Alternative formulae for the Transverse Mercator projection

A PDF version of this page is available. This page refers to other pages on this site:
Length of meridian arc on an ellipsoid (also available in a PDF version)
Converting isometric latitude to geographic latitude (also available in a PDF version)

Introduction

These notes describe an attempt to derive formulae for the Transverse Mercator projection that should be (a) somewhat simpler, for the same order of accuracy, than the Redfearn formulae [1, 2] used by the Ordnance Survey of Great Britain, and (b) capable of routine extension (perhaps with the help of computerized algebra) to higher orders of accuracy.

The formulae below, when applied to Great Britain, are similar in accuracy to the OS formulae if carried to the 6th order. If carried to the 12th order they agree with Karney’s GeographicLib [3] to a fraction of nanometre when applied to Great Britain. They seem, however, less accurate than Karney’s method when applied to a wider area.

Only formulae for coordinate conversion are given at present. Formulae for scale and convergence may be added later.

The method used here is not entirely original, as shown by the appendix to Deakin et al. [4]. The idea is to carry out the projection from the ellipsoid to the plane in two stages: (i) from the ellipsoid to an intermediate sphere, (ii) from the sphere to the plane. Part (ii) and its inverse can be done by simple closed formulae containing hyperbolic functions. The hard work lies in part (i).

Conformal mappings

Let S and Σ be surfaces (e.g. plane, surface of sphere). A differentiable mapping from S to Σ is called conformal iff it preserves the angle at which curves intersect, or (equivalently) iff the scale at any point is the same in all directions. Let (x,y) be rectangular coordinates in S such that at every point the scales in x and y are equal, i.e. if σ denotes distance in S then σ/x=σ/y. Let ,η) be similar coordinates in Σ. Then the mapping is conformal iff at every point it satisfies the Cauchy–Riemann conditions:

ξ /x=η/y,ξ /y=η/x.

Assuming that the mapping can be put into the form

ξ=f0(y)+xf1(y)+(x2/2!) f2(y)+(x3/3!) f3(y)+
η=g0(y)+xg1(y)+(x2/2!) g2(y)+(x3/3!) g3(y)+

the Cauchy–Riemann conditions are equivalent to

f1=g0(1),g1=f0(1)
f2=f0(2),g2=g0(2)
f3=g0(3),g3=f0(3)
f4=f0(4),g4=g0(4)

So writing f,g for f0,g0 we see that the mapping is conformal iff it is of the form

ξ=f (y)+xg(1)(y)(x2/2!) f (2)(y)(x3/3!)g(3)(y)+(x4/4!) f (4)(y)+
η=g(y)xf (1)(y)(x2/2!)g(2)(y)+(x3/3!) f (3)(y)+(x4/4!)g(4)(y)
(1)

Transverse Mercator projection

This is considered a suitable projection for a region such as Britain or New Zealand that extends more N–S than E–W. The Earth is taken to be an ellipsoid generated by rotating an ellipse about its smaller axis. A central meridian of longitude is chosen for the region to be mapped (e.g. for Britain the Ordnance Survey has chosen 2° W). The transverse Mercator projection from the ellipsoid to the (x,y) plane is such that

If μ1 and μ2 are two such mappings then μ11μ2 is a conformal mapping of the plane taking (0,y) to (0,my+n), where m and n are constants. In the notation of the preceding section, we must then have f (y)=0 and g(y)=my+n, so that (x,y) is sent to (mx,my+n). Hence the transverse Mercator projection defined above is unique up to a change of scale and a shift parallel to the y-axis.

Notation

Where lower-case and upper-case of the same letter are given, lower-case refers to the ellipsoid and upper-case refers to the intermediate sphere.

Geographical latitude φ at a point P on the ellipsoid is defined to be the angle between the normal at P and the plane of the equator. Isometric latitude ψ is defined in terms of φ by

ψ=tanh1(sinφ)etanh1(esinφ).
(2)

It is straightforward to verify that if σ denotes distance on the ellipsoid then σ/ω=σ/ψ at each point ,ψ). Hence the formulae for conformal mappings (first section) can be applied to the ,ψ) coordinate system.

