All,

I've uploaded the NMC-method background error stats package:
https://svnemc.ncep.noaa.gov/projects/gsi/branches/dtk-berr/util/NMC_Bkerror

You can check out the code, with the following command:
svn checkout https://svnemc.ncep.noaa.gov/projects/gsi/branches/dtk-berr/util/NMC_Bkerror NMC_Bkerror

Once you download/check out the code, you will see the following directories:
sorc, scripts, fix, and util

To compile, simple type "make" in the sorc code directory (this will generate the executable calcstats.exe).  

The driver for the program is statsmain.f90 which does nothing more than read in a namelist, some preliminary 
  stuff, and then calls the various component routines.  You should be able to follow the calling sequence fairly easily:
getcases: reads in the headers of the spectral files to see how many are available for processing
readpairs: reads in the 1,2 day lagged forecast pairs, convert to gridded fields, and right out subdomain scratch files
biascor: perform a systematic bias removal (this is mainly to remove huge biases near top of model)
balprojs: compute matrices that allow for transforms from streamfunction to balanced vel. pot, ps, and temperature
variances3d/variances: compute variance of each of the fields
vertsc: computes vertical length scales (in inverse model grid units) using recursive filter method (see Wu paper)
horizsc: computes horizontal length scales (km i think) as per Wu paper

The code should be fairly straightforward to follow, but feel free to pass along questions or comments.  
  I've set up a database of T126 GFS forecast pairs on vapor (450 pairs) for you to use/test the code.  If you look in 
  the scripts directory, you will see nmcstats_t126_vapor.sh

This script will compute the background error stats for the 450 forecast pairs at T126 (you just need to modify a few 
  things, like my user id to yours and whatever account you run under).  The test I ran took a little over 20 minutes 
  using two nodes on vapor.  If you get the code to run, there will be several things left over in the run directory upon completion:
gsir4.berror_stats (this is the file that is used by GSI...but this is the OLD, non-gcv format)
bgstats_sp.grd (this is a byte-addressable file that will allow you to view the 2d fields easily using GrADS....
  I need to find a control file for the T126 grid)
gsistats.out: standard output, there is nothing really of use in here unless diagnosing how/where things failed