Parallel Algorithm and Its Application of a Nonhydrostatic Semi-Implicit Semi-Lagrangian Global Model

GRAPES (Global/Regional Assimilation and PrEdiction System) is a new developed numerical weather prediction system and will be implemented operationally in the next few years at China Meteorological Administration (CMA). For a global semi-implicit semi-Lagrangian numerical prediction model formulated in spherical coordinates, due to the convergence of meridians, the longitudinal grid size decreases toward zero as the poles are approached. Therefore, parallelism near the poles is a tough issue. With efficiency, portability, maintainability and extensibility requirements, a cap-longitude-latitude decomposition parallel algorithm is proposed and realized adherence to the architectures of high-performance computers at CMA. The results indicate that the computing performance of the proposed algorithm is good on IBM-cluster 1600 at CMA. And it can resolve effectively the occurrence of calculated zonal wind exceeding the maximum of the halo regions when locating semi-Lagrangian departure points. The algorithm is efficient and stable and can meet the operational implementation requirement.