SC1.1

Data assimilation combines observational data with a numerical model. It is commonly used in numerical weather prediction, but is also applied in oceanography, hydrology and other areas of Earth system science. By integrating observations with models in a quantitative way, data assimilation allows to estimate model states with reduced uncertainty, e.g. to initialize model forecasts. Also, data assimilation can estimate parameters that control processes in the model or fluxes, which can be difficult or impossible to measure. As such, data assimilation can use observations to provide information about non-observable quantities if the model represents those. The combination of modelled and observed data requires error estimates for both sources of information. In ensemble data assimilation the error in the model state is estimated by an ensemble of model state realizations. This ensemble not only provides estimates of uncertainties, but also of cross-correlations between different model variables or parameters. The uncertainty estimate from the ensemble is then used by the assimilation method, and the most widely known is the ensemble Kalman filter.

To simplify the implementation and use of ensemble data assimilation, the Parallel Data Assimilation Framework - PDAF - has been developed. PDAF is a freely available open-source software (http://pdaf.awi.de) that provides ensemble-based data assimilation methods like the ensemble Kalman filter, but also allows to perform pure ensemble simulations. PDAF is designed such that it can be used from small toy problems running on notebook computers up to high-dimensional Earth system models running on supercomputers.

This course is both for the novices as well as for data-assimilation experts. It will be useful for novices who have a modelling application and observations and are interested in applying data assimilation, but haven't found a starting point yet. Data-assimilation experts who want to enhance the performance of their applications, or are keen to accelerate development of new data-assimilation methods and new applications will also benefit from the course.

The course will first provide an introduction to the ensemble data assimilation methodology. Then, it will explain the implementation concept of PDAF and finally provide a hands-on example of building a data assimilation system based on a numerical model. This practical introduction will prepare the participants to build a data assimilation system for their numerical models with PDAF and hence provide a quick start for applying ensemble data assimilation to their individual problems.

Participants are invited to bring their own notebook computer to run the hands-on examples themselves. For this, a Fortran compiler and the BLAS and LAPACK libraries are required. Matlab or Python would also be handy for plotting. Given the overall limited capacity of the Wifi network during the conference, it is recommended that you download PDAF from http://pdaf.awi.de before the short course if you like to do the hands-on example on your own notebook computer.

Public information:
Apart from the description above, we will provide in the Short Course a version of PDAF which only includes the relevant features for the hand-on examples and that does not require to register on the PDAF web site. If you like to run the hands-on example it would also be useful if you have OpenMPI installed (or any other MPI library), but there will also be an example that does not require MPI.

Share:
Co-organized as AS6.4/HS12.9/NP10.4/OS5.3
Convener: Lars Nerger | Co-conveners: Maria Broadbridge, Gernot Geppert, Peter Jan van Leeuwen
Programme
| Thu, 11 Apr, 14:00–15:45
 
Room -2.85
Data assimilation combines observational data with a numerical model. It is commonly used in numerical weather prediction, but is also applied in oceanography, hydrology and other areas of Earth system science. By integrating observations with models in a quantitative way, data assimilation allows to estimate model states with reduced uncertainty, e.g. to initialize model forecasts. Also, data assimilation can estimate parameters that control processes in the model or fluxes, which can be difficult or impossible to measure. As such, data assimilation can use observations to provide information about non-observable quantities if the model represents those. The combination of modelled and observed data requires error estimates for both sources of information. In ensemble data assimilation the error in the model state is estimated by an ensemble of model state realizations. This ensemble not only provides estimates of uncertainties, but also of cross-correlations between different model variables or parameters. The uncertainty estimate from the ensemble is then used by the assimilation method, and the most widely known is the ensemble Kalman filter.

To simplify the implementation and use of ensemble data assimilation, the Parallel Data Assimilation Framework - PDAF - has been developed. PDAF is a freely available open-source software (http://pdaf.awi.de) that provides ensemble-based data assimilation methods like the ensemble Kalman filter, but also allows to perform pure ensemble simulations. PDAF is designed such that it can be used from small toy problems running on notebook computers up to high-dimensional Earth system models running on supercomputers.

This course is both for the novices as well as for data-assimilation experts. It will be useful for novices who have a modelling application and observations and are interested in applying data assimilation, but haven't found a starting point yet. Data-assimilation experts who want to enhance the performance of their applications, or are keen to accelerate development of new data-assimilation methods and new applications will also benefit from the course.

The course will first provide an introduction to the ensemble data assimilation methodology. Then, it will explain the implementation concept of PDAF and finally provide a hands-on example of building a data assimilation system based on a numerical model. This practical introduction will prepare the participants to build a data assimilation system for their numerical models with PDAF and hence provide a quick start for applying ensemble data assimilation to their individual problems.

Participants are invited to bring their own notebook computer to run the hands-on examples themselves. For this, a Fortran compiler and the BLAS and LAPACK libraries are required. Matlab or Python would also be handy for plotting. Given the overall limited capacity of the Wifi network during the conference, it is recommended that you download PDAF from http://pdaf.awi.de before the short course if you like to do the hands-on example on your own notebook computer.
Public information: Apart from the description above, we will provide in the Short Course a version of PDAF which only includes the relevant features for the hand-on examples and that does not require to register on the PDAF web site. If you like to run the hands-on example it would also be useful if you have OpenMPI installed (or any other MPI library), but there will also be an example that does not require MPI.