Transforming Cartesian coordinates X,Y,Z

to Geographical coordinates φ, λ, h
George P. Gerdan and Rodney E. Deakin
Department of Land Information
Royal Melbourne Institute of Technology
GPO Box 2476V, MELBOURNE, VIC, 3001
AUSTRALIA

[published in The Australian Surveyor, Vol. 44, No. 1, pp. 55-63, June 1999]

ABSTRACT

In November 1994, the Intergovernmental Committee on Surveying and Mapping (ICSM) adopted a new
geodetic datum for Australia and recommended its progressive implementation nationally. This datum,
designated Geocentric Datum of Australia (GDA) – compatible with the Global Positioning System (GPS), a US
Department of Defense satellite positioning system – will eventually replace the existing Australian Geodetic
Datum (AGD). Consequently, “old” AGD coordinates need to be converted to “new” GDA coordinates, which
may require the transformation of Cartesian coordinates X,Y,Z to geographical coordinates φ , λ , h (latitude,
longitude, height) related to an ellipsoid of revolution.

Given X,Y,Z , longitude is easily derived, but not so latitude, which requires more sophisticated evaluation –
usually by iterative techniques, or complicated direct methods – the height following readily once latitude is
obtained. Thus Cartesian-to-geographic transformations revolve around the determination of latitude; this
paper reviews published techniques, some quite recent, which may be of use to practitioners.

INTRODUCTION

Conversion from geographic to Cartesian coordinates is a simple operation since closed formulae relate X,Y,Z to
φ , λ , h . Inverse computations, on the other hand, are more complex, since no simple relationships link φ to
X,Y,Z. Various techniques for computing φ are well documented in the literature, some more efficient than
others when comparing computational time; indirect (or iterative) techniques are often preferred for their relative
simplicity when compared with more complicated direct methods. This paper compares six techniques; two
direct, Paul (1973) and Ozone (1985) and four indirect, Bowring (1976), Borkowski (1989), Lin & Wang
(1995) and Simple Iteration. The last is described in several geodesy texts, for instance Bomford (1980, p. 679).

These are not the only techniques for computing φ given X,Y,Z; Borkowski (1989) refers to fourteen other
methods; three direct, five iterative, four using closed approximate formulae cast as series expansions and two
others, as well as another direct method of his own (Borkowski 1987). Vincenty (1985) claims that the earliest
solution to this problem was given by Doerrie in his book Kubische und biquadratische Gleichungen,
Muenchen, 1948.

This review is not definitive, since only a small selection of the published techniques are chosen, but amongst
them, Bowring’s method and Simple Iteration have often been used for comparative analysis. For instance,
Borkowski (1989) compares his iterative method against nine others, including three variants of Simple
Iteration, concluding that whilst simple iterative procedures are the easiest to program, his solution is more
appropriate – requiring fewer iterations to reach acceptable accuracy. Laskowski (1991) compares Borkowski’s
and Bowring’s methods, demonstrating that Bowring’s is faster whilst Lin & Wang (1995) find their method to
be faster than Bowring’s. Lin & Wang mention a review of methods by Rapp (1984) who concluded (at the
time) that Bowring’s procedure was the most efficient. Thus from the literature, it could be concluded that Lin
& Wang’s method is the most efficient Cartesian-to-geographic transformation. This is in fact confirmed by
computational analysis of the six techniques, which are regarded as representative of the numerous published
methods.

1
GEODETIC COORDINATE SYSTEMS IN USE IN AUSTRALIA

In 1966, under the direction of the National Mapping Council (NMC) all geodetic surveys in Australia were
recomputed and adjusted on the then new AGD, an astronomically derived topocentric datum having a physical
origin near the centroid of the geodetic network and fixing an ellipsoid of revolution, the Australian National
Spheroid (ANS), with respect to the Earth’s rotational axis. The national adjustment yielded an homogeneous
set of geographical coordinates (latitudes and longitudes) for the geodetic network. At the same time, the NMC
defined a system of rectangular grid coordinates (eastings and northings) known as the Australian Map Grid
(AMG), based on a Universal Transverse Mercator (UTM) projection of AGD latitudes and longitudes.

After 1966 there were several readjustments of the national geodetic network, densified and strengthened by the
inclusion of improved measurements, each readjustment referred to as a Geodetic Model of Australia (GMA).
In 1984 the NMC, recognizing the eventual need for Australia to convert to a geocentric datum, adopted the
latest readjustment at the time, GMA82, as an interim step in this process. This geographical coordinate set was
defined as AGD84 with AMG84 grid coordinates, and to avoid confusion, earlier coordinate sets derived from
the 1966 adjustment were defined as AGD66 and AMG66. Both ADG66 and AGD84 coordinates have a
common datum (defined in 1966) excepting that AGD84 coordinates were derived from an adjustment which
more correctly allowed for the separation between the geoid and the ANS over Australia (NMC 1986).

