesia), 300 in spherical co-ordinates, 53, 453
Jerry (GRACE satellite), 378 linearity, 42
local field behaviour, 42
K solving, 41
Kääriäinen, Jussi, 321 Laplace operator (∆), 416, 417
Kaivopuisto (Helsinki, Finland), definition, 16, 41
reference benchmark, 165, 166, linearity, 256
331 Laplace, Pierre-Simon, 16, 416
Kater, Henry, 300 Lapland, grade measurement, 131
Kaula, William, 287 latitude
Kepler, Johannes, 365 geocentric, 50
Simpson’s rule, 226 geodetic, 108
Kepler’s orbital elements, 366 reduced, 99, 108
Kepler’s second law, 408 definition, 52
figure, 409 types
Kepler’s third law, 365 figure, 100
octave script, 372 relationships, 99
kernel function grid matrix (FFT), 242 Laurentide ice sheet, 341
kernel modification law of motion, Newton’s, 4
advanced, 223 Legendre function, 54, 57
coefficients, 223 associated
degree, 221 algebraic-sign domains, 85
figure, 222 algebraic-sign intervals, 61
sharp cut-off, 222 figure, 60
soft cut-off, 222 fully normalised, 68

í ¤. û
490 ABCDEFGHIJKLMNOPQRSTUVWZ Index

symmetries, 60, 64, 84 Love number

table, 60 dependence on tidal period, 393
fully normalised, 436 determination, 393
of the second kind, 79 by GNSS, 393
table, 79 elastic, 346, 391
Legendre polynomial Hn , 392
algebraic-sign intervals, 59 Kn , 393
figure, 58 Ln , 392
fully normalised, 68 viscous, 347
symmetries, 59 Love, Augustus, 391
table, 58 LSC, see collocation, least-squares
Legendre polynomials lunar laser ranging (LLR), 308
as a basis, 434
generating function, 202 M
geometry, 202 Mäkinen, Jaakko, 169
orthogonality Maldives (Indian Ocean), 342
map projection co-ordinates (figure), 232
[︁ ]︁
on the interval −1, +1 , 67
on the unit sphere, 68 mareograph, 333
orthonormality on the unit sphere, Mars (planet), gravity field, 287
68 mass line, potential, 14
Legendre, Adrien-Marie, 57 mass point, underground, 297
Legendre’s equation, 458 mass surface density, SI unit, 19
Lego™ brick, 26 mass-density layer
Leibniz University (Hannover, Germany), double, 20, 33
402 Helmert condensation, 142
Leibniz, Gottfried Wilhelm, 1 single, 18, 33
level or equipotential surface, 92 mass-point set, potential, 11
level surface matter density function, 11
curvature, 92, 323 matter waves, coherence, 187
figure, 93 matter, conservation, 23
level surfaces matter-wave phase angle, 312
figure, 115 is a clock, 313
parallellity, 167 Mauna Kea (Hawaii), 162
levelling, 96 Maupertuis, Pierre de, 131
principle, 163 Maxwell, James Clerk, 17
figure, 164 mean geoid, 395
geostrophic, 337 mean sea level
relativistic, 186 definition, 329
steric, 337 concept, 330
levelling (gravimeter), 304 global, 333
levelling instrument, 163 mean location, 120
lever motion, at continent edge, 342 Meissner effect, 315
linearisation, of the free-air anomaly, 124 Melchior, Paul, 389
Liouville, Joseph, 433 meridian convergence, 237
longitude, of the Moon, 388 meridian ellipse
lookup table (FFT), 441 figure, 100
focal points, 53

