Instructures for setting up Wave Ensemble Reforecast
- Note this is a work in progress and the repo will be moved to a NOAA-EMC location
#Clone repo git clone https://github.com/JessicaMeixner-NOAA/wave_ens_reforecast
From sorc/ directory:
sh checkout.sh
From sorc/directory: sh build_ww3prepost.sh
Requests of features from Ricardo:
- One cycle per day (starting at 03Z), with restart file for +24h
- 5(or 11) ensemble members
- New list of point output (parm/wave/bstations_GEFSv12WW3grid.txt)
- New list of output fields -- Grib: HS FP DP -- Netcdf: WND HS FP T01 T02 DIR DP SPR PHS PTP PDIR
- DT point output is 1 hour
- DT for gridded output is 3 hours
- We can keep grib2 format for output fields, but netcdf format for the point output files will be much easier to handle and to compare with observations;
- Suggestion: 7 simulations (1week) per job, to optimize Orion queue time.
- 4 streams: blocks of 5 years, starting at 2000, 2005, 2010, and 2015.
Jesssica's notes:
GEFS winds and ice: /work/noaa/marine/ricardo.campos/data/GEFSv12 Point output list: /work/noaa/marine/ricardo.campos/data/buoys/bstations_GEFSv12WW3grid.txt
- Create mod_def once and store it
- Prep of wind and ice will be in workflow
- 1 forecast will use 10 nodes, 40 tasks/node takes ~32 min for 16 days or ~1 hour 7 min for 35 days
- Restart dt will be 24 hours
- Requested to not use check-point output and instead use one file for point/gridded output
- Point output should be netcdf spec and table (Example templates in parm/wave dir)
- Gridded, wants both netcdf and grib with different lists of files
- Can clean-up binary run data as we go
- Python post processing, Ricardo will provide script, module and resources
Issue: Need to get hpc-stack netcdf issue resolved