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

[netCDF #BCN-163009]: Best practice for encoding time

Hi David,

> I would like to encode data input and model output time values in a
> netCDF file consistent with ISO 8601 methods for specifying time, such as
> 2012-03-31T11 representing 11 am on March 31
> 2012-03-31T11.5  representing 11:30
> 2012-03-31T12  representing noon
> 2012-03-31T12.5
> 2012-03-31T13
> 2012-03-31T13.5  representing 1:30 pm
> etc
> 
> What is the best practice for encoding these in the time dimension
> values of a netCDF file.  (We will be reading and writing these files
> using the Fortran interface
> (http://www.unidata.ucar.edu/software/netcdf/docs/netcdf-f90.pdf)

To be CF-compliant, you should instead encode the times using units that
are supported by the udunits library, as described here:

  
http://cf-pcmdi.llnl.gov/documents/cf-conventions/1.6/cf-conventions.html#time-coordinate

Then if you use ncdump to look at the time values with the "-T" option, it 
shows them as ISO-8601 strings.

For example, if you have time values stored in a variable named "time" that
uses the dimension named "time" and has units:

        time:units = "seconds since 2009-01-01" ;

and you just use "ncdump" to see the time values, they might look like

        time = 7776000, 15597000 ;

but if you use "ncdump -T" instead, they will be shown as

        time = "2009-04-01", "2009-06-30T12:30" ;

There's no option to make the hours and minutes appear as "12.5" instead of
"12:30", but I think the latter is ISO-8601 standard.  Unfortunately,
the ncgen utility can't yet convert ISO-8601 strings to udunits quantities,
but there are other packages that do, e.g. the Python datetime module.

If you decide to store time as strings rather than as numeric quantities 
that can be used in calculations, you will need to define a character array 
variable for each time, but your data won't be CF-compliant.  In CDL
notation, you could use something like

dimensions:
   datelen = 15 ; // number of chars in every date-time string, blank padded
   time = unlimited;  // time dimension
variables:
   char time(time, datelen);  // date-time in ISO-8601 notation
data:
   time = "2012-03-31T11  ", "2012-03-31T11.5", "2012-03-31T12  ", ...

The above use only the netCDF-classic character arrays, which require that
an array of strings be represented as a multidimensional array of characters,
using a convention such as blank padding to make it possible to associate
strings with the associated dimension index.  You can do better in netCDF-4,
which supports real variable-length strings at the expense of some extra
complexity in programming for access.

--Russ







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



Ticket Details
===================
Ticket ID: BCN-163009
Department: Support netCDF
Priority: Normal
Status: Closed