grib_copy ICMGGftkm+001440 ICMGG_[typeOfLevel].grb

grib_to_netcdf ICMGG_hybrid.grb -o ICMGG_hybrid.nc

grib_to_netcdf -D NC_FLOAT ICMGG_hybrid.grb -o ICMGG_hybrid.nc

NCAR command language (NCL)

NCL provides a tool to convert both GRIB-1 & GRIB-2 to netCDF called ncl_convert2nc.

NCL example scripts to convert to netCDF are also available.

CDO: Climate Data Operators

These instructions assume the use of a recent version of Climate Data Operators (CDO) (available from Max-Planck-Institut, Germany).

CDO supports GRIB-2 but needs either the ecCodes library from ECMWF to be included. This is necessary in order to work correctly with OpenIFS model output.

cdo --version
Climate Data Operators version 1.9.6 (http://mpimet.mpg.de/cdo)
...........
Libraries: HDF5/1.10.4 proj/5.2 xml2/2.9.4
Filetypes: srv ext ieg grb1 grb2 nc1 nc2 nc4 nc4c nc5 
     CDI library version : 1.9.6
 cgribex library version : 1.9.2
 ecCodes library version : 2.12.5
  NetCDF library version : 4.6.2 of Feb 14 2019 10:13:27 $
    hdf5 library version : 1.10.4 threadsafe
    exse library version : 1.4.1
    FILE library version : 1.8.3
.........

If you find the CDO commands below do not work, you can either build CDO yourself, making sure a recent version of ecCodes is used, or you can use the workaround below.

Interpolate from reduced to regular grid

CDO can be used to interpolate from the reduced Gaussian grid to a regular Gaussian grid:

 cdo -R copy <input grib> <output grib>

The -R option only works for GRIB edition 1 (GRIB 1) data as it uses the CGRIBEX decoder.

For GRIB-2 messages it's better to use the setgridtype operator:

cdo setgridtype,regular        <input grib> <output grib>
cdo -f nc setgridtype,regular  <input grib> <output netcdf>

If these steps do not work, see workarounds below.

Steps to convert GRIB to netCDF

Split z axis

Before converting to netCDF, separate the different vertical axes using the generic command: cdo splitzaxis <input file> <output file pattern>.

This will ensure the following steps work correctly:

cdo splitzaxis ICMSHg4a4+000000.grb ICMSHg4a4+000000_split
ls ICMSH*split*
ICMSHg4a4+000000_split01.grb
ICMSHg4a4+000000_split02.grb
ICMSHg4a4+000000_split03.grb
ICMSHg4a4+000000_split04.grb

In this example, the different files might contain: '01' - fields on pressure levels, '02' - fields on model levels, '03' - hybrid levels, '04' - surface and so on. The number of files created depends on the number of different types of levels. Use the command grib_ls to inspect the contents of each split file.

Note that the filename ICMSH in this example, indicates it contains spectral and not gridpoint fields.

Another way to split the file would be using the grib_copy command from the ecCodes software as shown above:

grib_copy ICMSHg4a4+000000.grb ICMSHg4a4+000000_[typeOfLevel].grb

In this case 'typeOfLevel' is a GRIB key. The square brackets is a special syntax to the grib_copy command. This approach works with any GRIB key.

Convert spectral to gridpoint

CDO supports two options for converting spectral to gridpoint data:

• sp2gpl   - converts spectral to a linear Gaussian grid, appropriate for IFS data.
• sp2gp    - converts spectral to a quadratic Gaussian grid (only use this for T21/42 resolutions)

Both produce a regular Gaussian grid with equal number of longitudes on each latitude row. Remember that a Gaussian grid has irregularly spaced latitudes (but nearly regular). For more details please see the description of these operators on the CDO homepage.

cdo -f nc -sp2gpl <input spectral coefficent grib>  <output gridded netcdf>

Convert the resulting output files to netcdf

After any operation converting a GRIB file to any other GRIB file, the conversion to netCDF is just:

cdo -f nc copy <input grib> <output netcdf>

Convert vorticity and divergence to wind

To convert wind components u & v instead of vorticity and divergence, use:

cdo dv2uvl <input file> <output file>

Interpolate to pressure from model levels

To interpolate to pressure levels from model levels, use:

cdo ml2pl,92500,85000,50000,20000 <input file> <output file>

Viewing contents of GRIB files

There are various commands for inspecting the contents of a GRIB file. The ecCodes software installation required for OpenIFS has useful commands like grib_ls and grib_dump.

grib_ls ICMSHfrq2+000000.grb
grib_dump ICMSHfrq2+000000.grb

Here is another example using the cdo command:

$ cdo sinfo ICMSHfrq2+000000.grb 
   File format: GRIB
    -1 : Institut Source   Param       Time Typ  Grid Size Num  Levels Num
     1 : ECMWF    unknown  11.3        var  P16     65792   1      60   1
     2 : ECMWF    unknown  39.3        var  P16     65792   1      60   1
     3 : ECMWF    unknown  43.3        var  P16     65792   1      60   1
     4 : ECMWF    unknown  44.3        var  P16     65792   1      60   1
     5 : ECMWF    unknown  152.128     var  P16     65792   1       1   2
     6 : ECMWF    unknown  6.3         var  P16     65792   1       1   2
   Horizontal grids :
     1 : spectral     > size      : dim = 65792  truncation = 255  spc = 32896
                                    complexPacking = 1
   Vertical grids :
     1 : hybrid            level : 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 
                                   20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 
                                   36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 
                                   52 53 54 55 56 57 58 59 60 
     2 : hybrid            level : 1 
   Time axis :  1 step
     RefTime =  1999-12-24 12:00:00  Units = hours  Calendar = PROLEPTIC
  YYYY-MM-DD hh:mm:ss  YYYY-MM-DD hh:mm:ss  YYYY-MM-DD hh:mm:ss  YYYY-MM-DD hh:mm:ss
  1999-12-24 12:00:00
cdo sinfo: Processed 6 variables over 1 timestep. ( 0.04s )

Examples

Converting a single file

In this example a single GRIB file with model level data, on a regular Gaussian grid is converted to netCDF with vorticity & divergence replaced by u & v and model level data interpolated to pressure levels.

cdo -f nc copy file.grb file.nc
cdo dv2uvl file.nc file_uv.nc
cdo sp2gpl file_uv.nc file_uv_gg.nc
cdo ml2pl,92500,85000,50000,20000 file_uv_gg.nc file_uv_gg_pl.nc

Converting series of files

The following script shows how to loop over several output files using the steps above to convert to netCDF. Note the spectral parameter files are also converted but the extra command using the sp2gpl operator is first used to convert the variables from spectral to gridpoint on a regular, linear, Gaussian grid.

In this example, we use the pipelining feature of CDO to do both the 'copy' and 'sp2gpl' steps in one command, keeping the entire operations in memory for efficiency.

for i in 000 024 048 072
do

cdo -R copy ICMGGfpp4+000${i}.grb ICMGGfpp4+000${i}_R.grb
cdo splitzaxis ICMGGfpp4+000${i}_R.grb ICMGGfpp4+000${i}
cdo -f nc copy ICMGGfpp4+000${i}01.grb ICMGGfpp4+000${i}_PL_GP.nc		# pressure level gridpoint fields
cdo -f nc copy ICMGGfpp4+000${i}02.grb ICMGGfpp4+000${i}_ML_GP.nc       # model level gridpoint fields

cdo splitzaxis ICMSHfpp4+000${i}.grb ICMSHfpp4+000${i}
cdo -f nc copy -sp2gpl ICMSHfpp4+000${i}01.grb ICMSHfpp4+000${i}01_PL_GP.grb	# spectral fields on press levels transformed to gridpoint
cdo -f nc copy -sp2gpl ICMSHfpp4+000${i}03.grb ICMSHfpp4+000${i}03_ML_GP.grb	# spectral fields on model levels transformed to gridpoint
cdo -f nc copy -sp2gpl ICMSHfpp4+000${i}04.grb ICMSHfpp4+000${i}04_surf_GP.grb	# spectral fields on surface levels transformed to gridpoint

done

Using EMOSLIB to interpolate to regular grid

The ECMWF interpolation software library EMOSLIB also provides the capability to interpolate spectral data to regular gaussian grids or regular lat-lon grids, and interpolate regular gaussian grids to regular lat-lon grids.

