Datum transformation by Gauss projection of the ellipsoid onto the sphere

and bilinear function of complex variable: an application to ETRS89 ED50

J. B. Mena
Alcala University, Polytechnic School, Mathematics Department.
Aptdo 20, Alcal de Henares, E-28871 Madrid, Spain
Tel. +34 91 8854901
Fax +34 91 8854951
juan.mena@uah.es

Abstract.
In this paper a simple method for accurate transformation of geodetic coordinates is presented. The
process uses the regional network of points allocated in both geodetic systems researching only the
two dimensional problem (longitude and latitude). Basically, transformation of any point P consists to
determine first the three network vertexes nearest to P such as this point is located inside the triangle.
Computing both geodetic systems separately, these three vertexes are projected from the ellipsoid onto
the Gauss sphere tangent along a parallel. Given that both geodetic variables are isometric on the
sphere, and since only three network vertexes have been considered, it can find of closed way the bilinear transformation of complex variable which applies the triangle in study of a sphere in the other.
Starting from here, the P image is obtained projecting this point on the sphere corresponding to their
geodetic system, applying it later on the other sphere, and computing finally the inverse transformation
which projects this sphere on the second ellipsoid. This procedure achieves higher accuracy than matrix spatial methods of seven parameters, and it could be considered of the same order that methods
based on distortion analysis. However, besides being simpler, the presented process provides better
results there where regional reference network is not very dense.
Keywords: datum transformation, Gauss projection, bilinear function.

1. Introduction.
The establishment of WGS84 (World Geodetic System 1984) as world geodetic reference and
GPS (Global Positioning System) were fundamental reasons for developing accurate methods to transform geodetic coordinates between WGS84 and regional datums like ED50 (European Datum 1950).
Not only in Europe, regional ellipsoids are still used for different applications in other parts of the
world. The study of the mathematical relation between WGS84 and each one of the main regional
geodetic systems (NIMA, 2004) is a topic in continuous development in which progressively better
results are achieved.
Many of datum transformation methods are based on estimation of the unknown terms of a
matrix equation or a polynomial expression utilizing the least squares adjustment and starting of a
network of vertexes whose geodesic coordinates are known in both systems. This Double Vertexes
Network (DVN) is allocated in a limited region, reason for which the transformation equation only
will be correct for interior of this zone. Furthermore, the accuracy of each mathematical model is attached closely to the quality determination and local density of the network.
In Gonzalez-Matesanz et al. (2003) many methods developed in order to carry out transformation from WGS84 to other geodetic reference systems are presented. The seven-parameter models are
especially important for their simplicity, since infinitesimal character of transformation is assumed. An
example well known is the Molodensky Badekas model or Bursa Wolf model. Yet but simple are

-1-

the five parameters methods, where parallelism between corresponding ellipsoid axes is supposed.
With similar accuracy are the models based on determination by statistical regression of polynomial
coefficients in real or complex variable (Draper and Smidth, 1981). However, in a higher step for their
accuracy order are the methods whose intermediate results are corrected by means of distortion analysis (Collier et al., 1998). These corrections are achieved by techniques like the minimum curvature
surfaces (Briggs, 1974; Smith and Wessel, 1990), the rubber sheeting process (White and Griffin,
1985), the multiple statistic regression (Applebaum, 1982), or the least squares collocation (Tapley,
1976; Collier et al., 1998).
Researching the two dimensional geodetic problem, it is possible to generate a basic classification of different transformation methods by only two principal categories according to their accuracy
order. The first group involves the simple methods, which could be defined as those where distortion
analysis is not used. The second group includes any procedure where distortion modelling is applied.
Thereby, all method of the second group is assumed as higher accuracy order than any simple method.
According to this classification the process here presented is not belonging to any group, but rather it
can be classified in the intersection of both groups. In fact, this is a simple method for datum transition, but at the same time it is able to achieve an accuracy order comparable to the methods of the second group.
2. Process description.
2.1. The transformation function.
In order to generalize suppose that a DVN have been accurately determined in coordinates of
geodetic systems S1 and S2. In this frame, assume a point P which is included into region and whose
coordinates in the S1 and S2 systems are (1, 1) and (2, 2) respectively. The problem is to find the
most accurate bijection f between two dimensional spaces as:
[1]
2 , 2 = f 1 , 1

