# [CF-metadata] grid cells with a varying number of cell bounds

Karl Taylor taylor13 at llnl.gov
Wed Sep 6 10:53:43 MDT 2017

```Dear Jonathan,

I certainly always envisaged the possibility of cells of different
shapes, and we imply that this might be the case in the description of
cell bounds with the explicit mention of "maximum":

/"A boundary variable will have one more dimension than its associated
coordinate or auxiliary coordinate variable./ The additional dimension
should be the most rapidly varying one, and its size is the maximum
number of cell vertices."

I don't see any reason why the convention shouldn't provide for this.
As I recall (and a colleague of mine concurs), commonly used cube-sphere
grids have cells at the 8 "corners" that are a different shape from all
the other cells (and they have a different number of vertices).   Would
there be any reason not to explicitly say that for cells with fewer
vertices than the maximum, the missing_value should be used to fill up
the unneeded bounds elements?

Concerning missing values being permitted for "unused elements in
discrete sampling geometries", I agree when a variable is not formally a
function of time (as in examples H.3, H.16, and H.18, H.20, and H.21),
then time may have missing_values, but not when the variable is formally
a function of time as in H.4 and H.5, where time is actually a
coordinate variable.   I think in fact that the NUG forbids missing
values in coordinate variables.

best regards,
Karl

On 9/6/17 8:06 AM, Jonathan Gregory wrote:
> Dear Karl
>
>> Consider a grid consisting of both triangular and rectangular grid
>> cells with n total grid cells.  In this case the cell_bounds for
>> longitude could be dimensioned (n, 4) because there are a maximum of
>> 4 cell vertices.
>>
>> My question is, for the triangular grid cells do we *require* the
>> 4th element of the bounds to be set to missing_value?  If not, how
>> does software decide how many vertices each grid cell has?
> As Chris says, I don't think we envisaged a mixture of different sorts of
> polygons. The text at the end of sect 7.1 speaks of "p-sided cells" with
> "p the number of vertices of the cells". It doesn't suggest that p could
> be multivalued. We could clarify that with a defect ticket.
>
> The proposal by Dave Blodgett and others for simple geometries
> https://cf-trac.llnl.gov/trac/ticket/164
> does allow for a mixture of different polygons to define the cells of a grid.
>
>> Also, in H.2.3 examples H.4 and H.5, missing_value is defined for
>> the time coordinate, time(time).  I don't think coordinates are
>> allowed to include missing values.  Is this a "defect" in our
>> document?
> No. It's not normally allowed, but it is permitted for unused elements in
> discrete sampling geometries. See section 9.6.
>
> Best wishes
>
> Jonathan
> _______________________________________________