Studies On Parallel Algorithm Of Bicubic Numerical Prediction Model

The research and development on atmospheric numerical model is one of the core technologies of numerical weather prediction.This paper firstly introduces the background of research in numerical atmospheric prediction model and the characteristics of two current prediction models. Then it proposes this paper's research object - bicubic numerical model, the importance of parallel computing in numerical atmospheric prediction and basic parallel theory.Firstly, this paper studies the physical packages and mathematical truss of bicubic numerical model, and analyzes its mathematical specialty. And it puts forward model's optimizations by analyzing the model's main procedures and its wall time. According to the model's time integration scheme, this paper studies the model's feasibility of parallelism, and it's parallel scheme.Secondly, this paper puts forward the model's decomposition according its data flow and current data decompositions. To solve the global grids' data conversion problem, it introduces the parallel block data transposition algorithm. To solve the noncontiguous prediction data output in the process of parallelism, it uses collective I/O to reduce the I/O communication costs in the processes.Finally, the model's parallelism was implemented on the cluster systems, the experimental results show that there is a great decrease in the executed time gets and the parallel algorithm achieves load-balance easily and gets better in speedup and efficiency.