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
491
mesoscale eddy, 329, 340 National Map Grid Co-ordinate System
metal fatigue, 307 (KKJ), 124
metallurgy, 307 NAVD88, North American
Metsähovi research station (Finland), 316 Vertical Datum 1988, 165
N2000 reference benchmark, 165, 331 network hierarchy in gravimetry, 313
superconducting gravimeter, 315 Neumann, Carl Gottfried, 74
microgal (µGal), 120 Newton, Isaac, 3
microseismicity, effect on gravimeter, 306, Newton’s law of gravitation, 3
309 Newton’s law of motion, 4
Milan (Italy), ISG office, 402 Newton’s theory of gravitation, 1
milligal (mGal), 120 Niethammer, Theodor, 180
mixed covariance, 286 Niethammer’s method, 180
modal relaxation time, 347 NKG2004 (geoid model), 403
modal strength, 347 NKG2015 (geoid model), 403
Mohorovičić, Andrija, 159 NN (height system), 165
Mohorovičić discontinuity (“Moho”), 159, noise (definition), 264
252 noise variance matrix, 265, 275
Molodensky theory, 34, 123, 129, 168 Nordiska Kommissionen för Geodesi (NKG),
Molodensky, Mikhail Sergeevich, 34 403
book, 404 norm, of a vector, 5, 406, 424
photo, 170 Normaal Amsterdams Peil (NAP), 165
Molodensky’s method normal correction (NC)
evaluation point as reference level, equation, 185
214 for benchmark interval, 185
height anomaly, 215 normal gravitational potential,
linearisation, 213 spherical-harmonic expansion,
vertical gravity gradient, 211 104
Molodensky’s realisation, 170, 173 normal gravity
graphic cartoon, 173 definition, 87
Molodensky’s truncation coefficients, 224 at sea level, 125
monopole at Earth’s centre, 71 GRS80, 101
monsoon, 379 in a known location, 116, 117
Moritz, Helmut, 404 linearity along the plumb line, 170,
mu-metal, 315 173, 181
Munk, Walter, 353 on the reference ellipsoid, 99, 100
normal gravity field
N atmospheric mass, 318
N60 (height system), 164, 331 figure, 88
land uplift, 331 normal gravity vector, 171
reference surface, 247 normal height
N2000 (height system), 165, 247, 331 definition, 174
nabla (∇, operator), 10, 411 calculation, 181
National Geospatial-Intelligence Agency operationality, 175
(NGA), US, 76, 401 practical calculation, 185
National Imagery and Mapping Agency precise calculation, 181
(NIMA), US, 76 normal plumb line

í ¤. û
492 ABCDEFGHIJKLMNOPQRSTUVWZ Index

curvature, 95, 171 P

direction, 111 Paul’s coefficients, 224
normal potential Peltier effect, 342
definition, 87 Peltier, W. Richard, 342
GRS80, 101 pendulum
on the reference ellipsoid, 100, 171 absoluteness of measurement, 301
over the equator, 102, 103 period, 299
Norwegian Sea, 149 pivot, 301
Nottingham (Great Britain), 29 pendulum clock, 299
Nouvel, Henri SJ, 165 pendulum equation, 299
Peru, grade measurement, 131
O petroleum extraction industry,
obliquity, of the Earth’s rotation axis, 399 gravimetry, 247
ocean current physical geodesy
inversion problem, 381 geometry and physics, 50
transversal tilt, 338 potential convention, 10
unit, 339 textbooks, 404
variation, 339 physical theory, nature of, 1
ocean tidal loading, 398 Pizzetti, Paolo, 99
octave (programming language), 65 plasticity, 306
Ohio (USA), 146, 269 plate tectonics, 151
Ohio State University (OSU), 76, 404 Plesetsk Cosmodrome (Russia)
one-Earth problem, 259 CHAMP, 377
optical lattice clock, 186, 187 GOCE, 380
optimality, least-squares, 267 plumb line
orbit definition, 87
no-shadow, 369 bending towards mountain, 145
Sun-stationary, 369 curvature, 94, 95
order, harmonic, 54 figure, 94
orthometric correction (OC), 183 plumb-line deflection
equation, 184 definition, 88, 111
for benchmark interval, 185 and the geoid, 112, 127, 201
orthometric height as a functional, 257
definition, 97, 167 at sea, 351
calculation, 168 in Finland, 112
iteration, 168 inner zone, 227
practical calculation, 183 observed, 113
precise calculation, 178 plumb-line direction, 111
terrain density hypothesis, 175 absolute, 394
orthonormal basis (definition), 5, 410 Päijänne, Lake (Finland), 169
oscillation equation point mass
as an eigenvalue problem, 432 attraction vector, 10
astatised gravimeter, 306 underground, 34
quantisation, 433 point-mass assumption of
self-adjoint, 432 celestial mechanics, 13
spring gravimeter, 302 Poisson equation
OSU model, 76

