MPI Redecomposition and Remapping Algorithms Used within a Multigrid Approach to Modeling of the Background Error Covariance for High-Resolution Data Assimilation

The development of a new multigrid approach to modeling of the background covariance error for use in high resolution data assimilation applications, e.g., the 3D Real Time Mesoscale Analysis and FV3-based convection-allowing assimilation applications, involves several MPI computational challenges that will be described, and their solution discussed, in this talk.

One of the encountered issues is how to efficiently re-decompose and remap the variables between the final analysis grid and the intermediate filter grids where the covariance computations are performed, since these are defined at different resolutions and on different sets of computational tasks. The adopted solution to this problem will be the main subject of the talk. Another encountered issue is parallelization of the multigrid algorithm, and its future generalizations, that will provide the most optimal scaling as both the grid resolution and domain size increase.

The computational performance of the new approach in a preliminary application running at a very high horizontal resolution of about 2.5 km will be presented.