[Proj] Problems using grid shift files

[Mikael Rittri]

Paul Kelly wrote:

> On Wed, 1 Apr 2009, Ziegler Stefan wrote:
> 
> > Hi
> >
> > thanks for your answers and sorry for not replying for approx. 2
months.
> > I will now spend some time on my problems with grid files. I thought

> > first that the shift grid files can be used for any reprojection
(eg.
> > 21781 to 32632). But it's only usable for one specific reprojection 
> > (defined in the file header). The reprojection is more an
equalization 
> > than a "simple" reprojection since the old reference frame 
> > (CH1903/LV03) is inaccurate. By adding the +nadgrid parameter I'm
able 
> > to reproject correctly only to one reference system. Reprojections
to 
> > other systems are wrong.
> 
> Remember that we are slightly abusing the PROJ.4 API here by 
> using CH1903+ as a pivot datum instead of WGS84. So the +towgs84 
> parameters for other co-ordinate systems must be adjusted so they 
> describe a transformation to CH1903+, rather than a transformation 
> to WGS84, in exactly the same way CH1903+as we have used parameters 
> of +towgs84=0,0,0 for CH1903+. If you have an accurate set of +towgs84

> parameters for the CH1903+ --> WGS84 transformation, I suspect you
might 
> be able to determine the parameters for your target datum to CH1903+ 
> through some matrix multiplication (bear in mind that a set of
+towgs84 
> parameters consist of rotational, translational and scaling
components)
> but I'm not sure of the exact nature of the calculations you would
need
> to do and I haven't tried it.
> 
> In other words, you should be able to avoid a chain of re-projections
by 
> chaining the datum transformations before doing the re-projection.
That's
> how I understand it anyway but I could have missed something.
> 
> Paul

That's clever. I didn't think of that, but it sounds right. 
The currently recommended transform from CH1903+ to WGS84 (or to ETRS89,

technically) is actually a 3-parameter transform, for some reason: 

    +towgs84=674.374,15.056,405.346

EPSG coord. transformation code 1676 or 1647.  When the target is
regarded 
as ETRS89 rather than WGS84, then EPSG says the accuracy is 0.1 meter.
(Otherwise the continental drift makes the accuracy only about half a
meter.)  

This should make it easier to follow Paul's advice.  I suspect these 
three numbers should be added to the corresponding dx,dy,dz of all other
transformations, in order to make CH1903+ the pivot datum. Or possibly 
subtracted, of course...  (But I am not really certain, to be honest.) 

Best regards,
--
Mikael Rittri
Carmenta AB
SWEDEN
www.carmenta.com