In order to try running the GMAO software for simulating observations
from the ECMWF nature run, there are several files or pieces of
software that you will need in addition to our simulation software:

1. You will need a library for the BUFR routines.  We use a version
   that has integer*4 and real*8 as the kinds used for arguments, so
   if such a version is available to you, that is the one you should
   use. Let us know ahead of time what you have.

2, You will need a routine that determines eigenvalues and
   eigenvectors for a real symmetric matrix (probably need a version
   that uses real*8 arithmetic).

3. You will need a CRTM library.

4. You will need some observation data sets in NCEP BUFR format. If
   any of those files do not include the required BUFR table, then a
   separate file containing that table must also be available.

5. You need to generate files for reading the Nature Run fields in the
   format we use.  Later you can replace our reads if you want with
   something that does not require this sepcial format, but to start
   with, you should generate the files as Runhua Yang will instruct
   you. These read the ECMWF GRIB data files and create separate
   binary files for different times and 3d fields.

6. I have a 40-page description of the software, describing the
   purpose and approach, approximations, algorithms, software design
   and use, explanation of files and printed output, instructions for
   changing NR resolution, etc.  Whoever intends to use the code
   should read portions of it before hand to get a flavor for what the
   software contains and for what is in the documenation for later
   reference.

7. We are willing to sit down with users to go through the process of
   running the software.  We think that is better than saying "Here it
   is. Good luck." For the same reason, however, we prefer not to go
   through this exercise with anyone until they are actually ready to
   run the codes.

8. Although we have run a 2 month simulation, we must wait another
   week for a guy to return from vacation and fix a bug in the
   evaluation software that we are using. So, the GSI appears to read
   the data OK and to produce somewhat appropriate mean values of Jo,
   but that is a very gross evaluation.  We know that tuning will be
   necessary.

9. We give this preliminary software to others with the understanding
   that: (1) they agree to not submit any journal paper describing
   results from OSSE experiments using these observations or this
   software until either a paper reporting the tuned baseline results
   is submitted by the GMAO team or April 1, 2009, whichever
   date/event occurs earlier and (2) they agree to not present results
   produced using our untuned simulated data without clearly stating
   that the GMAO intended this early data for development purposes
   only. My guess is that tuned data sets may be available by December.