For this goal, assume that f has to verify the following conditions:
1. It has to be conformal, since angles are preserved in both geodetic systems.
2. The linear deformation coefficient in P has to be the corresponding of the DVN.
These conditions show that f has to be a likeness function into a differential environment of P, with
the same likeness factor in this point than DVN. However, f deduction involves two questions. The
first is that latitude variable is not isometric on the ellipsoid, and the second is that linear deformation
factor is unknown while f is not calculated.
For solving the first question we could opt for a variable change in order to transform the geodetic latitudes into isometric latitudes by usual way (Bugayevskiy and Snyder, 1995; Lauf, 1983).
However, longitude and latitude are isometric coordinates on the sphere. Then we can find the bijection of each ellipsoid on the corresponding sphere so that linear deformation between both projections
of P on these spheres is the same as the linear deformation between both positions of P on both ellipsoids. To achieve this goal, which solves our two questions, Gauss theory for ellipsoid to sphere projection has been applied (Zakatov, 1953). In this projection sphere and ellipsoid are tangents along a
parallel, and the sphere radius value is the mean of curvature of the ellipsoid in the selected parallel.
The Gauss projection is conformal on entire sphere and practically equidistant in two degrees to both
sides of tangent parallel. Denominating g1 and g2 these Gauss projections on respective parallels of P,
we write:
[2]
1 , 1 = g1 1 , 1 ,
2 , 2 = g 2 2 , 2

-2-

Here the first member coordinates are the corresponding P images on each sphere. Concretely, for an
ellipsoid where e is the first eccentricity and 0 is the geodetic latitude of tangent parallel, the corresponding g function is achieved computing previously the projection constants by means of the following expressions where 0 is the projection of 0:
[3]
1

e 2 cos 2 0 2
;
K = 1 +
2

1 e

1 e sin 0

k =
1 + e sin 0

Ke

sin 0

0 = asin

0
+
tan
2 4

0
+
tan
2 4

Then, the mathematical relations that define this function are:

[4]
1 1 e sin

tan + =
2 4 k 1 + e sin

= K ,

Ke

tan K +
2 4

Since the applied Gauss projection is conformal, the values of linear deformation factor on
meridian (m) and parallel (n) are the same in each point. In fact, equations [4] result by integration of
following differential equations where R is mean earths radius and M and N are the main radii of curvature of ellipsoid:
R d
M d

d = K

m = n d M cos
R cos d

=
n=
d
N cos
N cos d

m=

Besides, on tangent parallel and their vicinity both linear factors are equal to one:
m0 = n0 = K

R cos 0
N cos 0

=1

This equation can be extended up to two degrees to both sides of tangent parallel given that the values
of R, K and k are achieved computing also the following conditions (cf. Zakatov, 1953):

d 2m
2 =0
d 0

dm
d = 0,
0

Therefore, in the Gauss projection defined through Eqs. [2 4] angles and the linear deformation that
exists in P environment are preserved.
Once both equations [4] have been obtained, determination of a function f is reduced to find
the relation between spherical surfaces in P environment. This relation is now established through
isometric variables. Therefore we will opt for representing it by means of complex variable in the following way:
[5]
z = 1 + i 1
where
w = w (z )

w = 2 + i 2

-3-

For sought function w three basic conditions will be assumed:

1. Function w has to be conformal identically as a function f.
2. Function w has to be an union of several geometric elementary transformations.
3. Function w has to be specific for each point P for accuracy reasons.
Analyzing several complex variable transformations, bilinear model was selected because this function
is conformal and it includes rotation, translation, dilation (or contraction) and inversion:
[6]
z+
w=
z +1
Furthermore, in order to verify condition 3, we will compute the complex coefficients , and using
only the three DVN vertexes A, B and C nearest to P such as this point is located inside the triangle. In
this way the resolution of linear system:
[7]
zi + zi wi wi = 0
(i = A, B, C )
provides a close solution. This is more accurate for transformation of P than classic solution by the
least squares adjustment using all DVN vertexes (Higham, 1996).
Equations [1] [2] and [5] give rise to the following decomposition of bijections for function f
definition:
[8]

( 2 , 2 ) = (g 21 w g1 ) (1, 1 )

