[Mitgcm-support] Note: January 25, 2001

[mitgcm-support at dev.mitgcm.org]

HyperNews notification of new message.  See: 
  http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/wkmtg/80.html

     =======================================================================     
     
NOTES FROM THE ASSIMILATION MEETING : 25 January 2001    

1) K-filter 
------------
   - Degradation (east of Philippine) by the baroclinic filter is
     largely eliminated by assimilating detrended T/P and model sea
     level.  Skill of the barotropic filter does not depend much on
     trend removal.  See the skill (simulation residual variance minus
     dynup residual variance) of the baroclinic[1] and barotropic[2]
     filters.  Top left and top right panels are assimilation without
     and with removing trend, respectively.  (Lower panels use
     different process noise, Q.)

   - The skill of the filter is largely confined to regions spanned by
     the filter.  Moreover, the combined baroclinic and barotropic
     filter has skill that is practically the sum of the individual
     baroclinic and barotropic filters alone as seen here[3]; Top left
     (barotropic filter skill), top right (baroclinic filter), bottom
     left (combined baroclinic-barotropic filter), bottom right (sum
     of top left and top right).  These results indicate the separate
     filter estimations being linear and independent, and validate the
     assumption and skill of the hierarchical Kalman filter.

   - Process noise Q is being tuned; e.g., lower panels of first set
     of plots above.  We think we can do better.  See future postings.

   - Comparison of K-filtered results with XBTs show improvement in
     model skill, albeit small.  Raw differences are dominated by
     model bias.  XBT data are those from NCEP.  Results to be posted
     soon.  

2) Adjoint 
------------
   - Six iterations of the 1-year adjoint have been completed.  (Can
     do one iteration per day!) The cost function and comparisons with
     TAO T and U can be seen here[4].  The model still has SSS-SST
     relaxation, whereas the cost function only constrains the
     inhomogeneous surface heat and E-P forcing; the weights used in
     these are overestimates.  SST is somehow degraded; may have to do
     with time-invariant mixing parameters (KPP average).

   - A JPL buyer has been contacted to begin examining if we can buy
     TAF.  Buyer is concerned of purchasing a product from an
     ex-JPLer.  Johnny says we can't use P-card to purchase from
     foreign company.

===> Moto's term balance will be implemented on the ECCO code
     (Dimitris/Benny). 

===> Surface relaxation of SSS and SST need to be dumped for ECCO
     code (Dimitris). 

===> Implement time-varying subsurface T constraint so as to
     assimilate XBT data (N.Smith's gridded product)  (Benny). 

5) Other
---------
   - Model forcing has been updated.  We now have 1948 to Oct 2000.
     Tony has summarized the steps taken to get the forcing data; from
     ftp'ing Charmaine King's 1-deg format to interpolating to our
     grid.
===> A script will be written to process any future updating of
     forcing data (Zhangfan).  Updating no less frequently than every
     quarter.

   - T/P data has been extended to cycle 301. 
===> There are several data dropouts in the new T/P data set that need
     to be sorted out (Benny).   

   - Alistair's impldiff correction bombs when implemented to our
     higher resolution.  Cause still unknown. 

   - Tony found out his data disks were never backed up.
===> Zhangfan to 
        * write script to grep info of any failed disk backup.
        * automate e-mailing backup confirmation
        * report when last backup took place for all our disks. 

   - Nick finds optimal skill for his K-filter for a wind-driven
     barotropic model assuming process noise being 10 to 30% of NCEP
     wind covariance.   Somehow the MITgcm K-filter has a better skill
     improvement than Nick's K-filter when compared with bottom
     pressure measurements.  [Absolute skill is comparable.]

   - Detlef will have a Sun as their data server.  Zhangfan to
     implement the data server software[5] for their machine; we will
     have a copy running here, too. 

   - Postings on the forum can be expanded by clicking the "All"
     button near the top of the page, enabling search of strings
     within the postings' texts.   

   - The present forum will be retired as of end of January in favor
     of a newer version that has more features; in particular, string
     search without expanding all postings.

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

-----------------------------------------------------------------------
[1] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t73f5g.ps
[2] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t73f5g_bt.ps
[3] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t73f5h_b.ps
[4] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/assim/110.html
[5] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/mit-jpl/10/1.html