Karsten Engsager Senior advisor, Danish National Spacecenter, Danish Technical University (ke@space.dtu.dk).

Working with classical geodesy in the areas: Reference systems, Reference frames, Adjustment theory, Mappings, Transformations, Interpolation.

A HIGHLY ACCURATE WORLD WIDE ALGORITHM FOR THE TRANSVERSE MERCATOR MAPPING (ALMOST)
K. ENGSAGER1, K. PODER2 Danish National Space Center, Geodetic Department, Danish Technical University, Juliane Mariesvej 30, DK-2100 Copenhagen E, Denmark. 2 DK-3650 lstykke, Denmark
1

Introduction This presentation is in no way new or genius. It solely presents the ideas and developments from colleagues: Knig und Weise (1951). Map Projections Grafarend and Krumm (2006) appeared shortly before this paper was written, but some comments are included. The single steps in performing the mapping from geographical (geodetic) coordinates to Transverse Mercator coordinates (and reverse) are outlined but not proved. The formulas given are valid at high/low altitudes with only singularity at the equator 90 degrees from the central meridian. It is our hope that this presentation will convince the reader that the Transverse Mercator projection is a useful tool for world wide presentations. The Transverse Mercator mapping is a conformal mapping of the ellipsoid coordinates to a plane. Any mapping function satisfying the Cauchy-Riemann differential formula is conformal. It is therefore nearby to use series of complex functions to describe the Transverse Mercator projection. Some of the complex functions are given as Clenshaw sine summation where the coefficients are elaborated on the third flattening n which gives very efficient series. Our contribution is to extend these series from degree 4 to 7. Using series to degree 4 the accuracy is 0.03 mm up to 4400 km from the central meridian. Using the series to higher degree will increase this limit.

Coordinate designation

p = Geodetic latitude = Geodetic longitude = 2 = Geodetic co - latitude = r + ii = Complex geodetic latitude = 2 r ii = Complex geodetic co - latitude
e 1 + e cos p 2 ln tan p = 2 1 e cos p e 2 tan + 1 e cos p = ln 4 2 1 + e cos p = Isometriclatitude = + i = Complex Isometric coordinates

for the scale furthermore improves the efficiency of the series (b is the minor axis).
-1

a (1 + n ) = (a + b )/2

c
pc

Help functions
hypot( x, y ) = x 2 + y 2 atan 2(r sin p, r cos p ) = p ; in proper quadrant

Clenshaw sine accumulation (simple or complex) is used in calculation of the series:

CS( z ) = A2 sin ( 2 z )
=1
N

c u

y + ix = Complex (normalized) mapping coordinates N + iE = Complex (metric) mapping coordinates

Details may be found in Poder and Engsager (1998).

Derivatives of a mapping function Ellipsoidal definition

a f n = Equatorialradius = Flattening = f (2 f ) ; Third flattening
2 1 3

F ( ) = 1 + 2ncos(2 ) + n 2 M ( ) = a(1 n ) (1 + n ) F ( ) 2 ; Meridian curvature radius a 1 1 Q = 1 + 1 n 2 + 64 n 4 + 256 n 6 4 1+ n ; Meridian arc unit

The derivatives of a mapping function can be found as the complex differential with modulus m and argument g of the mapping function to the complex differential of isometric latitude and longitude. The simplification is possible because both systems are orthogonal and isometric.
du = dy + ix ds = r ( )(d + id ) du ds hypot(y , x ) m= r ( ) g = atan 2(x , y ) (Mapping fundamental form) (Geodetic fundamental form) (Complex scale) (Scale) (Meridian Convergence)

= m exp(ig ) =

The size of some coefficients to the series expansions have been calculated using the derived value of n in the Geodetic Reference System 1980 (GRS80). The meridian arc unit Q is the mean length of one radian of the meridian thus the length of the meridian quadrant is:
G

The Transverse Mercator mapping The Transverse Mercator is basically a generalisation of the meridian to the plane:

The last term being 0.1 10-18 shows the superiority of series expansion in powers of the third flattening n against using 2 given in e 2 = f (2 f ) = 4n(1 + n ) as Grafarend and Krumm (2006) (8.40). Using the mean axis

a ( 2 ) = 2 1 + n (1 +

1 4

1 1 n 2 + 64 n 4 + 256 n6

Point P with geodetic coordinates ( , + c ),

c = longitude of central meridian Genuine origo : (N, E ) = (0,0 ) at = 0 and = 0

ELLIPSOID Complex latitude : c = r + ii SPHERE c = r + i i

dc cos(c ) = 1 + n 2 + 2 cos(2c ) dc (1 n )2 Normalized transv. crd : u = y + ix du M (c ) = dc Q

