Example NetCDF files to be used for initial work can be found here:
ftp gmaoftp.gsfc.nasa.gov anonymous cd /pub/data/mccarty mget nc_4emily.diags.20171202.tar
The size of the tarball on FTP site is 11G and contains both guess and analysis diagnostic netCDF files for all 4 cycles of 20171202.
Guess diagnostic netCDF files for the 00Z cycle can be found here and its size is approximately 1.5G. The file structure is exactly the same, and the only discernible difference between the guess and the analysis diagnostic file is the global attribute of outer_iteration number. For the guess, this value is 1, for the analysis it is 3.
More information from about the bufr to netCDF converter here.
The diagnostic files, and the GMAO-developed GSI implementation of these files, are dependent on the 'ncdiag' libraries, which can be acquired via Will McCarty's ncdiag repository on GitHub.
the NRL team has created fortran interfaces for our H5 innovation (observation) files. The fortran and python readers are here: https://wiki.ucar.edu/x/2YSOF
A couple of sample innovation files are also included . We can easily generate them for any date time group upon request.