A COMPARISON OF METHODS USED IN RECTANGULAR TO GEODETIC COORDINATE TRANSFORMATIONS1

Robert Burtch Professor Surveying Engineering Department Ferris State University 915 Campus Drive, Swan 314 Big Rapids, MI 49307-2291

INTRODUCTION
The purpose of this paper is to gather and review the different methods of transforming geocentric coordinates to geodetic coordinates. The direct procedure is fairly simple and straightforward. Formulas for this transformation will be given in the next section. Problems exist for the reverse transformation since no pure inverse relationship exists. The methods presented can be broken down into two general categories: iterative and direct or non-iterative methods.

BASIC CONCEPTS
Before beginning the evaluation, some basic concepts of geometric geodesy are presented. For a comprehensive treatment of the topic the readers are encouraged to read Rapp [1998], Torge [2001], Vanicek and Krakiwsky [1991], or a number of other introductory geodesy textbooks. The following definitions will be used. a = semi-major axis of the ellipse (equatorial axis) b = semi-minor axis of the ellipse (polar axis) C = center of curvature of the meridian ellipse ab f = flattening: f = a a2 b2 2 e = first eccentricity: e = a2 a2 b2 e = second eccentricity: e' 2 = b2 Eh = error in height E = error in latitude h = geodetic height

Paper presented at the ACSM Annual Conference and Technology Exhibition, Orlando, FL, April 21-26, 2006.

M = radius of curvature in the meridian: M =

(1 e

a 1 e2
2

N = radius of curvature in the prime vertical: N = p = distance from the polar axis to the point: p =

sin 2 a

=
2

a 1 e2 W3
= a W

1 e 2 sin 2

X 2 + Y 2 = ( N + h) cos

R = distance from the origin to the point: R = X 2 + Y 2 + Z 2 W = constant: W 2 = 1 e 2 sin 2 X, Y, Z = Cartesian, geocentric coordinates of a point = reduced or parametric latitude (sometimes referred to as eccentric latitude) = geodetic longitude = geodetic latitude = geocentric latitude Looking at the meridional ellipse, one can see that three different latitudes can be displayed as shown in figure 1 [Rapp, 1989]. The geodetic latitude, , is the angle formed in the meridional plane where the normal to the ellipsoid intersects the equatorial axis. Note that this point is generally not at the center of the ellipse. The reduced latitude, , is the angle measured at the center of the equivalent circle from the equatorial plane to point P on the circumference of the circle (line OP in figure 1). The reduced latitude is frequently referred to as the parametric latitude. Finally, the geocentric latitude, , is the angle measured at the center of the ellipse from the equatorial axis to the point on the ellipse (line OP in figure 1). From Rapp [1989] numerous mathematical relationships between these latitudes can be developed. For example, one can write tan = Z = 1 e2 X
1 1 2 2

tan = 1 e 2 tan

tan = 1 e 2

( ) tan = (1 + e )
2

tan tan

CARTESIAN-GEODETIC COORDINATE RELATIONSHIPS

The direct problem involves the computation of the X-, Y-, and Z-coordinates of a point given the latitude, longitude, and height. These relationships are well known and can be derived with the help of the relationships shown in figure 1. The direct problem is solved using:
X = ( N + h ) cos cos Y = ( N + h ) cos sin Z = N 1 e 2 + h sin

[ (

) ]

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 2

Figure 1. Meridional Ellipse. The inverse association is not that easy, especially when solving for the latitude. The longitude can be easily computed by dividing Y by X. Hence, Y ( N + h ) cos cos = = tan X ( N + h ) cos sin Therefore,
Y X

= tan 1

The complexity of the latitude computation lies in the fact that the radius of curvature in the prime vertical contains the latitude and one cannot easily isolate that variable without making some simplification of the formula. Because of this, two approaches have been devised. The first is an iterative approach while the second uses a direct solution.

