Speaker
Description
Since their advent in the 1950s, Numerical Weather Prediction (NWP) models have used structured lat-lon mesh grids for forecasting, which subdivide the 2D (horizontal) global surface into square grid cells of equal size. This has historically resulted in issues such as unnecessarily high resolution at the poles (the ‘pole problem’), and flow distortions at nest boundaries with the use of regional nested models that involve abrupt mesh transitions between coarse and fine domains.
In recent years, there has been an interest in the use of unstructured Spherical Centroidal Voronoi Tessellation (SCVT) mesh grids for NWP forecasting. With 5-, 6- and 7-sided polygons of variable cell size and geometry, these mesh grids are a type of staggered Arakawa C-grid, with the primal grid being a centroidal Voronoi tessellation and the dual grid being a Delaunay triangulation. By solving for normal velocities on cell edges and conserved quantities at the centroids, these grids scale advantageously on massive parallel computing systems and potentially allow for higher forecast skill due to their superior isotropy.
I detail how these grids are generated using Lloyd’s algorithm under a user-defined density function, creating variable-resolution meshes that seamlessly transition between coarse and fine regions to eliminate the problems faced by conventional lat-lon grids. This can be particularly useful in the regional context where small cities and countries such as Singapore require extremely high resolutions of 100-300m over their urban areas, but deal with high Lateral Boundary Condition (LBC) error rates when using parasitic nested lat-lon grids. I also explore the generation of highly-refined SCVT meshes with the JIGSAW-GEO software, and their use with the MPAS-Atmosphere model to perform high-resolution NWP over Singapore and regional NWP over the Southeast-Asia region.
