Count parameters are stored as floats (NC_FLOAT); the range of NC_FLOAT values is too small. The proposal is to:
Store count values in 16 or 32-bit unsigned integer (NC_USHORT or NC_UINT) instead of 16-bit floating point (NC_FLOAT)
Use the standard NC_FILL_USHORT or NC_FILL_UINT as fill value.
As an example, the parameters RAWDOWNWELLING* and RAW_DOWNWELLING_PAR range of valid values is greater than the NC_FLOAT range. We should allow NC_UINT so that we can address [0 : (2^32)-1] = [0 : 4,294,967,295].
https://github.com/euroargodev/ArgoNetCDF/issues/1Status : ready, urgent
The “number of” parameters should also be stored as 16 or 32-bit integer, in NC_USHORT with NC_FILL_USHORT as fill value:
tcarval
commented
2 years ago
Count parameters are stored as floats (NC_FLOAT); the range of NC_FLOAT values is too small. The proposal is to:
As an example, the parameters RAWDOWNWELLING* and RAW_DOWNWELLING_PAR range of valid values is greater than the NC_FLOAT range. We should allow NC_UINT so that we can address [0 : (2^32)-1] = [0 : 4,294,967,295]. https://github.com/euroargodev/ArgoNetCDF/issues/1 Status : ready, urgent
The “number of” parameters should also be stored as 16 or 32-bit integer, in NC_USHORT with NC_FILL_USHORT as fill value: