[MITgcm-support] seaice anomalous advection in doubly periodic domain

Mon Nov 2 10:22:22 EST 2015

Hi Jean,

the LSR solver just has this issue that you stumbled over. In each non-linear iteration it solves a linearized set of equations by iteratively solving many tridiagonal systems locally on each tile, and these tridiagonal solvers require boundary values from the neighboring tiles (of the previous iteration step). If convergence (of the linearized system) is not reached (because the required tolerance is low, or there are not enough iterations allowed) the edge effects remain. You should be able to get rid of these stripes by decreasing LSR_ERROR (usally 1e-6 is enough, but that makes the model much slower). Alternatively you can improve the situation by increaseing the number of non-linear iterations ( NPSEUDOTIMESTEPS>2, e.g. 10 ), or you could increase the overlap from zero:
SEAICE_OLx = 1
SEAICE_OLy = 1
(requires OLx=3, OLy=3, because the the LSR solver requires 2 overlap points, but with advection 33 you should have that anyway, basically SEAICE_OLx <= OLx-2)
In my experience, you can reduce these edge effects sufficiently well, so that they do not (visibly) affect you simulation.

Alternatively you can try a different solver, e.g. the JFNK solver (unfortunately quite expensive) and the EVP solver (in its pure form EVP is likely to produce noise, so use the EVP* version, see recent documentation: <http://mitgcm.org/public/r2_manual/latest/online_documents/node254.html#SECTION00762460000000000000>)

I think your first question is unrelated and purely physical: in a doubly periodic domain it’s probably easier to resist deformation and to move the infinite ice cover according to some mean wind forcing (check if your wind forcing has a non-zero mean). You can test that by introducing walls or islands somewhere in the domain.

Martin

P.S. personally I would use SEAICE_EPS = 2e-9 or smaller (recent code has SEAICE_deltaMin as a replacement for this parameter), to give a little stronger ice.

P.P.S. I hope that Ryan forgives me for contradicting, but even though the VP rheology was originally meant for large scales, it appears to work very well at very high resolution, i.e. the simulated fields show very realistic deformation statistics. There appears to be some controversy in the community, and this is not yet settled. 

> On 29 Oct 2015, at 17:58, Ryan Abernathey <ryan.abernathey at gmail.com> wrote:
> 
> Hi Jean,
> 
> I think I remember Georgy Manucharyan encountered similar problems with sea ice on tile boundaries.
> 
> I don't know too much about the details of the sea ice model. But I do know that the sea ice viscous plastic rheology applies to the large-scale sea ice field (>100 km scales) is is very likely completely inaccurate for the 1km scales you are trying to model. This is a serious limitation for high-resolution ice-ocean modeling, and there is no obvious way around it until a completely new type of floe-scale sea ice model can be developed.
> 
> That being said, there are lots of people on this list doing such modeling (e.g. Dimitris). Maybe you can get some answers from them.
> 
> -Ryan Abernathey
> 
> 
> On Thu, Oct 29, 2015 at 11:34 AM, Jean Mensa <jean.mensa at yale.edu> wrote:
> Hello list,
> I am trying to setup an idealized ocean simulation with sea ice using the SEAICE package. The domain is a doubly periodic box (400km by 200km) which I force with a semi-idealized ECMWF forcing. Forcing is periodic in time (and space) and seems to generate the expected seasonal cycle of ice thickness and concentration. Nevertheless there seems to be something wrong with the ice advection.
> 
> 1 - ice velocities follow wind forcing for a while, and then, as ice starts growing (not sure if that is a coincidence), they stop being affected by wind forcing and transition to steady state as if purely driven by ice inertia. Ice moves at a reasonable speed, it just goes to steady state. I am not sure whether this is a numerical/physical problem or an consequence of the doubly periodic setup.
> 
> 2 - ice growth shows anomalous growth at the boundaries of the partitioning tiles. This results in a distinguishable thickness anomaly. Currently I use the same advection scheme than for tracers, a 3-rd order flux-limiter (33), and the problem persists when switching to a smaller stencil.
> 
> I am not sure whether the two problems are linked or not. The second one looks like a numerical issue but the first one could be due to my particular setup.
> Any idea on what's going on?
> Thanks!
> 
> It follows my data and data.seaice files,
> 
> ############# DATA
> 
> # Model parameters
> # Continuous equation parameters
>  &PARM01
>  no_slip_bottom=.FALSE.,
>  eosType='JMD95P',
>  diffKhT=0,
>  diffKzT=0,
>  rotationPeriod=86400.,
>  beta=0,
>  f0=1.4e-4,
>  gravity=9.81,
>  rigidLid=.FALSE.,
>  implicitFreeSurface=.TRUE.,
>  implicitViscosity = .TRUE.,
>  implicitDiffusion = .TRUE.,
>  tempAdvScheme=33,
>  saltAdvScheme=33,
>  viscC2Leith=1.0,
>  staggerTimeStep=.TRUE.,
>  nonHydrostatic=.FALSE.,
>  readBinaryPrec=64,
>  &
> # Elliptic solver parameters
>  &PARM02
>  cg2dMaxIters=1000,
>  cg2dTargetResidual=1.E-7,
>  &
> # Time stepping parameters
>  &PARM03
>  endTime=189216000,
>  deltaTmom=240.0,
>  deltaTtracer=240.0,
>  abEps=0.1,
>  pChkptFreq=8640000.0,
>  chkptFreq=1728000.0,
>  dumpFreq=86400.,
>  monitorFreq=864000.,
>  forcing_In_AB = .FALSE.,
>  pickupStrictlyMatch = .TRUE.,
>  &
> 
> # Gridding parameters
>  &PARM04
>  usingCartesianGrid=.TRUE.,
>  usingSphericalPolarGrid=.FALSE.,
>  delX=400*1000,
>  delY=200*1000,
>  delZ=2, 2, 2, 2, 2, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 10, 10, 10, 10, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 30, 50, 50, 50, 50, 50,
>  &
> 
> # Input datasets
>  &PARM05
>  bathyFile='topo.bin',
>  hydrogThetaFile='theta_100_DP.bin',
>  checkIniTemp=.FALSE.,
>  hydrogSaltFile='salinity_100_DP.bin',
>  uVelInitFile='u_100_DP.bin',
>  vVelInitFile='v_100_DP.bin',
>  &
> 
> 
> ############# DATA.SEAICE
> 
>  &SEAICE_PARM01
>   SEAICEwriteState   = .TRUE.,
>   SEAICEuseFlooding  = .TRUE.,
>   SEAICE_EPS         = 1.E-8,
>   SEAICE_area_floor  = 1.E-5,
>   SEAICE_area_reg    = 0.15,
>   SEAICE_hice_reg    = 0.10,
>   IMAX_TICE = 6,
>   SEAICE_availHeatFrac = 0.8,
>   SEAICEuseDYNAMICS  = .TRUE.,
>   HeffFile = 'ice_thick_S.bin',
>   AreaFile = 'ice_fract_S.bin',
>   SEAICEadvScheme    = 33,
>   useHB87StressCoupling=.TRUE.,
>  &
> 
>  &SEAICE_PARM03
>  &
> 
> 
> 
> 
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
> 
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support