Data Assimilation
Data assimilation combines observational information with numerical models to improve the model state and its prediction. The most common application of data assimilation is weather forecasting. However, also the state and prediction of ocean models can be improved by data assimilation, for example by utilizing satellite observations of sea surface temperature or sea surface height. Similarly, ocean-biogeochemical models can profit from the incorporation of satellite ocean chlorophyll data.
Sequential Data Assimilation
The research work in the Scientific Computing group at AWI focuses on sequential data assimilation methods. Parallel ensemble-based Kalman filter algorithms are well suited for parallel high-performance computers as they are highly scalable. Recent research includes:
- Studying the relation of different filter algorithms, like the Singular "Evolutive" Interpolated Kalman (SEIK) filter and the Ensemble Kalman Filter (EnKF). The study showed a particularly good performance of the SEIK filter. (Nerger et al. 2005a)
- Development of a localized variant (LSEIK). It results in further improved assimilation results by increasing the degrees of freedom for the correction of the model state compared to a global SEIK filter. (Nerger et al. 2006)
- Application of the LSEIK filter for state estimation in ocean models and coupled ocean-biogeochemical models. (Nerger et al. 2007; Nerger and Gregg 2007, 2008)
- Studying the relation of common localization methods (the so-called "covariance localization" and the "observation localization"). This research led to the developent of a new localization method, the "regulated localization". (Janjic et al. 2011; Nerger et al. 2012a)
- Studying the relation of the SEIK with to the ETKF (Ensemble Transform Kalman Filter). This research motivated the development of a new filter formulation, the "Error-subspace Transform Kalman Filter" (ESTKF). (Nerger et al. 2012b)
Parallel Data Assimilation Framework - PDAF
Related to the research projects the studied and developed filter algorithms, the data assimilation framework PDAF (Parallel Data Assimilation Framework) was developed. Initially, the framework allowed to easily compare different filter algorithms under identical conditions. However, its main task is now to simplify the implementation of parallelized data assimilation systems based on existing numerical models. PDAF provides complete implementations of sequential data assimilation algorithms, which are optimized for application on parallel computers. More information on PDAF can be found on the AWI web page on PDAF and on the project web pages of PDAF where also the source code package can be downloaded.
Projects
We participate in different research projects:
DeMarine Enviroment
Between 2008 and 2011 we have participated in this project to develop a data assimilation data system for the North and Baltic Seas for the German Maritime and Hydrographic Agency (BSH). The data assimilation system uses PDAF and the operational model BSHcmod of the BSH. The validation of the assimilation system has been performed in the section "Climate Dynamics" by Dr. S. Loza.
More information on the project is available on the web pages of DeMarine Environment.
Sangoma (EU FP7)
Since November 2011 we participate in the EU-funded project SANGOMA (Stochastic Assimilation for the Next Generation Ocean Model Applications). The aim of the project is to develop unified tools for data assimilation and new assimilation algorithms to support future operational systems with state-of-the-art data assimilation and related analysis tools.
More information is available on the web site of Sangoma.