The EMOSLIB library provides a Fortran library to enable users to write their own interpolation software. Please see the EMOSLIB website for more details and examples.

EMOSLIB also provides two command-line tools for using interpolation. Assuming a recent version of the EMOS library, there are two tools that can be used.

In the 'bin' directory there is a command 'emos_tool' which can be used as:

emos_tool --regular=256 [--area="40/-10/31/20"]  in.grib out.grib

The grid specified for the --regular option follows the grid naming convention for EMOSLIB here: Reduced Gaussian Grids.

Another command can be found in the 'tools' directory. This allows the interpolation function to be specified and is more flexible:

tools/int --INTOUT:gridname=F256 --INTOUT:area="40/-10/31/20" --input=1.grib --output=2.grib --intf2

Here 'F256' means 'full-grid' not reduced, and --intf2 is the EMOSLIB interpolation function to be used.

Possible problems and solutions

cdo -R option does not work with GRIB-2 fields to convert to regular grid

CDO's -R option, to convert from reduced Gaussian grid to regular Gaussian grid, only works with GRIB 1 format as the CGRIBEX decoder this uses does not work with GRIB-2 (see cdo man page). This is a problem as multi-level output from OpenIFS is encoded as GRIB-2 data.

The preferred way to deal with this is to use the setgridtype operator as described above:

cdo -f nc setgridtype,regular <input grib> <output netcdf>

However, if this does not work for any reason, a workaround is to temporarily change the edition number to 1 of the GRIB file. Use the grib_set command (from ecCodes) to change the GRIB edition number for all messages:

grib_set -s editionNumber=1  in.grb  in_1.grb
cdo -R copy  in_1.grb  in_1_R.grb
grib_set -s editionNumber=2  in_1_R.grb  in_2_R.grb

This does not actually convert the message to GRIB-1, it only changes the editionNumber. Any GRIB-2 specific elements are unaltered.

It's recommended that the editionNumber is changed back to 2 to ensure that the parameter names remain correct in any subsequent cdo / grib commands.

Workaround if using more than 128 levels

The workaround above will fail if the grib file contains field with more than 128 levels. This is because there are not enough bits available in the GRIB-1 parameter to encode the total number of model half levels (which would exceed 255).

In this case, the 'pv' GRIB parameter causes the problem (nothing to do with potential vorticity!). This is an array holding the half level values of the model's A & B coefficients that define the location of the levels.

The workaround above can be extended to delete this array while setting the edition number to 1, like this:

grib_set -s deletePV=1,editionNumber=1  in.grb  in_1.grb
cdo -R copy  in_1.grb  in_1_R.grb
grib_set -s editionNumber=2  in_1_R.grb  in_2_R.grb

The deletePV option is known as a 'concept' rather than a parameter contained in the GRIB file itself. It ensures the PV array is deleted correctly by grib_set. Now the cdo regridding will work correctly, but remember to reset the editionNumber back to 2.

The A & B half level coefficients can be recovered from the original file by:

cdo vct in.grb > pv.txt

Parameter names are lost for GRIB-1 fields

ECMWF GRIB-1 use keys which may not be recognised by CDO because they are locally defined (e.g. the shortName key) and not defined in the WMO GRIB tables that CDO uses.  This can cause parameter names to be lost or not recognised when using the CDO commands.

A workaround is to use the -t ecmwf option.  This will make CDO use ECMWF parameter table 128 for the definitions of variables but 128 is only for GRIB 1. This will work for the surface fields which are encoded by IFS as GRIB 1, but not for the multi-level fields which are encoded in GRIB 2. Variable names become 'unknown' if -t ecmwf is used with the multi-level fields. It is best to omit it completely, even for surface fields, and only use it for GRIB 1 data if you get missing variable names after CDO has converted the data (see CDO documentation for more details).

cdo -t ecmwf -f nc copy mygrib1.grb mygrib1.nc

Acknowledgements

Thanks to Paul Dando of User Support for help with the contents of this page.