Documentos de Académico
Documentos de Profesional
Documentos de Cultura
3230
Figure: Cluster of galaxies Abell 2218, distance 2 billion light years, acts as a gravitational lense, The geometry of space-time within the cluster is non-Euclidean. Image credit: NASA/ESA
Course Description
Workload 3 cr Teaching Period III Learning Outcomes After completing the course, the student Is able to do simple computations on the sphere and understands the geometry of the reference ellipsoid Is able to solve, using tools like MatlabTM , the geodetic forward and inverse problems etc. on the reference ellipsoid Masters the basics of global and local reference systems and is able to execute transformations Masters the basics of Gaussian and Riemannian surface theories and can derive the metric tensor, Christoel symbols and curvature tensor for simple surfaces Masters the basic math of map projections, esp. conformal ones, and the behaviour of map scale, and is able to compute the isometric latitude. Content Spherical trigonometry, geodetic co-ordinate computations in ellipsoidal and rectangular spatial co-ordinate systems, astronomical co-ordinates, co-ordinate system transformations, satellite orbits and computations, Gaussian and Riemannian surface theories, map projection computations. Foreknowledge Equivalences Replaces course Maa-6.230. Target Group Completion Completion in full consists of the exam, the exercise works and the calculation exercises. Workload by Component Lectures 16 2 h = 32 h Independent study 55 h Exercise works 2 12 h = 24 h (independent work) Calculation exercises at home 16, of which must be done 12 6 h = 72 h (independent work) Exam 3 h Total 186 h Grading The grade of the exam becomes the grade of the course, 1-5 Study Materials Lecture notes. Background material Hirvonen: Matemaattinen geodesia; Torge: Geodesy. Teaching Language Suomi Course Sta and Contact Info Martin Vermeer, room Gentti 4.143, name@aalto. Reception times Wednesdays at 12:15 CEFR-taso Listietoja a
Contents
1. Spherical trigonometry 1.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2. Spherical excess . . . . . . . . . . . . . . . . . . . . . . . . . 1.3. The surface area of a triangle on a sphere . . . . . . . . . . 1.4. A rectangular spherical triangle . . . . . . . . . . . . . . . . 1.5. A general spherical triangle . . . . . . . . . . . . . . . . . . 1.6. Deriving the formulas with the aid of vectors in space . . . 1.7. Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.8. Solving the spherical triangle by the method of additaments 1.9. Solving the spherical triangle by Legendres method . . . 1.10. The forward geodetic problem on the sphere . . . . . . . . . 1.11. The geodetic inverse problem on the sphere . . . . . . . . . 1.12. The half-angle cosine rule . . . . . . . . . . . . . . . . . . . 2. The 2.1. 2.2. 2.3. 2.4. 2.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 1 2 3 4 5 6 7 8 8 9 9 11 11 11 12 13 13 15 15 15 16 16 17 17 18 19 21 21 22 23 24 25 29 29 30 30 32 33 34 34
geometry of the reference ellipsoid Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . The geodesic as solution to a system of dierential equations An invariant . . . . . . . . . . . . . . . . . . . . . . . . . . . . The geodetic main problem . . . . . . . . . . . . . . . . . . . The geodetic inverse problem . . . . . . . . . . . . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
3. Co-ordinates on the reference ellipsoid 3.1. Representations of the sphere and the ellipsoid . . . . . . . 3.2. Various latitude types . . . . . . . . . . . . . . . . . . . . . 3.3. Measures for the attening . . . . . . . . . . . . . . . . . . 3.4. Relationships between dierent types of latitude . . . . . . 3.5. Co-ordinates in the meridional ellipse . . . . . . . . . . . . . 3.6. Three-dimensional rectangular co-ordinates on the reference 3.7. Computing geographic co-ordinates from rectangular ones . 3.8. Meridian arc length . . . . . . . . . . . . . . . . . . . . . . 4. Reference systems 4.1. The GRS80 system and geometric parameters 4.2. Gravimetric parameters . . . . . . . . . . . . 4.3. Reference frames . . . . . . . . . . . . . . . . 4.4. The orientation of the Earth . . . . . . . . . . 4.5. Transformations between systems . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . ellipsoid . . . . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
5. Using rotation matrices 5.1. General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2. Chaining matrices in three dimensions . . . . . . . . . . . . . . 5.3. Orthogonal matrices . . . . . . . . . . . . . . . . . . . . . . . . 5.4. Topocentric systems . . . . . . . . . . . . . . . . . . . . . . . . 5.5. From geocentric to topocentric and back . . . . . . . . . . . . . 5.6. The geodetic main and inverse problems with rotation matrices 5.6.1. Geodetic main problem . . . . . . . . . . . . . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
iii
Contents 5.6.2. Geodetic inverse problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6.3. Comparison with ellipsoidal surface solution . . . . . . . . . . . . . . . . . 6. Co-ordinate systems and transformations 6.1. Geocentric terrestrial systems . . . . . 6.2. Conventional Terrestrial System . . . . 6.3. Polar motion . . . . . . . . . . . . . . 6.4. The Instantaneous Terrestial System . 6.5. The quasi-inertial geocentric system . 35 35 37 37 37 37 38 38 41 41 41 42 42 42 43 44 47 47 49 50 52 52 53 54 54 57 57 58 60 61 65 67 71 71 75 77 78 78 78 78 80 81 84 86 89 89 89 90 90
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
7. Reference systems and realizations 7.1. Old and new reference systems; ED50 vs. WGS84/GRS80 7.2. WGS84 and ITRS . . . . . . . . . . . . . . . . . . . . . . 7.3. Co-ordinate system realizations . . . . . . . . . . . . . . . 7.4. Realization of WGS84 . . . . . . . . . . . . . . . . . . . . 7.5. Realizations of ITRS/ETRS systems . . . . . . . . . . . . 7.6. The three-dimensional Helmert transformation . . . . . . 7.7. Transformations between ITRF realizations . . . . . . . . 8. Celestial co-ordinate systems 8.1. Sidereal time . . . . . . . . . . . . . . . . . . . . 8.2. Trigonometry on the celestial sphere . . . . . . . 8.3. Using rotation matrices . . . . . . . . . . . . . . 8.4. On satellite orbits . . . . . . . . . . . . . . . . . 8.5. Crossing a given latitude in the inertial system . 8.6. The satellites topocentric co-ordinates . . . . . . 8.7. Crossing a given latitude in the terrestrial system 8.8. Determining the orbit from observations . . . . . 9. The 9.1. 9.2. 9.3. 9.4. 9.5. 9.6. surface theory of Gauss A curve in space . . . . . . . . . . . The rst fundamental form (metric) The second fundamental form . . . . Principal curvatures . . . . . . . . . A curve embedded in a surface . . . The geodesic . . . . . . . . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
10.The surface theory of Riemann 10.1. What is a tensor? . . . . . . . . . . . . . . . . . . 10.2. The metric tensor . . . . . . . . . . . . . . . . . . 10.3. The inverse metric tensor . . . . . . . . . . . . . 10.3.1. Raising or lowering sub- or superscripts of 10.3.2. The eigenvalues and -vectors of a tensor . 10.3.3. The graphic representation of a tensor . . 10.4. The Christoffel symbols . . . . . . . . . . . . 10.5. The geodesic revisited . . . . . . . . . . . . . . . 10.6. The curvature tensor . . . . . . . . . . . . . . . . 10.7. Gauss curvature and spherical excess . . . . . . 10.8. The curvature in quasi-Euclidean geometry . . . 11.Map projections 11.1. Map projections and scale . 11.1.1. On the Earth surface 11.1.2. In the map plane . . 11.1.3. The scale . . . . . .
. . . a . . . . . . .
. . . . . . . . . . . . tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
iv
Contents 11.1.4. The Tissot-indicatrix . . . 11.2. The Lambert projection (LCC) . 11.3. On the isometric latitude . . . . . 11.4. The Mercator projection . . . . 11.5. The stereographic projection . . . 11.6. The Gauss-Kr ger projection . . u 11.7. Curvature of the Earth surface and . . . . . . . . . . . . . . . . . . scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 . 93 . 95 . 96 . 97 . 99 . 102 105 105 105 106 106 109 111 113 115
12.Map projections in Finland 12.1. Traditional map projections . . . . . . . . . . . . . . . 12.2. Modern map projections . . . . . . . . . . . . . . . . . 12.3. The triangulated ane transformation used in Finland 12.3.1. Plane co-ordinates . . . . . . . . . . . . . . . . A. Isometric latitude on the ellipsoid B. Useful equations connecting the main radii of curvature C. The Christoel symbols from the metric D. The Riemann tensor from the Christoel symbols
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
A1 A2
A3 A4
Figure 1.1.: Spherical triangles on a semi-sphere. The back side surface of the sphere has been depicted in a lighter shade with its antipode triangles
Chapter 1. Spherical trigonometry i.e., A1 = + + = , where is called the spherical excess. If the radius of the sphere is not 1 but R, we obtain A1 = R2 = A1 . R2
where unit is the conversion factor of the unit considered, e.g., for degrees, 57.29577951308232087721 or for gons, 63.66197723675813430801. As we see is the spherical excess inversely proportional to R2 , i.e., directly proportional to the total curvature R2 . It is also directly proportional to the surface area of the triangle. This is a special case of a more general rule: The directional closing error of a vector which is transported in a parallel way around the closed edge of a surface is the same as the integral over the surface of the total curvature. As a formula: =
A
Kd,
where is the variable for surface integration, and K is the total curvature of the surface according to K.F. Gauss, which thus can vary from place to place. E.g., on the surface of an ellipsoid 1 K= , MN where M is the meridional curvature (in the North-South direction) and N the so-called transverse curvature in the East-West direction. Both depend on the latitude . In a smallish area, the internal geometry of the ellipsoidal surface does not dier noticeably from a spherical surface, the radius of which is R = M N . If the triangle of the surface of the sphere is small compared to the radius of the Earth, also the spherical excess will be small. In the limit we have 0 ja + + = exactly. We say that the a plane surface (or a very small part of a spherical surface) forms a Euclidean space, whereas a spherical surface is non-euclidean.
A=
A A D
cos b sin a cos b
1 b O b
sin b
cos b
a
cos a cos b
b O c a E .
B
. D .
a C O c
A
1
A
sin c sin c sin
sin c
cos c
sin c cos
Figure 1.2.: A rectangular spherical triangle. The partial triangles are lifted out for visibility
At least for computing the spherical excess, these approximate formulas are good enough: = A , R2
where also the approximate value for R, e.g., R a = 6378137 m, is completely sucient.
according to the last equation in group (1.1). By dividing the rst of group (1.2) by its second, we obtain cot = cot a sin b and the second of group (1.1) by its third, correspondingly cot = cot b sin a.
c h
b1 b2
xC
xA
b c
c a O
xB
B
y
x Figure 1.4.: The spherical triangle ABC, for deriving the cosine and sine rules using vectors in space
For comparison the corresponding formulas for a plane triangle: c2 = a2 + b2 2ab cos and a b c = = . sin sin sin
At least in the case of the sine rule it is clear, that in the limit for a small triangle sin a a etc., in other words, the spherical sine rule morphs into the one for a plane triangle. For the cosine rule this is not immediately clear.
sin A cos B sin B = sin A cos B cos B cos A sin B = cos A cos B sin B cos a sin b sin = cos a sin b cos sin a cos b . sin b sin a sin
c 2
a
2 2 b
a 2
When the third vector is 0 xC = 0 , 1 we obtain the volume of the parallelepiped spanned by the three vectors (i.e., twice the volume of the tetrahedron ABCO) as follows: Vol {xA , xB , xC } = (xA xB ) xC = sin b sin a sin . The volume contained by the three vectors does not however depend on the order of the vectors, so also Vol {xA , xB , xC } = sin b sin c sin = sin a sin c sin . Division yields sin a sin = sin c sin , sin b sin = sin c sin , i.e., sin b sin c sin a = = , sin sin sin the sine rule for a spherical triangle.
1.7. Polarization
For every corner of the triangle we may dene an equator or great circle, one pole of which is that corner point. If we do this, we obtain three equators, which themselves also form a triangle. This procedure is called polarization. Because the angular distance between the two corner points on the surface of the sphere is the length of the side, the angle between the planes of two such great circles must be the same as this length. And the polarization triangles angle is 180 minus this.
1.8. Solving the spherical triangle by the method of additaments The polarization method is symmetric: the original triangle is also the polarization of the polarization triangle. The intersection point of two edges of the polarization triangle is at a distance of 90 from both poles, i.e., the corners of the original triangle, and the edge between them thus is the equator of the intersection point. For symmetry reasons also the length of an edge of the polarization triangle equals 180 minus the corresponding angle of the original triangle. For an arbitrary angle we have: sin (180 ) = sin cos (180 ) = cos Because of this we obtain of the spherical trigonometry cosine rule (1.3) the following polarized version: cos = ( cos ) ( cos ) + sin sin ( cos c) or cos = cos cos + sin sin cos c, a formula with which one may calculate an angle if the two other angles and the side between them are given.
12
A12
1
2 12 1
.
1
or, in base-ten logarithms
10
ekvaattori
where M =10 log e = 0.43429448. In the age of logarithm tables this made the practical computations signicantly easier.
