[PROJ] Alternate use model on EPSG v10+ datum ensembles

[Even Rouault]

> That's a good point, but the use of a datum ensemble to model this is quite
> surprising to me. According to ISO-19111:2019, a datum ensemble is a "group
> of multiple realizations of the same terrestrial or vertical reference
> system that, for approximate spatial referencing purposes, are not
> significantly different". One consequence of that would seem that all
> members of a same ensemble should have the same extent (otherwise if you
> use a 'random' member and its extent is not compatible with your point of
> interest, it is significantly different...). I'll raise the matter to EPSG.

Just got a very comprehensive answer from Roger Lott, from IOGP:
"""
The ISO 19111 definition "group of multiple realizations of the same
terrestrial or vertical reference system that, for approximate spatial
referencing purposes, are not significantly different" needs to be
understood in the context of significant changes to geodetic practice for
realizing height systems in the past few years.

Historically, vertical CRSs were realized by levelling networks, with the
measurements observed through optical instruments ('levels') with ability to
measure over quite short distances. The levelling networks were unable to
cross wide stretches of water. As a consequence most islands had their own
vertical datum (one or more local tide gauges) and vertical CRS. In modern
geodesy it is possible to get levels through satellite-based techniques.
These satellite-based techniques are very significantly more cost effective
than the old levelling networks. As a consequence, there is a move to
redefine the old levelling networks through a satellite-based solution. The
satellite-based solution is not constrained by wide stretches of water. So
one satellite-based model can be used to define several adjacent island
datums.

This is what has happened in the British Isles. One geoid model covers all
of the islands and now defines all of what were once separate, isolated,
vertical datums. But it is complicated because of the historic horizontal
CRSs of the island of Ireland, and the single geoid model is distributed in
two different files, one grid for all Ireland (Republic and Northern Island
which is part of the UK) and another grid for the rest of the UK and the
Isle of Man. Because of the existence of this single geoid model, the three
national mapping authorities covering the British Isles now consider the old
datums to be part of a single conceptual vertical reference system.

Something similar has happened in Australia. Historically, Australia Height
Datum of 1971 covered only the mainland, with separate height datums on
Tasmania, Cocos Islands and Christmas Island. These are all now considered
to be part of AHD. In this case the geoid model data is went through two
phases of consolidation - first mainland and Tasmania were integrated
(AUSGeoid09) and in 2018 through AUSGeoid2020 Christmas Island and Cocos
Island was added. 

So what is the difference between what has happened in British Isles
compared to Australia? Technically, they are the same - independent
levelling networks replaced by realization through a geoid model. But the
national authorities have taken different approaches to naming. In
Australia, Australia Height Datum (Tasmania) is no longer referred to as
such; since its realization changed from through a levelling network to be
through a geoid model, it is just integrated into the AHD. But in the
British Isles, the names of the old networks are retained. So we have
Ordnance Datum (Newlyn),  Ordnance Datum (Orkney), Ordnance Datum
(Stornoway), etc. (Here is another idiosyncrasy: Newlyn Datum is now
referred to as Ordnance Datum (Newlyn) but Stornaway Datum is still referred
to as that name, despite it legally being Ordnance Datum for the Outer
Hebrides islands).

Australia has a second vertical CRS (Australian Vertical Working Surface).
This will be updated from time to time as new gravity data is incorporated
into a new geoid model.  There is an argument for having an ensemble for
these various AVWS realizations (although Australia will probably not do
this, I expect them to keep the same datum and CRS and replace the geoid
model defining it). AHD is onshore only, whilst AVWS covers onshore and
offshore. The onshore areas are common. But AHD and AVWS can never be part
of the same ensemble, not because they do not cover the same area but
because they are realizations of different conceptual vertical reference
systems (VRS - not to be confused with vertical CRS).

And why no transformations associated with the British Isles vertical
ensemble? There is a series of (geoid model) transformations to each of the
separate islands, but they all use the same model. They are documented this
way to reflect national authority documentation. The same is true for Spain
and the Canary islands. Why a vertical ensemble for the British Isles and
not for the Canary islands? Because the national mapping agencies for the
British Isles requested it (to simplify teir product delivery), the national
mapping agency for Spain has not done so. 

So the vertical case is somewhat different to the horizontal / geometric 3D
case. But both reflect geodetic techniques and practices, not extents. There
is no requirement for members of an ensemble to have the same extent. The
requirement is for ensemble members to be realizations of the same concept.

As far as Proj issue 9980 [note from E. Rouault: issue 9980 is this discussion 
thread about NOAA VDatum] is concerned, I would be very wary about using the
ensemble construct to include a mixture of onshore geodetic vertical datums
and offshore tidal datums - these have different conceptual reference
systems. One models the geoid, the other a tidal surface. Although the geoid
is a surface which best fits MSL, they are not the same or even parallel
with each other.

Datums without CRS. In the vertical, here we have a terminology problem.
Very few people working in the vertical domain use the term 'CRS'.
Colloquially 'Datum' is often used without distinction between (i) an origin
location, (ii) a surface promulgated away from the origin, and (iii) height
or depth coordinates referenced to the surface. But in terms of the ISO
19111 data model, all gravity-related heights and depths can only be
referenced to a CRS, not a datum. There is no coordinate system in a datum.

Why does EPSG have a datum (5107) and no associated CRS? Once upon a time
there was a vertical CRS, 5707, which was associated with datum 5107. Both
were invalidated in 1997. That was before EPSG had introduced its
deprecation policy of never removing data, even if invalid. At that time, in
general invalid data was removed from the Dataset. This is what happened to
CRS 5707. Datum 5107 was retained, because GeoTIFFv1.0 (inappropriately)
referred to EPSG datum codes. Some time later, CRS code 5707 came to be
reused for totally unrelated data (the fact that old use and new use both
were French is coincidence). Our registry software automatically assigns
codes, and it has no record of codes that were deleted prior to the
introduction of our deprecation policy in 2001.
"""

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