RE: redo striped.nc

• To: <Simon.Cox@xxxxxxxx>, <caron@xxxxxxxxxxxxxxxx>, <galeon@xxxxxxxxxxxxxxxx>
• Subject: RE: redo striped.nc
• From: "Woolf, A (Andrew)" <A.Woolf@xxxxxxxx>
• Date: Wed, 10 Aug 2005 19:53:22 +0100

Simon,

Your suggestion is essentially what we do in CSML for non-rectified grids:
- inherit from gml:Grid
- add an element to specify the (non-uniform) axis values at grid nodes for 
each CRS axis
- allow any CRS axis to span one or more grid dimensions (grid axes not 
necessarily aligned with - or affinely related to - CRS axes)

The last allows, for example, a two-dimensional grid to be embedded within 
three-space dimensions. Or - more typically - a two-dimensional grid to be 
arbitrarily placed within horizontal space. Numerical climate models often have 
grids with non-affine relationship to lat-lon (or any other standard) CRS.

Regards,
- Andrew

That's correct - the definition of the RectifiedGrid element follows ISO 19136, which specifies grids with constant spacing along each axis. There is just one offsetVector per dimension. The ISO 19123 ReferenceableGrid is a more general object, which maps grid coodinates (i.e. integer coordinates) to an external corodinate reference system, but it is defined in terms of operations only, with no limitation on how it might be implemented (function, lookup, etc). For obvious reasons, this is not implemented in GML

In order to specify a regular grid with non-constant offsets, a new specialization of CV_Grid/gml:Grid would be required. Not hard conceptually - just replace the single offsetVector on each axis with an ordered list of vectors. (Of course there are various kinds of constraints you might want to apply - e.g. are all vectors on one axis are parallel, or not?)

Simon