LAM not picking up global model land use fields in Nesting Suite

Unified Model
1

Hi,

I am attempting to replicate an old job, using the most recent version of the Nesting Suite (allowing the use of the most recent UM and JULES releases). My new job is u-ck718.
I had planned to point towards the set of ancillary files I’d generated for my old job, which does not include land use ancils (vegfrac, vegfunc, soil). As my simulations are over the Arctic Ocean, I am happy for the land use fields to be downscaled from the global run.

In the version of the Nesting Suite in which my old job was based there was a switch (rg01_rs01_vegsoil_anc) for defining whether or not to use land-use ancillaries in the LAMs. However, this switch does not seem to appear in the newest version of the NS.

So, I tried generating these ancillaries. However, it transpired that it was going to take an inordinate amount of time to generate these (vegfrac ancil generation timed-out after 4-hour wall time, and inspection of the job.out file suggests there was still a long way to go). I have tried various things - including adding rg01_rs01_vegsoil_anc=false to the rose-suite.conf file (rg01_rs01_vegsoil_anc is the switch that appeared in the older version of the Nesting Suite) - but these have not worked (the LAM reconfiguration fails as it’s looking for land use ancil files that don’t exist).

Do you have any suggestions on how I can force my LAM to downscale the land use fields from my global model?

Also, it would be useful to understand why generating the vegfrac ancil file is taking so long to run. A rose help entry on the previous version of the NS suggests it does indeed take a long time when your domain is composed predominantly of sea grid points. But this seems unintuitive to me, and I may be misinterpreting what was meant.

Thanks for your help.

Andy

2

Andy

I was looking at you suite output when it all changed – if the global dump contains land-use fields, the regional model should be configurable to simply interpolate from the global dump. Just remove the stash items from the list in Configure ancils and initialise … for the um model.

Grenville

3

Andy

ANTS is probably a better tool for creating ancillary files - have you tried that?

Grenville

4

Hi Grenville,

I have already tried running with the land use sections in Configure ancils and initialise … removed. My suite still failed trying to find a non-existent vegfrac ancil. I will try again, in case I made a mistake.

Andy

5

Hi Grenville,

I re-ran the job with the land use sections in Configure ancils and initialise … removed. The job is indeed still failing, but with a different error to that which I remembered. This error may not be related to the land use ancillaries. The error message is “Convergence failure in BiCGstab after 1 iterations: omg is too small”. I have re-run and re-created the error, but with extra diagnostics turned on, as advised in https://code.metoffice.gov.uk/trac/um/wiki/KnownUMFailurePoints

It’s not clear to me where the problem lies, so your help would be appreciated.

Andy

6

Andy

Hard to say what the problem is, could you try running the regional model with a shorter time step?

Grenville

7

Hi Grenville,

My intention was to set this job up to be identical to an older job, only updating to the most recent version of the Nesting Suite, since I require the most recent version - vn12.1 - of the UM.

I’ve had this error message in the past, usually after many timesteps or even hours into the run, where the problem has been a “grid-point storm”, preceded by unphysically large vertical velocities over steep orography. The solution has been to decrease the timestep and//or smooth the orography. But since the old job, with the same domain and spatial and temporal resolution, ran fine, and the error arises at only the second timestep, I was thinking it’s unlikely to be this same problem, or at least that the solution will be different.

Anyway, I have tried a few more things, including a small timestep of 15 seconds. The job still fails at the second timestep.

Thanks, Andy

8

Hi Grenville,

I started from scratch with a new job, and it is working. I don’t know what the problem was with the old job. The only change I’m aware of making is using a coarser-resolution global model.

Thanks, Andy

9

Andy

Thanks for letting us know - I’m glad it’s working.

Grenville