Diferencia entre revisiones de «WRF4L»

Revisión actual - 12:26 30 oct 2025

There is a far more powerful tool to manage work-flow of WRF4G.

Lluís expanded this work to other HPCs and models. See more details in this GIT repository WRFL

Lluís developed a less powerful one which is here described

WRF work-flow management is done via multiple scripts (these are the specifics for hydra [CIMA cluster]):

EXPERIMENTparameters.txt: General ASCII file which configures the experiment and chain of simulations (chunks). This is the unique file to modify
launch_experiment.pbs: PBS-queue job which prepares the experiment of the environment
wrf4l_prepare_WPS.pbs: PBS-queue job which prepares the WPS section of the model: ungrib.exe, metgrid.exe, real.exe
wrf4l_prepare_WRF.pbs: PBS-queue job which prepares the WRF section of the model: wrf.exe
wrf4l_prepare_nc2wps.pbs: PBS-queue job which prepare the execution of nc2wps for the chunk
wrf4l_prepare_ungrib.pbs: PBS-queue job which prepare the execution of ungrib for the chunk
wrf4l_launch_nc2wps.pbs: PBS-queue job which launches nc2wps: netcdf2wps
wrf4l_launch_ungrib.pbs: PBS-queue job which launches ungrib: ungrib.exe
wrf4l_launch_metgrid.pbs: PBS-queue job which launches metgrid: metgrid.exe
wrf4l_launch_real.pbs: PBS-queue job which launches real: real.exe
wrf4l_launch_wrf.pbs: PBS-queue job which launches wrf: wrf.exe
wrf4l_prepare-next_nc2wps.pbs: PBS-queue job which prepare the execution of nc2wps for the next chunk
wrf4l_prepare-next_ungrib.pbs: PBS-queue job which prepare the execution of ungrib for the next chunk
wrf4l_finishin_simulation.pbs: PBS-queue job which ends the simulation of the chunk

These PBS jobs are executed consecutively and they are put on hold until the precedent step is finished.

There is a folder called /share/tools/workflows/components with shell and python scripts necessary for the work-flow management

An experiment which contains a period of simulation is divided by chunks small pieces of times which are manageable by the model. The work-flow follows these steps using run_experiments.pbs:

Copy and link all the required files for a given chunk of the whole period of simulation following the content of EXPERIMENTparameters.txt

All the scripts are located in hydra at:

/share/tools/workflows/WRF4L/hydra

How to simulate

Assuming that we have already defined the domains of simulation and that we have a suited namelist.wps and namelist.input WRF namelist files in a folder called $GEOGRID. We will gather all this information in the storage folder for the experiment (let's assume called $STORAGEdir) and we create there a folder for the information of the domains.

mkdir ${STORAGEdir}/domains
cp ${GEOGRID}/geo_em.d* ${STORAGEdir}/domains
cp ${GEOGRID}/namelist.wps ${STORAGEdir}/domains
cp ${GEOGRID}/namelist.input ${STORAGEdir}/domains

We then move to the ${STORAGEdir}/domains and create a template file for the namelist.wps

cd ${STORAGEdir}/domains
cp namelist.wps namelist.wps_template
diff namelist.wps namelist.wps_template 
4,6c4,6
<  start_date = '2010-06-01_00:00:00', '2010-06-01_00:00:00'
<  end_date   = '2010-06-16_00:00:00', '2010-06-16_00:00:00'
<  interval_seconds = 10800,
---
>  start_date = 'iyr-imo-ida_iho:imi:ise', 'iyr-imo-ida_iho:imi:ise',
>  end_date   = 'eyr-emo-eda_eho:emi:ese', 'eyr-emo-eda_eho:emi:ese', 
>  interval_seconds = infreq

In case we are going to use nc2wps to generate the intermediate files (ungrib.exe output directly generated from netCDF files), we need to generate a template for the namelist.nc2wps. There are different options available in /share/tools/workflows/WRF4L/hydra. In this example we use the one prepared for ERA5 data as it is kept in Papa-Deimos:

cp /share/tools/workflows/WRF4L/hydra/namelist.nc2wps_template_ERA5 ${STORAGEdir}/domains/namelist.nc2wps_template

