vertical mixing & noise in a channel flow

[jabent]

Hi, I've been working on a zonal channel flow problem on a beta-plane. An initial surface mixed layer is specified to about 200 m deep, with a surface temperature gradient. A small perturbation is input into the zonal flow.

I've been having problems with 2-dx grid noise and I haven't been able to get rid of it, despite trying a range of Laplacian or Biharmonic mixing coefficients and grid resolution. From a previous forum post, I thought that this grid noise might be related to the vertical mixing scheme. Noise seems to be less in GLS than KPP but GLS tends to be restratifying. I've attached several figures of the zonal flow and the T field at the surface, as well as a y-z slice of T after 5 years.

I'm showing cases where there's no initial perturbation and 1) vertical mixing coefficients that have a tanh-profile ranging from 0.001 m^2/s at the surface to 10^-6 m^2/s below about 200 m,

and when there's an initial perturbation with

2) the same vertical mixing coefficients, 3) MY2.5 with N2S2_HORAVG, KANTHA_CLAYSON, 4)GLS with k-epsilon and N2S2_HORAVG, KANTHA_CLAYSON, 5) LMD with options CONVEC,RIMIX,SHAPIRO,SKPP,NONLOCAL.

For all cases, I'm using the same ocean_ini.nc file that has this input noise, Lm = 128, Mm = 128, N = 30

NtileI = 8, NtileJ = 4, DT = 1000 s

the channel is 1560 m deep, thetas = 5.0,thetab = 0, 36 deg x 36 deg, ie dx = dy = .28 deg resolution, beta = 2 x 10^-11 m^-1s^-1.

There's a nonzero temperature flux at the surface with a T*(y) profile that's equal to the initial temperature of the surface mixed layer.

tnu4 = visc4 = 2.0e11 m^4/s and tnu2 = visc2 = 800 m^2/s. I've done cases where I've increased visc2 to 2300 m^2/s helps the amplitude of the grid noise by about 1-2 cm/s but it's still prominent.

My .h-file has:

#define UV_ADV
#define UV_COR
#define UV_VIS2
#define UV_VIS4
#define TS_DIF2
#define TS_DIF4
#define UV_LDRAG

#define MIX_GEO_UV
#define MIX_GEO_TS

#define CURVGRID
#define SPHERICAL

#define WJ_GRADP
#define TS_MPDATA
#define SOLVE3D
#define SPLINES

Should I be trying a different advection scheme to eliminate this grid noise?

Any suggestions are greatly appreciated,

Jessica

[jcwarner]

try
#undef TS_MPDATA
this will default to the second order scheme, just to see if the noise goes away.
-j

[jabent]

With MPDATA turned off, the noise is worse. I can include a figure if you'd like. Any other possible advection schemes that might help?

Jessica

[jcwarner]

i guess i am not clear what exactly the issue is. So you have a simulation with constant vertical mixing and it does not produce the noise, but when you try to use a turb closure you do get the noise? I am not clear what is meant by an initial perturbation - do you get noise when you have an initial perturbation and use a const vert mix?

[jabent]

I'm trying to eliminate the 2-dx grid noise in the temperature and flow field. In order to examine the baroclinic instability problem, I am currently inputting an initial perturbation to the u-field in the ocean_ini.nc file. The initial perturbation is randomly generated via matlab and within [-0.01, +0.01] cm/s. I've also tried inputting a 4-dx grid wavelength perturbation in the u- or T-fields but the same 2-dx pattern emerges. In all cases, this noise in x appears after about 1-2 years.

In one case with constant vertical mixing, I didn't include the perturbation and no grid-noise appears. In another case, I put the perturbation in and grid-noise appears. Then, I went and tried different vertical mixing schemes to try to eliminate the noise because I thought it might be associated with unstable vertical density gradients that occured with constant vertical mixing. With a turbulent closure and the initial perturbation, I am still finding the 2-dx grid noise.

I am puzzled as to why this grid noise is so prominent in such a simple configuration. I've also tried grid spacing with dx of 0.07 deg but it's still there.

I'd be interested to hear when and where other configurations have given rise to this problem. Perhaps it'd provide some clues as to why it's happening here.

Jessica

[jcwarner]

"For all cases, I'm using the same ocean_ini.nc file that has this input noise, Lm = 128, Mm = 128, N = 30
NtileI = 8, NtileJ = 4, DT = 1000 s
"
Suggest you try a different tiling.
Try NtileI=1 NtileJ=8 and see if you get same results as the 8x4.
The NtileI=1 is the important one here.
I remember there was an issue with perturbations with MPI, but that occurred when the code generated the random seed. in your case, it seems that matlab is generating the seed for the ini file, so this may not be an issue.

-j

[kate]

I believe the vertical mixing schemes work pointwise, so that any 2dx junk will not be smoothed by it. You need to investigate horizontal mixing schemes.

[hetland]

Your noise looks like Rossby waves (note the pattern that the noise is least at the equator, and seems to be propagating faster to the south). I suspect that you have some strangeness at your open boundaries that is causing this noise.

What are your OBCs?

[jabent]

Hi, I appreciate the help. I really want to figure this out.

I was wondering if tiling was an issue. I tried NtileI = 1, NtileJ = 8 with the same ocean_ini.nc file and the model output appears to be the same.

The 2dx-noise appears when a larger scale mode structure develops after about 1.5 years. When I turn up the horizontal mixing, I can suppress the noise and this structure too but I'd like to see the instability develop. Yes, the noise does appear to arise first in the southern part of the domain. I have NS walls and EW periodic.

I tried turning off the ana_sst so that the surface heat flux is zero but that didn't help.

The last couple days, I've been going to finer x-resolution, dx = .14 deg, dy = .28 deg. and ramping up the horizontal mixing with the hope of damping this grid noise without damping the larger scale structure. Even VISC4 = 10^13 m^4/s, which seems quite large, and TNU4 = 10^10 m^4/s still is unable to remove it.

For one of the runs with LMD and convection on in the y-z plane at an x-grid point, there are alternating secondary cells (v,w) within the surface mixed layer. At one x-grid point over, these secondary cells have the same magnitude and similar structure but rotate in the opposite direction. From the surface, it appears as 2dx grid-noise in the surface v-field. I'm just not sure why these circulations would alternate direction every x-grid point. I'm not sure if it's related to the cells seen in this post: viewtopic.php?f=14&t=917.

I'm going to try Smagorinsky as well and see if that helps.

[jabent]

This 2-dx grid noise started turning up in one of my other set-ups, similar configuration but with a closed basin. The noise did not arise when I switched from spherical to cartesian coordinates. I need to test whether making this switch will fix the noise in this channel set-up, but I suspect it will. With cartesian coordinates, I'm now running the closed basin set-up with much smaller horizontal mixing coefficients than when I was running it in spherical coordinates and the 2-dx noise is still not appearing.

I'm not sure if there's a bug or if there's some aspect of my set-up in spherical coordinates that is triggering this problem. So, for now, I'm going to stick with cartesian coordinates!

[LiuZHQ]

I follow the question of Hetland, "what is your OBC?". EW_PERIODIC?
How about also try f plane, instead of beta plane?

[jabent]

Yes, I was using EW_PERIODIC. I could try the f-plane but I've decided to just use cartesian coordinates since it eliminated the noise that was present with spherical coordinates.

Jessica