Physical Geodesy

Physical Geodesy
Mass excess
Mass
deficit
−N
Martin Vermeer1
September 27, 2018
1
[email protected]
Preface
This book aims to present an overview over the current state of the study
of the Earth’s gravity field and those parts of geophysics closely related
to it, such as especially geodynamics, the study of the changing Earth.
It grew out of two decades of teaching in Helsinki’s two universities:
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. Also the author’s
own research, on gravimetric geoid determination, helped shape the
presentation. While there exist excellent textbooks on all parts of what is
presented here, he may still hope that this text will find a niche to fill.
Helsinki, September 27, 2018,
Martin Vermeer
Acknowledgements
Thanks are due to the many students and colleagues, both in academia
and at the Finnish Geodetic Institute, giving useful comments and cor-
rections over the course of many years of lecturing this course 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 prompted a basic revision also to 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 the Aalto University’s pedagogical training programme.
– iii –
Contents
Chapters
» 1. Fundamentals of the theory of gravitation . . . . . . . . . . . . 1

» 2. The Laplace equation and its solutions . . . . . . . . . . . . . . 31
» 3. The normal gravity field . . . . . . . . . . . . . . . . . . . . . . . 63
» 4. Anomalous quantities of the gravity field . . . . . . . . . . . . . 81
» 5. Geophysical reductions . . . . . . . . . . . . . . . . . . . . . . . . 95
» 6. Vertical reference systems . . . . . . . . . . . . . . . . . . . . . . 119
» 7. The Stokes equation and other integral equations . . . . . . . 139
» 8. Spectral techniques, FFT . . . . . . . . . . . . . . . . . . . . . . 163
» 9. Statistical methods . . . . . . . . . . . . . . . . . . . . . . . . . . 181
» 10. Gravimetric measurement devices . . . . . . . . . . . . . . . . . 213
» 11. The geoid, mean sea level, sea-surface topography . . . . . . . 235
» 12. Satellite altimetry and satellite gravity missions . . . . . . . . 251
» 13. Tides, atmosphere and Earth crustal movements . . . . . . . . 277
» 14. Earth gravity field research . . . . . . . . . . . . . . . . . . . . . 287
» A. Field theory and vector calculus — core knowledge . . . . . . . 291
» B. Function spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
» C. Why does FFT work? . . . . . . . . . . . . . . . . . . . . . . . . . 317
» D. Helmert condensation . . . . . . . . . . . . . . . . . . . . . . . . 319
» E. The Laplace equation in spherical co-ordinates . . . . . . . . . 327
Preface iii
List of Tables xii
–v–
vi Contents
List of Figures xiii
Abbreviations xix
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 . . . . . . . . . . . . . . . . . 4
1.4 Potential of a spherical shell . . . . . . . . . . . . . . . . . 5
1.5 Computing the attraction from the potential . . . . . . . . 7
1.6 Potential of a solid body . . . . . . . . . . . . . . . . . . . . 9
1.7 Example: The potential of a line of mass . . . . . . . . . . 11
1.8 Equations of Laplace and Poisson . . . . . . . . . . . . . . 12
1.9 Gauge invariance . . . . . . . . . . . . . . . . . . . . . . . . 13
1.10 Single mass density layer . . . . . . . . . . . . . . . . . . . 14
1.11 Double mass density layer . . . . . . . . . . . . . . . . . . . 15
1.12 The Gauss integral theorem . . . . . . . . . . . . . . . . . . 17
1.13 Green’s theorems . . . . . . . . . . . . . . . . . . . . . . . . 21
1.14 The Chasles theorem . . . . . . . . . . . . . . . . . . . . . . 24
1.15 Boundary-value problems . . . . . . . . . . . . . . . . . . . 26
1.16 What the boundary-value problem cannot compute . . . . 27
Self-test questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Exercise 1 – 1: Core of the Earth . . . . . . . . . . . . . . . . . . . 28
Exercise 1 – 2: Atmosphere . . . . . . . . . . . . . . . . . . . . . . . 29
Exercise 1 – 3: The Gauss theorem . . . . . . . . . . . . . . . . . . 29
2. The Laplace equation and its solutions 31

2.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 The Laplace equation in rectangular co-ordinates . . . . . 32
2.3 Example: The Laplace equation in polar co-ordinates . . 36
2.4 Spherical, geodetic, ellipsoidal co-ordinates . . . . . . . . 38
2.5 The Laplace equation in spherical co-ordinates . . . . . . 40
2.6 Dependence on height . . . . . . . . . . . . . . . . . . . . . 41
2.7 Legendre’s functions . . . . . . . . . . . . . . . . . . . . . . 41
Óîá .
Contents vii
2.8 Symmetry properties of spherical harmonics . . . . . . . . 46

2.9 Orthogonality, orthonormality of Legendre functions . . . 47
2.10 Splitting a function into degree constituents . . . . . . . . 49
2.11 Spectral representations of various quantities . . . . . . . 51
2.12 Low-degree spherical harmonics . . . . . . . . . . . . . . . 53
2.13 Often used spherical-harmonic expansions . . . . . . . . . 54
2.14 Ellipsoidal harmonics . . . . . . . . . . . . . . . . . . . . . 56
Exercise 2 – 1: Attenuation with height of a spherical-harmonic
expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Exercise 2 – 2: Symmetries of spherical harmonics . . . . . . . . 61
Exercise 2 – 3: Escape velocity . . . . . . . . . . . . . . . . . . . . . 62
3. The normal gravity field 63

3.1 The basic idea of a normal field . . . . . . . . . . . . . . . . 63
3.2 The centrifugal force and its potential . . . . . . . . . . . . 64
3.3 Level surfaces and plumb lines . . . . . . . . . . . . . . . . 66
3.4 Natural co-ordinates . . . . . . . . . . . . . . . . . . . . . . 68
3.5 The normal potential in ellipsoidal co-ordinates . . . . . . 70
3.6 Normal gravity on the reference ellipsoid . . . . . . . . . . 71
3.7 Numerical values and formulas . . . . . . . . . . . . . . . . 72
3.8 The normal potential as a spherical harmonic expansion 74
3.9 The disturbing potential . . . . . . . . . . . . . . . . . . . . 76
Exercise 3 – 1: The Somigliana–Pizetti equation . . . . . . . . . . 78
Exercise 3 – 2: The centrifugal force . . . . . . . . . . . . . . . . . 78
4. Anomalous quantities of the gravity field 81

4.1 Disturbing potential, geoid height. . . . . . . . . . . . . . . 81
4.2 Gravity disturbances . . . . . . . . . . . . . . . . . . . . . . 84
4.3 Gravity anomalies . . . . . . . . . . . . . . . . . . . . . . . . 85
4.4 Units used for gravity anomalies . . . . . . . . . . . . . . . 87
4.5 The boundary-value problem of physical geodesy . . . . . 88
4.6 The telluroid mapping and the “quasi-geoid” . . . . . . . . 90
Óîá .
viii Contents
4.7 Free-air anomalies . . . . . . . . . . . . . . . . . . . . . . . 91

Exercise 4 – 1: The spectrum of gravity anomalies . . . . . . . . . 93
Exercise 4 – 2: Deflections of the plumb line and geoid tilt . . . . 93
Exercise 4 – 3: Gravity anomaly, geoid height . . . . . . . . . . . . 94
5. Geophysical reductions 95
5.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.2 Bouguer anomalies . . . . . . . . . . . . . . . . . . . . . . . 96
5.3 Terrain effect and terrain correction . . . . . . . . . . . . . 99
5.4 Spherical Bouguer anomalies . . . . . . . . . . . . . . . . . 103
5.5 Helmert condensation . . . . . . . . . . . . . . . . . . . . . 104
5.6 Isostasy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.7 Isostatic reductions . . . . . . . . . . . . . . . . . . . . . . . 112
5.8 The “isostatic geoid” . . . . . . . . . . . . . . . . . . . . . . 113
Exercise 5 – 1: Gravity anomaly . . . . . . . . . . . . . . . . . . . . 117
Exercise 5 – 2: Bouguer reduction . . . . . . . . . . . . . . . . . . . 117
Exercise 5 – 3: Terrain correction, Bouguer . . . . . . . . . . . . . 118
Exercise 5 – 4: Isostasy . . . . . . . . . . . . . . . . . . . . . . . . . 118
6. Vertical reference systems 119

6.1 Levelling, orthometric heights and the geoid . . . . . . . . 119
6.2 Orthometric heights . . . . . . . . . . . . . . . . . . . . . . 121
6.3 Normal heights . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.4 Difference between geoid height and height anomaly . . . 128
6.5 Difference between orthometric and normal heights . . . 130
6.6 Calculating orthometric heights precisely . . . . . . . . . 130
6.7 Calculating normal heights precisely . . . . . . . . . . . . 132
6.8 Calculation example for heights . . . . . . . . . . . . . . . 133
6.9 Orthometric and normal corrections . . . . . . . . . . . . . 134
6.10 A vision for the future: relativistic levelling . . . . . . . . 136
Exercise 6 – 1: Calculating orthometric heights . . . . . . . . . . 138
Exercise 6 – 2: Calculating normal heights . . . . . . . . . . . . . 138
Exercise 6 – 3: Difference between orthometric and normal height 138
Óîá .
Contents ix
7. The Stokes equation and other integral equations 139

7.1 The Stokes equation and the Stokes integral kernel . . . 139
7.2 Plumb-line deflections and Vening Meinesz equations . . 142
7.3 The Poisson integral equation . . . . . . . . . . . . . . . . . 142
7.4 Gravity anomalies in the exterior space . . . . . . . . . . . 145
7.5 The vertical gradient of the gravity anomaly . . . . . . . . 146
7.6 Gravity reductions in geoid determination . . . . . . . . . 149
7.7 The Remove-Restore method . . . . . . . . . . . . . . . . . 153
7.8 Kernel modification in the Remove-Restore method . . . 154
7.9 Advanced kernel modifications . . . . . . . . . . . . . . . . 156
7.10 The effect of the local zone . . . . . . . . . . . . . . . . . . . 158
Exercise 7 – 1: The Stokes equation . . . . . . . . . . . . . . . . . . 161
8. Spectral techniques, FFT 163

8.1 The Stokes theorem as a convolution . . . . . . . . . . . . 163
8.2 Integration by FFT . . . . . . . . . . . . . . . . . . . . . . . 165
8.3 Solution in rectangular co-ordinates . . . . . . . . . . . . . 167
8.4 Bordering and tapering of the data area . . . . . . . . . . 172
8.5 Computing a geoid model with FFT . . . . . . . . . . . . . 174
8.6 Use of FFT in other contexts . . . . . . . . . . . . . . . . . 176
8.7 Computing terrain corrections with FFT . . . . . . . . . . 176
9. Statistical methods 181

9.1 The role of uncertainty in geophysics . . . . . . . . . . . . 181
9.2 Linear functionals . . . . . . . . . . . . . . . . . . . . . . . . 181
9.3 Statistics on the Earth’s surface . . . . . . . . . . . . . . . 183
9.4 The covariance function of the gravity field . . . . . . . . . 185
9.5 Least-squares collocation . . . . . . . . . . . . . . . . . . . 186
9.6 Prediction of gravity anomalies . . . . . . . . . . . . . . . . 196
9.7 Covariance function and degree variances . . . . . . . . . 197
9.8 Propagation of covariances . . . . . . . . . . . . . . . . . . 199
9.9 Global covariance functions . . . . . . . . . . . . . . . . . . 203
Óîá .
x Contents
9.10 Collocation and the spectral viewpoint . . . . . . . . . . . 204

Exercise 9 – 1: Variance of prediction . . . . . . . . . . . . . . . . . 208
Exercise 9 – 2: Hirvonen’s covariance formula and prediction . . 208
Exercise 9 – 3: Predicting gravity anomalies . . . . . . . . . . . . 209
Exercise 9 – 4: Predicting gravity anomalies (2) . . . . . . . . . . 210
Exercise 9 – 5: Propagation of covariances . . . . . . . . . . . . . . 210
Exercise 9 – 6: Kaula’s rule for gravity gradients . . . . . . . . . . 210
Exercise 9 – 7: Underground mass points . . . . . . . . . . . . . . 211
10. Gravimetric measurement devices 213

10.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
10.2 The relative or spring gravimeter . . . . . . . . . . . . . . 214
10.3 The absolute or ballistic gravimeter . . . . . . . . . . . . . 220
10.4 Network hierarchy in gravimetry . . . . . . . . . . . . . . 224
10.5 The superconducting gravimeter . . . . . . . . . . . . . . . 224
10.6 Atmospheric influence on gravity measurement . . . . . . 226
10.7 Airborne gravimetry and GNSS . . . . . . . . . . . . . . . 227
10.8 Measuring the gravity gradient . . . . . . . . . . . . . . . . 229
Exercise 10 – 1: Absolute gravimeter . . . . . . . . . . . . . . . . . 231
Exercise 10 – 2: Spring gravimeter . . . . . . . . . . . . . . . . . . 232
Exercise 10 – 3: Air pressure and gravity . . . . . . . . . . . . . . 232
11. The geoid, mean sea level, sea-surface topography 235

11.1 Basic concepts . . . . . . . . . . . . . . . . . . . . . . . . . . 235
11.2 Geoids and national height datums . . . . . . . . . . . . . 236
11.3 The geoid and post-glacial land uplift . . . . . . . . . . . . 238
11.4 Methods for determining the sea-surface topography . . . 239
11.5 Global sea-surface topography and heat transport . . . . 241
11.6 The global behaviour of sea level . . . . . . . . . . . . . . . 244
11.7 The sea-level equation . . . . . . . . . . . . . . . . . . . . . 245
Exercise 11 – 1: Coriolis force, ocean current . . . . . . . . . . . . 248
Exercise 11 – 2: Land subsidence and the mechanism of land uplift 248
Óîá .
Contents xi
12. Satellite altimetry and satellite gravity missions 251

12.1 Satellite altimetry . . . . . . . . . . . . . . . . . . . . . . . . 251
12.2 Crossover adjustment . . . . . . . . . . . . . . . . . . . . . 255
12.3 Choice of satellite orbit . . . . . . . . . . . . . . . . . . . . . 261
12.4 In-flight calibration . . . . . . . . . . . . . . . . . . . . . . . 266
12.5 Retracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
12.6 Oceanographic research using satellite altimetry . . . . . 268
12.7 Satellite gravity missions . . . . . . . . . . . . . . . . . . . 269
Exercise 12 – 1: Altimetry, crossover adjustment . . . . . . . . . . 273
Exercise 12 – 2: Satellite orbit . . . . . . . . . . . . . . . . . . . . . 274
Exercise 12 – 3: Kepler’s third law . . . . . . . . . . . . . . . . . . 274
13. Tides, atmosphere and Earth crustal movements 277

13.1 Theoretical tide . . . . . . . . . . . . . . . . . . . . . . . . . 277
13.2 Deformation caused by the tidal force . . . . . . . . . . . . 280
13.3 The permanent part of the tide . . . . . . . . . . . . . . . . 283
13.4 Tidal corrections between height systems . . . . . . . . . 283
13.5 Loading of the Earth’s crust by sea and atmosphere . . . 285
Exercise 13 – 1: Tide . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
14. Earth gravity field research 287

14.1 Internationally . . . . . . . . . . . . . . . . . . . . . . . . . . 287
14.2 Europe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
14.3 The Nordic countries . . . . . . . . . . . . . . . . . . . . . . 288
14.4 Finland . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
14.5 Textbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
A. Field theory and vector calculus — core knowledge 291

A.1 Vector calculus . . . . . . . . . . . . . . . . . . . . . . . . . . 291
A.2 Scalar and vector fields . . . . . . . . . . . . . . . . . . . . . 294
A.3 Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
A.4 The continuity of matter . . . . . . . . . . . . . . . . . . . . 302
Óîá .
xii List of Tables
B. Function spaces 305

B.1 An abstract vector space . . . . . . . . . . . . . . . . . . . . 305
B.2 Fourier function space . . . . . . . . . . . . . . . . . . . . . 306
B.3 Sturm-Liouville differential equations . . . . . . . . . . . 308
B.4 Legendre polynomials . . . . . . . . . . . . . . . . . . . . . 313
B.5 Spherical harmonics . . . . . . . . . . . . . . . . . . . . . . 313
Exercise B – 1: Orthonormality of the Fourier basis functions . . 315
C. Why does FFT work? 317
D. Helmert condensation 319

D.1 The interior potential of the topography . . . . . . . . . . 319
D.2 The exterior potential of the topography . . . . . . . . . . 320
D.3 The exterior potential of the condensation layer . . . . . . 321
D.4 The total potential of Helmert condensation . . . . . . . . 321
D.5 The dipole method . . . . . . . . . . . . . . . . . . . . . . . 324
E. The Laplace equation in spherical co-ordinates 327

E.1 Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
E.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
Bibliography 331
Index 339
List of Tables
2.1 Vertically shifting the potential field . . . . . . . . . . . . . . 36

2.2 Legendre polynomials . . . . . . . . . . . . . . . . . . . . . . . 42
2.3 Associated Legendre functions . . . . . . . . . . . . . . . . . . 43
2.4 Semi-wavelengths for different degrees and orders . . . . . 46
2.5 Plotting a surface spherical-harmonic map . . . . . . . . . . 48
2.6 EGM96 coefficients and mean errors . . . . . . . . . . . . . . 56
Óîá .
List of Figures xiii
2.7 Legendre functions of the second kind . . . . . . . . . . . . . 59
3.1 GRS80 normal potential spherical-harmonic coefficients . . 75
4.1 Orders of magnitude of gravity variations . . . . . . . . . . . 88
12.1 Altimetric satellites through the ages . . . . . . . . . . . . . 251

12.2 Calculating the height of a satellite from her period. . . . . 266
13.1 The various periods in the theoretical tide . . . . . . . . . . 280
List of Figures
1.1 Gravitation is universal . . . . . . . . . . . . . . . . . . . . . . 2

1.2 A thin spherical shell consists of rings . . . . . . . . . . . . . 5
1.3 Dependence of potential and attraction on distance. . . . . . 7
1.4 A double mass density layer . . . . . . . . . . . . . . . . . . . 16
1.5 A graphical explanation of the Gauss integral theorem . . . 18
1.6 A little rectangular box . . . . . . . . . . . . . . . . . . . . . . 19
1.7 Eight-unit cube . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.8 Green’s third theorem for an exterior point . . . . . . . . . . 23
1.9 Green’s third theorem for an interior point . . . . . . . . . . 24
1.10 Green’s third theorem for the space external to a body . . . 25
1.11 Iron-ore body . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.1 Attenuation of the gravitational field with height . . . . . . 35

2.2 Definition of spherical co-ordinates . . . . . . . . . . . . . . . 38
2.3 Definition of geodetic co-ordinates . . . . . . . . . . . . . . . 39
2.4 Some Legendre polynomials . . . . . . . . . . . . . . . . . . . 43
2.5 Associated Legendre functions . . . . . . . . . . . . . . . . . . 44
2.6 The algebraic signs of spherical harmonics . . . . . . . . . . 45
2.7 Surface spherical harmonics as maps . . . . . . . . . . . . . 45
2.8 Monopole, dipole, and quadrupole . . . . . . . . . . . . . . . . 54
3.1 The normal gravity field of the Earth . . . . . . . . . . . . . . 63

3.2 Gravitation and centrifugal force . . . . . . . . . . . . . . . . 65
Óîá .
xiv List of Figures
3.3 The curvature of level surfaces . . . . . . . . . . . . . . . . . 67

3.4 Natural co-ordinates . . . . . . . . . . . . . . . . . . . . . . . . 69
3.5 Meridian ellipse and latitude types . . . . . . . . . . . . . . . 72
3.6 The normal field’s potential curve over the equator . . . . . 75
4.1 Geoid undulations and deflections of the plumb line . . . . . 82

4.2 A Finnish geoid model . . . . . . . . . . . . . . . . . . . . . . . 83
4.3 Equipotential surfaces of gravity and normal gravity fields 84
4.4 Various reference surfaces . . . . . . . . . . . . . . . . . . . . 86
4.5 Free-air gravity anomalies for Southern Finland. . . . . . . . 93
5.1 The attraction of a Bouguer plate . . . . . . . . . . . . . . . . 97

5.2 The Bouguer plate as an approximation to the topography . 98
5.3 Behaviour of different anomaly types in the mountains . . . 99
5.4 Terrain corrected Bouguer anomalies for Southern Finland 100
5.5 Calculating the terrain correction by the prism method . . 101
5.6 The steps in computing the Bouguer anomaly . . . . . . . . 102
5.7 A special terrain shape . . . . . . . . . . . . . . . . . . . . . . 102
5.8 Helmert condensation and the gravity field . . . . . . . . . . 105
5.9 Friedrich Robert Helmert . . . . . . . . . . . . . . . . . . . . . 105
5.10 Isostasy and the bending of plumb lines . . . . . . . . . . . . 107
5.11 Pratt–Hayford isostatic hypothesis . . . . . . . . . . . . . . . 107
5.12 Airy–Heiskanen isostatic hypothesis . . . . . . . . . . . . . . 108
5.13 Quantities in isostatic compensation . . . . . . . . . . . . . . 109
5.14 The modern understanding of isostasy . . . . . . . . . . . . . 111
5.15 Isostatic gravity anomalies for Southern Finland. . . . . . . . 113
5.16 Isostatic reduction as a pair of surface density layers . . . . 115
6.1 The principle of levelling . . . . . . . . . . . . . . . . . . . . . 119

6.2 Height reference pillar in Kaivopuisto, Helsinki . . . . . . . 121
6.3 Levelled heights and geopotential numbers. . . . . . . . . . 122
6.4 Lake Päijänne: water flows “uphill” . . . . . . . . . . . . . . 123
6.5 Mikhail Sergeevich Molodensky . . . . . . . . . . . . . . . . . 124
6.6 A graphic cartoon of Molodensky’s proof . . . . . . . . . . . . 127
Óîá .
List of Figures xv
6.7 Geoid, quasi-geoid, telluroid and topography . . . . . . . . . 127

6.8 An optical lattice clock . . . . . . . . . . . . . . . . . . . . . . 137
7.1 The principle of gravimetric geoid determination . . . . . . 140

7.2 Integrating the Stokes equation geometrically . . . . . . . . 140
7.3 The Stokes kernel function . . . . . . . . . . . . . . . . . . . . 141
7.4 The geometry of the generating function of the Legendre
polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
7.5 The Poisson kernel for gravity anomalies . . . . . . . . . . . 147
7.6 Residual terrain modelling (RTM) . . . . . . . . . . . . . . . 151
7.7 Modified Stokes kernel functions . . . . . . . . . . . . . . . . 156
7.8 Simpson integration nodal weights in two dimensions . . . 159
8.1 Map projection co-ordinates in the local tangent plane . . . 164

8.2 “Tapering” 25% . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
8.3 Example images for FFT transform . . . . . . . . . . . . . . 174
8.4 The Finnish FIN2000 geoid . . . . . . . . . . . . . . . . . . . 175
9.1 Geocentric angular distance and azimuth . . . . . . . . . . . 185

9.2 Hirvonen’s covariance function in two dimensions . . . . . . 191
9.3 An example of least-squares collocation . . . . . . . . . . . . 191
9.4 Collocation example . . . . . . . . . . . . . . . . . . . . . . . . 193
9.5 Global covariance functions as degree variances . . . . . . . 204
9.6 Circular geometry . . . . . . . . . . . . . . . . . . . . . . . . . 206
10.1 Jean Richer’s report . . . . . . . . . . . . . . . . . . . . . . . . 214

10.2 Autograv CG5 spring gravimeter . . . . . . . . . . . . . . . . 215
10.3 Operating principle of spring gravimeter . . . . . . . . . . . 217
10.4 The idea of astatization . . . . . . . . . . . . . . . . . . . . . . 219
10.5 Operating principle of a ballistic absolute gravimeter . . . . 220
10.6 FG5 absolute gravimeter . . . . . . . . . . . . . . . . . . . . . 221
10.7 Principle of operation of an atomic gravimeter . . . . . . . . 223
10.8 International intercomparison of absolute gravimeters . . . 224
10.9 Principle of operation of a superconducting gravimeter . . . 225
11.1 The mechanism of post-glacial land uplift . . . . . . . . . . . 240
Óîá .
xvi List of Figures
11.2 The Fennoscandian gravity line on the 63rd parallel . . . . 240

11.3 Connection between sea-surface topography and ocean cur-
rents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
11.4 Sea-surface topography map produced by GOCE . . . . . . . 243
11.5 The sea-level equation . . . . . . . . . . . . . . . . . . . . . . 245
11.6 Sea-level rise after the last ice age . . . . . . . . . . . . . . . 247
12.1 Results from the TOPEX/Poseidon and Jason satellites . . . 253

12.2 Satellite altimetry as a measurement method . . . . . . . . 254
12.3 A simple crossover geometry . . . . . . . . . . . . . . . . . . . 256
12.4 Track geometry of satellite altimetry . . . . . . . . . . . . . . 260
12.5 Kepler’s orbital elements . . . . . . . . . . . . . . . . . . . . . 262
12.6 The mechanism of a Sun-stationary orbit . . . . . . . . . . . 263
12.7 Geometry of a “no-shadow” orbit . . . . . . . . . . . . . . . . 264
12.8 A satellite orbit example . . . . . . . . . . . . . . . . . . . . . 265
12.9 Analyzing the altimeter return pulse . . . . . . . . . . . . . . 267
12.10 Ice volume on the Arctic Ocean . . . . . . . . . . . . . . . . . 268
12.11 Gravity field determination from GPS orbital tracking . . . 270
12.12 The principle of the GRACE satellites . . . . . . . . . . . . . 271
12.13 Gravity field determination with GOCE . . . . . . . . . . . . 272
12.14 Satellite altimetric track geometry . . . . . . . . . . . . . . . 274
13.1 Theoretical tide . . . . . . . . . . . . . . . . . . . . . . . . . . . 277

13.2 The main components of the theoretical tide . . . . . . . . . 281
13.3 Constituents of the permanent tide . . . . . . . . . . . . . . . 284
A.1 Exterior or vectorial product . . . . . . . . . . . . . . . . . . . 293

A.2 Kepler’s second law . . . . . . . . . . . . . . . . . . . . . . . . 294
A.3 The gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
A.4 The divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
A.5 The curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
A.6 The Stokes curl theorem . . . . . . . . . . . . . . . . . . . . . 300
A.7 The Gauss integral theorem . . . . . . . . . . . . . . . . . . . 302
B.1 Fourier analysis on a step function . . . . . . . . . . . . . . . 307
E.1 Gauss integral theorem applied to a co-ordinate conformal

volume element . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
Óîá .
Abbreviations
AGU American Geophysical Union

BGI Bureau Gravimétrique International,
International Gravity Bureau
BVP boundary-value problem
CHAMP Challenging Minisatellite Payload for Geophysical Research
and Applications
DMA Defense Mapping Agency (U.S.A.)
DTM digital terrain model
EGM96 Earth Gravity Model 1996
EGM2008 Earth Gravity Model 2008
EGU European Geosciences Union
ENSO El Niño Southern Oscillation
ESA European Space Agency
FAS Member of the French Academy of Sciences
FIN2000 geoid model (Finland)
FIN2005N00
geoid model (Finland)
FFT fast Fourier transform
FGI Finnish Geospatial Research Institute, formerly Finnish
Geodetic Institute
FRS Fellow of the Royal Society (of London)
FRSE Fellow of the Royal Society of Edinburgh
GDR Geophysical Data Record
GFZ Geoforschungszentrum (Potsdam, Germany), German Re-
search Centre for Geosciences
GIA Glacial Isostatic Adjustment
GNSS Global Navigation Satellite Systems, comprise, besides the
American GPS, also the satellite positioning systems of other
countries, like the Russian GLONASS and the European
Galileo
GOCE Geopotential and Steady-state Ocean Circulation Explorer
GPS Global Positioning System
GRACE Gravity Recovery And Climate Experiment
– xvii –
xviii Abbreviations
GRAVSOFT
Geopotential determination software, mainly developed in
Denmark
GRS67 Geodetic Reference System 1967
GRS80 Geodetic Reference System 1980
HUT Helsinki University of Technology, today a part of Aalto Uni-
versity
IAG International Association of Geodesy
ICET International Center for Earth Tides
ICGEM International Center for Global Earth Models
IDEMS International Digital Elevation Model Service
IGeC International Geoid Commission (obsolete)
IGeS International Geoid Service (obsolete)
IGFS International Gravity Field Service
ISG International Service for the Geoid
IUGG International Union of Geodesy and Geophysics
JILA Joint Institute for Laboratory Astrophysics (Boulder, Col-
orado, USA)
KKJ National Grid Co-ordinate System (Finland)
Lego™ “Leg Godt”, en. “Play Well”, Danish toy brand
LLR Lunar laser ranging
LSC least-squares collocation
Mf Moon, fortnightly tide
moho Mohorovičić discontinuity
N60 height system (Finland)
N2000 height system (Finland)
NAP Normaal Amsterdams Peil, height system (Western Europe)
NASA National Aeronautics and Space Administration (USA)
NAVD88 North American Vertical Datum 1988
NC normal correction
NGA National Geospatial-Intelligence Agency (USA, formerly
NIMA)
NIMA National Imagery and Mapping Agency (USA, formerly DMA)
NKG Nordiska Kommissionen för Geodesi, Nordic Geodetic Com-
mission
NKG2004 geoid model (Nordic area)
NKG2015 geoid model (Nordic area)
NOAA National Oceanic and Atmospheric Administration (USA)
OSU Ohio State University, Columbus, Ohio, USA
OC orthometric correction
ppm parts per million
ppb parts per billion
RTM residual terrain modelling
SI Système International d’Unités
Óîá.
Abbreviations xix
SLR satellite laser ranging

Ssa Sun, semi-annual tide
SWH significant wave height
TC terrain correction
WGS84 World Geodetic System 1984
Óîá.
D Fundamentals of the theory of gravitation
1
D 1.1 General
In this chapter we discuss the foundations of Newton’s theory of gravi-
tation. 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 propor-
tional 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 describes gravitation as a phenomenon propagating through space,
a field. The propagation is described in the field equations. The field
approach isn’t quite as intuitive, but is a powerful theoretical tool1 .
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. Of
the practical and theoretical applications of these may be mentioned the
Bouguer plate and Helmert condensation. In the following we will discuss
their properties in detail. Mass density layers are also used in deriving
the theorems of Green. We will learn about important integral theorems
like the theorems of Gauss and Green, with the aid of which we may
infer the whole potential field in space from field values given only on a
certain surface. Other similar examples are the Chasles theorem and the
solution to Dirichlet’s problem.
In chapter 2 we apply these fundamentals of potential theory to derive
a spectral representation of the Earth’s gravitational field, a so-called
1
There is also a philosophical side to this. To many, e.g., 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 wasn’t until the advent of relativity theory, that the understanding gained
ground that a physical theory doesn’t have to satisfy our preconceptions about
what constitutes a “sensible explanation” — as long as it describes the physical
phenomena correctly.
–1–
2 Fundamentals of the theory of gravitation
Figure 1.1. Gravitation is universal. A gravitational lense imaged by the Hub-

ble Space Telescope, the cluster of galaxies Abell 1689 at a distance
of 2.2 billion light years. Source: NASA; ESA; L. Bradley (Johns
Hopkins University); R. Bouwens (University of California, Santa
Cruz); H. Ford (Johns Hopkins University); and G. Illingworth (Uni-
versity of California, Santa Cruz). Creative Commons Attribution
D 4.0 International license.
spherical-harmonic expansion.
Here, in the beginning of the text, we derive an extensive set of math-
ematical equations, such as well-known integral equations. This is an
unfortunately necessary preliminary work. These equations, however,
are no end in themselves and they are not worth committing to memory.
Try rather to understand their logic, and how historically these vari-
ous results have been arrived at. Try as well to acquire an intuition, a
fingertip feeling, about the nature of this theory.
Ó »îá .
Gravitation between two masses 3
D 1.2 Gravitation between two masses

We start the investigation of the Earth’s gravity field suitably with Isaac
Newton2 ’s general law of gravitation:
m1 m2
F =G . (1.1)
ℓ2
F is the attractional force between bodies 1 and 2, m 1 and m 2 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 = 6.6726 · 10−11 m3 kg−1 s−2 .
The value of G was determined for the first time by Henry Cavendish3
kiertoheiluri using a sensitive torsion balance (Cavendish, 1798).
Let us call the small body, the test mass, e.g., a satellite, m, and the
large mass, the planet or the Sun, M . Then, m 1 = M may be called the
attracting mass, and m 2 = m the attracted mass, and we obtain
mM
F =G .
ℓ2
According to Newton’s law of motion
F = ma,
where a is the acceleration of body m. From this follows
M
a=G .
ℓ2
From this equation, the quantity m = m 2 has vanished. This is the famous
observation by Galileo, that all bodies fall equally fast4 , irrespective of
their mass. This is also known as Einstein5 ’s principle of equivalence.
2
Sir Isaac Newton PRS (1642 – 1727) was an English universal genius who
derived the mathematical underpinning of astronomy, and much of geophysics, in
his main work “Philosophiæ Naturalis Principia Mathematica”, “Mathematical
Foundations of Physics”.
3
Henry Cavendish FRS (1731 – 1810) was a British natural scientist from a
wealthy nobility background. He did also pioneering work in chemistry. He was
extremely shy, and the renowned neurologist Oliver Sacks retrodiagnosed him
as afflicted with the Asperger syndrome (Sacks, 2001).
4
At least in vacuum. The Apollo astronauts showed impressively, how on the
Moon a feather and a hammer fall equally fast! https://www.youtube.com/watch?
feature=player_embedded&v=KDp1tiUsZw8#!.
5
Albert Einstein (1879 – 1955) was a theoretical physicist of Jewish German
descent, who created both the special and general theories of relativity, applying
the latter to cosmology, and did fundamental work in quantum theory.
Ó »îá .
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
where the three-dimensional vectors of place of both the attracting and
attracted masses are defined as follows in rectangular co-ordinates6 :
r = xi + yj + zk,
R = X i + Y j + Z k,
where the triad of unit vectors {i, j, k} is an orthonormal basis7 in Eu-

clidean space, R3 and
√
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.

Note that the vector equation 1.2 contains a minus sign! This only 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.
D 1.3 The potential of a point mass

The gravitational field is a special field: if it is stationary, i.e., not time
dependent, it is conservative. This means that a body moving inside the
field along a closed path will, at the end of the journey, not have lost or
gained energy. Because of this, one may attach to each point in the field
a “label” onto which is written the amount of energy gained or lost by a
unit or test mass, when travelling from an agreed-upon starting point to
the point under discussion. The value written on the label is called the
potential. (Note, that the choice of starting point is arbitrary! We will
return to this still.)
The potential function defined in this way for a pointlike body M is:
M GM
V =G = , (1.4)
ℓ ℓ
As vector notation, one may use either −

6 →
v (an arrow above) or v (bold). Here
we use the bold notation, except for vectors designated by Greek letters, which
cannot be bolded.
7
This means that ∥i∥ = ∥j∥ = ∥k∥ = 1 and 〈i · j〉 = 〈i · k〉 = 〈j · k〉 = 0, where the norm
def p
is defined as ∥a∥ = 〈a · a〉, and 〈a · b〉 , a, b ∈ E is the inner or scalar product of
the space.
Ó »îá .
Potential of a spherical shell 5
bd θ
p b
P
r
O
θ
ℓ
b
D Figure 1.2. A thin spherical shell consists of rings.
where ℓ is again, like above, the length of the vector r − R , ℓ = ∥r − R∥.

The constant GM has in case of the Earth (according to the GRS80
system, conventionally) the value:
GM⊕ = 3.986,005 · 1014 m3 /s2 .
The currently best available physical value again is:
GM⊕ = 3.986,004,40 · 1014 m3 /s2 .
D 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:
ˆ ˆ
dm (R) dm (R)
V (r) = G =G . (1.5)
M ℓ M ∥r − R∥
This is an integral over mass elements dm, where every such mass
element is located at place R. The potential V is evaluated at location r,
and the distance ℓ = ∥r − R∥.
We now derive the formula 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 · d θ , is 2π b sin θ , its
surface area is
(2π b sin θ ) ( b · d θ ) .
Ó »îá .
Let the thickness of the shell be p (small) and its density ρ . We obtain
for the total mass of the ring
2π pρ b2 sin θ d θ .
Because every point of the ring is at the same distance ℓ from point P,
we may write for the potential at point P :
2πG pρ b2 sin θ d θ
VP = .
ℓ
With the cosine rule,
ℓ2 = r 2 + b2 − 2 rb cos θ , (1.6)
we obtain, using equation 1.5, for the potential of the whole shell
ˆ
2 sin θ d θ
VP = 2πG ρ pb p .
r 2 + b2 − 2 rb cos θ
In order to evaluate this integral, we must replace the integration vari-
able θ by ℓ. Differentiating equation 1.6 yields
ℓ d ℓ = br sin θ d θ ,
p
and remembering that ℓ = r 2 + b2 − 2 rb cos θ we obtain
ˆ ℓ2
2 dℓ
VP = 2πG ρ pb .
ℓ1 br
In the case that point P is outside the shell, the integration bounds of ℓ
are ℓ1 = r − b and ℓ2 = r + b, and we obtain for the potential of point P
]ℓ=r+b
ℓ 4πG ρ pb2
[
2
VP = 2πG ρ pb = .
br ℓ= r − b r
Because the mass of the whole shell is M b = 4π b2 ρ p, it follows that the

potential of the shell is the same as that of an equal sized mass in its
centre O :
GM b
VP = ,
r
where r is now the distance of computation point P from the centre of the
sphere O . We see that this is the same as equation 1.4.
In the same way, the attraction, or rather, acceleration, caused by the
spherical shell is8
(−
→ ) rP − rO rP − rO
aP = ∇V = −4πG ρ pb2 = −GM b ,
P r3 r3
in which r = ∥rP − rO ∥ . This result is identical with the acceleration
caused by an equal sized point mass located in point O , see equation 1.2.
8
Here, the ∇ (nabla) operator is used, to be explained in section 1.5.
Ó »îá .
Computing the attraction from the potential 7
4πG ρ pb
Acceleration
2
4πG ρ p br2
4πG ρ pb br
Potential
0
0 b r
Figure 1.3. Dependence of potential and attraction on distance r from the centre
D of a spherical shell.
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
[
2
VP = 2πG ρ pb = 4πG ρ pb.
br ℓ= b − r
As we see, this is a constant and not dependent upon the location of

point P . Therefore ∇VP = 0 and the attraction, being the gradient of the
potential, vanishes.
The end result is, that the attraction of a spherical shell is, outside the
shell,
GM
a = ∥a∥ = 2 ,
r
where M is the total mass of the shell and r = ∥rP − rO ∥ the distance of
the observation point from the shell’s centre; and 0 inside the shell.
In figure 1.3 we have drawn the curves of potential and attraction (i.e.,
acceleration, attractive force per unit of mass). If a body consists of
many concentric spherical shells (like rather precisely the Earth and
many celestial bodies), then inside the body only those layers of mass
internal to the observation point participate in causing attraction, and
this attraction is the same as it would be if all the mass of 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.
D 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
Ó »îá .
acceleration vector by differentiating V with respect to place, i.e., by

applying the gradient operator:
−
→ −−→ ∂V ∂V ∂V
a = ∇ V = gradV = i +j +k . (1.7)
∂x ∂y ∂z
Here, the symbol ∇ (nabla), is a frequently used partial differential

operator
∂ ∂ ∂
∇ = i +j +k .
∂x ∂y ∂z
Here, {i, j, k} is again a basis of mutually orthogonal unit vectors in
Euclidean space R3 parallel to the ( x, y, z) axes.
Let us try this differentiation in the case of the potential field of the point
mass M . Substitute the above equations for V 1.4 and ℓ 1.39 :
∂V ∂V ∂ℓ 1 x− X x− X
= = GM · − = −GM 3 .
∂x ∂ℓ ∂ x ℓ 2 ℓ ℓ
Similarly we compute the y and z components:
∂V y − Y ∂V z−Z
= −GM ; = −GM 3 .
∂y ℓ 3 ∂z ℓ
These are the components of gravitational acceleration when the source of

the field is one point mass M. So, in this concrete case the vector equation
given above applies:
−−→ −
→
a = gradV = ∇ V .
Remark: in physical geodesy — unlike in physics — the potential

is reckoned always positive if the attracting mass M is positive (as
it is known to always be). However, the potential energy of body m
inside the field of mass M is negative! More precisely, the potential
energy of body m is
E pot = −V m.
In practice one calls the vector of gravitational acceleration the “gravita-
tional vector”.
9
From the equation
√ ] 12
(x − X )2 + (y − Y )2 + (z − Z)2 = (x − X )2 + (y − Y )2 + (z − Z)2
[
ℓ=
it follows with the chain rule, that

]1
∂ (x − X )2 + (y − Y )2 + (z − Z)2
[
∂ℓ 2
∂ (x − X )2
= [ ] · =
∂x ∂ (x − X )2 + (y − Y )2 + (z − Z) 2 ∂x
1[ ]− 1 x− X
= (x − X )2 + (y − Y )2 + (z − Z)2 2 · 2 (x − X ) = .
2 ℓ
Ó »îá .
Potential of a solid body 9
D 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. The Earth
serves as an example of this, as its mass distribution in space may be
described by a matter density function ρ :
dm ( x, y, z)
ρ ( x, y, z) = ,
d V ( x, y, z)
in which dm is a mass element and d V 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 force and acceleration vectors may be summed
linearly, it follows that also the total potential of the body can be obtained
by summing together the potentials of all its parts. E.g., the potential of
a set of n mass points is
n
∑ mi
V =G
ℓi
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. (Note that unfortunately almost the
same symbols V and V are used here for the potential and for volume,
respectively): ˚ ˚
dm ρ
V =G =G dV . (1.8)
body ℓ body ℓ
The symbol ρ inside the integral

√ designates the matter density at the
location of dm. ℓ = ∥r − R∥ = ( x − X )2 + ( y − Y )2 + ( z − Z )2 is the dis-
tance between point of measurement and attracting mass element. More
clearly:
˚
ρ (X , Y , Z)
V ( x, y, z) = G √ d X dY dZ .
body ( x − X ) + ( y − Y )2 + ( z − Z )2
2
As we showed already above for mass points, also the first derivative with
respect to place or gradient of the potential V of a solid body,
−−→ −
→
gradV = ∇ V = a, (1.9)
is the acceleration vector caused by the attraction of the body. This

applies generally.
Ó »îá .
D 1.6.1 Behaviour at infinity
If a body is of finite extent (i.e., it lies completely within a sphere of size

ϵ around the origin) and also its density is bounded everywhere, it follows
that
∥r∥ → ∞ =⇒ V (r) → 0,
because, according to the triangle inequality,
ℓ = ∥r − R∥ ≥ ∥r∥ − ∥R∥ ≥ ∥r∥ − ϵ
and thus
1
→ 0 when ∥r∥ → ∞.
ℓ
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 + /∥r∥ ℓ ∥r∥ 1 − ϵ/∥r∥
ϵ
It is seen, again with the notation r = ∥r∥, that
1 1
r → ∞ =⇒ → .
ℓ r
When we substitute this into the above integral 1.8, it follows that for
large distances r → ∞:
˚ ˚
ρ G GM
V =G dV ≈ ρ dV = ,
body ℓ r body r
where M, the integral of density over the volume of the body, is precisely
its total mass. From this we see, that at great distance, the field of a
finite sized body M is nearly identical with the field generated by a point
mass the total mass of which is equal to the total mass of the body. This
important observation was already made by Newton. As a result of this
phenomenon we can treat, in the celestial mechanics of the Solar system,
the Sun and planets10 as mass points, although we know that they are
not.
10
The only important exception is formed by the forces between a planet and its
moons, both due to the flattening of the planet and due to tidal effects.
Ó »îá .
Example: The potential of a line of mass 11
D 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
ˆ H
1
V ( x, y, z) = G √ dZ, (1.10)
0 ( X − x)2 + (Y − y)2 + ( Z − z)2
where ( X , Y ) is the location of the mass line, ( x, y, z) is the location of the

point in which the potential is being evaluated, and the mass line extends
from sea level Z = 0 to height Z = H .
Firstly we write ∆ x = X − x, ∆ y = Y − y, ∆ z = Z − z, and the potential
becomes
ˆ H−z
1
V (∆ x, ∆ y, ∆ z) = G √ d ∆ z.
−z ∆ x2 + ∆ y2 + ∆ z2
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
Now we can expand this into a Taylor series in H around the point H = 0:
the first derivative of equation 1.10 is
∂V G G
=√ =
∂H 2 2
( X − x ) + (Y − y ) + ( H − z )2 ℓ
√
in which ℓ ( H ) = ( X − x)2 + (Y − y)2 + ( H − z)2 . The second derivative is
(chain rule)
∂2 V ∂ 1 H−z
= ℓ ( H )−1 = − ℓ−3 · 2 ( H − z) = −G 3 .
∂H 2 ∂H 2 ℓ
The third derivative, in the same way:
∂3 V ∂ 3 ( H − z)2 ℓ2 3 ( H − z)2 − ℓ2
( )
H−z
= − 3 = − = G ,
∂H 3 ∂H ℓ ℓ5 ℓ5 ℓ5
and so on. The Taylor expansion is
1 1 z 1 3 z2 − ℓ20 3
V= 0 +G H + G 3 H2 + G H +... (1.11)
ˆ   ℓ0 2 ℓ0 6 ℓ50
H =0
1
dZ
0 ℓ
√
in which ℓ0 = ( X − x)2 + (Y − y)2 + z2 .
Ó »îá .
Question: how can we exploit this result for computing the gravitational
potential of a complete, realistic topography?
1 1 z
Answer: in this expansion, the coefficients , , . . . like ℓ0 , depend
ℓ0 2 ℓ30
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 for the given z value and for all
possible value pairs (∆ x, ∆ y). Then, if the grid is, e.g., N × N in size,
we need only N 2 operations for calculating every coefficient. The
brute-force evaluation of the Taylor expansion itself for the whole
topography, i.e., for every point of the terrain grid and every point of
the evaluation grid, requires after that N 4 operations, but those are
much simpler: the coefficients themselves have been precomputed.
And brute force isn’t even the best approach: as we shall see, the
above convolution can be computed much faster using the Fast
Fourier Transform.
We shall return to this subject more extensively in connection with
terrain effect evaluation, sections 5.3 and 8.7.
D 1.8 Equations of Laplace and Poisson

The second derivative with respect to place of the geopotential, the first
derivative with respect to place of the gravitational acceleration vector,
i.e., its divergence, is also of geophysical interest. We may write:
⟨−
→ ⟩ ⟨−
→ (−
→ )⟩ ⟨−
→ −
→⟩
div a = ∇ ·a = ∇· ∇V = ∇·∇ V=
∂2 ∂2 ∂2
= ∆V = V+ V + V, (1.12)
∂x 2 ∂ y2 ∂ z2
where
def
⟨−
→ −
→⟩ ∂2 ∂2 ∂2
∆= ∇·∇ = + +
∂x 2 ∂ y2 ∂ z2
is a well known symbol called the Laplace11 operator.
In equation 1.4 for the potential of a point mass we may show, by per-
forming all partial derivations 1.12, that
∆V = 0, (1.13)
11
Pierre-Simon, marquis de Laplace (1749 – 1827) was a French universal genius
who contributed to mathematics and natural sciences. He is one of the 72 French
scientists, engineers and mathematicians whose names were inscribed on the
Eiffel Tower, https://en.wikipedia.org/wiki/List_of_the_72_names_on_the_Eiffel_
Tower.
Ó »îá .
Gauge invariance 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 allowed to interchange integration
and partial differentiation, if both are defined.
A potential field for which the Laplace equation 1.13 applies, is called a
harmonic field or function.
In the case where the mass density doesn’t vanish everywhere, we have a
different equation, with ρ the mass density:
∆V = −4πG ρ . (1.14)
This equation is called the Poisson12 equation.

The pair of equations
−−→
gradV = a
div a = −4πG ρ
is known as the field equations of the gravitational field. They play the
same role as Maxwell’s13 field equations in electromagnetism. Unlike in
Maxwell’s equations, however, in the above there is no time co-ordinate.
Because of this, it is not possible to derive a formula for 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 approximate,
and that a more precise theory is Einstein’s general theory of relativity.
Nevertheless, in physical geodesy Newton’s theory is generally precise
enough and we shall use it exclusively.
D 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,
mittainvarianssi changes. This is called gauge invariance. Gravitation itself is obtained by
differentiation, an operation that eliminates the constant term. Therefore
12
Siméon Denis Poisson (1781 – 1840) was a French mathematician, physicist
and geodesist, one of the 72 names inscribed on the Eiffel Tower.
13
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.
Ó »îá .
the definition of potential is in a sense ambiguous: all potential fields V

obtained by different choices of C are equally valid.
Observations only give us potential differences, as spirit levellers know
all too well.
An often chosen definition of potential is based on requiring that, if
∥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 also that is
not free of problems.
E.g., for the mass of the Earth M⊕ a physically sensible form of the
potential is, in spherical approximation,
GM⊕ def
V= , r = ∥r∥
r
which vanishes at infinity r → ∞, when again a geodetically sensible
definition would be
GM⊕ GM⊕
V= − ,
r R
def
in which R = ∥R∥ is the radius of the Earth. The latter potential vanishes
where r = R , on the surface of the spherical Earth. In the limit r → ∞ its
value is − GM ⊕
R , not zero.
D 1.10 Single mass density layer

If we apply to the surface S of a body a “coating” of mass surface density,
of mass density value
dm
κ= ,
dS
we obtain for the potential an integral equation similar to equation 1.8,
but a surface integral:
¨ ¨
dm κ
V =G =G dS. (1.15)
surface ℓ surface ℓ
Here again ℓ is the distance between the point of consideration or test
mass P and the moving mass element in integration dm (or surface
element dS ). Note that the dimension of the mass surface density κ is
kg /m², different from the dimension of ordinary (volume) mass density.
This case is theoretically interesting, though physically unrealistic. The

function V is everywhere continuous, also at the surface S ; however al-
ready its first derivatives with place are discontinuous. The discontinuity
appears in the direction perpendicular to the surface, i.e., in the normal
derivative.
Let us look at the simple case where a sphere, radius R , has been coated
with a layer of standard surface density κ. By computing the above
Ó »îá .
Double mass density layer 15
integral 1.15 we may prove (in a complicated way) that the exterior
potential is the same as it would be if all of the mass of the body were
concentrated in the sphere’s centre. Earlier, in section 1.4, we proved
that the potential interior to the sphere is constant.
Thus, the exterior attraction (ℓ > R ) is
( )2
M κ · 4π R 2 R
a e (ℓ ) = G 2 = G = 4πG κ .
ℓ ℓ2 ℓ
The interior attraction (ℓ < R ) is
a i (ℓ) = 0.
This means that on the surface of the sphere, the attraction is discontin-
uous:
a e (R ) − a i (R ) = 4πG κ.
In this symmetric case we see, that
∂V
a = ∥a∥ = (1.16)
∂n
where the differentiation variable n represents the normal direction,
i.e., 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.
D 1.11 Double mass density layer

A double mass density layer may be interpreted as a dipole density layer.
The dipoles are oriented in the direction of the surface’s normal.
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
dM
µ= ,
dS
in which d M is a “dipole layer element”; this layer may be seen as made
up of two single layers. If we have a positive layer at density κ and a
negative layer at density −κ, and the distance between them is δ, we get
for small values of δ an approximate correspondence:
µ ≈ δκ.
Ó »îá .
P n
ℓ
δ κ
−κ
D Figure 1.4. A double mass density layer.
The combined potential of the two single mass density layers computed
as explained in the previous section is
¨ ( )
1 1
V =G κ − dS .
surface ℓ1 ℓ2
Between ℓ1 , ℓ2 and δ exists the following relationship (Taylor expansion

of function 1ℓ ):
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
¨ ¨
∂ (1) ∂ (1)
V =G κδ dS = G µ dS .
surface ∂n ℓ surface ∂n ℓ
If δ is small enough (and κ correspondingly large), this holds exactly.
One can easily show that the above potential isn’t even continuous. The
discontinuity happens at the surface S . Let us look again, for the sake of
simplicity, at a sphere, radius R , coated with a double layer of constant
mass density µ. The exterior potential is
¨
∂ (1)
Ve = −G µ dS = 0, (1.17)
surface ∂n ℓ
because the integral vanishes. To prove this, we need the Gauss integral
theorem, on which more below.
The interior potential is
¨ ( 1 )2 ⏐⏐
∂ (1) 2
Vi = −G µ dS = −4πR G µ ⏐ = −4πG µ,
surface ∂ n ℓ ℓ ⏐
ℓ= R
Ó »îá .
The Gauss integral theorem 17
by computing the surface integral using the sphere’s centre as the evalu-
ation point, and using the earlier established circumstance that inside a
sphere covered by a single mass density layer the potential is constant.
Now in the limit ℓ → R the result is different for the exterior and interior
potentials. The difference is
Ve (R ) − Vi (R ) = 4πG µ.
D 1.12 The Gauss integral theorem

D 1.12.1 Presentation
The Gauss14 integral theorem, famous from physics, looks in its vector
form like this: ˚ ¨
div a d V = 〈a · n〉 dS, (1.18)
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 surface of body V .
This theorem applies to all differentiable vector fields a and all “well
behaved” bodies V on whose surface ∂V , everywhere a normal direction
n exists. In other words, this is not a special property of the gravitational
acceleration vector, though it applies also to this vector field.
D 1.12.2 Intuitive description
Let us note that

div a = ∆V = −4πG ρ
is a source function. It describes the amount, in the part of space inside
lähteet, nielut surface S , of positive and negative “sources” and “sinks” of the gravita-
tional field.
The situation is fully analoguous with the flow pattern of a liquid: positive
charges correspond to points where liquid is added to the flow, negative
charges15 correspond to “sinks” through which liquid disappears. The
vector a is in this metaphor the velocity of flow; in the absence of “sources”
and “sinks” it satisfies the condition div a = 0, which describes the conser-
vation and incompressibility of matter.
On the other hand, the function
∂V
〈 a · n〉 =
∂n
vuo is often called the flux, in other words, how much field stuff “flows out” —
14
Johann Carl Friedrich Gauss (1777 – 1855) was a German mathematician and
universal genius. “Princeps mathematicorum”.
15
But the charges for gravitation, i.e., masses, are always positive.
Ó »îá .
Field line
.
Flux
.
.
.
. .
Body
Sources surface
.
.
. .
.
.
.
Figure 1.5. A graphical explanation of the Gauss integral theorem. The concept
D of field line was Michael Faraday’s invention.
just like a liquid flow — from the space inside the surface S to the outside
through S .
The Gauss integral theorem states the two amounts to be equal: it is
in a way a book-keeping statement demanding that everything which is
produced inside a surface — div a — has also 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, i.e., (+ + + . . .), has to be the same as the
∑
sum of “flux” (↑↑↑ . . .) over the whole boundary surface delimiting this
inner space.
D 1.12.3 The potential version of the Gauss theorem
Let us write the Gauss equation a bit differently, using the potential
instead of the gravitational vector:
˚ ¨
∂V
∆V d V = dS, (1.19)
V ∂V ∂n
where we have done the above substitutions. We also here see the popular
notation ∂V for the surface of the body V . The presentational forms 1.19
and 1.18 are connected by the equations 1.12 and 1.9, connecting V and
a.
D 1.12.4 Example 1: a little box
Let us look at a little rectangular box with sides ∆ x, ∆ y and ∆ z, so little,

that the field a ( x, y, z) is inside it an almost linear function of place. Let
Ó »îá .
The Gauss integral theorem 19
a+z
a+y
∆z a+
x
a−
x
a−y
a−z
∆y
∆x
D Figure 1.6. A little rectangular box.
us write a as the gradient of the potential V :

∂V ∂V ∂V
a = ∇V = i+ j+ k = a xi + a yj + a z k
∂x ∂y ∂z
where
∂V ∂V ∂V
ax = ,ay = , az = .
∂x ∂y ∂z
Now the volume integral
˚
∂a x ∂a y ∂a z
( )
div a d V ≈ + + ∆x ∆ y ∆z (1.20)
V ∂x ∂y ∂z
while the surface integral
¨
〈a · n〉 dS ≈ a+x − a−x ∆ y ∆ z + a+y − a−y ∆ x ∆ z + a+z − a−z ∆ x ∆ y .
( ) ( ) ( )
∂V
Here a+x is the value of component a x on the one face in the x direction
and a−x its value on the other face, etc. E.g., a+z is the value of a z on the
box’s upper and a−z on its lower face. A box has of course six faces, in each
of three co-ordinate directions both “up- and downstream”.
Then
∂a x
a+x − a−x ≈ ∆ x,
∂x
∂a y
a+y − a−y ≈ ∆ y,
∂y
∂a z
a+z − a−z ≈ ∆ z,
∂z
and by substitution we see that
¨
〈a · n〉 dS ≈ ∆x · ∆ y ∆z + ∆ y · ∆x ∆z + ∆z · ∆x ∆ y =
∂V ∂x ∂y ∂z
( )
= + + ∆ x ∆ y ∆ z,
∂x ∂y ∂z
Ó »îá .
the same equation as 1.20. So, in this simple case the Gauss equation
applies.
Obviously the equation works also, 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 bit harder to prove that the equation also applies to bodies having
inclined surfaces.
D 1.12.5 Example 2: The Poisson equation for a sphere
According to the Poisson equation 1.14 we have
∆V = −4πG ρ .
Assume a sphere, radius R , within which the mass density ρ is constant.

The volume integral thoughout the sphere gives
˚ ˚
∆V d V = −4πG ρ d V = −4πG ρ V = −4πGM, (1.21)
V V
in which M = ρ V is the total mass of the sphere.

On the surface of the sphere, the normal derivative is
∂V ∂ GM ⏐⏐
⏐
GM
= =− 2 ,
∂n ∂ r r r =R
⏐ R
a constant, and its integral over the surface of the sphere is
¨
∂V GM GM
dS = − · S = − 2 · 4πR 2 = −4πGM. (1.22)
∂V ∂n R 2 R
The results 1.22 and 1.21 are identical, as the Gauss theorem 1.19 re-
quires.
D 1.12.6 Example 3: a point mass in an eight-unit cube
See figure 1.7.

Let us assume that we have a point mass of size GM in the centre of a
cube bounded by the co-ordinate planes x, y, z ∈ {−1, +1}. In that case the
volume integral is
˚ ˚
∆V d V = −4πGM δ (r) d V = −4πGM,
V V
where δ (r) is Dirac16 ’s delta function in space, having an infinite spike at

the origin, being zero elsewhere, and producing a value of 1 upon volume
16
Paul Adrien 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. Physics Nobel laureate 1933, shared with
Erwin Schrödinger. He is also believed to have been on the autism spectrum
(https://en.wikipedia.org/wiki/The_Strangest_Man).
Ó »îá .
Green’s theorems 21
+1
z
0
GM, @ (0, 0, 0)
x
−1 y
0 −1
0
y +1
x
D Figure 1.7. Eight-unit cube.
integration. The surface integral is six times that over the top face
ˆ +1
[ˆ
+1
]
1
−GM ( )3/2 dx d y.
−1 −1 x2 + y2 + 1
Integrating with respect to x (expression in square brackets) yields

ˆ +1
⏐+1
1 x 2
⏐
)3/2 dx = ( 2 ) √ 2 ⏐ =( .
⏐
)√
2 2 y2 + 2
(
−1 x2 + y2 + 1 y +1 x + y +1⏐ y +1 −1
Integrating this with respect to y yields

ˆ +1
⏐+1
2 y ⏐
( )√ d y = 2 arctan √ ⏐ =
y2 + 1 y2 + 2 y2 + 2 −1
⏐
−1
1 π 2
= 4 arctan p = 4 · = π.
3 6 3
Adding the six faces together yields
ˆ +1
[ˆ
+1
]
1
−6 · GM ( )3/2 dx d y = −4πGM,
−1 −1 x 2 + y2 + 1
agreeing with the volume result above.
D 1.13 Green’s theorems

Apply the Gauss integral theorem to the vector field
−
→
F =U ∇ V.
Ó »îá .
Here U and V are two different scalar fields. We obtain:

˚ ˚ ⟨
−→ ( −
→ )⟩
div F d V = ∇ · U ∇V dV =
V
˚ V
˚ ⟨−
→ −
→ ⟩
= U ∆V d V + ∇ U · ∇ V dV =
˚V ˚V (
∂U ∂V ∂U ∂V ∂U ∂V
)
= U ∆V d V + + + dV
V V ∂x ∂x ∂y ∂y ∂z ∂z
and
¨ ¨ ⟨ −
→ ⟩
〈F · n〉 dS = U ∇ V · n dS =
∂V
¨ ¨∂V
⟨−
→ ⟩ ∂V
= U ∇ V · n dS = U dS .
∂V ∂V ∂n
The end result is Green17 ’s first theorem:

˚ ˚ ( ¨
∂U ∂V ∂U ∂V ∂U ∂V ∂V
)
U ∆ V dV + + + dV = U dS.
V V ∂x ∂x ∂y ∂y ∂z ∂z ∂V ∂n
This may be cleaned up, because the second term on the left is symmetric
for the interchange of 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:
˚ ¨
∂V ∂U
( )
(U ∆ V − V ∆ U ) d V = U −V dS.
V ∂V ∂n ∂n
We assume in all operations, that the functions U and V are “well be-
haved”, i.e., all necessary derivatives etc. 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 in point P , where
it is not defined.
In the case where point P is outside the surface ∂V , the result, Green’s
third theorem, is now obtained by substitution:
˚ ¨
1 ∂V ∂ (1)
( )
1
∆V d V = −V dS .
V ℓ ∂V ℓ ∂n ∂n ℓ
This case is depicted in figure 1.8.
17
George Green (1793 – 1841) was a British mathematical physicist, an au-
todidact, working as a miller near Nottingham. He also invented the word
“potential”.
http://www-history.mcs.st-and.ac.uk/Biographies/Green.html,
http://www.greensmill.org.uk/,
http://en.wikipedia.org/wiki/An_Essay_on_the_Application_of_Mathematical_
Analysis_to_the_Theories_of_Electricity_and_Magnetism.
Ó »îá .
Green’s theorems 23
Distance ℓ
Surface element d V
Surface
normal
n
Body V
Surface S = ∂V
Figure 1.8. Geometry for deriving Green’s third theorem if point P is outside
D surface ∂V .
In the case that point P is inside surface ∂V , the computation becomes a

little more complicated. One ought to learn about the clever technique
that — in this case as in others — comes to the rescue. This is why we
describe it shortly.
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:
˚ ˚ ˚
1 1 1
∆V d V = ∆V d V − ∆V d V ,
V ℓ V1 ℓ V2 ℓ
where the second term can be integrated in spherical co-ordinates:

˚ ˆ ϵ
1 1
∆V d V ≈ ∆VP 4πℓ2 d ℓ = 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 the Gauss integral theorem
1.19 :
¨ ¨ ˚
1 ∂V 1 ∂V 1 1 4
dS = dS = ∆V d V ≈ ∆VP · πϵ3 ,
∂V2 ℓ ∂n ϵ ∂V2 ∂n ϵ V2 ϵ 3
which also goes to zero for ϵ → 0.
The second surface integral (the normal is pointing away from P ):
¨ ¨
∂ (1) 1 2 1
− V dS = − V ·− dS ≈ 4 πϵ · 2 VP = 4πVP .
∂V2 ∂n ℓ ∂V2 ϵ2 ϵ
By combining all results with their correct algebraic signs we obtain —

for the case where P is inside surface ∂V1 ∼ ∂V — :
˚ ¨
1 ∂V ∂ (1)
( )
1
∆V d V = −4πVP + −V dS . (1.23)
Ó »îá .
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
D surface ∂V .
After this it must be intuitively clear, and we present without formal

proof, that
˚ ¨
1 ∂V ∂ (1)
( )
1
∆V d V = −2πVP + −V dS,
if point P is on the boundary surface of body V , i.e., on ∂V . This how-
ever presupposes that the normal derivative, and especially the normal
direction, actually exist in precisely point P !
In geodesy, the typical situation is that where the body V over the volume
of which one wants to evaluate the volume integral, is the whole space
outside the Earth. In this case, conveniently ∆V = 0 and the whole volume
integral appearing above vanishes.
The result 1.23 may be generalized to this case, where V is the whole
space outside surface S . This generalization is done by now choosing as
the surface S the three-part surface S = S 1 + S 2 + S 3 , in which S 3 is a
sphere of large radius centred on P . Its radius is then allowed to grow in
the limit to infinity, so that all integrals over both the surface S 3 and the
part of space outside it vanish. Also the normal direction to the surface
S 3 , like that of the little ball used earlier, is inverted, i.e., now aimed to
the inside, towards the Earth.
The end result is:
˚ ¨
1 ∂V ∂ (1)
( )
1
∆V d V = −4πVP − −V dS, (1.24)
Because in this case, where V is the part of the space external to the
Earth, the left-hand side integral vanishes, we may express the potential
at point P suitably as a two-term surface integral over surface ∂V . See
below.
D 1.14 The Chasles theorem

We study the above mentioned case where the “body” is the space outside
the surface ∂V — i.e., in practice: the space outside the Earth.
Ó »îá .
The Chasles theorem 25
Point P
Integration space V
Boundary ∂V , part 2
Matter
(Limit)
D Figure 1.10. Green’s third theorem for the space external to a body.
From the Green theorem 1.24 derived above, we may derive for a har-
monic function V (so, ∆V = 0) in the exterior space:
¨ ¨
1 1 ∂V 1 ∂ (1)
VP = − dS + V dS . (1.25)
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 by equation
1.15:
1 ∂V
κ=− ,
4πG ∂ n
the surface density of a double mass layer is obtained by equation
1.17:
V
µ= .
4πG
If we plug these into equation 1.25, we obtain
¨
κ ∂ (1)
[ ]
VP = G +µ dS .
∂V ℓ ∂n ℓ
In case that the surface ∂V is an equipotential surface of potential V , i.e.,

V = V0 , it follows that a single mass density layer suffices, because in
that case ¨ ¨
∂ (1) ∂ (1)
V dS = V0 dS = 0,
∂V ∂n ℓ ∂V ∂ n ℓ
Ó »îá .
the right-hand side integral vanishes based on the Gauss integral theo-
rem. This is because the function 1/ℓ (with ℓ the distance from point P ) is
harmonic inside the Earth’s body of which ∂V is the surface. This is the
Chasles theorem18 , also called the Green equivalent-layer theorem.
The theorem is used in Molodensky’s19 theory. Also the representation of
the Earth’s gravity field by underground point-mass layers, e.g., Vermeer
(1984), might be justified with this theorem.
The case where ∂V is an equipotential surface is realized 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. Also in
electrostatic theory, for a conductor in which the electrons can move
freely, the physical surface will become one equipotential surface. This is
why it is often stated that the electric charges are on the surface of the
conductor. This isn’t necessarily so, but from a practical viewpoint the
result is the same.
Equation 1.25 simplifies in this case as follows:
¨ ¨
1 1 ∂V κ
VP = − dS = G dS . (1.26)
4π ∂V ℓ ∂n ∂V ℓ
The equation tells us that we can compute the whole potential exterior to
the Earth, if only on the surface of the Earth (the shape of which we also
assume given in order to compute 1 /ℓ!) is given the gradient, in the normal,
i.e., vertical, direction, of the potential, ∂V /∂n. This gradient is precisely
the gravitational acceleration, a quantity derivable from measurement.
All of gravimetric geopotential determination (“geoid determination”),
ever since G. G. Stokes, is based on this.
D 1.15 Boundary-value problems

The boundary-value problem (BVP) is the problem of computing the
potential V throughout space (or throughout the body’s exterior or interior
part of space) from given values relating to V on the boundary surface,
e.g., on the surface of the Earth. The simplest boundary-value problem is
Dirichlet20 ’s problem: on the boundary surface the potential V itself is
given. More complicated boundary-value problems are based on linear
18
Michel Chasles (1793 – 1880) was a French mathematician and geometrician,
one of the 72 whose names are inscribed on the Eiffel Tower.
19
Mikhail Sergeevich Molodensky (1909 – 1991) was an illustrious Russian phys-
ical geodesist.
20
Peter Gustav Lejeune Dirichlet (1805 – 1859) was a German mathematician
also known for his contributions to number theory.
Ó »îá .
What the boundary-value problem cannot compute 27
functionals of the potential: on the boundary, some linear expression in V

is given, e.g., a derivative or a linear combination of derivatives, generally
L {V } ,
where L {·} is a linear operator.

The Dirichlet boundary-value problem in the form popular in geodesy is:
determine the potential field V if its values are given on a closed surface
S , and furthermore is given that V is harmonic (∆V = 0) outside surface
S . In the vacuum of space, the potential is always harmonic, as already
Gm
earlier noted: the potential of a point mass mP , V = ℓ p , 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.
D 1.16 What the boundary-value problem cannot compute

Based on the values of the potential function V on the surface S we
may thus compute the 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 is given that the mass distribution is
spherically symmetric, then nevertheless the density profile along the
radius remains indeterminate. All mass may be concentrated in the
centre, or it may be as a thin layer just under the sphere’s surface, or any
alternative between these two extremes. Without additional information
— e.g., from seismic studies or geophysical density models — we cannot
resolve this issue.
Also the Chasles theorem mentioned above, equation 1.25, and its special
case, equation 1.26, are examples of this: the theorem tells 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 origin is a mass
distribution extending throughout the body!
This is a fundamental limitation of all methods that try to obtain in-
formation on the situation inside the Earth based only on gravimetric
measurements on or outside the Earth.
Ó »îá .
D 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 acceleration
of free fall is 1 cm /s2 . What is the acceleration of free fall on another
asteroid that is otherwise similar, but that 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?
def
7. Is a linear potential, V ( x, y, z) = a + bx + c y + 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 gravi-
tational 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, section 1.11. What is the SI unit of this
quantity?
D 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 given are the density ρ
and radius R core .
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 g core .
Ó »îá .
Exercise 1 – 2: Atmosphere 29
g
g0
Σ1
d
∞ ∞
Σ2
D Figure 1.11. Iron-ore body.
3. What is the attraction g at the centre of the core? What can you say
in general about the geopotential in this point (don’t try to calculate
it)?
∂g
4. Derive the equation for the gravity gradient on the surface of a
∂r
homogeneous-density sphere of density ρ .
D Exercise 1 – 2: Atmosphere
1. The mean pressure of the atmosphere at sea level is 103 hPa (the
unit Pascal: Pa = Nm−2 .) On the Earth surface gravity is 10 m /s2 .
Calculate the mean surface density as a thin layer κ in units of
kg /m2 .
2. Calculate the total mass of the atmosphere using the spherical shell
approximation. You may take as its radius 6378 km.
3. Calculate the attraction generated by the atmosphere outside it,
both as acceleration and as a fraction of the total Earth attraction.
4. How much is the attraction from the atmosphere inside the atmo-
sphere?
D Exercise 1 – 3: The Gauss theorem
There is a deposit (body) of iron ore inside the Earth, which generates (in
the flat Earth approximation!) an attraction on the Earth surface, which
has been drawn as the g curve. See figure 1.11.
The true attraction curve is approximated by a simple function
{
g 0 if r≤d
g=
0 if r>d
Ó »îá .
(red dashed line), where r is the distance from the point on the Earth
surface straight above the ore deposit. So, the area where g ̸= 0 for a disk
of radius d on the surface of the Earth.
1. Compute, using the above approximation for g, the surface integral

¨
g d σ,
Σ1
where Σ1 is the surface of the Earth, see 1.11.

2. According to the Gauss equation
¨ ¨ ˚
′
∆V dV =
( )
(− g) d σ + + g dσ =
Σ1 Σ2 volume
˚
= −4πG ρ iron dV =
volume
= − 4πGMbody ,
where Σ1 + Σ2 is the (two part) closed surface around the body.

(The parts meet at infinity.) g and g′ are the gravity acceleration
functions on the surface of the Earth and on the surface Σ2 —
positive in the upward direction21 .
Assuming that
¨ ¨ ¨
′
g dσ − g dσ = 2 g d σ =⇒
Σ1 Σ2 Σ1
¨
=⇒ 2 g d σ = 4πGMbody ,
Σ1
calculate GMbody .
3. Assuming that the deposit is a sphere at depth d , calculate GM
using Newton’s law of gravitation from the value g 0 straight above
the deposit at the Earth surface.
4. Compare results 2. and 3. and draw conclusions. Is the function g
given above a good approximation?
21
Be careful with the algebraic signs!
Ó »îá .
D The Laplace equation and its solutions
2
D 2.1 General
An equation central to the study of the Earth’s gravitational field is the
Laplace equation,
∂2 ∂2 ∂2
( )
∆V = + + V = 0.
∂ x2 ∂ y2 ∂ z2
We call the symbol ∆ the Laplace operator. In many texts, the alternative
notation ∇2 is used.
If we study gravitation as a field, then the Laplace equation is more
natural than Newton’s formalism. Newton’s equations are 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 of the gravitational field thoughout 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-phased process. The conceptual difference is, that a certain
property, a field, is attributed to empty space, and we no longer talk about
action at a distance directly between two bodies.
Solving the Laplace equation in the general case may be difficult. The
approach generally taken is, that we choose some co-ordinate frame —
a rectangular frame (as above), spherical co-ordinates, cylindrical co-
ordinates, toroidal co-ordinates, or whatever — which fits best with
the geometry of the problem at hand. Then, we transform the Laplace
equation to those co-ordinates, we find special solutions of a certain form,
and finally we compose a general — or not-so-general — solution as a
linear combination of those special solutions, i.e., a series expansion.
Fortunately the theory of linear partial differential equations is well
developed. Similar theoretical problems are encountered in the theory
of the electromagnetic field (Maxwell theory) and quantum mechanics
– 31 –
32 The Laplace equation and its solutions
(Schrödinger1 equation), not to mention fluid and heat flow.

