search

travissluka / soca-tutorial

temporary testbed for a soca tutorial, before it gets absorbed into the correct JEDI repos later
Apache License 2.0
2 stars 2 forks source link

Questions related to tutorial: SOCA initialization #2

Closed gmao-cda closed 7 months ago

gmao-cda commented 8 months ago

Finish running the part init.

I had 2 problems (1 solved, 1 unsolved) for this part.

(Solved) 1. Insufficient stacksize when decreasing N < 10 for mpirun dirac test

Problem: When I changed the N from default 10 to 8 for the dirac test mpirun -n N ./soca_error_covariance_toolbox.x dirac.yaml_noSSH, I got the error 

FATAL from PE     1: MPP_DO_GLOBAL_FIELD user stack overflow: call mpp_domains_set_stack_size( 2160000) from all PEs.

....

Solution: Increasing the size of domains_stack_size under namelist $fms_nml in the file input.nml removes this error.

(Unsolved) 2. different vertical scales shown from reference figures

Problem:

  1. I can get similar hz scale as

  2. However, my vt scales in the southern ocean are far less spatially-varying than your reference figure. Based on the topography on your figure (especially based on the open channel for the islands in the southern Asia), I checked the vt scale at the model level 13. (colorbar range: 0-12, linear). Going deeper will make the band of larger vt scales narrower and move south, though no levels show similar detailed vt scales as your reference figure.

    Screenshot 2024-03-03 at 2 44 06 PM Screenshot 2024-03-03 at 3 21 33 PM
travissluka commented 8 months ago

for 1, thanks, I'll take note of that in the tutorial for 2 I think what you have is fine. The image I have on the tutorial is with the unclipped values, which range from 0-50. A correlation length of 50 levels is too much for the EXPLICIT_DIFFUSION to efficiently handle, hence why the values are clipped to 10. I'll make note of this in the documentation.