[Qgis-user] Z($geometry), CRS and transforms: buggy?

[Greg Troxel]

The TL:DR: is that I believe

  Z($geometry) returns the Z value from the geometry in the layer,
  without regard to the CRS.  It *should* return the Z component  after
  transforming from layer CRS to project CRS.  (qgis 3.34.10)

The longer story:

I have been gathering data for generation of elevation contours, using
an RTK-capable GNSS receiver and QField.  I'm capturing into a table
with a fid and a PointZM geometry.  The elevation is corrected to base
of rod from APC. I then use the Contour plugin.  Generally this works
well.

My RTK network is in NAD83(2011) epoch 2010.0, EPSG:6319.  That's my
national datum and plate fixed.

Due to not keeping up with a server IP address change at my RTK network
provider, I booted up the system after a long time not using it, and
failed to get an RTK solution.  I realized this, but decided it was a
good experiment to obtain a point without RTK, and assess accuracy.  I
captured a point as the displayed accuracy (ublox) crossed 0.600m;
QField displayed 0.424m H and 1.100 m V.  Unsurprisingly, the delta from
record coordinates (derived from RTK, that I believe, and replicated to
about 6 cm an hours later), and the point was about 1.5m horizontally,
and 8.5 m vertically.

I then realized that I was comparing NAD83(2011) record location
coordinates with probably ITRF2008 (you tell me what frame WAAS uses,
with a footnote!), or certainly something very close to some recent ITRF.

To resolve the, I created a scratch layer, copied the non-RTK point into
it, and relabeled the layer as ITRF2014, EPSG:7912.  Labeling it WGS84
would have run into ensemble confusion and null transforms, but calling
in 7912 led to a transform from ITRF2014(=ITRF2008) to NAD83(2011), and
there was about 1.1 m shift, as expected.

However, the Z values displayed by my label expression:

  coalesce(name, '_') || ' M: ' || round(M($geometry),3) || ' H: ' || round(Z($geometry),3)

were the same.  Even though the transform surely changed those.


So, I used gdal to create a layer natively in 7912, added that, and set
the same label expression.  Now, the Z value is about 1.2 m lower.
(This makes sense; the ellipsoids do not have the same center.)  Thus I
am not posting this query to gdal/proj lists; I can't argue they did
anything wrong.

So, it looks like Z($geometry) just pulls the value before transform,
and I think it should be afterwards, just like X and Y are used to place
the point on the canvas.

Agreed, or am I off base?

If I'm wrong about Z and the behavior is right, how do I write a label
expression for the transformed Z?

Thanks,
Greg