[Mitgcm-support] Note: February 1, 2001
mitgcm-support at dev.mitgcm.org
mitgcm-support at dev.mitgcm.org
Wed Jul 9 15:55:42 EDT 2003
HyperNews notification of new message. See:
http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/wkmtg/81.html
=======================================================================
NOTES FROM THE ASSIMILATION MEETING : 1 February 2001
1) Figure for Eric
-------------------
- Eric Lindstrom has requested a plot highlighting our work to be
included in a NASA Report (Earth Science Enterprise). The plot
is needed by Monday. He rejected the following suggestions;
* Trajectory plot
* Assimilation's reduction of model-data residuals
* Term balance
* ssh Hovmoeller diagram
He says these are "science journal stuff" and not for public
consumption. He wants "eye candy" and a "sense of usefulness."
The following were discussed as alternatives;
* Change of Equatorial Currents.
Undercurrent, Counter Current.
Show change as a function of xy, xz, or yz.
* Focus in the upper 150m where changes are largest.
* Plots should be of assimilation.
- This sort of request will recur and we need to be prepared. We
should collectively think of what to highlight and keep a copy of
plots handy.
2) K-filter
------------
- A summary of the effects of trend on the skill of assimilation
was posted[1]. Removing the trend in sea level reduces model
degradation by the assimilation.
- Model process noise is being calibrated so that uncertainty
estimates used in assimilation are consistent with actual
model-data differences. Run kf023a, in particular, described in
the summary above is not consistent in this regard even though it
results in little degradation. When process noise is calibrated
(matched) to be comparable to prior model error estimate
(kf024f), the assimilation once again leads to degradation off
Taiwan. It is speculated that the degradation is due to
unrealistic K-filter updates at the northwestern cell boundary;
The cellular filter's error estimate should taper to zero at open
boundaries due to representation error. When corrections are
forced to zero at the north-south boundary (20N, 20S) of the
Pacific cellular filter (kf024i), the degradation indeed is
diminished even when using the calibrated process noise.
[Addendum] This figure[2] compares skill of dynup over 1993-97 for
kf024f (top left; calibrated Q), kf024i (top right; calibrated Q,
tapered P at NS boundary), and kf023b3 (bottom left; uncalibrated
Q). kf024i has nearly as small a degradation as kf023b3.
Summary of this error calibration will be posted shortly.
- K-filter comparison with XBTs described last week have been
posted[3]. The largest improvement is found in the eastern
equatorial Pacific. There is non-trivial change outside the
tropics.
- T/P data for K-filtering has been extended to end of 1999. The
problem with the data reported last week was in Akiko's
processing (some of her input files were incorrect?). Benny has
posted a summary package of preparing the T/P data here[4].
3) Adjoint
-----------
- The one-year adjoint has completed 12 iterations. This analysis[5]
at the end of the 10th iteration shows improvement beyond the 5th
iteration (i.e., 10th better than the 5th iteration).
- Dimitris has implemented Moto's term-balance diagnostics to the
forward code of ECCO (the adjoint assimilation code). Benny is
making the implementation compatible with TAMC. The diagnostics
will allow term-balance analysis of the adjoint assimilation.
4) Other
---------
- Extension of control (c20000630) to end of 1999 is ongoing at
Ames.
- Zhangfan has implemented the automatic e-mailing of disk backup
information. Each user should keep an eye on it. If you don't
receive any e-mail, that means nothing is done. If you receive
the message, check if the disk was backed up successfully or not.
===> Everyone to monitor the e-mail and be aware of when your disks
were last backed up successfully and take corrective action if
necessary .
- Nick found assimilation skill of his barotropic model (with
respect to independent bottom pressure data) to be slightly
better than the MITgcm K-filter run when he reduced the magnitude
of his process noise.
However, this improvement is limited to periods longer than
15-days, similar to what was found with the MITgcm. This means
that there is no gain in doing assimilation as far as T/P
de-aliasing is concerned; simulation is good enough. However,
the assimilation was done with stationary model error P. Taking
into account the time variation in P (due to inhomogenous
T/P sampling) may allow improvements at higher frequencies.
- The new forum is not up yet.
=======================================================================
-----------------------------------------------------------------------
[1] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/assim/112.html
[2] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t73f5g_24i.ps
[3] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/assim/106/3/1.html
[4] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/assim/111.html
[5] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/Tony/tony0201.ps
More information about the MITgcm-support
mailing list