An important observation is, that the Laplace equation is linear. This
means that, if given are two solutions
∆V1 = 0 and ∆V2 = 0,
then also their linear combinations
V = αV1 + βV2 , α, β ∈ R
are good solutions, i.e., ∆V = 0. This linearity property 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 it a local equation. It describes the behaviour of the
potential field in one point and its small neighbourhood. However, the
solution is sought for a whole area. The solution technique generally used
is the so-called boundary-value problem. This means that the field values
(“boundary values”) have to be given only on the boundary of a certain
part of space, e.g., on the Earth’s surface. From this, one calculates the
values of the field in outer space — the behaviour of the field inside the
Earth remains outside the scope of our interest. From the perspective of
the exterior gravitational field one doesn’t 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!
D 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 analoguous to that of spherical
co-ordinates but the math is much simpler.
Assume that the Earth’s surface is the level surface for z co-ordinates
z = 0. Then
∂2 ∂2 ∂2
( )
∆V = + + V = ∆ ( X ( x) · Y ( y) · Z ( z)) ,
∂ x2 ∂ y2 ∂ z2
in which we have “experimentally” written
V ( x, y, z) = X ( x) · Y ( y) · Z ( z) .
1
Erwin Rudolf Josef Alexander Schrödinger (1887 – 1961) was a German physi-
cist 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.
Ó »îá .
The Laplace equation in rectangular co-ordinates 33
In other words, we write experimentally V as the product of three factor

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

∂ x2 ∂ y2 ∂ z2
+ + = 0.
X ( x) Y ( y) Z ( z)
Because this has to be true for all values x, y, z, it follows that each term
must be a constant. If we take for the first and second term − k21 and
− k22 , we get in conclusion for the third constant k21 + k22 . By writing this
definition and result out and moving the denominator to the other side,
we obtain
∂2
X ( x) = − k21 X ( x) ,
∂ x2
(why the minus sign? We shall presently see. . . )
∂2
Y ( y) = − k22 Y ( y) ,
∂ y2
and
∂2
Z ( z) = k21 + k22 Z ( z) .
( )
∂z 2
Now, the solution is readily found at least to the first two equations: they
harmoninen are harmonic oscillators, and their basis solutions2 are
värähtelijä
X ( x) = exp (± ik 1 x) ,
Y ( y) = exp (± ik 2 y) .
The solution of the Z equation on the other hand is exponential:

( √ )
Z ( z) = exp ± z k21 + k22 .
In principle we can now form the solution in space:

( √ )
Vk1 k2 ( x, y, z) = exp i (± k 1 x ± k 2 y) ± z k21 + k22 .
2
Alternative basis solutions are X (x) = sin k 1 x, X (x) = cos k 1 x etc. They
are equivalent with those presented because exp (ik 1 x) = cos k 1 x + i sin k 1 x,
exp (− ik 1 x) = cos k 1 x − i sin k 1 x.
Ó »îá .
The general solution is obtained by summing the terms Vk1 k2 with varying
coefficients, for different values of k 1 , k 2 .
We cannot choose the values k 1 , k 2 entirely freely. Which values are
allowed, will depend on the boundary conditions given.
Let us assume that both in the x and in the y direction the size of
our world is L (“shoebox world3 ”). Let us make things a little simpler
by assuming that, on the borders of our shoebox world, we have the
boundary conditions
V (0, y) = V (L, y) = V ( x, 0) = V ( x, L) = 0.
It then follows that the only pairs ( k 1 , k 2 ) yielding a solution that fits the
box are
πj πk
k1 = , k2 = , j, k ∈ Z,
L L
and the only suitable functions are sine functions. Thus we obtain as a
solution:
x) ( y) )z)
( ( √(
V jk ( x, y, z) = sin π j sin π k exp ±π j 2 + k2 .
L L L
This particular solution may now be generalized by multiplying it with
suitable coefficients, and summing it over different index values j =
0, ±1, ±2, . . . ; 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, . . . ; k = 1, 2, . . ..
Different boundary conditions will give slightly different general solu-
tions. Their general form is however always similar.
The zero-level z = 0 expansion resulting from the general solution is the
familiar Fourier4 sine expansion:
∞ ∑
∑ ∞
V ( x, y, 0) = v jk V jk ( x, y) =
j =1 k=1
∞ ∑ ∞ [ ( [ ]) ( [ ])]
∑ jx ky
= v jk sin π sin π , (2.1)
L L
j =1 k=1
where the v jk are Fourier coefficients, and the expressions

( [ ]) ( [ ])
def jx ky
V jk ( x, y) = sin π sin π
L L
3
. . . though real-world shoeboxes are rarely square.
4
Joseph Fourier (1768 – 1830) was a French mathematician and physicist — and
some would say, climatologist —, one of the Eiffel Tower’s 72 names.
Ó »îá .
The Laplace equation in rectangular co-ordinates 35
sin 13Lπ x
Sea level
sin 5Lπ x
Figure 2.1. The exponential attenuation of gravitational field Fourier waviness

with height. Rectangular geometry, one dimension. Long waves
(small wave numbers, red) attenuate slower with height than short
D waves (green), i.e., the height acts as a low-pass filter.
are two-dimensional basis functions on the Earth’s surface.

We refer to section B.2.2 in appendix B for a description with illustration
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
∞ ∑
∞
∑ ( √ z)
V ( x, y, z) = v jk V jk ( x, y) exp ±π j 2 + k2 = (2.2)
L
j =1 k=1
∞ ∑ ∞ ( [ ]) ( [ ])
∑ jx ky ( √ z)
= v jk sin π sin π exp ±π j 2 + k2 .
L L L
j =1 k=1
Note, that the z expression may have a positive as well as a negative

algebraic sign! Of course the solution with a positive sign goes to → ∞
when z → ∞, which is not physically realistic in the exterior space.
Note also that V ( x, y, 0) and v jk describe the same gravitational field in
two essentially different ways: in the space domain, and in the — spatial
— frequency, i.e., 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 of one surface
means — with the Laplace equation — knowing the potential throughout
space.
Ó »îá .
D Table 2.1. Vertically shifting the potential field V , in the space and frequency
domains.
Space domain Fourier Frequency domain
V (x, y, 0) F =⇒ v jk
↓ (hard) ⇓(× (easy) )
√
V (x, y, z) ⇐= F −1
v jk exp −π j 2 + k2 Lz
We summarize equations 2.1, 2.2 still in the commutative diagram of

table 2.1.
The takeaway from this is, that the operation of vertically shifting the
potential field V from zero level to the level z, which is not straightfor-
ward in the space domain, becomes simple — as in a straightforward
multiplication — in the frequency domain. The same applies in spherical
co-ordinates, where the frequency domain means spherical-harmonic
coefficients, as we shall see.
D 2.3 Example: The Laplace equation in polar co-ordinates

In polar co-ordinates, i.e., two-dimensionally, the Laplace equation is
∂2 V 1 ∂V 1 ∂2 V
∆V = + + 2 = 0.
∂r2 r ∂ r r ∂α2
Question:
Perform on this the same kind of separation of variables as in
section 2.2, i.e., write first
V ( r, α) = R ( r ) A (α)
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).
What form does the A (α) function of the general solution have?
Answer:
Substitution yields
∂2 R ( r ) A (α) ∂R ( r ) R ( r ) ∂2 A (α)
A (α) + + 2 = 0.
∂r2 r ∂r r ∂α2
r2
Multiply by the expression :
A (α) R ( r )
∂2 A(α)
r 2 ∂2 R ( r ) r ∂R ( r )
( )
∂α2
+ + = 0.
R (r) ∂r 2 R (r) ∂r A (α)
Ó »îá .
Example: The Laplace equation in polar co-ordinates 37
Both terms must be constant:

∂2 R ( r ) ∂R ( r )
[ ]
r r + − k 2 R ( r ) = 0,
∂r2 ∂r
∂2 A (α)
+ k2 A (α) = 0.
∂α2
Here the algebraic sign of k2 has been chosen so, that A (α) gets a
periodic solution. Such a general solution would be
A k (α) = a k cos ( kα) + b k sin ( kα) ,
where, because the angle α has a period of 2π, k has to be a non-

negative integer: k = 0, 1, 2, 3, . . . Negative k do not give different
solutions, because
a k cos ( kα) = a k cos ((− k) α) ,

b k sin ( kα) = (− b k ) sin ((− k) α) .
The other equation, in the function R ( r ), is harder to solve. A test solution

is a power law:
R (r) = r q .
Substitution yields
r rq ( q − 1)) r q−2 + qr q−1 − k2 r q = 0 =⇒ q2 − k2 = 0 =⇒ q2 = k2 .

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

we find
r − k2 r = 0 =⇒ k2 = 1 = q2 .
For q = 0 besides the trivial constant solution, the non-trivial solution
R ( r ) = ln r is found:
1 1][
r r·− 2 + − k2 ln r = 0 =⇒ k = 0.
r r
Thus we obtain the general solution
{
1 or ln r if k = 0,
R k (r) =
r k or r −k if k = 1, 2, . . . .
We see that, if we require the solution to exist at the origin r = 0, we need
the first solutions, obtaining
∞
∑
V1 ( r, α) = a 0 + r k (a k cos kα + b k sin kα) ,
k=1
but if we require existence — or a least good behaviour — at infinity5

r → ∞, we need the second solutions,
∞
∑
V2 ( r, α) = a 0 ln r + r −k (a k cos kα + b k sin kα) .
k=1
5 ∂V2
In fact, lim V2 → ∞ but lim = 0.
r →∞ r →∞ ∂r
Ó »îá .
Pole Z
r cos φ P
Equator
r r sin φ
φ Y
Greenwich meridian
D Figure 2.2. Definition of spherical co-ordinates.
There is a clear similarity here to the three-dimensional (spherical co-

ordinates) case.
D 2.4 Spherical, geodetic, ellipsoidal co-ordinates

In physical geodesy we use side by side geometrical and physical concepts.
E.g., co-ordinates of place can be given in the form ( X , Y , Z ) , which are in
principle geometric — except for the physical assumption that the origin
of the co-ordinate system is in the centre of mass of the Earth.
As the Earth is not precisely a sphere but rather an oblate ellipsoid
of revolution, one cannot use geographical co-ordinates as if they were
spherical co-ordinates. Because the flattening of the Earth — some
0.3% — cannot be ignored, this difference is significant. The connection
( )
between spherical co-ordinates r, φ, λ and rectangular ones ( X , Y , Z ) is
the following:
X = r cos φ cos λ,
Y = r cos φ sin λ, (2.3)
Z = r sin φ.
Here φ and λ are geocentric latitude and (ordinary, i.e., geocentric or

geodetic or geographic) longitude. r is the distance from the Earth’s
centre. The X axis points in the direction of the Greenwich meridian. See
figure 2.2.
On the Earth’s surface these spherical co-ordinates are not very useful
because of the Earth’s flattening, but in space, spherical co-ordinates are
Ó »îá .
Spherical, geodetic, ellipsoidal co-ordinates 39
Z Ellipsoidal
P normal
h
1 − e2 N + h sin ϕ
[( ) ]
ϕ
X,Y
O
Reference (N + h) cos ϕ
ellipsoid
D Figure 2.3. Definition of geodetic co-ordinates.
much used. On the Earth’s surface on the other hand, most often geodetic
— or geographical — co-ordinates ϕ, λ, h are used:
X = ( N + h) cos ϕ cos λ,
Y = ( N + h) cos ϕ sin λ, (2.4)
2
( )
Z = N + h − e N sin ϕ,
where
( ) a a2
N ϕ =√ =√ . (2.5)
1 − e2 sin2 ϕ a2 cos2 ϕ + b2 sin2 ϕ
The quantity N defined in equation 2.5 is the East-West direction, or
transversal, radius of curvature of the reference ellipsoid. In the equation,
2 2
a is the equatorial radius of the reference ellipsoid used, e2 = a a−2b is the
square of the so-called first eccentricity6 , and in equations 2.4, h is the
height of the point above the reference ellipsoid, see figure 2.3.
Converting rectangular co-ordinates into geodetic ones is easiest to do
iteratively, although the literature also offers closed formulas.
Spherical co-ordinates and geodetic, i.e., geographical, co-ordinates are
considerably different. In latitude, the difference is up to 11 minutes of
arc, or almost 20 km. This maximum is attained for latitudes ±45◦ .
In theoretical work one also uses ellipsoidal co-ordinates u and β. The
redukoitu co-ordinate β is called the reduced latitude. The relationship with rectan-
leveysaste
6
The parameter is connected to the Earth’s flattening f through the equation
e2 = 2 f − f 2 .
Ó »îá .
gular co-ordinates is
√
X= u2 + E 2 cos β cos λ,
√
Y= u2 + E 2 cos β sin λ, (2.6)
Z = u sin β.
If the semi-major axis of the Earth ellipsoid is a and its semi-minor axis
b, it follows — exercise! — that E 2 = a2 − b2 .
D 2.5 The Laplace equation in spherical co-ordinates

The Laplace equation transformed to spherical co-ordinates reads (see
appendix E for a simple geometric proof):
∂2 V 2 ∂V 1 ∂2 V tan φ ∂V 1 ∂2 V
∆V = + + 2 − + = 0, (2.7)
∂r2 r ∂ r r ∂φ2 r 2 ∂φ r 2 sin2 φ ∂λ2
in which φ is the (geocentric) latitude, λ is the longitude, and r is the
distance from the origin or centre of the Earth.
We shall here not derive the solution of this equation by separation of
variables, as it is pretty complicated and can be found in ready form in the
literature (Heiskanen and Moritz, 1967, section 1 – 9). What is significant
is, that the solution looks somewhat similar to the above solution in
rectangular co-ordinates. The basis solutions of the Laplace equation are
( )
( ) n
( ) ( ) Y n φ, λ
Vn,1 φ, λ, r = r Yn φ, λ , Vn,2 φ, λ, r = , (2.8)
r n+1
where the first is again nonphysical in outer space, because, unlike the
true geopotential, these expressions grow to infinity for r → ∞.
In the above equations, the functions Yn are called surface spherical pintapallofunktiot
harmonics, whereas the functions Vn are solid spherical harmonics. The avaruus-
latter are harmonic functions everywhere in space except at the origin pallofunktiot
(2.8, rightmost equation) or at infinity (leftmost, physically unrealistic
equation).
The functions Yn are
n
∑
( ) ( )
Y n φ, λ = P nm sin φ (a nm cos mλ + b nm sin mλ) . (2.9)
m=0
The functions P nm are so-called Legendre functions, on which more later

on. With the help of expression 2.9 we obtain, by using the second,
physically realistic alternative from equations 2.8, the following solution
or series expansion for the potential V in space: pallofunktio-
∞ n kehitelmä
( ) ∑ 1 ∑ ( )
V φ, λ, r = P nm sin φ (a nm cos mλ + b nm sin mλ) . (2.10)
r n+1
n=0 m=0
Ó »îá .
Dependence on height 41
The coefficients a nm and b nm are called the coefficients of the spherical-

harmonic expansion, shortly spectral coefficients. Together they describe
the function V , in somewhat the same way that the Fourier coefficients
v jk do in rectangular co-ordinates in equation 2.2. The subscripts n and
asteluku, m are called degree and order.
järjestysluku
Often we will be using a somewhat freeer notation for the functions Yn .
E.g., if we expand the disturbing potential T into spherical harmonics,
( )
we shall use the notation T n φ, λ for its surface harmonics. Similarly
∆ g n φ, λ is the surface harmonic of the gravity anomaly ∆ g for degree
( )
n, and so on.
D 2.6 Dependence on height

From the above equation 2.8 one sees that for different values of the
degree n the function Vn has a different dependence on the distance r
from the Earth’s centre, or equivalently, on the height H = r − R, if by R
we denote the radius of the Earth. The dependence is
( )
( ) Y n φ, λ
Vn φ, λ, r = .
r n+1
On the Earth’s surface
( )
( ) Y n φ, λ
Vn φ, λ, R = .
R n+1
Therefore we may write
( )n+1 ( )−(n+1)
( ) R ( ) R+H ( )
Vn φ, λ, r = Vn φ, λ, R = Vn φ, λ, R =
r R
( )−(n+1) ( )
H ( H) ( )
= 1+ Vn φ, λ, R ≈ exp − ( n + 1) Vn φ, λ, R .
R R
We see that the dependence of the potential on height is again exponential,
and the harmonic degree number n appears in the exponent, as did also
the wave number in rectangular geometry, see equation 2.2 and figure
2.1. The analogy works.
D 2.7 Legendre’s functions

In the above equations the functions P are so-called Legendre7 functions
that pop up whenever we solve a Laplace-like equation in spherical co-
ordinates. There exist various effective, so-called recursive algorithms,
7
Adrien-Marie Legendre (1752 – 1833) was a French mathematician known for
his work on number theory, statistics — he invented independently from Gauss
the method of least squares — and on elliptical functions. His name is inscribed
on the Eiffel Tower.
Ó »îá .
D Table 2.2. Legendre polynomials. t = sin φ.

( ) ( )
Function of t Function of sin nφ and cos nφ
( )
P0 (t) = 1 P0 sin φ = 1
( )
P1 (t) = t P1 sin φ = sin φ
P2 (t) = 23 t2 − 12 P2 sin φ = − 34 cos 2φ + 14
( )
P3 (t) = 52 t3 − 32 t P3 sin φ = − 58 sin 3φ + 38 sin φ

( )
P4 (t) = 18 35t4 − 30t2 + 3

( )
P5 (t) = 18 63t5 − 70t3 + 15t

( )
1
231t6 − 315t4 + 105t2 − 5
( )
P6 (t) = 16
e.g., the following (only for ordinary Legendre polynomials P n = P n0 ):
nP n ( t) = − ( n − 1) P n−2 ( t) + (2 n − 1) tP n−1 ( t) . (2.11)
Similar equations exist also for the functions P nm , m > 0. There are
even alternatives to choose from, though most equations are complicated.
One should be careful that in their computation, the factorials don’t go
overboard. Already 30! (factorial of 30) is a larger number than most
computers can handle as integers. . . not to mention 360!. Heiskanen
and Moritz (1967) equation 1 – 62, unlike is stated there, is not directy
suitable for computer use!
The first Legendre polynomials are listed in table 2.2.
Higher polynomials than this are rarely needed in manual computation.
Note that the even polynomials are mirror symmetric through the ori-
( ( )) ( )
gin, P n (− t) = P n ( t) — or equivalently P n sin −φ = P n sin φ — and
( ( ))
the odd ones are antisymmetric, i.e., P n (− t) = −P n ( t) or P n sin −φ =
( )
−P n sin φ .
For comparison: also the Fourier basis functions (like, in a somewhat
more complicated way, also sines and cosines!)
( x)
F j ( x) = exp 2π i j ,
L
in which i 2 = −1, can be computed recursively:
F j+1 ( x) = F j ( x) · F1 ( x) .
Of the associated Legendre functions P nm , m ̸= 0 we give several in table Legendren

2.3 for illustration. liitännäisfunktiot
One defining equation for these is

)m/2 d m P n ( t)
P nm ( t) = 1 − t2
(
. (2.12)
dt m
Ó »îá .
Legendre’s functions 43
−90◦ −30◦ 0◦ −→ φ 30◦ 90◦

1
P0 P0 P5
P1 P6
P2
P3 P10
0.5 P5 P4 P25
P6
−0.5 P4 P3
P2
P1
−1
−1 −0.5 0 −→ t 0.5 1
Figure 2.4. A number of Legendre polynomials P0 (t) . . . P25 (t) as functions of

D the argument t = sin φ.
Starting from equation 2.9 we may write

n
∑
( ) ( ( ) ( ) )
Y n φ, λ = a nm P nm sin φ cos mλ + a n,−m P n,m sin φ sin mλ =
m=0
∑n
( )
= a nm Ynm φ, λ ,
m=− n
where now m runs from − n to + n. Here

{ ( )
( ) P nm sin φ cos mλ if m ≥ 0,
Ynm φ, λ = ( )
P n|m| sin φ sin | m| λ if m < 0.
These are the surface spherical harmonics of degree n and order m.

Such surface spherical harmonics come in three kinds:
◦ Zonal functions: m = 0. These functions depend only on latitude.
D Table 2.3. Associated Legendre functions.
Function of t Trigonometric function

p ( )
P11 (t) = 1 − t2 P11 sin φ = cos φ
p ( )
P21 (t) = 3t 1 − t2 P21 sin φ = 3 sin φ cos φ
P22 (t) = 3 1 − t2 P22 sin φ = 3 cos2 φ
( ) ( )
)p
P31 (t) = 23 5t2 − 1 1 − t2 P31 sin φ = 23 5 sin2 φ − 1 cos φ
( ( ) ( )
P32 (t) = 15t 1 − t2 P32 sin φ = 15 sin φ cos2 φ

( ) ( )
)3/2
P33 (t) = 15 1 − t2 3
( ( )
P33 sin φ = 15 cos φ
Ó »îá .
5
P22
P21
P31 P11
−5 P32
P11
P21
−10 P22
P33 P31
P32
P33
−15
−1 −0.5 0 −→ t 0.5 1
D Figure 2.5. Associated Legendre functions.
◦ Sectorial functions: m = n. the algebraic signs of these functions

depend only on longitude and not on latitude. The functions them-
selves however do depend on both latitude and longitude!
◦ Tesseral functions: 0 < m < n. These functions, the algebraic sign
of which changes with both latitude and longitude, form a checker-
board pattern on the surface of the sphere, if the positive values
are painted white and the negative ones black (Lat. 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. The
are however not wave functions (sines or cosines), the connection to
those is complicated at least. It is nevertheless sensible to speak of their
wavelength.
In figure 2.6 is depicted how the algebraic signs of the different spherical etumerkit
harmonics behave on the Earth’s surface — and above.
When looking in equation 2.9 at the expressions cos mλ and sin mλ, we
observe that around a full circle, i.e., the equator, 0 ≤ λ < 2π they go
precisely 2 m times through zero. The “semi-wavelength” is thus puoliaallonpituus
2πR R
=π ,
2m m
where R is again the radius of the Earth.
( )
A similar formula applies also for functions P nm sin φ : as the function
passes through zero n − m times on the interval — from pole to pole —
Ó »îá .
Legendre’s functions 45
(a) (b) (c)

( )
Zonals: P5,0 sin φ Sectorials: Tesserals:
( ) ( )
P6,6 sin φ cos 6λ P11,6 sin φ cos 6λ
Figure 2.6. The algebraic signs of spherical harmonics on the Earth surface.
Grey means positive, white negative. The functions “wave” in a sine
D or cosine function like fashion.
− π2 ≤ φ < π2 , it follows that also here the semi-wavelength is

πR
.
n−m
If we plug various values for m and the expression n − m into this, we
obtain table 2.4.
This table also gives the resolution that can be achieved with a spherical-
harmonic expansion, i.e., in how detailed a fashion the expansion can
describe the Earth’s gravity field. The expansions available today, like
the model EGM2008, go to degree n = 2159; the “sharpness” of a geopo-
tential image based on them is thus 9 km. Models based on satellite orbit
perturbations often extend only to degree 40, meaning that only details
the size of continents — order of magnitude 500 km — will be visible.
On the other hand, experimental spherical-harmonic expansions of the
topography go even up to degree 10,800 (Balmino et al., 2012).
Figure 2.7. Surface spherical harmonics as maps. Horizontal axis λ ∈ [0, 360◦ ),
vertical axis φ ∈ [−90◦ , 90◦ ]. Functions depicted are
( ) ( ) ( )
P50 sin φ P66 sin φ cos 6λ P11,6 sin φ cos 6λ
( ) ( ) ( ) .
D P40 sin φ P65 sin φ cos 5λ P10,6 sin φ cos 6λ
Ó »îá .
D Table 2.4. Semi-wavelengths for different degrees and orders of spherical har-
monics.
m / n−m Semi-wavelength (km) In degrees
10 2000 18◦
40 500 4◦ .5
180 111 1◦
360 55 0◦ .5 = 30′
1800 11 0◦ .1 = 6′
10,800 1.85 1′
D 2.8 Symmetry properties of the spherical-harmonic

expansion
We recapitulate the spherical-harmonic expansion given at the beginning,

equation 2.10:
∞ n
( ) ∑ 1 ∑ ( )
V φ, λ, r = P nm sin φ (a nm cos mλ + b nm sin mλ) . (2.10)
r n+1
n=0 m=0
D 2.8.1 Dependence on latitude φ
It is seen that the dependence on φ only works through the associated

( )
Legendre function P nm sin φ . This function can be either symmetric in
φ, or antisymmetric in φ, meaning that either (symmetric case)
( ) ( ( ))
P nm sin φ = P nm sin −φ
or (antisymmetric case)
( ) ( ( ))
P nm sin φ = −P nm sin −φ .
Equivalently it means that, with t = sin φ, that either (symmetric case)
P nm ( t) = P nm (− t)
or (antisymmetric case)
P nm ( t) = −P nm (− t) .
Which case applies, depends on the values of both n and m. To figure it

out, one can look at, e.g., equation 2.12:
)m/2 d m P n ( t)
P nm ( t) = 1 − t2
(
. (2.12)
dt m
We need to answer two questions:
Ó »îá .
Orthogonality, orthonormality of Legendre functions 47
1. For which values n is the polynomial P n ( t) symmetric, for which is

it antisymmetric in t? For this, you need to look at the recursive
algorithm for computation of the polynomials, eq. 2.11. We already
know that P0 ( t) = 1 is symmetric, and that P1 ( t) = t is antisymmet-
ric. The rule for other n values follows recursively (or you could
cheat by looking at table 2.2).
d
2. What does differentiation dt do to the symmetry or antisymmetry
of the function?
p
(Multiplication by 1 − t2 = cos φ changes nothing, as this factor is
symmetric in t or φ.)
So, in order to make expansion 2.10 mirror symmetric between Northern
and Southern hemispheres, one has to set the coefficients a nm , b nm for
which the corresponding P nm is antisymmetric, to zero. The coefficients
remaining are those for which the corresponding P nm is symmetric.
In table 2.5 we give a code fragment in the octave rapid-prototyping
language to plot an arbitrary surface spherical harmonic, e.g., in order to
visually judge its symmetry properties. Don’t believe, test.
D 2.8.2 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 2.10 change when we change λ?
We see immediately that, for m ̸= 0, there will be dependence on λ if
any coefficient a nm , b nm is non-zero. So, in order to obtain rotational
symmetry, all coefficients a nm , b nm for values m > 0 must be suppressed:
a 11 = b 11 = a 21 = b 21 = a 22 = b 22 = · · · = 0.
Of the remaining coefficients, we can say that for m = 0, sin mλ = 0
identically, so the coefficients b 00 , b 10 , b 20 , . . . simply don’t matter. They
may be any value, including zero. The coefficients a 00 , a 10 , a 20 however
do matter, as for m = 0, cos mλ = 1 identically. So we obtain as the
rotationally symmetric expansion
∞
( ) ( ) ∑ 1 ( )
V φ, λ, r = V φ, r = a P sin φ ,
n+1 n n
r
n=0
in which P n = P n0 and a n = a n0 .
D 2.9 Orthogonality, orthonormality of Legendre functions

Legendre’s polynomials are orthogonal: the integral — formally a scalar
product of vectors —
ˆ +1 2
if n = n′
{
P n ( t) P n′ ( t) dt = 2n+1 . (2.13)
−1 0 if n ̸= n′
Ó »îá .
D Tableau 2.5. Plotting a surface spherical-harmonic map.
% Plotting surface spherical harmonics

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

functions and integrals over functions. There exists a useful analogy with
vector spaces: see appendix B.
Alternatively we may write, on the surface of a unit sphere σ, where we
( )
used a parametrization ψ, α :
¨ ˆ 2πˆ π
( ) ( ) ( ) ( )
P n cos ψ P n′ cos ψ d σ = P n cos ψ P n′ cos ψ sin ψ d ψ d α =
σ 0 0
ˆ −1 ˆ +1
= −2π P n ( t) P n′ ( t) dt = 2π P n ( t) P n′ ( t) dt,
+1 −1
in which t = cos ψ and d σ = sin ψ d ψ d α. So we have

¨ 4π
n = n′
{
( ) ( ) if
P n cos ψ P n′ cos ψ d σ = 2n+1 , (2.14)
σ 0 if n ̸= n′
in which ψ is the angular distance from some point on the surface of the
sphere. Equation 2.14 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 define also fully
normalized Legendre polynomials
( ) p ( )
P n cos ψ = 2 n + 1P n cos ψ , (2.15)
Ó »îá .
Splitting a function into degree constituents 49
in which case the now modified scalar product — the mean square value
over the unit sphere — is
¨
n = n′ ,
{
1 ( ) ( ) 1 if
P n cos ψ P n′ cos ψ d σ =
4π σ 0 if n ̸= n′ ,
showing the polynomials now to be orthonormal8 . Similarly fully nor-

malized associated Legendre functions also exist (without proof, see
Heiskanen and Moritz 1967, page 31):
√
( ) ( n − m)! ( )
P nm cos ψ = 2 (2 n + 1) P nm cos ψ .
( n + m)!
In this case, the orthonormal functions look like

( )
P nm cos ψ cos mα
{
( ) if m ≥ 0,
Y nm ψ, α = ( )
P n|m| cos ψ sin | m| α if m < 0.
The scalar product that applies is

¨
1 if n = n′ and m = m′ ,
{
1 ( ) ( )
Y nm ψ, α Y n′ m′ ψ, α d σ =
4π σ 0 otherwise.
D 2.10 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
formula is (Heiskanen and Moritz, 1967, equation 1 – 71, but using our
notation Yn −→ f n ):

¨
2n + 1
f φ′ , λ′ P n cos ψ d σ′ ,
( ) ( ) ( )
fn φ, λ = (2.16)
4π σ

where ψ is the geocentric angular distance between evaluation point

( ) ( )
φ, λ and moving data or integration point φ′ , λ′ , see figure 7.2. In
asteosuusyhtälö this degree constituent equation 2.16 there is a certain similarity with
the projection or coefficient computation formula B.10. Nevertheless,
here we don’t have a computation of spectral coefficients, but of “spectral
constituent functions” f n .
8
And also ˆ +1
n = n′ ,
{
1 1 if
P n (t) P n′ (t) dt =
2 −1 0 if n ̸= n′ ,
again the mean square value.
Ó »îá .
We bring to mind the core property of the functions f n

∞
∑
( ) ( )
f φ, λ = f n φ, λ
n=0
on the surface of the sphere.

For the proof, we choose as the “north pole” of the co-ordinate system the
( )
point φ, λ ; then, φ′ = 90◦ − ψ. By writing (see equation 2.10):
n
∞ ∑
∑
′ ′
P nm sin φ′ a nm cos mλ′ + b nm sin mλ′
( ) ( )( )
f φ ,λ =
n=0 m=0
and substituting this into the degree constituent equation 2.16, we obtain,
by exploiting the orthogonality of the Legendre functions, on the right-
hand side of the degree constituent equation:
¨
2n + 1
f φ′ , λ′ P n cos ψ d σ′ =
( ) ( )
IR =
4π σ
¨
2n + 1
P n2 cos ψ d σ′ =
( )
= a n0
4π σ
ˆ 2π [ ˆ 2π ]
2n + 1
P n2 ′
( )
= an cos ψ · sin ψ dλ dψ
4π 0 0
+1 ˆ [ ˆ 2π ][ ]
2n + 1 1
= an P n2 ( t) · sin ψ dλ ′
dt =
4π −1 0 sin ψ
2n + 1 2
= · 2π a n · = an,
4π 2n + 1
def
where we used the notation a n = a n0 as well as equation 2.13.
On the left-hand side of the degree constituent equation we obtain, be-
cause, on the assumed north pole, φ = 90◦ and thus sin φ = 1:
n
∑
( ) def ( ◦
)
I L = f n φ, λ = Yn 90 , λ = P nm (1) (a nm cos mλ + b nm sin mλ)
m=0
= P n0 (1) a n0 = a n ,
by using
P n0 (1) = 1,
P nm (1) = 0 if m ̸= 0.
( )
As this applies for every point φ, λ — and note that the values a n depend
on this choice! — it follows that the degree constituent equation 2.16 is
generally true.
Ó »îá .
Spectral representations of various quantities 51
D 2.11 Spectral representations of various quantities

D 2.11.1 The potential
Starting from equation 2.10 we write the following spectral expansion of

the geopotential V in space:
∞ ( ) n+1
( ) ∑ R ( )
V φ, λ, r = Vn φ, λ , (2.17)
r
n=0
where the spectral constituents Vn are the familiar Yn from equation 2.9:
n
∑
aVnm cos mλ + bVnm sin mλ =
( ) ( )( )
Vn φ, λ = P nm sin φ
m=0
∑n
( )
= vnm Ynm φ, λ ,
m=− n
in which the basis functions

( )
P nm sin φ cos mλ
{
( ) if m ≥ 0,
Ynm φ, λ = ( ) (2.18)
P n|m| sin φ sin | m| λ if m < 0,
and coefficients
aVnm if m ≥ 0,
{
vnm = (2.19)
bVn|m| if m < 0.
On the Earth’s surface ( r = R ) we obtain
∞
∑ n
∞ ∑
∑
( ) ( ) ( )
V φ, λ, R = Vn φ, λ = vnm Ynm φ, λ .
n=0 n=0 m=− n
We may summarize the relationships found in a commutative diagram.

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

˜ ∞
( ) ∑ ( )
V φ, λ, R = Vn φ, λ
−→
n=0
↓ (hard) ⇓ × (easy)
∑ ∞
( ) ∑ ( R )n+1 ( )
V φ, λ, r = r Vn φ, λ
←−
n=0
˜ ( ) 2n + 1 ˜ ( ′ ′ ) ( ) ′
: Vn φ, λ = σ V φ , λ , R P n cos ψ dσ
4π
Ó »îá .
D 2.11.2 Gravitation
In the Neumann9 boundary-value problem we solve a function V of which

the normal derivative, ∂∂Vn , is given on the surface of a body or a closed
surface in space.
In the case of a spherical body, we may assume ∂∂n = ∂∂r and work with
spherical-harmonic expansions. By differentiating equation 2.17 we
obtain
∞ ( )n+1
∂V ∑ n+1 R ( )
= − Vn φ, λ . (2.20)
∂r r r
n=0
On the spherical Earth’s surface this means

∞
∂V
( ) ∑ n+1 ( )
=− Vn φ, λ . (2.21)
∂r r =R R
n=0
If we also write on the Earth’s surface for the gravitation

∞
∂V
( ) ∑
( ) def def ( )
g φ, λ, R = = g n φ, λ ,
∂r r =R n=0
it follows by analogy that

( ) n+1 ( )
g n φ, λ = − Vn φ, λ ,
R
and conversely, that
( ) R ( )
Vn φ, λ = − g n φ, λ . (2.22)
n+1
As a result of this we obtain the spectral representation of the solution to
a certain Neumann problem:
∞ ( ) n+1 ( )
( ) ∑ R g n φ, λ
V φ, λ, r = −R . (2.23)
r n+1
n=0
We may write in an analogue fashion

∞
∑ n
∞ ∑
( ) ( ) def ∑ ( )
g φ, λ, R = g n φ, λ = g nm Ynm φ, λ ,
n=0 n=0 m=− n
where the definition used yields consistently
n+1
g nm = − vnm , (2.24)
R
see equation 2.19. This is an interesting result worth thinking about:
9
Carl Gottfried Neumann (1832 – 1925) was a German mathematician.
Ó »îá .
Low-degree spherical harmonics 53
1. Firstly, note how simple the connection 2.24 between potential vnm
and gravitation g nm has become in the frequency domain!
2. Secondly, if there are available, over the whole surface area of the
Earth, measurement values of gravitational acceleration g, we may
derive from these the spherical-harmonic coefficients g nm and the
( )
degree constituent functions g n φ, λ using the method explained
earlier. In this way we can then obtain the solution by means of
equation 2.23 for the whole exterior geopotential field! This is the
basic idea of geopotential — or geoid — determination, from the
spectral perspective.
D 2.12 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 2.10 for degree n = 0
is
1 ( ) a 00
V0 = a 00 P0 sin φ = ,
r r
from which
a 00 = GM.
So a 00 represents the force field of a point mass or spherically symmetric
mass distribution centred at the origin. The higher spherical-harmonic
coefficients are “perturbations” on top of this.
The expansion for the degree one coefficients looks as follows:
( ) 1 ( )
V1 φ, λ, r = 2
a 11 cos φ cos λ + b 11 cos φ sin λ + a 10 sin φ .
r
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
— where {i, j, k} is an orthonormal basis of the Euclidean space R3 —

yielding
1
V1 (r) = 3 〈(a 11 i + b 11 j + a 10 k) · r〉 .
r
Remember that the potential field of a dipole is
G
V (r) =
〈d · r〉 ,
r3
in which d is the dipole moment. Comparison yields
a 11 i + b 11 j + a 10 k = G d,
i.e., 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
Ó »îá .
Figure 2.8. Monopole, dipole, and quadrupole, at the Earth’s centre, and their
D effects on the geoid.
◦ a monopole at the origin of the co-ordinate system, size dm, and

◦ a dipole, size 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:
˚ ˚
d⊕ = r dm = ρ r d V = M⊕ · rcentre of mass ,
Earth Earth
in which, by definition, rcentre of mass 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 a 11 , b 11 , a 10 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 so-called quadrupole moment of the
Earth — corresponding to her inertial tensor —, etc.
D 2.13 Often used spherical-harmonic expansions

Of the existing global spherical-harmonic expansions must be mentioned
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 NGA, National
Geospatial-Intelligence Agency). This expansion goes up to harmonic
degree 360. Its standard presentation is10
10
Note that here is used a, the equatorial radius of the Earth’s reference ellipsoid,
( )
not R, and φ, the geocentric latitude. The co-ordinates r, φ, λ form a spherical
Ó »îá .
Often used spherical-harmonic expansions 55
360 ( ) ∑n
( )
GM⊕ ∑ a n ( )[ ]
V= 1+ P nm sin φ C nm cos mλ + S nm sin mλ . (2.25)
r r
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 paren-
theses which represents the point mass in the origin equal in magnitude
täysin to the total mass of the Earth, and the “fully normalized” coefficients C
normalisoidut and S — has been already for some time been an industry standard in the
kertoimet global research community in the field of computing spherical-harmonic
expansions as models of the Earth’s gravitational field. A pioneer has
been Professor Richard H. Rapp at Ohio State University, which is why
the models are often called OSU models.
Generally in these models the lower terms — 2 ≤ n ≤ 20 — are derived
ratahäiriöt 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 exposited earlier.
The higher coefficients again — 20 < n ≤ 360 — were before the year
2000 mostly the result of the analysis of gravimetric data (over land) and
satellite radar altimetric data (over the ocean). After the launches of
the gravimetric satellite missions CHAMP, GRACE and GOCE, and as a
result of their measurements, nowadays at least degree number interval
20 < n ≤ 200 is the product of space geodesy. Only the still higher degree
numbers — the new model EGM2008 (Pavlis et al., 2008, 2012) goes up
to degree 2159 — are still due to terrestrial data.
In table 2.6 we give the first and last coefficients of the EGM96 model,
the last and best model from the time just before the satellite gravity
missions. The values tabulated are n, m, C nm , S nm and the mean errors
(standard deviations) of both coefficients from their computation. Note
that all S n0 vanish!
Sometimes also non-normalized coefficients are used, and we write
n
( ∞ ( ) ∑
)
GM⊕ ∑ a n ( )
V= 1− P nm sin φ ( Jnm cos mλ + K nm sin mλ) . (2.26)
r r
n=2 m=0
def
Then we use the notation Jn = Jn0 , and J2 is the most important, the
parameter of the Earth’s gravity field describing the flattening of the
Earth. The relationship with the parameters C, S is
p
{ } { }
Jn0 C n0
= − 2n + 1 ,
K n0 S n0
co-ordinate system.
Ó »îá .
D Tableau 2.6. Coefficients and mean errors of the EGM96 spherical-harmonic

expansion.
n m C nm S nm C nm mean error S nm 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
{ } √ { }
Jnm ( n − m)! C nm
=− 2 (2 n + 1) , m ̸= 0.
K nm ( n + m)! S nm
D 2.14 Ellipsoidal harmonics

The Laplace differential equation 1.12 may be written, and solved, in-
stead of in spherical co-ordinates, in ellipsoidal co-ordinates. The result
Ó »îá .
Ellipsoidal harmonics 57
is known as an ellipsoidal-harmonic expansion11 . They are little used, be-

cause the math needed is more complicated. Also ellipsoidal co-ordinates
are mostly only theoretically interesting and not in any broad use within
geodesy.
The form of presentation is
n
∞ ∑
Q nm i Eu
( )
∑
( b ) P nm (sin β) aenm cos mλ + benm sin mλ ,
( ) ( )
V u, β, λ = (2.27)
n=0 m=0
Q nm i E
in which Q nm (z) are the so-called Legendre functions of the second kind,
sampled in table 2.7. Though the general argument z is complex, equation
2.27 gives a real result for real-valued coefficients aenm , benm .
Those interested in the derivation of the above equation can find it in
Heiskanen and Moritz (1967) or other textbooks on potential theory.
Heiskanen and Moritz give a slightly different form to the equation, the
auxiliary equations needed for the normalization used here can be found
on their pages 66 – 67.
D 2.14.1 The scaling to standard form of the expansion
Assume ae10 = ae11 = be11 = 0, i.e., the vanishing of the dipole moment12 .
We can also show that in the expansion 2.27 the first coefficient is
e GM⊕ E
a 00 = arctan
E b
and the expansion specialized for a rotationally symmetric field becomes
∞ ∞
Q n i Eu e
( )
( ) ∑ ( ) ∑ ( )
V u, β = Vn u, β = ( b ) a n0 P n sin β ,
n=0 n=0
Qn i E
yields now
Q 0 i Eu GM⊕
( )
E
V0 ( u) = ( b) arctan ,
Q0 i E E b
the gravitational potential of the field constituent of ellipsoidal degree
zero.
With the substitutions (Heiskanen and Moritz, 1967, page 66)
( u) ( )
E b E
Q0 i = − i arctan , Q 0 i = − i arctan (2.28)
E u E b
we obtain
GM⊕ E
V0 ( u) = arctan .
E u
11
This expansion for the ellipsoid of revolution differs from the expansion into
Lamé functions found for the triaxial ellipsoid.
12
Strictly speaking this works only in case of a spherical-harmonic expansion, or
in the limit E → 0.
Ó »îá .
This corresponds to the ”central field” of a spherical harmonic expansion

GM
r .
Using this, we may “scale” equation 2.27 as follows — by substituting the

above identities 2.28 and moving the constant expression
( )−1
GM⊕ E
arctan
E b
e e
into the new, standard scaled coefficients C nm , S nm :
( ) GM⊕ E
V u, β, λ = arctan ·
[ E∞ n u
( e )]
∑ ∑ arctan E Q nm i u
( )
b C nm cos m λ+
( Eb ) P nm sin β
( )
· 1+ e , (2.29)
arctan E
u Q nm i + S nm sin m λ
n=2 m=0 E
an ellipsoidal-harmonic expansion that agrees with the spherical-har-

monic expansion 2.25, with the total mass of the Earth outside brackets
and the coefficients dimensionless. This equation has however apparently
not been used for any geopotential determination.
D 2.14.2 Equivalence of the Rapp formula and the ellipsoidal

formula
We can demonstrate the equivalence of spherical equations 2.25, or 2.26,

and the ellipsoidal equation 2.27, if the flattening of the Earth → 0, and
thus also a, b → a, β → φ, and u → r . We assume that Heiskanen and
Moritz (1967) equation 1 – 112,
Q nm i Eu
( ) ( a )n+1
lim ( b
)=
E →0 Q nm iE r
is valid. Substitution into equation 2.27 yields

( ) ( )
V u, β, λ = V r, φ, λ =
n ( )
∞ ∑
∑ a n+1
aenm cos mλ + benm sin mλ ,
( )( )
= P nm sin φ
r
n=0 m=0
which, with the identifications ae00 = GM e e e

a , a 10 = a 11 = b 10 = 0 and
√
GM⊕ GM⊕ ( n + m)!
aenm =− Jnm = C nm ,
a a 2 (2 n + 1) ( n − m)!
√
GM⊕ GM⊕ ( n + m)!
benm =− K nm = S nm
a a 2 (2 n + 1) ( n − m)!
otherwise, corresponds to equations 2.25 and 2.26 for spherical harmon-

ics.
Ó »îá .
Self-test questions 59
D Table 2.7. Legendre functions of the second kind.
1 z+1
Q 0 (z) = ln (n + 1)Q n+1 (z) − (2n + 1) zQ n (z)+
2 z−1
z z+1 + nQ n−1 (z) = 0
Q 1 (z) = ln −1
2 z−1
3z2 − 1 z + 1 3z
Q 2 (z) = ln − )m/2 d m
Q mn (z) = 1 − z2
(
4 z−1 2 Q n (z)
3
5z − 3x z + 1 5z2 2 dzm
Q 3 (z) = ln − +
4 z−1 2 3
D 2.14.3 Advantages of using an ellipsoidal harmonic expansion:

1. The formula for the normal gravitational potential is in this form
of presentation simple, 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 3 to 4, i.e., an expansion up to J6 or J8 , will suffice.
2. The convergence will be more rapid, as less terms are needed. This
is because, due to the Earth’s flattening, the equator is some 23 km
farther from the Earth’s centre than the poles. Therefore, especially
high degree spherical harmonics will have difficulty converging
efficiently both at the poles and in the equatorial region. This
problem is worst for very high degree expansions (e.g., Wenzel,
1998). Already for a degree number of 360, the semi-wavelength of
a spherical harmonic will be only 55 km!
D 2.14.4 Disadvantage of using an ellipsoidal harmonic

expansion:
Evaluation of an ellipsoidal-harmonic expansion is clearly more laborious,

i.e., expensive, than a spherical-harmonic one, in terms of computer
resources.
1. How does separation of variables work?

2. Why does solving the Laplace equation require boundary condi-
tions?
3. What are the harmonic degree and harmonic order in a spherical-
harmonic expansion? How do they relate to the resolution of the
expansion on the Earth’s surface?
4. What types of spherical harmonics are there? Describe their depen-
dence on latitude and longitude.
Ó »îá .
5. 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.
6. How does the attenuation of spherical harmonics with height be-
have? Why does a gravimetric satellite that tries to map the gravity
field of the Earth at a high resolution, fly in as low an orbit as
possible?
7. What does the degree constituent equation describe?
8. Which spherical-harmonic coefficients are associated with the di-
pole moment of the Earth’s mass distribution? Why are they missing
from table 2.6?
D Exercise 2 – 1: Attenuation with height of a spherical-harmonic expansion
If
n
∞ ( ) n+1 ∑
( ) ∑ R ( )
V φ, λ, r = P nm sin φ (a nm sin mλ + b nm cos mλ) =
r
n=0 m=0
∞ ( )n+1
∑ R ( )
= Vn φ, λ, R , (2.30)
r
n=0
we may call ( ) ( )n+1

Vn φ, λ, r R
( )=
Vn φ, λ, R r
the attenuation factor of the potential with height.
Differentiation with respect to r yields
( ) ( )n+2
∂Vn φ, λ, r n+1 R ( )
=− Vn φ, λ, R , (2.31)
∂r R r
or, because, on the Earth’s surface, similarly
( )⏐
∂Vn φ, λ, r ⏐ n+1 ( )
⏐ =− Vn φ, λ, R , (2.32)
∂r ⏐
r =R R
it follows that the attenuation factor for the attraction is the ratio of
equations 2.31 and 2.32:
( )n+2
R
.
r
1. Draw a log-linear graph of both the potential’s and the attraction’s
attenuation factors for values n = 0, . . . , 100, by hand or by machine.
Choose R = 6378 km, r = 7378 km — i.e., the height is 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 ∂∂Vrn caused by the
attraction on satellite level be less than 1% of what they are on the
Earth’s surface?
Ó »îá .
Exercise 2 – 2: Symmetries of spherical harmonics 61
3. For what n value will they be less than 10−4 × of what they are on
the Earth’s surface?
D Exercise 2 – 2: Symmetries of spherical harmonics

( )
See equation 2.30 above. In it, P nm sin φ = P nm ( t) , t = sin φ is only a
function of latitude φ. When φ runs from the South pole through the
equator to the North pole, values −90◦ . . . 0◦ . . . + 90◦ , parameter t will run
through values −1 . . . 0 . . . + 1 on the interval [−1, 1].
For these Legendre functions exists closed expression:
( 2
)m/2 d m
P nm ( t) = 1 − t P n ( t) , (2.12)
dt m
in which the P n ( t) are the ordinary Legendre polynomials:
1 d n [( 2 )n ]
P n ( t) = n n
t −1 .
2 n! dt
1. Note first, that
(a) Differentiating a symmetric function of t will produce an anti-
symmetric function, and vice versa.
(b) The function t2 − 1 and its powers are symmetric.
( )
(c) Thus: for even n values P n ( t) = +P n (− t) — , i.e., P n is symmet-

ric between the Northern and Southern hemispheres — and
for odd n values P n ( t) = −P n (− t), i.e., P n is anti-symmetric
between hemispheres.
( ) ( ( ))
(d) Similarly, for even n, P n sin φ = +P n sin −φ , and for odd
( ) ( ( ))
n, P n sin φ = −P n sin −φ .
Question: what is the corresponding rule for the functions P nm ,
i.e., for which ( n, m) values is it symmetric and for which
values anti-symmetric?
(a) Fill out the diagram ( n = 0 . . . 5, m = 0 . . . n) with symbols
either ‘S’ or ‘A’ in each framed cell:
n= 0 1 2 3 4 5
m=0
1
2
3
4
5 ×
(b) What is the logic of symmetry?

2. If the field is symmetric between Northern and Southern hemi-
( ) ( )
spheres, i.e., V φ, λ, r = V −φ, λ, r , which of the spherical-har-
monic coefficients a nm , b nm drop out of the series expansion? Why?
Ó »îá .
( )
(Hint: see the example formulas and graphs for P nm sin φ in the
lecture notes and try to guess a general rule. Then, verify.)
3. The same question if the potential is rotationally symmetric about
( ) ( )
the Earth rotation axis, i.e., V φ, λ, r = V φ, r .
D Exercise 2 – 3: Escape velocity
1. Given a spherically symmetric planet, mass GM , radius R , from

the surface of which a cannon shoots projectiles at a flight velocity
v. What is the minimum value for v — the so-called escape velocity
— if it is desired that the projectile can travel to arbitrarily large
distances from the planet, and never falls back? The kinetic energy
of the projectile is E = 21 mv2 , in which m is the projectile’s mass.
2. Given, in two-dimensional geometry, a circularly symmetric planet,
mass GM , radius R . The gravitational field of the planet is de-
scribed by a potential V as described in section 2.3. What does V
look like, based on these assumptions?
3. There is again a cannon on the edge of the circle planet. What can
you now say about the escape velocity v?
Ó »îá .
D The normal gravity field
3
D 3.1 The basic idea of a normal field

Like the figure of the Earth can be approximated by an ellipsoid of revolu-
tion, also the gravity field of the Earth can be just as well approximated
by a field of which one equipotential surface or level surface is precisely
this ellipsoid of revolution, the reference ellipsoid.
This brings a logical idea to mind: why not define intercompatibly a refer-
ence ellipsoid, a geopotential or normal potential — one of the equipoten-
tial surfaces of which is the reference ellipsoid — and a gravity formula,
computed by taking the gradient of this normal potential?
After this we may define anomalous quantities, like disturbing potential
and gravity anomaly, which then again will be intercompatible, while
being numerically much smaller.
Ellipsoidal
Normal plumblines
gravity or normals
−
→
γ
−
→
γ
−
→
γ Equipotential surfaces of
the normal gravity field
Reference ellipsoid
(flattening exaggerated)
D Figure 3.1. The normal gravity field of the Earth.
– 63 –
64 The normal gravity field
Let the normal potential be U ( x, y, z) and the normal potential value at

the reference ellipsoid U0 . Then normal gravity will be
∂U
γ ( x, y, z) = − ,
∂n
in which ∂∂n denotes differentiation in the direction of the exterior surface

normal n to the ellipsoid. This normal will differ from the direction of the
normal to the level surfaces, or plumb line, by precisely the plumb-line luotiviiva,
deflection, typically a very small angle. luotiviivan
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 de-
scribed by a rotational potential Φ — also called centrifugal potential.
Also the normal potential U is defined in such a way, that the rotational
potential Φ is included in it: the normal potential is the reference po-
tential of the gravity field, not the gravitational field. If we denote the
normal gravitational potential by Ψ — a quantity rarely used in geodesy
—, then the normal gravity potential (“normal potential”) U is
U = Ψ + Φ,
in which Φ is the centrifugal potential. In other words: Ψ, like V , is

defined in a non-rotating (inertial) system, whereas U, like W, is defined
in a system that co-rotates with the Earth and is non-inertial. Just like
also the word gravity refers to a force acting in a co-rotating system, when painovoima
in an inertial system we use the word gravitation.
D 3.2 The centrifugal force and its potential

The rotation of the Earth affects the gravity field. In an inertial reference
system one speaks of gravitation and gravitational potential V , on the
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 3.2.
To derive the equation for centrifugal force, write first
p = X i + Y j,
when the vectors {i, j, k} form an orthonormal basis along the ( X , Y , Z )

axes. Then √ √
p = ∥p∥ = 〈p · p〉 = X 2 + Y 2 .
Now the centrifugal force — or rather, acceleration — is, in metres per

second squared,
f = ω2⊕ p = ω2⊕ ( X i + Y j) ,
Ó »îá .
The centrifugal force and its potential 65
p
Centrifugal force
k
Gravitation
Gravity
i j
Y
X
D Figure 3.2. Gravitation and centrifugal force.
with ω⊕ the rotation rate in radians per second.

Here on Earth, gravity measurements are generally done with an instru-
ment 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 Corio-
lis1 force. Also fluids — water, air — on the Earth’s surface, if they are
at rest, sense only the centrifugal force. Currents also sense the Coriolis
force, which deflects them sideways and causes the well known eddy
phenomena in the oceans and atmosphere.
If we forget for the moment about the Coriolis force, we may describe
the centrifugal force as the gradient of a potential. If we write for this
centrifugal potential
1
Φ = ω2⊕ ( X 2 + Y 2 ),
2
we may directly calculate the gradient
−
→ ∂Φ ∂Φ ∂Φ 1 1
f= ∇Φ= i+ j+ k = ω2⊕ · 2 X · i + ω2⊕ · 2Y · j + 0 = ω2⊕ ( X i + Y j) ,
∂X ∂Y ∂Z 2 2
which corresponds to the above centrifugal force equation.
If we start out with the gravitational potential V and add to it the
painovoima- centrifugal potential Φ, we obtain the gravity potential W :
potentiaali
W = V + Φ.
We may also derive from the centrifugal potential Φ the following equa-
tion by differentiating it twice:
∂ ∂
∆Φ = ∇2 Φ = ∇f = ω2⊕ X + ω2⊕ Y + 0 = 2ω2⊕ , (3.1)
∂X ∂Y
1
Gaspard-Gustave Coriolis (1792 – 1843) was a French mathematician, physicist
and mechanical engineer. His name is inscribed on the Eiffel Tower.
Ó »îá .
from which follows, with the Poisson equation 1.14,
∆W = −4πG ρ + 2ω2⊕ , (3.2)
the Poisson equation for the gravity potential.

The difference between gravitation and gravity is essential. The force, or
−
→
acceleration, of gravitation a = ∇ 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. Especially we may
mention the clever tests by the Hungarian baron Loránd Eötvös2 .
Water masses on the Earth’s surface, like 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 be-
tween attraction and centrifugal force. For this reason the sea surface
coincides within a metre or so with an equipotential or level surface of the
W function or geopotential. Also on dry land we measure heights from
this surface, the geoid (Gauss: “mathematical figure of the Earth”).
D 3.3 Level surfaces and plumb lines

Surfaces of the same gravity potential or geopotential, equipotential
surfaces or level surfaces are the following surfaces: tasopinnat
W ( x, y, z) = W0 = const.
Let {i, j, k} again be an orthonormal basis along the ( x, y, z) axes. Then, in

the direction of the unit vector
e = e1i + e2j + e3k
the potential changes as follows:

∂W ∂W ∂W ∂W
= e1 + e2 + e3 ,
∂e ∂x ∂y ∂z
which vanishes if and only if
⟨ −
→ ⟩
e · ∇ W = 0,
in other words, the potential is stationary only in directions that are

perpendicular to the Earth’s gravity vector
−
→
∇ W = g.
2
Loránd baron Eötvös de Vásárosnamény (1848 – 1919) was a Hungarian physi-
cist and student of gravitation.
Ó »îá .
Level surfaces and plumb lines 67
tangent plane
P
W = WP − δW
x0 W = WP
ρ1
D Figure 3.3. The curvature of level surfaces.

Level surfaces and gravity vectors, or plumb lines, are always
perpendicular to each other.

D 3.3.1 Curvature of level surfaces
Given in point P a plane that in P has the same direction as the level
surface, i.e., its tangent plane. If the local curvature of the level surface
in the x direction is ρ 1 , 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 .
2ρ 1
From this we obtain the difference in W values between the surfaces

def
( g = ∥g∥):
g
δW = −ϵ g ≈ − ( x − x0 )2 .
2ρ 1
By differentiating (note that W here is now the geopotential on the tangent
plane) we obtain
∂2 ∂2 g
δW = W = Wxx = −
∂x 2 ∂x 2 ρ1
from which
g
ρ1 = − ,
Wxx
Ó »îá .
where the W that is being differentiated with respect to the x co-ordinate

is its restriction to the tangent or horizontal plane. rajoittuma
By determining the curvature in the x direction
def 1 Wxx
κ1 = =− , (3.3)
ρ1 g
and similarly in the y direction
def 1 Wyy
κ2 = =− , (3.4)
ρ2 g
we obtain the mean or Germain3 curvature, in most locations a positive

number:
1 Wxx + Wyy
J = (κ1 + κ2 ) = − ,
2 2g
and by using the Poisson equation 3.2,
∆W = Wxx + Wyy + Wzz = −4πG ρ + 2ω2⊕ ,
we obtain
−2 gJ + Wzz = −4πG ρ + 2ω2⊕ .
By using
∂g ∂g
Wzz = − =−
∂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
= −2 gJ + 4πG ρ − 2ω2⊕ ,
∂H
an equation found by Ernst Heinrich Bruns.
D 3.4 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 from the reference ellipsoid h, a purely geometric
quantity.
In earlier times one could measure only the direction of the plumb line as
shown in figure 3.4, and the potential difference between an observation
point and sea level. The direction of the plumb line n was measured
3
Marie-Sophie Germain (1776 – 1831) was a brilliant French mathematician,
number theorist and student of elasticity. She corresponded with Gauss, among
others, on number theory, and did foundational work toward a proof of Fermat’s
last theorem. Her name is missing from the Eiffel Tower.
Ó »îá .
Natural co-ordinates 69
Astronomical
co-ordinates Φ, Λ
Plumb line
Greenwich
O Φ
Figure 3.4. Natural co-ordinates Φ, Λ. Additionally, a natural height co-ordi-

D nate, e.g., the geopotential W, is needed.
astronomically: astronomical latitude Φ (don’t confuse with the centrifu-

gal potential) and astronomical longitude Λ. The third co-ordinate, the
gravity potential difference W ( x, y, z) − W0 from that of sea level, was
determined by levelling. The co-ordinates Φ, Λ and W are called natural
co-ordinates.
Often, instead of the potential, orthometric height is used. Its definition
is easy to understand if one writes
ˆ WP
∂W ∂W 1 1
= = − g =⇒ dH = − dW =⇒ H P = − dW ′ ,
∂n ∂H g W0
′
g(W )
where the integral is taken along the plumb line of point P . ∂∂H = ∂∂n is
the derivative in the direction of the plumb line, i.e., 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 rock density. This
trickiness of their determination is a problem specific to orthometric
heights. We will return to this later on (Heiskanen and Moritz, 1967
chapter 4).
Also the co-ordinates Φ, Λ and H form a natural co-ordinate system.
Ó »îá .
D 3.5 The normal potential in ellipsoidal co-ordinates

We already presented an equation 2.27 for the expansion of the geopoten-
tial into ellipsoidal harmonics. It is demanded of the normal potential
U , that it is a constant on the reference ellipsoid u = b. We expand the
centrifugal potential Φ into ellipsoidal harmonics. We have
) 1 ( ) 1 (
Φ u, β = ω2⊕ x2 + y2 = ω2⊕ u2 + E 2 cos2 β =
( )
2 2
1 2( 2 2
1 − sin2 β =
)( )
= ω⊕ u + E
2
1 2( 2 2
)( 2 ( ) 2 ( ))
= ω⊕ u + E − P2 sin β + P0 sin β =
2 3 3
1 2( 2 2
)( ( ) ( ))
= − ω⊕ u + E P2 sin β − P0 sin β .
3
Additionally we have, based on equation 2.27, for a rotationally symmetric
gravitational potential Ψ:
∞ ∞
Q n i Eu
( )
∑ ∑
Ψ u, β = Ψn
( ) ( ) ( )
u, β = ( b ) A n P n sin β . (3.5)
n=0 n=0
Qn i E
Now
U u, β = Ψ u, β + Φ u, β .
( ) ( ) ( )
( )
On the reference ellipsoid u = b we have as a requirement U b, β = U0 ,
which is possible only if
1
U0 = A 0 + ω2⊕ b2 + E 2 ,
( )
3
0 = A1,
1
0 = A 2 − ω2⊕ b2 + E 2 ,
( )
3
0 = A 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, b are known.
The result, given in Heiskanen and Moritz (1967) as equation 2 – 61, is
GM⊕ E 1
U0 = arctan + ω2⊕ a2 .
E b 3
From this follows, using a2 = b2 + E 2 :
1 GM⊕ E
A 0 = U0 − ω2⊕ a2 = arctan .
3 E b
The normal gravity potential U is obtained as follows (remember the
Ó »îá .
Normal gravity on the reference ellipsoid 71
identities 2.28 and that a2 = b2 + E 2 ):
U ( u, β) = Ψ u, β + Φ u, β =
( ) ( )
Q 2 i Eu ( 3
( )
GM⊕ E 1 1)
= arctan + ω2⊕ a2 ( b) sin2 β − +
 E  u 3   Q 2 i E  2  2 
Ψ0 (u) A2 P2 (sin β)
1
+ ω2⊕ u2 + E 2 cos2 β =
( )
2  
Φ( u,β)
= C 1 ( u) sin2 β + C 2 ( u) cos2 β,
in which C 1 , C 2 are suitable functions of u. The function Ψ0 is the term

for n = 0 in expansion 3.5.
On the surface of the reference ellipsoid ( u = b):
( )
GM⊕ E 1 1 1
U ( b, β) = arctan − ω2⊕ a2 + ω2⊕ a2 sin2 β + ω2⊕ a2 cos2 β =
E b 6 2 2
GM⊕ E 1 2 2
= arctan + ω⊕ a ,
E b 3
a constant, as it better be!
D 3.6 Normal gravity on the reference ellipsoid

Without proof we mention that for normal gravity (the quantity γ = ∂∂Uh )
the following equation applies on the reference ellipsoid:
( ) aγb sin2 β + bγa cos2 β

γ β =√ .
a2 sin2 β + b2 cos2 β
By substitution we find immediately, that γa is normal gravity on the

equator (β = 0) and γb normal gravity on the poles (β = ±90◦ ).
Equations 2.4 and 2.6 yield
Z
sin β b a Z a
tan β = = p = p = tan φ
cos β X 2 +Y 2 b X2 +Y2 b
a
and
Z
sin ϕ (1− e2 ) N 1 Z a2
tan ϕ = = p 2 2 = p = tan φ,
cos ϕ X +Y 1 − e2 X 2 + Y 2 b2
N
in which φ is the geocentric latitude, see equation 2.3. From this follows
directly
b
tan β = tan ϕ,
a
Ó »îá .
.P
b
β
φ ϕ . X /Y
a O
D Figure 3.5. The geometry of the meridian ellipse and various types of latitude.
in which the latitude angle ϕ is the geodetic or geographic latitude. (β

is still the so-called reduced latitude). Now it is easy to show (exercise!)
that
( ) aγa cos2 ϕ + bγb sin2 ϕ
γ ϕ =√ . (3.6)
a2 cos2 ϕ + b2 sin2 ϕ
This is the famous Somigliana–Pizzetti4 equation. These geodesists de-
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 also in geographical co-ordinates the
gravity formula is a closed expression in latitude.
D 3.7 Numerical values and formulas

When the reference ellipsoid has been chosen, we may calculate the nor-
mal potential and normal gravity corresponding to it. The fundamental
quantities are
a the equatorial radius of the ellipsoid of revolution, i.e., its semi-major

axis
a− b
f the flattening, f = a , in which b is the polar radius or semi-minor
axis
4
Carlo Somigliana (1860 – 1955) was an Italian mathematician and physicist.
Paolo Pizzetti (1860 – 1918) was an Italian geodesist.
Ó »îá .
Numerical values and formulas 73
ω⊕ the rotation rate of the Earth

GM the total mass of the Earth (including the atmosphere).
Alternatively one may choose also γa , i.e., equatorial gravity.

Nowadays the most commonly used reference ellipsoid cum normal po-
tential is GRS80, the Geodetic Reference System 1980:
a = 6,378,137 m,
1
= 298.257,222,101,
f
ω⊕ = 7,292,115 · 10−11 s−1 ,
GM = 3,986,005 · 108 m3 s−2 .
In reality f is not a defining constant of GRS80, but instead is used J2 ,

which is a defining quantity for the gravity field, see equation 2.26.
WGS84 (World Geodetic System 1984) used by the GPS system has a
reference ellipsoid that is almost identical to that of GRS80.
The normal potential is (Heikkinen, 1981), units, in the SI system, are

metre [m] and second [s]:
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 −
[ ]
+ h2 −
− 0.000,000,10 · 10−4 sin4 ϕ
− −0.000,024,22 · 10−8 + 0.000,000,07 · 10−8 sin2 ϕ h3 ,
[ ]
and normal gravity (note the minus sign, U is positive and diminishes
going upward):
∂U
γ= − =
∂h
= + 9.780,326,77 + 0.051,630,75 sin2 ϕ +
+ 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−
− 0.000,000,200 · 10−4 sin4 ϕ
− −0.000,072,65 · 10−8 + 0.000,000,21 · 10−8 sin2 ϕ h2 .
[ ]
(3.7)
Here, the unit of potential is m2 /s2 , and the unit of gravity, m /s2 . More
precise equations can be found from Heikkinen (1981). In these equa-
tions, the coefficient 9.780,32 . . . m /s2 is equatorial gravity, and the value
0.030,87 . . . s−2 is the vertical gradient of gravity on the equator. ϕ is
geodetic latitude, h (in metres) is the height above the reference ellipsoid.
Ó »îá .
Other gravity formulas and reference ellipsoids still in legacy use (and
slowly vanishing) are Helmert’s 1906 ellipsoid, the Krasovsky ellipsoid or
SK-42 in the countries of Eastern Europe, the International or Hayford
ellipsoid (1924) and its gravity formula, and Geodetic Reference System
1967.
D 3.7.1 Numerical example
According to the above equation, the normal potential over the equator is
U = 62,636,860.8500 − 9.780,326,77 h + 0.015,438,99 · 10−4 h2 +

+ 0.000,024,22 · 10−8 h3 .
◦ Draw this function for values of h in the range 0 − 7000 km.

◦ Draw for comparison the quadratic version, from which the last
term is left off.
Questions:
1. What are the minima of these functions?

2. How physically realistic is this?
Answers:
1. See figure 3.6. The minima are at heights 3000 km and 2000 km,
approximately.
2. Not very physical: the stationary point for potential U (the normal
potential in a co-rotating reference system) should be located at
approx. 36,000 km height, at the geostationary orbit.
This tells us that polynomial approximation cannot be extrapolated
very far. In this case the interval of extrapolation is of the same
order as the radius of the Earth, and that won’t work any more.
D 3.8 The normal potential as a spherical harmonic expansion

The spherical-harmonic expansion of an ellipsoidal gravitational field
contains, besides the second degree harmonic, also higher degree harmon-
ics. If we write, as is customary, the potential outside the Earth in the
following form (Heiskanen and Moritz, 1967 section 2 – 39, also equation
2.26):
n
{ ∞ ( ) ∑
}
GM⊕ ∑ a n
V= 1− P nm (sin φ) [ Jnm cos mλ + K nm sin mλ] ,
r r
n=2 m=0
then we may also write the normal gravitational potential, Ψ, into the
form [ ]
∞ ( a )2n
GM⊕ ∑
Ψ=
( )
1− J2n P2n sin φ ,
r r
n=1
Ó »îá .
The normal potential as a spherical harmonic expansion 75
.
140,000,000
Cubic
120,000,000 Quadratic
Linear
100,000,000 Realistic
80,000,000
60,000,000
40,000,000
20,000,000
0 1000 2000 3000 4000 5000 6000 7000
Figure 3.6. The normal field’s potential curve over the equator. Heights in
D kilometres.
which contains only even coefficients J2n = J2n,0 as the normal field is
symmetric about the equatorial plane.
The coefficients for the GRS80 normal gravitational potential are
found5 in table 3.1. Higher terms are usually not needed. The
relationship between fully normalized and non-normalized coefficients is
p
J n = J n 2 n + 1.
Note for comparison that in the expansion of the same field into ellipsoidal
harmonics only the degree zero and degree two coefficients are non-zero!
5
They can also be calculated from equation (2 – 97) given in Heiskanen and
Moritz (1967) :
( )n
3 e2
( )
n+1 J2
J2n = (−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.
D Table 3.1. GRS80 normal potential spherical-harmonic coefficients (Heikkinen,

1981; Heiskanen and Moritz, 1967).
Non-normalized Fully normalized
J2 = J2,0 = 1082.63 · 10−6 J 2 = 484.166,854,90 · 10−6

J4 = J4,0 = −2.370,912,22 · 10−6 J 4 = −0.790,304,073 · 10−6
J6 = J6,0 = +0.006,083,47 · 10−6 J 6 = +0.001,687,251,00 · 10−6
J8 = J8,0 = −0.000,014,27 · 10−6 J 8 = −0.000,003,46 · 10−6
Ó »îá .
This is the main reason why these functions are used at all.
Instead of using an ellipsoidal model, we may use as a normal gravity
potential formula also the first two, three terms of the spherical-harmonic
expansion of the real geopotential. Then we obtain, taking the centrifugal
potential along:
Y0 Y2 (φ, λ) 1 2 2
U= + + ω⊕ ( X + Y 2 ),
r r3 2
with the corresponding equipotential surface being the “Bruns spheroid”,
or
Y0 Y2 (φ, λ) Y4 (φ, λ) 1 2 2
U= + + + ω⊕ ( X + Y 2 ),
r r3 r5 2
def ( ) ( )
the “Helmert spheroid”. Here, Y0 = GM and Y2 φ, λ , 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. The are, in fact, quite complicated surfaces (Heiskanen and
Moritz, 1967, 2 – 12)!
However, in geometric geodesy we always use a reference ellipsoid, so
this is also a wise thing to do in physical geodesy.
D 3.9 The disturbing potential

Write the gravity potential
W = V + Φ,
in which Φ is the centrifugal potential (see above), and the normal poten-
tial
U = Ψ + Φ.
The difference between them is
def
T = W − U = V − Ψ,
the disturbing potential. häiriöpotentiaali
Both V and Ψ can be expanded into spherical harmonics. If we write the

gravity potential
W = V +Φ =
n
{ ∞ ( ) ∑
}
GM⊕ ∑ a n
= Φ+
( )
1− P nm sin φ [ Jnm cos mλ + K nm sin mλ] ,
r r
n=2 m=0
and the normal potential

{ ∞
}
GM⊕ ∑ ( a )n
U = Φ+ 1− Jn∗ P n (sin φ) ,
r r
n=2,even
Ó »îá .
The disturbing potential 77
we obtain by subtraction for the disturbing potential
T = W −U =
n
{∞ }
GM⊕ ∑ ( a )n ∑ ( )
=− P nm sin φ [δ Jnm cos mλ + δK nm sin mλ] ,
r r
n=2 m=0
in which {
δ Jn0 = Jn0 − Jn∗ if n even,
δ Jnm = Jnm if n odd,
and δK nm = K nm .
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 = T n φ, λ , (3.8)
r
n=2
where, in every term, the degree constituent T n has the same dimension
as T , and
n
( ) GM⊕ ∑ ( )
Tn φ, λ = − P nm sin φ [δ Jnm cos mλ + δK nm sin mλ] .
a
m=0
On the surface of the reference sphere of radius a6 :

∞
∑
( ) ( )
T φ, λ = T n φ, λ , (3.9)
n=2
( )
from which we see, that on the reference level, the terms T n φ, λ are
really the degree constituents for a certain degree number n .
( )
The above expansion is 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, the value being proportional to
the cosine of the angular distance from the point on the Earth’s surface
pointed to by the dipole vector. Both vanish because it is assumed that
1. the total mass of the Earth GM⊕ assumed by the normal field is
realistic, and
2. the origin of the co-ordinate reference system is assumed to be at
the centre of mass of the Earth.
See section 2.12 for a discussion.

The higher T n are oscillating functions of ever shorter wavelengths.
6
Earlier on we have used for this reference radius (in spherical approximation)
also the symbol R.
Ó »îá .
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
1
Φ = ω 2 ( X 2 + Y 2 ),
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.
4. Explain the idea of natural co-ordinates.
5. What relationhip was there between Mr. Leblanc and C.F. Gauss?
Use Google.
6. What makes the Somigliana–Pizzetti equation (3.6) valuable?
7. What are the defining parameters of the Geodetic Reference System
1980?
8. Why does the spherical-harmonic expansion of the normal potential
contain only a small number of terms and coefficients?
9. Why doesn’t the spherical-harmonic expansion of the normal poten-
tial contain any terms with order m ̸= 0?
10. Why does the spherical-harmonic expansion of the normal potential
contain only terms with even degree numbers n?
D Exercise 3 – 1: The Somigliana–Pizetti equation
1. Given gravity on the equator γa and on the poles γb . What is gravity

on geodetic latitude ϕ = 45◦ ?
2. And what is gravity on the reduced latitude β = 45◦ ?
3. Given the semi-major axis a and semi-minor axis b, what are the
differences between the different latitudes (geodetic ϕ, geocentric
φ, and reduced β) at most? You may assume that the maximum
happens at latitudes ±45◦ .
4. Compute for the above quantities numerical values for the case of
the GRS80 reference ellipsoid.
D Exercise 3 – 2: The centrifugal force
Given is the rotation rate of the Earth, radians per second: ω⊕ = 7292115 ·
10−11 s−1 .
1. Compute (roughly) the centrifugal force caused by the Earth rota-

tion at Southern Finland (ϕ = 60◦ , R = 6378 km, spherical Earth).
In what direction does the force point (sketch!)?
Ó »îá .
Exercise 3 – 2: The centrifugal force 79
2. How much does the centrifugal force contribute to local gravity?

