Parallel Optimization of the MM5 Adjoint Model

The PSU/NCAR mesoscale model (MM5) and its adjoint model are widely used in some fields of meteorology such as adaptive observations for tropical cyclone prediction. Although the parallel optimization is provided in MM5, the corresponding adjoint model has not been parallelized, which limits the widespread use of MM5. So it is of great importance to develop a parallel MM5 adjoint model. In this paper, we propose a parallel implementation of the MM5 adjoint model by utilizing Runtime System Library (RSL) and Fortran Loop Index Converter (FLIC). Due to the character of the automatic transformation of FLIC, the workload of refactoring is far less than that based on classical MPI. Experiment results show that a near linear growth of acceleration could be achieved with appropriate grid decomposition.