rectangles delimited by anomalous values

[jmralves]

Hello

I am having difficulties with a grid-nesting configuration.
In the past, I successfully used ROMS in a 2-way configuration.
But now, shortly, after starting the run I get several rectangles delimited by anomalous values in the simulated variables, as you can see in the figure, and as a consequence, the simulation blows up.

I have made tests with 24 and with 48 CPUs. The rectangles appear in both, with more and smaller rectangles in the 48 case.
Could this issue be due to a cluster limitation, or am I missing something?

Thanks, for any hint.

[arango]

Hmm, I haven't seen this kind of problem in ROMS in a very long time. The squares clearly indicate a parallel partition bug in the code with the exchanges between tiles. Are you using distributed memory (MPI) or shared memory (OpenMP)? What version of the code are you using? And more importantly, did you modify any routine in the ROMS numerical kernel?

[jmralves]

I am using a ROMS version (ROMS 3.8 svn 1013) that is in COAWST v3.6, with a distributed memory (MPI).
The model was compiled with impi and I didn’t modify any routine in ROMS.
Although I am using ROMS that is in COAWST, here I am only using the ROMS model with atmospheric forcing from ERA5.

[jcwarner]

i have seen this problem if the ocean.in does not have the updated section to select the tracer advection schemes. Check to make sure your input file is current.

[jmralves]

Ups! When I changed from a one domain configuration to a nesting configuration, I forgot to update the tracer advection scheme in ocean.in

Thanks, John