Both as an acceleration and as a percentage.
3. Compute from the ω⊕ value given above, the rotation time of the
Earth in hours. Why is it not precisely 24h [difficult]?
Ó »îá .
D Anomalous quantities of the gravity field
4
D 4.1 Disturbing potential, geoid height, and deflections of the

plumb line
The first anomalous quantity, which we already discussed above, is the

difference between the true gravity potential W and the normal gravity
potential U , the disturbing potential:
def
T = W − U.
All other anomalous quantities are various functions of this, like the
geoid height N and the plumb-line deflections ξ, η. They are generally
obtained by subtracting from each other
1. a natural quantity related to the Earth’s real gravity field, and

2. a corresponding quantity related to the normal gravity field of the
reference ellipsoid of the Earth.
For example, deflections of the plumb line
ξ = Φ − ϕ,
η = (Λ − λ) cos ϕ,
in which (Φ, Λ) are astronomical latitude and longitude, i.e., the direction
( )
of the local plumb line, and the geodetic latitude and longitude ϕ, λ
form correspondingly the direction of the surface normal on the reference
ellipsoid. See figure 4.1.
The geoid height or geoid undulation is
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 plumbline are in Finland a few seconds of arc (′′ ) in
magnitude, geoid undulations range from 15 to 32 m (for comparison,
globally the range of variation is −107 . . . + 85 m), relative to the GRS80
– 81 –
82 Anomalous quantities of the gravity field
Plumb-line
deflection ξ, η Topography
Reference ellipsoid
Geoid Geoid height N
D Figure 4.1. Geoid undulations and deflections of the plumb line.
ellipsoid as is today customary. At sea level, the plumb-line deflections

equal the horizontal gradients of the geoid undulation. See figures 4.1,
4.2.
For any reference ellipsoid, e.g., 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. With
the aid of this field we may calculate for each gravity field quantity the
corresponding normal quantity, and by subtracting the two from each
other we obtain again the corresponding anomalous quantity.
For heights above the reference ellipsoid there exists an analoguous
formula to that for orthometric heights, where U is the normal potential
and γ normal gravity:
ˆ UP
1
hP = − dU.
U0 γ (U )
The geoid height in point P is now

ˆ WP ˆ UP
1 1
NP = h P − H P = dW − dU =
W0 g U0 γ
ˆ WP ˆ WP ˆ UP ˆ U0
1 1 1 1
= dW − dU − dU + dU
W0 g W0 γ WP γ W0 γ
ˆWP ˆ UP ˆ U0
γ− g 1 1
= dW ′ − dU + dU
W0 gγ WP γ W0 γ
ˆHP ˆ UP ˆ U0
γ− g 1 1
= dH − dU + dU, (4.1)
0 γ WP γ W0 γ
by re-naming the integration variables W,U → W ′ and changing it to a

metric one: dW ′ = gdH.
In equation 4.1 the last term vanishes if we assume1 U0 = W0 . Now in the
definition of geoid heights, point P is at mean sea level — the zero point
1
This is not self-evident! In a local vertical datum the potential of the zero point
could well differ by even as much as a metre from the normal potential of a
global reference ellipsoid.
Ó »îá .
Disturbing potential, geoid height. . . 83
Figure 4.2. A geoid model for Finland from 1984. Deflections of the plumb line
D from observations in red (Vermeer, 1984).
of the height system used. It follows that also the first term vanishes (it
would in any case always be small, except in the mountains). So
ˆ UP
1 1 TP T
NP = − dU ≈ (WP − UP ) = or N= . (4.2)
WP γ γP γP γ
where we have substituted T = W − U , the disturbing potential. All

quantities are assumed to be at sea level. This is the famous Bruns2
equation (Heiskanen and Moritz, 1967, equation 2 – 144).
2
Ernst Heinrich Bruns (1848 – 1919) was an eminent German mathematician
and mathematical geodesist.
Ó »îá .
−
→
γ = gradU
g = gradW
P
WP Geoid
UP
N Ellipsoid
UQ (= WP )
Q
Figure 4.3. Equipotential surfaces of the gravity field (W) and the normal
D gravity field (U).
Figure 4.3 depicts the situation still better. In this figure, the gradient
−−→ −−→
vectors g = grad W and −→
γ = grad U have lengths ∂∂W ∂U
H and ∂ H , 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 WP − UP T
N= = = .
γ γ γ
D 4.2 Gravity disturbances

The difference between the true and normal gravity accelerations is called
the gravity disturbance, δ g. An exact equation would be
∂W ∂U
( )
δg = − − ;
∂H ∂h
where differentiation takes place along the plumb line for W , and along
the ellipsoidal normal for U . The directions of plumb line and ellipsoidal
surface normal are actually very close to each other.
In spherical approximation we have
∂W ∂U ∂T
( )
δg = − − =− .
∂r ∂r ∂r
We already expanded the disturbing potential into components for differ-

ent spherical-harmonic degree numbers — equation 3.8 —, and now we
Ó »îá .
Gravity anomalies 85
obtain by differentiating with respect to r 3 :

[∞ ( ) ]
∑ R n+1
( )
( ) ∂T φ, λ, r ∂ ( )
δ g φ, λ, r = − =− T n φ, λ =
∂r ∂r r
n=2
∞ ( )n+1
1∑ R ( )
= ( n + 1) T n φ, λ , (4.3)
r r
n=2
or on the Earth’s surface ( r = R ):

∞
( ) 1∑ ( )
δ g φ, λ, R = ( n + 1) T n φ, λ . (4.4)
R
n=2
This is the spectral representation of the gravity disturbance on the

surface of the Earth — more precisely, on a 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 observe gravity disturbances only if, in addition to measuring the
acceleration of gravity g P = ∂∂W
( ⏐ )
H P in point P , we have a way to measure
⏐
P :s location in space, relative to the geocentre, so one may calculate
normal gravity γP = ∂∂Uh ⏐P in the same point. Nowadays this is even easy
⏐
using GPS, but traditionally it has been impossible. For this reason
gravity disturbances are little used. One rather uses gravity anomalies,
about which more below.
D 4.3 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, not any accurate height h above the reference
ellipsoid. One only had access to the height H above sea level (the geoid),
obtained, e.g., through a national levelling network — or, in the worst
case, barometrically.
This means that, though the true gravity g is measured in point P the
height of which above sea level is H P , normal gravity γ must of necessity
be calculated in another point Q , the height of which above the reference
ellipsoid is hQ = H P . See figure 4.4.
In other words, the measured height of point P above mean sea level
is substituted, brute-force style, into the normal gravity formula, that
3
Note that, if the normal field that defines T has a reference ellipsoid centred
on the Earth’s centre of mass, and the implied total mass GM of the normal
( )
field equals that of the true Earth, then the first two constituents T0 φ, λ =
( ) ∑∞
T1 φ, λ = 0, and the sum may be taken as n=2 . This was assumed here. See
section 2.12.
Ó »îá .
Topography
Telluroid P = observation point
ζ
Mean sea Q
level (geoid)
hQ HP
Ellipsoid
N
Figure 4.4. Reference ellipsoid, mean sea level (geoid), and gravity measure-
D ment.
however expects a height above the reference ellipsoid! This special trait
of the definition of gravity anomalies may be called a“free boundary-value
problem”.
According to this we calculate gravity anomalies as follows:
∆ g P = g P − γQ = g P − γP + γP − γQ =
( ) ( )
∂W ⏐⏐ ∂U ⏐⏐ ∂U ⏐⏐ ∂U ⏐⏐
( ⏐ ⏐ ) ( ⏐ ⏐ )
=− − − −
∂ H ⏐P ∂ h ⏐P ∂ h ⏐P ∂ h ⏐Q
∂ (W − U ) ⏐⏐ ) ∂γ ⏐
⏐ ⏐
(
≈− ⏐ + h P − hQ ∂ H ⏐ =
⏐
∂H P P
∂T ⏐⏐ ∂γ ⏐⏐
⏐ ⏐
=− + (h P − HP ) =
∂ H ⏐P ∂ H ⏐P
∂T ⏐⏐ ∂γ ⏐⏐
⏐ ⏐
=− + NP =
∂ H ⏐P ∂ H ⏐P
∂T T ∂γ ⏐⏐
( )⏐
− + ,
∂H γ ∂H ⏐ P
using almost all equations above.

The equation derived looks familiar: it is the boundary condition of the
third boundary-value problem (Heiskanen and Moritz, 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’s normal gravity field is spherically symmet-
ric, we may approximate (exercise: show this!):
∂T 2
∆g = − − T, (4.5)
∂r r
in which r = R + H is the distance from the Earth’s centre.
Ó »îá .
Units used for gravity anomalies 87
By substituting into this the equation for δ g, and r = R , we obtain on the

Earth’s surface:
2
∆ g = δ g − T. (4.6)
R
From this we obtain directly by using the above spectral representations
3.9, and 4.4 for T and δ g:
∞ ∞
1∑ 1∑
∆g = [( n + 1) − 2] T n = ( n − 1) T n .
R R
n=2 n=2
Choose the following notation, still on the Earth’s surface:

∞
∑
∆g = ∆ gn,
n=2
in which the degree constituents of the gravity anomaly are
def n−1
∆gn = Tn. (4.7)
R
The presence of the factor n − 1 shows that gravity anomalies cannot
contain n = 1 constituents, even if T would contain them. It is always
wise to choose the origin of the co-ordinate system to be in the centre
of mass of the Earth, but if it is not, at least gravity anomalies do not
change.
Equation 4.7 applies only on a spherical Earth of radius R . In the
space outside the Earth we obtain, using equations 4.3 and 4.5, the
corresponding equation
∞ ( )n+1
1∑ R
∆g = [( n + 1) − 2] Tn =
r r
n=2
∞ ( )n+1
1∑ R
= ( n − 1) Tn =
r r
n=2
∞ ( ) n+2
∑ R
= ∆ gn. (4.8)
r
n=2
D 4.4 Units used for gravity anomalies

The most common unit of measurement for gravity variations is the
milligal. The connection with the SI system is 1 mGal = 10−5 m /s². Also
µGal or 10−8 m /s² is used. In modern books are also used µm /s² and nm /s²,
which formally belong to the SI system. 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.
Ó »îá .
D Table 4.1. Orders of magnitude of gravity variations.
Phenomenon Fraction of gravity SI units mGal
Ambient gravity 1 9.8 980,000

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 ±10−8 − 10−7 ±10−7 − 10−6 ±0.01 − 0.1
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. In table 4.1 we
give a few values in order to get an idea of the orders of magnitude of
phenomena. On the Earth’s surface the vertical gradient ∂∂hg is on average
some −0.3 mGal /m = −3000 E.
D 4.5 The boundary-value problem of physical geodesy

As we explained in the above section, gravimetric measurement is more
complicated than just measuring the quantity ∂∂Wr . When we measure the
radial derivative of the geopotential, we do it in a place we don’t precisely
know. Even if we knew the height of the measurement location above sea
level, that still doesn’t give us the measurement point’s location in space.
This location depends additionally on the location in space of sea level,
i.e., the geoid, specifically its height above the reference ellipsoid.
This is how we arrive at the third boundary-value problem4 . The boun-
dary-value problem of physical geodesy is to determine the potential V
outside a body if given on its surface is the linear combination
∂V
aV + b ,
∂n
with a, b suitable constants. The variable n represents here differentia-
tion in the direction of the normal to the Earth surface, in practice the
same as r or h.
In physical geodesy is given, in spherical approximation, the following
linear combination (gravity anomaly, equation 4.6):

∂T 2
∆g = − − T. (4.9)
∂n R

4
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, the Neumann problem the second boundary-value problem.
Ó »îá .
The boundary-value problem of physical geodesy 89
The equation, or boundary condition, 4.9 is called the fundamental equa-

tion of physical geodesy.
Above we already obtained equations 2.17 and 2.20, that apply equally
well to the disturbing potential T as to the general potential V :
∞ ( )n+1
∂T ∑ n+1 R ( )
=− T n φ, λ ,
∂n r r
n=0
∞ ( )n+1
∑ R ( )
T= T n φ, λ .
r
n=0
By combining these we obtain

∞ [ ( )n+1
∑ n+1 2] R
∆g =
( )
− T n φ, λ ,
r R r
n=0
or on the Earth’s surface (R = r ):

∞ ∞
∑ n−1 ∑
∆g = ∆ g n φ, λ ,
( ) ( )
T n φ, λ =
R
n=0 n=0
where the quantities ∆ g n φ, λ = nR−1 T n φ, λ are defined in equation

( ) ( )
4.7. Remember that the functions ∆ g n φ, λ are computable with the

( )
help of the degree constituent equation 2.16 when ∆ g φ, λ is known all

( )
over the Earth.

Observe also that the term n = 1 vanishes: ∆ g 1 = 0. We assume also
∆ g 0 = − TR0 = 0, i.e., the true exterior potential, and thus the total mass
of the Earth GM , and her volume5 , is in global average the same as the
normal potential and its assumed 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 determinations6 .
Thus we obtain the solution also of this boundary-value problem in spec-
tral representation (which is thus valid in the whole exterior space) by
using the degree constituent equation7 2.16:
∞ ( ) n+1
∆ g n φ, λ
( )
( ) ∑ R
T φ, λ, r = R =
r n−1
n=2
∞ ( )n+1 ¨
R ∑ 2n + 1 R
∆ g φ′ , λ′ P n cos ψ d σ′ . (4.10)
( ) ( )
=
4π n−1 r σ
n=2
5
In fact, the atmosphere complicates this matter.
6
Note, however, that GRS80 has an equatorial radius of 6,378,137.0 m, while
the newer models like EGM2008 give a smaller value of 6,378,136.3 m as the
location of global mean sea level. Uncertainty continues to be in the decimetres.
) def
7
We write ∆ g φ′ , λ′ = ∆ g φ′ , λ′ , R , the integral being evaluated on the Earth
( ( )
sphere r = R.
Ó »îá .
This is precisely the boundary-value problem that is created if everywhere

on the Earth, land and sea, surface gravity anomalies are given.
The integral equation corresponding to the above spectral equation 4.10
is known as the Stokes8 equation:
¨
R
S ψ, r ∆ g φ′ , λ′ d σ′ ,
( ) ( )
T (φ, λ, r ) =
4π σ
in which the Stokes kernel is

∞ ( )n+1
( ) ∑ 2n + 1 R ( )
S ψ, r = P n cos ψ . (4.11)
n−1 r
n=2
In section 7.1 we will give a closed formula for this function (for the case
r = R ) and a graph.
D 4.6 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 refer-
ence ellipsoid h, we perform, as it were, a mapping, which 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 the telluroid mapping. The telluroid is the surface
that follows the shapes of Earth’s topography, but is everywhere below
the topography by an amount ζ if positive, or above it by an amount −ζ
otherwise. The quantity ζ is called a height anomaly.
The 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 surface, obtained by starting from the
assumptions that
◦ the true potential field of the Earth is the normal potential, and
◦ 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 in
normal potential with that of the reference ellipsoid.
8
Sir George Gabriel Stokes PRS (1819 – 1903) was an Irish-born, gifted mathe-
matician and physicist making his career in Cambridge.
Ó »îá .
Free-air anomalies 91
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 also that all variations in topographic height will be reflected
also as variations in the “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 describes the Earth’s
potential field. In it, variations in potential and variations in topographic
height are hopelessly mixed up.
This is why the quasi-geoid is an unfortunate compromise, a concession to
“reference-surface thinking”, which only really works within the classical
geoid concept. Better stick — within Molodensky’s theory — to the
concept height anomaly, which is a three-dimensional function or field
( )
ζ ( X , Y , Z ) = ζ ϕ, λ, h .
D 4.7 Free-air anomalies

If we measure gravity g in point P , the height of which over sea level is
H and its latitude Φ, we may calculate the gravity anomaly ∆ g in the
point as follows:
def
∆ g = g − γ ( H, Φ) ,
in which γ ( H, Φ) is normal gravity calculated according to its definition
at height H and latitude Φ.
This is how we define free-air anomalies.
We linearize this as follows:
∂γ ) ∂γ
∆ g = g − γ ( H, Φ) ≈ g − γ h, ϕ − ( H − h) − Φ−ϕ
( ) (
≈
dh ∂ϕ
( ) ∂γ ∂γ ( ) ∂γ
≈ g − γ 0, ϕ − h − ( H − h) = g − γ 0, ϕ − H ,
∂h dh dh
∂γ
where we make the approximation, that the vertical gradient ∂h
of normal
gravity is constant9 .
9
For greatest precision one should consider that also the latitude Φ may not be
a latitude on a geocentric reference ellipsoid, but, e.g., astronomical latitude, or
latitude in some old national co-ordinate system computed on a non-geocentric
ellipsoid, like in Finland KKJ, the National Grid Co-ordinate System, which was
computed on the Hayford ellipsoid. The error caused by this is however of order
a thousand times smaller than the effect caused by H − h.
Ó »îá .
Thus, free-air anomalies can be calculated in a simpler way. The gravity

formula of the normal field 3.7 gives for latitude 60◦ :
γ = 974,147.516 − 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
∆ g P = g P − γ0 ϕ + 0.3084 [mGal /m] H,

( )
(4.12)
( ) def ( )
in which γ0 ϕ = γ 0, ϕ , normal gravity at sea level, is only a function
of latitude. In a country like Finland, equation 4.12 is often sufficiently
precise, though today also the evaluation of original equation 3.7 is easy.
Free-air anomalies are widely used. Generally, when one discusses grav-
ity anomalies, one means just this, free-air anomalies. They describe the
Earth’s exterior gravity field, including mountains, valleys and every-
thing.
Questions:
1. If gravity on the Earth’s surface is 9.8 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:
105 × 9.8
1. At −0.3 mGal /m, it takes m = 3267 km to go to zero.
0. 3
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.
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
gradients? How are they related to the SI system?
5. How does the geoid height and the disturbing potential relate to
each other?
6. Explain the telluroid mapping and height anomalies.
Ó »îá .
Exercise 4 – 1: The spectrum of gravity anomalies 93
18 20 22 24 26 28 30 32
64 64
62 62
60 60
18 20 22 24 26 28 30 32
Figure 4.5. Free-air gravity anomalies for Southern Finland, computed from
the EGM2008 spherical-harmonic expansion. Data © Bureau
Gravimétrique International (BGI) / International Association of
Geodesy. Extraction address:
http://bgi.omp.obs-mip.fr/data-products/Toolbox/
D EGM2008-anomaly-maps-visualization.
D Exercise 4 – 1: The spectrum of gravity anomalies
Use equation 4.7. If we assume that the mean magnitude of the spectral
components ∆ g n of gravity anomalies
√ ¨
def 1
∆gn = ∆ g2n φ, λ d σ
( )
4π σ
does not depend on the chosen degree number n, how then does the
similar T n depend on n?
In other words: which degree numbers of the gravity field are relatively
strongest in the disturbing potential, and which in the gravity anomalies?
D Exercise 4 – 2: Deflections of the plumb line and geoid tilt
If, in the North-South components of plumb-line deflections in some

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, inter-point distance approx. 1000 km? See figure 4.1.
Ó »îá .
D Exercise 4 – 3: Gravity anomaly, geoid height
In Finland there is a place where the gravity anomaly (free air) is ∆ g =

100 mGal = 10−3 m /s2 . In the same place the disturbing potential T is
200 m2 s−2 .
1. Using the equation

∂T 2
∆g = − T,
∂n R
calculate ∂∂Tn , and compare it with the quantity 2
R T. Which of the
two ( ∂∂Tn or R2 T ) dominates?
2. Using the Bruns equation
T
N= ,
γ
where γ is average gravity 9.8 m /s2 , compute the geoid height N of

the point.
Ó »îá .
D
5 Geophysical reductions
D 5.1 General
We see that integral equations, like Green’s third theorem 1.24, offer a
possibility to calculate the whole exterior potential of the Earth — as
well as all quantities that may be calculated from the potential, like the
acceleration of gravity, etc. — from observed values V and ∂∂Vn on the
boundary surface only. Green’s third theorem is but one example out of
many: every integral theorem is the solution of some boundary-value
problem.
There are three alternatives concerning the choice of boundary surface:
1. Choose the topographic surface of the Earth.
2. Choose mean sea level, more precisely, a equipotential surface close
to mean sea level called the geoid.
3. Choose the reference ellipsoid.
◦ Alternative 1 has been developed most of all by the Molodensky

school (Molodensky et al., 1962) in the Soviet Union. The advantage
of the method is that we need no gravity reduction, as all significant
masses are already inside the boundary surface. Its disadvantage
is, that the, often complex, shape of the topography must be taken
into account when the boundary-value problem is formulated and
solved.
◦ 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 moved to the inside.
A further complication of the method then is, that the geopotential
or geoid solution obtained is not that of the original mass distribu-
tion, but of the reduced one. This surface is called the co-geoid. We
need a “restoration step” where this influence of the reduction step
epäsuora vaikutus on the geopotential and geoid is determined and reversed1 .
1
This influence is called the “indirect effect”.
– 95 –
In the literature this method is also referred to as the Remove-

Restore method.
◦ Alternative 3 has been used rarely, because it has not been tradi-
tionally possible to do gravity measurements in a location known
in the absolute sense, relative to the geocentre or the reference
ellipsoid. Nowadays this is possible using GNSS, e.g., 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, also
of gravimetric stations, are determined more and more directly with
GNSS. See, e.g., Märdla (2017).
D 5.2 Bouguer anomalies

Free-air anomalies depend on the topography. This is clear, because
gravity itself contains the attractive effect of topographic masses. A
map of free-air anomalies shows the same small details as seen in the
topography. One way of removing the effect of the topography is the
so-called Bouguer2 reduction.
D 5.2.1 Calculation
We calculate the effect of a homogeneous plate on gravity. Assume that

the plate is infinite in size; thickness d , matter density ρ , and height
of point P above the lower surface of the plate H . See figure 5.1. The
attraction in point P (which is directed straight downward for symmetry
reasons) is obtained by integrating. The volume integral to be computed
has a volume element
d V = ds · dz · s d α
( )
in the co-ordinates ( s, z, α). We transform this to the co-ordinates β, z, α .
We forget about α and study the surface element (figure5.1, top right)
ℓ
ds dz = d β dz,
cos β
in which the determinant of Jacobi needed is seen.
2
Pierre Bouguer (1698 – 1758) was a French professor in hydrography, who par-
ticipated 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 when De Maupertuis executed a
similar grade measurement in Lapland. In addition to geodesy, he was also
active in astronomy.
Ó »îá .
Bouguer anomalies 97
z
P
dz
ds ℓ
β cos β d β
H ℓ
d
s dm x
D Figure 5.1. The attraction of a Bouguer plate.
We carry out the integration:

˚
def cos β
a = ∥a∥ = G ρ dV =
ℓ2
ˆ 2π ˆ d ˆ ∞
cos β
= Gρ · ds dz · s d α =
0 0 0 ℓ2
ˆ 2π ˆ d ˆ π/2
cos β ℓ
= Gρ · d β dz · s d α =
0 0 0 ℓ 2 cos β
ˆ dˆ π/2
s
= 2πG ρ d β dz =
0 0 ℓ
ˆ d [ˆ π/2 ]
= 2πG ρ sin β d β dz .
0 0
Here, the integral

ˆ π/2
[ ]π/2
sin β d β = − cos β 0
= 1,
0
and the end result is

a = 2πG ρ d. (5.1)
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 :
ˆ β0 (z) [ ]β0 (z) ( )
sin β d β = − cos β 0
= 1 − cos β0 ( z) ,
0
and the whole integral

ˆ d[ ]
H−z
2πG ρ 1− √ dz .
0 ( H − z)2 + r 2
The indefinite integral is
ˆ
H−z √
√ dz = − ( H − z )2 + r 2 .
2
( H − z) + r2
Ó »îá .
Evaluation point P
II
Bouguer plate
I I
d=H
Topography
D Figure 5.2. The Bouguer plate as an approximation to the topography.
Substituting the bounds yields

ˆ d[ ]
H−z √ √
1− √ dz = d + ( H − d )2 + r 2 − H 2 + r2 .
0 ( H − z)2 + r 2
If we define √
def
ℓ ( z) = ( H − z)2 + r 2
we obtain for the whole integral
2πG ρ [ d + ℓ ( d ) − ℓ (0)] .
In the limit r → ∞, and thus ℓ ( d ) → ℓ (0) , this is identical to equation

5.1.
Bouguer anomalies are computed in order to remove the attraction of
masses of the Earth’s crust above sea level, i.e., the geoid. The true
topography is approximated by a Bouguer plate, see figure 5.2. There is
no standard way to treat sea-covered areas:
◦ Sometimes maps are drawn on which there are Bouguer anomalies

over land and free-air anomalies over the sea.
◦ A more correct way is to replace sea water by a rocky Bouguer
plate, the thickness of which equals the local sea depth, i.e., the
bathymetry.
The calculation goes as follows:
∆ g B = ∆ g FA − 2πG ρ H = ∆ g FA − 0.1119 H, (5.2)
where we assume for the density ρ of the plate an often used value for
the average density of the Earth’s crust, ρ = 2670 kg /m³. By substituting
into this equation 4.12 we obtain
∆ g B = g P − γ0 ϕ + [0.3084 − 0.1119] H = g P − γ0 ϕ + 0.1965 H. (5.3)

( ) ( )
The quantity ∆ g B is called a (simple) Bouguer anomaly.
Ó »îá .
Terrain effect and terrain correction 99
Topography
Free-air anomaly
Geoid
Bouguer anomaly
D Figure 5.3. The behaviour of different anomaly types in mountainous terrain.
D 5.2.2 Properties
Bouguer anomalies are, unlike free-air anomalies that vary on both sides
of zero, especially in the mountains strongly negative. E.g., if the mean
elevation of a mountain range is H = 1000 m, the Bouguer anomalies
systematiikka 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 interpolation and predic-
tion of gravity anomalies, in situations where the available gravimetric
material is sparse. However one then has to have access to topographic
heights.
D 5.3 Terrain effect and terrain correction

Using the simple Bouguer reduction does not remove precisely the attrac-
tive effect of the whole topography. Figure 5.2 shows that we make two
types of error:
◦ The attraction of volumes I is taken along, though there is nothing

there.
◦ The attraction of volumes II, where there actually is stuff, is ig-
nored.
Both errors work in the same direction! Because volumes I are below
the point of evaluation, their attraction — which the simple Bouguer
reduction considers present, and removes — would act downward. And
because volumes II are above the point of evaluation, their attraction —
which in the simple Bouguer reduction is not corrected for — acts upward.
The error made is in the same direction as in the previous case.

The terrain correction is always positive!

Ó »îá .
18 20 22 24 26 28 30 32
64 64
62 62
60 60
18 20 22 24 26 28 30 32
Figure 5.4. Terrain corrected Bouguer anomalies for Southern Finland,

computed from the spherical-harmonic expansion EGM2008. Data
© Bureau Gravimétrique International (BGI) / International
Association of Geodesy. Extraction address: http://bgi.omp.obs-mip.
fr/data-products/Toolbox/EGM2008-anomaly-maps-visualization.
Note, in comparison to figure 4.5 on page 93, the strong negative
bias of Bouguer anomalies — although part of this is due to
post-glacial isostatic unbalance and also visible in the free-air map.
Bouguer anomalies are also smoother, but that is harder to see
D here, as Southern Finland is already a smooth area.
We write
∆ g′B = ∆ g B + TC,
where TC — the “terrain correction” — is positive. ∆ g′B is called the
terrain corrected Bouguer anomaly.
The terrain correction is calculated by numerical integration. Figure
5.5 shows the prism integration method, and how both prisms, I and II,
lead to a positive correction, because prism I is computationally added
and prism II removed when applying the terrain correction. One needs
a digital terrain model, DTM, which must be, especially around the
evaluation point, extremely dense: according to experience, 500 m is
the maximum inter-point separation in a country like Finland, in the
mountains one needs even 50 m. The systematic nature of the terrain
correction makes a too sparse terrain model cause, possibly serious, biases
in the insufficiently corrected gravity anomalies.
Ó »îá .
Terrain effect and terrain correction 101
( )
H x′ , y′
Topography
θ II
H (x, y) P
I
( )
H x ′ , y′
Geoid
x, y x′ , y′
D Figure 5.5. Calculating the classical terrain correction by the prism method.
For computing 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:
ˆ +Dˆ +D [
1 ]2
H x′ , y′ − H ( x, y) ℓ−3 dx′ d y′ ,
( )
TC ( x, y) = G ρ
2 −D −D
where √
{1 }2
ℓ= ( x′ − x)2 + ( y′ − y) +2
[ H ( x′ , y′ ) − H ( x, y)]
2
[ ]T
is the distance between the evaluation point x y H ( x, y) and the
[ 1
[ ( )] ]T
centre point of the prism x y 2 H ( x, y) + H x′ , y . Of course ′
this is only an approximation, but it works well enough in terrain where

slopes generally do not exceed 45◦ . In the integral above, the limit D is
typically tens or hundreds of kilometres. In the latter case, the curvature
of the Earth already starts having an effect, which the formula does not
consider.
The values of the terrain correction TC vary from fractions of a milligal
(Southern Finland) to hundreds of milligals (high mountain ranges). In
the “arm” or Finland — the North-Western, somewhat mountainous
border area with Sweden and Norway —, the terrain correction may be
tens of milligals.
In figure 5.6 we depict the stages of calculating Bouguer anomalies from
gravity observations through terrain correction, Bouguer plate correction
and free-air reduction.
D 5.3.1 Example: applying the terrain correction in a special

case
Given the special terrain shape rendered in quasi 3D in figure 5.7. Here,
the height differences are PQ ′ = 300 m and QQ ′ = 200 m. Rock density is
the standard crustal density, 2670 kg /m³.
Questions:
Ó »îá .
Given: gravity g
on terrain Terrain Bouguer
correction plate
correction
Free-air
reduction Subtract normal gravity
to sea at sea level:
level −γ0 (ϕ)
Figure 5.6. The steps in calculating the Bouguer anomaly. Note that the reduc-
tion to sea level uses the standard free-air vertical gravity gradient,
D −0.3084 mGal /m, i.e., the vertical gradient of normal gravity.
1. Calculate the terrain correction at point P (hint: use the attraction

formula for the Bouguer plate). Algebraic sign?
2. Calculate the terrain correction at point Q . Algebraic sign?
3. If in point P is given that the free-air anomaly is 50 mGal, how
much is then the Bouguer anomaly in the point?
4. If in point Q is given that the Bouguer anomaly is 22 mGal, how
much is then the free-air anomaly in 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 level 300 metres. This
means the adding of half a Bouguer plate, thickness 100 m, below
the level of P . The effect (projected onto the vertical direction) is
1 1
TC = · 2πG ρ · H = · 0.1119 mGal /m · 100 m = 5.595 mGal.
2 2
2. The terrain correction at point Q is the change in gravity, if we
remove the half Bouguer plate to the right of the point, which is
Q′ Sea level
Figure 5.7. A special terrain shape. The vertical rock wall at PQ is also straight
D on a map and extends to infinity in both directions.
Ó »îá .
Spherical Bouguer anomalies 103
100 m thick. Its vertical gravity effect is, as calculated above,
TC = 5.595 mGal,
and, because a semi-plate is removed that is above the level of point

Q , the algebraic sign of TC is again positive.
3. Free air to Bouguer:
∆ g FA (P ) 50.000 mGal
TC +5.595 mGal
Bouguer plate removal, 300 m −33.570 mGal
∆ g B (P ) 22.025 mGal
∆ g FA +TC −Plate ∆ gB
4. Bouguer to free air:

∆ g B (Q ) 22.000 mGal
Bouguer plate addition, 200 m +23.800 mGal
TC “uncorrection” −5.595 mGal
∆ g FA (Q ) 40.205 mGal
∆ g B +Plate −TC ∆ g FA
D 5.4 Spherical Bouguer anomalies

More recently, also spherical Bouguer anomalies have been calculated,
e.g., 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). This causes four
differences with the Bouguer plate anomalies:
1. the attraction of a Bouguer shell of thickness H is 4πG ρ H , twice

as much as the corresponding Bouguer plate attraction.
2. The bathymetry of the oceans is accounted for3 by replacing the
water by standard-density crustal rock; this contribution to the
anomalies is positive.
3. Also the topography and bathymetry of remote parts of the globe
are taken into account realistically. As most of the Earth is covered
3
One can do so, and often does, also in connection with the Bouguer plate
correction.
Ó »îá .
by deep ocean, this causes a strong positive general bias, which in

moderately elevated areas like Southern Finland more than cancels
the negative one caused by the local topography!
4. As now also the terrain correction is calculated over the whole globe,
in spherical geometry, it is no longer a small number and may be
strongly negative as well as positive.
Between the planar and spherical Bouguer anomalies exists a large

systematic difference, 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 spher-
ical Bouguer reduction globally. The mean height of the land topogra-
phy 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 3700 × 2. 67 −1. 03
2. 67 m = 2272 m, assuming a density for crustal rock of
2670 kg /m3 and a sea-water density of 1030 kg /m3 , and ocean occupying 71%
of the globe. The sum weighted by area is thus
− (0.29 × 800 − 0.71 × 2272) m = 1381 m.
Interpretation: there is not enough topography to fill all of the oceans,

even if we’re allowed to compress sea water into standard crustal
rock. If we try this bulldozing experiment, we’ll end up 1381 m
short.
If, instead, we add standard crustal rock to end up at current sea
level — the definition of spherical Bouguer reduction! — we’ll
add to the Earth’s attraction as sensed from space an amount
4πG ρ × 1381 m = 309 mGal.
The global mean planar Bouguer reduction, as well as the difference
between planar and spherical Bouguer reductions, on average over
the globe, will be half of this, ≈ +155 mGal. Even coastal locations,
at zero height above sea level, will have similarly positive spherical
Bouguer anomalies!
D 5.5 Helmert condensation

An often used method, proposed by Friedrich Robert Helmert4 , for remov-
ing the effect of the masses exterior to the geoid is condensation. In this
method, we shift mathematically all the continental masses vertically
4
Friedrich Robert Helmert (1843 – 1917) was an eminent German geodesist
known for his work on mathematical and statistical geodesy.
Ó »îá .
Helmert condensation 105
Equipotential surface
Topography
g
g′
Condensation layer
D Figure 5.8. Helmert condensation and the changes it causes in the gravity field.
downward to mean sea level into a simple mass density layer κ = H ρ —

more precisely: ( )
H
κ = Hρ 1 + (5.4)
R
on a spherical Earth, radius R — where H is the height of the topography
above sea level and ρ its mean density. The advantage of Helmert con-
Figure 5.9. Friedrich Robert Helmert, © Humboldt University collection,

D Berlin.
Ó »îá .
densation 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
describe both the external and the internal 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, though we gloss over those
here.
D 5.6 Isostasy
D 5.6.1 Classical hypotheses
Already in the 18th and 19th centuries, e.g., 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 weren’t just piles
of rock on top of the Earth’s crust. The gravity field surrounding the
mountains, specifically the plumb line deflections, could only be 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 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 cor-
rect picture using physical methods (seismology) of how these mountain
roots were really shaped. That’s why simplified working hypotheses were
formulated.
One older isostatic hypothesis is the Pratt–Hayford hypothesis. This was
proposed by J. H. Pratt5 in the middle of the 19th century (Pratt, 1855,
1858, 1864), and J.F. Hayford6 developed the mathematical tools needed
for computation. According to this hypothesis, the density of the “root”
under a mountain would vary with the height of the mountain, so that
under the highest mountains would be the lightest material, and the
boundary between this light root material and the denser Earth mantle
5
John Henry Pratt (1809 – 1871) was a British clergyman and mathematician
who worked as the archdeacon of Kolkata, India.
https://en.wikipedia.org/wiki/John_Pratt_(Archdeacon_of_Calcutta).
6
John Fillmore Hayford (1868 – 1925) was a United States geodesist who studied
isostasy and the figure of the Earth.
Ó »îá .
Isostasy 107
Plumb-line deflections
Mountain Geoid
Earth’s crust
“Root”
Earth’s mantle
D Figure 5.10. Isostasy and the bending of plumb lines towards the mountain.
material would be at a fixed depth. This model, which nowadays finds

little acceptance anymore, is illustrated in figure 5.11.
Another classical isostatic hypothesis is due to G. B. Airy7 . Because V. A.
Heiskanen8 used it extensively and developed its mathematical form, it
is called the Airy–Heiskanen model. In this model it is assumed that the
7
George Biddell Airy PRS (1801 – 1892) was an English mathematician and
astronomer, “Astronomer Royal” 1835 – 1881.
8
Veikko Aleksanteri Heiskanen (1895 – 1971), ““the great Heiskanen” (http:
//en.wikipedia.org/wiki/Beyond_Sleep) was an eminent Finnish geodesist who
also worked in Ohio, USA He is known for his work on isostasy and global geoid
modelling. See Kakkuri (2008).
Mountains
Sea
Compen-
sation Earth’s crust
depth
Compen-
sation
level
Mantle
D Figure 5.11. Pratt–Hayford isostatic hypothesis.
Ó »îá .
Mountains
Sea σw
Earth’s crust
t0
σc
Anti-root
Mountain root
σm
Mantle
D Figure 5.12. Airy–Heiskanen isostatic hypothesis.
mass density of the “root” is fixed, and that the isostatic compensation
is realized 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 5.12.
D 5.6.2 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 crust be ρ c ,
the density of the mantle ρ m , and the density of sea water ρ w ; sea depth
d , crustal thickness t and topographic height H . We have
( )
d ρm − ρw + c
tρ c + d ρ w − ( t + d ) ρ m = c =⇒ t = −
ρm − ρc
on the sea, and
Hρm − c
tρ c − ( t − H ) ρ m = c =⇒ t =
ρm − ρc
on land. c is a suitable constant9 . Here we have conveniently forgotten

about the curvature of the Earth, i.e., we use the “flat Earth model”.
Under land, the depth of a mountain root is
H ρ m − c H ρ m − hρ c H ρ c − c
r = t−H = − = .
ρm − ρc ρm − ρc ρm − ρc
9
Its dimension 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.
Ó »îá .
Isostasy 109
t r
t0 t
r
Anti-root
Root
D Figure 5.13. Quantities in isostatic compensation.
Similarly under the sea

( ) ( )
d ρm − ρw + c dρm − dρc d ρc − ρw + c
r = t+d =− + =− .
ρm − ρc ρm − ρc ρm − ρc
Note that the constant c is arbitrary and expresses the fact that the
level from which one computes the depth of the root — less precisely, the
“average thickness of the crust” — can be chosen arbitrarily.
Another approach: instead of c, use the “zero topography compensation
level” t 0 , to be computed from the above equations by setting H = d = 0:
( )
t 0 ρ c − ρ m = c.
This yields under the land the root depth

( )
H ρ c − t0 ρ c − ρ m ρc
r= = t0 + H , (5.5)
ρm − ρc ρm − ρc
and under the sea
( ) ( )
d ρ c − ρ w + t0 ρ c − ρ m ρc − ρw
r=− = t0 − d , (5.6)
ρm − ρc ρm − ρc
somewhat simpler equations that are also more intuitive:
( )
H ρ c + (− r ) ρ m − ρ c = − t0 ;
( ) ( )
(− d ) ρ c − ρ w + (− r ) ρ m − ρ c = − t0 .
In other words,
∑
(deviation × density contrast) = const.
interfaces
However, the effect of the different isostatic hypotheses on gravity is

pretty much the same: the hypotheses can not be distinguished based on
gravity measurements only. The effect of the hypothesis on the geoid is
stronger.
Ó »îá .
D 5.6.3 Example: Norway
The Southern Norwegian Hardangar plateau (Hardangarvidda) 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 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 t 0 ?
2. What is the negative depth of the anti-root under the Norwegian
Sea, relative to the same compensation depth?
3. What is the relative depth of the root of the Hardanger plateau,
compared to the nearby Norwegian Sea?
Answers:
1. We use the equation 5.5, finding

ρc 2670 kg /m3
r − t0 = H = 1100 m × = 4196 m.
ρm − ρc 3370 − 2670 kg /m3
Here we have used standard densities for crustal and mantle rock,
respectively.
2. We use the equation 5.6, finding
ρc − ρw 2670 − 1027 kg /m3
r − t0 = −d = −2000 m × = −4694 m,
ρm − ρc 3370 − 2670 kg /m3
using the standard density value for sea water.
3. The depth contrast between root and anti-root is 4196 − (−4694) m =
8890 m (for perspective, Mount Everest is 8848 m above sea level).
D 5.6.4 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 picture of the internal structure of the Earth is given in figure
5.14.
An important subject for current research is the effect on vertical motion
of the Earth crust of the growing and melting of the ice masses of the
Earth, like the continental ice sheets. To this belongs both the direct mannerjäätiköt
effect of the varying ice masses, and the effect of the changes caused in
the water masses of the ocean. So-called paleo-research concentrates on
the changes over the glacial cycle, while modern retreats of glaciers, e.g.,
in Alaska and on Spitsbergen, cause their own, observable local uplift of
the Earth’s crust. More in chapter 11.
Ó »îá .
Isostasy 111
Mid-Atlantic ridge
Plate motion
Deep-sea trench Conrad
Earth’s crust discontinuity
Sea X
Mohorovičić
discontinuity
Lithosphere
Convection Bottom of
Subduction
lithosphere
Asthenosphere
Mantle Benioff zone

660 km
discontinuity
Figure 5.14. The modern understanding of isostasy and plate tectonics. Deep-
D sea trenches are known to be in isostatic disequilibrium.
D 5.6.5 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 maximal, at a rate of 10 mm /a. How long would it take
at this rate for the depression to disappear?
Answers:
1. We assume for the ice density a value of 920 kg /m3 . Then, with an
upper mantle density of 3370 kg /m3 — note that it is Earth’s mantle
material that is being displaced by the ice, the Earth’s crust just
transmits the load! See figure 11.1a — 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.
Part of this uplift has already taken place since the last deglacia-
tion.
In reality, of course, the rate has decreased substantially, and will
continue to decrease, over time.
Ó »îá .
D 5.7 Isostatic reductions

The computational removal of both the topography and its isostatic com-
pensation from the measured quantities of the gravity field is called
isostatic reduction. It serves two purposes.
1. 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.
2. These “superficial” effects are also generally very local: in spectral
language, very short wavelength. By removing those, we are left
with a residual field that is much smoother, and that can be inter-
polated 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.
For example, isostatic anomalies, i.e., free-air anomalies to which iso-

static reduction has been applied, are very smooth (like also Bouguer
anomalies), and their predictive properties are good. However, unlike
Bouguer anomalies, isostatic anomalies are on average zero. They lack
the large bias that makes Bouguer anomalies strongly negative especially
in mountainous areas (see section 5.2). This of course is because isostatic
reduction is only the shifting of masses from one place to another — from
mountains to roots beneath the same mountains, the mass deficit of which
is pretty precisely the same as the mass of the mountains themselves
sticking out above sea level — rather than removal of masses, which is
what Bouguer reduction does.
The reduction methods used in isostatic calculations are the same as in
other reductions, and 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 the spectral
domain (FFT, “Fast Collocation”, etc.).
The question of the hypothesis assumed to apply is a more interesting one.
Traditionally, the Pratt or Airy hypotheses have been used, developed
into quantitative methodologies by Hayford or Heiskanen or Vening
Meinesz10 . A newer approach has been to use real measurement 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.
10
Felix Andries Vening Meinesz (1887 – 1966) was a Dutch geophysicist, geode-
sist and gravimetrist. He wrote together with V. A. Heiskanen the textbook The
Earth and its Gravity Field (1958).
Ó »îá .
The “isostatic geoid” 113
18 20 22 24 26 28 30 32
64 64
62 62
60 60
18 20 22 24 26 28 30 32
Figure 5.15. 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. Extraction address:
http://bgi.omp.obs-mip.fr/data-products/Toolbox/
WGM2012-maps-vizualisation-extraction.
Note that, 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 4.5 on
D page 93.
D 5.8 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 the isostatic
co-geoid is under the continents as much as several metres below the
geoid, i.e., the indirect effect (“Restore” step) is of this order. Under the
oceans, similarly the isostatic co-geoid is somewhat above the geoid.
As one of the requirements for geoid determination methods is a small
indirect effect, it follows that isostatic methods are not (contrary to
what is said on Heiskanen and Moritz page 152) the best possible if the
intent is to calculate a model of the geoid or quasi-geoid representing
Ó »îá .
the exterior potential11 . However, isostatic methods are very suitable

for elucidating the interior structure of the Earth, because both the
topography and the “impression” it makes on the Earth’s mantle, the
isostatic compensation, are computationally removed. Research has
shown that the great topographic features of the Earth are some 85 −
90% isostatically compensated. 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
are especially convection currents in the Earth’s mantle as well as the
possible effect of the liquid 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:
1. computational removal of the topography

2. computational removal of the isostatic compensation of the topogra-
phy.
It is possible to calculate both these parts exactly using prism integration,

see section 5.3. Here however we shall gain understanding by a quali-
tative approach. We approximate both parts with a single mass density
layer, density, e.g., κ = ρ H for the topography. We place the first layer at
level H = 0, and the second, density -κ, at compensation depth H = −D .
The situation is depicted in figure 5.16.
In the following we use the “generating function” equation 7.4,
∞ ( )n+1
1 1∑ R ( )
= P n cos ψ ,
ℓ R r
n=0
together with the single mass density layer equation 1.15:

¨ ¨
κ 2 κ
V =G dS = GR d σ.
surface ℓ surface ℓ
We obtain for the potential field mass density layer at sea level, when
also the evaluation point is placed at at sea level, H = 0 =⇒ r = R :
¨ ∞
∑ ( )
Ttop = GR κ P n cos ψ d σ
σ n=0
11
Of course Bouguer reduction is even worse! The indirect effect can be hundreds
of metres.
Ó »îá .
The “isostatic geoid” 115
Sea level
Compensation depth
D Figure 5.16. Isostatic reduction as a pair of surface density layers.
and with the density layer at compensation depth (source level R − D ,

evaluation level R ):
¨ ∞ ( )n+1
GR 2 ∑ R−D ( )
Tcomp = (−κ) P n cos ψ d σ =
R−D σ R
n=0
¨ ∞ ( )n
∑ R−D ( )
= −GR κ P n cos ψ d σ,
σ R
n=0
from which the combined effect ( n = 0 drops out)

¨ ∞ [ ( )n ]
( ) ∑ R−D ( )
δTiso = − Ttop + Tcomp = −GR κ 1− P n cos ψ d σ.
σ R
n=1
(5.7)
Here, the mass density per unit of surface area κ is
{
ρc H if H ≥ 0,
κ= ( )
ρc − ρw H if H < 0,
i.e., we replace ocean depths by equivalent “dry” depths12 . Now we use

again the degree constituent equation, Heiskanen and Moritz (1967)
equation 1 – 71, or our equation 2.16, in the following form:
¨
) def 2 n + 1
κ φ′ , λ′ P n cos ψ d σ.
( ( ) ( )
κ n φ, λ =
4π σ
Multiplying both sides of this with the factor

)n ]
4π
[ (
R−D
1−
2n + 1 R
12
This works on dry land and on the ocean. Lakes, glaciers and areas like the
Dead Sea are more complicated.
Ó »îá .
and moving it inside the integral, we obtain

)n ]
4πGR
[ (
R−D ( )
− 1− κ n φ, λ =
2n + 1 R
¨ [ ( )n ]
( ′ ′
) R−D ( )
= −GR κ φ ,λ 1− P n cos ψ d σ.
σ R
Summation yields the expression 5.7 above:
∞ )n ]
4πGR
[ (
∑ R−D ( )
− 1− κ n φ, λ =
2n + 1 R
n=1
¨ ∞ [ ( )n ]
( ′ ′
)∑ R−D ( )
= −GR κ φ ,λ 1− P n cos ψ d σ,
σ R
n=1
so that
∞ [ ( )n ]
∑ 2 R−D
δTiso = − R 1− 2πG κn =
2n + 1 R
n=1
∞ [ ( )n ]
∑ 2 R−D
=− R 1− [ A B ]n .
2n + 1 R
n=1
Here we have used the notation A B = 2πG κ, the equivalent Bouguer

plate attraction of a mass density layer κ, and its degree constituent
[ A B ]n = 2πG κn .
R
Let us first look at the contribution from13 0 < n ≤ N = D. Then the
( −D ) n
following approximation holds, as R R ≈ 1 − nD
R :
N N
∑ 2 nD ∑
δTiso ≈ − [ A B ]n ≈ − D [ A B ]n ≈ −D A B ,
2n + 1
n=1 n=1
and
D AB δTiso
δ Niso = . ≈− (5.8)
γ γ
This is the indirect effect of isostatic reduction.
Let’s substitute realistic values. Let the Mohorovičić14 discontinuity’s
depth be on average ∼ 20 km15 .
13
The contribution from degree numbers n > R/D is
∞
∑ 2R
δTiso ≈ − [A B ]n ,
2n + 1
n= N +1
where the terms are small and rapidly falling to zero. In this degree range also
the mass density layer approximation for the topography breaks down.
14
Andrija Mohorovičić (1857 – 1936) was a Croatian meteorologist and a pioneer
of modern seismology.
15
Under the continents 35 km, under the oceans 7 km below the sea floor, accord-
ing to Encyclopædia Brittannica. Using these values, we find δ Niso = −3.2 m on
land, +2.8 m on the ocean.
Ó »îá .
On land: H ≈ 0.8 km (The Earth’s mean topographic height): =⇒ δ Niso ≈

−1.8 m.
2. 67−1. 03
On the ocean: H ≈ −3.7 km on average (multiply by 2. 67 because of
the water): =⇒ δ Niso ≈ +5.0 m.
In other words, this effect can be sizable!
Note that equation 5.8 is linear in the height H . This means that, under
the continents, the isostatic co-geoid will run on the order of a couple of
metres below the classical geoid, when on the oceans again it must be a
few metres above the geoid (mean sea level). We may also conclude that
in the isostatic reduction’s effect on the geoid – at least for longer wave-
lengths 2πR/n, longer than the compensation depth D — all wavelengths
are represented in the spectrum in approximately in the same propor-
tions as in the topography itself, and the effect is in fact proportional to
the topography.
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 (i.e., which additional information must be
available at the interpolation stage)?
4. How was it discovered that mountains have roots?
5. Describe the isostatic hypotheses of Pratt–Hayford and Airy–
Heiskanen.
D Exercise 5 – 1: Gravity anomaly
Given point P , height above sea level H = 500 m. Local gravity is

( )
g P = 9.82 m /s2 , normal gravity at sea level for local latitude ϕ is γ0 ϕ =
9.820192 m /s2 .
1. Compute point P ’s free-air anomaly ∆ g.
2. Compute point P ’s Bouguer anomaly (without terrain correction)
∆ gB.
D Exercise 5 – 2: Bouguer reduction
1. Point P is 500 m above sea level. Its free-air anomaly is ∆ g FA =

25 mGal. Calculate the Bouguer anomaly ∆ g B of the point. Forget
about the terrain correction.
Ó »îá .
2. See section 5.2: Bouguer anomalies. Derive the equations 5.2 and
5.3 anew, assuming that the mean density of the Earth crust were
ρ = 3370 kg /m3 .
D Exercise 5 – 3: Terrain correction, Bouguer
Given the terrain shape:
Q′ Sea level
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 in point P the terrain correction (hint: use the formula
for the attraction of a Bouguer plate. We have here a half Bouguer
plate, with only half the attraction of a full one.)
2. Compute in point Q the terrain correction. Algebraic sign?
3. If in point P is given that the free-air anomaly is 60 mGal, how
much is then the Bouguer anomaly in the point? (Use the complete
Bouguer reduction.)
4. If it is given in point Q that the Bouguer anomaly is 10 mGal, how
much is the point’s free-air anomaly?
D Exercise 5 – 4: Isostasy
Assume Airy-Heiskanen isostatic compensation (figure 5.12). Density

of Earth’s crust ρ c = 2670 kg /m3 , density of mantle ρ m = 3370 kg /m3 , i.e.,
the density contrast at the crust-mantle interface is 700 kg /m3 . Let the
reference level for the interface corresponding to zero topography be
−25 km, i.e. t 0 = 25 km.
1. Calculate the depth of the “root” of an 8 km high mountain below the

reference level −25 km, assuming it is isostatically compensated.
2. Mauna Kea is 4 km above sea level, however the surrounding sea is
5 km deep. How deep is the “root” of Mauna Kea below the reference
level?
3. How much is the “anti-root” of the surrounding sea above the refer-
ence level? Let the density of sea water be 1027 kg /m3 .
4. So, how deep is the “root” of Mauna Kea relative to its surroundings?
Ó »îá .
D Vertical reference systems
6
D 6.1 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 the effect of atmospheric refraction.
Longer distances are measured by repeat measurements using several
instrument stations and intermediate points.
The height differences ∆ H thus obtained are not, however, directly use-
able. The “height difference” between two points P and Q , obtained by
directly summing height differences ∆ H , depends namely also on the path
chosen when levelling from P to Q . Also the sum of height differences
Levelling staffs Horizontal

line of sight
00
10
1001 11
Level
back
11001100
00
00
11
fore
20
View
∆ H = back − fore
D Figure 6.1. The principle of levelling.
– 119 –
120 Vertical reference systems
⃝ ∆H
∑
around a closed path is generally not zero.

Geometric height is not a conservative field.

This is why, in precise levelling, the height differences are always con-
verted to potential differences: ∆W = −∆ H · g, in which g is the local
gravity, which is either measured or — e.g., in Finland — interpolated
from an existing gravity survey data base. The sum of potential differ-
ences around a closed loop is always zero: ⃝ ∆W = 0.
∑
For the potential of an arbitrary terrain point P we find

∑
WP = W0 − (∆ H · g ) ,
where the summation is done directly from sea level (potential W0 ) up to

point P . The quantity
P
∑
C P = − (WP − W0 ) = (∆ H · g ) ,
sea level
positive above sea level, is called the geopotential number of point P . geopotentiaaliluku
W0 is the potential of the national height reference level. In Finland,

the reference level of the old N60 system is in principle mean sea level
in Helsinki harbour at the beginning of 1960, which is why the system
is called N60. However the precise realization is a special pillar in the
garden of Helsinki astronomical observatory in Kaivopuisto1 . The new
Finnish height system is called N2000, and the realization 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 Amsterdam
datum NAP, Southern Europe has Trieste, North America the North
American Vertical Datum 1988 (NAVD88), datum point Father Point
(Pointe-au-Père)2 in Rimouski, Quebec, Canada, etc.
1
However, the value engraved in the pillar is the reference height of the still
older system NN, not of N60. The correct reference value for N60 for this pillar,
30.51376 m, is given in the publication Kääriäinen (1966).
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 na-
tive population of New France, today’s Quebec. Pointe-au-Père is also notorious
as the location of the RMS Empress of Ireland shipwreck in 1914, in which over
a thousand passengers perished.
Ó »îá .
Orthometric heights 121
Figure 6.2. Height reference pillar in the garden of Helsinki astronomical

observatory in Kaivopuisto, Kääriäinen (1966). Text:
Suomen Utgångspunkt för

tarkka precisionsnivellementet
vaakituksen i Finland
pääkiintopiste 30,4652 m öfver noll
30,4652 m yli nollan
(Reference bench mark of precise levelling of Finland, 30.4652 m

D above zero).
D 6.2 Orthometric heights

For creating a vertical reference, it would be simplest to use the original
geopotential differences from sea level, i.e., the geopotential numbers
defined 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 their 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!) flow always downward and
seek the state of minimum energy.
In Finland, as in many other countries, orthometric heights have been
long in use. They are physically defined heights above “mean sea level”
Ó »îá .
g
P
g WP
∆ H3
∆ H2 H ∆ H2′
∆ H1 ∆ H1′
W0
Geoid
Figure 6.3. Levelled heights and geopotential numbers. The height obtained by
∑3
summing levelled height differences, i=1 ∆ H i , is not the “correct”
∑3
height above the geoid, i.e., i=1 ∆ H ′i computed along the plumb
line.
Note how the equipotential or level surfaces of the geopotential are
not parallel: because of this, a journey along the Earth’s surface
may well go “upward”, to increasing heights above the geoid, al-
though the geopotential number decreases. Thus, water may flow
“upward”.
The gravity vector g is everywhere perpendicular to the level sur-
faces, and its length is inversely proportional to the distance sepa-
D rating the surfaces.
or the geoid. See figure 6.3.

The classical geoid is defined as

“The level surface of the Earth’s gravity field which fits on average
best to mean sea level.”

The orthometric height of point P is defined as the height obtained by

measuring along the plumb line the distance of P from the geoid.
This is a very physical definition, hoewever not a very operational one,
because we (generally) do not get to measure along a plumb line inside
the Earth, and the geoid isn’t visible there. This is why orthometric
heights are calculated from geopotential numbers: if the geopotential
number of point P is C P , we calculate the orthometric height with the
formula
CP
H= ,
g
where g, the average gravity along the plumb line, is
ˆ H
1
g= g ( z) dz,
H 0
Ó »îá .
Normal heights 123
North South
gS
Lake Päijänne: C = − (W − W0 ) = 76.9 GPU
gN
HS
Lake Päijänne
HN
Geoid: W = W0
Figure 6.4. Considered in terms of orthometric heights, water may sometimes

flow “upward”. Although the North and South ends of Lake Päi-
jänne 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 comm.). Calculation using the normal gravity field yields
6 mm. The balance of 2 mm comes from the difference between
D gravity anomalies at the Northern and Southern ends.
and z is the measured distance along the plumb line. Because the formula
for g already itself contains H , we obtain the solution iteratively, using
initially a crude estimate for H . The iteration converges fast.
We shall see that determining very precise orthometric heights is chal-
lenging, especially in the mountains.
D 6.3 Normal heights

In Finland, currently, with the height system N2000, 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.
D 6.3.1 Molodensky’s theory
M. S. Molodensky developed a theory in which the height of a point from

“mean sea level” would be defined by the following equation:
def C
H∗ = ,
γ0H
where γ0H is the average normal gravity computed between the zero level
(reference ellipsoid) and H ∗ along the ellipsoidal normal. So, the same
way of computing as in the case of orthometric heights, but using the
normal gravity field instead of the true gravity field.
Ó »îá .
Figure 6.5. Mikhail Sergeevich Molodensky, from an obscure Russian docu-

D ment.
Heights “above sea level” are for practical reasons given in metres. For
large, continental networks we want to give heights above a computa-
tional reference ellipsoid in metres, and thus also heights above “sea
level” have to be in metres.
Molodensky proposed also that instead of the geoid, height anomalies
would be used, the definition of which is
def T
ζ= , (6.1)
γHh
where now γHh is the average normal gravity at terrain level, more
precisely: between the heights H ∗ (but reckoned from the ellipsoid) and
h. T is the disturbing potential.
Based on these assumptions, he showed that
H ∗ + ζ = h,
where h is the height of a point above the reference ellipsoid. 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 ∗ close to H .
D 6.3.2 Molodensky’s proof
The realization of the Molodensky school was, that, because normal

gravity is along the plumb line very close to a linear function of place, one
could define a height type that can be computed directly from geopotential
Ó »îá .
Normal heights 125
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 indirectly, though the following
integral equation:
ˆ h
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 the local normal.
By defining
ˆ h
1
def
γ0h = γ ( z) dz
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 esthetic
flaw, that we divide here by the average normal gravity computed between
the levels 0 and h. This quantity is not operational without a means of
determining the ellipsoidal height h.
This suggests the following improvement based on the circumstance that
γ ( z) is a nearly linear function. This means that the vertical derivative
∂γ
∂r
is nearly constant in the height interval considered.
We define
ˆ h ˆ H+ ˆ h
1
def 1
def def1
γ0h = γ ( z) dz, γ0H = + γ ( z) dz, γHh = + γ ( z) dz .
h 0 H 0 N H+
Now
dγ N+
( )
1
γ0H ≈ γ0h − N + ≈ γ0h 1 − (6.2)
2 dr R
Ó »îá .
dγ 2γ
(R is the Earth’s radius, and dr ≈ R is assumed constant), and
dγ H+
( )
1
γHh ≈ γ0h + H + ≈ γ0h 1 + . (6.3)
2 dr R
N+ H+
Next, we also exploit that both R ≪ 1 and R ≪ 1, so
)−1 )−1
N+ N+ H+ H+
( ( ) ( ( )
1− ≈ 1+ , 1− ≈ 1+ ,
R R R R
and with equations 6.2, 6.3 and the definitions above of H + , N + ,
W − W0 γ0h N+ N + H+
( )
∗ W − W0
def
H =− =− · ≈ H+ 1 + = H+ + ,
γ0H γ0h γ0H R R
T γ0h H+ N + H+
( )
def T +
ζ= = · ≈N 1− = N+ − .
γHh γ0h γHh R R
N + H+
Because the, already small, correction terms R cancel, we finally
obtain
H ∗ + ζ = H + + N + = h. (6.4)
The quantity γ0H , and thus also normal height H ∗ , can be, unlike γ0h ,
computed using only information obtained by (spirit or trigonometric) lev-
elling, without having to know the height h above the reference ellipsoid,
which would require again knowledge of the local geoid.
This was Molodensky’s realization (Molodensky et al., 1962) already in
1945, long before the global positioning system GPS, or a global, geo-
centric reference ellipsoid, existed. Back then, continental triangulation
networks were computed on their own, regional reference ellipsoids.
+ +
The size of the correction term N RH is, taking for the 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 6.2 and 6.3 already assumed.
Figure 6.6 attempts to visualize the derivation.
D 6.3.3 Normal height and height anomaly
Normal height:
C W − W0
H∗ = =− , (6.5)
γ γ
where (recursive definition!)

ˆ H∗
1
γ = γ0H = ∗ γ ( z) dz.
H 0
Ó »îá .
Normal heights 127
N + H+
R
ζ N+
z
h
H+ H∗
N + H+
R
γ (z)
Figure 6.6. A graphic cartoon of Molodensky’s proof. 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 describe
the conversion process. The balls represent midpoints of averaging
D intervals for the function γ (z).
Height anomaly:
W −U
ζ=
γHh
where ˆ h
1
γHh = γ ( z) dz.
ζ H∗
The height anomaly ζ, which otherwise is a quantity similar to the
geoid height N , however is 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 the
topography), is called the telluroid. It is a mapping of sorts of the
topographic surface: the set of points Q , whose normal potential
UQ is the same as the true potential WP of the true topography’s
Topography
ζ
Telluroid
H∗ H h
Geoid H∗ Reference ellipsoid N ζ Quasi-geoid
Figure 6.7. Geoid, quasi-geoid, telluroid and topography. Note the correlation
D between quasi-geoid and topography.
Ó »îá .
corresponding point P . See figure 4.4.

Often, as a concession to old habits, we construct a surface that
is at a distance ζ above 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 ˆ h
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 differ-
ence 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).
D 6.4 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. Orthomet-
ric heights — e.g., Helmert heights — on the other hand are always used
together with geoid heights N . For computing both, H and N , one needs
the topographic mass density ρ , for which often a standard constant
value is assumed (2670 kg /m³), and the local vertical gradient of gravity,
for which generally the vertical normal gravity gradient (−0.3086 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 downward continued, exterior potential. If T is
the disturbing potential of the exterior, harmonically downward
continued field, then its difference between topography level and
sea level is: ˆ H
∂T
T H − T0 = dh ≈ −∆ g FA H, (6.6)
0 ∂h
Ó »îá .
Difference between geoid height and height anomaly 129
TH
and by using the Bruns equation twice, ζ = (height anomaly
γ
T0
or quasi-geoid height) and NFA = (”free-air geoid height”, FA =
γ
Free Air) we obtain3
∆ g FA H
ζ − NFA ≈ − . (6.7)
γ
2. Thus we have obtained the difference between height anomalies

and heights of the “free-air geoid”; what is left is determining the
separation between “free-air geoid” and geoid.
Let us approximate the topography by a Bouguer plate. Then
◦ in the case of the “free-air geoid” NFA the thickness of the plate
is the height H of point P . This is because the free-air geoid is
based on the harmonically downward continued exterior field,
meaning that also the Bouguer-plate attraction acting at P
must be continued downward, i.e., taken fully into account.
Because the surface mass density of the plate is H ρ , its as-
sumed attraction is everywhere on point P :s plumb line:
2πGH ρ .
◦ Now in the case of the geoid, we have to be physically realistic:

in an arbitrary point P ′ on the plumb line of point P, the
Bouguer plate is partly below the point, and partly above the
point. The attraction is then only
2πGH ′ ρ − 2πG H − H ′ ρ = 2πG 2 H ′ − H ρ ,

( ) ( )
in which H ′ is now the height of point P ′ .

By integrating the difference, like we did for equation 6.6, we obtain
ˆ H [(
2 H ′ − H − H dH ′ =
) ]
T − TFA = 2πG ρ
0
)2 ]H
H′ − 2 HH ′ = −2πG ρ H 2 ≈ − A B H,
[(
= 2πG ρ H ′ =0
in which A B is the attraction of a Bouguer plate of thickness H . We

obtain again by dividing by the average normal gravity:
AB H
N − NFA = − .
γ
By subtracting this latest result from equation 6.7 we find:

(−∆ g FA + A B ) H ∆ gB H
ζ− N = =− . (6.8)
γ γ
3
Here we made the approximation, that γ is the same on topography level as on
sea level.
Ó »îá .
See also Heiskanen and Moritz (1967, pages 8 – 13). As in the mountains
the Bouguer anomaly is strongly negative, it follows that the quasi-geoid
is there always above the geoid: approximately, using equation 5.2:
0.1119 [mGal /m] 2

H ≈ 10−7 m−1 H 2 .
[ ]
ζ− N ≈
9.8 [m /s²]
Or, if H is in units of [km] and ζ − N in units of [m]:
ζ − N [m] ≈ 0.1 H 2 [km] .
D 6.5 Difference between orthometric and normal heights

The geoid is the level from which orthometric heights are measured.
Therefore we may write
h = H + N,
where h is height above the reference ellipsoid and H orthometric height.

We may also bring back to memory equation 6.4:
h = H ∗ + ζ,
in which ζ is the height anomaly, and H ∗ the normal height.

We obtain simply:
∆ gB H
H − H∗ = ζ − N = − , (6.9)
γ
using equation 6.8.
D 6.6 Calculating orthometric heights precisely

Orthometric heights are a traditional way of expressing height “above
sea level”. Orthometric heights are heights above a real geoid, i.e., a level
surface inside the Earth and in the mean located at the same level as
mean sea level.
We may write
ˆ H
W = W0 − g ( z) dz
0
where g is the true gravity inside the topographic masses. From this we
obtain
C − (W − W0 )
H= = ,
g g
where mean gravity along the plumb line is
ˆ H
1
g= g ( z) dz.
H 0
Ó »îá .
Calculating orthometric heights precisely 131
The method is recursive: H appears both on the left and on the right side.
This is not a problem: both H and g are obtained iteratively. Convergence
is fast.
In practice one calculates orthometic height using an approximate for-
mula. In Finland, Helmert orthometric heights have long been used,
where gravity measured on the Earth’s surface, g ( H ), is extrapolated
downward by using the estimated vertical gravity gradient interior to
the rock. It is assumed that its standard value outside the rock, the
value, −0.3086 mGal /m (the free-air gradient), changes to a value that is
+0.2238 mGal /m greater (double Bouguer plate effect): the end result is the
total inside-rock gravity gradient, −0.0848 mGal /m.
This is called the Prey4 reduction. The end result are the following
equations (the coefficient is half the gradient, i.e., the mean gravity along
the plumb line is the same as gravity at the midpoint of the plumb line):
( 1 )
g = g ( H ) − 0.0848 mGal /m − H = g ( H ) + 0.0424 mGal /m · H, thus
2
C
H= , (6.10)
g ( H ) + 0.0424 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. Note that the term 0.0424 mGal /m · H is
typically much smaller than g ( H ) , which is about 9.8 m /s² = 980,000 mGal!
So, iteration, where the above denominator is first calculated using a
crude H value, converges really fast.
The use of Helmert heights as an approximation to orthometric heights
is imprecise for the following reasons:
◦ The assumption that gravity changes linearly along the plumb line.
This is not the case, especially not because of the terrain correction.
In the precise computation of orthometric heights, one ought to
compute the terrain correction separately for every point on the
plumb line.
◦ The assumption that the free-air vertical gravity gradient is a
constant, −0.3086 mGal /m. This is not the case, the gradient can
easily vary by ±10%.
◦ The assumption that rock density is ρ = 2.67 g /cm³. The true density
value may easily vary by ±10% or more around this assumed value.
The first approximation, neglecting the terrain effect, can be corrected by
using Niethammer5 ’s method (see Heiskanen and Moritz (1967) page 167).
4
Adalbert Prey (1873 – 1949) was an Austrian astronomer and geodesist and an
author of textbooks.
5
Theodor Niethammer (1876 – 1947) was a Swiss astronomer and geodesist who
was the first to map the gravity field of the Swiss Alps.
Ó »îá .
It requires that, also in geoid computation, the terrain is correspondingly

taken into account.
The third approximation, the density, can be removed as a problem by
conventionally agreeing to use also in the corresponding geoid computa-
tion a standard density ρ = 2.67 g /cm³. The surface thus obtained isn’t 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. However, the true gravity
gradient depends on local density variations. One can use, e.g., the
Poisson equation for computing the gradient, on which more later.
The precise calculation of orthometric heights is thus laborious. Just
as laborious as the precise determination of the geoid, and for the same
reasons. Fortunately in non-mountainous countries Helmert heights are
good enough. In Finland they were even computed using as ρ values
“true” crustal densities according to a geological map. . .
D 6.7 Calculating normal heights precisely

For this we use the equation 6.5:
C W − W0
H∗ = =− , (6.11)
γ γ
where the average value of normal gravity along the plumb line is
ˆ H∗
1
γ = γ0H = ∗ γ ( z) dz.
H 0
Because normal gravity is in good approximation a linear function of z,
we may write
( ) 1 ∂γ
γ = γ H∗ − H∗ ,
2 ∂z
∂γ
where ∂ z = −0.3086 mGal /m. We obtain
γ = γ H ∗ + 0.1543 mGal /m · H ∗ .
( )
The solution is again obtained iteratively:

C
H∗ = (6.12)
γ ( H ∗ ) + 0.1543 mGal /m · H ∗
where γ ( H ∗ ) can be calculated exactly when the height H ∗ (and the
local latitude) is known. H ∗ appears on both sides of the equation; it
converges fast because again, the first term of the denominator γ ( H ∗ ),
some 9.8 m /s² = 980,000 mGal, is a lot larger than 0.1543 mGal /m · H ∗ .
Calculating normal heights is not in the same way sensitive to Earth
crustal density and similar hypotheses, like calculation of orthometric
heights is. It depends however on the choice of normal field, i.e., the
reference ellipsoid.
Ó »îá .
Calculation example for heights 133
D 6.8 Calculation example for heights

At point P the potential difference with sea level is C = 5000 m2 /s2 . Local
gravity is g = 9.820,000 m /s2 .
Normal gravity calculated at the level of point P (i.e., calculated for a
point Q that is as high above the ellipsoid as P is above sea level) equals
γ = 9.820,492 m /s2 .
Questions:
1. Calculate the free-air anomaly ∆ g FA of point P .
2. Calculate the orthometric height of the point.
3. Calculate the Bouguer anomaly (without terrain correction)
∆ g B of the point.
4. Calculate the normal height of point P .
5. If the geoid height at point P is N = 25.000 m, how much is
then the height anomaly (“quasi-geoid height”) ζ?
Answers:
1. The free-air anomaly is
∆ g FA = (9.820,000 − 9.820,492) m /s2 = −49.2 mGal.

CC 5000
2. First attempt: H (0) = g /g = 9.82 m = 519.165 m. Second at-
tempt (equation 6.10):
(1) 5000 m2 /s2

H = =
9.820,000 m /s2 + 0.0424 · 10−5 s−2 · 519.165 m
= 509.154 m.
After that, the millimetres don’t change any more.

3. The Bouguer anomaly is (equation 5.2):
∆ g B = ∆ g FA − 0.1119 mGal /m H = −106.2 mGal.
4. The first attempt is again H ∗(0) = Cγ = 509.139 m. The second,

equation 6.12:
5000 m2 /s2
H ∗(1) = =
9.820,492 m /s2 + 0.1543 · 10−5 s−2 · 509.139 m
= 509.099 m,
also final on the millimetre level.

5. The difference equation 6.9 yields
∆ gB H
ζ− N = − = −0.055 m.
γ
Also (check) H ∗ − H = −0.055 m. So
ζ = N − (−0.055 m) = 25.055 m.
Ó »îá .
D 6.9 Orthometric and normal corrections

In practical orthometric height calculations, one often adds together at
first the height differences ∆ H measured by levelling (“staff reading
differences”) between points A and B as a tentative or crude height
difference
B
∑
∆ H,
A
after which the non-exactness of this method is accounted for by applying
the “orthometric correction” (OC ):
B
∑
HB = H A + ∆ H + OC AB .
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 C A , C B , ∆C the geopotential numbers at A and B, and the geopo-
∑B
tential differences along the levelling line, we have C B − C A − A ∆C = 0
because of the conservative nature of the geopotential. Dividing by a
constant γ0 yields
B
C B C A ∑ ∆C
− − = 0.
γ0 γ0 γ0
A
On the other hand we have
B B
∑ C B C A ∑ ∆C
OC AB = HB − H A − ∆H = − − ,
gB g A g
A A
with g A , gB average gravities along the plumb lines of A and B, g gravity

∑B
along the levelling line. In this expression, we compare A ∆ H , the
naively calculated sum of levelled height differences, with the difference
between the orthometric heights of the end points A and B, calculated
according to the definition.
Subtraction yields
B (
∆C ∆C
( ) ( ) ∑ )
CB CB CA CA
OC AB −0 = − − − − − ,
gB γ0 gA γ0 g γ0
A
in which
γ0 − g B C B γ0 − g B
( ) ( )
CB CB
− = = HB ,
gB γ0 γ0 gB γ0
γ0 − g A
( )
CA CA
− = HA,
gA γ0 γ0
∆C ∆C γ0 − g
( )
− = ∆ H,
g γ0 γ0
Ó »îá .
A vision for the future: relativistic levelling 135
yielding the orthometric correction
B (
g − γ0 g A − γ0 gB − γ0
∑ ) ( ) ( )
OC AB = ∆H + HA − HB , (6.13)
γ0 γ0 γ0
A
which is identical to Heiskanen and Moritz (1967) equation 4 – 33.

