Growing oscillations when using TS_MPDATA

[jwenegrat]

Hi All,

I'm running an idealized channel simulation (periodic in the y-direction), with a linear bottom slope in the x-direction. Domain is 60 km x 30 km x 1600 m (resolution 300 m x 300 m x 2-30 m). The simulation begins with uniform flow in the positive y-direction, and a uniform stratification. I'm looking at the Ekman spin-down and dynamics of the bottom boundary layer.

The simulation runs well, with processes evolving as expected, when using the default tracer advection scheme. However, when I switch to the TS_MPDATA scheme (which would be nice for studying some inert tracers) there begins to be large amplitude, low-vertical mode, oscillations (see attachments for plots of vertical velocity). These grow in time (for this simulation they become large after about 10 days of simulation time) and eventually crash the run by violating the vertical CFL condition. Without TS_MPDATA defined these waves don't appear in the simulation.

I've tried/confirmed a number of things:
- All slope and stability parameters are appear reasonable (rx0, rx1, barotropic Courant, CFL)
- Cranking up horizontal diffusivity delays the onset of the growing oscillation, and moves it to larger horizontal wavelength, but does not get rid of the growth.
- I've tried using TS_MPDATA_LIMIT, same behavior
- Played with timestepping to make the barotropic Courant number << 1
- Tried adjusting the vertical stretching, including switching to uniform grid spacing
- Tried altering the domain shape such that the topographic slope is confined to the middle of the domain, with flat regions near the walls.

I'd be grateful if there are any suggestions on other things to look at. Happy to provide more information on the configuration, etc. if useful.

Thank you very much!
Jacob Wenegrat

[jcwarner]

i cant open the ps file. how about a png?
how many grid cells is the domain. mpdata has a big footprint. needs at least 5 cells.
-j

[jwenegrat]

Sorry about the .ps files, attached here as jpegs. These are just quick plots from ncview, but it gives the general picture.

In terms of domain size it is 200 x 100 x 60 (NX, NY, NZ). The uniform vertical grid case I tried was NZ=250.

Thanks,
Jacob

[jcwarner]

are the files for this mostly ana*?
can you send them to me? i could try it here quick.
-j

[jwenegrat]

Sorry about the delay, and thanks for your offer of help. Before you spend any of your time, I think there's some more I can look at on my end, but it's good to know that this behavior is at least unexpected. My current working hypothesis is that the blow-up is a result of a growing instability that arises from phase-locking of baroclinic instabilities in the bottom boundary layer and gravity waves in the interior. My understanding is that the default advection scheme is dispersive, which perhaps is sufficient to stop the phase-locking and instability growth.

Regardless, I will update when I am able to return to this and make some progress.

Best,
Jacob