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
493
definition, 17 calculation of Fourier
for the geopotential, 91 basis functions, 59
Poisson integral calculation of Legendre polynomials,
for computing a harmonic field 57
from surface values, 204 definition of normal height, 174
for r ∆g, 205 definition of orthometric height, 179
spectral form, 206 reference benchmark
Poisson kernel for gravity anomalies, N60, 166, 331
205–207 N2000, 331
Poisson, Siméon Denis, 17 reference ellipsoid
polynomial fit, of geoid surface, 332 as a level surface, 87, 100, 112
potential legacy, 102
definition, 5 normal potential at surface, 171
harmonic upwards continuation, 281 reference surfaces in gravity
origin of word, 29 measurement (figure), 117
uniqueness, 416 reference-surface thinking, 123
potential energy, 11 regularisation of the exterior field, 215
Potsdam system, 301, 313 relativity theory, 1
powers of height, degree constituents, 448 relativity, general, 17, 186
PRARE (positioning instrument), 352 “remove” step, 216
Pratt, John Henry, 145 remove–restore method, 130, 218, 219
Pratt–Hayford hypothesis, 145, 146 research school, international, 247
precession, nodal or orbital, 367 residual gravity field, 218
precise levelling residual terrain modelling
between tide gauges, 337 indirect effect, 218
height system creation, 331 residual terrain modelling (RTM), 215,
Prey reduction, 179 217
Prey, Adalbert, 179 as an interpolation method, 218
Principia (book), 3 “restore” step, 218
propagation of variances, 276 reversion pendulum, Kater’s, 300
propeller aircraft (airborne gravimetry), Richer, Jean, 299
321 rising dough model, of land uplift, 335
proper time (relativity), 186 river basin, tropical, 379
Robin, Victor Gustave, 121
Q Romberg, Arnold, 305
quadrature, block average, 226 Romberg, Werner, 226
quadrupole at Earth’s centre, 71 root of mountain, 144
quasi-geoid depth, 146, 148
concept, 123, 175 matter density, 145
figure, 174 rotational potential, 88
Royal Society of Edinburgh, 17
R Royal Society of London, 3, 17, 122
radar-altimeter calibration, 372
in-flight, 372 S
Raman effect, 312 Sacks, Oliver, 4
Rapp, Richard H., 76 sampling density, spatial, 255
recursion San Francisco (USA), 402

í ¤. û
494 ABCDEFGHIJKLMNOPQRSTUVWZ Index

Sandwell, David, 249 Schrödinger’s cat, 42, 314

Sapporo (Japan), 401 Schwarzschild metric, 186
SARAL (satellite), 353 Schwarzschild, Karl, 186
saros (lunar motion periodicity), 331 Scripps Institute of Oceanography, 249
satellite altimetry sea gravimeter, 308
description, 351, 355 sea ice, remote sensing of, 374
and levelling, 359, 362 sea-floor pressure, 378
geoid, 375 sea-level equation, 342
in archipelagos, 374 convolution, 344
measurement method, 353 equation, 342
measurement reduction, 356 figure, 343
observation equation, 357 Green’s function, 344, 346
orbit correction, 354 sea-level rise, 333
over ice sheets, 374 global, 353
repeat tracks, 376 Holocene, figure, 345
results, 373 sea-level variation, 329
retracking, 373, 374 Seasat (satellite), 351
return pulse sea-surface residual variation, 357
analysis, 373 sea-surface topography
half-height point, 373 definition, 330
sea ice, 374 and heat transport, 337
sea-surface variability, 375 and ocean currents, 340
satellite gravity mission, 376 change over time, 334
differences with airborne, 322 determination, 336
FFT, 249 GOCE map, 341
orbit height, 322 mapping, 352
satellite orbit seismicity during deglaciation, 347
choice, 365 self-adjoint differential equation, 432
no-shadow, 368 self-adjoint operator
polar holes, 369, 370 definition, 429
repeat period, 367 symmetric matrix, 430
resonance, 369 semimajor axis, orbital, 366
retrograde, 368, 370 Sentinel-3A (satellite), 353
Sun-stationary, 368 separation of variables
satellite-to-satellite tracking (SST), 379 polar co-ordinates, 48
Saturn (planet), ring, 299 rectangular co-ordinates, 43
scalar field (definition), 410 spherical co-ordinates, 456
scalar product spherical surface co-ordinates, 458
definition, 405 Shida, Toshi, 394
commutativity, 423 shoebox world, 45
linearity, 423 signal (definition), 264
of Legendre polynomials, 67, 434 signal covariance matrix, 264
of two functions, 424 signal variance, 268
on the sphere, 435 signal variance matrix, 263
properties, 406 gravity anomalies, 269
Schrödinger, Erwin, 42 significant wave height (SWH), 355