2.2. Algorithm for a point set transformation.

Once equation [8] has been specified, for transforming a point set from the system S1 to S2, the
presented method involves applying for each one point P the following successive operations:
1. Determination of the minor area triangle ABC whose vertexes are DVN points and as it includes P.
This is a computational geometry problem whose solution can be consulted in De Berg et al. (1998).
2. Starting of S1 coordinates of A, B, C vertexes, computation of mean latitude parallel (01) and construction of the Gauss projection (g1) of ellipsoid 1 on sphere 1 (Eqs. [2], [3] and [4]). Since the geodetic latitude of P is known, this value could be used as latitude of tangent parallel.
3. Starting of S2 coordinates of A, B, C vertexes, computation of mean latitude parallel (02) and construction of the Gauss projection (g2) of ellipsoid 2 on sphere 2 (Eqs. [2], [3] and [4]). As latitude of
tangent parallel, geodetic value of P in S1 also could be used.
4. Projection of points A, B and C on sphere 1 (zi values) and sphere 2 (wi values) by means both equations [4] which have been obtained from steps 2 and 3.
5. Determination of the bilinear transformation [6] which applies zi values on wi values respectively.
For this goal linear system [7] has to be solved.
6. Application of the transformation [8] to S1 coordinates of P. That is to say: g1 projection, w transformation, and g2 inverse projection. Last operation involves successive approaches in the latitude
equation [4], but it converges quickly.

-4-

3. Results and conclusions.

The presented method has been applied to ETRS89 - ED50 transformation into Iberian Peninsula and Balearic Islands using as DVN the REGENTE network (Quiros et al., 2003). This network
embraces around thousand vertexes defined in two systems: ETRS89 (European Terrestrial Reference
System 1989) whose ellipsoid is the same as WGS84, and ED50 where the International Hayford ellipsoid is assumed. REGENTE network involves a convenient density since the mean distance between points is only of 30 40 kilometres. In this frame, once carried out different tests in several
points of the zone, it was obtained an estimated mean error of about 10 - 15 centimetres. This value is
enough for many applications of diverse nature and it suppose an accuracy which is characteristic for
some methods where distortion analysis is used. Table 1 shows some results which were achieved for
peripheral points of the region, this is to say, for points where least accuracy is obtained. In this table
corresponding results to a matrix method of seven parameters (simple method) and a method of distortion analysis (high accuracy) also are presented. It can observe that results of the new method (bilinear
function) are many more accurate than results of the simple method. In fact residual values distortion
analysis bilinear function are smaller than residual values distortion analysis simple method.
Name

CADIZ
ALMERIA
BARCELONA
VIZCAYA
PONTEVEDRA

ETRS89
06
36
02
36
02
41
02
43
08
42

17
31
06
45
09
20
30
22
46
35

02.926
18.163
26.179
08.985
37.816
56.596
36.291
38.228
00.460
49.383

ED50
(simple method)
W
N
W
N
E
N
W
N
W
N

06
36
02
36
02
41
02
43
08
42

16
31
06
45
09
21
30
22
45
35

58.164
22.855
21.811
13.560
42.013
00.610
31.611
42.126
55.189
53.542

W
N
W
N
E
N
W
N
W
N

ED50
(distortion analysis)
06
36
02
36
02
41
02
43
08
42

16
31
06
45
09
21
30
22
45
35

58.038
22.767
21.757
13.535
41.951
00.645
31.662
42.140
54.986
53.601

W
N
W
N
E
N
W
N
W
N

ED50
(bilinear function)
06
36
02
36
02
41
02
43
08
42

16
31
06
45
09
21
30
22
45
35

58.037
22.770
21.780
13.532
41.944
00.641
31.661
42.140
54.991
53.601

W
N
W
N
E
N
W
N
W
N

Table 1: Some results for peripheral points of the region.

