Satellite Radiance Data Assimilation: Code Migration to Scalable Architectures

摘要

We developed a highly scalable version of the MM5 4d-Var application. The application consists of a meteorological analysis code that provides accurate depictions of the state of the atmosphere, has been applied successfully to a large number of cases, and the results have been documented in peer reviewed forums. Prior to this project, this code was optimized for vector computer architectures. By rewriting this code in order to make it 'scale' (i.e. , performance increases linearly) as additional CPUs are devoted to the calculations, significant speedups were achieved on the class of computers referred to as Massively Parallel Processors. The 4d-Var technique performs a series of iterations and requires computational power measured in the tens of gigaflops for real-time application. Proven strategies and techniques were employed to develop the scaleable version of this code and to attain the objective performance metrics. Among the coding strategies employed was domain decomposition. A tested, scalable 4d-Var code was delivered, along with user documentation. During the later half of this contract, we investigated the potential impact of optical turbulence data on upper-air data analysis. The task involved the development of additional software for the MM5 4d-Var code and applying it to C(sup 2 sub n) data obtained by AFRL. The preliminary results from that exercise suggest that C(sup 2 sub n) data have the potential to improve upper-atmospheric analyses and the NWP model forecasts made from them.