Development of an Ensemble Data Assimilation System with Flexible Parallelism, with Applications to the Ionosphere

Complex geophysical models often benefit from parallelism. So too, do data assimilation algorithms. However, parallelism significantly increases the complexity of a computer code and introduces numerous implementation tradeoffs. In this presentation we discuss efforts to develop a flexible parallel data assimilation system, which can support disparate parallelization schemes in the physical model and the data assimilation algorithm. This approach enables the parallelism of the data assimilation algorithm to be optimized independently of the physical model. The current implementation uses an ensemble Kalman filter based on a perturbed observation scheme, but the system has a modular design intended to support parallel execution of any filter algorithm that can operate separately on subsets of the model domain. We explore the scalability of the system using different parallelization schemes, and test the system on a 1-D advection solver and on the ionospheric model SAMI3 (SAMI3 is Another Model of the Ionosphere).