[Liwg-core] CISM settings for Greenland coupled runs

William Lipscomb lipscomb at ucar.edu
Fri Jun 23 09:38:44 MDT 2017


Hi Jeremy,

I was in a rush to head home with Matthew yesterday afternoon, before we
had time to talk in detail about CISM config settings.  So I wanted to
follow up in an email.  My understanding is that there are three main
questions to settle:

(1) Is the isostasy scientifically validated?
(2) DIVA or Blatter-Pattyn?
(3) Which effective pressure parameterization for basal sliding?

On (1), I would defer to Sarah.  I'd just add that CISM isostasy is
numerically the same as Glimmer isostasy (with roundoff-level
differences).  The main difference is that now we can run it on multiple
processes with exact restart.

A question for Sarah: How would you like to handle the relx field?  Right
now we're assuming that the modern Greenland topography is in equilibrium
given the load.  Do you think this is a good enough assumption, or should
we try to acquire or compute a more accurate version of the relaxed
topography?

On (2), there are pros and cons each way, and it may come down to your
personal preference.  The main advantage of B-P is that it's formally a
more accurate approximation of Stokes flow, in that vertical variations in
membrane stresses are included in the momentum balance.  These vertical
variations can be important where the bed is frozen and the basal
topography is rough.  See Goldberg (2011, JGlac) Fig. 1 for an illustration
based on ISMIP-HOM Expt. B (the 2D version of Expt A).  Where there is fast
basal sliding as in Expts C/D, BP and depth-integrated results are very
similar (Goldberg Fig. 2).

The main advantages of DIVA are that it's faster (by an order of magnitude)
and has been run successfully for longer.  I and others have now run DIVA
for hundreds of thousands of years for Greenland, compared to just ~20,000
for BP.  There's no reason to expect one approximation to be more
numerically robust than the other, but we'd need to run BP for longer
before having the confidence we now have in DIVA.  As for cost, if you're
planning to run for ~1 M cpu-hr, then the cost of a 10,000-year BP run,
though greater than DIVA, would still be a small fraction of the total cost.

I suggest that you hold off on a decision until I can show you a clean
comparison of DIVA v. BP for a long Greenland spin-up with the parameter
settings we settle on.  Also, I can give you more precise numbers for
throughput.  I'll be working on this in the next couple of weeks.

I'm still working on (3).  My runs with the new RACMO2.3 SMB have a
Greenland volume that's too high by ~10%.  I talked yesterday with Jan and
Miren about this, and I think an important issue is that the new SMB has
positive values for peripheral glaciers that should be made negative for
purposes of an ice sheet spin-up exercise.  Once this issue is resolved, I
can go back to comparing effective pressure parameterizations, including
the fraction of the bed that is thawed and the frequency and magnitude of
oscillations.

On a closer look at Aschwanden et al. (2016), I see that the basal model
described in (6) is very simple, with a fixed, spatially uniform drainage
term (and no routing scheme). I think it would not be hard to implement
this formulation in CISM, so I'd like to try it and compare it to the
bmlt-based scheme I'm testing now.  The bmlt-based scheme is similar in
that it leads to a slippery bed where the basal melt rate exceeds 1 mm/yr,
but Andy's scheme may have more inertia (since you have to refreeze water
to increase the basal friction) and thus suppress oscillations.

The summary message is that I'm continuing to work to provide plots and
numbers to inform your decisions on config settings.  Please don't hesitate
to ask if you have more questions.

Bill L.

-- 
*****
William Lipscomb
Climate & Global Dynamics
National Center for Atmospheric Research
1850 Table Mesa Drive
Boulder, CO 80305
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cgd.ucar.edu/pipermail/liwg-core/attachments/20170623/bef873f4/attachment.html>


More information about the Liwg-core mailing list