Intermediate sphere

In the method of this web page, the transverse Mercator projection is split into two stages:

From (2) with e=0 we get Ψ as the inverse Gudermann function of Φ, which can be written in three equivalent ways:

tan=si,sec=co,sin=ta.

Stage (ii) above (transverse Mercator from sphere to plane) is therefore given by simple closed formulae

tanh(x/R)=sechΨ si=cosΦ si
tan(y/R)=sinhΨ se=tanΦ se,
(3)

whose equally simple inverse is given by (14) below.

We now need to work out stage (i), the projection from ellipsoid to sphere.

Outline of method

We are looking for a conformal mapping (1) with a change of notation: ,Ψ) for ,η), and ,ψ) for (x,y). Since the meridian ω=0 is projected to Ω=0, we have f=0, so that (1) becomes

Ω=ωg(1)(ψ)3/3!)g(3)(ψ)+5/5!)g(5)(ψ)
Ψ=g(ψ)2/2!)g(2)(ψ)+4/4!)g(4)(ψ)
(4)

The problem is to find g, and for this it suffices to consider the mapping of the central meridian, on which Ψ=g(ψ). Fix some geographical latitude φm near the centre of the region to be mapped (e.g. φm=55.6° is good for Britain) and let ψm,Φm,Ψm be the corresponding values of ψ,Φ,Ψ. Define δ=ψψm.

To simplify the working and the result, we assume that on the central meridian =0) Ψ is linear to third order in ψ around ψm, and write

Ψ|ω=0=g(ψ)=gm+δ)=Ψm+Δ,

where

Δ=a1δ+a4δ4+a5δ5+a6δ6+.
(5)

This is permissible because, as noted above, the projection has two degrees of freedom (scale and y-shift), so that we can choose it to make the terms in δ2 and δ3 vanish. If for convenience we define a0=Φm, a2=a3=0, then (4) and (5) can be developed as (cf. Deakin et al. [4], page 19)

Ψ+=g(ψ)+(iω)g(1)(ψ)+ (iω)2 g(2)(ψ)+ (iω)3 g(3)(ψ)+
2! 3!
Ψ+ =g+iω)=gm+δ+iω)
Ψ+ = an+iω)n= an n ( n ) δnm(iω)m
n=0 n=0 m=0 m
Ψ+ = (iω)m ( n ) anδnm= (iω)m ( m+r ) am+rδr,
m=0 n=m m m=0 r=0 m
(6)

from which Ψ and Ω can be found by equating real and imaginary parts.

To find the coefficients an we will calculate Φ as a power series in δ in two ways, going via φ and Ψ respectively (see diagram). Equating coefficients of δn will then give the result.

Isometric latitiude on ellipsoid to geograpic latitude on sphere

Details of method

φ from ψ. Recalling that δ=ψψm, we express φ as a Taylor series in δ. We require the first few dnφ/n. Differentiating (2) w.r.t. φ gives

/=(1e2)/(1e2sin2φ)co=1/cosφ(1+εcos2φ)

whence

/ =cosφ(1+εcos2φ)
d2φ/2 =sicosφ(1+εcos2φ)(1+cos2φ)
d3φ/3 =cosφ(1+εcos2φ)[1(212ε)cos2φ(112)cos4φ12cos6φ]
d4φ/4 =sicosφ(1+εcos2φ)[1(639ε)cos2φ(9132)cos4φ(232+103)cos6φ163cos8φ]

This and similar lists below are to be extended, preferably with the aid of computerized algebra, up to the order of accuracy desired.

Φ from φ. If σ denotes distance along the central meridian then /=ρ=const×(1+εcos2φ)3/2. By hypothesis, the scale along the central meridian is constant, so that

Φ=p(1+εcos2φ)3/2+q,
(7)

where p and q are constants. The first few deriatives of Φ w.r.t. φ are therefore

/ =p(1+εcos2φ)3/2
d2Φ/2 =3pεsicosφ(1+εcos2φ)5/2
d3Φ/3 =3pε(1+εcos2φ)7/2[1+(2+4ε)cos2φcos4φ]
d4Φ/4 =3pε(1+εcos2φ)9/2sicosφ[−(4+15ε)+(2+22)cos2φ2cos4φ]
d5Φ/5 =3pε(1+εcos2φ)−11/2[(4+15ε) (8+12+182)cos2φ +(11+362+123)cos4φ (162+133)cos6φ +23cos8φ],