The choice of γ0 is arbitrary; it is wise to choose it close to the average
gravity in the general area of ( A, B) to keep the numerics small.
Similarly we may also compute the normal correction ( NC ) in calculating
normal heights. Start from the equation
B B
∑ C B C A ∑ ∆C
∗
NC AB = HB − H ∗A − ∆H = − − ,
γB γ A g
A A
from which, like above, follows by subtraction
B (
g − γ0 γ A − γ0 γB − γ0
∑ ) ( ) ( )
NC AB = ∆H + H ∗A − ∗
HB . (6.14)
γ0 γ0 γ0
A
Note that the identical first term in both equation 6.13 and equation 6.14
derives from the term
B B
∑ ∆C ∑
= ∆ H,
g
A A
the naive summation of height differences ∆ H in the case of both ortho-

metric and normal correction, which is the generic basis of the concept of
both corrections.
Applying result 6.14:
B
∑
∗
HB = H ∗A + ∆ H + NC AB .
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.
Note that both the orthometric correction 6.13 and the normal correction
6.14 can be calculated one staff interval at a time: one must know, in
addition to the levelled height difference ∆ H , local gravity g along the
levelling line, and also at the end points g ( H ) or γ ( H ∗ ) for calculating
mean gravity g or γ along the plumb lines of those end points. This goes
well with the equations given above. Remember that gravity g along the
levelling line is needed also in order to reduce the individual levelled
height differences ∆ H to geopotential number differences ∆C .
Ó »îá .
D 6.10 A vision for the future: relativistic levelling

According to general relativity, the deeper a clock is inside the potential yleinen
well of masses, the slower it ticks. This is most easily seen by looking at suhteellisuusteoria
the Schwarzschild6 metric for a spherically symmetric field: metriikka
( ) ( )−1
2 2 2GM 2 2 2GM
dr 2 − r 2 d φ2 + cos2 φ d λ2 =
( )
c dτ = 1 − 2 c dt − 1 − 2
c r c r
( ) ( )−1
2W 2 2 2W
dr 2 − r 2 d φ2 + cos2 φ d λ2 ,
( )
= 1− 2 c dt − 1 − 2
c c
( )
in spherical co-ordinates plus time r, φ, λ, t . Here we see, how the rate
of the 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− ≈ 1− 2.
∂t c 2 c
Now c2 is 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, 1983). With the new optical clocks the objective should
be achievable and relativistic levelling may become a reality.
The clock works in this way, that an extremely cold, so-called Bose–
Einstein condensate of atoms is trapped inside an optical lattice formed optinen hila
by six laser beams, an electromagnetic pattern of standing waves. The
readout beam of the clock oscillation uses a different frequency. A Bose–
Einstein condensate has the property that all atoms are in precisely the
same quantum state — like the photons in an operating laser —, i.e.,
their matter waves are coherent. In a way, all atoms together act as one
virtual atom.
The condensate may consist of millions of atoms, and can actually be seen
through the window of the vacuum chamber as a small plasma blob.
Unfortunately it is not enough, that just one laboratory measures time
to extreme precision — one also has to be able to compare the ticking
rates of different clocks over geographical distances. Also for this, a
solution has been found: existing optic-fibre 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
6
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
solution to the field equation of Albert Einstein’s general theory of relativity, the
Schwarzschild metric.
Ó »îá .
Braunschweig
Garching
Figure 6.8. An optical lattice clock: the ultra-precise atomic clock of the future
operates at optical wavelengths. To the right, the trajectory of the
D Predehl et al. experiment.
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.
1. Why are heights calculated directly from levelled height differences

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 may water sometimes flow in the ”wrong” direction, i.e., to a
greater height?
9. What is the telluroid?
10. What are the orthometric correction and the normal correction?
Ó »îá .
D Exercise 6 – 1: Calculating orthometric heights
In Finland, in connection with the N60 system, heights used were so-
called Helmert heights, a good approximation of orthometric heights.
The potential difference with sea level at point P , − (W − W0 ), equals
1000 m2 s−2 . Gravity in the point is g P = 9.820000 m s−2 . Calculate the
precise orthometric height of the point.
D Exercise 6 – 2: Calculating normal heights
In point P the potential difference with sea level is
− (W − W0 ) = 5000 m2 /s2 .
In the point, normal gravity is γP = 9.820000 m /s2 . Calculate the normal

height of the point. Use for the gradient with height of normal gravity
the value −0.3086 mGal /m.
D Exercise 6 – 3: Difference between orthometric and normal height
In point P the Bouguer anomaly is ∆ g B = −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.
Ó »îá .
D
7 The Stokes equation and other
integral equations
D 7.1 The Stokes equation and the Stokes integral kernel

By suitably combining the equations in section 4.3 one obtains on the
surface of a spherical Earth
∞
∑ ∆gn
T=R ,
n−1
n=2
(in which, for the degree numbers n = 0, 1, the ∆ g n are assumed to vanish,
as n = 0 means a different total mass for the normal field than for the
Earth, and n = 1 an offset of the co-ordinate origin from the Earth’s centre
of mass, see section 2.12). This is now the Stokes equation’s spectral form.
Using the degree constituent equation 2.16 one obtains the integral
equation
∞ ¨
R ∑ 2n + 1
∆ gP n cos ψ d σ =
( )
T=
4π n−1 σ
n=2
¨ [∑∞
]
R 2n + 1
∆ gd σ =
( )
= P n cos ψ
4π σ n−1
n=2
¨
R
S ψ ∆ gd σ,
( )
=
4π σ
in which
∞
( ) ∑ 2n + 1 ( )
S ψ = P n cos ψ ,
n−1
n=2
the Stokes kernel function. The angle ψ is the geocentric angular distance
between evaluation point and moving observation point, see figure 7.2.
The equation above allows the calculation, from global gravimetric data
and for every point on the Earth surface, of the disturbing potential T ,
and from that the geoid height N using the Bruns equation 4.2, N = Tγ ,
with the result
– 139 –
140 The Stokes equation and other integral equations
Mass excess
Mass
deficit
−N
D Figure 7.1. The principle of gravimetric geoid determination.

¨
R
S ψ ∆ g φ′ , λ′ d σ′ ,
( ) ( ) ( )
N φ, λ = (7.1)
4πγ σ

( ) ( )
in which φ, λ and φ′ , λ′ are the evaluation point and the moving
point (“observation point”) respectively, and the distance between them
is ψ. Equation 7.1 is the classical Stokes equation of gravimetric geoid
( )
N φ, λ S(ψ)
Evaluation
point
∆ g φ′ , λ′ d σ
( )
ψ
Moving data or
integration point
Earth’s centre
D Figure 7.2. Integrating the Stokes equation geometrically.
Ó »îá .
The Stokes equation and the Stokes integral kernel 141
25
( )
S ψ
1
20
sin (ψ/2)
ψ
−6 sin 2 + 1 − 5 cos ψ
ψ ψ)
+ sin2
(
15 −3 cos ψ ln sin 2 2
10
5
S(ψ) →
ψ→
−5
0 0.5 1 1.5 2 2.5 3 3.5
( )
Figure 7.3. The Stokes kernel function S ψ . The argument ψ is in radians
[0, π). Also plotted are the three parts of the analytical expression
D 7.2 with their different asymptotic behaviours.
determination.
The above illustrates the correspondence between integral equations
and spectral equations. There are other examples of this. Earlier we
presented the spectral representation of the function 1ℓ , Heiskanen and
Moritz (1967) equation 1 – 81. Of course 1ℓ is also the kernel function of an
integral equation, the one yielding the potential V if given is the single
layer mass density κ.
Also a version of the Stokes equation for the exterior space exists. We
gave it earlier, equation 4.10. The spectral form of its kernel function, see
equation 4.11, is
∞ ( ) n+1
( ) ∑ R 2n + 1 ( )
S r, ψ, R = P n cos ψ .
r n−1
n=2
The Stokes kernel function on the Earth’s surface is depicted in figure

7.3, in which the angle ψ is in radians (1 rad ≈ 57◦.3).
This curve was calculated using the following closed expression (Heiska-
nen and Moritz, 1967, section 2 – 16, equation 2 – 164):
1 ψ ( ψ ψ)
S (ψ) = − 6 sin + 1 − 5 cos ψ − 3 cos ψ ln sin + sin2 . (7.2)
sin (ψ/2) 2 2 2
This closed expression helps us to understand better how the function

( ψ )−1
behaves close to the origin ψ = 0: the first term, sin 2 , goes to infinity
Ó »îá .
ψ
when ψ → 0. The next three terms, −6 sin 2 + 1 − 5 cos ψ, are all bounded
on the whole interval [0, π) and the limit for ψ → 0 is −4. The last, com-
( ψ ψ)
plicated term −3 cos ψ ln sin 2 + sin2 2 goes also to infinity — positive
infinity! — for ψ → 0, but much more slowly, because of the logarithm.
D 7.2 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 line
(Heiskanen and Moritz, 1967 equation 2 – 210’):
¨ ( ){
ξ dS ψ cos α
{ } }
1
= ∆g dσ =
η 4πγ σ dψ sin α
¨ ( ){
dS ψ cos α
}
1
= ∆g · sin ψ d α d ψ, (7.3)
4πγ σ dψ sin α
where ξ, η are the North-South and East-West direction deflections of

the plumb line, and the unit-sphere surface element d σ = sin ψ d α d ψ, in
which sin ψ is Jacobi1 ’s determinant of the α, ψ co-ordinates.
( )
These equations were derived for the first time by the Dutch geophysicist
F. A. Vening Meinesz. The angle α is the azimuth or direction angle
( )
between the calculation or evaluation point φ, λ and the moving inte-
( )
gration 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 anom-
aly, are all so-called isotropic quantities: they do not depend on the
azimuth, and therefore in the spectral representation the transformations
between them are only functions of harmonic degree n.
D 7.3 The Poisson integral equation

Look at figure 7.4. The point Q of the body is located at R, and the obser-
vation point P at location r. The geocentric angular distance between the
two location vectors, as seen from the origin, is ψ. The distance between
points P and Q is ℓ.
def def
With R = ∥R∥ and r = ∥r∥, we may write (cosine rule):
√
ℓ= r 2 + R 2 − 2 rR cos ψ.
1
We may also write the function ℓ
as the following expansion (for proof,
1
Carl Gustav Jacob Jacobi (1804 – 1851) was a German mathematician.
Ó »îá .
The Poisson integral equation 143
z
P
ℓ r
y
Q ψ
R
Figure 7.4. The geometry of the generating function of the Legendre polynomi-
D als.
see Heiskanen and Moritz (1967) page 33):

∞ ( )n+1
1 1 1∑ R ( )
=√ = P n cos ψ (7.4)
ℓ r 2 + R 2 − 2Rr cos ψ R r
n=0
in which r = ∥r∥ and R = ∥R∥ are the distances of points P and Q from the
origin, the centre of the Earth. The function 7.4 is called the generating
function of the Legendre polynomials.
Differentiating equation 7.4 with respect to r yields
∞ ( )n+1
r − R cos ψ 1 ∑ n+1 R
− = − P n (cos ψ).
ℓ3 R r r
n=0
This we multiply by 2 r :
∞ ( )n+1
2 r 2 − 2 rR cos ψ 1∑ R
− = − (2 n + 2) P n (cos ψ).
ℓ3 R r
n=0
Now we add together this equation and equation 7.4:

∞ ( )n+1
−2 r 2 + 2 rR cos ψ + ℓ2 1∑ R ( )
= − (2 n + 1) P n cos ψ .
ℓ3 R r
n=0
The left-hand side is simplified by using − r 2 + 2 rR cos ψ + ℓ2 = R 2 :
−2 r 2 + 2 rR cos ψ + ℓ2 R 2 − r 2
= ,
ℓ3 ℓ3
and the end result is, by multiplying with −R ,
∞ ( )n+1
R r2 − R 2
( )
∑ R ( )
= (2 n + 1) P n cos ψ . (7.5)
ℓ3 r
n=0
Ó »îá .
Applying now the degree constituent equation 2.16 to the harmonic po-
tential field V on the spherical Earth’s surface, radius R :
¨
2n + 1
V φ′ , λ′ , R P n cos ψ d σ′ ,
( ) ( ) ( )
Vn φ, λ =
4π σ
as well as the spectral expansion of the field in space 2.17:

∞ ( ) n+1
( ) ∑ R ( )
V φ, λ, r = Vn φ, λ ,
r
n=0
we obtain
∞ ( )n+1 ¨
1 ∑ R
V φ′ , λ′ , R P n cos ψ d σ′ =
( ) ( ) ( )
V φ, λ, r = (2 n + 1)
4π r σ
n=0
¨ [∞ ( )n+1 ]
1 ′ ′
) ∑ R
d σ′ =
( ( )
= V φ ,λ ,R (2 n + 1) P n cos ψ
4π σ r
n=0
¨ (
r 2 − R 2 V φ′ , λ′ , R
) ( )
R
= d σ′
4π σ ℓ3
by substituting the expression in square brackets directly into equation
7.5.
Thus we have obtained the Poisson equation for computing a harmonic
field V from values given on the Earth’s surface:
¨ (
r2 − R 2
)
R
VP = VQ d σQ , (7.6)
4π σ ℓ3
in which ℓ is again the straight 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 the data point Q are φ′ , λ′ , R .
Still a third way to write the same equation, useful when the function
or field V isn’t actually defined between the topographic Earth’s surface
and sea level, is
¨ ( 2
r − R2 ∗
)
R
V= V d σ,
4π σ ℓ3
in which V ∗ denotes the value of a harmonically downward continued
function V — downward 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 ,
that is harmonic, and that also exists between topography and sea level.
The existence of such a function has been a classical theoretical nut to
crack. . .
Equation 7.6 solves the so-called Dirichlet boundary-value problem, find-
ing a function in an area of space when the value of the function on its
boundary has been given.
Ó »îá .
Gravity anomalies in the exterior space 145
D 7.4 Gravity anomalies in the exterior space

The equation derived in the previous section 7.3, equation 7.6, applies
for an arbitrary harmonic field V , i.e., a field for which ∆V = 0. The
equation may be conveniently applied to the expression r ∆ g, i.e., the
gravity anomaly multiplied by the radius, which is also a harmonic field.
This is how we can express the gravity anomaly in the external space
∆ g φ, λ, r as a function of gravity anomalies ∆ g φ′ , λ′ , R on a sphere
( ) ( )
of radius R . The function r ∆ g is harmonic, because according to equation

4.8
∞ ( )n+1
1∑ R
∆g = ( n − 1) Tn, (7.7)
r r
n=2
i.e.,
∞ ( ) n+1 ∞ ( ) n+1
∑ R ∑ R
r∆ g = ( n − 1) T n = Sn,
r r
n=2 n=2
( ) def ( )
in which S n φ, λ = ( n − 1) T n φ, λ is a perfectly legal surface spherical
( )
harmonic just like T n φ, λ itself. Also, the dependence on the radius
( )n+1
r , the factor Rr , is the same as for the (harmonic) potential. So,
Poisson’s integral equation 7.6 applies to function r ∆ g:
¨ (
r2 − R 2 R ∆ g φ′ , λ′ , R
)[ ( )]
R
r ∆ g φ, λ, r d σ′ ,
[ ( )]
=
4π σ ℓ3
or
¨ (
r2 − R 2
)
R2
∆ g φ, λ, r ∆ g φ′ , λ′ , R d σ′ .
( ) ( )
= (7.8)
4π r σ ℓ3
An alternative notation:
¨
R2 r2 − R 2
∆g = ∆ g ∗ d σ,
4π r σ ℓ 3
in which ∆ g∗ denotes the gravity anomaly at sea level, again calculated

by harmonic downward continuation of the exterior field, in this case the
expression r ∆ g.
See also Heiskanen and Moritz (1967) equation 2 – 160. Using the ap-
proximation r + R ≈ 2 r yields still
¨
R2 (r − R )
∆ g φ, λ, r ≈ ∆
( ) ( ′ ′ ) ′
g φ , λ , R dσ .
2π σ ℓ3
Alternatively we derive the spectral form:

∞ ( )n+1 ∞ ( ) n+2
1∑ R ∑ R
∆g = ( n − 1) T n = ∆ gn.
r r r
n=2 n=2
Ó »îá .
The degree constituent equation 2.16 gives the functions ∆ g n :

¨
2n + 1
∆gn = ∆ gP n cos ψ d σ,
( )
4π σ
with the aid of which

∞ ( )n+2 ¨
1 ∑ R
∆g = (2 n + 1) ∆ gP n (cos ψ) d σ =
4π r σ
n=2
¨ [∑
∞ ( ) n+2
]
1 R
= (2 n + 1) P n (cos ψ) ∆ gd σ =
4π σ r
n=2
¨
1
= K ∆ gd σ, (7.9)
4π σ
in which
∞ ( ) n+2
( ) def ∑ R ( )
K r, ψ, R = (2 n + 1) P n cos ψ
r
n=2
is a (modified) Poisson kernel for gravity anomalies. Its closed form can
be lifted from equation 7.8:
( ) R 2 r2 − R 2
K r, ψ, R = .
r ℓ3
Compared to the Stokes kernel, the Poisson kernel drops off fast to zero
for growing ℓ values. In other words, the evaluation of the integral
equation may be restricted to a very local area, e.g., a cap of radius kalotti
1◦ . See figure 7.5. The main use of Poisson’s kernel is the harmonic
continuation, upward or downward, i.e., the shifting of gravity anomalies
measured and computed at various levels to the same reference level.
In the limit r → R (sea level becomes the level of evaluation) this kernel
function goes asymptotically to the Dirac δ function.
D 7.5 The vertical gradient of the gravity anomaly

Differentiate a formula obtained from formulas 4.7, 4.8:
∞ ( ) n+2 ∞ ( )n+3
∑ R ∂∆ g 1∑ R
∆g = ∆ g n =⇒ =− ( n + 2) ∆ g n .
r ∂r R r
n=2 n=2
This equation is exact in spherical approximation. Its kernel function is

well localized, in other words, it drops off to zero very fast, i.e., a small
“cap” also here suffices for calculation.
∆ g n is, as calculated according to the degree constituent equation 2.16
from the anomaly field at sea level:
¨
2n + 1
∆gn = ∆ g φ′ , λ′ , R P n cos ψ d σ′ ,
( ) ( )
4π σ
Ó »îá .
The vertical gradient of the gravity anomaly 147
0.2
Poisson kernel, dimensionless

0.1
1 km
2 km
0
Gravity gradient kernel, unit km−1

−0.1
1 km
1.25 km
−0.2 1.5 km
2 km
−0.3
Distance (km) −→
0 1 2 3 4 5 6 7
Figure 7.5. The Poisson kernel function for gravity anomalies as well as the
kernel for the anomalous vertical gravity gradient, both at various
height levels. These are kernel values to be used when evaluating
D the surface integral in map co-ordinates in km.
i.e.,
∞ ( )n+3
∂∆ g φ, λ, r
( )
1 ∑ R
=− (2 n + 1) ( n + 2) ·
∂r 4π R r
n=2
¨
∆ g φ′ , λ′ , R P n cos ψ d σ′ =
( ) ( )
·
σ
¨
1
K ′ r, ψ, R ∆ g φ′ , λ′ , R d σ′ ,
( ) ( )
= (7.10)
4πR σ
where now the kernel function is

∞ ( ) n+3
′
( ) ∑ R ( )
K r, ψ, R = − (2 n + 1) ( n + 2) P n cos ψ .
r
n=2
Alternatively we derive a closed expression. We start from the Poisson

equation 7.8 for gravity anomalies, and differentiate2 with respect to r :
[ ¨ ]
∂∆ g φ, λ, r r2 − R 2
( ) ( )
∂ R2
)3/2 ∆ g φ , λ , R d σ =
′ ′ ′
( )
=
∂r ∂ r 4π r
(
σ r 2 + R 2 − 2 rR cos ψ
r − R2 2
⎡ ⎤
¨ 2 − −
R2 1 ⎢ r 2
⎥ ∆ g φ , λ , R dσ =
⎥ ( ′ ′ ) ′
= ⎢
4π σ ℓ3 ⎣ 3 2 r − 2R cos ψ r 2 − R 2 ⎦
( )( )
−
ℓ2
2 r)(
2¨
( 2 2 2
r2 − R 2
)]
3 ℓ +r −R
[
R 1
∆ g φ′ , λ′ , R d σ′ −
( )
= 2−
4π σ ℓ 3 2r ℓ
2 2
2
Hint: use symbolic algebra software.
Ó »îá .
¨ (
r 2 − R 2 ∆ g φ′ , λ′ , R
) ( )
1 R2
− d σ′ =
r 4π r ℓ3
⎡σ
3 r2 − R 2
( ) ⎤
¨ 2− −
R2 1 ⎢ 2 r 2
⎥ ∆ g φ , λ , R dσ −
⎥ ( ′ ′ ) ′
= ⎢
4π ℓ 3⎣ 2 2 2 2 ⎦
( )( )
σ 3 r − R r − R
−
2 r 2 ℓ2
1
− ∆ g φ, λ, r =
( )
r
¨ [ )2 ]
3 r2 − R 2
(
R2 1
∆ g φ′ , λ′ , R d σ′ −
( )
= 2−
4π σℓ
3 2 r 2 ℓ2
[1
3]
∆ g φ, λ, r =
( )
− +
r 2 r[
¨ )2 ]
3 r2 − R 2
(
R2 1 ( ′ ′ ) ′ 5
∆ ∆
( )
= 2 − g φ , λ , R d σ − g φ , λ, r .
4π σ ℓ 3 2 r 2 ℓ2 2r
(7.11)
In the final right-hand side, the rightmost term is very small, typically
under one part in a thousand, compared to the leftmost term. Both terms
inside the square brackets are of the same order of magnitude.
Finally, note that, if we are integrating over the surface of the Earth
sphere, radius R , rather than the unit sphere σ, radius 1, the coefficient
R 2 drops out from both equation 7.8 and equation 7.11.
In Molodensky’s method this or similar equations can be rapidly evalu-

ated from very local gravimetric data.
The closed expression given in Heiskanen and Moritz (1967), expression
2 – 2173 , is the anomalous vertical gravity gradient evaluated at sea level
(on the reference sphere). That is why it differs from expression 7.11 given
above. Also in the expression given here, like in expression 7.10, we need
gravity anomalies at sea level. Available however are anomalies at the
topographic surface level. In practice we can proceed iteratively, by first
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 ,
( ) ( ) ( )
( ) ( )
where H = H φ′ , λ′ is the topographic height at point φ′ , λ′ . When the
first, crude anomalous gradient has been calculated, e.g., using equation
7.11, we may perform a real reduction to sea level, initially linearly:
⏐(0)
(1) ∂∆ g ⏐⏐
∆g φ′ , λ′ , R ≈ ∆ g φ′ , λ′ , R + H −
( ) ( )
H,
∂r ⏐
and so forth.
3
In the derivation there it is assumed that ∆ g is harmonic. Not so: r ∆ g is
harmonic. The error made is small.
Ó »îá .
Gravity reductions in geoid determination 149
D 7.6 Gravity reductions in geoid determination

D 7.6.1 Classical methods
Use of the Stokes equation for gravimetric geoid determination presup-

poses that all masses are inside the geoid — and the exterior field thus
harmonic. For this reason we move the topographic masses computation-
ally to inside the geoid, in a way that needs to be specified. The classical
methods for this are
◦ Helmert’s (second) condensation method, section 5.5: the masses

are shifted vertically down to the geoid into a surface density layer.
After this, shifting gravity down from the topographic surface to
sea level is easy. The indirect effect (the effect of the mass shifts on
the geoid, the “Restore” step) is small.
◦ Isostatic reduction, in which the effects of both the topography and
its compensation, i.e., the “roots” of mountains below sea level, are
computationally removed. The indirect effect of this method is
larger. See section 5.7 and equation 5.8.
◦ Bouguer reduction, section 5.2: 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 the reduction is used
more rarely.
D 7.6.2 The Residual Terrain Modelling (RTM) method
Imagine that, conceptually, the topographic masses are shifted into the
geoid, to below sea level, in a way that does not change the exterior field.
This is materially the same as determining the geoid associated with the
harmonically downward continued exterior field.
The problem here is, that such a mass distribution below sea level, which
produces the harmonically downward continued external potential in the
space between topographic surface and geoid, doesn’t always precisely
exist. Or, that a suitable mass distribution will contain extremely large
positive and negative masses close to each other, which are physically
unrealistic.
huonosti asetettu One expresses this by saying that the problem is “ill posed”. In such cases
one uses regularization: one changes a little — as little as possible —
the exterior field, so that it corresponds precisely to some sensible field
that is harmonically continued below the topographic surface, and some
sensible mass distribution interior to the geoid that produces it.
Ó »îá .
One can start, e.g., by filtering out the short-wave parts caused by the
topography using a high resolution digital terrain model. This is called
the RTM (residual terrain modelling) method.
In this method, we do not actually move all topographic masses to be-
low the geoid. Instead, only masses close to the topographic surface are
either removed or filled in, in a way which creates a smooth replace-
ment 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 downward 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
valleys, i.e., we do 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. In each point P we apply the terrain correction as described in
section 5.3 to the gravity anomalies.
2. Next, we remove the attraction of a Bouguer plate, thickness H −
HRTM , where H stands for the terrain height of point P , and HRTM
for the height of the smoothed, or low-pass filtered, terrain at the
location of P . This effect is, according to equation 5.2 on page 98,
equal to
2πG ρ ( H − HRTM ) ,
in which ρ is the rock density assumed in the calculation.
3. After this, the location of the gravity anomaly is moved down4 —
“downward continuation” — from the original terrain level H to the
surface of the new, smoothed terrain, HRTM . For this, the Poisson
equation may be used as described in section 7.8.
If the vertical gravity gradient of the terrain-reduced field equals
the normal vertical gradient of gravity — according to section 4.4,
0.3 mGal /m —, this will leave the anomaly unchanged. Typically,
there will be a small change: one may show — exercise 1 – 1 item
4 — that on the surface of a buried sphere of anomalous density
∆ρ , there will be a radial anomalous gravity gradient of 83 πG ∆ρ .
For ∆ρ ≪ ρ , this will be negligible compared to the Bouguer-plate
coefficient in item 2.
4. Rigorously speaking, an inverse terrain correction for the shapes
of the smoothed terrain should be applied, to arrive at gravity
anomalies realistic for this new replacement topography. Often this
step is left out as the effect is small.
4
Or up!
Ó »îá .
Gravity reductions in geoid determination 151
P −
− − P′ +
+ +
−
Bouguer plate, down- Inverse
Terrain correction ward continuation terrain correction
Figure 7.6. Residual terrain modelling (RTM). One removes from the terrain
computationally the short wavelengths, i.e., the differences from
the red dashed line: the masses rising above it are removed, the
valleys below it are filled. After reduction, the red line, smoother
than the original terrain, is the new terrain surface. The exterior
potential of the new mass distribution will differ only little from
the original one, but may be harmonically downward continued to
sea level.
Left, terrain correction for point P; middle, Bouguer plate and
gradient reduction to the level of the smoothed terrain point P ′ ;
D and right, the inverse terrain correction for point P ′ .
5. After that, harmonic downward continuation of the exterior field

succeeds: in the exterior field remain nearly only the long wave-
lengths.
Because the mass shifts in the RTM method are so small, take place over
such small distances, and are so 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 be often negligible. For the same
reason the effect of unknown topographic density will 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).
D 7.6.3 Downward continuation in linear approximation
The approach described above can, following Molodensky, be linearized:

¨ [
∂∆ g ∂T
]
R
∆g φ ,λ , H −
′ ′ ′ ′
S ψ d σ′ +
( ) ( ) ( )
T φ, λ, H = H H. (7.12)
4π σ ∂H ′ ∂H
So, first we reduce the ∆ g measured and calculated at the topographic

surface to sea level using the gradient of the anomalies and the terrain
height H ′ of the measurement point, with the result
∂∆ g
∆ g∗ = ∆ g − H′.
∂H ′
Ó »îá .
After this, we apply, at sea level, the Stokes equation, and obtain the
disturbing potential at sea level T ∗ . After this, the disturbing potential
is “unreduced” back to terrain level, to the evaluation point, with the
equation
∂T
T = T∗ + H.
∂H
In these equations T , its vertical derivative, and ∆ g and its vertical
derivative always belong to the exterior harmonic gravity field, and the
connection between them is the fundamental equation of physical geodesy,
equation 4.5, in spherical geometry:
∂T 2
∆g = − − T,
∂H r
in which r = R + H . Here, we need firstly the vertical derivative of the
disturbing potential. This is easy: we have
∂T 2
= −∆ g − T,
∂H r
where the first term on the right is directly measured, and the second
term’s T is obtained iteratively from the main product of the solution
process.
Calculating the vertical gradient of gravity anomalies, i.e., the anomalous
vertical gradient of gravity, is harder. For this, we have the integral
equation 7.10:
¨
∂∆ g φ, λ, r
( )
1
K ′ r, ψ, R ∆ g φ′ , λ′ , R d σ′ ,
( ) ( )
=
∂r 4π R σ
∞ ( ) n+3
′
( ) ∑ R ( )
K r, ψ, R = − (2 n + 1) ( n + 2) P n cos ψ .
r
n=2
Luckily for practical calculations, the kernel K ′ of this integral is very

localized and one does not need gravimetric data ∆ g from a very large
area.
D 7.6.4 The evaluation point as the reference level
In the above equation 7.12 we used as the reference level the sea surface.
This is arbitrary: we may use whatever reference level, e.g., H0 , in which
case
¨ (
) ∂∆ g ( ′
)
R + H0
∆ g φ′ , λ′ , H ′ − H − H 0 S ψ d σ′ +
( ) ( ) ( )
T φ, λ, H =
4π σ ∂H
∂T
+ ( H − H0 ) .
∂H
If we now choose H0 = H , the last term drops off, and we obtain
¨ (
∂∆ g (
)
R+H
∆g φ ,λ , H −′ ′ ′ ′
S ψ d σ′ .
( ) ( ) ) ( )
T φ, λ, H = H −H
4π σ ∂h
Ó »îá .
The Remove-Restore method 153
In this case the reduction takes place from the height of the ∆ g measure-
ment point to the height of the T evaluation point, probably a shorter
distance than from sea level to evaluation height, especially in the imme-
diate surrounding of the evaluation point. This means that the lineariza-
tion error will remain smaller. What is bad, on the other hand, is that
the expression in parentheses is now different for each evaluation point.
This complicates the use of FFT based computation techniques, on which
more later.
Here, we were all the time discussing the determination of the disturbing
( )
potential T φ, λ, H ; this is in practice the same as determining the
height anomaly
( ) ( )
( ) T φ, λ, H T φ, λ, H
ζ φ, λ, H = ≈ ) ,
γHh
(
γ φ, H
equation 6.1. Here, γ is normal gravity calculated for point latitude φ

(≈ ϕ) and height H .
D 7.7 The Remove-Restore method

All geoid determination methods currently in use are in one way or
another “Remove-Restore” methods, even in several different ways.
1. From the observed gravity values, first the effect of the a global
gravity field model is removed. This model is generally given in the
form of a spherical-harmonic expansion. Thus, a residual gravity
field is obtained
◦ that has numerically smaller values — easier to work with —,
and
◦ that is more local: the long “wavelengths”, the patterns ex-
tending 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. The purpose of this is to obtain a residual
gravity field
◦ on which the Stokes equation may be applied, because no
masses are left outside the boundary surface, and
◦ from which especially the very small “wavelengths” — details
the size of which is of order a few kilometres — caused by the
topography, are gone. After this, prediction of gravity values
from sparse measurement values will work better.
Some gravity reduction methods — i.e., methods which computationally

remove the gravity effect of the exterior masses — with good predic-
Ó »îá .
tion properties, were already presented in subsection 7.6.1: Helmert

condensation, Bouguer reduction, and isostatic reduction.
We may illustrate the Remove-Restore method by the following commu-
tative diagram: kommutoiva kaavio
“Remove” “Restore”
∆g (−→ “Brute force” −→) N
⇓− L99 Global grav. field model 99K +⇑
∆ g loc Nloc
⇓− L99 Exterior masses 99K +⇑
∆ g red =⇒ Stokes =⇒ Nred
In this diagram the thick arrows denote calculations that are recom-
mended, because they are easy and accurate. The arrows in parentheses
refer to direct computation, which again is troublesome and compute
intensive.
D 7.8 Kernel modification in the Remove-Restore method

In the Remove-Restore method described above, the handling of reduced
gravity anomalies ∆ g red and geoid heights Nred happens typically within
a relatively small area. E.g., 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 the geoid of which is being
computed.
Also if we compute the geoid directly by integrating the Stokes equation,
we will evaluate this integral, after removing the effect of the global model
from the given gravity data, only over a limited area or cap: evaluate the kalotti
equation ¨
R
S ψ ∆ g red d σ,
( )
Nred = (7.13)
4πγ σ0
where σ0 is a cap on the unit sphere, the radius of which is, say, ψ0 .
The assumption behind this is, that, outside the cap, ∆ g red is both small
and rapidly varying, because the longer wavelengths have been removed
from it with the global-model reduction. This may however be a dangerous
assumption.
Write, in the above equation 7.13,
∞
( ) ∑ 2n + 1 ( )
S ψ = P n cos ψ
n−1
n=2
and
∞
∑
∆ g red φ, λ = ∆ g n φ, λ ,
( ) ( )
n=L+1
Ó »îá .
Kernel modification in the Remove-Restore method 155
assuming that L is the largest degree number that is still along in the
global spherical-harmonic expansion, or gravity model, that was sub-
tracted from the data5 .
Now, because ∆ g n is a certain linear combination of the surface spherical
harmonics
{ ( )
( ) P n|m| cos ψ sin | m| α m = − n, . . . , −1,
Ynm ψ, α = ( )
P nm cos ψ cos mα m = 0, . . . , n,
i.e.,
n
∑
∆ g n ψ, α = ∆ g nm Ynm ψ, α ,
( ) ( )
m=− n
and also
( ) ( )
Yn0 ψ, α = P n cos ψ ,
it follows from the orthogonality of the Y functions, that
¨ ¨
( )
P n cos ψ Yn′ m d σ = Yn0 Yn′ m d σ = 0
σ σ
if n ̸= n′ or m ̸= 0. Now we may write — note that the terms n ≤ L drop

away:
S ψ ∆ g red φ, λ = S ψ ∆ g red ψ, α =
( ) ( ) ( ) ( )
n
[ ∞
] [ ∞ ]
∑ 2n + 1 ∑ ∑
∆ g nm Ynm ψ, α
( ) ( )
= P n cos ψ × =
n−1
n=2 n=L+1 m=− n
n
[ ∞ ] [ ∞ ]
∑ 2n + 1 ( ∑ ∑
∆ g nm Ynm ψ, α
) ( )
= P n cos ψ × =
n−1
n=L+1 n=L+1 m=− n
L
ψ ∆ g red φ, λ ,
( ) ( )
=S
in which
∞
L
( ) ∑ 2n + 1 ( )
S ψ = P n cos ψ
n−1
n=L+1
is a so-called modified Stokes kernel function. The 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 expan-
sion of the S function to higher degree numbers, is called the Wong-Gore6
modification (Wong and Gore, 1969). A desirable property of the new
kernel function S L is, that it would be — at least compared to the original
5
. . . and that the model is accurate!
6
L. Wong and R.C. Gore worked at the Aerospace Corporation, a space technology
research institution in California. http://en.wikipedia.org/wiki/The_Aerospace_
Corporation.
Ó »îá .
25
S(ψ)
S 3 (ψ)
20 S 4 (ψ)
S 5 (ψ)
15 S 6 (ψ)
10
Angular distance ψ (rad) −→

−5
0 0.5 1 1.5 2 2.5 3 3.5
Figure 7.7. Modified Stokes kernel functions. Note how the kernel value outside
D the local area goes to zero for higher L values.
function S — small outside the cap area σ0 . In that case restricting the
integral to the cap instead of the whole unit sphere (equation 7.13) does
not do much damage. It is clear that S L is much narrower than S , as in
it, only the higher harmonic degrees are represented. This can be verified
by plotting a graph of both curves. It doesn’t go totally to zero outside the
cap, however, but oscillates somewhat.
The reason for this oscillation is, that in the frequency (i.e., degree num-
ber) domain, the cut-off of the modified kernel is quite sharp. Trans-
forming such a sharp edge between space and frequency domains will
invariably produce an oscillation, which is related to the so-called Gibbs7
phenomenon. Gibbsin ilmiö
D 7.9 Advanced kernel modifications

In the literature, other kernel modification methods are found. Their
general form is
∞ L
L
( ) ∑ 2n + 1 ( ) ∑ 2n + 1 ( )
S ψ = P n cos ψ + (1 − s n ) P n cos ψ =
n−1 n−1
n=L+1 n=2
L
( ) ∑ 2n + 1 ( )
=S ψ − sn P n cos ψ , (7.14)
n−1
n=2
7
Josiah Willard Gibbs (1839 – 1903) was an American physicist, chemist, mathe-
matician and engineer.
Ó »îá .
Advanced kernel modifications 157
where the modification coefficients s n , n = 2, . . . , L can be chosen8 . They

are chosen so as to minimize the values of S L in the area outside the cap,
σ − σ0 . In this way one may eliminate the truncation error of equation
7.13, and the oscillation of the Wong-Gore modification, almost entirely.
Molodensky et al. (1962) developed already earlier such a method.
In the above equation 7.14 we want to minimize the function
L
( ) ∑ 2n + 1 (
SL ψ = S ψ −
( ) )
sn P n cos ψ
n−1
n=2
over the area outside a local cap, σ − σ0 . Let us multiply this expression
( )
with each of the Legendre polynomials P n cos ψ , n = 2, . . . , L in turn,
and integrate over the area σ − σ0 outside the local cap:
ˆ
( ) ( )
S ψ P n cos ψ d σ −
σ−σ0
L ˆ
∑ 2 n′ + 1 ( ) ( )
− s n′ ′ P n′ cos ψ P n cos ψ d σ = 0, n = 2, . . . , L,
n −1 σ−σ0
n′ =2
a system of L − 1 equations in the L − 1 unknowns s n′ :

L
∑
A nn′ s n′ = b n ,
n′ =2
with ˆ
( ) ( )
bn = S ψ P n cos ψ d σ
σ−σ0
and ˆ
2 n′ + 1 ( ) ( )
A nn′ = ′ P n′ cos ψ P n cos ψ d σ.
n −1 σ−σ0
From this we can solve the s n for every degree number n from 2 to L.
This solution sets to zero the expressions
ˆ
S L ψ P n cos ψ d σ,
( ) ( )
(7.15)
σ−σ0
also for all values n from 2 to L.

The expressions 7.15 can be understood as inner or scalar products,
between the functions S L and P n . Similarly the elements of A nn′ contain
the scalar products between the functions P n and P n′ . Note that these
scalar products do not vanish: when integrating over σ − σ0 , unlike over
σ, the Legendre polynomials are not mutually orthogonal. Therefore the
A matrix is a full, symmetric, positive-definite matrix, not a diagonal
matrix like when integrating over the full unit sphere σ.
8
The choice s n = 1 again gives the simply modified Stokes kernel from which the
low degrees have been completely removed.
Ó »îá .
The Legendre polynomials are however independent of each other also on

this domain, 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 ψ0 and 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 S L . This means that S L will be a linear combination of the
Legendre polynomials, i.e., an element of the vector space spanned by the
polynomials P n , n = 2, . . . , L. But if this is so, and the scalar products 7.15
with each of the basis vectors vanish, then S L must be the zero function
on σ − σ0 .
See also Featherstone (2003).
Appendix A.1 explains more about linear vector spaces and the scalar
product of vectors.
D 7.10 The effect of the local zone

In numerical gravimetric geoid computation one uses averages of anoma-
lies computed over standard-sized cells or blocks, generally 5′ × 5′ , 10′ × 10′ ,
30′ × 30′ etc. At European latitudes, often sizes 3′ × 5′ , 5′ × 10′ , 6′ × 10′ etc.,
are used, which are approximately square.
The following equation applies when evaluating an integral using block
averages: ∑
N= ci∆ gi,
i
in which ∆ g i is the mean of block i , and the weight

¨
( ) R ( )
c i λ, φ = S ψ d σ,
4πG σi
where σ i is the surface area of block i .

Numerical evaluation of such an integral, or quadrature, is done conve-
niently using Simpson’s rule9 :
ˆ λ2ˆ ϕ2
R
S λ, φ, λ′ , φ′ cos φ d φ′ d λ′ ≈
( ) ( )
c i λ, φ =
4πγ λ1 ϕ1
1 1
R ∆λ∆φ ∑ ∑ ( )
≈ wj wk S jk λ, φ ,
4πγ
j =−1 k=−1
9
Thomas Simpson FRS (1710 – 1761) was an English mathematician and text-
book writer. Actually Simpson’s rule was used already a century earlier by
Johannes Kepler.
Ó »îá .
The effect of the local zone 159
4
1 36 1
36 36
k=1
16
4 36 4
36 36 k=0
1 1
k = −1
36 4 36
36
j= -1 0 1
D Figure 7.8. Simpson integration nodal weights in two dimensions.
where ∆λ and ∆ϕ are the block sizes in the latitude and longitude direc-
tions, and w−1 = w1 = 16 , w0 = 46 are the weights.
S jk = S λ, φ, λ′0 + j ∆λ, φ′0 + k∆φ cos φ,

( )
j, k = −1, 0, 1
[ ( ) ]
are the values of expression S λ, φ, λ′ , φ cos φ at the nodal points used
′
in the evaluation, 3 × 3 of them. See figure 7.8. Also more complicated

formulas (repeated Simpson or Romberg) can be employed.
One can show that the effect of the local (inner) zone on the geoid in the
( )
evaluation point φ, λ is proportional to the gravity anomaly in the point
ψ )−1
itself, ∆ g. Starting from the Stokes equation 7.1 with S ψ ≈ sin 2
( ) (
≈
2
ψ
, we find, for a circular inner zone of radius ψ0 :
ˆ 2π ˆ ψ0
R 2 R d
Nint = ∆ g sin ψ d ψ d α ≈ · 2π · ∆ g 0 · 2ψ0 = ∆ g 0 .
4πγ 0 0 ψ 4πγ γ
Here d = R ψ0 is the radius of the local block or cap in metres. ∆ g 0 is a
special average of the gravity anomaly, the average of “ring averages” for
radii between zero and d . If d is small, one may take for this the anomaly
value at the centre without incurring much error.
The local contributions to the deflections of the plumb line again are
proportional to the horizontal gradients of gravity anomalies. We start
from the Vening-Meinesz equations 7.3, with the above approximations
for a local cap:
ˆ ψ0 ˆ 2π
ξint cos α
{ } ( ) { }
1 2
≈ − 2 ∆g · sin ψ d α d ψ .
η int 4πγ 0 0 ψ sin α
We expand ∆ g into local rectangular metric co-ordinates x, y:
∂∆ g ∂∆ g ∂∆ g ∂∆ g
( )
∆ g ≈ ∆ g0 + x +y = ∆ g 0 + R sin ψ cos α + sin α ,
∂x ∂y ∂x ∂y
Ó »îá .
and substitute:
ˆ ψ0 ˆ 2π
∂∆ g ∂∆ g
( )[ ( )]
1 2
ξint ≈ − 2 ∆ g 0 + R sin ψ cos α + sin α ·
4πγ 0 0 ψ ∂x ∂y
· cos α sin ψ d α d ψ,
ˆ ψ0 ˆ 2π
∂∆ g ∂∆ g
( )[ ( )]
1 2
η int ≈ − 2 ∆ g 0 + R sin ψ cos α + sin α ·
4πγ 0 0 ψ ∂x ∂y
· sin α sin ψ d α d ψ .
´ 2π
Here, the terms in ∆ g 0 drop out in α integration (because 0 sin α = 0),
as do the mixed terms in sin α cos α. What remains is
ˆ ψ0 ˆ 2π
1 2 ∂∆ g
ξint ≈ − R sin ψ cos α cos α · sin ψ d α d ψ ≈
4πγ 0 0 ψ2 ∂x
ˆ ψ0 ˆ 2π ˆ ψ0
R ∂∆ g 2 R ∂∆ g
≈− cos α · d α d ψ = − d ψ,
2πγ 0 0 ∂x 2γ 0 ∂x
ˆ ψ0 ˆ 2π
1 2 ∂∆ g
η int ≈ − R sin ψ sin α sin α · sin ψ d α d ψ ≈
4πγ 0 0 ψ2 ∂y
ˆ ψ0 ˆ 2π ˆ ψ0
R ∂∆ g 2 R ∂∆ g
≈− sin α · d α d ψ = − dψ .
2πγ 0 0 ∂y 2γ 0 ∂y
Executing the final ψ integration yields now, with R ψ0 = d :
d ∂∆ g d ∂∆ g
ξint = − , η int = − .
2γ ∂ x 2γ ∂ y
Sometimes these equations are useful, e.g., when estimating the errors of
grid based methods. Let the mesh size of a grid be ∆ x, then we may set
in the above equations d ≈ ∆2x , and for ∆ g we take
∆ gobs − ∆ ggrid ,
in which ∆ ggrid is the gravity anomaly value interpolated from the grid
file at the evaluation point. In this way one obtains a rough estimate of
how much error is due to the mesh size of the grid.
1. What do the Stokes equation and its spectral form look like?
( )
2. What does the Stokes kernel function S ψ look like when expanded
in Legendre polynomials?
3. What is a suitable aaproximation 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.
Ó »îá .
Exercise 7 – 1: The Stokes equation 161
5. What does the Poisson integral equation describe?

6. Why are gravity reductions necessary when using the Stokes equa-
tion for computing a geoid model?
7. Which various gravity reduction methods are available?
8. Describe the Residual Terrain Modelling (RTM) method.
9. Describe the Remove-Restore method.
10. Why, in geoid determination, is the Stokes kernel function often
midified? What does such a modification look like?
11. What is the Gibbs phenomenon?
D Exercise 7 – 1: The Stokes equation

( )
1. Derive a simpler form of the Stokes function S ψ , which is valid
when the angular distance ψ is small. Really only one term!
2. Using this term, write the integral equation
¨
R
S ψ ∆ g dσ
( )
N=
4πγ σ
into polar co-ordinates, i.e., an integral of the form

ˆ 2π ˆ ∞
· · · ds d α,
0 0
in which s = ψR is the metric distance from the evaluation point,

and α the azimuth angle (direction angle) from the evaluation point
for N to the moving integration point for ∆ g.
[Hint: you need to consider Jacobi’s determinant for the polar co-
ordinates ( s, α)]
3. Compute N (as a formula) if ∆ g = ∆ g 0 only within a circular disk
s ≤ s 0 , and outside it ∆ g = 0. (Assume that s 0 is small.)
Ó »îá .
D Spectral techniques, FFT
8
D 8.1 The Stokes theorem as a convolution

We start from the Stokes equation
¨
R
S ψ ∆ g φ′ , λ′ d σ′ ,
( ) ( ) ( )
T φ, λ =
4π σ
( )
in which φ′ , λ′ is the location of the moving point — i.e., the integration
( )
or observation point — on the Earth’s surface, and φ, λ the location
of the evaluation point, it too on the Earth’s surface. Generally the
( )
locations of both points are given in spherical co-ordinates φ, λ , and
correspondingly the integration is executed over the surface of the unit
sphere σ: a surface element is d σ = cos φ d φ d λ, in which the factor cos φ
represents the determinant of Jacobi, for these spherical co-ordinates
( )
φ, λ .
However locally, in a sufficiently small area, one may write the point
co-ordinates also in rectangular form, and then express also the integral
in rectangular co-ordinates. Suitable rectangular co-ordinates are, e.g.,
map projection co-ordinates, see figure 8.1.
A simple example of rectangular co-ordinates in the tangent plane would
be
x = ψR sin α,
(8.1)
y = ψR cos α,
where α is the azimuth of the connection between evaluation point and
moving data point. The centre of this projection is the point where
the tangent plane touches the sphere. The locations of other points are
measured by the angle at the Earth’s centre, ψ, i.e., the geocentric angular
distance, and by the direction angle in the tangent plane or azimuth α.
A more realistic example uses a popular conformal map projection, the
stereographic projection:
(ψ)
x = 2 sin 2 R sin α,
(ψ)
y = 2 sin 2 R cos α.
– 163 –
164 Spectral techniques, FFT
y
Data point
α
x
Evaluation point
R ψ
Earth’s centre
D Figure 8.1. Map projection co-ordinates x, y in the local tangent plane.
In the limit for small values of ψ this is the corresponds to equations 8.1.
Taking the squares of equations 8.1, summing them, and dividing them
by R 2 yields
x2 + y2
ψ2 ≈ .
R2
More generally ψ is the angular distance between the two points ( x, y)
( )
(evaluation point) and x′ , y′ (data, integration or moving point) seen
from the Earth’s centre, approximately
)2 ( )2
x − x′ y − y′
(
2
ψ ≈ + .
R R
Furthermore we must account for Jacobi’s determinant of the projection:
d σ = R −2 dx d y
and the Stokes equation becomes now

¨ ∞
1
S x − x′ , y − y′ ∆ g x′ , y′ dx′ d y′ ,
( ) ( )
T ( x, y) ≈ (8.2)
4π R −∞
a two-dimensional convolution1 .
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,
4π R
1
The integration extends from minus to plus infinity in both co-ordinates x and y.
This can only be kept physically realistic on a curved Earth if it is assumed that
the kernel S is of bounded support, i.e., it differs from zero only in a bounded
area. This is the case for the modified kernels discussed in section 7.8.
Ó »îá .
Integration by FFT 165
and according to the convolution theorem (“Fourier transforms a convolu-

tion into a multiplication”):
{ } 1 { } { }
F T = F S F ∆g .
4πR
This ( x, y) plane approximation works only, if integration can be restricted
to a local area, where the curvature of the Earth’s surface may be ne-
glected. This is possible thanks to the use of global spherical-harmonic
expansions, because these describe the long-wavelength part of the spa-
tial variability of the Earth’s gravity field. After we have removed from
the observed gravity anomalies ∆ g the effect of the global spherical-
harmonic model (the “Remove” step) we may safely forget the effect of
areas far removed from the evaluation point: after this removal, the
anomaly field will contain only the remaining short-wavelength parts,
the effect of which cancels out at greater distances. Of course, once the
integral has been computed and the local disturbing potential Tloc has
been obtained, we must remember to add to it again the Tglob effect of
the global spherical-harmonic expansion to be calculated separately (the
“Restore” step).
D 8.2 Integration by FFT

The Fourier transform needed for applying the convolution theorem is
calculated as a discrete Fourier transform. For this purpose exists the
highly efficient Fast Fourier Transform, FFT (e.g., Kakkuri, 1981 pp. 183 –
200). There are several slightly differing formulations of the discrete
Fourier transform to be found in the literature. It doesn’t really matter
which one is chosen, as long as it is a compatible pair of a forward Fourier
transform F and a reverse Fourier transform F −1 .
hilaesitys In preparation for this we first compute a discrete grid representation
of the function ∆ g ( x, y), a rectangular table of ∆ g values on an equidis-
tant ( x, y) grid of points. The values may be, e.g., the function values
themselves at the grid points2 :
∆ g i j = ∆ g xi , y j ,
( )
where the co-ordinates of the grid points are
N N
x i = i δ x, i=− , . . . , − 1,
2 2
N N
y j = j δ y, j = − , . . . , − 1,
2 2
2
There exist alternatives to this. E.g., one could calculate for every grid point
the average over a square cell surrounding the point.
Ó »îá .
for suitably chosen grid spacings (δ x, δ y) . The sequence of values of

the subscripts i and j has been chosen so, that the centre of the area
( x = y = 0) is also in the centre of the constructed data table ∆ g i j , ( i =
j = 0). The integer N, assumed even, is here the grid size, assumed for
simplicity the same in both directions.
Next, we do the same for the kernel function
S ψ = S x − x′ , y − y′ = S (∆ x, ∆ y) ,
( ) ( )
i.e., we write
S i j = S ∆xi , ∆ y j ,
( )
where again ( N being the grid size):

N N
∆ x i = i δ x, i=− , . . . , − 1,
2 2
N N
∆ y j = j δ y, j = − , . . . , − 1.
2 2
Also here, the choice of the value sequences for the i and j subscripts is
based in the wish to get the central peak at the origin of the S function —
S (∆ x, ∆ y) → ∞ when (∆ x, ∆ y) → (0, 0) — placed at the centre of the grid
of values S i j 3 .
Next:
1. The grid representations ∆ g i j and S i j thus obtained of the func-

tions S and ∆ g are transformed to the frequency domain — they
become functions S uv and G uv of the two “frequencies”, the wave
numbers u and v in the x and y directions. The spatial frequencies
are ωu = Lu , ωv = Lv , in which L is the size of the area (assumed
square).
2. They are multiplied with each other “one frequency pair at a time”,
i.e., we calculate
1
Tuv = S uv G uv , u, v = 0 . . . N − 1.
2π R
3. We transform the result, Tuv = F T i j , back to the space domain:

{ }
T i j = F −1 Tuv , i.e., to a grid T i j = T x i , y j describing the dis-

{ } ( )
turbing potential T . The disturbing potential of an arbitrary point

can be obtained from this grid by interpolation. The co-ordinates
x i , y j run as functions of i, j in the same way as described above for
∆ g4 .
3
Without this measure the result of the calculation would be correct, but in the
wrong place. . .
4
In fact, for both ∆ g and T we could choose the simpler grid geometry where the
subscript sequences are i, j = 0, · · · , N − 1; however, for S it is mandatory to have
the origin in the middle of the grid.
Ó »îá .
Solution in rectangular co-ordinates 167
This method is good for computing the disturbing potential T — and sim-
ilarly the geoid height N = Tγ — from gravity anomalies using the Stokes
equation. It is just as good for evaluating other quantities, like, e.g., the
vertical gradient of the gravity anomaly by the Poisson equation. The
only requirement is, that the equation can be expressed as a convolution.
Also the inversion calculation is 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 in a
field to be integrated — in the example, gravity anomalies — is given
on a regular grid covering the area of computation, or can be converted
to one. The result — e.g., the disturbing potential — is obtained on a
regular grid in the same geometry. Values can then be interpolated to
chosen locations.
kommutoiva kaavio The FFT method may again be depicted as a commutative diagram:
Free observation point

Interpolation =⇒ Point grid
selection
↓ (Direct solution) ⇓ FFT
Solution points in
⇐= Interpolation Point grid
their own places
Appendix C offers a short explanation of why FFT works and what makes
it as efficient as it is.
D 8.3 Solution in rectangular co-ordinates

In the above equation 8.2, the grid co-ordinates x and y are rectangular.
( )
For practical reasons, we would rather use latitude and longitude φ, λ
as grid co-ordinates. In that way, the need to generate a new ( x, y) point
( )
grid by interpolating from the given ϕ, λ one through a map projection
calculation is avoided. However, working in geographic co-ordinates
causes errors due to meridian convergence — as a latitude and longitude
co-ordinate system isn’t actually rectangular. Slightly more suitable
( )
would be the pair φ, λ cos φ .
The problem has also been addressed on a more conceptual level.
D 8.3.1 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 the observation point
Ó »îá .
( )
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
( )
( ′
) 2 λ′ − λ
cos λ − λ = 1 − 2 sin ,
2
ψ
cos ψ = 1 − 2 sin2
,
2
φ′ − φ
cos φ′ − φ = 1 − 2 sin2
( )
,
2
and obtain the half-angle cosine rule:
( ′
) ′ 2 λ′ − λ
cos ψ = cos φ − φ − 2 cos φ cos φ sin =⇒
2
ψ φ ′
− φ λ′
− λ
sin2 = sin2 + cos φ′ cos φ sin2 .
2 2 2
Here we may use the following approximation:
def
cos φ′ ≈ cos φ = cos φ0 ,
in which φ0 is a reference latitude in the middle of the calculation area.

Now the above equation becomes
ψ φ′ − φ λ′ − λ
sin2 ≈ sin2 + cos2 φ0 sin2 , (8.3)
2 2 2
def def
which depends only on the differences ∆φ = φ′ − φ and ∆λ = λ′ − λ, a
requirement for convolution.
( )5
After this, the FFT method may be applied by using co-ordinates φ, λ
and the re-written Stokes kernel
√
∆φ ∆λ
( )
) def 2
∗
∆φ, ∆λ = S 2 arcsin + cos2 φ0 sin2
(
S sin .
2 2
This clever way of using FFT in geographic co-ordinates was invented by

the Dutchman G. Strang van Hees6 in 1990.
D 8.3.2 “Spherical FFT”, multi-band model
We divide the area in several narrow latitude bands. In each band we

apply the Strang van Hees method using its own optimal central latitude.
5
In practice one uses the geodetic or geographic latitude ϕ instead of φ without
significant error.
6
Govert L. Strang van Hees (1932 – 2012) was a Dutch gravimetric geodesist.
Ó »îá .
Write the Stokes equation as follows:

¨
R
S φ − φ′ , λ − λ′ ; φ ∆ g φ′ , λ′ cos φ′ d φ′ d λ′ ,
( ) ( )[ ( ) ]
N φ, λ = (8.4)
4πγ
where we have expressed S (·) as a function of latitude difference, longi-
tude difference and evaluation latitude. Now, choose two support lati-
tudes: φ i and φ i+1 . Assume furthermore that S is between these a linear
function of φ. In that case we may write
φ − φ i S i+1 ∆φ, ∆λ + φ i+1 − φ S i ∆φ, ∆λ

( ) ( ) ( ) ( )
S ∆φ, ∆λ; φ =
( )
,
φ i+1 − φ i
where ∆φ = φ − φ′ , ∆λ = λ − λ′ and
S i ∆φ, ∆λ = S φ − φ′ , λ − λ′ ; φ i ,
( ) ( )
S i+1 ∆φ, ∆λ = S φ − φ′ , λ − λ′ ; φ i+1 .

( ) ( )
We obtain by substitution into integral equation 8.4
φ i+1 − φ φ − φi
{[ ] [ ] }
( R)
N φ, λ = Ii + I i+1 (8.5)
4πγ φ i+1 − φ i φ i+1 − φ i
with
¨
S i ∆φ, ∆λ ∆ g φ′ , λ′ cos φ′ d φ′ d λ′ ,
( )[ ( ) ]
Ii =
¨
S i+1 ∆φ, ∆λ ∆ g φ′ , λ′ cos φ′ d φ′ d λ′ .
( )[ ( ) ]
I i+1 =
This equation is the linear combination of two convolutions. Both are

evaluated by FFT and from the solutions obtained one forms the weighted
mean according to equation 8.5.
In this method we use, instead of the approximative equation 8.3, 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 − ∆φ cos φ sin2
( )
.
2 2
Here again, we calculate S i and S i+1 for the values φ = φ i and φ = φ i+1 ,
we evaluate the integrals with the aid of the convolution theorem, and
( )
interpolate N φ, λ according to equation 8.5 when φ i ≤ φ < φ i+1 . After
this, the solution isn’t entirely exact, because inside every band we still
use linear interpolation. However by making the bands narrower, we can
keep the error arbitrarily small.
D 8.3.3 “Spherical FFT”, Taylor expansion model
This somewhat more complicated but also more versatile approach ex-
pands the Stokes kernel into a Taylor expansion with respect to latitude
Ó »îá .
about a reference latitude located in the middle of the computation area7 .

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:
ˆ 2π ˆ +π/2
C ϕ, ϕ′ , ∆λ m ϕ′ , λ′ cos ϕ′ d ϕ′ d λ′ ,
( ) ( )[ ( ) ]
ℓ ϕ, λ =
0 −π/2
where ℓ contains values to be computed, m values given, and C is the

coefficient or kernel function. Here is assumed only rotational symmetry
around the Earth’s axis for the geometry, i.e., the kernel function depends
only on the difference between longitudes ∆λ rather than the absolute
longitudes λ, λ′ .
In a concrete case m contains for example gravity anomaly values ∆ g
( )
in various points ϕ′ , λ′ , ℓ contains geoid heights N in various points
( )
ϕ, λ , and C contains coefficients calculated using the Stokes kernel
function.
We first change the dependence upon ϕ, ϕ′ into a dependence upon ϕ, ∆ϕ:
C = C ϕ, ϕ′ , ∆λ = C ∆ϕ, ∆λ, ϕ .
( ) ( )
Linearize:
C = C 0 ∆ϕ, ∆λ + ϕ − ϕ0 C ϕ ∆ϕ, ∆λ + . . .
( ) ( ) ( )
where we define for a suitable reference latitude ϕ0 ,

) def (
C 0 ∆ϕ, ∆λ = C ∆ϕ, ∆λ, ϕ0 ,
( )
) def ∂
⏐
∆ϕ, ∆λ = C ∆ϕ, ∆λ, ϕ ⏐⏐
( )⏐ (
Cϕ .
∂ϕ ϕ=ϕ0
Note that this expansion into two terms will work only for a limited
range on ∆ϕ, and that the kernel function C is assumed to be of bounded
support. In this case, the integrals may be calculated over a limited area
instead of over the whole Earth.
Substitution yields
¨
C ∆ϕ, ∆λ, ϕ · m ϕ′ , λ′ cos ϕ′ d ϕ′ d λ′ =
( ) ( ) ( )
ℓ ϕ, λ =
¨
C 0 + ϕ − ϕ0 C ϕ · m cos ϕ′ d ϕ′ d λ′ =
[ ( ) ]
=
¨
= C 0 · m cos ϕ′ d ϕ′ d λ′ +
¨
C ϕ · m cos ϕ′ d ϕ′ d λ′ .
( )
+ ϕ − ϕ0 (8.6)
7
In the literature the method has been generalized by expanding the kernel also
with respect to height.
Ó »îá .
Important here is now, that the integrals in the first and second terms,
¨
C 0 ∆ϕ, ∆λ m ϕ′ , λ′ cos ϕ′ d ϕ′ d λ′ = C 0 ⊛ m cos ϕ ,
( )[ ( ) ] [ ]
¨
C ϕ ∆ϕ, ∆λ m ϕ′ , λ′ cos ϕ′ d ϕ′ d λ′ = C ϕ ⊛ m cos ϕ ,
( )[ ( ) ] [ ]
are both convolutions: both C functions depend only on ∆ϕ and ∆λ. Both
integrals can be calculated only if the corresponding ∆ϕ = ϕ − ϕ′ and
∆λ = λ − λ′ , and the corresponding coefficient grids C 0 , C ϕ , are calculated
first in preparation. After this (in principle expensive, but, thanks to
FFT and the convolution theorem, a lot cheaper) integration, computing
the compound 8.6 is cheap: one multiplication and one addition for each
( )
evaluation point ϕ, λ .
Example: let the evaluation area at latitude 60◦ be 10◦ × 20◦ in size. If
the grid mesh size is 5′ × 10′ , the number of cells is 120 × 120. Let
us choose, e.g., a 256 × 256 grid (i.e., N = 256) and fill the missing
values with extrapolated values.
Also the values of the kernel functions C 0 and C ϕ are calculated
on a 256 × 256 size ∆ϕ, ∆λ grid. The number of these is thus
( )
[ ]
also 65,536. Calculating the convolutions C 0 ⊛ m cos ϕ and C ϕ ⊛
[ ]
m cos ϕ by means of FFT — i.e.,
¨
C 0 ∆ϕ, ∆λ m ϕ′ , λ′ cos ϕ′ d ϕ′ d λ′ = C 0 ⊛ m cos ϕ =
( ) ( ) [ ]
{ { } { }}
= F −1 F C 0 F m cos ϕ ,
¨
C ϕ ∆ϕ, ∆λ m ϕ′ , λ′ cos ϕ′ d ϕ′ d λ′ = C ϕ ⊛ m cos ϕ =
( ) ( ) [ ]
{ { } { }}
= F −1 F C ϕ F m cos ϕ ,
requires N 2 ×2 log N 2 = 65,536 × 16 = more than a million op-

( ) ( )
( )
erations, multiplication with ϕ − ϕ0 and adding together, each
again 65,536 operations.
The grid matrices corresponding to functions C 0 and C ϕ are ob-
tained as follows: for three reference latitudes ϕ−1 , ϕ0 , ϕ+1 we com-
pute numerically the grids
C −1 = C ∆ϕ, ∆λ, ϕ−1 ,

( )
C 0 = C ∆ϕ, ∆λ, ϕ0 ,
( )
C +1 = C ∆ϕ, ∆λ, ϕ+1 ,

( )
where C 0 is directly available, and
C +1 − C −1
Cϕ ≈ .
ϕ+1 − ϕ−1
Ó »îá .
Also inversion calculation is thus directly feasible. Let ℓ be given in

suitable point grid form. We compute the first approximation to m as
follows:
F ℓ
{ } { } { } { { } }
](0)
F C 0 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) = C 0 ⊛ m cos ϕ
[ ) [
+ ϕ − ϕ0 · C ϕ ⊛ m cos ϕ ,
after which we make the improvement:
F ℓ − ℓ(0)
{ { }}
](1) ](0)
+ F −1
[ [
m cos ϕ = m cos ϕ ,
F C0
{ }
and so on, iteratively. Two, three interations are usually enough. This
method has been used to compute underground mass points from gravity
anomalies to represent the exterior gravity field of the Earth. More is
explained in Forsberg and Vermeer (1992).
D 8.3.4 “1D-FFT”
This is a limiting case of the previous ones, in which FFT is used only
in the longitude direction. In other words, a zones method where the
zones have a width of only a single grid row. This method is exact if all
longitudes (0◦ − 360◦ ) are along in the calculation. It requires a bit 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).
D 8.4 Bordering and tapering of the data area

The discrete Fourier transform presupposes the data to be periodically
continuous. In other words, it is assumed that when connecting the
Eastern edge of the data area to the Western edge, and the Northern edge
to the Southern edge, the data has to be continuous across these edges8 .
In practice this is not the case. We are faced with two different problems:
1. the data on the opposite side of an edge must be so far away to have
no noticeable influence on the result of the calculation, and
2. the data is continuous across the edges.
Therefore, always when using FFT with the convolution theorem 8.2, two
measures need to be taken.
8
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.
Ó »îá .
Bordering and tapering of the data area 173
25% 50% 25%

Data area
1
D Figure 8.2. “Tapering” 25%.
1. we continue the data by adding a border area to the data area,

so-called bordering. Often the border area is 25% of the size of the
data area; then, the size of the whole calculation area will become
four times that of the data area itself. The border is often filled
with zeroes, although predicted values — or even measured values,
if those exist — are a better choice.
Also the calculation area for the kernel function is made similarly
four times larger. In this case, as the function is symmetric, the
border area is filled with real (computable) values, making it auto-
matically periodically continuous.
2. Because the discrete Fourier transform assumes periodicity, one
must make sure that the data really is periodic. If the values at
the borders are not zero, they may be forced to zero by multiplying
the whole data area by a so-called tapering function, which goes
smoothly to zero towards the edges. Such a function can easily be
built, e.g., a cubic spline polynomial or a Tukey or cosine tapering.
See figure 8.2, showing a 25% tapering function, as well as example
images 8.3, where one sees how non-periodicity — differing left
and right, and upper and lower, edges — causes horizontal and
vertical artefacts in the Fourier transform. These artefacts are
called the Gibbs phenomenon, already mentioned in section 7.8: a
sharp cut-off or edge in the space domain will produce signal in all
frequencies up to the highest ones.
Many journal articles have appeared on these technicalities. Groups that

have been involved in early development of FFT geoid determination
already in the 1980s are Forsberg’s group in Copenhagen, the group of
Schwarz and Sideris in Calgary, Canada, the Delft group (Strang van
Hees, Haagmans, De Min, Van Gelderen), the Milanese (Sansò, Barzaghi,
Brovelli), Heiner Denker at the Hannover “Institut für Erdmessung”, and
many others.
Ó »îá .
Figure 8.3. Example images for FFT transform without (above) and with
(below) tapering. Used on-line FFT http://www.ejectamenta.com/
D Imaging-Experiments/fourierimagefiltering.html.
D 8.5 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 avail-
ability of precise geoid models an important issue, so that one can use
GNSS (Global Navigation Satellite Systems) technology for rapid and
inexpensive height determination.
D 8.5.1 The GRAVSOFT software
The GRAVSOFT geoid computation software has been mainly produced

in Denmark. Authors include Carl Christian Tscherning9 , René Fors-
berg, Per Knudsen, the Norwegian Dag Solheim, and the Greek Dimitris
Arabelos. The manual is Forsberg and Tscherning (2008).
This package is in widespread use and offers, in addition to variants of
9
Carl Christian Tscherning (1942 – 2014) was a Danish physical geodesist well
known for his research into the gravity field of the Earth. He did groundbreaking
work on statistical computation methods for modelling the Earth’s gravity field
from many different measurement types.
Ó »îá .
Computing a geoid model with FFT 175
20˚ 24˚ 28˚ 32˚
20
70˚ 70˚
19
68˚ 68˚
24 25
22
21
23
31
28
29
30
20
26
27
66˚ 19 66˚
17
18
18
64˚ 64˚
18
5
23 24 2
20 212
2
19
62˚ 62˚
18
17
19 16
60˚ 60˚
15
16
20˚ 32˚
24˚ 28˚
2010 Oct 20 13:27:28
Figure 8.4. The Finnish FIN2000 geoid. Data source: Finnish Geodetic Insti-
D tute.
FFT geoid determination, also, e.g., least-squares collocation, routines

for evaluation of various terrain effects, etc. Its spread can be partly
explained by it being free for scientific use, and being distributed as
source code. It is also well documented. Therefore it has also found
commercial use, e.g., in the petroleum extraction industry.
GRAVSOFT has been used a lot also for teaching, e.g., at many research
schools organized by the IAG (International Association of Geodesy) in
various countries. http://www.isgeoid.polimi.it/Schools/schools.html.
D 8.5.2 The Finnish FIN2000 geoid
Currently two geoid models are in use in Finland: FIN2000 (figure 8.4)
and FIN2005N00 (Bilker-Koivula and Ollikainen, 2009). The first model
is a reference surface for the N60 height system: using it together with
GNSS positioning allows determination of the N60 heights of points.
Ó »îá .
The model gives geoid heights above the GRS80 reference ellipsoid. The
second model is similarly a reference surface for the new N2000 height
system. It too gives heights from the GRS80 reference ellipsoid.
The precisions (mean errors) of FIN2000 and FIN2005N00 are on the
level of ±2 − 3 cm.
D 8.6 Use of FFT computation in other contexts

D 8.6.1 Satellite altimetry
The Danish researchers Per Knudsen and Ole Balthasar Andersen have
computed a gravity map of the world ocean by starting from satellite
altimetry derived “geoid heights” and inverting them to gravity anomalies
(Andersen et al., 2010). A pioneer of the method has been David Sandwell
from the Scripps Institute of Oceanography in California (e.g., Garcia
et al., 2014). The short-wavelength features in the map can tell us about
the sea-floor topography.
D 8.6.2 Satellite gravity missions and airborne gravimetry
Also the data from satellite gravity missions (like CHAMP, GRACE and
GOCE) can be regionally processed using the FFT method: in the case
of GOCE, the inversion of gradiometric measurements, i.e., calculating
geoid heights on the Earth’s surface from measurements made at satellite
level. Also airborne gravity measurements are processed in this way. The
problem is called “harmonic downward continuation” and is in principle
unstable.
Airborne gravimetry is a practical method for gravimetric mapping of
large areas. In the pioneering days, the gravity field over Greenland
was mapped, as well as many areas around the Arctic and Antarctic.
Later, areas were measured like the Brazilian Amazonas, Mongolia and
Ethiopia (Bedada, 2010), where no full-coverage terrestrial gravimetric
data existed. The advantage of this method is that one measures rapidly
large areas in a homogeneous way. Also for the processing of airborne
gravimetry data, FFT is suitable.
D 8.7 Computing terrain corrections with FFT

The terrain correction is a very localized phenomenon, the calculation
of which requires high-resolution terrain data from a relatively small
area surrounding the computation point. Thus, calculating the terrain
correction is ideally suited for the FFT method.
We show how, with FFT, we can simply and efficiently evaluate the
terrain correction. We make the following simplifying assumptions:
Ó »îá .
Computing terrain corrections with FFT 177
1. Terrain slopes are relatively small.

2. The density ρ of the Earth’s crust is constant.
3. 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 joint effect of all the topographic 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 describes the attraction of rock
columns projected vertically (figure 5.5):
¨ ( )
+∞ Gρ H′ − H
TC ( x, y) = cos θ dx′ d y′
−∞ ℓ2
¨ ( )
+∞ Gρ H′ − H 1 H′ − H
= · dx′ d y′
−∞ ℓ2 2 ℓ
¨ ( )2
1 +∞ H′ − H
= Gρ dx′ d y′ . (8.7)
2 −∞ ℓ3
Here G ρ H ′ − H ℓ−2 is the attraction of the column and 12 H ′ − H ℓ−1
( ) ( )
is the cosine of the angle θ between the force vector — assumed coming
from the midpoint of the rock column — and the vertical direction. This
is the so-called prism method.
We will make a linear approximation, wherein ℓ, the slant distance
( )
between the evaluation point ( x, y) and the moving data point x′ , y′ , is
also the horizontal distance:
)2 ( )2
ℓ2 ≈ x ′ − x + y′ − y
(
.
Equation 8.7 is easy to check straight from Newton’s law of gravitation.

