Getting (North/South halos)/Halos too small for advection unexpectedly in UKESM

Unified Model
1

Summary of error:
-Got the error ‘(North/South halos)/Halos too small for advection’ on a CRUN on the second cycle (201902; have run for 5 model years on first run)
-running on Monsoon
-UM version 11.1
-suite id is u-cd964
-log files are at /home/d00/mcoleman/cylc-run/u-cd964/log/job/20190201T0000Z/atmos_main/02

Hi there,

I got the above error, but after reading the suggestions at KnownUMFailurePoints – Unified Model (metoffice.gov.uk) I’m not really sure what to do with the max wind diagnostics or the extra diagnostics outputted. Having looked at the max wind diagnostics, it doesn’t look like the wind speeds started ‘spiralling’ into silly numbers towards the end - the highest wind speed in the CRUN was at timestep 132152 (which would be in the 20190101 cycle) while the model failed at timestep 133556, with wind speeds lower than they had been earlier in the CRUN. And I’m not really sure what to do with looking at physics increments as suggested on this page.

What also surprises me is this suite (u-cd964) is almost identical to another one I’ve done - u-br793 - which previously has run 35 years with no such problems. The only differences (unless I’ve made some mistake) are in the STASH requests and u-cd964 has a 1 month cycle and dumping while u-br793 is 3 months. I can’t think the STASH requests would affect this, but maybe the different cycles/dumping caused the difference in the simulation?

I’ve also tried running a CRUN from the same time with a similar suite (u-cm885; same STASH and cycling as u-cd964) but with different SO2 emissions, which managed to get to at least 202001 without encountering this error.

This makes me think the error maybe is just the model entering an unstable state by chance? But I’m not too sure.

So my questions are:
-is this likely to just be a chance error of the model entering an unstable state, and maybe I just need to change some innocuous setting to just avoid the model entering the unstable state?
-If necessary, how do I determine what caused the wind speed error from the extra diagnostics?
-or is there any otherwise recommended solution?

Thanks,
Max

2

Hi there,

Just as an update, on the advice of a colleague I have tried changing the number of convection calls (n_conv_calls) for just the failing cycle, on the basis the problem is likely a chance error, in order to avoid the model reaching the instability.

This seems to have fixed the error, with the model now succeeding in that cycle, and has continued on beyond that cycle fine.

Max

3

Glad you managed to solve this. Thanks for letting us know.

1 Like