Φ from ψ via φ. Using the above, express Φ as a Taylor series in φφm, and substitute for φφm a Taylor series in δ (=ψψm). Setting c=com,s=sim, we find after a routine calculation

Φ= Φm+rc(1+εc2)1/2δ 2/2!)s +3/3!)(12c2εc4) 4/4!)s(16c2c42c6) +5/5!)[120c2+(2458ε)c4 +(762)c6 +(7223)c8 +23c10] +.
(8)

Φ from ψ via Ψ. The first few derivatives of Φ w.r.t. Ψ, written in terms of Φ, are

/ =co,
d2Φ/2 =sico,
d3Φ/3 =co2cos3Φ,
d4Φ/4 =sinΦ(co6cos3Φ),
d5Φ/5 =co20cos3Φ+24cos5Φ.

So from (5) we get

Φ=Φm +δa1com 2/2!)a12simcom +3/3!)a13(com2cos3Φm)
  +δ4[a4com (1/4!)a14sim(com6cos3Φm)]
  +δ5[a5com a1a4simcom +(1/5!)a15(com20cos3Φm+24cos5Φm)
  +
(9)

On to the result. Equating coefficients of δ, δ2, and δ3 in (8) and (9) gives respectively

a1com=rc(1+εc2)1/2
 
a12simcom=rsc(1+εc2)1/2
 
a13(com2cos3Φm)=rc(12c2εc4)(1+εc2)1/2
 

For a given φm these three equations are to be solved for Φrmm, p, and a1. A straightforward calculation gives

tam=tam(1+εcos2φm)1/2
(10)
p=1+εcos2φm
(11)
a1=(1+εcos4φm)1/2.
 

To find a4, note that

a1com=rc(1+εc2)1/2=c(1+εc2)1/2,a1sim=s

so that equating coefficients of δ4 in (5) and (9) and dividing by a1com gives

−(1/4!)s(16c2c4)=a4/a1(1/4!)s[1+εc46c2(1+εc2)]

whence a4=(1/6)a1sεc4(1+εc2).

Similarly one calculates a5,a6, as far as desired. This is probably not the best method for finding the ai, as it is rather lengthy and does not explain why the factor (1+εc2) occurs in each ai. However, with the help of computerized algebra it yields the following up to 12th order, where k=a1εc4(1+εc2). Note that the factor s=sim appears in ai for even but not odd i.