When it is assumed that the terrain is relatively free of steep slopes, then
ℓ is large compared to H ′ − H .
From equation 8.7 we obtain by development into terms:
¨ +∞
1 1
TC ( x, y) = G ρ hH 2 dx′ d y′ −
2 ℓ3
¨ +∞
−∞
′
H
− Gρ H dx′ d y′ +
−∞ ℓ3
¨ ( )2
1 +∞ H′
+ Gρ dx′ d y′ , (8.8)
2 −∞ ℓ3
where every integral is a convolution with kernel ℓ−3 , and functions to be
( )2
integrated are 1, H ′ and H ′ .
Unfortunately the function ℓ−3 as implicitly defined above has no Fourier
transform, wherefore we change the above definition a tiny bit by adding
a small term:
ℓ2 = ( x′ − x)2 + ( y′ − y)2 + δ2 . (8.9)
Ó »îá .
Then, the terms in the above sum are large numbers that almost cancel
each other, giving a nearly correct result. Numerically this is however an
unpleasant situation.
If ℓ is defined according to equation 8.9, then the Fourier transform of
kernel ℓ−3 is (Harrison and Dickinson, 1989):
2π 2π 4π2 q2 δ2
{ } [ ]
−3
F ℓ = exp (−2πδ q) = 1 − 2πδ q + −... ,
δ δ 1·2
def p
in which q = u2 + v2 , and u, v are wave numbers (i.e., “frequencies”)
in the x and y directions in the ( x, y) plane. If we substitute this into
equation 8.8, we notice that the terms containing δ−1 sum to zero, and of
course also the terms containing positive powers of δ vanish when δ → 0.
As follows (Harrison and Dickinson, 1989):
{ } 1 { }[ 2π ]
2
F TC ≈ Gρ H F 1 (1 − 2πδ q) −
2 δ
{ }[ 2π ]
′
− Gρ HF H (1 − 2πδ q) +
δ
2 [ 2π
{( ) }
1 ]
+ GρF H′ (1 − 2πδ q)
2 δ
where we leave off all terms in higher powers of δ.

Re-order the terms:
{ } π [
2
{ } {
′
} {(
′ 2
}]
F TC = G ρ H F 1 − 2H F H +F
)
H +
δ
[ 1 { } { } 1 { ( ) }]
2 2 ′ 2
+ 4π q − G ρ H F 1 + G ρ H F H − GρF H′ .
2 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):
{ } π [ {
2 ′ ′ 2
}]
F TC = G ρ F H − 2 HH + H
( )
+
δ
2
[ {
′
} 1 { ( ) }]
2
+ 4π q G ρ H F H − GρF H′
2
and the reverse Fourier transform yields:
2πG ρ [ 1 2 1 ( ′ )2 ]
TC = H − H′ H + H +
δ 2 { {2 } }
+G ρ H F −1 F H ′ · 4π2 q −
1 { {( ) }
2
}
− G ρ F −1 F H′ · 4π 2 q .
2
In the first term
1 2 1 ( ′ )2 1 ( ′ )2
H − H′ H + H = H −H =0
2 2 2
Ó »îá .
in point ( x, y) where H ′ = H , and we obtain:

{ [ { } 1 { ( ′ )2 }]}
TC = 4π2 G ρ F −1 q · H F H ′ − F H ,
2
from which now the troublesome δ−1 has vanished.

A condition for this “regularization” or “renormalization” is, that at point
( x, y) H ′ = H , i.e., the evaluation happens at the Earth’s surface. The
convolutions above are evaluated by the FFT technique. A more detailed
account is found, e.g., in the article Vermeer (1992).
For calculating the terrain correction TC in the exterior space — airborne
gravimetry, but also the effect of the sea floor at the sea surface, or the
effect of the Mohorovičić discontinuity at the Earth’s surface — there are
techniques that express TC as a sum of convolutions, as a Taylor series
expansion. An early paper on this is Parker (1972).
1. What is the definition of a convolution?

2. Explain the convolution theorem.
3. Explain the basic idea of the Strang van Hees method.
4. What other approaches are there to apply the FFT method on a
curved (spherical or ellipsoidal) surface?
5. Why are bordering of the data area and tapering of the data neces-
sary?
6. In addition to geoid determination, where in physical geodesy is the
FFT method also used?
7. When computing the terrain correction on the Earth’s surface, de-
scribe the ”δ trick” used in the derivation. Why is it necessary, and
how does one make the δ vanish again?
Ó »îá .
D Statistical methods
9
D 9.1 The role of uncertainty in geophysics

In geophysics, we often obtain results based on uncertain, incomplete,
or otherwise deficient observational data. This applies also in the study
of the Earth’s gravity field: e.g., the density of gravity observations on
the Earth surface varies a lot, and large areas on 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 typically provide cover-
age of the whole globe, oceans, poles and all. They however don’t measure
at a very high resolution. Either the resolution of the method is limited
— e.g., the gravity field parameters calculated from satellite orbit per-
turbations — or the instruments measure only straight underneath the
satellite’s path — like satellite altimetry.
Another often relevant uncertainty factor is, that one can do precise
measurements on the Earth surface, but inside the Earth the uncertainty
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,
e.g., gravity anomalies are available everywhere on the Earth surface, and
with arbitrarily high resolution. In this chapter we look at mathematical
means to handle real-world situations where this is not the case.
D 9.2 Linear functionals

In mathematics, an operator or mapping that associates with every func-
tion in a given function space a certain numerical value is called a func-
tional. One such is, e.g., a (partial) derivative in a certain point:
∂
⏐
⏐
f→
↦ f ( x)⏐⏐ .
∂x x = x0
– 181 –
182 Statistical methods
A trivial functional is also the evaluation functional, the function value

itself for a certain argument value,
f ↦→ f ( x0 ) .
Other functionals are, e.g., the integral over a given area:

ˆ
f ↦→ f ( x) dx,
σ
and so on.
We may write symbolically
∂ ⏐⏐ ∂ f ⏐⏐
⏐ ⏐
L= , i.e., L { f } = .
∂ x ⏐ x= x0 ∂ x ⏐ x= x0
A functional or operator is linear if lineaariset

funktionaalit
L α f + β g = αL { f } + βL { g} , ∀α, β ∈ R.
{ }
Note that all partial derivatives, as also the Laplace operator ∆, are
linear.
In physical geodesy, all interesting functionals are functionals of the
( ) ( )⏐
function T φ, λ, R = T φ, λ, r ⏐r=R , i.e., the disturbing potential on the
Earth’s surface. The theory thus uses the spherical approximation1 ,
and the surface of the sphere, radius R , corresponds to mean sea level.
def ( )
E.g., the disturbing potential T P = T φP , λP , R in a point P at sea-level
( )
location φP , λP is such a functional:
( )
T (·, ·, R ) ↦→ T φ, λ, R .
Even if point P is not at sea level, a suitable functional exists:

( )
T (·, ·, R ) ↦→ T φ, λ, r .
Even if the quantity is not the disturbing potential, but, e.g., the gravity
anomaly or the deflection of the plumb line:
T (·, ·, R ) ↦→ ∆ g φ, λ, r ,
( )
( )
T (·, ·, R ) ↦→ ξ φ, λ, r ,
( )
T (·, ·, R ) ↦→ η φ, λ, r .
All these are also linear functionals. In fact, if we write

n
∞ ( ) n+1 ∑
∑ R ( )
T= P nm sin φ (a nm cos mλ + b nm sin mλ) ,
r
n=2 m=0
1
This is not mandatory, but the error of approximation is small.
Ó »îá .
Statistics on the Earth’s surface 183
then even the spherical-harmonic coefficients a nm , b nm are all linear

functionals of the disturbing potential T :
T (·, ·, R ) ↦−→ a nm ,
T (·, ·, R ) ↦−→ b nm .
Here, T (·, ·, R ) is shorthand for the whole function

( ) [ π π]
T φ, λ, R , φ ∈ − , + , λ ∈ [0, 2π) .
2 2
D 9.3 Statistics on the Earth’s surface

In statistics we define a stochastic process as a stochastic quantity, or
määrittelyjoukko random variable, the domain of which is a function space. In other
words, a random variable the realization values of which are functions. A
stochastic process may be a quantity developing over time, the precise
behaviour of which is uncertain, e.g., a satellite orbit. In the same way as
for a (real-valued) stochastic quantity x, we may calculate an expected
{ }
odotusarvo value or expectancy E x and a variance
{ } {[ { }]2 }
C xx = 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, e.g., the stochastic process x ( t) be a function of time. Then we may
compute its variance function as follows:
{ }
C xx ( t) = Var x ( t) .
For a stochastic process however, much more can be computed: e.g., the
covariance of values of the same function taken at different points in
time, the so-called autocovariance:
{ }
A xx ( t 1 , t 2 ) = Cov x ( t 1 ) , x ( t 2 ) =
{[ { }][ { }]}
=E x ( t1 ) − E x ( t1 ) x ( t2 ) − E x ( t2 ) .
Similarly if we have two different functions, we may compute between

them the so-called cross covariance, etc.
The argument of a stochastic process is commonly time, t. However in
geophysics we study stochastic processes, the arguments of which are
locations on the Earth’s surface, i.e., we talk of processes of the form
( )
x φ, λ . The definition of auto- and cross-covariances works otherwise
in the same way, but in case of the Earth we have a special problem.
A stochastic quantity is generally defined as a quantity x, from which
realizations x1 , x2 , x3 , . . . are obtained, which have certain statistical prop-
erties. The classical example is the dice throw. A die can be thrown again
Ó »îá .
and again, and one can practice 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 on the surface of the Earth, x φ, λ , we define
a quantity similar to the statistical expectancy E {·}, the geographic mean
¨ ˆ 2πˆ +π/2
def 1 ( ) 1 ( )
M { x} = x φ, λ d σ = x φ, λ cos φ d φ d λ. (9.1)
4π σ 4π 0 −π/2
( )
Here x φ, λ is the one and only realization of process x that we have
available on this Earth.
Clearly this definition makes sense only in the case where the statisti-
( )
cal behaviour of the process x φ, λ is the same everywhere on Earth,
( )
independently of the value of φ, λ . This is called the assumption of
homogeneity. It is in fact the assumption that the spherical symmetry of homogeenisuus
the Earth extends to the statistical behaviour of her gravity field.
Similarly to the statistical variance based on expectancy, we may define
the geographic variance:
) def )} def
= M [ x − M { x}]2 .
( { ( { }
C xx φ, λ = Var x φ, λ (9.2)
The global average of gravity anomalies ∆ g φ, λ vanishes based on their
( )
definition:
M {∆ g} = 0.
In that case, equation 9.2 is simplified as follows:
¨
2 1 )]2
C ∆ g∆ g φ, λ = Var ∆ g φ, λ = M ∆g ∆ g φ, λ
( ) { ( )} { } [ (
= d σ.
4π σ
The definition given here of the geographic mean M {·} is based on inte-
gration of the one and only realization 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 this means
E ∆ g , where ∆ g is the anomaly considered as a stochastic process, i.e.,
{ }
the series of values of ∆ g that results if we look at an infinitely long

series of randomly formed Earths. Not very practical!
If the expectancy of a stochastic process is the same as the mean of
one realization computed by integration, we speak of an ergodic process. ergodisuus
Establishing empirically that a process is ergodic is in geophysics typically
difficult to impossible.
Ó »îá .
The covariance function of the gravity field 185
α
P
Q
Earth’s centre of mass
D Figure 9.1. Definition of geocentric angular distance and azimuth.
D 9.4 The covariance function of the gravity field

Defining a covariance function between points P and Q is more compli-
cated. Something like equations 9.1, 9.2 cannot be used directly, because
both ∆ g P and ∆ g Q can move over the whole Earth’s surface. We have
∆ g P = ∆ g φP , λP ,
( )
∆ g Q = ∆ g φQ , λQ .
( )
In the following we assume that the covariance to be calculated will only

depend on the relative location of points P and Q . In a homogeneous
gravity field, the covariance function will not depend on the absolute
location of the points, but only on the difference in location between
points P and Q .
Write
( )
φQ = φQ φP , λP , ψPQ , αPQ ,
( )
λQ = λQ φP , λP , ψPQ , αPQ .
φQ , λQ can be computed2 , if we know φP , λP and both the geocentric

( ) ( )
angular distance ψPQ and the azimuth angle αPQ . See figure 9.1.
Now we may write
∆ g Q = ∆ g Q φQ φP , λP , ψPQ , αPQ , λQ φP , λP , ψPQ , αPQ

( ( ) ( ))
=
= ∆ g Q φP , λP , ψPQ , αPQ ,
( )
and we may define as the covariance function

) def
C ∆ g∆ g ψPQ , αPQ = M ∆ g P φP , λP ∆ g Q φP , λP , ψPQ , αPQ
( { ( ) ( )}
=
¨
1
∆ g P φP , λP ∆ g Q φP , λP , ψPQ , αPQ d σP .
( ) ( )
=
4π σ
2
This is called the geodetic forward problem on the sphere.
Ó »îá .
Also here, M is a geographic-mean operator. First we fix point Q in

relation to point P : both azimuth αPQ and distance ψPQ are held fixed.
The point P — and with it, point Q — is moved over the whole Earth’s
surface. We compute the corresponding integral over the unit sphere σP ,
and divide by 4π:
¨
1
C ∆ g∆ g ψPQ , αPQ = M ∆ g P ∆ g Q(P) ∆ g P ∆ g Q(P) d σP =
( ) { }
=
4π σ
ˆ +π/2ˆ 2π
1
= ∆ g P ∆ g Q(P) d λP cos φP d φP .
4π −π/2 0
In addition to the assumption of homogeneity, we may make still the

assumption of isotropy: the covariance function — or more generally, the
statistical behaviour of the gravity field — does not depend on the relative
direction or azimuth αPQ of point pair (P,Q ), but only on the angular
distance ψPQ between them. (This too is, like homogeneity, one of the
forms in which the Earth’s spherical symmetry is expressed.) In this case
we may compute the geographic mean in a slightly different way, by also
averaging over all azimuth angles αPQ ∈ [0, 2π):
ˆ 2π
1
′
∆ g P ∆ g Q(P) M ∆ g P ∆ g Q(P) d αPQ =
( ) { } { }
C ∆ g∆ g ψPQ = M =
2π 0
ˆ 2πˆ +π/2ˆ 2π
1
= ∆ g P ∆ g Q(P) d λP cos φP d φP d αPQ . (9.3)
8π2 0 −π/2 0
Remark. The true gravity field of the Earth isn’t terribly homogeneous
or isotropic, but in spite of this, both hypotheses are widely used.
D 9.5 Least-squares collocation

D 9.5.1 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 (e.g., mean neliösumman
errors) of the estimates. kollokaatio
Let s ( t) be a stochastic process, the autocovariance function of which is

( )
C t i , t j . Let the process furthermore be stationary, i.e., for any two mo-
ments in time t i , t j we have C t i , t j = C t j − t i = C (∆ t). The argument
( ) ( )
t is generally time, but could be any parameter, e.g., distance of a journey.

Of this process, we have observations made at times t 1 , t 2 , . . . , t N , when
the corresponding process values for those times are s ( t 1 ) , s ( t 2 ) , . . . , s ( t N ).
Let us assume, for the moment, that these values are error free observa-
tions. Then the observations are function values of process s, stochastic
quantities, the variance matrix of which we may write as follows (signal
Ó »îá .
Least-squares collocation 187
variance matrix):
⎡ ⎤
C ( t1 , t1 ) C ( t2 , t1 ) C ( t1 , t N )
···
⎢ .. ⎥
{ } ⎢ C ( t1 , t2 ) C ( t2 , t2 ) · · · . ⎥
Var s i =⎢ .. .. .. ..
⎥.
.
⎢ ⎥
⎣ . . . ⎦
C ( t1 , t N ) C ( t2 , t N ) · · · C ( t N , t N )
( )
We use for this the symbol C i j , both for one element C i j = C t i , t j of
[ ( ) ]
the matrix, and for the whole matrix, C i j = C t i , t j , i, j = 1, . . . , N . The
symbol s i again denotes a vector s ( t i ) , i = 1, . . . , N consisting of process
values — or one of its elements s ( t i ).
Note that, if the function C t i , t j or C (∆ t) is known, then the whole
( )
matrix and all of its elements can be calculated if only all argument
values (observation times) t i are known.
Let the shape of the problem now be, that one should estimate, i.e., predict,
the value of process s at the moment in time T , i.e., s (T ), based on our
knowledge of the above described observations s ( t i ) , i = 1, . . . , N .
In the same way as we calculated above the covariances between s ( t i )
( )
and s t j (elements of the signal variance matrix C i j ), we also compute
the covariances between s (T ) and all s ( t i ) , i = 1, . . . , N . We obtain
{ } [ ]
Cov s (T ) , s ( t i ) = C (T, t 1 ) C (T, t 2 ) · · · C (T, t N ) .
For this we may again use the notation C T j . It is assumed here, that there
is only one point in time T for which estimation is done. Generalization
to the case where there are several T p , 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 T p , s ( t i ) = ⎢ ⎥.
⎢ ⎥
.. .. .. ..
⎣ . . . . ⎦
C (T M , t 1 ) C (T M , t 2 ) · · · C (T M , t N )
For this we may use the more general notation C p j .
D 9.5.2 Signal and noise
The process s ( t) is called the signal. It is a physical phenomenon that we

are interested in. There exist also physical phenomena that are otherwise
similar, but that we are not interested in: to 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 ( t i ), we obtain in reality a value that is not absolutely
precise. The real observation thus is
ℓi = s (t i ) + ni . (9.4)
Ó »îá .
Here, n i is a stochastic quantity: observational error or noise. Let its

variance — or more precisely, the joint noise variance matrix of multiple
observations — be D i j . This is a very similar matrix to the above C i j ,
and also symmetric and positive definite. The only difference is that
D i j designates noise, which we are not interested in. Often it may be
assumed, that the errors n i , n j of two different observations ℓ i , ℓ j do not
correlate, in which case D i j is a diagonal matrix.
D 9.5.3 Estimator and variance of prediction
Now we construct an estimator

) def ∑
Λ pi ℓ i ,
(
ŝ T p =
i
a linear combination of the observations at our disposal ℓ i . The purpose

( )
in life of this estimator is to get as closely as possible to s T p . So, the
quantity to be minimized is the difference
ŝ T p − s T p = Λ pi ℓ i − s T p = Λ pi s ( t i ) + n i − s T p .
( ) ( ) ( ) [ ] ( )
∑
Here, for the sake of writing convenience, we left the summation sign
away (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:
def
Σ p p = Var ŝ T p − s T p
{ ( ) ( )}
.
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 (t i ) + ni , s t j + n j = Cov s ( t i ) , s t j + Cov n i , n j =
= Ci j + Di j,
and
Σ pq = Cov
{[ ( ) ( )] [ ( ) ( )]}
ŝ T p − s T p , ŝ T q − s T q =
= Λ pi Cov ΛTjq + Cov s T p , s T q
{[ ] [ ( ) ]} { ( ) ( )}
s (t i ) + ni , s t j + n j −
− Λ pi Cov s ( t i ) , s T q ΛTjq =
{ ( )} { ( ) ( )}
− Cov s T p , s t j
= Λ pi C i j + D i j ΛTjq + C pq − Λ pi C iq − C p j ΛTjq .
( )
(9.5)
The variances, or diagonal elements, Σ p p of the matrix are now obtained

by setting q = p.
Ó »îá .
D 9.5.4 Showing optimality
Here we show that the optimal estimator is indeed the one producing the
minimum possible variances.
Choose
def )−1
Λ p j = C pi C i j + D i j
(
.
Then, from equation 9.5, and exploiting the symmetry of the C and D
matrices, we obtain
)−1
Σ p p = C pi C i j + D i j
(
C j p + C p p−
( )−1 ( )−1
− C pi C i j + D i j C j p − C pi C i j + D i j C jp =
( )−1
= C p p − C pi C i j + D i j C j p. (9.6)
Let us study next the alternative choice

)−1
Λ p j = C pi C i j + D i j + δΛ p j .
(
In this case we obtain

)−1
Σ′p p = C p p − C pi C i j + D i j
(
C j p+
+ δΛ pi C i j + D i j ΛTj p + Λ pi C i j + D i j δΛTj p +
[( ) ] [ ( )]
+ δΛ pi C i j + D i j δΛTj p − δΛ pi C i p − C p j δΛTj p =
( )
)−1
C j p + δΛ pi C i p + C p j δΛTj p −
(
= C p p − C pi C i j + D i j
− δΛ pi C i p − C p j δΛTj p +δΛ pi C i j + D i j δΛTj p =
( )
)−1
C j p + δΛ pi C i j + D i j δΛTj p .
( ( )
= C p p − C pi C i j + D i j
Here, the last term — the only difference with result 9.6 — is positive,
because the matrices C i j and D i j are positive definite: Σ′p p > Σ p p , except
when δΛ pi = 0. In other words, the solution given above,
)−1 )−1
Λ p j = C pi C i j + D i j
( ( ) (
=⇒ ŝ T p = C pi C i j + D i j ℓ j,
is optimal in the sense of least squares — more precisely, in the sense of

minimizing the variance of prediction Σ p p , p = 1, . . . , M .
D 9.5.5 The covariance function of gravity anomalies
Least-squares collocation is used much to optimally estimate gravity

values and other functionals of the gravity field on the Earth’s surface.
If we have two points, P and Q , with measured gravity anomalies ∆ g P =
∆ g φP , λP and ∆ gQ = ∆ g φQ , λQ , we would like to have the covariance
( ) ( )
between these two anomalies,

{ }
Cov ∆ g P , ∆ gQ .
Ó »îá .
As argued in section 9.4, 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,Q . Then, we speak of an isotropic
process ∆ g φ, λ . Then also, the covariance will be
( )
{ }
Cov ∆ g P , ∆ gQ = M ′ ∆ g P ∆ g Q(P) = C ψPQ .
{ } ( )
A popular covariance function for gravity anomalies is Hirvonen3 ’s equa-

tion:
( ) C0
C ψ = , (9.7)
1 + (ψ/ψ0 )2
where C 0 = C (0) and ψ0 are parameters describing the behaviour of
the gravity field. C 0 is called the signal variance, ψ0 the correlation
length. ψ0 gives the distance at which the correlation between the gravity
anomalies in two points is still 50%.
In local applications, instead of the angular distance ψ one uses the
metric distance
s = ψR,
where R is the mean Earth radius. Then
C0
C ( s) = .
1 + (s/d )2
This equation was derived from gravimetric data for Ohio state, USA, but
it has broader validity. C (0) = C 0 , the signal variance when s = 0. Also
the variable d is called the correlation length. It is the distance d for
which C ( d ) = 12 C 0 , as seen from the equation.
The quantity C 0 varies considerably between areas, from hundreds to
thousands of mGal2 , and is largest in mountainous areas. The quantity
d is generally order of magnitude tens of km.
Warning: The Hirvonen covariance function is meant for use with (free-
air) gravity anomalies, i.e., quantities obtained by subtracting nor-
mal gravity from the measured gravity. Nowadays anomalies are
often obtained by subtracting from the observations a high-degree
“normal field”, i.e., a spherical-harmonic expansion. Then one uses
Hirvonen’s formula at one’s own risk!
Alternative functions that are also often used in local applications are the
covariance functions of first and second order autoregressive processes,
or AR(1) or AR(2) -processes:
C ψ = C 0 e−
ψ/ψ
C ψ = C 0 e−(
ψ/ψ )2 .
( ) ( )
0 or 0
3
Reino Antero Hirvonen (1908 – 1989) was a Finnish physical and mathematical
geodesist.
Ó »îá .
1
0.9
0.8
0.7
( ) 0.6
C ψ
0.5
0.4
0.3
0.2
0.1 4
2
−4 0
−2 0 −2 y
x 2 4 −4
Figure 9.2. Hirvonen’s covariance function in two dimensions. Assumed is

D C 0 = d = 1.
An AR(1) process is also called a Gauss–Markov process.
D 9.5.6 Least-squares collocation for gravity anomalies
If given are N points P i , i = 1, . . . , N , where were measured gravity values

— more precisely, anomalies — ∆ g i = ∆ g φ i , λ i , we may, like above,
( )
20
18
16
14
∆ g 12
10
8
6
4
2
35 40
30
y
25
20 25 30 35
15 15 20
10 10 x
5 5
Figure 9.3. An example of least-squares collocation. Here are given two data
points (stars); the surface plotted gives the estimated value ∆
ˆg P for
each point P in the area. Here we use least-squares collocation for
D inter- and extrapolating gravimetric data.
Ó »îá .
construct a signal variance matrix

( ) ( ) ⎤
C ψ21
· · · C ψ N1
⎡
C
( 0 ) ( )
⎢ C ψ12 · · · C ψ N2
C0 ⎥
Var ∆ g i = ⎢
{ }
⎥=
⎢ ⎥
.. .. ....
⎣ . . . . ⎦
( ) ( )
C ψ1N C ψ2N ··· C0
⎡ ⎤
C0 C 21 · · · C N1
⎢ C 12 C0 · · · C N2 ⎥
⎥ def
=⎢ ⎥ = Ci j,
⎢
.. .. .. ..
⎣ . . . . ⎦
C 1N C 2N · · · C0
( )
where all elements C ψ i j are calculated using the covariance function
9.7 given above.
If we also compute for the point P in which gravity is unknown:
) ] def
Cov ∆ g P , ∆ g i =
{ } [ ( ) ( ) (
C ψP1 C ψP2 · · · C ψP N = CP i ,
we obtain, in the same way as before, for the least-squares collocation

solution:
∆ˆg P = C P i C i j + D i j −1 ℓ j ≈ C P i C −i j1 ℓ j ,
( )
where the ℓ j = ∆ g j + n j are gravity anomaly observations made in points

j = 1, . . . , N . The matrix D i j (which we leave out of consideration here)
again describes the random observation error, observation uncertainty,
or noise n i associated with making those observations. Usually D i j is a
diagonal matrix, i.e., the observations are statistically independent and
don’t correlate with each other.
We may also compute a precision assessment of the above solution, i.e.,
the variance of prediction, equation 9.11: ennustusvarianssi
ΣPQ ≈ C PQ − C P i C −i j1 C jQ .
In the case of one unknown prediction point P , Q = P and
ΣPP = C 0 − C P i C −i j1 C jP .
Its square root √

σ∆ g P = ΣPP
is the mean error of estimator ∆
ˆg P .
D 9.5.7 Calculation example
See figure 9.4. Given two points 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. Assumed is that
Ó »îá .
y
1 (15 mGal)
30
P′
2 (20 mGal)
20
10 P
x
10 20 30
D Figure 9.4. Collocation example.
between the gravity anomalies of different points, Hirvonen’s covariance

function applies:
C0
C ( s) = , (9.8)
1 + s2/d 2
where d = 20 km and C 0 = ±1000 mGal2 . Additionally it is assumed that
the gravity measurements done (including height determination of the
gravity points!) were errorless. So, D i j = 0, i, j = 1, 2.
Calculate an estimate of the gravity anomaly ∆
ˆg P at point P and its
p
mean error σPP = ΣPP .
Calculate first the distances s and the corresponding covariances C .
s212 = (30 − 20)2 + (20 − 30)2 km2 = 200 km2 ,

( )
1000 mGal2
C 12 = C 21 = = 666.66 . . . mGal2 ,
1 + 200
400
s21P = (30 − 10)2 + (20 − 10)2 km2 = 500 km2 ,
( )
1000 mGal2
C 1P = = 444.44 . . . mGal2 ,
1 + 500
400
s22P = (20 − 10)2 + (30 − 10)2 km2 = 500 km2 ,
( )
1000 mGal2
C 2P = 500
= 444.44 . . . mGal2 .
1+ 400
From this follows

[ ] [ ]
C 11 C 12 1000 666.66
Ci j + Di j = Ci j = = mGal2 ,
C 21 C 22 666.66 1000
and its inverse matrix

[ ]
)−1 0.0018 −0.0012
mGal−2 .
(
Ci j + Di j =
−0.0012 0.0018
We also have
mGal2 .
[ ] [ ]
CP i = C P1 C P2 = 444.44 444.44
Ó »îá .
As the vector of observations is
∆ g1
[ ] [ ]
15
∆g j = = mGal,
∆ g2 20
we get the result

[ ][ ]
0.0018 −0.0012 15
∆
[ ]
ˆg P = 444.44 444.44 mGal =
−0.0012 0.0018 20
= 9.333 mGal.
Precision, i.e., the variance of prediction, equation 9.11:

)−1
ΣPP = C PP − C P i C i j + D i j
(
C jP =
[ ][ ]
0.0018 −0.0012 444.44
mGal2 =
[ ]
= C 0 − 444.44 444.44
−0.0012 0.0018 444.44
= 762.96 mGal2 ,
i.e., √
σ∆ g P = ΣPP = ±27.622 mGal.
Summarizing the result:
∆
ˆg P = 9.333 ± 27.622 mGal.
We may observe here, that the gravity anomaly estimate found is much
smaller than its own uncertainty, and thus does not differ significantly
from zero. In fact, not using the observational data at all would leave us
with the a priori estimate
p
∆
ˆg P = 0 ± 1000 mGal = 0 ± 31.623 mGal,
almost as good.
If, instead, we would choose to locate point P ′ in between points 1 and 2, at
1000 mGal2
location (25 km, 25 km) , then C P ′ 1 = C P ′ 2 = 50
= 888.89 mGal2
1 + 400
and ∆ˆg P ′ = 18.667 ± 7.201 mGal, which is clearly better than the a priori
estimate of zero.
And if we had chosen instead the Gauss–Markov covariance function
C = C 0 e− /d ,
s
we would have obtained the results ∆ ˆg P = 7.663 ± 29.272 mGal for the
original point location, and ∆
ˆg P ′ = 16.460 ± 18.426 mGal for the shifted
point location.
Ó »îá .
D 9.5.8 Theory of least-squares collocation
Above we presented one popular application of least-squares collocation.

Here we look at the method more generally. The basic equation is:
[ ]−1 [ ]
f̂ = C f g C g g + D g g g+n . (9.9)
The vector g contains observed quantities g i , the vector n contains the

observational noise, and f̂ is a vector of quantities fˆi to be predicted. The
hat is a commonly used symbol for an estimator.
Both vectors g and f̂ can, e.g., be gravity anomalies, in which case we have
homogeneous prediction, a type of interpolation or extrapolation. More
generally f̂ and g are of different type, e.g., f̂ consists of geoid heights N i ,
and g of gravity anomalies ∆ g i . 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:4
[ ] { }
Cf g ij
= M fi g j ,
[ ] { }
Cgg jk
= M g j gk ,
[ ] { }
D gg jk
= E n j nk ,
where n i , an element of vector n, represents the uncertainty of the obser-

vation process appearing in the observation equation 9.4:
ℓi = g i + ni ⇐⇒ ℓ = g + n.
ℓ is the vector of the observation values themselves, including observation

uncertainty n.
The D 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 ∆ g i ∆ g j can be as large as
{ }
1200 mGal2 , the values of E n i n j can be much smaller, depending on

{ }
the measurement technique used, e.g., as small as 0.01 mGal2 .

Not however in the case of block averages — e.g., averages over blocks of
size 1◦ × 1◦ , computed from scattered measurements — which often are
very imprecise.
The great advantage of least-squares collocation is its flexibility. Differ-
ent observation types may be handled with a single unified theory and
method, the locations of observation points (or blocks) are totally free, and
the result is obtained directly as freely choosable quantities in locations
where one wants them.
4
Note that 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
however assumed that their global geographic mean vanishes: M {f } = M {g} = 0.
Ó »îá .
D 9.6 Prediction of gravity anomalies

If the quantity to be calculated or estimated, f̂ , is of the same type as
the observed quantity, g, we often speak of (homogeneous) prediction.
E.g., the prediction equation for gravity anomalies already presented in
subsection 9.5.6 is obtained from equation 9.9 by substitution:
)−1
∆
(
ˆg P = C P i C i j + D i j ℓ j. (9.10)
Here are several points j where gravity is given: let us say, N observa-
tions ℓ j = ∆ g j + n j , j = 1, . . . , N . The number of points to be predicted may
be one, P , or also many. The matrices C i j and D i j are square, and the
inverse of their sum exists. C P i is a rectangular matrix. If there is only
one point P , it is a size 1 × N row matrix.
The prediction error is now the difference quantity5 ∆
ˆg P − ∆ g , and its
P
variance (“variance of prediction”) is ennustusvarianssi
{ }
def
ΣPP = Var ∆
ˆg P − ∆ g =
P
{} { } { }
= Var ∆ g P + Var ∆ g P − Cov ∆ g P , ∆ g P − Cov ∆ g P , ∆ g P .
{ }
ˆ ˆ ˆ
Here (propagation of variances applied to equation 9.10):

{ } )−1 (
Var ∆ g P = C P i C i j + D i j C jk + D jk (C kℓ + D kℓ )−1 C ℓP =
( )
ˆ
( )−1
= CP i C i j + D i j C ℓP
and
{ } { }
Cov ∆
ˆg P , ∆ g
P
= Cov ∆ g P , ∆
ˆg P = C P i (C i j + D i j )−1 C jP .
Here, C T T T
iP (or C jP , or C ℓP ) is the transpose of C P i . The matrix
( )−1
Ci j + Di j is symmetric and its own transpose.
The end result is (remember that the signal variance Var ∆ g P = C PP ):
{ }
ΣPP = C PP + C P i (C i j + D i j )−1 C jP −
− C P i (C i j + D i j )−1 C jP − C P i (C i j + D i j )−1 C jP =
= C PP − C P i (C i j + D i j )−1 C jP .
In case D i j ≪ C i j , we obtain a simpler, often used result:
ΣPP ≈ C PP − C P i C −i j1 C jP . (9.11)
Borderline cases:
Note that here, ∆ g P is the true value of the gravity anomaly at point P, which
5
we don’t know empirically. The measured value is ℓP = ∆ g P + nP , in which nP is

the random error or “noise” of the gravimetric observation.
Ó »îá .
Covariance function and degree variances 197
1. Point P is far from all points i . Then C P i ≈ 0 and ΣPP ≈ C PP , i.e.,

prediction is impossible in practice and the prediction equation 9.10
p
will yield the value zero. The mean error of prediction ΣPP is the
p
same as the variability C PP of the gravity anomaly signal.
2. Point P is identical with one of the points i . Then, if we use only
that point i , we obtain
1 T
ΣPP = C PP − C PP C −
PP C PP = 0,
no prediction error whatsoever — as the value at the prediction

point was already known!.
However, if D PP ̸= 0 (but small), the result is ΣPP ≈ D PP . Show.
D 9.7 Covariance function and degree variances

D 9.7.1 The covariance function of the disturbing potential
In theoretical work we use, instead of gravity anomalies, rather the

covariance function of the disturbing potential T on the Earth’s surface:
{ }
K (P,Q ) = M T P TQ(P) .
We write this in the following form using the definition of M ′ {·}, equation
9.3:
K ψPQ = M ′ T P TQ(P) =
( ) { }
ˆ 2π ˆ +π/2 ˆ 2π
1
= 2 T P TQ(P) d λP cos φP d φP d αPQ . (9.12)
8π 0 −π/2 0
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 = k nm Ynm α, ψ
n=2 m=− n
with Ynm defined as in equation 2.18.
Based on isotropy, all coefficients vanish6 for which m ̸= 0:
∞
∑ ∞
( ) ( ) def ∑ ( )
K ψ = k n0 Yn0 ψ = k n P n cos ψ .
n=2 n=2
astevarianssit The coefficients k n are called the degree variances (of the disturbing
6
because { ( )
( ) P nm cos ψ cos mα m ≥ 0,
Ynm α, ψ = ( )
P n|m| cos ψ sin | m| α m < 0,
Ó »îá .
( )
potential). For isotropic covariance functions K ψ the information
content of the degree variances k n , n = 2, 3, . . . is the same as that of the
function itself, and is in fact its spectral representation.
D 9.7.2 Degree variances and spherical-harmonic coefficients
We can in a simple way specialize the degree constituent equation 2.16:

¨ ˆ π
2n + 1 ( ) ( 2n + 1
) ( ) ( )
fn = f ψ Pn cos ψ d σ = f ψ P n cos ψ sin ψ d ψ
4π σ 2 0
if the expansion of function f is

∞
∑
( ) ( )
f ψ = f n P n cos ψ .
n=2
Comparison with the previous yields

ˆ π
2n + 1 ( ) ( )
kn = K ψ P n cos ψ sin ψ d ψ,
2 0
( )
i.e., if K ψ is given, we can calculate all k n .
( )
Substituting K ψPQ from equation 9.12 yields
ˆ +π/2ˆ 2π {ˆ πˆ 2π }
2n + 1 ( )
kn = TP TQ(P) d αPQ P n cos ψPQ sin ψPQ d ψPQ d λP cos φP d φP .
16π2 −π/2 0 0 0
Here we have already interchanged the order of the integrals, as is

allowed, and moved T P to another place.
The expression inside the curly braces is a surface integral over the unit
sphere
ˆ πˆ 2π ( )
TQ(P) P n cos ψPQ d αPQ sin ψPQ d ψPQ =
0 0 ¨
( ) def 4π
= TQ P n cos ψPQ d σQ = T n (P ) ,
σ 2n + 1
where T n is the constituent of T for the harmonic degree number n,

compare the degree constituent equation 2.16. Substitution yields
ˆ
+ /2 2π π ˆ
1
kn = TT n cos φ d λ d φ =
4π −π/2 0
¨ ¨
1 1
T n2 d σ = M T n2 ,
[ ]
= TT n d σ = M [TT n ] =
4π σ 4π σ
according to the definition of operator M , and considering the orthogonal-

ity of the functions T n .
an expression that can only be independent of α if m = 0.
Ó »îá .
Propagation of covariances 199
If we now write, with the familiar definitions,

∞
∑
( ) ( )
T φ, λ = T n φ, λ =
n=2
n
∞ ∑
∑ [ ( ) ( ) ]
= a nm P nm sin φ cos mλ + b nm P nm sin φ sin mλ =
n=2 m=0
∑∞ ∑ n
( )
= a nm Y nm φ, λ ,
n=2 m=− n
we obtain
∞ ∞ ( ¨ )
∑ ∑ 1
T n2 d σ
( ) ( ) ( )
K ψ = k n P n cos ψ = · P n cos ψ =
4π σ
n=2 n=2
∞
( n ) ∞
( n
)
∑ ∑[ 2
] ∑ ∑
a2nm + b nm a2nm
( ) ( )
= P n cos ψ = P n cos ψ .
n=2 m=0 n=2 m=− n
Here, we have exploited the orthonormality of the fully normalized basis

functions {
( ) cos mλ m≥0
Y nm = P n|m| sin φ
sin | m| λ m < 0
on the surface of unit sphere σ. One sees from the equation, that
n n
2
∑ ∑
kn = a2nm + b nm = a2nm , (9.13)
m=0 m=− n
i.e.,

The degree variances k n of the disturbing potential can be
calculated directly from the spherical-harmonic coefficients.

The literature offers many alternative notations for the degree variances,
like
def def
k n = σ2n = σTT
i .
D 9.8 Propagation of covariances

The covariance function K derived above can be used to also 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 Earth’s surface, as explained in section 9.2.
Ó »îá .
D 9.8.1 Example: upward 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 ) We know, with ylöspäin jatkaminen
the definition of T n , with equation 2.16, being
∞
( ) def ∑ ( )
T φ, λ, R = T n φ, λ ,
n=2
that
∞ ( ) n+1
( ) ∑ R ( )
T φ, λ, r = T n φ, λ .
r
n=2
Symbolically
( ) { ( )}
T φ, λ, r = L T φ, λ, R ,
Here, L is the linear operator
∞ ( ) n+1
∑ R
L{f } = f n,
r
n=2
where the f n are defined according to the degree constituent equation

2.16, so that on the surface of the sphere
∞
∑
f= f n.
n=2
Symbolically
∞
∑
L{f } = Ln f n,
n=2
where ( )n+1
n R
L =
r
is the spectral representation of the operator L.
( )
We may still write in a certain point P φP , λP , r P in space:
∞
∑
LP { f } = L nP f n ,
n=2
in which ( )n+1
R
L nP = .
rP
( )
Concretely, for the disturbing potential T φP , λP , r P in point P , this
means
( ) { ( )}
T φP , λP , r P = L P T φ, λ, R =
∞
∑
= L nP T n =
n=2
∞ ( )n+1
∑ R
= Tn.
rP
n=2
Ó »îá .
Propagation of covariances 201
Now the covariance function in space of T is obtained:

( ) { } { ( ) ( )}
K r P , r Q , ψPQ = M T P , TQ = M T φP , λP , r P T φQ , λQ , r Q =
{ { ( )} { ( )}}
= M L P T φ, λ, R L Q T φ, λ, R =
{∞ ∞ [
}
∑[ ∑ ′
]
L nP T n L nQ T n′
]
=M =
n=2 n′ =2
∞ ∑
∞
′
∑
= L nP L nQ M {T n T n′ } .
n=2 n′ =2
Based on the orthogonality of the functions T n it holds that

( )
k n P n cos ψPQ n = n′
{
if
M { T n T n′ } = ,
0 if n ̸= n′
i.e., the harmonic components of the surface covariance function

∞
∑ ∞
∑
( ) ( )
K ψPQ = M {T n T n } = k n P n cos ψPQ . (9.14)
n=2 n=2
Thus we obtain7
∞
∑
L nP L nQ k n P n cos ψPQ =
( ) ( )
K r P , r Q , ψPQ =
n=2
∞ ( )n+1 ( )n+1
∑ R R ( )
= k n P n cos ψPQ =
rP rQ
n=2
∞ ( )n+1
∑ R2 ( )
= k n P n cos ψPQ . (9.15)
r P rQ
n=2
Here we have expressed the covariance function of the disturbing poten-

( )
tial in space T φ, λ, r into an expansion into the degree variances k n
( )
of the corresponding Earth’s surface disturbing potential T φ, λ, R , by
applying propagation of covariances on the expansion 9.14 of the function
K . Thus we have obtained the three-dimensional covariance function for
the disturbing potential, needed, e.g., in mountainous countries and in
air and space applications.
D 9.8.2 Example: the covariance function of gravity anomalies
We know (equation 4.8) that there exists the following relationship be-
tween gravity anomalies and the disturbing potential:
∞ ( )n+1
1∑ R
∆g = ( n − 1) T n ,
r r
n=2
7
This works only this cleanly because in this case the operator L n is of multiplier
( )n+1
type, Rr .
Ó »îá .
symbolically: ∆ g = L g {T } for a suitable operator L g :

∞
∑
Lg {f } = L ng f n ,
n=2
where now ( )n+1

n−1 R
L ng = .
r r
Again, in a concrete point P ,
∆ g φP , λP , r P = L g,P T φ, λ, R
( ) { ( )}
=
∞
∑
= L ng,P T n =
n=2
∞ ( )n+1
∑ n−1 R
= Tn.
rP rP
n=2
Now we can show in the same way as above, that

{ } ∞
∑
Cov ∆ g P , ∆ gQ = M ∆ g P ∆ g Q = L ng,P L ng,Q M {T n T n } =
{ }
n=2
∞ ( )n+1 ( )n+1
∑ n−1 R n−1 R ( )
rP rP rQ rQ
n=2
∞ ( )n+2 (
∑ R2 n − 1 )2 ( )
= k n P n cos ψPQ .
r P rQ R
n=2
Often we write
∞ ( )n+2
) def ∑ R2
C ψPQ , r P , r Q = M ∆ g P ∆ g Q =
( { } ( )
c n P n cos ψPQ ,
r P rQ
n=2
where the degree variances of gravity anomalies are

( n − 1 )2
cn = kn.
R
Similarly we calculate also the “mixed covariances” between disturbing
potential and gravity anomaly:
{ } ∞
∑
Cov T P , ∆ gQ = M T P ∆ g Q = L nP L ng,Q M {T n T n } =
{ }
n=2
∞ ( )n+1 ( )n+1
∑ R n−1 R ( )
rP rQ rQ
n=2
∞ )n+1
R2
(
∑ n−1 ( )
= k n P n cos ψPQ .
rQ r P rQ
n=2
All these are examples of propagation of covariances, when applied to a kovarianssien

kulkeutuminen
Ó »îá .
Global covariance functions 203
series expansion:
∑
L n1,P L n2,Q M {T n T n } =
{ { } { }}
Cov L 1 T P , L 2 T Q =
n
∑
L n1,P L n2,Q k n P n cos ψPQ ,
( )
=
n
for arbitrary linear functionals

∞
∑ ∞
∑
L n1,P T n , L n2,Q T n ,
{ }
L 1 {T P } = L 2 TQ =
n=2 n=2
( )
where the T n = T n φ, λ 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 T n , and lifting the coefficient found from
the equation. These coefficients are indicated above by red and blue
colourings.
D 9.9 Global covariance functions

Empirical covariance functions have been calculated a lot. Empirical
covariance functions for the whole Earth there have been only a few.
Typically they are given in the form of a degree variance formula. The
best known is the rule observed by William Kaula8 :
k n = α n−4 .
By writing
( n − 1 )2
cn =
kn,
R
where c n are the degree variances of gravity anomalies, we obtain
α ( n − 1)2 α
cn = ≈ n−2 .
R2 n4 R2
Here, Rα2 is a planet specific constant, value about 1200 mGal2 for the
Earth.
The Kaula rule does not hold very precisely for very high degree numbers.
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 (Tscherning
and Rapp, 1974):
A ( n − 1) ( n − 1 )2
cn = = kn.
( n − 2) ( n + B) R
8
William M. Kaula (1926 – 2000) was an American geophysicist and space geode-
sist who studied the determination of the Earth’s gravity field by means of
satellite geodesy.
Ó »îá .
1010
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 k n
100
10−2
10−4
10−6
10−8
0 50 100 150 200 250 300 350 400
Degree n
Figure 9.5. Global covariance functions as degree variances. The GOCE model
D cuts off at degree 280.
The constants are, according to the authors, A = 425.28 mGal2 and B = 24

(exactly). As a technical detail, one usually chooses R = R B = 0.999R,
the radius of a Bjerhammar9 sphere inside the Earth (R is the Earth
mean radius). The form of the above equation is chosen so the covariance
functions of various quantities will be closed expressions.
D 9.10 Collocation and the spectral viewpoint

Also the calculations in least-squares collocation can 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, e.g., in
the direction of longitude on the whole Earth: nothing changes when we
turn the whole Earth by a certain angle θ around its axis of rotation: for
all longitudes, what happens is λ ↦→ λ + θ .
In the following we discuss a simplified example in two dimensions. Let
( )
observations ℓ i = g i + n i of a field g ψ , ψ ∈ [0, 2π) be given on the edge
def
of a circle, in points ψ i = 2π Ni , i = 0, 1, 2, . . . , N − 1. Let us assume that
also the results of the calculation, i.e., estimates fˆi of the result function
9
Arne Bjerhammar (1917 – 2011) was an eminent Swedish geodesist.
Ó .
»îá
Collocation and the spectral viewpoint 205
( )
f ψ are desired in the same points. Then equation 9.9 yields
[ ]−1 ( )
f̂ = C f g C g g + D g g g+n (9.16)
with
[ ] ( ( ) ( )) ( )
Cf g ij
= C f g f ψi , g ψ j = C f g ψi , ψ j ,
[ ] ( ( ) ( )) ( )
Cgg ij
= C g g g ψi , g ψ j = C g g ψi , ψ j ,
[ ] ( ( ) ( )) ( )
D gg ij
= D g g g ψi , g ψ j = D g g ψi , ψ j .
If the physics of the whole situation, including the physics of the mea-
surement process, is rotationally symmetric, we must have
N −1
[ ] { ( ) ( )} 1 ∑ ( ) ( )
Cf g ij
= M◦ f ψ i g ψ j(i) = f ψ i g ψ j(i) ,
N
i =0
in which j ( i ) = ( i + k) mod N . Here, the operator M◦ is again the “circle

average” of a function,
N −1
1 ∑ ( )
def
M ◦ { h} = h ψi ,
N
i =0
which, like the geographic average in section 9.4, replaces the statistical
average.
In the same way we obtain
N −1
[ ] { ( ) ( )} 1 ∑ ( ) ( )
Cgg ij
= M◦ g ψ i g ψ j(i) = g ψ i g ψ j(i) .
N
i =0
Now C f g , C g g are only functions of k, and we may write them
= C f g ψ i , ψ j = C f g ∆ψk = C f g [ k] ,
[ ] ( ) [ ]
Cf g ij
= C g g ψ i , ψ j = C g g ∆ψk = C g g [ k] ,
[ ] ( ) [ ]
Cgg ij
def
in which ∆ψk = ψ j − ψ i mod 2π and k = ( j − i ) mod N .
( )
Furthermore
= D g g ψ i , ψ j = D g g ∆ψk = D g g [ k] = E n i n j(i) ,
[ ] ( ) [ ] { }
D gg ij
the traditional statistical variance of the observation noise. Also because

generally the observations do not correlate with each other, we have10
D g g = σ2 I N ,
10
In fact, 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
F {I } = 1
(and yes, this extends to infinite frequencies, and contains an infinite amount of
power. . . ).
Ó »îá .
i 2
1
0
N −1
j N −2
∆ψk
ψj
ψi
D Figure 9.6. Circular geometry.
σ2 (the variance of observations, assumed equal for all) times the N × N

sized unit matrix.
Matrices of this form are called Toeplitz circulant11 . Thanks to this Toeplitz-sirkulantti
property, equation 9.16 is a string of convolutions. matriisi
Without proof we present that the spectral version of equation 9.16 looks
like this:
F Cf g
{} { }
F f̂ }F g +n =
{ }
=
F Cgg + F D gg
{ } {
F Cf g
{ }
F g+n .
{ }
= (9.17)
F Cgg + σ
{ }
2
This is an easy and rapid way to calculate the solution using FFT. In
the limit in which the observations are exact, i.e., σ2 = 0, by equation
9.17 f̂ follows straight from g + n = g. If for a suitable operator L we have
f = L {g}, the equation simplifies as follows:
{ } F { L } F {C } {
gg
F f̂ = F
}
g + n ,
F Cgg + σ
{ }
2
and if also σ2 = 0, i.e., n = 0, then

{}
F f̂ = F {L} F g
{ } { }
⇐⇒ f̂ = L g .
E.g., if the g are gravity anomalies and the f are values of the disturbing
potential, then12
R
F {L} = .
n−1
11
Otto Toeplitz (1881 – 1940) was a German Jewish mathematician who con-
tributed to functional analysis.
12
In real computation it is not so simple. . . the degree number n, which refers to
global spherical geometry, must first be converted to the Fourier wave number
expressed on the computational grid used.
Ó »îá .
The approach is called Fast Collocation, e.g., Bottoni and Barzaghi (1993).
Of course it is used in two dimensions on the Earth’s surface, though 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.
1. What is the difference between signal and noise? ⋆

(a) Signal is not a random stochastic process, whereas noise is.
(b) Signal is a stochastic process that we are interested in and
wish to estimate, while noise is a stochastic process that we
are not interested in and that we would like to filter out.
(c) Signal is a stochastic process with a greater variance and is
therefore more easily detectable than noise.
(d) Signal is a property of a real-world system, while noise is a
property of an observation instrument or method.
2. What is a functional? ⋆
(a) A mapping from a function space to a set of numbers, e.g., the
real numbers.
(b) A random-valued function.
(c) A functional associates with every (well behaved) function
defined on some domain, a number.
(d) A function of a vectorial argument.
3. What is a linear functional? ⋆
(a) A linear functional associates a number L { f } with any linear
function f { x} = a + bx defined on some domain
(b) If, for functions f and g, it holds for a functional L that
L {a f + b g} = aL { f } + bL { g}
for any real values a, b, then L is a linear functional.
(c) A linear functional associates with any (well behaved) function
defined on some domain, a linear expression L{ f } = a + bx.
4. The statistical behaviour of a stochastic process defined on the
Earth’s surface is the same independently of where on Earth you
are. This property is called isotropy | ergodicity | homogeneity |
stationarity. ⋆
5. The statistical behaviour of a stochastic process of time is the same
independently of where on the time axis you are. This property is
called isotropy | ergodicity | homogeneity | stationarity. ⋆
Ó »îá .
6. Why, in the study of the Earth’s gravity field, one uses 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 k n and c n ?
9. Describe Kaula’s rule.
10. What is a Toeplitz circulant matrix?
D Exercise 9 – 1: Variance of prediction
The equation for the variance of prediction in a point P is
ΣPP = C PP − C P i (C i j + D i j )−1 C jP ,
in which the observation points are i = 1, . . . , N . Assume there is only one
observation point, point P . Then
ΣPP = C PP − C PP (C PP + D PP )−1 C PP .
Show that, if D i j ̸= 0 but however D i j ≪ C i j ,
ΣPP ≈ D PP .
D Exercise 9 – 2: Hirvonen’s covariance formula and prediction
Hirvonen’s covariance equation is

C0
C ( s) = ,
1 + (s/d )2
with the Ohio parameters C 0 = 337 mGal2 and d = 40 km. The formula
gives the covariance between the gravity anomalies in two points P and
Q { }
C s PQ = Cov ∆ g P , ∆ gQ .
( )
s PQ is the metric distance between the points.

{ }
1. Calculate Var ∆ g P and Var ∆ gQ . Remember that according to
{ }
{ } { }
the definition Var x = Cov x, x !
{ }
2. Calculate Cov ∆ g P , ∆ gQ if s PQ = 20 km.
3. Calculate the correlation
{ }
{ }
def
Cov ∆ g P , ∆ gQ
Corr ∆ g P , ∆ gQ =√ { }.
Var ∆ g P Var ∆ gQ
{ }
Ó »îá .
Exercise 9 – 3: Predicting gravity anomalies 209
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 s PQ = 10 km from the (precisely!) given anomaly in
point P ? Apply equation 9.11 as follows:
2 1
σQQ = CQQ − CQP C −
PP C PQ .
5. And item 4 if the distance is s PQ = 80 km?
D Exercise 9 – 3: Predicting gravity anomalies
Let in two points 1 and 2 be given the measured gravity anomalies

ℓ1 = ∆ g1 + n1 and ℓ2 = ∆ g2 + n2 . The distance between the points is 80 km
and between them, at the same distance of 40 km from both, is located
point P . Compute the gravity anomaly of point P , ∆ g P by means of
prediction. The prediction equation is
)−1
∆
(
ˆg P = C P i C i j + D i j ℓ j,
where ℓ j = ∆ g j + n j is the (abstract) vector of gravity anomaly observa-

tions, ⎡ { } ⎤
Var ∆ g i Cov ∆ g i , ∆ g j
{ }
Ci j = ⎣ { } { } ⎦
Cov ∆ g i , ∆ g j Var ∆ g j
is the signal variance matrix of the vector ∆ g i , and
Cov ∆ g P , ∆ g1 Cov ∆ g P , ∆ g2
[ { } { } ]
CP i =
is the signal covariance matrix between ∆ g P and ∆ g i . D i j is the variance

matrix of the observation random uncertainty or noise n i , i = 1, 2:
[ [ ] { } ]
Var n i Cov n i , n j
Di j = { } { } .
Cov n i , n j Var n j
1. Compute the matrix C i j , assuming again Hirvonen’s covariance

formula (previous exercise) and a parameter value of d = 40 km.
2. Compute C P i .
3. Compute ∆ ˆg P expressed in the observed values ℓ1 and ℓ2 . Assume
D i j = 0 (and thus n i = 0). (Inverting the C i j matrix is possible by
hand, but just use Matlab.)
4. Compute the variance of prediction (note C jP = C T
P i ) using
σ2PP = C PP − C P i C − 1
i j C jP .
Ó »îá .
D Exercise 9 – 4: Predicting gravity anomalies (2)
Let us again have points 1 and 2 with measured gravity anomalies

ℓ1 = ∆ g1 and ℓ2 = ∆ g2 . Now however the points 1, 2 and P are in a
triangular configuration, with a right angle at point P , and the distances
from P to points 1 and 2 still 40 km. The distance between points 1 and 2
p
is now only 40 2 km.
1. Compute C i j , C P i , ∆
ˆg P and σ2 .
PP
2. Compare the result with the previous one. Conclusion?
D Exercise 9 – 5: Propagation of covariances
Given the covariance function 9.15 of the disturbing potential

∞ ( )n+1
{ } ∑ R2 ( )
Cov T P , T Q = k n P n cos ψPQ ,
r P rQ
n=2
1. calculate the covariance function of the gravity disturbance δ g

(equation 4.3). Hint: write first an expansion of form
∞
∑
δg = L nδ g T n
n=2
in order to find the expression for the coefficient L nδ g . After this
{ } ∞
∑
L nδ g,P L nδ g,Q k n P n cos ψPQ .
( )
Cov δ g P , δ gQ =
n=2
∂2 T
2. Compute the covariance function of the gravity gradient (i.e.,
∂r2
the vertical gradient of the gravity disturbance!).
D Exercise 9 – 6: Kaula’s rule for gravity gradients
For the disturbing potential

∞ ( ) n+1
( ) ∑ R ( )
T φ, λ, r = T n φ, λ (9.18)
r
n=2
or on the Earth’s surface ( r = R )

∞
∑
( ) ( )
T φ, λ, R = T n φ, λ
n=2
Kaula’s rule applies, with the degree variances
k n = α n−4 .
Ó »îá .
Exercise 9 – 7: Underground mass points 211
From these one can derive, using propagation of variances, the degree
variances of gravity anomalies
∞ ∞ (
∑ ∑ n −1)
∆g = L ng T n = Tn
R
n=2 n=2
as follows:
)2 ( n − 1 )2 α
c n = L ng n−2 .
(
kn = kn ≈
R R2
By differentiating the above expansion 9.18 for the disturbing potential
∞ ( ) n+1
∑ R
T ( r, ϕ, λ) = T n (ϕ, λ)
r
n=2
we obtain the second derivative

∞ ( )n+1
∂2 T ∑ ( n + 1) ( n + 2) R
= Tn,
∂r2 r2 r
n=2
the connection between the disturbing potential and the gravity gradient
in the spectral domain.
On the Earth’s surface r = R , or
∞ ∞
∂2 T ⏐⏐
⏐
∑ ( n + 1) ( n + 2) def
∑
= Tn = L ng g T n
∂ r 2 ⏐ r =R R2
n=2 n=2
with
( n + 1) ( n + 2)
L ng g = .
R2
1. Derive an (approximate) equation for the degree variances for the
gravity gradient, which we may call g n , in an analogue fashion as
above for the gravity anomaly degree variances c n :
g n =? · k n ≈? · n? .
2. Conclusion?
D Exercise 9 – 7: Underground mass points
1. If a mass point is placed inside the Earth at a depth D beneath

an observation point P , what then is the correlation length s of
the gravitational field it causes on the Earth’s surface, for which
C ( s) = 12 C 0 ?
2. Thus, if we wish to construct a model made of mass points, where
under each observation point ∆ g P there is one mass point, how
deep should we place them if the correlation length d is given?
Ó »îá .
D Gravimetric measurement devices
10
D 10.1 History
The first device ever built based on a pendulum was a clock. The pendu-
lum equation, √
ℓ
T = 2π ,
g
tells that the swinging time T of a pendulum of a given length is a con-
stant that depends only on the length ℓ and local gravity g, on condition
that the swings are small. The Dutch Christiaan Huygens1 built in 1657
the first useable pendulum clock based on this (http://en.wikipedia.org/
wiki/Pendulum_clock).
When the young French researcher Jean Richer2 visited French Guyana
in 1671 with a pendulum clock, he noticed that the clock ran clearly
slower. The matter was corrected simply by shortening the pendulum.
The cause of the effect could not be the climatic conditions in the tropics,
i.e., the thermal expansion of the pendulum. The right explanation was
that in the tropics, gravity g is weaker than in Europe. After return to
France, Richer had again to make his pendulum longer. The observation
is described in just one paragraph on pages 87 – 88 in his report “Obser-
vations astronomiques et physiques faites en l’isle de Caïenne”, Richer
(1731).
This is how the pendulum gravimeter was invented. Later, much more
precise special devices were built, e.g., Kater3 ’s reversion pendulum,
and the four-pendulum Von Sterneck4 device, which was also used in
1
Christiaan Huygens (1629 – 1695) was a leading Dutch natural scientist and
mathematician. Besides inventing the pendulum clock, he also was the first to
realize (in 1655) that the planet Saturn has a ring.
2
Jean Richer (1630 – 1696) was a French astronomer. He is really only remem-
bered for his pendulum finding.
3
Henry Kater (1777 – 1835) was an English physicist.
4
Robert von Sterneck (1839 – 1910) was an Austrian-Hungarian scientist.
– 213 –
214 Gravimetric measurement devices
D Figure 10.1. Jean Richer’s report.
Finland in the 1920’s and 1930’s. We must mention also the submarine
measurements, e.g., in the Java Sea by the Dutch F. A. Vening Meinesz in
which it was observed that above the trenches in the ocean floor there is syvänmeren hauta
a notable shortage of gravity, and that they thus are in a state of strong
isostatic disequilibrium (Vening Meinesz, 1928).
For production gravimetric observations, pendulum gravimeters are how-
ever too hard to operate and too slow. For that purpose the spring gravime-
ter has been developed, see section 10.2.
Pendulum gravimeters are in principle absolute measurement devices,
i.e., gravity is obtained directly as an acceleration. There are, however,
systematic effects associated with the suspension of the pendulum that
make that one cannot trust in the absoluteness of measurement after
all. One tried out solution is the very long wire pendulum, e.g., Hytönen
(1972). However, nowadays absolute measurements are made with bal-
listic gravimeters, cf. section 10.3. It has been observed that the older
measurements made with pendulum apparatus in the so-called Potsdam
system are systematically 14 mGal too large. . .
D 10.2 The relative or spring gravimeter

A spring gravimeter is at its simplest the same as a spring balance. jousivaaka
In a linear spring balance the equation of motion of the test mass is

d2ℓ
( )
m − g = − k (ℓ − ℓ0 ) ,
dt2
where m is the test mass, g the local (to be measured) gravity, k the spring
constant. The quantity ℓ0 is the “rest length” of the spring, its length if
there were no external forces acting on it. ℓ is the true, instantaneous
length of the string.
The equilibrium between the spring force and gravity is
d2ℓ ( )
= 0 ⇒ mg = k (ℓ − ℓ 0 ) = k ℓ − ℓ 0 , (10.1)
dt2
Ó »îá .
The relative or spring gravimeter 215
Figure 10.2. Autograv CG5 spring gravimeter. Image © 2011 David Monniaux,
D Wikimedia Commons, GNU Free Documentation License.
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 equilibrium
värähtely-yhtälö position. The oscillation equation, obtained by summing the above two
equations, is
d2 ( ) k( )
ℓ − ℓ = − ℓ − ℓ .
dt2 m
The period is √ √
ℓ − ℓ0 δℓ
√
m
T = 2π = 2π = 2π , (10.2)
k g g
in which δℓ = ℓ − ℓ0 denotes the difference between the equilibrium length

and the length in the state of rest, i.e., the lengthening of the spring by
gravity.
The sensitivity of the instrument is obtained by differentiating equation
10.1 in the form
( )
mg = k ℓ − ℓ0 = kδℓ
with the result

d ℓ d (δℓ) m T2
= = = 2. (10.3)
dg dg k 4π
Substitution, e.g., of δℓ = 5 cm and g = 10 m /s2 into equation 10.2 yields
T = 0.44 s. One milligal of change in gravity g produces according to
equation 10.3 a lengthening of only 5 · 10−8 m (check)! Clearly then, the
sensor observing or compensating this displacement must be extremely
sensitive!
Ó »îá .
D 10.2.1 Astatization
An astatized gravimeter offers a different measurement geometry. We

use as our example the LaCoste-Romberg gravimeter which long enjoyed
great popularity. In it, the test mass is at the end of a lever beam,
see figure 10.3. Two torques are operating on the beam, which are in
equilibrium. The torque by the spring is
( )
τs = k ℓ − ℓ0 b sin β,
where ℓ is the spring’s true, stretched equilibrium length, and ℓ0 the

theoretical or state-of-rest length without loading.
According to the sine rule
ℓ sin β = c sin 90◦ + ϵ = c cos ϵ,

( )
from which upon substitution in the previous:

( ) bc
τs = k ℓ − ℓ 0 cos ϵ.
ℓ
Gravity pulling at the mass again is mg, and the corresponding torque
τg = mg p cos ϵ.
Between these there has to be equilibrium:

( ) bc
τg − τs = mg p cos ϵ − k ℓ − ℓ0 cos ϵ = 0,
ℓ
or
( )
mg pℓ − kbc ℓ − ℓ0 = 0. (10.4)
By differentiation
mpℓ d g + mg p d ℓ − kbc d ℓ = 0
from which we obtain, by substituting equation 10.4, a sensitivity equa-

tion:
dℓ mpℓ mpℓ ℓ ℓ − ℓ0
=− =− = .
dg mg p − kbc mg p − mg p ℓ g ℓ0
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 solution5 .
Of course, levelling the instrument, with its bull’s eye level and three tasaus
rasiatasain
footscrews, is critical.
5
https://en.wikipedia.org/wiki/Spring_(device)#Zero-length_springs.
Ó »îá .
Reality
Working range
Force
Hooke’s law
Spring, length ℓ
c ( )
k ℓ − ℓ0 sin β
Length
β Test mass beam

ϵ
b
mg
Figure 10.3. Operating principle of spring gravimeter. On the right, how to

D build a “zero-length spring”.
E.g., assuming ℓ = 5 cm, ℓ0 = 0.1 cm, g = 10 m /s2 gives
dℓ
= 2.5 · 10−6 m /mGal,
dg
a 50 times6 better result than earlier! The improvement or astatization

ratio is precisely ℓ−ℓ0ℓ0 .
This is the operating principle of an astatized gravimeter, like the La-
Coste-Romberg7 .
D 10.2.2 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 10.4:
( )
mg pℓ − kbc ℓ − ℓ0 = 0, (10.5)
but for a state of disequilibrium. Then, the test mass will be undergoing
an acceleration a, and we have
m ( g − a) pℓ − kbc (ℓ − ℓ0 ) = 0,
6 p
For comparability we should still multiply by b sin β , if we measure the position
of the test mass.
7
Lucien LaCoste (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 astatized gravimeter and zero-length spring.
Ó »îá .
where, instead of the equilibrium spring length ℓ, we have the instanta-

neous length ℓ. Subtracting the above two equations yields
( ) ( )
mg p ℓ − ℓ − map ℓ − kbc ℓ − ℓ = 0.
We use equation 10.5 again to eliminate kbc, yielding

( ) ℓ ( )
mg p ℓ − ℓ − map ℓ − mg p ℓ − ℓ = 0.
ℓ − ℓ0
Rearranging terms gives
−ℓ0 ( )
map ℓ = mg p ℓ−ℓ
ℓ − ℓ0
or
g ℓ0 ( )
a= ℓ−ℓ .
ℓ ℓ − ℓ0
ℓ−ℓ0
Here we see again the “astatization ratio” 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 upward) 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.
dt2 ℓ ℓ − ℓ0 p
This is again an oscillation equation in z, with a period of
√
ℓ p ℓ − ℓ0
T = 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
T = 4. 4 s .
What this long oscillation period also means is, that the instrument is in-
sensitive to high-frequency vibrations by passing traffic, microseismicity,
etc. This is a significant operational advantage.
D 10.2.3 Practicalities of measurement
An ordinary spring gravimeter is based on elasticity. Because there is no

material that is perfectly elastic, but always also plastic8 (viscous), the
8
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, i.a., in aviation. https://en.wikipedia.org/wiki/Dislocation.
The art of making metals stronger by inhibiting the motion of dislocations,
e.g., by adding carbon to iron to form steel, forms a large part of metallurgy.
https://en.wikipedia.org/wiki/Strengthening_mechanisms_of_materials.
Ó »îá .
δ (ε)
)
F (ε
α F (ε) cos (α + δ + ε)
−ε
mg
.
mg
Figure 10.4. The idea of astatization. The elastic force of an ordinary spring
grows steeply with extension (left), whereas the weight of the
test mass is constant. The lever beam and diagonal arrangement
(right) causes the part of the force of the spring in the direction
of motion of the lever (red) to diminish with extension, while
the spring force itself grows similarly with extension. This near-
cancellation boosts sensitivity. The spring used is a zero-length
D spring.
gravimeter itself changes during the measurement process. This change

käynti is called drift. The drift is managed in practical measurements by the
following measures:
◦ we measure along lines starting from a known point and ending on

a known point, producing a closing error. The line is traversed as
rapidly as possible. The closing error is eliminated by adjusting the
values obtained from the measurement in proportion to their times
of measurement.
◦ The gravimeter is transported carefully without bumping it, and
arretointi ◦ we remember always to 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 having a

vaimennus powerful damping. This applies also for an airborne gravimeter. Both
types are mounted on a stabilized platform, keeping the axis of measure-
ment along the local vertical in spite of vehicle motions.
Ó »îá .
Vacuum pump system
Cage transporter
Prism protective cage
Falling prism
g
”Superspring”
Reference prism
Semi-transparent mirror
Laser
Mirror
Interference observation device
D Figure 10.5. Operating principle of a ballistic absolute gravimeter.
D 10.3 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
comprises a vacuum tube, inside of which an object, a prism reflecting
light, falls freely.
Here we describe shortly the JILA gravimeter, built at the University
of Colorado at Boulder by Jim Faller9 of which the Finnish Geodetic
Institute has acquired two. Figure 10.6 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 a model FG5X in 2012.
During the fall of the prism, a “cage” with a window in the bottom moves
along with the prism inside it without touching it. The purpose of the
cage is to prevent the last remaining traces of air from affecting the
motion of the prism. Approaching the bottom, the cage, which moves
along a rail under computer control, decelerates, and the prism lands
relatively softly on its bottom. 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;
9
James E. Faller (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.
Ó »îá .
The absolute or ballistic gravimeter 221
Figure 10.6. FG5 absolute gravimeter. Figure © National Oceanic and Atmos-
pheric Administration.
D
the measurements are repeated thousands of times to get a good precision

through averaging. Another prism, the reference prism, is suspended 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, e.g., the vibration
caused by the drop is controlled by a well-designed mount. Precisions are
of order several µGal, similar to what ordinary LaCoste-Romberg relative
gravimeters are capable of. The instrument is however large and, though
transportable, it 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 described 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. If we nevertheless take g to be constant,
we obtain by integration
d
z = v0 + gt,
dt
1
z = z0 + v0 t + gt2 ,
2
from which we obtain the observation equations of the measurement
Ó »îá .
process
⎡ ⎤
z0
1 2
[ ]
zi = 1 ti 2 ti
· v0 ⎦ + n i .
⎣ (10.6)
g
Here, the unknowns10 are zˆ0 , vˆ0 and ĝ. The quantities z i are the in-
terferometrically measured vertical locations of the falling prism, and
n i are the residuals of the measurements. Determining precisely the
corresponding measurement time or epoch t i is of course essential. The
volume of measurements obtained from each drop is large.
We write the observation equations in matric form:
ℓ = A x + n,
where
t21
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
z1 n1 1 t1
⎢
⎢ z2 ⎥
⎥
⎢
⎢ n2 ⎥
⎥
⎢
⎢ 1 t2 t22 ⎥
⎥ ⎡ ⎤
⎢ .. ⎥ ⎢ .. ⎥ ⎢ .. .. .. ⎥ z0
⎢ . ⎥ ⎢ . ⎥ ⎢ . . . ⎥
ℓ=⎢ ⎥, n=⎢ ⎥, A=⎢ ⎥ and x = v0 ⎦ .
⎣
⎢ zi ⎥ ⎢ ni ⎥ ⎢ 1 ti t2i ⎥
⎢
⎢ ..
⎥
⎥
⎢
⎢ ..
⎥
⎥
⎢
⎢ .. .. ..
⎥
⎥ g
⎣ . ⎦ ⎣ . ⎦ ⎣ . . . ⎦
zn nn 1 t n t2n
From this, the solution follows according to the method of least squares
adjustment, from the normal equations
A T A x̂ = A T ℓ
giving the solution (estimate)

]−1 T
x̂ = A T A
[
A ℓ.
The uncertainty of the estimates is given by the variance matrix

]−1
Var x̂ = σ2 A T A
{ } [
,
σ being the uncertainty (mean error) of a single observation, also known

as the mean error of unit weight. z i
An alternative type of absolute gravimeter throws the prism up (inside
the tube), after which it moves along a symmetric parabolic path. Such
a “rise-and-fall” instrument is, e.g., 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 case of the dropping
method. Intercomparisons between instruments of these two types have
helped to identify error sources.
10
It would be easy (exercise!) to add an unknown representing the vertical
gradient of gravity to this.
Ó »îá .
The absolute or ballistic gravimeter 223
Splitting Readout
beam beam
Mirroring
beam
Without gravity
In gravity field
t
D Figure 10.7. Principle of operation of an atomic gravimeter.
Recently also so-called atomic or quantum gravimeters have been built,

in which interferometrically the falling of individual atoms is measured
(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 an extremely cold, so-called Bose–Einstein condensate
is prepared; perhaps a million atoms in identical quantum states, with
the same phase angle like marching soldiers. The condensate is dropped,
and the first laser pulse splits it into two. Half of the atoms11 fall first
slowly, then faster; the other half fast at first and then slower. In order to
achieve this, a second laser pulse pair is used that acts like a mirror, or
perhaps a tennis racket. The third and last laser pulse is for reading out
interferometrically the phase difference between the two merging atomic
beams. The interaction between light and atoms is based on the Raman
effect.
As the atoms travel two different paths through space-time where the
gravity potential is different12 , a phase difference is formed between
these which can be measured. Without gravity (dashed lines) this phase
difference would be zero. See figure 10.7, where the horizontal axis is
time.
11
This is a quantum theoretically erroneous statement. The matter wave of each
individual atom splits into two!
12
In fact, the spinning of the atom’s phase angle acts like a clock, and the speed
at which time elapses depends on the local geopotential (Vermeer, 1983).
Ó »îá .
Figure 10.8. International intercomparison of absolute gravimeters. Image ©

D 2003 EGCS, Luxembourg.
D 10.4 Network hierarchy in gravimetry

In gravimetry, network hierarchy is just as important as in measure-
ments of location or height. The procedure has typically been, that the
highest measurement order consisted of points measured by absolute
gravimeters — in the old days this meant pendulum measurements.
Stepwise densification of this network, i.e., measurement of the base
network, was then done with relative or spring gravimeters, like also the
lowest-order measurements, gravity mapping surveys. In base network
measurement, fast transportation was used, such as aircraft: national or
regional reference points often were located at airports.
Because pendulum instruments were not genuinely absolute, the old,
so-called Potsdam system collected a 14 mGal systematic error: all values
were that much too high. Nowadays we use instead ballistic free-fall
gravimeters, the possible systematics of which are much smaller — but
not nonexistent, order of magnitude microgals. As there are no better,
i.e., more absolute, instruments than these, the issue cannot really be re-
solved. Nevertheless, international instrument intercomparisons, like the
International Intercomparison of Absolute Gravimeters, are organized
regularly and are valuable.
In Finland, regular absolute gravimetric measurements have been made
besides in Metsähovi, also in Vaasa (two points), Joensuu (two points),
Kuusamo, Sodankylä, Kevo and Eurajoki.
D 10.5 The superconducting gravimeter

This gravimeter type is based on a superconducting metal sphere levitat-
ing on a magnetic field, the precise place of which is measured electroni-
Ó »îá .
The superconducting gravimeter 225
Upper capacitor plate
Middle capac-
itor plate
Levitation coil
Feedback
coil
Lower capacitor plate
Levitation coil
Figure 10.9. Principle of operation of a superconducting gravimeter. Reading

D out the sphere position is done capacitively.
cally. Because a superconducting material is impenetrable by a magnetic

field, the sphere will remain forever in the same spot inside the field: the
Meissner effect. Of course the field itself must be constant; it is generated
käämet by superconducting solenoids inside a vessel made of mu-metal13 which
keeps out the Earth’s magnetic field.
Superconductivity in these applications still demands working at the
temperature of liquid helium (He). For this reason the device is not only
expensive, but requires also an expensive laboratory in an environment
where societal infrastructure works.
The number of these superconducting gravimeters in the world is over
twenty. One GWR 20 type instrument has worked from 1994 in Kirkko-
nummi at the Metsähovi research station of the then Finnish Geodetic
Institute, now the National Land Survey. See Virtanen and Kääriäinen
(1995), Virtanen (1998). The instrument was upgraded in 2014.
The most important property of a superconducting gravimeter is, in
addition to its precision14 , its stability, i.e., its small drift. For this reason
it is extremely suited for monitoring long period phenomena, like the
free oscillations of the solid Earth after large earthquakes 15 . Thus it is
13
http://en.wikipedia.org/wiki/Mu-metal.
14
Virtanen (2006) reports how the instrument at Metsähovi detected the change
in gravity as workmen cleared snow from its laboratory roof, including a tea
break! “Weighing” visitors to the lab by their gravitational attraction is also
standard fare.
15
Their periods range from under an hour to over twenty hours, and they are of
Ó »îá .
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 remotely
controllable superconducting gravimeters, e.g., the GWR iGrav, weighing
30 kg and not consuming any liquid helium at all. On the other hand it
needs over a kilowatt in grid power for its refrigeration system16 . Perhaps
this will lead to improvement over the current situation where the bulk
of instruments is located in Europe and North America.
D 10.6 Atmospheric influence on gravity measurement

The atmosphere has the following two effects on gravity:
1. Instrumental effects. These are due to the way the gravimeter is

constructed. By putting the instrument in a pressure chamber, one
makes these effects go away. In practice it is easier to calibrate
the instrument (in the laboratory) and calculate a correction term
according to the calibration certificate to be applied to the field
measurements.
2. Attraction of the atmosphere. This is real gravitation. It contains
irregular variations with place and time that we need to remove
from the observed gravity values.
The effect of the atmosphere can be evaluated with the aid of the
Bouguer plate approximation: if air pressure is p, then the surface
mass density of the atmosphere is
p
κ= ,
g
where g is a representative gravity value inside the atmosphere.

