Repeated failure of the rAM3 ancil_cap_smcsnow task for very small domains

Atmosphere Regional Nesting Suite
1

Hello all.

So I’ve been trying to build a small regional test domain for rCM3 development. This has proven problematic, because :

  1. The regional ancillary suite will not work unless you provide some land points
  2. The repeated failed of the rAM3 ancillary task : <Region>_<Resolution>_ancil_cap_smcsnow.

Based on the the job’s outputs, the task takes the coarse resolution soil moisture content climatology file (which is at 1 x 1 degree), compute a snow climatology, and then regrid the outputs to the target file qrclim.smow.

For small test domains over Tasmania, the snow task would sometimes fail, and sometimes succeed depending on the number of points and domain size.

So I decided to make a minimum working example based on the default domain for rAM3.

I checked out u-bu503/nci_access_ram3 and made the following changes to rose-suite.conf:

rg01_centre=-28.49,153.38
rg01_nreslns=2
rg01_rs01_npts=54,64
rg01_rs02_npts=40,40

The job failed. Stdout:

 SPIRAL SEARCH RADIUS SET TO                      5
 RESTRICTED IN ROWS TO                     20
 Reading target land sea mask                      0
 Attempting to open file 
 /home/548/pag548/cylc-run/u-bu503/share/data/ancils/Lismore/d1100/qrparm.mask

(and then the job ends).

Stderr:

forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image              PC                Routine            Line        Source             

Stack trace terminated abnormally.
[FAIL] rose mpi-launch -v central_ancillary.exe # return-code=174

Unfortunately the jobs uses the precompiled binary

/g/data/access/projects/access/apps/cap/9.2/bin/central_ancillary.exe

And it provides no useful diagnostic errors.

FWIW I got the job to succeed by increasing the domains slightly:

rg01_rs01_npts=64,74
rg01_rs02_npts=50,50

If anyone is interested in this, I can commit a non-working version to the roses repo for you to check out.

My hunch is that if your domain is too small, and it doesn’t completely bracket any of the 1 degree soil moisture data, the job fails.

2

CAP source code should be at trunk in main – ANCIL. I suspect you’re right that it couldn’t find a matching point in the source dataset. You could maybe try increasing the spiral search distance?

1 Like