[CF-metadata] The "grid" attribute seems to solve both unstructured and staggered grid issues

From: Jonathan Gregory <j.m.gregory>
Date: Thu, 25 Nov 2004 16:23:14 +0000

Dear All

I support Brian's proposal as well. Also following a comment of Steve, I think
it is better to make this kind of information subsidiary by putting it in a
container variable, since we are likely to need a number of new attributes for
it, just as with the grid mapping.

Minor comments:

(1) I think a more explicit attribute name than "grid" should be used to point
to the container variable. What about grid_topology? Would that be a correct
description of the intention of it?

(2) The attribute names horizontal_triangular_element_incidence_list and
boundary_segment_node_list should be all lower case, like other attributes.

As John says, we would need precise definitions of these to write into the
convention.

I agree with Steve's and Karl's caution, and I don't think we should try to
solve different problems with one mechanism. The grid topology (or whatever)
is about the construction of the grid of the data variable. The grid mapping
is about relating the grid points to latitude and longitude. The formula terms
relates vertical levels to physical vertical coordinates. Like all of of CF
which isn't in COARDS, all these features are optional.

The *relationship* between grids is yet another problem, different from all the
above because it relates different data variables. If we need a way to record
this, it should be separate, but Steve's experience suggests that it is hard
to deal with.

Cheers

Jonathan
Received on Thu Nov 25 2004 - 09:23:14 GMT