The WRF model has an idiosyncratic way of storing coordinate information in its netCDF output files, such that one needs WRF specific software to display WRF data. (Note that some users just deal with the GRIB output files from the WRF post-processor, which is a different animal).
Rich Signell made an NcML file that makes WRF files CF compliant. (See here for more info). What Rich is doing is adding coordinate system information, using CF attribute conventions. With these modification any CF-aware software can geolocate the data and, for example, display it with a map background, or overlay different data sources together (an IDV specialty).
The CDM has a Coordinate System Builder plug-in layer to identify coordinate systems, and it has such a plug-in for WRF netCDF output files. So any application that uses the CDM (like IDV and ToolsUI) can geolocate WRF data without the need for the CF modifications.
Rich's NcML uses the 2D lat/lon coordinates that are in the WRF output file. The CDM synthesizes projection coordinates using the projection parameters that are stored in the global attributes. Here are pictures of the two results:
Figure 1: Lat/Lon Coordinates (above)
(Click to enlarge)
(Click to enlarge)
Figure 2: Projection Coordinates
(Click to enlarge)
(Click to enlarge)
The first figure is on a lat/lon coordinate system, and one can see that the data covers a curved region in lat/lon coordinates. The second uses the data projection and therefore covers a rectangular region on that projection. We call this the natural projection of the data. It's often (but not always) also the coordinate system in which the numerical model does its computations. Please note that the data in the two figures is identical, we are just seeing differences in the way it is displayed.
The WRF group has expressed interest in making WRF output data CF compliant. Which way should they do it, with 2D lat/lon coordinates, or with 1D projection coordinates? If it was up to me, I'd use the 1D projection coordinates, because the lat/lon coordinates are derived from the projection coordinates, and because having the actual projection gives you more information than just having the lat/lon values of your grid. (One can also draw pictures faster when the pixels are square). To be CF-compliant, however, one has to have lat/lon coordinates, even if they are 2D, so that software that doesnt know how to deal with the projection can still work. So the best thing to do would be to add both.
