def backup_scripts():
current = cluster.scriptsdir
main_a = cluster.archivedir() + '/DART-WRF/'
old_a = main_a + '/old/'
os.makedirs(cluster.archivedir(), exist_ok=True)
os.makedirs(main_a, exist_ok=True)
os.makedirs(old_a, exist_ok=True)
def func(a, b, method): # call method if not link or directory
if os.path.islink(a) or os.path.isdir(a):
pass
else:
method(a, b)
# archive existing files
for f in os.listdir(main_a):
func(os.path.join(main_a, f), old_a + '/' + f, shutil.move)
# reproducibility
for f in ['scheduler.py', 'config/clusters.py', 'config/cfg.py']:
fname = os.path.basename(f)
func(current + '/../' + f, main_a + '/' + fname, shutil.copy)
for f in os.listdir(current):
func(os.path.join(current, f), main_a + '/', shutil.copy)
def run(iens, begin, end, hist_interval=5, radt=5, archive=True):
"""
Args:
archive (bool): if True, write to archivedir of experiment
if False, write to WRF run directory
"""
rundir = cluster.wrf_rundir(iens)
print(rundir)
copy(cluster.namelist, rundir + '/namelist.input')
sed_inplace(rundir + '/namelist.input', '<dx>', str(int(exp.model_dx)))
#sed_inplace(rundir+'/namelist.input', '<timestep>', str(int(exp.timestep)))
sed_inplace(rundir + '/namelist.input', '<hist_interval>',
str(int(hist_interval)))
sed_inplace(rundir + '/namelist.input', '<radt>', str(int(radt)))
if archive:
archdir = cluster.archivedir() + begin.strftime('/%Y-%m-%d_%H:%M/' +
str(iens) + '/')
os.makedirs(archdir, exist_ok=True)
else:
archdir = './'
print('namelist for run from', begin, end, 'output to', archdir)
sed_inplace(rundir + '/namelist.input', '<archivedir>', archdir)
# set times
for k, v in {
'<y1>': '%Y',
'<m1>': '%m',
'<d1>': '%d',
'<HH1>': '%H',
'<MM1>': '%M'
}.items():
sed_inplace(rundir + '/namelist.input', k, begin.strftime(v))
for k, v in {
'<y2>': '%Y',
'<m2>': '%m',
'<d2>': '%d',
'<HH2>': '%H',
'<MM2>': '%M'
}.items():
sed_inplace(rundir + '/namelist.input', k, end.strftime(v))
#########################
if archive:
init_dir = cluster.archivedir() + begin.strftime(
'/%Y-%m-%d_%H:%M/') + str(iens)
os.makedirs(init_dir, exist_ok=True)
try:
print('copy wrfinput of this run to archive')
wrfin_old = rundir + '/wrfinput_d01'
wrfin_arch = init_dir + '/wrfinput_d01'
copy(wrfin_old, wrfin_arch)
print('copy namelist to archive')
copy(rundir + '/namelist.input', init_dir + '/namelist.input')
except Exception as e:
warnings.warn(str(e))
def create_satimages(init_time, depends_on=None):
s = my_Slurm("pRTTOV", cfg_update={"ntasks": "48", "time": "30"})
s.run(cluster.python + ' /home/fs71386/lkugler/RTTOV-WRF/run_init.py ' +
cluster.archivedir() + init_time.strftime('/%Y-%m-%d_%H:%M/'),
depends_on=[depends_on])
def assimilate(assim_time,
prior_init_time,
prior_path_exp=False,
depends_on=None):
"""Creates observations from a nature run and assimilates them.
Args:
assim_time (dt.datetime): timestamp of prior wrfout files
prior_init_time (dt.datetime):
timestamp to find the directory where the prior wrfout files are
prior_path_exp (bool or str):
put a `str` to take the prior from a different experiment
if False: use `archivedir` (defined in config) to get prior state
if str: use this directory to get prior state
"""
if not prior_path_exp:
prior_path_exp = cluster.archivedir()
elif not isinstance(prior_path_exp, str):
raise TypeError('prior_path_exp either str or False, is ' +
str(type(prior_path_exp)))
# prepare state of nature run, from which observation is sampled
#s = my_Slurm("prepNature", cfg_update=dict(time="2"))
#id = s.run(cluster.python+' '+cluster.scriptsdir+'/prepare_nature.py '
# +time.strftime('%Y-%m-%d_%H:%M'), depends_on=[depends_on])
# prepare prior model state
s = my_Slurm("preAssim", cfg_update=dict(time="2"))
id = s.run(cluster.python + ' ' + cluster.scriptsdir + '/pre_assim.py ' +
assim_time.strftime('%Y-%m-%d_%H:%M ') +
prior_init_time.strftime('%Y-%m-%d_%H:%M ') + prior_path_exp,
depends_on=[depends_on])
# prepare nature run, generate observations
s = my_Slurm("Assim",
cfg_update={
"nodes": "1",
"ntasks": "96",
"time": "30",
"mem": "300G",
"ntasks-per-node": "96",
"ntasks-per-core": "2"
})
id = s.run(cluster.python + ' ' + cluster.scriptsdir +
'/assim_synth_obs.py ' + assim_time.strftime('%Y-%m-%d_%H:%M'),
depends_on=[id])