In 1988, the NMC was superseded by the ICSM, representing the mapping organizations of the States and
Territories of the Commonwealth of Australia and New Zealand. The GDA was adopted by the ICSM in
November 1994 in response to anticipated demand by major users of GPS technology such as the Australian
Defence Force, the International Civil Aviation Organization, the International Hydrographic Organization and
the International Association of Geodesy (Steed 1996, p. 24). The new datum is primarily based on the
coordinates of eight geologically stable sites across Australia with permanent GPS tracking facilities known as
the Australian Fiducial Network (AFN), supplemented by a network of seventy survey stations (covering
Australia at approximately 500km intervals) which together form the Australian National Network (ANN).
Geocentric Cartesian coordinates of these stations were derived from an adjustment of precise GPS observations
obtained from – (i) a two week global observation period in 1992 conducted by the International GPS
Geodynamics Service at approximately two hundred sites around the world (including all the AFN sites) and (ii)
ICSM campaigns in 1992, ’93 and ’94 linking all AFN and ANN sites. These coordinates are related to the
International Earth Rotation Service (IERS) Terrestrial Reference Frame for 1992 (ITRF92) at epoch 1994.0
[The epoch 1994.0 (1st Jan. 1994) reflects the fact that monitoring stations used by IERS are moving with
respect to each other due to earth crustal motion; the epoch date indicating the datum is ITRF92 adjusted for
station motion in the intervening period]. The ICSM has defined GDA94 coordinates as latitudes and longitudes
related to the ellipsoid of the Geodetic Reference System 1980 (GRS80) [BG 1988, p. 348] and Map Grid
Australia 1994 (MGA94) grid coordinates as a UTM projection of those latitudes and longitudes.

The GRS80 was adopted by the International Union of Geodesy and Geophysics (IUGG) at its XVIITH General
Assembly in Canberra, December 1979, as the best representation of the size and shape of the Earth. The
GRS80 is based on the theory of a geocentric equipotential ellipsoid and is defined by four physical parameters
of the Earth – (i) a the equatorial radius, (ii) GM the geocentric gravitational constant, (iii) J2 the dynamical
form factor, and (iv) ω the angular velocity – from which geometric constants, e 2 (square of the first
eccentricity) and f (flattening) are derived. The World Geodetic System 1984 (WGS84), the datum for GPS, is
also based on GRS80, except that the dynamical form factor of the Earth is expressed in a modified form,
causing very small differences between derived constants of the WGS84 and GRS80 ellipsoids. These
differences can be regarded as negligible for all practical purposes and GDA94 is considered compatible with
WGS84.

Geometric parameters of relevant ellipsoids are:

Ellipsoid semi-major axis a flattening f semi-minor axis b = a (1 − f )
ANS 6378160 m 1/298.25 6356774.7192 m
GRS80 6378137 m 1/298.257222101 6356752.3141 m
WGS84 6378137 m 1/298.257223563 6356752.3142 m

2
THE GENERAL METHOD OF CONVERSION BETWEEN COORDINATE DATUMS

X,Y,Z Cartesian coordinates have an origin at the centre of the ellipsoid. The Z-axis is in the direction of the
rotational axis of the ellipsoid of revolution, the X-Z plane is the Greenwich meridian plane (the origin of
longitudes), the X-Y plane is the equatorial plane of the ellipsoid (the origin of latitudes), the X-axis is in the
direction of the intersection of the Greenwich meridian plane and the equatorial plane and the Y-axis is advanced
90° east along the equator. In Australia, there are two reference ellipsoids to be considered, the ANS (for
AGD66 and AGD84) and the GRS80 (for GDA94) and three derived sets of Cartesian coordinates X , Y , Z AGD66 ,
X , Y , Z AGD84 and X , Y , Z GDA94 , the last properly termed geocentric. It is known that the origin of the ANS is not
coincident with the geocentre (the origin of the GRS80 ellipsoid) and also that the Cartesian axes of the ANS are
slightly rotated with respect to the axes of GRS80 ellipsoid. A generally accepted method of conversion consists
of the following steps (Featherstone 1994, pp. 8-10)

