Simple method for calculating Numerically Induced Mixing

General scientific issues regarding ROMS

Moderators: arango, robertson

Post Reply
Message
Author
rsignell
Posts: 123
Joined: Fri Apr 25, 2003 9:22 pm
Location: USGS

Simple method for calculating Numerically Induced Mixing

#1 Post by rsignell » Tue Jun 15, 2010 7:26 pm

Burchard & Rennau (Ocean Modeling 20 (2008) 293-311) have a simple way to calculate physical and numerical mixing of tracers in ocean models. The physical mixing is simply the rate of change in the turbulent mean tracer variance and the numerical mixing is simply the rate of change of the difference between the advected square of the tracer and the square of the advected tracer. All one has to do is add one more state variable (the square of a tracer) and a few easy calculations.

Rob Hetland and I took a stab at hacking this in a while ago, got close, but didn't finish it. So I'm posting a message here to see if by chance anyone has already done this.

Thanks,
Rich

User avatar
gouillon
Posts: 9
Joined: Tue Sep 12, 2006 2:45 pm
Location: SHOM - Toulouse
Contact:

Re: Simple method for calculating Numerically Induced Mixing

#2 Post by gouillon » Thu Jun 17, 2010 4:10 pm

Hi Rich,

I have just finished my PhD and my dissertation focus was about the numerical mixing. Specifically, I look at how different numerical choices (e.g., grid horizontal and vertical resolution, advection scheme) impact on the magnitude of the spurious mixing generated by internal wave propagation in fixed coordinate ocean model. I did a multi-model comparison study using HYCOM (fully z- or fully sigma), ROMS and the MITgcm for very idealized internal wave scenarios (a lock exchange and a barotropic tide interacting with a ridge).
I have used the method of Winters and D'Asaro (APE based, paper of 95) which is also the tracer flux method used by Griffies et al. (2000). I compare my numerical tracer variance decay (using the diapycnal diffusivities computed from the tracer flux method) with the ones found by Burchard and Rennau (using the same simulations for the lock exchange of course). I sent a draft of this to John Warner and Hernan and will be happy to share it with you if you are interested.

Now, in collaboration with Florian Lemarié (from UCLA, florian@atmos.ucla.edu , http://www.atmos.ucla.edu/~florian/), we are implementing the Burchard and Rennau diagnostic. However, we first try to understand if this quantity is very pertinent for all the cases and if it can be apply to ANY advection schemes which for now we are still trying to figure out and are not sure. Thus we're conducting some 1D experiments. I know that Florian has this diagnostic coded for the UP3 (not sure if it is ROMS-agrif or rutgers) but it becomes tricky when using the flux corrector scheme (MPDATA)...

Anyway, feel free to contact any of us (both) for more info if you'd like,

Cheers,,

Flav

======================
Flavien Gouillon
OMP/SHOM
Physical Oceanography
http://www.coaps.fsu.edu/~gouillon

14 av Edouard Belin
31401 Toulouse France
Phone: 05 61 33 29 11
Fax : 05 61 33 29 16

rsignell
Posts: 123
Joined: Fri Apr 25, 2003 9:22 pm
Location: USGS

Re: Simple method for calculating Numerically Induced Mixing

#3 Post by rsignell » Thu Jun 17, 2010 8:49 pm

Flavien,

I'm glad to see that you and others are working on this. This subject came up when discussing improvements to Elizabeth North's LTRANS particle tracking model, which would use the horizontal mixing in the deterministic model (e.g. ROMS) to specify the horizontal random walk parameters for the particles. It seems that if you wanted to compare the particle concentrations from LTRANS to a dye release in ROMS, you would want to use the actual numerical mixing.

-Rich

Post Reply