#2896 closed help (answered)

settings for switched off reconfiguration

Reported by: emb66 Owned by: um_support
Component: Rose/Cylc Keywords: UKCA Release Job, CRUN
Cc: Platform: PUMA
UM Version: 11.0


Dear CMS,
I was running a version of the vn11.0 UKCA release job but with changes in chemistry scheme (following vn11.0_double_extended_StratChem), u-bh059. At some stage things went wrong while I was trying to change some cycling settings.

I’ve figured it would be best to copy the suite into a new one (u-bi005), use a dump file from bh059 from a couple of model months before things go wrong as start dump, switch off reconfiguration and run a quasi c-run but in a new rose suite.

My results don’t seem to be reproduceable – could I check please I have done it correctly?
I have:

  • set ASTART to the required dump (bh059a.da19960901_00)
  • set AINITIAL to bh059a.da19960901_00 too (although my understanding is this wouldn’t matter anyway as reconfiguration is off?)
  • switched reconfiguration task off; switched the build of reconfiguration executable off (but I’ve left the build of atmosphere executable on)
  • don’t override the date in start dump
  • left all entries in → um → Reconfiguration and Ancillary control → Configure ancils and initialise dump fields unchanged, apart from age of air (34150) and all my new UKCA tracers - for these I changed the source of initialisation to ‘allow reconfiguration to find default source’ (again, my understanding is these changes would be ignored anyway?)

Is this what I should have done?

Thank you,

Change History (6)

comment:1 Changed 14 months ago by grenville

Hi Ewa

This all sounds OK - but you'd need to double check. Run the model for as short a time as possible to compare configurations.

I'm not sure what you mean by "My results don’t seem to be reproduceable" - I can't find comparable output.


comment:2 Changed 14 months ago by emb66

Hi Grenville,

So things went wrong with u-bh059 ~Nov-Dec 1996, and u-bi005 started on 1 Sep 1996 (I archive dumps seasonally).

I'd expect the output for the first month of the new run (ARCHER: /nerc/n02/n02/emb66/archive/u-bi005/19960901T0000Z/bi005a.pm1996sep.pp) to be identical to the corresponding output from u-bh059 (/nerc/n02/n02/emb66/archive/u-bh059/19960901T0000Z/bh059a.pm1996sep.pp), and it's not.

Having said that, I've now realised that I have the BITCOMP_NRUN in rose set to false - is this what forces the results to be identical to the previous run? Does this only affect the random noise/internal variability part of the calculations (and so is mainly important if one wants to exactly reproduce the previous results), or if set to false in my case would lead to scientifically incorrect results?

Is there a good way to compare these configurations if the results are not exactly reproducable? (other than just looking at the plots)?

Thank you,

comment:3 Changed 14 months ago by grenville

Hi Ewa

The help for BITCOMP_NRUN explains briefly what this setting does - switching it off will result in different stochastic physics in addition to some differences in initialisation. Given the nature of the model, I don't think its scientific correctness depends on the value of BITCOMP_NRUN.

There are tools to compare models - I'm not sure how to advise on this topic; I suggest you contact other climate scientists for their experience.


comment:4 Changed 14 months ago by emb66

Hi Grenville,

That's fine, thank you.

Could I just double check:

  • the value of AINITIAL is only used by the reconfiguration task (and so it is ignored if reconfiguration is off)
  • for the entries in: " → um → Reconfiguration and Ancillary control → Configure ancils and initialise dump fields", the entries that deal with Ancillary Control should be left as is, and the entries that deal with initialisation of fields can either be left as is or changed to source=1 (default) as they are ignored anyway? (for switched off refconfiguration)

Thank you,

comment:5 Changed 14 months ago by grenville


In later versions of the model if the reconfiguration is off, the model will use AINITIAL as its start file - I'd double check with a test run. Leave the ancil settings as is.


comment:6 Changed 14 months ago by emb66

  • Resolution set to answered
  • Status changed from new to closed
Note: See TracTickets for help on using tickets.