1.11. The geodetic inverse problem on the sphere and sin (2 1 ) sin 12 = sin A12 cos 2 sin 12 sin A12 cos 2 .
2 = 1 + arcsin
cos 12 = sin 1 sin 2 + cos 1 cos 2 cos (2 1 ) , sin (2 1 ) . sin A12 = cos 2 sin 12
(1.5) (1.6)
and cos b cos c + sin b sin c cos = (cos b cos c + sin b sin c) + sin b sin c (cos 1) = = (cos b cos c + sin b sin c) 2 sin b sin c sin2 ; 2 bc cos b cos c + sin b sin c = cos (b c) = 1 2 sin2 ; 2 after a few rearrangements we obtain the half-angle cosine rule for a spherical triangle, which also for small triangles is wel behaved1 : sin2 a bc = sin2 + sin b sin c sin2 . 2 2 2
(2.1)
Group 2.1 is valid not only on the ellipsoid of revolution; it applies to all gures of revolution. On a rotationally symmetric body we have p () = N () cos , i.e., d ds dA ds = = sin A , N () cos tan sin A . N ()
If the initial condition is given as 1 , 1 , A12 , we may obtain the geodesic (s) , (s) , A (s) as a solution parametrized by arc length s. Numerically computing the solution using the
11
dA
N co t
p dp
dx 2
p dp
dx
p=
Nc
d
os
M
A ds dy
ds dy
dx
MatLab software is also fairly easy thanks to the ODE routines (Ordinary Dierential Equation) provided. Transformation to rectangular form is easy: X (s) = N ( (s)) cos (s) cos (s) , Y (s) = N ( (s)) cos (s) sin (s) , b2 N () sin (s) . Z (s) = a2
2.3. An invariant
Let us look closer at the quantity p () = N () cos , the distance of a point from the rotation axis. let us compute the s derivative: dp dp dx = = sin cos A. ds dx ds Now, with the equation 2.1 dA sin sin A = ds p we obtain by division
12
2.4. The geodetic main problem d (p sin A) dA dp sin A + p cos A = dA cos A = p sin A + p cos A = sin A = p ( cos A + cos A) = 0. =
Result: the expression p sin A is an invariant.1 This applies on all rotationally symmetric bodies, i.e., also on the ellipsoid of revolution where this is called the Clairaut equation , and of course on the plane. This invariant can be used to eliminate the dierential equation in A from the system 2.1. This yields d ds d ds = = cos A () , M () sin A () , p () (2.2) (2.3)
where A () is obtained from the invariant formula sin A () = sin A12 p (1 ) . p () (2.4)
,
1/2
p () = N () cos = a 1 e2 sin2
cos .
(1)
(1)
(1)
(1)
(1) (1)
2 2 , 2 2 .
(1)
(1)
13
Chapter 2. The geometry of the reference ellipsoid Next, these closing errors could be used for computing improved values A12 , s12 , and so on. The nearly linear dependence between (A12 , s12 ) and (2 , 2 ) can be approximated using spherical geometry. The formulas needed (1.5, 1.6):
(2) (2)
cos s12 = sin 1 sin 2 + cos 1 cos 2 cos (2 1 ) , sin s12 sin A12 = cos 2 sin (2 1 ) . From this, by dierentiation: sin s12 s12 = [sin 1 cos 2 cos 1 sin 2 cos (2 1 )] 2 cos 1 cos 2 sin (2 1 ) 2 , sin s12 cos A12 A12 + cos s12 sin A12 s12 = sin 2 sin (2 1 ) 2 + + cos 2 cos (2 1 ) 2 . So, if we write A= sin 1 cos 2 cos 1 sin 2 cos (2 1 ) cos 1 cos 2 sin (2 1 ) sin 2 sin (2 1 ) cos 2 cos (2 1 ) B= we obtain as iteration formulas: s12 (i+1) A12
(i+1)
0 sin s12 sin s12 cos A12 cos s12 sin A12
=
(i+1)
(i)
(i)
(i)
+ B1 A
2 (i) 2
(i)
to obtain new values 2 , 2 until convergence. The matrices A, B can be recomputed if needed using better approximate values.
(i+1)
14
y b
= sin2 + cos2 = 1
x2 y 2 + 2 1=0 a2 b
15
.P
b
a O
.
R U V
a2 b2 . b2
i.e.,
16
3.5. Co-ordinates in the meridional ellipse i.e., cot PR b = = , cot QR a eli tan = By combining still tan = a2 tan . b2 a tan . b
The equations 3.1, 3.2 represent a description of the meridian ellipse as a function of geodetic latitude . Remember that N is a function of too, so not a constant! In fact x2 y 2 + 2 a2 b = = = the latter condition yields N= using the denition e2 a2 b2 . a2 N 2 1 e2 N2 cos2 + sin2 = a2 b2 N2 N 2 b2 cos2 + 2 2 sin2 = a2 a a N2 b2 cos2 + 2 sin2 = 1; a2 a a 1 e2 sin2
2
17
.
Q
.
S
.
O
If we look at points not on the surface of the reference ellipsoid but above or below it in space, we may write X = (N + h) cos cos , Y = (N + h) cos sin , N 1e
2
(3.3)
Z =
+ h sin .
Here, h is the straight distance of the point from the surface of the ellipsoid (ellipsoidal height). This quantity is interesting because satellite positioning devices can be said to directly measure precisely this quantity (more precisely, they measure X, Y, Z and compute from these h).
A possible stumbling block is identifying the correct quadrant for . and h are more complicated. See gure 3.2, where point P s rectangular co-ordinates X, Y, Z are known and the geographical , , h are to be computed. Let us rst compute the geocentric latitude using the formula: tan = Z +Y2
X2
If point P would be located on the reference ellipsoid, we could determine its geographical latitude by the following formula: Z tan P = (1 e2 ) X 2 + Y 2
18
3.8. Meridian arc length (proof directly from eqs. (3.3).) Now that P is located above the ellipsoid, we obtain in this way the latitude Q of point Q , where Q is the intersection of the ellipsoid and the radius of P , for which (geometrically obviously) the above ratio is the same as for P . Now we compute XQ = N cos Q cos , YQ = N cos Q sin , ZQ = N 1 e2 sin Q , from which OQ = and thus P Q = OP OQ. Additionally, in the little triangle T QP we may compute (Q abbreviated to ): T QP = and thus TP = P Q sin ( ) ,
2 2 2 XQ + YQ + ZQ
TP , PO
(3.4)
P Q
TP cos ( ) . PO
(3.5)
This procedure is in practice fairly precise. If h = 8000 m and = 45 , then T P 26 m, between the P solutions (3.4) and (3.5) there is a dierence of 0.1 mm2 , which is also the order of magnitude of the error that is possibly present in the dierent solutions. The approximation P S T Q contains 0.05 mm of error.3 .
Or then, not. Instead of P O:n one should take M () + h, where M the meridional radius of curvature. Linearly T P (1 cos ( )). 2 3 1 TP . 2 OP
19
Chapter 3. Co-ordinates on the reference ellipsoid Now we may calculate the length of a meridian arc as follows: 0 M d. s (0 ) =
0
(3.6)
a 1 e2 1 e2 sin2
0 3/2
1 e2 sin2
3/2
d.
Here, the last factor can be expanded into a series because e2 sin2 1 and integrated termwise. See the literature. Of course also a numeric approach is possible, nowaday it may even be the superior alternative. MatLab oers for this purpose its QUAD (quadrature) routines.
20
Some of the parameters are geometric (a), some are dynamic (J2 ,). Other geometric and dynamic parameters may be derived as follows ((Moritz, 1992), following (Heiskanen and Moritz, 1967, eqs. 2-90, 2-92)): J2 = e2 2 me 15 q0 2me e2 . = 3J2 + 15q0 1 2 a2 b GM e2 3
Here(Heiskanen and Moritz, 1967, eq. 2-70)) m= and (based on the denitions) be = ae 4 2 a3 e3 . 15 GM 2q0 Furthermore we know ((Heiskanen and Moritz, 1967, eq. 2-58)) e2 = 3J2 + 2q0 e = ja e (e) = 1+ 3 e2 arctan e 3 e with the aid of which (4.1)
e . 1 e2
Now we may compute e2 iteratively using eq. (4.1) , which computes q0 e (e) anew in every step. The result is (table 4.2): Quantity a GM J2 Value 6378137 m 3986005 108 m3 s2 108263 108 7292115 1011 rad s1 Explanation semi-major axis Earth mass G Dynamic form factor Angular rotation rate
21
Chapter 4. Reference systems Quantity e2 2 e b 1/f Value 0.00669438002290 0.00673949677548 6356752.314140 m 298.257222101 Explanation Ensimminen eksentrisyys nelin a oo Toinen eksentrisyys nelin oo Lyhyt akselipuolikas Knteinen litistyssuhde aa
Table 4.2.: GRS80 derived parameters Quantity a 1/f b Value (WGS84) 6378137 m 298.257223563 6356752.314245 m Remark same dierent! di. 0.1 mm
Here we used 1 e2 = 1
a2 b2 b2 1 ab b = 2 ja 1 = 1 = , i.e., 2 a a f a a 1 e2 = 1 1 f
2
1 =1 f
1 e2 .
Often we use (a, f ) together to dene the GRS80 reference ellipsoid. The ocial reference system of the GPS system is the World Geodetic System 1984 (WGS84), whose reference ellipsoid is almost identical with GRS80. However, not exactly, table 4.3:
Most often the dierence, a bit over 0.1 mm at most, can be neglected. It is apparently due to imprecise numerics.
e = p =
GM ab GM a2
1m
m e q0 6 q0 m e q0 1+ , 3 q0
p e ): e
22
4.3. Reference frames Quantity e p f Value 9.7803267715 ms2 9.8321863685 ms2 0.00530244011229 Explanation Normal gravity at equator Normal gravity at poles Gravity attening
As a check, we may still compute Clairauts equation in its precise form ((Heiskanen and Moritz, 1967, eq. 2-75)): 2b q f + f = 1+e 0 . e 2q0 A simple closed, beautiful formula for normal gravity on the reference ellipsoid is SomiglianaPizzettis formula: ae cos2 + bp sin2 = . a2 cos2 + b2 sin2
A Conventional Terrestrial System (CTS) is dened as follows: the origin of the frame coincides with the centre of mass of the Earth
23
Rotational motion
Greenwich
Y
Spring Equinox
X X
sidereal time
Figure 4.1.: Geocentric reference frames
the Z axis is directed along the rotation axis of the Earth, more precisely the Conventional International Origin (CIO) , i.e., the average direction of the axis over the time span 19001905 the XZ-plane is parallel to the zero meridian as dened by Greenwich, more precisely by: earlier the BIH (Bureau International de lHeure, International Time Bureau), today the IERS (International Earth Rotation and Reference Systems Service), based on their precise monitoring of the Earth rotation. In gure 4.2 we see the Earth orbit or ecliptic, the Earth axis tilt relative to the ecliptic plane, and how this tilt causes the most impressive climating variation observable to human beings: the four seasons.
If we study the motion of the Earths axis, and Earth rotation in general, relative to a reference frame connected to the solid Earth itself, we nd dierent quantitities: Polar motion: this consists of an annual (forced) component and a 14-months component called the Chandler wobble. Length of Day.
24
Autumnal equinox True Earth orbit Summer Winter solstice Autumn Apparent Solar path
Ecliptic (zodiac)
Together these are called Earth Orientation Parameters (EOP). They are nowadays monitored routinely, and available after the fact from the International Earth Rotation Service over the Internet. The variation of these parameters is geophysically well understood, e.g., for the Chandler wobble it is mainly the pressure of the Earths oceans and atmosphere that is responsible ((Gross, 2000)).
Local Astronomical
Conventional Terrestrial
Instantaneous Terrestrial
Real Astronomical
Here, , are local astronomical co-ordinates (direction of the plumbline), while xp , yp are the co-ordinates of the pole in the CIO system. 0 is Greenwich Apparent Sidereal Time (GAST). In the sequel we shall show how these transformations are realized in practice.
25
Cyg (Deneb)
Lyra
Lyr (Vega) Umi
Umi (Polaris)
Ecliptic Pole 24 hr
Figure 4.3.: Precession, nutation and the torques from Sun and Moon
26
Plumbline 90 A
A
27
The place of the minus sign is the easiest to obtain by sketching both pairs of axes on paper, mark the angle , and infer graphically whether the for a point on the positive x axis (i.e.: y = 0) the new y co-ordinate is positive or negative: in the above case y = sin cos x 0 = sin x ,
i.e., y < 0, i.e., the minus sign is indeed in the lower left corner of the matrix.
y Q
P
x
V T O S
U x R
29
SR z R y x x x
Figure 5.2.: The associativity of rotations
z=z
S y y z
z y y
x =x
(5.1)
Two dimensional rotations are in fact commutative; they can be described also by complex numbers.
30
completely understandable, because this is a rotation around the same axis, by the same amount, but in the opposite direction. The above formula written in the following way:
n
Rij Rik =
i=1
1 if j = k . 0 if j = k
The columns of a rotation matrix are orthonormal, their norm (length) is 1 and they are mutually orthogonal. This can be seen for the case of our example matrix: cos2 + sin2 = 1, cos sin + ( sin ) cos = 0. Often we encounter other orthogonal matrices: 1. The mirror matrix for an axis, e.g.: 1 0 0 M2 0 1 0 , 0 0 1 which inverts the direction, or algebraic sign, of the y co-ordinate. 2. The axes interchange matrix (permutation): 0 1 0 P12 1 0 0 . 0 0 1 3. Inversion of all axes: 1 0 0 X = 0 1 0 . 0 0 1 Both M and P dier from rotation matrices in this way, that their determinant is 1, when for rotation matrices it is +1. The determinant of the X matrix is (1)n , with n the number of dimensions (i.e., 3 in the above case). A determinant of 1 means that the transformation changes a right handed co-ordinate frame into a left handed one, and conversely. If we multiply, e.g., M2 and P12 , we obtain M2 P12 0 1 0 = 1 0 0 . 0 0 1
The determinant of this is +1. However, it is again a rotation matrix: cos 90 sin 90 0 R3 (90 ) = sin 90 cos 90 0 ! 0 0 1 All orthogonal transformations having positive determinants are rotations. (Without proof still, that all orthogonal transformations can be written as either a rotation around a certain axis, or a mirroring through a certain plane.)