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
495
Simpson, Thomas, 226 figure, 7
Simpson’s rule potential, 6, 7
block average, 226 potential and attraction, figure, 9
nodal weights, 226 spherical-harmonic coefficient
sink (vector field), 22, 23, 413 as a functional, 257
figure, 414 fully normalised, 436
Skylab (space station), 351 spherical-harmonic expansion
snow clearing, 316 coefficients, 54
solar time, 369 degree-one part, 69
Solheim, Dag, 247 degree-zero part, 69
solid body, 11 first terms, 105
attraction, 12 global, 76
field at infinity, 12 model, 76
potential, 11 resolution, 63
total mass, 13 rotational symmetry, 66, 67
solid spherical harmonic, 53 spheroid
Somigliana, Carlo, 99 Bruns, 105
Somigliana–Pizzetti equation, 99 Helmert, 105
source (vector field), 22, 23, 413 spring balance, linear, 302
figure, 414 stabilised platform (gravimeter), 308, 319
space domain (Fourier), 46, 236 staff-reading difference, 183
spatial frequency (Fourier), 235, 251 steel manufacture, 307
spatial wavelength (Fourier), 235 stereographic map projection, 232
spectral coefficients, 54 Sterneck, Robert von, 300
spectral constituent function, 71 stochastic process
sphere, coated definition, 257
exterior attraction, 19 ergodicity, 260
exterior potential, 21 on the Earth’s surface, 258, 259
interior attraction, 19 stationarity, 263
interior potential, 21 variance function, 258
spherical co-ordinates, 50 Stokes curl theorem, 418
figure, 51 figure, 419
spherical harmonic Stokes equation
algebraic sign, 62 2-D simulation, 200
sectorial, 61, 62 and harmonicity, 212
semi-wavelength, 62 convolution, 231
table, 64 differentiation, 201
symmetries, 62 disturbing potential, 122
tesseral, 61, 62 exterior space, 193
wavelength, 62 geoid height, 193
zero points, 62 geoid rise, 334
zonal, 61, 62 in collocation, 275
spherical harmonics in plane co-ordinates, 233
of Laplace, 54 inner zone, 227
spherical shell integration geometry, 193
attraction, 8 spectral form, 191

