Dear all

Before the meeting yesterday I was arguing, like Karl @taylor13, to retain the 
presumption that if the `grid_mapping` and `crs_wkt` are inconsistent, the 
data-user should presume the `grid_mapping` is correct. I was persuaded by the 
discussion that this isn't generally helpful, because the data-user might well 
think it was unsafe to proceed on that basis. In particular, the data-user 
might be aware that the WKT information had come first, and therefore suspect 
the `grid_mapping` of being an incorrect translation. Karl and I had argued 
also that the presumption of `grid_mapping` being correct gives an extra 
incentive to the data-producer to make sure the two are consistent. However, 
this also probably doesn't work; if the data-producer had thought about the 
consequence of misinterpretation, they would have tried to avoid inconsistency 
anyway.

Therefore I support the change to remove this assumption, and state that the 
metadata is invalid if `grid_mapping` and `crs_wkt` are inconsistent. In 
response to Karl, I agree with Alan @snowman2 that this is an error, and the 
file is not compliant, because the convention states the two kinds of metadata 
must be consistent.

Unfortunately, the CF checker won't be able to detect this error unless we 
write down the mapping between `grid_mapping` attributes and `crs_wkt` in the 
conformance document (or some document it can refer to). To make the check, you 
have to be able to interpret both kinds and compare them. One of Alan's 
concerns was that data-users felt they had to do that. We agreed that they 
don't. They can read one or the other, assuming they agree. However, it would 
be good if this could be checked routinely. As I argued before, I strongly feel 
that it would improve the convention if we could write down the equivalence. 
Note that we don't have to consider _all_ aspects of WKT, but only those 
aspects which people want to write in CF-netCDF files. Doing this would cause 
us to identify what has to be added to `grid_mapping` to give it the required 
capabilities, and whether there are inconsistencies between the CF data model 
and WKT. I wouldn't be surprised if there are, and we need to know, because 
it's not safe to treat `crs_wkt` as a black box if it might conflict with other 
CF metadata (not just in the `grid_mapping`, but for instance in the units and 
standard name of coordinate variables).

I would be concerned about adopting Philip @cameronsmith1's suggestion, because 
I fear that might lead to data-producers not being so careful with one or the 
other of the representations, thinking that they could set the flag to indicate 
it's not to be trusted.

Jonathan

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/cf-convention/cf-conventions/issues/222#issuecomment-643267858
This list forwards relevant notifications from Github.  It is distinct from 
[email protected], although if you do nothing, a subscription to the 
UCAR list will result in a subscription to this list.
To unsubscribe from this list only, send a message to 
[email protected].

Reply via email to