ITERATIVE METHODS Borkowskis Iterative Method

Borkowski [1989] presented an iterative approach where the successive approximations are performed on the reduced latitude (see also Gerdan and Deakin [1999] for a detailed description). Beginning with

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 3

p = a cos + h cos Z = b sin + h sin

Rewrite these equations in terms of the height and geodetic latitude terms then divide the second with the first equation. Recognizing the relationship for the geodetic latitude in terms of the reduced latitude, we have a Z b sin tan = b p a cos Eliminating the denominator and multiplying the expression by 2, 2ap sin 2bZ cos 2 a 2 b 2 cos sin = 0 Borkowski defines = tan 1 bZ ap for ap and sin and
c= a2 b2

(ap )

+ (bZ )

Substitute

for bZ (Note the Gerdan and (ap )2 + (bZ )2 + (bZ ) Deakin [1999] designate the denominator of both of these substitutions as q), we arrive at

cos

(ap )

f ( ) = 2 sin( ) c sin(2 ) = 0

(1)

Solve this equation using the Newton-Raphson iterative technique. The first derivative of the function is

f ' ( ) = 2 [cos( ) c cos(2 )]

For an initial estimate of the reduced latitude, Borkowski suggests using

(2)

o tan 1

aZ bp

(3)

For points on the surface of the ellipsoid, (3) is exact. Evaluate the original function and the first derivative, equations (1) and (2), to arrive at the new estimate of the reduced latitude.

1 = o

f ( ) f ' ( )

(4)

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 4

If the second term on the right hand side of (4) is sufficiently small f ( ) criteria then the reduced latitude is 1 as shown in (4). Otherwise, f ' ( ) recalculate (1) and (2) again, this time using the updated value for to arrive at a new estimate for . Continue to iterate until the criteria is met. Once determined, the geodetic latitude is found using the formula

= tan 1 tan

a b

(5)

Although the height can be determined using a number of different formulae, Borkowski suggests using the next formula because its accuracy over all ranges of latitudes.

h = ( p a cos ) cos + (Z b sin )sin

(6)

Borkowski [1989] claims that accuracies for the geodetic latitude at the 1 x 10-9 rad (0.0002) can be achieved in just two iterations for points that are over 1,000 km from the origin of the coordinate system. The results of his comparison with other methods are shown in figure 2. Note that the method identified at Heiskanen and Moritz [1967] is referred to as the Hirvonen and Moritz method in this paper and acknowledged as such by Heiskanen and Moritz (see also Rapp [1989]). Other methods mentioned here that will be discussed in this paper are those by Long [1975], Pick [1985], Heikkinen [Rapp, 1989], and Borkowskis exact method [1989].

Bowrings Iterative Method

Bowring [1976] presented an iterative procedure which has been described as a standard upon which other methods have been based. The principles are succinctly described in Rapp [1989]. Figure 3 shows the meridian ellipse. Point P is the location of the point and P is the corresponding point located at the intersection of the ellipsoid with the spheroidal normal. M is the radius of curvature in the meridian and C is the center of curvature of the meridian ellipse. From figure 3, xC = x P" M cos (7 )

From Rapp [1989], equation (3.42)

x P" = a cos a cos = W 1 e 2 sin 2

Substitute the value of xP into (7) and recalling the equation for the radius of curvature in the meridian,

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 5

Figure 2. Results from Borkowski's evaluation of transformations from Cartesian to geodetic coordinates [Borkowski, 1989, p.53].

z
P

h
P"

M
f

x
C

Figure 3. Meridian ellipse showing the relationship between the center of curvature in the meridian.

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 6

xC =

a cos a 1 e 2 cos W W3 a cos 1 e 2 a cos e 2 1 sin 2 = 1 = W W W2 W2 2 3 2 3 ae cos ae cos = = 3 3 W 1 e 2 sin 2 2

(8)

Again from Rapp, equation (3.66)

cos = W cos

Then, substitute this into (8) will give us the x-coordinate of the center of curvature of the meridian ellipse. xC = ae 2W 3 cos 3 = ae 2 cos 3 3 W (9)

From figure 3, we can develop a similar formula for the z-coordinate of C. Write z C = z P" M sin From Rapp, equation (3.43) z P" a 1 e 2 sin a 1 e 2 sin = = W 1 e 2 sin 2 (10)

which, when substituted into (10) yields

zC = a 1 e 2 sin a 1 e 2 sin W W3 1 a 1 e 2 sin = 1 2 W W

( (

) )

(11)

In Rapp, equation (3,41), W2 can be expressed as

W2 = 1 1 + e' sin 2
2

Substitute into (11) yields

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 7

a 1 e 2 sin zC = 1 1 + e' 2 sin 2 W a 1 e 2 sin = e' 2 sin 2 W

( (

[ (
(

)]

