气象资料三维变分同化阶段区域分解并行实现

Implementation of Phase Domain Decomposition Parallel Algorithm of Three-Dimensional Variational Data Assimilation

摘要: 变分同化由于能明显改善同化质量，正在成为数值天气预报的主流同化方法.研究三维变分同化的并行计算，提出了三维变分同化的阶段区域分解、观测资料的自适应划分算法、计算与通信重叠的矩阵转置和周边区域通信以及文件I/O方法，在此基础上实现了MPI并行三维变分原型系统，在由8个双CPU节点组成的Linux机群上并行加速比达到了11.9.

Abstract: The principles of 3DVAR(three-dimensional variational) data assimilation of meteorological observations are introduced, and parallel strategies of 3DVAR systems, such as phase domain decomposition, loading balance and message communication, etc. are studied. Based on the SPMD programming model and a message passing interface, the 3DVAR parallel program, is designed and implemented, and some analysis of the results are finally presented.