Open znichollscr opened 3 months ago
I will rely on @durack1 to answer this drawing on his experience from CMIP6.
@durack1 any thoughts here?
The bounds are required for CF, and they should be the same type as the coordinate type, int64 seems a little quirky to me, these are normally floating point/float
The bounds are required for CF, and they should be the same type as the coordinate type, int64 seems a little quirky to me, these are normally floating point/float
Ok I'll change sector to floating point so then the bounds are the same
Feedback from @vnaik60:
I'm simply using the cf-xarray library to do this, specifically here. I assume they are following some cf convention hence why it is now bounds rather than bnds. If it is meant to be bnds, I can make a PR into that library.
I believe that "int64 bounds(bounds)" denotes that the bounds variable is a 64-byte integer whose only dimension is bounds. I think the fact that this was missing in the CMIP6 data was a data writing error. As far as I understand, all the variables (including dimensions) are meant to appear in the file's metadata/variable section, but maybe this is just how xarray does it and isn't actually required.