Fix ensemble mode stochastic pattern seed values for tendency-based and SPP stochastic physics. #466
Conversation
… into feature/ensemble_seed
JeffBeck-NOAA
requested review from
BenjaminBlake-NOAA,
gsketefian,
JulieSchramm and
mkavulich
as code owners
… into feature/ensemble_seed
There was a problem hiding this comment.
I tested on hera (20200615, 36-h fcst; 10 members; GSD_SAR physics suite at 25km). Members 1-7,10 ran as expected. Seed values were generated appropriately and output looked good. Members 8 and 9 failed when running the forecast due to an issue with generating the unique seed values. The problem lies in: regional_workflow/ush/set_FV3nml_stoch_params.sh on lines 98-101. The error has to do with the leading zero with ensmem_num (bash treats numbers preceded with a zero as an octal and apparently 08 and 09 are out of range for octal). I modified the script to remove the leading zero in ensmem_num and was able to run mem08 and mem09 successfully. As a note, my config.sh: /scratch2/BMC/det/harrold/ufs-srw_seed_retest/expt_dirs/test_ens_seed/config.sh
… into feature/ensemble_seed
|
@michelleharrold, thank you for finding the octal bug! I have added bash syntax to convert the ensemble member numbers from octal to base 10 prior to computing the seed values. If you have time, could you please do a "git pull" and retest for the members which were failing for you? Thanks! |
There was a problem hiding this comment.
@JeffBeck-NOAA -- I tested the latest mods to account for the issue with octal. The namelist was generated appropriately, and everything ran as expected. Looks good to me!
Thanks, @michelleharrold! |
…chemes. (ufs-community#466) * Change the way the seed values are defined across cycles. * Remove references to stochastic seed script. * Update the generate script. * Source the stochastic seed script in the run_fcst ex-script. * Reduce default SPPT magnitude to 0.7. * Reduce the default SPPT magnitude to 0.7 in the config_defaults.sh file. * Update cycle directory path to work with both NCO and community modes. * Remove loop from seed creation, since the XML handles the ensemble members individually. * Reset sppt magnitude to 1.0 and turn on surface limiter. Tested on Hera. * Add bash syntax to remove leading zeros from member values when creating seeds.
* Adding configs for NSSL physics. * Add mem2 * Adding config files from AWS. * Fix ensemble mode stochastic pattern seed values for ad-hoc and SPP schemes. (#466) * Change the way the seed values are defined across cycles. * Remove references to stochastic seed script. * Update the generate script. * Source the stochastic seed script in the run_fcst ex-script. * Reduce default SPPT magnitude to 0.7. * Reduce the default SPPT magnitude to 0.7 in the config_defaults.sh file. * Update cycle directory path to work with both NCO and community modes. * Remove loop from seed creation, since the XML handles the ensemble members individually. * Reset sppt magnitude to 1.0 and turn on surface limiter. Tested on Hera. * Add bash syntax to remove leading zeros from member values when creating seeds. * Scripts need to differentiate from list of suites. * Set up namelists appropriately for each member * Suites are now housed in the model repo. * Update namelist parameters to match HWT experiment * Updating ensemble configuration * Need configurable SPPT param. * Updating for new member 2 config file. * ICS/LBCS tasks need to know which ens member. * Adding new physics suite option. * GEFS02 needs its own dir struct. * Update the workflow env version of Rocoto. * Removing temporary files added for comparison. Co-authored-by: JeffBeck-NOAA <55201531+JeffBeck-NOAA@users.noreply.github.com>
JeffBeck-NOAA
changed the title
JeffBeck-NOAA
changed the title
* Continue to refactor configure files.
* Add configure files for RRFS_A on CONUS 3km on wcoss2.
* The following updates are made for running system on WCOSS2
with 24-h ensemble cycles and ensemble forecast in certain cycles:
1) refactor the configure files: collect all machine and grid releated
parameters in one file. All configure files read this file for i
grid and machine related setups. Tune for CONUS 13km runs.
2). change file name of "rrfs_glb_berror.l127y194.f77" to i
"rrfs_glb_berror.l127y770.f77". Updated FIX_RRFS.
3). Added configure files for rrfs_a.
4). Add run enviromental parameter TPP_* for wcoss2, which should set with PPN_*
5). In GSI analysis, added check to run reflectivity DA only
when RRFSE ensembles are available.
6) update bufrsnd scripts and added fix_bufrsnd.
Co-authored-by: Shun Liu <Shun.Liu@noaa.gov>
DESCRIPTION OF CHANGES:
This PR fixes the stochastic pattern seed values for SPP, SPPT, SKEB, and SHUM when running in ensemble mode. Previously, the cycle date that was used to create the seed values was not being correctly updated with the CDATE of each cycle. Seed values are now set as part of the forecast task instead of being set during either workflow generation or the make_grid task, as was done before. This PR also reduces the default SPPT magnitude to 0.7 given recent testing showing instabilities with anything larger.
TESTS CONDUCTED:
Tested on Jet.
CONTRIBUTORS (optional):
@gsketefian and @chunhuazhou