1. Compute Cartesian coordinates from geographical coordinates using

X = (ν + h) cos φ cos λ (1.1)
Y = (ν + h) cos φ sin λ (1.2)
X = (ν (1 − e 2 ) + h)sin φ (1.3)
where ν (nu), the radius of curvature in the prime vertical plane, and e 2 are given by
a
ν = (1.4)
1 − e sin 2 φ
2

e 2 = f (2 − f ) (1.5)

2. Convert Cartesian coordinates from AGD to GDA94 via a 7-parameter conformal transformation given
in matrix notation as

LM X OP LM 1 RZ − RY OP LM X OP LM
XT OP
MMY PP MM
= (1 + ds ) − RZ 1 RX PP MMY PP MM
+ YT PP (1.6)
NZ Q GDA 94 NRY − RX 1 Q NZ Q AGD N
ZT Q
where ds is a scale change, RX , RY , RZ are small rotations about the X,Y,Z axes respectively and
XT , YT , ZT are translations between the two ellipsoid origins.

3. Convert Cartesian coordinates to geographical coordinates using

Y
tan λ = (1.7)
X
Z + ν e 2 sin φ
tan φ = (1.8)
p
p
h = −ν (1.9)
cos φ
where p is the perpendicular distance from the rotational axis
p = X2 + Y 2 (1.10)

Note 1: If AMG66 or AMG84 coordinates are to be transformed to MGA94 coordinates then they must be
converted from grid to geographic prior to step 1 and from geographic to grid after step 3 using
Redfearn’s formulae (NMC 1986).
Note 2: In (1.1) to (1.3), the ellipsoidal height h, the height above the ellipsoid measured along the normal, is
required. In Australia, all heights are related to the Australian Height Datum (AHD), a practical
approximation of the geoid over Australia, and the change to GDA94 will not affect AHD heights.
Cartesian coordinates for P on the Earth’s surface must be calculated using h (obtained by adding the
geoid-ellipsoid separation N to the AHD value). Geoid models, such as AUSGEOID93 (Steed &
Holtznagel 1994) can be used to determine N. [AUSGEOID93 is a grid of computed N values,
produced by the Australian Surveying and Land Information Group (AUSLIG), which can be
interpolated to give values at desired locations].

3
Note 3: Sets of transformation parameters (ds, RX , RY , RZ and XT , YT , ZT ) for AGD66 and AGD84 are
available from AUSLIG.

The Cartesian-to-geographic conversion problem is embodied in (1.8) where functions of φ ( tan φ , ν and sin φ )
appear on both sides of the equation making the evaluation of φ difficult. The following techniques, direct and
iterative, are solutions to this problem.

Z
N

λ Q

P ellipsoid
ian

•
erid

PE •

h
Greenwich M

al
rm
no

O a Y
•
φ
a •
E
H C D •

R
X M
G Equator

Figure 1 Section of ellipsoid

Z
auxiliary circle
S
N PC ·P
·· P
h

Q E
l
ma
ν
a
n or

θ
O ψ φ R T M

·C D
H
·
Figure 2 Meridian ellipse

4
ELLIPSOID RELATIONSHIPS

Referring to Figures 1 and 2, the following relationships will be of use in the derivations that follow.

(i) OE = OG = a and ON = b are the semi-major and semi-minor axes of the ellipsoid
(ii) PH is the ellipsoidal normal and P PE = h. PE is the projection of P on the ellipsoid and PC is the
projection of PE on the auxiliary circle of the meridian ellipse
(iii) H PE = ν and C PE = ρ are the radii of curvature in the prime vertical and meridian planes of P
respectively
(iv) OH = ν e 2 sin φ and DH = ν e 2
(v) latitudes: PDM = φ (geodetic), PC O M = ψ (parametric), PE O M = θ (geocentric)
b b b2
(vi) latitudes are related by: tan ψ = tan φ and tan θ = tan ψ = 2 tan φ
a a a
(vii) PS = p = X + Y and PT = Z
2 2

(viii) the ratio PC R PE R = a b and PE Q = pE = a cos ψ , PE R = Z E = b sin ψ

(ix) perpendicular distances from C (the centre of curvature of the meridian ellipse) to the Z- and OM-axes
respectively are:
F a2 − b2 I F
a2 − b2 I
GH a 2 JK
a cos3 ψ = e 2 a cos3 ψ and − GH
b2
JK
b sin 3 ψ = − e ′ 2 b sin 3 ψ

The following relations between geometric constants of the ellipsoid are also of use.

