sdfwrite

sdfwrite varname

This command will write out a defined variable varname into a NetCDF formatted data file.

Usage Notes

The sdfwrite command was initially implemented in GrADS version 2.0.a3. Additional features added in subsequent releases are as follows:

Options to control the name and format of the output file are implemented with the set sdfwrite command. The q sdfwrite command returns the status of the sdfwrite options.

The name of the output file will be grads.sdfwrite.nc unless specified otherwise with the set sdfwrite command. If the file exists, it will be replaced. It is not possible to append variables to an existing file.

The dimensions of the variable written to file correspond to the dimension environment that was set when the variable was defined. The dimension environment that is set when the 'sdfwrite' command is invoked is ignored. Note this behavior is different from the fwrite command.

By default, the output file will have a coordinate variable only for varying dimensions in the defined variable; non-varying dimensions will not appear as a coordinate variable with a size of 1. However, options have been added to the set sdfwrite command to force the variable in the output file to have at least 3, 4, or all 5 dimensions. When either of these options to set sdfwrite are used, the output file will retain information about the values of the dimensions that were fixed when the variable was defined; non-varying dimensions will appear as a coordinate variable of size 1. For example, if your defined variable is 500mb height on14may2002 (a 2D variable that varies only in lon and lat), and you use the -4d option, the output file with show height as a 4D variable with a Z dimension of size 1 called 'lev' with a value "500 mb", and a T dimension of size 1 called 'time' with a value of "0 minutes since 2002-05-14 00:00".

The coordinate variables will be type 'double' and will have two default attributes ("units" and "long_name") with values that are based on the GrADS 5-dimensional gridded data model. The data variable will also be type 'double' by default; beginning with version 2.0.a8, you can also write out data of type 'float' if you use the -flt option with set sdfwrite command. Data variables will have a "missing _value" or "_FillValue" attribute for undefined gridpoints. The undefined value is controlled with the set undef command.

For example, if varname is called "testvar" and varies in X, Y, Z, T, and E, then the output file might have a NetCDF header that looks like this:

netcdf foo {
dimensions:
  longitude = 9 ;
  latitude = 9 ;
  level = 9 ;
  time = 9 ;
  ensemble = 9 ;
variables:
  double longitude(longitude) ;
     longitude:units = "degrees_east" ;
     longitude:long_name = "Longitude" ;
  double latitude(latitude) ;
     latitude:units = "degrees_north" ;
     latitude:long_name = "Latitude" ;
  double level(level) ;
     level:units = "millibar" ;
     level:long_name = "Level" ;
  double time(time) ;
     time:units = "minutes since 01-01-0001 00:00" ;
     time:long_name = "Time" ;
  double ensemble(ensemble) ;
     ensemble:grads_dim = "e" ;
     ensemble:long_name = "Ensemble member" ;
  double testvar(ensemble, time, level, latitude, longitude) ;
     testvar:missing_value = -888. ;
}

The time axis units will always be "minutes since ..." and the date of the time axis origin will correspond to the initial time of the defined variable.

If the variable has an ensemble dimension, the attribute "grads_dim" with the value "e" will always be present so that the resulting output file can be opened with GrADS using the 'sdfopen' command.

To supplement or override the default attributes of the output file, use the set sdfattr command.

Beginning with version 2.0.a8, the output file may also be a compressed netCDF file. Use the -zip option with the set sdfwrite command to enable compression. Please see the documentation on compression for more details.

Examples

The following commands produce a regional subset of a global precip forecast :

open gfs.ctl
set lon -111.3 -103.8 
set lat 40.9 45.0 
set t 1 last
define precip = ptot
set sdfwrite wyoming_precip.nc
sdfwrite precip

Appending Variables to a File
The sdfwrite command does not allow you to write out more than one variable to a file. The way to do achieve this is to write out each variable in a separate file, then merge them with the NetCDF Operators (NCO). Use the ncks command to append variables defined on the same grid. The GrADS commands might look like this: 

open gfs.ctl
set x 1 720
set y 1 361
set t 1 last
define slp = mslp
define precip = ptot
set sdfwrite -flt slp.nc
sdfwrite slp
sdt sdfwrite -flt precip.nc
sdfwrite precip
Then, outside GrADS, or using the ! in front of the command to send it to the shell from within GrADS, merge these two files with ncks: 
ncks -h -A precip.nc slp.nc

The -A is for appending to the output file, and the -h does not add anything to the output file's global 'history' attribute. Please see the NCO User's Guide for more information.

Concatenating Files
Suppose you want to write out a long time series with the sdfwrite command, but your system does not have enough memory to define the whole grid. Use GrADS to write out the data in smaller time segments (e.g. one year, or one month, or one day), and then concatenate them together with the NetCDF Operators (NCO). Concatenation in the time dimension only works properly if time is defined to be a record (unlimited) dimension. If you are using GrADS version 2.1.a2+, you can use the -rt option to set the time axes as a record dimension; otherwise, you must use the ncks command to change the time dimension to a record dimension. When the inidividual files have been created, use the ncrcat command to concatenate them together. The GrADS commands might look like this: 

open long_time_series.ctl
set x 1 720
set y 1 361
set time 01jan2001 31dec2001
define slp = mslp
set sdfwrite -flt slp_2001.nc
sdfwrite slp
set time 01jan2002 31dec2002
define slp = mslp
set sdfwrite -flt slp_2002.nc
sdfwrite slp
set time 01jan2003 31dec2003
define slp = mslp
set sdfwrite -flt slp_2003.nc
sdfwrite slp

If you are using GrADS version 2.1.a2+, the 'set sdfwrite' commands should look like this instead:
set sdfwrite -flt -rt slp_2001.nc
set sdfwrite -flt -rt slp_2001.nc
set sdfwrite -flt -rt slp_2001.nc

If you are using an earlier verison of GrADS, you must modify the time dimension with ncks -- do this outside of GrADS, or use the ! in front of the command to send it to the shell from within GrADS (the --mk_rec_dmn option modifies the time dimension): 

ncks -O -h --mk_rec_dmn time slp_2001.nc
ncks -O -h --mk_rec_dmn time slp_2002.nc
ncks -O -h --mk_rec_dmn time slp_2003.nc

Now you are ready to concatenate the files with ncrcat:

ncrcat -O -h slp_2001.nc slp_2002.nc slp_2003.nc slp.nc

The -O is for overwriting existing output file, and the -h does not add anything to the output file's global 'history' attribute. Please see the NCO User's Guide for more information -- there are shortcuts for ncrcat to specify the long list of files to concatenate.

N.B. (For compressed NetCDF-4 files): If the NetCDF files you create with GrADS are compressed netCDF-4 format, then you must use NCO version 4.0.4 (or higher) in order to maintain the compression in the output files. The chunk sizes may also be altered by ncks or ncrcat. Use the --cnk_dmn option to specify the name and chunk size of each dimension to manually override the defaults. For example: 

ncrcat -O -h --cnk_dmn lev,1 --cnk_dmn lat,361 --cnk_dmn lon,720 slp_2001.nc slp_2002.nc slp_2003.nc slp.nc