only needed for internal models (only homogenous available)

numsamples and numtraces can be read from member matrix (seismogram matrix)

ModelRead=0 will generate a model on the fly
ModelRead=1 will read a regular grid model and mesh onto discontinuous
grid if variable grid is used
ModelRead=2 will directly read a discontinuous grid, this only works for the
variable grid

even if the bottom layer of the staggered grid cell is not adressed  and has no effect on the simulation.
The ghostpoints Vy,Sxy,Syz in the elastic and
Vx,Vx,Vz in the acoustic case are not linked to the coarse grid and fine grid of the dicontinous grid

In the var grid not every grid coordinate is at a gridpoint
no automatic shifting to right position.

this is not necessary for the simulation.
Syy will be set implicitly during the update of the velocities to zero.
But at the free surface Ryy,p and Syy will have arbitrary values.
For further use of the wavefields (i.e. in FWI) this causes problems.

and matrizes

even if Syy at y=0 is not used to update velocities. The values at y=0 are visible at y=0
For furthe use of the wavefield (i.e. FWI) this values disturb the gradient.