)
U = Y + iX

Mapping sequence for the Transverse Mercator mapping The mapping for the Transverse Mercator is split up in several conformal mappings during the development of the mapping. This is done to ensure as efficient expansion of series and a simple development strategy. Finally some of the mappings are united again ending up with only three mappings: 1. Ellipsoidal coordinates to the Soldner Sphere 2. Soldner Sphere to Complex Gaussian coordinates 3. Complex Gaussian coordinates to Transverse Mercator coordinates Direct mapping ,: Geodetic coord (1) (2) ,: Soldner Sphere (3) (4) Y,X: Complex Gaussian coord. (5) (6) N,E: Transversal coord. Inverse mapping

Mapping Ellipsoid Soldner Sphere The Soldner Sphere is a pure parameter sphere of unity size. The Geodetic coordinates of the ellipsoid is mapped on the sphere with the simple and virtually unique mapping equations as shown below. Strictly speaking we should account for the Riemann leaves arising from the periodicity of the complex functions.
Mapping equations : = cos( )(d + id ) cos( ) = = =m ; N ( )cos( )(d + id ) N ( )cos( ) Cauchy - Riemann : = g =0

= =1
Function : 1 + e cos( p ) exp( ) = tan ( p 2 ) 1 e cos( p )
e2

= =0
= tan (P 2 )

This may be reformulated into a Clenshaw sine summation:

= + e2 sin (2 )
1 7

Array e[7] = {e2 , e4 , e6 , e8 , e10 , e12 , e14 } e2 = 2n + 2 n 2 + 4 n3 82 n 4 + 32 n5 + 4642 n6 8384 n7 3 3 45 45 4725 4725 e4 = e6 = e8 = e10 = e12 = e14 =
2288 + 5 n 2 16 n3 13 n 4 + 904 n5 1522 n6 1575 n7 3 15 9 315 945 26 44644 15 n3 + 34 n 4 + 8 n5 12686 n 6 14175 n 7 21 5 2835 24832 + 1237 n 4 12 n5 14175 n 6 + 1077964 n 7 630 5 155925

734 n5 + 109598 n 6 + 1040 n 7 315 31185 567

444337 941912 + 155925 n 6 184275 n 7

2405834 n 7 675675

(1) Geodetic coordinates Soldner Sphere

The last coefficient is e14 = -1.3 10 Limiting the series to n4 gives the last coefficient in e8 = 1.6 10-11
= + G2 sin (2 )
1 7

-19

Array G[7] = {G2 , G4 , G6 , G8 , G10 , G12 , G14 } +

26 45

G2 = +2n 2 n 2 2n3 + 116 n 4 3 45 G4 = G6 = G8 = G10 = G12 = G14 = + 7 n 2 8 n3 3 5 +

227 45

n5

2854 675

n6

+ 16822 n 7 4725 31256 n 7 1575

98738 + 14175 n 7 11763988 155925 2046082 31185

n4 + n
4

2704 315

n5 +
5

The figure shows the latitudes and the longitude and the auxiliary parameter t, which in fact is the virtual latitude, while r is the virtual longitude in a Mercator mapping, where the central meridian acts as the virtual equator and the virtual poles are found in the real equator 90 degrees from the central meridian. The singularity at the (true) Poles is handled by using the atan2 function.
Input : , ; on the soldner sphere Y = r = atan 2(sin ( ), cos( )cos( ))

2323 945 73814 2835

n6 n6 n +
6

56 + 15 n3 136 n 4 1262 n5 + 35 105 4279 630 332 35 4174 315

n +

399572 14175

n5 144838 n 6 6237
601676 22275

n7 n
7

n
6

115444544 2027025 38341552 675675

t = atan 2(cos( )sin ( ), hypot(sin ( ), cos( )cos( ))) X = i = ln (tan ( 4 + t 2 ))

n7

U = Y + iX = r + i ; Complex Gaussian Coordinates

(2) Soldner coordinates

Sphere

Geodetic

(3) Soldner Sphere Complex Gaussian coordinates The parameter t (atan2()) is crucial for the accuracy of the entire Transverse Mercator mapping. For geodetic purposes up with series expantion to degree 4 t up to 40 is acceptable giving 0.03 mm in accuracy. For cartographic purposes it is possible to go close 90 and still getting an accuracy of one meter.
U = Y + iX = r + i i t = 2 atan (exp(X )) 2 = atan 2(sin (t ), cos(t ) cos(Y )) = atan 2(sin (Y ) cos(t ), hypot (sin (t ), cos(t ) cos(Y ))) ; Complex Gaussian Coordinates

The last coefficient is G14 = 2.1 10-18 Limiting the series to n4 gives the last coefficient in G8 = 5.4 10-11

Soldner Sphere Complex Gaussian coordinates The Complex Gaussian coordinates gives a transverse mapping, where the central meridian is mapped with unity scale i.e. it behaves as if the central meridian is the virtual equator and the normal to the central meridian plane passing through the centre of the sphere is the virtual rotation axis. A point with the Gaussian coordinates (, ) should be mapped to the Complex Gaussian coordinates c = r + ii.

