Title: Generic parallelization strategies for data assimilation applications

Authors: Nils van Velzen (Vortech and TU Delft, The Netherlands)
Martin Verlaan (Deltares and TU Delft)

Parallel computing is used in many complex simulation models in order to reduce the computational time to acceptable size. The computation time is often increased by several orders when a simulation model is extended with data assimilation techniques. This increase of computational time can be reduced to an acceptable level by using parallel computing for the data assimilation algorithm.

Parallel computing can be used within the model, but also in the data assimilation algorithm. An efficient parallel computing strategy depends on various aspects such as the kind of model and the parallelization that it uses, the data assimilation algorithm and the amount of observations that are assimilated.

The open-source data assimilation toolbox OpenDA (www.openda.org) offers support for various kinds of parallel computing. It can accommodate models with various forms of parallel computing. OpenDA aims at providing a generic parallel performance that is comparable with dedicated implementations of parallel data assimilation systems for a single model taking all complexities and challenges of parallel data assimilation methods into account.

Here, the efficiency of parallel computing is illustrated with data-assimilation using Kalman type algorithms and calibration algorithms for the SWAN wave model, the WAQUA shallow water model and the LOTOS-EUROS air quality model. For most applications some degree of parallel computing can be achieved with the generic tools independent of the specific model or application.

For the attainable speedup, the type of data assimilation algorithm is very important. In general, algorithms that are very computationally efficient when run sequentially are not always the best choice for parallel computing. They may contain dependencies that can not be parallelized efficiently.


nasaLogo
GMAO Head: Michele Rienecker
Global Modeling and Assimilation Office
NASA Goddard Space Flight Center
Curator: Nikki Privé
Last Updated: May 27 2011