a4 = ks/6,
a5 = −(k/30)[(5 − 6c2) + εc2(6 − 7c2)],
a6 = (ks/180)[(23 − 39c2) + εc2(66 − 104c2) + ε2c4(48 − 70c2)],
a7 = −(k/1 260)[ (97 − 366c2 + 285c4)
   + εc2 (534 − 1 834c2 + 1 354c4)
   + ε2c4 (912 − 2 826c2 + 1 974c4)
   + ε3c6 (480 − 1 368c2 + 910c4)],
a8 = (ks/10 080)[ (399 − 2 259c2 + 2 340c4)
   + εc2 (3 78618 201c2 + 17 091c4)
   + ε2c4 (11 56847 664c2 + 41 176c4)
   + ε3c6 (13 92050 832c2 + 40 964c4)
   + ε4c8 (5 76019 152c2 + 14 560c4)],
a9 = −(k/90 720)[ (1 61715 822c2 + 35 145c421 420c6)
   + εc2 (25 110197 721c2 + 390 216c4220 761c6)
   + ε2c4 (122 832810 108c2 + 1 456 888c4777 368c6)
   + ε3c6 (254 8801 476 432c2 + 2 465 576c41 253 156c6)
   + ε4c8 (236 1601 240 992c2 + 1 951 360c4951 748c6)
   + ε5c10 (80 640392 832c2 + 587 664c4276 640c6)],
a10 = (ks/907 200)[ (6 51196 003c2 + 284 580c4216 720c6)
   + εc2 (160 3621 741 062c2 + 4 357 044c42 979 576c6)
   + ε2c4 (1 183 53610 218 210c2 + 22 614 831c414 293 933c6)
   + ε3c6 (3 777 12027 694 080c2 + 55 715 568c433 050 400c6)
   + ε4c8 (5 892 48038 260 512c2 + 71 319 600c440 132 128c6)
   + ε5c10 (4 435 20026 210 304c2 + 45 899 424c424 698 800c6)
   + ε6c12 (1 290 2407 070 976c2 + 11 753 280c46 086 080c6)],
a11 = −(k/9 979 200)[ (26 129612 942c2 + 2 973 465c44 769 100c6 + 2 404 080c8)
   + εc2 (1 001 23816 079 440c2 + 64 079 286c491 147 284c6 + 42 371 064c8)
   + ε2c4 (10 751 184133 032 942c2 + 462 342 495c4603 358 772c6 + 264 215 043c8)
   + ε3c6 (49 606 560513 290 016c2 + 1 608 671 808c41 961 068 704c6 + 818 045 920c8)
   + ε4c8 (116 035 2001 054 446 048c2 + 3 047 708 544c43 514 163 808c6 + 1 407 298 464c8)
   + ε5c10 (144 587 5201 190 221 056c2 + 3 223 093 824c43 547 726 976c6 + 1 372 021 728c8)
   + ε6c12 (91 607 040697 432 320c2 + 1 789 882 560c41 894 042 304c6 + 710 673 040c8)
   + ε7c14 (23 224 320165 934 080c2 + 407 062 656c4416 391 360c6 + 152 152 000c8)],
a12 = (ks/119 750 400)[ (104 6873 695 043c2 + 23 932 260c447 995 920c6 + 29 030 400c8)
   + εc2 (6 164 202134 900 843c2 + 679 980 393c41 168 763 544c6 + 636 663 600c8)
   + ε2c4 (94 019 3761 503 338 308c2 + 6 400 822 965c49 876 665 101c6 + 4 988 315 196c8)
   + ε3c6 (603 577 4407 805 457 864c2 + 29 310 906 258c441 588 254 065c6 + 19 773 581 543c8)
   + ε4c8 (1 987 701 12022 054 179 456c2 + 75 126 837 408c499 606 227 856c6 + 45 043 964 496c8)
   + ε5c10 (3 648 718 08036 037 592 064c2 + 113 548 973 184c4142 326 334 672c6 + 61 685 211 984c8)
   + ε6c12 (3 779 112 96034 051 258 368c2 + 100 637 983 296c4120 305 983 424c6 + 50 268 203 216c8)
   + ε7c14 (2 066 964 48017 285 068 800c2 + 48 419 217 792c455 579 694 016c6 + 22 494 311 840c8)
   + ε8c16 (464 486 4003 650 549 760c2 + 9 769 503 744c410 826 175 360c6 + 4 260 256 000c8)].

Radius of intermediate sphere

We chose above that the projection from intermediate sphere to plane should have unit scale on the central meridian. We therefore need to choose R (the radius of the intermediate sphere) so that the projection from ellipsoid to intermediate sphere has the required constant scale F0 on the central meridian. It suffices to do this at the base latitude φm. If σ denotes distance along the central meridian, then at φm we have

dφ/dσ = ρ−1 = a−1(1 + ε)−1/2(1 + εc2)3/2
dΦ/dφ = p(1 + εc2)−3/2 = (1 + εc2)−1/2   from (7) and (11)

whence the scale on the central meridian is Ra−1(1 + ε)−1/2(1 + εc2). Equating this to F0 gives

R = aF0(1 + ε)1/2(1 + εc2)−1.
(12)

Geographical to grid

The method for converting geographical coordinates (φ, ω) on the ellipsoid to grid coordinates (EN) can now be given.

Find the isometric latitude ψ from (2), and set δ = ψ − ψm. Calculate R from (12) and calculate as many of a1, a4, a5, … as desired.

Longitude and isometric latitude on the sphere can now be found from from (6). If we equate real and imaginary parts, and evaluate the binomial coefficients, this gives the following series for Ψ and Ω. Terms containing ai beyond those calculated should be ignored.

