[MetaCRS] Coordinate offset after transformation from 4267 to 4326

[Martin Desruisseaux]

Hello Jody

The NADCON grid files are from NAD27 to NAD83. The coordinate
transformations that Betsey wanted to do are from NAD27 to WGS84, which
is considered as a different operation in the EPSG database. This
difference was often ignored because we were used to consider NAD83 as
equivalent to WGS84. This was approximatively true 20 years ago, but
there is now a difference of about 1.5 metres between NAD83 and latest
realizations of WGS84.

The key point is that GeoTools will not necessarily provide the same
transformation for NAD27 to WGS84 than for NAD27 to NAD83. It depends on
what GeoTools finds in the EPSG database, and how the referencing module
is designed. Last time I touched to the GeoTools referencing module, it
had a bug: when it found more than one operation for the same sourceCRS
and targetCRS, GeoTools unconditionally selected the most accurate one,
regardless its area of validity. This explain why the TOWGS84 parameters
are for Cuba: this is the operation with the smallest "positional
accuracy" found in the EPSG database, but this accuracy is true only in
a relatively small geographic area.

The above bug exists (or existed) because at the time I wrote the
referencing module, I though that a sourceCRS and a targetCRS were
sufficient for uniquely identifying a coordinate operation. I didn't
realized that we can have as much as 80 different operations for the
same source and target CRS depending on the geographic area. The fix for
GeoTools would be that the CRS.findMathTransform(...) method takes
another argument, which is the geographic area where the user intend to
apply the coordinate operation. This problem has been fixed in Apache
SIS (among other issues discovered as I continue to learn new aspects).

Even if GeoTools uses the grids, the grids are not the same for USA than
for Canada, so a hint about geographic area is still desirable. For
Texas, if the desired accuracy is 1 metre then there is two possible
operations depending on whether the points are on the west side or the
east side of 100°W: EPSG:8624 and EPSG:8625. If an accuracy of 10 metres
is sufficient then the TOWGS84 parameters for USA given in my previous
emails should be okay.

As a side note: the above also explain why the TOWGS84 element in WKT
has been removed in the new WKT 2 specification (ISO 19162): the TOWGS84
element can not handle the real-world complexity and is actually
dangerous since it can be misleading when given without information
about its domain of validity, like what happened with the TOWGS84
parameters for Cuba.

    Martin


Le 26/05/16 à 12:39, Jody Garnett a écrit :
> G'Day Martin:
>
> The GeoTools has been configured with GridShft files for NAD27 - this
> is not something I have personally tried before so I am not quite sure
> how it works.
>
> I had a look with Betsey, the data being projected is in Texas (so
> slightly different from Cuba so the TOWGS84 parameters sound legit).

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/metacrs/attachments/20160526/64c8f722/attachment.html>