Data assimilation applications with large-scale numerical modelsudexhibit extreme requirements on computational resources. Goodudscalability of the assimilation system is necessary to make theseudapplications feasible. Sequential data assimilation methods based onudensemble forecasts, like ensemble-based Kalman filters, provide suchudgood scalability, because the forecast of each ensemble member can beudperformed independently. However, this parallelism has to be combinedudwith the parallelization of both the numerical model and the data udassimilation algorithm. In order to simplify the implementation ofudscalable data assimilation systems based on existing numerical models,udthe Parallel Data Assimilation Framework PDAF (http://pdaf.awi.de) has udbeen developed. PDAF provides support for implementing a data udassimilation system with parallel ensemble forecasts and parallel udnumerical models. Further, it includes several optimized parallel udfilter algorithms, like the Ensemble Transform Kalman Filter. udWe will discuss the philosophy behind PDAF as well as features and udscalability of data assimilation systems based on PDAF on the example udof data assimilation with the finite element ocean model FEOM.ud
展开▼