2.6 Using hierarchical time-stepping to utilize MPAS-A computational resources for customized extreme variable-resolution meshes.
Ng, Ka-Ki, Kwan-Shu Tse, Yuk Sing Lui, Wai-Nang Leung, and Chi Chiu Cheung, ClusterTech Limited, Hong Kong
High-resolution
regional/local weather forecast has long been a simulation challenge due to limited
computational resources. In order to achieve higher accuracy within a limited
budget, focusing computational resources on the region of interest is desired.
The use of unstructured spherical centroidal Voronoi tessellation in the Model
for Prediction Across Scales - Atmosphere (MPAS-A) gives flexibility to use
customized variable-resolution meshes with refinements for regions of interest.
However, time integration becomes expensive when using a small
globally-constant time-step determined by CFL condition on the smallest cell.
We have implemented hierarchical time-stepping (HTS) to save computational
resources by using inhomogeneous time-step. MPAS-A integrations with and
without HTS were carried out on both standard and customized variable-resolution
meshes for three historical cases. By comparing with reanalysis and station
data, it was discovered that simulations with and without HTS have comparable
performance. With HTS, computational cost is reduced considerably and it is
more practical to use customized extreme variable-resolution meshes.