Creation of a new folder from where launch the experiment [ExperimentName] (g.e. somewhere at salidas/ folder)

$ mkdir salidas/[ExperimentName]
cd salidas/[ExperimentName]

copy essential WRF4L files to this folder

cp /share/tools/workflows/WRF4L/hydra/EXPERIMENTparameters.txt ./
cp /share/tools/workflows/WRF4L/hydra/launch_experiment.pbs ./

Edit the configuration/set-up of the simulation of the experiment (e.g. period of simulation, version of WPS and WRF to use, additional namelist.input parameters, multiple folders: storageHOME, runHOME, MPI configuration, ...)

$ vim EXPERIMENTparameters.txt

Change the name of the user in the launch_experiment.pbs

diff launch_experiment.pbs /share/tools/workflows/WRF4L/hydra/launch_experiment.pbs
11c11
< #PBS -M lluis.fita@cima.fcen.uba.ar
---
> #PBS -M [user]@cima.fcen.uba.ar

Launch the experiment

$ qsub launch_experiment.pbs

When it is running one would have multiple jobs, see for example for an experiment called 'MPIsimplest' and it simulation labelled 'urban' (finished `C', runnig `R', and hold `H'):

$ qstat -u $USER

hydra: 
                                                                         Req'd  Req'd   Elap
Job ID               Username Queue    Jobname          SessID NDS   TSK Memory Time  S Time
-------------------- -------- -------- ---------------- ------ ----- --- ------ ----- - -----
72507.hydra                experiment_relay lluis.fita      00:00:07 C larga          
72508.hydra                ...implest-urban lluis.fita      00:00:05 C larga          
72509.hydra                ...implest-urban lluis.fita      00:00:05 C larga          
72510.hydra                ...implest-urban lluis.fita      00:00:08 C larga          
72511.hydra                ...implest-urban lluis.fita             0 R larga          
72512.hydra                ...implest-urban lluis.fita             0 H larga          
72513.hydra                ...implest-urban lluis.fita             0 H larga          
72514.hydra                ...implest-urban lluis.fita             0 H larga          
72515.hydra                ...implest-urban lluis.fita             0 H larga          
72516.hydra                ...implest-urban lluis.fita             0 H larga          
72518.hydra                ...implest-urban lluis.fita             0 H larga          
72519.hydra                experiment_relay lluis.fita             0 H larga

In case of crash of the simulation, after fixing the issue, go to [expHOME]/[ExpName]/[SimName] and re-launch the experiment (after the first run the scratch is switched automatically to `false')

$ qsub launch_experiment.pbs

Checking the experiment

Once the experiment runs, one needs to look on (following name of the variables from EXPERIMENTparameters.txt

[runHOME]/[ExpName]/[SimName]: Will content the copies of the templates namelist.wps, namelist.input and files:
- chunk_attemps.inf which counts how many times a chunk has been attempted to be run (if it reached 4 times, the WRF4L is stopped)
- chunk_nums.inf which keeps track of the jobs and periods used for each step for a given chunk
[runHOME]/[ExpName]/[SimName]/ungrib: actual folder where the computing nodes run ungrib (either ungrib.exe or netcdf2wps)
[runHOME]/[ExpName]/[SimName]/metgrid: actual folder where the computing nodes run metgrd (metgrid.exe)
[runHOME]/[ExpName]/[SimName]/run: actual folder where the computing nodes run the model (real.exe and wrf.exe)
[runHOME]/[ExpName]/[SimName]/run/outreal/: folder with the output of real.exe
[storageHOME]/[ExpName]/[SimName]/outwrf/: folder with the output of wrf.exe

NOTE: ungrib, metgrid and real outputs are not kept in the storage folder

When something went wrong

If there has been any problem check the last chunk to try to understand what happens and where the problem comes from:

rsl.[error/out].[nnnn]: These are the files which content the standard output while running the model. One file for each process. If the problem was something related to model execution and it has been prepared for the error, a correct message must appear. (look first for the largest files...