31
Z z s y A K O x T
T z K y
The latter formula is known as the half-angle formula and avoids the problem of nding the correct quadrant for A. The result is in the interval (180 , 180 ] and negative values may be incremented by 360 to make them positive.
32
z y x z y
Geoid prole
Z Y X
Figure 5.4.: From the geocentric to the topocentric system. The matrix R1 mentioned in the text was left out here
R3
rotates the co-ordinate frame around the z axis from the Greenwich meridian to the local meridian of the observation site, rotation angle : cos + sin 0 R3 = sin cos 0 . (5.2) 0 0 1 Seen from the direction of the z axis we see (gure), that x y = x cos + y sin , = x sin + y cos .
(The correct algebraic signs should always be established by the aid of a sketch! Also the directional conventions of dierent countries may dier.)
33
x
z
R2
turns the co-ordinate frame around the y axis, in such a way that the z axis points to the North celestial pole instead of to the zenoth. The rotation angle needed for this is 90 . sin 0 cos . 0 1 0 R2 = (5.3) + cos 0 sin rotates the co-ordinate frame around the new z axis or vertical axis by the amount A0 , after which the x axis points to the azimuth of the zero point of the instruments horizontal circle: cos A0 + sin A0 0 R1 = sin A0 cos A0 0 . 0 0 1
R1
5.6. The geodetic main and inverse problems with rotation matrices
It is possible to solve the geodetic main and inverse problems three-dimensionally, without using the surface geometry of the ellipsoid of revolution. The idea is based on that the three-dimensional co-ordinate of a point or points are given, e.g., in the form (, , h) relative to some reference ellipsoid; and that is given or to be computed the azimuth, elevation angle and distance of a second point as seen from the rst point. In the forward problem one should compute the secont points ellipsoidal co-ordinates (, , h).
More precisely, into co-ordinate dierences in the geocentric orientation as in this case the origin is not the Earths centre of mass!
34
5.6. The geodetic main and inverse problems with rotation matrices 3. Add to the result the geocentric co-ordinates of point A: XB = XA + XAB , jossa (N (A ) + hA ) cos A cos A XA = (N (A ) + hA ) cos A sin A . N (A ) 1 e2 + hA sin A 4. Transform the geocentric co-ordinates of B obtained back to ellipsoidal form (3.7) in the way depicted: XB (B , B , hB ) .
yAB
AB
2 x2 + yAB AB
tan zAB =
zAB
35
37
North pole
Greenwich meridian
h .
Ellipsoidal P normal
. .
0 sin xP 1 0 0 cos xP
0 sin yP . cos yP
Because the angles xp and yp are very small, order of magnitude second of arc, we may approximate sin xp xp and cos xp 1 (same for yp ), as well as xP yP 0, obtaining 1 0 xp 1 0 0 1 0 xp 0 0 1 yp = 0 1 yp . RY (xp ) RX (yp ) 0 1 xp 0 1 0 yp 1 xp yp 1
38
6.5. The quasi-inertial geocentric system 1. The origin of the co-ordinate system again coincides with the Earths centre of mass. 2. The Z axis of the co-ordinate system is again oriented in the direction of the Earths rotation axis, i.e., the North Pole. 3. But: The X axis is pointing to the vernal equinox. Such a reference system doesnt rotate along with the solid Earth. It is (to good approximation) inertial. We also refer to it as an equatorial co-ordinate system. In this system the direction co-ordinates are the astronomical right ascention and declination , . If the celestial spherical co-ordinates , are known, we may compute the unit direction vector as follows: cos cos X Y = cos sin . sin Z RA Here we must consider, however, that, while is given in degrees, minutes and seconds, is given in time units. They must rst be converted to degrees etc. One hour corresponds to 15 degrees, one minute to 15 minutes of arc, and one second to 15 seconds of arc. Going from rectangular to spherical again requires the following formulae (note the use of the half-angle formula for , which is precise over the whole range and avoids the problem of identifying the right quadrant): = arcsin Z = 2 arctan X+ Y . X2 + Y 2
Again, negative values for can be made positive by adding 24 h . The co-ordinates, or places, of stars are apparent, a technical term meaning as they appear at a certain point in time 1 . The places , read from a celestial chart are not apparent. They refer to a certain point in time, e.g., 1950.0 or 2000.0. Obtaining apparent places requires a long reduction chain, taking into account precession, nutation, the variations in Earth rotation, and also the annual parallax and the possible proper motion of the star. The apparent places of stars are found from the reference work Apparent Places of Fundamental Stars precalculated and tabulated according to date. We can transform between RA and X Y = RZ (0 ) Z RA IT (instantaneous terrestrial) co-ordinates as follows: X cos (0 ) sin (0 ) 0 X Y = sin (0 ) cos (0 ) 0 Y . Z IT 0 0 1 Z IT
. . . from the centre of the Earth. The xed stars are so far way, however, that the location of the observer doesnt matter.
39
41
Chapter 7. Reference systems and realizations The scale derives from the SI system. This is realized by using range measurements by propagation of electromagnetic waves. The velocity of these waves in vacuum is conventionally xed to 299 792 458 m s1 . Thus, range measurement becomes time measurement by atomic clock, which is very precise. Orientation: originally the Conventional International Origin (CIO) of the Earth axis, i.e., the mean orientation over the years 1900-1905, and the direction of the Greenwich Meridian, i.e., the plumbline of the Greenwich transit circle. Currently, as the orientation is maintained by the International Earth Rotation and Reference Systems Service (IERS) using VLBI and GPS, this is no longer the formal denition; but continuity is maintained. The current denition uses the BIH (Bureau International de lHeure) 1984 denition of the conventional pole, and their 1984 denition of the zero meridian plane. Together, X, Y and Z form a right-handed system.
42
7.6. The three-dimensional Helmert transformation frame before actual realization. With every ITRS corresponds a number of ITRFs (Frames), which are realizations, i.e., co-ordinate solutions for networks of ground stations computed from sets of actual measurements. Same for ETRS/ETRF, which are the corresponding things for the European area, where the eect of the slow motion of the rigid part of the Eurasian tectonic plate has been corrected out in order to obtain approximately constant co-ordinates. Data used for realizing ITRF/ETRF frames: mostly GPS, but also Very long Baseline Interferometry (VLBI) providing a strong orientation; satellite and lunar laser ranging (SLR, LLR) contributing to the right scale, and the French DORIS satellite system. Nowadays also GLONASS is used. Currently the following realizations exist for ITRS: ITRF88, 89, 90, 91, 92, 93, 94; 96, 97; ITRF2000, ITRF2005 and ITRF2008. The denition of an ITRFyy is as follows (McCarthy, 1996): The mean rotation of the Earths crust in the reference frame will vanish globally (cf. for ETRF: on the Eurasian plate). Obviously then, co-ordinates of points on the Earths surface will slowly change due to the motion of the plate that the point is on. Unfortunately at the current level of geodetic precision, it is not possible to dene a global co-ordinate frame in which points are xed. The Z-axis corresponds to the IERS Reference pole (IRP) which corresponds to the BIH Conventional terrestrial Pole of 1984, with an uncertainty of 0.005 The X-axis, or IERS Reference Meridian, similarly corresponds to the BIH zero meridian of 1984, with the same uncertainty. Finally, note that the Precise Ephemeris which are computed by IGS (the International GPS Geodynamics Service) and distributed over the Internet, are referred to the current (newest) ITRF, and are computed using these co-ordinates for the tracking stations used.
(7.1)
where tx ty tz is the translation vector of the origin, m is the scale factor correction from unity, and (ex , ey , ez ) are the (small) rotation angles about the respective axes. Together we thus have seven parameters. The superscripts (1) and (2) refer to the old and new systems, respectively. Eq. (7.1) can be re-written and linearized as follows, using mex = mey = mez = 0, and replacing the vector X (1) Y (1) Z (1) by approximate values m and the e angles are all assumed small.
T
X0 Y 0 Z0
. This is allowed as
43
Chapter 7. Reference systems and realizations An elaborate rearranging yields 0 (2) (1) Xi Xi Xi 0 Zi0 +Yi0 (2) (1) 0 0 0 Xi0 = Yi +Zi Yi Yi (2) (1) Zi0 Yi0 +Xi0 0 Zi Zi 1 0 0 0 1 0 0 0 1 m ex ey ez
. tx ty tz
(7.2)
Here we have added for generality a point index i, i = 1, . . . , n. The number of points is then n, the number of observations (available co-ordinate dierences) is 3n. The full set of these observation equations then becomes X1 X1 (2) (1) Y1 Y1 (2) (1) Z1 Z1 . . . Xi Xi (2) (1) Yi Yi (2) (1) Zi Zi . . .
(2) Xn X (1) (2) (1) Yn Yn (2) (1) Zn Zn (2) (1) (2) (1)
. . .
. . .
. . .
. . .
. . .
. . .
. . .
m ex ey ez
. tx ty tz
(7.3)
This is a set of observation equations of form + v = Ax (but without the residuals vector v which is needed to make the equality true in the presence of observational uncertainty). There are seven unknowns x on the right. They can be solved in the least-squares sense if we have co-ordinates (X, Y, Z) in both the old (1) and the new (2) system for at least three points, i.e., nine observations in the observation vector . In fact, two points and one co-ordinate from a third point would suce. However, it is always good to have redundancy.
44
7.7. Transformations between ITRF realizations These parameters2 are to be used as follows: X D R3 R2 X Y D R1 Y (t) = 1 + R3 (t) + Z IT RF 2005 R2 R1 D Z IT RF 2008 D R3 R2 X d D R1 Y (t) + + (t t0 ) R3 dt R2 R1 D Z IT RF 2008 T1 T d 1 T2 = + T2 + (t t0 ) dt T3 T3 0.94 0 0 X 0.94 0 109 Y = 1+ 0 (t) + 0 0 0.94 Z IT RF 2008 0.5 + 0.3 (t 2005.0) 0.9 + 4.7 d with the numbers given, and forgetting about the uncertainties. Here refers to the rates, of dt which only that of T1 is non-vanishing in this example.
Note the change in parameter names compared to the previous. For (tx , ty , tz ) we now have (T 1, T 2, T 3); m is now called D; and (ex , ey , ez ) became (R1, R2, R3).
45
In constructing this table, the following knowledge was used: on March 21 at 12 UTC in Greenwich, the hour angle of the Sun, i.e., of the vernal equinox, i.e., sidereal time, is 0h . This Greenwich sidereal time 0 consists of two parts: an annual part a , and a clock time (UTC or Greenwich Mean Time). So we obtain the annual part of sidereal time by subtracting: a = 0 = 12h , i.e., 12h after adding a full turn, 24h . If on March 21, or more precisely at midnight after March 20, sidereal time is 12h 00m , then sidereal time for March 0 is 12h 00m 4 20m = 10h 40m ; Remember that one day corresponds to about four minutes. We may ll out the table using the rule that 1 month 2t . (In principle we could slightly improve the tables accuracy by taking into account the varying lengths of the months. However, the cycle of leap years causes error of similar magnitude.)
47
oW 165
180 W
165 o E
15 0o E
0 15
5 13
12 o 0W
o
hGr
30 N
o
Right ascension
5 13
o
h
o 45 N
E
o 0E 12
o 60 N
105 o W
105
oE
o 75 N
30 o W
15 oW
0o
o 15 E
30
45
60 o E
Greenwich
45
After this, local sidereal time is obtained as follows: = m + d + + hms = a + + hms = = 0 + hms where m the value taken from the above table d the day number within a month a = m + d annual part of sidereal time time (UTC) the longitude of the Earth station converted to hours, minute and seconds (15 = 1h , 1 = 4m , 1 = 4s ).
See the pretty gure 8.1. We have the following quantities: GAST = Greenwich Apparent Sidereal Time (= 0 )
48
75 oE
Earth rotation
90oE
90oW
o 75 W
60
o
o
W
W
Meridian Equator
Figure 8.2.: Hour angle and other co-ordinates on the celestial sphere
LMST = Local Mean Sidereal Time (= ) the equinox varies irregularly with time due to precession and nutation. Thats why we distinguish Mean and Apparent. The dierence is called the equation of equinoxes, ee. h is the hour angle hGr is the Greenwich hour angle is the right ascension (of a celestial object) is the longitude (of a terrestrial object). h = , hGr = 0 , = 0 + , = 0 + ; = 0 0 = ee.
