I keep getting the EG_BICGSTAB (NaNs in error term in BiCGstab after 1 iterations) or the North/South halos too small for advection error when attempting to prescribe PFT fraction (stash 216) in a Archer2 run (u-cp508). I have run with Diagnostic Print Status but can’t see any obvious solution. The suite runs for ~10 mins before crashing.
u-cp508 is coupled and I am overwriting the land use generated by Triffid to investigate a particular LU scenario. I know this general approach works as I have had coupled Archer2 suites (with Triffid on) (e.g. u-cp448) succeed when prescribing LULC derived from UKESM CMIP6 model output (e.g. u-be690_m01s00i216_2015-2100_monthly_timeseries_land_cover_frac.anc.nc).
However, in this case I am trying to prescribe LULC from a scenario I have developed from an external source (CESM). I did the necessary PFT conversions, created a netCDF file and used xancil to make the ancil file (/work/n02/n02/jweber/Ancils_CESM_Land/Timeseries/maxforest_216_timeseries_OUT.anc). I followed Jeff’s advice in Using xancil and xconv to create timeslice ancillary files for this. The maxforest_216_timeseries_OUT.anc fails on Archer2 but succeeds when run in coupled suites on Monsoon (e.g. u-cn793). I have looked at the ancil in xconv and it looks reasonable.
To test my processing code, I extracted the raw data from the UKESM model derived stash (u-be690_m01s00i216_2015-2100_monthly_timeseries_land_cover_frac.anc.nc) and ran this through my processing script - u-cp508 ran fine with the resulting ancil. I’ve also done further tests where I duplicated the UKESM data and replaced one PFT field with the field corresponding field from my new data - this also caused the model to crash.
This is quite confusing as the new ancil file runs on Monsoon and makes me think I may need to adjust my Archer2 suite.
Would you have any suggestions?
Many thanks for your help,