a−b
f = flattening (2.1)
a
b = a (1 − f ) semi-minor axis (2.2)
a2 − b2
e2 = = f (2 − f ) first eccentricity squared (2.3)
a2
a2 − b2 e2
e′ 2 = = second eccentricity squared (2.4)
b 2
1 − e2

SIMPLE ITERATION

This technique, described in various forms in geodesy texts, owes its popularity to its programming simplicity.
The basis for the method is (1.8)
Z + ν e 2 sin φ
tan φ = (1.8)
p
An approximate value φ 0 is used in the right hand side (RHS) of (1.8) to evaluate tan φ 1 , (and hence φ 1 ) on the
left hand side (LHS). This “new” value, φ 1 is then used in the RHS to give the next value, tan φ 2 . This
procedure is repeated until the difference between successive LHS values, tan φ i , tan φ i +1, reaches an acceptable
limit; the iteration thus converges to a solution of tan φ .

Z
The starting value φ 0 is obtained from tan θ ′ = , θ ′ being a good approximation of the geocentric latitude θ .
p
a2 a2
Hence, since tan φ = tan θ ≈ tan θ ′ , using (2.4) gives
b2 b2
Z (1 + e ′ )
2
tan φ 0 ≈ (3.1)
p

After evaluating φ the height h is calculated from (1.9).

5
PAUL’S METHOD

This method (Paul 1973) is direct in so far as tan φ is obtained from a simple closed equation, but only after
several intermediate variables have been evaluated. An outline of the development of the necessary equations is
given below.

From the ellipsoid relationships

p tan φ − Z = ν e 2 sin φ (4.1)
After manipulations involving the equation for υ , a quartic equation in tan φ is obtained
2 p3 Z p2 Z 2
p 4 tan 4 φ − 2 p3 Z tan 3 φ + ( Z 2 + β ) p 2 tan 2 φ − 2 tan φ + = 0 (4.2)
1− e 1 − e2
p2 − a 2 e 4
with β = (4.3)
1 − e2
Now, in common with all direct methods, (4.2) is solved by successive substitutions which eventually yield a
cubic equation in a subsidiary variable, for which standard solutions exist (Spiegel 1968, p. 32). Making the
substitution
Z
p tan φ = ζ + (4.4)
2
expressions for p 2 tan 2 φ , p3 tan 3 φ and p 4 tan 4 φ are substituted into (4.2) giving
F Z2 2 I β Z2 F I
GH
ζ4 + β −
2 JK
ζ − α Zζ + Z2 +
4 16 GH JK = 0 (4.5)

p2 + a2e 4
where α = (4.6)
1 − e2
Assuming a solution of (4.5) is ζ = t1 + t2 + t3 , t1 , t2 , t3 being the roots of the cubic equation of the form
t + a1 t + a2 t + a3 = 0 , expressions for ζ 2 and ζ 4 are substituted into (4.5), giving
3 2

ζ 4 − 2 (t1 + t2 + t3 ) ζ 2 − 8 t1t2 t3 ζ + t1 + t2 + t3 {b g 2
b
− 4 t1t2 + t1t3 + t2 t3 g} = 0 (4.7)
Comparing coefficients of (4.5) and (4.7) gives expressions for t1 + t2 + t3 , t1 t2 + t1 t3 + t2 t3 and t1 t2 t3 (the
coefficients − a1, a2 and − a3 ) respectively. Using the sums and products, and manipulating the expression for ζ
gives a solution in terms of the single real root t1
Z2 β αZ
ζ = t1 + − − t1 + (4.8)
4 2 4 t1
Substituting expressions for a1, a2 and a3 into the general cubic equation gives
Fβ + Z It + Fβ − βZ It − α Z
2 2 2 2 2
GH 2 4 JK GH 16 8 JK 64
t3 + 2
= 0 (4.9)

A further substitution

t = G
Fβ+Z Iρ + Z − β 2 2

H 6 JK 12 6 (4.10)

enables (4.9) to be reduced to a standard form

4 ρ 3 − 3ρ = q (4.11)
with a single real root ρ1 having the solution
1 RSF q + IK 1
FH IK UV− 13

TH
3
ρ1 = q2 − 1 + q + q2 − 1 (4.12)
2 W
27 Z 2 (α 2 − β 2 )
where q = 1+ (4.13)
2 ( β + Z 2 )3

Thus, having X,Y,Z (hence p) for a point related to an ellipsoid, tan φ is obtained by computing the following
variables in order: α from (4.6), β from (4.3), q from (4.13), ρ1 from (4.12), t1 from (4.10), ζ from (4.8)
noting that all square-roots in (4.8) have the same sign as Z, and finally tan φ from (4.4). [ ρ1 and t1 are the real
roots of (4.11) and (4.9) respectively]

6
In common with many direct solutions, the method fails when Z approaches zero, in which case q and ρ1 both
approach unity, t1 becomes infinitely small and ζ is indeterminate. In such cases, Paul (1973, p. 136) gives an
approximate equation for the calculation of tan φ .

