Parallelization of a Global Circulation Model

Issues regarding the development and implementation of parallel algorithms for global circulation models (GCMs) will be discussed. GCMs are numerical simulations for weather forecasting and climate prediction. Namely, we are interested in developing a parallel Semi-Lagrangian algorithm for GCMs with a view to large-grain parallelism, such as what can be exploited on the Kendall Square architecture. Sequential methods have already been demonstrated on simplified models, and are currently being implemented for full GCMs. Our research focus is on both the computation of the trajectories for the Semi-Lagrangian method and on the numerical techniques involved with solving the linear systems arising in fluid-dynamics applications. We study GMRES, a Krylov-based iterative method for solving these systems. In order to compete with other existing methods, such as multigrid solvers, that are employed in current Semi-Lagrangian models, effective parallel preconditioners must be constructed.