We don’t make a very large error by assuming
g ≈ 9.8 m /s²,
giving us on sea level17 κ ≈ 10,000 kg /m2 . The effect of the Bouguer

plate is
−2πG κ = −0.43 mGal.
Variations in air pressure affect proportionally. If the air pres-

sure disturbance is ∆ p = p − p 0 , in which p 0 is mean air pressure,
considerable geophysical interest.

16
http://www.gwrinstruments.com/igrav-gravity-sensors.html.
17
So yes, the force acting on a standard 14-inch laptop screen is 540 kg. . . fortu-
nately it’s not an oldfashioned vacuum image tube.
Ó »îá .
Airborne gravimetry and GNSS 227
1015 hP, its effect on gravity measurement will be

∆p
δ g A = −0.43 mGal.
p0
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!
3. Including the atmosphere into the mass of the Earth. This is not
a correction to be applied to gravity measurements. It is a reduc-
tion which is applied in the calculation of gravity anomalies, if we
want anomalies from which the effect of the atmosphere has been
removed.
Remember that the reference or normal gravity field of GRS80
is defined in such a way, that the parameter GM contains the
whole mass of the Earth including atmosphere, i.e., the Earth’s
gravitational field as satellites are observing it (Heikkinen, 1981).
Therefore, also when calculating gravity anomalies ∆ g, one should
reduce gravity by computationally moving the whole atmosphere
above the point of measurement to below the measurement point,
e.g., to sea level.
The total mass of the atmosphere is
p
M A = 4πκR 2 = 4π R 2 .
g
According to Newton its attraction is
GM A 4πG p
= ,
R2 g
twice the atmospheric reduction given above. At sea level, the effect
is 0.86 mGal. At height, the effect is
p (H )
0.86 mGal,
p0
in which p ( H ) and p 0 are the air pressures at height H and at sea
level, respectively.
D 10.7 Airborne gravimetry and GNSS

In the early years of the 1990s GPS, the Global Positioning System,
and more generally, satellite positioning, has changed airborne gravime-
try from a difficult technology to something completely operational. To
understand this, one must know the principle of operation of airborne
gravimetry.
An aircraft carries an airborne gravimeter, an instrument that, in the
vaimennus same way as a sea gravimeter, is strongly damped. The measurement
Ó »îá .
is done automatically, generally using electrostatic compensation. The

instrument is mounted on a stabilized platform that follows the local
vertical.
During flight, the gravimeter measures total gravity on board the aircraft,
consisting of two parts:
1. gravity proper — i.e., gravity as felt in a reference frame connected

to the Earth’s surface —, and
2. the pseudo-forces caused by the inevitable accelerations of the
aircraft even in cruise flight.
Attached to the aircraft are a number of GNSS antennas. With these,

and a geodetic GNSS instrument, the motions of the aircraft can be
monitored with centrimetre accuracy. From these can then be calculated
the pseudo-forces mentioned above under item 2.
If we measure the position of the plane (or instrument) x i at moment t i ,
∆ t = t i+1 − t i , we obtain estimated acceleration values as follows:
(x i+1 + x i−1 − 2x i )
ai ≈ . (10.7)
(∆ t)2
When the acceleration measured by the gravimeter is Γ i and the direction
of the local plumb line n i , local gravity g i follows:
g i = Γ i − 〈a i · n i 〉 .
Critical in the whole method is the choice of the time constant ∆ t. It is

best to choose it as long as possible; then, the precision of the calculated
GNSS accelerations a i is as good as possible. Also the damping of the
gravimeter is chosen in accordance with ∆ t, and the observations are
filtered digitally: all frequencies above the bound ∆ t−1 are removed,
because they are largely caused by the motions of the aircraft.
In practice, often the high-frequency part removed from the signal is
10,000 times larger than the gravity signal we are after!
If the uncertainty (mean error) of one GNSS position co-ordinate mea-
surement is σ x (and the different co-ordinates don’t correlate with each
other!), then according to equation 10.7 the uncertainty of the vertical
acceleration is p
σx 6
σa = .
(∆ t )2
Making the time interval ∆ t as long as possible without resolution suffer-
ing, requires a low flight speed. Generally a propeller aircraft or even a
helicopter is used. Of course the price of the measurement grows with
the duration of the flight — a helicopter rotor hour is expensive!
For the flight height H we choose in accordance with resolution ∆ x:
H ∼ ∆ x = v∆ t,
Ó »îá .
Measuring the gravity gradient 229
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 a radar altimeter (Ekholm et al., 1995).
After that, also other large uninhabited areas in the Arctic and Antarctic
regions have been mapped, see Brozena et al. (1996), Brozena and Peters
(1994). Already in subsection 8.6.2 on page 176 we made mention of other
large surveys. Activity continues, see Coakley et al. (2013), Kenyon et al.
(2012). The method is well suitable for large, uninhabited areas, but also,
e.g., for sea areas close to the coast where ship gravimeters would have
difficulty navigating long straight tracks. In 1999 an airborne gravimetry
campaign was undertaken over the Baltic, including the Gulf of Finland
(Jussi Kääriäinen, personal comm.).
In addition to the economic viewpoint, an important advantage of air-
borne gravimetry is, that a 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.
Also 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 5.3 on page 99), does not come into play for airborne
gravimetry.
The operating principle of satellite gravimetry, e.g., GOCE (Geopotential
and Steady-state Ocean Circulation Explorer) is similar. An essential
difference is however, that the instrumentation on the satellite is in a
state of weightlessness: Γ = 0 (in a high orbit, or when using an air
ilmanvastus drag compensation mechanism), or Γ is small and is measured using a
sensitive accelerometer (in a low orbit, where air drag is notable).
The greatest challenge in planning a satellite mission is choosing the
flight height. The lowest possible height is some 150 km. At that height,
ajoaine already a tankload of propellant is needed, or the flight will not last long.
erotuskyky However, the resolution of the measurements on the Earth’s surface is
limited, e.g., the smallest details in the Earth’s gravity field “seen” by the
GOCE satellite are 50 − 100 km in diameter.
D 10.8 Measuring the gravity gradient

The acceleration of gravity g is the gradient of the geopotential W . It
varies with place, especially close to masses. We speak of the gravity-
Ó »îá .
gradient tensor or Eötvös tensor:

⎡ ⎤
∂2 ∂2 ∂2
∂ x2 ∂ x∂ y ∂ x∂ z
def ∂2 ∂2 ∂2
⎢ ⎥
M=⎢
⎣ ∂ y∂ x ∂ y2 ∂ y∂ z
⎥ W.
⎦
∂2 ∂2 ∂2
∂ z∂ x ∂ z∂ y ∂ z2
We know that gravity increases going down, at least in free air. Going up,
gravity diminishes, about 0.3 mGal for every metre of height.
In topocentric co-ordinates ( x, y, z), where z points to the zenith, this
matrix is approximately
⎡ ⎤
−0.15 0 0
M≈ ⎣ 0 −0.15 0 ⎦ mGal /m,
0 0 0.3
2
where ∂∂ zW2 = ∂∂z g z ≈ 0.3 mGal /m is the standard value for the free-air verti-
cal gravity gradient: Newton’s law gives for a spherical Earth (the minus
sign is because g points downward while the z co-ordinate increases going
up):
GM
gz = − .
(R + z)2
Derivation gives
∂ GM ∂ (R + z) 2gz
gz = = 2 =− ≈ 3 · 10−6 m /s2 /m = 0.3 mGal /m.
∂z (R + z) 3 ∂z (R + z)
2 2
The quantities ∂∂ xW2 and ∂∂ yW2 again describe the curvatures of the equipo-
tential or level surfaces in the x and y directions, equations 3.3, 3.4:
∂2 W g ∂2 W g
=− and =− ,
∂ x2 ρ1 ∂y 2 ρ2
where ρ 1 and ρ 2 are the radii of curvature in the x and y directions.

Substitution ρ 1 = ρ 2 = R ≈ 6378 km yields
∂2 W ∂2 W
= ≈ −1.5 · 10−6 m /s2 /m = −0.15 mGal /m.
∂ x2 ∂ y2
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, as the gravity gradient as a mea-
sured quantity is very sensitive to local variations in matter density in
the Earth’s crust.
In honour of Eötvös we use as the unit of gravity gradient the Eötvös,
symbol E:
1 E = 10−9 m/s2 /m = 10−4 mGal /m.
Ó »îá .
The above tensor is now

⎡ ⎤
−1500 0 0
M ≈⎣ 0 −1500 0 ⎦ E.
0 0 3000
Note that
∂2 W ∂2 W ∂2 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, 2ω2⊕ , must be added, equation 3.1.
The gravity-gradient field of Sun and Moon is known on the Earth’s
surface as the tidal field, see section 13.1.
1. For the spring gravimeter described in section 10.2 , one milligal of

change in gravity g produces according to equation 10.3 a lengthen-
ing 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. What is, in an absolute or ballistic gravimeter, the role of:
(a) the “cage” surrounding the falling prism, and
(b) the “superspring”?
5. According to Google
◦ the Gulf War from 1990 to 1991 was the first conflict in which
the military widely used GPS,
◦ by December 1993, GPS achieved initial operational capabil-
ity (IOC), indicating a full constellation (24 satellites) was
available, and
◦ 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?
D Exercise 10 – 1: Absolute gravimeter
The observation process of absolute gravimetry is described by

1
z = z0 + v0 t + gt2 .
2
Ó »îá .
Let us assume that the distance of falling is 30 cm.
1. How much is the time of falling?

2. If we aim at an accuracy of ±10µGal, how precisely should the laser
interferometer then measure the falling distance of the prism? (A
very crude order-of-magnitude guesstimate is enough!)
3. Same question for the time registration of the falling time.
D Exercise 10 – 2: Spring gravimeter
When we use a spring gravimeter in the field, we always take care that
◦ the device is arrested during transport, i.e., the beam is clamped to

be motionless;
◦ the internal temperature of the device is kept constant by a ther-
mostat system.
The reason for this is that the functioning of a spring gravimeter depends
on the properties of the spring material, which may change as a result of
careless handling or temperature variations.
Furthermore a gravimeter always has a drift, i.e., the relationship be-
tween measured value and true value changes slowly over time. In a non-
factory fresh gravimeter this drift is however very regular and almost
linear.
As a result of the drift, a spring gravimeter cannot be used for absolute
gravity measurement and is therefore called a relative gravimeter.
Question: how is the relative nature of a spring gravimeter and its drift
taken into account
1. in planning the topology of the measurement network?

2. In planning the time order of the different measurements in a
network?
3. In the choice of vehicles and point locations?
D Exercise 10 – 3: Air pressure and gravity
1. How much does a low-pressure zone of 100 hPa (i.e., the pressure is
100 hPa lower than average air pressure 1015 hPa) affect gravity
measured on the Earth surface? (You may assume the low-pressure
zone to be very extended in area.)
2. How much does sea water rise due to the “upside-down barometer
effect” under a low-pressure zone?
3. How much does the effect from point 2 amount to in local gravity
measured on a ship? On the open sea, with a free-air gravity
Ó »îá .
Exercise 10 – 3: Air pressure and gravity 233
gradient of −0.3 mGal /m, density of sea water 1000 kg /m3 . Analyze the
situation carefully18 .
18
And I mean really carefully.
Ó »îá .
D The geoid, mean sea level,
11 sea-surface topography
D 11.1 Basic concepts

On the ocean, the geoid is on average at the same level as mean sea level,
the surface obtained by removing from the instantaneous sea surface all
periodic and quasi-periodic variations. These variations are, for example:
◦ tidal phenomena, caused by Sun and Moon, order of magnitude
±1 m, locally even more
◦ variations caused by air pressure variations (“inverted barometer
effect”). Typically of order decimetres, but up to metres under
tropical cyclones
◦ “wind pile-up”, water being pushed by winds
◦ littoral seas: variation in the volume of sweet water flowing out
from rivers into the sea
◦ eddies that are formed in the oceans in connection with, e.g., the
Gulf Stream and the Agulhas Stream (“mesoscale eddies”) that may
live for months, and inside of which the sea surface may be even
decimetres above or below that of the surroundings
◦ the continual shifting of ocean currents from one place to another
◦ 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 phenomena
worldwide. The time scale of variability ranges from two to seven
years.
If we remove all these periodic and quasi-periodic variations, we are left
with mean sea level. If the water of the seas was in a state of equilibrium,
then 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 the following phenomena:
◦ Permanent ocean currents cause, though the Coriolis force, perma-
nent differences in mean water level.
– 235 –
236 The geoid, mean sea level, sea-surface topography
◦ Also permanent differences in temperature and salinity cause per-

manent differences in mean water level, the latter, e.g., in front of
the mouths of rivers.
These physical phenomena, among others, cause the so-called sea-surface

topography, a permanent separation between sea surface and geoid.
The classical definition of the geoid is

“the level surface of the Earth’s gravity field that agrees most
closely with mean sea level.”

The practical problem with this definition is, that determining the correct
level of the geoid requires knowledge of mean sea level everywhere on the
world ocean. This is why many “geoid” models in practice don’t 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 has been computationally removed all periodic effects — but
who can know if a so-called secular effect in reality is perhaps long
period? A sensible compromise is the average sea level over 18 years —
an important periodicity, saros1 , in the orbital motion of the Moon.
The sea-surface topography again is defined as that part of the difference meritopografia
between mean sea level and the geoid, which is permanent. Also here,
the measure of permanency is the time series that are available, as tide
gauges have been widely operating already for about a century, when mareografi
again many satellite time series — TOPEX/Poseidon and its successors —
are just about a quarter of a century long. See figure 12.1.
D 11.2 Geoids and national height datums

A locally determined geoid model is generally relative. Locally, at the
current state of the art, one has no access to global mean sea level at an
acceptable precision. This may 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 global mean sea level. In the case
of Finland, the difference is about 30 cm, almost entirely caused by the
sea-surface topography in the Baltic Sea, see figure 11.4.
In Finland, heights were determined for a long time in the N60 height
system, which is tied to mean sea level in Helsinki harbour at the start
1
https://en.wikipedia.org/wiki/Saros_(astronomy).
Ó »îá .
Geoids and national height datums 237
of 1960. The reference benchmark however is located in nearby Kaivo-

puisto. Precise levelling disseminated heights from here all over Finland.
The modern Finnish height system is N2000, which is in principle tied
to sea level in Amsterdam, but 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 post-glacial land uplift, that is no longer the case:
the post-glacial land uplift varies from some four millimetres per year
in the Helsinki area to some ten millimetres per year in the area of
maximum land uplift near Ostrobothnia. This is the main reason why
in Fennoscandia height systems have a “best before” date and must be
modernized a couple of times per century.
Generally, geoid maps for practical use, like FIN2000, the Finnish geoid
model (figure 8.4), are constructed so, that they transform heights in the
national height system, e.g., N60 heights (Helmert heights) above “mean
sea level” to heights above the GRS80 reference ellipsoid. As, however,
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, several sources).
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 constructed generally in the following
way:
1. We calculate a gravimetric geoid model by using the Stokes method

and Remove-Restore, e.g., by the FFT method.
2. We fit this geoid surface solution to a number of comparison points,
in which both the height from levelling — “above sea level” — and
from the GNSS method — above the reference ellipsoid — are
known. The fit takes place, e.g., by describing the differerences by
a polynomial function:
( )
δ N = a + b (λ − λ0 ) + c ϕ − ϕ0 + ...
or something more complicated, and solving the coefficients a, b, c

from the geoid differences in the known comparison points by using
the least-squares method.
Ó »îá .
D 11.3 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, e.g., 1.6 mm /a (Wöppelmann et al., 2009).
Over the last couple of decades, the rate has accelerated and is now over
3 mm /a, see figure 12.1.
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
thermal expansion of sea water. A precise value for the eustatic rise
is hard to determine: almost all tide gauges used for monitoring sea
level have their own vertical motions, and distinguishing these from
the rise of sea level requires a representative geographic distribution
of measurement locations. Especially the ongoing response of the solid
Earth to the end of the last ice age, the latest deglaciation: so-called GIA
(Glacial Isostatic Adjustment) is a global phenomenon that it only in the
latest decades has been possible to observe by satellite positioning.
Because of eustatic sea-level rise, a distinction must be made between
absolute and relative land uplift:
Absolute land uplift is the motion of the Earth’s crust relative to the
centre of mass of the Earth. This land uplift is measured when
using satellites the orbits of which are determined in a co-ordinate
reference system tied to the Earth’s centre of mass. E.g., GNSS
positioning of tide gauges.
Relative land uplift is the motion of the Earth’s crust relative to 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 in-
ternal to the Earth from one place to another, it is clear that also
the geoid must change. The geoid rise is however small compared
to the land uplift, only a few percent of it.
d
Equation (the point above a quantity denotes the time derivative dt ):
ḣ = Ḣr + Ḣe + Ḣt + Ṅ,
in which
ḣ is the absolute land uplift,

Ḣr is the relative land uplift,
Ḣe is the eustatic (mean sea level) rise,
Ḣt is the change over time of the sea-surface topography (likely small),
Ṅ is the geoid rise.
Ó »îá .
Methods for determining the sea-surface topography 239
The rise in the geoid as a result of land uplift can be simply calculated
with the Stokes equation:
¨ ( )
dN R d
∆ g d σ.
( )
= S ψ
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 into the land uplift area in the Earth’s mantle. We may write
d dh
∆g = c ,
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 corre-

sponds to the situation where upper mantle material flows into the
space freed up underneath the rising Earth’s crust, in order to fill
it.
◦ The value −0.31 mGal /m is the opposite extreme, the “free-air hypoth-
esis”. 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 into its former volume (“rising dough model”).
Up until fairly recently, the most likely value was about −0.2 mGal /m, with
substantial uncertainty. The latest results (Mäkinen et al., 2010) give
−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 happens probably within the asthenosphere.
This problem has been studied much in the Nordic countries. The method
has been gravimetric measurement along the 63◦ N parallel (“Blue Road
Geotraverse” project). The measurement stations extend from the Norwe-
gian coast to the Russian border, and have been chosen so, that 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 pre-
cision spring or relative gravimeters. In recent years, there has been a
shift to using absolute gravimeters, obviating the need for measurement
lines.
D 11.4 Methods for determining the sea-surface topography

In principle three geodetic methods exist:
1. satellite radar altimetry and gravimetric geoid determination
Ó »îá .
Earth’s crust
Asthenosphere
(a)
Bouguer hypothesis. . .
Earth’s crust
Upper mantle
(b)
. . . and free-air hypothesis.
Figure 11.1. The two different hypotheses on the mechanism of post-glacial

D land uplift.
2. GNSS positioning along the coast (tide gauges) and gravimetric

geoid determination
3. precise levelling along the coast.
In addition to this, we still have the oceanographic method, i.e., physical

modelling. The method is termed steric levelling if temperature and
65 5 35
65
10 30
15 20 25
Föllinge
Meldal Stugun
Kopperå Vaasa Joensuu
Vågstranda Kramfors
Äänekoski
60 60
5 35
10 30
15 20 25
D Figure 11.2. The Fennoscandian 63◦ N parallel gravity line.
Ó »îá .
Global sea-surface topography and heat transport 241
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 methods have been used. A result was 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 of the Gulf of
Finland and the Bothnian Bay by some 25 − 30 cm.
Oceanographic model calculations show, that this tilt is mainly due to a
salinity gradient: in the Atlantic Ocean, salinity is 30 − 35 o /oo, when in
the Baltic it drops to 5 − 10 o /oo due to the massive production of sweet
water (Ekman, 1992). Of course on top of this come temporal variations,
like oscillations like 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.
D 11.5 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 things the study of
which is helped by precise knowledge of ocean currents: carbon dioxide
dissolved into the water, chlorophyll (phytoplankton), salinity, etc.
The Coriolis force, or acceleration, caused by the Earth’s rotation is:
a = 2 v×−
→ ,
⟨ ⟩
ω ⊕ (11.1)
in which v is the velocity vector in a system attached to the rotating

Earth, and −
→ is the rotation vector of the Earth.
ω ⊕
If a fluid flows on the Earth’s surface, then, in the above equation 11.1,
only the part of − → normal to the surface will have an effect: this part
ω ⊕
has a length of ω→
⟨− ⟩
⊕ · n = ω⊕ sin ϕ, and the vector equation 11.1 may be
replaced by a simpler scalar equation:
a = 2vω⊕ sin ϕ,
def
where a = ∥a − 〈a · n〉∥, i.e., the length of the projection of a onto the
tangent plane to the Earth, and v = ∥v∥ , ω⊕ = −
def def
 →
ω⊕  etc. 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 on
the Northern hemisphere, to the left on the Southern hemisphere.
Ó »îá .
Figure 11.3. Connection between sea-surface topography and ocean currents.

Red arrows depict the ocean currents; curves, the sea-surface
D topography.
As a result of the Coriolis force, the sea surface in the area of an ocean
current is tilted sideways with respect to the current, at an angle
a 2 v ω⊕
= sin ϕ.
γ γ
This equilibrium between Coriolis force and the horizontal gradient of

pressure is called the geostrophic equilibrium. On the equator it can be
seen from the equation that the tilt is zero, but everywhere else, ocean
currents are tilted. E.g., in case of the Gulf Stream, the height variation
caused by this effect is several decimetres. If we define a local ( x, y)
( ) ( )
co-ordinate system where 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⊕
= −2v y sin ϕ, = +2v x sin ϕ, (11.2)
∂x γ ∂y γ
As we will see in chapter 12, we can measure the location in space of

the sea surface at this precision 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 11.2 solve for the flow velocity
]T ) ]T
field2
[ [ ( ) (
v x ( x, y) v y ( x, y) = v x ϕ, λ v y ϕ, λ .
An elegant property of these equations is, that we don’t even have to know
( )
the absolute level of the field H ( x, y) = H ϕ, λ because that vanishes in
differentiation.
2
A popular, though unofficial, unit for ocean current is the sverdrup (http:
//en.wikipedia.org/wiki/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.
Ó »îá .
Global sea-surface topography and heat transport 243
Figure 11.4. Sea-surface topography map produced by GOCE. © European

Space Agency. Unit: cm. Ocean currents superimposed: NOAA /
D Rick Lumpkin (Lumpkin and Garraffo, 2005).
The described method requires a sufficiently precise geoid map of the

oceans of the world. To this need, the GOCE satellite fits like a glove, see
section 12.7. One objective of the mission was, as the name indicates, to
get a full picture of ocean currents and especially their capacity for heat
transport. This knowledge helps understand how the Earth’s climate
functions and how it is changing, also as a result of human activity. This
is for Europe and Fennoscandia, and also Finland, an important issue,
as the heat energy brought by the Gulf Stream helps to keep these areas
habitable.
Even without a geoid model we can study, using satellite altimetry, the
variations of ocean currents. It has been known for long that in the North
Atlantic Ocean, mesoscale eddies have been moving alongside the Gulf
Stream, eddies of size 10 − 100 km which show up in altimetric imagery.
Interesting is 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 (Godø et al., 2012). The life span of the eddies
can be weeks, even months.
A good, though somewhat dated, introduction into “geodetic oceanogra-
phy” is given by Rummel and Sansó (1992).
Ó »îá .
D 11.6 The global behaviour of sea level

Water exists on the Earth in three phases: liquid, ice, and vapour. During
geological history, especially the ratio between liquid water and ice has
varied substantially. Also today, a large amount of ice is tied up in
continental ice sheets, specifically Antarctica and Greenland. Of these, mannerjäätikkö
Eastern Antarctica is the overwhelmingly largest.
When the amount of water tied up in continental ice sheets varies, so
does sea level. The end of the last ice age has raised sea level by as
much as 120 m, a process that ran to completion some 6000 years ago3
(http://en.wikipedia.org/wiki/Current_sea_level_rise). Not until the last
century or two has 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 molten away, like in Fennoscandia mannerjäätikkö
and Canada (the so-called Laurentide ice sheet), the land is still rising
at an even pace, in places even 10 millimetres per year. Around the
land uplift areas, in central Europe and the United States, again takes
place a subsidence of the land at an annual pace of 1 − 1.5 mm, as directly maan vajoaminen
underneath the hard crust of the Earth or lithosphere, in the upper
mantle layer called asthenosphere, material is flowing slowly inward
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 that is what interests us, as it always
does in climate research, this Peltier correction must still be added to the
observation values.
The subsidence of the sea floor hasn’t even been globally uniform: at
the edges of the continents happens a “lever motion” when the sea floor vipuliike
subsides but dry land doesn’t. And in the tropics in the Indian and Pacific
Oceans, sea level reached 6000 years ago its maximum level, the so-called
mid-Holocene highstand, relative to the Earth’s crust. After this, local sea
level has subsided and the coral formations from that age have remained,
dead, some 2 − 3 m above modern sea level. This is how, e.g., Tuvalu and
the Maldives were formed, which are now again being threatened by
modern sea-level rise.
3
6000 years “before present”, 6 ka BP. BP conventionally means: before 1950.
Nowadays is also used b2k, before the year 2000.
Ó »îá .
The sea-level equation 245
Sea level
drops Sea level
Sea level
rises
drops
Greenland
Antarctica
Figure 11.5. The sea-level equation. Sea level reacts in a complicated way when
D continental ice sheets melt.
D 11.7 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 W. Richard
Peltier4 , 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 equation5 is
ρi [ ] ρo [ ]
S = SE + G s ⊛i I − G s ⊛i I + G s ⊛o S − G s ⊛o S , (11.3)
γ0 γ0
where
( )
◦ S = S (ω, t) = S ϕ, λ, t describes the variations of sea level as a
( )
function of place ω = ϕ, λ and time t,
◦ I = I (ω, t) is similarly a function of place and time describing the
geometry of ice sheets and glaciers,
◦ S E is the eustatic term, i.e., the variation in ice volume converted to
“equivalent global sea-level variation”, in an equation
m i ( t)
S E ( t) = ,
ρo Ao
where m i ( t) is the variation in total ice mass as a function of time,

ρ o the density of sea water, and A o the total surface area of the
oceans,
◦ ρ is the density of matter: ρ i that of ice, and ρ o that of sea water,
◦ ⊛ is the symbol of a convolution on the surface of the Earth and
the time axis, ⊛i over land ice, ⊛o over the oceans — i.e., Green’s
4
http://www.atmosp.physics.utoronto.ca/∼peltier/data.php.
5
https://geodynamics.org/cig/software/selen/selen-manual.pdf.
Ó »îá .
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 7.1, e.g.:
ˆ t ¨
G s ψ ω, ω′ , t − t ′ S ω′ , t′ d ω′ dt′ ,
{ ( ) ( )} ( )
{G s ⊛o S } (ω, t) =
−∞ ocean
( )
where ψ ω, ω′ is the geocentric angular distance between evalua-
( ) ( )
tion point ω = ϕ, λ and data point ω′ = ϕ′ , λ′ . The measure of the
( ) ( )
surface integral is d ω = N ϕ M ϕ cos ϕ d ϕ d λ, where N, M ≈ R
are the principal radii of curvature of the Earth ellipsoid. 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 ocean surface,
◦ γ0 is an average acceleration of gravity,
◦ G s is the so-called Green’s function of sea level:
G s = G V − γ0 G u ,
where the Green’s function of the geopotential is
G V ψ, t = G rV ψ, t + G eV ψ, t + G vV ψ, t
( ) ( ) ( ) ( )
where again ψ is the distance of evaluation point from integration

point, and G rV ,G eV ja G vV are the rigid, elastic and plastic (“viscous”)
partial Green’s functions of deformation. These thus describe the
rheological behaviour of the Earth, and their theoretical calcula-
tion requires the internal viscosity distribution η ( r ) of the Earth,
assuming it is isotropic, i.e., only dependent upon r .
G u ψ, t = G eu ψ, t + G vu ψ, t
( ) ( ) ( )
is again similarly Green’s function function of vertical displacement,

in the same way split into elastic and plastic parts.
The behaviour of sea level can now be computed in this way, that one
first tries to construct an “ice-load history”, i.e., I (ω, t); Then, from this
one tries to calulate iteratively, using the sea-level equation 11.3, S (ω, t).
Note that S describes relative sea-level variation, i.e., changes in the
relative positions of sea level and the Earth’s solid body or Earth’s crust.
It is a function of place: one may not assume that it would be the same
everywhere. In the article Mitrovica et al. (2001) it is shown how, e.g., the
melting water from Greenland flees to the Southern hemisphere, when
the melting water from Antarctica again comes similarly to the North.
This is a consequence from the change in the Earth’s gravity field and
the geoid, when large volumes of ice melt. Another factor is that also the
physical shape of the Earth changes, when the ice load changes: so-called
Glacial Isostatic Adjustment, GIA.
Ó »îá .
The sea-level equation 247
Figure 11.6. Sea-level rise after the last ice age (Wikimedia Commons, © Robert
D A. Rohde, GNU Free Documentation License).
This also complicates the monitoring of global mean sea level from local
measurements: the problem is familiar in Fennoscandia, as the Earth’s
crust, for now, moves up faster than global sea-level rise. . .
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, e.g.6
I
∞
( )
R γ0 ∑ ∑
G vV ψ, ∆ t = H (∆ t) k ℓ i e−sℓ i ∆ t
( ) ( )
Pℓ cos ψ ,
M
ℓ=1 i =1
where H is a step function, the “Heaviside function”, which is zero for

negative and +1 for positive arguments. The index i counts the so-called
viscous relaxation modes for every degree number ℓ; k ℓ i are “viscous load-
ing deformation coefficients” and τℓ i = 1/sℓ i correspond to the relaxation
times in which the mode in question will decay over time. Generally
the modes that are of large spatial extent — i.e., low ℓ values — decay
slower, when again the local modes — high ℓ values — tend to decay
faster, and the local modes of the last deglaciation have today already
vanished: the geographic pattern of the Fennoscandian land uplift is
already very smooth, and the seismicity accompanying the deglaciation
is pretty much over. Back then, immediately after the retreat of the ice
sheet at its edge, there were strong earthquakes, the traces of which are
visible in the landscape (Kuivamäki et al., 1998). The now dominant
6
We consider here only the plastic or viscous deformation.
Ó »îá .
viscoelastic modes are many hundreds of kilometres in geographic extent,

and correspondingly of time scales thousands of years.
1. List all the causes you are aware of of sea-level variations.

2. What is the sea-surface topography?
3. What is eustatic sea-level rise?
4. What is absolute, what relative land uplift? What does the differ-
ence between the two consist of?
5. What two main models are on offer for the mechanism of land
uplift?
6. What three geodetic techniques are available for determining the
sea-surface topography?
7. What is the shape of the sea-surface topography of the Baltic Sea,
and what is its cause?
8. What is the Coriolis force, and how does it affect ocean currents?
9. What is the geostrophic balance?
10. How can one invert a map of the sea-surface topography into a map
of ocean currents? Where on Earth does this not work?
11. What is the Peltier effect? What is the mid-Holocene highstand?
12. What does the sea-level equation describe?
13. Why does mean sea level in the Baltic Sea not rise when the Green-
land continental ice sheet melts? What will happen in the Baltic
Sea when the West Antarctic ice sheet melts?
D Exercise 11 – 1: Coriolis force, ocean current
If the velocity of flow of an ocean current is 0.1 m /s and its width 100 km,
compute:
1. How much at latitude 45◦ N is the height difference between its left
and right edges?
2. If the same current was 200 km broad and the velocity of flow
0.05 m /s (i.e., assuming the same depth, also the transport of water
would be the same), compute for that case the height difference
between the left and the right edges.
3. [For fun] if the depth of the current is 1 km, what is the water
transport in sverdrup?
D Exercise 11 – 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
Ó »îá .
Exercise 11 – 2: Land subsidence and the mechanism of land uplift 249
(figure 11.1a), rather than a free-air mechanism?
Ó »îá .
D Satellite altimetry and satellite gravity
12 missions
D 12.1 Satellite altimetry

Satellite altimetry measures, using microwave radar, the distance from a
satellite straight downward to the sea surface. Historically there have
been many satellites carrying an altimetry radar, see table 12.1, 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 rather poor. Before
that, satellite altimetry was also tested with a device on board the
orbital laboratory Skylab (1973-027A).
D Table 12.1. Altimetric satellites through the ages.
Satellite Orbital Orbital Repeat Measure- Positioning technique

inclina- height periods ment
tion (◦ ) (km) (days) precision
(m)
GEOS-3 1975 115.00 0843 − 0.200

Seasat 1978 108.00 0780 3 (17) 0.080
Geosat 1985 108.00 0780 3, 17 0.040
ERS-1 1991 098.50 0780 3, 35, 2 × 168 0.030
TOPEX/Poseidon 1992 066.00 1337 10 0.033 GPS, DORIS
ERS-2 1995 098.50 0780 35 0.030 PRARE
Geosat follow-on 1998 108.00 0800 17 0.035
Envisat 2001 098.50 0784 35 0.045 GPS, DORIS
Jason-1 2001 066.10 1336 9.9156 0.025 GPS, DORIS
Jason-2 2008 066.04 1336 9.9156 0.025 GPS, DORIS
Cryosat-2 2010 092.00 0725 369, 30 DORIS
HY-2A 2011 099.30 0970 14, 168 0.085 DORIS, GPS
SARAL/AltiKa 2013 098.50 0781 35 DORIS
Jason-3 2016 066.04 1338 9.9927 0.025 GPS, DORIS
Sentinel-3A 2016 098.62 0804 27 0.030 DORIS, SLR, GNSS
– 251 –
252 Satellite altimetry and satellite gravity missions
◦ Seasat (1978-064A) broke down only three months after launch,

probably due to a short-circuit1 . However, the data from Seasat
was the first large satellite altimetry data set used for determining
the mean sea surface, also of the Baltic Sea.
◦ Geosat (1985-021A) was a satellite launched by the U.S. Navy, in-
tended to map the gravity field on the world’s oceans, more precisely
the deflections of the plumb line, which are needed to impart the
correct departure direction to ballistic missiles launched from sub-
marines. The 17-day repeat data from the geodetic mission was
initially classified. Later however, the data from the Southern hemi-
sphere was published for scientists to use, and currently the whole
data set is public.
◦ The satellites ERS-1/2 (1991-050A, 1995-021A) and Envisat (2002-
009A) were launched by ESA, the European Space Agency. The
altimeter was just one among many packages. On the ERS satellites
a German positioning device called PRARE was along, but only on
ERS-2 it functioned after launch.
◦ TOPEX/Poseidon (1992-052A) was an American-French collab-
oration, 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 of
the sea surface geocentrically. Together with her 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 sea level over the last 20 years, about 3 mm per
year. See figure 12.1.
The famous oceanographer Walter Munk described TOPEX/Posei-
don in 2002 as “the most successful ocean experiment of all time”
(http://en.wikipedia.org/wiki/TOPEX/Poseidon).
◦ HY-2A (2011-043A) is a Chinese satellite also launched by China.
◦ SARAL/AltiKa (2013-009A) is a satellite launched by India. The
altimeter and DORIS are French contributions.
◦ Cryosat-2 (2010-013A) is a satellite launched by the ESA, the Euro-
pean Space Agency, to study polar sea ice. Especially of interest is
the so-called freeboard, the amount by which the ice sticks out of
the water. From this, the thickness, and with surface area, the total
volume may be calculated. The launch of Cryosat-1 failed. In-orbit
positioning is done with the French DORIS system.
◦ Sentinel 3A (2016-011A) is a versatile ESA remote-sensing satellite,
the first of a planned constellation. She carries several instruments,
among them SRAL, Synthetic Aperture Radar Altimeter.
1
But read this.
Ó »îá .
Satellite altimetry 253
Figure 12.1. Results from the TOPEX/Poseidon and Jason satellites. Left,
global mean sea-level rise; right, correlation between global
mean sea level and ENSO (“El Niño”). © Colorado University at
Boulder’s Sea Level Research Group http://sealevel.colorado.edu/,
D Nerem et al. (2010).
The measurement method of satellite radar altimetry is depicted in figure

12.2. Here we see all the quantities that are along in altimetry: the
measured range ℓ is the height h of the satellite above the reference
ellipsoid, corrected for the geoid height N , sea-surface topography H , and
variations of the sea surface, like tides, eddies, annual variation etc.
Furthermore, if the satellite does not contain a precise positioning device,
the true orbit of the satellite will differ from the calculated orbit – even
from the orbit calculated afterward. Therefore
h sat = h 0,sat + ∆ h,
where h 0,sat is the calculated orbit, and ∆ h the orbit-error correction.

The measurements are performed by sending thousands of pulses per
second down, measuring and averaging the travel times of the reflected
return pulses on-board the satellite down to a rate of 10 – 20 values per
second, and transmitting these to Earth. Of these values, the largest and
smallest are thrown away as possibly erroneous, and from the remainder,
a mean value is calculated for the central epoch of the pulse train using
linear regression. The value thus obtained from the regression line
is the actual “measurement”: one every second, making the effective
measurement frequency 1 Hz.
The details will vary from satellite to satellite. The pulse shape is never
quite crisp; the place of the reflection on the ocean surface, or footprint,
has a diameter of several kilometres. Especially if the ocean has wave
motion (significant wave height, SWH), then in the processing phase one
should make a careful correction so no bias is created: if the SWH is large,
also the altimeter footprint (the area on the sea from which radio energy
Ó »îá .
True orbit
Calculated orbit
h sat ℓ
footprint
Sea-surface topography H
Geoid
Reference ellipsoid
Sea surface Mean sea surface

Geoid height N
D Figure 12.2. Satellite altimetry as a measurement method; concepts.
returns to the receiver) will be larger, and the distance travelled by the
radio waves will on average be longer.
Newer satellites use an interferometric technique that differs somewhat
from the description above.
Of all the corrections related to instrumentation, atmosphere, ocean, and
solid Earth, we mention
◦ the height of sea waves (SWH)

◦ solid-Earth tides
◦ ocean tides
◦ the “wet” tropospheric propagation delay, best computed from water
vapour content measured with a water vapour radiometer on the
satellite, otherwise from an atmospheric model
◦ the “dry” tropospheric propagation delay
◦ the ionospheric delay, only for the part of the ionosphere below the
satellite, depending on flight height
◦ the altimeter’s own calibration correction. Nowadays “in-flight”
calibration is always strived for.
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
Ó »îá .
Crossover adjustment 255
experimentation, e.g., the replacement of a correction by one calculated

from improved models, etc.
D 12.2 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
sea level is the same for both satellite overflights, then this forms a
condition that can be used to adjust orbit errors.
The observation equations are
ha = N + H + ∆ h + ϵ + n,
in which ha is the altimetric measurement of the height of the sea surface,

N is the geoid height, H is the sea-surface topography, or the permanent
deviation of the sea surface from an equipotential surface, ∆ h is the
orbit-error correction, ϵ is the variability of the sea surface due to, e.g.,
the tides among other causes, and n is the noise in the radar altimetry
observations.
From this we obtain in the crossing point of tracks i and j :
def
ℓk = hai − haj = ∆ h i − ∆ h j + ϵ i − ϵ j + n i − n j .
( ) ( ) ( )
This is the observation equation of crossover adjustment. Here we see

the complication that in both sea-surface variability and orbit errors
are along in the equation in the same way. They cannot be separately
determined by crossover adjustment.
If we forget for now the sea-surface variability — or assume that it
behaves randomly, in other words it is part of the noise n —, we may
write more simply
ℓk = ∆h i − ∆h j + n k .
The index k counts crossover points, the indices i, j count tracks.

Next, we choose a suitable model for the satellite orbit error. The simplest
choice, sufficient for a small area, is the assumption that the orbit error
is a constant for each track. See a simple example, figure 12.3.
D 12.2.1 A simple example
In the figure we have three tracks and two crossing points. The observa-
tion equations, which describe the discrepancies in the known crossover
points as functions of the orbit errors, are
ℓ1 = ∆ h 2 − ∆ h 3 + n 1
ℓ2 = ∆ h 1 − ∆ h 3 + n 2
Ó »îá .
∆h2 ∆h3
∆h2 − ∆h3
∆h1
∆h1 − ∆h3
Crossover 1
Crossover 2
2
1
D Figure 12.3. A simple crossover geometry.
or in matric form2
x
ℓ A⎡  ⎤ n
[  ] [  ] ∆h1 [  ]
ℓ1 0 1 −1 ⎣ n1
= ∆h2 ⎦ + , (12.1)
ℓ2 1 0 −1 n2
∆h3
symbolically
ℓ = A x + n.
If one now tries to calculate the solution with ordinary least squares,
)−1 T
x̂ = A T A
(
A ℓ,
one will notice that this doesn’t work. The normal matrix A T 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. E.g., one
track, or, more democratically, the mean level of all tracks. This fixing is
2
Note the similarity with the observation equations for levelling! Instead of
benchmarks, we have tracks, instead of levelling lines, crossover points.
Ó »îá .
achieved by adding the following “observation equation”:

def [ ]
ℓ3 = 0 = c c c · x, (12.2)
where c is some suitable constant. Then the matrix A becomes

⎡ ⎤
0 1 −1
A = 1 0 −1 ⎦ ,
⎣
c c c
and the least-squares solution

⎡ ⎤
∆ ℓ1
ˆh1 ⎡ ⎤
⎢ ˆ ⎥ ( T )−1 T ( T )−1 T
x̂ = ⎣ ∆h2 ⎦ = A A A ℓ= A A A ⎣ ℓ2 ⎦ ,
∆
ˆh3 0
where the matrix inversion is now possible. Note that x̂ = A −1 ℓ will in

this case give the same solution, as A is square and invertible:
T
)−1 T −1 T −1 T −1
[(
T −1 T
]
ℓ = A −1 ℓ.
( ( ) )
A A A ℓ= A A A ℓ= A A A
Now the symbolic algebra system maxima3 — or brute-force calculation —

gives the readily verified inverse
⎡ ⎤−1
0 1 −1
A −1 = ⎣ 1 0 −1 ⎦ =
c c c
⎡⎡ ⎤⎡ ⎤⎤−1
1 0 1 −1
= ⎣⎣ 1 ⎦ ⎣ 1 0 −1 ⎦⎦ =
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 1 ⎦ ⎣ 1 ⎦=
3 1
−1 −1 1 c
⎡ 1 ⎤
−1 2 c
1⎣ 1 ⎦,
= 2 −1 c
3 1
−1 −1 c
and the solution is

⎡ ⎤
∆ 1
ℓ1
ˆh1 ⎡ ⎤⎡ ⎤ ⎡ ⎤
−1 2 −1 2
c ℓ
[ ]
−1 1⎣ 1 1 1
⎣ ∆ h2 ⎦ = A ℓ = 2 −1 ⎦ ⎣ ℓ2 ⎦ = ⎣ 2 −1 ⎦ ,
⎢ ˆ ⎥
3 c
1 3 ℓ2
∆
ˆh3 −1 −1 c 0 −1 −1
3
http://maxima.sourceforge.net/.
Ó »îá .
from which c has vanished.

Another way to look at this is to first write the observation equations 12.1
and 12.2 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 both sides, and both terms on the right, with the
diagonal matrix
⎡ ⎤
1 0 0
def
D=⎣ 0 1 0 ⎦.
0 0 1c
The result is
Dℓ DA Dn
⎡  ⎤   ⎤ ⎡ ⎤ ⎡  ⎤
ℓ1 ∆h1
⎡
0 1 −1 n1
⎣ ℓ2 ⎦ = ⎣ 1 0 −1 ⎦ ⎣ ∆h2 ⎦ + ⎣ n2 ⎦,
0 1 1 1 ∆h3 0
from which also c has vanished.

The principle applies generally:

Minimal constraints added to observation equations with a datum
defect do not essentially change the solution.

D 12.2.2 A more advanced orbit correction model
A more advanced representation of orbit errors more suitable for use in a

larger area, is a linear function:
∆ h = a + b τ,
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
angular distance (degrees). Now the set of observation equations for the
situation described above is
x
⎡  ⎤
a1
ℓ A n
[  ] [ 
⎢
⎢ b1 ⎥
⎥ [  ]
ℓ1 1 τ21 −1 −τ31 ⎢
]
0 0 a2 n1
⎢ ⎥
= ⎥+ .
⎥
ℓ2 1 τ12 0 0 −1 −τ32 ⎢ b2 n2
⎢
⎥
⎢ ⎥
⎣ a3 ⎦
b3
Ó »îá .
Here the design matrix contains, besides the values 1 and −1, also the
expressions τki , in which i is the number of the track, k that of the
crossover point. These are computable when the geometry of the tracks
is known.
Now there are two unknowns for every track, a and b, a constant and
a trend. Of course also this system will prove to be singular. Removing
the singularity can be done by fixing all three parameters b and one
parameter a4 .
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 error parameter per track, this transformation is
a simple parallel shift or translation of all tracks up or down.
The situation is somewhat similar as when defining a height or vertical
reference system, for a country. One needs to fix one point, e.g., Helsinki
harbour. If alternatively one fixes another point, e.g., 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 Northgoing and ten Southgoing tracks, crossing in 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. It gets complicated, but a simple approach is to attach a
priori uncertainties to the unknowns a i , b i to be estimated, e.g., from the
known uncertainties of the orbit prediction available. The least-squares
adjustment equation then becomes
)−1 T
x̂ = A T A + Σ−1
(
A ℓ,
where Σ is the diagonal matrix containing the a priori variances5 σ2a,i ,
σ2b,i of the parameters of each track i . This is referred to as Tikhonov6
regularization.
D 12.2.3 Another example
In diagram 12.4 describing a satellite altimetry geometry, there are 16

crossover points. We attempt a crossover adjustment.
4
In order to understand this, build, e.g., a three track “wire-frame model” from
pieces of iron wire, tied together by pieces of string at the crossover points.
Crossover conditions don’t in any way fix the values of the trends b, and the
whole absolute level of the frame continues to be unconstrained.
5
We assume the mean error of unit weight to be 1.
6
Andrey Nikolayevich Tikhonov (1906 – 1993) was a Russian Soviet mathemati-
cian and geophysicist.
Ó »îá .
D Figure 12.4. Example of track geometry of satellite altimetry
Questions:
1. If the orbital error ∆ h of each satellite track is described by a model

with a single bias term, how many unknowns are there?
2. If we have available 16 “observations”, i.e., crossover differences,
how many of them are redundant?
3. Is it geometrically possible to calculate this network?
4. If we fix one track in advance (so-called a priori information), how
many redundant observations are there? Can this network be
calculated?
5. If every track has two unknowns, a bias as well as an error growing
linearly with time, i.e., a trend, what then needs to be fixed in order
to make the network calculable? How many redundancies are there
then?
6. If, in case 3, we fix one track, which one would you choose? Propose
alternatively a solution where you do not have to make a choice.
Answers:
1. As many as there are tracks: 8.

2. 16 − 8 = 8.
3. No, because the absolute level of the whole network is indetermi-
nate.
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,
Ó »îá .
Choice of satellite orbit 261
then the set of allowable transformations on the whole network is
∆ h = a 00 + a 10 x + a 01 y + a 11 x y
with four degrees of freedom. So, fix one bias and three trends, not
all North- or all Southgoing.
6. Any such choice would be arbitrary. Rather use the method de-
scribed above instead, Tikhonov regularization.
D 12.2.4 Global crossover adjustment
In a global crossover adjustment, often a still more sophisticated model

is used,
∆ h = a + b sin τ + c cos τ, (12.3)
where now τ is an angular measure, e.g., the place along the track mea-
sured from the last South-North equator crossing. See Schrama (1989),
where this problem is treated more extensively. In this model, a repre-
sents the size of the orbit, while ( b, c) describe 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:
with ( X , Y , Z ) geocentric co-ordinates, we have
∆h = a0 + a1 X + a2 Y + a3 Z (12.4)
with four degrees of freedom7 .
D 12.3 Choice of satellite orbit

In choosing a satellite orbit, Kepler’s orbital laws are central. Kepler’s
third law says:
GM P 2 = 4π2 a3 , (12.5)
where a = a E + h is the satellite orbit’s semi-major axis (i.e., the mean

distance from the geocentre), while h is called the satellite’s mean height.
P is the orbital period, a E the equatorial radius of the Earth ellipsoid.
From equation 12.5 one can already infer that using satellite observations
one can precisely determine the quantity GM , the mass of the Earth
7
One could argue that, in eq. 12.3, the parameter a should be zero, as Kepler’s
third law allows a very precise determination of the orbital size, see section 12.3.
Then, also a 0 = 0 in equation 12.4.
Ó »îá .
z
Earth’s
rotation E
Perigee Nodal line
axis
y
Nodal line
ν ω
ν
Satellite i Apogee Perigee
a
Ω b
ω X
Ω̇
θ
Apsidal
Earth
line ea
x rotation
′
x (Greenwich)
(Vernal equinox)
Ascending node
Apogee
Figure 12.5. Kepler’s orbital elements: a — semi-major axis, e — eccentricity,

i — inclination, Ω — right ascension (celestial longitude) of the
D ascending node, ω — argument of perigee, and ν — true anomaly.
multiplied by Newton’s universal gravitational constant8 . The period P

can be precisely determined from long observation series, and also the
size of the orbit a can be obtained very precisely, e.g., from satellite laser
ranging (SLR) observations. For this purpose have been used, e.g., the
well known Lageos (Laser Geodynamic Satellite) satellites, which orbit
the Earth at a height of 6000 km. Ranges are nowadays obtained with
better than centimetre precision.
The orbits of altimetric satellites are chosen much lower, as is seen from
table 12.1 at the start of the chapter. The height is fine tuned using
on-board thrusters, so that the satellite passes over the same place, e.g., rakettimoottorit
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.
8
This is 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. The challenge was separating G and the mass of the Earth M,
obtaining the latter in ordinary units of mass.
Ó .
»îá
Attraction
Solar
caused by Satellite
(apparent)
Earth’s orbital
daily
flattening motion
motion
X
Daily motion of satellite

orbit ascending node
D Figure 12.6. The mechanism of a Sun-stationary orbit.
Also parameters describing the figure of the Earth affect satellite mo-
tion, e.g., the quantity J2 , the dynamic flattening, having a value of
J2 = 1082.6267 · 10−6 . It is just one of many so-called spherical-harmonic
coefficients that describe the figure of the Earth and affect satellite orbits.
In the case of J2 the effect is, that the plane of the satellite orbit rotates
at a certain rate (orbital precession), which make the satellite, if she flies
over the same location the next day, do so several minutes earlier. The
equation is, for a circular orbit of radius a:
√
dΩ 3 GM a2E
=− J2 cos i,
dt 2 a3 a2
in which again a E is the equatorial radius of the Earth reference ellipsoid,
and i the inclination of the orbital plane relative to the equator. If we
substitute numerical values into this, we obtain
dΩ cos i
= −1.318,95 · 1018
[ 3.5 −1 ]
m s ,
dt (a E + h)3.5
where h is the mean height of the satellite orbit, conventionally above a
sphere of size equatorial radius a E . If we substitute into this, e.g., the
satellite height h = 800 km (and use a E = 6,378,137 m) we obtain
dΩ ◦
= −1.331,02 · 10−6 cos i rad s−1 = −6day
.589
[ ]
· cos i. (12.6)
dt
For practical reasons — solar panels! — we often choose the satellite
orbit such, that the orbital plane turns along with the annual apparent
◦
0◦. 9856
motion of the Sun, 365.360
25 days = day . See figure 12.6.
If the inclination i is chosen in the range 96◦ − 102◦ , depending on the

orbital height, then the Earth’s dynamic flattening J2 will cause just
the suitable rotational motion of the orbital plane (“no-shadow / Sun-
synchronous / Sun-stationary orbit”), see figure 12.7.
Ó »îá .
Spring
Summer Winter
Autumn
D Figure 12.7. Geometry of a “no-shadow” orbit. Season names are boreal.
An orbit with an inclination i > 90◦ is called a retrograde orbit: the

satellite is moving Westward in longitude, opposite to the direction of the
Earth’s rotation, which is Eastward. The orbital inclination i , or for a
retrograde orbit, its supplement 180◦ − i , is also the greatest Northern
or Southern latitude a satellite can fly over. This means that, 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 made at always the same local time of day. E.g., the diurnal and semi-
diurnal tides caused by the Sun will always have the same phase angle,
and thus they cannot be observed with this kind of satellite (“resonance”).
Therefore the oceanographic satellite TOPEX/Poseidon, and her follow-up
Jason satellites, were placed in non-Sun-stationary orbits.
D 12.3.1 Example
A satellite moves in an Sun-stationary orbit, i.e., she always, day after

day, flies over the same latitude at the same local (mean) Solar time.
Questions:
1. What is the period of the satellite if she always after 14 revolutions

flies again over the same spot?
2. Same question if she always flies over the same spot after 43 revo-
lutions (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 12.5. GM = 3,986,005 · 108 m3 s−2 , and
the height of the satellite is h = a S − a, where a = 6,378,137 m.
Ó »îá .
i
3 2 1
Figure 12.8. A satellite in a retrograde orbit, 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
D by white dashed lines.
5. What is the satellite height in a “35-day orbit”? And the height

difference with the previous question?
6. What is, for the three-day orbit, the mean separation between
North-going orbital tracks (i.e., at what level of detail is the altime-
ter able to image the sea surface!)?
7. Same question for a 35-day orbit.
8. Questions for reflection:
(a) for what purpose would you use a 35-day orbit, for what pur-
pose a three-day orbit?
(b) Would it be possible, or easy, to fly both orbits with the same
satellite (see question 5)?
Answers:
1. The satellite completes 14 orbits per day, i.e., per 1440 minutes:
P = 1440
14 = 102.857 min.
2. The satellite completes 43 orbits in three days, i.e., per 3 × 1440
minutes: P = 3×43
1440
min = 100.465 min.
3. The satellite completes 502 orbits in 35 days, i.e., per 35 × 1440
minutes: P = 35×502
1440
min = 100.398 min.
4. Execute the octave code in table 12.2. The result is 780.604 km.
Ó »îá .
D Tableau 12.2. Calculating the height of a satellite from her period.
%format long
GM=3986005e8;
ae=6378137;
P=100.465*60; % seconds
fac=4*pi*pi; % four pi square
a=(GM*P*P/fac)^0.33333333;
h = a - ae;
printf(’\n\nOrbital height: %8.3f km.\n’, h/1000);
5. The same code, with P=100.398*60, yields 777.421 km. The differ-
ence with the previous is 3.183 km.
6. There are 43 orbits with different ground tracks. That means a
40,000
separation of 360
43 = 8.372 degrees, or at the equator, 43 = 930 km;
less at higher latitudes.
360 40,000
7. 502 = 0.717 degrees, or 502 = 80 km.
8. (a) The 35-day orbit would be excellent for detailed mapping. The
three-day orbit would be able to see, e.g., tides or weather-
related phenomena, but at poor 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 thrusters. So, yes.
D 12.4 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).
One reason for in-flight calibration is the circumstance that radar al-
timeters 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.
Ó »îá .
Retracking 267
Travel time
Half-
height
rule
Sent pulse Received pulse
Figure 12.9. Analyzing the altimeter return pulse. The classical return pulse
D time measurement uses the “half-height point”.
D 12.5 Retracking
The results of a satellite altimetry mission are published already during
flight in the form of a so-called geophysical data record (GDR) file, con-
taining everything related to the measurement and, e.g., atmospheric
correction terms, tidal corrections, wave parameters, etc.
It is common practice today to process again altimetry measurements al-
ready collected earlier, in order to extract further useful information. The
complete return pulse is analyzed again in a method called retracking9 .
The standard method of analysis is based on that point on leading edge
of the return pulse, which is at half height from the maximum value of
the pulse. This is according to experience a good way to get the travel
time associated with the point in 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 however two situations where this method doesn’t work prop-
erly during flight, and a more careful a posteriori analysis of the pulse is
worthwhile:
1. Archipelagos like Indonesia, Åland, . . . Here it may happen, e.g.,

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.
2. 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, i.e., how much the ice sticks out of the
water.
In both cases the traditional real-time processing on-board produces

erroneous measurements, as the travel time of the return pulse varies
too rapidly as the satellite flies on. With retracking, such measurements
have been saved, and the area covered by altimetric measurements has
been extended into the Arctic and Antarctic Oceans.
9
http://www.altimetry.info/radar-altimetry-tutorial/data-flow/data-processing/
retracking/
Ó »îá .
Figure 12.10. Ice volume on the Arctic Ocean. © PIOMASa .
a
http://psc.apl.washington.edu/research/projects/arctic-sea-ice-volume-
D anomaly/.
Freeboard is an important quantity in determining the thickness of the

ice. As the density of ice is about 920 kg /m3 and the density of sea water
about 1027 kg /m3 , the ice thickness is about 8× freeboard. If additionally
there is remote-sensing data on the area of ice cover, one can calculate
the total volume and mass of sea ice.
The Arctic ice cover has diminished radically over the last decades. How-
ever, the most radical reduction has been that of ice volume, see figure
12.10: in addition to surface area, also thickness goes down, and espe-
cially of the multi-year, thicker ice, a large part has already vanished.
D 12.6 Oceanographic research using satellite altimetry

The interest of geodesy into satellite altimetry has traditionally been into
its use for determining the geoid. Altimetric geoid determination works
only is we assume that the sea surface
1. is constant
2. coincides with a level surface, i.e., is the same as the geoid.
In practice however the ocean surface is variable in time and is also not a
level surface. Therefore, other approaches have appeared.
1. The variability of the sea surface can be studied by satellite altime-

try using three methods:
(a) Repeat tracks from the same satellite. The tracks can be
Ó »îá .
Satellite gravity missions 269
stacked and adjusted using a simple orbit error model, and

the remaining per-track residuals tell something (but, not
everything!) about the variability of the sea surface.
(b) Also the crossovers may provide information on sea-surface
variability. When the sea surface varies, the results from the
crossover adjustment will get poorer: the root-mean-square a
posteriori (after calculation) crossover difference will become
larger. Using this method to actually study sea-surface vari-
ability is more difficult: it is mostly just able to establish that
it exists, and estimate its magnitude.
(c) Nowadays altimetric satellites always carry a GNSS position-
ing instrument, providing the absolute, geocentric location of
the microwave radar device at the moment of measurement.
With it, the variations of sea level can be monitored by di-
rect measurement, assuming that both temporal and spatial
measurement densities are sufficient.
2. The deviations of sea level from a level surface — the geoid — can
be studied only, if we have access to independent information on
the true geoid surface. If dense, high-quality gravity measurements
are available for an area, this is the case, and we may estimate the
sea-surface topography.
Collecting sufficiently precise and dense gravimetric data is pos-
sible with a sea gravimeter or with airborne gravimetry. Also,
measurement with a special satellite (gravitational gradiometry,
GOCE satellite) has long been planned and was finally realized, see
section 12.7.
D 12.7 Satellite gravity missions

During the early years of the 21st century three satellites were launched
for investigating the fine structure of the Earth’s gravity field or geopo-
tential, i.e., for determining a global, high resolution model of the geoid.
CHAMP (Challenging Minisatellite Payload for Geophysical Research

and Applications, 2000-099A) was a German satellite project under
the auspices of the German Research Centre for Geosciences GFZ.
She was launched into orbit from Plesetsk, Russia, in 2000. The
orbit height of CHAMP was initially 454 km, diminishing over the
mission time to ∼ 300 km due to atmospheric drag. The orbital
inclination was 87◦ . On September 19, 2010 the satellite returned
into the atmosphere. A project description is found at the address
http://op.gfz-potsdam.de/champ.
CHAMP contained a GPS receiver. This allowed the determination
Ó »îá .
GPS-2 GPS-3
Magnetometer GPS-4
GPS-1
beam
GPS antenna
Acceleration vector
Nom
inal
orbi
t
”CHAMP” Tru
e
orb
Solar it
cells
Earth internal
mass density variations (example)
Figure 12.11. Determining the Earth’s gravity field from GPS orbital tracking
D of a low flying satellite.
of her location in space x ( t) for any moment in time t. From a

succession of such locations one may calculate the geometric accel-
eration a ( t) by differentiation:
d2
a ( t) = x ( t) .
dt2
The differentiation is done numerically in the way that was de-
scribed in the part on airborne gravimetry, equation 10.7.
The satellite also contained an accelerometer, which eliminated
the satellite accelerations caused by the atmosphere’s aerodynamic
forces, i.e., the deviations from free-fall motion. Then, only the
accelerations caused by the Earth’s gravitational field remain, from
which a precise geopotential or geoid model may be calculated by
the techniques described earlier.
A number of global geopotential models based on CHAMP data
have been calculated and published.
GRACE (Gravity Recovery And Climate Experiment Mission, 2002-
012 A and B) measured temporal changes in the Earth’s grav-
ity field at intervals of about a month extremely precisely, but
at a rather crude geographic resolution. These temporal changes
are caused by motions in the Earth’s “blue film”, i.e., 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 of a column of air and wa-
ter. The project is described at the University of Texas web site
http://www.csr.utexas.edu/grace/. It was a collaborative Ameri-
can – 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
Ó »îá .
Satellite gravity missions 271
Difference between line-of-sight accelerations
Sate
te 1 llite
li 2
Satel Distance 220 km
Precise ranging, wavelength 1.5 cm
km
450
Figure 12.12. 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
D variations in “total sea-floor pressure” (↓).
height, at an inter-satellite separation of 220 km. The orbital incli-

nation was 89◦ , i.e., 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 µm /s. Both
satellites also carried sensitive accelerometers for measuring and
eliminating the effect of atmospheric drag.
The measurement system was so sensitive, that even the movement
of a water layer of one millimetre thickness could be noticed, as long
as it extended over an area the size of a continent (some 500 km).
Published results show impressively, e.g., the wet and dry monsoons,
seasonal variations in opposite phase in the Northern and Southern
hemispheres, in the great tropical river basins: Amazonas, Congo,
the Mekong, India, Indonesia. . . see http://grace.jpl.nasa.gov/. Ani-
mation: GIF10 .
The mission ended in 2017 after 15 years, three times the planned
mission duration. A GRACE follow-on mission was launched in
2018.
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 she was launched successfully from
Plesetsk in March 2009. The orbital height was only 270 − 235 km
during the mission and the satellite contained an ionic rocket engine
10
http://commons.wikimedia.org/wiki/File:Global_Gravity_Anomaly_
Animation_over_LAND.gif
Ó »îá .
- Measurement of
GOCE satellite acceleration
1 differences
5 -
4 X
3
6
Gradiometer
Accelerometer
Unknown
density variations
Figure 12.13. Determining the Earth’s gravity field with the gravitational gra-
D diometer on the GOCE satellite.
with a stock of propellant in order to maintain the orbit against

atmospheric drag. The orbital inclination was 96.7◦ , i.e., the orbit
was Sun-stationary11 .
GOCE carried a very sensitive gravitational gradiometer, a device
that measured precisely components of the gradient of the Earth’s
attraction, i.e., the dependence of components of the attraction vec-
tor on the various co-ordinates of place. The gradiometer consisted
of six extremely sensitive, three-axes accelerometers mounted pair-
wise on a frame. The mission ended in 2013 and the satellite was
seen to burn up in the atmosphere November 11 over the Falkland
islands12 .
Theoretical analysis has shown that a gravitational gradiometer
is the best way to measure the very local features of the Earth’s
11
Because of this inclination angle, there was a cap of radius 3.3◦ at each pole
within which no measurements were obtained.
12
http://blogs.esa.int/rocketscience/2013/11/11/goce-burning-last-orbital-view/
Ó »îá .
gravity field, better than orbital tracking by GNSS. The smallest

details in the geoid map seen by GOCE are only 100 km in diameter,
and their precision is as good as ±2 cm.
With a global geoid model this precise, we may calculate the devi-
ations of the sea surface from the geoid, or equipotential or level
surface, at the same precision. We saw that the true location in
space of the sea surface is obtained from satellite radar altimetry,
also at a few centimetres precision. This separation between sea
surface and equipotential surface can again be inverted to ocean
currents, see section 11.5. This is the background for the name of
the GOCE satellite.
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 correction
exist?
4. What is, in the case of satellite altimetry, a datum defect, and how
can it be fixed?
5. How can Kepler’s third law be used to determine the mean height
of a satellite if its 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. With which three satellite altimetric methods can one study sea-
surface variability?
12. For the study of the fine structure of the Earth’s gravity field and
its temporal variability, so far three satellite missions have been
launched. Describe them and the methods used by them.
D Exercise 12 – 1: Altimetry, crossover adjustment
Given are two Northgoing satellite tracks and three Southgoing ones.
There are six crossover sites, see figure 12.14.
Ó »îá .
D Figure 12.14. Example of satellite altimetric track geometry.
1. If the orbit errors of every track are described as a linear function

of place:
∆ h = a + b τ,
how many coefficients a and b are then needed?
2. Write out the observation equations. The observations are the
crossover differences, the unknowns are the coefficients a and b for
the different tracks.
3. Can these observation equations give a unique solution? Why / why
not?
D Exercise 12 – 2: Satellite orbit
A satellite moves in a Sun-synchronous orbit, where after 419 orbits and

30 days, she again moves over exactly the same spot.
1. How much is the period of the satellite?

2. How long is the distance (West to East), in kilometres, between the
Northgoing tracks at the equator?
3. What is the highest Northern latitude that the satellite can fly
over? And. . .
4. . . . in what compass direction does the satellite fly at that point?
D Exercise 12 – 3: Kepler’s third law
What is the satellite height if the period is 98 minutes? Use Kepler’s

third law
GMP 2 = 4π2 a3S ,
Ó »îá .
Exercise 12 – 3: Kepler’s third law 275
GM = 3986005 · 108 m3 s−2 , and the height of a satellite is h S = a S − a,

where a = 6378137 m.
Ó »îá .
D Tides, atmosphere and
13 Earth crustal movements
D 13.1 Theoretical tide

We may write the tidal potential W as follows:
GMR 2 GMR 2 ( 2
)
W= P 2 (cos ζ ) + . . . = 3 cos ζ − 1 +...,
d3 2d 3
where d is the distance to either the Moon or the Sun, R the radius of
the Earth, and ζ the local zenith angle of the Sun or Moon. P2 (cos ζ) is
the Legendre polynomial of degree two. GM is the mass of the Sun or
Moon multiplied by Newton’s gravitational constant. In the case of Sun
and Moon the extra terms (. . .) can be neglected, because these are such
remote bodies: d ≫ R .
According to spherical trigonometry (cosine rule on the sphere)
cos ζ = sin φ sin δ + cos φ cos δ cos h,
Low tide
High tide
High tide
Low tide
D Figure 13.1. Theoretical tide. ζ is the local zenith angle of the Moon (or Sun).
– 277 –
278 Tides, atmosphere and Earth crustal movements
where φ is latitude, δ is the declination1 of the Moon, and h is the hour

angle2 of the Moon.
According to the spherical-harmonic addition theorem3 we have now
( )
P n (cos ζ) = P n sin φ P n (sin δ) +
n
∑ ( n − m)! ( )
+2 P nm sin φ P nm (sin δ) cos mh,
( n + m)!
m=1
or for n = 2,
( )
P2 (cos ζ) = P2 sin φ P2 (sin δ) +
1 ( )
+ P21 sin φ P21 (sin δ) cos h+
3
1 ( )
+ P22 sin φ P22 (sin δ) cos 2 h.
12
In this, according to table 2.3,
( )
P21 sin φ = 3 sin φ cos φ,
P22 sin φ = 3 cos2 φ,
( )
and we obtain
1(
3 cos2 ζ − 1 = P2 sin φ P2 (sin δ) +
) ( )
P2 (cos ζ) =
2
+ 3 sin φ cos φ sin δ cos δ cos h+
3
+ cos2 φ cos2 δ cos 2 h =
4
1( )1(
3 sin2 φ − 1 3 sin2 δ − 1 +
)
=
2 2
3
+ sin 2φ sin 2δ cos h+
4
3
+ cos2 φ cos2 δ cos 2 h.
4
From this
3 sin2 φ − 1 3 sin2 δ − 1 +
⎡ ( )( ) ⎤
2
GMR ⎣
W= + 3 sin 2φ sin 2δ cos h+ ⎦ .
4d 3
+ 3 cos2 φ cos2 δ cos 2 h
This is the so-called Laplace tidal decomposition equation.