But, Rapp shows that sin = V sin in (3.65), where V = (1 + e' 2 cos 2 ) (3.40). This means zC = Rapp shows in (3.49) that a 1 e 2 V sin e' 2 sin 2 W

as shown in

(12)

V a = . Substitute this into (12) yields the formula for the zW b coordinate for the center of curvature of the meridian ellipse.
z C = be' 2 sin 3 B (13)

From figure 3, one can write the following relationship for the geodetic latitude.

tan =

z P zC x P xC

(14)

Substitute the values of xC and zC in equations (9) and (13) respectively which yields z P + be' 2 sin 3 tan = x P ae 2 cos 3 Express the values for xP and zP in terms of the Cartesian coordinates of the point. The solution for the latitude then can be expressed as [see Bowring equation (4)]: tan = Z + be' 2 sin 3 p ae 2 cos 3 (15)

This formula is clearly iterative since both the geodetic and the reduced or parametric latitude are unknown. Bowring suggests using the next formula to determine the initial estimate of . Z a tan = p b (16)

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 8

Thus, once (16) is computed, the value for is substituted into (15) to arrive at the geodetic latitude. If a new value for the reduced latitude is needed then Rapp [1989] shows that it can be computed as
tan = (1 f ) tan = b tan a (17)

where is determined from (15). This value for is then inserted into (15) to compute a new estimate of . Once the latitude is determined, the height can be computed from one of the following [see Bowring (8) and Rapp (6.98) and (6.99)]. h= p Z N = N + e2 N cos sin (18)

As both Bowring [1976] and Rapp [1989] point out, this method is very accurate for most terrestrial points. For example, Bowring shows that the error in latitude using his method can be defined as (Bowrings equation 6)

3 6 ah 2 sin 3 cos 3 E = e 3 2 (a + h )

(19)

Then, for earth-based points, the main error will be 0.000 000 030 for a single iteration. The maximum latitude error is shown to be 0.0018 when h = 2a. As Rapp points out, the error in height using the Bowring method is around 39 mm, at a height of 5,000 km. As Bowring [1985] states, the error in his method is not appropriate for outer-spaced scenarios. He identified the error in height to be

p Eh = e' M cos sin 2 cos

(20)

Then, at the maximum latitude error of 0.0018, the error in height is 17 cm. To alleviate this large of discrepancy, Bowring [1985] suggests using the following formula to calculate the height. a2 h = p cos + Z sin N a2 = R cos( ) N where tan = Z (21)

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 9

Despite the fact that the latitude is a part of equation (21), Bowring shows that the error in height is insensitive to any error in latitude to the first order differential. In fact, he also shows that the worst case scenario for error is when h = 4a and even then the error is -9.1 x 10-10 m. Bowring also addressed refining the algorithm for computing the latitude. In his 1976 paper, the initial estimate of the reduced latitude is as shown in (16). For a better approximation, Bowring [1985] suggests
tan = b bZ 1 + e' R ap ( 22)

Then using (15) solve for the geodetic latitude. This algorithm is estimated to be accurate to 0.000 000 1 for any point, either terrestrial or outer space. Rapp [1989] presents an alternative formula for the computation of height, which was developed by Bartelme and Meissl. The height is shown as
h 2 = ( p a cos ) + (Z b sin )
2 2

(23)

This formula is stable, although it does fail at the poles.

Fukushimas Iterative Method

As will be describe later in the section on closed form methods, Borkowski [1987, 1989] presents an alternative to the iterative method already discussed. He shows that the solution involves finding roots of a quartic equation shown as (see later section for what the variables represent): t 4 + 2 Et 3 + 2 Ft 1 = 0 (24)

His solution was to employ Ferraris formula. But, as Fukushima [1999] points out, the distance from the polar axis, p, which is a part of E and F, is ill-conditioned as p gets smaller. As the co-latitude tends towards zero, the error grows significantly. To solve the problem, Fukushima offers a different form for equation (24). We can write f (t ) pt 4 + ut 3 + vt p = 0
1 1 where: t tan = tan tan 1 e' tan 4 2 2