OZONE’S METHOD

This method (Ozone 1985) is direct in so far as tan φ is obtained from a simple closed equation, but only after
several intermediate variables have been evaluated. An outline of the development of the necessary equations is
given below.

From the ellipsoid relationships, where subscripts E refer to points on the ellipsoid
Z − ZE a
tan φ = = tan ψ (5.1)
p − pE b
which, with pE = a cos ψ and Z E = b sin ψ can be rearranged to give
ap bZ
− = a2 − b2 (5.2)
cos ψ sin ψ
ψ 1
Making the substitution tan = and using trigonometric identities leads to
2 u
u2 − 1 2u 2u
cos ψ = 2 , sin ψ = 2 , tan ψ = 2 (5.3)
u +1 u +1 u −1
and substituting expressions for cosψ and sin ψ into (5.2) gives
ap (u 2 + 1) b Z (u 2 + 1)
− = a2 − b2
u2 − 1 2u
which can be simplified to give a quartic equation in u
u 4 − 4 M u3 − 4 N u − 1 = 0 (5.4)
a p − (a 2 − b 2 )
where M = (5.5)
2bZ
a p + (a2 − b 2 )
N = (5.6)
2bZ
Introducing variables I, J, K enables (5.4) to be expressed in two forms
(i) the product of two quadratic equations
u 2 − ( 2 M − J ) u + ( I + J ) u 2 − (2 M + J ) u + ( I − J ) = 0 (5.7)
(ii) the difference of two squares
du 2
− 2Mu + I i − bJ u + Kg
2 2
= 0 (5.8)
Both (5.7) and (5.8) give
u 4 − 4 M u 3 + (2 I − J 2 + 4 M 2 ) u 2 − ( 2 K J + 4 M I ) u + ( I 2 − K 2 ) = 0 (5.9)
Comparing coefficients of u and u 2 , and the constant terms of (5.4) and (5.9) gives
J 2 = 2 I + 4 M 2 , K 2 = I 2 + 1 , J K = 2 ( N − MI ) (5.10)
which can be rearranged to give a cubic equation in I
I3 + V I − W = 0 (5.11)
for which standard solutions exist for the real root of I (Spiegel 1968, p. 32)
F V W I 1
F V W I 1

I = G F I +F I − G F I +F I
3 3
3 2 3 2

GH H 3 K H 2 K +
W
JJ GH H 3 K H 2 K −
W
JJ (5.12)
2 K 2 K
where V = 4NM + 1 (5.13)
W = 2 (N 2 − M2 ) (5.14)
After solving for I, J and K follow from (5.10)
J = 2I + 4M 2 (5.15)
2 ( N − MI )
K = (5.16)
J

7
The parameter u can now be obtained from the real solution of (5.7), a quadratic equation in u for which the
positive root is
2M + J + G
u = (5.17)
2
where G = (2 M + J )2 − 4 ( I − K) (5.18)
a 2u
Now since tan φ = tan ψ and from (5.3), where tan ψ = 2 then
b u −1
2 au
tan φ = (5.19)
b (u 2 − 1)

Thus, having X,Y,Z (hence p) for a point related to an ellipsoid, tan φ is obtained by computing the following
variables in order: M and N from (5.5) and (5.6), noting that Z is taken as positive, V and W from (5.13) and
(5.14), I from (5.12), J and K from (5.15) and (5.16), G from (5.18), then u from (5.17) and finally tan φ
from (5.19).
After calculating φ , its sign is determined by inspection of Z.

BOWRING’S METHOD

This method (Bowring 1976) is iterative, but owing to the nature of the equation used can be regarded as exact,
since for all practical purposes, second or third iterations are not required (Bowring 1976, p.326).

