open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello,

Me again. I have a problem where, when using a less refined mesh my free surface behaves normally - it reaches an expected height and remains there whilst the solution converges. However using a more refined mesh but the same initial conditions the simulation behaves the same until ~165s when the elevation suddenly drops (see below). The log file shows this is the key timestep when the mass balance drops.

I cannot fathom what's making the free surface change so suddenly at that time-step. The only thing I'm changing is the mesh refinement - comparing the two the geometry are close matches and there aren't any holes in the mesh as far as I can see. I've sued auto-adjust nodes in blue kenue to try and reduce error, and CHECKING THE MESH option hasn't thrown up any issues as far as I can see.

Does anyone have any ideas as why there might be a sudden change in elevation like this?

Less refined mesh:


More refined mesh:

Refined mesh file attached, if anyone wants to have a look...

I cannot open your mesh here with BlueKenue - an invalid format.

But anyway, getting different results with different mesh resolutions is not good news, which suggests your model is sensitive to mesh resolution. What if you use a even finer grid, do you see the sudden drop or not?

Have you compared the elevations near your observation point in your coarse and finer meshes? Are there any steep slope that is missed in your coarse mesh but caught in your finer mesh?

Hailiang

Hi Shenh,

Thanks for the reply!

The mesh is double-precision selafin, which may be why if you're not using the newest beta version? Unsure. I've attached a single-precision .slf file just in case.

I've also attached the .cas file, as I realise it's unhelpful to discuss parameters without a reference!

What's puzzling me is that the free surface elevation is as expected for the first 165s or so but the model doesn't seem to be able to sustain it.The main thing is that the high res mesh requires a very high drag coeff' for the free surface to reach the expected level.

I do expect some differences between my low res and high res mesh, but it's how the free surface is behaving which is puzzling me. My current mesh is already very high resolution (edge length = 0.1m) so I'm testing with a 'halfway' resolution between the two currently. Comparing the meshes, the biggest differences are around 0.01m in some of the pools in the channel, with most of the differences concentrated in a pool near the outlet of the river. Most of the differences in channel geometry is millimeter in scale which I thought acceptable.


Given the above I assume the issue is stemming from model sensitivity. I'm trying different solvers and combinations of preconditioning. I've increased the accuracy of K-E and velocities to 1.E-8. I don't suppose you could advise which might be a stable combination, or what else should really need to be checked beyond the solvers? Perhaps different numbers for KMIN and EMIN might help...

Using my two refined meshes I get a similar elevation. My rough mesh has a different elevation. Assuming the more refined meshes are more accurate, what is confusing is that the rough mesh has the better calibrated free surface rather than the refined mesh. My questions are:

- How can the refined mesh have a lower, less accurate free surface than the rough one?
- Changes to my drag coeffcient results in very slight changes to the free surface compared to larger ones for the rough mesh. What about the mesh controls this sensitivity?

Either I have to accept unrealistic drag coefficent for my free surface, or find a way to rectify this. I've thought of testing sensitivity to the time-step (I usually use 0.4s) and perhaps introducing more diffusion...

Anyone know why this might happen?

Progress!

My free surface sensitivity seems to be dependant on the number of layers.Testing the refined mesh with a different number of layers but with the same roughness coefficient seems to support this: less layers = more sensitivity to the roughness coefficients. Free surface sensitivity diminishes with more layers.

My reasoning is when introducing more horizintal levels onto the mesh the influence of the roughness decreases as vertical resolution increases, as more vertical structure in the fluid flow accounts for more of the bathymetry. Therefore the free surface is less sensitive to roughness as the effects of bathymetry dominates with more levels. However I'm puzzled as I don't think would alter the height of the roughness layer.

Is this accurate for changes in free surface sensitivity, or is the cause mis-identified as the number of layers?