[PROJ] DerivedProjectedCRS — database support gap and proposed PR

[Ben Griffin]

Hi again Even, 

Thank you for the clear and constructive feedback — and for confirming that the general plan for DerivedProjectedCRS in the database is sound. That's the core of the PR and it's good to have your endorsement of the approach.

I'd like to separate two things your reply touches on, because I think they're independent. 

The first is the PR itself. The CS in all the examples is plain 2D Cartesian — no face identifier appears as a coordinate axis. The derived_projected_crs table would register planar expressions of a projected CRS as distinct entries, in the same way UTM zones are registered as distinct ProjectedCRS entries: As you point out, a face or zone selector is a CRS selector, not a coordinate axis. 
As an illustration: if one wanted to apply UTM-style zoning to a Lambert Conformal Conic, the natural expression would be a DerivedProjectedCRS derived from the base LCC — with the zone identity captured in the CRS registration and the output CS remaining plain 2D Cartesian. 

The second is whether a face identifier could legitimately appear as a coordinate component in ISO 19111. Your point about the absence of a mixed linear/ordinal CS type is well taken, and I would like to discuss that separately if you would be so kind — but I don't such issues block the PR which doesn't depend on that interpretation.

One observation in passing: The challenge in finding a clean registered example of this pattern may itself reflect the gap — without a derived_projected_crs table, these CRS types have had no correct home, and have either been approximated via text_definition or not registered at all.

Best regards,
Ben


> On 27 Mar 2026, at 13:50, Even Rouault <even.rouault at spatialys.com> wrote:
> 
> Ben,
> 
> the general plan if we wanted to introduce support for DerivedProjectedCRS in the database is sound, but I still don't think that using that concept for polyhedral projections with a face identifier is ISO-19111 compliant. AFAICS, there is no Coordinate System subclass that allows mixing linear axis (X,Y) and an ordinal one (face number). In particular a Cartesian CS is "a coordinate system in Euclidean space which gives the position of points relative to n mutually perpendicular straight axes all having the same unit of measure".
> 
> Even
> 
> Le 27/03/2026 à 12:13, Ben Griffin via PROJ a écrit :
>> Hi all,
>> 
>> Following the earlier thread on face/octant coordinates and the UTM zone analogy, I've been investigating whether DERIVEDPROJCRS could be used to formally express planar net projections as derived from  intermediate octahedral ProjectedCRS […]