(4) Complex Gaussian coordinates Soldner Sphere

Complex Gaussian coordinates Transverse Coordinates The Complex Gaussian coordinates are mapped to Transverse Coordinates in two steps via the Normalized Transverse Coordinates. The two Clenshaw summations have been reformulated to one which is presented below. Interested readers may find details in Poder and Engsager (1998). The normalized transversal coordinates are then scaled to Transversal Mercator coordinates.

input : Gaussian complex coordinates : U = Y + iX = r + i i normalized transversal coordinates : u = U + U 2 sin (2U )

1 7

Implementation details All coordinates except the Transversal Mercator coordinates may be considered be angular units in radians. The Meridian arc unit Q is used to transform the Transverse Mercator coordinates into angular units called Normalized Transverse Mercator coordinates. Using the scaled meridian arc unit Qm instead introduce the scale in a simple way: Scale on central meridian : m0 Qm = Q m0 Accuracy check Taking the difference between the input coordinates and the result of the backward_trf(forward_trf(input coordinates)) the accuracy may be checked against the accuracy required for transformation. In the figure below is the limit 0.03 mm shown at series to degree 4 and 5. This limit may be controlled in (3) by fabs ( ) 0.810 and in (4) by fabs ((U )) 0.810 as a very coarse limit (i.e. E 5100 km ). The limit is in fact a function of the latitude and longitude. For degree 5 the limit will be 1.144 (i.e. E 7300 km ). An accuracy limit of 30 m will give a limiting factor of 2.290 (i.e. E 14500 km ). The singularity points may be controlled by the limiting factor above or a coarser one.

output : metric scaled transversal coordinates : N + iE = uQ U Array U [7] = { 2 , U 4 , U 6 , U 8 , U10 , U12 , U14 }
7891 72161 5 41 U 2 = + 1 n 2 n2 + 16 n3 + 180 n4 127 n5 + 37800 n6 + 387072 n7 288 2 3

U4 = U6 = U8 = U10 = U12 = U14 =

3 557 281 1983433 + 13 n2 5 n3 + 1440 n4 + 630 n5 1935360 n6 + 13769 n7 48 28800 61 103 167603 + 240 n3 140 n4 + 15061 n5 + 181440 n6 67102379 n7 26880 29030400 49561 6601661 + 161280 n4 179 n5 + 7257600 n6 + 97445 n7 168 49896 34729 3418889 + 80640 n5 1995840 n6 + 14644087 n7 9123840 212378941 30705481 + 319334400 n6 10378368 n7

+ 1522256789 n7 1383782400

(5) Complex Gaussian coordinates Transversal Mercator coordinates

The last coefficient is U14 = 4.1 10-20 Limiting the series to n4 gives the last coefficient in U8 = 2.4 10-12 NB.: Clenshaw Complex sine summation is used in (5) and (6).
input : metric scaled transversal coordinates : N, E normalized transversal coordinates : u = (N + iE ) / Q output : Gaussian complex coordinates : U = u + u2 sin (2u ) Array u[7] = {u2 , u4 , u6 , u8 , u10 , u12 , u14 }
81 96199 5406467 1 u2 = 1 n + 2 n 2 37 n3 + 360 n 4 + 512 n5 604800 n 6 + 38707200 n 7 2 3 96 1 7

u4 = u6 = u8 = u10 = u12 = u14 =

437 46 51841 1 1 - 48 n 2 15 n3 + 1440 n 4 105 n5 + 1118711 n 6 1209600 n 7 3870720 17 37 - 480 n3 + 840 n 4 + 209 4480 5569 9261899 n5 90720 n 6 58060800 n 7

4397 830251 466511 11 - 161280 n 4 + 504 n5 + 7257600 n 6 294800 n 7 4583 108847 8005831 161280 n5 + 3991680 n 6 + 63866880 n 7 20648693 16363163 638668800 n 6 + 518918400 n 7 219941297 5535129600 n 7

(6) Transversal Mercator coordinates Complex Gaussian coordinates The last coefficient is u14 = -1.5 10-21 Limiting the series to n4 gives the last coefficient in u8 = -2.2 10-13