$ ls -lrS rsl.error.*

run_wrf.log: These are the files which content the standard output of the model. Search for `segmentation faults' in form of (it might differ):

forrtl: error (63): output conversion error, unit -5, file Internal Formatted Write
Image              PC                Routine            Line        Source
wrf.exe            00000000032B736A  Unknown               Unknown  Unknown
wrf.exe            00000000032B5EE5  Unknown               Unknown  Unknown
wrf.exe            0000000003265966  Unknown               Unknown  Unknown
wrf.exe            0000000003226EB5  Unknown               Unknown  Unknown
wrf.exe            0000000003226671  Unknown               Unknown  Unknown
wrf.exe            000000000324BC3C  Unknown               Unknown  Unknown
wrf.exe            0000000003249C94  Unknown               Unknown  Unknown
wrf.exe            00000000004184DC  Unknown               Unknown  Unknown
libc.so.6          000000319021ECDD  Unknown               Unknown  Unknown
wrf.exe            00000000004183D9  Unknown               Unknown  Unknown

in [expHOME]/[SimName], check the output of the PBS jobs. Which are called:
- experiment_relay.o[nnnn]: output of the launch_experiment.pbs
- prepare_wps_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_prepare_WPS.pbs
- prepare_wrf_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_prepare_WRF.pbs
- prep_ncwps_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_prepare_nc2wps.pbs
- prep_ungrib_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_prepare_ungrib.pbs
- prep-nxt_ncwps_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_prepare_nc2wps.pbs for the following chunk
- prep-nxt_ungrib_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_prepare_ungrib.pbs for the following chunk
- finish_sim_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_finishing_simulation.pbs

in [runHOME]/[SimName], check the output of the PBS jobs. Which are called:
- ungrib/nc2wps_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_launch_nc2wps.pbs
- ungrib/ungrib_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_launch_ungrib.pbs
- metgrid/metgrid_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_launch_metgrid.pbs
- run/wrf_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_launch_real.pbs
- run/wrf_[ExpName]-[SimName].o[nnnn]: output of the wrf4l_launch_wrf.pbs

All the logs and job outputs are kept in ${storageHOME}/[ExpName]/[SimName]/joblogs with a compressed file for ecah chunk and attemp as ${chunk_period}-${Nattempts}_joblogs.tar

EXPERIMENTSparameters.txt

This ASCII file configures all the simulation. It assumes:

Required files, forcings, storage, compiled version of the code might be at different machines.
There is a folder with a given template version of the namelist.input which will be used and changed accordingly to the requirement of the experiments

Location of the WRF4L main folder (example for hydra)

# Home of the WRF4L
wrf4lHOME=/share/tools/workflows/WRF4L

Name of the machine where the experiment is running (NOTE: a folder with specific work-flow must exist as $wrf4lHOME/${HPC})

# Machine specific work-flow files for $HPC (a folder with specific work-flow must exist as $wrf4lHOME/${HPC})
HPC=hydra

Name of the compiler used to compile the model (NOTE: a file called $wrf4lHOME/arch/${HPC}_${compiler}.env must exist)

# Compilation (a file called $wrf4lHOME/arch/${HPC}_${compiler}.env must exist)
compiler=intel

Name of the experiment

# Experiment name
ExpName = WRFsensSFC

Name of the simulation. Here is understood that a given experiment could have the model configured with different set-ups (here identified with a different name of simulation)

# Simulation name
SimName = control

Should this simulation be run from the beginning or not. If it is set to `true', it will remove all the pre-existing content of the folder [ExpName]/[SimName] in the running and in the storage spaces. Be careful. In case of `false' simulation will continue from the last successful ran chunk (checking the restart files).

# Start from the beginning (keeping folder structure)
scratch = false

Period of the simulation of the simulation (In this example from 1958 Jan 1st to 2015 Dec 31)

# Experiment starting date
exp_start_date = 19790101000000
# Experiment ending date
exp_end_date = 20150101000000

Length of the chunks (do not make chunks larger than 1-month!!)

