dome
The dome tasks in polaris.tasks.landice.dome evolve an ice dome on a flat
bed with MALI’s shallow-ice and first-order velocity solvers (see
dome). add_dome_tasks() adds a smoke_test, decomp and
restart task for each solver.
framework
The tasks share a config file, dome.cfg, and the steps below.
init
The class polaris.tasks.landice.dome.init.Init is a step shared
by all the tasks. It makes a planar hexagonal mesh, culls and converts it,
turns it into a MALI mesh with polaris.landice.mesh.make_mali_mesh()
and adds the dome’s thickness, a flat bed and zero surface mass balance.
forward
The class polaris.tasks.landice.dome.forward.Forward is a
polaris.landice.model.MaliModelStep that runs MALI on the dome
mesh with a given velocity solver, from namelist.landice and
streams.landice in the package. For the first-order solver, it uses the
package’s albany_input.yaml.
validate
The class polaris.tasks.landice.dome.validate.Validate compares
the thickness and normal velocity in the output of two forward steps,
within given norms.
viz
The class polaris.tasks.landice.dome.viz.Viz plots the ice
thickness and surface speed at the end of a run.
smoke_test
The class polaris.tasks.landice.dome.smoke_test.SmokeTest runs
MALI on 4 cores, for 200 years with the shallow-ice solver and 2 years with
the first-order solver, and plots the results.
decomp
The class polaris.tasks.landice.dome.decomp.Decomp runs MALI on
1 and 4 cores and validates that the results are bit-for-bit with the
shallow-ice solver, and agree within Compass’s norms with the first-order
solver.
restart
The class polaris.tasks.landice.dome.restart.Restart runs MALI
for 2 years in full_run, and for 1 year in restart_run from the restart
file full_run writes after its first year. Both runs write output only at
the end of year 2, using streams.restart in the package, and the outputs
must be bit-for-bit with the shallow-ice solver. With the first-order
solver, Albany’s velocity depends slightly on how MPI tasks are placed on
cores (by about 1e-21 m/s), and the two runs are usually placed differently,
so the validate step uses the norms of decomp instead.