Random inconsistencies on the temperature and salinity variables

[mmontano]

Hi, community!

As the subject suggests we recently noticed that our forecast showed random inconsistencies in the temperature variable. At first, we thought it was due to atmospheric forcing but looking further into it we noticed the "inconsistency" is present through the water column. It is also reflected in the salinity. By inconsistency, I am referring to a grid point where values (of salinity or temperature) drop. There are a few mysteries here.

• These "inconsistencies" appear after 30 minutes, and they completely disappear after 12 hours within a 7-day forecast
• They are not big enough to cause a model blow-up
• It is not exclusive to temperature as it also shows in the salinity
• It seems to propagate throughout the water column
• These spots appear randomly throughout the domain, although around the same regions (as seen below)
• There are no clear bathymetric features that might be to blame

Forecast for the 9th of July forward, SST. Three inconsistencies through the northern region of the domain.

Forecast for the 8th of July forward, SST. The colour map was changed to make the inconsistencies more noticeable. Five-ish inconsistencies through the northern region of the domain.

Even though this doesn't blow our run and doesn't seem to have a great impact on our output, we would like to get it right to be able to couple our run to an atmospheric model and to not propagate the error. Our guess is that it is an internal computation kind of issue, if anybody has encountered this issue before or have any input I would really appreciate it!

[jcwarner]

what kind of vertical mixing do you have on?
do you have any wave-ocn coupling?

[mmontano]

Hi @jcwarner

what kind of vertical mixing do you have on?
do you have any wave-ocn coupling?

- The vertical mixing for this run is resolved using “generic length scale” (GLS) turbulence model configured as a k-kl – equivalent to Mellor–Yamada 2.5 (Warner et al., 2005)
- This run in particular has no coupling

[jcwarner]

if you set the coefficients to use k-e or k-gen, does the spots still appear?

[mmontano]

Hi @jcwarner,
First a disclaimer, we are actually using k-e. An older configuration was using k-kl so I got mixed up there.
Anyway, I ran the tests and compared k-e, k-kl and k-gen. Thirty minutes into the run k-kl seem to not show random "inconsistencies", however an hour into the run the coastal regions were showing a lot of noise (as shown below, bottom panel). K-e, as in the examples above, only shows some random "inconsistencies" (top panel). Meanwhile, k-gen presents more "inconsistencies" throughout the region.

7-day forecast cycle starting on the 11th of July, SST time stamp 11th of July 01:24:00. The top panel shows the run with k-e coefficients. The bottom panel shows the run with k-kl coefficients.

7-day forecast cycle starting on the 11th of July, SST time stamp 11th of July 00:28:00 using k-gen coefficients

These tests suggest that playing around with the vertical mixing coefficients might be the answer to removing those spots. Therefore I have some additional questions, do you have any insight on where to start (other papers)? Is it okay to play around with the parameters, if so, is there a threshold of how much can these parameters change before affecting the physics of the domain?

Thank you very much for your insights so far!

[jcwarner]

ok do me a favor.
cd ROMS/Nonlinear
mv gls_corstep.F gls_corstep.F_orig
go here:
https://myroms.org/forum/viewtopic.php?p=24570#p24570
and download their alt_gls_corstep.F,
mv alt_gls_corstep.F gls_corstep.F
then recompile and see if that helps.
-j