[Proj] Datum shift for Dutch maps 1850-2000

[Jan Hartmann]

Yes, an automatic datum shift to WGS84 seems to be causing the trouble, 
but I don't see the system in this madness:

Test point 9018/375648  (Dutch RD, EPSG:28992 without WGS84 datum shift)
The source PROJ string is:
+proj=sterea +lat_0=52.15616055555555 +lon_0=5.38763888888889 
+k=0.999908 +x_0=155000 +y_0=463000 +ellps=bessel +units=m +no_defs.
I tested this string both without and with +nadgrids=@null on four 
target projections:

1) to latlon Bessel (+proj=longlat +ellps=bessel +no_defs)
     --> (without nadgrids in source) 3.291612 51.352060
     --> (with nadgrids null in source) 3.291612 51.351476
2) to latlon Bessel with nadgrids null (+proj=longlat +ellps=bessel 
+no_defs +nadgrids=@null)
     --> (without nadgrids in source) 3.291612 51.352643
     --> (with nadgrids null in source) 3.291612 51.352060
3) to latlon Bessel on the WGS84 datum (+proj=longlat +ellps=bessel 
+no_defs +datum=WGS84)
     --> (without nadgrids in source) 3.291612 51.352060
     --> (with nadgrids null in source) 3.291612 51.351476
4) to latlon on the WGS84 datum (+proj=longlat +no_defs +datum=WGS84)
     --> (without nadgrids in source) 3.291612 51.352643
     --> (with nadgrids null in source) 3.291612 51.352060

1) and 3) give the same results; so do 2) and 4), which seems to inicate 
that an automatic datum shift to WGS84 takes place, even when not 
requested. I'm not sure about the effects of +nadgrids=@null. 
Specifying +ellips=bessel without adding +nadgrids=@null seems to cause 
an automatic WGS84 datum shift, whether or no datum=WGS84 is specified 
(1a = 3a). If +nadgrids=@null is specified in the the source projection, 
but not in the target projection (1b = 3b), the latlon coordinates shift 
about 65m to the south. When +nadgrids=@null is omitted in the source 
projection, but added in the target projection, the resulting 
coordinates are shifted of 65M to the *north* (2a). Finally, when 
+nadgrids=@null is specified in both source and target projections, the 
datum shift is applied nevertheless (2b is equal to 1a and 3a). Finally, 
specifying a WGS84 datum shift for the target projection without the 
Bessel ellipsoide gives the same values as 2. I don't understand this, 
but filed it as a bug nevertheless (http://trac.osgeo.org/proj/ticket/22).

Finally I would like to quote Frank Warmerdam 
(http://article.gmane.org/gmane.comp.gis.gdal.devel/17261)

<quote FrankW>
       Grr.  I'm so sick of coordinate system related issues!
</quote FrankW>

I agree :-)

Jan


Frank Warmerdam wrote:
 > Jan Hartmann wrote:
 >> This certainly makes a difference! Adding +nadgrids=@null to the 
original definition gives a map with a 120m South shift, still too much 
but nearing a solution.  Isn't this a bug in PROJ? I did not request a 
datum shift, and neither the from-projection, nor the to-projection had 
a datum defined, only ellipsoids. I am using Proj 4.6.1.
 >
 > Jan,
 >
 > An ellipsoid/geocentric based datum shift should not be done if no 
towgs84
 > or nadgrids value is provided with 4.6.0 or later. If you believe 
there is
 > a bug please file a detailed ticket showing exactly what you were doing.
 >
 > Best regards,