[MetaCRS] Motion Vote: CS-Map RFC 2 - Redesign Datum Transformation Engine

[Martin Desruisseaux]

Hello Frank

Le 07/07/10 17:39, Frank Warmerdam a écrit :
> I wasn't thinking the multiple shifts could be reduced to a single Bursa-Wolf
> transformation, but rather that each datum shift would still occur by passing
> through WGS84.

Yes I understood that. What I was trying to said (maybe badly) is that the 
result can be mathematically different, unless you can ensure that the matrix 
meet some conditions.

To follow on the made up example, if the EPSG database defines two datum shifts:

   (1) Estonia 1992 to ETRS89
   (2) Estonia 1992 to WGS84

Then, considering two possible transformation paths:

   (1) Estonia 1992 to ETRS89
   (2) (Estonia 1992 to WGS84) followed by (WGS84 to ETRS89)

it may be mathematically impossible for (1) to give the same result than (2). 
However it is always possible for (2) to produce the same result than (1), but 
in order to simulate the numerical result of (1) you may need to use different 
coefficients then the one declared in the EPSG database for (2). I don't know if 
it would happen often (e.g. you will not need to mind for every datum shifts 
containing only translation terms) - but it is a possibility.


> So for a rather made up example lets assume there is a translation from
> Estonia 1992 to ETRS89 and a transformation from ETRS89 to WGS84, I would
> like to be able to do the datum transformation from Estonia 1992 to ETRS89
> by going from Estonia to ETRS89 to WGS84 and then back to ETRS89.

I see no mathematical downsides with this approach.

	Regards,

		Martin