Understanding COSMO#

operational settings#

  • itype_turb = 3

  • itype_vdif = -1

  • itype_canopy=2

all physics settings in sfc_terra.f90#

  • itype_canopy = 2 (type of canopy parameterisation with respect to the surface energy balance)

  • itype_eisa = 2 (type of evaporation from impervious surfaces)

  • itype_evsl = 4 (type of parametrization of bare soil evaporation)

  • itype_heatcond = 3 (type of soil heat conductivity)

  • itype_hydbound = (type of soil water transport and ground water runoff)

  • itype_hydmod

  • itype_interception = 1 ( type of plant interception)

  • itype_mire = 0 ( Approach from Alla Yurova et al. 2014 - activate mire param. )

  • itype_root = 2 ( type of root distrubution )

  • itype_snow = we know about this

  • itype_trvg = 2 (type of vegetation transpiration parameterization)

  • itype_interception = 1 (type of plant interception)

some important subroutines#

  • turbdiff

  • turbtran

to add a new field#

  • I chose rain_gsp as my template variable. rain_gsp variable is a 2d diagnostic variable. I am copying exactly the same implementation but for 4 new variables

  • variables scratch_a, scratch_b, scratch_c, scratch_d

  • rain_gsp is used in ( i found it using grep -i -r 'rain_gsp' . | cut -d':' -f1 | sort | uniq) :

  • ./acc_global_data.f90

  • ./data_fields.f90

  • ./dfi_initialization.f90 ( not needed for us)

  • ./near_surface.f90 ( not needed for us )

  • ./organize_data.f90 ( only for predefined i/o - not needed for us)

  • ./src_allocation.f90

  • ./src_gridpoints.f90 (not for us)

  • ./src_meanvalues.f90 (not needed for us)

  • ./src_output.f90 (surprisingly, no snowpolino variables )

  • ./src_setup_vartab.f90

  • ./src_sing_local.f90 ( not needed for us)

  • On second though - I changed my mind and decided to follow sfhl_s as my template variable - why ? because rain_gsp does not exist as blocks and at the interface level (sfc_interface.f90), everything is like blocks.

  • ./acc_global_data.f90

  • ./data_block_fields.f90

  • ./data_fields.f90

  • ./organize_data.f90 ( not needed only for initial or restart etc )

  • ./sfc_interface.f90

  • ./sfc_tile_approach.f90

  • ./src_allocation.f90

  • ./src_block_fields_org.f90

  • ./src_input.f90

  • ./src_setup_vartab.f90 ( not sure if I have to do this)

  • ./src_slow_tendencies_rk.f90 ( not needed )

  • ./turb_diffusion.f90 ( not needed )

  • ./turb_interface.f90 ( not needed )

  • ./turb_transfer.f90 ( not needed )

  • ./turb_vertdiff.f90( not needed )

COMPILING COSMO#

module load python/3.7.4
. /project/g110/spack/user/tsa/spack/share/spack/setup-env.sh

COSMO_SPEC="cosmo@dev-build%gcc real_type=double cosmo_target=cpu ~cppdycore"
spack devbuildcosmo $COSMO_SPEC

# get and source env
spack build-env --dump cosmo.env $COSMO_SPEC -- #!one space after --

In the sbatch script

source cosmo.env
export REAL_TYPE=DOUBLE #FLOAT for float

RUNNING DDT FOR COSMO#

  1. get an allocation

salloc --ntasks=96 --partition=postproc --time=00:30:00

  1. do things similar to sbatch script

 ulimit -s unlimited
 ulimit -c unlimited
 source ./cosmo_cpu_sp.env
 export REAL_TYPE=FLOAT
rm YU* M*

  1. load and launch ddt

module load ddt
ddt

  1. Within DDT, make sure to

  • select MPI , and select SLURM generic

That is it !

Tips on DDT#

  • to step through the code, use Step over, step within and step out.

  • basically, let the code run a bit, then step out … keep stepping out until you reach a reasonable level in the code heirarcy. At that point, keep stepping over.