49
East
90
90
Zenith Star
An intermediate form of co-ordinates is: hour angle and declination, h, . The hour angle is dened as shown in gure 8.3, the angular distance around the Earths rotation axis (or at the celestial pole) from the meridian of the location. Equation: h = 0 + , where 0 is Greenwich apparent sidereal time (GAST), is the local (astronomical) longitude, and is the right ascension of the star. If the star is in the meridian, we have h = 0 and = 0 + . On this is based the use of a transit instrument: if, of the three quantities 0 , or , two are knownthe third may be computed. According to the application we speak of astronomical position determination, time keeping of determination of the places of stars. One mans noise is another mans signal . On the celestial sphere there is a fundamental triangle of astronomy: it consists of the star, the celestial North pole, and the zenith. Of the angles of the triangle we mention t (North pole) and A (the zenith), of its sides, 90 (pole-zenith), 90 (star-pole) and (star-zenith). The sine rule: sin A sin h = . cos sin The cosine rule: cos = sin sin + cos cos cos h, sin = sin cos z + cos sin z cos A. We compute rst, using the cosine rule, either or z, and then, using the sine rule, either h or A. Thus we obtain either (A, ) (h, ) in both directions.
50
8.3. Using rotation matrices Let the topocentric co-ordinate vector (length 1) be r = x y z this is x cos A sin r = y = sin A sin . z cos
T
. In spherical co-ordinates
The same vector we may also write in local astronomical co-ordinates: x cos cos r = y = sin cos . z sin The transformation between them is: 1. the direction of the x axis is changed from North to South 2. the new xz axis pair is rotated by an amount 90 to the South, being the astronomical latitude. 3. the new xy axis pair is rotated to the West (clockwise) from the local meridian to the vernal equinox by an amount , the local (apparent) sidereal time. The matrices are: M1 1 0 0 = 0 1 0 , 0 0 1 cos (90 ) 0 sin (90 ) sin 0 cos = 0 , 0 1 0 1 0 = sin (90 ) 0 cos (90 ) cos 0 sin cos sin 0 = sin cos 0 . 0 0 1
R2
R3
Let us compute R3 R2 M1 cos sin = sin cos 0 0 cos sin = sin sin cos 0 sin 0 cos 0 0 1 0 = 1 cos 0 sin sin cos cos cos sin cos . 0 sin
After this: cos cos cos sin sin cos cos cos A sin sin cos = sin sin cos sin cos sin A sin , sin cos 0 sin cos where we identify immediately sin = sin cos + cos sin cos A, the cosine rule in the triangle star-celestial pole-zenith. The inverse transformation is, based on orthogonality (the transpose!) cos A sin cos sin sin sin cos cos cos sin A sin = sin cos 0 sin cos . cos cos cos sin cos sin sin With these, we can do the transformation of spherical co-ordinates with the aid of threedimensional direction cosines.
51
= + . Here, is the right ascension of the ascending node of the satellite orbit. Both longitude and right ascension (ja ) are reckoned positive to the East.
52
R Kevta tasaus i
r A
Figure 8.4.: The places of satellite and Earth station in the inertial system
53
Chapter 8. Celestial co-ordinate systems These are thus the satellites right ascension and declination as seen against the starry sky. With the aid of a star chart and binoculars we may now wait for and observe the satellite. We also obtain the satellites distance: d= d2 + d2 + d2 . 3 2 1
Using the distance we may compute the satellites visual brightness of magnitude. The distances are generally of order 500-1000 km and the magnitudes 2-5.
From this again we obtain the terrestial topocentric vector: d cos T cos ( tT ) d = r R d cos T sin ( tT ) , d sin T from which we may solve the topocentric declination T and hour angle tT . If we then want the right ascension for use with a celestial chart, all we need to do is subtract it from the local sidereal time: T = tT = (0 + ) + tT = 0 + ( tT .) . In this, 0 is Greenwich sidereal time, and ( tT ) the hour angle of Greenwich.
54
8.8. Determining the orbit from observations When also the ground station vector R has been computed, we may compute d1 , d2 by the cosine rule if a suitable value1 satelliitin korkeudelle h tai vastaavasti, satelliitin radan steelle a r = R + h has been given: r2 = R2 + d2 + 2Rd cos (e, R) d2 + 2Rd sin + R2 r2 = 0 2R sin 4R2 sin2 4 (R2 r2 ) = 2 r2 R2 1 sin2 ,
d1,2 =
= R sin
in which sin = cos (e, R) = e eU is the projection of the satellites direction vector onto the local vertical, and cos cos R eU = = cos sin , R sin the zenith vector of the observer, a unit vector pointing straight upward. The value sin , the sine of the elevation angle, may be calculated directly as the dot product e eU when both vectors are given in rectangular form. Only the positive solution makes physical sense: d= r2 R2 1 sin2 R sin .
We know however what the velocity for a circular orbit should be at height h: according to Keplers third law 2 (R + h) 4 2 (R + h)3 vk = = GM P GM . R+h
P =
500 7612.609
750 7477.921
1000 7350.139
1500 7113.071
We see that the satellites linear speed of ight decreases only slowly with height. Therefore we may use the observed velocity v for correcting the height h according to the following formula: h =h vk , v
where vk is the velocity according to Kepler (from the table for the value h), v the calculated velocity, and h the improved value for the satellite height. This process converges already in one step to almost the correct height. Note that the height thus obtained is, in the case of an elliptical orbit, only (approximately) the overight height! The real height will vary along the orbit. For the same reason, the height obtained will not be good enough to calculate the period P (or equivalently: the semi-major
1
An experienced observer can guess a satellites height from the observed motion in the sky surprisingly precisely.
55
Chapter 8. Celestial co-ordinate systems axis a) from. The real period can be inferred only of the satellite has been observed for at least a couple of successive days. The computed vector for the change in position between the two moments of observation, d2 d1 = r2 r1 also tells about the inclination of the orbit, by calculating its vectorial product with, e.g., the satellites geocentric location vector r1 , as follows: (d2 d1 ) r1 = d2 d1 r1 cos i, from which i may be solved. And if i and some footpoint (, ) of the satellite is known, also and (with P ) 0 may be computed. Then we can already generate predictions! When the height of the satellite (and its approximate period) as well as the inclination are known, we can also compute the rapid precessional rate of the ascending node2 : d = d 3 GM = (0 ) ( 0 ) 2 a3
( ) = (0 ) ( 0 )
ae a
J2 cos i,
in which J2 is the dynamic attening of the Earth, value J2 = 1082.62 106 . One of the rst achievements of the satellite era was the precise determination of J2 3 . As a numerical formula: cos i d = 6.52927 1024 3.5 m3.5 degrees/day d a (note the unit!) Thus we obtain the following table (unit degree/days): Ht./Incl. 0 56 65 74 81 90 0.9856 500 -7.651 -4.278 -3.233 -2.109 -1.197 0 97 .401 750 -6.752 -3.776 -2.854 -1.861 -1.056 0 98 .394 1000 -5.985 -3.347 -2.529 -1.650 -0.936 0 99 .478 1500 -4.758 -2.661 -2.011 -1.311 -0.744 0 101 .955
In the table the last row is special. The value 0.9856 degree/day is the Suns apparent angular velocity relative to the background of the stars. If the precessional velocity of the satellites orbital plane is set to this value, the satellite will always y over the same area at the same local solar time. In this way we achieve a so-called heliostationary orbit, also known as noshadow orbit, the advantages of which are continuous light on the solar cells, and in remote sensing, always the same elevation angle of the Sun when imaging the Earth surface. Given is the inclination angle which, for each value of the mean height for each column of the table, gives precisely such a sun-stationary orbit. The precession rate of the satellites orbital plane relative to the sun is now d q 0.9856 degree/day d E.g., for a satellite whose height is 500 km and inclination 56 : q = 4.278 0.9856 = 5.2636 degree/day, and the period of one cycle is 360/q = 68 days 2.2 months. This is the time span after which the satellite appears again, e.g., in the evening sky of the same latitude.
2 3
The formula applies for circular orbits. In fact, this motion of the ascending node is so rapid, that without considering it it is not possible to generate usable orbit predictions.
56
Carl Friedrich Gauss, (1777 1855), was also among the rst to speculate on the possibility of non-Euclidean geometry. Others were Jnos Bolyai and Nikolai I. Lobachevsky. a
57
Chapter 9. The surface theory of Gauss Let us assume in the sequel, that the parametrization used is unambiguous and dierentiable (and thus continuous). The tangent of the curve C is obtained by dierentiation: t (s) = the length of the tangent is t = dx ds
2
dx (s) = xs ; ds
dy ds
dz ds
dx2 + dy 2 + dz 2 = 1 = 1. ds2
This applies only if s is a parametrization by distance. An arbitrary parametrization t can always be converted into a parametrization by distance in the following way: s(t) = =
0 0 t t
ds d = d dx d
0 2
dx2 + dy 2 + dz 2 d = d dy d
2
(9.1)
dz d
d,
dy (t) dt
dz (t) dt
from which s (t) can always be computed by integration 9.1. Another dierentiation yields the curvature vector : dt (s) d2 = 2 x (s) . ds ds
k (s) =
These vectors are called the tangent vectors of surface S and parametrization (u, v).
58
v + dv
xv
G /2
1
E /2
1
xu x u O
Figure 9.1.: The Gauss rst fundamental form
u + du
From these we we obtain, as dot products of two vectors in space, the elements of the fundamental form: E = F = G = xu xu , xu xv , xv xv .
In gure 9.1 we have depicted the Gauss rst fundamental form and the tangent vectors xu , xv . It is easy to show (chain rule in three dimensions (x, y, z)), that ds2 = = +2 + = dx2 + dy 2 + dz 2 = x u
2
+ + +
y u
z u + z v
du2 + z z u v
2
x x u v x v
2
y y u v y v
2
dudv +
dv 2 =
xu xu du2 + 2 xu xv dudv + xv xv dv 2 ,
from which (9.2) follows directly. We see, e.g., that in the direction of the v curves (dv = 0): ds2 = Edu2 , in other words, E represents the metric distance between two successive curves (u, u + 1). Similarly G represents the distance between two successive v curves. The closer the curves are to each other, the larger xu or xv and also the larger E or G. F again represents the angle between the u and v curves: it vanishes if the angle is straight.
59
See gure 9.2. When moving from location to location, the normal vectors direction changes: when we move from point x (u, v) to point x (u + du, v), the normal changes n n = n + dn. In the same way, when moving from point x to x (u, v + dv), the normal changes n n = n + dn . As a formula: dn = n n du + dv = nu du + nv dv. u v
60
gdv n + nv dv
f du
edu f dv n + nu du n
xv v + dv x O v u
xu
u + du
The norm of the normal vector, i.e., its length, is always 1, like we saw above; thats why the vector can change only in two directions, either in the direction of tangent vector xu , or in the direction of tangent vector xv . Let us separate them by projection: dn xu dn xv = = nu xu du + nv xu dv, nu xv du + nv xv dv,
in which we may directly identify the elements of the second fundamental form e, f, g with the aid of (9.5): dn xu dn xv = edu + f dv, = f du + gdv.
Contrary to the rst fundamental form, the second fundamental form has no corresponding object in the Riemann surface theory. It exists only for surfaces embedded in a surrounding (Euclidean) space. Often we can also nd a tensorial notation: 11 = e, 12 = 21 = f, 22 = g.
61
Chapter 9. The surface theory of Gauss The latter means that if the direction of the normal vector n changes by an amount dn when we travel a distance du along the v co-ordinate curve, we still dont know how many metres the distance du corresponds to. If it is a long distance, then the same change in the normal vector dn means only a small curvature of the surface; if it is a short distance, then the same change dn corresponds to a large surface curvature. The fact that the parameter curves u = constant and v = constant generally not are perpendicular to each other, makes this problem even trickier. Write the rst and second fundamental forms in matrix form: H= Form the matrix: E F F G ,B= e f f g .
C H 1 B = =
G F
F E
1 EG F 2
e f f g
=.
1 EG F 2
Ge F f Gf F g Ef F e Eg F f
This is called the shape operator, and the above, the Weingarten2 equations, see http://en. wikipedia.org/wiki/Differential_geometry_of_surfaces#Shape_operator We can say that multiplication by the inverse of the H matrix, i.e., the rst fundamental form, which describes the length of a distance element, performs a metric scaling of the B matrix 3 . Principal curvatures: The matrix C has two eigenvalues: the values 1,2 for which (C I) x = 0
T
(9.6)
. The solutions 1,2 are called the principal curvatures for suitable value pairs x = du dv of surface S. They are invariant with respect to the chosen parametrization (u, v). The corT T responding value pairs x1 = du1 dv1 and x2 = du2 dv2 dene the local principal directions of curvature on the surface. Other invariants: det B eg f 2 = is the Gaussian curvature. det H EG F 2 1 1 1 eG + gE 2f F 2. The half-sum (1 + 2 ) = (C11 + C22 ) = is the mean curvature. 2 2 2 EG F 2 1. The product 1 2 = det C = Principal directions of curvature: We may also study the eigenvectors of C, which are called principal directions of curvature. For this, write 0 = H (C 1 I) x1 = (B 1 H) x1 , 0 = H (C 2 I) x2 = (B 2 H) x2 .
2 3
Julius Weingarten, German mathematician 1836-1910 A more technical description: B is a covariant tensor ij , and H the covariant metric tensor gij . H 1 again i corresponds to the contravariant tensor g ij . C is now the mixed tensor k = g ij jk , whose tensorial eigenvalue problem is i i j j xj = 0, the same equation as (9.6). See chapter 10.