From the ellipsoid relationships, perpendicular distances from C (the centre of curvature of the meridian ellipse)
to the Z- and OM-axes are e 2 a cos3 ψ and − e ′ 2 b sin 3 ψ respectively, thus an expression for tan φ in terms of
the parametric latitude ψ is
Z + b e ′ 2 sin 3 ψ
tan φ = (6.1)
p − a e 2 cos3 ψ
This equation, having functions of φ on both sides of the equation, clearly requires an iterative solution for
Z
tan φ . An initial value ψ 0 is obtained from tan θ ′ = , θ ′ being a good approximation of the geocentric
p
a a
latitude θ , and since tan ψ = tan θ ≈ tan θ ′ , then
b b
aZ
tan ψ 0 ≈ (6.2)
bp

Bowring (1976) shows that for all Earth-bound points ( −5,000 m ≤ h ≤ 10,000 m ) the maximum error in φ ,
induced by using only a single iteration, is 0.000 000 030 ′′ . Thus for all practical purposes, the evaluation of
tan φ by (6.1), with a first approximation of ψ from (6.2), can be regarded as exact.

BORKOWSKI’S METHOD

This is an indirect method using Newton’s iterative technique to solve for the parametric latitude ψ (Borkowski
1989). An outline of the development of the necessary equations is given below.

From the ellipsoid relationships, the rectangular components of h in the meridian plane are given by
h sin φ = Z − b sin ψ (7.1)
h cos φ = p − a cos ψ (7.2)
Eliminating h by division and substituting for tan φ gives
a Z − b sin ψ
tan ψ = (7.3)
b p − a cos ψ
which can be rearranged as
2 ap sin ψ − 2 bZ cos ψ − 2 sin ψ cos ψ ( a 2 − b 2 ) = 0 (7.4)
Replacing ap and bZ with q cos Ω and q sin Ω respectively, gives
2 q ( cos Ω sin ψ − sin Ω cos ψ ) − 2 sin ψ cos ψ ( a 2 − b 2 ) = 0 (7.5)

8
 bZ
where tan Ω = (7.6)
ap
q = ( ap)2 + (bZ )2 (7.7)
Simplifying (7.5) using trigonometric identities gives a function of ψ as
f (ψ ) = 2 sin (ψ − Ω) − c sin 2ψ = 0 (7.8)
a2 − b2 a2 − b2
where c = = (7.9)
q ( ap)2 + (bZ )2
and ψ can be computed by Newton’s iterative technique
f (ψ n )
ψ n +1 = ψ n − (7.10)
f ′(ψ n )
where the function and its derivative are
f (ψ n ) = 2 sin (ψ n − Ω) − c sin 2ψ n (7.11)
l
f ′ (ψ n ) = 2 cos(ψ n − Ω) − c cos 2ψ n q (7.12)

An initial approximation ψ 0 can be obtained from

aZ
tan ψ 0 ≈ (7.13)
bp
After solving for ψ , tan φ is obtained from
a
tan φ = tan ψ (7.14)
b

Thus, having X,Y,Z (hence p) for a point related to an ellipsoid, ψ is obtained by iteration after first calculating
constants Ω , q and c from (7.6), (7.7) and (7.9) respectively, then using (7.10), with (7.11) and (7.12) calculated
using successively improved values of ψ .

The standard method of computing h once φ is known is via (1.9); which unfortunately fails at the poles
(division by zero). Borkowski (1989, p. 51, eq. 6) gives an alternative, obtained by multiplying (7.1) and (7.2)
by sin φ and cos φ respectively and adding the resultants to give
b g b g
h = Z − b sin ψ sin φ + p − a cos ψ cos φ (7.15)
This formula, whilst more robust, has more trigonometric functions to evaluate, and thus is demanding of
computer resources; (1.9) is more efficient in non-polar regions.

LIN AND WANG’S METHOD

This elegant method (Lin & Wang 1995) uses Newton’s iterative technique to evaluate a scalar multiplier m of
the unit vector n , normal to the ellipsoid. Once obtained, simple relationships between Cartesian coordinates of
P and its projection on the ellipsoid PE are used to evaluate X E , YE , Z E giving the geocentric latitude θ directly
and thus φ . An outline of the development of the necessary equations is given below.

The Cartesian equation of the ellipsoid of revolution is

X E2 YE2 Z E2
2 + 2 + = 1 (8.1)
a a b2
which, when partially differentiated, gives the normal unit vector n
2 XE 2Y 2Z
n = i + 2E j + 2E k (8.2)
a2 a b
The vector equation of h is
h = ( X − X E ) i + (Y − YE ) j + ( Z − Z E ) k (8.3)
h can also be given by
h = mn
where m is a scalar multiplier of the normal unit vector, hence h is also given by
2m XE 2 m YE 2m ZE
h = i+ j+ k (8.4)
a2 a2 b2
Equating the vector components of (8.3) and (8.4) and rearranging, gives

