[MITgcm-support] Temperature Advection Scheme : Internal Gravity Wave Simulation

[Jody Klymak]

On 26 Jun 2017, at 10:22, Dimitris Menemenlis wrote:

> Maybe try using Crank Nickelson time stepping with >=5 min time step.
> Here is copy of old email from Ayan Chaudhuri on topic:
>
> When we ran our tidal forcing experiments with the default 
> timestepping scheme, we noticed weaker amplitudes mostly because the 
> fast gravity waves were getting  damped (even with timesteps of 2 
> min). Crank-Nickelson overcomes that short period damping which affect 
> the tidal solution . The 5 min recommendation is time-stepping scheme 
> dependent (not model resolution dependent) as Crank-Nickelson does not 
> work for large time steps. This time step limitation makes it more 
> expensive to run and hence is not default. I don't think there are any 
> other considerable side effects.

5 minutes!  My 100-m dx simulations are run at 10 seconds.

To directly answer Hanut’s questions, we’ve used advection scheme 2, 
33, and 77.  The flux-limiting schemes change how much energy you can 
directly diagnose as turbulent because they are quite numerically 
diffusive, but the actual energy lost in the simulation is not any 
different than the second-order scheme, and they tend to be more stable, 
and hence more convenient for many simulations.  I usually use 77.  The 
idea here is that, yes, the diffusion and dissipation is numeric, but it 
is working to smooth out shocks and strong shears which is exactly what 
any other diffusive scheme will do.  This implies that you are largely 
resolving the outer scale of the turbulence in your simulation, which in 
our case is usually breaking waves or strong shear layers.

Cheers,   Jody



>
> And probably best to use rStar if your configuration allows it.
> Here is some old notes from Jean-Michel on the topic:
>
> Regarding Crank-Nickolson (as in 
> verification/adjustment.cs-32x32x1/input/data:
> implicSurfPress=0.5,
> implicDiv2DFlow=0.5,
> these 2 parameters are just 1/2. You can try with 0.6 instead, but 
> it's
> no longer 2nd order (and will add some dissipation).
> And there is typo in your data file since implicSurfPress is repeated 
> 2 times
> but implicDiv2DFlow is missing.
>
> Regarding Non-Lin Free-Surf: What is the minimum depth in this domain 
> ?
> If it's not too small (e.g., at least 20.m I would say), I would 
> advise
> to start with rStar (this is sometime more stable than pure z-coord),
> especially if you add tides, rStar will reduce vertical velocity
> (and related spurious vertical mixing) which is associated to 
> d.Eta/dt.
> select_rStar=2,
> nonlinFreeSurf=4,
>
>> On Jun 26, 2017, at 10:00 AM, Hanut Vemulapalli <hanut2 at gmail.com> 
>> wrote:
>>
>> Hi all,
>>
>> I am trying to simulate gravity waves produced by an isolated 
>> gaussian topography in a uniformly stratified ocean. I have a 2D 
>> simulation with unit depth in the y-coordinate. I am applying the 
>> tidal forcing by prescribing the U-velocity at the boundaries (@Jody: 
>> Thank you for your suggestions) . I am able to simulate gravity 
>> waves, but could anyone please advice as to which temperature 
>> advection scheme would be best suited for my case and to what extent 
>> the scheme used can effect the wave field produced.
>> I have tried using schemes 2,3,33 and 77 but there seems to be no 
>> difference in the wave field produced and the gravity wave beam is 
>> not sharp/poorly resolved.
>> Thanks in advance.
>>
>> Best regards,
>> Hanut V


> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support