trun-3.rst - pism - [fork] customized build of PISM, the parallel ice sheet model (tillflux branch)
 (HTM) git clone git://src.adamsgaard.dk/pism
 (DIR) Log
 (DIR) Files
 (DIR) Refs
 (DIR) LICENSE
       ---
       trun-3.rst (2797B)
       ---
            1 .. include:: ../../global.txt
            2 
            3 .. _sec-higherresrun:
            4 
            5 Third run: higher resolution
            6 ----------------------------
            7 
            8 Now we change one key parameter, the grid resolution. Model results differ even when the
            9 only change is the resolution. Using higher resolution "picks up" more detail in the bed
           10 elevation and climate data.
           11 
           12 If you can let it run overnight, do
           13 
           14 .. literalinclude:: scripts/run-3.sh
           15    :language: bash
           16    :lines: 3-
           17 
           18 This run might take 4 to 6 hours. However, supposing you have a larger parallel computer,
           19 you can change "``mpiexec -n 4``" to "``mpiexec -n N``" where ``N`` is a substantially
           20 larger number, up to 100 or so with an expectation of reasonable scaling on this grid
           21 :cite:`BBssasliding`, :cite:`DickensMorey2013`.
           22 
           23 .. figure:: figures/g10km-10ka-hy-usurf-csurf-cbase.png
           24    :name: fig-secondoutputfiner
           25 
           26    Fields from output file ``g10km_10ka_hy.nc``.
           27 
           28    Compare to :numref:`fig-secondoutputcoarse`, which only differs by resolution.
           29 
           30    :Left: :var:`usurf` in meters.
           31    :Middle: :var:`velsurf_mag` in m/year.
           32    :Right: :var:`velbase_mag` in m/year.
           33 
           34 Some fields from the result ``g10km_10ka_hy.nc`` are shown in
           35 :numref:`fig-secondoutputfiner`. :numref:`fig-csurfvsobserved` also compares observed
           36 velocity to the model results from 20 km and 10 km grids. As a different comparison,
           37 :numref:`fig-ivolboth` shows ice volume time series ``ice_volume_glacierized`` for 20 km and
           38 10 km runs done here. We see that this result depends on resolution, in particular because
           39 higher resolution grids allow the model to better resolve the flux through
           40 topographically-controlled outlet glaciers (compare :cite:`Pfefferetal2008`). However,
           41 because the total ice sheet volume is a highly-averaged quantity, the
           42 ``ice_volume_glacierized`` difference from 20 km and 10 km resolution runs is only about one
           43 part in 60 (about 1.5%) at the final time. By contrast, as is seen in the near-margin ice
           44 in various locations shown in :numref:`fig-csurfvsobserved`, the ice velocity at a
           45 particular location may change by 100\% when the resolution changes from 20 km to 10 km.
           46 
           47 Roughly speaking, the reader should only consider trusting model results which are
           48 demonstrated to be robust across a range of model parameters, and, in particular, which
           49 are shown to be relatively-stable among relatively-high resolution results for a
           50 particular case. Using a supercomputer is justified merely to confirm that
           51 lower-resolution runs were already "getting" a given feature or result.
           52 
           53 .. figure:: figures/ivol-both-g20km-g10km.png
           54    :name: fig-ivolboth
           55 
           56    Time series of modeled ice sheet volume ``ice_volume_glacierized`` on 20km and 10km grids.
           57    The present-day ice sheet has volume about `2.9\times 10^6\,\text{km}^3`
           58    :cite:`BamberLayberryGogenini`, the initial value seen in both runs.