Mesoscale predictability and ensemble prediction

Principal Investigator

Prof. Dr. Martin Ehrendorfer
Institut für Meteorologie und Geophysik der
Universität Innsbruck
Innrain 52
A-6020 Innsbruck
Austria

martin.ehrendorfer@uibk.ac.at

Other researchers: Dr Alexander Gohm, Felix Schüller, Andreas Mühlbauer

Project description

Research in this Special Project is concerned with atmospheric predictability on the mesoscale with particular focus on: (i) mesoscale and small-scale predictability, (ii) ensemble prediction, and (iii) data assimilation. Through the potential of rapid growth of errors in the assimilated initial state of atmospheric models, a strong connection exists between atmospheric predictability and data assimilation research; thus, also the above topics are strongly connected.

In topic (i) it is planned to carry out predictability experiments with a regional mesoscale model at the Institute for Meteorology and Geophysics, Univ. Innsbruck (IMGI), on the basis of initial and boundary perturbations obtainable from ECMWF. Such experiments should be complemented with high-resolution integrations within the ECMWF ensemble prediction system (EPS). Another aspect of this topic is the investigation of the scale dependence of measures of predictability, to be extended into increasingly smaller scales, such as error doubling times for specific model variables. Again, to assess predictability estimates for such variables, both integrations at ECMWF and IMGI are necessary.

In topic (ii) it is planned to extend the numerical experimentation performed in the previous ECMWF special project Singular vector-based multivariate normal sampling in ensemble prediction to assess the properties of initial singular vector (SV)-based ensemble perturbations for the ECMWF EPS. The basic idea of this SV-based technique is to refer to the SV-decomposition of the analysis error covariance matrix P^a provided by N SVs as the basis to generate a set of M ensemble perturbations. That SV-decomposition is used to generate a multivariately normal sample from the estimate of the initial probability density function.

In topic (iii) it is planned to explore data assimilation methodology. Investigations will focus on experimentation with intermediate models, for which Kalman filter methods in their full and approximated forms, such as, e.g., ensemble methods or the reduced-rank Kalman filter (RRKF), can be directly compared. Experimentation with such an intermediate model has demonstrated the positive properties of the RRKF. Beyond insight that can be obtained from such intermediate model experiments, it seems in addition desirable to exploit further within the ECMWF system the capability to extract P^a from a variational assimilation experiment (through differences of cost function gradients), as well as to bring knowledge of P^f back into the variational cost function. That capability has already been configured at ECMWF in a low-resolution context in an experimental phase. Its extension is planned to be investigated in this special project, as a means of bringing flow-dependence into the background error statistics.

The proposed project duration is 2004-2007. Estimated computer allocations are 8000 HPCF units per year.

Final Report

Additional information

Project started in 2007.