62
9.4. Principal curvatures Multiply the rst from the left with xT , and the second with xT and transposing: 2 1 0 =xT Bx1 1 xT Hx1 , 2 2 0 = xT Bx2 2 xT Hx2 1 1
T
= xT Bx1 2 xT Hx1 , 2 2
as, both B and H being symmetric matrices, we have xT Bx2 1 xT Hx2 1 Subtraction of the two yields (2 1 ) xT Hx1 = 0. 2 This shows that if the principal radii of curvature are dierent 4 , then the expression xT Hx1 2 vanishes. This expression5 can be interpreted as an inner product: in fact, in Cartesian plane co-ordinates in the tangent plane, the matrix H = I, and we have xT x1 = 0, or x1 x2 in the 2 Euclidean sense. The principal directions of curvature are mutually perpendicular. This is just a special case of self-adjoint (symmetric) operators having mutually orthogonal eigenvectors, e.g., the eigenfunctions of Sturm-Liouville theory (http://en.wikipedia.org/ wiki/Sturm%E2%80%93Liouville_theory). Example: The co-ordinates of a point on the surface of the ellipsoidal Earth are N () cos cos x = N () cos sin . N () 1 e2 sin From this x x = = d (N () cos ) d d sin (N () cos ) d d 1 e2 (N () sin ) d cos ;
T T
Using the formulas derived in the Appendix B we obtain: sin cos x = M () sin sin . + cos cos sin x = N () + cos cos . x = 0 The surface normal is obtained as the vectorial product, normalized: n=
4
x x , x x
And if they are not, any linear combination of x1 and x2 will again be an eigenvector, and we can always choose two that are mutually perpendicular. 5 In index notation: gij xi xj . 2 1
63
Chapter 9. The surface theory of Gauss where x x cos2 cos = = NM cos2 sin 2 2 sin cos cos sin cos sin cos cos2 cos = N M cos2 sin = N M cos2 sin , tan sin cos x x = N M cos2 Thus
not a surprising result. . . Let us compute the rst fundamental form: E = F = G = x x = M 2 sin2 sin2 + cos2 + cos2 = M 2 , x x = 0, x x = N 2 cos2 = p2 .
We calculate for use in calculating the second fundamental form + cos sin + sin cos n = + sin sin ; n = cos cos 0 cos and thus (equations 9.5) e = n x = +M, f In other words: H = B = E F F G e f f g = = M2 0 0 N 2 cos2 M 0 0 N cos2 , = n x = n x = 0, g = n x = +N cos2 .
, and
As could be expected. . .
64
In other words, the curvature vectors components in the (u, v) co-ordinate system ontain other things besides just the derivatives of the component values dt1 /ds ja dt2 /ds. We write xuu = 1 xu + 2 xv + xuu n n, 11 11 xuv = 1 xu + 2 xv + xuv n n, 12 12 xvv = 1 xu 22 + 2 xv 22 + xvv n n; (9.7)
i.e., we develop the three-dimensional vectors on the base (xu , xv , n) of the space. Here appear or naturally arise the symbols or Christoel symbols which we discuss later (Chapter 10). They illustrate the reality that dierentiating a vector (also known as parallel transport, see Chapter 10) in curvilinear co-ordinates on a curved surface is not a trivial thing. The third term on the right hand side however represents, according to denition (9.4) , the elements of the second fundamental form e, f, g. We obtain k = + + dt1 + 1 t1 11 ds dt2 + 2 t1 11 ds e t1
2 2
+ 21 t1 t2 + 1 t2 12 22 + 22 t1 t2 + 2 t2 12 22
2
xu + xv +
+ 2f t1 t2 + g t2
n.
Here, the rst two terms represent the internal curvature kint of the curve C, the curvature inside surface S; the latter term represents the exterior curvature kext , the curvature of the curve along with the itself curved surface. The internal curvature again has two components in (u, v) co-ordinates, which can be read from the above equation. We write kint = k 1 xu + k 2 xv ,
6
Dont get nervous about the use of a superscript (upper index). Its just a writing convention, the good sense of which we will argue later on
65
2 2
+ 21 t1 t2 + 1 t2 12 22 + 22 t1 t2 + 2 t2 12 22
2 2
1 ti tj ij j=1 , 2 2 i j ij t t
i=1 j=1
where we have economized the formulas by using summation signs. The external curvature again is kext = k n n, where k n = e t1 Example: a latitude circle on the Earth surface.
2
+ 2f t1 t2 + g t2
N cos sin sin 1 dx d dx = = +N cos cos = + cos ; t= ds d ds N cos 0 0 cos dt dt d 1 k= = = sin . ds d ds N cos 0 External curvature: cos cos n = cos sin , sin so 1 n cos2 + sin2 n = . N N Because the vector n is a unit vector, we may infer, that the exterior curvature of the curve is precisely the inverse of the transverse radius of curvature, and directed inward. This is the same as the curvature of the surface taken in the direction of the curve. kext = k n n = The internal curvature is kint = k kext 1 = N cos + cos cos cos sin + cos sin cos + sin tan cos sin = sin sin . N + cos
66
p/sin
= N cot kint
p k kext
N = p/cos
Figure 9.3.: The curvatures and radii of curvature of a latitude circle 1 1 = ; its internal part kint , N cos p () tan 1 1 sin = ; and its external part kext , length k cos = cos = length k sin = p () N () p () 1 . In the gure one also sees how the distance from the Earths rotation axis is in every N () direction (k, kint ja kext ) the inverse of the curvature. This is also intuitively clear from rotational symmetry. In Figure 9.3 we see the curvature vector k, length k =
i tj tk = 0. jk
j=1 k=1
(9.8)
This approach is developed further later on in chapter 10.5. Externally, using vectors in space An alternative, three-dimensional (exterior) form is obtained by observing, that the geodesic is only externally curved, i.e., by writing dt = kext = e t1 ds Because t xu t xv = t1 xu xu + t2 xu xv = Et1 + F t2 , = t1 xv xu + t2 xv xv = F t1 + Gt2 ,
2
+ 2f t1 t2 + g t2
n=
t1 t2
e f f g
t1 t2
n.
67
Chapter 9. The surface theory of Gauss it follows that t1 t2 If we dene t1 t2 we may write dt = ds Here E F F G t xu t xv E F F G
1 1
E F F G
t xu t xv
t xu t xv
e f f g
1
E F F G
t xu t xv
n.
e f f g
E F F G
= H 1 BH 1
following the earlier used notation7 . Example: on the ellipsoidal surface of the Earth, we have H 1 BH 1 = = and t xu t xv from which we obtain dt ds = = Here, the unit vectors eN = xu xu sin cos cos sin xv = sin sin , eE = = + cos cos . xv cos 0 t eN 1 t eN M t eE
2
M 2 0 0 N 2 cos2 M 3 0 0 N 3 cos2 =
M N cos
2
M 2 0 0 N 2 cos2
1/M 0
2
0 cos /p n.
t eN t eE
n=
1 t eE N
1 The approach is geometrically intuitive. The expression in wave brackets t eN 2 + M 1 1 1 t eE 2 = cos2 A+ sin2 A is precisely the curvature of the surface in the direction N M N of the curve. dx In addition we have the equations = t, altogether 2 3 = 6 ordinary dierential ds equations. The method of space vectors has both advantages and disadvantages compared to the surface co-ordinate method (e.g., equations (2.1)). Advantage: in surface co-ordinates (, ) there are inevitably two poles, singularities where the curvature of the latitude circles goes to innity and numerical methods can be ill-behaved. This will not happen in rectangular space co-ordinates. Disadvantages:
7
In index notation: g ij jk g kl , see chapter 10. A logical notation for this would be il .
68
9.6. The geodesic 1. More equations, more computational eort. 2. At every point, we must compute M and N and for this = arctan p= X 2 + Y 2. Z , where (1 e2 ) p
3. The roll-in and roll-out of the computation of the geodesic presupposes the transformation of (, ) to (X, Y, Z) co-ordinates in the starting point, and back in the end point.
69
Note that here we have both a symbolic notation (v) and a component notation (v i , i = 1, 2). The components depend on the chosen co-ordinate system (x = x, y = y, although 2 2 always x2 + y 2 = x + y = constant an invariant), but the symbolic notation does not depend on this. A vector is always the same thing, even if its co-ordinates transform with the co-ordinate system. A vector is always the same arrow in space. There exists the following transformation formula for changing the components of a vector: vi =
i
1 2
i i v i ,
(10.2)
Georg Friedrich Bernhard Riemann (1826 1866), German mathematician. Albert Einstein, 1879 1955, German-born theoretical physicist, discoverer of relativity theory, and icon of science.
71
Chapter 10. The surface theory of Riemann or in matrix language v = Av, where v = vi = v1 v2 = x y
is the component matrix of the transformation operator. Every quantity that transforms according to equation (10.2) is called a vector. Familiar vector quantities are velocity, acceleration, force, . . . what they have in common is, that they can be graphically presented as arrows. Tensors A tensor is just a square object a matrix which has the same transformation property from on co-ordinate system to another, but for every index : Ti j =
i,j i j i j T ij .
T = AT AT , where A is the same as dened above. We have already encountered many tensors in geodetic theory: 1. The inertial tensor of the Earth. 2. The gravity gradient or Marussi tensor g x g M = = y g z 2W x2 2W yx 2W zx 2W xy 2W y 2 2W zy 2W xy 2W xz 2W z 2 .
2 x xy , where x and y are the 2 xy y mean errors of the co-ordinates, i.e., the components of x, and xy is the covariance.
4. In the earlier discussed surface theory of Gauss, the rst, H = B= e f f g fundamental form, as well as C = H 1 B;
5. Also the elements of the map projection fundamental form (to be discussed later) E, F , G E F constitute a tensor H = , a metric tensor of sorts. The object H 1 H = F G F E M2 Mp may again be called the scale tensor. F G M p p2
3
Verify that the matrix equation produces the same index summations as the index equation!
72
10.1. What is a tensor? The geometric representation of a tensor and invariants In the same way that the geometric depiction of a vector is an arrow, is the geometric depiction of a tensor an ellipse (in two dimensions) or an ellipsoid (in three dimensions).The lengths of the principal axes of the ellipse/ellipsoid depict the eigenvalues4 of the tensor; the directions of the principal axes again are the directions of the tensors eigenvectors. In Euclidean space and in rectangular co-ordinates, tensors T ij are typically symmetric; therefore, to dierent eigenvalues i , j , i = j belong eigenvectors xi , xj that are mutually perpendicular, as proven in mathematics textbooks. In an n-dimensional space, a tensor T has n independent invariants. The eigenvalues i , i = 1, . . . , n are of course invariants. So are their sum and product. i =
i i
T ii ,
on the two-dimensional plane 1 + 2 = T 11 + T 22 , the sum of the diagonal elements or trace 5 ; ja i = det (T ) ,
i
Mii =
i
2W 2W 2W + + W, x2 y 2 z 2
the Laplace operator. Tensors in general co-ordinates And what about non-Euclidean spaces, with non-rectangular co-ordinate systems? Well, in that case the dierence between sub- and superscripts becomes meaningful, and the reason we write superscripts may nally be appreciated. A contravariant vector transforms as follows: vi =
i i i v i
(10.3)
(10.4)
More precisely, the lengths of the semi-axes are the square roots of the eigenvalues i . germ. Spur.
73
Chapter 10. The surface theory of Riemann Where the prototype of a contravariant vector is the co-ordinate dierences if two (nearby) points, as above: x vi = , y is the prototype of a covariant vector the gradient operator : vj = V (x, y) is some scalar eld in space. If we take for a model vector vi = we obtain vi = also with the chain rule V /x V /y The coecient matrices =
x/x y/x x/y y/y
V = xj
V /x V /y
(10.5)
dx dy
dx dy
x /x y /x
x /y y /y
dx dy
V /x V /y
xi xi i i = i are apparently each others inverse matrices. = i and xi xi So, if the matrix form of the covariant transformation equations (10.4) transformation parami eters i is A (i row and i column index), then the matrix of the contravariant transformation i parameters (equation 10.3) i must be A1 (i row and i column index). From this circumstance, the names covariant and contravariant derive. A tensor may have both super- and subscripts. Under a co-ordinate transformation, each index changes according to its character. There may even be more than two indices.
j j i because if the matrix form of j is A, then that of i or correspondingly, that of j is A1 , as shown above.
6
74
10.2. The metric tensor 2. The corkscrew tensor of Tullio Levi-Civita7 in three dimensions: if, of ijk, two are the same 0 1 if ijk is an even permutation of the numbers (123) = ijk 1 if ijk is an odd permutation of the numbers (123) That is,
123
231
312
= 1,
132
321
213
= 1, all others = 0.
Ks. http://folk.uio.no/patricg/teaching/a112/levi-civita/.
This is called the metric of the rectangular co-ordinate system in the Euclidean plane. In equation (10.6) it is assumed that we sum over the indices i and j; we call this the Einstein summation convention. Always when in an equation we have a subscript and a superscript with the same name, we sum over it. I.e., in this case
2 2
ds =
i=1 j=1
An alternative way of writing using matrix notation is a quadratic form: ds2 = = dx Hdx = dxT Hdx = dxT dx1 dx2 H g11 g12 g21 g22 dx dx1 dx2 .
If the surface is not curved, one may always nd a co-ordinate system which is everywhere rectangular and both co-ordinates scaled correctly such that to a co-ordinate dierence of 1 m corresponds also a location dierence of 1 m. Then the gij matrix or metric tensor has the form of the unit matrix, like above. If the surface is curved, we can nd a unit matrix only in some places. E.g., on the surface of the Earth, only within a strip in the vicinity of the equator. It wont be possible in a unied manner over the whole surface of the Earth.
