citibeth / icebin Goto Github PK

Set up a restart framework in GLINT2 that is compatible with ModelE, and works with PISM.

We shouldn't need to save much in GLINT2 (maybe just the reference date). We do need some API support in GLINT2. And we need to step PISM to the restart timepoint, and save in PISM. The PISM restart file needs to be as-of the LAST PISM TIMESTEP (which may be up to 1 month before the restart file time).

1. Add a --with-eigen compiler flag
2. If eigen is NOT being used, then don't link to it. And conditional-out code that's currently commented out in SparseMatrix.hpp and sparsemult.cpp

Currently, projection correction always happens. This is good for conserved quantities, but not non-conserved quantities. Search for EvEp and EpvE

The reference date of ModelE is not being succesfully transferred to PISM. This is due to problems on the PISM side (pism/pism#225), and also on the ModelE side (iyear1 is passed to GLINT2, but it's not set properly at the time it is passed).

Pism's IceGrid has the following: g.x, g.y, g.Mx, g.My, etc. Make sure these match with GLINT2's idea of the grid. Especially the actual x and y coordinates of the grid cells.

1. Things won't work so well if the run lasts long enough for the ice extent to change. I'm currently tracking this, so it will be easy to see when it does change. But to make the system run longer than that, Glint2/IceMint will need to load up the new mask, and then ModelE's Stieglitz model will have to be regridded from the old to the new.

This is currently hard-coded to 5m. It should really be made as a IceMint parameter, that gets set by the GCM.

Many changes, much will break... Reasons to upgrade:

1. Allows Neumann boundary condition. (But so far, the changes across the Stieglitz/PISM boundary are so small I don't think it would matter).
2. Get up-to-date on PISM error reporting, so I can get PISM stack traces working in a permanent way.
3. Good for the long term.

Consider using a PISM-like system that will write output / log files in glaciological units.

Currently the coupling assumes a Dirichlet boundary condition for PISM (which is the default). Re-do this to use PISM's Neumann Boundary Condition, which should be preferable in a coupled case where fluxes are passed between the two models.

gather_msg_array() template cleans up code from GCMCoupler::couple_to_ice(). But couple_to_ice() doesn't yet use it.

Bob,

Would you mind rephrasing your question so as to make it a bit more
specific, please? As it is I don't quite know what it is you would
like to know.

Thanks!

PS: I would start by looking at the initialization flow-chart
http://www.pism-docs.org/doxy/html/classIceModel.html#a74bc46b66b8383203287417335787c93
and following the path corresponding to "Is -i set? -> yes".

-- Constantine On Wed, May 7, 2014 at 10:45 PM, Fischer, Robert P (GISS-611.0)[COLUMBIA UNIVERSITY] [email protected] wrote:

Constantine,

Could you please give me a quick pointer to where/how restarts are
accomplished in PISM? At some point, I will want to couple this to ModelE's
restart capabilities. (Currently, I can't do coupled runs involving
restarts).

Thanks,
-- Bob

Currently, a number of variables have enthalpy units [J m-2]. This makes it hard to interpret output / log files. Consider converting to specific enthalpy [J m-2 kg-1] --- if not for internal variables, at least in output files.

Glint2 should provide surface temperature in both icy and ice-free areas.

PISM uses surface temperature in ice-free areas as a boundary condition for the heat equation modeling the temperature of the thin (~1 km) "thermal layer" of bedrock. This "bedrock thermal layer" covers the whole computational domain.

See the PISM's source code browser for details.

Right now the desm executable reports surface temperatures of 0 Kelvin in ice-free areas:

1. I'm not entirely happy with the fields produced when regridding from ice to elevation grids. It seems too choppy. This could (maybe) be fixed by adding a smoothness component to the objective function. It's worth playing around with.

Use the reshape() template in giss/blitz.hpp instead.

PlotterI and PlotterA work. But plotterE does not; needs to be converted to "new" regridder generation scheme.