[gdal-dev] New OGC standard about WKT for projections

[Even Rouault]

Selon Andre Vautour <andre.vautour at caris.com>:

> The way I see it, the new spec does clear up some big holes like:
>
>  1. Being able to specify an unambiguous operation/projection method (by
>     identifier) for projections instead of loosely relying on the
>     projection's name (9.3.3)
>  2. Knowing the units of projection parameters (9.3.4)

Yes, there are of course interesting improvements. I'm often playing the devil's
adovcate.

>
> The transformation is not part of the coordinate reference system in
> both the EPSG registry and the ISO 19111 specification. What was weird
> with the older WKT spec is that you could have an identified CRS that
> had TOWGS84 specified, even though that definition is technically not
> part of the identified object. That being said, for a lot of practical
> applications, the transformation is essential, so it'll be interesting
> to see how it all plays out.
>
> > 18.1 Bound CRS
> > The definition of a CRS is not dependent upon any relationship to an
> > independent CRS. However in an implementation that merges datasets
> > referenced to differing CRSs, it is sometimes useful to associate the
> > definition of the transformation that has been used with the CRS
> > definition.
> >   AFAIU, the BOUNDCRS concept which can embrace the
> > concept of TOWGS84 through a Coordinate Frame method is to be used when the
> > transform to the TARGETCRS has already been done and document what was the
> > SOURCRS and the transformation applied.
> Even, I'm not sure if I completely understood that last part. It's
> entirely possible that what follows is what you were trying to say. If
> that is the case, fell free to just ignore me. :)
>
> The way I understand the BOUNDCRS concept is that it binds a
> transformation to a given CRS. That is, even though the transformation
> is not part of the CRS, you can use that construct to associate the
> transformation with a CRS (as it was previously done with TOWGS84). So,
> for a TOWGS84 equivalence, you'd have a BOUNDCRS where the source is the
> CRS being defined, the target would be WGS 84, and the operation method
> would be a 7 parameter geocentric transformations with the rotation
> convention (position vector/coordinate frame) explicitly defined.

Hum, apparently I didn't understand the spec the way you explain it (I wouldn't
bet much on the correctness of my interpretation), although I would prefer your
interpretation to mine. My understanding was that if you have a BOUNDCRS the CRS
that would apply would be the TARGETCRS and you would know that the data
originaly was in SOURCECRS (this comes from how I understand the sentence "it is
sometimes useful to associate the definition of the transformation that has been
used with the CRS definition."). But you seem to understand it the other way,
i.e the SOURCECRS would still apply ?

Anyway it doesn't seem likely that most data would come with a BOUNDCRS
expliciting the tranform to be used to go to WGS 84, so that would mean that
GDAL for example would have to query it from its own EPSG database. Actually,
that something we should ideally already do for example with the ESRI WKT of
.prj files, since they do not have TOWG84 nodes.

Even


-- 
Spatialys - Geospatial professional services
http://www.spatialys.com