It has three parts:
1
The declination is the geocentric latitude of the Moon.
2
The hour angle is the difference in longitude between the Moon and the local
meridian. It vanishes when the Moon is in upper culmination, i.e., due South
when seen from Northern non-tropical latitudes.
3
http://mathworld.wolfram.com/SphericalHarmonicAdditionTheorem.html.
Ó »îá .
Theoretical tide 279
1. A slowly varying part,
GMR 2 [(
3 sin2 φ − 1 3 sin2 δ − 1 ,
)( )]
W1 = 3
4d
that still depends on δ and therefore is periodic with a 14-day
(half-month) period. Again by using spherical trigonometry:
1 ) (1
2 2 2 2
sin δ = sin ϵ sin ℓ =⇒ sin δ = sin ϵ sin ℓ = sin ϵ − cos 2ℓ ,
2 2
(13.1)
where ℓ is the longitude of the Moon in its orbit, reckoned from the
ascending node (equator crossing), and ϵ is the inclination of the
Moon’s orbit with respect to the equator, on average 23◦ but rather
variable, between 18◦ .3 and 28◦ .6 . Thus we obtain
GMR 2 [( 2
)( 2
(1 1 ) )]
W1 = 3 sin φ − 1 3 sin ϵ − cos 2 ℓ −1 ,
4d 3 2 2
where we have used result 13.1. We split W1 = W1a + W1b into two
parts, a constant4 and a periodic or semi-monthly (“fortnightly”)
part:
GMR 2 [( 2
)(3 2 )]
W1a = 3 sin φ − 1 sin ϵ − 1 , (13.2)
4d 3 2
GMR 2 [( 2
)(3 2 )]
W1b = − 3 sin φ − 1 sin ϵ cos 2 ℓ .
4d 3 2
2. Additionally we have a couple of terms in which the hour angle h

appears (periods roughly a day and roughly half a day):
GMR 2 [ ]
W2 = 3 sin 2 φ sin 2 δ cos h ,
4d 3
GMR 2 [ 2 2
]
W3 = 3 cos φ cos δ cos 2 h .
4d 3
In both, we have in addition to h, still δ as a “slow” variable. These
equations could be written out as sums of various functions of the
longitude of the Moon ℓ.
Use again basic trigonometry, equation 13.1:
1 ] [1
2 2 2 2 2
cos δ = 1 − sin δ = 1 − sin ϵ sin ℓ = 1 − sin ϵ − cos 2ℓ ,
2 2
1
cos 2ℓ cos 2 h = [cos(2ℓ + 2 h) + cos(2ℓ − 2 h)] ,
2
√
sin 2δ = 2 sin δ cos δ = 2 sin δ cos2 δ =
√ [1
2 1 ]
= 2 sin ϵ sin ℓ 1 − sin ϵ − cos 2ℓ ,
2 2
4
Not precisely, because ϵ is (slowly) time dependent.
Ó »îá .
D Table 13.1. The various periods in the theoretical tide. The widely used symbols
were standardized by George Darwin.
Changing Period Darwin symbol

Name
function
Moon Sun Moon Sun
W1a - - - M0 S0 Permanent tide

W1b cos 2ℓ 14d 182d Mfa Ssab Declination tide
W2 cos h 24h 50m 24h K 1 , O1 S 1 , P1 Diurnal
W3 cos 2h 12h 25m 12h M2 S2 Semi-diurnal
a
Lunar fortnightly
b
Solar semi-annual
leading to a trigonometric expansion in ℓ, and so on. See, e.g.,

Melchior’s5 famous book Melchior (1978).
From the above equations, often the coefficient
def 3GMR 2
D= , (13.3)
4d 3
“Doodson6 ’s constant” is taken separately. The value for the Moon equals
D = 26.75 cm × g and for the Sun 12.3 cm × g. See figure 13.2.
The periods are tabulated in table 13.1 with their Darwin7 symbols.
In practice, the diurnal and semi-diurnal tides can be divided further
into many “spectral lines” close to each other, also because the lunar orbit
(like the Earth’s orbit) is an ellipse, not a circle.
D 13.2 Deformation caused by the tidal force

The tidal force, or theoretical tide, of which we spoke above, is not the
same as the deformation it causes in the solid Earth. This deformation
will depend upon the elastic properties inside the Earth. These elas-
tic properties are often described by so-called (elastic) Love8 numbers
(Melchior, 1978).
5
Paul Melchior (1925 – 2004) was an eminent Belgian geophysicist and Earth
tides researcher.
6
Arthur Thomas Doodson (1890 – 1968) was a British oceanographer, a pioneer
of tidal theory, also involved in designing machines for computing the tides. He
was stone deaf.
7
Sir George Howard Darwin (1845 – 1912) was an English astronomer and
mathematician, son of Charles Darwin of Origin of Species fame.
8
Augustus Edward Hough Love (1863 – 1940) was a British mathematician and
student of Earth elasticity.
Ó »îá .
Deformation caused by the tidal force 281
W1a , permanent W1b , fortnightly, ϵ = 23◦

80 0.2 80 0.1
60 0.1 60
40 0 40 0.05
Latitude ( ◦ )
20 −0.1 20
0
0 −0.2 0 X
−20 −0.3 −20 −0.05
−40 −0.4 −40
−60 −0.5 −60 −0.1
−80 −0.6 −80 −0.15
Obliquity ϵ, 0◦ − 90◦ Ecliptic longitude, 0◦ − 360◦
W2 , diurnal, δ = 23◦ W3 , semi-diurnal, δ = 23◦

80 80 0.6
60 0.4 60 0.4
40 0.2 40
Latitude ( ◦ )
0.2
20 20
0 0
0 0
−20 −0.2 −20 −0.2
−40 −0.4 −40 −0.4
−60 −60 −0.6
−0.6
−80 −80 −0.8
0 5 10 15 20 0 5 10 15 20
Hour angle Hour angle
Figure 13.2. The main components of the theoretical tide. These values must
D still be multiplied by Doodson’s constant D.
Let us first write the external (tidal, or generally, disturbing) potential in

the following way:
∞ ( )
∑ r n
W= Wn ,
R
n=2
in which the index n denotes the degree number of spherical harmonics.
Call the displacement of an element of matter of the solid Earth in the
radial direction, u r , in the North direction, u φ , and in the East direction,
u λ . The following equations apply:
∞
∑ Wn
ur = H n (r) ,
g
n=2
∞
1 ∑ ∂Wn
uφ = L n (r) ,
g ∂φ
n=2
∞
1 ∑ ∂Wn
uλ = L n (r) .
g cos φ ∂λ
n=2
Here r is the distance from the geocentre. It is assumed here that the
Love numbers H n , L n depend only on r , i.e., the elastic properties of the
Ó »îá .
Earth are spherically symmetric.

The deformation of the Earth causes also a change (the “indirect effect”,
in addition to the Moon’s original potential W ) in the gravity potential.
We write
∞
∑
δW = K n ( r ) Wn ,
n=2
where we use already a third type of Love numbers.
On the Earth’s surface r = R we make the following specialization:
def
h n = H n (R ) ,
def
ℓn = L n (R ) ,
def
k n = K n (R ) .
In practice, because of the large distances to Sun and Moon, the only
important part of the tidal potential W is the part for the degree number
n = 2, i.e., W2 .
The Love numbers will depend still on the frequency, i.e., the tidal period
P:
h n = h n (P ) ,
ℓn = ℓn (P ) ,
k n = k n (P ) .
The tides offer an excellent means of determining all these Love numbers
h 2 (P ) , ℓ2 (P ) and k 2 (P ) empirically, because, being periodic variations,
they cause in the Earth deformations at the same periods, but differ-
ent amplitudes and phase angles. In this way we may determine at
least those Love numbers that correspond to periods occurring in the
theoretical tide.
The h and ℓ numbers are nowadays obtained, e.g., by GNSS position-
ing. The GNSS processing software contains a built-in reduction for this
phenomenon. From gravity measurements one obtains information on
a certain linear combination of h and k (vertical displacement changes
gravity though its gradient, and deformation of the Earth, the shifting
of masses, also changes gravity). A useful research instrument is also
the long water-tube clinometer, like the tube of the Finnish Geodetic
Institute that has long been in use in the Tytyri limestone mine 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 pendu-
lum etc. A clinometer measures the change in orientation between the
Earth’s crust and the local plumb line.
Measuring the absolute direction of the plumb line, e.g., with a zenith
tube, can again give information on a certain linear combination of ℓ and
k, but only after various reductions (Earth polar motion).
Ó »îá .
The permanent part of the tide 283
D 13.3 The permanent part of the tide

As shown above, the theoretical tide equation contains a constant part
that doesn’t even vary in a long-period way. Of course the Earth responds
also to this part of the tidal force. However, because the deformation isn’t
periodic, it is not possible to measure it. And the mechanical theory of
the solid Earth, and our knowledge of the state of matter inside of the
Earth, just aren’t good enough for prediction 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). However, often, e.g.,
in the processing of GNSS observations or in defining spherical-harmonic
expansions of the Earth’s gravity field, the tidal reduction does include
this term which it is theoretically and practically impossible to know. See
Poutanen et al. (1996).
More generally we can say, that a geodetic quantity, e.g., the height of the
geoid, can be reduced for the permanent part of the tide 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, and is therefore
the best surface to use in oceanography.
◦ The effect of the gravitational field emanating from celestial objects
is removed in its entirety from the quantity, but the Earth’s defor-
mation it causes is left completely uncorrected; the quantity thus
obtained is called the “zero geoid”.
◦ Both the gravitational effect of a celestial body, and the indirect
effect of the deformation it causes, can be calculated according
to a certain deformation model (Love number), and corrected for.
The result obtained is called the “tide-free geoid”. Its problem is
precisely the empirical indeterminacy of the elasticity model used.
See figure 13.3. It is good to be critical and precisely analyze in which

way the data reduction has been done!
D 13.4 Tidal corrections between height systems

We see from equation 13.2 that, with ϵ = 23◦.5, the permanent part of the
tidal potential is equal to
GMR 2 [( 2
)(3 2 )]
Wperm = 3 sin φ − 1 sin ϵ − 1 ≈
4d 3 2
3GMR 2 ( 2 1)
≈ − 0.7615 sin φ − .
4d 3 3
Ó »îá .
Earth permanent tidal deformation Tide-free Earth crust

Direct effect on geoid of Moon and Sun Mean Earth crust
Effect of Earth permanent tidal Tide-free geoid
deformation (mass displacement) on Zero geoid
geoid Mean geoid
Reference ellipsoid
Figure 13.3. Conceptual diagram showing the constituents of the permanent

D tide.
With the combined Doodson’s constant 13.3 for Sun and Moon equal to
3GMSun R 2 3GMMoon R 2
D= 3
+ 3
= (12.3 cm + 26.75 cm) × g = 39.05 cm × g
4 d Sun 4 d Moon
we obtain (1 )
2
Wperm = 29.74 cm × − sin φ × g.
3
We can express this, with the Bruns equation 4.2, into a permanent tidal
geoid effect:
(1 )
2
Nperm = 29.74 cm × − sin φ .
3
From this, Nperm (0◦ ) = 9.91 cm on the equator, and Nperm (90◦ ) =
−29.74 cm on the poles.
This, the geoid effect of the permanent part of the external potential of
Sun and Moon, is also equal to the difference between the mean geoid
and the zero geoid as defined above:
(1 )
def
∆mean
zero N = Nmean − Nzero = 29.74 cm × 2
− sin φ .
3
For heights H above sea level, with H = h − N , we have
(1 )
def
∆mean
zero H = Hmean − Hzero = −29.74 cm × 2
− sin φ ,
3
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 = 29.74 cm × sin φ2 − sin φ1 .
( ) ( ) ( )
Ó »îá .
Loading of the Earth’s crust by sea and atmosphere 285
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 enters into the picture, we need a value for
the Love number k 2 for the permanent tidal deformation, describing the
potential of this deformation as a fraction of the original external tidal
potential. As we have seen, this number cannot really be empirically
determined; a value often used is k 2 ≈ 0.3. With this, the above equations
apply with the coefficient 29.68 cm multiplied by 1 + k 2 , yielding
(1 )
def
∆mean
tidefree H = Hmean − Htidefree = −38.66 cm × − sin2 φ ,
3
∆mean mean 2 2
tidefree H φ2 − ∆tidefree H φ1 = 38.66 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 = 8.92 cm × sin φ2 − sin φ1 .
( ) ( ) ( )
D 13.5 Loading of the Earth’s crust by sea and atmosphere

In addition to the deformation caused by the tidal force, the Earth’s crust
also deforms due to the loading by sea and atmosphere. Especially close
to the coast, the tidal motion of the sea causes a multi-period deformation
that moves the Earth’s crust up and down by as much as centimetres.
This phenomenon can be computationally modelled if the elastic proper-
ties of the solid Earth, the tidal motion of the sea, and the precise shape
of the coastline are known. One known programme for this purpose is
the package Eterna written by the German H.-G. Wenzel9 , which has been
used also in Finland.
On the other hand, when such a tool exists, then tidal loading offers
also an excellent opportunity for studying precisely the very local elastic
properties of the Earth’s crust.
For measuring the deformation, generally a registering gravimeter is
used. The Earth’s crust moves up and down, which changes gravity in
proportion to the free-air gradient value −0.3 mGal /m (Torge, 1992).
The use of GNSS for measuring the ocean tidal loading has not yet become
common.
Like the ocean, also the atmosphere causes, through changes in air
pressure, varying deformations of the Earth’s crust. The phenomenon
is very small, at most a couple of cm. Gravity measurement is not a
9
Hans-Georg Wenzel (1945 – 1999) was a German physical geodesist and geo-
physicist.
Ó »îá .
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.
1. Describe 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? Describe the parts in words.
3. What are the declination and hour angle of a celestial body, e.g., the
Moon?
4. What is Doodson’s constant?
5. What do Love numbers describe?
6. Why is it not possible to empirically determine the deformation
caused by the permanent part of the tide?
7. Describe three different ways to take the permanent part of the
tide into account when defining the geoid.
D Exercise 13 – 1: Tide
The formula for the permanent tide is
GMR 2 [( 2
)(3 2 )]
W1a = 3 sin ϕ − 1 sin ϵ − 1 ,
4d 3 2
where ϕ is latitude and ϵ is the obliquity of the Earth’s axis of rotation,
currently about 23◦ .
1. For what value ϕ the permanent tide vanishes? Interpretation?

2. For what value ϵ the permanent tide vanishes? Interpretation?
Ó »îá .
D Earth gravity field research
14
D 14.1 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. 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 IAG service of great reputation is the International Gravity
Bureau, BGI, Bureau Gravimétrique International, located in Toulouse,
France (http://bgi.omp.obs-mip.fr/). The bureau works as an international
broker to which countries can submit their gravimetric materials. If some
researcher needs gravimetric material from another country, e.g., for
geoid computation, he 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 already operated since 1992 under
the name International Geoid Service (IGeS), the executive arm of the
International Geoid Commission (IGeC). The ISG office is located in
Milano (http://www.isgeoid.polimi.it/) also with substantial support by the
Italian state. The task of this service is to support geoid determination
in different countries, for which purpose existing geoid solutions are
collected in a common data base, and international research schools
are organized to develop knowledgability and skills in the art of geoid
computation, especially in developing countries.
Both services, BGI and ISG, are under the auspices of the International
Gravity Field Service IGFS, as two of the many official services of the
IAG. Other IGFS services are the International Center for Earth Tides
– 287 –
288 Earth gravity field research
(ICET), the International Center for Global Earth Models (ICGEM), and
the International Digital Elevation Model Service (IDEMS).
D 14.2 Europe
In Europe operates the EGU, the European Geosciences Union, in the
frame of which much publication and meeting activities relating to gravity
field and geoid are being co-ordinated. The EGU organizes annually
symposia, where always also sessions are included on subjects related to
gravity field and geoid. Also American scientists participate. Conversely
the American Geophysical Union’s (AGU) fall and spring meetings1 are
also favoured by European researchers.
To be mentioned is the Geodetic Institute (“Institut für Erdmessung”) of
Leibniz University in Hannover, Germany, which since 1990 has acted as
the European computing centre of the International Geoid Commission
(IGeC, and produced high quality geoid models for use in Europe (Denker,
1998); (http://www.ife.uni-hannover.de/gravity.html?&L=1).
D 14.3 The Nordic countries

In the Nordic countries, important work is being co-ordinated by the
NKG, Nordiska Kommissionen för Geodesiand its Working Group for
Geoid and Height Systems. To its activities belongs geoid determina-
tion, 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, at its computing centre in Copen-
hagen, high-quality geoid models, the next to last such being NKG2004
(Forsberg and Kaminskis, 1996), https://goo.gl/dTFfjz. 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.
D 14.4 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 af-
ter Finnish independence. The institute has been responsible for the
national fundamental levelling and gravimetric networks and their in-
ternational connections. In 2001 the Finnish Geodetic Institute’s gravity
1
Fall (autumn) meetings in San Francisco, spring meetings somewhere in the
world. The AGU, though American, is a very cosmopolitan player.
Ó »îá .
Textbooks 289
and geodesy departments were joined into a new department of geodesy

and geodynamics, to which also gravity research belongs. Among matters
studied are, e.g., solid-Earth tides, the free oscillations of the solid Earth,
post-glacial land uplift, and vertical reference or height systems.
Geoid models have been computed all the time, starting with Hirvonen’s
global model (Hirvonen, 1934) and ending, for now, with the Finnish
model FIN2005N00 (Bilker-Koivula, 2010). These geoid models are actu-
ally based on the Nordic NKG2004 gravimetric geoid, and are only fitted
to a Finnish set of GNSS levelling control points through a transformation
surface.
In 2015, the Finnish Geodetic Institute was merged into the National
Land Survey as its geospatial data centre and research facility. The
English-language acronym continues as FGI, the Finnish Geospatial
Research Institute (http://www.fgi.fi/fgi).
Also Helsinki University of Technology (today part of Aalto University)
has been active in research on the Earth’s gravity field. V. A. Heiskanen,
who was a professor at HUT in 1928 – 1949, acted 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).
D 14.5 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
Torge (1989). Difficult but good is Moritz (1980). Similarly difficult is
Molodensky et al. (1962). Worth reading also from the perspective of
physical geodesy is Vaníček and Krakiwsky (1986). A new book in this
field is Hofmann-Wellenhof and Moritz (2006).
Ó »îá .
D Field theory and vector calculus —
A core knowledge
D A.1 Vector calculus

In physics, we decribe many quantities as vector quantities. E.g., force,
velocity, electrostatic field, . . . A vector behaves in the same way as the
location difference between two neighbouring points. Velocity v, force F,
location difference ∆r = r2 − r1 , where 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 numeric values of its components
are co-ordinate system dependent, see subsection A.2.2.
About notation: in printed text, vectors are most often written in bold.
In handwritten text one may use an arrow above the symbol: − →
v.
D A.1.1 Scalar product
Between two vectors, a scalar product or dot product can be defined,

which is itself a scalar value. A scalar is in physics a single numeric
value, e.g., pressure or temperature. In the case of a scalar product of
two vector fields, the value is tied to a location, but, even if a co-ordinate
transformation changes the co-ordinate values of the location, the scalar
itself remains unchanged.
An example of a scalar product: work ∆E is
∆E = 〈F · ∆r〉 ,
the scalar product of force F and path ∆r. Often, also in the sequel, we
leave the angle brackets 〈·〉 off.
Later we shall see that if the points 1 and 2, ∆r = r2 − r1 , are very close
to each other, we may write
dE = 〈F · d r〉 ,
where d r and dE are infinitesimal elements of path and energy. If now

there is between the points A and B a curved path, we may get from this
an integral equation:
ˆ B
∆E AB = F · d r.
A
– 291 –
292 Field theory and vector calculus — core knowledge
This is the work integral.
D A.1.2 The scalar product, formally
Let
def
s = 〈a · b〉
be the scalar product of the vectors a and b. Then
⟨ ⟩ ⟨ ⟩
µa · b = a · µb = µ 〈 a · b〉 ,
〈a · b〉 = 〈b · a〉 ,
and often we call
def
√
∥a∥ = 〈 a · a〉
the norm or length of vector a.
The following also applies:
〈a · b〉 = ∥a∥ ∥b∥ cos α,
where α is the angle between vectors a and b.
D A.1.3 Exterior or vectorial product
The exterior product, or cross product, of two vectors is itself a vector

called the vectorial product (at least in three-dimensional Euclidean space
E). E.g., the angular momentum q:
q = 〈 r × p〉 ,
dr
where p = mv is linear momentum, m the mass of the body, and v =
dt
is the time derivative of the location, or velocity. We write
⟨ ⟩
dr
q = m r× .
dt
D A.1.4 The vectorial product, formally
Let
def
c = 〈a × b〉
be the vectorial product of the two vectors a and b. Then (µ ∈ R):
⟨ ⟩ ⟨ ⟩
µa × b = a × µb = µ 〈 a × b〉 ,
〈a × b〉 = − 〈b × a〉 ,
and thus 〈a × a〉 = 0.
The resulting vector c is always orthogonal to the vectors a and b; the
length of vector c corresponds to the surface area of the parallellogram suunnikas
spanned by the vectors a and b. In a formula:
∥c∥ = ∥a∥ ∥b∥ sin α,
where again α is the angle between vectors a and b. If the angle is zero,
then also the vectorial product is zero (because then, a = µb for some
suitable value of µ).
Ó »îá .
Vector calculus 293
c = 〈 b × a〉
α
∥c∥
. . b
a
D Figure A.1. Exterior or vectorial product.
D A.1.5 Kepler’s second law
If r is the distance of the body (planet) from the centre of motion (the
dr
Sun), and is the distance travelled in a unit of time, then the product
dt
⟨ ⟩
dr
r× (A.1)
dt
is precisely twice the surface area of the triangle or “area” swept over.
Let us take the time derivative of this product, i.e., the expression A.1:
d2r
⟨ ⟩ ⟨ ⟩ ⟨ ⟩
d dr dr dr
r× = × + r× 2 .
dt dt dt dt dt
Here the first term vanishes, because 〈a × a〉 = 0. In the second term we

can exploit our knowledge, that the attractive force F emanating from
the Sun that causes planetary orbital motion — and also the acceleration
2
def d r
it causes, a = 2 — is central:
dt
GMm
F = ma = − r.
∥r∥3
Substitute into the above:

⟨ ⟩
d dr GM
r× = 0− 〈r × r〉 = 0.
dt dt ∥r∥3
⟨ ⟩
dr
So: the quantity r × — angular momentum per unit of mass q/m —
dt
is conserved. Like, e.g., the total amount of energy, electric charge and
many other quantitites, the amount of angular momentum in a closed
system is constant.
G is the universal gravitational constant, M is the mass of the Sun, m
the mass of the planet.
Ó »îá .
Angular momentum
〈r × v〉
Planet
Sun
v
1
2 r×v
∥〈 〉∥
Velocity
vector
Radius 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
D angular momentum.
D A.2 Scalar and vector fields

D A.2.1 Definitions
In the space R3 we may define functions, or fields.

A scalar field is a scalar-valued function, which is defined throughout the
space (or a part of it), e.g., temperature T :
T (r) .
I.e., for every value of the location vector r there is a temperature value
T.
A vector field is a vector-valued function that again is defined throughout
space, e.g., the electrostatic field E:
E (r) .
D A.2.2 A basis in space
In the space R3 we may choose a basis made up of three vectors which

span the space in question. Generally we choose three basis vectors i, j,
k, that are orthogonal to each other, and the norms, or lengths, of which
are 1, a so-called orthonormal basis:
i ⊥ j, i ⊥ k, j ⊥ k; ∥i∥ = ∥j∥ = ∥k∥ = 1.
Now we may write vectors out into their components:
a = a1 i + a2 j + a3 k
and also scalar and vectorial products can now be calculated with the aid
of their components:
s = 〈a · b〉 = 〈(a 1 i + a 2 j + a 3 k) · ( b 1 i + b 2 j + b 3 k)〉 =
3
∑
= a1 b1 + a2 b2 + a3 b3 = ai bi,
i =1
Ó »îá .
Scalar and vector fields 295
using the above identities for the basis vectors.

For the vectorial product, the calculation is more involved; we get as the
final outcome the determinant
c = 〈a × b〉 =
⏐ ⏐
⏐ i j k ⏐⏐
⏐
= ⏐⏐ a 1 a 2 a 3 ⏐⏐ =
⏐ b b b ⏐
1 2 3
= (a 2 b 3 − a 3 b 2 ) i + (a 3 b 1 − a 1 b 3 ) j + (a 1 b 2 − a 2 b 1 ) k.
I.e.,
c1 = a2 b3 − a3 b2 ,
c2 = a3 b1 − a1 b3 ,
c3 = a1 b2 − a2 b1 .
Also these expressions are determinants:

⎡ ⎤
c1 [
⏐ a2 a3 ⏐ ⏐ a3 a1 ⏐ ⏐ a1 a2 ⏐ T
⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ]
⎣ c2 ⎦ = ⏐ ⏐ ⏐ ⏐ ⏐ ⏐
⏐ b2 b3 ⏐ ⏐ b3 b1 ⏐ ⏐ b1 b2 ⏐ .
c3
D A.2.3 The nabla operator
The location vector r can be written on the {i, j, k} basis as follows:
r = xi + yj + zk,
which defines ( x, y, z) co-ordinates in space.

Let us now define a vector operator called nabla (∇) as follows:
def ∂ ∂ ∂
∇=i +j +k .
∂x ∂y ∂z
The operator, or function, 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.
D A.2.4 The gradient
Let F (r) = F ( x, y, z) be a scalar field in space. The nabla operator will

give its gradient g, a vector field in the same space:
∂F ∂F ∂F
g = grad F = ∇F = i +j +k .
∂x ∂y ∂z
So, the field g (r) = g ( x, y, z) is a vector field in the same space, the gradi-
ent field of F .
Ó »îá .
D Figure A.3. The gradient. The level curves of the scalar field in blue.
Interpretation: the gradient describes the slope of the scalar field. The
direction of the vector is the direction in which the value of the
scalar field changes fastest, and its length describes the rate of
change with location. Imaging a hilly landscape: the height if the
ground above sea level is the scalar field, and its gradient is pointing
everywhere uphill, away from the valleys toward the hilltops. The g
arrows are the longer, the steeper is the slope of the ground surface.
The gradient operator (like also the divergence and the curl, see
later) is linear:
grad (F + G ) = grad F + grad G.
D A.2.5 The divergence
Given a vector field a ( x, y, z) . We form the scalar product s of this and

the nabla operator:
∂a 1 ∂a 2 ∂a 3
s = div a = 〈∇ · a〉 = + + .
∂x ∂y ∂z
Interpretation: The divergence describes the “sources” of a vector field,

both the positive and the negative ones. Imagine the velocity of
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 appear out of
nothing or disappear into nothing).
Ó »îá .
Scalar and vector fields 297
Figure A.4. The divergence. Positive divergences (“sources”) and negative ones
D (“sinks”).
D A.2.6 The curl
Given a vector field a ( x, y, z). We form the vectorial product c of this and
the nabla operator, producing again a vector field:
⏐ ⏐
⏐ i j k
⏐
∂ ∂ ∂
⏐ ⏐
⏐ ⏐
c = rot a = ∇ × a = ⏐⏐ ⏐=
⏐ ∂x ∂y ∂z
⏐
⏐
⏐ a1 a2 a3
⏐
⏐ ∂ ∂ ⏐ ⏐ ∂ ∂ ⏐ ⏐⏐ ∂ ∂
⏐ ⏐ ⏐ ⏐ ⏐ ⏐
⏐
⏐ ⏐ ⏐ ⏐
= ⏐⏐ ∂ x ∂ y ⏐⏐ i − ⏐ ∂x ∂z ⏐ j + ⏐ ∂ y ∂z ⏐k =
⏐ ⏐
⏐
⏐ a1 a2 ⏐ ⏐ a1 a3 ⏐ ⏐ a2 a3 ⏐
∂a 3 ∂a 2 ∂a 1 ∂a 3 ∂a 2 ∂a 1
( ) ( ) ( )
= − i+ − j+ − k,
∂y ∂z ∂z ∂x ∂x ∂y
using the evaluation rules for determinants.
Interpretation: the curl describes the eddiness or turbulence present

in a vector field.
Imagine a weather map, where low- and high-pressure zones are drawn.
Our vector field is the wind field. The wind circulates (on the Northern
hemisphere) clockwise around the high-pressure zones, and counterclock-
wise 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 describe a
rotational motion, when in R3 we need the three Euler angles.)
Ó »îá .
Figure A.5. The curl. Positive (clockwise) and negative (counterclockwise) ed-
D dies.
D A.2.7 Conservative fields
What happens if a vector field a is the gradient of a scalar field F , and

we try to calculate its curl? As follows:
⏐ ⏐
⏐
⏐ i j k ⏐
⏐
⏐
⏐∂ ∂ ∂ ⏐
⏐
rot a = rot grad F = ⏐⏐∂x ∂ y ∂z
⏐=
⏐
∂F ∂F ∂F
⏐
⏐
⏐
⏐
∂x ∂ y ∂z
⏐ ⏐
∂ ∂ ∂ ∂
( )
= F− F i+
∂ y ∂z ∂z ∂ y
∂ ∂ ∂ ∂
( )
+ F− F j+
∂z ∂x ∂x ∂z
∂ ∂ ∂ ∂
( )
+ F− F k = 0!
∂x ∂ y ∂ y ∂x
In other words, if the vector field a ( x, y, z) is the gradient of the scalar

field F ( x, y, z), its curl will vanish.
Definition: this kind of vector field a is called conservative, and the

corresponding scalar field F , a = grad F , is called the potential of
field a.
Note that if
a ( x, y, z) = grad F ( x, y, z) ,
then also
a ( x, y, z) = grad (F ( x, y, z) + F0 ) ,
Ó »îá .
Integrals 299
where F0 is a constant, because
∂ F0 ∂ F0 ∂ F0
grad F0 = i +j +k = 0.
∂x ∂y ∂z
So the potential is not uniquely defined.
D A.2.8 The Laplace operator
Assume a conservative field a, i.e., rot a = 0. Then we may write
a = grad F = ∇F,
where F is the potential.

Let us now express the divergence of field a into the potential:
∂ ∂ ∂ ∂ ∂ ∂
div a = ∇a = ∇∇F = F+ F+ F=
∂x ∂x ∂y ∂y ∂z ∂z
∂2 ∂2 ∂2
( )
def
= + + F = ∆F,
∂ x2 ∂ y2 ∂ z2
where we have introduced a new differential operator: the Delta operator

invented by the French Pierre-Simon de Laplace,
def ∂2 ∂2 ∂2
∆= + + .
∂x 2 ∂ y2 ∂ z2
For the potential of a “source free” field — e.g., for the gravitational
potential in vacuum, the electrostatic potential in an area of space free of
electric charges — this Delta, or Laplace, operator vanishes.
D A.3 Integrals
D 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 · d r,
from which one obtains the integral form, the work integral
ˆ B
∆E AB = F · d r.
A
Here is computed the amount of work needed to move a body from point
A to point B by integrating F · d r along the path AB.
Ó »îá .
Tangent vector t
rot a
Integral
¸
S 〈a · t〉 ds
Closed
path S
D Figure A.6. The Stokes curl theorem.
If we parametrize the path according to arc length s, and the tangent

vector to the path is called
dx dy dz
t= i+ j+ k,
ds ds ds
we may also write
ˆ B
∆E AB = 〈F · t〉 ds,
A
the parametrized version of the integral.
D A.3.2 The surface integral
Assume given again some vector field a and a surface in space S . Often,
one needs to integrate over the surface S the normal component of a
vector field, the projection of a onto the normal vector of the surface.
Let the normal vector of the surface be n. Then we must integrate
¨
〈a · n〉 dS,
S
symbolically written ¨
a · d S,
S
where the notation d S is called an oriented surface element. It is a vector
pointing in the same direction as the normal vector n.
Like a curve, also a surface can be parametrized. E.g., the Earth’s surface
(assumed a sphere) can be parametrized by latitude φ and longitude λ:
( )
r = r φ, λ . In this case we write as the surface element
dS = R 2 cos φ d φ d λ,
Ó »îá .
Integrals 301
where R 2 cos φ is Jacobi’s determinant, of the parameter pair φ, λ . In

( )
this parametrization, the integral is calculated as follows:

¨ ¨
a · dS = 〈a · n〉 R 2 cos φ d φ d λ.
S S
Other surfaces and parametrizations have other Jacobi’s determinants.

The determinant always describes the true area of a “parameter surface
element” d φ d λ “in nature”. E.g., on the Earth’s surface, a degree times
( )
degree patch is the 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) parametrization in the plane, it is 1 and thus can be left
out altogether.
D A.3.3 The Stokes curl theorem
Let S be a surface in space (not necessarily flat) and ∂S its edge curve.
Assume that the surface and its edge are well-behaved enough for all
necessary integations and differentiations to be possible. Then (Stokes):
¨ ˛
rot a · d S = a · d r.
S ∂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: If rot a = 0 everywhere (a conservative vector field) then
˛
a · d r = 0,
∂S
i.e., also
ˆ B ˆ B
a · dr = a · d r.
A, path 1 A, path 2
In other words, 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 explains perhaps better the essence of a conservative force
field. A conservative field can be represented as the gradient of
a potential: a = grad F, where F is the potential of the field. The
Earth’s gravity vector field g ( x, y, z) is the gradient of the Earth’s
gravity potential W ( x, y, z). At mean sea level — more precisely, at
the geoid — the gravity potential is constant; the gravity vector g
stands everywhere orthogonally on the geoid.
D A.3.4 The Gauss integral theorem
Let V be a certain volume of space, and ∂V its closed boundary, a union

of surfaces. Assume again that both are mathematically well behaved.
Ó »îá .
n
a
∂V
div a
Figure A.7. The Gauss integral theorem. n is the normal vector to the exte-
rior surface. Note that the Gauss intregral theorem can also be
formulated with the aid of (Michael Faraday’s) field lines: a field
line starts or terminates on an electric charge (i.e., a place where
D div a ̸= 0) or runs to infinity (i.e., through the surface ∂V ).
Then the following theorem applies (Gauss):

˚ ¨ ¨
div a dV = a · dS = 〈a · n〉 dS.
V ∂V ∂V
In words: what is created inside a body (i.e., “sources”, divergence) must

come out through its surfaces.
Remark: generally the orientation of surface ∂V is taken as positive on
the outside, i.e., the normal vector n points outward.
D A.4 The continuity of matter

An often used equation in, e.g., hydro- or aerodynamics is the continuity
equation. This describes that matter cannot just disappear or increase in
amount. In the general case, this equation looks like this:
( ) d
div ρ v + ρ = 0.
dt
Here, the expression ρ v describes mass currents; ρ is the matter density,
( )
v is the current flow velocity. The term div ρ v describes how much
more matter, in a unit of time, exits the volume element than enters it,
Ó »îá .
The continuity of matter 303
d
per unit of volume. The second term again, ρ , describes the change
dt
in the amount of matter inside the volume element over time. The two
terms must balance for the “matter accounting” to close.
If the moving fluid is incompressible, then ρ is constant and
d ( )
ρ=0 and div ρ v = ρ div v,
dt
and we obtain
ρ div v = 0 =⇒ div v = 0.
Remember, however, that not necessarily rot v = 0 — i.e., the flow isn’t
necessarily eddy free —, i.e., a potential F for which v = grad F does not
necessarily exist.
Ó »îá .
D Function spaces
B
D B.1 An abstract vector space

In an abstract vector space we may create a basis, with the help of which
each vector can be written as a linear combination of the basis vectors:
e.g., if the basis, in a three-dimensional space, is {e1 , e2 , e3 }, we may write
an arbitrary vector r in the form
3
∑
r = r 1 e1 + r 2 e2 + r 3 e3 = r i ei .
i =1
Precisely because three basis vectors is always enough, we call the ordi-
nary (Euclidean) space E three-dimensional, also R3 .
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〉 ,
and commutativity, that
〈r · s〉 = 〈s · r〉
⟨ ⟩
If the basis vectors are orthogonal to each other, in other words, e i · e j =
0 if i ̸= j , we may calculate the coefficients r i simply as
3
∑ 〈r · e i 〉
r= ei (B.1)
〈e i · e i 〉
i =1
If, additionally, also 〈e i · e i 〉 = ∥e i ∥2 = 1 ∀ i ∈ {1, 2, 3}, i.e., the basis vectors

are orthonormal, equation B.1 becomes simpler still:
3
∑ 3
∑
r= 〈r · e i 〉 e i = r i ei . (B.2)
i =1 i =1
Here, the coefficients are, according to definition, r i = 〈r · e i 〉. The quan-
tity √
∥e i ∥ = 〈e i · e i 〉
is called the norm of the vector e i .
– 305 –
306 Function spaces
D B.2 Fourier function space

D B.2.1 Description
Also functions can be considered elements in a vector space. If we define

the scalar product of two functions f , g as the following integral:
⟨−
→ ⟩ 1ˆ 2π
−
→ def
f · g = f ( x) g ( x) dx, (B.3)
π 0
it is easily verified that the above requirements for a scalar product are
met.
One basis in this vector space (a function space) is formed by the so-called
Fourier basis functions,
−
→ 1p
e0 = 2 ( k = 0)
2
−
→
e = cos kx, k = 1, 2, 3, ... (B.4)
k
−
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), i.e.,
∞
1 p ∑
f ( x) = a 0 2 + (a k cos kx + b k sin kx)
2
k=1
— the familiar Fourier expansion —, in which the coefficients are

⟨−
→ ⟩ 1 p ˆ 2π p
a0 = −→
f · e0 = 2 f ( x) dx = 2 · f ( x),
2π 0
⟨− ˆ 2π
→ − 1
f ·→
⟩
ak = e =k f ( x) cos kxdx, k = 1, 2, . . .
π 0
⟨− ˆ 2π
→ −→⟩ 1
bk = f · e −k = f ( x) sin kxdx, k = 1, 2, . . .
π 0
This is the familiar way in which the coefficients of a Fourier series are
calculated.
D 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)
Ó »îá .
Fourier function space 307
K =1
K = 25 K =3 K =5
1.0
a0
0.8
f (x)
0.6
X
0.4 a b
0.2 b5
b3
0
b1
−0.2
0 1 2 3 4 5 6
Figure B.1. Fourier analysis on a step function. Plotted are the Fourier expan-
def p ∑K ∑K
sions f (K ) (x) = 12 a 0 2 + k=1 b k sin kx = 12 − π2 k=1,3,5,... 1k sin kx,
for values of K of 1, 3, 5, and 25. The inset gives the spectrum of
D the function.
We can derive the Fourier coefficients of this function as follows:

ˆ 2π
1 p 1 p 1p
a0 = 2· f ( x) dx = 2·π = 2,
2π 0 2π 2
ˆ ˆ
1 2π 1 2π
ak = f ( x) cos kxdx = cos kxdx =
π 0 π π
1 [1 ]2π 1
= sin kx = {sin 2 kπ − sin kπ} = 0,
π k π kπ
ˆ ˆ
1 2π 1 2π
bk = f ( x) sin kxdx = sin kxdx =
π 0 π π
1[ 1 ]2π 1
= − cos kx = {cos kπ − cos 2 kπ}
π k π kπ
{
1 { 0 k even,
(−1)k − 1 =
}
=
kπ 2
− kπ k odd.
p
Or, in numbers, a 0 = 21 2 = 0.70710, b 1 = − π2 = −0.63662 . . . , b 3 = − 32π =
−0.21220, b 5 = −0.12732, and so forth. The expansion becomes now
∞ ∞
1p ∑ 1 2 ∑ 1
f ( x) = 2a 0 + b k sin kx = − sin kx =
2 2 π k
k=1 k=1,3,5,...
∞
1 2 ∑ 1
= − sin (2 n − 1) x.
2 π 2n − 1
n=1
Note 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.
Ó »îá .
308 Function spaces
D B.2.3 Convergence
The Fourier expansion converges in the square integral sense, i.e., if we

define the truncated expansion
K
(K) def1 p ∑
f ( x) = a 0 2 + (a k cos kx + b k sin kx) ,
2
k=1
then ˆ 2π {
1 }2
lim f (K) ( x) − f ( x) dx = 0.
K →∞ π 0
Note that this does not mean that, for every x ∈ [0, 2π), f (K) ( x) → f ( x)
when K → ∞. Looking at figure B.1, there will always remain a small
neighbourhood of x = π where the difference f (K) ( x)− f ( x) will be almost 12 ,
even for arbitrarily large values of K . We say that the Fourier expansion
is convergent, but not uniformly convergent.
The Fourier series converges pointwise “almost everywhere” in x ∈ [0, 2π):
in all points except in the two special points x = 0 and x = π.
Also, note the “shoulder” of the expansion even for K = 25. This shoulder
will get narrower for higher K , but not any lower. It is known as the
Gibbs phenomenon.
D B.3 Sturm-Liouville differential equations

D B.3.1 The eigenvalue problem
In an abstract vector space we may formulate an eigenvalue problem: if

there exists a linear operator (mapping) L, we may write
Lx − λx = 0,
where the problem consists of determining the values λ i for which solu-
tions x i exist.
In a concrete n-dimensional vector space we may write the vector
n
∑
x= xi ei ,
i =1
and, thanks to linearity,

{ n } n
∑ ∑
Lx = L xi ei = x i · L {e i } ;
i =1 i =1
on the other hand we may write n different vectors L {e i } on the basis

{ }
e j in the following way:
n
∑
L {e i } = ai je j.
j =1
Ó »îá .
Sturm-Liouville differential equations 309
This defines the coefficients a i j , which may be collected into an n × n

matrix A .
Now by substitution
n
[ n ]
∑ ∑
Lx = a i j xi e j , (B.5)
j =1 i =1
when also
n
∑ n
∑ [ ]
λx = λ xi ei = λx j e j . (B.6)
i =1 j =1
By combining equations B.5, B.6, of which all coefficients must be identi-

cal, we obtain
n
∑
a i j x i − λ x j = 0, j = 1, . . . , n,
i =1
or, as a matric equation,
A x − λx = 0, (B.7)
where A is a matrix consisting of the coefficients a i j , and x a column
[ ]T
vector consisting of the coefficients x i : x = x1 x2 · · · xn .
Of course also equation B.7 represents an eigenvalue problem, but now
in the linear vector space consisting of all coefficient vectors x. Every x is
the numerical representation of a vector x on the chosen basis {e i }. The
matrix A again is the numerical representation of operator L on the same
basis1 .
D B.3.2 A self-adjoint operator
Let L be a linear operator in a vector space where there exists a scalar

product, i.e., a bilinear form 〈x · y〉 which is symmetric or commutative.
Then L is self-adjoint, if for each pair of vectors x, y it holds that
〈x · Ly〉 = 〈Lx · y〉 .
If the matrix A is self-adjoint, that means that
〈x · A y〉 = 〈 A x · y〉
i.e.,
n n n
[ ] [ n ]
∑ ∑ ∑ ∑
xi ai j yj = a i j x j yi ,
i =1 j =1 i =1 j =1
which is trivially true if, for all i, j values 1, . . . , n,
a i j = a ji , i.e., A = A T .
In other words:
1
An advantage of the numerical representations is of course that one can really
calculate with them.
Ó »îá .
310 Function spaces

A symmetric matrix is a self-adjoint operator.

From linear algebra it is undoubtedly familiar, that the eigenvectors x p , x q

belonging to different eigenvalues λ p ̸= λ q of a symmetric n × n matrix
are mutually orthogonal: x p ⊥ x q . If all eigenvalues λ p , p = 1, . . . , n are
different, then the eigenvectors x p , p = 1, . . . , n will constitute a complete
orthogonal basis2 in the vector space Rn .
The proof is not hard. We start from the equation for the eigenvalue
problem for eigenvectors and -values x p , λ p :
Lx p = λ p x p ,
and multiply from the left by vector x q :

⟨ ⟩ ⟨ ⟩
x q · Lx p = λ p x q · x p .
Similarly for eigenvectors and -values x q , λ q multiplied from the left by

vector x p :
⟨ ⟩ ⟨ ⟩
x p · Lx q = λ q x p · x q .
If now L is self-adjoint, then
⟨ ⟩ ⟨ ⟩ ⟨ ⟩
x q · Lx p = Lx q · x p = x p · Lx q
(remember that the scalar product is symmetric) and thus, that

⟨ ⟩ ⟨ ⟩
λ p xq · x p = λq x p · xq
i.e.,
( )⟨ ⟩
λ p − λq x p · x q = 0.
⟨ ⟩
If λ p ̸= λ q , we thus must have x p · x q = 0, or x p ⊥ x q . 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
σ2x σ x y
[ ]
Var (xP ) = ΣPP = ,
σ x y σ2y
a symmetric matrix. Here, σ2x and σ2y are the variances, or squares
of the mean errors, of the x and y co-ordinates, whereas σ x y is the
covariance between the co-ordinates.
The eigenvalues of this matrix ΣPP are the solutions of the charac-
teristic equation
σ2x − λ σx y
[ ]
det = 0,
σx y σ2y − λ
2
Actually the eigenvectors may be arbitrarily re-scaled: if x is an eigenvector,
def
then also e = ∥xx∥ is. Thus we obtain an orthonormal basis.
Ó »îá .
Sturm-Liouville differential equations 311
i.e.,
σ2x − λ σ2y − λ − σ2x y = 0.
( )( )
This yields
1( 2 ) 1 [ √]2
σ x + σ2y ±
[ ]
λ1,2 = σ2x + σ2y − 4 σ2x σ2y − σ2x y =
2 2√
1( 2 1 [ 2 ]2
σ x + σ2y ±
)
= σ x − σ2y + 4σ2x y .
2 2
The variance matrix has a variance or error ellipse. The semi-
p p
lengths of its principal axes are λ1 , λ2 and the directions of
the principal axes are the eigenvectors of ΣPP , x1 , x2 , mutually
orthogonal. If the co-ordinate axes are turned into the directions of
x1,2 , then the matrix ΣPP will assume the form
σ2x′ λ1 0
[ ] [ ]
0
Σ′PP = = .
0 σ2y′ 0 λ2
The sum of the eigenvalues (and the trace of this matrix), λ1 + λ2 =

σ2x + σ2y , is an invariant called the point variance.
D B.3.3 Self-adjoint differential equations
In a function space there are also self-adjoint or “symmetric” differential

equations. In fact, the most famous equations of physics are of this type.
Take a good look at, e.g., the oscillation equation
d2
2
x ( t ) − ω 2 x ( t ) = 0. (B.8)
dt
The solution has the general form (α amplitude, φ phase constant)
( )
x ( t) = α sin tω − φ .
⏐ ⏐
d ⏐⏐ d ⏐⏐
On the interval t ∈ [0, T ] we require x (0) = x (T ) , x⏐ = x , i.e.,
dt x=0 dt ⏐ x=T
periodicity. These boundary conditions are an essential part of being
self-adjoint. Then, a solution is found only for certain values of ω —
quantization.
Note that equation B.8 is an eigenvalue problem, form-wise:
Lx − ω2 x = 0,
where 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:
ˆ T
⟨−
→
x ·−
→ ⟩ def
y = x ( t) y ( t) dt,
0
Ó »îá .
312 Function spaces
it holds that (integration by parts):

ˆ T
⟨−
→ d2
x · L−
→ ⟩
y = x ( t) y ( t) dt =
0 dt2
[ ]T ˆ T
d d d
= x ( t) y ( t) − x ( t) y ( t) dt,
dt 0 0 dt dt
ˆ T( 2
)
⟨ − d
L→
x ·−
→ ⟩
y = x ( t) y ( t) dt =
0 dt2
[( ) ]T ˆ T
d d d
= x ( t) y ( t) − x ( t) y ( t) dt.
dt 0 0 dt dt
As on the right-hand side the first terms vanish and the second terms are
identical, we have
⟨−
→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 ω3 . For the
oscillation equation with the above periodicity conditions they are just
the solution functions
2π k
( )
( )
sin ωk t − φ = sin t−φ , (B.9)
T
in which the frequency

2π k
ωk =
T
is quantized 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 “Sturm4 -
Liouville5 type problems”. To it belongs, e.g., the oscillation equation,
3
In fact, for the same value ω there exist two mutually orthogonal periodic
solutions,
2π kt
sin ωk t = sin ,
T
2π kt
cos ωk t = cos .
T
Any linear combination of these is also a valid solution, and is of the general
form B.9.
4
Jacques Charles François Sturm FRS FAS (1803 – 1855) was an eminent French
mathematician, one of the 72 names engraved on the Eiffel Tower.
5
Joseph Liouville FRS FRSE FAS (1809 – 1882) was an eminent French mathe-
matician.
Ó »îá .
Legendre polynomials 313
Legendre’s equation, 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 the basis functions for the general solution of
many partial differential equations.
D B.4 Legendre polynomials

Also the ordinary Legendre polynomials P n ( t) constitute a basis in a
function space, with the scalar product definition
⟨−
→ ⟩ ˆ +1
−
→ def
f · g = f ( t) g ( t) dt.
−1
They don’t however constitute an orthonormal basis, but only an or-

thogonal one:
ˆ +1
−
 p→n 2 = −
 ⟨→ − →⟩ = 1
pn · p n P n2 ( t) dt = .
−1 2n + 1
Unlike ordinary space, which is three-dimensional, a function space is an
∞-dimensional, abstract vector space, that nevertheless helps us make
more concrete certain abstract, but very useful fundamentals of function
theory!
D B.5 Spherical harmonics

On the surface of the sphere also all functions can be considered elements
of a function space. Every function meeting certain “well-behavedness
properties” — like integrability — is an element. The functions
( ) ( )
R nm φ, λ = P nm sin φ cos mλ,
( ) ( )
S nm φ, λ = P nm sin φ sin mλ,
together, for the values n = 0 . . . ∞, m = 0 . . . n, 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.
An alternative, more compact way of writing this is
{ ( )
( ) P nm sin φ cos mλ m ≥ 0,
Ynm φ, λ = ( )
P n|m| sin φ sin | m| λ m < 0,
for values n = 0 . . . ∞, m = − n . . . n.
In this function space, a scalar product is defined:
¨
⟨−
→ ⟩ 1
v ·−
→ ( ) ( )
w = V φ, λ W φ, λ d σ ,
4π σ
Ó »îá .
314 Function spaces
in which σ is the surface of the unit sphere (“directional sphere”, or even

“celestial sphere”). According to this definition we can show, that two
different functions, namely Ynm , Ysr , are orthogonal with respect to each
other: ¨
⟨−→ − → ⟩ 1 ( ) ( )
ynm · ysr = Ynm φ, λ Ysr φ, λ d σ = 0
4π σ
if n ̸= s or m ̸= r , etc.
The basis −→
{ }
ynm is orthogonal but not orthonormal: the “length” of every
vector differs from 1.
¨ {
1
−→2 ⟨−→ −→⟩ 1 ( )2 2n+1 m = 0,
 ynm  = ynm · ynm = Ynm φ, λ dσ =
4π σ
1 (n+| m|)!
m ̸= 0,
2(2n+1) (n−| m|)!
see also Heiskanen and Moritz (1967). Proving this with the help of
equation 2.13 is a long process.
If we now divide the functions Ynm (or, equivalently, R nm , S nm ) by the
square roots of the above factors, we obtain the fully normalized surface
spherical harmonics Y nm .
With those it is again easy to calculate the coefficients f nm of a given gen-
( )
eral function f φ, λ (the overline means that these are fully normalized
coefficients):
¨
⟨ −−→⟩ 1 ( ) ( )
f nm = f · ynm = f φ, λ Y nm φ, λ d σ. (B.10)
4π σ
This is a straightforward projection on the unit vectors of the basis (geo-

metric analogue).
( )
In the above integral, f φ, λ is the function f on the Earth’s surface, i.e.,
( ) ( )
if the mean radius of the Earth is R , f φ, λ = f φ, λ, R .
The equation corresponding to expansion 2.10 is
∞ n
( ) ∑ 1 ∑ ( )( )
V φ, λ, r = P nm sin φ a nm cos mλ + b nm sin mλ .
r n+1
n=0 m=0
We may write also

{ ( )
( ) P nm sin φ cos mλ m ≥ 0,
Y nm φ, λ = ( )
P n|m| sin φ sin | m| λ m < 0,
which corresponds to the definition of the fully normalized Legendre

functions:
( ) p ( )
P n0 sin φ = 2 n + 1 P n0 sin φ ,
√
( ) ( n − m)! ( )
P nm sin φ = 2(2 n + 1) P nm sin φ , m > 0.
( n + m)!
Ó »îá .
Then the above equation for the potential becomes

∞ n
( ) ∑ 1 ∑ ( )
V φ, λ, r = ynm Y nm φ, λ ,
r n+1 m=− n
n=0
in which {
a nm m ≥ 0,
ynm =
b n| m| m < 0.
1. The identity 〈r · s〉 = 〈s · r〉, for two elements r and s of a vector space,

expresses the property of: linearity | commutativity | associativity.
⋆
D 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.
Ó »îá .
D Why does FFT work?
C
FFT is a factorization method for computing the discrete Fourier trans-

form that spectacularly reduces the number of calculations needed and
speeds up the calculation. It requires the number of grid points to be a
factorizable 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, e.g., N = 360 = 2 × 2 × 2 × 3 × 3 × 5. If the grid
size is a prime number, FFT does not give any advantage over ordinary
discrete FFT.
If the function f ( x) is defined on the interval1 x ∈ [0, 1) on an equi-spaced
grid as values f ( xk ) , k = 0, . . . , N − 1, then the discrete Fourier transform
in one dimension is { }
F f ( x) = F (ω) ,
in which
N −1 ( )
( ) 1 ∑ jk
F ωj = f ( xk ) exp 2π i , j = 0, . . . , N − 1. (C.1)
N N
k=0
Here, also the frequency argument ω j , j = 0, . . . , N − 1 is defined on the

interval ω ∈ [0, 1]. Here, i is the imaginary unit i 2 = −1. We use exp ( x) to
denote e x .
Correspondingly, the inverse discrete Fourier transform
{ }
−1
F F (ω)
is
N −1 ( )
∑ ( ) jk
f ( xk ) = F ωj exp −2π i , k = 0, . . . , N − 1. (C.2)
N
j =0
1
For simplicity’s sake. An arbitrary interval will work.
– 317 –
318 Why does FFT work?
FFT is just a very efficient method for computing both these formulas
C.1, C.2. A brute-force calculation of these formulas requires of order N 2
“standard operations”, each of them a single multiplication plus a single
addition or subtraction. If N is even, we may write
⎡N ⎤
2 −1 ( ) N −1 ( )
( ) 1 ∑ jk ∑ jk ⎦
F ωj = ⎣ f ( xk ) exp −2π i + f ( xk ) exp −2π i =
N N N
k=0 k= N
2
⎡N N
⎤
2 −1 ( ) ( 2 −1 (
N)∑ ( ′
)
1 ∑ jk j ) jk ⎦
= ⎣ f ( xk ) exp −2π i + exp −2π i 2 f xk′ + N exp −2π i =
N N N 2 N
k=0 k′ =0
⎡N N
⎤
2 −1 ( ) 2 −1 ( ( )
1 ∑ jk ∑ ) jk ⎦
= ⎣ f ( xk ) exp −2π i + exp (−π i j ) f x k+ N exp −2π i =
N N 2 N
k=0 k=0
N
2 −1 [ )] ( ) ( )
1 ∑ j ( jk + if j even
= f ( x k ) ± f x k+ N exp −2π i , (C.3)
N 2 N − if j odd
k=0
the computation of which sum requires only N · N N

2 multiplications and 2 +
N
2 additions and subtractions. Here we used Euler’s identity exp (−π i ) =
)j
−1, i.e., e−π i j = e−π i = (−1) j , either +1 or −1.
(
Note that the expression in square brackets, for each k value

1
k = 0, 1, . . . , N − 1, is either a summation, for even values of j , or a
subtraction, for odd values of j . In total, 2 · N
2 = N different values are
2
computed, N N N N
2 sums and 2 differences. After that, there are N · 2 = 2
multiplications. We assume that the exp expressions are pre-calculated
2
into a lookup table. Altogether some N2 standard operations are needed,
half the original number.
Equation C.3 is itself recognised as a Fourier series, but the number of
support points is instead of N only N N
2 . If also 2 is even, we may repeat
2
the above trick, resulting in an expression requiring only of order N4
operations. Lather, rinse, repeat, and the number of operations becomes
N2 N2 N2
8 , 16 , 32 , etc. . . A more precise analysis shows that, if N is a power of
2, then the whole discrete Fourier transform may be computed in order
N × 2 log N operations!
In the literature, smart algorithms are found implementing the above
method, e.g., fftw (“Fastest Fourier Transform in the West”, http://www.
fftw.org).
Ó »îá .
D Helmert condensation
D
D D.1 The interior potential of the topography

In order to derive the equation for Helmert condensation, we first derive
the equation for the interior potential of the topography, i.e., the masses
between sea level and the terrain surface:
˚ ( )
ρ r ′ , φ′ , λ′
) dV ,
( )
Tt r, φ, λ = G (
top ℓ r, r ′ , ψ
( )
where ψ is the angular distance between the evaluation point r, φ, λ
( )
and the observation point r ′ , φ′ , λ′ . The spatial distance ℓ between
those points again is written using the interior expansion (equation 7.4):
∞
1 1 ∑ ( r )n+1 ( )
= P n cos ψ .
ℓ r r′
n=0
This expansion converges uniformly1 with respect to ψ if r < r ′ .

Substitute:
˚ ∞
1 ∑ ( r )n+1 (
Ttint P n cos ψ d V =
( ) )
r, φ, λ = G ρ ′
top r r
n=0
¨ ˆ R + H (φ′ ,λ′ ) ∞
1 ∑ ( r )n+1 ( ′ )2 ′ ( )
= Gρ ′
r dr P n cos ψ d σ =
σ R r r
n=0
⎡ ∞
⎤R + H
¨
∑ ( 1 ( ′ )−(n−2) )
rn − r + ⎥
n−2
⎢ ( )
= Gρ ⎢ n=0
⎥ P n cos ψ d σ =
σ n̸=2
⎣ ⎦
+ r 2 ln r ′ R
1
Uniform convergence means that, given r, r ′ , for every ϵ > 0 there is an Nmin
for which
⏐ ⏐
N ( )
r n+1
⏐1 1 ∑ ( )⏐⏐
⏐ − P n 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.
– 319 –
320 Helmert condensation
¨ ∑
⎡ −(n−2) ⎤
∞ n R −
r ⎣
− (R + H )−(n−2) +
( )
= Gρ ⎦ P n cos ψ d σ.
σ n=0 n−2
n̸=2
+ r 2 ln R +
R
H
Here, we shall expand the following expression into a Taylor series:
H ( n − 2) ( n − 1) H 2
⎡ ⎤
1 − ( n − 2) + −
(R + H )−(n−2) = R −(n−2) ⎣
⎢ R 2 R2 ⎥ ⎦.
3
( n − 2) ( n − 1) n H
− + . . .
2·3 R3
Also the special case n = 2,
H 1 H2 1 H3 1 H4
[ ]
2 R+H 2
r ln = r − + − +... =
R R 2 R2 3 R3 4 R4
H n − 1 H 2 ( n − 1) n H 3
⎡ ⎤
r ⎢ R − 2 R2 + 2 · 3 R4 − ⎥
n
= ⎦,
R n−2 ( n − 1) n( n + 1) H 4
⎣
− +...
2·3·4 R4
is cleanly included into the following expression obtained by substitution:
H n − 1 H2
⎡ ⎤
¨ ∑
∞
r ⎢ R − 2 R2 +
n
Ttint r, φ, λ = G ρ
( ) ⎥ ( )
n−2 ⎣ 3 ⎦P n cos ψ d σ.
σ n=0 R ( n − 1) n H
+ −...
6 R3
(D.1)
D D.2 The exterior potential of the topography

In the same way we may derive the exterior potential of the topography.
Now we use for the expansion of the inverse distance (equation 7.4):
∞ ( ′ )n+1 ∞ ( )n
1 ∑1 r ( ) ∑ 1 r′ ( )
= P n cos ψ = P n cos ψ ,
ℓ r′ r r r
n=0 n=0
which converges uniformly if r > r ′ .

Substitution yields
˚ ∞ ( )n
∑ 1 r′
Ttext P n cos ψ d V =
( )
= Gρ
top n=0 r r
¨ ˆ R +H (φ′ ,λ′ ) ∑
∞ ( ′ )n
1 r ( ′ )2
dr ′ P n cos ψ d σ =
( )
= Gρ r
σ R r r
n=0
¨ [∑
∞
]R + H
1 1 ( ′ )n+3 ( )
= Gρ r P n cos ψ d σ =
σ r n+1 n + 3
n=0 R
¨ ∞
∑ 1 1 [
(R + H )n+3 − R n+3 P n cos ψ d σ.
] ( )
= Gρ
σ n=0 r n+1 n + 3
Ó »îá .
The exterior potential of the condensation layer 321
Here again we use the Taylor expansion:
H ( n + 3) ( n + 2) H 2
⎡ ⎤
⎢ 1 + (n + 3) R + 2
+
R2 ⎥
(R + H )n+3 = R n+3 ⎣ 3 ⎦.
( n + 3) ( n + 2) ( n + 1) H
+...
2·3 R3
Substitution yields
H n + 2 H2
⎡ ⎤
¨ ∑
∞ ( ) n+1
R ⎢ R + 2 R2 +
Ttext = G ρ R 2
⎥ ( )
3 ⎦ P n cos ψ d σ.
r ( n + 2) ( n + 1) H
⎣
σ n=0
+...
6 R3
(D.2)
This is thus the exterior potential of the topography, or, inside the topo-
graphic masses, the harmonic downward continuation of the exterior
potential, assuming that this is mathematically possible (in the case of
mountainous topography, generally not) and doesn’t diverge.
D D.3 The exterior potential of the condensation layer

This is derived by specializing equation D.2 to the case H → 0, but
nevertheless ρ → ∞, so that κ = ρ H is finite. in this limit, all terms
containing H 2 , H 3 etc. are going to zero. The result is then
¨ ∑
∞ ( ) n+1
R H
Tcext 2
( )
= GρR P n cos ψ d σ =
σ n=0 r R
¨ ∑
∞ ( ) n+1
R ( )
= GR κ P n cos ψ d σ.
σ n=0 r
Earlier on we had a more precise formula 5.4 for κ on the surface of a

spherical Earth: ( )
H
κ = ρH 1 + ;
R
by substituting this in the previous, we obtain
¨ ∑
∞ ( ) n+1 [
H2
]
R H
Tcext 2
( )
= GρR + 2 P n cos ψ d σ. (D.3)
σ n=0 r R R
D D.4 The total potential of Helmert condensation

This is obtained by subtracting equations D.2 and D.3 from each other.
The result (which thus applies in the exterior space) is
ext
THelmert = Ttext − Tcext =
Ó »îá .
⎡ {
n+2 } H2 ⎤
¨ ∑
∞ ( ) n+1 −1 +
R 2 R2
= GρR2
⎥ ( )
⎦ P n cos ψ d σ =
⎢
3
r ( n + 2) ( n + 1) H
⎣
σ n=0
+ +...
6 R3
¨ ∑∞ ( ) n+1 [
n 2 ( n + 2) ( n + 1) H 3
]
R ( )
= Gρ H + + . . . P n cos ψ d σ.
σ r 2 6 R
n=0
(D.4)
Often we define the degree constituents of powers of height H (compare

the degree constituent equation 2.16), as follows:
¨
2n + 1
H nν H ν P n cos ψ d σ,
( )
= (D.5)
4π σ
after which we may expand

∞
ν
H nν .
∑
H =
n=0
Then
∞ ( ) n+1 [
( n + 2) ( n + 1) H n3
]
ext
∑ R n
= 4πG ρ
THelmert H n2 + +... .
r 2 (2 n + 1) 6 (2 n + 1) R
n=0
(D.6)
If the topography is constant, then all terms for which n ̸= 0 vanish; in
the above expression, also the first term vanishes, and the second and
further terms are very small. So, in practice2
ext 4 3 Gρ
THelmert = πH · +... ≈ 0
3 r
as was to be expected. The exterior field remains in this special case
almost unchanged as a result of Helmert condensation.
D D.4.1 The gravity effect of Helmert condensation
Let us calculate gravity anomalies from the Helmert potential, but only
using the first term of equation D.6:
∂T 2
∆ g Helmert = + T=
∂r r
∞ [ ]( )n+1
∑ n − ( n + 1) 2 R
= 2πG ρ + H n2 =
2n + 1 r r r
n=0
∞ ( )n+1
1 ∑ n ( n − 1) R
= −2πG ρ H n2 .
r 2n + 1 r
n=0
2
As a curiosity, this result can be interpreted as the potential of a sphere of
crustal matter with radius H (the average height of the topography) located at
the geocentre. Even for a topographic mean height of 10 km the effect on the
geoid would only be 12 mm (exercise!).
Ó »îá .
The total potential of Helmert condensation 323
Now, also n = 1 gives a zero result, expected as gravity anomalies do not

contain any components of degree number 1.
Note in this equation also a dependence upon n: The gravity effect of
Helmert condensation is dominated by short wavelengths or the very
local features of the topography. The appearance of the square of height
in this equation is again related to the terrain correction, in which also
the square of the terrain height figures. When we are summing squares,
leaving out terms will always cause a systematic error: also very short
wavelengths, i.e., high values of n, must be taken along in the summation.
D D.4.2 The interior potential of Helmert condensation
This quantity is evaluated on the level of the geoid. It represents the

indirect effect of Helmert condensation, i.e., the shift of the geoid surface
caused by the mass shift in space.
int
THelmert = T tint − T cext =
H n − 1 H2
⎡ ⎤
¨ ∑
∞
⎢ R − 2 R2 +
= GρR2
⎥ ( )
3 ⎦ P n cos ψ d σ−
( n − 1) n H
⎣
σ n=0
−...
6 R3
¨ ∑∞ [
H H2
]
2
( )
− GρR + 2 P n cos ψ d σ =
σ R R
n=0
¨ ∑
∞ [
( n − 1) n H 3
]
n+1 2
( )
= Gρ − H + − . . . P n cos ψ d σ.
σ n=0 2 6 R
Using again the definition of the degree constituents of the powers of

height H , equation D.5:
¨
2n + 1
H nν H ν P n cos ψ d σ,
( )
=
4π σ
we obtain
∞ [
( n − 1) n H n3
]
int
∑ n+1
THelmert = 4πG ρ − H n2 + −... ,
2 (2 n + 1) 6 (2 n + 1) R
n=0
from which one obtains the indirect effect of Helmert condensation:

int ∞
4πG ρ ∑ ( n − 1) n H n3
[ ]
THelmert n+1 2
δ NHC = − =− − H + −... =
γ γ 2 (2 n + 1) n 6 (2 n + 1) R
n=0
[ ∞ ∞
]
4πG ρ 1 ∑ n+1 2 1 ∑ ( n − 1) n 3
= · Hn − H −... . (D.7)
γ 2 2n + 1 6R 2n + 1 n
n=0 n=0
The term n = 0 yields the indirect effect of a constant terrain H = H = H0 :

using only the first term inside the brackets yields
2πG ρ 2
δ NHC,Const = H ,
γ
which can not be neglected.
Ó »îá .
D 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, density κ = H ρ , moves downward by on average 12 H . The effect
would be the same if mean sea level3 was covered by a double mass
density layer
1
µ = H2ρ. (D.8)
2
The potential of this layer is, in spherical approximation,
¨ ¨
∂ (1) 2 ∂ (1)
T =G µ dS ≈ GR µ d σ.
S ∂n ℓ σ ∂n ℓ
Written more explicitly:
¨
∂
( )
2 1
T P = GR µQ d σQ .
σ ∂r Q ℓPQ
We use the expansion into Legendre polynomials, equation 7.4:

∞ ( )n+1
1 1 ∑ rQ ( )
= P n cos ψPQ ,
ℓPQ rQ rP
n=0
differentiate with respect to r Q , and substitute:

¨ ∞ ( )n+1
2 1 ∑ rQ ( )
T P = GR 2
µQ n P n cos ψPQ d σQ .
σ rQ rP
n=0
By substituting into this the equation D.8 for the µQ double mass density
layer, we obtain, by taking the limit r P , r Q ↓ R :
∞ ¨
1 ∑ ( ) ( )
T = n 2πGH ρ HP n cos ψ d σ =
4π σ
n=0
∞ ¨
1 ∑ ( )
= n A B HP n cos ψ d σ.
4π
n=0
Here, we have left the designations P,Q again off as no longer needed for
clarity.
The symbol A B denotes the attraction of a Bouguer plate of thickness H
and density ρ .
Let us develop the quantity [ A B H ] into a spherical-harmonic expan-
sion (Heiskanen and Moritz, 1967 equation 1-71). then, each degree
constituent is (equation 2.16):
¨
2n + 1 ( )
[ A B H ]n = [ A B H ] P n cos ψ d σ,
4π σ
3 1
In fact, a better place for this layer would be level 4 H. This is one of the
approximations made here.
Ó »îá .
The dipole method 325
in which case (note that the term n = 0 vanishes):

∞
∑ n 1
T= [ A B H ]n ≈ [ A B H ] ,
2n + 1 2
n=1
at least for the higher n values, i.e., regionally if not globally.

Thus is obtained again the indirect effect of Helmert condensation, in
geoid computation by means of this method the shift in geoid surface
caused by the condensation, which must be accounted for with opposite
algebraic sign. In other words, looked upon as a Remove-Restore method,
its “Restore” step:
T 1 A B H πG ρ H 2
δ NHC = ≈ = .
γ 2 γ γ
For comparison, the more precise expansion D.7 yields in approximation

for larger n values
∞
4πG ρ 1 ∑ n + 1 2 πG ρ H 2
δ NHC ≈ · H ≈ ,
γ 2 2n + 1 n γ
n=0
essentially the same result.
Ó »îá .
D The Laplace equation in spherical
E co-ordinates
D E.1 Derivation
Consider a small volume element with sizes in co-ordinate directions of
def
∆ r , ∆φ, ∆λ. Look at the difference in flux of vector field a = ∇V between
what comes in and what goes out through opposite faces.
We do the analogue of what was shown in section 1.12.4, using a body
with surfaces aligned along co-ordinate lines, allowing the size of the
body to go to zero in the limit, and exploiting the Gauss integral theorem
1.18. The quantity div a = ∆V is a volume density, and its average value
multiplied by the volume of a body must equal the total flux through the
surface of the body.
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:
f + − f − ≈ ω a + − a − + a ω+ − ω− .
( ) ( )
See figure E.1.
1. In the radial direction, the surface areas of opposing faces — “inner –

outer” — are
2
r = ( r + ∆ r ) cos φ∆φ∆λ,
ω+
2
r = r cos φ∆φ∆λ,
ω−
difference
r − ω r ≈ 2 r ∆ r cos φ∆φ∆λ.
ω+ −
This is be multiplied by a r = ∂∂Vr , and divided by the volume of the

body r 2 cos φ∆ r ∆φ∆λ, yielding for the contribution to the Laplacian
2 ∂V
∆r′ V = .
r ∂r
This in addition to the “regular” contribution
a+r − a−r ∂2 V
∆r V = ≈ .
∆r ∂r2
– 327 –
328 The Laplace equation in spherical co-ordinates
r ∆φ
r cos φ ∆r
r r ∆λ cos φ
∆φ
∆λ
λ
Equatorial plane
Figure E.1. Gauss integral theorem applied to a co-ordinate conformal volume

D element.
2. Longitudinal direction, λ, “West – East”: no change in surface area

ωλ = r ∆ r ∆φ because of rotational symmetry. We only have
a+ −
∂V + ∂V −
( )
λ − aλ 1
∆λ V = = − ≈
r cos φ∆λ r cos φ∆λ ∂ r λ cos φ ∂ r λ cos φ
1 ∂2 V
≈ .
r 2 cos2 φ ∂λ2
3. Latitudinal direction, φ, “South – North”:
φ = r cos φ + ∆φ ∆ r ∆λ,
ω+
( )
φ = r cos φ∆ r ∆λ,
ω−
difference
φ − ωφ ≈ − r sin φ∆φ∆ r ∆λ.
ω+ −
∂V
Multiply by a φ = ∂rφ
and divide by body volume r 2 cos φ∆ r ∆φ∆λ,
yielding
tan φ ∂V
∆φ′ V = − .
r 2 ∂φ
This of course in addition to the regular contribution
a+ −
φ − aφ ∂V + ∂V − 1 ∂2 V
( )
1
∆φ V = = − ≈ .
r ∆φ r ∆φ ∂rφ ∂rφ r 2 ∂φ2
Ó »îá .
Solution 329
All of this gives us the end result
∆V = ∆r V + ∆λ V + ∆φ V + ∆r′ V + ∆φ′ V =
∂2 V 1 ∂2 V 1 ∂2 V 2 ∂V tan φ ∂V
= + + + − 2 ,
∂r2 r 2 cos2 φ ∂λ2 r 2 ∂φ2 r ∂ r r ∂φ
equivalent to equation 2.7.
D E.2 Solution
D E.2.1 Separating the radial dependency
Let us attempt separation of variables as follows:

( ) ( )
V r, φ, λ = R ( r ) Y φ, λ .
r2
Substitution into equation 2.7 and multiplication by RY yields
∂2 R ∂R 1 ∂2 Y ∂2 Y ∂Y
( ) ( )
1 1
r2 2 + 2r =− + − tan φ .
R ∂r ∂r Y cos φ ∂λ
2 2 ∂φ 2 ∂φ
This must again apply for all values r, φ, λ and thus can only be a constant,
p. This yields two equations
2
2∂ ∂R
( )
R
r + 2r − pR = 0,
∂r 2 ∂r
1 ∂2 Y ∂2 Y ∂Y
( )
+ − tan φ + pY = 0.
cos φ ∂λ
2 2 ∂φ 2 ∂φ
For the first equation we try a power law,
R = rq,
yielding
q ( q − 1)) r q + 2 qr q − pr q = 0
with the solution
p = q ( q + 1) .
( )
Solving the second equation for Y φ, λ ,
1 ∂2 Y ∂2 Y ∂Y
( )
+ − tan φ + q ( q + 1) Y = 0, (E.1)
cos φ ∂λ
2 2 ∂φ 2 ∂φ
is trickier. It turns out that q must be an integer. One finds, for n ∈
N0 , that there are positive solutions1 q = n and negative solutions q =
− ( n + 1), with n = 0, 1, 2, . . . With this, the full set of special solutions is
( )
Y n φ, λ
Vn,1 = r n Yn φ, λ , Vn,2 = , n ∈ N0 .
( )
r n+1
i.e., equations 2.8.
1
The above derivation logic doesn’t work for the exponent values q = 1 or q = 0,
but the result holds also for these values.
Ó »îá .
330 The Laplace equation in spherical co-ordinates
D E.2.2 Solving for surface harmonics
Note that both solutions q, the positive and the negative one, produce on
substitution into equation E.1 the same equation for n.
1 ∂2 Y ∂2 Y ∂Y
( )
+ − tan φ + n ( n + 1) Y = 0.
cos φ ∂λ
2 2 ∂φ 2 ∂φ
We attempt again separation of variables:

( ) ( )
Yn φ, λ = F φ L (λ) .
cos2 φ
Substitution and multiplication by FL yields
cos2 φ ∂2 F ∂F 1 ∂2 L
( )
− tan φ + n ( n + 1) F =− .
F ∂φ2 ∂φ L ∂λ2
Both sides must be again equal to the same constant, which we shall
assume positive and call m:
∂2 F ∂F m2
( )
− tan φ + n ( n + 1) − F = 0,
∂φ2 ∂φ cos2 φ
∂2 L
+ m 2 L = 0.
∂λ2
The first equation is known as Legendre’s equation. Its solutions are the
( )
Legendre functions P nm sin φ , with the integer m = 0, 1, 2, . . . , n. The
second is the classical harmonic oscillator, with solutions2
L m,1 (λ) = cos mλ, L m,2 = sin mλ .
With this, we find for the surface spherical harmonics

( ) ( )
Yn φ, λ = P nm sin φ cos mλ,
( ) ( )
Yn φ, λ = P nm sin φ sin mλ.
The general solution is now formed as a linear combination

∞
∑ n
∑
n
( ) ( )
V1 r, φ, λ = r P nm sin φ [a nm cos mλ + b nm sin mλ] ,
n=0 m=0
∞ n
( ) ∑ 1 ∑ ( )
V2 r, φ, λ = P nm sin φ [a nm cos mλ + b nm sin mλ] .
r n+1
n=0 m=0
Here, a nm and b nm are the spherical-harmonic coefficients describing

the linear combination of special solutions. For describing the Earth’s
gravitational field, only the second solution is physically realistic, going
to zero at infinity r → ∞.
2
This also explains why m must be an integer.
Ó »îá .
Bibliography
Ole Balthasar Andersen, Per Knudsen, and Philippa Berry. The DNSC08GRA
global marine gravity field from double retracked satellite altimetry. Journal
of Geodesy, 84(3), 2010. DOI: 10.1007/s00190-009-0355-9. 176
Georges Balmino, N. Vales, S. Bonvalot, and A Briais. Spherical harmonic

modeling to ultra-high degree of Bouguer and isostatic anomalies. Journal of
Geodesy, 2012. doi: DOI10.1007/s00190-011-0533-4. http://bgi.omp.obs-mip.
fr/content/download/1043/7111/file/2011_Balmino_JOGEOD.pdf. 45, 103
Tulu Besha Bedada. Absolute geopotential height system for Ethiopia. PhD
thesis, University of Edinburgh, 2010. URL: https://www.era.lib.ed.ac.uk/
bitstream/1842/4726/3/Bedada2010_small.pdf. 176
Mirjam Bilker-Koivula. Development of the Finnish Height Conversion Surface

FIN2005N00. Nordic Journal of Surveying and Real Estate Research, 7(1):
76 – 88, 2010.
http://ojs.tsv.fi/index.php/njs/article/download/3663/3432. 289
Mirjam Bilker-Koivula and Matti Ollikainen. Suomen geoidimallit ja niiden

käyttäminen korkeuden muunnoksissa. Tiedote 29, Geodeettinen laitos,
2009.
http://www.fgi.fi/fgi/sites/default/files/publications/gltiedote/GLtiedote29.pdf.
175
Gian Paolo Bottoni and Riccardo Barzaghi. Fast collocation. Bulletin géodésique,
67:119- 126, 1993. 207
J. M. Brozena. The Greenland Aerogeophysics Project - Airborne gravity, topo-

graphic and magnetic mapping of an entire continent. In O. L. Colombo, editor,
From Mars to Greenland: Charting Gravity with Space, pages 203 – 214, 1992.
229
J. M. Brozena, M. F. Peters, and R. Salman. Arctic airborne gravity measure-

ments program. In J. Segawa, H. Fujimoto, and S. Okubo, editors, Proceedings,
IAG International Symposium on Gravity, Geoid and Marine Geodesy (GraGeo-
Mar96), Tokyo, Sep. 30 – Oct. 5, 1996, number 117 in International Association
of Geodesy Symposia, pages 131 – 146, Tokyo, 1996. International Association
of Geodesy, Springer Verlag. 229
– 331 –
332 Bibliography
John M. Brozena and Mary F. Peters. State-of-the-art airborne gravimetry. In

Hans Sünkel and Iginio Marson, editors, Gravity and Geoid, number 113
in International Association of Geodesy Symposia, pages 187 – 197, Graz,
Austria, 1994. International Association of Geodesy, Springer Verlag. 229
Henry Cavendish. Experiments to determine the Density of the Earth.

