[MITgcm-devel] Hack to increase mixing near bottom/surface

[Jean-Michel Campin]

Hi Dan and Martin,

I run few tests using a 2-D (y-z) shelfice set-up from Dan, to try to evaluate
how this pCellMix hack applies to pkg/shelfice (to avoid using SHELFICEboundaryLayer);
and here is a summary.

1) the set-up:
 I started from the set-up Dan gave me, without rotation, similar to the one used
 here: http://mitgcm.org/pipermail/mitgcm-devel/2015-April/006435.html
 but without updating the ice-shelf thickness (SHELFICEMassStepping=F)
 and with Linear Free-Surf.

 One limitation of this testing-set-up is that hFac below the ice-shelf is never
 smaller than 0.2, even when I set hFacMin to 0.05 (this has to do with
 input files SHELFICEtopoFile). This limit the range of the test/validation.
 Also the shape of the ice-shelf is not perfectly regular, as shown by the
 bottom-depth slope of the ice-shelf:
  http://mitgcm.org/~jmc/figs_iceShelf2d/ice_bottom_slope.jpg
 (blue curve = slope black curve = 10 x ice-bottom-hFac)
 so that, even without any numerical problem, we could expect to get some
 irregular feature (e.g., in melt-rate).

2) the current pCellMix code (using pCellMix_select=10) is not helping much,
 and the melt-rate is almost as noisy as without (could expect larger improvement
 where/if smaller hFac); this was tested with SHELFICEuseGammaFrict=F.
 I made some changes to pCellMix hack to increase even more viscosity and
 diffusivity by (1/hFac)^mixSurf, and I can get a smooth solution,
 as smooth as with SHELFICEboundaryLayer=T., using mixSurf=3 or 4.
 Will need to check how it performs with smaller hFac (smaller hFacMin) and
 with rotation.

3) With SHELFICEuseGammaFrict=T, the modified pCellMix hack does not address
 the problem of ustar variations next to a step (averaging to grid-cell center)
 as anticipated in Feb.:
> 3) might need to test different averaging (e.g. wet-point) of the velocity
>    to the grid-cell center (in shelfice_thermodynamics.F)
>    I don't know if there is a diagnostic for this (but could be useful)
 I added an alternative expression for uStar (just checked-in today), using
 wet-point averaging, and manage to get a smooth melt-rate when using mixSurf=3 or 4.,
 again as smooth as with SHELFICEboundaryLayer=T.
 However, the melt rate is significantly larger than with SHELFICEboundaryLayer=T.
 (with original uStar expression), and this comes from increased melting
 near the deepest part of the ice-shelf where the slope is large with many steps
 in ice-shelf k-bottom.
 Note that, with original uStar expression, the melt-rate was significantly lower
 with SHELFICEboundaryLayer=F (m34) than with SHELFICEboundaryLayer=T.

I made 3 other plots (all in http://mitgcm.org/~jmc/figs_iceShelf2d/) showing
the melt-rate fct of lat after 1.month, on left axis, from different experiments
(all using SHELFICEuseGammaFrict=T) and with ice-bottom hFac (in red) on right axis:
 m3b : with SHELFICEboundaryLayer=T and original uStar
 m34 : mixSurf=4 and original uStar
 m44 : mixSurf=4 and wet-point aver uStar
 m43 : mixSurf=3 and wet-point aver uStar
 m42 : mixSurf=2 and wet-point aver uStar
 m40 : no pCellMix + wet-point aver uStar

I wonder if this large sensitivity of melt-rate when using SHELFICEuseGammaFrict=T
 is common ? or is it due to no-rotation ? or maybe there is something wrong in
my wet-point averaging code (now cheked-in) ?

Cheers,
Jean-Michel

On Fri, Jan 16, 2015 at 02:05:08PM -0500, Jean-Michel Campin wrote:
> Hi Dan and Martin,
> 
> I made some simple test with a hack to increase mixing near above bottom and/or 
> below surface when the bottom/surface grid cell is thin (small hFac).
> 
> I put some files here: http://mitgcm.org/~jmc/pCellMix/
> with some documentation in file "pCellMix.notes" 
>  and some plots from my simple test.
> 
> the hack: it should move to the main code at some point, but
> 1) need more tests (and feedback)
> 2) it involves changing more S/R (> 7 but only 2 right now) since it 
>  needs to be pushed where vertical diffusion and viscosity are applied:
>  ( mom_u/v_implicit_r.F mom_u/v_rviscflux.F 
>    impldiff.F gad_diff_r.F gad_implicit_r.F )
>  and passing more arguments to these S/R (-> also changing the calling S/R)
> 
> Some remarks:
> 1) we should probably always use selectBotDragQuadr=1 or 2 instead of the original
>   discretisation (selectBotDragQuadr=0), see http://mitgcm.org/~jmc/pCellMix/fig_wt4.ps
> 2) accounting for partial cell in the interior (interViscAr_pCell=T, 
>   interDiffKr_pCell=T) is safe when using implicit visc/diff ; 
>   but also accounting for hFac in bottom friction (bottomVisc_pCell=T) when 
>   using no-slip-bottom=T can be unstable; and we don't have code for implicit 
>   bottom friction + it would not work with present solve_for_pressure code.
> 
> Regarding how this could be use with pkg/shelfice (to avoid using SHELFICEboundaryLayer):
> 1) in isomip test experiment, no_slip_shelfice=F and this is good, otherwise
>    the velocity below ice-shelf is strongly reduced when hFac is small (and it's
>    physical) and make the melt-rate very dependent on hFac.
> 2) using a type of quadratic drag: the equivalent of selectBotDragQuadr > 0 is not
>    coded in shelfice_u/v_drag.F but should not be hard (+ I added the new argument)
> 3) might need to test different averaging (e.g. wet-point) of the velocity 
>    to the grid-cell center (in shelfice_thermodynamics.F)
>    I don't know if there is a diagnostic for this (but could be useful)
> 4) would be interesting to try SHELFICEboundaryLayer=F and pCellMix_select=10 ;
>  
> Cheers,
> Jean-Michel
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel