dpath2o / AFIM

Australian Fast Ice Modelling
1 stars 0 forks source link

ERA5 forcing for CICE6 standalone #6

Closed dpath2o closed 1 month ago

dpath2o commented 10 months ago

Edited src/CICE/my_editted_files/ice_forcing.F90 to take hourly ERA5 data.

Observed that runs were failing during the dynamic time step as documented here: logs/CICE/afim_era5/cice.runlog.231024-152839. In this file I note that the "Tair" minimum values are 0 (zero), which is not right as the value is Kelvin. Therefore, using src/python/cice_prep.ipynb began investigating why this might be the case. I determined that for some reason when using xesmf "bilinear" to regrid ERA5 data the minimum would goto 0. After a bit further investigation, I found that "nearest_s2d" provided a regridding that perseved the original minimum/maximums as well as slightly improved upon regridding speed/efficiency.

Used src/python/regrid_era5_for_cice6.py (via PBS job src/sh/regrid_era5_for_cice6.sh) to regrid ERA5 files. However, I after several attempts to regrid yearly concatenated ERA5 dataset, whereby the error I would get was either out of memory or used up the walltime. I have subsequently opened an NCI help request (see NCI help request HELP-191457) to assist in getting the writing of these files.

In the meantime I was able to do one month of ERA5 data (this still took roughly 20 CPU hours or 13 real-time hours), and the file is only 30GB in size -- something is really wrong with the way I'm writing this out, and I need help.

All that being said, once I did get a new regridded ERA5 forcing file I was able to succesfully run CICE6 being forced by ERA5 as documented here: logs/CICE/afim_era5/cice.runlog.231025-165520 . Of note, I only ran this as a test case for one month because that is all I wanted to initially do, but also, at the moment, I can't seem to write out a larger re-gridded ERA5 file, which is really frustrating.

Versions of ERA5 regridded files can be found in /home/581/da1339/cice-dirs/input/AFIM/forcing/0p25/ERA5/24XDAILY

dpath2o commented 10 months ago

Re-rote src/sh/regrid_era5_for_cice6.sh to use NCO calls and OpenMPI instead of using Python. I have broken the re-gridding into single months and single variables. Then concat these monthly files into one yearly file, again, using NCO. Job has been set to run and is currently in the queue.

dpath2o commented 1 month ago

see http://131.217.175.138/~ec2-user/research/cice6_standalone.html for aditional information