open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi all,

we made a very large 3D model of the Scheldt estuary. On the seaboundary we impose water levels we extracted from a bigger model. This approach works fine. in our project we are now in the phase of running several scenario's with this model and in one of the scenrio's we got high instabilities on the seaboundary. We got high velocities. This was expected since the water levels we impose are including a sea level rise and are no longer in equilibrium with the current bathymetry. So we tried using the Thompson boundary. And this solved the instabilities and gave acceptable water levels in our model.

The problem is now that when running different scenario's (and in these different scenario's we only changed the bathymetry a little far upstream in the estuary) we get differences in water levels on the open sea boundary. the differences are small in the order of 1 or 2 cm but so are the differences we expect from the small bathymetry changes we made far upstream. If we want to compare scenario's, it is difficult to say if the small change in water level is due to the change in bathymetry or due to the different solution the thompson boundary imposes on the seaboundary.

My question is:
1. is there a way to control the thompson solution better?
2. I guess that hte partitioning of the mesh in subdomains influences the Thompson solution. Is there a way or has anyone already wrote some script to control the partitioning of the mesh into subdomains so that we can do this everytime in the same way?

Thanks

Sven

Hello Sven,

The Thompson boundary conditions are based on the theory of Riemann invariants for the Saint-Venant equations. The idea is to solve a linearised Riemann problem at the boundary, taking into account: 1. the exterior state -> what you want to prescribe, 2. the interior state -> coming from the computational domain.

So I don't think you can "force" a water depth with the Thompson formulation. The bathymetry change must modify the solution inside the domain, which yields different behaviour at the boundary...
I don't think the partitionning has an influence on this, it would seem surprising to me.

So I don't really come with a solution for you...

By the way, we tried a similar thing for a simulation in the Pacific Ocean and got instability issues too. Using Thompson in our case (where the fields are considered as constant on the vertical), while the water depths were very important with velocity/temperature changes only very close to the free-surface did not seem adapted. What HR Wallingford tried was to prescribe the velocities from the other model at the boundary, and then do a smoothing between the boundary and the interior on a certain width (they called it sponge layer). However, this kind of solution is not conservative and we could probably improve it.

This kind of problems is far from easy to treat and worth investigation. I think quite some people working on oceanic simulations have proposed approaches to prescribe boundary conditions coming from other models but I don't have much experience on this topic...

If you want to know more about this "sponge layer" approach, I have an svn branch with the code for it, and you can ask Sébastien Bourban about it too, he developped that approach.

Best regards,
Agnès

hello aleroy,
I want to read the subroutine named thomps.f, but i don't know about the theory of Riemann invariants for the Saint-Venant equations.Where i can get the introduce about the theory in the programming.
my best regards.

Hello,

You can find documentation about this in the book:

Hydrodynamics of free surface flows by Jean-Michel Hervouet.

It is described the chapter 4.6.

Best regards,
Agnès