[CF-metadata] Vertical datums (again)

From: Jonathan Gregory <j.m.gregory>
Date: Mon, 10 Feb 2014 18:30:15 +0000

Dear Jim and Rich

Many thanks for your helpful comments. I see a prospect of my understanding
things a bit better than before!

Jim says that a vertical datum always has a reference ellipsoid. Sometimes a
vertical datum might *be* a reference ellipsoid. Sometimes it is a geoid, and
in that case, is it accompanied by a reference ellipsoid as part of the
definition of the vertical datum?

Rich comments that a vertical datum could be orthometric. If I've understood
Jim correctly, orthometric describes how you measure the height wrt the
reference surface. It is not a third type of surface, in addition to geoid
and reference ellipsoid. Is that right?

Tides define a different sort of reference surface from geoid and ellipsoid.
Are there also vertical datums which involve tidal levels in their definition?

> why can't we just say
> "sea_surface_height_above_datum" or just "sea_surface_height" and then
> specify the vertical datum, no matter what it is?

I don't think we should do so because height wrt geoid and height wrt ellipsoid
are rather different quantities. For that reason they have different standard
names (altitude and height_above_reference_ellipsoid, and there is also a
standard name of geoid_height_above_reference_ellipsoid). They are seriously
different in value, aren't they? - by 100s of metres, so you have to know which
one you are dealing with. If they had the same standard name, a height wrt
geoid from one data source and a height wrt ref ellipsoid from another might
be regarded as comparable quantities, which could be a serious error. Of course
I recognise that the stdname is not the only metadata one should consult, but
it is the first point of call.

To make an analogy, suppose we just defined height as "vertical distance above
something", with something defined elsewhere. Then altitude and height above
sea floor would be synomymous standard names. I don't think that would be as
helpful to the data-analyst.

I do think, however, that it's acceptable to define the geoid or reference
ellipsoid in another place (the grid mapping) from the standard name. This is
still a risk, because heights on different vertical datums might be treated as
comparable they aren't, but on the other hand there are cases where heights on
different vertical datums could be compared e.g. if they come from models with
a different shape for the Earth.

We can meet Rich's need, I think, if we provide a way for the grid_mapping to
specify vertical datums which involve the geoid being implied.

Best wishes

Jonathan
Received on Mon Feb 10 2014 - 11:30:15 GMT