Spatial oscillation in bottom stress for a steady flow

[mapeifeng]

Dear all,

I got a difficulty when simulating a steady flow in an open channel of 100km length, 10km width and 10m still water depth. "Lm == 50, Mm == 3, N == 20" are specified. Uniform condition is assumed in the north-south direction. The steady flow is driven by prescribed constant flow discharges at the western and eastern open boundaries, equivalent to specifying depth mean velocity of ubar = 0.5/(1+zeta/h) m/s at the boundaries. In the simulation, T and S were fixed and MY25 turbulence scheme is chosen. The quadratic bottom friction condition with RDRG2 = 0.1 (equivalent to a reasonable apparent roughness of 7cm) is used. Bottom stress is expected to be around 1 N/m^2 in the conditions.

Tests show that stable results can only be obtained when a very small time step of DT = 36sec (NDTFAST == 20) is used (DX = 2km), which is much smaller than the time step limit from CFL condition. With DT = 36sec, the results of a well established stable state show pronounced spatial oscillations in bottom stresses, near bottom velocities and depth mean velocity. The oscillations disappear when DT reduces to 18sec. If DT = 24sec, the oscillations occur only near the eastern boundary where bottom shear stresses are larger.

Another simulation with a particular bed profile (smaller water depth at the middle of the domain) yields oscillations only around the center where bottom stresses are larger. So the oscillations should not be caused by the boundary conditions.

A few simulations in 2D depth averaged mode show no such oscillations, even bottom stress is up to ~ 5 N/m^2.

I have to look at (relatively) large bottom stress cases due to working on sediment transport modeling in shallow waters. I will appreciate any comments and suggestions on eliminating such oscillations.

I have attached a few plots and necessary input and header files.


Peifeng

[mapeifeng]

Since I have been using POM, I ran the same test with POM and found that, for this particular case, POM can predict non-oscillating bottom stress at a larger time step of DT = 120sec (The bottom stress along the center line of the channel is shown below). This makes me a bit confused as, due to the more advanced algorithms, I expected ROMS to work fine at a larger time step than POM, which I did observe in some idealized baroclinic tests where water depth is 50m and therefore bottom friction is less important. So I am thinking of two possible reasons which may cause the oscillations: (1) Inappropriate specifications in my input or header files; (2) Imperfect numerical implementation associated with bottom stress. I hope to get any help.

[jcwarner]

did you use
#define UV_SMAGRINSKY
because that is spelled incorrectly so it did not do anything.

[mapeifeng]

I noted that after uploading the header file. So I ran a case with correct spelling: #define UV_SMAGORINSKY. The results are slightly different, but the oscillations remain. I also tried different advection schemes and got a little different but still oscillating results.

I also ran a few tests in a smaller domain of 20km long with same grid numbers and got improvements. The results show no oscillations for RDRG2 = 0.1 and DT = 20s, but similar oscillations show up when RDRG2 = 0.2 (equivalent to an apparent roughness of 11cm) DT = 20s are used.

[jcwarner]

i monkeyed around with this setup for awhile. here are a few things:
- when i changed to log_drag, that seemed to help.
- i changed theta_s and theta_b to both = 0 and the oscillations went away. so maybe it is a vertical advection issue.

[mapeifeng]

Thanks for the help.

I also tried a couple of tests with uniform vertical spacing, which does help improve the results. When using 20 uniform sigma layers, for the case of L=100km and RDRG2=0.1, oscillations do not show up until DT=60s.

I have looked at some cases with LOGDRAG. The results depend on the specified roughness zob as expected. When zob=1cm, the bottom stress is about 0.6N/m^2 and no oscillations take place. And oscillations show up when zob=2cm where bottom stress is over 1N/m^2, similar to the bottom stress for the QDRAG case with RDRG2=0.1. So essentially, this does not help.

I agree that the oscillations associated with large bottom stress may be related to numerical treatment of vertical advection or viscous terms, but it is obviously beyond my capability to check it out.

Now I have two ways to improve the results: using FINER horizontal grids and using COARSER vertical spacings. For example, a test, with L = 20km (DX = 400m), 20 uniform sigma layers, RDRG2=0.2 and DT=30s, shows no oscillations where the bottom stress is about 2N/m^2. Hope these can help avoid oscillations in most of my simulations.

-- Peifeng