75
s
q v v
pu
u + u
Nevertheless we an choose also on a non-curved surface a co-ordinate system that isnt rectangular but skewed, and where the co-ordinates are scaled arbitrarily. See gure 10.1. In this case we nd with the aid of the cosine rule: s2 = p2 u2 + q 2 v 2 + 2puqv cos , or, dierentially ds2 = p2 du2 + q 2 dv 2 + 2pq cos dudv, or, in index notation ds2 = gij dxi dxj where gij = The matrix representation: dxi du dv gij , i, j p2 pq cos pq cos q2 dxj du dv . p2 pq cos pq cos q2 , dxi = du dv
ds2 =
On a curved surface gij will depend on place, gij xi , where xi is a vector of parameters describing the surface, xi = u v . Also on a non-curved surface can gij depend on place, e.g., when choosing curvilinear, e.g., polar, co-ordinates. Examples: 1. The surface of a spherical Earth: ds2 = R2 d2 + R2 cos2 d2 , i.e., gij (, ) =
7
R2 0 2 0 R cos2
, dxi =
d d
76
10.3. The inverse metric tensor from which in matrix language dxi d d gij , i, j R2 0 0 R2 cos2 dxj d d .
ds2 =
2. Polar co-ordinates (, ) in the plane: ds2 = d2 + 2 d2 , i.e., gij = from which ds2 = dxi d d gij , i, j 1 0 0 2 1 0 0 2 , dxi = d d , dxj d d .
3. In three dimensions in the air space using aviation co-ordinates: nautical miles North dN , nautical miles East dE, feet above sea level dH; in metres ds2 = (1852)2 dN 2 + (1852)2 dE 2 + (0.3048)2 dH 2 , eli 18522 0 0 dN dxi = dE , gij = 0 18522 0 dH 0 0 0.30482 gij , i, j ds2 = dN dxi dE dH dxj
g 11 g 12 g 21 g 22
1 0 0 1
77
All three forms are equivalent, which is easily proved. For eigenvectors we have xi = gij xj . If the tensor T ij (or Tji , or Tij ) is symmetric (meaning T ij = T ji etc.), then the eigenvalues are real and the eigenvectors mutually orthogonal: if xi , y i are dierent eigenvectors, then gij xi y j = 0. There are as many eigenvalues as there are dimensions in the space, i.e., in the plane R2 two.
78
v1 v v2 yi v1 v v2
u +y
u2 u1 u1 + y 1
Figure 10.2.: Christoffel symbols and parallel transport The Christoel symbols i do not form a tensor like the metric tensor does. They describe the jk curvilinearity of a curvilinear co-ordinate system, i.e., a property of the co-ordinate system. A co-ordinate transformation that removes at least locally if not everywhere the curvature of the co-ordinate curves, also makes the elements of i vanish in that point (for tensors, such a jk local transforming away will never succeed!). E.g., on the Earth surface in the (, ) co-ordinate system on the equator the co-ordinate curves are locally non-curved and later we shall show, theat there indeed all i vanish. Anjk other example is from the general theory of relativity, where the components of acceleration are Christoffel symbols: acceleration can be transformed away by moving to a falling along reference system. In Einsteins falling elevator the people inside the elevator accelerate with respect the Earth surface but are weightless (i.e., the acceleration vanishes) in a reference system connected to the elevator. The Christoel symbols can be computed from the metric tensor; the equation is (complicated proof; see appendix C): g j gjk 1 gk i = g i + . (10.7) jk j k 2 x x x (g ij (gij )1 .) From this is can be seen, that the Christoel symbols are, like, the rst derivatives of the metric. In a straight-lined co-ordinate system on a non-curved surface gij is a constant and thus all i vanish. Also, because gij is symmetric (gij = gji ), we obtain jk i = i . jk kj The Christoel symbols are useful when writing the equation of the geodesic 9 in this formalism (cf. equation (9.8)): d i t + i tj tk = 0. jk ds Here, ti is the tangent vector of the geodesic ti = dxi/ds.
9
In fact we write
Dti dti + i tj tk , jk ds ds the so-called absolute or covariant derivative which is a tensor; and Dti =0 ds is then the equation for the geodesic, which thus applies in curvilinear co-ordinates.
79
Chapter 10. The surface theory of Riemann Examples: 1. On the surface of the spherical Earth (, ). We use a notation where and symbolize the index values 1 and 2 (for these symbols Einsteins summation convention thus fails to work!): gk gk 0 0 0 0 = , = , 2 0 0 0 R sin 2 so only g g22 = = R2 sin 2 is non-vanishing. Then, remembering that g = (gi )
i 1
R2 0 0 R2 cos2
R2 0 0 R2 cos2
the only non-zero elements are: = = 1 g (g )1 2 g = 1 1 == R2 R2 sin 2 = + sin 2 = sin cos , 2 2 1 R2 cos2 2 1 (g )1 2 R2 sin 2 = = tan . sin 2 = tan , 2 cos2
1 (g )1 2
Note that at the equator = 0 all i = 0. jk 2. Polar co-ordinates in the plane (, ): gk = yielding g22 g = = 2, so = = = 1 g (g )1 2 1 g (g )1 + 2 1 2 = , 2 1 1 1 = 2 2 = + . 2 = 0 0 0 2 , gk = 0 0 0 0 ,
80
10.6. The curvature tensor Simultaneous integration of the dierential equations would give cos A R = sin A , R cos as a function of s, from which A and may be computed. To this set we add the denition equations of the tangent d = = t1 , ds d = = t2 , ds In this way also comes along. Perhaps this approach seems overly complicated; its major theoretical advantage is, that it will work for all curvilinear co-ordinate systems on the surface considered, also, e.g., for stereographic map projection co-ordinates used in the polar areas, as long as we rst manage to write down the metric tensor of the co-ordinate curves. The length of the tangent vector
T
is computed as follows:
2
cos A R
+ R2 cos2
sin A R cos
= 1,
gij =
R2 0 2 0 R cos2
on the surface of the sphere, The lengths of the tangent vectors have always to be 1; this requirement is fullled if s is a parametrization of the curve by distance.
1. the orientation of the closed rectangular path, the side indices k and ; 2. the size of the rectangle, measured in curvilinear co-ordinates: let the length of the uk side beuk and the length of the u side, u ; 3. the transported vector v i itself. In the following way:
i v i = Rjk v j uk u .
(10.8)
i Here, the creature Rjk is called the Riemann curvature tensor. In two-dimensional space (i.e., on a surface) it has 24 = 16 elements.
i i + i m i m m . km j jk k j x x jk
(10.9)
81
D u + u
C v i vi uk A B u
uk
uk + uk
Figure 10.3.: The curvature tensor and parallel transport around a closed grid path
There exist many more such symmetries; the number of independent components is actually small. Esimerkkej: a 1. Surface of a spherical Earth (, ): Based on the antisymmetry property we may say that only elements for which k = l can dier from zero. Additionally the Christoffel symbols depend only on . Thus we obtain the auxiliary terms:
= = =
The terms i m are obtained in the following way: km jl = = = sin2 , = = + tan2 . By combining
R = R = R = R = R = R = R = R =
82
= 1, 1 = = 2 ,
= = = 1, 1 = = + 2 . After this:
R = R = 0; R = R = R = R R = R
= = 0.
1 1 + m m = 2 + 2 = 0; m m + m m = 1 + 1 = 0; m m
So the whole Riemann tensor vanishes, as it should, because the surface is not curved. From the Riemann tensor we obtain the smaller Ricci12 -tensor in the following way:
2
i Rjik
=
i=1
i Rjik .
(10.10)
1. Let us continue with the computation of Rij for the case of a spherical surface:
R = R + R = +1; R = R + R = 0 + 0 = 0; R = R + R = cos2 .
2. For polar co-ordinates in the plane, all Rij = 0. We may continue this process to obtain the curvature scalar 13 :
2
R = g Rji =
j=1
ij
(gij )1 Rji .
(10.11)
Gregorio Ricci-Curbastro (1853 1925), Italian mathematician, inventor of tensor calculus i i This is in fact the trace of Rij , more precisely, of Rj g ik Rkj , written Ri , see above. This also serves as an example of how in general curvilinear co-ordinates an index may be raised from covariant to contravariant, or lowered, using the metric tensor gij or its inverse g ij . In rectangular co-ordinates these are unit matrices and the dierence between super- and subscripts is without consequence.
83
Chapter 10. The surface theory of Riemann 1. Spherical surface: 1 R2 = 0 0 1 2 cos2 R 1 0 0 + cos2 = R2 0 0 R2 ,
i Rk = g ij Rjk
i.e.14 R=
i i Ri =
1 cos2 2 + 2 = 2. 2 2 R R cos R
2. In polar co-ordinates Rij = 0 , i.e., R = 0. More generally we have (without proof): R = 2K = 21 2 = 2 , R1 R2 (10.12)
twice the Gauss total curvature K, the inverse of the product of the two principal radii of curvature.
This is not hard to show: 1. in the equation (10.8) uv describe the sides of a small diamond shape, around which the vector v i is transported in a parallel fashion. In two dimensions there are only two choices for the sides in the uk , u directions: uk u and u uk . One represents clockwise i i transport, the other, counterclockwise. The corresponding elements Rjk = Rj k thus are essentially the same. 2. In the same equation v j and v i are mutually perpendicular. v i represents a small rotation of vector v j . On a surface, in two dimensions, the rotation is described by one angle. Because the angle between two parallelly transported vectors doesnt change, we may infer that this rotation angle is the same for all vectors v i . Again we nd only one independent parameter. See gure 10.4.
i In other words, when give a u1 u2 diamond shape, the expression Rj12 u1 u2 = cos sin i Rj21 u2 u1 is a 2 2 rotation matrix , containing only one free pa sin cos rameter.
However, for higher dimensionalities, the number of independent elements in the tensors of Riemann and Ricci is larger. This case is interesting because of General Relativity (four dimensions!) though not for geodesy. We can remark here, that the spherical excess already discussed in section 1.2 is a special case of the change in direction of a vector that is parallelly transported around: the spherical excess is the small change of direction of a vector transported around a closed triangle!
14
Here we use the symbol R both for the radius of the Earth and for the curvature scalar. We hope it doesnt cause confusion. 15 The radii of curvature exist only in the three-dimensional space surrounding the Earth surface, in which it is embedded. R on the other hand is an intrinsic property of the surface.
84
C 3 D 4 ABCD 2 1 B
A
Figure 10.5.: Transport of a vector around a larger surface area. ABCD = 1 + 2 + 3 + 4 . When we transport a vector around a larger surface area, it is the same as if we had transported it successively around all of the little patches making up the surface area. This is depicted in gure 10.5. In every little patch du1 du2 , area dS, the change in direction of the vector amounts to KdS, where K is the Gauss total curvature at the patch location. By generalization one obtains from this the following integral equation (Gau-Bonnet16 for the triangle, Theorema elegantissimum): = K (, ) dS,
which is the surface area of the corresponding triangle, but on the unit sphere.17 . From this follows that
16 17
Pierre Ossian Bonnet (1819 1892), French mathematician This is not quite exactly true. . . the triangle symbols under the integral sign dd and under the integral sign
85
Chapter 10. The surface theory of Riemann on the reference ellipsoid the spherical excess depends only on the directions of the normals in the corner points (i , i ) , i = 1, 2, 3, not on the shape of the ellipsoidal surface. Shorter: spherical excesses are computed on the sphere, even while all other computations are done on the ellipsoid. It suces that the geographical co-ordinates of the corner points, which describe the ellipsoidal normal, are known A fast geometric way of computing the spherical excess of a triangle is the following: Let the corner points cos i cos i xi = cos i sin i , i = 1, 2, 3; sin i we use the polarization method (1.7) in three dimensions, in order to nd the poles of the triangles sides: y1 = x2 x3 x3 x1 x1 x 2 , y2 = , y3 = . x2 x3 x3 x1 x1 x 2
Now the inter-pole distances correspond to the angles of the spherical triangle (more precisely, minus those angles): 1 = 2 arctan 2 3 y2 y 3 , y2 + y 3 y1 y 3 , = 2 arctan y1 + y 3 y y2 = 2 arctan 1 , y1 + y2
numerically strong equations18 . Here i is the angle at corner point i. The spherical excess is now
3
=
i=1
i .
1 if i = j 0 otherwise,
86
10.8. The curvature in quasi-Euclidean geometry In this case the Christoffel symbols (equation 10.7) are i = jk = gjk g j 1 i gk = g + j k 2 x x x gjk 1 gki gij + 0 j k 2 x xi x
based on the above assumption of stationarity. Of the Riemann curvature tensor (equation 10.9) the last two terms vanish:
i Rjk
= = =
i xk j 1 2 xk 1 2 x 1 2 xj
1 2
(xi )2
+ = .