(25)

u 2 pE = 2(Z 'c ) v 2 pF = 2(Z '+c ) c ae 2 Z ' e' Z

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 10

Z 'c p Z '+ c F p

The solution is referred to as an iterative trigonometric procedure performed using the Newton-Raphson method. Like any iterative approach, convergence can be accelerated by selecting a better estimate of the unknown parameters. Here, as in Borkowski, the evaluation is based on t which is the tangent of half the complement of the reduced latitude to 90. Fukushima evaluates the first and second derivatives of (25), which yields (Fukushima equations 11 and 12)

f ' (t ) = 4 pt 3 + 3ut 2 + v f " (t ) = 12 pt 2 + 6ut

From this, it can be explicitly shown that the zeros for the second derivative are (Fukushima equation 12) t = 0, t = tM u c Z' = p 2p

Then, three different cases were identified depending upon TM. In most cases, t M 0 . In this case, the upper bound can be used as the initial estimate for t in the NewtonRaphson iteration. The initial estimate for t then becomes (Fukushima equation 13) t= p c + Z' p c + 2Z ' (26)

To look at the initial estimates for the other cases the reader is referred to Fukushima [1999], p.605. The new estimate of t is
t t + t ( 27 )

f (t ) p ( pt 4 + ut 3 + vt ) = . Equation (27) is iterated until t is sufficiently f ' (t ) 4 pt 3 + 3ut 2 + v small for the problem. Then, the latitude is found using: where t

= tan 1

2e' t 2 1 t

( 28)

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 11

and the height is given as (Fukushima equation 9): h= 2 pe' t + Z (1 t 2 ) ae' (1 + t 2 ) (29)

(1 + t )

2 2

4e 2 t 2

or by using one of the forms of the height equation (21) given earlier. Fukushima [1999] explains that the approach has several advantages to both the Borkowski and Bowring methods used in the comparison. First, the approach is faster than Bowrings despite the fact that it requires a few iterations to complete while Bowrings method needs, for all practical purposes, only a single iteration to find the latitude and height. A big part of the reason is that Fukushima requires no calls for transcendental functions in the iteration while Bowring requires 8-9 calls, including atan, tan, sin/cos, and sqrt. Secondly, Fukushimas method is reportedly accurate to 10-15 or 10 nm on the surface of the earth. Finally, this method is very stable for all types of inputs, from heights close to the geocenter to satellite altitudes, such as the Global Positioning Satellite constellation.

Hirvonen and Moritz Iterative Method2

z
P h P"
Q

f
N+h (N + h) sin f

f
O

x
f
Of

Figure 4. Meridian section showing the radius of curvature in the prime vertical (adapted from Rapp, 1989 and Pollard, 2002). Many authors refer to this as the Heiskanen and Moritz algorithm or method because it appears in their book. The authors themselves refer to the work by Hirvonen and Moritz within the textbook.
2

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 12

The iterative solution by Hirvonen and Moritz is described in Heiskanen and Moritz [1967] and Rapp [1989] and the geometry is presented in figure 4. From basic trigonometry we can write [Rapp, equation 6.94]: tan =

(N + h )sin
p

(30)

Recall earlier that it was shown that Z = N (1 e 2 ) + h sin , which, after rearranging, yields

(N + h )sin = Z + e 2 N sin
Substitute this into (30) gives us

tan =

Z + e 2 N sin p

(31)

This is the basic equation to iterate. Heiskanen and Moritz [1967] and Torge [2001] show the equation in a slightly different form:
tan =
Z N 2 1 e p N +h
1

(32)

In (32), both (which is imbedded in N) and h are unknowns on the right hand side of the equation. By setting h = 0, (32) becomes Z 1 tan 0 = p 1 e2 (33)

Here, 0 indicates the initial estimate of the latitude. Use this value for the latitude to find the radius of curvature in the prime vertical. Then upgrade the estimate of the geodetic latitude using (31). If using the Heiskanen and Moritz form of the tan , one needs to calculate the geodetic height using (18) [Heiskanen and Moritz give the first p form of (18), namely h = N . Evaluate = 0 . If is less than the cos desired criteria, the last value of the latitude is the correct value and a new value for the elevation is determined using (18). Otherwise, 0 is replaced by and the process continues.