# Chunk Length [N]@[unit]
#  [unit]=[year, month, week, day, hour, minute, second]
chunk_length = 1@month

Selection of the machines and users to each machine where the different requirement files are located and the output should be placed.

NOTE: this will only work if one set-up the .ssh public/private keys in each involved USER/HOST.
NOTE 2: All the forcings, compiled code, ... are already at hydra at the common space called share
NOTE 3: From the computing nodes, one can not access to the /share folder and to any of the CIMA's storage machines: skogul, freyja, ... For that reason, one need to use these system of [USER]@[HOST] accounts. *.pbs scripts uses a series of wrappers of the standard functions: cp, ln, ls, mv, .... which manage them `from' and `to' different pairs of [USER]@[HOST]. NOTE: This will only work if the public/private ssh key pairs have been set-up (see more details at llaves_ssh)

# Hosts
#   list of different hosts and specific user
#     [USER]@[HOST]
#   NOTE: this will only work if public keys have been set-up
##
# Host with compiled code, namelist templates
codeHOST=lluis.fita@hydra
# forcing Host with forcings (atmospherics and morphologicals)
forcingHOST=lluis.fita@hydra
# output Host with storage of output (including restarts)
outHOST=lluis.fita@hydra

Templates of the configuration of WRF: namelist.wps, namelist.input files. NOTE: they will be changed according to the content of EXPERIMENTparameters.txt like period of the chunk, atmospheric forcing, differences of the set-up, ... (located in the [codeHOST]

# Folder with the `namelist.wps', `namelist.input' and `geo_em.d[nn].nc' of the experiment
domainHOME = /home/lluis.fita/salidas/EXPS/estudios/dominmios/SA50k

Folder where the workflow will be launched (on top of that there will be one more [SimName])

# Experiment folder
expHOME=/home/lluis.fita/salidas/estudios/MPIsimplest/exp

Folder where the WRF model will run in the computing nodes (on top of that there will be two folders [ExpName]/[SimName]). WRF will run at the folder [SimName]/run

# Running folder
runHOME = /home/lluis.fita/salidas/RUNNING

Folder with the compiled version of the WPS (located at [codeHOST])

# Folder with the compiled source of WPS
wpsHOME = /opt/wrf/WRF-4.5/intel/2021.4.0/dmpar/WPS

Folder with the compiled version of the WRF (located at [codeHOST])

# Folder with the compiled source of WRF
wrfHOME = /share/WRF/WRFV3.9.1/ifort/dmpar/WRFV3

Folder to storage all the output of the model (history files, restarts and compressed file with content of the configuration and the standard output of the given run). The content of the folder will be organized by chunks (located at [storageHOST]) with two folders called [ExpName]/[SimName].

# Storage folder of the output
storageHOME = /home/lluis.fita/salidas/EXPS

Wether modules should be load (not used for hydra)

# Modules to load ('None' for any)
modulesLOAD = None

Names of the files used to check that the chunk has properly ran

# Model reference output names (to be used as checking file names)
nameLISTfile = namelist.input # namelist
nameRSTfile = wrfrst_d01_ # restart file
nameOUTfile = wfrout_d01_ # output file

Extensions of the files with the configuration of WRF (to be retrieved from codeHOST and domainHOME)

# Extensions of the files with the configuration of the model
configEXTS = wps:input

To continue from a previous chunk one needs to use the `restart' files. But they need to be renamed, because otherwise they will be re-written. Here one specifies the original name of the file [origFile] and the name to be used to avoid the re-writting [destFile]. It uses a complex bash script which even can deal with the change of dates according to the period of the chunk (':' list of [origFile]@[destFile]). They will located at the [storageHOST]