9
 XE X Y Y ZE Z
= , E = , = (8.5)
a 2 m a 2 m b 2m
a+ a+ b+
a a b
which, when squared and substituted into (8.1), and simplified, gives a function of m as
p2 Z2
f ( m) = 2 + −1 = 0
F 2m I F
2m 2 I (8.6)

Ha+
a K b+
b H K
and m can be obtained by Newton’s iterative technique
f ( mn )
mn +1 = mn − (8.7)
f ′( mn )
where the function and its derivative are
p2 Z2
f ( mn ) = 2 + −1
F 2 mn I 2 mn 2 F I (8.8)
a+
H a K b+
b H K
R| U|
f ′( m ) = − 4 S
| p 2
+
Z |V2

|| a FH a + 2m IK bFb +
2m I
(8.9)

H b K ||W
n 3 3
n n
T a

An initial approximation m0 can be obtained from (Lin & Wang 1995, p. 301)
d i d i
3
ab a 2 Z 2 + b 2 p 2 − a 2 b 2 a 2 Z 2 + b 2 p2
2

m0 = (8.10)
d
2 a Z +b p
4 2 4 2
i
After calculating m, pE and Z E are obtained from (8.5) as
p
pE = (8.11)
2m
1+ 2
a
Z
ZE = (8.12)
2m
1+ 2
b
and the latitude φ and height h obtained from
a2 Z
tan φ = 2 E (8.13)
b pE
h = ± b p − p g + bZ − Z g
E
2
E
2
(8.14)
Note: h is negative if ( p + Z ) is less than ( pE + Z E )

Thus, having X,Y,Z (hence p) for a point related to an ellipsoid, m is obtained by iteration after first calculating a
starting value m0 from (8.10) then using (8.7), with (8.8) and (8.9) calculated using successively improved
values of m. Having calculated m the coordinates of PE on the ellipsoid are obtained and φ follows from (8.13).
The attraction of this method, compared to the other iterative techniques (Simple Iteration, Bowring’s and
Borkowski’s) is that no trigonometric functions are used in the calculation of tan φ or h.

COMPUTER TESTING OF THE TRANSFORMATION METHODS

The preference for one transformation method over another depends on the answer to three questions relevant to
computer evaluation – (i) is it easy to code as a computer algorithm? (ii) is it accurate? and (iii) is it fast? In
answer to the first two questions; all the methods are simple to code as computer algorithms, requiring similar
numbers of lines of source code and each yielding accurate results, providing that Paul’s and Ozone’s direct
methods are not used when Z = 0 or very small, in which cases they either fail or give unreliable results. The
answer to the last question, often the determining factor in the adoption of a particular method, depends on two
factors, (a) the number square roots, cube roots and trigonometric functions requiring evaluation and (b) the

10
number of iterations required for the indirect methods. With regard to these factors, Lin & Wang’s method
requires only a single evaluation of a square root to compute tan φ and, if points are restricted to an Earth-bound
region ( −5,000 m ≤ h ≤ 10,000 m ) , the analysis shows that Simple Iteration is the only indirect method actually
requiring multiple iterations to reach an acceptable accuracy for φ , the other indirect methods (Bowring’s,
Borkowski’s and Lin & Wang’s) giving, for all practical purposes, “exact” solutions for tan φ without iteration.

In order to evaluate accuracy and relative speeds, each method was coded as a computer algorithm (using
Borland’s C++ version 5.0A compiler on a PC with a 133MHz Pentium processor) and tested over a grid of
points at 0.1° intervals between latitudes 5°S and 50°S and longitudes 110°E and 160°E on the GRS80 ellipsoid,
each assigned an ellipsoidal height h = 10,000 m (225,450 points in total, covering the Australian continent,
Tasmania and offshore islands). For each algorithm, the following procedure was adopted

• compute X,Y,Z coordinates for each point ( φ , λ , h ) using (1.1), (1.2) and (1.3) [these Cartesian and
geographic values regarded as “exact”] then, using the X,Y,Z coordinates
• compute the longitude λ C using (1.7)
• compute the latitude φ C of each point using the relevant equations for each method
• compute the height hC using (1.9) for each method, excepting Lin and Wang’s, where (8.14) was used
• compare φ C and hC with the exact values to determine the magnitude of the “geographical errors”
• compute X,Y,Z again, using (1.1), (1.2) and (1.3) with computed values φ C , λ C , hC and compare with exact
values to determine the magnitude of the “Cartesian errors”.

Note: Simple Iteration was the only indirect method where iteration was used, the others, Bowring’s,
Borkowski’s and Lin & Wang’s, were used without iteration.

