[Mitgcm-support] Note: November 9, 2000
mitgcm-support at dev.mitgcm.org
mitgcm-support at dev.mitgcm.org
Wed Jul 9 15:54:51 EDT 2003
HyperNews notification of new message. See:
http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/wkmtg/73.html
=======================================================================
NOTES FROM THE ASSIMILATION MEETING : 09 November 2000
1) BC K-filter
---------------
- Two bugs were found in the transition matrix derivation; The
fine-to-coarse transformation used incorrect layer
thicknesses. The time integration of displacement had an extra
time-step.
--> Leading (~100) eigenvalues of the corrected transition matrix
are now nearly identical with those of MOM (JGR '99) as shown in
this figure[1]. The leading eigenvalues are what dictate the
asymptotic error. Left column is eigenvalues of 1-day transition
matrix. Right column is same but for 3-days. Lower figures are
close-ups of upper figures. Solid black is MOM. Solid blue is
corrected MITgcm. Solid red is old 1-day transition matrix.
(Dash red is red times 1.06 [See 11/02/00[2]], dash blue is 3-day
transition matrix of 11/02/00.) The trailing eigenvalues of MOM
are still larger than MITgcm; Is this due to the "ringing"?
--> The spatial structure of the time-asymptotic simulation error
estimate (based on 1-day transition above and NCEP covariance) is
sensible as shown in this figure[3]. (Assumed 3-day wind error
decorrelation.)
--> The assimilation based on the corrected transition matrix shows
improvements as measured by the difference between simulation
model-data residuals and the innovation (dynamic update
residuals). The measure is shown as a function of time[4] and space[5].
The green curve in the temporal skill is the simulation residual
variance minus innovation variance. (Blue is the same but for the
barotropic filter. Purple is the difference between innovation
variance and filtered residual of the baroclinic filter. Red is
the same for the barotropic filter.) The negative skill away from
the tropical Pacific is due to sea level not being compared in the
plot at the correct time; Simulation is based on 12-hourly outputs
whereas assimilation is carried out every 6-hours. Significant
sea level variations occur over 6-hours in shallow regions that
were not sampled in the simulation. The negative skill east of
Taiwan may be due to excessive model error estimates in this
region, as seen in the simulation error estimate.
2) Adjoint
-----------
- The data uncertainty weights based on model-data comparisons did
make the 3-month assimilation better after all. This figure[6]
compares model estimates with TAO. Assimilation invariably is
closer to TAO than simulation.
- There was a bug in the flux constraint. Instead of adding the
cost every 10-days, the cost was added every time-step, thus
effectively reducing the uncertainty estimate of the flux. A
single adjoint iteration after correcting this bug lead to a 25%
decrease in cost. Much larger correction and better fit is
expected. Will need 10 more iterations for the 3-month
assimilation before the T/P meeting.
- The adjoint of the new KPP package now works as well as the old
package did. The adjoint has run OK for at least 3-days when
convection and shear instability are turned off.
3) BT filter
-------------
- Comparisons between results of the barotropic filter (kf014) and
bottom pressure gauges show improvements in coherence[7] at periods
longer than 20-days. The degradation at shorter periods may be
due to the comparisons being done with the filtered results, which
have discontinuities every 6-hours. Better comparisons may need
to wait until the smoother is run and/or after finer error
calibration.
=======================================================================
-----------------------------------------------------------------------
[1] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum_guppy/t20c.ps
[2] http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/wkmtg/72.html
[3] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t72.ps
[4] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t71d_1993.ps
[5] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/t71e_1993.ps
[6] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/group_poster_swt3.ps
[7] http://escher.JPL.NASA.GOV:2000/hosts/escher/escher2/medea/if/forum/CohSq.ps
More information about the MITgcm-support
mailing list