[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[netCDF #JNE-101118]: limit digits after decimal point in file

Hi Jerry,

> I am converting an ascii dataset that has float values, into a netcdf file.
> I am facing an issue I try to solve, that is to limit the number of digits
> after the decimal point, as I have only 4 significant digits scientifically.
> If I choose to use Float as the storage unit, the conversion process adds
> some digits to the end.
> 
> my questions 1) is it possible to limit the number of digits after the
> decimal point  for "float" or "double" values in the file, 2) if not, is
> there a standard attribute to indicate the significant digits are 4 -- I
> have seen something like c_format and f_format attribute for c or fortran
> based client, but is there one attribute that indicate the this property
> when the nc file is accessed?  By the way I know I could do "compact" and
> use integer and scale factor to achieve that, but I want to know if there is
> a general solution for floating point data.

There's nothing in the netCDF library that will change the number of 
significant 
values used to store a value, and the values are stored in binary, rather than 
as 
digits.

The C_format and FORTRAN_format attributes are defined in the Attribute 
Conventions
section of the NetCDF Users Guide:

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

as indicating to applications how many digits of precision to display, but
whether applications use that information is up to the application.  The 
"ncdump" application honors the C_format attribute and only displays the number 
of
digits specified.

It's also possible to use the C pow(), floor(), and round() functions to modify
a floating-point number to limit the number of significant digits it represents
using some arithmetic, but numbers like "0.1" can't be exactly represented in
binary, so your intention may still be thwarted.

If you want to limit the precision with which the data is stored, the only way I
know to do that is as you have pointed out, by packing the data and indicating
the packing parameters with the "scale_factor" and "add_offset" variable 
attributes, as discussed in the same section.

--Russ

Russ Rew                                         UCAR Unidata Program
address@hidden                      http://www.unidata.ucar.edu



Ticket Details
===================
Ticket ID: JNE-101118
Department: Support netCDF
Priority: Normal
Status: Closed