open TELEMAC-MASCARET The mathematically superior suite of solvers

Dear all,

I have simulation of an estuary and a part of a coastal sea. When I run it with salinity as a passive tracer, the model works fine. However, when I switch on the baroclinic pressure gradient by setting DENSITY LAW = 2, the simulation gives NaN values as output. With respect to this, I have two questions:

1.) Is there in Telemac a time step restriction when using DENISTY LAW = 2?
2.) What are the recommanded numerical settings when using DENISTY LAW = 2 ( I added the settings I used below)?
3.) Is there some way of checking whether the simulation goes wrong in a particular cell in the model?

Kind regards,

Alexander


TREATMENT OF NEGATIVE DEPTHS = 2
TIDAL FLATS = YES
OPTION FOR THE TREATMENT OF TIDAL FLATS = 1


SCHEME FOR ADVECTION OF VELOCITIES = 5
SCHEME FOR ADVECTION OF TRACERS = 14

SUPG OPTION = 0;0;0;0
IMPLICITATION FOR DEPTH = 1.0
IMPLICITATION FOR VELOCITIES = 1.0
IMPLICITATION FOR DIFFUSION = 1.0

MASS-LUMPING FOR DEPTH = 1.
MASS-LUMPING FOR VELOCITIES = 1.
MASS-LUMPING FOR DIFFUSION = 1.
FREE SURFACE GRADIENT COMPATIBILITY = 0.0

SOLVER FOR DIFFUSION OF VELOCITIES = 7
SOLVER FOR DIFFUSION OF TRACERS = 1
SOLVER FOR PROPAGATION = 7
OPTION OF SOLVER FOR PROPAGATION = 3

MAXIMUM NUMBER OF ITERATIONS FOR DIFFUSION OF TRACERS = 50
MAXIMUM NUMBER OF ITERATIONS FOR PROPAGATION = 20
MAXIMUM NUMBER OF ITERATIONS FOR DIFFUSION OF VELOCITIES = 10

ACCURACY FOR DIFFUSION OF VELOCITIES =1.0E-05
ACCURACY FOR DIFFUSION OF TRACERS = 1.0E-09
ACCURACY FOR PROPAGATION = 1.0E-05

INITIAL GUESS FOR DEPTH = 1 / Previous time step
MATRIX STORAGE = 3 / edge based storage

Hello,

If you have really tidal flats, the PSI scheme 5 for advection of velocity should not work and should be replaced by 1 (characteristics) or 14 (N scheme adapted to tidal flats). This has nothing to see with the active tracer but may be a first thing to look at.

With best regards,

Jean-Michel Hervouet

Hello,

Thank you for your reply. I changed the advection scheme to 1 (characteristic method), but unfortunately, I have the same problem.

Kind regards,

Alexander

Hello,

Now the next thing to check is whether you have prescribed the:

STANDARD VALUES FOR TRACERS (by default 0., there must be several if there are several tracers)

and the:

DENSITY FOR STANDARD VALUE (by default 1025.)

Regards,

JMH

Hello,

I used the default values for both of these variables (STANDARD VALUES FOR TRACER, DENSITY FOR STANDARD VALUE). I just have salt as tracer. I used INITIAL VALUES OF TRACERS = 33.0 (in PSU) as initial condition (salt water everywhere). I prescribe a salinity of 0 psu at the river boundaries (in combination with a discharge). the simulation is ment as a spin-up run , in oder to let the salinity gradient in the estuary develop.

Kind regards,

Alexander

Hello,

I enclose an example (in French) of a parameter file for a lake with salinity and temperature (and wind, rain and evaporation, etc.). You can inspire from it though you will have to translate.

I would try in particular :

HYDROSTATIC INCONSISTENCY FILTER : YES

With tidal flats you may have unphysical huge gradients of salinity on tidal flats, which may cause the crash.

TREATMENT OF FLUXES AT THE BOUNDARIES : 2;2;2;2;... (as many as open boundaries)

so that the salt flux at open boundaries is the water discharge times the prescribed value of salinity (instead of imposing brutally at once the prescribed value, which may cause an artificial flux). This is not a cause of crash, only of artificial fluxes at the beginning of computations.

With best regards,

Jean-Michel Hervouet

Hello,

Thank you very much for the example and your suggestions. I tried the two sugestions, and the simulation remeianed stable for a longer time, but unfortunatley not until the end of the run. In order to verify that the crash is the results of huge gradients on the tidal flats, I did a second simulation, in which I had adapted the grid, such there were no tidal flats (i.e. some high points in my domain were lowered), and indeed, that simulation remained stable. Is there other parameter, wihich I can use to diminish these gradients?

I noticed in the example that the key word MESH TRANSFORMATION is set to 3, while in my simulation, it was set to 2 (non-uniform sigma coordinates programmed in Fortran). I wonder whether you think this might have an effect, and it would be worthwhile to change this?

Kind regards,

Alexander Breugem

Hello Alexander,

There are other formulas for filtering, that are available in subroutine trisou.f. If you take trisou.f (library telemac3d, in directory sources), you will find a line :

! IF(INCHYD) FORMUL(6:6)='4'

and you can try to remove the comment ! so that it is enabled.

MESH TRANSFORMATION = 3 is for fixed planes in the mesh, while 2 is for evenly spaced planes. I don't think it will help to have fixed planes, though it generally triggers a lot of crushed elements.

I hope it works, I have not many ideas left after this...

Jean-Michel Hervouet