# # actuall assimilation step
# s = my_Slurm("Assim", cfg_update=dict(nodes="1", ntasks="48", time="50", mem="200G"))
# cmd = 'cd '+cluster.dartrundir+'; mpirun -np 48 ./filter; rm obs_seq_all.out'
# id = s.run(cmd, depends_on=[id])
# s = my_Slurm("archiveAssim", cfg_update=dict(time="10"))
# id = s.run(cluster.python+' '+cluster.scriptsdir+'/archive_assim.py '
# + assim_time.strftime('%Y-%m-%d_%H:%M'), depends_on=[id])
s = my_Slurm("updateIC", cfg_update=dict(time="8"))
id = s.run(cluster.python + ' ' + cluster.scriptsdir +
'/update_wrfinput_from_filteroutput.py ' +
assim_time.strftime('%Y-%m-%d_%H:%M ') +
prior_init_time.strftime('%Y-%m-%d_%H:%M ') + prior_path_exp,
depends_on=[id])
return id
submitting jobs into SLURM queue
"""
import os, sys, shutil, glob
import datetime as dt
from slurmpy import Slurm
from config.cfg import exp, cluster
from scripts.utils import script_to_str, symlink
# necessary to find modules in folder, since SLURM runs the script elsewhere
sys.path.append(os.getcwd())
# allow scripts to access the configuration
symlink(cluster.scriptsdir + '/../config', cluster.scriptsdir + '/config')
log_dir = cluster.archivedir() + '/logs/'
slurm_scripts_dir = cluster.archivedir() + '/slurm-scripts/'
print('logging to', log_dir)
print('scripts, which are submitted to SLURM:', slurm_scripts_dir)
def my_Slurm(*args, cfg_update=dict(), **kwargs):
"""Shortcut to slurmpy's class; keep certain default kwargs
and only update some with kwarg `cfg_update`
see https://github.com/brentp/slurmpy
"""
return Slurm(*args,
slurm_kwargs=dict(cluster.slurm_cfg, **cfg_update),
log_dir=log_dir,
scripts_dir=slurm_scripts_dir,
**kwargs)
1) prepare nature run for DART
optional: 2) calculate obs-error from parametrization
3) create obs_seq.in with obs-errors from 2)
4) generate actual observations (obs_seq.out) with obs_seq.in from 3)
- calculate obs-error from parametrization
1) create obs_seq.in with obs error=0
2) calculate y_nat = H(x_nature) and y_ens = H(x_ensemble)
3) calculate obs error as function of y_nat, y_ensmean
Assumptions:
- x_ensemble is already linked for DART to advance_temp<iens>/wrfout_d01
"""
time = dt.datetime.strptime(sys.argv[1], '%Y-%m-%d_%H:%M')
archive_time = cluster.archivedir()+time.strftime('/%Y-%m-%d_%H:%M/')
os.chdir(cluster.dartrundir)
os.system('rm -f obs_seq.in obs_seq.out obs_seq.final') # remove any existing observation files
n_stages = len(exp.observations)
for istage, obscfg in enumerate(exp.observations):
print('running observation stage', istage, obscfg)
archive_stage = archive_time + '/assim_stage'+str(istage)+'/'
n_obs = obscfg['n_obs']
n_obs_3d = n_obs * len(obscfg['heights'])
sat_channel = obscfg.get('sat_channel', False)
error_generate = obscfg['error_generate']
error_assimilate = obscfg['error_assimilate']
import os, sys, warnings, glob
import datetime as dt
from config.cfg import exp, cluster
from utils import symlink, copy_scp_srvx8, copy, mkdir, mkdir_srvx8, clean_wrfdir
# if cluster.name != 'srvx8':
# copy = copy_scp_srvx8
# mkdir = mkdir_srvx8
time = dt.datetime.strptime(sys.argv[1], '%Y-%m-%d_%H:%M')
try:
print('archive obs space diagnostics')
savedir = cluster.archivedir() + '/obs_seq_final/'
mkdir(savedir)
copy(cluster.dartrundir + '/obs_seq.final',
savedir + time.strftime('/%Y-%m-%d_%H:%M_obs_seq.final'))
except Exception as e:
warnings.warn(str(e))
try:
print('archive regression diagnostics')
savedir = cluster.archivedir() + '/reg_factor/'
mkdir(savedir)
copy(cluster.dartrundir + '/reg_diagnostics',
savedir + time.strftime('/%Y-%m-%d_%H:%M_reg_diagnostics'))
except Exception as e:
warnings.warn(str(e))
print('archive model state')
try:
"""Apply observation operator to some ensemble state
i.e. wrfout files in an archive directory
output: saves obs_seq.final files
these contain the Hx values
usually applied to 1 min forecasts to assess the posterior analysis quality
(analysis+1minute = 'posterior')
"""
if __name__ == '__main__':
prev_forecast_init = dt.datetime.strptime(sys.argv[1], '%Y-%m-%d_%H:%M')
time = dt.datetime.strptime(sys.argv[2], '%Y-%m-%d_%H:%M')
exppath_firstguess = cluster.archivedir()
print(prev_forecast_init, time)
# link ensemble states to run_DART directory
# we want the observation operator applied to these states!
pre_assim.run(time, prev_forecast_init, exppath_firstguess)
savedir = cluster.archivedir()+'/obs_seq_final_1min/'
n_stages = len(exp.observations)
for istage, obscfg in enumerate(exp.observations):
n_obs = obscfg['n_obs']
sat_channel = obscfg.get('sat_channel', False)
obscfg['folder_obs_coords'] = False