# restart file names
# ':' list of [tmplrstfilen|[NNNNN1]?[val1]#[...[NNNNNn]?[valn]]@[tmpllinkname]|[NNNNN1]?[val1]#[...[NNNNNn]?[valn]]
#    [tmplrstfilen]: template name of the restart file (if necessary with [NNNNN] variables to be substituted)
#      [NNNNN]: section of the file name to be automatically substituted
#        `[YYYY]': year in 4 digits
#        `[YY]': year in 2 digits
#        `[MM]': month in 2 digits
#        `[DD]': day in 2 digits
#        `[HH]': hour in 2 digits
#        `[SS]': second in 2 digits
#        `[JJJ]': julian day in 3 digits
#      [val]: value to use (which is systematically defined in `run_OR.pbs')
#        `%Y%': year in 4 digits
#        `%y%': year in 2 digits
#        `%m%': month in 2 digits
#        `%d%': day in 2 digits
#        `%h%': hour in 2 digits
#        `%s%': second in 2 digits
#        `%j%': julian day in 3 digits
#    [tmpllinkname]: template name of the link of the restart file (if necessary with [NNNNN] variables to be substituted)
rstFILES=wrfrst_d01_[YYYY]-[MM]-[DD]_[HH]:[MI]:[SS]|YYYY?%Y#MM?%m#DD?%d#HH?%H#MI?%M#SS?%S@wrfrst_d01_[YYYY]-[MM]-[DD]_[HH]:[MI]:[SS]|YYYY?%Y#MM?%m#DD?%d#HH?%H#MI?%M#SS?%S

Folder with the input data (located at [forcingHOST]).

# Folder with the input morphological forcing data
indataHOME = /share/DATA/re-analysis/ERA-Interim

Format of the input data and name of files

# Data format (grib, nc)
indataFMT= grib
# For `grib' format
#   Head and tail of indata files names.
#     Assuming ${indataFheader}*[YYYY][MM]*${indataFtail}.[grib/nc]
indataFheader=ERAI_
indataFtail=

In case of netCDF input data, there is a bash script which transforms the data to grib, to be used later by ungrib

Variable table to use in ungrib

#   Type of Vtable for ungrib as Vtable.[VtableType]
VtableType=ERA-interim.pl

Folder with the atmospheric forcing data (located at [forcingHOST]).

# For `nc' format
#   Folder which contents the atmospheric data to generate the initial state
iniatmosHOME = ./
#   Type of atmospheric data to generate the initial state
#     `ECMWFstd': ECMWF 'standard' way ERAI_[pl/sfc][YYYY][MM]_[var1]-[var2].grib
#     `ERAI-IPSL': ECMWF ERA-INTERIM stored in the common IPSL way (.../4xdaily/[AN\_PL/AN\_SF])
iniatmosTYPE = 'ECMWFstd'

Here on can change values on the template namelist.input. It will change the values of the provided parameters with a new value. If the given parameter is not in the template of the namelist.input it will be automatically added.

## Namelist changes
nlparameters = ra_sw_physics;4,ra_lw_physics;4,time_step;180

Name of WRF's executable (to be localized at [orHOME] folder from [codeHOST])

# Name of the exectuable
nameEXEC=wrf.exe

':' separated list of netCDF file names from WRF's output which do not need to be kept

# netCDF Files which will not be kept anywhere
NokeptfileNAMES=''

':' separated list of headers of netCDF file names from WRF's output which need to be kept

# Headers of netCDF files need to be kept
HkeptfileNAMES=wrfout_d:wrfxtrm_d:wrfpress_d:wrfcdx_d:wrfhfcdx_d

':' separated list of headers of restarts netCDF file names from WRF's output which need to be kept

# Headers of netCDF restart files need to be kept
HrstfileNAMES=wrfrst_d

Parallel configuration of the run.

# WRF parallel run configuration
## Number of nodes
Nnodes = 1
## Number of mpi procs
Nmpiprocs = 16
## Number of shared memory threads ('None' for no openMP threads)
Nopenthreads = None
## Memory size of shared memory threads
SIZEopenthreads = 200M
## Memory for PBS jobs
MEMjobs = 30gb

Generic definitions

## Generic
errormsg=ERROR -- error -- ERROR -- error
warnmsg=WARNING -- warning -- WARNING -- warning