Ψ = Ψm + a1δ + a4δ4 + a5δ5 + a6δ6 + a7δ7 + a8δ8 + a9δ9 + a10δ10 + a11δ11 + a12δ12 + …
   − ω2(6a4δ2 + 10a5δ3 + 15a6δ4 + 21a7δ5 + 28a8δ6 + 36a9δ7 + 45a10δ8 + 55a11δ9 + 66a12δ10 + …)
   + ω4(a4 + 5a5δ + 15a6δ2 + 35a7δ3 + 70a8δ4 + 126a9δ5 + 210a10δ6 + 330a11δ7 + 495a12δ8 + …)
   − ω6(a6 + 7a7δ + 28a8δ2 + 84a9δ3 + 210a10δ4 + 462a11δ5 + 924a12δ6 + …)
   + ω8(a8 + 9a9δ + 45a10δ2 + 165a11δ3 + 495a12δ4 + …)
   − ω10(a10 + 11a11δ + 66a12δ2 + …)
   + ω12(a12 + …)
   − …
Ω = ω(a1 + 4a4δ3 + 5a5δ4 + 6a6δ5 + 7a7δ6 + 8a8δ7 + 9a9δ8 + 10a10δ9 + 11a11δ10 + 12a12δ11 + …
   − ω3(4a4δ + 10a5δ2 + 20a6δ3 + 35a7δ4 + 56a8δ5 + 84a9δ6 + 120a10δ7 + 165a11δ8 + 220a12δ9 + …)
   + ω5(a5 + 6a6δ + 21a7δ2 + 56a8δ3 + 126a9δ4 + 252a10δ5 + 462a11δ6 + 792a12δ7 + …)
   − ω7(a7 + 8a8δ + 36a9δ2 + 120a10δ3 + 330a11δ4 + 792a12δ5 + …)
   + ω9(a9 + 10a10δ + 55a11δ2 + 210a12δ3 + …)
   − ω11(a11 + 12a12δ + …)
   + …
(13)

Having found Ψ and Ω, calculate the grid coordinates as

E = R tanh−1(sinΩ/coshΨ) + Eoff ,
N = R tan−1(sinhΨ/cosΩ) + Noff .

It remains to find the constant offsets Eoff and Noff. There is no difficulty with Eoff, which is simply the conventional easting of points on the central meridian (e.g. for Great Britain the OS uses Eoff = 400 000 m).

Noff is implicitly defined by choosing a point on the central meridian, say with latitude φ0, and specifying its conventional northing, say N0. (E.g. for Great Britain the OS chooses φ0 = 49° and N0 = −100000 m.) One could estimate Noff by feeding φ = φ0 into the above procedure, but since φ0 may be outside the region of greatest accuracy the following method is preferable.

Let Nm be the northing of the point on the central meridian with latitude φm, before applying the conventional offset. Let D be the distance along the central meridian measured northwards from φ0 to φm, perhaps calculated by one of the methods for meridian arc suggested on this website. Then we require Noff = F0D + N0 − Nm. The value of Nm can be found from (10), whence

Noff = F0D + N0R tan−1((1 + εc2)−1/2tanφm).

Grid to geographical

The problem here is: Given grid coordinates E and N, find geographical latitude φ and longitude λ. It is a question of reversing the above-described method of finding grid from geographical coordinates.

Having found the constants Eoff and Noff as above, define

x= E − Eoff ,   y= N − Noff .

The inverse of (3) above is given by

tanhΨ = sech(x/R)sin(y/R),
tanΩ = sinh(x/R)sec(y/R),
(14)

from which we get the isometric latitude Ψ and longitude Ω on the intermediate sphere.

The projection from sphere to ellipsoid is analogous to the projection from ellipsoid to sphere. We first need to invert the power series (5), so as to get δ as a power series in Δ = Ψ − Ψm. The absence of terms in δ2 and δ3 simplifies the result, which can be written

δ = b1Δ − b4Δ4b5Δ5b6Δ6 − …,
 

where up to 12th order

b1 = 1 / a1,
b4 = a4 / a15,
b5 = a5 / a16,
b6 = a6 / a17,
b7 = (a1a7 − 4a42) / a19,
b8 = (a1a8 − 9a4a5) / a110,
b9 = (a1a9 − 5a52 − 10a4a6) / a111,
b10 = (a12a10 − 11a1a5a6 − 11a1a4a7 + 22a43) / a113,
b11 = (a12a11 − 6a1a62 − 12a1a5a7 − 12a1a4a8 + 78a42a5) / a114,
b12 = (a12a12 − 13a1a6a7 − 13a1a5a8 − 13a1a4a9 + 91a4a52 + 91a42a6) / a115.