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 13

From Rapp (equation 3.18) we can write (1 e 2 ) 2 = (1 + e' 2 ) be written as (see Gerdan and Deakin, 1999, equation 3.1):
1

. Therefore, (33) can also

tan 0 =

Z (1 + e' 2 ) p

Lin and Wang Iterative Method

Like Fukushima and others, Lin and Wang [1995] recognized the computation burden that the Bowring method requires due to the use of trigonometric and inverse functions. Thus, they presented a more efficient method based employing the Newton-Raphson iterative procedure. The Lin and Wang approach also involves solving a quartic equation like Fukushima and Borkowski [Pollard, 2002]. To begin, the ellipsoid is modeled using
2 2 X P" YP2" Z P" + 2 + 2 =1 a2 a b

(34)

where XP, YP, ZP are the geocentric coordinates of point P on the ellipsoid. The coordinates of any point on the normal to the ellipse passing through point P can be described from the parametric equation as:

x=
y=

X 1 + 2m
Y 1 + 2m

a2
a2

z=

Z 1 + 2m

b2

where m is a parameter that describes where along the normal the point (x, y, z) is located. For example, when m = 0, x = X, y = Y, and z = Z. Lin and Wang describe the problem as one of finding that value of m where (x, y, z) = (XP, YP, ZP). Substitute the values for x, y, z into (34) gives us
f ( m) =

X2 a

a + 2m p2

) (
2 2

Y2 a

a + 2m Z2

) (
2 2

Z2 b

b + 2m

1 = 0

(a + 2m a ) (b + 2m b )

1 = 0

(35)

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 14

Equation (35) is solved iteratively using the Newton-Raphson method. estimate for m is given by Lin and Wang as
ab a 2 Z 2 + b 2 p 2 2 a 2 b 2 a 2 Z 2 + b 2 p 2 m0 = 2 a4Z 2 + b4 p2

An initial

(36)

To begin the iteration process take the derivative of (35). df (m) p2 f ' (m ) = = 4 dm 2m a a + a + 3 b b + 2m b Z2

(37)

Compute a new estimate of m.

mi = mi 1

f (mi 1 ) f ' (mi 1 )

(38)

Equations (35), (37) and (38) are solved for iteratively until f(m) converges towards zero (the criteria established by the user). Then, using the most current value for m, compute

p P" =

p 2m 1 + 2 a
2 p P" a2

(39)

Z P" = b 1

(40)

The latitude is found using

= tan 1
and the height is

a 2 Z P" 2 b p P"

(41)

h=

( p p P " )2 + (Z Z P " )2

(42)

According to Lin and Wang, the error in their algorithm has the same magnitude as the error found in Bowrings approach. Both methods also converge at about the same rate, but the algorithm here is a bout 50% faster than Bowrings approach. Gerdan and Deakin

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 15

[1999] conclude that from the methods they tested, the Lin and Wang approach was the most efficient method in terms of execution time.

Pollards Height First Iterative Method

Pollard [2002] presented two vector methods from converting Cartesian coordinates to geodetic coordinates. As he points out that while the vector method is simple to follow because it does not require the solution of a quartic equation, it does require the separation of the height from the latitude determination thus making this approach iterative. The first method presented is based on the computation of the height first through an iterative process and, once determined, to compute the latitude directly. The theory of Pollards method begins with the general equation of an ellipse of revolution for the earth, written in general form as [note that this is a slightly different form of equation (34)]
a x + y + z = a2 b
2 2 2 2

Designate k P" = a

( b) .
2

Then differentiate the general ellipsoid formula and substitute

into the results the coordinates of point P, which lies on the surface of the ellipsoid. This yields the direction numbers of the normal at P [Pollard, 2002]. Then, from figure 4, the distance QO = k P" Z P" and
N=
2 2 2 X P" + YP2" + k P" Z P"

The unit normal rectangular coordinates, designated by Pollard as l, m, and n, are found using the following relationships

l=

X P" N

m=

YP " N

n=

k P" Z P" N

The coordinates of P can then be shown in terms of the height, h, to be X P" = X lh YP" = Y mh Z P" = Z nh Write the equation of the ellipse in terms of the coordinates of P.
2 2 X P" + YP2" + k P" Z P" = a 2

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 16

Substitute the values of XP, YP, and ZP yields

( X lh )2 + (Y mh )2 + k P" (Z nh )2
This can be shown in a quadratic form as

= a2

(l

+ m 2 + k P" n 2 )h 2 2 (lX + mY + k P" nZ ) h + (X 2 + Y 2 + k P" Z 2 a 2 ) = 0

Pollard [2002] defines r, s, and t as

r = l 2 + m 2 + k P" n 2 = 1 + (k P" 1) n 2 = 1 + e' 2 n 2 s = lX + mY + k P" nZ t = X 2 + Y 2 + k P" Z 2 a 2

