Test variable grid spacing for C grid

[JFLemieux73]

No description provided.

[JFLemieux73]

I used the following setup:

./cice.setup --case Bvargrid -g gbox80 -m ppp6 -e intel -p 1x1 -s boxforcee,boxclosed,vargrid,buildincremental

I then changed things in the namelist:
npt=10 (days)
ndte = 1200
deltaminEVP = 2e-9
capping_method = 'sum'
kridge = 1
ktransport = 1
atm_data_type = 'uniform_east'
ocn_data_type = 'calm'
bgc_data_type = 'default'
fe_data_type = 'default'
ice_data_type = 'uniform'
ice_data_conc = 'p5'
ice_data_dist = 'uniform'

The runs are 10 days long. I look at the results after 10 days.

[JFLemieux73]

Here is hi after 10 days for B-evp:

for C-evp:

The difference (C-B):

Notice that the scales goes from -2 to 2 cm...kind of small differences.

[eclare108213]

They are small, and it makes sense that the differences would first show up on the right (east) side, not the others.
It's difficult to tell whether the trend is leaning toward more or less ice volume. There is a hint of thicker ice building out from the corners in the C-grid. Is the thermodynamics on? If so, then the trend might depend on what season the forcing is set for...

[JFLemieux73]

Experiments are repeated with ice_data_conc = 'p8'. Note that hbar=1 here instead of hbar=2 for ice_data_conc = 'p5'.

Here is C-B:

[JFLemieux73]

Hi Elizabeth,

Thermo is off. Can I have thermo "on" with these idealized tests?

[eclare108213]

You can always turn it on and see what happens :)
Looks like you're making good progress with the dynamics by itself.

[JFLemieux73]

I also want to verify the pressure fields. Here is for the B grid:

and the C grid:

[JFLemieux73]

Going back to ice_data_conc='p5' and now setting atm_data_type = 'uniform_northeast'

For the B grid:

and the C grid:

[JFLemieux73]

Looking at differences we have (C-B):

note that the scale goes from -5 to 5 cm.

[JFLemieux73]

Do we see a thickening with the C grid (north-east corner)? Is the thickening in gx3-gx1 experiments related to corners?

[JFLemieux73]

Now with ice_data_conc = 'p8'.

B grid:

C grid:

difference (C-B ):

Again we kind of see a thickening with the C grid. Do we get the same thing with a uniform grid?

[JFLemieux73]

Here is sigP for the B grid:

\

and for the C grid:

[JFLemieux73]

We see something in the northeast corner for both grids. But clearly there is something weird with the C grid corner.

[eclare108213]

There is something weird with the C grid corner, but is the B-grid corner right, sigP=0 there? Maybe that's just because of which value (ne?) is chosen to be written to history?
There are differences in all 4 corners in this test, which is interesting. Have you tried cranking up the subcycling even further, and/or comparing with VP on the B grid? It wouldn't be surprising for differences in residual elastic waves to be amplified near corners and also in areas with steep thickness or pressure gradients (like the nw and se corners in this test).

[JFLemieux73]

Thanks Elizabeth for your input. I think you are right for sigP=0 for the B-grid.

I am already using 1200 subcycles. I could try also to reduce the damping time scale for the elastic waves.

I will also compare against the VP solution and use the upwind for the advection (Dave's idea).

[JFLemieux73]

Before doing all these tests I have decided to run these for 30 days instead to make the problem more obvious (ice_data_conc = 'p8'). Here is hi (m) after 30 days for the B grid:

and for the C grid:

And the diff (C-B):

[JFLemieux73]

There is an asymmetry in the C grid hi field. We also see the thickening close to the corner when compared to the B grid.

[JFLemieux73]

I am taking a step back. I want to see if we get the asymmetry and the thickening with the constant grid. The only differences with the previous runs is that

scale_dxdy = .false.
dxscale = 1.d0
dyscale = 1.d0

Instead of:

scale_dxdy = .true.
dxscale = 1.02d0
dyscale = 1.01d0

[eclare108213]

scale_dxdy = .true.
dxscale = 1.02d0
dyscale = 1.01d0

Is the non-symmetry in dxscale and dyscale intentional? It looks like there is a slight nonsymmetry in the B-grid test, maybe coming from this. But see also #694 #316

So now there are two mysteries:

1. the solution with variable grid sizing is very different from a constant grid, for either C- or B-grid

Changing the scaling factor by 1 or 2% will change the domain size, which will affect the overall rate of convergence/divergence at the boundaries, but I wouldn't expect it to be big.

We've always suspected that the ice pressure formulation is too grid dependent. Try keeping it constant but allow everything else to use the spatial gradients that are calculated? Not sure if that will even run with advection on, but you could turn advection off and compare the resulting velocities and stresses.

2. C-grid results are more nonsymmetric with the variable grid than when using B-grid

Is this still true on the constant grid? If not, then at least we can narrow the search.

[JFLemieux73]

I made a mistake with the plotting and have edited the comment...stay tuned I will update the figures.

[JFLemieux73]

I am closing this issue and will start a new one with:

scale_dxdy = .true.
dxscale = 1.01d0
dyscale = 1.01d0

I also want to to get the VP solution and crank up the EVP convergence.