Skip to content

Output in netcdf

Branch develop/netcdf includes now a basic way to write output as netcdf based on the netcdf_writer module which has been added to the code. The entire operation is carried out by the nc_write, which is currently called every 6hrs.

The subroutine includes a table which defines which variables should be saved. Similarly, all the output are currently saved in a single file data.nc at daily frequency for the entire experiment.

The subroutine then create the netcdf files, generating dimensions and variables, and then in a second instance append a new record in the file. If the file exists, data are only appended reading the last record of the time dimension. Variable names is trying to follow cmor-like standard. Both 2d and 3d files can be stored: of course the assignment of each model field to the netcdf file has to be hard-coded in the subroutine (since no pointer including all the diagnostic variables is present in bolam.F90). For 3d data, we exploited the interpolation capabilities of the wrshf_s2s subroutine which has been structured into a plev_interpolation subroutine which can potentially be used also on a different set of vertical levels. A short function to estimate julian date is used to provide the correct calendar in netcdf format. Metadata are provided in a decent way by multiple subroutines in the netcdf_writer module, and global attributes can be easily expanded. An important function is provided by nc_info, which provides the metadata and dimensional information for each variable.

IMPORTANT: the devel/netcdf builds on the devel/mod_model branch which split the mod_model from bolam.f90 and allows for separated compilation of the netcdf_writer module

As short guide, if you want to add a new variable you should

  1. add it to the savelist table, adjusting its dimension
  2. add it to the select case statement of nc_info to provide correct metadata
  3. add it to the select case statement of nc_write to assign the correct field
  4. add the required variables in the call of the mod_model

Things that can be improved

  • Read variable names and number of vertical levels from namelist
  • Split 2d and 3d field data data in multiple files
  • Improve handling of surface fields introducing a proper vertical axis
  • Generalize the nc_write function so that can be called at different time intervals
  • Produce cumulated fields as daily fields
Edited by Paolo Davini