The reasoning that led to (13) above leads to exactly analogous formulae for the ellipsoid coordinates ψ and ω:

ψ = ψm + b1Δ + b4Δ4 + b5Δ5 + b6Δ6 + b7Δ7 + b8Δ8 + b9Δ9 + b10Δ10 + b11Δ11 + b12Δ12 + …
   − Ω2(6b4Δ2 + 10b5Δ3 + 15b6Δ4 + 21b7Δ5 + 28b8Δ6 + 36b9Δ7 + 45b10Δ8 + 55b11Δ9 + 66b12Δ10 + …)
   + Ω4(b4 + 5b5Δ + 15b6Δ2 + 35b7Δ3 + 70b8Δ4 + 126b9Δ5 + 210b10Δ6 + 330b11Δ7 + 495b12Δ8 + …)
   − Ω6(b6 + 7b7Δ + 28b8Δ2 + 84b9Δ3 + 210b10Δ4 + 462b11Δ5 + 924b12Δ6 + …)
   + Ω8(b8 + 9b9Δ + 45b10Δ2 + 165b11Δ3 + 495b12Δ4 + …)
   − Ω10(b10 + 11b11Δ + 66b12Δ2 + …)
   + Ω12(b12 + …)
   − …
ω = Ω(b1 + 4b4Δ3 + 5b5Δ4 + 6b6Δ5 + 7b7Δ6 + 8b8Δ7 + 9b9Δ8 + 10b10Δ9 + 11b11Δ10 + 12b12Δ11 + …
   − Ω3(4b4Δ + 10b5Δ2 + 20b6Δ3 + 35b7Δ4 + 56b8Δ5 + 84b9Δ6 + 120b10Δ7 + 165b11Δ8 + 220b12Δ9 + …)
   + Ω5(b5 + 6b6Δ + 21b7Δ2 + 56b8Δ3 + 126b9Δ4 + 252b10Δ5 + 462b11Δ6 + 792b12Δ7 + …)
   − Ω7(b7 + 8b8Δ + 36b9Δ2 + 120b10Δ3 + 330b11Δ4 + 792b12Δ5 + …)
   + Ω9(b9 + 10b10Δ + 55b11Δ2 + 210b12Δ3 + …)
   − Ω11(b11 + 12b12Δ + …)
   + …

The isometric latitude ψ on the ellipsoid needs to be converted to geographical latitude φ by inverting equation (2). This can be done by iteration or one of the other methods for isometric to geographic latitude suggested on this website.

Finally the longitude λ with respect to Greenwich is given by λ = λ0 + ω.

Michael Behrend, October 2011

 

References

[1] Ordnance Survey (UK), A guide to coordinate systems in Great Britain. Downloaded 2011-09-09 from
http://www.ordnancesurvey.co.uk/oswebsite/gps/docs/A_Guide_to_Coordinate_Systems_in_Great_Britain.pdf
2015-09-24 Now at https://www.ordnancesurvey.co.uk/docs/support/guide-coordinate-systems-great-britain.pdf

[2] https://en.wikipedia.org/wiki/Transverse_Mercator:_Redfearn_series

[3] C.F.F. Karney, GeographicLib. Available from http://geographiclib.sourceforge.net

[4] R.E. Deakin, M.N. Hunter, & C.F.F. Karney, “The Gauss–Krüger projection”. Presented at the Victoria Regional Survey Conference, Warrnambool, 10–12 September 2010. Downloaded 2011-08-30 from
http://user.gs.rmit.edu.au/rod/files/publications/Gauss-Krueger%20Warrnambool%20Conference.pdf
2015-09-24 Now at http://www.mygeodesy.id.au/documents/Gauss-Krueger%20Warrnambool%20Conference.pdf