Modelling differences based on computer?

Dominik · 2019-05-17 11:22:57

I've been running a few simulations on different computers. When running the exact same simulation on different computers I could notice that the output slightly differs. Does this imply a problem with my model or is this a normal thing?

Matthias Bürgler · 2019-05-20 08:19:00

Hi Dominik

Are you using the same operating system and the same BASEMENT version on both computers? Can you also find differences if you run the same simulation on the same computer, and what order of magnitude are the deviations?

Best regards
Matthias

Dominik · 2019-05-20 12:15:36

yes, everything is the same. I'm working with Basement v2.8
The latest two simulations (on the same computer) show the following differences:

TIME = 1.800000e+04 hyd. Time Step = 6.413185e-03 (SD: Hasliaare)
Qin[1] = 2.000000e+02 [m3/s] (HasliaareInflow)
Qout[2] = 1.300675e+02 [m3/s] (HasliaareOutflow)
Qout[3] = 2.241300e+02 [m3/s] (LakeBrienz)
-> Domain: Run Time of the simulation= 3567.76 sec
-> Domain: Finished simulation.

TIME = 1.800000e+04 hyd. Time Step = 8.250480e-03 (SD: Hasliaare)
Qin[1] = 2.000000e+02 [m3/s] (HasliaareInflow)
Qout[2] = 1.232932e+02 [m3/s] (HasliaareOutflow)
Qout[3] = 1.614050e+02 [m3/s] (LakeBrienz)
-> Domain: Run Time of the simulation= 3791.15 sec
-> Domain: Finished simulation.

when comparing the generated shapefiles with maximum water depth and time of maximum waterdepth a delay of about 30sek in time (t_max_depth) and differences in waterdepth varying from -7cm to +4cm per node (t_max_depth) could be observed.

Matthias Bürgler · 2019-05-23 08:35:59

Hi Dominik

This could be related to distortions of the mesh (small element angles, irregular elements, two breaklines very close together or ontop of eachother) or issues at your boundary conditions.

Did you generate the mesh with BASEmesh? If so what minimum angle did you use?

Also check if the output time step and calculation time step is the same, or when the calculation time step starts diverging.

Also: did you run the simulation in parallel?

Best regards
Matthias

Dominik · 2019-05-27 12:50:22

Hi Matthias,
thanks for your reply. I did indeed generate the mesh with BASEmesh based on the default angle configuration of 28° as I tried to avoid too sharp triangles. Further as indicated in the User manual ("smaller angles lead to less elements, while larger angles lead to more elements", p. 54) I didn't want to have more elements as this would increase the computation time of the simulation.

What do you mean with the comparison between the calculation time step and the output time step? Did you mean the comparison of the line 1 in the post #3 further up in this topic (TIME = 1.800000e+04 hyd. Time Step = 8.250480e-03 (SD: Hasliaare))? If so, the latest simulation I ran started diverging the TIME of 1.2e+3 corresponding to a hydrological time step of 7.995779e-02 and 7.846531e-03, respectivly.

Yes, I ran the simulation on 15 threads.

Thank you very much for your time
Dominik

Matthias Bürgler · 2019-05-27 14:23:48

Hi Dominik

What do you mean with the comparison between the calculation time step and the output time step?

What I meant is a comparison of the computational timestep between the two simulations, resp. if they also converge. But this seems to be the case.

Regarding the element angle: 28° is a minimum that should be satisfied. Larger angles indeed lead to more elements, but also to more numerical stability.
In another post you mentioned element areas of 0. Did you solve this issue? If so, what was the problem? Maybe the deviations you observe could still be related to this. Knowing what your mesh looks like could possibly resolve the observed problems.

Best regards
Matthias

Dominik · 2019-05-27 15:39:51

Hi Matthias,

yes, I could get rid of the element areas with 0. The problem was (at least in my opinion) the different XY-Tolerance and -Resolution between ArcGIS and QGIS. ArcGIS can differ two points down to 0.001m whereas QGIS works with 10^-6m. Therefore many breaklines I snapped in ArcGIS were not snapped in QGIS anymore. Therefore, I rebuilt the breaklines in QGIS and always ensured an absolute correct snapping (until 10^-6m); this did the trick

For the mesh, I'll send you an email in a few minutes.

Thanks very much
Dominik

User Forum of Software BASEMENT

#1 2019-05-17 11:22:57

Modelling differences based on computer?

#2 2019-05-20 08:19:00

Re: Modelling differences based on computer?

#3 2019-05-20 12:15:36

Re: Modelling differences based on computer?

#4 2019-05-23 08:35:59

Re: Modelling differences based on computer?

#5 2019-05-27 12:50:22

Re: Modelling differences based on computer?

#6 2019-05-27 14:23:48

Re: Modelling differences based on computer?

#7 2019-05-27 15:39:51

Re: Modelling differences based on computer?

Board footer