It has been demonstrated that it is possible in a fairly simple sequence of very efficient mappings to transform the geographical coordinates to Transversal Mercator Coordinates. It has been pointed out that there is a substantial difference in expanding series in powers of the third flattening n against in powers of the eccentricity e2. Due to the relation between the two parameters any series expansion may be reformulated to the other. To illustrate the accuracy the size of some coefficients depending on the third flattening n have been calculated to the derived value of n in the Geodetic Reference System 1980 (GRS80). Implementation details have also been given. The C-code for the Transversal Mercator mapping is free and may be found on the homepage spacecenter.dk under Research/Geodesy/Mapping (the URL is not ready at writing time) or will be forwarded on request to ke@spacecenter.dk

UTM zone 32 coordinates : E coordinate is given +E0 (=500km) Performance The mapping implemented to 4th degree run on a PC with Intel Pentium 4 cpu 2.66 GHz makes more than 224000 transformations pr. second in accuracy checking mode: backward_trf(forward_trf((input coordinates)). An implementation to 5th degree decreases the throughput to 222000. When the mapping is run trough our general transformation manager the throughput decreases by further 10%. More convolutions The formulas presented are valid for the latitude + and for the + + , where longitude 2 2 may be related to the control setup above. When the input latitude goes beyond the limits it is forced to the interval by adding/subtracting 2 to the latitude and the northing N is either increased or decreased by 4Qm . The reverse situation is handled in a similar way. Conclusion

Appendix Scaled meridian arc length from equator to the latitude

7 G ( ) = Qm + p 2 sin (2 ) =1 15 135 9 3 57 p 2 = 3 n + 16 n 3 32 n 5 + 2048 n 7 p 4 = 16 n 2 15 n 4 + 2048 n 6 32 2

p6 = p10 = p14 =

105 35 n 3 + 105 n 5 2048 n 7 p8 = 48 256 693 693 1280 n 5 + 2048 n 7 p12 = 6435 14336 n 7

315 + 512 n 4 189 n 6 512

+ 1001 n 6 2048

The last coefficient is p14 = 1.7 10-20

Limiting the series to n4 gives the last coefficient in p8 = 4.9 10-12 Latitude at the normalized arc length
A = G / Qm

Normalized Transversal coordinates :

u = ( N + iE )Qm1 = y + ix

= latitude of the point (N, E)

(, ) = coordinates on the Soldner sphere of (N, E)

MERIDIAN CONVERGENCE

A = G ( )Qm1 ; Normalized scaled meridian length

( A) = A + q2 sin (2A)
=1
27 32

q2 = + n
3 2

n
3

269 512

n
5

6607 24576

q4 = + n
21 16 2

55 32

n +
4

6759 4096

= atan 2(sin ( y ) tanh ( x ), cos( y )) C2 cos(2u )

=1 4 = atan 2(sin ( )sin ( ), cos( )) C2 cos(2u ) =1

q6 = q10 = q14 =

417 + 151 n3 128 n5 + 87963 n 7 q8 = 96 2048

+ 1097 n 4 15543 n 6 512 2560 +

293393 61440

8011 2560

n5 69119 n 7 q12 = 6144 + 6459601 n 7 860160

n6

= m0 [hypot(cos( y ), sinh ( x ))] cos( A)

LOCAL SCALE

The last coefficient is q14 = 2.8 10-19 Limiting the series to n4 gives the last coefficient in q8 = 1.7 10-11 Meridian convergence and local scale

4 4 exp C2 cos(2u ) cos(2A) =1 =1 = m0 1 cos 2 ( )sin 2 ( )

4 4 exp C2 cos(2u ) cos(2A) =1 =1

4 55 3 3 3 2 C 2 = + 1 n 8 n + 32 n 1152 n 2 3 5 2 C 4 = + 16 n 1 n + 3 2421 772

C6 =

83 + 480 n3 173 n 4 899

C8 =

+ 1531 n 4 336

The last coefficient is C8 = 3.6 10-11 and in the Clenshaw summation is used both cosine and complex cosine.

References Bomford, G (1962): Geodesy, second edition, Oxford Bougayevskiy Lev M, Snyder John P (1995) Map Projections. A reference Manual. Taylor and Francis, London Grarafend E.W., Krumm F.W. (2006): Map Projections. Springer, Berlin/Heidelberg/New York Krger L (1912): Konforme Abbildung des Erdellipsoids in der Ebene. Verffentlichung des Kniglisch Preuszischen Geodtischen Institutes. Neue Folge No. 52, Potsdam Knig R, Weise K.H. (1951): Mathematische Grundlagen der Hheren Geodsie und Kartographie, Erster Band, Springer, Berlin/Gttingen/Heidelberg Poder Knud, Engsager Karsten (1998): Some Conformal Mappings and Transformations for Geodesy and Topographic Cartography. National Survey and Cadastre, Denmark, Publications 4. series vol. 6, Copenhagen

Richardus Peter, Adler Ron K (1972): Map projections, North Holland, Amsterdam Journal of Geodesy (2004): The Geodesists Handbook 2004, Springer