By way an extreme test, the method was deformed using as reference only the three network
vertexes located in the northeast, northwest and south limits of territory. In this case, it would have
been more convenient applying the least squares adjustment for computing the bilinear transformation.
However, errors around two meters were obtained for triangle inside (typical accuracy of the simple
methods). This fact confirms that the method in this way deformed achieves a similar accuracy than
some simple matrix methods. In Table 2 some results of the simple method, distortion analysis and
deformed method are presented.
Name

P001
P002
P003
P004
P005

ETRS89
01
41
02
40
03
40
04
38
06
38

18
30
15
22
12
01
12
55
16
15

12.435
19.123
33.239
18.355
17.211
53.544
22.221
12.258
10.362
11.343

ED50
(simple method)
W
N
W
N
W
N
W
N
W
N

01
41
02
40
03
40
04
38
06
38

18
30
15
22
12
01
12
55
16
15

07.967
23.200
28.720
22.577
12.611
57.824
17.567
16.675
05.535
15.875

W
N
W
N
W
N
W
N
W
N

ED50
(distortion analysis)
01
41
02
40
03
40
04
38
06
38

18
30
15
22
12
01
12
55
16
15

07.930
23.214
28.675
22.584
12.555
57.836
17.507
16.665
05.522
15.825

W
N
W
N
W
N
W
N
W
N

ED50
(deformed method)
01
41
02
40
03
40
04
38
06
38

18
30
15
22
12
01
12
55
16
15

07.922
23.182
28.645
22.536
12.513
57.787
17.441
16.639
05.340
15.885

W
N
W
N
W
N
W
N
W
N

Table 2: Some results obtained by deformed method.

Our main conclusion is that the presented method shows a simple and novel algorithm for
datum transformation that solves this problem using classical theories. When a DVN is dense then the
process is of similar accuracy order than methods based on distortion analysis. But if a DVN is poor
then it is of similar accuracy order than simple methods applied with a dense DVN. Therefore high
density of reference network is not required necessarily so that it can be applied for datum transformation in widespread zones with low density of observations.

-5-

4. References.

Applebaum, L.T. (1982). Geodetic datum transformation by multiple regression equations. Proceedings of the Third International Geodetic Symposium on Satellite Doppler Positioning. Las Cruces,
New Mexico, pp. 207-223.
Briggs, I.C. (1974). Machine contouring using minimum curvature. Geophysics, Volume 39, Issue 1,
pp. 39-48 (February).
Bugayevskiy, Lev M. and John P. Snyder. (1995). Map Projections: A Reference Manual. London:
Taylor and Francis.
Collier, P.A., Argeseanu, V., Leahy, F. (1998). Distortion Modelling and the transition to GDA94.
The Australian Surveyor, Vol. 43, No. 1, March, pp. 29-40.
De Berg, M, Van Kreveld, M., Overmars, M., Schwarzkopf, O. (1998). Computational Geometry.
Algorithms and Applications. Springer Verlag. Berlin, Heidelberg, New York.
NIMA (2004). WGS84 definition/relationships with Geodetic Systems. Technical Report, No. 8350.2,
National Imagery and Mapping Agency, Washington.
Draper, N.R., Smidth, H. (1981). Applied regression analysis. Wiley.
Gonzalez-Matesanz, J., Dalda, A., Quirs, R., Celada, J. (2003). ED50-ETRS89 Transition models for
the spanish geodetic network. Report on the Symposium of the IAG Subcommission for Europe
(EUREF). Toledo, Spain, 4 7, June. Vol. 13, IAG publications.
Higham, N.J. (1996). Accuracy and Stability of Numerical Algorithms. SIAM.
Lauf, G B (1983). Geodesy and Map Projections, Royal Melbourne Institute of Technology.
Quiros, R., Sobrino, J.A.S., Valds, M, Cano, M.A., Lozano, L. (2003). National Report of Spain.
Symposium of the IAG Subcommission for Europe (EUREF). Toledo, Spain, 4 7, June. Vol. 13,
IAG publications.
Smith, W.H.F., Wessel, P. (1990). Gridding with continuous curvature splines in tension. Geophysics,
Vol. 55, Issue 3, pp. 293-305.
Tapley, B.D. (1976). On the interpretation of least squares collocation. Long time predictions in dynamics. Reidel Publishing Co., pp. 165-172.
White, M.S., Griffin, P. (1985). Piecewise linear rubber-sheet map transformation. American Cartographer 12 (1985), pp. 123-131.
Zakatov, P.S. (1953). A Course in Higher Geodesy. Translated from Russian by the Israel Program for
Scientific Translations for the Office of Technical Services, U.S. Department of Commerce, Washington, D.C., U.S.A., 1962.

-6-