A parallel semi-Lagrangian algorithm for limited area weather models; Scalability and comparison with the Eulerian formulation

摘要

Massively parallel processing is expected to be of major importance in numerical weather prediction. This is due to the time constraints under which forecasts are produced, combined with the increased work load resulting from more accurate computations. The Shallow Water equations serve as a primary test problem for numerical methods used in modeling atmospheric flow. We present a parallel semi-Lagrangian algorithm for these equations and compare its parallel properties to those of the Eulerian approach. We argue that the two approaches scale equally well on parallel systems. The semi-Lagrangian (SLT) algorithm has the additional advantage of allowing a longer time step.