Usage Examples

1. get an overview of times, layers and grid used in the felt-file
2. group the parameters so that all parameters within an group use the same grid, times and the same layers (its possible to have a file with both sigma and pressure layers, but then all sigma-variables should have the same number of sigma layers and all pressure-variables should have the same number of pressure layers. Surface only variable, usually sigma=1000 felt variables, are layer-group of their own)
3. create a felt2nc_variables.xml file for each variable group. Copy parameters from existing examples, or lookup variable and parameter names in the CF standard name documentation. If you have to generate several groups, it might be useful to split the file into axes, global attributes and variables, see i.e. damocles_felt2nc_variables.xml If you have two paramters which map to the same standard_name, i.e. make sure to give them different variable names if they are in the same group. Adjust the datatype according to your data. float will always work fine, use short if you know, that the variable even fits as 4-dimensional data into shorts. Use eventually a cdmWriterConfig.xml to process data as float and write the data as short.
4. validate the felt2nc_variables.xml file, i.e. with

   xmllint --xinclude --postvalid --noout ../../share/etc/felt2nc_variables.xml

5. adjust the projection and area you want to extract in fimex_example.cfg (or use command line arguments to fimex)
6. run

   fimex -c fimex_example.cfg

Zarr does not support unlimited dimensions. You may need to set unlimited=false for some dimensions, e.g. with ncml like

<?xml version="1.0" encoding="UTF-8"?>
<netcdf xmlns="http://www.unidata.ucar.edu/namespaces/netcdf/ncml-2.2"
        xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
  <dimension name="time" length="61" isUnlimited="false" />
</netcdf>

In version 4.8.1, libnetcdf does not support compressed zarr chunks. You probably have to disable compression with a netcdf writer config xml file similar to:

<?xml version="1.0" encoding="UTF-8"?>
<cdm_ncwriter_config>
  <default compressionLevel="0" />
</cdm_ncwriter_config>

The fimex command would then be something like

fimex \
    --input.file "input.nc" \
    --input.config "no_unlimited_dim.ncml" \
    --output.file "file:///path/to/output#mode=nczarr,file" \
    --output.config "nc_writer_no_compression.xml" \
    --output.type nc4

Please note that using libnetcdf to write zarr will still use a lock on libnetcdf function calls, i.e. there will only be one thread at a time reading or writing.

Datasets that were written like this do not seem to be readable with the same url (observed with netcdf 4.8.1), but reading appears to work with mode=zarr,file (i.e., not nczarr).

See the libnetcdf zarr documentation for more details on zarr support in libnetcdf.