[PROJ] "Trustworthiness" of vertical transformations

[Even Rouault]

On jeudi 28 novembre 2019 01:49:35 CET Chris Crook wrote:
> Am I right in understanding that where proj (6) determines two or more
> possible transformation paths between CRSs it uses the accuracy to decide
> which is preferable?

The long answer would be long :-) Trying to do a medium one. There are several 
levels of API. In the low level of API, PROJ enumerates the possible 
transformation paths from its database and a set of heuristics, and then 
return them in a sorted way, trying to put the most relevant at top. The 
criterion include area of use, accuracy, availability of grids. The user can 
then decide which one is appropriate for its use case
The higher level of API ("black box") will use those results to automatically 
select the one that is "best" for the coordinate to transform.
So yes, given 2 transformation paths that have the same are of use and where 
the needed grid(s) are available, advertized accuracy will decide which one is 
to be used.
I didn't mention deprecation and supersession information coming from EPSG 
which are also taken into account.

> Also geodetic agencies defining parameters will need to be careful that
> accuracies are configured to ensure the desired preference is achieved.  In
> the past it is likely that not much attention would have been given to this
> as it isn't necessarily very meaningful (certainly not as a single value)
> and until now possibly didn't make much difference to anything...

Yes, an example of the consequence of this in:
https://github.com/OSGeo/PROJ/issues/1580#issuecomment-528042944

Even

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