[Mitgcm-support] Note: January 4, 2001

mitgcm-support at dev.mitgcm.org mitgcm-support at dev.mitgcm.org
Wed Jul 9 15:51:51 EDT 2003


HyperNews notification of new message.  See: 
  http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/wkmtg/77.html

     =======================================================================     
     
NOTES FROM THE ASSIMILATION MEETING : 04 January 2001

1) K-filter
------------
   - A month's worth of achievements were lost during the holiday
     break by alhena's /tmp being scrubbed.  The loss included four of
     the latest K-filter runs (2 baroclinic and 2 barotropic) and the
     sole copy of the working source code.  This is a serious setback.
     Results must be backed up timely and duplicate copies of source
     code must always be maintained.

   - The K-filter results comparison with bottom pressure gauges were
     updated with more data from P.Woodworth/R.Peterson[1].

2) Adjoint
-----------
   - The 3-month assimilation was revisited using different
     preconditioning factors in the minimization algorithm.  With
     10-iterations each, the best result (w.r.t. TAO and cost
     function) is obtained when correction is preferentially made to
     the wind.  This indicates the importance of judiciously choosing
     the preconditioning factors.  The best preconditioning factors
     will be used with the 1-year adjoint. 

   - A few more runs will be conducted to assess the prior errors of
     the initial hydrography.  Present error estimates are based on
     hydrography's spatial variability averaged horizontally over the
     globe; (i.e., prior error is a function of depth alone.)  Largest
     values of error (variability) are found in the surface.  However,
     with Reynolds relaxation, the model's surface has a smaller
     difference with Levitus than the interior; i.e., the error near
     the surface should probably be smaller than those at depth.

   - The packaged KPP works with zero or constant dQdT (T relaxation
     constant) but not so with variable dQdT.  The problem may be with
     large dQdT; the constant dQdT that was tried may have been too
     small.  The test is with full KPP (i.e., with shear instability
     on etc) but with no GM.   For the time-being, adjoint will be
     done without KPP; KPP requires 60% overhead.  

3) Others
----------
   - Alistair's new impldiff subroutine corrects a bug in vertical
     mask and has a smaller memory requirement.  The difference with
     the old routine seems only important at shallow locations.

   - The TAF (TAMC's successor) license agreement has been received.
     The license should be reviewed by Procurement.  

   - Dimitris is putting together a poster of our work and those of
     MIT's biogeochemical group for the GSFC CO2 meeting.  

   - The ames O2K is now available.  However, the machine is 2/3 the
     speed of alhena and is crowded.   See Benny's posting[2]. 

=======================================================================     

-----------------------------------------------------------------------
[1] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/assim/106/1/1/1.html
[2] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/comp/21.html



More information about the MITgcm-support mailing list