Converting isometric latitude to geographic latitude

A PDF version of this page is available.

If the Earth is modelled as an oblate spheroid with eccentricity e, then the isometric latitude ψ is defined in terms of the geographic latitude φ by

ψ= tanh−1(sinφ) − e tanh−1(e sinφ).
(1)

The problem is to invert (1) and find φ when ψ is given. We do this here by finding sinφ. For other approaches, see e.g. Wolfram MathWorld [1].

If we define s = sinφ then (1) becomes

ψ= tanh−1(s) − e tanh−1(es),
(2)

and we require s given ψ. Three methods are briefly described below.

Method 1: Iteration

This is straightforward. Starting with s0 = tanh ψ, or better s0 = tanh ψ / (1 − e2), we iterate

sn+1= tanh(ψ + e tanh−1(esn))

until the desired accuracy has been achieved.

Method 2: Power series obtained by iteration

Write t = tanh ψ. From (2)

ψ= tanh−1(s) − e2se4s3/3 − e6s5/5 − …

whence

s= tanh(ψ + e2s + e4s3/3 + e6s5/5 + …).

Substituting the first approximation s = t in the RHS gives

s= tanh(ψ + e2t) + O(e4) = (1 + e2)te2t3 + O(e4).

Substituting the second approximation s = (1 + e2)te2t3 gives a third approximation with error O(e6), and so on. Taking the approximation up to terms in e12, with error O(e14), we find

sinφ= F1(e)tF3(e)t3 + F5(e)t5F7(e)t7 + F9(e)t9F11(e)t11 + F13(e)t13 − …

where

F1(e)= 1 + e2 + e4 + e6 + e8 + e10 + e12 + …
F3(e)= e2 + (8/3)e4 + 5e6 + 8e8 + (35/3)e10 + 16e12 + …
F5(e)= (5/3)e4 + (36/5)e6 + (293/15)e8 + (127/3)e10 + 80e12 + …
F7(e)= (16/5)e6 + (6 017/315)e8 + (21 319/315)e10 + (58 111/315)e12 + …
F9(e)= (2 069/315)e8 + (1 751/35)e10 + (619 831/2 835)e12 + …
F11(e)= (883/63)e10 + (2 892 031/22 275)e12 + …
F13(e)= (1 594 444/51 975)e12 + …
 ….

It may be more convenient to rearrange this as a power series in e2. Method 2 thus becomes:

sinφ= t + e2t (1 − t2)[1 + G1(t)e2 + G2(t)e4 + G3(t)e6 + …]

where

G1(t)= 1 − (5/3)t2,
G2(t)= 1 − 4t2 + (16/5)t4,
G3(t)= 1 − 7t2 + (188/15)t4 − (2069/315)t6,
G4(t)= 1 − (32/3)t2 + (95/3)t4 − (11 344/315)t6 + (883/63)t8,
G5(t)= 1 − 15t2 + 65t4 − (37 636/315)t6 + (281 107/2 835)t8 − (1 594 444/51 975)t10,
 ….

2016-10-22 Corrected the following errors.
The coefficient 16/5 in F7 was mistyped as16/15.
The coefficient 188/15 in G3 appeared as 1316/105 (correct value, but not in its lowest terms).

Method 3: Another power series method

If we define ε = e2 / (1 − e2), then up to O(e14) the first two F’s in the preceding method are given by

F1(e)= 1 + ε,      F3(e)= (1 + ε)(3 + 2ε)ε/3.

In fact these expressions are exact, and all the F’s are finite polynomials in ε. To see this, note that (2) gives

d s = (1 − s2)(1 − e2s2) = (1 − s2)(1 + ε − ε s2) .
d t (1 − e2)(1 − t2) (1 − t2)

Consequently, if we set

dns = (1 − s2)(1 + ε − ε s2) c(n; i, j)sitj
d tn (1 − t2)n

then after some routine calculation we find the recurrence

c(n + 1; i, j)= ε(i + 1)c(n; i − 3, j) − (i + 1)(1 + 2ε)c(n; i − 1, j) + (i + 1)(1 + ε)c(n; i + 1, j)
    + (2nj + 1)c(n; i, j − 1) + (j + 1)c(n; i, j + 1),

from which the  dns/d tn can be calculated for n = 1, 2, 3, …. Hence we obtain s as a power series in t,

sinφ= (1 + ε) [ t  −  εP3(ε)t3  +  ε2P5(ε)t5  −  ε3P7(ε)t7  +  ε4P9(ε)t9  − … ] ,
(3)

where

3P3(ε)= 3 + 2ε,
15P5(ε)= 25 + 33ε + 11ε2,
315P7(ε)= 1 008 + 1 985ε + 1 314ε2 + 292ε3,
2 835P9(ε)= 18 621 + 48 726ε + 48 160ε2 + 21 288ε3 + 3 548ε4,
155 925P11(ε)= 2 185 425 + 7 131 667ε + 9 368 049ε2 + 6 187 111ε3 + 2 053 245ε4 + 273 766ε5,
6 081 075P13(ε)= 186 549 948 + 729 283 824ε + 1 194 598 888ε2 + 1 048 861 221ε3 + 520 345 408ε4 + 138 241 602ε5 + 15 360 178ε6,
638 512 875P15(ε)= 43 645 743 540 + 198 807 682 245ε + 390 064 483 656ε2 + 427 111 938 764ε3 + 281 769 302 772ε4 + 111 954 199 526ε5 + 24 798 632 628ε6 + 2 361 774 536ε7,
10 854 718 875P17(ε)= 1 675 357 363 125 + 8 712 734 726 988ε + 19 914 368 526 088ε2 + 26 118 610 367 256ε3 + 21 491 782 658 226ε4 + 11 358 058 960 776ε5 + 3 763 853 754 228ε6 + 714 894 255 024ε7 + 59 574 521 252ε8,
1 856 156 927 625P19(ε)= 653 523 328 781 091 + 3 820 411 433 944 980ε + 9 967 712 402 926 242ε2 + 15 228 851 331 982 346ε3 + 15 010 468 930 152 492ε4 + 9 895 994 198 275 836ε5 + 4 362 794 587 093 944ε6 + 1 240 022 346 019 032ε7 + 206 147 974 037 364ε8 + 15 270 220 299 064ε9,
 ….

It may be more convenient to rearrange this as a power series in ε. Method 3 thus becomes:

sinφ= t + ε t (1 − t2)[1 + Q1(t)ε + Q2(t2 + Q3(t3 + …]

where

Q1(t)= −(5/3) t2,
Q2(t)= −(2/15) t2 (5 − 24t2),
Q3(t)= (1/315) t4 (924 − 2 069t2),
Q4(t)= (1/315) t4 (231 − 3 068t2 + 4 415t4),
Q5(t)= −(2/155 925) t6 (397 4852 266 880t2 + 2 391 666t4),
 ….

Discussion

The diagrams show the error on the ground when one includes terms up to and including e8 and e10 respectively. Method 3 has a lower error than Method 2 up to about 30° latitude, but is much worse around 70°–80°.

Error versus latitude for series up to 8th power of eccentricity Error versus latitude for series up to 10th power of eccentricity

 

For a method that suits all latitudes, it seems best therefore to use either iteration or Method 2. The maximum error on the ground with Method 2 is as follows:

Highest power of e included 6 8 10 12
Maximum error on ground in mm 1.9 0.010 5.9 × 10−5 3.4 × 10−7

Reference

[1] http://mathworld.wolfram.com/IsometricLatitude.html