[Mitgcm-support] Note: January 18, 2001

[mitgcm-support at dev.mitgcm.org]

HyperNews notification of new message.  See: 
  http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/wkmtg/79.html

     =======================================================================     
     
NOTES FROM THE ASSIMILATION MEETING : 18 January 2001    

1) American Meteorological Society Meeting 
-------------------------------------------

   - Detlef and Ichiro presented progress of ECCO at the AMS meeting.
     The participation was a request by NOPP.  For the JPL part, skill
     of the K-filter and adjoint was presented based on comparisons
     with independent data (T/P, bottom pressure, TAO T, surface U).
     Adjoint sensitivity study of the ITF was also described.

     Validating the skill of the filter per se, although a
     prerequisite, does not pique the general interest including those
     of HQ managers.  We need to collectively think of how to make a
     bigger impact.  Following points were brought up; 

     a) It is unlikely that a zero order qualitative change in our
        understanding/description of ocean circulation is going to
        result from assimilation.
     b) What can we (assimilation) do that nobody else can do (e.g.,
        ad hoc assimilation, modeling, data analyses)?  One example is
        that we can produce error bars to our estimate.
     c) Focus analyses on specific features that assimilation made a
        difference; e.g., surface equatorial circulation changes from
        Jan-Mar 1997 estimated by adjoint.
     d) Examine what difference there are between simulation and
        assimilation. 
     e) Begin analyzing the oceanographic features of the assimilation
        regardless of specific impact of the assimilation.

2) K-filter
------------
    - The detrended baroclinic filter does appear to get rid of the
      large degradation.  This figure[1] compares the skill of one of the
      baroclinic filter runs without removing trend (left) and with
      removing trend (right).  The skill is measured by the decrease
      (shown as positive number) in model-data residual variance
      (cm^2) from the unconstrained control run. [The figure shown at
      the meeting was in error.  The figure above is the corrected
      one, and for the full 5-year period 1993-1997.]  The large
      degradation east of Taiwan has significantly diminished in the
      detrended assimilation. 

    - All lost runs have now been recovered. The barotropic and
      baroclinic filters have each been run with two different prior
      errors (process noise); NCEP wind covariance, scaled NCEP wind
      covariance to match simulation-data difference.  In addition, a
      detrended version of the NCEP wind covariance runs are being
      carried out (baroclinic is done.  barotropic is nearly
      complete.)

    - The combined barotropic and baroclinic filter program has been
      debugged.  The code will now be tested for a particular choice
      of prior errors.

    - The assimilation code with the detrending algorithm will be
      placed on CVS.

3) Adjoint
-----------
   - Two additional variations of the 3-month adjoint did not improve
     the solution much;
      a) Changed wind preconditioning factor to be twice as sensitive 
      b) Reduced temperature weights near the surface based on
         comparisons with TAO

   - The 1-year adjoint assimilation will commence using the weights
     and preconditioners used in the 3-month assimilation.
     Calculations will be conducted at Ames.  The optimized solution
     to the 3-month adjoint has a 25% smaller initial cost function
     than the control.

4) Ames 
---------
   - The "test" queue at Ames is no longer available.  

   - Command
          acct_ytd
     provides information of year-to-date usage.  However, this usage
     reflects that of the JPL-Goddard allocation as a whole.
     L.Eversole says not to worry about how much we've used; if we run
     out, he will try to get some more.

   - Oversubscription makes steger unsuitable to do adjoint; the
     calculation takes too long.  However, the 8-hour queue would work
     OK for the K-filter.

5) Other
---------
   - A 50+ year run (1947-2000) is being prepared.  Forcing from
     1952-2000 is present on HPSS.  Those of 1947-1951 will be
     retrieved from NCAR.  A 2-deg by 2-deg with 23-levels (SIO
     configuration) will be done first, then one in c20000630
     configuration.  A northern and southern sponge layer may need to
     be implemented to minimize the trend arising from lack of deep
     water production over the long integration.  Comparisons will be
     made between using the latest MIT release and our c20000630 code.

   - Forcing for '99 to Oct '00 is being retrieved from NCAR.  Tony is
     streamlining the code for somebody else to take over for future
     retrievals.

   - Results Dimitris put together for the CO2 meeting at GSFC
     regarding MIT's biogeochemical studies using our control results
     will be added to our web-site.

   - Purchasing TAF through procurement may take several months due to
     complications associated with the foreign contract.  We could
     expedite the purchase by using P-card, except for P-card's upper
     limit on cost (several $K).  We need to see if we can purchase
     TAF by ~4 installments.

   - Description of optimization flags used in c20000630 has been
     posted[2].

   - The OS on the three linux boxes (nautilus, stormy, giant) is 
     being upgraded to RedHat 7.0.  This is to prevent break-ins and to
     make system administration simple.  However, the upgrade for
     nautilus encountered problems; the partition table of the main
     disk was lost.  Home directory is being recovered from backup,
     but some directories are inaccessible.  Will attempt recovery by
     outside contractor.  All users should double check that their
     files are regularly backed up.

   - Lee is comparing c20000630 and T/P sea level in the North Pacific
     associated with reflected signals from the east coast.  There is
     coherence between the two.  Although the reflected signal is
     small, it appears to be amplified from time-to-time by wind
     events.

   - Tony and Xiaoyun are examining reflection and transmission of
     equatorial waves at the west end of the tropical Pacific. 

   - Nick has concluded that estimating pressure driven non-IB signal
     would be more complicated than estimating wind-forced solutions
     because of the dominance of IB being a static response. 

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

-----------------------------------------------------------------------
[1] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t73f5e_0111.ps
[2] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/comp/22.html