Unclear interpretation of CF trajectories

[d70-t]

I am wondering how the definition of trajectories in CF conventions is to be interpreted. Here I am referring to the CF 1.7 standard, as that seems to be new newest one supported by the compliance-checker, but I think the same applies to at least CF-1.6 ... CF-1.9.

The standard says the following about dimensions of trajectory features:

data(i,o), x(i,o) y(i,o) t(i,o)

and

However a featureType may also include other space-time coordinates which are not mandatory (notably the z coordinate).

From these I read that data, x, y and t must be present and must be exactly along (i, o). Also z may be present, but no requirements on z's dimension are posed.

On the contrary, the checker assumes that z is also along (i, o) (if z is present). I am wondering if that is a correct interpretation of the CF conventions.

I have a current example of Radiosoundings which are reachable via opendap at https://observations.ipsl.fr/thredds/dodsC/EUREC4A/PRODUCTS/MERGED-MEASUREMENTS/RADIOSOUNDINGS/v3.0.0/level2/EUREC4A_BCO_Vaisala-RS_L2_v3.0.0.nc. This example does not validate according to the checker due to the failing is_cf_trajectory check, because the coordinate_dimension_matrix is the following:

In [5]: compliance_checker.cfutil.coordinate_dimension_matrix(ds)
Out[5]: 
{'x': ('sounding', 'alt'),
 'y': ('sounding', 'alt'),
 'z': ('alt',),
 't': ('sounding', 'alt')}

To my interpretation, this should be a valid CF Trajectory and that representation is also useful, as the intention of the dataset explicitly is to provide data which is resampled on a uniform altitude grid.

[DWesl]

There's a few examples in the CF document that allow storing CF features coordinates with fewer than the required dimensions: the time series profile section of the appendix has the closest example I can find.