Thus, the height can be found as

h=

s s 2 rt r

As Pollard point out, the root with the minus sign yields the right answer. The algorithm employed by Pollard can be outlined in the following steps. The problem is to find a solution to ZP for the point on the ellipsoid where the normal to the ellipsoid passing through point P intersects the ellipsoid. For most points on the earths surface, the ZP coordinate can be approximated by the Z-coordinate of the point in question. For other points located farther away from the ellipsoid, Pollard recommends scaling the Zcoordinate and using an initial value of
Z P"ini = bZ R

(43)

Make the initial value of Z on the ellipse, Z P "ini = Z P" .

k P" = a

( b)

The estimated distance from P to O is found using

P" O = X 2 + Y 2 + (Z + e' 2 Z P" )

l= X P" O m= Y P" O

(44)
Z + e' 2 Z P " P" O (45)

n=

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 17

r = l 2 + m 2 + k P" n 2 s = lX + mY + k P" Z t = X 2 + Y 2 + k P" Z 2 a 2

(46)

s s 2 rt h= r

(47)

Z P" = Z nh

(48)

Compare this new estimate for the Z-coordinate for point P with the initial estimate. If the difference is within the desired criteria, then the geodetic height is as determined in (46). Otherwise, using the new estimate for ZP and recalculate the variables in equations (43) (47). This iterative process is continued until the solution converges. Then, once the geodetic height has been finalized, compute the geodetic latitude using the algorithm as follows. Calculate the coordinates of the point of intersection of the normal to the ellipse passing through the point P with the ellipse (XP and YP). X P" = X lh YP" = Y mh Next, compute the length of the normal, N, using
N = P" O h

Then, the latitude is found using

= tan 1

Z + e' 2 Z P " p

(49)

Pollards Latitude First Iterative Method

Pollard [2002] presented an alternative iterative vector approach to finding the geocentric to geodetic coordinate transformation where the latitude is created first. The algorithm presented by Pollard is as follows. Begin by computing the initial value of the Zcoordinate along the normal at the ellipsoid (43). Then compute the tangent of the parametric latitude, t.

b Z + e' 2 Z P " t = tan = a p

(50)

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 18

Update the estimate of ZP.

Z P" = b

t 1+ t2

(51)

Iterate on ZP using either a simple iteration or, as recommended by Pollard, a NewtonRaphson method were the new estimate of ZP is found by
bt Z P"New = 1+ t 1
2

( 1+ t )
2

bZ P"t '

( 1+ t )
2

bt '

(52)

dt be' 2 where, t ' = = . dZ P" ap Iterate until the discrepancy between the estimate on the right hand side of (52) for ZP and the new value for ZP on the left hand side are sufficiently small. Once done, compute the latitude and height using:

= tan 1
h=

Z + e' 2 Z P " p
p N cos

(53)

(54)

Pollard [2002] compared his methods with Bowrings approach in terms of errors and speed. For terrestrial points, the error in latitudes are comparable. In terms of height errors, Pollard found the errors in his approach are lower than those found in Bowrings formulas. In terms of execution time, all algorithms had similar levels of performance. Pollards vector methods were slightly better when the programs were optimized.

Barteleme and Meissls Iterative Method

Rapp [1976], in his documentation for the Fortran program XYZ2PLH, gives a concise overview of the algorithm attributed to Barteleme and Meissl3. The basic equation to iterate is

Although not identified specifically, it is believed that the reference is (see for example the references in Sjberg [1999]): Bartelme, N., and P. Meissl, 1975. Ein einfaches, rasches und numerisch stabiles Verfahren zur Bestimmung des krzesten Abstandes eines Punktes von einem sphroidischen Rotationsellipsoid, Allg. Verm. Nachr., 12: 436-439.

