[Mitgcm-support] Note: February 15, 2001
mitgcm-support at dev.mitgcm.org
mitgcm-support at dev.mitgcm.org
Wed Jul 9 15:56:27 EDT 2003
HyperNews notification of new message. See:
http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/wkmtg/83.html
=======================================================================
NOTES FROM THE ASSIMILATION MEETING : 15 February 2001
1) K-filter
------------
- The combined bc-bt K-filtering has been run for 1993-2000 (3rd
quarter) with different process noise (Q);
* NCEP wind covariance (kf026b)
* Scaled NCEP wind covariance to match estimate of simulation
P from model-data comparison (kf029a)
Both have similar dynup skill as seen in this bottom panel[1] as a
function of time; kf026b (black), kf029a (blue). The sudden
decrease in skill at the beginning of 1998 corresponds to a
decrease in signal amplitude (and model error to be corrected)
associated with the transition from El Nino to La Nina.
- Simulation error appears to be non-stationary especially in the
tropical Pacific. This figure[2] shows estimates of simulation
error comparing data and simulation assuming independence among
data error, model error, and the common signal. Process noise of
kf026b may not necessarily be too small.
!!! HOWEVER, modal amplitudes are correlated among different modes as !!!
!!! described below. Therefore, it may be best to forgo any further !!!
!!! adjustment of Q for the single baroclinic mode filter. !!!
- The 5-baroclinic mode equatorial Pacific filter has been
derived. This figure[3] shows the simulation sea level error
variance estimate (cm^2) based on NCEP wind covariance. The
figure shows;
* First two modes have largest errors.
* The first mode is largest in the western Pacific, whereas
the second mode sea level error is largest in the eastern
Pacific.
The errors are also correlated among different modes. This
figure[4] shows columns of the corresponding measurement updated
state error covariance matrix, corresponding to displacements in
the middle of the Pacific on the equator (coarse grid 63); 1st
mode (top), second mode (middle), third mode (bottom). The
coarse state is ordered by mode (elements 1-396 is 1st mode,
397-792 is 2nd mode, etc, delineated by the magenta lines).
Within each mode, first 132 elements are displacement, 2nd 132
are U, and 3rd 132 elements are V. Elements are modal amplitudes.
- Additional K-filter runs (1st baroclinic and barotropic) have
been compared with bottom pressure and TAO. For bottom pressure[5],
the non-calibrated Q (NCEP wind covariance) fair best (runs
kf021c, kf026a). Similarly, non-calibrated Q runs compare better
with TAO-T[6] below 200m. All runs lead to better TAO T above 200m
but show degradation at 300m. The K-filter runs show improved U
above 200m but have degradation below that.
===> Time-series should be compared to see how degradation occur.
[Dimitris]
- Akiko's T/P data is somehow different from what we had before.
===> Why the data is different needs to be understood. [Benny/Ichiro]
2) Adjoint
-----------
- Several postings have been made assessing the adjoint;
Subsurface T[7] is improved but SST isn't. The cause was initially
thought to be due to SSS, but instead it appears to be due to
lack of KPP[8] in the adjoint code. However, curiously, the simple
time-invariant prescribed mixing results in a better subsurface T[9]
simulation even though it has worse SST simulation than the
simulation with KPP. Cause is unknown.
- The wind and surface heat flux correction amounts to ~1 dyne/cm^2
in the tropics and 40 W/m^2, respectively; both are reasonable
with regards to expected errors in NCEP forcing.
- The gsgt error in surface temperature balance was corrected. Now
the sum of forcing terms match actual temperature change as in
this plot[10]. However, the ECCO gsgt package was found to have a
bug that results in a significant trend in sea level (off
Kuroshio). The bug was introduced somewhere along the main ECCO
branch after Tony's assimilation code branched off; i.e., the bug
is absent in Tony's code. (Regardless of this bug, the ECCO KPP
remains different from MITgcm KPP; somehow ECCO KPP requires
dQ/dt to be smaller for it to run.)
===> The ECCO gsgt needs to be debugged. [Benny/Dimitris]
3) Other
---------
- The control run (c20000630) has been extended to the 3rd quarter
of 2000. This extension is called c20010119 and is saved on
/silo.
- Transferring numerous large files to /silo (alhena) by ftp can be
too costly. Rather do rcp by asking Cris to enable rhost.
Requests must be made each time such large transfers are done.
- The backup software (legato) now works except for nautilus
(license problem) and guppy (rcp problem).
- VNC has been put on guppy and nautilus. VNC allows remote
display and operation of SUN/Linux/Windows window systems.
See posting here[11].
- Nick's assimilation with time-dependent Kalman filtering
(periodic P) did not improve the K-filter's skill by much. The
limitation in improving the high-frequency bottom pressure
(periods
-----------------------------------------------------------------------
[1] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t73h2d_c.ps
[2] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t73f5h_g_trend2.ps
[3] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t73gb_0214a_sim.ps
[4] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t83e_3.ps
[5] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/assim/106/1/2.html
[6] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/assim/106/2/3.html
[7] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/assim/110/3.html
[8] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/assim/110/4.html
[9] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/assim/110/5.html
[10] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/Benny/gsgt_1.gif
[11] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/comp/24.html
More information about the MITgcm-support
mailing list