Alkim:

Ok, I finally have a "concurrent model processor" (modelproc) from
which you can get NO+. This code executes time-gcm version 20 for a
single time step, saving NO+ (and optionally other fields), then
the processor part makes the plots and/or ascii data.

Take the following steps from your home on cedar:

1. mkdir ~/modelproc
2. cd ~/modelproc
3. cp ~foster/modelproc/newuser/README .
4. cp ~foster/modelproc/newuser/modelproc_tgcm20.alkim modelproc_tgcm20

Read the README. The file modelproc_tgcm20 is the script to
submit. I have set it up as an example of how to run for one
of the histories you have been looking at. Let me know when you
get it and I will remove it from ~foster/modelproc/newuser.
Don't worry about TSEYDRx volumes not being tgcm20 -- they should
work ok. You might want to ask Ray about new runs, I think he may
have made some new runs with tgcm20.

The inputs are in two parts, the first for the processor, second 
for the model.  The processor inputs look similiar to tgcmproc. 
The inputs for the model are lexical reads rather than namelist,
but just follow the obvious syntax.  For successive runs, you 
should only have to change SOURCE, START, STOP, etc., in the model 
inputs. (don't forget to set sendcgm and senddat in the processor 
the first time).

Submit like tgcmproc, i.e., "nrnet shjob modelproc_tgcm20".

This is probably a bit buggy. Let me know when you have problems.
Also beware it requires a lot more memory than tgcmproc, i.e.,
keep your QSUB -lM up around 7.5 to 8Mw. Good luck -- I will 
probably log in from home on Sunday, so send me mail if you
try it out this weekend.

--Ben

BTW, if you have occaision to make several successive runs of
tgcmproc, you can job-step it, i.e., make several executions with
different inputs from the same script, i.e., in the same batch
job, rather than waiting for each job separately. Call me if you 
want to do this.