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 19

1+

k +1
R2

2 1+ p = a 1+

k
R2 k a2

k 2 1+ Z R2 + k b a+ 2 b

(55)

Iterations begins with k = 0 . Rapp states that the speed of the convergence depends on the selection of R. The method for calculating R from Bartelme and Meissl is

R2 =

a2 e' 2 Z 2 1+ 2 Z + p2

(56)

or using the approximation R 2 = ab

(57 )

Rapp states the Vincenty offers an alternative form that speeds up the convergence. This is 2 2 R = a 1 b 2 2 1 Z a 4 2 b 2 p +Z a

(58)

Once the algorithm converges, then the geodetic height is found using

h=

( p p E ) 2 + (Z Z E )2

(59)

where p E = p 1 + 2 a

Z E = Z 1 + 2 b Finally, the geodetic latitude is found using

ZE a2 = tan p b2 E
1

(60)

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 20

Sjbergs Iterative Method

The iterative method developed by Sjberg [1999] begins in a similar fashion as that by Hirvonen and Moritz. He begins by expressing equation (31) in the following form p tan Z = e 2 a sin 1 e 2 sin 2

(61)

Squaring both sides of (61) and making the following substitutions

= tan

oo = Z p

e2a p

We can, after some manipulation, arrive at Sjbergs equation (7):

( oo )2 [1 + (1 e 2 ) 2 ] 2 2 = 0

(62)

Sjberg acknowledges that equation (62) does become unstable in polar regions. In those cases, he suggests making the following substitutions:

= 1 = cot
Then, (62) can be shown as

oo = p Z

e2a Z

( oo )2 [ 2 + (1 e 2 )] 2 2 = 0

(63)

Sjbergs method is iterative in finding an acceptable value for either or . For nonpolar regions, an initial estimate for is found using equation (33) noting that Sjbergs notation denotes = tan . Then, the iteration is shown as

k +1 = oo +

k
1 + A k2

(64)

where oo and are given above and A = 1 e2. The main difference between (63) and (32) is that Sjbergs method does not require iterations involving transcendental functions, like the arctangent. Instead, each iteration uses the tangent value for the latitude and only at the end does one find the actual angular value for the latitude. This speeds up the algorithm processing and Sjberg [1999] indicates that this new method performed faster than the Hirvonen and Moritz method. For the polar regions, Sjberg suggests using

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 21

0 =

p 1 e2 Z

(65)

which is the cotangent of 0 as identified earlier in equation (33). Then the equation to iterate is

k +1 = 0

k
A + k2

(66)

where the variables have been defined above. Sjberg [1999] tested this method with those iterative methods identified in Borkowski [1989] and found that his method performed better. It works for all latitudes and heights close to the surface of the earth. He also compared his iterative method to the two closed methods developed by Fotiou [1998], which will be presented later. He found the height method performed much better than the distance method and recommends the height method as a possible alternative approach to the method he presented.

Seemkooeis Iterative Method

Seemkooei [2002] presented an iterative method that is very similar to that of Hirvonen and Moritz. Recall from (31) that

tan =

Z + e 2 N sin p

(31)

This can be rewritten as (Seemkooei equation 11) p sin = Z cos + Ne 2 sin cos (sin 2 + cos 2 ) or as tan = Z + Ne 2 sin 3 p Ne 2 cos 3 (67)

Equation (67) is solved iteratively. An initial estimate of the latitude is given in equation (33). Note that this equation can be easily transformed into the form depicted by Hirvonen and Moritz. Seemkooei [2002] compared his method with those of Bowring, Borkowski (iterative), Lin and Wang, and Hirvonen and Moritz, along with a closed form where he solves the quartic equation. The accuracy, as measured in terms of the latitude, were slightly better using Seemkooeis method, but not significantly. He did find in his experiments that the closed form was considerably less precise than the iterative methods tested. In terms of

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 22

speed in calculations4, Seemkooei found that Bowrings method only required one iteration and was the fastest algorithm employed, about 35% faster than his own method (using average times). The next fastest was Seemkooei, which required 2.3 iterations on average, followed by Borkowski, which averages 2 iterations, followed by the closed form solution, then Hirvonen and Moritz, which required about 3.2 iterations as an average, with Lin and Wangs algorithm being the slowest, requiring only about 1.4 iterations on average.

CLOSED FORM METHODS Borkowskis Direct Method

In addition to the iterative solution, Borkowski [1987, 1989] developed a closed or direct solution to the conversion of Cartesian to geodetic coordinates. It was shown in Borkowskis Iterative Method section in equation (1) that

2 sin ( ) c sin (2 ) = 0
Expressing this equation in terms of tan t 5, a fourth-degree quartic or 4 2 biquadratic equation can be developed. This is shown as t 4 + 2 Et 3 + 2 Ft 1 = 0 where Borkowski defines (68)

bZ a 2 b 2 E= ap bZ + a 2 b 2 F= ap

)
)

(69a ) (69b)

Using Ferraris formula for the solution, we have

t = G2 +
where
v=3 G=

F vG G 2G E
D +G

(70)
(71a ) (71b)

D Q E2 + v + E 2

4 5

See Seemkooei [2002] for specifics on experimentation he performed. This is a tangent function of half the complement of the reduced latitude to 90o.