í ¤. û
496 ABCDEFGHIJKLMNOPQRSTUVWZ Index

Stokes kernel, 122, 192 terrain correction (T C), 162

as a function of deltas, 238 definition, 134
closed expression, 194 algebraic sign, 135
modified, 221 bias, 137
figure, 222 convolution, 251
on the Earth’s surface, 194 equation, 137
smoothness, 224 evaluation point, 252
spectral form, 193 example, 138
Taylor-series expansion, 240 FFT, 249, 250
two-dimensional, 198 in spherical geometry, 141
figure, 200 in the exterior space, 252
Stokes, George Gabriel, 122 prism method, 135, 137, 250
Strang van Hees, Govert, 238 values, 138
Sturm, Jacques, 433 terrain effect in airborne gravimetry, 322
Sturm-Liouville problem, 434 terrain point, potential at, 164
submarine-launched missile, 351 theoretical tide
subsidence, land, 342 periods, 390
surface thermostat (gravimeter), 308
normal derivative of the potential, 19 tidal decomposition of Laplace, 388
normal direction, 19 tidal field, 324
normal vector, 418 deformation of the Earth, 391
orientation, 420 of Sun and Moon, 395
surface element, oriented, 418 tidal potential, 387
surface spherical harmonic, 53, 458 degree number, 393
presentation, 61 indirect effect, 392, 395
as a map, 63 tidal reduction, permanent deformation,
fully normalised, 435 394
notation, 54 tide
plotting, 65, 66 diurnal, 389, 390
surface spherical harmonics as a basis, fortnightly, 389
434 permanent part, 389, 394
sverdrup (unit), 339 concepts, 396
symmetric matrix deformation, 394
eigenvectors, 430 effect on height difference, 397
is a self-adjoint operator, 430 effect on the geoid, 396
Synthetic Aperture Radar Altimeter reduction, 395
(SRAL), 353 value, 395
zero points, 399
T semidiurnal, 389, 390
tangent-plane co-ordinates, 231 theoretical, 386
tapering function (Fourier), 245 figure, 391
tapering, of data (Fourier), 245 tide gauge
tea break, 316 land uplift, 333
telluroid observable, 343
definition, 122 tide-free Earth’s crust, 397
figure, 174 tide-free geoid, 395, 397
telluroid mapping, 122, 123, 174

í ¤. û
Index ABCDEFGHIJKLMNOPQRSTUVWZ
497
Tikhonov regularisation, 363 normal component, 417
Tikhonov, Andrey Nikolayevich, 363 vector space
tilt meter, long water-tube, 393 abstract, 423
time slowing-down ratio (relativity), 186 basis, 423
Toeplitz circulant matrix, 290 bilinear form, 423
Toeplitz, Otto, 290 orthogonal basis, 424
Tom (GRACE satellite), 378 orthonormal basis, 424
tomography, seismic, 153 scalar product, 5, 423
TOPEX/Poseidon (satellite) vector, informal definition, 405
description, 352 vectorial product, 407
mean sea level, 331 figure, 408
orbit choice, 369 properties, 407
results, 354 Vening Meinesz integral equations, 201
topography Vening Meinesz, Felix Andries, 153, 201
exterior potential, 444 submarine measurement, 300
exterior Taylor-series expansion, 444 Verbaandert–Melchior pendulum, 394
interior potential, 445 vertical gravity gradient, 102
Taylor-series expansion, 446 anomalous
potential, 15 kernel, 207, 208
spherical-harmonic expansion, 64 reduction to sea level, 211
topography shift to inside geoid, 212 free-air, 323
Torge, Wolfgang, 404 inside-rock, 179
torsion balance viscous relaxation mode, 347
Cavendish, 4 Von Sterneck device, 300
Eötvös, 323 vortex phenomenon, 90
Toulouse (France), BGI office, 401 vorticity
trace, of a matrix, 432 in a vector field, 414
trench, ocean, 300 of flow, 422
Trieste (Italy) datum, 165
true anomaly, 366 W
Tscherning, Carl Christian, 247 Walferdange (Luxembourg), 315
Tscherning-Rapp formula, 287 water flowing upwards, 169
Tuvalu (Pacific Ocean), 342 water vapour radiometer, 356
twiddle factor (FFT), 441 water, phases, 340
Tytyri limestone mine (Lohja, Finland), wave equation
394 of matter, 42
relativistic, for the electron, 27
U weighing the Earth, 366
unit sphere, 435 weighing visitors, 316
upper culmination, of the Moon, 387 Wenzel, Hans-Georg, 398
wind pile-up, variation, 329
V wire pendulum, very long, 301
variance of prediction, 271, 276 work as a scalar product, 406
definition, 265 work integral, 406, 417
minimisation, 267 is path independent, 420
vector field parametrised, 417
definition, 410

í ¤. û
498 ABCDEFGHIJKLMNOPQRSTUVWZ Index

Working Group for Geoid and Height

Systems (NKG), 403
world aether, 1
World Geodetic System 1984 (WGS84),
101

Z
zenith tube, 394
zero geoid, 395
zero potential, convention
at infinity, 18
at mean sea surface, 18
zero-length spring, 304, 306
how to build, 304
invention, 305

í ¤. û