For each algorithm, the time required to perform the Cartesian-to-geographic transformations for all points was
recorded (in processor “clock-ticks”) and converted to relative speed units, with Bowring’s method given a
value of 50. Results of the comparative tests are given in Table 1.

Conversion Maximum Errors (magnitudes) Relative

Method φ (sec) h ( m) X ( m) Y ( m) Z ( m) Speed
Lin & Wang 6.87e-11 2.53e-09 9.31e-10 9.31e-10 1.86e-09 42
Bowring 2.88e-08 1.01e-06 9.31e-10 9.31e-10 1.32e-06 50
Paul 3.90e-07 1.14e-06 9.31e-10 9.31e-10 1.21e-05 52
Ozone 6.87e-11 2.42e-09 9.31e-10 9.31e-10 2.33e-09 56
Simple Iteration 1.35e-05 6.12e-05 9.31e-10 9.31e-10 4.19e-04 62
Borkowski 2.88e-08 1.01e-06 9.31e-10 9.31e-10 1.32e-06 63

Table 1
Cartesian-to-geographic conversion algorithms ranked from
fastest (Lin & Wang) to slowest (Borkowski).

CONCLUSION AND RECOMMENDATION

This paper has presented the mathematical background to six Cartesian-to-geographic transformation
techniques, two direct and four indirect. In theory, the indirect methods require iteration for an acceptable
solution of φ , but in practice, by restricting the use of the techniques to Earth-bound regions
( −5,000 m ≤ h ≤ 10,000 m ) , only the Simple Iteration technique actually requires iteration, the other indirect
methods giving, for all practical purposes, “exact” solutions for tan φ without iteration. All the methods were
coded as computer algorithms, tested (on a region covering Australia) and found to give acceptable results for
the computation of φ and h. Lin & Wang’s method is appreciably faster than the other five methods and is
recommended for use in Australia.

11
REFERENCES

BG, 1988. The Geodesist’s Handbook 1988, Bulletin Géodésique, Vol. 62, No. 3.

Bomford, G, 1980. Geodesy, 4th ed, Clarendon Press, Oxford.

Borkowski, K.M, (1987). Transformation of Geocentric to Geodetic Coordinates without Approximations,

Astrophys. Space Sci, No. 139, pp. 1-4 and No. 146 (1988), p. 201.

Borkowski, K.M, (1989). Accurate Algorithms to Transform Geocentric to Geodetic Coordinates, Bulletin
Géodésique, Vol. 63, pp. 50-56.

Bowring, B.R, (1976). Transformation from Spatial to Geographical Coordinates, Survey Review, Vol. XXIII,
No. 181, pp. 323-327.

Featherstone, W.E, 1994. An explanation of the Geocentric Datum of Australia and its effect upon future
mapping, Cartography, Vol. 23, No. 2, December 1994 and Vol. 24,
No. 1, June 1995

Laskowski, P, (1991). Is Newton’s iteration faster than simple iteration for transformation between geocentric
and geodetic coordinates? Bulletin Géodésique, Vol. 65, pp. 14-17.

Lin, K. -C and Wang, J, (1995). Transformation from geocentric to geodetic coordinates using Newton’s
iteration, Bulletin Géodésique, Vol. 69, pp. 300-303.

NMC, (1986). The Australian Geodetic Datum Technical Manual, Special Publication 10, National Mapping
Council of Australia, Canberra.

Ozone, M. I, (1985). Non-Iterative Solution of the φ Equation, Surveying and Mapping, Vol. 45, No. 2, pp.
169-171.

Paul, M.K, (1973). A Note on Computation of Geodetic Coordinates from Geocentric (Cartesian) Coordinates,
Bulletin Géodésique, No. 108, pp. 134-139.

Rapp, R.H, (1984). Geometric Geodesy - Part I, Department of Geodetic Science and Surveying, The Ohio
State University, pp. 121-126.

Spiegel, M.R, (1968). Mathematical Handbook of Formulas and Tables, Schaum’s Outline Series, McGraw-
Hill, New York.

Steed, J, (1996). The Geocentric Datum of Australia, Azimuth, Mar 1996, pp. 24-26 and Apr 1996, p. 23.

Steed, J. & Holtznagel, S, (1994). AHD Heights from GPS using AUSGEOID93, The Australian Surveyor,
Vol. 39, No. 1, pp. 21-27.

Vincenty, T, (1985). Comment and Discussion, Re: Non-Iterative Solution of the φ Equation, by Mohammed
Id Ozone, Surveying and Mapping, Vol. 45, No. 3, pp. 265-267.

12