[CF-metadata] axis attribute

[Simon Wood]

Jonathan,

So to provide a concrete example, would the following correctly reflect 
what you are suggesting?

dimensions:
   time = 1;
   nx = 5008; ny = 5536;

variables:
   // coordinate variables
   float time(time);	

   float nx(nx);
         nx:standard_name = “projection_x_coordinate”;
         nx:units = “km”;
         nx:axis = “X”;		// ###

   float ny(ny);
         ny:standard_name = “projection_y_coordinate”;
         ny:units = “km”;
         ny:axis = “Y”;		// ###

   float lat(ny, nx);
         lat:standard_name = “latitude”;
         lat:units = “degrees_north”;

   float lon(ny, nx);
         lon:standard_name = “longitude”;
         lon:units = “degrees_east”;

   // observation data
   float sst(time, ny, nx);
         sst:standard_name = “sea_surface_temperature”;
         sst:coordinates = “lat lon”;
         ...

This makes the most sense to me, although I suspect it is not what 
others are suggesting...

cheers

Simon


Jonathan Gregory wrote:
> Dear All
> 
> I've argued so far in favour of restricting axis X and Y to 1D lon and lat
> coord variables. Now I'd like to argue the alternative of generalising them
> to be used with any 1D horizontal coord variables i.e. removing the exclusion
> that they can't be used for rotated lon and lat.
> 
> For 1D lon and lat, the axis attribute is still redundant while we adhere to
> COARDS, because such coord vars must be identifiable by units (and can
> optionally also be identified by standard name). However for other horizontal
> coordinates the axis attribute is not redundant if generalised. It tells you
> that this axis is a horizontal axis. If you had some non-spatiotemporal
> axes e.g. realization or land cover type, it might be useful to be told which
> of the axes are the horizontal coordinates, since this is not easy to deduce
> from the standard names without some intelligence. Is it useful to know?
> 
> I think the choice of X and Y is arbitrary. Both of them indicate horizontal,
> and the only difference is a preferred choice for plotting, I imagine. Rotated
> lon is not necessarily X, for instance, if the pole is tipped over a long way.
> 
> I would not favour allowing X and Y on 2D aux coord variables. I think that
> in cases where there are 2D horizontal aux coord variables, the X and Y
> designations still properly belong to the 1D axes e.g. the projection coords.
> Because these axes are thus designated, you can tell which 2D coord vars are
> horizontal; they have dimensions belonging to axes which are labelled with
> X and Y. Perhaps we should make it mandatory to attach standard_names of
> longitude and latitude to 2D auxiliary lon and lat variables so that they can
> be easily identified as such.
> 
> Cheers
> 
> Jonathan
> _______________________________________________
> CF-metadata mailing list
> CF-metadata at cgd.ucar.edu
> http://www.cgd.ucar.edu/mailman/listinfo/cf-metadata


-- 
Simon Wood
National Institute of Water and Atmospheric Research, Wellington, NZ	
simon.wood at niwa.co.nz
http://www.niwa.co.nz