This issue was of interest to us at the NOAA-CIRES Climate Diagnostics
Center, when we went about storing NCEP-DOE Reanalysis 2 data in NetCDF
form. We wanted to be CF compliant, which said, to us at least, that
valid_range had to match the packed type. We liked it being of the
unpacked type, for purposes of human readability, and that is how we had
done it in previous (COARDS-compliant) data sets. Our compromise was
to render valid_range in the packed type and add the attribute
"unpacked_valid_range" to handle the more human-readable form. We felt
this way we could satisfy the needs of coders who wanted valid_range
as another way of determining "missing-ness", while maintaining a more
human-readable valid_range for users who weren't writing such applications.
-Hoop
