[netCDF #UPA-771360]: netCDF: Representing floating point with Compressed Integer?

Hi,

> If I want to reduce the data volume by representing floating point values 
> using 'compressed integer'(?), does netCDF have an API for that or would I 
> have to that myself before writing to the array to the netCDF4/HDF5 file?

There is no API for packing floats or doubles in smaller integers 
in the netCDF C or Fortran libraries, but there is a convention

  http://www.unidata.ucar.edu/netcdf/docs/netcdf/Attribute-Conventions.html

for using the "scale_factor" and "add_offset" attributes to 
implement packing. A detailed description of how to handle 
this convention is in the Best Practices document here:

  http://www.unidata.ucar.edu/netcdf/docs/BestPractices.html#Packed Data Values

The netCDF-Java API implements packing and unpacking according to 
the above convention automatically.  Also the NCO (NetCDF Operators)
package implements packing according to that convention:

  http://nco.sourceforge.net/nco.html#Packed-data

> Would it be the same for both classic & enhanced models?

That simple packing convention will work for both classic
and enhanced models, but it's not implemented in the C or
Fortran libraries. The netCDF-4 library implements zlib
compression as well, for both netCDF-4 and netCDF-4 
classic-model formats, and that can sometimes achieve 
better compression than packing.

--Russ

> Thanks,
> 
> 
> ------------------------------------------------------------------------
> Oh-ig Kwoun
> 
> NASA/JPL, California Institute of Technology
> 4800 Oak Grove Drive, MS-T264-867
> Pasadena, CA 91109-8099, USA
> (Cell) (818) 399-6998
> Tel:   (818) 354-3496
> Fax:  (818) TBD
> ------------------------------------------------------------------------
> 
Russ Rew                                         UCAR Unidata Program
address@hidden                      http://www.unidata.ucar.edu



Ticket Details
===================
Ticket ID: UPA-771360
Department: Support netCDF
Priority: Normal
Status: Closed