1 2
2 gji 2 gii 2g i + xi xj x xj x xi
1 1 = gj + 2 2
2 gij 2 gii 2g i + i xi xj x x x xj
This already looks a lot more symmetric. The symbol here means the Laplace operator =
i
2 (xi )2
= = =
1 2 1 2
gjj +
j i j
(xj )2
gjj +
j i j
gii =
i
gii +
i i j
2 gij . xi xj
In the special case that the co-ordinate curves are (everywhere, not just in point P ) orthogonal, we obtain gij = 0 if i = j with its derivatives of place, i.e. R=
i
gii +
i
2 gii (xi )2
g11 (x1 )2
g22 (x2 )2 +
g11 (x1 )2 .
g11 (x2 )2
g22 (x1 )2
87
(11.1)
Here we have suitably dened two ways of writing, the index notation gij , dxi and the matrixvector notation H, x. The elements of the matrix are also the elements of the Gauss First Fundamental Form on the Earth surface E, F, G. We see that E = M 2 , F = 0 ja G = p2 . For the azimuth A again we obtain the following equation: tan A = from which, with the above, dS sin A = pd, dS cos A = M d. (11.4) (11.5) pd , M d (11.3)
89
x d + y d +
x d, y d,
y y d + d
= (11.6)
G =
E, F ja G are the Gauss First Fundamental Form in the map plane. If we interpret the element of distance in the map plane ds2 as a metric of the Earth surface, then we obtain also here a metric tensor, E F H. (11.7) gij = F G The corresponding metric is ds2 =
i,j
where dx1 = d and dx2 = d, , i.e., dxi = dxj = x = d d . Also here we see as alternatives the index notation and the matrix-vector notation, which describe the same thing.
ds2 = dS 2
i,j i,j
90
11.1. Map projections and scale The same equation in matrix language, if x = dx1 dx2
T
d d
xT H m2 H x = 0. Equations (11.4, 11.5) The read: dS sin A = pd, dS cos A = M d. Let us now look at all vectors that are of form x= d d cos A M = sin A . p (11.9)
+ p2
sin A p
= cos2 A + sin2 A = 1. So: on the surface of the Earth these vectors form a circle of unit radius. Substitute (11.9) into equation (11.8), using (11.6): E cos A M
2
+G
sin A p
2
+ 2F
cos A sin A M p
2
cos A M
+ p2
sin A p
=0
(11.10)
d 2 m = dA
E G 2 2 p M
G p2
sin 2A +
F cos 2A, Mp
F Mp
Ep2 GM 2 FMp
This yields two maximum and two minumum values, which all four are at a distance of 90 from each other1 . These eigenvalues are obtained by writing the equation (11.10) as follows: E F 2 m cos A 2 Mp cos A sin A M sin A = 0; F G 2 m Mp p2
1
91
Chapter 11. Map projections This presupposes that the determinant of the matrix in the middle vanishes: 0 = det H 1 H m2 I = = E m2 M2 G m2 p2 m2 + F2 = M 2 p2 1 M 2 p2 EG F 2 .
E G 2 2 M p
G p2
E M2
G p2
4 M 1 p2 E G F 2 2
These two solutions are called the principal scale factors m1 , m2 . If the H matrix is a diagonal matrix: H= we obtain det H 1 H m2 I = E m2 M2 E G , . M 2 p2 G m2 p2 =0 E 0 0 G ,
m2 1,2 =
m1 =
E G is the meridional scale factor, m2 = is the scale factor in the direction of the M2 p2 parallel. In intermediate directions A (azimuth) the magnication factor is then m= m2 cos2 A + m2 sin2 A. 1 2
is often visualized as an ellipse on the Earth surface (or, correspondingly, in the map plane). The eigenvalues of the matrix are m2 and m2 ; the axes of the visualizing ellipse are m1 in the 1 2 meridional direction and m2 in the direction of the parallel. This ellipse is called the indicatrix of Tissot2 . See subsection 10.3.3. Of the many map projections used, we need to mention especially those that are conformal, i.e., the scale in a point is the same in all directions: m1 = m2 = m,
2
92
dt ds ds 0 dt
The Tissot indicatrix is a circle. With a conformal projection, small circles, squares and local angles and length ratios are mapped true. Surface areas are distorted, however. Conformal projections have the useful property, that the projection formulas in one co-ordinate direction can be derived when the formulas for the other co-ordinate direction are given. A sometimes useful property is, that surface areas are mapped true 3 , even though the shapes of small circles or squares are distorted. This requirement is det H 1 H = m2 m2 = constant. 1 2
E.g. when depicting the surface density of population or some other phenomenon. Johann Heinrich Lambert (1728 1777) Swiss methematician, physicist and astronomer.
93
N N N N
45 o W
o
30 o W
15 o W
0o
75
60
15 E
o
oE
30 E
o 45 E
Figure 11.3.: Example of a Lambert projektion (software used: m_map (http://www2.ocgy. ubc.ca/~rich/map.html))
The images of the meridians in the map are straight lines, which intersect in one point, which is also the common centre of all latitude circle images. Polar co-ordinates in the map plane are x = sin , y = 0 cos , where = 0 describes the latitude circle = 0 . Also = n. (11.11)
Let ds be the distance element on the Earth surface and dt the corresponding element in the map plane; then along the latitude circle ds = p () d dt = nd and by dividing dt n = . ds p Based on conformality this will also be valid along the meridian. Now we have ds = M () d, i.e. dt ds n dt = = M. d ds d p If we reckon positive toward the South, i.e., d = dt, we obtain: d M () = n () . d p () (11.12)
Using equations (11.11, 11.12) we may compute (, ) if given (, ). However, let us rst move to the left hand side: d M M ( ) ln = n = 0 exp n d . d p 0 p ( ) If we dene ()
0
M ( ) d , p ( )
94
90 o E
11.3. On the isometric latitude the so-called isometric latitude, we obtain = 0 exp {n ( () (0 ))} = 0 exp {n ()} . exp {n (0 )} (11.13)
Furthermore we must choose a value n. We do this so, that the scale is stationary at the reference latitude 0 : dm = 0 jos = 0 . d The scale is m i.e., dm d n d n dp 2 = p d p d n n n2 = 2 M + 2 M sin = 2 M (n sin ) , p p p = d dt n = = , ds ds p
(using 11.12 and Appendix B) which ought to vanish for = 0 . It does by choosing n = sin 0 . As an initial condition we must still choose 0 ; it yields for the scale at the reference latitude m (0 ) = n 0 . p (0 )
from which 1 (1 ) follows. Now () is obtained by integrating (11.12) from a starting value, either 0 or 1 , with the integration interval being either [0 , ] or [1 , ]. The inverse operation is easy for ; solving in reverse the dierential equation (11.12) (computing when is given) can be done as follows: 1. invert analytically equation (11.13): () = (0 ) (ln ln 0 ) ; n
M () d, p ()
can be computed numerically (quadrature; the QUAD routines of Matlab). However, for the spherical case a closed solution exists = ln tan + , 4 2 (11.14)
95
Chapter 11. Map projections easily proven by dierentiating using the chain rule. Even for the ellipsoid of revolution a closed solution exists: = ln tan + 4 2 See appendix A. Reverse operation: if is given and to be computed, we may take the equation d M () = d p () and turn it upside down: d p () = . d M () This equation has the general form dy = f (y, t) dt and may be numerically solved using Matlabs ODE routines. An alternative way for the sphere is to analytically invert equation (11.14): = 2 arctan exp . 4 1 + e sin 1 e sin
e 2
(11.15)
This is also useful as the rst iteration step for the ellipsoidal case: (0) = 2 arctan exp , 4 after which (i+1) This converges rapidly. 1 e sin (i) = 2 arctan 1 + e sin (i)
e 2
exp . 4
M ( ) d . p ( )
Here we see the isometric latitude in its most simple glory: in the case of spherical geometry y = ln tan + . 4 2 Mercator is not a lamp projection: There is no projection centre from which the light emanates.
96
y dt ds ds x dt
75 N
60oN
45 N
30 N
15oN
o
60 W 45 W 30 W 15 W
15 E
30 E
45 E
60 E
75 E
90 E
97
Projection plane
0 Equator
/2
( )/2 2
Projection centre
Figure 11.6.: The stereographic projection
0/tan( + )
This case is depicted in gure 11.6. 0 is the distance from the central point (South pole) to the projection plane. In this case the projection is truly a lamp projection. . . but unfortunately only in the case of spherical geometry. Let us derive the scale of the stereographical projection: d d d d M 1 exp { ()} = = 0 exp {} = = 0 . dx d d dx p M p p () The value is negative because x grows to the North and to the South. The co-ordinate x is the metric northing. By using equation 11.15 we obtain 0 d = cot + dx N cos () 4 2 or close to the pole d 0 dx 2 1+e 1e 1 + e sin 1 e sin
e 2 e 2
In case of the Earth where (GRS80) e = 0.08181919104281097693, we obtain for the last term in e 0.99331307907268199009. If we want unity for the scale at the pole, we must set 0 = 2.01346387371353381594.
98
165
oW
165 o E
15 o 0 E
15
o
oW
o 30 N
5 13
12 o 0 W
W
o 45 N
13 5
o
E
0 12
oE
o 60 N
105 o W
oE 105
o 75 N
30 o W
30
15 oW
0
o
oE
o 15 E
Figure 11.7.: An example of the stereographic projection Reference latitude 1 61 70 70 65 Latitude interval 1 20 61 80 61 80 61 80 61 70 Number of support points 20 20 20 39 19 Computation point (, ) (19 .333, 10 .0) (79 .333, 20 .0) (79 .333, 20 .0) (79 .333, 20 .0) (60 .333, 10 .0) Number of terms 12 16 15 15 14
Table 11.1.: Results of Gauss-Kr ger test computations. Number of terms: the number of u terms in the polynomial expansion needed to achieve a change under 1 mm in the computed Gauss-Kr ger co-ordinates u
45 o E
60 o E
75 oE
90 E
90 W
o 75 W
60
oW
45
99
M d .
(11.16)
Furthermore we have on the central meridian: y = 0. Now we have a boundary value problem: the sought for complex map co-ordinate z x + iy is given as a function of the starting projections map co-ordinate w u+iv on the edge y = v = 0 i.e., the real axis. The problem is to determine the function in the whole complex plane. See gure 11.8. Intermezzo. In complex analysis we talk of analytical functions. An analytical function is such a mapping x = f (w) which is dierentiable. Not just once, but innitely many times. In this case, the CauchyRiemann conditions apply: x y y x = , = , u v u v if x = x + iy, w = u + iv. This means that the small vector to the following equation: dx dy = a b b a du dv du dv
T
dx dy du dv
according
=K
x y y x where a = = and b = = , precisely as the Cauchy-Riemann conditions u v u v require. From this it is seen that the mapping of local vectors (du, dv) (dx, dy) is a scaling and rotation; i.e., we have a conformal mapping. Almost all familiar functions are analytical in the complex plane: the exponent, the logarithm, trigonometric and hyperbolic functions, and especially power series expansions. The sum and product of two analytical functions is again analytic. From the Cauchy-Riemann conditions we still derive the Laplace equations: 2x 2x 2y 2y + 2 = 0, + 2 = 0. u2 v u2 v Let us try, as a general solution, a series expansion:
z = a0 + a1 w + a2 w2 + a3 w3 + . . . =
k=0
ak w k .
(11.17)
Here we dene for the sake of generality z (x x0 ) + iy, w (u u0 ) + iv. The values x0 u0
0 0 0
M () d, M () d p ()
100
Figure 11.8.: The Gauss-Kr ger projection as a boundary value problem. The boundary u values at the central meridian are marked with crosses, the directions of integration with arrows
have been chosen to be compatible with a suitable reference latitude 0 . These series expansions are thus meant to be used only in a relatively small area, not the whole projection zone. Thus the number of terms needed also remains smaller. The meridian conditions (11.16) now form a set of observation equations: xi = a0 + a1 ui + a2 u2 + a3 u3 + . . . i i from which the coecients aj can be solved, if a sucient number of support points (ui , xi ) is given (the crosses in gure 11.8). Applying the found coecients aj to equation (11.17) yields the solution for the whole complex plane. This solution may be squeezed into the following, more computationally suitable, form (so-called Clenshaw summation): z = a0 + w (a1 + w (a2 + w (a3 + w ( + wan )))) . The computing sequence is multiplication-addition-multiplication-addition. . . the powers of1 w need not be computed separately. Remember that also the intermediate results are complex numbers! Fortunately complex numbers belong to the popular programming languages either as a xed part (Matlab, Fortran) or a standard library (C++). From the same equation we obtain also easily the scale and the meridian convergence. At said, a complex analytical map is a scaling plus a rotation. When the equation of the map is (11.17), its dierential version is dz z = w, (11.18) dw where dz = a1 + 2a2 w + 3a3 w2 + 4a4 w4 + = kak wk1 . dw
k=1
101
Chapter 11. Map projections Now we may write the complex equation (11.18) in real-valued matrix form dz dz Im Re dw dx dw du = dv , dz dz dy Re Im dw dw where dx = dx dy
T
, dw = Re
du dv dz dw dz dw
x y = = K cos , u w y x = = K sin , u v
from which , the meridian convergence, may be resolved. The choice of the classical Mercator as a starting projection is not the only alternative. Probably the number of terms of the series expansion (11.17) could be reduced by using a Lambert projection for the reference latitude 0 . Whether the saving achievable is worth the trouble, is a so-called Good Question.
Because according to equation 10.12 we have R = 2K, we obtain K= 1 g11 g22 + . 2 )2 2 (x (x1 )2
Let us now take a rectangular co-ordinate system in the map plane (x, y) and transfer its coordinate lines back to the curved surface of the Earth, forming a curvilinear co-ordinate system (, ). At the origin or central meridian in the map plane the metric of this co-ordinate system gij is, at least for ordinary map projections, quasi-Euclidean, in other words, the metric is the unit matrix and stationary at the origin. In this case the theory in the previous chapter applies. In the case of a conformal projection the form of the metric is gij = m2 1 0 0 1 ,
where m is the scale of the map projection, which thus depends on location xi . Derive K= Now m2 = 2 2 + 2 m2 (, ) = 2 = 2 m3 m 2 m3 m = 2m, 1 2 2 g 2 g + 2 2 1 = m2 . 2
102
11.7. Curvature of the Earth surface and scale if we consider the stationarity of m and m 1. So K = m. So, there is a simple relationship between the Gauss curvature radius and the second derivative of the scale (more precisely, the Laplace operator ) As a consequence, there also is a relationship between the scales second derivatives in the North-South and East-West directions: if, e.g., 2m 2m =0 =K 2 2 etc. In the below table we give some examples remember that the area considered is always a neighbourhood of the central point or central meridian or standard parallel, where m 1 and stationary! Projection Mercator Lambert conical Oblique stereographic Gauss-Kr ger, UTM u 2 m/x2 K K K 2 0 2 m/y 2 0 0 K 2 K
In the table we used again, instead of (, ),(x, y), i.e., we substituted 2m 2m 2 x2 which is allowed based on quasi-Euclidicity. The scale of Mercator and Lambert projections is a constant in the direction of the y coordinate, in other words, from map West to map East. In transversal cylindrical projections the scale again is constant in the direction of the x co-ordinate, i.e., the direction of the central meridian. The oblique stereographic projection again is symmetric and the scale behaves in the same way in all compass directions from the central point. Thus, we may use the second derivatives of place of the scale for classifying map projections. E.g., Lambert is most suited for countries extending in the West-East direction (Estonia), whereas Gauss-Kr ger again is best for countries extending in the North-South direction u (New Zealand). The conformal azimuthal projection called oblique stereographic is suitable for square countries (The Netherlands). and 2m 2m , 2 y 2
103
(c) Stereographic
Figure 11.9.: The classication of map projections as vertical, horizontal and square projections
104
x y
+
ED50
61.571 m 95.693 m
yy
with tC observations central epoch, yy = (19)96. The values T96 and Ri,96 are tabulated in (Boucher and Altamimi, ) Tables 3 and 4.
105
Chapter 12. Map projections in Finland For small-scale and topographic maps, the UTM projection is used with a central meridian of 27 (zone 35) for the whole country, producing the ETRS-TM35FIN plane co-ordinate system. This also denes the map sheet division. However, on maps in parts of Finland where another central meridian would be more appropriate (like zone 34, central meridian 21 ), the corresponding co-ordinate grid is also printed on the map, in purple (Anon., 2003).
For large scale maps, such as used for planning and cadastral work, the Gau-Krger u projection continues to be used (but based on the above reference ellipsoid and datum), with a central meridian interval of only one degree: ETRS-GKn, where n designates the central meridian longitude. This avoids the problem of signicant scale distortions.
Most often the co-ordinates in the (1)and (2) datums are close to each other, i.e., are small. In that case we may write the shifts x x(2) x(1) = x + (a1 1) x(1) + a2 y (1) , y y (2) y (1) = y + b1 x(1) + (b2 1) y (1) .
x y
106
The Lappeenranta base network Delaunay triangulation (blue) Voronoi diagram (red)
Figure 12.1.: Lappeenranta densication of the national triangular grid
If we now dene x we obtain the short form x = x + Ax(1) . Also in this generally, if the co-ordinates are close together, the elements of A will be numerically small. Let there be a triangle ABC. Then we have given the shift vectors of the corners xA = x + AxA , xB = x + AxB , xC = x + AxC .
(1) (1) (1)
x y
, A=
a1 1 a2 b1 b2 1
Write this out in components, with x, A on the right hand side: xA = x + a11 xA + a12 yA yA = y + a21 xA + a22 yA
(1) (1) (1) (1) (1)
(1) (1)
(1) (1)
(1)
107
111111111111111 000000000000000C 11111111 00000000 111111111111111 000000000000000 111 000 ( APC ) 00000000 B 111111111111111 000000000000000 p = 000 = 111 ( ABC ) 11111111 11111111 00000000 111111111111111 000000000000000 111111111111111 000000000000000 11111111 00000000 111111111111111 000000000000000 11111111 00000000 111111111111111 111 000000000000000 000 = pA = 111111111111111 000000000000000 11111111 00000000 111111111111111 000000000000000 11111111 00000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 111111111111111 000000000000000( 1111111111111111111111111111111 0000000000000000000000000000000PBC ) P 00000000 11111111 1111111111111111111111111111111 0000000000000000000000000000000 111111111111111 000000000000000 = 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 111111111111111 000000000000000( 1111111111111111111111111111111 0000000000000000000000000000000ABC ) 11111111 00000000 1111111111111111111111111111111 0000000000000000000000000000000 A 1111111111111111111111111111111 0000000000000000000000000000000
1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111 0000 1111111111111111111111111111111 0000000000000000000000000000000 1111 0000 1111111111111111111111111111111 0000000000000000000000000000000 1111 0000 C 1111111111111111111111111111111 0000000000000000000000000000000 1111 p = 0000 = ( ABP ) 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 ( ABC ) 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 B 1111111111111111111111111111111 0000000000000000000000000000000
Figure 12.2.: Computing barycentric co-ordinates as the ratio of the surface areas of two triangles
11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000
or in matrix form xA yA xB yB xC yC = 1 0 1 0 1 0 (1) (1) 0 xA 0 yA 0 x (1) (1) 1 0 xA 0 yA y (1) (1) 0 xB 0 yB 0 a11 (1) (1) 1 0 xB 0 yB a21 (1) (1) a12 0 xC 0 yC 0 a22 (1) (1) 1 0 xC 0 yC ,
from which they can all be solved. Let us write the coordinates (x, y) as follows: x = pA xA + pB xB + pC xC , y = pA yA + pB yB + pC yC, with the further condition pA + pB + pC = 1. Then also x = pA xA + pB xB + pC xC , y = p yA + p yB + p yC.
A B C
(12.1) (12.2)
The set of three numbers pA , pB , pC is called the barycentric co-ordinates of point P See gure 12.2. (BCP ) etc., where is the surface area They can be found as follows (geometrically pA = (ABC) of the triangle) using determinants: xB xC yB yC 1 1 x y 1 xC yC 1 xA x yA y 1 1 xA xB x yA yB y 1 1 1 xA xB xC yA yB yC 1 1 1
pA =
xA xB xC yA yB yC 1 1 1
, pB =
xA xB xC yA yB yC 1 1 1
, pC =
108
M ( ) d . p ( )
The integrand is decomposed into partial fractions: 1 e2 1 e2 sin2 cos = = = = The integral of the rst term is =
0
1 e2 sin2 e2 cos2 = 1 e2 sin2 cos 1 e2 cos = cos 1 e2 sin2 1 e2 cos (1 e sin ) + e2 cos (1 + e sin ) = cos 2 (1 + e sin ) (1 e sin ) e e cos e cos 1 + . cos 2 1 + e sin 1 e sin
1 d = ln tan + . cos 4 2
Proof by using the chain rule: d d = = = d ln tan + d tan + d + 4 2 4 2 4 2 = d d tan + d + 4 2 4 2 1 1 1 = 2 + 2 tan 4 + 2 cos 4 2 1 1 1 = . = cos 2 sin 4 + 2 cos 4 + 2 sin 2 +
This is the full solution in the case that e = 0 (solution for the sphere). In the case of the ellipsoid the second integral f () e cos d = d = ln f () = ln (1 + e sin ) , 1 + e sin f () where we designate f 1 + e sin . In the same way e cos d = ln (1 e sin ) , 1 e sin
109
Appendix A. Isometric latitude on the ellipsoid and the end result is = ln tan + + 4 2 + = ln tan 4 2 e (ln (1 + e sin ) ln (1 e sin )) = 2 e 1 + e sin 2 . 1 e sin
110
,
3/2
1 e2 sin2
111
where ui = u1 , u2 is the parametrization (co-ordinate frame) of a curved surface in space (or more generally a sub-space). Dierentiate: gjk = ui = ui x x = uj uk 2x x 2x x k + ui uj u ui uk uj 2x x j uk ui u 2x x k ui uj u 2x x j ui uk u 2x x k uj ui u
(C.1)
Compute equation (C.1) plus equation (C.2) minus equation (C.3): gjk + j gki k gij = 2 ui u u 2x x ui uj uk x x , ,n u1 u2 . (C.4) as we did in equation
Lets write the second derivatives of x on the local base (9.7), even if in a slightly dierent notation:
2x x ij + ij n, i uj u u which implicitly denes the symbols. Substitution into equation (C.4) yields ij Here we recognise x x u uk or ij g
k
(C.5)
x x u uk
+ ij
x uk
1 2
g + j gki k gij . i jk u u u x uk
=g
ja
= 0,
= =
ij
113
x v i x using (C.5). Here, the v derivative consists of two parts: the interior part, +i v k i , jk j ui u u embedded in the surface, and the exterior part, ij n, aperpendicular to the surface. When the surface S has been given, we can only zero the internal part, i.e. v i + i v k = 0 jk uj describes the parallel transport of the vector v i within the surface. Let us now consider a small rectangle ABCD, side lengths uk and u (see Fig. 10.3), along co-ordinate curves. The sides AB and CD are on opposite sides, the running co-ordinate being uk . Similarly BC and AD are opposite, the running co-ordinate being u . Derive the change in v i over the distance AB : AB v i = In the same way CD v i = +i v m uk . km For the side BC we obtain BC v i = and DA v i = +i m v m u . v i u = i m v m u u v i uk = i v m uk . km uk (D.1)
115
Appendix D. The Riemann tensor from the Christoffel symbols sum together these four terms: ABCD v i = = = i v m km
CD
i v m km
AB
uk
i m v m
DA
i m v m
BC
u =
i v m u uk i m v m uk u = km u uk i m i v m v m km v m + i i m k u uk . km k u u u u
v m v m = m v h , = m v h ; h kh u uk i i lm km ul uk i kj u i j uk
substituting: ABCD v i = = v m + i m i m v h u uk = lm kh km lh v j u uk ,
+ i m m i m kj km j
where we have changed the names of the indices m j (in the rst two terms) ja h j (in the last two terms). Here we see in ready form the Riemann curvature tensor :
i Rj k
i kj ul
i j uk
+ i m m i m , km j kj
aapart from the names of the indices and interchanges of type i = i , just what already was kj jk given in eq. (10.9).
116
Bibliography
Anon. (2003). JHS154. ETRS89 -jrjestelmn liittyvt karttaprojektiot, tasokoordinaatistot a aa a ja karttalehtijako (Map projections, plane co-ordinates and map sheet division in relation to the ETRS89 system). Web site, Finnish National Land Survey. URL: http://www. jhs-suositukset.fi/suomi/jhs154, accessed August 30, 2005. Anon. (2008). JHS153. ETRS89-jjestelm mukaiset koordinaatit Suomessa. Web site, Finnish r n National Land Survey. URL: http://www.jhs-suositukset.fi/suomi/jhs153, accessed Oct. 19, 2010. Bomford, G. (1963). Report of Study Group No. 10. The Geoid in Europe and Connected Countries. In Traveaux de lAssociation Internationale de Godsie, number 22. e e Boucher, C. and Altamimi, Z. Specications for reference frame xing in the analysis of a EUREF GPS campaign. Version 4, 24-10-2008, http://etrs89.ensg.ign.fr/memo-V7. pdf. Boucher, C. and Altamimi, Z. (1995). Specications for Reference Frame Fixing in the Analysis of a EUREF GPS Campaign. In VI Meeting of the TWG in Bern, March 9 and 10, Ver. o der Bayerischen Kommission fr die Internationale Erdmessung, number 56, pages 265267, u Mnchen. u Gross, R. S. (2000). The excitation of the Chandler wobble. Geophys. Res. Lett., 27:23292332. Grossman, W. (1964). Geodtische Rechnungen und Abbildungen. Verlag Konrad Wittwer, a Stuttgart. Heikkinen, M. (1981). Solving the shape of the Earth by using digital density models. Report 81:2, Finnish Geodetic Institute, Helsinki. Heiskanen, W. A. and Moritz, H. (1967). Physical Geodesy. W.H. Freeman and Company, San Francisco. Kumar, M. and Reilly, J. P. (2006). Is denition of WGS 84 correct. http://mycoordinates.org/is-denition-of-wgs-84-correct/. URL:
McCarthy, D. D. (1996). IERS Conventions (1996). (IERS Technical Note ; 21). Technical report, Central Bureau of IERS - Observatoire de Paris. URL: http://www.iers.org/nn 11216/IERS/EN/Publications/TechnicalNotes/tn21.html. Moritz, H. (1992). Geodetic Reference System 1980. In Tscherning, C. C., editor, Geodesists Handbook 1992, pages 187192. International Association of Geodesy, Copenhagen. Ollikainen, M. (1993). GPS-koordinaattien muuntaminen Kartastokoordinaateiksi (Transforming GPS co-ordinates to the National Grid Co-ordinate System). Tiedote 8, Finnish Geodetic Institute, Helsinki. Ollikainen, M., Koivula, H., and Poutanen, M. (2000). The densication of the EUREF network in Finland. Publication 129, Finnish Geodetic Institute, Kirkkonummi. Parm, T. (1988). Kansallisen koordinaattijrjestelmn luominen Suomessa (The creation of a a a national co-ordinate system in Finland). Maanmittaus, (1). Stevenson, P. (2008). The Truth About WGS84. URL: http://www.ashgps.com/out3/General
117