vertical velocity artifacts caused by j-grid decomposition

[sunnylandjim]

In an mpi implementation of ROMS 3.4 (32 processors; NtileI=4; NtileJ=8) for the southeast Indian Ocean, I have noticed artifacts in vertical velocity that appear to be associated with the grid decomposition. The model is set-up on a curvilinear horizontal grid (Lm=192;Mm=352) with 30 vertical levels (maximum depth 2000 m), and has been run for 4 years with 6 hourly output. The attached figure shows monthly average w (m/d) for a selected month at vertical level 15 as a typical example. The discontinuities in w appear to correspond with the j-grid decomposition (which is indicated on the figure by the position of the y tick marks). It looks like the w values are not being shared correctly between the j-direction partitions? Other fields (e.g. u,v) appear to be smooth. Is this a known problem?

Thanks for your help
Jim Greenwood

[arango]

Which vertical velocity? omega or true vertical velocity? It looks like noise in the j-partitions. If it is a parallel bug you will see it in all fields (momentum and tracers). I have never seen this type of structure in omega. Have you tried to run with the lastest version of the code?

What advection option are you using for momentum and tracers?

[sunnylandjim]

Thanks for taking the time to reply. The field shown is true vertical velocity. Omega was not included in the model output so I can't tell you what that looks like. The default advection scheme (4th order centered?) was used for momentum, and the 3rd order upstream scheme (TS_U3HADVECTION) was used for tracers. This run was completed (as part of a larger project) some time ago (hence the old version of the code), but the anomaly in w was not picked up until now; as I mentioned, the other fields look ok. I have not tried to re-run with the latest version. I will no doubt have to do that at some time. In the meantime, I thought it was worth reporting.

Jim Greenwood

[arango]

The true vertical velocity is a diagnostic quantity that it is not use in the ROMS numerical kernel. It is only computed for output. It is possible that there is a parallel bug in wvelocity.F. I will check and debug this routine when I get some time. I need to find an application that show this problem. Thank you for reporting this.

[jabent]

Hi, I'd like to follow-up on this topic if it's still being investigated.

I'm using this model configuraton (probably different model forcing) and the tiling issue appeared when I started using terms from the temperature equation (ocean_dia.nc). Below is a figures of the depth-averaged (upper 60 m), and time-integrated (over 60 days) terms from the temperature equation.

The sum of the advection terms (which largely cancel) lead to the lower, left hand panel, which shows the tiling. It looks like this issue is in the horizontal advection terms as well as the vertical advection term. If there's been any update, please let me know.

Cheers,

Jessica