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 23

D = P3 + Q2 4 P = E2 F 2 3 Q = 2 E2 F 2

(71c )

(71d ) (71e)

It is clear from the in (70) and (71b) that the quartic solution contains four solutions. While not significant, if D < 0, which occurs when the point is less than about 45 km from the center, Bowkowski does suggest using the following equation for v: v=2

( P ) cos 1 cos 1 Q ( P )
3 P

The final solution for the latitude and height can be found using:

a 1 t2 2bt h = ( p at ) cos + (Z b )sin

= tan 1

(72)
(73)

Borkowski states that using the positive square roots in (70) and (71b) will give the desired answer for most applications provided that a > b and > 0. The algorithm does not work when = 90. For latitudes in the southern hemisphere the user must apply a negative value to the semi-minor axis (i.e., -b). Another weakness in the algorithm is the cube root required in (53a). The approach by Borkowski found round-off problems close to the Z-axis. He points out that Heikkinens approach is free from this problem, even though the cube root is used in Heikkinens method, but the trade-off is that Heikkinens algorithm cannot be used when D < 0. To solve Borkowskis problem, he suggests using the cubic resolvent, v (one of the three roots) to compute v as

(v' +2Q ) v=
3

3P

CONCLUSION
This paper presents a general overview of some methods used in the transformation from Cartesian coordinates to geodetic coordinates. Two general approaches are employed in these types of transformations: iterative or direct. The latter approaches employ some form of approximation to arrive at the desired solution.

REFERENCES

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 24

Borkowski, K.M., 1987. Transformation of Geocentric to Geodetic Coordinates Without Approximations, Astrophysics and Space Science, 139: 1-4. Borkowski, K.M., 1989. Accurate Algorithms to Transform Geocentric to Geodetic Coordinates, Bulletin Godsique, 63: 50-56. Bowring, B.R., 1976. Transformation from Spatial to geographical Coordinates, Survey Review, 23(181): 323-327. Bowring, B.R., 1985. The Accuracy of Geodetic Latitude and Height Equations, Survey Review, 28(218): 202-206. Fukushima, T., 1999. Fast Transform from Geocentric to Geodetic Coordinates, Journal of Geodesy, 73: 603-610. Gerdan, G.P., R.E. Deakin, 1999. Transforming Cartesian Coordinates X, Y, Z to Geographical Coordinates, The Australian Surveyor, 44(1): 55-63. Heiskanen, W. and H. Moritz, 1967. Physical Geodesy, W.H. Freeman and Company, San Francisco, CA. Lin, K-C. and J. Wang, 1995. Transformation from Geocentric to Geodetic Coordinates using Newtons Iteration, Bulletin Godsique, 69(4): 300-303. Pollard, J., 2002. Iterative Vector Methods for Computing Geodetic Latitude and Height from Rectangular Coordinates, Journal of Geodesy, 76: 36-40. Rapp, R.H., 1976. Unpublished notes for Fortran Program XYZ2PHL, Department of Geodetic Science, The Ohio State University, Columbus, OH. Rapp, R.H., 1989. Geometric Geodesy Part 1, Department of Geodetic Science and Surveying, The Ohio State University, Columbus, OH. Seemkooei, A.A., 2002. Comparison of Different Algorithms to Transform Geocentric to Geodetic Coordinates, Survey Review, 36(286): 627-633. Sjberg, L.E., 1999. An Efficient Iterative Solution to Transform Rectangular Geocentric Coordinates to Geodetic Coordinates, Zeitschrift fr Vermessungswesen, 124(9): 295-297. Torge, W., 2001. Germany. Geodesy, 3rd edition, Walter de Gruyter GmbH & Co., Berlin,

Vanicek, P. and E. Krakiwsky, 1991. Geodesy: The Concepts, 2nd edition, 3rd impression, Elsevier, Amsterdam, Holland.

A Comparison of Methods Used in Rectangular to Geodetic Coordinate Transformations

Page 25