Philosophical Transactions of the Royal Society, 1798. http://rstl.
royalsocietypublishing.org/content/88/469. 3
B. 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. http://adsabs.harvard.edu/abs/
2013AGUFM.G13C..03C. 229
Giancarlo d’Agostino, Sergio Desogus, Alessandro Germak, Claudio Origlia,

Danilo Quagliotti, Giovanna Berrino, Gennaro Corrado, Vincenzo d’Errico,
and Giuseppe Ricciardi. The new IMGC-02 transportable absolute gravimeter:
measurement apparatus and applications in geophysics and volcanology. An-
nals of Geophysics, 51(1):39–49, 2008. doi: http://dx.doi.org/10.4401/ag-3038.
URL: http://www.annalsofgeophysics.eu/index.php/annals/article/download/
3038/3081. 222
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. http:
//stacks.iop.org/0957-0233/20/i=2/a=022001. 223
Heiner Denker. Evaluation and Improvement of the EGG97 Quasigeoid Model

for Europe by GPS and Leveling Data. In Vermeer and Ádám (1998), pages
53 – 61. 288
Simon Ekholm, René Forsberg, and John M. Brozena. Accuracy of satellite

altimeter observations over the Greenland ice sheet. Journal of Geophysical
Research, 100(C2):2687 – 2696, 1995. http://onlinelibrary.wiley.com/doi/10.
1029/94JC03042/pdf. 229
Martin Ekman. Postglacial Rebound and Sea Level Phenomena with Special
Reference to Fennoscandia and the Baltic Sea. In Juhani Kakkuri, editor,
Geodesy and Geophysics, lecture notes, NKG Autumn School. Finnish Geodetic
Institute Publication 115, 1992. 241, 283
Loránd Eötvös. Three Fundamental Papers of Loránd Eötvös. Loránd Eötvös

Geophysical Institute of Hungary, 1998. ISBN 963-7135-02-2. 299 pp. 230
W. E. Featherstone. Software for computing five existing types of determin-

istically modified integration kernel for gravimetric geoid determination.
Computers and Geosciences, (29):183 – 193, 2003. http://citeseerx.ist.psu.edu/
viewdoc/download?doi=10.1.1.579.4845&rep=rep1&type=pdf. 158
R. Forsberg and M. Vermeer. A Generalized Strang van Hees Approach to Fast

Geopotential Inversion. Manuscripta geodaetica, 17:302 – 314, 1992. 172
Óîá .
Bibliography 333
René Forsberg and Jānis Kaminskis. Geoid of the Nordic and Baltic region from
gravimetry and satellite altimetry. In J. Segawa, H. Fujimoto, and S. Okubo,
editors, Proceedings, IAG International Symposium on Gravity, Geoid and
Marine Geodesy (GraGeoMar96), Tokyo, Sep. 30 – Oct. 5, 1996, number 117 in
International Association of Geodesy Symposia, pages 540 – 547, Tokyo, 1996.
International Association of Geodesy, Springer Verlag. 288
René Forsberg and Carl Christian Tscherning. An overview man-

ual for the GRAVSOFT Geodetic Gravity Field Modelling Programs,
2008. https://www.academia.edu/9206363/An_overview_manual_for_the_
GRAVSOFT_Geodetic_Gravity_Field_Modelling_Programs. 174
Emmanuel S. Garcia, David T. Sandwell, and Walter H.F. Smith. Retracking

CryoSat-2, Envisat and Jason-1 radar altimetry waveforms for improved
gravity field recovery. Geophysical Journal International, 2014. doi: 10.1093/
gji/ggt469. URL: http://topex.ucsd.edu/sandwell/publications/142.pdf. 176
A. Gatti, M. Reguzzoni, F. Migliaccio, and F. Sansò. Space-wise grids of gravity

gradients from GOCE data at nominal satellite altitude. Presented at the 5th
International GOCE User Workshop, November 25-28, 2014, UNESCO, Paris,
France, 2014. http://icgem.gfz-potsdam.de/ICGEM/shms/go_cons_gcf_2_spw_
r4.gfc. 204
Olav R. Godø, Annette Samuelsen, Gavin J. Macaulay, Ruben Patel, Sol-

frid Sætre Hjøllo, John Horne, Stein Kaartvedt, and Johnny A. Jo-
hannessen. Mesoscale Eddies Are Oases for Higher Trophic Ma-
rine Life. PLoS One, 7(1):e30161, 2012. doi: 10.1371/journal.pone.
0030161. http://www.plosone.org/article/fetchObject.action?uri=info:doi/10.
1371/journal.pone.0030161&representation=PDF. 243
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. 172
J. C. Harrison and M. Dickinson. Fourier transform methods in local gravity

modelling. Bulletin Géodésique, 63:149 – 166, 1989. 178
Markku Heikkinen. Solving the Shape of the Earth by Using Digital Density
Models. Report 81:2, Finnish Geodetic Institute, Helsinki, 1981. 73, 75, 227
W. E. Heiskanen and H. Moritz. Physical Geodesy. W.H. Freeman and Company,

San Francisco, London, 1967. 27, 40, 42, 49, 57, 58, 59, 68, 69, 70, 74, 75, 76,
77, 83, 86, 113, 115, 130, 131, 135, 141, 142, 143, 145, 148, 289, 314, 324
C. Hirt and M. Kuhn. Band-limited topographic mass distribution gener-

ates full-spectrum gravity field: Gravity forward modeling in the spectral
and spatial domains revisited. J. Geophys. Res. Solid Earth, 119, 2014.
doi:10.1002/2013JB010900. 103
R.A. Hirvonen. The continental undulations of the geoid. PhD thesis, Helsinki
University of Technology, 1934. Finnish Geodetic Institute Publication 19.
289
Óîá .
334 Bibliography
Bernhard Hofmann-Wellenhof and Helmut Moritz. Physical geodesy. Springer-

Verlag Wien GmbH, 2006. Second, revised edition. 289
Erkki Hytönen. Absolute gravity measurement with long wire pendulum. PhD
thesis, University of Helsinki, 1972. Finnish Geodetic Institute Publication
75. 214
J. Kakkuri, editor. Geodesy and Geophysics, lecture notes, NKG Autumn School
1992, Finnish Geodetic Institute Publication 115, 1981. 165
Juhani Kakkuri. Surveyor of the Globe – Story of the Life of V.A. Heiskanen.
Maanmittauslaitos, 2008. URL: https://readymag.com/u95015526/508134/.
107, 289
S. C. Kenyon, R. Forsberg, A. V. Olesen, and S. A. Holmes. NGA’s use of

aerogravity to advance the next generation of Earth Gravitational Models.
AGU Fall Meeting Abstracts, page A4, December 2012. http://adsabs.harvard.
edu/abs/2012AGUFM.G12A..04K. 229
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. 114
Michael Kuhn, Will Featherstone, and Jonathan Kirby. Complete spherical

Bouguer gravity anomalies over Australia. Australian Journal of Earth
Sciences, 56(2):213 – 223, 2009. http://espace.library.curtin.edu.au/cgi-bin/
espace.pdf?file=/2010/06/21/file_1/135131. 103, 104
Aimo Kuivamäki, Paavo Vuorela, and Markku Paananen. Indications of post-

glacial and recent bedrock movements in Finland and Russian Karelia. Tech-
nical Report YST-99, Geological Survey of Finland, 1998. 247
Erkki Kääriäinen. The Second Levelling of Finland in 1935–1955. Publication 61,

Finnish Geodetic Institute, Helsinki, 1966. 120, 121
Jussi Kääriäinen and Hannu Ruotsalainen. Tilt measurements in the under-

ground laboratory Lohja 2, Finland, in 1977–1987. Publication 110, Finnish
Geodetic Institute, Helsinki, 1989. 282
Rick Lumpkin and Zulema Garraffo. Evaluating the decomposition of tropical

atlantic drifter observations. Journal of Atmospheric and Oceanic Technol-
ogy, pages 1403 – 1415, 2005. http://journals.ametsoc.org/doi/pdf/10.1175/
JTECH1793.1. 243
P. Melchior. The Tides of the Planet Earth. Pergamon Press, Oxford, 1978. 280
Jerry X. Mitrovica, Mark E. Tamisiea, James L. Davis, and Glenn A. Milne.

Recent mass balance of polar ice sheets inferred from patterns of global sea
level change. Nature, 409:1026 – 1029, February 2001. 246
Óîá .
Bibliography 335
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). 95, 126, 157,
289
Helmut Moritz. Advanced Physical Geodesy. H. Wichman Verlag, Karlsruhe,

1980. 289
J. Mäkinen, A. Engfeldt, L. Engman, B.G. Harsson, T. Oja, S. Rekkedal,

K. Røthing, P. Rouhiainen, H. Ruotsalainen, H. Skatt, G. Strykowski, H. Virta-
nen, K. Wieczerkowski, and D.Wolf. The Fennoscandian Land Uplift Gravity
Lines: comparison of observed gravity change with observed vertical motion
and with GIA models. 2010. http://www.nkg.fi/nkg/sites/default/files/Makinen_
et_al_land_uplift_gravity_lines.pdf. 239
Silja Märdla. Regional Geoid Modelling by the Least Squares Modified Hotine
Formula Using Gridded Gravity Disturbances. PhD thesis, Tallinn University
of Technology, 2017. ISBN 978-9949-83-186-9. 96, 151
R.S. Nerem, D. 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. 253
R.L. Parker. The rapid calculation of potential anomalies. Geophys J R Astr Soc,
31:447 – 455, 1972. http://topex.ucsd.edu/geodynamics/parker.pdf. 179
Nikolaos K. Pavlis, Simon A. Holmes, Steve C. Kenyon, and John K. Fac-

tor. The development and evaluation of the Earth Gravitational Model
2008 (EGM2008). Journal of Geophysical Research, 117(B4), 2012. http:
//onlinelibrary.wiley.com/doi/10.1029/2011JB008916/epdf. 55
N.K. Pavlis, S.A. Holmes, S.C. Kenyon, and J.K. Factor. An Earth Gravitational
Model to Degree 2160: EGM2008. In EGU General Assembly 2008, Vienna,
Austria, April 13-18, 2008, 2008. http://earth-info.nga.mil/GandG/wgs84/
gravitymod/egm2008. 55
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. ISSN 0277-3791. doi: http://dx.doi.org/10.1016/j.
quascirev.2009.04.004. URL http://www.sciencedirect.com/science/article/pii/
S0277379109001218. Quaternary Ice Sheet-Ocean Interactions and Land-
scape Responses. 244
M. Poutanen, M. Vermeer, and J. Mäkinen. The Permanent Tide in GPS Posi-

tioning. Journal of Geodesy, 70:499 – 504, 1996. 283
John Henry Pratt. The attraction of the Himalaya mountains upon the plumbline
in India. Philosophical Transactions of the Royal Society of London, CXLV:
53 – 100, 1855. 106
Óîá .
336 Bibliography
John Henry Pratt. The deflection of the plumb-line in India and the com-
pensatory effect of a deficiency of matter below the Himalaya mountains.
Philosophical Transactions of the Royal Society of London, CXLIX:745 – 778,
1858. 106
John Henry Pratt. Speculations on the constitution of the Earth’s crust. Pro-
ceedings of the Royal Society of London, XIII:253 – 276, 1864. 106
K. Predehl, G. Grosche, S.M.F. 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. April 27 2012.
http://www.sciencemag.org/content/336/6080/441.abstract. 137
I. Prutkin. Gravitational and magnetic models of the core-mantle boundary and

their correlation. Journal of Geodynamics, 45:146 – 153, 2008.
http://www.researchgate.net/profile/Ilya_Prutkin/publication/257097255_
Gravitational_and_magnetic_models_of_the_coremantle_boundary_and_
their_correlation/links/0c960527cd74c9de59000000.pdf. 114
Jean Richer. Observations astronomiques et physiques faites en l’isle de Caïenne.

In Ouvrages de Mathematique de M. Picard. P. Gosse and J. Neaulme, 1731.
http://www.e-rara.ch/zut/content/pageview/815403. 213
Reiner Rummel and Fernando Sansó, editors. Satellite Altimetry in Geodesy

and Oceanography. Proceedings, International Summer School of Theoretical
Geodesy, May 25 – June 6, 1992, Lecture Notes in Earth Sciences, Heidelberg,
1992. Springer-Verlag. 243
Hannu Ruotsalainen. Interferometric Water Level Tilt Meter Development

in Finland and Comparison with Combined Earth Tide and Ocean Loading
Models. Pure and Applied Geophysics, 2017. https://link.springer.com/article/
10.1007/s00024-017-1562-6, DOI 10.1007/s00024-017-1562-6. 282
Oliver Sacks. Henry Cavendish: An early case of Asperger’s syndrome? Neurol-

ogy, 57(7):1347–1347, 2001. ISSN 0028-3878. doi: 10.1212/WNL.57.7.1347.
URL http://n.neurology.org/content/57/7/1347. 3
E. J. O. Schrama. The Role of Orbit Errors in Processing of Satellite Altimeter

Data. PhD thesis, Delft University of Technology, 1989. https://www.ncgeo.nl/
downloads/33Schrama.pdf. 261
W. Torge. Gravity and Tectonics. In Geodesy and Geophysics, lecture notes, NKG
Autumn School, Finnish Geodetic Institute Publication 115, 1992. 285
Wolfgang Torge. Gravimetry. W. de Gruyter, Berlin – New York, 1989. 289
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. Technical report, Dept. of Geodetic Science and
Surveying, The Ohio State University, Columbus, OH, USA, 1974. 203
Óîá .
Bibliography 337
Peter Vaníček and Edward Krakiwsky. Geodesy – The Concepts. Elsevier Science
Publishers, Amsterdam, 1986. 289
Felix Andries Vening Meinesz. Gravity Survey by Submarine Via Panama

to Java. The Geographical Journal, 71(2):144–156, 1928. ISSN 00167398,
14754959. URL: http://www.jstor.org/stable/1782700. 214
M. Vermeer. Chronometric levelling. Report 83:2, Finnish Geodetic Institute,

1983. ISSN 0355-1962. 136, 223
Martin Vermeer. Geoid studies on Finland and the Baltic. PhD thesis, University
of Helsinki, 1984. 26, 83
Martin Vermeer. Terrain Reduction and Gridding Techniques for Geoid Deter-
mination. In Geodesy and Geophysics, lecture notes, NKG Autumn School,
Finnish Geodetic Institute Publication 115, 1992. 179
Martin Vermeer and József Ádám, editors. Proceedings, Second Continental

Workshop on the Geoid in Europe, March 10 – 14, 1998, Report 98:4, Finnish
Geodetic Institute, Masala, 1998. 332, 337
H. Virtanen. On superconducting gravimeter observations above 8 mHz at the

Metsähovi station. Report 98:5, Finnish Geodetic Institute, Masala, 1998. 19
p. 225
Heikki Virtanen. Studies of earth dynamics with superconducting gravimeter.

PhD thesis, University of Helsinki, 2006. http://ethesis.helsinki.fi/julkaisut/
mat/fysik/vk/virtanen/studieso.pdf. 225
Heikki Virtanen and Jussi Kääriäinen. The installation and first results from
the superconducting gravimeter GWR20 at the Metsähovi station, Finland.
Report 95:1, Finnish Geodetic Institute, Helsinki, 1995. 15 p. 225
Phuong Lan Vu, Frédéric Frappart, José Darrozes, Vincent Marieu, Fabien
Blarel, Guillaume Ramillien, Pascal Bonnefond, and Florence 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://www.mdpi.com/2072-4292/10/1/93/pdf. 266
Lianxing Wen and Don 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. doi: http://dx.doi.org/10.1016/S0012-821X(96)
00238-5. http://222.195.83.195/wen/Reprints/WenAnderson97EPSL.pdf. 114
Hans-Georg Wenzel. Ultra High Degree Geopotential Model GPM3E97A to

Degree and Order 1800 Tailored to Europe. In Vermeer and Ádám (1998),
pages 71 – 80. 59
L. Wong and R. Gore. Accuracy of geoid heights from modified Stokes kernels.
Geophys J R Astr Soc, 18:81 – 91, 1969. 155
Óîá .
338 Bibliography
G. Wöppelmann, C. Letetrel, A. Santamaria, M.-N. Bouin, X. Collilieux, Z. Al-

tamimi, S. D. P. Williams, and B. Martin Miguez. Rates of sea-level change
over the past century in a geocentric reference frame. Geophysical Research
Letters, 36(12), 2009. DOI: 10.1029/2009GL038720, http://onlinelibrary.wiley.
com/doi/10.1029/2009GL038720/full. 238
Dah-Ning Yuan, William L. Sjogren, Alex S. Konopliv, and Algis B. Kucinskas.

Gravity field of Mars: A 75th Degree and Order Model. Journal of Geophysical
Research, 106(E10):23377 – 23401, 2001. http://onlinelibrary.wiley.com/doi/10.
1029/2000JE001302/pdf. 203
Óîá .
Index
ABCDEFGHIJKLM degree constituent equation, 49

NOPQRSTUVWXYZ forward geodetic problem, 185
generating function, 142
A Helmert condensation, 319
a priori information, 260 sea-level equation, 246
a priori variance, 259 Stokes equation, 139
Aalto University, 289 Stokes kernel, 167
acceleration, 3 tangent plane, 163
geometric, 270 anisotropy, 142
measured by GNSS, 228 anomalous quantity, 63
measured by gravimeter, 228 Antarctic Ocean, 267
of aircraft, 228 Antarctica, 96, 176, 229, 244, 246
of free fall, 220 antimatter, 20
satellite, 270 anti-root, of mountain, 110
accelerometer, 229, 270, 272 Apollo project, 3, 220
on GRACE, 271 Arabelos, Dimitris, 174
action at a distance, 1 Archimedes’ law, 108
Agulhas Stream, 235 archipelago, 267
air drag compensation, 229 Arctic, 176, 229
air pressure, variations, 226 Arctic Ocean, 267
airborne gravimetry, 176, 269 ice cover, 268
description, 227 ice volume, 268
homogeneity, 229 argument of perigee, 262
Airy, George Biddell, 107 arrest (gravimeter), 219
Airy–Heiskanen hypothesis, 108, 113 ascending node, 262
Airy–Heiskanen model, 107 of the Moon, 279
Åland, 267 Asperger syndrome, 3
Alaska (USA), 110 associated Legendre function, 42–44
altimetric satellite, 251, 262 fully normalized, 49
Amazonas (Brazil), 176, 271 symmetry, 46
American Geophysical Union (AGU), 288 astatization, 219
amplifier, in optic fible cable, 136 astatization ratio, 217, 218
Amsterdam (The Netherlands), 120, 237 asthenosphere, 239, 244
analysis, a posteriori, 267 Atlantic Ocean
angular distance, geocentric North, 243
definition, 185 salinity, 241
covariance function, 186, 190 atmosphere
– 339 –
340 Index
attraction, 226 Bouguer anomaly, 98

surface mass density, 226 bias, 99, 100
total mass, 227 calculation steps, 101, 102
atmospheric drag, 269, 271, 272 example, 102
atmospheric loading, 285 properties, 99
atmospheric refraction, 119 simple, 98
atomic clock, 136 Southern Finland, 100
attenuation, gravitation, with height, 35 spherical, 103
attraction bias, 104
exterior, 15 terrain corrected, 100
interior, 15 Bouguer hypothesis, 239
spherical shell, 6, 7 Bouguer hypothesis, of land uplift, 239,
attraction, gravitational, 4 240
autocovariance, 183 Bouguer plate, 1
autoregressive process, 190 as approximation, 98, 129
average density, of Earth’s crust, 98 attraction, 97, 324
averaging over ocean surface, 246 double, 131
azimuth, 142, 163, 185, 186 half, 102
definition, 185 of air, 226
Bouguer reduction, 96, 149, 154
B simple, 99
Baltic Sea Bouguer shell
airborne gravimetry, 229 attraction, 103
mean sea surface, 252 Bouguer, Pierre, 96, 106
salinity, 241 Boulder, University of Colorado at, USA,
sea-surface topography, 236, 241 220, 253
barometer, 85 boundary condition, 34, 89, 311
Barzaghi, Riccardo, 173 boundary surface
base network measurement, 224 choice, 95
basis boundary-value problem, 26, 27, 32, 95
complete, 306, 313 definition, 26
in a function space, 313 free, 86
in a vector space, 305, 309 of Dirichlet, 26, 27, 88, 144
basis vector, 305, 306 of Neumann, 52, 88
bathymetry, 98, 103 of physical geodesy, 88
benchmark, 256 spectral solution, 89
Bergensbanen, 110 third, 86, 88
Bessel’s equation, 313 bounded support, 164
BGI, 287 box, rectangular, 18, 19
bias, of measurement, 253 Brovelli, Maria, 173
bilinear form, 305, 309 Bruns equation, 83, 129, 139, 284
Bjerhammar sphere, 204 Bruns vertical-gradient equation, 68
Bjerhammar, Arne, 204 Bruns, Ernst Heinrich, 68, 83
block surface area, 158
block weight, 158 C
blue film, Earth’s, 270 cage, in absolute gravimeter, 220
Blue Road Geotraverse project, 239 Calgary (Canada), 173
book-keeping, 18 calibration
bordering, 173 gravimeter, 226
Bose–Einstein condensate, 136, 223 in-flight, 254, 266
Bothnian Bay (Finland, Sweden), 241 radar altimeter, 254, 266
Bouguer anomalies calibration certificate, 226
interpolation, 99 cannon, 62
prediction, 99 carbon dioxide, 241
Óîá .
Index 341
Cavendish, Henry, 3, 262 convolution theorem, 165, 172

celestial mechanics, 10 co-ordinate conversion, 39
central force field, 293 co-ordinate reference system
centrifugal acceleration, 64 co-rotating, 64, 231
centrifugal force, 64, 65, 231 inertial, 64
centrifugal potential, 64 co-ordinate time, 136
expression, 65 co-ordinate transformation, 291
CHAMP (satellite), 55, 176, 269, 270 co-ordinates
characteristic equation, 310 cylindrical, 31
Chasles theorem, 1, 26, 27 ellipsoidal, 39
Chasles, Michel, 26 geodetic, 39
checkerboard, 44 definition, 39
chlorophyll, 241 geographic, 168
circular disk, attraction, 97 geographical, 38
climate research, 244 natural, 69
climate, of Earth, 243 polar, 301
clinometer, 282 rectangular, 4
clock, 136 spherical, 23, 31, 38, 41
pendulum, 213 definition, 38
closing error, 219 topocentric, 230
coastline, 267, 285 toroidal, 31
coefficient vector, 309 Copenhagen (Denmark), 173, 288
co-geoid, 95 coral, 244
of isostatic reduction, 113, 117 Coriolis acceleration, 241
coherence, of matter waves, 136 direction, 241
collocation, least-squares (LSC), 112, 175, Coriolis effect, 241
189, 191, 195 Coriolis force, 65, 235, 241
description, 186, 192 Coriolis, Gaspard-Gustave, 65
FFT, 204 correlation, 188
flexibility, 195 correlation length, 190, 211
Columbus geoid (model), 289 correlation, quasi-geoid & topography, 127,
commutative diagram, 36, 51, 154, 167 128
comparison point, for geoid determination, correspondence, integral & spectral
237 equations, 141
compensation depth, 113, 114 cosine rule, 168
component, of a vector, 294 half-angle, 168
Congo (Africa), 271 cosine taper, 173
conservation, of matter, 17 covariance function, 185, 195
conservative field, 120 definition, 185
definition, 4, 298 empirical, 203
curl, 301 Gauss–Markov, 194
potential, 301 global, 204
continental ice sheet, 110, 111, 238, 244 gradient of gravity, 210
continental ice sheets, total mass, 245 isotropic, 198
continental shelf, 110 of gravity anomalies, 201
continuity equation, 302 of Hirvonen, 190–193
convection, in the Earth’s mantle, 114 of the disturbing potential, 197, 210
convolution, 171, 206 in space, 201
calculation, 171 on the Earth surface, 201
kirjoitustapa, 164 spectral representation, 198
linear combination, 169 cross covariance, 183
terrain correction, 177 cross product, 292
two-dimensional, 164 crossover
Óîá .
342 Index
geometry, 256 degrees of freedom, 261

crossover adjustment, 259, 269, 273 Delft (The Netherlands), 173
global, 261 delta function, Dirac’s, 20, 146, 205
crossover condition, 261 Denker, Heiner, 173
crossover difference, 260, 269, 274 density
crossover point, 255, 256, 259 ice, 111, 245, 268
crustal density, 108, 177 mantle, 108
Cryosat-2 (satellite), 252 rock, 69, 131
curl (operator), 297, 298 sea water, 108, 110, 245, 268
interpretation, 297 standard crustal, 132
lineaarisuus, 296 topography, 128, 151
of gradient, 298 upper mantle, 111
of wind field, 297 density model, 27
curvature density profile, 27
of a level surface, 67, 230 density, SI unit, 9
of the Earth, 101, 108, 165 developing country, education, 287
cyclone, tropical, 235 dice throw, 183
difference, geoid – free-air geoid, 129
D difference, height anomaly – geoid height,
damping, of gravimeter, 219, 227, 228 128
Danish straits, 241 difference, orthometric height – normal
Darwin, Sir George, 280 height, 130
datum, 259 difference, quasi-geoid – geoid, 129
datum defect, 259 differential operator, 299
datum point, 120 digital terrain model (DTM), 100, 150
datum transformation, 259, 261 dipole, 15, 54
De Maupertuis, Pierre, 96 dipole density layer, 15, 324
Dead Sea, 115 dipole field, 77
declination, of the Moon, 278 dipole layer element, 15
deformation dipole moment, 53
of the Earth, 280 definition, 15
plastic, 247 of the Earth, 53, 54
deformation coefficient, for viscous vanishing, 57
loading, 247 dipole surface density, 15
deglaciation, last, 111, 238, 247 Dirac, Paul, 20
degree constituent directional sphere, 314
disturbing potential, 77 Dirichlet, Peter Gustav Lejeune, 26
gravity anomaly, 87 Dirichlet’s problem, 1
powers of height, 322, 323 dislocation (crystal), 218
degree constituent equation, 49, 50, 89, disturbing potential, 81, 89, 93
139 definition, 76
data point, 49 at terrain level, 152
evaluation point, 49 in spherical harmonics, 77
harmonic field, 144 local, 165
degree number surface harmonics, 41
of tidal force, 282 divergence (operator), 12, 17, 297, 299, 302
degree of freedom, 261 interpretation, 296
degree variance lineaarisuus, 296
disturbing potential, 197, 199 DMA (Defense Mapping Agency, U.S.), 54
on the Earth surface, 201 Doodson, Arthur Thomas, 280
gravity anomalies, 202, 203 Doodson’s constant, 280, 284
notation, 199 DORIS (instrument), 252
degree variance formula, 203 dot product, 291
degree, harmonic, 41
Óîá .
Index 343
downward continuation, harmonic, 128, computation, 59

129, 149–151, 176, 321 RMS Empress of Ireland, 120
existence, 144, 149 energy, conservation, 293
of r ∆ g, 145 energy, of place, 121
drift (gravimeter), 219, 225, 226 Envisat (satellite), 252
Eötvös (unit), 88, 230
E Eötvös tensor, 230
Earth centre of mass, 38, 54, 55, 77, 87 Eötvös, Loránd, 66
Earth magnetic field, 114 epoch, of land uplift, 237
Earth radius, 14 equations of motion, of satellites, 54
Earth rotation rate, 65, 73 equatorial radius, 39, 72, 261
earthquake, 225, 247 equilibrium length, of spring, 215, 216
Earth’s flattening, 38, 39, 55, 72 equipotential surface, 15
Earth’s gravitational field, 31, 227, 270 as boundary, 25, 26
spectral representation, 1 figure, 84
Earth’s gravity field, 55 equivalence principle, 3, 66
eccentricity ergodicity, 184
orbital, 262 erotus, height anomaly – free-air geoid,
eddiness, in a vector field, 297 129
eddy, 253 error ellipse, 311
eddy phenomena, 65 ERS-1 (satellite), 252
eddy-free flow, 303 ERS-2 (satellite), 252
EGM96 (geopotential model), 54 escape velocity, 62
coefficients, mean errors, 56 estimation, 187
EGM2008 (geopotential model), 45, 55, 93, estimator, 188
100 mean error, 192
Eiffel Tower, 68 optimal, 189
Eiffel Tower’s 72 names, 12, 13, 26, 34, 41, Eterna (software), 285
65, 312 Ethiopia, 176
eigenvalue, 310 Euclidean space, 4, 8, 305
eigenvalue problem, 308–310 Euler angle, 297
eigenvector, 310, 311 Euler’s identity, 318
eight-unit cube, 21 Eurajoki (Finland), 224
Einstein summation convention, 188 European climate, 243
Einstein, Albert, 3 European Geosciences Union (EGU), 288
El Niño Southern Oscillation (ENSO), 235, European Space Agency (ESA), 252, 271
253 eustatic rise, of mean sea level, 238, 245
elasticity, 68, 218 evaluation functional, 182
of Earth, 280 evaluation latitude, 169
of the Earth, 280, 285 Everest, Mount, 110
of the Earth’s crust, 285 expectancy
elasticity model, 283 of a stochastic process, 184
electric charge, conservation, 293 statistical, 184
electric currents, in the Earth’s core, 114 exterior product, 292
electrostatic compensation, 228 exterior surface normal, 17
electrostatic potential, 299 exterior surface normal, ellipsoidal, 64
ellipsoid of revolution, 38
ellipsoidal harmonic, 70 F
ellipsoidal harmonic expansion factorial, 42
centrifugal potential, 70 Falkland islands, 272
convergence, 59 Faller, James E., 220
normal potential, 59 Faraday, Michael, 18, 302
ellipsoidal-harmonic expansion, 58 Fast Collocation, 112, 207
definition, 57 Fast Fourier Transform (FFT), 112
Óîá .
344 Index
and convolution, 165 Fourier transform, 35

and tapering, 174 artefacts, 173
collocation, 204, 206 discrete, 165, 167, 317, 318
mixed-radix, 317 periodicity, 172, 173
radix 2, 317 reverse, 317
terrain correction, 176, 179 forward, 165
fast Fourier transform (FFT) notation, 164
algorithms, 318 of ℓ−3 , 178
Fastest Fourier Transform in the West reverse, 165, 178
(fftw, software), 318 Fourier, Joseph, 34
Father Point / Pointe-au-Père (Rimouski, France, 287
Quebec, Canada), 120 free oscillations, of the solid Earth, 225,
Fennoscandia, 111, 237, 247 289
Fennoscandian Shield, 113 periods, 225
field, 8, 17 free-air anomaly
field equations, 1 definition, 91
of electromagnetism, 13 calculation, 92
of gravitation, 13 linearization, 91
field line, 18, 302 Southern Finland, 93
field theory use, 92
electromagnetism, 31 free-air hypothesis, of land uplift, 239, 240
gravitation, 1 free-air reduction, 101
field, the concept, 31 freeboard, 252, 267, 268
figure of the Earth, 63, 96, 106, 263 French Academy of Sciences, 96
mathematical, 66 frequency domain, 35, 166, 167
FIN2000 (geoid model), 175, 237 function space, 181, 183, 306, 311–313
precision, 176 on the sphre, 313
FIN2005N00 (geoid model), 175, 289 function theory, 313
precision, 176 functional, 200
finite elements method (FEM), 112 definition, 181
Finland, 120, 131, 175, 220, 236 linear, 27, 182, 199, 203
Finnish climate, 243 definition, 182
Finnish Geodetic Institute (FGI), 220, 225, of the disturbin potential, 183
282, 288, 289 of the disturbing potential, 182
Finnish Geospatial Research Institute fundamental equation of physical geodesy,
(FGI), 289 89, 152
first eccentricity, 39
flat Earth model, 108 G
flattening Galilei, Galileo, 3
of a planet, 10 gauge invariance, 13
flight height, 228 Gauss integral theorem, 1, 16, 18, 21, 23,
flow velocity, 17, 296, 302 26, 301, 327, 328
flow velocity field, 242 presentation, 17
fluid motion, 121 in terms of potential, 18
flux, 17 figure, 18, 302
footprint, radar altimeter, 253, 267 Gauss, Carl Friedrich, 17
footscrew, 216 Gauss–Markov process, 191
Forsberg, René, 173, 174 Gelderen, Martin van, 173
Fourier basis function, 42, 47, 306 general relativity, 13, 136
Fourier coefficient, 34, 41 generating function
Fourier expansion, 306 geometry, 143
Fourier series, 306, 318 of the Legendre polynomials, 114, 143
Fourier sine expansion, 34 geodetic forward problem on the sphere,
185
Óîá .
Index 345
Geodetic Reference System 1967 (GRS67), geopotential image, sharpness, 45

74 geopotential model, 270
geographic mean, 184, 186 global, 270
definition, 184 geopotential number, 121, 122
geographic variance, 184 definition, 120
geoid, 86, 114, 123, 235, 322 GEOS-3 (satellite), 251
definition, 66, 122 Geosat (satellite), 252
classical, 236 geostrophic equilibrium, 242
fake, 132 Germain curvature, 68
free-air, 128, 129 Germain, Marie-Sophie, 68
true, 269 German Research Centre for Geosciences
geoid computation, 154 (GFZ), 269
gravimetric, 158 Gibbs phenomenon, 156, 173, 308
software, 174 Gibbs, Josiah Willard, 156
geoid determination, 26, 53, 268, 287 Glacial Isostatic Adjustment (GIA), 238,
1D-FFT, 172 246
classical, 95 glacial maximum, last, 111
FFT, 173, 175 glacier, 115, 238, 245
gravimetric, 141, 239, 240 retreat, 110
principle, 140 global warming, 244
NKG, 288 GM⊕ , best value, 5
precise, 132 GNSS
Spherical FFT height of gravimetric stations, 96
multi-band, 168 in airborne gravimetry, 228
Taylor expansion, 169 in height determination, 174
geoid height, 81, 88, 128, 133, 253, 255 measuring atmospheric loading, 286
definition, 81 measuring ocean tidal loading, 285
from satellite altimetry, 176 orbital tracking, 273
redukoitu, 154 positioning of tide gauges, 238, 240
geoid map, 273 GNSS antenna, 228
geoid model, 113, 270 GNSS instrument
computation, 174 in aircraft, 228
Finland, 83 in altimetric satellite, 269
global GNSS levelling, 237
high resolution, 269 GOCE, 55, 176, 229, 243, 272
precise, 273 description, 271
gravimetric, 237 name, 273
local, 236 GPS
geoid modelling on satellite, 252, 269
global, 107 orbital tracking, 270
geoid rise, 238, 239 reference system, 73
geoid undulation, 81, 82 GRACE (satellite pair)
globally, 81 description, 270
in Finland, 81 GRACE (satellites), 55, 176
geological map, 132 principle, 271
geometric geodesy, 76 GRACE follow-on mission, 271
geophysical data record (GDR), 254, 267 grade measurement, 96
geophysical reduction, 95 gradient
geopotential, 70 of Earth attraction, 272
gradient, 229 of gravity disturbance, 210
level surface, 66 of the potential
on the tangent plane, 67 in the normal direction, 26
spectral expansion, 51 gradient (operator), 8, 9, 19, 296, 298
Óîá .
346 Index
interpretation, 296 measurements, 65

linearity, 296 prediction, 153
of scalar field, 295 total, 228
gradient vector, 84 gravity acceleration, 66
gravimeter average, 246
absolute, 220, 239 from pendulum, 214
principle of operation, 220 on the plumb line, 69
airborne, 219, 227 gravity anomaly, 197
astatized, 216, 217 a priori estimate, 194
atomic, 223 as a boundary condition, 88
principle of operation, 223 as a functional, 182
ballistic, 214, 220 at sea level, 145, 148
FG5, 220, 221 at topography level, 148
IMGC-02, 222 atmospheric reduction, 227
JILA, 220 availability, 181
LaCoste-Romberg, 216, 217, 221 block average, 158, 159, 195
pendulum, 213 calculation, 86, 91
quantum, 223 change, 239
registering, 285 estimate, 193, 194
relative, 239 from satellite altimetry, 176
sea, 219, 269 global average, 184
spring, 214, 215, 217, 218, 224 in the external space, 145
superconducting, 224–226 interpolated from grid, 160
gravimetry mean error, 193
limitation, 27 observations, 192
satellite, 229 reduced, 154
gravitation, 2 surface harmonics, 41
gravitational acceleration, 9, 10, 26, 66 gravity disturbance, 210
measurements, 53 definition, 84
gravitational acceleration vector, 8, 12 observing, 85
gravitational constant, universal, 3, 277, spectral representation, 85
293 gravity field, 3, 63, 122
gravitational field, 4, 17 behaviour, 190
attenuation with height, 35 change, 246
of celestial objects, 283 determination, 203, 270, 272
gravitational gradiometer (GOCE), 269 exterior, 92, 152
figure, 272 fine structure, 269
gravitational lense, 2 GOCE resolution, 229
gravitational potential, 1 observations, 181
in vacuum, 299 oceanic, 252
rotationally symmetric, 70 of mountains, 106
gravitational vector, 8 research in Europe, 288
gravitational wave, 13 research in Finland, 288
gravity, 96, 214, 221, 223, 228, 239 research in HUT, 289
definition, 220 research internationally, 287
absolute measurement, Finland, 224 residual, 153
along levelling line, 135 spatial variability, 165
equatorial, 73 temporal change, 270
in airborne gravimetry, 228 textbooks, 289
in the tropics, 213 very local features, 273
local, 123, 213, 228 gravity formula, 63, 72, 74, 92
measured, 214 gravity gradient, 88, 230, 282
measurement, 86 free air, 285
Óîá .
Index 347
gravity mapping survey, 224 Hannover (Germany), 173

gravity potential Hardanger plateau (Norway), 110
as sum of gravitational and harmonic continuation
centrifugal potentials, 65 of gravity anomalies, 146
gradient, 301 harmonic continuation, of the potential,
in spherical harmonics, 76 200
of falling atoms, 223 harmonic field
gravity vector, 122 definition, 13
gravity vector field, 301 r ∆ g, 145
gravity versus gravitation, 64 harmonic oscillator, 33, 330
gravity-gradient field hat notation, 195
of Sun and Moon, 231 Hayford ellipsoid, 74, 91
gravity-gradient tensor, 230 Hayford, John Fillmore, 106
GRAVSOFT (software), 174 heat transport, 241, 243
Green equivalent-layer theorem, 26 Heaviside function, 247
Green theorem hegiht
for external space, 25 above mean sea level, 123
Green, George, 22 height
Greenland, 96, 176, 244, 246 above mean sea level, 121
Greenland Aerogeophysics Project (GAP), above the geoid, 122
229 above the reference ellipsoid, 39, 82,
Green’s function, 246, 247 124, 253
of sea level, 246 height anomaly, 90, 91, 124, 127, 128
of the geopotential, 246 definition, 124
of vertical displacement, 246 height system, 289
partial, of deformation, 246 national, 236, 237
Green’s theorem, 1 of a country, 259
exterior point, 23 Heiskanen, Veikko Aleksanteri, 107, 289
first, 22 helicopter, 228
interior point, 24 helium, liquid, 225, 226
second, 22 Helmert condensation, 1, 104–106, 149,
third, 22, 95 319, 323, 324
Green’s theorems, 1 gravity effect, 323
Greenwich meridian, 38 Helmert ellipsoid, 74
grid Helmert height, 128, 131, 138, 237
disturbing potential, 166 as approximation, 131
gravity anomaly, 165 Helmert, Friedrich Robert, 104
Stokes kernel, 166 Helsinki astronomical observatory, 120,
grid integration, 112 121
grid matrix calculation, 171 Helsinki harbour, 120, 236, 259
grid representation, 165 Helsinki University of Technology (HUT,
GRS80 TKK), 289
definition, 73 Helsinki, Finland, 237
GM⊕ , 5 Himalayas, 106
Gulf of Finland, 241 Hirvonen, Reino Antero, 190
airborne gravimetry, 229 Hirvonen’s geoid model, 289
Gulf Stream, 235, 242, 243 Hofmann-Wellenhof, Bernhard, 289
Guyana, French, 213 homogeneity
GWR 20 (gravimeter), 225 of data precision, 207
GWR iGrav (gravimeter), 226 of gravimetric data, 229
homogeneity assumption, 184, 185
H horizontal gradient, of gravity anomalies,
Haagmans, Roger, 173 159
half-height point, 267
Óîá .
348 Index
hour angle, of the Moon, 278 inversion calculation, 167, 172

Hubble Space Telescope, 2 inverted barometer, 235
Huygens, Christiaan, 213 ISG, 287
HY-2A (satellite), 252 isostasy, 106, 107
hydrodynamics, 283 modern understanding, 110, 111
hydrosphere, 270 isostasy hypothesis, 106
isostatic anomaly
I definition, 112
IAG (International Association of interpolation, 112
Geodesy), 175, 287 prediction, 112
ice age, last, 239, 244, 247 Southern Finland, 113
ice cap, 238 isostatic compensation, 108, 109
ice load, 239, 246 definition, 106
history, 246 percentage, 114
ice sheet, 245, 247 isostatic equilibrium, 214
Laurentide, 244 isostatic geoid, 113
ice, multi-year, 268 interest, 114
ICET, 288 isostatic hypothesis, 106, 108, 112
ICGEM, 288 isostatic reduction, 112, 154
IDEMS, 288 description, 114
identity matrix, 205 and density layers, 115
IGeC, 287, 288 residual field, 112
IGeS, 287 isotropic density distribution, 7
IGFS, 287 isotropic process, 190
ill-posed problem, 149 isotropy, 142
inclination, orbital, 262–265 of the disturbing potential, 197
of the Moon, 279 of the viscosity, 246
incompressibility, 17, 303 isotropy and spectral representation, 142
independence, statistical, 192 isotropy assumption, 186
India, 252 Italy, 287
indirect effect, 95 iteration
of a constant terrain, 323 calculation of normal height, 132
of Bouguer reduction, 114, 149 calculation of orthometric height, 131
of Helmert condensation, 149, 323, orthometric height calculation, 123
325
of isostatic reduction, 113, 116, 149 J
of tidal potential, 282, 283 J2 (dynamic flattening), 55, 73, 263
Indonesia, 267, 271 Jacobi, Carl Gustav Jacob, 142
inertial tensor, 54 Jacobi’s determinant
instantaneous length, of spring, 218 definition, 301
Institut für Erdmessung (Hannover, map projection co-ordinates, 164
Germany), 173 polar co-ordinates, 161, 301
Institut für Erdmessung, Hannover spherical co-ordinates, 142, 163, 301
(Germany), 288 Jason (satellites), 252, 264
integrability, 313 Java Sea (Dutch Indies, Indonesia), 214
integral equation, 2 Jerry (GRACE satellite), 270
integration by parts, 312 Joensuu (Finland), 224
intercomparison, absolute gravimeters,
222, 224 K
International Isostatic Institute, 289 Kaivopuisto (Helsinki, Finland), 120, 121,
International Union of Geodesy and 237
Geophysics (IUGG), 287 Kater, Henry, 213
interpolation from grid, 166, 167 Kater’s reversion pendulum, 213
invariant, 311 Kaula, William, 203
Óîá .
Index 349
Kepler, Johannes, 158, 261 reduced, 39, 72, 78

Kepler’s laws, 261 types, 72
Kepler’s orbital elements, 262 least-squares adjustment, 259
Kepler’s second law, 293, 294 least-squares method, 222, 237
Kepler’s third law, 261, 264, 274 ordinary, 256
kernel modification Legendre function, 40, 41, 330
advanced, 156 fully normalized, 314
coefficients, 157 of the second kind, 57, 59
degree, 155 Legendre polynomials
sharp cut-off, 156 as a basis, 313
Wong-Gore, 155, 157 figure, 43
Kevo (Finland), 224 fully normalized, 48
Kirkkonummi (Finland), 225, 237 orthogonality on the interval [−1, +1],
KKJ (National Grid Co-ordinate System), 47
91 orthogonality on the unit sphere, 48
Knudsen, Per, 174 orthonormality on the unit sphere, 49
Kolkata (India), 106 table, 42
Krasovsky ellipsoid, 74 Legendre, Adrien-Marie, 41
Kronecker delta, 205 Legendre’s equation, 313, 330
Kronstadt (Russia), 120 Lego™ brick, 20
Kuusamo (Finland), 224 Leibniz University (Hannover, Germany),
Kääriäinen, Jussi, 229 288
Leibniz, Gottfried Wilhelm, 1
L level surface, 67, 122, 236, 273
LaCoste, Lucien, 217 definition, 66
Lageos (satellite), 262 level, bull’s eye, 216
land uplift levelling, 69, 122, 256
absolute, 238 principle, 119
Fennoscandian, 247 geostrophic, 241
post-glacial, 237, 238, 288, 289 new technologies, 288
mechanism, 240 relativistic, 136
relative, 238 steric, 240
Laplace equation, 31, 231 levelling instrument, 119
definition, 31 levelling line, 135, 256
definition, 13 levelling staff, 119
basis solutions, 40 levelling, of gravimeter, 216
in ellipsoidal co-ordinates, 56 lever beam, 216, 219
in polar co-ordinates, 36 lever motion, 244
in rectangular co-ordinates, 32 linear partial differential equations,
in spherical co-ordinates, 40, 327 theory of, 31
linearity, 32, 33 linear regression, 253
solution, 31 Liouville, Joseph, 312
transformation, 31 local terrain effect, 229
Laplace operator (∆), 299 localized kernel, 146, 152
definition, 12, 31 location vector, 53
linearity, 182 of a mass element, 54
Laplace, Pierre-Simon de, 12, 299 Lohja (Finland), 282
Lapland (Northern Europe), 96 long water-tube clinometer, 282
laser interferometer, 220 longitude
latitude astronomical, 69
astronomical, 69, 91 geocentric, 38
geocentric, 38, 40, 71 of the Moon, 279
geodetic, 72, 78, 168 lookup table, 318
geographic, 72
Óîá.
350 Index
Love number, 280–283, 285 Min, Erik de, 173

determination, 282 minimum energy state, 121
GNSS, 282 mirror antisymmetry, 42
Love, Augustus, 280 mirror symmetry, 42, 47
lunar laser ranging (LLR), 220 missile, submarine-launched ballistic, 252
mixed covariance, 202
M Mohorovičić, Andrija, 116
magnetic field, 224 Mohorovičić discontinuity (“Moho”), 116,
of Earth, 225 179
Maldives (Indian Ocean), 244 Molodensky theory, 26, 90
map projection co-ordinates, 163, 164 Molodensky, Mikhail Sergeevich, 26, 90,
map projection plane, 154 95, 124, 289
mareograph, see tide gauge, 238 Molodensky’s method, 148
Mars (planet), gravity field, 203 height anomaly, 153
mass density, 13 linearization, 151
mass density layer, 1, 149 linearization error, 153
double, 15, 16, 25, 324 Molodensky’s proof, 124, 127
single, 14, 25, 105, 114, 141 Molodensky’s realization, 126
mass distribution inside the Earth, 27 momentum
mass point angular, 292
underground, 211 conservation, 293, 294
mass point, underground, 172 per unit of mass, 293
mass surface density, 14 linear, 292
matter density, 9, 302 Mongolia, 176
Mauna Kea, 118 monopole, 54
Maxwell, James Clerk, 13 monsoon, 271
mean geoid, 283 Moritz, Helmut, 289
mean height, of satellite orbit, 261, 263 mosaic, 44
mean sea level, 86, 122, 236 mu-metal, 225
definition, 235, 236 Munk, Walter, 252
global, 236, 238, 253 Mäkinen, Jaakko, 123
global location, 89
mean sea surface, 262 N
measurement axis, 219 N60 (height system), 120, 175, 236, 237
measuring telescope, 119 N2000 (height system), 120, 176, 237
mechanics, of the solid Earth, 283 nabla (∇, operator), definition, 8, 295
Meissner effect, 225 National Geospatial-Intelligence Agency
Mekong (river), 271 (NGA), 287
Melchior, Paul, 280 National Land Survey of Finland, 225, 289
meridian convergence, 167 Navy, United States, 252
meridian ellipse, 72 network hierarchy, 224
mesh size, of grid, 160 Neumann, Carl Gottfried, 52
mesoscale eddy, 235, 243 Newton, Isaac, 3, 10
metal fatigue, 218 Newton’s law of gravitation, 1, 177
metallurgy, 218 definition, 3
Metsähovi research station (Finland), 120, spherical Earth, 230
224, 225, 237 Newton’s law of motion, 3
microgal (µGal), 87 Newton’s theory of gravitation, 1, 13
microseismicity, effect on gravimeter, 218, NGA (National Geospatial-Intelligence
221 Agency, U.S.), 54
microwave link (GRACE), 271 Niethammer, Theodor, 131
mid-Holocene highstand, 244 Niethammer’s method, 131
Milano (Italy), 173, 287 NIMA (National Imagery and Mapping
milligal (mGal), 87 Agency, U.S.), 54
Óîá .
Index 351
NKG WG for Geoid and Height Systems, in a co-rotating system, 74

288 in spherical harmonics, 76
NKG2004 (geoid model), 288, 289 on the reference ellipsoid, 64, 71, 72,
NKG2015 (geoid model), 288 125
NN (height system), 120 over the equator, 74, 75
noise normal vector, of a surface, 300, 302
definition, 187 North American Vertical Datum 1988
in altimetry observations, 255 (NAVD88), 120
observational, 188 Norwegian Sea, 110
noise variance matrix, 188, 195 Nottingham (Great Britain), 22
Nordiska Kommissionen för Geodesi Nouvel, Henri SJ, 120
(NKG), 288
norm, of a vector, 292, 305 O
Normaal Amsterdams Peil (NAP), 120 obliquity, of Earth rotation axis, 286
normal component, of a vector field, 300 observation equation, 195, 256
normal correction ( NC ), 135 adjustment constraint, 257
equation, 135 of ballistic gravimetry, 221, 222
staff interval, 135 of crossover adjustment, 255, 258,
normal derivative, 14 274
existence, 24 of satellite altimetry, 255
of potential, 15 ocean current, 242, 243
on surface of sphere, 20 GOCE mission, 243
normal direction, 15 inversion problem, 273
existence, 24 measurements, 241
normal equation, 222 permanent, 235
normal field tilt, 242
choice, 132 unit, 242
high degree, 190 variation, 243
normal gravitational potential, 64 oceanography, 283
normal gravity, 190 octave (programming language), 47, 265
definition, 64 Ohio (USA), 107, 190
at sea level, 92 Ohio State University, 54, 55, 289
calculation, 85 one-Earth problem, 184
GRS80, 73 operator
linearity along the plumb line, 124, linear, 27, 308, 309
126, 132 self-adjoint, 309, 310, 312
on the equator, 71 spectral representation, 200
on the poles, 71 optical lattice clock, 136, 137
on the reference ellipsoid, 71, 72 optic-fibre cable, 136
normal gravity field, 63 optimality, least-squares, 189
and reference ellipsoid, 82 orbit
ellipsoidal, 72 35-day, 266
GRS80, 227 no-shadow, 263, 264
normal gravity potential, 64, 70 retrograde, 264, 265
normal height, 123, 126, 128 satellite, 261
calculation, 135 Sun-stationary, 263, 264, 272
precise calculation, 132 Sun-synchronous, 263, 274
normal matrix, 256 three-day, 266
normal potential, 64 orbit error, 255, 258
definition, 63 bias, 260, 261
global average, 89 correction, 253, 269
gradient, 63 of satellite, 255
GRS80, 73 trend, 259–261
orbit prediction, 259
Óîá .
352 Index
orbit-error correction, 255 planet as a mass point, 10

order, harmonic, 41 plasma, 136
orientation, of a surface, 302 plasticity, 218
orthogonal basis, 305, 314 Plesetsk Cosmodrome (Russia), 269, 271
complete, 310 plumb line, 67
Legendre polynomials, 313 definition, 64
orthogonality bending, 107
of degree constituents, 198, 201 plumb line deflection, 142
of functions, 313 plumb-line deflection, 81, 82, 106
of surface spherical harmonics, 155 definition, 64
orthometric correction (OC ), 134, 135 and the geoid, 82
equation, 135 as a functional, 182
staff interval, 135 at sea, 252
orthometric height, 121, 128 in Finland, 81
definition, 69, 122 local contribution, 159
calculation, 122, 134 observed, 83
precise calculation, 130, 132 definition, 81
orthonormal basis, 8, 53, 64, 66, 305, 306, plumb-line direction, 81, 84, 228
310 absolute, 282
definition, 4, 294 measurement, 68
orthonormality, of surface harmonics, 199 point mass, underground, 26
oscillation equation, 215, 218, 311, 312 point variance, 311
as an eigenvalue problem, 311 Poisson equation, 20, 132, 167
Ostrobothnia (Finland), 237 definition, 13
OSU model, 55 data point, 144
evaluation point, 144
P for computing a harmonic field from
parallellogram, 292 surface values, 144
Peltier effect, 244 for r ∆ g, 145
Peltier, W. Richard, 245 gravity potential, 66
pendulum clock, 213 spectral form, 145
pendulum equation, 213 Poisson kernel for gravity anomalies, 146,
peneplain, 110 147
period Poisson, Siméon Denis, 13
fortnightly tide, 279 polar holes, 264, 265
orbital, 261, 262 polar motion, Earth, 282
oscillation, 215, 218 polar radius, 72
theoretical tide, 280, 282 polynomial fit, of geoid surface, 237
Peru (South America), 96 potential
petroleum extraction industry, 175 definition, 4
phase extended body, 5
of water, 244 exterior, 16, 27, 32, 89, 95, 114
phase angle of the topography, 320
of atom, 223 interior, 16
of matter waves, 223 of the topography, 319
tidal, 264 mass line, 11
physical geodesy of a conservative field, 298
geometry and physics, 38 point mass, 8, 12, 27, 53
potential convention, 8 pointlike body, 4
textbook, 289 set of mass points, 9
physical modelling, 240 solid body, 9
physical theory, nature, 1 spherical shell, 6, 7
phytoplankton, 241 terrain point, 120
Pizzetti, Paolo, 72
Óîá .
Index 353
topography, 12 quasi-geoid, 91, 123, 127, 128

uniqueness, 299 quasi-geoid height, 124, 128
potential difference, 14, 69, 120 quasi-geoid model, 91, 113
potential energy, 8 computation, 174
potential field, 1, 31
local behaviour, 32 R
of a dipole, 53 radar, microwave, 251, 269
of a mass density layer, 1 radio energy, 253
vertical shift, 36 radius of curvature
potential theory, 1 principal, 246
Potsdam system, 214, 224 transversal, 39
PRARE (instrument), 252 Raman effect, 223
Pratt, John Henry, 106 Rapp, Richard H., 55, 203
Pratt–Hayford hypothesis, 106 recursion
Pratt-Hayford Pratt–Hayford hypothesis, calculation of orthometric height, 131
107 computation of Fourier basis
precession, orbital, 263 functions, 42
precise levelling, 120, 237, 240 computation of Legendre polynomials,
prediction, 187 47
homogeneous, 195, 196 definition of normal height, 126
Prey reduction, 131 recursive algorithm, 41
Prey, Adalbert, 131 reduction to sea level, 148
principal axes, of error ellipse, 311 redundancy, 260
Principia (book), 3 reference benchmark, 121
prism integration, of terrain correction, N60, 237
100, 101 N2000, 237
prism method, of terrain correction, 177 reference ellipsoid, 39, 86
propagation delay as a level surface, 63, 72
ionospheric, 254 reference latitude, 168, 170
tropospheric reference radius, 77
dry, 254 reference-surface thinking, 91
wet, 254 regularization, 149, 179
propagation of covariances, 201, 202 relativity theory, 1
propagation of variances, 188, 196 relaxation time, 247
propellant, GOCE, 229 remote sensing of sea ice, 268
propeller aircraft, 228 “Remove” step, 150, 165
proper time, 136 Remove-Restore method, 96, 153, 154, 325
pseudo-force, 228 renormalization, 179
aircraft motions, 228 repeat period, of a satellite orbit, 262
Earth rotation, 64 repeat track, 268
moving on a rotating Earth, 65 repulsion, 4
Päijänne, Lake (Finland), 123 research school, international, 175, 287
Pythagoras theorem, 4 residual, 222
residual terrain modelling (RTM), 150,
Q 151
quadrature, 158 indirect effect, 151
quadrupole, 54 resolution, of measurement, 181
quadrupole moment resonance, 264
of the Earth, 54 rest length, of spring, 214, 216
quantization, 311 “Restore” step, 151, 165, 325
quantum mechanics, 31 retracking, 267
quantum number, 312 return pulse
quantum state, 136, 223 analysis, 267
quantum theory, 223 time measurement, 267
Óîá .
354 Index
travel time, 267 of vector fields, 291

rheology, 246 of vectors, 292
Richer, Jean, 213 on the sphere, 313
rising dough model, 239 Schrödinger equation, 32
Robin, Victor Gustave, 88 Schrödinger, Erwin, 32
rocket engine Schrödinger’s cat, 32
ionic, 271 Schwarz, Klaus Peter, 173
Romberg, Arnold, 217 Schwarzschild metric, 136
Romberg, Werner, 159 sea ice, 267
root, of mountain, 106, 112 sea-floor pressure, 270
density, 106, 108 sea-level equation, 245
depth, 108, 109 convolution, 245, 246
rotational potential, 64 data point, 246
rotational symmetry, 170, 204, 205, 328 evaluation point, 246
Royal Society of Edinburgh, 13 integration point, 246
Royal Society of London, 3, 13, 90 sea-level oscillation, amplitude, 241
sea-level rise, 238, 252
S global, 253
Sacks, Oliver, 3 Holocene, 247
salinity (sea water), 236, 241 Seasat (satellite), 252
salinity gradient, 241 sea-surface topography, 236, 242, 253, 255,
sampling, spatial, 181 269
San Francisco (USA), 288 definition, 236
Sandwell, David, 176 change, 238
Sansò, Fernando, 173 determination, 239
SARAL/AltiKa (satellite), 252 global, 241
saros (lunar motion periodicity), 236 map, 243
satellite altimetry, 239, 251 mapping, 252
concepts, 254 sea-surface variability, 255, 262, 268
geoid, 268 secular effect, in sea level, 236
location of sea surface, 273 seismicity, 247
measured range, 253 seismology, 27, 106
measurement method, 253 seismometer, 226
orbital geometry, 260 self-adjoint differential equation, 311, 312
results, 267 semi-major axis, 78, 261, 262
sea-surface variability, 268 Earth ellipsoid, 40, 72
satellite geodesy, 203 semi-minor axis, 78
satellite laser ranging (SLR), 262 semi-minor axis, of the Earth ellipsoid, 40,
satellite orbit 72
Sun-stationary, 263 sensitivity, instrument, 215, 216
satellite orbit perturbations, 55, 181 Sentinel 3A (satellite), 252
satellite-to-satellite tracking (SST), 271 separation of variables, 33, 36, 40
Saturn (planet), ring, 213 shoebox world, 34, 177
scalar field, definition, 294 Sideris, Michael, 173
scalar product, 157 sight axis, of a level, 119
definition, 291 signal covariance matrix, 187
commutativity, 305 signal variance, 190
in a function space, 311 signal variance matrix, 187
in a vector space, 4, 305, 309 of gravity anomalies, 192
linearity, 305 signal, definition, 187
of force and path, 291 significant wave height (SWH), 253, 254
of Legendre polynomials, 47, 313 Simpson integration, 159
of operator and vector, 296 Simpson, Thomas, 158
of two functions, 306
Óîá .
Index 355
Simpson’s rule, 158 ellipsoidal gravitational field, 74

singularity, of normal matrix, 256, 259 first terms, 76
sink (vector field), 17, 297 global, 54
sink, of vector field, 296 high degree, 59
Skylab (space station), 251 model, 55, 155
slowing-down ratio, of time, 136 normal gravitational potential, 74
snow clearing, 225 of the topography, 45
Sodankylä (Finland), 224 resolution, 45
solar panel, 263 rotational symmetry, 47
Solar time, 264 spheroid
Solheim, Dag, 174 Bruns, 76
solid body, 9 Helmert, 76
solid spherical harmonic, 40 spirit level, 119
Somigliana, Carlo, 72 Spitsbergen (island), 110
Somigliana–Pizzetti equation, 72 spline, 173
source spring balance, linear, 214
of a field, 8 spring constant, 214
source (vector field), 17, 297 spring lengthening, 215
source function, 17 stability (gravimeter), 225
source, of vector field, 296 stabilized platform, 219, 228
space domain, 35, 166 staff reading difference, 134
space geodesy, 55 stationarity, of a stochastic process, 186
spatial frequency, 166 stationary field, 4
spectral coefficients, 41 steel, 218
spectral constituent function, 49 stepwise network densification, 224
sphere stereographic map projection, 163
celestial, 314 Sterneck, Robert von, 213
spherical Bouguer reduction, 104 stochastic process
mass effect, 104 definition, 183
spherical cap, 154 ergodic, 184
spherical harmonic of location, 183
algebraic sign, 45 on the Earth’s surface, 184
sectorial, 44, 45 stochastic quantity, 183
semi-wavelength, 44, 46, 59 realization, 183
tesseral, 44, 45 Stokes curl theorem, 300, 301
wavelength, 44 Stokes equation, 90, 140, 154, 163, 239
zonal, 43, 45 at sea level, 152
spherical shell, 5, 7 differentiation, 142
potential of, 5 evaluation point, 140, 163
spherical symmetry, 247, 282 exterior space, 141
of mass distribution, 27 geoid height, 139
of the Earth, 184, 186 in collocation, 195
spherical-cap in plane co-ordinates, 164
integration, 112 inner zone, 159
spherical-harmonic coefficient integration, 140
fully normalized, 314 observation point, 140
spherical-harmonic coefficients, 199, 263 spectral form, 139
as linear functionals, 183 Stokes equation and harmonicity, 149
fully normalized, 55 Stokes kernel, 90, 139, 141, 169
GRS80, 75 closed expression, 141
spherical-harmonic expansion, 2, 46, 165, modified, 155, 156
190 on the Earth’s surface, 141
coefficients, 41 re-written, 168
Óîá .
356 Index
smoothness, 158 in the exterior space, 179

spectral form, 141 spherical geometry, 104
Taylor series expansion, 169 values, 101
Stokes, George Gabriel, 26, 90 thermal expansion
Strang van Hees, Govert, 168, 173 of pendulum, 213
Sturm, Jacques, 312 of sea water, 238
Sturm-Liouville problem, 312 thermostat (gravimeter), 219
submarine measurements, 214 thruster, on-board, 262, 266
subsidence, land, 244 tidal decomposition equation, of Laplace,
superconduction, 224 278
superconductivity, 225 tidal field, 231
superspring, 221 tidal force, 280, 283
support latitude, 169 tidal loading
surface element, oriented, 300 atmospheric, 285
surface normal, Earth, 88 ocean, 285
surface normal, ellipsoidal, 81, 84 tidal loading, ocean, 285
surface spherical harmonic, 40, 43 tidal potential, 277, 282
as map, 45 tidal reduction, permanent deformation,
fully normalized, 314 283
plotting, 47 tide, 10, 235, 253, 255
surface spherical harmonics, linear amplitude, 282
combination of, 155 diurnal, 264, 279, 280
surface spherical-harmonic fortnightly, 279
plotting, 48 frequency, 282
sverdrup (unit), 242 ocean, 254, 285
sweet water, 235 period, 282
swinging time, 213 permanent part, 283, 286
symbolic algebra software, 147 deformation, 283
symmetric matrix, eigenvectors, 310 effect on height difference, 284
Synthetic Aperture Radar Altimeter geoid effect, 284
(SRAL), 252 phase angle, 282
semidiurnal, 279
T semi-diurnal, 264, 280
tangent plane, 163, 164 solid-Earth, 254, 289
to a level surface, 67 theoretical, 277, 280, 281, 283
to the Earth, 241 tide gauge, 236, 238
tapering, 173 tide-free geoid, 283, 285
tapering function, 173 Tikhonov regularization, 259, 261
25%, 173 Tikhonov, Andrey Nikolayevich, 259
Taylor series expansion, 11, 12, 16 Toeplitz circulant matrix, 206
Helmert condensation, 320, 321 Toeplitz, Otto, 206
tea break, 225 Tom (GRACE satellite), 270
telluroid, 90, 127 tomography, seismic, 112
definition, 90 TOPEX/Poseidon (satellite), 236, 252, 253,
telluroid mapping, 90, 127 264
temperature (sea water), 236 topographic-potential integral
terrain correction (TC ), 99, 100, 102, 131, evaluation point, 319
176, 177, 323 observation point, 319
bias, 100 topography, 98, 103
calculation, 100, 101 topography shift to inside geoid, 149
data point, 177 Torge, Wolfgang, 289
equation, 101 torque, 216
evaluation point, 177, 178 torsion balance, 3, 230
example, 102
Óîá .
Index 357
total mass n-dimensional, 308

body, 10 vector sum, 66
of a column of air and water, 270 vectorial product, 292, 293
of a column of matter, 108 of operator and vector, 297
of the Earth, 55, 73, 77, 89 Vening Meinesz, Felix Andries, 112, 142
Toulouse (France), 287 submarine measurements, 214
trace, of a matrix, 311 Vening-Meinesz equations, 142, 159
transformation surface, geoidal, 237, 289 Verbaandert–Melchior pendulum, 282
translation (parallel shift), 259 vertical displacement, of test mass, 218
trench, ocean, 214 vertical gravity gradient, 73, 92, 222
triangle inequality, 10 anomalous, 148, 151, 152, 167
Trieste (Italy), 120 at sea level, 148
true anomaly, 262 kernel, 147
Tscherning, Carl Christian, 174, 203 Earth surface, 88
Tscherning-Rapp formula, 203 free air, 230
Tukey taper, 173 free-air, 131
turbulence, in a vector field, 297 inside rock, 131
Turku harbour, 259 vertical normal gravity gradient, 91, 102,
Turku, Finland, 259 128
Tuvalu (Pacific Ocean), 244 vertical reference system, see height
Tytyri limestone mine (Lohja, Finland), system, 259
282 viscosity, 246
viscous relaxation mode, 247
U Von Sterneck device, 213
uncertainty
a priori, 259 W
inside the Earth, 181 water vapour radiometer, 254
of co-ordinate measurement, 228 water, flowing upward, 123
of estimate, 186, 194 wave equation
of observation, 181, 192, 195 of matter, 32
of vertical acceleration, 228 relativistic, for the electron, 20
uniform convergence, 308, 319, 320 wave number, 178
unit matrix, 205 weighing visitors, 225
unit sphere, 314 weightlessness, 229
University of Texas, 270 Wenzel, Hans-Georg, 285
unknown (adjustment parameter), 222, wind field, 297
260 wind pile-up, 235
upper culmination, of the Moon, 278 wire pendulum, very long, 214
wire-frame model, 259
V work integral, 292, 299, 301
Vaasa (Finland), 224 world aether, 1
variance, 183 World Geodetic System 1984 (WGS84), 54,
variance function, of a stochastic process, 73
183
variance matrix, of location, 310 Z
variance of prediction, 192, 194, 196, 209 zenith angle, of the Moon, 277
definition, 188 zenith tube, 282
minimization, 189 zero geoid, 283
vector, 291 zero potential
vector field at infinity, 14
definition, 294 at mean sea surface, 14
differentiable, 17 zero topography compensation level, 109
vector space, 306 zero-length spring, 216–219
abstract, 305, 313 how to build, 217
Óîá .

Physical Geodesy

Uploaded by

Copyright:

Available Formats

Physical Geodesy

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Physical Geodesy

Uploaded by

Copyright:

Available Formats

Physical Geodesy

September 27, 2018

Helsinki, September 27, 2018,

»  1. Fundamentals of the theory of gravitation . . . . . . . . . . . . 1

List of Tables xii

List of Figures xiii

1. Fundamentals of the theory of gravitation 1

2. The Laplace equation and its solutions 31

2.8 Symmetry properties of spherical harmonics . . . . . . . . 46

3. The normal gravity field 63

4. Anomalous quantities of the gravity field 81

4.7 Free-air anomalies . . . . . . . . . . . . . . . . . . . . . . . 91

6. Vertical reference systems 119

7. The Stokes equation and other integral equations 139

8. Spectral techniques, FFT 163

9. Statistical methods 181

9.10 Collocation and the spectral viewpoint . . . . . . . . . . . 204

10. Gravimetric measurement devices 213

11. The geoid, mean sea level, sea-surface topography 235

12. Satellite altimetry and satellite gravity missions 251

13. Tides, atmosphere and Earth crustal movements 277

14. Earth gravity field research 287

A. Field theory and vector calculus — core knowledge 291

B. Function spaces 305

C. Why does FFT work? 317

D. Helmert condensation 319

E. The Laplace equation in spherical co-ordinates 327

2.1 Vertically shifting the potential field . . . . . . . . . . . . . . 36

2.7 Legendre functions of the second kind . . . . . . . . . . . . . 59

3.1 GRS80 normal potential spherical-harmonic coefficients . . 75

4.1 Orders of magnitude of gravity variations . . . . . . . . . . . 88

12.1 Altimetric satellites through the ages . . . . . . . . . . . . . 251

13.1 The various periods in the theoretical tide . . . . . . . . . . 280

1.1 Gravitation is universal . . . . . . . . . . . . . . . . . . . . . . 2

2.1 Attenuation of the gravitational field with height . . . . . . 35

3.1 The normal gravity field of the Earth . . . . . . . . . . . . . . 63

3.3 The curvature of level surfaces . . . . . . . . . . . . . . . . . 67

4.1 Geoid undulations and deflections of the plumb line . . . . . 82

5.1 The attraction of a Bouguer plate . . . . . . . . . . . . . . . . 97

6.1 The principle of levelling . . . . . . . . . . . . . . . . . . . . . 119

6.7 Geoid, quasi-geoid, telluroid and topography . . . . . . . . . 127

7.1 The principle of gravimetric geoid determination . . . . . . 140

8.1 Map projection co-ordinates in the local tangent plane . . . 164

9.1 Geocentric angular distance and azimuth . . . . . . . . . . . 185

10.1 Jean Richer’s report . . . . . . . . . . . . . . . . . . . . . . . . 214

11.1 The mechanism of post-glacial land uplift . . . . . . . . . . . 240

11.2 The Fennoscandian gravity line on the 63rd parallel . . . . 240

12.1 Results from the TOPEX/Poseidon and Jason satellites . . . 253

13.1 Theoretical tide . . . . . . . . . . . . . . . . . . . . . . . . . . . 277

A.1 Exterior or vectorial product . . . . . . . . . . . . . . . . . . . 293

B.1 Fourier analysis on a step function . . . . . . . . . . . . . . . 307

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

AGU American Geophysical Union

SLR satellite laser ranging

Figure 1.1. Gravitation is universal. A gravitational lense imaged by the Hub-

D 1.2 Gravitation between two masses

where a is the acceleration of body m. From this follows

where the triad of unit vectors {i, j, k} is an orthonormal basis7 in Eu-

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

D 1.3 The potential of a point mass

» 1. Fundamentals of the theory of gravitation . . . . . . . . . . . . 1