search

ACCESS-NRI / accessdev-Trac-archive

Archive accessdev Trac contents as issues
Apache License 2.0
0 stars 0 forks source link

Starting from another run #344

Closed penguian closed 6 years ago

penguian commented 6 years ago

resolution_fixed | by mrd599@nci.org.au

For CMIP6 it will be necessary to start a new run from a given year of the PI control run.

This capability would also be useful for testing.

In the old script environment there was a separate script for this, but it should be possible to make it part of the suite.

Issue migrated from trac:344 at 2024-01-31 18:32:19 +1100

penguian commented 6 years ago

@martin.dix@anu.edu.au commented

In current suites, the first run uses the INSTALL_COLD task which copies OASIS files (a2i.nc, i2a.nc, o2i.nc), ICE_RUNDIR/INPUT/monthly_sstsss.nc and OCN_RUNDIR/INPUT/ocean_temp_salt.res.nc from ~access/access-cm2/input_b. The UM uses share/data/RUNID.astart.

For later runs the access_coupled script copies the OASIS, CICE and MOM restart files from the archive directory. The UM restart comes from share/data/History_Data where it was written by the previous run.

For UM environment variable CONTINUE is unset for the first run and true for later runs. This controls whether the model uses the xhist file to get the name of the restart file.

For CICE, JOBNUM=1 on the first run and 2 later. This controls whether or not the model is initialised from monthly_sstsss.nc. There are also ICE_RESTART and ICE_RUNTYPE variables.

For MOM the TRUNCATE environment variable is set true for the first year.

For restarting from another run, the UM CONTINUE needs to be set false, but CICE variables should be set as for a restarted run. CONTINUE=false means that the access_coupled script won't try to copy the restart files.

penguian commented 6 years ago

@martin.dix@anu.edu.au commented

There are two cases of interest

  1. Exact restart of another run that reproduces exactly (e.g. for debugging)
  2. Restart with changed model year (e.g. start from year 101 of PI control as 1851 for historical).
  1. Suite u-ar277 is a copy of u-al623. Adds a new script warm_restart.sh and variables in rose-suite.conf 
    WARM_RESTART=true
WARM_RESTART_DIR='/short/p66/pbd562/archive/al623'
WARM_RESTART_RUNID='al623'
WARM_RESTART_DATE='02010101'

For an exact restart, set RECON=false. The install_warm task replaces install_cold and the warm_restart.sh script copies the required restart files from WARM_RESTART_DIR.

In this case a restarted one day run matches the original run exactly (comparing solver diagnostics with original model output from /short/p66/pbd562/cylc-run/u-al623/work/02010101T0000Z/coupled/ATM_RUNDIR/pe_output/al623.fort6.pe000.

Note that in this case the run length (RUNLEN) is relative to the restart date. E.g. with

RUNLEN='P100Y'
WARM_RESTART_DATE='02010101'

it will run an extra 100 years to year 300.

  1. To restart with changed model year, set BASIS='18510101T0000Z' and RECON=true.
penguian commented 6 years ago

@martin.dix@anu.edu.au changed _comment0 which not transferred by tractive

penguian commented 6 years ago

@martin.dix@anu.edu.au commented

Testing with the reconfiguration as in 2. above caused the model to crash when writing the first dump file. This was fixed by the Met Office in vn10.7 https://code.metoffice.gov.uk/trac/um/ticket/2657.

I added this fix to the vn10.6_access_cm2 branch as [45508].

penguian commented 6 years ago

@martin.dix@anu.edu.au set owner to mrd599

penguian commented 6 years ago

@martin.dix@anu.edu.au changed status from assigned to closed

penguian commented 6 years ago

@martin.dix@anu.edu.au set resolution to fixed