Tuesday, 31 July 2001
A MM5-based four-dimensional variational analysis system developed for distributed memory multiprocessor computers
This presentation provides an overview of a project underway to develop an MM5-based mesoscale four-dimensional variational analysis (4DVAR) system that is optimized to run on distributed-memory multiple processor machines. The current version of the MM5 4DVAR system, which is based on MM5 version 1, is coded for single processor machines. In addition, the 4DVAR system is being developed to extensively make use of satellite observations so that it will be useful in regions where conventional observations are not available. The first step of the project involves upgrading the non-linear, tangent-linear, and adjoint models of the MM5 4DVAR system to be compatible with the current release of MM5 which is now at version 3.4. The tangent-linear and adjoint components have been created using the Tangent-linear and Adjoint Compiler (TAMC) developed by Giering (1998). (Details on the development of the tangent-linear and adjoint models will be discussed in detail in another presentation by Nehrkorn et al. in this conference). The non-linear model of the MM5 (version 3.4) already is designed to run on distributed memory processor machines. The design of the parallel version of the tangent-linear model will closely follow that used for the non-linear model since both are similar in structure. In general, the parallel version of the adjoint model will make use of the same parallel programming principles employed in the forward models (i.e. horizontal domain decomposition, MPI-based communications), however issues of establishing owner computes, possible false recursion, and disk I/O will need to be addressed. When complete, the components will be controlled with an incremental assimilation driver that will allow the 4DVAR system to be compatible with other mesoscale models such as the WRF. Project completion is scheduled for early 2003.
Reference: R. Giering and T. Kaminski, 1998: ACM Trans. on Math. Software, 24(4): 437-474