MPAS variable-resolution mesh generation

[Lixiang]

Hi~

I have been working on variable-resolution meshes for MPAS these days. I want to use a 60km-3km mesh. I understand there is one on the MPAS-A website. However, I hope the 3km resolution can be expanded to a larger area, more like the 60km-10km case (please see the pictures attached).

I wonder if there is any way that I can generate the mesh myself, or if I can get help from experts on this forum. Thanks a lot!

 

[mgduda]

Unfortunately, we don't yet support (or even make available) the software that we use internally to generate MPAS-A meshes. However, I can look into generating a 60 - 3 km mesh with a larger refinement region. Would you be interested in a refinement region similar to the 60 - 10 km mesh, or would a region approximately the same size as in the 15 - 3 km (circular) mesh be sufficient?

 

[mgduda]

Also, note that the lead time to generate this mesh will likely be several months. So, if there's any chance that one of the existing meshes would work for your use case, that would certainly allow you to do your studies much sooner.

 

[Lixiang]

Also, note that the lead time to generate this mesh will likely be several months. So, if there's any chance that one of the existing meshes would work for your use case, that would certainly allow you to do your studies much sooner.

Thank you for the reply! In this case, I think I should find another way.
Also, may I ask why it would take such a long time to generate meshes? Is it because the resolution is too high and the refinement region is too large (I am interested in a refinement region similar to 60-10km mesh)?

 

[mgduda]

The generation of an SCVT can be viewed as an optimization problem, and the method that we use to perform this optimization -- Lloyd's method -- requires a very large number of iterations (typically O(10^6)), and each iteration can take several seconds. Part of each iteration can be parallelized, but there is still a part that is done in serial with our current software.

 

[Lixiang]

The generation of an SCVT can be viewed as an optimization problem, and the method that we use to perform this optimization -- Lloyd's method -- requires a very large number of iterations (typically O(10^6)), and each iteration can take several seconds. Part of each iteration can be parallelized, but there is still a part that is done in serial with our current software.

Thank you very much for the reply! It helps a lot.

 

[Lixiang]

The generation of an SCVT can be viewed as an optimization problem, and the method that we use to perform this optimization -- Lloyd's method -- requires a very large number of iterations (typically O(10^6)), and each iteration can take several seconds. Part of each iteration can be parallelized, but there is still a part that is done in serial with our current software.

Hi~
I have been using the Quasi-uniform meshes on the MPAS website. I noticed that the lowest resolution is 480 km. I was wondering how much time it would take to generate a 960 km mesh and is it possible for you to help me generate one. Thanks a lot for all the patient replies!